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 BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
39 DEFINE_PER_CPU(int, bpf_prog_active);
40 static DEFINE_IDR(prog_idr);
41 static DEFINE_SPINLOCK(prog_idr_lock);
42 static DEFINE_IDR(map_idr);
43 static DEFINE_SPINLOCK(map_idr_lock);
45 int sysctl_unprivileged_bpf_disabled __read_mostly;
47 static const struct bpf_map_ops * const bpf_map_types[] = {
48 #define BPF_PROG_TYPE(_id, _ops)
49 #define BPF_MAP_TYPE(_id, _ops) \
51 #include <linux/bpf_types.h>
57 * If we're handed a bigger struct than we know of, ensure all the unknown bits
58 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
59 * we don't know about yet.
61 * There is a ToCToU between this function call and the following
62 * copy_from_user() call. However, this is not a concern since this function is
63 * meant to be a future-proofing of bits.
65 static int check_uarg_tail_zero(void __user *uaddr,
69 unsigned char __user *addr;
70 unsigned char __user *end;
74 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
77 if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
80 if (actual_size <= expected_size)
83 addr = uaddr + expected_size;
84 end = uaddr + actual_size;
86 for (; addr < end; addr++) {
87 err = get_user(val, addr);
97 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
101 if (attr->map_type >= ARRAY_SIZE(bpf_map_types) ||
102 !bpf_map_types[attr->map_type])
103 return ERR_PTR(-EINVAL);
105 map = bpf_map_types[attr->map_type]->map_alloc(attr);
108 map->ops = bpf_map_types[attr->map_type];
109 map->map_type = attr->map_type;
113 void *bpf_map_area_alloc(size_t size, int numa_node)
115 /* We definitely need __GFP_NORETRY, so OOM killer doesn't
116 * trigger under memory pressure as we really just want to
119 const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
122 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
123 area = kmalloc_node(size, GFP_USER | flags, numa_node);
128 return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
129 __builtin_return_address(0));
132 void bpf_map_area_free(void *area)
137 int bpf_map_precharge_memlock(u32 pages)
139 struct user_struct *user = get_current_user();
140 unsigned long memlock_limit, cur;
142 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
143 cur = atomic_long_read(&user->locked_vm);
145 if (cur + pages > memlock_limit)
150 static int bpf_map_charge_memlock(struct bpf_map *map)
152 struct user_struct *user = get_current_user();
153 unsigned long memlock_limit;
155 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
157 atomic_long_add(map->pages, &user->locked_vm);
159 if (atomic_long_read(&user->locked_vm) > memlock_limit) {
160 atomic_long_sub(map->pages, &user->locked_vm);
168 static void bpf_map_uncharge_memlock(struct bpf_map *map)
170 struct user_struct *user = map->user;
172 atomic_long_sub(map->pages, &user->locked_vm);
176 static int bpf_map_alloc_id(struct bpf_map *map)
180 spin_lock_bh(&map_idr_lock);
181 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
184 spin_unlock_bh(&map_idr_lock);
186 if (WARN_ON_ONCE(!id))
189 return id > 0 ? 0 : id;
192 static void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
197 spin_lock_irqsave(&map_idr_lock, flags);
199 __acquire(&map_idr_lock);
201 idr_remove(&map_idr, map->id);
204 spin_unlock_irqrestore(&map_idr_lock, flags);
206 __release(&map_idr_lock);
209 /* called from workqueue */
210 static void bpf_map_free_deferred(struct work_struct *work)
212 struct bpf_map *map = container_of(work, struct bpf_map, work);
214 bpf_map_uncharge_memlock(map);
215 security_bpf_map_free(map);
216 /* implementation dependent freeing */
217 map->ops->map_free(map);
220 static void bpf_map_put_uref(struct bpf_map *map)
222 if (atomic_dec_and_test(&map->usercnt)) {
223 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
224 bpf_fd_array_map_clear(map);
228 /* decrement map refcnt and schedule it for freeing via workqueue
229 * (unrelying map implementation ops->map_free() might sleep)
231 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
233 if (atomic_dec_and_test(&map->refcnt)) {
234 /* bpf_map_free_id() must be called first */
235 bpf_map_free_id(map, do_idr_lock);
236 INIT_WORK(&map->work, bpf_map_free_deferred);
237 schedule_work(&map->work);
241 void bpf_map_put(struct bpf_map *map)
243 __bpf_map_put(map, true);
246 void bpf_map_put_with_uref(struct bpf_map *map)
248 bpf_map_put_uref(map);
252 static int bpf_map_release(struct inode *inode, struct file *filp)
254 struct bpf_map *map = filp->private_data;
256 if (map->ops->map_release)
257 map->ops->map_release(map, filp);
259 bpf_map_put_with_uref(map);
263 #ifdef CONFIG_PROC_FS
264 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
266 const struct bpf_map *map = filp->private_data;
267 const struct bpf_array *array;
268 u32 owner_prog_type = 0;
271 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
272 array = container_of(map, struct bpf_array, map);
273 owner_prog_type = array->owner_prog_type;
274 owner_jited = array->owner_jited;
289 map->pages * 1ULL << PAGE_SHIFT);
291 if (owner_prog_type) {
292 seq_printf(m, "owner_prog_type:\t%u\n",
294 seq_printf(m, "owner_jited:\t%u\n",
300 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
303 /* We need this handler such that alloc_file() enables
304 * f_mode with FMODE_CAN_READ.
