1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of version 2 of the GNU General Public
5 * License as published by the Free Software Foundation.
7 * This program is distributed in the hope that it will be useful, but
8 * WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
10 * General Public License for more details.
12 #include <linux/bpf.h>
13 #include <linux/bpf_trace.h>
14 #include <linux/bpf_lirc.h>
15 #include <linux/btf.h>
16 #include <linux/syscalls.h>
17 #include <linux/slab.h>
18 #include <linux/sched/signal.h>
19 #include <linux/vmalloc.h>
20 #include <linux/mmzone.h>
21 #include <linux/anon_inodes.h>
22 #include <linux/fdtable.h>
23 #include <linux/file.h>
25 #include <linux/license.h>
26 #include <linux/filter.h>
27 #include <linux/version.h>
28 #include <linux/kernel.h>
29 #include <linux/idr.h>
30 #include <linux/cred.h>
31 #include <linux/timekeeping.h>
32 #include <linux/ctype.h>
33 #include <linux/btf.h>
34 #include <linux/nospec.h>
36 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
37 (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
38 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
39 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
40 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
41 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
43 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
45 DEFINE_PER_CPU(int, bpf_prog_active);
46 static DEFINE_IDR(prog_idr);
47 static DEFINE_SPINLOCK(prog_idr_lock);
48 static DEFINE_IDR(map_idr);
49 static DEFINE_SPINLOCK(map_idr_lock);
51 int sysctl_unprivileged_bpf_disabled __read_mostly;
53 static const struct bpf_map_ops * const bpf_map_types[] = {
54 #define BPF_PROG_TYPE(_id, _ops)
55 #define BPF_MAP_TYPE(_id, _ops) \
57 #include <linux/bpf_types.h>
63 * If we're handed a bigger struct than we know of, ensure all the unknown bits
64 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
65 * we don't know about yet.
67 * There is a ToCToU between this function call and the following
68 * copy_from_user() call. However, this is not a concern since this function is
69 * meant to be a future-proofing of bits.
71 int bpf_check_uarg_tail_zero(void __user *uaddr,
75 unsigned char __user *addr;
76 unsigned char __user *end;
80 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
83 if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
86 if (actual_size <= expected_size)
89 addr = uaddr + expected_size;
90 end = uaddr + actual_size;
92 for (; addr < end; addr++) {
93 err = get_user(val, addr);
103 const struct bpf_map_ops bpf_map_offload_ops = {
104 .map_alloc = bpf_map_offload_map_alloc,
105 .map_free = bpf_map_offload_map_free,
106 .map_check_btf = map_check_no_btf,
109 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
111 const struct bpf_map_ops *ops;
112 u32 type = attr->map_type;
116 if (type >= ARRAY_SIZE(bpf_map_types))
117 return ERR_PTR(-EINVAL);
118 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
119 ops = bpf_map_types[type];
121 return ERR_PTR(-EINVAL);
123 if (ops->map_alloc_check) {
124 err = ops->map_alloc_check(attr);
128 if (attr->map_ifindex)
129 ops = &bpf_map_offload_ops;
130 map = ops->map_alloc(attr);
134 map->map_type = type;
138 void *bpf_map_area_alloc(size_t size, int numa_node)
140 /* We definitely need __GFP_NORETRY, so OOM killer doesn't
141 * trigger under memory pressure as we really just want to
144 const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
147 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
148 area = kmalloc_node(size, GFP_USER | flags, numa_node);
153 return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
154 __builtin_return_address(0));
157 void bpf_map_area_free(void *area)
162 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
164 map->map_type = attr->map_type;
165 map->key_size = attr->key_size;
166 map->value_size = attr->value_size;
167 map->max_entries = attr->max_entries;
168 map->map_flags = attr->map_flags;
169 map->numa_node = bpf_map_attr_numa_node(attr);
172 int bpf_map_precharge_memlock(u32 pages)
174 struct user_struct *user = get_current_user();
175 unsigned long memlock_limit, cur;
177 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
178 cur = atomic_long_read(&user->locked_vm);
180 if (cur + pages > memlock_limit)
185 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
187 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
189 if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
190 atomic_long_sub(pages, &user->locked_vm);
196 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
198 atomic_long_sub(pages, &user->locked_vm);
201 static int bpf_map_init_memlock(struct bpf_map *map)
203 struct user_struct *user = get_current_user();
206 ret = bpf_charge_memlock(user, map->pages);
215 static void bpf_map_release_memlock(struct bpf_map *map)
217 struct user_struct *user = map->user;
218 bpf_uncharge_memlock(user, map->pages);
222 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
226 ret = bpf_charge_memlock(map->user, pages);
233 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
235 bpf_uncharge_memlock(map->user, pages);
239 static int bpf_map_alloc_id(struct bpf_map *map)
243 idr_preload(GFP_KERNEL);
244 spin_lock_bh(&map_idr_lock);
245 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
248 spin_unlock_bh(&map_idr_lock);
251 if (WARN_ON_ONCE(!id))
254 return id > 0 ? 0 : id;
257 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
261 /* Offloaded maps are removed from the IDR store when their device
262 * disappears - even if someone holds an fd to them they are unusable,
263 * the memory is gone, all ops will fail; they are simply waiting for
264 * refcnt to drop to be freed.
