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(VERIFY_READ, 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_type *key_type,
460 const struct btf_type *value_type)
465 static int map_check_btf(const struct bpf_map *map, const struct btf *btf,
466 u32 btf_key_id, u32 btf_value_id)
468 const struct btf_type *key_type, *value_type;
469 u32 key_size, value_size;
472 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
473 if (!key_type || key_size != map->key_size)
476 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
477 if (!value_type || value_size != map->value_size)
480 if (map->ops->map_check_btf)
481 ret = map->ops->map_check_btf(map, key_type, value_type);
486 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
487 /* called via syscall */
488 static int map_create(union bpf_attr *attr)
490 int numa_node = bpf_map_attr_numa_node(attr);
495 err = CHECK_ATTR(BPF_MAP_CREATE);
499 f_flags = bpf_get_file_flag(attr->map_flags);
503 if (numa_node != NUMA_NO_NODE &&
504 ((unsigned int)numa_node >= nr_node_ids ||
505 !node_online(numa_node)))
508 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
509 map = find_and_alloc_map(attr);
513 err = bpf_obj_name_cpy(map->name, attr->map_name);
515 goto free_map_nouncharge;
517 atomic_set(&map->refcnt, 1);
518 atomic_set(&map->usercnt, 1);
520 if (attr->btf_key_type_id || attr->btf_value_type_id) {
523 if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
525 goto free_map_nouncharge;
528 btf = btf_get_by_fd(attr->btf_fd);
531 goto free_map_nouncharge;
534 err = map_check_btf(map, btf, attr->btf_key_type_id,
535 attr->btf_value_type_id);
538 goto free_map_nouncharge;
542 map->btf_key_type_id = attr->btf_key_type_id;
543 map->btf_value_type_id = attr->btf_value_type_id;
546 err = security_bpf_map_alloc(map);
548 goto free_map_nouncharge;
550 err = bpf_map_init_memlock(map);
554 err = bpf_map_alloc_id(map);
558 err = bpf_map_new_fd(map, f_flags);
560 /* failed to allocate fd.
561 * bpf_map_put() is needed because the above
562 * bpf_map_alloc_id() has published the map
563 * to the userspace and the userspace may
564 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
573 bpf_map_release_memlock(map);
575 security_bpf_map_free(map);
578 map->ops->map_free(map);
582 /* if error is returned, fd is released.
583 * On success caller should complete fd access with matching fdput()
585 struct bpf_map *__bpf_map_get(struct fd f)
588 return ERR_PTR(-EBADF);
589 if (f.file->f_op != &bpf_map_fops) {
591 return ERR_PTR(-EINVAL);
594 return f.file->private_data;
597 /* prog's and map's refcnt limit */
598 #define BPF_MAX_REFCNT 32768
600 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
602 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
603 atomic_dec(&map->refcnt);
604 return ERR_PTR(-EBUSY);
607 atomic_inc(&map->usercnt);
610 EXPORT_SYMBOL_GPL(bpf_map_inc);
612 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
614 struct fd f = fdget(ufd);
617 map = __bpf_map_get(f);
621 map = bpf_map_inc(map, true);
627 /* map_idr_lock should have been held */
628 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
633 refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
635 if (refold >= BPF_MAX_REFCNT) {
636 __bpf_map_put(map, false);
637 return ERR_PTR(-EBUSY);
641 return ERR_PTR(-ENOENT);
644 atomic_inc(&map->usercnt);
649 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
654 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
657 return memdup_user(ukey, key_size);
660 return ERR_PTR(-EINVAL);
665 /* last field in 'union bpf_attr' used by this command */
666 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
668 static int map_lookup_elem(union bpf_attr *attr)
670 void __user *ukey = u64_to_user_ptr(attr->key);
671 void __user *uvalue = u64_to_user_ptr(attr->value);
672 int ufd = attr->map_fd;
674 void *key, *value, *ptr;
679 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
683 map = __bpf_map_get(f);
687 if (!(f.file->f_mode & FMODE_CAN_READ)) {
692 key = __bpf_copy_key(ukey, map->key_size);
698 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
699 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
700 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
701 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
702 value_size = round_up(map->value_size, 8) * num_possible_cpus();
703 else if (IS_FD_MAP(map))
704 value_size = sizeof(u32);
706 value_size = map->value_size;
709 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
713 if (bpf_map_is_dev_bound(map)) {
714 err = bpf_map_offload_lookup_elem(map, key, value);
715 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
716 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
717 err = bpf_percpu_hash_copy(map, key, value);
718 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
719 err = bpf_percpu_array_copy(map, key, value);
720 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
721 err = bpf_percpu_cgroup_storage_copy(map, key, value);
722 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
723 err = bpf_stackmap_copy(map, key, value);
724 } else if (IS_FD_ARRAY(map)) {
725 err = bpf_fd_array_map_lookup_elem(map, key, value);
726 } else if (IS_FD_HASH(map)) {
727 err = bpf_fd_htab_map_lookup_elem(map, key, value);
728 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
729 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
730 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
731 map->map_type == BPF_MAP_TYPE_STACK) {
732 err = map->ops->map_peek_elem(map, value);
735 ptr = map->ops->map_lookup_elem(map, key);
742 memcpy(value, ptr, value_size);
751 if (copy_to_user(uvalue, value, value_size) != 0)
765 static void maybe_wait_bpf_programs(struct bpf_map *map)
767 /* Wait for any running BPF programs to complete so that
768 * userspace, when we return to it, knows that all programs
769 * that could be running use the new map value.
