1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of version 2 of the GNU General Public
5 * License as published by the Free Software Foundation.
7 * This program is distributed in the hope that it will be useful, but
8 * WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
10 * General Public License for more details.
12 #include <linux/bpf.h>
13 #include <linux/bpf_trace.h>
14 #include <linux/bpf_lirc.h>
15 #include <linux/btf.h>
16 #include <linux/syscalls.h>
17 #include <linux/slab.h>
18 #include <linux/sched/signal.h>
19 #include <linux/vmalloc.h>
20 #include <linux/mmzone.h>
21 #include <linux/anon_inodes.h>
22 #include <linux/fdtable.h>
23 #include <linux/file.h>
25 #include <linux/license.h>
26 #include <linux/filter.h>
27 #include <linux/version.h>
28 #include <linux/kernel.h>
29 #include <linux/idr.h>
30 #include <linux/cred.h>
31 #include <linux/timekeeping.h>
32 #include <linux/ctype.h>
33 #include <linux/nospec.h>
35 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
36 (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
37 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
38 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
39 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
40 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
42 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
44 DEFINE_PER_CPU(int, bpf_prog_active);
45 static DEFINE_IDR(prog_idr);
46 static DEFINE_SPINLOCK(prog_idr_lock);
47 static DEFINE_IDR(map_idr);
48 static DEFINE_SPINLOCK(map_idr_lock);
50 int sysctl_unprivileged_bpf_disabled __read_mostly;
52 static const struct bpf_map_ops * const bpf_map_types[] = {
53 #define BPF_PROG_TYPE(_id, _ops)
54 #define BPF_MAP_TYPE(_id, _ops) \
56 #include <linux/bpf_types.h>
62 * If we're handed a bigger struct than we know of, ensure all the unknown bits
63 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
64 * we don't know about yet.
66 * There is a ToCToU between this function call and the following
67 * copy_from_user() call. However, this is not a concern since this function is
68 * meant to be a future-proofing of bits.
70 int bpf_check_uarg_tail_zero(void __user *uaddr,
74 unsigned char __user *addr;
75 unsigned char __user *end;
79 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
82 if (unlikely(!access_ok(uaddr, actual_size)))
85 if (actual_size <= expected_size)
88 addr = uaddr + expected_size;
89 end = uaddr + actual_size;
91 for (; addr < end; addr++) {
92 err = get_user(val, addr);
102 const struct bpf_map_ops bpf_map_offload_ops = {
103 .map_alloc = bpf_map_offload_map_alloc,
104 .map_free = bpf_map_offload_map_free,
105 .map_check_btf = map_check_no_btf,
108 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
110 const struct bpf_map_ops *ops;
111 u32 type = attr->map_type;
115 if (type >= ARRAY_SIZE(bpf_map_types))
116 return ERR_PTR(-EINVAL);
117 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
118 ops = bpf_map_types[type];
120 return ERR_PTR(-EINVAL);
122 if (ops->map_alloc_check) {
123 err = ops->map_alloc_check(attr);
127 if (attr->map_ifindex)
128 ops = &bpf_map_offload_ops;
129 map = ops->map_alloc(attr);
133 map->map_type = type;
137 void *bpf_map_area_alloc(size_t size, int numa_node)
139 /* We really just want to fail instead of triggering OOM killer
140 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
141 * which is used for lower order allocation requests.
143 * It has been observed that higher order allocation requests done by
144 * vmalloc with __GFP_NORETRY being set might fail due to not trying
145 * to reclaim memory from the page cache, thus we set
146 * __GFP_RETRY_MAYFAIL to avoid such situations.
149 const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
152 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
153 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
159 return __vmalloc_node_flags_caller(size, numa_node,
160 GFP_KERNEL | __GFP_RETRY_MAYFAIL |
161 flags, __builtin_return_address(0));
164 void bpf_map_area_free(void *area)
169 static u32 bpf_map_flags_retain_permanent(u32 flags)
171 /* Some map creation flags are not tied to the map object but
172 * rather to the map fd instead, so they have no meaning upon
173 * map object inspection since multiple file descriptors with
174 * different (access) properties can exist here. Thus, given
175 * this has zero meaning for the map itself, lets clear these
178 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
181 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
183 map->map_type = attr->map_type;
184 map->key_size = attr->key_size;
185 map->value_size = attr->value_size;
186 map->max_entries = attr->max_entries;
187 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
188 map->numa_node = bpf_map_attr_numa_node(attr);
191 int bpf_map_precharge_memlock(u32 pages)
193 struct user_struct *user = get_current_user();
194 unsigned long memlock_limit, cur;
196 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
197 cur = atomic_long_read(&user->locked_vm);
199 if (cur + pages > memlock_limit)
204 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
206 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
208 if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
209 atomic_long_sub(pages, &user->locked_vm);
215 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
217 atomic_long_sub(pages, &user->locked_vm);
220 static int bpf_map_init_memlock(struct bpf_map *map)
222 struct user_struct *user = get_current_user();
225 ret = bpf_charge_memlock(user, map->pages);
234 static void bpf_map_release_memlock(struct bpf_map *map)
236 struct user_struct *user = map->user;
237 bpf_uncharge_memlock(user, map->pages);
241 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
245 ret = bpf_charge_memlock(map->user, pages);
252 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
254 bpf_uncharge_memlock(map->user, pages);
258 static int bpf_map_alloc_id(struct bpf_map *map)
262 idr_preload(GFP_KERNEL);
263 spin_lock_bh(&map_idr_lock);
264 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
267 spin_unlock_bh(&map_idr_lock);
270 if (WARN_ON_ONCE(!id))
273 return id > 0 ? 0 : id;
276 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
280 /* Offloaded maps are removed from the IDR store when their device
281 * disappears - even if someone holds an fd to them they are unusable,
282 * the memory is gone, all ops will fail; they are simply waiting for
283 * refcnt to drop to be freed.
289 spin_lock_irqsave(&map_idr_lock, flags);
291 __acquire(&map_idr_lock);
293 idr_remove(&map_idr, map->id);
297 spin_unlock_irqrestore(&map_idr_lock, flags);
299 __release(&map_idr_lock);
302 /* called from workqueue */
303 static void bpf_map_free_deferred(struct work_struct *work)
305 struct bpf_map *map = container_of(work, struct bpf_map, work);
307 bpf_map_release_memlock(map);
308 security_bpf_map_free(map);
309 /* implementation dependent freeing */
310 map->ops->map_free(map);
313 static void bpf_map_put_uref(struct bpf_map *map)
315 if (atomic_dec_and_test(&map->usercnt)) {
316 if (map->ops->map_release_uref)
317 map->ops->map_release_uref(map);
321 /* decrement map refcnt and schedule it for freeing via workqueue
322 * (unrelying map implementation ops->map_free() might sleep)
324 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
326 if (atomic_dec_and_test(&map->refcnt)) {
327 /* bpf_map_free_id() must be called first */
328 bpf_map_free_id(map, do_idr_lock);
330 INIT_WORK(&map->work, bpf_map_free_deferred);
331 schedule_work(&map->work);
335 void bpf_map_put(struct bpf_map *map)
337 __bpf_map_put(map, true);
339 EXPORT_SYMBOL_GPL(bpf_map_put);
341 void bpf_map_put_with_uref(struct bpf_map *map)
343 bpf_map_put_uref(map);
347 static int bpf_map_release(struct inode *inode, struct file *filp)
349 struct bpf_map *map = filp->private_data;
351 if (map->ops->map_release)
352 map->ops->map_release(map, filp);
354 bpf_map_put_with_uref(map);
358 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
360 fmode_t mode = f.file->f_mode;
362 /* Our file permissions may have been overridden by global
363 * map permissions facing syscall side.