270 spin_lock_irqsave(&map_idr_lock, flags);
272 __acquire(&map_idr_lock);
274 idr_remove(&map_idr, map->id);
278 spin_unlock_irqrestore(&map_idr_lock, flags);
280 __release(&map_idr_lock);
283 /* called from workqueue */
284 static void bpf_map_free_deferred(struct work_struct *work)
286 struct bpf_map *map = container_of(work, struct bpf_map, work);
288 bpf_map_release_memlock(map);
289 security_bpf_map_free(map);
290 /* implementation dependent freeing */
291 map->ops->map_free(map);
294 static void bpf_map_put_uref(struct bpf_map *map)
296 if (atomic_dec_and_test(&map->usercnt)) {
297 if (map->ops->map_release_uref)
298 map->ops->map_release_uref(map);
302 /* decrement map refcnt and schedule it for freeing via workqueue
303 * (unrelying map implementation ops->map_free() might sleep)
305 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
307 if (atomic_dec_and_test(&map->refcnt)) {
308 /* bpf_map_free_id() must be called first */
309 bpf_map_free_id(map, do_idr_lock);
311 INIT_WORK(&map->work, bpf_map_free_deferred);
312 schedule_work(&map->work);
316 void bpf_map_put(struct bpf_map *map)
318 __bpf_map_put(map, true);
320 EXPORT_SYMBOL_GPL(bpf_map_put);
322 void bpf_map_put_with_uref(struct bpf_map *map)
324 bpf_map_put_uref(map);
328 static int bpf_map_release(struct inode *inode, struct file *filp)
330 struct bpf_map *map = filp->private_data;
332 if (map->ops->map_release)
333 map->ops->map_release(map, filp);
335 bpf_map_put_with_uref(map);
339 #ifdef CONFIG_PROC_FS
340 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
342 const struct bpf_map *map = filp->private_data;
343 const struct bpf_array *array;
344 u32 owner_prog_type = 0;
347 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
348 array = container_of(map, struct bpf_array, map);
349 owner_prog_type = array->owner_prog_type;
350 owner_jited = array->owner_jited;
366 map->pages * 1ULL << PAGE_SHIFT,
369 if (owner_prog_type) {
370 seq_printf(m, "owner_prog_type:\t%u\n",
372 seq_printf(m, "owner_jited:\t%u\n",
378 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
381 /* We need this handler such that alloc_file() enables
382 * f_mode with FMODE_CAN_READ.
387 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
388 size_t siz, loff_t *ppos)
390 /* We need this handler such that alloc_file() enables
391 * f_mode with FMODE_CAN_WRITE.
396 const struct file_operations bpf_map_fops = {
397 #ifdef CONFIG_PROC_FS
398 .show_fdinfo = bpf_map_show_fdinfo,
400 .release = bpf_map_release,
401 .read = bpf_dummy_read,
402 .write = bpf_dummy_write,
405 int bpf_map_new_fd(struct bpf_map *map, int flags)
409 ret = security_bpf_map(map, OPEN_FMODE(flags));
413 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
417 int bpf_get_file_flag(int flags)
419 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
421 if (flags & BPF_F_RDONLY)
423 if (flags & BPF_F_WRONLY)
428 /* helper macro to check that unused fields 'union bpf_attr' are zero */
429 #define CHECK_ATTR(CMD) \
430 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
431 sizeof(attr->CMD##_LAST_FIELD), 0, \
433 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
434 sizeof(attr->CMD##_LAST_FIELD)) != NULL
436 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
437 * Return 0 on success and < 0 on error.
439 static int bpf_obj_name_cpy(char *dst, const char *src)
441 const char *end = src + BPF_OBJ_NAME_LEN;
443 memset(dst, 0, BPF_OBJ_NAME_LEN);
445 /* Copy all isalnum() and '_' char */
446 while (src < end && *src) {
447 if (!isalnum(*src) && *src != '_')
452 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
459 int map_check_no_btf(const struct bpf_map *map,
460 const struct btf_type *key_type,
461 const struct btf_type *value_type)
466 static int map_check_btf(const struct bpf_map *map, const struct btf *btf,
467 u32 btf_key_id, u32 btf_value_id)
469 const struct btf_type *key_type, *value_type;
470 u32 key_size, value_size;
473 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
474 if (!key_type || key_size != map->key_size)
477 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
478 if (!value_type || value_size != map->value_size)
481 if (map->ops->map_check_btf)
482 ret = map->ops->map_check_btf(map, key_type, value_type);
487 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
488 /* called via syscall */
489 static int map_create(union bpf_attr *attr)
491 int numa_node = bpf_map_attr_numa_node(attr);
496 err = CHECK_ATTR(BPF_MAP_CREATE);
500 f_flags = bpf_get_file_flag(attr->map_flags);
504 if (numa_node != NUMA_NO_NODE &&
505 ((unsigned int)numa_node >= nr_node_ids ||
506 !node_online(numa_node)))
509 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
510 map = find_and_alloc_map(attr);
514 err = bpf_obj_name_cpy(map->name, attr->map_name);
516 goto free_map_nouncharge;
518 atomic_set(&map->refcnt, 1);
519 atomic_set(&map->usercnt, 1);
521 if (attr->btf_key_type_id || attr->btf_value_type_id) {
524 if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
526 goto free_map_nouncharge;
529 btf = btf_get_by_fd(attr->btf_fd);
532 goto free_map_nouncharge;
535 err = map_check_btf(map, btf, attr->btf_key_type_id,
536 attr->btf_value_type_id);
539 goto free_map_nouncharge;
543 map->btf_key_type_id = attr->btf_key_type_id;
544 map->btf_value_type_id = attr->btf_value_type_id;
547 err = security_bpf_map_alloc(map);
549 goto free_map_nouncharge;
551 err = bpf_map_init_memlock(map);
555 err = bpf_map_alloc_id(map);
559 err = bpf_map_new_fd(map, f_flags);
561 /* failed to allocate fd.
562 * bpf_map_put_with_uref() is needed because the above
563 * bpf_map_alloc_id() has published the map
564 * to the userspace and the userspace may
565 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
567 bpf_map_put_with_uref(map);
574 bpf_map_release_memlock(map);
576 security_bpf_map_free(map);
579 map->ops->map_free(map);
583 /* if error is returned, fd is released.
