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/btf.h>
34 #include <linux/nospec.h>
36 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
37 (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
38 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
39 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
40 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
41 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
43 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
45 DEFINE_PER_CPU(int, bpf_prog_active);
46 static DEFINE_IDR(prog_idr);
47 static DEFINE_SPINLOCK(prog_idr_lock);
48 static DEFINE_IDR(map_idr);
49 static DEFINE_SPINLOCK(map_idr_lock);
51 int sysctl_unprivileged_bpf_disabled __read_mostly;
53 static const struct bpf_map_ops * const bpf_map_types[] = {
54 #define BPF_PROG_TYPE(_id, _ops)
55 #define BPF_MAP_TYPE(_id, _ops) \
57 #include <linux/bpf_types.h>
63 * If we're handed a bigger struct than we know of, ensure all the unknown bits
64 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
65 * we don't know about yet.
67 * There is a ToCToU between this function call and the following
68 * copy_from_user() call. However, this is not a concern since this function is
69 * meant to be a future-proofing of bits.
71 int bpf_check_uarg_tail_zero(void __user *uaddr,
75 unsigned char __user *addr;
76 unsigned char __user *end;
80 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
83 if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
86 if (actual_size <= expected_size)
89 addr = uaddr + expected_size;
90 end = uaddr + actual_size;
92 for (; addr < end; addr++) {
93 err = get_user(val, addr);
103 const struct bpf_map_ops bpf_map_offload_ops = {
104 .map_alloc = bpf_map_offload_map_alloc,
105 .map_free = bpf_map_offload_map_free,
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 definitely need __GFP_NORETRY, so OOM killer doesn't
140 * trigger under memory pressure as we really just want to
143 const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
146 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
147 area = kmalloc_node(size, GFP_USER | flags, numa_node);
152 return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
153 __builtin_return_address(0));
156 void bpf_map_area_free(void *area)
161 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
163 map->map_type = attr->map_type;
164 map->key_size = attr->key_size;
165 map->value_size = attr->value_size;
166 map->max_entries = attr->max_entries;
167 map->map_flags = attr->map_flags;
168 map->numa_node = bpf_map_attr_numa_node(attr);
171 int bpf_map_precharge_memlock(u32 pages)
173 struct user_struct *user = get_current_user();
174 unsigned long memlock_limit, cur;
176 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
177 cur = atomic_long_read(&user->locked_vm);
179 if (cur + pages > memlock_limit)
184 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
186 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
188 if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
189 atomic_long_sub(pages, &user->locked_vm);
195 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
197 atomic_long_sub(pages, &user->locked_vm);
200 static int bpf_map_init_memlock(struct bpf_map *map)
202 struct user_struct *user = get_current_user();
205 ret = bpf_charge_memlock(user, map->pages);
214 static void bpf_map_release_memlock(struct bpf_map *map)
216 struct user_struct *user = map->user;
217 bpf_uncharge_memlock(user, map->pages);
221 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
225 ret = bpf_charge_memlock(map->user, pages);
232 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
234 bpf_uncharge_memlock(map->user, pages);
238 static int bpf_map_alloc_id(struct bpf_map *map)
242 idr_preload(GFP_KERNEL);
243 spin_lock_bh(&map_idr_lock);
244 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
247 spin_unlock_bh(&map_idr_lock);
250 if (WARN_ON_ONCE(!id))
253 return id > 0 ? 0 : id;
256 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
260 /* Offloaded maps are removed from the IDR store when their device
261 * disappears - even if someone holds an fd to them they are unusable,
262 * the memory is gone, all ops will fail; they are simply waiting for
263 * refcnt to drop to be freed.
269 spin_lock_irqsave(&map_idr_lock, flags);
271 __acquire(&map_idr_lock);
273 idr_remove(&map_idr, map->id);
277 spin_unlock_irqrestore(&map_idr_lock, flags);
279 __release(&map_idr_lock);
282 /* called from workqueue */
283 static void bpf_map_free_deferred(struct work_struct *work)
285 struct bpf_map *map = container_of(work, struct bpf_map, work);
287 bpf_map_release_memlock(map);
288 security_bpf_map_free(map);
289 /* implementation dependent freeing */
290 map->ops->map_free(map);
293 static void bpf_map_put_uref(struct bpf_map *map)
295 if (atomic_dec_and_test(&map->usercnt)) {
296 if (map->ops->map_release_uref)
297 map->ops->map_release_uref(map);
301 /* decrement map refcnt and schedule it for freeing via workqueue
302 * (unrelying map implementation ops->map_free() might sleep)
304 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
306 if (atomic_dec_and_test(&map->refcnt)) {
307 /* bpf_map_free_id() must be called first */
308 bpf_map_free_id(map, do_idr_lock);
310 INIT_WORK(&map->work, bpf_map_free_deferred);
311 schedule_work(&map->work);
315 void bpf_map_put(struct bpf_map *map)
317 __bpf_map_put(map, true);
319 EXPORT_SYMBOL_GPL(bpf_map_put);
321 void bpf_map_put_with_uref(struct bpf_map *map)
323 bpf_map_put_uref(map);
327 static int bpf_map_release(struct inode *inode, struct file *filp)
329 struct bpf_map *map = filp->private_data;
331 if (map->ops->map_release)
332 map->ops->map_release(map, filp);
334 bpf_map_put_with_uref(map);
338 #ifdef CONFIG_PROC_FS
339 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
341 const struct bpf_map *map = filp->private_data;
342 const struct bpf_array *array;
343 u32 owner_prog_type = 0;
346 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
347 array = container_of(map, struct bpf_array, map);
348 owner_prog_type = array->owner_prog_type;
349 owner_jited = array->owner_jited;
365 map->pages * 1ULL << PAGE_SHIFT,
368 if (owner_prog_type) {
369 seq_printf(m, "owner_prog_type:\t%u\n",
371 seq_printf(m, "owner_jited:\t%u\n",
377 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
380 /* We need this handler such that alloc_file() enables
381 * f_mode with FMODE_CAN_READ.
