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 struct bpf_map *map = filp->private_data;
129 if (map->ops->map_release)
130 map->ops->map_release(map, filp);
132 bpf_map_put_with_uref(map);
136 #ifdef CONFIG_PROC_FS
137 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
139 const struct bpf_map *map = filp->private_data;
155 static const struct file_operations bpf_map_fops = {
156 #ifdef CONFIG_PROC_FS
157 .show_fdinfo = bpf_map_show_fdinfo,
159 .release = bpf_map_release,
162 int bpf_map_new_fd(struct bpf_map *map)
164 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
168 /* helper macro to check that unused fields 'union bpf_attr' are zero */
169 #define CHECK_ATTR(CMD) \
170 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
171 sizeof(attr->CMD##_LAST_FIELD), 0, \
173 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
174 sizeof(attr->CMD##_LAST_FIELD)) != NULL
176 #define BPF_MAP_CREATE_LAST_FIELD map_flags
177 /* called via syscall */
178 static int map_create(union bpf_attr *attr)
183 err = CHECK_ATTR(BPF_MAP_CREATE);
187 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
188 map = find_and_alloc_map(attr);
192 atomic_set(&map->refcnt, 1);
193 atomic_set(&map->usercnt, 1);
195 err = bpf_map_charge_memlock(map);
197 goto free_map_nouncharge;
199 err = bpf_map_new_fd(map);
201 /* failed to allocate fd */
207 bpf_map_uncharge_memlock(map);
209 map->ops->map_free(map);
213 /* if error is returned, fd is released.
214 * On success caller should complete fd access with matching fdput()
216 struct bpf_map *__bpf_map_get(struct fd f)
219 return ERR_PTR(-EBADF);
220 if (f.file->f_op != &bpf_map_fops) {
222 return ERR_PTR(-EINVAL);
225 return f.file->private_data;
228 /* prog's and map's refcnt limit */
229 #define BPF_MAX_REFCNT 32768
231 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
233 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
234 atomic_dec(&map->refcnt);
235 return ERR_PTR(-EBUSY);
238 atomic_inc(&map->usercnt);
242 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
244 struct fd f = fdget(ufd);
247 map = __bpf_map_get(f);
251 map = bpf_map_inc(map, true);
257 /* helper to convert user pointers passed inside __aligned_u64 fields */
258 static void __user *u64_to_ptr(__u64 val)
260 return (void __user *) (unsigned long) val;
263 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
268 /* last field in 'union bpf_attr' used by this command */
269 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
271 static int map_lookup_elem(union bpf_attr *attr)
273 void __user *ukey = u64_to_ptr(attr->key);
274 void __user *uvalue = u64_to_ptr(attr->value);
275 int ufd = attr->map_fd;
277 void *key, *value, *ptr;
282 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
286 map = __bpf_map_get(f);
291 key = kmalloc(map->key_size, GFP_USER);
296 if (copy_from_user(key, ukey, map->key_size) != 0)
299 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
300 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
301 value_size = round_up(map->value_size, 8) * num_possible_cpus();
303 value_size = map->value_size;
306 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
310 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) {
311 err = bpf_percpu_hash_copy(map, key, value);
312 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
313 err = bpf_percpu_array_copy(map, key, value);
314 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
315 err = bpf_stackmap_copy(map, key, value);
318 ptr = map->ops->map_lookup_elem(map, key);
320 memcpy(value, ptr, value_size);
322 err = ptr ? 0 : -ENOENT;
329 if (copy_to_user(uvalue, value, value_size) != 0)
343 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
345 static int map_update_elem(union bpf_attr *attr)
347 void __user *ukey = u64_to_ptr(attr->key);
348 void __user *uvalue = u64_to_ptr(attr->value);
349 int ufd = attr->map_fd;
356 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
360 map = __bpf_map_get(f);
365 key = kmalloc(map->key_size, GFP_USER);
370 if (copy_from_user(key, ukey, map->key_size) != 0)
373 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
374 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
375 value_size = round_up(map->value_size, 8) * num_possible_cpus();
377 value_size = map->value_size;
380 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
385 if (copy_from_user(value, uvalue, value_size) != 0)
388 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
389 * inside bpf map update or delete otherwise deadlocks are possible
392 __this_cpu_inc(bpf_prog_active);
