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 =
52 IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
54 static const struct bpf_map_ops * const bpf_map_types[] = {
55 #define BPF_PROG_TYPE(_id, _ops)
56 #define BPF_MAP_TYPE(_id, _ops) \
58 #include <linux/bpf_types.h>
64 * If we're handed a bigger struct than we know of, ensure all the unknown bits
65 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
66 * we don't know about yet.
68 * There is a ToCToU between this function call and the following
69 * copy_from_user() call. However, this is not a concern since this function is
70 * meant to be a future-proofing of bits.
72 int bpf_check_uarg_tail_zero(void __user *uaddr,
76 unsigned char __user *addr;
77 unsigned char __user *end;
81 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
84 if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
87 if (actual_size <= expected_size)
90 addr = uaddr + expected_size;
91 end = uaddr + actual_size;
93 for (; addr < end; addr++) {
94 err = get_user(val, addr);
104 const struct bpf_map_ops bpf_map_offload_ops = {
105 .map_alloc = bpf_map_offload_map_alloc,
106 .map_free = bpf_map_offload_map_free,
107 .map_check_btf = map_check_no_btf,
110 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
112 const struct bpf_map_ops *ops;
113 u32 type = attr->map_type;
117 if (type >= ARRAY_SIZE(bpf_map_types))
118 return ERR_PTR(-EINVAL);
119 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
120 ops = bpf_map_types[type];
122 return ERR_PTR(-EINVAL);
124 if (ops->map_alloc_check) {
125 err = ops->map_alloc_check(attr);
129 if (attr->map_ifindex)
130 ops = &bpf_map_offload_ops;
131 map = ops->map_alloc(attr);
135 map->map_type = type;
139 void *bpf_map_area_alloc(size_t size, int numa_node)
141 /* We definitely need __GFP_NORETRY, so OOM killer doesn't
142 * trigger under memory pressure as we really just want to
145 const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
148 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
149 area = kmalloc_node(size, GFP_USER | flags, numa_node);
154 return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
155 __builtin_return_address(0));
158 void bpf_map_area_free(void *area)
163 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
165 map->map_type = attr->map_type;
166 map->key_size = attr->key_size;
167 map->value_size = attr->value_size;
168 map->max_entries = attr->max_entries;
169 map->map_flags = attr->map_flags;
170 map->numa_node = bpf_map_attr_numa_node(attr);
173 int bpf_map_precharge_memlock(u32 pages)
175 struct user_struct *user = get_current_user();
176 unsigned long memlock_limit, cur;
178 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
179 cur = atomic_long_read(&user->locked_vm);
181 if (cur + pages > memlock_limit)
186 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
188 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
190 if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
191 atomic_long_sub(pages, &user->locked_vm);
197 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
199 atomic_long_sub(pages, &user->locked_vm);
202 static int bpf_map_init_memlock(struct bpf_map *map)
204 struct user_struct *user = get_current_user();
207 ret = bpf_charge_memlock(user, map->pages);
216 static void bpf_map_release_memlock(struct bpf_map *map)
218 struct user_struct *user = map->user;
219 bpf_uncharge_memlock(user, map->pages);
223 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
227 ret = bpf_charge_memlock(map->user, pages);
234 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
236 bpf_uncharge_memlock(map->user, pages);
240 static int bpf_map_alloc_id(struct bpf_map *map)
244 idr_preload(GFP_KERNEL);
245 spin_lock_bh(&map_idr_lock);
246 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
249 spin_unlock_bh(&map_idr_lock);
252 if (WARN_ON_ONCE(!id))
255 return id > 0 ? 0 : id;
258 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
262 /* Offloaded maps are removed from the IDR store when their device
263 * disappears - even if someone holds an fd to them they are unusable,
264 * the memory is gone, all ops will fail; they are simply waiting for
265 * refcnt to drop to be freed.
271 spin_lock_irqsave(&map_idr_lock, flags);
273 __acquire(&map_idr_lock);
275 idr_remove(&map_idr, map->id);
279 spin_unlock_irqrestore(&map_idr_lock, flags);
281 __release(&map_idr_lock);
284 /* called from workqueue */
285 static void bpf_map_free_deferred(struct work_struct *work)
287 struct bpf_map *map = container_of(work, struct bpf_map, work);
289 bpf_map_release_memlock(map);
290 security_bpf_map_free(map);
291 /* implementation dependent freeing */
292 map->ops->map_free(map);
295 static void bpf_map_put_uref(struct bpf_map *map)
297 if (atomic_dec_and_test(&map->usercnt)) {
298 if (map->ops->map_release_uref)
299 map->ops->map_release_uref(map);
303 /* decrement map refcnt and schedule it for freeing via workqueue
304 * (unrelying map implementation ops->map_free() might sleep)
306 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
308 if (atomic_dec_and_test(&map->refcnt)) {
309 /* bpf_map_free_id() must be called first */
310 bpf_map_free_id(map, do_idr_lock);
312 INIT_WORK(&map->work, bpf_map_free_deferred);
313 schedule_work(&map->work);
317 void bpf_map_put(struct bpf_map *map)
319 __bpf_map_put(map, true);
321 EXPORT_SYMBOL_GPL(bpf_map_put);
323 void bpf_map_put_with_uref(struct bpf_map *map)
325 bpf_map_put_uref(map);
329 static int bpf_map_release(struct inode *inode, struct file *filp)
331 struct bpf_map *map = filp->private_data;
333 if (map->ops->map_release)
334 map->ops->map_release(map, filp);
336 bpf_map_put_with_uref(map);
340 #ifdef CONFIG_PROC_FS
341 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
343 const struct bpf_map *map = filp->private_data;
344 const struct bpf_array *array;
345 u32 owner_prog_type = 0;
348 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
349 array = container_of(map, struct bpf_array, map);
350 owner_prog_type = array->owner_prog_type;
351 owner_jited = array->owner_jited;
367 map->pages * 1ULL << PAGE_SHIFT,
370 if (owner_prog_type) {
371 seq_printf(m, "owner_prog_type:\t%u\n",
373 seq_printf(m, "owner_jited:\t%u\n",
379 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
382 /* We need this handler such that alloc_file() enables
383 * f_mode with FMODE_CAN_READ.