386 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
387 size_t siz, loff_t *ppos)
389 /* We need this handler such that alloc_file() enables
390 * f_mode with FMODE_CAN_WRITE.
395 const struct file_operations bpf_map_fops = {
396 #ifdef CONFIG_PROC_FS
397 .show_fdinfo = bpf_map_show_fdinfo,
399 .release = bpf_map_release,
400 .read = bpf_dummy_read,
401 .write = bpf_dummy_write,
404 int bpf_map_new_fd(struct bpf_map *map, int flags)
408 ret = security_bpf_map(map, OPEN_FMODE(flags));
412 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
416 int bpf_get_file_flag(int flags)
418 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
420 if (flags & BPF_F_RDONLY)
422 if (flags & BPF_F_WRONLY)
427 /* helper macro to check that unused fields 'union bpf_attr' are zero */
428 #define CHECK_ATTR(CMD) \
429 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
430 sizeof(attr->CMD##_LAST_FIELD), 0, \
432 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
433 sizeof(attr->CMD##_LAST_FIELD)) != NULL
435 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
436 * Return 0 on success and < 0 on error.
438 static int bpf_obj_name_cpy(char *dst, const char *src)
440 const char *end = src + BPF_OBJ_NAME_LEN;
442 memset(dst, 0, BPF_OBJ_NAME_LEN);
444 /* Copy all isalnum() and '_' char */
445 while (src < end && *src) {
446 if (!isalnum(*src) && *src != '_')
451 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
458 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
459 /* called via syscall */
460 static int map_create(union bpf_attr *attr)
462 int numa_node = bpf_map_attr_numa_node(attr);
467 err = CHECK_ATTR(BPF_MAP_CREATE);
471 f_flags = bpf_get_file_flag(attr->map_flags);
475 if (numa_node != NUMA_NO_NODE &&
476 ((unsigned int)numa_node >= nr_node_ids ||
477 !node_online(numa_node)))
480 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
481 map = find_and_alloc_map(attr);
485 err = bpf_obj_name_cpy(map->name, attr->map_name);
487 goto free_map_nouncharge;
489 atomic_set(&map->refcnt, 1);
490 atomic_set(&map->usercnt, 1);
492 if (bpf_map_support_seq_show(map) &&
493 (attr->btf_key_type_id || attr->btf_value_type_id)) {
496 if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
498 goto free_map_nouncharge;
501 btf = btf_get_by_fd(attr->btf_fd);
504 goto free_map_nouncharge;
507 err = map->ops->map_check_btf(map, btf, attr->btf_key_type_id,
508 attr->btf_value_type_id);
511 goto free_map_nouncharge;
515 map->btf_key_type_id = attr->btf_key_type_id;
516 map->btf_value_type_id = attr->btf_value_type_id;
519 err = security_bpf_map_alloc(map);
521 goto free_map_nouncharge;
523 err = bpf_map_init_memlock(map);
527 err = bpf_map_alloc_id(map);
531 err = bpf_map_new_fd(map, f_flags);
533 /* failed to allocate fd.
534 * bpf_map_put() is needed because the above
535 * bpf_map_alloc_id() has published the map
536 * to the userspace and the userspace may
537 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
546 bpf_map_release_memlock(map);
548 security_bpf_map_free(map);
551 map->ops->map_free(map);
555 /* if error is returned, fd is released.
556 * On success caller should complete fd access with matching fdput()
558 struct bpf_map *__bpf_map_get(struct fd f)
561 return ERR_PTR(-EBADF);
562 if (f.file->f_op != &bpf_map_fops) {
564 return ERR_PTR(-EINVAL);
567 return f.file->private_data;
570 /* prog's and map's refcnt limit */
571 #define BPF_MAX_REFCNT 32768
573 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
575 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
576 atomic_dec(&map->refcnt);
577 return ERR_PTR(-EBUSY);
580 atomic_inc(&map->usercnt);
583 EXPORT_SYMBOL_GPL(bpf_map_inc);
585 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
587 struct fd f = fdget(ufd);
590 map = __bpf_map_get(f);
594 map = bpf_map_inc(map, true);
600 /* map_idr_lock should have been held */
601 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
606 refold = __atomic_add_unless(&map->refcnt, 1, 0);
608 if (refold >= BPF_MAX_REFCNT) {
609 __bpf_map_put(map, false);
610 return ERR_PTR(-EBUSY);
614 return ERR_PTR(-ENOENT);
617 atomic_inc(&map->usercnt);
622 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
627 /* last field in 'union bpf_attr' used by this command */
628 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
630 static int map_lookup_elem(union bpf_attr *attr)
632 void __user *ukey = u64_to_user_ptr(attr->key);
633 void __user *uvalue = u64_to_user_ptr(attr->value);
634 int ufd = attr->map_fd;
636 void *key, *value, *ptr;
641 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
645 map = __bpf_map_get(f);
649 if (!(f.file->f_mode & FMODE_CAN_READ)) {
654 key = memdup_user(ukey, map->key_size);
660 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
661 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
662 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
663 value_size = round_up(map->value_size, 8) * num_possible_cpus();
664 else if (IS_FD_MAP(map))
665 value_size = sizeof(u32);
667 value_size = map->value_size;
670 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
674 if (bpf_map_is_dev_bound(map)) {
675 err = bpf_map_offload_lookup_elem(map, key, value);
676 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
677 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
678 err = bpf_percpu_hash_copy(map, key, value);
679 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
680 err = bpf_percpu_array_copy(map, key, value);
681 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
682 err = bpf_stackmap_copy(map, key, value);
683 } else if (IS_FD_ARRAY(map)) {
684 err = bpf_fd_array_map_lookup_elem(map, key, value);
685 } else if (IS_FD_HASH(map)) {
