1
2
3
4
5
6
7
8
9
10
11
12#include <linux/bpf.h>
13#include <linux/bpf_trace.h>
14#include <linux/syscalls.h>
15#include <linux/slab.h>
16#include <linux/sched/signal.h>
17#include <linux/vmalloc.h>
18#include <linux/mmzone.h>
19#include <linux/anon_inodes.h>
20#include <linux/file.h>
21#include <linux/license.h>
22#include <linux/filter.h>
23#include <linux/version.h>
24#include <linux/kernel.h>
25#include <linux/idr.h>
26
27#define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
28 (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
29 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
30 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
31#define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
32#define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
33
34DEFINE_PER_CPU(int, bpf_prog_active);
35static DEFINE_IDR(prog_idr);
36static DEFINE_SPINLOCK(prog_idr_lock);
37static DEFINE_IDR(map_idr);
38static DEFINE_SPINLOCK(map_idr_lock);
39
40int sysctl_unprivileged_bpf_disabled __read_mostly;
41
42static const struct bpf_map_ops * const bpf_map_types[] = {
43#define BPF_PROG_TYPE(_id, _ops)
44#define BPF_MAP_TYPE(_id, _ops) \
45 [_id] = &_ops,
46#include <linux/bpf_types.h>
47#undef BPF_PROG_TYPE
48#undef BPF_MAP_TYPE
49};
50
51
52
53
54
55
56
57
58
59
60static int check_uarg_tail_zero(void __user *uaddr,
61 size_t expected_size,
62 size_t actual_size)
63{
64 unsigned char __user *addr;
65 unsigned char __user *end;
66 unsigned char val;
67 int err;
68
69 if (unlikely(actual_size > PAGE_SIZE))
70 return -E2BIG;
71
72 if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
73 return -EFAULT;
74
75 if (actual_size <= expected_size)
76 return 0;
77
78 addr = uaddr + expected_size;
79 end = uaddr + actual_size;
80
81 for (; addr < end; addr++) {
82 err = get_user(val, addr);
83 if (err)
84 return err;
85 if (val)
86 return -E2BIG;
87 }
88
89 return 0;
90}
91
92static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
93{
94 struct bpf_map *map;
95
96 if (attr->map_type >= ARRAY_SIZE(bpf_map_types) ||
97 !bpf_map_types[attr->map_type])
98 return ERR_PTR(-EINVAL);
99
100 map = bpf_map_types[attr->map_type]->map_alloc(attr);
101 if (IS_ERR(map))
102 return map;
103 map->ops = bpf_map_types[attr->map_type];
104 map->map_type = attr->map_type;
105 return map;
106}
107
108void *bpf_map_area_alloc(size_t size, int numa_node)
109{
110
111
112
113
114 const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
115 void *area;
116
117 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
118 area = kmalloc_node(size, GFP_USER | flags, numa_node);
119 if (area != NULL)
120 return area;
121 }
122
123 return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
124 __builtin_return_address(0));
125}
126
127void bpf_map_area_free(void *area)
128{
129 kvfree(area);
130}
131
132int bpf_map_precharge_memlock(u32 pages)
133{
134 struct user_struct *user = get_current_user();
135 unsigned long memlock_limit, cur;
136
137 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
138 cur = atomic_long_read(&user->locked_vm);
139 free_uid(user);
140 if (cur + pages > memlock_limit)
141 return -EPERM;
142 return 0;
143}
144
145static int bpf_map_charge_memlock(struct bpf_map *map)
146{
147 struct user_struct *user = get_current_user();
148 unsigned long memlock_limit;
149
150 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
151
152 atomic_long_add(map->pages, &user->locked_vm);
153
154 if (atomic_long_read(&user->locked_vm) > memlock_limit) {
155 atomic_long_sub(map->pages, &user->locked_vm);
156 free_uid(user);
157 return -EPERM;
158 }
159 map->user = user;
160 return 0;
161}
162
163static void bpf_map_uncharge_memlock(struct bpf_map *map)
164{
165 struct user_struct *user = map->user;
166
167 atomic_long_sub(map->pages, &user->locked_vm);
168 free_uid(user);
169}
170
171static int bpf_map_alloc_id(struct bpf_map *map)
172{
173 int id;
174
175 spin_lock_bh(&map_idr_lock);
176 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
177 if (id > 0)
178 map->id = id;
179 spin_unlock_bh(&map_idr_lock);
180
181 if (WARN_ON_ONCE(!id))
182 return -ENOSPC;
183
184 return id > 0 ? 0 : id;
185}
186
187static void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
188{
189 unsigned long flags;
190
191 if (do_idr_lock)
192 spin_lock_irqsave(&map_idr_lock, flags);
193 else
194 __acquire(&map_idr_lock);
195
196 idr_remove(&map_idr, map->id);
197
198 if (do_idr_lock)
199 spin_unlock_irqrestore(&map_idr_lock, flags);
200 else
201 __release(&map_idr_lock);
202}
203
204
205static void bpf_map_free_deferred(struct work_struct *work)
206{
207 struct bpf_map *map = container_of(work, struct bpf_map, work);
208
209 bpf_map_uncharge_memlock(map);
210
211 map->ops->map_free(map);
212}
213
214static void bpf_map_put_uref(struct bpf_map *map)
215{
216 if (atomic_dec_and_test(&map->usercnt)) {
217 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
218 bpf_fd_array_map_clear(map);
219 }
220}
221
222
223
224
