1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of version 2 of the GNU General Public
5 * License as published by the Free Software Foundation.
7 * This program is distributed in the hope that it will be useful, but
8 * WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
10 * General Public License for more details.
12 #include <linux/bpf.h>
13 #include <linux/bpf_trace.h>
14 #include <linux/bpf_lirc.h>
15 #include <linux/btf.h>
16 #include <linux/syscalls.h>
17 #include <linux/slab.h>
18 #include <linux/sched/signal.h>
19 #include <linux/vmalloc.h>
20 #include <linux/mmzone.h>
21 #include <linux/anon_inodes.h>
22 #include <linux/fdtable.h>
23 #include <linux/file.h>
25 #include <linux/license.h>
26 #include <linux/filter.h>
27 #include <linux/version.h>
28 #include <linux/kernel.h>
29 #include <linux/idr.h>
30 #include <linux/cred.h>
31 #include <linux/timekeeping.h>
32 #include <linux/ctype.h>
33 #include <linux/nospec.h>
35 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
36 (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
37 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
38 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
39 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
40 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
42 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
44 DEFINE_PER_CPU(int, bpf_prog_active);
45 static DEFINE_IDR(prog_idr);
46 static DEFINE_SPINLOCK(prog_idr_lock);
47 static DEFINE_IDR(map_idr);
48 static DEFINE_SPINLOCK(map_idr_lock);
50 int sysctl_unprivileged_bpf_disabled __read_mostly;
52 static const struct bpf_map_ops * const bpf_map_types[] = {
53 #define BPF_PROG_TYPE(_id, _ops)
54 #define BPF_MAP_TYPE(_id, _ops) \
56 #include <linux/bpf_types.h>
62 * If we're handed a bigger struct than we know of, ensure all the unknown bits
63 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
64 * we don't know about yet.
66 * There is a ToCToU between this function call and the following
67 * copy_from_user() call. However, this is not a concern since this function is
68 * meant to be a future-proofing of bits.
70 int bpf_check_uarg_tail_zero(void __user *uaddr,
74 unsigned char __user *addr;
75 unsigned char __user *end;
79 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
82 if (unlikely(!access_ok(uaddr, actual_size)))
85 if (actual_size <= expected_size)
88 addr = uaddr + expected_size;
89 end = uaddr + actual_size;
91 for (; addr < end; addr++) {
92 err = get_user(val, addr);
102 const struct bpf_map_ops bpf_map_offload_ops = {
103 .map_alloc = bpf_map_offload_map_alloc,
104 .map_free = bpf_map_offload_map_free,
105 .map_check_btf = map_check_no_btf,
108 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
110 const struct bpf_map_ops *ops;
111 u32 type = attr->map_type;
115 if (type >= ARRAY_SIZE(bpf_map_types))
116 return ERR_PTR(-EINVAL);
117 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
118 ops = bpf_map_types[type];
120 return ERR_PTR(-EINVAL);
122 if (ops->map_alloc_check) {
123 err = ops->map_alloc_check(attr);
127 if (attr->map_ifindex)
128 ops = &bpf_map_offload_ops;
129 map = ops->map_alloc(attr);
133 map->map_type = type;
137 void *bpf_map_area_alloc(size_t size, int numa_node)
139 /* We definitely need __GFP_NORETRY, so OOM killer doesn't
140 * trigger under memory pressure as we really just want to
143 const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
146 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
147 area = kmalloc_node(size, GFP_USER | flags, numa_node);
152 return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
153 __builtin_return_address(0));
156 void bpf_map_area_free(void *area)
161 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
163 map->map_type = attr->map_type;
164 map->key_size = attr->key_size;
165 map->value_size = attr->value_size;
166 map->max_entries = attr->max_entries;
167 map->map_flags = attr->map_flags;
168 map->numa_node = bpf_map_attr_numa_node(attr);
171 int bpf_map_precharge_memlock(u32 pages)
173 struct user_struct *user = get_current_user();
174 unsigned long memlock_limit, cur;
176 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
177 cur = atomic_long_read(&user->locked_vm);
179 if (cur + pages > memlock_limit)
184 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
186 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
188 if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
189 atomic_long_sub(pages, &user->locked_vm);
195 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
197 atomic_long_sub(pages, &user->locked_vm);
200 static int bpf_map_init_memlock(struct bpf_map *map)
202 struct user_struct *user = get_current_user();
205 ret = bpf_charge_memlock(user, map->pages);
214 static void bpf_map_release_memlock(struct bpf_map *map)
216 struct user_struct *user = map->user;
217 bpf_uncharge_memlock(user, map->pages);
221 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
225 ret = bpf_charge_memlock(map->user, pages);
232 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
234 bpf_uncharge_memlock(map->user, pages);
238 static int bpf_map_alloc_id(struct bpf_map *map)
242 idr_preload(GFP_KERNEL);
243 spin_lock_bh(&map_idr_lock);
244 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
247 spin_unlock_bh(&map_idr_lock);
250 if (WARN_ON_ONCE(!id))
253 return id > 0 ? 0 : id;
256 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
260 /* Offloaded maps are removed from the IDR store when their device
261 * disappears - even if someone holds an fd to them they are unusable,
262 * the memory is gone, all ops will fail; they are simply waiting for
263 * refcnt to drop to be freed.
269 spin_lock_irqsave(&map_idr_lock, flags);
271 __acquire(&map_idr_lock);
273 idr_remove(&map_idr, map->id);
277 spin_unlock_irqrestore(&map_idr_lock, flags);
279 __release(&map_idr_lock);
282 /* called from workqueue */
283 static void bpf_map_free_deferred(struct work_struct *work)
285 struct bpf_map *map = container_of(work, struct bpf_map, work);
287 bpf_map_release_memlock(map);
288 security_bpf_map_free(map);
289 /* implementation dependent freeing */
290 map->ops->map_free(map);
293 static void bpf_map_put_uref(struct bpf_map *map)
295 if (atomic_dec_and_test(&map->usercnt)) {
296 if (map->ops->map_release_uref)
297 map->ops->map_release_uref(map);
301 /* decrement map refcnt and schedule it for freeing via workqueue
302 * (unrelying map implementation ops->map_free() might sleep)
304 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
306 if (atomic_dec_and_test(&map->refcnt)) {
307 /* bpf_map_free_id() must be called first */
308 bpf_map_free_id(map, do_idr_lock);
310 INIT_WORK(&map->work, bpf_map_free_deferred);
311 schedule_work(&map->work);
315 void bpf_map_put(struct bpf_map *map)
317 __bpf_map_put(map, true);
319 EXPORT_SYMBOL_GPL(bpf_map_put);
321 void bpf_map_put_with_uref(struct bpf_map *map)
323 bpf_map_put_uref(map);
327 static int bpf_map_release(struct inode *inode, struct file *filp)
329 struct bpf_map *map = filp->private_data;
331 if (map->ops->map_release)
332 map->ops->map_release(map, filp);
334 bpf_map_put_with_uref(map);
338 #ifdef CONFIG_PROC_FS
339 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
341 const struct bpf_map *map = filp->private_data;
342 const struct bpf_array *array;
343 u32 owner_prog_type = 0;
346 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
347 array = container_of(map, struct bpf_array, map);
348 owner_prog_type = array->owner_prog_type;
349 owner_jited = array->owner_jited;
365 map->pages * 1ULL << PAGE_SHIFT,
368 if (owner_prog_type) {
369 seq_printf(m, "owner_prog_type:\t%u\n",
371 seq_printf(m, "owner_jited:\t%u\n",
377 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
380 /* We need this handler such that alloc_file() enables
381 * f_mode with FMODE_CAN_READ.
