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;
75 static const struct file_operations bpf_map_fops = {
76 .release = bpf_map_release,
79 /* helper macro to check that unused fields 'union bpf_attr' are zero */
80 #define CHECK_ATTR(CMD) \
81 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
82 sizeof(attr->CMD##_LAST_FIELD), 0, \
84 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
85 sizeof(attr->CMD##_LAST_FIELD)) != NULL
87 #define BPF_MAP_CREATE_LAST_FIELD max_entries
88 /* called via syscall */
89 static int map_create(union bpf_attr *attr)
94 err = CHECK_ATTR(BPF_MAP_CREATE);
98 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
99 map = find_and_alloc_map(attr);
103 atomic_set(&map->refcnt, 1);
105 err = anon_inode_getfd("bpf-map", &bpf_map_fops, map, O_RDWR | O_CLOEXEC);
108 /* failed to allocate fd */
114 map->ops->map_free(map);
118 /* if error is returned, fd is released.
119 * On success caller should complete fd access with matching fdput()
121 struct bpf_map *bpf_map_get(struct fd f)
126 return ERR_PTR(-EBADF);
128 if (f.file->f_op != &bpf_map_fops) {
130 return ERR_PTR(-EINVAL);
133 map = f.file->private_data;
138 /* helper to convert user pointers passed inside __aligned_u64 fields */
139 static void __user *u64_to_ptr(__u64 val)
141 return (void __user *) (unsigned long) val;
144 /* last field in 'union bpf_attr' used by this command */
145 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
147 static int map_lookup_elem(union bpf_attr *attr)
149 void __user *ukey = u64_to_ptr(attr->key);
150 void __user *uvalue = u64_to_ptr(attr->value);
151 int ufd = attr->map_fd;
152 struct fd f = fdget(ufd);
154 void *key, *value, *ptr;
157 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
160 map = bpf_map_get(f);
165 key = kmalloc(map->key_size, GFP_USER);
170 if (copy_from_user(key, ukey, map->key_size) != 0)
174 value = kmalloc(map->value_size, GFP_USER);
179 ptr = map->ops->map_lookup_elem(map, key);
181 memcpy(value, ptr, map->value_size);
189 if (copy_to_user(uvalue, value, map->value_size) != 0)
203 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
205 static int map_update_elem(union bpf_attr *attr)
207 void __user *ukey = u64_to_ptr(attr->key);
208 void __user *uvalue = u64_to_ptr(attr->value);
209 int ufd = attr->map_fd;
210 struct fd f = fdget(ufd);
215 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
218 map = bpf_map_get(f);
223 key = kmalloc(map->key_size, GFP_USER);
228 if (copy_from_user(key, ukey, map->key_size) != 0)
232 value = kmalloc(map->value_size, GFP_USER);
237 if (copy_from_user(value, uvalue, map->value_size) != 0)
240 /* eBPF program that use maps are running under rcu_read_lock(),
241 * therefore all map accessors rely on this fact, so do the same here
244 err = map->ops->map_update_elem(map, key, value, attr->flags);
256 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
258 static int map_delete_elem(union bpf_attr *attr)
260 void __user *ukey = u64_to_ptr(attr->key);
261 int ufd = attr->map_fd;
262 struct fd f = fdget(ufd);
267 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
270 map = bpf_map_get(f);
275 key = kmalloc(map->key_size, GFP_USER);
280 if (copy_from_user(key, ukey, map->key_size) != 0)
284 err = map->ops->map_delete_elem(map, key);
294 /* last field in 'union bpf_attr' used by this command */
295 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
297 static int map_get_next_key(union bpf_attr *attr)
299 void __user *ukey = u64_to_ptr(attr->key);
300 void __user *unext_key = u64_to_ptr(attr->next_key);
301 int ufd = attr->map_fd;
302 struct fd f = fdget(ufd);
304 void *key, *next_key;
307 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
310 map = bpf_map_get(f);
315 key = kmalloc(map->key_size, GFP_USER);
320 if (copy_from_user(key, ukey, map->key_size) != 0)
324 next_key = kmalloc(map->key_size, GFP_USER);