365 if (READ_ONCE(map->frozen))
366 mode &= ~FMODE_CAN_WRITE;
370 #ifdef CONFIG_PROC_FS
371 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
373 const struct bpf_map *map = filp->private_data;
374 const struct bpf_array *array;
375 u32 owner_prog_type = 0;
378 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
379 array = container_of(map, struct bpf_array, map);
380 owner_prog_type = array->owner_prog_type;
381 owner_jited = array->owner_jited;
398 map->pages * 1ULL << PAGE_SHIFT,
400 READ_ONCE(map->frozen));
402 if (owner_prog_type) {
403 seq_printf(m, "owner_prog_type:\t%u\n",
405 seq_printf(m, "owner_jited:\t%u\n",
411 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
414 /* We need this handler such that alloc_file() enables
415 * f_mode with FMODE_CAN_READ.
420 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
421 size_t siz, loff_t *ppos)
423 /* We need this handler such that alloc_file() enables
424 * f_mode with FMODE_CAN_WRITE.
429 const struct file_operations bpf_map_fops = {
430 #ifdef CONFIG_PROC_FS
431 .show_fdinfo = bpf_map_show_fdinfo,
433 .release = bpf_map_release,
434 .read = bpf_dummy_read,
435 .write = bpf_dummy_write,
438 int bpf_map_new_fd(struct bpf_map *map, int flags)
442 ret = security_bpf_map(map, OPEN_FMODE(flags));
446 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
450 int bpf_get_file_flag(int flags)
452 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
454 if (flags & BPF_F_RDONLY)
456 if (flags & BPF_F_WRONLY)
461 /* helper macro to check that unused fields 'union bpf_attr' are zero */
462 #define CHECK_ATTR(CMD) \
463 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
464 sizeof(attr->CMD##_LAST_FIELD), 0, \
466 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
467 sizeof(attr->CMD##_LAST_FIELD)) != NULL
469 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
470 * Return 0 on success and < 0 on error.
472 static int bpf_obj_name_cpy(char *dst, const char *src)
474 const char *end = src + BPF_OBJ_NAME_LEN;
476 memset(dst, 0, BPF_OBJ_NAME_LEN);
477 /* Copy all isalnum(), '_' and '.' chars. */
478 while (src < end && *src) {
479 if (!isalnum(*src) &&
480 *src != '_' && *src != '.')
485 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
492 int map_check_no_btf(const struct bpf_map *map,
493 const struct btf *btf,
494 const struct btf_type *key_type,
495 const struct btf_type *value_type)
500 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
501 u32 btf_key_id, u32 btf_value_id)
503 const struct btf_type *key_type, *value_type;
504 u32 key_size, value_size;
507 /* Some maps allow key to be unspecified. */
509 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
510 if (!key_type || key_size != map->key_size)
513 key_type = btf_type_by_id(btf, 0);
514 if (!map->ops->map_check_btf)
518 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
519 if (!value_type || value_size != map->value_size)
522 map->spin_lock_off = btf_find_spin_lock(btf, value_type);
524 if (map_value_has_spin_lock(map)) {
525 if (map->map_flags & BPF_F_RDONLY_PROG)
527 if (map->map_type != BPF_MAP_TYPE_HASH &&
528 map->map_type != BPF_MAP_TYPE_ARRAY &&
529 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE)
531 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
534 "verifier bug spin_lock_off %d value_size %d\n",
535 map->spin_lock_off, map->value_size);
540 if (map->ops->map_check_btf)
541 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
546 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
547 /* called via syscall */
548 static int map_create(union bpf_attr *attr)
550 int numa_node = bpf_map_attr_numa_node(attr);
555 err = CHECK_ATTR(BPF_MAP_CREATE);
559 f_flags = bpf_get_file_flag(attr->map_flags);
563 if (numa_node != NUMA_NO_NODE &&
564 ((unsigned int)numa_node >= nr_node_ids ||
565 !node_online(numa_node)))
568 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
569 map = find_and_alloc_map(attr);
573 err = bpf_obj_name_cpy(map->name, attr->map_name);
575 goto free_map_nouncharge;
577 atomic_set(&map->refcnt, 1);
578 atomic_set(&map->usercnt, 1);
580 if (attr->btf_key_type_id || attr->btf_value_type_id) {
583 if (!attr->btf_value_type_id) {
585 goto free_map_nouncharge;
588 btf = btf_get_by_fd(attr->btf_fd);
591 goto free_map_nouncharge;
594 err = map_check_btf(map, btf, attr->btf_key_type_id,
595 attr->btf_value_type_id);
598 goto free_map_nouncharge;
602 map->btf_key_type_id = attr->btf_key_type_id;
603 map->btf_value_type_id = attr->btf_value_type_id;
605 map->spin_lock_off = -EINVAL;
608 err = security_bpf_map_alloc(map);
610 goto free_map_nouncharge;
612 err = bpf_map_init_memlock(map);
616 err = bpf_map_alloc_id(map);
620 err = bpf_map_new_fd(map, f_flags);
622 /* failed to allocate fd.
623 * bpf_map_put_with_uref() is needed because the above
624 * bpf_map_alloc_id() has published the map
625 * to the userspace and the userspace may
626 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
628 bpf_map_put_with_uref(map);
635 bpf_map_release_memlock(map);
637 security_bpf_map_free(map);
640 map->ops->map_free(map);
644 /* if error is returned, fd is released.