386 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
387 size_t siz, loff_t *ppos)
389 /* We need this handler such that alloc_file() enables
390 * f_mode with FMODE_CAN_WRITE.
395 const struct file_operations bpf_map_fops = {
396 #ifdef CONFIG_PROC_FS
397 .show_fdinfo = bpf_map_show_fdinfo,
399 .release = bpf_map_release,
400 .read = bpf_dummy_read,
401 .write = bpf_dummy_write,
404 int bpf_map_new_fd(struct bpf_map *map, int flags)
408 ret = security_bpf_map(map, OPEN_FMODE(flags));
412 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
416 int bpf_get_file_flag(int flags)
418 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
420 if (flags & BPF_F_RDONLY)
422 if (flags & BPF_F_WRONLY)
427 /* helper macro to check that unused fields 'union bpf_attr' are zero */
428 #define CHECK_ATTR(CMD) \
429 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
430 sizeof(attr->CMD##_LAST_FIELD), 0, \
432 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
433 sizeof(attr->CMD##_LAST_FIELD)) != NULL
435 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
436 * Return 0 on success and < 0 on error.
438 static int bpf_obj_name_cpy(char *dst, const char *src)
440 const char *end = src + BPF_OBJ_NAME_LEN;
442 memset(dst, 0, BPF_OBJ_NAME_LEN);
444 /* Copy all isalnum() and '_' char */
445 while (src < end && *src) {
446 if (!isalnum(*src) && *src != '_')
451 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
458 int map_check_no_btf(const struct bpf_map *map,
459 const struct btf *btf,
460 const struct btf_type *key_type,
461 const struct btf_type *value_type)
466 static int map_check_btf(const struct bpf_map *map, const struct btf *btf,
467 u32 btf_key_id, u32 btf_value_id)
469 const struct btf_type *key_type, *value_type;
470 u32 key_size, value_size;
473 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
474 if (!key_type || key_size != map->key_size)
477 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
478 if (!value_type || value_size != map->value_size)
481 if (map->ops->map_check_btf)
482 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
487 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
488 /* called via syscall */
489 static int map_create(union bpf_attr *attr)
491 int numa_node = bpf_map_attr_numa_node(attr);
496 err = CHECK_ATTR(BPF_MAP_CREATE);
500 f_flags = bpf_get_file_flag(attr->map_flags);
504 if (numa_node != NUMA_NO_NODE &&
505 ((unsigned int)numa_node >= nr_node_ids ||
506 !node_online(numa_node)))
509 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
510 map = find_and_alloc_map(attr);
514 err = bpf_obj_name_cpy(map->name, attr->map_name);
516 goto free_map_nouncharge;
518 atomic_set(&map->refcnt, 1);
519 atomic_set(&map->usercnt, 1);
521 if (attr->btf_key_type_id || attr->btf_value_type_id) {
524 if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
526 goto free_map_nouncharge;
529 btf = btf_get_by_fd(attr->btf_fd);
532 goto free_map_nouncharge;
535 err = map_check_btf(map, btf, attr->btf_key_type_id,
536 attr->btf_value_type_id);
539 goto free_map_nouncharge;
543 map->btf_key_type_id = attr->btf_key_type_id;
544 map->btf_value_type_id = attr->btf_value_type_id;
547 err = security_bpf_map_alloc(map);
549 goto free_map_nouncharge;
551 err = bpf_map_init_memlock(map);
555 err = bpf_map_alloc_id(map);
559 err = bpf_map_new_fd(map, f_flags);
561 /* failed to allocate fd.
562 * bpf_map_put() is needed because the above
563 * bpf_map_alloc_id() has published the map
564 * to the userspace and the userspace may
565 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
574 bpf_map_release_memlock(map);
576 security_bpf_map_free(map);
579 map->ops->map_free(map);
583 /* if error is returned, fd is released.
584 * On success caller should complete fd access with matching fdput()
586 struct bpf_map *__bpf_map_get(struct fd f)
589 return ERR_PTR(-EBADF);
590 if (f.file->f_op != &bpf_map_fops) {
592 return ERR_PTR(-EINVAL);
595 return f.file->private_data;
598 /* prog's and map's refcnt limit */
599 #define BPF_MAX_REFCNT 32768
601 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
603 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
604 atomic_dec(&map->refcnt);
605 return ERR_PTR(-EBUSY);
608 atomic_inc(&map->usercnt);
611 EXPORT_SYMBOL_GPL(bpf_map_inc);
613 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
615 struct fd f = fdget(ufd);
618 map = __bpf_map_get(f);
622 map = bpf_map_inc(map, true);
628 /* map_idr_lock should have been held */
629 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
634 refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
636 if (refold >= BPF_MAX_REFCNT) {
637 __bpf_map_put(map, false);
638 return ERR_PTR(-EBUSY);
642 return ERR_PTR(-ENOENT);
645 atomic_inc(&map->usercnt);
650 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
655 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
658 return memdup_user(ukey, key_size);
661 return ERR_PTR(-EINVAL);
666 /* last field in 'union bpf_attr' used by this command */
667 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
669 static int map_lookup_elem(union bpf_attr *attr)
671 void __user *ukey = u64_to_user_ptr(attr->key);
672 void __user *uvalue = u64_to_user_ptr(attr->value);
673 int ufd = attr->map_fd;
675 void *key, *value, *ptr;
680 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
684 map = __bpf_map_get(f);
688 if (!(f.file->f_mode & FMODE_CAN_READ)) {
693 key = __bpf_copy_key(ukey, map->key_size);
699 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
700 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
701 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
702 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
703 value_size = round_up(map->value_size, 8) * num_possible_cpus();
704 else if (IS_FD_MAP(map))
705 value_size = sizeof(u32);
707 value_size = map->value_size;
710 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
714 if (bpf_map_is_dev_bound(map)) {
715 err = bpf_map_offload_lookup_elem(map, key, value);
716 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
717 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
718 err = bpf_percpu_hash_copy(map, key, value);
719 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
720 err = bpf_percpu_array_copy(map, key, value);
721 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
722 err = bpf_percpu_cgroup_storage_copy(map, key, value);
723 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
724 err = bpf_stackmap_copy(map, key, value);
725 } else if (IS_FD_ARRAY(map)) {
726 err = bpf_fd_array_map_lookup_elem(map, key, value);
727 } else if (IS_FD_HASH(map)) {
728 err = bpf_fd_htab_map_lookup_elem(map, key, value);
729 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
730 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
731 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
732 map->map_type == BPF_MAP_TYPE_STACK) {
733 err = map->ops->map_peek_elem(map, value);
736 ptr = map->ops->map_lookup_elem(map, key);
743 memcpy(value, ptr, value_size);
752 if (copy_to_user(uvalue, value, value_size) != 0)
766 static void maybe_wait_bpf_programs(struct bpf_map *map)
768 /* Wait for any running BPF programs to complete so that
769 * userspace, when we return to it, knows that all programs
770 * that could be running use the new map value.