388 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
389 size_t siz, loff_t *ppos)
391 /* We need this handler such that alloc_file() enables
392 * f_mode with FMODE_CAN_WRITE.
397 const struct file_operations bpf_map_fops = {
398 #ifdef CONFIG_PROC_FS
399 .show_fdinfo = bpf_map_show_fdinfo,
401 .release = bpf_map_release,
402 .read = bpf_dummy_read,
403 .write = bpf_dummy_write,
406 int bpf_map_new_fd(struct bpf_map *map, int flags)
410 ret = security_bpf_map(map, OPEN_FMODE(flags));
414 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
418 int bpf_get_file_flag(int flags)
420 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
422 if (flags & BPF_F_RDONLY)
424 if (flags & BPF_F_WRONLY)
429 /* helper macro to check that unused fields 'union bpf_attr' are zero */
430 #define CHECK_ATTR(CMD) \
431 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
432 sizeof(attr->CMD##_LAST_FIELD), 0, \
434 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
435 sizeof(attr->CMD##_LAST_FIELD)) != NULL
437 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
438 * Return 0 on success and < 0 on error.
440 static int bpf_obj_name_cpy(char *dst, const char *src)
442 const char *end = src + BPF_OBJ_NAME_LEN;
444 memset(dst, 0, BPF_OBJ_NAME_LEN);
446 /* Copy all isalnum() and '_' char */
447 while (src < end && *src) {
448 if (!isalnum(*src) && *src != '_')
453 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
460 int map_check_no_btf(const struct bpf_map *map,
461 const struct btf_type *key_type,
462 const struct btf_type *value_type)
467 static int map_check_btf(const struct bpf_map *map, const struct btf *btf,
468 u32 btf_key_id, u32 btf_value_id)
470 const struct btf_type *key_type, *value_type;
471 u32 key_size, value_size;
474 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
475 if (!key_type || key_size != map->key_size)
478 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
479 if (!value_type || value_size != map->value_size)
482 if (map->ops->map_check_btf)
483 ret = map->ops->map_check_btf(map, key_type, value_type);
488 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
489 /* called via syscall */
490 static int map_create(union bpf_attr *attr)
492 int numa_node = bpf_map_attr_numa_node(attr);
497 err = CHECK_ATTR(BPF_MAP_CREATE);
501 f_flags = bpf_get_file_flag(attr->map_flags);
505 if (numa_node != NUMA_NO_NODE &&
506 ((unsigned int)numa_node >= nr_node_ids ||
507 !node_online(numa_node)))
510 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
511 map = find_and_alloc_map(attr);
515 err = bpf_obj_name_cpy(map->name, attr->map_name);
517 goto free_map_nouncharge;
519 atomic_set(&map->refcnt, 1);
520 atomic_set(&map->usercnt, 1);
522 if (attr->btf_key_type_id || attr->btf_value_type_id) {
525 if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
527 goto free_map_nouncharge;
530 btf = btf_get_by_fd(attr->btf_fd);
533 goto free_map_nouncharge;
536 err = map_check_btf(map, btf, attr->btf_key_type_id,
537 attr->btf_value_type_id);
540 goto free_map_nouncharge;
544 map->btf_key_type_id = attr->btf_key_type_id;
545 map->btf_value_type_id = attr->btf_value_type_id;
548 err = security_bpf_map_alloc(map);
550 goto free_map_nouncharge;
552 err = bpf_map_init_memlock(map);
556 err = bpf_map_alloc_id(map);
560 err = bpf_map_new_fd(map, f_flags);
562 /* failed to allocate fd.
563 * bpf_map_put_with_uref() is needed because the above
564 * bpf_map_alloc_id() has published the map
565 * to the userspace and the userspace may
566 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
568 bpf_map_put_with_uref(map);
575 bpf_map_release_memlock(map);
577 security_bpf_map_free(map);
580 map->ops->map_free(map);
584 /* if error is returned, fd is released.
585 * On success caller should complete fd access with matching fdput()
587 struct bpf_map *__bpf_map_get(struct fd f)
590 return ERR_PTR(-EBADF);
591 if (f.file->f_op != &bpf_map_fops) {
593 return ERR_PTR(-EINVAL);
596 return f.file->private_data;
599 /* prog's and map's refcnt limit */
600 #define BPF_MAX_REFCNT 32768
602 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
604 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
605 atomic_dec(&map->refcnt);
606 return ERR_PTR(-EBUSY);
609 atomic_inc(&map->usercnt);
612 EXPORT_SYMBOL_GPL(bpf_map_inc);
614 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
616 struct fd f = fdget(ufd);
619 map = __bpf_map_get(f);
623 map = bpf_map_inc(map, true);
629 /* map_idr_lock should have been held */
630 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
635 refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
637 if (refold >= BPF_MAX_REFCNT) {
638 __bpf_map_put(map, false);
639 return ERR_PTR(-EBUSY);
643 return ERR_PTR(-ENOENT);
646 atomic_inc(&map->usercnt);
651 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
656 /* last field in 'union bpf_attr' used by this command */
657 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
659 static int map_lookup_elem(union bpf_attr *attr)
661 void __user *ukey = u64_to_user_ptr(attr->key);
662 void __user *uvalue = u64_to_user_ptr(attr->value);
663 int ufd = attr->map_fd;
665 void *key, *value, *ptr;
670 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
674 map = __bpf_map_get(f);
678 if (!(f.file->f_mode & FMODE_CAN_READ)) {
683 key = memdup_user(ukey, map->key_size);
689 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
690 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
691 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
692 value_size = round_up(map->value_size, 8) * num_possible_cpus();
693 else if (IS_FD_MAP(map))
694 value_size = sizeof(u32);
696 value_size = map->value_size;
699 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
703 if (bpf_map_is_dev_bound(map)) {
704 err = bpf_map_offload_lookup_elem(map, key, value);
709 this_cpu_inc(bpf_prog_active);
710 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
711 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
712 err = bpf_percpu_hash_copy(map, key, value);
713 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
714 err = bpf_percpu_array_copy(map, key, value);
715 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
716 err = bpf_stackmap_copy(map, key, value);
717 } else if (IS_FD_ARRAY(map)) {
718 err = bpf_fd_array_map_lookup_elem(map, key, value);
719 } else if (IS_FD_HASH(map)) {
720 err = bpf_fd_htab_map_lookup_elem(map, key, value);
721 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
722 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
725 if (map->ops->map_lookup_elem_sys_only)
726 ptr = map->ops->map_lookup_elem_sys_only(map, key);
728 ptr = map->ops->map_lookup_elem(map, key);
730 memcpy(value, ptr, value_size);
732 err = ptr ? 0 : -ENOENT;
734 this_cpu_dec(bpf_prog_active);
742 if (copy_to_user(uvalue, value, value_size) != 0)
756 static void maybe_wait_bpf_programs(struct bpf_map *map)
758 /* Wait for any running BPF programs to complete so that
759 * userspace, when we return to it, knows that all programs
760 * that could be running use the new map value.