584 * On success caller should complete fd access with matching fdput()
586 struct bpf_map *__bpf_map_get(struct fd f)
589 return ERR_PTR(-EBADF);
590 if (f.file->f_op != &bpf_map_fops) {
592 return ERR_PTR(-EINVAL);
595 return f.file->private_data;
598 /* prog's and map's refcnt limit */
599 #define BPF_MAX_REFCNT 32768
601 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
603 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
604 atomic_dec(&map->refcnt);
605 return ERR_PTR(-EBUSY);
608 atomic_inc(&map->usercnt);
611 EXPORT_SYMBOL_GPL(bpf_map_inc);
613 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
615 struct fd f = fdget(ufd);
618 map = __bpf_map_get(f);
622 map = bpf_map_inc(map, true);
628 /* map_idr_lock should have been held */
629 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
634 refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
636 if (refold >= BPF_MAX_REFCNT) {
637 __bpf_map_put(map, false);
638 return ERR_PTR(-EBUSY);
642 return ERR_PTR(-ENOENT);
645 atomic_inc(&map->usercnt);
650 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
655 /* last field in 'union bpf_attr' used by this command */
656 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
658 static int map_lookup_elem(union bpf_attr *attr)
660 void __user *ukey = u64_to_user_ptr(attr->key);
661 void __user *uvalue = u64_to_user_ptr(attr->value);
662 int ufd = attr->map_fd;
664 void *key, *value, *ptr;
669 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
673 map = __bpf_map_get(f);
677 if (!(f.file->f_mode & FMODE_CAN_READ)) {
682 key = memdup_user(ukey, map->key_size);
688 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
689 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
690 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
691 value_size = round_up(map->value_size, 8) * num_possible_cpus();
692 else if (IS_FD_MAP(map))
693 value_size = sizeof(u32);
695 value_size = map->value_size;
698 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
702 if (bpf_map_is_dev_bound(map)) {
703 err = bpf_map_offload_lookup_elem(map, key, value);
708 this_cpu_inc(bpf_prog_active);
709 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
710 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
711 err = bpf_percpu_hash_copy(map, key, value);
712 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
713 err = bpf_percpu_array_copy(map, key, value);
714 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
715 err = bpf_stackmap_copy(map, key, value);
716 } else if (IS_FD_ARRAY(map)) {
717 err = bpf_fd_array_map_lookup_elem(map, key, value);
718 } else if (IS_FD_HASH(map)) {
719 err = bpf_fd_htab_map_lookup_elem(map, key, value);
720 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
721 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
724 if (map->ops->map_lookup_elem_sys_only)
725 ptr = map->ops->map_lookup_elem_sys_only(map, key);
727 ptr = map->ops->map_lookup_elem(map, key);
729 memcpy(value, ptr, value_size);
731 err = ptr ? 0 : -ENOENT;
733 this_cpu_dec(bpf_prog_active);
741 if (copy_to_user(uvalue, value, value_size) != 0)
755 static void maybe_wait_bpf_programs(struct bpf_map *map)
757 /* Wait for any running BPF programs to complete so that
758 * userspace, when we return to it, knows that all programs
759 * that could be running use the new map value.
761 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
762 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
766 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
768 static int map_update_elem(union bpf_attr *attr)
770 void __user *ukey = u64_to_user_ptr(attr->key);
771 void __user *uvalue = u64_to_user_ptr(attr->value);
772 int ufd = attr->map_fd;
779 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
783 map = __bpf_map_get(f);
787 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
792 key = memdup_user(ukey, map->key_size);
798 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
799 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
800 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
801 value_size = round_up(map->value_size, 8) * num_possible_cpus();
803 value_size = map->value_size;
806 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
811 if (copy_from_user(value, uvalue, value_size) != 0)
814 /* Need to create a kthread, thus must support schedule */
815 if (bpf_map_is_dev_bound(map)) {
816 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
818 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
819 map->map_type == BPF_MAP_TYPE_SOCKHASH ||
820 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
821 err = map->ops->map_update_elem(map, key, value, attr->flags);
825 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
826 * inside bpf map update or delete otherwise deadlocks are possible
829 __this_cpu_inc(bpf_prog_active);
830 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
831 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
832 err = bpf_percpu_hash_update(map, key, value, attr->flags);
833 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
834 err = bpf_percpu_array_update(map, key, value, attr->flags);
835 } else if (IS_FD_ARRAY(map)) {
837 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
840 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
842 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
845 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
846 /* rcu_read_lock() is not needed */
847 err = bpf_fd_reuseport_array_update_elem(map, key, value,
851 err = map->ops->map_update_elem(map, key, value, attr->flags);
854 __this_cpu_dec(bpf_prog_active);
856 maybe_wait_bpf_programs(map);
867 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
869 static int map_delete_elem(union bpf_attr *attr)
871 void __user *ukey = u64_to_user_ptr(attr->key);
872 int ufd = attr->map_fd;
878 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
882 map = __bpf_map_get(f);
886 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
891 key = memdup_user(ukey, map->key_size);
897 if (bpf_map_is_dev_bound(map)) {
898 err = bpf_map_offload_delete_elem(map, key);
903 __this_cpu_inc(bpf_prog_active);
905 err = map->ops->map_delete_elem(map, key);
907 __this_cpu_dec(bpf_prog_active);
909 maybe_wait_bpf_programs(map);
917 /* last field in 'union bpf_attr' used by this command */
918 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
920 static int map_get_next_key(union bpf_attr *attr)
922 void __user *ukey = u64_to_user_ptr(attr->key);
923 void __user *unext_key = u64_to_user_ptr(attr->next_key);
924 int ufd = attr->map_fd;
926 void *key, *next_key;
930 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
934 map = __bpf_map_get(f);
938 if (!(f.file->f_mode & FMODE_CAN_READ)) {
944 key = memdup_user(ukey, map->key_size);
954 next_key = kmalloc(map->key_size, GFP_USER);
958 if (bpf_map_is_dev_bound(map)) {
959 err = bpf_map_offload_get_next_key(map, key, next_key);
964 err = map->ops->map_get_next_key(map, key, next_key);
971 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
985 static const struct bpf_prog_ops * const bpf_prog_types[] = {
986 #define BPF_PROG_TYPE(_id, _name) \
987 [_id] = & _name ## _prog_ops,
988 #define BPF_MAP_TYPE(_id, _ops)
989 #include <linux/bpf_types.h>
994 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
996 const struct bpf_prog_ops *ops;
998 if (type >= ARRAY_SIZE(bpf_prog_types))
1000 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1001 ops = bpf_prog_types[type];
1005 if (!bpf_prog_is_dev_bound(prog->aux))
1006 prog->aux->ops = ops;
1008 prog->aux->ops = &bpf_offload_prog_ops;
1013 /* drop refcnt on maps used by eBPF program and free auxilary data */
1014 static void free_used_maps(struct bpf_prog_aux *aux)
1018 if (aux->cgroup_storage)
1019 bpf_cgroup_storage_release(aux->prog, aux->cgroup_storage);
1021 for (i = 0; i < aux->used_map_cnt; i++)
1022 bpf_map_put(aux->used_maps[i]);
1024 kfree(aux->used_maps);
1027 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1029 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1030 unsigned long user_bufs;
1033 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1034 if (user_bufs > memlock_limit) {
1035 atomic_long_sub(pages, &user->locked_vm);
1043 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1046 atomic_long_sub(pages, &user->locked_vm);
1049 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1051 struct user_struct *user = get_current_user();
1054 ret = __bpf_prog_charge(user, prog->pages);
1060 prog->aux->user = user;
1064 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1066 struct user_struct *user = prog->aux->user;
1068 __bpf_prog_uncharge(user, prog->pages);
1072 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1076 idr_preload(GFP_KERNEL);
1077 spin_lock_bh(&prog_idr_lock);
1078 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1081 spin_unlock_bh(&prog_idr_lock);
1084 /* id is in [1, INT_MAX) */
1085 if (WARN_ON_ONCE(!id))
1088 return id > 0 ? 0 : id;
1091 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1093 /* cBPF to eBPF migrations are currently not in the idr store.