772 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
773 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
777 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
779 static int map_update_elem(union bpf_attr *attr)
781 void __user *ukey = u64_to_user_ptr(attr->key);
782 void __user *uvalue = u64_to_user_ptr(attr->value);
783 int ufd = attr->map_fd;
790 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
794 map = __bpf_map_get(f);
798 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
803 key = __bpf_copy_key(ukey, map->key_size);
809 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
810 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
811 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
812 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
813 value_size = round_up(map->value_size, 8) * num_possible_cpus();
815 value_size = map->value_size;
818 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
823 if (copy_from_user(value, uvalue, value_size) != 0)
826 /* Need to create a kthread, thus must support schedule */
827 if (bpf_map_is_dev_bound(map)) {
828 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
830 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
831 map->map_type == BPF_MAP_TYPE_SOCKHASH ||
832 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
833 err = map->ops->map_update_elem(map, key, value, attr->flags);
837 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
838 * inside bpf map update or delete otherwise deadlocks are possible
841 __this_cpu_inc(bpf_prog_active);
842 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
843 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
844 err = bpf_percpu_hash_update(map, key, value, attr->flags);
845 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
846 err = bpf_percpu_array_update(map, key, value, attr->flags);
847 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
848 err = bpf_percpu_cgroup_storage_update(map, key, value,
850 } else if (IS_FD_ARRAY(map)) {
852 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
855 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
857 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
860 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
861 /* rcu_read_lock() is not needed */
862 err = bpf_fd_reuseport_array_update_elem(map, key, value,
864 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
865 map->map_type == BPF_MAP_TYPE_STACK) {
866 err = map->ops->map_push_elem(map, value, attr->flags);
869 err = map->ops->map_update_elem(map, key, value, attr->flags);
872 __this_cpu_dec(bpf_prog_active);
874 maybe_wait_bpf_programs(map);
885 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
887 static int map_delete_elem(union bpf_attr *attr)
889 void __user *ukey = u64_to_user_ptr(attr->key);
890 int ufd = attr->map_fd;
896 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
900 map = __bpf_map_get(f);
904 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
909 key = __bpf_copy_key(ukey, map->key_size);
915 if (bpf_map_is_dev_bound(map)) {
916 err = bpf_map_offload_delete_elem(map, key);
921 __this_cpu_inc(bpf_prog_active);
923 err = map->ops->map_delete_elem(map, key);
925 __this_cpu_dec(bpf_prog_active);
927 maybe_wait_bpf_programs(map);
935 /* last field in 'union bpf_attr' used by this command */
936 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
938 static int map_get_next_key(union bpf_attr *attr)
940 void __user *ukey = u64_to_user_ptr(attr->key);
941 void __user *unext_key = u64_to_user_ptr(attr->next_key);
942 int ufd = attr->map_fd;
944 void *key, *next_key;
948 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
952 map = __bpf_map_get(f);
956 if (!(f.file->f_mode & FMODE_CAN_READ)) {
962 key = __bpf_copy_key(ukey, map->key_size);
972 next_key = kmalloc(map->key_size, GFP_USER);
976 if (bpf_map_is_dev_bound(map)) {
977 err = bpf_map_offload_get_next_key(map, key, next_key);
982 err = map->ops->map_get_next_key(map, key, next_key);
989 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1003 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1005 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1007 void __user *ukey = u64_to_user_ptr(attr->key);
1008 void __user *uvalue = u64_to_user_ptr(attr->value);
1009 int ufd = attr->map_fd;
1010 struct bpf_map *map;
1016 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1020 map = __bpf_map_get(f);
1022 return PTR_ERR(map);
1024 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
1029 key = __bpf_copy_key(ukey, map->key_size);
1035 value_size = map->value_size;
1038 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1042 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1043 map->map_type == BPF_MAP_TYPE_STACK) {
1044 err = map->ops->map_pop_elem(map, value);
1052 if (copy_to_user(uvalue, value, value_size) != 0)
1066 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1067 #define BPF_PROG_TYPE(_id, _name) \
1068 [_id] = & _name ## _prog_ops,
1069 #define BPF_MAP_TYPE(_id, _ops)
1070 #include <linux/bpf_types.h>
1071 #undef BPF_PROG_TYPE
1075 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1077 const struct bpf_prog_ops *ops;
1079 if (type >= ARRAY_SIZE(bpf_prog_types))
1081 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1082 ops = bpf_prog_types[type];
1086 if (!bpf_prog_is_dev_bound(prog->aux))
1087 prog->aux->ops = ops;
1089 prog->aux->ops = &bpf_offload_prog_ops;
1094 /* drop refcnt on maps used by eBPF program and free auxilary data */
1095 static void free_used_maps(struct bpf_prog_aux *aux)
1097 enum bpf_cgroup_storage_type stype;
1100 for_each_cgroup_storage_type(stype) {
1101 if (!aux->cgroup_storage[stype])
1103 bpf_cgroup_storage_release(aux->prog,
1104 aux->cgroup_storage[stype]);
1107 for (i = 0; i < aux->used_map_cnt; i++)
1108 bpf_map_put(aux->used_maps[i]);
1110 kfree(aux->used_maps);
1113 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1115 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1116 unsigned long user_bufs;
1119 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1120 if (user_bufs > memlock_limit) {
1121 atomic_long_sub(pages, &user->locked_vm);
1129 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1132 atomic_long_sub(pages, &user->locked_vm);
1135 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1137 struct user_struct *user = get_current_user();
1140 ret = __bpf_prog_charge(user, prog->pages);
1146 prog->aux->user = user;
1150 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1152 struct user_struct *user = prog->aux->user;
1154 __bpf_prog_uncharge(user, prog->pages);
1158 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1162 idr_preload(GFP_KERNEL);
1163 spin_lock_bh(&prog_idr_lock);
1164 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1167 spin_unlock_bh(&prog_idr_lock);
1170 /* id is in [1, INT_MAX) */
1171 if (WARN_ON_ONCE(!id))
1174 return id > 0 ? 0 : id;
1177 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1179 /* cBPF to eBPF migrations are currently not in the idr store.
