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>
27 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
28 (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
29 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
30 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
31 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
32 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
34 DEFINE_PER_CPU(int, bpf_prog_active);
35 static DEFINE_IDR(prog_idr);
36 static DEFINE_SPINLOCK(prog_idr_lock);
37 static DEFINE_IDR(map_idr);
38 static DEFINE_SPINLOCK(map_idr_lock);
40 int sysctl_unprivileged_bpf_disabled __read_mostly;
42 static const struct bpf_map_ops * const bpf_map_types[] = {
43 #define BPF_PROG_TYPE(_id, _ops)
44 #define BPF_MAP_TYPE(_id, _ops) \
46 #include <linux/bpf_types.h>
52 * If we're handed a bigger struct than we know of, ensure all the unknown bits
53 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
54 * we don't know about yet.
56 * There is a ToCToU between this function call and the following
57 * copy_from_user() call. However, this is not a concern since this function is
58 * meant to be a future-proofing of bits.
60 static int check_uarg_tail_zero(void __user *uaddr,
64 unsigned char __user *addr;
65 unsigned char __user *end;
69 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
72 if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
75 if (actual_size <= expected_size)
78 addr = uaddr + expected_size;
79 end = uaddr + actual_size;
81 for (; addr < end; addr++) {
82 err = get_user(val, addr);
92 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
96 if (attr->map_type >= ARRAY_SIZE(bpf_map_types) ||
97 !bpf_map_types[attr->map_type])
98 return ERR_PTR(-EINVAL);
100 map = bpf_map_types[attr->map_type]->map_alloc(attr);
103 map->ops = bpf_map_types[attr->map_type];
104 map->map_type = attr->map_type;
108 void *bpf_map_area_alloc(size_t size, int numa_node)
110 /* We definitely need __GFP_NORETRY, so OOM killer doesn't
111 * trigger under memory pressure as we really just want to
114 const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
117 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
118 area = kmalloc_node(size, GFP_USER | flags, numa_node);
123 return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
124 __builtin_return_address(0));
127 void bpf_map_area_free(void *area)
132 int bpf_map_precharge_memlock(u32 pages)
134 struct user_struct *user = get_current_user();
135 unsigned long memlock_limit, cur;
137 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
138 cur = atomic_long_read(&user->locked_vm);
140 if (cur + pages > memlock_limit)
145 static int bpf_map_charge_memlock(struct bpf_map *map)
147 struct user_struct *user = get_current_user();
148 unsigned long memlock_limit;
150 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
152 atomic_long_add(map->pages, &user->locked_vm);
154 if (atomic_long_read(&user->locked_vm) > memlock_limit) {
155 atomic_long_sub(map->pages, &user->locked_vm);
163 static void bpf_map_uncharge_memlock(struct bpf_map *map)
165 struct user_struct *user = map->user;
167 atomic_long_sub(map->pages, &user->locked_vm);
171 static int bpf_map_alloc_id(struct bpf_map *map)
175 spin_lock_bh(&map_idr_lock);
176 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
179 spin_unlock_bh(&map_idr_lock);
181 if (WARN_ON_ONCE(!id))
184 return id > 0 ? 0 : id;
187 static void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
192 spin_lock_irqsave(&map_idr_lock, flags);
194 __acquire(&map_idr_lock);
196 idr_remove(&map_idr, map->id);
199 spin_unlock_irqrestore(&map_idr_lock, flags);
201 __release(&map_idr_lock);
204 /* called from workqueue */
205 static void bpf_map_free_deferred(struct work_struct *work)
207 struct bpf_map *map = container_of(work, struct bpf_map, work);
209 bpf_map_uncharge_memlock(map);
210 /* implementation dependent freeing */
211 map->ops->map_free(map);
214 static void bpf_map_put_uref(struct bpf_map *map)
216 if (atomic_dec_and_test(&map->usercnt)) {
217 if (map->ops->map_release_uref)
218 map->ops->map_release_uref(map);
222 /* decrement map refcnt and schedule it for freeing via workqueue
223 * (unrelying map implementation ops->map_free() might sleep)
225 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
