tools/lib/bpf/libbpf.c

   1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
   2
   3 /*
   4  * Common eBPF ELF object loading operations.
   5  *
   6  * Copyright (C) 2013-2015 Alexei Starovoitov <[email protected]>
   7  * Copyright (C) 2015 Wang Nan <[email protected]>
   8  * Copyright (C) 2015 Huawei Inc.
   9  * Copyright (C) 2017 Nicira, Inc.
  10  * Copyright (C) 2019 Isovalent, Inc.
  11  */
  12
  13 #ifndef _GNU_SOURCE
  14 #define _GNU_SOURCE
  15 #endif
  16 #include <stdlib.h>
  17 #include <stdio.h>
  18 #include <stdarg.h>
  19 #include <libgen.h>
  20 #include <inttypes.h>
  21 #include <limits.h>
  22 #include <string.h>
  23 #include <unistd.h>
  24 #include <endian.h>
  25 #include <fcntl.h>
  26 #include <errno.h>
  27 #include <ctype.h>
  28 #include <asm/unistd.h>
  29 #include <linux/err.h>
  30 #include <linux/kernel.h>
  31 #include <linux/bpf.h>
  32 #include <linux/btf.h>
  33 #include <linux/filter.h>
  34 #include <linux/limits.h>
  35 #include <linux/perf_event.h>
  36 #include <linux/ring_buffer.h>
  37 #include <sys/epoll.h>
  38 #include <sys/ioctl.h>
  39 #include <sys/mman.h>
  40 #include <sys/stat.h>
  41 #include <sys/types.h>
  42 #include <sys/vfs.h>
  43 #include <sys/utsname.h>
  44 #include <sys/resource.h>
  45 #include <libelf.h>
  46 #include <gelf.h>
  47 #include <zlib.h>
  48
  49 #include "libbpf.h"
  50 #include "bpf.h"
  51 #include "btf.h"
  52 #include "str_error.h"
  53 #include "libbpf_internal.h"
  54 #include "hashmap.h"
  55 #include "bpf_gen_internal.h"
  56 #include "zip.h"
  57
  58 #ifndef BPF_FS_MAGIC
  59 #define BPF_FS_MAGIC            0xcafe4a11
  60 #endif
  61
  62 #define BPF_INSN_SZ (sizeof(struct bpf_insn))
  63
  64 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
  65  * compilation if user enables corresponding warning. Disable it explicitly.
  66  */
  67 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
  68
  69 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  70
  71 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj);
  72 static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog);
  73
  74 static const char * const attach_type_name[] = {
  75         [BPF_CGROUP_INET_INGRESS]       = "cgroup_inet_ingress",
  76         [BPF_CGROUP_INET_EGRESS]        = "cgroup_inet_egress",
  77         [BPF_CGROUP_INET_SOCK_CREATE]   = "cgroup_inet_sock_create",
  78         [BPF_CGROUP_INET_SOCK_RELEASE]  = "cgroup_inet_sock_release",
  79         [BPF_CGROUP_SOCK_OPS]           = "cgroup_sock_ops",
  80         [BPF_CGROUP_DEVICE]             = "cgroup_device",
  81         [BPF_CGROUP_INET4_BIND]         = "cgroup_inet4_bind",
  82         [BPF_CGROUP_INET6_BIND]         = "cgroup_inet6_bind",
  83         [BPF_CGROUP_INET4_CONNECT]      = "cgroup_inet4_connect",
  84         [BPF_CGROUP_INET6_CONNECT]      = "cgroup_inet6_connect",
  85         [BPF_CGROUP_UNIX_CONNECT]       = "cgroup_unix_connect",
  86         [BPF_CGROUP_INET4_POST_BIND]    = "cgroup_inet4_post_bind",
  87         [BPF_CGROUP_INET6_POST_BIND]    = "cgroup_inet6_post_bind",
  88         [BPF_CGROUP_INET4_GETPEERNAME]  = "cgroup_inet4_getpeername",
  89         [BPF_CGROUP_INET6_GETPEERNAME]  = "cgroup_inet6_getpeername",
  90         [BPF_CGROUP_UNIX_GETPEERNAME]   = "cgroup_unix_getpeername",
  91         [BPF_CGROUP_INET4_GETSOCKNAME]  = "cgroup_inet4_getsockname",
  92         [BPF_CGROUP_INET6_GETSOCKNAME]  = "cgroup_inet6_getsockname",
  93         [BPF_CGROUP_UNIX_GETSOCKNAME]   = "cgroup_unix_getsockname",
  94         [BPF_CGROUP_UDP4_SENDMSG]       = "cgroup_udp4_sendmsg",
  95         [BPF_CGROUP_UDP6_SENDMSG]       = "cgroup_udp6_sendmsg",
  96         [BPF_CGROUP_UNIX_SENDMSG]       = "cgroup_unix_sendmsg",
  97         [BPF_CGROUP_SYSCTL]             = "cgroup_sysctl",
  98         [BPF_CGROUP_UDP4_RECVMSG]       = "cgroup_udp4_recvmsg",
  99         [BPF_CGROUP_UDP6_RECVMSG]       = "cgroup_udp6_recvmsg",
 100         [BPF_CGROUP_UNIX_RECVMSG]       = "cgroup_unix_recvmsg",
 101         [BPF_CGROUP_GETSOCKOPT]         = "cgroup_getsockopt",
 102         [BPF_CGROUP_SETSOCKOPT]         = "cgroup_setsockopt",
 103         [BPF_SK_SKB_STREAM_PARSER]      = "sk_skb_stream_parser",
 104         [BPF_SK_SKB_STREAM_VERDICT]     = "sk_skb_stream_verdict",
 105         [BPF_SK_SKB_VERDICT]            = "sk_skb_verdict",
 106         [BPF_SK_MSG_VERDICT]            = "sk_msg_verdict",
 107         [BPF_LIRC_MODE2]                = "lirc_mode2",
 108         [BPF_FLOW_DISSECTOR]            = "flow_dissector",
 109         [BPF_TRACE_RAW_TP]              = "trace_raw_tp",
 110         [BPF_TRACE_FENTRY]              = "trace_fentry",
 111         [BPF_TRACE_FEXIT]               = "trace_fexit",
 112         [BPF_MODIFY_RETURN]             = "modify_return",
 113         [BPF_LSM_MAC]                   = "lsm_mac",
 114         [BPF_LSM_CGROUP]                = "lsm_cgroup",
 115         [BPF_SK_LOOKUP]                 = "sk_lookup",
 116         [BPF_TRACE_ITER]                = "trace_iter",
 117         [BPF_XDP_DEVMAP]                = "xdp_devmap",
 118         [BPF_XDP_CPUMAP]                = "xdp_cpumap",
 119         [BPF_XDP]                       = "xdp",
 120         [BPF_SK_REUSEPORT_SELECT]       = "sk_reuseport_select",
 121         [BPF_SK_REUSEPORT_SELECT_OR_MIGRATE]    = "sk_reuseport_select_or_migrate",
 122         [BPF_PERF_EVENT]                = "perf_event",
 123         [BPF_TRACE_KPROBE_MULTI]        = "trace_kprobe_multi",
 124         [BPF_STRUCT_OPS]                = "struct_ops",
 125         [BPF_NETFILTER]                 = "netfilter",
 126         [BPF_TCX_INGRESS]               = "tcx_ingress",
 127         [BPF_TCX_EGRESS]                = "tcx_egress",
 128         [BPF_TRACE_UPROBE_MULTI]        = "trace_uprobe_multi",
 129         [BPF_NETKIT_PRIMARY]            = "netkit_primary",
 130         [BPF_NETKIT_PEER]               = "netkit_peer",
 131 };
 132
 133 static const char * const link_type_name[] = {
 134         [BPF_LINK_TYPE_UNSPEC]                  = "unspec",
 135         [BPF_LINK_TYPE_RAW_TRACEPOINT]          = "raw_tracepoint",
 136         [BPF_LINK_TYPE_TRACING]                 = "tracing",
 137         [BPF_LINK_TYPE_CGROUP]                  = "cgroup",
 138         [BPF_LINK_TYPE_ITER]                    = "iter",
 139         [BPF_LINK_TYPE_NETNS]                   = "netns",
 140         [BPF_LINK_TYPE_XDP]                     = "xdp",
 141         [BPF_LINK_TYPE_PERF_EVENT]              = "perf_event",
 142         [BPF_LINK_TYPE_KPROBE_MULTI]            = "kprobe_multi",
 143         [BPF_LINK_TYPE_STRUCT_OPS]              = "struct_ops",
 144         [BPF_LINK_TYPE_NETFILTER]               = "netfilter",
 145         [BPF_LINK_TYPE_TCX]                     = "tcx",
 146         [BPF_LINK_TYPE_UPROBE_MULTI]            = "uprobe_multi",
 147         [BPF_LINK_TYPE_NETKIT]                  = "netkit",
 148 };
 149
 150 static const char * const map_type_name[] = {
 151         [BPF_MAP_TYPE_UNSPEC]                   = "unspec",
 152         [BPF_MAP_TYPE_HASH]                     = "hash",
 153         [BPF_MAP_TYPE_ARRAY]                    = "array",
 154         [BPF_MAP_TYPE_PROG_ARRAY]               = "prog_array",
 155         [BPF_MAP_TYPE_PERF_EVENT_ARRAY]         = "perf_event_array",
 156         [BPF_MAP_TYPE_PERCPU_HASH]              = "percpu_hash",
 157         [BPF_MAP_TYPE_PERCPU_ARRAY]             = "percpu_array",
 158         [BPF_MAP_TYPE_STACK_TRACE]              = "stack_trace",
 159         [BPF_MAP_TYPE_CGROUP_ARRAY]             = "cgroup_array",
 160         [BPF_MAP_TYPE_LRU_HASH]                 = "lru_hash",
 161         [BPF_MAP_TYPE_LRU_PERCPU_HASH]          = "lru_percpu_hash",
 162         [BPF_MAP_TYPE_LPM_TRIE]                 = "lpm_trie",
 163         [BPF_MAP_TYPE_ARRAY_OF_MAPS]            = "array_of_maps",
 164         [BPF_MAP_TYPE_HASH_OF_MAPS]             = "hash_of_maps",
 165         [BPF_MAP_TYPE_DEVMAP]                   = "devmap",
 166         [BPF_MAP_TYPE_DEVMAP_HASH]              = "devmap_hash",
 167         [BPF_MAP_TYPE_SOCKMAP]                  = "sockmap",
 168         [BPF_MAP_TYPE_CPUMAP]                   = "cpumap",
 169         [BPF_MAP_TYPE_XSKMAP]                   = "xskmap",
 170         [BPF_MAP_TYPE_SOCKHASH]                 = "sockhash",
 171         [BPF_MAP_TYPE_CGROUP_STORAGE]           = "cgroup_storage",
 172         [BPF_MAP_TYPE_REUSEPORT_SOCKARRAY]      = "reuseport_sockarray",
 173         [BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE]    = "percpu_cgroup_storage",
 174         [BPF_MAP_TYPE_QUEUE]                    = "queue",
 175         [BPF_MAP_TYPE_STACK]                    = "stack",
 176         [BPF_MAP_TYPE_SK_STORAGE]               = "sk_storage",
 177         [BPF_MAP_TYPE_STRUCT_OPS]               = "struct_ops",
 178         [BPF_MAP_TYPE_RINGBUF]                  = "ringbuf",
 179         [BPF_MAP_TYPE_INODE_STORAGE]            = "inode_storage",
 180         [BPF_MAP_TYPE_TASK_STORAGE]             = "task_storage",
 181         [BPF_MAP_TYPE_BLOOM_FILTER]             = "bloom_filter",
 182         [BPF_MAP_TYPE_USER_RINGBUF]             = "user_ringbuf",
 183         [BPF_MAP_TYPE_CGRP_STORAGE]             = "cgrp_storage",
 184 };
 185
 186 static const char * const prog_type_name[] = {
 187         [BPF_PROG_TYPE_UNSPEC]                  = "unspec",
 188         [BPF_PROG_TYPE_SOCKET_FILTER]           = "socket_filter",
 189         [BPF_PROG_TYPE_KPROBE]                  = "kprobe",
 190         [BPF_PROG_TYPE_SCHED_CLS]               = "sched_cls",
 191         [BPF_PROG_TYPE_SCHED_ACT]               = "sched_act",
 192         [BPF_PROG_TYPE_TRACEPOINT]              = "tracepoint",
 193         [BPF_PROG_TYPE_XDP]                     = "xdp",
 194         [BPF_PROG_TYPE_PERF_EVENT]              = "perf_event",
 195         [BPF_PROG_TYPE_CGROUP_SKB]              = "cgroup_skb",
 196         [BPF_PROG_TYPE_CGROUP_SOCK]             = "cgroup_sock",
 197         [BPF_PROG_TYPE_LWT_IN]                  = "lwt_in",
 198         [BPF_PROG_TYPE_LWT_OUT]                 = "lwt_out",
 199         [BPF_PROG_TYPE_LWT_XMIT]                = "lwt_xmit",
 200         [BPF_PROG_TYPE_SOCK_OPS]                = "sock_ops",
 201         [BPF_PROG_TYPE_SK_SKB]                  = "sk_skb",
 202         [BPF_PROG_TYPE_CGROUP_DEVICE]           = "cgroup_device",
 203         [BPF_PROG_TYPE_SK_MSG]                  = "sk_msg",
 204         [BPF_PROG_TYPE_RAW_TRACEPOINT]          = "raw_tracepoint",
 205         [BPF_PROG_TYPE_CGROUP_SOCK_ADDR]        = "cgroup_sock_addr",
 206         [BPF_PROG_TYPE_LWT_SEG6LOCAL]           = "lwt_seg6local",
 207         [BPF_PROG_TYPE_LIRC_MODE2]              = "lirc_mode2",
 208         [BPF_PROG_TYPE_SK_REUSEPORT]            = "sk_reuseport",
 209         [BPF_PROG_TYPE_FLOW_DISSECTOR]          = "flow_dissector",
 210         [BPF_PROG_TYPE_CGROUP_SYSCTL]           = "cgroup_sysctl",
 211         [BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE] = "raw_tracepoint_writable",
 212         [BPF_PROG_TYPE_CGROUP_SOCKOPT]          = "cgroup_sockopt",
 213         [BPF_PROG_TYPE_TRACING]                 = "tracing",
 214         [BPF_PROG_TYPE_STRUCT_OPS]              = "struct_ops",
 215         [BPF_PROG_TYPE_EXT]                     = "ext",
 216         [BPF_PROG_TYPE_LSM]                     = "lsm",
 217         [BPF_PROG_TYPE_SK_LOOKUP]               = "sk_lookup",
 218         [BPF_PROG_TYPE_SYSCALL]                 = "syscall",
 219         [BPF_PROG_TYPE_NETFILTER]               = "netfilter",
 220 };
 221
 222 static int __base_pr(enum libbpf_print_level level, const char *format,
 223                      va_list args)
 224 {
 225         if (level == LIBBPF_DEBUG)
 226                 return 0;
 227
 228         return vfprintf(stderr, format, args);
 229 }
 230
 231 static libbpf_print_fn_t __libbpf_pr = __base_pr;
 232
 233 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
 234 {
 235         libbpf_print_fn_t old_print_fn;
 236
 237         old_print_fn = __atomic_exchange_n(&__libbpf_pr, fn, __ATOMIC_RELAXED);
 238
 239         return old_print_fn;
 240 }
 241
 242 __printf(2, 3)
 243 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
 244 {
 245         va_list args;
 246         int old_errno;
 247         libbpf_print_fn_t print_fn;
 248
 249         print_fn = __atomic_load_n(&__libbpf_pr, __ATOMIC_RELAXED);
 250         if (!print_fn)
 251                 return;
 252
 253         old_errno = errno;
 254
 255         va_start(args, format);
 256         __libbpf_pr(level, format, args);
 257         va_end(args);
 258
 259         errno = old_errno;
 260 }
 261
 262 static void pr_perm_msg(int err)
 263 {
 264         struct rlimit limit;
 265         char buf[100];
 266
 267         if (err != -EPERM || geteuid() != 0)
 268                 return;
 269
 270         err = getrlimit(RLIMIT_MEMLOCK, &limit);
 271         if (err)
 272                 return;
 273
 274         if (limit.rlim_cur == RLIM_INFINITY)
 275                 return;
 276
 277         if (limit.rlim_cur < 1024)
 278                 snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur);
 279         else if (limit.rlim_cur < 1024*1024)
 280                 snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024);
 281         else
 282                 snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024));
 283
 284         pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n",
 285                 buf);
 286 }
 287
 288 #define STRERR_BUFSIZE  128
 289
 290 /* Copied from tools/perf/util/util.h */
 291 #ifndef zfree
 292 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 293 #endif
 294
 295 #ifndef zclose
 296 # define zclose(fd) ({                  \
 297         int ___err = 0;                 \
 298         if ((fd) >= 0)                  \
 299                 ___err = close((fd));   \
 300         fd = -1;                        \
 301         ___err; })
 302 #endif
 303
 304 static inline __u64 ptr_to_u64(const void *ptr)
 305 {
 306         return (__u64) (unsigned long) ptr;
 307 }
 308
 309 int libbpf_set_strict_mode(enum libbpf_strict_mode mode)
 310 {
 311         /* as of v1.0 libbpf_set_strict_mode() is a no-op */
 312         return 0;
 313 }
 314
 315 __u32 libbpf_major_version(void)
 316 {
 317         return LIBBPF_MAJOR_VERSION;
 318 }
 319
 320 __u32 libbpf_minor_version(void)
 321 {
 322         return LIBBPF_MINOR_VERSION;
 323 }
 324
 325 const char *libbpf_version_string(void)
 326 {
 327 #define __S(X) #X
 328 #define _S(X) __S(X)
 329         return  "v" _S(LIBBPF_MAJOR_VERSION) "." _S(LIBBPF_MINOR_VERSION);
 330 #undef _S
 331 #undef __S
 332 }
 333
 334 enum reloc_type {
 335         RELO_LD64,
 336         RELO_CALL,
 337         RELO_DATA,
 338         RELO_EXTERN_LD64,
 339         RELO_EXTERN_CALL,
 340         RELO_SUBPROG_ADDR,
 341         RELO_CORE,
 342 };
 343
 344 struct reloc_desc {
 345         enum reloc_type type;
 346         int insn_idx;
 347         union {
 348                 const struct bpf_core_relo *core_relo; /* used when type == RELO_CORE */
 349                 struct {
 350                         int map_idx;
 351                         int sym_off;
 352                         int ext_idx;
 353                 };
 354         };
 355 };
 356
 357 /* stored as sec_def->cookie for all libbpf-supported SEC()s */
 358 enum sec_def_flags {
 359         SEC_NONE = 0,
 360         /* expected_attach_type is optional, if kernel doesn't support that */
 361         SEC_EXP_ATTACH_OPT = 1,
 362         /* legacy, only used by libbpf_get_type_names() and
 363          * libbpf_attach_type_by_name(), not used by libbpf itself at all.
 364          * This used to be associated with cgroup (and few other) BPF programs
 365          * that were attachable through BPF_PROG_ATTACH command. Pretty
 366          * meaningless nowadays, though.
 367          */
 368         SEC_ATTACHABLE = 2,
 369         SEC_ATTACHABLE_OPT = SEC_ATTACHABLE | SEC_EXP_ATTACH_OPT,
 370         /* attachment target is specified through BTF ID in either kernel or
 371          * other BPF program's BTF object
 372          */
 373         SEC_ATTACH_BTF = 4,
 374         /* BPF program type allows sleeping/blocking in kernel */
 375         SEC_SLEEPABLE = 8,
 376         /* BPF program support non-linear XDP buffer */
 377         SEC_XDP_FRAGS = 16,
 378         /* Setup proper attach type for usdt probes. */
 379         SEC_USDT = 32,
 380 };
 381
 382 struct bpf_sec_def {
 383         char *sec;
 384         enum bpf_prog_type prog_type;
 385         enum bpf_attach_type expected_attach_type;
 386         long cookie;
 387         int handler_id;
 388
 389         libbpf_prog_setup_fn_t prog_setup_fn;
 390         libbpf_prog_prepare_load_fn_t prog_prepare_load_fn;
 391         libbpf_prog_attach_fn_t prog_attach_fn;
 392 };
 393
 394 /*
 395  * bpf_prog should be a better name but it has been used in
 396  * linux/filter.h.
 397  */
 398 struct bpf_program {
 399         char *name;
 400         char *sec_name;
 401         size_t sec_idx;
 402         const struct bpf_sec_def *sec_def;
 403         /* this program's instruction offset (in number of instructions)
 404          * within its containing ELF section
 405          */
 406         size_t sec_insn_off;
 407         /* number of original instructions in ELF section belonging to this
 408          * program, not taking into account subprogram instructions possible
 409          * appended later during relocation
 410          */
 411         size_t sec_insn_cnt;
 412         /* Offset (in number of instructions) of the start of instruction
 413          * belonging to this BPF program  within its containing main BPF
 414          * program. For the entry-point (main) BPF program, this is always
 415          * zero. For a sub-program, this gets reset before each of main BPF
 416          * programs are processed and relocated and is used to determined
 417          * whether sub-program was already appended to the main program, and
 418          * if yes, at which instruction offset.
 419          */
 420         size_t sub_insn_off;
 421
 422         /* instructions that belong to BPF program; insns[0] is located at
 423          * sec_insn_off instruction within its ELF section in ELF file, so
 424          * when mapping ELF file instruction index to the local instruction,
 425          * one needs to subtract sec_insn_off; and vice versa.
 426          */
 427         struct bpf_insn *insns;
 428         /* actual number of instruction in this BPF program's image; for
 429          * entry-point BPF programs this includes the size of main program
 430          * itself plus all the used sub-programs, appended at the end
 431          */
 432         size_t insns_cnt;
 433
 434         struct reloc_desc *reloc_desc;
 435         int nr_reloc;
 436
 437         /* BPF verifier log settings */
 438         char *log_buf;
 439         size_t log_size;
 440         __u32 log_level;
 441
 442         struct bpf_object *obj;
 443
 444         int fd;
 445         bool autoload;
 446         bool autoattach;
 447         bool sym_global;
 448         bool mark_btf_static;
 449         enum bpf_prog_type type;
 450         enum bpf_attach_type expected_attach_type;
 451         int exception_cb_idx;
 452
 453         int prog_ifindex;
 454         __u32 attach_btf_obj_fd;
 455         __u32 attach_btf_id;
 456         __u32 attach_prog_fd;
 457
 458         void *func_info;
 459         __u32 func_info_rec_size;
 460         __u32 func_info_cnt;
 461
 462         void *line_info;
 463         __u32 line_info_rec_size;
 464         __u32 line_info_cnt;
 465         __u32 prog_flags;
 466 };
 467
 468 struct bpf_struct_ops {
 469         const char *tname;
 470         const struct btf_type *type;
 471         struct bpf_program **progs;
 472         __u32 *kern_func_off;
 473         /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */
 474         void *data;
 475         /* e.g. struct bpf_struct_ops_tcp_congestion_ops in
 476          *      btf_vmlinux's format.
 477          * struct bpf_struct_ops_tcp_congestion_ops {
 478          *      [... some other kernel fields ...]
 479          *      struct tcp_congestion_ops data;
 480          * }
 481          * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops)
 482          * bpf_map__init_kern_struct_ops() will populate the "kern_vdata"
 483          * from "data".
 484          */
 485         void *kern_vdata;
 486         __u32 type_id;
 487 };
 488
 489 #define DATA_SEC ".data"
 490 #define BSS_SEC ".bss"
 491 #define RODATA_SEC ".rodata"
 492 #define KCONFIG_SEC ".kconfig"
 493 #define KSYMS_SEC ".ksyms"
 494 #define STRUCT_OPS_SEC ".struct_ops"
 495 #define STRUCT_OPS_LINK_SEC ".struct_ops.link"
 496
 497 enum libbpf_map_type {
 498         LIBBPF_MAP_UNSPEC,
 499         LIBBPF_MAP_DATA,
 500         LIBBPF_MAP_BSS,
 501         LIBBPF_MAP_RODATA,
 502         LIBBPF_MAP_KCONFIG,
 503 };
 504
 505 struct bpf_map_def {
 506         unsigned int type;
 507         unsigned int key_size;
 508         unsigned int value_size;
 509         unsigned int max_entries;
 510         unsigned int map_flags;
 511 };
 512
 513 struct bpf_map {
 514         struct bpf_object *obj;
 515         char *name;
 516         /* real_name is defined for special internal maps (.rodata*,
 517          * .data*, .bss, .kconfig) and preserves their original ELF section
 518          * name. This is important to be able to find corresponding BTF
 519          * DATASEC information.
 520          */
 521         char *real_name;
 522         int fd;
 523         int sec_idx;
 524         size_t sec_offset;
 525         int map_ifindex;
 526         int inner_map_fd;
 527         struct bpf_map_def def;
 528         __u32 numa_node;
 529         __u32 btf_var_idx;
 530         __u32 btf_key_type_id;
 531         __u32 btf_value_type_id;
 532         __u32 btf_vmlinux_value_type_id;
 533         enum libbpf_map_type libbpf_type;
 534         void *mmaped;
 535         struct bpf_struct_ops *st_ops;
 536         struct bpf_map *inner_map;
 537         void **init_slots;
 538         int init_slots_sz;
 539         char *pin_path;
 540         bool pinned;
 541         bool reused;
 542         bool autocreate;
 543         __u64 map_extra;
 544 };
 545
 546 enum extern_type {
 547         EXT_UNKNOWN,
 548         EXT_KCFG,
 549         EXT_KSYM,
 550 };
 551
 552 enum kcfg_type {
 553         KCFG_UNKNOWN,
 554         KCFG_CHAR,
 555         KCFG_BOOL,
 556         KCFG_INT,
 557         KCFG_TRISTATE,
 558         KCFG_CHAR_ARR,
 559 };
 560
 561 struct extern_desc {
 562         enum extern_type type;
 563         int sym_idx;
 564         int btf_id;
 565         int sec_btf_id;
 566         const char *name;
 567         char *essent_name;
 568         bool is_set;
 569         bool is_weak;
 570         union {
 571                 struct {
 572                         enum kcfg_type type;
 573                         int sz;
 574                         int align;
 575                         int data_off;
 576                         bool is_signed;
 577                 } kcfg;
 578                 struct {
 579                         unsigned long long addr;
 580
 581                         /* target btf_id of the corresponding kernel var. */
 582                         int kernel_btf_obj_fd;
 583                         int kernel_btf_id;
 584
 585                         /* local btf_id of the ksym extern's type. */
 586                         __u32 type_id;
 587                         /* BTF fd index to be patched in for insn->off, this is
 588                          * 0 for vmlinux BTF, index in obj->fd_array for module
 589                          * BTF
 590                          */
 591                         __s16 btf_fd_idx;
 592                 } ksym;
 593         };
 594 };
 595
 596 struct module_btf {
 597         struct btf *btf;
 598         char *name;
 599         __u32 id;
 600         int fd;
 601         int fd_array_idx;
 602 };
 603
 604 enum sec_type {
 605         SEC_UNUSED = 0,
 606         SEC_RELO,
 607         SEC_BSS,
 608         SEC_DATA,
 609         SEC_RODATA,
 610 };
 611
 612 struct elf_sec_desc {
 613         enum sec_type sec_type;
 614         Elf64_Shdr *shdr;
 615         Elf_Data *data;
 616 };
 617
 618 struct elf_state {
 619         int fd;
 620         const void *obj_buf;
 621         size_t obj_buf_sz;
 622         Elf *elf;
 623         Elf64_Ehdr *ehdr;
 624         Elf_Data *symbols;
 625         Elf_Data *st_ops_data;
 626         Elf_Data *st_ops_link_data;
 627         size_t shstrndx; /* section index for section name strings */
 628         size_t strtabidx;
 629         struct elf_sec_desc *secs;
 630         size_t sec_cnt;
 631         int btf_maps_shndx;
 632         __u32 btf_maps_sec_btf_id;
 633         int text_shndx;
 634         int symbols_shndx;
 635         int st_ops_shndx;
 636         int st_ops_link_shndx;
 637 };
 638
 639 struct usdt_manager;
 640
 641 struct bpf_object {
 642         char name[BPF_OBJ_NAME_LEN];
 643         char license[64];
 644         __u32 kern_version;
 645
 646         struct bpf_program *programs;
 647         size_t nr_programs;
 648         struct bpf_map *maps;
 649         size_t nr_maps;
 650         size_t maps_cap;
 651
 652         char *kconfig;
 653         struct extern_desc *externs;
 654         int nr_extern;
 655         int kconfig_map_idx;
 656
 657         bool loaded;
 658         bool has_subcalls;
 659         bool has_rodata;
 660
 661         struct bpf_gen *gen_loader;
 662
 663         /* Information when doing ELF related work. Only valid if efile.elf is not NULL */
 664         struct elf_state efile;
 665
 666         struct btf *btf;
 667         struct btf_ext *btf_ext;
 668
 669         /* Parse and load BTF vmlinux if any of the programs in the object need
 670          * it at load time.
 671          */
 672         struct btf *btf_vmlinux;
 673         /* Path to the custom BTF to be used for BPF CO-RE relocations as an
 674          * override for vmlinux BTF.
 675          */
 676         char *btf_custom_path;
 677         /* vmlinux BTF override for CO-RE relocations */
 678         struct btf *btf_vmlinux_override;
 679         /* Lazily initialized kernel module BTFs */
 680         struct module_btf *btf_modules;
 681         bool btf_modules_loaded;
 682         size_t btf_module_cnt;
 683         size_t btf_module_cap;
 684
 685         /* optional log settings passed to BPF_BTF_LOAD and BPF_PROG_LOAD commands */
 686         char *log_buf;
 687         size_t log_size;
 688         __u32 log_level;
 689
 690         int *fd_array;
 691         size_t fd_array_cap;
 692         size_t fd_array_cnt;
 693
 694         struct usdt_manager *usdt_man;
 695
 696         char path[];
 697 };
 698
 699 static const char *elf_sym_str(const struct bpf_object *obj, size_t off);
 700 static const char *elf_sec_str(const struct bpf_object *obj, size_t off);
 701 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx);
 702 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name);
 703 static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn);
 704 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn);
 705 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn);
 706 static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx);
 707 static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx);
 708
 709 void bpf_program__unload(struct bpf_program *prog)
 710 {
 711         if (!prog)
 712                 return;
 713
 714         zclose(prog->fd);
 715
 716         zfree(&prog->func_info);
 717         zfree(&prog->line_info);
 718 }
 719
 720 static void bpf_program__exit(struct bpf_program *prog)
 721 {
 722         if (!prog)
 723                 return;
 724
 725         bpf_program__unload(prog);
 726         zfree(&prog->name);
 727         zfree(&prog->sec_name);
 728         zfree(&prog->insns);
 729         zfree(&prog->reloc_desc);
 730
 731         prog->nr_reloc = 0;
 732         prog->insns_cnt = 0;
 733         prog->sec_idx = -1;
 734 }
 735
 736 static bool insn_is_subprog_call(const struct bpf_insn *insn)
 737 {
 738         return BPF_CLASS(insn->code) == BPF_JMP &&
 739                BPF_OP(insn->code) == BPF_CALL &&
 740                BPF_SRC(insn->code) == BPF_K &&
 741                insn->src_reg == BPF_PSEUDO_CALL &&
 742                insn->dst_reg == 0 &&
 743                insn->off == 0;
 744 }
 745
 746 static bool is_call_insn(const struct bpf_insn *insn)
 747 {
 748         return insn->code == (BPF_JMP | BPF_CALL);
 749 }
 750
 751 static bool insn_is_pseudo_func(struct bpf_insn *insn)
 752 {
 753         return is_ldimm64_insn(insn) && insn->src_reg == BPF_PSEUDO_FUNC;
 754 }
 755
 756 static int
 757 bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog,
 758                       const char *name, size_t sec_idx, const char *sec_name,
 759                       size_t sec_off, void *insn_data, size_t insn_data_sz)
 760 {
 761         if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) {
 762                 pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n",
 763                         sec_name, name, sec_off, insn_data_sz);
 764                 return -EINVAL;
 765         }
 766
 767         memset(prog, 0, sizeof(*prog));
 768         prog->obj = obj;
 769
 770         prog->sec_idx = sec_idx;
 771         prog->sec_insn_off = sec_off / BPF_INSN_SZ;
 772         prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ;
 773         /* insns_cnt can later be increased by appending used subprograms */
 774         prog->insns_cnt = prog->sec_insn_cnt;
 775
 776         prog->type = BPF_PROG_TYPE_UNSPEC;
 777         prog->fd = -1;
 778         prog->exception_cb_idx = -1;
 779
 780         /* libbpf's convention for SEC("?abc...") is that it's just like
 781          * SEC("abc...") but the corresponding bpf_program starts out with
 782          * autoload set to false.
 783          */
 784         if (sec_name[0] == '?') {
 785                 prog->autoload = false;
 786                 /* from now on forget there was ? in section name */
 787                 sec_name++;
 788         } else {
 789                 prog->autoload = true;
 790         }
 791
 792         prog->autoattach = true;
 793
 794         /* inherit object's log_level */
 795         prog->log_level = obj->log_level;
 796
 797         prog->sec_name = strdup(sec_name);
 798         if (!prog->sec_name)
 799                 goto errout;
 800
 801         prog->name = strdup(name);
 802         if (!prog->name)
 803                 goto errout;
 804
 805         prog->insns = malloc(insn_data_sz);
 806         if (!prog->insns)
 807                 goto errout;
 808         memcpy(prog->insns, insn_data, insn_data_sz);
 809
 810         return 0;
 811 errout:
 812         pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name);
 813         bpf_program__exit(prog);
 814         return -ENOMEM;
 815 }
 816
 817 static int
 818 bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data,
 819                          const char *sec_name, int sec_idx)
 820 {
 821         Elf_Data *symbols = obj->efile.symbols;
 822         struct bpf_program *prog, *progs;
 823         void *data = sec_data->d_buf;
 824         size_t sec_sz = sec_data->d_size, sec_off, prog_sz, nr_syms;
 825         int nr_progs, err, i;
 826         const char *name;
 827         Elf64_Sym *sym;
 828
 829         progs = obj->programs;
 830         nr_progs = obj->nr_programs;
 831         nr_syms = symbols->d_size / sizeof(Elf64_Sym);
 832
 833         for (i = 0; i < nr_syms; i++) {
 834                 sym = elf_sym_by_idx(obj, i);
 835
 836                 if (sym->st_shndx != sec_idx)
 837                         continue;
 838                 if (ELF64_ST_TYPE(sym->st_info) != STT_FUNC)
 839                         continue;
 840
 841                 prog_sz = sym->st_size;
 842                 sec_off = sym->st_value;
 843
 844                 name = elf_sym_str(obj, sym->st_name);
 845                 if (!name) {
 846                         pr_warn("sec '%s': failed to get symbol name for offset %zu\n",
 847                                 sec_name, sec_off);
 848                         return -LIBBPF_ERRNO__FORMAT;
 849                 }
 850
 851                 if (sec_off + prog_sz > sec_sz) {
 852                         pr_warn("sec '%s': program at offset %zu crosses section boundary\n",
 853                                 sec_name, sec_off);
 854                         return -LIBBPF_ERRNO__FORMAT;
 855                 }
 856
 857                 if (sec_idx != obj->efile.text_shndx && ELF64_ST_BIND(sym->st_info) == STB_LOCAL) {
 858                         pr_warn("sec '%s': program '%s' is static and not supported\n", sec_name, name);
 859                         return -ENOTSUP;
 860                 }
 861
 862                 pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n",
 863                          sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz);
 864
 865                 progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs));
 866                 if (!progs) {
 867                         /*
 868                          * In this case the original obj->programs
 869                          * is still valid, so don't need special treat for
 870                          * bpf_close_object().
 871                          */
 872                         pr_warn("sec '%s': failed to alloc memory for new program '%s'\n",
 873                                 sec_name, name);
 874                         return -ENOMEM;
 875                 }
 876                 obj->programs = progs;
 877
 878                 prog = &progs[nr_progs];
 879
 880                 err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name,
 881                                             sec_off, data + sec_off, prog_sz);
 882                 if (err)
 883                         return err;
 884
 885                 if (ELF64_ST_BIND(sym->st_info) != STB_LOCAL)
 886                         prog->sym_global = true;
 887
 888                 /* if function is a global/weak symbol, but has restricted
 889                  * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF FUNC
 890                  * as static to enable more permissive BPF verification mode
 891                  * with more outside context available to BPF verifier
 892                  */
 893                 if (prog->sym_global && (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN
 894                     || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL))
 895                         prog->mark_btf_static = true;
 896
 897                 nr_progs++;
 898                 obj->nr_programs = nr_progs;
 899         }
 900
 901         return 0;
 902 }
 903
 904 static const struct btf_member *
 905 find_member_by_offset(const struct btf_type *t, __u32 bit_offset)
 906 {
 907         struct btf_member *m;
 908         int i;
 909
 910         for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
 911                 if (btf_member_bit_offset(t, i) == bit_offset)
 912                         return m;
 913         }
 914
 915         return NULL;
 916 }
 917
 918 static const struct btf_member *
 919 find_member_by_name(const struct btf *btf, const struct btf_type *t,
 920                     const char *name)
 921 {
 922         struct btf_member *m;
 923         int i;
 924
 925         for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
 926                 if (!strcmp(btf__name_by_offset(btf, m->name_off), name))
 927                         return m;
 928         }
 929
 930         return NULL;
 931 }
 932
 933 #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_"
 934 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
 935                                    const char *name, __u32 kind);
 936
 937 static int
 938 find_struct_ops_kern_types(const struct btf *btf, const char *tname,
 939                            const struct btf_type **type, __u32 *type_id,
 940                            const struct btf_type **vtype, __u32 *vtype_id,
 941                            const struct btf_member **data_member)
 942 {
 943         const struct btf_type *kern_type, *kern_vtype;
 944         const struct btf_member *kern_data_member;
 945         __s32 kern_vtype_id, kern_type_id;
 946         __u32 i;
 947
 948         kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT);
 949         if (kern_type_id < 0) {
 950                 pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n",
 951                         tname);
 952                 return kern_type_id;
 953         }
 954         kern_type = btf__type_by_id(btf, kern_type_id);
 955
 956         /* Find the corresponding "map_value" type that will be used
 957          * in map_update(BPF_MAP_TYPE_STRUCT_OPS).  For example,
 958          * find "struct bpf_struct_ops_tcp_congestion_ops" from the
 959          * btf_vmlinux.
 960          */
 961         kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX,
 962                                                 tname, BTF_KIND_STRUCT);
 963         if (kern_vtype_id < 0) {
 964                 pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n",
 965                         STRUCT_OPS_VALUE_PREFIX, tname);
 966                 return kern_vtype_id;
 967         }
 968         kern_vtype = btf__type_by_id(btf, kern_vtype_id);
 969
 970         /* Find "struct tcp_congestion_ops" from
 971          * struct bpf_struct_ops_tcp_congestion_ops {
 972          *      [ ... ]
 973          *      struct tcp_congestion_ops data;
 974          * }
 975          */
 976         kern_data_member = btf_members(kern_vtype);
 977         for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) {
 978                 if (kern_data_member->type == kern_type_id)
 979                         break;
 980         }
 981         if (i == btf_vlen(kern_vtype)) {
 982                 pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n",
 983                         tname, STRUCT_OPS_VALUE_PREFIX, tname);
 984                 return -EINVAL;
 985         }
 986
 987         *type = kern_type;
 988         *type_id = kern_type_id;
 989         *vtype = kern_vtype;
 990         *vtype_id = kern_vtype_id;
 991         *data_member = kern_data_member;
 992
 993         return 0;
 994 }
 995
 996 static bool bpf_map__is_struct_ops(const struct bpf_map *map)
 997 {
 998         return map->def.type == BPF_MAP_TYPE_STRUCT_OPS;
 999 }
1000
1001 /* Init the map's fields that depend on kern_btf */
1002 static int bpf_map__init_kern_struct_ops(struct bpf_map *map,
1003                                          const struct btf *btf,
1004                                          const struct btf *kern_btf)
1005 {
1006         const struct btf_member *member, *kern_member, *kern_data_member;
1007         const struct btf_type *type, *kern_type, *kern_vtype;
1008         __u32 i, kern_type_id, kern_vtype_id, kern_data_off;
1009         struct bpf_struct_ops *st_ops;
1010         void *data, *kern_data;
1011         const char *tname;
1012         int err;
1013
1014         st_ops = map->st_ops;
1015         type = st_ops->type;
1016         tname = st_ops->tname;
1017         err = find_struct_ops_kern_types(kern_btf, tname,
1018                                          &kern_type, &kern_type_id,
1019                                          &kern_vtype, &kern_vtype_id,
1020                                          &kern_data_member);
1021         if (err)
1022                 return err;
1023
1024         pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n",
1025                  map->name, st_ops->type_id, kern_type_id, kern_vtype_id);
1026
1027         map->def.value_size = kern_vtype->size;
1028         map->btf_vmlinux_value_type_id = kern_vtype_id;
1029
1030         st_ops->kern_vdata = calloc(1, kern_vtype->size);
1031         if (!st_ops->kern_vdata)
1032                 return -ENOMEM;
1033
1034         data = st_ops->data;
1035         kern_data_off = kern_data_member->offset / 8;
1036         kern_data = st_ops->kern_vdata + kern_data_off;
1037
1038         member = btf_members(type);
1039         for (i = 0; i < btf_vlen(type); i++, member++) {
1040                 const struct btf_type *mtype, *kern_mtype;
1041                 __u32 mtype_id, kern_mtype_id;
1042                 void *mdata, *kern_mdata;
1043                 __s64 msize, kern_msize;
1044                 __u32 moff, kern_moff;
1045                 __u32 kern_member_idx;
1046                 const char *mname;
1047
1048                 mname = btf__name_by_offset(btf, member->name_off);
1049                 kern_member = find_member_by_name(kern_btf, kern_type, mname);
1050                 if (!kern_member) {
1051                         pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n",
1052                                 map->name, mname);
1053                         return -ENOTSUP;
1054                 }
1055
1056                 kern_member_idx = kern_member - btf_members(kern_type);
1057                 if (btf_member_bitfield_size(type, i) ||
1058                     btf_member_bitfield_size(kern_type, kern_member_idx)) {
1059                         pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n",
1060                                 map->name, mname);
1061                         return -ENOTSUP;
1062                 }
1063
1064                 moff = member->offset / 8;
1065                 kern_moff = kern_member->offset / 8;
1066
1067                 mdata = data + moff;
1068                 kern_mdata = kern_data + kern_moff;
1069
1070                 mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id);
1071                 kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type,
1072                                                     &kern_mtype_id);
1073                 if (BTF_INFO_KIND(mtype->info) !=
1074                     BTF_INFO_KIND(kern_mtype->info)) {
1075                         pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n",
1076                                 map->name, mname, BTF_INFO_KIND(mtype->info),
1077                                 BTF_INFO_KIND(kern_mtype->info));
1078                         return -ENOTSUP;
1079                 }
1080
1081                 if (btf_is_ptr(mtype)) {
1082                         struct bpf_program *prog;
1083
1084                         prog = st_ops->progs[i];
1085                         if (!prog)
1086                                 continue;
1087
1088                         kern_mtype = skip_mods_and_typedefs(kern_btf,
1089                                                             kern_mtype->type,
1090                                                             &kern_mtype_id);
1091
1092                         /* mtype->type must be a func_proto which was
1093                          * guaranteed in bpf_object__collect_st_ops_relos(),
1094                          * so only check kern_mtype for func_proto here.
1095                          */
1096                         if (!btf_is_func_proto(kern_mtype)) {
1097                                 pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n",
1098                                         map->name, mname);
1099                                 return -ENOTSUP;
1100                         }
1101
1102                         prog->attach_btf_id = kern_type_id;
1103                         prog->expected_attach_type = kern_member_idx;
1104
1105                         st_ops->kern_func_off[i] = kern_data_off + kern_moff;
1106
1107                         pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n",
1108                                  map->name, mname, prog->name, moff,
1109                                  kern_moff);
1110
1111                         continue;
1112                 }
1113
1114                 msize = btf__resolve_size(btf, mtype_id);
1115                 kern_msize = btf__resolve_size(kern_btf, kern_mtype_id);
1116                 if (msize < 0 || kern_msize < 0 || msize != kern_msize) {
1117                         pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n",
1118                                 map->name, mname, (ssize_t)msize,
1119                                 (ssize_t)kern_msize);
1120                         return -ENOTSUP;
1121                 }
1122
1123                 pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n",
1124                          map->name, mname, (unsigned int)msize,
1125                          moff, kern_moff);
1126                 memcpy(kern_mdata, mdata, msize);
1127         }
1128
1129         return 0;
1130 }
1131
1132 static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj)
1133 {
1134         struct bpf_map *map;
1135         size_t i;
1136         int err;
1137
1138         for (i = 0; i < obj->nr_maps; i++) {
1139                 map = &obj->maps[i];
1140
1141                 if (!bpf_map__is_struct_ops(map))
1142                         continue;
1143
1144                 err = bpf_map__init_kern_struct_ops(map, obj->btf,
1145                                                     obj->btf_vmlinux);
1146                 if (err)
1147                         return err;
1148         }
1149
1150         return 0;
1151 }
1152
1153 static int init_struct_ops_maps(struct bpf_object *obj, const char *sec_name,
1154                                 int shndx, Elf_Data *data, __u32 map_flags)
1155 {
1156         const struct btf_type *type, *datasec;
1157         const struct btf_var_secinfo *vsi;
1158         struct bpf_struct_ops *st_ops;
1159         const char *tname, *var_name;
1160         __s32 type_id, datasec_id;
1161         const struct btf *btf;
1162         struct bpf_map *map;
1163         __u32 i;
1164
1165         if (shndx == -1)
1166                 return 0;
1167
1168         btf = obj->btf;
1169         datasec_id = btf__find_by_name_kind(btf, sec_name,
1170                                             BTF_KIND_DATASEC);
1171         if (datasec_id < 0) {
1172                 pr_warn("struct_ops init: DATASEC %s not found\n",
1173                         sec_name);
1174                 return -EINVAL;
1175         }
1176
1177         datasec = btf__type_by_id(btf, datasec_id);
1178         vsi = btf_var_secinfos(datasec);
1179         for (i = 0; i < btf_vlen(datasec); i++, vsi++) {
1180                 type = btf__type_by_id(obj->btf, vsi->type);
1181                 var_name = btf__name_by_offset(obj->btf, type->name_off);
1182
1183                 type_id = btf__resolve_type(obj->btf, vsi->type);
1184                 if (type_id < 0) {
1185                         pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n",
1186                                 vsi->type, sec_name);
1187                         return -EINVAL;
1188                 }
1189
1190                 type = btf__type_by_id(obj->btf, type_id);
1191                 tname = btf__name_by_offset(obj->btf, type->name_off);
1192                 if (!tname[0]) {
1193                         pr_warn("struct_ops init: anonymous type is not supported\n");
1194                         return -ENOTSUP;
1195                 }
1196                 if (!btf_is_struct(type)) {
1197                         pr_warn("struct_ops init: %s is not a struct\n", tname);
1198                         return -EINVAL;
1199                 }
1200
1201                 map = bpf_object__add_map(obj);
1202                 if (IS_ERR(map))
1203                         return PTR_ERR(map);
1204
1205                 map->sec_idx = shndx;
1206                 map->sec_offset = vsi->offset;
1207                 map->name = strdup(var_name);
1208                 if (!map->name)
1209                         return -ENOMEM;
1210
1211                 map->def.type = BPF_MAP_TYPE_STRUCT_OPS;
1212                 map->def.key_size = sizeof(int);
1213                 map->def.value_size = type->size;
1214                 map->def.max_entries = 1;
1215                 map->def.map_flags = map_flags;
1216
1217                 map->st_ops = calloc(1, sizeof(*map->st_ops));
1218                 if (!map->st_ops)
1219                         return -ENOMEM;
1220                 st_ops = map->st_ops;
1221                 st_ops->data = malloc(type->size);
1222                 st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs));
1223                 st_ops->kern_func_off = malloc(btf_vlen(type) *
1224                                                sizeof(*st_ops->kern_func_off));
1225                 if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off)
1226                         return -ENOMEM;
1227
1228                 if (vsi->offset + type->size > data->d_size) {
1229                         pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n",
1230                                 var_name, sec_name);
1231                         return -EINVAL;
1232                 }
1233
1234                 memcpy(st_ops->data,
1235                        data->d_buf + vsi->offset,
1236                        type->size);
1237                 st_ops->tname = tname;
1238                 st_ops->type = type;
1239                 st_ops->type_id = type_id;
1240
1241                 pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n",
1242                          tname, type_id, var_name, vsi->offset);
1243         }
1244
1245         return 0;
1246 }
1247
1248 static int bpf_object_init_struct_ops(struct bpf_object *obj)
1249 {
1250         int err;
1251
1252         err = init_struct_ops_maps(obj, STRUCT_OPS_SEC, obj->efile.st_ops_shndx,
1253                                    obj->efile.st_ops_data, 0);
1254         err = err ?: init_struct_ops_maps(obj, STRUCT_OPS_LINK_SEC,
1255                                           obj->efile.st_ops_link_shndx,
1256                                           obj->efile.st_ops_link_data,
1257                                           BPF_F_LINK);
1258         return err;
1259 }
1260
1261 static struct bpf_object *bpf_object__new(const char *path,
1262                                           const void *obj_buf,
1263                                           size_t obj_buf_sz,
1264                                           const char *obj_name)
1265 {
1266         struct bpf_object *obj;
1267         char *end;
1268
1269         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
1270         if (!obj) {
1271                 pr_warn("alloc memory failed for %s\n", path);
1272                 return ERR_PTR(-ENOMEM);
1273         }
1274
1275         strcpy(obj->path, path);
1276         if (obj_name) {
1277                 libbpf_strlcpy(obj->name, obj_name, sizeof(obj->name));
1278         } else {
1279                 /* Using basename() GNU version which doesn't modify arg. */
1280                 libbpf_strlcpy(obj->name, basename((void *)path), sizeof(obj->name));
1281                 end = strchr(obj->name, '.');
1282                 if (end)
1283                         *end = 0;
1284         }
1285
1286         obj->efile.fd = -1;
1287         /*
1288          * Caller of this function should also call
1289          * bpf_object__elf_finish() after data collection to return
1290          * obj_buf to user. If not, we should duplicate the buffer to
1291          * avoid user freeing them before elf finish.
1292          */
1293         obj->efile.obj_buf = obj_buf;
1294         obj->efile.obj_buf_sz = obj_buf_sz;
1295         obj->efile.btf_maps_shndx = -1;
1296         obj->efile.st_ops_shndx = -1;
1297         obj->efile.st_ops_link_shndx = -1;
1298         obj->kconfig_map_idx = -1;
1299
1300         obj->kern_version = get_kernel_version();
1301         obj->loaded = false;
1302
1303         return obj;
1304 }
1305
1306 static void bpf_object__elf_finish(struct bpf_object *obj)
1307 {
1308         if (!obj->efile.elf)
1309                 return;
1310
1311         elf_end(obj->efile.elf);
1312         obj->efile.elf = NULL;
1313         obj->efile.symbols = NULL;
1314         obj->efile.st_ops_data = NULL;
1315         obj->efile.st_ops_link_data = NULL;
1316
1317         zfree(&obj->efile.secs);
1318         obj->efile.sec_cnt = 0;
1319         zclose(obj->efile.fd);
1320         obj->efile.obj_buf = NULL;
1321         obj->efile.obj_buf_sz = 0;
1322 }
1323
1324 static int bpf_object__elf_init(struct bpf_object *obj)
1325 {
1326         Elf64_Ehdr *ehdr;
1327         int err = 0;
1328         Elf *elf;
1329
1330         if (obj->efile.elf) {
1331                 pr_warn("elf: init internal error\n");
1332                 return -LIBBPF_ERRNO__LIBELF;
1333         }
1334
1335         if (obj->efile.obj_buf_sz > 0) {
1336                 /* obj_buf should have been validated by bpf_object__open_mem(). */
1337                 elf = elf_memory((char *)obj->efile.obj_buf, obj->efile.obj_buf_sz);
1338         } else {
1339                 obj->efile.fd = open(obj->path, O_RDONLY | O_CLOEXEC);
1340                 if (obj->efile.fd < 0) {
1341                         char errmsg[STRERR_BUFSIZE], *cp;
1342
1343                         err = -errno;
1344                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
1345                         pr_warn("elf: failed to open %s: %s\n", obj->path, cp);
1346                         return err;
1347                 }
1348
1349                 elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL);
1350         }
1351
1352         if (!elf) {
1353                 pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1));
1354                 err = -LIBBPF_ERRNO__LIBELF;
1355                 goto errout;
1356         }
1357
1358         obj->efile.elf = elf;
1359
1360         if (elf_kind(elf) != ELF_K_ELF) {
1361                 err = -LIBBPF_ERRNO__FORMAT;
1362                 pr_warn("elf: '%s' is not a proper ELF object\n", obj->path);
1363                 goto errout;
1364         }
1365
1366         if (gelf_getclass(elf) != ELFCLASS64) {
1367                 err = -LIBBPF_ERRNO__FORMAT;
1368                 pr_warn("elf: '%s' is not a 64-bit ELF object\n", obj->path);
1369                 goto errout;
1370         }
1371
1372         obj->efile.ehdr = ehdr = elf64_getehdr(elf);
1373         if (!obj->efile.ehdr) {
1374                 pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1));
1375                 err = -LIBBPF_ERRNO__FORMAT;
1376                 goto errout;
1377         }
1378
1379         if (elf_getshdrstrndx(elf, &obj->efile.shstrndx)) {
1380                 pr_warn("elf: failed to get section names section index for %s: %s\n",
1381                         obj->path, elf_errmsg(-1));
1382                 err = -LIBBPF_ERRNO__FORMAT;
1383                 goto errout;
1384         }
1385
1386         /* ELF is corrupted/truncated, avoid calling elf_strptr. */
1387         if (!elf_rawdata(elf_getscn(elf, obj->efile.shstrndx), NULL)) {
1388                 pr_warn("elf: failed to get section names strings from %s: %s\n",
1389                         obj->path, elf_errmsg(-1));
1390                 err = -LIBBPF_ERRNO__FORMAT;
1391                 goto errout;
1392         }
1393
1394         /* Old LLVM set e_machine to EM_NONE */
1395         if (ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF)) {
1396                 pr_warn("elf: %s is not a valid eBPF object file\n", obj->path);
1397                 err = -LIBBPF_ERRNO__FORMAT;
1398                 goto errout;
1399         }
1400
1401         return 0;
1402 errout:
1403         bpf_object__elf_finish(obj);
1404         return err;
1405 }
1406
1407 static int bpf_object__check_endianness(struct bpf_object *obj)
1408 {
1409 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
1410         if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2LSB)
1411                 return 0;
1412 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
1413         if (obj->efile.ehdr->e_ident[EI_DATA] == ELFDATA2MSB)
1414                 return 0;
1415 #else
1416 # error "Unrecognized __BYTE_ORDER__"
1417 #endif
1418         pr_warn("elf: endianness mismatch in %s.\n", obj->path);
1419         return -LIBBPF_ERRNO__ENDIAN;
1420 }
1421
1422 static int
1423 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
1424 {
1425         if (!data) {
1426                 pr_warn("invalid license section in %s\n", obj->path);
1427                 return -LIBBPF_ERRNO__FORMAT;
1428         }
1429         /* libbpf_strlcpy() only copies first N - 1 bytes, so size + 1 won't
1430          * go over allowed ELF data section buffer
1431          */
1432         libbpf_strlcpy(obj->license, data, min(size + 1, sizeof(obj->license)));
1433         pr_debug("license of %s is %s\n", obj->path, obj->license);
1434         return 0;
1435 }
1436
1437 static int
1438 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
1439 {
1440         __u32 kver;
1441
1442         if (!data || size != sizeof(kver)) {
1443                 pr_warn("invalid kver section in %s\n", obj->path);
1444                 return -LIBBPF_ERRNO__FORMAT;
1445         }
1446         memcpy(&kver, data, sizeof(kver));
1447         obj->kern_version = kver;
1448         pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
1449         return 0;
1450 }
1451
1452 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
1453 {
1454         if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
1455             type == BPF_MAP_TYPE_HASH_OF_MAPS)
1456                 return true;
1457         return false;
1458 }
1459
1460 static int find_elf_sec_sz(const struct bpf_object *obj, const char *name, __u32 *size)
1461 {
1462         Elf_Data *data;
1463         Elf_Scn *scn;
1464
1465         if (!name)
1466                 return -EINVAL;
1467
1468         scn = elf_sec_by_name(obj, name);
1469         data = elf_sec_data(obj, scn);
1470         if (data) {
1471                 *size = data->d_size;
1472                 return 0; /* found it */
1473         }
1474
1475         return -ENOENT;
1476 }
1477
1478 static Elf64_Sym *find_elf_var_sym(const struct bpf_object *obj, const char *name)
1479 {
1480         Elf_Data *symbols = obj->efile.symbols;
1481         const char *sname;
1482         size_t si;
1483
1484         for (si = 0; si < symbols->d_size / sizeof(Elf64_Sym); si++) {
1485                 Elf64_Sym *sym = elf_sym_by_idx(obj, si);
1486
1487                 if (ELF64_ST_TYPE(sym->st_info) != STT_OBJECT)
1488                         continue;
1489
1490                 if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL &&
1491                     ELF64_ST_BIND(sym->st_info) != STB_WEAK)
1492                         continue;
1493
1494                 sname = elf_sym_str(obj, sym->st_name);
1495                 if (!sname) {
1496                         pr_warn("failed to get sym name string for var %s\n", name);
1497                         return ERR_PTR(-EIO);
1498                 }
1499                 if (strcmp(name, sname) == 0)
1500                         return sym;
1501         }
1502
1503         return ERR_PTR(-ENOENT);
1504 }
1505
1506 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
1507 {
1508         struct bpf_map *map;
1509         int err;
1510
1511         err = libbpf_ensure_mem((void **)&obj->maps, &obj->maps_cap,
1512                                 sizeof(*obj->maps), obj->nr_maps + 1);
1513         if (err)
1514                 return ERR_PTR(err);
1515
1516         map = &obj->maps[obj->nr_maps++];
1517         map->obj = obj;
1518         map->fd = -1;
1519         map->inner_map_fd = -1;
1520         map->autocreate = true;
1521
1522         return map;
1523 }
1524
1525 static size_t bpf_map_mmap_sz(unsigned int value_sz, unsigned int max_entries)
1526 {
1527         const long page_sz = sysconf(_SC_PAGE_SIZE);
1528         size_t map_sz;
1529
1530         map_sz = (size_t)roundup(value_sz, 8) * max_entries;
1531         map_sz = roundup(map_sz, page_sz);
1532         return map_sz;
1533 }
1534
1535 static int bpf_map_mmap_resize(struct bpf_map *map, size_t old_sz, size_t new_sz)
1536 {
1537         void *mmaped;
1538
1539         if (!map->mmaped)
1540                 return -EINVAL;
1541
1542         if (old_sz == new_sz)
1543                 return 0;
1544
1545         mmaped = mmap(NULL, new_sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1546         if (mmaped == MAP_FAILED)
1547                 return -errno;
1548
1549         memcpy(mmaped, map->mmaped, min(old_sz, new_sz));
1550         munmap(map->mmaped, old_sz);
1551         map->mmaped = mmaped;
1552         return 0;
1553 }
1554
1555 static char *internal_map_name(struct bpf_object *obj, const char *real_name)
1556 {
1557         char map_name[BPF_OBJ_NAME_LEN], *p;
1558         int pfx_len, sfx_len = max((size_t)7, strlen(real_name));
1559
1560         /* This is one of the more confusing parts of libbpf for various
1561          * reasons, some of which are historical. The original idea for naming
1562          * internal names was to include as much of BPF object name prefix as
1563          * possible, so that it can be distinguished from similar internal
1564          * maps of a different BPF object.
1565          * As an example, let's say we have bpf_object named 'my_object_name'
1566          * and internal map corresponding to '.rodata' ELF section. The final
1567          * map name advertised to user and to the kernel will be
1568          * 'my_objec.rodata', taking first 8 characters of object name and
1569          * entire 7 characters of '.rodata'.
1570          * Somewhat confusingly, if internal map ELF section name is shorter
1571          * than 7 characters, e.g., '.bss', we still reserve 7 characters
1572          * for the suffix, even though we only have 4 actual characters, and
1573          * resulting map will be called 'my_objec.bss', not even using all 15
1574          * characters allowed by the kernel. Oh well, at least the truncated
1575          * object name is somewhat consistent in this case. But if the map
1576          * name is '.kconfig', we'll still have entirety of '.kconfig' added
1577          * (8 chars) and thus will be left with only first 7 characters of the
1578          * object name ('my_obje'). Happy guessing, user, that the final map
1579          * name will be "my_obje.kconfig".
