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 static LIST_HEAD(bpf_map_types);
23 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
25 struct bpf_map_type_list *tl;
28 list_for_each_entry(tl, &bpf_map_types, list_node) {
29 if (tl->type == attr->map_type) {
30 map = tl->ops->map_alloc(attr);
34 map->map_type = attr->map_type;
38 return ERR_PTR(-EINVAL);
41 /* boot time registration of different map implementations */
42 void bpf_register_map_type(struct bpf_map_type_list *tl)
44 list_add(&tl->list_node, &bpf_map_types);
47 /* called from workqueue */
48 static void bpf_map_free_deferred(struct work_struct *work)
50 struct bpf_map *map = container_of(work, struct bpf_map, work);
52 /* implementation dependent freeing */
53 map->ops->map_free(map);
56 /* decrement map refcnt and schedule it for freeing via workqueue
57 * (unrelying map implementation ops->map_free() might sleep)
59 void bpf_map_put(struct bpf_map *map)
61 if (atomic_dec_and_test(&map->refcnt)) {
62 INIT_WORK(&map->work, bpf_map_free_deferred);
63 schedule_work(&map->work);
67 static int bpf_map_release(struct inode *inode, struct file *filp)
69 struct bpf_map *map = filp->private_data;
71 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
72 /* prog_array stores refcnt-ed bpf_prog pointers
73 * release them all when user space closes prog_array_fd
75 bpf_fd_array_map_clear(map);
81 static const struct file_operations bpf_map_fops = {
82 .release = bpf_map_release,
85 /* helper macro to check that unused fields 'union bpf_attr' are zero */
86 #define CHECK_ATTR(CMD) \
87 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
88 sizeof(attr->CMD##_LAST_FIELD), 0, \
90 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
91 sizeof(attr->CMD##_LAST_FIELD)) != NULL
93 #define BPF_MAP_CREATE_LAST_FIELD max_entries
94 /* called via syscall */
95 static int map_create(union bpf_attr *attr)
100 err = CHECK_ATTR(BPF_MAP_CREATE);
104 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
105 map = find_and_alloc_map(attr);
109 atomic_set(&map->refcnt, 1);
111 err = anon_inode_getfd("bpf-map", &bpf_map_fops, map, O_RDWR | O_CLOEXEC);
114 /* failed to allocate fd */
120 map->ops->map_free(map);
124 /* if error is returned, fd is released.
125 * On success caller should complete fd access with matching fdput()
127 struct bpf_map *bpf_map_get(struct fd f)
132 return ERR_PTR(-EBADF);
134 if (f.file->f_op != &bpf_map_fops) {
136 return ERR_PTR(-EINVAL);
139 map = f.file->private_data;
144 /* helper to convert user pointers passed inside __aligned_u64 fields */
145 static void __user *u64_to_ptr(__u64 val)
147 return (void __user *) (unsigned long) val;
150 /* last field in 'union bpf_attr' used by this command */
151 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
153 static int map_lookup_elem(union bpf_attr *attr)
155 void __user *ukey = u64_to_ptr(attr->key);
156 void __user *uvalue = u64_to_ptr(attr->value);
157 int ufd = attr->map_fd;
159 void *key, *value, *ptr;
163 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
167 map = bpf_map_get(f);
172 key = kmalloc(map->key_size, GFP_USER);
177 if (copy_from_user(key, ukey, map->key_size) != 0)
181 value = kmalloc(map->value_size, GFP_USER);
186 ptr = map->ops->map_lookup_elem(map, key);
188 memcpy(value, ptr, map->value_size);
196 if (copy_to_user(uvalue, value, map->value_size) != 0)
210 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
212 static int map_update_elem(union bpf_attr *attr)
214 void __user *ukey = u64_to_ptr(attr->key);
215 void __user *uvalue = u64_to_ptr(attr->value);
216 int ufd = attr->map_fd;
222 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
226 map = bpf_map_get(f);
231 key = kmalloc(map->key_size, GFP_USER);
