1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
5 #include <linux/bpf_trace.h>
6 #include <linux/bpf_lirc.h>
8 #include <linux/syscalls.h>
9 #include <linux/slab.h>
10 #include <linux/sched/signal.h>
11 #include <linux/vmalloc.h>
12 #include <linux/mmzone.h>
13 #include <linux/anon_inodes.h>
14 #include <linux/fdtable.h>
15 #include <linux/file.h>
17 #include <linux/license.h>
18 #include <linux/filter.h>
19 #include <linux/version.h>
20 #include <linux/kernel.h>
21 #include <linux/idr.h>
22 #include <linux/cred.h>
23 #include <linux/timekeeping.h>
24 #include <linux/ctype.h>
25 #include <linux/nospec.h>
27 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
28 (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
29 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
30 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
31 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
32 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
34 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
36 DEFINE_PER_CPU(int, bpf_prog_active);
37 static DEFINE_IDR(prog_idr);
38 static DEFINE_SPINLOCK(prog_idr_lock);
39 static DEFINE_IDR(map_idr);
40 static DEFINE_SPINLOCK(map_idr_lock);
42 int sysctl_unprivileged_bpf_disabled __read_mostly;
44 static const struct bpf_map_ops * const bpf_map_types[] = {
45 #define BPF_PROG_TYPE(_id, _ops)
46 #define BPF_MAP_TYPE(_id, _ops) \
48 #include <linux/bpf_types.h>
54 * If we're handed a bigger struct than we know of, ensure all the unknown bits
55 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
56 * we don't know about yet.
58 * There is a ToCToU between this function call and the following
59 * copy_from_user() call. However, this is not a concern since this function is
60 * meant to be a future-proofing of bits.
62 int bpf_check_uarg_tail_zero(void __user *uaddr,
66 unsigned char __user *addr;
67 unsigned char __user *end;
71 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
74 if (unlikely(!access_ok(uaddr, actual_size)))
77 if (actual_size <= expected_size)
80 addr = uaddr + expected_size;
81 end = uaddr + actual_size;
83 for (; addr < end; addr++) {
84 err = get_user(val, addr);
94 const struct bpf_map_ops bpf_map_offload_ops = {
95 .map_alloc = bpf_map_offload_map_alloc,
96 .map_free = bpf_map_offload_map_free,
97 .map_check_btf = map_check_no_btf,
100 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
102 const struct bpf_map_ops *ops;
103 u32 type = attr->map_type;
107 if (type >= ARRAY_SIZE(bpf_map_types))
108 return ERR_PTR(-EINVAL);
109 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
110 ops = bpf_map_types[type];
112 return ERR_PTR(-EINVAL);
114 if (ops->map_alloc_check) {
115 err = ops->map_alloc_check(attr);
119 if (attr->map_ifindex)
120 ops = &bpf_map_offload_ops;
121 map = ops->map_alloc(attr);
125 map->map_type = type;
129 void *bpf_map_area_alloc(size_t size, int numa_node)
131 /* We really just want to fail instead of triggering OOM killer
132 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
133 * which is used for lower order allocation requests.
135 * It has been observed that higher order allocation requests done by
136 * vmalloc with __GFP_NORETRY being set might fail due to not trying
137 * to reclaim memory from the page cache, thus we set
138 * __GFP_RETRY_MAYFAIL to avoid such situations.
141 const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
144 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
145 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
151 return __vmalloc_node_flags_caller(size, numa_node,
152 GFP_KERNEL | __GFP_RETRY_MAYFAIL |
153 flags, __builtin_return_address(0));
156 void bpf_map_area_free(void *area)
161 static u32 bpf_map_flags_retain_permanent(u32 flags)
163 /* Some map creation flags are not tied to the map object but
164 * rather to the map fd instead, so they have no meaning upon
165 * map object inspection since multiple file descriptors with
166 * different (access) properties can exist here. Thus, given
167 * this has zero meaning for the map itself, lets clear these
170 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
173 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
175 map->map_type = attr->map_type;
176 map->key_size = attr->key_size;
177 map->value_size = attr->value_size;
178 map->max_entries = attr->max_entries;
179 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
180 map->numa_node = bpf_map_attr_numa_node(attr);
183 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
185 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
187 if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
188 atomic_long_sub(pages, &user->locked_vm);
194 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
197 atomic_long_sub(pages, &user->locked_vm);
200 int bpf_map_charge_init(struct bpf_map_memory *mem, size_t size)
202 u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
203 struct user_struct *user;
206 if (size >= U32_MAX - PAGE_SIZE)
209 user = get_current_user();
210 ret = bpf_charge_memlock(user, pages);
222 void bpf_map_charge_finish(struct bpf_map_memory *mem)
224 bpf_uncharge_memlock(mem->user, mem->pages);
228 void bpf_map_charge_move(struct bpf_map_memory *dst,
229 struct bpf_map_memory *src)
233 /* Make sure src will not be used for the redundant uncharging. */
234 memset(src, 0, sizeof(struct bpf_map_memory));
237 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
241 ret = bpf_charge_memlock(map->memory.user, pages);
244 map->memory.pages += pages;
248 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
250 bpf_uncharge_memlock(map->memory.user, pages);
251 map->memory.pages -= pages;
254 static int bpf_map_alloc_id(struct bpf_map *map)
258 idr_preload(GFP_KERNEL);
259 spin_lock_bh(&map_idr_lock);
260 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
263 spin_unlock_bh(&map_idr_lock);
266 if (WARN_ON_ONCE(!id))
269 return id > 0 ? 0 : id;
272 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
276 /* Offloaded maps are removed from the IDR store when their device
277 * disappears - even if someone holds an fd to them they are unusable,
278 * the memory is gone, all ops will fail; they are simply waiting for
279 * refcnt to drop to be freed.
285 spin_lock_irqsave(&map_idr_lock, flags);
287 __acquire(&map_idr_lock);
289 idr_remove(&map_idr, map->id);
293 spin_unlock_irqrestore(&map_idr_lock, flags);
295 __release(&map_idr_lock);
298 /* called from workqueue */
299 static void bpf_map_free_deferred(struct work_struct *work)
301 struct bpf_map *map = container_of(work, struct bpf_map, work);
302 struct bpf_map_memory mem;
304 bpf_map_charge_move(&mem, &map->memory);
305 security_bpf_map_free(map);
306 /* implementation dependent freeing */
307 map->ops->map_free(map);
308 bpf_map_charge_finish(&mem);
311 static void bpf_map_put_uref(struct bpf_map *map)
313 if (atomic_dec_and_test(&map->usercnt)) {
314 if (map->ops->map_release_uref)
315 map->ops->map_release_uref(map);
319 /* decrement map refcnt and schedule it for freeing via workqueue
320 * (unrelying map implementation ops->map_free() might sleep)
322 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
324 if (atomic_dec_and_test(&map->refcnt)) {
325 /* bpf_map_free_id() must be called first */
326 bpf_map_free_id(map, do_idr_lock);
328 INIT_WORK(&map->work, bpf_map_free_deferred);
329 schedule_work(&map->work);
333 void bpf_map_put(struct bpf_map *map)
335 __bpf_map_put(map, true);
337 EXPORT_SYMBOL_GPL(bpf_map_put);
339 void bpf_map_put_with_uref(struct bpf_map *map)
341 bpf_map_put_uref(map);
345 static int bpf_map_release(struct inode *inode, struct file *filp)
347 struct bpf_map *map = filp->private_data;
349 if (map->ops->map_release)
350 map->ops->map_release(map, filp);
352 bpf_map_put_with_uref(map);
356 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
358 fmode_t mode = f.file->f_mode;
360 /* Our file permissions may have been overridden by global
361 * map permissions facing syscall side.
