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>
26 #include <uapi/linux/btf.h>
28 #define IS_FD_ARRAY(map) ((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_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
32 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
33 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
36 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
38 DEFINE_PER_CPU(int, bpf_prog_active);
39 static DEFINE_IDR(prog_idr);
40 static DEFINE_SPINLOCK(prog_idr_lock);
41 static DEFINE_IDR(map_idr);
42 static DEFINE_SPINLOCK(map_idr_lock);
44 int sysctl_unprivileged_bpf_disabled __read_mostly;
46 static const struct bpf_map_ops * const bpf_map_types[] = {
47 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
48 #define BPF_MAP_TYPE(_id, _ops) \
50 #include <linux/bpf_types.h>
56 * If we're handed a bigger struct than we know of, ensure all the unknown bits
57 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
58 * we don't know about yet.
60 * There is a ToCToU between this function call and the following
61 * copy_from_user() call. However, this is not a concern since this function is
62 * meant to be a future-proofing of bits.
64 int bpf_check_uarg_tail_zero(void __user *uaddr,
68 unsigned char __user *addr;
69 unsigned char __user *end;
73 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
76 if (unlikely(!access_ok(uaddr, actual_size)))
79 if (actual_size <= expected_size)
82 addr = uaddr + expected_size;
83 end = uaddr + actual_size;
85 for (; addr < end; addr++) {
86 err = get_user(val, addr);
96 const struct bpf_map_ops bpf_map_offload_ops = {
97 .map_alloc = bpf_map_offload_map_alloc,
98 .map_free = bpf_map_offload_map_free,
99 .map_check_btf = map_check_no_btf,
102 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
104 const struct bpf_map_ops *ops;
105 u32 type = attr->map_type;
109 if (type >= ARRAY_SIZE(bpf_map_types))
110 return ERR_PTR(-EINVAL);
111 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
112 ops = bpf_map_types[type];
114 return ERR_PTR(-EINVAL);
116 if (ops->map_alloc_check) {
117 err = ops->map_alloc_check(attr);
121 if (attr->map_ifindex)
122 ops = &bpf_map_offload_ops;
123 map = ops->map_alloc(attr);
127 map->map_type = type;
131 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
133 /* We really just want to fail instead of triggering OOM killer
134 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
135 * which is used for lower order allocation requests.
137 * It has been observed that higher order allocation requests done by
138 * vmalloc with __GFP_NORETRY being set might fail due to not trying
139 * to reclaim memory from the page cache, thus we set
140 * __GFP_RETRY_MAYFAIL to avoid such situations.
143 const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
146 if (size >= SIZE_MAX)
149 /* kmalloc()'ed memory can't be mmap()'ed */
150 if (!mmapable && size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
151 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
157 BUG_ON(!PAGE_ALIGNED(size));
158 return vmalloc_user_node_flags(size, numa_node, GFP_KERNEL |
159 __GFP_RETRY_MAYFAIL | flags);
161 return __vmalloc_node_flags_caller(size, numa_node,
162 GFP_KERNEL | __GFP_RETRY_MAYFAIL |
163 flags, __builtin_return_address(0));
166 void *bpf_map_area_alloc(u64 size, int numa_node)
168 return __bpf_map_area_alloc(size, numa_node, false);
171 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
173 return __bpf_map_area_alloc(size, numa_node, true);
176 void bpf_map_area_free(void *area)
181 static u32 bpf_map_flags_retain_permanent(u32 flags)
183 /* Some map creation flags are not tied to the map object but
184 * rather to the map fd instead, so they have no meaning upon
185 * map object inspection since multiple file descriptors with
186 * different (access) properties can exist here. Thus, given
187 * this has zero meaning for the map itself, lets clear these
190 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
193 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
195 map->map_type = attr->map_type;
196 map->key_size = attr->key_size;
197 map->value_size = attr->value_size;
198 map->max_entries = attr->max_entries;
199 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
200 map->numa_node = bpf_map_attr_numa_node(attr);
203 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
205 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
207 if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
208 atomic_long_sub(pages, &user->locked_vm);
214 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
217 atomic_long_sub(pages, &user->locked_vm);
220 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size)
222 u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
223 struct user_struct *user;
226 if (size >= U32_MAX - PAGE_SIZE)
229 user = get_current_user();
230 ret = bpf_charge_memlock(user, pages);
242 void bpf_map_charge_finish(struct bpf_map_memory *mem)
244 bpf_uncharge_memlock(mem->user, mem->pages);
248 void bpf_map_charge_move(struct bpf_map_memory *dst,
249 struct bpf_map_memory *src)
253 /* Make sure src will not be used for the redundant uncharging. */
254 memset(src, 0, sizeof(struct bpf_map_memory));
257 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
261 ret = bpf_charge_memlock(map->memory.user, pages);
264 map->memory.pages += pages;
268 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
270 bpf_uncharge_memlock(map->memory.user, pages);
271 map->memory.pages -= pages;
274 static int bpf_map_alloc_id(struct bpf_map *map)
278 idr_preload(GFP_KERNEL);
279 spin_lock_bh(&map_idr_lock);
280 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
283 spin_unlock_bh(&map_idr_lock);
286 if (WARN_ON_ONCE(!id))
289 return id > 0 ? 0 : id;
292 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
296 /* Offloaded maps are removed from the IDR store when their device
297 * disappears - even if someone holds an fd to them they are unusable,
298 * the memory is gone, all ops will fail; they are simply waiting for
299 * refcnt to drop to be freed.
305 spin_lock_irqsave(&map_idr_lock, flags);
307 __acquire(&map_idr_lock);
309 idr_remove(&map_idr, map->id);
313 spin_unlock_irqrestore(&map_idr_lock, flags);
315 __release(&map_idr_lock);
318 /* called from workqueue */
319 static void bpf_map_free_deferred(struct work_struct *work)
321 struct bpf_map *map = container_of(work, struct bpf_map, work);
322 struct bpf_map_memory mem;
324 bpf_map_charge_move(&mem, &map->memory);
325 security_bpf_map_free(map);
326 /* implementation dependent freeing */
327 map->ops->map_free(map);
328 bpf_map_charge_finish(&mem);
331 static void bpf_map_put_uref(struct bpf_map *map)
333 if (atomic64_dec_and_test(&map->usercnt)) {
334 if (map->ops->map_release_uref)
335 map->ops->map_release_uref(map);
339 /* decrement map refcnt and schedule it for freeing via workqueue
340 * (unrelying map implementation ops->map_free() might sleep)
342 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
344 if (atomic64_dec_and_test(&map->refcnt)) {
345 /* bpf_map_free_id() must be called first */
346 bpf_map_free_id(map, do_idr_lock);
348 INIT_WORK(&map->work, bpf_map_free_deferred);
349 schedule_work(&map->work);
353 void bpf_map_put(struct bpf_map *map)
355 __bpf_map_put(map, true);
357 EXPORT_SYMBOL_GPL(bpf_map_put);
359 void bpf_map_put_with_uref(struct bpf_map *map)
361 bpf_map_put_uref(map);
365 static int bpf_map_release(struct inode *inode, struct file *filp)
367 struct bpf_map *map = filp->private_data;
369 if (map->ops->map_release)
370 map->ops->map_release(map, filp);
372 bpf_map_put_with_uref(map);
376 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
378 fmode_t mode = f.file->f_mode;
380 /* Our file permissions may have been overridden by global
381 * map permissions facing syscall side.
383 if (READ_ONCE(map->frozen))
384 mode &= ~FMODE_CAN_WRITE;
388 #ifdef CONFIG_PROC_FS
389 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
391 const struct bpf_map *map = filp->private_data;
392 const struct bpf_array *array;
393 u32 type = 0, jited = 0;
395 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
396 array = container_of(map, struct bpf_array, map);
397 type = array->aux->type;
398 jited = array->aux->jited;
415 map->memory.pages * 1ULL << PAGE_SHIFT,
417 READ_ONCE(map->frozen));
419 seq_printf(m, "owner_prog_type:\t%u\n", type);
420 seq_printf(m, "owner_jited:\t%u\n", jited);
425 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
428 /* We need this handler such that alloc_file() enables
429 * f_mode with FMODE_CAN_READ.