309 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
310 size_t siz, loff_t *ppos)
312 /* We need this handler such that alloc_file() enables
313 * f_mode with FMODE_CAN_WRITE.
318 const struct file_operations bpf_map_fops = {
319 #ifdef CONFIG_PROC_FS
320 .show_fdinfo = bpf_map_show_fdinfo,
322 .release = bpf_map_release,
323 .read = bpf_dummy_read,
324 .write = bpf_dummy_write,
327 int bpf_map_new_fd(struct bpf_map *map, int flags)
331 ret = security_bpf_map(map, OPEN_FMODE(flags));
335 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
339 int bpf_get_file_flag(int flags)
341 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
343 if (flags & BPF_F_RDONLY)
345 if (flags & BPF_F_WRONLY)
350 /* helper macro to check that unused fields 'union bpf_attr' are zero */
351 #define CHECK_ATTR(CMD) \
352 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
353 sizeof(attr->CMD##_LAST_FIELD), 0, \
355 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
356 sizeof(attr->CMD##_LAST_FIELD)) != NULL
358 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
359 * Return 0 on success and < 0 on error.
361 static int bpf_obj_name_cpy(char *dst, const char *src)
363 const char *end = src + BPF_OBJ_NAME_LEN;
365 memset(dst, 0, BPF_OBJ_NAME_LEN);
367 /* Copy all isalnum() and '_' char */
368 while (src < end && *src) {
369 if (!isalnum(*src) && *src != '_')
374 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
381 #define BPF_MAP_CREATE_LAST_FIELD map_name
382 /* called via syscall */
383 static int map_create(union bpf_attr *attr)
385 int numa_node = bpf_map_attr_numa_node(attr);
390 err = CHECK_ATTR(BPF_MAP_CREATE);
394 f_flags = bpf_get_file_flag(attr->map_flags);
398 if (numa_node != NUMA_NO_NODE &&
399 ((unsigned int)numa_node >= nr_node_ids ||
400 !node_online(numa_node)))
403 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
404 map = find_and_alloc_map(attr);
408 err = bpf_obj_name_cpy(map->name, attr->map_name);
410 goto free_map_nouncharge;
412 atomic_set(&map->refcnt, 1);
413 atomic_set(&map->usercnt, 1);
415 err = security_bpf_map_alloc(map);
417 goto free_map_nouncharge;
419 err = bpf_map_charge_memlock(map);
423 err = bpf_map_alloc_id(map);
427 err = bpf_map_new_fd(map, f_flags);
429 /* failed to allocate fd.
430 * bpf_map_put() is needed because the above
431 * bpf_map_alloc_id() has published the map
432 * to the userspace and the userspace may
433 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
439 trace_bpf_map_create(map, err);
443 bpf_map_uncharge_memlock(map);
445 security_bpf_map_free(map);
447 map->ops->map_free(map);
451 /* if error is returned, fd is released.
