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/syscalls.h>
15 #include <linux/slab.h>
16 #include <linux/sched/signal.h>
17 #include <linux/vmalloc.h>
18 #include <linux/mmzone.h>
19 #include <linux/anon_inodes.h>
20 #include <linux/file.h>
21 #include <linux/license.h>
22 #include <linux/filter.h>
23 #include <linux/version.h>
24 #include <linux/kernel.h>
25 #include <linux/idr.h>
26 #include <linux/cred.h>
27 #include <linux/timekeeping.h>
28 #include <linux/ctype.h>
30 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
31 (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
32 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
33 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
34 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
35 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
37 DEFINE_PER_CPU(int, bpf_prog_active);
38 static DEFINE_IDR(prog_idr);
39 static DEFINE_SPINLOCK(prog_idr_lock);
40 static DEFINE_IDR(map_idr);
41 static DEFINE_SPINLOCK(map_idr_lock);
43 int sysctl_unprivileged_bpf_disabled __read_mostly;
45 static const struct bpf_map_ops * const bpf_map_types[] = {
46 #define BPF_PROG_TYPE(_id, _ops)
47 #define BPF_MAP_TYPE(_id, _ops) \
49 #include <linux/bpf_types.h>
55 * If we're handed a bigger struct than we know of, ensure all the unknown bits
56 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
57 * we don't know about yet.
59 * There is a ToCToU between this function call and the following
60 * copy_from_user() call. However, this is not a concern since this function is
61 * meant to be a future-proofing of bits.
63 static int check_uarg_tail_zero(void __user *uaddr,
67 unsigned char __user *addr;
68 unsigned char __user *end;
72 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
75 if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
78 if (actual_size <= expected_size)
81 addr = uaddr + expected_size;
82 end = uaddr + actual_size;
84 for (; addr < end; addr++) {
85 err = get_user(val, addr);
95 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
99 if (attr->map_type >= ARRAY_SIZE(bpf_map_types) ||
100 !bpf_map_types[attr->map_type])
101 return ERR_PTR(-EINVAL);
103 map = bpf_map_types[attr->map_type]->map_alloc(attr);
106 map->ops = bpf_map_types[attr->map_type];
107 map->map_type = attr->map_type;
111 void *bpf_map_area_alloc(size_t size, int numa_node)
113 /* We definitely need __GFP_NORETRY, so OOM killer doesn't
114 * trigger under memory pressure as we really just want to
117 const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
120 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
121 area = kmalloc_node(size, GFP_USER | flags, numa_node);
126 return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
127 __builtin_return_address(0));
130 void bpf_map_area_free(void *area)
135 int bpf_map_precharge_memlock(u32 pages)
137 struct user_struct *user = get_current_user();
138 unsigned long memlock_limit, cur;
140 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
141 cur = atomic_long_read(&user->locked_vm);
143 if (cur + pages > memlock_limit)
148 static int bpf_map_charge_memlock(struct bpf_map *map)
150 struct user_struct *user = get_current_user();
151 unsigned long memlock_limit;
153 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
155 atomic_long_add(map->pages, &user->locked_vm);
157 if (atomic_long_read(&user->locked_vm) > memlock_limit) {
158 atomic_long_sub(map->pages, &user->locked_vm);
166 static void bpf_map_uncharge_memlock(struct bpf_map *map)
168 struct user_struct *user = map->user;
170 atomic_long_sub(map->pages, &user->locked_vm);
174 static int bpf_map_alloc_id(struct bpf_map *map)
178 spin_lock_bh(&map_idr_lock);
179 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
182 spin_unlock_bh(&map_idr_lock);
184 if (WARN_ON_ONCE(!id))
187 return id > 0 ? 0 : id;
190 static void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
195 spin_lock_irqsave(&map_idr_lock, flags);
