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_map_charge_memlock(struct bpf_map *map)
186 struct user_struct *user = get_current_user();
187 unsigned long memlock_limit;
189 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
191 atomic_long_add(map->pages, &user->locked_vm);
193 if (atomic_long_read(&user->locked_vm) > memlock_limit) {
194 atomic_long_sub(map->pages, &user->locked_vm);
202 static void bpf_map_uncharge_memlock(struct bpf_map *map)
204 struct user_struct *user = map->user;
206 atomic_long_sub(map->pages, &user->locked_vm);
210 static int bpf_map_alloc_id(struct bpf_map *map)
214 idr_preload(GFP_KERNEL);
215 spin_lock_bh(&map_idr_lock);
216 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
219 spin_unlock_bh(&map_idr_lock);
222 if (WARN_ON_ONCE(!id))
225 return id > 0 ? 0 : id;
228 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
232 /* Offloaded maps are removed from the IDR store when their device
233 * disappears - even if someone holds an fd to them they are unusable,
234 * the memory is gone, all ops will fail; they are simply waiting for
235 * refcnt to drop to be freed.
241 spin_lock_irqsave(&map_idr_lock, flags);
243 __acquire(&map_idr_lock);
245 idr_remove(&map_idr, map->id);
249 spin_unlock_irqrestore(&map_idr_lock, flags);
251 __release(&map_idr_lock);
254 /* called from workqueue */
255 static void bpf_map_free_deferred(struct work_struct *work)
257 struct bpf_map *map = container_of(work, struct bpf_map, work);
259 bpf_map_uncharge_memlock(map);
260 security_bpf_map_free(map);
261 /* implementation dependent freeing */
262 map->ops->map_free(map);
265 static void bpf_map_put_uref(struct bpf_map *map)
267 if (atomic_dec_and_test(&map->usercnt)) {
268 if (map->ops->map_release_uref)
269 map->ops->map_release_uref(map);
273 /* decrement map refcnt and schedule it for freeing via workqueue
274 * (unrelying map implementation ops->map_free() might sleep)
276 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
278 if (atomic_dec_and_test(&map->refcnt)) {
279 /* bpf_map_free_id() must be called first */
280 bpf_map_free_id(map, do_idr_lock);
282 INIT_WORK(&map->work, bpf_map_free_deferred);
283 schedule_work(&map->work);
287 void bpf_map_put(struct bpf_map *map)
289 __bpf_map_put(map, true);
291 EXPORT_SYMBOL_GPL(bpf_map_put);
293 void bpf_map_put_with_uref(struct bpf_map *map)
295 bpf_map_put_uref(map);
299 static int bpf_map_release(struct inode *inode, struct file *filp)
301 struct bpf_map *map = filp->private_data;
303 if (map->ops->map_release)
304 map->ops->map_release(map, filp);
306 bpf_map_put_with_uref(map);
310 #ifdef CONFIG_PROC_FS
311 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
313 const struct bpf_map *map = filp->private_data;
314 const struct bpf_array *array;
315 u32 owner_prog_type = 0;
318 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
319 array = container_of(map, struct bpf_array, map);
320 owner_prog_type = array->owner_prog_type;
321 owner_jited = array->owner_jited;
337 map->pages * 1ULL << PAGE_SHIFT,
340 if (owner_prog_type) {
341 seq_printf(m, "owner_prog_type:\t%u\n",
343 seq_printf(m, "owner_jited:\t%u\n",
349 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
352 /* We need this handler such that alloc_file() enables
353 * f_mode with FMODE_CAN_READ.
358 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
359 size_t siz, loff_t *ppos)
361 /* We need this handler such that alloc_file() enables
362 * f_mode with FMODE_CAN_WRITE.
367 const struct file_operations bpf_map_fops = {
368 #ifdef CONFIG_PROC_FS
369 .show_fdinfo = bpf_map_show_fdinfo,
371 .release = bpf_map_release,
372 .read = bpf_dummy_read,
373 .write = bpf_dummy_write,
376 int bpf_map_new_fd(struct bpf_map *map, int flags)
380 ret = security_bpf_map(map, OPEN_FMODE(flags));
384 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
388 int bpf_get_file_flag(int flags)
390 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
392 if (flags & BPF_F_RDONLY)
394 if (flags & BPF_F_WRONLY)
399 /* helper macro to check that unused fields 'union bpf_attr' are zero */
400 #define CHECK_ATTR(CMD) \
401 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
402 sizeof(attr->CMD##_LAST_FIELD), 0, \
404 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
405 sizeof(attr->CMD##_LAST_FIELD)) != NULL
407 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
408 * Return 0 on success and < 0 on error.
