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 DEFINE_PER_CPU(int, bpf_prog_active
);
23 int sysctl_unprivileged_bpf_disabled __read_mostly
;
25 static LIST_HEAD(bpf_map_types
);
27 static struct bpf_map
*find_and_alloc_map(union bpf_attr
*attr
)
29 struct bpf_map_type_list
*tl
;
32 list_for_each_entry(tl
, &bpf_map_types
, list_node
) {
33 if (tl
->type
== attr
->map_type
) {
34 map
= tl
->ops
->map_alloc(attr
);
38 map
->map_type
= attr
->map_type
;
42 return ERR_PTR(-EINVAL
);
45 /* boot time registration of different map implementations */
46 void bpf_register_map_type(struct bpf_map_type_list
*tl
)
48 list_add(&tl
->list_node
, &bpf_map_types
);
51 int bpf_map_precharge_memlock(u32 pages
)
53 struct user_struct
*user
= get_current_user();
54 unsigned long memlock_limit
, cur
;
56 memlock_limit
= rlimit(RLIMIT_MEMLOCK
) >> PAGE_SHIFT
;
57 cur
= atomic_long_read(&user
->locked_vm
);
59 if (cur
+ pages
> memlock_limit
)
64 static int bpf_map_charge_memlock(struct bpf_map
*map
)
66 struct user_struct
*user
= get_current_user();
67 unsigned long memlock_limit
;
69 memlock_limit
= rlimit(RLIMIT_MEMLOCK
) >> PAGE_SHIFT
;
71 atomic_long_add(map
->pages
, &user
->locked_vm
);
73 if (atomic_long_read(&user
->locked_vm
) > memlock_limit
) {
74 atomic_long_sub(map
->pages
, &user
->locked_vm
);
82 static void bpf_map_uncharge_memlock(struct bpf_map
*map
)
84 struct user_struct
*user
= map
->user
;
86 atomic_long_sub(map
->pages
, &user
->locked_vm
);
90 /* called from workqueue */
91 static void bpf_map_free_deferred(struct work_struct
*work
)
93 struct bpf_map
*map
= container_of(work
, struct bpf_map
, work
);
95 bpf_map_uncharge_memlock(map
);
96 /* implementation dependent freeing */
97 map
->ops
->map_free(map
);
100 static void bpf_map_put_uref(struct bpf_map
*map
)
102 if (atomic_dec_and_test(&map
->usercnt
)) {
103 if (map
->map_type
== BPF_MAP_TYPE_PROG_ARRAY
)
104 bpf_fd_array_map_clear(map
);
108 /* decrement map refcnt and schedule it for freeing via workqueue
109 * (unrelying map implementation ops->map_free() might sleep)
111 void bpf_map_put(struct bpf_map
*map
)
113 if (atomic_dec_and_test(&map
->refcnt
)) {
114 INIT_WORK(&map
->work
, bpf_map_free_deferred
);
115 schedule_work(&map
->work
);
119 void bpf_map_put_with_uref(struct bpf_map
*map
)
121 bpf_map_put_uref(map
);
125 static int bpf_map_release(struct inode
*inode
, struct file
*filp
)
127 bpf_map_put_with_uref(filp
->private_data
);
131 #ifdef CONFIG_PROC_FS
132 static void bpf_map_show_fdinfo(struct seq_file
*m
, struct file
*filp
)
134 const struct bpf_map
*map
= filp
->private_data
;
140 "max_entries:\t%u\n",
148 static const struct file_operations bpf_map_fops
= {
149 #ifdef CONFIG_PROC_FS
150 .show_fdinfo
= bpf_map_show_fdinfo
,
152 .release
= bpf_map_release
,
155 int bpf_map_new_fd(struct bpf_map
*map
)
157 return anon_inode_getfd("bpf-map", &bpf_map_fops
, map
,
161 /* helper macro to check that unused fields 'union bpf_attr' are zero */
162 #define CHECK_ATTR(CMD) \
163 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
164 sizeof(attr->CMD##_LAST_FIELD), 0, \
166 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
167 sizeof(attr->CMD##_LAST_FIELD)) != NULL
169 #define BPF_MAP_CREATE_LAST_FIELD map_flags
170 /* called via syscall */
171 static int map_create(union bpf_attr
*attr
)
176 err
= CHECK_ATTR(BPF_MAP_CREATE
);
180 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
181 map
= find_and_alloc_map(attr
);
185 atomic_set(&map
->refcnt
, 1);
186 atomic_set(&map
->usercnt
, 1);
188 err
= bpf_map_charge_memlock(map
);
192 err
= bpf_map_new_fd(map
);
194 /* failed to allocate fd */
200 map
->ops
->map_free(map
);
204 /* if error is returned, fd is released.