452 * On success caller should complete fd access with matching fdput()
454 struct bpf_map *__bpf_map_get(struct fd f)
457 return ERR_PTR(-EBADF);
458 if (f.file->f_op != &bpf_map_fops) {
460 return ERR_PTR(-EINVAL);
463 return f.file->private_data;
466 /* prog's and map's refcnt limit */
467 #define BPF_MAX_REFCNT 32768
469 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
471 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
472 atomic_dec(&map->refcnt);
473 return ERR_PTR(-EBUSY);
476 atomic_inc(&map->usercnt);
480 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
482 struct fd f = fdget(ufd);
485 map = __bpf_map_get(f);
489 map = bpf_map_inc(map, true);
495 /* map_idr_lock should have been held */
496 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
501 refold = __atomic_add_unless(&map->refcnt, 1, 0);
503 if (refold >= BPF_MAX_REFCNT) {
504 __bpf_map_put(map, false);
505 return ERR_PTR(-EBUSY);
509 return ERR_PTR(-ENOENT);
512 atomic_inc(&map->usercnt);
517 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
522 /* last field in 'union bpf_attr' used by this command */
523 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
525 static int map_lookup_elem(union bpf_attr *attr)
527 void __user *ukey = u64_to_user_ptr(attr->key);
528 void __user *uvalue = u64_to_user_ptr(attr->value);
529 int ufd = attr->map_fd;
531 void *key, *value, *ptr;
536 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
540 map = __bpf_map_get(f);
544 if (!(f.file->f_mode & FMODE_CAN_READ)) {
549 key = memdup_user(ukey, map->key_size);
555 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
556 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
557 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
558 value_size = round_up(map->value_size, 8) * num_possible_cpus();
559 else if (IS_FD_MAP(map))
560 value_size = sizeof(u32);
562 value_size = map->value_size;
565 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
569 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
570 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
571 err = bpf_percpu_hash_copy(map, key, value);
572 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
573 err = bpf_percpu_array_copy(map, key, value);
574 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
575 err = bpf_stackmap_copy(map, key, value);
576 } else if (IS_FD_ARRAY(map)) {
577 err = bpf_fd_array_map_lookup_elem(map, key, value);
578 } else if (IS_FD_HASH(map)) {
579 err = bpf_fd_htab_map_lookup_elem(map, key, value);
582 ptr = map->ops->map_lookup_elem(map, key);
584 memcpy(value, ptr, value_size);
586 err = ptr ? 0 : -ENOENT;
593 if (copy_to_user(uvalue, value, value_size) != 0)
596 trace_bpf_map_lookup_elem(map, ufd, key, value);
608 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
610 static int map_update_elem(union bpf_attr *attr)
612 void __user *ukey = u64_to_user_ptr(attr->key);
613 void __user *uvalue = u64_to_user_ptr(attr->value);
614 int ufd = attr->map_fd;
621 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
625 map = __bpf_map_get(f);
629 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
634 key = memdup_user(ukey, map->key_size);
640 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
641 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
642 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
643 value_size = round_up(map->value_size, 8) * num_possible_cpus();
645 value_size = map->value_size;
648 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
653 if (copy_from_user(value, uvalue, value_size) != 0)
656 /* Need to create a kthread, thus must support schedule */
657 if (map->map_type == BPF_MAP_TYPE_CPUMAP) {
658 err = map->ops->map_update_elem(map, key, value, attr->flags);
662 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
663 * inside bpf map update or delete otherwise deadlocks are possible
666 __this_cpu_inc(bpf_prog_active);
667 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
668 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
669 err = bpf_percpu_hash_update(map, key, value, attr->flags);
670 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
671 err = bpf_percpu_array_update(map, key, value, attr->flags);
672 } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
673 map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
674 map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY ||
675 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) {
677 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
680 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
682 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
687 err = map->ops->map_update_elem(map, key, value, attr->flags);
690 __this_cpu_dec(bpf_prog_active);
694 trace_bpf_map_update_elem(map, ufd, key, value);
704 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
706 static int map_delete_elem(union bpf_attr *attr)
708 void __user *ukey = u64_to_user_ptr(attr->key);
709 int ufd = attr->map_fd;
715 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
719 map = __bpf_map_get(f);
723 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
728 key = memdup_user(ukey, map->key_size);
735 __this_cpu_inc(bpf_prog_active);
737 err = map->ops->map_delete_elem(map, key);