645 * On success caller should complete fd access with matching fdput()
647 struct bpf_map *__bpf_map_get(struct fd f)
650 return ERR_PTR(-EBADF);
651 if (f.file->f_op != &bpf_map_fops) {
653 return ERR_PTR(-EINVAL);
656 return f.file->private_data;
659 /* prog's and map's refcnt limit */
660 #define BPF_MAX_REFCNT 32768
662 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
664 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
665 atomic_dec(&map->refcnt);
666 return ERR_PTR(-EBUSY);
669 atomic_inc(&map->usercnt);
672 EXPORT_SYMBOL_GPL(bpf_map_inc);
674 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
676 struct fd f = fdget(ufd);
679 map = __bpf_map_get(f);
683 map = bpf_map_inc(map, true);
689 /* map_idr_lock should have been held */
690 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
695 refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
697 if (refold >= BPF_MAX_REFCNT) {
698 __bpf_map_put(map, false);
699 return ERR_PTR(-EBUSY);
703 return ERR_PTR(-ENOENT);
706 atomic_inc(&map->usercnt);
711 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
716 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
719 return memdup_user(ukey, key_size);
722 return ERR_PTR(-EINVAL);
727 /* last field in 'union bpf_attr' used by this command */
728 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
730 static int map_lookup_elem(union bpf_attr *attr)
732 void __user *ukey = u64_to_user_ptr(attr->key);
733 void __user *uvalue = u64_to_user_ptr(attr->value);
734 int ufd = attr->map_fd;
736 void *key, *value, *ptr;
741 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
744 if (attr->flags & ~BPF_F_LOCK)
748 map = __bpf_map_get(f);
751 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
756 if ((attr->flags & BPF_F_LOCK) &&
757 !map_value_has_spin_lock(map)) {
762 key = __bpf_copy_key(ukey, map->key_size);
768 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
769 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
770 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
771 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
772 value_size = round_up(map->value_size, 8) * num_possible_cpus();
773 else if (IS_FD_MAP(map))
774 value_size = sizeof(u32);
776 value_size = map->value_size;
779 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
783 if (bpf_map_is_dev_bound(map)) {
784 err = bpf_map_offload_lookup_elem(map, key, value);
789 this_cpu_inc(bpf_prog_active);
790 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
791 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
792 err = bpf_percpu_hash_copy(map, key, value);
793 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
794 err = bpf_percpu_array_copy(map, key, value);
795 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
796 err = bpf_percpu_cgroup_storage_copy(map, key, value);
797 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
798 err = bpf_stackmap_copy(map, key, value);
799 } else if (IS_FD_ARRAY(map)) {
800 err = bpf_fd_array_map_lookup_elem(map, key, value);
801 } else if (IS_FD_HASH(map)) {
802 err = bpf_fd_htab_map_lookup_elem(map, key, value);
803 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
804 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
805 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
806 map->map_type == BPF_MAP_TYPE_STACK) {
807 err = map->ops->map_peek_elem(map, value);
810 ptr = map->ops->map_lookup_elem(map, key);
817 if (attr->flags & BPF_F_LOCK)
818 /* lock 'ptr' and copy everything but lock */
819 copy_map_value_locked(map, value, ptr, true);
821 copy_map_value(map, value, ptr);
822 /* mask lock, since value wasn't zero inited */
823 check_and_init_map_lock(map, value);
827 this_cpu_dec(bpf_prog_active);
835 if (copy_to_user(uvalue, value, value_size) != 0)
849 static void maybe_wait_bpf_programs(struct bpf_map *map)
851 /* Wait for any running BPF programs to complete so that
852 * userspace, when we return to it, knows that all programs
853 * that could be running use the new map value.
855 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
856 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
860 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
862 static int map_update_elem(union bpf_attr *attr)
864 void __user *ukey = u64_to_user_ptr(attr->key);
865 void __user *uvalue = u64_to_user_ptr(attr->value);
866 int ufd = attr->map_fd;
873 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
877 map = __bpf_map_get(f);
880 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
885 if ((attr->flags & BPF_F_LOCK) &&
886 !map_value_has_spin_lock(map)) {
891 key = __bpf_copy_key(ukey, map->key_size);
897 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
898 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
899 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
900 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
901 value_size = round_up(map->value_size, 8) * num_possible_cpus();
903 value_size = map->value_size;
906 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
911 if (copy_from_user(value, uvalue, value_size) != 0)
914 /* Need to create a kthread, thus must support schedule */
915 if (bpf_map_is_dev_bound(map)) {
916 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
918 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
919 map->map_type == BPF_MAP_TYPE_SOCKHASH ||
920 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
921 err = map->ops->map_update_elem(map, key, value, attr->flags);
925 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
926 * inside bpf map update or delete otherwise deadlocks are possible
929 __this_cpu_inc(bpf_prog_active);
930 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
931 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
932 err = bpf_percpu_hash_update(map, key, value, attr->flags);
933 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
934 err = bpf_percpu_array_update(map, key, value, attr->flags);
935 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
936 err = bpf_percpu_cgroup_storage_update(map, key, value,
938 } else if (IS_FD_ARRAY(map)) {
940 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
943 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
945 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
948 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
949 /* rcu_read_lock() is not needed */
950 err = bpf_fd_reuseport_array_update_elem(map, key, value,
952 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
953 map->map_type == BPF_MAP_TYPE_STACK) {
954 err = map->ops->map_push_elem(map, value, attr->flags);
957 err = map->ops->map_update_elem(map, key, value, attr->flags);
960 __this_cpu_dec(bpf_prog_active);
962 maybe_wait_bpf_programs(map);
973 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
975 static int map_delete_elem(union bpf_attr *attr)
977 void __user *ukey = u64_to_user_ptr(attr->key);
978 int ufd = attr->map_fd;
984 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
988 map = __bpf_map_get(f);
991 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
996 key = __bpf_copy_key(ukey, map->key_size);
1002 if (bpf_map_is_dev_bound(map)) {
1003 err = bpf_map_offload_delete_elem(map, key);
1008 __this_cpu_inc(bpf_prog_active);
1010 err = map->ops->map_delete_elem(map, key);
1012 __this_cpu_dec(bpf_prog_active);
1014 maybe_wait_bpf_programs(map);
1022 /* last field in 'union bpf_attr' used by this command */
1023 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1025 static int map_get_next_key(union bpf_attr *attr)
1027 void __user *ukey = u64_to_user_ptr(attr->key);
1028 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1029 int ufd = attr->map_fd;
1030 struct bpf_map *map;
1031 void *key, *next_key;
1035 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1039 map = __bpf_map_get(f);
1041 return PTR_ERR(map);
1042 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1048 key = __bpf_copy_key(ukey, map->key_size);
1058 next_key = kmalloc(map->key_size, GFP_USER);
1062 if (bpf_map_is_dev_bound(map)) {
1063 err = bpf_map_offload_get_next_key(map, key, next_key);
1068 err = map->ops->map_get_next_key(map, key, next_key);
1075 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1089 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1091 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1093 void __user *ukey = u64_to_user_ptr(attr->key);
1094 void __user *uvalue = u64_to_user_ptr(attr->value);
1095 int ufd = attr->map_fd;
1096 struct bpf_map *map;
1102 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1106 map = __bpf_map_get(f);
1108 return PTR_ERR(map);
1109 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1114 key = __bpf_copy_key(ukey, map->key_size);
1120 value_size = map->value_size;
1123 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1127 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1128 map->map_type == BPF_MAP_TYPE_STACK) {
1129 err = map->ops->map_pop_elem(map, value);
1137 if (copy_to_user(uvalue, value, value_size) != 0)
1151 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1153 static int map_freeze(const union bpf_attr *attr)
1155 int err = 0, ufd = attr->map_fd;
1156 struct bpf_map *map;
1159 if (CHECK_ATTR(BPF_MAP_FREEZE))
1163 map = __bpf_map_get(f);
1165 return PTR_ERR(map);
1166 if (READ_ONCE(map->frozen)) {
1170 if (!capable(CAP_SYS_ADMIN)) {
1175 WRITE_ONCE(map->frozen, true);
1181 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1182 #define BPF_PROG_TYPE(_id, _name) \
1183 [_id] = & _name ## _prog_ops,
1184 #define BPF_MAP_TYPE(_id, _ops)
1185 #include <linux/bpf_types.h>
1186 #undef BPF_PROG_TYPE
1190 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1192 const struct bpf_prog_ops *ops;
1194 if (type >= ARRAY_SIZE(bpf_prog_types))
1196 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1197 ops = bpf_prog_types[type];
1201 if (!bpf_prog_is_dev_bound(prog->aux))
1202 prog->aux->ops = ops;
1204 prog->aux->ops = &bpf_offload_prog_ops;
1209 /* drop refcnt on maps used by eBPF program and free auxilary data */
1210 static void free_used_maps(struct bpf_prog_aux *aux)
1212 enum bpf_cgroup_storage_type stype;
1215 for_each_cgroup_storage_type(stype) {
1216 if (!aux->cgroup_storage[stype])
1218 bpf_cgroup_storage_release(aux->prog,
1219 aux->cgroup_storage[stype]);
1222 for (i = 0; i < aux->used_map_cnt; i++)
1223 bpf_map_put(aux->used_maps[i]);
1225 kfree(aux->used_maps);
1228 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1230 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1231 unsigned long user_bufs;
1234 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1235 if (user_bufs > memlock_limit) {
1236 atomic_long_sub(pages, &user->locked_vm);
1244 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1247 atomic_long_sub(pages, &user->locked_vm);
1250 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1252 struct user_struct *user = get_current_user();
1255 ret = __bpf_prog_charge(user, prog->pages);
1261 prog->aux->user = user;
1265 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1267 struct user_struct *user = prog->aux->user;
1269 __bpf_prog_uncharge(user, prog->pages);
1273 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1277 idr_preload(GFP_KERNEL);
1278 spin_lock_bh(&prog_idr_lock);
1279 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1282 spin_unlock_bh(&prog_idr_lock);
1285 /* id is in [1, INT_MAX) */
1286 if (WARN_ON_ONCE(!id))
1289 return id > 0 ? 0 : id;
1292 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1294 /* cBPF to eBPF migrations are currently not in the idr store.