410 static int bpf_obj_name_cpy(char *dst, const char *src)
412 const char *end = src + BPF_OBJ_NAME_LEN;
414 memset(dst, 0, BPF_OBJ_NAME_LEN);
416 /* Copy all isalnum() and '_' char */
417 while (src < end && *src) {
418 if (!isalnum(*src) && *src != '_')
423 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
430 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
431 /* called via syscall */
432 static int map_create(union bpf_attr *attr)
434 int numa_node = bpf_map_attr_numa_node(attr);
439 err = CHECK_ATTR(BPF_MAP_CREATE);
443 f_flags = bpf_get_file_flag(attr->map_flags);
447 if (numa_node != NUMA_NO_NODE &&
448 ((unsigned int)numa_node >= nr_node_ids ||
449 !node_online(numa_node)))
452 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
453 map = find_and_alloc_map(attr);
457 err = bpf_obj_name_cpy(map->name, attr->map_name);
459 goto free_map_nouncharge;
461 atomic_set(&map->refcnt, 1);
462 atomic_set(&map->usercnt, 1);
464 if (bpf_map_support_seq_show(map) &&
465 (attr->btf_key_type_id || attr->btf_value_type_id)) {
468 if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
470 goto free_map_nouncharge;
473 btf = btf_get_by_fd(attr->btf_fd);
476 goto free_map_nouncharge;
479 err = map->ops->map_check_btf(map, btf, attr->btf_key_type_id,
480 attr->btf_value_type_id);
483 goto free_map_nouncharge;
487 map->btf_key_type_id = attr->btf_key_type_id;
488 map->btf_value_type_id = attr->btf_value_type_id;
491 err = security_bpf_map_alloc(map);
493 goto free_map_nouncharge;
495 err = bpf_map_charge_memlock(map);
499 err = bpf_map_alloc_id(map);
503 err = bpf_map_new_fd(map, f_flags);
505 /* failed to allocate fd.
506 * bpf_map_put() is needed because the above
507 * bpf_map_alloc_id() has published the map
508 * to the userspace and the userspace may
509 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
518 bpf_map_uncharge_memlock(map);
520 security_bpf_map_free(map);
523 map->ops->map_free(map);
527 /* if error is returned, fd is released.
528 * On success caller should complete fd access with matching fdput()
530 struct bpf_map *__bpf_map_get(struct fd f)
533 return ERR_PTR(-EBADF);
534 if (f.file->f_op != &bpf_map_fops) {
536 return ERR_PTR(-EINVAL);
539 return f.file->private_data;
542 /* prog's and map's refcnt limit */
543 #define BPF_MAX_REFCNT 32768
545 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
547 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
548 atomic_dec(&map->refcnt);
549 return ERR_PTR(-EBUSY);
552 atomic_inc(&map->usercnt);
555 EXPORT_SYMBOL_GPL(bpf_map_inc);
557 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
559 struct fd f = fdget(ufd);
562 map = __bpf_map_get(f);
566 map = bpf_map_inc(map, true);
572 /* map_idr_lock should have been held */
573 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
578 refold = __atomic_add_unless(&map->refcnt, 1, 0);
580 if (refold >= BPF_MAX_REFCNT) {
581 __bpf_map_put(map, false);
582 return ERR_PTR(-EBUSY);
586 return ERR_PTR(-ENOENT);
589 atomic_inc(&map->usercnt);
594 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
599 /* last field in 'union bpf_attr' used by this command */
600 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
602 static int map_lookup_elem(union bpf_attr *attr)
604 void __user *ukey = u64_to_user_ptr(attr->key);
605 void __user *uvalue = u64_to_user_ptr(attr->value);
606 int ufd = attr->map_fd;
608 void *key, *value, *ptr;
613 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
617 map = __bpf_map_get(f);
621 if (!(f.file->f_mode & FMODE_CAN_READ)) {
626 key = memdup_user(ukey, map->key_size);
632 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
633 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
634 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
635 value_size = round_up(map->value_size, 8) * num_possible_cpus();
636 else if (IS_FD_MAP(map))
637 value_size = sizeof(u32);
639 value_size = map->value_size;
642 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
646 if (bpf_map_is_dev_bound(map)) {
647 err = bpf_map_offload_lookup_elem(map, key, value);
648 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
649 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
650 err = bpf_percpu_hash_copy(map, key, value);
651 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
652 err = bpf_percpu_array_copy(map, key, value);
653 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
654 err = bpf_stackmap_copy(map, key, value);
655 } else if (IS_FD_ARRAY(map)) {
656 err = bpf_fd_array_map_lookup_elem(map, key, value);
657 } else if (IS_FD_HASH(map)) {
658 err = bpf_fd_htab_map_lookup_elem(map, key, value);
661 ptr = map->ops->map_lookup_elem(map, key);
663 memcpy(value, ptr, value_size);
665 err = ptr ? 0 : -ENOENT;
672 if (copy_to_user(uvalue, value, value_size) != 0)
686 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
688 static int map_update_elem(union bpf_attr *attr)
690 void __user *ukey = u64_to_user_ptr(attr->key);
691 void __user *uvalue = u64_to_user_ptr(attr->value);
692 int ufd = attr->map_fd;
699 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
703 map = __bpf_map_get(f);
707 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
712 key = memdup_user(ukey, map->key_size);
718 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
719 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
720 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
721 value_size = round_up(map->value_size, 8) * num_possible_cpus();
723 value_size = map->value_size;
726 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
731 if (copy_from_user(value, uvalue, value_size) != 0)
734 /* Need to create a kthread, thus must support schedule */
735 if (bpf_map_is_dev_bound(map)) {
736 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
738 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
739 map->map_type == BPF_MAP_TYPE_SOCKHASH ||
740 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
741 err = map->ops->map_update_elem(map, key, value, attr->flags);
745 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
746 * inside bpf map update or delete otherwise deadlocks are possible
749 __this_cpu_inc(bpf_prog_active);
750 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
751 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
752 err = bpf_percpu_hash_update(map, key, value, attr->flags);
