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
;
245 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM
))
249 map
= __bpf_map_get(f
);
254 key
= kmalloc(map
->key_size
, GFP_USER
);
259 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
263 value
= kmalloc(map
->value_size
, GFP_USER
| __GFP_NOWARN
);
268 ptr
= map
->ops
->map_lookup_elem(map
, key
);
270 memcpy(value
, ptr
, map
->value_size
);
278 if (copy_to_user(uvalue
, value
, map
->value_size
) != 0)
292 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
294 static int map_update_elem(union bpf_attr
*attr
)
296 void __user
*ukey
= u64_to_ptr(attr
->key
);
297 void __user
*uvalue
= u64_to_ptr(attr
->value
);
298 int ufd
= attr
->map_fd
;
304 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM
))
308 map
= __bpf_map_get(f
);
313 key
= kmalloc(map
->key_size
, GFP_USER
);
318 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
322 value
= kmalloc(map
->value_size
, GFP_USER
| __GFP_NOWARN
);
327 if (copy_from_user(value
, uvalue
, map
->value_size
) != 0)
330 /* eBPF program that use maps are running under rcu_read_lock(),
331 * therefore all map accessors rely on this fact, so do the same here
334 err
= map
->ops
->map_update_elem(map
, key
, value
, attr
->flags
);
346 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
348 static int map_delete_elem(union bpf_attr
*attr
)
350 void __user
*ukey
= u64_to_ptr(attr
->key
);
351 int ufd
= attr
->map_fd
;
357 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM
))
361 map
= __bpf_map_get(f
);
366 key
= kmalloc(map
->key_size
, GFP_USER
);
371 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
375 err
= map
->ops
->map_delete_elem(map
, key
);
385 /* last field in 'union bpf_attr' used by this command */
386 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
388 static int map_get_next_key(union bpf_attr
*attr
)
390 void __user
*ukey
= u64_to_ptr(attr
->key
);
391 void __user
*unext_key
= u64_to_ptr(attr
->next_key
);
392 int ufd
= attr
->map_fd
;
394 void *key
, *next_key
;
398 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY
))
402 map
= __bpf_map_get(f
);
407 key
= kmalloc(map
->key_size
, GFP_USER
);
412 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
416 next_key
= kmalloc(map
->key_size
, GFP_USER
);
421 err
= map
->ops
->map_get_next_key(map
, key
, next_key
);
427 if (copy_to_user(unext_key
, next_key
, map
->key_size
) != 0)
441 static LIST_HEAD(bpf_prog_types
);
443 static int find_prog_type(enum bpf_prog_type type
, struct bpf_prog
*prog
)
445 struct bpf_prog_type_list
*tl
;
447 list_for_each_entry(tl
, &bpf_prog_types
, list_node
) {
448 if (tl
->type
== type
) {
449 prog
->aux
->ops
= tl
->ops
;
458 void bpf_register_prog_type(struct bpf_prog_type_list
*tl
)
460 list_add(&tl
->list_node
, &bpf_prog_types
);
463 /* fixup insn->imm field of bpf_call instructions:
464 * if (insn->imm == BPF_FUNC_map_lookup_elem)
465 * insn->imm = bpf_map_lookup_elem - __bpf_call_base;
466 * else if (insn->imm == BPF_FUNC_map_update_elem)
467 * insn->imm = bpf_map_update_elem - __bpf_call_base;
470 * this function is called after eBPF program passed verification
472 static void fixup_bpf_calls(struct bpf_prog
*prog
)
474 const struct bpf_func_proto
*fn
;
477 for (i
= 0; i
< prog
->len
; i
++) {