1295 * Offloaded programs are removed from the store when their device
1296 * disappears - even if someone grabs an fd to them they are unusable,
1297 * simply waiting for refcnt to drop to be freed.
1303 spin_lock_bh(&prog_idr_lock);
1305 __acquire(&prog_idr_lock);
1307 idr_remove(&prog_idr, prog->aux->id);
1311 spin_unlock_bh(&prog_idr_lock);
1313 __release(&prog_idr_lock);
1316 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1318 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1320 free_used_maps(aux);
1321 bpf_prog_uncharge_memlock(aux->prog);
1322 security_bpf_prog_free(aux);
1323 bpf_prog_free(aux->prog);
1326 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1328 if (atomic_dec_and_test(&prog->aux->refcnt)) {
1329 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1330 /* bpf_prog_free_id() must be called first */
1331 bpf_prog_free_id(prog, do_idr_lock);
1332 bpf_prog_kallsyms_del_all(prog);
1333 btf_put(prog->aux->btf);
1334 kvfree(prog->aux->func_info);
1335 bpf_prog_free_linfo(prog);
1337 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1341 void bpf_prog_put(struct bpf_prog *prog)
1343 __bpf_prog_put(prog, true);
1345 EXPORT_SYMBOL_GPL(bpf_prog_put);
1347 static int bpf_prog_release(struct inode *inode, struct file *filp)
1349 struct bpf_prog *prog = filp->private_data;
1355 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1356 struct bpf_prog_stats *stats)
1358 u64 nsecs = 0, cnt = 0;
1361 for_each_possible_cpu(cpu) {
1362 const struct bpf_prog_stats *st;
1366 st = per_cpu_ptr(prog->aux->stats, cpu);
1368 start = u64_stats_fetch_begin_irq(&st->syncp);
1371 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1375 stats->nsecs = nsecs;
1379 #ifdef CONFIG_PROC_FS
1380 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1382 const struct bpf_prog *prog = filp->private_data;
1383 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1384 struct bpf_prog_stats stats;
1386 bpf_prog_get_stats(prog, &stats);
1387 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1394 "run_time_ns:\t%llu\n"
1399 prog->pages * 1ULL << PAGE_SHIFT,
1406 const struct file_operations bpf_prog_fops = {
1407 #ifdef CONFIG_PROC_FS
1408 .show_fdinfo = bpf_prog_show_fdinfo,
1410 .release = bpf_prog_release,
1411 .read = bpf_dummy_read,
1412 .write = bpf_dummy_write,
1415 int bpf_prog_new_fd(struct bpf_prog *prog)
1419 ret = security_bpf_prog(prog);
1423 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1424 O_RDWR | O_CLOEXEC);
1427 static struct bpf_prog *____bpf_prog_get(struct fd f)
1430 return ERR_PTR(-EBADF);
1431 if (f.file->f_op != &bpf_prog_fops) {
1433 return ERR_PTR(-EINVAL);
1436 return f.file->private_data;
1439 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1441 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1442 atomic_sub(i, &prog->aux->refcnt);
1443 return ERR_PTR(-EBUSY);
1447 EXPORT_SYMBOL_GPL(bpf_prog_add);
1449 void bpf_prog_sub(struct bpf_prog *prog, int i)
1451 /* Only to be used for undoing previous bpf_prog_add() in some
1452 * error path. We still know that another entity in our call
1453 * path holds a reference to the program, thus atomic_sub() can
1454 * be safely used in such cases!
1456 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1458 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1460 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1462 return bpf_prog_add(prog, 1);
1464 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1466 /* prog_idr_lock should have been held */
1467 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1471 refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1473 if (refold >= BPF_MAX_REFCNT) {
1474 __bpf_prog_put(prog, false);
1475 return ERR_PTR(-EBUSY);
1479 return ERR_PTR(-ENOENT);
1483 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1485 bool bpf_prog_get_ok(struct bpf_prog *prog,
1486 enum bpf_prog_type *attach_type, bool attach_drv)
1488 /* not an attachment, just a refcount inc, always allow */
1492 if (prog->type != *attach_type)
1494 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1500 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1503 struct fd f = fdget(ufd);
1504 struct bpf_prog *prog;
1506 prog = ____bpf_prog_get(f);
1509 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1510 prog = ERR_PTR(-EINVAL);
1514 prog = bpf_prog_inc(prog);
1520 struct bpf_prog *bpf_prog_get(u32 ufd)
1522 return __bpf_prog_get(ufd, NULL, false);
1525 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1528 return __bpf_prog_get(ufd, &type, attach_drv);
1530 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1532 /* Initially all BPF programs could be loaded w/o specifying
1533 * expected_attach_type. Later for some of them specifying expected_attach_type
1534 * at load time became required so that program could be validated properly.
1535 * Programs of types that are allowed to be loaded both w/ and w/o (for
1536 * backward compatibility) expected_attach_type, should have the default attach
1537 * type assigned to expected_attach_type for the latter case, so that it can be
1538 * validated later at attach time.
1540 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1541 * prog type requires it but has some attach types that have to be backward
1544 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1546 switch (attr->prog_type) {
1547 case BPF_PROG_TYPE_CGROUP_SOCK:
1548 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1549 * exist so checking for non-zero is the way to go here.