686 err = bpf_fd_htab_map_lookup_elem(map, key, value);
689 ptr = map->ops->map_lookup_elem(map, key);
691 memcpy(value, ptr, value_size);
693 err = ptr ? 0 : -ENOENT;
700 if (copy_to_user(uvalue, value, value_size) != 0)
714 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
716 static int map_update_elem(union bpf_attr *attr)
718 void __user *ukey = u64_to_user_ptr(attr->key);
719 void __user *uvalue = u64_to_user_ptr(attr->value);
720 int ufd = attr->map_fd;
727 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
731 map = __bpf_map_get(f);
735 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
740 key = memdup_user(ukey, map->key_size);
746 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
747 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
748 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
749 value_size = round_up(map->value_size, 8) * num_possible_cpus();
751 value_size = map->value_size;
754 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
759 if (copy_from_user(value, uvalue, value_size) != 0)
762 /* Need to create a kthread, thus must support schedule */
763 if (bpf_map_is_dev_bound(map)) {
764 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
766 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
767 map->map_type == BPF_MAP_TYPE_SOCKHASH ||
768 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
769 err = map->ops->map_update_elem(map, key, value, attr->flags);
773 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
774 * inside bpf map update or delete otherwise deadlocks are possible
777 __this_cpu_inc(bpf_prog_active);
778 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
779 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
780 err = bpf_percpu_hash_update(map, key, value, attr->flags);
781 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
782 err = bpf_percpu_array_update(map, key, value, attr->flags);
783 } else if (IS_FD_ARRAY(map)) {
785 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
788 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
790 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
795 err = map->ops->map_update_elem(map, key, value, attr->flags);
798 __this_cpu_dec(bpf_prog_active);
810 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
812 static int map_delete_elem(union bpf_attr *attr)
814 void __user *ukey = u64_to_user_ptr(attr->key);
815 int ufd = attr->map_fd;
821 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
825 map = __bpf_map_get(f);
829 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
834 key = memdup_user(ukey, map->key_size);
840 if (bpf_map_is_dev_bound(map)) {
841 err = bpf_map_offload_delete_elem(map, key);
846 __this_cpu_inc(bpf_prog_active);
848 err = map->ops->map_delete_elem(map, key);
850 __this_cpu_dec(bpf_prog_active);
859 /* last field in 'union bpf_attr' used by this command */
860 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
862 static int map_get_next_key(union bpf_attr *attr)
864 void __user *ukey = u64_to_user_ptr(attr->key);
865 void __user *unext_key = u64_to_user_ptr(attr->next_key);
866 int ufd = attr->map_fd;
868 void *key, *next_key;
872 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
876 map = __bpf_map_get(f);
880 if (!(f.file->f_mode & FMODE_CAN_READ)) {
886 key = memdup_user(ukey, map->key_size);
896 next_key = kmalloc(map->key_size, GFP_USER);
900 if (bpf_map_is_dev_bound(map)) {
901 err = bpf_map_offload_get_next_key(map, key, next_key);
906 err = map->ops->map_get_next_key(map, key, next_key);
913 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
927 static const struct bpf_prog_ops * const bpf_prog_types[] = {
928 #define BPF_PROG_TYPE(_id, _name) \
929 [_id] = & _name ## _prog_ops,
930 #define BPF_MAP_TYPE(_id, _ops)
931 #include <linux/bpf_types.h>
936 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
938 const struct bpf_prog_ops *ops;
940 if (type >= ARRAY_SIZE(bpf_prog_types))
942 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
943 ops = bpf_prog_types[type];
947 if (!bpf_prog_is_dev_bound(prog->aux))
948 prog->aux->ops = ops;
950 prog->aux->ops = &bpf_offload_prog_ops;
955 /* drop refcnt on maps used by eBPF program and free auxilary data */
956 static void free_used_maps(struct bpf_prog_aux *aux)
960 for (i = 0; i < aux->used_map_cnt; i++)
961 bpf_map_put(aux->used_maps[i]);
963 kfree(aux->used_maps);
966 int __bpf_prog_charge(struct user_struct *user, u32 pages)
968 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
969 unsigned long user_bufs;
972 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
973 if (user_bufs > memlock_limit) {
974 atomic_long_sub(pages, &user->locked_vm);
982 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
985 atomic_long_sub(pages, &user->locked_vm);
988 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
990 struct user_struct *user = get_current_user();
993 ret = __bpf_prog_charge(user, prog->pages);
999 prog->aux->user = user;
1003 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1005 struct user_struct *user = prog->aux->user;
1007 __bpf_prog_uncharge(user, prog->pages);
1011 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1015 idr_preload(GFP_KERNEL);
1016 spin_lock_bh(&prog_idr_lock);
1017 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1020 spin_unlock_bh(&prog_idr_lock);
1023 /* id is in [1, INT_MAX) */
1024 if (WARN_ON_ONCE(!id))
1027 return id > 0 ? 0 : id;
1030 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1032 /* cBPF to eBPF migrations are currently not in the idr store.
1033 * Offloaded programs are removed from the store when their device
1034 * disappears - even if someone grabs an fd to them they are unusable,
1035 * simply waiting for refcnt to drop to be freed.
