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 <string.h>
  22 #include <unistd.h>
  23 #include <fcntl.h>
  24 #include <errno.h>
  25 #include <asm/unistd.h>
  26 #include <linux/err.h>
  27 #include <linux/kernel.h>
  28 #include <linux/bpf.h>
  29 #include <linux/btf.h>
  30 #include <linux/filter.h>
  31 #include <linux/list.h>
  32 #include <linux/limits.h>
  33 #include <linux/perf_event.h>
  34 #include <linux/ring_buffer.h>
  35 #include <sys/stat.h>
  36 #include <sys/types.h>
  37 #include <sys/vfs.h>
  38 #include <tools/libc_compat.h>
  39 #include <libelf.h>
  40 #include <gelf.h>
  41
  42 #include "libbpf.h"
  43 #include "bpf.h"
  44 #include "btf.h"
  45 #include "str_error.h"
  46 #include "libbpf_internal.h"
  47
  48 #ifndef EM_BPF
  49 #define EM_BPF 247
  50 #endif
  51
  52 #ifndef BPF_FS_MAGIC
  53 #define BPF_FS_MAGIC            0xcafe4a11
  54 #endif
  55
  56 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
  57  * compilation if user enables corresponding warning. Disable it explicitly.
  58  */
  59 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
  60
  61 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  62
  63 static int __base_pr(enum libbpf_print_level level, const char *format,
  64                      va_list args)
  65 {
  66         if (level == LIBBPF_DEBUG)
  67                 return 0;
  68
  69         return vfprintf(stderr, format, args);
  70 }
  71
  72 static libbpf_print_fn_t __libbpf_pr = __base_pr;
  73
  74 void libbpf_set_print(libbpf_print_fn_t fn)
  75 {
  76         __libbpf_pr = fn;
  77 }
  78
  79 __printf(2, 3)
  80 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
  81 {
  82         va_list args;
  83
  84         if (!__libbpf_pr)
  85                 return;
  86
  87         va_start(args, format);
  88         __libbpf_pr(level, format, args);
  89         va_end(args);
  90 }
  91
  92 #define STRERR_BUFSIZE  128
  93
  94 #define CHECK_ERR(action, err, out) do {        \
  95         err = action;                   \
  96         if (err)                        \
  97                 goto out;               \
  98 } while(0)
  99
 100
 101 /* Copied from tools/perf/util/util.h */
 102 #ifndef zfree
 103 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 104 #endif
 105
 106 #ifndef zclose
 107 # define zclose(fd) ({                  \
 108         int ___err = 0;                 \
 109         if ((fd) >= 0)                  \
 110                 ___err = close((fd));   \
 111         fd = -1;                        \
 112         ___err; })
 113 #endif
 114
 115 #ifdef HAVE_LIBELF_MMAP_SUPPORT
 116 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
 117 #else
 118 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 119 #endif
 120
 121 static inline __u64 ptr_to_u64(const void *ptr)
 122 {
 123         return (__u64) (unsigned long) ptr;
 124 }
 125
 126 struct bpf_capabilities {
 127         /* v4.14: kernel support for program & map names. */
 128         __u32 name:1;
 129         /* v5.2: kernel support for global data sections. */
 130         __u32 global_data:1;
 131         /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
 132         __u32 btf_func:1;
 133         /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
 134         __u32 btf_datasec:1;
 135 };
 136
 137 /*
 138  * bpf_prog should be a better name but it has been used in
 139  * linux/filter.h.
 140  */
 141 struct bpf_program {
 142         /* Index in elf obj file, for relocation use. */
 143         int idx;
 144         char *name;
 145         int prog_ifindex;
 146         char *section_name;
 147         /* section_name with / replaced by _; makes recursive pinning
 148          * in bpf_object__pin_programs easier
 149          */
 150         char *pin_name;
 151         struct bpf_insn *insns;
 152         size_t insns_cnt, main_prog_cnt;
 153         enum bpf_prog_type type;
 154
 155         struct reloc_desc {
 156                 enum {
 157                         RELO_LD64,
 158                         RELO_CALL,
 159                         RELO_DATA,
 160                 } type;
 161                 int insn_idx;
 162                 union {
 163                         int map_idx;
 164                         int text_off;
 165                 };
 166         } *reloc_desc;
 167         int nr_reloc;
 168         int log_level;
 169
 170         struct {
 171                 int nr;
 172                 int *fds;
 173         } instances;
 174         bpf_program_prep_t preprocessor;
 175
 176         struct bpf_object *obj;
 177         void *priv;
 178         bpf_program_clear_priv_t clear_priv;
 179
 180         enum bpf_attach_type expected_attach_type;
 181         int btf_fd;
 182         void *func_info;
 183         __u32 func_info_rec_size;
 184         __u32 func_info_cnt;
 185
 186         struct bpf_capabilities *caps;
 187
 188         void *line_info;
 189         __u32 line_info_rec_size;
 190         __u32 line_info_cnt;
 191         __u32 prog_flags;
 192 };
 193
 194 enum libbpf_map_type {
 195         LIBBPF_MAP_UNSPEC,
 196         LIBBPF_MAP_DATA,
 197         LIBBPF_MAP_BSS,
 198         LIBBPF_MAP_RODATA,
 199 };
 200
 201 static const char * const libbpf_type_to_btf_name[] = {
 202         [LIBBPF_MAP_DATA]       = ".data",
 203         [LIBBPF_MAP_BSS]        = ".bss",
 204         [LIBBPF_MAP_RODATA]     = ".rodata",
 205 };
 206
 207 struct bpf_map {
 208         int fd;
 209         char *name;
 210         int sec_idx;
 211         size_t sec_offset;
 212         int map_ifindex;
 213         int inner_map_fd;
 214         struct bpf_map_def def;
 215         __u32 btf_key_type_id;
 216         __u32 btf_value_type_id;
 217         void *priv;
 218         bpf_map_clear_priv_t clear_priv;
 219         enum libbpf_map_type libbpf_type;
 220 };
 221
 222 struct bpf_secdata {
 223         void *rodata;
 224         void *data;
 225 };
 226
 227 static LIST_HEAD(bpf_objects_list);
 228
 229 struct bpf_object {
 230         char name[BPF_OBJ_NAME_LEN];
 231         char license[64];
 232         __u32 kern_version;
 233
 234         struct bpf_program *programs;
 235         size_t nr_programs;
 236         struct bpf_map *maps;
 237         size_t nr_maps;
 238         size_t maps_cap;
 239         struct bpf_secdata sections;
 240
 241         bool loaded;
 242         bool has_pseudo_calls;
 243
 244         /*
 245          * Information when doing elf related work. Only valid if fd
 246          * is valid.
 247          */
 248         struct {
 249                 int fd;
 250                 void *obj_buf;
 251                 size_t obj_buf_sz;
 252                 Elf *elf;
 253                 GElf_Ehdr ehdr;
 254                 Elf_Data *symbols;
 255                 Elf_Data *data;
 256                 Elf_Data *rodata;
 257                 Elf_Data *bss;
 258                 size_t strtabidx;
 259                 struct {
 260                         GElf_Shdr shdr;
 261                         Elf_Data *data;
 262                 } *reloc;
 263                 int nr_reloc;
 264                 int maps_shndx;
 265                 int btf_maps_shndx;
 266                 int text_shndx;
 267                 int data_shndx;
 268                 int rodata_shndx;
 269                 int bss_shndx;
 270         } efile;
 271         /*
 272          * All loaded bpf_object is linked in a list, which is
 273          * hidden to caller. bpf_objects__<func> handlers deal with
 274          * all objects.
 275          */
 276         struct list_head list;
 277
 278         struct btf *btf;
 279         struct btf_ext *btf_ext;
 280
 281         void *priv;
 282         bpf_object_clear_priv_t clear_priv;
 283
 284         struct bpf_capabilities caps;
 285
 286         char path[];
 287 };
 288 #define obj_elf_valid(o)        ((o)->efile.elf)
 289
 290 void bpf_program__unload(struct bpf_program *prog)
 291 {
 292         int i;
 293
 294         if (!prog)
 295                 return;
 296
 297         /*
 298          * If the object is opened but the program was never loaded,
 299          * it is possible that prog->instances.nr == -1.
 300          */
 301         if (prog->instances.nr > 0) {
 302                 for (i = 0; i < prog->instances.nr; i++)
 303                         zclose(prog->instances.fds[i]);
 304         } else if (prog->instances.nr != -1) {
 305                 pr_warning("Internal error: instances.nr is %d\n",
 306                            prog->instances.nr);
 307         }
 308
 309         prog->instances.nr = -1;
 310         zfree(&prog->instances.fds);
 311
 312         zclose(prog->btf_fd);
 313         zfree(&prog->func_info);
 314         zfree(&prog->line_info);
 315 }
 316
 317 static void bpf_program__exit(struct bpf_program *prog)
 318 {
 319         if (!prog)
 320                 return;
 321
 322         if (prog->clear_priv)
 323                 prog->clear_priv(prog, prog->priv);
 324
 325         prog->priv = NULL;
 326         prog->clear_priv = NULL;
 327
 328         bpf_program__unload(prog);
 329         zfree(&prog->name);
 330         zfree(&prog->section_name);
 331         zfree(&prog->pin_name);
 332         zfree(&prog->insns);
 333         zfree(&prog->reloc_desc);
 334
 335         prog->nr_reloc = 0;
 336         prog->insns_cnt = 0;
 337         prog->idx = -1;
 338 }
 339
 340 static char *__bpf_program__pin_name(struct bpf_program *prog)
 341 {
 342         char *name, *p;
 343
 344         name = p = strdup(prog->section_name);
 345         while ((p = strchr(p, '/')))
 346                 *p = '_';
 347
 348         return name;
 349 }
 350
 351 static int
 352 bpf_program__init(void *data, size_t size, char *section_name, int idx,
 353                   struct bpf_program *prog)
 354 {
 355         const size_t bpf_insn_sz = sizeof(struct bpf_insn);
 356
 357         if (size == 0 || size % bpf_insn_sz) {
 358                 pr_warning("corrupted section '%s', size: %zu\n",
 359                            section_name, size);
 360                 return -EINVAL;
 361         }
 362
 363         memset(prog, 0, sizeof(*prog));
 364
 365         prog->section_name = strdup(section_name);
 366         if (!prog->section_name) {
 367                 pr_warning("failed to alloc name for prog under section(%d) %s\n",
 368                            idx, section_name);
 369                 goto errout;
 370         }
 371
 372         prog->pin_name = __bpf_program__pin_name(prog);
 373         if (!prog->pin_name) {
 374                 pr_warning("failed to alloc pin name for prog under section(%d) %s\n",
 375                            idx, section_name);
 376                 goto errout;
 377         }
 378
 379         prog->insns = malloc(size);
 380         if (!prog->insns) {
 381                 pr_warning("failed to alloc insns for prog under section %s\n",
 382                            section_name);
 383                 goto errout;
 384         }
 385         prog->insns_cnt = size / bpf_insn_sz;
 386         memcpy(prog->insns, data, size);
 387         prog->idx = idx;
 388         prog->instances.fds = NULL;
 389         prog->instances.nr = -1;
 390         prog->type = BPF_PROG_TYPE_UNSPEC;
 391         prog->btf_fd = -1;
 392
 393         return 0;
 394 errout:
 395         bpf_program__exit(prog);
 396         return -ENOMEM;
 397 }
 398
 399 static int
 400 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 401                         char *section_name, int idx)
 402 {
 403         struct bpf_program prog, *progs;
 404         int nr_progs, err;
 405
 406         err = bpf_program__init(data, size, section_name, idx, &prog);
 407         if (err)
 408                 return err;
 409
 410         prog.caps = &obj->caps;
 411         progs = obj->programs;
 412         nr_progs = obj->nr_programs;
 413
 414         progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
 415         if (!progs) {
 416                 /*
 417                  * In this case the original obj->programs
 418                  * is still valid, so don't need special treat for
 419                  * bpf_close_object().
 420                  */
 421                 pr_warning("failed to alloc a new program under section '%s'\n",
 422                            section_name);
 423                 bpf_program__exit(&prog);
 424                 return -ENOMEM;
 425         }
 426
 427         pr_debug("found program %s\n", prog.section_name);
 428         obj->programs = progs;
 429         obj->nr_programs = nr_progs + 1;
 430         prog.obj = obj;
 431         progs[nr_progs] = prog;
 432         return 0;
 433 }
 434
 435 static int
 436 bpf_object__init_prog_names(struct bpf_object *obj)
 437 {
 438         Elf_Data *symbols = obj->efile.symbols;
 439         struct bpf_program *prog;
 440         size_t pi, si;
 441
 442         for (pi = 0; pi < obj->nr_programs; pi++) {
 443                 const char *name = NULL;
 444
 445                 prog = &obj->programs[pi];
 446
 447                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 448                      si++) {
 449                         GElf_Sym sym;
 450
 451                         if (!gelf_getsym(symbols, si, &sym))
 452                                 continue;
 453                         if (sym.st_shndx != prog->idx)
 454                                 continue;
 455                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 456                                 continue;
 457
 458                         name = elf_strptr(obj->efile.elf,
 459                                           obj->efile.strtabidx,
 460                                           sym.st_name);
 461                         if (!name) {
 462                                 pr_warning("failed to get sym name string for prog %s\n",
 463                                            prog->section_name);
 464                                 return -LIBBPF_ERRNO__LIBELF;
 465                         }
 466                 }
 467
 468                 if (!name && prog->idx == obj->efile.text_shndx)
 469                         name = ".text";
 470
 471                 if (!name) {
 472                         pr_warning("failed to find sym for prog %s\n",
 473                                    prog->section_name);
 474                         return -EINVAL;
 475                 }
 476
 477                 prog->name = strdup(name);
 478                 if (!prog->name) {
 479                         pr_warning("failed to allocate memory for prog sym %s\n",
 480                                    name);
 481                         return -ENOMEM;
 482                 }
 483         }
 484
 485         return 0;
 486 }
 487
 488 static struct bpf_object *bpf_object__new(const char *path,
 489                                           void *obj_buf,
 490                                           size_t obj_buf_sz)
 491 {
 492         struct bpf_object *obj;
 493         char *end;
 494
 495         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
 496         if (!obj) {
 497                 pr_warning("alloc memory failed for %s\n", path);
 498                 return ERR_PTR(-ENOMEM);
 499         }
 500
 501         strcpy(obj->path, path);
 502         /* Using basename() GNU version which doesn't modify arg. */
 503         strncpy(obj->name, basename((void *)path), sizeof(obj->name) - 1);
 504         end = strchr(obj->name, '.');
 505         if (end)
 506                 *end = 0;
 507
 508         obj->efile.fd = -1;
 509         /*
 510          * Caller of this function should also call
 511          * bpf_object__elf_finish() after data collection to return
 512          * obj_buf to user. If not, we should duplicate the buffer to
 513          * avoid user freeing them before elf finish.
 514          */
 515         obj->efile.obj_buf = obj_buf;
 516         obj->efile.obj_buf_sz = obj_buf_sz;
 517         obj->efile.maps_shndx = -1;
 518         obj->efile.btf_maps_shndx = -1;
 519         obj->efile.data_shndx = -1;
 520         obj->efile.rodata_shndx = -1;
 521         obj->efile.bss_shndx = -1;
 522
 523         obj->loaded = false;
 524
 525         INIT_LIST_HEAD(&obj->list);
 526         list_add(&obj->list, &bpf_objects_list);
 527         return obj;
 528 }
 529
 530 static void bpf_object__elf_finish(struct bpf_object *obj)
 531 {
 532         if (!obj_elf_valid(obj))
 533                 return;
 534
 535         if (obj->efile.elf) {
 536                 elf_end(obj->efile.elf);
 537                 obj->efile.elf = NULL;
 538         }
 539         obj->efile.symbols = NULL;
 540         obj->efile.data = NULL;
 541         obj->efile.rodata = NULL;
 542         obj->efile.bss = NULL;
 543
 544         zfree(&obj->efile.reloc);
 545         obj->efile.nr_reloc = 0;
 546         zclose(obj->efile.fd);
 547         obj->efile.obj_buf = NULL;
 548         obj->efile.obj_buf_sz = 0;
 549 }
 550
 551 static int bpf_object__elf_init(struct bpf_object *obj)
 552 {
 553         int err = 0;
 554         GElf_Ehdr *ep;
 555
 556         if (obj_elf_valid(obj)) {
 557                 pr_warning("elf init: internal error\n");
 558                 return -LIBBPF_ERRNO__LIBELF;
 559         }
 560
 561         if (obj->efile.obj_buf_sz > 0) {
 562                 /*
 563                  * obj_buf should have been validated by
 564                  * bpf_object__open_buffer().