434 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
435 size_t siz, loff_t *ppos)
437 /* We need this handler such that alloc_file() enables
438 * f_mode with FMODE_CAN_WRITE.
443 /* called for any extra memory-mapped regions (except initial) */
444 static void bpf_map_mmap_open(struct vm_area_struct *vma)
446 struct bpf_map *map = vma->vm_file->private_data;
448 bpf_map_inc_with_uref(map);
450 if (vma->vm_flags & VM_WRITE) {
451 mutex_lock(&map->freeze_mutex);
453 mutex_unlock(&map->freeze_mutex);
457 /* called for all unmapped memory region (including initial) */
458 static void bpf_map_mmap_close(struct vm_area_struct *vma)
460 struct bpf_map *map = vma->vm_file->private_data;
462 if (vma->vm_flags & VM_WRITE) {
463 mutex_lock(&map->freeze_mutex);
465 mutex_unlock(&map->freeze_mutex);
468 bpf_map_put_with_uref(map);
471 static const struct vm_operations_struct bpf_map_default_vmops = {
472 .open = bpf_map_mmap_open,
473 .close = bpf_map_mmap_close,
476 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
478 struct bpf_map *map = filp->private_data;
481 if (!map->ops->map_mmap || map_value_has_spin_lock(map))
484 if (!(vma->vm_flags & VM_SHARED))
487 mutex_lock(&map->freeze_mutex);
489 if ((vma->vm_flags & VM_WRITE) && map->frozen) {
494 /* set default open/close callbacks */
495 vma->vm_ops = &bpf_map_default_vmops;
496 vma->vm_private_data = map;
498 err = map->ops->map_mmap(map, vma);
502 bpf_map_inc_with_uref(map);
504 if (vma->vm_flags & VM_WRITE)
507 mutex_unlock(&map->freeze_mutex);
511 const struct file_operations bpf_map_fops = {
512 #ifdef CONFIG_PROC_FS
513 .show_fdinfo = bpf_map_show_fdinfo,
515 .release = bpf_map_release,
516 .read = bpf_dummy_read,
517 .write = bpf_dummy_write,
518 .mmap = bpf_map_mmap,
521 int bpf_map_new_fd(struct bpf_map *map, int flags)
525 ret = security_bpf_map(map, OPEN_FMODE(flags));
529 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
533 int bpf_get_file_flag(int flags)
535 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
537 if (flags & BPF_F_RDONLY)
539 if (flags & BPF_F_WRONLY)
544 /* helper macro to check that unused fields 'union bpf_attr' are zero */
545 #define CHECK_ATTR(CMD) \
546 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
547 sizeof(attr->CMD##_LAST_FIELD), 0, \
549 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
550 sizeof(attr->CMD##_LAST_FIELD)) != NULL
552 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
553 * Return 0 on success and < 0 on error.
555 static int bpf_obj_name_cpy(char *dst, const char *src)
557 const char *end = src + BPF_OBJ_NAME_LEN;
559 memset(dst, 0, BPF_OBJ_NAME_LEN);
560 /* Copy all isalnum(), '_' and '.' chars. */
561 while (src < end && *src) {
562 if (!isalnum(*src) &&
563 *src != '_' && *src != '.')
568 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
575 int map_check_no_btf(const struct bpf_map *map,
576 const struct btf *btf,
577 const struct btf_type *key_type,
578 const struct btf_type *value_type)
583 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
584 u32 btf_key_id, u32 btf_value_id)
586 const struct btf_type *key_type, *value_type;
587 u32 key_size, value_size;
590 /* Some maps allow key to be unspecified. */
592 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
593 if (!key_type || key_size != map->key_size)
596 key_type = btf_type_by_id(btf, 0);
597 if (!map->ops->map_check_btf)
601 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
602 if (!value_type || value_size != map->value_size)
605 map->spin_lock_off = btf_find_spin_lock(btf, value_type);
607 if (map_value_has_spin_lock(map)) {
608 if (map->map_flags & BPF_F_RDONLY_PROG)
610 if (map->map_type != BPF_MAP_TYPE_HASH &&
611 map->map_type != BPF_MAP_TYPE_ARRAY &&
612 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
613 map->map_type != BPF_MAP_TYPE_SK_STORAGE)
615 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
618 "verifier bug spin_lock_off %d value_size %d\n",
619 map->spin_lock_off, map->value_size);
624 if (map->ops->map_check_btf)
625 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
630 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
631 /* called via syscall */
632 static int map_create(union bpf_attr *attr)
634 int numa_node = bpf_map_attr_numa_node(attr);
635 struct bpf_map_memory mem;
640 err = CHECK_ATTR(BPF_MAP_CREATE);
644 f_flags = bpf_get_file_flag(attr->map_flags);
648 if (numa_node != NUMA_NO_NODE &&
649 ((unsigned int)numa_node >= nr_node_ids ||
650 !node_online(numa_node)))
653 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
654 map = find_and_alloc_map(attr);
658 err = bpf_obj_name_cpy(map->name, attr->map_name);
662 atomic64_set(&map->refcnt, 1);
663 atomic64_set(&map->usercnt, 1);
664 mutex_init(&map->freeze_mutex);
666 if (attr->btf_key_type_id || attr->btf_value_type_id) {
669 if (!attr->btf_value_type_id) {
674 btf = btf_get_by_fd(attr->btf_fd);
680 err = map_check_btf(map, btf, attr->btf_key_type_id,
681 attr->btf_value_type_id);
688 map->btf_key_type_id = attr->btf_key_type_id;
689 map->btf_value_type_id = attr->btf_value_type_id;
691 map->spin_lock_off = -EINVAL;
694 err = security_bpf_map_alloc(map);
698 err = bpf_map_alloc_id(map);
702 err = bpf_map_new_fd(map, f_flags);
704 /* failed to allocate fd.
705 * bpf_map_put_with_uref() is needed because the above
706 * bpf_map_alloc_id() has published the map
707 * to the userspace and the userspace may
708 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
710 bpf_map_put_with_uref(map);
717 security_bpf_map_free(map);
720 bpf_map_charge_move(&mem, &map->memory);
721 map->ops->map_free(map);
722 bpf_map_charge_finish(&mem);
726 /* if error is returned, fd is released.
