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 const struct bpf_map_ops bpf_map_offload_ops
= {
98 .map_alloc
= bpf_map_offload_map_alloc
,
99 .map_free
= bpf_map_offload_map_free
,
102 static struct bpf_map
*find_and_alloc_map(union bpf_attr
*attr
)
104 const struct bpf_map_ops
*ops
;
108 if (attr
->map_type
>= ARRAY_SIZE(bpf_map_types
))
109 return ERR_PTR(-EINVAL
);
110 ops
= bpf_map_types
[attr
->map_type
];
112 return ERR_PTR(-EINVAL
);
114 if (ops
->map_alloc_check
) {
115 err
= ops
->map_alloc_check(attr
);
119 if (attr
->map_ifindex
)
120 ops
= &bpf_map_offload_ops
;
121 map
= ops
->map_alloc(attr
);
125 map
->map_type
= attr
->map_type
;
129 void *bpf_map_area_alloc(size_t size
, int numa_node
)
131 /* We definitely need __GFP_NORETRY, so OOM killer doesn't
132 * trigger under memory pressure as we really just want to
135 const gfp_t flags
= __GFP_NOWARN
| __GFP_NORETRY
| __GFP_ZERO
;
138 if (size
<= (PAGE_SIZE
<< PAGE_ALLOC_COSTLY_ORDER
)) {
139 area
= kmalloc_node(size
, GFP_USER
| flags
, numa_node
);
144 return __vmalloc_node_flags_caller(size
, numa_node
, GFP_KERNEL
| flags
,
145 __builtin_return_address(0));
148 void bpf_map_area_free(void *area
)
153 void bpf_map_init_from_attr(struct bpf_map
*map
, union bpf_attr
*attr
)
155 map
->map_type
= attr
->map_type
;
156 map
->key_size
= attr
->key_size
;
157 map
->value_size
= attr
->value_size
;
158 map
->max_entries
= attr
->max_entries
;
159 map
->map_flags
= attr
->map_flags
;
160 map
->numa_node
= bpf_map_attr_numa_node(attr
);
163 int bpf_map_precharge_memlock(u32 pages
)
165 struct user_struct
*user
= get_current_user();
166 unsigned long memlock_limit
, cur
;
168 memlock_limit
= rlimit(RLIMIT_MEMLOCK
) >> PAGE_SHIFT
;
169 cur
= atomic_long_read(&user
->locked_vm
);
171 if (cur
+ pages
> memlock_limit
)
176 static int bpf_map_charge_memlock(struct bpf_map
*map
)
178 struct user_struct
*user
= get_current_user();
179 unsigned long memlock_limit
;
181 memlock_limit
= rlimit(RLIMIT_MEMLOCK
) >> PAGE_SHIFT
;
183 atomic_long_add(map
->pages
, &user
->locked_vm
);
185 if (atomic_long_read(&user
->locked_vm
) > memlock_limit
) {
186 atomic_long_sub(map
->pages
, &user
->locked_vm
);
194 static void bpf_map_uncharge_memlock(struct bpf_map
*map
)
196 struct user_struct
*user
= map
->user
;
198 atomic_long_sub(map
->pages
, &user
->locked_vm
);
202 static int bpf_map_alloc_id(struct bpf_map
*map
)
206 idr_preload(GFP_KERNEL
);
207 spin_lock_bh(&map_idr_lock
);
208 id
= idr_alloc_cyclic(&map_idr
, map
, 1, INT_MAX
, GFP_ATOMIC
);
211 spin_unlock_bh(&map_idr_lock
);
214 if (WARN_ON_ONCE(!id
))
217 return id
> 0 ? 0 : id
;
220 void bpf_map_free_id(struct bpf_map
*map
, bool do_idr_lock
)
224 /* Offloaded maps are removed from the IDR store when their device
225 * disappears - even if someone holds an fd to them they are unusable,
226 * the memory is gone, all ops will fail; they are simply waiting for
227 * refcnt to drop to be freed.
233 spin_lock_irqsave(&map_idr_lock
, flags
);
235 __acquire(&map_idr_lock
);
237 idr_remove(&map_idr
, map
->id
);
241 spin_unlock_irqrestore(&map_idr_lock
, flags
);
243 __release(&map_idr_lock
);
246 /* called from workqueue */
247 static void bpf_map_free_deferred(struct work_struct
*work
)
249 struct bpf_map
*map
= container_of(work
, struct bpf_map
, work
);
251 bpf_map_uncharge_memlock(map
);
252 security_bpf_map_free(map
);
253 /* implementation dependent freeing */
254 map
->ops
->map_free(map
);
257 static void bpf_map_put_uref(struct bpf_map
*map
)
259 if (atomic_dec_and_test(&map
->usercnt
)) {
260 if (map
->ops
->map_release_uref
)
261 map
->ops
->map_release_uref(map
);
265 /* decrement map refcnt and schedule it for freeing via workqueue
266 * (unrelying map implementation ops->map_free() might sleep)
268 static void __bpf_map_put(struct bpf_map
*map
, bool do_idr_lock
)
270 if (atomic_dec_and_test(&map
->refcnt
)) {
271 /* bpf_map_free_id() must be called first */
272 bpf_map_free_id(map
, do_idr_lock
);
273 INIT_WORK(&map
->work
, bpf_map_free_deferred
);
274 schedule_work(&map
->work
);
278 void bpf_map_put(struct bpf_map
*map
)
280 __bpf_map_put(map
, true);
283 void bpf_map_put_with_uref(struct bpf_map
*map
)
285 bpf_map_put_uref(map
);
289 static int bpf_map_release(struct inode
*inode
, struct file
*filp
)
291 struct bpf_map
*map
= filp
->private_data
;
293 if (map
->ops
->map_release
)
294 map
->ops
->map_release(map
, filp
);
296 bpf_map_put_with_uref(map
);
300 #ifdef CONFIG_PROC_FS
301 static void bpf_map_show_fdinfo(struct seq_file
*m
, struct file
*filp
)
303 const struct bpf_map
*map
= filp
->private_data
;
304 const struct bpf_array
*array
;
305 u32 owner_prog_type
= 0;
308 if (map
->map_type
== BPF_MAP_TYPE_PROG_ARRAY
) {
309 array
= container_of(map
, struct bpf_array
, map
);
310 owner_prog_type
= array
->owner_prog_type
;
311 owner_jited
= array
->owner_jited
;
326 map
->pages
* 1ULL << PAGE_SHIFT
);
328 if (owner_prog_type
) {
329 seq_printf(m
, "owner_prog_type:\t%u\n",
331 seq_printf(m
, "owner_jited:\t%u\n",
337 static ssize_t
bpf_dummy_read(struct file
*filp
, char __user
*buf
, size_t siz
,
340 /* We need this handler such that alloc_file() enables
341 * f_mode with FMODE_CAN_READ.