1094 * Offloaded programs are removed from the store when their device
1095 * disappears - even if someone grabs an fd to them they are unusable,
1096 * simply waiting for refcnt to drop to be freed.
1102 spin_lock_bh(&prog_idr_lock);
1104 __acquire(&prog_idr_lock);
1106 idr_remove(&prog_idr, prog->aux->id);
1110 spin_unlock_bh(&prog_idr_lock);
1112 __release(&prog_idr_lock);
1115 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1117 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1119 free_used_maps(aux);
1120 bpf_prog_uncharge_memlock(aux->prog);
1121 security_bpf_prog_free(aux);
1122 bpf_prog_free(aux->prog);
1125 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1127 if (atomic_dec_and_test(&prog->aux->refcnt)) {
1128 /* bpf_prog_free_id() must be called first */
1129 bpf_prog_free_id(prog, do_idr_lock);
1130 bpf_prog_kallsyms_del_all(prog);
1132 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1136 void bpf_prog_put(struct bpf_prog *prog)
1138 __bpf_prog_put(prog, true);
1140 EXPORT_SYMBOL_GPL(bpf_prog_put);
1142 static int bpf_prog_release(struct inode *inode, struct file *filp)
1144 struct bpf_prog *prog = filp->private_data;
1150 #ifdef CONFIG_PROC_FS
1151 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1153 const struct bpf_prog *prog = filp->private_data;
1154 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1156 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1166 prog->pages * 1ULL << PAGE_SHIFT,
1171 const struct file_operations bpf_prog_fops = {
1172 #ifdef CONFIG_PROC_FS
1173 .show_fdinfo = bpf_prog_show_fdinfo,
1175 .release = bpf_prog_release,
1176 .read = bpf_dummy_read,
1177 .write = bpf_dummy_write,
1180 int bpf_prog_new_fd(struct bpf_prog *prog)
1184 ret = security_bpf_prog(prog);
1188 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1189 O_RDWR | O_CLOEXEC);
1192 static struct bpf_prog *____bpf_prog_get(struct fd f)
1195 return ERR_PTR(-EBADF);
1196 if (f.file->f_op != &bpf_prog_fops) {
1198 return ERR_PTR(-EINVAL);
1201 return f.file->private_data;
1204 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1206 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1207 atomic_sub(i, &prog->aux->refcnt);
1208 return ERR_PTR(-EBUSY);
1212 EXPORT_SYMBOL_GPL(bpf_prog_add);
1214 void bpf_prog_sub(struct bpf_prog *prog, int i)
1216 /* Only to be used for undoing previous bpf_prog_add() in some
1217 * error path. We still know that another entity in our call
1218 * path holds a reference to the program, thus atomic_sub() can
1219 * be safely used in such cases!
1221 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1223 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1225 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1227 return bpf_prog_add(prog, 1);
1229 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1231 /* prog_idr_lock should have been held */
1232 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1236 refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1238 if (refold >= BPF_MAX_REFCNT) {
1239 __bpf_prog_put(prog, false);
1240 return ERR_PTR(-EBUSY);
1244 return ERR_PTR(-ENOENT);
1248 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1250 bool bpf_prog_get_ok(struct bpf_prog *prog,
1251 enum bpf_prog_type *attach_type, bool attach_drv)
1253 /* not an attachment, just a refcount inc, always allow */
1257 if (prog->type != *attach_type)
1259 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1265 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1268 struct fd f = fdget(ufd);
1269 struct bpf_prog *prog;
1271 prog = ____bpf_prog_get(f);
1274 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1275 prog = ERR_PTR(-EINVAL);
1279 prog = bpf_prog_inc(prog);
1285 struct bpf_prog *bpf_prog_get(u32 ufd)
1287 return __bpf_prog_get(ufd, NULL, false);
1290 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1293 return __bpf_prog_get(ufd, &type, attach_drv);
1295 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1297 /* Initially all BPF programs could be loaded w/o specifying
1298 * expected_attach_type. Later for some of them specifying expected_attach_type
1299 * at load time became required so that program could be validated properly.
1300 * Programs of types that are allowed to be loaded both w/ and w/o (for
1301 * backward compatibility) expected_attach_type, should have the default attach
1302 * type assigned to expected_attach_type for the latter case, so that it can be
1303 * validated later at attach time.
1305 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1306 * prog type requires it but has some attach types that have to be backward
1309 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1311 switch (attr->prog_type) {
1312 case BPF_PROG_TYPE_CGROUP_SOCK:
1313 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1314 * exist so checking for non-zero is the way to go here.