727 * On success caller should complete fd access with matching fdput()
729 struct bpf_map *__bpf_map_get(struct fd f)
732 return ERR_PTR(-EBADF);
733 if (f.file->f_op != &bpf_map_fops) {
735 return ERR_PTR(-EINVAL);
738 return f.file->private_data;
741 void bpf_map_inc(struct bpf_map *map)
743 atomic64_inc(&map->refcnt);
745 EXPORT_SYMBOL_GPL(bpf_map_inc);
747 void bpf_map_inc_with_uref(struct bpf_map *map)
749 atomic64_inc(&map->refcnt);
750 atomic64_inc(&map->usercnt);
752 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
754 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
756 struct fd f = fdget(ufd);
759 map = __bpf_map_get(f);
763 bpf_map_inc_with_uref(map);
769 /* map_idr_lock should have been held */
770 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
774 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
776 return ERR_PTR(-ENOENT);
778 atomic64_inc(&map->usercnt);
783 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
785 spin_lock_bh(&map_idr_lock);
786 map = __bpf_map_inc_not_zero(map, false);
787 spin_unlock_bh(&map_idr_lock);
791 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
793 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
798 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
801 return memdup_user(ukey, key_size);
804 return ERR_PTR(-EINVAL);
809 /* last field in 'union bpf_attr' used by this command */
810 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
812 static int map_lookup_elem(union bpf_attr *attr)
814 void __user *ukey = u64_to_user_ptr(attr->key);
815 void __user *uvalue = u64_to_user_ptr(attr->value);
816 int ufd = attr->map_fd;
818 void *key, *value, *ptr;
823 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
826 if (attr->flags & ~BPF_F_LOCK)
830 map = __bpf_map_get(f);
833 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
838 if ((attr->flags & BPF_F_LOCK) &&
839 !map_value_has_spin_lock(map)) {
844 key = __bpf_copy_key(ukey, map->key_size);
850 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
851 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
852 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
853 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
854 value_size = round_up(map->value_size, 8) * num_possible_cpus();
855 else if (IS_FD_MAP(map))
856 value_size = sizeof(u32);
858 value_size = map->value_size;
861 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
865 if (bpf_map_is_dev_bound(map)) {
866 err = bpf_map_offload_lookup_elem(map, key, value);
871 this_cpu_inc(bpf_prog_active);
872 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
873 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
874 err = bpf_percpu_hash_copy(map, key, value);
875 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
876 err = bpf_percpu_array_copy(map, key, value);
877 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
878 err = bpf_percpu_cgroup_storage_copy(map, key, value);
879 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
880 err = bpf_stackmap_copy(map, key, value);
881 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
882 err = bpf_fd_array_map_lookup_elem(map, key, value);
883 } else if (IS_FD_HASH(map)) {
884 err = bpf_fd_htab_map_lookup_elem(map, key, value);
885 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
886 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
887 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
888 map->map_type == BPF_MAP_TYPE_STACK) {
889 err = map->ops->map_peek_elem(map, value);
892 if (map->ops->map_lookup_elem_sys_only)
893 ptr = map->ops->map_lookup_elem_sys_only(map, key);
895 ptr = map->ops->map_lookup_elem(map, key);
902 if (attr->flags & BPF_F_LOCK)
903 /* lock 'ptr' and copy everything but lock */
904 copy_map_value_locked(map, value, ptr, true);
906 copy_map_value(map, value, ptr);
907 /* mask lock, since value wasn't zero inited */
908 check_and_init_map_lock(map, value);
912 this_cpu_dec(bpf_prog_active);
920 if (copy_to_user(uvalue, value, value_size) != 0)
934 static void maybe_wait_bpf_programs(struct bpf_map *map)
936 /* Wait for any running BPF programs to complete so that
937 * userspace, when we return to it, knows that all programs
938 * that could be running use the new map value.
940 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
941 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
945 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
947 static int map_update_elem(union bpf_attr *attr)
949 void __user *ukey = u64_to_user_ptr(attr->key);
950 void __user *uvalue = u64_to_user_ptr(attr->value);
951 int ufd = attr->map_fd;
958 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
962 map = __bpf_map_get(f);
965 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
970 if ((attr->flags & BPF_F_LOCK) &&
971 !map_value_has_spin_lock(map)) {
976 key = __bpf_copy_key(ukey, map->key_size);
982 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
983 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
984 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
985 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
986 value_size = round_up(map->value_size, 8) * num_possible_cpus();
988 value_size = map->value_size;
991 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
996 if (copy_from_user(value, uvalue, value_size) != 0)
999 /* Need to create a kthread, thus must support schedule */
1000 if (bpf_map_is_dev_bound(map)) {
1001 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
1003 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
1004 map->map_type == BPF_MAP_TYPE_SOCKHASH ||
1005 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
1006 err = map->ops->map_update_elem(map, key, value, attr->flags);
1008 } else if (IS_FD_PROG_ARRAY(map)) {
1009 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
1014 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
1015 * inside bpf map update or delete otherwise deadlocks are possible
1018 __this_cpu_inc(bpf_prog_active);
1019 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1020 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1021 err = bpf_percpu_hash_update(map, key, value, attr->flags);
1022 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
1023 err = bpf_percpu_array_update(map, key, value, attr->flags);
1024 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
1025 err = bpf_percpu_cgroup_storage_update(map, key, value,
1027 } else if (IS_FD_ARRAY(map)) {
1029 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
1032 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
1034 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
1037 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
1038 /* rcu_read_lock() is not needed */
1039 err = bpf_fd_reuseport_array_update_elem(map, key, value,
1041 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1042 map->map_type == BPF_MAP_TYPE_STACK) {
1043 err = map->ops->map_push_elem(map, value, attr->flags);
1046 err = map->ops->map_update_elem(map, key, value, attr->flags);
1049 __this_cpu_dec(bpf_prog_active);
1051 maybe_wait_bpf_programs(map);
1062 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1064 static int map_delete_elem(union bpf_attr *attr)
1066 void __user *ukey = u64_to_user_ptr(attr->key);
1067 int ufd = attr->map_fd;
1068 struct bpf_map *map;
1073 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1077 map = __bpf_map_get(f);
1079 return PTR_ERR(map);
1080 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1085 key = __bpf_copy_key(ukey, map->key_size);
1091 if (bpf_map_is_dev_bound(map)) {
1092 err = bpf_map_offload_delete_elem(map, key);
1094 } else if (IS_FD_PROG_ARRAY(map)) {
1095 err = map->ops->map_delete_elem(map, key);
1100 __this_cpu_inc(bpf_prog_active);
1102 err = map->ops->map_delete_elem(map, key);
1104 __this_cpu_dec(bpf_prog_active);
1106 maybe_wait_bpf_programs(map);
1114 /* last field in 'union bpf_attr' used by this command */
1115 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1117 static int map_get_next_key(union bpf_attr *attr)
1119 void __user *ukey = u64_to_user_ptr(attr->key);
1120 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1121 int ufd = attr->map_fd;
1122 struct bpf_map *map;
1123 void *key, *next_key;
1127 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1131 map = __bpf_map_get(f);
1133 return PTR_ERR(map);
1134 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1140 key = __bpf_copy_key(ukey, map->key_size);
1150 next_key = kmalloc(map->key_size, GFP_USER);
1154 if (bpf_map_is_dev_bound(map)) {
1155 err = bpf_map_offload_get_next_key(map, key, next_key);
1160 err = map->ops->map_get_next_key(map, key, next_key);
1167 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1181 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1183 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1185 void __user *ukey = u64_to_user_ptr(attr->key);
1186 void __user *uvalue = u64_to_user_ptr(attr->value);
1187 int ufd = attr->map_fd;
1188 struct bpf_map *map;
1194 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1198 map = __bpf_map_get(f);
1200 return PTR_ERR(map);
1201 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1206 key = __bpf_copy_key(ukey, map->key_size);
1212 value_size = map->value_size;
1215 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1219 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1220 map->map_type == BPF_MAP_TYPE_STACK) {
1221 err = map->ops->map_pop_elem(map, value);
1229 if (copy_to_user(uvalue, value, value_size) != 0)
1243 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1245 static int map_freeze(const union bpf_attr *attr)
1247 int err = 0, ufd = attr->map_fd;
1248 struct bpf_map *map;
1251 if (CHECK_ATTR(BPF_MAP_FREEZE))
1255 map = __bpf_map_get(f);
1257 return PTR_ERR(map);
1259 mutex_lock(&map->freeze_mutex);
1261 if (map->writecnt) {
1265 if (READ_ONCE(map->frozen)) {
1269 if (!capable(CAP_SYS_ADMIN)) {
1274 WRITE_ONCE(map->frozen, true);
1276 mutex_unlock(&map->freeze_mutex);
1281 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1282 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1283 [_id] = & _name ## _prog_ops,
1284 #define BPF_MAP_TYPE(_id, _ops)
1285 #include <linux/bpf_types.h>
1286 #undef BPF_PROG_TYPE
1290 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1292 const struct bpf_prog_ops *ops;
1294 if (type >= ARRAY_SIZE(bpf_prog_types))
1296 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1297 ops = bpf_prog_types[type];
1301 if (!bpf_prog_is_dev_bound(prog->aux))
1302 prog->aux->ops = ops;
1304 prog->aux->ops = &bpf_offload_prog_ops;
1309 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1311 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1312 unsigned long user_bufs;
1315 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1316 if (user_bufs > memlock_limit) {
1317 atomic_long_sub(pages, &user->locked_vm);
1325 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1328 atomic_long_sub(pages, &user->locked_vm);
1331 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1333 struct user_struct *user = get_current_user();
1336 ret = __bpf_prog_charge(user, prog->pages);
1342 prog->aux->user = user;
1346 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1348 struct user_struct *user = prog->aux->user;
1350 __bpf_prog_uncharge(user, prog->pages);
1354 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1358 idr_preload(GFP_KERNEL);
1359 spin_lock_bh(&prog_idr_lock);
1360 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1363 spin_unlock_bh(&prog_idr_lock);
1366 /* id is in [1, INT_MAX) */
1367 if (WARN_ON_ONCE(!id))
1370 return id > 0 ? 0 : id;
1373 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1375 /* cBPF to eBPF migrations are currently not in the idr store.