205 * On success caller should complete fd access with matching fdput()
207 struct bpf_map
*__bpf_map_get(struct fd f
)
210 return ERR_PTR(-EBADF
);
211 if (f
.file
->f_op
!= &bpf_map_fops
) {
213 return ERR_PTR(-EINVAL
);
216 return f
.file
->private_data
;
219 void bpf_map_inc(struct bpf_map
*map
, bool uref
)
221 atomic_inc(&map
->refcnt
);
223 atomic_inc(&map
->usercnt
);
226 struct bpf_map
*bpf_map_get_with_uref(u32 ufd
)
228 struct fd f
= fdget(ufd
);
231 map
= __bpf_map_get(f
);
235 bpf_map_inc(map
, true);
241 /* helper to convert user pointers passed inside __aligned_u64 fields */
242 static void __user
*u64_to_ptr(__u64 val
)
244 return (void __user
*) (unsigned long) val
;
247 int __weak
bpf_stackmap_copy(struct bpf_map
*map
, void *key
, void *value
)
252 /* last field in 'union bpf_attr' used by this command */
253 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
255 static int map_lookup_elem(union bpf_attr
*attr
)
257 void __user
*ukey
= u64_to_ptr(attr
->key
);
258 void __user
*uvalue
= u64_to_ptr(attr
->value
);
259 int ufd
= attr
->map_fd
;
261 void *key
, *value
, *ptr
;
266 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM
))
270 map
= __bpf_map_get(f
);
275 key
= kmalloc(map
->key_size
, GFP_USER
);
280 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
283 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
||
284 map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
)
285 value_size
= round_up(map
->value_size
, 8) * num_possible_cpus();
287 value_size
= map
->value_size
;
290 value
= kmalloc(value_size
, GFP_USER
| __GFP_NOWARN
);
294 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
) {
295 err
= bpf_percpu_hash_copy(map
, key
, value
);
296 } else if (map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
) {
297 err
= bpf_percpu_array_copy(map
, key
, value
);
298 } else if (map
->map_type
== BPF_MAP_TYPE_STACK_TRACE
) {
299 err
= bpf_stackmap_copy(map
, key
, value
);
302 ptr
= map
->ops
->map_lookup_elem(map
, key
);
304 memcpy(value
, ptr
, value_size
);
306 err
= ptr
? 0 : -ENOENT
;
313 if (copy_to_user(uvalue
, value
, value_size
) != 0)
327 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
329 static int map_update_elem(union bpf_attr
*attr
)
331 void __user
*ukey
= u64_to_ptr(attr
->key
);
332 void __user
*uvalue
= u64_to_ptr(attr
->value
);
333 int ufd
= attr
->map_fd
;
340 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM
))
344 map
= __bpf_map_get(f
);
349 key
= kmalloc(map
->key_size
, GFP_USER
);
354 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
357 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
||
358 map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
)
359 value_size
= round_up(map
->value_size
, 8) * num_possible_cpus();
361 value_size
= map
->value_size
;
364 value
= kmalloc(value_size
, GFP_USER
| __GFP_NOWARN
);
369 if (copy_from_user(value
, uvalue
, value_size
) != 0)
372 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
373 * inside bpf map update or delete otherwise deadlocks are possible