1551 if (!attr->expected_attach_type)
1552 attr->expected_attach_type =
1553 BPF_CGROUP_INET_SOCK_CREATE;
1559 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1560 enum bpf_attach_type expected_attach_type)
1562 switch (prog_type) {
1563 case BPF_PROG_TYPE_CGROUP_SOCK:
1564 switch (expected_attach_type) {
1565 case BPF_CGROUP_INET_SOCK_CREATE:
1566 case BPF_CGROUP_INET4_POST_BIND:
1567 case BPF_CGROUP_INET6_POST_BIND:
1572 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1573 switch (expected_attach_type) {
1574 case BPF_CGROUP_INET4_BIND:
1575 case BPF_CGROUP_INET6_BIND:
1576 case BPF_CGROUP_INET4_CONNECT:
1577 case BPF_CGROUP_INET6_CONNECT:
1578 case BPF_CGROUP_UDP4_SENDMSG:
1579 case BPF_CGROUP_UDP6_SENDMSG:
1589 /* last field in 'union bpf_attr' used by this command */
1590 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1592 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1594 enum bpf_prog_type type = attr->prog_type;
1595 struct bpf_prog *prog;
1600 if (CHECK_ATTR(BPF_PROG_LOAD))
1603 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT))
1606 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1607 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1608 !capable(CAP_SYS_ADMIN))
1611 /* copy eBPF program license from user space */
1612 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1613 sizeof(license) - 1) < 0)
1615 license[sizeof(license) - 1] = 0;
1617 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1618 is_gpl = license_is_gpl_compatible(license);
1620 if (attr->insn_cnt == 0 ||
1621 attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1623 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1624 type != BPF_PROG_TYPE_CGROUP_SKB &&
1625 !capable(CAP_SYS_ADMIN))
1628 bpf_prog_load_fixup_attach_type(attr);
1629 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1632 /* plain bpf_prog allocation */
1633 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1637 prog->expected_attach_type = attr->expected_attach_type;
1639 prog->aux->offload_requested = !!attr->prog_ifindex;
1641 err = security_bpf_prog_alloc(prog->aux);
1643 goto free_prog_nouncharge;
1645 err = bpf_prog_charge_memlock(prog);
1649 prog->len = attr->insn_cnt;
1652 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1653 bpf_prog_insn_size(prog)) != 0)
1656 prog->orig_prog = NULL;
1659 atomic_set(&prog->aux->refcnt, 1);
1660 prog->gpl_compatible = is_gpl ? 1 : 0;
1662 if (bpf_prog_is_dev_bound(prog->aux)) {
1663 err = bpf_prog_offload_init(prog, attr);
1668 /* find program type: socket_filter vs tracing_filter */
1669 err = find_prog_type(type, prog);
1673 prog->aux->load_time = ktime_get_boot_ns();
1674 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1678 /* run eBPF verifier */
1679 err = bpf_check(&prog, attr, uattr);
1681 goto free_used_maps;
1683 prog = bpf_prog_select_runtime(prog, &err);
1685 goto free_used_maps;
1687 err = bpf_prog_alloc_id(prog);
1689 goto free_used_maps;
1691 err = bpf_prog_new_fd(prog);
1693 /* failed to allocate fd.
1694 * bpf_prog_put() is needed because the above
1695 * bpf_prog_alloc_id() has published the prog
1696 * to the userspace and the userspace may
1697 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1703 bpf_prog_kallsyms_add(prog);
1704 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1708 bpf_prog_free_linfo(prog);
1709 kvfree(prog->aux->func_info);
1710 btf_put(prog->aux->btf);
1711 bpf_prog_kallsyms_del_subprogs(prog);
1712 free_used_maps(prog->aux);
1714 bpf_prog_uncharge_memlock(prog);
1716 security_bpf_prog_free(prog->aux);
1717 free_prog_nouncharge:
1718 bpf_prog_free(prog);
1722 #define BPF_OBJ_LAST_FIELD file_flags
1724 static int bpf_obj_pin(const union bpf_attr *attr)
1726 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1729 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1732 static int bpf_obj_get(const union bpf_attr *attr)
1734 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1735 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1738 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1742 struct bpf_raw_tracepoint {
1743 struct bpf_raw_event_map *btp;
1744 struct bpf_prog *prog;
1747 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1749 struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1752 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1753 bpf_prog_put(raw_tp->prog);
1755 bpf_put_raw_tracepoint(raw_tp->btp);
1760 static const struct file_operations bpf_raw_tp_fops = {
1761 .release = bpf_raw_tracepoint_release,
1762 .read = bpf_dummy_read,
1763 .write = bpf_dummy_write,
1766 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1768 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1770 struct bpf_raw_tracepoint *raw_tp;
1771 struct bpf_raw_event_map *btp;
1772 struct bpf_prog *prog;
1776 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1777 sizeof(tp_name) - 1) < 0)
1779 tp_name[sizeof(tp_name) - 1] = 0;
1781 btp = bpf_get_raw_tracepoint(tp_name);
1785 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1792 prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1793 BPF_PROG_TYPE_RAW_TRACEPOINT);
1795 err = PTR_ERR(prog);
1799 err = bpf_probe_register(raw_tp->btp, prog);
1803 raw_tp->prog = prog;
1804 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1807 bpf_probe_unregister(raw_tp->btp, prog);
1818 bpf_put_raw_tracepoint(btp);
1822 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1823 enum bpf_attach_type attach_type)
1825 switch (prog->type) {
1826 case BPF_PROG_TYPE_CGROUP_SOCK:
1827 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1828 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1834 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1836 #define BPF_F_ATTACH_MASK \
1837 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1839 static int bpf_prog_attach(const union bpf_attr *attr)
1841 enum bpf_prog_type ptype;
1842 struct bpf_prog *prog;
1845 if (!capable(CAP_NET_ADMIN))
1848 if (CHECK_ATTR(BPF_PROG_ATTACH))
1851 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1854 switch (attr->attach_type) {
1855 case BPF_CGROUP_INET_INGRESS:
1856 case BPF_CGROUP_INET_EGRESS:
1857 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1859 case BPF_CGROUP_INET_SOCK_CREATE:
1860 case BPF_CGROUP_INET4_POST_BIND:
1861 case BPF_CGROUP_INET6_POST_BIND:
1862 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1864 case BPF_CGROUP_INET4_BIND:
1865 case BPF_CGROUP_INET6_BIND:
1866 case BPF_CGROUP_INET4_CONNECT:
1867 case BPF_CGROUP_INET6_CONNECT:
1868 case BPF_CGROUP_UDP4_SENDMSG:
1869 case BPF_CGROUP_UDP6_SENDMSG:
1870 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1872 case BPF_CGROUP_SOCK_OPS:
1873 ptype = BPF_PROG_TYPE_SOCK_OPS;
1875 case BPF_CGROUP_DEVICE:
1876 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1878 case BPF_SK_MSG_VERDICT:
1879 ptype = BPF_PROG_TYPE_SK_MSG;
1881 case BPF_SK_SKB_STREAM_PARSER:
1882 case BPF_SK_SKB_STREAM_VERDICT:
1883 ptype = BPF_PROG_TYPE_SK_SKB;
1885 case BPF_LIRC_MODE2:
1886 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1888 case BPF_FLOW_DISSECTOR:
1889 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1895 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1897 return PTR_ERR(prog);
1899 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1905 case BPF_PROG_TYPE_SK_SKB:
1906 case BPF_PROG_TYPE_SK_MSG:
1907 ret = sock_map_get_from_fd(attr, prog);
1909 case BPF_PROG_TYPE_LIRC_MODE2:
1910 ret = lirc_prog_attach(attr, prog);
1912 case BPF_PROG_TYPE_FLOW_DISSECTOR:
1913 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1916 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1924 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1926 static int bpf_prog_detach(const union bpf_attr *attr)
1928 enum bpf_prog_type ptype;
1930 if (!capable(CAP_NET_ADMIN))
1933 if (CHECK_ATTR(BPF_PROG_DETACH))
1936 switch (attr->attach_type) {
1937 case BPF_CGROUP_INET_INGRESS:
1938 case BPF_CGROUP_INET_EGRESS:
1939 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1941 case BPF_CGROUP_INET_SOCK_CREATE:
1942 case BPF_CGROUP_INET4_POST_BIND:
1943 case BPF_CGROUP_INET6_POST_BIND:
1944 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1946 case BPF_CGROUP_INET4_BIND:
1947 case BPF_CGROUP_INET6_BIND:
1948 case BPF_CGROUP_INET4_CONNECT:
1949 case BPF_CGROUP_INET6_CONNECT:
1950 case BPF_CGROUP_UDP4_SENDMSG:
1951 case BPF_CGROUP_UDP6_SENDMSG:
1952 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1954 case BPF_CGROUP_SOCK_OPS:
1955 ptype = BPF_PROG_TYPE_SOCK_OPS;
1957 case BPF_CGROUP_DEVICE:
1958 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1960 case BPF_SK_MSG_VERDICT:
1961 return sock_map_get_from_fd(attr, NULL);
1962 case BPF_SK_SKB_STREAM_PARSER:
1963 case BPF_SK_SKB_STREAM_VERDICT:
1964 return sock_map_get_from_fd(attr, NULL);
1965 case BPF_LIRC_MODE2:
1966 return lirc_prog_detach(attr);
1967 case BPF_FLOW_DISSECTOR:
1968 return skb_flow_dissector_bpf_prog_detach(attr);
1973 return cgroup_bpf_prog_detach(attr, ptype);
1976 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1978 static int bpf_prog_query(const union bpf_attr *attr,
1979 union bpf_attr __user *uattr)
1981 if (!capable(CAP_NET_ADMIN))
1983 if (CHECK_ATTR(BPF_PROG_QUERY))
1985 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1988 switch (attr->query.attach_type) {
1989 case BPF_CGROUP_INET_INGRESS:
1990 case BPF_CGROUP_INET_EGRESS:
1991 case BPF_CGROUP_INET_SOCK_CREATE:
1992 case BPF_CGROUP_INET4_BIND:
1993 case BPF_CGROUP_INET6_BIND:
1994 case BPF_CGROUP_INET4_POST_BIND:
1995 case BPF_CGROUP_INET6_POST_BIND:
1996 case BPF_CGROUP_INET4_CONNECT:
1997 case BPF_CGROUP_INET6_CONNECT:
1998 case BPF_CGROUP_UDP4_SENDMSG:
1999 case BPF_CGROUP_UDP6_SENDMSG:
2000 case BPF_CGROUP_SOCK_OPS:
2001 case BPF_CGROUP_DEVICE:
2003 case BPF_LIRC_MODE2:
2004 return lirc_prog_query(attr, uattr);
2009 return cgroup_bpf_prog_query(attr, uattr);
2012 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2014 static int bpf_prog_test_run(const union bpf_attr *attr,
2015 union bpf_attr __user *uattr)
2017 struct bpf_prog *prog;
2018 int ret = -ENOTSUPP;
2020 if (!capable(CAP_SYS_ADMIN))
2022 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2025 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2026 (!attr->test.ctx_size_in && attr->test.ctx_in))
2029 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2030 (!attr->test.ctx_size_out && attr->test.ctx_out))
2033 prog = bpf_prog_get(attr->test.prog_fd);
2035 return PTR_ERR(prog);
2037 if (prog->aux->ops->test_run)
2038 ret = prog->aux->ops->test_run(prog, attr, uattr);
2044 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2046 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2047 union bpf_attr __user *uattr,
2051 u32 next_id = attr->start_id;
2054 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2057 if (!capable(CAP_SYS_ADMIN))
2062 if (!idr_get_next(idr, &next_id))
2064 spin_unlock_bh(lock);
2067 err = put_user(next_id, &uattr->next_id);
2072 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2074 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2076 struct bpf_prog *prog;
2077 u32 id = attr->prog_id;
2080 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2083 if (!capable(CAP_SYS_ADMIN))
2086 spin_lock_bh(&prog_idr_lock);
2087 prog = idr_find(&prog_idr, id);
2089 prog = bpf_prog_inc_not_zero(prog);
2091 prog = ERR_PTR(-ENOENT);
2092 spin_unlock_bh(&prog_idr_lock);
2095 return PTR_ERR(prog);
2097 fd = bpf_prog_new_fd(prog);
2104 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2106 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2108 struct bpf_map *map;
2109 u32 id = attr->map_id;
2113 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2114 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2117 if (!capable(CAP_SYS_ADMIN))
2120 f_flags = bpf_get_file_flag(attr->open_flags);
2124 spin_lock_bh(&map_idr_lock);
2125 map = idr_find(&map_idr, id);
2127 map = bpf_map_inc_not_zero(map, true);
2129 map = ERR_PTR(-ENOENT);
2130 spin_unlock_bh(&map_idr_lock);
2133 return PTR_ERR(map);
2135 fd = bpf_map_new_fd(map, f_flags);
2137 bpf_map_put_with_uref(map);
2142 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2143 unsigned long addr, u32 *off,
2146 const struct bpf_map *map;
2149 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2150 map = prog->aux->used_maps[i];
2151 if (map == (void *)addr) {
2152 *type = BPF_PSEUDO_MAP_FD;
2155 if (!map->ops->map_direct_value_meta)
2157 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2158 *type = BPF_PSEUDO_MAP_VALUE;
2166 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2168 const struct bpf_map *map;
2169 struct bpf_insn *insns;
2174 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2179 for (i = 0; i < prog->len; i++) {
2180 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2181 insns[i].code = BPF_JMP | BPF_CALL;
2182 insns[i].imm = BPF_FUNC_tail_call;
2185 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2186 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2187 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2188 insns[i].code = BPF_JMP | BPF_CALL;
2189 if (!bpf_dump_raw_ok())
2194 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2197 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2198 map = bpf_map_from_imm(prog, imm, &off, &type);
2200 insns[i].src_reg = type;
2201 insns[i].imm = map->id;
2202 insns[i + 1].imm = off;
2210 static int set_info_rec_size(struct bpf_prog_info *info)
2213 * Ensure info.*_rec_size is the same as kernel expected size
2217 * Only allow zero *_rec_size if both _rec_size and _cnt are
2218 * zero. In this case, the kernel will set the expected
2219 * _rec_size back to the info.
