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/syscalls.h>
14 #include <linux/slab.h>
15 #include <linux/anon_inodes.h>
16 #include <linux/file.h>
17 #include <linux/license.h>
18 #include <linux/filter.h>
19 #include <linux/version.h>
21 int sysctl_unprivileged_bpf_disabled __read_mostly
;
23 static LIST_HEAD(bpf_map_types
);
25 static struct bpf_map
*find_and_alloc_map(union bpf_attr
*attr
)
27 struct bpf_map_type_list
*tl
;
30 list_for_each_entry(tl
, &bpf_map_types
, list_node
) {
31 if (tl
->type
== attr
->map_type
) {
32 map
= tl
->ops
->map_alloc(attr
);
36 map
->map_type
= attr
->map_type
;
40 return ERR_PTR(-EINVAL
);
43 /* boot time registration of different map implementations */
44 void bpf_register_map_type(struct bpf_map_type_list
*tl
)
46 list_add(&tl
->list_node
, &bpf_map_types
);
49 static int bpf_map_charge_memlock(struct bpf_map
*map
)
51 struct user_struct
*user
= get_current_user();
52 unsigned long memlock_limit
;
54 memlock_limit
= rlimit(RLIMIT_MEMLOCK
) >> PAGE_SHIFT
;
56 atomic_long_add(map
->pages
, &user
->locked_vm
);
58 if (atomic_long_read(&user
->locked_vm
) > memlock_limit
) {
59 atomic_long_sub(map
->pages
, &user
->locked_vm
);
67 static void bpf_map_uncharge_memlock(struct bpf_map
*map
)
69 struct user_struct
*user
= map
->user
;
71 atomic_long_sub(map
->pages
, &user
->locked_vm
);
75 /* called from workqueue */
76 static void bpf_map_free_deferred(struct work_struct
*work
)
78 struct bpf_map
*map
= container_of(work
, struct bpf_map
, work
);
80 bpf_map_uncharge_memlock(map
);
81 /* implementation dependent freeing */
82 map
->ops
->map_free(map
);
85 static void bpf_map_put_uref(struct bpf_map
*map
)
87 if (atomic_dec_and_test(&map
->usercnt
)) {
88 if (map
->map_type
== BPF_MAP_TYPE_PROG_ARRAY
)
89 bpf_fd_array_map_clear(map
);
93 /* decrement map refcnt and schedule it for freeing via workqueue
94 * (unrelying map implementation ops->map_free() might sleep)
96 void bpf_map_put(struct bpf_map
*map
)
98 if (atomic_dec_and_test(&map
->refcnt
)) {
99 INIT_WORK(&map
->work
, bpf_map_free_deferred
);
100 schedule_work(&map
->work
);
104 void bpf_map_put_with_uref(struct bpf_map
*map
)
106 bpf_map_put_uref(map
);
110 static int bpf_map_release(struct inode
*inode
, struct file
*filp
)
112 bpf_map_put_with_uref(filp
->private_data
);
116 #ifdef CONFIG_PROC_FS
117 static void bpf_map_show_fdinfo(struct seq_file
*m
, struct file
*filp
)
119 const struct bpf_map
*map
= filp
->private_data
;
125 "max_entries:\t%u\n",
133 static const struct file_operations bpf_map_fops
= {
134 #ifdef CONFIG_PROC_FS
135 .show_fdinfo
= bpf_map_show_fdinfo
,
137 .release
= bpf_map_release
,
140 int bpf_map_new_fd(struct bpf_map
*map
)
142 return anon_inode_getfd("bpf-map", &bpf_map_fops
, map
,
146 /* helper macro to check that unused fields 'union bpf_attr' are zero */
147 #define CHECK_ATTR(CMD) \
148 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
149 sizeof(attr->CMD##_LAST_FIELD), 0, \
151 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
152 sizeof(attr->CMD##_LAST_FIELD)) != NULL
154 #define BPF_MAP_CREATE_LAST_FIELD max_entries
155 /* called via syscall */
156 static int map_create(union bpf_attr
*attr
)
161 err
= CHECK_ATTR(BPF_MAP_CREATE
);
165 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
166 map
= find_and_alloc_map(attr
);
170 atomic_set(&map
->refcnt
, 1);
171 atomic_set(&map
->usercnt
, 1);
173 err
= bpf_map_charge_memlock(map
);
177 err
= bpf_map_new_fd(map
);
179 /* failed to allocate fd */
185 map
->ops
->map_free(map
);
189 /* if error is returned, fd is released.