739 __this_cpu_dec(bpf_prog_active);
743 trace_bpf_map_delete_elem(map, ufd, key);
750 /* last field in 'union bpf_attr' used by this command */
751 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
753 static int map_get_next_key(union bpf_attr *attr)
755 void __user *ukey = u64_to_user_ptr(attr->key);
756 void __user *unext_key = u64_to_user_ptr(attr->next_key);
757 int ufd = attr->map_fd;
759 void *key, *next_key;
763 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
767 map = __bpf_map_get(f);
771 if (!(f.file->f_mode & FMODE_CAN_READ)) {
777 key = memdup_user(ukey, map->key_size);
787 next_key = kmalloc(map->key_size, GFP_USER);
792 err = map->ops->map_get_next_key(map, key, next_key);
798 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
801 trace_bpf_map_next_key(map, ufd, key, next_key);
813 static const struct bpf_prog_ops * const bpf_prog_types[] = {
814 #define BPF_PROG_TYPE(_id, _name) \
815 [_id] = & _name ## _prog_ops,
816 #define BPF_MAP_TYPE(_id, _ops)
817 #include <linux/bpf_types.h>
822 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
824 if (type >= ARRAY_SIZE(bpf_prog_types) || !bpf_prog_types[type])
827 if (!bpf_prog_is_dev_bound(prog->aux))
828 prog->aux->ops = bpf_prog_types[type];
830 prog->aux->ops = &bpf_offload_prog_ops;
835 /* drop refcnt on maps used by eBPF program and free auxilary data */
836 static void free_used_maps(struct bpf_prog_aux *aux)
840 for (i = 0; i < aux->used_map_cnt; i++)
841 bpf_map_put(aux->used_maps[i]);
843 kfree(aux->used_maps);
846 int __bpf_prog_charge(struct user_struct *user, u32 pages)
848 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
849 unsigned long user_bufs;
852 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
853 if (user_bufs > memlock_limit) {
854 atomic_long_sub(pages, &user->locked_vm);
862 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
865 atomic_long_sub(pages, &user->locked_vm);
868 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
870 struct user_struct *user = get_current_user();
873 ret = __bpf_prog_charge(user, prog->pages);
879 prog->aux->user = user;
883 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
885 struct user_struct *user = prog->aux->user;
887 __bpf_prog_uncharge(user, prog->pages);
891 static int bpf_prog_alloc_id(struct bpf_prog *prog)
895 spin_lock_bh(&prog_idr_lock);
896 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
899 spin_unlock_bh(&prog_idr_lock);
901 /* id is in [1, INT_MAX) */
902 if (WARN_ON_ONCE(!id))
905 return id > 0 ? 0 : id;
908 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
910 /* cBPF to eBPF migrations are currently not in the idr store.
911 * Offloaded programs are removed from the store when their device
912 * disappears - even if someone grabs an fd to them they are unusable,
913 * simply waiting for refcnt to drop to be freed.
919 spin_lock_bh(&prog_idr_lock);
921 __acquire(&prog_idr_lock);
923 idr_remove(&prog_idr, prog->aux->id);
927 spin_unlock_bh(&prog_idr_lock);
929 __release(&prog_idr_lock);
932 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
934 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
937 bpf_prog_uncharge_memlock(aux->prog);
938 security_bpf_prog_free(aux);
939 bpf_prog_free(aux->prog);
942 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
944 if (atomic_dec_and_test(&prog->aux->refcnt)) {
947 trace_bpf_prog_put_rcu(prog);
948 /* bpf_prog_free_id() must be called first */
949 bpf_prog_free_id(prog, do_idr_lock);
951 for (i = 0; i < prog->aux->func_cnt; i++)
952 bpf_prog_kallsyms_del(prog->aux->func[i]);
953 bpf_prog_kallsyms_del(prog);
955 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
959 void bpf_prog_put(struct bpf_prog *prog)
961 __bpf_prog_put(prog, true);
963 EXPORT_SYMBOL_GPL(bpf_prog_put);
965 static int bpf_prog_release(struct inode *inode, struct file *filp)
967 struct bpf_prog *prog = filp->private_data;
973 #ifdef CONFIG_PROC_FS
974 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
976 const struct bpf_prog *prog = filp->private_data;
977 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
979 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
988 prog->pages * 1ULL << PAGE_SHIFT);
992 const struct file_operations bpf_prog_fops = {
993 #ifdef CONFIG_PROC_FS
994 .show_fdinfo = bpf_prog_show_fdinfo,
996 .release = bpf_prog_release,
997 .read = bpf_dummy_read,
998 .write = bpf_dummy_write,
1001 int bpf_prog_new_fd(struct bpf_prog *prog)
1005 ret = security_bpf_prog(prog);
1009 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1010 O_RDWR | O_CLOEXEC);
1013 static struct bpf_prog *____bpf_prog_get(struct fd f)
1016 return ERR_PTR(-EBADF);
1017 if (f.file->f_op != &bpf_prog_fops) {
1019 return ERR_PTR(-EINVAL);
1022 return f.file->private_data;
1025 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1027 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1028 atomic_sub(i, &prog->aux->refcnt);
1029 return ERR_PTR(-EBUSY);
1033 EXPORT_SYMBOL_GPL(bpf_prog_add);
1035 void bpf_prog_sub(struct bpf_prog *prog, int i)
1037 /* Only to be used for undoing previous bpf_prog_add() in some
1038 * error path. We still know that another entity in our call
1039 * path holds a reference to the program, thus atomic_sub() can
1040 * be safely used in such cases!