197 __acquire(&map_idr_lock);
199 idr_remove(&map_idr, map->id);
202 spin_unlock_irqrestore(&map_idr_lock, flags);
204 __release(&map_idr_lock);
207 /* called from workqueue */
208 static void bpf_map_free_deferred(struct work_struct *work)
210 struct bpf_map *map = container_of(work, struct bpf_map, work);
212 bpf_map_uncharge_memlock(map);
213 /* implementation dependent freeing */
214 map->ops->map_free(map);
217 static void bpf_map_put_uref(struct bpf_map *map)
219 if (atomic_dec_and_test(&map->usercnt)) {
220 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
221 bpf_fd_array_map_clear(map);
225 /* decrement map refcnt and schedule it for freeing via workqueue
226 * (unrelying map implementation ops->map_free() might sleep)
228 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
230 if (atomic_dec_and_test(&map->refcnt)) {
231 /* bpf_map_free_id() must be called first */
232 bpf_map_free_id(map, do_idr_lock);
233 INIT_WORK(&map->work, bpf_map_free_deferred);
234 schedule_work(&map->work);
238 void bpf_map_put(struct bpf_map *map)
240 __bpf_map_put(map, true);
243 void bpf_map_put_with_uref(struct bpf_map *map)
245 bpf_map_put_uref(map);
249 static int bpf_map_release(struct inode *inode, struct file *filp)
251 struct bpf_map *map = filp->private_data;
253 if (map->ops->map_release)
254 map->ops->map_release(map, filp);
256 bpf_map_put_with_uref(map);
260 #ifdef CONFIG_PROC_FS
261 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
263 const struct bpf_map *map = filp->private_data;
264 const struct bpf_array *array;
265 u32 owner_prog_type = 0;
268 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
269 array = container_of(map, struct bpf_array, map);
270 owner_prog_type = array->owner_prog_type;
271 owner_jited = array->owner_jited;
286 map->pages * 1ULL << PAGE_SHIFT);
288 if (owner_prog_type) {
289 seq_printf(m, "owner_prog_type:\t%u\n",
291 seq_printf(m, "owner_jited:\t%u\n",
297 static const struct file_operations bpf_map_fops = {
298 #ifdef CONFIG_PROC_FS
299 .show_fdinfo = bpf_map_show_fdinfo,
301 .release = bpf_map_release,
304 int bpf_map_new_fd(struct bpf_map *map)
306 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
310 /* helper macro to check that unused fields 'union bpf_attr' are zero */
311 #define CHECK_ATTR(CMD) \
312 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
313 sizeof(attr->CMD##_LAST_FIELD), 0, \
315 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
316 sizeof(attr->CMD##_LAST_FIELD)) != NULL
318 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
319 * Return 0 on success and < 0 on error.
321 static int bpf_obj_name_cpy(char *dst, const char *src)
323 const char *end = src + BPF_OBJ_NAME_LEN;
325 /* Copy all isalnum() and '_' char */
326 while (src < end && *src) {
327 if (!isalnum(*src) && *src != '_')
332 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
336 /* '\0' terminates dst */
342 #define BPF_MAP_CREATE_LAST_FIELD map_name
343 /* called via syscall */
344 static int map_create(union bpf_attr *attr)
346 int numa_node = bpf_map_attr_numa_node(attr);
350 err = CHECK_ATTR(BPF_MAP_CREATE);
354 if (numa_node != NUMA_NO_NODE &&
355 ((unsigned int)numa_node >= nr_node_ids ||
356 !node_online(numa_node)))
359 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
360 map = find_and_alloc_map(attr);
364 err = bpf_obj_name_cpy(map->name, attr->map_name);
366 goto free_map_nouncharge;
368 atomic_set(&map->refcnt, 1);
369 atomic_set(&map->usercnt, 1);
371 err = bpf_map_charge_memlock(map);
373 goto free_map_nouncharge;
375 err = bpf_map_alloc_id(map);
379 err = bpf_map_new_fd(map);
381 /* failed to allocate fd.
382 * bpf_map_put() is needed because the above
383 * bpf_map_alloc_id() has published the map
384 * to the userspace and the userspace may
385 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
391 trace_bpf_map_create(map, err);
395 bpf_map_uncharge_memlock(map);
397 map->ops->map_free(map);
401 /* if error is returned, fd is released.