236 if (copy_from_user(key, ukey, map->key_size) != 0)
240 value = kmalloc(map->value_size, GFP_USER);
245 if (copy_from_user(value, uvalue, map->value_size) != 0)
248 /* eBPF program that use maps are running under rcu_read_lock(),
249 * therefore all map accessors rely on this fact, so do the same here
252 err = map->ops->map_update_elem(map, key, value, attr->flags);
264 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
266 static int map_delete_elem(union bpf_attr *attr)
268 void __user *ukey = u64_to_ptr(attr->key);
269 int ufd = attr->map_fd;
275 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
279 map = bpf_map_get(f);
284 key = kmalloc(map->key_size, GFP_USER);
289 if (copy_from_user(key, ukey, map->key_size) != 0)
293 err = map->ops->map_delete_elem(map, key);
303 /* last field in 'union bpf_attr' used by this command */
304 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
306 static int map_get_next_key(union bpf_attr *attr)
308 void __user *ukey = u64_to_ptr(attr->key);
309 void __user *unext_key = u64_to_ptr(attr->next_key);
310 int ufd = attr->map_fd;
312 void *key, *next_key;
316 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
320 map = bpf_map_get(f);
325 key = kmalloc(map->key_size, GFP_USER);
330 if (copy_from_user(key, ukey, map->key_size) != 0)
334 next_key = kmalloc(map->key_size, GFP_USER);
339 err = map->ops->map_get_next_key(map, key, next_key);
345 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
359 static LIST_HEAD(bpf_prog_types);
361 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
363 struct bpf_prog_type_list *tl;
365 list_for_each_entry(tl, &bpf_prog_types, list_node) {
366 if (tl->type == type) {
367 prog->aux->ops = tl->ops;
376 void bpf_register_prog_type(struct bpf_prog_type_list *tl)
378 list_add(&tl->list_node, &bpf_prog_types);
381 /* fixup insn->imm field of bpf_call instructions:
382 * if (insn->imm == BPF_FUNC_map_lookup_elem)
383 * insn->imm = bpf_map_lookup_elem - __bpf_call_base;
384 * else if (insn->imm == BPF_FUNC_map_update_elem)
385 * insn->imm = bpf_map_update_elem - __bpf_call_base;
388 * this function is called after eBPF program passed verification
390 static void fixup_bpf_calls(struct bpf_prog *prog)
392 const struct bpf_func_proto *fn;
395 for (i = 0; i < prog->len; i++) {
396 struct bpf_insn *insn = &prog->insnsi[i];
398 if (insn->code == (BPF_JMP | BPF_CALL)) {
399 /* we reach here when program has bpf_call instructions
400 * and it passed bpf_check(), means that
401 * ops->get_func_proto must have been supplied, check it
403 BUG_ON(!prog->aux->ops->get_func_proto);
405 if (insn->imm == BPF_FUNC_get_route_realm)
406 prog->dst_needed = 1;
407 if (insn->imm == BPF_FUNC_tail_call) {
408 /* mark bpf_tail_call as different opcode
409 * to avoid conditional branch in
410 * interpeter for every normal call
411 * and to prevent accidental JITing by
412 * JIT compiler that doesn't support
420 fn = prog->aux->ops->get_func_proto(insn->imm);
421 /* all functions that have prototype and verifier allowed
422 * programs to call them, must be real in-kernel functions
425 insn->imm = fn->func - __bpf_call_base;
430 /* drop refcnt on maps used by eBPF program and free auxilary data */
431 static void free_used_maps(struct bpf_prog_aux *aux)
435 for (i = 0; i < aux->used_map_cnt; i++)
436 bpf_map_put(aux->used_maps[i]);
438 kfree(aux->used_maps);
441 static void __prog_put_rcu(struct rcu_head *rcu)
443 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
446 bpf_prog_free(aux->prog);
449 /* version of bpf_prog_put() that is called after a grace period */
450 void bpf_prog_put_rcu(struct bpf_prog *prog)
452 if (atomic_dec_and_test(&prog->aux->refcnt)) {
453 prog->aux->prog = prog;