478 struct bpf_insn
*insn
= &prog
->insnsi
[i
];
480 if (insn
->code
== (BPF_JMP
| BPF_CALL
)) {
481 /* we reach here when program has bpf_call instructions
482 * and it passed bpf_check(), means that
483 * ops->get_func_proto must have been supplied, check it
485 BUG_ON(!prog
->aux
->ops
->get_func_proto
);
487 if (insn
->imm
== BPF_FUNC_get_route_realm
)
488 prog
->dst_needed
= 1;
489 if (insn
->imm
== BPF_FUNC_get_prandom_u32
)
490 bpf_user_rnd_init_once();
491 if (insn
->imm
== BPF_FUNC_tail_call
) {
492 /* mark bpf_tail_call as different opcode
493 * to avoid conditional branch in
494 * interpeter for every normal call
495 * and to prevent accidental JITing by
496 * JIT compiler that doesn't support
504 fn
= prog
->aux
->ops
->get_func_proto(insn
->imm
);
505 /* all functions that have prototype and verifier allowed
506 * programs to call them, must be real in-kernel functions
509 insn
->imm
= fn
->func
- __bpf_call_base
;
514 /* drop refcnt on maps used by eBPF program and free auxilary data */
515 static void free_used_maps(struct bpf_prog_aux
*aux
)
519 for (i
= 0; i
< aux
->used_map_cnt
; i
++)
520 bpf_map_put(aux
->used_maps
[i
]);
522 kfree(aux
->used_maps
);
525 static int bpf_prog_charge_memlock(struct bpf_prog
*prog
)
527 struct user_struct
*user
= get_current_user();
528 unsigned long memlock_limit
;
530 memlock_limit
= rlimit(RLIMIT_MEMLOCK
) >> PAGE_SHIFT
;
532 atomic_long_add(prog
->pages
, &user
->locked_vm
);
533 if (atomic_long_read(&user
->locked_vm
) > memlock_limit
) {
534 atomic_long_sub(prog
->pages
, &user
->locked_vm
);
538 prog
->aux
->user
= user
;
542 static void bpf_prog_uncharge_memlock(struct bpf_prog
*prog
)
544 struct user_struct
*user
= prog
->aux
->user
;
546 atomic_long_sub(prog
->pages
, &user
->locked_vm
);
550 static void __prog_put_common(struct rcu_head
*rcu
)
552 struct bpf_prog_aux
*aux
= container_of(rcu
, struct bpf_prog_aux
, rcu
);
555 bpf_prog_uncharge_memlock(aux
->prog
);
556 bpf_prog_free(aux
->prog
);
559 /* version of bpf_prog_put() that is called after a grace period */
560 void bpf_prog_put_rcu(struct bpf_prog
*prog
)
562 if (atomic_dec_and_test(&prog
->aux
->refcnt
))
563 call_rcu(&prog
->aux
->rcu
, __prog_put_common
);
566 void bpf_prog_put(struct bpf_prog
*prog
)
568 if (atomic_dec_and_test(&prog
->aux
->refcnt
))
569 __prog_put_common(&prog
->aux
->rcu
);
571 EXPORT_SYMBOL_GPL(bpf_prog_put
);
573 static int bpf_prog_release(struct inode
*inode
, struct file
*filp
)
575 struct bpf_prog
*prog
= filp
->private_data
;
577 bpf_prog_put_rcu(prog
);
581 static const struct file_operations bpf_prog_fops
= {
582 .release
= bpf_prog_release
,
585 int bpf_prog_new_fd(struct bpf_prog
*prog
)
587 return anon_inode_getfd("bpf-prog", &bpf_prog_fops
, prog
,
591 static struct bpf_prog
*__bpf_prog_get(struct fd f
)
594 return ERR_PTR(-EBADF
);
595 if (f
.file
->f_op
!= &bpf_prog_fops
) {
597 return ERR_PTR(-EINVAL
);
600 return f
.file
->private_data
;
603 /* called by sockets/tracing/seccomp before attaching program to an event
604 * pairs with bpf_prog_put()
606 struct bpf_prog
*bpf_prog_get(u32 ufd
)
608 struct fd f
= fdget(ufd
);
609 struct bpf_prog
*prog
;
611 prog
= __bpf_prog_get(f
);
615 atomic_inc(&prog
->aux
->refcnt
);
620 EXPORT_SYMBOL_GPL(bpf_prog_get
);