771 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
772 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
776 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
778 static int map_update_elem(union bpf_attr *attr)
780 void __user *ukey = u64_to_user_ptr(attr->key);
781 void __user *uvalue = u64_to_user_ptr(attr->value);
782 int ufd = attr->map_fd;
789 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
793 map = __bpf_map_get(f);
797 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
802 key = __bpf_copy_key(ukey, map->key_size);
808 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
809 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
810 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
811 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
812 value_size = round_up(map->value_size, 8) * num_possible_cpus();
814 value_size = map->value_size;
817 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
822 if (copy_from_user(value, uvalue, value_size) != 0)
825 /* Need to create a kthread, thus must support schedule */
826 if (bpf_map_is_dev_bound(map)) {
827 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
829 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
830 map->map_type == BPF_MAP_TYPE_SOCKHASH ||
831 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
832 err = map->ops->map_update_elem(map, key, value, attr->flags);
836 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
837 * inside bpf map update or delete otherwise deadlocks are possible
840 __this_cpu_inc(bpf_prog_active);
841 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
842 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
843 err = bpf_percpu_hash_update(map, key, value, attr->flags);
844 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
845 err = bpf_percpu_array_update(map, key, value, attr->flags);
846 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
847 err = bpf_percpu_cgroup_storage_update(map, key, value,
849 } else if (IS_FD_ARRAY(map)) {
851 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
854 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
856 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
859 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
860 /* rcu_read_lock() is not needed */
861 err = bpf_fd_reuseport_array_update_elem(map, key, value,
863 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
864 map->map_type == BPF_MAP_TYPE_STACK) {
865 err = map->ops->map_push_elem(map, value, attr->flags);
868 err = map->ops->map_update_elem(map, key, value, attr->flags);
871 __this_cpu_dec(bpf_prog_active);
873 maybe_wait_bpf_programs(map);
884 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
886 static int map_delete_elem(union bpf_attr *attr)
888 void __user *ukey = u64_to_user_ptr(attr->key);
889 int ufd = attr->map_fd;
895 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
899 map = __bpf_map_get(f);
903 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
908 key = __bpf_copy_key(ukey, map->key_size);
914 if (bpf_map_is_dev_bound(map)) {
915 err = bpf_map_offload_delete_elem(map, key);
920 __this_cpu_inc(bpf_prog_active);
922 err = map->ops->map_delete_elem(map, key);
924 __this_cpu_dec(bpf_prog_active);
926 maybe_wait_bpf_programs(map);
934 /* last field in 'union bpf_attr' used by this command */
935 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
937 static int map_get_next_key(union bpf_attr *attr)
939 void __user *ukey = u64_to_user_ptr(attr->key);
940 void __user *unext_key = u64_to_user_ptr(attr->next_key);
941 int ufd = attr->map_fd;
943 void *key, *next_key;
947 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
951 map = __bpf_map_get(f);
955 if (!(f.file->f_mode & FMODE_CAN_READ)) {
961 key = __bpf_copy_key(ukey, map->key_size);
971 next_key = kmalloc(map->key_size, GFP_USER);
975 if (bpf_map_is_dev_bound(map)) {
976 err = bpf_map_offload_get_next_key(map, key, next_key);
981 err = map->ops->map_get_next_key(map, key, next_key);
988 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1002 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1004 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1006 void __user *ukey = u64_to_user_ptr(attr->key);
1007 void __user *uvalue = u64_to_user_ptr(attr->value);
1008 int ufd = attr->map_fd;
1009 struct bpf_map *map;
1015 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1019 map = __bpf_map_get(f);
1021 return PTR_ERR(map);
1023 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
1028 key = __bpf_copy_key(ukey, map->key_size);
1034 value_size = map->value_size;
1037 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1041 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1042 map->map_type == BPF_MAP_TYPE_STACK) {
1043 err = map->ops->map_pop_elem(map, value);
1051 if (copy_to_user(uvalue, value, value_size) != 0)
1065 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1066 #define BPF_PROG_TYPE(_id, _name) \
1067 [_id] = & _name ## _prog_ops,
1068 #define BPF_MAP_TYPE(_id, _ops)
1069 #include <linux/bpf_types.h>
1070 #undef BPF_PROG_TYPE
1074 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1076 const struct bpf_prog_ops *ops;
1078 if (type >= ARRAY_SIZE(bpf_prog_types))
1080 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1081 ops = bpf_prog_types[type];
1085 if (!bpf_prog_is_dev_bound(prog->aux))
1086 prog->aux->ops = ops;
1088 prog->aux->ops = &bpf_offload_prog_ops;
1093 /* drop refcnt on maps used by eBPF program and free auxilary data */
1094 static void free_used_maps(struct bpf_prog_aux *aux)
1096 enum bpf_cgroup_storage_type stype;
1099 for_each_cgroup_storage_type(stype) {
1100 if (!aux->cgroup_storage[stype])
1102 bpf_cgroup_storage_release(aux->prog,
1103 aux->cgroup_storage[stype]);
1106 for (i = 0; i < aux->used_map_cnt; i++)
1107 bpf_map_put(aux->used_maps[i]);
1109 kfree(aux->used_maps);
1112 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1114 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1115 unsigned long user_bufs;
1118 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1119 if (user_bufs > memlock_limit) {
1120 atomic_long_sub(pages, &user->locked_vm);
1128 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1131 atomic_long_sub(pages, &user->locked_vm);
1134 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1136 struct user_struct *user = get_current_user();
1139 ret = __bpf_prog_charge(user, prog->pages);
1145 prog->aux->user = user;
1149 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1151 struct user_struct *user = prog->aux->user;
1153 __bpf_prog_uncharge(user, prog->pages);
1157 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1161 idr_preload(GFP_KERNEL);
1162 spin_lock_bh(&prog_idr_lock);
1163 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1166 spin_unlock_bh(&prog_idr_lock);
1169 /* id is in [1, INT_MAX) */
1170 if (WARN_ON_ONCE(!id))
1173 return id > 0 ? 0 : id;
1176 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1178 /* cBPF to eBPF migrations are currently not in the idr store.