227 if (atomic_dec_and_test(&map->refcnt)) {
228 /* bpf_map_free_id() must be called first */
229 bpf_map_free_id(map, do_idr_lock);
230 INIT_WORK(&map->work, bpf_map_free_deferred);
231 schedule_work(&map->work);
235 void bpf_map_put(struct bpf_map *map)
237 __bpf_map_put(map, true);
240 void bpf_map_put_with_uref(struct bpf_map *map)
242 bpf_map_put_uref(map);
246 static int bpf_map_release(struct inode *inode, struct file *filp)
248 struct bpf_map *map = filp->private_data;
250 if (map->ops->map_release)
251 map->ops->map_release(map, filp);
253 bpf_map_put_with_uref(map);
257 #ifdef CONFIG_PROC_FS
258 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
260 const struct bpf_map *map = filp->private_data;
261 const struct bpf_array *array;
262 u32 owner_prog_type = 0;
265 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
266 array = container_of(map, struct bpf_array, map);
267 owner_prog_type = array->owner_prog_type;
268 owner_jited = array->owner_jited;
283 map->pages * 1ULL << PAGE_SHIFT);
285 if (owner_prog_type) {
286 seq_printf(m, "owner_prog_type:\t%u\n",
288 seq_printf(m, "owner_jited:\t%u\n",
294 static const struct file_operations bpf_map_fops = {
295 #ifdef CONFIG_PROC_FS
296 .show_fdinfo = bpf_map_show_fdinfo,
298 .release = bpf_map_release,
301 int bpf_map_new_fd(struct bpf_map *map)
303 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
307 /* helper macro to check that unused fields 'union bpf_attr' are zero */
308 #define CHECK_ATTR(CMD) \
309 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
310 sizeof(attr->CMD##_LAST_FIELD), 0, \
312 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
313 sizeof(attr->CMD##_LAST_FIELD)) != NULL
315 #define BPF_MAP_CREATE_LAST_FIELD numa_node
316 /* called via syscall */
317 static int map_create(union bpf_attr *attr)
319 int numa_node = bpf_map_attr_numa_node(attr);
323 err = CHECK_ATTR(BPF_MAP_CREATE);
327 if (numa_node != NUMA_NO_NODE &&
328 ((unsigned int)numa_node >= nr_node_ids ||
329 !node_online(numa_node)))
332 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
333 map = find_and_alloc_map(attr);
337 atomic_set(&map->refcnt, 1);
338 atomic_set(&map->usercnt, 1);
340 err = bpf_map_charge_memlock(map);
342 goto free_map_nouncharge;
344 err = bpf_map_alloc_id(map);
348 err = bpf_map_new_fd(map);
350 /* failed to allocate fd.
351 * bpf_map_put_with_uref() is needed because the above
352 * bpf_map_alloc_id() has published the map
353 * to the userspace and the userspace may
354 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
356 bpf_map_put_with_uref(map);
360 trace_bpf_map_create(map, err);
364 bpf_map_uncharge_memlock(map);
366 map->ops->map_free(map);
370 /* if error is returned, fd is released.
371 * On success caller should complete fd access with matching fdput()
373 struct bpf_map *__bpf_map_get(struct fd f)
376 return ERR_PTR(-EBADF);
377 if (f.file->f_op != &bpf_map_fops) {
379 return ERR_PTR(-EINVAL);
382 return f.file->private_data;
385 /* prog's and map's refcnt limit */
386 #define BPF_MAX_REFCNT 32768
388 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
390 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
391 atomic_dec(&map->refcnt);
392 return ERR_PTR(-EBUSY);
395 atomic_inc(&map->usercnt);
399 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
401 struct fd f = fdget(ufd);
404 map = __bpf_map_get(f);
408 map = bpf_map_inc(map, true);
414 /* map_idr_lock should have been held */
415 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
420 refold = __atomic_add_unless(&map->refcnt, 1, 0);
422 if (refold >= BPF_MAX_REFCNT) {
423 __bpf_map_put(map, false);
424 return ERR_PTR(-EBUSY);
428 return ERR_PTR(-ENOENT);
431 atomic_inc(&map->usercnt);
436 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
441 /* last field in 'union bpf_attr' used by this command */
442 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
444 static int map_lookup_elem(union bpf_attr *attr)
446 void __user *ukey = u64_to_user_ptr(attr->key);
447 void __user *uvalue = u64_to_user_ptr(attr->value);
448 int ufd = attr->map_fd;
450 void *key, *value, *ptr;
455 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
459 map = __bpf_map_get(f);
463 key = memdup_user(ukey, map->key_size);