376 __this_cpu_inc(bpf_prog_active
);
377 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
) {
378 err
= bpf_percpu_hash_update(map
, key
, value
, attr
->flags
);
379 } else if (map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
) {
380 err
= bpf_percpu_array_update(map
, key
, value
, attr
->flags
);
383 err
= map
->ops
->map_update_elem(map
, key
, value
, attr
->flags
);
386 __this_cpu_dec(bpf_prog_active
);
398 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
400 static int map_delete_elem(union bpf_attr
*attr
)
402 void __user
*ukey
= u64_to_ptr(attr
->key
);
403 int ufd
= attr
->map_fd
;
409 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM
))
413 map
= __bpf_map_get(f
);
418 key
= kmalloc(map
->key_size
, GFP_USER
);
423 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
427 __this_cpu_inc(bpf_prog_active
);
429 err
= map
->ops
->map_delete_elem(map
, key
);
431 __this_cpu_dec(bpf_prog_active
);
441 /* last field in 'union bpf_attr' used by this command */
442 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
444 static int map_get_next_key(union bpf_attr
*attr
)
446 void __user
*ukey
= u64_to_ptr(attr
->key
);
447 void __user
*unext_key
= u64_to_ptr(attr
->next_key
);
448 int ufd
= attr
->map_fd
;
450 void *key
, *next_key
;
454 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY
))
458 map
= __bpf_map_get(f
);
463 key
= kmalloc(map
->key_size
, GFP_USER
);
468 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
472 next_key
= kmalloc(map
->key_size
, GFP_USER
);
477 err
= map
->ops
->map_get_next_key(map
, key
, next_key
);
483 if (copy_to_user(unext_key
, next_key
, map
->key_size
) != 0)
497 static LIST_HEAD(bpf_prog_types
);
499 static int find_prog_type(enum bpf_prog_type type
, struct bpf_prog
*prog
)
501 struct bpf_prog_type_list
*tl
;
503 list_for_each_entry(tl
, &bpf_prog_types
, list_node
) {
504 if (tl
->type
== type
) {
505 prog
->aux
->ops
= tl
->ops
;
514 void bpf_register_prog_type(struct bpf_prog_type_list
*tl
)
516 list_add(&tl
->list_node
, &bpf_prog_types
);
519 /* fixup insn->imm field of bpf_call instructions:
520 * if (insn->imm == BPF_FUNC_map_lookup_elem)
521 * insn->imm = bpf_map_lookup_elem - __bpf_call_base;
522 * else if (insn->imm == BPF_FUNC_map_update_elem)
523 * insn->imm = bpf_map_update_elem - __bpf_call_base;
526 * this function is called after eBPF program passed verification
528 static void fixup_bpf_calls(struct bpf_prog
*prog
)
530 const struct bpf_func_proto
*fn
;
533 for (i
= 0; i
< prog
->len
; i
++) {
534 struct bpf_insn
*insn
= &prog
->insnsi
[i
];
536 if (insn
->code
== (BPF_JMP
| BPF_CALL
)) {
537 /* we reach here when program has bpf_call instructions
538 * and it passed bpf_check(), means that
539 * ops->get_func_proto must have been supplied, check it
541 BUG_ON(!prog
->aux
->ops
->get_func_proto
);
543 if (insn
->imm
== BPF_FUNC_get_route_realm
)
544 prog
->dst_needed
= 1;
545 if (insn
->imm
== BPF_FUNC_get_prandom_u32
)
546 bpf_user_rnd_init_once();
547 if (insn
->imm
== BPF_FUNC_tail_call