1179 * Offloaded programs are removed from the store when their device
1180 * disappears - even if someone grabs an fd to them they are unusable,
1181 * simply waiting for refcnt to drop to be freed.
1187 spin_lock_bh(&prog_idr_lock);
1189 __acquire(&prog_idr_lock);
1191 idr_remove(&prog_idr, prog->aux->id);
1195 spin_unlock_bh(&prog_idr_lock);
1197 __release(&prog_idr_lock);
1200 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1202 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1204 free_used_maps(aux);
1205 bpf_prog_uncharge_memlock(aux->prog);
1206 security_bpf_prog_free(aux);
1207 bpf_prog_free(aux->prog);
1210 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1212 if (atomic_dec_and_test(&prog->aux->refcnt)) {
1213 /* bpf_prog_free_id() must be called first */
1214 bpf_prog_free_id(prog, do_idr_lock);
1215 bpf_prog_kallsyms_del_all(prog);
1216 btf_put(prog->aux->btf);
1217 kvfree(prog->aux->func_info);
1218 bpf_prog_free_linfo(prog);
1220 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1224 void bpf_prog_put(struct bpf_prog *prog)
1226 __bpf_prog_put(prog, true);
1228 EXPORT_SYMBOL_GPL(bpf_prog_put);
1230 static int bpf_prog_release(struct inode *inode, struct file *filp)
1232 struct bpf_prog *prog = filp->private_data;
1238 #ifdef CONFIG_PROC_FS
1239 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1241 const struct bpf_prog *prog = filp->private_data;
1242 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1244 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1254 prog->pages * 1ULL << PAGE_SHIFT,
1259 const struct file_operations bpf_prog_fops = {
1260 #ifdef CONFIG_PROC_FS
1261 .show_fdinfo = bpf_prog_show_fdinfo,
1263 .release = bpf_prog_release,
1264 .read = bpf_dummy_read,
1265 .write = bpf_dummy_write,
1268 int bpf_prog_new_fd(struct bpf_prog *prog)
1272 ret = security_bpf_prog(prog);
1276 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1277 O_RDWR | O_CLOEXEC);
1280 static struct bpf_prog *____bpf_prog_get(struct fd f)
1283 return ERR_PTR(-EBADF);
1284 if (f.file->f_op != &bpf_prog_fops) {
1286 return ERR_PTR(-EINVAL);
1289 return f.file->private_data;
1292 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1294 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1295 atomic_sub(i, &prog->aux->refcnt);
1296 return ERR_PTR(-EBUSY);
1300 EXPORT_SYMBOL_GPL(bpf_prog_add);
1302 void bpf_prog_sub(struct bpf_prog *prog, int i)
1304 /* Only to be used for undoing previous bpf_prog_add() in some
1305 * error path. We still know that another entity in our call
1306 * path holds a reference to the program, thus atomic_sub() can
1307 * be safely used in such cases!
1309 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1311 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1313 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1315 return bpf_prog_add(prog, 1);
1317 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1319 /* prog_idr_lock should have been held */
1320 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1324 refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1326 if (refold >= BPF_MAX_REFCNT) {
1327 __bpf_prog_put(prog, false);
1328 return ERR_PTR(-EBUSY);
1332 return ERR_PTR(-ENOENT);
1336 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1338 bool bpf_prog_get_ok(struct bpf_prog *prog,
1339 enum bpf_prog_type *attach_type, bool attach_drv)
1341 /* not an attachment, just a refcount inc, always allow */
1345 if (prog->type != *attach_type)
1347 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1353 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1356 struct fd f = fdget(ufd);
1357 struct bpf_prog *prog;
1359 prog = ____bpf_prog_get(f);
1362 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1363 prog = ERR_PTR(-EINVAL);
1367 prog = bpf_prog_inc(prog);
1373 struct bpf_prog *bpf_prog_get(u32 ufd)
1375 return __bpf_prog_get(ufd, NULL, false);
1378 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1381 return __bpf_prog_get(ufd, &type, attach_drv);
1383 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1385 /* Initially all BPF programs could be loaded w/o specifying
1386 * expected_attach_type. Later for some of them specifying expected_attach_type
1387 * at load time became required so that program could be validated properly.
1388 * Programs of types that are allowed to be loaded both w/ and w/o (for
1389 * backward compatibility) expected_attach_type, should have the default attach
1390 * type assigned to expected_attach_type for the latter case, so that it can be
1391 * validated later at attach time.
1393 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1394 * prog type requires it but has some attach types that have to be backward
1397 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1399 switch (attr->prog_type) {
1400 case BPF_PROG_TYPE_CGROUP_SOCK:
1401 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1402 * exist so checking for non-zero is the way to go here.