363 if (READ_ONCE(map->frozen))
364 mode &= ~FMODE_CAN_WRITE;
368 #ifdef CONFIG_PROC_FS
369 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
371 const struct bpf_map *map = filp->private_data;
372 const struct bpf_array *array;
373 u32 owner_prog_type = 0;
376 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
377 array = container_of(map, struct bpf_array, map);
378 owner_prog_type = array->owner_prog_type;
379 owner_jited = array->owner_jited;
396 map->memory.pages * 1ULL << PAGE_SHIFT,
398 READ_ONCE(map->frozen));
400 if (owner_prog_type) {
401 seq_printf(m, "owner_prog_type:\t%u\n",
403 seq_printf(m, "owner_jited:\t%u\n",
409 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
412 /* We need this handler such that alloc_file() enables
413 * f_mode with FMODE_CAN_READ.
418 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
419 size_t siz, loff_t *ppos)
421 /* We need this handler such that alloc_file() enables
422 * f_mode with FMODE_CAN_WRITE.
427 const struct file_operations bpf_map_fops = {
428 #ifdef CONFIG_PROC_FS
429 .show_fdinfo = bpf_map_show_fdinfo,
431 .release = bpf_map_release,
432 .read = bpf_dummy_read,
433 .write = bpf_dummy_write,
436 int bpf_map_new_fd(struct bpf_map *map, int flags)
440 ret = security_bpf_map(map, OPEN_FMODE(flags));
444 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
448 int bpf_get_file_flag(int flags)
450 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
452 if (flags & BPF_F_RDONLY)
454 if (flags & BPF_F_WRONLY)
459 /* helper macro to check that unused fields 'union bpf_attr' are zero */
460 #define CHECK_ATTR(CMD) \
461 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
462 sizeof(attr->CMD##_LAST_FIELD), 0, \
464 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
465 sizeof(attr->CMD##_LAST_FIELD)) != NULL
467 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
468 * Return 0 on success and < 0 on error.
470 static int bpf_obj_name_cpy(char *dst, const char *src)
472 const char *end = src + BPF_OBJ_NAME_LEN;
474 memset(dst, 0, BPF_OBJ_NAME_LEN);
475 /* Copy all isalnum(), '_' and '.' chars. */
476 while (src < end && *src) {
477 if (!isalnum(*src) &&
478 *src != '_' && *src != '.')
483 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
490 int map_check_no_btf(const struct bpf_map *map,
491 const struct btf *btf,
492 const struct btf_type *key_type,
493 const struct btf_type *value_type)
498 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
499 u32 btf_key_id, u32 btf_value_id)
501 const struct btf_type *key_type, *value_type;
502 u32 key_size, value_size;
505 /* Some maps allow key to be unspecified. */
507 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
508 if (!key_type || key_size != map->key_size)
511 key_type = btf_type_by_id(btf, 0);
512 if (!map->ops->map_check_btf)
516 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
517 if (!value_type || value_size != map->value_size)
520 map->spin_lock_off = btf_find_spin_lock(btf, value_type);
522 if (map_value_has_spin_lock(map)) {
523 if (map->map_flags & BPF_F_RDONLY_PROG)
525 if (map->map_type != BPF_MAP_TYPE_HASH &&
526 map->map_type != BPF_MAP_TYPE_ARRAY &&
527 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
528 map->map_type != BPF_MAP_TYPE_SK_STORAGE)
530 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
533 "verifier bug spin_lock_off %d value_size %d\n",
534 map->spin_lock_off, map->value_size);
539 if (map->ops->map_check_btf)
540 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
545 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
546 /* called via syscall */
547 static int map_create(union bpf_attr *attr)
549 int numa_node = bpf_map_attr_numa_node(attr);
550 struct bpf_map_memory mem;
555 err = CHECK_ATTR(BPF_MAP_CREATE);
559 f_flags = bpf_get_file_flag(attr->map_flags);
563 if (numa_node != NUMA_NO_NODE &&
564 ((unsigned int)numa_node >= nr_node_ids ||
565 !node_online(numa_node)))
568 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
569 map = find_and_alloc_map(attr);
573 err = bpf_obj_name_cpy(map->name, attr->map_name);
577 atomic_set(&map->refcnt, 1);
578 atomic_set(&map->usercnt, 1);
580 if (attr->btf_key_type_id || attr->btf_value_type_id) {
583 if (!attr->btf_value_type_id) {
588 btf = btf_get_by_fd(attr->btf_fd);
594 err = map_check_btf(map, btf, attr->btf_key_type_id,
595 attr->btf_value_type_id);
602 map->btf_key_type_id = attr->btf_key_type_id;
603 map->btf_value_type_id = attr->btf_value_type_id;
605 map->spin_lock_off = -EINVAL;
608 err = security_bpf_map_alloc(map);
612 err = bpf_map_alloc_id(map);
616 err = bpf_map_new_fd(map, f_flags);
618 /* failed to allocate fd.
619 * bpf_map_put_with_uref() is needed because the above
620 * bpf_map_alloc_id() has published the map
621 * to the userspace and the userspace may
622 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
624 bpf_map_put_with_uref(map);
631 security_bpf_map_free(map);
634 bpf_map_charge_move(&mem, &map->memory);
635 map->ops->map_free(map);
636 bpf_map_charge_finish(&mem);
640 /* if error is returned, fd is released.
641 * On success caller should complete fd access with matching fdput()
643 struct bpf_map *__bpf_map_get(struct fd f)
646 return ERR_PTR(-EBADF);
647 if (f.file->f_op != &bpf_map_fops) {
649 return ERR_PTR(-EINVAL);
652 return f.file->private_data;
655 /* prog's and map's refcnt limit */
656 #define BPF_MAX_REFCNT 32768
658 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
660 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
661 atomic_dec(&map->refcnt);
662 return ERR_PTR(-EBUSY);
665 atomic_inc(&map->usercnt);
668 EXPORT_SYMBOL_GPL(bpf_map_inc);
670 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
672 struct fd f = fdget(ufd);
675 map = __bpf_map_get(f);
679 map = bpf_map_inc(map, true);
685 /* map_idr_lock should have been held */
686 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
691 refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
693 if (refold >= BPF_MAX_REFCNT) {
694 __bpf_map_put(map, false);
695 return ERR_PTR(-EBUSY);
699 return ERR_PTR(-ENOENT);
702 atomic_inc(&map->usercnt);
707 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
712 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
715 return memdup_user(ukey, key_size);
718 return ERR_PTR(-EINVAL);
723 /* last field in 'union bpf_attr' used by this command */
724 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
726 static int map_lookup_elem(union bpf_attr *attr)
728 void __user *ukey = u64_to_user_ptr(attr->key);
729 void __user *uvalue = u64_to_user_ptr(attr->value);
730 int ufd = attr->map_fd;
732 void *key, *value, *ptr;
737 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
740 if (attr->flags & ~BPF_F_LOCK)
744 map = __bpf_map_get(f);
747 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
752 if ((attr->flags & BPF_F_LOCK) &&
753 !map_value_has_spin_lock(map)) {
758 key = __bpf_copy_key(ukey, map->key_size);
764 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
765 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
766 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
767 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
768 value_size = round_up(map->value_size, 8) * num_possible_cpus();
769 else if (IS_FD_MAP(map))
770 value_size = sizeof(u32);
772 value_size = map->value_size;
775 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
779 if (bpf_map_is_dev_bound(map)) {
780 err = bpf_map_offload_lookup_elem(map, key, value);
785 this_cpu_inc(bpf_prog_active);
786 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
787 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
788 err = bpf_percpu_hash_copy(map, key, value);
789 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
790 err = bpf_percpu_array_copy(map, key, value);
791 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
792 err = bpf_percpu_cgroup_storage_copy(map, key, value);
793 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
794 err = bpf_stackmap_copy(map, key, value);
795 } else if (IS_FD_ARRAY(map)) {
796 err = bpf_fd_array_map_lookup_elem(map, key, value);
797 } else if (IS_FD_HASH(map)) {
798 err = bpf_fd_htab_map_lookup_elem(map, key, value);
799 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
800 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
801 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
802 map->map_type == BPF_MAP_TYPE_STACK) {
803 err = map->ops->map_peek_elem(map, value);
806 if (map->ops->map_lookup_elem_sys_only)
807 ptr = map->ops->map_lookup_elem_sys_only(map, key);
809 ptr = map->ops->map_lookup_elem(map, key);
816 if (attr->flags & BPF_F_LOCK)
817 /* lock 'ptr' and copy everything but lock */
818 copy_map_value_locked(map, value, ptr, true);
820 copy_map_value(map, value, ptr);
821 /* mask lock, since value wasn't zero inited */
822 check_and_init_map_lock(map, value);
826 this_cpu_dec(bpf_prog_active);
834 if (copy_to_user(uvalue, value, value_size) != 0)
848 static void maybe_wait_bpf_programs(struct bpf_map *map)
850 /* Wait for any running BPF programs to complete so that
851 * userspace, when we return to it, knows that all programs
852 * that could be running use the new map value.