 565                  */
 566                 obj->efile.elf = elf_memory(obj->efile.obj_buf,
 567                                             obj->efile.obj_buf_sz);
 568         } else {
 569                 obj->efile.fd = open(obj->path, O_RDONLY);
 570                 if (obj->efile.fd < 0) {
 571                         char errmsg[STRERR_BUFSIZE], *cp;
 572
 573                         err = -errno;
 574                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
 575                         pr_warning("failed to open %s: %s\n", obj->path, cp);
 576                         return err;
 577                 }
 578
 579                 obj->efile.elf = elf_begin(obj->efile.fd,
 580                                            LIBBPF_ELF_C_READ_MMAP, NULL);
 581         }
 582
 583         if (!obj->efile.elf) {
 584                 pr_warning("failed to open %s as ELF file\n", obj->path);
 585                 err = -LIBBPF_ERRNO__LIBELF;
 586                 goto errout;
 587         }
 588
 589         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
 590                 pr_warning("failed to get EHDR from %s\n", obj->path);
 591                 err = -LIBBPF_ERRNO__FORMAT;
 592                 goto errout;
 593         }
 594         ep = &obj->efile.ehdr;
 595
 596         /* Old LLVM set e_machine to EM_NONE */
 597         if (ep->e_type != ET_REL ||
 598             (ep->e_machine && ep->e_machine != EM_BPF)) {
 599                 pr_warning("%s is not an eBPF object file\n", obj->path);
 600                 err = -LIBBPF_ERRNO__FORMAT;
 601                 goto errout;
 602         }
 603
 604         return 0;
 605 errout:
 606         bpf_object__elf_finish(obj);
 607         return err;
 608 }
 609
 610 static int bpf_object__check_endianness(struct bpf_object *obj)
 611 {
 612 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 613         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
 614                 return 0;
 615 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 616         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
 617                 return 0;
 618 #else
 619 # error "Unrecognized __BYTE_ORDER__"
 620 #endif
 621         pr_warning("endianness mismatch.\n");
 622         return -LIBBPF_ERRNO__ENDIAN;
 623 }
 624
 625 static int
 626 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
 627 {
 628         memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
 629         pr_debug("license of %s is %s\n", obj->path, obj->license);
 630         return 0;
 631 }
 632
 633 static int
 634 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
 635 {
 636         __u32 kver;
 637
 638         if (size != sizeof(kver)) {
 639                 pr_warning("invalid kver section in %s\n", obj->path);
 640                 return -LIBBPF_ERRNO__FORMAT;
 641         }
 642         memcpy(&kver, data, sizeof(kver));
 643         obj->kern_version = kver;
 644         pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
 645         return 0;
 646 }
 647
 648 static int compare_bpf_map(const void *_a, const void *_b)
 649 {
 650         const struct bpf_map *a = _a;
 651         const struct bpf_map *b = _b;
 652
 653         if (a->sec_idx != b->sec_idx)
 654                 return a->sec_idx - b->sec_idx;
 655         return a->sec_offset - b->sec_offset;
 656 }
 657
 658 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
 659 {
 660         if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
 661             type == BPF_MAP_TYPE_HASH_OF_MAPS)
 662                 return true;
 663         return false;
 664 }
 665
 666 static int bpf_object_search_section_size(const struct bpf_object *obj,
 667                                           const char *name, size_t *d_size)
 668 {
 669         const GElf_Ehdr *ep = &obj->efile.ehdr;
 670         Elf *elf = obj->efile.elf;
 671         Elf_Scn *scn = NULL;
 672         int idx = 0;
 673
 674         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 675                 const char *sec_name;
 676                 Elf_Data *data;
 677                 GElf_Shdr sh;
 678
 679                 idx++;
 680                 if (gelf_getshdr(scn, &sh) != &sh) {
 681                         pr_warning("failed to get section(%d) header from %s\n",
 682                                    idx, obj->path);
 683                         return -EIO;
 684                 }
 685
 686                 sec_name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 687                 if (!sec_name) {
 688                         pr_warning("failed to get section(%d) name from %s\n",
 689                                    idx, obj->path);
 690                         return -EIO;
 691                 }
 692
 693                 if (strcmp(name, sec_name))
 694                         continue;
 695
 696                 data = elf_getdata(scn, 0);
 697                 if (!data) {
 698                         pr_warning("failed to get section(%d) data from %s(%s)\n",
 699                                    idx, name, obj->path);
 700                         return -EIO;
 701                 }
 702
 703                 *d_size = data->d_size;
 704                 return 0;
 705         }
 706
 707         return -ENOENT;
 708 }
 709
 710 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
 711                              __u32 *size)
 712 {
 713         int ret = -ENOENT;
 714         size_t d_size;
 715
 716         *size = 0;
 717         if (!name) {
 718                 return -EINVAL;
 719         } else if (!strcmp(name, ".data")) {
 720                 if (obj->efile.data)
 721                         *size = obj->efile.data->d_size;
 722         } else if (!strcmp(name, ".bss")) {
 723                 if (obj->efile.bss)
 724                         *size = obj->efile.bss->d_size;
 725         } else if (!strcmp(name, ".rodata")) {
 726                 if (obj->efile.rodata)
 727                         *size = obj->efile.rodata->d_size;
 728         } else {
 729                 ret = bpf_object_search_section_size(obj, name, &d_size);
 730                 if (!ret)
 731                         *size = d_size;
 732         }
 733
 734         return *size ? 0 : ret;
 735 }
 736
 737 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
 738                                 __u32 *off)
 739 {
 740         Elf_Data *symbols = obj->efile.symbols;
 741         const char *sname;
 742         size_t si;
 743
 744         if (!name || !off)
 745                 return -EINVAL;
 746
 747         for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
 748                 GElf_Sym sym;
 749
 750                 if (!gelf_getsym(symbols, si, &sym))
 751                         continue;
 752                 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
 753                     GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
 754                         continue;
 755
 756                 sname = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
 757                                    sym.st_name);
 758                 if (!sname) {
 759                         pr_warning("failed to get sym name string for var %s\n",
 760                                    name);
 761                         return -EIO;
 762                 }
 763                 if (strcmp(name, sname) == 0) {
 764                         *off = sym.st_value;
 765                         return 0;
 766                 }
 767         }
 768
 769         return -ENOENT;
 770 }
 771
 772 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
 773 {
 774         struct bpf_map *new_maps;
 775         size_t new_cap;
 776         int i;
 777
 778         if (obj->nr_maps < obj->maps_cap)
 779                 return &obj->maps[obj->nr_maps++];
 780
 781         new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
 782         new_maps = realloc(obj->maps, new_cap * sizeof(*obj->maps));
 783         if (!new_maps) {
 784                 pr_warning("alloc maps for object failed\n");
 785                 return ERR_PTR(-ENOMEM);
 786         }
 787
 788         obj->maps_cap = new_cap;
 789         obj->maps = new_maps;
 790
 791         /* zero out new maps */
 792         memset(obj->maps + obj->nr_maps, 0,
 793                (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
 794         /*
 795          * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
 796          * when failure (zclose won't close negative fd)).
 797          */
 798         for (i = obj->nr_maps; i < obj->maps_cap; i++) {
 799                 obj->maps[i].fd = -1;
 800                 obj->maps[i].inner_map_fd = -1;
 801         }
 802
 803         return &obj->maps[obj->nr_maps++];
 804 }
 805
 806 static int
 807 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
 808                               int sec_idx, Elf_Data *data, void **data_buff)
 809 {
 810         char map_name[BPF_OBJ_NAME_LEN];
 811         struct bpf_map_def *def;
 812         struct bpf_map *map;
 813
 814         map = bpf_object__add_map(obj);
 815         if (IS_ERR(map))
 816                 return PTR_ERR(map);
 817
 818         map->libbpf_type = type;
 819         map->sec_idx = sec_idx;
 820         map->sec_offset = 0;
 821         snprintf(map_name, sizeof(map_name), "%.8s%.7s", obj->name,
 822                  libbpf_type_to_btf_name[type]);
 823         map->name = strdup(map_name);
 824         if (!map->name) {
 825                 pr_warning("failed to alloc map name\n");
 826                 return -ENOMEM;
 827         }
 828         pr_debug("map '%s' (global data): at sec_idx %d, offset %zu.\n",
 829                  map_name, map->sec_idx, map->sec_offset);
 830
 831         def = &map->def;
 832         def->type = BPF_MAP_TYPE_ARRAY;
 833         def->key_size = sizeof(int);
 834         def->value_size = data->d_size;
 835         def->max_entries = 1;
 836         def->map_flags = type == LIBBPF_MAP_RODATA ? BPF_F_RDONLY_PROG : 0;
 837         if (data_buff) {
 838                 *data_buff = malloc(data->d_size);
 839                 if (!*data_buff) {
 840                         zfree(&map->name);
 841                         pr_warning("failed to alloc map content buffer\n");
 842                         return -ENOMEM;
 843                 }
 844                 memcpy(*data_buff, data->d_buf, data->d_size);
 845         }
 846
 847         pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
 848         return 0;
 849 }
 850
 851 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
 852 {
 853         int err;
 854
 855         if (!obj->caps.global_data)
 856                 return 0;
 857         /*
 858          * Populate obj->maps with libbpf internal maps.
 859          */
 860         if (obj->efile.data_shndx >= 0) {
 861                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
 862                                                     obj->efile.data_shndx,
 863                                                     obj->efile.data,
 864                                                     &obj->sections.data);
 865                 if (err)
 866                         return err;
 867         }
 868         if (obj->efile.rodata_shndx >= 0) {
 869                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
 870                                                     obj->efile.rodata_shndx,
 871                                                     obj->efile.rodata,
 872                                                     &obj->sections.rodata);
 873                 if (err)
 874                         return err;
 875         }
 876         if (obj->efile.bss_shndx >= 0) {
 877                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
 878                                                     obj->efile.bss_shndx,
 879                                                     obj->efile.bss, NULL);
 880                 if (err)
 881                         return err;
 882         }
 883         return 0;
 884 }
 885
 886 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
 887 {
 888         Elf_Data *symbols = obj->efile.symbols;
 889         int i, map_def_sz = 0, nr_maps = 0, nr_syms;
 890         Elf_Data *data = NULL;
 891         Elf_Scn *scn;
 892
 893         if (obj->efile.maps_shndx < 0)
 894                 return 0;
 895
 896         if (!symbols)
 897                 return -EINVAL;
 898
 899         scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
 900         if (scn)
 901                 data = elf_getdata(scn, NULL);
 902         if (!scn || !data) {
 903                 pr_warning("failed to get Elf_Data from map section %d\n",
 904                            obj->efile.maps_shndx);
 905                 return -EINVAL;
 906         }
 907
 908         /*
 909          * Count number of maps. Each map has a name.
 910          * Array of maps is not supported: only the first element is
 911          * considered.
 912          *
 913          * TODO: Detect array of map and report error.
 914          */
 915         nr_syms = symbols->d_size / sizeof(GElf_Sym);
 916         for (i = 0; i < nr_syms; i++) {
 917                 GElf_Sym sym;
 918
 919                 if (!gelf_getsym(symbols, i, &sym))
 920                         continue;
 921                 if (sym.st_shndx != obj->efile.maps_shndx)
 922                         continue;
 923                 nr_maps++;
 924         }
 925         /* Assume equally sized map definitions */
 926         pr_debug("maps in %s: %d maps in %zd bytes\n",
 927                  obj->path, nr_maps, data->d_size);
 928
 929         map_def_sz = data->d_size / nr_maps;
 930         if (!data->d_size || (data->d_size % nr_maps) != 0) {
 931                 pr_warning("unable to determine map definition size "
 932                            "section %s, %d maps in %zd bytes\n",
 933                            obj->path, nr_maps, data->d_size);
 934                 return -EINVAL;
 935         }
 936
 937         /* Fill obj->maps using data in "maps" section.  */
 938         for (i = 0; i < nr_syms; i++) {
 939                 GElf_Sym sym;
 940                 const char *map_name;
 941                 struct bpf_map_def *def;
 942                 struct bpf_map *map;
 943
 944                 if (!gelf_getsym(symbols, i, &sym))
 945                         continue;
 946                 if (sym.st_shndx != obj->efile.maps_shndx)
 947                         continue;
 948
 949                 map = bpf_object__add_map(obj);
 950                 if (IS_ERR(map))
 951                         return PTR_ERR(map);
 952
 953                 map_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
 954                                       sym.st_name);
 955                 if (!map_name) {
 956                         pr_warning("failed to get map #%d name sym string for obj %s\n",
 957                                    i, obj->path);
 958                         return -LIBBPF_ERRNO__FORMAT;
 959                 }
 960
 961                 map->libbpf_type = LIBBPF_MAP_UNSPEC;
 962                 map->sec_idx = sym.st_shndx;
 963                 map->sec_offset = sym.st_value;
 964                 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
 965                          map_name, map->sec_idx, map->sec_offset);
 966                 if (sym.st_value + map_def_sz > data->d_size) {
 967                         pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
 968                                    obj->path, map_name);
 969                         return -EINVAL;
 970                 }
 971
 972                 map->name = strdup(map_name);
 973                 if (!map->name) {
 974                         pr_warning("failed to alloc map name\n");
 975                         return -ENOMEM;
 976                 }
 977                 pr_debug("map %d is \"%s\"\n", i, map->name);
 978                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
 979                 /*
 980                  * If the definition of the map in the object file fits in
 981                  * bpf_map_def, copy it.  Any extra fields in our version
 982                  * of bpf_map_def will default to zero as a result of the
 983                  * calloc above.
 984                  */
 985                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
 986                         memcpy(&map->def, def, map_def_sz);
 987                 } else {
 988                         /*
 989                          * Here the map structure being read is bigger than what
 990                          * we expect, truncate if the excess bits are all zero.
 991                          * If they are not zero, reject this map as
 992                          * incompatible.