753 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
754 err = bpf_percpu_array_update(map, key, value, attr->flags);
755 } else if (IS_FD_ARRAY(map)) {
757 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
760 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
762 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
767 err = map->ops->map_update_elem(map, key, value, attr->flags);
770 __this_cpu_dec(bpf_prog_active);
782 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
784 static int map_delete_elem(union bpf_attr *attr)
786 void __user *ukey = u64_to_user_ptr(attr->key);
787 int ufd = attr->map_fd;
793 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
797 map = __bpf_map_get(f);
801 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
806 key = memdup_user(ukey, map->key_size);
812 if (bpf_map_is_dev_bound(map)) {
813 err = bpf_map_offload_delete_elem(map, key);
818 __this_cpu_inc(bpf_prog_active);
820 err = map->ops->map_delete_elem(map, key);
822 __this_cpu_dec(bpf_prog_active);
831 /* last field in 'union bpf_attr' used by this command */
832 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
834 static int map_get_next_key(union bpf_attr *attr)
836 void __user *ukey = u64_to_user_ptr(attr->key);
837 void __user *unext_key = u64_to_user_ptr(attr->next_key);
838 int ufd = attr->map_fd;
840 void *key, *next_key;
844 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
848 map = __bpf_map_get(f);
852 if (!(f.file->f_mode & FMODE_CAN_READ)) {
858 key = memdup_user(ukey, map->key_size);
868 next_key = kmalloc(map->key_size, GFP_USER);
872 if (bpf_map_is_dev_bound(map)) {
873 err = bpf_map_offload_get_next_key(map, key, next_key);
878 err = map->ops->map_get_next_key(map, key, next_key);
885 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
899 static const struct bpf_prog_ops * const bpf_prog_types[] = {
900 #define BPF_PROG_TYPE(_id, _name) \
901 [_id] = & _name ## _prog_ops,
902 #define BPF_MAP_TYPE(_id, _ops)
903 #include <linux/bpf_types.h>
908 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
910 const struct bpf_prog_ops *ops;
912 if (type >= ARRAY_SIZE(bpf_prog_types))
914 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
915 ops = bpf_prog_types[type];
919 if (!bpf_prog_is_dev_bound(prog->aux))
920 prog->aux->ops = ops;
922 prog->aux->ops = &bpf_offload_prog_ops;
927 /* drop refcnt on maps used by eBPF program and free auxilary data */
928 static void free_used_maps(struct bpf_prog_aux *aux)
932 for (i = 0; i < aux->used_map_cnt; i++)
933 bpf_map_put(aux->used_maps[i]);
935 kfree(aux->used_maps);
938 int __bpf_prog_charge(struct user_struct *user, u32 pages)
940 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
941 unsigned long user_bufs;
944 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
945 if (user_bufs > memlock_limit) {
946 atomic_long_sub(pages, &user->locked_vm);
954 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
957 atomic_long_sub(pages, &user->locked_vm);
960 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
962 struct user_struct *user = get_current_user();
965 ret = __bpf_prog_charge(user, prog->pages);
971 prog->aux->user = user;
975 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
977 struct user_struct *user = prog->aux->user;
979 __bpf_prog_uncharge(user, prog->pages);
983 static int bpf_prog_alloc_id(struct bpf_prog *prog)
987 idr_preload(GFP_KERNEL);
988 spin_lock_bh(&prog_idr_lock);
989 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
992 spin_unlock_bh(&prog_idr_lock);
995 /* id is in [1, INT_MAX) */
996 if (WARN_ON_ONCE(!id))
999 return id > 0 ? 0 : id;
1002 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1004 /* cBPF to eBPF migrations are currently not in the idr store.
1005 * Offloaded programs are removed from the store when their device
1006 * disappears - even if someone grabs an fd to them they are unusable,
1007 * simply waiting for refcnt to drop to be freed.
1013 spin_lock_bh(&prog_idr_lock);
1015 __acquire(&prog_idr_lock);
1017 idr_remove(&prog_idr, prog->aux->id);
1021 spin_unlock_bh(&prog_idr_lock);
1023 __release(&prog_idr_lock);
1026 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1028 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1030 free_used_maps(aux);
1031 bpf_prog_uncharge_memlock(aux->prog);
1032 security_bpf_prog_free(aux);
1033 bpf_prog_free(aux->prog);
1036 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1038 if (atomic_dec_and_test(&prog->aux->refcnt)) {
1039 /* bpf_prog_free_id() must be called first */
1040 bpf_prog_free_id(prog, do_idr_lock);
1041 bpf_prog_kallsyms_del_all(prog);
1043 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1047 void bpf_prog_put(struct bpf_prog *prog)
1049 __bpf_prog_put(prog, true);
1051 EXPORT_SYMBOL_GPL(bpf_prog_put);
1053 static int bpf_prog_release(struct inode *inode, struct file *filp)
1055 struct bpf_prog *prog = filp->private_data;
1061 #ifdef CONFIG_PROC_FS
1062 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1064 const struct bpf_prog *prog = filp->private_data;
1065 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1067 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1077 prog->pages * 1ULL << PAGE_SHIFT,
1082 const struct file_operations bpf_prog_fops = {
1083 #ifdef CONFIG_PROC_FS
1084 .show_fdinfo = bpf_prog_show_fdinfo,
1086 .release = bpf_prog_release,
1087 .read = bpf_dummy_read,
1088 .write = bpf_dummy_write,
1091 int bpf_prog_new_fd(struct bpf_prog *prog)
1095 ret = security_bpf_prog(prog);
1099 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1100 O_RDWR | O_CLOEXEC);
1103 static struct bpf_prog *____bpf_prog_get(struct fd f)
1106 return ERR_PTR(-EBADF);
1107 if (f.file->f_op != &bpf_prog_fops) {
1109 return ERR_PTR(-EINVAL);
1112 return f.file->private_data;
1115 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1117 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1118 atomic_sub(i, &prog->aux->refcnt);
1119 return ERR_PTR(-EBUSY);
1123 EXPORT_SYMBOL_GPL(bpf_prog_add);
1125 void bpf_prog_sub(struct bpf_prog *prog, int i)
1127 /* Only to be used for undoing previous bpf_prog_add() in some
1128 * error path. We still know that another entity in our call
1129 * path holds a reference to the program, thus atomic_sub() can
1130 * be safely used in such cases!