1376 * Offloaded programs are removed from the store when their device
1377 * disappears - even if someone grabs an fd to them they are unusable,
1378 * simply waiting for refcnt to drop to be freed.
1384 spin_lock_bh(&prog_idr_lock);
1386 __acquire(&prog_idr_lock);
1388 idr_remove(&prog_idr, prog->aux->id);
1392 spin_unlock_bh(&prog_idr_lock);
1394 __release(&prog_idr_lock);
1397 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1399 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1401 kvfree(aux->func_info);
1402 kfree(aux->func_info_aux);
1403 bpf_prog_uncharge_memlock(aux->prog);
1404 security_bpf_prog_free(aux);
1405 bpf_prog_free(aux->prog);
1408 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1410 bpf_prog_kallsyms_del_all(prog);
1411 btf_put(prog->aux->btf);
1412 bpf_prog_free_linfo(prog);
1415 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1417 __bpf_prog_put_rcu(&prog->aux->rcu);
1420 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1422 if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1423 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1424 /* bpf_prog_free_id() must be called first */
1425 bpf_prog_free_id(prog, do_idr_lock);
1426 __bpf_prog_put_noref(prog, true);
1430 void bpf_prog_put(struct bpf_prog *prog)
1432 __bpf_prog_put(prog, true);
1434 EXPORT_SYMBOL_GPL(bpf_prog_put);
1436 static int bpf_prog_release(struct inode *inode, struct file *filp)
1438 struct bpf_prog *prog = filp->private_data;
1444 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1445 struct bpf_prog_stats *stats)
1447 u64 nsecs = 0, cnt = 0;
1450 for_each_possible_cpu(cpu) {
1451 const struct bpf_prog_stats *st;
1455 st = per_cpu_ptr(prog->aux->stats, cpu);
1457 start = u64_stats_fetch_begin_irq(&st->syncp);
1460 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1464 stats->nsecs = nsecs;
1468 #ifdef CONFIG_PROC_FS
1469 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1471 const struct bpf_prog *prog = filp->private_data;
1472 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1473 struct bpf_prog_stats stats;
1475 bpf_prog_get_stats(prog, &stats);
1476 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1483 "run_time_ns:\t%llu\n"
1488 prog->pages * 1ULL << PAGE_SHIFT,
1495 const struct file_operations bpf_prog_fops = {
1496 #ifdef CONFIG_PROC_FS
1497 .show_fdinfo = bpf_prog_show_fdinfo,
1499 .release = bpf_prog_release,
1500 .read = bpf_dummy_read,
1501 .write = bpf_dummy_write,
1504 int bpf_prog_new_fd(struct bpf_prog *prog)
1508 ret = security_bpf_prog(prog);
1512 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1513 O_RDWR | O_CLOEXEC);
1516 static struct bpf_prog *____bpf_prog_get(struct fd f)
1519 return ERR_PTR(-EBADF);
1520 if (f.file->f_op != &bpf_prog_fops) {
1522 return ERR_PTR(-EINVAL);
1525 return f.file->private_data;
1528 void bpf_prog_add(struct bpf_prog *prog, int i)
1530 atomic64_add(i, &prog->aux->refcnt);
1532 EXPORT_SYMBOL_GPL(bpf_prog_add);
1534 void bpf_prog_sub(struct bpf_prog *prog, int i)
1536 /* Only to be used for undoing previous bpf_prog_add() in some
1537 * error path. We still know that another entity in our call
1538 * path holds a reference to the program, thus atomic_sub() can
1539 * be safely used in such cases!
1541 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1543 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1545 void bpf_prog_inc(struct bpf_prog *prog)
1547 atomic64_inc(&prog->aux->refcnt);
1549 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1551 /* prog_idr_lock should have been held */
1552 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1556 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1559 return ERR_PTR(-ENOENT);
1563 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1565 bool bpf_prog_get_ok(struct bpf_prog *prog,
1566 enum bpf_prog_type *attach_type, bool attach_drv)
1568 /* not an attachment, just a refcount inc, always allow */
1572 if (prog->type != *attach_type)
1574 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1580 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1583 struct fd f = fdget(ufd);
1584 struct bpf_prog *prog;
1586 prog = ____bpf_prog_get(f);
1589 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1590 prog = ERR_PTR(-EINVAL);
1600 struct bpf_prog *bpf_prog_get(u32 ufd)
1602 return __bpf_prog_get(ufd, NULL, false);
1605 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1608 return __bpf_prog_get(ufd, &type, attach_drv);
1610 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1612 /* Initially all BPF programs could be loaded w/o specifying
1613 * expected_attach_type. Later for some of them specifying expected_attach_type
1614 * at load time became required so that program could be validated properly.
1615 * Programs of types that are allowed to be loaded both w/ and w/o (for
1616 * backward compatibility) expected_attach_type, should have the default attach
1617 * type assigned to expected_attach_type for the latter case, so that it can be
1618 * validated later at attach time.
1620 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1621 * prog type requires it but has some attach types that have to be backward
1624 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1626 switch (attr->prog_type) {
1627 case BPF_PROG_TYPE_CGROUP_SOCK:
1628 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1629 * exist so checking for non-zero is the way to go here.