 993                          */
 994                         char *b;
 995                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
 996                              b < ((char *)def) + map_def_sz; b++) {
 997                                 if (*b != 0) {
 998                                         pr_warning("maps section in %s: \"%s\" "
 999                                                    "has unrecognized, non-zero "
1000                                                    "options\n",
1001                                                    obj->path, map_name);
1002                                         if (strict)
1003                                                 return -EINVAL;
1004                                 }
1005                         }
1006                         memcpy(&map->def, def, sizeof(struct bpf_map_def));
1007                 }
1008         }
1009         return 0;
1010 }
1011
1012 static const struct btf_type *skip_mods_and_typedefs(const struct btf *btf,
1013                                                      __u32 id)
1014 {
1015         const struct btf_type *t = btf__type_by_id(btf, id);
1016
1017         while (true) {
1018                 switch (BTF_INFO_KIND(t->info)) {
1019                 case BTF_KIND_VOLATILE:
1020                 case BTF_KIND_CONST:
1021                 case BTF_KIND_RESTRICT:
1022                 case BTF_KIND_TYPEDEF:
1023                         t = btf__type_by_id(btf, t->type);
1024                         break;
1025                 default:
1026                         return t;
1027                 }
1028         }
1029 }
1030
1031 static bool get_map_field_int(const char *map_name,
1032                               const struct btf *btf,
1033                               const struct btf_type *def,
1034                               const struct btf_member *m,
1035                               const void *data, __u32 *res) {
1036         const struct btf_type *t = skip_mods_and_typedefs(btf, m->type);
1037         const char *name = btf__name_by_offset(btf, m->name_off);
1038         __u32 int_info = *(const __u32 *)(const void *)(t + 1);
1039
1040         if (BTF_INFO_KIND(t->info) != BTF_KIND_INT) {
1041                 pr_warning("map '%s': attr '%s': expected INT, got %u.\n",
1042                            map_name, name, BTF_INFO_KIND(t->info));
1043                 return false;
1044         }
1045         if (t->size != 4 || BTF_INT_BITS(int_info) != 32 ||
1046             BTF_INT_OFFSET(int_info)) {
1047                 pr_warning("map '%s': attr '%s': expected 32-bit non-bitfield integer, "
1048                            "got %u-byte (%d-bit) one with bit offset %d.\n",
1049                            map_name, name, t->size, BTF_INT_BITS(int_info),
1050                            BTF_INT_OFFSET(int_info));
1051                 return false;
1052         }
1053         if (BTF_INFO_KFLAG(def->info) && BTF_MEMBER_BITFIELD_SIZE(m->offset)) {
1054                 pr_warning("map '%s': attr '%s': bitfield is not supported.\n",
1055                            map_name, name);
1056                 return false;
1057         }
1058         if (m->offset % 32) {
1059                 pr_warning("map '%s': attr '%s': unaligned fields are not supported.\n",
1060                            map_name, name);
1061                 return false;
1062         }
1063
1064         *res = *(const __u32 *)(data + m->offset / 8);
1065         return true;
1066 }
1067
1068 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
1069                                          const struct btf_type *sec,
1070                                          int var_idx, int sec_idx,
1071                                          const Elf_Data *data, bool strict)
1072 {
1073         const struct btf_type *var, *def, *t;
1074         const struct btf_var_secinfo *vi;
1075         const struct btf_var *var_extra;
1076         const struct btf_member *m;
1077         const void *def_data;
1078         const char *map_name;
1079         struct bpf_map *map;
1080         int vlen, i;
1081
1082         vi = (const struct btf_var_secinfo *)(const void *)(sec + 1) + var_idx;
1083         var = btf__type_by_id(obj->btf, vi->type);
1084         var_extra = (const void *)(var + 1);
1085         map_name = btf__name_by_offset(obj->btf, var->name_off);
1086         vlen = BTF_INFO_VLEN(var->info);
1087
1088         if (map_name == NULL || map_name[0] == '\0') {
1089                 pr_warning("map #%d: empty name.\n", var_idx);
1090                 return -EINVAL;
1091         }
1092         if ((__u64)vi->offset + vi->size > data->d_size) {
1093                 pr_warning("map '%s' BTF data is corrupted.\n", map_name);
1094                 return -EINVAL;
1095         }
1096         if (BTF_INFO_KIND(var->info) != BTF_KIND_VAR) {
1097                 pr_warning("map '%s': unexpected var kind %u.\n",
1098                            map_name, BTF_INFO_KIND(var->info));
1099                 return -EINVAL;
1100         }
1101         if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED &&
1102             var_extra->linkage != BTF_VAR_STATIC) {
1103                 pr_warning("map '%s': unsupported var linkage %u.\n",
1104                            map_name, var_extra->linkage);
1105                 return -EOPNOTSUPP;
1106         }
1107
1108         def = skip_mods_and_typedefs(obj->btf, var->type);
1109         if (BTF_INFO_KIND(def->info) != BTF_KIND_STRUCT) {
1110                 pr_warning("map '%s': unexpected def kind %u.\n",
1111                            map_name, BTF_INFO_KIND(var->info));
1112                 return -EINVAL;
1113         }
1114         if (def->size > vi->size) {
1115                 pr_warning("map '%s': invalid def size.\n", map_name);
1116                 return -EINVAL;
1117         }
1118
1119         map = bpf_object__add_map(obj);
1120         if (IS_ERR(map))
1121                 return PTR_ERR(map);
1122         map->name = strdup(map_name);
1123         if (!map->name) {
1124                 pr_warning("map '%s': failed to alloc map name.\n", map_name);
1125                 return -ENOMEM;
1126         }
1127         map->libbpf_type = LIBBPF_MAP_UNSPEC;
1128         map->def.type = BPF_MAP_TYPE_UNSPEC;
1129         map->sec_idx = sec_idx;
1130         map->sec_offset = vi->offset;
1131         pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
1132                  map_name, map->sec_idx, map->sec_offset);
1133
1134         def_data = data->d_buf + vi->offset;
1135         vlen = BTF_INFO_VLEN(def->info);
1136         m = (const void *)(def + 1);
1137         for (i = 0; i < vlen; i++, m++) {
1138                 const char *name = btf__name_by_offset(obj->btf, m->name_off);
1139
1140                 if (!name) {
1141                         pr_warning("map '%s': invalid field #%d.\n",
1142                                    map_name, i);
1143                         return -EINVAL;
1144                 }
1145                 if (strcmp(name, "type") == 0) {
1146                         if (!get_map_field_int(map_name, obj->btf, def, m,
1147                                                def_data, &map->def.type))
1148                                 return -EINVAL;
1149                         pr_debug("map '%s': found type = %u.\n",
1150                                  map_name, map->def.type);
1151                 } else if (strcmp(name, "max_entries") == 0) {
1152                         if (!get_map_field_int(map_name, obj->btf, def, m,
1153                                                def_data, &map->def.max_entries))
1154                                 return -EINVAL;
1155                         pr_debug("map '%s': found max_entries = %u.\n",
1156                                  map_name, map->def.max_entries);
1157                 } else if (strcmp(name, "map_flags") == 0) {
1158                         if (!get_map_field_int(map_name, obj->btf, def, m,
1159                                                def_data, &map->def.map_flags))
1160                                 return -EINVAL;
1161                         pr_debug("map '%s': found map_flags = %u.\n",
1162                                  map_name, map->def.map_flags);
1163                 } else if (strcmp(name, "key_size") == 0) {
1164                         __u32 sz;
1165
1166                         if (!get_map_field_int(map_name, obj->btf, def, m,
1167                                                def_data, &sz))
1168                                 return -EINVAL;
1169                         pr_debug("map '%s': found key_size = %u.\n",
1170                                  map_name, sz);
1171                         if (map->def.key_size && map->def.key_size != sz) {
1172                                 pr_warning("map '%s': conflicting key size %u != %u.\n",
1173                                            map_name, map->def.key_size, sz);
1174                                 return -EINVAL;
1175                         }
1176                         map->def.key_size = sz;
1177                 } else if (strcmp(name, "key") == 0) {
1178                         __s64 sz;
1179
1180                         t = btf__type_by_id(obj->btf, m->type);
1181                         if (!t) {
1182                                 pr_warning("map '%s': key type [%d] not found.\n",
1183                                            map_name, m->type);
1184                                 return -EINVAL;
1185                         }
1186                         if (BTF_INFO_KIND(t->info) != BTF_KIND_PTR) {
1187                                 pr_warning("map '%s': key spec is not PTR: %u.\n",
1188                                            map_name, BTF_INFO_KIND(t->info));
1189                                 return -EINVAL;
1190                         }
1191                         sz = btf__resolve_size(obj->btf, t->type);
1192                         if (sz < 0) {
1193                                 pr_warning("map '%s': can't determine key size for type [%u]: %lld.\n",
1194                                            map_name, t->type, sz);
1195                                 return sz;
1196                         }
1197                         pr_debug("map '%s': found key [%u], sz = %lld.\n",
1198                                  map_name, t->type, sz);
1199                         if (map->def.key_size && map->def.key_size != sz) {
1200                                 pr_warning("map '%s': conflicting key size %u != %lld.\n",
1201                                            map_name, map->def.key_size, sz);
1202                                 return -EINVAL;
1203                         }
1204                         map->def.key_size = sz;
1205                         map->btf_key_type_id = t->type;
1206                 } else if (strcmp(name, "value_size") == 0) {
1207                         __u32 sz;
1208
1209                         if (!get_map_field_int(map_name, obj->btf, def, m,
1210                                                def_data, &sz))
1211                                 return -EINVAL;
1212                         pr_debug("map '%s': found value_size = %u.\n",
1213                                  map_name, sz);
1214                         if (map->def.value_size && map->def.value_size != sz) {
1215                                 pr_warning("map '%s': conflicting value size %u != %u.\n",
1216                                            map_name, map->def.value_size, sz);
1217                                 return -EINVAL;
1218                         }
1219                         map->def.value_size = sz;
1220                 } else if (strcmp(name, "value") == 0) {
1221                         __s64 sz;
1222
1223                         t = btf__type_by_id(obj->btf, m->type);
1224                         if (!t) {
1225                                 pr_warning("map '%s': value type [%d] not found.\n",
1226                                            map_name, m->type);
1227                                 return -EINVAL;
1228                         }
1229                         if (BTF_INFO_KIND(t->info) != BTF_KIND_PTR) {
1230                                 pr_warning("map '%s': value spec is not PTR: %u.\n",
1231                                            map_name, BTF_INFO_KIND(t->info));
1232                                 return -EINVAL;
1233                         }
1234                         sz = btf__resolve_size(obj->btf, t->type);
1235                         if (sz < 0) {
1236                                 pr_warning("map '%s': can't determine value size for type [%u]: %lld.\n",
1237                                            map_name, t->type, sz);
1238                                 return sz;
1239                         }
1240                         pr_debug("map '%s': found value [%u], sz = %lld.\n",
1241                                  map_name, t->type, sz);
1242                         if (map->def.value_size && map->def.value_size != sz) {
1243                                 pr_warning("map '%s': conflicting value size %u != %lld.\n",
1244                                            map_name, map->def.value_size, sz);
1245                                 return -EINVAL;
1246                         }
1247                         map->def.value_size = sz;
1248                         map->btf_value_type_id = t->type;
1249                 } else {
1250                         if (strict) {
1251                                 pr_warning("map '%s': unknown field '%s'.\n",
1252                                            map_name, name);
1253                                 return -ENOTSUP;
1254                         }
1255                         pr_debug("map '%s': ignoring unknown field '%s'.\n",
1256                                  map_name, name);
1257                 }
1258         }
1259
1260         if (map->def.type == BPF_MAP_TYPE_UNSPEC) {
1261                 pr_warning("map '%s': map type isn't specified.\n", map_name);
1262                 return -EINVAL;
1263         }
1264
1265         return 0;
1266 }
1267
1268 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict)
1269 {
1270         const struct btf_type *sec = NULL;
1271         int nr_types, i, vlen, err;
1272         const struct btf_type *t;
1273         const char *name;
1274         Elf_Data *data;
1275         Elf_Scn *scn;
1276
1277         if (obj->efile.btf_maps_shndx < 0)
1278                 return 0;
1279
1280         scn = elf_getscn(obj->efile.elf, obj->efile.btf_maps_shndx);
1281         if (scn)
1282                 data = elf_getdata(scn, NULL);
1283         if (!scn || !data) {
1284                 pr_warning("failed to get Elf_Data from map section %d (%s)\n",
1285                            obj->efile.maps_shndx, MAPS_ELF_SEC);
1286                 return -EINVAL;
1287         }
1288
1289         nr_types = btf__get_nr_types(obj->btf);
1290         for (i = 1; i <= nr_types; i++) {
1291                 t = btf__type_by_id(obj->btf, i);
1292                 if (BTF_INFO_KIND(t->info) != BTF_KIND_DATASEC)
1293                         continue;
1294                 name = btf__name_by_offset(obj->btf, t->name_off);
1295                 if (strcmp(name, MAPS_ELF_SEC) == 0) {
1296                         sec = t;
1297                         break;
1298                 }
1299         }
1300
1301         if (!sec) {
1302                 pr_warning("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
1303                 return -ENOENT;
1304         }
1305
1306         vlen = BTF_INFO_VLEN(sec->info);
1307         for (i = 0; i < vlen; i++) {
1308                 err = bpf_object__init_user_btf_map(obj, sec, i,
1309                                                     obj->efile.btf_maps_shndx,
1310                                                     data, strict);
1311                 if (err)
1312                         return err;
1313         }
1314
1315         return 0;
1316 }
1317
1318 static int bpf_object__init_maps(struct bpf_object *obj, int flags)
1319 {
1320         bool strict = !(flags & MAPS_RELAX_COMPAT);
1321         int err;
1322
1323         err = bpf_object__init_user_maps(obj, strict);
1324         if (err)
1325                 return err;
1326
1327         err = bpf_object__init_user_btf_maps(obj, strict);
1328         if (err)
1329                 return err;
1330
1331         err = bpf_object__init_global_data_maps(obj);
1332         if (err)
1333                 return err;
1334
1335         if (obj->nr_maps) {
1336                 qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]),
1337                       compare_bpf_map);
1338         }
1339         return 0;
1340 }
1341
1342 static bool section_have_execinstr(struct bpf_object *obj, int idx)
1343 {
1344         Elf_Scn *scn;
1345         GElf_Shdr sh;
1346
1347         scn = elf_getscn(obj->efile.elf, idx);
1348         if (!scn)
1349                 return false;
1350
1351         if (gelf_getshdr(scn, &sh) != &sh)
1352                 return false;
1353
1354         if (sh.sh_flags & SHF_EXECINSTR)
1355                 return true;
1356
1357         return false;
1358 }
1359
1360 static void bpf_object__sanitize_btf(struct bpf_object *obj)
1361 {
1362         bool has_datasec = obj->caps.btf_datasec;
1363         bool has_func = obj->caps.btf_func;
1364         struct btf *btf = obj->btf;
1365         struct btf_type *t;
1366         int i, j, vlen;
1367         __u16 kind;
1368
1369         if (!obj->btf || (has_func && has_datasec))
1370                 return;
1371
1372         for (i = 1; i <= btf__get_nr_types(btf); i++) {
1373                 t = (struct btf_type *)btf__type_by_id(btf, i);
1374                 kind = BTF_INFO_KIND(t->info);
1375
1376                 if (!has_datasec && kind == BTF_KIND_VAR) {
1377                         /* replace VAR with INT */
1378                         t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
1379                         t->size = sizeof(int);
1380                         *(int *)(t+1) = BTF_INT_ENC(0, 0, 32);
1381                 } else if (!has_datasec && kind == BTF_KIND_DATASEC) {
1382                         /* replace DATASEC with STRUCT */
1383                         struct btf_var_secinfo *v = (void *)(t + 1);
1384                         struct btf_member *m = (void *)(t + 1);
1385                         struct btf_type *vt;
1386                         char *name;
1387
1388                         name = (char *)btf__name_by_offset(btf, t->name_off);
1389                         while (*name) {
1390                                 if (*name == '.')
1391                                         *name = '_';
1392                                 name++;
1393                         }
1394
1395                         vlen = BTF_INFO_VLEN(t->info);
1396                         t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
1397                         for (j = 0; j < vlen; j++, v++, m++) {
1398                                 /* order of field assignments is important */
1399                                 m->offset = v->offset * 8;
1400                                 m->type = v->type;
1401                                 /* preserve variable name as member name */
1402                                 vt = (void *)btf__type_by_id(btf, v->type);
1403                                 m->name_off = vt->name_off;
1404                         }
1405                 } else if (!has_func && kind == BTF_KIND_FUNC_PROTO) {
1406                         /* replace FUNC_PROTO with ENUM */
1407                         vlen = BTF_INFO_VLEN(t->info);
1408                         t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
1409                         t->size = sizeof(__u32); /* kernel enforced */
1410                 } else if (!has_func && kind == BTF_KIND_FUNC) {
1411                         /* replace FUNC with TYPEDEF */
1412                         t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
1413                 }
1414         }
1415 }
1416
1417 static void bpf_object__sanitize_btf_ext(struct bpf_object *obj)
1418 {
1419         if (!obj->btf_ext)
1420                 return;
1421
1422         if (!obj->caps.btf_func) {
1423                 btf_ext__free(obj->btf_ext);
1424                 obj->btf_ext = NULL;
1425         }
1426 }
1427
1428 static bool bpf_object__is_btf_mandatory(const struct bpf_object *obj)
1429 {
1430         return obj->efile.btf_maps_shndx >= 0;
1431 }
1432
1433 static int bpf_object__init_btf(struct bpf_object *obj,
1434                                 Elf_Data *btf_data,
1435                                 Elf_Data *btf_ext_data)
1436 {
1437         bool btf_required = bpf_object__is_btf_mandatory(obj);
1438         int err = 0;
1439
1440         if (btf_data) {
1441                 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
1442                 if (IS_ERR(obj->btf)) {
1443                         pr_warning("Error loading ELF section %s: %d.\n",
1444                                    BTF_ELF_SEC, err);
1445                         goto out;
1446                 }
1447                 err = btf__finalize_data(obj, obj->btf);
1448                 if (err) {
1449                         pr_warning("Error finalizing %s: %d.\n",
1450                                    BTF_ELF_SEC, err);
1451                         goto out;
1452                 }
1453         }
1454         if (btf_ext_data) {
1455                 if (!obj->btf) {
1456                         pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
1457                                  BTF_EXT_ELF_SEC, BTF_ELF_SEC);
1458                         goto out;
1459                 }
1460                 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
1461                                             btf_ext_data->d_size);
1462                 if (IS_ERR(obj->btf_ext)) {
1463                         pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
1464                                    BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
1465                         obj->btf_ext = NULL;
1466                         goto out;
1467                 }
1468         }
1469 out:
1470         if (err || IS_ERR(obj->btf)) {
1471                 if (btf_required)
1472                         err = err ? : PTR_ERR(obj->btf);
1473                 else
1474                         err = 0;
1475                 if (!IS_ERR_OR_NULL(obj->btf))
1476                         btf__free(obj->btf);
1477                 obj->btf = NULL;
1478         }
1479         if (btf_required && !obj->btf) {
1480                 pr_warning("BTF is required, but is missing or corrupted.\n");
1481                 return err == 0 ? -ENOENT : err;
1482         }
1483         return 0;
1484 }
1485
1486 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
1487 {
1488         int err = 0;
1489
1490         if (!obj->btf)
1491                 return 0;
1492
1493         bpf_object__sanitize_btf(obj);
1494         bpf_object__sanitize_btf_ext(obj);
1495
1496         err = btf__load(obj->btf);
1497         if (err) {
1498                 pr_warning("Error loading %s into kernel: %d.\n",
1499                            BTF_ELF_SEC, err);
1500                 btf__free(obj->btf);
1501                 obj->btf = NULL;
1502                 if (bpf_object__is_btf_mandatory(obj))
1503                         return err;
1504         }
1505         return 0;
1506 }
1507
1508 static int bpf_object__elf_collect(struct bpf_object *obj, int flags)
1509 {
1510         Elf *elf = obj->efile.elf;
1511         GElf_Ehdr *ep = &obj->efile.ehdr;
1512         Elf_Data *btf_ext_data = NULL;
1513         Elf_Data *btf_data = NULL;
1514         Elf_Scn *scn = NULL;
1515         int idx = 0, err = 0;
1516
1517         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1518         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
1519                 pr_warning("failed to get e_shstrndx from %s\n", obj->path);
1520                 return -LIBBPF_ERRNO__FORMAT;
1521         }
1522
1523         while ((scn = elf_nextscn(elf, scn)) != NULL) {
1524                 char *name;
1525                 GElf_Shdr sh;
1526                 Elf_Data *data;
1527
1528                 idx++;
1529                 if (gelf_getshdr(scn, &sh) != &sh) {
