kernel/bpf/syscall.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
bpf_check_uarg_tail_zero
find_and_alloc_map
bpf_map_area_alloc
bpf_map_area_free
bpf_map_flags_retain_permanent
bpf_map_init_from_attr
bpf_charge_memlock
bpf_uncharge_memlock
bpf_map_charge_init
bpf_map_charge_finish
bpf_map_charge_move
bpf_map_charge_memlock
bpf_map_uncharge_memlock
bpf_map_alloc_id
bpf_map_free_id
bpf_map_free_deferred
bpf_map_put_uref
__bpf_map_put
bpf_map_put
bpf_map_put_with_uref
bpf_map_release
map_get_sys_perms
bpf_map_show_fdinfo
bpf_dummy_read
bpf_dummy_write
bpf_map_new_fd
bpf_get_file_flag
bpf_obj_name_cpy
map_check_no_btf
map_check_btf
map_create
__bpf_map_get
bpf_map_inc
bpf_map_get_with_uref
__bpf_map_inc_not_zero
bpf_map_inc_not_zero
bpf_stackmap_copy
__bpf_copy_key
map_lookup_elem
maybe_wait_bpf_programs
map_update_elem
map_delete_elem
map_get_next_key
map_lookup_and_delete_elem
map_freeze
find_prog_type
free_used_maps
__bpf_prog_charge
__bpf_prog_uncharge
bpf_prog_charge_memlock
bpf_prog_uncharge_memlock
bpf_prog_alloc_id
bpf_prog_free_id
__bpf_prog_put_rcu
__bpf_prog_put_noref
__bpf_prog_put
bpf_prog_put
bpf_prog_release
bpf_prog_get_stats
bpf_prog_show_fdinfo
bpf_prog_new_fd
____bpf_prog_get
bpf_prog_add
bpf_prog_sub
bpf_prog_inc
bpf_prog_inc_not_zero
bpf_prog_get_ok
__bpf_prog_get
bpf_prog_get
bpf_prog_get_type_dev
bpf_prog_load_fixup_attach_type
bpf_prog_load_check_attach_type
bpf_prog_load
bpf_obj_pin
bpf_obj_get
bpf_raw_tracepoint_release
bpf_raw_tracepoint_open
bpf_prog_attach_check_attach_type
bpf_prog_attach
bpf_prog_detach
bpf_prog_query
bpf_prog_test_run
bpf_obj_get_next_id
bpf_prog_get_fd_by_id
bpf_map_get_fd_by_id
bpf_map_from_imm
bpf_insn_prepare_dump
set_info_rec_size
bpf_prog_get_info_by_fd
bpf_map_get_info_by_fd
bpf_btf_get_info_by_fd
bpf_obj_get_info_by_fd
bpf_btf_load
bpf_btf_get_fd_by_id
bpf_task_fd_query_copy
bpf_task_fd_query
SYSCALL_DEFINE3
   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_trace.h>
   6 #include <linux/bpf_lirc.h>
   7 #include <linux/btf.h>
   8 #include <linux/syscalls.h>
   9 #include <linux/slab.h>
  10 #include <linux/sched/signal.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/mmzone.h>
  13 #include <linux/anon_inodes.h>
  14 #include <linux/fdtable.h>
  15 #include <linux/file.h>
  16 #include <linux/fs.h>
  17 #include <linux/license.h>
  18 #include <linux/filter.h>
  19 #include <linux/version.h>
  20 #include <linux/kernel.h>
  21 #include <linux/idr.h>
  22 #include <linux/cred.h>
  23 #include <linux/timekeeping.h>
  24 #include <linux/ctype.h>
  25 #include <linux/nospec.h>
  26 
  27 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
  28                            (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  29                            (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  30                            (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  31 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  32 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
  33 
  34 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  35 
  36 DEFINE_PER_CPU(int, bpf_prog_active);
  37 static DEFINE_IDR(prog_idr);
  38 static DEFINE_SPINLOCK(prog_idr_lock);
  39 static DEFINE_IDR(map_idr);
  40 static DEFINE_SPINLOCK(map_idr_lock);
  41 
  42 int sysctl_unprivileged_bpf_disabled __read_mostly;
  43 
  44 static const struct bpf_map_ops * const bpf_map_types[] = {
  45 #define BPF_PROG_TYPE(_id, _ops)
  46 #define BPF_MAP_TYPE(_id, _ops) \
  47         [_id] = &_ops,
  48 #include <linux/bpf_types.h>
  49 #undef BPF_PROG_TYPE
  50 #undef BPF_MAP_TYPE
  51 };
  52 
  53 /*
  54  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  55  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  56  * we don't know about yet.
  57  *
  58  * There is a ToCToU between this function call and the following
  59  * copy_from_user() call. However, this is not a concern since this function is
  60  * meant to be a future-proofing of bits.
  61  */
  62 int bpf_check_uarg_tail_zero(void __user *uaddr,
  63                              size_t expected_size,
  64                              size_t actual_size)
  65 {
  66         unsigned char __user *addr;
  67         unsigned char __user *end;
  68         unsigned char val;
  69         int err;
  70 
  71         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  72                 return -E2BIG;
  73 
  74         if (unlikely(!access_ok(uaddr, actual_size)))
  75                 return -EFAULT;
  76 
  77         if (actual_size <= expected_size)
  78                 return 0;
  79 
  80         addr = uaddr + expected_size;
  81         end  = uaddr + actual_size;
  82 
  83         for (; addr < end; addr++) {
  84                 err = get_user(val, addr);
  85                 if (err)
  86                         return err;
  87                 if (val)
  88                         return -E2BIG;
  89         }
  90 
  91         return 0;
  92 }
  93 
  94 const struct bpf_map_ops bpf_map_offload_ops = {
  95         .map_alloc = bpf_map_offload_map_alloc,
  96         .map_free = bpf_map_offload_map_free,
  97         .map_check_btf = map_check_no_btf,
  98 };
  99 
 100 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 101 {
 102         const struct bpf_map_ops *ops;
 103         u32 type = attr->map_type;
 104         struct bpf_map *map;
 105         int err;
 106 
 107         if (type >= ARRAY_SIZE(bpf_map_types))
 108                 return ERR_PTR(-EINVAL);
 109         type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
 110         ops = bpf_map_types[type];
 111         if (!ops)
 112                 return ERR_PTR(-EINVAL);
 113 
 114         if (ops->map_alloc_check) {
 115                 err = ops->map_alloc_check(attr);
 116                 if (err)
 117                         return ERR_PTR(err);
 118         }
 119         if (attr->map_ifindex)
 120                 ops = &bpf_map_offload_ops;
 121         map = ops->map_alloc(attr);
 122         if (IS_ERR(map))
 123                 return map;
 124         map->ops = ops;
 125         map->map_type = type;
 126         return map;
 127 }
 128 
 129 void *bpf_map_area_alloc(u64 size, int numa_node)
 130 {
 131         /* We really just want to fail instead of triggering OOM killer
 132          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 133          * which is used for lower order allocation requests.
 134          *
 135          * It has been observed that higher order allocation requests done by
 136          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 137          * to reclaim memory from the page cache, thus we set
 138          * __GFP_RETRY_MAYFAIL to avoid such situations.
 139          */
 140 
 141         const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
 142         void *area;
 143 
 144         if (size >= SIZE_MAX)
 145                 return NULL;
 146 
 147         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 148                 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
 149                                     numa_node);
 150                 if (area != NULL)
 151                         return area;
 152         }
 153 
 154         return __vmalloc_node_flags_caller(size, numa_node,
 155                                            GFP_KERNEL | __GFP_RETRY_MAYFAIL |
 156                                            flags, __builtin_return_address(0));
 157 }
 158 
 159 void bpf_map_area_free(void *area)
 160 {
 161         kvfree(area);
 162 }
 163 
 164 static u32 bpf_map_flags_retain_permanent(u32 flags)
 165 {
 166         /* Some map creation flags are not tied to the map object but
 167          * rather to the map fd instead, so they have no meaning upon
 168          * map object inspection since multiple file descriptors with
 169          * different (access) properties can exist here. Thus, given
 170          * this has zero meaning for the map itself, lets clear these
 171          * from here.
 172          */
 173         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 174 }
 175 
 176 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 177 {
 178         map->map_type = attr->map_type;
 179         map->key_size = attr->key_size;
 180         map->value_size = attr->value_size;
 181         map->max_entries = attr->max_entries;
 182         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 183         map->numa_node = bpf_map_attr_numa_node(attr);
 184 }
 185 
 186 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
 187 {
 188         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 189 
 190         if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
 191                 atomic_long_sub(pages, &user->locked_vm);
 192                 return -EPERM;
 193         }
 194         return 0;
 195 }
 196 
 197 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
 198 {
 199         if (user)
 200                 atomic_long_sub(pages, &user->locked_vm);
 201 }
 202 
 203 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size)
 204 {
 205         u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
 206         struct user_struct *user;
 207         int ret;
 208 
 209         if (size >= U32_MAX - PAGE_SIZE)
 210                 return -E2BIG;
 211 
 212         user = get_current_user();
 213         ret = bpf_charge_memlock(user, pages);
 214         if (ret) {
 215                 free_uid(user);
 216                 return ret;
 217         }
 218 
 219         mem->pages = pages;
 220         mem->user = user;
 221 
 222         return 0;
 223 }
 224 
 225 void bpf_map_charge_finish(struct bpf_map_memory *mem)
 226 {
 227         bpf_uncharge_memlock(mem->user, mem->pages);
 228         free_uid(mem->user);
 229 }
 230 
 231 void bpf_map_charge_move(struct bpf_map_memory *dst,
 232                          struct bpf_map_memory *src)
 233 {
 234         *dst = *src;
 235 
 236         /* Make sure src will not be used for the redundant uncharging. */
 237         memset(src, 0, sizeof(struct bpf_map_memory));
 238 }
 239 
 240 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
 241 {
 242         int ret;
 243 
 244         ret = bpf_charge_memlock(map->memory.user, pages);
 245         if (ret)
 246                 return ret;
 247         map->memory.pages += pages;
 248         return ret;
 249 }
 250 
 251 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
 252 {
 253         bpf_uncharge_memlock(map->memory.user, pages);
 254         map->memory.pages -= pages;
 255 }
 256 
 257 static int bpf_map_alloc_id(struct bpf_map *map)
 258 {
 259         int id;
 260 
 261         idr_preload(GFP_KERNEL);
 262         spin_lock_bh(&map_idr_lock);
 263         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 264         if (id > 0)
 265                 map->id = id;
 266         spin_unlock_bh(&map_idr_lock);
 267         idr_preload_end();
 268 
 269         if (WARN_ON_ONCE(!id))
 270                 return -ENOSPC;
 271 
 272         return id > 0 ? 0 : id;
 273 }
 274 
 275 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 276 {
 277         unsigned long flags;
 278 
 279         /* Offloaded maps are removed from the IDR store when their device
 280          * disappears - even if someone holds an fd to them they are unusable,
 281          * the memory is gone, all ops will fail; they are simply waiting for
 282          * refcnt to drop to be freed.