1180 * Offloaded programs are removed from the store when their device
1181 * disappears - even if someone grabs an fd to them they are unusable,
1182 * simply waiting for refcnt to drop to be freed.
1188 spin_lock_bh(&prog_idr_lock);
1190 __acquire(&prog_idr_lock);
1192 idr_remove(&prog_idr, prog->aux->id);
1196 spin_unlock_bh(&prog_idr_lock);
1198 __release(&prog_idr_lock);
1201 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1203 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1205 free_used_maps(aux);
1206 bpf_prog_uncharge_memlock(aux->prog);
1207 security_bpf_prog_free(aux);
1208 bpf_prog_free(aux->prog);
1211 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1213 if (atomic_dec_and_test(&prog->aux->refcnt)) {
1214 /* bpf_prog_free_id() must be called first */
1215 bpf_prog_free_id(prog, do_idr_lock);
1216 bpf_prog_kallsyms_del_all(prog);
1217 btf_put(prog->aux->btf);
1218 kvfree(prog->aux->func_info);
1219 bpf_prog_free_linfo(prog);
1221 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1225 void bpf_prog_put(struct bpf_prog *prog)
1227 __bpf_prog_put(prog, true);
1229 EXPORT_SYMBOL_GPL(bpf_prog_put);
1231 static int bpf_prog_release(struct inode *inode, struct file *filp)
1233 struct bpf_prog *prog = filp->private_data;
1239 #ifdef CONFIG_PROC_FS
1240 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1242 const struct bpf_prog *prog = filp->private_data;
1243 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1245 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1255 prog->pages * 1ULL << PAGE_SHIFT,
1260 const struct file_operations bpf_prog_fops = {
1261 #ifdef CONFIG_PROC_FS
1262 .show_fdinfo = bpf_prog_show_fdinfo,
1264 .release = bpf_prog_release,
1265 .read = bpf_dummy_read,
1266 .write = bpf_dummy_write,
1269 int bpf_prog_new_fd(struct bpf_prog *prog)
1273 ret = security_bpf_prog(prog);
1277 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1278 O_RDWR | O_CLOEXEC);
1281 static struct bpf_prog *____bpf_prog_get(struct fd f)
1284 return ERR_PTR(-EBADF);
1285 if (f.file->f_op != &bpf_prog_fops) {
1287 return ERR_PTR(-EINVAL);
1290 return f.file->private_data;
1293 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1295 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1296 atomic_sub(i, &prog->aux->refcnt);
1297 return ERR_PTR(-EBUSY);
1301 EXPORT_SYMBOL_GPL(bpf_prog_add);
1303 void bpf_prog_sub(struct bpf_prog *prog, int i)
1305 /* Only to be used for undoing previous bpf_prog_add() in some
1306 * error path. We still know that another entity in our call
1307 * path holds a reference to the program, thus atomic_sub() can
1308 * be safely used in such cases!
1310 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1312 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1314 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1316 return bpf_prog_add(prog, 1);
1318 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1320 /* prog_idr_lock should have been held */
1321 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1325 refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1327 if (refold >= BPF_MAX_REFCNT) {
1328 __bpf_prog_put(prog, false);
1329 return ERR_PTR(-EBUSY);
1333 return ERR_PTR(-ENOENT);
1337 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1339 bool bpf_prog_get_ok(struct bpf_prog *prog,
1340 enum bpf_prog_type *attach_type, bool attach_drv)
1342 /* not an attachment, just a refcount inc, always allow */
1346 if (prog->type != *attach_type)
1348 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1354 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1357 struct fd f = fdget(ufd);
1358 struct bpf_prog *prog;
1360 prog = ____bpf_prog_get(f);
1363 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1364 prog = ERR_PTR(-EINVAL);
1368 prog = bpf_prog_inc(prog);
1374 struct bpf_prog *bpf_prog_get(u32 ufd)
1376 return __bpf_prog_get(ufd, NULL, false);
1379 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1382 return __bpf_prog_get(ufd, &type, attach_drv);
1384 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1386 /* Initially all BPF programs could be loaded w/o specifying
1387 * expected_attach_type. Later for some of them specifying expected_attach_type
1388 * at load time became required so that program could be validated properly.
1389 * Programs of types that are allowed to be loaded both w/ and w/o (for
1390 * backward compatibility) expected_attach_type, should have the default attach
1391 * type assigned to expected_attach_type for the latter case, so that it can be
1392 * validated later at attach time.
1394 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1395 * prog type requires it but has some attach types that have to be backward
1398 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1400 switch (attr->prog_type) {
1401 case BPF_PROG_TYPE_CGROUP_SOCK:
1402 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1403 * exist so checking for non-zero is the way to go here.
