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 struct bpf_map
*map
= filp
->private_data
;
129 if (map
->ops
->map_release
)
130 map
->ops
->map_release(map
, filp
);
132 bpf_map_put_with_uref(map
);
136 #ifdef CONFIG_PROC_FS
137 static void bpf_map_show_fdinfo(struct seq_file
*m
, struct file
*filp
)
139 const struct bpf_map
*map
= filp
->private_data
;
155 static const struct file_operations bpf_map_fops
= {
156 #ifdef CONFIG_PROC_FS
157 .show_fdinfo
= bpf_map_show_fdinfo
,
159 .release
= bpf_map_release
,
162 int bpf_map_new_fd(struct bpf_map
*map
)
164 return anon_inode_getfd("bpf-map", &bpf_map_fops
, map
,
168 /* helper macro to check that unused fields 'union bpf_attr' are zero */
169 #define CHECK_ATTR(CMD) \
170 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
171 sizeof(attr->CMD##_LAST_FIELD), 0, \
173 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
174 sizeof(attr->CMD##_LAST_FIELD)) != NULL
176 #define BPF_MAP_CREATE_LAST_FIELD map_flags
177 /* called via syscall */
178 static int map_create(union bpf_attr
*attr
)
183 err
= CHECK_ATTR(BPF_MAP_CREATE
);
187 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
188 map
= find_and_alloc_map(attr
);
192 atomic_set(&map
->refcnt
, 1);
193 atomic_set(&map
->usercnt
, 1);
195 err
= bpf_map_charge_memlock(map
);
199 err
= bpf_map_new_fd(map
);
201 /* failed to allocate fd */
207 map
->ops
->map_free(map
);
211 /* if error is returned, fd is released.
212 * On success caller should complete fd access with matching fdput()
214 struct bpf_map
*__bpf_map_get(struct fd f
)
217 return ERR_PTR(-EBADF
);
218 if (f
.file
->f_op
!= &bpf_map_fops
) {
220 return ERR_PTR(-EINVAL
);
223 return f
.file
->private_data
;
226 /* prog's and map's refcnt limit */
227 #define BPF_MAX_REFCNT 32768
229 struct bpf_map
*bpf_map_inc(struct bpf_map
*map
, bool uref
)
231 if (atomic_inc_return(&map
->refcnt
) > BPF_MAX_REFCNT
) {
232 atomic_dec(&map
->refcnt
);
233 return ERR_PTR(-EBUSY
);
236 atomic_inc(&map
->usercnt
);
240 struct bpf_map
*bpf_map_get_with_uref(u32 ufd
)
242 struct fd f
= fdget(ufd
);
245 map
= __bpf_map_get(f
);
249 map
= bpf_map_inc(map
, true);
255 /* helper to convert user pointers passed inside __aligned_u64 fields */
256 static void __user
*u64_to_ptr(__u64 val
)
258 return (void __user
*) (unsigned long) val
;
261 int __weak
bpf_stackmap_copy(struct bpf_map
*map
, void *key
, void *value
)
266 /* last field in 'union bpf_attr' used by this command */
267 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
269 static int map_lookup_elem(union bpf_attr
*attr
)
271 void __user
*ukey
= u64_to_ptr(attr
->key
);
272 void __user
*uvalue
= u64_to_ptr(attr
->value
);
273 int ufd
= attr
->map_fd
;
275 void *key
, *value
, *ptr
;
280 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM
))
284 map
= __bpf_map_get(f
);
289 key
= kmalloc(map
->key_size
, GFP_USER
);
294 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
297 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
||
298 map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
)
299 value_size
= round_up(map
->value_size
, 8) * num_possible_cpus();
301 value_size
= map
->value_size
;
304 value
= kmalloc(value_size
, GFP_USER
| __GFP_NOWARN
);
308 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
) {
309 err
= bpf_percpu_hash_copy(map
, key
, value
);
310 } else if (map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
) {
311 err
= bpf_percpu_array_copy(map
, key
, value
);
312 } else if (map
->map_type
== BPF_MAP_TYPE_STACK_TRACE
) {
313 err