1404 if (!attr->expected_attach_type)
1405 attr->expected_attach_type =
1406 BPF_CGROUP_INET_SOCK_CREATE;
1412 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1413 enum bpf_attach_type expected_attach_type)
1415 switch (prog_type) {
1416 case BPF_PROG_TYPE_CGROUP_SOCK:
1417 switch (expected_attach_type) {
1418 case BPF_CGROUP_INET_SOCK_CREATE:
1419 case BPF_CGROUP_INET4_POST_BIND:
1420 case BPF_CGROUP_INET6_POST_BIND:
1425 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1426 switch (expected_attach_type) {
1427 case BPF_CGROUP_INET4_BIND:
1428 case BPF_CGROUP_INET6_BIND:
1429 case BPF_CGROUP_INET4_CONNECT:
1430 case BPF_CGROUP_INET6_CONNECT:
1431 case BPF_CGROUP_UDP4_SENDMSG:
1432 case BPF_CGROUP_UDP6_SENDMSG:
1442 /* last field in 'union bpf_attr' used by this command */
1443 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1445 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1447 enum bpf_prog_type type = attr->prog_type;
1448 struct bpf_prog *prog;
1453 if (CHECK_ATTR(BPF_PROG_LOAD))
1456 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT))
1459 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1460 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1461 !capable(CAP_SYS_ADMIN))
1464 /* copy eBPF program license from user space */
1465 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1466 sizeof(license) - 1) < 0)
1468 license[sizeof(license) - 1] = 0;
1470 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1471 is_gpl = license_is_gpl_compatible(license);
1473 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1476 if (type == BPF_PROG_TYPE_KPROBE &&
1477 attr->kern_version != LINUX_VERSION_CODE)
1480 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1481 type != BPF_PROG_TYPE_CGROUP_SKB &&
1482 !capable(CAP_SYS_ADMIN))
1485 bpf_prog_load_fixup_attach_type(attr);
1486 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1489 /* plain bpf_prog allocation */
1490 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1494 prog->expected_attach_type = attr->expected_attach_type;
1496 prog->aux->offload_requested = !!attr->prog_ifindex;
1498 err = security_bpf_prog_alloc(prog->aux);
1500 goto free_prog_nouncharge;
1502 err = bpf_prog_charge_memlock(prog);
1506 prog->len = attr->insn_cnt;
1509 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1510 bpf_prog_insn_size(prog)) != 0)
1513 prog->orig_prog = NULL;
1516 atomic_set(&prog->aux->refcnt, 1);
1517 prog->gpl_compatible = is_gpl ? 1 : 0;
1519 if (bpf_prog_is_dev_bound(prog->aux)) {
1520 err = bpf_prog_offload_init(prog, attr);
1525 /* find program type: socket_filter vs tracing_filter */
1526 err = find_prog_type(type, prog);
1530 prog->aux->load_time = ktime_get_boot_ns();
1531 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1535 /* run eBPF verifier */
1536 err = bpf_check(&prog, attr, uattr);
1538 goto free_used_maps;
1540 prog = bpf_prog_select_runtime(prog, &err);
1542 goto free_used_maps;
1544 err = bpf_prog_alloc_id(prog);
1546 goto free_used_maps;
1548 err = bpf_prog_new_fd(prog);
1550 /* failed to allocate fd.
1551 * bpf_prog_put() is needed because the above
1552 * bpf_prog_alloc_id() has published the prog
1553 * to the userspace and the userspace may
1554 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1560 bpf_prog_kallsyms_add(prog);
1564 bpf_prog_free_linfo(prog);
1565 kvfree(prog->aux->func_info);
1566 btf_put(prog->aux->btf);
1567 bpf_prog_kallsyms_del_subprogs(prog);
1568 free_used_maps(prog->aux);
1570 bpf_prog_uncharge_memlock(prog);
1572 security_bpf_prog_free(prog->aux);
1573 free_prog_nouncharge:
1574 bpf_prog_free(prog);
1578 #define BPF_OBJ_LAST_FIELD file_flags
1580 static int bpf_obj_pin(const union bpf_attr *attr)
1582 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1585 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1588 static int bpf_obj_get(const union bpf_attr *attr)
1590 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1591 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1594 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1598 struct bpf_raw_tracepoint {
1599 struct bpf_raw_event_map *btp;
1600 struct bpf_prog *prog;
1603 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1605 struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1608 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1609 bpf_prog_put(raw_tp->prog);
1615 static const struct file_operations bpf_raw_tp_fops = {
1616 .release = bpf_raw_tracepoint_release,
1617 .read = bpf_dummy_read,
1618 .write = bpf_dummy_write,
1621 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1623 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1625 struct bpf_raw_tracepoint *raw_tp;
1626 struct bpf_raw_event_map *btp;
1627 struct bpf_prog *prog;
1631 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1632 sizeof(tp_name) - 1) < 0)
1634 tp_name[sizeof(tp_name) - 1] = 0;
1636 btp = bpf_find_raw_tracepoint(tp_name);
1640 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1645 prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1646 BPF_PROG_TYPE_RAW_TRACEPOINT);
1648 err = PTR_ERR(prog);
1652 err = bpf_probe_register(raw_tp->btp, prog);
1656 raw_tp->prog = prog;
1657 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1660 bpf_probe_unregister(raw_tp->btp, prog);
