kernel/bpf/syscall.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   3  */
   4 #include <linux/bpf.h>
   5 #include <linux/bpf-cgroup.h>
   6 #include <linux/bpf_trace.h>
   7 #include <linux/bpf_lirc.h>
   8 #include <linux/bpf_verifier.h>
   9 #include <linux/bsearch.h>
  10 #include <linux/btf.h>
  11 #include <linux/syscalls.h>
  12 #include <linux/slab.h>
  13 #include <linux/sched/signal.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/mmzone.h>
  16 #include <linux/anon_inodes.h>
  17 #include <linux/fdtable.h>
  18 #include <linux/file.h>
  19 #include <linux/fs.h>
  20 #include <linux/license.h>
  21 #include <linux/filter.h>
  22 #include <linux/kernel.h>
  23 #include <linux/idr.h>
  24 #include <linux/cred.h>
  25 #include <linux/timekeeping.h>
  26 #include <linux/ctype.h>
  27 #include <linux/nospec.h>
  28 #include <linux/audit.h>
  29 #include <uapi/linux/btf.h>
  30 #include <linux/pgtable.h>
  31 #include <linux/bpf_lsm.h>
  32 #include <linux/poll.h>
  33 #include <linux/sort.h>
  34 #include <linux/bpf-netns.h>
  35 #include <linux/rcupdate_trace.h>
  36 #include <linux/memcontrol.h>
  37 #include <linux/trace_events.h>
  38 #include <linux/tracepoint.h>
  39
  40 #include <net/netfilter/nf_bpf_link.h>
  41 #include <net/netkit.h>
  42 #include <net/tcx.h>
  43
  44 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  45                           (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  46                           (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  47 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
  48 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  49 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
  50                         IS_FD_HASH(map))
  51
  52 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  53
  54 DEFINE_PER_CPU(int, bpf_prog_active);
  55 static DEFINE_IDR(prog_idr);
  56 static DEFINE_SPINLOCK(prog_idr_lock);
  57 static DEFINE_IDR(map_idr);
  58 static DEFINE_SPINLOCK(map_idr_lock);
  59 static DEFINE_IDR(link_idr);
  60 static DEFINE_SPINLOCK(link_idr_lock);
  61
  62 int sysctl_unprivileged_bpf_disabled __read_mostly =
  63         IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
  64
  65 static const struct bpf_map_ops * const bpf_map_types[] = {
  66 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
  67 #define BPF_MAP_TYPE(_id, _ops) \
  68         [_id] = &_ops,
  69 #define BPF_LINK_TYPE(_id, _name)
  70 #include <linux/bpf_types.h>
  71 #undef BPF_PROG_TYPE
  72 #undef BPF_MAP_TYPE
  73 #undef BPF_LINK_TYPE
  74 };
  75
  76 /*
  77  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  78  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  79  * we don't know about yet.
  80  *
  81  * There is a ToCToU between this function call and the following
  82  * copy_from_user() call. However, this is not a concern since this function is
  83  * meant to be a future-proofing of bits.
  84  */
  85 int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
  86                              size_t expected_size,
  87                              size_t actual_size)
  88 {
  89         int res;
  90
  91         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  92                 return -E2BIG;
  93
  94         if (actual_size <= expected_size)
  95                 return 0;
  96
  97         if (uaddr.is_kernel)
  98                 res = memchr_inv(uaddr.kernel + expected_size, 0,
  99                                  actual_size - expected_size) == NULL;
 100         else
 101                 res = check_zeroed_user(uaddr.user + expected_size,
 102                                         actual_size - expected_size);
 103         if (res < 0)
 104                 return res;
 105         return res ? 0 : -E2BIG;
 106 }
 107
 108 const struct bpf_map_ops bpf_map_offload_ops = {
 109         .map_meta_equal = bpf_map_meta_equal,
 110         .map_alloc = bpf_map_offload_map_alloc,
 111         .map_free = bpf_map_offload_map_free,
 112         .map_check_btf = map_check_no_btf,
 113         .map_mem_usage = bpf_map_offload_map_mem_usage,
 114 };
 115
 116 static void bpf_map_write_active_inc(struct bpf_map *map)
 117 {
 118         atomic64_inc(&map->writecnt);
 119 }
 120
 121 static void bpf_map_write_active_dec(struct bpf_map *map)
 122 {
 123         atomic64_dec(&map->writecnt);
 124 }
 125
 126 bool bpf_map_write_active(const struct bpf_map *map)
 127 {
 128         return atomic64_read(&map->writecnt) != 0;
 129 }
 130
 131 static u32 bpf_map_value_size(const struct bpf_map *map)
 132 {
 133         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 134             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 135             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 136             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 137                 return round_up(map->value_size, 8) * num_possible_cpus();
 138         else if (IS_FD_MAP(map))
 139                 return sizeof(u32);
 140         else
 141                 return  map->value_size;
 142 }
 143
 144 static void maybe_wait_bpf_programs(struct bpf_map *map)
 145 {
 146         /* Wait for any running non-sleepable BPF programs to complete so that
 147          * userspace, when we return to it, knows that all non-sleepable
 148          * programs that could be running use the new map value. For sleepable
 149          * BPF programs, synchronize_rcu_tasks_trace() should be used to wait
 150          * for the completions of these programs, but considering the waiting
 151          * time can be very long and userspace may think it will hang forever,
 152          * so don't handle sleepable BPF programs now.
 153          */
 154         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 155             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 156                 synchronize_rcu();
 157 }
 158
 159 static void unpin_uptr_kaddr(void *kaddr)
 160 {
 161         if (kaddr)
 162                 unpin_user_page(virt_to_page(kaddr));
 163 }
 164
 165 static void __bpf_obj_unpin_uptrs(struct btf_record *rec, u32 cnt, void *obj)
 166 {
 167         const struct btf_field *field;
 168         void **uptr_addr;
 169         int i;
 170
 171         for (i = 0, field = rec->fields; i < cnt; i++, field++) {
 172                 if (field->type != BPF_UPTR)
 173                         continue;
 174
 175                 uptr_addr = obj + field->offset;
 176                 unpin_uptr_kaddr(*uptr_addr);
 177         }
 178 }
 179
 180 static void bpf_obj_unpin_uptrs(struct btf_record *rec, void *obj)
 181 {
 182         if (!btf_record_has_field(rec, BPF_UPTR))
 183                 return;
 184
 185         __bpf_obj_unpin_uptrs(rec, rec->cnt, obj);
 186 }
 187
 188 static int bpf_obj_pin_uptrs(struct btf_record *rec, void *obj)
 189 {
 190         const struct btf_field *field;
 191         const struct btf_type *t;
 192         unsigned long start, end;
 193         struct page *page;
 194         void **uptr_addr;
 195         int i, err;
 196
 197         if (!btf_record_has_field(rec, BPF_UPTR))
 198                 return 0;
 199
 200         for (i = 0, field = rec->fields; i < rec->cnt; i++, field++) {
 201                 if (field->type != BPF_UPTR)
 202                         continue;
 203
 204                 uptr_addr = obj + field->offset;
 205                 start = *(unsigned long *)uptr_addr;
 206                 if (!start)
 207                         continue;
 208
 209                 t = btf_type_by_id(field->kptr.btf, field->kptr.btf_id);
 210                 /* t->size was checked for zero before */
 211                 if (check_add_overflow(start, t->size - 1, &end)) {
 212                         err = -EFAULT;
 213                         goto unpin_all;
 214                 }
 215
 216                 /* The uptr's struct cannot span across two pages */
 217                 if ((start & PAGE_MASK) != (end & PAGE_MASK)) {
 218                         err = -EOPNOTSUPP;
 219                         goto unpin_all;
 220                 }
 221
 222                 err = pin_user_pages_fast(start, 1, FOLL_LONGTERM | FOLL_WRITE, &page);
 223                 if (err != 1)
 224                         goto unpin_all;
 225
 226                 if (PageHighMem(page)) {
 227                         err = -EOPNOTSUPP;
 228                         unpin_user_page(page);
 229                         goto unpin_all;
 230                 }
 231
 232                 *uptr_addr = page_address(page) + offset_in_page(start);
 233         }
 234
 235         return 0;
 236
 237 unpin_all:
 238         __bpf_obj_unpin_uptrs(rec, i, obj);
 239         return err;
 240 }
 241
 242 static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
 243                                 void *key, void *value, __u64 flags)
 244 {
 245         int err;
 246
 247         /* Need to create a kthread, thus must support schedule */
 248         if (bpf_map_is_offloaded(map)) {
 249                 return bpf_map_offload_update_elem(map, key, value, flags);
 250         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 251                    map->map_type == BPF_MAP_TYPE_ARENA ||
 252                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 253                 return map->ops->map_update_elem(map, key, value, flags);
 254         } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 255                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 256                 return sock_map_update_elem_sys(map, key, value, flags);
 257         } else if (IS_FD_PROG_ARRAY(map)) {
 258                 return bpf_fd_array_map_update_elem(map, map_file, key, value,
 259                                                     flags);
 260         }
 261
 262         bpf_disable_instrumentation();
 263         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 264             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 265                 err = bpf_percpu_hash_update(map, key, value, flags);
 266         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 267                 err = bpf_percpu_array_update(map, key, value, flags);
 268         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 269                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 270                                                        flags);
 271         } else if (IS_FD_ARRAY(map)) {
 272                 err = bpf_fd_array_map_update_elem(map, map_file, key, value,
 273                                                    flags);
 274         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 275                 err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
 276                                                   flags);
 277         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 278                 /* rcu_read_lock() is not needed */
 279                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 280                                                          flags);
 281         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 282                    map->map_type == BPF_MAP_TYPE_STACK ||
 283                    map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 284                 err = map->ops->map_push_elem(map, value, flags);
 285         } else {
 286                 err = bpf_obj_pin_uptrs(map->record, value);
 287                 if (!err) {
 288                         rcu_read_lock();
 289                         err = map->ops->map_update_elem(map, key, value, flags);
 290                         rcu_read_unlock();
 291                         if (err)
 292                                 bpf_obj_unpin_uptrs(map->record, value);
 293                 }
 294         }
 295         bpf_enable_instrumentation();
 296
 297         return err;
 298 }
 299
 300 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 301                               __u64 flags)
 302 {
 303         void *ptr;
 304         int err;
 305
 306         if (bpf_map_is_offloaded(map))
 307                 return bpf_map_offload_lookup_elem(map, key, value);
 308
 309         bpf_disable_instrumentation();
 310         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 311             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 312                 err = bpf_percpu_hash_copy(map, key, value);
 313         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 314                 err = bpf_percpu_array_copy(map, key, value);
 315         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 316                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 317         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 318                 err = bpf_stackmap_copy(map, key, value);
 319         } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 320                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 321         } else if (IS_FD_HASH(map)) {
 322                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 323         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 324                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 325         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 326                    map->map_type == BPF_MAP_TYPE_STACK ||
 327                    map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 328                 err = map->ops->map_peek_elem(map, value);
 329         } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 330                 /* struct_ops map requires directly updating "value" */
 331                 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
 332         } else {
 333                 rcu_read_lock();
 334                 if (map->ops->map_lookup_elem_sys_only)
 335                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 336                 else
 337                         ptr = map->ops->map_lookup_elem(map, key);
 338                 if (IS_ERR(ptr)) {
 339                         err = PTR_ERR(ptr);
 340                 } else if (!ptr) {
 341                         err = -ENOENT;
 342                 } else {
 343                         err = 0;
 344                         if (flags & BPF_F_LOCK)
 345                                 /* lock 'ptr' and copy everything but lock */
 346                                 copy_map_value_locked(map, value, ptr, true);
 347                         else
 348                                 copy_map_value(map, value, ptr);
 349                         /* mask lock and timer, since value wasn't zero inited */
 350                         check_and_init_map_value(map, value);
 351                 }
 352                 rcu_read_unlock();
 353         }
 354
 355         bpf_enable_instrumentation();
 356
 357         return err;
 358 }
 359
 360 /* Please, do not use this function outside from the map creation path
 361  * (e.g. in map update path) without taking care of setting the active
 362  * memory cgroup (see at bpf_map_kmalloc_node() for example).
 363  */
 364 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 365 {
 366         /* We really just want to fail instead of triggering OOM killer
 367          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 368          * which is used for lower order allocation requests.
 369          *
 370          * It has been observed that higher order allocation requests done by
 371          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 372          * to reclaim memory from the page cache, thus we set
 373          * __GFP_RETRY_MAYFAIL to avoid such situations.
 374          */
 375
 376         gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
 377         unsigned int flags = 0;
 378         unsigned long align = 1;
 379         void *area;
 380
 381         if (size >= SIZE_MAX)
 382                 return NULL;
 383
 384         /* kmalloc()'ed memory can't be mmap()'ed */
 385         if (mmapable) {
 386                 BUG_ON(!PAGE_ALIGNED(size));
 387                 align = SHMLBA;
 388                 flags = VM_USERMAP;
 389         } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 390                 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
 391                                     numa_node);
 392                 if (area != NULL)
 393                         return area;
 394         }
 395
 396         return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
 397                         gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
 398                         flags, numa_node, __builtin_return_address(0));
 399 }
 400
 401 void *bpf_map_area_alloc(u64 size, int numa_node)
 402 {
 403         return __bpf_map_area_alloc(size, numa_node, false);
 404 }
 405
 406 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 407 {
 408         return __bpf_map_area_alloc(size, numa_node, true);
 409 }
 410
 411 void bpf_map_area_free(void *area)
 412 {
 413         kvfree(area);
 414 }
 415
 416 static u32 bpf_map_flags_retain_permanent(u32 flags)
 417 {
 418         /* Some map creation flags are not tied to the map object but
 419          * rather to the map fd instead, so they have no meaning upon
 420          * map object inspection since multiple file descriptors with
 421          * different (access) properties can exist here. Thus, given
 422          * this has zero meaning for the map itself, lets clear these
 423          * from here.
 424          */
 425         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 426 }
 427
 428 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 429 {
 430         map->map_type = attr->map_type;
 431         map->key_size = attr->key_size;
 432         map->value_size = attr->value_size;
 433         map->max_entries = attr->max_entries;
 434         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 435         map->numa_node = bpf_map_attr_numa_node(attr);
 436         map->map_extra = attr->map_extra;
 437 }
 438
 439 static int bpf_map_alloc_id(struct bpf_map *map)
 440 {
 441         int id;
 442
 443         idr_preload(GFP_KERNEL);
 444         spin_lock_bh(&map_idr_lock);
 445         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 446         if (id > 0)
 447                 map->id = id;
 448         spin_unlock_bh(&map_idr_lock);
 449         idr_preload_end();
 450
 451         if (WARN_ON_ONCE(!id))
 452                 return -ENOSPC;
 453
 454         return id > 0 ? 0 : id;
 455 }
 456
 457 void bpf_map_free_id(struct bpf_map *map)
 458 {
 459         unsigned long flags;
 460
 461         /* Offloaded maps are removed from the IDR store when their device
 462          * disappears - even if someone holds an fd to them they are unusable,
 463          * the memory is gone, all ops will fail; they are simply waiting for
 464          * refcnt to drop to be freed.
 465          */
 466         if (!map->id)
 467                 return;
 468
 469         spin_lock_irqsave(&map_idr_lock, flags);
 470
 471         idr_remove(&map_idr, map->id);
 472         map->id = 0;
 473
 474         spin_unlock_irqrestore(&map_idr_lock, flags);
 475 }
 476
 477 #ifdef CONFIG_MEMCG
 478 static void bpf_map_save_memcg(struct bpf_map *map)
 479 {
 480         /* Currently if a map is created by a process belonging to the root
 481          * memory cgroup, get_obj_cgroup_from_current() will return NULL.
 482          * So we have to check map->objcg for being NULL each time it's
 483          * being used.
 484          */
 485         if (memcg_bpf_enabled())
 486                 map->objcg = get_obj_cgroup_from_current();
 487 }
 488
 489 static void bpf_map_release_memcg(struct bpf_map *map)
 490 {
 491         if (map->objcg)
 492                 obj_cgroup_put(map->objcg);
 493 }
 494
 495 static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
 496 {
 497         if (map->objcg)
 498                 return get_mem_cgroup_from_objcg(map->objcg);
 499
 500         return root_mem_cgroup;
 501 }
 502
 503 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
 504                            int node)
 505 {
 506         struct mem_cgroup *memcg, *old_memcg;
 507         void *ptr;
 508
 509         memcg = bpf_map_get_memcg(map);
 510         old_memcg = set_active_memcg(memcg);
 511         ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
 512         set_active_memcg(old_memcg);
 513         mem_cgroup_put(memcg);
 514
 515         return ptr;
 516 }
 517
 518 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
 519 {
 520         struct mem_cgroup *memcg, *old_memcg;
 521         void *ptr;
 522
 523         memcg = bpf_map_get_memcg(map);
 524         old_memcg = set_active_memcg(memcg);
 525         ptr = kzalloc(size, flags | __GFP_ACCOUNT);
 526         set_active_memcg(old_memcg);
 527         mem_cgroup_put(memcg);
 528
 529         return ptr;
 530 }
 531
 532 void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
 533                        gfp_t flags)
 534 {
 535         struct mem_cgroup *memcg, *old_memcg;
 536         void *ptr;
 537
 538         memcg = bpf_map_get_memcg(map);
 539         old_memcg = set_active_memcg(memcg);
 540         ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
 541         set_active_memcg(old_memcg);
 542         mem_cgroup_put(memcg);
 543
 544         return ptr;
 545 }
 546
 547 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
 548                                     size_t align, gfp_t flags)
 549 {
 550         struct mem_cgroup *memcg, *old_memcg;
 551         void __percpu *ptr;
 552
 553         memcg = bpf_map_get_memcg(map);
 554         old_memcg = set_active_memcg(memcg);
 555         ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
 556         set_active_memcg(old_memcg);
 557         mem_cgroup_put(memcg);
 558
 559         return ptr;
 560 }
 561
 562 #else
 563 static void bpf_map_save_memcg(struct bpf_map *map)
 564 {
 565 }
 566
 567 static void bpf_map_release_memcg(struct bpf_map *map)
 568 {
 569 }
 570 #endif
 571
 572 int bpf_map_alloc_pages(const struct bpf_map *map, gfp_t gfp, int nid,
 573                         unsigned long nr_pages, struct page **pages)
 574 {
 575         unsigned long i, j;
 576         struct page *pg;
 577         int ret = 0;
 578 #ifdef CONFIG_MEMCG
 579         struct mem_cgroup *memcg, *old_memcg;
 580
 581         memcg = bpf_map_get_memcg(map);
 582         old_memcg = set_active_memcg(memcg);
 583 #endif
 584         for (i = 0; i < nr_pages; i++) {
 585                 pg = alloc_pages_node(nid, gfp | __GFP_ACCOUNT, 0);
 586
 587                 if (pg) {
 588                         pages[i] = pg;
 589                         continue;
 590                 }
 591                 for (j = 0; j < i; j++)
 592                         __free_page(pages[j]);
 593                 ret = -ENOMEM;
 594                 break;
 595         }
 596
 597 #ifdef CONFIG_MEMCG
 598         set_active_memcg(old_memcg);
 599         mem_cgroup_put(memcg);
 600 #endif
 601         return ret;
 602 }
 603
 604
 605 static int btf_field_cmp(const void *a, const void *b)
 606 {
 607         const struct btf_field *f1 = a, *f2 = b;
 608
 609         if (f1->offset < f2->offset)
 610                 return -1;
 611         else if (f1->offset > f2->offset)
 612                 return 1;
 613         return 0;
 614 }
 615
 616 struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
 617                                   u32 field_mask)
 618 {
 619         struct btf_field *field;
 620
 621         if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & field_mask))
 622                 return NULL;
 623         field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
 624         if (!field || !(field->type & field_mask))
 625                 return NULL;
 626         return field;
 627 }
 628
 629 void btf_record_free(struct btf_record *rec)
 630 {
 631         int i;
 632
 633         if (IS_ERR_OR_NULL(rec))
 634                 return;
 635         for (i = 0; i < rec->cnt; i++) {
 636                 switch (rec->fields[i].type) {
 637                 case BPF_KPTR_UNREF:
 638                 case BPF_KPTR_REF:
 639                 case BPF_KPTR_PERCPU:
 640                 case BPF_UPTR:
 641                         if (rec->fields[i].kptr.module)
 642                                 module_put(rec->fields[i].kptr.module);
 643                         if (btf_is_kernel(rec->fields[i].kptr.btf))
 644                                 btf_put(rec->fields[i].kptr.btf);
 645                         break;
 646                 case BPF_LIST_HEAD:
 647                 case BPF_LIST_NODE:
 648                 case BPF_RB_ROOT:
 649                 case BPF_RB_NODE:
 650                 case BPF_SPIN_LOCK:
 651                 case BPF_TIMER:
 652                 case BPF_REFCOUNT:
 653                 case BPF_WORKQUEUE:
 654                         /* Nothing to release */
 655                         break;
 656                 default:
 657                         WARN_ON_ONCE(1);
 658                         continue;
 659                 }
 660         }
 661         kfree(rec);
 662 }
 663
 664 void bpf_map_free_record(struct bpf_map *map)
 665 {
 666         btf_record_free(map->record);
 667         map->record = NULL;
 668 }
 669
 670 struct btf_record *btf_record_dup(const struct btf_record *rec)
 671 {
 672         const struct btf_field *fields;
 673         struct btf_record *new_rec;
 674         int ret, size, i;
 675
 676         if (IS_ERR_OR_NULL(rec))
 677                 return NULL;
 678         size = offsetof(struct btf_record, fields[rec->cnt]);
 679         new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
 680         if (!new_rec)
 681                 return ERR_PTR(-ENOMEM);
 682         /* Do a deep copy of the btf_record */
 683         fields = rec->fields;
 684         new_rec->cnt = 0;
 685         for (i = 0; i < rec->cnt; i++) {
 686                 switch (fields[i].type) {
 687                 case BPF_KPTR_UNREF:
 688                 case BPF_KPTR_REF:
 689                 case BPF_KPTR_PERCPU:
 690                 case BPF_UPTR:
 691                         if (btf_is_kernel(fields[i].kptr.btf))
 692                                 btf_get(fields[i].kptr.btf);
 693                         if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
 694                                 ret = -ENXIO;
 695                                 goto free;
 696                         }
 697                         break;
 698                 case BPF_LIST_HEAD:
 699                 case BPF_LIST_NODE:
 700                 case BPF_RB_ROOT:
 701                 case BPF_RB_NODE:
 702                 case BPF_SPIN_LOCK:
 703                 case BPF_TIMER:
 704                 case BPF_REFCOUNT:
 705                 case BPF_WORKQUEUE:
 706                         /* Nothing to acquire */
 707                         break;
 708                 default:
 709                         ret = -EFAULT;
 710                         WARN_ON_ONCE(1);
 711                         goto free;
 712                 }
 713                 new_rec->cnt++;
 714         }
 715         return new_rec;
 716 free:
 717         btf_record_free(new_rec);
 718         return ERR_PTR(ret);
 719 }
 720
 721 bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
 722 {
 723         bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
 724         int size;
 725
 726         if (!a_has_fields && !b_has_fields)
 727                 return true;
 728         if (a_has_fields != b_has_fields)
 729                 return false;
 730         if (rec_a->cnt != rec_b->cnt)
 731                 return false;
 732         size = offsetof(struct btf_record, fields[rec_a->cnt]);
 733         /* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
 734          * members are zeroed out. So memcmp is safe to do without worrying
 735          * about padding/unused fields.
 736          *
 737          * While spin_lock, timer, and kptr have no relation to map BTF,
 738          * list_head metadata is specific to map BTF, the btf and value_rec
 739          * members in particular. btf is the map BTF, while value_rec points to
 740          * btf_record in that map BTF.
 741          *
 742          * So while by default, we don't rely on the map BTF (which the records
 743          * were parsed from) matching for both records, which is not backwards
 744          * compatible, in case list_head is part of it, we implicitly rely on
 745          * that by way of depending on memcmp succeeding for it.
 746          */
 747         return !memcmp(rec_a, rec_b, size);
 748 }
 749
 750 void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
 751 {
 752         if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
 753                 return;
 754         bpf_timer_cancel_and_free(obj + rec->timer_off);
 755 }
 756
 757 void bpf_obj_free_workqueue(const struct btf_record *rec, void *obj)
 758 {
 759         if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_WORKQUEUE)))
 760                 return;
 761         bpf_wq_cancel_and_free(obj + rec->wq_off);
 762 }
 763
 764 void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
 765 {
 766         const struct btf_field *fields;
 767         int i;
 768
 769         if (IS_ERR_OR_NULL(rec))
 770                 return;
 771         fields = rec->fields;
 772         for (i = 0; i < rec->cnt; i++) {
 773                 struct btf_struct_meta *pointee_struct_meta;
 774                 const struct btf_field *field = &fields[i];
 775                 void *field_ptr = obj + field->offset;
 776                 void *xchgd_field;
 777
 778                 switch (fields[i].type) {
 779                 case BPF_SPIN_LOCK:
 780                         break;
 781                 case BPF_TIMER:
 782                         bpf_timer_cancel_and_free(field_ptr);
 783                         break;
 784                 case BPF_WORKQUEUE:
 785                         bpf_wq_cancel_and_free(field_ptr);
 786                         break;
 787                 case BPF_KPTR_UNREF:
 788                         WRITE_ONCE(*(u64 *)field_ptr, 0);
 789                         break;
 790                 case BPF_KPTR_REF:
 791                 case BPF_KPTR_PERCPU:
 792                         xchgd_field = (void *)xchg((unsigned long *)field_ptr, 0);
 793                         if (!xchgd_field)
 794                                 break;
 795
 796                         if (!btf_is_kernel(field->kptr.btf)) {
 797                                 pointee_struct_meta = btf_find_struct_meta(field->kptr.btf,
 798                                                                            field->kptr.btf_id);
 799                                 migrate_disable();
 800                                 __bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
 801                                                                  pointee_struct_meta->record : NULL,
 802                                                                  fields[i].type == BPF_KPTR_PERCPU);
 803                                 migrate_enable();
 804                         } else {
 805                                 field->kptr.dtor(xchgd_field);
 806                         }
 807                         break;
 808                 case BPF_UPTR:
 809                         /* The caller ensured that no one is using the uptr */
 810                         unpin_uptr_kaddr(*(void **)field_ptr);
 811                         break;
 812                 case BPF_LIST_HEAD:
 813                         if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 814                                 continue;
 815                         bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
 816                         break;
 817                 case BPF_RB_ROOT:
 818                         if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 819                                 continue;
 820                         bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
 821                         break;
 822                 case BPF_LIST_NODE:
 823                 case BPF_RB_NODE:
 824                 case BPF_REFCOUNT:
 825                         break;
 826                 default:
 827                         WARN_ON_ONCE(1);
 828                         continue;
 829                 }
 830         }
 831 }
 832
 833 static void bpf_map_free(struct bpf_map *map)
 834 {
 835         struct btf_record *rec = map->record;
 836         struct btf *btf = map->btf;
 837
 838         /* implementation dependent freeing */
 839         map->ops->map_free(map);
 840         /* Delay freeing of btf_record for maps, as map_free
 841          * callback usually needs access to them. It is better to do it here
 842          * than require each callback to do the free itself manually.