402 * On success caller should complete fd access with matching fdput()
404 struct bpf_map *__bpf_map_get(struct fd f)
407 return ERR_PTR(-EBADF);
408 if (f.file->f_op != &bpf_map_fops) {
410 return ERR_PTR(-EINVAL);
413 return f.file->private_data;
416 /* prog's and map's refcnt limit */
417 #define BPF_MAX_REFCNT 32768
419 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
421 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
422 atomic_dec(&map->refcnt);
423 return ERR_PTR(-EBUSY);
426 atomic_inc(&map->usercnt);
430 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
432 struct fd f = fdget(ufd);
435 map = __bpf_map_get(f);
439 map = bpf_map_inc(map, true);
445 /* map_idr_lock should have been held */
446 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
451 refold = __atomic_add_unless(&map->refcnt, 1, 0);
453 if (refold >= BPF_MAX_REFCNT) {
454 __bpf_map_put(map, false);
455 return ERR_PTR(-EBUSY);
459 return ERR_PTR(-ENOENT);
462 atomic_inc(&map->usercnt);
467 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
472 /* last field in 'union bpf_attr' used by this command */
473 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
475 static int map_lookup_elem(union bpf_attr *attr)
477 void __user *ukey = u64_to_user_ptr(attr->key);
478 void __user *uvalue = u64_to_user_ptr(attr->value);
479 int ufd = attr->map_fd;
481 void *key, *value, *ptr;
486 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
490 map = __bpf_map_get(f);
494 key = memdup_user(ukey, map->key_size);
500 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
501 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
502 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
503 value_size = round_up(map->value_size, 8) * num_possible_cpus();
504 else if (IS_FD_MAP(map))
505 value_size = sizeof(u32);
507 value_size = map->value_size;
510 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
514 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
515 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
516 err = bpf_percpu_hash_copy(map, key, value);
517 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
518 err = bpf_percpu_array_copy(map, key, value);
519 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
520 err = bpf_stackmap_copy(map, key, value);
521 } else if (IS_FD_ARRAY(map)) {
522 err = bpf_fd_array_map_lookup_elem(map, key, value);
523 } else if (IS_FD_HASH(map)) {
524 err = bpf_fd_htab_map_lookup_elem(map, key, value);
527 ptr = map->ops->map_lookup_elem(map, key);
529 memcpy(value, ptr, value_size);
531 err = ptr ? 0 : -ENOENT;
538 if (copy_to_user(uvalue, value, value_size) != 0)
541 trace_bpf_map_lookup_elem(map, ufd, key, value);
553 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
555 static int map_update_elem(union bpf_attr *attr)
557 void __user *ukey = u64_to_user_ptr(attr->key);
558 void __user *uvalue = u64_to_user_ptr(attr->value);
559 int ufd = attr->map_fd;
566 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
570 map = __bpf_map_get(f);
574 key = memdup_user(ukey, map->key_size);
580 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
581 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
582 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
583 value_size = round_up(map->value_size, 8) * num_possible_cpus();
585 value_size = map->value_size;
588 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
593 if (copy_from_user(value, uvalue, value_size) != 0)
596 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
597 * inside bpf map update or delete otherwise deadlocks are possible
600 __this_cpu_inc(bpf_prog_active);
601 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
602 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
603 err = bpf_percpu_hash_update(map, key, value, attr->flags);
604 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
605 err = bpf_percpu_array_update(map, key, value, attr->flags);
606 } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
607 map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
608 map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY ||
609 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) {
611 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
614 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
616 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
621 err = map->ops->map_update_elem(map, key, value, attr->flags);
624 __this_cpu_dec(bpf_prog_active);
628 trace_bpf_map_update_elem(map, ufd, key, value);
638 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
640 static int map_delete_elem(union bpf_attr *attr)
642 void __user *ukey = u64_to_user_ptr(attr->key);
643 int ufd = attr->map_fd;
649 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
653 map = __bpf_map_get(f);