762 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
763 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
767 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
769 static int map_update_elem(union bpf_attr *attr)
771 void __user *ukey = u64_to_user_ptr(attr->key);
772 void __user *uvalue = u64_to_user_ptr(attr->value);
773 int ufd = attr->map_fd;
780 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
784 map = __bpf_map_get(f);
788 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
793 key = memdup_user(ukey, map->key_size);
799 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
800 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
801 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
802 value_size = round_up(map->value_size, 8) * num_possible_cpus();
804 value_size = map->value_size;
807 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
812 if (copy_from_user(value, uvalue, value_size) != 0)
815 /* Need to create a kthread, thus must support schedule */
816 if (bpf_map_is_dev_bound(map)) {
817 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
819 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
820 map->map_type == BPF_MAP_TYPE_SOCKHASH ||
821 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
822 err = map->ops->map_update_elem(map, key, value, attr->flags);
826 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
827 * inside bpf map update or delete otherwise deadlocks are possible
830 __this_cpu_inc(bpf_prog_active);
831 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
832 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
833 err = bpf_percpu_hash_update(map, key, value, attr->flags);
834 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
835 err = bpf_percpu_array_update(map, key, value, attr->flags);
836 } else if (IS_FD_ARRAY(map)) {
838 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
841 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
843 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
846 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
847 /* rcu_read_lock() is not needed */
848 err = bpf_fd_reuseport_array_update_elem(map, key, value,
852 err = map->ops->map_update_elem(map, key, value, attr->flags);
855 __this_cpu_dec(bpf_prog_active);
857 maybe_wait_bpf_programs(map);
868 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
870 static int map_delete_elem(union bpf_attr *attr)
872 void __user *ukey = u64_to_user_ptr(attr->key);
873 int ufd = attr->map_fd;
879 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
883 map = __bpf_map_get(f);
887 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
892 key = memdup_user(ukey, map->key_size);
898 if (bpf_map_is_dev_bound(map)) {
899 err = bpf_map_offload_delete_elem(map, key);
904 __this_cpu_inc(bpf_prog_active);
906 err = map->ops->map_delete_elem(map, key);
908 __this_cpu_dec(bpf_prog_active);
910 maybe_wait_bpf_programs(map);
918 /* last field in 'union bpf_attr' used by this command */
919 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
921 static int map_get_next_key(union bpf_attr *attr)
923 void __user *ukey = u64_to_user_ptr(attr->key);
924 void __user *unext_key = u64_to_user_ptr(attr->next_key);
925 int ufd = attr->map_fd;
927 void *key, *next_key;
931 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
935 map = __bpf_map_get(f);
939 if (!(f.file->f_mode & FMODE_CAN_READ)) {
945 key = memdup_user(ukey, map->key_size);
955 next_key = kmalloc(map->key_size, GFP_USER);
959 if (bpf_map_is_dev_bound(map)) {
960 err = bpf_map_offload_get_next_key(map, key, next_key);
965 err = map->ops->map_get_next_key(map, key, next_key);
972 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
986 static const struct bpf_prog_ops * const bpf_prog_types[] = {
987 #define BPF_PROG_TYPE(_id, _name) \
988 [_id] = & _name ## _prog_ops,
989 #define BPF_MAP_TYPE(_id, _ops)
990 #include <linux/bpf_types.h>
995 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
997 const struct bpf_prog_ops *ops;
999 if (type >= ARRAY_SIZE(bpf_prog_types))
1001 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1002 ops = bpf_prog_types[type];
1006 if (!bpf_prog_is_dev_bound(prog->aux))
1007 prog->aux->ops = ops;
1009 prog->aux->ops = &bpf_offload_prog_ops;
1014 /* drop refcnt on maps used by eBPF program and free auxilary data */
1015 static void free_used_maps(struct bpf_prog_aux *aux)
1019 if (aux->cgroup_storage)
1020 bpf_cgroup_storage_release(aux->prog, aux->cgroup_storage);
1022 for (i = 0; i < aux->used_map_cnt; i++)
1023 bpf_map_put(aux->used_maps[i]);
1025 kfree(aux->used_maps);
1028 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1030 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1031 unsigned long user_bufs;
1034 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1035 if (user_bufs > memlock_limit) {
1036 atomic_long_sub(pages, &user->locked_vm);
1044 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1047 atomic_long_sub(pages, &user->locked_vm);
1050 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1052 struct user_struct *user = get_current_user();
1055 ret = __bpf_prog_charge(user, prog->pages);
1061 prog->aux->user = user;
1065 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1067 struct user_struct *user = prog->aux->user;
1069 __bpf_prog_uncharge(user, prog->pages);
1073 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1077 idr_preload(GFP_KERNEL);
1078 spin_lock_bh(&prog_idr_lock);
1079 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1082 spin_unlock_bh(&prog_idr_lock);
1085 /* id is in [1, INT_MAX) */
1086 if (WARN_ON_ONCE(!id))
1089 return id > 0 ? 0 : id;
1092 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1094 /* cBPF to eBPF migrations are currently not in the idr store.
1095 * Offloaded programs are removed from the store when their device
1096 * disappears - even if someone grabs an fd to them they are unusable,
1097 * simply waiting for refcnt to drop to be freed.