 843          *
 844          * Note that the btf_record stashed in map->inner_map_meta->record was
 845          * already freed using the map_free callback for map in map case which
 846          * eventually calls bpf_map_free_meta, since inner_map_meta is only a
 847          * template bpf_map struct used during verification.
 848          */
 849         btf_record_free(rec);
 850         /* Delay freeing of btf for maps, as map_free callback may need
 851          * struct_meta info which will be freed with btf_put().
 852          */
 853         btf_put(btf);
 854 }
 855
 856 /* called from workqueue */
 857 static void bpf_map_free_deferred(struct work_struct *work)
 858 {
 859         struct bpf_map *map = container_of(work, struct bpf_map, work);
 860
 861         security_bpf_map_free(map);
 862         bpf_map_release_memcg(map);
 863         bpf_map_free(map);
 864 }
 865
 866 static void bpf_map_put_uref(struct bpf_map *map)
 867 {
 868         if (atomic64_dec_and_test(&map->usercnt)) {
 869                 if (map->ops->map_release_uref)
 870                         map->ops->map_release_uref(map);
 871         }
 872 }
 873
 874 static void bpf_map_free_in_work(struct bpf_map *map)
 875 {
 876         INIT_WORK(&map->work, bpf_map_free_deferred);
 877         /* Avoid spawning kworkers, since they all might contend
 878          * for the same mutex like slab_mutex.
 879          */
 880         queue_work(system_unbound_wq, &map->work);
 881 }
 882
 883 static void bpf_map_free_rcu_gp(struct rcu_head *rcu)
 884 {
 885         bpf_map_free_in_work(container_of(rcu, struct bpf_map, rcu));
 886 }
 887
 888 static void bpf_map_free_mult_rcu_gp(struct rcu_head *rcu)
 889 {
 890         if (rcu_trace_implies_rcu_gp())
 891                 bpf_map_free_rcu_gp(rcu);
 892         else
 893                 call_rcu(rcu, bpf_map_free_rcu_gp);
 894 }
 895
 896 /* decrement map refcnt and schedule it for freeing via workqueue
 897  * (underlying map implementation ops->map_free() might sleep)
 898  */
 899 void bpf_map_put(struct bpf_map *map)
 900 {
 901         if (atomic64_dec_and_test(&map->refcnt)) {
 902                 /* bpf_map_free_id() must be called first */
 903                 bpf_map_free_id(map);
 904
 905                 WARN_ON_ONCE(atomic64_read(&map->sleepable_refcnt));
 906                 if (READ_ONCE(map->free_after_mult_rcu_gp))
 907                         call_rcu_tasks_trace(&map->rcu, bpf_map_free_mult_rcu_gp);
 908                 else if (READ_ONCE(map->free_after_rcu_gp))
 909                         call_rcu(&map->rcu, bpf_map_free_rcu_gp);
 910                 else
 911                         bpf_map_free_in_work(map);
 912         }
 913 }
 914 EXPORT_SYMBOL_GPL(bpf_map_put);
 915
 916 void bpf_map_put_with_uref(struct bpf_map *map)
 917 {
 918         bpf_map_put_uref(map);
 919         bpf_map_put(map);
 920 }
 921
 922 static int bpf_map_release(struct inode *inode, struct file *filp)
 923 {
 924         struct bpf_map *map = filp->private_data;
 925
 926         if (map->ops->map_release)
 927                 map->ops->map_release(map, filp);
 928
 929         bpf_map_put_with_uref(map);
 930         return 0;
 931 }
 932
 933 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 934 {
 935         fmode_t mode = fd_file(f)->f_mode;
 936
 937         /* Our file permissions may have been overridden by global
 938          * map permissions facing syscall side.
 939          */
 940         if (READ_ONCE(map->frozen))
 941                 mode &= ~FMODE_CAN_WRITE;
 942         return mode;
 943 }
 944
 945 #ifdef CONFIG_PROC_FS
 946 /* Show the memory usage of a bpf map */
 947 static u64 bpf_map_memory_usage(const struct bpf_map *map)
 948 {
 949         return map->ops->map_mem_usage(map);
 950 }
 951
 952 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 953 {
 954         struct bpf_map *map = filp->private_data;
 955         u32 type = 0, jited = 0;
 956
 957         if (map_type_contains_progs(map)) {
 958                 spin_lock(&map->owner.lock);
 959                 type  = map->owner.type;
 960                 jited = map->owner.jited;
 961                 spin_unlock(&map->owner.lock);
 962         }
 963
 964         seq_printf(m,
 965                    "map_type:\t%u\n"
 966                    "key_size:\t%u\n"
 967                    "value_size:\t%u\n"
 968                    "max_entries:\t%u\n"
 969                    "map_flags:\t%#x\n"
 970                    "map_extra:\t%#llx\n"
 971                    "memlock:\t%llu\n"
 972                    "map_id:\t%u\n"
 973                    "frozen:\t%u\n",
 974                    map->map_type,
 975                    map->key_size,
 976                    map->value_size,
 977                    map->max_entries,
 978                    map->map_flags,
 979                    (unsigned long long)map->map_extra,
 980                    bpf_map_memory_usage(map),
 981                    map->id,
 982                    READ_ONCE(map->frozen));
 983         if (type) {
 984                 seq_printf(m, "owner_prog_type:\t%u\n", type);
 985                 seq_printf(m, "owner_jited:\t%u\n", jited);
 986         }
 987 }
 988 #endif
 989
 990 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 991                               loff_t *ppos)
 992 {
 993         /* We need this handler such that alloc_file() enables
 994          * f_mode with FMODE_CAN_READ.
 995          */
 996         return -EINVAL;
 997 }
 998
 999 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
1000                                size_t siz, loff_t *ppos)
1001 {
1002         /* We need this handler such that alloc_file() enables
1003          * f_mode with FMODE_CAN_WRITE.
1004          */
1005         return -EINVAL;
1006 }
1007
1008 /* called for any extra memory-mapped regions (except initial) */
1009 static void bpf_map_mmap_open(struct vm_area_struct *vma)
1010 {
1011         struct bpf_map *map = vma->vm_file->private_data;
1012
1013         if (vma->vm_flags & VM_MAYWRITE)
1014                 bpf_map_write_active_inc(map);
1015 }
1016
1017 /* called for all unmapped memory region (including initial) */
1018 static void bpf_map_mmap_close(struct vm_area_struct *vma)
1019 {
1020         struct bpf_map *map = vma->vm_file->private_data;
1021
1022         if (vma->vm_flags & VM_MAYWRITE)
1023                 bpf_map_write_active_dec(map);
1024 }
1025
1026 static const struct vm_operations_struct bpf_map_default_vmops = {
1027         .open           = bpf_map_mmap_open,
1028         .close          = bpf_map_mmap_close,
1029 };
1030
1031 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
1032 {
1033         struct bpf_map *map = filp->private_data;
1034         int err;
1035
1036         if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
1037                 return -ENOTSUPP;
1038
1039         if (!(vma->vm_flags & VM_SHARED))
1040                 return -EINVAL;
1041
1042         mutex_lock(&map->freeze_mutex);
1043
1044         if (vma->vm_flags & VM_WRITE) {
1045                 if (map->frozen) {
1046                         err = -EPERM;
1047                         goto out;
1048                 }
1049                 /* map is meant to be read-only, so do not allow mapping as
1050                  * writable, because it's possible to leak a writable page
1051                  * reference and allows user-space to still modify it after
1052                  * freezing, while verifier will assume contents do not change
1053                  */
1054                 if (map->map_flags & BPF_F_RDONLY_PROG) {
1055                         err = -EACCES;
1056                         goto out;
1057                 }
1058         }
1059
1060         /* set default open/close callbacks */
1061         vma->vm_ops = &bpf_map_default_vmops;
1062         vma->vm_private_data = map;
1063         vm_flags_clear(vma, VM_MAYEXEC);
1064         if (!(vma->vm_flags & VM_WRITE))
1065                 /* disallow re-mapping with PROT_WRITE */
1066                 vm_flags_clear(vma, VM_MAYWRITE);
1067
1068         err = map->ops->map_mmap(map, vma);
1069         if (err)
1070                 goto out;
1071
1072         if (vma->vm_flags & VM_MAYWRITE)
1073                 bpf_map_write_active_inc(map);
1074 out:
1075         mutex_unlock(&map->freeze_mutex);
1076         return err;
1077 }
1078
1079 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
1080 {
1081         struct bpf_map *map = filp->private_data;
1082
1083         if (map->ops->map_poll)
1084                 return map->ops->map_poll(map, filp, pts);
1085
1086         return EPOLLERR;
1087 }
1088
1089 static unsigned long bpf_get_unmapped_area(struct file *filp, unsigned long addr,
1090                                            unsigned long len, unsigned long pgoff,
1091                                            unsigned long flags)
1092 {
1093         struct bpf_map *map = filp->private_data;
1094
1095         if (map->ops->map_get_unmapped_area)
1096                 return map->ops->map_get_unmapped_area(filp, addr, len, pgoff, flags);
1097 #ifdef CONFIG_MMU
1098         return mm_get_unmapped_area(current->mm, filp, addr, len, pgoff, flags);
1099 #else
1100         return addr;
1101 #endif
1102 }
1103
1104 const struct file_operations bpf_map_fops = {
1105 #ifdef CONFIG_PROC_FS
1106         .show_fdinfo    = bpf_map_show_fdinfo,
1107 #endif
1108         .release        = bpf_map_release,
1109         .read           = bpf_dummy_read,
1110         .write          = bpf_dummy_write,
1111         .mmap           = bpf_map_mmap,
1112         .poll           = bpf_map_poll,
1113         .get_unmapped_area = bpf_get_unmapped_area,
1114 };
1115
1116 int bpf_map_new_fd(struct bpf_map *map, int flags)
1117 {
1118         int ret;
1119
1120         ret = security_bpf_map(map, OPEN_FMODE(flags));
1121         if (ret < 0)
1122                 return ret;
1123
1124         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
1125                                 flags | O_CLOEXEC);
1126 }
1127
1128 int bpf_get_file_flag(int flags)
1129 {
1130         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
1131                 return -EINVAL;
1132         if (flags & BPF_F_RDONLY)
1133                 return O_RDONLY;
1134         if (flags & BPF_F_WRONLY)
1135                 return O_WRONLY;
1136         return O_RDWR;
1137 }
1138
1139 /* helper macro to check that unused fields 'union bpf_attr' are zero */
1140 #define CHECK_ATTR(CMD) \
1141         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
1142                    sizeof(attr->CMD##_LAST_FIELD), 0, \
1143                    sizeof(*attr) - \
1144                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
1145                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
1146
1147 /* dst and src must have at least "size" number of bytes.
1148  * Return strlen on success and < 0 on error.
1149  */
1150 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
1151 {
1152         const char *end = src + size;
1153         const char *orig_src = src;
1154
1155         memset(dst, 0, size);
1156         /* Copy all isalnum(), '_' and '.' chars. */
1157         while (src < end && *src) {
1158                 if (!isalnum(*src) &&
1159                     *src != '_' && *src != '.')
1160                         return -EINVAL;
1161                 *dst++ = *src++;
1162         }
1163
1164         /* No '\0' found in "size" number of bytes */
1165         if (src == end)
1166                 return -EINVAL;
1167
1168         return src - orig_src;
1169 }
1170
1171 int map_check_no_btf(const struct bpf_map *map,
1172                      const struct btf *btf,
1173                      const struct btf_type *key_type,
1174                      const struct btf_type *value_type)
1175 {
1176         return -ENOTSUPP;
1177 }
1178
1179 static int map_check_btf(struct bpf_map *map, struct bpf_token *token,
1180                          const struct btf *btf, u32 btf_key_id, u32 btf_value_id)
1181 {
1182         const struct btf_type *key_type, *value_type;
1183         u32 key_size, value_size;
1184         int ret = 0;
1185
1186         /* Some maps allow key to be unspecified. */
1187         if (btf_key_id) {
1188                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
1189                 if (!key_type || key_size != map->key_size)
1190                         return -EINVAL;
1191         } else {
1192                 key_type = btf_type_by_id(btf, 0);
1193                 if (!map->ops->map_check_btf)
1194                         return -EINVAL;
1195         }
1196
1197         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1198         if (!value_type || value_size != map->value_size)
1199                 return -EINVAL;
1200
1201         map->record = btf_parse_fields(btf, value_type,
1202                                        BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1203                                        BPF_RB_ROOT | BPF_REFCOUNT | BPF_WORKQUEUE | BPF_UPTR,
1204                                        map->value_size);
1205         if (!IS_ERR_OR_NULL(map->record)) {
1206                 int i;
1207
1208                 if (!bpf_token_capable(token, CAP_BPF)) {
1209                         ret = -EPERM;
1210                         goto free_map_tab;
1211                 }
1212                 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1213                         ret = -EACCES;
1214                         goto free_map_tab;
1215                 }
1216                 for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1217                         switch (map->record->field_mask & (1 << i)) {
1218                         case 0:
1219                                 continue;
1220                         case BPF_SPIN_LOCK:
1221                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1222                                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1223                                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1224                                     map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1225                                     map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1226                                     map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1227                                     map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1228                                         ret = -EOPNOTSUPP;
1229                                         goto free_map_tab;
1230                                 }
1231                                 break;
1232                         case BPF_TIMER:
1233                         case BPF_WORKQUEUE:
1234                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1235                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1236                                     map->map_type != BPF_MAP_TYPE_ARRAY) {
1237                                         ret = -EOPNOTSUPP;
1238                                         goto free_map_tab;
1239                                 }
1240                                 break;
1241                         case BPF_KPTR_UNREF:
1242                         case BPF_KPTR_REF:
1243                         case BPF_KPTR_PERCPU:
1244                         case BPF_REFCOUNT:
1245                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1246                                     map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
1247                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1248                                     map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH &&
1249                                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1250                                     map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY &&
1251                                     map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1252                                     map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1253                                     map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1254                                     map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1255                                         ret = -EOPNOTSUPP;
1256                                         goto free_map_tab;
1257                                 }
1258                                 break;
1259                         case BPF_UPTR:
1260                                 if (map->map_type != BPF_MAP_TYPE_TASK_STORAGE) {
1261                                         ret = -EOPNOTSUPP;
1262                                         goto free_map_tab;
1263                                 }
1264                                 break;
1265                         case BPF_LIST_HEAD:
1266                         case BPF_RB_ROOT:
1267                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1268                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1269                                     map->map_type != BPF_MAP_TYPE_ARRAY) {
1270                                         ret = -EOPNOTSUPP;
1271                                         goto free_map_tab;
1272                                 }
1273                                 break;
1274                         default:
1275                                 /* Fail if map_type checks are missing for a field type */
1276                                 ret = -EOPNOTSUPP;
1277                                 goto free_map_tab;
1278                         }
1279                 }
1280         }
1281
1282         ret = btf_check_and_fixup_fields(btf, map->record);
1283         if (ret < 0)
1284                 goto free_map_tab;
1285
1286         if (map->ops->map_check_btf) {
1287                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1288                 if (ret < 0)
1289                         goto free_map_tab;
1290         }
1291
1292         return ret;
1293 free_map_tab:
1294         bpf_map_free_record(map);
1295         return ret;
1296 }
1297
1298 static bool bpf_net_capable(void)
1299 {
1300         return capable(CAP_NET_ADMIN) || capable(CAP_SYS_ADMIN);
1301 }
1302
1303 #define BPF_MAP_CREATE_LAST_FIELD map_token_fd
1304 /* called via syscall */
1305 static int map_create(union bpf_attr *attr)
1306 {
1307         const struct bpf_map_ops *ops;
1308         struct bpf_token *token = NULL;
1309         int numa_node = bpf_map_attr_numa_node(attr);
1310         u32 map_type = attr->map_type;
1311         struct bpf_map *map;
1312         bool token_flag;
1313         int f_flags;
1314         int err;
1315
1316         err = CHECK_ATTR(BPF_MAP_CREATE);
1317         if (err)
1318                 return -EINVAL;
1319
1320         /* check BPF_F_TOKEN_FD flag, remember if it's set, and then clear it
1321          * to avoid per-map type checks tripping on unknown flag
1322          */
1323         token_flag = attr->map_flags & BPF_F_TOKEN_FD;
1324         attr->map_flags &= ~BPF_F_TOKEN_FD;
1325
1326         if (attr->btf_vmlinux_value_type_id) {
1327                 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1328                     attr->btf_key_type_id || attr->btf_value_type_id)
1329                         return -EINVAL;
1330         } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1331                 return -EINVAL;
1332         }
1333
1334         if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1335             attr->map_type != BPF_MAP_TYPE_ARENA &&
1336             attr->map_extra != 0)
1337                 return -EINVAL;
1338
1339         f_flags = bpf_get_file_flag(attr->map_flags);
1340         if (f_flags < 0)
1341                 return f_flags;
1342
1343         if (numa_node != NUMA_NO_NODE &&
1344             ((unsigned int)numa_node >= nr_node_ids ||
1345              !node_online(numa_node)))
1346                 return -EINVAL;
1347
1348         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1349         map_type = attr->map_type;
1350         if (map_type >= ARRAY_SIZE(bpf_map_types))
1351                 return -EINVAL;
1352         map_type = array_index_nospec(map_type, ARRAY_SIZE(bpf_map_types));
1353         ops = bpf_map_types[map_type];
1354         if (!ops)
1355                 return -EINVAL;
1356
1357         if (ops->map_alloc_check) {
1358                 err = ops->map_alloc_check(attr);
1359                 if (err)
1360                         return err;
1361         }
1362         if (attr->map_ifindex)
1363                 ops = &bpf_map_offload_ops;
1364         if (!ops->map_mem_usage)
1365                 return -EINVAL;
1366
1367         if (token_flag) {
1368                 token = bpf_token_get_from_fd(attr->map_token_fd);
1369                 if (IS_ERR(token))
1370                         return PTR_ERR(token);
1371
1372                 /* if current token doesn't grant map creation permissions,
1373                  * then we can't use this token, so ignore it and rely on
1374                  * system-wide capabilities checks
1375                  */
1376                 if (!bpf_token_allow_cmd(token, BPF_MAP_CREATE) ||
1377                     !bpf_token_allow_map_type(token, attr->map_type)) {
1378                         bpf_token_put(token);
1379                         token = NULL;
1380                 }
1381         }
1382
1383         err = -EPERM;
1384
1385         /* Intent here is for unprivileged_bpf_disabled to block BPF map
1386          * creation for unprivileged users; other actions depend
1387          * on fd availability and access to bpffs, so are dependent on
1388          * object creation success. Even with unprivileged BPF disabled,
1389          * capability checks are still carried out.
1390          */
1391         if (sysctl_unprivileged_bpf_disabled && !bpf_token_capable(token, CAP_BPF))
1392                 goto put_token;
1393
1394         /* check privileged map type permissions */
1395         switch (map_type) {
1396         case BPF_MAP_TYPE_ARRAY:
1397         case BPF_MAP_TYPE_PERCPU_ARRAY:
1398         case BPF_MAP_TYPE_PROG_ARRAY:
1399         case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1400         case BPF_MAP_TYPE_CGROUP_ARRAY:
1401         case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1402         case BPF_MAP_TYPE_HASH:
1403         case BPF_MAP_TYPE_PERCPU_HASH:
1404         case BPF_MAP_TYPE_HASH_OF_MAPS:
1405         case BPF_MAP_TYPE_RINGBUF:
1406         case BPF_MAP_TYPE_USER_RINGBUF:
1407         case BPF_MAP_TYPE_CGROUP_STORAGE:
1408         case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
1409                 /* unprivileged */
1410                 break;
1411         case BPF_MAP_TYPE_SK_STORAGE:
1412         case BPF_MAP_TYPE_INODE_STORAGE:
1413         case BPF_MAP_TYPE_TASK_STORAGE:
1414         case BPF_MAP_TYPE_CGRP_STORAGE:
1415         case BPF_MAP_TYPE_BLOOM_FILTER:
1416         case BPF_MAP_TYPE_LPM_TRIE:
1417         case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
1418         case BPF_MAP_TYPE_STACK_TRACE:
1419         case BPF_MAP_TYPE_QUEUE:
1420         case BPF_MAP_TYPE_STACK:
1421         case BPF_MAP_TYPE_LRU_HASH:
1422         case BPF_MAP_TYPE_LRU_PERCPU_HASH:
1423         case BPF_MAP_TYPE_STRUCT_OPS:
1424         case BPF_MAP_TYPE_CPUMAP:
1425         case BPF_MAP_TYPE_ARENA:
1426                 if (!bpf_token_capable(token, CAP_BPF))
1427                         goto put_token;
1428                 break;
1429         case BPF_MAP_TYPE_SOCKMAP:
1430         case BPF_MAP_TYPE_SOCKHASH:
1431         case BPF_MAP_TYPE_DEVMAP:
1432         case BPF_MAP_TYPE_DEVMAP_HASH:
1433         case BPF_MAP_TYPE_XSKMAP:
1434                 if (!bpf_token_capable(token, CAP_NET_ADMIN))
1435                         goto put_token;
1436                 break;
1437         default:
1438                 WARN(1, "unsupported map type %d", map_type);
1439                 goto put_token;
1440         }
1441
1442         map = ops->map_alloc(attr);
1443         if (IS_ERR(map)) {
1444                 err = PTR_ERR(map);
1445                 goto put_token;
1446         }
1447         map->ops = ops;
1448         map->map_type = map_type;
1449
1450         err = bpf_obj_name_cpy(map->name, attr->map_name,
1451                                sizeof(attr->map_name));
1452         if (err < 0)
1453                 goto free_map;
1454
1455         atomic64_set(&map->refcnt, 1);
1456         atomic64_set(&map->usercnt, 1);
1457         mutex_init(&map->freeze_mutex);
1458         spin_lock_init(&map->owner.lock);
1459
1460         if (attr->btf_key_type_id || attr->btf_value_type_id ||
1461             /* Even the map's value is a kernel's struct,
1462              * the bpf_prog.o must have BTF to begin with
1463              * to figure out the corresponding kernel's
1464              * counter part.  Thus, attr->btf_fd has
1465              * to be valid also.
1466              */
1467             attr->btf_vmlinux_value_type_id) {
1468                 struct btf *btf;
1469
1470                 btf = btf_get_by_fd(attr->btf_fd);
1471                 if (IS_ERR(btf)) {
1472                         err = PTR_ERR(btf);
1473                         goto free_map;
1474                 }
1475                 if (btf_is_kernel(btf)) {
1476                         btf_put(btf);
1477                         err = -EACCES;
1478                         goto free_map;
1479                 }
1480                 map->btf = btf;
1481
1482                 if (attr->btf_value_type_id) {
1483                         err = map_check_btf(map, token, btf, attr->btf_key_type_id,
1484                                             attr->btf_value_type_id);
1485                         if (err)
1486                                 goto free_map;
1487                 }
1488
1489                 map->btf_key_type_id = attr->btf_key_type_id;
1490                 map->btf_value_type_id = attr->btf_value_type_id;
1491                 map->btf_vmlinux_value_type_id =
1492                         attr->btf_vmlinux_value_type_id;
1493         }
1494
1495         err = security_bpf_map_create(map, attr, token);
1496         if (err)
1497                 goto free_map_sec;
1498
1499         err = bpf_map_alloc_id(map);
1500         if (err)
1501                 goto free_map_sec;
1502
1503         bpf_map_save_memcg(map);
1504         bpf_token_put(token);
1505
1506         err = bpf_map_new_fd(map, f_flags);
1507         if (err < 0) {
1508                 /* failed to allocate fd.
1509                  * bpf_map_put_with_uref() is needed because the above
1510                  * bpf_map_alloc_id() has published the map
1511                  * to the userspace and the userspace may
1512                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1513                  */
1514                 bpf_map_put_with_uref(map);
1515                 return err;
1516         }
1517
1518         return err;
1519
1520 free_map_sec:
1521         security_bpf_map_free(map);
1522 free_map:
1523         bpf_map_free(map);
1524 put_token:
1525         bpf_token_put(token);
1526         return err;
1527 }
1528
1529 void bpf_map_inc(struct bpf_map *map)
1530 {
1531         atomic64_inc(&map->refcnt);
1532 }
1533 EXPORT_SYMBOL_GPL(bpf_map_inc);
1534
1535 void bpf_map_inc_with_uref(struct bpf_map *map)
1536 {
1537         atomic64_inc(&map->refcnt);
1538         atomic64_inc(&map->usercnt);
1539 }
1540 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1541
1542 struct bpf_map *bpf_map_get(u32 ufd)
1543 {
1544         CLASS(fd, f)(ufd);
1545         struct bpf_map *map = __bpf_map_get(f);
1546
1547         if (!IS_ERR(map))
1548                 bpf_map_inc(map);
1549
1550         return map;
1551 }
1552 EXPORT_SYMBOL(bpf_map_get);
1553
1554 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1555 {
1556         CLASS(fd, f)(ufd);
1557         struct bpf_map *map = __bpf_map_get(f);
1558
1559         if (!IS_ERR(map))
1560                 bpf_map_inc_with_uref(map);
1561
1562         return map;
1563 }
1564
1565 /* map_idr_lock should have been held or the map should have been
1566  * protected by rcu read lock.