1631 if (!attr->expected_attach_type)
1632 attr->expected_attach_type =
1633 BPF_CGROUP_INET_SOCK_CREATE;
1639 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1640 enum bpf_attach_type expected_attach_type,
1641 u32 btf_id, u32 prog_fd)
1643 switch (prog_type) {
1644 case BPF_PROG_TYPE_TRACING:
1645 if (btf_id > BTF_MAX_TYPE)
1649 if (btf_id || prog_fd)
1654 switch (prog_type) {
1655 case BPF_PROG_TYPE_CGROUP_SOCK:
1656 switch (expected_attach_type) {
1657 case BPF_CGROUP_INET_SOCK_CREATE:
1658 case BPF_CGROUP_INET4_POST_BIND:
1659 case BPF_CGROUP_INET6_POST_BIND:
1664 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1665 switch (expected_attach_type) {
1666 case BPF_CGROUP_INET4_BIND:
1667 case BPF_CGROUP_INET6_BIND:
1668 case BPF_CGROUP_INET4_CONNECT:
1669 case BPF_CGROUP_INET6_CONNECT:
1670 case BPF_CGROUP_UDP4_SENDMSG:
1671 case BPF_CGROUP_UDP6_SENDMSG:
1672 case BPF_CGROUP_UDP4_RECVMSG:
1673 case BPF_CGROUP_UDP6_RECVMSG:
1678 case BPF_PROG_TYPE_CGROUP_SKB:
1679 switch (expected_attach_type) {
1680 case BPF_CGROUP_INET_INGRESS:
1681 case BPF_CGROUP_INET_EGRESS:
1686 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1687 switch (expected_attach_type) {
1688 case BPF_CGROUP_SETSOCKOPT:
1689 case BPF_CGROUP_GETSOCKOPT:
1699 /* last field in 'union bpf_attr' used by this command */
1700 #define BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
1702 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1704 enum bpf_prog_type type = attr->prog_type;
1705 struct bpf_prog *prog;
1710 if (CHECK_ATTR(BPF_PROG_LOAD))
1713 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
1714 BPF_F_ANY_ALIGNMENT |
1715 BPF_F_TEST_STATE_FREQ |
1716 BPF_F_TEST_RND_HI32))
1719 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1720 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1721 !capable(CAP_SYS_ADMIN))
1724 /* copy eBPF program license from user space */
1725 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1726 sizeof(license) - 1) < 0)
1728 license[sizeof(license) - 1] = 0;
1730 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1731 is_gpl = license_is_gpl_compatible(license);
1733 if (attr->insn_cnt == 0 ||
1734 attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1736 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1737 type != BPF_PROG_TYPE_CGROUP_SKB &&
1738 !capable(CAP_SYS_ADMIN))
1741 bpf_prog_load_fixup_attach_type(attr);
1742 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
1743 attr->attach_btf_id,
1744 attr->attach_prog_fd))
1747 /* plain bpf_prog allocation */
1748 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1752 prog->expected_attach_type = attr->expected_attach_type;
1753 prog->aux->attach_btf_id = attr->attach_btf_id;
1754 if (attr->attach_prog_fd) {
1755 struct bpf_prog *tgt_prog;
1757 tgt_prog = bpf_prog_get(attr->attach_prog_fd);
1758 if (IS_ERR(tgt_prog)) {
1759 err = PTR_ERR(tgt_prog);
1760 goto free_prog_nouncharge;
1762 prog->aux->linked_prog = tgt_prog;
1765 prog->aux->offload_requested = !!attr->prog_ifindex;
1767 err = security_bpf_prog_alloc(prog->aux);
1769 goto free_prog_nouncharge;
1771 err = bpf_prog_charge_memlock(prog);
1775 prog->len = attr->insn_cnt;
1778 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1779 bpf_prog_insn_size(prog)) != 0)
1782 prog->orig_prog = NULL;
1785 atomic64_set(&prog->aux->refcnt, 1);
1786 prog->gpl_compatible = is_gpl ? 1 : 0;
1788 if (bpf_prog_is_dev_bound(prog->aux)) {
1789 err = bpf_prog_offload_init(prog, attr);
1794 /* find program type: socket_filter vs tracing_filter */
1795 err = find_prog_type(type, prog);
1799 prog->aux->load_time = ktime_get_boottime_ns();
1800 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1804 /* run eBPF verifier */
1805 err = bpf_check(&prog, attr, uattr);
1807 goto free_used_maps;
1809 prog = bpf_prog_select_runtime(prog, &err);
1811 goto free_used_maps;
1813 err = bpf_prog_alloc_id(prog);
1815 goto free_used_maps;
1817 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
1818 * effectively publicly exposed. However, retrieving via
1819 * bpf_prog_get_fd_by_id() will take another reference,
1820 * therefore it cannot be gone underneath us.
1822 * Only for the time /after/ successful bpf_prog_new_fd()
1823 * and before returning to userspace, we might just hold
1824 * one reference and any parallel close on that fd could
1825 * rip everything out. Hence, below notifications must
1826 * happen before bpf_prog_new_fd().
1828 * Also, any failure handling from this point onwards must
1829 * be using bpf_prog_put() given the program is exposed.
1831 bpf_prog_kallsyms_add(prog);
1832 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1834 err = bpf_prog_new_fd(prog);
1840 /* In case we have subprogs, we need to wait for a grace
1841 * period before we can tear down JIT memory since symbols
1842 * are already exposed under kallsyms.
1844 __bpf_prog_put_noref(prog, prog->aux->func_cnt);
1847 bpf_prog_uncharge_memlock(prog);
1849 security_bpf_prog_free(prog->aux);
1850 free_prog_nouncharge:
1851 bpf_prog_free(prog);
1855 #define BPF_OBJ_LAST_FIELD file_flags
1857 static int bpf_obj_pin(const union bpf_attr *attr)
1859 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1862 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1865 static int bpf_obj_get(const union bpf_attr *attr)
1867 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1868 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1871 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1875 static int bpf_tracing_prog_release(struct inode *inode, struct file *filp)
1877 struct bpf_prog *prog = filp->private_data;
1879 WARN_ON_ONCE(bpf_trampoline_unlink_prog(prog));
1884 static const struct file_operations bpf_tracing_prog_fops = {
1885 .release = bpf_tracing_prog_release,
1886 .read = bpf_dummy_read,
1887 .write = bpf_dummy_write,
1890 static int bpf_tracing_prog_attach(struct bpf_prog *prog)
1894 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
1895 prog->expected_attach_type != BPF_TRACE_FEXIT) {
1900 err = bpf_trampoline_link_prog(prog);
1904 tr_fd = anon_inode_getfd("bpf-tracing-prog", &bpf_tracing_prog_fops,
1907 WARN_ON_ONCE(bpf_trampoline_unlink_prog(prog));
1918 struct bpf_raw_tracepoint {
1919 struct bpf_raw_event_map *btp;
1920 struct bpf_prog *prog;
1923 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1925 struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1928 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1929 bpf_prog_put(raw_tp->prog);
1931 bpf_put_raw_tracepoint(raw_tp->btp);
1936 static const struct file_operations bpf_raw_tp_fops = {
1937 .release = bpf_raw_tracepoint_release,
1938 .read = bpf_dummy_read,
1939 .write = bpf_dummy_write,
1942 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1944 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1946 struct bpf_raw_tracepoint *raw_tp;
1947 struct bpf_raw_event_map *btp;
1948 struct bpf_prog *prog;
1949 const char *tp_name;
1953 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
1956 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1958 return PTR_ERR(prog);
1960 if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1961 prog->type != BPF_PROG_TYPE_TRACING &&
1962 prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1967 if (prog->type == BPF_PROG_TYPE_TRACING) {
1968 if (attr->raw_tracepoint.name) {
1969 /* The attach point for this category of programs
1970 * should be specified via btf_id during program load.
1975 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
1976 tp_name = prog->aux->attach_func_name;
1978 return bpf_tracing_prog_attach(prog);
1980 if (strncpy_from_user(buf,
1981 u64_to_user_ptr(attr->raw_tracepoint.name),
1982 sizeof(buf) - 1) < 0) {
1986 buf[sizeof(buf) - 1] = 0;
1990 btp = bpf_get_raw_tracepoint(tp_name);
1996 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
2002 raw_tp->prog = prog;
2004 err = bpf_probe_register(raw_tp->btp, prog);
2008 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
2011 bpf_probe_unregister(raw_tp->btp, prog);
2020 bpf_put_raw_tracepoint(btp);
2026 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2027 enum bpf_attach_type attach_type)
2029 switch (prog->type) {
2030 case BPF_PROG_TYPE_CGROUP_SOCK:
2031 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2032 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2033 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2034 case BPF_PROG_TYPE_CGROUP_SKB:
2035 return prog->enforce_expected_attach_type &&
2036 prog->expected_attach_type != attach_type ?