1580          * Now, with libbpf starting to support arbitrarily named .rodata.*
1581          * and .data.* data sections, it's possible that ELF section name is
1582          * longer than allowed 15 chars, so we now need to be careful to take
1583          * only up to 15 first characters of ELF name, taking no BPF object
1584          * name characters at all. So '.rodata.abracadabra' will result in
1585          * '.rodata.abracad' kernel and user-visible name.
1586          * We need to keep this convoluted logic intact for .data, .bss and
1587          * .rodata maps, but for new custom .data.custom and .rodata.custom
1588          * maps we use their ELF names as is, not prepending bpf_object name
1589          * in front. We still need to truncate them to 15 characters for the
1590          * kernel. Full name can be recovered for such maps by using DATASEC
1591          * BTF type associated with such map's value type, though.
1592          */
1593         if (sfx_len >= BPF_OBJ_NAME_LEN)
1594                 sfx_len = BPF_OBJ_NAME_LEN - 1;
1595
1596         /* if there are two or more dots in map name, it's a custom dot map */
1597         if (strchr(real_name + 1, '.') != NULL)
1598                 pfx_len = 0;
1599         else
1600                 pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, strlen(obj->name));
1601
1602         snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name,
1603                  sfx_len, real_name);
1604
1605         /* sanitise map name to characters allowed by kernel */
1606         for (p = map_name; *p && p < map_name + sizeof(map_name); p++)
1607                 if (!isalnum(*p) && *p != '_' && *p != '.')
1608                         *p = '_';
1609
1610         return strdup(map_name);
1611 }
1612
1613 static int
1614 map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map);
1615
1616 /* Internal BPF map is mmap()'able only if at least one of corresponding
1617  * DATASEC's VARs are to be exposed through BPF skeleton. I.e., it's a GLOBAL
1618  * variable and it's not marked as __hidden (which turns it into, effectively,
1619  * a STATIC variable).
1620  */
1621 static bool map_is_mmapable(struct bpf_object *obj, struct bpf_map *map)
1622 {
1623         const struct btf_type *t, *vt;
1624         struct btf_var_secinfo *vsi;
1625         int i, n;
1626
1627         if (!map->btf_value_type_id)
1628                 return false;
1629
1630         t = btf__type_by_id(obj->btf, map->btf_value_type_id);
1631         if (!btf_is_datasec(t))
1632                 return false;
1633
1634         vsi = btf_var_secinfos(t);
1635         for (i = 0, n = btf_vlen(t); i < n; i++, vsi++) {
1636                 vt = btf__type_by_id(obj->btf, vsi->type);
1637                 if (!btf_is_var(vt))
1638                         continue;
1639
1640                 if (btf_var(vt)->linkage != BTF_VAR_STATIC)
1641                         return true;
1642         }
1643
1644         return false;
1645 }
1646
1647 static int
1648 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
1649                               const char *real_name, int sec_idx, void *data, size_t data_sz)
1650 {
1651         struct bpf_map_def *def;
1652         struct bpf_map *map;
1653         size_t mmap_sz;
1654         int err;
1655
1656         map = bpf_object__add_map(obj);
1657         if (IS_ERR(map))
1658                 return PTR_ERR(map);
1659
1660         map->libbpf_type = type;
1661         map->sec_idx = sec_idx;
1662         map->sec_offset = 0;
1663         map->real_name = strdup(real_name);
1664         map->name = internal_map_name(obj, real_name);
1665         if (!map->real_name || !map->name) {
1666                 zfree(&map->real_name);
1667                 zfree(&map->name);
1668                 return -ENOMEM;
1669         }
1670
1671         def = &map->def;
1672         def->type = BPF_MAP_TYPE_ARRAY;
1673         def->key_size = sizeof(int);
1674         def->value_size = data_sz;
1675         def->max_entries = 1;
1676         def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG
1677                          ? BPF_F_RDONLY_PROG : 0;
1678
1679         /* failures are fine because of maps like .rodata.str1.1 */
1680         (void) map_fill_btf_type_info(obj, map);
1681
1682         if (map_is_mmapable(obj, map))
1683                 def->map_flags |= BPF_F_MMAPABLE;
1684
1685         pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
1686                  map->name, map->sec_idx, map->sec_offset, def->map_flags);
1687
1688         mmap_sz = bpf_map_mmap_sz(map->def.value_size, map->def.max_entries);
1689         map->mmaped = mmap(NULL, mmap_sz, PROT_READ | PROT_WRITE,
1690                            MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1691         if (map->mmaped == MAP_FAILED) {
1692                 err = -errno;
1693                 map->mmaped = NULL;
1694                 pr_warn("failed to alloc map '%s' content buffer: %d\n",
1695                         map->name, err);
1696                 zfree(&map->real_name);
1697                 zfree(&map->name);
1698                 return err;
1699         }
1700
1701         if (data)
1702                 memcpy(map->mmaped, data, data_sz);
1703
1704         pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
1705         return 0;
1706 }
1707
1708 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
1709 {
1710         struct elf_sec_desc *sec_desc;
1711         const char *sec_name;
1712         int err = 0, sec_idx;
1713
1714         /*
1715          * Populate obj->maps with libbpf internal maps.
1716          */
1717         for (sec_idx = 1; sec_idx < obj->efile.sec_cnt; sec_idx++) {
1718                 sec_desc = &obj->efile.secs[sec_idx];
1719
1720                 /* Skip recognized sections with size 0. */
1721                 if (!sec_desc->data || sec_desc->data->d_size == 0)
1722                         continue;
1723
1724                 switch (sec_desc->sec_type) {
1725                 case SEC_DATA:
1726                         sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1727                         err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
1728                                                             sec_name, sec_idx,
1729                                                             sec_desc->data->d_buf,
1730                                                             sec_desc->data->d_size);
1731                         break;
1732                 case SEC_RODATA:
1733                         obj->has_rodata = true;
1734                         sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1735                         err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
1736                                                             sec_name, sec_idx,
1737                                                             sec_desc->data->d_buf,
1738                                                             sec_desc->data->d_size);
1739                         break;
1740                 case SEC_BSS:
1741                         sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1742                         err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
1743                                                             sec_name, sec_idx,
1744                                                             NULL,
1745                                                             sec_desc->data->d_size);
1746                         break;
1747                 default:
1748                         /* skip */
1749                         break;
1750                 }
1751                 if (err)
1752                         return err;
1753         }
1754         return 0;
1755 }
1756
1757
1758 static struct extern_desc *find_extern_by_name(const struct bpf_object *obj,
1759                                                const void *name)
1760 {
1761         int i;
1762
1763         for (i = 0; i < obj->nr_extern; i++) {
1764                 if (strcmp(obj->externs[i].name, name) == 0)
1765                         return &obj->externs[i];
1766         }
1767         return NULL;
1768 }
1769
1770 static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val,
1771                               char value)
1772 {
1773         switch (ext->kcfg.type) {
1774         case KCFG_BOOL:
1775                 if (value == 'm') {
1776                         pr_warn("extern (kcfg) '%s': value '%c' implies tristate or char type\n",
1777                                 ext->name, value);
1778                         return -EINVAL;
1779                 }
1780                 *(bool *)ext_val = value == 'y' ? true : false;
1781                 break;
1782         case KCFG_TRISTATE:
1783                 if (value == 'y')
1784                         *(enum libbpf_tristate *)ext_val = TRI_YES;
1785                 else if (value == 'm')
1786                         *(enum libbpf_tristate *)ext_val = TRI_MODULE;
1787                 else /* value == 'n' */
1788                         *(enum libbpf_tristate *)ext_val = TRI_NO;
1789                 break;
1790         case KCFG_CHAR:
1791                 *(char *)ext_val = value;
1792                 break;
1793         case KCFG_UNKNOWN:
1794         case KCFG_INT:
1795         case KCFG_CHAR_ARR:
1796         default:
1797                 pr_warn("extern (kcfg) '%s': value '%c' implies bool, tristate, or char type\n",
1798                         ext->name, value);
1799                 return -EINVAL;
1800         }
1801         ext->is_set = true;
1802         return 0;
1803 }
1804
1805 static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val,
1806                               const char *value)
1807 {
1808         size_t len;
1809
1810         if (ext->kcfg.type != KCFG_CHAR_ARR) {
1811                 pr_warn("extern (kcfg) '%s': value '%s' implies char array type\n",
1812                         ext->name, value);
1813                 return -EINVAL;
1814         }
1815
1816         len = strlen(value);
1817         if (value[len - 1] != '"') {
1818                 pr_warn("extern (kcfg) '%s': invalid string config '%s'\n",
1819                         ext->name, value);
1820                 return -EINVAL;
1821         }
1822
1823         /* strip quotes */
1824         len -= 2;
1825         if (len >= ext->kcfg.sz) {
1826                 pr_warn("extern (kcfg) '%s': long string '%s' of (%zu bytes) truncated to %d bytes\n",
1827                         ext->name, value, len, ext->kcfg.sz - 1);
1828                 len = ext->kcfg.sz - 1;
1829         }
1830         memcpy(ext_val, value + 1, len);
1831         ext_val[len] = '\0';
1832         ext->is_set = true;
1833         return 0;
1834 }
1835
1836 static int parse_u64(const char *value, __u64 *res)
1837 {
1838         char *value_end;
1839         int err;
1840
1841         errno = 0;
1842         *res = strtoull(value, &value_end, 0);
1843         if (errno) {
1844                 err = -errno;
1845                 pr_warn("failed to parse '%s' as integer: %d\n", value, err);
1846                 return err;
1847         }
1848         if (*value_end) {
1849                 pr_warn("failed to parse '%s' as integer completely\n", value);
1850                 return -EINVAL;
1851         }
1852         return 0;
1853 }
1854
1855 static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v)
1856 {
1857         int bit_sz = ext->kcfg.sz * 8;
1858
1859         if (ext->kcfg.sz == 8)
1860                 return true;
1861
1862         /* Validate that value stored in u64 fits in integer of `ext->sz`
1863          * bytes size without any loss of information. If the target integer
1864          * is signed, we rely on the following limits of integer type of
1865          * Y bits and subsequent transformation:
1866          *
1867          *     -2^(Y-1) <= X           <= 2^(Y-1) - 1
1868          *            0 <= X + 2^(Y-1) <= 2^Y - 1
1869          *            0 <= X + 2^(Y-1) <  2^Y
1870          *
1871          *  For unsigned target integer, check that all the (64 - Y) bits are
1872          *  zero.
1873          */
1874         if (ext->kcfg.is_signed)
1875                 return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz);
1876         else
1877                 return (v >> bit_sz) == 0;
1878 }
1879
1880 static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val,
1881                               __u64 value)
1882 {
1883         if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR &&
1884             ext->kcfg.type != KCFG_BOOL) {
1885                 pr_warn("extern (kcfg) '%s': value '%llu' implies integer, char, or boolean type\n",
1886                         ext->name, (unsigned long long)value);
1887                 return -EINVAL;
1888         }
1889         if (ext->kcfg.type == KCFG_BOOL && value > 1) {
1890                 pr_warn("extern (kcfg) '%s': value '%llu' isn't boolean compatible\n",
1891                         ext->name, (unsigned long long)value);
1892                 return -EINVAL;
1893
1894         }
1895         if (!is_kcfg_value_in_range(ext, value)) {
1896                 pr_warn("extern (kcfg) '%s': value '%llu' doesn't fit in %d bytes\n",
1897                         ext->name, (unsigned long long)value, ext->kcfg.sz);
1898                 return -ERANGE;
1899         }
1900         switch (ext->kcfg.sz) {
1901         case 1:
1902                 *(__u8 *)ext_val = value;
1903                 break;
1904         case 2:
1905                 *(__u16 *)ext_val = value;
1906                 break;
1907         case 4:
1908                 *(__u32 *)ext_val = value;
1909                 break;
1910         case 8:
1911                 *(__u64 *)ext_val = value;
1912                 break;
1913         default:
1914                 return -EINVAL;
1915         }
1916         ext->is_set = true;
1917         return 0;
1918 }
1919
1920 static int bpf_object__process_kconfig_line(struct bpf_object *obj,
1921                                             char *buf, void *data)
1922 {
1923         struct extern_desc *ext;
1924         char *sep, *value;
1925         int len, err = 0;
1926         void *ext_val;
1927         __u64 num;
1928
1929         if (!str_has_pfx(buf, "CONFIG_"))
1930                 return 0;
1931
1932         sep = strchr(buf, '=');
1933         if (!sep) {
1934                 pr_warn("failed to parse '%s': no separator\n", buf);
1935                 return -EINVAL;
1936         }
1937
1938         /* Trim ending '\n' */
1939         len = strlen(buf);
1940         if (buf[len - 1] == '\n')
1941                 buf[len - 1] = '\0';
1942         /* Split on '=' and ensure that a value is present. */
1943         *sep = '\0';
1944         if (!sep[1]) {
1945                 *sep = '=';
1946                 pr_warn("failed to parse '%s': no value\n", buf);
1947                 return -EINVAL;
1948         }
1949
1950         ext = find_extern_by_name(obj, buf);
1951         if (!ext || ext->is_set)
1952                 return 0;
1953
1954         ext_val = data + ext->kcfg.data_off;
1955         value = sep + 1;
1956
1957         switch (*value) {
1958         case 'y': case 'n': case 'm':
1959                 err = set_kcfg_value_tri(ext, ext_val, *value);
1960                 break;
1961         case '"':
1962                 err = set_kcfg_value_str(ext, ext_val, value);
1963                 break;
1964         default:
1965                 /* assume integer */
1966                 err = parse_u64(value, &num);
1967                 if (err) {
1968                         pr_warn("extern (kcfg) '%s': value '%s' isn't a valid integer\n", ext->name, value);
1969                         return err;
1970                 }
1971                 if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) {
1972                         pr_warn("extern (kcfg) '%s': value '%s' implies integer type\n", ext->name, value);
1973                         return -EINVAL;
1974                 }
1975                 err = set_kcfg_value_num(ext, ext_val, num);
1976                 break;
1977         }
1978         if (err)
1979                 return err;
1980         pr_debug("extern (kcfg) '%s': set to %s\n", ext->name, value);
1981         return 0;
1982 }
1983
1984 static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data)
1985 {
1986         char buf[PATH_MAX];
1987         struct utsname uts;
1988         int len, err = 0;
1989         gzFile file;
1990
1991         uname(&uts);
1992         len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release);
1993         if (len < 0)
1994                 return -EINVAL;
1995         else if (len >= PATH_MAX)
1996                 return -ENAMETOOLONG;
1997
1998         /* gzopen also accepts uncompressed files. */
1999         file = gzopen(buf, "re");
2000         if (!file)
2001                 file = gzopen("/proc/config.gz", "re");
2002
2003         if (!file) {
2004                 pr_warn("failed to open system Kconfig\n");
2005                 return -ENOENT;
2006         }
2007
2008         while (gzgets(file, buf, sizeof(buf))) {
2009                 err = bpf_object__process_kconfig_line(obj, buf, data);
2010                 if (err) {
2011                         pr_warn("error parsing system Kconfig line '%s': %d\n",
2012                                 buf, err);
2013                         goto out;
2014                 }
2015         }
2016
2017 out:
2018         gzclose(file);
2019         return err;
2020 }
2021
2022 static int bpf_object__read_kconfig_mem(struct bpf_object *obj,
2023                                         const char *config, void *data)
2024 {
2025         char buf[PATH_MAX];
2026         int err = 0;
2027         FILE *file;
2028
2029         file = fmemopen((void *)config, strlen(config), "r");
2030         if (!file) {
2031                 err = -errno;
2032                 pr_warn("failed to open in-memory Kconfig: %d\n", err);
2033                 return err;
2034         }
2035
2036         while (fgets(buf, sizeof(buf), file)) {
2037                 err = bpf_object__process_kconfig_line(obj, buf, data);
2038                 if (err) {
2039                         pr_warn("error parsing in-memory Kconfig line '%s': %d\n",
2040                                 buf, err);
2041                         break;
2042                 }
2043         }
2044
2045         fclose(file);
2046         return err;
2047 }
2048
2049 static int bpf_object__init_kconfig_map(struct bpf_object *obj)
2050 {
2051         struct extern_desc *last_ext = NULL, *ext;
2052         size_t map_sz;
2053         int i, err;
2054
2055         for (i = 0; i < obj->nr_extern; i++) {
2056                 ext = &obj->externs[i];
2057                 if (ext->type == EXT_KCFG)
2058                         last_ext = ext;
2059         }
2060
2061         if (!last_ext)
2062                 return 0;
2063
2064         map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz;
2065         err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG,
2066                                             ".kconfig", obj->efile.symbols_shndx,
2067                                             NULL, map_sz);
2068         if (err)
2069                 return err;
2070
2071         obj->kconfig_map_idx = obj->nr_maps - 1;
2072
2073         return 0;
2074 }
2075
2076 const struct btf_type *
2077 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
2078 {
2079         const struct btf_type *t = btf__type_by_id(btf, id);
2080
2081         if (res_id)
2082                 *res_id = id;
2083
2084         while (btf_is_mod(t) || btf_is_typedef(t)) {
2085                 if (res_id)
2086                         *res_id = t->type;
2087                 t = btf__type_by_id(btf, t->type);
2088         }
2089
2090         return t;
2091 }
2092
2093 static const struct btf_type *
2094 resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id)
2095 {
2096         const struct btf_type *t;
2097
2098         t = skip_mods_and_typedefs(btf, id, NULL);
2099         if (!btf_is_ptr(t))
2100                 return NULL;
2101
2102         t = skip_mods_and_typedefs(btf, t->type, res_id);
2103
2104         return btf_is_func_proto(t) ? t : NULL;
2105 }
2106
2107 static const char *__btf_kind_str(__u16 kind)
2108 {
2109         switch (kind) {
2110         case BTF_KIND_UNKN: return "void";
2111         case BTF_KIND_INT: return "int";
2112         case BTF_KIND_PTR: return "ptr";
2113         case BTF_KIND_ARRAY: return "array";
2114         case BTF_KIND_STRUCT: return "struct";
2115         case BTF_KIND_UNION: return "union";
2116         case BTF_KIND_ENUM: return "enum";
2117         case BTF_KIND_FWD: return "fwd";
2118         case BTF_KIND_TYPEDEF: return "typedef";
2119         case BTF_KIND_VOLATILE: return "volatile";
2120         case BTF_KIND_CONST: return "const";
2121         case BTF_KIND_RESTRICT: return "restrict";
2122         case BTF_KIND_FUNC: return "func";
2123         case BTF_KIND_FUNC_PROTO: return "func_proto";
2124         case BTF_KIND_VAR: return "var";
2125         case BTF_KIND_DATASEC: return "datasec";
2126         case BTF_KIND_FLOAT: return "float";
2127         case BTF_KIND_DECL_TAG: return "decl_tag";
2128         case BTF_KIND_TYPE_TAG: return "type_tag";
2129         case BTF_KIND_ENUM64: return "enum64";
2130         default: return "unknown";
2131         }
2132 }
2133
2134 const char *btf_kind_str(const struct btf_type *t)
2135 {
2136         return __btf_kind_str(btf_kind(t));
2137 }
2138
2139 /*
2140  * Fetch integer attribute of BTF map definition. Such attributes are
2141  * represented using a pointer to an array, in which dimensionality of array
2142  * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
2143  * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
2144  * type definition, while using only sizeof(void *) space in ELF data section.
2145  */
2146 static bool get_map_field_int(const char *map_name, const struct btf *btf,
2147                               const struct btf_member *m, __u32 *res)
2148 {
2149         const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
2150         const char *name = btf__name_by_offset(btf, m->name_off);
2151         const struct btf_array *arr_info;
2152         const struct btf_type *arr_t;
2153
2154         if (!btf_is_ptr(t)) {
2155                 pr_warn("map '%s': attr '%s': expected PTR, got %s.\n",
2156                         map_name, name, btf_kind_str(t));
2157                 return false;
2158         }
2159
2160         arr_t = btf__type_by_id(btf, t->type);
2161         if (!arr_t) {
2162                 pr_warn("map '%s': attr '%s': type [%u] not found.\n",
2163                         map_name, name, t->type);
2164                 return false;
2165         }
2166         if (!btf_is_array(arr_t)) {
2167                 pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n",
2168                         map_name, name, btf_kind_str(arr_t));
2169                 return false;
2170         }
2171         arr_info = btf_array(arr_t);
2172         *res = arr_info->nelems;
2173         return true;
2174 }
2175
2176 static int pathname_concat(char *buf, size_t buf_sz, const char *path, const char *name)
2177 {
2178         int len;
2179
2180         len = snprintf(buf, buf_sz, "%s/%s", path, name);
2181         if (len < 0)
2182                 return -EINVAL;
2183         if (len >= buf_sz)
2184                 return -ENAMETOOLONG;
2185
2186         return 0;
2187 }
2188
2189 static int build_map_pin_path(struct bpf_map *map, const char *path)
2190 {
2191         char buf[PATH_MAX];
2192         int err;
2193
2194         if (!path)
2195                 path = "/sys/fs/bpf";
2196
2197         err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
2198         if (err)
2199                 return err;
2200
2201         return bpf_map__set_pin_path(map, buf);
2202 }
2203
2204 /* should match definition in bpf_helpers.h */
2205 enum libbpf_pin_type {
2206         LIBBPF_PIN_NONE,
2207         /* PIN_BY_NAME: pin maps by name (in /sys/fs/bpf by default) */
2208         LIBBPF_PIN_BY_NAME,
2209 };
2210
2211 int parse_btf_map_def(const char *map_name, struct btf *btf,
2212                       const struct btf_type *def_t, bool strict,
2213                       struct btf_map_def *map_def, struct btf_map_def *inner_def)
2214 {
2215         const struct btf_type *t;
2216         const struct btf_member *m;
2217         bool is_inner = inner_def == NULL;
2218         int vlen, i;
2219
2220         vlen = btf_vlen(def_t);
2221         m = btf_members(def_t);
2222         for (i = 0; i < vlen; i++, m++) {
2223                 const char *name = btf__name_by_offset(btf, m->name_off);
2224
2225                 if (!name) {
2226                         pr_warn("map '%s': invalid field #%d.\n", map_name, i);
2227                         return -EINVAL;
2228                 }
2229                 if (strcmp(name, "type") == 0) {
2230                         if (!get_map_field_int(map_name, btf, m, &map_def->map_type))
2231                                 return -EINVAL;
2232                         map_def->parts |= MAP_DEF_MAP_TYPE;
2233                 } else if (strcmp(name, "max_entries") == 0) {
2234                         if (!get_map_field_int(map_name, btf, m, &map_def->max_entries))
2235                                 return -EINVAL;
2236                         map_def->parts |= MAP_DEF_MAX_ENTRIES;
2237                 } else if (strcmp(name, "map_flags") == 0) {
2238                         if (!get_map_field_int(map_name, btf, m, &map_def->map_flags))
2239                                 return -EINVAL;
2240                         map_def->parts |= MAP_DEF_MAP_FLAGS;
2241                 } else if (strcmp(name, "numa_node") == 0) {
2242                         if (!get_map_field_int(map_name, btf, m, &map_def->numa_node))
2243                                 return -EINVAL;
2244                         map_def->parts |= MAP_DEF_NUMA_NODE;
2245                 } else if (strcmp(name, "key_size") == 0) {
2246                         __u32 sz;
2247
2248                         if (!get_map_field_int(map_name, btf, m, &sz))
2249                                 return -EINVAL;
2250                         if (map_def->key_size && map_def->key_size != sz) {
2251                                 pr_warn("map '%s': conflicting key size %u != %u.\n",
2252                                         map_name, map_def->key_size, sz);
2253                                 return -EINVAL;
2254                         }
2255                         map_def->key_size = sz;
2256                         map_def->parts |= MAP_DEF_KEY_SIZE;
2257                 } else if (strcmp(name, "key") == 0) {
2258                         __s64 sz;
2259
2260                         t = btf__type_by_id(btf, m->type);
2261                         if (!t) {
2262                                 pr_warn("map '%s': key type [%d] not found.\n",
2263                                         map_name, m->type);
2264                                 return -EINVAL;
2265                         }
2266                         if (!btf_is_ptr(t)) {
2267                                 pr_warn("map '%s': key spec is not PTR: %s.\n",
2268                                         map_name, btf_kind_str(t));
2269                                 return -EINVAL;
2270                         }
2271                         sz = btf__resolve_size(btf, t->type);
2272                         if (sz < 0) {
2273                                 pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n",
2274                                         map_name, t->type, (ssize_t)sz);
2275                                 return sz;
2276                         }
2277                         if (map_def->key_size && map_def->key_size != sz) {
2278                                 pr_warn("map '%s': conflicting key size %u != %zd.\n",
2279                                         map_name, map_def->key_size, (ssize_t)sz);
2280                                 return -EINVAL;
2281                         }
2282                         map_def->key_size = sz;
2283                         map_def->key_type_id = t->type;
2284                         map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE;
2285                 } else if (strcmp(name, "value_size") == 0) {
2286                         __u32 sz;
2287
2288                         if (!get_map_field_int(map_name, btf, m, &sz))
2289                                 return -EINVAL;
2290                         if (map_def->value_size && map_def->value_size != sz) {
2291                                 pr_warn("map '%s': conflicting value size %u != %u.\n",
2292                                         map_name, map_def->value_size, sz);
2293                                 return -EINVAL;
2294                         }
2295                         map_def->value_size = sz;
2296                         map_def->parts |= MAP_DEF_VALUE_SIZE;
2297                 } else if (strcmp(name, "value") == 0) {
2298                         __s64 sz;
2299
2300                         t = btf__type_by_id(btf, m->type);
2301                         if (!t) {
2302                                 pr_warn("map '%s': value type [%d] not found.\n",
2303                                         map_name, m->type);
2304                                 return -EINVAL;
2305                         }
2306                         if (!btf_is_ptr(t)) {
2307                                 pr_warn("map '%s': value spec is not PTR: %s.\n",
2308                                         map_name, btf_kind_str(t));
2309                                 return -EINVAL;
2310                         }
2311                         sz = btf__resolve_size(btf, t->type);
2312                         if (sz < 0) {
2313                                 pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n",
2314                                         map_name, t->type, (ssize_t)sz);
2315                                 return sz;
2316                         }
2317                         if (map_def->value_size && map_def->value_size != sz) {
2318                                 pr_warn("map '%s': conflicting value size %u != %zd.\n",
2319                                         map_name, map_def->value_size, (ssize_t)sz);
2320                                 return -EINVAL;
2321                         }
2322                         map_def->value_size = sz;
2323                         map_def->value_type_id = t->type;
2324                         map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE;
2325                 }
2326                 else if (strcmp(name, "values") == 0) {
2327                         bool is_map_in_map = bpf_map_type__is_map_in_map(map_def->map_type);
2328                         bool is_prog_array = map_def->map_type == BPF_MAP_TYPE_PROG_ARRAY;
2329                         const char *desc = is_map_in_map ? "map-in-map inner" : "prog-array value";
2330                         char inner_map_name[128];
2331                         int err;
2332
2333                         if (is_inner) {
2334                                 pr_warn("map '%s': multi-level inner maps not supported.\n",
2335                                         map_name);
2336                                 return -ENOTSUP;
2337                         }
2338                         if (i != vlen - 1) {
2339                                 pr_warn("map '%s': '%s' member should be last.\n",
2340                                         map_name, name);
2341                                 return -EINVAL;
2342                         }
2343                         if (!is_map_in_map && !is_prog_array) {
2344                                 pr_warn("map '%s': should be map-in-map or prog-array.\n",
2345                                         map_name);
2346                                 return -ENOTSUP;
2347                         }
2348                         if (map_def->value_size && map_def->value_size != 4) {
2349                                 pr_warn("map '%s': conflicting value size %u != 4.\n",
2350                                         map_name, map_def->value_size);
2351                                 return -EINVAL;
2352                         }
2353                         map_def->value_size = 4;
2354                         t = btf__type_by_id(btf, m->type);
2355                         if (!t) {
2356                                 pr_warn("map '%s': %s type [%d] not found.\n",
2357                                         map_name, desc, m->type);
2358                                 return -EINVAL;
2359                         }
2360                         if (!btf_is_array(t) || btf_array(t)->nelems) {
2361                                 pr_warn("map '%s': %s spec is not a zero-sized array.\n",
2362                                         map_name, desc);
2363                                 return -EINVAL;
2364                         }
2365                         t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL);
2366                         if (!btf_is_ptr(t)) {
2367                                 pr_warn("map '%s': %s def is of unexpected kind %s.\n",
2368                                         map_name, desc, btf_kind_str(t));
2369                                 return -EINVAL;
2370                         }
2371                         t = skip_mods_and_typedefs(btf, t->type, NULL);
2372                         if (is_prog_array) {
2373                                 if (!btf_is_func_proto(t)) {
2374                                         pr_warn("map '%s': prog-array value def is of unexpected kind %s.\n",
2375                                                 map_name, btf_kind_str(t));
2376                                         return -EINVAL;
2377                                 }
2378                                 continue;
2379                         }
2380                         if (!btf_is_struct(t)) {
2381                                 pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n",
2382                                         map_name, btf_kind_str(t));
2383                                 return -EINVAL;
2384                         }
2385
2386                         snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name);
2387                         err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL);
2388                         if (err)
2389                                 return err;
2390
2391                         map_def->parts |= MAP_DEF_INNER_MAP;
2392                 } else if (strcmp(name, "pinning") == 0) {
2393                         __u32 val;
2394
2395                         if (is_inner) {
2396                                 pr_warn("map '%s': inner def can't be pinned.\n", map_name);
2397                                 return -EINVAL;
2398                         }
2399                         if (!get_map_field_int(map_name, btf, m, &val))
2400                                 return -EINVAL;
2401                         if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) {
2402                                 pr_warn("map '%s': invalid pinning value %u.\n",
2403                                         map_name, val);
2404                                 return -EINVAL;
2405                         }
2406                         map_def->pinning = val;
2407                         map_def->parts |= MAP_DEF_PINNING;
2408                 } else if (strcmp(name, "map_extra") == 0) {
2409                         __u32 map_extra;
2410
2411                         if (!get_map_field_int(map_name, btf, m, &map_extra))
2412                                 return -EINVAL;
2413                         map_def->map_extra = map_extra;
2414                         map_def->parts |= MAP_DEF_MAP_EXTRA;
2415                 } else {
2416                         if (strict) {
2417                                 pr_warn("map '%s': unknown field '%s'.\n", map_name, name);
2418                                 return -ENOTSUP;
2419                         }
2420                         pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name);
2421                 }
2422         }
2423
2424         if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) {
2425                 pr_warn("map '%s': map type isn't specified.\n", map_name);
2426                 return -EINVAL;
2427         }
2428
2429         return 0;
2430 }
2431
2432 static size_t adjust_ringbuf_sz(size_t sz)
2433 {
2434         __u32 page_sz = sysconf(_SC_PAGE_SIZE);
2435         __u32 mul;
2436
2437         /* if user forgot to set any size, make sure they see error */
2438         if (sz == 0)
2439                 return 0;
2440         /* Kernel expects BPF_MAP_TYPE_RINGBUF's max_entries to be
2441          * a power-of-2 multiple of kernel's page size. If user diligently
2442          * satisified these conditions, pass the size through.
2443          */
2444         if ((sz % page_sz) == 0 && is_pow_of_2(sz / page_sz))
2445                 return sz;
2446
2447         /* Otherwise find closest (page_sz * power_of_2) product bigger than
2448          * user-set size to satisfy both user size request and kernel
2449          * requirements and substitute correct max_entries for map creation.
2450          */
2451         for (mul = 1; mul <= UINT_MAX / page_sz; mul <<= 1) {
2452                 if (mul * page_sz > sz)
2453                         return mul * page_sz;
2454         }
2455
2456         /* if it's impossible to satisfy the conditions (i.e., user size is
2457          * very close to UINT_MAX but is not a power-of-2 multiple of
2458          * page_size) then just return original size and let kernel reject it
2459          */
2460         return sz;
2461 }
2462
2463 static bool map_is_ringbuf(const struct bpf_map *map)
2464 {
2465         return map->def.type == BPF_MAP_TYPE_RINGBUF ||
2466                map->def.type == BPF_MAP_TYPE_USER_RINGBUF;
2467 }
2468
2469 static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def)
2470 {
2471         map->def.type = def->map_type;
2472         map->def.key_size = def->key_size;
2473         map->def.value_size = def->value_size;
2474         map->def.max_entries = def->max_entries;
2475         map->def.map_flags = def->map_flags;
2476         map->map_extra = def->map_extra;
2477
2478         map->numa_node = def->numa_node;
2479         map->btf_key_type_id = def->key_type_id;
2480         map->btf_value_type_id = def->value_type_id;
2481
2482         /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */
2483         if (map_is_ringbuf(map))
2484                 map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries);
2485
2486         if (def->parts & MAP_DEF_MAP_TYPE)
2487                 pr_debug("map '%s': found type = %u.\n", map->name, def->map_type);
2488
2489         if (def->parts & MAP_DEF_KEY_TYPE)
2490                 pr_debug("map '%s': found key [%u], sz = %u.\n",
2491                          map->name, def->key_type_id, def->key_size);
2492         else if (def->parts & MAP_DEF_KEY_SIZE)
2493                 pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size);
2494
2495         if (def->parts & MAP_DEF_VALUE_TYPE)
2496                 pr_debug("map '%s': found value [%u], sz = %u.\n",
2497                          map->name, def->value_type_id, def->value_size);
2498         else if (def->parts & MAP_DEF_VALUE_SIZE)
2499                 pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size);
2500
2501         if (def->parts & MAP_DEF_MAX_ENTRIES)
2502                 pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries);
2503         if (def->parts & MAP_DEF_MAP_FLAGS)
2504                 pr_debug("map '%s': found map_flags = 0x%x.\n", map->name, def->map_flags);
2505         if (def->parts & MAP_DEF_MAP_EXTRA)
2506                 pr_debug("map '%s': found map_extra = 0x%llx.\n", map->name,
2507                          (unsigned long long)def->map_extra);
2508         if (def->parts & MAP_DEF_PINNING)
2509                 pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning);
2510         if (def->parts & MAP_DEF_NUMA_NODE)
2511                 pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node);
2512
2513         if (def->parts & MAP_DEF_INNER_MAP)
2514                 pr_debug("map '%s': found inner map definition.\n", map->name);
2515 }
2516
2517 static const char *btf_var_linkage_str(__u32 linkage)
2518 {
2519         switch (linkage) {
2520         case BTF_VAR_STATIC: return "static";
2521         case BTF_VAR_GLOBAL_ALLOCATED: return "global";
2522         case BTF_VAR_GLOBAL_EXTERN: return "extern";
2523         default: return "unknown";
2524         }
2525 }
2526
2527 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
2528                                          const struct btf_type *sec,
2529                                          int var_idx, int sec_idx,
2530                                          const Elf_Data *data, bool strict,
2531                                          const char *pin_root_path)
2532 {
2533         struct btf_map_def map_def = {}, inner_def = {};
2534         const struct btf_type *var, *def;
2535         const struct btf_var_secinfo *vi;
2536         const struct btf_var *var_extra;
2537         const char *map_name;
2538         struct bpf_map *map;
2539         int err;
2540
2541         vi = btf_var_secinfos(sec) + var_idx;
2542         var = btf__type_by_id(obj->btf, vi->type);
2543         var_extra = btf_var(var);
2544         map_name = btf__name_by_offset(obj->btf, var->name_off);
2545
2546         if (map_name == NULL || map_name[0] == '\0') {
2547                 pr_warn("map #%d: empty name.\n", var_idx);
2548                 return -EINVAL;
2549         }
2550         if ((__u64)vi->offset + vi->size > data->d_size) {
2551                 pr_warn("map '%s' BTF data is corrupted.\n", map_name);
2552                 return -EINVAL;
2553         }
2554         if (!btf_is_var(var)) {
2555                 pr_warn("map '%s': unexpected var kind %s.\n",
2556                         map_name, btf_kind_str(var));
2557                 return -EINVAL;
2558         }
2559         if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) {
2560                 pr_warn("map '%s': unsupported map linkage %s.\n",
2561                         map_name, btf_var_linkage_str(var_extra->linkage));
2562                 return -EOPNOTSUPP;
2563         }
2564
2565         def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
2566         if (!btf_is_struct(def)) {
2567                 pr_warn("map '%s': unexpected def kind %s.\n",
2568                         map_name, btf_kind_str(var));
2569                 return -EINVAL;
2570         }
2571         if (def->size > vi->size) {
2572                 pr_warn("map '%s': invalid def size.\n", map_name);
2573                 return -EINVAL;
2574         }
2575
2576         map = bpf_object__add_map(obj);
2577         if (IS_ERR(map))
2578                 return PTR_ERR(map);
2579         map->name = strdup(map_name);
2580         if (!map->name) {
2581                 pr_warn("map '%s': failed to alloc map name.\n", map_name);
2582                 return -ENOMEM;
2583         }
2584         map->libbpf_type = LIBBPF_MAP_UNSPEC;
2585         map->def.type = BPF_MAP_TYPE_UNSPEC;
2586         map->sec_idx = sec_idx;
2587         map->sec_offset = vi->offset;
2588         map->btf_var_idx = var_idx;
2589         pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
2590                  map_name, map->sec_idx, map->sec_offset);
2591
2592         err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def);
2593         if (err)
2594                 return err;
2595
2596         fill_map_from_def(map, &map_def);
2597
2598         if (map_def.pinning == LIBBPF_PIN_BY_NAME) {
2599                 err = build_map_pin_path(map, pin_root_path);
2600                 if (err) {
2601                         pr_warn("map '%s': couldn't build pin path.\n", map->name);
2602                         return err;
2603                 }
2604         }
2605
2606         if (map_def.parts & MAP_DEF_INNER_MAP) {
2607                 map->inner_map = calloc(1, sizeof(*map->inner_map));
2608                 if (!map->inner_map)
2609                         return -ENOMEM;
2610                 map->inner_map->fd = -1;
2611                 map->inner_map->sec_idx = sec_idx;
2612                 map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1);
2613                 if (!map->inner_map->name)
2614                         return -ENOMEM;
2615                 sprintf(map->inner_map->name, "%s.inner", map_name);
2616
2617                 fill_map_from_def(map->inner_map, &inner_def);
2618         }
2619
2620         err = map_fill_btf_type_info(obj, map);
2621         if (err)
2622                 return err;
2623
2624         return 0;
2625 }
2626
2627 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
2628                                           const char *pin_root_path)
2629 {
2630         const struct btf_type *sec = NULL;
2631         int nr_types, i, vlen, err;
2632         const struct btf_type *t;
2633         const char *name;
2634         Elf_Data *data;
2635         Elf_Scn *scn;
2636
2637         if (obj->efile.btf_maps_shndx < 0)
2638                 return 0;
2639
2640         scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx);
2641         data = elf_sec_data(obj, scn);
2642         if (!scn || !data) {
2643                 pr_warn("elf: failed to get %s map definitions for %s\n",
2644                         MAPS_ELF_SEC, obj->path);
2645                 return -EINVAL;
2646         }
2647
2648         nr_types = btf__type_cnt(obj->btf);
2649         for (i = 1; i < nr_types; i++) {
2650                 t = btf__type_by_id(obj->btf, i);
2651                 if (!btf_is_datasec(t))
2652                         continue;
2653                 name = btf__name_by_offset(obj->btf, t->name_off);
2654                 if (strcmp(name, MAPS_ELF_SEC) == 0) {
2655                         sec = t;
2656                         obj->efile.btf_maps_sec_btf_id = i;
2657                         break;
2658                 }
2659         }
2660
2661         if (!sec) {
2662                 pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
2663                 return -ENOENT;
2664         }
2665
2666         vlen = btf_vlen(sec);
2667         for (i = 0; i < vlen; i++) {
2668                 err = bpf_object__init_user_btf_map(obj, sec, i,
2669                                                     obj->efile.btf_maps_shndx,
2670                                                     data, strict,
2671                                                     pin_root_path);
2672                 if (err)
2673                         return err;
2674         }
2675
2676         return 0;
2677 }
2678
2679 static int bpf_object__init_maps(struct bpf_object *obj,
2680                                  const struct bpf_object_open_opts *opts)
2681 {
2682         const char *pin_root_path;
2683         bool strict;
2684         int err = 0;
2685
2686         strict = !OPTS_GET(opts, relaxed_maps, false);
2687         pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
2688
2689         err = bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
2690         err = err ?: bpf_object__init_global_data_maps(obj);
2691         err = err ?: bpf_object__init_kconfig_map(obj);
2692         err = err ?: bpf_object_init_struct_ops(obj);
2693
2694         return err;
2695 }
2696
2697 static bool section_have_execinstr(struct bpf_object *obj, int idx)
2698 {
2699         Elf64_Shdr *sh;
2700
2701         sh = elf_sec_hdr(obj, elf_sec_by_idx(obj, idx));
2702         if (!sh)
2703                 return false;
2704
2705         return sh->sh_flags & SHF_EXECINSTR;
2706 }
2707
2708 static bool btf_needs_sanitization(struct bpf_object *obj)
2709 {
2710         bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
2711         bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
2712         bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
2713         bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
2714         bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
2715         bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
2716         bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64);
2717
2718         return !has_func || !has_datasec || !has_func_global || !has_float ||
2719                !has_decl_tag || !has_type_tag || !has_enum64;
2720 }
2721
2722 static int bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf)
2723 {
2724         bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
2725         bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
2726         bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
2727         bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
2728         bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
2729         bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
2730         bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64);
2731         int enum64_placeholder_id = 0;
2732         struct btf_type *t;
2733         int i, j, vlen;
2734
2735         for (i = 1; i < btf__type_cnt(btf); i++) {
2736                 t = (struct btf_type *)btf__type_by_id(btf, i);
2737
2738                 if ((!has_datasec && btf_is_var(t)) || (!has_decl_tag && btf_is_decl_tag(t))) {
2739                         /* replace VAR/DECL_TAG with INT */
2740                         t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
2741                         /*
2742                          * using size = 1 is the safest choice, 4 will be too
2743                          * big and cause kernel BTF validation failure if
2744                          * original variable took less than 4 bytes
2745                          */
2746                         t->size = 1;
2747                         *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
2748                 } else if (!has_datasec && btf_is_datasec(t)) {
2749                         /* replace DATASEC with STRUCT */
2750                         const struct btf_var_secinfo *v = btf_var_secinfos(t);
2751                         struct btf_member *m = btf_members(t);
2752                         struct btf_type *vt;
2753                         char *name;
2754
2755                         name = (char *)btf__name_by_offset(btf, t->name_off);
2756                         while (*name) {
2757                                 if (*name == '.')
2758                                         *name = '_';
2759                                 name++;
2760                         }
2761
2762                         vlen = btf_vlen(t);
2763                         t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
2764                         for (j = 0; j < vlen; j++, v++, m++) {
2765                                 /* order of field assignments is important */
2766                                 m->offset = v->offset * 8;
2767                                 m->type = v->type;
2768                                 /* preserve variable name as member name */
2769                                 vt = (void *)btf__type_by_id(btf, v->type);
2770                                 m->name_off = vt->name_off;
2771                         }
2772                 } else if (!has_func && btf_is_func_proto(t)) {
2773                         /* replace FUNC_PROTO with ENUM */
2774                         vlen = btf_vlen(t);
2775                         t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
2776                         t->size = sizeof(__u32); /* kernel enforced */
2777                 } else if (!has_func && btf_is_func(t)) {
2778                         /* replace FUNC with TYPEDEF */
2779                         t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
2780                 } else if (!has_func_global && btf_is_func(t)) {
2781                         /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */
2782                         t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0);
2783                 } else if (!has_float && btf_is_float(t)) {
2784                         /* replace FLOAT with an equally-sized empty STRUCT;
2785                          * since C compilers do not accept e.g. "float" as a
2786                          * valid struct name, make it anonymous
2787                          */
2788                         t->name_off = 0;
2789                         t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0);
2790                 } else if (!has_type_tag && btf_is_type_tag(t)) {
2791                         /* replace TYPE_TAG with a CONST */
2792                         t->name_off = 0;
2793                         t->info = BTF_INFO_ENC(BTF_KIND_CONST, 0, 0);
2794                 } else if (!has_enum64 && btf_is_enum(t)) {
2795                         /* clear the kflag */
2796                         t->info = btf_type_info(btf_kind(t), btf_vlen(t), false);
2797                 } else if (!has_enum64 && btf_is_enum64(t)) {
2798                         /* replace ENUM64 with a union */
2799                         struct btf_member *m;
2800
2801                         if (enum64_placeholder_id == 0) {
2802                                 enum64_placeholder_id = btf__add_int(btf, "enum64_placeholder", 1, 0);
2803                                 if (enum64_placeholder_id < 0)
2804                                         return enum64_placeholder_id;
2805
2806                                 t = (struct btf_type *)btf__type_by_id(btf, i);
2807                         }
2808
2809                         m = btf_members(t);
2810                         vlen = btf_vlen(t);
2811                         t->info = BTF_INFO_ENC(BTF_KIND_UNION, 0, vlen);
2812                         for (j = 0; j < vlen; j++, m++) {
2813                                 m->type = enum64_placeholder_id;
2814                                 m->offset = 0;
2815                         }
2816                 }
2817         }
2818
2819         return 0;
2820 }
2821
2822 static bool libbpf_needs_btf(const struct bpf_object *obj)
2823 {
2824         return obj->efile.btf_maps_shndx >= 0 ||
2825                obj->efile.st_ops_shndx >= 0 ||
2826                obj->efile.st_ops_link_shndx >= 0 ||
2827                obj->nr_extern > 0;
2828 }
2829
2830 static bool kernel_needs_btf(const struct bpf_object *obj)
2831 {
2832         return obj->efile.st_ops_shndx >= 0 || obj->efile.st_ops_link_shndx >= 0;
2833 }
2834
2835 static int bpf_object__init_btf(struct bpf_object *obj,
2836                                 Elf_Data *btf_data,
2837                                 Elf_Data *btf_ext_data)
2838 {
2839         int err = -ENOENT;
2840
2841         if (btf_data) {
2842                 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
2843                 err = libbpf_get_error(obj->btf);
2844                 if (err) {
2845                         obj->btf = NULL;
2846                         pr_warn("Error loading ELF section %s: %d.\n", BTF_ELF_SEC, err);
2847                         goto out;
2848                 }
2849                 /* enforce 8-byte pointers for BPF-targeted BTFs */
2850                 btf__set_pointer_size(obj->btf, 8);
2851         }
2852         if (btf_ext_data) {
2853                 struct btf_ext_info *ext_segs[3];
2854                 int seg_num, sec_num;
2855
2856                 if (!obj->btf) {
2857                         pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
2858                                  BTF_EXT_ELF_SEC, BTF_ELF_SEC);
2859                         goto out;
2860                 }
2861                 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size);
2862                 err = libbpf_get_error(obj->btf_ext);
2863                 if (err) {
2864                         pr_warn("Error loading ELF section %s: %d. Ignored and continue.\n",
2865                                 BTF_EXT_ELF_SEC, err);
2866                         obj->btf_ext = NULL;
2867                         goto out;
2868                 }
2869
2870                 /* setup .BTF.ext to ELF section mapping */
2871                 ext_segs[0] = &obj->btf_ext->func_info;
2872                 ext_segs[1] = &obj->btf_ext->line_info;
2873                 ext_segs[2] = &obj->btf_ext->core_relo_info;
2874                 for (seg_num = 0; seg_num < ARRAY_SIZE(ext_segs); seg_num++) {
2875                         struct btf_ext_info *seg = ext_segs[seg_num];
2876                         const struct btf_ext_info_sec *sec;
2877                         const char *sec_name;
2878                         Elf_Scn *scn;
2879
2880                         if (seg->sec_cnt == 0)
2881                                 continue;
2882
2883                         seg->sec_idxs = calloc(seg->sec_cnt, sizeof(*seg->sec_idxs));
2884                         if (!seg->sec_idxs) {
2885                                 err = -ENOMEM;
2886                                 goto out;
2887                         }
2888
2889                         sec_num = 0;
2890                         for_each_btf_ext_sec(seg, sec) {
2891                                 /* preventively increment index to avoid doing
2892                                  * this before every continue below
2893                                  */
2894                                 sec_num++;
2895
2896                                 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
2897                                 if (str_is_empty(sec_name))
2898                                         continue;
2899                                 scn = elf_sec_by_name(obj, sec_name);
2900                                 if (!scn)
2901                                         continue;
2902
2903                                 seg->sec_idxs[sec_num - 1] = elf_ndxscn(scn);
2904                         }
2905                 }
2906         }
2907 out:
2908         if (err && libbpf_needs_btf(obj)) {
2909                 pr_warn("BTF is required, but is missing or corrupted.\n");
2910                 return err;
2911         }
2912         return 0;
2913 }
2914
2915 static int compare_vsi_off(const void *_a, const void *_b)
2916 {
2917         const struct btf_var_secinfo *a = _a;
2918         const struct btf_var_secinfo *b = _b;
2919
2920         return a->offset - b->offset;
2921 }
2922
2923 static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
2924                              struct btf_type *t)
2925 {
2926         __u32 size = 0, i, vars = btf_vlen(t);
2927         const char *sec_name = btf__name_by_offset(btf, t->name_off);
2928         struct btf_var_secinfo *vsi;
2929         bool fixup_offsets = false;
2930         int err;
2931
2932         if (!sec_name) {
2933                 pr_debug("No name found in string section for DATASEC kind.\n");
2934                 return -ENOENT;
2935         }
2936
2937         /* Extern-backing datasecs (.ksyms, .kconfig) have their size and
2938          * variable offsets set at the previous step. Further, not every
2939          * extern BTF VAR has corresponding ELF symbol preserved, so we skip
2940          * all fixups altogether for such sections and go straight to sorting
2941          * VARs within their DATASEC.
2942          */
2943         if (strcmp(sec_name, KCONFIG_SEC) == 0 || strcmp(sec_name, KSYMS_SEC) == 0)
2944                 goto sort_vars;
2945
2946         /* Clang leaves DATASEC size and VAR offsets as zeroes, so we need to
2947          * fix this up. But BPF static linker already fixes this up and fills
2948          * all the sizes and offsets during static linking. So this step has
2949          * to be optional. But the STV_HIDDEN handling is non-optional for any
2950          * non-extern DATASEC, so the variable fixup loop below handles both
2951          * functions at the same time, paying the cost of BTF VAR <-> ELF
2952          * symbol matching just once.
2953          */
2954         if (t->size == 0) {
2955                 err = find_elf_sec_sz(obj, sec_name, &size);
2956                 if (err || !size) {
2957                         pr_debug("sec '%s': failed to determine size from ELF: size %u, err %d\n",
2958                                  sec_name, size, err);
2959                         return -ENOENT;
2960                 }
2961
2962                 t->size = size;
2963                 fixup_offsets = true;
2964         }
2965
2966         for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
2967                 const struct btf_type *t_var;
2968                 struct btf_var *var;
2969                 const char *var_name;
2970                 Elf64_Sym *sym;
2971
2972                 t_var = btf__type_by_id(btf, vsi->type);
2973                 if (!t_var || !btf_is_var(t_var)) {
2974                         pr_debug("sec '%s': unexpected non-VAR type found\n", sec_name);
2975                         return -EINVAL;
2976                 }
2977
2978                 var = btf_var(t_var);
2979                 if (var->linkage == BTF_VAR_STATIC || var->linkage == BTF_VAR_GLOBAL_EXTERN)
2980                         continue;
2981
2982                 var_name = btf__name_by_offset(btf, t_var->name_off);
2983                 if (!var_name) {
2984                         pr_debug("sec '%s': failed to find name of DATASEC's member #%d\n",
2985                                  sec_name, i);
2986                         return -ENOENT;
2987                 }
2988
2989                 sym = find_elf_var_sym(obj, var_name);
2990                 if (IS_ERR(sym)) {
2991                         pr_debug("sec '%s': failed to find ELF symbol for VAR '%s'\n",
2992                                  sec_name, var_name);
2993                         return -ENOENT;
2994                 }
2995
2996                 if (fixup_offsets)
2997                         vsi->offset = sym->st_value;
2998
2999                 /* if variable is a global/weak symbol, but has restricted
3000                  * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF VAR
3001                  * as static. This follows similar logic for functions (BPF
3002                  * subprogs) and influences libbpf's further decisions about
3003                  * whether to make global data BPF array maps as
3004                  * BPF_F_MMAPABLE.
3005                  */
3006                 if (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN
3007                     || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL)
3008                         var->linkage = BTF_VAR_STATIC;
3009         }
3010
3011 sort_vars:
3012         qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off);
3013         return 0;
3014 }
3015
3016 static int bpf_object_fixup_btf(struct bpf_object *obj)
3017 {
3018         int i, n, err = 0;
3019
3020         if (!obj->btf)
3021                 return 0;
3022
3023         n = btf__type_cnt(obj->btf);
3024         for (i = 1; i < n; i++) {
3025                 struct btf_type *t = btf_type_by_id(obj->btf, i);
3026
3027                 /* Loader needs to fix up some of the things compiler
3028                  * couldn't get its hands on while emitting BTF. This
3029                  * is section size and global variable offset. We use
3030                  * the info from the ELF itself for this purpose.
3031                  */
3032                 if (btf_is_datasec(t)) {
3033                         err = btf_fixup_datasec(obj, obj->btf, t);
3034                         if (err)
3035                                 return err;
3036                 }
3037         }
3038
3039         return 0;
3040 }
3041
3042 static bool prog_needs_vmlinux_btf(struct bpf_program *prog)
3043 {
3044         if (prog->type == BPF_PROG_TYPE_STRUCT_OPS ||
3045             prog->type == BPF_PROG_TYPE_LSM)
3046                 return true;
3047
3048         /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs
3049          * also need vmlinux BTF
3050          */
3051         if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd)
3052                 return true;
3053
3054         return false;
3055 }
3056
3057 static bool obj_needs_vmlinux_btf(const struct bpf_object *obj)
3058 {
3059         struct bpf_program *prog;
3060         int i;
3061
3062         /* CO-RE relocations need kernel BTF, only when btf_custom_path
3063          * is not specified
3064          */
3065         if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path)
3066                 return true;
3067
3068         /* Support for typed ksyms needs kernel BTF */
3069         for (i = 0; i < obj->nr_extern; i++) {
3070                 const struct extern_desc *ext;
3071
3072                 ext = &obj->externs[i];
3073                 if (ext->type == EXT_KSYM && ext->ksym.type_id)
3074                         return true;
3075         }
3076
3077         bpf_object__for_each_program(prog, obj) {
3078                 if (!prog->autoload)
3079                         continue;
3080                 if (prog_needs_vmlinux_btf(prog))
3081                         return true;
3082         }
3083
3084         return false;
3085 }
3086
3087 static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force)
3088 {
3089         int err;
3090
3091         /* btf_vmlinux could be loaded earlier */
3092         if (obj->btf_vmlinux || obj->gen_loader)
3093                 return 0;
3094
3095         if (!force && !obj_needs_vmlinux_btf(obj))
3096                 return 0;
3097
3098         obj->btf_vmlinux = btf__load_vmlinux_btf();
3099         err = libbpf_get_error(obj->btf_vmlinux);
3100         if (err) {
3101                 pr_warn("Error loading vmlinux BTF: %d\n", err);
3102                 obj->btf_vmlinux = NULL;
3103                 return err;
3104         }
3105         return 0;
3106 }
3107
3108 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
3109 {
3110         struct btf *kern_btf = obj->btf;
3111         bool btf_mandatory, sanitize;
3112         int i, err = 0;
3113
3114         if (!obj->btf)
3115                 return 0;
3116
3117         if (!kernel_supports(obj, FEAT_BTF)) {
3118                 if (kernel_needs_btf(obj)) {
3119                         err = -EOPNOTSUPP;
3120                         goto report;
3121                 }
3122                 pr_debug("Kernel doesn't support BTF, skipping uploading it.\n");
3123                 return 0;
3124         }
3125
3126         /* Even though some subprogs are global/weak, user might prefer more
3127          * permissive BPF verification process that BPF verifier performs for
3128          * static functions, taking into account more context from the caller
3129          * functions. In such case, they need to mark such subprogs with
3130          * __attribute__((visibility("hidden"))) and libbpf will adjust
3131          * corresponding FUNC BTF type to be marked as static and trigger more
3132          * involved BPF verification process.
3133          */
3134         for (i = 0; i < obj->nr_programs; i++) {
3135                 struct bpf_program *prog = &obj->programs[i];
3136                 struct btf_type *t;
3137                 const char *name;
3138                 int j, n;
3139
3140                 if (!prog->mark_btf_static || !prog_is_subprog(obj, prog))
3141                         continue;
3142
3143                 n = btf__type_cnt(obj->btf);
3144                 for (j = 1; j < n; j++) {
3145                         t = btf_type_by_id(obj->btf, j);
3146                         if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL)
3147                                 continue;
3148
3149                         name = btf__str_by_offset(obj->btf, t->name_off);
3150                         if (strcmp(name, prog->name) != 0)
3151                                 continue;
3152
3153                         t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0);
3154                         break;
3155                 }
3156         }
3157
3158         if (!kernel_supports(obj, FEAT_BTF_DECL_TAG))
3159                 goto skip_exception_cb;
3160         for (i = 0; i < obj->nr_programs; i++) {
3161                 struct bpf_program *prog = &obj->programs[i];
3162                 int j, k, n;
3163
3164                 if (prog_is_subprog(obj, prog))
3165                         continue;
3166                 n = btf__type_cnt(obj->btf);
3167                 for (j = 1; j < n; j++) {
3168                         const char *str = "exception_callback:", *name;
3169                         size_t len = strlen(str);
3170                         struct btf_type *t;
3171
3172                         t = btf_type_by_id(obj->btf, j);
3173                         if (!btf_is_decl_tag(t) || btf_decl_tag(t)->component_idx != -1)
3174                                 continue;
3175
3176                         name = btf__str_by_offset(obj->btf, t->name_off);
3177                         if (strncmp(name, str, len))
3178                                 continue;
3179
3180                         t = btf_type_by_id(obj->btf, t->type);
3181                         if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL) {
3182                                 pr_warn("prog '%s': exception_callback:<value> decl tag not applied to the main program\n",
3183                                         prog->name);
3184                                 return -EINVAL;
3185                         }
3186                         if (strcmp(prog->name, btf__str_by_offset(obj->btf, t->name_off)))
3187                                 continue;
3188                         /* Multiple callbacks are specified for the same prog,
3189                          * the verifier will eventually return an error for this
3190                          * case, hence simply skip appending a subprog.
3191                          */
3192                         if (prog->exception_cb_idx >= 0) {
3193                                 prog->exception_cb_idx = -1;
3194                                 break;
3195                         }
3196
3197                         name += len;
3198                         if (str_is_empty(name)) {
3199                                 pr_warn("prog '%s': exception_callback:<value> decl tag contains empty value\n",
3200                                         prog->name);
3201                                 return -EINVAL;
3202                         }
3203
3204                         for (k = 0; k < obj->nr_programs; k++) {
3205                                 struct bpf_program *subprog = &obj->programs[k];
3206
3207                                 if (!prog_is_subprog(obj, subprog))
3208                                         continue;
3209                                 if (strcmp(name, subprog->name))
3210                                         continue;
3211                                 /* Enforce non-hidden, as from verifier point of
3212                                  * view it expects global functions, whereas the
3213                                  * mark_btf_static fixes up linkage as static.