1567  */
1568 struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1569 {
1570         int refold;
1571
1572         refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1573         if (!refold)
1574                 return ERR_PTR(-ENOENT);
1575         if (uref)
1576                 atomic64_inc(&map->usercnt);
1577
1578         return map;
1579 }
1580
1581 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1582 {
1583         spin_lock_bh(&map_idr_lock);
1584         map = __bpf_map_inc_not_zero(map, false);
1585         spin_unlock_bh(&map_idr_lock);
1586
1587         return map;
1588 }
1589 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1590
1591 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1592 {
1593         return -ENOTSUPP;
1594 }
1595
1596 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1597 {
1598         if (key_size)
1599                 return vmemdup_user(ukey, key_size);
1600
1601         if (ukey)
1602                 return ERR_PTR(-EINVAL);
1603
1604         return NULL;
1605 }
1606
1607 static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1608 {
1609         if (key_size)
1610                 return kvmemdup_bpfptr(ukey, key_size);
1611
1612         if (!bpfptr_is_null(ukey))
1613                 return ERR_PTR(-EINVAL);
1614
1615         return NULL;
1616 }
1617
1618 /* last field in 'union bpf_attr' used by this command */
1619 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1620
1621 static int map_lookup_elem(union bpf_attr *attr)
1622 {
1623         void __user *ukey = u64_to_user_ptr(attr->key);
1624         void __user *uvalue = u64_to_user_ptr(attr->value);
1625         struct bpf_map *map;
1626         void *key, *value;
1627         u32 value_size;
1628         int err;
1629
1630         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1631                 return -EINVAL;
1632
1633         if (attr->flags & ~BPF_F_LOCK)
1634                 return -EINVAL;
1635
1636         CLASS(fd, f)(attr->map_fd);
1637         map = __bpf_map_get(f);
1638         if (IS_ERR(map))
1639                 return PTR_ERR(map);
1640         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ))
1641                 return -EPERM;
1642
1643         if ((attr->flags & BPF_F_LOCK) &&
1644             !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1645                 return -EINVAL;
1646
1647         key = __bpf_copy_key(ukey, map->key_size);
1648         if (IS_ERR(key))
1649                 return PTR_ERR(key);
1650
1651         value_size = bpf_map_value_size(map);
1652
1653         err = -ENOMEM;
1654         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1655         if (!value)
1656                 goto free_key;
1657
1658         if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1659                 if (copy_from_user(value, uvalue, value_size))
1660                         err = -EFAULT;
1661                 else
1662                         err = bpf_map_copy_value(map, key, value, attr->flags);
1663                 goto free_value;
1664         }
1665
1666         err = bpf_map_copy_value(map, key, value, attr->flags);
1667         if (err)
1668                 goto free_value;
1669
1670         err = -EFAULT;
1671         if (copy_to_user(uvalue, value, value_size) != 0)
1672                 goto free_value;
1673
1674         err = 0;
1675
1676 free_value:
1677         kvfree(value);
1678 free_key:
1679         kvfree(key);
1680         return err;
1681 }
1682
1683
1684 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1685
1686 static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1687 {
1688         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1689         bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1690         struct bpf_map *map;
1691         void *key, *value;
1692         u32 value_size;
1693         int err;
1694
1695         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1696                 return -EINVAL;
1697
1698         CLASS(fd, f)(attr->map_fd);
1699         map = __bpf_map_get(f);
1700         if (IS_ERR(map))
1701                 return PTR_ERR(map);
1702         bpf_map_write_active_inc(map);
1703         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1704                 err = -EPERM;
1705                 goto err_put;
1706         }
1707
1708         if ((attr->flags & BPF_F_LOCK) &&
1709             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1710                 err = -EINVAL;
1711                 goto err_put;
1712         }
1713
1714         key = ___bpf_copy_key(ukey, map->key_size);
1715         if (IS_ERR(key)) {
1716                 err = PTR_ERR(key);
1717                 goto err_put;
1718         }
1719
1720         value_size = bpf_map_value_size(map);
1721         value = kvmemdup_bpfptr(uvalue, value_size);
1722         if (IS_ERR(value)) {
1723                 err = PTR_ERR(value);
1724                 goto free_key;
1725         }
1726
1727         err = bpf_map_update_value(map, fd_file(f), key, value, attr->flags);
1728         if (!err)
1729                 maybe_wait_bpf_programs(map);
1730
1731         kvfree(value);
1732 free_key:
1733         kvfree(key);
1734 err_put:
1735         bpf_map_write_active_dec(map);
1736         return err;
1737 }
1738
1739 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1740
1741 static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1742 {
1743         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1744         struct bpf_map *map;
1745         void *key;
1746         int err;
1747
1748         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1749                 return -EINVAL;
1750
1751         CLASS(fd, f)(attr->map_fd);
1752         map = __bpf_map_get(f);
1753         if (IS_ERR(map))
1754                 return PTR_ERR(map);
1755         bpf_map_write_active_inc(map);
1756         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1757                 err = -EPERM;
1758                 goto err_put;
1759         }
1760
1761         key = ___bpf_copy_key(ukey, map->key_size);
1762         if (IS_ERR(key)) {
1763                 err = PTR_ERR(key);
1764                 goto err_put;
1765         }
1766
1767         if (bpf_map_is_offloaded(map)) {
1768                 err = bpf_map_offload_delete_elem(map, key);
1769                 goto out;
1770         } else if (IS_FD_PROG_ARRAY(map) ||
1771                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1772                 /* These maps require sleepable context */
1773                 err = map->ops->map_delete_elem(map, key);
1774                 goto out;
1775         }
1776
1777         bpf_disable_instrumentation();
1778         rcu_read_lock();
1779         err = map->ops->map_delete_elem(map, key);
1780         rcu_read_unlock();
1781         bpf_enable_instrumentation();
1782         if (!err)
1783                 maybe_wait_bpf_programs(map);
1784 out:
1785         kvfree(key);
1786 err_put:
1787         bpf_map_write_active_dec(map);
1788         return err;
1789 }
1790
1791 /* last field in 'union bpf_attr' used by this command */
1792 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1793
1794 static int map_get_next_key(union bpf_attr *attr)
1795 {
1796         void __user *ukey = u64_to_user_ptr(attr->key);
1797         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1798         struct bpf_map *map;
1799         void *key, *next_key;
1800         int err;
1801
1802         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1803                 return -EINVAL;
1804
1805         CLASS(fd, f)(attr->map_fd);
1806         map = __bpf_map_get(f);
1807         if (IS_ERR(map))
1808                 return PTR_ERR(map);
1809         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ))
1810                 return -EPERM;
1811
1812         if (ukey) {
1813                 key = __bpf_copy_key(ukey, map->key_size);
1814                 if (IS_ERR(key))
1815                         return PTR_ERR(key);
1816         } else {
1817                 key = NULL;
1818         }
1819
1820         err = -ENOMEM;
1821         next_key = kvmalloc(map->key_size, GFP_USER);
1822         if (!next_key)
1823                 goto free_key;
1824
1825         if (bpf_map_is_offloaded(map)) {
1826                 err = bpf_map_offload_get_next_key(map, key, next_key);
1827                 goto out;
1828         }
1829
1830         rcu_read_lock();
1831         err = map->ops->map_get_next_key(map, key, next_key);
1832         rcu_read_unlock();
1833 out:
1834         if (err)
1835                 goto free_next_key;
1836
1837         err = -EFAULT;
1838         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1839                 goto free_next_key;
1840
1841         err = 0;
1842
1843 free_next_key:
1844         kvfree(next_key);
1845 free_key:
1846         kvfree(key);
1847         return err;
1848 }
1849
1850 int generic_map_delete_batch(struct bpf_map *map,
1851                              const union bpf_attr *attr,
1852                              union bpf_attr __user *uattr)
1853 {
1854         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1855         u32 cp, max_count;
1856         int err = 0;
1857         void *key;
1858
1859         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1860                 return -EINVAL;
1861
1862         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1863             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1864                 return -EINVAL;
1865         }
1866
1867         max_count = attr->batch.count;
1868         if (!max_count)
1869                 return 0;
1870
1871         if (put_user(0, &uattr->batch.count))
1872                 return -EFAULT;
1873
1874         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1875         if (!key)
1876                 return -ENOMEM;
1877
1878         for (cp = 0; cp < max_count; cp++) {
1879                 err = -EFAULT;
1880                 if (copy_from_user(key, keys + cp * map->key_size,
1881                                    map->key_size))
1882                         break;
1883
1884                 if (bpf_map_is_offloaded(map)) {
1885                         err = bpf_map_offload_delete_elem(map, key);
1886                         break;
1887                 }
1888
1889                 bpf_disable_instrumentation();
1890                 rcu_read_lock();
1891                 err = map->ops->map_delete_elem(map, key);
1892                 rcu_read_unlock();
1893                 bpf_enable_instrumentation();
1894                 if (err)
1895                         break;
1896                 cond_resched();
1897         }
1898         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1899                 err = -EFAULT;
1900
1901         kvfree(key);
1902
1903         return err;
1904 }
1905
1906 int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1907                              const union bpf_attr *attr,
1908                              union bpf_attr __user *uattr)
1909 {
1910         void __user *values = u64_to_user_ptr(attr->batch.values);
1911         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1912         u32 value_size, cp, max_count;
1913         void *key, *value;
1914         int err = 0;
1915
1916         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1917                 return -EINVAL;
1918
1919         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1920             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1921                 return -EINVAL;
1922         }
1923
1924         value_size = bpf_map_value_size(map);
1925
1926         max_count = attr->batch.count;
1927         if (!max_count)
1928                 return 0;
1929
1930         if (put_user(0, &uattr->batch.count))
1931                 return -EFAULT;
1932
1933         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1934         if (!key)
1935                 return -ENOMEM;
1936
1937         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1938         if (!value) {
1939                 kvfree(key);
1940                 return -ENOMEM;
1941         }
1942
1943         for (cp = 0; cp < max_count; cp++) {
1944                 err = -EFAULT;
1945                 if (copy_from_user(key, keys + cp * map->key_size,
1946                     map->key_size) ||
1947                     copy_from_user(value, values + cp * value_size, value_size))
1948                         break;
1949
1950                 err = bpf_map_update_value(map, map_file, key, value,
1951                                            attr->batch.elem_flags);
1952
1953                 if (err)
1954                         break;
1955                 cond_resched();
1956         }
1957
1958         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1959                 err = -EFAULT;
1960
1961         kvfree(value);
1962         kvfree(key);
1963
1964         return err;
1965 }
1966
1967 #define MAP_LOOKUP_RETRIES 3
1968
1969 int generic_map_lookup_batch(struct bpf_map *map,
1970                                     const union bpf_attr *attr,
1971                                     union bpf_attr __user *uattr)
1972 {
1973         void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1974         void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1975         void __user *values = u64_to_user_ptr(attr->batch.values);
1976         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1977         void *buf, *buf_prevkey, *prev_key, *key, *value;
1978         int err, retry = MAP_LOOKUP_RETRIES;
1979         u32 value_size, cp, max_count;
1980
1981         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1982                 return -EINVAL;
1983
1984         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1985             !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1986                 return -EINVAL;
1987
1988         value_size = bpf_map_value_size(map);
1989
1990         max_count = attr->batch.count;
1991         if (!max_count)
1992                 return 0;
1993
1994         if (put_user(0, &uattr->batch.count))
1995                 return -EFAULT;
1996
1997         buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1998         if (!buf_prevkey)
1999                 return -ENOMEM;
2000
2001         buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
2002         if (!buf) {
2003                 kvfree(buf_prevkey);
2004                 return -ENOMEM;
2005         }
2006
2007         err = -EFAULT;
2008         prev_key = NULL;
2009         if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
2010                 goto free_buf;
2011         key = buf;
2012         value = key + map->key_size;
2013         if (ubatch)
2014                 prev_key = buf_prevkey;
2015
2016         for (cp = 0; cp < max_count;) {
2017                 rcu_read_lock();
2018                 err = map->ops->map_get_next_key(map, prev_key, key);
2019                 rcu_read_unlock();
2020                 if (err)
2021                         break;
2022                 err = bpf_map_copy_value(map, key, value,
2023                                          attr->batch.elem_flags);
2024
2025                 if (err == -ENOENT) {
2026                         if (retry) {
2027                                 retry--;
2028                                 continue;
2029                         }
2030                         err = -EINTR;
2031                         break;
2032                 }
2033
2034                 if (err)
2035                         goto free_buf;
2036
2037                 if (copy_to_user(keys + cp * map->key_size, key,
2038                                  map->key_size)) {
2039                         err = -EFAULT;
2040                         goto free_buf;
2041                 }
2042                 if (copy_to_user(values + cp * value_size, value, value_size)) {
2043                         err = -EFAULT;
2044                         goto free_buf;
2045                 }
2046
2047                 if (!prev_key)
2048                         prev_key = buf_prevkey;
2049
2050                 swap(prev_key, key);
2051                 retry = MAP_LOOKUP_RETRIES;
2052                 cp++;
2053                 cond_resched();
2054         }
2055
2056         if (err == -EFAULT)
2057                 goto free_buf;
2058
2059         if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
2060                     (cp && copy_to_user(uobatch, prev_key, map->key_size))))
2061                 err = -EFAULT;
2062
2063 free_buf:
2064         kvfree(buf_prevkey);
2065         kvfree(buf);
2066         return err;
2067 }
2068
2069 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
2070
2071 static int map_lookup_and_delete_elem(union bpf_attr *attr)
2072 {
2073         void __user *ukey = u64_to_user_ptr(attr->key);
2074         void __user *uvalue = u64_to_user_ptr(attr->value);
2075         struct bpf_map *map;
2076         void *key, *value;
2077         u32 value_size;
2078         int err;
2079
2080         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
2081                 return -EINVAL;
2082
2083         if (attr->flags & ~BPF_F_LOCK)
2084                 return -EINVAL;
2085
2086         CLASS(fd, f)(attr->map_fd);
2087         map = __bpf_map_get(f);
2088         if (IS_ERR(map))
2089                 return PTR_ERR(map);
2090         bpf_map_write_active_inc(map);
2091         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
2092             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
2093                 err = -EPERM;
2094                 goto err_put;
2095         }
2096
2097         if (attr->flags &&
2098             (map->map_type == BPF_MAP_TYPE_QUEUE ||
2099              map->map_type == BPF_MAP_TYPE_STACK)) {
2100                 err = -EINVAL;
2101                 goto err_put;
2102         }
2103
2104         if ((attr->flags & BPF_F_LOCK) &&
2105             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
2106                 err = -EINVAL;
2107                 goto err_put;
2108         }
2109
2110         key = __bpf_copy_key(ukey, map->key_size);
2111         if (IS_ERR(key)) {
2112                 err = PTR_ERR(key);
2113                 goto err_put;
2114         }
2115
2116         value_size = bpf_map_value_size(map);
2117
2118         err = -ENOMEM;
2119         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
2120         if (!value)
2121                 goto free_key;
2122
2123         err = -ENOTSUPP;
2124         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
2125             map->map_type == BPF_MAP_TYPE_STACK) {
2126                 err = map->ops->map_pop_elem(map, value);
2127         } else if (map->map_type == BPF_MAP_TYPE_HASH ||
2128                    map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
2129                    map->map_type == BPF_MAP_TYPE_LRU_HASH ||
2130                    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
2131                 if (!bpf_map_is_offloaded(map)) {
2132                         bpf_disable_instrumentation();
2133                         rcu_read_lock();
2134                         err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
2135                         rcu_read_unlock();
2136                         bpf_enable_instrumentation();
2137                 }
2138         }
2139
2140         if (err)
2141                 goto free_value;
2142
2143         if (copy_to_user(uvalue, value, value_size) != 0) {
2144                 err = -EFAULT;
2145                 goto free_value;
2146         }
2147
2148         err = 0;
2149
2150 free_value:
2151         kvfree(value);
2152 free_key:
2153         kvfree(key);
2154 err_put:
2155         bpf_map_write_active_dec(map);
2156         return err;
2157 }
2158
2159 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
2160
2161 static int map_freeze(const union bpf_attr *attr)
2162 {
2163         int err = 0;
2164         struct bpf_map *map;
2165
2166         if (CHECK_ATTR(BPF_MAP_FREEZE))
2167                 return -EINVAL;
2168
2169         CLASS(fd, f)(attr->map_fd);
2170         map = __bpf_map_get(f);
2171         if (IS_ERR(map))
2172                 return PTR_ERR(map);
2173
2174         if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record))
2175                 return -ENOTSUPP;
2176
2177         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE))
2178                 return -EPERM;
2179
2180         mutex_lock(&map->freeze_mutex);
2181         if (bpf_map_write_active(map)) {
2182                 err = -EBUSY;
2183                 goto err_put;
2184         }
2185         if (READ_ONCE(map->frozen)) {
2186                 err = -EBUSY;
2187                 goto err_put;
2188         }
2189
2190         WRITE_ONCE(map->frozen, true);
2191 err_put:
2192         mutex_unlock(&map->freeze_mutex);
2193         return err;
2194 }
2195
2196 static const struct bpf_prog_ops * const bpf_prog_types[] = {
2197 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
2198         [_id] = & _name ## _prog_ops,
2199 #define BPF_MAP_TYPE(_id, _ops)
2200 #define BPF_LINK_TYPE(_id, _name)
2201 #include <linux/bpf_types.h>
2202 #undef BPF_PROG_TYPE
2203 #undef BPF_MAP_TYPE
2204 #undef BPF_LINK_TYPE
2205 };
2206
2207 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
2208 {
2209         const struct bpf_prog_ops *ops;
2210
2211         if (type >= ARRAY_SIZE(bpf_prog_types))
2212                 return -EINVAL;
2213         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
2214         ops = bpf_prog_types[type];
2215         if (!ops)
2216                 return -EINVAL;
2217
2218         if (!bpf_prog_is_offloaded(prog->aux))
2219                 prog->aux->ops = ops;
2220         else
2221                 prog->aux->ops = &bpf_offload_prog_ops;
2222         prog->type = type;
2223         return 0;
2224 }
2225
2226 enum bpf_audit {
2227         BPF_AUDIT_LOAD,
2228         BPF_AUDIT_UNLOAD,
2229         BPF_AUDIT_MAX,
2230 };
2231
2232 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
2233         [BPF_AUDIT_LOAD]   = "LOAD",
2234         [BPF_AUDIT_UNLOAD] = "UNLOAD",
2235 };
2236
2237 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
2238 {
2239         struct audit_context *ctx = NULL;
2240         struct audit_buffer *ab;
2241
2242         if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2243                 return;
2244         if (audit_enabled == AUDIT_OFF)
2245                 return;
2246         if (!in_irq() && !irqs_disabled())
2247                 ctx = audit_context();
2248         ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2249         if (unlikely(!ab))
2250                 return;
2251         audit_log_format(ab, "prog-id=%u op=%s",
2252                          prog->aux->id, bpf_audit_str[op]);
2253         audit_log_end(ab);
2254 }
2255
2256 static int bpf_prog_alloc_id(struct bpf_prog *prog)
2257 {
2258         int id;
2259
2260         idr_preload(GFP_KERNEL);
2261         spin_lock_bh(&prog_idr_lock);
2262         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2263         if (id > 0)
2264                 prog->aux->id = id;
2265         spin_unlock_bh(&prog_idr_lock);
2266         idr_preload_end();
2267
2268         /* id is in [1, INT_MAX) */
2269         if (WARN_ON_ONCE(!id))
2270                 return -ENOSPC;
2271
2272         return id > 0 ? 0 : id;
2273 }
2274
2275 void bpf_prog_free_id(struct bpf_prog *prog)
2276 {
2277         unsigned long flags;
2278
2279         /* cBPF to eBPF migrations are currently not in the idr store.
2280          * Offloaded programs are removed from the store when their device
2281          * disappears - even if someone grabs an fd to them they are unusable,
2282          * simply waiting for refcnt to drop to be freed.
2283          */
2284         if (!prog->aux->id)
2285                 return;
2286
2287         spin_lock_irqsave(&prog_idr_lock, flags);
2288         idr_remove(&prog_idr, prog->aux->id);
2289         prog->aux->id = 0;
2290         spin_unlock_irqrestore(&prog_idr_lock, flags);
2291 }
2292
2293 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2294 {
2295         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2296
2297         kvfree(aux->func_info);
2298         kfree(aux->func_info_aux);
2299         free_uid(aux->user);
2300         security_bpf_prog_free(aux->prog);
2301         bpf_prog_free(aux->prog);
2302 }
2303
2304 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2305 {
2306         bpf_prog_kallsyms_del_all(prog);
2307         btf_put(prog->aux->btf);
2308         module_put(prog->aux->mod);
2309         kvfree(prog->aux->jited_linfo);
2310         kvfree(prog->aux->linfo);
2311         kfree(prog->aux->kfunc_tab);
2312         if (prog->aux->attach_btf)
2313                 btf_put(prog->aux->attach_btf);
2314
2315         if (deferred) {
2316                 if (prog->sleepable)
2317                         call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2318                 else
2319                         call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2320         } else {
2321                 __bpf_prog_put_rcu(&prog->aux->rcu);
2322         }
2323 }
2324
2325 static void bpf_prog_put_deferred(struct work_struct *work)
2326 {
2327         struct bpf_prog_aux *aux;
2328         struct bpf_prog *prog;
2329
2330         aux = container_of(work, struct bpf_prog_aux, work);
2331         prog = aux->prog;
2332         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2333         bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2334         bpf_prog_free_id(prog);
2335         __bpf_prog_put_noref(prog, true);
2336 }
2337
2338 static void __bpf_prog_put(struct bpf_prog *prog)
2339 {
2340         struct bpf_prog_aux *aux = prog->aux;
2341
2342         if (atomic64_dec_and_test(&aux->refcnt)) {
2343                 if (in_irq() || irqs_disabled()) {
2344                         INIT_WORK(&aux->work, bpf_prog_put_deferred);
2345                         schedule_work(&aux->work);
2346                 } else {
2347                         bpf_prog_put_deferred(&aux->work);
2348                 }
2349         }
2350 }
2351
2352 void bpf_prog_put(struct bpf_prog *prog)
2353 {
2354         __bpf_prog_put(prog);
2355 }
2356 EXPORT_SYMBOL_GPL(bpf_prog_put);
2357
2358 static int bpf_prog_release(struct inode *inode, struct file *filp)
2359 {
2360         struct bpf_prog *prog = filp->private_data;
2361
2362         bpf_prog_put(prog);
2363         return 0;
2364 }
2365
2366 struct bpf_prog_kstats {
2367         u64 nsecs;
2368         u64 cnt;
2369         u64 misses;
2370 };
2371
2372 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2373 {
2374         struct bpf_prog_stats *stats;
2375         unsigned int flags;
2376
2377         stats = this_cpu_ptr(prog->stats);
2378         flags = u64_stats_update_begin_irqsave(&stats->syncp);
2379         u64_stats_inc(&stats->misses);
2380         u64_stats_update_end_irqrestore(&stats->syncp, flags);
2381 }
2382
2383 static void bpf_prog_get_stats(const struct bpf_prog *prog,
2384                                struct bpf_prog_kstats *stats)
2385 {
2386         u64 nsecs = 0, cnt = 0, misses = 0;
2387         int cpu;
2388
2389         for_each_possible_cpu(cpu) {
2390                 const struct bpf_prog_stats *st;
2391                 unsigned int start;
2392                 u64 tnsecs, tcnt, tmisses;
2393
2394                 st = per_cpu_ptr(prog->stats, cpu);
2395                 do {
2396                         start = u64_stats_fetch_begin(&st->syncp);
2397                         tnsecs = u64_stats_read(&st->nsecs);
2398                         tcnt = u64_stats_read(&st->cnt);
2399                         tmisses = u64_stats_read(&st->misses);
2400                 } while (u64_stats_fetch_retry(&st->syncp, start));
2401                 nsecs += tnsecs;
2402                 cnt += tcnt;
2403                 misses += tmisses;
2404         }
2405         stats->nsecs = nsecs;
2406         stats->cnt = cnt;
2407         stats->misses = misses;
2408 }
2409
2410 #ifdef CONFIG_PROC_FS
2411 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2412 {
2413         const struct bpf_prog *prog = filp->private_data;
2414         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2415         struct bpf_prog_kstats stats;
2416
2417         bpf_prog_get_stats(prog, &stats);
2418         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2419         seq_printf(m,
2420                    "prog_type:\t%u\n"
2421                    "prog_jited:\t%u\n"
2422                    "prog_tag:\t%s\n"
2423                    "memlock:\t%llu\n"
2424                    "prog_id:\t%u\n"
2425                    "run_time_ns:\t%llu\n"
2426                    "run_cnt:\t%llu\n"
2427                    "recursion_misses:\t%llu\n"
2428                    "verified_insns:\t%u\n",
2429                    prog->type,
2430                    prog->jited,
2431                    prog_tag,
2432                    prog->pages * 1ULL << PAGE_SHIFT,
2433                    prog->aux->id,
2434                    stats.nsecs,
2435                    stats.cnt,
2436                    stats.misses,
2437                    prog->aux->verified_insns);
2438 }
2439 #endif
2440
2441 const struct file_operations bpf_prog_fops = {
2442 #ifdef CONFIG_PROC_FS
2443         .show_fdinfo    = bpf_prog_show_fdinfo,
2444 #endif
2445         .release        = bpf_prog_release,
2446         .read           = bpf_dummy_read,
2447         .write          = bpf_dummy_write,
2448 };
2449
2450 int bpf_prog_new_fd(struct bpf_prog *prog)
2451 {
2452         int ret;
2453
2454         ret = security_bpf_prog(prog);
2455         if (ret < 0)
2456                 return ret;
2457
2458         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2459                                 O_RDWR | O_CLOEXEC);
2460 }
2461
2462 void bpf_prog_add(struct bpf_prog *prog, int i)
2463 {
2464         atomic64_add(i, &prog->aux->refcnt);
2465 }
2466 EXPORT_SYMBOL_GPL(bpf_prog_add);
2467
2468 void bpf_prog_sub(struct bpf_prog *prog, int i)
2469 {
2470         /* Only to be used for undoing previous bpf_prog_add() in some
2471          * error path. We still know that another entity in our call
2472          * path holds a reference to the program, thus atomic_sub() can
2473          * be safely used in such cases!