1316 if (!attr->expected_attach_type)
1317 attr->expected_attach_type =
1318 BPF_CGROUP_INET_SOCK_CREATE;
1324 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1325 enum bpf_attach_type expected_attach_type)
1327 switch (prog_type) {
1328 case BPF_PROG_TYPE_CGROUP_SOCK:
1329 switch (expected_attach_type) {
1330 case BPF_CGROUP_INET_SOCK_CREATE:
1331 case BPF_CGROUP_INET4_POST_BIND:
1332 case BPF_CGROUP_INET6_POST_BIND:
1337 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1338 switch (expected_attach_type) {
1339 case BPF_CGROUP_INET4_BIND:
1340 case BPF_CGROUP_INET6_BIND:
1341 case BPF_CGROUP_INET4_CONNECT:
1342 case BPF_CGROUP_INET6_CONNECT:
1343 case BPF_CGROUP_UDP4_SENDMSG:
1344 case BPF_CGROUP_UDP6_SENDMSG:
1345 case BPF_CGROUP_UDP4_RECVMSG:
1346 case BPF_CGROUP_UDP6_RECVMSG:
1356 /* last field in 'union bpf_attr' used by this command */
1357 #define BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1359 static int bpf_prog_load(union bpf_attr *attr)
1361 enum bpf_prog_type type = attr->prog_type;
1362 struct bpf_prog *prog;
1367 if (CHECK_ATTR(BPF_PROG_LOAD))
1370 if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1373 /* copy eBPF program license from user space */
1374 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1375 sizeof(license) - 1) < 0)
1377 license[sizeof(license) - 1] = 0;
1379 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1380 is_gpl = license_is_gpl_compatible(license);
1382 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1385 if (type == BPF_PROG_TYPE_KPROBE &&
1386 attr->kern_version != LINUX_VERSION_CODE)
1389 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1390 type != BPF_PROG_TYPE_CGROUP_SKB &&
1391 !capable(CAP_SYS_ADMIN))
1394 bpf_prog_load_fixup_attach_type(attr);
1395 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1398 /* plain bpf_prog allocation */
1399 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1403 prog->expected_attach_type = attr->expected_attach_type;
1405 prog->aux->offload_requested = !!attr->prog_ifindex;
1407 err = security_bpf_prog_alloc(prog->aux);
1409 goto free_prog_nouncharge;
1411 err = bpf_prog_charge_memlock(prog);
1415 prog->len = attr->insn_cnt;
1418 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1419 bpf_prog_insn_size(prog)) != 0)
1422 prog->orig_prog = NULL;
1425 atomic_set(&prog->aux->refcnt, 1);
1426 prog->gpl_compatible = is_gpl ? 1 : 0;
1428 if (bpf_prog_is_dev_bound(prog->aux)) {
1429 err = bpf_prog_offload_init(prog, attr);
1434 /* find program type: socket_filter vs tracing_filter */
1435 err = find_prog_type(type, prog);
1439 prog->aux->load_time = ktime_get_boot_ns();
1440 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1444 /* run eBPF verifier */
1445 err = bpf_check(&prog, attr);
1447 goto free_used_maps;
1449 prog = bpf_prog_select_runtime(prog, &err);
1451 goto free_used_maps;
1453 err = bpf_prog_alloc_id(prog);
1455 goto free_used_maps;
1457 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
1458 * effectively publicly exposed. However, retrieving via
1459 * bpf_prog_get_fd_by_id() will take another reference,
1460 * therefore it cannot be gone underneath us.
1462 * Only for the time /after/ successful bpf_prog_new_fd()
1463 * and before returning to userspace, we might just hold
1464 * one reference and any parallel close on that fd could
1465 * rip everything out. Hence, below notifications must
1466 * happen before bpf_prog_new_fd().
1468 * Also, any failure handling from this point onwards must
1469 * be using bpf_prog_put() given the program is exposed.
1471 bpf_prog_kallsyms_add(prog);
1473 err = bpf_prog_new_fd(prog);
1479 bpf_prog_kallsyms_del_subprogs(prog);
1480 free_used_maps(prog->aux);
1482 bpf_prog_uncharge_memlock(prog);
1484 security_bpf_prog_free(prog->aux);
1485 free_prog_nouncharge:
1486 bpf_prog_free(prog);
1490 #define BPF_OBJ_LAST_FIELD file_flags
1492 static int bpf_obj_pin(const union bpf_attr *attr)
1494 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1497 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1500 static int bpf_obj_get(const union bpf_attr *attr)
1502 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1503 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1506 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1510 struct bpf_raw_tracepoint {
1511 struct bpf_raw_event_map *btp;
1512 struct bpf_prog *prog;
1515 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1517 struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1520 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1521 bpf_prog_put(raw_tp->prog);
1527 static const struct file_operations bpf_raw_tp_fops = {
1528 .release = bpf_raw_tracepoint_release,
1529 .read = bpf_dummy_read,
1530 .write = bpf_dummy_write,
1533 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1535 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1537 struct bpf_raw_tracepoint *raw_tp;
1538 struct bpf_raw_event_map *btp;
1539 struct bpf_prog *prog;
1543 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1544 sizeof(tp_name) - 1) < 0)
1546 tp_name[sizeof(tp_name) - 1] = 0;
1548 btp = bpf_find_raw_tracepoint(tp_name);
1552 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1557 prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1558 BPF_PROG_TYPE_RAW_TRACEPOINT);
1560 err = PTR_ERR(prog);
1564 err = bpf_probe_register(raw_tp->btp, prog);
1568 raw_tp->prog = prog;
1569 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1572 bpf_probe_unregister(raw_tp->btp, prog);
1585 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1586 enum bpf_attach_type attach_type)
1588 switch (prog->type) {
1589 case BPF_PROG_TYPE_CGROUP_SOCK:
1590 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1591 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1597 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1599 #define BPF_F_ATTACH_MASK \