1041 spin_lock_bh(&prog_idr_lock);
1043 __acquire(&prog_idr_lock);
1045 idr_remove(&prog_idr, prog->aux->id);
1049 spin_unlock_bh(&prog_idr_lock);
1051 __release(&prog_idr_lock);
1054 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1056 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1058 free_used_maps(aux);
1059 bpf_prog_uncharge_memlock(aux->prog);
1060 security_bpf_prog_free(aux);
1061 bpf_prog_free(aux->prog);
1064 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1066 if (atomic_dec_and_test(&prog->aux->refcnt)) {
1067 /* bpf_prog_free_id() must be called first */
1068 bpf_prog_free_id(prog, do_idr_lock);
1069 bpf_prog_kallsyms_del_all(prog);
1071 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1075 void bpf_prog_put(struct bpf_prog *prog)
1077 __bpf_prog_put(prog, true);
1079 EXPORT_SYMBOL_GPL(bpf_prog_put);
1081 static int bpf_prog_release(struct inode *inode, struct file *filp)
1083 struct bpf_prog *prog = filp->private_data;
1089 #ifdef CONFIG_PROC_FS
1090 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1092 const struct bpf_prog *prog = filp->private_data;
1093 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1095 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1105 prog->pages * 1ULL << PAGE_SHIFT,
1110 const struct file_operations bpf_prog_fops = {
1111 #ifdef CONFIG_PROC_FS
1112 .show_fdinfo = bpf_prog_show_fdinfo,
1114 .release = bpf_prog_release,
1115 .read = bpf_dummy_read,
1116 .write = bpf_dummy_write,
1119 int bpf_prog_new_fd(struct bpf_prog *prog)
1123 ret = security_bpf_prog(prog);
1127 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1128 O_RDWR | O_CLOEXEC);
1131 static struct bpf_prog *____bpf_prog_get(struct fd f)
1134 return ERR_PTR(-EBADF);
1135 if (f.file->f_op != &bpf_prog_fops) {
1137 return ERR_PTR(-EINVAL);
1140 return f.file->private_data;
1143 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1145 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1146 atomic_sub(i, &prog->aux->refcnt);
1147 return ERR_PTR(-EBUSY);
1151 EXPORT_SYMBOL_GPL(bpf_prog_add);
1153 void bpf_prog_sub(struct bpf_prog *prog, int i)
1155 /* Only to be used for undoing previous bpf_prog_add() in some
1156 * error path. We still know that another entity in our call
1157 * path holds a reference to the program, thus atomic_sub() can
1158 * be safely used in such cases!
1160 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1162 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1164 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1166 return bpf_prog_add(prog, 1);
1168 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1170 /* prog_idr_lock should have been held */
1171 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1175 refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
1177 if (refold >= BPF_MAX_REFCNT) {
1178 __bpf_prog_put(prog, false);
1179 return ERR_PTR(-EBUSY);
1183 return ERR_PTR(-ENOENT);
1187 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1189 bool bpf_prog_get_ok(struct bpf_prog *prog,
1190 enum bpf_prog_type *attach_type, bool attach_drv)
1192 /* not an attachment, just a refcount inc, always allow */
1196 if (prog->type != *attach_type)
1198 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1204 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1207 struct fd f = fdget(ufd);
1208 struct bpf_prog *prog;
1210 prog = ____bpf_prog_get(f);
1213 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1214 prog = ERR_PTR(-EINVAL);
1218 prog = bpf_prog_inc(prog);
1224 struct bpf_prog *bpf_prog_get(u32 ufd)
1226 return __bpf_prog_get(ufd, NULL, false);
1229 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1232 return __bpf_prog_get(ufd, &type, attach_drv);
1234 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1236 /* Initially all BPF programs could be loaded w/o specifying
1237 * expected_attach_type. Later for some of them specifying expected_attach_type
1238 * at load time became required so that program could be validated properly.
1239 * Programs of types that are allowed to be loaded both w/ and w/o (for
1240 * backward compatibility) expected_attach_type, should have the default attach
1241 * type assigned to expected_attach_type for the latter case, so that it can be
1242 * validated later at attach time.
1244 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1245 * prog type requires it but has some attach types that have to be backward
1248 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1250 switch (attr->prog_type) {
1251 case BPF_PROG_TYPE_CGROUP_SOCK:
1252 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1253 * exist so checking for non-zero is the way to go here.
1255 if (!attr->expected_attach_type)
1256 attr->expected_attach_type =
1257 BPF_CGROUP_INET_SOCK_CREATE;
1263 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1264 enum bpf_attach_type expected_attach_type)
1266 switch (prog_type) {
1267 case BPF_PROG_TYPE_CGROUP_SOCK:
1268 switch (expected_attach_type) {
1269 case BPF_CGROUP_INET_SOCK_CREATE:
1270 case BPF_CGROUP_INET4_POST_BIND:
1271 case BPF_CGROUP_INET6_POST_BIND:
1276 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1277 switch (expected_attach_type) {
1278 case BPF_CGROUP_INET4_BIND:
1279 case BPF_CGROUP_INET6_BIND:
1280 case BPF_CGROUP_INET4_CONNECT:
1281 case BPF_CGROUP_INET6_CONNECT:
1282 case BPF_CGROUP_UDP4_SENDMSG:
1283 case BPF_CGROUP_UDP6_SENDMSG:
1293 /* last field in 'union bpf_attr' used by this command */
1294 #define BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1296 static int bpf_prog_load(union bpf_attr *attr)
1298 enum bpf_prog_type type = attr->prog_type;
1299 struct bpf_prog *prog;
1304 if (CHECK_ATTR(BPF_PROG_LOAD))
1307 if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1310 /* copy eBPF program license from user space */
1311 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1312 sizeof(license) - 1) < 0)
1314 license[sizeof(license) - 1] = 0;
1316 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1317 is_gpl = license_is_gpl_compatible(license);
1319 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1322 if (type == BPF_PROG_TYPE_KPROBE &&
1323 attr->kern_version != LINUX_VERSION_CODE)
1326 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1327 type != BPF_PROG_TYPE_CGROUP_SKB &&
1328 !capable(CAP_SYS_ADMIN))
1331 bpf_prog_load_fixup_attach_type(attr);
1332 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1335 /* plain bpf_prog allocation */
1336 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1340 prog->expected_attach_type = attr->expected_attach_type;
1342 prog->aux->offload_requested = !!attr->prog_ifindex;
1344 err = security_bpf_prog_alloc(prog->aux);
1346 goto free_prog_nouncharge;
1348 err = bpf_prog_charge_memlock(prog);
1352 prog->len = attr->insn_cnt;
1355 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1356 bpf_prog_insn_size(prog)) != 0)
1359 prog->orig_prog = NULL;
1362 atomic_set(&prog->aux->refcnt, 1);
1363 prog->gpl_compatible = is_gpl ? 1 : 0;
1365 if (bpf_prog_is_dev_bound(prog->aux)) {
1366 err = bpf_prog_offload_init(prog, attr);
1371 /* find program type: socket_filter vs tracing_filter */
1372 err = find_prog_type(type, prog);
1376 prog->aux->load_time = ktime_get_boot_ns();
1377 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1381 /* run eBPF verifier */
1382 err = bpf_check(&prog, attr);
1384 goto free_used_maps;
1386 prog = bpf_prog_select_runtime(prog, &err);
1388 goto free_used_maps;
1390 err = bpf_prog_alloc_id(prog);
1392 goto free_used_maps;
1394 err = bpf_prog_new_fd(prog);
1396 /* failed to allocate fd.