) {
548 /* mark bpf_tail_call as different opcode
549 * to avoid conditional branch in
550 * interpeter for every normal call
551 * and to prevent accidental JITing by
552 * JIT compiler that doesn't support
560 fn
= prog
->aux
->ops
->get_func_proto(insn
->imm
);
561 /* all functions that have prototype and verifier allowed
562 * programs to call them, must be real in-kernel functions
565 insn
->imm
= fn
->func
- __bpf_call_base
;
570 /* drop refcnt on maps used by eBPF program and free auxilary data */
571 static void free_used_maps(struct bpf_prog_aux
*aux
)
575 for (i
= 0; i
< aux
->used_map_cnt
; i
++)
576 bpf_map_put(aux
->used_maps
[i
]);
578 kfree(aux
->used_maps
);
581 static int bpf_prog_charge_memlock(struct bpf_prog
*prog
)
583 struct user_struct
*user
= get_current_user();
584 unsigned long memlock_limit
;
586 memlock_limit
= rlimit(RLIMIT_MEMLOCK
) >> PAGE_SHIFT
;
588 atomic_long_add(prog
->pages
, &user
->locked_vm
);
589 if (atomic_long_read(&user
->locked_vm
) > memlock_limit
) {
590 atomic_long_sub(prog
->pages
, &user
->locked_vm
);
594 prog
->aux
->user
= user
;
598 static void bpf_prog_uncharge_memlock(struct bpf_prog
*prog
)
600 struct user_struct
*user
= prog
->aux
->user
;
602 atomic_long_sub(prog
->pages
, &user
->locked_vm
);
606 static void __prog_put_common(struct rcu_head
*rcu
)
608 struct bpf_prog_aux
*aux
= container_of(rcu
, struct bpf_prog_aux
, rcu
);
611 bpf_prog_uncharge_memlock(aux
->prog
);
612 bpf_prog_free(aux
->prog
);
615 /* version of bpf_prog_put() that is called after a grace period */
616 void bpf_prog_put_rcu(struct bpf_prog
*prog
)
618 if (atomic_dec_and_test(&prog
->aux
->refcnt
))
619 call_rcu(&prog
->aux
->rcu
, __prog_put_common
);
622 void bpf_prog_put(struct bpf_prog
*prog
)
624 if (atomic_dec_and_test(&prog
->aux
->refcnt
))
625 __prog_put_common(&prog
->aux
->rcu
);
627 EXPORT_SYMBOL_GPL(bpf_prog_put
);
629 static int bpf_prog_release(struct inode
*inode
, struct file
*filp
)
631 struct bpf_prog
*prog
= filp
->private_data
;
633 bpf_prog_put_rcu(prog
);
637 static const struct file_operations bpf_prog_fops
= {
638 .release
= bpf_prog_release
,
641 int bpf_prog_new_fd(struct bpf_prog
*prog
)
643 return anon_inode_getfd("bpf-prog", &bpf_prog_fops
, prog
,
647 static struct bpf_prog
*__bpf_prog_get(struct fd f
)
650 return ERR_PTR(-EBADF
);
651 if (f
.file
->f_op
!= &bpf_prog_fops
) {
653 return ERR_PTR(-EINVAL
);
656 return f
.file
->private_data
;
659 /* called by sockets/tracing/seccomp before attaching program to an event
660 * pairs with bpf_prog_put()
662 struct bpf_prog
*bpf_prog_get(u32 ufd
)
664 struct fd f
= fdget(ufd
);
665 struct bpf_prog
*prog
;
667 prog
= __bpf_prog_get(f
);
671 atomic_inc(&prog
->aux
->refcnt
);
676 EXPORT_SYMBOL_GPL(bpf_prog_get
);
678 /* last field in 'union bpf_attr' used by this command */
679 #define BPF_PROG_LOAD_LAST_FIELD kern_version
681 static int bpf_prog_load(union bpf_attr
*attr
)
683 enum bpf_prog_type type
= attr
->prog_type
;
684 struct bpf_prog
*prog
;
689 if (CHECK_ATTR(BPF_PROG_LOAD