= bpf_stackmap_copy(map
, key
, value
);
316 ptr
= map
->ops
->map_lookup_elem(map
, key
);
318 memcpy(value
, ptr
, value_size
);
320 err
= ptr
? 0 : -ENOENT
;
327 if (copy_to_user(uvalue
, value
, value_size
) != 0)
341 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
343 static int map_update_elem(union bpf_attr
*attr
)
345 void __user
*ukey
= u64_to_ptr(attr
->key
);
346 void __user
*uvalue
= u64_to_ptr(attr
->value
);
347 int ufd
= attr
->map_fd
;
354 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM
))
358 map
= __bpf_map_get(f
);
363 key
= kmalloc(map
->key_size
, GFP_USER
);
368 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
371 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
||
372 map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
)
373 value_size
= round_up(map
->value_size
, 8) * num_possible_cpus();
375 value_size
= map
->value_size
;
378 value
= kmalloc(value_size
, GFP_USER
| __GFP_NOWARN
);
383 if (copy_from_user(value
, uvalue
, value_size
) != 0)
386 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
387 * inside bpf map update or delete otherwise deadlocks are possible
390 __this_cpu_inc(bpf_prog_active
);
391 if (map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
) {
392 err
= bpf_percpu_hash_update(map
, key
, value
, attr
->flags
);
393 } else if (map
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
) {
394 err
= bpf_percpu_array_update(map
, key
, value
, attr
->flags
);
395 } else if (map
->map_type
== BPF_MAP_TYPE_PERF_EVENT_ARRAY
||
396 map
->map_type
== BPF_MAP_TYPE_PROG_ARRAY
) {
398 err
= bpf_fd_array_map_update_elem(map
, f
.file
, key
, value
,
403 err
= map
->ops
->map_update_elem(map
, key
, value
, attr
->flags
);
406 __this_cpu_dec(bpf_prog_active
);
418 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
420 static int map_delete_elem(union bpf_attr
*attr
)
422 void __user
*ukey
= u64_to_ptr(attr
->key
);
423 int ufd
= attr
->map_fd
;
429 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM
))
433 map
= __bpf_map_get(f
);
438 key
= kmalloc(map
->key_size
, GFP_USER
);
443 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
447 __this_cpu_inc(bpf_prog_active
);
449 err
= map
->ops
->map_delete_elem(map
, key
);
451 __this_cpu_dec(bpf_prog_active
);
461 /* last field in 'union bpf_attr' used by this command */
462 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
464 static int map_get_next_key(union bpf_attr
*attr
)
466 void __user
*ukey
= u64_to_ptr(attr
->key
);
467 void __user
*unext_key
= u64_to_ptr(attr
->next_key
);
468 int ufd
= attr
->map_fd
;
470 void *key
, *next_key
;
474 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY
))
478 map
= __bpf_map_get(f
);
483 key
= kmalloc(map
->key_size
, GFP_USER
);
488 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
492 next_key
= kmalloc(map
->key_size
, GFP_USER
);
497 err
= map
->ops
->map_get_next_key(map
, key
, next_key
);
503 if (copy_to_user(unext_key
, next_key
, map
->key_size
) != 0)
517 static LIST_HEAD(bpf_prog_types
);
519 static int find_prog_type(enum bpf_prog_type type
, struct bpf_prog
*prog
)
521 struct bpf_prog_type_list
*tl
;
523 list_for_each_entry(tl
, &bpf_prog_types
, list_node
) {
524 if (tl
->type
== type
) {
525 prog
->aux
->ops
= tl
->ops
;
534 void bpf_register_prog_type(struct bpf_prog_type_list
*tl
)
536 list_add(&tl
->list_node
, &bpf_prog_types
);
539 /* fixup insn->imm field of bpf_call instructions:
540 * if (insn->imm == BPF_FUNC_map_lookup_elem)
541 * insn->imm = bpf_map_lookup_elem - __bpf_call_base;