1530                         pr_warning("failed to get section(%d) header from %s\n",
1531                                    idx, obj->path);
1532                         return -LIBBPF_ERRNO__FORMAT;
1533                 }
1534
1535                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
1536                 if (!name) {
1537                         pr_warning("failed to get section(%d) name from %s\n",
1538                                    idx, obj->path);
1539                         return -LIBBPF_ERRNO__FORMAT;
1540                 }
1541
1542                 data = elf_getdata(scn, 0);
1543                 if (!data) {
1544                         pr_warning("failed to get section(%d) data from %s(%s)\n",
1545                                    idx, name, obj->path);
1546                         return -LIBBPF_ERRNO__FORMAT;
1547                 }
1548                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
1549                          idx, name, (unsigned long)data->d_size,
1550                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
1551                          (int)sh.sh_type);
1552
1553                 if (strcmp(name, "license") == 0) {
1554                         err = bpf_object__init_license(obj,
1555                                                        data->d_buf,
1556                                                        data->d_size);
1557                         if (err)
1558                                 return err;
1559                 } else if (strcmp(name, "version") == 0) {
1560                         err = bpf_object__init_kversion(obj,
1561                                                         data->d_buf,
1562                                                         data->d_size);
1563                         if (err)
1564                                 return err;
1565                 } else if (strcmp(name, "maps") == 0) {
1566                         obj->efile.maps_shndx = idx;
1567                 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
1568                         obj->efile.btf_maps_shndx = idx;
1569                 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
1570                         btf_data = data;
1571                 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
1572                         btf_ext_data = data;
1573                 } else if (sh.sh_type == SHT_SYMTAB) {
1574                         if (obj->efile.symbols) {
1575                                 pr_warning("bpf: multiple SYMTAB in %s\n",
1576                                            obj->path);
1577                                 return -LIBBPF_ERRNO__FORMAT;
1578                         }
1579                         obj->efile.symbols = data;
1580                         obj->efile.strtabidx = sh.sh_link;
1581                 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
1582                         if (sh.sh_flags & SHF_EXECINSTR) {
1583                                 if (strcmp(name, ".text") == 0)
1584                                         obj->efile.text_shndx = idx;
1585                                 err = bpf_object__add_program(obj, data->d_buf,
1586                                                               data->d_size, name, idx);
1587                                 if (err) {
1588                                         char errmsg[STRERR_BUFSIZE];
1589                                         char *cp = libbpf_strerror_r(-err, errmsg,
1590                                                                      sizeof(errmsg));
1591
1592                                         pr_warning("failed to alloc program %s (%s): %s",
1593                                                    name, obj->path, cp);
1594                                         return err;
1595                                 }
1596                         } else if (strcmp(name, ".data") == 0) {
1597                                 obj->efile.data = data;
1598                                 obj->efile.data_shndx = idx;
1599                         } else if (strcmp(name, ".rodata") == 0) {
1600                                 obj->efile.rodata = data;
1601                                 obj->efile.rodata_shndx = idx;
1602                         } else {
1603                                 pr_debug("skip section(%d) %s\n", idx, name);
1604                         }
1605                 } else if (sh.sh_type == SHT_REL) {
1606                         int nr_reloc = obj->efile.nr_reloc;
1607                         void *reloc = obj->efile.reloc;
1608                         int sec = sh.sh_info; /* points to other section */
1609
1610                         /* Only do relo for section with exec instructions */
1611                         if (!section_have_execinstr(obj, sec)) {
1612                                 pr_debug("skip relo %s(%d) for section(%d)\n",
1613                                          name, idx, sec);
1614                                 continue;
1615                         }
1616
1617                         reloc = reallocarray(reloc, nr_reloc + 1,
1618                                              sizeof(*obj->efile.reloc));
1619                         if (!reloc) {
1620                                 pr_warning("realloc failed\n");
1621                                 return -ENOMEM;
1622                         }
1623
1624                         obj->efile.reloc = reloc;
1625                         obj->efile.nr_reloc++;
1626
1627                         obj->efile.reloc[nr_reloc].shdr = sh;
1628                         obj->efile.reloc[nr_reloc].data = data;
1629                 } else if (sh.sh_type == SHT_NOBITS && strcmp(name, ".bss") == 0) {
1630                         obj->efile.bss = data;
1631                         obj->efile.bss_shndx = idx;
1632                 } else {
1633                         pr_debug("skip section(%d) %s\n", idx, name);
1634                 }
1635         }
1636
1637         if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
1638                 pr_warning("Corrupted ELF file: index of strtab invalid\n");
1639                 return -LIBBPF_ERRNO__FORMAT;
1640         }
1641         err = bpf_object__init_btf(obj, btf_data, btf_ext_data);
1642         if (!err)
1643                 err = bpf_object__init_maps(obj, flags);
1644         if (!err)
1645                 err = bpf_object__sanitize_and_load_btf(obj);
1646         if (!err)
1647                 err = bpf_object__init_prog_names(obj);
1648         return err;
1649 }
1650
1651 static struct bpf_program *
1652 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
1653 {
1654         struct bpf_program *prog;
1655         size_t i;
1656
1657         for (i = 0; i < obj->nr_programs; i++) {
1658                 prog = &obj->programs[i];
1659                 if (prog->idx == idx)
1660                         return prog;
1661         }
1662         return NULL;
1663 }
1664
1665 struct bpf_program *
1666 bpf_object__find_program_by_title(const struct bpf_object *obj,
1667                                   const char *title)
1668 {
1669         struct bpf_program *pos;
1670
1671         bpf_object__for_each_program(pos, obj) {
1672                 if (pos->section_name && !strcmp(pos->section_name, title))
1673                         return pos;
1674         }
1675         return NULL;
1676 }
1677
1678 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
1679                                       int shndx)
1680 {
1681         return shndx == obj->efile.data_shndx ||
1682                shndx == obj->efile.bss_shndx ||
1683                shndx == obj->efile.rodata_shndx;
1684 }
1685
1686 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
1687                                       int shndx)
1688 {
1689         return shndx == obj->efile.maps_shndx ||
1690                shndx == obj->efile.btf_maps_shndx;
1691 }
1692
1693 static bool bpf_object__relo_in_known_section(const struct bpf_object *obj,
1694                                               int shndx)
1695 {
1696         return shndx == obj->efile.text_shndx ||
1697                bpf_object__shndx_is_maps(obj, shndx) ||
1698                bpf_object__shndx_is_data(obj, shndx);
1699 }
1700
1701 static enum libbpf_map_type
1702 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
1703 {
1704         if (shndx == obj->efile.data_shndx)
1705                 return LIBBPF_MAP_DATA;
1706         else if (shndx == obj->efile.bss_shndx)
1707                 return LIBBPF_MAP_BSS;
1708         else if (shndx == obj->efile.rodata_shndx)
1709                 return LIBBPF_MAP_RODATA;
1710         else
1711                 return LIBBPF_MAP_UNSPEC;
1712 }
1713
1714 static int
1715 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
1716                            Elf_Data *data, struct bpf_object *obj)
1717 {
1718         Elf_Data *symbols = obj->efile.symbols;
1719         struct bpf_map *maps = obj->maps;
1720         size_t nr_maps = obj->nr_maps;
1721         int i, nrels;
1722
1723         pr_debug("collecting relocating info for: '%s'\n", prog->section_name);
1724         nrels = shdr->sh_size / shdr->sh_entsize;
1725
1726         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
1727         if (!prog->reloc_desc) {
1728                 pr_warning("failed to alloc memory in relocation\n");
1729                 return -ENOMEM;
1730         }
1731         prog->nr_reloc = nrels;
1732
1733         for (i = 0; i < nrels; i++) {
1734                 struct bpf_insn *insns = prog->insns;
1735                 enum libbpf_map_type type;
1736                 unsigned int insn_idx;
1737                 unsigned int shdr_idx;
1738                 const char *name;
1739                 size_t map_idx;
1740                 GElf_Sym sym;
1741                 GElf_Rel rel;
1742
1743                 if (!gelf_getrel(data, i, &rel)) {
1744                         pr_warning("relocation: failed to get %d reloc\n", i);
1745                         return -LIBBPF_ERRNO__FORMAT;
1746                 }
1747
1748                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
1749                         pr_warning("relocation: symbol %"PRIx64" not found\n",
1750                                    GELF_R_SYM(rel.r_info));
1751                         return -LIBBPF_ERRNO__FORMAT;
1752                 }
1753
1754                 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1755                                   sym.st_name) ? : "<?>";
1756
1757                 pr_debug("relo for %lld value %lld name %d (\'%s\')\n",
1758                          (long long) (rel.r_info >> 32),
1759                          (long long) sym.st_value, sym.st_name, name);
1760
1761                 shdr_idx = sym.st_shndx;
1762                 if (!bpf_object__relo_in_known_section(obj, shdr_idx)) {
1763                         pr_warning("Program '%s' contains unrecognized relo data pointing to section %u\n",
1764                                    prog->section_name, shdr_idx);
1765                         return -LIBBPF_ERRNO__RELOC;
1766                 }
1767
1768                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
1769                 pr_debug("relocation: insn_idx=%u\n", insn_idx);
1770
1771                 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
1772                         if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
1773                                 pr_warning("incorrect bpf_call opcode\n");
1774                                 return -LIBBPF_ERRNO__RELOC;
1775                         }
1776                         prog->reloc_desc[i].type = RELO_CALL;
1777                         prog->reloc_desc[i].insn_idx = insn_idx;
1778                         prog->reloc_desc[i].text_off = sym.st_value;
1779                         obj->has_pseudo_calls = true;
1780                         continue;
1781                 }
1782
1783                 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
1784                         pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
1785                                    insn_idx, insns[insn_idx].code);
1786                         return -LIBBPF_ERRNO__RELOC;
1787                 }
1788
1789                 if (bpf_object__shndx_is_maps(obj, shdr_idx) ||
1790                     bpf_object__shndx_is_data(obj, shdr_idx)) {
1791                         type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
1792                         if (type != LIBBPF_MAP_UNSPEC) {
1793                                 if (GELF_ST_BIND(sym.st_info) == STB_GLOBAL) {
1794                                         pr_warning("bpf: relocation: not yet supported relo for non-static global \'%s\' variable found in insns[%d].code 0x%x\n",
1795                                                    name, insn_idx, insns[insn_idx].code);
1796                                         return -LIBBPF_ERRNO__RELOC;
1797                                 }
1798                                 if (!obj->caps.global_data) {
1799                                         pr_warning("bpf: relocation: kernel does not support global \'%s\' variable access in insns[%d]\n",
1800                                                    name, insn_idx);
1801                                         return -LIBBPF_ERRNO__RELOC;
1802                                 }
1803                         }
1804
1805                         for (map_idx = 0; map_idx < nr_maps; map_idx++) {
1806                                 if (maps[map_idx].libbpf_type != type)
1807                                         continue;
1808                                 if (type != LIBBPF_MAP_UNSPEC ||
1809                                     (maps[map_idx].sec_idx == sym.st_shndx &&
1810                                      maps[map_idx].sec_offset == sym.st_value)) {
1811                                         pr_debug("relocation: found map %zd (%s, sec_idx %d, offset %zu) for insn %u\n",
1812                                                  map_idx, maps[map_idx].name,
1813                                                  maps[map_idx].sec_idx,
1814                                                  maps[map_idx].sec_offset,
1815                                                  insn_idx);
1816                                         break;
1817                                 }
1818                         }
1819
1820                         if (map_idx >= nr_maps) {
1821                                 pr_warning("bpf relocation: map_idx %d larger than %d\n",
1822                                            (int)map_idx, (int)nr_maps - 1);
1823                                 return -LIBBPF_ERRNO__RELOC;
1824                         }
1825
1826                         prog->reloc_desc[i].type = type != LIBBPF_MAP_UNSPEC ?
1827                                                    RELO_DATA : RELO_LD64;
1828                         prog->reloc_desc[i].insn_idx = insn_idx;
1829                         prog->reloc_desc[i].map_idx = map_idx;
1830                 }
1831         }
1832         return 0;
1833 }
1834
1835 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
1836 {
1837         struct bpf_map_def *def = &map->def;
1838         __u32 key_type_id = 0, value_type_id = 0;
1839         int ret;
1840
1841         /* if it's BTF-defined map, we don't need to search for type IDs */
1842         if (map->sec_idx == obj->efile.btf_maps_shndx)
1843                 return 0;
1844
1845         if (!bpf_map__is_internal(map)) {
1846                 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
1847                                            def->value_size, &key_type_id,
1848                                            &value_type_id);
1849         } else {
1850                 /*
1851                  * LLVM annotates global data differently in BTF, that is,
1852                  * only as '.data', '.bss' or '.rodata'.
1853                  */
1854                 ret = btf__find_by_name(obj->btf,
1855                                 libbpf_type_to_btf_name[map->libbpf_type]);
1856         }
1857         if (ret < 0)
1858                 return ret;
1859
1860         map->btf_key_type_id = key_type_id;
1861         map->btf_value_type_id = bpf_map__is_internal(map) ?
1862                                  ret : value_type_id;
1863         return 0;
1864 }
1865
1866 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
1867 {
1868         struct bpf_map_info info = {};
1869         __u32 len = sizeof(info);
1870         int new_fd, err;
1871         char *new_name;
1872
1873         err = bpf_obj_get_info_by_fd(fd, &info, &len);
1874         if (err)
1875                 return err;
1876
1877         new_name = strdup(info.name);
1878         if (!new_name)
1879                 return -errno;
1880
1881         new_fd = open("/", O_RDONLY | O_CLOEXEC);
1882         if (new_fd < 0)
1883                 goto err_free_new_name;
1884
1885         new_fd = dup3(fd, new_fd, O_CLOEXEC);
1886         if (new_fd < 0)
1887                 goto err_close_new_fd;
1888
1889         err = zclose(map->fd);
1890         if (err)
1891                 goto err_close_new_fd;
1892         free(map->name);
1893
1894         map->fd = new_fd;
1895         map->name = new_name;
1896         map->def.type = info.type;
1897         map->def.key_size = info.key_size;
1898         map->def.value_size = info.value_size;
1899         map->def.max_entries = info.max_entries;
1900         map->def.map_flags = info.map_flags;
1901         map->btf_key_type_id = info.btf_key_type_id;
1902         map->btf_value_type_id = info.btf_value_type_id;
1903
1904         return 0;
1905
1906 err_close_new_fd:
1907         close(new_fd);
1908 err_free_new_name:
1909         free(new_name);
1910         return -errno;
1911 }
1912
1913 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
1914 {
1915         if (!map || !max_entries)
1916                 return -EINVAL;
1917
1918         /* If map already created, its attributes can't be changed. */
1919         if (map->fd >= 0)
1920                 return -EBUSY;
1921
1922         map->def.max_entries = max_entries;
1923
1924         return 0;
1925 }
1926
1927 static int
1928 bpf_object__probe_name(struct bpf_object *obj)
1929 {
1930         struct bpf_load_program_attr attr;
1931         char *cp, errmsg[STRERR_BUFSIZE];
1932         struct bpf_insn insns[] = {
1933                 BPF_MOV64_IMM(BPF_REG_0, 0),
1934                 BPF_EXIT_INSN(),
1935         };
1936         int ret;
1937
1938         /* make sure basic loading works */
1939
1940         memset(&attr, 0, sizeof(attr));
1941         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
1942         attr.insns = insns;
1943         attr.insns_cnt = ARRAY_SIZE(insns);
1944         attr.license = "GPL";
1945
1946         ret = bpf_load_program_xattr(&attr, NULL, 0);
1947         if (ret < 0) {
1948                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1949                 pr_warning("Error in %s():%s(%d). Couldn't load basic 'r0 = 0' BPF program.\n",
1950                            __func__, cp, errno);
1951                 return -errno;
1952         }
1953         close(ret);
1954
1955         /* now try the same program, but with the name */
1956
1957         attr.name = "test";
1958         ret = bpf_load_program_xattr(&attr, NULL, 0);
1959         if (ret >= 0) {
1960                 obj->caps.name = 1;
1961                 close(ret);
1962         }
1963
1964         return 0;
1965 }
1966
1967 static int
1968 bpf_object__probe_global_data(struct bpf_object *obj)
1969 {
1970         struct bpf_load_program_attr prg_attr;
1971         struct bpf_create_map_attr map_attr;
1972         char *cp, errmsg[STRERR_BUFSIZE];
1973         struct bpf_insn insns[] = {
1974                 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
1975                 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
1976                 BPF_MOV64_IMM(BPF_REG_0, 0),
1977                 BPF_EXIT_INSN(),
1978         };
1979         int ret, map;
1980
1981         memset(&map_attr, 0, sizeof(map_attr));
1982         map_attr.map_type = BPF_MAP_TYPE_ARRAY;
1983         map_attr.key_size = sizeof(int);
1984         map_attr.value_size = 32;
1985         map_attr.max_entries = 1;
1986
1987         map = bpf_create_map_xattr(&map_attr);
1988         if (map < 0) {
1989                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1990                 pr_warning("Error in %s():%s(%d). Couldn't create simple array map.\n",
1991                            __func__, cp, errno);
1992                 return -errno;
1993         }
1994
1995         insns[0].imm = map;
1996
1997         memset(&prg_attr, 0, sizeof(prg_attr));
1998         prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
1999         prg_attr.insns = insns;
2000         prg_attr.insns_cnt = ARRAY_SIZE(insns);
2001         prg_attr.license = "GPL";
2002
2003         ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
2004         if (ret >= 0) {
2005                 obj->caps.global_data = 1;
2006                 close(ret);
2007         }
2008
2009         close(map);
2010         return 0;
2011 }
2012
2013 static int bpf_object__probe_btf_func(struct bpf_object *obj)
2014 {
2015         const char strs[] = "\0int\0x\0a";
2016         /* void x(int a) {} */
2017         __u32 types[] = {
2018                 /* int */
2019                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
2020                 /* FUNC_PROTO */                                /* [2] */
2021                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
2022                 BTF_PARAM_ENC(7, 1),
2023                 /* FUNC x */                                    /* [3] */
2024                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
2025         };
2026         int btf_fd;
2027
2028         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2029                                       strs, sizeof(strs));
2030         if (btf_fd >= 0) {
2031                 obj->caps.btf_func = 1;
2032                 close(btf_fd);
2033                 return 1;
2034         }
2035
2036         return 0;
2037 }
2038
2039 static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
2040 {
2041         const char strs[] = "\0x\0.data";
2042         /* static int a; */
2043         __u32 types[] = {
2044                 /* int */
2045                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
2046                 /* VAR x */                                     /* [2] */
2047                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
2048                 BTF_VAR_STATIC,
2049                 /* DATASEC val */                               /* [3] */
2050                 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
2051                 BTF_VAR_SECINFO_ENC(2, 0, 4),
2052         };
2053         int btf_fd;
2054
2055         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2056                                       strs, sizeof(strs));
2057         if (btf_fd >= 0) {
2058                 obj->caps.btf_datasec = 1;
2059                 close(btf_fd);
2060                 return 1;
2061         }
2062
2063         return 0;
2064 }
2065
2066 static int
2067 bpf_object__probe_caps(struct bpf_object *obj)
2068 {
2069         int (*probe_fn[])(struct bpf_object *obj) = {
2070                 bpf_object__probe_name,
2071                 bpf_object__probe_global_data,
2072                 bpf_object__probe_btf_func,
2073                 bpf_object__probe_btf_datasec,
2074         };
2075         int i, ret;
2076
2077         for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
2078                 ret = probe_fn[i](obj);
2079                 if (ret < 0)
2080                         pr_debug("Probe #%d failed with %d.\n", i, ret);
2081         }
2082
2083         return 0;
2084 }
2085
2086 static int
2087 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
2088 {
2089         char *cp, errmsg[STRERR_BUFSIZE];
2090         int err, zero = 0;
2091         __u8 *data;
2092
2093         /* Nothing to do here since kernel already zero-initializes .bss map. */
2094         if (map->libbpf_type == LIBBPF_MAP_BSS)
2095                 return 0;
2096
2097         data = map->libbpf_type == LIBBPF_MAP_DATA ?
