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>
20 static LIST_HEAD(bpf_map_types
);
22 static struct bpf_map
*find_and_alloc_map(union bpf_attr
*attr
)
24 struct bpf_map_type_list
*tl
;
27 list_for_each_entry(tl
, &bpf_map_types
, list_node
) {
28 if (tl
->type
== attr
->map_type
) {
29 map
= tl
->ops
->map_alloc(attr
);
33 map
->map_type
= attr
->map_type
;
37 return ERR_PTR(-EINVAL
);
40 /* boot time registration of different map implementations */
41 void bpf_register_map_type(struct bpf_map_type_list
*tl
)
43 list_add(&tl
->list_node
, &bpf_map_types
);
46 /* called from workqueue */
47 static void bpf_map_free_deferred(struct work_struct
*work
)
49 struct bpf_map
*map
= container_of(work
, struct bpf_map
, work
);
51 /* implementation dependent freeing */
52 map
->ops
->map_free(map
);
55 /* decrement map refcnt and schedule it for freeing via workqueue
56 * (unrelying map implementation ops->map_free() might sleep)
58 void bpf_map_put(struct bpf_map
*map
)
60 if (atomic_dec_and_test(&map
->refcnt
)) {
61 INIT_WORK(&map
->work
, bpf_map_free_deferred
);
62 schedule_work(&map
->work
);
66 static int bpf_map_release(struct inode
*inode
, struct file
*filp
)
68 struct bpf_map
*map
= filp
->private_data
;
74 static const struct file_operations bpf_map_fops
= {
75 .release
= bpf_map_release
,
78 /* helper macro to check that unused fields 'union bpf_attr' are zero */
79 #define CHECK_ATTR(CMD) \
80 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
81 sizeof(attr->CMD##_LAST_FIELD), 0, \
83 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
84 sizeof(attr->CMD##_LAST_FIELD)) != NULL
86 #define BPF_MAP_CREATE_LAST_FIELD max_entries
87 /* called via syscall */
88 static int map_create(union bpf_attr
*attr
)
93 err
= CHECK_ATTR(BPF_MAP_CREATE
);
97 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
98 map
= find_and_alloc_map(attr
);
102 atomic_set(&map
->refcnt
, 1);
104 err
= anon_inode_getfd("bpf-map", &bpf_map_fops
, map
, O_RDWR
| O_CLOEXEC
);
107 /* failed to allocate fd */
113 map
->ops
->map_free(map
);
117 /* if error is returned, fd is released.
118 * On success caller should complete fd access with matching fdput()
120 struct bpf_map
*bpf_map_get(struct fd f
)
125 return ERR_PTR(-EBADF
);
127 if (f
.file
->f_op
!= &bpf_map_fops
) {
129 return ERR_PTR(-EINVAL
);
132 map
= f
.file
->private_data
;
137 /* helper to convert user pointers passed inside __aligned_u64 fields */
138 static void __user
*u64_to_ptr(__u64 val
)
140 return (void __user
*) (unsigned long) val
;
143 /* last field in 'union bpf_attr' used by this command */
144 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
146 static int map_lookup_elem(union bpf_attr
*attr
)
148 void __user
*ukey
= u64_to_ptr(attr
->key
);
149 void __user
*uvalue
= u64_to_ptr(attr
->value
);
150 int ufd
= attr
->map_fd
;
151 struct fd f
= fdget(ufd
);
153 void *key
, *value
, *ptr
;
156 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM
))
159 map
= bpf_map_get(f
);
164 key
= kmalloc(map
->key_size
, GFP_USER
);
169 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
173 value
= kmalloc(map
->value_size
, GFP_USER
);
178 ptr
= map
->ops
->map_lookup_elem(map
, key
);
180 memcpy(value
, ptr
, map
->value_size
);
188 if (copy_to_user(uvalue
, value
, map
->value_size
) != 0)
202 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
204 static int map_update_elem(union bpf_attr
*attr
)
206 void __user
*ukey
= u64_to_ptr(attr
->key
);
207 void __user
*uvalue
= u64_to_ptr(attr
->value
);
208 int ufd
= attr
->map_fd
;