393 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) {
394 err = bpf_percpu_hash_update(map, key, value, attr->flags);
395 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
396 err = bpf_percpu_array_update(map, key, value, attr->flags);
397 } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
398 map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
399 map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY) {
401 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
406 err = map->ops->map_update_elem(map, key, value, attr->flags);
409 __this_cpu_dec(bpf_prog_active);
421 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
423 static int map_delete_elem(union bpf_attr *attr)
425 void __user *ukey = u64_to_ptr(attr->key);
426 int ufd = attr->map_fd;
432 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
436 map = __bpf_map_get(f);
441 key = kmalloc(map->key_size, GFP_USER);
446 if (copy_from_user(key, ukey, map->key_size) != 0)
450 __this_cpu_inc(bpf_prog_active);
452 err = map->ops->map_delete_elem(map, key);
454 __this_cpu_dec(bpf_prog_active);
464 /* last field in 'union bpf_attr' used by this command */
465 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
467 static int map_get_next_key(union bpf_attr *attr)
469 void __user *ukey = u64_to_ptr(attr->key);
470 void __user *unext_key = u64_to_ptr(attr->next_key);
471 int ufd = attr->map_fd;
473 void *key, *next_key;
477 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
481 map = __bpf_map_get(f);
486 key = kmalloc(map->key_size, GFP_USER);
491 if (copy_from_user(key, ukey, map->key_size) != 0)
495 next_key = kmalloc(map->key_size, GFP_USER);
500 err = map->ops->map_get_next_key(map, key, next_key);
506 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
520 static LIST_HEAD(bpf_prog_types);
522 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
524 struct bpf_prog_type_list *tl;
526 list_for_each_entry(tl, &bpf_prog_types, list_node) {
527 if (tl->type == type) {
528 prog->aux->ops = tl->ops;
537 void bpf_register_prog_type(struct bpf_prog_type_list *tl)
539 list_add(&tl->list_node, &bpf_prog_types);
542 /* fixup insn->imm field of bpf_call instructions:
543 * if (insn->imm == BPF_FUNC_map_lookup_elem)
544 * insn->imm = bpf_map_lookup_elem - __bpf_call_base;
545 * else if (insn->imm == BPF_FUNC_map_update_elem)
546 * insn->imm = bpf_map_update_elem - __bpf_call_base;
549 * this function is called after eBPF program passed verification
551 static void fixup_bpf_calls(struct bpf_prog *prog)
553 const struct bpf_func_proto *fn;
556 for (i = 0; i < prog->len; i++) {
557 struct bpf_insn *insn = &prog->insnsi[i];
559 if (insn->code == (BPF_JMP | BPF_CALL)) {
560 /* we reach here when program has bpf_call instructions
561 * and it passed bpf_check(), means that
562 * ops->get_func_proto must have been supplied, check it
564 BUG_ON(!prog->aux->ops->get_func_proto);
566 if (insn->imm == BPF_FUNC_get_route_realm)
567 prog->dst_needed = 1;
568 if (insn->imm == BPF_FUNC_get_prandom_u32)
569 bpf_user_rnd_init_once();
570 if (insn->imm == BPF_FUNC_tail_call) {
571 /* mark bpf_tail_call as different opcode
572 * to avoid conditional branch in
573 * interpeter for every normal call
574 * and to prevent accidental JITing by
575 * JIT compiler that doesn't support
583 fn = prog->aux->ops->get_func_proto(insn->imm);
584 /* all functions that have prototype and verifier allowed
585 * programs to call them, must be real in-kernel functions
588 insn->imm = fn->func - __bpf_call_base;
593 /* drop refcnt on maps used by eBPF program and free auxilary data */
594 static void free_used_maps(struct bpf_prog_aux *aux)
598 for (i = 0; i < aux->used_map_cnt; i++)
599 bpf_map_put(aux->used_maps[i]);
601 kfree(aux->used_maps);
604 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
606 struct user_struct *user = get_current_user();
607 unsigned long memlock_limit;
609 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
611 atomic_long_add(prog->pages, &user->locked_vm);
612 if (atomic_long_read(&user->locked_vm) > memlock_limit) {
613 atomic_long_sub(prog->pages, &user->locked_vm);
617 prog->aux->user = user;
621 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
623 struct user_struct *user = prog->aux->user;
625 atomic_long_sub(prog->pages, &user->locked_vm);
629 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
631 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
634 bpf_prog_uncharge_memlock(aux->prog);
635 bpf_prog_free(aux->prog);
638 void bpf_prog_put(struct bpf_prog *prog)
640 if (atomic_dec_and_test(&prog->aux->refcnt))