622 /* last field in 'union bpf_attr' used by this command */
623 #define BPF_PROG_LOAD_LAST_FIELD kern_version
625 static int bpf_prog_load(union bpf_attr
*attr
)
627 enum bpf_prog_type type
= attr
->prog_type
;
628 struct bpf_prog
*prog
;
633 if (CHECK_ATTR(BPF_PROG_LOAD
))
636 /* copy eBPF program license from user space */
637 if (strncpy_from_user(license
, u64_to_ptr(attr
->license
),
638 sizeof(license
) - 1) < 0)
640 license
[sizeof(license
) - 1] = 0;
642 /* eBPF programs must be GPL compatible to use GPL-ed functions */
643 is_gpl
= license_is_gpl_compatible(license
);
645 if (attr
->insn_cnt
>= BPF_MAXINSNS
)
648 if (type
== BPF_PROG_TYPE_KPROBE
&&
649 attr
->kern_version
!= LINUX_VERSION_CODE
)
652 if (type
!= BPF_PROG_TYPE_SOCKET_FILTER
&& !capable(CAP_SYS_ADMIN
))
655 /* plain bpf_prog allocation */
656 prog
= bpf_prog_alloc(bpf_prog_size(attr
->insn_cnt
), GFP_USER
);
660 err
= bpf_prog_charge_memlock(prog
);
662 goto free_prog_nouncharge
;
664 prog
->len
= attr
->insn_cnt
;
667 if (copy_from_user(prog
->insns
, u64_to_ptr(attr
->insns
),
668 prog
->len
* sizeof(struct bpf_insn
)) != 0)
671 prog
->orig_prog
= NULL
;
674 atomic_set(&prog
->aux
->refcnt
, 1);
675 prog
->gpl_compatible
= is_gpl
? 1 : 0;
677 /* find program type: socket_filter vs tracing_filter */
678 err
= find_prog_type(type
, prog
);
682 /* run eBPF verifier */
683 err
= bpf_check(&prog
, attr
);
687 /* fixup BPF_CALL->imm field */
688 fixup_bpf_calls(prog
);
690 /* eBPF program is ready to be JITed */
691 err
= bpf_prog_select_runtime(prog
);
695 err
= bpf_prog_new_fd(prog
);
697 /* failed to allocate fd */
703 free_used_maps(prog
->aux
);
705 bpf_prog_uncharge_memlock(prog
);
706 free_prog_nouncharge
:
711 #define BPF_OBJ_LAST_FIELD bpf_fd
713 static int bpf_obj_pin(const union bpf_attr
*attr
)
715 if (CHECK_ATTR(BPF_OBJ
))
718 return bpf_obj_pin_user(attr
->bpf_fd
, u64_to_ptr(attr
->pathname
));
721 static int bpf_obj_get(const union bpf_attr
*attr
)
723 if (CHECK_ATTR(BPF_OBJ
) || attr
->bpf_fd
!= 0)
726 return bpf_obj_get_user(u64_to_ptr(attr
->pathname
));
729 SYSCALL_DEFINE3(bpf
, int, cmd
, union bpf_attr __user
*, uattr
, unsigned int, size
)
731 union bpf_attr attr
= {};
734 if (!capable(CAP_SYS_ADMIN
) && sysctl_unprivileged_bpf_disabled
)
737 if (!access_ok(VERIFY_READ
, uattr
, 1))
740 if (size
> PAGE_SIZE
) /* silly large */
743 /* If we're handed a bigger struct than we know of,
744 * ensure all the unknown bits are 0 - i.e. new
745 * user-space does not rely on any kernel feature
746 * extensions we dont know about yet.
748 if (size
> sizeof(attr
)) {
749 unsigned char __user
*addr
;
750 unsigned char __user
*end
;
753 addr
= (void __user
*)uattr
+ sizeof(attr
);
754 end
= (void __user
*)uattr
+ size
;
756 for (; addr
< end
; addr
++) {
757 err
= get_user(val
, addr
);
766 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
767 if (copy_from_user(&attr
, uattr
, size
) != 0)
772 err
= map_create(&attr
);
774 case BPF_MAP_LOOKUP_ELEM
:
775 err
= map_lookup_elem(&attr
);
777 case BPF_MAP_UPDATE_ELEM
:
778 err
= map_update_elem(&attr
);
780 case BPF_MAP_DELETE_ELEM
:
781 err
= map_delete_elem(&attr
);
783 case BPF_MAP_GET_NEXT_KEY
:
784 err
= map_get_next_key(&attr
);
787 err
= bpf_prog_load(&attr
);
790 err
= bpf_obj_pin(&attr
);
793 err
= bpf_obj_get(&attr
);