1042 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1044 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1046 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1048 return bpf_prog_add(prog, 1);
1050 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1052 /* prog_idr_lock should have been held */
1053 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1057 refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
1059 if (refold >= BPF_MAX_REFCNT) {
1060 __bpf_prog_put(prog, false);
1061 return ERR_PTR(-EBUSY);
1065 return ERR_PTR(-ENOENT);
1069 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1071 static bool bpf_prog_get_ok(struct bpf_prog *prog,
1072 enum bpf_prog_type *attach_type, bool attach_drv)
1074 /* not an attachment, just a refcount inc, always allow */
1078 if (prog->type != *attach_type)
1080 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1086 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1089 struct fd f = fdget(ufd);
1090 struct bpf_prog *prog;
1092 prog = ____bpf_prog_get(f);
1095 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1096 prog = ERR_PTR(-EINVAL);
1100 prog = bpf_prog_inc(prog);
1106 struct bpf_prog *bpf_prog_get(u32 ufd)
1108 return __bpf_prog_get(ufd, NULL, false);
1111 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1114 struct bpf_prog *prog = __bpf_prog_get(ufd, &type, attach_drv);
1117 trace_bpf_prog_get_type(prog);
1120 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1122 /* last field in 'union bpf_attr' used by this command */
1123 #define BPF_PROG_LOAD_LAST_FIELD prog_ifindex
1125 static int bpf_prog_load(union bpf_attr *attr)
1127 enum bpf_prog_type type = attr->prog_type;
1128 struct bpf_prog *prog;
1133 if (CHECK_ATTR(BPF_PROG_LOAD))
1136 if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1139 /* copy eBPF program license from user space */
1140 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1141 sizeof(license) - 1) < 0)
1143 license[sizeof(license) - 1] = 0;
1145 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1146 is_gpl = license_is_gpl_compatible(license);
1148 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1151 if (type == BPF_PROG_TYPE_KPROBE &&
1152 attr->kern_version != LINUX_VERSION_CODE)
1155 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1156 type != BPF_PROG_TYPE_CGROUP_SKB &&
1157 !capable(CAP_SYS_ADMIN))
1160 /* plain bpf_prog allocation */
1161 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1165 prog->aux->offload_requested = !!attr->prog_ifindex;
1167 err = security_bpf_prog_alloc(prog->aux);
1169 goto free_prog_nouncharge;
1171 err = bpf_prog_charge_memlock(prog);
1175 prog->len = attr->insn_cnt;
1178 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1179 bpf_prog_insn_size(prog)) != 0)
1182 prog->orig_prog = NULL;
1185 atomic_set(&prog->aux->refcnt, 1);
1186 prog->gpl_compatible = is_gpl ? 1 : 0;
1188 if (bpf_prog_is_dev_bound(prog->aux)) {
1189 err = bpf_prog_offload_init(prog, attr);
1194 /* find program type: socket_filter vs tracing_filter */
1195 err = find_prog_type(type, prog);
1199 prog->aux->load_time = ktime_get_boot_ns();
1200 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1204 /* run eBPF verifier */
1205 err = bpf_check(&prog, attr);
1207 goto free_used_maps;
1209 /* eBPF program is ready to be JITed */
1210 if (!prog->bpf_func)
1211 prog = bpf_prog_select_runtime(prog, &err);
1213 goto free_used_maps;
1215 err = bpf_prog_alloc_id(prog);
1217 goto free_used_maps;
1219 err = bpf_prog_new_fd(prog);
1221 /* failed to allocate fd.