657 key = memdup_user(ukey, map->key_size);
664 __this_cpu_inc(bpf_prog_active);
666 err = map->ops->map_delete_elem(map, key);
668 __this_cpu_dec(bpf_prog_active);
672 trace_bpf_map_delete_elem(map, ufd, key);
679 /* last field in 'union bpf_attr' used by this command */
680 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
682 static int map_get_next_key(union bpf_attr *attr)
684 void __user *ukey = u64_to_user_ptr(attr->key);
685 void __user *unext_key = u64_to_user_ptr(attr->next_key);
686 int ufd = attr->map_fd;
688 void *key, *next_key;
692 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
696 map = __bpf_map_get(f);
701 key = memdup_user(ukey, map->key_size);
711 next_key = kmalloc(map->key_size, GFP_USER);
716 err = map->ops->map_get_next_key(map, key, next_key);
722 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
725 trace_bpf_map_next_key(map, ufd, key, next_key);
737 static const struct bpf_verifier_ops * const bpf_prog_types[] = {
738 #define BPF_PROG_TYPE(_id, _ops) \
740 #define BPF_MAP_TYPE(_id, _ops)
741 #include <linux/bpf_types.h>
746 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
748 if (type >= ARRAY_SIZE(bpf_prog_types) || !bpf_prog_types[type])
751 prog->aux->ops = bpf_prog_types[type];
756 /* drop refcnt on maps used by eBPF program and free auxilary data */
757 static void free_used_maps(struct bpf_prog_aux *aux)
761 for (i = 0; i < aux->used_map_cnt; i++)
762 bpf_map_put(aux->used_maps[i]);
764 kfree(aux->used_maps);
767 int __bpf_prog_charge(struct user_struct *user, u32 pages)
769 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
770 unsigned long user_bufs;
773 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
774 if (user_bufs > memlock_limit) {
775 atomic_long_sub(pages, &user->locked_vm);
783 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
786 atomic_long_sub(pages, &user->locked_vm);
789 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
791 struct user_struct *user = get_current_user();
794 ret = __bpf_prog_charge(user, prog->pages);
800 prog->aux->user = user;
804 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
806 struct user_struct *user = prog->aux->user;
808 __bpf_prog_uncharge(user, prog->pages);
812 static int bpf_prog_alloc_id(struct bpf_prog *prog)
816 spin_lock_bh(&prog_idr_lock);
817 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
820 spin_unlock_bh(&prog_idr_lock);
822 /* id is in [1, INT_MAX) */
823 if (WARN_ON_ONCE(!id))
826 return id > 0 ? 0 : id;
829 static void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
831 /* cBPF to eBPF migrations are currently not in the idr store. */
836 spin_lock_bh(&prog_idr_lock);
838 __acquire(&prog_idr_lock);
840 idr_remove(&prog_idr, prog->aux->id);
843 spin_unlock_bh(&prog_idr_lock);
845 __release(&prog_idr_lock);
848 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
850 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
853 bpf_prog_uncharge_memlock(aux->prog);
854 bpf_prog_free(aux->prog);
857 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
859 if (atomic_dec_and_test(&prog->aux->refcnt)) {
860 trace_bpf_prog_put_rcu(prog);
861 /* bpf_prog_free_id() must be called first */
862 bpf_prog_free_id(prog, do_idr_lock);
863 bpf_prog_kallsyms_del(prog);
864 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
868 void bpf_prog_put(struct bpf_prog *prog)
870 __bpf_prog_put(prog, true);
872 EXPORT_SYMBOL_GPL(bpf_prog_put);
874 static int bpf_prog_release(struct inode *inode, struct file *filp)
876 struct bpf_prog *prog = filp->private_data;
882 #ifdef CONFIG_PROC_FS
883 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
885 const struct bpf_prog *prog = filp->private_data;
886 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
888 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
897 prog->pages * 1ULL << PAGE_SHIFT);
901 static const struct file_operations bpf_prog_fops = {
902 #ifdef CONFIG_PROC_FS
903 .show_fdinfo = bpf_prog_show_fdinfo,
905 .release = bpf_prog_release,
908 int bpf_prog_new_fd(struct bpf_prog *prog)
910 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
914 static struct bpf_prog *____bpf_prog_get(struct fd f)
917 return ERR_PTR(-EBADF);
918 if (f.file->f_op != &bpf_prog_fops) {
920 return ERR_PTR(-EINVAL);
923 return f.file->private_data;
926 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
928 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
929 atomic_sub(i, &prog->aux->refcnt);
930 return ERR_PTR(-EBUSY);
934 EXPORT_SYMBOL_GPL(bpf_prog_add);
936 void bpf_prog_sub(struct bpf_prog *prog, int i)
938 /* Only to be used for undoing previous bpf_prog_add() in some
939 * error path. We still know that another entity in our call
940 * path holds a reference to the program, thus atomic_sub() can
941 * be safely used in such cases!