1132 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1134 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1136 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1138 return bpf_prog_add(prog, 1);
1140 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1142 /* prog_idr_lock should have been held */
1143 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1147 refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
1149 if (refold >= BPF_MAX_REFCNT) {
1150 __bpf_prog_put(prog, false);
1151 return ERR_PTR(-EBUSY);
1155 return ERR_PTR(-ENOENT);
1159 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1161 bool bpf_prog_get_ok(struct bpf_prog *prog,
1162 enum bpf_prog_type *attach_type, bool attach_drv)
1164 /* not an attachment, just a refcount inc, always allow */
1168 if (prog->type != *attach_type)
1170 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1176 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1179 struct fd f = fdget(ufd);
1180 struct bpf_prog *prog;
1182 prog = ____bpf_prog_get(f);
1185 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1186 prog = ERR_PTR(-EINVAL);
1190 prog = bpf_prog_inc(prog);
1196 struct bpf_prog *bpf_prog_get(u32 ufd)
1198 return __bpf_prog_get(ufd, NULL, false);
1201 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1204 return __bpf_prog_get(ufd, &type, attach_drv);
1206 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1208 /* Initially all BPF programs could be loaded w/o specifying
1209 * expected_attach_type. Later for some of them specifying expected_attach_type
1210 * at load time became required so that program could be validated properly.
1211 * Programs of types that are allowed to be loaded both w/ and w/o (for
1212 * backward compatibility) expected_attach_type, should have the default attach
1213 * type assigned to expected_attach_type for the latter case, so that it can be
1214 * validated later at attach time.
1216 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1217 * prog type requires it but has some attach types that have to be backward
1220 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1222 switch (attr->prog_type) {
1223 case BPF_PROG_TYPE_CGROUP_SOCK:
1224 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1225 * exist so checking for non-zero is the way to go here.
1227 if (!attr->expected_attach_type)
1228 attr->expected_attach_type =
1229 BPF_CGROUP_INET_SOCK_CREATE;
1235 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1236 enum bpf_attach_type expected_attach_type)
1238 switch (prog_type) {
1239 case BPF_PROG_TYPE_CGROUP_SOCK:
1240 switch (expected_attach_type) {
1241 case BPF_CGROUP_INET_SOCK_CREATE:
1242 case BPF_CGROUP_INET4_POST_BIND:
1243 case BPF_CGROUP_INET6_POST_BIND:
1248 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1249 switch (expected_attach_type) {
1250 case BPF_CGROUP_INET4_BIND:
1251 case BPF_CGROUP_INET6_BIND:
1252 case BPF_CGROUP_INET4_CONNECT:
1253 case BPF_CGROUP_INET6_CONNECT:
1254 case BPF_CGROUP_UDP4_SENDMSG:
1255 case BPF_CGROUP_UDP6_SENDMSG:
1265 /* last field in 'union bpf_attr' used by this command */
1266 #define BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1268 static int bpf_prog_load(union bpf_attr *attr)
1270 enum bpf_prog_type type = attr->prog_type;
1271 struct bpf_prog *prog;
1276 if (CHECK_ATTR(BPF_PROG_LOAD))
1279 if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1282 /* copy eBPF program license from user space */
1283 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1284 sizeof(license) - 1) < 0)
1286 license[sizeof(license) - 1] = 0;
1288 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1289 is_gpl = license_is_gpl_compatible(license);
1291 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1294 if (type == BPF_PROG_TYPE_KPROBE &&
1295 attr->kern_version != LINUX_VERSION_CODE)
1298 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1299 type != BPF_PROG_TYPE_CGROUP_SKB &&
1300 !capable(CAP_SYS_ADMIN))
1303 bpf_prog_load_fixup_attach_type(attr);
1304 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1307 /* plain bpf_prog allocation */
1308 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1312 prog->expected_attach_type = attr->expected_attach_type;
1314 prog->aux->offload_requested = !!attr->prog_ifindex;
1316 err = security_bpf_prog_alloc(prog->aux);
1318 goto free_prog_nouncharge;
1320 err = bpf_prog_charge_memlock(prog);
1324 prog->len = attr->insn_cnt;
1327 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1328 bpf_prog_insn_size(prog)) != 0)
1331 prog->orig_prog = NULL;
1334 atomic_set(&prog->aux->refcnt, 1);
1335 prog->gpl_compatible = is_gpl ? 1 : 0;
1337 if (bpf_prog_is_dev_bound(prog->aux)) {
1338 err = bpf_prog_offload_init(prog, attr);
1343 /* find program type: socket_filter vs tracing_filter */
1344 err = find_prog_type(type, prog);
1348 prog->aux->load_time = ktime_get_boot_ns();
1349 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1353 /* run eBPF verifier */
1354 err = bpf_check(&prog, attr);
1356 goto free_used_maps;
1358 prog = bpf_prog_select_runtime(prog, &err);
1360 goto free_used_maps;
1362 err = bpf_prog_alloc_id(prog);
1364 goto free_used_maps;
1366 err = bpf_prog_new_fd(prog);
1368 /* failed to allocate fd.