 283          */
 284         if (!map->id)
 285                 return;
 286 
 287         if (do_idr_lock)
 288                 spin_lock_irqsave(&map_idr_lock, flags);
 289         else
 290                 __acquire(&map_idr_lock);
 291 
 292         idr_remove(&map_idr, map->id);
 293         map->id = 0;
 294 
 295         if (do_idr_lock)
 296                 spin_unlock_irqrestore(&map_idr_lock, flags);
 297         else
 298                 __release(&map_idr_lock);
 299 }
 300 
 301 /* called from workqueue */
 302 static void bpf_map_free_deferred(struct work_struct *work)
 303 {
 304         struct bpf_map *map = container_of(work, struct bpf_map, work);
 305         struct bpf_map_memory mem;
 306 
 307         bpf_map_charge_move(&mem, &map->memory);
 308         security_bpf_map_free(map);
 309         /* implementation dependent freeing */
 310         map->ops->map_free(map);
 311         bpf_map_charge_finish(&mem);
 312 }
 313 
 314 static void bpf_map_put_uref(struct bpf_map *map)
 315 {
 316         if (atomic_dec_and_test(&map->usercnt)) {
 317                 if (map->ops->map_release_uref)
 318                         map->ops->map_release_uref(map);
 319         }
 320 }
 321 
 322 /* decrement map refcnt and schedule it for freeing via workqueue
 323  * (unrelying map implementation ops->map_free() might sleep)
 324  */
 325 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 326 {
 327         if (atomic_dec_and_test(&map->refcnt)) {
 328                 /* bpf_map_free_id() must be called first */
 329                 bpf_map_free_id(map, do_idr_lock);
 330                 btf_put(map->btf);
 331                 INIT_WORK(&map->work, bpf_map_free_deferred);
 332                 schedule_work(&map->work);
 333         }
 334 }
 335 
 336 void bpf_map_put(struct bpf_map *map)
 337 {
 338         __bpf_map_put(map, true);
 339 }
 340 EXPORT_SYMBOL_GPL(bpf_map_put);
 341 
 342 void bpf_map_put_with_uref(struct bpf_map *map)
 343 {
 344         bpf_map_put_uref(map);
 345         bpf_map_put(map);
 346 }
 347 
 348 static int bpf_map_release(struct inode *inode, struct file *filp)
 349 {
 350         struct bpf_map *map = filp->private_data;
 351 
 352         if (map->ops->map_release)
 353                 map->ops->map_release(map, filp);
 354 
 355         bpf_map_put_with_uref(map);
 356         return 0;
 357 }
 358 
 359 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 360 {
 361         fmode_t mode = f.file->f_mode;
 362 
 363         /* Our file permissions may have been overridden by global
 364          * map permissions facing syscall side.
 365          */
 366         if (READ_ONCE(map->frozen))
 367                 mode &= ~FMODE_CAN_WRITE;
 368         return mode;
 369 }
 370 
 371 #ifdef CONFIG_PROC_FS
 372 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 373 {
 374         const struct bpf_map *map = filp->private_data;
 375         const struct bpf_array *array;
 376         u32 owner_prog_type = 0;
 377         u32 owner_jited = 0;
 378 
 379         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 380                 array = container_of(map, struct bpf_array, map);
 381                 owner_prog_type = array->owner_prog_type;
 382                 owner_jited = array->owner_jited;
 383         }
 384 
 385         seq_printf(m,
 386                    "map_type:\t%u\n"
 387                    "key_size:\t%u\n"
 388                    "value_size:\t%u\n"
 389                    "max_entries:\t%u\n"
 390                    "map_flags:\t%#x\n"
 391                    "memlock:\t%llu\n"
 392                    "map_id:\t%u\n"
 393                    "frozen:\t%u\n",
 394                    map->map_type,
 395                    map->key_size,
 396                    map->value_size,
 397                    map->max_entries,
 398                    map->map_flags,
 399                    map->memory.pages * 1ULL << PAGE_SHIFT,
 400                    map->id,
 401                    READ_ONCE(map->frozen));
 402 
 403         if (owner_prog_type) {
 404                 seq_printf(m, "owner_prog_type:\t%u\n",
 405                            owner_prog_type);
 406                 seq_printf(m, "owner_jited:\t%u\n",
 407                            owner_jited);
 408         }
 409 }
 410 #endif
 411 
 412 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 413                               loff_t *ppos)
 414 {
 415         /* We need this handler such that alloc_file() enables
 416          * f_mode with FMODE_CAN_READ.
 417          */
 418         return -EINVAL;
 419 }
 420 
 421 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 422                                size_t siz, loff_t *ppos)
 423 {
 424         /* We need this handler such that alloc_file() enables
 425          * f_mode with FMODE_CAN_WRITE.
 426          */
 427         return -EINVAL;
 428 }
 429 
 430 const struct file_operations bpf_map_fops = {
 431 #ifdef CONFIG_PROC_FS
 432         .show_fdinfo    = bpf_map_show_fdinfo,
 433 #endif
 434         .release        = bpf_map_release,
 435         .read           = bpf_dummy_read,
 436         .write          = bpf_dummy_write,
 437 };
 438 
 439 int bpf_map_new_fd(struct bpf_map *map, int flags)
 440 {
 441         int ret;
 442 
 443         ret = security_bpf_map(map, OPEN_FMODE(flags));
 444         if (ret < 0)
 445                 return ret;
 446 
 447         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 448                                 flags | O_CLOEXEC);
 449 }
 450 
 451 int bpf_get_file_flag(int flags)
 452 {
 453         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 454                 return -EINVAL;
 455         if (flags & BPF_F_RDONLY)
 456                 return O_RDONLY;
 457         if (flags & BPF_F_WRONLY)
 458                 return O_WRONLY;
 459         return O_RDWR;
 460 }
 461 
 462 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 463 #define CHECK_ATTR(CMD) \
 464         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 465                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 466                    sizeof(*attr) - \
 467                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 468                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 469 
 470 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
 471  * Return 0 on success and < 0 on error.
 472  */
 473 static int bpf_obj_name_cpy(char *dst, const char *src)
 474 {
 475         const char *end = src + BPF_OBJ_NAME_LEN;
 476 
 477         memset(dst, 0, BPF_OBJ_NAME_LEN);
 478         /* Copy all isalnum(), '_' and '.' chars. */
 479         while (src < end && *src) {
 480                 if (!isalnum(*src) &&
 481                     *src != '_' && *src != '.')
 482                         return -EINVAL;
 483                 *dst++ = *src++;
 484         }
 485 
 486         /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
 487         if (src == end)
 488                 return -EINVAL;
 489 
 490         return 0;
 491 }
 492 
 493 int map_check_no_btf(const struct bpf_map *map,
 494                      const struct btf *btf,
 495                      const struct btf_type *key_type,
 496                      const struct btf_type *value_type)
 497 {
 498         return -ENOTSUPP;
 499 }
 500 
 501 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 502                          u32 btf_key_id, u32 btf_value_id)
 503 {
 504         const struct btf_type *key_type, *value_type;
 505         u32 key_size, value_size;
 506         int ret = 0;
 507 
 508         /* Some maps allow key to be unspecified. */
 509         if (btf_key_id) {
 510                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 511                 if (!key_type || key_size != map->key_size)
 512                         return -EINVAL;
 513         } else {
 514                 key_type = btf_type_by_id(btf, 0);
 515                 if (!map->ops->map_check_btf)
 516                         return -EINVAL;
 517         }
 518 
 519         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
 520         if (!value_type || value_size != map->value_size)
 521                 return -EINVAL;
 522 
 523         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
 524 
 525         if (map_value_has_spin_lock(map)) {
 526                 if (map->map_flags & BPF_F_RDONLY_PROG)
 527                         return -EACCES;
 528                 if (map->map_type != BPF_MAP_TYPE_HASH &&
 529                     map->map_type != BPF_MAP_TYPE_ARRAY &&
 530                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
 531                     map->map_type != BPF_MAP_TYPE_SK_STORAGE)
 532                         return -ENOTSUPP;
 533                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
 534                     map->value_size) {
 535                         WARN_ONCE(1,
 536                                   "verifier bug spin_lock_off %d value_size %d\n",
 537                                   map->spin_lock_off, map->value_size);
 538                         return -EFAULT;
 539                 }
 540         }
 541 
 542         if (map->ops->map_check_btf)
 543                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
 544 
 545         return ret;
 546 }
 547 
 548 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
 549 /* called via syscall */
 550 static int map_create(union bpf_attr *attr)
 551 {
 552         int numa_node = bpf_map_attr_numa_node(attr);
 553         struct bpf_map_memory mem;
 554         struct bpf_map *map;
 555         int f_flags;
 556         int err;
 557 
 558         err = CHECK_ATTR(BPF_MAP_CREATE);
 559         if (err)
 560                 return -EINVAL;
 561 
 562         f_flags = bpf_get_file_flag(attr->map_flags);
 563         if (f_flags < 0)
 564                 return f_flags;
 565 
 566         if (numa_node != NUMA_NO_NODE &&
 567             ((unsigned int)numa_node >= nr_node_ids ||
 568              !node_online(numa_node)))
 569                 return -EINVAL;
 570 
 571         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 572         map = find_and_alloc_map(attr);
 573         if (IS_ERR(map))
 574                 return PTR_ERR(map);
 575 
 576         err = bpf_obj_name_cpy(map->name, attr->map_name);
 577         if (err)
 578                 goto free_map;
 579 
 580         atomic_set(&map->refcnt, 1);
 581         atomic_set(&map->usercnt, 1);
 582 
 583         if (attr->btf_key_type_id || attr->btf_value_type_id) {
 584                 struct btf *btf;
 585 
 586                 if (!attr->btf_value_type_id) {
 587                         err = -EINVAL;
 588                         goto free_map;
 589                 }
 590 
 591                 btf = btf_get_by_fd(attr->btf_fd);
 592                 if (IS_ERR(btf)) {
 593                         err = PTR_ERR(btf);
 594                         goto free_map;
 595                 }
 596 
 597                 err = map_check_btf(map, btf, attr->btf_key_type_id,
 598                                     attr->btf_value_type_id);
 599                 if (err) {
 600                         btf_put(btf);
 601                         goto free_map;
 602                 }
 603 
 604                 map->btf = btf;
 605                 map->btf_key_type_id = attr->btf_key_type_id;
 606                 map->btf_value_type_id = attr->btf_value_type_id;
 607         } else {
 608                 map->spin_lock_off = -EINVAL;
 609         }
 610 
 611         err = security_bpf_map_alloc(map);
 612         if (err)
 613                 goto free_map;
 614 
 615         err = bpf_map_alloc_id(map);
 616         if (err)
 617                 goto free_map_sec;
 618 
 619         err = bpf_map_new_fd(map, f_flags);
 620         if (err < 0) {
 621                 /* failed to allocate fd.
 622                  * bpf_map_put_with_uref() is needed because the above
 623                  * bpf_map_alloc_id() has published the map
 624                  * to the userspace and the userspace may
 625                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 626                  */
 627                 bpf_map_put_with_uref(map);
 628                 return err;
 629         }
 630 
 631         return err;
 632 
 633 free_map_sec:
 634         security_bpf_map_free(map);
 635 free_map:
 636         btf_put(map->btf);
 637         bpf_map_charge_move(&mem, &map->memory);
 638         map->ops->map_free(map);
 639         bpf_map_charge_finish(&mem);
 640         return err;
 641 }
 642 
 643 /* if error is returned, fd is released.