469 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
470 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
471 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
472 value_size = round_up(map->value_size, 8) * num_possible_cpus();
473 else if (IS_FD_MAP(map))
474 value_size = sizeof(u32);
476 value_size = map->value_size;
479 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
483 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
484 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
485 err = bpf_percpu_hash_copy(map, key, value);
486 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
487 err = bpf_percpu_array_copy(map, key, value);
488 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
489 err = bpf_stackmap_copy(map, key, value);
490 } else if (IS_FD_ARRAY(map)) {
491 err = bpf_fd_array_map_lookup_elem(map, key, value);
492 } else if (IS_FD_HASH(map)) {
493 err = bpf_fd_htab_map_lookup_elem(map, key, value);
496 if (map->ops->map_lookup_elem_sys_only)
497 ptr = map->ops->map_lookup_elem_sys_only(map, key);
499 ptr = map->ops->map_lookup_elem(map, key);
501 memcpy(value, ptr, value_size);
503 err = ptr ? 0 : -ENOENT;
510 if (copy_to_user(uvalue, value, value_size) != 0)
513 trace_bpf_map_lookup_elem(map, ufd, key, value);
525 static void maybe_wait_bpf_programs(struct bpf_map *map)
527 /* Wait for any running BPF programs to complete so that
528 * userspace, when we return to it, knows that all programs
529 * that could be running use the new map value.
531 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
532 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
536 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
538 static int map_update_elem(union bpf_attr *attr)
540 void __user *ukey = u64_to_user_ptr(attr->key);
541 void __user *uvalue = u64_to_user_ptr(attr->value);
542 int ufd = attr->map_fd;
549 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
553 map = __bpf_map_get(f);
557 key = memdup_user(ukey, map->key_size);
563 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
564 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
565 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
566 value_size = round_up(map->value_size, 8) * num_possible_cpus();
568 value_size = map->value_size;
571 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
576 if (copy_from_user(value, uvalue, value_size) != 0)
579 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
580 * inside bpf map update or delete otherwise deadlocks are possible
583 __this_cpu_inc(bpf_prog_active);
584 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
585 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
586 err = bpf_percpu_hash_update(map, key, value, attr->flags);
587 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
588 err = bpf_percpu_array_update(map, key, value, attr->flags);
589 } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
590 map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
591 map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY ||
592 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) {
594 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
597 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
599 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
604 err = map->ops->map_update_elem(map, key, value, attr->flags);
607 __this_cpu_dec(bpf_prog_active);
609 maybe_wait_bpf_programs(map);
612 trace_bpf_map_update_elem(map, ufd, key, value);
622 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
624 static int map_delete_elem(union bpf_attr *attr)
626 void __user *ukey = u64_to_user_ptr(attr->key);
627 int ufd = attr->map_fd;
633 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
637 map = __bpf_map_get(f);
641 key = memdup_user(ukey, map->key_size);
648 __this_cpu_inc(bpf_prog_active);
650 err = map->ops->map_delete_elem(map, key);
652 __this_cpu_dec(bpf_prog_active);
654 maybe_wait_bpf_programs(map);
657 trace_bpf_map_delete_elem(map, ufd, key);
664 /* last field in 'union bpf_attr' used by this command */
665 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
667 static int map_get_next_key(union bpf_attr *attr)
669 void __user *ukey = u64_to_user_ptr(attr->key);
670 void __user *unext_key = u64_to_user_ptr(attr->next_key);
671 int ufd = attr->map_fd;