1600 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1602 static int bpf_prog_attach(const union bpf_attr *attr)
1604 enum bpf_prog_type ptype;
1605 struct bpf_prog *prog;
1608 if (!capable(CAP_NET_ADMIN))
1611 if (CHECK_ATTR(BPF_PROG_ATTACH))
1614 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1617 switch (attr->attach_type) {
1618 case BPF_CGROUP_INET_INGRESS:
1619 case BPF_CGROUP_INET_EGRESS:
1620 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1622 case BPF_CGROUP_INET_SOCK_CREATE:
1623 case BPF_CGROUP_INET4_POST_BIND:
1624 case BPF_CGROUP_INET6_POST_BIND:
1625 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1627 case BPF_CGROUP_INET4_BIND:
1628 case BPF_CGROUP_INET6_BIND:
1629 case BPF_CGROUP_INET4_CONNECT:
1630 case BPF_CGROUP_INET6_CONNECT:
1631 case BPF_CGROUP_UDP4_SENDMSG:
1632 case BPF_CGROUP_UDP6_SENDMSG:
1633 case BPF_CGROUP_UDP4_RECVMSG:
1634 case BPF_CGROUP_UDP6_RECVMSG:
1635 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1637 case BPF_CGROUP_SOCK_OPS:
1638 ptype = BPF_PROG_TYPE_SOCK_OPS;
1640 case BPF_CGROUP_DEVICE:
1641 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1643 case BPF_SK_MSG_VERDICT:
1644 ptype = BPF_PROG_TYPE_SK_MSG;
1646 case BPF_SK_SKB_STREAM_PARSER:
1647 case BPF_SK_SKB_STREAM_VERDICT:
1648 ptype = BPF_PROG_TYPE_SK_SKB;
1650 case BPF_LIRC_MODE2:
1651 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1657 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1659 return PTR_ERR(prog);
1661 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1667 case BPF_PROG_TYPE_SK_SKB:
1668 case BPF_PROG_TYPE_SK_MSG:
1669 ret = sockmap_get_from_fd(attr, ptype, prog);
1671 case BPF_PROG_TYPE_LIRC_MODE2:
1672 ret = lirc_prog_attach(attr, prog);
1675 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1683 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1685 static int bpf_prog_detach(const union bpf_attr *attr)
1687 enum bpf_prog_type ptype;
1689 if (!capable(CAP_NET_ADMIN))
1692 if (CHECK_ATTR(BPF_PROG_DETACH))
1695 switch (attr->attach_type) {
1696 case BPF_CGROUP_INET_INGRESS:
1697 case BPF_CGROUP_INET_EGRESS:
1698 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1700 case BPF_CGROUP_INET_SOCK_CREATE:
1701 case BPF_CGROUP_INET4_POST_BIND:
1702 case BPF_CGROUP_INET6_POST_BIND:
1703 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1705 case BPF_CGROUP_INET4_BIND:
1706 case BPF_CGROUP_INET6_BIND:
1707 case BPF_CGROUP_INET4_CONNECT:
1708 case BPF_CGROUP_INET6_CONNECT:
1709 case BPF_CGROUP_UDP4_SENDMSG:
1710 case BPF_CGROUP_UDP6_SENDMSG:
1711 case BPF_CGROUP_UDP4_RECVMSG:
1712 case BPF_CGROUP_UDP6_RECVMSG:
1713 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1715 case BPF_CGROUP_SOCK_OPS:
1716 ptype = BPF_PROG_TYPE_SOCK_OPS;
1718 case BPF_CGROUP_DEVICE:
1719 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1721 case BPF_SK_MSG_VERDICT:
1722 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, NULL);
1723 case BPF_SK_SKB_STREAM_PARSER:
1724 case BPF_SK_SKB_STREAM_VERDICT:
1725 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, NULL);
1726 case BPF_LIRC_MODE2:
1727 return lirc_prog_detach(attr);
1732 return cgroup_bpf_prog_detach(attr, ptype);
1735 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1737 static int bpf_prog_query(const union bpf_attr *attr,
1738 union bpf_attr __user *uattr)
1740 if (!capable(CAP_NET_ADMIN))
1742 if (CHECK_ATTR(BPF_PROG_QUERY))
1744 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1747 switch (attr->query.attach_type) {
1748 case BPF_CGROUP_INET_INGRESS:
1749 case BPF_CGROUP_INET_EGRESS:
1750 case BPF_CGROUP_INET_SOCK_CREATE:
1751 case BPF_CGROUP_INET4_BIND:
1752 case BPF_CGROUP_INET6_BIND:
1753 case BPF_CGROUP_INET4_POST_BIND:
1754 case BPF_CGROUP_INET6_POST_BIND:
1755 case BPF_CGROUP_INET4_CONNECT:
1756 case BPF_CGROUP_INET6_CONNECT:
1757 case BPF_CGROUP_UDP4_SENDMSG:
1758 case BPF_CGROUP_UDP6_SENDMSG:
1759 case BPF_CGROUP_UDP4_RECVMSG:
1760 case BPF_CGROUP_UDP6_RECVMSG:
1761 case BPF_CGROUP_SOCK_OPS:
1762 case BPF_CGROUP_DEVICE:
1764 case BPF_LIRC_MODE2:
1765 return lirc_prog_query(attr, uattr);
1770 return cgroup_bpf_prog_query(attr, uattr);
1773 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1775 static int bpf_prog_test_run(const union bpf_attr *attr,
1776 union bpf_attr __user *uattr)
1778 struct bpf_prog *prog;
1779 int ret = -ENOTSUPP;
1781 if (!capable(CAP_SYS_ADMIN))
1783 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1786 prog = bpf_prog_get(attr->test.prog_fd);
1788 return PTR_ERR(prog);
1790 if (prog->aux->ops->test_run)
1791 ret = prog->aux->ops->test_run(prog, attr, uattr);
1797 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1799 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1800 union bpf_attr __user *uattr,
1804 u32 next_id = attr->start_id;
1807 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1810 if (!capable(CAP_SYS_ADMIN))
1815 if (!idr_get_next(idr, &next_id))
1817 spin_unlock_bh(lock);
1820 err = put_user(next_id, &uattr->next_id);
1825 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1827 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1829 struct bpf_prog *prog;
1830 u32 id = attr->prog_id;
1833 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1836 if (!capable(CAP_SYS_ADMIN))
1839 spin_lock_bh(&prog_idr_lock);
1840 prog = idr_find(&prog_idr, id);
1842 prog = bpf_prog_inc_not_zero(prog);
1844 prog = ERR_PTR(-ENOENT);
1845 spin_unlock_bh(&prog_idr_lock);
1848 return PTR_ERR(prog);
1850 fd = bpf_prog_new_fd(prog);
1857 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1859 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1861 struct bpf_map *map;
1862 u32 id = attr->map_id;
1866 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1867 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1870 if (!capable(CAP_SYS_ADMIN))
1873 f_flags = bpf_get_file_flag(attr->open_flags);
1877 spin_lock_bh(&map_idr_lock);
1878 map = idr_find(&map_idr, id);