2474          */
2475         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2476 }
2477 EXPORT_SYMBOL_GPL(bpf_prog_sub);
2478
2479 void bpf_prog_inc(struct bpf_prog *prog)
2480 {
2481         atomic64_inc(&prog->aux->refcnt);
2482 }
2483 EXPORT_SYMBOL_GPL(bpf_prog_inc);
2484
2485 /* prog_idr_lock should have been held */
2486 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2487 {
2488         int refold;
2489
2490         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2491
2492         if (!refold)
2493                 return ERR_PTR(-ENOENT);
2494
2495         return prog;
2496 }
2497 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2498
2499 bool bpf_prog_get_ok(struct bpf_prog *prog,
2500                             enum bpf_prog_type *attach_type, bool attach_drv)
2501 {
2502         /* not an attachment, just a refcount inc, always allow */
2503         if (!attach_type)
2504                 return true;
2505
2506         if (prog->type != *attach_type)
2507                 return false;
2508         if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2509                 return false;
2510
2511         return true;
2512 }
2513
2514 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2515                                        bool attach_drv)
2516 {
2517         CLASS(fd, f)(ufd);
2518         struct bpf_prog *prog;
2519
2520         if (fd_empty(f))
2521                 return ERR_PTR(-EBADF);
2522         if (fd_file(f)->f_op != &bpf_prog_fops)
2523                 return ERR_PTR(-EINVAL);
2524
2525         prog = fd_file(f)->private_data;
2526         if (!bpf_prog_get_ok(prog, attach_type, attach_drv))
2527                 return ERR_PTR(-EINVAL);
2528
2529         bpf_prog_inc(prog);
2530         return prog;
2531 }
2532
2533 struct bpf_prog *bpf_prog_get(u32 ufd)
2534 {
2535         return __bpf_prog_get(ufd, NULL, false);
2536 }
2537
2538 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2539                                        bool attach_drv)
2540 {
2541         return __bpf_prog_get(ufd, &type, attach_drv);
2542 }
2543 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2544
2545 /* Initially all BPF programs could be loaded w/o specifying
2546  * expected_attach_type. Later for some of them specifying expected_attach_type
2547  * at load time became required so that program could be validated properly.
2548  * Programs of types that are allowed to be loaded both w/ and w/o (for
2549  * backward compatibility) expected_attach_type, should have the default attach
2550  * type assigned to expected_attach_type for the latter case, so that it can be
2551  * validated later at attach time.
2552  *
2553  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2554  * prog type requires it but has some attach types that have to be backward
2555  * compatible.
2556  */
2557 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2558 {
2559         switch (attr->prog_type) {
2560         case BPF_PROG_TYPE_CGROUP_SOCK:
2561                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2562                  * exist so checking for non-zero is the way to go here.
2563                  */
2564                 if (!attr->expected_attach_type)
2565                         attr->expected_attach_type =
2566                                 BPF_CGROUP_INET_SOCK_CREATE;
2567                 break;
2568         case BPF_PROG_TYPE_SK_REUSEPORT:
2569                 if (!attr->expected_attach_type)
2570                         attr->expected_attach_type =
2571                                 BPF_SK_REUSEPORT_SELECT;
2572                 break;
2573         }
2574 }
2575
2576 static int
2577 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2578                            enum bpf_attach_type expected_attach_type,
2579                            struct btf *attach_btf, u32 btf_id,
2580                            struct bpf_prog *dst_prog)
2581 {
2582         if (btf_id) {
2583                 if (btf_id > BTF_MAX_TYPE)
2584                         return -EINVAL;
2585
2586                 if (!attach_btf && !dst_prog)
2587                         return -EINVAL;
2588
2589                 switch (prog_type) {
2590                 case BPF_PROG_TYPE_TRACING:
2591                 case BPF_PROG_TYPE_LSM:
2592                 case BPF_PROG_TYPE_STRUCT_OPS:
2593                 case BPF_PROG_TYPE_EXT:
2594                         break;
2595                 default:
2596                         return -EINVAL;
2597                 }
2598         }
2599
2600         if (attach_btf && (!btf_id || dst_prog))
2601                 return -EINVAL;
2602
2603         if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2604             prog_type != BPF_PROG_TYPE_EXT)
2605                 return -EINVAL;
2606
2607         switch (prog_type) {
2608         case BPF_PROG_TYPE_CGROUP_SOCK:
2609                 switch (expected_attach_type) {
2610                 case BPF_CGROUP_INET_SOCK_CREATE:
2611                 case BPF_CGROUP_INET_SOCK_RELEASE:
2612                 case BPF_CGROUP_INET4_POST_BIND:
2613                 case BPF_CGROUP_INET6_POST_BIND:
2614                         return 0;
2615                 default:
2616                         return -EINVAL;
2617                 }
2618         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2619                 switch (expected_attach_type) {
2620                 case BPF_CGROUP_INET4_BIND:
2621                 case BPF_CGROUP_INET6_BIND:
2622                 case BPF_CGROUP_INET4_CONNECT:
2623                 case BPF_CGROUP_INET6_CONNECT:
2624                 case BPF_CGROUP_UNIX_CONNECT:
2625                 case BPF_CGROUP_INET4_GETPEERNAME:
2626                 case BPF_CGROUP_INET6_GETPEERNAME:
2627                 case BPF_CGROUP_UNIX_GETPEERNAME:
2628                 case BPF_CGROUP_INET4_GETSOCKNAME:
2629                 case BPF_CGROUP_INET6_GETSOCKNAME:
2630                 case BPF_CGROUP_UNIX_GETSOCKNAME:
2631                 case BPF_CGROUP_UDP4_SENDMSG:
2632                 case BPF_CGROUP_UDP6_SENDMSG:
2633                 case BPF_CGROUP_UNIX_SENDMSG:
2634                 case BPF_CGROUP_UDP4_RECVMSG:
2635                 case BPF_CGROUP_UDP6_RECVMSG:
2636                 case BPF_CGROUP_UNIX_RECVMSG:
2637                         return 0;
2638                 default:
2639                         return -EINVAL;
2640                 }
2641         case BPF_PROG_TYPE_CGROUP_SKB:
2642                 switch (expected_attach_type) {
2643                 case BPF_CGROUP_INET_INGRESS:
2644                 case BPF_CGROUP_INET_EGRESS:
2645                         return 0;
2646                 default:
2647                         return -EINVAL;
2648                 }
2649         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2650                 switch (expected_attach_type) {
2651                 case BPF_CGROUP_SETSOCKOPT:
2652                 case BPF_CGROUP_GETSOCKOPT:
2653                         return 0;
2654                 default:
2655                         return -EINVAL;
2656                 }
2657         case BPF_PROG_TYPE_SK_LOOKUP:
2658                 if (expected_attach_type == BPF_SK_LOOKUP)
2659                         return 0;
2660                 return -EINVAL;
2661         case BPF_PROG_TYPE_SK_REUSEPORT:
2662                 switch (expected_attach_type) {
2663                 case BPF_SK_REUSEPORT_SELECT:
2664                 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2665                         return 0;
2666                 default:
2667                         return -EINVAL;
2668                 }
2669         case BPF_PROG_TYPE_NETFILTER:
2670                 if (expected_attach_type == BPF_NETFILTER)
2671                         return 0;
2672                 return -EINVAL;
2673         case BPF_PROG_TYPE_SYSCALL:
2674         case BPF_PROG_TYPE_EXT:
2675                 if (expected_attach_type)
2676                         return -EINVAL;
2677                 fallthrough;
2678         default:
2679                 return 0;
2680         }
2681 }
2682
2683 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2684 {
2685         switch (prog_type) {
2686         case BPF_PROG_TYPE_SCHED_CLS:
2687         case BPF_PROG_TYPE_SCHED_ACT:
2688         case BPF_PROG_TYPE_XDP:
2689         case BPF_PROG_TYPE_LWT_IN:
2690         case BPF_PROG_TYPE_LWT_OUT:
2691         case BPF_PROG_TYPE_LWT_XMIT:
2692         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2693         case BPF_PROG_TYPE_SK_SKB:
2694         case BPF_PROG_TYPE_SK_MSG:
2695         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2696         case BPF_PROG_TYPE_CGROUP_DEVICE:
2697         case BPF_PROG_TYPE_CGROUP_SOCK:
2698         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2699         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2700         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2701         case BPF_PROG_TYPE_SOCK_OPS:
2702         case BPF_PROG_TYPE_EXT: /* extends any prog */
2703         case BPF_PROG_TYPE_NETFILTER:
2704                 return true;
2705         case BPF_PROG_TYPE_CGROUP_SKB:
2706                 /* always unpriv */
2707         case BPF_PROG_TYPE_SK_REUSEPORT:
2708                 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2709         default:
2710                 return false;
2711         }
2712 }
2713
2714 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2715 {
2716         switch (prog_type) {
2717         case BPF_PROG_TYPE_KPROBE:
2718         case BPF_PROG_TYPE_TRACEPOINT:
2719         case BPF_PROG_TYPE_PERF_EVENT:
2720         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2721         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2722         case BPF_PROG_TYPE_TRACING:
2723         case BPF_PROG_TYPE_LSM:
2724         case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2725         case BPF_PROG_TYPE_EXT: /* extends any prog */
2726                 return true;
2727         default:
2728                 return false;
2729         }
2730 }
2731
2732 /* last field in 'union bpf_attr' used by this command */
2733 #define BPF_PROG_LOAD_LAST_FIELD prog_token_fd
2734
2735 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
2736 {
2737         enum bpf_prog_type type = attr->prog_type;
2738         struct bpf_prog *prog, *dst_prog = NULL;
2739         struct btf *attach_btf = NULL;
2740         struct bpf_token *token = NULL;
2741         bool bpf_cap;
2742         int err;
2743         char license[128];
2744
2745         if (CHECK_ATTR(BPF_PROG_LOAD))
2746                 return -EINVAL;
2747
2748         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2749                                  BPF_F_ANY_ALIGNMENT |
2750                                  BPF_F_TEST_STATE_FREQ |
2751                                  BPF_F_SLEEPABLE |
2752                                  BPF_F_TEST_RND_HI32 |
2753                                  BPF_F_XDP_HAS_FRAGS |
2754                                  BPF_F_XDP_DEV_BOUND_ONLY |
2755                                  BPF_F_TEST_REG_INVARIANTS |
2756                                  BPF_F_TOKEN_FD))
2757                 return -EINVAL;
2758
2759         bpf_prog_load_fixup_attach_type(attr);
2760
2761         if (attr->prog_flags & BPF_F_TOKEN_FD) {
2762                 token = bpf_token_get_from_fd(attr->prog_token_fd);
2763                 if (IS_ERR(token))
2764                         return PTR_ERR(token);
2765                 /* if current token doesn't grant prog loading permissions,
2766                  * then we can't use this token, so ignore it and rely on
2767                  * system-wide capabilities checks
2768                  */
2769                 if (!bpf_token_allow_cmd(token, BPF_PROG_LOAD) ||
2770                     !bpf_token_allow_prog_type(token, attr->prog_type,
2771                                                attr->expected_attach_type)) {
2772                         bpf_token_put(token);
2773                         token = NULL;
2774                 }
2775         }
2776
2777         bpf_cap = bpf_token_capable(token, CAP_BPF);
2778         err = -EPERM;
2779
2780         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2781             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2782             !bpf_cap)
2783                 goto put_token;
2784
2785         /* Intent here is for unprivileged_bpf_disabled to block BPF program
2786          * creation for unprivileged users; other actions depend
2787          * on fd availability and access to bpffs, so are dependent on
2788          * object creation success. Even with unprivileged BPF disabled,
2789          * capability checks are still carried out for these
2790          * and other operations.
2791          */
2792         if (sysctl_unprivileged_bpf_disabled && !bpf_cap)
2793                 goto put_token;
2794
2795         if (attr->insn_cnt == 0 ||
2796             attr->insn_cnt > (bpf_cap ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS)) {
2797                 err = -E2BIG;
2798                 goto put_token;
2799         }
2800         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2801             type != BPF_PROG_TYPE_CGROUP_SKB &&
2802             !bpf_cap)
2803                 goto put_token;
2804
2805         if (is_net_admin_prog_type(type) && !bpf_token_capable(token, CAP_NET_ADMIN))
2806                 goto put_token;
2807         if (is_perfmon_prog_type(type) && !bpf_token_capable(token, CAP_PERFMON))
2808                 goto put_token;
2809
2810         /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2811          * or btf, we need to check which one it is
2812          */
2813         if (attr->attach_prog_fd) {
2814                 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2815                 if (IS_ERR(dst_prog)) {
2816                         dst_prog = NULL;
2817                         attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2818                         if (IS_ERR(attach_btf)) {
2819                                 err = -EINVAL;
2820                                 goto put_token;
2821                         }
2822                         if (!btf_is_kernel(attach_btf)) {
2823                                 /* attaching through specifying bpf_prog's BTF
2824                                  * objects directly might be supported eventually
2825                                  */
2826                                 btf_put(attach_btf);
2827                                 err = -ENOTSUPP;
2828                                 goto put_token;
2829                         }
2830                 }
2831         } else if (attr->attach_btf_id) {
2832                 /* fall back to vmlinux BTF, if BTF type ID is specified */
2833                 attach_btf = bpf_get_btf_vmlinux();
2834                 if (IS_ERR(attach_btf)) {
2835                         err = PTR_ERR(attach_btf);
2836                         goto put_token;
2837                 }
2838                 if (!attach_btf) {
2839                         err = -EINVAL;
2840                         goto put_token;
2841                 }
2842                 btf_get(attach_btf);
2843         }
2844
2845         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2846                                        attach_btf, attr->attach_btf_id,
2847                                        dst_prog)) {
2848                 if (dst_prog)
2849                         bpf_prog_put(dst_prog);
2850                 if (attach_btf)
2851                         btf_put(attach_btf);
2852                 err = -EINVAL;
2853                 goto put_token;
2854         }
2855
2856         /* plain bpf_prog allocation */
2857         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2858         if (!prog) {
2859                 if (dst_prog)
2860                         bpf_prog_put(dst_prog);
2861                 if (attach_btf)
2862                         btf_put(attach_btf);
2863                 err = -EINVAL;
2864                 goto put_token;
2865         }
2866
2867         prog->expected_attach_type = attr->expected_attach_type;
2868         prog->sleepable = !!(attr->prog_flags & BPF_F_SLEEPABLE);
2869         prog->aux->attach_btf = attach_btf;
2870         prog->aux->attach_btf_id = attr->attach_btf_id;
2871         prog->aux->dst_prog = dst_prog;
2872         prog->aux->dev_bound = !!attr->prog_ifindex;
2873         prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2874
2875         /* move token into prog->aux, reuse taken refcnt */
2876         prog->aux->token = token;
2877         token = NULL;
2878
2879         prog->aux->user = get_current_user();
2880         prog->len = attr->insn_cnt;
2881
2882         err = -EFAULT;
2883         if (copy_from_bpfptr(prog->insns,
2884                              make_bpfptr(attr->insns, uattr.is_kernel),
2885                              bpf_prog_insn_size(prog)) != 0)
2886                 goto free_prog;
2887         /* copy eBPF program license from user space */
2888         if (strncpy_from_bpfptr(license,
2889                                 make_bpfptr(attr->license, uattr.is_kernel),
2890                                 sizeof(license) - 1) < 0)
2891                 goto free_prog;
2892         license[sizeof(license) - 1] = 0;
2893
2894         /* eBPF programs must be GPL compatible to use GPL-ed functions */
2895         prog->gpl_compatible = license_is_gpl_compatible(license) ? 1 : 0;
2896
2897         prog->orig_prog = NULL;
2898         prog->jited = 0;
2899
2900         atomic64_set(&prog->aux->refcnt, 1);
2901
2902         if (bpf_prog_is_dev_bound(prog->aux)) {
2903                 err = bpf_prog_dev_bound_init(prog, attr);
2904                 if (err)
2905                         goto free_prog;
2906         }
2907
2908         if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2909             bpf_prog_is_dev_bound(dst_prog->aux)) {
2910                 err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2911                 if (err)
2912                         goto free_prog;
2913         }
2914
2915         /*
2916          * Bookkeeping for managing the program attachment chain.
2917          *
2918          * It might be tempting to set attach_tracing_prog flag at the attachment
2919          * time, but this will not prevent from loading bunch of tracing prog
2920          * first, then attach them one to another.
2921          *
2922          * The flag attach_tracing_prog is set for the whole program lifecycle, and
2923          * doesn't have to be cleared in bpf_tracing_link_release, since tracing
2924          * programs cannot change attachment target.
2925          */
2926         if (type == BPF_PROG_TYPE_TRACING && dst_prog &&
2927             dst_prog->type == BPF_PROG_TYPE_TRACING) {
2928                 prog->aux->attach_tracing_prog = true;
2929         }
2930
2931         /* find program type: socket_filter vs tracing_filter */
2932         err = find_prog_type(type, prog);
2933         if (err < 0)
2934                 goto free_prog;
2935
2936         prog->aux->load_time = ktime_get_boottime_ns();
2937         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2938                                sizeof(attr->prog_name));
2939         if (err < 0)
2940                 goto free_prog;
2941
2942         err = security_bpf_prog_load(prog, attr, token);
2943         if (err)
2944                 goto free_prog_sec;
2945
2946         /* run eBPF verifier */
2947         err = bpf_check(&prog, attr, uattr, uattr_size);
2948         if (err < 0)
2949                 goto free_used_maps;
2950
2951         prog = bpf_prog_select_runtime(prog, &err);
2952         if (err < 0)
2953                 goto free_used_maps;
2954
2955         err = bpf_prog_alloc_id(prog);
2956         if (err)
2957                 goto free_used_maps;
2958
2959         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2960          * effectively publicly exposed. However, retrieving via
2961          * bpf_prog_get_fd_by_id() will take another reference,
2962          * therefore it cannot be gone underneath us.
2963          *
2964          * Only for the time /after/ successful bpf_prog_new_fd()
2965          * and before returning to userspace, we might just hold
2966          * one reference and any parallel close on that fd could
2967          * rip everything out. Hence, below notifications must
2968          * happen before bpf_prog_new_fd().
2969          *
2970          * Also, any failure handling from this point onwards must
2971          * be using bpf_prog_put() given the program is exposed.
2972          */
2973         bpf_prog_kallsyms_add(prog);
2974         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2975         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2976
2977         err = bpf_prog_new_fd(prog);
2978         if (err < 0)
2979                 bpf_prog_put(prog);
2980         return err;
2981
2982 free_used_maps:
2983         /* In case we have subprogs, we need to wait for a grace
2984          * period before we can tear down JIT memory since symbols
2985          * are already exposed under kallsyms.
2986          */
2987         __bpf_prog_put_noref(prog, prog->aux->real_func_cnt);
2988         return err;
2989
2990 free_prog_sec:
2991         security_bpf_prog_free(prog);
2992 free_prog:
2993         free_uid(prog->aux->user);
2994         if (prog->aux->attach_btf)
2995                 btf_put(prog->aux->attach_btf);
2996         bpf_prog_free(prog);
2997 put_token:
2998         bpf_token_put(token);
2999         return err;
3000 }
3001
3002 #define BPF_OBJ_LAST_FIELD path_fd
3003
3004 static int bpf_obj_pin(const union bpf_attr *attr)
3005 {
3006         int path_fd;
3007
3008         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags & ~BPF_F_PATH_FD)
3009                 return -EINVAL;
3010
3011         /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
3012         if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
3013                 return -EINVAL;
3014
3015         path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
3016         return bpf_obj_pin_user(attr->bpf_fd, path_fd,
3017                                 u64_to_user_ptr(attr->pathname));
3018 }
3019
3020 static int bpf_obj_get(const union bpf_attr *attr)
3021 {
3022         int path_fd;
3023
3024         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
3025             attr->file_flags & ~(BPF_OBJ_FLAG_MASK | BPF_F_PATH_FD))
3026                 return -EINVAL;
3027
3028         /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
3029         if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
3030                 return -EINVAL;
3031
3032         path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
3033         return bpf_obj_get_user(path_fd, u64_to_user_ptr(attr->pathname),
3034                                 attr->file_flags);
3035 }
3036
3037 /* bpf_link_init_sleepable() allows to specify whether BPF link itself has
3038  * "sleepable" semantics, which normally would mean that BPF link's attach
3039  * hook can dereference link or link's underlying program for some time after
3040  * detachment due to RCU Tasks Trace-based lifetime protection scheme.
3041  * BPF program itself can be non-sleepable, yet, because it's transitively
3042  * reachable through BPF link, its freeing has to be delayed until after RCU
3043  * Tasks Trace GP.
3044  */
3045 void bpf_link_init_sleepable(struct bpf_link *link, enum bpf_link_type type,
3046                              const struct bpf_link_ops *ops, struct bpf_prog *prog,
3047                              bool sleepable)
3048 {
3049         WARN_ON(ops->dealloc && ops->dealloc_deferred);
3050         atomic64_set(&link->refcnt, 1);
3051         link->type = type;
3052         link->sleepable = sleepable;
3053         link->id = 0;
3054         link->ops = ops;
3055         link->prog = prog;
3056 }
3057
3058 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
3059                    const struct bpf_link_ops *ops, struct bpf_prog *prog)
3060 {
3061         bpf_link_init_sleepable(link, type, ops, prog, false);
3062 }
3063
3064 static void bpf_link_free_id(int id)
3065 {
3066         if (!id)
3067                 return;
3068
3069         spin_lock_bh(&link_idr_lock);
3070         idr_remove(&link_idr, id);
3071         spin_unlock_bh(&link_idr_lock);
3072 }
3073
3074 /* Clean up bpf_link and corresponding anon_inode file and FD. After
3075  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
3076  * anon_inode's release() call. This helper marks bpf_link as
3077  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
3078  * is not decremented, it's the responsibility of a calling code that failed
3079  * to complete bpf_link initialization.
3080  * This helper eventually calls link's dealloc callback, but does not call
3081  * link's release callback.
3082  */
3083 void bpf_link_cleanup(struct bpf_link_primer *primer)
3084 {
3085         primer->link->prog = NULL;
3086         bpf_link_free_id(primer->id);
3087         fput(primer->file);
3088         put_unused_fd(primer->fd);
3089 }
3090
3091 void bpf_link_inc(struct bpf_link *link)
3092 {
3093         atomic64_inc(&link->refcnt);
3094 }
3095
3096 static void bpf_link_dealloc(struct bpf_link *link)
3097 {
3098         /* now that we know that bpf_link itself can't be reached, put underlying BPF program */
3099         if (link->prog)
3100                 bpf_prog_put(link->prog);
3101
3102         /* free bpf_link and its containing memory */
3103         if (link->ops->dealloc_deferred)
3104                 link->ops->dealloc_deferred(link);
3105         else
3106                 link->ops->dealloc(link);
3107 }
3108
3109 static void bpf_link_defer_dealloc_rcu_gp(struct rcu_head *rcu)
3110 {
3111         struct bpf_link *link = container_of(rcu, struct bpf_link, rcu);
3112
3113         bpf_link_dealloc(link);
3114 }
3115
3116 static void bpf_link_defer_dealloc_mult_rcu_gp(struct rcu_head *rcu)
3117 {
3118         if (rcu_trace_implies_rcu_gp())
3119                 bpf_link_defer_dealloc_rcu_gp(rcu);
3120         else
3121                 call_rcu(rcu, bpf_link_defer_dealloc_rcu_gp);
3122 }
3123
3124 /* bpf_link_free is guaranteed to be called from process context */
3125 static void bpf_link_free(struct bpf_link *link)
3126 {
3127         const struct bpf_link_ops *ops = link->ops;
3128
3129         bpf_link_free_id(link->id);
3130         /* detach BPF program, clean up used resources */
3131         if (link->prog)
3132                 ops->release(link);
3133         if (ops->dealloc_deferred) {
3134                 /* Schedule BPF link deallocation, which will only then
3135                  * trigger putting BPF program refcount.
3136                  * If underlying BPF program is sleepable or BPF link's target
3137                  * attach hookpoint is sleepable or otherwise requires RCU GPs
3138                  * to ensure link and its underlying BPF program is not
3139                  * reachable anymore, we need to first wait for RCU tasks
3140                  * trace sync, and then go through "classic" RCU grace period
3141                  */
3142                 if (link->sleepable || (link->prog && link->prog->sleepable))
3143                         call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp);
3144                 else
3145                         call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
3146         } else if (ops->dealloc) {
3147                 bpf_link_dealloc(link);
3148         }
3149 }
3150
3151 static void bpf_link_put_deferred(struct work_struct *work)
3152 {
3153         struct bpf_link *link = container_of(work, struct bpf_link, work);
3154
3155         bpf_link_free(link);
3156 }
3157
3158 /* bpf_link_put might be called from atomic context. It needs to be called
3159  * from sleepable context in order to acquire sleeping locks during the process.
