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 /* decrement map refcnt and schedule it for freeing via workqueue
86 * (unrelying map implementation ops->map_free() might sleep)
88 void bpf_map_put(struct bpf_map *map)
90 if (atomic_dec_and_test(&map->refcnt)) {
91 INIT_WORK(&map->work, bpf_map_free_deferred);
92 schedule_work(&map->work);
96 static int bpf_map_release(struct inode *inode, struct file *filp)
98 struct bpf_map *map = filp->private_data;
100 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
101 /* prog_array stores refcnt-ed bpf_prog pointers
102 * release them all when user space closes prog_array_fd
104 bpf_fd_array_map_clear(map);
110 static const struct file_operations bpf_map_fops = {
111 .release = bpf_map_release,
114 int bpf_map_new_fd(struct bpf_map *map)
116 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
120 /* helper macro to check that unused fields 'union bpf_attr' are zero */
121 #define CHECK_ATTR(CMD) \
122 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
123 sizeof(attr->CMD##_LAST_FIELD), 0, \
125 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
126 sizeof(attr->CMD##_LAST_FIELD)) != NULL
128 #define BPF_MAP_CREATE_LAST_FIELD max_entries
129 /* called via syscall */
130 static int map_create(union bpf_attr *attr)
135 err = CHECK_ATTR(BPF_MAP_CREATE);
139 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
140 map = find_and_alloc_map(attr);
144 atomic_set(&map->refcnt, 1);
146 err = bpf_map_charge_memlock(map);
150 err = bpf_map_new_fd(map);
152 /* failed to allocate fd */
158 map->ops->map_free(map);
162 /* if error is returned, fd is released.
163 * On success caller should complete fd access with matching fdput()
165 struct bpf_map *__bpf_map_get(struct fd f)
168 return ERR_PTR(-EBADF);
169 if (f.file->f_op != &bpf_map_fops) {
171 return ERR_PTR(-EINVAL);
174 return f.file->private_data;
177 struct bpf_map *bpf_map_get(u32 ufd)
179 struct fd f = fdget(ufd);
182 map = __bpf_map_get(f);
186 atomic_inc(&map->refcnt);
192 /* helper to convert user pointers passed inside __aligned_u64 fields */
193 static void __user *u64_to_ptr(__u64 val)
195 return (void __user *) (unsigned long) val;
198 /* last field in 'union bpf_attr' used by this command */
199 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
201 static int map_lookup_elem(union bpf_attr *attr)
203 void __user *ukey = u64_to_ptr(attr->key);
204 void __user *uvalue = u64_to_ptr(attr->value);
205 int ufd = attr->map_fd;
207 void *key, *value, *ptr;
211 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
215 map = __bpf_map_get(f);
220 key = kmalloc(map->key_size, GFP_USER);
225 if (copy_from_user(key, ukey, map->key_size) != 0)
229 value = kmalloc(map->value_size, GFP_USER);
234 ptr = map->ops->map_lookup_elem(map, key);
236 memcpy(value, ptr, map->value_size);
244 if (copy_to_user(uvalue, value, map->value_size) != 0)
258 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
260 static int map_update_elem(union bpf_attr *attr)
262 void __user *ukey = u64_to_ptr(attr->key);
263 void __user *uvalue = u64_to_ptr(attr->value);
264 int ufd = attr->map_fd;
270 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
274 map = __bpf_map_get(f);
279 key = kmalloc(map->key_size, GFP_USER);
284 if (copy_from_user(key, ukey, map->key_size) != 0)
288 value = kmalloc(map->value_size, GFP_USER);
293 if (copy_from_user(value, uvalue, map->value_size) != 0)
296 /* eBPF program that use maps are running under rcu_read_lock(),
297 * therefore all map accessors rely on this fact, so do the same here
300 err = map->ops->map_update_elem(map, key, value, attr->flags);
312 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
314 static int map_delete_elem(union bpf_attr *attr)
316 void __user *ukey = u64_to_ptr(attr->key);
317 int ufd = attr->map_fd;
323 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
327 map = __bpf_map_get(f);
332 key = kmalloc(map->key_size, GFP_USER);