3214                                  */
3215                                 if (!subprog->sym_global || subprog->mark_btf_static) {
3216                                         pr_warn("prog '%s': exception callback %s must be a global non-hidden function\n",
3217                                                 prog->name, subprog->name);
3218                                         return -EINVAL;
3219                                 }
3220                                 /* Let's see if we already saw a static exception callback with the same name */
3221                                 if (prog->exception_cb_idx >= 0) {
3222                                         pr_warn("prog '%s': multiple subprogs with same name as exception callback '%s'\n",
3223                                                 prog->name, subprog->name);
3224                                         return -EINVAL;
3225                                 }
3226                                 prog->exception_cb_idx = k;
3227                                 break;
3228                         }
3229
3230                         if (prog->exception_cb_idx >= 0)
3231                                 continue;
3232                         pr_warn("prog '%s': cannot find exception callback '%s'\n", prog->name, name);
3233                         return -ENOENT;
3234                 }
3235         }
3236 skip_exception_cb:
3237
3238         sanitize = btf_needs_sanitization(obj);
3239         if (sanitize) {
3240                 const void *raw_data;
3241                 __u32 sz;
3242
3243                 /* clone BTF to sanitize a copy and leave the original intact */
3244                 raw_data = btf__raw_data(obj->btf, &sz);
3245                 kern_btf = btf__new(raw_data, sz);
3246                 err = libbpf_get_error(kern_btf);
3247                 if (err)
3248                         return err;
3249
3250                 /* enforce 8-byte pointers for BPF-targeted BTFs */
3251                 btf__set_pointer_size(obj->btf, 8);
3252                 err = bpf_object__sanitize_btf(obj, kern_btf);
3253                 if (err)
3254                         return err;
3255         }
3256
3257         if (obj->gen_loader) {
3258                 __u32 raw_size = 0;
3259                 const void *raw_data = btf__raw_data(kern_btf, &raw_size);
3260
3261                 if (!raw_data)
3262                         return -ENOMEM;
3263                 bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size);
3264                 /* Pretend to have valid FD to pass various fd >= 0 checks.
3265                  * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
3266                  */
3267                 btf__set_fd(kern_btf, 0);
3268         } else {
3269                 /* currently BPF_BTF_LOAD only supports log_level 1 */
3270                 err = btf_load_into_kernel(kern_btf, obj->log_buf, obj->log_size,
3271                                            obj->log_level ? 1 : 0);
3272         }
3273         if (sanitize) {
3274                 if (!err) {
3275                         /* move fd to libbpf's BTF */
3276                         btf__set_fd(obj->btf, btf__fd(kern_btf));
3277                         btf__set_fd(kern_btf, -1);
3278                 }
3279                 btf__free(kern_btf);
3280         }
3281 report:
3282         if (err) {
3283                 btf_mandatory = kernel_needs_btf(obj);
3284                 pr_warn("Error loading .BTF into kernel: %d. %s\n", err,
3285                         btf_mandatory ? "BTF is mandatory, can't proceed."
3286                                       : "BTF is optional, ignoring.");
3287                 if (!btf_mandatory)
3288                         err = 0;
3289         }
3290         return err;
3291 }
3292
3293 static const char *elf_sym_str(const struct bpf_object *obj, size_t off)
3294 {
3295         const char *name;
3296
3297         name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off);
3298         if (!name) {
3299                 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3300                         off, obj->path, elf_errmsg(-1));
3301                 return NULL;
3302         }
3303
3304         return name;
3305 }
3306
3307 static const char *elf_sec_str(const struct bpf_object *obj, size_t off)
3308 {
3309         const char *name;
3310
3311         name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off);
3312         if (!name) {
3313                 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3314                         off, obj->path, elf_errmsg(-1));
3315                 return NULL;
3316         }
3317
3318         return name;
3319 }
3320
3321 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx)
3322 {
3323         Elf_Scn *scn;
3324
3325         scn = elf_getscn(obj->efile.elf, idx);
3326         if (!scn) {
3327                 pr_warn("elf: failed to get section(%zu) from %s: %s\n",
3328                         idx, obj->path, elf_errmsg(-1));
3329                 return NULL;
3330         }
3331         return scn;
3332 }
3333
3334 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name)
3335 {
3336         Elf_Scn *scn = NULL;
3337         Elf *elf = obj->efile.elf;
3338         const char *sec_name;
3339
3340         while ((scn = elf_nextscn(elf, scn)) != NULL) {
3341                 sec_name = elf_sec_name(obj, scn);
3342                 if (!sec_name)
3343                         return NULL;
3344
3345                 if (strcmp(sec_name, name) != 0)
3346                         continue;
3347
3348                 return scn;
3349         }
3350         return NULL;
3351 }
3352
3353 static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn)
3354 {
3355         Elf64_Shdr *shdr;
3356
3357         if (!scn)
3358                 return NULL;
3359
3360         shdr = elf64_getshdr(scn);
3361         if (!shdr) {
3362                 pr_warn("elf: failed to get section(%zu) header from %s: %s\n",
3363                         elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3364                 return NULL;
3365         }
3366
3367         return shdr;
3368 }
3369
3370 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn)
3371 {
3372         const char *name;
3373         Elf64_Shdr *sh;
3374
3375         if (!scn)
3376                 return NULL;
3377
3378         sh = elf_sec_hdr(obj, scn);
3379         if (!sh)
3380                 return NULL;
3381
3382         name = elf_sec_str(obj, sh->sh_name);
3383         if (!name) {
3384                 pr_warn("elf: failed to get section(%zu) name from %s: %s\n",
3385                         elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3386                 return NULL;
3387         }
3388
3389         return name;
3390 }
3391
3392 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn)
3393 {
3394         Elf_Data *data;
3395
3396         if (!scn)
3397                 return NULL;
3398
3399         data = elf_getdata(scn, 0);
3400         if (!data) {
3401                 pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n",
3402                         elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>",
3403                         obj->path, elf_errmsg(-1));
3404                 return NULL;
3405         }
3406
3407         return data;
3408 }
3409
3410 static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx)
3411 {
3412         if (idx >= obj->efile.symbols->d_size / sizeof(Elf64_Sym))
3413                 return NULL;
3414
3415         return (Elf64_Sym *)obj->efile.symbols->d_buf + idx;
3416 }
3417
3418 static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx)
3419 {
3420         if (idx >= data->d_size / sizeof(Elf64_Rel))
3421                 return NULL;
3422
3423         return (Elf64_Rel *)data->d_buf + idx;
3424 }
3425
3426 static bool is_sec_name_dwarf(const char *name)
3427 {
3428         /* approximation, but the actual list is too long */
3429         return str_has_pfx(name, ".debug_");
3430 }
3431
3432 static bool ignore_elf_section(Elf64_Shdr *hdr, const char *name)
3433 {
3434         /* no special handling of .strtab */
3435         if (hdr->sh_type == SHT_STRTAB)
3436                 return true;
3437
3438         /* ignore .llvm_addrsig section as well */
3439         if (hdr->sh_type == SHT_LLVM_ADDRSIG)
3440                 return true;
3441
3442         /* no subprograms will lead to an empty .text section, ignore it */
3443         if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 &&
3444             strcmp(name, ".text") == 0)
3445                 return true;
3446
3447         /* DWARF sections */
3448         if (is_sec_name_dwarf(name))
3449                 return true;
3450
3451         if (str_has_pfx(name, ".rel")) {
3452                 name += sizeof(".rel") - 1;
3453                 /* DWARF section relocations */
3454                 if (is_sec_name_dwarf(name))
3455                         return true;
3456
3457                 /* .BTF and .BTF.ext don't need relocations */
3458                 if (strcmp(name, BTF_ELF_SEC) == 0 ||
3459                     strcmp(name, BTF_EXT_ELF_SEC) == 0)
3460                         return true;
3461         }
3462
3463         return false;
3464 }
3465
3466 static int cmp_progs(const void *_a, const void *_b)
3467 {
3468         const struct bpf_program *a = _a;
3469         const struct bpf_program *b = _b;
3470
3471         if (a->sec_idx != b->sec_idx)
3472                 return a->sec_idx < b->sec_idx ? -1 : 1;
3473
3474         /* sec_insn_off can't be the same within the section */
3475         return a->sec_insn_off < b->sec_insn_off ? -1 : 1;
3476 }
3477
3478 static int bpf_object__elf_collect(struct bpf_object *obj)
3479 {
3480         struct elf_sec_desc *sec_desc;
3481         Elf *elf = obj->efile.elf;
3482         Elf_Data *btf_ext_data = NULL;
3483         Elf_Data *btf_data = NULL;
3484         int idx = 0, err = 0;
3485         const char *name;
3486         Elf_Data *data;
3487         Elf_Scn *scn;
3488         Elf64_Shdr *sh;
3489
3490         /* ELF section indices are 0-based, but sec #0 is special "invalid"
3491          * section. Since section count retrieved by elf_getshdrnum() does
3492          * include sec #0, it is already the necessary size of an array to keep
3493          * all the sections.
3494          */
3495         if (elf_getshdrnum(obj->efile.elf, &obj->efile.sec_cnt)) {
3496                 pr_warn("elf: failed to get the number of sections for %s: %s\n",
3497                         obj->path, elf_errmsg(-1));
3498                 return -LIBBPF_ERRNO__FORMAT;
3499         }
3500         obj->efile.secs = calloc(obj->efile.sec_cnt, sizeof(*obj->efile.secs));
3501         if (!obj->efile.secs)
3502                 return -ENOMEM;
3503
3504         /* a bunch of ELF parsing functionality depends on processing symbols,
3505          * so do the first pass and find the symbol table
3506          */
3507         scn = NULL;
3508         while ((scn = elf_nextscn(elf, scn)) != NULL) {
3509                 sh = elf_sec_hdr(obj, scn);
3510                 if (!sh)
3511                         return -LIBBPF_ERRNO__FORMAT;
3512
3513                 if (sh->sh_type == SHT_SYMTAB) {
3514                         if (obj->efile.symbols) {
3515                                 pr_warn("elf: multiple symbol tables in %s\n", obj->path);
3516                                 return -LIBBPF_ERRNO__FORMAT;
3517                         }
3518
3519                         data = elf_sec_data(obj, scn);
3520                         if (!data)
3521                                 return -LIBBPF_ERRNO__FORMAT;
3522
3523                         idx = elf_ndxscn(scn);
3524
3525                         obj->efile.symbols = data;
3526                         obj->efile.symbols_shndx = idx;
3527                         obj->efile.strtabidx = sh->sh_link;
3528                 }
3529         }
3530
3531         if (!obj->efile.symbols) {
3532                 pr_warn("elf: couldn't find symbol table in %s, stripped object file?\n",
3533                         obj->path);
3534                 return -ENOENT;
3535         }
3536
3537         scn = NULL;
3538         while ((scn = elf_nextscn(elf, scn)) != NULL) {
3539                 idx = elf_ndxscn(scn);
3540                 sec_desc = &obj->efile.secs[idx];
3541
3542                 sh = elf_sec_hdr(obj, scn);
3543                 if (!sh)
3544                         return -LIBBPF_ERRNO__FORMAT;
3545
3546                 name = elf_sec_str(obj, sh->sh_name);
3547                 if (!name)
3548                         return -LIBBPF_ERRNO__FORMAT;
3549
3550                 if (ignore_elf_section(sh, name))
3551                         continue;
3552
3553                 data = elf_sec_data(obj, scn);
3554                 if (!data)
3555                         return -LIBBPF_ERRNO__FORMAT;
3556
3557                 pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
3558                          idx, name, (unsigned long)data->d_size,
3559                          (int)sh->sh_link, (unsigned long)sh->sh_flags,
3560                          (int)sh->sh_type);
3561
3562                 if (strcmp(name, "license") == 0) {
3563                         err = bpf_object__init_license(obj, data->d_buf, data->d_size);
3564                         if (err)
3565                                 return err;
3566                 } else if (strcmp(name, "version") == 0) {
3567                         err = bpf_object__init_kversion(obj, data->d_buf, data->d_size);
3568                         if (err)
3569                                 return err;
3570                 } else if (strcmp(name, "maps") == 0) {
3571                         pr_warn("elf: legacy map definitions in 'maps' section are not supported by libbpf v1.0+\n");
3572                         return -ENOTSUP;
3573                 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
3574                         obj->efile.btf_maps_shndx = idx;
3575                 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
3576                         if (sh->sh_type != SHT_PROGBITS)
3577                                 return -LIBBPF_ERRNO__FORMAT;
3578                         btf_data = data;
3579                 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
3580                         if (sh->sh_type != SHT_PROGBITS)
3581                                 return -LIBBPF_ERRNO__FORMAT;
3582                         btf_ext_data = data;
3583                 } else if (sh->sh_type == SHT_SYMTAB) {
3584                         /* already processed during the first pass above */
3585                 } else if (sh->sh_type == SHT_PROGBITS && data->d_size > 0) {
3586                         if (sh->sh_flags & SHF_EXECINSTR) {
3587                                 if (strcmp(name, ".text") == 0)
3588                                         obj->efile.text_shndx = idx;
3589                                 err = bpf_object__add_programs(obj, data, name, idx);
3590                                 if (err)
3591                                         return err;
3592                         } else if (strcmp(name, DATA_SEC) == 0 ||
3593                                    str_has_pfx(name, DATA_SEC ".")) {
3594                                 sec_desc->sec_type = SEC_DATA;
3595                                 sec_desc->shdr = sh;
3596                                 sec_desc->data = data;
3597                         } else if (strcmp(name, RODATA_SEC) == 0 ||
3598                                    str_has_pfx(name, RODATA_SEC ".")) {
3599                                 sec_desc->sec_type = SEC_RODATA;
3600                                 sec_desc->shdr = sh;
3601                                 sec_desc->data = data;
3602                         } else if (strcmp(name, STRUCT_OPS_SEC) == 0) {
3603                                 obj->efile.st_ops_data = data;
3604                                 obj->efile.st_ops_shndx = idx;
3605                         } else if (strcmp(name, STRUCT_OPS_LINK_SEC) == 0) {
3606                                 obj->efile.st_ops_link_data = data;
3607                                 obj->efile.st_ops_link_shndx = idx;
3608                         } else {
3609                                 pr_info("elf: skipping unrecognized data section(%d) %s\n",
3610                                         idx, name);
3611                         }
3612                 } else if (sh->sh_type == SHT_REL) {
3613                         int targ_sec_idx = sh->sh_info; /* points to other section */
3614
3615                         if (sh->sh_entsize != sizeof(Elf64_Rel) ||
3616                             targ_sec_idx >= obj->efile.sec_cnt)
3617                                 return -LIBBPF_ERRNO__FORMAT;
3618
3619                         /* Only do relo for section with exec instructions */
3620                         if (!section_have_execinstr(obj, targ_sec_idx) &&
3621                             strcmp(name, ".rel" STRUCT_OPS_SEC) &&
3622                             strcmp(name, ".rel" STRUCT_OPS_LINK_SEC) &&
3623                             strcmp(name, ".rel" MAPS_ELF_SEC)) {
3624                                 pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n",
3625                                         idx, name, targ_sec_idx,
3626                                         elf_sec_name(obj, elf_sec_by_idx(obj, targ_sec_idx)) ?: "<?>");
3627                                 continue;
3628                         }
3629
3630                         sec_desc->sec_type = SEC_RELO;
3631                         sec_desc->shdr = sh;
3632                         sec_desc->data = data;
3633                 } else if (sh->sh_type == SHT_NOBITS && (strcmp(name, BSS_SEC) == 0 ||
3634                                                          str_has_pfx(name, BSS_SEC "."))) {
3635                         sec_desc->sec_type = SEC_BSS;
3636                         sec_desc->shdr = sh;
3637                         sec_desc->data = data;
3638                 } else {
3639                         pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name,
3640                                 (size_t)sh->sh_size);
3641                 }
3642         }
3643
3644         if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
3645                 pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path);
3646                 return -LIBBPF_ERRNO__FORMAT;
3647         }
3648
3649         /* sort BPF programs by section name and in-section instruction offset
3650          * for faster search
3651          */
3652         if (obj->nr_programs)
3653                 qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs);
3654
3655         return bpf_object__init_btf(obj, btf_data, btf_ext_data);
3656 }
3657
3658 static bool sym_is_extern(const Elf64_Sym *sym)
3659 {
3660         int bind = ELF64_ST_BIND(sym->st_info);
3661         /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
3662         return sym->st_shndx == SHN_UNDEF &&
3663                (bind == STB_GLOBAL || bind == STB_WEAK) &&
3664                ELF64_ST_TYPE(sym->st_info) == STT_NOTYPE;
3665 }
3666
3667 static bool sym_is_subprog(const Elf64_Sym *sym, int text_shndx)
3668 {
3669         int bind = ELF64_ST_BIND(sym->st_info);
3670         int type = ELF64_ST_TYPE(sym->st_info);
3671
3672         /* in .text section */
3673         if (sym->st_shndx != text_shndx)
3674                 return false;
3675
3676         /* local function */
3677         if (bind == STB_LOCAL && type == STT_SECTION)
3678                 return true;
3679
3680         /* global function */
3681         return bind == STB_GLOBAL && type == STT_FUNC;
3682 }
3683
3684 static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
3685 {
3686         const struct btf_type *t;
3687         const char *tname;
3688         int i, n;
3689
3690         if (!btf)
3691                 return -ESRCH;
3692
3693         n = btf__type_cnt(btf);
3694         for (i = 1; i < n; i++) {
3695                 t = btf__type_by_id(btf, i);
3696
3697                 if (!btf_is_var(t) && !btf_is_func(t))
3698                         continue;
3699
3700                 tname = btf__name_by_offset(btf, t->name_off);
3701                 if (strcmp(tname, ext_name))
3702                         continue;
3703
3704                 if (btf_is_var(t) &&
3705                     btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
3706                         return -EINVAL;
3707
3708                 if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN)
3709                         return -EINVAL;
3710
3711                 return i;
3712         }
3713
3714         return -ENOENT;
3715 }
3716
3717 static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) {
3718         const struct btf_var_secinfo *vs;
3719         const struct btf_type *t;
3720         int i, j, n;
3721
3722         if (!btf)
3723                 return -ESRCH;
3724
3725         n = btf__type_cnt(btf);
3726         for (i = 1; i < n; i++) {
3727                 t = btf__type_by_id(btf, i);
3728
3729                 if (!btf_is_datasec(t))
3730                         continue;
3731
3732                 vs = btf_var_secinfos(t);
3733                 for (j = 0; j < btf_vlen(t); j++, vs++) {
3734                         if (vs->type == ext_btf_id)
3735                                 return i;
3736                 }
3737         }
3738
3739         return -ENOENT;
3740 }
3741
3742 static enum kcfg_type find_kcfg_type(const struct btf *btf, int id,
3743                                      bool *is_signed)
3744 {
3745         const struct btf_type *t;
3746         const char *name;
3747
3748         t = skip_mods_and_typedefs(btf, id, NULL);
3749         name = btf__name_by_offset(btf, t->name_off);
3750
3751         if (is_signed)
3752                 *is_signed = false;
3753         switch (btf_kind(t)) {
3754         case BTF_KIND_INT: {
3755                 int enc = btf_int_encoding(t);
3756
3757                 if (enc & BTF_INT_BOOL)
3758                         return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN;
3759                 if (is_signed)
3760                         *is_signed = enc & BTF_INT_SIGNED;
3761                 if (t->size == 1)
3762                         return KCFG_CHAR;
3763                 if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
3764                         return KCFG_UNKNOWN;
3765                 return KCFG_INT;
3766         }
3767         case BTF_KIND_ENUM:
3768                 if (t->size != 4)
3769                         return KCFG_UNKNOWN;
3770                 if (strcmp(name, "libbpf_tristate"))
3771                         return KCFG_UNKNOWN;
3772                 return KCFG_TRISTATE;
3773         case BTF_KIND_ENUM64:
3774                 if (strcmp(name, "libbpf_tristate"))
3775                         return KCFG_UNKNOWN;
3776                 return KCFG_TRISTATE;
3777         case BTF_KIND_ARRAY:
3778                 if (btf_array(t)->nelems == 0)
3779                         return KCFG_UNKNOWN;
3780                 if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR)
3781                         return KCFG_UNKNOWN;
3782                 return KCFG_CHAR_ARR;
3783         default:
3784                 return KCFG_UNKNOWN;
3785         }
3786 }
3787
3788 static int cmp_externs(const void *_a, const void *_b)
3789 {
3790         const struct extern_desc *a = _a;
3791         const struct extern_desc *b = _b;
3792
3793         if (a->type != b->type)
3794                 return a->type < b->type ? -1 : 1;
3795
3796         if (a->type == EXT_KCFG) {
3797                 /* descending order by alignment requirements */
3798                 if (a->kcfg.align != b->kcfg.align)
3799                         return a->kcfg.align > b->kcfg.align ? -1 : 1;
3800                 /* ascending order by size, within same alignment class */
3801                 if (a->kcfg.sz != b->kcfg.sz)
3802                         return a->kcfg.sz < b->kcfg.sz ? -1 : 1;
3803         }
3804
3805         /* resolve ties by name */
3806         return strcmp(a->name, b->name);
3807 }
3808
3809 static int find_int_btf_id(const struct btf *btf)
3810 {
3811         const struct btf_type *t;
3812         int i, n;
3813
3814         n = btf__type_cnt(btf);
3815         for (i = 1; i < n; i++) {
3816                 t = btf__type_by_id(btf, i);
3817
3818                 if (btf_is_int(t) && btf_int_bits(t) == 32)
3819                         return i;
3820         }
3821
3822         return 0;
3823 }
3824
3825 static int add_dummy_ksym_var(struct btf *btf)
3826 {
3827         int i, int_btf_id, sec_btf_id, dummy_var_btf_id;
3828         const struct btf_var_secinfo *vs;
3829         const struct btf_type *sec;
3830
3831         if (!btf)
3832                 return 0;
3833
3834         sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC,
3835                                             BTF_KIND_DATASEC);
3836         if (sec_btf_id < 0)
3837                 return 0;
3838
3839         sec = btf__type_by_id(btf, sec_btf_id);
3840         vs = btf_var_secinfos(sec);
3841         for (i = 0; i < btf_vlen(sec); i++, vs++) {
3842                 const struct btf_type *vt;
3843
3844                 vt = btf__type_by_id(btf, vs->type);
3845                 if (btf_is_func(vt))
3846                         break;
3847         }
3848
3849         /* No func in ksyms sec.  No need to add dummy var. */
3850         if (i == btf_vlen(sec))
3851                 return 0;
3852
3853         int_btf_id = find_int_btf_id(btf);
3854         dummy_var_btf_id = btf__add_var(btf,
3855                                         "dummy_ksym",
3856                                         BTF_VAR_GLOBAL_ALLOCATED,
3857                                         int_btf_id);
3858         if (dummy_var_btf_id < 0)
3859                 pr_warn("cannot create a dummy_ksym var\n");
3860
3861         return dummy_var_btf_id;
3862 }
3863
3864 static int bpf_object__collect_externs(struct bpf_object *obj)
3865 {
3866         struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL;
3867         const struct btf_type *t;
3868         struct extern_desc *ext;
3869         int i, n, off, dummy_var_btf_id;
3870         const char *ext_name, *sec_name;
3871         size_t ext_essent_len;
3872         Elf_Scn *scn;
3873         Elf64_Shdr *sh;
3874
3875         if (!obj->efile.symbols)
3876                 return 0;
3877
3878         scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx);
3879         sh = elf_sec_hdr(obj, scn);
3880         if (!sh || sh->sh_entsize != sizeof(Elf64_Sym))
3881                 return -LIBBPF_ERRNO__FORMAT;
3882
3883         dummy_var_btf_id = add_dummy_ksym_var(obj->btf);
3884         if (dummy_var_btf_id < 0)
3885                 return dummy_var_btf_id;
3886
3887         n = sh->sh_size / sh->sh_entsize;
3888         pr_debug("looking for externs among %d symbols...\n", n);
3889
3890         for (i = 0; i < n; i++) {
3891                 Elf64_Sym *sym = elf_sym_by_idx(obj, i);
3892
3893                 if (!sym)
3894                         return -LIBBPF_ERRNO__FORMAT;
3895                 if (!sym_is_extern(sym))
3896                         continue;
3897                 ext_name = elf_sym_str(obj, sym->st_name);
3898                 if (!ext_name || !ext_name[0])
3899                         continue;
3900
3901                 ext = obj->externs;
3902                 ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
3903                 if (!ext)
3904                         return -ENOMEM;
3905                 obj->externs = ext;
3906                 ext = &ext[obj->nr_extern];
3907                 memset(ext, 0, sizeof(*ext));
3908                 obj->nr_extern++;
3909
3910                 ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
3911                 if (ext->btf_id <= 0) {
3912                         pr_warn("failed to find BTF for extern '%s': %d\n",
3913                                 ext_name, ext->btf_id);
3914                         return ext->btf_id;
3915                 }
3916                 t = btf__type_by_id(obj->btf, ext->btf_id);
3917                 ext->name = btf__name_by_offset(obj->btf, t->name_off);
3918                 ext->sym_idx = i;
3919                 ext->is_weak = ELF64_ST_BIND(sym->st_info) == STB_WEAK;
3920
3921                 ext_essent_len = bpf_core_essential_name_len(ext->name);
3922                 ext->essent_name = NULL;
3923                 if (ext_essent_len != strlen(ext->name)) {
3924                         ext->essent_name = strndup(ext->name, ext_essent_len);
3925                         if (!ext->essent_name)
3926                                 return -ENOMEM;
3927                 }
3928
3929                 ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id);
3930                 if (ext->sec_btf_id <= 0) {
3931                         pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n",
3932                                 ext_name, ext->btf_id, ext->sec_btf_id);
3933                         return ext->sec_btf_id;
3934                 }
3935                 sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id);
3936                 sec_name = btf__name_by_offset(obj->btf, sec->name_off);
3937
3938                 if (strcmp(sec_name, KCONFIG_SEC) == 0) {
3939                         if (btf_is_func(t)) {
3940                                 pr_warn("extern function %s is unsupported under %s section\n",
3941                                         ext->name, KCONFIG_SEC);
3942                                 return -ENOTSUP;
3943                         }
3944                         kcfg_sec = sec;
3945                         ext->type = EXT_KCFG;
3946                         ext->kcfg.sz = btf__resolve_size(obj->btf, t->type);
3947                         if (ext->kcfg.sz <= 0) {
3948                                 pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n",
3949                                         ext_name, ext->kcfg.sz);
3950                                 return ext->kcfg.sz;
3951                         }
3952                         ext->kcfg.align = btf__align_of(obj->btf, t->type);
3953                         if (ext->kcfg.align <= 0) {
3954                                 pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n",
3955                                         ext_name, ext->kcfg.align);
3956                                 return -EINVAL;
3957                         }
3958                         ext->kcfg.type = find_kcfg_type(obj->btf, t->type,
3959                                                         &ext->kcfg.is_signed);
3960                         if (ext->kcfg.type == KCFG_UNKNOWN) {
3961                                 pr_warn("extern (kcfg) '%s': type is unsupported\n", ext_name);
3962                                 return -ENOTSUP;
3963                         }
3964                 } else if (strcmp(sec_name, KSYMS_SEC) == 0) {
3965                         ksym_sec = sec;
3966                         ext->type = EXT_KSYM;
3967                         skip_mods_and_typedefs(obj->btf, t->type,
3968                                                &ext->ksym.type_id);
3969                 } else {
3970                         pr_warn("unrecognized extern section '%s'\n", sec_name);
3971                         return -ENOTSUP;
3972                 }
3973         }
3974         pr_debug("collected %d externs total\n", obj->nr_extern);
3975
3976         if (!obj->nr_extern)
3977                 return 0;
3978
3979         /* sort externs by type, for kcfg ones also by (align, size, name) */
3980         qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
3981
3982         /* for .ksyms section, we need to turn all externs into allocated
3983          * variables in BTF to pass kernel verification; we do this by
3984          * pretending that each extern is a 8-byte variable
3985          */
3986         if (ksym_sec) {
3987                 /* find existing 4-byte integer type in BTF to use for fake
3988                  * extern variables in DATASEC
3989                  */
3990                 int int_btf_id = find_int_btf_id(obj->btf);
3991                 /* For extern function, a dummy_var added earlier
3992                  * will be used to replace the vs->type and
3993                  * its name string will be used to refill
3994                  * the missing param's name.
3995                  */
3996                 const struct btf_type *dummy_var;
3997
3998                 dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id);
3999                 for (i = 0; i < obj->nr_extern; i++) {
4000                         ext = &obj->externs[i];
4001                         if (ext->type != EXT_KSYM)
4002                                 continue;
4003                         pr_debug("extern (ksym) #%d: symbol %d, name %s\n",
4004                                  i, ext->sym_idx, ext->name);
4005                 }
4006
4007                 sec = ksym_sec;
4008                 n = btf_vlen(sec);
4009                 for (i = 0, off = 0; i < n; i++, off += sizeof(int)) {
4010                         struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
4011                         struct btf_type *vt;
4012
4013                         vt = (void *)btf__type_by_id(obj->btf, vs->type);
4014                         ext_name = btf__name_by_offset(obj->btf, vt->name_off);
4015                         ext = find_extern_by_name(obj, ext_name);
4016                         if (!ext) {
4017                                 pr_warn("failed to find extern definition for BTF %s '%s'\n",
4018                                         btf_kind_str(vt), ext_name);
4019                                 return -ESRCH;
4020                         }
4021                         if (btf_is_func(vt)) {
4022                                 const struct btf_type *func_proto;
4023                                 struct btf_param *param;
4024                                 int j;
4025
4026                                 func_proto = btf__type_by_id(obj->btf,
4027                                                              vt->type);
4028                                 param = btf_params(func_proto);
4029                                 /* Reuse the dummy_var string if the
4030                                  * func proto does not have param name.
4031                                  */
4032                                 for (j = 0; j < btf_vlen(func_proto); j++)
4033                                         if (param[j].type && !param[j].name_off)
4034                                                 param[j].name_off =
4035                                                         dummy_var->name_off;
4036                                 vs->type = dummy_var_btf_id;
4037                                 vt->info &= ~0xffff;
4038                                 vt->info |= BTF_FUNC_GLOBAL;
4039                         } else {
4040                                 btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
4041                                 vt->type = int_btf_id;
4042                         }
4043                         vs->offset = off;
4044                         vs->size = sizeof(int);
4045                 }
4046                 sec->size = off;
4047         }
4048
4049         if (kcfg_sec) {
4050                 sec = kcfg_sec;
4051                 /* for kcfg externs calculate their offsets within a .kconfig map */
4052                 off = 0;
4053                 for (i = 0; i < obj->nr_extern; i++) {
4054                         ext = &obj->externs[i];
4055                         if (ext->type != EXT_KCFG)
4056                                 continue;
4057
4058                         ext->kcfg.data_off = roundup(off, ext->kcfg.align);
4059                         off = ext->kcfg.data_off + ext->kcfg.sz;
4060                         pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n",
4061                                  i, ext->sym_idx, ext->kcfg.data_off, ext->name);
4062                 }
4063                 sec->size = off;
4064                 n = btf_vlen(sec);
4065                 for (i = 0; i < n; i++) {
4066                         struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
4067
4068                         t = btf__type_by_id(obj->btf, vs->type);
4069                         ext_name = btf__name_by_offset(obj->btf, t->name_off);
4070                         ext = find_extern_by_name(obj, ext_name);
4071                         if (!ext) {
4072                                 pr_warn("failed to find extern definition for BTF var '%s'\n",
4073                                         ext_name);
4074                                 return -ESRCH;
4075                         }
4076                         btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
4077                         vs->offset = ext->kcfg.data_off;
4078                 }
4079         }
4080         return 0;
4081 }
4082
4083 static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog)
4084 {
4085         return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1;
4086 }
4087
4088 struct bpf_program *
4089 bpf_object__find_program_by_name(const struct bpf_object *obj,
4090                                  const char *name)
4091 {
4092         struct bpf_program *prog;
4093
4094         bpf_object__for_each_program(prog, obj) {
4095                 if (prog_is_subprog(obj, prog))
4096                         continue;
4097                 if (!strcmp(prog->name, name))
4098                         return prog;
4099         }
4100         return errno = ENOENT, NULL;
4101 }
4102
4103 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
4104                                       int shndx)
4105 {
4106         switch (obj->efile.secs[shndx].sec_type) {
4107         case SEC_BSS:
4108         case SEC_DATA:
4109         case SEC_RODATA:
4110                 return true;
4111         default:
4112                 return false;
4113         }
4114 }
4115
4116 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
4117                                       int shndx)
4118 {
4119         return shndx == obj->efile.btf_maps_shndx;
4120 }
4121
4122 static enum libbpf_map_type
4123 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
4124 {
4125         if (shndx == obj->efile.symbols_shndx)
4126                 return LIBBPF_MAP_KCONFIG;
4127
4128         switch (obj->efile.secs[shndx].sec_type) {
4129         case SEC_BSS:
4130                 return LIBBPF_MAP_BSS;
4131         case SEC_DATA:
4132                 return LIBBPF_MAP_DATA;
4133         case SEC_RODATA:
4134                 return LIBBPF_MAP_RODATA;
4135         default:
4136                 return LIBBPF_MAP_UNSPEC;
4137         }
4138 }
4139
4140 static int bpf_program__record_reloc(struct bpf_program *prog,
4141                                      struct reloc_desc *reloc_desc,
4142                                      __u32 insn_idx, const char *sym_name,
4143                                      const Elf64_Sym *sym, const Elf64_Rel *rel)
4144 {
4145         struct bpf_insn *insn = &prog->insns[insn_idx];
4146         size_t map_idx, nr_maps = prog->obj->nr_maps;
4147         struct bpf_object *obj = prog->obj;
4148         __u32 shdr_idx = sym->st_shndx;
4149         enum libbpf_map_type type;
4150         const char *sym_sec_name;
4151         struct bpf_map *map;
4152
4153         if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) {
4154                 pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n",
4155                         prog->name, sym_name, insn_idx, insn->code);
4156                 return -LIBBPF_ERRNO__RELOC;
4157         }
4158
4159         if (sym_is_extern(sym)) {
4160                 int sym_idx = ELF64_R_SYM(rel->r_info);
4161                 int i, n = obj->nr_extern;
4162                 struct extern_desc *ext;
4163
4164                 for (i = 0; i < n; i++) {
4165                         ext = &obj->externs[i];
4166                         if (ext->sym_idx == sym_idx)
4167                                 break;
4168                 }
4169                 if (i >= n) {
4170                         pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n",
4171                                 prog->name, sym_name, sym_idx);
4172                         return -LIBBPF_ERRNO__RELOC;
4173                 }
4174                 pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n",
4175                          prog->name, i, ext->name, ext->sym_idx, insn_idx);
4176                 if (insn->code == (BPF_JMP | BPF_CALL))
4177                         reloc_desc->type = RELO_EXTERN_CALL;
4178                 else
4179                         reloc_desc->type = RELO_EXTERN_LD64;
4180                 reloc_desc->insn_idx = insn_idx;
4181                 reloc_desc->ext_idx = i;
4182                 return 0;
4183         }
4184
4185         /* sub-program call relocation */
4186         if (is_call_insn(insn)) {
4187                 if (insn->src_reg != BPF_PSEUDO_CALL) {
4188                         pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name);
4189                         return -LIBBPF_ERRNO__RELOC;
4190                 }
4191                 /* text_shndx can be 0, if no default "main" program exists */
4192                 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
4193                         sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
4194                         pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n",
4195                                 prog->name, sym_name, sym_sec_name);
4196                         return -LIBBPF_ERRNO__RELOC;
4197                 }
4198                 if (sym->st_value % BPF_INSN_SZ) {
4199                         pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n",
4200                                 prog->name, sym_name, (size_t)sym->st_value);
4201                         return -LIBBPF_ERRNO__RELOC;
4202                 }
4203                 reloc_desc->type = RELO_CALL;
4204                 reloc_desc->insn_idx = insn_idx;
4205                 reloc_desc->sym_off = sym->st_value;
4206                 return 0;
4207         }
4208
4209         if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
4210                 pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n",
4211                         prog->name, sym_name, shdr_idx);
4212                 return -LIBBPF_ERRNO__RELOC;
4213         }
4214
4215         /* loading subprog addresses */
4216         if (sym_is_subprog(sym, obj->efile.text_shndx)) {
4217                 /* global_func: sym->st_value = offset in the section, insn->imm = 0.
4218                  * local_func: sym->st_value = 0, insn->imm = offset in the section.
4219                  */
4220                 if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) {
4221                         pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n",
4222                                 prog->name, sym_name, (size_t)sym->st_value, insn->imm);
4223                         return -LIBBPF_ERRNO__RELOC;
4224                 }
4225
4226                 reloc_desc->type = RELO_SUBPROG_ADDR;
4227                 reloc_desc->insn_idx = insn_idx;
4228                 reloc_desc->sym_off = sym->st_value;
4229                 return 0;
4230         }
4231
4232         type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
4233         sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
4234
4235         /* generic map reference relocation */
4236         if (type == LIBBPF_MAP_UNSPEC) {
4237                 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
4238                         pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n",
4239                                 prog->name, sym_name, sym_sec_name);
4240                         return -LIBBPF_ERRNO__RELOC;
4241                 }
4242                 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4243                         map = &obj->maps[map_idx];
4244                         if (map->libbpf_type != type ||
4245                             map->sec_idx != sym->st_shndx ||
4246                             map->sec_offset != sym->st_value)
4247                                 continue;
4248                         pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n",
4249                                  prog->name, map_idx, map->name, map->sec_idx,
4250                                  map->sec_offset, insn_idx);
4251                         break;
4252                 }
4253                 if (map_idx >= nr_maps) {
4254                         pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n",
4255                                 prog->name, sym_sec_name, (size_t)sym->st_value);
4256                         return -LIBBPF_ERRNO__RELOC;
4257                 }
4258                 reloc_desc->type = RELO_LD64;
4259                 reloc_desc->insn_idx = insn_idx;
4260                 reloc_desc->map_idx = map_idx;
4261                 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
4262                 return 0;
4263         }
4264
4265         /* global data map relocation */
4266         if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
4267                 pr_warn("prog '%s': bad data relo against section '%s'\n",
4268                         prog->name, sym_sec_name);
4269                 return -LIBBPF_ERRNO__RELOC;
4270         }
4271         for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4272                 map = &obj->maps[map_idx];
4273                 if (map->libbpf_type != type || map->sec_idx != sym->st_shndx)
4274                         continue;
4275                 pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n",
4276                          prog->name, map_idx, map->name, map->sec_idx,
4277                          map->sec_offset, insn_idx);
4278                 break;
4279         }
4280         if (map_idx >= nr_maps) {
4281                 pr_warn("prog '%s': data relo failed to find map for section '%s'\n",
4282                         prog->name, sym_sec_name);
4283                 return -LIBBPF_ERRNO__RELOC;
4284         }
4285
4286         reloc_desc->type = RELO_DATA;
4287         reloc_desc->insn_idx = insn_idx;
4288         reloc_desc->map_idx = map_idx;
4289         reloc_desc->sym_off = sym->st_value;
4290         return 0;
4291 }
4292
4293 static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx)
4294 {
4295         return insn_idx >= prog->sec_insn_off &&
4296                insn_idx < prog->sec_insn_off + prog->sec_insn_cnt;
4297 }
4298
4299 static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj,
4300                                                  size_t sec_idx, size_t insn_idx)
4301 {
4302         int l = 0, r = obj->nr_programs - 1, m;
4303         struct bpf_program *prog;
4304
4305         if (!obj->nr_programs)
4306                 return NULL;
4307
4308         while (l < r) {
4309                 m = l + (r - l + 1) / 2;
4310                 prog = &obj->programs[m];
4311
4312                 if (prog->sec_idx < sec_idx ||
4313                     (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx))
4314                         l = m;
4315                 else
4316                         r = m - 1;
4317         }
4318         /* matching program could be at index l, but it still might be the
4319          * wrong one, so we need to double check conditions for the last time
4320          */
4321         prog = &obj->programs[l];
4322         if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx))
4323                 return prog;
4324         return NULL;
4325 }
4326
4327 static int
4328 bpf_object__collect_prog_relos(struct bpf_object *obj, Elf64_Shdr *shdr, Elf_Data *data)
4329 {
4330         const char *relo_sec_name, *sec_name;
4331         size_t sec_idx = shdr->sh_info, sym_idx;
4332         struct bpf_program *prog;
4333         struct reloc_desc *relos;
4334         int err, i, nrels;
4335         const char *sym_name;
4336         __u32 insn_idx;
4337         Elf_Scn *scn;
4338         Elf_Data *scn_data;
4339         Elf64_Sym *sym;
4340         Elf64_Rel *rel;
4341
4342         if (sec_idx >= obj->efile.sec_cnt)
4343                 return -EINVAL;
4344
4345         scn = elf_sec_by_idx(obj, sec_idx);
4346         scn_data = elf_sec_data(obj, scn);
4347
4348         relo_sec_name = elf_sec_str(obj, shdr->sh_name);
4349         sec_name = elf_sec_name(obj, scn);
4350         if (!relo_sec_name || !sec_name)
4351                 return -EINVAL;
4352
4353         pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n",
4354                  relo_sec_name, sec_idx, sec_name);
4355         nrels = shdr->sh_size / shdr->sh_entsize;
4356
4357         for (i = 0; i < nrels; i++) {
4358                 rel = elf_rel_by_idx(data, i);
4359                 if (!rel) {
4360                         pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i);
4361                         return -LIBBPF_ERRNO__FORMAT;
4362                 }
4363
4364                 sym_idx = ELF64_R_SYM(rel->r_info);
4365                 sym = elf_sym_by_idx(obj, sym_idx);
4366                 if (!sym) {
4367                         pr_warn("sec '%s': symbol #%zu not found for relo #%d\n",
4368                                 relo_sec_name, sym_idx, i);
4369                         return -LIBBPF_ERRNO__FORMAT;
4370                 }
4371
4372                 if (sym->st_shndx >= obj->efile.sec_cnt) {
4373                         pr_warn("sec '%s': corrupted symbol #%zu pointing to invalid section #%zu for relo #%d\n",
4374                                 relo_sec_name, sym_idx, (size_t)sym->st_shndx, i);
4375                         return -LIBBPF_ERRNO__FORMAT;
4376                 }
4377
4378                 if (rel->r_offset % BPF_INSN_SZ || rel->r_offset >= scn_data->d_size) {
4379                         pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n",
4380                                 relo_sec_name, (size_t)rel->r_offset, i);
4381                         return -LIBBPF_ERRNO__FORMAT;
4382                 }
4383
4384                 insn_idx = rel->r_offset / BPF_INSN_SZ;
4385                 /* relocations against static functions are recorded as
4386                  * relocations against the section that contains a function;
4387                  * in such case, symbol will be STT_SECTION and sym.st_name
4388                  * will point to empty string (0), so fetch section name
4389                  * instead
4390                  */
4391                 if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION && sym->st_name == 0)
4392                         sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym->st_shndx));
4393                 else
4394                         sym_name = elf_sym_str(obj, sym->st_name);
4395                 sym_name = sym_name ?: "<?";
4396
4397                 pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n",
4398                          relo_sec_name, i, insn_idx, sym_name);
4399
4400                 prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
4401                 if (!prog) {
4402                         pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n",
4403                                 relo_sec_name, i, sec_name, insn_idx);
4404                         continue;
4405                 }
4406
4407                 relos = libbpf_reallocarray(prog->reloc_desc,
4408                                             prog->nr_reloc + 1, sizeof(*relos));
4409                 if (!relos)
4410                         return -ENOMEM;
4411                 prog->reloc_desc = relos;
4412
4413                 /* adjust insn_idx to local BPF program frame of reference */
4414                 insn_idx -= prog->sec_insn_off;
4415                 err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc],
4416                                                 insn_idx, sym_name, sym, rel);
4417                 if (err)
4418                         return err;
4419
4420                 prog->nr_reloc++;
4421         }
4422         return 0;
4423 }
4424
4425 static int map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map)
4426 {
4427         int id;
4428
4429         if (!obj->btf)
4430                 return -ENOENT;
4431
4432         /* if it's BTF-defined map, we don't need to search for type IDs.
4433          * For struct_ops map, it does not need btf_key_type_id and
4434          * btf_value_type_id.
4435          */
4436         if (map->sec_idx == obj->efile.btf_maps_shndx || bpf_map__is_struct_ops(map))
4437                 return 0;
4438
4439         /*
4440          * LLVM annotates global data differently in BTF, that is,
4441          * only as '.data', '.bss' or '.rodata'.
4442          */
4443         if (!bpf_map__is_internal(map))
4444                 return -ENOENT;
4445
4446         id = btf__find_by_name(obj->btf, map->real_name);
4447         if (id < 0)
4448                 return id;
4449
4450         map->btf_key_type_id = 0;
4451         map->btf_value_type_id = id;
4452         return 0;
4453 }
4454
4455 static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info)
4456 {
4457         char file[PATH_MAX], buff[4096];
4458         FILE *fp;
4459         __u32 val;
4460         int err;
4461
4462         snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd);
4463         memset(info, 0, sizeof(*info));
4464
4465         fp = fopen(file, "re");
4466         if (!fp) {
4467                 err = -errno;
4468                 pr_warn("failed to open %s: %d. No procfs support?\n", file,
4469                         err);
4470                 return err;
4471         }
4472
4473         while (fgets(buff, sizeof(buff), fp)) {
4474                 if (sscanf(buff, "map_type:\t%u", &val) == 1)
4475                         info->type = val;
4476                 else if (sscanf(buff, "key_size:\t%u", &val) == 1)
4477                         info->key_size = val;
4478                 else if (sscanf(buff, "value_size:\t%u", &val) == 1)
4479                         info->value_size = val;
4480                 else if (sscanf(buff, "max_entries:\t%u", &val) == 1)
4481                         info->max_entries = val;
4482                 else if (sscanf(buff, "map_flags:\t%i", &val) == 1)
4483                         info->map_flags = val;
4484         }
4485
4486         fclose(fp);
4487
4488         return 0;
4489 }
4490
4491 bool bpf_map__autocreate(const struct bpf_map *map)
4492 {
4493         return map->autocreate;
4494 }
4495
4496 int bpf_map__set_autocreate(struct bpf_map *map, bool autocreate)
4497 {
4498         if (map->obj->loaded)
4499                 return libbpf_err(-EBUSY);
4500
4501         map->autocreate = autocreate;
4502         return 0;
4503 }
4504
4505 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
4506 {
4507         struct bpf_map_info info;
4508         __u32 len = sizeof(info), name_len;
4509         int new_fd, err;
4510         char *new_name;
4511
4512         memset(&info, 0, len);
4513         err = bpf_map_get_info_by_fd(fd, &info, &len);
4514         if (err && errno == EINVAL)
4515                 err = bpf_get_map_info_from_fdinfo(fd, &info);
4516         if (err)
4517                 return libbpf_err(err);
4518
4519         name_len = strlen(info.name);
4520         if (name_len == BPF_OBJ_NAME_LEN - 1 && strncmp(map->name, info.name, name_len) == 0)
4521                 new_name = strdup(map->name);
4522         else
4523                 new_name = strdup(info.name);
4524
4525         if (!new_name)
4526                 return libbpf_err(-errno);
4527
4528         /*
4529          * Like dup(), but make sure new FD is >= 3 and has O_CLOEXEC set.
4530          * This is similar to what we do in ensure_good_fd(), but without
4531          * closing original FD.
4532          */
4533         new_fd = fcntl(fd, F_DUPFD_CLOEXEC, 3);
4534         if (new_fd < 0) {
4535                 err = -errno;
4536                 goto err_free_new_name;
4537         }
4538
4539         err = zclose(map->fd);
4540         if (err) {
4541                 err = -errno;
4542                 goto err_close_new_fd;
4543         }
4544         free(map->name);
4545
4546         map->fd = new_fd;
4547         map->name = new_name;
4548         map->def.type = info.type;
4549         map->def.key_size = info.key_size;
4550         map->def.value_size = info.value_size;
4551         map->def.max_entries = info.max_entries;
4552         map->def.map_flags = info.map_flags;
4553         map->btf_key_type_id = info.btf_key_type_id;
4554         map->btf_value_type_id = info.btf_value_type_id;
4555         map->reused = true;
4556         map->map_extra = info.map_extra;
4557
4558         return 0;
4559
4560 err_close_new_fd:
4561         close(new_fd);
4562 err_free_new_name:
4563         free(new_name);
4564         return libbpf_err(err);
4565 }
4566
4567 __u32 bpf_map__max_entries(const struct bpf_map *map)
4568 {
4569         return map->def.max_entries;
4570 }
4571
4572 struct bpf_map *bpf_map__inner_map(struct bpf_map *map)
4573 {
4574         if (!bpf_map_type__is_map_in_map(map->def.type))
4575                 return errno = EINVAL, NULL;
4576
4577         return map->inner_map;
4578 }
4579
4580 int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries)
4581 {
4582         if (map->obj->loaded)
4583                 return libbpf_err(-EBUSY);
4584
4585         map->def.max_entries = max_entries;
4586
4587         /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */
4588         if (map_is_ringbuf(map))
4589                 map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries);
4590
4591         return 0;
4592 }
4593
4594 static int
4595 bpf_object__probe_loading(struct bpf_object *obj)
4596 {
4597         char *cp, errmsg[STRERR_BUFSIZE];
4598         struct bpf_insn insns[] = {
4599                 BPF_MOV64_IMM(BPF_REG_0, 0),
4600                 BPF_EXIT_INSN(),
4601         };
4602         int ret, insn_cnt = ARRAY_SIZE(insns);
4603
4604         if (obj->gen_loader)
4605                 return 0;
4606
4607         ret = bump_rlimit_memlock();
4608         if (ret)
4609                 pr_warn("Failed to bump RLIMIT_MEMLOCK (err = %d), you might need to do it explicitly!\n", ret);
4610
4611         /* make sure basic loading works */
4612         ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4613         if (ret < 0)
4614                 ret = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, NULL);
4615         if (ret < 0) {
4616                 ret = errno;
4617                 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4618                 pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF "
4619                         "program. Make sure your kernel supports BPF "
4620                         "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is "
4621                         "set to big enough value.\n", __func__, cp, ret);
4622                 return -ret;
4623         }
4624         close(ret);
4625
4626         return 0;
4627 }
4628
4629 static int probe_fd(int fd)
4630 {
4631         if (fd >= 0)
4632                 close(fd);
4633         return fd >= 0;
4634 }
4635
4636 static int probe_kern_prog_name(void)
4637 {
4638         const size_t attr_sz = offsetofend(union bpf_attr, prog_name);
4639         struct bpf_insn insns[] = {
4640                 BPF_MOV64_IMM(BPF_REG_0, 0),
4641                 BPF_EXIT_INSN(),
4642         };
4643         union bpf_attr attr;
4644         int ret;
4645
4646         memset(&attr, 0, attr_sz);
4647         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
4648         attr.license = ptr_to_u64("GPL");
4649         attr.insns = ptr_to_u64(insns);
4650         attr.insn_cnt = (__u32)ARRAY_SIZE(insns);
4651         libbpf_strlcpy(attr.prog_name, "libbpf_nametest", sizeof(attr.prog_name));
4652
4653         /* make sure loading with name works */
4654         ret = sys_bpf_prog_load(&attr, attr_sz, PROG_LOAD_ATTEMPTS);
4655         return probe_fd(ret);
4656 }
4657
4658 static int probe_kern_global_data(void)
4659 {
4660         char *cp, errmsg[STRERR_BUFSIZE];
4661         struct bpf_insn insns[] = {
4662                 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
4663                 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
4664                 BPF_MOV64_IMM(BPF_REG_0, 0),
4665                 BPF_EXIT_INSN(),
4666         };
4667         int ret, map, insn_cnt = ARRAY_SIZE(insns);
4668
4669         map = bpf_map_create(BPF_MAP_TYPE_ARRAY, "libbpf_global", sizeof(int), 32, 1, NULL);
4670         if (map < 0) {
4671                 ret = -errno;
4672                 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4673                 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
4674                         __func__, cp, -ret);
4675                 return ret;
4676         }
4677
4678         insns[0].imm = map;
4679
4680         ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4681         close(map);
4682         return probe_fd(ret);
4683 }
4684
4685 static int probe_kern_btf(void)
4686 {
4687         static const char strs[] = "\0int";
4688         __u32 types[] = {
4689                 /* int */
4690                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
4691         };
4692
4693         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4694                                              strs, sizeof(strs)));
4695 }
4696
4697 static int probe_kern_btf_func(void)
4698 {
4699         static const char strs[] = "\0int\0x\0a";
4700         /* void x(int a) {} */
4701         __u32 types[] = {
4702                 /* int */
4703                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
4704                 /* FUNC_PROTO */                                /* [2] */
4705                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
4706                 BTF_PARAM_ENC(7, 1),
4707                 /* FUNC x */                                    /* [3] */
4708                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
4709         };
4710
4711         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4712                                              strs, sizeof(strs)));
4713 }
4714
4715 static int probe_kern_btf_func_global(void)
4716 {
4717         static const char strs[] = "\0int\0x\0a";
4718         /* static void x(int a) {} */
4719         __u32 types[] = {
4720                 /* int */
4721                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
4722                 /* FUNC_PROTO */                                /* [2] */
4723                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
4724                 BTF_PARAM_ENC(7, 1),
4725                 /* FUNC x BTF_FUNC_GLOBAL */                    /* [3] */
4726                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2),
4727         };
4728
4729         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4730                                              strs, sizeof(strs)));
4731 }
4732
4733 static int probe_kern_btf_datasec(void)
4734 {
4735         static const char strs[] = "\0x\0.data";
4736         /* static int a; */
4737         __u32 types[] = {
4738                 /* int */
4739                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
4740                 /* VAR x */                                     /* [2] */
4741                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
4742                 BTF_VAR_STATIC,
4743                 /* DATASEC val */                               /* [3] */
4744                 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
4745                 BTF_VAR_SECINFO_ENC(2, 0, 4),
4746         };
4747
4748         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4749                                              strs, sizeof(strs)));
4750 }
4751
4752 static int probe_kern_btf_float(void)
4753 {
4754         static const char strs[] = "\0float";
4755         __u32 types[] = {
4756                 /* float */
4757                 BTF_TYPE_FLOAT_ENC(1, 4),
4758         };
4759
4760         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4761                                              strs, sizeof(strs)));
4762 }
4763
4764 static int probe_kern_btf_decl_tag(void)
4765 {
4766         static const char strs[] = "\0tag";
4767         __u32 types[] = {
4768                 /* int */
4769                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
4770                 /* VAR x */                                     /* [2] */
4771                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
4772                 BTF_VAR_STATIC,
4773                 /* attr */
4774                 BTF_TYPE_DECL_TAG_ENC(1, 2, -1),
4775         };
4776
4777         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4778                                              strs, sizeof(strs)));
4779 }
4780
4781 static int probe_kern_btf_type_tag(void)
4782 {
4783         static const char strs[] = "\0tag";
4784         __u32 types[] = {
4785                 /* int */
4786                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),          /* [1] */
4787                 /* attr */
4788                 BTF_TYPE_TYPE_TAG_ENC(1, 1),                            /* [2] */
4789                 /* ptr */
4790                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_PTR, 0, 0), 2),   /* [3] */
4791         };
4792
4793         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4794                                              strs, sizeof(strs)));
4795 }
4796
4797 static int probe_kern_array_mmap(void)
4798 {
4799         LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE);
4800         int fd;
4801
4802         fd = bpf_map_create(BPF_MAP_TYPE_ARRAY, "libbpf_mmap", sizeof(int), sizeof(int), 1, &opts);
4803         return probe_fd(fd);
4804 }
4805
4806 static int probe_kern_exp_attach_type(void)
4807 {
4808         LIBBPF_OPTS(bpf_prog_load_opts, opts, .expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE);
4809         struct bpf_insn insns[] = {
4810                 BPF_MOV64_IMM(BPF_REG_0, 0),
4811                 BPF_EXIT_INSN(),
4812         };
4813         int fd, insn_cnt = ARRAY_SIZE(insns);
4814
4815         /* use any valid combination of program type and (optional)
4816          * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS)
4817          * to see if kernel supports expected_attach_type field for
4818          * BPF_PROG_LOAD command
4819          */
4820         fd = bpf_prog_load(BPF_PROG_TYPE_CGROUP_SOCK, NULL, "GPL", insns, insn_cnt, &opts);
4821         return probe_fd(fd);
4822 }
4823
4824 static int probe_kern_probe_read_kernel(void)
4825 {
4826         struct bpf_insn insns[] = {
4827                 BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),   /* r1 = r10 (fp) */
4828                 BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -8),  /* r1 += -8 */
4829                 BPF_MOV64_IMM(BPF_REG_2, 8),            /* r2 = 8 */
4830                 BPF_MOV64_IMM(BPF_REG_3, 0),            /* r3 = 0 */
4831                 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_kernel),
4832                 BPF_EXIT_INSN(),
4833         };
4834         int fd, insn_cnt = ARRAY_SIZE(insns);
4835
4836         fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, NULL);
4837         return probe_fd(fd);
4838 }
4839
4840 static int probe_prog_bind_map(void)
4841 {
4842         char *cp, errmsg[STRERR_BUFSIZE];
4843         struct bpf_insn insns[] = {
4844                 BPF_MOV64_IMM(BPF_REG_0, 0),
4845                 BPF_EXIT_INSN(),
4846         };
4847         int ret, map, prog, insn_cnt = ARRAY_SIZE(insns);
4848
4849         map = bpf_map_create(BPF_MAP_TYPE_ARRAY, "libbpf_det_bind", sizeof(int), 32, 1, NULL);
4850         if (map < 0) {
4851                 ret = -errno;
4852                 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
4853                 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
4854                         __func__, cp, -ret);
4855                 return ret;
4856         }
4857
4858         prog = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
4859         if (prog < 0) {
4860                 close(map);
4861                 return 0;
4862         }
4863
4864         ret = bpf_prog_bind_map(prog, map, NULL);
4865
4866         close(map);
4867         close(prog);
4868
4869         return ret >= 0;
4870 }
4871
4872 static int probe_module_btf(void)
4873 {
4874         static const char strs[] = "\0int";
4875         __u32 types[] = {
4876                 /* int */
4877                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
4878         };
4879         struct bpf_btf_info info;
4880         __u32 len = sizeof(info);
4881         char name[16];
4882         int fd, err;
4883
4884         fd = libbpf__load_raw_btf((char *)types, sizeof(types), strs, sizeof(strs));
4885         if (fd < 0)
4886                 return 0; /* BTF not supported at all */
4887
4888         memset(&info, 0, sizeof(info));
4889         info.name = ptr_to_u64(name);
4890         info.name_len = sizeof(name);
4891
4892         /* check that BPF_OBJ_GET_INFO_BY_FD supports specifying name pointer;
4893          * kernel's module BTF support coincides with support for
4894          * name/name_len fields in struct bpf_btf_info.