 644  * On success caller should complete fd access with matching fdput()
 645  */
 646 struct bpf_map *__bpf_map_get(struct fd f)
 647 {
 648         if (!f.file)
 649                 return ERR_PTR(-EBADF);
 650         if (f.file->f_op != &bpf_map_fops) {
 651                 fdput(f);
 652                 return ERR_PTR(-EINVAL);
 653         }
 654 
 655         return f.file->private_data;
 656 }
 657 
 658 /* prog's and map's refcnt limit */
 659 #define BPF_MAX_REFCNT 32768
 660 
 661 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
 662 {
 663         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
 664                 atomic_dec(&map->refcnt);
 665                 return ERR_PTR(-EBUSY);
 666         }
 667         if (uref)
 668                 atomic_inc(&map->usercnt);
 669         return map;
 670 }
 671 EXPORT_SYMBOL_GPL(bpf_map_inc);
 672 
 673 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 674 {
 675         struct fd f = fdget(ufd);
 676         struct bpf_map *map;
 677 
 678         map = __bpf_map_get(f);
 679         if (IS_ERR(map))
 680                 return map;
 681 
 682         map = bpf_map_inc(map, true);
 683         fdput(f);
 684 
 685         return map;
 686 }
 687 
 688 /* map_idr_lock should have been held */
 689 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map,
 690                                               bool uref)
 691 {
 692         int refold;
 693 
 694         refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
 695 
 696         if (refold >= BPF_MAX_REFCNT) {
 697                 __bpf_map_put(map, false);
 698                 return ERR_PTR(-EBUSY);
 699         }
 700 
 701         if (!refold)
 702                 return ERR_PTR(-ENOENT);
 703 
 704         if (uref)
 705                 atomic_inc(&map->usercnt);
 706 
 707         return map;
 708 }
 709 
 710 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
 711 {
 712         spin_lock_bh(&map_idr_lock);
 713         map = __bpf_map_inc_not_zero(map, uref);
 714         spin_unlock_bh(&map_idr_lock);
 715 
 716         return map;
 717 }
 718 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
 719 
 720 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 721 {
 722         return -ENOTSUPP;
 723 }
 724 
 725 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 726 {
 727         if (key_size)
 728                 return memdup_user(ukey, key_size);
 729 
 730         if (ukey)
 731                 return ERR_PTR(-EINVAL);
 732 
 733         return NULL;
 734 }
 735 
 736 /* last field in 'union bpf_attr' used by this command */
 737 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
 738 
 739 static int map_lookup_elem(union bpf_attr *attr)
 740 {
 741         void __user *ukey = u64_to_user_ptr(attr->key);
 742         void __user *uvalue = u64_to_user_ptr(attr->value);
 743         int ufd = attr->map_fd;
 744         struct bpf_map *map;
 745         void *key, *value, *ptr;
 746         u32 value_size;
 747         struct fd f;
 748         int err;
 749 
 750         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 751                 return -EINVAL;
 752 
 753         if (attr->flags & ~BPF_F_LOCK)
 754                 return -EINVAL;
 755 
 756         f = fdget(ufd);
 757         map = __bpf_map_get(f);
 758         if (IS_ERR(map))
 759                 return PTR_ERR(map);
 760         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
 761                 err = -EPERM;
 762                 goto err_put;
 763         }
 764 
 765         if ((attr->flags & BPF_F_LOCK) &&
 766             !map_value_has_spin_lock(map)) {
 767                 err = -EINVAL;
 768                 goto err_put;
 769         }
 770 
 771         key = __bpf_copy_key(ukey, map->key_size);
 772         if (IS_ERR(key)) {
 773                 err = PTR_ERR(key);
 774                 goto err_put;
 775         }
 776 
 777         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 778             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 779             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 780             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 781                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 782         else if (IS_FD_MAP(map))
 783                 value_size = sizeof(u32);
 784         else
 785                 value_size = map->value_size;
 786 
 787         err = -ENOMEM;
 788         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 789         if (!value)
 790                 goto free_key;
 791 
 792         if (bpf_map_is_dev_bound(map)) {
 793                 err = bpf_map_offload_lookup_elem(map, key, value);
 794                 goto done;
 795         }
 796 
 797         preempt_disable();
 798         this_cpu_inc(bpf_prog_active);
 799         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 800             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 801                 err = bpf_percpu_hash_copy(map, key, value);
 802         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 803                 err = bpf_percpu_array_copy(map, key, value);
 804         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 805                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 806         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 807                 err = bpf_stackmap_copy(map, key, value);
 808         } else if (IS_FD_ARRAY(map)) {
 809                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 810         } else if (IS_FD_HASH(map)) {
 811                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 812         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 813                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 814         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 815                    map->map_type == BPF_MAP_TYPE_STACK) {
 816                 err = map->ops->map_peek_elem(map, value);
 817         } else {
 818                 rcu_read_lock();
 819                 if (map->ops->map_lookup_elem_sys_only)
 820                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 821                 else
 822                         ptr = map->ops->map_lookup_elem(map, key);
 823                 if (IS_ERR(ptr)) {
 824                         err = PTR_ERR(ptr);
 825                 } else if (!ptr) {
 826                         err = -ENOENT;
 827                 } else {
 828                         err = 0;
 829                         if (attr->flags & BPF_F_LOCK)
 830                                 /* lock 'ptr' and copy everything but lock */
 831                                 copy_map_value_locked(map, value, ptr, true);
 832                         else
 833                                 copy_map_value(map, value, ptr);
 834                         /* mask lock, since value wasn't zero inited */
 835                         check_and_init_map_lock(map, value);
 836                 }
 837                 rcu_read_unlock();
 838         }
 839         this_cpu_dec(bpf_prog_active);
 840         preempt_enable();
 841 
 842 done:
 843         if (err)
 844                 goto free_value;
 845 
 846         err = -EFAULT;
 847         if (copy_to_user(uvalue, value, value_size) != 0)
 848                 goto free_value;
 849 
 850         err = 0;
 851 
 852 free_value:
 853         kfree(value);
 854 free_key:
 855         kfree(key);
 856 err_put:
 857         fdput(f);
 858         return err;
 859 }
 860 
 861 static void maybe_wait_bpf_programs(struct bpf_map *map)
 862 {
 863         /* Wait for any running BPF programs to complete so that
 864          * userspace, when we return to it, knows that all programs
 865          * that could be running use the new map value.
 866          */
 867         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 868             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 869                 synchronize_rcu();
 870 }
 871 
 872 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 873 
 874 static int map_update_elem(union bpf_attr *attr)
 875 {
 876         void __user *ukey = u64_to_user_ptr(attr->key);
 877         void __user *uvalue = u64_to_user_ptr(attr->value);
 878         int ufd = attr->map_fd;
 879         struct bpf_map *map;
 880         void *key, *value;
 881         u32 value_size;
 882         struct fd f;
 883         int err;
 884 
 885         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 886                 return -EINVAL;
 887 
 888         f = fdget(ufd);
 889         map = __bpf_map_get(f);
 890         if (IS_ERR(map))
 891                 return PTR_ERR(map);
 892         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
 893                 err = -EPERM;
 894                 goto err_put;
 895         }
 896 
 897         if ((attr->flags & BPF_F_LOCK) &&
 898             !map_value_has_spin_lock(map)) {
 899                 err = -EINVAL;
 900                 goto err_put;
 901         }
 902 
 903         key = __bpf_copy_key(ukey, map->key_size);
 904         if (IS_ERR(key)) {
 905                 err = PTR_ERR(key);
 906                 goto err_put;
 907         }
 908 
 909         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 910             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 911             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 912             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 913                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 914         else
 915                 value_size = map->value_size;
 916 
 917         err = -ENOMEM;
 918         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 919         if (!value)
 920                 goto free_key;
 921 
 922         err = -EFAULT;
 923         if (copy_from_user(value, uvalue, value_size) != 0)
 924                 goto free_value;
 925 
 926         /* Need to create a kthread, thus must support schedule */
 927         if (bpf_map_is_dev_bound(map)) {
 928                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
 929                 goto out;
 930         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 931                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 932                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 933                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 934                 goto out;
 935         }
 936 
 937         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 938          * inside bpf map update or delete otherwise deadlocks are possible
 939          */
 940         preempt_disable();
 941         __this_cpu_inc(bpf_prog_active);
 942         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 943             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 944                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 945         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 946                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 947         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 948                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 949                                                        attr->flags);
 950         } else if (IS_FD_ARRAY(map)) {
 951                 rcu_read_lock();
 952                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 953                                                    attr->flags);
 954                 rcu_read_unlock();
 955         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 956                 rcu_read_lock();
 957                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 958                                                   attr->flags);
 959                 rcu_read_unlock();
 960         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 961                 /* rcu_read_lock() is not needed */
 962                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 963                                                          attr->flags);
 964         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 965                    map->map_type == BPF_MAP_TYPE_STACK) {
 966                 err = map->ops->map_push_elem(map, value, attr->flags);
 967         } else {
 968                 rcu_read_lock();
 969                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 970                 rcu_read_unlock();
 971         }
 972         __this_cpu_dec(bpf_prog_active);
 973         preempt_enable();
 974         maybe_wait_bpf_programs(map);
 975 out:
 976 free_value:
 977         kfree(value);
 978 free_key:
 979         kfree(key);
 980 err_put:
 981         fdput(f);
 982         return err;
 983 }
 984 
 985 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 986 
 987 static int map_delete_elem(union bpf_attr *attr)
 988 {
 989         void __user *ukey = u64_to_user_ptr(attr->key);
 990         int ufd = attr->map_fd;
 991         struct bpf_map *map;
 992         struct fd f;
 993         void *key;
 994         int err;
 995 
 996         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 997                 return -EINVAL;
 998 
 999         f = fdget(ufd);
1000         map = __bpf_map_get(f);
1001         if (IS_ERR(map))
1002                 return PTR_ERR(map);
1003         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1004                 err = -EPERM;
1005                 goto err_put;
1006         }
1007 
1008         key = __bpf_copy_key(ukey, map->key_size);
1009         if (IS_ERR(key)) {
1010                 err = PTR_ERR(key);
1011                 goto err_put;
1012         }
1013 
1014         if (bpf_map_is_dev_bound(map)) {
1015                 err = bpf_map_offload_delete_elem(map, key);
1016                 goto out;
1017         }
1018 
1019         preempt_disable();
1020         __this_cpu_inc(bpf_prog_active);
1021         rcu_read_lock();
1022         err = map->ops->map_delete_elem(map, key);
1023         rcu_read_unlock();
1024         __this_cpu_dec(bpf_prog_active);
1025         preempt_enable();
1026         maybe_wait_bpf_programs(map);