854 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
855 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
859 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
861 static int map_update_elem(union bpf_attr *attr)
863 void __user *ukey = u64_to_user_ptr(attr->key);
864 void __user *uvalue = u64_to_user_ptr(attr->value);
865 int ufd = attr->map_fd;
872 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
876 map = __bpf_map_get(f);
879 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
884 if ((attr->flags & BPF_F_LOCK) &&
885 !map_value_has_spin_lock(map)) {
890 key = __bpf_copy_key(ukey, map->key_size);
896 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
897 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
898 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
899 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
900 value_size = round_up(map->value_size, 8) * num_possible_cpus();
902 value_size = map->value_size;
905 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
910 if (copy_from_user(value, uvalue, value_size) != 0)
913 /* Need to create a kthread, thus must support schedule */
914 if (bpf_map_is_dev_bound(map)) {
915 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
917 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
918 map->map_type == BPF_MAP_TYPE_SOCKHASH ||
919 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
920 err = map->ops->map_update_elem(map, key, value, attr->flags);
924 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
925 * inside bpf map update or delete otherwise deadlocks are possible
928 __this_cpu_inc(bpf_prog_active);
929 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
930 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
931 err = bpf_percpu_hash_update(map, key, value, attr->flags);
932 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
933 err = bpf_percpu_array_update(map, key, value, attr->flags);
934 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
935 err = bpf_percpu_cgroup_storage_update(map, key, value,
937 } else if (IS_FD_ARRAY(map)) {
939 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
942 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
944 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
947 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
948 /* rcu_read_lock() is not needed */
949 err = bpf_fd_reuseport_array_update_elem(map, key, value,
951 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
952 map->map_type == BPF_MAP_TYPE_STACK) {
953 err = map->ops->map_push_elem(map, value, attr->flags);
956 err = map->ops->map_update_elem(map, key, value, attr->flags);
959 __this_cpu_dec(bpf_prog_active);
961 maybe_wait_bpf_programs(map);
972 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
974 static int map_delete_elem(union bpf_attr *attr)
976 void __user *ukey = u64_to_user_ptr(attr->key);
977 int ufd = attr->map_fd;
983 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
987 map = __bpf_map_get(f);
990 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
995 key = __bpf_copy_key(ukey, map->key_size);
1001 if (bpf_map_is_dev_bound(map)) {
1002 err = bpf_map_offload_delete_elem(map, key);
1007 __this_cpu_inc(bpf_prog_active);
1009 err = map->ops->map_delete_elem(map, key);
1011 __this_cpu_dec(bpf_prog_active);
1013 maybe_wait_bpf_programs(map);
1021 /* last field in 'union bpf_attr' used by this command */
1022 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1024 static int map_get_next_key(union bpf_attr *attr)
1026 void __user *ukey = u64_to_user_ptr(attr->key);
1027 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1028 int ufd = attr->map_fd;
1029 struct bpf_map *map;
1030 void *key, *next_key;
1034 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1038 map = __bpf_map_get(f);
1040 return PTR_ERR(map);
1041 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1047 key = __bpf_copy_key(ukey, map->key_size);
1057 next_key = kmalloc(map->key_size, GFP_USER);
1061 if (bpf_map_is_dev_bound(map)) {
1062 err = bpf_map_offload_get_next_key(map, key, next_key);
1067 err = map->ops->map_get_next_key(map, key, next_key);
1074 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1088 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1090 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1092 void __user *ukey = u64_to_user_ptr(attr->key);
1093 void __user *uvalue = u64_to_user_ptr(attr->value);
1094 int ufd = attr->map_fd;
1095 struct bpf_map *map;
1101 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1105 map = __bpf_map_get(f);
1107 return PTR_ERR(map);
1108 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1113 key = __bpf_copy_key(ukey, map->key_size);
1119 value_size = map->value_size;
1122 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1126 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1127 map->map_type == BPF_MAP_TYPE_STACK) {
1128 err = map->ops->map_pop_elem(map, value);
1136 if (copy_to_user(uvalue, value, value_size) != 0)
1150 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1152 static int map_freeze(const union bpf_attr *attr)
1154 int err = 0, ufd = attr->map_fd;
1155 struct bpf_map *map;
1158 if (CHECK_ATTR(BPF_MAP_FREEZE))
1162 map = __bpf_map_get(f);
1164 return PTR_ERR(map);
1165 if (READ_ONCE(map->frozen)) {
1169 if (!capable(CAP_SYS_ADMIN)) {
1174 WRITE_ONCE(map->frozen, true);
1180 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1181 #define BPF_PROG_TYPE(_id, _name) \
1182 [_id] = & _name ## _prog_ops,
1183 #define BPF_MAP_TYPE(_id, _ops)
1184 #include <linux/bpf_types.h>
1185 #undef BPF_PROG_TYPE
1189 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1191 const struct bpf_prog_ops *ops;
1193 if (type >= ARRAY_SIZE(bpf_prog_types))
1195 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1196 ops = bpf_prog_types[type];
1200 if (!bpf_prog_is_dev_bound(prog->aux))
1201 prog->aux->ops = ops;
1203 prog->aux->ops = &bpf_offload_prog_ops;
1208 /* drop refcnt on maps used by eBPF program and free auxilary data */
1209 static void free_used_maps(struct bpf_prog_aux *aux)
1211 enum bpf_cgroup_storage_type stype;
1214 for_each_cgroup_storage_type(stype) {
1215 if (!aux->cgroup_storage[stype])
1217 bpf_cgroup_storage_release(aux->prog,
1218 aux->cgroup_storage[stype]);
1221 for (i = 0; i < aux->used_map_cnt; i++)
1222 bpf_map_put(aux->used_maps[i]);
1224 kfree(aux->used_maps);
1227 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1229 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1230 unsigned long user_bufs;
1233 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1234 if (user_bufs > memlock_limit) {
1235 atomic_long_sub(pages, &user->locked_vm);
1243 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1246 atomic_long_sub(pages, &user->locked_vm);
1249 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1251 struct user_struct *user = get_current_user();
1254 ret = __bpf_prog_charge(user, prog->pages);
1260 prog->aux->user = user;
1264 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1266 struct user_struct *user = prog->aux->user;
1268 __bpf_prog_uncharge(user, prog->pages);
1272 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1276 idr_preload(GFP_KERNEL);
1277 spin_lock_bh(&prog_idr_lock);
1278 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1281 spin_unlock_bh(&prog_idr_lock);
1284 /* id is in [1, INT_MAX) */
1285 if (WARN_ON_ONCE(!id))
1288 return id > 0 ? 0 : id;
1291 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1293 /* cBPF to eBPF migrations are currently not in the idr store.