943 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
945 EXPORT_SYMBOL_GPL(bpf_prog_sub);
947 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
949 return bpf_prog_add(prog, 1);
951 EXPORT_SYMBOL_GPL(bpf_prog_inc);
953 /* prog_idr_lock should have been held */
954 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
958 refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
960 if (refold >= BPF_MAX_REFCNT) {
961 __bpf_prog_put(prog, false);
962 return ERR_PTR(-EBUSY);
966 return ERR_PTR(-ENOENT);
970 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
972 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type)
974 struct fd f = fdget(ufd);
975 struct bpf_prog *prog;
977 prog = ____bpf_prog_get(f);
980 if (type && prog->type != *type) {
981 prog = ERR_PTR(-EINVAL);
985 prog = bpf_prog_inc(prog);
991 struct bpf_prog *bpf_prog_get(u32 ufd)
993 return __bpf_prog_get(ufd, NULL);
996 struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type)
998 struct bpf_prog *prog = __bpf_prog_get(ufd, &type);
1001 trace_bpf_prog_get_type(prog);
1004 EXPORT_SYMBOL_GPL(bpf_prog_get_type);
1006 /* last field in 'union bpf_attr' used by this command */
1007 #define BPF_PROG_LOAD_LAST_FIELD prog_name
1009 static int bpf_prog_load(union bpf_attr *attr)
1011 enum bpf_prog_type type = attr->prog_type;
1012 struct bpf_prog *prog;
1017 if (CHECK_ATTR(BPF_PROG_LOAD))
1020 if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1023 /* copy eBPF program license from user space */
1024 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1025 sizeof(license) - 1) < 0)
1027 license[sizeof(license) - 1] = 0;
1029 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1030 is_gpl = license_is_gpl_compatible(license);
1032 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1035 if (type == BPF_PROG_TYPE_KPROBE &&
1036 attr->kern_version != LINUX_VERSION_CODE)
1039 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1040 type != BPF_PROG_TYPE_CGROUP_SKB &&
1041 !capable(CAP_SYS_ADMIN))
1044 /* plain bpf_prog allocation */
1045 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1049 err = bpf_prog_charge_memlock(prog);
1051 goto free_prog_nouncharge;
1053 prog->len = attr->insn_cnt;
1056 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1057 bpf_prog_insn_size(prog)) != 0)
1060 prog->orig_prog = NULL;
1063 atomic_set(&prog->aux->refcnt, 1);
1064 prog->gpl_compatible = is_gpl ? 1 : 0;
1066 /* find program type: socket_filter vs tracing_filter */
1067 err = find_prog_type(type, prog);
1071 prog->aux->load_time = ktime_get_boot_ns();
1072 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1076 /* run eBPF verifier */
1077 err = bpf_check(&prog, attr);
1079 goto free_used_maps;
1081 /* eBPF program is ready to be JITed */
1082 prog = bpf_prog_select_runtime(prog, &err);
1084 goto free_used_maps;
1086 err = bpf_prog_alloc_id(prog);
1088 goto free_used_maps;
1090 err = bpf_prog_new_fd(prog);
1092 /* failed to allocate fd.
1093 * bpf_prog_put() is needed because the above
1094 * bpf_prog_alloc_id() has published the prog
1095 * to the userspace and the userspace may
1096 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1102 bpf_prog_kallsyms_add(prog);
1103 trace_bpf_prog_load(prog, err);
1107 free_used_maps(prog->aux);
1109 bpf_prog_uncharge_memlock(prog);
1110 free_prog_nouncharge:
1111 bpf_prog_free(prog);
1115 #define BPF_OBJ_LAST_FIELD bpf_fd
1117 static int bpf_obj_pin(const union bpf_attr *attr)
1119 if (CHECK_ATTR(BPF_OBJ))
1122 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1125 static int bpf_obj_get(const union bpf_attr *attr)
1127 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0)
1130 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname));
1133 #ifdef CONFIG_CGROUP_BPF
1135 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1137 static int sockmap_get_from_fd(const union bpf_attr *attr, bool attach)
1139 struct bpf_prog *prog = NULL;
1140 int ufd = attr->target_fd;
1141 struct bpf_map *map;
1146 map = __bpf_map_get(f);
1148 return PTR_ERR(map);