2222 if ((info->nr_func_info || info->func_info_rec_size) &&
2223 info->func_info_rec_size != sizeof(struct bpf_func_info))
2226 if ((info->nr_line_info || info->line_info_rec_size) &&
2227 info->line_info_rec_size != sizeof(struct bpf_line_info))
2230 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2231 info->jited_line_info_rec_size != sizeof(__u64))
2234 info->func_info_rec_size = sizeof(struct bpf_func_info);
2235 info->line_info_rec_size = sizeof(struct bpf_line_info);
2236 info->jited_line_info_rec_size = sizeof(__u64);
2241 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2242 const union bpf_attr *attr,
2243 union bpf_attr __user *uattr)
2245 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2246 struct bpf_prog_info info = {};
2247 u32 info_len = attr->info.info_len;
2248 struct bpf_prog_stats stats;
2249 char __user *uinsns;
2253 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2256 info_len = min_t(u32, sizeof(info), info_len);
2258 if (copy_from_user(&info, uinfo, info_len))
2261 info.type = prog->type;
2262 info.id = prog->aux->id;
2263 info.load_time = prog->aux->load_time;
2264 info.created_by_uid = from_kuid_munged(current_user_ns(),
2265 prog->aux->user->uid);
2266 info.gpl_compatible = prog->gpl_compatible;
2268 memcpy(info.tag, prog->tag, sizeof(prog->tag));
2269 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2271 ulen = info.nr_map_ids;
2272 info.nr_map_ids = prog->aux->used_map_cnt;
2273 ulen = min_t(u32, info.nr_map_ids, ulen);
2275 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2278 for (i = 0; i < ulen; i++)
2279 if (put_user(prog->aux->used_maps[i]->id,
2284 err = set_info_rec_size(&info);
2288 bpf_prog_get_stats(prog, &stats);
2289 info.run_time_ns = stats.nsecs;
2290 info.run_cnt = stats.cnt;
2292 if (!capable(CAP_SYS_ADMIN)) {
2293 info.jited_prog_len = 0;
2294 info.xlated_prog_len = 0;
2295 info.nr_jited_ksyms = 0;
2296 info.nr_jited_func_lens = 0;
2297 info.nr_func_info = 0;
2298 info.nr_line_info = 0;
2299 info.nr_jited_line_info = 0;
2303 ulen = info.xlated_prog_len;
2304 info.xlated_prog_len = bpf_prog_insn_size(prog);
2305 if (info.xlated_prog_len && ulen) {
2306 struct bpf_insn *insns_sanitized;
2309 if (prog->blinded && !bpf_dump_raw_ok()) {
2310 info.xlated_prog_insns = 0;
2313 insns_sanitized = bpf_insn_prepare_dump(prog);
2314 if (!insns_sanitized)
2316 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2317 ulen = min_t(u32, info.xlated_prog_len, ulen);
2318 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2319 kfree(insns_sanitized);
2324 if (bpf_prog_is_dev_bound(prog->aux)) {
2325 err = bpf_prog_offload_info_fill(&info, prog);
2331 /* NOTE: the following code is supposed to be skipped for offload.
2332 * bpf_prog_offload_info_fill() is the place to fill similar fields
2335 ulen = info.jited_prog_len;
2336 if (prog->aux->func_cnt) {
2339 info.jited_prog_len = 0;
2340 for (i = 0; i < prog->aux->func_cnt; i++)
2341 info.jited_prog_len += prog->aux->func[i]->jited_len;
2343 info.jited_prog_len = prog->jited_len;
2346 if (info.jited_prog_len && ulen) {
2347 if (bpf_dump_raw_ok()) {
2348 uinsns = u64_to_user_ptr(info.jited_prog_insns);
2349 ulen = min_t(u32, info.jited_prog_len, ulen);
2351 /* for multi-function programs, copy the JITed
2352 * instructions for all the functions
2354 if (prog->aux->func_cnt) {
2359 for (i = 0; i < prog->aux->func_cnt; i++) {
2360 len = prog->aux->func[i]->jited_len;
2361 len = min_t(u32, len, free);
2362 img = (u8 *) prog->aux->func[i]->bpf_func;
2363 if (copy_to_user(uinsns, img, len))
2371 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2375 info.jited_prog_insns = 0;
2379 ulen = info.nr_jited_ksyms;
2380 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2382 if (bpf_dump_raw_ok()) {
2383 unsigned long ksym_addr;
2384 u64 __user *user_ksyms;
2387 /* copy the address of the kernel symbol
2388 * corresponding to each function
2390 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2391 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2392 if (prog->aux->func_cnt) {
2393 for (i = 0; i < ulen; i++) {
2394 ksym_addr = (unsigned long)
2395 prog->aux->func[i]->bpf_func;
2396 if (put_user((u64) ksym_addr,
2401 ksym_addr = (unsigned long) prog->bpf_func;
2402 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2406 info.jited_ksyms = 0;
2410 ulen = info.nr_jited_func_lens;
2411 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2413 if (bpf_dump_raw_ok()) {
2414 u32 __user *user_lens;
2417 /* copy the JITed image lengths for each function */
2418 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2419 user_lens = u64_to_user_ptr(info.jited_func_lens);
2420 if (prog->aux->func_cnt) {
2421 for (i = 0; i < ulen; i++) {
2423 prog->aux->func[i]->jited_len;
2424 if (put_user(func_len, &user_lens[i]))
2428 func_len = prog->jited_len;
2429 if (put_user(func_len, &user_lens[0]))
2433 info.jited_func_lens = 0;
2438 info.btf_id = btf_id(prog->aux->btf);
2440 ulen = info.nr_func_info;
2441 info.nr_func_info = prog->aux->func_info_cnt;
2442 if (info.nr_func_info && ulen) {
2443 char __user *user_finfo;
2445 user_finfo = u64_to_user_ptr(info.func_info);
2446 ulen = min_t(u32, info.nr_func_info, ulen);
2447 if (copy_to_user(user_finfo, prog->aux->func_info,
2448 info.func_info_rec_size * ulen))
2452 ulen = info.nr_line_info;
2453 info.nr_line_info = prog->aux->nr_linfo;
2454 if (info.nr_line_info && ulen) {
2455 __u8 __user *user_linfo;
2457 user_linfo = u64_to_user_ptr(info.line_info);
2458 ulen = min_t(u32, info.nr_line_info, ulen);
2459 if (copy_to_user(user_linfo, prog->aux->linfo,
2460 info.line_info_rec_size * ulen))
2464 ulen = info.nr_jited_line_info;
2465 if (prog->aux->jited_linfo)
2466 info.nr_jited_line_info = prog->aux->nr_linfo;
2468 info.nr_jited_line_info = 0;
2469 if (info.nr_jited_line_info && ulen) {
2470 if (bpf_dump_raw_ok()) {
2471 __u64 __user *user_linfo;
2474 user_linfo = u64_to_user_ptr(info.jited_line_info);
2475 ulen = min_t(u32, info.nr_jited_line_info, ulen);
2476 for (i = 0; i < ulen; i++) {
2477 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2482 info.jited_line_info = 0;
2486 ulen = info.nr_prog_tags;
2487 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2489 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2492 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2493 ulen = min_t(u32, info.nr_prog_tags, ulen);
2494 if (prog->aux->func_cnt) {
2495 for (i = 0; i < ulen; i++) {
2496 if (copy_to_user(user_prog_tags[i],
2497 prog->aux->func[i]->tag,
2502 if (copy_to_user(user_prog_tags[0],
2503 prog->tag, BPF_TAG_SIZE))
2509 if (copy_to_user(uinfo, &info, info_len) ||