1673 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1674 enum bpf_attach_type attach_type)
1676 switch (prog->type) {
1677 case BPF_PROG_TYPE_CGROUP_SOCK:
1678 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1679 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1685 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1687 #define BPF_F_ATTACH_MASK \
1688 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1690 static int bpf_prog_attach(const union bpf_attr *attr)
1692 enum bpf_prog_type ptype;
1693 struct bpf_prog *prog;
1696 if (!capable(CAP_NET_ADMIN))
1699 if (CHECK_ATTR(BPF_PROG_ATTACH))
1702 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1705 switch (attr->attach_type) {
1706 case BPF_CGROUP_INET_INGRESS:
1707 case BPF_CGROUP_INET_EGRESS:
1708 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1710 case BPF_CGROUP_INET_SOCK_CREATE:
1711 case BPF_CGROUP_INET4_POST_BIND:
1712 case BPF_CGROUP_INET6_POST_BIND:
1713 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1715 case BPF_CGROUP_INET4_BIND:
1716 case BPF_CGROUP_INET6_BIND:
1717 case BPF_CGROUP_INET4_CONNECT:
1718 case BPF_CGROUP_INET6_CONNECT:
1719 case BPF_CGROUP_UDP4_SENDMSG:
1720 case BPF_CGROUP_UDP6_SENDMSG:
1721 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1723 case BPF_CGROUP_SOCK_OPS:
1724 ptype = BPF_PROG_TYPE_SOCK_OPS;
1726 case BPF_CGROUP_DEVICE:
1727 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1729 case BPF_SK_MSG_VERDICT:
1730 ptype = BPF_PROG_TYPE_SK_MSG;
1732 case BPF_SK_SKB_STREAM_PARSER:
1733 case BPF_SK_SKB_STREAM_VERDICT:
1734 ptype = BPF_PROG_TYPE_SK_SKB;
1736 case BPF_LIRC_MODE2:
1737 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1739 case BPF_FLOW_DISSECTOR:
1740 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1746 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1748 return PTR_ERR(prog);
1750 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1756 case BPF_PROG_TYPE_SK_SKB:
1757 case BPF_PROG_TYPE_SK_MSG:
1758 ret = sock_map_get_from_fd(attr, prog);
1760 case BPF_PROG_TYPE_LIRC_MODE2:
1761 ret = lirc_prog_attach(attr, prog);
1763 case BPF_PROG_TYPE_FLOW_DISSECTOR:
1764 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1767 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1775 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1777 static int bpf_prog_detach(const union bpf_attr *attr)
1779 enum bpf_prog_type ptype;
1781 if (!capable(CAP_NET_ADMIN))
1784 if (CHECK_ATTR(BPF_PROG_DETACH))
1787 switch (attr->attach_type) {
1788 case BPF_CGROUP_INET_INGRESS:
1789 case BPF_CGROUP_INET_EGRESS:
1790 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1792 case BPF_CGROUP_INET_SOCK_CREATE:
1793 case BPF_CGROUP_INET4_POST_BIND:
1794 case BPF_CGROUP_INET6_POST_BIND:
1795 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1797 case BPF_CGROUP_INET4_BIND:
1798 case BPF_CGROUP_INET6_BIND:
1799 case BPF_CGROUP_INET4_CONNECT:
1800 case BPF_CGROUP_INET6_CONNECT:
1801 case BPF_CGROUP_UDP4_SENDMSG:
1802 case BPF_CGROUP_UDP6_SENDMSG:
1803 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1805 case BPF_CGROUP_SOCK_OPS:
1806 ptype = BPF_PROG_TYPE_SOCK_OPS;
1808 case BPF_CGROUP_DEVICE:
1809 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1811 case BPF_SK_MSG_VERDICT:
1812 return sock_map_get_from_fd(attr, NULL);
1813 case BPF_SK_SKB_STREAM_PARSER:
1814 case BPF_SK_SKB_STREAM_VERDICT:
1815 return sock_map_get_from_fd(attr, NULL);
1816 case BPF_LIRC_MODE2:
1817 return lirc_prog_detach(attr);
1818 case BPF_FLOW_DISSECTOR:
1819 return skb_flow_dissector_bpf_prog_detach(attr);
1824 return cgroup_bpf_prog_detach(attr, ptype);
1827 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1829 static int bpf_prog_query(const union bpf_attr *attr,
1830 union bpf_attr __user *uattr)
1832 if (!capable(CAP_NET_ADMIN))
1834 if (CHECK_ATTR(BPF_PROG_QUERY))
1836 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1839 switch (attr->query.attach_type) {
1840 case BPF_CGROUP_INET_INGRESS:
1841 case BPF_CGROUP_INET_EGRESS:
1842 case BPF_CGROUP_INET_SOCK_CREATE:
1843 case BPF_CGROUP_INET4_BIND:
1844 case BPF_CGROUP_INET6_BIND:
1845 case BPF_CGROUP_INET4_POST_BIND:
1846 case BPF_CGROUP_INET6_POST_BIND:
1847 case BPF_CGROUP_INET4_CONNECT:
1848 case BPF_CGROUP_INET6_CONNECT:
1849 case BPF_CGROUP_UDP4_SENDMSG:
1850 case BPF_CGROUP_UDP6_SENDMSG:
1851 case BPF_CGROUP_SOCK_OPS:
1852 case BPF_CGROUP_DEVICE:
1854 case BPF_LIRC_MODE2:
1855 return lirc_prog_query(attr, uattr);
1860 return cgroup_bpf_prog_query(attr, uattr);
1863 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1865 static int bpf_prog_test_run(const union bpf_attr *attr,
1866 union bpf_attr __user *uattr)
1868 struct bpf_prog *prog;
1869 int ret = -ENOTSUPP;
1871 if (!capable(CAP_SYS_ADMIN))
1873 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1876 prog = bpf_prog_get(attr->test.prog_fd);
1878 return PTR_ERR(prog);
1880 if (prog->aux->ops->test_run)
1881 ret = prog->aux->ops->test_run(prog, attr, uattr);
1887 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1889 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1890 union bpf_attr __user *uattr,
1894 u32 next_id = attr->start_id;
1897 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1900 if (!capable(CAP_SYS_ADMIN))