1294 * Offloaded programs are removed from the store when their device
1295 * disappears - even if someone grabs an fd to them they are unusable,
1296 * simply waiting for refcnt to drop to be freed.
1302 spin_lock_bh(&prog_idr_lock);
1304 __acquire(&prog_idr_lock);
1306 idr_remove(&prog_idr, prog->aux->id);
1310 spin_unlock_bh(&prog_idr_lock);
1312 __release(&prog_idr_lock);
1315 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1317 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1319 free_used_maps(aux);
1320 bpf_prog_uncharge_memlock(aux->prog);
1321 security_bpf_prog_free(aux);
1322 bpf_prog_free(aux->prog);
1325 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1327 if (atomic_dec_and_test(&prog->aux->refcnt)) {
1328 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1329 /* bpf_prog_free_id() must be called first */
1330 bpf_prog_free_id(prog, do_idr_lock);
1331 bpf_prog_kallsyms_del_all(prog);
1332 btf_put(prog->aux->btf);
1333 kvfree(prog->aux->func_info);
1334 bpf_prog_free_linfo(prog);
1336 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1340 void bpf_prog_put(struct bpf_prog *prog)
1342 __bpf_prog_put(prog, true);
1344 EXPORT_SYMBOL_GPL(bpf_prog_put);
1346 static int bpf_prog_release(struct inode *inode, struct file *filp)
1348 struct bpf_prog *prog = filp->private_data;
1354 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1355 struct bpf_prog_stats *stats)
1357 u64 nsecs = 0, cnt = 0;
1360 for_each_possible_cpu(cpu) {
1361 const struct bpf_prog_stats *st;
1365 st = per_cpu_ptr(prog->aux->stats, cpu);
1367 start = u64_stats_fetch_begin_irq(&st->syncp);
1370 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1374 stats->nsecs = nsecs;
1378 #ifdef CONFIG_PROC_FS
1379 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1381 const struct bpf_prog *prog = filp->private_data;
1382 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1383 struct bpf_prog_stats stats;
1385 bpf_prog_get_stats(prog, &stats);
1386 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1393 "run_time_ns:\t%llu\n"
1398 prog->pages * 1ULL << PAGE_SHIFT,
1405 const struct file_operations bpf_prog_fops = {
1406 #ifdef CONFIG_PROC_FS
1407 .show_fdinfo = bpf_prog_show_fdinfo,
1409 .release = bpf_prog_release,
1410 .read = bpf_dummy_read,
1411 .write = bpf_dummy_write,
1414 int bpf_prog_new_fd(struct bpf_prog *prog)
1418 ret = security_bpf_prog(prog);
1422 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1423 O_RDWR | O_CLOEXEC);
1426 static struct bpf_prog *____bpf_prog_get(struct fd f)
1429 return ERR_PTR(-EBADF);
1430 if (f.file->f_op != &bpf_prog_fops) {
1432 return ERR_PTR(-EINVAL);
1435 return f.file->private_data;
1438 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1440 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1441 atomic_sub(i, &prog->aux->refcnt);
1442 return ERR_PTR(-EBUSY);
1446 EXPORT_SYMBOL_GPL(bpf_prog_add);
1448 void bpf_prog_sub(struct bpf_prog *prog, int i)
1450 /* Only to be used for undoing previous bpf_prog_add() in some
1451 * error path. We still know that another entity in our call
1452 * path holds a reference to the program, thus atomic_sub() can
1453 * be safely used in such cases!
1455 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1457 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1459 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1461 return bpf_prog_add(prog, 1);
1463 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1465 /* prog_idr_lock should have been held */
1466 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1470 refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1472 if (refold >= BPF_MAX_REFCNT) {
1473 __bpf_prog_put(prog, false);
1474 return ERR_PTR(-EBUSY);
1478 return ERR_PTR(-ENOENT);
1482 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1484 bool bpf_prog_get_ok(struct bpf_prog *prog,
1485 enum bpf_prog_type *attach_type, bool attach_drv)
1487 /* not an attachment, just a refcount inc, always allow */
1491 if (prog->type != *attach_type)
1493 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1499 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1502 struct fd f = fdget(ufd);
1503 struct bpf_prog *prog;
1505 prog = ____bpf_prog_get(f);
1508 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1509 prog = ERR_PTR(-EINVAL);
1513 prog = bpf_prog_inc(prog);
1519 struct bpf_prog *bpf_prog_get(u32 ufd)
1521 return __bpf_prog_get(ufd, NULL, false);
1524 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1527 return __bpf_prog_get(ufd, &type, attach_drv);
1529 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1531 /* Initially all BPF programs could be loaded w/o specifying
1532 * expected_attach_type. Later for some of them specifying expected_attach_type
1533 * at load time became required so that program could be validated properly.
1534 * Programs of types that are allowed to be loaded both w/ and w/o (for
1535 * backward compatibility) expected_attach_type, should have the default attach
1536 * type assigned to expected_attach_type for the latter case, so that it can be
1537 * validated later at attach time.
1539 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1540 * prog type requires it but has some attach types that have to be backward
1543 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1545 switch (attr->prog_type) {
1546 case BPF_PROG_TYPE_CGROUP_SOCK:
1547 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1548 * exist so checking for non-zero is the way to go here.
