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 bpf_map_put_with_uref(filp->private_data);
131 #ifdef CONFIG_PROC_FS
132 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
134 const struct bpf_map *map = filp->private_data;
150 static const struct file_operations bpf_map_fops = {
151 #ifdef CONFIG_PROC_FS
152 .show_fdinfo = bpf_map_show_fdinfo,
154 .release = bpf_map_release,
157 int bpf_map_new_fd(struct bpf_map *map)
159 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
163 /* helper macro to check that unused fields 'union bpf_attr' are zero */
164 #define CHECK_ATTR(CMD) \
165 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
166 sizeof(attr->CMD##_LAST_FIELD), 0, \
168 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
169 sizeof(attr->CMD##_LAST_FIELD)) != NULL
171 #define BPF_MAP_CREATE_LAST_FIELD map_flags
172 /* called via syscall */
173 static int map_create(union bpf_attr *attr)
178 err = CHECK_ATTR(BPF_MAP_CREATE);
182 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
183 map = find_and_alloc_map(attr);
187 atomic_set(&map->refcnt, 1);
188 atomic_set(&map->usercnt, 1);
190 err = bpf_map_charge_memlock(map);
194 err = bpf_map_new_fd(map);
196 /* failed to allocate fd */
202 map->ops->map_free(map);
206 /* if error is returned, fd is released.
207 * On success caller should complete fd access with matching fdput()
209 struct bpf_map *__bpf_map_get(struct fd f)
212 return ERR_PTR(-EBADF);
213 if (f.file->f_op != &bpf_map_fops) {
215 return ERR_PTR(-EINVAL);
218 return f.file->private_data;
221 void bpf_map_inc(struct bpf_map *map, bool uref)
223 atomic_inc(&map->refcnt);
225 atomic_inc(&map->usercnt);
228 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
230 struct fd f = fdget(ufd);
233 map = __bpf_map_get(f);
237 bpf_map_inc(map, true);
243 /* helper to convert user pointers passed inside __aligned_u64 fields */
244 static void __user *u64_to_ptr(__u64 val)
246 return (void __user *) (unsigned long) val;
249 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
254 /* last field in 'union bpf_attr' used by this command */
255 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
257 static int map_lookup_elem(union bpf_attr *attr)
259 void __user *ukey = u64_to_ptr(attr->key);
260 void __user *uvalue = u64_to_ptr(attr->value);
261 int ufd = attr->map_fd;
263 void *key, *value, *ptr;
268 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
272 map = __bpf_map_get(f);
277 key = kmalloc(map->key_size, GFP_USER);
282 if (copy_from_user(key, ukey, map->key_size) != 0)
285 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
286 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
287 value_size = round_up(map->value_size, 8) * num_possible_cpus();
289 value_size = map->value_size;
292 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
296 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) {
297 err = bpf_percpu_hash_copy(map, key, value);
298 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
299 err = bpf_percpu_array_copy(map, key, value);
300 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
301 err = bpf_stackmap_copy(map, key, value);
304 ptr = map->ops->map_lookup_elem(map, key);
306 memcpy(value, ptr, value_size);
308 err = ptr ? 0 : -ENOENT;
315 if (copy_to_user(uvalue, value, value_size) != 0)
329 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
331 static int map_update_elem(union bpf_attr *attr)
333 void __user *ukey = u64_to_ptr(attr->key);
334 void __user *uvalue = u64_to_ptr(attr->value);
335 int ufd = attr->map_fd;
342 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
346 map = __bpf_map_get(f);
351 key = kmalloc(map->key_size, GFP_USER);
356 if (copy_from_user(key, ukey, map->key_size) != 0)
359 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
360 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
361 value_size = round_up(map->value_size, 8) * num_possible_cpus();
363 value_size = map->value_size;
366 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
371 if (copy_from_user(value, uvalue, value_size) != 0)
374 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
375 * inside bpf map update or delete otherwise deadlocks are possible
378 __this_cpu_inc(bpf_prog_active);
379 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) {