2043 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
2045 #define BPF_F_ATTACH_MASK \
2046 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
2048 static int bpf_prog_attach(const union bpf_attr *attr)
2050 enum bpf_prog_type ptype;
2051 struct bpf_prog *prog;
2054 if (!capable(CAP_NET_ADMIN))
2057 if (CHECK_ATTR(BPF_PROG_ATTACH))
2060 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
2063 switch (attr->attach_type) {
2064 case BPF_CGROUP_INET_INGRESS:
2065 case BPF_CGROUP_INET_EGRESS:
2066 ptype = BPF_PROG_TYPE_CGROUP_SKB;
2068 case BPF_CGROUP_INET_SOCK_CREATE:
2069 case BPF_CGROUP_INET4_POST_BIND:
2070 case BPF_CGROUP_INET6_POST_BIND:
2071 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2073 case BPF_CGROUP_INET4_BIND:
2074 case BPF_CGROUP_INET6_BIND:
2075 case BPF_CGROUP_INET4_CONNECT:
2076 case BPF_CGROUP_INET6_CONNECT:
2077 case BPF_CGROUP_UDP4_SENDMSG:
2078 case BPF_CGROUP_UDP6_SENDMSG:
2079 case BPF_CGROUP_UDP4_RECVMSG:
2080 case BPF_CGROUP_UDP6_RECVMSG:
2081 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2083 case BPF_CGROUP_SOCK_OPS:
2084 ptype = BPF_PROG_TYPE_SOCK_OPS;
2086 case BPF_CGROUP_DEVICE:
2087 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2089 case BPF_SK_MSG_VERDICT:
2090 ptype = BPF_PROG_TYPE_SK_MSG;
2092 case BPF_SK_SKB_STREAM_PARSER:
2093 case BPF_SK_SKB_STREAM_VERDICT:
2094 ptype = BPF_PROG_TYPE_SK_SKB;
2096 case BPF_LIRC_MODE2:
2097 ptype = BPF_PROG_TYPE_LIRC_MODE2;
2099 case BPF_FLOW_DISSECTOR:
2100 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
2102 case BPF_CGROUP_SYSCTL:
2103 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2105 case BPF_CGROUP_GETSOCKOPT:
2106 case BPF_CGROUP_SETSOCKOPT:
2107 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2113 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
2115 return PTR_ERR(prog);
2117 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
2123 case BPF_PROG_TYPE_SK_SKB:
2124 case BPF_PROG_TYPE_SK_MSG:
2125 ret = sock_map_get_from_fd(attr, prog);
2127 case BPF_PROG_TYPE_LIRC_MODE2:
2128 ret = lirc_prog_attach(attr, prog);
2130 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2131 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
2134 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
2142 #define BPF_PROG_DETACH_LAST_FIELD attach_type
2144 static int bpf_prog_detach(const union bpf_attr *attr)
2146 enum bpf_prog_type ptype;
2148 if (!capable(CAP_NET_ADMIN))
2151 if (CHECK_ATTR(BPF_PROG_DETACH))
2154 switch (attr->attach_type) {
2155 case BPF_CGROUP_INET_INGRESS:
2156 case BPF_CGROUP_INET_EGRESS:
2157 ptype = BPF_PROG_TYPE_CGROUP_SKB;
2159 case BPF_CGROUP_INET_SOCK_CREATE:
2160 case BPF_CGROUP_INET4_POST_BIND:
2161 case BPF_CGROUP_INET6_POST_BIND:
2162 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2164 case BPF_CGROUP_INET4_BIND:
2165 case BPF_CGROUP_INET6_BIND:
2166 case BPF_CGROUP_INET4_CONNECT:
2167 case BPF_CGROUP_INET6_CONNECT:
2168 case BPF_CGROUP_UDP4_SENDMSG:
2169 case BPF_CGROUP_UDP6_SENDMSG:
2170 case BPF_CGROUP_UDP4_RECVMSG:
2171 case BPF_CGROUP_UDP6_RECVMSG:
2172 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2174 case BPF_CGROUP_SOCK_OPS:
2175 ptype = BPF_PROG_TYPE_SOCK_OPS;
2177 case BPF_CGROUP_DEVICE:
2178 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2180 case BPF_SK_MSG_VERDICT:
2181 return sock_map_get_from_fd(attr, NULL);
2182 case BPF_SK_SKB_STREAM_PARSER:
2183 case BPF_SK_SKB_STREAM_VERDICT:
2184 return sock_map_get_from_fd(attr, NULL);
2185 case BPF_LIRC_MODE2:
2186 return lirc_prog_detach(attr);
2187 case BPF_FLOW_DISSECTOR:
2188 return skb_flow_dissector_bpf_prog_detach(attr);
2189 case BPF_CGROUP_SYSCTL:
2190 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2192 case BPF_CGROUP_GETSOCKOPT:
2193 case BPF_CGROUP_SETSOCKOPT:
2194 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2200 return cgroup_bpf_prog_detach(attr, ptype);
2203 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2205 static int bpf_prog_query(const union bpf_attr *attr,
2206 union bpf_attr __user *uattr)
2208 if (!capable(CAP_NET_ADMIN))
2210 if (CHECK_ATTR(BPF_PROG_QUERY))
2212 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2215 switch (attr->query.attach_type) {
2216 case BPF_CGROUP_INET_INGRESS:
2217 case BPF_CGROUP_INET_EGRESS:
2218 case BPF_CGROUP_INET_SOCK_CREATE:
2219 case BPF_CGROUP_INET4_BIND:
2220 case BPF_CGROUP_INET6_BIND:
2221 case BPF_CGROUP_INET4_POST_BIND:
2222 case BPF_CGROUP_INET6_POST_BIND:
2223 case BPF_CGROUP_INET4_CONNECT:
2224 case BPF_CGROUP_INET6_CONNECT:
2225 case BPF_CGROUP_UDP4_SENDMSG:
2226 case BPF_CGROUP_UDP6_SENDMSG:
2227 case BPF_CGROUP_UDP4_RECVMSG:
2228 case BPF_CGROUP_UDP6_RECVMSG:
2229 case BPF_CGROUP_SOCK_OPS:
2230 case BPF_CGROUP_DEVICE:
2231 case BPF_CGROUP_SYSCTL:
2232 case BPF_CGROUP_GETSOCKOPT:
2233 case BPF_CGROUP_SETSOCKOPT:
2235 case BPF_LIRC_MODE2:
2236 return lirc_prog_query(attr, uattr);
2237 case BPF_FLOW_DISSECTOR:
2238 return skb_flow_dissector_prog_query(attr, uattr);
2243 return cgroup_bpf_prog_query(attr, uattr);
2246 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2248 static int bpf_prog_test_run(const union bpf_attr *attr,
2249 union bpf_attr __user *uattr)
2251 struct bpf_prog *prog;
2252 int ret = -ENOTSUPP;
2254 if (!capable(CAP_SYS_ADMIN))
2256 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2259 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2260 (!attr->test.ctx_size_in && attr->test.ctx_in))
2263 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2264 (!attr->test.ctx_size_out && attr->test.ctx_out))
2267 prog = bpf_prog_get(attr->test.prog_fd);
2269 return PTR_ERR(prog);
2271 if (prog->aux->ops->test_run)
2272 ret = prog->aux->ops->test_run(prog, attr, uattr);
2278 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2280 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2281 union bpf_attr __user *uattr,
2285 u32 next_id = attr->start_id;
2288 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2291 if (!capable(CAP_SYS_ADMIN))
2296 if (!idr_get_next(idr, &next_id))
2298 spin_unlock_bh(lock);
2301 err = put_user(next_id, &uattr->next_id);
2306 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2308 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2310 struct bpf_prog *prog;
2311 u32 id = attr->prog_id;
2314 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2317 if (!capable(CAP_SYS_ADMIN))
2320 spin_lock_bh(&prog_idr_lock);
2321 prog = idr_find(&prog_idr, id);
2323 prog = bpf_prog_inc_not_zero(prog);
2325 prog = ERR_PTR(-ENOENT);
2326 spin_unlock_bh(&prog_idr_lock);
2329 return PTR_ERR(prog);
2331 fd = bpf_prog_new_fd(prog);
2338 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2340 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2342 struct bpf_map *map;
2343 u32 id = attr->map_id;
2347 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2348 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2351 if (!capable(CAP_SYS_ADMIN))
2354 f_flags = bpf_get_file_flag(attr->open_flags);
2358 spin_lock_bh(&map_idr_lock);
2359 map = idr_find(&map_idr, id);
2361 map = __bpf_map_inc_not_zero(map, true);
2363 map = ERR_PTR(-ENOENT);
2364 spin_unlock_bh(&map_idr_lock);
2367 return PTR_ERR(map);
2369 fd = bpf_map_new_fd(map, f_flags);
2371 bpf_map_put_with_uref(map);
2376 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2377 unsigned long addr, u32 *off,
2380 const struct bpf_map *map;
2383 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2384 map = prog->aux->used_maps[i];
2385 if (map == (void *)addr) {
2386 *type = BPF_PSEUDO_MAP_FD;
2389 if (!map->ops->map_direct_value_meta)
2391 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2392 *type = BPF_PSEUDO_MAP_VALUE;
2400 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2402 const struct bpf_map *map;
2403 struct bpf_insn *insns;
2408 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2413 for (i = 0; i < prog->len; i++) {
2414 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2415 insns[i].code = BPF_JMP | BPF_CALL;
2416 insns[i].imm = BPF_FUNC_tail_call;
2419 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2420 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2421 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2422 insns[i].code = BPF_JMP | BPF_CALL;
2423 if (!bpf_dump_raw_ok())
2428 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2431 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2432 map = bpf_map_from_imm(prog, imm, &off, &type);
2434 insns[i].src_reg = type;
2435 insns[i].imm = map->id;
2436 insns[i + 1].imm = off;
2444 static int set_info_rec_size(struct bpf_prog_info *info)
2447 * Ensure info.*_rec_size is the same as kernel expected size
2451 * Only allow zero *_rec_size if both _rec_size and _cnt are
2452 * zero. In this case, the kernel will set the expected
2453 * _rec_size back to the info.