673 void *key, *next_key;
677 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
681 map = __bpf_map_get(f);
686 key = memdup_user(ukey, map->key_size);
696 next_key = kmalloc(map->key_size, GFP_USER);
701 err = map->ops->map_get_next_key(map, key, next_key);
707 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
710 trace_bpf_map_next_key(map, ufd, key, next_key);
722 static const struct bpf_verifier_ops * const bpf_prog_types[] = {
723 #define BPF_PROG_TYPE(_id, _ops) \
725 #define BPF_MAP_TYPE(_id, _ops)
726 #include <linux/bpf_types.h>
731 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
733 if (type >= ARRAY_SIZE(bpf_prog_types) || !bpf_prog_types[type])
736 prog->aux->ops = bpf_prog_types[type];
741 /* drop refcnt on maps used by eBPF program and free auxilary data */
742 static void free_used_maps(struct bpf_prog_aux *aux)
746 for (i = 0; i < aux->used_map_cnt; i++)
747 bpf_map_put(aux->used_maps[i]);
749 kfree(aux->used_maps);
752 int __bpf_prog_charge(struct user_struct *user, u32 pages)
754 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
755 unsigned long user_bufs;
758 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
759 if (user_bufs > memlock_limit) {
760 atomic_long_sub(pages, &user->locked_vm);
768 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
771 atomic_long_sub(pages, &user->locked_vm);
774 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
776 struct user_struct *user = get_current_user();
779 ret = __bpf_prog_charge(user, prog->pages);
785 prog->aux->user = user;
789 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
791 struct user_struct *user = prog->aux->user;
793 __bpf_prog_uncharge(user, prog->pages);
797 static int bpf_prog_alloc_id(struct bpf_prog *prog)
801 spin_lock_bh(&prog_idr_lock);
802 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
805 spin_unlock_bh(&prog_idr_lock);
807 /* id is in [1, INT_MAX) */
808 if (WARN_ON_ONCE(!id))
811 return id > 0 ? 0 : id;
814 static void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
816 /* cBPF to eBPF migrations are currently not in the idr store. */
821 spin_lock_bh(&prog_idr_lock);
823 __acquire(&prog_idr_lock);
825 idr_remove(&prog_idr, prog->aux->id);
828 spin_unlock_bh(&prog_idr_lock);
830 __release(&prog_idr_lock);
833 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
835 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
838 bpf_prog_uncharge_memlock(aux->prog);
839 bpf_prog_free(aux->prog);
842 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
844 if (atomic_dec_and_test(&prog->aux->refcnt)) {
845 trace_bpf_prog_put_rcu(prog);
846 /* bpf_prog_free_id() must be called first */
847 bpf_prog_free_id(prog, do_idr_lock);
848 bpf_prog_kallsyms_del(prog);
849 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
853 void bpf_prog_put(struct bpf_prog *prog)
855 __bpf_prog_put(prog, true);
857 EXPORT_SYMBOL_GPL(bpf_prog_put);
859 static int bpf_prog_release(struct inode *inode, struct file *filp)
861 struct bpf_prog *prog = filp->private_data;
867 #ifdef CONFIG_PROC_FS
868 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
870 const struct bpf_prog *prog = filp->private_data;
871 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
873 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
882 prog->pages * 1ULL << PAGE_SHIFT);
886 static const struct file_operations bpf_prog_fops = {
887 #ifdef CONFIG_PROC_FS
888 .show_fdinfo = bpf_prog_show_fdinfo,
890 .release = bpf_prog_release,
893 int bpf_prog_new_fd(struct bpf_prog *prog)
895 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
899 static struct bpf_prog *____bpf_prog_get(struct fd f)
902 return ERR_PTR(-EBADF);
903 if (f.file->f_op != &bpf_prog_fops) {
905 return ERR_PTR(-EINVAL);
908 return f.file->private_data;
911 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
913 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
914 atomic_sub(i, &prog->aux->refcnt);
915 return ERR_PTR(-EBUSY);
919 EXPORT_SYMBOL_GPL(bpf_prog_add);
921 void bpf_prog_sub(struct bpf_prog *prog, int i)
923 /* Only to be used for undoing previous bpf_prog_add() in some
924 * error path. We still know that another entity in our call
925 * path holds a reference to the program, thus atomic_sub() can
926 * be safely used in such cases!