1905 if (!idr_get_next(idr, &next_id))
1907 spin_unlock_bh(lock);
1910 err = put_user(next_id, &uattr->next_id);
1915 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1917 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1919 struct bpf_prog *prog;
1920 u32 id = attr->prog_id;
1923 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1926 if (!capable(CAP_SYS_ADMIN))
1929 spin_lock_bh(&prog_idr_lock);
1930 prog = idr_find(&prog_idr, id);
1932 prog = bpf_prog_inc_not_zero(prog);
1934 prog = ERR_PTR(-ENOENT);
1935 spin_unlock_bh(&prog_idr_lock);
1938 return PTR_ERR(prog);
1940 fd = bpf_prog_new_fd(prog);
1947 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1949 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1951 struct bpf_map *map;
1952 u32 id = attr->map_id;
1956 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1957 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1960 if (!capable(CAP_SYS_ADMIN))
1963 f_flags = bpf_get_file_flag(attr->open_flags);
1967 spin_lock_bh(&map_idr_lock);
1968 map = idr_find(&map_idr, id);
1970 map = bpf_map_inc_not_zero(map, true);
1972 map = ERR_PTR(-ENOENT);
1973 spin_unlock_bh(&map_idr_lock);
1976 return PTR_ERR(map);
1978 fd = bpf_map_new_fd(map, f_flags);
1985 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1990 for (i = 0; i < prog->aux->used_map_cnt; i++)
1991 if (prog->aux->used_maps[i] == (void *)addr)
1992 return prog->aux->used_maps[i];
1996 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1998 const struct bpf_map *map;
1999 struct bpf_insn *insns;
2003 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2008 for (i = 0; i < prog->len; i++) {
2009 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2010 insns[i].code = BPF_JMP | BPF_CALL;
2011 insns[i].imm = BPF_FUNC_tail_call;
2014 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2015 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2016 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2017 insns[i].code = BPF_JMP | BPF_CALL;
2018 if (!bpf_dump_raw_ok())
2023 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2026 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2027 map = bpf_map_from_imm(prog, imm);
2029 insns[i].src_reg = BPF_PSEUDO_MAP_FD;
2030 insns[i].imm = map->id;
2031 insns[i + 1].imm = 0;
2035 if (!bpf_dump_raw_ok() &&
2036 imm == (unsigned long)prog->aux) {
2038 insns[i + 1].imm = 0;
2046 static int set_info_rec_size(struct bpf_prog_info *info)
2049 * Ensure info.*_rec_size is the same as kernel expected size
2053 * Only allow zero *_rec_size if both _rec_size and _cnt are
2054 * zero. In this case, the kernel will set the expected
2055 * _rec_size back to the info.
2058 if ((info->func_info_cnt || info->func_info_rec_size) &&
2059 info->func_info_rec_size != sizeof(struct bpf_func_info))
2062 if ((info->line_info_cnt || info->line_info_rec_size) &&
2063 info->line_info_rec_size != sizeof(struct bpf_line_info))
2066 if ((info->jited_line_info_cnt || info->jited_line_info_rec_size) &&
2067 info->jited_line_info_rec_size != sizeof(__u64))
2070 info->func_info_rec_size = sizeof(struct bpf_func_info);
2071 info->line_info_rec_size = sizeof(struct bpf_line_info);
2072 info->jited_line_info_rec_size = sizeof(__u64);
2077 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2078 const union bpf_attr *attr,
2079 union bpf_attr __user *uattr)
2081 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2082 struct bpf_prog_info info = {};
2083 u32 info_len = attr->info.info_len;
2084 char __user *uinsns;
2088 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2091 info_len = min_t(u32, sizeof(info), info_len);
2093 if (copy_from_user(&info, uinfo, info_len))
2096 info.type = prog->type;
2097 info.id = prog->aux->id;
2098 info.load_time = prog->aux->load_time;
2099 info.created_by_uid = from_kuid_munged(current_user_ns(),
2100 prog->aux->user->uid);
2101 info.gpl_compatible = prog->gpl_compatible;
2103 memcpy(info.tag, prog->tag, sizeof(prog->tag));
2104 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2106 ulen = info.nr_map_ids;
2107 info.nr_map_ids = prog->aux->used_map_cnt;
2108 ulen = min_t(u32, info.nr_map_ids, ulen);
2110 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2113 for (i = 0; i < ulen; i++)
2114 if (put_user(prog->aux->used_maps[i]->id,
2119 err = set_info_rec_size(&info);
2123 if (!capable(CAP_SYS_ADMIN)) {
2124 info.jited_prog_len = 0;
2125 info.xlated_prog_len = 0;
2126 info.nr_jited_ksyms = 0;
2127 info.nr_jited_func_lens = 0;
2128 info.func_info_cnt = 0;
2129 info.line_info_cnt = 0;
2130 info.jited_line_info_cnt = 0;
2134 ulen = info.xlated_prog_len;
2135 info.xlated_prog_len = bpf_prog_insn_size(prog);
2136 if (info.xlated_prog_len && ulen) {
2137 struct bpf_insn *insns_sanitized;
2140 if (prog->blinded && !bpf_dump_raw_ok()) {
2141 info.xlated_prog_insns = 0;
2144 insns_sanitized = bpf_insn_prepare_dump(prog);
2145 if (!insns_sanitized)
2147 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2148 ulen = min_t(u32, info.xlated_prog_len, ulen);
2149 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2150 kfree(insns_sanitized);
2155 if (bpf_prog_is_dev_bound(prog->aux)) {
2156 err = bpf_prog_offload_info_fill(&info, prog);
2162 /* NOTE: the following code is supposed to be skipped for offload.