2098                obj->sections.data : obj->sections.rodata;
2099
2100         err = bpf_map_update_elem(map->fd, &zero, data, 0);
2101         /* Freeze .rodata map as read-only from syscall side. */
2102         if (!err && map->libbpf_type == LIBBPF_MAP_RODATA) {
2103                 err = bpf_map_freeze(map->fd);
2104                 if (err) {
2105                         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2106                         pr_warning("Error freezing map(%s) as read-only: %s\n",
2107                                    map->name, cp);
2108                         err = 0;
2109                 }
2110         }
2111         return err;
2112 }
2113
2114 static int
2115 bpf_object__create_maps(struct bpf_object *obj)
2116 {
2117         struct bpf_create_map_attr create_attr = {};
2118         unsigned int i;
2119         int err;
2120
2121         for (i = 0; i < obj->nr_maps; i++) {
2122                 struct bpf_map *map = &obj->maps[i];
2123                 struct bpf_map_def *def = &map->def;
2124                 char *cp, errmsg[STRERR_BUFSIZE];
2125                 int *pfd = &map->fd;
2126
2127                 if (map->fd >= 0) {
2128                         pr_debug("skip map create (preset) %s: fd=%d\n",
2129                                  map->name, map->fd);
2130                         continue;
2131                 }
2132
2133                 if (obj->caps.name)
2134                         create_attr.name = map->name;
2135                 create_attr.map_ifindex = map->map_ifindex;
2136                 create_attr.map_type = def->type;
2137                 create_attr.map_flags = def->map_flags;
2138                 create_attr.key_size = def->key_size;
2139                 create_attr.value_size = def->value_size;
2140                 create_attr.max_entries = def->max_entries;
2141                 create_attr.btf_fd = 0;
2142                 create_attr.btf_key_type_id = 0;
2143                 create_attr.btf_value_type_id = 0;
2144                 if (bpf_map_type__is_map_in_map(def->type) &&
2145                     map->inner_map_fd >= 0)
2146                         create_attr.inner_map_fd = map->inner_map_fd;
2147
2148                 if (obj->btf && !bpf_map_find_btf_info(obj, map)) {
2149                         create_attr.btf_fd = btf__fd(obj->btf);
2150                         create_attr.btf_key_type_id = map->btf_key_type_id;
2151                         create_attr.btf_value_type_id = map->btf_value_type_id;
2152                 }
2153
2154                 *pfd = bpf_create_map_xattr(&create_attr);
2155                 if (*pfd < 0 && (create_attr.btf_key_type_id ||
2156                                  create_attr.btf_value_type_id)) {
2157                         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2158                         pr_warning("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
2159                                    map->name, cp, errno);
2160                         create_attr.btf_fd = 0;
2161                         create_attr.btf_key_type_id = 0;
2162                         create_attr.btf_value_type_id = 0;
2163                         map->btf_key_type_id = 0;
2164                         map->btf_value_type_id = 0;
2165                         *pfd = bpf_create_map_xattr(&create_attr);
2166                 }
2167
2168                 if (*pfd < 0) {
2169                         size_t j;
2170
2171                         err = *pfd;
2172 err_out:
2173                         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2174                         pr_warning("failed to create map (name: '%s'): %s\n",
2175                                    map->name, cp);
2176                         for (j = 0; j < i; j++)
2177                                 zclose(obj->maps[j].fd);
2178                         return err;
2179                 }
2180
2181                 if (bpf_map__is_internal(map)) {
2182                         err = bpf_object__populate_internal_map(obj, map);
2183                         if (err < 0) {
2184                                 zclose(*pfd);
2185                                 goto err_out;
2186                         }
2187                 }
2188
2189                 pr_debug("created map %s: fd=%d\n", map->name, *pfd);
2190         }
2191
2192         return 0;
2193 }
2194
2195 static int
2196 check_btf_ext_reloc_err(struct bpf_program *prog, int err,
2197                         void *btf_prog_info, const char *info_name)
2198 {
2199         if (err != -ENOENT) {
2200                 pr_warning("Error in loading %s for sec %s.\n",
2201                            info_name, prog->section_name);
2202                 return err;
2203         }
2204
2205         /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
2206
2207         if (btf_prog_info) {
2208                 /*
2209                  * Some info has already been found but has problem
2210                  * in the last btf_ext reloc. Must have to error out.
2211                  */
2212                 pr_warning("Error in relocating %s for sec %s.\n",
2213                            info_name, prog->section_name);
2214                 return err;
2215         }
2216
2217         /* Have problem loading the very first info. Ignore the rest. */
2218         pr_warning("Cannot find %s for main program sec %s. Ignore all %s.\n",
2219                    info_name, prog->section_name, info_name);
2220         return 0;
2221 }
2222
2223 static int
2224 bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
2225                           const char *section_name,  __u32 insn_offset)
2226 {
2227         int err;
2228
2229         if (!insn_offset || prog->func_info) {
2230                 /*
2231                  * !insn_offset => main program
2232                  *
2233                  * For sub prog, the main program's func_info has to
2234                  * be loaded first (i.e. prog->func_info != NULL)
2235                  */
2236                 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
2237                                                section_name, insn_offset,
2238                                                &prog->func_info,
2239                                                &prog->func_info_cnt);
2240                 if (err)
2241                         return check_btf_ext_reloc_err(prog, err,
2242                                                        prog->func_info,
2243                                                        "bpf_func_info");
2244
2245                 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
2246         }
2247
2248         if (!insn_offset || prog->line_info) {
2249                 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
2250                                                section_name, insn_offset,
2251                                                &prog->line_info,
2252                                                &prog->line_info_cnt);
2253                 if (err)
2254                         return check_btf_ext_reloc_err(prog, err,
2255                                                        prog->line_info,
2256                                                        "bpf_line_info");
2257
2258                 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
2259         }
2260
2261         if (!insn_offset)
2262                 prog->btf_fd = btf__fd(obj->btf);
2263
2264         return 0;
2265 }
2266
2267 static int
2268 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
2269                         struct reloc_desc *relo)
2270 {
2271         struct bpf_insn *insn, *new_insn;
2272         struct bpf_program *text;
2273         size_t new_cnt;
2274         int err;
2275
2276         if (relo->type != RELO_CALL)
2277                 return -LIBBPF_ERRNO__RELOC;
2278
2279         if (prog->idx == obj->efile.text_shndx) {
2280                 pr_warning("relo in .text insn %d into off %d\n",
2281                            relo->insn_idx, relo->text_off);
2282                 return -LIBBPF_ERRNO__RELOC;
2283         }
2284
2285         if (prog->main_prog_cnt == 0) {
2286                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
2287                 if (!text) {
2288                         pr_warning("no .text section found yet relo into text exist\n");
2289                         return -LIBBPF_ERRNO__RELOC;
2290                 }
2291                 new_cnt = prog->insns_cnt + text->insns_cnt;
2292                 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
2293                 if (!new_insn) {
2294                         pr_warning("oom in prog realloc\n");
2295                         return -ENOMEM;
2296                 }
2297
2298                 if (obj->btf_ext) {
2299                         err = bpf_program_reloc_btf_ext(prog, obj,
2300                                                         text->section_name,
2301                                                         prog->insns_cnt);
2302                         if (err)
2303                                 return err;
2304                 }
2305
2306                 memcpy(new_insn + prog->insns_cnt, text->insns,
2307                        text->insns_cnt * sizeof(*insn));
2308                 prog->insns = new_insn;
2309                 prog->main_prog_cnt = prog->insns_cnt;
2310                 prog->insns_cnt = new_cnt;
2311                 pr_debug("added %zd insn from %s to prog %s\n",
2312                          text->insns_cnt, text->section_name,
2313                          prog->section_name);
2314         }
2315         insn = &prog->insns[relo->insn_idx];
2316         insn->imm += prog->main_prog_cnt - relo->insn_idx;
2317         return 0;
2318 }
2319
2320 static int
2321 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
2322 {
2323         int i, err;
2324
2325         if (!prog)
2326                 return 0;
2327
2328         if (obj->btf_ext) {
2329                 err = bpf_program_reloc_btf_ext(prog, obj,
2330                                                 prog->section_name, 0);
2331                 if (err)
2332                         return err;
2333         }
2334
2335         if (!prog->reloc_desc)
2336                 return 0;
2337
2338         for (i = 0; i < prog->nr_reloc; i++) {
2339                 if (prog->reloc_desc[i].type == RELO_LD64 ||
2340                     prog->reloc_desc[i].type == RELO_DATA) {
2341                         bool relo_data = prog->reloc_desc[i].type == RELO_DATA;
2342                         struct bpf_insn *insns = prog->insns;
2343                         int insn_idx, map_idx;
2344
2345                         insn_idx = prog->reloc_desc[i].insn_idx;
2346                         map_idx = prog->reloc_desc[i].map_idx;
2347
2348                         if (insn_idx + 1 >= (int)prog->insns_cnt) {
2349                                 pr_warning("relocation out of range: '%s'\n",
2350                                            prog->section_name);
2351                                 return -LIBBPF_ERRNO__RELOC;
2352                         }
2353
2354                         if (!relo_data) {
2355                                 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
2356                         } else {
2357                                 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_VALUE;
2358                                 insns[insn_idx + 1].imm = insns[insn_idx].imm;
2359                         }
2360                         insns[insn_idx].imm = obj->maps[map_idx].fd;
2361                 } else if (prog->reloc_desc[i].type == RELO_CALL) {
2362                         err = bpf_program__reloc_text(prog, obj,
2363                                                       &prog->reloc_desc[i]);
2364                         if (err)
2365                                 return err;
2366                 }
2367         }
2368
2369         zfree(&prog->reloc_desc);
2370         prog->nr_reloc = 0;
2371         return 0;
2372 }
2373
2374
2375 static int
2376 bpf_object__relocate(struct bpf_object *obj)
2377 {
2378         struct bpf_program *prog;
2379         size_t i;
2380         int err;
2381
2382         for (i = 0; i < obj->nr_programs; i++) {
2383                 prog = &obj->programs[i];
2384
2385                 err = bpf_program__relocate(prog, obj);
2386                 if (err) {
2387                         pr_warning("failed to relocate '%s'\n",
2388                                    prog->section_name);
2389                         return err;
2390                 }
2391         }
2392         return 0;
2393 }
2394
2395 static int bpf_object__collect_reloc(struct bpf_object *obj)
2396 {
2397         int i, err;
2398
2399         if (!obj_elf_valid(obj)) {
2400                 pr_warning("Internal error: elf object is closed\n");
2401                 return -LIBBPF_ERRNO__INTERNAL;
2402         }
2403
2404         for (i = 0; i < obj->efile.nr_reloc; i++) {
2405                 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
2406                 Elf_Data *data = obj->efile.reloc[i].data;
2407                 int idx = shdr->sh_info;
2408                 struct bpf_program *prog;
2409
2410                 if (shdr->sh_type != SHT_REL) {
2411                         pr_warning("internal error at %d\n", __LINE__);
2412                         return -LIBBPF_ERRNO__INTERNAL;
2413                 }
2414
2415                 prog = bpf_object__find_prog_by_idx(obj, idx);
2416                 if (!prog) {
2417                         pr_warning("relocation failed: no section(%d)\n", idx);
2418                         return -LIBBPF_ERRNO__RELOC;
2419                 }
2420
2421                 err = bpf_program__collect_reloc(prog, shdr, data, obj);
2422                 if (err)
2423                         return err;
2424         }
2425         return 0;
2426 }
2427
2428 static int
2429 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
2430              char *license, __u32 kern_version, int *pfd)
2431 {
2432         struct bpf_load_program_attr load_attr;
2433         char *cp, errmsg[STRERR_BUFSIZE];
2434         int log_buf_size = BPF_LOG_BUF_SIZE;
2435         char *log_buf;
2436         int ret;
2437
2438         if (!insns || !insns_cnt)
2439                 return -EINVAL;
2440
2441         memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
2442         load_attr.prog_type = prog->type;
2443         load_attr.expected_attach_type = prog->expected_attach_type;
2444         if (prog->caps->name)
2445                 load_attr.name = prog->name;
2446         load_attr.insns = insns;
2447         load_attr.insns_cnt = insns_cnt;
2448         load_attr.license = license;
2449         load_attr.kern_version = kern_version;
2450         load_attr.prog_ifindex = prog->prog_ifindex;
2451         load_attr.prog_btf_fd = prog->btf_fd >= 0 ? prog->btf_fd : 0;
2452         load_attr.func_info = prog->func_info;
2453         load_attr.func_info_rec_size = prog->func_info_rec_size;
2454         load_attr.func_info_cnt = prog->func_info_cnt;
2455         load_attr.line_info = prog->line_info;
2456         load_attr.line_info_rec_size = prog->line_info_rec_size;
2457         load_attr.line_info_cnt = prog->line_info_cnt;
2458         load_attr.log_level = prog->log_level;
2459         load_attr.prog_flags = prog->prog_flags;
2460
2461 retry_load:
2462         log_buf = malloc(log_buf_size);
2463         if (!log_buf)
2464                 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
2465
2466         ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
2467
2468         if (ret >= 0) {
2469                 if (load_attr.log_level)
2470                         pr_debug("verifier log:\n%s", log_buf);
2471                 *pfd = ret;
2472                 ret = 0;
2473                 goto out;
2474         }
2475
2476         if (errno == ENOSPC) {
2477                 log_buf_size <<= 1;
2478                 free(log_buf);
2479                 goto retry_load;
2480         }
2481         ret = -LIBBPF_ERRNO__LOAD;
2482         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2483         pr_warning("load bpf program failed: %s\n", cp);
2484
2485         if (log_buf && log_buf[0] != '\0') {
2486                 ret = -LIBBPF_ERRNO__VERIFY;
2487                 pr_warning("-- BEGIN DUMP LOG ---\n");
2488                 pr_warning("\n%s\n", log_buf);
2489                 pr_warning("-- END LOG --\n");
2490         } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
2491                 pr_warning("Program too large (%zu insns), at most %d insns\n",
2492                            load_attr.insns_cnt, BPF_MAXINSNS);
2493                 ret = -LIBBPF_ERRNO__PROG2BIG;
2494         } else {
2495                 /* Wrong program type? */
2496                 if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
2497                         int fd;
2498
2499                         load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
2500                         load_attr.expected_attach_type = 0;
2501                         fd = bpf_load_program_xattr(&load_attr, NULL, 0);
2502                         if (fd >= 0) {
2503                                 close(fd);
2504                                 ret = -LIBBPF_ERRNO__PROGTYPE;
2505                                 goto out;
2506                         }
2507                 }
2508
2509                 if (log_buf)
2510                         ret = -LIBBPF_ERRNO__KVER;
2511         }
2512
2513 out:
2514         free(log_buf);
2515         return ret;
2516 }
2517
2518 int
2519 bpf_program__load(struct bpf_program *prog,
2520                   char *license, __u32 kern_version)
2521 {
2522         int err = 0, fd, i;
2523
2524         if (prog->instances.nr < 0 || !prog->instances.fds) {
2525                 if (prog->preprocessor) {
2526                         pr_warning("Internal error: can't load program '%s'\n",
2527                                    prog->section_name);
2528                         return -LIBBPF_ERRNO__INTERNAL;
2529                 }
2530
2531                 prog->instances.fds = malloc(sizeof(int));
2532                 if (!prog->instances.fds) {
2533                         pr_warning("Not enough memory for BPF fds\n");
2534                         return -ENOMEM;
2535                 }
2536                 prog->instances.nr = 1;
2537                 prog->instances.fds[0] = -1;
2538         }
2539
2540         if (!prog->preprocessor) {
2541                 if (prog->instances.nr != 1) {
2542                         pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
2543                                    prog->section_name, prog->instances.nr);
2544                 }
2545                 err = load_program(prog, prog->insns, prog->insns_cnt,
2546                                    license, kern_version, &fd);
2547                 if (!err)
2548                         prog->instances.fds[0] = fd;
2549                 goto out;
2550         }
2551
2552         for (i = 0; i < prog->instances.nr; i++) {
2553                 struct bpf_prog_prep_result result;
2554                 bpf_program_prep_t preprocessor = prog->preprocessor;
2555
2556                 memset(&result, 0, sizeof(result));
2557                 err = preprocessor(prog, i, prog->insns,
2558                                    prog->insns_cnt, &result);
2559                 if (err) {
2560                         pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
2561                                    i, prog->section_name);
2562                         goto out;
2563                 }
2564
2565                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
2566                         pr_debug("Skip loading the %dth instance of program '%s'\n",
2567                                  i, prog->section_name);
2568                         prog->instances.fds[i] = -1;
2569                         if (result.pfd)
2570                                 *result.pfd = -1;
2571                         continue;
2572                 }
2573
2574                 err = load_program(prog, result.new_insn_ptr,
2575                                    result.new_insn_cnt,
2576                                    license, kern_version, &fd);
2577
2578                 if (err) {
2579                         pr_warning("Loading the %dth instance of program '%s' failed\n",
2580                                         i, prog->section_name);
2581                         goto out;
2582                 }
2583
2584                 if (result.pfd)
2585                         *result.pfd = fd;
2586                 prog->instances.fds[i] = fd;
2587         }
2588 out:
2589         if (err)
2590                 pr_warning("failed to load program '%s'\n",
2591                            prog->section_name);
2592         zfree(&prog->insns);
2593         prog->insns_cnt = 0;
2594         return err;
2595 }
2596
2597 static bool bpf_program__is_function_storage(const struct bpf_program *prog,
2598                                              const struct bpf_object *obj)
2599 {
2600         return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
2601 }
2602
2603 static int
2604 bpf_object__load_progs(struct bpf_object *obj, int log_level)
2605 {
2606         size_t i;
2607         int err;
2608
2609         for (i = 0; i < obj->nr_programs; i++) {
2610                 if (bpf_program__is_function_storage(&obj->programs[i], obj))
2611                         continue;
2612                 obj->programs[i].log_level |= log_level;
2613                 err = bpf_program__load(&obj->programs[i],
2614                                         obj->license,
2615                                         obj->kern_version);
2616                 if (err)
2617                         return err;
2618         }
2619         return 0;
2620 }
2621
2622 static bool bpf_prog_type__needs_kver(enum bpf_prog_type type)
2623 {