1103 spin_lock_bh(&prog_idr_lock);
1105 __acquire(&prog_idr_lock);
1107 idr_remove(&prog_idr, prog->aux->id);
1111 spin_unlock_bh(&prog_idr_lock);
1113 __release(&prog_idr_lock);
1116 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1118 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1120 free_used_maps(aux);
1121 bpf_prog_uncharge_memlock(aux->prog);
1122 security_bpf_prog_free(aux);
1123 bpf_prog_free(aux->prog);
1126 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1128 if (atomic_dec_and_test(&prog->aux->refcnt)) {
1129 /* bpf_prog_free_id() must be called first */
1130 bpf_prog_free_id(prog, do_idr_lock);
1131 bpf_prog_kallsyms_del_all(prog);
1133 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1137 void bpf_prog_put(struct bpf_prog *prog)
1139 __bpf_prog_put(prog, true);
1141 EXPORT_SYMBOL_GPL(bpf_prog_put);
1143 static int bpf_prog_release(struct inode *inode, struct file *filp)
1145 struct bpf_prog *prog = filp->private_data;
1151 #ifdef CONFIG_PROC_FS
1152 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1154 const struct bpf_prog *prog = filp->private_data;
1155 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1157 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1167 prog->pages * 1ULL << PAGE_SHIFT,
1172 const struct file_operations bpf_prog_fops = {
1173 #ifdef CONFIG_PROC_FS
1174 .show_fdinfo = bpf_prog_show_fdinfo,
1176 .release = bpf_prog_release,
1177 .read = bpf_dummy_read,
1178 .write = bpf_dummy_write,
1181 int bpf_prog_new_fd(struct bpf_prog *prog)
1185 ret = security_bpf_prog(prog);
1189 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1190 O_RDWR | O_CLOEXEC);
1193 static struct bpf_prog *____bpf_prog_get(struct fd f)
1196 return ERR_PTR(-EBADF);
1197 if (f.file->f_op != &bpf_prog_fops) {
1199 return ERR_PTR(-EINVAL);
1202 return f.file->private_data;
1205 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1207 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1208 atomic_sub(i, &prog->aux->refcnt);
1209 return ERR_PTR(-EBUSY);
1213 EXPORT_SYMBOL_GPL(bpf_prog_add);
1215 void bpf_prog_sub(struct bpf_prog *prog, int i)
1217 /* Only to be used for undoing previous bpf_prog_add() in some
1218 * error path. We still know that another entity in our call
1219 * path holds a reference to the program, thus atomic_sub() can
1220 * be safely used in such cases!
1222 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1224 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1226 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1228 return bpf_prog_add(prog, 1);
1230 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1232 /* prog_idr_lock should have been held */
1233 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1237 refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1239 if (refold >= BPF_MAX_REFCNT) {
1240 __bpf_prog_put(prog, false);
1241 return ERR_PTR(-EBUSY);
1245 return ERR_PTR(-ENOENT);
1249 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1251 bool bpf_prog_get_ok(struct bpf_prog *prog,
1252 enum bpf_prog_type *attach_type, bool attach_drv)
1254 /* not an attachment, just a refcount inc, always allow */
1258 if (prog->type != *attach_type)
1260 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1266 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1269 struct fd f = fdget(ufd);
1270 struct bpf_prog *prog;
1272 prog = ____bpf_prog_get(f);
1275 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1276 prog = ERR_PTR(-EINVAL);
1280 prog = bpf_prog_inc(prog);
1286 struct bpf_prog *bpf_prog_get(u32 ufd)
1288 return __bpf_prog_get(ufd, NULL, false);
1291 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1294 return __bpf_prog_get(ufd, &type, attach_drv);
1296 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1298 /* Initially all BPF programs could be loaded w/o specifying
1299 * expected_attach_type. Later for some of them specifying expected_attach_type
1300 * at load time became required so that program could be validated properly.
1301 * Programs of types that are allowed to be loaded both w/ and w/o (for
1302 * backward compatibility) expected_attach_type, should have the default attach
1303 * type assigned to expected_attach_type for the latter case, so that it can be
1304 * validated later at attach time.
1306 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1307 * prog type requires it but has some attach types that have to be backward
1310 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1312 switch (attr->prog_type) {
1313 case BPF_PROG_TYPE_CGROUP_SOCK:
1314 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1315 * exist so checking for non-zero is the way to go here.