1369 * bpf_prog_put() is needed because the above
1370 * bpf_prog_alloc_id() has published the prog
1371 * to the userspace and the userspace may
1372 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1378 bpf_prog_kallsyms_add(prog);
1382 bpf_prog_kallsyms_del_subprogs(prog);
1383 free_used_maps(prog->aux);
1385 bpf_prog_uncharge_memlock(prog);
1387 security_bpf_prog_free(prog->aux);
1388 free_prog_nouncharge:
1389 bpf_prog_free(prog);
1393 #define BPF_OBJ_LAST_FIELD file_flags
1395 static int bpf_obj_pin(const union bpf_attr *attr)
1397 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1400 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1403 static int bpf_obj_get(const union bpf_attr *attr)
1405 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1406 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1409 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1413 struct bpf_raw_tracepoint {
1414 struct bpf_raw_event_map *btp;
1415 struct bpf_prog *prog;
1418 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1420 struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1423 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1424 bpf_prog_put(raw_tp->prog);
1430 static const struct file_operations bpf_raw_tp_fops = {
1431 .release = bpf_raw_tracepoint_release,
1432 .read = bpf_dummy_read,
1433 .write = bpf_dummy_write,
1436 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1438 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1440 struct bpf_raw_tracepoint *raw_tp;
1441 struct bpf_raw_event_map *btp;
1442 struct bpf_prog *prog;
1446 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1447 sizeof(tp_name) - 1) < 0)
1449 tp_name[sizeof(tp_name) - 1] = 0;
1451 btp = bpf_find_raw_tracepoint(tp_name);
1455 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1460 prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1461 BPF_PROG_TYPE_RAW_TRACEPOINT);
1463 err = PTR_ERR(prog);
1467 err = bpf_probe_register(raw_tp->btp, prog);
1471 raw_tp->prog = prog;
1472 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1475 bpf_probe_unregister(raw_tp->btp, prog);
1488 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1489 enum bpf_attach_type attach_type)
1491 switch (prog->type) {
1492 case BPF_PROG_TYPE_CGROUP_SOCK:
1493 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1494 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1500 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1502 #define BPF_F_ATTACH_MASK \
1503 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1505 static int bpf_prog_attach(const union bpf_attr *attr)
1507 enum bpf_prog_type ptype;
1508 struct bpf_prog *prog;
1511 if (!capable(CAP_NET_ADMIN))
1514 if (CHECK_ATTR(BPF_PROG_ATTACH))
1517 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1520 switch (attr->attach_type) {
1521 case BPF_CGROUP_INET_INGRESS:
1522 case BPF_CGROUP_INET_EGRESS:
1523 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1525 case BPF_CGROUP_INET_SOCK_CREATE:
1526 case BPF_CGROUP_INET4_POST_BIND:
1527 case BPF_CGROUP_INET6_POST_BIND:
1528 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1530 case BPF_CGROUP_INET4_BIND:
1531 case BPF_CGROUP_INET6_BIND:
1532 case BPF_CGROUP_INET4_CONNECT:
1533 case BPF_CGROUP_INET6_CONNECT:
1534 case BPF_CGROUP_UDP4_SENDMSG:
1535 case BPF_CGROUP_UDP6_SENDMSG:
1536 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1538 case BPF_CGROUP_SOCK_OPS:
1539 ptype = BPF_PROG_TYPE_SOCK_OPS;
1541 case BPF_CGROUP_DEVICE:
1542 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1544 case BPF_SK_MSG_VERDICT:
1545 ptype = BPF_PROG_TYPE_SK_MSG;
1547 case BPF_SK_SKB_STREAM_PARSER:
1548 case BPF_SK_SKB_STREAM_VERDICT:
1549 ptype = BPF_PROG_TYPE_SK_SKB;
1551 case BPF_LIRC_MODE2:
1552 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1558 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1560 return PTR_ERR(prog);
1562 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1568 case BPF_PROG_TYPE_SK_SKB:
1569 case BPF_PROG_TYPE_SK_MSG:
1570 ret = sockmap_get_from_fd(attr, ptype, prog);
1572 case BPF_PROG_TYPE_LIRC_MODE2:
1573 ret = lirc_prog_attach(attr, prog);
1576 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1584 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1586 static int bpf_prog_detach(const union bpf_attr *attr)
1588 enum bpf_prog_type ptype;
1590 if (!capable(CAP_NET_ADMIN))
1593 if (CHECK_ATTR(BPF_PROG_DETACH))
1596 switch (attr->attach_type) {
1597 case BPF_CGROUP_INET_INGRESS:
1598 case BPF_CGROUP_INET_EGRESS:
1599 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1601 case BPF_CGROUP_INET_SOCK_CREATE:
1602 case BPF_CGROUP_INET4_POST_BIND:
1603 case BPF_CGROUP_INET6_POST_BIND:
1604 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1606 case BPF_CGROUP_INET4_BIND:
1607 case BPF_CGROUP_INET6_BIND:
1608 case BPF_CGROUP_INET4_CONNECT:
1609 case BPF_CGROUP_INET6_CONNECT:
1610 case BPF_CGROUP_UDP4_SENDMSG:
1611 case BPF_CGROUP_UDP6_SENDMSG:
1612 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1614 case BPF_CGROUP_SOCK_OPS:
1615 ptype = BPF_PROG_TYPE_SOCK_OPS;
1617 case BPF_CGROUP_DEVICE:
1618 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1620 case BPF_SK_MSG_VERDICT:
1621 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, NULL);
1622 case BPF_SK_SKB_STREAM_PARSER:
1623 case BPF_SK_SKB_STREAM_VERDICT:
1624 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, NULL);
1625 case BPF_LIRC_MODE2:
1626 return lirc_prog_detach(attr);
1631 return cgroup_bpf_prog_detach(attr, ptype);
1634 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1636 static int bpf_prog_query(const union bpf_attr *attr,
1637 union bpf_attr __user *uattr)
1639 if (!capable(CAP_NET_ADMIN))