329 err = map->ops->map_get_next_key(map, key, next_key);
335 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
349 static LIST_HEAD(bpf_prog_types);
351 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
353 struct bpf_prog_type_list *tl;
355 list_for_each_entry(tl, &bpf_prog_types, list_node) {
356 if (tl->type == type) {
357 prog->aux->ops = tl->ops;
358 prog->aux->prog_type = type;
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);
423 static int bpf_prog_release(struct inode *inode, struct file *filp)
425 struct bpf_prog *prog = filp->private_data;
431 static const struct file_operations bpf_prog_fops = {
432 .release = bpf_prog_release,
435 static struct bpf_prog *get_prog(struct fd f)
437 struct bpf_prog *prog;
440 return ERR_PTR(-EBADF);
442 if (f.file->f_op != &bpf_prog_fops) {
444 return ERR_PTR(-EINVAL);
447 prog = f.file->private_data;
452 /* called by sockets/tracing/seccomp before attaching program to an event
453 * pairs with bpf_prog_put()
455 struct bpf_prog *bpf_prog_get(u32 ufd)
457 struct fd f = fdget(ufd);
458 struct bpf_prog *prog;
465 atomic_inc(&prog->aux->refcnt);
470 /* last field in 'union bpf_attr' used by this command */
471 #define BPF_PROG_LOAD_LAST_FIELD kern_version
473 static int bpf_prog_load(union bpf_attr *attr)
475 enum bpf_prog_type type = attr->prog_type;
476 struct bpf_prog *prog;
481 if (CHECK_ATTR(BPF_PROG_LOAD))
484 /* copy eBPF program license from user space */
485 if (strncpy_from_user(license, u64_to_ptr(attr->license),
486 sizeof(license) - 1) < 0)
488 license[sizeof(license) - 1] = 0;
490 /* eBPF programs must be GPL compatible to use GPL-ed functions */
491 is_gpl = license_is_gpl_compatible(license);
493 if (attr->insn_cnt >= BPF_MAXINSNS)
496 if (type == BPF_PROG_TYPE_KPROBE &&
497 attr->kern_version != LINUX_VERSION_CODE)
500 /* plain bpf_prog allocation */
501 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
505 prog->len = attr->insn_cnt;
508 if (copy_from_user(prog->insns, u64_to_ptr(attr->insns),
509 prog->len * sizeof(struct bpf_insn)) != 0)
512 prog->orig_prog = NULL;
515 atomic_set(&prog->aux->refcnt, 1);
516 prog->aux->is_gpl_compatible = is_gpl;
518 /* find program type: socket_filter vs tracing_filter */
519 err = find_prog_type(type, prog);
523 /* run eBPF verifier */
524 err = bpf_check(prog, attr);
529 /* fixup BPF_CALL->imm field */
530 fixup_bpf_calls(prog);
532 /* eBPF program is ready to be JITed */
533 bpf_prog_select_runtime(prog);
535 err = anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, O_RDWR | O_CLOEXEC);
538 /* failed to allocate fd */
544 free_used_maps(prog->aux);
550 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
552 union bpf_attr attr = {};
555 /* the syscall is limited to root temporarily. This restriction will be
556 * lifted when security audit is clean. Note that eBPF+tracing must have
557 * this restriction, since it may pass kernel data to user space
559 if (!capable(CAP_SYS_ADMIN))
562 if (!access_ok(VERIFY_READ, uattr, 1))
565 if (size > PAGE_SIZE) /* silly large */
568 /* If we're handed a bigger struct than we know of,
569 * ensure all the unknown bits are 0 - i.e. new
570 * user-space does not rely on any kernel feature
571 * extensions we dont know about yet.
573 if (size > sizeof(attr)) {
574 unsigned char __user *addr;
575 unsigned char __user *end;
578 addr = (void __user *)uattr + sizeof(attr);
579 end = (void __user *)uattr + size;
581 for (; addr < end; addr++) {
582 err = get_user(val, addr);
591 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
592 if (copy_from_user(&attr, uattr, size) != 0)
597 err = map_create(&attr);
599 case BPF_MAP_LOOKUP_ELEM:
600 err = map_lookup_elem(&attr);
602 case BPF_MAP_UPDATE_ELEM:
603 err = map_update_elem(&attr);
605 case BPF_MAP_DELETE_ELEM:
606 err = map_delete_elem(&attr);
608 case BPF_MAP_GET_NEXT_KEY:
609 err = map_get_next_key(&attr);
612 err = bpf_prog_load(&attr);