1151 prog = bpf_prog_get_type(attr->attach_bpf_fd,
1152 BPF_PROG_TYPE_SK_SKB);
1155 return PTR_ERR(prog);
1159 err = sock_map_prog(map, prog, attr->attach_type);
1171 static int bpf_prog_attach(const union bpf_attr *attr)
1173 enum bpf_prog_type ptype;
1174 struct bpf_prog *prog;
1175 struct cgroup *cgrp;
1178 if (!capable(CAP_NET_ADMIN))
1181 if (CHECK_ATTR(BPF_PROG_ATTACH))
1184 if (attr->attach_flags & ~BPF_F_ALLOW_OVERRIDE)
1187 switch (attr->attach_type) {
1188 case BPF_CGROUP_INET_INGRESS:
1189 case BPF_CGROUP_INET_EGRESS:
1190 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1192 case BPF_CGROUP_INET_SOCK_CREATE:
1193 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1195 case BPF_CGROUP_SOCK_OPS:
1196 ptype = BPF_PROG_TYPE_SOCK_OPS;
1198 case BPF_SK_SKB_STREAM_PARSER:
1199 case BPF_SK_SKB_STREAM_VERDICT:
1200 return sockmap_get_from_fd(attr, true);
1205 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1207 return PTR_ERR(prog);
1209 cgrp = cgroup_get_from_fd(attr->target_fd);
1212 return PTR_ERR(cgrp);
1215 ret = cgroup_bpf_update(cgrp, prog, attr->attach_type,
1216 attr->attach_flags & BPF_F_ALLOW_OVERRIDE);
1224 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1226 static int bpf_prog_detach(const union bpf_attr *attr)
1228 struct cgroup *cgrp;
1231 if (!capable(CAP_NET_ADMIN))
1234 if (CHECK_ATTR(BPF_PROG_DETACH))
1237 switch (attr->attach_type) {
1238 case BPF_CGROUP_INET_INGRESS:
1239 case BPF_CGROUP_INET_EGRESS:
1240 case BPF_CGROUP_INET_SOCK_CREATE:
1241 case BPF_CGROUP_SOCK_OPS:
1242 cgrp = cgroup_get_from_fd(attr->target_fd);
1244 return PTR_ERR(cgrp);
1246 ret = cgroup_bpf_update(cgrp, NULL, attr->attach_type, false);
1249 case BPF_SK_SKB_STREAM_PARSER:
1250 case BPF_SK_SKB_STREAM_VERDICT:
1251 ret = sockmap_get_from_fd(attr, false);
1260 #endif /* CONFIG_CGROUP_BPF */
1262 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1264 static int bpf_prog_test_run(const union bpf_attr *attr,
1265 union bpf_attr __user *uattr)
1267 struct bpf_prog *prog;
1268 int ret = -ENOTSUPP;
1270 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1273 prog = bpf_prog_get(attr->test.prog_fd);
1275 return PTR_ERR(prog);
1277 if (prog->aux->ops->test_run)
1278 ret = prog->aux->ops->test_run(prog, attr, uattr);
1284 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1286 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1287 union bpf_attr __user *uattr,
1291 u32 next_id = attr->start_id;
1294 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1297 if (!capable(CAP_SYS_ADMIN))
1302 if (!idr_get_next(idr, &next_id))
1304 spin_unlock_bh(lock);
1307 err = put_user(next_id, &uattr->next_id);
1312 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1314 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1316 struct bpf_prog *prog;
1317 u32 id = attr->prog_id;
1320 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1323 if (!capable(CAP_SYS_ADMIN))
1326 spin_lock_bh(&prog_idr_lock);
1327 prog = idr_find(&prog_idr, id);
1329 prog = bpf_prog_inc_not_zero(prog);
1331 prog = ERR_PTR(-ENOENT);
1332 spin_unlock_bh(&prog_idr_lock);
1335 return PTR_ERR(prog);
1337 fd = bpf_prog_new_fd(prog);
1344 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD map_id
1346 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1348 struct bpf_map *map;
1349 u32 id = attr->map_id;
1352 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID))
1355 if (!capable(CAP_SYS_ADMIN))
1358 spin_lock_bh(&map_idr_lock);
1359 map = idr_find(&map_idr, id);
1361 map = bpf_map_inc_not_zero(map, true);
1363 map = ERR_PTR(-ENOENT);
1364 spin_unlock_bh(&map_idr_lock);
1367 return PTR_ERR(map);
1369 fd = bpf_map_new_fd(map);
1376 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1377 const union bpf_attr *attr,
1378 union bpf_attr __user *uattr)
1380 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1381 struct bpf_prog_info info = {};
1382 u32 info_len = attr->info.info_len;
1383 char __user *uinsns;
1387 err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1390 info_len = min_t(u32, sizeof(info), info_len);