1641 if (CHECK_ATTR(BPF_PROG_QUERY))
1643 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1646 switch (attr->query.attach_type) {
1647 case BPF_CGROUP_INET_INGRESS:
1648 case BPF_CGROUP_INET_EGRESS:
1649 case BPF_CGROUP_INET_SOCK_CREATE:
1650 case BPF_CGROUP_INET4_BIND:
1651 case BPF_CGROUP_INET6_BIND:
1652 case BPF_CGROUP_INET4_POST_BIND:
1653 case BPF_CGROUP_INET6_POST_BIND:
1654 case BPF_CGROUP_INET4_CONNECT:
1655 case BPF_CGROUP_INET6_CONNECT:
1656 case BPF_CGROUP_UDP4_SENDMSG:
1657 case BPF_CGROUP_UDP6_SENDMSG:
1658 case BPF_CGROUP_SOCK_OPS:
1659 case BPF_CGROUP_DEVICE:
1661 case BPF_LIRC_MODE2:
1662 return lirc_prog_query(attr, uattr);
1667 return cgroup_bpf_prog_query(attr, uattr);
1670 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1672 static int bpf_prog_test_run(const union bpf_attr *attr,
1673 union bpf_attr __user *uattr)
1675 struct bpf_prog *prog;
1676 int ret = -ENOTSUPP;
1678 if (!capable(CAP_SYS_ADMIN))
1680 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1683 prog = bpf_prog_get(attr->test.prog_fd);
1685 return PTR_ERR(prog);
1687 if (prog->aux->ops->test_run)
1688 ret = prog->aux->ops->test_run(prog, attr, uattr);
1694 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1696 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1697 union bpf_attr __user *uattr,
1701 u32 next_id = attr->start_id;
1704 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1707 if (!capable(CAP_SYS_ADMIN))
1712 if (!idr_get_next(idr, &next_id))
1714 spin_unlock_bh(lock);
1717 err = put_user(next_id, &uattr->next_id);
1722 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1724 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1726 struct bpf_prog *prog;
1727 u32 id = attr->prog_id;
1730 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1733 if (!capable(CAP_SYS_ADMIN))
1736 spin_lock_bh(&prog_idr_lock);
1737 prog = idr_find(&prog_idr, id);
1739 prog = bpf_prog_inc_not_zero(prog);
1741 prog = ERR_PTR(-ENOENT);
1742 spin_unlock_bh(&prog_idr_lock);
1745 return PTR_ERR(prog);
1747 fd = bpf_prog_new_fd(prog);
1754 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1756 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1758 struct bpf_map *map;
1759 u32 id = attr->map_id;
1763 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1764 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1767 if (!capable(CAP_SYS_ADMIN))
1770 f_flags = bpf_get_file_flag(attr->open_flags);
1774 spin_lock_bh(&map_idr_lock);
1775 map = idr_find(&map_idr, id);
1777 map = bpf_map_inc_not_zero(map, true);
1779 map = ERR_PTR(-ENOENT);
1780 spin_unlock_bh(&map_idr_lock);
1783 return PTR_ERR(map);
1785 fd = bpf_map_new_fd(map, f_flags);
1792 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1797 for (i = 0; i < prog->aux->used_map_cnt; i++)
1798 if (prog->aux->used_maps[i] == (void *)addr)
1799 return prog->aux->used_maps[i];
1803 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1805 const struct bpf_map *map;
1806 struct bpf_insn *insns;
1810 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1815 for (i = 0; i < prog->len; i++) {
1816 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1817 insns[i].code = BPF_JMP | BPF_CALL;
1818 insns[i].imm = BPF_FUNC_tail_call;
1821 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1822 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1823 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1824 insns[i].code = BPF_JMP | BPF_CALL;
1825 if (!bpf_dump_raw_ok())
1830 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1833 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1834 map = bpf_map_from_imm(prog, imm);
1836 insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1837 insns[i].imm = map->id;
1838 insns[i + 1].imm = 0;
1842 if (!bpf_dump_raw_ok() &&
1843 imm == (unsigned long)prog->aux) {
1845 insns[i + 1].imm = 0;
1853 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1854 const union bpf_attr *attr,
1855 union bpf_attr __user *uattr)
1857 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1858 struct bpf_prog_info info = {};
1859 u32 info_len = attr->info.info_len;
1860 char __user *uinsns;
1864 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1867 info_len = min_t(u32, sizeof(info), info_len);
1869 if (copy_from_user(&info, uinfo, info_len))
1872 info.type = prog->type;
1873 info.id = prog->aux->id;
1874 info.load_time = prog->aux->load_time;
1875 info.created_by_uid = from_kuid_munged(current_user_ns(),
1876 prog->aux->user->uid);
1877 info.gpl_compatible = prog->gpl_compatible;
1879 memcpy(info.tag, prog->tag, sizeof(prog->tag));
1880 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1882 ulen = info.nr_map_ids;
1883 info.nr_map_ids = prog->aux->used_map_cnt;
1884 ulen = min_t(u32, info.nr_map_ids, ulen);
1886 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1889 for (i = 0; i < ulen; i++)
1890 if (put_user(prog->aux->used_maps[i]->id,
1895 if (!capable(CAP_SYS_ADMIN)) {
1896 info.jited_prog_len = 0;
1897 info.xlated_prog_len = 0;
1898 info.nr_jited_ksyms = 0;
1902 ulen = info.xlated_prog_len;
1903 info.xlated_prog_len = bpf_prog_insn_size(prog);
1904 if (info.xlated_prog_len && ulen) {
1905 struct bpf_insn *insns_sanitized;
1908 if (prog->blinded && !bpf_dump_raw_ok()) {
1909 info.xlated_prog_insns = 0;
1912 insns_sanitized = bpf_insn_prepare_dump(prog);
1913 if (!insns_sanitized)
1915 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1916 ulen = min_t(u32, info.xlated_prog_len, ulen);
1917 fault = copy_to_user(uinsns, insns_sanitized, ulen);
1918 kfree(insns_sanitized);
1923 if (bpf_prog_is_dev_bound(prog->aux)) {
1924 err = bpf_prog_offload_info_fill(&info, prog);
1930 /* NOTE: the following code is supposed to be skipped for offload.