346 static ssize_t
bpf_dummy_write(struct file
*filp
, const char __user
*buf
,
347 size_t siz
, loff_t
*ppos
)
349 /* We need this handler such that alloc_file() enables
350 * f_mode with FMODE_CAN_WRITE.
355 const struct file_operations bpf_map_fops
= {
356 #ifdef CONFIG_PROC_FS
357 .show_fdinfo
= bpf_map_show_fdinfo
,
359 .release
= bpf_map_release
,
360 .read
= bpf_dummy_read
,
361 .write
= bpf_dummy_write
,
364 int bpf_map_new_fd(struct bpf_map
*map
, int flags
)
368 ret
= security_bpf_map(map
, OPEN_FMODE(flags
));
372 return anon_inode_getfd("bpf-map", &bpf_map_fops
, map
,
376 int bpf_get_file_flag(int flags
)
378 if ((flags
& BPF_F_RDONLY
) && (flags
& BPF_F_WRONLY
))
380 if (flags
& BPF_F_RDONLY
)
382 if (flags
& BPF_F_WRONLY
)
387 /* helper macro to check that unused fields 'union bpf_attr' are zero */
388 #define CHECK_ATTR(CMD) \
389 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
390 sizeof(attr->CMD##_LAST_FIELD), 0, \
392 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
393 sizeof(attr->CMD##_LAST_FIELD)) != NULL
395 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
396 * Return 0 on success and < 0 on error.
398 static int bpf_obj_name_cpy(char *dst
, const char *src
)
400 const char *end
= src
+ BPF_OBJ_NAME_LEN
;
402 memset(dst
, 0, BPF_OBJ_NAME_LEN
);
404 /* Copy all isalnum() and '_' char */
405 while (src
< end
&& *src
) {
406 if (!isalnum(*src
) && *src
!= '_')
411 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
418 #define BPF_MAP_CREATE_LAST_FIELD map_ifindex
419 /* called via syscall */
420 static int map_create(union bpf_attr
*attr
)
422 int numa_node
= bpf_map_attr_numa_node(attr
);
427 err
= CHECK_ATTR(BPF_MAP_CREATE
);
431 f_flags
= bpf_get_file_flag(attr
->map_flags
);
435 if (numa_node
!= NUMA_NO_NODE
&&
436 ((unsigned int)numa_node
>= nr_node_ids
||
437 !node_online(numa_node
)))
440 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
441 map
= find_and_alloc_map(attr
);
445 err
= bpf_obj_name_cpy(map
->name
, attr
->map_name
);
447 goto free_map_nouncharge
;
449 atomic_set(&map
->refcnt
, 1);
450 atomic_set(&map
->usercnt
, 1);
452 err
= security_bpf_map_alloc(map
);
454 goto free_map_nouncharge
;
456 err
= bpf_map_charge_memlock(map
);
460 err
= bpf_map_alloc_id(map
);
464 err
= bpf_map_new_fd(map
, f_flags
);
466 /* failed to allocate fd.
467 * bpf_map_put() is needed because the above
468 * bpf_map_alloc_id() has published the map
469 * to the userspace and the userspace may
470 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
476 trace_bpf_map_create(map
, err
);
480 bpf_map_uncharge_memlock(map
);
482 security_bpf_map_free(map
);
484 map
->ops
->map_free(map
);
488 /* if error is returned, fd is released.
489 * On success caller should complete fd access with matching fdput()
491 struct bpf_map
*__bpf_map_get(struct fd f
)
494 return ERR_PTR(-EBADF
);
495 if (f
.file
->f_op
!= &bpf_map_fops
) {
497 return ERR_PTR(-EINVAL
);
500 return f
.file
->private_data
;
503 /* prog's and map's refcnt limit */
504 #define BPF_MAX_REFCNT 32768
506 struct bpf_map
*bpf_map_inc(struct bpf_map
*map
, bool uref
)
508 if (atomic_inc_return(&map
->refcnt
) > BPF_MAX_REFCNT
) {
509 atomic_dec(&map
->refcnt
);
510 return ERR_PTR(-EBUSY
);
513 atomic_inc(&map
->usercnt
);
517 struct bpf_map
*bpf_map_get_with_uref(u32 ufd
)
519 struct fd f
= fdget(ufd
);
522 map
= __bpf_map_get(f
);
526 map
= bpf_map_inc(map
, true);
532 /* map_idr_lock should have been held */
533 static struct bpf_map
*bpf_map_inc_not_zero(struct bpf_map
*map
,
538 refold
= __atomic_add_unless(&map
->refcnt
, 1, 0);
540 if (refold
>= BPF_MAX_REFCNT
) {
541 __bpf_map_put(map
, false);
542 return ERR_PTR(-EBUSY
);
546 return ERR_PTR(-ENOENT
);
549 atomic_inc(&map
->usercnt
);
554 int __weak
bpf_stackmap_copy(struct bpf_map
*map
, void *key
, void *value
)
559 /* last field in 'union bpf_attr' used by this command */
560 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
562 static int map_lookup_elem(union bpf_attr
*attr
)
564 void __user
*ukey
= u64_to_user_ptr(attr
->key
);
565 void __user
*uvalue
= u64_to_user_ptr(attr
->value
);
566 int ufd
= attr
->map_fd
;
568 void *key
, *value
, *ptr
;
573 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM
))
577 map
= __bpf_map_get(f
);
581 if (!(f
.file
->f_mode
& FMODE_CAN_READ
)) {
586 key
= memdup_user(ukey
, map
->key_size
);
592 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
||
593 map
->map_type
== BPF_MAP_TYPE_LRU_PERCPU_HASH
||
594 map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
)
595 value_size
= round_up(map
->value_size
, 8) * num_possible_cpus();
596 else if (IS_FD_MAP(map
))
597 value_size
= sizeof(u32
);
599 value_size
= map
->value_size
;
602 value
= kmalloc(value_size
, GFP_USER
| __GFP_NOWARN
);
606 if (bpf_map_is_dev_bound(map
)) {
607 err
= bpf_map_offload_lookup_elem(map
, key
, value
);
608 } else if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
||
609 map
->map_type
== BPF_MAP_TYPE_LRU_PERCPU_HASH
) {
610 err
= bpf_percpu_hash_copy(map
, key
, value
);
611 } else if (map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
) {
612 err
= bpf_percpu_array_copy(map
, key
, value
);
613 } else if (map
->map_type
== BPF_MAP_TYPE_STACK_TRACE
) {
614 err
= bpf_stackmap_copy(map
, key
, value
);
615 } else if (IS_FD_ARRAY(map
)) {
616 err
= bpf_fd_array_map_lookup_elem(map
, key
, value
);
617 } else if (IS_FD_HASH(map
)) {
618 err
= bpf_fd_htab_map_lookup_elem(map
, key
, value
);
621 ptr
= map
->ops
->map_lookup_elem(map
, key
);
623 memcpy(value
, ptr
, value_size
);
625 err
= ptr
? 0 : -ENOENT
;
632 if (copy_to_user(uvalue
, value
, value_size