1222 * bpf_prog_put() is needed because the above
1223 * bpf_prog_alloc_id() has published the prog
1224 * to the userspace and the userspace may
1225 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1231 bpf_prog_kallsyms_add(prog);
1232 trace_bpf_prog_load(prog, err);
1236 free_used_maps(prog->aux);
1238 bpf_prog_uncharge_memlock(prog);
1240 security_bpf_prog_free(prog->aux);
1241 free_prog_nouncharge:
1242 bpf_prog_free(prog);
1246 #define BPF_OBJ_LAST_FIELD file_flags
1248 static int bpf_obj_pin(const union bpf_attr *attr)
1250 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1253 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1256 static int bpf_obj_get(const union bpf_attr *attr)
1258 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1259 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1262 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1266 #ifdef CONFIG_CGROUP_BPF
1268 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1270 static int sockmap_get_from_fd(const union bpf_attr *attr, bool attach)
1272 struct bpf_prog *prog = NULL;
1273 int ufd = attr->target_fd;
1274 struct bpf_map *map;
1279 map = __bpf_map_get(f);
1281 return PTR_ERR(map);
1284 prog = bpf_prog_get_type(attr->attach_bpf_fd,
1285 BPF_PROG_TYPE_SK_SKB);
1288 return PTR_ERR(prog);
1292 err = sock_map_prog(map, prog, attr->attach_type);
1304 #define BPF_F_ATTACH_MASK \
1305 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1307 static int bpf_prog_attach(const union bpf_attr *attr)
1309 enum bpf_prog_type ptype;
1310 struct bpf_prog *prog;
1311 struct cgroup *cgrp;
1314 if (!capable(CAP_NET_ADMIN))
1317 if (CHECK_ATTR(BPF_PROG_ATTACH))
1320 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1323 switch (attr->attach_type) {
1324 case BPF_CGROUP_INET_INGRESS:
1325 case BPF_CGROUP_INET_EGRESS:
1326 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1328 case BPF_CGROUP_INET_SOCK_CREATE:
1329 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1331 case BPF_CGROUP_SOCK_OPS:
1332 ptype = BPF_PROG_TYPE_SOCK_OPS;
1334 case BPF_CGROUP_DEVICE:
1335 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1337 case BPF_SK_SKB_STREAM_PARSER:
1338 case BPF_SK_SKB_STREAM_VERDICT:
1339 return sockmap_get_from_fd(attr, true);
1344 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1346 return PTR_ERR(prog);
1348 cgrp = cgroup_get_from_fd(attr->target_fd);
1351 return PTR_ERR(cgrp);
1354 ret = cgroup_bpf_attach(cgrp, prog, attr->attach_type,
1355 attr->attach_flags);
1363 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1365 static int bpf_prog_detach(const union bpf_attr *attr)
1367 enum bpf_prog_type ptype;
1368 struct bpf_prog *prog;
1369 struct cgroup *cgrp;
1372 if (!capable(CAP_NET_ADMIN))
1375 if (CHECK_ATTR(BPF_PROG_DETACH))
1378 switch (attr->attach_type) {
1379 case BPF_CGROUP_INET_INGRESS:
1380 case BPF_CGROUP_INET_EGRESS:
1381 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1383 case BPF_CGROUP_INET_SOCK_CREATE:
1384 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1386 case BPF_CGROUP_SOCK_OPS:
1387 ptype = BPF_PROG_TYPE_SOCK_OPS;
1389 case BPF_CGROUP_DEVICE:
1390 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1392 case BPF_SK_SKB_STREAM_PARSER:
1393 case BPF_SK_SKB_STREAM_VERDICT:
1394 return sockmap_get_from_fd(attr, false);
1399 cgrp = cgroup_get_from_fd(attr->target_fd);
1401 return PTR_ERR(cgrp);
1403 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1407 ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type, 0);
1414 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1416 static int bpf_prog_query(const union bpf_attr *attr,
1417 union bpf_attr __user *uattr)
1419 struct cgroup *cgrp;
1422 if (!capable(CAP_NET_ADMIN))
1424 if (CHECK_ATTR(BPF_PROG_QUERY))
1426 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1429 switch (attr->query.attach_type) {
1430 case BPF_CGROUP_INET_INGRESS:
1431 case BPF_CGROUP_INET_EGRESS:
1432 case BPF_CGROUP_INET_SOCK_CREATE:
1433 case BPF_CGROUP_SOCK_OPS:
1434 case BPF_CGROUP_DEVICE:
1439 cgrp = cgroup_get_from_fd(attr->query.target_fd);