1027 out:
1028         kfree(key);
1029 err_put:
1030         fdput(f);
1031         return err;
1032 }
1033 
1034 /* last field in 'union bpf_attr' used by this command */
1035 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1036 
1037 static int map_get_next_key(union bpf_attr *attr)
1038 {
1039         void __user *ukey = u64_to_user_ptr(attr->key);
1040         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1041         int ufd = attr->map_fd;
1042         struct bpf_map *map;
1043         void *key, *next_key;
1044         struct fd f;
1045         int err;
1046 
1047         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1048                 return -EINVAL;
1049 
1050         f = fdget(ufd);
1051         map = __bpf_map_get(f);
1052         if (IS_ERR(map))
1053                 return PTR_ERR(map);
1054         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1055                 err = -EPERM;
1056                 goto err_put;
1057         }
1058 
1059         if (ukey) {
1060                 key = __bpf_copy_key(ukey, map->key_size);
1061                 if (IS_ERR(key)) {
1062                         err = PTR_ERR(key);
1063                         goto err_put;
1064                 }
1065         } else {
1066                 key = NULL;
1067         }
1068 
1069         err = -ENOMEM;
1070         next_key = kmalloc(map->key_size, GFP_USER);
1071         if (!next_key)
1072                 goto free_key;
1073 
1074         if (bpf_map_is_dev_bound(map)) {
1075                 err = bpf_map_offload_get_next_key(map, key, next_key);
1076                 goto out;
1077         }
1078 
1079         rcu_read_lock();
1080         err = map->ops->map_get_next_key(map, key, next_key);
1081         rcu_read_unlock();
1082 out:
1083         if (err)
1084                 goto free_next_key;
1085 
1086         err = -EFAULT;
1087         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1088                 goto free_next_key;
1089 
1090         err = 0;
1091 
1092 free_next_key:
1093         kfree(next_key);
1094 free_key:
1095         kfree(key);
1096 err_put:
1097         fdput(f);
1098         return err;
1099 }
1100 
1101 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1102 
1103 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1104 {
1105         void __user *ukey = u64_to_user_ptr(attr->key);
1106         void __user *uvalue = u64_to_user_ptr(attr->value);
1107         int ufd = attr->map_fd;
1108         struct bpf_map *map;
1109         void *key, *value;
1110         u32 value_size;
1111         struct fd f;
1112         int err;
1113 
1114         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1115                 return -EINVAL;
1116 
1117         f = fdget(ufd);
1118         map = __bpf_map_get(f);
1119         if (IS_ERR(map))
1120                 return PTR_ERR(map);
1121         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1122                 err = -EPERM;
1123                 goto err_put;
1124         }
1125 
1126         key = __bpf_copy_key(ukey, map->key_size);
1127         if (IS_ERR(key)) {
1128                 err = PTR_ERR(key);
1129                 goto err_put;
1130         }
1131 
1132         value_size = map->value_size;
1133 
1134         err = -ENOMEM;
1135         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1136         if (!value)
1137                 goto free_key;
1138 
1139         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1140             map->map_type == BPF_MAP_TYPE_STACK) {
1141                 err = map->ops->map_pop_elem(map, value);
1142         } else {
1143                 err = -ENOTSUPP;
1144         }
1145 
1146         if (err)
1147                 goto free_value;
1148 
1149         if (copy_to_user(uvalue, value, value_size) != 0) {
1150                 err = -EFAULT;
1151                 goto free_value;
1152         }
1153 
1154         err = 0;
1155 
1156 free_value:
1157         kfree(value);
1158 free_key:
1159         kfree(key);
1160 err_put:
1161         fdput(f);
1162         return err;
1163 }
1164 
1165 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1166 
1167 static int map_freeze(const union bpf_attr *attr)
1168 {
1169         int err = 0, ufd = attr->map_fd;
1170         struct bpf_map *map;
1171         struct fd f;
1172 
1173         if (CHECK_ATTR(BPF_MAP_FREEZE))
1174                 return -EINVAL;
1175 
1176         f = fdget(ufd);
1177         map = __bpf_map_get(f);
1178         if (IS_ERR(map))
1179                 return PTR_ERR(map);
1180         if (READ_ONCE(map->frozen)) {
1181                 err = -EBUSY;
1182                 goto err_put;
1183         }
1184         if (!capable(CAP_SYS_ADMIN)) {
1185                 err = -EPERM;
1186                 goto err_put;
1187         }
1188 
1189         WRITE_ONCE(map->frozen, true);
1190 err_put:
1191         fdput(f);
1192         return err;
1193 }
1194 
1195 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1196 #define BPF_PROG_TYPE(_id, _name) \
1197         [_id] = & _name ## _prog_ops,
1198 #define BPF_MAP_TYPE(_id, _ops)
1199 #include <linux/bpf_types.h>
1200 #undef BPF_PROG_TYPE
1201 #undef BPF_MAP_TYPE
1202 };
1203 
1204 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1205 {
1206         const struct bpf_prog_ops *ops;
1207 
1208         if (type >= ARRAY_SIZE(bpf_prog_types))
1209                 return -EINVAL;
1210         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1211         ops = bpf_prog_types[type];
1212         if (!ops)
1213                 return -EINVAL;
1214 
1215         if (!bpf_prog_is_dev_bound(prog->aux))
1216                 prog->aux->ops = ops;
1217         else
1218                 prog->aux->ops = &bpf_offload_prog_ops;
1219         prog->type = type;
1220         return 0;
1221 }
1222 
1223 /* drop refcnt on maps used by eBPF program and free auxilary data */
1224 static void free_used_maps(struct bpf_prog_aux *aux)
1225 {
1226         enum bpf_cgroup_storage_type stype;
1227         int i;
1228 
1229         for_each_cgroup_storage_type(stype) {
1230                 if (!aux->cgroup_storage[stype])
1231                         continue;
1232                 bpf_cgroup_storage_release(aux->prog,
1233                                            aux->cgroup_storage[stype]);
1234         }
1235 
1236         for (i = 0; i < aux->used_map_cnt; i++)
1237                 bpf_map_put(aux->used_maps[i]);
1238 
1239         kfree(aux->used_maps);
1240 }
1241 
1242 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1243 {
1244         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1245         unsigned long user_bufs;
1246 
1247         if (user) {
1248                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1249                 if (user_bufs > memlock_limit) {
1250                         atomic_long_sub(pages, &user->locked_vm);
1251                         return -EPERM;
1252                 }
1253         }
1254 
1255         return 0;
1256 }
1257 
1258 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1259 {
1260         if (user)
1261                 atomic_long_sub(pages, &user->locked_vm);
1262 }
1263 
1264 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1265 {
1266         struct user_struct *user = get_current_user();
1267         int ret;
1268 
1269         ret = __bpf_prog_charge(user, prog->pages);
1270         if (ret) {
1271                 free_uid(user);
1272                 return ret;
1273         }
1274 
1275         prog->aux->user = user;
1276         return 0;
1277 }
1278 
1279 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1280 {
1281         struct user_struct *user = prog->aux->user;
1282 
1283         __bpf_prog_uncharge(user, prog->pages);
1284         free_uid(user);
1285 }
1286 
1287 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1288 {
1289         int id;
1290 
1291         idr_preload(GFP_KERNEL);
1292         spin_lock_bh(&prog_idr_lock);
1293         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1294         if (id > 0)
1295                 prog->aux->id = id;
1296         spin_unlock_bh(&prog_idr_lock);
1297         idr_preload_end();
1298 
1299         /* id is in [1, INT_MAX) */
1300         if (WARN_ON_ONCE(!id))
1301                 return -ENOSPC;
1302 
1303         return id > 0 ? 0 : id;
1304 }
1305 
1306 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1307 {
1308         /* cBPF to eBPF migrations are currently not in the idr store.
1309          * Offloaded programs are removed from the store when their device
1310          * disappears - even if someone grabs an fd to them they are unusable,
1311          * simply waiting for refcnt to drop to be freed.
1312          */
1313         if (!prog->aux->id)
1314                 return;
1315 
1316         if (do_idr_lock)
1317                 spin_lock_bh(&prog_idr_lock);
1318         else
1319                 __acquire(&prog_idr_lock);
1320 
1321         idr_remove(&prog_idr, prog->aux->id);
1322         prog->aux->id = 0;
1323 
1324         if (do_idr_lock)
1325                 spin_unlock_bh(&prog_idr_lock);
1326         else
1327                 __release(&prog_idr_lock);
1328 }
1329 
1330 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1331 {
1332         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1333 
1334         kvfree(aux->func_info);
1335         free_used_maps(aux);
1336         bpf_prog_uncharge_memlock(aux->prog);
1337         security_bpf_prog_free(aux);
1338         bpf_prog_free(aux->prog);
1339 }
1340 
1341 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1342 {
1343         bpf_prog_kallsyms_del_all(prog);
1344         btf_put(prog->aux->btf);
1345         bpf_prog_free_linfo(prog);
1346 
1347         if (deferred)
1348                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1349         else
1350                 __bpf_prog_put_rcu(&prog->aux->rcu);
1351 }
1352 
1353 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1354 {
1355         if (atomic_dec_and_test(&prog->aux->refcnt)) {
1356                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1357                 /* bpf_prog_free_id() must be called first */
1358                 bpf_prog_free_id(prog, do_idr_lock);
1359                 __bpf_prog_put_noref(prog, true);
1360         }
1361 }
1362 
1363 void bpf_prog_put(struct bpf_prog *prog)
1364 {
1365         __bpf_prog_put(prog, true);
1366 }
1367 EXPORT_SYMBOL_GPL(bpf_prog_put);
1368 
1369 static int bpf_prog_release(struct inode *inode, struct file *filp)
1370 {
1371         struct bpf_prog *prog = filp->private_data;
1372 
1373         bpf_prog_put(prog);
1374         return 0;
1375 }
1376 
1377 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1378                                struct bpf_prog_stats *stats)
1379 {
1380         u64 nsecs = 0, cnt = 0;
1381         int cpu;
1382 
1383         for_each_possible_cpu(cpu) {
1384                 const struct bpf_prog_stats *st;
1385                 unsigned int start;
1386                 u64 tnsecs, tcnt;
1387 
1388                 st = per_cpu_ptr(prog->aux->stats, cpu);
1389                 do {
1390                         start = u64_stats_fetch_begin_irq(&st->syncp);
1391                         tnsecs = st->nsecs;
1392                         tcnt = st->cnt;
1393                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1394                 nsecs += tnsecs;
1395                 cnt += tcnt;
1396         }
1397         stats->nsecs = nsecs;
1398         stats->cnt = cnt;
1399 }
1400 
1401 #ifdef CONFIG_PROC_FS
1402 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1403 {
1404         const struct bpf_prog *prog = filp->private_data;
1405         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1406         struct bpf_prog_stats stats;
1407 
1408         bpf_prog_get_stats(prog, &stats);
1409         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1410         seq_printf(m,
1411                    "prog_type:\t%u\n"
1412                    "prog_jited:\t%u\n"
1413                    "prog_tag:\t%s\n"
1414                    "memlock:\t%llu\n"
1415                    "prog_id:\t%u\n"
1416                    "run_time_ns:\t%llu\n"
1417                    "run_cnt:\t%llu\n",
1418                    prog->type,
1419                    prog->jited,
1420                    prog_tag,
1421                    prog->pages * 1ULL << PAGE_SHIFT,
1422                    prog->aux->id,
1423                    stats.nsecs,
1424                    stats.cnt);
1425 }
1426 #endif
1427 
1428 const struct file_operations bpf_prog_fops = {
1429 #ifdef CONFIG_PROC_FS
1430         .show_fdinfo    = bpf_prog_show_fdinfo,
1431 #endif
1432         .release        = bpf_prog_release,
1433         .read           = bpf_dummy_read,
1434         .write          = bpf_dummy_write,
1435 };
1436 
1437 int bpf_prog_new_fd(struct bpf_prog *prog)
1438 {
1439         int ret;
1440 
1441         ret = security_bpf_prog(prog);
1442         if (ret < 0)
1443                 return ret;
1444 
1445         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1446                                 O_RDWR | O_CLOEXEC);
1447 }
1448 
1449 static struct bpf_prog *____bpf_prog_get(struct fd f)
1450 {
1451         if (!f.file)
1452                 return ERR_PTR(-EBADF);
1453         if (f.file->f_op != &bpf_prog_fops) {
1454                 fdput(f);
1455                 return ERR_PTR(-EINVAL);
1456         }
1457 
1458         return f.file->private_data;
1459 }
1460 
1461 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1462 {
1463         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1464                 atomic_sub(i, &prog->aux->refcnt);
1465                 return ERR_PTR(-EBUSY);
1466         }
1467         return prog;
1468 }
1469 EXPORT_SYMBOL_GPL(bpf_prog_add);
1470 
1471 void bpf_prog_sub(struct bpf_prog *prog, int i)
1472 {
1473         /* Only to be used for undoing previous bpf_prog_add() in some
1474          * error path. We still know that another entity in our call
1475          * path holds a reference to the program, thus atomic_sub() can
1476          * be safely used in such cases!