454 call_rcu(&prog->aux->rcu, __prog_put_rcu);
458 void bpf_prog_put(struct bpf_prog *prog)
460 if (atomic_dec_and_test(&prog->aux->refcnt)) {
461 free_used_maps(prog->aux);
465 EXPORT_SYMBOL_GPL(bpf_prog_put);
467 static int bpf_prog_release(struct inode *inode, struct file *filp)
469 struct bpf_prog *prog = filp->private_data;
471 bpf_prog_put_rcu(prog);
475 static const struct file_operations bpf_prog_fops = {
476 .release = bpf_prog_release,
479 static struct bpf_prog *get_prog(struct fd f)
481 struct bpf_prog *prog;
484 return ERR_PTR(-EBADF);
486 if (f.file->f_op != &bpf_prog_fops) {
488 return ERR_PTR(-EINVAL);
491 prog = f.file->private_data;
496 /* called by sockets/tracing/seccomp before attaching program to an event
497 * pairs with bpf_prog_put()
499 struct bpf_prog *bpf_prog_get(u32 ufd)
501 struct fd f = fdget(ufd);
502 struct bpf_prog *prog;
509 atomic_inc(&prog->aux->refcnt);
513 EXPORT_SYMBOL_GPL(bpf_prog_get);
515 /* last field in 'union bpf_attr' used by this command */
516 #define BPF_PROG_LOAD_LAST_FIELD kern_version
518 static int bpf_prog_load(union bpf_attr *attr)
520 enum bpf_prog_type type = attr->prog_type;
521 struct bpf_prog *prog;
526 if (CHECK_ATTR(BPF_PROG_LOAD))
529 /* copy eBPF program license from user space */
530 if (strncpy_from_user(license, u64_to_ptr(attr->license),
531 sizeof(license) - 1) < 0)
533 license[sizeof(license) - 1] = 0;
535 /* eBPF programs must be GPL compatible to use GPL-ed functions */
536 is_gpl = license_is_gpl_compatible(license);
538 if (attr->insn_cnt >= BPF_MAXINSNS)
541 if (type == BPF_PROG_TYPE_KPROBE &&
542 attr->kern_version != LINUX_VERSION_CODE)
545 /* plain bpf_prog allocation */
546 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
550 prog->len = attr->insn_cnt;
553 if (copy_from_user(prog->insns, u64_to_ptr(attr->insns),
554 prog->len * sizeof(struct bpf_insn)) != 0)
557 prog->orig_prog = NULL;
560 atomic_set(&prog->aux->refcnt, 1);
561 prog->gpl_compatible = is_gpl ? 1 : 0;
563 /* find program type: socket_filter vs tracing_filter */
564 err = find_prog_type(type, prog);
568 /* run eBPF verifier */
569 err = bpf_check(&prog, attr);
573 /* fixup BPF_CALL->imm field */
574 fixup_bpf_calls(prog);
576 /* eBPF program is ready to be JITed */
577 err = bpf_prog_select_runtime(prog);
581 err = anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, O_RDWR | O_CLOEXEC);
583 /* failed to allocate fd */
589 free_used_maps(prog->aux);
595 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
597 union bpf_attr attr = {};
600 /* the syscall is limited to root temporarily. This restriction will be
601 * lifted when security audit is clean. Note that eBPF+tracing must have
602 * this restriction, since it may pass kernel data to user space
604 if (!capable(CAP_SYS_ADMIN))
607 if (!access_ok(VERIFY_READ, uattr, 1))
610 if (size > PAGE_SIZE) /* silly large */
613 /* If we're handed a bigger struct than we know of,
614 * ensure all the unknown bits are 0 - i.e. new
615 * user-space does not rely on any kernel feature
616 * extensions we dont know about yet.
618 if (size > sizeof(attr)) {
619 unsigned char __user *addr;
620 unsigned char __user *end;
623 addr = (void __user *)uattr + sizeof(attr);
624 end = (void __user *)uattr + size;
626 for (; addr < end; addr++) {
627 err = get_user(val, addr);
636 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
637 if (copy_from_user(&attr, uattr, size) != 0)
642 err = map_create(&attr);
644 case BPF_MAP_LOOKUP_ELEM:
645 err = map_lookup_elem(&attr);
647 case BPF_MAP_UPDATE_ELEM:
648 err = map_update_elem(&attr);
650 case BPF_MAP_DELETE_ELEM:
651 err = map_delete_elem(&attr);
653 case BPF_MAP_GET_NEXT_KEY:
654 err = map_get_next_key(&attr);
657 err = bpf_prog_load(&attr);