))
692 /* copy eBPF program license from user space */
693 if (strncpy_from_user(license
, u64_to_ptr(attr
->license
),
694 sizeof(license
) - 1) < 0)
696 license
[sizeof(license
) - 1] = 0;
698 /* eBPF programs must be GPL compatible to use GPL-ed functions */
699 is_gpl
= license_is_gpl_compatible(license
);
701 if (attr
->insn_cnt
>= BPF_MAXINSNS
)
704 if (type
== BPF_PROG_TYPE_KPROBE
&&
705 attr
->kern_version
!= LINUX_VERSION_CODE
)
708 if (type
!= BPF_PROG_TYPE_SOCKET_FILTER
&& !capable(CAP_SYS_ADMIN
))
711 /* plain bpf_prog allocation */
712 prog
= bpf_prog_alloc(bpf_prog_size(attr
->insn_cnt
), GFP_USER
);
716 err
= bpf_prog_charge_memlock(prog
);
718 goto free_prog_nouncharge
;
720 prog
->len
= attr
->insn_cnt
;
723 if (copy_from_user(prog
->insns
, u64_to_ptr(attr
->insns
),
724 prog
->len
* sizeof(struct bpf_insn
)) != 0)
727 prog
->orig_prog
= NULL
;
730 atomic_set(&prog
->aux
->refcnt
, 1);
731 prog
->gpl_compatible
= is_gpl
? 1 : 0;
733 /* find program type: socket_filter vs tracing_filter */
734 err
= find_prog_type(type
, prog
);
738 /* run eBPF verifier */
739 err
= bpf_check(&prog
, attr
);
743 /* fixup BPF_CALL->imm field */
744 fixup_bpf_calls(prog
);
746 /* eBPF program is ready to be JITed */
747 err
= bpf_prog_select_runtime(prog
);
751 err
= bpf_prog_new_fd(prog
);
753 /* failed to allocate fd */
759 free_used_maps(prog
->aux
);
761 bpf_prog_uncharge_memlock(prog
);
762 free_prog_nouncharge
:
767 #define BPF_OBJ_LAST_FIELD bpf_fd
769 static int bpf_obj_pin(const union bpf_attr
*attr
)
771 if (CHECK_ATTR(BPF_OBJ
))
774 return bpf_obj_pin_user(attr
->bpf_fd
, u64_to_ptr(attr
->pathname
));
777 static int bpf_obj_get(const union bpf_attr
*attr
)
779 if (CHECK_ATTR(BPF_OBJ
) || attr
->bpf_fd
!= 0)
782 return bpf_obj_get_user(u64_to_ptr(attr
->pathname
));
785 SYSCALL_DEFINE3(bpf
, int, cmd
, union bpf_attr __user
*, uattr
, unsigned int, size
)
787 union bpf_attr attr
= {};
790 if (!capable(CAP_SYS_ADMIN
) && sysctl_unprivileged_bpf_disabled
)
793 if (!access_ok(VERIFY_READ
, uattr
, 1))
796 if (size
> PAGE_SIZE
) /* silly large */
799 /* If we're handed a bigger struct than we know of,
800 * ensure all the unknown bits are 0 - i.e. new
801 * user-space does not rely on any kernel feature
802 * extensions we dont know about yet.
804 if (size
> sizeof(attr
)) {
805 unsigned char __user
*addr
;
806 unsigned char __user
*end
;
809 addr
= (void __user
*)uattr
+ sizeof(attr
);
810 end
= (void __user
*)uattr
+ size
;
812 for (; addr
< end
; addr
++) {
813 err
= get_user(val
, addr
);
822 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
823 if (copy_from_user(&attr
, uattr
, size
) != 0)
828 err
= map_create(&attr
);
830 case BPF_MAP_LOOKUP_ELEM
:
831 err
= map_lookup_elem(&attr
);
833 case BPF_MAP_UPDATE_ELEM
:
834 err
= map_update_elem(&attr
);
836 case BPF_MAP_DELETE_ELEM
:
837 err
= map_delete_elem(&attr
);
839 case BPF_MAP_GET_NEXT_KEY
:
840 err
= map_get_next_key(&attr
);
843 err
= bpf_prog_load(&attr
);
846 err
= bpf_obj_pin(&attr
);
849 err
= bpf_obj_get(&attr
);