4895          */
4896         err = bpf_btf_get_info_by_fd(fd, &info, &len);
4897         close(fd);
4898         return !err;
4899 }
4900
4901 static int probe_perf_link(void)
4902 {
4903         struct bpf_insn insns[] = {
4904                 BPF_MOV64_IMM(BPF_REG_0, 0),
4905                 BPF_EXIT_INSN(),
4906         };
4907         int prog_fd, link_fd, err;
4908
4909         prog_fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL",
4910                                 insns, ARRAY_SIZE(insns), NULL);
4911         if (prog_fd < 0)
4912                 return -errno;
4913
4914         /* use invalid perf_event FD to get EBADF, if link is supported;
4915          * otherwise EINVAL should be returned
4916          */
4917         link_fd = bpf_link_create(prog_fd, -1, BPF_PERF_EVENT, NULL);
4918         err = -errno; /* close() can clobber errno */
4919
4920         if (link_fd >= 0)
4921                 close(link_fd);
4922         close(prog_fd);
4923
4924         return link_fd < 0 && err == -EBADF;
4925 }
4926
4927 static int probe_uprobe_multi_link(void)
4928 {
4929         LIBBPF_OPTS(bpf_prog_load_opts, load_opts,
4930                 .expected_attach_type = BPF_TRACE_UPROBE_MULTI,
4931         );
4932         LIBBPF_OPTS(bpf_link_create_opts, link_opts);
4933         struct bpf_insn insns[] = {
4934                 BPF_MOV64_IMM(BPF_REG_0, 0),
4935                 BPF_EXIT_INSN(),
4936         };
4937         int prog_fd, link_fd, err;
4938         unsigned long offset = 0;
4939
4940         prog_fd = bpf_prog_load(BPF_PROG_TYPE_KPROBE, NULL, "GPL",
4941                                 insns, ARRAY_SIZE(insns), &load_opts);
4942         if (prog_fd < 0)
4943                 return -errno;
4944
4945         /* Creating uprobe in '/' binary should fail with -EBADF. */
4946         link_opts.uprobe_multi.path = "/";
4947         link_opts.uprobe_multi.offsets = &offset;
4948         link_opts.uprobe_multi.cnt = 1;
4949
4950         link_fd = bpf_link_create(prog_fd, -1, BPF_TRACE_UPROBE_MULTI, &link_opts);
4951         err = -errno; /* close() can clobber errno */
4952
4953         if (link_fd >= 0)
4954                 close(link_fd);
4955         close(prog_fd);
4956
4957         return link_fd < 0 && err == -EBADF;
4958 }
4959
4960 static int probe_kern_bpf_cookie(void)
4961 {
4962         struct bpf_insn insns[] = {
4963                 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_get_attach_cookie),
4964                 BPF_EXIT_INSN(),
4965         };
4966         int ret, insn_cnt = ARRAY_SIZE(insns);
4967
4968         ret = bpf_prog_load(BPF_PROG_TYPE_KPROBE, NULL, "GPL", insns, insn_cnt, NULL);
4969         return probe_fd(ret);
4970 }
4971
4972 static int probe_kern_btf_enum64(void)
4973 {
4974         static const char strs[] = "\0enum64";
4975         __u32 types[] = {
4976                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_ENUM64, 0, 0), 8),
4977         };
4978
4979         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
4980                                              strs, sizeof(strs)));
4981 }
4982
4983 static int probe_kern_syscall_wrapper(void);
4984
4985 enum kern_feature_result {
4986         FEAT_UNKNOWN = 0,
4987         FEAT_SUPPORTED = 1,
4988         FEAT_MISSING = 2,
4989 };
4990
4991 typedef int (*feature_probe_fn)(void);
4992
4993 static struct kern_feature_desc {
4994         const char *desc;
4995         feature_probe_fn probe;
4996         enum kern_feature_result res;
4997 } feature_probes[__FEAT_CNT] = {
4998         [FEAT_PROG_NAME] = {
4999                 "BPF program name", probe_kern_prog_name,
5000         },
5001         [FEAT_GLOBAL_DATA] = {
5002                 "global variables", probe_kern_global_data,
5003         },
5004         [FEAT_BTF] = {
5005                 "minimal BTF", probe_kern_btf,
5006         },
5007         [FEAT_BTF_FUNC] = {
5008                 "BTF functions", probe_kern_btf_func,
5009         },
5010         [FEAT_BTF_GLOBAL_FUNC] = {
5011                 "BTF global function", probe_kern_btf_func_global,
5012         },
5013         [FEAT_BTF_DATASEC] = {
5014                 "BTF data section and variable", probe_kern_btf_datasec,
5015         },
5016         [FEAT_ARRAY_MMAP] = {
5017                 "ARRAY map mmap()", probe_kern_array_mmap,
5018         },
5019         [FEAT_EXP_ATTACH_TYPE] = {
5020                 "BPF_PROG_LOAD expected_attach_type attribute",
5021                 probe_kern_exp_attach_type,
5022         },
5023         [FEAT_PROBE_READ_KERN] = {
5024                 "bpf_probe_read_kernel() helper", probe_kern_probe_read_kernel,
5025         },
5026         [FEAT_PROG_BIND_MAP] = {
5027                 "BPF_PROG_BIND_MAP support", probe_prog_bind_map,
5028         },
5029         [FEAT_MODULE_BTF] = {
5030                 "module BTF support", probe_module_btf,
5031         },
5032         [FEAT_BTF_FLOAT] = {
5033                 "BTF_KIND_FLOAT support", probe_kern_btf_float,
5034         },
5035         [FEAT_PERF_LINK] = {
5036                 "BPF perf link support", probe_perf_link,
5037         },
5038         [FEAT_BTF_DECL_TAG] = {
5039                 "BTF_KIND_DECL_TAG support", probe_kern_btf_decl_tag,
5040         },
5041         [FEAT_BTF_TYPE_TAG] = {
5042                 "BTF_KIND_TYPE_TAG support", probe_kern_btf_type_tag,
5043         },
5044         [FEAT_MEMCG_ACCOUNT] = {
5045                 "memcg-based memory accounting", probe_memcg_account,
5046         },
5047         [FEAT_BPF_COOKIE] = {
5048                 "BPF cookie support", probe_kern_bpf_cookie,
5049         },
5050         [FEAT_BTF_ENUM64] = {
5051                 "BTF_KIND_ENUM64 support", probe_kern_btf_enum64,
5052         },
5053         [FEAT_SYSCALL_WRAPPER] = {
5054                 "Kernel using syscall wrapper", probe_kern_syscall_wrapper,
5055         },
5056         [FEAT_UPROBE_MULTI_LINK] = {
5057                 "BPF multi-uprobe link support", probe_uprobe_multi_link,
5058         },
5059 };
5060
5061 bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id)
5062 {
5063         struct kern_feature_desc *feat = &feature_probes[feat_id];
5064         int ret;
5065
5066         if (obj && obj->gen_loader)
5067                 /* To generate loader program assume the latest kernel
5068                  * to avoid doing extra prog_load, map_create syscalls.
5069                  */
5070                 return true;
5071
5072         if (READ_ONCE(feat->res) == FEAT_UNKNOWN) {
5073                 ret = feat->probe();
5074                 if (ret > 0) {
5075                         WRITE_ONCE(feat->res, FEAT_SUPPORTED);
5076                 } else if (ret == 0) {
5077                         WRITE_ONCE(feat->res, FEAT_MISSING);
5078                 } else {
5079                         pr_warn("Detection of kernel %s support failed: %d\n", feat->desc, ret);
5080                         WRITE_ONCE(feat->res, FEAT_MISSING);
5081                 }
5082         }
5083
5084         return READ_ONCE(feat->res) == FEAT_SUPPORTED;
5085 }
5086
5087 static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
5088 {
5089         struct bpf_map_info map_info;
5090         char msg[STRERR_BUFSIZE];
5091         __u32 map_info_len = sizeof(map_info);
5092         int err;
5093
5094         memset(&map_info, 0, map_info_len);
5095         err = bpf_map_get_info_by_fd(map_fd, &map_info, &map_info_len);
5096         if (err && errno == EINVAL)
5097                 err = bpf_get_map_info_from_fdinfo(map_fd, &map_info);
5098         if (err) {
5099                 pr_warn("failed to get map info for map FD %d: %s\n", map_fd,
5100                         libbpf_strerror_r(errno, msg, sizeof(msg)));
5101                 return false;
5102         }
5103
5104         return (map_info.type == map->def.type &&
5105                 map_info.key_size == map->def.key_size &&
5106                 map_info.value_size == map->def.value_size &&
5107                 map_info.max_entries == map->def.max_entries &&
5108                 map_info.map_flags == map->def.map_flags &&
5109                 map_info.map_extra == map->map_extra);
5110 }
5111
5112 static int
5113 bpf_object__reuse_map(struct bpf_map *map)
5114 {
5115         char *cp, errmsg[STRERR_BUFSIZE];
5116         int err, pin_fd;
5117
5118         pin_fd = bpf_obj_get(map->pin_path);
5119         if (pin_fd < 0) {
5120                 err = -errno;
5121                 if (err == -ENOENT) {
5122                         pr_debug("found no pinned map to reuse at '%s'\n",
5123                                  map->pin_path);
5124                         return 0;
5125                 }
5126
5127                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
5128                 pr_warn("couldn't retrieve pinned map '%s': %s\n",
5129                         map->pin_path, cp);
5130                 return err;
5131         }
5132
5133         if (!map_is_reuse_compat(map, pin_fd)) {
5134                 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
5135                         map->pin_path);
5136                 close(pin_fd);
5137                 return -EINVAL;
5138         }
5139
5140         err = bpf_map__reuse_fd(map, pin_fd);
5141         close(pin_fd);
5142         if (err)
5143                 return err;
5144
5145         map->pinned = true;
5146         pr_debug("reused pinned map at '%s'\n", map->pin_path);
5147
5148         return 0;
5149 }
5150
5151 static int
5152 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
5153 {
5154         enum libbpf_map_type map_type = map->libbpf_type;
5155         char *cp, errmsg[STRERR_BUFSIZE];
5156         int err, zero = 0;
5157
5158         if (obj->gen_loader) {
5159                 bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps,
5160                                          map->mmaped, map->def.value_size);
5161                 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG)
5162                         bpf_gen__map_freeze(obj->gen_loader, map - obj->maps);
5163                 return 0;
5164         }
5165         err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
5166         if (err) {
5167                 err = -errno;
5168                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5169                 pr_warn("Error setting initial map(%s) contents: %s\n",
5170                         map->name, cp);
5171                 return err;
5172         }
5173
5174         /* Freeze .rodata and .kconfig map as read-only from syscall side. */
5175         if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
5176                 err = bpf_map_freeze(map->fd);
5177                 if (err) {
5178                         err = -errno;
5179                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5180                         pr_warn("Error freezing map(%s) as read-only: %s\n",
5181                                 map->name, cp);
5182                         return err;
5183                 }
5184         }
5185         return 0;
5186 }
5187
5188 static void bpf_map__destroy(struct bpf_map *map);
5189
5190 static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner)
5191 {
5192         LIBBPF_OPTS(bpf_map_create_opts, create_attr);
5193         struct bpf_map_def *def = &map->def;
5194         const char *map_name = NULL;
5195         int err = 0;
5196
5197         if (kernel_supports(obj, FEAT_PROG_NAME))
5198                 map_name = map->name;
5199         create_attr.map_ifindex = map->map_ifindex;
5200         create_attr.map_flags = def->map_flags;
5201         create_attr.numa_node = map->numa_node;
5202         create_attr.map_extra = map->map_extra;
5203
5204         if (bpf_map__is_struct_ops(map))
5205                 create_attr.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
5206
5207         if (obj->btf && btf__fd(obj->btf) >= 0) {
5208                 create_attr.btf_fd = btf__fd(obj->btf);
5209                 create_attr.btf_key_type_id = map->btf_key_type_id;
5210                 create_attr.btf_value_type_id = map->btf_value_type_id;
5211         }
5212
5213         if (bpf_map_type__is_map_in_map(def->type)) {
5214                 if (map->inner_map) {
5215                         err = bpf_object__create_map(obj, map->inner_map, true);
5216                         if (err) {
5217                                 pr_warn("map '%s': failed to create inner map: %d\n",
5218                                         map->name, err);
5219                                 return err;
5220                         }
5221                         map->inner_map_fd = bpf_map__fd(map->inner_map);
5222                 }
5223                 if (map->inner_map_fd >= 0)
5224                         create_attr.inner_map_fd = map->inner_map_fd;
5225         }
5226
5227         switch (def->type) {
5228         case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
5229         case BPF_MAP_TYPE_CGROUP_ARRAY:
5230         case BPF_MAP_TYPE_STACK_TRACE:
5231         case BPF_MAP_TYPE_ARRAY_OF_MAPS:
5232         case BPF_MAP_TYPE_HASH_OF_MAPS:
5233         case BPF_MAP_TYPE_DEVMAP:
5234         case BPF_MAP_TYPE_DEVMAP_HASH:
5235         case BPF_MAP_TYPE_CPUMAP:
5236         case BPF_MAP_TYPE_XSKMAP:
5237         case BPF_MAP_TYPE_SOCKMAP:
5238         case BPF_MAP_TYPE_SOCKHASH:
5239         case BPF_MAP_TYPE_QUEUE:
5240         case BPF_MAP_TYPE_STACK:
5241                 create_attr.btf_fd = 0;
5242                 create_attr.btf_key_type_id = 0;
5243                 create_attr.btf_value_type_id = 0;
5244                 map->btf_key_type_id = 0;
5245                 map->btf_value_type_id = 0;
5246         default:
5247                 break;
5248         }
5249
5250         if (obj->gen_loader) {
5251                 bpf_gen__map_create(obj->gen_loader, def->type, map_name,
5252                                     def->key_size, def->value_size, def->max_entries,
5253                                     &create_attr, is_inner ? -1 : map - obj->maps);
5254                 /* Pretend to have valid FD to pass various fd >= 0 checks.
5255                  * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
5256                  */
5257                 map->fd = 0;
5258         } else {
5259                 map->fd = bpf_map_create(def->type, map_name,
5260                                          def->key_size, def->value_size,
5261                                          def->max_entries, &create_attr);
5262         }
5263         if (map->fd < 0 && (create_attr.btf_key_type_id ||
5264                             create_attr.btf_value_type_id)) {
5265                 char *cp, errmsg[STRERR_BUFSIZE];
5266
5267                 err = -errno;
5268                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5269                 pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
5270                         map->name, cp, err);
5271                 create_attr.btf_fd = 0;
5272                 create_attr.btf_key_type_id = 0;
5273                 create_attr.btf_value_type_id = 0;
5274                 map->btf_key_type_id = 0;
5275                 map->btf_value_type_id = 0;
5276                 map->fd = bpf_map_create(def->type, map_name,
5277                                          def->key_size, def->value_size,
5278                                          def->max_entries, &create_attr);
5279         }
5280
5281         err = map->fd < 0 ? -errno : 0;
5282
5283         if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) {
5284                 if (obj->gen_loader)
5285                         map->inner_map->fd = -1;
5286                 bpf_map__destroy(map->inner_map);
5287                 zfree(&map->inner_map);
5288         }
5289
5290         return err;
5291 }
5292
5293 static int init_map_in_map_slots(struct bpf_object *obj, struct bpf_map *map)
5294 {
5295         const struct bpf_map *targ_map;
5296         unsigned int i;
5297         int fd, err = 0;
5298
5299         for (i = 0; i < map->init_slots_sz; i++) {
5300                 if (!map->init_slots[i])
5301                         continue;
5302
5303                 targ_map = map->init_slots[i];
5304                 fd = bpf_map__fd(targ_map);
5305
5306                 if (obj->gen_loader) {
5307                         bpf_gen__populate_outer_map(obj->gen_loader,
5308                                                     map - obj->maps, i,
5309                                                     targ_map - obj->maps);
5310                 } else {
5311                         err = bpf_map_update_elem(map->fd, &i, &fd, 0);
5312                 }
5313                 if (err) {
5314                         err = -errno;
5315                         pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n",
5316                                 map->name, i, targ_map->name, fd, err);
5317                         return err;
5318                 }
5319                 pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n",
5320                          map->name, i, targ_map->name, fd);
5321         }
5322
5323         zfree(&map->init_slots);
5324         map->init_slots_sz = 0;
5325
5326         return 0;
5327 }
5328
5329 static int init_prog_array_slots(struct bpf_object *obj, struct bpf_map *map)
5330 {
5331         const struct bpf_program *targ_prog;
5332         unsigned int i;
5333         int fd, err;
5334
5335         if (obj->gen_loader)
5336                 return -ENOTSUP;
5337
5338         for (i = 0; i < map->init_slots_sz; i++) {
5339                 if (!map->init_slots[i])
5340                         continue;
5341
5342                 targ_prog = map->init_slots[i];
5343                 fd = bpf_program__fd(targ_prog);
5344
5345                 err = bpf_map_update_elem(map->fd, &i, &fd, 0);
5346                 if (err) {
5347                         err = -errno;
5348                         pr_warn("map '%s': failed to initialize slot [%d] to prog '%s' fd=%d: %d\n",
5349                                 map->name, i, targ_prog->name, fd, err);
5350                         return err;
5351                 }
5352                 pr_debug("map '%s': slot [%d] set to prog '%s' fd=%d\n",
5353                          map->name, i, targ_prog->name, fd);
5354         }
5355
5356         zfree(&map->init_slots);
5357         map->init_slots_sz = 0;
5358
5359         return 0;
5360 }
5361
5362 static int bpf_object_init_prog_arrays(struct bpf_object *obj)
5363 {
5364         struct bpf_map *map;
5365         int i, err;
5366
5367         for (i = 0; i < obj->nr_maps; i++) {
5368                 map = &obj->maps[i];
5369
5370                 if (!map->init_slots_sz || map->def.type != BPF_MAP_TYPE_PROG_ARRAY)
5371                         continue;
5372
5373                 err = init_prog_array_slots(obj, map);
5374                 if (err < 0) {
5375                         zclose(map->fd);
5376                         return err;
5377                 }
5378         }
5379         return 0;
5380 }
5381
5382 static int map_set_def_max_entries(struct bpf_map *map)
5383 {
5384         if (map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !map->def.max_entries) {
5385                 int nr_cpus;
5386
5387                 nr_cpus = libbpf_num_possible_cpus();
5388                 if (nr_cpus < 0) {
5389                         pr_warn("map '%s': failed to determine number of system CPUs: %d\n",
5390                                 map->name, nr_cpus);
5391                         return nr_cpus;
5392                 }
5393                 pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus);
5394                 map->def.max_entries = nr_cpus;
5395         }
5396
5397         return 0;
5398 }
5399
5400 static int
5401 bpf_object__create_maps(struct bpf_object *obj)
5402 {
5403         struct bpf_map *map;
5404         char *cp, errmsg[STRERR_BUFSIZE];
5405         unsigned int i, j;
5406         int err;
5407         bool retried;
5408
5409         for (i = 0; i < obj->nr_maps; i++) {
5410                 map = &obj->maps[i];
5411
5412                 /* To support old kernels, we skip creating global data maps
5413                  * (.rodata, .data, .kconfig, etc); later on, during program
5414                  * loading, if we detect that at least one of the to-be-loaded
5415                  * programs is referencing any global data map, we'll error
5416                  * out with program name and relocation index logged.
5417                  * This approach allows to accommodate Clang emitting
5418                  * unnecessary .rodata.str1.1 sections for string literals,
5419                  * but also it allows to have CO-RE applications that use
5420                  * global variables in some of BPF programs, but not others.
5421                  * If those global variable-using programs are not loaded at
5422                  * runtime due to bpf_program__set_autoload(prog, false),
5423                  * bpf_object loading will succeed just fine even on old
5424                  * kernels.
5425                  */
5426                 if (bpf_map__is_internal(map) && !kernel_supports(obj, FEAT_GLOBAL_DATA))
5427                         map->autocreate = false;
5428
5429                 if (!map->autocreate) {
5430                         pr_debug("map '%s': skipped auto-creating...\n", map->name);
5431                         continue;
5432                 }
5433
5434                 err = map_set_def_max_entries(map);
5435                 if (err)
5436                         goto err_out;
5437
5438                 retried = false;
5439 retry:
5440                 if (map->pin_path) {
5441                         err = bpf_object__reuse_map(map);
5442                         if (err) {
5443                                 pr_warn("map '%s': error reusing pinned map\n",
5444                                         map->name);
5445                                 goto err_out;
5446                         }
5447                         if (retried && map->fd < 0) {
5448                                 pr_warn("map '%s': cannot find pinned map\n",
5449                                         map->name);
5450                                 err = -ENOENT;
5451                                 goto err_out;
5452                         }
5453                 }
5454
5455                 if (map->fd >= 0) {
5456                         pr_debug("map '%s': skipping creation (preset fd=%d)\n",
5457                                  map->name, map->fd);
5458                 } else {
5459                         err = bpf_object__create_map(obj, map, false);
5460                         if (err)
5461                                 goto err_out;
5462
5463                         pr_debug("map '%s': created successfully, fd=%d\n",
5464                                  map->name, map->fd);
5465
5466                         if (bpf_map__is_internal(map)) {
5467                                 err = bpf_object__populate_internal_map(obj, map);
5468                                 if (err < 0) {
5469                                         zclose(map->fd);
5470                                         goto err_out;
5471                                 }
5472                         }
5473
5474                         if (map->init_slots_sz && map->def.type != BPF_MAP_TYPE_PROG_ARRAY) {
5475                                 err = init_map_in_map_slots(obj, map);
5476                                 if (err < 0) {
5477                                         zclose(map->fd);
5478                                         goto err_out;
5479                                 }
5480                         }
5481                 }
5482
5483                 if (map->pin_path && !map->pinned) {
5484                         err = bpf_map__pin(map, NULL);
5485                         if (err) {
5486                                 zclose(map->fd);
5487                                 if (!retried && err == -EEXIST) {
5488                                         retried = true;
5489                                         goto retry;
5490                                 }
5491                                 pr_warn("map '%s': failed to auto-pin at '%s': %d\n",
5492                                         map->name, map->pin_path, err);
5493                                 goto err_out;
5494                         }
5495                 }
5496         }
5497
5498         return 0;
5499
5500 err_out:
5501         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
5502         pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err);
5503         pr_perm_msg(err);
5504         for (j = 0; j < i; j++)
5505                 zclose(obj->maps[j].fd);
5506         return err;
5507 }
5508
5509 static bool bpf_core_is_flavor_sep(const char *s)
5510 {
5511         /* check X___Y name pattern, where X and Y are not underscores */
5512         return s[0] != '_' &&                                 /* X */
5513                s[1] == '_' && s[2] == '_' && s[3] == '_' &&   /* ___ */
5514                s[4] != '_';                                   /* Y */
5515 }
5516
5517 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
5518  * before last triple underscore. Struct name part after last triple
5519  * underscore is ignored by BPF CO-RE relocation during relocation matching.
5520  */
5521 size_t bpf_core_essential_name_len(const char *name)
5522 {
5523         size_t n = strlen(name);
5524         int i;
5525
5526         for (i = n - 5; i >= 0; i--) {
5527                 if (bpf_core_is_flavor_sep(name + i))
5528                         return i + 1;
5529         }
5530         return n;
5531 }
5532
5533 void bpf_core_free_cands(struct bpf_core_cand_list *cands)
5534 {
5535         if (!cands)
5536                 return;
5537
5538         free(cands->cands);
5539         free(cands);
5540 }
5541
5542 int bpf_core_add_cands(struct bpf_core_cand *local_cand,
5543                        size_t local_essent_len,
5544                        const struct btf *targ_btf,
5545                        const char *targ_btf_name,
5546                        int targ_start_id,
5547                        struct bpf_core_cand_list *cands)
5548 {
5549         struct bpf_core_cand *new_cands, *cand;
5550         const struct btf_type *t, *local_t;
5551         const char *targ_name, *local_name;
5552         size_t targ_essent_len;
5553         int n, i;
5554
5555         local_t = btf__type_by_id(local_cand->btf, local_cand->id);
5556         local_name = btf__str_by_offset(local_cand->btf, local_t->name_off);
5557
5558         n = btf__type_cnt(targ_btf);
5559         for (i = targ_start_id; i < n; i++) {
5560                 t = btf__type_by_id(targ_btf, i);
5561                 if (!btf_kind_core_compat(t, local_t))
5562                         continue;
5563
5564                 targ_name = btf__name_by_offset(targ_btf, t->name_off);
5565                 if (str_is_empty(targ_name))
5566                         continue;
5567
5568                 targ_essent_len = bpf_core_essential_name_len(targ_name);
5569                 if (targ_essent_len != local_essent_len)
5570                         continue;
5571
5572                 if (strncmp(local_name, targ_name, local_essent_len) != 0)
5573                         continue;
5574
5575                 pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n",
5576                          local_cand->id, btf_kind_str(local_t),
5577                          local_name, i, btf_kind_str(t), targ_name,
5578                          targ_btf_name);
5579                 new_cands = libbpf_reallocarray(cands->cands, cands->len + 1,
5580                                               sizeof(*cands->cands));
5581                 if (!new_cands)
5582                         return -ENOMEM;
5583
5584                 cand = &new_cands[cands->len];
5585                 cand->btf = targ_btf;
5586                 cand->id = i;
5587
5588                 cands->cands = new_cands;
5589                 cands->len++;
5590         }
5591         return 0;
5592 }
5593
5594 static int load_module_btfs(struct bpf_object *obj)
5595 {
5596         struct bpf_btf_info info;
5597         struct module_btf *mod_btf;
5598         struct btf *btf;
5599         char name[64];
5600         __u32 id = 0, len;
5601         int err, fd;
5602
5603         if (obj->btf_modules_loaded)
5604                 return 0;
5605
5606         if (obj->gen_loader)
5607                 return 0;
5608
5609         /* don't do this again, even if we find no module BTFs */
5610         obj->btf_modules_loaded = true;
5611
5612         /* kernel too old to support module BTFs */
5613         if (!kernel_supports(obj, FEAT_MODULE_BTF))
5614                 return 0;
5615
5616         while (true) {
5617                 err = bpf_btf_get_next_id(id, &id);
5618                 if (err && errno == ENOENT)
5619                         return 0;
5620                 if (err && errno == EPERM) {
5621                         pr_debug("skipping module BTFs loading, missing privileges\n");
5622                         return 0;
5623                 }
5624                 if (err) {
5625                         err = -errno;
5626                         pr_warn("failed to iterate BTF objects: %d\n", err);
5627                         return err;
5628                 }
5629
5630                 fd = bpf_btf_get_fd_by_id(id);
5631                 if (fd < 0) {
5632                         if (errno == ENOENT)
5633                                 continue; /* expected race: BTF was unloaded */
5634                         err = -errno;
5635                         pr_warn("failed to get BTF object #%d FD: %d\n", id, err);
5636                         return err;
5637                 }
5638
5639                 len = sizeof(info);
5640                 memset(&info, 0, sizeof(info));
5641                 info.name = ptr_to_u64(name);
5642                 info.name_len = sizeof(name);
5643
5644                 err = bpf_btf_get_info_by_fd(fd, &info, &len);
5645                 if (err) {
5646                         err = -errno;
5647                         pr_warn("failed to get BTF object #%d info: %d\n", id, err);
5648                         goto err_out;
5649                 }
5650
5651                 /* ignore non-module BTFs */
5652                 if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) {
5653                         close(fd);
5654                         continue;
5655                 }
5656
5657                 btf = btf_get_from_fd(fd, obj->btf_vmlinux);
5658                 err = libbpf_get_error(btf);
5659                 if (err) {
5660                         pr_warn("failed to load module [%s]'s BTF object #%d: %d\n",
5661                                 name, id, err);
5662                         goto err_out;
5663                 }
5664
5665                 err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap,
5666                                         sizeof(*obj->btf_modules), obj->btf_module_cnt + 1);
5667                 if (err)
5668                         goto err_out;
5669
5670                 mod_btf = &obj->btf_modules[obj->btf_module_cnt++];
5671
5672                 mod_btf->btf = btf;
5673                 mod_btf->id = id;
5674                 mod_btf->fd = fd;
5675                 mod_btf->name = strdup(name);
5676                 if (!mod_btf->name) {
5677                         err = -ENOMEM;
5678                         goto err_out;
5679                 }
5680                 continue;
5681
5682 err_out:
5683                 close(fd);
5684                 return err;
5685         }
5686
5687         return 0;
5688 }
5689
5690 static struct bpf_core_cand_list *
5691 bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id)
5692 {
5693         struct bpf_core_cand local_cand = {};
5694         struct bpf_core_cand_list *cands;
5695         const struct btf *main_btf;
5696         const struct btf_type *local_t;
5697         const char *local_name;
5698         size_t local_essent_len;
5699         int err, i;
5700
5701         local_cand.btf = local_btf;
5702         local_cand.id = local_type_id;
5703         local_t = btf__type_by_id(local_btf, local_type_id);
5704         if (!local_t)
5705                 return ERR_PTR(-EINVAL);
5706
5707         local_name = btf__name_by_offset(local_btf, local_t->name_off);
5708         if (str_is_empty(local_name))
5709                 return ERR_PTR(-EINVAL);
5710         local_essent_len = bpf_core_essential_name_len(local_name);
5711
5712         cands = calloc(1, sizeof(*cands));
5713         if (!cands)
5714                 return ERR_PTR(-ENOMEM);
5715
5716         /* Attempt to find target candidates in vmlinux BTF first */
5717         main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux;
5718         err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands);
5719         if (err)
5720                 goto err_out;
5721
5722         /* if vmlinux BTF has any candidate, don't got for module BTFs */
5723         if (cands->len)
5724                 return cands;
5725
5726         /* if vmlinux BTF was overridden, don't attempt to load module BTFs */
5727         if (obj->btf_vmlinux_override)
5728                 return cands;
5729
5730         /* now look through module BTFs, trying to still find candidates */
5731         err = load_module_btfs(obj);
5732         if (err)
5733                 goto err_out;
5734
5735         for (i = 0; i < obj->btf_module_cnt; i++) {
5736                 err = bpf_core_add_cands(&local_cand, local_essent_len,
5737                                          obj->btf_modules[i].btf,
5738                                          obj->btf_modules[i].name,
5739                                          btf__type_cnt(obj->btf_vmlinux),
5740                                          cands);
5741                 if (err)
5742                         goto err_out;
5743         }
5744
5745         return cands;
5746 err_out:
5747         bpf_core_free_cands(cands);
5748         return ERR_PTR(err);
5749 }
5750
5751 /* Check local and target types for compatibility. This check is used for
5752  * type-based CO-RE relocations and follow slightly different rules than
5753  * field-based relocations. This function assumes that root types were already
5754  * checked for name match. Beyond that initial root-level name check, names
5755  * are completely ignored. Compatibility rules are as follows:
5756  *   - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but
5757  *     kind should match for local and target types (i.e., STRUCT is not
5758  *     compatible with UNION);
5759  *   - for ENUMs, the size is ignored;
5760  *   - for INT, size and signedness are ignored;
5761  *   - for ARRAY, dimensionality is ignored, element types are checked for
5762  *     compatibility recursively;
5763  *   - CONST/VOLATILE/RESTRICT modifiers are ignored;
5764  *   - TYPEDEFs/PTRs are compatible if types they pointing to are compatible;
5765  *   - FUNC_PROTOs are compatible if they have compatible signature: same
5766  *     number of input args and compatible return and argument types.
5767  * These rules are not set in stone and probably will be adjusted as we get
5768  * more experience with using BPF CO-RE relocations.
5769  */
5770 int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id,
5771                               const struct btf *targ_btf, __u32 targ_id)
5772 {
5773         return __bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id, 32);
5774 }
5775
5776 int bpf_core_types_match(const struct btf *local_btf, __u32 local_id,
5777                          const struct btf *targ_btf, __u32 targ_id)
5778 {
5779         return __bpf_core_types_match(local_btf, local_id, targ_btf, targ_id, false, 32);
5780 }
5781
5782 static size_t bpf_core_hash_fn(const long key, void *ctx)
5783 {
5784         return key;
5785 }
5786
5787 static bool bpf_core_equal_fn(const long k1, const long k2, void *ctx)
5788 {
5789         return k1 == k2;
5790 }
5791
5792 static int record_relo_core(struct bpf_program *prog,
5793                             const struct bpf_core_relo *core_relo, int insn_idx)
5794 {
5795         struct reloc_desc *relos, *relo;
5796
5797         relos = libbpf_reallocarray(prog->reloc_desc,
5798                                     prog->nr_reloc + 1, sizeof(*relos));
5799         if (!relos)
5800                 return -ENOMEM;
5801         relo = &relos[prog->nr_reloc];
5802         relo->type = RELO_CORE;
5803         relo->insn_idx = insn_idx;
5804         relo->core_relo = core_relo;
5805         prog->reloc_desc = relos;
5806         prog->nr_reloc++;
5807         return 0;
5808 }
5809
5810 static const struct bpf_core_relo *find_relo_core(struct bpf_program *prog, int insn_idx)
5811 {
5812         struct reloc_desc *relo;
5813         int i;
5814
5815         for (i = 0; i < prog->nr_reloc; i++) {
5816                 relo = &prog->reloc_desc[i];
5817                 if (relo->type != RELO_CORE || relo->insn_idx != insn_idx)
5818                         continue;
5819
5820                 return relo->core_relo;
5821         }
5822
5823         return NULL;
5824 }
5825
5826 static int bpf_core_resolve_relo(struct bpf_program *prog,
5827                                  const struct bpf_core_relo *relo,
5828                                  int relo_idx,
5829                                  const struct btf *local_btf,
5830                                  struct hashmap *cand_cache,
5831                                  struct bpf_core_relo_res *targ_res)
5832 {
5833         struct bpf_core_spec specs_scratch[3] = {};
5834         struct bpf_core_cand_list *cands = NULL;
5835         const char *prog_name = prog->name;
5836         const struct btf_type *local_type;
5837         const char *local_name;
5838         __u32 local_id = relo->type_id;
5839         int err;
5840
5841         local_type = btf__type_by_id(local_btf, local_id);
5842         if (!local_type)
5843                 return -EINVAL;
5844
5845         local_name = btf__name_by_offset(local_btf, local_type->name_off);
5846         if (!local_name)
5847                 return -EINVAL;
5848
5849         if (relo->kind != BPF_CORE_TYPE_ID_LOCAL &&
5850             !hashmap__find(cand_cache, local_id, &cands)) {
5851                 cands = bpf_core_find_cands(prog->obj, local_btf, local_id);
5852                 if (IS_ERR(cands)) {
5853                         pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n",
5854                                 prog_name, relo_idx, local_id, btf_kind_str(local_type),
5855                                 local_name, PTR_ERR(cands));
5856                         return PTR_ERR(cands);
5857                 }
5858                 err = hashmap__set(cand_cache, local_id, cands, NULL, NULL);
5859                 if (err) {
5860                         bpf_core_free_cands(cands);
5861                         return err;
5862                 }
5863         }
5864
5865         return bpf_core_calc_relo_insn(prog_name, relo, relo_idx, local_btf, cands, specs_scratch,
5866                                        targ_res);
5867 }
5868
5869 static int
5870 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
5871 {
5872         const struct btf_ext_info_sec *sec;
5873         struct bpf_core_relo_res targ_res;
5874         const struct bpf_core_relo *rec;
5875         const struct btf_ext_info *seg;
5876         struct hashmap_entry *entry;
5877         struct hashmap *cand_cache = NULL;
5878         struct bpf_program *prog;
5879         struct bpf_insn *insn;
5880         const char *sec_name;
5881         int i, err = 0, insn_idx, sec_idx, sec_num;
5882
5883         if (obj->btf_ext->core_relo_info.len == 0)
5884                 return 0;
5885
5886         if (targ_btf_path) {
5887                 obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL);
5888                 err = libbpf_get_error(obj->btf_vmlinux_override);
5889                 if (err) {
5890                         pr_warn("failed to parse target BTF: %d\n", err);
5891                         return err;
5892                 }
5893         }
5894
5895         cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
5896         if (IS_ERR(cand_cache)) {
5897                 err = PTR_ERR(cand_cache);
5898                 goto out;
5899         }
5900
5901         seg = &obj->btf_ext->core_relo_info;
5902         sec_num = 0;
5903         for_each_btf_ext_sec(seg, sec) {
5904                 sec_idx = seg->sec_idxs[sec_num];
5905                 sec_num++;
5906
5907                 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
5908                 if (str_is_empty(sec_name)) {
5909                         err = -EINVAL;
5910                         goto out;
5911                 }
5912
5913                 pr_debug("sec '%s': found %d CO-RE relocations\n", sec_name, sec->num_info);
5914
5915                 for_each_btf_ext_rec(seg, sec, i, rec) {
5916                         if (rec->insn_off % BPF_INSN_SZ)
5917                                 return -EINVAL;
5918                         insn_idx = rec->insn_off / BPF_INSN_SZ;
5919                         prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
5920                         if (!prog) {
5921                                 /* When __weak subprog is "overridden" by another instance
5922                                  * of the subprog from a different object file, linker still
5923                                  * appends all the .BTF.ext info that used to belong to that
5924                                  * eliminated subprogram.
5925                                  * This is similar to what x86-64 linker does for relocations.
5926                                  * So just ignore such relocations just like we ignore
5927                                  * subprog instructions when discovering subprograms.
5928                                  */
5929                                 pr_debug("sec '%s': skipping CO-RE relocation #%d for insn #%d belonging to eliminated weak subprogram\n",
5930                                          sec_name, i, insn_idx);
5931                                 continue;
5932                         }
5933                         /* no need to apply CO-RE relocation if the program is
5934                          * not going to be loaded
5935                          */
5936                         if (!prog->autoload)
5937                                 continue;
5938
5939                         /* adjust insn_idx from section frame of reference to the local
5940                          * program's frame of reference; (sub-)program code is not yet
5941                          * relocated, so it's enough to just subtract in-section offset
5942                          */
5943                         insn_idx = insn_idx - prog->sec_insn_off;
5944                         if (insn_idx >= prog->insns_cnt)
5945                                 return -EINVAL;
5946                         insn = &prog->insns[insn_idx];
5947
5948                         err = record_relo_core(prog, rec, insn_idx);
5949                         if (err) {
5950                                 pr_warn("prog '%s': relo #%d: failed to record relocation: %d\n",
5951                                         prog->name, i, err);
5952                                 goto out;
5953                         }
5954
5955                         if (prog->obj->gen_loader)
5956                                 continue;
5957
5958                         err = bpf_core_resolve_relo(prog, rec, i, obj->btf, cand_cache, &targ_res);
5959                         if (err) {
5960                                 pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
5961                                         prog->name, i, err);
5962                                 goto out;
5963                         }
5964
5965                         err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res);
5966                         if (err) {
5967                                 pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %d\n",
5968                                         prog->name, i, insn_idx, err);
5969                                 goto out;
5970                         }
5971                 }
5972         }
5973
5974 out:
5975         /* obj->btf_vmlinux and module BTFs are freed after object load */
5976         btf__free(obj->btf_vmlinux_override);
5977         obj->btf_vmlinux_override = NULL;
5978
5979         if (!IS_ERR_OR_NULL(cand_cache)) {
5980                 hashmap__for_each_entry(cand_cache, entry, i) {
5981                         bpf_core_free_cands(entry->pvalue);
5982                 }
5983                 hashmap__free(cand_cache);
5984         }
5985         return err;
5986 }
5987
5988 /* base map load ldimm64 special constant, used also for log fixup logic */
5989 #define POISON_LDIMM64_MAP_BASE 2001000000
5990 #define POISON_LDIMM64_MAP_PFX "200100"
5991
5992 static void poison_map_ldimm64(struct bpf_program *prog, int relo_idx,
5993                                int insn_idx, struct bpf_insn *insn,
5994                                int map_idx, const struct bpf_map *map)
5995 {
5996         int i;
5997
5998         pr_debug("prog '%s': relo #%d: poisoning insn #%d that loads map #%d '%s'\n",
5999                  prog->name, relo_idx, insn_idx, map_idx, map->name);
6000
6001         /* we turn single ldimm64 into two identical invalid calls */
6002         for (i = 0; i < 2; i++) {
6003                 insn->code = BPF_JMP | BPF_CALL;
6004                 insn->dst_reg = 0;
6005                 insn->src_reg = 0;
6006                 insn->off = 0;
6007                 /* if this instruction is reachable (not a dead code),
6008                  * verifier will complain with something like:
6009                  * invalid func unknown#2001000123
6010                  * where lower 123 is map index into obj->maps[] array
6011                  */
6012                 insn->imm = POISON_LDIMM64_MAP_BASE + map_idx;
6013
6014                 insn++;
6015         }
6016 }
6017
6018 /* unresolved kfunc call special constant, used also for log fixup logic */
6019 #define POISON_CALL_KFUNC_BASE 2002000000
6020 #define POISON_CALL_KFUNC_PFX "2002"
6021
6022 static void poison_kfunc_call(struct bpf_program *prog, int relo_idx,
6023                               int insn_idx, struct bpf_insn *insn,
6024                               int ext_idx, const struct extern_desc *ext)
6025 {
6026         pr_debug("prog '%s': relo #%d: poisoning insn #%d that calls kfunc '%s'\n",
6027                  prog->name, relo_idx, insn_idx, ext->name);
6028
6029         /* we turn kfunc call into invalid helper call with identifiable constant */
6030         insn->code = BPF_JMP | BPF_CALL;
6031         insn->dst_reg = 0;
6032         insn->src_reg = 0;
6033         insn->off = 0;
6034         /* if this instruction is reachable (not a dead code),
6035          * verifier will complain with something like:
6036          * invalid func unknown#2001000123
6037          * where lower 123 is extern index into obj->externs[] array
6038          */
6039         insn->imm = POISON_CALL_KFUNC_BASE + ext_idx;
6040 }
6041
6042 /* Relocate data references within program code:
6043  *  - map references;
6044  *  - global variable references;
6045  *  - extern references.
6046  */
6047 static int
6048 bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog)
6049 {
6050         int i;
6051
6052         for (i = 0; i < prog->nr_reloc; i++) {
6053                 struct reloc_desc *relo = &prog->reloc_desc[i];
6054                 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
6055                 const struct bpf_map *map;
6056                 struct extern_desc *ext;
6057
6058                 switch (relo->type) {
6059                 case RELO_LD64:
6060                         map = &obj->maps[relo->map_idx];
6061                         if (obj->gen_loader) {
6062                                 insn[0].src_reg = BPF_PSEUDO_MAP_IDX;
6063                                 insn[0].imm = relo->map_idx;
6064                         } else if (map->autocreate) {
6065                                 insn[0].src_reg = BPF_PSEUDO_MAP_FD;
6066                                 insn[0].imm = map->fd;
6067                         } else {
6068                                 poison_map_ldimm64(prog, i, relo->insn_idx, insn,
6069                                                    relo->map_idx, map);
6070                         }
6071                         break;
6072                 case RELO_DATA:
6073                         map = &obj->maps[relo->map_idx];
6074                         insn[1].imm = insn[0].imm + relo->sym_off;
6075                         if (obj->gen_loader) {
6076                                 insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
6077                                 insn[0].imm = relo->map_idx;
6078                         } else if (map->autocreate) {
6079                                 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
6080                                 insn[0].imm = map->fd;
6081                         } else {
6082                                 poison_map_ldimm64(prog, i, relo->insn_idx, insn,
6083                                                    relo->map_idx, map);
6084                         }
6085                         break;
6086                 case RELO_EXTERN_LD64:
6087                         ext = &obj->externs[relo->ext_idx];
6088                         if (ext->type == EXT_KCFG) {
6089                                 if (obj->gen_loader) {
6090                                         insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
6091                                         insn[0].imm = obj->kconfig_map_idx;
6092                                 } else {
6093                                         insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
6094                                         insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
6095                                 }
6096                                 insn[1].imm = ext->kcfg.data_off;
6097                         } else /* EXT_KSYM */ {
6098                                 if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */
6099                                         insn[0].src_reg = BPF_PSEUDO_BTF_ID;
6100                                         insn[0].imm = ext->ksym.kernel_btf_id;
6101                                         insn[1].imm = ext->ksym.kernel_btf_obj_fd;
6102                                 } else { /* typeless ksyms or unresolved typed ksyms */
6103                                         insn[0].imm = (__u32)ext->ksym.addr;
6104                                         insn[1].imm = ext->ksym.addr >> 32;
6105                                 }
6106                         }
6107                         break;
6108                 case RELO_EXTERN_CALL:
6109                         ext = &obj->externs[relo->ext_idx];
6110                         insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL;
6111                         if (ext->is_set) {
6112                                 insn[0].imm = ext->ksym.kernel_btf_id;
6113                                 insn[0].off = ext->ksym.btf_fd_idx;
6114                         } else { /* unresolved weak kfunc call */
6115                                 poison_kfunc_call(prog, i, relo->insn_idx, insn,
6116                                                   relo->ext_idx, ext);
6117                         }
6118                         break;
6119                 case RELO_SUBPROG_ADDR:
6120                         if (insn[0].src_reg != BPF_PSEUDO_FUNC) {
6121                                 pr_warn("prog '%s': relo #%d: bad insn\n",
6122                                         prog->name, i);
6123                                 return -EINVAL;
6124                         }
6125                         /* handled already */
6126                         break;
6127                 case RELO_CALL:
6128                         /* handled already */
6129                         break;
6130                 case RELO_CORE:
6131                         /* will be handled by bpf_program_record_relos() */
6132                         break;
6133                 default:
6134                         pr_warn("prog '%s': relo #%d: bad relo type %d\n",
6135                                 prog->name, i, relo->type);
6136                         return -EINVAL;
6137                 }
6138         }
6139
6140         return 0;
6141 }
6142
6143 static int adjust_prog_btf_ext_info(const struct bpf_object *obj,
6144                                     const struct bpf_program *prog,
6145                                     const struct btf_ext_info *ext_info,
6146                                     void **prog_info, __u32 *prog_rec_cnt,
6147                                     __u32 *prog_rec_sz)
6148 {
6149         void *copy_start = NULL, *copy_end = NULL;
6150         void *rec, *rec_end, *new_prog_info;
6151         const struct btf_ext_info_sec *sec;
6152         size_t old_sz, new_sz;
6153         int i, sec_num, sec_idx, off_adj;
6154
6155         sec_num = 0;
6156         for_each_btf_ext_sec(ext_info, sec) {
6157                 sec_idx = ext_info->sec_idxs[sec_num];
6158                 sec_num++;
6159                 if (prog->sec_idx != sec_idx)
6160                         continue;
6161
6162                 for_each_btf_ext_rec(ext_info, sec, i, rec) {
6163                         __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ;
6164
6165                         if (insn_off < prog->sec_insn_off)
6166                                 continue;
6167                         if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt)
6168                                 break;
6169
6170                         if (!copy_start)
6171                                 copy_start = rec;
6172                         copy_end = rec + ext_info->rec_size;
6173                 }
6174
6175                 if (!copy_start)
6176                         return -ENOENT;
6177
6178                 /* append func/line info of a given (sub-)program to the main
6179                  * program func/line info
6180                  */
6181                 old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size;
6182                 new_sz = old_sz + (copy_end - copy_start);
6183                 new_prog_info = realloc(*prog_info, new_sz);
6184                 if (!new_prog_info)
6185                         return -ENOMEM;
6186                 *prog_info = new_prog_info;
6187                 *prog_rec_cnt = new_sz / ext_info->rec_size;
6188                 memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start);
6189
6190                 /* Kernel instruction offsets are in units of 8-byte
6191                  * instructions, while .BTF.ext instruction offsets generated
6192                  * by Clang are in units of bytes. So convert Clang offsets
6193                  * into kernel offsets and adjust offset according to program
6194                  * relocated position.
6195                  */
6196                 off_adj = prog->sub_insn_off - prog->sec_insn_off;
6197                 rec = new_prog_info + old_sz;
6198                 rec_end = new_prog_info + new_sz;
6199                 for (; rec < rec_end; rec += ext_info->rec_size) {
6200                         __u32 *insn_off = rec;
6201
6202                         *insn_off = *insn_off / BPF_INSN_SZ + off_adj;
6203                 }
6204                 *prog_rec_sz = ext_info->rec_size;
6205                 return 0;
6206         }
6207
6208         return -ENOENT;
6209 }
6210
6211 static int
6212 reloc_prog_func_and_line_info(const struct bpf_object *obj,
6213                               struct bpf_program *main_prog,
6214                               const struct bpf_program *prog)
6215 {
6216         int err;
6217
6218         /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't
6219          * supprot func/line info
6220          */
6221         if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC))
6222                 return 0;
6223
6224         /* only attempt func info relocation if main program's func_info
6225          * relocation was successful
6226          */
6227         if (main_prog != prog && !main_prog->func_info)
6228                 goto line_info;
6229
6230         err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info,
6231                                        &main_prog->func_info,
6232                                        &main_prog->func_info_cnt,
6233                                        &main_prog->func_info_rec_size);
6234         if (err) {
6235                 if (err != -ENOENT) {
6236                         pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n",
6237                                 prog->name, err);
6238                         return err;
6239                 }
6240                 if (main_prog->func_info) {
6241                         /*
6242                          * Some info has already been found but has problem
6243                          * in the last btf_ext reloc. Must have to error out.
6244                          */
6245                         pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name);
6246                         return err;
6247                 }
6248                 /* Have problem loading the very first info. Ignore the rest. */
6249                 pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n",
6250                         prog->name);
6251         }
6252
6253 line_info:
6254         /* don't relocate line info if main program's relocation failed */
6255         if (main_prog != prog && !main_prog->line_info)
6256                 return 0;
6257
6258         err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info,
6259                                        &main_prog->line_info,
6260                                        &main_prog->line_info_cnt,
6261                                        &main_prog->line_info_rec_size);
6262         if (err) {
6263                 if (err != -ENOENT) {
6264                         pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n",
6265                                 prog->name, err);
6266                         return err;
6267                 }
6268                 if (main_prog->line_info) {
6269                         /*
6270                          * Some info has already been found but has problem
6271                          * in the last btf_ext reloc. Must have to error out.
6272                          */
6273                         pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name);
6274                         return err;
6275                 }
6276                 /* Have problem loading the very first info. Ignore the rest. */
6277                 pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n",
6278                         prog->name);
6279         }
6280         return 0;
6281 }
6282
6283 static int cmp_relo_by_insn_idx(const void *key, const void *elem)
6284 {
6285         size_t insn_idx = *(const size_t *)key;
6286         const struct reloc_desc *relo = elem;
6287
6288         if (insn_idx == relo->insn_idx)
6289                 return 0;
6290         return insn_idx < relo->insn_idx ? -1 : 1;
6291 }
6292
6293 static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx)
6294 {
6295         if (!prog->nr_reloc)
6296                 return NULL;
6297         return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc,
6298                        sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx);
6299 }
6300
6301 static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog)
6302 {
6303         int new_cnt = main_prog->nr_reloc + subprog->nr_reloc;
6304         struct reloc_desc *relos;
6305         int i;
6306
6307         if (main_prog == subprog)
6308                 return 0;
6309         relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos));
6310         /* if new count is zero, reallocarray can return a valid NULL result;
6311          * in this case the previous pointer will be freed, so we *have to*
6312          * reassign old pointer to the new value (even if it's NULL)
6313          */
6314         if (!relos && new_cnt)
6315                 return -ENOMEM;
6316         if (subprog->nr_reloc)
6317                 memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc,
6318                        sizeof(*relos) * subprog->nr_reloc);
6319
6320         for (i = main_prog->nr_reloc; i < new_cnt; i++)
6321                 relos[i].insn_idx += subprog->sub_insn_off;
6322         /* After insn_idx adjustment the 'relos' array is still sorted
6323          * by insn_idx and doesn't break bsearch.
6324          */
6325         main_prog->reloc_desc = relos;
6326         main_prog->nr_reloc = new_cnt;
6327         return 0;
6328 }
6329
6330 static int
6331 bpf_object__append_subprog_code(struct bpf_object *obj, struct bpf_program *main_prog,
6332                                 struct bpf_program *subprog)
6333 {
6334        struct bpf_insn *insns;
6335        size_t new_cnt;
6336        int err;
6337
6338        subprog->sub_insn_off = main_prog->insns_cnt;
6339
6340        new_cnt = main_prog->insns_cnt + subprog->insns_cnt;
6341        insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns));
6342        if (!insns) {
6343                pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name);
6344                return -ENOMEM;
6345        }
6346        main_prog->insns = insns;
6347        main_prog->insns_cnt = new_cnt;
6348
6349        memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns,
6350               subprog->insns_cnt * sizeof(*insns));
6351
6352        pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n",
6353                 main_prog->name, subprog->insns_cnt, subprog->name);
6354
6355        /* The subprog insns are now appended. Append its relos too. */
6356        err = append_subprog_relos(main_prog, subprog);
6357        if (err)
6358                return err;
6359        return 0;
6360 }
6361
6362 static int
6363 bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog,
6364                        struct bpf_program *prog)
6365 {
6366         size_t sub_insn_idx, insn_idx;
6367         struct bpf_program *subprog;
6368         struct reloc_desc *relo;
6369         struct bpf_insn *insn;
6370         int err;
6371
6372         err = reloc_prog_func_and_line_info(obj, main_prog, prog);
6373         if (err)
6374                 return err;
6375
6376         for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) {
6377                 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6378                 if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn))
6379                         continue;
6380
6381                 relo = find_prog_insn_relo(prog, insn_idx);
6382                 if (relo && relo->type == RELO_EXTERN_CALL)
6383                         /* kfunc relocations will be handled later
6384                          * in bpf_object__relocate_data()
6385                          */
6386                         continue;
6387                 if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) {
6388                         pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n",
6389                                 prog->name, insn_idx, relo->type);
6390                         return -LIBBPF_ERRNO__RELOC;
6391                 }
6392                 if (relo) {
6393                         /* sub-program instruction index is a combination of
6394                          * an offset of a symbol pointed to by relocation and
6395                          * call instruction's imm field; for global functions,
6396                          * call always has imm = -1, but for static functions
6397                          * relocation is against STT_SECTION and insn->imm
6398                          * points to a start of a static function
6399                          *
6400                          * for subprog addr relocation, the relo->sym_off + insn->imm is
6401                          * the byte offset in the corresponding section.
6402                          */
6403                         if (relo->type == RELO_CALL)
6404                                 sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1;
6405                         else
6406                                 sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ;
6407                 } else if (insn_is_pseudo_func(insn)) {
6408                         /*
6409                          * RELO_SUBPROG_ADDR relo is always emitted even if both
6410                          * functions are in the same section, so it shouldn't reach here.
6411                          */
6412                         pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n",
6413                                 prog->name, insn_idx);
6414                         return -LIBBPF_ERRNO__RELOC;
6415                 } else {
6416                         /* if subprogram call is to a static function within
6417                          * the same ELF section, there won't be any relocation
6418                          * emitted, but it also means there is no additional
6419                          * offset necessary, insns->imm is relative to
6420                          * instruction's original position within the section
6421                          */
6422                         sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1;
6423                 }
6424
6425                 /* we enforce that sub-programs should be in .text section */
6426                 subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx);
6427                 if (!subprog) {
6428                         pr_warn("prog '%s': no .text section found yet sub-program call exists\n",
6429                                 prog->name);
6430                         return -LIBBPF_ERRNO__RELOC;
6431                 }
6432
6433                 /* if it's the first call instruction calling into this
6434                  * subprogram (meaning this subprog hasn't been processed
6435                  * yet) within the context of current main program:
6436                  *   - append it at the end of main program's instructions blog;
6437                  *   - process is recursively, while current program is put on hold;
6438                  *   - if that subprogram calls some other not yet processes
6439                  *   subprogram, same thing will happen recursively until
6440                  *   there are no more unprocesses subprograms left to append
6441                  *   and relocate.
6442                  */
6443                 if (subprog->sub_insn_off == 0) {
6444                         err = bpf_object__append_subprog_code(obj, main_prog, subprog);
6445                         if (err)
6446                                 return err;
6447                         err = bpf_object__reloc_code(obj, main_prog, subprog);
6448                         if (err)
6449                                 return err;
6450                 }
6451
6452                 /* main_prog->insns memory could have been re-allocated, so
6453                  * calculate pointer again
6454                  */
6455                 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6456                 /* calculate correct instruction position within current main
6457                  * prog; each main prog can have a different set of
6458                  * subprograms appended (potentially in different order as
6459                  * well), so position of any subprog can be different for
6460                  * different main programs
6461                  */
6462                 insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1;
6463
6464                 pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n",
6465                          prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off);
6466         }
6467
6468         return 0;
6469 }
6470
6471 /*
6472  * Relocate sub-program calls.
6473  *
6474  * Algorithm operates as follows. Each entry-point BPF program (referred to as
6475  * main prog) is processed separately. For each subprog (non-entry functions,
6476  * that can be called from either entry progs or other subprogs) gets their
6477  * sub_insn_off reset to zero. This serves as indicator that this subprogram
6478  * hasn't been yet appended and relocated within current main prog. Once its
6479  * relocated, sub_insn_off will point at the position within current main prog
6480  * where given subprog was appended. This will further be used to relocate all
6481  * the call instructions jumping into this subprog.
6482  *
6483  * We start with main program and process all call instructions. If the call
6484  * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off
6485  * is zero), subprog instructions are appended at the end of main program's
6486  * instruction array. Then main program is "put on hold" while we recursively
6487  * process newly appended subprogram. If that subprogram calls into another
6488  * subprogram that hasn't been appended, new subprogram is appended again to
6489  * the *main* prog's instructions (subprog's instructions are always left
6490  * untouched, as they need to be in unmodified state for subsequent main progs
6491  * and subprog instructions are always sent only as part of a main prog) and
6492  * the process continues recursively. Once all the subprogs called from a main
6493  * prog or any of its subprogs are appended (and relocated), all their
6494  * positions within finalized instructions array are known, so it's easy to
6495  * rewrite call instructions with correct relative offsets, corresponding to
6496  * desired target subprog.
6497  *
6498  * Its important to realize that some subprogs might not be called from some
6499  * main prog and any of its called/used subprogs. Those will keep their
6500  * subprog->sub_insn_off as zero at all times and won't be appended to current
6501  * main prog and won't be relocated within the context of current main prog.
6502  * They might still be used from other main progs later.
6503  *
6504  * Visually this process can be shown as below. Suppose we have two main
6505  * programs mainA and mainB and BPF object contains three subprogs: subA,
6506  * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and
6507  * subC both call subB:
6508  *
6509  *        +--------+ +-------+
6510  *        |        v v       |
6511  *     +--+---+ +--+-+-+ +---+--+
6512  *     | subA | | subB | | subC |
6513  *     +--+---+ +------+ +---+--+
6514  *        ^                  ^
6515  *        |                  |
6516  *    +---+-------+   +------+----+
6517  *    |   mainA   |   |   mainB   |
6518  *    +-----------+   +-----------+
6519  *
6520  * We'll start relocating mainA, will find subA, append it and start
6521  * processing sub A recursively:
6522  *
6523  *    +-----------+------+
6524  *    |   mainA   | subA |
6525  *    +-----------+------+
6526  *
6527  * At this point we notice that subB is used from subA, so we append it and
6528  * relocate (there are no further subcalls from subB):
6529  *
6530  *    +-----------+------+------+
6531  *    |   mainA   | subA | subB |
6532  *    +-----------+------+------+
6533  *
6534  * At this point, we relocate subA calls, then go one level up and finish with
6535  * relocatin mainA calls. mainA is done.
6536  *
6537  * For mainB process is similar but results in different order. We start with
6538  * mainB and skip subA and subB, as mainB never calls them (at least
6539  * directly), but we see subC is needed, so we append and start processing it:
6540  *
6541  *    +-----------+------+
6542  *    |   mainB   | subC |
6543  *    +-----------+------+
6544  * Now we see subC needs subB, so we go back to it, append and relocate it:
6545  *
6546  *    +-----------+------+------+
6547  *    |   mainB   | subC | subB |
6548  *    +-----------+------+------+
6549  *
6550  * At this point we unwind recursion, relocate calls in subC, then in mainB.
6551  */
6552 static int
6553 bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog)
6554 {
6555         struct bpf_program *subprog;
6556         int i, err;
6557
6558         /* mark all subprogs as not relocated (yet) within the context of
6559          * current main program
6560          */
6561         for (i = 0; i < obj->nr_programs; i++) {
6562                 subprog = &obj->programs[i];
6563                 if (!prog_is_subprog(obj, subprog))
6564                         continue;
6565
6566                 subprog->sub_insn_off = 0;
6567         }
6568
6569         err = bpf_object__reloc_code(obj, prog, prog);
6570         if (err)
6571                 return err;
6572
6573         return 0;
6574 }
6575
6576 static void
6577 bpf_object__free_relocs(struct bpf_object *obj)
6578 {
6579         struct bpf_program *prog;
6580         int i;
6581
6582         /* free up relocation descriptors */
6583         for (i = 0; i < obj->nr_programs; i++) {
6584                 prog = &obj->programs[i];
6585                 zfree(&prog->reloc_desc);
6586                 prog->nr_reloc = 0;
6587         }
6588 }
6589
6590 static int cmp_relocs(const void *_a, const void *_b)
6591 {
6592         const struct reloc_desc *a = _a;
6593         const struct reloc_desc *b = _b;
6594
6595         if (a->insn_idx != b->insn_idx)
6596                 return a->insn_idx < b->insn_idx ? -1 : 1;
6597
6598         /* no two relocations should have the same insn_idx, but ... */
6599         if (a->type != b->type)
6600                 return a->type < b->type ? -1 : 1;
6601
6602         return 0;
6603 }
6604
6605 static void bpf_object__sort_relos(struct bpf_object *obj)
6606 {
6607         int i;
6608
6609         for (i = 0; i < obj->nr_programs; i++) {
6610                 struct bpf_program *p = &obj->programs[i];
6611
6612                 if (!p->nr_reloc)
6613                         continue;
6614
6615                 qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs);
6616         }
6617 }
6618
6619 static int
6620 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
6621 {
6622         struct bpf_program *prog;
6623         size_t i, j;
6624         int err;
6625
6626         if (obj->btf_ext) {
6627                 err = bpf_object__relocate_core(obj, targ_btf_path);
6628                 if (err) {
6629                         pr_warn("failed to perform CO-RE relocations: %d\n",
6630                                 err);
6631                         return err;
6632                 }
6633                 bpf_object__sort_relos(obj);
6634         }
6635
6636         /* Before relocating calls pre-process relocations and mark
6637          * few ld_imm64 instructions that points to subprogs.
6638          * Otherwise bpf_object__reloc_code() later would have to consider
6639          * all ld_imm64 insns as relocation candidates. That would
6640          * reduce relocation speed, since amount of find_prog_insn_relo()
6641          * would increase and most of them will fail to find a relo.
6642          */
6643         for (i = 0; i < obj->nr_programs; i++) {
6644                 prog = &obj->programs[i];
6645                 for (j = 0; j < prog->nr_reloc; j++) {
6646                         struct reloc_desc *relo = &prog->reloc_desc[j];
6647                         struct bpf_insn *insn = &prog->insns[relo->insn_idx];
6648
6649                         /* mark the insn, so it's recognized by insn_is_pseudo_func() */
6650                         if (relo->type == RELO_SUBPROG_ADDR)
6651                                 insn[0].src_reg = BPF_PSEUDO_FUNC;
6652                 }
6653         }
6654
6655         /* relocate subprogram calls and append used subprograms to main
6656          * programs; each copy of subprogram code needs to be relocated
6657          * differently for each main program, because its code location might
6658          * have changed.