1317 if (!attr->expected_attach_type)
1318 attr->expected_attach_type =
1319 BPF_CGROUP_INET_SOCK_CREATE;
1325 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1326 enum bpf_attach_type expected_attach_type)
1328 switch (prog_type) {
1329 case BPF_PROG_TYPE_CGROUP_SOCK:
1330 switch (expected_attach_type) {
1331 case BPF_CGROUP_INET_SOCK_CREATE:
1332 case BPF_CGROUP_INET4_POST_BIND:
1333 case BPF_CGROUP_INET6_POST_BIND:
1338 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1339 switch (expected_attach_type) {
1340 case BPF_CGROUP_INET4_BIND:
1341 case BPF_CGROUP_INET6_BIND:
1342 case BPF_CGROUP_INET4_CONNECT:
1343 case BPF_CGROUP_INET6_CONNECT:
1344 case BPF_CGROUP_UDP4_SENDMSG:
1345 case BPF_CGROUP_UDP6_SENDMSG:
1346 case BPF_CGROUP_UDP4_RECVMSG:
1347 case BPF_CGROUP_UDP6_RECVMSG:
1357 /* last field in 'union bpf_attr' used by this command */
1358 #define BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1360 static int bpf_prog_load(union bpf_attr *attr)
1362 enum bpf_prog_type type = attr->prog_type;
1363 struct bpf_prog *prog;
1368 if (CHECK_ATTR(BPF_PROG_LOAD))
1371 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT))
1374 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1375 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1376 !capable(CAP_SYS_ADMIN))
1379 /* copy eBPF program license from user space */
1380 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1381 sizeof(license) - 1) < 0)
1383 license[sizeof(license) - 1] = 0;
1385 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1386 is_gpl = license_is_gpl_compatible(license);
1388 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1391 if (type == BPF_PROG_TYPE_KPROBE &&
1392 attr->kern_version != LINUX_VERSION_CODE)
1395 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1396 type != BPF_PROG_TYPE_CGROUP_SKB &&
1397 !capable(CAP_SYS_ADMIN))
1400 bpf_prog_load_fixup_attach_type(attr);
1401 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1404 /* plain bpf_prog allocation */
1405 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1409 prog->expected_attach_type = attr->expected_attach_type;
1411 prog->aux->offload_requested = !!attr->prog_ifindex;
1413 err = security_bpf_prog_alloc(prog->aux);
1415 goto free_prog_nouncharge;
1417 err = bpf_prog_charge_memlock(prog);
1421 prog->len = attr->insn_cnt;
1424 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1425 bpf_prog_insn_size(prog)) != 0)
1428 prog->orig_prog = NULL;
1431 atomic_set(&prog->aux->refcnt, 1);
1432 prog->gpl_compatible = is_gpl ? 1 : 0;
1434 if (bpf_prog_is_dev_bound(prog->aux)) {
1435 err = bpf_prog_offload_init(prog, attr);
1440 /* find program type: socket_filter vs tracing_filter */
1441 err = find_prog_type(type, prog);
1445 prog->aux->load_time = ktime_get_boot_ns();
1446 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1450 /* run eBPF verifier */
1451 err = bpf_check(&prog, attr);
1453 goto free_used_maps;
1455 prog = bpf_prog_select_runtime(prog, &err);
1457 goto free_used_maps;
1459 err = bpf_prog_alloc_id(prog);
1461 goto free_used_maps;
1463 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
1464 * effectively publicly exposed. However, retrieving via
1465 * bpf_prog_get_fd_by_id() will take another reference,
1466 * therefore it cannot be gone underneath us.
1468 * Only for the time /after/ successful bpf_prog_new_fd()
1469 * and before returning to userspace, we might just hold
1470 * one reference and any parallel close on that fd could
1471 * rip everything out. Hence, below notifications must
1472 * happen before bpf_prog_new_fd().
1474 * Also, any failure handling from this point onwards must
1475 * be using bpf_prog_put() given the program is exposed.
1477 bpf_prog_kallsyms_add(prog);
1479 err = bpf_prog_new_fd(prog);
1485 bpf_prog_kallsyms_del_subprogs(prog);
1486 free_used_maps(prog->aux);
1488 bpf_prog_uncharge_memlock(prog);
1490 security_bpf_prog_free(prog->aux);
1491 free_prog_nouncharge:
1492 bpf_prog_free(prog);
1496 #define BPF_OBJ_LAST_FIELD file_flags
1498 static int bpf_obj_pin(const union bpf_attr *attr)
1500 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1503 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1506 static int bpf_obj_get(const union bpf_attr *attr)
1508 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1509 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1512 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1516 struct bpf_raw_tracepoint {
1517 struct bpf_raw_event_map *btp;
1518 struct bpf_prog *prog;
1521 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1523 struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1526 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1527 bpf_prog_put(raw_tp->prog);
1533 static const struct file_operations bpf_raw_tp_fops = {
1534 .release = bpf_raw_tracepoint_release,
1535 .read = bpf_dummy_read,
1536 .write = bpf_dummy_write,
1539 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1541 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1543 struct bpf_raw_tracepoint *raw_tp;
1544 struct bpf_raw_event_map *btp;
1545 struct bpf_prog *prog;
1549 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1550 sizeof(tp_name) - 1) < 0)
1552 tp_name[sizeof(tp_name) - 1] = 0;
1554 btp = bpf_find_raw_tracepoint(tp_name);
1558 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1563 prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1564 BPF_PROG_TYPE_RAW_TRACEPOINT);
1566 err = PTR_ERR(prog);
1570 err = bpf_probe_register(raw_tp->btp, prog);
1574 raw_tp->prog = prog;
1575 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1578 bpf_probe_unregister(raw_tp->btp, prog);
1591 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1592 enum bpf_attach_type attach_type)
1594 switch (prog->type) {
1595 case BPF_PROG_TYPE_CGROUP_SOCK:
1596 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1597 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1603 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1605 #define BPF_F_ATTACH_MASK \