1550 if (!attr->expected_attach_type)
1551 attr->expected_attach_type =
1552 BPF_CGROUP_INET_SOCK_CREATE;
1558 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1559 enum bpf_attach_type expected_attach_type)
1561 switch (prog_type) {
1562 case BPF_PROG_TYPE_CGROUP_SOCK:
1563 switch (expected_attach_type) {
1564 case BPF_CGROUP_INET_SOCK_CREATE:
1565 case BPF_CGROUP_INET4_POST_BIND:
1566 case BPF_CGROUP_INET6_POST_BIND:
1571 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1572 switch (expected_attach_type) {
1573 case BPF_CGROUP_INET4_BIND:
1574 case BPF_CGROUP_INET6_BIND:
1575 case BPF_CGROUP_INET4_CONNECT:
1576 case BPF_CGROUP_INET6_CONNECT:
1577 case BPF_CGROUP_UDP4_SENDMSG:
1578 case BPF_CGROUP_UDP6_SENDMSG:
1579 case BPF_CGROUP_UDP4_RECVMSG:
1580 case BPF_CGROUP_UDP6_RECVMSG:
1585 case BPF_PROG_TYPE_CGROUP_SKB:
1586 switch (expected_attach_type) {
1587 case BPF_CGROUP_INET_INGRESS:
1588 case BPF_CGROUP_INET_EGRESS:
1593 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1594 switch (expected_attach_type) {
1595 case BPF_CGROUP_SETSOCKOPT:
1596 case BPF_CGROUP_GETSOCKOPT:
1606 /* last field in 'union bpf_attr' used by this command */
1607 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1609 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1611 enum bpf_prog_type type = attr->prog_type;
1612 struct bpf_prog *prog;
1617 if (CHECK_ATTR(BPF_PROG_LOAD))
1620 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
1621 BPF_F_ANY_ALIGNMENT |
1622 BPF_F_TEST_RND_HI32))
1625 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1626 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1627 !capable(CAP_SYS_ADMIN))
1630 /* copy eBPF program license from user space */
1631 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1632 sizeof(license) - 1) < 0)
1634 license[sizeof(license) - 1] = 0;
1636 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1637 is_gpl = license_is_gpl_compatible(license);
1639 if (attr->insn_cnt == 0 ||
1640 attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1642 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1643 type != BPF_PROG_TYPE_CGROUP_SKB &&
1644 !capable(CAP_SYS_ADMIN))
1647 bpf_prog_load_fixup_attach_type(attr);
1648 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1651 /* plain bpf_prog allocation */
1652 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1656 prog->expected_attach_type = attr->expected_attach_type;
1658 prog->aux->offload_requested = !!attr->prog_ifindex;
1660 err = security_bpf_prog_alloc(prog->aux);
1662 goto free_prog_nouncharge;
1664 err = bpf_prog_charge_memlock(prog);
1668 prog->len = attr->insn_cnt;
1671 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1672 bpf_prog_insn_size(prog)) != 0)
1675 prog->orig_prog = NULL;
1678 atomic_set(&prog->aux->refcnt, 1);
1679 prog->gpl_compatible = is_gpl ? 1 : 0;
1681 if (bpf_prog_is_dev_bound(prog->aux)) {
1682 err = bpf_prog_offload_init(prog, attr);
1687 /* find program type: socket_filter vs tracing_filter */
1688 err = find_prog_type(type, prog);
1692 prog->aux->load_time = ktime_get_boottime_ns();
1693 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1697 /* run eBPF verifier */
1698 err = bpf_check(&prog, attr, uattr);
1700 goto free_used_maps;
1702 prog = bpf_prog_select_runtime(prog, &err);
1704 goto free_used_maps;
1706 err = bpf_prog_alloc_id(prog);
1708 goto free_used_maps;
1710 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
1711 * effectively publicly exposed. However, retrieving via
1712 * bpf_prog_get_fd_by_id() will take another reference,
1713 * therefore it cannot be gone underneath us.
1715 * Only for the time /after/ successful bpf_prog_new_fd()
1716 * and before returning to userspace, we might just hold
1717 * one reference and any parallel close on that fd could
1718 * rip everything out. Hence, below notifications must
1719 * happen before bpf_prog_new_fd().
1721 * Also, any failure handling from this point onwards must
1722 * be using bpf_prog_put() given the program is exposed.
1724 bpf_prog_kallsyms_add(prog);
1725 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1727 err = bpf_prog_new_fd(prog);
1733 bpf_prog_free_linfo(prog);
1734 kvfree(prog->aux->func_info);
1735 btf_put(prog->aux->btf);
1736 bpf_prog_kallsyms_del_subprogs(prog);
1737 free_used_maps(prog->aux);
1739 bpf_prog_uncharge_memlock(prog);
1741 security_bpf_prog_free(prog->aux);
1742 free_prog_nouncharge:
1743 bpf_prog_free(prog);
1747 #define BPF_OBJ_LAST_FIELD file_flags
1749 static int bpf_obj_pin(const union bpf_attr *attr)
1751 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1754 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1757 static int bpf_obj_get(const union bpf_attr *attr)
1759 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1760 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1763 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1767 struct bpf_raw_tracepoint {
1768 struct bpf_raw_event_map *btp;
1769 struct bpf_prog *prog;
1772 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1774 struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1777 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1778 bpf_prog_put(raw_tp->prog);
1780 bpf_put_raw_tracepoint(raw_tp->btp);
1785 static const struct file_operations bpf_raw_tp_fops = {
1786 .release = bpf_raw_tracepoint_release,
1787 .read = bpf_dummy_read,
1788 .write = bpf_dummy_write,
1791 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1793 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1795 struct bpf_raw_tracepoint *raw_tp;
1796 struct bpf_raw_event_map *btp;
1797 struct bpf_prog *prog;
1801 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1802 sizeof(tp_name) - 1) < 0)
1804 tp_name[sizeof(tp_name) - 1] = 0;
1806 btp = bpf_get_raw_tracepoint(tp_name);
1810 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1817 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1819 err = PTR_ERR(prog);
1822 if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1823 prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1828 err = bpf_probe_register(raw_tp->btp, prog);
1832 raw_tp->prog = prog;
1833 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1836 bpf_probe_unregister(raw_tp->btp, prog);
1847 bpf_put_raw_tracepoint(btp);
1851 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1852 enum bpf_attach_type attach_type)
1854 switch (prog->type) {
1855 case BPF_PROG_TYPE_CGROUP_SOCK:
1856 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1857 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1858 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1859 case BPF_PROG_TYPE_CGROUP_SKB:
1860 return prog->enforce_expected_attach_type &&
1861 prog->expected_attach_type != attach_type ?