3160  */
3161 void bpf_link_put(struct bpf_link *link)
3162 {
3163         if (!atomic64_dec_and_test(&link->refcnt))
3164                 return;
3165
3166         INIT_WORK(&link->work, bpf_link_put_deferred);
3167         schedule_work(&link->work);
3168 }
3169 EXPORT_SYMBOL(bpf_link_put);
3170
3171 static void bpf_link_put_direct(struct bpf_link *link)
3172 {
3173         if (!atomic64_dec_and_test(&link->refcnt))
3174                 return;
3175         bpf_link_free(link);
3176 }
3177
3178 static int bpf_link_release(struct inode *inode, struct file *filp)
3179 {
3180         struct bpf_link *link = filp->private_data;
3181
3182         bpf_link_put_direct(link);
3183         return 0;
3184 }
3185
3186 #ifdef CONFIG_PROC_FS
3187 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
3188 #define BPF_MAP_TYPE(_id, _ops)
3189 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
3190 static const char *bpf_link_type_strs[] = {
3191         [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
3192 #include <linux/bpf_types.h>
3193 };
3194 #undef BPF_PROG_TYPE
3195 #undef BPF_MAP_TYPE
3196 #undef BPF_LINK_TYPE
3197
3198 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
3199 {
3200         const struct bpf_link *link = filp->private_data;
3201         const struct bpf_prog *prog = link->prog;
3202         enum bpf_link_type type = link->type;
3203         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
3204
3205         if (type < ARRAY_SIZE(bpf_link_type_strs) && bpf_link_type_strs[type]) {
3206                 seq_printf(m, "link_type:\t%s\n", bpf_link_type_strs[type]);
3207         } else {
3208                 WARN_ONCE(1, "missing BPF_LINK_TYPE(...) for link type %u\n", type);
3209                 seq_printf(m, "link_type:\t<%u>\n", type);
3210         }
3211         seq_printf(m, "link_id:\t%u\n", link->id);
3212
3213         if (prog) {
3214                 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
3215                 seq_printf(m,
3216                            "prog_tag:\t%s\n"
3217                            "prog_id:\t%u\n",
3218                            prog_tag,
3219                            prog->aux->id);
3220         }
3221         if (link->ops->show_fdinfo)
3222                 link->ops->show_fdinfo(link, m);
3223 }
3224 #endif
3225
3226 static __poll_t bpf_link_poll(struct file *file, struct poll_table_struct *pts)
3227 {
3228         struct bpf_link *link = file->private_data;
3229
3230         return link->ops->poll(file, pts);
3231 }
3232
3233 static const struct file_operations bpf_link_fops = {
3234 #ifdef CONFIG_PROC_FS
3235         .show_fdinfo    = bpf_link_show_fdinfo,
3236 #endif
3237         .release        = bpf_link_release,
3238         .read           = bpf_dummy_read,
3239         .write          = bpf_dummy_write,
3240 };
3241
3242 static const struct file_operations bpf_link_fops_poll = {
3243 #ifdef CONFIG_PROC_FS
3244         .show_fdinfo    = bpf_link_show_fdinfo,
3245 #endif
3246         .release        = bpf_link_release,
3247         .read           = bpf_dummy_read,
3248         .write          = bpf_dummy_write,
3249         .poll           = bpf_link_poll,
3250 };
3251
3252 static int bpf_link_alloc_id(struct bpf_link *link)
3253 {
3254         int id;
3255
3256         idr_preload(GFP_KERNEL);
3257         spin_lock_bh(&link_idr_lock);
3258         id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
3259         spin_unlock_bh(&link_idr_lock);
3260         idr_preload_end();
3261
3262         return id;
3263 }
3264
3265 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
3266  * reserving unused FD and allocating ID from link_idr. This is to be paired
3267  * with bpf_link_settle() to install FD and ID and expose bpf_link to
3268  * user-space, if bpf_link is successfully attached. If not, bpf_link and
3269  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
3270  * transient state is passed around in struct bpf_link_primer.
3271  * This is preferred way to create and initialize bpf_link, especially when
3272  * there are complicated and expensive operations in between creating bpf_link
3273  * itself and attaching it to BPF hook. By using bpf_link_prime() and
3274  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
3275  * expensive (and potentially failing) roll back operations in a rare case
3276  * that file, FD, or ID can't be allocated.
3277  */
3278 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
3279 {
3280         struct file *file;
3281         int fd, id;
3282
3283         fd = get_unused_fd_flags(O_CLOEXEC);
3284         if (fd < 0)
3285                 return fd;
3286
3287
3288         id = bpf_link_alloc_id(link);
3289         if (id < 0) {
3290                 put_unused_fd(fd);
3291                 return id;
3292         }
3293
3294         file = anon_inode_getfile("bpf_link",
3295                                   link->ops->poll ? &bpf_link_fops_poll : &bpf_link_fops,
3296                                   link, O_CLOEXEC);
3297         if (IS_ERR(file)) {
3298                 bpf_link_free_id(id);
3299                 put_unused_fd(fd);
3300                 return PTR_ERR(file);
3301         }
3302
3303         primer->link = link;
3304         primer->file = file;
3305         primer->fd = fd;
3306         primer->id = id;
3307         return 0;
3308 }
3309
3310 int bpf_link_settle(struct bpf_link_primer *primer)
3311 {
3312         /* make bpf_link fetchable by ID */
3313         spin_lock_bh(&link_idr_lock);
3314         primer->link->id = primer->id;
3315         spin_unlock_bh(&link_idr_lock);
3316         /* make bpf_link fetchable by FD */
3317         fd_install(primer->fd, primer->file);
3318         /* pass through installed FD */
3319         return primer->fd;
3320 }
3321
3322 int bpf_link_new_fd(struct bpf_link *link)
3323 {
3324         return anon_inode_getfd("bpf-link",
3325                                 link->ops->poll ? &bpf_link_fops_poll : &bpf_link_fops,
3326                                 link, O_CLOEXEC);
3327 }
3328
3329 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
3330 {
3331         CLASS(fd, f)(ufd);
3332         struct bpf_link *link;
3333
3334         if (fd_empty(f))
3335                 return ERR_PTR(-EBADF);
3336         if (fd_file(f)->f_op != &bpf_link_fops && fd_file(f)->f_op != &bpf_link_fops_poll)
3337                 return ERR_PTR(-EINVAL);
3338
3339         link = fd_file(f)->private_data;
3340         bpf_link_inc(link);
3341         return link;
3342 }
3343 EXPORT_SYMBOL(bpf_link_get_from_fd);
3344
3345 static void bpf_tracing_link_release(struct bpf_link *link)
3346 {
3347         struct bpf_tracing_link *tr_link =
3348                 container_of(link, struct bpf_tracing_link, link.link);
3349
3350         WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
3351                                                 tr_link->trampoline,
3352                                                 tr_link->tgt_prog));
3353
3354         bpf_trampoline_put(tr_link->trampoline);
3355
3356         /* tgt_prog is NULL if target is a kernel function */
3357         if (tr_link->tgt_prog)
3358                 bpf_prog_put(tr_link->tgt_prog);
3359 }
3360
3361 static void bpf_tracing_link_dealloc(struct bpf_link *link)
3362 {
3363         struct bpf_tracing_link *tr_link =
3364                 container_of(link, struct bpf_tracing_link, link.link);
3365
3366         kfree(tr_link);
3367 }
3368
3369 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
3370                                          struct seq_file *seq)
3371 {
3372         struct bpf_tracing_link *tr_link =
3373                 container_of(link, struct bpf_tracing_link, link.link);
3374         u32 target_btf_id, target_obj_id;
3375
3376         bpf_trampoline_unpack_key(tr_link->trampoline->key,
3377                                   &target_obj_id, &target_btf_id);
3378         seq_printf(seq,
3379                    "attach_type:\t%d\n"
3380                    "target_obj_id:\t%u\n"
3381                    "target_btf_id:\t%u\n",
3382                    tr_link->attach_type,
3383                    target_obj_id,
3384                    target_btf_id);
3385 }
3386
3387 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
3388                                            struct bpf_link_info *info)
3389 {
3390         struct bpf_tracing_link *tr_link =
3391                 container_of(link, struct bpf_tracing_link, link.link);
3392
3393         info->tracing.attach_type = tr_link->attach_type;
3394         bpf_trampoline_unpack_key(tr_link->trampoline->key,
3395                                   &info->tracing.target_obj_id,
3396                                   &info->tracing.target_btf_id);
3397
3398         return 0;
3399 }
3400
3401 static const struct bpf_link_ops bpf_tracing_link_lops = {
3402         .release = bpf_tracing_link_release,
3403         .dealloc = bpf_tracing_link_dealloc,
3404         .show_fdinfo = bpf_tracing_link_show_fdinfo,
3405         .fill_link_info = bpf_tracing_link_fill_link_info,
3406 };
3407
3408 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
3409                                    int tgt_prog_fd,
3410                                    u32 btf_id,
3411                                    u64 bpf_cookie)
3412 {
3413         struct bpf_link_primer link_primer;
3414         struct bpf_prog *tgt_prog = NULL;
3415         struct bpf_trampoline *tr = NULL;
3416         struct bpf_tracing_link *link;
3417         u64 key = 0;
3418         int err;
3419
3420         switch (prog->type) {
3421         case BPF_PROG_TYPE_TRACING:
3422                 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3423                     prog->expected_attach_type != BPF_TRACE_FEXIT &&
3424                     prog->expected_attach_type != BPF_MODIFY_RETURN) {
3425                         err = -EINVAL;
3426                         goto out_put_prog;
3427                 }
3428                 break;
3429         case BPF_PROG_TYPE_EXT:
3430                 if (prog->expected_attach_type != 0) {
3431                         err = -EINVAL;
3432                         goto out_put_prog;
3433                 }
3434                 break;
3435         case BPF_PROG_TYPE_LSM:
3436                 if (prog->expected_attach_type != BPF_LSM_MAC) {
3437                         err = -EINVAL;
3438                         goto out_put_prog;
3439                 }
3440                 break;
3441         default:
3442                 err = -EINVAL;
3443                 goto out_put_prog;
3444         }
3445
3446         if (!!tgt_prog_fd != !!btf_id) {
3447                 err = -EINVAL;
3448                 goto out_put_prog;
3449         }
3450
3451         if (tgt_prog_fd) {
3452                 /*
3453                  * For now we only allow new targets for BPF_PROG_TYPE_EXT. If this
3454                  * part would be changed to implement the same for
3455                  * BPF_PROG_TYPE_TRACING, do not forget to update the way how
3456                  * attach_tracing_prog flag is set.
3457                  */
3458                 if (prog->type != BPF_PROG_TYPE_EXT) {
3459                         err = -EINVAL;
3460                         goto out_put_prog;
3461                 }
3462
3463                 tgt_prog = bpf_prog_get(tgt_prog_fd);
3464                 if (IS_ERR(tgt_prog)) {
3465                         err = PTR_ERR(tgt_prog);
3466                         tgt_prog = NULL;
3467                         goto out_put_prog;
3468                 }
3469
3470                 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3471         }
3472
3473         link = kzalloc(sizeof(*link), GFP_USER);
3474         if (!link) {
3475                 err = -ENOMEM;
3476                 goto out_put_prog;
3477         }
3478         bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3479                       &bpf_tracing_link_lops, prog);
3480         link->attach_type = prog->expected_attach_type;
3481         link->link.cookie = bpf_cookie;
3482
3483         mutex_lock(&prog->aux->dst_mutex);
3484
3485         /* There are a few possible cases here:
3486          *
3487          * - if prog->aux->dst_trampoline is set, the program was just loaded
3488          *   and not yet attached to anything, so we can use the values stored
3489          *   in prog->aux
3490          *
3491          * - if prog->aux->dst_trampoline is NULL, the program has already been
3492          *   attached to a target and its initial target was cleared (below)
3493          *
3494          * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3495          *   target_btf_id using the link_create API.
3496          *
3497          * - if tgt_prog == NULL when this function was called using the old
3498          *   raw_tracepoint_open API, and we need a target from prog->aux
3499          *
3500          * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3501          *   was detached and is going for re-attachment.
3502          *
3503          * - if prog->aux->dst_trampoline is NULL and tgt_prog and prog->aux->attach_btf
3504          *   are NULL, then program was already attached and user did not provide
3505          *   tgt_prog_fd so we have no way to find out or create trampoline
3506          */
3507         if (!prog->aux->dst_trampoline && !tgt_prog) {
3508                 /*
3509                  * Allow re-attach for TRACING and LSM programs. If it's
3510                  * currently linked, bpf_trampoline_link_prog will fail.
3511                  * EXT programs need to specify tgt_prog_fd, so they
3512                  * re-attach in separate code path.
3513                  */
3514                 if (prog->type != BPF_PROG_TYPE_TRACING &&
3515                     prog->type != BPF_PROG_TYPE_LSM) {
3516                         err = -EINVAL;
3517                         goto out_unlock;
3518                 }
3519                 /* We can allow re-attach only if we have valid attach_btf. */
3520                 if (!prog->aux->attach_btf) {
3521                         err = -EINVAL;
3522                         goto out_unlock;
3523                 }
3524                 btf_id = prog->aux->attach_btf_id;
3525                 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3526         }
3527
3528         if (!prog->aux->dst_trampoline ||
3529             (key && key != prog->aux->dst_trampoline->key)) {
3530                 /* If there is no saved target, or the specified target is
3531                  * different from the destination specified at load time, we
3532                  * need a new trampoline and a check for compatibility
3533                  */
3534                 struct bpf_attach_target_info tgt_info = {};
3535
3536                 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3537                                               &tgt_info);
3538                 if (err)
3539                         goto out_unlock;
3540
3541                 if (tgt_info.tgt_mod) {
3542                         module_put(prog->aux->mod);
3543                         prog->aux->mod = tgt_info.tgt_mod;
3544                 }
3545
3546                 tr = bpf_trampoline_get(key, &tgt_info);
3547                 if (!tr) {
3548                         err = -ENOMEM;
3549                         goto out_unlock;
3550                 }
3551         } else {
3552                 /* The caller didn't specify a target, or the target was the
3553                  * same as the destination supplied during program load. This
3554                  * means we can reuse the trampoline and reference from program
3555                  * load time, and there is no need to allocate a new one. This
3556                  * can only happen once for any program, as the saved values in
3557                  * prog->aux are cleared below.
3558                  */
3559                 tr = prog->aux->dst_trampoline;
3560                 tgt_prog = prog->aux->dst_prog;
3561         }
3562
3563         err = bpf_link_prime(&link->link.link, &link_primer);
3564         if (err)
3565                 goto out_unlock;
3566
3567         err = bpf_trampoline_link_prog(&link->link, tr, tgt_prog);
3568         if (err) {
3569                 bpf_link_cleanup(&link_primer);
3570                 link = NULL;
3571                 goto out_unlock;
3572         }
3573
3574         link->tgt_prog = tgt_prog;
3575         link->trampoline = tr;
3576
3577         /* Always clear the trampoline and target prog from prog->aux to make
3578          * sure the original attach destination is not kept alive after a
3579          * program is (re-)attached to another target.
3580          */
3581         if (prog->aux->dst_prog &&
3582             (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3583                 /* got extra prog ref from syscall, or attaching to different prog */
3584                 bpf_prog_put(prog->aux->dst_prog);
3585         if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3586                 /* we allocated a new trampoline, so free the old one */
3587                 bpf_trampoline_put(prog->aux->dst_trampoline);
3588
3589         prog->aux->dst_prog = NULL;
3590         prog->aux->dst_trampoline = NULL;
3591         mutex_unlock(&prog->aux->dst_mutex);
3592
3593         return bpf_link_settle(&link_primer);
3594 out_unlock:
3595         if (tr && tr != prog->aux->dst_trampoline)
3596                 bpf_trampoline_put(tr);
3597         mutex_unlock(&prog->aux->dst_mutex);
3598         kfree(link);
3599 out_put_prog:
3600         if (tgt_prog_fd && tgt_prog)
3601                 bpf_prog_put(tgt_prog);
3602         return err;
3603 }
3604
3605 static void bpf_raw_tp_link_release(struct bpf_link *link)
3606 {
3607         struct bpf_raw_tp_link *raw_tp =
3608                 container_of(link, struct bpf_raw_tp_link, link);
3609
3610         bpf_probe_unregister(raw_tp->btp, raw_tp);
3611         bpf_put_raw_tracepoint(raw_tp->btp);
3612 }
3613
3614 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3615 {
3616         struct bpf_raw_tp_link *raw_tp =
3617                 container_of(link, struct bpf_raw_tp_link, link);
3618
3619         kfree(raw_tp);
3620 }
3621
3622 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3623                                         struct seq_file *seq)
3624 {
3625         struct bpf_raw_tp_link *raw_tp_link =
3626                 container_of(link, struct bpf_raw_tp_link, link);
3627
3628         seq_printf(seq,
3629                    "tp_name:\t%s\n",
3630                    raw_tp_link->btp->tp->name);
3631 }
3632
3633 static int bpf_copy_to_user(char __user *ubuf, const char *buf, u32 ulen,
3634                             u32 len)
3635 {
3636         if (ulen >= len + 1) {
3637                 if (copy_to_user(ubuf, buf, len + 1))
3638                         return -EFAULT;
3639         } else {
3640                 char zero = '\0';
3641
3642                 if (copy_to_user(ubuf, buf, ulen - 1))
3643                         return -EFAULT;
3644                 if (put_user(zero, ubuf + ulen - 1))
3645                         return -EFAULT;
3646                 return -ENOSPC;
3647         }
3648
3649         return 0;
3650 }
3651
3652 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3653                                           struct bpf_link_info *info)
3654 {
3655         struct bpf_raw_tp_link *raw_tp_link =
3656                 container_of(link, struct bpf_raw_tp_link, link);
3657         char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3658         const char *tp_name = raw_tp_link->btp->tp->name;
3659         u32 ulen = info->raw_tracepoint.tp_name_len;
3660         size_t tp_len = strlen(tp_name);
3661
3662         if (!ulen ^ !ubuf)
3663                 return -EINVAL;
3664
3665         info->raw_tracepoint.tp_name_len = tp_len + 1;
3666
3667         if (!ubuf)
3668                 return 0;
3669
3670         return bpf_copy_to_user(ubuf, tp_name, ulen, tp_len);
3671 }
3672
3673 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3674         .release = bpf_raw_tp_link_release,
3675         .dealloc_deferred = bpf_raw_tp_link_dealloc,
3676         .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3677         .fill_link_info = bpf_raw_tp_link_fill_link_info,
3678 };
3679
3680 #ifdef CONFIG_PERF_EVENTS
3681 struct bpf_perf_link {
3682         struct bpf_link link;
3683         struct file *perf_file;
3684 };
3685
3686 static void bpf_perf_link_release(struct bpf_link *link)
3687 {
3688         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3689         struct perf_event *event = perf_link->perf_file->private_data;
3690
3691         perf_event_free_bpf_prog(event);
3692         fput(perf_link->perf_file);
3693 }
3694
3695 static void bpf_perf_link_dealloc(struct bpf_link *link)
3696 {
3697         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3698
3699         kfree(perf_link);
3700 }
3701
3702 static int bpf_perf_link_fill_common(const struct perf_event *event,
3703                                      char __user *uname, u32 *ulenp,
3704                                      u64 *probe_offset, u64 *probe_addr,
3705                                      u32 *fd_type, unsigned long *missed)
3706 {
3707         const char *buf;
3708         u32 prog_id, ulen;
3709         size_t len;
3710         int err;
3711
3712         ulen = *ulenp;
3713         if (!ulen ^ !uname)
3714                 return -EINVAL;
3715
3716         err = bpf_get_perf_event_info(event, &prog_id, fd_type, &buf,
3717                                       probe_offset, probe_addr, missed);
3718         if (err)
3719                 return err;
3720
3721         if (buf) {
3722                 len = strlen(buf);
3723                 *ulenp = len + 1;
3724         } else {
3725                 *ulenp = 1;
3726         }
3727         if (!uname)
3728                 return 0;
3729
3730         if (buf) {
3731                 err = bpf_copy_to_user(uname, buf, ulen, len);
3732                 if (err)
3733                         return err;
3734         } else {
3735                 char zero = '\0';
3736
3737                 if (put_user(zero, uname))
3738                         return -EFAULT;
3739         }
3740         return 0;
3741 }
3742
3743 #ifdef CONFIG_KPROBE_EVENTS
3744 static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
3745                                      struct bpf_link_info *info)
3746 {
3747         unsigned long missed;
3748         char __user *uname;
3749         u64 addr, offset;
3750         u32 ulen, type;
3751         int err;
3752
3753         uname = u64_to_user_ptr(info->perf_event.kprobe.func_name);
3754         ulen = info->perf_event.kprobe.name_len;
3755         err = bpf_perf_link_fill_common(event, uname, &ulen, &offset, &addr,
3756                                         &type, &missed);
3757         if (err)
3758                 return err;
3759         if (type == BPF_FD_TYPE_KRETPROBE)
3760                 info->perf_event.type = BPF_PERF_EVENT_KRETPROBE;
3761         else
3762                 info->perf_event.type = BPF_PERF_EVENT_KPROBE;
3763         info->perf_event.kprobe.name_len = ulen;
3764         info->perf_event.kprobe.offset = offset;
3765         info->perf_event.kprobe.missed = missed;
3766         if (!kallsyms_show_value(current_cred()))
3767                 addr = 0;
3768         info->perf_event.kprobe.addr = addr;
3769         info->perf_event.kprobe.cookie = event->bpf_cookie;
3770         return 0;
3771 }
3772 #endif
3773
3774 #ifdef CONFIG_UPROBE_EVENTS
3775 static int bpf_perf_link_fill_uprobe(const struct perf_event *event,
3776                                      struct bpf_link_info *info)
3777 {
3778         char __user *uname;
3779         u64 addr, offset;
3780         u32 ulen, type;
3781         int err;
3782
3783         uname = u64_to_user_ptr(info->perf_event.uprobe.file_name);
3784         ulen = info->perf_event.uprobe.name_len;
3785         err = bpf_perf_link_fill_common(event, uname, &ulen, &offset, &addr,
3786                                         &type, NULL);
3787         if (err)
3788                 return err;
3789
3790         if (type == BPF_FD_TYPE_URETPROBE)
3791                 info->perf_event.type = BPF_PERF_EVENT_URETPROBE;
3792         else
3793                 info->perf_event.type = BPF_PERF_EVENT_UPROBE;
3794         info->perf_event.uprobe.name_len = ulen;
3795         info->perf_event.uprobe.offset = offset;
3796         info->perf_event.uprobe.cookie = event->bpf_cookie;
3797         return 0;
3798 }
3799 #endif
3800
3801 static int bpf_perf_link_fill_probe(const struct perf_event *event,
3802                                     struct bpf_link_info *info)
3803 {
3804 #ifdef CONFIG_KPROBE_EVENTS
3805         if (event->tp_event->flags & TRACE_EVENT_FL_KPROBE)
3806                 return bpf_perf_link_fill_kprobe(event, info);
3807 #endif
3808 #ifdef CONFIG_UPROBE_EVENTS
3809         if (event->tp_event->flags & TRACE_EVENT_FL_UPROBE)
3810                 return bpf_perf_link_fill_uprobe(event, info);
3811 #endif
3812         return -EOPNOTSUPP;
3813 }
3814
3815 static int bpf_perf_link_fill_tracepoint(const struct perf_event *event,
3816                                          struct bpf_link_info *info)
3817 {
3818         char __user *uname;
3819         u32 ulen;
3820         int err;
3821
3822         uname = u64_to_user_ptr(info->perf_event.tracepoint.tp_name);
3823         ulen = info->perf_event.tracepoint.name_len;
3824         err = bpf_perf_link_fill_common(event, uname, &ulen, NULL, NULL, NULL, NULL);
3825         if (err)
3826                 return err;
3827
3828         info->perf_event.type = BPF_PERF_EVENT_TRACEPOINT;
3829         info->perf_event.tracepoint.name_len = ulen;
3830         info->perf_event.tracepoint.cookie = event->bpf_cookie;
3831         return 0;
3832 }
3833
3834 static int bpf_perf_link_fill_perf_event(const struct perf_event *event,
3835                                          struct bpf_link_info *info)
3836 {
3837         info->perf_event.event.type = event->attr.type;
3838         info->perf_event.event.config = event->attr.config;
3839         info->perf_event.event.cookie = event->bpf_cookie;
3840         info->perf_event.type = BPF_PERF_EVENT_EVENT;
3841         return 0;
3842 }
3843
3844 static int bpf_perf_link_fill_link_info(const struct bpf_link *link,
3845                                         struct bpf_link_info *info)
3846 {
3847         struct bpf_perf_link *perf_link;
3848         const struct perf_event *event;
3849
3850         perf_link = container_of(link, struct bpf_perf_link, link);
3851         event = perf_get_event(perf_link->perf_file);
3852         if (IS_ERR(event))
3853                 return PTR_ERR(event);
3854
3855         switch (event->prog->type) {
3856         case BPF_PROG_TYPE_PERF_EVENT:
3857                 return bpf_perf_link_fill_perf_event(event, info);
3858         case BPF_PROG_TYPE_TRACEPOINT:
3859                 return bpf_perf_link_fill_tracepoint(event, info);
3860         case BPF_PROG_TYPE_KPROBE:
3861                 return bpf_perf_link_fill_probe(event, info);
3862         default:
3863                 return -EOPNOTSUPP;
3864         }
3865 }
3866
3867 static const struct bpf_link_ops bpf_perf_link_lops = {
3868         .release = bpf_perf_link_release,
3869         .dealloc = bpf_perf_link_dealloc,
3870         .fill_link_info = bpf_perf_link_fill_link_info,
3871 };
3872
3873 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3874 {
3875         struct bpf_link_primer link_primer;
3876         struct bpf_perf_link *link;
3877         struct perf_event *event;
3878         struct file *perf_file;
3879         int err;
3880
3881         if (attr->link_create.flags)
3882                 return -EINVAL;
3883
3884         perf_file = perf_event_get(attr->link_create.target_fd);
3885         if (IS_ERR(perf_file))
3886                 return PTR_ERR(perf_file);
3887
3888         link = kzalloc(sizeof(*link), GFP_USER);
3889         if (!link) {
3890                 err = -ENOMEM;
3891                 goto out_put_file;
3892         }
3893         bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3894         link->perf_file = perf_file;
3895
3896         err = bpf_link_prime(&link->link, &link_primer);
3897         if (err) {
3898                 kfree(link);
3899                 goto out_put_file;
3900         }
3901
3902         event = perf_file->private_data;
3903         err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3904         if (err) {
3905                 bpf_link_cleanup(&link_primer);
3906                 goto out_put_file;
3907         }
3908         /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3909         bpf_prog_inc(prog);
3910
3911         return bpf_link_settle(&link_primer);
3912
3913 out_put_file:
3914         fput(perf_file);
3915         return err;
3916 }
3917 #else
3918 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3919 {
3920         return -EOPNOTSUPP;
3921 }
3922 #endif /* CONFIG_PERF_EVENTS */
3923
3924 static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3925                                   const char __user *user_tp_name, u64 cookie)
3926 {
3927         struct bpf_link_primer link_primer;
3928         struct bpf_raw_tp_link *link;
3929         struct bpf_raw_event_map *btp;
3930         const char *tp_name;
3931         char buf[128];
3932         int err;
3933
3934         switch (prog->type) {
3935         case BPF_PROG_TYPE_TRACING:
3936         case BPF_PROG_TYPE_EXT:
3937         case BPF_PROG_TYPE_LSM:
3938                 if (user_tp_name)
3939                         /* The attach point for this category of programs
3940                          * should be specified via btf_id during program load.