1477          */
1478         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1479 }
1480 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1481 
1482 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1483 {
1484         return bpf_prog_add(prog, 1);
1485 }
1486 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1487 
1488 /* prog_idr_lock should have been held */
1489 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1490 {
1491         int refold;
1492 
1493         refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1494 
1495         if (refold >= BPF_MAX_REFCNT) {
1496                 __bpf_prog_put(prog, false);
1497                 return ERR_PTR(-EBUSY);
1498         }
1499 
1500         if (!refold)
1501                 return ERR_PTR(-ENOENT);
1502 
1503         return prog;
1504 }
1505 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1506 
1507 bool bpf_prog_get_ok(struct bpf_prog *prog,
1508                             enum bpf_prog_type *attach_type, bool attach_drv)
1509 {
1510         /* not an attachment, just a refcount inc, always allow */
1511         if (!attach_type)
1512                 return true;
1513 
1514         if (prog->type != *attach_type)
1515                 return false;
1516         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1517                 return false;
1518 
1519         return true;
1520 }
1521 
1522 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1523                                        bool attach_drv)
1524 {
1525         struct fd f = fdget(ufd);
1526         struct bpf_prog *prog;
1527 
1528         prog = ____bpf_prog_get(f);
1529         if (IS_ERR(prog))
1530                 return prog;
1531         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1532                 prog = ERR_PTR(-EINVAL);
1533                 goto out;
1534         }
1535 
1536         prog = bpf_prog_inc(prog);
1537 out:
1538         fdput(f);
1539         return prog;
1540 }
1541 
1542 struct bpf_prog *bpf_prog_get(u32 ufd)
1543 {
1544         return __bpf_prog_get(ufd, NULL, false);
1545 }
1546 
1547 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1548                                        bool attach_drv)
1549 {
1550         return __bpf_prog_get(ufd, &type, attach_drv);
1551 }
1552 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1553 
1554 /* Initially all BPF programs could be loaded w/o specifying
1555  * expected_attach_type. Later for some of them specifying expected_attach_type
1556  * at load time became required so that program could be validated properly.
1557  * Programs of types that are allowed to be loaded both w/ and w/o (for
1558  * backward compatibility) expected_attach_type, should have the default attach
1559  * type assigned to expected_attach_type for the latter case, so that it can be
1560  * validated later at attach time.
1561  *
1562  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1563  * prog type requires it but has some attach types that have to be backward
1564  * compatible.
1565  */
1566 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1567 {
1568         switch (attr->prog_type) {
1569         case BPF_PROG_TYPE_CGROUP_SOCK:
1570                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1571                  * exist so checking for non-zero is the way to go here.
1572                  */
1573                 if (!attr->expected_attach_type)
1574                         attr->expected_attach_type =
1575                                 BPF_CGROUP_INET_SOCK_CREATE;
1576                 break;
1577         }
1578 }
1579 
1580 static int
1581 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1582                                 enum bpf_attach_type expected_attach_type)
1583 {
1584         switch (prog_type) {
1585         case BPF_PROG_TYPE_CGROUP_SOCK:
1586                 switch (expected_attach_type) {
1587                 case BPF_CGROUP_INET_SOCK_CREATE:
1588                 case BPF_CGROUP_INET4_POST_BIND:
1589                 case BPF_CGROUP_INET6_POST_BIND:
1590                         return 0;
1591                 default:
1592                         return -EINVAL;
1593                 }
1594         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1595                 switch (expected_attach_type) {
1596                 case BPF_CGROUP_INET4_BIND:
1597                 case BPF_CGROUP_INET6_BIND:
1598                 case BPF_CGROUP_INET4_CONNECT:
1599                 case BPF_CGROUP_INET6_CONNECT:
1600                 case BPF_CGROUP_UDP4_SENDMSG:
1601                 case BPF_CGROUP_UDP6_SENDMSG:
1602                 case BPF_CGROUP_UDP4_RECVMSG:
1603                 case BPF_CGROUP_UDP6_RECVMSG:
1604                         return 0;
1605                 default:
1606                         return -EINVAL;
1607                 }
1608         case BPF_PROG_TYPE_CGROUP_SKB:
1609                 switch (expected_attach_type) {
1610                 case BPF_CGROUP_INET_INGRESS:
1611                 case BPF_CGROUP_INET_EGRESS:
1612                         return 0;
1613                 default:
1614                         return -EINVAL;
1615                 }
1616         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1617                 switch (expected_attach_type) {
1618                 case BPF_CGROUP_SETSOCKOPT:
1619                 case BPF_CGROUP_GETSOCKOPT:
1620                         return 0;
1621                 default:
1622                         return -EINVAL;
1623                 }
1624         default:
1625                 return 0;
1626         }
1627 }
1628 
1629 /* last field in 'union bpf_attr' used by this command */
1630 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1631 
1632 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1633 {
1634         enum bpf_prog_type type = attr->prog_type;
1635         struct bpf_prog *prog;
1636         int err;
1637         char license[128];
1638         bool is_gpl;
1639 
1640         if (CHECK_ATTR(BPF_PROG_LOAD))
1641                 return -EINVAL;
1642 
1643         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
1644                                  BPF_F_ANY_ALIGNMENT |
1645                                  BPF_F_TEST_STATE_FREQ |
1646                                  BPF_F_TEST_RND_HI32))
1647                 return -EINVAL;
1648 
1649         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1650             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1651             !capable(CAP_SYS_ADMIN))
1652                 return -EPERM;
1653 
1654         /* copy eBPF program license from user space */
1655         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1656                               sizeof(license) - 1) < 0)
1657                 return -EFAULT;
1658         license[sizeof(license) - 1] = 0;
1659 
1660         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1661         is_gpl = license_is_gpl_compatible(license);
1662 
1663         if (attr->insn_cnt == 0 ||
1664             attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1665                 return -E2BIG;
1666         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1667             type != BPF_PROG_TYPE_CGROUP_SKB &&
1668             !capable(CAP_SYS_ADMIN))
1669                 return -EPERM;
1670 
1671         bpf_prog_load_fixup_attach_type(attr);
1672         if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1673                 return -EINVAL;
1674 
1675         /* plain bpf_prog allocation */
1676         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1677         if (!prog)
1678                 return -ENOMEM;
1679 
1680         prog->expected_attach_type = attr->expected_attach_type;
1681 
1682         prog->aux->offload_requested = !!attr->prog_ifindex;
1683 
1684         err = security_bpf_prog_alloc(prog->aux);
1685         if (err)
1686                 goto free_prog_nouncharge;
1687 
1688         err = bpf_prog_charge_memlock(prog);
1689         if (err)
1690                 goto free_prog_sec;
1691 
1692         prog->len = attr->insn_cnt;
1693 
1694         err = -EFAULT;
1695         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1696                            bpf_prog_insn_size(prog)) != 0)
1697                 goto free_prog;
1698 
1699         prog->orig_prog = NULL;
1700         prog->jited = 0;
1701 
1702         atomic_set(&prog->aux->refcnt, 1);
1703         prog->gpl_compatible = is_gpl ? 1 : 0;
1704 
1705         if (bpf_prog_is_dev_bound(prog->aux)) {
1706                 err = bpf_prog_offload_init(prog, attr);
1707                 if (err)
1708                         goto free_prog;
1709         }
1710 
1711         /* find program type: socket_filter vs tracing_filter */
1712         err = find_prog_type(type, prog);
1713         if (err < 0)
1714                 goto free_prog;
1715 
1716         prog->aux->load_time = ktime_get_boottime_ns();
1717         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1718         if (err)
1719                 goto free_prog;
1720 
1721         /* run eBPF verifier */
1722         err = bpf_check(&prog, attr, uattr);
1723         if (err < 0)
1724                 goto free_used_maps;
1725 
1726         prog = bpf_prog_select_runtime(prog, &err);
1727         if (err < 0)
1728                 goto free_used_maps;
1729 
1730         err = bpf_prog_alloc_id(prog);
1731         if (err)
1732                 goto free_used_maps;
1733 
1734         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
1735          * effectively publicly exposed. However, retrieving via
1736          * bpf_prog_get_fd_by_id() will take another reference,
1737          * therefore it cannot be gone underneath us.
1738          *
1739          * Only for the time /after/ successful bpf_prog_new_fd()
1740          * and before returning to userspace, we might just hold
1741          * one reference and any parallel close on that fd could
1742          * rip everything out. Hence, below notifications must
1743          * happen before bpf_prog_new_fd().
1744          *
1745          * Also, any failure handling from this point onwards must
1746          * be using bpf_prog_put() given the program is exposed.
1747          */
1748         bpf_prog_kallsyms_add(prog);
1749         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1750 
1751         err = bpf_prog_new_fd(prog);
1752         if (err < 0)
1753                 bpf_prog_put(prog);
1754         return err;
1755 
1756 free_used_maps:
1757         /* In case we have subprogs, we need to wait for a grace
1758          * period before we can tear down JIT memory since symbols
1759          * are already exposed under kallsyms.
1760          */
1761         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
1762         return err;
1763 free_prog:
1764         bpf_prog_uncharge_memlock(prog);
1765 free_prog_sec:
1766         security_bpf_prog_free(prog->aux);
1767 free_prog_nouncharge:
1768         bpf_prog_free(prog);
1769         return err;
1770 }
1771 
1772 #define BPF_OBJ_LAST_FIELD file_flags
1773 
1774 static int bpf_obj_pin(const union bpf_attr *attr)
1775 {
1776         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1777                 return -EINVAL;
1778 
1779         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1780 }
1781 
1782 static int bpf_obj_get(const union bpf_attr *attr)
1783 {
1784         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1785             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1786                 return -EINVAL;
1787 
1788         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1789                                 attr->file_flags);
1790 }
1791 
1792 struct bpf_raw_tracepoint {
1793         struct bpf_raw_event_map *btp;
1794         struct bpf_prog *prog;
1795 };
1796 
1797 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1798 {
1799         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1800 
1801         if (raw_tp->prog) {
1802                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1803                 bpf_prog_put(raw_tp->prog);
1804         }
1805         bpf_put_raw_tracepoint(raw_tp->btp);
1806         kfree(raw_tp);
1807         return 0;
1808 }
1809 
1810 static const struct file_operations bpf_raw_tp_fops = {
1811         .release        = bpf_raw_tracepoint_release,
1812         .read           = bpf_dummy_read,
1813         .write          = bpf_dummy_write,
1814 };
1815 
1816 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1817 
1818 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1819 {
1820         struct bpf_raw_tracepoint *raw_tp;
1821         struct bpf_raw_event_map *btp;
1822         struct bpf_prog *prog;
1823         char tp_name[128];
1824         int tp_fd, err;
1825 
1826         if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1827                               sizeof(tp_name) - 1) < 0)
1828                 return -EFAULT;
1829         tp_name[sizeof(tp_name) - 1] = 0;
1830 
1831         btp = bpf_get_raw_tracepoint(tp_name);
1832         if (!btp)
1833                 return -ENOENT;
1834 
1835         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1836         if (!raw_tp) {
1837                 err = -ENOMEM;
1838                 goto out_put_btp;
1839         }
1840         raw_tp->btp = btp;
1841 
1842         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1843         if (IS_ERR(prog)) {
1844                 err = PTR_ERR(prog);
1845                 goto out_free_tp;
1846         }
1847         if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1848             prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1849                 err = -EINVAL;
1850                 goto out_put_prog;
1851         }
1852 
1853         err = bpf_probe_register(raw_tp->btp, prog);
1854         if (err)
1855                 goto out_put_prog;
1856 
1857         raw_tp->prog = prog;
1858         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1859                                  O_CLOEXEC);
1860         if (tp_fd < 0) {
1861                 bpf_probe_unregister(raw_tp->btp, prog);
1862                 err = tp_fd;
1863                 goto out_put_prog;
1864         }
1865         return tp_fd;
1866 
1867 out_put_prog:
1868         bpf_prog_put(prog);
1869 out_free_tp:
1870         kfree(raw_tp);
1871 out_put_btp:
1872         bpf_put_raw_tracepoint(btp);
1873         return err;
1874 }
1875 
1876 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1877                                              enum bpf_attach_type attach_type)
1878 {
1879         switch (prog->type) {
1880         case BPF_PROG_TYPE_CGROUP_SOCK:
1881         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1882         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1883                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1884         case BPF_PROG_TYPE_CGROUP_SKB:
1885                 return prog->enforce_expected_attach_type &&
1886                         prog->expected_attach_type != attach_type ?