1868 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1870 #define BPF_F_ATTACH_MASK \
1871 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1873 static int bpf_prog_attach(const union bpf_attr *attr)
1875 enum bpf_prog_type ptype;
1876 struct bpf_prog *prog;
1879 if (!capable(CAP_NET_ADMIN))
1882 if (CHECK_ATTR(BPF_PROG_ATTACH))
1885 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1888 switch (attr->attach_type) {
1889 case BPF_CGROUP_INET_INGRESS:
1890 case BPF_CGROUP_INET_EGRESS:
1891 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1893 case BPF_CGROUP_INET_SOCK_CREATE:
1894 case BPF_CGROUP_INET4_POST_BIND:
1895 case BPF_CGROUP_INET6_POST_BIND:
1896 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1898 case BPF_CGROUP_INET4_BIND:
1899 case BPF_CGROUP_INET6_BIND:
1900 case BPF_CGROUP_INET4_CONNECT:
1901 case BPF_CGROUP_INET6_CONNECT:
1902 case BPF_CGROUP_UDP4_SENDMSG:
1903 case BPF_CGROUP_UDP6_SENDMSG:
1904 case BPF_CGROUP_UDP4_RECVMSG:
1905 case BPF_CGROUP_UDP6_RECVMSG:
1906 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1908 case BPF_CGROUP_SOCK_OPS:
1909 ptype = BPF_PROG_TYPE_SOCK_OPS;
1911 case BPF_CGROUP_DEVICE:
1912 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1914 case BPF_SK_MSG_VERDICT:
1915 ptype = BPF_PROG_TYPE_SK_MSG;
1917 case BPF_SK_SKB_STREAM_PARSER:
1918 case BPF_SK_SKB_STREAM_VERDICT:
1919 ptype = BPF_PROG_TYPE_SK_SKB;
1921 case BPF_LIRC_MODE2:
1922 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1924 case BPF_FLOW_DISSECTOR:
1925 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1927 case BPF_CGROUP_SYSCTL:
1928 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1930 case BPF_CGROUP_GETSOCKOPT:
1931 case BPF_CGROUP_SETSOCKOPT:
1932 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
1938 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1940 return PTR_ERR(prog);
1942 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1948 case BPF_PROG_TYPE_SK_SKB:
1949 case BPF_PROG_TYPE_SK_MSG:
1950 ret = sock_map_get_from_fd(attr, prog);
1952 case BPF_PROG_TYPE_LIRC_MODE2:
1953 ret = lirc_prog_attach(attr, prog);
1955 case BPF_PROG_TYPE_FLOW_DISSECTOR:
1956 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1959 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1967 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1969 static int bpf_prog_detach(const union bpf_attr *attr)
1971 enum bpf_prog_type ptype;
1973 if (!capable(CAP_NET_ADMIN))
1976 if (CHECK_ATTR(BPF_PROG_DETACH))
1979 switch (attr->attach_type) {
1980 case BPF_CGROUP_INET_INGRESS:
1981 case BPF_CGROUP_INET_EGRESS:
1982 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1984 case BPF_CGROUP_INET_SOCK_CREATE:
1985 case BPF_CGROUP_INET4_POST_BIND:
1986 case BPF_CGROUP_INET6_POST_BIND:
1987 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1989 case BPF_CGROUP_INET4_BIND:
1990 case BPF_CGROUP_INET6_BIND:
1991 case BPF_CGROUP_INET4_CONNECT:
1992 case BPF_CGROUP_INET6_CONNECT:
1993 case BPF_CGROUP_UDP4_SENDMSG:
1994 case BPF_CGROUP_UDP6_SENDMSG:
1995 case BPF_CGROUP_UDP4_RECVMSG:
1996 case BPF_CGROUP_UDP6_RECVMSG:
1997 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1999 case BPF_CGROUP_SOCK_OPS:
2000 ptype = BPF_PROG_TYPE_SOCK_OPS;
2002 case BPF_CGROUP_DEVICE:
2003 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2005 case BPF_SK_MSG_VERDICT:
2006 return sock_map_get_from_fd(attr, NULL);
2007 case BPF_SK_SKB_STREAM_PARSER:
2008 case BPF_SK_SKB_STREAM_VERDICT:
2009 return sock_map_get_from_fd(attr, NULL);
2010 case BPF_LIRC_MODE2:
2011 return lirc_prog_detach(attr);
2012 case BPF_FLOW_DISSECTOR:
2013 return skb_flow_dissector_bpf_prog_detach(attr);
2014 case BPF_CGROUP_SYSCTL:
2015 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2017 case BPF_CGROUP_GETSOCKOPT:
2018 case BPF_CGROUP_SETSOCKOPT:
2019 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2025 return cgroup_bpf_prog_detach(attr, ptype);
2028 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2030 static int bpf_prog_query(const union bpf_attr *attr,
2031 union bpf_attr __user *uattr)
2033 if (!capable(CAP_NET_ADMIN))
2035 if (CHECK_ATTR(BPF_PROG_QUERY))
2037 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2040 switch (attr->query.attach_type) {
2041 case BPF_CGROUP_INET_INGRESS:
2042 case BPF_CGROUP_INET_EGRESS:
2043 case BPF_CGROUP_INET_SOCK_CREATE:
2044 case BPF_CGROUP_INET4_BIND:
2045 case BPF_CGROUP_INET6_BIND:
2046 case BPF_CGROUP_INET4_POST_BIND:
2047 case BPF_CGROUP_INET6_POST_BIND:
2048 case BPF_CGROUP_INET4_CONNECT:
2049 case BPF_CGROUP_INET6_CONNECT:
2050 case BPF_CGROUP_UDP4_SENDMSG:
2051 case BPF_CGROUP_UDP6_SENDMSG:
2052 case BPF_CGROUP_UDP4_RECVMSG:
2053 case BPF_CGROUP_UDP6_RECVMSG:
2054 case BPF_CGROUP_SOCK_OPS:
2055 case BPF_CGROUP_DEVICE:
2056 case BPF_CGROUP_SYSCTL:
2057 case BPF_CGROUP_GETSOCKOPT:
2058 case BPF_CGROUP_SETSOCKOPT:
2060 case BPF_LIRC_MODE2:
2061 return lirc_prog_query(attr, uattr);
2062 case BPF_FLOW_DISSECTOR:
2063 return skb_flow_dissector_prog_query(attr, uattr);
2068 return cgroup_bpf_prog_query(attr, uattr);
2071 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2073 static int bpf_prog_test_run(const union bpf_attr *attr,
2074 union bpf_attr __user *uattr)
2076 struct bpf_prog *prog;
2077 int ret = -ENOTSUPP;
2079 if (!capable(CAP_SYS_ADMIN))
2081 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2084 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2085 (!attr->test.ctx_size_in && attr->test.ctx_in))
2088 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2089 (!attr->test.ctx_size_out && attr->test.ctx_out))
2092 prog = bpf_prog_get(attr->test.prog_fd);
2094 return PTR_ERR(prog);
2096 if (prog->aux->ops->test_run)
2097 ret = prog->aux->ops->test_run(prog, attr, uattr);
2103 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2105 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2106 union bpf_attr __user *uattr,
2110 u32 next_id = attr->start_id;
2113 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2116 if (!capable(CAP_SYS_ADMIN))
2121 if (!idr_get_next(idr, &next_id))
2123 spin_unlock_bh(lock);
2126 err = put_user(next_id, &uattr->next_id);
2131 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2133 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2135 struct bpf_prog *prog;
2136 u32 id = attr->prog_id;
2139 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2142 if (!capable(CAP_SYS_ADMIN))
2145 spin_lock_bh(&prog_idr_lock);
2146 prog = idr_find(&prog_idr, id);
2148 prog = bpf_prog_inc_not_zero(prog);
2150 prog = ERR_PTR(-ENOENT);
2151 spin_unlock_bh(&prog_idr_lock);
2154 return PTR_ERR(prog);
2156 fd = bpf_prog_new_fd(prog);
2163 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2165 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2167 struct bpf_map *map;
2168 u32 id = attr->map_id;
2172 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2173 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2176 if (!capable(CAP_SYS_ADMIN))
2179 f_flags = bpf_get_file_flag(attr->open_flags);
2183 spin_lock_bh(&map_idr_lock);
2184 map = idr_find(&map_idr, id);
2186 map = bpf_map_inc_not_zero(map, true);
2188 map = ERR_PTR(-ENOENT);
2189 spin_unlock_bh(&map_idr_lock);
2192 return PTR_ERR(map);
2194 fd = bpf_map_new_fd(map, f_flags);
2196 bpf_map_put_with_uref(map);
2201 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2202 unsigned long addr, u32 *off,
2205 const struct bpf_map *map;
2208 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2209 map = prog->aux->used_maps[i];
2210 if (map == (void *)addr) {
2211 *type = BPF_PSEUDO_MAP_FD;
2214 if (!map->ops->map_direct_value_meta)
2216 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2217 *type = BPF_PSEUDO_MAP_VALUE;
2225 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2227 const struct bpf_map *map;
2228 struct bpf_insn *insns;
2233 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2238 for (i = 0; i < prog->len; i++) {
2239 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2240 insns[i].code = BPF_JMP | BPF_CALL;
2241 insns[i].imm = BPF_FUNC_tail_call;
2244 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2245 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2246 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2247 insns[i].code = BPF_JMP | BPF_CALL;
2248 if (!bpf_dump_raw_ok())
2253 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2256 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2257 map = bpf_map_from_imm(prog, imm, &off, &type);
2259 insns[i].src_reg = type;
2260 insns[i].imm = map->id;
2261 insns[i + 1].imm = off;
2269 static int set_info_rec_size(struct bpf_prog_info *info)
2272 * Ensure info.*_rec_size is the same as kernel expected size
2276 * Only allow zero *_rec_size if both _rec_size and _cnt are
2277 * zero. In this case, the kernel will set the expected
2278 * _rec_size back to the info.