3941                          */
3942                         return -EINVAL;
3943                 if (prog->type == BPF_PROG_TYPE_TRACING &&
3944                     prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3945                         tp_name = prog->aux->attach_func_name;
3946                         break;
3947                 }
3948                 return bpf_tracing_prog_attach(prog, 0, 0, 0);
3949         case BPF_PROG_TYPE_RAW_TRACEPOINT:
3950         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3951                 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3952                         return -EFAULT;
3953                 buf[sizeof(buf) - 1] = 0;
3954                 tp_name = buf;
3955                 break;
3956         default:
3957                 return -EINVAL;
3958         }
3959
3960         btp = bpf_get_raw_tracepoint(tp_name);
3961         if (!btp)
3962                 return -ENOENT;
3963
3964         link = kzalloc(sizeof(*link), GFP_USER);
3965         if (!link) {
3966                 err = -ENOMEM;
3967                 goto out_put_btp;
3968         }
3969         bpf_link_init_sleepable(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3970                                 &bpf_raw_tp_link_lops, prog,
3971                                 tracepoint_is_faultable(btp->tp));
3972         link->btp = btp;
3973         link->cookie = cookie;
3974
3975         err = bpf_link_prime(&link->link, &link_primer);
3976         if (err) {
3977                 kfree(link);
3978                 goto out_put_btp;
3979         }
3980
3981         err = bpf_probe_register(link->btp, link);
3982         if (err) {
3983                 bpf_link_cleanup(&link_primer);
3984                 goto out_put_btp;
3985         }
3986
3987         return bpf_link_settle(&link_primer);
3988
3989 out_put_btp:
3990         bpf_put_raw_tracepoint(btp);
3991         return err;
3992 }
3993
3994 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.cookie
3995
3996 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3997 {
3998         struct bpf_prog *prog;
3999         void __user *tp_name;
4000         __u64 cookie;
4001         int fd;
4002
4003         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
4004                 return -EINVAL;
4005
4006         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
4007         if (IS_ERR(prog))
4008                 return PTR_ERR(prog);
4009
4010         tp_name = u64_to_user_ptr(attr->raw_tracepoint.name);
4011         cookie = attr->raw_tracepoint.cookie;
4012         fd = bpf_raw_tp_link_attach(prog, tp_name, cookie);
4013         if (fd < 0)
4014                 bpf_prog_put(prog);
4015         return fd;
4016 }
4017
4018 static enum bpf_prog_type
4019 attach_type_to_prog_type(enum bpf_attach_type attach_type)
4020 {
4021         switch (attach_type) {
4022         case BPF_CGROUP_INET_INGRESS:
4023         case BPF_CGROUP_INET_EGRESS:
4024                 return BPF_PROG_TYPE_CGROUP_SKB;
4025         case BPF_CGROUP_INET_SOCK_CREATE:
4026         case BPF_CGROUP_INET_SOCK_RELEASE:
4027         case BPF_CGROUP_INET4_POST_BIND:
4028         case BPF_CGROUP_INET6_POST_BIND:
4029                 return BPF_PROG_TYPE_CGROUP_SOCK;
4030         case BPF_CGROUP_INET4_BIND:
4031         case BPF_CGROUP_INET6_BIND:
4032         case BPF_CGROUP_INET4_CONNECT:
4033         case BPF_CGROUP_INET6_CONNECT:
4034         case BPF_CGROUP_UNIX_CONNECT:
4035         case BPF_CGROUP_INET4_GETPEERNAME:
4036         case BPF_CGROUP_INET6_GETPEERNAME:
4037         case BPF_CGROUP_UNIX_GETPEERNAME:
4038         case BPF_CGROUP_INET4_GETSOCKNAME:
4039         case BPF_CGROUP_INET6_GETSOCKNAME:
4040         case BPF_CGROUP_UNIX_GETSOCKNAME:
4041         case BPF_CGROUP_UDP4_SENDMSG:
4042         case BPF_CGROUP_UDP6_SENDMSG:
4043         case BPF_CGROUP_UNIX_SENDMSG:
4044         case BPF_CGROUP_UDP4_RECVMSG:
4045         case BPF_CGROUP_UDP6_RECVMSG:
4046         case BPF_CGROUP_UNIX_RECVMSG:
4047                 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
4048         case BPF_CGROUP_SOCK_OPS:
4049                 return BPF_PROG_TYPE_SOCK_OPS;
4050         case BPF_CGROUP_DEVICE:
4051                 return BPF_PROG_TYPE_CGROUP_DEVICE;
4052         case BPF_SK_MSG_VERDICT:
4053                 return BPF_PROG_TYPE_SK_MSG;
4054         case BPF_SK_SKB_STREAM_PARSER:
4055         case BPF_SK_SKB_STREAM_VERDICT:
4056         case BPF_SK_SKB_VERDICT:
4057                 return BPF_PROG_TYPE_SK_SKB;
4058         case BPF_LIRC_MODE2:
4059                 return BPF_PROG_TYPE_LIRC_MODE2;
4060         case BPF_FLOW_DISSECTOR:
4061                 return BPF_PROG_TYPE_FLOW_DISSECTOR;
4062         case BPF_CGROUP_SYSCTL:
4063                 return BPF_PROG_TYPE_CGROUP_SYSCTL;
4064         case BPF_CGROUP_GETSOCKOPT:
4065         case BPF_CGROUP_SETSOCKOPT:
4066                 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
4067         case BPF_TRACE_ITER:
4068         case BPF_TRACE_RAW_TP:
4069         case BPF_TRACE_FENTRY:
4070         case BPF_TRACE_FEXIT:
4071         case BPF_MODIFY_RETURN:
4072                 return BPF_PROG_TYPE_TRACING;
4073         case BPF_LSM_MAC:
4074                 return BPF_PROG_TYPE_LSM;
4075         case BPF_SK_LOOKUP:
4076                 return BPF_PROG_TYPE_SK_LOOKUP;
4077         case BPF_XDP:
4078                 return BPF_PROG_TYPE_XDP;
4079         case BPF_LSM_CGROUP:
4080                 return BPF_PROG_TYPE_LSM;
4081         case BPF_TCX_INGRESS:
4082         case BPF_TCX_EGRESS:
4083         case BPF_NETKIT_PRIMARY:
4084         case BPF_NETKIT_PEER:
4085                 return BPF_PROG_TYPE_SCHED_CLS;
4086         default:
4087                 return BPF_PROG_TYPE_UNSPEC;
4088         }
4089 }
4090
4091 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
4092                                              enum bpf_attach_type attach_type)
4093 {
4094         enum bpf_prog_type ptype;
4095
4096         switch (prog->type) {
4097         case BPF_PROG_TYPE_CGROUP_SOCK:
4098         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4099         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4100         case BPF_PROG_TYPE_SK_LOOKUP:
4101                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
4102         case BPF_PROG_TYPE_CGROUP_SKB:
4103                 if (!bpf_token_capable(prog->aux->token, CAP_NET_ADMIN))
4104                         /* cg-skb progs can be loaded by unpriv user.
4105                          * check permissions at attach time.
4106                          */
4107                         return -EPERM;
4108
4109                 ptype = attach_type_to_prog_type(attach_type);
4110                 if (prog->type != ptype)
4111                         return -EINVAL;
4112
4113                 return prog->enforce_expected_attach_type &&
4114                         prog->expected_attach_type != attach_type ?
4115                         -EINVAL : 0;
4116         case BPF_PROG_TYPE_EXT:
4117                 return 0;
4118         case BPF_PROG_TYPE_NETFILTER:
4119                 if (attach_type != BPF_NETFILTER)
4120                         return -EINVAL;
4121                 return 0;
4122         case BPF_PROG_TYPE_PERF_EVENT:
4123         case BPF_PROG_TYPE_TRACEPOINT:
4124                 if (attach_type != BPF_PERF_EVENT)
4125                         return -EINVAL;
4126                 return 0;
4127         case BPF_PROG_TYPE_KPROBE:
4128                 if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI &&
4129                     attach_type != BPF_TRACE_KPROBE_MULTI)
4130                         return -EINVAL;
4131                 if (prog->expected_attach_type == BPF_TRACE_KPROBE_SESSION &&
4132                     attach_type != BPF_TRACE_KPROBE_SESSION)
4133                         return -EINVAL;
4134                 if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI &&
4135                     attach_type != BPF_TRACE_UPROBE_MULTI)
4136                         return -EINVAL;
4137                 if (prog->expected_attach_type == BPF_TRACE_UPROBE_SESSION &&
4138                     attach_type != BPF_TRACE_UPROBE_SESSION)
4139                         return -EINVAL;
4140                 if (attach_type != BPF_PERF_EVENT &&
4141                     attach_type != BPF_TRACE_KPROBE_MULTI &&
4142                     attach_type != BPF_TRACE_KPROBE_SESSION &&
4143                     attach_type != BPF_TRACE_UPROBE_MULTI &&
4144                     attach_type != BPF_TRACE_UPROBE_SESSION)
4145                         return -EINVAL;
4146                 return 0;
4147         case BPF_PROG_TYPE_SCHED_CLS:
4148                 if (attach_type != BPF_TCX_INGRESS &&
4149                     attach_type != BPF_TCX_EGRESS &&
4150                     attach_type != BPF_NETKIT_PRIMARY &&
4151                     attach_type != BPF_NETKIT_PEER)
4152                         return -EINVAL;
4153                 return 0;
4154         default:
4155                 ptype = attach_type_to_prog_type(attach_type);
4156                 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type)
4157                         return -EINVAL;
4158                 return 0;
4159         }
4160 }
4161
4162 #define BPF_PROG_ATTACH_LAST_FIELD expected_revision
4163
4164 #define BPF_F_ATTACH_MASK_BASE  \
4165         (BPF_F_ALLOW_OVERRIDE | \
4166          BPF_F_ALLOW_MULTI |    \
4167          BPF_F_REPLACE)
4168
4169 #define BPF_F_ATTACH_MASK_MPROG \
4170         (BPF_F_REPLACE |        \
4171          BPF_F_BEFORE |         \
4172          BPF_F_AFTER |          \
4173          BPF_F_ID |             \
4174          BPF_F_LINK)
4175
4176 static int bpf_prog_attach(const union bpf_attr *attr)
4177 {
4178         enum bpf_prog_type ptype;
4179         struct bpf_prog *prog;
4180         int ret;
4181
4182         if (CHECK_ATTR(BPF_PROG_ATTACH))
4183                 return -EINVAL;
4184
4185         ptype = attach_type_to_prog_type(attr->attach_type);
4186         if (ptype == BPF_PROG_TYPE_UNSPEC)
4187                 return -EINVAL;
4188         if (bpf_mprog_supported(ptype)) {
4189                 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4190                         return -EINVAL;
4191         } else {
4192                 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_BASE)
4193                         return -EINVAL;
4194                 if (attr->relative_fd ||
4195                     attr->expected_revision)
4196                         return -EINVAL;
4197         }
4198
4199         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4200         if (IS_ERR(prog))
4201                 return PTR_ERR(prog);
4202
4203         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
4204                 bpf_prog_put(prog);
4205                 return -EINVAL;
4206         }
4207
4208         switch (ptype) {
4209         case BPF_PROG_TYPE_SK_SKB:
4210         case BPF_PROG_TYPE_SK_MSG:
4211                 ret = sock_map_get_from_fd(attr, prog);
4212                 break;
4213         case BPF_PROG_TYPE_LIRC_MODE2:
4214                 ret = lirc_prog_attach(attr, prog);
4215                 break;
4216         case BPF_PROG_TYPE_FLOW_DISSECTOR:
4217                 ret = netns_bpf_prog_attach(attr, prog);
4218                 break;
4219         case BPF_PROG_TYPE_CGROUP_DEVICE:
4220         case BPF_PROG_TYPE_CGROUP_SKB:
4221         case BPF_PROG_TYPE_CGROUP_SOCK:
4222         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4223         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4224         case BPF_PROG_TYPE_CGROUP_SYSCTL:
4225         case BPF_PROG_TYPE_SOCK_OPS:
4226         case BPF_PROG_TYPE_LSM:
4227                 if (ptype == BPF_PROG_TYPE_LSM &&
4228                     prog->expected_attach_type != BPF_LSM_CGROUP)
4229                         ret = -EINVAL;
4230                 else
4231                         ret = cgroup_bpf_prog_attach(attr, ptype, prog);
4232                 break;
4233         case BPF_PROG_TYPE_SCHED_CLS:
4234                 if (attr->attach_type == BPF_TCX_INGRESS ||
4235                     attr->attach_type == BPF_TCX_EGRESS)
4236                         ret = tcx_prog_attach(attr, prog);
4237                 else
4238                         ret = netkit_prog_attach(attr, prog);
4239                 break;
4240         default:
4241                 ret = -EINVAL;
4242         }
4243
4244         if (ret)
4245                 bpf_prog_put(prog);
4246         return ret;
4247 }
4248
4249 #define BPF_PROG_DETACH_LAST_FIELD expected_revision
4250
4251 static int bpf_prog_detach(const union bpf_attr *attr)
4252 {
4253         struct bpf_prog *prog = NULL;
4254         enum bpf_prog_type ptype;
4255         int ret;
4256
4257         if (CHECK_ATTR(BPF_PROG_DETACH))
4258                 return -EINVAL;
4259
4260         ptype = attach_type_to_prog_type(attr->attach_type);
4261         if (bpf_mprog_supported(ptype)) {
4262                 if (ptype == BPF_PROG_TYPE_UNSPEC)
4263                         return -EINVAL;
4264                 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4265                         return -EINVAL;
4266                 if (attr->attach_bpf_fd) {
4267                         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4268                         if (IS_ERR(prog))
4269                                 return PTR_ERR(prog);
4270                 }
4271         } else if (attr->attach_flags ||
4272                    attr->relative_fd ||
4273                    attr->expected_revision) {
4274                 return -EINVAL;
4275         }
4276
4277         switch (ptype) {
4278         case BPF_PROG_TYPE_SK_MSG:
4279         case BPF_PROG_TYPE_SK_SKB:
4280                 ret = sock_map_prog_detach(attr, ptype);
4281                 break;
4282         case BPF_PROG_TYPE_LIRC_MODE2:
4283                 ret = lirc_prog_detach(attr);
4284                 break;
4285         case BPF_PROG_TYPE_FLOW_DISSECTOR:
4286                 ret = netns_bpf_prog_detach(attr, ptype);
4287                 break;
4288         case BPF_PROG_TYPE_CGROUP_DEVICE:
4289         case BPF_PROG_TYPE_CGROUP_SKB:
4290         case BPF_PROG_TYPE_CGROUP_SOCK:
4291         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4292         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4293         case BPF_PROG_TYPE_CGROUP_SYSCTL:
4294         case BPF_PROG_TYPE_SOCK_OPS:
4295         case BPF_PROG_TYPE_LSM:
4296                 ret = cgroup_bpf_prog_detach(attr, ptype);
4297                 break;
4298         case BPF_PROG_TYPE_SCHED_CLS:
4299                 if (attr->attach_type == BPF_TCX_INGRESS ||
4300                     attr->attach_type == BPF_TCX_EGRESS)
4301                         ret = tcx_prog_detach(attr, prog);
4302                 else
4303                         ret = netkit_prog_detach(attr, prog);
4304                 break;
4305         default:
4306                 ret = -EINVAL;
4307         }
4308
4309         if (prog)
4310                 bpf_prog_put(prog);
4311         return ret;
4312 }
4313
4314 #define BPF_PROG_QUERY_LAST_FIELD query.revision
4315
4316 static int bpf_prog_query(const union bpf_attr *attr,
4317                           union bpf_attr __user *uattr)
4318 {
4319         if (!bpf_net_capable())
4320                 return -EPERM;
4321         if (CHECK_ATTR(BPF_PROG_QUERY))
4322                 return -EINVAL;
4323         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
4324                 return -EINVAL;
4325
4326         switch (attr->query.attach_type) {
4327         case BPF_CGROUP_INET_INGRESS:
4328         case BPF_CGROUP_INET_EGRESS:
4329         case BPF_CGROUP_INET_SOCK_CREATE:
4330         case BPF_CGROUP_INET_SOCK_RELEASE:
4331         case BPF_CGROUP_INET4_BIND:
4332         case BPF_CGROUP_INET6_BIND:
4333         case BPF_CGROUP_INET4_POST_BIND:
4334         case BPF_CGROUP_INET6_POST_BIND:
4335         case BPF_CGROUP_INET4_CONNECT:
4336         case BPF_CGROUP_INET6_CONNECT:
4337         case BPF_CGROUP_UNIX_CONNECT:
4338         case BPF_CGROUP_INET4_GETPEERNAME:
4339         case BPF_CGROUP_INET6_GETPEERNAME:
4340         case BPF_CGROUP_UNIX_GETPEERNAME:
4341         case BPF_CGROUP_INET4_GETSOCKNAME:
4342         case BPF_CGROUP_INET6_GETSOCKNAME:
4343         case BPF_CGROUP_UNIX_GETSOCKNAME:
4344         case BPF_CGROUP_UDP4_SENDMSG:
4345         case BPF_CGROUP_UDP6_SENDMSG:
4346         case BPF_CGROUP_UNIX_SENDMSG:
4347         case BPF_CGROUP_UDP4_RECVMSG:
4348         case BPF_CGROUP_UDP6_RECVMSG:
4349         case BPF_CGROUP_UNIX_RECVMSG:
4350         case BPF_CGROUP_SOCK_OPS:
4351         case BPF_CGROUP_DEVICE:
4352         case BPF_CGROUP_SYSCTL:
4353         case BPF_CGROUP_GETSOCKOPT:
4354         case BPF_CGROUP_SETSOCKOPT:
4355         case BPF_LSM_CGROUP:
4356                 return cgroup_bpf_prog_query(attr, uattr);
4357         case BPF_LIRC_MODE2:
4358                 return lirc_prog_query(attr, uattr);
4359         case BPF_FLOW_DISSECTOR:
4360         case BPF_SK_LOOKUP:
4361                 return netns_bpf_prog_query(attr, uattr);
4362         case BPF_SK_SKB_STREAM_PARSER:
4363         case BPF_SK_SKB_STREAM_VERDICT:
4364         case BPF_SK_MSG_VERDICT:
4365         case BPF_SK_SKB_VERDICT:
4366                 return sock_map_bpf_prog_query(attr, uattr);
4367         case BPF_TCX_INGRESS:
4368         case BPF_TCX_EGRESS:
4369                 return tcx_prog_query(attr, uattr);
4370         case BPF_NETKIT_PRIMARY:
4371         case BPF_NETKIT_PEER:
4372                 return netkit_prog_query(attr, uattr);
4373         default:
4374                 return -EINVAL;
4375         }
4376 }
4377
4378 #define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
4379
4380 static int bpf_prog_test_run(const union bpf_attr *attr,
4381                              union bpf_attr __user *uattr)
4382 {
4383         struct bpf_prog *prog;
4384         int ret = -ENOTSUPP;
4385
4386         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
4387                 return -EINVAL;
4388
4389         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
4390             (!attr->test.ctx_size_in && attr->test.ctx_in))
4391                 return -EINVAL;
4392
4393         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
4394             (!attr->test.ctx_size_out && attr->test.ctx_out))
4395                 return -EINVAL;
4396
4397         prog = bpf_prog_get(attr->test.prog_fd);
4398         if (IS_ERR(prog))
4399                 return PTR_ERR(prog);
4400
4401         if (prog->aux->ops->test_run)
4402                 ret = prog->aux->ops->test_run(prog, attr, uattr);
4403
4404         bpf_prog_put(prog);
4405         return ret;
4406 }
4407
4408 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
4409
4410 static int bpf_obj_get_next_id(const union bpf_attr *attr,
4411                                union bpf_attr __user *uattr,
4412                                struct idr *idr,
4413                                spinlock_t *lock)
4414 {
4415         u32 next_id = attr->start_id;
4416         int err = 0;
4417
4418         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
4419                 return -EINVAL;
4420
4421         if (!capable(CAP_SYS_ADMIN))
4422                 return -EPERM;
4423
4424         next_id++;
4425         spin_lock_bh(lock);
4426         if (!idr_get_next(idr, &next_id))
4427                 err = -ENOENT;
4428         spin_unlock_bh(lock);
4429
4430         if (!err)
4431                 err = put_user(next_id, &uattr->next_id);
4432
4433         return err;
4434 }
4435
4436 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
4437 {
4438         struct bpf_map *map;
4439
4440         spin_lock_bh(&map_idr_lock);
4441 again:
4442         map = idr_get_next(&map_idr, id);
4443         if (map) {
4444                 map = __bpf_map_inc_not_zero(map, false);
4445                 if (IS_ERR(map)) {
4446                         (*id)++;
4447                         goto again;
4448                 }
4449         }
4450         spin_unlock_bh(&map_idr_lock);
4451
4452         return map;
4453 }
4454
4455 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
4456 {
4457         struct bpf_prog *prog;
4458
4459         spin_lock_bh(&prog_idr_lock);
4460 again:
4461         prog = idr_get_next(&prog_idr, id);
4462         if (prog) {
4463                 prog = bpf_prog_inc_not_zero(prog);
4464                 if (IS_ERR(prog)) {
4465                         (*id)++;
4466                         goto again;
4467                 }
4468         }
4469         spin_unlock_bh(&prog_idr_lock);
4470
4471         return prog;
4472 }
4473
4474 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
4475
4476 struct bpf_prog *bpf_prog_by_id(u32 id)
4477 {
4478         struct bpf_prog *prog;
4479
4480         if (!id)
4481                 return ERR_PTR(-ENOENT);
4482
4483         spin_lock_bh(&prog_idr_lock);
4484         prog = idr_find(&prog_idr, id);
4485         if (prog)
4486                 prog = bpf_prog_inc_not_zero(prog);
4487         else
4488                 prog = ERR_PTR(-ENOENT);
4489         spin_unlock_bh(&prog_idr_lock);
4490         return prog;
4491 }
4492
4493 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
4494 {
4495         struct bpf_prog *prog;
4496         u32 id = attr->prog_id;
4497         int fd;
4498
4499         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
4500                 return -EINVAL;
4501
4502         if (!capable(CAP_SYS_ADMIN))
4503                 return -EPERM;
4504
4505         prog = bpf_prog_by_id(id);
4506         if (IS_ERR(prog))
4507                 return PTR_ERR(prog);
4508
4509         fd = bpf_prog_new_fd(prog);
4510         if (fd < 0)
4511                 bpf_prog_put(prog);
4512
4513         return fd;
4514 }
4515
4516 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
4517
4518 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
4519 {
4520         struct bpf_map *map;
4521         u32 id = attr->map_id;
4522         int f_flags;
4523         int fd;
4524
4525         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
4526             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
4527                 return -EINVAL;
4528
4529         if (!capable(CAP_SYS_ADMIN))
4530                 return -EPERM;
4531
4532         f_flags = bpf_get_file_flag(attr->open_flags);
4533         if (f_flags < 0)
4534                 return f_flags;
4535
4536         spin_lock_bh(&map_idr_lock);
4537         map = idr_find(&map_idr, id);
4538         if (map)
4539                 map = __bpf_map_inc_not_zero(map, true);
4540         else
4541                 map = ERR_PTR(-ENOENT);
4542         spin_unlock_bh(&map_idr_lock);
4543
4544         if (IS_ERR(map))
4545                 return PTR_ERR(map);
4546
4547         fd = bpf_map_new_fd(map, f_flags);
4548         if (fd < 0)
4549                 bpf_map_put_with_uref(map);
4550
4551         return fd;
4552 }
4553
4554 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
4555                                               unsigned long addr, u32 *off,
4556                                               u32 *type)
4557 {
4558         const struct bpf_map *map;
4559         int i;
4560
4561         mutex_lock(&prog->aux->used_maps_mutex);
4562         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
4563                 map = prog->aux->used_maps[i];
4564                 if (map == (void *)addr) {
4565                         *type = BPF_PSEUDO_MAP_FD;
4566                         goto out;
4567                 }
4568                 if (!map->ops->map_direct_value_meta)
4569                         continue;
4570                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
4571                         *type = BPF_PSEUDO_MAP_VALUE;
4572                         goto out;
4573                 }
4574         }
4575         map = NULL;
4576
4577 out:
4578         mutex_unlock(&prog->aux->used_maps_mutex);
4579         return map;
4580 }
4581
4582 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
4583                                               const struct cred *f_cred)
4584 {
4585         const struct bpf_map *map;
4586         struct bpf_insn *insns;
4587         u32 off, type;
4588         u64 imm;
4589         u8 code;
4590         int i;
4591
4592         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
4593                         GFP_USER);
4594         if (!insns)
4595                 return insns;
4596
4597         for (i = 0; i < prog->len; i++) {
4598                 code = insns[i].code;
4599
4600                 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
4601                         insns[i].code = BPF_JMP | BPF_CALL;
4602                         insns[i].imm = BPF_FUNC_tail_call;
4603                         /* fall-through */
4604                 }
4605                 if (code == (BPF_JMP | BPF_CALL) ||
4606                     code == (BPF_JMP | BPF_CALL_ARGS)) {
4607                         if (code == (BPF_JMP | BPF_CALL_ARGS))
4608                                 insns[i].code = BPF_JMP | BPF_CALL;
4609                         if (!bpf_dump_raw_ok(f_cred))
4610                                 insns[i].imm = 0;
4611                         continue;
4612                 }
4613                 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
4614                         insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
4615                         continue;
4616                 }
4617
4618                 if ((BPF_CLASS(code) == BPF_LDX || BPF_CLASS(code) == BPF_STX ||
4619                      BPF_CLASS(code) == BPF_ST) && BPF_MODE(code) == BPF_PROBE_MEM32) {
4620                         insns[i].code = BPF_CLASS(code) | BPF_SIZE(code) | BPF_MEM;
4621                         continue;
4622                 }
4623
4624                 if (code != (BPF_LD | BPF_IMM | BPF_DW))
4625                         continue;
4626
4627                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
4628                 map = bpf_map_from_imm(prog, imm, &off, &type);
4629                 if (map) {
4630                         insns[i].src_reg = type;
4631                         insns[i].imm = map->id;
4632                         insns[i + 1].imm = off;
4633                         continue;
4634                 }
4635         }
4636
4637         return insns;
4638 }
4639
4640 static int set_info_rec_size(struct bpf_prog_info *info)
4641 {
4642         /*
4643          * Ensure info.*_rec_size is the same as kernel expected size
4644          *
4645          * or
4646          *
4647          * Only allow zero *_rec_size if both _rec_size and _cnt are
4648          * zero.  In this case, the kernel will set the expected
4649          * _rec_size back to the info.