1441 return PTR_ERR(cgrp);
1442 ret = cgroup_bpf_query(cgrp, attr, uattr);
1446 #endif /* CONFIG_CGROUP_BPF */
1448 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1450 static int bpf_prog_test_run(const union bpf_attr *attr,
1451 union bpf_attr __user *uattr)
1453 struct bpf_prog *prog;
1454 int ret = -ENOTSUPP;
1456 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1459 prog = bpf_prog_get(attr->test.prog_fd);
1461 return PTR_ERR(prog);
1463 if (prog->aux->ops->test_run)
1464 ret = prog->aux->ops->test_run(prog, attr, uattr);
1470 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1472 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1473 union bpf_attr __user *uattr,
1477 u32 next_id = attr->start_id;
1480 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1483 if (!capable(CAP_SYS_ADMIN))
1488 if (!idr_get_next(idr, &next_id))
1490 spin_unlock_bh(lock);
1493 err = put_user(next_id, &uattr->next_id);
1498 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1500 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1502 struct bpf_prog *prog;
1503 u32 id = attr->prog_id;
1506 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1509 if (!capable(CAP_SYS_ADMIN))
1512 spin_lock_bh(&prog_idr_lock);
1513 prog = idr_find(&prog_idr, id);
1515 prog = bpf_prog_inc_not_zero(prog);
1517 prog = ERR_PTR(-ENOENT);
1518 spin_unlock_bh(&prog_idr_lock);
1521 return PTR_ERR(prog);
1523 fd = bpf_prog_new_fd(prog);
1530 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1532 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1534 struct bpf_map *map;
1535 u32 id = attr->map_id;
1539 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1540 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1543 if (!capable(CAP_SYS_ADMIN))
1546 f_flags = bpf_get_file_flag(attr->open_flags);
1550 spin_lock_bh(&map_idr_lock);
1551 map = idr_find(&map_idr, id);
1553 map = bpf_map_inc_not_zero(map, true);
1555 map = ERR_PTR(-ENOENT);
1556 spin_unlock_bh(&map_idr_lock);
1559 return PTR_ERR(map);
1561 fd = bpf_map_new_fd(map, f_flags);
1568 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1573 for (i = 0; i < prog->aux->used_map_cnt; i++)
1574 if (prog->aux->used_maps[i] == (void *)addr)
1575 return prog->aux->used_maps[i];
1579 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1581 const struct bpf_map *map;
1582 struct bpf_insn *insns;
1586 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1591 for (i = 0; i < prog->len; i++) {
1592 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1593 insns[i].code = BPF_JMP | BPF_CALL;
1594 insns[i].imm = BPF_FUNC_tail_call;
1597 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1598 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1599 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1600 insns[i].code = BPF_JMP | BPF_CALL;
1601 if (!bpf_dump_raw_ok())
1606 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1609 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1610 map = bpf_map_from_imm(prog, imm);
1612 insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1613 insns[i].imm = map->id;
1614 insns[i + 1].imm = 0;
1618 if (!bpf_dump_raw_ok() &&
1619 imm == (unsigned long)prog->aux) {
1621 insns[i + 1].imm = 0;
1629 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1630 const union bpf_attr *attr,
1631 union bpf_attr __user *uattr)
1633 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1634 struct bpf_prog_info info = {};
1635 u32 info_len = attr->info.info_len;
1636 char __user *uinsns;
1640 err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1643 info_len = min_t(u32, sizeof(info), info_len);
1645 if (copy_from_user(&info, uinfo, info_len))
1648 info.type = prog->type;
1649 info.id = prog->aux->id;
1650 info.load_time = prog->aux->load_time;
1651 info.created_by_uid = from_kuid_munged(current_user_ns(),
1652 prog->aux->user->uid);
1654 memcpy(info.tag, prog->tag, sizeof(prog->tag));
1655 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1657 ulen = info.nr_map_ids;
1658 info.nr_map_ids = prog->aux->used_map_cnt;