1606 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1608 static int bpf_prog_attach(const union bpf_attr *attr)
1610 enum bpf_prog_type ptype;
1611 struct bpf_prog *prog;
1614 if (!capable(CAP_NET_ADMIN))
1617 if (CHECK_ATTR(BPF_PROG_ATTACH))
1620 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1623 switch (attr->attach_type) {
1624 case BPF_CGROUP_INET_INGRESS:
1625 case BPF_CGROUP_INET_EGRESS:
1626 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1628 case BPF_CGROUP_INET_SOCK_CREATE:
1629 case BPF_CGROUP_INET4_POST_BIND:
1630 case BPF_CGROUP_INET6_POST_BIND:
1631 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1633 case BPF_CGROUP_INET4_BIND:
1634 case BPF_CGROUP_INET6_BIND:
1635 case BPF_CGROUP_INET4_CONNECT:
1636 case BPF_CGROUP_INET6_CONNECT:
1637 case BPF_CGROUP_UDP4_SENDMSG:
1638 case BPF_CGROUP_UDP6_SENDMSG:
1639 case BPF_CGROUP_UDP4_RECVMSG:
1640 case BPF_CGROUP_UDP6_RECVMSG:
1641 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1643 case BPF_CGROUP_SOCK_OPS:
1644 ptype = BPF_PROG_TYPE_SOCK_OPS;
1646 case BPF_CGROUP_DEVICE:
1647 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1649 case BPF_SK_MSG_VERDICT:
1650 ptype = BPF_PROG_TYPE_SK_MSG;
1652 case BPF_SK_SKB_STREAM_PARSER:
1653 case BPF_SK_SKB_STREAM_VERDICT:
1654 ptype = BPF_PROG_TYPE_SK_SKB;
1656 case BPF_LIRC_MODE2:
1657 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1663 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1665 return PTR_ERR(prog);
1667 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1673 case BPF_PROG_TYPE_SK_SKB:
1674 case BPF_PROG_TYPE_SK_MSG:
1675 ret = sockmap_get_from_fd(attr, ptype, prog);
1677 case BPF_PROG_TYPE_LIRC_MODE2:
1678 ret = lirc_prog_attach(attr, prog);
1681 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1689 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1691 static int bpf_prog_detach(const union bpf_attr *attr)
1693 enum bpf_prog_type ptype;
1695 if (!capable(CAP_NET_ADMIN))
1698 if (CHECK_ATTR(BPF_PROG_DETACH))
1701 switch (attr->attach_type) {
1702 case BPF_CGROUP_INET_INGRESS:
1703 case BPF_CGROUP_INET_EGRESS:
1704 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1706 case BPF_CGROUP_INET_SOCK_CREATE:
1707 case BPF_CGROUP_INET4_POST_BIND:
1708 case BPF_CGROUP_INET6_POST_BIND:
1709 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1711 case BPF_CGROUP_INET4_BIND:
1712 case BPF_CGROUP_INET6_BIND:
1713 case BPF_CGROUP_INET4_CONNECT:
1714 case BPF_CGROUP_INET6_CONNECT:
1715 case BPF_CGROUP_UDP4_SENDMSG:
1716 case BPF_CGROUP_UDP6_SENDMSG:
1717 case BPF_CGROUP_UDP4_RECVMSG:
1718 case BPF_CGROUP_UDP6_RECVMSG:
1719 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1721 case BPF_CGROUP_SOCK_OPS:
1722 ptype = BPF_PROG_TYPE_SOCK_OPS;
1724 case BPF_CGROUP_DEVICE:
1725 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1727 case BPF_SK_MSG_VERDICT:
1728 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, NULL);
1729 case BPF_SK_SKB_STREAM_PARSER:
1730 case BPF_SK_SKB_STREAM_VERDICT:
1731 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, NULL);
1732 case BPF_LIRC_MODE2:
1733 return lirc_prog_detach(attr);
1738 return cgroup_bpf_prog_detach(attr, ptype);
1741 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1743 static int bpf_prog_query(const union bpf_attr *attr,
1744 union bpf_attr __user *uattr)
1746 if (!capable(CAP_NET_ADMIN))
1748 if (CHECK_ATTR(BPF_PROG_QUERY))
1750 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1753 switch (attr->query.attach_type) {
1754 case BPF_CGROUP_INET_INGRESS:
1755 case BPF_CGROUP_INET_EGRESS:
1756 case BPF_CGROUP_INET_SOCK_CREATE:
1757 case BPF_CGROUP_INET4_BIND:
1758 case BPF_CGROUP_INET6_BIND:
1759 case BPF_CGROUP_INET4_POST_BIND:
1760 case BPF_CGROUP_INET6_POST_BIND:
1761 case BPF_CGROUP_INET4_CONNECT:
1762 case BPF_CGROUP_INET6_CONNECT:
1763 case BPF_CGROUP_UDP4_SENDMSG:
1764 case BPF_CGROUP_UDP6_SENDMSG:
1765 case BPF_CGROUP_UDP4_RECVMSG:
1766 case BPF_CGROUP_UDP6_RECVMSG:
1767 case BPF_CGROUP_SOCK_OPS:
1768 case BPF_CGROUP_DEVICE:
1770 case BPF_LIRC_MODE2:
1771 return lirc_prog_query(attr, uattr);
1776 return cgroup_bpf_prog_query(attr, uattr);
1779 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1781 static int bpf_prog_test_run(const union bpf_attr *attr,
1782 union bpf_attr __user *uattr)
1784 struct bpf_prog *prog;
1785 int ret = -ENOTSUPP;
1787 if (!capable(CAP_SYS_ADMIN))
1789 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1792 prog = bpf_prog_get(attr->test.prog_fd);
1794 return PTR_ERR(prog);
1796 if (prog->aux->ops->test_run)
1797 ret = prog->aux->ops->test_run(prog, attr, uattr);
1803 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1805 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1806 union bpf_attr __user *uattr,
1810 u32 next_id = attr->start_id;
1813 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1816 if (!capable(CAP_SYS_ADMIN))
1821 if (!idr_get_next(idr, &next_id))
1823 spin_unlock_bh(lock);
1826 err = put_user(next_id, &uattr->next_id);
1831 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1833 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1835 struct bpf_prog *prog;
1836 u32 id = attr->prog_id;
1839 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1842 if (!capable(CAP_SYS_ADMIN))
1845 spin_lock_bh(&prog_idr_lock);
1846 prog = idr_find(&prog_idr, id);
1848 prog = bpf_prog_inc_not_zero(prog);
1850 prog = ERR_PTR(-ENOENT);
1851 spin_unlock_bh(&prog_idr_lock);
1854 return PTR_ERR(prog);
1856 fd = bpf_prog_new_fd(prog);
1863 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1865 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1867 struct bpf_map *map;
1868 u32 id = attr->map_id;
1872 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1873 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1876 if (!capable(CAP_SYS_ADMIN))
1879 f_flags = bpf_get_file_flag(attr->open_flags);
1883 spin_lock_bh(&map_idr_lock);
1884 map = idr_find(&map_idr, id);
1886 map = bpf_map_inc_not_zero(map, true);