4650          */
4651
4652         if ((info->nr_func_info || info->func_info_rec_size) &&
4653             info->func_info_rec_size != sizeof(struct bpf_func_info))
4654                 return -EINVAL;
4655
4656         if ((info->nr_line_info || info->line_info_rec_size) &&
4657             info->line_info_rec_size != sizeof(struct bpf_line_info))
4658                 return -EINVAL;
4659
4660         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
4661             info->jited_line_info_rec_size != sizeof(__u64))
4662                 return -EINVAL;
4663
4664         info->func_info_rec_size = sizeof(struct bpf_func_info);
4665         info->line_info_rec_size = sizeof(struct bpf_line_info);
4666         info->jited_line_info_rec_size = sizeof(__u64);
4667
4668         return 0;
4669 }
4670
4671 static int bpf_prog_get_info_by_fd(struct file *file,
4672                                    struct bpf_prog *prog,
4673                                    const union bpf_attr *attr,
4674                                    union bpf_attr __user *uattr)
4675 {
4676         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4677         struct btf *attach_btf = bpf_prog_get_target_btf(prog);
4678         struct bpf_prog_info info;
4679         u32 info_len = attr->info.info_len;
4680         struct bpf_prog_kstats stats;
4681         char __user *uinsns;
4682         u32 ulen;
4683         int err;
4684
4685         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4686         if (err)
4687                 return err;
4688         info_len = min_t(u32, sizeof(info), info_len);
4689
4690         memset(&info, 0, sizeof(info));
4691         if (copy_from_user(&info, uinfo, info_len))
4692                 return -EFAULT;
4693
4694         info.type = prog->type;
4695         info.id = prog->aux->id;
4696         info.load_time = prog->aux->load_time;
4697         info.created_by_uid = from_kuid_munged(current_user_ns(),
4698                                                prog->aux->user->uid);
4699         info.gpl_compatible = prog->gpl_compatible;
4700
4701         memcpy(info.tag, prog->tag, sizeof(prog->tag));
4702         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
4703
4704         mutex_lock(&prog->aux->used_maps_mutex);
4705         ulen = info.nr_map_ids;
4706         info.nr_map_ids = prog->aux->used_map_cnt;
4707         ulen = min_t(u32, info.nr_map_ids, ulen);
4708         if (ulen) {
4709                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
4710                 u32 i;
4711
4712                 for (i = 0; i < ulen; i++)
4713                         if (put_user(prog->aux->used_maps[i]->id,
4714                                      &user_map_ids[i])) {
4715                                 mutex_unlock(&prog->aux->used_maps_mutex);
4716                                 return -EFAULT;
4717                         }
4718         }
4719         mutex_unlock(&prog->aux->used_maps_mutex);
4720
4721         err = set_info_rec_size(&info);
4722         if (err)
4723                 return err;
4724
4725         bpf_prog_get_stats(prog, &stats);
4726         info.run_time_ns = stats.nsecs;
4727         info.run_cnt = stats.cnt;
4728         info.recursion_misses = stats.misses;
4729
4730         info.verified_insns = prog->aux->verified_insns;
4731
4732         if (!bpf_capable()) {
4733                 info.jited_prog_len = 0;
4734                 info.xlated_prog_len = 0;
4735                 info.nr_jited_ksyms = 0;
4736                 info.nr_jited_func_lens = 0;
4737                 info.nr_func_info = 0;
4738                 info.nr_line_info = 0;
4739                 info.nr_jited_line_info = 0;
4740                 goto done;
4741         }
4742
4743         ulen = info.xlated_prog_len;
4744         info.xlated_prog_len = bpf_prog_insn_size(prog);
4745         if (info.xlated_prog_len && ulen) {
4746                 struct bpf_insn *insns_sanitized;
4747                 bool fault;
4748
4749                 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
4750                         info.xlated_prog_insns = 0;
4751                         goto done;
4752                 }
4753                 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4754                 if (!insns_sanitized)
4755                         return -ENOMEM;
4756                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4757                 ulen = min_t(u32, info.xlated_prog_len, ulen);
4758                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
4759                 kfree(insns_sanitized);
4760                 if (fault)
4761                         return -EFAULT;
4762         }
4763
4764         if (bpf_prog_is_offloaded(prog->aux)) {
4765                 err = bpf_prog_offload_info_fill(&info, prog);
4766                 if (err)
4767                         return err;
4768                 goto done;
4769         }
4770
4771         /* NOTE: the following code is supposed to be skipped for offload.
4772          * bpf_prog_offload_info_fill() is the place to fill similar fields
4773          * for offload.
4774          */
4775         ulen = info.jited_prog_len;
4776         if (prog->aux->func_cnt) {
4777                 u32 i;
4778
4779                 info.jited_prog_len = 0;
4780                 for (i = 0; i < prog->aux->func_cnt; i++)
4781                         info.jited_prog_len += prog->aux->func[i]->jited_len;
4782         } else {
4783                 info.jited_prog_len = prog->jited_len;
4784         }
4785
4786         if (info.jited_prog_len && ulen) {
4787                 if (bpf_dump_raw_ok(file->f_cred)) {
4788                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
4789                         ulen = min_t(u32, info.jited_prog_len, ulen);
4790
4791                         /* for multi-function programs, copy the JITed
4792                          * instructions for all the functions
4793                          */
4794                         if (prog->aux->func_cnt) {
4795                                 u32 len, free, i;
4796                                 u8 *img;
4797
4798                                 free = ulen;
4799                                 for (i = 0; i < prog->aux->func_cnt; i++) {
4800                                         len = prog->aux->func[i]->jited_len;
4801                                         len = min_t(u32, len, free);
4802                                         img = (u8 *) prog->aux->func[i]->bpf_func;
4803                                         if (copy_to_user(uinsns, img, len))
4804                                                 return -EFAULT;
4805                                         uinsns += len;
4806                                         free -= len;
4807                                         if (!free)
4808                                                 break;
4809                                 }
4810                         } else {
4811                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4812                                         return -EFAULT;
4813                         }
4814                 } else {
4815                         info.jited_prog_insns = 0;
4816                 }
4817         }
4818
4819         ulen = info.nr_jited_ksyms;
4820         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4821         if (ulen) {
4822                 if (bpf_dump_raw_ok(file->f_cred)) {
4823                         unsigned long ksym_addr;
4824                         u64 __user *user_ksyms;
4825                         u32 i;
4826
4827                         /* copy the address of the kernel symbol
4828                          * corresponding to each function
4829                          */
4830                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4831                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4832                         if (prog->aux->func_cnt) {
4833                                 for (i = 0; i < ulen; i++) {
4834                                         ksym_addr = (unsigned long)
4835                                                 prog->aux->func[i]->bpf_func;
4836                                         if (put_user((u64) ksym_addr,
4837                                                      &user_ksyms[i]))
4838                                                 return -EFAULT;
4839                                 }
4840                         } else {
4841                                 ksym_addr = (unsigned long) prog->bpf_func;
4842                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
4843                                         return -EFAULT;
4844                         }
4845                 } else {
4846                         info.jited_ksyms = 0;
4847                 }
4848         }
4849
4850         ulen = info.nr_jited_func_lens;
4851         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4852         if (ulen) {
4853                 if (bpf_dump_raw_ok(file->f_cred)) {
4854                         u32 __user *user_lens;
4855                         u32 func_len, i;
4856
4857                         /* copy the JITed image lengths for each function */
4858                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4859                         user_lens = u64_to_user_ptr(info.jited_func_lens);
4860                         if (prog->aux->func_cnt) {
4861                                 for (i = 0; i < ulen; i++) {
4862                                         func_len =
4863                                                 prog->aux->func[i]->jited_len;
4864                                         if (put_user(func_len, &user_lens[i]))
4865                                                 return -EFAULT;
4866                                 }
4867                         } else {
4868                                 func_len = prog->jited_len;
4869                                 if (put_user(func_len, &user_lens[0]))
4870                                         return -EFAULT;
4871                         }
4872                 } else {
4873                         info.jited_func_lens = 0;
4874                 }
4875         }
4876
4877         if (prog->aux->btf)
4878                 info.btf_id = btf_obj_id(prog->aux->btf);
4879         info.attach_btf_id = prog->aux->attach_btf_id;
4880         if (attach_btf)
4881                 info.attach_btf_obj_id = btf_obj_id(attach_btf);
4882
4883         ulen = info.nr_func_info;
4884         info.nr_func_info = prog->aux->func_info_cnt;
4885         if (info.nr_func_info && ulen) {
4886                 char __user *user_finfo;
4887
4888                 user_finfo = u64_to_user_ptr(info.func_info);
4889                 ulen = min_t(u32, info.nr_func_info, ulen);
4890                 if (copy_to_user(user_finfo, prog->aux->func_info,
4891                                  info.func_info_rec_size * ulen))
4892                         return -EFAULT;
4893         }
4894
4895         ulen = info.nr_line_info;
4896         info.nr_line_info = prog->aux->nr_linfo;
4897         if (info.nr_line_info && ulen) {
4898                 __u8 __user *user_linfo;
4899
4900                 user_linfo = u64_to_user_ptr(info.line_info);
4901                 ulen = min_t(u32, info.nr_line_info, ulen);
4902                 if (copy_to_user(user_linfo, prog->aux->linfo,
4903                                  info.line_info_rec_size * ulen))
4904                         return -EFAULT;
4905         }
4906
4907         ulen = info.nr_jited_line_info;
4908         if (prog->aux->jited_linfo)
4909                 info.nr_jited_line_info = prog->aux->nr_linfo;
4910         else
4911                 info.nr_jited_line_info = 0;
4912         if (info.nr_jited_line_info && ulen) {
4913                 if (bpf_dump_raw_ok(file->f_cred)) {
4914                         unsigned long line_addr;
4915                         __u64 __user *user_linfo;
4916                         u32 i;
4917
4918                         user_linfo = u64_to_user_ptr(info.jited_line_info);
4919                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
4920                         for (i = 0; i < ulen; i++) {
4921                                 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4922                                 if (put_user((__u64)line_addr, &user_linfo[i]))
4923                                         return -EFAULT;
4924                         }
4925                 } else {
4926                         info.jited_line_info = 0;
4927                 }
4928         }
4929
4930         ulen = info.nr_prog_tags;
4931         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4932         if (ulen) {
4933                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4934                 u32 i;
4935
4936                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4937                 ulen = min_t(u32, info.nr_prog_tags, ulen);
4938                 if (prog->aux->func_cnt) {
4939                         for (i = 0; i < ulen; i++) {
4940                                 if (copy_to_user(user_prog_tags[i],
4941                                                  prog->aux->func[i]->tag,
4942                                                  BPF_TAG_SIZE))
4943                                         return -EFAULT;
4944                         }
4945                 } else {
4946                         if (copy_to_user(user_prog_tags[0],
4947                                          prog->tag, BPF_TAG_SIZE))
4948                                 return -EFAULT;
4949                 }
4950         }
4951
4952 done:
4953         if (copy_to_user(uinfo, &info, info_len) ||
4954             put_user(info_len, &uattr->info.info_len))
4955                 return -EFAULT;
4956
4957         return 0;
4958 }
4959
4960 static int bpf_map_get_info_by_fd(struct file *file,
4961                                   struct bpf_map *map,
4962                                   const union bpf_attr *attr,
4963                                   union bpf_attr __user *uattr)
4964 {
4965         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4966         struct bpf_map_info info;
4967         u32 info_len = attr->info.info_len;
4968         int err;
4969
4970         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4971         if (err)
4972                 return err;
4973         info_len = min_t(u32, sizeof(info), info_len);
4974
4975         memset(&info, 0, sizeof(info));
4976         info.type = map->map_type;
4977         info.id = map->id;
4978         info.key_size = map->key_size;
4979         info.value_size = map->value_size;
4980         info.max_entries = map->max_entries;
4981         info.map_flags = map->map_flags;
4982         info.map_extra = map->map_extra;
4983         memcpy(info.name, map->name, sizeof(map->name));
4984
4985         if (map->btf) {
4986                 info.btf_id = btf_obj_id(map->btf);
4987                 info.btf_key_type_id = map->btf_key_type_id;
4988                 info.btf_value_type_id = map->btf_value_type_id;
4989         }
4990         info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4991         if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS)
4992                 bpf_map_struct_ops_info_fill(&info, map);
4993
4994         if (bpf_map_is_offloaded(map)) {
4995                 err = bpf_map_offload_info_fill(&info, map);
4996                 if (err)
4997                         return err;
4998         }
4999
5000         if (copy_to_user(uinfo, &info, info_len) ||
5001             put_user(info_len, &uattr->info.info_len))
5002                 return -EFAULT;
5003
5004         return 0;
5005 }
5006
5007 static int bpf_btf_get_info_by_fd(struct file *file,
5008                                   struct btf *btf,
5009                                   const union bpf_attr *attr,
5010                                   union bpf_attr __user *uattr)
5011 {
5012         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
5013         u32 info_len = attr->info.info_len;
5014         int err;
5015
5016         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
5017         if (err)
5018                 return err;
5019
5020         return btf_get_info_by_fd(btf, attr, uattr);
5021 }
5022
5023 static int bpf_link_get_info_by_fd(struct file *file,
5024                                   struct bpf_link *link,
5025                                   const union bpf_attr *attr,
5026                                   union bpf_attr __user *uattr)
5027 {
5028         struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
5029         struct bpf_link_info info;
5030         u32 info_len = attr->info.info_len;
5031         int err;
5032
5033         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
5034         if (err)
5035                 return err;
5036         info_len = min_t(u32, sizeof(info), info_len);
5037
5038         memset(&info, 0, sizeof(info));
5039         if (copy_from_user(&info, uinfo, info_len))
5040                 return -EFAULT;
5041
5042         info.type = link->type;
5043         info.id = link->id;
5044         if (link->prog)
5045                 info.prog_id = link->prog->aux->id;
5046
5047         if (link->ops->fill_link_info) {
5048                 err = link->ops->fill_link_info(link, &info);
5049                 if (err)
5050                         return err;
5051         }
5052
5053         if (copy_to_user(uinfo, &info, info_len) ||
5054             put_user(info_len, &uattr->info.info_len))
5055                 return -EFAULT;
5056
5057         return 0;
5058 }
5059
5060
5061 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
5062
5063 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
5064                                   union bpf_attr __user *uattr)
5065 {
5066         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
5067                 return -EINVAL;
5068
5069         CLASS(fd, f)(attr->info.bpf_fd);
5070         if (fd_empty(f))
5071                 return -EBADFD;
5072
5073         if (fd_file(f)->f_op == &bpf_prog_fops)
5074                 return bpf_prog_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr,
5075                                               uattr);
5076         else if (fd_file(f)->f_op == &bpf_map_fops)
5077                 return bpf_map_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr,
5078                                              uattr);
5079         else if (fd_file(f)->f_op == &btf_fops)
5080                 return bpf_btf_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr, uattr);
5081         else if (fd_file(f)->f_op == &bpf_link_fops || fd_file(f)->f_op == &bpf_link_fops_poll)
5082                 return bpf_link_get_info_by_fd(fd_file(f), fd_file(f)->private_data,
5083                                               attr, uattr);
5084         return -EINVAL;
5085 }
5086
5087 #define BPF_BTF_LOAD_LAST_FIELD btf_token_fd
5088
5089 static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
5090 {
5091         struct bpf_token *token = NULL;
5092
5093         if (CHECK_ATTR(BPF_BTF_LOAD))
5094                 return -EINVAL;
5095
5096         if (attr->btf_flags & ~BPF_F_TOKEN_FD)
5097                 return -EINVAL;
5098
5099         if (attr->btf_flags & BPF_F_TOKEN_FD) {
5100                 token = bpf_token_get_from_fd(attr->btf_token_fd);
5101                 if (IS_ERR(token))
5102                         return PTR_ERR(token);
5103                 if (!bpf_token_allow_cmd(token, BPF_BTF_LOAD)) {
5104                         bpf_token_put(token);
5105                         token = NULL;
5106                 }
5107         }
5108
5109         if (!bpf_token_capable(token, CAP_BPF)) {
5110                 bpf_token_put(token);
5111                 return -EPERM;
5112         }
5113
5114         bpf_token_put(token);
5115
5116         return btf_new_fd(attr, uattr, uattr_size);
5117 }
5118
5119 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
5120
5121 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
5122 {
5123         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
5124                 return -EINVAL;
5125
5126         if (!capable(CAP_SYS_ADMIN))
5127                 return -EPERM;
5128
5129         return btf_get_fd_by_id(attr->btf_id);
5130 }
5131
5132 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
5133                                     union bpf_attr __user *uattr,
5134                                     u32 prog_id, u32 fd_type,
5135                                     const char *buf, u64 probe_offset,
5136                                     u64 probe_addr)
5137 {
5138         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
5139         u32 len = buf ? strlen(buf) : 0, input_len;
5140         int err = 0;
5141
5142         if (put_user(len, &uattr->task_fd_query.buf_len))
5143                 return -EFAULT;
5144         input_len = attr->task_fd_query.buf_len;
5145         if (input_len && ubuf) {
5146                 if (!len) {
5147                         /* nothing to copy, just make ubuf NULL terminated */
5148                         char zero = '\0';
5149
5150                         if (put_user(zero, ubuf))
5151                                 return -EFAULT;
5152                 } else if (input_len >= len + 1) {
5153                         /* ubuf can hold the string with NULL terminator */
5154                         if (copy_to_user(ubuf, buf, len + 1))
5155                                 return -EFAULT;
5156                 } else {
5157                         /* ubuf cannot hold the string with NULL terminator,
5158                          * do a partial copy with NULL terminator.