1397 * bpf_prog_put() is needed because the above
1398 * bpf_prog_alloc_id() has published the prog
1399 * to the userspace and the userspace may
1400 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1406 bpf_prog_kallsyms_add(prog);
1410 bpf_prog_kallsyms_del_subprogs(prog);
1411 free_used_maps(prog->aux);
1413 bpf_prog_uncharge_memlock(prog);
1415 security_bpf_prog_free(prog->aux);
1416 free_prog_nouncharge:
1417 bpf_prog_free(prog);
1421 #define BPF_OBJ_LAST_FIELD file_flags
1423 static int bpf_obj_pin(const union bpf_attr *attr)
1425 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1428 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1431 static int bpf_obj_get(const union bpf_attr *attr)
1433 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1434 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1437 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1441 struct bpf_raw_tracepoint {
1442 struct bpf_raw_event_map *btp;
1443 struct bpf_prog *prog;
1446 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1448 struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1451 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1452 bpf_prog_put(raw_tp->prog);
1458 static const struct file_operations bpf_raw_tp_fops = {
1459 .release = bpf_raw_tracepoint_release,
1460 .read = bpf_dummy_read,
1461 .write = bpf_dummy_write,
1464 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1466 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1468 struct bpf_raw_tracepoint *raw_tp;
1469 struct bpf_raw_event_map *btp;
1470 struct bpf_prog *prog;
1474 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1475 sizeof(tp_name) - 1) < 0)
1477 tp_name[sizeof(tp_name) - 1] = 0;
1479 btp = bpf_find_raw_tracepoint(tp_name);
1483 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1488 prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1489 BPF_PROG_TYPE_RAW_TRACEPOINT);
1491 err = PTR_ERR(prog);
1495 err = bpf_probe_register(raw_tp->btp, prog);
1499 raw_tp->prog = prog;
1500 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1503 bpf_probe_unregister(raw_tp->btp, prog);
1516 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1517 enum bpf_attach_type attach_type)
1519 switch (prog->type) {
1520 case BPF_PROG_TYPE_CGROUP_SOCK:
1521 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1522 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1528 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1530 #define BPF_F_ATTACH_MASK \
1531 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1533 static int bpf_prog_attach(const union bpf_attr *attr)
1535 enum bpf_prog_type ptype;
1536 struct bpf_prog *prog;
1539 if (!capable(CAP_NET_ADMIN))
1542 if (CHECK_ATTR(BPF_PROG_ATTACH))
1545 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1548 switch (attr->attach_type) {
1549 case BPF_CGROUP_INET_INGRESS:
1550 case BPF_CGROUP_INET_EGRESS:
1551 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1553 case BPF_CGROUP_INET_SOCK_CREATE:
1554 case BPF_CGROUP_INET4_POST_BIND:
1555 case BPF_CGROUP_INET6_POST_BIND:
1556 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1558 case BPF_CGROUP_INET4_BIND:
1559 case BPF_CGROUP_INET6_BIND:
1560 case BPF_CGROUP_INET4_CONNECT:
1561 case BPF_CGROUP_INET6_CONNECT:
1562 case BPF_CGROUP_UDP4_SENDMSG:
1563 case BPF_CGROUP_UDP6_SENDMSG:
1564 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1566 case BPF_CGROUP_SOCK_OPS:
1567 ptype = BPF_PROG_TYPE_SOCK_OPS;
1569 case BPF_CGROUP_DEVICE:
1570 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1572 case BPF_SK_MSG_VERDICT:
1573 ptype = BPF_PROG_TYPE_SK_MSG;
1575 case BPF_SK_SKB_STREAM_PARSER:
1576 case BPF_SK_SKB_STREAM_VERDICT:
1577 ptype = BPF_PROG_TYPE_SK_SKB;
1579 case BPF_LIRC_MODE2:
1580 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1586 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1588 return PTR_ERR(prog);
1590 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1596 case BPF_PROG_TYPE_SK_SKB:
1597 case BPF_PROG_TYPE_SK_MSG:
1598 ret = sockmap_get_from_fd(attr, ptype, prog);
1600 case BPF_PROG_TYPE_LIRC_MODE2:
1601 ret = lirc_prog_attach(attr, prog);
1604 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1612 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1614 static int bpf_prog_detach(const union bpf_attr *attr)
1616 enum bpf_prog_type ptype;
1618 if (!capable(CAP_NET_ADMIN))
1621 if (CHECK_ATTR(BPF_PROG_DETACH))
1624 switch (attr->attach_type) {
1625 case BPF_CGROUP_INET_INGRESS:
1626 case BPF_CGROUP_INET_EGRESS:
1627 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1629 case BPF_CGROUP_INET_SOCK_CREATE:
1630 case BPF_CGROUP_INET4_POST_BIND:
1631 case BPF_CGROUP_INET6_POST_BIND:
1632 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1634 case BPF_CGROUP_INET4_BIND:
1635 case BPF_CGROUP_INET6_BIND:
1636 case BPF_CGROUP_INET4_CONNECT:
1637 case BPF_CGROUP_INET6_CONNECT:
1638 case BPF_CGROUP_UDP4_SENDMSG:
1639 case BPF_CGROUP_UDP6_SENDMSG:
1640 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1642 case BPF_CGROUP_SOCK_OPS:
1643 ptype = BPF_PROG_TYPE_SOCK_OPS;
1645 case BPF_CGROUP_DEVICE:
1646 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1648 case BPF_SK_MSG_VERDICT:
1649 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, NULL);
1650 case BPF_SK_SKB_STREAM_PARSER:
1651 case BPF_SK_SKB_STREAM_VERDICT:
1652 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, NULL);
1653 case BPF_LIRC_MODE2:
1654 return lirc_prog_detach(attr);
1659 return cgroup_bpf_prog_detach(attr, ptype);
1662 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1664 static int bpf_prog_query(const union bpf_attr *attr,
1665 union bpf_attr __user *uattr)
1667 if (!capable(CAP_NET_ADMIN))
1669 if (CHECK_ATTR(BPF_PROG_QUERY))
1671 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1674 switch (attr->query.attach_type) {
1675 case BPF_CGROUP_INET_INGRESS:
1676 case BPF_CGROUP_INET_EGRESS:
1677 case BPF_CGROUP_INET_SOCK_CREATE:
1678 case BPF_CGROUP_INET4_BIND:
1679 case BPF_CGROUP_INET6_BIND:
1680 case BPF_CGROUP_INET4_POST_BIND:
1681 case BPF_CGROUP_INET6_POST_BIND:
1682 case BPF_CGROUP_INET4_CONNECT:
1683 case BPF_CGROUP_INET6_CONNECT:
1684 case BPF_CGROUP_UDP4_SENDMSG:
1685 case BPF_CGROUP_UDP6_SENDMSG:
1686 case BPF_CGROUP_SOCK_OPS:
1687 case BPF_CGROUP_DEVICE:
1689 case BPF_LIRC_MODE2:
1690 return lirc_prog_query(attr, uattr);
1695 return cgroup_bpf_prog_query(attr, uattr);
1698 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1700 static int bpf_prog_test_run(const union bpf_attr *attr,
1701 union bpf_attr __user *uattr)
1703 struct bpf_prog *prog;
1704 int ret = -ENOTSUPP;
1706 if (!capable(CAP_SYS_ADMIN))
1708 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1711 prog = bpf_prog_get(attr->test.prog_fd);
1713 return PTR_ERR(prog);
1715 if (prog->aux->ops->test_run)
1716 ret = prog->aux->ops->test_run(prog, attr, uattr);
1722 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1724 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1725 union bpf_attr __user *uattr,
1729 u32 next_id = attr->start_id;
1732 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1735 if (!capable(CAP_SYS_ADMIN))
1740 if (!idr_get_next(idr, &next_id))
1742 spin_unlock_bh(lock);
1745 err = put_user(next_id, &uattr->next_id);
1750 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1752 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1754 struct bpf_prog *prog;
1755 u32 id = attr->prog_id;
1758 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1761 if (!capable(CAP_SYS_ADMIN))
1764 spin_lock_bh(&prog_idr_lock);
1765 prog = idr_find(&prog_idr, id);
1767 prog = bpf_prog_inc_not_zero(prog);
1769 prog = ERR_PTR(-ENOENT);
1770 spin_unlock_bh(&prog_idr_lock);
1773 return PTR_ERR(prog);
1775 fd = bpf_prog_new_fd(prog);
1782 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1784 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1786 struct bpf_map *map;
1787 u32 id = attr->map_id;
1791 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1792 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1795 if (!capable(CAP_SYS_ADMIN))
1798 f_flags = bpf_get_file_flag(attr->open_flags);
1802 spin_lock_bh(&map_idr_lock);
1803 map = idr_find(&map_idr, id);
1805 map = bpf_map_inc_not_zero(map, true);
1807 map = ERR_PTR(-ENOENT);
1808 spin_unlock_bh(&map_idr_lock);
1811 return PTR_ERR(map);
1813 fd = bpf_map_new_fd(map, f_flags);
1820 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1825 for (i = 0; i < prog->aux->used_map_cnt; i++)
1826 if (prog->aux->used_maps[i] == (void *)addr)
1827 return prog->aux->used_maps[i];
1831 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1833 const struct bpf_map *map;
1834 struct bpf_insn *insns;
1838 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1843 for (i = 0; i < prog->len; i++) {
1844 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1845 insns[i].code = BPF_JMP | BPF_CALL;
1846 insns[i].imm = BPF_FUNC_tail_call;
1849 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1850 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1851 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1852 insns[i].code = BPF_JMP | BPF_CALL;
1853 if (!bpf_dump_raw_ok())
1858 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1861 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1862 map = bpf_map_from_imm(prog, imm);
1864 insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1865 insns[i].imm = map->id;
1866 insns[i + 1].imm = 0;
1870 if (!bpf_dump_raw_ok() &&
1871 imm == (unsigned long)prog->aux) {
1873 insns[i + 1].imm = 0;
1881 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1882 const union bpf_attr *attr,
1883 union bpf_attr __user *uattr)
1885 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1886 struct bpf_prog_info info = {};
1887 u32 info_len = attr->info.info_len;
1888 char __user *uinsns;
1892 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1895 info_len = min_t(u32, sizeof(info), info_len);
1897 if (copy_from_user(&info, uinfo, info_len))