6659          * Append subprog relos to main programs to allow data relos to be
6660          * processed after text is completely relocated.
6661          */
6662         for (i = 0; i < obj->nr_programs; i++) {
6663                 prog = &obj->programs[i];
6664                 /* sub-program's sub-calls are relocated within the context of
6665                  * its main program only
6666                  */
6667                 if (prog_is_subprog(obj, prog))
6668                         continue;
6669                 if (!prog->autoload)
6670                         continue;
6671
6672                 err = bpf_object__relocate_calls(obj, prog);
6673                 if (err) {
6674                         pr_warn("prog '%s': failed to relocate calls: %d\n",
6675                                 prog->name, err);
6676                         return err;
6677                 }
6678
6679                 /* Now, also append exception callback if it has not been done already. */
6680                 if (prog->exception_cb_idx >= 0) {
6681                         struct bpf_program *subprog = &obj->programs[prog->exception_cb_idx];
6682
6683                         /* Calling exception callback directly is disallowed, which the
6684                          * verifier will reject later. In case it was processed already,
6685                          * we can skip this step, otherwise for all other valid cases we
6686                          * have to append exception callback now.
6687                          */
6688                         if (subprog->sub_insn_off == 0) {
6689                                 err = bpf_object__append_subprog_code(obj, prog, subprog);
6690                                 if (err)
6691                                         return err;
6692                                 err = bpf_object__reloc_code(obj, prog, subprog);
6693                                 if (err)
6694                                         return err;
6695                         }
6696                 }
6697         }
6698         /* Process data relos for main programs */
6699         for (i = 0; i < obj->nr_programs; i++) {
6700                 prog = &obj->programs[i];
6701                 if (prog_is_subprog(obj, prog))
6702                         continue;
6703                 if (!prog->autoload)
6704                         continue;
6705                 err = bpf_object__relocate_data(obj, prog);
6706                 if (err) {
6707                         pr_warn("prog '%s': failed to relocate data references: %d\n",
6708                                 prog->name, err);
6709                         return err;
6710                 }
6711         }
6712
6713         return 0;
6714 }
6715
6716 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
6717                                             Elf64_Shdr *shdr, Elf_Data *data);
6718
6719 static int bpf_object__collect_map_relos(struct bpf_object *obj,
6720                                          Elf64_Shdr *shdr, Elf_Data *data)
6721 {
6722         const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *);
6723         int i, j, nrels, new_sz;
6724         const struct btf_var_secinfo *vi = NULL;
6725         const struct btf_type *sec, *var, *def;
6726         struct bpf_map *map = NULL, *targ_map = NULL;
6727         struct bpf_program *targ_prog = NULL;
6728         bool is_prog_array, is_map_in_map;
6729         const struct btf_member *member;
6730         const char *name, *mname, *type;
6731         unsigned int moff;
6732         Elf64_Sym *sym;
6733         Elf64_Rel *rel;
6734         void *tmp;
6735
6736         if (!obj->efile.btf_maps_sec_btf_id || !obj->btf)
6737                 return -EINVAL;
6738         sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id);
6739         if (!sec)
6740                 return -EINVAL;
6741
6742         nrels = shdr->sh_size / shdr->sh_entsize;
6743         for (i = 0; i < nrels; i++) {
6744                 rel = elf_rel_by_idx(data, i);
6745                 if (!rel) {
6746                         pr_warn(".maps relo #%d: failed to get ELF relo\n", i);
6747                         return -LIBBPF_ERRNO__FORMAT;
6748                 }
6749
6750                 sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
6751                 if (!sym) {
6752                         pr_warn(".maps relo #%d: symbol %zx not found\n",
6753                                 i, (size_t)ELF64_R_SYM(rel->r_info));
6754                         return -LIBBPF_ERRNO__FORMAT;
6755                 }
6756                 name = elf_sym_str(obj, sym->st_name) ?: "<?>";
6757
6758                 pr_debug(".maps relo #%d: for %zd value %zd rel->r_offset %zu name %d ('%s')\n",
6759                          i, (ssize_t)(rel->r_info >> 32), (size_t)sym->st_value,
6760                          (size_t)rel->r_offset, sym->st_name, name);
6761
6762                 for (j = 0; j < obj->nr_maps; j++) {
6763                         map = &obj->maps[j];
6764                         if (map->sec_idx != obj->efile.btf_maps_shndx)
6765                                 continue;
6766
6767                         vi = btf_var_secinfos(sec) + map->btf_var_idx;
6768                         if (vi->offset <= rel->r_offset &&
6769                             rel->r_offset + bpf_ptr_sz <= vi->offset + vi->size)
6770                                 break;
6771                 }
6772                 if (j == obj->nr_maps) {
6773                         pr_warn(".maps relo #%d: cannot find map '%s' at rel->r_offset %zu\n",
6774                                 i, name, (size_t)rel->r_offset);
6775                         return -EINVAL;
6776                 }
6777
6778                 is_map_in_map = bpf_map_type__is_map_in_map(map->def.type);
6779                 is_prog_array = map->def.type == BPF_MAP_TYPE_PROG_ARRAY;
6780                 type = is_map_in_map ? "map" : "prog";
6781                 if (is_map_in_map) {
6782                         if (sym->st_shndx != obj->efile.btf_maps_shndx) {
6783                                 pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n",
6784                                         i, name);
6785                                 return -LIBBPF_ERRNO__RELOC;
6786                         }
6787                         if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS &&
6788                             map->def.key_size != sizeof(int)) {
6789                                 pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n",
6790                                         i, map->name, sizeof(int));
6791                                 return -EINVAL;
6792                         }
6793                         targ_map = bpf_object__find_map_by_name(obj, name);
6794                         if (!targ_map) {
6795                                 pr_warn(".maps relo #%d: '%s' isn't a valid map reference\n",
6796                                         i, name);
6797                                 return -ESRCH;
6798                         }
6799                 } else if (is_prog_array) {
6800                         targ_prog = bpf_object__find_program_by_name(obj, name);
6801                         if (!targ_prog) {
6802                                 pr_warn(".maps relo #%d: '%s' isn't a valid program reference\n",
6803                                         i, name);
6804                                 return -ESRCH;
6805                         }
6806                         if (targ_prog->sec_idx != sym->st_shndx ||
6807                             targ_prog->sec_insn_off * 8 != sym->st_value ||
6808                             prog_is_subprog(obj, targ_prog)) {
6809                                 pr_warn(".maps relo #%d: '%s' isn't an entry-point program\n",
6810                                         i, name);
6811                                 return -LIBBPF_ERRNO__RELOC;
6812                         }
6813                 } else {
6814                         return -EINVAL;
6815                 }
6816
6817                 var = btf__type_by_id(obj->btf, vi->type);
6818                 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
6819                 if (btf_vlen(def) == 0)
6820                         return -EINVAL;
6821                 member = btf_members(def) + btf_vlen(def) - 1;
6822                 mname = btf__name_by_offset(obj->btf, member->name_off);
6823                 if (strcmp(mname, "values"))
6824                         return -EINVAL;
6825
6826                 moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8;
6827                 if (rel->r_offset - vi->offset < moff)
6828                         return -EINVAL;
6829
6830                 moff = rel->r_offset - vi->offset - moff;
6831                 /* here we use BPF pointer size, which is always 64 bit, as we
6832                  * are parsing ELF that was built for BPF target
6833                  */
6834                 if (moff % bpf_ptr_sz)
6835                         return -EINVAL;
6836                 moff /= bpf_ptr_sz;
6837                 if (moff >= map->init_slots_sz) {
6838                         new_sz = moff + 1;
6839                         tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz);
6840                         if (!tmp)
6841                                 return -ENOMEM;
6842                         map->init_slots = tmp;
6843                         memset(map->init_slots + map->init_slots_sz, 0,
6844                                (new_sz - map->init_slots_sz) * host_ptr_sz);
6845                         map->init_slots_sz = new_sz;
6846                 }
6847                 map->init_slots[moff] = is_map_in_map ? (void *)targ_map : (void *)targ_prog;
6848
6849                 pr_debug(".maps relo #%d: map '%s' slot [%d] points to %s '%s'\n",
6850                          i, map->name, moff, type, name);
6851         }
6852
6853         return 0;
6854 }
6855
6856 static int bpf_object__collect_relos(struct bpf_object *obj)
6857 {
6858         int i, err;
6859
6860         for (i = 0; i < obj->efile.sec_cnt; i++) {
6861                 struct elf_sec_desc *sec_desc = &obj->efile.secs[i];
6862                 Elf64_Shdr *shdr;
6863                 Elf_Data *data;
6864                 int idx;
6865
6866                 if (sec_desc->sec_type != SEC_RELO)
6867                         continue;
6868
6869                 shdr = sec_desc->shdr;
6870                 data = sec_desc->data;
6871                 idx = shdr->sh_info;
6872
6873                 if (shdr->sh_type != SHT_REL) {
6874                         pr_warn("internal error at %d\n", __LINE__);
6875                         return -LIBBPF_ERRNO__INTERNAL;
6876                 }
6877
6878                 if (idx == obj->efile.st_ops_shndx || idx == obj->efile.st_ops_link_shndx)
6879                         err = bpf_object__collect_st_ops_relos(obj, shdr, data);
6880                 else if (idx == obj->efile.btf_maps_shndx)
6881                         err = bpf_object__collect_map_relos(obj, shdr, data);
6882                 else
6883                         err = bpf_object__collect_prog_relos(obj, shdr, data);
6884                 if (err)
6885                         return err;
6886         }
6887
6888         bpf_object__sort_relos(obj);
6889         return 0;
6890 }
6891
6892 static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id)
6893 {
6894         if (BPF_CLASS(insn->code) == BPF_JMP &&
6895             BPF_OP(insn->code) == BPF_CALL &&
6896             BPF_SRC(insn->code) == BPF_K &&
6897             insn->src_reg == 0 &&
6898             insn->dst_reg == 0) {
6899                     *func_id = insn->imm;
6900                     return true;
6901         }
6902         return false;
6903 }
6904
6905 static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog)
6906 {
6907         struct bpf_insn *insn = prog->insns;
6908         enum bpf_func_id func_id;
6909         int i;
6910
6911         if (obj->gen_loader)
6912                 return 0;
6913
6914         for (i = 0; i < prog->insns_cnt; i++, insn++) {
6915                 if (!insn_is_helper_call(insn, &func_id))
6916                         continue;
6917
6918                 /* on kernels that don't yet support
6919                  * bpf_probe_read_{kernel,user}[_str] helpers, fall back
6920                  * to bpf_probe_read() which works well for old kernels
6921                  */
6922                 switch (func_id) {
6923                 case BPF_FUNC_probe_read_kernel:
6924                 case BPF_FUNC_probe_read_user:
6925                         if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
6926                                 insn->imm = BPF_FUNC_probe_read;
6927                         break;
6928                 case BPF_FUNC_probe_read_kernel_str:
6929                 case BPF_FUNC_probe_read_user_str:
6930                         if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
6931                                 insn->imm = BPF_FUNC_probe_read_str;
6932                         break;
6933                 default:
6934                         break;
6935                 }
6936         }
6937         return 0;
6938 }
6939
6940 static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
6941                                      int *btf_obj_fd, int *btf_type_id);
6942
6943 /* this is called as prog->sec_def->prog_prepare_load_fn for libbpf-supported sec_defs */
6944 static int libbpf_prepare_prog_load(struct bpf_program *prog,
6945                                     struct bpf_prog_load_opts *opts, long cookie)
6946 {
6947         enum sec_def_flags def = cookie;
6948
6949         /* old kernels might not support specifying expected_attach_type */
6950         if ((def & SEC_EXP_ATTACH_OPT) && !kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE))
6951                 opts->expected_attach_type = 0;
6952
6953         if (def & SEC_SLEEPABLE)
6954                 opts->prog_flags |= BPF_F_SLEEPABLE;
6955
6956         if (prog->type == BPF_PROG_TYPE_XDP && (def & SEC_XDP_FRAGS))
6957                 opts->prog_flags |= BPF_F_XDP_HAS_FRAGS;
6958
6959         /* special check for usdt to use uprobe_multi link */
6960         if ((def & SEC_USDT) && kernel_supports(prog->obj, FEAT_UPROBE_MULTI_LINK))
6961                 prog->expected_attach_type = BPF_TRACE_UPROBE_MULTI;
6962
6963         if ((def & SEC_ATTACH_BTF) && !prog->attach_btf_id) {
6964                 int btf_obj_fd = 0, btf_type_id = 0, err;
6965                 const char *attach_name;
6966
6967                 attach_name = strchr(prog->sec_name, '/');
6968                 if (!attach_name) {
6969                         /* if BPF program is annotated with just SEC("fentry")
6970                          * (or similar) without declaratively specifying
6971                          * target, then it is expected that target will be
6972                          * specified with bpf_program__set_attach_target() at
6973                          * runtime before BPF object load step. If not, then
6974                          * there is nothing to load into the kernel as BPF
6975                          * verifier won't be able to validate BPF program
6976                          * correctness anyways.
6977                          */
6978                         pr_warn("prog '%s': no BTF-based attach target is specified, use bpf_program__set_attach_target()\n",
6979                                 prog->name);
6980                         return -EINVAL;
6981                 }
6982                 attach_name++; /* skip over / */
6983
6984                 err = libbpf_find_attach_btf_id(prog, attach_name, &btf_obj_fd, &btf_type_id);
6985                 if (err)
6986                         return err;
6987
6988                 /* cache resolved BTF FD and BTF type ID in the prog */
6989                 prog->attach_btf_obj_fd = btf_obj_fd;
6990                 prog->attach_btf_id = btf_type_id;
6991
6992                 /* but by now libbpf common logic is not utilizing
6993                  * prog->atach_btf_obj_fd/prog->attach_btf_id anymore because
6994                  * this callback is called after opts were populated by
6995                  * libbpf, so this callback has to update opts explicitly here
6996                  */
6997                 opts->attach_btf_obj_fd = btf_obj_fd;
6998                 opts->attach_btf_id = btf_type_id;
6999         }
7000         return 0;
7001 }
7002
7003 static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz);
7004
7005 static int bpf_object_load_prog(struct bpf_object *obj, struct bpf_program *prog,
7006                                 struct bpf_insn *insns, int insns_cnt,
7007                                 const char *license, __u32 kern_version, int *prog_fd)
7008 {
7009         LIBBPF_OPTS(bpf_prog_load_opts, load_attr);
7010         const char *prog_name = NULL;
7011         char *cp, errmsg[STRERR_BUFSIZE];
7012         size_t log_buf_size = 0;
7013         char *log_buf = NULL, *tmp;
7014         int btf_fd, ret, err;
7015         bool own_log_buf = true;
7016         __u32 log_level = prog->log_level;
7017
7018         if (prog->type == BPF_PROG_TYPE_UNSPEC) {
7019                 /*
7020                  * The program type must be set.  Most likely we couldn't find a proper
7021                  * section definition at load time, and thus we didn't infer the type.
7022                  */
7023                 pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n",
7024                         prog->name, prog->sec_name);
7025                 return -EINVAL;
7026         }
7027
7028         if (!insns || !insns_cnt)
7029                 return -EINVAL;
7030
7031         if (kernel_supports(obj, FEAT_PROG_NAME))
7032                 prog_name = prog->name;
7033         load_attr.attach_prog_fd = prog->attach_prog_fd;
7034         load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd;
7035         load_attr.attach_btf_id = prog->attach_btf_id;
7036         load_attr.kern_version = kern_version;
7037         load_attr.prog_ifindex = prog->prog_ifindex;
7038
7039         /* specify func_info/line_info only if kernel supports them */
7040         btf_fd = bpf_object__btf_fd(obj);
7041         if (btf_fd >= 0 && kernel_supports(obj, FEAT_BTF_FUNC)) {
7042                 load_attr.prog_btf_fd = btf_fd;
7043                 load_attr.func_info = prog->func_info;
7044                 load_attr.func_info_rec_size = prog->func_info_rec_size;
7045                 load_attr.func_info_cnt = prog->func_info_cnt;
7046                 load_attr.line_info = prog->line_info;
7047                 load_attr.line_info_rec_size = prog->line_info_rec_size;
7048                 load_attr.line_info_cnt = prog->line_info_cnt;
7049         }
7050         load_attr.log_level = log_level;
7051         load_attr.prog_flags = prog->prog_flags;
7052         load_attr.fd_array = obj->fd_array;
7053
7054         /* adjust load_attr if sec_def provides custom preload callback */
7055         if (prog->sec_def && prog->sec_def->prog_prepare_load_fn) {
7056                 err = prog->sec_def->prog_prepare_load_fn(prog, &load_attr, prog->sec_def->cookie);
7057                 if (err < 0) {
7058                         pr_warn("prog '%s': failed to prepare load attributes: %d\n",
7059                                 prog->name, err);
7060                         return err;
7061                 }
7062                 insns = prog->insns;
7063                 insns_cnt = prog->insns_cnt;
7064         }
7065
7066         /* allow prog_prepare_load_fn to change expected_attach_type */
7067         load_attr.expected_attach_type = prog->expected_attach_type;
7068
7069         if (obj->gen_loader) {
7070                 bpf_gen__prog_load(obj->gen_loader, prog->type, prog->name,
7071                                    license, insns, insns_cnt, &load_attr,
7072                                    prog - obj->programs);
7073                 *prog_fd = -1;
7074                 return 0;
7075         }
7076
7077 retry_load:
7078         /* if log_level is zero, we don't request logs initially even if
7079          * custom log_buf is specified; if the program load fails, then we'll
7080          * bump log_level to 1 and use either custom log_buf or we'll allocate
7081          * our own and retry the load to get details on what failed
7082          */
7083         if (log_level) {
7084                 if (prog->log_buf) {
7085                         log_buf = prog->log_buf;
7086                         log_buf_size = prog->log_size;
7087                         own_log_buf = false;
7088                 } else if (obj->log_buf) {
7089                         log_buf = obj->log_buf;
7090                         log_buf_size = obj->log_size;
7091                         own_log_buf = false;
7092                 } else {
7093                         log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, log_buf_size * 2);
7094                         tmp = realloc(log_buf, log_buf_size);
7095                         if (!tmp) {
7096                                 ret = -ENOMEM;
7097                                 goto out;
7098                         }
7099                         log_buf = tmp;
7100                         log_buf[0] = '\0';
7101                         own_log_buf = true;
7102                 }
7103         }
7104
7105         load_attr.log_buf = log_buf;
7106         load_attr.log_size = log_buf_size;
7107         load_attr.log_level = log_level;
7108
7109         ret = bpf_prog_load(prog->type, prog_name, license, insns, insns_cnt, &load_attr);
7110         if (ret >= 0) {
7111                 if (log_level && own_log_buf) {
7112                         pr_debug("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
7113                                  prog->name, log_buf);
7114                 }
7115
7116                 if (obj->has_rodata && kernel_supports(obj, FEAT_PROG_BIND_MAP)) {
7117                         struct bpf_map *map;
7118                         int i;
7119
7120                         for (i = 0; i < obj->nr_maps; i++) {
7121                                 map = &prog->obj->maps[i];
7122                                 if (map->libbpf_type != LIBBPF_MAP_RODATA)
7123                                         continue;
7124
7125                                 if (bpf_prog_bind_map(ret, bpf_map__fd(map), NULL)) {
7126                                         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
7127                                         pr_warn("prog '%s': failed to bind map '%s': %s\n",
7128                                                 prog->name, map->real_name, cp);
7129                                         /* Don't fail hard if can't bind rodata. */
7130                                 }
7131                         }
7132                 }
7133
7134                 *prog_fd = ret;
7135                 ret = 0;
7136                 goto out;
7137         }
7138
7139         if (log_level == 0) {
7140                 log_level = 1;
7141                 goto retry_load;
7142         }
7143         /* On ENOSPC, increase log buffer size and retry, unless custom
7144          * log_buf is specified.
7145          * Be careful to not overflow u32, though. Kernel's log buf size limit
7146          * isn't part of UAPI so it can always be bumped to full 4GB. So don't
7147          * multiply by 2 unless we are sure we'll fit within 32 bits.
7148          * Currently, we'll get -EINVAL when we reach (UINT_MAX >> 2).
7149          */
7150         if (own_log_buf && errno == ENOSPC && log_buf_size <= UINT_MAX / 2)
7151                 goto retry_load;
7152
7153         ret = -errno;
7154
7155         /* post-process verifier log to improve error descriptions */
7156         fixup_verifier_log(prog, log_buf, log_buf_size);
7157
7158         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
7159         pr_warn("prog '%s': BPF program load failed: %s\n", prog->name, cp);
7160         pr_perm_msg(ret);
7161
7162         if (own_log_buf && log_buf && log_buf[0] != '\0') {
7163                 pr_warn("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
7164                         prog->name, log_buf);
7165         }
7166
7167 out:
7168         if (own_log_buf)
7169                 free(log_buf);
7170         return ret;
7171 }
7172
7173 static char *find_prev_line(char *buf, char *cur)
7174 {
7175         char *p;
7176
7177         if (cur == buf) /* end of a log buf */
7178                 return NULL;
7179
7180         p = cur - 1;
7181         while (p - 1 >= buf && *(p - 1) != '\n')
7182                 p--;
7183
7184         return p;
7185 }
7186
7187 static void patch_log(char *buf, size_t buf_sz, size_t log_sz,
7188                       char *orig, size_t orig_sz, const char *patch)
7189 {
7190         /* size of the remaining log content to the right from the to-be-replaced part */
7191         size_t rem_sz = (buf + log_sz) - (orig + orig_sz);
7192         size_t patch_sz = strlen(patch);
7193
7194         if (patch_sz != orig_sz) {
7195                 /* If patch line(s) are longer than original piece of verifier log,
7196                  * shift log contents by (patch_sz - orig_sz) bytes to the right
7197                  * starting from after to-be-replaced part of the log.
7198                  *
7199                  * If patch line(s) are shorter than original piece of verifier log,
7200                  * shift log contents by (orig_sz - patch_sz) bytes to the left
7201                  * starting from after to-be-replaced part of the log
7202                  *
7203                  * We need to be careful about not overflowing available
7204                  * buf_sz capacity. If that's the case, we'll truncate the end
7205                  * of the original log, as necessary.
7206                  */
7207                 if (patch_sz > orig_sz) {
7208                         if (orig + patch_sz >= buf + buf_sz) {
7209                                 /* patch is big enough to cover remaining space completely */
7210                                 patch_sz -= (orig + patch_sz) - (buf + buf_sz) + 1;
7211                                 rem_sz = 0;
7212                         } else if (patch_sz - orig_sz > buf_sz - log_sz) {
7213                                 /* patch causes part of remaining log to be truncated */
7214                                 rem_sz -= (patch_sz - orig_sz) - (buf_sz - log_sz);
7215                         }
7216                 }
7217                 /* shift remaining log to the right by calculated amount */
7218                 memmove(orig + patch_sz, orig + orig_sz, rem_sz);
7219         }
7220
7221         memcpy(orig, patch, patch_sz);
7222 }
7223
7224 static void fixup_log_failed_core_relo(struct bpf_program *prog,
7225                                        char *buf, size_t buf_sz, size_t log_sz,
7226                                        char *line1, char *line2, char *line3)
7227 {
7228         /* Expected log for failed and not properly guarded CO-RE relocation:
7229          * line1 -> 123: (85) call unknown#195896080
7230          * line2 -> invalid func unknown#195896080
7231          * line3 -> <anything else or end of buffer>
7232          *
7233          * "123" is the index of the instruction that was poisoned. We extract
7234          * instruction index to find corresponding CO-RE relocation and
7235          * replace this part of the log with more relevant information about
7236          * failed CO-RE relocation.
7237          */
7238         const struct bpf_core_relo *relo;
7239         struct bpf_core_spec spec;
7240         char patch[512], spec_buf[256];
7241         int insn_idx, err, spec_len;
7242
7243         if (sscanf(line1, "%d: (%*d) call unknown#195896080\n", &insn_idx) != 1)
7244                 return;
7245
7246         relo = find_relo_core(prog, insn_idx);
7247         if (!relo)
7248                 return;
7249
7250         err = bpf_core_parse_spec(prog->name, prog->obj->btf, relo, &spec);
7251         if (err)
7252                 return;
7253
7254         spec_len = bpf_core_format_spec(spec_buf, sizeof(spec_buf), &spec);
7255         snprintf(patch, sizeof(patch),
7256                  "%d: <invalid CO-RE relocation>\n"
7257                  "failed to resolve CO-RE relocation %s%s\n",
7258                  insn_idx, spec_buf, spec_len >= sizeof(spec_buf) ? "..." : "");
7259
7260         patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
7261 }
7262
7263 static void fixup_log_missing_map_load(struct bpf_program *prog,
7264                                        char *buf, size_t buf_sz, size_t log_sz,
7265                                        char *line1, char *line2, char *line3)
7266 {
7267         /* Expected log for failed and not properly guarded map reference:
7268          * line1 -> 123: (85) call unknown#2001000345
7269          * line2 -> invalid func unknown#2001000345
7270          * line3 -> <anything else or end of buffer>
7271          *
7272          * "123" is the index of the instruction that was poisoned.
7273          * "345" in "2001000345" is a map index in obj->maps to fetch map name.
7274          */
7275         struct bpf_object *obj = prog->obj;
7276         const struct bpf_map *map;
7277         int insn_idx, map_idx;
7278         char patch[128];
7279
7280         if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &map_idx) != 2)
7281                 return;
7282
7283         map_idx -= POISON_LDIMM64_MAP_BASE;
7284         if (map_idx < 0 || map_idx >= obj->nr_maps)
7285                 return;
7286         map = &obj->maps[map_idx];
7287
7288         snprintf(patch, sizeof(patch),
7289                  "%d: <invalid BPF map reference>\n"
7290                  "BPF map '%s' is referenced but wasn't created\n",
7291                  insn_idx, map->name);
7292
7293         patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
7294 }
7295
7296 static void fixup_log_missing_kfunc_call(struct bpf_program *prog,
7297                                          char *buf, size_t buf_sz, size_t log_sz,
7298                                          char *line1, char *line2, char *line3)
7299 {
7300         /* Expected log for failed and not properly guarded kfunc call:
7301          * line1 -> 123: (85) call unknown#2002000345
7302          * line2 -> invalid func unknown#2002000345
7303          * line3 -> <anything else or end of buffer>
7304          *
7305          * "123" is the index of the instruction that was poisoned.
7306          * "345" in "2002000345" is an extern index in obj->externs to fetch kfunc name.
7307          */
7308         struct bpf_object *obj = prog->obj;
7309         const struct extern_desc *ext;
7310         int insn_idx, ext_idx;
7311         char patch[128];
7312
7313         if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &ext_idx) != 2)
7314                 return;
7315
7316         ext_idx -= POISON_CALL_KFUNC_BASE;
7317         if (ext_idx < 0 || ext_idx >= obj->nr_extern)
7318                 return;
7319         ext = &obj->externs[ext_idx];
7320
7321         snprintf(patch, sizeof(patch),
7322                  "%d: <invalid kfunc call>\n"
7323                  "kfunc '%s' is referenced but wasn't resolved\n",
7324                  insn_idx, ext->name);
7325
7326         patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
7327 }
7328
7329 static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz)
7330 {
7331         /* look for familiar error patterns in last N lines of the log */
7332         const size_t max_last_line_cnt = 10;
7333         char *prev_line, *cur_line, *next_line;
7334         size_t log_sz;
7335         int i;
7336
7337         if (!buf)
7338                 return;
7339
7340         log_sz = strlen(buf) + 1;
7341         next_line = buf + log_sz - 1;
7342
7343         for (i = 0; i < max_last_line_cnt; i++, next_line = cur_line) {
7344                 cur_line = find_prev_line(buf, next_line);
7345                 if (!cur_line)
7346                         return;
7347
7348                 if (str_has_pfx(cur_line, "invalid func unknown#195896080\n")) {
7349                         prev_line = find_prev_line(buf, cur_line);
7350                         if (!prev_line)
7351                                 continue;
7352
7353                         /* failed CO-RE relocation case */
7354                         fixup_log_failed_core_relo(prog, buf, buf_sz, log_sz,
7355                                                    prev_line, cur_line, next_line);
7356                         return;
7357                 } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_LDIMM64_MAP_PFX)) {
7358                         prev_line = find_prev_line(buf, cur_line);
7359                         if (!prev_line)
7360                                 continue;
7361
7362                         /* reference to uncreated BPF map */
7363                         fixup_log_missing_map_load(prog, buf, buf_sz, log_sz,
7364                                                    prev_line, cur_line, next_line);
7365                         return;
7366                 } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_CALL_KFUNC_PFX)) {
7367                         prev_line = find_prev_line(buf, cur_line);
7368                         if (!prev_line)
7369                                 continue;
7370
7371                         /* reference to unresolved kfunc */
7372                         fixup_log_missing_kfunc_call(prog, buf, buf_sz, log_sz,
7373                                                      prev_line, cur_line, next_line);
7374                         return;
7375                 }
7376         }
7377 }
7378
7379 static int bpf_program_record_relos(struct bpf_program *prog)
7380 {
7381         struct bpf_object *obj = prog->obj;
7382         int i;
7383
7384         for (i = 0; i < prog->nr_reloc; i++) {
7385                 struct reloc_desc *relo = &prog->reloc_desc[i];
7386                 struct extern_desc *ext = &obj->externs[relo->ext_idx];
7387                 int kind;
7388
7389                 switch (relo->type) {
7390                 case RELO_EXTERN_LD64:
7391                         if (ext->type != EXT_KSYM)
7392                                 continue;
7393                         kind = btf_is_var(btf__type_by_id(obj->btf, ext->btf_id)) ?
7394                                 BTF_KIND_VAR : BTF_KIND_FUNC;
7395                         bpf_gen__record_extern(obj->gen_loader, ext->name,
7396                                                ext->is_weak, !ext->ksym.type_id,
7397                                                true, kind, relo->insn_idx);
7398                         break;
7399                 case RELO_EXTERN_CALL:
7400                         bpf_gen__record_extern(obj->gen_loader, ext->name,
7401                                                ext->is_weak, false, false, BTF_KIND_FUNC,
7402                                                relo->insn_idx);
7403                         break;
7404                 case RELO_CORE: {
7405                         struct bpf_core_relo cr = {
7406                                 .insn_off = relo->insn_idx * 8,
7407                                 .type_id = relo->core_relo->type_id,
7408                                 .access_str_off = relo->core_relo->access_str_off,
7409                                 .kind = relo->core_relo->kind,
7410                         };
7411
7412                         bpf_gen__record_relo_core(obj->gen_loader, &cr);
7413                         break;
7414                 }
7415                 default:
7416                         continue;
7417                 }
7418         }
7419         return 0;
7420 }
7421
7422 static int
7423 bpf_object__load_progs(struct bpf_object *obj, int log_level)
7424 {
7425         struct bpf_program *prog;
7426         size_t i;
7427         int err;
7428
7429         for (i = 0; i < obj->nr_programs; i++) {
7430                 prog = &obj->programs[i];
7431                 err = bpf_object__sanitize_prog(obj, prog);
7432                 if (err)
7433                         return err;
7434         }
7435
7436         for (i = 0; i < obj->nr_programs; i++) {
7437                 prog = &obj->programs[i];
7438                 if (prog_is_subprog(obj, prog))
7439                         continue;
7440                 if (!prog->autoload) {
7441                         pr_debug("prog '%s': skipped loading\n", prog->name);
7442                         continue;
7443                 }
7444                 prog->log_level |= log_level;
7445
7446                 if (obj->gen_loader)
7447                         bpf_program_record_relos(prog);
7448
7449                 err = bpf_object_load_prog(obj, prog, prog->insns, prog->insns_cnt,
7450                                            obj->license, obj->kern_version, &prog->fd);
7451                 if (err) {
7452                         pr_warn("prog '%s': failed to load: %d\n", prog->name, err);
7453                         return err;
7454                 }
7455         }
7456
7457         bpf_object__free_relocs(obj);
7458         return 0;
7459 }
7460
7461 static const struct bpf_sec_def *find_sec_def(const char *sec_name);
7462
7463 static int bpf_object_init_progs(struct bpf_object *obj, const struct bpf_object_open_opts *opts)
7464 {
7465         struct bpf_program *prog;
7466         int err;
7467
7468         bpf_object__for_each_program(prog, obj) {
7469                 prog->sec_def = find_sec_def(prog->sec_name);
7470                 if (!prog->sec_def) {
7471                         /* couldn't guess, but user might manually specify */
7472                         pr_debug("prog '%s': unrecognized ELF section name '%s'\n",
7473                                 prog->name, prog->sec_name);
7474                         continue;
7475                 }
7476
7477                 prog->type = prog->sec_def->prog_type;
7478                 prog->expected_attach_type = prog->sec_def->expected_attach_type;
7479
7480                 /* sec_def can have custom callback which should be called
7481                  * after bpf_program is initialized to adjust its properties
7482                  */
7483                 if (prog->sec_def->prog_setup_fn) {
7484                         err = prog->sec_def->prog_setup_fn(prog, prog->sec_def->cookie);
7485                         if (err < 0) {
7486                                 pr_warn("prog '%s': failed to initialize: %d\n",
7487                                         prog->name, err);
7488                                 return err;
7489                         }
7490                 }
7491         }
7492
7493         return 0;
7494 }
7495
7496 static struct bpf_object *bpf_object_open(const char *path, const void *obj_buf, size_t obj_buf_sz,
7497                                           const struct bpf_object_open_opts *opts)
7498 {
7499         const char *obj_name, *kconfig, *btf_tmp_path;
7500         struct bpf_object *obj;
7501         char tmp_name[64];
7502         int err;
7503         char *log_buf;
7504         size_t log_size;
7505         __u32 log_level;
7506
7507         if (elf_version(EV_CURRENT) == EV_NONE) {
7508                 pr_warn("failed to init libelf for %s\n",
7509                         path ? : "(mem buf)");
7510                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
7511         }
7512
7513         if (!OPTS_VALID(opts, bpf_object_open_opts))
7514                 return ERR_PTR(-EINVAL);
7515
7516         obj_name = OPTS_GET(opts, object_name, NULL);
7517         if (obj_buf) {
7518                 if (!obj_name) {
7519                         snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
7520                                  (unsigned long)obj_buf,
7521                                  (unsigned long)obj_buf_sz);
7522                         obj_name = tmp_name;
7523                 }
7524                 path = obj_name;
7525                 pr_debug("loading object '%s' from buffer\n", obj_name);
7526         }
7527
7528         log_buf = OPTS_GET(opts, kernel_log_buf, NULL);
7529         log_size = OPTS_GET(opts, kernel_log_size, 0);
7530         log_level = OPTS_GET(opts, kernel_log_level, 0);
7531         if (log_size > UINT_MAX)
7532                 return ERR_PTR(-EINVAL);
7533         if (log_size && !log_buf)
7534                 return ERR_PTR(-EINVAL);
7535
7536         obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
7537         if (IS_ERR(obj))
7538                 return obj;
7539
7540         obj->log_buf = log_buf;
7541         obj->log_size = log_size;
7542         obj->log_level = log_level;
7543
7544         btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL);
7545         if (btf_tmp_path) {
7546                 if (strlen(btf_tmp_path) >= PATH_MAX) {
7547                         err = -ENAMETOOLONG;
7548                         goto out;
7549                 }
7550                 obj->btf_custom_path = strdup(btf_tmp_path);
7551                 if (!obj->btf_custom_path) {
7552                         err = -ENOMEM;
7553                         goto out;
7554                 }
7555         }
7556
7557         kconfig = OPTS_GET(opts, kconfig, NULL);
7558         if (kconfig) {
7559                 obj->kconfig = strdup(kconfig);
7560                 if (!obj->kconfig) {
7561                         err = -ENOMEM;
7562                         goto out;
7563                 }
7564         }
7565
7566         err = bpf_object__elf_init(obj);
7567         err = err ? : bpf_object__check_endianness(obj);
7568         err = err ? : bpf_object__elf_collect(obj);
7569         err = err ? : bpf_object__collect_externs(obj);
7570         err = err ? : bpf_object_fixup_btf(obj);
7571         err = err ? : bpf_object__init_maps(obj, opts);
7572         err = err ? : bpf_object_init_progs(obj, opts);
7573         err = err ? : bpf_object__collect_relos(obj);
7574         if (err)
7575                 goto out;
7576
7577         bpf_object__elf_finish(obj);
7578
7579         return obj;
7580 out:
7581         bpf_object__close(obj);
7582         return ERR_PTR(err);
7583 }
7584
7585 struct bpf_object *
7586 bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
7587 {
7588         if (!path)
7589                 return libbpf_err_ptr(-EINVAL);
7590
7591         pr_debug("loading %s\n", path);
7592
7593         return libbpf_ptr(bpf_object_open(path, NULL, 0, opts));
7594 }
7595
7596 struct bpf_object *bpf_object__open(const char *path)
7597 {
7598         return bpf_object__open_file(path, NULL);
7599 }
7600
7601 struct bpf_object *
7602 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
7603                      const struct bpf_object_open_opts *opts)
7604 {
7605         if (!obj_buf || obj_buf_sz == 0)
7606                 return libbpf_err_ptr(-EINVAL);
7607
7608         return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, opts));
7609 }
7610
7611 static int bpf_object_unload(struct bpf_object *obj)
7612 {
7613         size_t i;
7614
7615         if (!obj)
7616                 return libbpf_err(-EINVAL);
7617
7618         for (i = 0; i < obj->nr_maps; i++) {
7619                 zclose(obj->maps[i].fd);
7620                 if (obj->maps[i].st_ops)
7621                         zfree(&obj->maps[i].st_ops->kern_vdata);
7622         }
7623
7624         for (i = 0; i < obj->nr_programs; i++)
7625                 bpf_program__unload(&obj->programs[i]);
7626
7627         return 0;
7628 }
7629
7630 static int bpf_object__sanitize_maps(struct bpf_object *obj)
7631 {
7632         struct bpf_map *m;
7633
7634         bpf_object__for_each_map(m, obj) {
7635                 if (!bpf_map__is_internal(m))
7636                         continue;
7637                 if (!kernel_supports(obj, FEAT_ARRAY_MMAP))
7638                         m->def.map_flags &= ~BPF_F_MMAPABLE;
7639         }
7640
7641         return 0;
7642 }
7643
7644 int libbpf_kallsyms_parse(kallsyms_cb_t cb, void *ctx)
7645 {
7646         char sym_type, sym_name[500];
7647         unsigned long long sym_addr;
7648         int ret, err = 0;
7649         FILE *f;
7650
7651         f = fopen("/proc/kallsyms", "re");
7652         if (!f) {
7653                 err = -errno;
7654                 pr_warn("failed to open /proc/kallsyms: %d\n", err);
7655                 return err;
7656         }
7657
7658         while (true) {
7659                 ret = fscanf(f, "%llx %c %499s%*[^\n]\n",
7660                              &sym_addr, &sym_type, sym_name);
7661                 if (ret == EOF && feof(f))
7662                         break;
7663                 if (ret != 3) {
7664                         pr_warn("failed to read kallsyms entry: %d\n", ret);
7665                         err = -EINVAL;
7666                         break;
7667                 }
7668
7669                 err = cb(sym_addr, sym_type, sym_name, ctx);
7670                 if (err)
7671                         break;
7672         }
7673
7674         fclose(f);
7675         return err;
7676 }
7677
7678 static int kallsyms_cb(unsigned long long sym_addr, char sym_type,
7679                        const char *sym_name, void *ctx)
7680 {
7681         struct bpf_object *obj = ctx;
7682         const struct btf_type *t;
7683         struct extern_desc *ext;
7684
7685         ext = find_extern_by_name(obj, sym_name);
7686         if (!ext || ext->type != EXT_KSYM)
7687                 return 0;
7688
7689         t = btf__type_by_id(obj->btf, ext->btf_id);
7690         if (!btf_is_var(t))
7691                 return 0;
7692
7693         if (ext->is_set && ext->ksym.addr != sym_addr) {
7694                 pr_warn("extern (ksym) '%s': resolution is ambiguous: 0x%llx or 0x%llx\n",
7695                         sym_name, ext->ksym.addr, sym_addr);
7696                 return -EINVAL;
7697         }
7698         if (!ext->is_set) {
7699                 ext->is_set = true;
7700                 ext->ksym.addr = sym_addr;
7701                 pr_debug("extern (ksym) '%s': set to 0x%llx\n", sym_name, sym_addr);
7702         }
7703         return 0;
7704 }
7705
7706 static int bpf_object__read_kallsyms_file(struct bpf_object *obj)
7707 {
7708         return libbpf_kallsyms_parse(kallsyms_cb, obj);
7709 }
7710
7711 static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name,
7712                             __u16 kind, struct btf **res_btf,
7713                             struct module_btf **res_mod_btf)
7714 {
7715         struct module_btf *mod_btf;
7716         struct btf *btf;
7717         int i, id, err;
7718
7719         btf = obj->btf_vmlinux;
7720         mod_btf = NULL;
7721         id = btf__find_by_name_kind(btf, ksym_name, kind);
7722
7723         if (id == -ENOENT) {
7724                 err = load_module_btfs(obj);
7725                 if (err)
7726                         return err;
7727
7728                 for (i = 0; i < obj->btf_module_cnt; i++) {
7729                         /* we assume module_btf's BTF FD is always >0 */
7730                         mod_btf = &obj->btf_modules[i];
7731                         btf = mod_btf->btf;
7732                         id = btf__find_by_name_kind_own(btf, ksym_name, kind);
7733                         if (id != -ENOENT)
7734                                 break;
7735                 }
7736         }
7737         if (id <= 0)
7738                 return -ESRCH;
7739
7740         *res_btf = btf;
7741         *res_mod_btf = mod_btf;
7742         return id;
7743 }
7744
7745 static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj,
7746                                                struct extern_desc *ext)
7747 {
7748         const struct btf_type *targ_var, *targ_type;
7749         __u32 targ_type_id, local_type_id;
7750         struct module_btf *mod_btf = NULL;
7751         const char *targ_var_name;
7752         struct btf *btf = NULL;
7753         int id, err;
7754
7755         id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &mod_btf);
7756         if (id < 0) {
7757                 if (id == -ESRCH && ext->is_weak)
7758                         return 0;
7759                 pr_warn("extern (var ksym) '%s': not found in kernel BTF\n",
7760                         ext->name);
7761                 return id;
7762         }
7763
7764         /* find local type_id */
7765         local_type_id = ext->ksym.type_id;
7766
7767         /* find target type_id */
7768         targ_var = btf__type_by_id(btf, id);
7769         targ_var_name = btf__name_by_offset(btf, targ_var->name_off);
7770         targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id);
7771
7772         err = bpf_core_types_are_compat(obj->btf, local_type_id,
7773                                         btf, targ_type_id);
7774         if (err <= 0) {
7775                 const struct btf_type *local_type;
7776                 const char *targ_name, *local_name;
7777
7778                 local_type = btf__type_by_id(obj->btf, local_type_id);
7779                 local_name = btf__name_by_offset(obj->btf, local_type->name_off);
7780                 targ_name = btf__name_by_offset(btf, targ_type->name_off);
7781
7782                 pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n",
7783                         ext->name, local_type_id,
7784                         btf_kind_str(local_type), local_name, targ_type_id,
7785                         btf_kind_str(targ_type), targ_name);
7786                 return -EINVAL;
7787         }
7788
7789         ext->is_set = true;
7790         ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0;
7791         ext->ksym.kernel_btf_id = id;
7792         pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n",
7793                  ext->name, id, btf_kind_str(targ_var), targ_var_name);
7794
7795         return 0;
7796 }
7797
7798 static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj,
7799                                                 struct extern_desc *ext)
7800 {
7801         int local_func_proto_id, kfunc_proto_id, kfunc_id;
7802         struct module_btf *mod_btf = NULL;
7803         const struct btf_type *kern_func;
7804         struct btf *kern_btf = NULL;
7805         int ret;
7806
7807         local_func_proto_id = ext->ksym.type_id;
7808
7809         kfunc_id = find_ksym_btf_id(obj, ext->essent_name ?: ext->name, BTF_KIND_FUNC, &kern_btf,
7810                                     &mod_btf);
7811         if (kfunc_id < 0) {
7812                 if (kfunc_id == -ESRCH && ext->is_weak)
7813                         return 0;
7814                 pr_warn("extern (func ksym) '%s': not found in kernel or module BTFs\n",
7815                         ext->name);
7816                 return kfunc_id;
7817         }
7818
7819         kern_func = btf__type_by_id(kern_btf, kfunc_id);
7820         kfunc_proto_id = kern_func->type;
7821
7822         ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id,
7823                                         kern_btf, kfunc_proto_id);
7824         if (ret <= 0) {
7825                 if (ext->is_weak)
7826                         return 0;
7827
7828                 pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with %s [%d]\n",
7829                         ext->name, local_func_proto_id,
7830                         mod_btf ? mod_btf->name : "vmlinux", kfunc_proto_id);
7831                 return -EINVAL;
7832         }
7833
7834         /* set index for module BTF fd in fd_array, if unset */
7835         if (mod_btf && !mod_btf->fd_array_idx) {
7836                 /* insn->off is s16 */
7837                 if (obj->fd_array_cnt == INT16_MAX) {
7838                         pr_warn("extern (func ksym) '%s': module BTF fd index %d too big to fit in bpf_insn offset\n",
7839                                 ext->name, mod_btf->fd_array_idx);
7840                         return -E2BIG;
7841                 }
7842                 /* Cannot use index 0 for module BTF fd */
7843                 if (!obj->fd_array_cnt)
7844                         obj->fd_array_cnt = 1;
7845
7846                 ret = libbpf_ensure_mem((void **)&obj->fd_array, &obj->fd_array_cap, sizeof(int),
7847                                         obj->fd_array_cnt + 1);
7848                 if (ret)
7849                         return ret;
7850                 mod_btf->fd_array_idx = obj->fd_array_cnt;
7851                 /* we assume module BTF FD is always >0 */
7852                 obj->fd_array[obj->fd_array_cnt++] = mod_btf->fd;
7853         }
7854
7855         ext->is_set = true;
7856         ext->ksym.kernel_btf_id = kfunc_id;
7857         ext->ksym.btf_fd_idx = mod_btf ? mod_btf->fd_array_idx : 0;
7858         /* Also set kernel_btf_obj_fd to make sure that bpf_object__relocate_data()
7859          * populates FD into ld_imm64 insn when it's used to point to kfunc.
7860          * {kernel_btf_id, btf_fd_idx} -> fixup bpf_call.
7861          * {kernel_btf_id, kernel_btf_obj_fd} -> fixup ld_imm64.
7862          */
7863         ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0;
7864         pr_debug("extern (func ksym) '%s': resolved to %s [%d]\n",
7865                  ext->name, mod_btf ? mod_btf->name : "vmlinux", kfunc_id);
7866
7867         return 0;
7868 }
7869
7870 static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj)
7871 {
7872         const struct btf_type *t;
7873         struct extern_desc *ext;
7874         int i, err;
7875
7876         for (i = 0; i < obj->nr_extern; i++) {
7877                 ext = &obj->externs[i];
7878                 if (ext->type != EXT_KSYM || !ext->ksym.type_id)
7879                         continue;
7880
7881                 if (obj->gen_loader) {
7882                         ext->is_set = true;
7883                         ext->ksym.kernel_btf_obj_fd = 0;
7884                         ext->ksym.kernel_btf_id = 0;
7885                         continue;
7886                 }
7887                 t = btf__type_by_id(obj->btf, ext->btf_id);
7888                 if (btf_is_var(t))
7889                         err = bpf_object__resolve_ksym_var_btf_id(obj, ext);
7890                 else
7891                         err = bpf_object__resolve_ksym_func_btf_id(obj, ext);
7892                 if (err)
7893                         return err;
7894         }
7895         return 0;
7896 }
7897
7898 static int bpf_object__resolve_externs(struct bpf_object *obj,
7899                                        const char *extra_kconfig)
7900 {
7901         bool need_config = false, need_kallsyms = false;
7902         bool need_vmlinux_btf = false;
7903         struct extern_desc *ext;
7904         void *kcfg_data = NULL;
7905         int err, i;
7906
7907         if (obj->nr_extern == 0)
7908                 return 0;
7909
7910         if (obj->kconfig_map_idx >= 0)
7911                 kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped;
7912
7913         for (i = 0; i < obj->nr_extern; i++) {
7914                 ext = &obj->externs[i];
7915
7916                 if (ext->type == EXT_KSYM) {
7917                         if (ext->ksym.type_id)
7918                                 need_vmlinux_btf = true;
7919                         else
7920                                 need_kallsyms = true;
7921                         continue;
7922                 } else if (ext->type == EXT_KCFG) {
7923                         void *ext_ptr = kcfg_data + ext->kcfg.data_off;
7924                         __u64 value = 0;
7925
7926                         /* Kconfig externs need actual /proc/config.gz */
7927                         if (str_has_pfx(ext->name, "CONFIG_")) {
7928                                 need_config = true;
7929                                 continue;
7930                         }
7931
7932                         /* Virtual kcfg externs are customly handled by libbpf */
7933                         if (strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
7934                                 value = get_kernel_version();
7935                                 if (!value) {
7936                                         pr_warn("extern (kcfg) '%s': failed to get kernel version\n", ext->name);
7937                                         return -EINVAL;
7938                                 }
7939                         } else if (strcmp(ext->name, "LINUX_HAS_BPF_COOKIE") == 0) {
7940                                 value = kernel_supports(obj, FEAT_BPF_COOKIE);
7941                         } else if (strcmp(ext->name, "LINUX_HAS_SYSCALL_WRAPPER") == 0) {
7942                                 value = kernel_supports(obj, FEAT_SYSCALL_WRAPPER);
7943                         } else if (!str_has_pfx(ext->name, "LINUX_") || !ext->is_weak) {
7944                                 /* Currently libbpf supports only CONFIG_ and LINUX_ prefixed
7945                                  * __kconfig externs, where LINUX_ ones are virtual and filled out
7946                                  * customly by libbpf (their values don't come from Kconfig).
7947                                  * If LINUX_xxx variable is not recognized by libbpf, but is marked
7948                                  * __weak, it defaults to zero value, just like for CONFIG_xxx
7949                                  * externs.
7950                                  */
7951                                 pr_warn("extern (kcfg) '%s': unrecognized virtual extern\n", ext->name);
7952                                 return -EINVAL;
7953                         }
7954
7955                         err = set_kcfg_value_num(ext, ext_ptr, value);
7956                         if (err)
7957                                 return err;
7958                         pr_debug("extern (kcfg) '%s': set to 0x%llx\n",
7959                                  ext->name, (long long)value);
7960                 } else {
7961                         pr_warn("extern '%s': unrecognized extern kind\n", ext->name);
7962                         return -EINVAL;
7963                 }
7964         }
7965         if (need_config && extra_kconfig) {
7966                 err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data);
7967                 if (err)
7968                         return -EINVAL;
7969                 need_config = false;
7970                 for (i = 0; i < obj->nr_extern; i++) {
7971                         ext = &obj->externs[i];
7972                         if (ext->type == EXT_KCFG && !ext->is_set) {
7973                                 need_config = true;
7974                                 break;
7975                         }
7976                 }
7977         }
7978         if (need_config) {
7979                 err = bpf_object__read_kconfig_file(obj, kcfg_data);
7980                 if (err)
7981                         return -EINVAL;
7982         }
7983         if (need_kallsyms) {
7984                 err = bpf_object__read_kallsyms_file(obj);
7985                 if (err)
7986                         return -EINVAL;
7987         }
7988         if (need_vmlinux_btf) {
7989                 err = bpf_object__resolve_ksyms_btf_id(obj);
7990                 if (err)
7991                         return -EINVAL;
7992         }
7993         for (i = 0; i < obj->nr_extern; i++) {
7994                 ext = &obj->externs[i];
7995
7996                 if (!ext->is_set && !ext->is_weak) {
7997                         pr_warn("extern '%s' (strong): not resolved\n", ext->name);
7998                         return -ESRCH;
7999                 } else if (!ext->is_set) {
8000                         pr_debug("extern '%s' (weak): not resolved, defaulting to zero\n",
8001                                  ext->name);
8002                 }
8003         }
8004
8005         return 0;
8006 }
8007
8008 static void bpf_map_prepare_vdata(const struct bpf_map *map)
8009 {
8010         struct bpf_struct_ops *st_ops;
8011         __u32 i;
8012
8013         st_ops = map->st_ops;
8014         for (i = 0; i < btf_vlen(st_ops->type); i++) {
8015                 struct bpf_program *prog = st_ops->progs[i];
8016                 void *kern_data;
8017                 int prog_fd;
8018
8019                 if (!prog)
8020                         continue;
8021
8022                 prog_fd = bpf_program__fd(prog);
8023                 kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i];
8024                 *(unsigned long *)kern_data = prog_fd;
8025         }
8026 }
8027
8028 static int bpf_object_prepare_struct_ops(struct bpf_object *obj)
8029 {
8030         int i;
8031
8032         for (i = 0; i < obj->nr_maps; i++)
8033                 if (bpf_map__is_struct_ops(&obj->maps[i]))
8034                         bpf_map_prepare_vdata(&obj->maps[i]);
8035
8036         return 0;
8037 }
8038
8039 static int bpf_object_load(struct bpf_object *obj, int extra_log_level, const char *target_btf_path)
8040 {
8041         int err, i;
8042
8043         if (!obj)
8044                 return libbpf_err(-EINVAL);
8045
8046         if (obj->loaded) {
8047                 pr_warn("object '%s': load can't be attempted twice\n", obj->name);
8048                 return libbpf_err(-EINVAL);
8049         }
8050
8051         if (obj->gen_loader)
8052                 bpf_gen__init(obj->gen_loader, extra_log_level, obj->nr_programs, obj->nr_maps);
8053
8054         err = bpf_object__probe_loading(obj);
8055         err = err ? : bpf_object__load_vmlinux_btf(obj, false);
8056         err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
8057         err = err ? : bpf_object__sanitize_and_load_btf(obj);
8058         err = err ? : bpf_object__sanitize_maps(obj);
8059         err = err ? : bpf_object__init_kern_struct_ops_maps(obj);
8060         err = err ? : bpf_object__create_maps(obj);
8061         err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : target_btf_path);
8062         err = err ? : bpf_object__load_progs(obj, extra_log_level);
8063         err = err ? : bpf_object_init_prog_arrays(obj);
8064         err = err ? : bpf_object_prepare_struct_ops(obj);
8065
8066         if (obj->gen_loader) {
8067                 /* reset FDs */
8068                 if (obj->btf)
8069                         btf__set_fd(obj->btf, -1);
8070                 for (i = 0; i < obj->nr_maps; i++)
8071                         obj->maps[i].fd = -1;
8072                 if (!err)
8073                         err = bpf_gen__finish(obj->gen_loader, obj->nr_programs, obj->nr_maps);
8074         }
8075
8076         /* clean up fd_array */
8077         zfree(&obj->fd_array);
8078
8079         /* clean up module BTFs */
8080         for (i = 0; i < obj->btf_module_cnt; i++) {
8081                 close(obj->btf_modules[i].fd);
8082                 btf__free(obj->btf_modules[i].btf);
8083                 free(obj->btf_modules[i].name);
8084         }
8085         free(obj->btf_modules);
8086
8087         /* clean up vmlinux BTF */
8088         btf__free(obj->btf_vmlinux);
8089         obj->btf_vmlinux = NULL;
8090
8091         obj->loaded = true; /* doesn't matter if successfully or not */
8092
8093         if (err)
8094                 goto out;
8095
8096         return 0;
8097 out:
8098         /* unpin any maps that were auto-pinned during load */
8099         for (i = 0; i < obj->nr_maps; i++)
8100                 if (obj->maps[i].pinned && !obj->maps[i].reused)
8101                         bpf_map__unpin(&obj->maps[i], NULL);
8102
8103         bpf_object_unload(obj);
8104         pr_warn("failed to load object '%s'\n", obj->path);
8105         return libbpf_err(err);
8106 }
8107
8108 int bpf_object__load(struct bpf_object *obj)
8109 {
8110         return bpf_object_load(obj, 0, NULL);
8111 }
8112
8113 static int make_parent_dir(const char *path)
8114 {
8115         char *cp, errmsg[STRERR_BUFSIZE];
8116         char *dname, *dir;
8117         int err = 0;
8118
8119         dname = strdup(path);
8120         if (dname == NULL)
8121                 return -ENOMEM;
8122
8123         dir = dirname(dname);
8124         if (mkdir(dir, 0700) && errno != EEXIST)
8125                 err = -errno;
8126
8127         free(dname);
8128         if (err) {
8129                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
8130                 pr_warn("failed to mkdir %s: %s\n", path, cp);
8131         }
8132         return err;
8133 }
8134
8135 static int check_path(const char *path)
8136 {
8137         char *cp, errmsg[STRERR_BUFSIZE];
8138         struct statfs st_fs;
8139         char *dname, *dir;
8140         int err = 0;
8141
8142         if (path == NULL)
8143                 return -EINVAL;
8144
8145         dname = strdup(path);
8146         if (dname == NULL)
8147                 return -ENOMEM;
8148
8149         dir = dirname(dname);
8150         if (statfs(dir, &st_fs)) {
8151                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
8152                 pr_warn("failed to statfs %s: %s\n", dir, cp);
8153                 err = -errno;
8154         }
8155         free(dname);
8156
8157         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
8158                 pr_warn("specified path %s is not on BPF FS\n", path);
8159                 err = -EINVAL;
8160         }
8161
8162         return err;
8163 }
8164
8165 int bpf_program__pin(struct bpf_program *prog, const char *path)
8166 {
8167         char *cp, errmsg[STRERR_BUFSIZE];
8168         int err;
8169
8170         if (prog->fd < 0) {
8171                 pr_warn("prog '%s': can't pin program that wasn't loaded\n", prog->name);
8172                 return libbpf_err(-EINVAL);
8173         }
8174
8175         err = make_parent_dir(path);
8176         if (err)
8177                 return libbpf_err(err);
8178
8179         err = check_path(path);
8180         if (err)
8181                 return libbpf_err(err);
8182
8183         if (bpf_obj_pin(prog->fd, path)) {
8184                 err = -errno;
8185                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
8186                 pr_warn("prog '%s': failed to pin at '%s': %s\n", prog->name, path, cp);
8187                 return libbpf_err(err);
8188         }
8189
8190         pr_debug("prog '%s': pinned at '%s'\n", prog->name, path);
8191         return 0;
8192 }
8193
8194 int bpf_program__unpin(struct bpf_program *prog, const char *path)
8195 {
8196         int err;
8197
8198         if (prog->fd < 0) {
8199                 pr_warn("prog '%s': can't unpin program that wasn't loaded\n", prog->name);
8200                 return libbpf_err(-EINVAL);
8201         }
8202
8203         err = check_path(path);
8204         if (err)
8205                 return libbpf_err(err);
8206
8207         err = unlink(path);
8208         if (err)
8209                 return libbpf_err(-errno);
8210
8211         pr_debug("prog '%s': unpinned from '%s'\n", prog->name, path);
8212         return 0;
8213 }
8214
8215 int bpf_map__pin(struct bpf_map *map, const char *path)
8216 {
8217         char *cp, errmsg[STRERR_BUFSIZE];
8218         int err;
8219
8220         if (map == NULL) {
8221                 pr_warn("invalid map pointer\n");
8222                 return libbpf_err(-EINVAL);
8223         }
8224
8225         if (map->pin_path) {
8226                 if (path && strcmp(path, map->pin_path)) {
8227                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
8228                                 bpf_map__name(map), map->pin_path, path);
8229                         return libbpf_err(-EINVAL);
8230                 } else if (map->pinned) {
8231                         pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
8232                                  bpf_map__name(map), map->pin_path);
8233                         return 0;
8234                 }
8235         } else {
8236                 if (!path) {
8237                         pr_warn("missing a path to pin map '%s' at\n",
8238                                 bpf_map__name(map));
8239                         return libbpf_err(-EINVAL);
8240                 } else if (map->pinned) {
8241                         pr_warn("map '%s' already pinned\n", bpf_map__name(map));
8242                         return libbpf_err(-EEXIST);
8243                 }
8244
8245                 map->pin_path = strdup(path);
8246                 if (!map->pin_path) {
8247                         err = -errno;
8248                         goto out_err;
8249                 }
8250         }
8251
8252         err = make_parent_dir(map->pin_path);
8253         if (err)
8254                 return libbpf_err(err);
8255
8256         err = check_path(map->pin_path);
8257         if (err)
8258                 return libbpf_err(err);
8259
8260         if (bpf_obj_pin(map->fd, map->pin_path)) {
8261                 err = -errno;
8262                 goto out_err;
8263         }
8264
8265         map->pinned = true;
8266         pr_debug("pinned map '%s'\n", map->pin_path);
8267
8268         return 0;
8269
8270 out_err:
8271         cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
8272         pr_warn("failed to pin map: %s\n", cp);
8273         return libbpf_err(err);
8274 }
8275
8276 int bpf_map__unpin(struct bpf_map *map, const char *path)
8277 {
8278         int err;
8279
8280         if (map == NULL) {
8281                 pr_warn("invalid map pointer\n");
8282                 return libbpf_err(-EINVAL);
8283         }
8284
8285         if (map->pin_path) {
8286                 if (path && strcmp(path, map->pin_path)) {
8287                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
8288                                 bpf_map__name(map), map->pin_path, path);
8289                         return libbpf_err(-EINVAL);
8290                 }
8291                 path = map->pin_path;
8292         } else if (!path) {
8293                 pr_warn("no path to unpin map '%s' from\n",
8294                         bpf_map__name(map));
8295                 return libbpf_err(-EINVAL);
8296         }
8297
8298         err = check_path(path);
8299         if (err)
8300                 return libbpf_err(err);
8301
8302         err = unlink(path);
8303         if (err != 0)
8304                 return libbpf_err(-errno);
8305
8306         map->pinned = false;
8307         pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
8308
8309         return 0;
8310 }
8311
8312 int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
8313 {
8314         char *new = NULL;
8315
8316         if (path) {
8317                 new = strdup(path);
8318                 if (!new)
8319                         return libbpf_err(-errno);
8320         }
8321
8322         free(map->pin_path);
8323         map->pin_path = new;
8324         return 0;
8325 }
8326
8327 __alias(bpf_map__pin_path)
8328 const char *bpf_map__get_pin_path(const struct bpf_map *map);
8329
8330 const char *bpf_map__pin_path(const struct bpf_map *map)
8331 {
8332         return map->pin_path;
8333 }
8334
8335 bool bpf_map__is_pinned(const struct bpf_map *map)
8336 {
8337         return map->pinned;
8338 }
8339
8340 static void sanitize_pin_path(char *s)
8341 {
8342         /* bpffs disallows periods in path names */
8343         while (*s) {
8344                 if (*s == '.')