1405 if (!attr->expected_attach_type)
1406 attr->expected_attach_type =
1407 BPF_CGROUP_INET_SOCK_CREATE;
1413 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1414 enum bpf_attach_type expected_attach_type)
1416 switch (prog_type) {
1417 case BPF_PROG_TYPE_CGROUP_SOCK:
1418 switch (expected_attach_type) {
1419 case BPF_CGROUP_INET_SOCK_CREATE:
1420 case BPF_CGROUP_INET4_POST_BIND:
1421 case BPF_CGROUP_INET6_POST_BIND:
1426 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1427 switch (expected_attach_type) {
1428 case BPF_CGROUP_INET4_BIND:
1429 case BPF_CGROUP_INET6_BIND:
1430 case BPF_CGROUP_INET4_CONNECT:
1431 case BPF_CGROUP_INET6_CONNECT:
1432 case BPF_CGROUP_UDP4_SENDMSG:
1433 case BPF_CGROUP_UDP6_SENDMSG:
1443 /* last field in 'union bpf_attr' used by this command */
1444 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1446 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1448 enum bpf_prog_type type = attr->prog_type;
1449 struct bpf_prog *prog;
1454 if (CHECK_ATTR(BPF_PROG_LOAD))
1457 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT))
1460 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1461 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1462 !capable(CAP_SYS_ADMIN))
1465 /* copy eBPF program license from user space */
1466 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1467 sizeof(license) - 1) < 0)
1469 license[sizeof(license) - 1] = 0;
1471 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1472 is_gpl = license_is_gpl_compatible(license);
1474 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1476 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1477 type != BPF_PROG_TYPE_CGROUP_SKB &&
1478 !capable(CAP_SYS_ADMIN))
1481 bpf_prog_load_fixup_attach_type(attr);
1482 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1485 /* plain bpf_prog allocation */
1486 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1490 prog->expected_attach_type = attr->expected_attach_type;
1492 prog->aux->offload_requested = !!attr->prog_ifindex;
1494 err = security_bpf_prog_alloc(prog->aux);
1496 goto free_prog_nouncharge;
1498 err = bpf_prog_charge_memlock(prog);
1502 prog->len = attr->insn_cnt;
1505 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1506 bpf_prog_insn_size(prog)) != 0)
1509 prog->orig_prog = NULL;
1512 atomic_set(&prog->aux->refcnt, 1);
1513 prog->gpl_compatible = is_gpl ? 1 : 0;
1515 if (bpf_prog_is_dev_bound(prog->aux)) {
1516 err = bpf_prog_offload_init(prog, attr);
1521 /* find program type: socket_filter vs tracing_filter */
1522 err = find_prog_type(type, prog);
1526 prog->aux->load_time = ktime_get_boot_ns();
1527 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1531 /* run eBPF verifier */
1532 err = bpf_check(&prog, attr, uattr);
1534 goto free_used_maps;
1536 prog = bpf_prog_select_runtime(prog, &err);
1538 goto free_used_maps;
1540 err = bpf_prog_alloc_id(prog);
1542 goto free_used_maps;
1544 err = bpf_prog_new_fd(prog);
1546 /* failed to allocate fd.
1547 * bpf_prog_put() is needed because the above
1548 * bpf_prog_alloc_id() has published the prog
1549 * to the userspace and the userspace may
1550 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1556 bpf_prog_kallsyms_add(prog);
1560 bpf_prog_free_linfo(prog);
1561 kvfree(prog->aux->func_info);
1562 btf_put(prog->aux->btf);
1563 bpf_prog_kallsyms_del_subprogs(prog);
1564 free_used_maps(prog->aux);
1566 bpf_prog_uncharge_memlock(prog);
1568 security_bpf_prog_free(prog->aux);
1569 free_prog_nouncharge:
1570 bpf_prog_free(prog);
1574 #define BPF_OBJ_LAST_FIELD file_flags
1576 static int bpf_obj_pin(const union bpf_attr *attr)
1578 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1581 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1584 static int bpf_obj_get(const union bpf_attr *attr)
1586 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1587 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1590 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1594 struct bpf_raw_tracepoint {
1595 struct bpf_raw_event_map *btp;
1596 struct bpf_prog *prog;
1599 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1601 struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1604 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1605 bpf_prog_put(raw_tp->prog);
1607 bpf_put_raw_tracepoint(raw_tp->btp);
1612 static const struct file_operations bpf_raw_tp_fops = {
1613 .release = bpf_raw_tracepoint_release,
1614 .read = bpf_dummy_read,
1615 .write = bpf_dummy_write,
1618 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1620 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1622 struct bpf_raw_tracepoint *raw_tp;
1623 struct bpf_raw_event_map *btp;
1624 struct bpf_prog *prog;
1628 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1629 sizeof(tp_name) - 1) < 0)
1631 tp_name[sizeof(tp_name) - 1] = 0;
1633 btp = bpf_get_raw_tracepoint(tp_name);
1637 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1644 prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1645 BPF_PROG_TYPE_RAW_TRACEPOINT);
1647 err = PTR_ERR(prog);
1651 err = bpf_probe_register(raw_tp->btp, prog);
1655 raw_tp->prog = prog;
1656 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1659 bpf_probe_unregister(raw_tp->btp, prog);
1670 bpf_put_raw_tracepoint(btp);
1674 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1675 enum bpf_attach_type attach_type)
1677 switch (prog->type) {
1678 case BPF_PROG_TYPE_CGROUP_SOCK:
1679 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1680 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1686 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1688 #define BPF_F_ATTACH_MASK \
1689 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1691 static int bpf_prog_attach(const union bpf_attr *attr)
1693 enum bpf_prog_type ptype;
1694 struct bpf_prog *prog;
1697 if (!capable(CAP_NET_ADMIN))
1700 if (CHECK_ATTR(BPF_PROG_ATTACH))
1703 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1706 switch (attr->attach_type) {
1707 case BPF_CGROUP_INET_INGRESS:
1708 case BPF_CGROUP_INET_EGRESS:
1709 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1711 case BPF_CGROUP_INET_SOCK_CREATE:
1712 case BPF_CGROUP_INET4_POST_BIND:
1713 case BPF_CGROUP_INET6_POST_BIND:
1714 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1716 case BPF_CGROUP_INET4_BIND:
1717 case BPF_CGROUP_INET6_BIND:
1718 case BPF_CGROUP_INET4_CONNECT:
1719 case BPF_CGROUP_INET6_CONNECT:
1720 case BPF_CGROUP_UDP4_SENDMSG:
1721 case BPF_CGROUP_UDP6_SENDMSG:
1722 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1724 case BPF_CGROUP_SOCK_OPS:
1725 ptype = BPF_PROG_TYPE_SOCK_OPS;
1727 case BPF_CGROUP_DEVICE:
1728 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1730 case BPF_SK_MSG_VERDICT:
1731 ptype = BPF_PROG_TYPE_SK_MSG;
1733 case BPF_SK_SKB_STREAM_PARSER:
1734 case BPF_SK_SKB_STREAM_VERDICT:
1735 ptype = BPF_PROG_TYPE_SK_SKB;
1737 case BPF_LIRC_MODE2:
1738 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1740 case BPF_FLOW_DISSECTOR:
1741 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1747 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1749 return PTR_ERR(prog);
1751 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1757 case BPF_PROG_TYPE_SK_SKB:
1758 case BPF_PROG_TYPE_SK_MSG:
1759 ret = sock_map_get_from_fd(attr, prog);
1761 case BPF_PROG_TYPE_LIRC_MODE2:
1762 ret = lirc_prog_attach(attr, prog);
1764 case BPF_PROG_TYPE_FLOW_DISSECTOR:
1765 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1768 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1776 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1778 static int bpf_prog_detach(const union bpf_attr *attr)
1780 enum bpf_prog_type ptype;
1782 if (!capable(CAP_NET_ADMIN))
1785 if (CHECK_ATTR(BPF_PROG_DETACH))
1788 switch (attr->attach_type) {
1789 case BPF_CGROUP_INET_INGRESS:
1790 case BPF_CGROUP_INET_EGRESS:
1791 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1793 case BPF_CGROUP_INET_SOCK_CREATE:
1794 case BPF_CGROUP_INET4_POST_BIND:
1795 case BPF_CGROUP_INET6_POST_BIND:
1796 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1798 case BPF_CGROUP_INET4_BIND:
1799 case BPF_CGROUP_INET6_BIND:
1800 case BPF_CGROUP_INET4_CONNECT:
1801 case BPF_CGROUP_INET6_CONNECT:
1802 case BPF_CGROUP_UDP4_SENDMSG:
1803 case BPF_CGROUP_UDP6_SENDMSG:
1804 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1806 case BPF_CGROUP_SOCK_OPS:
1807 ptype = BPF_PROG_TYPE_SOCK_OPS;
1809 case BPF_CGROUP_DEVICE:
1810 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1812 case BPF_SK_MSG_VERDICT:
1813 return sock_map_get_from_fd(attr, NULL);
1814 case BPF_SK_SKB_STREAM_PARSER:
1815 case BPF_SK_SKB_STREAM_VERDICT:
1816 return sock_map_get_from_fd(attr, NULL);
1817 case BPF_LIRC_MODE2:
1818 return lirc_prog_detach(attr);
1819 case BPF_FLOW_DISSECTOR:
1820 return skb_flow_dissector_bpf_prog_detach(attr);
1825 return cgroup_bpf_prog_detach(attr, ptype);
1828 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1830 static int bpf_prog_query(const union bpf_attr *attr,
1831 union bpf_attr __user *uattr)
1833 if (!capable(CAP_NET_ADMIN))
1835 if (CHECK_ATTR(BPF_PROG_QUERY))
1837 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1840 switch (attr->query.attach_type) {
1841 case BPF_CGROUP_INET_INGRESS:
1842 case BPF_CGROUP_INET_EGRESS:
1843 case BPF_CGROUP_INET_SOCK_CREATE:
1844 case BPF_CGROUP_INET4_BIND:
1845 case BPF_CGROUP_INET6_BIND:
1846 case BPF_CGROUP_INET4_POST_BIND:
1847 case BPF_CGROUP_INET6_POST_BIND:
1848 case BPF_CGROUP_INET4_CONNECT:
1849 case BPF_CGROUP_INET6_CONNECT:
1850 case BPF_CGROUP_UDP4_SENDMSG:
1851 case BPF_CGROUP_UDP6_SENDMSG:
1852 case BPF_CGROUP_SOCK_OPS:
1853 case BPF_CGROUP_DEVICE:
1855 case BPF_LIRC_MODE2:
1856 return lirc_prog_query(attr, uattr);
1861 return cgroup_bpf_prog_query(attr, uattr);
1864 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1866 static int bpf_prog_test_run(const union bpf_attr *attr,
1867 union bpf_attr __user *uattr)
1869 struct bpf_prog *prog;
1870 int ret = -ENOTSUPP;
1872 if (!capable(CAP_SYS_ADMIN))
1874 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1877 prog = bpf_prog_get(attr->test.prog_fd);
1879 return PTR_ERR(prog);
1881 if (prog->aux->ops->test_run)
1882 ret = prog->aux->ops->test_run(prog, attr, uattr);
1888 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1890 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1891 union bpf_attr __user *uattr,
1895 u32 next_id = attr->start_id;
1898 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1901 if (!capable(CAP_SYS_ADMIN))
1906 if (!idr_get_next(idr, &next_id))
1908 spin_unlock_bh(lock);
1911 err = put_user(next_id, &uattr->next_id);
1916 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1918 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1920 struct bpf_prog *prog;
1921 u32 id = attr->prog_id;
1924 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1927 if (!capable(CAP_SYS_ADMIN))
1930 spin_lock_bh(&prog_idr_lock);
1931 prog = idr_find(&prog_idr, id);
1933 prog = bpf_prog_inc_not_zero(prog);
1935 prog = ERR_PTR(-ENOENT);
1936 spin_unlock_bh(&prog_idr_lock);
1939 return PTR_ERR(prog);
1941 fd = bpf_prog_new_fd(prog);
1948 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1950 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1952 struct bpf_map *map;
1953 u32 id = attr->map_id;
1957 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1958 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1961 if (!capable(CAP_SYS_ADMIN))
1964 f_flags = bpf_get_file_flag(attr->open_flags);
1968 spin_lock_bh(&map_idr_lock);
1969 map = idr_find(&map_idr, id);
1971 map = bpf_map_inc_not_zero(map, true);
1973 map = ERR_PTR(-ENOENT);
1974 spin_unlock_bh(&map_idr_lock);
1977 return PTR_ERR(map);
1979 fd = bpf_map_new_fd(map, f_flags);
1986 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1991 for (i = 0; i < prog->aux->used_map_cnt; i++)
1992 if (prog->aux->used_maps[i] == (void *)addr)
1993 return prog->aux->used_maps[i];
1997 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1999 const struct bpf_map *map;
2000 struct bpf_insn *insns;
2004 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2009 for (i = 0; i < prog->len; i++) {
2010 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2011 insns[i].code = BPF_JMP | BPF_CALL;
2012 insns[i].imm = BPF_FUNC_tail_call;
2015 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2016 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2017 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2018 insns[i].code = BPF_JMP | BPF_CALL;
2019 if (!bpf_dump_raw_ok())
2024 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2027 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2028 map = bpf_map_from_imm(prog, imm);
2030 insns[i].src_reg = BPF_PSEUDO_MAP_FD;
2031 insns[i].imm = map->id;
2032 insns[i + 1].imm = 0;
2040 static int set_info_rec_size(struct bpf_prog_info *info)
2043 * Ensure info.*_rec_size is the same as kernel expected size
2047 * Only allow zero *_rec_size if both _rec_size and _cnt are
2048 * zero. In this case, the kernel will set the expected
2049 * _rec_size back to the info.