2281 if ((info->nr_func_info || info->func_info_rec_size) &&
2282 info->func_info_rec_size != sizeof(struct bpf_func_info))
2285 if ((info->nr_line_info || info->line_info_rec_size) &&
2286 info->line_info_rec_size != sizeof(struct bpf_line_info))
2289 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2290 info->jited_line_info_rec_size != sizeof(__u64))
2293 info->func_info_rec_size = sizeof(struct bpf_func_info);
2294 info->line_info_rec_size = sizeof(struct bpf_line_info);
2295 info->jited_line_info_rec_size = sizeof(__u64);
2300 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2301 const union bpf_attr *attr,
2302 union bpf_attr __user *uattr)
2304 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2305 struct bpf_prog_info info = {};
2306 u32 info_len = attr->info.info_len;
2307 struct bpf_prog_stats stats;
2308 char __user *uinsns;
2312 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2315 info_len = min_t(u32, sizeof(info), info_len);
2317 if (copy_from_user(&info, uinfo, info_len))
2320 info.type = prog->type;
2321 info.id = prog->aux->id;
2322 info.load_time = prog->aux->load_time;
2323 info.created_by_uid = from_kuid_munged(current_user_ns(),
2324 prog->aux->user->uid);
2325 info.gpl_compatible = prog->gpl_compatible;
2327 memcpy(info.tag, prog->tag, sizeof(prog->tag));
2328 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2330 ulen = info.nr_map_ids;
2331 info.nr_map_ids = prog->aux->used_map_cnt;
2332 ulen = min_t(u32, info.nr_map_ids, ulen);
2334 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2337 for (i = 0; i < ulen; i++)
2338 if (put_user(prog->aux->used_maps[i]->id,
2343 err = set_info_rec_size(&info);
2347 bpf_prog_get_stats(prog, &stats);
2348 info.run_time_ns = stats.nsecs;
2349 info.run_cnt = stats.cnt;
2351 if (!capable(CAP_SYS_ADMIN)) {
2352 info.jited_prog_len = 0;
2353 info.xlated_prog_len = 0;
2354 info.nr_jited_ksyms = 0;
2355 info.nr_jited_func_lens = 0;
2356 info.nr_func_info = 0;
2357 info.nr_line_info = 0;
2358 info.nr_jited_line_info = 0;
2362 ulen = info.xlated_prog_len;
2363 info.xlated_prog_len = bpf_prog_insn_size(prog);
2364 if (info.xlated_prog_len && ulen) {
2365 struct bpf_insn *insns_sanitized;
2368 if (prog->blinded && !bpf_dump_raw_ok()) {
2369 info.xlated_prog_insns = 0;
2372 insns_sanitized = bpf_insn_prepare_dump(prog);
2373 if (!insns_sanitized)
2375 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2376 ulen = min_t(u32, info.xlated_prog_len, ulen);
2377 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2378 kfree(insns_sanitized);
2383 if (bpf_prog_is_dev_bound(prog->aux)) {
2384 err = bpf_prog_offload_info_fill(&info, prog);
2390 /* NOTE: the following code is supposed to be skipped for offload.
2391 * bpf_prog_offload_info_fill() is the place to fill similar fields
2394 ulen = info.jited_prog_len;
2395 if (prog->aux->func_cnt) {
2398 info.jited_prog_len = 0;
2399 for (i = 0; i < prog->aux->func_cnt; i++)
2400 info.jited_prog_len += prog->aux->func[i]->jited_len;
2402 info.jited_prog_len = prog->jited_len;
2405 if (info.jited_prog_len && ulen) {
2406 if (bpf_dump_raw_ok()) {
2407 uinsns = u64_to_user_ptr(info.jited_prog_insns);
2408 ulen = min_t(u32, info.jited_prog_len, ulen);
2410 /* for multi-function programs, copy the JITed
2411 * instructions for all the functions
2413 if (prog->aux->func_cnt) {
2418 for (i = 0; i < prog->aux->func_cnt; i++) {
2419 len = prog->aux->func[i]->jited_len;
2420 len = min_t(u32, len, free);
2421 img = (u8 *) prog->aux->func[i]->bpf_func;
2422 if (copy_to_user(uinsns, img, len))
2430 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2434 info.jited_prog_insns = 0;
2438 ulen = info.nr_jited_ksyms;
2439 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2441 if (bpf_dump_raw_ok()) {
2442 unsigned long ksym_addr;
2443 u64 __user *user_ksyms;
2446 /* copy the address of the kernel symbol
2447 * corresponding to each function
2449 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2450 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2451 if (prog->aux->func_cnt) {
2452 for (i = 0; i < ulen; i++) {
2453 ksym_addr = (unsigned long)
2454 prog->aux->func[i]->bpf_func;
2455 if (put_user((u64) ksym_addr,
2460 ksym_addr = (unsigned long) prog->bpf_func;
2461 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2465 info.jited_ksyms = 0;
2469 ulen = info.nr_jited_func_lens;
2470 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2472 if (bpf_dump_raw_ok()) {
2473 u32 __user *user_lens;
2476 /* copy the JITed image lengths for each function */
2477 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2478 user_lens = u64_to_user_ptr(info.jited_func_lens);
2479 if (prog->aux->func_cnt) {
2480 for (i = 0; i < ulen; i++) {
2482 prog->aux->func[i]->jited_len;
2483 if (put_user(func_len, &user_lens[i]))
2487 func_len = prog->jited_len;
2488 if (put_user(func_len, &user_lens[0]))
2492 info.jited_func_lens = 0;
2497 info.btf_id = btf_id(prog->aux->btf);
2499 ulen = info.nr_func_info;
2500 info.nr_func_info = prog->aux->func_info_cnt;
2501 if (info.nr_func_info && ulen) {
2502 char __user *user_finfo;
2504 user_finfo = u64_to_user_ptr(info.func_info);
2505 ulen = min_t(u32, info.nr_func_info, ulen);
2506 if (copy_to_user(user_finfo, prog->aux->func_info,
2507 info.func_info_rec_size * ulen))
2511 ulen = info.nr_line_info;
2512 info.nr_line_info = prog->aux->nr_linfo;
2513 if (info.nr_line_info && ulen) {
2514 __u8 __user *user_linfo;
2516 user_linfo = u64_to_user_ptr(info.line_info);
2517 ulen = min_t(u32, info.nr_line_info, ulen);
2518 if (copy_to_user(user_linfo, prog->aux->linfo,
2519 info.line_info_rec_size * ulen))
2523 ulen = info.nr_jited_line_info;
2524 if (prog->aux->jited_linfo)
2525 info.nr_jited_line_info = prog->aux->nr_linfo;
2527 info.nr_jited_line_info = 0;
2528 if (info.nr_jited_line_info && ulen) {
2529 if (bpf_dump_raw_ok()) {
2530 __u64 __user *user_linfo;
2533 user_linfo = u64_to_user_ptr(info.jited_line_info);
2534 ulen = min_t(u32, info.nr_jited_line_info, ulen);
2535 for (i = 0; i < ulen; i++) {
2536 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2541 info.jited_line_info = 0;
2545 ulen = info.nr_prog_tags;
2546 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2548 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2551 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2552 ulen = min_t(u32, info.nr_prog_tags, ulen);
2553 if (prog->aux->func_cnt) {
2554 for (i = 0; i < ulen; i++) {
2555 if (copy_to_user(user_prog_tags[i],
2556 prog->aux->func[i]->tag,
2561 if (copy_to_user(user_prog_tags[0],
2562 prog->tag, BPF_TAG_SIZE))
2568 if (copy_to_user(uinfo, &info, info_len) ||