380 err = bpf_percpu_hash_update(map, key, value, attr->flags);
381 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
382 err = bpf_percpu_array_update(map, key, value, attr->flags);
385 err = map->ops->map_update_elem(map, key, value, attr->flags);
388 __this_cpu_dec(bpf_prog_active);
400 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
402 static int map_delete_elem(union bpf_attr *attr)
404 void __user *ukey = u64_to_ptr(attr->key);
405 int ufd = attr->map_fd;
411 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
415 map = __bpf_map_get(f);
420 key = kmalloc(map->key_size, GFP_USER);
425 if (copy_from_user(key, ukey, map->key_size) != 0)
429 __this_cpu_inc(bpf_prog_active);
431 err = map->ops->map_delete_elem(map, key);
433 __this_cpu_dec(bpf_prog_active);
443 /* last field in 'union bpf_attr' used by this command */
444 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
446 static int map_get_next_key(union bpf_attr *attr)
448 void __user *ukey = u64_to_ptr(attr->key);
449 void __user *unext_key = u64_to_ptr(attr->next_key);
450 int ufd = attr->map_fd;
452 void *key, *next_key;
456 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
460 map = __bpf_map_get(f);
465 key = kmalloc(map->key_size, GFP_USER);
470 if (copy_from_user(key, ukey, map->key_size) != 0)
474 next_key = kmalloc(map->key_size, GFP_USER);
479 err = map->ops->map_get_next_key(map, key, next_key);
485 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
499 static LIST_HEAD(bpf_prog_types);
501 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
503 struct bpf_prog_type_list *tl;
505 list_for_each_entry(tl, &bpf_prog_types, list_node) {
506 if (tl->type == type) {
507 prog->aux->ops = tl->ops;
516 void bpf_register_prog_type(struct bpf_prog_type_list *tl)
518 list_add(&tl->list_node, &bpf_prog_types);
521 /* fixup insn->imm field of bpf_call instructions:
522 * if (insn->imm == BPF_FUNC_map_lookup_elem)
523 * insn->imm = bpf_map_lookup_elem - __bpf_call_base;
524 * else if (insn->imm == BPF_FUNC_map_update_elem)
525 * insn->imm = bpf_map_update_elem - __bpf_call_base;
528 * this function is called after eBPF program passed verification
530 static void fixup_bpf_calls(struct bpf_prog *prog)
532 const struct bpf_func_proto *fn;
535 for (i = 0; i < prog->len; i++) {
536 struct bpf_insn *insn = &prog->insnsi[i];
538 if (insn->code == (BPF_JMP | BPF_CALL)) {
539 /* we reach here when program has bpf_call instructions
540 * and it passed bpf_check(), means that
541 * ops->get_func_proto must have been supplied, check it
543 BUG_ON(!prog->aux->ops->get_func_proto);
545 if (insn->imm == BPF_FUNC_get_route_realm)
546 prog->dst_needed = 1;
547 if (insn->imm == BPF_FUNC_get_prandom_u32)
548 bpf_user_rnd_init_once();
549 if (insn->imm == BPF_FUNC_tail_call) {
550 /* mark bpf_tail_call as different opcode
551 * to avoid conditional branch in
552 * interpeter for every normal call
553 * and to prevent accidental JITing by
554 * JIT compiler that doesn't support
562 fn = prog->aux->ops->get_func_proto(insn->imm);
563 /* all functions that have prototype and verifier allowed
564 * programs to call them, must be real in-kernel functions
567 insn->imm = fn->func - __bpf_call_base;
572 /* drop refcnt on maps used by eBPF program and free auxilary data */
573 static void free_used_maps(struct bpf_prog_aux *aux)
577 for (i = 0; i < aux->used_map_cnt; i++)
578 bpf_map_put(aux->used_maps[i]);
580 kfree(aux->used_maps);
583 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
585 struct user_struct *user = get_current_user();
586 unsigned long memlock_limit;
588 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
590 atomic_long_add(prog->pages, &user->locked_vm);
591 if (atomic_long_read(&user->locked_vm) > memlock_limit) {
592 atomic_long_sub(prog->pages, &user->locked_vm);
596 prog->aux->user = user;
600 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
602 struct user_struct *user = prog->aux->user;
604 atomic_long_sub(prog->pages, &user->locked_vm);
608 static void __prog_put_common(struct rcu_head *rcu)
610 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
613 bpf_prog_uncharge_memlock(aux->prog);
614 bpf_prog_free(aux->prog);
617 /* version of bpf_prog_put() that is called after a grace period */
618 void bpf_prog_put_rcu(struct bpf_prog *prog)