542 * else if (insn->imm == BPF_FUNC_map_update_elem)
543 * insn->imm = bpf_map_update_elem - __bpf_call_base;
546 * this function is called after eBPF program passed verification
548 static void fixup_bpf_calls(struct bpf_prog
*prog
)
550 const struct bpf_func_proto
*fn
;
553 for (i
= 0; i
< prog
->len
; i
++) {
554 struct bpf_insn
*insn
= &prog
->insnsi
[i
];
556 if (insn
->code
== (BPF_JMP
| BPF_CALL
)) {
557 /* we reach here when program has bpf_call instructions
558 * and it passed bpf_check(), means that
559 * ops->get_func_proto must have been supplied, check it
561 BUG_ON(!prog
->aux
->ops
->get_func_proto
);
563 if (insn
->imm
== BPF_FUNC_get_route_realm
)
564 prog
->dst_needed
= 1;
565 if (insn
->imm
== BPF_FUNC_get_prandom_u32
)
566 bpf_user_rnd_init_once();
567 if (insn
->imm
== BPF_FUNC_tail_call
) {
568 /* mark bpf_tail_call as different opcode
569 * to avoid conditional branch in
570 * interpeter for every normal call
571 * and to prevent accidental JITing by
572 * JIT compiler that doesn't support
580 fn
= prog
->aux
->ops
->get_func_proto(insn
->imm
);
581 /* all functions that have prototype and verifier allowed
582 * programs to call them, must be real in-kernel functions
585 insn
->imm
= fn
->func
- __bpf_call_base
;
590 /* drop refcnt on maps used by eBPF program and free auxilary data */
591 static void free_used_maps(struct bpf_prog_aux
*aux
)
595 for (i
= 0; i
< aux
->used_map_cnt
; i
++)
596 bpf_map_put(aux
->used_maps
[i
]);
598 kfree(aux
->used_maps
);
601 static int bpf_prog_charge_memlock(struct bpf_prog
*prog
)
603 struct user_struct
*user
= get_current_user();
604 unsigned long memlock_limit
;
606 memlock_limit
= rlimit(RLIMIT_MEMLOCK
) >> PAGE_SHIFT
;
608 atomic_long_add(prog
->pages
, &user
->locked_vm
);
609 if (atomic_long_read(&user
->locked_vm
) > memlock_limit
) {
610 atomic_long_sub(prog
->pages
, &user
->locked_vm
);
614 prog
->aux
->user
= user
;
618 static void bpf_prog_uncharge_memlock(struct bpf_prog
*prog
)
620 struct user_struct
*user
= prog
->aux
->user
;
622 atomic_long_sub(prog
->pages
, &user
->locked_vm
);
626 static void __prog_put_common(struct rcu_head
*rcu
)
628 struct bpf_prog_aux
*aux
= container_of(rcu
, struct bpf_prog_aux
, rcu
);
631 bpf_prog_uncharge_memlock(aux
->prog
);
632 bpf_prog_free(aux
->prog
);
635 /* version of bpf_prog_put() that is called after a grace period */
636 void bpf_prog_put_rcu(struct bpf_prog
*prog
)
638 if (atomic_dec_and_test(&prog
->aux
->refcnt
))
639 call_rcu(&prog
->aux
->rcu
, __prog_put_common
);
642 void bpf_prog_put(struct bpf_prog
*prog
)
644 if (atomic_dec_and_test(&prog
->aux
->refcnt
))
645 __prog_put_common(&prog
->aux
->rcu
);
647 EXPORT_SYMBOL_GPL(bpf_prog_put
);
649 static int bpf_prog_release(struct inode
*inode
, struct file
*filp
)
651 struct bpf_prog
*prog
= filp
->private_data
;
653 bpf_prog_put_rcu(prog
);
657 static const struct file_operations bpf_prog_fops
= {
658 .release
= bpf_prog_release
,
661 int bpf_prog_new_fd(struct bpf_prog
*prog
)
663 return anon_inode_getfd("bpf-prog", &bpf_prog_fops
, prog
,
667 static struct bpf_prog
*__bpf_prog_get(struct fd f
)
670 return ERR_PTR(-EBADF
);
671 if (f
.file
->f_op
!= &bpf_prog_fops
) {
673 return ERR_PTR(-EINVAL
);
676 return f
.file
->private_data
;
679 struct bpf_prog
*bpf_prog_inc(struct bpf_prog
*prog
)
681 if (atomic_inc_return(&prog
->aux
->refcnt
) > BPF_MAX_REFCNT
) {
682 atomic_dec(&prog
->aux
->refcnt
);
683 return ERR_PTR(-EBUSY
);
688 /* called by sockets/tracing/seccomp before attaching program to an event