2163 * bpf_prog_offload_info_fill() is the place to fill similar fields
2166 ulen = info.jited_prog_len;
2167 if (prog->aux->func_cnt) {
2170 info.jited_prog_len = 0;
2171 for (i = 0; i < prog->aux->func_cnt; i++)
2172 info.jited_prog_len += prog->aux->func[i]->jited_len;
2174 info.jited_prog_len = prog->jited_len;
2177 if (info.jited_prog_len && ulen) {
2178 if (bpf_dump_raw_ok()) {
2179 uinsns = u64_to_user_ptr(info.jited_prog_insns);
2180 ulen = min_t(u32, info.jited_prog_len, ulen);
2182 /* for multi-function programs, copy the JITed
2183 * instructions for all the functions
2185 if (prog->aux->func_cnt) {
2190 for (i = 0; i < prog->aux->func_cnt; i++) {
2191 len = prog->aux->func[i]->jited_len;
2192 len = min_t(u32, len, free);
2193 img = (u8 *) prog->aux->func[i]->bpf_func;
2194 if (copy_to_user(uinsns, img, len))
2202 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2206 info.jited_prog_insns = 0;
2210 ulen = info.nr_jited_ksyms;
2211 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2213 if (bpf_dump_raw_ok()) {
2214 unsigned long ksym_addr;
2215 u64 __user *user_ksyms;
2218 /* copy the address of the kernel symbol
2219 * corresponding to each function
2221 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2222 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2223 if (prog->aux->func_cnt) {
2224 for (i = 0; i < ulen; i++) {
2225 ksym_addr = (unsigned long)
2226 prog->aux->func[i]->bpf_func;
2227 if (put_user((u64) ksym_addr,
2232 ksym_addr = (unsigned long) prog->bpf_func;
2233 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2237 info.jited_ksyms = 0;
2241 ulen = info.nr_jited_func_lens;
2242 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2244 if (bpf_dump_raw_ok()) {
2245 u32 __user *user_lens;
2248 /* copy the JITed image lengths for each function */
2249 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2250 user_lens = u64_to_user_ptr(info.jited_func_lens);
2251 if (prog->aux->func_cnt) {
2252 for (i = 0; i < ulen; i++) {
2254 prog->aux->func[i]->jited_len;
2255 if (put_user(func_len, &user_lens[i]))
2259 func_len = prog->jited_len;
2260 if (put_user(func_len, &user_lens[0]))
2264 info.jited_func_lens = 0;
2269 info.btf_id = btf_id(prog->aux->btf);
2271 ulen = info.func_info_cnt;
2272 info.func_info_cnt = prog->aux->func_info_cnt;
2273 if (info.func_info_cnt && ulen) {
2274 if (bpf_dump_raw_ok()) {
2275 char __user *user_finfo;
2277 user_finfo = u64_to_user_ptr(info.func_info);
2278 ulen = min_t(u32, info.func_info_cnt, ulen);
2279 if (copy_to_user(user_finfo, prog->aux->func_info,
2280 info.func_info_rec_size * ulen))
2287 ulen = info.line_info_cnt;
2288 info.line_info_cnt = prog->aux->nr_linfo;
2289 if (info.line_info_cnt && ulen) {
2290 if (bpf_dump_raw_ok()) {
2291 __u8 __user *user_linfo;
2293 user_linfo = u64_to_user_ptr(info.line_info);
2294 ulen = min_t(u32, info.line_info_cnt, ulen);
2295 if (copy_to_user(user_linfo, prog->aux->linfo,
2296 info.line_info_rec_size * ulen))
2303 ulen = info.jited_line_info_cnt;
2304 if (prog->aux->jited_linfo)
2305 info.jited_line_info_cnt = prog->aux->nr_linfo;
2307 info.jited_line_info_cnt = 0;
2308 if (info.jited_line_info_cnt && ulen) {
2309 if (bpf_dump_raw_ok()) {
2310 __u64 __user *user_linfo;
2313 user_linfo = u64_to_user_ptr(info.jited_line_info);
2314 ulen = min_t(u32, info.jited_line_info_cnt, ulen);
2315 for (i = 0; i < ulen; i++) {
2316 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2321 info.jited_line_info = 0;
2326 if (copy_to_user(uinfo, &info, info_len) ||
2327 put_user(info_len, &uattr->info.info_len))
2333 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2334 const union bpf_attr *attr,
2335 union bpf_attr __user *uattr)
2337 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2338 struct bpf_map_info info = {};
2339 u32 info_len = attr->info.info_len;
2342 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2345 info_len = min_t(u32, sizeof(info), info_len);
2347 info.type = map->map_type;
2349 info.key_size = map->key_size;
2350 info.value_size = map->value_size;
2351 info.max_entries = map->max_entries;
2352 info.map_flags = map->map_flags;
2353 memcpy(info.name, map->name, sizeof(map->name));
2356 info.btf_id = btf_id(map->btf);
2357 info.btf_key_type_id = map->btf_key_type_id;
2358 info.btf_value_type_id = map->btf_value_type_id;
2361 if (bpf_map_is_dev_bound(map)) {
2362 err = bpf_map_offload_info_fill(&info, map);
2367 if (copy_to_user(uinfo, &info, info_len) ||
2368 put_user(info_len, &uattr->info.info_len))
2374 static int bpf_btf_get_info_by_fd(struct btf *btf,
2375 const union bpf_attr *attr,
2376 union bpf_attr __user *uattr)
2378 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2379 u32 info_len = attr->info.info_len;
2382 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2386 return btf_get_info_by_fd(btf, attr, uattr);
2389 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2391 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2392 union bpf_attr __user *uattr)
2394 int ufd = attr->info.bpf_fd;
2398 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2405 if (f.file->f_op == &bpf_prog_fops)
2406 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2408 else if (f.file->f_op == &bpf_map_fops)
2409 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2411 else if (f.file->f_op == &btf_fops)
2412 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2420 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2422 static int bpf_btf_load(const union bpf_attr *attr)
2424 if (CHECK_ATTR(BPF_BTF_LOAD))
2427 if (!capable(CAP_SYS_ADMIN))
2430 return btf_new_fd(attr);
2433 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2435 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2437 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2440 if (!capable(CAP_SYS_ADMIN))
2443 return btf_get_fd_by_id(attr->btf_id);
2446 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2447 union bpf_attr __user *uattr,
2448 u32 prog_id, u32 fd_type,
2449 const char *buf, u64 probe_offset,
2452 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2453 u32 len = buf ? strlen(buf) : 0, input_len;
2456 if (put_user(len, &uattr->task_fd_query.buf_len))
2458 input_len = attr->task_fd_query.buf_len;
2459 if (input_len && ubuf) {
2461 /* nothing to copy, just make ubuf NULL terminated */
2464 if (put_user(zero, ubuf))
2466 } else if (input_len >= len + 1) {
2467 /* ubuf can hold the string with NULL terminator */
2468 if (copy_to_user(ubuf, buf, len + 1))
2471 /* ubuf cannot hold the string with NULL terminator,
2472 * do a partial copy with NULL terminator.