8345                         *s = '_';
8346                 s++;
8347         }
8348 }
8349
8350 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
8351 {
8352         struct bpf_map *map;
8353         int err;
8354
8355         if (!obj)
8356                 return libbpf_err(-ENOENT);
8357
8358         if (!obj->loaded) {
8359                 pr_warn("object not yet loaded; load it first\n");
8360                 return libbpf_err(-ENOENT);
8361         }
8362
8363         bpf_object__for_each_map(map, obj) {
8364                 char *pin_path = NULL;
8365                 char buf[PATH_MAX];
8366
8367                 if (!map->autocreate)
8368                         continue;
8369
8370                 if (path) {
8371                         err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
8372                         if (err)
8373                                 goto err_unpin_maps;
8374                         sanitize_pin_path(buf);
8375                         pin_path = buf;
8376                 } else if (!map->pin_path) {
8377                         continue;
8378                 }
8379
8380                 err = bpf_map__pin(map, pin_path);
8381                 if (err)
8382                         goto err_unpin_maps;
8383         }
8384
8385         return 0;
8386
8387 err_unpin_maps:
8388         while ((map = bpf_object__prev_map(obj, map))) {
8389                 if (!map->pin_path)
8390                         continue;
8391
8392                 bpf_map__unpin(map, NULL);
8393         }
8394
8395         return libbpf_err(err);
8396 }
8397
8398 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
8399 {
8400         struct bpf_map *map;
8401         int err;
8402
8403         if (!obj)
8404                 return libbpf_err(-ENOENT);
8405
8406         bpf_object__for_each_map(map, obj) {
8407                 char *pin_path = NULL;
8408                 char buf[PATH_MAX];
8409
8410                 if (path) {
8411                         err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
8412                         if (err)
8413                                 return libbpf_err(err);
8414                         sanitize_pin_path(buf);
8415                         pin_path = buf;
8416                 } else if (!map->pin_path) {
8417                         continue;
8418                 }
8419
8420                 err = bpf_map__unpin(map, pin_path);
8421                 if (err)
8422                         return libbpf_err(err);
8423         }
8424
8425         return 0;
8426 }
8427
8428 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
8429 {
8430         struct bpf_program *prog;
8431         char buf[PATH_MAX];
8432         int err;
8433
8434         if (!obj)
8435                 return libbpf_err(-ENOENT);
8436
8437         if (!obj->loaded) {
8438                 pr_warn("object not yet loaded; load it first\n");
8439                 return libbpf_err(-ENOENT);
8440         }
8441
8442         bpf_object__for_each_program(prog, obj) {
8443                 err = pathname_concat(buf, sizeof(buf), path, prog->name);
8444                 if (err)
8445                         goto err_unpin_programs;
8446
8447                 err = bpf_program__pin(prog, buf);
8448                 if (err)
8449                         goto err_unpin_programs;
8450         }
8451
8452         return 0;
8453
8454 err_unpin_programs:
8455         while ((prog = bpf_object__prev_program(obj, prog))) {
8456                 if (pathname_concat(buf, sizeof(buf), path, prog->name))
8457                         continue;
8458
8459                 bpf_program__unpin(prog, buf);
8460         }
8461
8462         return libbpf_err(err);
8463 }
8464
8465 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
8466 {
8467         struct bpf_program *prog;
8468         int err;
8469
8470         if (!obj)
8471                 return libbpf_err(-ENOENT);
8472
8473         bpf_object__for_each_program(prog, obj) {
8474                 char buf[PATH_MAX];
8475
8476                 err = pathname_concat(buf, sizeof(buf), path, prog->name);
8477                 if (err)
8478                         return libbpf_err(err);
8479
8480                 err = bpf_program__unpin(prog, buf);
8481                 if (err)
8482                         return libbpf_err(err);
8483         }
8484
8485         return 0;
8486 }
8487
8488 int bpf_object__pin(struct bpf_object *obj, const char *path)
8489 {
8490         int err;
8491
8492         err = bpf_object__pin_maps(obj, path);
8493         if (err)
8494                 return libbpf_err(err);
8495
8496         err = bpf_object__pin_programs(obj, path);
8497         if (err) {
8498                 bpf_object__unpin_maps(obj, path);
8499                 return libbpf_err(err);
8500         }
8501
8502         return 0;
8503 }
8504
8505 int bpf_object__unpin(struct bpf_object *obj, const char *path)
8506 {
8507         int err;
8508
8509         err = bpf_object__unpin_programs(obj, path);
8510         if (err)
8511                 return libbpf_err(err);
8512
8513         err = bpf_object__unpin_maps(obj, path);
8514         if (err)
8515                 return libbpf_err(err);
8516
8517         return 0;
8518 }
8519
8520 static void bpf_map__destroy(struct bpf_map *map)
8521 {
8522         if (map->inner_map) {
8523                 bpf_map__destroy(map->inner_map);
8524                 zfree(&map->inner_map);
8525         }
8526
8527         zfree(&map->init_slots);
8528         map->init_slots_sz = 0;
8529
8530         if (map->mmaped) {
8531                 size_t mmap_sz;
8532
8533                 mmap_sz = bpf_map_mmap_sz(map->def.value_size, map->def.max_entries);
8534                 munmap(map->mmaped, mmap_sz);
8535                 map->mmaped = NULL;
8536         }
8537
8538         if (map->st_ops) {
8539                 zfree(&map->st_ops->data);
8540                 zfree(&map->st_ops->progs);
8541                 zfree(&map->st_ops->kern_func_off);
8542                 zfree(&map->st_ops);
8543         }
8544
8545         zfree(&map->name);
8546         zfree(&map->real_name);
8547         zfree(&map->pin_path);
8548
8549         if (map->fd >= 0)
8550                 zclose(map->fd);
8551 }
8552
8553 void bpf_object__close(struct bpf_object *obj)
8554 {
8555         size_t i;
8556
8557         if (IS_ERR_OR_NULL(obj))
8558                 return;
8559
8560         usdt_manager_free(obj->usdt_man);
8561         obj->usdt_man = NULL;
8562
8563         bpf_gen__free(obj->gen_loader);
8564         bpf_object__elf_finish(obj);
8565         bpf_object_unload(obj);
8566         btf__free(obj->btf);
8567         btf__free(obj->btf_vmlinux);
8568         btf_ext__free(obj->btf_ext);
8569
8570         for (i = 0; i < obj->nr_maps; i++)
8571                 bpf_map__destroy(&obj->maps[i]);
8572
8573         zfree(&obj->btf_custom_path);
8574         zfree(&obj->kconfig);
8575
8576         for (i = 0; i < obj->nr_extern; i++)
8577                 zfree(&obj->externs[i].essent_name);
8578
8579         zfree(&obj->externs);
8580         obj->nr_extern = 0;
8581
8582         zfree(&obj->maps);
8583         obj->nr_maps = 0;
8584
8585         if (obj->programs && obj->nr_programs) {
8586                 for (i = 0; i < obj->nr_programs; i++)
8587                         bpf_program__exit(&obj->programs[i]);
8588         }
8589         zfree(&obj->programs);
8590
8591         free(obj);
8592 }
8593
8594 const char *bpf_object__name(const struct bpf_object *obj)
8595 {
8596         return obj ? obj->name : libbpf_err_ptr(-EINVAL);
8597 }
8598
8599 unsigned int bpf_object__kversion(const struct bpf_object *obj)
8600 {
8601         return obj ? obj->kern_version : 0;
8602 }
8603
8604 struct btf *bpf_object__btf(const struct bpf_object *obj)
8605 {
8606         return obj ? obj->btf : NULL;
8607 }
8608
8609 int bpf_object__btf_fd(const struct bpf_object *obj)
8610 {
8611         return obj->btf ? btf__fd(obj->btf) : -1;
8612 }
8613
8614 int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version)
8615 {
8616         if (obj->loaded)
8617                 return libbpf_err(-EINVAL);
8618
8619         obj->kern_version = kern_version;
8620
8621         return 0;
8622 }
8623
8624 int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts)
8625 {
8626         struct bpf_gen *gen;
8627
8628         if (!opts)
8629                 return -EFAULT;
8630         if (!OPTS_VALID(opts, gen_loader_opts))
8631                 return -EINVAL;
8632         gen = calloc(sizeof(*gen), 1);
8633         if (!gen)
8634                 return -ENOMEM;
8635         gen->opts = opts;
8636         obj->gen_loader = gen;
8637         return 0;
8638 }
8639
8640 static struct bpf_program *
8641 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
8642                     bool forward)
8643 {
8644         size_t nr_programs = obj->nr_programs;
8645         ssize_t idx;
8646
8647         if (!nr_programs)
8648                 return NULL;
8649
8650         if (!p)
8651                 /* Iter from the beginning */
8652                 return forward ? &obj->programs[0] :
8653                         &obj->programs[nr_programs - 1];
8654
8655         if (p->obj != obj) {
8656                 pr_warn("error: program handler doesn't match object\n");
8657                 return errno = EINVAL, NULL;
8658         }
8659
8660         idx = (p - obj->programs) + (forward ? 1 : -1);
8661         if (idx >= obj->nr_programs || idx < 0)
8662                 return NULL;
8663         return &obj->programs[idx];
8664 }
8665
8666 struct bpf_program *
8667 bpf_object__next_program(const struct bpf_object *obj, struct bpf_program *prev)
8668 {
8669         struct bpf_program *prog = prev;
8670
8671         do {
8672                 prog = __bpf_program__iter(prog, obj, true);
8673         } while (prog && prog_is_subprog(obj, prog));
8674
8675         return prog;
8676 }
8677
8678 struct bpf_program *
8679 bpf_object__prev_program(const struct bpf_object *obj, struct bpf_program *next)
8680 {
8681         struct bpf_program *prog = next;
8682
8683         do {
8684                 prog = __bpf_program__iter(prog, obj, false);
8685         } while (prog && prog_is_subprog(obj, prog));
8686
8687         return prog;
8688 }
8689
8690 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
8691 {
8692         prog->prog_ifindex = ifindex;
8693 }
8694
8695 const char *bpf_program__name(const struct bpf_program *prog)
8696 {
8697         return prog->name;
8698 }
8699
8700 const char *bpf_program__section_name(const struct bpf_program *prog)
8701 {
8702         return prog->sec_name;
8703 }
8704
8705 bool bpf_program__autoload(const struct bpf_program *prog)
8706 {
8707         return prog->autoload;
8708 }
8709
8710 int bpf_program__set_autoload(struct bpf_program *prog, bool autoload)
8711 {
8712         if (prog->obj->loaded)
8713                 return libbpf_err(-EINVAL);
8714
8715         prog->autoload = autoload;
8716         return 0;
8717 }
8718
8719 bool bpf_program__autoattach(const struct bpf_program *prog)
8720 {
8721         return prog->autoattach;
8722 }
8723
8724 void bpf_program__set_autoattach(struct bpf_program *prog, bool autoattach)
8725 {
8726         prog->autoattach = autoattach;
8727 }
8728
8729 const struct bpf_insn *bpf_program__insns(const struct bpf_program *prog)
8730 {
8731         return prog->insns;
8732 }
8733
8734 size_t bpf_program__insn_cnt(const struct bpf_program *prog)
8735 {
8736         return prog->insns_cnt;
8737 }
8738
8739 int bpf_program__set_insns(struct bpf_program *prog,
8740                            struct bpf_insn *new_insns, size_t new_insn_cnt)
8741 {
8742         struct bpf_insn *insns;
8743
8744         if (prog->obj->loaded)
8745                 return -EBUSY;
8746
8747         insns = libbpf_reallocarray(prog->insns, new_insn_cnt, sizeof(*insns));
8748         /* NULL is a valid return from reallocarray if the new count is zero */
8749         if (!insns && new_insn_cnt) {
8750                 pr_warn("prog '%s': failed to realloc prog code\n", prog->name);
8751                 return -ENOMEM;
8752         }
8753         memcpy(insns, new_insns, new_insn_cnt * sizeof(*insns));
8754
8755         prog->insns = insns;
8756         prog->insns_cnt = new_insn_cnt;
8757         return 0;
8758 }
8759
8760 int bpf_program__fd(const struct bpf_program *prog)
8761 {
8762         if (!prog)
8763                 return libbpf_err(-EINVAL);
8764
8765         if (prog->fd < 0)
8766                 return libbpf_err(-ENOENT);
8767
8768         return prog->fd;
8769 }
8770
8771 __alias(bpf_program__type)
8772 enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog);
8773
8774 enum bpf_prog_type bpf_program__type(const struct bpf_program *prog)
8775 {
8776         return prog->type;
8777 }
8778
8779 static size_t custom_sec_def_cnt;
8780 static struct bpf_sec_def *custom_sec_defs;
8781 static struct bpf_sec_def custom_fallback_def;
8782 static bool has_custom_fallback_def;
8783 static int last_custom_sec_def_handler_id;
8784
8785 int bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
8786 {
8787         if (prog->obj->loaded)
8788                 return libbpf_err(-EBUSY);
8789
8790         /* if type is not changed, do nothing */
8791         if (prog->type == type)
8792                 return 0;
8793
8794         prog->type = type;
8795
8796         /* If a program type was changed, we need to reset associated SEC()
8797          * handler, as it will be invalid now. The only exception is a generic
8798          * fallback handler, which by definition is program type-agnostic and
8799          * is a catch-all custom handler, optionally set by the application,
8800          * so should be able to handle any type of BPF program.
8801          */
8802         if (prog->sec_def != &custom_fallback_def)
8803                 prog->sec_def = NULL;
8804         return 0;
8805 }
8806
8807 __alias(bpf_program__expected_attach_type)
8808 enum bpf_attach_type bpf_program__get_expected_attach_type(const struct bpf_program *prog);
8809
8810 enum bpf_attach_type bpf_program__expected_attach_type(const struct bpf_program *prog)
8811 {
8812         return prog->expected_attach_type;
8813 }
8814
8815 int bpf_program__set_expected_attach_type(struct bpf_program *prog,
8816                                            enum bpf_attach_type type)
8817 {
8818         if (prog->obj->loaded)
8819                 return libbpf_err(-EBUSY);
8820
8821         prog->expected_attach_type = type;
8822         return 0;
8823 }
8824
8825 __u32 bpf_program__flags(const struct bpf_program *prog)
8826 {
8827         return prog->prog_flags;
8828 }
8829
8830 int bpf_program__set_flags(struct bpf_program *prog, __u32 flags)
8831 {
8832         if (prog->obj->loaded)
8833                 return libbpf_err(-EBUSY);
8834
8835         prog->prog_flags = flags;
8836         return 0;
8837 }
8838
8839 __u32 bpf_program__log_level(const struct bpf_program *prog)
8840 {
8841         return prog->log_level;
8842 }
8843
8844 int bpf_program__set_log_level(struct bpf_program *prog, __u32 log_level)
8845 {
8846         if (prog->obj->loaded)
8847                 return libbpf_err(-EBUSY);
8848
8849         prog->log_level = log_level;
8850         return 0;
8851 }
8852
8853 const char *bpf_program__log_buf(const struct bpf_program *prog, size_t *log_size)
8854 {
8855         *log_size = prog->log_size;
8856         return prog->log_buf;
8857 }
8858
8859 int bpf_program__set_log_buf(struct bpf_program *prog, char *log_buf, size_t log_size)
8860 {
8861         if (log_size && !log_buf)
8862                 return -EINVAL;
8863         if (prog->log_size > UINT_MAX)
8864                 return -EINVAL;
8865         if (prog->obj->loaded)
8866                 return -EBUSY;
8867
8868         prog->log_buf = log_buf;
8869         prog->log_size = log_size;
8870         return 0;
8871 }
8872
8873 #define SEC_DEF(sec_pfx, ptype, atype, flags, ...) {                        \
8874         .sec = (char *)sec_pfx,                                             \
8875         .prog_type = BPF_PROG_TYPE_##ptype,                                 \
8876         .expected_attach_type = atype,                                      \
8877         .cookie = (long)(flags),                                            \
8878         .prog_prepare_load_fn = libbpf_prepare_prog_load,                   \
8879         __VA_ARGS__                                                         \
8880 }
8881
8882 static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8883 static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8884 static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8885 static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8886 static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8887 static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8888 static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8889 static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8890 static int attach_uprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8891 static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8892 static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link);
8893
8894 static const struct bpf_sec_def section_defs[] = {
8895         SEC_DEF("socket",               SOCKET_FILTER, 0, SEC_NONE),
8896         SEC_DEF("sk_reuseport/migrate", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, SEC_ATTACHABLE),
8897         SEC_DEF("sk_reuseport",         SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT, SEC_ATTACHABLE),
8898         SEC_DEF("kprobe+",              KPROBE, 0, SEC_NONE, attach_kprobe),
8899         SEC_DEF("uprobe+",              KPROBE, 0, SEC_NONE, attach_uprobe),
8900         SEC_DEF("uprobe.s+",            KPROBE, 0, SEC_SLEEPABLE, attach_uprobe),
8901         SEC_DEF("kretprobe+",           KPROBE, 0, SEC_NONE, attach_kprobe),
8902         SEC_DEF("uretprobe+",           KPROBE, 0, SEC_NONE, attach_uprobe),
8903         SEC_DEF("uretprobe.s+",         KPROBE, 0, SEC_SLEEPABLE, attach_uprobe),
8904         SEC_DEF("kprobe.multi+",        KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi),
8905         SEC_DEF("kretprobe.multi+",     KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi),
8906         SEC_DEF("uprobe.multi+",        KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_NONE, attach_uprobe_multi),
8907         SEC_DEF("uretprobe.multi+",     KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_NONE, attach_uprobe_multi),
8908         SEC_DEF("uprobe.multi.s+",      KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_SLEEPABLE, attach_uprobe_multi),
8909         SEC_DEF("uretprobe.multi.s+",   KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_SLEEPABLE, attach_uprobe_multi),
8910         SEC_DEF("ksyscall+",            KPROBE, 0, SEC_NONE, attach_ksyscall),
8911         SEC_DEF("kretsyscall+",         KPROBE, 0, SEC_NONE, attach_ksyscall),
8912         SEC_DEF("usdt+",                KPROBE, 0, SEC_USDT, attach_usdt),
8913         SEC_DEF("usdt.s+",              KPROBE, 0, SEC_USDT | SEC_SLEEPABLE, attach_usdt),
8914         SEC_DEF("tc/ingress",           SCHED_CLS, BPF_TCX_INGRESS, SEC_NONE), /* alias for tcx */
8915         SEC_DEF("tc/egress",            SCHED_CLS, BPF_TCX_EGRESS, SEC_NONE),  /* alias for tcx */
8916         SEC_DEF("tcx/ingress",          SCHED_CLS, BPF_TCX_INGRESS, SEC_NONE),
8917         SEC_DEF("tcx/egress",           SCHED_CLS, BPF_TCX_EGRESS, SEC_NONE),
8918         SEC_DEF("tc",                   SCHED_CLS, 0, SEC_NONE), /* deprecated / legacy, use tcx */
8919         SEC_DEF("classifier",           SCHED_CLS, 0, SEC_NONE), /* deprecated / legacy, use tcx */
8920         SEC_DEF("action",               SCHED_ACT, 0, SEC_NONE), /* deprecated / legacy, use tcx */
8921         SEC_DEF("netkit/primary",       SCHED_CLS, BPF_NETKIT_PRIMARY, SEC_NONE),
8922         SEC_DEF("netkit/peer",          SCHED_CLS, BPF_NETKIT_PEER, SEC_NONE),
8923         SEC_DEF("tracepoint+",          TRACEPOINT, 0, SEC_NONE, attach_tp),
8924         SEC_DEF("tp+",                  TRACEPOINT, 0, SEC_NONE, attach_tp),
8925         SEC_DEF("raw_tracepoint+",      RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
8926         SEC_DEF("raw_tp+",              RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
8927         SEC_DEF("raw_tracepoint.w+",    RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
8928         SEC_DEF("raw_tp.w+",            RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
8929         SEC_DEF("tp_btf+",              TRACING, BPF_TRACE_RAW_TP, SEC_ATTACH_BTF, attach_trace),
8930         SEC_DEF("fentry+",              TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF, attach_trace),
8931         SEC_DEF("fmod_ret+",            TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF, attach_trace),
8932         SEC_DEF("fexit+",               TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF, attach_trace),
8933         SEC_DEF("fentry.s+",            TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8934         SEC_DEF("fmod_ret.s+",          TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8935         SEC_DEF("fexit.s+",             TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
8936         SEC_DEF("freplace+",            EXT, 0, SEC_ATTACH_BTF, attach_trace),
8937         SEC_DEF("lsm+",                 LSM, BPF_LSM_MAC, SEC_ATTACH_BTF, attach_lsm),
8938         SEC_DEF("lsm.s+",               LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm),
8939         SEC_DEF("lsm_cgroup+",          LSM, BPF_LSM_CGROUP, SEC_ATTACH_BTF),
8940         SEC_DEF("iter+",                TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter),
8941         SEC_DEF("iter.s+",              TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_iter),
8942         SEC_DEF("syscall",              SYSCALL, 0, SEC_SLEEPABLE),
8943         SEC_DEF("xdp.frags/devmap",     XDP, BPF_XDP_DEVMAP, SEC_XDP_FRAGS),
8944         SEC_DEF("xdp/devmap",           XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE),
8945         SEC_DEF("xdp.frags/cpumap",     XDP, BPF_XDP_CPUMAP, SEC_XDP_FRAGS),
8946         SEC_DEF("xdp/cpumap",           XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE),
8947         SEC_DEF("xdp.frags",            XDP, BPF_XDP, SEC_XDP_FRAGS),
8948         SEC_DEF("xdp",                  XDP, BPF_XDP, SEC_ATTACHABLE_OPT),
8949         SEC_DEF("perf_event",           PERF_EVENT, 0, SEC_NONE),
8950         SEC_DEF("lwt_in",               LWT_IN, 0, SEC_NONE),
8951         SEC_DEF("lwt_out",              LWT_OUT, 0, SEC_NONE),
8952         SEC_DEF("lwt_xmit",             LWT_XMIT, 0, SEC_NONE),
8953         SEC_DEF("lwt_seg6local",        LWT_SEG6LOCAL, 0, SEC_NONE),
8954         SEC_DEF("sockops",              SOCK_OPS, BPF_CGROUP_SOCK_OPS, SEC_ATTACHABLE_OPT),
8955         SEC_DEF("sk_skb/stream_parser", SK_SKB, BPF_SK_SKB_STREAM_PARSER, SEC_ATTACHABLE_OPT),
8956         SEC_DEF("sk_skb/stream_verdict",SK_SKB, BPF_SK_SKB_STREAM_VERDICT, SEC_ATTACHABLE_OPT),
8957         SEC_DEF("sk_skb",               SK_SKB, 0, SEC_NONE),
8958         SEC_DEF("sk_msg",               SK_MSG, BPF_SK_MSG_VERDICT, SEC_ATTACHABLE_OPT),
8959         SEC_DEF("lirc_mode2",           LIRC_MODE2, BPF_LIRC_MODE2, SEC_ATTACHABLE_OPT),
8960         SEC_DEF("flow_dissector",       FLOW_DISSECTOR, BPF_FLOW_DISSECTOR, SEC_ATTACHABLE_OPT),
8961         SEC_DEF("cgroup_skb/ingress",   CGROUP_SKB, BPF_CGROUP_INET_INGRESS, SEC_ATTACHABLE_OPT),
8962         SEC_DEF("cgroup_skb/egress",    CGROUP_SKB, BPF_CGROUP_INET_EGRESS, SEC_ATTACHABLE_OPT),
8963         SEC_DEF("cgroup/skb",           CGROUP_SKB, 0, SEC_NONE),
8964         SEC_DEF("cgroup/sock_create",   CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE),
8965         SEC_DEF("cgroup/sock_release",  CGROUP_SOCK, BPF_CGROUP_INET_SOCK_RELEASE, SEC_ATTACHABLE),
8966         SEC_DEF("cgroup/sock",          CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE_OPT),
8967         SEC_DEF("cgroup/post_bind4",    CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND, SEC_ATTACHABLE),
8968         SEC_DEF("cgroup/post_bind6",    CGROUP_SOCK, BPF_CGROUP_INET6_POST_BIND, SEC_ATTACHABLE),
8969         SEC_DEF("cgroup/bind4",         CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND, SEC_ATTACHABLE),
8970         SEC_DEF("cgroup/bind6",         CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_BIND, SEC_ATTACHABLE),
8971         SEC_DEF("cgroup/connect4",      CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_CONNECT, SEC_ATTACHABLE),
8972         SEC_DEF("cgroup/connect6",      CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_CONNECT, SEC_ATTACHABLE),
8973         SEC_DEF("cgroup/connect_unix",  CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_CONNECT, SEC_ATTACHABLE),
8974         SEC_DEF("cgroup/sendmsg4",      CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_SENDMSG, SEC_ATTACHABLE),
8975         SEC_DEF("cgroup/sendmsg6",      CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_SENDMSG, SEC_ATTACHABLE),
8976         SEC_DEF("cgroup/sendmsg_unix",  CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_SENDMSG, SEC_ATTACHABLE),
8977         SEC_DEF("cgroup/recvmsg4",      CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_RECVMSG, SEC_ATTACHABLE),
8978         SEC_DEF("cgroup/recvmsg6",      CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_RECVMSG, SEC_ATTACHABLE),
8979         SEC_DEF("cgroup/recvmsg_unix",  CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_RECVMSG, SEC_ATTACHABLE),
8980         SEC_DEF("cgroup/getpeername4",  CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETPEERNAME, SEC_ATTACHABLE),
8981         SEC_DEF("cgroup/getpeername6",  CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETPEERNAME, SEC_ATTACHABLE),
8982         SEC_DEF("cgroup/getpeername_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_GETPEERNAME, SEC_ATTACHABLE),
8983         SEC_DEF("cgroup/getsockname4",  CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETSOCKNAME, SEC_ATTACHABLE),
8984         SEC_DEF("cgroup/getsockname6",  CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETSOCKNAME, SEC_ATTACHABLE),
8985         SEC_DEF("cgroup/getsockname_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_GETSOCKNAME, SEC_ATTACHABLE),
8986         SEC_DEF("cgroup/sysctl",        CGROUP_SYSCTL, BPF_CGROUP_SYSCTL, SEC_ATTACHABLE),
8987         SEC_DEF("cgroup/getsockopt",    CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT, SEC_ATTACHABLE),
8988         SEC_DEF("cgroup/setsockopt",    CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE),
8989         SEC_DEF("cgroup/dev",           CGROUP_DEVICE, BPF_CGROUP_DEVICE, SEC_ATTACHABLE_OPT),
8990         SEC_DEF("struct_ops+",          STRUCT_OPS, 0, SEC_NONE),
8991         SEC_DEF("struct_ops.s+",        STRUCT_OPS, 0, SEC_SLEEPABLE),
8992         SEC_DEF("sk_lookup",            SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE),
8993         SEC_DEF("netfilter",            NETFILTER, BPF_NETFILTER, SEC_NONE),
8994 };
8995
8996 int libbpf_register_prog_handler(const char *sec,
8997                                  enum bpf_prog_type prog_type,
8998                                  enum bpf_attach_type exp_attach_type,
8999                                  const struct libbpf_prog_handler_opts *opts)
9000 {
9001         struct bpf_sec_def *sec_def;
9002
9003         if (!OPTS_VALID(opts, libbpf_prog_handler_opts))
9004                 return libbpf_err(-EINVAL);
9005
9006         if (last_custom_sec_def_handler_id == INT_MAX) /* prevent overflow */
9007                 return libbpf_err(-E2BIG);
9008
9009         if (sec) {
9010                 sec_def = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt + 1,
9011                                               sizeof(*sec_def));
9012                 if (!sec_def)
9013                         return libbpf_err(-ENOMEM);
9014
9015                 custom_sec_defs = sec_def;
9016                 sec_def = &custom_sec_defs[custom_sec_def_cnt];
9017         } else {
9018                 if (has_custom_fallback_def)
9019                         return libbpf_err(-EBUSY);
9020
9021                 sec_def = &custom_fallback_def;
9022         }
9023
9024         sec_def->sec = sec ? strdup(sec) : NULL;
9025         if (sec && !sec_def->sec)
9026                 return libbpf_err(-ENOMEM);
9027
9028         sec_def->prog_type = prog_type;
9029         sec_def->expected_attach_type = exp_attach_type;
9030         sec_def->cookie = OPTS_GET(opts, cookie, 0);
9031
9032         sec_def->prog_setup_fn = OPTS_GET(opts, prog_setup_fn, NULL);
9033         sec_def->prog_prepare_load_fn = OPTS_GET(opts, prog_prepare_load_fn, NULL);
9034         sec_def->prog_attach_fn = OPTS_GET(opts, prog_attach_fn, NULL);
9035
9036         sec_def->handler_id = ++last_custom_sec_def_handler_id;
9037
9038         if (sec)
9039                 custom_sec_def_cnt++;
9040         else
9041                 has_custom_fallback_def = true;
9042
9043         return sec_def->handler_id;
9044 }
9045
9046 int libbpf_unregister_prog_handler(int handler_id)
9047 {
9048         struct bpf_sec_def *sec_defs;
9049         int i;
9050
9051         if (handler_id <= 0)
9052                 return libbpf_err(-EINVAL);
9053
9054         if (has_custom_fallback_def && custom_fallback_def.handler_id == handler_id) {
9055                 memset(&custom_fallback_def, 0, sizeof(custom_fallback_def));
9056                 has_custom_fallback_def = false;
9057                 return 0;
9058         }
9059
9060         for (i = 0; i < custom_sec_def_cnt; i++) {
9061                 if (custom_sec_defs[i].handler_id == handler_id)
9062                         break;
9063         }
9064
9065         if (i == custom_sec_def_cnt)
9066                 return libbpf_err(-ENOENT);
9067
9068         free(custom_sec_defs[i].sec);
9069         for (i = i + 1; i < custom_sec_def_cnt; i++)
9070                 custom_sec_defs[i - 1] = custom_sec_defs[i];
9071         custom_sec_def_cnt--;
9072
9073         /* try to shrink the array, but it's ok if we couldn't */
9074         sec_defs = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt, sizeof(*sec_defs));
9075         /* if new count is zero, reallocarray can return a valid NULL result;
9076          * in this case the previous pointer will be freed, so we *have to*
9077          * reassign old pointer to the new value (even if it's NULL)
9078          */
9079         if (sec_defs || custom_sec_def_cnt == 0)
9080                 custom_sec_defs = sec_defs;
9081
9082         return 0;
9083 }
9084
9085 static bool sec_def_matches(const struct bpf_sec_def *sec_def, const char *sec_name)
9086 {
9087         size_t len = strlen(sec_def->sec);
9088
9089         /* "type/" always has to have proper SEC("type/extras") form */
9090         if (sec_def->sec[len - 1] == '/') {
9091                 if (str_has_pfx(sec_name, sec_def->sec))
9092                         return true;
9093                 return false;
9094         }
9095
9096         /* "type+" means it can be either exact SEC("type") or
9097          * well-formed SEC("type/extras") with proper '/' separator
9098          */
9099         if (sec_def->sec[len - 1] == '+') {
9100                 len--;
9101                 /* not even a prefix */
9102                 if (strncmp(sec_name, sec_def->sec, len) != 0)
9103                         return false;
9104                 /* exact match or has '/' separator */
9105                 if (sec_name[len] == '\0' || sec_name[len] == '/')
9106                         return true;
9107                 return false;
9108         }
9109
9110         return strcmp(sec_name, sec_def->sec) == 0;
9111 }
9112
9113 static const struct bpf_sec_def *find_sec_def(const char *sec_name)
9114 {
9115         const struct bpf_sec_def *sec_def;
9116         int i, n;
9117
9118         n = custom_sec_def_cnt;
9119         for (i = 0; i < n; i++) {
9120                 sec_def = &custom_sec_defs[i];
9121                 if (sec_def_matches(sec_def, sec_name))
9122                         return sec_def;
9123         }
9124
9125         n = ARRAY_SIZE(section_defs);
9126         for (i = 0; i < n; i++) {
9127                 sec_def = &section_defs[i];
9128                 if (sec_def_matches(sec_def, sec_name))
9129                         return sec_def;
9130         }
9131
9132         if (has_custom_fallback_def)
9133                 return &custom_fallback_def;
9134
9135         return NULL;
9136 }
9137
9138 #define MAX_TYPE_NAME_SIZE 32
9139
9140 static char *libbpf_get_type_names(bool attach_type)
9141 {
9142         int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
9143         char *buf;
9144
9145         buf = malloc(len);
9146         if (!buf)
9147                 return NULL;
9148
9149         buf[0] = '\0';
9150         /* Forge string buf with all available names */
9151         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
9152                 const struct bpf_sec_def *sec_def = &section_defs[i];
9153
9154                 if (attach_type) {
9155                         if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load)
9156                                 continue;
9157
9158                         if (!(sec_def->cookie & SEC_ATTACHABLE))
9159                                 continue;
9160                 }
9161
9162                 if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
9163                         free(buf);
9164                         return NULL;
9165                 }
9166                 strcat(buf, " ");
9167                 strcat(buf, section_defs[i].sec);
9168         }
9169
9170         return buf;
9171 }
9172
9173 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
9174                              enum bpf_attach_type *expected_attach_type)
9175 {
9176         const struct bpf_sec_def *sec_def;
9177         char *type_names;
9178
9179         if (!name)
9180                 return libbpf_err(-EINVAL);
9181
9182         sec_def = find_sec_def(name);
9183         if (sec_def) {
9184                 *prog_type = sec_def->prog_type;
9185                 *expected_attach_type = sec_def->expected_attach_type;
9186                 return 0;
9187         }
9188
9189         pr_debug("failed to guess program type from ELF section '%s'\n", name);
9190         type_names = libbpf_get_type_names(false);
9191         if (type_names != NULL) {
9192                 pr_debug("supported section(type) names are:%s\n", type_names);
9193                 free(type_names);
9194         }
9195
9196         return libbpf_err(-ESRCH);
9197 }
9198
9199 const char *libbpf_bpf_attach_type_str(enum bpf_attach_type t)
9200 {
9201         if (t < 0 || t >= ARRAY_SIZE(attach_type_name))
9202                 return NULL;
9203
9204         return attach_type_name[t];
9205 }
9206
9207 const char *libbpf_bpf_link_type_str(enum bpf_link_type t)
9208 {
9209         if (t < 0 || t >= ARRAY_SIZE(link_type_name))
9210                 return NULL;
9211
9212         return link_type_name[t];
9213 }
9214
9215 const char *libbpf_bpf_map_type_str(enum bpf_map_type t)
9216 {
9217         if (t < 0 || t >= ARRAY_SIZE(map_type_name))
9218                 return NULL;
9219
9220         return map_type_name[t];
9221 }
9222
9223 const char *libbpf_bpf_prog_type_str(enum bpf_prog_type t)
9224 {
9225         if (t < 0 || t >= ARRAY_SIZE(prog_type_name))
9226                 return NULL;
9227
9228         return prog_type_name[t];
9229 }
9230
9231 static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj,
9232                                                      int sec_idx,
9233                                                      size_t offset)
9234 {
9235         struct bpf_map *map;
9236         size_t i;
9237
9238         for (i = 0; i < obj->nr_maps; i++) {
9239                 map = &obj->maps[i];
9240                 if (!bpf_map__is_struct_ops(map))
9241                         continue;
9242                 if (map->sec_idx == sec_idx &&
9243                     map->sec_offset <= offset &&
9244                     offset - map->sec_offset < map->def.value_size)
9245                         return map;
9246         }
9247
9248         return NULL;
9249 }
9250
9251 /* Collect the reloc from ELF and populate the st_ops->progs[] */
9252 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
9253                                             Elf64_Shdr *shdr, Elf_Data *data)
9254 {
9255         const struct btf_member *member;
9256         struct bpf_struct_ops *st_ops;
9257         struct bpf_program *prog;
9258         unsigned int shdr_idx;
9259         const struct btf *btf;
9260         struct bpf_map *map;
9261         unsigned int moff, insn_idx;
9262         const char *name;
9263         __u32 member_idx;
9264         Elf64_Sym *sym;
9265         Elf64_Rel *rel;
9266         int i, nrels;
9267
9268         btf = obj->btf;
9269         nrels = shdr->sh_size / shdr->sh_entsize;
9270         for (i = 0; i < nrels; i++) {
9271                 rel = elf_rel_by_idx(data, i);
9272                 if (!rel) {
9273                         pr_warn("struct_ops reloc: failed to get %d reloc\n", i);
9274                         return -LIBBPF_ERRNO__FORMAT;
9275                 }
9276
9277                 sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
9278                 if (!sym) {
9279                         pr_warn("struct_ops reloc: symbol %zx not found\n",
9280                                 (size_t)ELF64_R_SYM(rel->r_info));
9281                         return -LIBBPF_ERRNO__FORMAT;
9282                 }
9283
9284                 name = elf_sym_str(obj, sym->st_name) ?: "<?>";
9285                 map = find_struct_ops_map_by_offset(obj, shdr->sh_info, rel->r_offset);
9286                 if (!map) {
9287                         pr_warn("struct_ops reloc: cannot find map at rel->r_offset %zu\n",
9288                                 (size_t)rel->r_offset);
9289                         return -EINVAL;
9290                 }
9291
9292                 moff = rel->r_offset - map->sec_offset;
9293                 shdr_idx = sym->st_shndx;
9294                 st_ops = map->st_ops;
9295                 pr_debug("struct_ops reloc %s: for %lld value %lld shdr_idx %u rel->r_offset %zu map->sec_offset %zu name %d (\'%s\')\n",
9296                          map->name,
9297                          (long long)(rel->r_info >> 32),
9298                          (long long)sym->st_value,
9299                          shdr_idx, (size_t)rel->r_offset,
9300                          map->sec_offset, sym->st_name, name);
9301
9302                 if (shdr_idx >= SHN_LORESERVE) {
9303                         pr_warn("struct_ops reloc %s: rel->r_offset %zu shdr_idx %u unsupported non-static function\n",
9304                                 map->name, (size_t)rel->r_offset, shdr_idx);
9305                         return -LIBBPF_ERRNO__RELOC;
9306                 }
9307                 if (sym->st_value % BPF_INSN_SZ) {
9308                         pr_warn("struct_ops reloc %s: invalid target program offset %llu\n",
9309                                 map->name, (unsigned long long)sym->st_value);
9310                         return -LIBBPF_ERRNO__FORMAT;
9311                 }
9312                 insn_idx = sym->st_value / BPF_INSN_SZ;
9313
9314                 member = find_member_by_offset(st_ops->type, moff * 8);
9315                 if (!member) {
9316                         pr_warn("struct_ops reloc %s: cannot find member at moff %u\n",
9317                                 map->name, moff);
9318                         return -EINVAL;
9319                 }
9320                 member_idx = member - btf_members(st_ops->type);
9321                 name = btf__name_by_offset(btf, member->name_off);
9322
9323                 if (!resolve_func_ptr(btf, member->type, NULL)) {
9324                         pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n",
9325                                 map->name, name);
9326                         return -EINVAL;
9327                 }
9328
9329                 prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx);
9330                 if (!prog) {
9331                         pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n",
9332                                 map->name, shdr_idx, name);
9333                         return -EINVAL;
9334                 }
9335
9336                 /* prevent the use of BPF prog with invalid type */
9337                 if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) {
9338                         pr_warn("struct_ops reloc %s: prog %s is not struct_ops BPF program\n",
9339                                 map->name, prog->name);
9340                         return -EINVAL;
9341                 }
9342
9343                 /* if we haven't yet processed this BPF program, record proper
9344                  * attach_btf_id and member_idx
9345                  */
9346                 if (!prog->attach_btf_id) {
9347                         prog->attach_btf_id = st_ops->type_id;
9348                         prog->expected_attach_type = member_idx;
9349                 }
9350
9351                 /* struct_ops BPF prog can be re-used between multiple
9352                  * .struct_ops & .struct_ops.link as long as it's the
9353                  * same struct_ops struct definition and the same
9354                  * function pointer field
9355                  */
9356                 if (prog->attach_btf_id != st_ops->type_id ||
9357                     prog->expected_attach_type != member_idx) {
9358                         pr_warn("struct_ops reloc %s: cannot use prog %s in sec %s with type %u attach_btf_id %u expected_attach_type %u for func ptr %s\n",
9359                                 map->name, prog->name, prog->sec_name, prog->type,
9360                                 prog->attach_btf_id, prog->expected_attach_type, name);
9361                         return -EINVAL;
9362                 }
9363
9364                 st_ops->progs[member_idx] = prog;
9365         }
9366
9367         return 0;
9368 }
9369
9370 #define BTF_TRACE_PREFIX "btf_trace_"
9371 #define BTF_LSM_PREFIX "bpf_lsm_"
9372 #define BTF_ITER_PREFIX "bpf_iter_"
9373 #define BTF_MAX_NAME_SIZE 128
9374
9375 void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type,
9376                                 const char **prefix, int *kind)
9377 {
9378         switch (attach_type) {
9379         case BPF_TRACE_RAW_TP:
9380                 *prefix = BTF_TRACE_PREFIX;
9381                 *kind = BTF_KIND_TYPEDEF;
9382                 break;
9383         case BPF_LSM_MAC:
9384         case BPF_LSM_CGROUP:
9385                 *prefix = BTF_LSM_PREFIX;
9386                 *kind = BTF_KIND_FUNC;
9387                 break;
9388         case BPF_TRACE_ITER:
9389                 *prefix = BTF_ITER_PREFIX;
9390                 *kind = BTF_KIND_FUNC;
9391                 break;
9392         default:
9393                 *prefix = "";
9394                 *kind = BTF_KIND_FUNC;
9395         }
9396 }
9397
9398 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
9399                                    const char *name, __u32 kind)
9400 {
9401         char btf_type_name[BTF_MAX_NAME_SIZE];
9402         int ret;
9403
9404         ret = snprintf(btf_type_name, sizeof(btf_type_name),
9405                        "%s%s", prefix, name);
9406         /* snprintf returns the number of characters written excluding the
9407          * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it
9408          * indicates truncation.
9409          */
9410         if (ret < 0 || ret >= sizeof(btf_type_name))
9411                 return -ENAMETOOLONG;
9412         return btf__find_by_name_kind(btf, btf_type_name, kind);
9413 }
9414
9415 static inline int find_attach_btf_id(struct btf *btf, const char *name,
9416                                      enum bpf_attach_type attach_type)
9417 {
9418         const char *prefix;
9419         int kind;
9420
9421         btf_get_kernel_prefix_kind(attach_type, &prefix, &kind);
9422         return find_btf_by_prefix_kind(btf, prefix, name, kind);
9423 }
9424
9425 int libbpf_find_vmlinux_btf_id(const char *name,
9426                                enum bpf_attach_type attach_type)
9427 {
9428         struct btf *btf;
9429         int err;
9430
9431         btf = btf__load_vmlinux_btf();
9432         err = libbpf_get_error(btf);
9433         if (err) {
9434                 pr_warn("vmlinux BTF is not found\n");
9435                 return libbpf_err(err);
9436         }
9437
9438         err = find_attach_btf_id(btf, name, attach_type);
9439         if (err <= 0)
9440                 pr_warn("%s is not found in vmlinux BTF\n", name);
9441
9442         btf__free(btf);
9443         return libbpf_err(err);
9444 }
9445
9446 static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
9447 {
9448         struct bpf_prog_info info;
9449         __u32 info_len = sizeof(info);
9450         struct btf *btf;
9451         int err;
9452
9453         memset(&info, 0, info_len);
9454         err = bpf_prog_get_info_by_fd(attach_prog_fd, &info, &info_len);
9455         if (err) {
9456                 pr_warn("failed bpf_prog_get_info_by_fd for FD %d: %d\n",
9457                         attach_prog_fd, err);
9458                 return err;
9459         }
9460
9461         err = -EINVAL;
9462         if (!info.btf_id) {
9463                 pr_warn("The target program doesn't have BTF\n");
9464                 goto out;
9465         }
9466         btf = btf__load_from_kernel_by_id(info.btf_id);
9467         err = libbpf_get_error(btf);
9468         if (err) {
9469                 pr_warn("Failed to get BTF %d of the program: %d\n", info.btf_id, err);
9470                 goto out;
9471         }
9472         err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
9473         btf__free(btf);
9474         if (err <= 0) {
9475                 pr_warn("%s is not found in prog's BTF\n", name);
9476                 goto out;
9477         }
9478 out:
9479         return err;
9480 }
9481
9482 static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name,
9483                               enum bpf_attach_type attach_type,
9484                               int *btf_obj_fd, int *btf_type_id)
9485 {
9486         int ret, i;
9487
9488         ret = find_attach_btf_id(obj->btf_vmlinux, attach_name, attach_type);
9489         if (ret > 0) {
9490                 *btf_obj_fd = 0; /* vmlinux BTF */
9491                 *btf_type_id = ret;
9492                 return 0;
9493         }
9494         if (ret != -ENOENT)
9495                 return ret;
9496
9497         ret = load_module_btfs(obj);
9498         if (ret)
9499                 return ret;
9500
9501         for (i = 0; i < obj->btf_module_cnt; i++) {
9502                 const struct module_btf *mod = &obj->btf_modules[i];
9503
9504                 ret = find_attach_btf_id(mod->btf, attach_name, attach_type);
9505                 if (ret > 0) {
9506                         *btf_obj_fd = mod->fd;
9507                         *btf_type_id = ret;
9508                         return 0;
9509                 }
9510                 if (ret == -ENOENT)
9511                         continue;
9512
9513                 return ret;
9514         }
9515
9516         return -ESRCH;
9517 }
9518
9519 static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
9520                                      int *btf_obj_fd, int *btf_type_id)
9521 {
9522         enum bpf_attach_type attach_type = prog->expected_attach_type;
9523         __u32 attach_prog_fd = prog->attach_prog_fd;
9524         int err = 0;
9525
9526         /* BPF program's BTF ID */
9527         if (prog->type == BPF_PROG_TYPE_EXT || attach_prog_fd) {
9528                 if (!attach_prog_fd) {
9529                         pr_warn("prog '%s': attach program FD is not set\n", prog->name);
9530                         return -EINVAL;
9531                 }
9532                 err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd);
9533                 if (err < 0) {
9534                         pr_warn("prog '%s': failed to find BPF program (FD %d) BTF ID for '%s': %d\n",
9535                                  prog->name, attach_prog_fd, attach_name, err);
9536                         return err;
9537                 }
9538                 *btf_obj_fd = 0;
9539                 *btf_type_id = err;
9540                 return 0;
9541         }
9542
9543         /* kernel/module BTF ID */
9544         if (prog->obj->gen_loader) {
9545                 bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type);
9546                 *btf_obj_fd = 0;
9547                 *btf_type_id = 1;
9548         } else {
9549                 err = find_kernel_btf_id(prog->obj, attach_name, attach_type, btf_obj_fd, btf_type_id);
9550         }
9551         if (err) {
9552                 pr_warn("prog '%s': failed to find kernel BTF type ID of '%s': %d\n",
9553                         prog->name, attach_name, err);
9554                 return err;
9555         }
9556         return 0;
9557 }
9558
9559 int libbpf_attach_type_by_name(const char *name,
9560                                enum bpf_attach_type *attach_type)
9561 {
9562         char *type_names;
9563         const struct bpf_sec_def *sec_def;
9564
9565         if (!name)
9566                 return libbpf_err(-EINVAL);
9567
9568         sec_def = find_sec_def(name);
9569         if (!sec_def) {
9570                 pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
9571                 type_names = libbpf_get_type_names(true);
9572                 if (type_names != NULL) {
9573                         pr_debug("attachable section(type) names are:%s\n", type_names);
9574                         free(type_names);
9575                 }
9576
9577                 return libbpf_err(-EINVAL);
9578         }
9579
9580         if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load)
9581                 return libbpf_err(-EINVAL);
9582         if (!(sec_def->cookie & SEC_ATTACHABLE))
9583                 return libbpf_err(-EINVAL);
9584
9585         *attach_type = sec_def->expected_attach_type;
9586         return 0;
9587 }
9588
9589 int bpf_map__fd(const struct bpf_map *map)
9590 {
9591         return map ? map->fd : libbpf_err(-EINVAL);
9592 }
9593
9594 static bool map_uses_real_name(const struct bpf_map *map)
9595 {
9596         /* Since libbpf started to support custom .data.* and .rodata.* maps,
9597          * their user-visible name differs from kernel-visible name. Users see
9598          * such map's corresponding ELF section name as a map name.
9599          * This check distinguishes .data/.rodata from .data.* and .rodata.*
9600          * maps to know which name has to be returned to the user.
9601          */
9602         if (map->libbpf_type == LIBBPF_MAP_DATA && strcmp(map->real_name, DATA_SEC) != 0)
9603                 return true;
9604         if (map->libbpf_type == LIBBPF_MAP_RODATA && strcmp(map->real_name, RODATA_SEC) != 0)
9605                 return true;
9606         return false;
9607 }
9608
9609 const char *bpf_map__name(const struct bpf_map *map)
9610 {
9611         if (!map)
9612                 return NULL;
9613
9614         if (map_uses_real_name(map))
9615                 return map->real_name;
9616
9617         return map->name;
9618 }
9619
9620 enum bpf_map_type bpf_map__type(const struct bpf_map *map)
9621 {
9622         return map->def.type;
9623 }
9624
9625 int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type)
9626 {
9627         if (map->fd >= 0)
9628                 return libbpf_err(-EBUSY);
9629         map->def.type = type;
9630         return 0;
9631 }
9632
9633 __u32 bpf_map__map_flags(const struct bpf_map *map)
9634 {
9635         return map->def.map_flags;
9636 }
9637
9638 int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags)
9639 {
9640         if (map->fd >= 0)
9641                 return libbpf_err(-EBUSY);
9642         map->def.map_flags = flags;
9643         return 0;
9644 }
9645
9646 __u64 bpf_map__map_extra(const struct bpf_map *map)
9647 {
9648         return map->map_extra;
9649 }
9650
9651 int bpf_map__set_map_extra(struct bpf_map *map, __u64 map_extra)
9652 {
9653         if (map->fd >= 0)
9654                 return libbpf_err(-EBUSY);
9655         map->map_extra = map_extra;
9656         return 0;
9657 }
9658
9659 __u32 bpf_map__numa_node(const struct bpf_map *map)
9660 {
9661         return map->numa_node;
9662 }
9663
9664 int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node)
9665 {
9666         if (map->fd >= 0)
9667                 return libbpf_err(-EBUSY);
9668         map->numa_node = numa_node;
9669         return 0;
9670 }
9671
9672 __u32 bpf_map__key_size(const struct bpf_map *map)
9673 {
9674         return map->def.key_size;
9675 }
9676
9677 int bpf_map__set_key_size(struct bpf_map *map, __u32 size)
9678 {
9679         if (map->fd >= 0)
9680                 return libbpf_err(-EBUSY);
9681         map->def.key_size = size;
9682         return 0;
9683 }
9684
9685 __u32 bpf_map__value_size(const struct bpf_map *map)
9686 {
9687         return map->def.value_size;
9688 }
9689
9690 static int map_btf_datasec_resize(struct bpf_map *map, __u32 size)
9691 {
9692         struct btf *btf;
9693         struct btf_type *datasec_type, *var_type;
9694         struct btf_var_secinfo *var;
9695         const struct btf_type *array_type;
9696         const struct btf_array *array;
9697         int vlen, element_sz, new_array_id;
9698         __u32 nr_elements;
9699
9700         /* check btf existence */
9701         btf = bpf_object__btf(map->obj);
9702         if (!btf)
9703                 return -ENOENT;
9704
9705         /* verify map is datasec */
9706         datasec_type = btf_type_by_id(btf, bpf_map__btf_value_type_id(map));
9707         if (!btf_is_datasec(datasec_type)) {
9708                 pr_warn("map '%s': cannot be resized, map value type is not a datasec\n",
9709                         bpf_map__name(map));
9710                 return -EINVAL;
9711         }
9712
9713         /* verify datasec has at least one var */
9714         vlen = btf_vlen(datasec_type);
9715         if (vlen == 0) {
9716                 pr_warn("map '%s': cannot be resized, map value datasec is empty\n",
9717                         bpf_map__name(map));
9718                 return -EINVAL;
9719         }
9720
9721         /* verify last var in the datasec is an array */
9722         var = &btf_var_secinfos(datasec_type)[vlen - 1];
9723         var_type = btf_type_by_id(btf, var->type);
9724         array_type = skip_mods_and_typedefs(btf, var_type->type, NULL);
9725         if (!btf_is_array(array_type)) {
9726                 pr_warn("map '%s': cannot be resized, last var must be an array\n",
9727                         bpf_map__name(map));
9728                 return -EINVAL;
9729         }
9730
9731         /* verify request size aligns with array */
9732         array = btf_array(array_type);
9733         element_sz = btf__resolve_size(btf, array->type);
9734         if (element_sz <= 0 || (size - var->offset) % element_sz != 0) {
9735                 pr_warn("map '%s': cannot be resized, element size (%d) doesn't align with new total size (%u)\n",
9736                         bpf_map__name(map), element_sz, size);
9737                 return -EINVAL;
9738         }
9739
9740         /* create a new array based on the existing array, but with new length */
9741         nr_elements = (size - var->offset) / element_sz;
9742         new_array_id = btf__add_array(btf, array->index_type, array->type, nr_elements);
9743         if (new_array_id < 0)
9744                 return new_array_id;
9745
9746         /* adding a new btf type invalidates existing pointers to btf objects,
9747          * so refresh pointers before proceeding
9748          */
9749         datasec_type = btf_type_by_id(btf, map->btf_value_type_id);
9750         var = &btf_var_secinfos(datasec_type)[vlen - 1];
9751         var_type = btf_type_by_id(btf, var->type);
9752
9753         /* finally update btf info */
9754         datasec_type->size = size;
9755         var->size = size - var->offset;
9756         var_type->type = new_array_id;
9757
9758         return 0;
9759 }
9760
9761 int bpf_map__set_value_size(struct bpf_map *map, __u32 size)
9762 {
9763         if (map->fd >= 0)
9764                 return libbpf_err(-EBUSY);
9765
9766         if (map->mmaped) {
9767                 int err;
9768                 size_t mmap_old_sz, mmap_new_sz;
9769
9770                 mmap_old_sz = bpf_map_mmap_sz(map->def.value_size, map->def.max_entries);
9771                 mmap_new_sz = bpf_map_mmap_sz(size, map->def.max_entries);
9772                 err = bpf_map_mmap_resize(map, mmap_old_sz, mmap_new_sz);
9773                 if (err) {
9774                         pr_warn("map '%s': failed to resize memory-mapped region: %d\n",
9775                                 bpf_map__name(map), err);
9776                         return err;
9777                 }
9778                 err = map_btf_datasec_resize(map, size);
9779                 if (err && err != -ENOENT) {
9780                         pr_warn("map '%s': failed to adjust resized BTF, clearing BTF key/value info: %d\n",
9781                                 bpf_map__name(map), err);
9782                         map->btf_value_type_id = 0;
9783                         map->btf_key_type_id = 0;
9784                 }
9785         }
9786
9787         map->def.value_size = size;
9788         return 0;
9789 }
9790
9791 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
9792 {
9793         return map ? map->btf_key_type_id : 0;
9794 }
9795
9796 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
9797 {
9798         return map ? map->btf_value_type_id : 0;
9799 }
9800
9801 int bpf_map__set_initial_value(struct bpf_map *map,
9802                                const void *data, size_t size)
9803 {
9804         if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG ||
9805             size != map->def.value_size || map->fd >= 0)
9806                 return libbpf_err(-EINVAL);
9807
9808         memcpy(map->mmaped, data, size);
9809         return 0;
9810 }
9811
9812 void *bpf_map__initial_value(struct bpf_map *map, size_t *psize)
9813 {
9814         if (!map->mmaped)
9815                 return NULL;
9816         *psize = map->def.value_size;
9817         return map->mmaped;
9818 }
9819
9820 bool bpf_map__is_internal(const struct bpf_map *map)
9821 {
9822         return map->libbpf_type != LIBBPF_MAP_UNSPEC;
9823 }
9824
9825 __u32 bpf_map__ifindex(const struct bpf_map *map)
9826 {
9827         return map->map_ifindex;
9828 }
9829
9830 int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
9831 {
9832         if (map->fd >= 0)
9833                 return libbpf_err(-EBUSY);
9834         map->map_ifindex = ifindex;
9835         return 0;
9836 }
9837
9838 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
9839 {
9840         if (!bpf_map_type__is_map_in_map(map->def.type)) {
9841                 pr_warn("error: unsupported map type\n");
9842                 return libbpf_err(-EINVAL);
9843         }
9844         if (map->inner_map_fd != -1) {
9845                 pr_warn("error: inner_map_fd already specified\n");
9846                 return libbpf_err(-EINVAL);
9847         }
9848         if (map->inner_map) {
9849                 bpf_map__destroy(map->inner_map);
9850                 zfree(&map->inner_map);
9851         }
9852         map->inner_map_fd = fd;
9853         return 0;
9854 }
9855
9856 static struct bpf_map *
9857 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
9858 {
9859         ssize_t idx;
9860         struct bpf_map *s, *e;
9861
9862         if (!obj || !obj->maps)
9863                 return errno = EINVAL, NULL;
9864
9865         s = obj->maps;
9866         e = obj->maps + obj->nr_maps;
9867
9868         if ((m < s) || (m >= e)) {
9869                 pr_warn("error in %s: map handler doesn't belong to object\n",
9870                          __func__);
9871                 return errno = EINVAL, NULL;
9872         }
9873
9874         idx = (m - obj->maps) + i;
9875         if (idx >= obj->nr_maps || idx < 0)
9876                 return NULL;
9877         return &obj->maps[idx];
9878 }
9879
9880 struct bpf_map *
9881 bpf_object__next_map(const struct bpf_object *obj, const struct bpf_map *prev)
9882 {
9883         if (prev == NULL)
9884                 return obj->maps;
9885
9886         return __bpf_map__iter(prev, obj, 1);
9887 }
9888
9889 struct bpf_map *
9890 bpf_object__prev_map(const struct bpf_object *obj, const struct bpf_map *next)
9891 {
9892         if (next == NULL) {
9893                 if (!obj->nr_maps)
9894                         return NULL;
9895                 return obj->maps + obj->nr_maps - 1;
9896         }
9897
9898         return __bpf_map__iter(next, obj, -1);
9899 }
9900
9901 struct bpf_map *
9902 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
9903 {
9904         struct bpf_map *pos;
9905
9906         bpf_object__for_each_map(pos, obj) {
9907                 /* if it's a special internal map name (which always starts
9908                  * with dot) then check if that special name matches the
9909                  * real map name (ELF section name)
9910                  */
9911                 if (name[0] == '.') {
9912                         if (pos->real_name && strcmp(pos->real_name, name) == 0)
9913                                 return pos;
9914                         continue;
9915                 }
9916                 /* otherwise map name has to be an exact match */
9917                 if (map_uses_real_name(pos)) {
9918                         if (strcmp(pos->real_name, name) == 0)
9919                                 return pos;
9920                         continue;
9921                 }
9922                 if (strcmp(pos->name, name) == 0)
9923                         return pos;
9924         }
9925         return errno = ENOENT, NULL;
9926 }
9927
9928 int
9929 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
9930 {
9931         return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
9932 }
9933
9934 static int validate_map_op(const struct bpf_map *map, size_t key_sz,
9935                            size_t value_sz, bool check_value_sz)
9936 {
9937         if (map->fd <= 0)
9938                 return -ENOENT;
9939
9940         if (map->def.key_size != key_sz) {
9941                 pr_warn("map '%s': unexpected key size %zu provided, expected %u\n",
9942                         map->name, key_sz, map->def.key_size);
9943                 return -EINVAL;
9944         }
9945
9946         if (!check_value_sz)
9947                 return 0;
9948
9949         switch (map->def.type) {
9950         case BPF_MAP_TYPE_PERCPU_ARRAY:
9951         case BPF_MAP_TYPE_PERCPU_HASH:
9952         case BPF_MAP_TYPE_LRU_PERCPU_HASH:
9953         case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: {
9954                 int num_cpu = libbpf_num_possible_cpus();
9955                 size_t elem_sz = roundup(map->def.value_size, 8);
9956
9957                 if (value_sz != num_cpu * elem_sz) {
9958                         pr_warn("map '%s': unexpected value size %zu provided for per-CPU map, expected %d * %zu = %zd\n",
9959                                 map->name, value_sz, num_cpu, elem_sz, num_cpu * elem_sz);
9960                         return -EINVAL;
9961                 }
9962                 break;
9963         }
9964         default:
9965                 if (map->def.value_size != value_sz) {
9966                         pr_warn("map '%s': unexpected value size %zu provided, expected %u\n",
9967                                 map->name, value_sz, map->def.value_size);
9968                         return -EINVAL;
9969                 }
9970                 break;
9971         }
9972         return 0;
9973 }
9974
9975 int bpf_map__lookup_elem(const struct bpf_map *map,
9976                          const void *key, size_t key_sz,
9977                          void *value, size_t value_sz, __u64 flags)
9978 {
9979         int err;
9980
9981         err = validate_map_op(map, key_sz, value_sz, true);
9982         if (err)
9983                 return libbpf_err(err);
9984
9985         return bpf_map_lookup_elem_flags(map->fd, key, value, flags);
9986 }
9987
9988 int bpf_map__update_elem(const struct bpf_map *map,
9989                          const void *key, size_t key_sz,
9990                          const void *value, size_t value_sz, __u64 flags)
9991 {
9992         int err;
9993
9994         err = validate_map_op(map, key_sz, value_sz, true);
9995         if (err)
9996                 return libbpf_err(err);
9997
9998         return bpf_map_update_elem(map->fd, key, value, flags);
9999 }
10000
10001 int bpf_map__delete_elem(const struct bpf_map *map,
10002                          const void *key, size_t key_sz, __u64 flags)
10003 {
10004         int err;
10005
10006         err = validate_map_op(map, key_sz, 0, false /* check_value_sz */);
10007         if (err)
10008                 return libbpf_err(err);
10009
10010         return bpf_map_delete_elem_flags(map->fd, key, flags);
10011 }
10012
10013 int bpf_map__lookup_and_delete_elem(const struct bpf_map *map,
10014                                     const void *key, size_t key_sz,
10015                                     void *value, size_t value_sz, __u64 flags)
10016 {
10017         int err;
10018
10019         err = validate_map_op(map, key_sz, value_sz, true);
10020         if (err)
10021                 return libbpf_err(err);
10022
10023         return bpf_map_lookup_and_delete_elem_flags(map->fd, key, value, flags);
10024 }
10025
10026 int bpf_map__get_next_key(const struct bpf_map *map,
10027                           const void *cur_key, void *next_key, size_t key_sz)
10028 {
10029         int err;
10030
10031         err = validate_map_op(map, key_sz, 0, false /* check_value_sz */);
10032         if (err)
10033                 return libbpf_err(err);
10034
10035         return bpf_map_get_next_key(map->fd, cur_key, next_key);
10036 }
10037
10038 long libbpf_get_error(const void *ptr)
10039 {
10040         if (!IS_ERR_OR_NULL(ptr))
10041                 return 0;
10042
10043         if (IS_ERR(ptr))
10044                 errno = -PTR_ERR(ptr);
10045
10046         /* If ptr == NULL, then errno should be already set by the failing
10047          * API, because libbpf never returns NULL on success and it now always
10048          * sets errno on error. So no extra errno handling for ptr == NULL
10049          * case.