2456 if ((info->nr_func_info || info->func_info_rec_size) &&
2457 info->func_info_rec_size != sizeof(struct bpf_func_info))
2460 if ((info->nr_line_info || info->line_info_rec_size) &&
2461 info->line_info_rec_size != sizeof(struct bpf_line_info))
2464 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2465 info->jited_line_info_rec_size != sizeof(__u64))
2468 info->func_info_rec_size = sizeof(struct bpf_func_info);
2469 info->line_info_rec_size = sizeof(struct bpf_line_info);
2470 info->jited_line_info_rec_size = sizeof(__u64);
2475 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2476 const union bpf_attr *attr,
2477 union bpf_attr __user *uattr)
2479 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2480 struct bpf_prog_info info = {};
2481 u32 info_len = attr->info.info_len;
2482 struct bpf_prog_stats stats;
2483 char __user *uinsns;
2487 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2490 info_len = min_t(u32, sizeof(info), info_len);
2492 if (copy_from_user(&info, uinfo, info_len))
2495 info.type = prog->type;
2496 info.id = prog->aux->id;
2497 info.load_time = prog->aux->load_time;
2498 info.created_by_uid = from_kuid_munged(current_user_ns(),
2499 prog->aux->user->uid);
2500 info.gpl_compatible = prog->gpl_compatible;
2502 memcpy(info.tag, prog->tag, sizeof(prog->tag));
2503 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2505 ulen = info.nr_map_ids;
2506 info.nr_map_ids = prog->aux->used_map_cnt;
2507 ulen = min_t(u32, info.nr_map_ids, ulen);
2509 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2512 for (i = 0; i < ulen; i++)
2513 if (put_user(prog->aux->used_maps[i]->id,
2518 err = set_info_rec_size(&info);
2522 bpf_prog_get_stats(prog, &stats);
2523 info.run_time_ns = stats.nsecs;
2524 info.run_cnt = stats.cnt;
2526 if (!capable(CAP_SYS_ADMIN)) {
2527 info.jited_prog_len = 0;
2528 info.xlated_prog_len = 0;
2529 info.nr_jited_ksyms = 0;
2530 info.nr_jited_func_lens = 0;
2531 info.nr_func_info = 0;
2532 info.nr_line_info = 0;
2533 info.nr_jited_line_info = 0;
2537 ulen = info.xlated_prog_len;
2538 info.xlated_prog_len = bpf_prog_insn_size(prog);
2539 if (info.xlated_prog_len && ulen) {
2540 struct bpf_insn *insns_sanitized;
2543 if (prog->blinded && !bpf_dump_raw_ok()) {
2544 info.xlated_prog_insns = 0;
2547 insns_sanitized = bpf_insn_prepare_dump(prog);
2548 if (!insns_sanitized)
2550 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2551 ulen = min_t(u32, info.xlated_prog_len, ulen);
2552 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2553 kfree(insns_sanitized);
2558 if (bpf_prog_is_dev_bound(prog->aux)) {
2559 err = bpf_prog_offload_info_fill(&info, prog);
2565 /* NOTE: the following code is supposed to be skipped for offload.
2566 * bpf_prog_offload_info_fill() is the place to fill similar fields
2569 ulen = info.jited_prog_len;
2570 if (prog->aux->func_cnt) {
2573 info.jited_prog_len = 0;
2574 for (i = 0; i < prog->aux->func_cnt; i++)
2575 info.jited_prog_len += prog->aux->func[i]->jited_len;
2577 info.jited_prog_len = prog->jited_len;
2580 if (info.jited_prog_len && ulen) {
2581 if (bpf_dump_raw_ok()) {
2582 uinsns = u64_to_user_ptr(info.jited_prog_insns);
2583 ulen = min_t(u32, info.jited_prog_len, ulen);
2585 /* for multi-function programs, copy the JITed
2586 * instructions for all the functions
2588 if (prog->aux->func_cnt) {
2593 for (i = 0; i < prog->aux->func_cnt; i++) {
2594 len = prog->aux->func[i]->jited_len;
2595 len = min_t(u32, len, free);
2596 img = (u8 *) prog->aux->func[i]->bpf_func;
2597 if (copy_to_user(uinsns, img, len))
2605 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2609 info.jited_prog_insns = 0;
2613 ulen = info.nr_jited_ksyms;
2614 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2616 if (bpf_dump_raw_ok()) {
2617 unsigned long ksym_addr;
2618 u64 __user *user_ksyms;
2621 /* copy the address of the kernel symbol
2622 * corresponding to each function
2624 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2625 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2626 if (prog->aux->func_cnt) {
2627 for (i = 0; i < ulen; i++) {
2628 ksym_addr = (unsigned long)
2629 prog->aux->func[i]->bpf_func;
2630 if (put_user((u64) ksym_addr,
2635 ksym_addr = (unsigned long) prog->bpf_func;
2636 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2640 info.jited_ksyms = 0;
2644 ulen = info.nr_jited_func_lens;
2645 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2647 if (bpf_dump_raw_ok()) {
2648 u32 __user *user_lens;
2651 /* copy the JITed image lengths for each function */
2652 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2653 user_lens = u64_to_user_ptr(info.jited_func_lens);
2654 if (prog->aux->func_cnt) {
2655 for (i = 0; i < ulen; i++) {
2657 prog->aux->func[i]->jited_len;
2658 if (put_user(func_len, &user_lens[i]))
2662 func_len = prog->jited_len;
2663 if (put_user(func_len, &user_lens[0]))
2667 info.jited_func_lens = 0;
2672 info.btf_id = btf_id(prog->aux->btf);
2674 ulen = info.nr_func_info;
2675 info.nr_func_info = prog->aux->func_info_cnt;
2676 if (info.nr_func_info && ulen) {
2677 char __user *user_finfo;
2679 user_finfo = u64_to_user_ptr(info.func_info);
2680 ulen = min_t(u32, info.nr_func_info, ulen);
2681 if (copy_to_user(user_finfo, prog->aux->func_info,
2682 info.func_info_rec_size * ulen))
2686 ulen = info.nr_line_info;
2687 info.nr_line_info = prog->aux->nr_linfo;
2688 if (info.nr_line_info && ulen) {
2689 __u8 __user *user_linfo;
2691 user_linfo = u64_to_user_ptr(info.line_info);
2692 ulen = min_t(u32, info.nr_line_info, ulen);
2693 if (copy_to_user(user_linfo, prog->aux->linfo,
2694 info.line_info_rec_size * ulen))
2698 ulen = info.nr_jited_line_info;
2699 if (prog->aux->jited_linfo)
2700 info.nr_jited_line_info = prog->aux->nr_linfo;
2702 info.nr_jited_line_info = 0;
2703 if (info.nr_jited_line_info && ulen) {
2704 if (bpf_dump_raw_ok()) {
2705 __u64 __user *user_linfo;
2708 user_linfo = u64_to_user_ptr(info.jited_line_info);
2709 ulen = min_t(u32, info.nr_jited_line_info, ulen);
2710 for (i = 0; i < ulen; i++) {
2711 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2716 info.jited_line_info = 0;
2720 ulen = info.nr_prog_tags;
2721 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2723 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2726 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2727 ulen = min_t(u32, info.nr_prog_tags, ulen);
2728 if (prog->aux->func_cnt) {
2729 for (i = 0; i < ulen; i++) {
2730 if (copy_to_user(user_prog_tags[i],
2731 prog->aux->func[i]->tag,
2736 if (copy_to_user(user_prog_tags[0],
2737 prog->tag, BPF_TAG_SIZE))
2743 if (copy_to_user(uinfo, &info, info_len) ||
2744 put_user(info_len, &uattr->info.info_len))