1888 map = ERR_PTR(-ENOENT);
1889 spin_unlock_bh(&map_idr_lock);
1892 return PTR_ERR(map);
1894 fd = bpf_map_new_fd(map, f_flags);
1896 bpf_map_put_with_uref(map);
1901 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1906 for (i = 0; i < prog->aux->used_map_cnt; i++)
1907 if (prog->aux->used_maps[i] == (void *)addr)
1908 return prog->aux->used_maps[i];
1912 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
1913 const struct cred *f_cred)
1915 const struct bpf_map *map;
1916 struct bpf_insn *insns;
1920 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1925 for (i = 0; i < prog->len; i++) {
1926 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1927 insns[i].code = BPF_JMP | BPF_CALL;
1928 insns[i].imm = BPF_FUNC_tail_call;
1931 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1932 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1933 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1934 insns[i].code = BPF_JMP | BPF_CALL;
1935 if (!bpf_dump_raw_ok(f_cred))
1940 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1943 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1944 map = bpf_map_from_imm(prog, imm);
1946 insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1947 insns[i].imm = map->id;
1948 insns[i + 1].imm = 0;
1952 if (!bpf_dump_raw_ok(f_cred) &&
1953 imm == (unsigned long)prog->aux) {
1955 insns[i + 1].imm = 0;
1963 static int bpf_prog_get_info_by_fd(struct file *file,
1964 struct bpf_prog *prog,
1965 const union bpf_attr *attr,
1966 union bpf_attr __user *uattr)
1968 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1969 struct bpf_prog_info info;
1970 u32 info_len = attr->info.info_len;
1971 char __user *uinsns;
1975 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1978 info_len = min_t(u32, sizeof(info), info_len);
1980 memset(&info, 0, sizeof(info));
1981 if (copy_from_user(&info, uinfo, info_len))
1984 info.type = prog->type;
1985 info.id = prog->aux->id;
1986 info.load_time = prog->aux->load_time;
1987 info.created_by_uid = from_kuid_munged(current_user_ns(),
1988 prog->aux->user->uid);
1989 info.gpl_compatible = prog->gpl_compatible;
1991 memcpy(info.tag, prog->tag, sizeof(prog->tag));
1992 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1994 ulen = info.nr_map_ids;
1995 info.nr_map_ids = prog->aux->used_map_cnt;
1996 ulen = min_t(u32, info.nr_map_ids, ulen);
1998 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2001 for (i = 0; i < ulen; i++)
2002 if (put_user(prog->aux->used_maps[i]->id,
2007 if (!capable(CAP_SYS_ADMIN)) {
2008 info.jited_prog_len = 0;
2009 info.xlated_prog_len = 0;
2010 info.nr_jited_ksyms = 0;
2011 info.nr_jited_func_lens = 0;
2015 ulen = info.xlated_prog_len;
2016 info.xlated_prog_len = bpf_prog_insn_size(prog);
2017 if (info.xlated_prog_len && ulen) {
2018 struct bpf_insn *insns_sanitized;
2021 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
2022 info.xlated_prog_insns = 0;
2025 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
2026 if (!insns_sanitized)
2028 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2029 ulen = min_t(u32, info.xlated_prog_len, ulen);
2030 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2031 kfree(insns_sanitized);
2036 if (bpf_prog_is_dev_bound(prog->aux)) {
2037 err = bpf_prog_offload_info_fill(&info, prog);
2043 /* NOTE: the following code is supposed to be skipped for offload.
2044 * bpf_prog_offload_info_fill() is the place to fill similar fields
2047 ulen = info.jited_prog_len;
2048 if (prog->aux->func_cnt) {
2051 info.jited_prog_len = 0;
2052 for (i = 0; i < prog->aux->func_cnt; i++)
2053 info.jited_prog_len += prog->aux->func[i]->jited_len;
2055 info.jited_prog_len = prog->jited_len;
2058 if (info.jited_prog_len && ulen) {
2059 if (bpf_dump_raw_ok(file->f_cred)) {
2060 uinsns = u64_to_user_ptr(info.jited_prog_insns);
2061 ulen = min_t(u32, info.jited_prog_len, ulen);
2063 /* for multi-function programs, copy the JITed
2064 * instructions for all the functions
2066 if (prog->aux->func_cnt) {
2071 for (i = 0; i < prog->aux->func_cnt; i++) {
2072 len = prog->aux->func[i]->jited_len;
2073 len = min_t(u32, len, free);
2074 img = (u8 *) prog->aux->func[i]->bpf_func;
2075 if (copy_to_user(uinsns, img, len))
2083 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2087 info.jited_prog_insns = 0;
2091 ulen = info.nr_jited_ksyms;
2092 info.nr_jited_ksyms = prog->aux->func_cnt;
2093 if (info.nr_jited_ksyms && ulen) {
2094 if (bpf_dump_raw_ok(file->f_cred)) {
2095 u64 __user *user_ksyms;
2099 /* copy the address of the kernel symbol
2100 * corresponding to each function
2102 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2103 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2104 for (i = 0; i < ulen; i++) {
2105 ksym_addr = (ulong) prog->aux->func[i]->bpf_func;
2106 ksym_addr &= PAGE_MASK;
2107 if (put_user((u64) ksym_addr, &user_ksyms[i]))
2111 info.jited_ksyms = 0;
2115 ulen = info.nr_jited_func_lens;
2116 info.nr_jited_func_lens = prog->aux->func_cnt;
2117 if (info.nr_jited_func_lens && ulen) {
2118 if (bpf_dump_raw_ok(file->f_cred)) {
2119 u32 __user *user_lens;
2122 /* copy the JITed image lengths for each function */
2123 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2124 user_lens = u64_to_user_ptr(info.jited_func_lens);
2125 for (i = 0; i < ulen; i++) {
2126 func_len = prog->aux->func[i]->jited_len;
2127 if (put_user(func_len, &user_lens[i]))
2131 info.jited_func_lens = 0;
2136 if (copy_to_user(uinfo, &info, info_len) ||
2137 put_user(info_len, &uattr->info.info_len))
2143 static int bpf_map_get_info_by_fd(struct file *file,
2144 struct bpf_map *map,
2145 const union bpf_attr *attr,
2146 union bpf_attr __user *uattr)
2148 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2149 struct bpf_map_info info;
2150 u32 info_len = attr->info.info_len;
2153 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2156 info_len = min_t(u32, sizeof(info), info_len);
2158 memset(&info, 0, sizeof(info));
2159 info.type = map->map_type;
2161 info.key_size = map->key_size;