2510 put_user(info_len, &uattr->info.info_len))
2516 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2517 const union bpf_attr *attr,
2518 union bpf_attr __user *uattr)
2520 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2521 struct bpf_map_info info = {};
2522 u32 info_len = attr->info.info_len;
2525 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2528 info_len = min_t(u32, sizeof(info), info_len);
2530 info.type = map->map_type;
2532 info.key_size = map->key_size;
2533 info.value_size = map->value_size;
2534 info.max_entries = map->max_entries;
2535 info.map_flags = map->map_flags;
2536 memcpy(info.name, map->name, sizeof(map->name));
2539 info.btf_id = btf_id(map->btf);
2540 info.btf_key_type_id = map->btf_key_type_id;
2541 info.btf_value_type_id = map->btf_value_type_id;
2544 if (bpf_map_is_dev_bound(map)) {
2545 err = bpf_map_offload_info_fill(&info, map);
2550 if (copy_to_user(uinfo, &info, info_len) ||
2551 put_user(info_len, &uattr->info.info_len))
2557 static int bpf_btf_get_info_by_fd(struct btf *btf,
2558 const union bpf_attr *attr,
2559 union bpf_attr __user *uattr)
2561 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2562 u32 info_len = attr->info.info_len;
2565 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2569 return btf_get_info_by_fd(btf, attr, uattr);
2572 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2574 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2575 union bpf_attr __user *uattr)
2577 int ufd = attr->info.bpf_fd;
2581 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2588 if (f.file->f_op == &bpf_prog_fops)
2589 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2591 else if (f.file->f_op == &bpf_map_fops)
2592 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2594 else if (f.file->f_op == &btf_fops)
2595 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2603 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2605 static int bpf_btf_load(const union bpf_attr *attr)
2607 if (CHECK_ATTR(BPF_BTF_LOAD))
2610 if (!capable(CAP_SYS_ADMIN))
2613 return btf_new_fd(attr);
2616 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2618 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2620 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2623 if (!capable(CAP_SYS_ADMIN))
2626 return btf_get_fd_by_id(attr->btf_id);
2629 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2630 union bpf_attr __user *uattr,
2631 u32 prog_id, u32 fd_type,
2632 const char *buf, u64 probe_offset,
2635 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2636 u32 len = buf ? strlen(buf) : 0, input_len;
2639 if (put_user(len, &uattr->task_fd_query.buf_len))
2641 input_len = attr->task_fd_query.buf_len;
2642 if (input_len && ubuf) {
2644 /* nothing to copy, just make ubuf NULL terminated */
2647 if (put_user(zero, ubuf))
2649 } else if (input_len >= len + 1) {
2650 /* ubuf can hold the string with NULL terminator */
2651 if (copy_to_user(ubuf, buf, len + 1))
2654 /* ubuf cannot hold the string with NULL terminator,
2655 * do a partial copy with NULL terminator.
2660 if (copy_to_user(ubuf, buf, input_len - 1))
2662 if (put_user(zero, ubuf + input_len - 1))
2667 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2668 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2669 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2670 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2676 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2678 static int bpf_task_fd_query(const union bpf_attr *attr,
2679 union bpf_attr __user *uattr)
2681 pid_t pid = attr->task_fd_query.pid;
2682 u32 fd = attr->task_fd_query.fd;
2683 const struct perf_event *event;
2684 struct files_struct *files;
2685 struct task_struct *task;
2689 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2692 if (!capable(CAP_SYS_ADMIN))
2695 if (attr->task_fd_query.flags != 0)
2698 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2702 files = get_files_struct(task);
2703 put_task_struct(task);
2708 spin_lock(&files->file_lock);
2709 file = fcheck_files(files, fd);
2714 spin_unlock(&files->file_lock);
2715 put_files_struct(files);
2720 if (file->f_op == &bpf_raw_tp_fops) {
2721 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2722 struct bpf_raw_event_map *btp = raw_tp->btp;
2724 err = bpf_task_fd_query_copy(attr, uattr,
2725 raw_tp->prog->aux->id,
2726 BPF_FD_TYPE_RAW_TRACEPOINT,
2727 btp->tp->name, 0, 0);
2731 event = perf_get_event(file);
2732 if (!IS_ERR(event)) {
2733 u64 probe_offset, probe_addr;
2734 u32 prog_id, fd_type;
2737 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2738 &buf, &probe_offset,
2741 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2755 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2757 union bpf_attr attr = {};
2760 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2763 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2766 size = min_t(u32, size, sizeof(attr));
2768 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2769 if (copy_from_user(&attr, uattr, size) != 0)
2772 err = security_bpf(cmd, &attr, size);
2777 case BPF_MAP_CREATE:
2778 err = map_create(&attr);
2780 case BPF_MAP_LOOKUP_ELEM:
2781 err = map_lookup_elem(&attr);
2783 case BPF_MAP_UPDATE_ELEM:
2784 err = map_update_elem(&attr);
2786 case BPF_MAP_DELETE_ELEM:
2787 err = map_delete_elem(&attr);
2789 case BPF_MAP_GET_NEXT_KEY:
2790 err = map_get_next_key(&attr);
2792 case BPF_MAP_FREEZE:
2793 err = map_freeze(&attr);
2796 err = bpf_prog_load(&attr, uattr);
2799 err = bpf_obj_pin(&attr);
2802 err = bpf_obj_get(&attr);
2804 case BPF_PROG_ATTACH:
2805 err = bpf_prog_attach(&attr);
2807 case BPF_PROG_DETACH:
2808 err = bpf_prog_detach(&attr);
2810 case BPF_PROG_QUERY:
2811 err = bpf_prog_query(&attr, uattr);
2813 case BPF_PROG_TEST_RUN:
2814 err = bpf_prog_test_run(&attr, uattr);
2816 case BPF_PROG_GET_NEXT_ID:
2817 err = bpf_obj_get_next_id(&attr, uattr,
2818 &prog_idr, &prog_idr_lock);
2820 case BPF_MAP_GET_NEXT_ID:
2821 err = bpf_obj_get_next_id(&attr, uattr,
2822 &map_idr, &map_idr_lock);
2824 case BPF_PROG_GET_FD_BY_ID:
2825 err = bpf_prog_get_fd_by_id(&attr);
2827 case BPF_MAP_GET_FD_BY_ID:
2828 err = bpf_map_get_fd_by_id(&attr);
2830 case BPF_OBJ_GET_INFO_BY_FD:
2831 err = bpf_obj_get_info_by_fd(&attr, uattr);
2833 case BPF_RAW_TRACEPOINT_OPEN:
2834 err = bpf_raw_tracepoint_open(&attr);
2837 err = bpf_btf_load(&attr);
2839 case BPF_BTF_GET_FD_BY_ID:
2840 err = bpf_btf_get_fd_by_id(&attr);
2842 case BPF_TASK_FD_QUERY:
2843 err = bpf_task_fd_query(&attr, uattr);
2845 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2846 err = map_lookup_and_delete_elem(&attr);