1931 * bpf_prog_offload_info_fill() is the place to fill similar fields
1934 ulen = info.jited_prog_len;
1935 if (prog->aux->func_cnt) {
1938 info.jited_prog_len = 0;
1939 for (i = 0; i < prog->aux->func_cnt; i++)
1940 info.jited_prog_len += prog->aux->func[i]->jited_len;
1942 info.jited_prog_len = prog->jited_len;
1945 if (info.jited_prog_len && ulen) {
1946 if (bpf_dump_raw_ok()) {
1947 uinsns = u64_to_user_ptr(info.jited_prog_insns);
1948 ulen = min_t(u32, info.jited_prog_len, ulen);
1950 /* for multi-function programs, copy the JITed
1951 * instructions for all the functions
1953 if (prog->aux->func_cnt) {
1958 for (i = 0; i < prog->aux->func_cnt; i++) {
1959 len = prog->aux->func[i]->jited_len;
1960 len = min_t(u32, len, free);
1961 img = (u8 *) prog->aux->func[i]->bpf_func;
1962 if (copy_to_user(uinsns, img, len))
1970 if (copy_to_user(uinsns, prog->bpf_func, ulen))
1974 info.jited_prog_insns = 0;
1978 ulen = info.nr_jited_ksyms;
1979 info.nr_jited_ksyms = prog->aux->func_cnt;
1980 if (info.nr_jited_ksyms && ulen) {
1981 if (bpf_dump_raw_ok()) {
1982 u64 __user *user_ksyms;
1986 /* copy the address of the kernel symbol
1987 * corresponding to each function
1989 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
1990 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
1991 for (i = 0; i < ulen; i++) {
1992 ksym_addr = (ulong) prog->aux->func[i]->bpf_func;
1993 ksym_addr &= PAGE_MASK;
1994 if (put_user((u64) ksym_addr, &user_ksyms[i]))
1998 info.jited_ksyms = 0;
2002 ulen = info.nr_jited_func_lens;
2003 info.nr_jited_func_lens = prog->aux->func_cnt;
2004 if (info.nr_jited_func_lens && ulen) {
2005 if (bpf_dump_raw_ok()) {
2006 u32 __user *user_lens;
2009 /* copy the JITed image lengths for each function */
2010 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2011 user_lens = u64_to_user_ptr(info.jited_func_lens);
2012 for (i = 0; i < ulen; i++) {
2013 func_len = prog->aux->func[i]->jited_len;
2014 if (put_user(func_len, &user_lens[i]))
2018 info.jited_func_lens = 0;
2023 if (copy_to_user(uinfo, &info, info_len) ||
2024 put_user(info_len, &uattr->info.info_len))
2030 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2031 const union bpf_attr *attr,
2032 union bpf_attr __user *uattr)
2034 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2035 struct bpf_map_info info = {};
2036 u32 info_len = attr->info.info_len;
2039 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2042 info_len = min_t(u32, sizeof(info), info_len);
2044 info.type = map->map_type;
2046 info.key_size = map->key_size;
2047 info.value_size = map->value_size;
2048 info.max_entries = map->max_entries;
2049 info.map_flags = map->map_flags;
2050 memcpy(info.name, map->name, sizeof(map->name));
2053 info.btf_id = btf_id(map->btf);
2054 info.btf_key_type_id = map->btf_key_type_id;
2055 info.btf_value_type_id = map->btf_value_type_id;
2058 if (bpf_map_is_dev_bound(map)) {
2059 err = bpf_map_offload_info_fill(&info, map);
2064 if (copy_to_user(uinfo, &info, info_len) ||
2065 put_user(info_len, &uattr->info.info_len))
2071 static int bpf_btf_get_info_by_fd(struct btf *btf,
2072 const union bpf_attr *attr,
2073 union bpf_attr __user *uattr)
2075 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2076 u32 info_len = attr->info.info_len;
2079 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2083 return btf_get_info_by_fd(btf, attr, uattr);
2086 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2088 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2089 union bpf_attr __user *uattr)
2091 int ufd = attr->info.bpf_fd;
2095 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2102 if (f.file->f_op == &bpf_prog_fops)
2103 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2105 else if (f.file->f_op == &bpf_map_fops)
2106 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2108 else if (f.file->f_op == &btf_fops)
2109 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2117 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2119 static int bpf_btf_load(const union bpf_attr *attr)
2121 if (CHECK_ATTR(BPF_BTF_LOAD))
2124 if (!capable(CAP_SYS_ADMIN))
2127 return btf_new_fd(attr);
2130 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2132 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2134 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2137 if (!capable(CAP_SYS_ADMIN))
2140 return btf_get_fd_by_id(attr->btf_id);
2143 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2144 union bpf_attr __user *uattr,
2145 u32 prog_id, u32 fd_type,
2146 const char *buf, u64 probe_offset,
2149 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2150 u32 len = buf ? strlen(buf) : 0, input_len;
2153 if (put_user(len, &uattr->task_fd_query.buf_len))
2155 input_len = attr->task_fd_query.buf_len;
2156 if (input_len && ubuf) {
2158 /* nothing to copy, just make ubuf NULL terminated */
2161 if (put_user(zero, ubuf))
2163 } else if (input_len >= len + 1) {
2164 /* ubuf can hold the string with NULL terminator */
2165 if (copy_to_user(ubuf, buf, len + 1))
2168 /* ubuf cannot hold the string with NULL terminator,
2169 * do a partial copy with NULL terminator.