10050          */
10051         return -errno;
10052 }
10053
10054 /* Replace link's underlying BPF program with the new one */
10055 int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog)
10056 {
10057         int ret;
10058
10059         ret = bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL);
10060         return libbpf_err_errno(ret);
10061 }
10062
10063 /* Release "ownership" of underlying BPF resource (typically, BPF program
10064  * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
10065  * link, when destructed through bpf_link__destroy() call won't attempt to
10066  * detach/unregisted that BPF resource. This is useful in situations where,
10067  * say, attached BPF program has to outlive userspace program that attached it
10068  * in the system. Depending on type of BPF program, though, there might be
10069  * additional steps (like pinning BPF program in BPF FS) necessary to ensure
10070  * exit of userspace program doesn't trigger automatic detachment and clean up
10071  * inside the kernel.
10072  */
10073 void bpf_link__disconnect(struct bpf_link *link)
10074 {
10075         link->disconnected = true;
10076 }
10077
10078 int bpf_link__destroy(struct bpf_link *link)
10079 {
10080         int err = 0;
10081
10082         if (IS_ERR_OR_NULL(link))
10083                 return 0;
10084
10085         if (!link->disconnected && link->detach)
10086                 err = link->detach(link);
10087         if (link->pin_path)
10088                 free(link->pin_path);
10089         if (link->dealloc)
10090                 link->dealloc(link);
10091         else
10092                 free(link);
10093
10094         return libbpf_err(err);
10095 }
10096
10097 int bpf_link__fd(const struct bpf_link *link)
10098 {
10099         return link->fd;
10100 }
10101
10102 const char *bpf_link__pin_path(const struct bpf_link *link)
10103 {
10104         return link->pin_path;
10105 }
10106
10107 static int bpf_link__detach_fd(struct bpf_link *link)
10108 {
10109         return libbpf_err_errno(close(link->fd));
10110 }
10111
10112 struct bpf_link *bpf_link__open(const char *path)
10113 {
10114         struct bpf_link *link;
10115         int fd;
10116
10117         fd = bpf_obj_get(path);
10118         if (fd < 0) {
10119                 fd = -errno;
10120                 pr_warn("failed to open link at %s: %d\n", path, fd);
10121                 return libbpf_err_ptr(fd);
10122         }
10123
10124         link = calloc(1, sizeof(*link));
10125         if (!link) {
10126                 close(fd);
10127                 return libbpf_err_ptr(-ENOMEM);
10128         }
10129         link->detach = &bpf_link__detach_fd;
10130         link->fd = fd;
10131
10132         link->pin_path = strdup(path);
10133         if (!link->pin_path) {
10134                 bpf_link__destroy(link);
10135                 return libbpf_err_ptr(-ENOMEM);
10136         }
10137
10138         return link;
10139 }
10140
10141 int bpf_link__detach(struct bpf_link *link)
10142 {
10143         return bpf_link_detach(link->fd) ? -errno : 0;
10144 }
10145
10146 int bpf_link__pin(struct bpf_link *link, const char *path)
10147 {
10148         int err;
10149
10150         if (link->pin_path)
10151                 return libbpf_err(-EBUSY);
10152         err = make_parent_dir(path);
10153         if (err)
10154                 return libbpf_err(err);
10155         err = check_path(path);
10156         if (err)
10157                 return libbpf_err(err);
10158
10159         link->pin_path = strdup(path);
10160         if (!link->pin_path)
10161                 return libbpf_err(-ENOMEM);
10162
10163         if (bpf_obj_pin(link->fd, link->pin_path)) {
10164                 err = -errno;
10165                 zfree(&link->pin_path);
10166                 return libbpf_err(err);
10167         }
10168
10169         pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path);
10170         return 0;
10171 }
10172
10173 int bpf_link__unpin(struct bpf_link *link)
10174 {
10175         int err;
10176
10177         if (!link->pin_path)
10178                 return libbpf_err(-EINVAL);
10179
10180         err = unlink(link->pin_path);
10181         if (err != 0)
10182                 return -errno;
10183
10184         pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path);
10185         zfree(&link->pin_path);
10186         return 0;
10187 }
10188
10189 struct bpf_link_perf {
10190         struct bpf_link link;
10191         int perf_event_fd;
10192         /* legacy kprobe support: keep track of probe identifier and type */
10193         char *legacy_probe_name;
10194         bool legacy_is_kprobe;
10195         bool legacy_is_retprobe;
10196 };
10197
10198 static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe);
10199 static int remove_uprobe_event_legacy(const char *probe_name, bool retprobe);
10200
10201 static int bpf_link_perf_detach(struct bpf_link *link)
10202 {
10203         struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
10204         int err = 0;
10205
10206         if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0)
10207                 err = -errno;
10208
10209         if (perf_link->perf_event_fd != link->fd)
10210                 close(perf_link->perf_event_fd);
10211         close(link->fd);
10212
10213         /* legacy uprobe/kprobe needs to be removed after perf event fd closure */
10214         if (perf_link->legacy_probe_name) {
10215                 if (perf_link->legacy_is_kprobe) {
10216                         err = remove_kprobe_event_legacy(perf_link->legacy_probe_name,
10217                                                          perf_link->legacy_is_retprobe);
10218                 } else {
10219                         err = remove_uprobe_event_legacy(perf_link->legacy_probe_name,
10220                                                          perf_link->legacy_is_retprobe);
10221                 }
10222         }
10223
10224         return err;
10225 }
10226
10227 static void bpf_link_perf_dealloc(struct bpf_link *link)
10228 {
10229         struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
10230
10231         free(perf_link->legacy_probe_name);
10232         free(perf_link);
10233 }
10234
10235 struct bpf_link *bpf_program__attach_perf_event_opts(const struct bpf_program *prog, int pfd,
10236                                                      const struct bpf_perf_event_opts *opts)
10237 {
10238         char errmsg[STRERR_BUFSIZE];
10239         struct bpf_link_perf *link;
10240         int prog_fd, link_fd = -1, err;
10241         bool force_ioctl_attach;
10242
10243         if (!OPTS_VALID(opts, bpf_perf_event_opts))
10244                 return libbpf_err_ptr(-EINVAL);
10245
10246         if (pfd < 0) {
10247                 pr_warn("prog '%s': invalid perf event FD %d\n",
10248                         prog->name, pfd);
10249                 return libbpf_err_ptr(-EINVAL);
10250         }
10251         prog_fd = bpf_program__fd(prog);
10252         if (prog_fd < 0) {
10253                 pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n",
10254                         prog->name);
10255                 return libbpf_err_ptr(-EINVAL);
10256         }
10257
10258         link = calloc(1, sizeof(*link));
10259         if (!link)
10260                 return libbpf_err_ptr(-ENOMEM);
10261         link->link.detach = &bpf_link_perf_detach;
10262         link->link.dealloc = &bpf_link_perf_dealloc;
10263         link->perf_event_fd = pfd;
10264
10265         force_ioctl_attach = OPTS_GET(opts, force_ioctl_attach, false);
10266         if (kernel_supports(prog->obj, FEAT_PERF_LINK) && !force_ioctl_attach) {
10267                 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts,
10268                         .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0));
10269
10270                 link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts);
10271                 if (link_fd < 0) {
10272                         err = -errno;
10273                         pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %d (%s)\n",
10274                                 prog->name, pfd,
10275                                 err, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10276                         goto err_out;
10277                 }
10278                 link->link.fd = link_fd;
10279         } else {
10280                 if (OPTS_GET(opts, bpf_cookie, 0)) {
10281                         pr_warn("prog '%s': user context value is not supported\n", prog->name);
10282                         err = -EOPNOTSUPP;
10283                         goto err_out;
10284                 }
10285
10286                 if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
10287                         err = -errno;
10288                         pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n",
10289                                 prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10290                         if (err == -EPROTO)
10291                                 pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n",
10292                                         prog->name, pfd);
10293                         goto err_out;
10294                 }
10295                 link->link.fd = pfd;
10296         }
10297         if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
10298                 err = -errno;
10299                 pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n",
10300                         prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10301                 goto err_out;
10302         }
10303
10304         return &link->link;
10305 err_out:
10306         if (link_fd >= 0)
10307                 close(link_fd);
10308         free(link);
10309         return libbpf_err_ptr(err);
10310 }
10311
10312 struct bpf_link *bpf_program__attach_perf_event(const struct bpf_program *prog, int pfd)
10313 {
10314         return bpf_program__attach_perf_event_opts(prog, pfd, NULL);
10315 }
10316
10317 /*
10318  * this function is expected to parse integer in the range of [0, 2^31-1] from
10319  * given file using scanf format string fmt. If actual parsed value is
10320  * negative, the result might be indistinguishable from error
10321  */
10322 static int parse_uint_from_file(const char *file, const char *fmt)
10323 {
10324         char buf[STRERR_BUFSIZE];
10325         int err, ret;
10326         FILE *f;
10327
10328         f = fopen(file, "re");
10329         if (!f) {
10330                 err = -errno;
10331                 pr_debug("failed to open '%s': %s\n", file,
10332                          libbpf_strerror_r(err, buf, sizeof(buf)));
10333                 return err;
10334         }
10335         err = fscanf(f, fmt, &ret);
10336         if (err != 1) {
10337                 err = err == EOF ? -EIO : -errno;
10338                 pr_debug("failed to parse '%s': %s\n", file,
10339                         libbpf_strerror_r(err, buf, sizeof(buf)));
10340                 fclose(f);
10341                 return err;
10342         }
10343         fclose(f);
10344         return ret;
10345 }
10346
10347 static int determine_kprobe_perf_type(void)
10348 {
10349         const char *file = "/sys/bus/event_source/devices/kprobe/type";
10350
10351         return parse_uint_from_file(file, "%d\n");
10352 }
10353
10354 static int determine_uprobe_perf_type(void)
10355 {
10356         const char *file = "/sys/bus/event_source/devices/uprobe/type";
10357
10358         return parse_uint_from_file(file, "%d\n");
10359 }
10360
10361 static int determine_kprobe_retprobe_bit(void)
10362 {
10363         const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
10364
10365         return parse_uint_from_file(file, "config:%d\n");
10366 }
10367
10368 static int determine_uprobe_retprobe_bit(void)
10369 {
10370         const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
10371
10372         return parse_uint_from_file(file, "config:%d\n");
10373 }
10374
10375 #define PERF_UPROBE_REF_CTR_OFFSET_BITS 32
10376 #define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32
10377
10378 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
10379                                  uint64_t offset, int pid, size_t ref_ctr_off)
10380 {
10381         const size_t attr_sz = sizeof(struct perf_event_attr);
10382         struct perf_event_attr attr;
10383         char errmsg[STRERR_BUFSIZE];
10384         int type, pfd;
10385
10386         if ((__u64)ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS))
10387                 return -EINVAL;
10388
10389         memset(&attr, 0, attr_sz);
10390
10391         type = uprobe ? determine_uprobe_perf_type()
10392                       : determine_kprobe_perf_type();
10393         if (type < 0) {
10394                 pr_warn("failed to determine %s perf type: %s\n",
10395                         uprobe ? "uprobe" : "kprobe",
10396                         libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
10397                 return type;
10398         }
10399         if (retprobe) {
10400                 int bit = uprobe ? determine_uprobe_retprobe_bit()
10401                                  : determine_kprobe_retprobe_bit();
10402
10403                 if (bit < 0) {
10404                         pr_warn("failed to determine %s retprobe bit: %s\n",
10405                                 uprobe ? "uprobe" : "kprobe",
10406                                 libbpf_strerror_r(bit, errmsg, sizeof(errmsg)));
10407                         return bit;
10408                 }
10409                 attr.config |= 1 << bit;
10410         }
10411         attr.size = attr_sz;
10412         attr.type = type;
10413         attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT;
10414         attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
10415         attr.config2 = offset;           /* kprobe_addr or probe_offset */
10416
10417         /* pid filter is meaningful only for uprobes */
10418         pfd = syscall(__NR_perf_event_open, &attr,
10419                       pid < 0 ? -1 : pid /* pid */,
10420                       pid == -1 ? 0 : -1 /* cpu */,
10421                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
10422         return pfd >= 0 ? pfd : -errno;
10423 }
10424
10425 static int append_to_file(const char *file, const char *fmt, ...)
10426 {
10427         int fd, n, err = 0;
10428         va_list ap;
10429         char buf[1024];
10430
10431         va_start(ap, fmt);
10432         n = vsnprintf(buf, sizeof(buf), fmt, ap);
10433         va_end(ap);
10434
10435         if (n < 0 || n >= sizeof(buf))
10436                 return -EINVAL;
10437
10438         fd = open(file, O_WRONLY | O_APPEND | O_CLOEXEC, 0);
10439         if (fd < 0)
10440                 return -errno;
10441
10442         if (write(fd, buf, n) < 0)
10443                 err = -errno;
10444
10445         close(fd);
10446         return err;
10447 }
10448
10449 #define DEBUGFS "/sys/kernel/debug/tracing"
10450 #define TRACEFS "/sys/kernel/tracing"
10451
10452 static bool use_debugfs(void)
10453 {
10454         static int has_debugfs = -1;
10455
10456         if (has_debugfs < 0)
10457                 has_debugfs = faccessat(AT_FDCWD, DEBUGFS, F_OK, AT_EACCESS) == 0;
10458
10459         return has_debugfs == 1;
10460 }
10461
10462 static const char *tracefs_path(void)
10463 {
10464         return use_debugfs() ? DEBUGFS : TRACEFS;
10465 }
10466
10467 static const char *tracefs_kprobe_events(void)
10468 {
10469         return use_debugfs() ? DEBUGFS"/kprobe_events" : TRACEFS"/kprobe_events";
10470 }
10471
10472 static const char *tracefs_uprobe_events(void)
10473 {
10474         return use_debugfs() ? DEBUGFS"/uprobe_events" : TRACEFS"/uprobe_events";
10475 }
10476
10477 static const char *tracefs_available_filter_functions(void)
10478 {
10479         return use_debugfs() ? DEBUGFS"/available_filter_functions"
10480                              : TRACEFS"/available_filter_functions";
10481 }
10482
10483 static const char *tracefs_available_filter_functions_addrs(void)
10484 {
10485         return use_debugfs() ? DEBUGFS"/available_filter_functions_addrs"
10486                              : TRACEFS"/available_filter_functions_addrs";
10487 }
10488
10489 static void gen_kprobe_legacy_event_name(char *buf, size_t buf_sz,
10490                                          const char *kfunc_name, size_t offset)
10491 {
10492         static int index = 0;
10493         int i;
10494
10495         snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx_%d", getpid(), kfunc_name, offset,
10496                  __sync_fetch_and_add(&index, 1));
10497
10498         /* sanitize binary_path in the probe name */
10499         for (i = 0; buf[i]; i++) {
10500                 if (!isalnum(buf[i]))
10501                         buf[i] = '_';
10502         }
10503 }
10504
10505 static int add_kprobe_event_legacy(const char *probe_name, bool retprobe,
10506                                    const char *kfunc_name, size_t offset)
10507 {
10508         return append_to_file(tracefs_kprobe_events(), "%c:%s/%s %s+0x%zx",
10509                               retprobe ? 'r' : 'p',
10510                               retprobe ? "kretprobes" : "kprobes",
10511                               probe_name, kfunc_name, offset);
10512 }
10513
10514 static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe)
10515 {
10516         return append_to_file(tracefs_kprobe_events(), "-:%s/%s",
10517                               retprobe ? "kretprobes" : "kprobes", probe_name);
10518 }
10519
10520 static int determine_kprobe_perf_type_legacy(const char *probe_name, bool retprobe)
10521 {
10522         char file[256];
10523
10524         snprintf(file, sizeof(file), "%s/events/%s/%s/id",
10525                  tracefs_path(), retprobe ? "kretprobes" : "kprobes", probe_name);
10526
10527         return parse_uint_from_file(file, "%d\n");
10528 }
10529
10530 static int perf_event_kprobe_open_legacy(const char *probe_name, bool retprobe,
10531                                          const char *kfunc_name, size_t offset, int pid)
10532 {
10533         const size_t attr_sz = sizeof(struct perf_event_attr);
10534         struct perf_event_attr attr;
10535         char errmsg[STRERR_BUFSIZE];
10536         int type, pfd, err;
10537
10538         err = add_kprobe_event_legacy(probe_name, retprobe, kfunc_name, offset);
10539         if (err < 0) {
10540                 pr_warn("failed to add legacy kprobe event for '%s+0x%zx': %s\n",
10541                         kfunc_name, offset,
10542                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10543                 return err;
10544         }
10545         type = determine_kprobe_perf_type_legacy(probe_name, retprobe);
10546         if (type < 0) {
10547                 err = type;
10548                 pr_warn("failed to determine legacy kprobe event id for '%s+0x%zx': %s\n",
10549                         kfunc_name, offset,
10550                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10551                 goto err_clean_legacy;
10552         }
10553
10554         memset(&attr, 0, attr_sz);
10555         attr.size = attr_sz;
10556         attr.config = type;
10557         attr.type = PERF_TYPE_TRACEPOINT;
10558
10559         pfd = syscall(__NR_perf_event_open, &attr,
10560                       pid < 0 ? -1 : pid, /* pid */
10561                       pid == -1 ? 0 : -1, /* cpu */
10562                       -1 /* group_fd */,  PERF_FLAG_FD_CLOEXEC);
10563         if (pfd < 0) {
10564                 err = -errno;
10565                 pr_warn("legacy kprobe perf_event_open() failed: %s\n",
10566                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10567                 goto err_clean_legacy;
10568         }
10569         return pfd;
10570
10571 err_clean_legacy:
10572         /* Clear the newly added legacy kprobe_event */
10573         remove_kprobe_event_legacy(probe_name, retprobe);
10574         return err;
10575 }
10576
10577 static const char *arch_specific_syscall_pfx(void)
10578 {
10579 #if defined(__x86_64__)
10580         return "x64";
10581 #elif defined(__i386__)
10582         return "ia32";
10583 #elif defined(__s390x__)
10584         return "s390x";
10585 #elif defined(__s390__)
10586         return "s390";
10587 #elif defined(__arm__)
10588         return "arm";
10589 #elif defined(__aarch64__)
10590         return "arm64";
10591 #elif defined(__mips__)
10592         return "mips";
10593 #elif defined(__riscv)
10594         return "riscv";
10595 #elif defined(__powerpc__)
10596         return "powerpc";
10597 #elif defined(__powerpc64__)
10598         return "powerpc64";
10599 #else
10600         return NULL;
10601 #endif
10602 }
10603
10604 static int probe_kern_syscall_wrapper(void)
10605 {
10606         char syscall_name[64];
10607         const char *ksys_pfx;
10608
10609         ksys_pfx = arch_specific_syscall_pfx();
10610         if (!ksys_pfx)
10611                 return 0;
10612
10613         snprintf(syscall_name, sizeof(syscall_name), "__%s_sys_bpf", ksys_pfx);
10614
10615         if (determine_kprobe_perf_type() >= 0) {
10616                 int pfd;
10617
10618                 pfd = perf_event_open_probe(false, false, syscall_name, 0, getpid(), 0);
10619                 if (pfd >= 0)
10620                         close(pfd);
10621
10622                 return pfd >= 0 ? 1 : 0;
10623         } else { /* legacy mode */
10624                 char probe_name[128];
10625
10626                 gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name), syscall_name, 0);
10627                 if (add_kprobe_event_legacy(probe_name, false, syscall_name, 0) < 0)
10628                         return 0;
10629
10630                 (void)remove_kprobe_event_legacy(probe_name, false);
10631                 return 1;
10632         }
10633 }
10634
10635 struct bpf_link *
10636 bpf_program__attach_kprobe_opts(const struct bpf_program *prog,
10637                                 const char *func_name,
10638                                 const struct bpf_kprobe_opts *opts)
10639 {
10640         DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
10641         enum probe_attach_mode attach_mode;
10642         char errmsg[STRERR_BUFSIZE];
10643         char *legacy_probe = NULL;
10644         struct bpf_link *link;
10645         size_t offset;
10646         bool retprobe, legacy;
10647         int pfd, err;
10648
10649         if (!OPTS_VALID(opts, bpf_kprobe_opts))
10650                 return libbpf_err_ptr(-EINVAL);
10651
10652         attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT);
10653         retprobe = OPTS_GET(opts, retprobe, false);
10654         offset = OPTS_GET(opts, offset, 0);
10655         pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
10656
10657         legacy = determine_kprobe_perf_type() < 0;
10658         switch (attach_mode) {
10659         case PROBE_ATTACH_MODE_LEGACY:
10660                 legacy = true;
10661                 pe_opts.force_ioctl_attach = true;
10662                 break;
10663         case PROBE_ATTACH_MODE_PERF:
10664                 if (legacy)
10665                         return libbpf_err_ptr(-ENOTSUP);
10666                 pe_opts.force_ioctl_attach = true;
10667                 break;
10668         case PROBE_ATTACH_MODE_LINK:
10669                 if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK))
10670                         return libbpf_err_ptr(-ENOTSUP);
10671                 break;
10672         case PROBE_ATTACH_MODE_DEFAULT:
10673                 break;
10674         default:
10675                 return libbpf_err_ptr(-EINVAL);
10676         }
10677
10678         if (!legacy) {
10679                 pfd = perf_event_open_probe(false /* uprobe */, retprobe,
10680                                             func_name, offset,
10681                                             -1 /* pid */, 0 /* ref_ctr_off */);
10682         } else {
10683                 char probe_name[256];
10684
10685                 gen_kprobe_legacy_event_name(probe_name, sizeof(probe_name),
10686                                              func_name, offset);
10687
10688                 legacy_probe = strdup(probe_name);
10689                 if (!legacy_probe)
10690                         return libbpf_err_ptr(-ENOMEM);
10691
10692                 pfd = perf_event_kprobe_open_legacy(legacy_probe, retprobe, func_name,
10693                                                     offset, -1 /* pid */);
10694         }
10695         if (pfd < 0) {
10696                 err = -errno;
10697                 pr_warn("prog '%s': failed to create %s '%s+0x%zx' perf event: %s\n",
10698                         prog->name, retprobe ? "kretprobe" : "kprobe",
10699                         func_name, offset,
10700                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10701                 goto err_out;
10702         }
10703         link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
10704         err = libbpf_get_error(link);
10705         if (err) {
10706                 close(pfd);
10707                 pr_warn("prog '%s': failed to attach to %s '%s+0x%zx': %s\n",
10708                         prog->name, retprobe ? "kretprobe" : "kprobe",
10709                         func_name, offset,
10710                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
10711                 goto err_clean_legacy;
10712         }
10713         if (legacy) {
10714                 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
10715
10716                 perf_link->legacy_probe_name = legacy_probe;
10717                 perf_link->legacy_is_kprobe = true;
10718                 perf_link->legacy_is_retprobe = retprobe;
10719         }
10720
10721         return link;
10722
10723 err_clean_legacy:
10724         if (legacy)
10725                 remove_kprobe_event_legacy(legacy_probe, retprobe);
10726 err_out:
10727         free(legacy_probe);
10728         return libbpf_err_ptr(err);
10729 }
10730
10731 struct bpf_link *bpf_program__attach_kprobe(const struct bpf_program *prog,
10732                                             bool retprobe,
10733                                             const char *func_name)
10734 {
10735         DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts,
10736                 .retprobe = retprobe,
10737         );
10738
10739         return bpf_program__attach_kprobe_opts(prog, func_name, &opts);
10740 }
10741
10742 struct bpf_link *bpf_program__attach_ksyscall(const struct bpf_program *prog,
10743                                               const char *syscall_name,
10744                                               const struct bpf_ksyscall_opts *opts)
10745 {
10746         LIBBPF_OPTS(bpf_kprobe_opts, kprobe_opts);
10747         char func_name[128];
10748
10749         if (!OPTS_VALID(opts, bpf_ksyscall_opts))
10750                 return libbpf_err_ptr(-EINVAL);
10751
10752         if (kernel_supports(prog->obj, FEAT_SYSCALL_WRAPPER)) {
10753                 /* arch_specific_syscall_pfx() should never return NULL here
10754                  * because it is guarded by kernel_supports(). However, since
10755                  * compiler does not know that we have an explicit conditional
10756                  * as well.
10757                  */
10758                 snprintf(func_name, sizeof(func_name), "__%s_sys_%s",
10759                          arch_specific_syscall_pfx() ? : "", syscall_name);
10760         } else {
10761                 snprintf(func_name, sizeof(func_name), "__se_sys_%s", syscall_name);
10762         }
10763
10764         kprobe_opts.retprobe = OPTS_GET(opts, retprobe, false);
10765         kprobe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
10766
10767         return bpf_program__attach_kprobe_opts(prog, func_name, &kprobe_opts);
10768 }
10769
10770 /* Adapted from perf/util/string.c */
10771 bool glob_match(const char *str, const char *pat)
10772 {
10773         while (*str && *pat && *pat != '*') {
10774                 if (*pat == '?') {      /* Matches any single character */
10775                         str++;
10776                         pat++;
10777                         continue;
10778                 }
10779                 if (*str != *pat)
10780                         return false;
10781                 str++;
10782                 pat++;
10783         }
10784         /* Check wild card */
10785         if (*pat == '*') {
10786                 while (*pat == '*')
10787                         pat++;
10788                 if (!*pat) /* Tail wild card matches all */
10789                         return true;
10790                 while (*str)
10791                         if (glob_match(str++, pat))
10792                                 return true;
10793         }
10794         return !*str && !*pat;
10795 }
10796
10797 struct kprobe_multi_resolve {
10798         const char *pattern;
10799         unsigned long *addrs;
10800         size_t cap;
10801         size_t cnt;
10802 };
10803
10804 struct avail_kallsyms_data {
10805         char **syms;
10806         size_t cnt;
10807         struct kprobe_multi_resolve *res;
10808 };
10809
10810 static int avail_func_cmp(const void *a, const void *b)
10811 {
10812         return strcmp(*(const char **)a, *(const char **)b);
10813 }
10814
10815 static int avail_kallsyms_cb(unsigned long long sym_addr, char sym_type,
10816                              const char *sym_name, void *ctx)
10817 {
10818         struct avail_kallsyms_data *data = ctx;
10819         struct kprobe_multi_resolve *res = data->res;
10820         int err;
10821
10822         if (!bsearch(&sym_name, data->syms, data->cnt, sizeof(*data->syms), avail_func_cmp))
10823                 return 0;
10824
10825         err = libbpf_ensure_mem((void **)&res->addrs, &res->cap, sizeof(*res->addrs), res->cnt + 1);
10826         if (err)
10827                 return err;
10828
10829         res->addrs[res->cnt++] = (unsigned long)sym_addr;
10830         return 0;
10831 }
10832
10833 static int libbpf_available_kallsyms_parse(struct kprobe_multi_resolve *res)
10834 {
10835         const char *available_functions_file = tracefs_available_filter_functions();
10836         struct avail_kallsyms_data data;
10837         char sym_name[500];
10838         FILE *f;
10839         int err = 0, ret, i;
10840         char **syms = NULL;
10841         size_t cap = 0, cnt = 0;
10842
10843         f = fopen(available_functions_file, "re");
10844         if (!f) {
10845                 err = -errno;
10846                 pr_warn("failed to open %s: %d\n", available_functions_file, err);
10847                 return err;
10848         }
10849
10850         while (true) {
10851                 char *name;
10852
10853                 ret = fscanf(f, "%499s%*[^\n]\n", sym_name);
10854                 if (ret == EOF && feof(f))
10855                         break;
10856
10857                 if (ret != 1) {
10858                         pr_warn("failed to parse available_filter_functions entry: %d\n", ret);
10859                         err = -EINVAL;
10860                         goto cleanup;
10861                 }
10862
10863                 if (!glob_match(sym_name, res->pattern))
10864                         continue;
10865
10866                 err = libbpf_ensure_mem((void **)&syms, &cap, sizeof(*syms), cnt + 1);
10867                 if (err)
10868                         goto cleanup;
10869
10870                 name = strdup(sym_name);
10871                 if (!name) {
10872                         err = -errno;
10873                         goto cleanup;
10874                 }
10875
10876                 syms[cnt++] = name;
10877         }
10878
10879         /* no entries found, bail out */
10880         if (cnt == 0) {
10881                 err = -ENOENT;
10882                 goto cleanup;
10883         }
10884
10885         /* sort available functions */
10886         qsort(syms, cnt, sizeof(*syms), avail_func_cmp);
10887
10888         data.syms = syms;
10889         data.res = res;
10890         data.cnt = cnt;
10891         libbpf_kallsyms_parse(avail_kallsyms_cb, &data);
10892
10893         if (res->cnt == 0)
10894                 err = -ENOENT;
10895
10896 cleanup:
10897         for (i = 0; i < cnt; i++)
10898                 free((char *)syms[i]);
10899         free(syms);
10900
10901         fclose(f);
10902         return err;
10903 }
10904
10905 static bool has_available_filter_functions_addrs(void)
10906 {
10907         return access(tracefs_available_filter_functions_addrs(), R_OK) != -1;
10908 }
10909
10910 static int libbpf_available_kprobes_parse(struct kprobe_multi_resolve *res)
10911 {
10912         const char *available_path = tracefs_available_filter_functions_addrs();
10913         char sym_name[500];
10914         FILE *f;
10915         int ret, err = 0;
10916         unsigned long long sym_addr;
10917
10918         f = fopen(available_path, "re");
10919         if (!f) {
10920                 err = -errno;
10921                 pr_warn("failed to open %s: %d\n", available_path, err);
10922                 return err;
10923         }
10924
10925         while (true) {
10926                 ret = fscanf(f, "%llx %499s%*[^\n]\n", &sym_addr, sym_name);
10927                 if (ret == EOF && feof(f))
10928                         break;
10929
10930                 if (ret != 2) {
10931                         pr_warn("failed to parse available_filter_functions_addrs entry: %d\n",
10932                                 ret);
10933                         err = -EINVAL;
10934                         goto cleanup;
10935                 }
10936
10937                 if (!glob_match(sym_name, res->pattern))
10938                         continue;
10939
10940                 err = libbpf_ensure_mem((void **)&res->addrs, &res->cap,
10941                                         sizeof(*res->addrs), res->cnt + 1);
10942                 if (err)
10943                         goto cleanup;
10944
10945                 res->addrs[res->cnt++] = (unsigned long)sym_addr;
10946         }
10947
10948         if (res->cnt == 0)
10949                 err = -ENOENT;
10950
10951 cleanup:
10952         fclose(f);
10953         return err;
10954 }
10955
10956 struct bpf_link *
10957 bpf_program__attach_kprobe_multi_opts(const struct bpf_program *prog,
10958                                       const char *pattern,
10959                                       const struct bpf_kprobe_multi_opts *opts)
10960 {
10961         LIBBPF_OPTS(bpf_link_create_opts, lopts);
10962         struct kprobe_multi_resolve res = {
10963                 .pattern = pattern,
10964         };
10965         struct bpf_link *link = NULL;
10966         char errmsg[STRERR_BUFSIZE];
10967         const unsigned long *addrs;
10968         int err, link_fd, prog_fd;
10969         const __u64 *cookies;
10970         const char **syms;
10971         bool retprobe;
10972         size_t cnt;
10973
10974         if (!OPTS_VALID(opts, bpf_kprobe_multi_opts))
10975                 return libbpf_err_ptr(-EINVAL);
10976
10977         syms    = OPTS_GET(opts, syms, false);
10978         addrs   = OPTS_GET(opts, addrs, false);
10979         cnt     = OPTS_GET(opts, cnt, false);
10980         cookies = OPTS_GET(opts, cookies, false);
10981
10982         if (!pattern && !addrs && !syms)
10983                 return libbpf_err_ptr(-EINVAL);
10984         if (pattern && (addrs || syms || cookies || cnt))
10985                 return libbpf_err_ptr(-EINVAL);
10986         if (!pattern && !cnt)
10987                 return libbpf_err_ptr(-EINVAL);
10988         if (addrs && syms)
10989                 return libbpf_err_ptr(-EINVAL);
10990
10991         if (pattern) {
10992                 if (has_available_filter_functions_addrs())
10993                         err = libbpf_available_kprobes_parse(&res);
10994                 else
10995                         err = libbpf_available_kallsyms_parse(&res);
10996                 if (err)
10997                         goto error;
10998                 addrs = res.addrs;
10999                 cnt = res.cnt;
11000         }
11001
11002         retprobe = OPTS_GET(opts, retprobe, false);
11003
11004         lopts.kprobe_multi.syms = syms;
11005         lopts.kprobe_multi.addrs = addrs;
11006         lopts.kprobe_multi.cookies = cookies;
11007         lopts.kprobe_multi.cnt = cnt;
11008         lopts.kprobe_multi.flags = retprobe ? BPF_F_KPROBE_MULTI_RETURN : 0;
11009
11010         link = calloc(1, sizeof(*link));
11011         if (!link) {
11012                 err = -ENOMEM;
11013                 goto error;
11014         }
11015         link->detach = &bpf_link__detach_fd;
11016
11017         prog_fd = bpf_program__fd(prog);
11018         link_fd = bpf_link_create(prog_fd, 0, BPF_TRACE_KPROBE_MULTI, &lopts);
11019         if (link_fd < 0) {
11020                 err = -errno;
11021                 pr_warn("prog '%s': failed to attach: %s\n",
11022                         prog->name, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11023                 goto error;
11024         }
11025         link->fd = link_fd;
11026         free(res.addrs);
11027         return link;
11028
11029 error:
11030         free(link);
11031         free(res.addrs);
11032         return libbpf_err_ptr(err);
11033 }
11034
11035 static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11036 {
11037         DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts);
11038         unsigned long offset = 0;
11039         const char *func_name;
11040         char *func;
11041         int n;
11042
11043         *link = NULL;
11044
11045         /* no auto-attach for SEC("kprobe") and SEC("kretprobe") */
11046         if (strcmp(prog->sec_name, "kprobe") == 0 || strcmp(prog->sec_name, "kretprobe") == 0)
11047                 return 0;
11048
11049         opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe/");
11050         if (opts.retprobe)
11051                 func_name = prog->sec_name + sizeof("kretprobe/") - 1;
11052         else
11053                 func_name = prog->sec_name + sizeof("kprobe/") - 1;
11054
11055         n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset);
11056         if (n < 1) {
11057                 pr_warn("kprobe name is invalid: %s\n", func_name);
11058                 return -EINVAL;
11059         }
11060         if (opts.retprobe && offset != 0) {
11061                 free(func);
11062                 pr_warn("kretprobes do not support offset specification\n");
11063                 return -EINVAL;
11064         }
11065
11066         opts.offset = offset;
11067         *link = bpf_program__attach_kprobe_opts(prog, func, &opts);
11068         free(func);
11069         return libbpf_get_error(*link);
11070 }
11071
11072 static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11073 {
11074         LIBBPF_OPTS(bpf_ksyscall_opts, opts);
11075         const char *syscall_name;
11076
11077         *link = NULL;
11078
11079         /* no auto-attach for SEC("ksyscall") and SEC("kretsyscall") */
11080         if (strcmp(prog->sec_name, "ksyscall") == 0 || strcmp(prog->sec_name, "kretsyscall") == 0)
11081                 return 0;
11082
11083         opts.retprobe = str_has_pfx(prog->sec_name, "kretsyscall/");
11084         if (opts.retprobe)
11085                 syscall_name = prog->sec_name + sizeof("kretsyscall/") - 1;
11086         else
11087                 syscall_name = prog->sec_name + sizeof("ksyscall/") - 1;
11088
11089         *link = bpf_program__attach_ksyscall(prog, syscall_name, &opts);
11090         return *link ? 0 : -errno;
11091 }
11092
11093 static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11094 {
11095         LIBBPF_OPTS(bpf_kprobe_multi_opts, opts);
11096         const char *spec;
11097         char *pattern;
11098         int n;
11099
11100         *link = NULL;
11101
11102         /* no auto-attach for SEC("kprobe.multi") and SEC("kretprobe.multi") */
11103         if (strcmp(prog->sec_name, "kprobe.multi") == 0 ||
11104             strcmp(prog->sec_name, "kretprobe.multi") == 0)
11105                 return 0;
11106
11107         opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe.multi/");
11108         if (opts.retprobe)
11109                 spec = prog->sec_name + sizeof("kretprobe.multi/") - 1;
11110         else
11111                 spec = prog->sec_name + sizeof("kprobe.multi/") - 1;
11112
11113         n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern);
11114         if (n < 1) {
11115                 pr_warn("kprobe multi pattern is invalid: %s\n", pattern);
11116                 return -EINVAL;
11117         }
11118
11119         *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts);
11120         free(pattern);
11121         return libbpf_get_error(*link);
11122 }
11123
11124 static int attach_uprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11125 {
11126         char *probe_type = NULL, *binary_path = NULL, *func_name = NULL;
11127         LIBBPF_OPTS(bpf_uprobe_multi_opts, opts);
11128         int n, ret = -EINVAL;
11129
11130         *link = NULL;
11131
11132         n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[^\n]",
11133                    &probe_type, &binary_path, &func_name);
11134         switch (n) {
11135         case 1:
11136                 /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */
11137                 ret = 0;
11138                 break;
11139         case 3:
11140                 opts.retprobe = strcmp(probe_type, "uretprobe.multi") == 0;
11141                 *link = bpf_program__attach_uprobe_multi(prog, -1, binary_path, func_name, &opts);
11142                 ret = libbpf_get_error(*link);
11143                 break;
11144         default:
11145                 pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name,
11146                         prog->sec_name);
11147                 break;
11148         }
11149         free(probe_type);
11150         free(binary_path);
11151         free(func_name);
11152         return ret;
11153 }
11154
11155 static void gen_uprobe_legacy_event_name(char *buf, size_t buf_sz,
11156                                          const char *binary_path, uint64_t offset)
11157 {
11158         int i;
11159
11160         snprintf(buf, buf_sz, "libbpf_%u_%s_0x%zx", getpid(), binary_path, (size_t)offset);
11161
11162         /* sanitize binary_path in the probe name */
11163         for (i = 0; buf[i]; i++) {
11164                 if (!isalnum(buf[i]))
11165                         buf[i] = '_';
11166         }
11167 }
11168
11169 static inline int add_uprobe_event_legacy(const char *probe_name, bool retprobe,
11170                                           const char *binary_path, size_t offset)
11171 {
11172         return append_to_file(tracefs_uprobe_events(), "%c:%s/%s %s:0x%zx",
11173                               retprobe ? 'r' : 'p',
11174                               retprobe ? "uretprobes" : "uprobes",
11175                               probe_name, binary_path, offset);
11176 }
11177
11178 static inline int remove_uprobe_event_legacy(const char *probe_name, bool retprobe)
11179 {
11180         return append_to_file(tracefs_uprobe_events(), "-:%s/%s",
11181                               retprobe ? "uretprobes" : "uprobes", probe_name);
11182 }
11183
11184 static int determine_uprobe_perf_type_legacy(const char *probe_name, bool retprobe)
11185 {
11186         char file[512];
11187
11188         snprintf(file, sizeof(file), "%s/events/%s/%s/id",
11189                  tracefs_path(), retprobe ? "uretprobes" : "uprobes", probe_name);
11190
11191         return parse_uint_from_file(file, "%d\n");
11192 }
11193
11194 static int perf_event_uprobe_open_legacy(const char *probe_name, bool retprobe,
11195                                          const char *binary_path, size_t offset, int pid)
11196 {
11197         const size_t attr_sz = sizeof(struct perf_event_attr);
11198         struct perf_event_attr attr;
11199         int type, pfd, err;
11200
11201         err = add_uprobe_event_legacy(probe_name, retprobe, binary_path, offset);
11202         if (err < 0) {
11203                 pr_warn("failed to add legacy uprobe event for %s:0x%zx: %d\n",
11204                         binary_path, (size_t)offset, err);
11205                 return err;
11206         }
11207         type = determine_uprobe_perf_type_legacy(probe_name, retprobe);
11208         if (type < 0) {
11209                 err = type;
11210                 pr_warn("failed to determine legacy uprobe event id for %s:0x%zx: %d\n",
11211                         binary_path, offset, err);
11212                 goto err_clean_legacy;
11213         }
11214
11215         memset(&attr, 0, attr_sz);
11216         attr.size = attr_sz;
11217         attr.config = type;
11218         attr.type = PERF_TYPE_TRACEPOINT;
11219
11220         pfd = syscall(__NR_perf_event_open, &attr,
11221                       pid < 0 ? -1 : pid, /* pid */
11222                       pid == -1 ? 0 : -1, /* cpu */
11223                       -1 /* group_fd */,  PERF_FLAG_FD_CLOEXEC);
11224         if (pfd < 0) {
11225                 err = -errno;
11226                 pr_warn("legacy uprobe perf_event_open() failed: %d\n", err);
11227                 goto err_clean_legacy;
11228         }
11229         return pfd;
11230
11231 err_clean_legacy:
11232         /* Clear the newly added legacy uprobe_event */
11233         remove_uprobe_event_legacy(probe_name, retprobe);
11234         return err;
11235 }
11236
11237 /* Find offset of function name in archive specified by path. Currently
11238  * supported are .zip files that do not compress their contents, as used on
11239  * Android in the form of APKs, for example. "file_name" is the name of the ELF
11240  * file inside the archive. "func_name" matches symbol name or name@@LIB for
11241  * library functions.
11242  *
11243  * An overview of the APK format specifically provided here:
11244  * https://en.wikipedia.org/w/index.php?title=Apk_(file_format)&oldid=1139099120#Package_contents
11245  */
11246 static long elf_find_func_offset_from_archive(const char *archive_path, const char *file_name,
11247                                               const char *func_name)
11248 {
11249         struct zip_archive *archive;
11250         struct zip_entry entry;
11251         long ret;
11252         Elf *elf;
11253
11254         archive = zip_archive_open(archive_path);
11255         if (IS_ERR(archive)) {
11256                 ret = PTR_ERR(archive);
11257                 pr_warn("zip: failed to open %s: %ld\n", archive_path, ret);
11258                 return ret;
11259         }
11260
11261         ret = zip_archive_find_entry(archive, file_name, &entry);
11262         if (ret) {
11263                 pr_warn("zip: could not find archive member %s in %s: %ld\n", file_name,
11264                         archive_path, ret);
11265                 goto out;
11266         }
11267         pr_debug("zip: found entry for %s in %s at 0x%lx\n", file_name, archive_path,
11268                  (unsigned long)entry.data_offset);
11269
11270         if (entry.compression) {
11271                 pr_warn("zip: entry %s of %s is compressed and cannot be handled\n", file_name,
11272                         archive_path);
11273                 ret = -LIBBPF_ERRNO__FORMAT;
11274                 goto out;
11275         }
11276
11277         elf = elf_memory((void *)entry.data, entry.data_length);
11278         if (!elf) {
11279                 pr_warn("elf: could not read elf file %s from %s: %s\n", file_name, archive_path,
11280                         elf_errmsg(-1));
11281                 ret = -LIBBPF_ERRNO__LIBELF;
11282                 goto out;
11283         }
11284
11285         ret = elf_find_func_offset(elf, file_name, func_name);
11286         if (ret > 0) {
11287                 pr_debug("elf: symbol address match for %s of %s in %s: 0x%x + 0x%lx = 0x%lx\n",
11288                          func_name, file_name, archive_path, entry.data_offset, ret,
11289                          ret + entry.data_offset);
11290                 ret += entry.data_offset;
11291         }
11292         elf_end(elf);
11293
11294 out:
11295         zip_archive_close(archive);
11296         return ret;
11297 }
11298
11299 static const char *arch_specific_lib_paths(void)
11300 {
11301         /*
11302          * Based on https://packages.debian.org/sid/libc6.
11303          *
11304          * Assume that the traced program is built for the same architecture
11305          * as libbpf, which should cover the vast majority of cases.
11306          */
11307 #if defined(__x86_64__)
11308         return "/lib/x86_64-linux-gnu";
11309 #elif defined(__i386__)
11310         return "/lib/i386-linux-gnu";
11311 #elif defined(__s390x__)
11312         return "/lib/s390x-linux-gnu";
11313 #elif defined(__s390__)
11314         return "/lib/s390-linux-gnu";
11315 #elif defined(__arm__) && defined(__SOFTFP__)
11316         return "/lib/arm-linux-gnueabi";
11317 #elif defined(__arm__) && !defined(__SOFTFP__)
11318         return "/lib/arm-linux-gnueabihf";
11319 #elif defined(__aarch64__)
11320         return "/lib/aarch64-linux-gnu";
11321 #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 64
11322         return "/lib/mips64el-linux-gnuabi64";
11323 #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 32
11324         return "/lib/mipsel-linux-gnu";
11325 #elif defined(__powerpc64__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
11326         return "/lib/powerpc64le-linux-gnu";
11327 #elif defined(__sparc__) && defined(__arch64__)
11328         return "/lib/sparc64-linux-gnu";
11329 #elif defined(__riscv) && __riscv_xlen == 64
11330         return "/lib/riscv64-linux-gnu";
11331 #else
11332         return NULL;
11333 #endif
11334 }
11335
11336 /* Get full path to program/shared library. */
11337 static int resolve_full_path(const char *file, char *result, size_t result_sz)
11338 {
11339         const char *search_paths[3] = {};
11340         int i, perm;
11341
11342         if (str_has_sfx(file, ".so") || strstr(file, ".so.")) {
11343                 search_paths[0] = getenv("LD_LIBRARY_PATH");
11344                 search_paths[1] = "/usr/lib64:/usr/lib";
11345                 search_paths[2] = arch_specific_lib_paths();
11346                 perm = R_OK;
11347         } else {
11348                 search_paths[0] = getenv("PATH");
11349                 search_paths[1] = "/usr/bin:/usr/sbin";
11350                 perm = R_OK | X_OK;
11351         }
11352
11353         for (i = 0; i < ARRAY_SIZE(search_paths); i++) {
11354                 const char *s;
11355
11356                 if (!search_paths[i])
11357                         continue;
11358                 for (s = search_paths[i]; s != NULL; s = strchr(s, ':')) {
11359                         char *next_path;
11360                         int seg_len;
11361
11362                         if (s[0] == ':')
11363                                 s++;
11364                         next_path = strchr(s, ':');
11365                         seg_len = next_path ? next_path - s : strlen(s);
11366                         if (!seg_len)
11367                                 continue;
11368                         snprintf(result, result_sz, "%.*s/%s", seg_len, s, file);
11369                         /* ensure it has required permissions */
11370                         if (faccessat(AT_FDCWD, result, perm, AT_EACCESS) < 0)
11371                                 continue;
11372                         pr_debug("resolved '%s' to '%s'\n", file, result);
11373                         return 0;
11374                 }
11375         }
11376         return -ENOENT;
11377 }
11378
11379 struct bpf_link *
11380 bpf_program__attach_uprobe_multi(const struct bpf_program *prog,
11381                                  pid_t pid,
11382                                  const char *path,
11383                                  const char *func_pattern,
11384                                  const struct bpf_uprobe_multi_opts *opts)
11385 {
11386         const unsigned long *ref_ctr_offsets = NULL, *offsets = NULL;
11387         LIBBPF_OPTS(bpf_link_create_opts, lopts);
11388         unsigned long *resolved_offsets = NULL;
11389         int err = 0, link_fd, prog_fd;
11390         struct bpf_link *link = NULL;
11391         char errmsg[STRERR_BUFSIZE];
11392         char full_path[PATH_MAX];
11393         const __u64 *cookies;
11394         const char **syms;
11395         size_t cnt;
11396
11397         if (!OPTS_VALID(opts, bpf_uprobe_multi_opts))
11398                 return libbpf_err_ptr(-EINVAL);
11399
11400         syms = OPTS_GET(opts, syms, NULL);
11401         offsets = OPTS_GET(opts, offsets, NULL);
11402         ref_ctr_offsets = OPTS_GET(opts, ref_ctr_offsets, NULL);
11403         cookies = OPTS_GET(opts, cookies, NULL);
11404         cnt = OPTS_GET(opts, cnt, 0);
11405
11406         /*
11407          * User can specify 2 mutually exclusive set of inputs:
11408          *
11409          * 1) use only path/func_pattern/pid arguments
11410          *
11411          * 2) use path/pid with allowed combinations of:
11412          *    syms/offsets/ref_ctr_offsets/cookies/cnt
11413          *
11414          *    - syms and offsets are mutually exclusive
11415          *    - ref_ctr_offsets and cookies are optional
11416          *
11417          * Any other usage results in error.
11418          */
11419
11420         if (!path)
11421                 return libbpf_err_ptr(-EINVAL);
11422         if (!func_pattern && cnt == 0)
11423                 return libbpf_err_ptr(-EINVAL);
11424
11425         if (func_pattern) {
11426                 if (syms || offsets || ref_ctr_offsets || cookies || cnt)
11427                         return libbpf_err_ptr(-EINVAL);
11428         } else {
11429                 if (!!syms == !!offsets)
11430                         return libbpf_err_ptr(-EINVAL);
11431         }
11432
11433         if (func_pattern) {
11434                 if (!strchr(path, '/')) {
11435                         err = resolve_full_path(path, full_path, sizeof(full_path));
11436                         if (err) {
11437                                 pr_warn("prog '%s': failed to resolve full path for '%s': %d\n",
11438                                         prog->name, path, err);
11439                                 return libbpf_err_ptr(err);
11440                         }
11441                         path = full_path;
11442                 }
11443
11444                 err = elf_resolve_pattern_offsets(path, func_pattern,
11445                                                   &resolved_offsets, &cnt);
11446                 if (err < 0)
11447                         return libbpf_err_ptr(err);
11448                 offsets = resolved_offsets;
11449         } else if (syms) {
11450                 err = elf_resolve_syms_offsets(path, cnt, syms, &resolved_offsets);
11451                 if (err < 0)
11452                         return libbpf_err_ptr(err);
11453                 offsets = resolved_offsets;
11454         }
11455
11456         lopts.uprobe_multi.path = path;
11457         lopts.uprobe_multi.offsets = offsets;
11458         lopts.uprobe_multi.ref_ctr_offsets = ref_ctr_offsets;
11459         lopts.uprobe_multi.cookies = cookies;
11460         lopts.uprobe_multi.cnt = cnt;
11461         lopts.uprobe_multi.flags = OPTS_GET(opts, retprobe, false) ? BPF_F_UPROBE_MULTI_RETURN : 0;
11462
11463         if (pid == 0)
11464                 pid = getpid();
11465         if (pid > 0)
11466                 lopts.uprobe_multi.pid = pid;
11467
11468         link = calloc(1, sizeof(*link));
11469         if (!link) {
11470                 err = -ENOMEM;
11471                 goto error;
11472         }
11473         link->detach = &bpf_link__detach_fd;
11474
11475         prog_fd = bpf_program__fd(prog);
11476         link_fd = bpf_link_create(prog_fd, 0, BPF_TRACE_UPROBE_MULTI, &lopts);
11477         if (link_fd < 0) {
11478                 err = -errno;
11479                 pr_warn("prog '%s': failed to attach multi-uprobe: %s\n",
11480                         prog->name, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11481                 goto error;
11482         }
11483         link->fd = link_fd;
11484         free(resolved_offsets);
11485         return link;
11486
11487 error:
11488         free(resolved_offsets);
11489         free(link);
11490         return libbpf_err_ptr(err);
11491 }
11492
11493 LIBBPF_API struct bpf_link *
11494 bpf_program__attach_uprobe_opts(const struct bpf_program *prog, pid_t pid,
11495                                 const char *binary_path, size_t func_offset,
11496                                 const struct bpf_uprobe_opts *opts)
11497 {
11498         const char *archive_path = NULL, *archive_sep = NULL;
11499         char errmsg[STRERR_BUFSIZE], *legacy_probe = NULL;
11500         DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
11501         enum probe_attach_mode attach_mode;
11502         char full_path[PATH_MAX];
11503         struct bpf_link *link;
11504         size_t ref_ctr_off;
11505         int pfd, err;
11506         bool retprobe, legacy;
11507         const char *func_name;
11508
11509         if (!OPTS_VALID(opts, bpf_uprobe_opts))
11510                 return libbpf_err_ptr(-EINVAL);
11511
11512         attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT);
11513         retprobe = OPTS_GET(opts, retprobe, false);
11514         ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0);
11515         pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
11516
11517         if (!binary_path)
11518                 return libbpf_err_ptr(-EINVAL);
11519
11520         /* Check if "binary_path" refers to an archive. */
11521         archive_sep = strstr(binary_path, "!/");
11522         if (archive_sep) {
11523                 full_path[0] = '\0';
11524                 libbpf_strlcpy(full_path, binary_path,
11525                                min(sizeof(full_path), (size_t)(archive_sep - binary_path + 1)));
11526                 archive_path = full_path;
11527                 binary_path = archive_sep + 2;
11528         } else if (!strchr(binary_path, '/')) {
11529                 err = resolve_full_path(binary_path, full_path, sizeof(full_path));
11530                 if (err) {
11531                         pr_warn("prog '%s': failed to resolve full path for '%s': %d\n",
11532                                 prog->name, binary_path, err);
11533                         return libbpf_err_ptr(err);
11534                 }
11535                 binary_path = full_path;
11536         }
11537         func_name = OPTS_GET(opts, func_name, NULL);
11538         if (func_name) {
11539                 long sym_off;
11540
11541                 if (archive_path) {
11542                         sym_off = elf_find_func_offset_from_archive(archive_path, binary_path,
11543                                                                     func_name);
11544                         binary_path = archive_path;
11545                 } else {
11546                         sym_off = elf_find_func_offset_from_file(binary_path, func_name);
11547                 }
11548                 if (sym_off < 0)
11549                         return libbpf_err_ptr(sym_off);
11550                 func_offset += sym_off;
11551         }
11552
11553         legacy = determine_uprobe_perf_type() < 0;
11554         switch (attach_mode) {
11555         case PROBE_ATTACH_MODE_LEGACY:
11556                 legacy = true;
11557                 pe_opts.force_ioctl_attach = true;
11558                 break;
11559         case PROBE_ATTACH_MODE_PERF:
11560                 if (legacy)
11561                         return libbpf_err_ptr(-ENOTSUP);
11562                 pe_opts.force_ioctl_attach = true;
11563                 break;
11564         case PROBE_ATTACH_MODE_LINK:
11565                 if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK))
11566                         return libbpf_err_ptr(-ENOTSUP);
11567                 break;
11568         case PROBE_ATTACH_MODE_DEFAULT:
11569                 break;
11570         default:
11571                 return libbpf_err_ptr(-EINVAL);
11572         }
11573
11574         if (!legacy) {
11575                 pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path,
11576                                             func_offset, pid, ref_ctr_off);
11577         } else {
11578                 char probe_name[PATH_MAX + 64];
11579
11580                 if (ref_ctr_off)
11581                         return libbpf_err_ptr(-EINVAL);
11582
11583                 gen_uprobe_legacy_event_name(probe_name, sizeof(probe_name),
11584                                              binary_path, func_offset);
11585
11586                 legacy_probe = strdup(probe_name);
11587                 if (!legacy_probe)
11588                         return libbpf_err_ptr(-ENOMEM);
11589
11590                 pfd = perf_event_uprobe_open_legacy(legacy_probe, retprobe,
11591                                                     binary_path, func_offset, pid);
11592         }
11593         if (pfd < 0) {
11594                 err = -errno;
11595                 pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
11596                         prog->name, retprobe ? "uretprobe" : "uprobe",
11597                         binary_path, func_offset,
11598                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11599                 goto err_out;
11600         }
11601
11602         link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
11603         err = libbpf_get_error(link);
11604         if (err) {
11605                 close(pfd);
11606                 pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n",
11607                         prog->name, retprobe ? "uretprobe" : "uprobe",
11608                         binary_path, func_offset,
11609                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11610                 goto err_clean_legacy;
11611         }
11612         if (legacy) {
11613                 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
11614
11615                 perf_link->legacy_probe_name = legacy_probe;
11616                 perf_link->legacy_is_kprobe = false;
11617                 perf_link->legacy_is_retprobe = retprobe;
11618         }
11619         return link;
11620
11621 err_clean_legacy:
11622         if (legacy)
11623                 remove_uprobe_event_legacy(legacy_probe, retprobe);
11624 err_out:
11625         free(legacy_probe);
11626         return libbpf_err_ptr(err);
11627 }
11628
11629 /* Format of u[ret]probe section definition supporting auto-attach:
11630  * u[ret]probe/binary:function[+offset]
11631  *
11632  * binary can be an absolute/relative path or a filename; the latter is resolved to a
11633  * full binary path via bpf_program__attach_uprobe_opts.