2750 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2751 const union bpf_attr *attr,
2752 union bpf_attr __user *uattr)
2754 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2755 struct bpf_map_info info = {};
2756 u32 info_len = attr->info.info_len;
2759 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2762 info_len = min_t(u32, sizeof(info), info_len);
2764 info.type = map->map_type;
2766 info.key_size = map->key_size;
2767 info.value_size = map->value_size;
2768 info.max_entries = map->max_entries;
2769 info.map_flags = map->map_flags;
2770 memcpy(info.name, map->name, sizeof(map->name));
2773 info.btf_id = btf_id(map->btf);
2774 info.btf_key_type_id = map->btf_key_type_id;
2775 info.btf_value_type_id = map->btf_value_type_id;
2778 if (bpf_map_is_dev_bound(map)) {
2779 err = bpf_map_offload_info_fill(&info, map);
2784 if (copy_to_user(uinfo, &info, info_len) ||
2785 put_user(info_len, &uattr->info.info_len))
2791 static int bpf_btf_get_info_by_fd(struct btf *btf,
2792 const union bpf_attr *attr,
2793 union bpf_attr __user *uattr)
2795 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2796 u32 info_len = attr->info.info_len;
2799 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2803 return btf_get_info_by_fd(btf, attr, uattr);
2806 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2808 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2809 union bpf_attr __user *uattr)
2811 int ufd = attr->info.bpf_fd;
2815 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2822 if (f.file->f_op == &bpf_prog_fops)
2823 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2825 else if (f.file->f_op == &bpf_map_fops)
2826 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2828 else if (f.file->f_op == &btf_fops)
2829 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2837 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2839 static int bpf_btf_load(const union bpf_attr *attr)
2841 if (CHECK_ATTR(BPF_BTF_LOAD))
2844 if (!capable(CAP_SYS_ADMIN))
2847 return btf_new_fd(attr);
2850 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2852 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2854 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2857 if (!capable(CAP_SYS_ADMIN))
2860 return btf_get_fd_by_id(attr->btf_id);
2863 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2864 union bpf_attr __user *uattr,
2865 u32 prog_id, u32 fd_type,
2866 const char *buf, u64 probe_offset,
2869 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2870 u32 len = buf ? strlen(buf) : 0, input_len;
2873 if (put_user(len, &uattr->task_fd_query.buf_len))
2875 input_len = attr->task_fd_query.buf_len;
2876 if (input_len && ubuf) {
2878 /* nothing to copy, just make ubuf NULL terminated */
2881 if (put_user(zero, ubuf))
2883 } else if (input_len >= len + 1) {
2884 /* ubuf can hold the string with NULL terminator */
2885 if (copy_to_user(ubuf, buf, len + 1))
2888 /* ubuf cannot hold the string with NULL terminator,
2889 * do a partial copy with NULL terminator.
2894 if (copy_to_user(ubuf, buf, input_len - 1))
2896 if (put_user(zero, ubuf + input_len - 1))
2901 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2902 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2903 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2904 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2910 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2912 static int bpf_task_fd_query(const union bpf_attr *attr,
2913 union bpf_attr __user *uattr)
2915 pid_t pid = attr->task_fd_query.pid;
2916 u32 fd = attr->task_fd_query.fd;
2917 const struct perf_event *event;
2918 struct files_struct *files;
2919 struct task_struct *task;
2923 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2926 if (!capable(CAP_SYS_ADMIN))
2929 if (attr->task_fd_query.flags != 0)
2932 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2936 files = get_files_struct(task);
2937 put_task_struct(task);
2942 spin_lock(&files->file_lock);
2943 file = fcheck_files(files, fd);
2948 spin_unlock(&files->file_lock);
2949 put_files_struct(files);
2954 if (file->f_op == &bpf_raw_tp_fops) {
2955 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2956 struct bpf_raw_event_map *btp = raw_tp->btp;
2958 err = bpf_task_fd_query_copy(attr, uattr,
2959 raw_tp->prog->aux->id,
2960 BPF_FD_TYPE_RAW_TRACEPOINT,
2961 btp->tp->name, 0, 0);
2965 event = perf_get_event(file);
2966 if (!IS_ERR(event)) {
2967 u64 probe_offset, probe_addr;
2968 u32 prog_id, fd_type;
2971 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2972 &buf, &probe_offset,
2975 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2989 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2991 union bpf_attr attr = {};
2994 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2997 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
3000 size = min_t(u32, size, sizeof(attr));
3002 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
3003 if (copy_from_user(&attr, uattr, size) != 0)
3006 err = security_bpf(cmd, &attr, size);
3011 case BPF_MAP_CREATE:
3012 err = map_create(&attr);
3014 case BPF_MAP_LOOKUP_ELEM:
3015 err = map_lookup_elem(&attr);
3017 case BPF_MAP_UPDATE_ELEM:
3018 err = map_update_elem(&attr);
3020 case BPF_MAP_DELETE_ELEM:
3021 err = map_delete_elem(&attr);
3023 case BPF_MAP_GET_NEXT_KEY:
3024 err = map_get_next_key(&attr);
3026 case BPF_MAP_FREEZE:
3027 err = map_freeze(&attr);
3030 err = bpf_prog_load(&attr, uattr);
3033 err = bpf_obj_pin(&attr);
3036 err = bpf_obj_get(&attr);
3038 case BPF_PROG_ATTACH:
3039 err = bpf_prog_attach(&attr);
3041 case BPF_PROG_DETACH:
3042 err = bpf_prog_detach(&attr);
3044 case BPF_PROG_QUERY:
3045 err = bpf_prog_query(&attr, uattr);
3047 case BPF_PROG_TEST_RUN:
3048 err = bpf_prog_test_run(&attr, uattr);
3050 case BPF_PROG_GET_NEXT_ID:
3051 err = bpf_obj_get_next_id(&attr, uattr,
3052 &prog_idr, &prog_idr_lock);
3054 case BPF_MAP_GET_NEXT_ID:
3055 err = bpf_obj_get_next_id(&attr, uattr,
3056 &map_idr, &map_idr_lock);
3058 case BPF_BTF_GET_NEXT_ID:
3059 err = bpf_obj_get_next_id(&attr, uattr,
3060 &btf_idr, &btf_idr_lock);
3062 case BPF_PROG_GET_FD_BY_ID:
3063 err = bpf_prog_get_fd_by_id(&attr);
3065 case BPF_MAP_GET_FD_BY_ID:
3066 err = bpf_map_get_fd_by_id(&attr);
3068 case BPF_OBJ_GET_INFO_BY_FD:
3069 err = bpf_obj_get_info_by_fd(&attr, uattr);
3071 case BPF_RAW_TRACEPOINT_OPEN:
3072 err = bpf_raw_tracepoint_open(&attr);
3075 err = bpf_btf_load(&attr);
3077 case BPF_BTF_GET_FD_BY_ID:
3078 err = bpf_btf_get_fd_by_id(&attr);
3080 case BPF_TASK_FD_QUERY:
3081 err = bpf_task_fd_query(&attr, uattr);
3083 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
3084 err = map_lookup_and_delete_elem(&attr);