337 if (copy_from_user(key, ukey, map->key_size) != 0)
341 err = map->ops->map_delete_elem(map, key);
351 /* last field in 'union bpf_attr' used by this command */
352 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
354 static int map_get_next_key(union bpf_attr *attr)
356 void __user *ukey = u64_to_ptr(attr->key);
357 void __user *unext_key = u64_to_ptr(attr->next_key);
358 int ufd = attr->map_fd;
360 void *key, *next_key;
364 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
368 map = __bpf_map_get(f);
373 key = kmalloc(map->key_size, GFP_USER);
378 if (copy_from_user(key, ukey, map->key_size) != 0)
382 next_key = kmalloc(map->key_size, GFP_USER);
387 err = map->ops->map_get_next_key(map, key, next_key);
393 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
407 static LIST_HEAD(bpf_prog_types);
409 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
411 struct bpf_prog_type_list *tl;
413 list_for_each_entry(tl, &bpf_prog_types, list_node) {
414 if (tl->type == type) {
415 prog->aux->ops = tl->ops;
424 void bpf_register_prog_type(struct bpf_prog_type_list *tl)
426 list_add(&tl->list_node, &bpf_prog_types);
429 /* fixup insn->imm field of bpf_call instructions:
430 * if (insn->imm == BPF_FUNC_map_lookup_elem)
431 * insn->imm = bpf_map_lookup_elem - __bpf_call_base;
432 * else if (insn->imm == BPF_FUNC_map_update_elem)
433 * insn->imm = bpf_map_update_elem - __bpf_call_base;
436 * this function is called after eBPF program passed verification
438 static void fixup_bpf_calls(struct bpf_prog *prog)
440 const struct bpf_func_proto *fn;
443 for (i = 0; i < prog->len; i++) {
444 struct bpf_insn *insn = &prog->insnsi[i];
446 if (insn->code == (BPF_JMP | BPF_CALL)) {
447 /* we reach here when program has bpf_call instructions
448 * and it passed bpf_check(), means that
449 * ops->get_func_proto must have been supplied, check it
451 BUG_ON(!prog->aux->ops->get_func_proto);
453 if (insn->imm == BPF_FUNC_get_route_realm)
454 prog->dst_needed = 1;
455 if (insn->imm == BPF_FUNC_get_prandom_u32)
456 bpf_user_rnd_init_once();
457 if (insn->imm == BPF_FUNC_tail_call) {
458 /* mark bpf_tail_call as different opcode
459 * to avoid conditional branch in
460 * interpeter for every normal call
461 * and to prevent accidental JITing by
462 * JIT compiler that doesn't support
470 fn = prog->aux->ops->get_func_proto(insn->imm);
471 /* all functions that have prototype and verifier allowed
472 * programs to call them, must be real in-kernel functions
475 insn->imm = fn->func - __bpf_call_base;
480 /* drop refcnt on maps used by eBPF program and free auxilary data */
481 static void free_used_maps(struct bpf_prog_aux *aux)
485 for (i = 0; i < aux->used_map_cnt; i++)
486 bpf_map_put(aux->used_maps[i]);
488 kfree(aux->used_maps);
491 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
493 struct user_struct *user = get_current_user();
494 unsigned long memlock_limit;
496 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
498 atomic_long_add(prog->pages, &user->locked_vm);
499 if (atomic_long_read(&user->locked_vm) > memlock_limit) {
500 atomic_long_sub(prog->pages, &user->locked_vm);
504 prog->aux->user = user;
508 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
510 struct user_struct *user = prog->aux->user;
512 atomic_long_sub(prog->pages, &user->locked_vm);
516 static void __prog_put_common(struct rcu_head *rcu)
518 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
521 bpf_prog_uncharge_memlock(aux->prog);
522 bpf_prog_free(aux->prog);
525 /* version of bpf_prog_put() that is called after a grace period */
526 void bpf_prog_put_rcu(struct bpf_prog *prog)
528 if (atomic_dec_and_test(&prog->aux->refcnt))
529 call_rcu(&prog->aux->rcu, __prog_put_common);
532 void bpf_prog_put(struct bpf_prog *prog)
534 if (atomic_dec_and_test(&prog->aux->refcnt))
535 __prog_put_common(&prog->aux->rcu);
537 EXPORT_SYMBOL_GPL(bpf_prog_put);
539 static int bpf_prog_release(struct inode *inode, struct file *filp)