209 struct fd f
= fdget(ufd
);
214 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM
))
217 map
= bpf_map_get(f
);
222 key
= kmalloc(map
->key_size
, GFP_USER
);
227 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
231 value
= kmalloc(map
->value_size
, GFP_USER
);
236 if (copy_from_user(value
, uvalue
, map
->value_size
) != 0)
239 /* eBPF program that use maps are running under rcu_read_lock(),
240 * therefore all map accessors rely on this fact, so do the same here
243 err
= map
->ops
->map_update_elem(map
, key
, value
, attr
->flags
);
255 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
257 static int map_delete_elem(union bpf_attr
*attr
)
259 void __user
*ukey
= u64_to_ptr(attr
->key
);
260 int ufd
= attr
->map_fd
;
261 struct fd f
= fdget(ufd
);
266 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM
))
269 map
= bpf_map_get(f
);
274 key
= kmalloc(map
->key_size
, GFP_USER
);
279 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
283 err
= map
->ops
->map_delete_elem(map
, key
);
293 /* last field in 'union bpf_attr' used by this command */
294 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
296 static int map_get_next_key(union bpf_attr
*attr
)
298 void __user
*ukey
= u64_to_ptr(attr
->key
);
299 void __user
*unext_key
= u64_to_ptr(attr
->next_key
);
300 int ufd
= attr
->map_fd
;
301 struct fd f
= fdget(ufd
);
303 void *key
, *next_key
;
306 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY
))
309 map
= bpf_map_get(f
);
314 key
= kmalloc(map
->key_size
, GFP_USER
);
319 if (copy_from_user(key
, ukey
, map
->key_size
) != 0)
323 next_key
= kmalloc(map
->key_size
, GFP_USER
);
328 err
= map
->ops
->map_get_next_key(map
, key
, next_key
);
334 if (copy_to_user(unext_key
, next_key
, map
->key_size
) != 0)
348 static LIST_HEAD(bpf_prog_types
);
350 static int find_prog_type(enum bpf_prog_type type
, struct bpf_prog
*prog
)
352 struct bpf_prog_type_list
*tl
;
354 list_for_each_entry(tl
, &bpf_prog_types
, list_node
) {
355 if (tl
->type
== type
) {
356 prog
->aux
->ops
= tl
->ops
;
365 void bpf_register_prog_type(struct bpf_prog_type_list
*tl
)
367 list_add(&tl
->list_node
, &bpf_prog_types
);
370 /* fixup insn->imm field of bpf_call instructions:
371 * if (insn->imm == BPF_FUNC_map_lookup_elem)
372 * insn->imm = bpf_map_lookup_elem - __bpf_call_base;
373 * else if (insn->imm == BPF_FUNC_map_update_elem)
374 * insn->imm = bpf_map_update_elem - __bpf_call_base;
377 * this function is called after eBPF program passed verification
379 static void fixup_bpf_calls(struct bpf_prog
*prog
)
381 const struct bpf_func_proto
*fn
;
384 for (i
= 0; i
< prog
->len
; i
++) {
385 struct bpf_insn
*insn
= &prog
->insnsi
[i
];
387 if (insn
->code
== (BPF_JMP
| BPF_CALL
)) {
388 /* we reach here when program has bpf_call instructions
389 * and it passed bpf_check(), means that
390 * ops->get_func_proto must have been supplied, check it
392 BUG_ON(!prog
->aux
->ops
->get_func_proto
);
394 fn
= prog
->aux
->ops
->get_func_proto(insn
->imm
);
395 /* all functions that have prototype and verifier allowed
396 * programs to call them, must be real in-kernel functions
399 insn
->imm
= fn
->func
- __bpf_call_base
;
404 /* drop refcnt on maps used by eBPF program and free auxilary data */
405 static void free_used_maps(struct bpf_prog_aux
*aux
)
409 for (i
= 0; i
< aux
->used_map_cnt
; i
++)
410 bpf_map_put(aux
->used_maps
[i
]);
412 kfree(aux
->used_maps
);
415 void bpf_prog_put(struct bpf_prog
*prog
)
417 if (atomic_dec_and_test(&prog
->aux
->refcnt
)) {
418 free_used_maps(prog
->aux
);