) != 0)
635 trace_bpf_map_lookup_elem(map
, ufd
, key
, value
);
647 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
649 static int map_update_elem(union bpf_attr
*attr
)
651 void __user
*ukey
= u64_to_user_ptr(attr
->key
);
652 void __user
*uvalue
= u64_to_user_ptr(attr
->value
);
653 int ufd
= attr
->map_fd
;
660 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM
))
664 map
= __bpf_map_get(f
);
668 if (!(f
.file
->f_mode
& FMODE_CAN_WRITE
)) {
673 key
= memdup_user(ukey
, map
->key_size
);
679 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
||
680 map
->map_type
== BPF_MAP_TYPE_LRU_PERCPU_HASH
||
681 map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
)
682 value_size
= round_up(map
->value_size
, 8) * num_possible_cpus();
684 value_size
= map
->value_size
;
687 value
= kmalloc(value_size
, GFP_USER
| __GFP_NOWARN
);
692 if (copy_from_user(value
, uvalue
, value_size
) != 0)
695 /* Need to create a kthread, thus must support schedule */
696 if (bpf_map_is_dev_bound(map
)) {
697 err
= bpf_map_offload_update_elem(map
, key
, value
, attr
->flags
);
699 } else if (map
->map_type
== BPF_MAP_TYPE_CPUMAP
) {
700 err
= map
->ops
->map_update_elem(map
, key
, value
, attr
->flags
);
704 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
705 * inside bpf map update or delete otherwise deadlocks are possible
708 __this_cpu_inc(bpf_prog_active
);
709 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
||
710 map
->map_type
== BPF_MAP_TYPE_LRU_PERCPU_HASH
) {
711 err
= bpf_percpu_hash_update(map
, key
, value
, attr
->flags
);
712 } else if (map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
) {
713 err
= bpf_percpu_array_update(map
, key
, value
, attr
->flags
);
714 } else if (IS_FD_ARRAY(map
)) {
716 err
= bpf_fd_array_map_update_elem(map
, f
.file
, key
, value
,
719 } else if (map
->map_type
== BPF_MAP_TYPE_HASH_OF_MAPS
) {
721 err
= bpf_fd_htab_map_update_elem(map
, f
.file
, key
, value
,
726 err
= map
->ops
->map_update_elem(map
, key
, value
, attr
->flags
);
729 __this_cpu_dec(bpf_prog_active
);
733 trace_bpf_map_update_elem(map
, ufd
, key
, value
);
743 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
745 static int map_delete_elem(union bpf_attr
*attr
)
747 void __user
*ukey
= u64_to_user_ptr(attr
->key
);
748 int ufd
= attr
->map_fd
;
754 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM
))
758 map
= __bpf_map_get(f
);
762 if (!(f
.file
->f_mode
& FMODE_CAN_WRITE
)) {
767 key
= memdup_user(ukey
, map
->key_size
);
773 if (bpf_map_is_dev_bound(map
)) {
774 err
= bpf_map_offload_delete_elem(map
, key
);
779 __this_cpu_inc(bpf_prog_active
);
781 err
= map
->ops
->map_delete_elem(map
, key
);
783 __this_cpu_dec(bpf_prog_active
);
787 trace_bpf_map_delete_elem(map
, ufd
, key
);
794 /* last field in 'union bpf_attr' used by this command */
795 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
797 static int map_get_next_key(union bpf_attr
*attr
)
799 void __user
*ukey
= u64_to_user_ptr(attr
->key
);
800 void __user
*unext_key
= u64_to_user_ptr(attr
->next_key
);
801 int ufd
= attr
->map_fd
;
803 void *key
, *next_key
;
807 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY
))
811 map
= __bpf_map_get(f
);
815 if (!(f
.file
->f_mode
& FMODE_CAN_READ
)) {
821 key
= memdup_user(ukey
, map
->key_size
);
831 next_key
= kmalloc(map
->key_size
, GFP_USER
);
835 if (bpf_map_is_dev_bound(map
)) {
836 err
= bpf_map_offload_get_next_key(map
, key
, next_key
);
841 err
= map
->ops
->map_get_next_key(map
, key
, next_key
);
848 if (copy_to_user(unext_key
, next_key
, map
->key_size
) != 0)
851 trace_bpf_map_next_key(map
, ufd
, key
, next_key
);
863 static const struct bpf_prog_ops
* const bpf_prog_types
[] = {
864 #define BPF_PROG_TYPE(_id, _name) \
865 [_id] = & _name ## _prog_ops,
866 #define BPF_MAP_TYPE(_id, _ops)
867 #include <linux/bpf_types.h>
872 static int find_prog_type(enum bpf_prog_type type
, struct bpf_prog
*prog
)
874 if (type
>= ARRAY_SIZE(bpf_prog_types
) || !bpf_prog_types
[type
])
877 if (!bpf_prog_is_dev_bound(prog
->aux
))
878 prog
->aux
->ops
= bpf_prog_types
[type
];
880 prog
->aux
->ops
= &bpf_offload_prog_ops
;
885 /* drop refcnt on maps used by eBPF program and free auxilary data */
886 static void free_used_maps(struct bpf_prog_aux
*aux
)
890 for (i
= 0; i
< aux
->used_map_cnt
; i
++)
891 bpf_map_put(aux
->used_maps
[i
]);
893 kfree(aux
->used_maps
);
896 int __bpf_prog_charge(struct user_struct
*user
, u32 pages
)
898 unsigned long memlock_limit
= rlimit(RLIMIT_MEMLOCK
) >> PAGE_SHIFT
;
899 unsigned long user_bufs
;
902 user_bufs
= atomic_long_add_return(pages
, &user
->locked_vm
);
903 if (user_bufs
> memlock_limit
) {
904 atomic_long_sub(pages
, &user
->locked_vm
);
912 void __bpf_prog_uncharge(struct user_struct
*user
, u32 pages
)
915 atomic_long_sub(pages
, &user
->locked_vm
);
918 static int bpf_prog_charge_memlock(struct bpf_prog
*prog
)
920 struct user_struct
*user
= get_current_user();
923 ret
= __bpf_prog_charge(user
, prog
->pages
);
929 prog
->aux
->user
= user
;
933 static void bpf_prog_uncharge_memlock(struct bpf_prog
*prog
)
935 struct user_struct
*user
= prog
->aux
->user
;
937 __bpf_prog_uncharge(user
, prog
->pages
);
941 static int bpf_prog_alloc_id(struct bpf_prog
*prog
)
945 idr_preload(GFP_KERNEL
);
946 spin_lock_bh(&prog_idr_lock
);
947 id
= idr_alloc_cyclic(&prog_idr
, prog
, 1, INT_MAX
, GFP_ATOMIC
);
950 spin_unlock_bh(&prog_idr_lock
);
953 /* id is in [1, INT_MAX) */
954 if (WARN_ON_ONCE(!id
))
957 return id
> 0 ? 0 : id
;
960 void bpf_prog_free_id(struct bpf_prog
*prog
, bool do_idr_lock
)
962 /* cBPF to eBPF migrations are currently not in the idr store.