928 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
930 EXPORT_SYMBOL_GPL(bpf_prog_sub);
932 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
934 return bpf_prog_add(prog, 1);
936 EXPORT_SYMBOL_GPL(bpf_prog_inc);
938 /* prog_idr_lock should have been held */
939 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
943 refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
945 if (refold >= BPF_MAX_REFCNT) {
946 __bpf_prog_put(prog, false);
947 return ERR_PTR(-EBUSY);
951 return ERR_PTR(-ENOENT);
955 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
957 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type)
959 struct fd f = fdget(ufd);
960 struct bpf_prog *prog;
962 prog = ____bpf_prog_get(f);
965 if (type && prog->type != *type) {
966 prog = ERR_PTR(-EINVAL);
970 prog = bpf_prog_inc(prog);
976 struct bpf_prog *bpf_prog_get(u32 ufd)
978 return __bpf_prog_get(ufd, NULL);
981 struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type)
983 struct bpf_prog *prog = __bpf_prog_get(ufd, &type);
986 trace_bpf_prog_get_type(prog);
989 EXPORT_SYMBOL_GPL(bpf_prog_get_type);
991 /* last field in 'union bpf_attr' used by this command */
992 #define BPF_PROG_LOAD_LAST_FIELD prog_flags
994 static int bpf_prog_load(union bpf_attr *attr)
996 enum bpf_prog_type type = attr->prog_type;
997 struct bpf_prog *prog;
1002 if (CHECK_ATTR(BPF_PROG_LOAD))
1005 if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1008 /* copy eBPF program license from user space */
1009 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1010 sizeof(license) - 1) < 0)
1012 license[sizeof(license) - 1] = 0;
1014 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1015 is_gpl = license_is_gpl_compatible(license);
1017 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1020 if (type == BPF_PROG_TYPE_KPROBE &&
1021 attr->kern_version != LINUX_VERSION_CODE)
1024 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1025 type != BPF_PROG_TYPE_CGROUP_SKB &&
1026 !capable(CAP_SYS_ADMIN))
1029 /* plain bpf_prog allocation */
1030 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1034 err = bpf_prog_charge_memlock(prog);
1036 goto free_prog_nouncharge;
1038 prog->len = attr->insn_cnt;
1041 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1042 bpf_prog_insn_size(prog)) != 0)
1045 prog->orig_prog = NULL;
1048 atomic_set(&prog->aux->refcnt, 1);
1049 prog->gpl_compatible = is_gpl ? 1 : 0;
1051 /* find program type: socket_filter vs tracing_filter */
1052 err = find_prog_type(type, prog);
1056 /* run eBPF verifier */
1057 err = bpf_check(&prog, attr);
1059 goto free_used_maps;
1061 /* eBPF program is ready to be JITed */
1062 prog = bpf_prog_select_runtime(prog, &err);
1064 goto free_used_maps;
1066 err = bpf_prog_alloc_id(prog);
1068 goto free_used_maps;
1070 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
1071 * effectively publicly exposed. However, retrieving via
1072 * bpf_prog_get_fd_by_id() will take another reference,
1073 * therefore it cannot be gone underneath us.
1075 * Only for the time /after/ successful bpf_prog_new_fd()
1076 * and before returning to userspace, we might just hold
1077 * one reference and any parallel close on that fd could
1078 * rip everything out. Hence, below notifications must
1079 * happen before bpf_prog_new_fd().
1081 * Also, any failure handling from this point onwards must
1082 * be using bpf_prog_put() given the program is exposed.
1084 bpf_prog_kallsyms_add(prog);
1085 trace_bpf_prog_load(prog, err);
1087 err = bpf_prog_new_fd(prog);
1093 free_used_maps(prog->aux);
1095 bpf_prog_uncharge_memlock(prog);
1096 free_prog_nouncharge:
1097 bpf_prog_free(prog);
1101 #define BPF_OBJ_LAST_FIELD bpf_fd
1103 static int bpf_obj_pin(const union bpf_attr *attr)
1105 if (CHECK_ATTR(BPF_OBJ))
1108 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1111 static int bpf_obj_get(const union bpf_attr *attr)
1113 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0)
1116 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname));