1887                         -EINVAL : 0;
1888         default:
1889                 return 0;
1890         }
1891 }
1892 
1893 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1894 
1895 #define BPF_F_ATTACH_MASK \
1896         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1897 
1898 static int bpf_prog_attach(const union bpf_attr *attr)
1899 {
1900         enum bpf_prog_type ptype;
1901         struct bpf_prog *prog;
1902         int ret;
1903 
1904         if (!capable(CAP_NET_ADMIN))
1905                 return -EPERM;
1906 
1907         if (CHECK_ATTR(BPF_PROG_ATTACH))
1908                 return -EINVAL;
1909 
1910         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1911                 return -EINVAL;
1912 
1913         switch (attr->attach_type) {
1914         case BPF_CGROUP_INET_INGRESS:
1915         case BPF_CGROUP_INET_EGRESS:
1916                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1917                 break;
1918         case BPF_CGROUP_INET_SOCK_CREATE:
1919         case BPF_CGROUP_INET4_POST_BIND:
1920         case BPF_CGROUP_INET6_POST_BIND:
1921                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1922                 break;
1923         case BPF_CGROUP_INET4_BIND:
1924         case BPF_CGROUP_INET6_BIND:
1925         case BPF_CGROUP_INET4_CONNECT:
1926         case BPF_CGROUP_INET6_CONNECT:
1927         case BPF_CGROUP_UDP4_SENDMSG:
1928         case BPF_CGROUP_UDP6_SENDMSG:
1929         case BPF_CGROUP_UDP4_RECVMSG:
1930         case BPF_CGROUP_UDP6_RECVMSG:
1931                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1932                 break;
1933         case BPF_CGROUP_SOCK_OPS:
1934                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1935                 break;
1936         case BPF_CGROUP_DEVICE:
1937                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1938                 break;
1939         case BPF_SK_MSG_VERDICT:
1940                 ptype = BPF_PROG_TYPE_SK_MSG;
1941                 break;
1942         case BPF_SK_SKB_STREAM_PARSER:
1943         case BPF_SK_SKB_STREAM_VERDICT:
1944                 ptype = BPF_PROG_TYPE_SK_SKB;
1945                 break;
1946         case BPF_LIRC_MODE2:
1947                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1948                 break;
1949         case BPF_FLOW_DISSECTOR:
1950                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1951                 break;
1952         case BPF_CGROUP_SYSCTL:
1953                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1954                 break;
1955         case BPF_CGROUP_GETSOCKOPT:
1956         case BPF_CGROUP_SETSOCKOPT:
1957                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
1958                 break;
1959         default:
1960                 return -EINVAL;
1961         }
1962 
1963         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1964         if (IS_ERR(prog))
1965                 return PTR_ERR(prog);
1966 
1967         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1968                 bpf_prog_put(prog);
1969                 return -EINVAL;
1970         }
1971 
1972         switch (ptype) {
1973         case BPF_PROG_TYPE_SK_SKB:
1974         case BPF_PROG_TYPE_SK_MSG:
1975                 ret = sock_map_get_from_fd(attr, prog);
1976                 break;
1977         case BPF_PROG_TYPE_LIRC_MODE2:
1978                 ret = lirc_prog_attach(attr, prog);
1979                 break;
1980         case BPF_PROG_TYPE_FLOW_DISSECTOR:
1981                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1982                 break;
1983         default:
1984                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1985         }
1986 
1987         if (ret)
1988                 bpf_prog_put(prog);
1989         return ret;
1990 }
1991 
1992 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1993 
1994 static int bpf_prog_detach(const union bpf_attr *attr)
1995 {
1996         enum bpf_prog_type ptype;
1997 
1998         if (!capable(CAP_NET_ADMIN))
1999                 return -EPERM;
2000 
2001         if (CHECK_ATTR(BPF_PROG_DETACH))
2002                 return -EINVAL;
2003 
2004         switch (attr->attach_type) {
2005         case BPF_CGROUP_INET_INGRESS:
2006         case BPF_CGROUP_INET_EGRESS:
2007                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
2008                 break;
2009         case BPF_CGROUP_INET_SOCK_CREATE:
2010         case BPF_CGROUP_INET4_POST_BIND:
2011         case BPF_CGROUP_INET6_POST_BIND:
2012                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2013                 break;
2014         case BPF_CGROUP_INET4_BIND:
2015         case BPF_CGROUP_INET6_BIND:
2016         case BPF_CGROUP_INET4_CONNECT:
2017         case BPF_CGROUP_INET6_CONNECT:
2018         case BPF_CGROUP_UDP4_SENDMSG:
2019         case BPF_CGROUP_UDP6_SENDMSG:
2020         case BPF_CGROUP_UDP4_RECVMSG:
2021         case BPF_CGROUP_UDP6_RECVMSG:
2022                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2023                 break;
2024         case BPF_CGROUP_SOCK_OPS:
2025                 ptype = BPF_PROG_TYPE_SOCK_OPS;
2026                 break;
2027         case BPF_CGROUP_DEVICE:
2028                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2029                 break;
2030         case BPF_SK_MSG_VERDICT:
2031                 return sock_map_get_from_fd(attr, NULL);
2032         case BPF_SK_SKB_STREAM_PARSER:
2033         case BPF_SK_SKB_STREAM_VERDICT:
2034                 return sock_map_get_from_fd(attr, NULL);
2035         case BPF_LIRC_MODE2:
2036                 return lirc_prog_detach(attr);
2037         case BPF_FLOW_DISSECTOR:
2038                 return skb_flow_dissector_bpf_prog_detach(attr);
2039         case BPF_CGROUP_SYSCTL:
2040                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2041                 break;
2042         case BPF_CGROUP_GETSOCKOPT:
2043         case BPF_CGROUP_SETSOCKOPT:
2044                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2045                 break;
2046         default:
2047                 return -EINVAL;
2048         }
2049 
2050         return cgroup_bpf_prog_detach(attr, ptype);
2051 }
2052 
2053 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2054 
2055 static int bpf_prog_query(const union bpf_attr *attr,
2056                           union bpf_attr __user *uattr)
2057 {
2058         if (!capable(CAP_NET_ADMIN))
2059                 return -EPERM;
2060         if (CHECK_ATTR(BPF_PROG_QUERY))
2061                 return -EINVAL;
2062         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2063                 return -EINVAL;
2064 
2065         switch (attr->query.attach_type) {
2066         case BPF_CGROUP_INET_INGRESS:
2067         case BPF_CGROUP_INET_EGRESS:
2068         case BPF_CGROUP_INET_SOCK_CREATE:
2069         case BPF_CGROUP_INET4_BIND:
2070         case BPF_CGROUP_INET6_BIND:
2071         case BPF_CGROUP_INET4_POST_BIND:
2072         case BPF_CGROUP_INET6_POST_BIND:
2073         case BPF_CGROUP_INET4_CONNECT:
2074         case BPF_CGROUP_INET6_CONNECT:
2075         case BPF_CGROUP_UDP4_SENDMSG:
2076         case BPF_CGROUP_UDP6_SENDMSG:
2077         case BPF_CGROUP_UDP4_RECVMSG:
2078         case BPF_CGROUP_UDP6_RECVMSG:
2079         case BPF_CGROUP_SOCK_OPS:
2080         case BPF_CGROUP_DEVICE:
2081         case BPF_CGROUP_SYSCTL:
2082         case BPF_CGROUP_GETSOCKOPT:
2083         case BPF_CGROUP_SETSOCKOPT:
2084                 break;
2085         case BPF_LIRC_MODE2:
2086                 return lirc_prog_query(attr, uattr);
2087         case BPF_FLOW_DISSECTOR:
2088                 return skb_flow_dissector_prog_query(attr, uattr);
2089         default:
2090                 return -EINVAL;
2091         }
2092 
2093         return cgroup_bpf_prog_query(attr, uattr);
2094 }
2095 
2096 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2097 
2098 static int bpf_prog_test_run(const union bpf_attr *attr,
2099                              union bpf_attr __user *uattr)
2100 {
2101         struct bpf_prog *prog;
2102         int ret = -ENOTSUPP;
2103 
2104         if (!capable(CAP_SYS_ADMIN))
2105                 return -EPERM;
2106         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2107                 return -EINVAL;
2108 
2109         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2110             (!attr->test.ctx_size_in && attr->test.ctx_in))
2111                 return -EINVAL;
2112 
2113         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2114             (!attr->test.ctx_size_out && attr->test.ctx_out))
2115                 return -EINVAL;
2116 
2117         prog = bpf_prog_get(attr->test.prog_fd);
2118         if (IS_ERR(prog))
2119                 return PTR_ERR(prog);
2120 
2121         if (prog->aux->ops->test_run)
2122                 ret = prog->aux->ops->test_run(prog, attr, uattr);
2123 
2124         bpf_prog_put(prog);
2125         return ret;
2126 }
2127 
2128 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2129 
2130 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2131                                union bpf_attr __user *uattr,
2132                                struct idr *idr,
2133                                spinlock_t *lock)
2134 {
2135         u32 next_id = attr->start_id;
2136         int err = 0;
2137 
2138         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2139                 return -EINVAL;
2140 
2141         if (!capable(CAP_SYS_ADMIN))
2142                 return -EPERM;
2143 
2144         next_id++;
2145         spin_lock_bh(lock);
2146         if (!idr_get_next(idr, &next_id))
2147                 err = -ENOENT;
2148         spin_unlock_bh(lock);
2149 
2150         if (!err)
2151                 err = put_user(next_id, &uattr->next_id);
2152 
2153         return err;
2154 }
2155 
2156 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2157 
2158 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2159 {
2160         struct bpf_prog *prog;
2161         u32 id = attr->prog_id;
2162         int fd;
2163 
2164         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2165                 return -EINVAL;
2166 
2167         if (!capable(CAP_SYS_ADMIN))
2168                 return -EPERM;
2169 
2170         spin_lock_bh(&prog_idr_lock);
2171         prog = idr_find(&prog_idr, id);
2172         if (prog)
2173                 prog = bpf_prog_inc_not_zero(prog);
2174         else
2175                 prog = ERR_PTR(-ENOENT);
2176         spin_unlock_bh(&prog_idr_lock);
2177 
2178         if (IS_ERR(prog))
2179                 return PTR_ERR(prog);
2180 
2181         fd = bpf_prog_new_fd(prog);
2182         if (fd < 0)
2183                 bpf_prog_put(prog);
2184 
2185         return fd;
2186 }
2187 
2188 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2189 
2190 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2191 {
2192         struct bpf_map *map;
2193         u32 id = attr->map_id;
2194         int f_flags;
2195         int fd;
2196 
2197         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2198             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2199                 return -EINVAL;
2200 
2201         if (!capable(CAP_SYS_ADMIN))
2202                 return -EPERM;
2203 
2204         f_flags = bpf_get_file_flag(attr->open_flags);
2205         if (f_flags < 0)
2206                 return f_flags;
2207 
2208         spin_lock_bh(&map_idr_lock);
2209         map = idr_find(&map_idr, id);
2210         if (map)
2211                 map = __bpf_map_inc_not_zero(map, true);
2212         else
2213                 map = ERR_PTR(-ENOENT);
2214         spin_unlock_bh(&map_idr_lock);
2215 
2216         if (IS_ERR(map))
2217                 return PTR_ERR(map);
2218 
2219         fd = bpf_map_new_fd(map, f_flags);
2220         if (fd < 0)
2221                 bpf_map_put_with_uref(map);
2222 
2223         return fd;
2224 }
2225 
2226 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2227                                               unsigned long addr, u32 *off,
2228                                               u32 *type)
2229 {
2230         const struct bpf_map *map;
2231         int i;
2232 
2233         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2234                 map = prog->aux->used_maps[i];
2235                 if (map == (void *)addr) {
2236                         *type = BPF_PSEUDO_MAP_FD;
2237                         return map;
2238                 }
2239                 if (!map->ops->map_direct_value_meta)
2240                         continue;
2241                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2242                         *type = BPF_PSEUDO_MAP_VALUE;
2243                         return map;
2244                 }
2245         }
2246 
2247         return NULL;
2248 }
2249 
2250 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2251 {
2252         const struct bpf_map *map;
2253         struct bpf_insn *insns;
2254         u32 off, type;
2255         u64 imm;
2256         int i;
2257 
2258         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2259                         GFP_USER);
2260         if (!insns)
2261                 return insns;
2262 
2263         for (i = 0; i < prog->len; i++) {
2264                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2265                         insns[i].code = BPF_JMP | BPF_CALL;
2266                         insns[i].imm = BPF_FUNC_tail_call;
2267                         /* fall-through */
2268                 }
2269                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2270                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2271                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2272                                 insns[i].code = BPF_JMP | BPF_CALL;
2273                         if (!bpf_dump_raw_ok())
2274                                 insns[i].imm = 0;
2275                         continue;
2276                 }
2277 
2278                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2279                         continue;
2280 
2281                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2282                 map = bpf_map_from_imm(prog, imm, &off, &type);
2283                 if (map) {
2284                         insns[i].src_reg = type;
2285                         insns[i].imm = map->id;
2286                         insns[i + 1].imm = off;
2287                         continue;
2288                 }
2289         }
2290 
2291         return insns;
2292 }
2293 
2294 static int set_info_rec_size(struct bpf_prog_info *info)
2295 {
2296         /*
2297          * Ensure info.*_rec_size is the same as kernel expected size
2298          *
2299          * or
2300          *
2301          * Only allow zero *_rec_size if both _rec_size and _cnt are
2302          * zero.  In this case, the kernel will set the expected
2303          * _rec_size back to the info.