963 * Offloaded programs are removed from the store when their device
964 * disappears - even if someone grabs an fd to them they are unusable,
965 * simply waiting for refcnt to drop to be freed.
971 spin_lock_bh(&prog_idr_lock
);
973 __acquire(&prog_idr_lock
);
975 idr_remove(&prog_idr
, prog
->aux
->id
);
979 spin_unlock_bh(&prog_idr_lock
);
981 __release(&prog_idr_lock
);
984 static void __bpf_prog_put_rcu(struct rcu_head
*rcu
)
986 struct bpf_prog_aux
*aux
= container_of(rcu
, struct bpf_prog_aux
, rcu
);
989 bpf_prog_uncharge_memlock(aux
->prog
);
990 security_bpf_prog_free(aux
);
991 bpf_prog_free(aux
->prog
);
994 static void __bpf_prog_put(struct bpf_prog
*prog
, bool do_idr_lock
)
996 if (atomic_dec_and_test(&prog
->aux
->refcnt
)) {
999 trace_bpf_prog_put_rcu(prog
);
1000 /* bpf_prog_free_id() must be called first */
1001 bpf_prog_free_id(prog
, do_idr_lock
);
1003 for (i
= 0; i
< prog
->aux
->func_cnt
; i
++)
1004 bpf_prog_kallsyms_del(prog
->aux
->func
[i
]);
1005 bpf_prog_kallsyms_del(prog
);
1007 call_rcu(&prog
->aux
->rcu
, __bpf_prog_put_rcu
);
1011 void bpf_prog_put(struct bpf_prog
*prog
)
1013 __bpf_prog_put(prog
, true);
1015 EXPORT_SYMBOL_GPL(bpf_prog_put
);
1017 static int bpf_prog_release(struct inode
*inode
, struct file
*filp
)
1019 struct bpf_prog
*prog
= filp
->private_data
;
1025 #ifdef CONFIG_PROC_FS
1026 static void bpf_prog_show_fdinfo(struct seq_file
*m
, struct file
*filp
)
1028 const struct bpf_prog
*prog
= filp
->private_data
;
1029 char prog_tag
[sizeof(prog
->tag
) * 2 + 1] = { };
1031 bin2hex(prog_tag
, prog
->tag
, sizeof(prog
->tag
));
1040 prog
->pages
* 1ULL << PAGE_SHIFT
);
1044 const struct file_operations bpf_prog_fops
= {
1045 #ifdef CONFIG_PROC_FS
1046 .show_fdinfo
= bpf_prog_show_fdinfo
,
1048 .release
= bpf_prog_release
,
1049 .read
= bpf_dummy_read
,
1050 .write
= bpf_dummy_write
,
1053 int bpf_prog_new_fd(struct bpf_prog
*prog
)
1057 ret
= security_bpf_prog(prog
);
1061 return anon_inode_getfd("bpf-prog", &bpf_prog_fops
, prog
,
1062 O_RDWR
| O_CLOEXEC
);
1065 static struct bpf_prog
*____bpf_prog_get(struct fd f
)
1068 return ERR_PTR(-EBADF
);
1069 if (f
.file
->f_op
!= &bpf_prog_fops
) {
1071 return ERR_PTR(-EINVAL
);
1074 return f
.file
->private_data
;
1077 struct bpf_prog
*bpf_prog_add(struct bpf_prog
*prog
, int i
)
1079 if (atomic_add_return(i
, &prog
->aux
->refcnt
) > BPF_MAX_REFCNT
) {
1080 atomic_sub(i
, &prog
->aux
->refcnt
);
1081 return ERR_PTR(-EBUSY
);
1085 EXPORT_SYMBOL_GPL(bpf_prog_add
);
1087 void bpf_prog_sub(struct bpf_prog
*prog
, int i
)
1089 /* Only to be used for undoing previous bpf_prog_add() in some
1090 * error path. We still know that another entity in our call
1091 * path holds a reference to the program, thus atomic_sub() can
1092 * be safely used in such cases!
1094 WARN_ON(atomic_sub_return(i
, &prog
->aux
->refcnt
) == 0);
1096 EXPORT_SYMBOL_GPL(bpf_prog_sub
);
1098 struct bpf_prog
*bpf_prog_inc(struct bpf_prog
*prog
)
1100 return bpf_prog_add(prog
, 1);
1102 EXPORT_SYMBOL_GPL(bpf_prog_inc
);
1104 /* prog_idr_lock should have been held */
1105 struct bpf_prog
*bpf_prog_inc_not_zero(struct bpf_prog
*prog
)
1109 refold
= __atomic_add_unless(&prog
->aux
->refcnt
, 1, 0);
1111 if (refold
>= BPF_MAX_REFCNT
) {
1112 __bpf_prog_put(prog
, false);
1113 return ERR_PTR(-EBUSY
);
1117 return ERR_PTR(-ENOENT
);
1121 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero
);
1123 bool bpf_prog_get_ok(struct bpf_prog
*prog
,
1124 enum bpf_prog_type
*attach_type
, bool attach_drv
)
1126 /* not an attachment, just a refcount inc, always allow */
1130 if (prog
->type
!= *attach_type
)
1132 if (bpf_prog_is_dev_bound(prog
->aux
) && !attach_drv
)
1138 static struct bpf_prog
*__bpf_prog_get(u32 ufd
, enum bpf_prog_type
*attach_type
,
1141 struct fd f
= fdget(ufd
);
1142 struct bpf_prog
*prog
;
1144 prog
= ____bpf_prog_get(f
);
1147 if (!bpf_prog_get_ok(prog
, attach_type
, attach_drv
)) {
1148 prog
= ERR_PTR(-EINVAL
);
1152 prog
= bpf_prog_inc(prog
);
1158 struct bpf_prog
*bpf_prog_get(u32 ufd
)
1160 return __bpf_prog_get(ufd
, NULL
, false);
1163 struct bpf_prog
*bpf_prog_get_type_dev(u32 ufd
, enum bpf_prog_type type
,
1166 struct bpf_prog
*prog
= __bpf_prog_get(ufd
, &type
, attach_drv
);
1169 trace_bpf_prog_get_type(prog
);
1172 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev
);
1174 /* Initially all BPF programs could be loaded w/o specifying
1175 * expected_attach_type. Later for some of them specifying expected_attach_type
1176 * at load time became required so that program could be validated properly.
1177 * Programs of types that are allowed to be loaded both w/ and w/o (for
1178 * backward compatibility) expected_attach_type, should have the default attach
1179 * type assigned to expected_attach_type for the latter case, so that it can be
1180 * validated later at attach time.