11634  *
11635  * Specifying uprobe+ ensures we carry out strict matching; either "uprobe" must be
11636  * specified (and auto-attach is not possible) or the above format is specified for
11637  * auto-attach.
11638  */
11639 static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11640 {
11641         DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts);
11642         char *probe_type = NULL, *binary_path = NULL, *func_name = NULL, *func_off;
11643         int n, c, ret = -EINVAL;
11644         long offset = 0;
11645
11646         *link = NULL;
11647
11648         n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[^\n]",
11649                    &probe_type, &binary_path, &func_name);
11650         switch (n) {
11651         case 1:
11652                 /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */
11653                 ret = 0;
11654                 break;
11655         case 2:
11656                 pr_warn("prog '%s': section '%s' missing ':function[+offset]' specification\n",
11657                         prog->name, prog->sec_name);
11658                 break;
11659         case 3:
11660                 /* check if user specifies `+offset`, if yes, this should be
11661                  * the last part of the string, make sure sscanf read to EOL
11662                  */
11663                 func_off = strrchr(func_name, '+');
11664                 if (func_off) {
11665                         n = sscanf(func_off, "+%li%n", &offset, &c);
11666                         if (n == 1 && *(func_off + c) == '\0')
11667                                 func_off[0] = '\0';
11668                         else
11669                                 offset = 0;
11670                 }
11671                 opts.retprobe = strcmp(probe_type, "uretprobe") == 0 ||
11672                                 strcmp(probe_type, "uretprobe.s") == 0;
11673                 if (opts.retprobe && offset != 0) {
11674                         pr_warn("prog '%s': uretprobes do not support offset specification\n",
11675                                 prog->name);
11676                         break;
11677                 }
11678                 opts.func_name = func_name;
11679                 *link = bpf_program__attach_uprobe_opts(prog, -1, binary_path, offset, &opts);
11680                 ret = libbpf_get_error(*link);
11681                 break;
11682         default:
11683                 pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name,
11684                         prog->sec_name);
11685                 break;
11686         }
11687         free(probe_type);
11688         free(binary_path);
11689         free(func_name);
11690
11691         return ret;
11692 }
11693
11694 struct bpf_link *bpf_program__attach_uprobe(const struct bpf_program *prog,
11695                                             bool retprobe, pid_t pid,
11696                                             const char *binary_path,
11697                                             size_t func_offset)
11698 {
11699         DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe);
11700
11701         return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts);
11702 }
11703
11704 struct bpf_link *bpf_program__attach_usdt(const struct bpf_program *prog,
11705                                           pid_t pid, const char *binary_path,
11706                                           const char *usdt_provider, const char *usdt_name,
11707                                           const struct bpf_usdt_opts *opts)
11708 {
11709         char resolved_path[512];
11710         struct bpf_object *obj = prog->obj;
11711         struct bpf_link *link;
11712         __u64 usdt_cookie;
11713         int err;
11714
11715         if (!OPTS_VALID(opts, bpf_uprobe_opts))
11716                 return libbpf_err_ptr(-EINVAL);
11717
11718         if (bpf_program__fd(prog) < 0) {
11719                 pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n",
11720                         prog->name);
11721                 return libbpf_err_ptr(-EINVAL);
11722         }
11723
11724         if (!binary_path)
11725                 return libbpf_err_ptr(-EINVAL);
11726
11727         if (!strchr(binary_path, '/')) {
11728                 err = resolve_full_path(binary_path, resolved_path, sizeof(resolved_path));
11729                 if (err) {
11730                         pr_warn("prog '%s': failed to resolve full path for '%s': %d\n",
11731                                 prog->name, binary_path, err);
11732                         return libbpf_err_ptr(err);
11733                 }
11734                 binary_path = resolved_path;
11735         }
11736
11737         /* USDT manager is instantiated lazily on first USDT attach. It will
11738          * be destroyed together with BPF object in bpf_object__close().
11739          */
11740         if (IS_ERR(obj->usdt_man))
11741                 return libbpf_ptr(obj->usdt_man);
11742         if (!obj->usdt_man) {
11743                 obj->usdt_man = usdt_manager_new(obj);
11744                 if (IS_ERR(obj->usdt_man))
11745                         return libbpf_ptr(obj->usdt_man);
11746         }
11747
11748         usdt_cookie = OPTS_GET(opts, usdt_cookie, 0);
11749         link = usdt_manager_attach_usdt(obj->usdt_man, prog, pid, binary_path,
11750                                         usdt_provider, usdt_name, usdt_cookie);
11751         err = libbpf_get_error(link);
11752         if (err)
11753                 return libbpf_err_ptr(err);
11754         return link;
11755 }
11756
11757 static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11758 {
11759         char *path = NULL, *provider = NULL, *name = NULL;
11760         const char *sec_name;
11761         int n, err;
11762
11763         sec_name = bpf_program__section_name(prog);
11764         if (strcmp(sec_name, "usdt") == 0) {
11765                 /* no auto-attach for just SEC("usdt") */
11766                 *link = NULL;
11767                 return 0;
11768         }
11769
11770         n = sscanf(sec_name, "usdt/%m[^:]:%m[^:]:%m[^:]", &path, &provider, &name);
11771         if (n != 3) {
11772                 pr_warn("invalid section '%s', expected SEC(\"usdt/<path>:<provider>:<name>\")\n",
11773                         sec_name);
11774                 err = -EINVAL;
11775         } else {
11776                 *link = bpf_program__attach_usdt(prog, -1 /* any process */, path,
11777                                                  provider, name, NULL);
11778                 err = libbpf_get_error(*link);
11779         }
11780         free(path);
11781         free(provider);
11782         free(name);
11783         return err;
11784 }
11785
11786 static int determine_tracepoint_id(const char *tp_category,
11787                                    const char *tp_name)
11788 {
11789         char file[PATH_MAX];
11790         int ret;
11791
11792         ret = snprintf(file, sizeof(file), "%s/events/%s/%s/id",
11793                        tracefs_path(), tp_category, tp_name);
11794         if (ret < 0)
11795                 return -errno;
11796         if (ret >= sizeof(file)) {
11797                 pr_debug("tracepoint %s/%s path is too long\n",
11798                          tp_category, tp_name);
11799                 return -E2BIG;
11800         }
11801         return parse_uint_from_file(file, "%d\n");
11802 }
11803
11804 static int perf_event_open_tracepoint(const char *tp_category,
11805                                       const char *tp_name)
11806 {
11807         const size_t attr_sz = sizeof(struct perf_event_attr);
11808         struct perf_event_attr attr;
11809         char errmsg[STRERR_BUFSIZE];
11810         int tp_id, pfd, err;
11811
11812         tp_id = determine_tracepoint_id(tp_category, tp_name);
11813         if (tp_id < 0) {
11814                 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
11815                         tp_category, tp_name,
11816                         libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
11817                 return tp_id;
11818         }
11819
11820         memset(&attr, 0, attr_sz);
11821         attr.type = PERF_TYPE_TRACEPOINT;
11822         attr.size = attr_sz;
11823         attr.config = tp_id;
11824
11825         pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
11826                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
11827         if (pfd < 0) {
11828                 err = -errno;
11829                 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
11830                         tp_category, tp_name,
11831                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11832                 return err;
11833         }
11834         return pfd;
11835 }
11836
11837 struct bpf_link *bpf_program__attach_tracepoint_opts(const struct bpf_program *prog,
11838                                                      const char *tp_category,
11839                                                      const char *tp_name,
11840                                                      const struct bpf_tracepoint_opts *opts)
11841 {
11842         DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
11843         char errmsg[STRERR_BUFSIZE];
11844         struct bpf_link *link;
11845         int pfd, err;
11846
11847         if (!OPTS_VALID(opts, bpf_tracepoint_opts))
11848                 return libbpf_err_ptr(-EINVAL);
11849
11850         pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
11851
11852         pfd = perf_event_open_tracepoint(tp_category, tp_name);
11853         if (pfd < 0) {
11854                 pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
11855                         prog->name, tp_category, tp_name,
11856                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
11857                 return libbpf_err_ptr(pfd);
11858         }
11859         link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
11860         err = libbpf_get_error(link);
11861         if (err) {
11862                 close(pfd);
11863                 pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n",
11864                         prog->name, tp_category, tp_name,
11865                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
11866                 return libbpf_err_ptr(err);
11867         }
11868         return link;
11869 }
11870
11871 struct bpf_link *bpf_program__attach_tracepoint(const struct bpf_program *prog,
11872                                                 const char *tp_category,
11873                                                 const char *tp_name)
11874 {
11875         return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL);
11876 }
11877
11878 static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11879 {
11880         char *sec_name, *tp_cat, *tp_name;
11881
11882         *link = NULL;
11883
11884         /* no auto-attach for SEC("tp") or SEC("tracepoint") */
11885         if (strcmp(prog->sec_name, "tp") == 0 || strcmp(prog->sec_name, "tracepoint") == 0)
11886                 return 0;
11887
11888         sec_name = strdup(prog->sec_name);
11889         if (!sec_name)
11890                 return -ENOMEM;
11891
11892         /* extract "tp/<category>/<name>" or "tracepoint/<category>/<name>" */
11893         if (str_has_pfx(prog->sec_name, "tp/"))
11894                 tp_cat = sec_name + sizeof("tp/") - 1;
11895         else
11896                 tp_cat = sec_name + sizeof("tracepoint/") - 1;
11897         tp_name = strchr(tp_cat, '/');
11898         if (!tp_name) {
11899                 free(sec_name);
11900                 return -EINVAL;
11901         }
11902         *tp_name = '\0';
11903         tp_name++;
11904
11905         *link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
11906         free(sec_name);
11907         return libbpf_get_error(*link);
11908 }
11909
11910 struct bpf_link *bpf_program__attach_raw_tracepoint(const struct bpf_program *prog,
11911                                                     const char *tp_name)
11912 {
11913         char errmsg[STRERR_BUFSIZE];
11914         struct bpf_link *link;
11915         int prog_fd, pfd;
11916
11917         prog_fd = bpf_program__fd(prog);
11918         if (prog_fd < 0) {
11919                 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
11920                 return libbpf_err_ptr(-EINVAL);
11921         }
11922
11923         link = calloc(1, sizeof(*link));
11924         if (!link)
11925                 return libbpf_err_ptr(-ENOMEM);
11926         link->detach = &bpf_link__detach_fd;
11927
11928         pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
11929         if (pfd < 0) {
11930                 pfd = -errno;
11931                 free(link);
11932                 pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n",
11933                         prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
11934                 return libbpf_err_ptr(pfd);
11935         }
11936         link->fd = pfd;
11937         return link;
11938 }
11939
11940 static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link)
11941 {
11942         static const char *const prefixes[] = {
11943                 "raw_tp",
11944                 "raw_tracepoint",
11945                 "raw_tp.w",
11946                 "raw_tracepoint.w",
11947         };
11948         size_t i;
11949         const char *tp_name = NULL;
11950
11951         *link = NULL;
11952
11953         for (i = 0; i < ARRAY_SIZE(prefixes); i++) {
11954                 size_t pfx_len;
11955
11956                 if (!str_has_pfx(prog->sec_name, prefixes[i]))
11957                         continue;
11958
11959                 pfx_len = strlen(prefixes[i]);
11960                 /* no auto-attach case of, e.g., SEC("raw_tp") */
11961                 if (prog->sec_name[pfx_len] == '\0')
11962                         return 0;
11963
11964                 if (prog->sec_name[pfx_len] != '/')
11965                         continue;
11966
11967                 tp_name = prog->sec_name + pfx_len + 1;
11968                 break;
11969         }
11970
11971         if (!tp_name) {
11972                 pr_warn("prog '%s': invalid section name '%s'\n",
11973                         prog->name, prog->sec_name);
11974                 return -EINVAL;
11975         }
11976
11977         *link = bpf_program__attach_raw_tracepoint(prog, tp_name);
11978         return libbpf_get_error(*link);
11979 }
11980
11981 /* Common logic for all BPF program types that attach to a btf_id */
11982 static struct bpf_link *bpf_program__attach_btf_id(const struct bpf_program *prog,
11983                                                    const struct bpf_trace_opts *opts)
11984 {
11985         LIBBPF_OPTS(bpf_link_create_opts, link_opts);
11986         char errmsg[STRERR_BUFSIZE];
11987         struct bpf_link *link;
11988         int prog_fd, pfd;
11989
11990         if (!OPTS_VALID(opts, bpf_trace_opts))
11991                 return libbpf_err_ptr(-EINVAL);
11992
11993         prog_fd = bpf_program__fd(prog);
11994         if (prog_fd < 0) {
11995                 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
11996                 return libbpf_err_ptr(-EINVAL);
11997         }
11998
11999         link = calloc(1, sizeof(*link));
12000         if (!link)
12001                 return libbpf_err_ptr(-ENOMEM);
12002         link->detach = &bpf_link__detach_fd;
12003
12004         /* libbpf is smart enough to redirect to BPF_RAW_TRACEPOINT_OPEN on old kernels */
12005         link_opts.tracing.cookie = OPTS_GET(opts, cookie, 0);
12006         pfd = bpf_link_create(prog_fd, 0, bpf_program__expected_attach_type(prog), &link_opts);
12007         if (pfd < 0) {
12008                 pfd = -errno;
12009                 free(link);
12010                 pr_warn("prog '%s': failed to attach: %s\n",
12011                         prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
12012                 return libbpf_err_ptr(pfd);
12013         }
12014         link->fd = pfd;
12015         return link;
12016 }
12017
12018 struct bpf_link *bpf_program__attach_trace(const struct bpf_program *prog)
12019 {
12020         return bpf_program__attach_btf_id(prog, NULL);
12021 }
12022
12023 struct bpf_link *bpf_program__attach_trace_opts(const struct bpf_program *prog,
12024                                                 const struct bpf_trace_opts *opts)
12025 {
12026         return bpf_program__attach_btf_id(prog, opts);
12027 }
12028
12029 struct bpf_link *bpf_program__attach_lsm(const struct bpf_program *prog)
12030 {
12031         return bpf_program__attach_btf_id(prog, NULL);
12032 }
12033
12034 static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12035 {
12036         *link = bpf_program__attach_trace(prog);
12037         return libbpf_get_error(*link);
12038 }
12039
12040 static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12041 {
12042         *link = bpf_program__attach_lsm(prog);
12043         return libbpf_get_error(*link);
12044 }
12045
12046 static struct bpf_link *
12047 bpf_program_attach_fd(const struct bpf_program *prog,
12048                       int target_fd, const char *target_name,
12049                       const struct bpf_link_create_opts *opts)
12050 {
12051         enum bpf_attach_type attach_type;
12052         char errmsg[STRERR_BUFSIZE];
12053         struct bpf_link *link;
12054         int prog_fd, link_fd;
12055
12056         prog_fd = bpf_program__fd(prog);
12057         if (prog_fd < 0) {
12058                 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
12059                 return libbpf_err_ptr(-EINVAL);
12060         }
12061
12062         link = calloc(1, sizeof(*link));
12063         if (!link)
12064                 return libbpf_err_ptr(-ENOMEM);
12065         link->detach = &bpf_link__detach_fd;
12066
12067         attach_type = bpf_program__expected_attach_type(prog);
12068         link_fd = bpf_link_create(prog_fd, target_fd, attach_type, opts);
12069         if (link_fd < 0) {
12070                 link_fd = -errno;
12071                 free(link);
12072                 pr_warn("prog '%s': failed to attach to %s: %s\n",
12073                         prog->name, target_name,
12074                         libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
12075                 return libbpf_err_ptr(link_fd);
12076         }
12077         link->fd = link_fd;
12078         return link;
12079 }
12080
12081 struct bpf_link *
12082 bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd)
12083 {
12084         return bpf_program_attach_fd(prog, cgroup_fd, "cgroup", NULL);
12085 }
12086
12087 struct bpf_link *
12088 bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd)
12089 {
12090         return bpf_program_attach_fd(prog, netns_fd, "netns", NULL);
12091 }
12092
12093 struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex)
12094 {
12095         /* target_fd/target_ifindex use the same field in LINK_CREATE */
12096         return bpf_program_attach_fd(prog, ifindex, "xdp", NULL);
12097 }
12098
12099 struct bpf_link *
12100 bpf_program__attach_tcx(const struct bpf_program *prog, int ifindex,
12101                         const struct bpf_tcx_opts *opts)
12102 {
12103         LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
12104         __u32 relative_id;
12105         int relative_fd;
12106
12107         if (!OPTS_VALID(opts, bpf_tcx_opts))
12108                 return libbpf_err_ptr(-EINVAL);
12109
12110         relative_id = OPTS_GET(opts, relative_id, 0);
12111         relative_fd = OPTS_GET(opts, relative_fd, 0);
12112
12113         /* validate we don't have unexpected combinations of non-zero fields */
12114         if (!ifindex) {
12115                 pr_warn("prog '%s': target netdevice ifindex cannot be zero\n",
12116                         prog->name);
12117                 return libbpf_err_ptr(-EINVAL);
12118         }
12119         if (relative_fd && relative_id) {
12120                 pr_warn("prog '%s': relative_fd and relative_id cannot be set at the same time\n",
12121                         prog->name);
12122                 return libbpf_err_ptr(-EINVAL);
12123         }
12124
12125         link_create_opts.tcx.expected_revision = OPTS_GET(opts, expected_revision, 0);
12126         link_create_opts.tcx.relative_fd = relative_fd;
12127         link_create_opts.tcx.relative_id = relative_id;
12128         link_create_opts.flags = OPTS_GET(opts, flags, 0);
12129
12130         /* target_fd/target_ifindex use the same field in LINK_CREATE */
12131         return bpf_program_attach_fd(prog, ifindex, "tcx", &link_create_opts);
12132 }
12133
12134 struct bpf_link *
12135 bpf_program__attach_netkit(const struct bpf_program *prog, int ifindex,
12136                            const struct bpf_netkit_opts *opts)
12137 {
12138         LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
12139         __u32 relative_id;
12140         int relative_fd;
12141
12142         if (!OPTS_VALID(opts, bpf_netkit_opts))
12143                 return libbpf_err_ptr(-EINVAL);
12144
12145         relative_id = OPTS_GET(opts, relative_id, 0);
12146         relative_fd = OPTS_GET(opts, relative_fd, 0);
12147
12148         /* validate we don't have unexpected combinations of non-zero fields */
12149         if (!ifindex) {
12150                 pr_warn("prog '%s': target netdevice ifindex cannot be zero\n",
12151                         prog->name);
12152                 return libbpf_err_ptr(-EINVAL);
12153         }
12154         if (relative_fd && relative_id) {
12155                 pr_warn("prog '%s': relative_fd and relative_id cannot be set at the same time\n",
12156                         prog->name);
12157                 return libbpf_err_ptr(-EINVAL);
12158         }
12159
12160         link_create_opts.netkit.expected_revision = OPTS_GET(opts, expected_revision, 0);
12161         link_create_opts.netkit.relative_fd = relative_fd;
12162         link_create_opts.netkit.relative_id = relative_id;
12163         link_create_opts.flags = OPTS_GET(opts, flags, 0);
12164
12165         return bpf_program_attach_fd(prog, ifindex, "netkit", &link_create_opts);
12166 }
12167
12168 struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog,
12169                                               int target_fd,
12170                                               const char *attach_func_name)
12171 {
12172         int btf_id;
12173
12174         if (!!target_fd != !!attach_func_name) {
12175                 pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n",
12176                         prog->name);
12177                 return libbpf_err_ptr(-EINVAL);
12178         }
12179
12180         if (prog->type != BPF_PROG_TYPE_EXT) {
12181                 pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace",
12182                         prog->name);
12183                 return libbpf_err_ptr(-EINVAL);
12184         }
12185
12186         if (target_fd) {
12187                 LIBBPF_OPTS(bpf_link_create_opts, target_opts);
12188
12189                 btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd);
12190                 if (btf_id < 0)
12191                         return libbpf_err_ptr(btf_id);
12192
12193                 target_opts.target_btf_id = btf_id;
12194
12195                 return bpf_program_attach_fd(prog, target_fd, "freplace",
12196                                              &target_opts);
12197         } else {
12198                 /* no target, so use raw_tracepoint_open for compatibility
12199                  * with old kernels
12200                  */
12201                 return bpf_program__attach_trace(prog);
12202         }
12203 }
12204
12205 struct bpf_link *
12206 bpf_program__attach_iter(const struct bpf_program *prog,
12207                          const struct bpf_iter_attach_opts *opts)
12208 {
12209         DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
12210         char errmsg[STRERR_BUFSIZE];
12211         struct bpf_link *link;
12212         int prog_fd, link_fd;
12213         __u32 target_fd = 0;
12214
12215         if (!OPTS_VALID(opts, bpf_iter_attach_opts))
12216                 return libbpf_err_ptr(-EINVAL);
12217
12218         link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0);
12219         link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0);
12220
12221         prog_fd = bpf_program__fd(prog);
12222         if (prog_fd < 0) {
12223                 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
12224                 return libbpf_err_ptr(-EINVAL);
12225         }
12226
12227         link = calloc(1, sizeof(*link));
12228         if (!link)
12229                 return libbpf_err_ptr(-ENOMEM);
12230         link->detach = &bpf_link__detach_fd;
12231
12232         link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER,
12233                                   &link_create_opts);
12234         if (link_fd < 0) {
12235                 link_fd = -errno;
12236                 free(link);
12237                 pr_warn("prog '%s': failed to attach to iterator: %s\n",
12238                         prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
12239                 return libbpf_err_ptr(link_fd);
12240         }
12241         link->fd = link_fd;
12242         return link;
12243 }
12244
12245 static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12246 {
12247         *link = bpf_program__attach_iter(prog, NULL);
12248         return libbpf_get_error(*link);
12249 }
12250
12251 struct bpf_link *bpf_program__attach_netfilter(const struct bpf_program *prog,
12252                                                const struct bpf_netfilter_opts *opts)
12253 {
12254         LIBBPF_OPTS(bpf_link_create_opts, lopts);
12255         struct bpf_link *link;
12256         int prog_fd, link_fd;
12257
12258         if (!OPTS_VALID(opts, bpf_netfilter_opts))
12259                 return libbpf_err_ptr(-EINVAL);
12260
12261         prog_fd = bpf_program__fd(prog);
12262         if (prog_fd < 0) {
12263                 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
12264                 return libbpf_err_ptr(-EINVAL);
12265         }
12266
12267         link = calloc(1, sizeof(*link));
12268         if (!link)
12269                 return libbpf_err_ptr(-ENOMEM);
12270
12271         link->detach = &bpf_link__detach_fd;
12272
12273         lopts.netfilter.pf = OPTS_GET(opts, pf, 0);
12274         lopts.netfilter.hooknum = OPTS_GET(opts, hooknum, 0);
12275         lopts.netfilter.priority = OPTS_GET(opts, priority, 0);
12276         lopts.netfilter.flags = OPTS_GET(opts, flags, 0);
12277
12278         link_fd = bpf_link_create(prog_fd, 0, BPF_NETFILTER, &lopts);
12279         if (link_fd < 0) {
12280                 char errmsg[STRERR_BUFSIZE];
12281
12282                 link_fd = -errno;
12283                 free(link);
12284                 pr_warn("prog '%s': failed to attach to netfilter: %s\n",
12285                         prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
12286                 return libbpf_err_ptr(link_fd);
12287         }
12288         link->fd = link_fd;
12289
12290         return link;
12291 }
12292
12293 struct bpf_link *bpf_program__attach(const struct bpf_program *prog)
12294 {
12295         struct bpf_link *link = NULL;
12296         int err;
12297
12298         if (!prog->sec_def || !prog->sec_def->prog_attach_fn)
12299                 return libbpf_err_ptr(-EOPNOTSUPP);
12300
12301         err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, &link);
12302         if (err)
12303                 return libbpf_err_ptr(err);
12304
12305         /* When calling bpf_program__attach() explicitly, auto-attach support
12306          * is expected to work, so NULL returned link is considered an error.
12307          * This is different for skeleton's attach, see comment in
12308          * bpf_object__attach_skeleton().
12309          */
12310         if (!link)
12311                 return libbpf_err_ptr(-EOPNOTSUPP);
12312
12313         return link;
12314 }
12315
12316 struct bpf_link_struct_ops {
12317         struct bpf_link link;
12318         int map_fd;
12319 };
12320
12321 static int bpf_link__detach_struct_ops(struct bpf_link *link)
12322 {
12323         struct bpf_link_struct_ops *st_link;
12324         __u32 zero = 0;
12325
12326         st_link = container_of(link, struct bpf_link_struct_ops, link);
12327
12328         if (st_link->map_fd < 0)
12329                 /* w/o a real link */
12330                 return bpf_map_delete_elem(link->fd, &zero);
12331
12332         return close(link->fd);
12333 }
12334
12335 struct bpf_link *bpf_map__attach_struct_ops(const struct bpf_map *map)
12336 {
12337         struct bpf_link_struct_ops *link;
12338         __u32 zero = 0;
12339         int err, fd;
12340
12341         if (!bpf_map__is_struct_ops(map) || map->fd == -1)
12342                 return libbpf_err_ptr(-EINVAL);
12343
12344         link = calloc(1, sizeof(*link));
12345         if (!link)
12346                 return libbpf_err_ptr(-EINVAL);
12347
12348         /* kern_vdata should be prepared during the loading phase. */
12349         err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0);
12350         /* It can be EBUSY if the map has been used to create or
12351          * update a link before.  We don't allow updating the value of
12352          * a struct_ops once it is set.  That ensures that the value
12353          * never changed.  So, it is safe to skip EBUSY.
12354          */
12355         if (err && (!(map->def.map_flags & BPF_F_LINK) || err != -EBUSY)) {
12356                 free(link);
12357                 return libbpf_err_ptr(err);
12358         }
12359
12360         link->link.detach = bpf_link__detach_struct_ops;
12361
12362         if (!(map->def.map_flags & BPF_F_LINK)) {
12363                 /* w/o a real link */
12364                 link->link.fd = map->fd;
12365                 link->map_fd = -1;
12366                 return &link->link;
12367         }
12368
12369         fd = bpf_link_create(map->fd, 0, BPF_STRUCT_OPS, NULL);
12370         if (fd < 0) {
12371                 free(link);
12372                 return libbpf_err_ptr(fd);
12373         }
12374
12375         link->link.fd = fd;
12376         link->map_fd = map->fd;
12377
12378         return &link->link;
12379 }
12380
12381 /*
12382  * Swap the back struct_ops of a link with a new struct_ops map.
12383  */
12384 int bpf_link__update_map(struct bpf_link *link, const struct bpf_map *map)
12385 {
12386         struct bpf_link_struct_ops *st_ops_link;
12387         __u32 zero = 0;
12388         int err;
12389
12390         if (!bpf_map__is_struct_ops(map) || map->fd < 0)
12391                 return -EINVAL;
12392
12393         st_ops_link = container_of(link, struct bpf_link_struct_ops, link);
12394         /* Ensure the type of a link is correct */
12395         if (st_ops_link->map_fd < 0)
12396                 return -EINVAL;
12397
12398         err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0);
12399         /* It can be EBUSY if the map has been used to create or
12400          * update a link before.  We don't allow updating the value of
12401          * a struct_ops once it is set.  That ensures that the value
12402          * never changed.  So, it is safe to skip EBUSY.
12403          */
12404         if (err && err != -EBUSY)
12405                 return err;
12406
12407         err = bpf_link_update(link->fd, map->fd, NULL);
12408         if (err < 0)
12409                 return err;
12410
12411         st_ops_link->map_fd = map->fd;
12412
12413         return 0;
12414 }
12415
12416 typedef enum bpf_perf_event_ret (*bpf_perf_event_print_t)(struct perf_event_header *hdr,
12417                                                           void *private_data);
12418
12419 static enum bpf_perf_event_ret
12420 perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
12421                        void **copy_mem, size_t *copy_size,
12422                        bpf_perf_event_print_t fn, void *private_data)
12423 {
12424         struct perf_event_mmap_page *header = mmap_mem;
12425         __u64 data_head = ring_buffer_read_head(header);
12426         __u64 data_tail = header->data_tail;
12427         void *base = ((__u8 *)header) + page_size;
12428         int ret = LIBBPF_PERF_EVENT_CONT;
12429         struct perf_event_header *ehdr;
12430         size_t ehdr_size;
12431
12432         while (data_head != data_tail) {
12433                 ehdr = base + (data_tail & (mmap_size - 1));
12434                 ehdr_size = ehdr->size;
12435
12436                 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
12437                         void *copy_start = ehdr;
12438                         size_t len_first = base + mmap_size - copy_start;
12439                         size_t len_secnd = ehdr_size - len_first;
12440
12441                         if (*copy_size < ehdr_size) {
12442                                 free(*copy_mem);
12443                                 *copy_mem = malloc(ehdr_size);
12444                                 if (!*copy_mem) {
12445                                         *copy_size = 0;
12446                                         ret = LIBBPF_PERF_EVENT_ERROR;
12447                                         break;
12448                                 }
12449                                 *copy_size = ehdr_size;
12450                         }
12451
12452                         memcpy(*copy_mem, copy_start, len_first);
12453                         memcpy(*copy_mem + len_first, base, len_secnd);
12454                         ehdr = *copy_mem;
12455                 }
12456
12457                 ret = fn(ehdr, private_data);
12458                 data_tail += ehdr_size;
12459                 if (ret != LIBBPF_PERF_EVENT_CONT)
12460                         break;
12461         }
12462
12463         ring_buffer_write_tail(header, data_tail);
12464         return libbpf_err(ret);
12465 }
12466
12467 struct perf_buffer;
12468
12469 struct perf_buffer_params {
12470         struct perf_event_attr *attr;
12471         /* if event_cb is specified, it takes precendence */
12472         perf_buffer_event_fn event_cb;
12473         /* sample_cb and lost_cb are higher-level common-case callbacks */
12474         perf_buffer_sample_fn sample_cb;
12475         perf_buffer_lost_fn lost_cb;
12476         void *ctx;
12477         int cpu_cnt;
12478         int *cpus;
12479         int *map_keys;
12480 };
12481
12482 struct perf_cpu_buf {
12483         struct perf_buffer *pb;
12484         void *base; /* mmap()'ed memory */
12485         void *buf; /* for reconstructing segmented data */
12486         size_t buf_size;
12487         int fd;
12488         int cpu;
12489         int map_key;
12490 };
12491
12492 struct perf_buffer {
12493         perf_buffer_event_fn event_cb;
12494         perf_buffer_sample_fn sample_cb;
12495         perf_buffer_lost_fn lost_cb;
12496         void *ctx; /* passed into callbacks */
12497
12498         size_t page_size;
12499         size_t mmap_size;
12500         struct perf_cpu_buf **cpu_bufs;
12501         struct epoll_event *events;
12502         int cpu_cnt; /* number of allocated CPU buffers */
12503         int epoll_fd; /* perf event FD */
12504         int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
12505 };
12506
12507 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
12508                                       struct perf_cpu_buf *cpu_buf)
12509 {
12510         if (!cpu_buf)
12511                 return;
12512         if (cpu_buf->base &&
12513             munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
12514                 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
12515         if (cpu_buf->fd >= 0) {
12516                 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
12517                 close(cpu_buf->fd);
12518         }
12519         free(cpu_buf->buf);
12520         free(cpu_buf);
12521 }
12522
12523 void perf_buffer__free(struct perf_buffer *pb)
12524 {
12525         int i;
12526
12527         if (IS_ERR_OR_NULL(pb))
12528                 return;
12529         if (pb->cpu_bufs) {
12530                 for (i = 0; i < pb->cpu_cnt; i++) {
12531                         struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
12532
12533                         if (!cpu_buf)
12534                                 continue;
12535
12536                         bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
12537                         perf_buffer__free_cpu_buf(pb, cpu_buf);
12538                 }
12539                 free(pb->cpu_bufs);
12540         }
12541         if (pb->epoll_fd >= 0)
12542                 close(pb->epoll_fd);
12543         free(pb->events);
12544         free(pb);
12545 }
12546
12547 static struct perf_cpu_buf *
12548 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
12549                           int cpu, int map_key)
12550 {
12551         struct perf_cpu_buf *cpu_buf;
12552         char msg[STRERR_BUFSIZE];
12553         int err;
12554
12555         cpu_buf = calloc(1, sizeof(*cpu_buf));
12556         if (!cpu_buf)
12557                 return ERR_PTR(-ENOMEM);
12558
12559         cpu_buf->pb = pb;
12560         cpu_buf->cpu = cpu;
12561         cpu_buf->map_key = map_key;
12562
12563         cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
12564                               -1, PERF_FLAG_FD_CLOEXEC);
12565         if (cpu_buf->fd < 0) {
12566                 err = -errno;
12567                 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
12568                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
12569                 goto error;
12570         }
12571
12572         cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
12573                              PROT_READ | PROT_WRITE, MAP_SHARED,
12574                              cpu_buf->fd, 0);
12575         if (cpu_buf->base == MAP_FAILED) {
12576                 cpu_buf->base = NULL;
12577                 err = -errno;
12578                 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
12579                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
12580                 goto error;
12581         }
12582
12583         if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
12584                 err = -errno;
12585                 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
12586                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
12587                 goto error;
12588         }
12589
12590         return cpu_buf;
12591
12592 error:
12593         perf_buffer__free_cpu_buf(pb, cpu_buf);
12594         return (struct perf_cpu_buf *)ERR_PTR(err);
12595 }
12596
12597 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
12598                                               struct perf_buffer_params *p);
12599
12600 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
12601                                      perf_buffer_sample_fn sample_cb,
12602                                      perf_buffer_lost_fn lost_cb,
12603                                      void *ctx,
12604                                      const struct perf_buffer_opts *opts)
12605 {
12606         const size_t attr_sz = sizeof(struct perf_event_attr);
12607         struct perf_buffer_params p = {};
12608         struct perf_event_attr attr;
12609         __u32 sample_period;
12610
12611         if (!OPTS_VALID(opts, perf_buffer_opts))
12612                 return libbpf_err_ptr(-EINVAL);
12613
12614         sample_period = OPTS_GET(opts, sample_period, 1);
12615         if (!sample_period)
12616                 sample_period = 1;
12617
12618         memset(&attr, 0, attr_sz);
12619         attr.size = attr_sz;
12620         attr.config = PERF_COUNT_SW_BPF_OUTPUT;
12621         attr.type = PERF_TYPE_SOFTWARE;
12622         attr.sample_type = PERF_SAMPLE_RAW;
12623         attr.sample_period = sample_period;
12624         attr.wakeup_events = sample_period;
12625
12626         p.attr = &attr;
12627         p.sample_cb = sample_cb;
12628         p.lost_cb = lost_cb;
12629         p.ctx = ctx;
12630
12631         return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
12632 }
12633
12634 struct perf_buffer *perf_buffer__new_raw(int map_fd, size_t page_cnt,
12635                                          struct perf_event_attr *attr,
12636                                          perf_buffer_event_fn event_cb, void *ctx,
12637                                          const struct perf_buffer_raw_opts *opts)
12638 {
12639         struct perf_buffer_params p = {};
12640
12641         if (!attr)
12642                 return libbpf_err_ptr(-EINVAL);
12643
12644         if (!OPTS_VALID(opts, perf_buffer_raw_opts))
12645                 return libbpf_err_ptr(-EINVAL);
12646
12647         p.attr = attr;
12648         p.event_cb = event_cb;
12649         p.ctx = ctx;
12650         p.cpu_cnt = OPTS_GET(opts, cpu_cnt, 0);
12651         p.cpus = OPTS_GET(opts, cpus, NULL);
12652         p.map_keys = OPTS_GET(opts, map_keys, NULL);
12653
12654         return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
12655 }
12656
12657 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
12658                                               struct perf_buffer_params *p)
12659 {
12660         const char *online_cpus_file = "/sys/devices/system/cpu/online";
12661         struct bpf_map_info map;
12662         char msg[STRERR_BUFSIZE];
12663         struct perf_buffer *pb;
12664         bool *online = NULL;
12665         __u32 map_info_len;
12666         int err, i, j, n;
12667
12668         if (page_cnt == 0 || (page_cnt & (page_cnt - 1))) {
12669                 pr_warn("page count should be power of two, but is %zu\n",
12670                         page_cnt);
12671                 return ERR_PTR(-EINVAL);
12672         }
12673
12674         /* best-effort sanity checks */
12675         memset(&map, 0, sizeof(map));
12676         map_info_len = sizeof(map);
12677         err = bpf_map_get_info_by_fd(map_fd, &map, &map_info_len);
12678         if (err) {
12679                 err = -errno;
12680                 /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return
12681                  * -EBADFD, -EFAULT, or -E2BIG on real error
12682                  */
12683                 if (err != -EINVAL) {
12684                         pr_warn("failed to get map info for map FD %d: %s\n",
12685                                 map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
12686                         return ERR_PTR(err);
12687                 }
12688                 pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n",
12689                          map_fd);
12690         } else {
12691                 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
12692                         pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
12693                                 map.name);
12694                         return ERR_PTR(-EINVAL);
12695                 }
12696         }
12697
12698         pb = calloc(1, sizeof(*pb));
12699         if (!pb)
12700                 return ERR_PTR(-ENOMEM);
12701
12702         pb->event_cb = p->event_cb;
12703         pb->sample_cb = p->sample_cb;
12704         pb->lost_cb = p->lost_cb;
12705         pb->ctx = p->ctx;
12706
12707         pb->page_size = getpagesize();
12708         pb->mmap_size = pb->page_size * page_cnt;
12709         pb->map_fd = map_fd;
12710
12711         pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
12712         if (pb->epoll_fd < 0) {
12713                 err = -errno;
12714                 pr_warn("failed to create epoll instance: %s\n",
12715                         libbpf_strerror_r(err, msg, sizeof(msg)));
12716                 goto error;
12717         }
12718
12719         if (p->cpu_cnt > 0) {
12720                 pb->cpu_cnt = p->cpu_cnt;
12721         } else {
12722                 pb->cpu_cnt = libbpf_num_possible_cpus();
12723                 if (pb->cpu_cnt < 0) {
12724                         err = pb->cpu_cnt;
12725                         goto error;
12726                 }
12727                 if (map.max_entries && map.max_entries < pb->cpu_cnt)
12728                         pb->cpu_cnt = map.max_entries;
12729         }
12730
12731         pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
12732         if (!pb->events) {
12733                 err = -ENOMEM;
12734                 pr_warn("failed to allocate events: out of memory\n");
12735                 goto error;
12736         }
12737         pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
12738         if (!pb->cpu_bufs) {
12739                 err = -ENOMEM;
12740                 pr_warn("failed to allocate buffers: out of memory\n");
12741                 goto error;
12742         }
12743
12744         err = parse_cpu_mask_file(online_cpus_file, &online, &n);
12745         if (err) {
12746                 pr_warn("failed to get online CPU mask: %d\n", err);
12747                 goto error;
12748         }
12749
12750         for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
12751                 struct perf_cpu_buf *cpu_buf;
12752                 int cpu, map_key;
12753
12754                 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
12755                 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
12756
12757                 /* in case user didn't explicitly requested particular CPUs to
12758                  * be attached to, skip offline/not present CPUs
12759                  */
12760                 if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
12761                         continue;
12762
12763                 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
12764                 if (IS_ERR(cpu_buf)) {
12765                         err = PTR_ERR(cpu_buf);
12766                         goto error;
12767                 }
12768
12769                 pb->cpu_bufs[j] = cpu_buf;
12770
12771                 err = bpf_map_update_elem(pb->map_fd, &map_key,
12772                                           &cpu_buf->fd, 0);
12773                 if (err) {
12774                         err = -errno;
12775                         pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
12776                                 cpu, map_key, cpu_buf->fd,
12777                                 libbpf_strerror_r(err, msg, sizeof(msg)));
12778                         goto error;
12779                 }
12780
12781                 pb->events[j].events = EPOLLIN;
12782                 pb->events[j].data.ptr = cpu_buf;
12783                 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
12784                               &pb->events[j]) < 0) {
12785                         err = -errno;
12786                         pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
12787                                 cpu, cpu_buf->fd,
12788                                 libbpf_strerror_r(err, msg, sizeof(msg)));
12789                         goto error;
12790                 }
12791                 j++;
12792         }
12793         pb->cpu_cnt = j;
12794         free(online);
12795
12796         return pb;
12797
12798 error:
12799         free(online);
12800         if (pb)
12801                 perf_buffer__free(pb);
12802         return ERR_PTR(err);
12803 }
12804
12805 struct perf_sample_raw {
12806         struct perf_event_header header;
12807         uint32_t size;
12808         char data[];
12809 };
12810
12811 struct perf_sample_lost {
12812         struct perf_event_header header;
12813         uint64_t id;
12814         uint64_t lost;
12815         uint64_t sample_id;
12816 };
12817
12818 static enum bpf_perf_event_ret
12819 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
12820 {
12821         struct perf_cpu_buf *cpu_buf = ctx;
12822         struct perf_buffer *pb = cpu_buf->pb;
12823         void *data = e;
12824
12825         /* user wants full control over parsing perf event */
12826         if (pb->event_cb)
12827                 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
12828
12829         switch (e->type) {
12830         case PERF_RECORD_SAMPLE: {
12831                 struct perf_sample_raw *s = data;
12832
12833                 if (pb->sample_cb)
12834                         pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
12835                 break;
12836         }
12837         case PERF_RECORD_LOST: {
12838                 struct perf_sample_lost *s = data;
12839
12840                 if (pb->lost_cb)
12841                         pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
12842                 break;
12843         }
12844         default:
12845                 pr_warn("unknown perf sample type %d\n", e->type);
12846                 return LIBBPF_PERF_EVENT_ERROR;
12847         }
12848         return LIBBPF_PERF_EVENT_CONT;
12849 }
12850
12851 static int perf_buffer__process_records(struct perf_buffer *pb,
12852                                         struct perf_cpu_buf *cpu_buf)
12853 {
12854         enum bpf_perf_event_ret ret;
12855
12856         ret = perf_event_read_simple(cpu_buf->base, pb->mmap_size,
12857                                      pb->page_size, &cpu_buf->buf,
12858                                      &cpu_buf->buf_size,
12859                                      perf_buffer__process_record, cpu_buf);
12860         if (ret != LIBBPF_PERF_EVENT_CONT)
12861                 return ret;
12862         return 0;
12863 }
12864
12865 int perf_buffer__epoll_fd(const struct perf_buffer *pb)
12866 {
12867         return pb->epoll_fd;
12868 }
12869
12870 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
12871 {
12872         int i, cnt, err;
12873
12874         cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
12875         if (cnt < 0)
12876                 return -errno;
12877
12878         for (i = 0; i < cnt; i++) {
12879                 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
12880
12881                 err = perf_buffer__process_records(pb, cpu_buf);
12882                 if (err) {
12883                         pr_warn("error while processing records: %d\n", err);
12884                         return libbpf_err(err);
12885                 }
12886         }
12887         return cnt;
12888 }
12889
12890 /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer
12891  * manager.
12892  */
12893 size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb)
12894 {
12895         return pb->cpu_cnt;
12896 }
12897
12898 /*
12899  * Return perf_event FD of a ring buffer in *buf_idx* slot of
12900  * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using
12901  * select()/poll()/epoll() Linux syscalls.
12902  */
12903 int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx)
12904 {
12905         struct perf_cpu_buf *cpu_buf;
12906
12907         if (buf_idx >= pb->cpu_cnt)
12908                 return libbpf_err(-EINVAL);
12909
12910         cpu_buf = pb->cpu_bufs[buf_idx];
12911         if (!cpu_buf)
12912                 return libbpf_err(-ENOENT);
12913
12914         return cpu_buf->fd;
12915 }
12916
12917 int perf_buffer__buffer(struct perf_buffer *pb, int buf_idx, void **buf, size_t *buf_size)
12918 {
12919         struct perf_cpu_buf *cpu_buf;
12920
12921         if (buf_idx >= pb->cpu_cnt)
12922                 return libbpf_err(-EINVAL);
12923
12924         cpu_buf = pb->cpu_bufs[buf_idx];
12925         if (!cpu_buf)
12926                 return libbpf_err(-ENOENT);
12927
12928         *buf = cpu_buf->base;
12929         *buf_size = pb->mmap_size;
12930         return 0;
12931 }
12932
12933 /*
12934  * Consume data from perf ring buffer corresponding to slot *buf_idx* in
12935  * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to
12936  * consume, do nothing and return success.
12937  * Returns:
12938  *   - 0 on success;
12939  *   - <0 on failure.
12940  */
12941 int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx)
12942 {
12943         struct perf_cpu_buf *cpu_buf;
12944
12945         if (buf_idx >= pb->cpu_cnt)
12946                 return libbpf_err(-EINVAL);
12947
12948         cpu_buf = pb->cpu_bufs[buf_idx];
12949         if (!cpu_buf)
12950                 return libbpf_err(-ENOENT);
12951
12952         return perf_buffer__process_records(pb, cpu_buf);
12953 }
12954
12955 int perf_buffer__consume(struct perf_buffer *pb)
12956 {
12957         int i, err;
12958
12959         for (i = 0; i < pb->cpu_cnt; i++) {
12960                 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
12961
12962                 if (!cpu_buf)
12963                         continue;
12964
12965                 err = perf_buffer__process_records(pb, cpu_buf);
12966                 if (err) {
12967                         pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err);
12968                         return libbpf_err(err);
12969                 }
12970         }
12971         return 0;
12972 }
12973
12974 int bpf_program__set_attach_target(struct bpf_program *prog,
12975                                    int attach_prog_fd,
12976                                    const char *attach_func_name)
12977 {
12978         int btf_obj_fd = 0, btf_id = 0, err;
12979
12980         if (!prog || attach_prog_fd < 0)
12981                 return libbpf_err(-EINVAL);
12982
12983         if (prog->obj->loaded)
12984                 return libbpf_err(-EINVAL);
12985
12986         if (attach_prog_fd && !attach_func_name) {
12987                 /* remember attach_prog_fd and let bpf_program__load() find
12988                  * BTF ID during the program load
12989                  */
12990                 prog->attach_prog_fd = attach_prog_fd;
12991                 return 0;
12992         }
12993
12994         if (attach_prog_fd) {
12995                 btf_id = libbpf_find_prog_btf_id(attach_func_name,
12996                                                  attach_prog_fd);
12997                 if (btf_id < 0)
12998                         return libbpf_err(btf_id);
12999         } else {
13000                 if (!attach_func_name)
13001                         return libbpf_err(-EINVAL);
13002
13003                 /* load btf_vmlinux, if not yet */
13004                 err = bpf_object__load_vmlinux_btf(prog->obj, true);
13005                 if (err)
13006                         return libbpf_err(err);
13007                 err = find_kernel_btf_id(prog->obj, attach_func_name,
13008                                          prog->expected_attach_type,
13009                                          &btf_obj_fd, &btf_id);
13010                 if (err)
13011                         return libbpf_err(err);
13012         }
13013
13014         prog->attach_btf_id = btf_id;
13015         prog->attach_btf_obj_fd = btf_obj_fd;
13016         prog->attach_prog_fd = attach_prog_fd;
13017         return 0;
13018 }
13019
13020 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
13021 {
13022         int err = 0, n, len, start, end = -1;
13023         bool *tmp;
13024
13025         *mask = NULL;
13026         *mask_sz = 0;
13027
13028         /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
13029         while (*s) {
13030                 if (*s == ',' || *s == '\n') {
13031                         s++;
13032                         continue;
13033                 }
13034                 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
13035                 if (n <= 0 || n > 2) {
13036                         pr_warn("Failed to get CPU range %s: %d\n", s, n);
13037                         err = -EINVAL;
13038                         goto cleanup;
13039                 } else if (n == 1) {
13040                         end = start;
13041                 }
13042                 if (start < 0 || start > end) {
13043                         pr_warn("Invalid CPU range [%d,%d] in %s\n",
13044                                 start, end, s);
13045                         err = -EINVAL;
13046                         goto cleanup;
13047                 }
13048                 tmp = realloc(*mask, end + 1);
13049                 if (!tmp) {
13050                         err = -ENOMEM;
13051                         goto cleanup;
13052                 }
13053                 *mask = tmp;
13054                 memset(tmp + *mask_sz, 0, start - *mask_sz);
13055                 memset(tmp + start, 1, end - start + 1);
13056                 *mask_sz = end + 1;
13057                 s += len;
13058         }
13059         if (!*mask_sz) {
13060                 pr_warn("Empty CPU range\n");
13061                 return -EINVAL;
13062         }
13063         return 0;
13064 cleanup:
13065         free(*mask);
13066         *mask = NULL;
13067         return err;
13068 }
13069
13070 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
13071 {
13072         int fd, err = 0, len;
13073         char buf[128];
13074
13075         fd = open(fcpu, O_RDONLY | O_CLOEXEC);
13076         if (fd < 0) {
13077                 err = -errno;
13078                 pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err);
13079                 return err;
13080         }
13081         len = read(fd, buf, sizeof(buf));
13082         close(fd);
13083         if (len <= 0) {
13084                 err = len ? -errno : -EINVAL;
13085                 pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err);
13086                 return err;
13087         }
13088         if (len >= sizeof(buf)) {
13089                 pr_warn("CPU mask is too big in file %s\n", fcpu);
13090                 return -E2BIG;
13091         }
13092         buf[len] = '\0';
13093
13094         return parse_cpu_mask_str(buf, mask, mask_sz);
13095 }
13096
13097 int libbpf_num_possible_cpus(void)
13098 {
13099         static const char *fcpu = "/sys/devices/system/cpu/possible";
13100         static int cpus;
13101         int err, n, i, tmp_cpus;
13102         bool *mask;
13103
13104         tmp_cpus = READ_ONCE(cpus);
13105         if (tmp_cpus > 0)
13106                 return tmp_cpus;
13107
13108         err = parse_cpu_mask_file(fcpu, &mask, &n);
13109         if (err)
13110                 return libbpf_err(err);
13111
13112         tmp_cpus = 0;
13113         for (i = 0; i < n; i++) {
13114                 if (mask[i])
13115                         tmp_cpus++;
13116         }
13117         free(mask);
13118
13119         WRITE_ONCE(cpus, tmp_cpus);
13120         return tmp_cpus;
13121 }
13122
13123 static int populate_skeleton_maps(const struct bpf_object *obj,
13124                                   struct bpf_map_skeleton *maps,
13125                                   size_t map_cnt)
13126 {
13127         int i;
13128
13129         for (i = 0; i < map_cnt; i++) {
13130                 struct bpf_map **map = maps[i].map;
13131                 const char *name = maps[i].name;
13132                 void **mmaped = maps[i].mmaped;
13133
13134                 *map = bpf_object__find_map_by_name(obj, name);
13135                 if (!*map) {
13136                         pr_warn("failed to find skeleton map '%s'\n", name);
13137                         return -ESRCH;
13138                 }
13139
13140                 /* externs shouldn't be pre-setup from user code */
13141                 if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
13142                         *mmaped = (*map)->mmaped;
13143         }
13144         return 0;
13145 }
13146
13147 static int populate_skeleton_progs(const struct bpf_object *obj,
13148                                    struct bpf_prog_skeleton *progs,
13149                                    size_t prog_cnt)
13150 {
13151         int i;
13152
13153         for (i = 0; i < prog_cnt; i++) {
13154                 struct bpf_program **prog = progs[i].prog;
13155                 const char *name = progs[i].name;
13156
13157                 *prog = bpf_object__find_program_by_name(obj, name);
13158                 if (!*prog) {
13159                         pr_warn("failed to find skeleton program '%s'\n", name);
13160                         return -ESRCH;
13161                 }
13162         }
13163         return 0;
13164 }
13165
13166 int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
13167                               const struct bpf_object_open_opts *opts)
13168 {
13169         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts,
13170                 .object_name = s->name,
13171         );
13172         struct bpf_object *obj;
13173         int err;
13174
13175         /* Attempt to preserve opts->object_name, unless overriden by user
13176          * explicitly. Overwriting object name for skeletons is discouraged,
13177          * as it breaks global data maps, because they contain object name
13178          * prefix as their own map name prefix. When skeleton is generated,
13179          * bpftool is making an assumption that this name will stay the same.
13180          */
13181         if (opts) {
13182                 memcpy(&skel_opts, opts, sizeof(*opts));
13183                 if (!opts->object_name)
13184                         skel_opts.object_name = s->name;
13185         }
13186
13187         obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts);
13188         err = libbpf_get_error(obj);
13189         if (err) {
13190                 pr_warn("failed to initialize skeleton BPF object '%s': %d\n",
13191                         s->name, err);
13192                 return libbpf_err(err);
13193         }
13194
13195         *s->obj = obj;
13196         err = populate_skeleton_maps(obj, s->maps, s->map_cnt);
13197         if (err) {
13198                 pr_warn("failed to populate skeleton maps for '%s': %d\n", s->name, err);
13199                 return libbpf_err(err);
13200         }
13201
13202         err = populate_skeleton_progs(obj, s->progs, s->prog_cnt);
13203         if (err) {
13204                 pr_warn("failed to populate skeleton progs for '%s': %d\n", s->name, err);
13205                 return libbpf_err(err);
13206         }
13207
13208         return 0;
13209 }
13210
13211 int bpf_object__open_subskeleton(struct bpf_object_subskeleton *s)
13212 {
13213         int err, len, var_idx, i;
13214         const char *var_name;
13215         const struct bpf_map *map;
13216         struct btf *btf;
13217         __u32 map_type_id;
13218         const struct btf_type *map_type, *var_type;
13219         const struct bpf_var_skeleton *var_skel;
13220         struct btf_var_secinfo *var;
13221
13222         if (!s->obj)
13223                 return libbpf_err(-EINVAL);
13224
13225         btf = bpf_object__btf(s->obj);
13226         if (!btf) {
13227                 pr_warn("subskeletons require BTF at runtime (object %s)\n",
13228                         bpf_object__name(s->obj));
13229                 return libbpf_err(-errno);
13230         }
13231
13232         err = populate_skeleton_maps(s->obj, s->maps, s->map_cnt);
13233         if (err) {
13234                 pr_warn("failed to populate subskeleton maps: %d\n", err);
13235                 return libbpf_err(err);
13236         }
13237
13238         err = populate_skeleton_progs(s->obj, s->progs, s->prog_cnt);
13239         if (err) {
13240                 pr_warn("failed to populate subskeleton maps: %d\n", err);
13241                 return libbpf_err(err);
13242         }
13243
13244         for (var_idx = 0; var_idx < s->var_cnt; var_idx++) {
13245                 var_skel = &s->vars[var_idx];
13246                 map = *var_skel->map;
13247                 map_type_id = bpf_map__btf_value_type_id(map);
13248                 map_type = btf__type_by_id(btf, map_type_id);
13249
13250                 if (!btf_is_datasec(map_type)) {
13251                         pr_warn("type for map '%1$s' is not a datasec: %2$s",
13252                                 bpf_map__name(map),
13253                                 __btf_kind_str(btf_kind(map_type)));
13254                         return libbpf_err(-EINVAL);
13255                 }
13256
13257                 len = btf_vlen(map_type);
13258                 var = btf_var_secinfos(map_type);
13259                 for (i = 0; i < len; i++, var++) {
13260                         var_type = btf__type_by_id(btf, var->type);
13261                         var_name = btf__name_by_offset(btf, var_type->name_off);
13262                         if (strcmp(var_name, var_skel->name) == 0) {
13263                                 *var_skel->addr = map->mmaped + var->offset;
13264                                 break;
13265                         }
13266                 }
13267         }
13268         return 0;
13269 }
13270
13271 void bpf_object__destroy_subskeleton(struct bpf_object_subskeleton *s)
13272 {
13273         if (!s)
13274                 return;
13275         free(s->maps);
13276         free(s->progs);
13277         free(s->vars);
13278         free(s);
13279 }
13280
13281 int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
13282 {
13283         int i, err;
13284
13285         err = bpf_object__load(*s->obj);
13286         if (err) {
13287                 pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err);
13288                 return libbpf_err(err);
13289         }
13290
13291         for (i = 0; i < s->map_cnt; i++) {
13292                 struct bpf_map *map = *s->maps[i].map;
13293                 size_t mmap_sz = bpf_map_mmap_sz(map->def.value_size, map->def.max_entries);
13294                 int prot, map_fd = bpf_map__fd(map);
13295                 void **mmaped = s->maps[i].mmaped;
13296
13297                 if (!mmaped)
13298                         continue;
13299
13300                 if (!(map->def.map_flags & BPF_F_MMAPABLE)) {
13301                         *mmaped = NULL;
13302                         continue;
13303                 }
13304
13305                 if (map->def.map_flags & BPF_F_RDONLY_PROG)
13306                         prot = PROT_READ;
13307                 else
13308                         prot = PROT_READ | PROT_WRITE;
13309
13310                 /* Remap anonymous mmap()-ed "map initialization image" as
13311                  * a BPF map-backed mmap()-ed memory, but preserving the same
13312                  * memory address. This will cause kernel to change process'
13313                  * page table to point to a different piece of kernel memory,
13314                  * but from userspace point of view memory address (and its
13315                  * contents, being identical at this point) will stay the
13316                  * same. This mapping will be released by bpf_object__close()
13317                  * as per normal clean up procedure, so we don't need to worry
13318                  * about it from skeleton's clean up perspective.
13319                  */
13320                 *mmaped = mmap(map->mmaped, mmap_sz, prot, MAP_SHARED | MAP_FIXED, map_fd, 0);
13321                 if (*mmaped == MAP_FAILED) {
13322                         err = -errno;
13323                         *mmaped = NULL;
13324                         pr_warn("failed to re-mmap() map '%s': %d\n",
13325                                  bpf_map__name(map), err);
13326                         return libbpf_err(err);
13327                 }
13328         }
13329
13330         return 0;
13331 }
13332
13333 int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
13334 {
13335         int i, err;
13336
13337         for (i = 0; i < s->prog_cnt; i++) {
13338                 struct bpf_program *prog = *s->progs[i].prog;
13339                 struct bpf_link **link = s->progs[i].link;
13340
13341                 if (!prog->autoload || !prog->autoattach)
13342                         continue;
13343
13344                 /* auto-attaching not supported for this program */
13345                 if (!prog->sec_def || !prog->sec_def->prog_attach_fn)
13346                         continue;
13347
13348                 /* if user already set the link manually, don't attempt auto-attach */
13349                 if (*link)
13350                         continue;
13351
13352                 err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, link);
13353                 if (err) {
13354                         pr_warn("prog '%s': failed to auto-attach: %d\n",
13355                                 bpf_program__name(prog), err);
13356                         return libbpf_err(err);
13357                 }
13358
13359                 /* It's possible that for some SEC() definitions auto-attach
13360                  * is supported in some cases (e.g., if definition completely
13361                  * specifies target information), but is not in other cases.
13362                  * SEC("uprobe") is one such case. If user specified target
13363                  * binary and function name, such BPF program can be
13364                  * auto-attached. But if not, it shouldn't trigger skeleton's
13365                  * attach to fail. It should just be skipped.
13366                  * attach_fn signals such case with returning 0 (no error) and
13367                  * setting link to NULL.
13368                  */
13369         }
13370
13371         return 0;
13372 }
13373
13374 void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
13375 {
13376         int i;
13377
13378         for (i = 0; i < s->prog_cnt; i++) {
13379                 struct bpf_link **link = s->progs[i].link;
13380
13381                 bpf_link__destroy(*link);
13382                 *link = NULL;
13383         }
13384 }
13385
13386 void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
13387 {
13388         if (!s)
13389                 return;
13390
13391         if (s->progs)
13392                 bpf_object__detach_skeleton(s);
13393         if (s->obj)
13394                 bpf_object__close(*s->obj);
13395         free(s->maps);
13396         free(s->progs);
13397         free(s);
13398 }