1659 ulen = min_t(u32, info.nr_map_ids, ulen);
1661 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1664 for (i = 0; i < ulen; i++)
1665 if (put_user(prog->aux->used_maps[i]->id,
1670 if (!capable(CAP_SYS_ADMIN)) {
1671 info.jited_prog_len = 0;
1672 info.xlated_prog_len = 0;
1676 ulen = info.jited_prog_len;
1677 info.jited_prog_len = prog->jited_len;
1678 if (info.jited_prog_len && ulen) {
1679 if (bpf_dump_raw_ok()) {
1680 uinsns = u64_to_user_ptr(info.jited_prog_insns);
1681 ulen = min_t(u32, info.jited_prog_len, ulen);
1682 if (copy_to_user(uinsns, prog->bpf_func, ulen))
1685 info.jited_prog_insns = 0;
1689 ulen = info.xlated_prog_len;
1690 info.xlated_prog_len = bpf_prog_insn_size(prog);
1691 if (info.xlated_prog_len && ulen) {
1692 struct bpf_insn *insns_sanitized;
1695 if (prog->blinded && !bpf_dump_raw_ok()) {
1696 info.xlated_prog_insns = 0;
1699 insns_sanitized = bpf_insn_prepare_dump(prog);
1700 if (!insns_sanitized)
1702 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1703 ulen = min_t(u32, info.xlated_prog_len, ulen);
1704 fault = copy_to_user(uinsns, insns_sanitized, ulen);
1705 kfree(insns_sanitized);
1710 if (bpf_prog_is_dev_bound(prog->aux)) {
1711 err = bpf_prog_offload_info_fill(&info, prog);
1717 if (copy_to_user(uinfo, &info, info_len) ||
1718 put_user(info_len, &uattr->info.info_len))
1724 static int bpf_map_get_info_by_fd(struct bpf_map *map,
1725 const union bpf_attr *attr,
1726 union bpf_attr __user *uattr)
1728 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1729 struct bpf_map_info info = {};
1730 u32 info_len = attr->info.info_len;
1733 err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1736 info_len = min_t(u32, sizeof(info), info_len);
1738 info.type = map->map_type;
1740 info.key_size = map->key_size;
1741 info.value_size = map->value_size;
1742 info.max_entries = map->max_entries;
1743 info.map_flags = map->map_flags;
1744 memcpy(info.name, map->name, sizeof(map->name));
1746 if (copy_to_user(uinfo, &info, info_len) ||
1747 put_user(info_len, &uattr->info.info_len))
1753 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
1755 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
1756 union bpf_attr __user *uattr)
1758 int ufd = attr->info.bpf_fd;
1762 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
1769 if (f.file->f_op == &bpf_prog_fops)
1770 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
1772 else if (f.file->f_op == &bpf_map_fops)
1773 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
1782 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
1784 union bpf_attr attr = {};
1787 if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
1790 err = check_uarg_tail_zero(uattr, sizeof(attr), size);
1793 size = min_t(u32, size, sizeof(attr));
1795 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
1796 if (copy_from_user(&attr, uattr, size) != 0)
1799 err = security_bpf(cmd, &attr, size);
1804 case BPF_MAP_CREATE:
1805 err = map_create(&attr);
1807 case BPF_MAP_LOOKUP_ELEM:
1808 err = map_lookup_elem(&attr);
1810 case BPF_MAP_UPDATE_ELEM:
1811 err = map_update_elem(&attr);
1813 case BPF_MAP_DELETE_ELEM:
1814 err = map_delete_elem(&attr);
1816 case BPF_MAP_GET_NEXT_KEY:
1817 err = map_get_next_key(&attr);
1820 err = bpf_prog_load(&attr);
1823 err = bpf_obj_pin(&attr);
1826 err = bpf_obj_get(&attr);
1828 #ifdef CONFIG_CGROUP_BPF
1829 case BPF_PROG_ATTACH:
1830 err = bpf_prog_attach(&attr);
1832 case BPF_PROG_DETACH:
1833 err = bpf_prog_detach(&attr);
1835 case BPF_PROG_QUERY:
1836 err = bpf_prog_query(&attr, uattr);
1839 case BPF_PROG_TEST_RUN:
1840 err = bpf_prog_test_run(&attr, uattr);
1842 case BPF_PROG_GET_NEXT_ID:
1843 err = bpf_obj_get_next_id(&attr, uattr,
1844 &prog_idr, &prog_idr_lock);
1846 case BPF_MAP_GET_NEXT_ID:
1847 err = bpf_obj_get_next_id(&attr, uattr,
1848 &map_idr, &map_idr_lock);
1850 case BPF_PROG_GET_FD_BY_ID:
1851 err = bpf_prog_get_fd_by_id(&attr);
1853 case BPF_MAP_GET_FD_BY_ID:
1854 err = bpf_map_get_fd_by_id(&attr);
1856 case BPF_OBJ_GET_INFO_BY_FD:
1857 err = bpf_obj_get_info_by_fd(&attr, uattr);