1182 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1183 * prog type requires it but has some attach types that have to be backward
1186 static void bpf_prog_load_fixup_attach_type(union bpf_attr
*attr
)
1188 switch (attr
->prog_type
) {
1189 case BPF_PROG_TYPE_CGROUP_SOCK
:
1190 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1191 * exist so checking for non-zero is the way to go here.
1193 if (!attr
->expected_attach_type
)
1194 attr
->expected_attach_type
=
1195 BPF_CGROUP_INET_SOCK_CREATE
;
1201 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type
,
1202 enum bpf_attach_type expected_attach_type
)
1204 switch (prog_type
) {
1205 case BPF_PROG_TYPE_CGROUP_SOCK
:
1206 switch (expected_attach_type
) {
1207 case BPF_CGROUP_INET_SOCK_CREATE
:
1208 case BPF_CGROUP_INET4_POST_BIND
:
1209 case BPF_CGROUP_INET6_POST_BIND
:
1214 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR
:
1215 switch (expected_attach_type
) {
1216 case BPF_CGROUP_INET4_BIND
:
1217 case BPF_CGROUP_INET6_BIND
:
1218 case BPF_CGROUP_INET4_CONNECT
:
1219 case BPF_CGROUP_INET6_CONNECT
:
1229 /* last field in 'union bpf_attr' used by this command */
1230 #define BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1232 static int bpf_prog_load(union bpf_attr
*attr
)
1234 enum bpf_prog_type type
= attr
->prog_type
;
1235 struct bpf_prog
*prog
;
1240 if (CHECK_ATTR(BPF_PROG_LOAD
))
1243 if (attr
->prog_flags
& ~BPF_F_STRICT_ALIGNMENT
)
1246 /* copy eBPF program license from user space */
1247 if (strncpy_from_user(license
, u64_to_user_ptr(attr
->license
),
1248 sizeof(license
) - 1) < 0)
1250 license
[sizeof(license
) - 1] = 0;
1252 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1253 is_gpl
= license_is_gpl_compatible(license
);
1255 if (attr
->insn_cnt
== 0 || attr
->insn_cnt
> BPF_MAXINSNS
)
1258 if (type
== BPF_PROG_TYPE_KPROBE
&&
1259 attr
->kern_version
!= LINUX_VERSION_CODE
)
1262 if (type
!= BPF_PROG_TYPE_SOCKET_FILTER
&&
1263 type
!= BPF_PROG_TYPE_CGROUP_SKB
&&
1264 !capable(CAP_SYS_ADMIN
))
1267 bpf_prog_load_fixup_attach_type(attr
);
1268 if (bpf_prog_load_check_attach_type(type
, attr
->expected_attach_type
))
1271 /* plain bpf_prog allocation */
1272 prog
= bpf_prog_alloc(bpf_prog_size(attr
->insn_cnt
), GFP_USER
);
1276 prog
->expected_attach_type
= attr
->expected_attach_type
;
1278 prog
->aux
->offload_requested
= !!attr
->prog_ifindex
;
1280 err
= security_bpf_prog_alloc(prog
->aux
);
1282 goto free_prog_nouncharge
;
1284 err
= bpf_prog_charge_memlock(prog
);
1288 prog
->len
= attr
->insn_cnt
;
1291 if (copy_from_user(prog
->insns
, u64_to_user_ptr(attr
->insns
),
1292 bpf_prog_insn_size(prog
)) != 0)
1295 prog
->orig_prog
= NULL
;
1298 atomic_set(&prog
->aux
->refcnt
, 1);
1299 prog
->gpl_compatible
= is_gpl
? 1 : 0;
1301 if (bpf_prog_is_dev_bound(prog
->aux
)) {
1302 err
= bpf_prog_offload_init(prog
, attr
);
1307 /* find program type: socket_filter vs tracing_filter */
1308 err
= find_prog_type(type
, prog
);
1312 prog
->aux
->load_time
= ktime_get_boot_ns();
1313 err
= bpf_obj_name_cpy(prog
->aux
->name
, attr
->prog_name
);
1317 /* run eBPF verifier */
1318 err
= bpf_check(&prog
, attr
);
1320 goto free_used_maps
;
1322 /* eBPF program is ready to be JITed */
1323 if (!prog
->bpf_func
)
1324 prog
= bpf_prog_select_runtime(prog
, &err
);
1326 goto free_used_maps
;
1328 err
= bpf_prog_alloc_id(prog
);
1330 goto free_used_maps
;
1332 err
= bpf_prog_new_fd(prog
);
1334 /* failed to allocate fd.
1335 * bpf_prog_put() is needed because the above
1336 * bpf_prog_alloc_id() has published the prog
1337 * to the userspace and the userspace may
1338 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1344 bpf_prog_kallsyms_add(prog
);
1345 trace_bpf_prog_load(prog
, err
);
1349 free_used_maps(prog
->aux
);
1351 bpf_prog_uncharge_memlock(prog
);
1353 security_bpf_prog_free(prog
->aux
);
1354 free_prog_nouncharge
:
1355 bpf_prog_free(prog
);
1359 #define BPF_OBJ_LAST_FIELD file_flags
1361 static int bpf_obj_pin(const union bpf_attr
*attr
)
1363 if (CHECK_ATTR(BPF_OBJ
) || attr
->file_flags
!= 0)
1366 return bpf_obj_pin_user(attr
->bpf_fd
, u64_to_user_ptr(attr
->pathname
));
1369 static int bpf_obj_get(const union bpf_attr
*attr
)
1371 if (CHECK_ATTR(BPF_OBJ
) || attr
->bpf_fd
!= 0 ||
1372 attr
->file_flags
& ~BPF_OBJ_FLAG_MASK
)
1375 return bpf_obj_get_user(u64_to_user_ptr(attr
->pathname
),
1379 struct bpf_raw_tracepoint
{
1380 struct bpf_raw_event_map
*btp
;
1381 struct bpf_prog
*prog
;
1384 static int bpf_raw_tracepoint_release(struct inode
*inode
, struct file
*filp
)
1386 struct bpf_raw_tracepoint
*raw_tp
= filp
->private_data
;
1389 bpf_probe_unregister(raw_tp
->btp
, raw_tp
->prog
);
1390 bpf_prog_put(raw_tp
->prog
);
1396 static const struct file_operations bpf_raw_tp_fops
= {
1397 .release
= bpf_raw_tracepoint_release
,
1398 .read
= bpf_dummy_read
,
1399 .write
= bpf_dummy_write
,
1402 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1404 static int bpf_raw_tracepoint_open(const union bpf_attr
*attr
)
1406 struct bpf_raw_tracepoint
*raw_tp
;
1407 struct bpf_raw_event_map
*btp
;
1408 struct bpf_prog
*prog
;
1412 if (strncpy_from_user(tp_name
, u64_to_user_ptr(attr
->raw_tracepoint
.name
),
1413 sizeof(tp_name