541 struct bpf_prog *prog = filp->private_data;
543 bpf_prog_put_rcu(prog);
547 static const struct file_operations bpf_prog_fops = {
548 .release = bpf_prog_release,
551 int bpf_prog_new_fd(struct bpf_prog *prog)
553 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
557 static struct bpf_prog *__bpf_prog_get(struct fd f)
560 return ERR_PTR(-EBADF);
561 if (f.file->f_op != &bpf_prog_fops) {
563 return ERR_PTR(-EINVAL);
566 return f.file->private_data;
569 /* called by sockets/tracing/seccomp before attaching program to an event
570 * pairs with bpf_prog_put()
572 struct bpf_prog *bpf_prog_get(u32 ufd)
574 struct fd f = fdget(ufd);
575 struct bpf_prog *prog;
577 prog = __bpf_prog_get(f);
581 atomic_inc(&prog->aux->refcnt);
586 EXPORT_SYMBOL_GPL(bpf_prog_get);
588 /* last field in 'union bpf_attr' used by this command */
589 #define BPF_PROG_LOAD_LAST_FIELD kern_version
591 static int bpf_prog_load(union bpf_attr *attr)
593 enum bpf_prog_type type = attr->prog_type;
594 struct bpf_prog *prog;
599 if (CHECK_ATTR(BPF_PROG_LOAD))
602 /* copy eBPF program license from user space */
603 if (strncpy_from_user(license, u64_to_ptr(attr->license),
604 sizeof(license) - 1) < 0)
606 license[sizeof(license) - 1] = 0;
608 /* eBPF programs must be GPL compatible to use GPL-ed functions */
609 is_gpl = license_is_gpl_compatible(license);
611 if (attr->insn_cnt >= BPF_MAXINSNS)
614 if (type == BPF_PROG_TYPE_KPROBE &&
615 attr->kern_version != LINUX_VERSION_CODE)
618 if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN))
621 /* plain bpf_prog allocation */
622 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
626 err = bpf_prog_charge_memlock(prog);
628 goto free_prog_nouncharge;
630 prog->len = attr->insn_cnt;
633 if (copy_from_user(prog->insns, u64_to_ptr(attr->insns),
634 prog->len * sizeof(struct bpf_insn)) != 0)
637 prog->orig_prog = NULL;
640 atomic_set(&prog->aux->refcnt, 1);
641 prog->gpl_compatible = is_gpl ? 1 : 0;
643 /* find program type: socket_filter vs tracing_filter */
644 err = find_prog_type(type, prog);
648 /* run eBPF verifier */
649 err = bpf_check(&prog, attr);
653 /* fixup BPF_CALL->imm field */
654 fixup_bpf_calls(prog);
656 /* eBPF program is ready to be JITed */
657 err = bpf_prog_select_runtime(prog);
661 err = bpf_prog_new_fd(prog);
663 /* failed to allocate fd */
669 free_used_maps(prog->aux);
671 bpf_prog_uncharge_memlock(prog);
672 free_prog_nouncharge:
677 #define BPF_OBJ_LAST_FIELD bpf_fd
679 static int bpf_obj_pin(const union bpf_attr *attr)
681 if (CHECK_ATTR(BPF_OBJ))
684 return bpf_obj_pin_user(attr->bpf_fd, u64_to_ptr(attr->pathname));
687 static int bpf_obj_get(const union bpf_attr *attr)
689 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0)
692 return bpf_obj_get_user(u64_to_ptr(attr->pathname));
695 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
697 union bpf_attr attr = {};
700 if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
703 if (!access_ok(VERIFY_READ, uattr, 1))
706 if (size > PAGE_SIZE) /* silly large */
709 /* If we're handed a bigger struct than we know of,
710 * ensure all the unknown bits are 0 - i.e. new
711 * user-space does not rely on any kernel feature
712 * extensions we dont know about yet.
714 if (size > sizeof(attr)) {
715 unsigned char __user *addr;
716 unsigned char __user *end;
719 addr = (void __user *)uattr + sizeof(attr);
720 end = (void __user *)uattr + size;
722 for (; addr < end; addr++) {
723 err = get_user(val, addr);
732 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
733 if (copy_from_user(&attr, uattr, size) != 0)
738 err = map_create(&attr);
740 case BPF_MAP_LOOKUP_ELEM:
741 err = map_lookup_elem(&attr);
743 case BPF_MAP_UPDATE_ELEM:
744 err = map_update_elem(&attr);
746 case BPF_MAP_DELETE_ELEM:
747 err = map_delete_elem(&attr);
749 case BPF_MAP_GET_NEXT_KEY:
750 err = map_get_next_key(&attr);
753 err = bpf_prog_load(&attr);
756 err = bpf_obj_pin(&attr);
759 err = bpf_obj_get(&attr);