2569 put_user(info_len, &uattr->info.info_len))
2575 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2576 const union bpf_attr *attr,
2577 union bpf_attr __user *uattr)
2579 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2580 struct bpf_map_info info = {};
2581 u32 info_len = attr->info.info_len;
2584 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2587 info_len = min_t(u32, sizeof(info), info_len);
2589 info.type = map->map_type;
2591 info.key_size = map->key_size;
2592 info.value_size = map->value_size;
2593 info.max_entries = map->max_entries;
2594 info.map_flags = map->map_flags;
2595 memcpy(info.name, map->name, sizeof(map->name));
2598 info.btf_id = btf_id(map->btf);
2599 info.btf_key_type_id = map->btf_key_type_id;
2600 info.btf_value_type_id = map->btf_value_type_id;
2603 if (bpf_map_is_dev_bound(map)) {
2604 err = bpf_map_offload_info_fill(&info, map);
2609 if (copy_to_user(uinfo, &info, info_len) ||
2610 put_user(info_len, &uattr->info.info_len))
2616 static int bpf_btf_get_info_by_fd(struct btf *btf,
2617 const union bpf_attr *attr,
2618 union bpf_attr __user *uattr)
2620 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2621 u32 info_len = attr->info.info_len;
2624 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2628 return btf_get_info_by_fd(btf, attr, uattr);
2631 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2633 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2634 union bpf_attr __user *uattr)
2636 int ufd = attr->info.bpf_fd;
2640 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2647 if (f.file->f_op == &bpf_prog_fops)
2648 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2650 else if (f.file->f_op == &bpf_map_fops)
2651 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2653 else if (f.file->f_op == &btf_fops)
2654 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2662 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2664 static int bpf_btf_load(const union bpf_attr *attr)
2666 if (CHECK_ATTR(BPF_BTF_LOAD))
2669 if (!capable(CAP_SYS_ADMIN))
2672 return btf_new_fd(attr);
2675 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2677 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2679 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2682 if (!capable(CAP_SYS_ADMIN))
2685 return btf_get_fd_by_id(attr->btf_id);
2688 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2689 union bpf_attr __user *uattr,
2690 u32 prog_id, u32 fd_type,
2691 const char *buf, u64 probe_offset,
2694 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2695 u32 len = buf ? strlen(buf) : 0, input_len;
2698 if (put_user(len, &uattr->task_fd_query.buf_len))
2700 input_len = attr->task_fd_query.buf_len;
2701 if (input_len && ubuf) {
2703 /* nothing to copy, just make ubuf NULL terminated */
2706 if (put_user(zero, ubuf))
2708 } else if (input_len >= len + 1) {
2709 /* ubuf can hold the string with NULL terminator */
2710 if (copy_to_user(ubuf, buf, len + 1))
2713 /* ubuf cannot hold the string with NULL terminator,
2714 * do a partial copy with NULL terminator.
2719 if (copy_to_user(ubuf, buf, input_len - 1))
2721 if (put_user(zero, ubuf + input_len - 1))
2726 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2727 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2728 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2729 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2735 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2737 static int bpf_task_fd_query(const union bpf_attr *attr,
2738 union bpf_attr __user *uattr)
2740 pid_t pid = attr->task_fd_query.pid;
2741 u32 fd = attr->task_fd_query.fd;
2742 const struct perf_event *event;
2743 struct files_struct *files;
2744 struct task_struct *task;
2748 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2751 if (!capable(CAP_SYS_ADMIN))
2754 if (attr->task_fd_query.flags != 0)
2757 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2761 files = get_files_struct(task);
2762 put_task_struct(task);
2767 spin_lock(&files->file_lock);
2768 file = fcheck_files(files, fd);
2773 spin_unlock(&files->file_lock);
2774 put_files_struct(files);
2779 if (file->f_op == &bpf_raw_tp_fops) {
2780 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2781 struct bpf_raw_event_map *btp = raw_tp->btp;
2783 err = bpf_task_fd_query_copy(attr, uattr,
2784 raw_tp->prog->aux->id,
2785 BPF_FD_TYPE_RAW_TRACEPOINT,
2786 btp->tp->name, 0, 0);
2790 event = perf_get_event(file);
2791 if (!IS_ERR(event)) {
2792 u64 probe_offset, probe_addr;
2793 u32 prog_id, fd_type;
2796 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2797 &buf, &probe_offset,
2800 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2814 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2816 union bpf_attr attr = {};
2819 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2822 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2825 size = min_t(u32, size, sizeof(attr));
2827 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2828 if (copy_from_user(&attr, uattr, size) != 0)
2831 err = security_bpf(cmd, &attr, size);
2836 case BPF_MAP_CREATE:
2837 err = map_create(&attr);
2839 case BPF_MAP_LOOKUP_ELEM:
2840 err = map_lookup_elem(&attr);
2842 case BPF_MAP_UPDATE_ELEM:
2843 err = map_update_elem(&attr);
2845 case BPF_MAP_DELETE_ELEM:
2846 err = map_delete_elem(&attr);
2848 case BPF_MAP_GET_NEXT_KEY:
2849 err = map_get_next_key(&attr);
2851 case BPF_MAP_FREEZE:
2852 err = map_freeze(&attr);
2855 err = bpf_prog_load(&attr, uattr);
2858 err = bpf_obj_pin(&attr);
2861 err = bpf_obj_get(&attr);
2863 case BPF_PROG_ATTACH:
2864 err = bpf_prog_attach(&attr);
2866 case BPF_PROG_DETACH:
2867 err = bpf_prog_detach(&attr);
2869 case BPF_PROG_QUERY:
2870 err = bpf_prog_query(&attr, uattr);
2872 case BPF_PROG_TEST_RUN:
2873 err = bpf_prog_test_run(&attr, uattr);
2875 case BPF_PROG_GET_NEXT_ID:
2876 err = bpf_obj_get_next_id(&attr, uattr,
2877 &prog_idr, &prog_idr_lock);
2879 case BPF_MAP_GET_NEXT_ID:
2880 err = bpf_obj_get_next_id(&attr, uattr,
2881 &map_idr, &map_idr_lock);
2883 case BPF_PROG_GET_FD_BY_ID:
2884 err = bpf_prog_get_fd_by_id(&attr);
2886 case BPF_MAP_GET_FD_BY_ID:
2887 err = bpf_map_get_fd_by_id(&attr);
2889 case BPF_OBJ_GET_INFO_BY_FD:
2890 err = bpf_obj_get_info_by_fd(&attr, uattr);
2892 case BPF_RAW_TRACEPOINT_OPEN:
2893 err = bpf_raw_tracepoint_open(&attr);
2896 err = bpf_btf_load(&attr);
2898 case BPF_BTF_GET_FD_BY_ID:
2899 err = bpf_btf_get_fd_by_id(&attr);
2901 case BPF_TASK_FD_QUERY:
2902 err = bpf_task_fd_query(&attr, uattr);
2904 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2905 err = map_lookup_and_delete_elem(&attr);