) - 1) < 0)
1415 tp_name
[sizeof(tp_name
) - 1] = 0;
1417 btp
= bpf_find_raw_tracepoint(tp_name
);
1421 raw_tp
= kzalloc(sizeof(*raw_tp
), GFP_USER
);
1426 prog
= bpf_prog_get_type(attr
->raw_tracepoint
.prog_fd
,
1427 BPF_PROG_TYPE_RAW_TRACEPOINT
);
1429 err
= PTR_ERR(prog
);
1433 err
= bpf_probe_register(raw_tp
->btp
, prog
);
1437 raw_tp
->prog
= prog
;
1438 tp_fd
= anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops
, raw_tp
,
1441 bpf_probe_unregister(raw_tp
->btp
, prog
);
1454 #ifdef CONFIG_CGROUP_BPF
1456 static int bpf_prog_attach_check_attach_type(const struct bpf_prog
*prog
,
1457 enum bpf_attach_type attach_type
)
1459 switch (prog
->type
) {
1460 case BPF_PROG_TYPE_CGROUP_SOCK
:
1461 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR
:
1462 return attach_type
== prog
->expected_attach_type
? 0 : -EINVAL
;
1468 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1470 static int sockmap_get_from_fd(const union bpf_attr
*attr
,
1471 int type
, bool attach
)
1473 struct bpf_prog
*prog
= NULL
;
1474 int ufd
= attr
->target_fd
;
1475 struct bpf_map
*map
;
1480 map
= __bpf_map_get(f
);
1482 return PTR_ERR(map
);
1485 prog
= bpf_prog_get_type(attr
->attach_bpf_fd
, type
);
1488 return PTR_ERR(prog
);
1492 err
= sock_map_prog(map
, prog
, attr
->attach_type
);
1504 #define BPF_F_ATTACH_MASK \
1505 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1507 static int bpf_prog_attach(const union bpf_attr
*attr
)
1509 enum bpf_prog_type ptype
;
1510 struct bpf_prog
*prog
;
1511 struct cgroup
*cgrp
;
1514 if (!capable(CAP_NET_ADMIN
))
1517 if (CHECK_ATTR(BPF_PROG_ATTACH
))
1520 if (attr
->attach_flags
& ~BPF_F_ATTACH_MASK
)
1523 switch (attr
->attach_type
) {
1524 case BPF_CGROUP_INET_INGRESS
:
1525 case BPF_CGROUP_INET_EGRESS
:
1526 ptype
= BPF_PROG_TYPE_CGROUP_SKB
;
1528 case BPF_CGROUP_INET_SOCK_CREATE
:
1529 case BPF_CGROUP_INET4_POST_BIND
:
1530 case BPF_CGROUP_INET6_POST_BIND
:
1531 ptype
= BPF_PROG_TYPE_CGROUP_SOCK
;
1533 case BPF_CGROUP_INET4_BIND
:
1534 case BPF_CGROUP_INET6_BIND
:
1535 case BPF_CGROUP_INET4_CONNECT
:
1536 case BPF_CGROUP_INET6_CONNECT
:
1537 ptype
= BPF_PROG_TYPE_CGROUP_SOCK_ADDR
;
1539 case BPF_CGROUP_SOCK_OPS
:
1540 ptype
= BPF_PROG_TYPE_SOCK_OPS
;
1542 case BPF_CGROUP_DEVICE
:
1543 ptype
= BPF_PROG_TYPE_CGROUP_DEVICE
;
1545 case BPF_SK_MSG_VERDICT
:
1546 return sockmap_get_from_fd(attr
, BPF_PROG_TYPE_SK_MSG
, true);
1547 case BPF_SK_SKB_STREAM_PARSER
:
1548 case BPF_SK_SKB_STREAM_VERDICT
:
1549 return sockmap_get_from_fd(attr
, BPF_PROG_TYPE_SK_SKB
, true);
1554 prog
= bpf_prog_get_type(attr
->attach_bpf_fd
, ptype
);
1556 return PTR_ERR(prog
);
1558 if (bpf_prog_attach_check_attach_type(prog
, attr
->attach_type
)) {
1563 cgrp
= cgroup_get_from_fd(attr
->target_fd
);
1566 return PTR_ERR(cgrp
);
1569 ret
= cgroup_bpf_attach(cgrp
, prog
, attr
->attach_type
,
1570 attr
->attach_flags
);
1578 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1580 static int bpf_prog_detach(const union bpf_attr
*attr
)
1582 enum bpf_prog_type ptype
;
1583 struct bpf_prog
*prog
;
1584 struct cgroup
*cgrp
;
1587 if (!capable(CAP_NET_ADMIN
))
1590 if (CHECK_ATTR(BPF_PROG_DETACH
))
1593 switch (attr
->attach_type
) {
1594 case BPF_CGROUP_INET_INGRESS
:
1595 case BPF_CGROUP_INET_EGRESS
:
1596 ptype
= BPF_PROG_TYPE_CGROUP_SKB
;
1598 case BPF_CGROUP_INET_SOCK_CREATE
:
1599 case BPF_CGROUP_INET4_POST_BIND
:
1600 case BPF_CGROUP_INET6_POST_BIND
:
1601 ptype
= BPF_PROG_TYPE_CGROUP_SOCK
;
1603 case BPF_CGROUP_INET4_BIND
:
1604 case BPF_CGROUP_INET6_BIND
:
1605 case BPF_CGROUP_INET4_CONNECT
:
1606 case BPF_CGROUP_INET6_CONNECT
:
1607 ptype
= BPF_PROG_TYPE_CGROUP_SOCK_ADDR
;
1609 case BPF_CGROUP_SOCK_OPS
:
1610 ptype
= BPF_PROG_TYPE_SOCK_OPS
;
1612 case BPF_CGROUP_DEVICE
:
1613 ptype
= BPF_PROG_TYPE_CGROUP_DEVICE
;
1615 case BPF_SK_MSG_VERDICT
:
1616 return sockmap_get_from_fd(attr
, BPF_PROG_TYPE_SK_MSG
, false);
1617 case BPF_SK_SKB_STREAM_PARSER
:
1618 case BPF_SK_SKB_STREAM_VERDICT
:
1619 return sockmap_get_from_fd(attr
, BPF_PROG_TYPE_SK_SKB
, false);
1624 cgrp
= cgroup_get_from_fd(attr
->target_fd
);
1626 return PTR_ERR(cgrp
);
1628 prog
= bpf_prog_get_type(attr
->attach_bpf_fd
, ptype
);
1632 ret
= cgroup_bpf_detach(cgrp
, prog
, attr
->attach_type
, 0);
1639 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1641 static int bpf_prog_query(const union bpf_attr
*attr
,
1642 union bpf_attr __user
*uattr
)
1644 struct cgroup
*cgrp
;
1647 if (!capable(CAP_NET_ADMIN
))
1649 if (CHECK_ATTR(BPF_PROG_QUERY
))
1651 if (attr
->query
.query_flags
& ~BPF_F_QUERY_EFFECTIVE
)
1654 switch (attr
->query
.attach_type
) {
1655 case BPF_CGROUP_INET_INGRESS
:
1656 case BPF_CGROUP_INET_EGRESS
:
1657 case BPF_CGROUP_INET_SOCK_CREATE
:
1658 case BPF_CGROUP_INET4_BIND
:
1659 case BPF_CGROUP_INET6_BIND
:
1660 case BPF_CGROUP_INET4_POST_BIND
:
1661 case BPF_CGROUP_INET6_POST_BIND
:
1662 case BPF_CGROUP_INET4_CONNECT
:
1663 case BPF_CGROUP_INET6_CONNECT
:
1664 case BPF_CGROUP_SOCK_OPS
:
1665 case BPF_CGROUP_DEVICE
:
1670 cgrp
= cgroup_get_from_fd(attr
->query
.target_fd
);
1672 return PTR_ERR(cgrp
);
1673 ret
= cgroup_bpf_query(cgrp
, attr
, uattr
);
1677 #endif /* CONFIG_CGROUP_BPF */