2052 if ((info->nr_func_info || info->func_info_rec_size) &&
2053 info->func_info_rec_size != sizeof(struct bpf_func_info))
2056 if ((info->nr_line_info || info->line_info_rec_size) &&
2057 info->line_info_rec_size != sizeof(struct bpf_line_info))
2060 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2061 info->jited_line_info_rec_size != sizeof(__u64))
2064 info->func_info_rec_size = sizeof(struct bpf_func_info);
2065 info->line_info_rec_size = sizeof(struct bpf_line_info);
2066 info->jited_line_info_rec_size = sizeof(__u64);
2071 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2072 const union bpf_attr *attr,
2073 union bpf_attr __user *uattr)
2075 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2076 struct bpf_prog_info info = {};
2077 u32 info_len = attr->info.info_len;
2078 char __user *uinsns;
2082 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2085 info_len = min_t(u32, sizeof(info), info_len);
2087 if (copy_from_user(&info, uinfo, info_len))
2090 info.type = prog->type;
2091 info.id = prog->aux->id;
2092 info.load_time = prog->aux->load_time;
2093 info.created_by_uid = from_kuid_munged(current_user_ns(),
2094 prog->aux->user->uid);
2095 info.gpl_compatible = prog->gpl_compatible;
2097 memcpy(info.tag, prog->tag, sizeof(prog->tag));
2098 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2100 ulen = info.nr_map_ids;
2101 info.nr_map_ids = prog->aux->used_map_cnt;
2102 ulen = min_t(u32, info.nr_map_ids, ulen);
2104 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2107 for (i = 0; i < ulen; i++)
2108 if (put_user(prog->aux->used_maps[i]->id,
2113 err = set_info_rec_size(&info);
2117 if (!capable(CAP_SYS_ADMIN)) {
2118 info.jited_prog_len = 0;
2119 info.xlated_prog_len = 0;
2120 info.nr_jited_ksyms = 0;
2121 info.nr_jited_func_lens = 0;
2122 info.nr_func_info = 0;
2123 info.nr_line_info = 0;
2124 info.nr_jited_line_info = 0;
2128 ulen = info.xlated_prog_len;
2129 info.xlated_prog_len = bpf_prog_insn_size(prog);
2130 if (info.xlated_prog_len && ulen) {
2131 struct bpf_insn *insns_sanitized;
2134 if (prog->blinded && !bpf_dump_raw_ok()) {
2135 info.xlated_prog_insns = 0;
2138 insns_sanitized = bpf_insn_prepare_dump(prog);
2139 if (!insns_sanitized)
2141 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2142 ulen = min_t(u32, info.xlated_prog_len, ulen);
2143 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2144 kfree(insns_sanitized);
2149 if (bpf_prog_is_dev_bound(prog->aux)) {
2150 err = bpf_prog_offload_info_fill(&info, prog);
2156 /* NOTE: the following code is supposed to be skipped for offload.
2157 * bpf_prog_offload_info_fill() is the place to fill similar fields
2160 ulen = info.jited_prog_len;
2161 if (prog->aux->func_cnt) {
2164 info.jited_prog_len = 0;
2165 for (i = 0; i < prog->aux->func_cnt; i++)
2166 info.jited_prog_len += prog->aux->func[i]->jited_len;
2168 info.jited_prog_len = prog->jited_len;
2171 if (info.jited_prog_len && ulen) {
2172 if (bpf_dump_raw_ok()) {
2173 uinsns = u64_to_user_ptr(info.jited_prog_insns);
2174 ulen = min_t(u32, info.jited_prog_len, ulen);
2176 /* for multi-function programs, copy the JITed
2177 * instructions for all the functions
2179 if (prog->aux->func_cnt) {
2184 for (i = 0; i < prog->aux->func_cnt; i++) {
2185 len = prog->aux->func[i]->jited_len;
2186 len = min_t(u32, len, free);
2187 img = (u8 *) prog->aux->func[i]->bpf_func;
2188 if (copy_to_user(uinsns, img, len))
2196 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2200 info.jited_prog_insns = 0;
2204 ulen = info.nr_jited_ksyms;
2205 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2207 if (bpf_dump_raw_ok()) {
2208 unsigned long ksym_addr;
2209 u64 __user *user_ksyms;
2212 /* copy the address of the kernel symbol
2213 * corresponding to each function
2215 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2216 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2217 if (prog->aux->func_cnt) {
2218 for (i = 0; i < ulen; i++) {
2219 ksym_addr = (unsigned long)
2220 prog->aux->func[i]->bpf_func;
2221 if (put_user((u64) ksym_addr,
2226 ksym_addr = (unsigned long) prog->bpf_func;
2227 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2231 info.jited_ksyms = 0;
2235 ulen = info.nr_jited_func_lens;
2236 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2238 if (bpf_dump_raw_ok()) {
2239 u32 __user *user_lens;
2242 /* copy the JITed image lengths for each function */
2243 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2244 user_lens = u64_to_user_ptr(info.jited_func_lens);
2245 if (prog->aux->func_cnt) {
2246 for (i = 0; i < ulen; i++) {
2248 prog->aux->func[i]->jited_len;
2249 if (put_user(func_len, &user_lens[i]))
2253 func_len = prog->jited_len;
2254 if (put_user(func_len, &user_lens[0]))
2258 info.jited_func_lens = 0;
2263 info.btf_id = btf_id(prog->aux->btf);
2265 ulen = info.nr_func_info;
2266 info.nr_func_info = prog->aux->func_info_cnt;
2267 if (info.nr_func_info && ulen) {
2268 char __user *user_finfo;
2270 user_finfo = u64_to_user_ptr(info.func_info);
2271 ulen = min_t(u32, info.nr_func_info, ulen);
2272 if (copy_to_user(user_finfo, prog->aux->func_info,
2273 info.func_info_rec_size * ulen))
2277 ulen = info.nr_line_info;
2278 info.nr_line_info = prog->aux->nr_linfo;
2279 if (info.nr_line_info && ulen) {
2280 __u8 __user *user_linfo;
2282 user_linfo = u64_to_user_ptr(info.line_info);
2283 ulen = min_t(u32, info.nr_line_info, ulen);
2284 if (copy_to_user(user_linfo, prog->aux->linfo,
2285 info.line_info_rec_size * ulen))
2289 ulen = info.nr_jited_line_info;
2290 if (prog->aux->jited_linfo)
2291 info.nr_jited_line_info = prog->aux->nr_linfo;
2293 info.nr_jited_line_info = 0;
2294 if (info.nr_jited_line_info && ulen) {
2295 if (bpf_dump_raw_ok()) {
2296 __u64 __user *user_linfo;
2299 user_linfo = u64_to_user_ptr(info.jited_line_info);
2300 ulen = min_t(u32, info.nr_jited_line_info, ulen);
2301 for (i = 0; i < ulen; i++) {
2302 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2307 info.jited_line_info = 0;
2311 ulen = info.nr_prog_tags;
2312 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2314 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2317 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2318 ulen = min_t(u32, info.nr_prog_tags, ulen);
2319 if (prog->aux->func_cnt) {
2320 for (i = 0; i < ulen; i++) {
2321 if (copy_to_user(user_prog_tags[i],
2322 prog->aux->func[i]->tag,
2327 if (copy_to_user(user_prog_tags[0],
2328 prog->tag, BPF_TAG_SIZE))
2334 if (copy_to_user(uinfo, &info, info_len) ||
2335 put_user(info_len, &uattr->info.info_len))