1880 map = bpf_map_inc_not_zero(map, true);
1882 map = ERR_PTR(-ENOENT);
1883 spin_unlock_bh(&map_idr_lock);
1886 return PTR_ERR(map);
1888 fd = bpf_map_new_fd(map, f_flags);
1890 bpf_map_put_with_uref(map);
1895 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1900 for (i = 0; i < prog->aux->used_map_cnt; i++)
1901 if (prog->aux->used_maps[i] == (void *)addr)
1902 return prog->aux->used_maps[i];
1906 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1908 const struct bpf_map *map;
1909 struct bpf_insn *insns;
1913 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1918 for (i = 0; i < prog->len; i++) {
1919 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1920 insns[i].code = BPF_JMP | BPF_CALL;
1921 insns[i].imm = BPF_FUNC_tail_call;
1924 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1925 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1926 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1927 insns[i].code = BPF_JMP | BPF_CALL;
1928 if (!bpf_dump_raw_ok())
1933 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1936 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1937 map = bpf_map_from_imm(prog, imm);
1939 insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1940 insns[i].imm = map->id;
1941 insns[i + 1].imm = 0;
1945 if (!bpf_dump_raw_ok() &&
1946 imm == (unsigned long)prog->aux) {
1948 insns[i + 1].imm = 0;
1956 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1957 const union bpf_attr *attr,
1958 union bpf_attr __user *uattr)
1960 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1961 struct bpf_prog_info info = {};
1962 u32 info_len = attr->info.info_len;
1963 char __user *uinsns;
1967 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1970 info_len = min_t(u32, sizeof(info), info_len);
1972 if (copy_from_user(&info, uinfo, info_len))
1975 info.type = prog->type;
1976 info.id = prog->aux->id;
1977 info.load_time = prog->aux->load_time;
1978 info.created_by_uid = from_kuid_munged(current_user_ns(),
1979 prog->aux->user->uid);
1980 info.gpl_compatible = prog->gpl_compatible;
1982 memcpy(info.tag, prog->tag, sizeof(prog->tag));
1983 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1985 ulen = info.nr_map_ids;
1986 info.nr_map_ids = prog->aux->used_map_cnt;
1987 ulen = min_t(u32, info.nr_map_ids, ulen);
1989 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1992 for (i = 0; i < ulen; i++)
1993 if (put_user(prog->aux->used_maps[i]->id,
1998 if (!capable(CAP_SYS_ADMIN)) {
1999 info.jited_prog_len = 0;
2000 info.xlated_prog_len = 0;
2001 info.nr_jited_ksyms = 0;
2002 info.nr_jited_func_lens = 0;
2006 ulen = info.xlated_prog_len;
2007 info.xlated_prog_len = bpf_prog_insn_size(prog);
2008 if (info.xlated_prog_len && ulen) {
2009 struct bpf_insn *insns_sanitized;
2012 if (prog->blinded && !bpf_dump_raw_ok()) {
2013 info.xlated_prog_insns = 0;
2016 insns_sanitized = bpf_insn_prepare_dump(prog);
2017 if (!insns_sanitized)
2019 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2020 ulen = min_t(u32, info.xlated_prog_len, ulen);
2021 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2022 kfree(insns_sanitized);
2027 if (bpf_prog_is_dev_bound(prog->aux)) {
2028 err = bpf_prog_offload_info_fill(&info, prog);
2034 /* NOTE: the following code is supposed to be skipped for offload.
2035 * bpf_prog_offload_info_fill() is the place to fill similar fields
2038 ulen = info.jited_prog_len;
2039 if (prog->aux->func_cnt) {
2042 info.jited_prog_len = 0;
2043 for (i = 0; i < prog->aux->func_cnt; i++)
2044 info.jited_prog_len += prog->aux->func[i]->jited_len;
2046 info.jited_prog_len = prog->jited_len;
2049 if (info.jited_prog_len && ulen) {
2050 if (bpf_dump_raw_ok()) {
2051 uinsns = u64_to_user_ptr(info.jited_prog_insns);
2052 ulen = min_t(u32, info.jited_prog_len, ulen);
2054 /* for multi-function programs, copy the JITed
2055 * instructions for all the functions
2057 if (prog->aux->func_cnt) {
2062 for (i = 0; i < prog->aux->func_cnt; i++) {
2063 len = prog->aux->func[i]->jited_len;
2064 len = min_t(u32, len, free);
2065 img = (u8 *) prog->aux->func[i]->bpf_func;
2066 if (copy_to_user(uinsns, img, len))
2074 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2078 info.jited_prog_insns = 0;
2082 ulen = info.nr_jited_ksyms;
2083 info.nr_jited_ksyms = prog->aux->func_cnt;
2084 if (info.nr_jited_ksyms && ulen) {
2085 if (bpf_dump_raw_ok()) {
2086 u64 __user *user_ksyms;
2090 /* copy the address of the kernel symbol
2091 * corresponding to each function
2093 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2094 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2095 for (i = 0; i < ulen; i++) {
2096 ksym_addr = (ulong) prog->aux->func[i]->bpf_func;
2097 ksym_addr &= PAGE_MASK;
2098 if (put_user((u64) ksym_addr, &user_ksyms[i]))
2102 info.jited_ksyms = 0;
2106 ulen = info.nr_jited_func_lens;
2107 info.nr_jited_func_lens = prog->aux->func_cnt;
2108 if (info.nr_jited_func_lens && ulen) {
2109 if (bpf_dump_raw_ok()) {
2110 u32 __user *user_lens;
2113 /* copy the JITed image lengths for each function */
2114 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2115 user_lens = u64_to_user_ptr(info.jited_func_lens);
2116 for (i = 0; i < ulen; i++) {
2117 func_len = prog->aux->func[i]->jited_len;
2118 if (put_user(func_len, &user_lens[i]))
2122 info.jited_func_lens = 0;
2127 if (copy_to_user(uinfo, &info, info_len) ||
2128 put_user(info_len, &uattr->info.info_len))
2134 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2135 const union bpf_attr *attr,
2136 union bpf_attr __user *uattr)
2138 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2139 struct bpf_map_info info = {};
2140 u32 info_len = attr->info.info_len;
2143 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2146 info_len = min_t(u32, sizeof(info), info_len);
2148 info.type = map->map_type;
2150 info.key_size = map->key_size;
2151 info.value_size = map->value_size;
2152 info.max_entries = map->max_entries;