1679 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1681 static int bpf_prog_test_run(const union bpf_attr
*attr
,
1682 union bpf_attr __user
*uattr
)
1684 struct bpf_prog
*prog
;
1685 int ret
= -ENOTSUPP
;
1687 if (!capable(CAP_SYS_ADMIN
))
1689 if (CHECK_ATTR(BPF_PROG_TEST_RUN
))
1692 prog
= bpf_prog_get(attr
->test
.prog_fd
);
1694 return PTR_ERR(prog
);
1696 if (prog
->aux
->ops
->test_run
)
1697 ret
= prog
->aux
->ops
->test_run(prog
, attr
, uattr
);
1703 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1705 static int bpf_obj_get_next_id(const union bpf_attr
*attr
,
1706 union bpf_attr __user
*uattr
,
1710 u32 next_id
= attr
->start_id
;
1713 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID
) || next_id
>= INT_MAX
)
1716 if (!capable(CAP_SYS_ADMIN
))
1721 if (!idr_get_next(idr
, &next_id
))
1723 spin_unlock_bh(lock
);
1726 err
= put_user(next_id
, &uattr
->next_id
);
1731 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1733 static int bpf_prog_get_fd_by_id(const union bpf_attr
*attr
)
1735 struct bpf_prog
*prog
;
1736 u32 id
= attr
->prog_id
;
1739 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID
))
1742 if (!capable(CAP_SYS_ADMIN
))
1745 spin_lock_bh(&prog_idr_lock
);
1746 prog
= idr_find(&prog_idr
, id
);
1748 prog
= bpf_prog_inc_not_zero(prog
);
1750 prog
= ERR_PTR(-ENOENT
);
1751 spin_unlock_bh(&prog_idr_lock
);
1754 return PTR_ERR(prog
);
1756 fd
= bpf_prog_new_fd(prog
);
1763 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1765 static int bpf_map_get_fd_by_id(const union bpf_attr
*attr
)
1767 struct bpf_map
*map
;
1768 u32 id
= attr
->map_id
;
1772 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID
) ||
1773 attr
->open_flags
& ~BPF_OBJ_FLAG_MASK
)
1776 if (!capable(CAP_SYS_ADMIN
))
1779 f_flags
= bpf_get_file_flag(attr
->open_flags
);
1783 spin_lock_bh(&map_idr_lock
);
1784 map
= idr_find(&map_idr
, id
);
1786 map
= bpf_map_inc_not_zero(map
, true);
1788 map
= ERR_PTR(-ENOENT
);
1789 spin_unlock_bh(&map_idr_lock
);
1792 return PTR_ERR(map
);
1794 fd
= bpf_map_new_fd(map
, f_flags
);
1801 static const struct bpf_map
*bpf_map_from_imm(const struct bpf_prog
*prog
,
1806 for (i
= 0; i
< prog
->aux
->used_map_cnt
; i
++)
1807 if (prog
->aux
->used_maps
[i
] == (void *)addr
)
1808 return prog
->aux
->used_maps
[i
];
1812 static struct bpf_insn
*bpf_insn_prepare_dump(const struct bpf_prog
*prog
)
1814 const struct bpf_map
*map
;
1815 struct bpf_insn
*insns
;
1819 insns
= kmemdup(prog
->insnsi
, bpf_prog_insn_size(prog
),
1824 for (i
= 0; i
< prog
->len
; i
++) {
1825 if (insns
[i
].code
== (BPF_JMP
| BPF_TAIL_CALL
)) {
1826 insns
[i
].code
= BPF_JMP
| BPF_CALL
;
1827 insns
[i
].imm
= BPF_FUNC_tail_call
;
1830 if (insns
[i
].code
== (BPF_JMP
| BPF_CALL
) ||
1831 insns
[i
].code
== (BPF_JMP
| BPF_CALL_ARGS
)) {
1832 if (insns
[i
].code
== (BPF_JMP
| BPF_CALL_ARGS
))
1833 insns
[i
].code
= BPF_JMP
| BPF_CALL
;
1834 if (!bpf_dump_raw_ok())
1839 if (insns
[i
].code
!= (BPF_LD
| BPF_IMM
| BPF_DW
))
1842 imm
= ((u64
)insns
[i
+ 1].imm
<< 32) | (u32
)insns
[i
].imm
;
1843 map
= bpf_map_from_imm(prog
, imm
);
1845 insns
[i
].src_reg
= BPF_PSEUDO_MAP_FD
;
1846 insns
[i
].imm
= map
->id
;
1847 insns
[i
+ 1].imm
= 0;
1851 if (!bpf_dump_raw_ok() &&
1852 imm
== (unsigned long)prog
->aux
) {
1854 insns
[i
+ 1].imm
= 0;
1862 static int bpf_prog_get_info_by_fd(struct bpf_prog
*prog
,
1863 const union bpf_attr
*attr
,
1864 union bpf_attr __user
*uattr
)
1866 struct bpf_prog_info __user
*uinfo
= u64_to_user_ptr(attr
->info
.info
);
1867 struct bpf_prog_info info
= {};
1868 u32 info_len
= attr
->info
.info_len
;
1869 char __user
*uinsns
;
1873 err
= check_uarg_tail_zero(uinfo
, sizeof(info
), info_len
);
1876 info_len
= min_t(u32
, sizeof(info
), info_len
);
1878 if (copy_from_user(&info
, uinfo
, info_len
))
1881 info
.type
= prog
->type
;
1882 info
.id
= prog
->aux
->id
;
1883 info
.load_time
= prog
->aux
->load_time
;
1884 info
.created_by_uid
= from_kuid_munged(current_user_ns(),
1885 prog
->aux
->user
->uid
);
1887 memcpy(info
.tag
, prog
->tag
, sizeof(prog
->tag
));
1888 memcpy(info
.name
, prog
->aux
->name
, sizeof(prog
->aux
->name
));
1890 ulen
= info
.nr_map_ids
;
1891 info
.nr_map_ids
= prog
->aux
->used_map_cnt
;
1892 ulen
= min_t(u32
, info
.nr_map_ids
, ulen
);
1894 u32 __user
*user_map_ids
= u64_to_user_ptr(info
.map_ids
);
1897 for (i
= 0; i
< ulen
; i
++)
1898 if (put_user(prog
->aux
->used_maps
[i
]->id
,
1903 if (!capable(CAP_SYS_ADMIN
)) {
1904 info
.jited_prog_len
= 0;
1905 info
.xlated_prog_len
= 0;
1909 ulen
= info
.xlated_prog_len
;
1910 info
.xlated_prog_len
= bpf_prog_insn_size(prog
);
1911 if (info
.xlated_prog_len
&& ulen
) {
1912 struct bpf_insn
*insns_sanitized
;
1915 if (prog
->blinded
&& !bpf_dump_raw_ok()) {
1916 info
.xlated_prog_insns
= 0;
1919 insns_sanitized
= bpf_insn_prepare_dump(prog
);
1920 if (!insns_sanitized
)
1922 uinsns
= u64_to_user_ptr(info
.xlated_prog_insns
);
1923 ulen
= min_t(u32
, info
.xlated_prog_len
, ulen
);
1924 fault
= copy_to_user(uinsns
, insns_sanitized
, ulen
);
1925 kfree(insns_sanitized
);
1930 if (bpf_prog_is_dev_bound(prog
->aux
)) {
1931 err
= bpf_prog_offload_info_fill(&info
, prog
);
1937 /* NOTE: the following code is supposed to be skipped for offload.