2174 if (copy_to_user(ubuf, buf, input_len - 1))
2176 if (put_user(zero, ubuf + input_len - 1))
2181 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2182 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2183 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2184 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2190 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2192 static int bpf_task_fd_query(const union bpf_attr *attr,
2193 union bpf_attr __user *uattr)
2195 pid_t pid = attr->task_fd_query.pid;
2196 u32 fd = attr->task_fd_query.fd;
2197 const struct perf_event *event;
2198 struct files_struct *files;
2199 struct task_struct *task;
2203 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2206 if (!capable(CAP_SYS_ADMIN))
2209 if (attr->task_fd_query.flags != 0)
2212 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2216 files = get_files_struct(task);
2217 put_task_struct(task);
2222 spin_lock(&files->file_lock);
2223 file = fcheck_files(files, fd);
2228 spin_unlock(&files->file_lock);
2229 put_files_struct(files);
2234 if (file->f_op == &bpf_raw_tp_fops) {
2235 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2236 struct bpf_raw_event_map *btp = raw_tp->btp;
2238 err = bpf_task_fd_query_copy(attr, uattr,
2239 raw_tp->prog->aux->id,
2240 BPF_FD_TYPE_RAW_TRACEPOINT,
2241 btp->tp->name, 0, 0);
2245 event = perf_get_event(file);
2246 if (!IS_ERR(event)) {
2247 u64 probe_offset, probe_addr;
2248 u32 prog_id, fd_type;
2251 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2252 &buf, &probe_offset,
2255 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2269 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2271 union bpf_attr attr = {};
2274 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2277 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2280 size = min_t(u32, size, sizeof(attr));
2282 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2283 if (copy_from_user(&attr, uattr, size) != 0)
2286 err = security_bpf(cmd, &attr, size);
2291 case BPF_MAP_CREATE:
2292 err = map_create(&attr);
2294 case BPF_MAP_LOOKUP_ELEM:
2295 err = map_lookup_elem(&attr);
2297 case BPF_MAP_UPDATE_ELEM:
2298 err = map_update_elem(&attr);
2300 case BPF_MAP_DELETE_ELEM:
2301 err = map_delete_elem(&attr);
2303 case BPF_MAP_GET_NEXT_KEY:
2304 err = map_get_next_key(&attr);
2307 err = bpf_prog_load(&attr);
2310 err = bpf_obj_pin(&attr);
2313 err = bpf_obj_get(&attr);
2315 case BPF_PROG_ATTACH:
2316 err = bpf_prog_attach(&attr);
2318 case BPF_PROG_DETACH:
2319 err = bpf_prog_detach(&attr);
2321 case BPF_PROG_QUERY:
2322 err = bpf_prog_query(&attr, uattr);
2324 case BPF_PROG_TEST_RUN:
2325 err = bpf_prog_test_run(&attr, uattr);
2327 case BPF_PROG_GET_NEXT_ID:
2328 err = bpf_obj_get_next_id(&attr, uattr,
2329 &prog_idr, &prog_idr_lock);
2331 case BPF_MAP_GET_NEXT_ID:
2332 err = bpf_obj_get_next_id(&attr, uattr,
2333 &map_idr, &map_idr_lock);
2335 case BPF_PROG_GET_FD_BY_ID:
2336 err = bpf_prog_get_fd_by_id(&attr);
2338 case BPF_MAP_GET_FD_BY_ID:
2339 err = bpf_map_get_fd_by_id(&attr);
2341 case BPF_OBJ_GET_INFO_BY_FD:
2342 err = bpf_obj_get_info_by_fd(&attr, uattr);
2344 case BPF_RAW_TRACEPOINT_OPEN:
2345 err = bpf_raw_tracepoint_open(&attr);
2348 err = bpf_btf_load(&attr);
2350 case BPF_BTF_GET_FD_BY_ID:
2351 err = bpf_btf_get_fd_by_id(&attr);
2353 case BPF_TASK_FD_QUERY:
2354 err = bpf_task_fd_query(&attr, uattr);