2624         switch (type) {
2625         case BPF_PROG_TYPE_SOCKET_FILTER:
2626         case BPF_PROG_TYPE_SCHED_CLS:
2627         case BPF_PROG_TYPE_SCHED_ACT:
2628         case BPF_PROG_TYPE_XDP:
2629         case BPF_PROG_TYPE_CGROUP_SKB:
2630         case BPF_PROG_TYPE_CGROUP_SOCK:
2631         case BPF_PROG_TYPE_LWT_IN:
2632         case BPF_PROG_TYPE_LWT_OUT:
2633         case BPF_PROG_TYPE_LWT_XMIT:
2634         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2635         case BPF_PROG_TYPE_SOCK_OPS:
2636         case BPF_PROG_TYPE_SK_SKB:
2637         case BPF_PROG_TYPE_CGROUP_DEVICE:
2638         case BPF_PROG_TYPE_SK_MSG:
2639         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2640         case BPF_PROG_TYPE_LIRC_MODE2:
2641         case BPF_PROG_TYPE_SK_REUSEPORT:
2642         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2643         case BPF_PROG_TYPE_UNSPEC:
2644         case BPF_PROG_TYPE_TRACEPOINT:
2645         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2646         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2647         case BPF_PROG_TYPE_PERF_EVENT:
2648         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2649         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2650                 return false;
2651         case BPF_PROG_TYPE_KPROBE:
2652         default:
2653                 return true;
2654         }
2655 }
2656
2657 static int bpf_object__validate(struct bpf_object *obj, bool needs_kver)
2658 {
2659         if (needs_kver && obj->kern_version == 0) {
2660                 pr_warning("%s doesn't provide kernel version\n",
2661                            obj->path);
2662                 return -LIBBPF_ERRNO__KVERSION;
2663         }
2664         return 0;
2665 }
2666
2667 static struct bpf_object *
2668 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz,
2669                    bool needs_kver, int flags)
2670 {
2671         struct bpf_object *obj;
2672         int err;
2673
2674         if (elf_version(EV_CURRENT) == EV_NONE) {
2675                 pr_warning("failed to init libelf for %s\n", path);
2676                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
2677         }
2678
2679         obj = bpf_object__new(path, obj_buf, obj_buf_sz);
2680         if (IS_ERR(obj))
2681                 return obj;
2682
2683         CHECK_ERR(bpf_object__elf_init(obj), err, out);
2684         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
2685         CHECK_ERR(bpf_object__probe_caps(obj), err, out);
2686         CHECK_ERR(bpf_object__elf_collect(obj, flags), err, out);
2687         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
2688         CHECK_ERR(bpf_object__validate(obj, needs_kver), err, out);
2689
2690         bpf_object__elf_finish(obj);
2691         return obj;
2692 out:
2693         bpf_object__close(obj);
2694         return ERR_PTR(err);
2695 }
2696
2697 struct bpf_object *__bpf_object__open_xattr(struct bpf_object_open_attr *attr,
2698                                             int flags)
2699 {
2700         /* param validation */
2701         if (!attr->file)
2702                 return NULL;
2703
2704         pr_debug("loading %s\n", attr->file);
2705
2706         return __bpf_object__open(attr->file, NULL, 0,
2707                                   bpf_prog_type__needs_kver(attr->prog_type),
2708                                   flags);
2709 }
2710
2711 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
2712 {
2713         return __bpf_object__open_xattr(attr, 0);
2714 }
2715
2716 struct bpf_object *bpf_object__open(const char *path)
2717 {
2718         struct bpf_object_open_attr attr = {
2719                 .file           = path,
2720                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
2721         };
2722
2723         return bpf_object__open_xattr(&attr);
2724 }
2725
2726 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
2727                                            size_t obj_buf_sz,
2728                                            const char *name)
2729 {
2730         char tmp_name[64];
2731
2732         /* param validation */
2733         if (!obj_buf || obj_buf_sz <= 0)
2734                 return NULL;
2735
2736         if (!name) {
2737                 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
2738                          (unsigned long)obj_buf,
2739                          (unsigned long)obj_buf_sz);
2740                 name = tmp_name;
2741         }
2742         pr_debug("loading object '%s' from buffer\n", name);
2743
2744         return __bpf_object__open(name, obj_buf, obj_buf_sz, true, true);
2745 }
2746
2747 int bpf_object__unload(struct bpf_object *obj)
2748 {
2749         size_t i;
2750
2751         if (!obj)
2752                 return -EINVAL;
2753
2754         for (i = 0; i < obj->nr_maps; i++)
2755                 zclose(obj->maps[i].fd);
2756
2757         for (i = 0; i < obj->nr_programs; i++)
2758                 bpf_program__unload(&obj->programs[i]);
2759
2760         return 0;
2761 }
2762
2763 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
2764 {
2765         struct bpf_object *obj;
2766         int err;
2767
2768         if (!attr)
2769                 return -EINVAL;
2770         obj = attr->obj;
2771         if (!obj)
2772                 return -EINVAL;
2773
2774         if (obj->loaded) {
2775                 pr_warning("object should not be loaded twice\n");
2776                 return -EINVAL;
2777         }
2778
2779         obj->loaded = true;
2780
2781         CHECK_ERR(bpf_object__create_maps(obj), err, out);
2782         CHECK_ERR(bpf_object__relocate(obj), err, out);
2783         CHECK_ERR(bpf_object__load_progs(obj, attr->log_level), err, out);
2784
2785         return 0;
2786 out:
2787         bpf_object__unload(obj);
2788         pr_warning("failed to load object '%s'\n", obj->path);
2789         return err;
2790 }
2791
2792 int bpf_object__load(struct bpf_object *obj)
2793 {
2794         struct bpf_object_load_attr attr = {
2795                 .obj = obj,
2796         };
2797
2798         return bpf_object__load_xattr(&attr);
2799 }
2800
2801 static int check_path(const char *path)
2802 {
2803         char *cp, errmsg[STRERR_BUFSIZE];
2804         struct statfs st_fs;
2805         char *dname, *dir;
2806         int err = 0;
2807
2808         if (path == NULL)
2809                 return -EINVAL;
2810
2811         dname = strdup(path);
2812         if (dname == NULL)
2813                 return -ENOMEM;
2814
2815         dir = dirname(dname);
2816         if (statfs(dir, &st_fs)) {
2817                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2818                 pr_warning("failed to statfs %s: %s\n", dir, cp);
2819                 err = -errno;
2820         }
2821         free(dname);
2822
2823         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
2824                 pr_warning("specified path %s is not on BPF FS\n", path);
2825                 err = -EINVAL;
2826         }
2827
2828         return err;
2829 }
2830
2831 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
2832                               int instance)
2833 {
2834         char *cp, errmsg[STRERR_BUFSIZE];
2835         int err;
2836
2837         err = check_path(path);
2838         if (err)
2839                 return err;
2840
2841         if (prog == NULL) {
2842                 pr_warning("invalid program pointer\n");
2843                 return -EINVAL;
2844         }
2845
2846         if (instance < 0 || instance >= prog->instances.nr) {
2847                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
2848                            instance, prog->section_name, prog->instances.nr);
2849                 return -EINVAL;
2850         }
2851
2852         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
2853                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2854                 pr_warning("failed to pin program: %s\n", cp);
2855                 return -errno;
2856         }
2857         pr_debug("pinned program '%s'\n", path);
2858
2859         return 0;
2860 }
2861
2862 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
2863                                 int instance)
2864 {
2865         int err;
2866
2867         err = check_path(path);
2868         if (err)
2869                 return err;
2870
2871         if (prog == NULL) {
2872                 pr_warning("invalid program pointer\n");
2873                 return -EINVAL;
2874         }
2875
2876         if (instance < 0 || instance >= prog->instances.nr) {
2877                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
2878                            instance, prog->section_name, prog->instances.nr);
2879                 return -EINVAL;
2880         }
2881
2882         err = unlink(path);
2883         if (err != 0)
2884                 return -errno;
2885         pr_debug("unpinned program '%s'\n", path);
2886
2887         return 0;
2888 }
2889
2890 static int make_dir(const char *path)
2891 {
2892         char *cp, errmsg[STRERR_BUFSIZE];
2893         int err = 0;
2894
2895         if (mkdir(path, 0700) && errno != EEXIST)
2896                 err = -errno;
2897
2898         if (err) {
2899                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
2900                 pr_warning("failed to mkdir %s: %s\n", path, cp);
2901         }
2902         return err;
2903 }
2904
2905 int bpf_program__pin(struct bpf_program *prog, const char *path)
2906 {
2907         int i, err;
2908
2909         err = check_path(path);
2910         if (err)
2911                 return err;
2912
2913         if (prog == NULL) {
2914                 pr_warning("invalid program pointer\n");
2915                 return -EINVAL;
2916         }
2917
2918         if (prog->instances.nr <= 0) {
2919                 pr_warning("no instances of prog %s to pin\n",
2920                            prog->section_name);
2921                 return -EINVAL;
2922         }
2923
2924         if (prog->instances.nr == 1) {
2925                 /* don't create subdirs when pinning single instance */
2926                 return bpf_program__pin_instance(prog, path, 0);
2927         }
2928
2929         err = make_dir(path);
2930         if (err)
2931                 return err;
2932
2933         for (i = 0; i < prog->instances.nr; i++) {
2934                 char buf[PATH_MAX];
2935                 int len;
2936
2937                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
2938                 if (len < 0) {
2939                         err = -EINVAL;
2940                         goto err_unpin;
2941                 } else if (len >= PATH_MAX) {
2942                         err = -ENAMETOOLONG;
2943                         goto err_unpin;
2944                 }
2945
2946                 err = bpf_program__pin_instance(prog, buf, i);
2947                 if (err)
2948                         goto err_unpin;
2949         }
2950
2951         return 0;
2952
2953 err_unpin:
2954         for (i = i - 1; i >= 0; i--) {
2955                 char buf[PATH_MAX];
2956                 int len;
2957
2958                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
2959                 if (len < 0)
2960                         continue;
2961                 else if (len >= PATH_MAX)
2962                         continue;
2963
2964                 bpf_program__unpin_instance(prog, buf, i);
2965         }
2966
2967         rmdir(path);
2968
2969         return err;
2970 }
2971
2972 int bpf_program__unpin(struct bpf_program *prog, const char *path)
2973 {
2974         int i, err;
2975
2976         err = check_path(path);
2977         if (err)
2978                 return err;
2979
2980         if (prog == NULL) {
2981                 pr_warning("invalid program pointer\n");
2982                 return -EINVAL;
2983         }
2984
2985         if (prog->instances.nr <= 0) {
2986                 pr_warning("no instances of prog %s to pin\n",
2987                            prog->section_name);
2988                 return -EINVAL;
2989         }
2990
2991         if (prog->instances.nr == 1) {
2992                 /* don't create subdirs when pinning single instance */
2993                 return bpf_program__unpin_instance(prog, path, 0);
2994         }
2995
2996         for (i = 0; i < prog->instances.nr; i++) {
2997                 char buf[PATH_MAX];
2998                 int len;
2999
3000                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
3001                 if (len < 0)
3002                         return -EINVAL;
3003                 else if (len >= PATH_MAX)
3004                         return -ENAMETOOLONG;
3005
3006                 err = bpf_program__unpin_instance(prog, buf, i);
3007                 if (err)
3008                         return err;
3009         }
3010
3011         err = rmdir(path);
3012         if (err)
3013                 return -errno;
3014
3015         return 0;
3016 }
3017
3018 int bpf_map__pin(struct bpf_map *map, const char *path)
3019 {
3020         char *cp, errmsg[STRERR_BUFSIZE];
3021         int err;
3022
3023         err = check_path(path);
3024         if (err)
3025                 return err;
3026
3027         if (map == NULL) {
3028                 pr_warning("invalid map pointer\n");
3029                 return -EINVAL;
3030         }
3031
3032         if (bpf_obj_pin(map->fd, path)) {
3033                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
3034                 pr_warning("failed to pin map: %s\n", cp);
3035                 return -errno;
3036         }
3037
3038         pr_debug("pinned map '%s'\n", path);
3039
3040         return 0;
3041 }
3042
3043 int bpf_map__unpin(struct bpf_map *map, const char *path)
3044 {
3045         int err;
3046
3047         err = check_path(path);
3048         if (err)
3049                 return err;
3050
3051         if (map == NULL) {
3052                 pr_warning("invalid map pointer\n");
3053                 return -EINVAL;
3054         }
3055
3056         err = unlink(path);
3057         if (err != 0)
3058                 return -errno;
3059         pr_debug("unpinned map '%s'\n", path);
3060
3061         return 0;
3062 }
3063
3064 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
3065 {
3066         struct bpf_map *map;
3067         int err;
3068
3069         if (!obj)
3070                 return -ENOENT;
3071
3072         if (!obj->loaded) {
3073                 pr_warning("object not yet loaded; load it first\n");
3074                 return -ENOENT;
3075         }
3076
3077         err = make_dir(path);
3078         if (err)
3079                 return err;
3080
3081         bpf_object__for_each_map(map, obj) {
3082                 char buf[PATH_MAX];
3083                 int len;
3084
3085                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
3086                                bpf_map__name(map));
3087                 if (len < 0) {
3088                         err = -EINVAL;
3089                         goto err_unpin_maps;
3090                 } else if (len >= PATH_MAX) {
3091                         err = -ENAMETOOLONG;
3092                         goto err_unpin_maps;
3093                 }
3094
3095                 err = bpf_map__pin(map, buf);
3096                 if (err)
3097                         goto err_unpin_maps;
3098         }
3099
3100         return 0;
3101
3102 err_unpin_maps:
3103         while ((map = bpf_map__prev(map, obj))) {
3104                 char buf[PATH_MAX];
3105                 int len;
3106
3107                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
3108                                bpf_map__name(map));
3109                 if (len < 0)
3110                         continue;
3111                 else if (len >= PATH_MAX)
3112                         continue;
3113
3114                 bpf_map__unpin(map, buf);
3115         }
3116
3117         return err;
3118 }
3119
3120 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
3121 {
3122         struct bpf_map *map;
3123         int err;
3124
3125         if (!obj)
3126                 return -ENOENT;
3127
3128         bpf_object__for_each_map(map, obj) {
3129                 char buf[PATH_MAX];
3130                 int len;
3131
3132                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
3133                                bpf_map__name(map));
3134                 if (len < 0)
3135                         return -EINVAL;
3136                 else if (len >= PATH_MAX)
3137                         return -ENAMETOOLONG;
3138
3139                 err = bpf_map__unpin(map, buf);
3140                 if (err)
3141                         return err;
3142         }
3143
3144         return 0;
3145 }
3146
3147 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
3148 {
3149         struct bpf_program *prog;
3150         int err;
3151
3152         if (!obj)
3153                 return -ENOENT;
3154
3155         if (!obj->loaded) {
3156                 pr_warning("object not yet loaded; load it first\n");
3157                 return -ENOENT;
3158         }
3159
3160         err = make_dir(path);
3161         if (err)
3162                 return err;
3163
3164         bpf_object__for_each_program(prog, obj) {
3165                 char buf[PATH_MAX];
3166                 int len;
3167
3168                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
3169                                prog->pin_name);
3170                 if (len < 0) {
3171                         err = -EINVAL;
3172                         goto err_unpin_programs;
3173                 } else if (len >= PATH_MAX) {
3174                         err = -ENAMETOOLONG;
3175                         goto err_unpin_programs;
3176                 }
3177
3178                 err = bpf_program__pin(prog, buf);
3179                 if (err)
3180                         goto err_unpin_programs;
3181         }
3182
3183         return 0;
3184
3185 err_unpin_programs:
3186         while ((prog = bpf_program__prev(prog, obj))) {
3187                 char buf[PATH_MAX];
3188                 int len;
3189
3190                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
3191                                prog->pin_name);
3192                 if (len < 0)
3193                         continue;
3194                 else if (len >= PATH_MAX)
3195                         continue;
3196
3197                 bpf_program__unpin(prog, buf);
3198         }
3199
3200         return err;
3201 }
3202
3203 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
3204 {
3205         struct bpf_program *prog;
3206         int err;
3207
3208         if (!obj)
3209                 return -ENOENT;
3210
3211         bpf_object__for_each_program(prog, obj) {
3212                 char buf[PATH_MAX];
3213                 int len;
3214
3215                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
3216                                prog->pin_name);
3217                 if (len < 0)
3218                         return -EINVAL;
3219                 else if (len >= PATH_MAX)
3220                         return -ENAMETOOLONG;
3221
3222                 err = bpf_program__unpin(prog, buf);
3223                 if (err)
3224                         return err;
3225         }
3226
3227         return 0;
3228 }
3229
3230 int bpf_object__pin(struct bpf_object *obj, const char *path)
3231 {
3232         int err;
3233
3234         err = bpf_object__pin_maps(obj, path);
3235         if (err)
3236                 return err;
3237
3238         err = bpf_object__pin_programs(obj, path);
3239         if (err) {
3240                 bpf_object__unpin_maps(obj, path);
3241                 return err;
3242         }
3243
3244         return 0;
3245 }
3246
3247 void bpf_object__close(struct bpf_object *obj)
3248 {
3249         size_t i;
3250
3251         if (!obj)
3252                 return;
3253
3254         if (obj->clear_priv)
3255                 obj->clear_priv(obj, obj->priv);
3256
3257         bpf_object__elf_finish(obj);
3258         bpf_object__unload(obj);
3259         btf__free(obj->btf);
3260         btf_ext__free(obj->btf_ext);
3261
3262         for (i = 0; i < obj->nr_maps; i++) {
3263                 zfree(&obj->maps[i].name);
3264                 if (obj->maps[i].clear_priv)
3265                         obj->maps[i].clear_priv(&obj->maps[i],
3266                                                 obj->maps[i].priv);
3267                 obj->maps[i].priv = NULL;
3268                 obj->maps[i].clear_priv = NULL;
3269         }
3270
3271         zfree(&obj->sections.rodata);
3272         zfree(&obj->sections.data);
3273         zfree(&obj->maps);
3274         obj->nr_maps = 0;
3275
3276         if (obj->programs && obj->nr_programs) {
3277                 for (i = 0; i < obj->nr_programs; i++)
3278                         bpf_program__exit(&obj->programs[i]);
3279         }
3280         zfree(&obj->programs);
3281
3282         list_del(&obj->list);
3283         free(obj);
3284 }
3285
3286 struct bpf_object *
3287 bpf_object__next(struct bpf_object *prev)
3288 {
3289         struct bpf_object *next;
3290
3291         if (!prev)
3292                 next = list_first_entry(&bpf_objects_list,
3293                                         struct bpf_object,
3294                                         list);
3295         else
3296                 next = list_next_entry(prev, list);
3297
3298         /* Empty list is noticed here so don't need checking on entry. */
3299         if (&next->list == &bpf_objects_list)
3300                 return NULL;
3301
3302         return next;
3303 }
3304
3305 const char *bpf_object__name(const struct bpf_object *obj)
3306 {
3307         return obj ? obj->path : ERR_PTR(-EINVAL);
3308 }
3309
3310 unsigned int bpf_object__kversion(const struct bpf_object *obj)
3311 {
3312         return obj ? obj->kern_version : 0;
3313 }
3314
3315 struct btf *bpf_object__btf(const struct bpf_object *obj)
3316 {
3317         return obj ? obj->btf : NULL;
3318 }
3319