1900 info.type = prog->type;
1901 info.id = prog->aux->id;
1902 info.load_time = prog->aux->load_time;
1903 info.created_by_uid = from_kuid_munged(current_user_ns(),
1904 prog->aux->user->uid);
1905 info.gpl_compatible = prog->gpl_compatible;
1907 memcpy(info.tag, prog->tag, sizeof(prog->tag));
1908 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1910 ulen = info.nr_map_ids;
1911 info.nr_map_ids = prog->aux->used_map_cnt;
1912 ulen = min_t(u32, info.nr_map_ids, ulen);
1914 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1917 for (i = 0; i < ulen; i++)
1918 if (put_user(prog->aux->used_maps[i]->id,
1923 if (!capable(CAP_SYS_ADMIN)) {
1924 info.jited_prog_len = 0;
1925 info.xlated_prog_len = 0;
1926 info.nr_jited_ksyms = 0;
1930 ulen = info.xlated_prog_len;
1931 info.xlated_prog_len = bpf_prog_insn_size(prog);
1932 if (info.xlated_prog_len && ulen) {
1933 struct bpf_insn *insns_sanitized;
1936 if (prog->blinded && !bpf_dump_raw_ok()) {
1937 info.xlated_prog_insns = 0;
1940 insns_sanitized = bpf_insn_prepare_dump(prog);
1941 if (!insns_sanitized)
1943 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1944 ulen = min_t(u32, info.xlated_prog_len, ulen);
1945 fault = copy_to_user(uinsns, insns_sanitized, ulen);
1946 kfree(insns_sanitized);
1951 if (bpf_prog_is_dev_bound(prog->aux)) {
1952 err = bpf_prog_offload_info_fill(&info, prog);
1958 /* NOTE: the following code is supposed to be skipped for offload.
1959 * bpf_prog_offload_info_fill() is the place to fill similar fields
1962 ulen = info.jited_prog_len;
1963 if (prog->aux->func_cnt) {
1966 info.jited_prog_len = 0;
1967 for (i = 0; i < prog->aux->func_cnt; i++)
1968 info.jited_prog_len += prog->aux->func[i]->jited_len;
1970 info.jited_prog_len = prog->jited_len;
1973 if (info.jited_prog_len && ulen) {
1974 if (bpf_dump_raw_ok()) {
1975 uinsns = u64_to_user_ptr(info.jited_prog_insns);
1976 ulen = min_t(u32, info.jited_prog_len, ulen);
1978 /* for multi-function programs, copy the JITed
1979 * instructions for all the functions
1981 if (prog->aux->func_cnt) {
1986 for (i = 0; i < prog->aux->func_cnt; i++) {
1987 len = prog->aux->func[i]->jited_len;
1988 len = min_t(u32, len, free);
1989 img = (u8 *) prog->aux->func[i]->bpf_func;
1990 if (copy_to_user(uinsns, img, len))
1998 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2002 info.jited_prog_insns = 0;
2006 ulen = info.nr_jited_ksyms;
2007 info.nr_jited_ksyms = prog->aux->func_cnt;
2008 if (info.nr_jited_ksyms && ulen) {
2009 if (bpf_dump_raw_ok()) {
2010 u64 __user *user_ksyms;
2014 /* copy the address of the kernel symbol
2015 * corresponding to each function
2017 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2018 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2019 for (i = 0; i < ulen; i++) {
2020 ksym_addr = (ulong) prog->aux->func[i]->bpf_func;
2021 ksym_addr &= PAGE_MASK;
2022 if (put_user((u64) ksym_addr, &user_ksyms[i]))
2026 info.jited_ksyms = 0;
2030 ulen = info.nr_jited_func_lens;
2031 info.nr_jited_func_lens = prog->aux->func_cnt;
2032 if (info.nr_jited_func_lens && ulen) {
2033 if (bpf_dump_raw_ok()) {
2034 u32 __user *user_lens;
2037 /* copy the JITed image lengths for each function */
2038 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2039 user_lens = u64_to_user_ptr(info.jited_func_lens);
2040 for (i = 0; i < ulen; i++) {
2041 func_len = prog->aux->func[i]->jited_len;
2042 if (put_user(func_len, &user_lens[i]))
2046 info.jited_func_lens = 0;
2051 if (copy_to_user(uinfo, &info, info_len) ||
2052 put_user(info_len, &uattr->info.info_len))
2058 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2059 const union bpf_attr *attr,
2060 union bpf_attr __user *uattr)
2062 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2063 struct bpf_map_info info = {};
2064 u32 info_len = attr->info.info_len;
2067 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2070 info_len = min_t(u32, sizeof(info), info_len);
2072 info.type = map->map_type;
2074 info.key_size = map->key_size;
2075 info.value_size = map->value_size;
2076 info.max_entries = map->max_entries;
2077 info.map_flags = map->map_flags;
2078 memcpy(info.name, map->name, sizeof(map->name));
2081 info.btf_id = btf_id(map->btf);
2082 info.btf_key_type_id = map->btf_key_type_id;
2083 info.btf_value_type_id = map->btf_value_type_id;
2086 if (bpf_map_is_dev_bound(map)) {
2087 err = bpf_map_offload_info_fill(&info, map);
2092 if (copy_to_user(uinfo, &info, info_len) ||
2093 put_user(info_len, &uattr->info.info_len))
2099 static int bpf_btf_get_info_by_fd(struct btf *btf,
2100 const union bpf_attr *attr,
2101 union bpf_attr __user *uattr)
2103 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2104 u32 info_len = attr->info.info_len;
2107 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2111 return btf_get_info_by_fd(btf, attr, uattr);
2114 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2116 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2117 union bpf_attr __user *uattr)