2162 info.value_size = map->value_size;
2163 info.max_entries = map->max_entries;
2164 info.map_flags = map->map_flags;
2165 memcpy(info.name, map->name, sizeof(map->name));
2168 info.btf_id = btf_id(map->btf);
2169 info.btf_key_type_id = map->btf_key_type_id;
2170 info.btf_value_type_id = map->btf_value_type_id;
2173 if (bpf_map_is_dev_bound(map)) {
2174 err = bpf_map_offload_info_fill(&info, map);
2179 if (copy_to_user(uinfo, &info, info_len) ||
2180 put_user(info_len, &uattr->info.info_len))
2186 static int bpf_btf_get_info_by_fd(struct file *file,
2188 const union bpf_attr *attr,
2189 union bpf_attr __user *uattr)
2191 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2192 u32 info_len = attr->info.info_len;
2195 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2199 return btf_get_info_by_fd(btf, attr, uattr);
2202 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2204 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2205 union bpf_attr __user *uattr)
2207 int ufd = attr->info.bpf_fd;
2211 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2218 if (f.file->f_op == &bpf_prog_fops)
2219 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
2221 else if (f.file->f_op == &bpf_map_fops)
2222 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
2224 else if (f.file->f_op == &btf_fops)
2225 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
2233 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2235 static int bpf_btf_load(const union bpf_attr *attr)
2237 if (CHECK_ATTR(BPF_BTF_LOAD))
2240 if (!capable(CAP_SYS_ADMIN))
2243 return btf_new_fd(attr);
2246 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2248 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2250 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2253 if (!capable(CAP_SYS_ADMIN))
2256 return btf_get_fd_by_id(attr->btf_id);
2259 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2260 union bpf_attr __user *uattr,
2261 u32 prog_id, u32 fd_type,
2262 const char *buf, u64 probe_offset,
2265 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2266 u32 len = buf ? strlen(buf) : 0, input_len;
2269 if (put_user(len, &uattr->task_fd_query.buf_len))
2271 input_len = attr->task_fd_query.buf_len;
2272 if (input_len && ubuf) {
2274 /* nothing to copy, just make ubuf NULL terminated */
2277 if (put_user(zero, ubuf))
2279 } else if (input_len >= len + 1) {
2280 /* ubuf can hold the string with NULL terminator */
2281 if (copy_to_user(ubuf, buf, len + 1))
2284 /* ubuf cannot hold the string with NULL terminator,
2285 * do a partial copy with NULL terminator.
2290 if (copy_to_user(ubuf, buf, input_len - 1))
2292 if (put_user(zero, ubuf + input_len - 1))
2297 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2298 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2299 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2300 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2306 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2308 static int bpf_task_fd_query(const union bpf_attr *attr,
2309 union bpf_attr __user *uattr)
2311 pid_t pid = attr->task_fd_query.pid;
2312 u32 fd = attr->task_fd_query.fd;
2313 const struct perf_event *event;
2314 struct files_struct *files;
2315 struct task_struct *task;
2319 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2322 if (!capable(CAP_SYS_ADMIN))
2325 if (attr->task_fd_query.flags != 0)
2329 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2334 files = get_files_struct(task);
2335 put_task_struct(task);
2340 spin_lock(&files->file_lock);
2341 file = fcheck_files(files, fd);
2346 spin_unlock(&files->file_lock);
2347 put_files_struct(files);
2352 if (file->f_op == &bpf_raw_tp_fops) {
2353 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2354 struct bpf_raw_event_map *btp = raw_tp->btp;
2356 err = bpf_task_fd_query_copy(attr, uattr,
2357 raw_tp->prog->aux->id,
2358 BPF_FD_TYPE_RAW_TRACEPOINT,
2359 btp->tp->name, 0, 0);
2363 event = perf_get_event(file);
2364 if (!IS_ERR(event)) {
2365 u64 probe_offset, probe_addr;
2366 u32 prog_id, fd_type;
2369 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2370 &buf, &probe_offset,
2373 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2387 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2389 union bpf_attr attr;
2392 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2395 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2398 size = min_t(u32, size, sizeof(attr));
2400 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2401 memset(&attr, 0, sizeof(attr));
2402 if (copy_from_user(&attr, uattr, size) != 0)
2405 err = security_bpf(cmd, &attr, size);
2410 case BPF_MAP_CREATE:
2411 err = map_create(&attr);
2413 case BPF_MAP_LOOKUP_ELEM:
2414 err = map_lookup_elem(&attr);
2416 case BPF_MAP_UPDATE_ELEM:
2417 err = map_update_elem(&attr);
2419 case BPF_MAP_DELETE_ELEM:
2420 err = map_delete_elem(&attr);
2422 case BPF_MAP_GET_NEXT_KEY:
2423 err = map_get_next_key(&attr);
2426 err = bpf_prog_load(&attr);
2429 err = bpf_obj_pin(&attr);
2432 err = bpf_obj_get(&attr);
2434 case BPF_PROG_ATTACH:
2435 err = bpf_prog_attach(&attr);
2437 case BPF_PROG_DETACH:
2438 err = bpf_prog_detach(&attr);
2440 case BPF_PROG_QUERY:
2441 err = bpf_prog_query(&attr, uattr);
2443 case BPF_PROG_TEST_RUN:
2444 err = bpf_prog_test_run(&attr, uattr);
2446 case BPF_PROG_GET_NEXT_ID:
2447 err = bpf_obj_get_next_id(&attr, uattr,
2448 &prog_idr, &prog_idr_lock);
2450 case BPF_MAP_GET_NEXT_ID:
2451 err = bpf_obj_get_next_id(&attr, uattr,
2452 &map_idr, &map_idr_lock);
2454 case BPF_PROG_GET_FD_BY_ID:
2455 err = bpf_prog_get_fd_by_id(&attr);
2457 case BPF_MAP_GET_FD_BY_ID:
2458 err = bpf_map_get_fd_by_id(&attr);
2460 case BPF_OBJ_GET_INFO_BY_FD:
2461 err = bpf_obj_get_info_by_fd(&attr, uattr);
2463 case BPF_RAW_TRACEPOINT_OPEN:
2464 err = bpf_raw_tracepoint_open(&attr);
2467 err = bpf_btf_load(&attr);
2469 case BPF_BTF_GET_FD_BY_ID:
2470 err = bpf_btf_get_fd_by_id(&attr);
2472 case BPF_TASK_FD_QUERY:
2473 err = bpf_task_fd_query(&attr, uattr);