422 EXPORT_SYMBOL_GPL(bpf_prog_put
);
424 static int bpf_prog_release(struct inode
*inode
, struct file
*filp
)
426 struct bpf_prog
*prog
= filp
->private_data
;
432 static const struct file_operations bpf_prog_fops
= {
433 .release
= bpf_prog_release
,
436 static struct bpf_prog
*get_prog(struct fd f
)
438 struct bpf_prog
*prog
;
441 return ERR_PTR(-EBADF
);
443 if (f
.file
->f_op
!= &bpf_prog_fops
) {
445 return ERR_PTR(-EINVAL
);
448 prog
= f
.file
->private_data
;
453 /* called by sockets/tracing/seccomp before attaching program to an event
454 * pairs with bpf_prog_put()
456 struct bpf_prog
*bpf_prog_get(u32 ufd
)
458 struct fd f
= fdget(ufd
);
459 struct bpf_prog
*prog
;
466 atomic_inc(&prog
->aux
->refcnt
);
470 EXPORT_SYMBOL_GPL(bpf_prog_get
);
472 /* last field in 'union bpf_attr' used by this command */
473 #define BPF_PROG_LOAD_LAST_FIELD log_buf
475 static int bpf_prog_load(union bpf_attr
*attr
)
477 enum bpf_prog_type type
= attr
->prog_type
;
478 struct bpf_prog
*prog
;
483 if (CHECK_ATTR(BPF_PROG_LOAD
))
486 /* copy eBPF program license from user space */
487 if (strncpy_from_user(license
, u64_to_ptr(attr
->license
),
488 sizeof(license
) - 1) < 0)
490 license
[sizeof(license
) - 1] = 0;
492 /* eBPF programs must be GPL compatible to use GPL-ed functions */
493 is_gpl
= license_is_gpl_compatible(license
);
495 if (attr
->insn_cnt
>= BPF_MAXINSNS
)
498 /* plain bpf_prog allocation */
499 prog
= bpf_prog_alloc(bpf_prog_size(attr
->insn_cnt
), GFP_USER
);
503 prog
->len
= attr
->insn_cnt
;
506 if (copy_from_user(prog
->insns
, u64_to_ptr(attr
->insns
),
507 prog
->len
* sizeof(struct bpf_insn
)) != 0)
510 prog
->orig_prog
= NULL
;
513 atomic_set(&prog
->aux
->refcnt
, 1);
514 prog
->gpl_compatible
= is_gpl
;
516 /* find program type: socket_filter vs tracing_filter */
517 err
= find_prog_type(type
, prog
);
521 /* run eBPF verifier */
522 err
= bpf_check(prog
, attr
);
526 /* fixup BPF_CALL->imm field */
527 fixup_bpf_calls(prog
);
529 /* eBPF program is ready to be JITed */
530 bpf_prog_select_runtime(prog
);
532 err
= anon_inode_getfd("bpf-prog", &bpf_prog_fops
, prog
, O_RDWR
| O_CLOEXEC
);
534 /* failed to allocate fd */
540 free_used_maps(prog
->aux
);
546 SYSCALL_DEFINE3(bpf
, int, cmd
, union bpf_attr __user
*, uattr
, unsigned int, size
)
548 union bpf_attr attr
= {};
551 /* the syscall is limited to root temporarily. This restriction will be
552 * lifted when security audit is clean. Note that eBPF+tracing must have
553 * this restriction, since it may pass kernel data to user space
555 if (!capable(CAP_SYS_ADMIN
))
558 if (!access_ok(VERIFY_READ
, uattr
, 1))
561 if (size
> PAGE_SIZE
) /* silly large */
564 /* If we're handed a bigger struct than we know of,
565 * ensure all the unknown bits are 0 - i.e. new
566 * user-space does not rely on any kernel feature
567 * extensions we dont know about yet.
569 if (size
> sizeof(attr
)) {
570 unsigned char __user
*addr
;
571 unsigned char __user
*end
;
574 addr
= (void __user
*)uattr
+ sizeof(attr
);
575 end
= (void __user
*)uattr
+ size
;
577 for (; addr
< end
; addr
++) {
578 err
= get_user(val
, addr
);
587 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
588 if (copy_from_user(&attr
, uattr
, size
) != 0)
593 err
= map_create(&attr
);
595 case BPF_MAP_LOOKUP_ELEM
:
596 err
= map_lookup_elem(&attr
);
598 case BPF_MAP_UPDATE_ELEM
:
599 err
= map_update_elem(&attr
);
601 case BPF_MAP_DELETE_ELEM
:
602 err
= map_delete_elem(&attr
);
604 case BPF_MAP_GET_NEXT_KEY
:
605 err
= map_get_next_key(&attr
);
608 err
= bpf_prog_load(&attr
);