1392 if (copy_from_user(&info, uinfo, info_len))
1395 info.type = prog->type;
1396 info.id = prog->aux->id;
1397 info.load_time = prog->aux->load_time;
1398 info.created_by_uid = from_kuid_munged(current_user_ns(),
1399 prog->aux->user->uid);
1401 memcpy(info.tag, prog->tag, sizeof(prog->tag));
1402 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1404 ulen = info.nr_map_ids;
1405 info.nr_map_ids = prog->aux->used_map_cnt;
1406 ulen = min_t(u32, info.nr_map_ids, ulen);
1408 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1411 for (i = 0; i < ulen; i++)
1412 if (put_user(prog->aux->used_maps[i]->id,
1417 if (!capable(CAP_SYS_ADMIN)) {
1418 info.jited_prog_len = 0;
1419 info.xlated_prog_len = 0;
1423 ulen = info.jited_prog_len;
1424 info.jited_prog_len = prog->jited_len;
1425 if (info.jited_prog_len && ulen) {
1426 uinsns = u64_to_user_ptr(info.jited_prog_insns);
1427 ulen = min_t(u32, info.jited_prog_len, ulen);
1428 if (copy_to_user(uinsns, prog->bpf_func, ulen))
1432 ulen = info.xlated_prog_len;
1433 info.xlated_prog_len = bpf_prog_insn_size(prog);
1434 if (info.xlated_prog_len && ulen) {
1435 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1436 ulen = min_t(u32, info.xlated_prog_len, ulen);
1437 if (copy_to_user(uinsns, prog->insnsi, ulen))
1442 if (copy_to_user(uinfo, &info, info_len) ||
1443 put_user(info_len, &uattr->info.info_len))
1449 static int bpf_map_get_info_by_fd(struct bpf_map *map,
1450 const union bpf_attr *attr,
1451 union bpf_attr __user *uattr)
1453 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1454 struct bpf_map_info info = {};
1455 u32 info_len = attr->info.info_len;
1458 err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1461 info_len = min_t(u32, sizeof(info), info_len);
1463 info.type = map->map_type;
1465 info.key_size = map->key_size;
1466 info.value_size = map->value_size;
1467 info.max_entries = map->max_entries;
1468 info.map_flags = map->map_flags;
1469 memcpy(info.name, map->name, sizeof(map->name));
1471 if (copy_to_user(uinfo, &info, info_len) ||
1472 put_user(info_len, &uattr->info.info_len))
1478 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
1480 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
1481 union bpf_attr __user *uattr)
1483 int ufd = attr->info.bpf_fd;
1487 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
1494 if (f.file->f_op == &bpf_prog_fops)
1495 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
1497 else if (f.file->f_op == &bpf_map_fops)
1498 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
1507 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
1509 union bpf_attr attr = {};
1512 if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
1515 err = check_uarg_tail_zero(uattr, sizeof(attr), size);
1518 size = min_t(u32, size, sizeof(attr));
1520 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
1521 if (copy_from_user(&attr, uattr, size) != 0)
1525 case BPF_MAP_CREATE:
1526 err = map_create(&attr);
1528 case BPF_MAP_LOOKUP_ELEM:
1529 err = map_lookup_elem(&attr);
1531 case BPF_MAP_UPDATE_ELEM:
1532 err = map_update_elem(&attr);
1534 case BPF_MAP_DELETE_ELEM:
1535 err = map_delete_elem(&attr);
1537 case BPF_MAP_GET_NEXT_KEY:
1538 err = map_get_next_key(&attr);
1541 err = bpf_prog_load(&attr);
1544 err = bpf_obj_pin(&attr);
1547 err = bpf_obj_get(&attr);
1549 #ifdef CONFIG_CGROUP_BPF
1550 case BPF_PROG_ATTACH:
1551 err = bpf_prog_attach(&attr);
1553 case BPF_PROG_DETACH:
1554 err = bpf_prog_detach(&attr);
1557 case BPF_PROG_TEST_RUN:
1558 err = bpf_prog_test_run(&attr, uattr);
1560 case BPF_PROG_GET_NEXT_ID:
1561 err = bpf_obj_get_next_id(&attr, uattr,
1562 &prog_idr, &prog_idr_lock);
1564 case BPF_MAP_GET_NEXT_ID:
1565 err = bpf_obj_get_next_id(&attr, uattr,
1566 &map_idr, &map_idr_lock);
1568 case BPF_PROG_GET_FD_BY_ID:
1569 err = bpf_prog_get_fd_by_id(&attr);
1571 case BPF_MAP_GET_FD_BY_ID:
1572 err = bpf_map_get_fd_by_id(&attr);
1574 case BPF_OBJ_GET_INFO_BY_FD:
1575 err = bpf_obj_get_info_by_fd(&attr, uattr);