2304          */
2305 
2306         if ((info->nr_func_info || info->func_info_rec_size) &&
2307             info->func_info_rec_size != sizeof(struct bpf_func_info))
2308                 return -EINVAL;
2309 
2310         if ((info->nr_line_info || info->line_info_rec_size) &&
2311             info->line_info_rec_size != sizeof(struct bpf_line_info))
2312                 return -EINVAL;
2313 
2314         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2315             info->jited_line_info_rec_size != sizeof(__u64))
2316                 return -EINVAL;
2317 
2318         info->func_info_rec_size = sizeof(struct bpf_func_info);
2319         info->line_info_rec_size = sizeof(struct bpf_line_info);
2320         info->jited_line_info_rec_size = sizeof(__u64);
2321 
2322         return 0;
2323 }
2324 
2325 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2326                                    const union bpf_attr *attr,
2327                                    union bpf_attr __user *uattr)
2328 {
2329         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2330         struct bpf_prog_info info;
2331         u32 info_len = attr->info.info_len;
2332         struct bpf_prog_stats stats;
2333         char __user *uinsns;
2334         u32 ulen;
2335         int err;
2336 
2337         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2338         if (err)
2339                 return err;
2340         info_len = min_t(u32, sizeof(info), info_len);
2341 
2342         memset(&info, 0, sizeof(info));
2343         if (copy_from_user(&info, uinfo, info_len))
2344                 return -EFAULT;
2345 
2346         info.type = prog->type;
2347         info.id = prog->aux->id;
2348         info.load_time = prog->aux->load_time;
2349         info.created_by_uid = from_kuid_munged(current_user_ns(),
2350                                                prog->aux->user->uid);
2351         info.gpl_compatible = prog->gpl_compatible;
2352 
2353         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2354         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2355 
2356         ulen = info.nr_map_ids;
2357         info.nr_map_ids = prog->aux->used_map_cnt;
2358         ulen = min_t(u32, info.nr_map_ids, ulen);
2359         if (ulen) {
2360                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2361                 u32 i;
2362 
2363                 for (i = 0; i < ulen; i++)
2364                         if (put_user(prog->aux->used_maps[i]->id,
2365                                      &user_map_ids[i]))
2366                                 return -EFAULT;
2367         }
2368 
2369         err = set_info_rec_size(&info);
2370         if (err)
2371                 return err;
2372 
2373         bpf_prog_get_stats(prog, &stats);
2374         info.run_time_ns = stats.nsecs;
2375         info.run_cnt = stats.cnt;
2376 
2377         if (!capable(CAP_SYS_ADMIN)) {
2378                 info.jited_prog_len = 0;
2379                 info.xlated_prog_len = 0;
2380                 info.nr_jited_ksyms = 0;
2381                 info.nr_jited_func_lens = 0;
2382                 info.nr_func_info = 0;
2383                 info.nr_line_info = 0;
2384                 info.nr_jited_line_info = 0;
2385                 goto done;
2386         }
2387 
2388         ulen = info.xlated_prog_len;
2389         info.xlated_prog_len = bpf_prog_insn_size(prog);
2390         if (info.xlated_prog_len && ulen) {
2391                 struct bpf_insn *insns_sanitized;
2392                 bool fault;
2393 
2394                 if (prog->blinded && !bpf_dump_raw_ok()) {
2395                         info.xlated_prog_insns = 0;
2396                         goto done;
2397                 }
2398                 insns_sanitized = bpf_insn_prepare_dump(prog);
2399                 if (!insns_sanitized)
2400                         return -ENOMEM;
2401                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2402                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2403                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2404                 kfree(insns_sanitized);
2405                 if (fault)
2406                         return -EFAULT;
2407         }
2408 
2409         if (bpf_prog_is_dev_bound(prog->aux)) {
2410                 err = bpf_prog_offload_info_fill(&info, prog);
2411                 if (err)
2412                         return err;
2413                 goto done;
2414         }
2415 
2416         /* NOTE: the following code is supposed to be skipped for offload.
2417          * bpf_prog_offload_info_fill() is the place to fill similar fields
2418          * for offload.
2419          */
2420         ulen = info.jited_prog_len;
2421         if (prog->aux->func_cnt) {
2422                 u32 i;
2423 
2424                 info.jited_prog_len = 0;
2425                 for (i = 0; i < prog->aux->func_cnt; i++)
2426                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2427         } else {
2428                 info.jited_prog_len = prog->jited_len;
2429         }
2430 
2431         if (info.jited_prog_len && ulen) {
2432                 if (bpf_dump_raw_ok()) {
2433                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2434                         ulen = min_t(u32, info.jited_prog_len, ulen);
2435 
2436                         /* for multi-function programs, copy the JITed
2437                          * instructions for all the functions
2438                          */
2439                         if (prog->aux->func_cnt) {
2440                                 u32 len, free, i;
2441                                 u8 *img;
2442 
2443                                 free = ulen;
2444                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2445                                         len = prog->aux->func[i]->jited_len;
2446                                         len = min_t(u32, len, free);
2447                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2448                                         if (copy_to_user(uinsns, img, len))
2449                                                 return -EFAULT;
2450                                         uinsns += len;
2451                                         free -= len;
2452                                         if (!free)
2453                                                 break;
2454                                 }
2455                         } else {
2456                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2457                                         return -EFAULT;
2458                         }
2459                 } else {
2460                         info.jited_prog_insns = 0;
2461                 }
2462         }
2463 
2464         ulen = info.nr_jited_ksyms;
2465         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2466         if (ulen) {
2467                 if (bpf_dump_raw_ok()) {
2468                         unsigned long ksym_addr;
2469                         u64 __user *user_ksyms;
2470                         u32 i;
2471 
2472                         /* copy the address of the kernel symbol
2473                          * corresponding to each function
2474                          */
2475                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2476                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2477                         if (prog->aux->func_cnt) {
2478                                 for (i = 0; i < ulen; i++) {
2479                                         ksym_addr = (unsigned long)
2480                                                 prog->aux->func[i]->bpf_func;
2481                                         if (put_user((u64) ksym_addr,
2482                                                      &user_ksyms[i]))
2483                                                 return -EFAULT;
2484                                 }
2485                         } else {
2486                                 ksym_addr = (unsigned long) prog->bpf_func;
2487                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2488                                         return -EFAULT;
2489                         }
2490                 } else {
2491                         info.jited_ksyms = 0;
2492                 }
2493         }
2494 
2495         ulen = info.nr_jited_func_lens;
2496         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2497         if (ulen) {
2498                 if (bpf_dump_raw_ok()) {
2499                         u32 __user *user_lens;
2500                         u32 func_len, i;
2501 
2502                         /* copy the JITed image lengths for each function */
2503                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2504                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2505                         if (prog->aux->func_cnt) {
2506                                 for (i = 0; i < ulen; i++) {
2507                                         func_len =
2508                                                 prog->aux->func[i]->jited_len;
2509                                         if (put_user(func_len, &user_lens[i]))
2510                                                 return -EFAULT;
2511                                 }
2512                         } else {
2513                                 func_len = prog->jited_len;
2514                                 if (put_user(func_len, &user_lens[0]))
2515                                         return -EFAULT;
2516                         }
2517                 } else {
2518                         info.jited_func_lens = 0;
2519                 }
2520         }
2521 
2522         if (prog->aux->btf)
2523                 info.btf_id = btf_id(prog->aux->btf);
2524 
2525         ulen = info.nr_func_info;
2526         info.nr_func_info = prog->aux->func_info_cnt;
2527         if (info.nr_func_info && ulen) {
2528                 char __user *user_finfo;
2529 
2530                 user_finfo = u64_to_user_ptr(info.func_info);
2531                 ulen = min_t(u32, info.nr_func_info, ulen);
2532                 if (copy_to_user(user_finfo, prog->aux->func_info,
2533                                  info.func_info_rec_size * ulen))
2534                         return -EFAULT;
2535         }
2536 
2537         ulen = info.nr_line_info;
2538         info.nr_line_info = prog->aux->nr_linfo;
2539         if (info.nr_line_info && ulen) {
2540                 __u8 __user *user_linfo;
2541 
2542                 user_linfo = u64_to_user_ptr(info.line_info);
2543                 ulen = min_t(u32, info.nr_line_info, ulen);
2544                 if (copy_to_user(user_linfo, prog->aux->linfo,
2545                                  info.line_info_rec_size * ulen))
2546                         return -EFAULT;
2547         }
2548 
2549         ulen = info.nr_jited_line_info;
2550         if (prog->aux->jited_linfo)
2551                 info.nr_jited_line_info = prog->aux->nr_linfo;
2552         else
2553                 info.nr_jited_line_info = 0;
2554         if (info.nr_jited_line_info && ulen) {
2555                 if (bpf_dump_raw_ok()) {
2556                         __u64 __user *user_linfo;
2557                         u32 i;
2558 
2559                         user_linfo = u64_to_user_ptr(info.jited_line_info);
2560                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
2561                         for (i = 0; i < ulen; i++) {
2562                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2563                                              &user_linfo[i]))
2564                                         return -EFAULT;
2565                         }
2566                 } else {
2567                         info.jited_line_info = 0;
2568                 }
2569         }
2570 
2571         ulen = info.nr_prog_tags;
2572         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2573         if (ulen) {
2574                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2575                 u32 i;
2576 
2577                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2578                 ulen = min_t(u32, info.nr_prog_tags, ulen);
2579                 if (prog->aux->func_cnt) {
2580                         for (i = 0; i < ulen; i++) {
2581                                 if (copy_to_user(user_prog_tags[i],
2582                                                  prog->aux->func[i]->tag,
2583                                                  BPF_TAG_SIZE))
2584                                         return -EFAULT;
2585                         }
2586                 } else {
2587                         if (copy_to_user(user_prog_tags[0],
2588                                          prog->tag, BPF_TAG_SIZE))
2589                                 return -EFAULT;
2590                 }
2591         }
2592 
2593 done:
2594         if (copy_to_user(uinfo, &info, info_len) ||
2595             put_user(info_len, &uattr->info.info_len))
2596                 return -EFAULT;
2597 
2598         return 0;