2119 int ufd = attr->info.bpf_fd;
2123 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2130 if (f.file->f_op == &bpf_prog_fops)
2131 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2133 else if (f.file->f_op == &bpf_map_fops)
2134 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2136 else if (f.file->f_op == &btf_fops)
2137 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2145 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2147 static int bpf_btf_load(const union bpf_attr *attr)
2149 if (CHECK_ATTR(BPF_BTF_LOAD))
2152 if (!capable(CAP_SYS_ADMIN))
2155 return btf_new_fd(attr);
2158 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2160 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2162 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2165 if (!capable(CAP_SYS_ADMIN))
2168 return btf_get_fd_by_id(attr->btf_id);
2171 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2172 union bpf_attr __user *uattr,
2173 u32 prog_id, u32 fd_type,
2174 const char *buf, u64 probe_offset,
2177 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2178 u32 len = buf ? strlen(buf) : 0, input_len;
2181 if (put_user(len, &uattr->task_fd_query.buf_len))
2183 input_len = attr->task_fd_query.buf_len;
2184 if (input_len && ubuf) {
2186 /* nothing to copy, just make ubuf NULL terminated */
2189 if (put_user(zero, ubuf))
2191 } else if (input_len >= len + 1) {
2192 /* ubuf can hold the string with NULL terminator */
2193 if (copy_to_user(ubuf, buf, len + 1))
2196 /* ubuf cannot hold the string with NULL terminator,
2197 * do a partial copy with NULL terminator.
2202 if (copy_to_user(ubuf, buf, input_len - 1))
2204 if (put_user(zero, ubuf + input_len - 1))
2209 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2210 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2211 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2212 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2218 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2220 static int bpf_task_fd_query(const union bpf_attr *attr,
2221 union bpf_attr __user *uattr)
2223 pid_t pid = attr->task_fd_query.pid;
2224 u32 fd = attr->task_fd_query.fd;
2225 const struct perf_event *event;
2226 struct files_struct *files;
2227 struct task_struct *task;
2231 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2234 if (!capable(CAP_SYS_ADMIN))
2237 if (attr->task_fd_query.flags != 0)
2240 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2244 files = get_files_struct(task);
2245 put_task_struct(task);
2250 spin_lock(&files->file_lock);
2251 file = fcheck_files(files, fd);
2256 spin_unlock(&files->file_lock);
2257 put_files_struct(files);
2262 if (file->f_op == &bpf_raw_tp_fops) {
2263 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2264 struct bpf_raw_event_map *btp = raw_tp->btp;
2266 err = bpf_task_fd_query_copy(attr, uattr,
2267 raw_tp->prog->aux->id,
2268 BPF_FD_TYPE_RAW_TRACEPOINT,
2269 btp->tp->name, 0, 0);
2273 event = perf_get_event(file);
2274 if (!IS_ERR(event)) {
2275 u64 probe_offset, probe_addr;
2276 u32 prog_id, fd_type;
2279 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2280 &buf, &probe_offset,
2283 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2297 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2299 union bpf_attr attr = {};
2302 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2305 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2308 size = min_t(u32, size, sizeof(attr));
2310 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2311 if (copy_from_user(&attr, uattr, size) != 0)
2314 err = security_bpf(cmd, &attr, size);
2319 case BPF_MAP_CREATE:
2320 err = map_create(&attr);
2322 case BPF_MAP_LOOKUP_ELEM:
2323 err = map_lookup_elem(&attr);
2325 case BPF_MAP_UPDATE_ELEM:
2326 err = map_update_elem(&attr);
2328 case BPF_MAP_DELETE_ELEM:
2329 err = map_delete_elem(&attr);
2331 case BPF_MAP_GET_NEXT_KEY:
2332 err = map_get_next_key(&attr);
2335 err = bpf_prog_load(&attr);
2338 err = bpf_obj_pin(&attr);
2341 err = bpf_obj_get(&attr);
2343 case BPF_PROG_ATTACH:
2344 err = bpf_prog_attach(&attr);
2346 case BPF_PROG_DETACH:
2347 err = bpf_prog_detach(&attr);
2349 case BPF_PROG_QUERY:
2350 err = bpf_prog_query(&attr, uattr);
2352 case BPF_PROG_TEST_RUN:
2353 err = bpf_prog_test_run(&attr, uattr);
2355 case BPF_PROG_GET_NEXT_ID:
2356 err = bpf_obj_get_next_id(&attr, uattr,
2357 &prog_idr, &prog_idr_lock);
2359 case BPF_MAP_GET_NEXT_ID:
2360 err = bpf_obj_get_next_id(&attr, uattr,
2361 &map_idr, &map_idr_lock);
2363 case BPF_PROG_GET_FD_BY_ID:
2364 err = bpf_prog_get_fd_by_id(&attr);
2366 case BPF_MAP_GET_FD_BY_ID:
2367 err = bpf_map_get_fd_by_id(&attr);
2369 case BPF_OBJ_GET_INFO_BY_FD:
2370 err = bpf_obj_get_info_by_fd(&attr, uattr);
2372 case BPF_RAW_TRACEPOINT_OPEN:
2373 err = bpf_raw_tracepoint_open(&attr);
2376 err = bpf_btf_load(&attr);
2378 case BPF_BTF_GET_FD_BY_ID:
2379 err = bpf_btf_get_fd_by_id(&attr);
2381 case BPF_TASK_FD_QUERY:
2382 err = bpf_task_fd_query(&attr, uattr);