2477 if (copy_to_user(ubuf, buf, input_len - 1))
2479 if (put_user(zero, ubuf + input_len - 1))
2484 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2485 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2486 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2487 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2493 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2495 static int bpf_task_fd_query(const union bpf_attr *attr,
2496 union bpf_attr __user *uattr)
2498 pid_t pid = attr->task_fd_query.pid;
2499 u32 fd = attr->task_fd_query.fd;
2500 const struct perf_event *event;
2501 struct files_struct *files;
2502 struct task_struct *task;
2506 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2509 if (!capable(CAP_SYS_ADMIN))
2512 if (attr->task_fd_query.flags != 0)
2515 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2519 files = get_files_struct(task);
2520 put_task_struct(task);
2525 spin_lock(&files->file_lock);
2526 file = fcheck_files(files, fd);
2531 spin_unlock(&files->file_lock);
2532 put_files_struct(files);
2537 if (file->f_op == &bpf_raw_tp_fops) {
2538 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2539 struct bpf_raw_event_map *btp = raw_tp->btp;
2541 err = bpf_task_fd_query_copy(attr, uattr,
2542 raw_tp->prog->aux->id,
2543 BPF_FD_TYPE_RAW_TRACEPOINT,
2544 btp->tp->name, 0, 0);
2548 event = perf_get_event(file);
2549 if (!IS_ERR(event)) {
2550 u64 probe_offset, probe_addr;
2551 u32 prog_id, fd_type;
2554 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2555 &buf, &probe_offset,
2558 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2572 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2574 union bpf_attr attr = {};
2577 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2580 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2583 size = min_t(u32, size, sizeof(attr));
2585 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2586 if (copy_from_user(&attr, uattr, size) != 0)
2589 err = security_bpf(cmd, &attr, size);
2594 case BPF_MAP_CREATE:
2595 err = map_create(&attr);
2597 case BPF_MAP_LOOKUP_ELEM:
2598 err = map_lookup_elem(&attr);
2600 case BPF_MAP_UPDATE_ELEM:
2601 err = map_update_elem(&attr);
2603 case BPF_MAP_DELETE_ELEM:
2604 err = map_delete_elem(&attr);
2606 case BPF_MAP_GET_NEXT_KEY:
2607 err = map_get_next_key(&attr);
2610 err = bpf_prog_load(&attr, uattr);
2613 err = bpf_obj_pin(&attr);
2616 err = bpf_obj_get(&attr);
2618 case BPF_PROG_ATTACH:
2619 err = bpf_prog_attach(&attr);
2621 case BPF_PROG_DETACH:
2622 err = bpf_prog_detach(&attr);
2624 case BPF_PROG_QUERY:
2625 err = bpf_prog_query(&attr, uattr);
2627 case BPF_PROG_TEST_RUN:
2628 err = bpf_prog_test_run(&attr, uattr);
2630 case BPF_PROG_GET_NEXT_ID:
2631 err = bpf_obj_get_next_id(&attr, uattr,
2632 &prog_idr, &prog_idr_lock);
2634 case BPF_MAP_GET_NEXT_ID:
2635 err = bpf_obj_get_next_id(&attr, uattr,
2636 &map_idr, &map_idr_lock);
2638 case BPF_PROG_GET_FD_BY_ID:
2639 err = bpf_prog_get_fd_by_id(&attr);
2641 case BPF_MAP_GET_FD_BY_ID:
2642 err = bpf_map_get_fd_by_id(&attr);
2644 case BPF_OBJ_GET_INFO_BY_FD:
2645 err = bpf_obj_get_info_by_fd(&attr, uattr);
2647 case BPF_RAW_TRACEPOINT_OPEN:
2648 err = bpf_raw_tracepoint_open(&attr);
2651 err = bpf_btf_load(&attr);
2653 case BPF_BTF_GET_FD_BY_ID:
2654 err = bpf_btf_get_fd_by_id(&attr);
2656 case BPF_TASK_FD_QUERY:
2657 err = bpf_task_fd_query(&attr, uattr);
2659 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2660 err = map_lookup_and_delete_elem(&attr);