2599 }
2600 
2601 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2602                                   const union bpf_attr *attr,
2603                                   union bpf_attr __user *uattr)
2604 {
2605         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2606         struct bpf_map_info info;
2607         u32 info_len = attr->info.info_len;
2608         int err;
2609 
2610         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2611         if (err)
2612                 return err;
2613         info_len = min_t(u32, sizeof(info), info_len);
2614 
2615         memset(&info, 0, sizeof(info));
2616         info.type = map->map_type;
2617         info.id = map->id;
2618         info.key_size = map->key_size;
2619         info.value_size = map->value_size;
2620         info.max_entries = map->max_entries;
2621         info.map_flags = map->map_flags;
2622         memcpy(info.name, map->name, sizeof(map->name));
2623 
2624         if (map->btf) {
2625                 info.btf_id = btf_id(map->btf);
2626                 info.btf_key_type_id = map->btf_key_type_id;
2627                 info.btf_value_type_id = map->btf_value_type_id;
2628         }
2629 
2630         if (bpf_map_is_dev_bound(map)) {
2631                 err = bpf_map_offload_info_fill(&info, map);
2632                 if (err)
2633                         return err;
2634         }
2635 
2636         if (copy_to_user(uinfo, &info, info_len) ||
2637             put_user(info_len, &uattr->info.info_len))
2638                 return -EFAULT;
2639 
2640         return 0;
2641 }
2642 
2643 static int bpf_btf_get_info_by_fd(struct btf *btf,
2644                                   const union bpf_attr *attr,
2645                                   union bpf_attr __user *uattr)
2646 {
2647         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2648         u32 info_len = attr->info.info_len;
2649         int err;
2650 
2651         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2652         if (err)
2653                 return err;
2654 
2655         return btf_get_info_by_fd(btf, attr, uattr);
2656 }
2657 
2658 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2659 
2660 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2661                                   union bpf_attr __user *uattr)
2662 {
2663         int ufd = attr->info.bpf_fd;
2664         struct fd f;
2665         int err;
2666 
2667         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2668                 return -EINVAL;
2669 
2670         f = fdget(ufd);
2671         if (!f.file)
2672                 return -EBADFD;
2673 
2674         if (f.file->f_op == &bpf_prog_fops)
2675                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2676                                               uattr);
2677         else if (f.file->f_op == &bpf_map_fops)
2678                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2679                                              uattr);
2680         else if (f.file->f_op == &btf_fops)
2681                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2682         else
2683                 err = -EINVAL;
2684 
2685         fdput(f);
2686         return err;
2687 }
2688 
2689 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2690 
2691 static int bpf_btf_load(const union bpf_attr *attr)
2692 {
2693         if (CHECK_ATTR(BPF_BTF_LOAD))
2694                 return -EINVAL;
2695 
2696         if (!capable(CAP_SYS_ADMIN))
2697                 return -EPERM;
2698 
2699         return btf_new_fd(attr);
2700 }
2701 
2702 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2703 
2704 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2705 {
2706         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2707                 return -EINVAL;
2708 
2709         if (!capable(CAP_SYS_ADMIN))
2710                 return -EPERM;
2711 
2712         return btf_get_fd_by_id(attr->btf_id);
2713 }
2714 
2715 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2716                                     union bpf_attr __user *uattr,
2717                                     u32 prog_id, u32 fd_type,
2718                                     const char *buf, u64 probe_offset,
2719                                     u64 probe_addr)
2720 {
2721         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2722         u32 len = buf ? strlen(buf) : 0, input_len;
2723         int err = 0;
2724 
2725         if (put_user(len, &uattr->task_fd_query.buf_len))
2726                 return -EFAULT;
2727         input_len = attr->task_fd_query.buf_len;
2728         if (input_len && ubuf) {
2729                 if (!len) {
2730                         /* nothing to copy, just make ubuf NULL terminated */
2731                         char zero = '\0';
2732 
2733                         if (put_user(zero, ubuf))
2734                                 return -EFAULT;
2735                 } else if (input_len >= len + 1) {
2736                         /* ubuf can hold the string with NULL terminator */
2737                         if (copy_to_user(ubuf, buf, len + 1))
2738                                 return -EFAULT;
2739                 } else {
2740                         /* ubuf cannot hold the string with NULL terminator,
2741                          * do a partial copy with NULL terminator.
2742                          */
2743                         char zero = '\0';
2744 
2745                         err = -ENOSPC;
2746                         if (copy_to_user(ubuf, buf, input_len - 1))
2747                                 return -EFAULT;
2748                         if (put_user(zero, ubuf + input_len - 1))
2749                                 return -EFAULT;
2750                 }
2751         }
2752 
2753         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2754             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2755             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2756             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2757                 return -EFAULT;
2758 
2759         return err;
2760 }
2761 
2762 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2763 
2764 static int bpf_task_fd_query(const union bpf_attr *attr,
2765                              union bpf_attr __user *uattr)
2766 {
2767         pid_t pid = attr->task_fd_query.pid;
2768         u32 fd = attr->task_fd_query.fd;
2769         const struct perf_event *event;
2770         struct files_struct *files;
2771         struct task_struct *task;
2772         struct file *file;
2773         int err;
2774 
2775         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2776                 return -EINVAL;
2777 
2778         if (!capable(CAP_SYS_ADMIN))
2779                 return -EPERM;
2780 
2781         if (attr->task_fd_query.flags != 0)
2782                 return -EINVAL;
2783 
2784         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2785         if (!task)
2786                 return -ENOENT;
2787 
2788         files = get_files_struct(task);
2789         put_task_struct(task);
2790         if (!files)
2791                 return -ENOENT;
2792 
2793         err = 0;
2794         spin_lock(&files->file_lock);
2795         file = fcheck_files(files, fd);
2796         if (!file)
2797                 err = -EBADF;
2798         else
2799                 get_file(file);
2800         spin_unlock(&files->file_lock);
2801         put_files_struct(files);
2802 
2803         if (err)
2804                 goto out;
2805 
2806         if (file->f_op == &bpf_raw_tp_fops) {
2807                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2808                 struct bpf_raw_event_map *btp = raw_tp->btp;
2809 
2810                 err = bpf_task_fd_query_copy(attr, uattr,
2811                                              raw_tp->prog->aux->id,
2812                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2813                                              btp->tp->name, 0, 0);
2814                 goto put_file;
2815         }
2816 
2817         event = perf_get_event(file);
2818         if (!IS_ERR(event)) {
2819                 u64 probe_offset, probe_addr;
2820                 u32 prog_id, fd_type;
2821                 const char *buf;
2822 
2823                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2824                                               &buf, &probe_offset,
2825                                               &probe_addr);
2826                 if (!err)
2827                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2828                                                      fd_type, buf,
2829                                                      probe_offset,
2830                                                      probe_addr);
2831                 goto put_file;
2832         }
2833 
2834         err = -ENOTSUPP;
2835 put_file:
2836         fput(file);
2837 out:
2838         return err;
2839 }
2840 
2841 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2842 {
2843         union bpf_attr attr;
2844         int err;
2845 
2846         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2847                 return -EPERM;
2848 
2849         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2850         if (err)
2851                 return err;
2852         size = min_t(u32, size, sizeof(attr));
2853 
2854         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2855         memset(&attr, 0, sizeof(attr));
2856         if (copy_from_user(&attr, uattr, size) != 0)
2857                 return -EFAULT;
2858 
2859         err = security_bpf(cmd, &attr, size);
2860         if (err < 0)
2861                 return err;
2862 
2863         switch (cmd) {
2864         case BPF_MAP_CREATE:
2865                 err = map_create(&attr);
2866                 break;
2867         case BPF_MAP_LOOKUP_ELEM:
2868                 err = map_lookup_elem(&attr);
2869                 break;
2870         case BPF_MAP_UPDATE_ELEM:
2871                 err = map_update_elem(&attr);
2872                 break;
2873         case BPF_MAP_DELETE_ELEM:
2874                 err = map_delete_elem(&attr);
2875                 break;
2876         case BPF_MAP_GET_NEXT_KEY:
2877                 err = map_get_next_key(&attr);
2878                 break;
2879         case BPF_MAP_FREEZE:
2880                 err = map_freeze(&attr);
2881                 break;
2882         case BPF_PROG_LOAD:
2883                 err = bpf_prog_load(&attr, uattr);
2884                 break;
2885         case BPF_OBJ_PIN:
2886                 err = bpf_obj_pin(&attr);
2887                 break;
2888         case BPF_OBJ_GET:
2889                 err = bpf_obj_get(&attr);
2890                 break;
2891         case BPF_PROG_ATTACH:
2892                 err = bpf_prog_attach(&attr);
2893                 break;
2894         case BPF_PROG_DETACH:
2895                 err = bpf_prog_detach(&attr);
2896                 break;
2897         case BPF_PROG_QUERY:
2898                 err = bpf_prog_query(&attr, uattr);
2899                 break;
2900         case BPF_PROG_TEST_RUN:
2901                 err = bpf_prog_test_run(&attr, uattr);
2902                 break;
2903         case BPF_PROG_GET_NEXT_ID:
2904                 err = bpf_obj_get_next_id(&attr, uattr,
2905                                           &prog_idr, &prog_idr_lock);
2906                 break;
2907         case BPF_MAP_GET_NEXT_ID:
2908                 err = bpf_obj_get_next_id(&attr, uattr,
2909                                           &map_idr, &map_idr_lock);
2910                 break;
2911         case BPF_BTF_GET_NEXT_ID:
2912                 err = bpf_obj_get_next_id(&attr, uattr,
2913                                           &btf_idr, &btf_idr_lock);
2914                 break;
2915         case BPF_PROG_GET_FD_BY_ID:
2916                 err = bpf_prog_get_fd_by_id(&attr);
2917                 break;
2918         case BPF_MAP_GET_FD_BY_ID:
2919                 err = bpf_map_get_fd_by_id(&attr);
2920                 break;
2921         case BPF_OBJ_GET_INFO_BY_FD:
2922                 err = bpf_obj_get_info_by_fd(&attr, uattr);
2923                 break;
2924         case BPF_RAW_TRACEPOINT_OPEN:
2925                 err = bpf_raw_tracepoint_open(&attr);
2926                 break;
2927         case BPF_BTF_LOAD:
2928                 err = bpf_btf_load(&attr);
2929                 break;
2930         case BPF_BTF_GET_FD_BY_ID:
2931                 err = bpf_btf_get_fd_by_id(&attr);
2932                 break;
2933         case BPF_TASK_FD_QUERY:
2934                 err = bpf_task_fd_query(&attr, uattr);
2935                 break;
2936         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2937                 err = map_lookup_and_delete_elem(&attr);
2938                 break;
2939         default:
2940                 err = -EINVAL;
2941                 break;
2942         }
2943 
2944         return err;
2945 }
/* [<][>][^][v][top][bottom][index][help] */
root/kernel/bpf/syscall.c

DEFINITIONS