190 * On success caller should complete fd access with matching fdput()
192 struct bpf_map
*__bpf_map_get(struct fd f
)
195 return ERR_PTR(-EBADF
);
196 if (f
.file
->f_op
!= &bpf_map_fops
) {
198 return ERR_PTR(-EINVAL
);
201 return f
.file
->private_data
;
204 void bpf_map_inc(struct bpf_map
*map
, bool uref
)
206 atomic_inc(&map
->refcnt
);
208 atomic_inc(&map
->usercnt
);
211 struct bpf_map
*bpf_map_get_with_uref(u32 ufd
)
213 struct fd f
= fdget(ufd
);
216 map
= __bpf_map_get(f
);
220 bpf_map_inc(map
, true);
226 /* helper to convert user pointers passed inside __aligned_u64 fields */
227 static void __user
*u64_to_ptr(__u64 val
)
229 return (void __user
*) (unsigned long) val
;
232 /* last field in 'union bpf_attr' used by this command */
233 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
235 static int map_lookup_elem(union bpf_attr
*attr
)
237 void __user
*ukey
= u64_to_ptr(attr
->key
);
238 void __user
*uvalue
= u64_to_ptr(attr
->value
);
239 int ufd
= attr
->map_fd
;
241 void *key
, *value
, *ptr
;
246 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM
))
250 map
= __bpf_map_get(f
);
255 key
= kmalloc(map
->key_size
, GFP_USER
);
260 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
263 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
||
264 map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
)
265 value_size
= round_up(map
->value_size
, 8) * num_possible_cpus();
267 value_size
= map
->value_size
;
270 value
= kmalloc(value_size
, GFP_USER
| __GFP_NOWARN
);
274 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
) {
275 err
= bpf_percpu_hash_copy(map
, key
, value
);
276 } else if (map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
) {
277 err
= bpf_percpu_array_copy(map
, key
, value
);
280 ptr
= map
->ops
->map_lookup_elem(map
, key
);
282 memcpy(value
, ptr
, value_size
);
284 err
= ptr
? 0 : -ENOENT
;
291 if (copy_to_user(uvalue
, value
, value_size
) != 0)
305 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
307 static int map_update_elem(union bpf_attr
*attr
)
309 void __user
*ukey
= u64_to_ptr(attr
->key
);
310 void __user
*uvalue
= u64_to_ptr(attr
->value
);
311 int ufd
= attr
->map_fd
;
318 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM
))
322 map
= __bpf_map_get(f
);
327 key
= kmalloc(map
->key_size
, GFP_USER
);
332 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
335 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
||
336 map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
)
337 value_size
= round_up(map
->value_size
, 8) * num_possible_cpus();
339 value_size
= map
->value_size
;
342 value
= kmalloc(value_size
, GFP_USER
| __GFP_NOWARN
);
347 if (copy_from_user(value
, uvalue
, value_size
) != 0)
350 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
) {
351 err
= bpf_percpu_hash_update(map
, key
, value
, attr
->flags
);
352 } else if (map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
) {
353 err
= bpf_percpu_array_update(map
, key
, value
, attr
->flags
);
356 err
= map
->ops
->map_update_elem(map
, key
, value
, attr
->flags
);
369 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
371 static int map_delete_elem(union bpf_attr
*attr
)
373 void __user
*ukey
= u64_to_ptr(attr
->key
);
374 int ufd
= attr
->map_fd
;
380 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM
))
384 map
= __bpf_map_get(f
);
389 key
= kmalloc(map
->key_size
, GFP_USER
);
394 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
398 err
= map
->ops
->map_delete_elem(map
, key
);
408 /* last field in 'union bpf_attr' used by this command */
409 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
411 static int map_get_next_key(union bpf_attr
*attr
)
413 void __user
*ukey
= u64_to_ptr(attr
->key
);
414 void __user
*unext_key
= u64_to_ptr(attr
->next_key
);
415 int ufd
= attr
->map_fd
;
417 void *key
, *next_key
;
421 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY
))
425 map
= __bpf_map_get(f
);
430 key
= kmalloc(map
->key_size
, GFP_USER
);
435 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
439 next_key
= kmalloc(map
->key_size
, GFP_USER
);
444 err
= map
->ops
->map_get_next_key(map
, key
, next_key
);
450 if (copy_to_user(unext_key
, next_key
, map
->key_size
) != 0)
464 static LIST_HEAD(bpf_prog_types
);
466 static int find_prog_type(enum bpf_prog_type type
, struct bpf_prog
*prog
)
468 struct bpf_prog_type_list
*tl
;
470 list_for_each_entry(tl
, &bpf_prog_types
, list_node
) {
471 if (tl
->type
== type
) {
472 prog
->aux
->ops
= tl
->ops
;
481 void bpf_register_prog_type(struct bpf_prog_type_list
*tl
)
483 list_add(&tl
->list_node
, &bpf_prog_types
);
486 /* fixup insn->imm field of bpf_call instructions:
487 * if (insn->imm == BPF_FUNC_map_lookup_elem)
488 * insn->imm = bpf_map_lookup_elem - __bpf_call_base;
489 * else if (insn->imm == BPF_FUNC_map_update_elem)
490 * insn->imm = bpf_map_update_elem - __bpf_call_base;
493 * this function is called after eBPF program passed verification
495 static void fixup_bpf_calls(struct bpf_prog
*prog
)
497 const struct bpf_func_proto
*fn
;
500 for (i
= 0; i
< prog
->len
; i
++) {
501 struct bpf_insn
*insn
= &prog
->insnsi
[i
];
503 if (insn
->code
== (BPF_JMP
| BPF_CALL
)) {
504 /* we reach here when program has bpf_call instructions
505 * and it passed bpf_check(), means that
506 * ops->get_func_proto must have been supplied, check it
508 BUG_ON(!prog
->aux
->ops
->get_func_proto
);
510 if (insn
->imm
== BPF_FUNC_get_route_realm
)
511 prog
->dst_needed
= 1;
512 if (insn
->imm
== BPF_FUNC_get_prandom_u32
)
513 bpf_user_rnd_init_once();
514 if (insn
->imm
== BPF_FUNC_tail_call
) {
515 /* mark bpf_tail_call as different opcode
516 * to avoid conditional branch in
517 * interpeter for every normal call
518 * and to prevent accidental JITing by
519 * JIT compiler that doesn't support
527 fn
= prog
->aux
->ops
->get_func_proto(insn
->imm
);
528 /* all functions that have prototype and verifier allowed
529 * programs to call them, must be real in-kernel functions
532 insn
->imm
= fn
->func
- __bpf_call_base
;
537 /* drop refcnt on maps used by eBPF program and free auxilary data */
538 static void free_used_maps(struct bpf_prog_aux
*aux
)
542 for (i
= 0; i
< aux
->used_map_cnt
; i
++)
543 bpf_map_put(aux
->used_maps
[i
]);
545 kfree(aux
->used_maps
);
548 static int bpf_prog_charge_memlock(struct bpf_prog
*prog
)
550 struct user_struct
*user
= get_current_user();
551 unsigned long memlock_limit
;
553 memlock_limit
= rlimit(RLIMIT_MEMLOCK
) >> PAGE_SHIFT
;
555 atomic_long_add(prog
->pages
, &user
->locked_vm
);
556 if (atomic_long_read(&user
->locked_vm
) > memlock_limit
) {
557 atomic_long_sub(prog
->pages
, &user
->locked_vm
);
561 prog
->aux
->user
= user
;
565 static void bpf_prog_uncharge_memlock(struct bpf_prog
*prog
)
567 struct user_struct
*user
= prog
->aux
->user
;
569 atomic_long_sub(prog
->pages
, &user
->locked_vm
);
573 static void __prog_put_common(struct rcu_head
*rcu
)
575 struct bpf_prog_aux
*aux
= container_of(rcu
, struct bpf_prog_aux
, rcu
);
578 bpf_prog_uncharge_memlock(aux
->prog
);
579 bpf_prog_free(aux
->prog
);
582 /* version of bpf_prog_put() that is called after a grace period */
583 void bpf_prog_put_rcu(struct bpf_prog
*prog
)
585 if (atomic_dec_and_test(&prog
->aux
->refcnt
))
586 call_rcu(&prog
->aux
->rcu
, __prog_put_common
);
589 void bpf_prog_put(struct bpf_prog
*prog
)
591 if (atomic_dec_and_test(&prog
->aux
->refcnt
))
592 __prog_put_common(&prog
->aux
->rcu
);
594 EXPORT_SYMBOL_GPL(bpf_prog_put
);
596 static int bpf_prog_release(struct inode
*inode
, struct file
*filp
)
598 struct bpf_prog
*prog
= filp
->private_data
;
600 bpf_prog_put_rcu(prog
);
604 static const struct file_operations bpf_prog_fops
= {
605 .release
= bpf_prog_release
,
608 int bpf_prog_new_fd(struct bpf_prog
*prog
)