641 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
643 EXPORT_SYMBOL_GPL(bpf_prog_put);
645 static int bpf_prog_release(struct inode *inode, struct file *filp)
647 struct bpf_prog *prog = filp->private_data;
653 static const struct file_operations bpf_prog_fops = {
654 .release = bpf_prog_release,
657 int bpf_prog_new_fd(struct bpf_prog *prog)
659 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
663 static struct bpf_prog *____bpf_prog_get(struct fd f)
666 return ERR_PTR(-EBADF);
667 if (f.file->f_op != &bpf_prog_fops) {
669 return ERR_PTR(-EINVAL);
672 return f.file->private_data;
675 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
677 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
678 atomic_sub(i, &prog->aux->refcnt);
679 return ERR_PTR(-EBUSY);
683 EXPORT_SYMBOL_GPL(bpf_prog_add);
685 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
687 return bpf_prog_add(prog, 1);
690 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type)
692 struct fd f = fdget(ufd);
693 struct bpf_prog *prog;
695 prog = ____bpf_prog_get(f);
698 if (type && prog->type != *type) {
699 prog = ERR_PTR(-EINVAL);
703 prog = bpf_prog_inc(prog);
709 struct bpf_prog *bpf_prog_get(u32 ufd)
711 return __bpf_prog_get(ufd, NULL);
714 struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type)
716 return __bpf_prog_get(ufd, &type);
718 EXPORT_SYMBOL_GPL(bpf_prog_get_type);
720 /* last field in 'union bpf_attr' used by this command */
721 #define BPF_PROG_LOAD_LAST_FIELD kern_version
723 static int bpf_prog_load(union bpf_attr *attr)
725 enum bpf_prog_type type = attr->prog_type;
726 struct bpf_prog *prog;
731 if (CHECK_ATTR(BPF_PROG_LOAD))
734 /* copy eBPF program license from user space */
735 if (strncpy_from_user(license, u64_to_ptr(attr->license),
736 sizeof(license) - 1) < 0)
738 license[sizeof(license) - 1] = 0;
740 /* eBPF programs must be GPL compatible to use GPL-ed functions */
741 is_gpl = license_is_gpl_compatible(license);
743 if (attr->insn_cnt >= BPF_MAXINSNS)
746 if (type == BPF_PROG_TYPE_KPROBE &&
747 attr->kern_version != LINUX_VERSION_CODE)
750 if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN))
753 /* plain bpf_prog allocation */
754 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
758 err = bpf_prog_charge_memlock(prog);
760 goto free_prog_nouncharge;
762 prog->len = attr->insn_cnt;
765 if (copy_from_user(prog->insns, u64_to_ptr(attr->insns),
766 prog->len * sizeof(struct bpf_insn)) != 0)
769 prog->orig_prog = NULL;
772 atomic_set(&prog->aux->refcnt, 1);
773 prog->gpl_compatible = is_gpl ? 1 : 0;
775 /* find program type: socket_filter vs tracing_filter */
776 err = find_prog_type(type, prog);
780 /* run eBPF verifier */
781 err = bpf_check(&prog, attr);
785 /* fixup BPF_CALL->imm field */
786 fixup_bpf_calls(prog);
788 /* eBPF program is ready to be JITed */
789 prog = bpf_prog_select_runtime(prog, &err);
793 err = bpf_prog_new_fd(prog);
795 /* failed to allocate fd */
801 free_used_maps(prog->aux);
803 bpf_prog_uncharge_memlock(prog);
804 free_prog_nouncharge:
809 #define BPF_OBJ_LAST_FIELD bpf_fd
811 static int bpf_obj_pin(const union bpf_attr *attr)
813 if (CHECK_ATTR(BPF_OBJ))
816 return bpf_obj_pin_user(attr->bpf_fd, u64_to_ptr(attr->pathname));
819 static int bpf_obj_get(const union bpf_attr *attr)
821 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0)
824 return bpf_obj_get_user(u64_to_ptr(attr->pathname));
827 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
829 union bpf_attr attr = {};
832 if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
835 if (!access_ok(VERIFY_READ, uattr, 1))
838 if (size > PAGE_SIZE) /* silly large */
841 /* If we're handed a bigger struct than we know of,
842 * ensure all the unknown bits are 0 - i.e. new
843 * user-space does not rely on any kernel feature
844 * extensions we dont know about yet.
846 if (size > sizeof(attr)) {
847 unsigned char __user *addr;
848 unsigned char __user *end;
851 addr = (void __user *)uattr + sizeof(attr);
852 end = (void __user *)uattr + size;
854 for (; addr < end; addr++) {
855 err = get_user(val, addr);
864 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
865 if (copy_from_user(&attr, uattr, size) != 0)
870 err = map_create(&attr);
872 case BPF_MAP_LOOKUP_ELEM:
873 err = map_lookup_elem(&attr);
875 case BPF_MAP_UPDATE_ELEM:
876 err = map_update_elem(&attr);
878 case BPF_MAP_DELETE_ELEM:
879 err = map_delete_elem(&attr);
881 case BPF_MAP_GET_NEXT_KEY:
882 err = map_get_next_key(&attr);
885 err = bpf_prog_load(&attr);
888 err = bpf_obj_pin(&attr);
891 err = bpf_obj_get(&attr);