3320 int bpf_object__btf_fd(const struct bpf_object *obj)
3321 {
3322         return obj->btf ? btf__fd(obj->btf) : -1;
3323 }
3324
3325 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
3326                          bpf_object_clear_priv_t clear_priv)
3327 {
3328         if (obj->priv && obj->clear_priv)
3329                 obj->clear_priv(obj, obj->priv);
3330
3331         obj->priv = priv;
3332         obj->clear_priv = clear_priv;
3333         return 0;
3334 }
3335
3336 void *bpf_object__priv(const struct bpf_object *obj)
3337 {
3338         return obj ? obj->priv : ERR_PTR(-EINVAL);
3339 }
3340
3341 static struct bpf_program *
3342 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
3343                     bool forward)
3344 {
3345         size_t nr_programs = obj->nr_programs;
3346         ssize_t idx;
3347
3348         if (!nr_programs)
3349                 return NULL;
3350
3351         if (!p)
3352                 /* Iter from the beginning */
3353                 return forward ? &obj->programs[0] :
3354                         &obj->programs[nr_programs - 1];
3355
3356         if (p->obj != obj) {
3357                 pr_warning("error: program handler doesn't match object\n");
3358                 return NULL;
3359         }
3360
3361         idx = (p - obj->programs) + (forward ? 1 : -1);
3362         if (idx >= obj->nr_programs || idx < 0)
3363                 return NULL;
3364         return &obj->programs[idx];
3365 }
3366
3367 struct bpf_program *
3368 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
3369 {
3370         struct bpf_program *prog = prev;
3371
3372         do {
3373                 prog = __bpf_program__iter(prog, obj, true);
3374         } while (prog && bpf_program__is_function_storage(prog, obj));
3375
3376         return prog;
3377 }
3378
3379 struct bpf_program *
3380 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
3381 {
3382         struct bpf_program *prog = next;
3383
3384         do {
3385                 prog = __bpf_program__iter(prog, obj, false);
3386         } while (prog && bpf_program__is_function_storage(prog, obj));
3387
3388         return prog;
3389 }
3390
3391 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
3392                           bpf_program_clear_priv_t clear_priv)
3393 {
3394         if (prog->priv && prog->clear_priv)
3395                 prog->clear_priv(prog, prog->priv);
3396
3397         prog->priv = priv;
3398         prog->clear_priv = clear_priv;
3399         return 0;
3400 }
3401
3402 void *bpf_program__priv(const struct bpf_program *prog)
3403 {
3404         return prog ? prog->priv : ERR_PTR(-EINVAL);
3405 }
3406
3407 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
3408 {
3409         prog->prog_ifindex = ifindex;
3410 }
3411
3412 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
3413 {
3414         const char *title;
3415
3416         title = prog->section_name;
3417         if (needs_copy) {
3418                 title = strdup(title);
3419                 if (!title) {
3420                         pr_warning("failed to strdup program title\n");
3421                         return ERR_PTR(-ENOMEM);
3422                 }
3423         }
3424
3425         return title;
3426 }
3427
3428 int bpf_program__fd(const struct bpf_program *prog)
3429 {
3430         return bpf_program__nth_fd(prog, 0);
3431 }
3432
3433 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
3434                           bpf_program_prep_t prep)
3435 {
3436         int *instances_fds;
3437
3438         if (nr_instances <= 0 || !prep)
3439                 return -EINVAL;
3440
3441         if (prog->instances.nr > 0 || prog->instances.fds) {
3442                 pr_warning("Can't set pre-processor after loading\n");
3443                 return -EINVAL;
3444         }
3445
3446         instances_fds = malloc(sizeof(int) * nr_instances);
3447         if (!instances_fds) {
3448                 pr_warning("alloc memory failed for fds\n");
3449                 return -ENOMEM;
3450         }
3451
3452         /* fill all fd with -1 */
3453         memset(instances_fds, -1, sizeof(int) * nr_instances);
3454
3455         prog->instances.nr = nr_instances;
3456         prog->instances.fds = instances_fds;
3457         prog->preprocessor = prep;
3458         return 0;
3459 }
3460
3461 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
3462 {
3463         int fd;
3464
3465         if (!prog)
3466                 return -EINVAL;
3467
3468         if (n >= prog->instances.nr || n < 0) {
3469                 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
3470                            n, prog->section_name, prog->instances.nr);
3471                 return -EINVAL;
3472         }
3473
3474         fd = prog->instances.fds[n];
3475         if (fd < 0) {
3476                 pr_warning("%dth instance of program '%s' is invalid\n",
3477                            n, prog->section_name);
3478                 return -ENOENT;
3479         }
3480
3481         return fd;
3482 }
3483
3484 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
3485 {
3486         prog->type = type;
3487 }
3488
3489 static bool bpf_program__is_type(const struct bpf_program *prog,
3490                                  enum bpf_prog_type type)
3491 {
3492         return prog ? (prog->type == type) : false;
3493 }
3494
3495 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                           \
3496 int bpf_program__set_##NAME(struct bpf_program *prog)           \
3497 {                                                               \
3498         if (!prog)                                              \
3499                 return -EINVAL;                                 \
3500         bpf_program__set_type(prog, TYPE);                      \
3501         return 0;                                               \
3502 }                                                               \
3503                                                                 \
3504 bool bpf_program__is_##NAME(const struct bpf_program *prog)     \
3505 {                                                               \
3506         return bpf_program__is_type(prog, TYPE);                \
3507 }                                                               \
3508
3509 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
3510 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
3511 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
3512 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
3513 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
3514 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
3515 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
3516 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
3517
3518 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
3519                                            enum bpf_attach_type type)
3520 {
3521         prog->expected_attach_type = type;
3522 }
3523
3524 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, is_attachable, atype) \
3525         { string, sizeof(string) - 1, ptype, eatype, is_attachable, atype }
3526
3527 /* Programs that can NOT be attached. */
3528 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0)
3529
3530 /* Programs that can be attached. */
3531 #define BPF_APROG_SEC(string, ptype, atype) \
3532         BPF_PROG_SEC_IMPL(string, ptype, 0, 1, atype)
3533
3534 /* Programs that must specify expected attach type at load time. */
3535 #define BPF_EAPROG_SEC(string, ptype, eatype) \
3536         BPF_PROG_SEC_IMPL(string, ptype, eatype, 1, eatype)
3537
3538 /* Programs that can be attached but attach type can't be identified by section
3539  * name. Kept for backward compatibility.
3540  */
3541 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
3542
3543 static const struct {
3544         const char *sec;
3545         size_t len;
3546         enum bpf_prog_type prog_type;
3547         enum bpf_attach_type expected_attach_type;
3548         int is_attachable;
3549         enum bpf_attach_type attach_type;
3550 } section_names[] = {
3551         BPF_PROG_SEC("socket",                  BPF_PROG_TYPE_SOCKET_FILTER),
3552         BPF_PROG_SEC("kprobe/",                 BPF_PROG_TYPE_KPROBE),
3553         BPF_PROG_SEC("kretprobe/",              BPF_PROG_TYPE_KPROBE),
3554         BPF_PROG_SEC("classifier",              BPF_PROG_TYPE_SCHED_CLS),
3555         BPF_PROG_SEC("action",                  BPF_PROG_TYPE_SCHED_ACT),
3556         BPF_PROG_SEC("tracepoint/",             BPF_PROG_TYPE_TRACEPOINT),
3557         BPF_PROG_SEC("raw_tracepoint/",         BPF_PROG_TYPE_RAW_TRACEPOINT),
3558         BPF_PROG_SEC("xdp",                     BPF_PROG_TYPE_XDP),
3559         BPF_PROG_SEC("perf_event",              BPF_PROG_TYPE_PERF_EVENT),
3560         BPF_PROG_SEC("lwt_in",                  BPF_PROG_TYPE_LWT_IN),
3561         BPF_PROG_SEC("lwt_out",                 BPF_PROG_TYPE_LWT_OUT),
3562         BPF_PROG_SEC("lwt_xmit",                BPF_PROG_TYPE_LWT_XMIT),
3563         BPF_PROG_SEC("lwt_seg6local",           BPF_PROG_TYPE_LWT_SEG6LOCAL),
3564         BPF_APROG_SEC("cgroup_skb/ingress",     BPF_PROG_TYPE_CGROUP_SKB,
3565                                                 BPF_CGROUP_INET_INGRESS),
3566         BPF_APROG_SEC("cgroup_skb/egress",      BPF_PROG_TYPE_CGROUP_SKB,
3567                                                 BPF_CGROUP_INET_EGRESS),
3568         BPF_APROG_COMPAT("cgroup/skb",          BPF_PROG_TYPE_CGROUP_SKB),
3569         BPF_APROG_SEC("cgroup/sock",            BPF_PROG_TYPE_CGROUP_SOCK,
3570                                                 BPF_CGROUP_INET_SOCK_CREATE),
3571         BPF_EAPROG_SEC("cgroup/post_bind4",     BPF_PROG_TYPE_CGROUP_SOCK,
3572                                                 BPF_CGROUP_INET4_POST_BIND),
3573         BPF_EAPROG_SEC("cgroup/post_bind6",     BPF_PROG_TYPE_CGROUP_SOCK,
3574                                                 BPF_CGROUP_INET6_POST_BIND),
3575         BPF_APROG_SEC("cgroup/dev",             BPF_PROG_TYPE_CGROUP_DEVICE,
3576                                                 BPF_CGROUP_DEVICE),
3577         BPF_APROG_SEC("sockops",                BPF_PROG_TYPE_SOCK_OPS,
3578                                                 BPF_CGROUP_SOCK_OPS),
3579         BPF_APROG_SEC("sk_skb/stream_parser",   BPF_PROG_TYPE_SK_SKB,
3580                                                 BPF_SK_SKB_STREAM_PARSER),
3581         BPF_APROG_SEC("sk_skb/stream_verdict",  BPF_PROG_TYPE_SK_SKB,
3582                                                 BPF_SK_SKB_STREAM_VERDICT),
3583         BPF_APROG_COMPAT("sk_skb",              BPF_PROG_TYPE_SK_SKB),
3584         BPF_APROG_SEC("sk_msg",                 BPF_PROG_TYPE_SK_MSG,
3585                                                 BPF_SK_MSG_VERDICT),
3586         BPF_APROG_SEC("lirc_mode2",             BPF_PROG_TYPE_LIRC_MODE2,
3587                                                 BPF_LIRC_MODE2),
3588         BPF_APROG_SEC("flow_dissector",         BPF_PROG_TYPE_FLOW_DISSECTOR,
3589                                                 BPF_FLOW_DISSECTOR),
3590         BPF_EAPROG_SEC("cgroup/bind4",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3591                                                 BPF_CGROUP_INET4_BIND),
3592         BPF_EAPROG_SEC("cgroup/bind6",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3593                                                 BPF_CGROUP_INET6_BIND),
3594         BPF_EAPROG_SEC("cgroup/connect4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3595                                                 BPF_CGROUP_INET4_CONNECT),
3596         BPF_EAPROG_SEC("cgroup/connect6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3597                                                 BPF_CGROUP_INET6_CONNECT),
3598         BPF_EAPROG_SEC("cgroup/sendmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3599                                                 BPF_CGROUP_UDP4_SENDMSG),
3600         BPF_EAPROG_SEC("cgroup/sendmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3601                                                 BPF_CGROUP_UDP6_SENDMSG),
3602         BPF_EAPROG_SEC("cgroup/recvmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3603                                                 BPF_CGROUP_UDP4_RECVMSG),
3604         BPF_EAPROG_SEC("cgroup/recvmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
3605                                                 BPF_CGROUP_UDP6_RECVMSG),
3606         BPF_EAPROG_SEC("cgroup/sysctl",         BPF_PROG_TYPE_CGROUP_SYSCTL,
3607                                                 BPF_CGROUP_SYSCTL),
3608         BPF_EAPROG_SEC("cgroup/getsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
3609                                                 BPF_CGROUP_GETSOCKOPT),
3610         BPF_EAPROG_SEC("cgroup/setsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
3611                                                 BPF_CGROUP_SETSOCKOPT),
3612 };
3613
3614 #undef BPF_PROG_SEC_IMPL
3615 #undef BPF_PROG_SEC
3616 #undef BPF_APROG_SEC
3617 #undef BPF_EAPROG_SEC
3618 #undef BPF_APROG_COMPAT
3619
3620 #define MAX_TYPE_NAME_SIZE 32
3621
3622 static char *libbpf_get_type_names(bool attach_type)
3623 {
3624         int i, len = ARRAY_SIZE(section_names) * MAX_TYPE_NAME_SIZE;
3625         char *buf;
3626
3627         buf = malloc(len);
3628         if (!buf)
3629                 return NULL;
3630
3631         buf[0] = '\0';
3632         /* Forge string buf with all available names */
3633         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
3634                 if (attach_type && !section_names[i].is_attachable)
3635                         continue;
3636
3637                 if (strlen(buf) + strlen(section_names[i].sec) + 2 > len) {
3638                         free(buf);
3639                         return NULL;
3640                 }
3641                 strcat(buf, " ");
3642                 strcat(buf, section_names[i].sec);
3643         }
3644
3645         return buf;
3646 }
3647
3648 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
3649                              enum bpf_attach_type *expected_attach_type)
3650 {
3651         char *type_names;
3652         int i;
3653
3654         if (!name)
3655                 return -EINVAL;
3656
3657         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
3658                 if (strncmp(name, section_names[i].sec, section_names[i].len))
3659                         continue;
3660                 *prog_type = section_names[i].prog_type;
3661                 *expected_attach_type = section_names[i].expected_attach_type;
3662                 return 0;
3663         }
3664         pr_warning("failed to guess program type based on ELF section name '%s'\n", name);
3665         type_names = libbpf_get_type_names(false);
3666         if (type_names != NULL) {
3667                 pr_info("supported section(type) names are:%s\n", type_names);
3668                 free(type_names);
3669         }
3670
3671         return -EINVAL;
3672 }
3673
3674 int libbpf_attach_type_by_name(const char *name,
3675                                enum bpf_attach_type *attach_type)
3676 {
3677         char *type_names;
3678         int i;
3679
3680         if (!name)
3681                 return -EINVAL;
3682
3683         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
3684                 if (strncmp(name, section_names[i].sec, section_names[i].len))
3685                         continue;
3686                 if (!section_names[i].is_attachable)
3687                         return -EINVAL;
3688                 *attach_type = section_names[i].attach_type;
3689                 return 0;
3690         }
3691         pr_warning("failed to guess attach type based on ELF section name '%s'\n", name);
3692         type_names = libbpf_get_type_names(true);
3693         if (type_names != NULL) {
3694                 pr_info("attachable section(type) names are:%s\n", type_names);
3695                 free(type_names);
3696         }
3697
3698         return -EINVAL;
3699 }
3700
3701 static int
3702 bpf_program__identify_section(struct bpf_program *prog,
3703                               enum bpf_prog_type *prog_type,
3704                               enum bpf_attach_type *expected_attach_type)
3705 {
3706         return libbpf_prog_type_by_name(prog->section_name, prog_type,
3707                                         expected_attach_type);
3708 }
3709
3710 int bpf_map__fd(const struct bpf_map *map)
3711 {
3712         return map ? map->fd : -EINVAL;
3713 }
3714
3715 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
3716 {
3717         return map ? &map->def : ERR_PTR(-EINVAL);
3718 }
3719
3720 const char *bpf_map__name(const struct bpf_map *map)
3721 {
3722         return map ? map->name : NULL;
3723 }
3724
3725 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
3726 {
3727         return map ? map->btf_key_type_id : 0;
3728 }
3729
3730 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
3731 {
3732         return map ? map->btf_value_type_id : 0;
3733 }
3734
3735 int bpf_map__set_priv(struct bpf_map *map, void *priv,
3736                      bpf_map_clear_priv_t clear_priv)
3737 {
3738         if (!map)
3739                 return -EINVAL;
3740
3741         if (map->priv) {
3742                 if (map->clear_priv)
3743                         map->clear_priv(map, map->priv);
3744         }
3745
3746         map->priv = priv;
3747         map->clear_priv = clear_priv;
3748         return 0;
3749 }
3750
3751 void *bpf_map__priv(const struct bpf_map *map)
3752 {
3753         return map ? map->priv : ERR_PTR(-EINVAL);
3754 }
3755
3756 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
3757 {
3758         return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
3759 }
3760
3761 bool bpf_map__is_internal(const struct bpf_map *map)
3762 {
3763         return map->libbpf_type != LIBBPF_MAP_UNSPEC;
3764 }
3765
3766 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
3767 {
3768         map->map_ifindex = ifindex;
3769 }
3770
3771 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
3772 {
3773         if (!bpf_map_type__is_map_in_map(map->def.type)) {
3774                 pr_warning("error: unsupported map type\n");
3775                 return -EINVAL;
3776         }
3777         if (map->inner_map_fd != -1) {
3778                 pr_warning("error: inner_map_fd already specified\n");
3779                 return -EINVAL;
3780         }
3781         map->inner_map_fd = fd;
3782         return 0;
3783 }
3784
3785 static struct bpf_map *
3786 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
3787 {
3788         ssize_t idx;
3789         struct bpf_map *s, *e;
3790
3791         if (!obj || !obj->maps)
3792                 return NULL;
3793
3794         s = obj->maps;
3795         e = obj->maps + obj->nr_maps;
3796
3797         if ((m < s) || (m >= e)) {
3798                 pr_warning("error in %s: map handler doesn't belong to object\n",
3799                            __func__);
3800                 return NULL;
3801         }
3802
3803         idx = (m - obj->maps) + i;
3804         if (idx >= obj->nr_maps || idx < 0)
3805                 return NULL;
3806         return &obj->maps[idx];
3807 }
3808
3809 struct bpf_map *
3810 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
3811 {
3812         if (prev == NULL)
3813                 return obj->maps;
3814
3815         return __bpf_map__iter(prev, obj, 1);
3816 }
3817
3818 struct bpf_map *
3819 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
3820 {
3821         if (next == NULL) {
3822                 if (!obj->nr_maps)
3823                         return NULL;
3824                 return obj->maps + obj->nr_maps - 1;
3825         }
3826
3827         return __bpf_map__iter(next, obj, -1);
3828 }
3829
3830 struct bpf_map *
3831 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
3832 {
3833         struct bpf_map *pos;
3834
3835         bpf_object__for_each_map(pos, obj) {
3836                 if (pos->name && !strcmp(pos->name, name))
3837                         return pos;
3838         }
3839         return NULL;
3840 }
3841
3842 int
3843 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
3844 {
3845         return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
3846 }
3847
3848 struct bpf_map *
3849 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
3850 {
3851         return ERR_PTR(-ENOTSUP);
3852 }
3853
3854 long libbpf_get_error(const void *ptr)
3855 {
3856         return PTR_ERR_OR_ZERO(ptr);
3857 }
3858
3859 int bpf_prog_load(const char *file, enum bpf_prog_type type,
3860                   struct bpf_object **pobj, int *prog_fd)
3861 {
3862         struct bpf_prog_load_attr attr;
3863
3864         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
3865         attr.file = file;
3866         attr.prog_type = type;
3867         attr.expected_attach_type = 0;
3868
3869         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
3870 }
3871
3872 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
3873                         struct bpf_object **pobj, int *prog_fd)
3874 {
3875         struct bpf_object_open_attr open_attr = {
3876                 .file           = attr->file,
3877                 .prog_type      = attr->prog_type,
3878         };
3879         struct bpf_program *prog, *first_prog = NULL;
3880         enum bpf_attach_type expected_attach_type;
3881         enum bpf_prog_type prog_type;
3882         struct bpf_object *obj;
3883         struct bpf_map *map;
3884         int err;
3885
3886         if (!attr)
3887                 return -EINVAL;
3888         if (!attr->file)
3889                 return -EINVAL;
3890
3891         obj = bpf_object__open_xattr(&open_attr);
3892         if (IS_ERR_OR_NULL(obj))
3893                 return -ENOENT;
3894
3895         bpf_object__for_each_program(prog, obj) {
3896                 /*
3897                  * If type is not specified, try to guess it based on
3898                  * section name.