610 return anon_inode_getfd("bpf-prog", &bpf_prog_fops
, prog
,
614 static struct bpf_prog
*__bpf_prog_get(struct fd f
)
617 return ERR_PTR(-EBADF
);
618 if (f
.file
->f_op
!= &bpf_prog_fops
) {
620 return ERR_PTR(-EINVAL
);
623 return f
.file
->private_data
;
626 /* called by sockets/tracing/seccomp before attaching program to an event
627 * pairs with bpf_prog_put()
629 struct bpf_prog
*bpf_prog_get(u32 ufd
)
631 struct fd f
= fdget(ufd
);
632 struct bpf_prog
*prog
;
634 prog
= __bpf_prog_get(f
);
638 atomic_inc(&prog
->aux
->refcnt
);
643 EXPORT_SYMBOL_GPL(bpf_prog_get
);
645 /* last field in 'union bpf_attr' used by this command */
646 #define BPF_PROG_LOAD_LAST_FIELD kern_version
648 static int bpf_prog_load(union bpf_attr
*attr
)
650 enum bpf_prog_type type
= attr
->prog_type
;
651 struct bpf_prog
*prog
;
656 if (CHECK_ATTR(BPF_PROG_LOAD
))
659 /* copy eBPF program license from user space */
660 if (strncpy_from_user(license
, u64_to_ptr(attr
->license
),
661 sizeof(license
) - 1) < 0)
663 license
[sizeof(license
) - 1] = 0;
665 /* eBPF programs must be GPL compatible to use GPL-ed functions */
666 is_gpl
= license_is_gpl_compatible(license
);
668 if (attr
->insn_cnt
>= BPF_MAXINSNS
)
671 if (type
== BPF_PROG_TYPE_KPROBE
&&
672 attr
->kern_version
!= LINUX_VERSION_CODE
)
675 if (type
!= BPF_PROG_TYPE_SOCKET_FILTER
&& !capable(CAP_SYS_ADMIN
))
678 /* plain bpf_prog allocation */
679 prog
= bpf_prog_alloc(bpf_prog_size(attr
->insn_cnt
), GFP_USER
);
683 err
= bpf_prog_charge_memlock(prog
);
685 goto free_prog_nouncharge
;
687 prog
->len
= attr
->insn_cnt
;
690 if (copy_from_user(prog
->insns
, u64_to_ptr(attr
->insns
),
691 prog
->len
* sizeof(struct bpf_insn
)) != 0)
694 prog
->orig_prog
= NULL
;
697 atomic_set(&prog
->aux
->refcnt
, 1);
698 prog
->gpl_compatible
= is_gpl
? 1 : 0;
700 /* find program type: socket_filter vs tracing_filter */
701 err
= find_prog_type(type
, prog
);
705 /* run eBPF verifier */
706 err
= bpf_check(&prog
, attr
);
710 /* fixup BPF_CALL->imm field */
711 fixup_bpf_calls(prog
);
713 /* eBPF program is ready to be JITed */
714 err
= bpf_prog_select_runtime(prog
);
718 err
= bpf_prog_new_fd(prog
);
720 /* failed to allocate fd */
726 free_used_maps(prog
->aux
);
728 bpf_prog_uncharge_memlock(prog
);
729 free_prog_nouncharge
:
734 #define BPF_OBJ_LAST_FIELD bpf_fd
736 static int bpf_obj_pin(const union bpf_attr
*attr
)
738 if (CHECK_ATTR(BPF_OBJ
))
741 return bpf_obj_pin_user(attr
->bpf_fd
, u64_to_ptr(attr
->pathname
));
744 static int bpf_obj_get(const union bpf_attr
*attr
)
746 if (CHECK_ATTR(BPF_OBJ
) || attr
->bpf_fd
!= 0)
749 return bpf_obj_get_user(u64_to_ptr(attr
->pathname
));
752 SYSCALL_DEFINE3(bpf
, int, cmd
, union bpf_attr __user
*, uattr
, unsigned int, size
)
754 union bpf_attr attr
= {};
757 if (!capable(CAP_SYS_ADMIN
) && sysctl_unprivileged_bpf_disabled
)
760 if (!access_ok(VERIFY_READ
, uattr
, 1))
763 if (size
> PAGE_SIZE
) /* silly large */
766 /* If we're handed a bigger struct than we know of,
767 * ensure all the unknown bits are 0 - i.e. new
768 * user-space does not rely on any kernel feature
769 * extensions we dont know about yet.
771 if (size
> sizeof(attr
)) {
772 unsigned char __user
*addr
;
773 unsigned char __user
*end
;
776 addr
= (void __user
*)uattr
+ sizeof(attr
);
777 end
= (void __user
*)uattr
+ size
;
779 for (; addr
< end
; addr
++) {
780 err
= get_user(val
, addr
);
789 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
790 if (copy_from_user(&attr
, uattr
, size
) != 0)
795 err
= map_create(&attr
);
797 case BPF_MAP_LOOKUP_ELEM
:
798 err
= map_lookup_elem(&attr
);
800 case BPF_MAP_UPDATE_ELEM
:
801 err
= map_update_elem(&attr
);
803 case BPF_MAP_DELETE_ELEM
:
804 err
= map_delete_elem(&attr
);
806 case BPF_MAP_GET_NEXT_KEY
:
807 err
= map_get_next_key(&attr
);
810 err
= bpf_prog_load(&attr
);
813 err
= bpf_obj_pin(&attr
);
816 err
= bpf_obj_get(&attr
);