1119 #ifdef CONFIG_CGROUP_BPF
1121 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1123 static int sockmap_get_from_fd(const union bpf_attr *attr, bool attach)
1125 struct bpf_prog *prog = NULL;
1126 int ufd = attr->target_fd;
1127 struct bpf_map *map;
1132 map = __bpf_map_get(f);
1134 return PTR_ERR(map);
1137 prog = bpf_prog_get_type(attr->attach_bpf_fd,
1138 BPF_PROG_TYPE_SK_SKB);
1141 return PTR_ERR(prog);
1145 err = sock_map_prog(map, prog, attr->attach_type);
1157 static int bpf_prog_attach(const union bpf_attr *attr)
1159 enum bpf_prog_type ptype;
1160 struct bpf_prog *prog;
1161 struct cgroup *cgrp;
1164 if (!capable(CAP_NET_ADMIN))
1167 if (CHECK_ATTR(BPF_PROG_ATTACH))
1170 if (attr->attach_flags & ~BPF_F_ALLOW_OVERRIDE)
1173 switch (attr->attach_type) {
1174 case BPF_CGROUP_INET_INGRESS:
1175 case BPF_CGROUP_INET_EGRESS:
1176 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1178 case BPF_CGROUP_INET_SOCK_CREATE:
1179 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1181 case BPF_CGROUP_SOCK_OPS:
1182 ptype = BPF_PROG_TYPE_SOCK_OPS;
1184 case BPF_SK_SKB_STREAM_PARSER:
1185 case BPF_SK_SKB_STREAM_VERDICT:
1186 return sockmap_get_from_fd(attr, true);
1191 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1193 return PTR_ERR(prog);
1195 cgrp = cgroup_get_from_fd(attr->target_fd);
1198 return PTR_ERR(cgrp);
1201 ret = cgroup_bpf_update(cgrp, prog, attr->attach_type,
1202 attr->attach_flags & BPF_F_ALLOW_OVERRIDE);
1210 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1212 static int bpf_prog_detach(const union bpf_attr *attr)
1214 struct cgroup *cgrp;
1217 if (!capable(CAP_NET_ADMIN))
1220 if (CHECK_ATTR(BPF_PROG_DETACH))
1223 switch (attr->attach_type) {
1224 case BPF_CGROUP_INET_INGRESS:
1225 case BPF_CGROUP_INET_EGRESS:
1226 case BPF_CGROUP_INET_SOCK_CREATE:
1227 case BPF_CGROUP_SOCK_OPS:
1228 cgrp = cgroup_get_from_fd(attr->target_fd);
1230 return PTR_ERR(cgrp);
1232 ret = cgroup_bpf_update(cgrp, NULL, attr->attach_type, false);
1235 case BPF_SK_SKB_STREAM_PARSER:
1236 case BPF_SK_SKB_STREAM_VERDICT:
1237 ret = sockmap_get_from_fd(attr, false);
1246 #endif /* CONFIG_CGROUP_BPF */
1248 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1250 static int bpf_prog_test_run(const union bpf_attr *attr,
1251 union bpf_attr __user *uattr)
1253 struct bpf_prog *prog;
1254 int ret = -ENOTSUPP;
1256 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1259 prog = bpf_prog_get(attr->test.prog_fd);
1261 return PTR_ERR(prog);
1263 if (prog->aux->ops->test_run)
1264 ret = prog->aux->ops->test_run(prog, attr, uattr);
1270 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1272 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1273 union bpf_attr __user *uattr,
1277 u32 next_id = attr->start_id;
1280 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1283 if (!capable(CAP_SYS_ADMIN))
1288 if (!idr_get_next(idr, &next_id))
1290 spin_unlock_bh(lock);
1293 err = put_user(next_id, &uattr->next_id);
1298 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1300 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1302 struct bpf_prog *prog;
1303 u32 id = attr->prog_id;
1306 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1309 if (!capable(CAP_SYS_ADMIN))
1312 spin_lock_bh(&prog_idr_lock);
1313 prog = idr_find(&prog_idr, id);
1315 prog = bpf_prog_inc_not_zero(prog);
1317 prog = ERR_PTR(-ENOENT);
1318 spin_unlock_bh(&prog_idr_lock);
1321 return PTR_ERR(prog);
1323 fd = bpf_prog_new_fd(prog);
1330 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD map_id
1332 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1334 struct bpf_map *map;
1335 u32 id = attr->map_id;
1338 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID))
1341 if (!capable(CAP_SYS_ADMIN))
1344 spin_lock_bh(&map_idr_lock);
1345 map = idr_find(&map_idr, id);
1347 map = bpf_map_inc_not_zero(map, true);
1349 map = ERR_PTR(-ENOENT);
1350 spin_unlock_bh(&map_idr_lock);
1353 return PTR_ERR(map);
1355 fd = bpf_map_new_fd(map);