2153 info.map_flags = map->map_flags;
2154 memcpy(info.name, map->name, sizeof(map->name));
2157 info.btf_id = btf_id(map->btf);
2158 info.btf_key_type_id = map->btf_key_type_id;
2159 info.btf_value_type_id = map->btf_value_type_id;
2162 if (bpf_map_is_dev_bound(map)) {
2163 err = bpf_map_offload_info_fill(&info, map);
2168 if (copy_to_user(uinfo, &info, info_len) ||
2169 put_user(info_len, &uattr->info.info_len))
2175 static int bpf_btf_get_info_by_fd(struct btf *btf,
2176 const union bpf_attr *attr,
2177 union bpf_attr __user *uattr)
2179 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2180 u32 info_len = attr->info.info_len;
2183 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2187 return btf_get_info_by_fd(btf, attr, uattr);
2190 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2192 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2193 union bpf_attr __user *uattr)
2195 int ufd = attr->info.bpf_fd;
2199 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2206 if (f.file->f_op == &bpf_prog_fops)
2207 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2209 else if (f.file->f_op == &bpf_map_fops)
2210 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2212 else if (f.file->f_op == &btf_fops)
2213 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2221 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2223 static int bpf_btf_load(const union bpf_attr *attr)
2225 if (CHECK_ATTR(BPF_BTF_LOAD))
2228 if (!capable(CAP_SYS_ADMIN))
2231 return btf_new_fd(attr);
2234 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2236 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2238 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2241 if (!capable(CAP_SYS_ADMIN))
2244 return btf_get_fd_by_id(attr->btf_id);
2247 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2248 union bpf_attr __user *uattr,
2249 u32 prog_id, u32 fd_type,
2250 const char *buf, u64 probe_offset,
2253 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2254 u32 len = buf ? strlen(buf) : 0, input_len;
2257 if (put_user(len, &uattr->task_fd_query.buf_len))
2259 input_len = attr->task_fd_query.buf_len;
2260 if (input_len && ubuf) {
2262 /* nothing to copy, just make ubuf NULL terminated */
2265 if (put_user(zero, ubuf))
2267 } else if (input_len >= len + 1) {
2268 /* ubuf can hold the string with NULL terminator */
2269 if (copy_to_user(ubuf, buf, len + 1))
2272 /* ubuf cannot hold the string with NULL terminator,
2273 * do a partial copy with NULL terminator.
2278 if (copy_to_user(ubuf, buf, input_len - 1))
2280 if (put_user(zero, ubuf + input_len - 1))
2285 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2286 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2287 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2288 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2294 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2296 static int bpf_task_fd_query(const union bpf_attr *attr,
2297 union bpf_attr __user *uattr)
2299 pid_t pid = attr->task_fd_query.pid;
2300 u32 fd = attr->task_fd_query.fd;
2301 const struct perf_event *event;
2302 struct files_struct *files;
2303 struct task_struct *task;
2307 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2310 if (!capable(CAP_SYS_ADMIN))
2313 if (attr->task_fd_query.flags != 0)
2316 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2320 files = get_files_struct(task);
2321 put_task_struct(task);
2326 spin_lock(&files->file_lock);
2327 file = fcheck_files(files, fd);
2332 spin_unlock(&files->file_lock);
2333 put_files_struct(files);
2338 if (file->f_op == &bpf_raw_tp_fops) {
2339 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2340 struct bpf_raw_event_map *btp = raw_tp->btp;
2342 err = bpf_task_fd_query_copy(attr, uattr,
2343 raw_tp->prog->aux->id,
2344 BPF_FD_TYPE_RAW_TRACEPOINT,
2345 btp->tp->name, 0, 0);
2349 event = perf_get_event(file);
2350 if (!IS_ERR(event)) {
2351 u64 probe_offset, probe_addr;
2352 u32 prog_id, fd_type;
2355 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2356 &buf, &probe_offset,
2359 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2373 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2375 union bpf_attr attr = {};
2378 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2381 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2384 size = min_t(u32, size, sizeof(attr));
2386 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2387 if (copy_from_user(&attr, uattr, size) != 0)
2390 err = security_bpf(cmd, &attr, size);
2395 case BPF_MAP_CREATE:
2396 err = map_create(&attr);
2398 case BPF_MAP_LOOKUP_ELEM:
2399 err = map_lookup_elem(&attr);
2401 case BPF_MAP_UPDATE_ELEM:
2402 err = map_update_elem(&attr);
2404 case BPF_MAP_DELETE_ELEM:
2405 err = map_delete_elem(&attr);
2407 case BPF_MAP_GET_NEXT_KEY:
2408 err = map_get_next_key(&attr);
2411 err = bpf_prog_load(&attr);
2414 err = bpf_obj_pin(&attr);
2417 err = bpf_obj_get(&attr);
2419 case BPF_PROG_ATTACH:
2420 err = bpf_prog_attach(&attr);
2422 case BPF_PROG_DETACH:
2423 err = bpf_prog_detach(&attr);
2425 case BPF_PROG_QUERY:
2426 err = bpf_prog_query(&attr, uattr);
2428 case BPF_PROG_TEST_RUN:
2429 err = bpf_prog_test_run(&attr, uattr);
2431 case BPF_PROG_GET_NEXT_ID:
2432 err = bpf_obj_get_next_id(&attr, uattr,
2433 &prog_idr, &prog_idr_lock);
2435 case BPF_MAP_GET_NEXT_ID:
2436 err = bpf_obj_get_next_id(&attr, uattr,
2437 &map_idr, &map_idr_lock);
2439 case BPF_PROG_GET_FD_BY_ID:
2440 err = bpf_prog_get_fd_by_id(&attr);
2442 case BPF_MAP_GET_FD_BY_ID:
2443 err = bpf_map_get_fd_by_id(&attr);
2445 case BPF_OBJ_GET_INFO_BY_FD:
2446 err = bpf_obj_get_info_by_fd(&attr, uattr);
2448 case BPF_RAW_TRACEPOINT_OPEN:
2449 err = bpf_raw_tracepoint_open(&attr);
2452 err = bpf_btf_load(&attr);
2454 case BPF_BTF_GET_FD_BY_ID:
2455 err = bpf_btf_get_fd_by_id(&attr);
2457 case BPF_TASK_FD_QUERY:
2458 err = bpf_task_fd_query(&attr, uattr);