2341 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2342 const union bpf_attr *attr,
2343 union bpf_attr __user *uattr)
2345 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2346 struct bpf_map_info info = {};
2347 u32 info_len = attr->info.info_len;
2350 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2353 info_len = min_t(u32, sizeof(info), info_len);
2355 info.type = map->map_type;
2357 info.key_size = map->key_size;
2358 info.value_size = map->value_size;
2359 info.max_entries = map->max_entries;
2360 info.map_flags = map->map_flags;
2361 memcpy(info.name, map->name, sizeof(map->name));
2364 info.btf_id = btf_id(map->btf);
2365 info.btf_key_type_id = map->btf_key_type_id;
2366 info.btf_value_type_id = map->btf_value_type_id;
2369 if (bpf_map_is_dev_bound(map)) {
2370 err = bpf_map_offload_info_fill(&info, map);
2375 if (copy_to_user(uinfo, &info, info_len) ||
2376 put_user(info_len, &uattr->info.info_len))
2382 static int bpf_btf_get_info_by_fd(struct btf *btf,
2383 const union bpf_attr *attr,
2384 union bpf_attr __user *uattr)
2386 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2387 u32 info_len = attr->info.info_len;
2390 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2394 return btf_get_info_by_fd(btf, attr, uattr);
2397 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2399 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2400 union bpf_attr __user *uattr)
2402 int ufd = attr->info.bpf_fd;
2406 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2413 if (f.file->f_op == &bpf_prog_fops)
2414 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2416 else if (f.file->f_op == &bpf_map_fops)
2417 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2419 else if (f.file->f_op == &btf_fops)
2420 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2428 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2430 static int bpf_btf_load(const union bpf_attr *attr)
2432 if (CHECK_ATTR(BPF_BTF_LOAD))
2435 if (!capable(CAP_SYS_ADMIN))
2438 return btf_new_fd(attr);
2441 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2443 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2445 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2448 if (!capable(CAP_SYS_ADMIN))
2451 return btf_get_fd_by_id(attr->btf_id);
2454 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2455 union bpf_attr __user *uattr,
2456 u32 prog_id, u32 fd_type,
2457 const char *buf, u64 probe_offset,
2460 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2461 u32 len = buf ? strlen(buf) : 0, input_len;
2464 if (put_user(len, &uattr->task_fd_query.buf_len))
2466 input_len = attr->task_fd_query.buf_len;
2467 if (input_len && ubuf) {
2469 /* nothing to copy, just make ubuf NULL terminated */
2472 if (put_user(zero, ubuf))
2474 } else if (input_len >= len + 1) {
2475 /* ubuf can hold the string with NULL terminator */
2476 if (copy_to_user(ubuf, buf, len + 1))
2479 /* ubuf cannot hold the string with NULL terminator,
2480 * do a partial copy with NULL terminator.
2485 if (copy_to_user(ubuf, buf, input_len - 1))
2487 if (put_user(zero, ubuf + input_len - 1))
2492 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2493 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2494 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2495 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2501 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2503 static int bpf_task_fd_query(const union bpf_attr *attr,
2504 union bpf_attr __user *uattr)
2506 pid_t pid = attr->task_fd_query.pid;
2507 u32 fd = attr->task_fd_query.fd;
2508 const struct perf_event *event;
2509 struct files_struct *files;
2510 struct task_struct *task;
2514 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2517 if (!capable(CAP_SYS_ADMIN))
2520 if (attr->task_fd_query.flags != 0)
2523 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2527 files = get_files_struct(task);
2528 put_task_struct(task);
2533 spin_lock(&files->file_lock);
2534 file = fcheck_files(files, fd);
2539 spin_unlock(&files->file_lock);
2540 put_files_struct(files);
2545 if (file->f_op == &bpf_raw_tp_fops) {
2546 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2547 struct bpf_raw_event_map *btp = raw_tp->btp;
2549 err = bpf_task_fd_query_copy(attr, uattr,
2550 raw_tp->prog->aux->id,
2551 BPF_FD_TYPE_RAW_TRACEPOINT,
2552 btp->tp->name, 0, 0);
2556 event = perf_get_event(file);
2557 if (!IS_ERR(event)) {
2558 u64 probe_offset, probe_addr;
2559 u32 prog_id, fd_type;
2562 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2563 &buf, &probe_offset,
2566 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2580 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2582 union bpf_attr attr = {};
2585 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2588 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2591 size = min_t(u32, size, sizeof(attr));
2593 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2594 if (copy_from_user(&attr, uattr, size) != 0)
2597 err = security_bpf(cmd, &attr, size);
2602 case BPF_MAP_CREATE:
2603 err = map_create(&attr);
2605 case BPF_MAP_LOOKUP_ELEM:
2606 err = map_lookup_elem(&attr);
2608 case BPF_MAP_UPDATE_ELEM:
2609 err = map_update_elem(&attr);
2611 case BPF_MAP_DELETE_ELEM:
2612 err = map_delete_elem(&attr);
2614 case BPF_MAP_GET_NEXT_KEY:
2615 err = map_get_next_key(&attr);
2618 err = bpf_prog_load(&attr, uattr);
2621 err = bpf_obj_pin(&attr);
2624 err = bpf_obj_get(&attr);
2626 case BPF_PROG_ATTACH:
2627 err = bpf_prog_attach(&attr);
2629 case BPF_PROG_DETACH:
2630 err = bpf_prog_detach(&attr);
2632 case BPF_PROG_QUERY:
2633 err = bpf_prog_query(&attr, uattr);
2635 case BPF_PROG_TEST_RUN:
2636 err = bpf_prog_test_run(&attr, uattr);
2638 case BPF_PROG_GET_NEXT_ID:
2639 err = bpf_obj_get_next_id(&attr, uattr,
2640 &prog_idr, &prog_idr_lock);
2642 case BPF_MAP_GET_NEXT_ID:
2643 err = bpf_obj_get_next_id(&attr, uattr,
2644 &map_idr, &map_idr_lock);
2646 case BPF_PROG_GET_FD_BY_ID:
2647 err = bpf_prog_get_fd_by_id(&attr);
2649 case BPF_MAP_GET_FD_BY_ID:
2650 err = bpf_map_get_fd_by_id(&attr);
2652 case BPF_OBJ_GET_INFO_BY_FD:
2653 err = bpf_obj_get_info_by_fd(&attr, uattr);
2655 case BPF_RAW_TRACEPOINT_OPEN:
2656 err = bpf_raw_tracepoint_open(&attr);
2659 err = bpf_btf_load(&attr);
2661 case BPF_BTF_GET_FD_BY_ID:
2662 err = bpf_btf_get_fd_by_id(&attr);
2664 case BPF_TASK_FD_QUERY:
2665 err = bpf_task_fd_query(&attr, uattr);
2667 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2668 err = map_lookup_and_delete_elem(&attr);