1357 bpf_map_put_with_uref(map);
1362 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1363 const union bpf_attr *attr,
1364 union bpf_attr __user *uattr)
1366 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1367 struct bpf_prog_info info;
1368 u32 info_len = attr->info.info_len;
1369 char __user *uinsns;
1373 err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1376 info_len = min_t(u32, sizeof(info), info_len);
1378 memset(&info, 0, sizeof(info));
1379 if (copy_from_user(&info, uinfo, info_len))
1382 info.type = prog->type;
1383 info.id = prog->aux->id;
1385 memcpy(info.tag, prog->tag, sizeof(prog->tag));
1387 if (!capable(CAP_SYS_ADMIN)) {
1388 info.jited_prog_len = 0;
1389 info.xlated_prog_len = 0;
1393 ulen = info.jited_prog_len;
1394 info.jited_prog_len = prog->jited_len;
1395 if (info.jited_prog_len && ulen) {
1396 uinsns = u64_to_user_ptr(info.jited_prog_insns);
1397 ulen = min_t(u32, info.jited_prog_len, ulen);
1398 if (copy_to_user(uinsns, prog->bpf_func, ulen))
1402 ulen = info.xlated_prog_len;
1403 info.xlated_prog_len = bpf_prog_insn_size(prog);
1404 if (info.xlated_prog_len && ulen) {
1405 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1406 ulen = min_t(u32, info.xlated_prog_len, ulen);
1407 if (copy_to_user(uinsns, prog->insnsi, ulen))
1412 if (copy_to_user(uinfo, &info, info_len) ||
1413 put_user(info_len, &uattr->info.info_len))
1419 static int bpf_map_get_info_by_fd(struct bpf_map *map,
1420 const union bpf_attr *attr,
1421 union bpf_attr __user *uattr)
1423 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1424 struct bpf_map_info info;
1425 u32 info_len = attr->info.info_len;
1428 err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1431 info_len = min_t(u32, sizeof(info), info_len);
1433 memset(&info, 0, sizeof(info));
1434 info.type = map->map_type;
1436 info.key_size = map->key_size;
1437 info.value_size = map->value_size;
1438 info.max_entries = map->max_entries;
1439 info.map_flags = map->map_flags;
1441 if (copy_to_user(uinfo, &info, info_len) ||
1442 put_user(info_len, &uattr->info.info_len))
1448 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
1450 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
1451 union bpf_attr __user *uattr)
1453 int ufd = attr->info.bpf_fd;
1457 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
1464 if (f.file->f_op == &bpf_prog_fops)
1465 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
1467 else if (f.file->f_op == &bpf_map_fops)
1468 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
1477 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
1479 union bpf_attr attr;
1482 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
1485 err = check_uarg_tail_zero(uattr, sizeof(attr), size);
1488 size = min_t(u32, size, sizeof(attr));
1490 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
1491 memset(&attr, 0, sizeof(attr));
1492 if (copy_from_user(&attr, uattr, size) != 0)
1496 case BPF_MAP_CREATE:
1497 err = map_create(&attr);
1499 case BPF_MAP_LOOKUP_ELEM:
1500 err = map_lookup_elem(&attr);
1502 case BPF_MAP_UPDATE_ELEM:
1503 err = map_update_elem(&attr);
1505 case BPF_MAP_DELETE_ELEM:
1506 err = map_delete_elem(&attr);
1508 case BPF_MAP_GET_NEXT_KEY:
1509 err = map_get_next_key(&attr);
1512 err = bpf_prog_load(&attr);
1515 err = bpf_obj_pin(&attr);
1518 err = bpf_obj_get(&attr);
1520 #ifdef CONFIG_CGROUP_BPF
1521 case BPF_PROG_ATTACH:
1522 err = bpf_prog_attach(&attr);
1524 case BPF_PROG_DETACH:
1525 err = bpf_prog_detach(&attr);
1528 case BPF_PROG_TEST_RUN:
1529 err = bpf_prog_test_run(&attr, uattr);
1531 case BPF_PROG_GET_NEXT_ID:
1532 err = bpf_obj_get_next_id(&attr, uattr,
1533 &prog_idr, &prog_idr_lock);
1535 case BPF_MAP_GET_NEXT_ID:
1536 err = bpf_obj_get_next_id(&attr, uattr,
1537 &map_idr, &map_idr_lock);
1539 case BPF_PROG_GET_FD_BY_ID:
1540 err = bpf_prog_get_fd_by_id(&attr);
1542 case BPF_MAP_GET_FD_BY_ID:
1543 err = bpf_map_get_fd_by_id(&attr);
1545 case BPF_OBJ_GET_INFO_BY_FD:
1546 err = bpf_obj_get_info_by_fd(&attr, uattr);