689 * pairs with bpf_prog_put()
691 struct bpf_prog
*bpf_prog_get(u32 ufd
)
693 struct fd f
= fdget(ufd
);
694 struct bpf_prog
*prog
;
696 prog
= __bpf_prog_get(f
);
700 prog
= bpf_prog_inc(prog
);
705 EXPORT_SYMBOL_GPL(bpf_prog_get
);
707 /* last field in 'union bpf_attr' used by this command */
708 #define BPF_PROG_LOAD_LAST_FIELD kern_version
710 static int bpf_prog_load(union bpf_attr
*attr
)
712 enum bpf_prog_type type
= attr
->prog_type
;
713 struct bpf_prog
*prog
;
718 if (CHECK_ATTR(BPF_PROG_LOAD
))
721 /* copy eBPF program license from user space */
722 if (strncpy_from_user(license
, u64_to_ptr(attr
->license
),
723 sizeof(license
) - 1) < 0)
725 license
[sizeof(license
) - 1] = 0;
727 /* eBPF programs must be GPL compatible to use GPL-ed functions */
728 is_gpl
= license_is_gpl_compatible(license
);
730 if (attr
->insn_cnt
>= BPF_MAXINSNS
)
733 if (type
== BPF_PROG_TYPE_KPROBE
&&
734 attr
->kern_version
!= LINUX_VERSION_CODE
)
737 if (type
!= BPF_PROG_TYPE_SOCKET_FILTER
&& !capable(CAP_SYS_ADMIN
))
740 /* plain bpf_prog allocation */
741 prog
= bpf_prog_alloc(bpf_prog_size(attr
->insn_cnt
), GFP_USER
);
745 err
= bpf_prog_charge_memlock(prog
);
747 goto free_prog_nouncharge
;
749 prog
->len
= attr
->insn_cnt
;
752 if (copy_from_user(prog
->insns
, u64_to_ptr(attr
->insns
),
753 prog
->len
* sizeof(struct bpf_insn
)) != 0)
756 prog
->orig_prog
= NULL
;
759 atomic_set(&prog
->aux
->refcnt
, 1);
760 prog
->gpl_compatible
= is_gpl
? 1 : 0;
762 /* find program type: socket_filter vs tracing_filter */
763 err
= find_prog_type(type
, prog
);
767 /* run eBPF verifier */
768 err
= bpf_check(&prog
, attr
);
772 /* fixup BPF_CALL->imm field */
773 fixup_bpf_calls(prog
);
775 /* eBPF program is ready to be JITed */
776 prog
= bpf_prog_select_runtime(prog
, &err
);
780 err
= bpf_prog_new_fd(prog
);
782 /* failed to allocate fd */
788 free_used_maps(prog
->aux
);
790 bpf_prog_uncharge_memlock(prog
);
791 free_prog_nouncharge
:
796 #define BPF_OBJ_LAST_FIELD bpf_fd
798 static int bpf_obj_pin(const union bpf_attr
*attr
)
800 if (CHECK_ATTR(BPF_OBJ
))
803 return bpf_obj_pin_user(attr
->bpf_fd
, u64_to_ptr(attr
->pathname
));
806 static int bpf_obj_get(const union bpf_attr
*attr
)
808 if (CHECK_ATTR(BPF_OBJ
) || attr
->bpf_fd
!= 0)
811 return bpf_obj_get_user(u64_to_ptr(attr
->pathname
));
814 SYSCALL_DEFINE3(bpf
, int, cmd
, union bpf_attr __user
*, uattr
, unsigned int, size
)
816 union bpf_attr attr
= {};
819 if (!capable(CAP_SYS_ADMIN
) && sysctl_unprivileged_bpf_disabled
)
822 if (!access_ok(VERIFY_READ
, uattr
, 1))
825 if (size
> PAGE_SIZE
) /* silly large */
828 /* If we're handed a bigger struct than we know of,
829 * ensure all the unknown bits are 0 - i.e. new
830 * user-space does not rely on any kernel feature
831 * extensions we dont know about yet.
833 if (size
> sizeof(attr
)) {
834 unsigned char __user
*addr
;
835 unsigned char __user
*end
;
838 addr
= (void __user
*)uattr
+ sizeof(attr
);
839 end
= (void __user
*)uattr
+ size
;
841 for (; addr
< end
; addr
++) {
842 err
= get_user(val
, addr
);
851 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
852 if (copy_from_user(&attr
, uattr
, size
) != 0)
857 err
= map_create(&attr
);
859 case BPF_MAP_LOOKUP_ELEM
:
860 err
= map_lookup_elem(&attr
);
862 case BPF_MAP_UPDATE_ELEM
:
863 err
= map_update_elem(&attr
);
865 case BPF_MAP_DELETE_ELEM
:
866 err
= map_delete_elem(&attr
);
868 case BPF_MAP_GET_NEXT_KEY
:
869 err
= map_get_next_key(&attr
);
872 err
= bpf_prog_load(&attr
);
875 err
= bpf_obj_pin(&attr
);
878 err
= bpf_obj_get(&attr
);