1938 * bpf_prog_offload_info_fill() is the place to fill similar fields
1941 ulen
= info
.jited_prog_len
;
1942 info
.jited_prog_len
= prog
->jited_len
;
1943 if (info
.jited_prog_len
&& ulen
) {
1944 if (bpf_dump_raw_ok()) {
1945 uinsns
= u64_to_user_ptr(info
.jited_prog_insns
);
1946 ulen
= min_t(u32
, info
.jited_prog_len
, ulen
);
1947 if (copy_to_user(uinsns
, prog
->bpf_func
, ulen
))
1950 info
.jited_prog_insns
= 0;
1955 if (copy_to_user(uinfo
, &info
, info_len
) ||
1956 put_user(info_len
, &uattr
->info
.info_len
))
1962 static int bpf_map_get_info_by_fd(struct bpf_map
*map
,
1963 const union bpf_attr
*attr
,
1964 union bpf_attr __user
*uattr
)
1966 struct bpf_map_info __user
*uinfo
= u64_to_user_ptr(attr
->info
.info
);
1967 struct bpf_map_info info
= {};
1968 u32 info_len
= attr
->info
.info_len
;
1971 err
= check_uarg_tail_zero(uinfo
, sizeof(info
), info_len
);
1974 info_len
= min_t(u32
, sizeof(info
), info_len
);
1976 info
.type
= map
->map_type
;
1978 info
.key_size
= map
->key_size
;
1979 info
.value_size
= map
->value_size
;
1980 info
.max_entries
= map
->max_entries
;
1981 info
.map_flags
= map
->map_flags
;
1982 memcpy(info
.name
, map
->name
, sizeof(map
->name
));
1984 if (bpf_map_is_dev_bound(map
)) {
1985 err
= bpf_map_offload_info_fill(&info
, map
);
1990 if (copy_to_user(uinfo
, &info
, info_len
) ||
1991 put_user(info_len
, &uattr
->info
.info_len
))
1997 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
1999 static int bpf_obj_get_info_by_fd(const union bpf_attr
*attr
,
2000 union bpf_attr __user
*uattr
)
2002 int ufd
= attr
->info
.bpf_fd
;
2006 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD
))
2013 if (f
.file
->f_op
== &bpf_prog_fops
)
2014 err
= bpf_prog_get_info_by_fd(f
.file
->private_data
, attr
,
2016 else if (f
.file
->f_op
== &bpf_map_fops
)
2017 err
= bpf_map_get_info_by_fd(f
.file
->private_data
, attr
,
2026 SYSCALL_DEFINE3(bpf
, int, cmd
, union bpf_attr __user
*, uattr
, unsigned int, size
)
2028 union bpf_attr attr
= {};
2031 if (sysctl_unprivileged_bpf_disabled
&& !capable(CAP_SYS_ADMIN
))
2034 err
= check_uarg_tail_zero(uattr
, sizeof(attr
), size
);
2037 size
= min_t(u32
, size
, sizeof(attr
));
2039 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2040 if (copy_from_user(&attr
, uattr
, size
) != 0)
2043 err
= security_bpf(cmd
, &attr
, size
);
2048 case BPF_MAP_CREATE
:
2049 err
= map_create(&attr
);
2051 case BPF_MAP_LOOKUP_ELEM
:
2052 err
= map_lookup_elem(&attr
);
2054 case BPF_MAP_UPDATE_ELEM
:
2055 err
= map_update_elem(&attr
);
2057 case BPF_MAP_DELETE_ELEM
:
2058 err
= map_delete_elem(&attr
);
2060 case BPF_MAP_GET_NEXT_KEY
:
2061 err
= map_get_next_key(&attr
);
2064 err
= bpf_prog_load(&attr
);
2067 err
= bpf_obj_pin(&attr
);
2070 err
= bpf_obj_get(&attr
);
2072 #ifdef CONFIG_CGROUP_BPF
2073 case BPF_PROG_ATTACH
:
2074 err
= bpf_prog_attach(&attr
);
2076 case BPF_PROG_DETACH
:
2077 err
= bpf_prog_detach(&attr
);
2079 case BPF_PROG_QUERY
:
2080 err
= bpf_prog_query(&attr
, uattr
);
2083 case BPF_PROG_TEST_RUN
:
2084 err
= bpf_prog_test_run(&attr
, uattr
);
2086 case BPF_PROG_GET_NEXT_ID
:
2087 err
= bpf_obj_get_next_id(&attr
, uattr
,
2088 &prog_idr
, &prog_idr_lock
);
2090 case BPF_MAP_GET_NEXT_ID
:
2091 err
= bpf_obj_get_next_id(&attr
, uattr
,
2092 &map_idr
, &map_idr_lock
);
2094 case BPF_PROG_GET_FD_BY_ID
:
2095 err
= bpf_prog_get_fd_by_id(&attr
);
2097 case BPF_MAP_GET_FD_BY_ID
:
2098 err
= bpf_map_get_fd_by_id(&attr
);
2100 case BPF_OBJ_GET_INFO_BY_FD
:
2101 err
= bpf_obj_get_info_by_fd(&attr
, uattr
);
2103 case BPF_RAW_TRACEPOINT_OPEN
:
2104 err
= bpf_raw_tracepoint_open(&attr
);