620 if (atomic_dec_and_test(&prog->aux->refcnt))
621 call_rcu(&prog->aux->rcu, __prog_put_common);
624 void bpf_prog_put(struct bpf_prog *prog)
626 if (atomic_dec_and_test(&prog->aux->refcnt))
627 __prog_put_common(&prog->aux->rcu);
629 EXPORT_SYMBOL_GPL(bpf_prog_put);
631 static int bpf_prog_release(struct inode *inode, struct file *filp)
633 struct bpf_prog *prog = filp->private_data;
635 bpf_prog_put_rcu(prog);
639 static const struct file_operations bpf_prog_fops = {
640 .release = bpf_prog_release,
643 int bpf_prog_new_fd(struct bpf_prog *prog)
645 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
649 static struct bpf_prog *__bpf_prog_get(struct fd f)
652 return ERR_PTR(-EBADF);
653 if (f.file->f_op != &bpf_prog_fops) {
655 return ERR_PTR(-EINVAL);
658 return f.file->private_data;
661 /* called by sockets/tracing/seccomp before attaching program to an event
662 * pairs with bpf_prog_put()
664 struct bpf_prog *bpf_prog_get(u32 ufd)
666 struct fd f = fdget(ufd);
667 struct bpf_prog *prog;
669 prog = __bpf_prog_get(f);
673 atomic_inc(&prog->aux->refcnt);
678 EXPORT_SYMBOL_GPL(bpf_prog_get);
680 /* last field in 'union bpf_attr' used by this command */
681 #define BPF_PROG_LOAD_LAST_FIELD kern_version
683 static int bpf_prog_load(union bpf_attr *attr)
685 enum bpf_prog_type type = attr->prog_type;
686 struct bpf_prog *prog;
691 if (CHECK_ATTR(BPF_PROG_LOAD))
694 /* copy eBPF program license from user space */
695 if (strncpy_from_user(license, u64_to_ptr(attr->license),
696 sizeof(license) - 1) < 0)
698 license[sizeof(license) - 1] = 0;
700 /* eBPF programs must be GPL compatible to use GPL-ed functions */
701 is_gpl = license_is_gpl_compatible(license);
703 if (attr->insn_cnt >= BPF_MAXINSNS)
706 if (type == BPF_PROG_TYPE_KPROBE &&
707 attr->kern_version != LINUX_VERSION_CODE)
710 if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN))
713 /* plain bpf_prog allocation */
714 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
718 err = bpf_prog_charge_memlock(prog);
720 goto free_prog_nouncharge;
722 prog->len = attr->insn_cnt;
725 if (copy_from_user(prog->insns, u64_to_ptr(attr->insns),
726 prog->len * sizeof(struct bpf_insn)) != 0)
729 prog->orig_prog = NULL;
732 atomic_set(&prog->aux->refcnt, 1);
733 prog->gpl_compatible = is_gpl ? 1 : 0;
735 /* find program type: socket_filter vs tracing_filter */
736 err = find_prog_type(type, prog);
740 /* run eBPF verifier */
741 err = bpf_check(&prog, attr);
745 /* fixup BPF_CALL->imm field */
746 fixup_bpf_calls(prog);
748 /* eBPF program is ready to be JITed */
749 err = bpf_prog_select_runtime(prog);
753 err = bpf_prog_new_fd(prog);
755 /* failed to allocate fd */
761 free_used_maps(prog->aux);
763 bpf_prog_uncharge_memlock(prog);
764 free_prog_nouncharge:
769 #define BPF_OBJ_LAST_FIELD bpf_fd
771 static int bpf_obj_pin(const union bpf_attr *attr)
773 if (CHECK_ATTR(BPF_OBJ))
776 return bpf_obj_pin_user(attr->bpf_fd, u64_to_ptr(attr->pathname));
779 static int bpf_obj_get(const union bpf_attr *attr)
781 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0)
784 return bpf_obj_get_user(u64_to_ptr(attr->pathname));
787 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
789 union bpf_attr attr = {};
792 if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
795 if (!access_ok(VERIFY_READ, uattr, 1))
798 if (size > PAGE_SIZE) /* silly large */
801 /* If we're handed a bigger struct than we know of,
802 * ensure all the unknown bits are 0 - i.e. new
803 * user-space does not rely on any kernel feature
804 * extensions we dont know about yet.
806 if (size > sizeof(attr)) {
807 unsigned char __user *addr;
808 unsigned char __user *end;
811 addr = (void __user *)uattr + sizeof(attr);
812 end = (void __user *)uattr + size;
814 for (; addr < end; addr++) {
815 err = get_user(val, addr);
824 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
825 if (copy_from_user(&attr, uattr, size) != 0)
830 err = map_create(&attr);
832 case BPF_MAP_LOOKUP_ELEM:
833 err = map_lookup_elem(&attr);
835 case BPF_MAP_UPDATE_ELEM:
836 err = map_update_elem(&attr);
838 case BPF_MAP_DELETE_ELEM:
839 err = map_delete_elem(&attr);
841 case BPF_MAP_GET_NEXT_KEY:
842 err = map_get_next_key(&attr);
845 err = bpf_prog_load(&attr);
848 err = bpf_obj_pin(&attr);
851 err = bpf_obj_get(&attr);