5159                          */
5160                         char zero = '\0';
5161
5162                         err = -ENOSPC;
5163                         if (copy_to_user(ubuf, buf, input_len - 1))
5164                                 return -EFAULT;
5165                         if (put_user(zero, ubuf + input_len - 1))
5166                                 return -EFAULT;
5167                 }
5168         }
5169
5170         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
5171             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
5172             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
5173             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
5174                 return -EFAULT;
5175
5176         return err;
5177 }
5178
5179 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
5180
5181 static int bpf_task_fd_query(const union bpf_attr *attr,
5182                              union bpf_attr __user *uattr)
5183 {
5184         pid_t pid = attr->task_fd_query.pid;
5185         u32 fd = attr->task_fd_query.fd;
5186         const struct perf_event *event;
5187         struct task_struct *task;
5188         struct file *file;
5189         int err;
5190
5191         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
5192                 return -EINVAL;
5193
5194         if (!capable(CAP_SYS_ADMIN))
5195                 return -EPERM;
5196
5197         if (attr->task_fd_query.flags != 0)
5198                 return -EINVAL;
5199
5200         rcu_read_lock();
5201         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
5202         rcu_read_unlock();
5203         if (!task)
5204                 return -ENOENT;
5205
5206         err = 0;
5207         file = fget_task(task, fd);
5208         put_task_struct(task);
5209         if (!file)
5210                 return -EBADF;
5211
5212         if (file->f_op == &bpf_link_fops || file->f_op == &bpf_link_fops_poll) {
5213                 struct bpf_link *link = file->private_data;
5214
5215                 if (link->ops == &bpf_raw_tp_link_lops) {
5216                         struct bpf_raw_tp_link *raw_tp =
5217                                 container_of(link, struct bpf_raw_tp_link, link);
5218                         struct bpf_raw_event_map *btp = raw_tp->btp;
5219
5220                         err = bpf_task_fd_query_copy(attr, uattr,
5221                                                      raw_tp->link.prog->aux->id,
5222                                                      BPF_FD_TYPE_RAW_TRACEPOINT,
5223                                                      btp->tp->name, 0, 0);
5224                         goto put_file;
5225                 }
5226                 goto out_not_supp;
5227         }
5228
5229         event = perf_get_event(file);
5230         if (!IS_ERR(event)) {
5231                 u64 probe_offset, probe_addr;
5232                 u32 prog_id, fd_type;
5233                 const char *buf;
5234
5235                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
5236                                               &buf, &probe_offset,
5237                                               &probe_addr, NULL);
5238                 if (!err)
5239                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
5240                                                      fd_type, buf,
5241                                                      probe_offset,
5242                                                      probe_addr);
5243                 goto put_file;
5244         }
5245
5246 out_not_supp:
5247         err = -ENOTSUPP;
5248 put_file:
5249         fput(file);
5250         return err;
5251 }
5252
5253 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
5254
5255 #define BPF_DO_BATCH(fn, ...)                   \
5256         do {                                    \
5257                 if (!fn) {                      \
5258                         err = -ENOTSUPP;        \
5259                         goto err_put;           \
5260                 }                               \
5261                 err = fn(__VA_ARGS__);          \
5262         } while (0)
5263
5264 static int bpf_map_do_batch(const union bpf_attr *attr,
5265                             union bpf_attr __user *uattr,
5266                             int cmd)
5267 {
5268         bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
5269                          cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
5270         bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
5271         struct bpf_map *map;
5272         int err;
5273
5274         if (CHECK_ATTR(BPF_MAP_BATCH))
5275                 return -EINVAL;
5276
5277         CLASS(fd, f)(attr->batch.map_fd);
5278
5279         map = __bpf_map_get(f);
5280         if (IS_ERR(map))
5281                 return PTR_ERR(map);
5282         if (has_write)
5283                 bpf_map_write_active_inc(map);
5284         if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
5285                 err = -EPERM;
5286                 goto err_put;
5287         }
5288         if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
5289                 err = -EPERM;
5290                 goto err_put;
5291         }
5292
5293         if (cmd == BPF_MAP_LOOKUP_BATCH)
5294                 BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
5295         else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
5296                 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
5297         else if (cmd == BPF_MAP_UPDATE_BATCH)
5298                 BPF_DO_BATCH(map->ops->map_update_batch, map, fd_file(f), attr, uattr);
5299         else
5300                 BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
5301 err_put:
5302         if (has_write) {
5303                 maybe_wait_bpf_programs(map);
5304                 bpf_map_write_active_dec(map);
5305         }
5306         return err;
5307 }
5308
5309 #define BPF_LINK_CREATE_LAST_FIELD link_create.uprobe_multi.pid
5310 static int link_create(union bpf_attr *attr, bpfptr_t uattr)
5311 {
5312         struct bpf_prog *prog;
5313         int ret;
5314
5315         if (CHECK_ATTR(BPF_LINK_CREATE))
5316                 return -EINVAL;
5317
5318         if (attr->link_create.attach_type == BPF_STRUCT_OPS)
5319                 return bpf_struct_ops_link_create(attr);
5320
5321         prog = bpf_prog_get(attr->link_create.prog_fd);
5322         if (IS_ERR(prog))
5323                 return PTR_ERR(prog);
5324
5325         ret = bpf_prog_attach_check_attach_type(prog,
5326                                                 attr->link_create.attach_type);
5327         if (ret)
5328                 goto out;
5329
5330         switch (prog->type) {
5331         case BPF_PROG_TYPE_CGROUP_SKB:
5332         case BPF_PROG_TYPE_CGROUP_SOCK:
5333         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
5334         case BPF_PROG_TYPE_SOCK_OPS:
5335         case BPF_PROG_TYPE_CGROUP_DEVICE:
5336         case BPF_PROG_TYPE_CGROUP_SYSCTL:
5337         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
5338                 ret = cgroup_bpf_link_attach(attr, prog);
5339                 break;
5340         case BPF_PROG_TYPE_EXT:
5341                 ret = bpf_tracing_prog_attach(prog,
5342                                               attr->link_create.target_fd,
5343                                               attr->link_create.target_btf_id,
5344                                               attr->link_create.tracing.cookie);
5345                 break;
5346         case BPF_PROG_TYPE_LSM:
5347         case BPF_PROG_TYPE_TRACING:
5348                 if (attr->link_create.attach_type != prog->expected_attach_type) {
5349                         ret = -EINVAL;
5350                         goto out;
5351                 }
5352                 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
5353                         ret = bpf_raw_tp_link_attach(prog, NULL, attr->link_create.tracing.cookie);
5354                 else if (prog->expected_attach_type == BPF_TRACE_ITER)
5355                         ret = bpf_iter_link_attach(attr, uattr, prog);
5356                 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
5357                         ret = cgroup_bpf_link_attach(attr, prog);
5358                 else
5359                         ret = bpf_tracing_prog_attach(prog,
5360                                                       attr->link_create.target_fd,
5361                                                       attr->link_create.target_btf_id,
5362                                                       attr->link_create.tracing.cookie);
5363                 break;
5364         case BPF_PROG_TYPE_FLOW_DISSECTOR:
5365         case BPF_PROG_TYPE_SK_LOOKUP:
5366                 ret = netns_bpf_link_create(attr, prog);
5367                 break;
5368         case BPF_PROG_TYPE_SK_MSG:
5369         case BPF_PROG_TYPE_SK_SKB:
5370                 ret = sock_map_link_create(attr, prog);
5371                 break;
5372 #ifdef CONFIG_NET
5373         case BPF_PROG_TYPE_XDP:
5374                 ret = bpf_xdp_link_attach(attr, prog);
5375                 break;
5376         case BPF_PROG_TYPE_SCHED_CLS:
5377                 if (attr->link_create.attach_type == BPF_TCX_INGRESS ||
5378                     attr->link_create.attach_type == BPF_TCX_EGRESS)
5379                         ret = tcx_link_attach(attr, prog);
5380                 else
5381                         ret = netkit_link_attach(attr, prog);
5382                 break;
5383         case BPF_PROG_TYPE_NETFILTER:
5384                 ret = bpf_nf_link_attach(attr, prog);
5385                 break;
5386 #endif
5387         case BPF_PROG_TYPE_PERF_EVENT:
5388         case BPF_PROG_TYPE_TRACEPOINT:
5389                 ret = bpf_perf_link_attach(attr, prog);
5390                 break;
5391         case BPF_PROG_TYPE_KPROBE:
5392                 if (attr->link_create.attach_type == BPF_PERF_EVENT)
5393                         ret = bpf_perf_link_attach(attr, prog);
5394                 else if (attr->link_create.attach_type == BPF_TRACE_KPROBE_MULTI ||
5395                          attr->link_create.attach_type == BPF_TRACE_KPROBE_SESSION)
5396                         ret = bpf_kprobe_multi_link_attach(attr, prog);
5397                 else if (attr->link_create.attach_type == BPF_TRACE_UPROBE_MULTI ||
5398                          attr->link_create.attach_type == BPF_TRACE_UPROBE_SESSION)
5399                         ret = bpf_uprobe_multi_link_attach(attr, prog);
5400                 break;
5401         default:
5402                 ret = -EINVAL;
5403         }
5404
5405 out:
5406         if (ret < 0)
5407                 bpf_prog_put(prog);
5408         return ret;
5409 }
5410
5411 static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
5412 {
5413         struct bpf_map *new_map, *old_map = NULL;
5414         int ret;
5415
5416         new_map = bpf_map_get(attr->link_update.new_map_fd);
5417         if (IS_ERR(new_map))
5418                 return PTR_ERR(new_map);
5419
5420         if (attr->link_update.flags & BPF_F_REPLACE) {
5421                 old_map = bpf_map_get(attr->link_update.old_map_fd);
5422                 if (IS_ERR(old_map)) {
5423                         ret = PTR_ERR(old_map);
5424                         goto out_put;
5425                 }
5426         } else if (attr->link_update.old_map_fd) {
5427                 ret = -EINVAL;
5428                 goto out_put;
5429         }
5430
5431         ret = link->ops->update_map(link, new_map, old_map);
5432
5433         if (old_map)
5434                 bpf_map_put(old_map);
5435 out_put:
5436         bpf_map_put(new_map);
5437         return ret;
5438 }
5439
5440 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
5441
5442 static int link_update(union bpf_attr *attr)
5443 {
5444         struct bpf_prog *old_prog = NULL, *new_prog;
5445         struct bpf_link *link;
5446         u32 flags;
5447         int ret;
5448
5449         if (CHECK_ATTR(BPF_LINK_UPDATE))
5450                 return -EINVAL;
5451
5452         flags = attr->link_update.flags;
5453         if (flags & ~BPF_F_REPLACE)
5454                 return -EINVAL;
5455
5456         link = bpf_link_get_from_fd(attr->link_update.link_fd);
5457         if (IS_ERR(link))
5458                 return PTR_ERR(link);
5459
5460         if (link->ops->update_map) {
5461                 ret = link_update_map(link, attr);
5462                 goto out_put_link;
5463         }
5464
5465         new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
5466         if (IS_ERR(new_prog)) {
5467                 ret = PTR_ERR(new_prog);
5468                 goto out_put_link;
5469         }
5470
5471         if (flags & BPF_F_REPLACE) {
5472                 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
5473                 if (IS_ERR(old_prog)) {
5474                         ret = PTR_ERR(old_prog);
5475                         old_prog = NULL;
5476                         goto out_put_progs;
5477                 }
5478         } else if (attr->link_update.old_prog_fd) {
5479                 ret = -EINVAL;
5480                 goto out_put_progs;
5481         }
5482
5483         if (link->ops->update_prog)
5484                 ret = link->ops->update_prog(link, new_prog, old_prog);
5485         else
5486                 ret = -EINVAL;
5487
5488 out_put_progs:
5489         if (old_prog)
5490                 bpf_prog_put(old_prog);
5491         if (ret)
5492                 bpf_prog_put(new_prog);
5493 out_put_link:
5494         bpf_link_put_direct(link);
5495         return ret;
5496 }
5497
5498 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
5499
5500 static int link_detach(union bpf_attr *attr)
5501 {
5502         struct bpf_link *link;
5503         int ret;
5504
5505         if (CHECK_ATTR(BPF_LINK_DETACH))
5506                 return -EINVAL;
5507
5508         link = bpf_link_get_from_fd(attr->link_detach.link_fd);
5509         if (IS_ERR(link))
5510                 return PTR_ERR(link);
5511
5512         if (link->ops->detach)
5513                 ret = link->ops->detach(link);
5514         else
5515                 ret = -EOPNOTSUPP;
5516
5517         bpf_link_put_direct(link);
5518         return ret;
5519 }
5520
5521 struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
5522 {
5523         return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
5524 }
5525 EXPORT_SYMBOL(bpf_link_inc_not_zero);
5526
5527 struct bpf_link *bpf_link_by_id(u32 id)
5528 {
5529         struct bpf_link *link;
5530
5531         if (!id)
5532                 return ERR_PTR(-ENOENT);
5533
5534         spin_lock_bh(&link_idr_lock);
5535         /* before link is "settled", ID is 0, pretend it doesn't exist yet */
5536         link = idr_find(&link_idr, id);
5537         if (link) {
5538                 if (link->id)
5539                         link = bpf_link_inc_not_zero(link);
5540                 else
5541                         link = ERR_PTR(-EAGAIN);
5542         } else {
5543                 link = ERR_PTR(-ENOENT);
5544         }
5545         spin_unlock_bh(&link_idr_lock);
5546         return link;
5547 }
5548
5549 struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
5550 {
5551         struct bpf_link *link;
5552
5553         spin_lock_bh(&link_idr_lock);
5554 again:
5555         link = idr_get_next(&link_idr, id);
5556         if (link) {
5557                 link = bpf_link_inc_not_zero(link);
5558                 if (IS_ERR(link)) {
5559                         (*id)++;
5560                         goto again;
5561                 }
5562         }
5563         spin_unlock_bh(&link_idr_lock);
5564
5565         return link;
5566 }
5567
5568 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
5569
5570 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
5571 {
5572         struct bpf_link *link;
5573         u32 id = attr->link_id;
5574         int fd;
5575
5576         if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
5577                 return -EINVAL;
5578
5579         if (!capable(CAP_SYS_ADMIN))
5580                 return -EPERM;
5581
5582         link = bpf_link_by_id(id);
5583         if (IS_ERR(link))
5584                 return PTR_ERR(link);
5585
5586         fd = bpf_link_new_fd(link);
5587         if (fd < 0)
5588                 bpf_link_put_direct(link);
5589
5590         return fd;
5591 }
5592
5593 DEFINE_MUTEX(bpf_stats_enabled_mutex);
5594
5595 static int bpf_stats_release(struct inode *inode, struct file *file)
5596 {
5597         mutex_lock(&bpf_stats_enabled_mutex);
5598         static_key_slow_dec(&bpf_stats_enabled_key.key);
5599         mutex_unlock(&bpf_stats_enabled_mutex);
5600         return 0;
5601 }
5602
5603 static const struct file_operations bpf_stats_fops = {
5604         .release = bpf_stats_release,
5605 };
5606
5607 static int bpf_enable_runtime_stats(void)
5608 {
5609         int fd;
5610
5611         mutex_lock(&bpf_stats_enabled_mutex);
5612
5613         /* Set a very high limit to avoid overflow */
5614         if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
5615                 mutex_unlock(&bpf_stats_enabled_mutex);
5616                 return -EBUSY;
5617         }
5618
5619         fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
5620         if (fd >= 0)
5621                 static_key_slow_inc(&bpf_stats_enabled_key.key);
5622
5623         mutex_unlock(&bpf_stats_enabled_mutex);
5624         return fd;
5625 }
5626
5627 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
5628
5629 static int bpf_enable_stats(union bpf_attr *attr)
5630 {
5631
5632         if (CHECK_ATTR(BPF_ENABLE_STATS))
5633                 return -EINVAL;
5634
5635         if (!capable(CAP_SYS_ADMIN))
5636                 return -EPERM;
5637
5638         switch (attr->enable_stats.type) {
5639         case BPF_STATS_RUN_TIME:
5640                 return bpf_enable_runtime_stats();
5641         default:
5642                 break;
5643         }
5644         return -EINVAL;
5645 }
5646
5647 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
5648
5649 static int bpf_iter_create(union bpf_attr *attr)
5650 {
5651         struct bpf_link *link;
5652         int err;
5653
5654         if (CHECK_ATTR(BPF_ITER_CREATE))
5655                 return -EINVAL;
5656
5657         if (attr->iter_create.flags)
5658                 return -EINVAL;
5659
5660         link = bpf_link_get_from_fd(attr->iter_create.link_fd);
5661         if (IS_ERR(link))
5662                 return PTR_ERR(link);
5663
5664         err = bpf_iter_new_fd(link);
5665         bpf_link_put_direct(link);
5666
5667         return err;
5668 }
5669
5670 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
5671
5672 static int bpf_prog_bind_map(union bpf_attr *attr)
5673 {
5674         struct bpf_prog *prog;
5675         struct bpf_map *map;
5676         struct bpf_map **used_maps_old, **used_maps_new;
5677         int i, ret = 0;
5678
5679         if (CHECK_ATTR(BPF_PROG_BIND_MAP))
5680                 return -EINVAL;
5681
5682         if (attr->prog_bind_map.flags)
5683                 return -EINVAL;
5684
5685         prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
5686         if (IS_ERR(prog))
5687                 return PTR_ERR(prog);
5688
5689         map = bpf_map_get(attr->prog_bind_map.map_fd);
5690         if (IS_ERR(map)) {
5691                 ret = PTR_ERR(map);
5692                 goto out_prog_put;
5693         }
5694
5695         mutex_lock(&prog->aux->used_maps_mutex);
5696
5697         used_maps_old = prog->aux->used_maps;
5698
5699         for (i = 0; i < prog->aux->used_map_cnt; i++)
5700                 if (used_maps_old[i] == map) {
5701                         bpf_map_put(map);
5702                         goto out_unlock;
5703                 }
5704
5705         used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
5706                                       sizeof(used_maps_new[0]),
5707                                       GFP_KERNEL);
5708         if (!used_maps_new) {
5709                 ret = -ENOMEM;
5710                 goto out_unlock;
5711         }
5712
5713         /* The bpf program will not access the bpf map, but for the sake of
5714          * simplicity, increase sleepable_refcnt for sleepable program as well.
5715          */
5716         if (prog->sleepable)
5717                 atomic64_inc(&map->sleepable_refcnt);
5718         memcpy(used_maps_new, used_maps_old,
5719                sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
5720         used_maps_new[prog->aux->used_map_cnt] = map;
5721
5722         prog->aux->used_map_cnt++;
5723         prog->aux->used_maps = used_maps_new;
5724
5725         kfree(used_maps_old);
5726
5727 out_unlock:
5728         mutex_unlock(&prog->aux->used_maps_mutex);
5729
5730         if (ret)
5731                 bpf_map_put(map);
5732 out_prog_put:
5733         bpf_prog_put(prog);
5734         return ret;
5735 }
5736
5737 #define BPF_TOKEN_CREATE_LAST_FIELD token_create.bpffs_fd
5738
5739 static int token_create(union bpf_attr *attr)
5740 {
5741         if (CHECK_ATTR(BPF_TOKEN_CREATE))
5742                 return -EINVAL;
5743
5744         /* no flags are supported yet */
5745         if (attr->token_create.flags)
5746                 return -EINVAL;
5747
5748         return bpf_token_create(attr);
5749 }
5750
5751 static int __sys_bpf(enum bpf_cmd cmd, bpfptr_t uattr, unsigned int size)
5752 {
5753         union bpf_attr attr;
5754         int err;
5755
5756         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
5757         if (err)
5758                 return err;
5759         size = min_t(u32, size, sizeof(attr));
5760
5761         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
5762         memset(&attr, 0, sizeof(attr));
5763         if (copy_from_bpfptr(&attr, uattr, size) != 0)
5764                 return -EFAULT;
5765
5766         err = security_bpf(cmd, &attr, size);
5767         if (err < 0)
5768                 return err;
5769
5770         switch (cmd) {
5771         case BPF_MAP_CREATE:
5772                 err = map_create(&attr);
5773                 break;
5774         case BPF_MAP_LOOKUP_ELEM:
5775                 err = map_lookup_elem(&attr);
5776                 break;
5777         case BPF_MAP_UPDATE_ELEM:
5778                 err = map_update_elem(&attr, uattr);
5779                 break;
5780         case BPF_MAP_DELETE_ELEM:
5781                 err = map_delete_elem(&attr, uattr);
5782                 break;
5783         case BPF_MAP_GET_NEXT_KEY:
5784                 err = map_get_next_key(&attr);
5785                 break;
5786         case BPF_MAP_FREEZE:
5787                 err = map_freeze(&attr);
5788                 break;
5789         case BPF_PROG_LOAD:
5790                 err = bpf_prog_load(&attr, uattr, size);
5791                 break;
5792         case BPF_OBJ_PIN:
5793                 err = bpf_obj_pin(&attr);
5794                 break;
5795         case BPF_OBJ_GET:
5796                 err = bpf_obj_get(&attr);
5797                 break;
5798         case BPF_PROG_ATTACH:
5799                 err = bpf_prog_attach(&attr);
5800                 break;
5801         case BPF_PROG_DETACH:
5802                 err = bpf_prog_detach(&attr);
5803                 break;
5804         case BPF_PROG_QUERY:
5805                 err = bpf_prog_query(&attr, uattr.user);
5806                 break;
5807         case BPF_PROG_TEST_RUN:
5808                 err = bpf_prog_test_run(&attr, uattr.user);
5809                 break;
5810         case BPF_PROG_GET_NEXT_ID:
5811                 err = bpf_obj_get_next_id(&attr, uattr.user,
5812                                           &prog_idr, &prog_idr_lock);
5813                 break;
5814         case BPF_MAP_GET_NEXT_ID:
5815                 err = bpf_obj_get_next_id(&attr, uattr.user,
5816                                           &map_idr, &map_idr_lock);
5817                 break;
5818         case BPF_BTF_GET_NEXT_ID:
5819                 err = bpf_obj_get_next_id(&attr, uattr.user,
5820                                           &btf_idr, &btf_idr_lock);
5821                 break;
5822         case BPF_PROG_GET_FD_BY_ID:
5823                 err = bpf_prog_get_fd_by_id(&attr);
5824                 break;
5825         case BPF_MAP_GET_FD_BY_ID:
5826                 err = bpf_map_get_fd_by_id(&attr);
5827                 break;
5828         case BPF_OBJ_GET_INFO_BY_FD:
5829                 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5830                 break;
5831         case BPF_RAW_TRACEPOINT_OPEN:
5832                 err = bpf_raw_tracepoint_open(&attr);
5833                 break;
5834         case BPF_BTF_LOAD:
5835                 err = bpf_btf_load(&attr, uattr, size);
5836                 break;
5837         case BPF_BTF_GET_FD_BY_ID:
5838                 err = bpf_btf_get_fd_by_id(&attr);
5839                 break;
5840         case BPF_TASK_FD_QUERY:
5841                 err = bpf_task_fd_query(&attr, uattr.user);
5842                 break;
5843         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5844                 err = map_lookup_and_delete_elem(&attr);
5845                 break;
5846         case BPF_MAP_LOOKUP_BATCH:
5847                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5848                 break;
5849         case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5850                 err = bpf_map_do_batch(&attr, uattr.user,
5851                                        BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5852                 break;
5853         case BPF_MAP_UPDATE_BATCH:
5854                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5855                 break;
5856         case BPF_MAP_DELETE_BATCH:
5857                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5858                 break;
5859         case BPF_LINK_CREATE:
5860                 err = link_create(&attr, uattr);
5861                 break;
5862         case BPF_LINK_UPDATE:
5863                 err = link_update(&attr);
5864                 break;
5865         case BPF_LINK_GET_FD_BY_ID:
5866                 err = bpf_link_get_fd_by_id(&attr);
5867                 break;
5868         case BPF_LINK_GET_NEXT_ID:
5869                 err = bpf_obj_get_next_id(&attr, uattr.user,
5870                                           &link_idr, &link_idr_lock);
5871                 break;
5872         case BPF_ENABLE_STATS:
5873                 err = bpf_enable_stats(&attr);
5874                 break;
5875         case BPF_ITER_CREATE:
5876                 err = bpf_iter_create(&attr);
5877                 break;
5878         case BPF_LINK_DETACH:
5879                 err = link_detach(&attr);
5880                 break;
5881         case BPF_PROG_BIND_MAP:
5882                 err = bpf_prog_bind_map(&attr);
5883                 break;
5884         case BPF_TOKEN_CREATE:
5885                 err = token_create(&attr);
5886                 break;
5887         default:
5888                 err = -EINVAL;
5889                 break;
5890         }
5891
5892         return err;
5893 }
5894
5895 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5896 {
5897         return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5898 }
5899
5900 static bool syscall_prog_is_valid_access(int off, int size,
5901                                          enum bpf_access_type type,
5902                                          const struct bpf_prog *prog,
5903                                          struct bpf_insn_access_aux *info)
5904 {
5905         if (off < 0 || off >= U16_MAX)
5906                 return false;
5907         if (off % size != 0)
5908                 return false;
5909         return true;
5910 }
5911
5912 BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5913 {
5914         switch (cmd) {
5915         case BPF_MAP_CREATE:
5916         case BPF_MAP_DELETE_ELEM:
5917         case BPF_MAP_UPDATE_ELEM:
5918         case BPF_MAP_FREEZE:
5919         case BPF_MAP_GET_FD_BY_ID:
5920         case BPF_PROG_LOAD:
5921         case BPF_BTF_LOAD:
5922         case BPF_LINK_CREATE:
5923         case BPF_RAW_TRACEPOINT_OPEN:
5924                 break;
5925         default:
5926                 return -EINVAL;
5927         }
5928         return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5929 }
5930
5931
5932 /* To shut up -Wmissing-prototypes.
5933  * This function is used by the kernel light skeleton
5934  * to load bpf programs when modules are loaded or during kernel boot.
5935  * See tools/lib/bpf/skel_internal.h
5936  */
5937 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5938
5939 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5940 {
5941         struct bpf_prog * __maybe_unused prog;
5942         struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5943
5944         switch (cmd) {
5945 #ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5946         case BPF_PROG_TEST_RUN:
5947                 if (attr->test.data_in || attr->test.data_out ||
5948                     attr->test.ctx_out || attr->test.duration ||
5949                     attr->test.repeat || attr->test.flags)
5950                         return -EINVAL;
5951
5952                 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5953                 if (IS_ERR(prog))
5954                         return PTR_ERR(prog);
5955
5956                 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5957                     attr->test.ctx_size_in > U16_MAX) {
5958                         bpf_prog_put(prog);
5959                         return -EINVAL;
5960                 }
5961
5962                 run_ctx.bpf_cookie = 0;
5963                 if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5964                         /* recursion detected */
5965                         __bpf_prog_exit_sleepable_recur(prog, 0, &run_ctx);
5966                         bpf_prog_put(prog);
5967                         return -EBUSY;
5968                 }
5969                 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5970                 __bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5971                                                 &run_ctx);
5972                 bpf_prog_put(prog);
5973                 return 0;
5974 #endif
5975         default:
5976                 return ____bpf_sys_bpf(cmd, attr, size);
5977         }
5978 }
5979 EXPORT_SYMBOL(kern_sys_bpf);
5980
5981 static const struct bpf_func_proto bpf_sys_bpf_proto = {
5982         .func           = bpf_sys_bpf,
5983         .gpl_only       = false,
5984         .ret_type       = RET_INTEGER,
5985         .arg1_type      = ARG_ANYTHING,
5986         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5987         .arg3_type      = ARG_CONST_SIZE,
5988 };
5989
5990 const struct bpf_func_proto * __weak
5991 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5992 {
5993         return bpf_base_func_proto(func_id, prog);
5994 }
5995
5996 BPF_CALL_1(bpf_sys_close, u32, fd)
5997 {
5998         /* When bpf program calls this helper there should not be
5999          * an fdget() without matching completed fdput().
6000          * This helper is allowed in the following callchain only:
6001          * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
6002          */
6003         return close_fd(fd);
6004 }
6005
6006 static const struct bpf_func_proto bpf_sys_close_proto = {
6007         .func           = bpf_sys_close,
6008         .gpl_only       = false,
6009         .ret_type       = RET_INTEGER,
6010         .arg1_type      = ARG_ANYTHING,
6011 };
6012
6013 BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
6014 {
6015         *res = 0;
6016         if (flags)
6017                 return -EINVAL;
6018
6019         if (name_sz <= 1 || name[name_sz - 1])
6020                 return -EINVAL;
6021
6022         if (!bpf_dump_raw_ok(current_cred()))
6023                 return -EPERM;
6024
6025         *res = kallsyms_lookup_name(name);
6026         return *res ? 0 : -ENOENT;
6027 }
6028
6029 static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
6030         .func           = bpf_kallsyms_lookup_name,
6031         .gpl_only       = false,
6032         .ret_type       = RET_INTEGER,
6033         .arg1_type      = ARG_PTR_TO_MEM,
6034         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
6035         .arg3_type      = ARG_ANYTHING,
6036         .arg4_type      = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_WRITE | MEM_ALIGNED,
6037         .arg4_size      = sizeof(u64),
6038 };
6039
6040 static const struct bpf_func_proto *
6041 syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6042 {
6043         switch (func_id) {
6044         case BPF_FUNC_sys_bpf:
6045                 return !bpf_token_capable(prog->aux->token, CAP_PERFMON)
6046                        ? NULL : &bpf_sys_bpf_proto;
6047         case BPF_FUNC_btf_find_by_name_kind:
6048                 return &bpf_btf_find_by_name_kind_proto;
6049         case BPF_FUNC_sys_close:
6050                 return &bpf_sys_close_proto;
6051         case BPF_FUNC_kallsyms_lookup_name:
6052                 return &bpf_kallsyms_lookup_name_proto;
6053         default:
6054                 return tracing_prog_func_proto(func_id, prog);
6055         }
6056 }
6057
6058 const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
6059         .get_func_proto  = syscall_prog_func_proto,
6060         .is_valid_access = syscall_prog_is_valid_access,
6061 };
6062
6063 const struct bpf_prog_ops bpf_syscall_prog_ops = {
6064         .test_run = bpf_prog_test_run_syscall,
6065 };
6066
6067 #ifdef CONFIG_SYSCTL
6068 static int bpf_stats_handler(const struct ctl_table *table, int write,
6069                              void *buffer, size_t *lenp, loff_t *ppos)
6070 {
6071         struct static_key *key = (struct static_key *)table->data;
6072         static int saved_val;
6073         int val, ret;
6074         struct ctl_table tmp = {
6075                 .data   = &val,
6076                 .maxlen = sizeof(val),
6077                 .mode   = table->mode,
6078                 .extra1 = SYSCTL_ZERO,
6079                 .extra2 = SYSCTL_ONE,
6080         };
6081
6082         if (write && !capable(CAP_SYS_ADMIN))
6083                 return -EPERM;
6084
6085         mutex_lock(&bpf_stats_enabled_mutex);
6086         val = saved_val;
6087         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
6088         if (write && !ret && val != saved_val) {
6089                 if (val)
6090                         static_key_slow_inc(key);
6091                 else
6092                         static_key_slow_dec(key);
6093                 saved_val = val;
6094         }
6095         mutex_unlock(&bpf_stats_enabled_mutex);
6096         return ret;
6097 }
6098
6099 void __weak unpriv_ebpf_notify(int new_state)
6100 {
6101 }
6102
6103 static int bpf_unpriv_handler(const struct ctl_table *table, int write,
6104                               void *buffer, size_t *lenp, loff_t *ppos)
6105 {
6106         int ret, unpriv_enable = *(int *)table->data;
6107         bool locked_state = unpriv_enable == 1;
6108         struct ctl_table tmp = *table;
6109
6110         if (write && !capable(CAP_SYS_ADMIN))
6111                 return -EPERM;
6112
6113         tmp.data = &unpriv_enable;
6114         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
6115         if (write && !ret) {
6116                 if (locked_state && unpriv_enable != 1)
6117                         return -EPERM;
6118                 *(int *)table->data = unpriv_enable;
6119         }
6120
6121         if (write)
6122                 unpriv_ebpf_notify(unpriv_enable);
6123
6124         return ret;
6125 }
6126
6127 static struct ctl_table bpf_syscall_table[] = {
6128         {
6129                 .procname       = "unprivileged_bpf_disabled",
6130                 .data           = &sysctl_unprivileged_bpf_disabled,
6131                 .maxlen         = sizeof(sysctl_unprivileged_bpf_disabled),
6132                 .mode           = 0644,
6133                 .proc_handler   = bpf_unpriv_handler,
6134                 .extra1         = SYSCTL_ZERO,
6135                 .extra2         = SYSCTL_TWO,
6136         },
6137         {
6138                 .procname       = "bpf_stats_enabled",
6139                 .data           = &bpf_stats_enabled_key.key,
6140                 .mode           = 0644,
6141                 .proc_handler   = bpf_stats_handler,
6142         },
6143 };
6144
6145 static int __init bpf_syscall_sysctl_init(void)
6146 {
6147         register_sysctl_init("kernel", bpf_syscall_table);
6148         return 0;
6149 }
6150 late_initcall(bpf_syscall_sysctl_init);
6151 #endif /* CONFIG_SYSCTL */