3899                  */
3900                 prog_type = attr->prog_type;
3901                 prog->prog_ifindex = attr->ifindex;
3902                 expected_attach_type = attr->expected_attach_type;
3903                 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
3904                         err = bpf_program__identify_section(prog, &prog_type,
3905                                                             &expected_attach_type);
3906                         if (err < 0) {
3907                                 bpf_object__close(obj);
3908                                 return -EINVAL;
3909                         }
3910                 }
3911
3912                 bpf_program__set_type(prog, prog_type);
3913                 bpf_program__set_expected_attach_type(prog,
3914                                                       expected_attach_type);
3915
3916                 prog->log_level = attr->log_level;
3917                 prog->prog_flags = attr->prog_flags;
3918                 if (!first_prog)
3919                         first_prog = prog;
3920         }
3921
3922         bpf_object__for_each_map(map, obj) {
3923                 if (!bpf_map__is_offload_neutral(map))
3924                         map->map_ifindex = attr->ifindex;
3925         }
3926
3927         if (!first_prog) {
3928                 pr_warning("object file doesn't contain bpf program\n");
3929                 bpf_object__close(obj);
3930                 return -ENOENT;
3931         }
3932
3933         err = bpf_object__load(obj);
3934         if (err) {
3935                 bpf_object__close(obj);
3936                 return -EINVAL;
3937         }
3938
3939         *pobj = obj;
3940         *prog_fd = bpf_program__fd(first_prog);
3941         return 0;
3942 }
3943
3944 enum bpf_perf_event_ret
3945 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
3946                            void **copy_mem, size_t *copy_size,
3947                            bpf_perf_event_print_t fn, void *private_data)
3948 {
3949         struct perf_event_mmap_page *header = mmap_mem;
3950         __u64 data_head = ring_buffer_read_head(header);
3951         __u64 data_tail = header->data_tail;
3952         void *base = ((__u8 *)header) + page_size;
3953         int ret = LIBBPF_PERF_EVENT_CONT;
3954         struct perf_event_header *ehdr;
3955         size_t ehdr_size;
3956
3957         while (data_head != data_tail) {
3958                 ehdr = base + (data_tail & (mmap_size - 1));
3959                 ehdr_size = ehdr->size;
3960
3961                 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
3962                         void *copy_start = ehdr;
3963                         size_t len_first = base + mmap_size - copy_start;
3964                         size_t len_secnd = ehdr_size - len_first;
3965
3966                         if (*copy_size < ehdr_size) {
3967                                 free(*copy_mem);
3968                                 *copy_mem = malloc(ehdr_size);
3969                                 if (!*copy_mem) {
3970                                         *copy_size = 0;
3971                                         ret = LIBBPF_PERF_EVENT_ERROR;
3972                                         break;
3973                                 }
3974                                 *copy_size = ehdr_size;
3975                         }
3976
3977                         memcpy(*copy_mem, copy_start, len_first);
3978                         memcpy(*copy_mem + len_first, base, len_secnd);
3979                         ehdr = *copy_mem;
3980                 }
3981
3982                 ret = fn(ehdr, private_data);
3983                 data_tail += ehdr_size;
3984                 if (ret != LIBBPF_PERF_EVENT_CONT)
3985                         break;
3986         }
3987
3988         ring_buffer_write_tail(header, data_tail);
3989         return ret;
3990 }
3991
3992 struct bpf_prog_info_array_desc {
3993         int     array_offset;   /* e.g. offset of jited_prog_insns */
3994         int     count_offset;   /* e.g. offset of jited_prog_len */
3995         int     size_offset;    /* > 0: offset of rec size,
3996                                  * < 0: fix size of -size_offset
3997                                  */
3998 };
3999
4000 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
4001         [BPF_PROG_INFO_JITED_INSNS] = {
4002                 offsetof(struct bpf_prog_info, jited_prog_insns),
4003                 offsetof(struct bpf_prog_info, jited_prog_len),
4004                 -1,
4005         },
4006         [BPF_PROG_INFO_XLATED_INSNS] = {
4007                 offsetof(struct bpf_prog_info, xlated_prog_insns),
4008                 offsetof(struct bpf_prog_info, xlated_prog_len),
4009                 -1,
4010         },
4011         [BPF_PROG_INFO_MAP_IDS] = {
4012                 offsetof(struct bpf_prog_info, map_ids),
4013                 offsetof(struct bpf_prog_info, nr_map_ids),
4014                 -(int)sizeof(__u32),
4015         },
4016         [BPF_PROG_INFO_JITED_KSYMS] = {
4017                 offsetof(struct bpf_prog_info, jited_ksyms),
4018                 offsetof(struct bpf_prog_info, nr_jited_ksyms),
4019                 -(int)sizeof(__u64),
4020         },
4021         [BPF_PROG_INFO_JITED_FUNC_LENS] = {
4022                 offsetof(struct bpf_prog_info, jited_func_lens),
4023                 offsetof(struct bpf_prog_info, nr_jited_func_lens),
4024                 -(int)sizeof(__u32),
4025         },
4026         [BPF_PROG_INFO_FUNC_INFO] = {
4027                 offsetof(struct bpf_prog_info, func_info),
4028                 offsetof(struct bpf_prog_info, nr_func_info),
4029                 offsetof(struct bpf_prog_info, func_info_rec_size),
4030         },
4031         [BPF_PROG_INFO_LINE_INFO] = {
4032                 offsetof(struct bpf_prog_info, line_info),
4033                 offsetof(struct bpf_prog_info, nr_line_info),
4034                 offsetof(struct bpf_prog_info, line_info_rec_size),
4035         },
4036         [BPF_PROG_INFO_JITED_LINE_INFO] = {
4037                 offsetof(struct bpf_prog_info, jited_line_info),
4038                 offsetof(struct bpf_prog_info, nr_jited_line_info),
4039                 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
4040         },
4041         [BPF_PROG_INFO_PROG_TAGS] = {
4042                 offsetof(struct bpf_prog_info, prog_tags),
4043                 offsetof(struct bpf_prog_info, nr_prog_tags),
4044                 -(int)sizeof(__u8) * BPF_TAG_SIZE,
4045         },
4046
4047 };
4048
4049 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info, int offset)
4050 {
4051         __u32 *array = (__u32 *)info;
4052
4053         if (offset >= 0)
4054                 return array[offset / sizeof(__u32)];
4055         return -(int)offset;
4056 }
4057
4058 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info, int offset)
4059 {
4060         __u64 *array = (__u64 *)info;
4061
4062         if (offset >= 0)
4063                 return array[offset / sizeof(__u64)];
4064         return -(int)offset;
4065 }
4066
4067 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
4068                                          __u32 val)
4069 {
4070         __u32 *array = (__u32 *)info;
4071
4072         if (offset >= 0)
4073                 array[offset / sizeof(__u32)] = val;
4074 }
4075
4076 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
4077                                          __u64 val)
4078 {
4079         __u64 *array = (__u64 *)info;
4080
4081         if (offset >= 0)
4082                 array[offset / sizeof(__u64)] = val;
4083 }
4084
4085 struct bpf_prog_info_linear *
4086 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
4087 {
4088         struct bpf_prog_info_linear *info_linear;
4089         struct bpf_prog_info info = {};
4090         __u32 info_len = sizeof(info);
4091         __u32 data_len = 0;
4092         int i, err;
4093         void *ptr;
4094
4095         if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
4096                 return ERR_PTR(-EINVAL);
4097
4098         /* step 1: get array dimensions */
4099         err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
4100         if (err) {
4101                 pr_debug("can't get prog info: %s", strerror(errno));
4102                 return ERR_PTR(-EFAULT);
4103         }
4104
4105         /* step 2: calculate total size of all arrays */
4106         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
4107                 bool include_array = (arrays & (1UL << i)) > 0;
4108                 struct bpf_prog_info_array_desc *desc;
4109                 __u32 count, size;
4110
4111                 desc = bpf_prog_info_array_desc + i;
4112
4113                 /* kernel is too old to support this field */
4114                 if (info_len < desc->array_offset + sizeof(__u32) ||
4115                     info_len < desc->count_offset + sizeof(__u32) ||
4116                     (desc->size_offset > 0 && info_len < desc->size_offset))
4117                         include_array = false;
4118
4119                 if (!include_array) {
4120                         arrays &= ~(1UL << i);  /* clear the bit */
4121                         continue;
4122                 }
4123
4124                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
4125                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
4126
4127                 data_len += count * size;
4128         }
4129
4130         /* step 3: allocate continuous memory */
4131         data_len = roundup(data_len, sizeof(__u64));
4132         info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
4133         if (!info_linear)
4134                 return ERR_PTR(-ENOMEM);
4135
4136         /* step 4: fill data to info_linear->info */
4137         info_linear->arrays = arrays;
4138         memset(&info_linear->info, 0, sizeof(info));
4139         ptr = info_linear->data;
4140
4141         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
4142                 struct bpf_prog_info_array_desc *desc;
4143                 __u32 count, size;
4144
4145                 if ((arrays & (1UL << i)) == 0)
4146                         continue;
4147
4148                 desc  = bpf_prog_info_array_desc + i;
4149                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
4150                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
4151                 bpf_prog_info_set_offset_u32(&info_linear->info,
4152                                              desc->count_offset, count);
4153                 bpf_prog_info_set_offset_u32(&info_linear->info,
4154                                              desc->size_offset, size);
4155                 bpf_prog_info_set_offset_u64(&info_linear->info,
4156                                              desc->array_offset,
4157                                              ptr_to_u64(ptr));
4158                 ptr += count * size;
4159         }
4160
4161         /* step 5: call syscall again to get required arrays */
4162         err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
4163         if (err) {
4164                 pr_debug("can't get prog info: %s", strerror(errno));
4165                 free(info_linear);
4166                 return ERR_PTR(-EFAULT);
4167         }
4168
4169         /* step 6: verify the data */
4170         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
4171                 struct bpf_prog_info_array_desc *desc;
4172                 __u32 v1, v2;
4173
4174                 if ((arrays & (1UL << i)) == 0)
4175                         continue;
4176
4177                 desc = bpf_prog_info_array_desc + i;
4178                 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
4179                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
4180                                                    desc->count_offset);
4181                 if (v1 != v2)
4182                         pr_warning("%s: mismatch in element count\n", __func__);
4183
4184                 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
4185                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
4186                                                    desc->size_offset);
4187                 if (v1 != v2)
4188                         pr_warning("%s: mismatch in rec size\n", __func__);
4189         }
4190
4191         /* step 7: update info_len and data_len */
4192         info_linear->info_len = sizeof(struct bpf_prog_info);
4193         info_linear->data_len = data_len;
4194
4195         return info_linear;
4196 }
4197
4198 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
4199 {
4200         int i;
4201
4202         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
4203                 struct bpf_prog_info_array_desc *desc;
4204                 __u64 addr, offs;
4205
4206                 if ((info_linear->arrays & (1UL << i)) == 0)
4207                         continue;
4208
4209                 desc = bpf_prog_info_array_desc + i;
4210                 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
4211                                                      desc->array_offset);
4212                 offs = addr - ptr_to_u64(info_linear->data);
4213                 bpf_prog_info_set_offset_u64(&info_linear->info,
4214                                              desc->array_offset, offs);
4215         }
4216 }
4217
4218 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
4219 {
4220         int i;
4221
4222         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
4223                 struct bpf_prog_info_array_desc *desc;
4224                 __u64 addr, offs;
4225
4226                 if ((info_linear->arrays & (1UL << i)) == 0)
4227                         continue;
4228
4229                 desc = bpf_prog_info_array_desc + i;
4230                 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
4231                                                      desc->array_offset);
4232                 addr = offs + ptr_to_u64(info_linear->data);
4233                 bpf_prog_info_set_offset_u64(&info_linear->info,
4234                                              desc->array_offset, addr);
4235         }
4236 }
4237
4238 int libbpf_num_possible_cpus(void)
4239 {
4240         static const char *fcpu = "/sys/devices/system/cpu/possible";
4241         int len = 0, n = 0, il = 0, ir = 0;
4242         unsigned int start = 0, end = 0;
4243         static int cpus;
4244         char buf[128];
4245         int error = 0;
4246         int fd = -1;
4247
4248         if (cpus > 0)
4249                 return cpus;
4250
4251         fd = open(fcpu, O_RDONLY);
4252         if (fd < 0) {
4253                 error = errno;
4254                 pr_warning("Failed to open file %s: %s\n",
4255                            fcpu, strerror(error));
4256                 return -error;
4257         }
4258         len = read(fd, buf, sizeof(buf));
4259         close(fd);
4260         if (len <= 0) {
4261                 error = len ? errno : EINVAL;
4262                 pr_warning("Failed to read # of possible cpus from %s: %s\n",
4263                            fcpu, strerror(error));
4264                 return -error;
4265         }
4266         if (len == sizeof(buf)) {
4267                 pr_warning("File %s size overflow\n", fcpu);
4268                 return -EOVERFLOW;
4269         }
4270         buf[len] = '\0';
4271
4272         for (ir = 0, cpus = 0; ir <= len; ir++) {
4273                 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
4274                 if (buf[ir] == ',' || buf[ir] == '\0') {
4275                         buf[ir] = '\0';
4276                         n = sscanf(&buf[il], "%u-%u", &start, &end);
4277                         if (n <= 0) {
4278                                 pr_warning("Failed to get # CPUs from %s\n",
4279                                            &buf[il]);
4280                                 return -EINVAL;
4281                         } else if (n == 1) {
4282                                 end = start;
4283                         }
4284                         cpus += end - start + 1;
4285                         il = ir + 1;
4286                 }
4287         }
4288         if (cpus <= 0) {
4289                 pr_warning("Invalid #CPUs %d from %s\n", cpus, fcpu);
4290                 return -EINVAL;
4291         }
4292         return cpus;
4293 }