225static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
226{
227 if (atomic_dec_and_test(&map->refcnt)) {
228
229 bpf_map_free_id(map, do_idr_lock);
230 INIT_WORK(&map->work, bpf_map_free_deferred);
231 schedule_work(&map->work);
232 }
233}
234
235void bpf_map_put(struct bpf_map *map)
236{
237 __bpf_map_put(map, true);
238}
239
240void bpf_map_put_with_uref(struct bpf_map *map)
241{
242 bpf_map_put_uref(map);
243 bpf_map_put(map);
244}
245
246static int bpf_map_release(struct inode *inode, struct file *filp)
247{
248 struct bpf_map *map = filp->private_data;
249
250 if (map->ops->map_release)
251 map->ops->map_release(map, filp);
252
253 bpf_map_put_with_uref(map);
254 return 0;
255}
256
257#ifdef CONFIG_PROC_FS
258static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
259{
260 const struct bpf_map *map = filp->private_data;
261 const struct bpf_array *array;
262 u32 owner_prog_type = 0;
263 u32 owner_jited = 0;
264
265 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
266 array = container_of(map, struct bpf_array, map);
267 owner_prog_type = array->owner_prog_type;
268 owner_jited = array->owner_jited;
269 }
270
271 seq_printf(m,
272 "map_type:\t%u\n"
273 "key_size:\t%u\n"
274 "value_size:\t%u\n"
275 "max_entries:\t%u\n"
276 "map_flags:\t%#x\n"
277 "memlock:\t%llu\n",
278 map->map_type,
279 map->key_size,
280 map->value_size,
281 map->max_entries,
282 map->map_flags,
283 map->pages * 1ULL << PAGE_SHIFT);
284
285 if (owner_prog_type) {
286 seq_printf(m, "owner_prog_type:\t%u\n",
287 owner_prog_type);
288 seq_printf(m, "owner_jited:\t%u\n",
289 owner_jited);
290 }
291}
292#endif
293
294static const struct file_operations bpf_map_fops = {
295#ifdef CONFIG_PROC_FS
296 .show_fdinfo = bpf_map_show_fdinfo,
297#endif
298 .release = bpf_map_release,
299};
300
301int bpf_map_new_fd(struct bpf_map *map)
302{
303 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
304 O_RDWR | O_CLOEXEC);
305}
306
307
308#define CHECK_ATTR(CMD) \
309 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
310 sizeof(attr->CMD##_LAST_FIELD), 0, \
311 sizeof(*attr) - \
312 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
313 sizeof(attr->CMD##_LAST_FIELD)) != NULL
314
315#define BPF_MAP_CREATE_LAST_FIELD numa_node
316
317static int map_create(union bpf_attr *attr)
318{
319 int numa_node = bpf_map_attr_numa_node(attr);
320 struct bpf_map *map;
321 int err;
322
323 err = CHECK_ATTR(BPF_MAP_CREATE);
324 if (err)
325 return -EINVAL;
326
327 if (numa_node != NUMA_NO_NODE &&
328 ((unsigned int)numa_node >= nr_node_ids ||
329 !node_online(numa_node)))
330 return -EINVAL;
331
332
333 map = find_and_alloc_map(attr);
334 if (IS_ERR(map))
335 return PTR_ERR(map);
336
337 atomic_set(&map->refcnt, 1);
338 atomic_set(&map->usercnt, 1);
339
340 err = bpf_map_charge_memlock(map);
341 if (err)
342 goto free_map_nouncharge;
343
344 err = bpf_map_alloc_id(map);
345 if (err)
346 goto free_map;
347
348 err = bpf_map_new_fd(map);
349 if (err < 0) {
350
351
352
353
354
355
356 bpf_map_put(map);
357 return err;
358 }
359
360 trace_bpf_map_create(map, err);
361 return err;
362
363free_map:
364 bpf_map_uncharge_memlock(map);
365free_map_nouncharge:
366 map->ops->map_free(map);
367 return err;
368}
369
370
371
372
373struct bpf_map *__bpf_map_get(struct fd f)
374{
375 if (!f.file)
376 return ERR_PTR(-EBADF);
377 if (f.file->f_op != &bpf_map_fops) {
378 fdput(f);
379 return ERR_PTR(-EINVAL);
380 }
381
382 return f.file->private_data;
383}
384
385
386#define BPF_MAX_REFCNT 32768
387
388struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
389{
390 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
391 atomic_dec(&map->refcnt);
392 return ERR_PTR(-EBUSY);
393 }
394 if (uref)
395 atomic_inc(&map->usercnt);
396 return map;
397}
398
399struct bpf_map *bpf_map_get_with_uref(u32 ufd)
400{
401 struct fd f = fdget(ufd);
402 struct bpf_map *map;
403
404 map = __bpf_map_get(f);
405 if (IS_ERR(map))
406 return map;
407
408 map = bpf_map_inc(map, true);
409 fdput(f);
410
411 return map;
412}
413
414
415static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
416 bool uref)
417{
418 int refold;
419
420 refold = __atomic_add_unless(&map->refcnt, 1, 0);
421
422 if (refold >= BPF_MAX_REFCNT) {
423 __bpf_map_put(map, false);
424 return ERR_PTR(-EBUSY);
425 }
426
427 if (!refold)
428 return ERR_PTR(-ENOENT);
429
430 if (uref)
431 atomic_inc(&map->usercnt);
432
433 return map;
434}
435
436int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
437{
438 return -ENOTSUPP;
439}
440
441
442#define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
443
444static int map_lookup_elem(union bpf_attr *attr)
445{
446 void __user *ukey = u64_to_user_ptr(attr->key);
447 void __user *uvalue = u64_to_user_ptr(attr->value);
448 int ufd = attr->map_fd;
449 struct bpf_map *map;
450 void *key, *value, *ptr;
451 u32 value_size;
452 struct fd f;
453 int err;
454
455 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
456 return -EINVAL;
457
458 f = fdget(ufd);
459 map = __bpf_map_get(f);
460 if (IS_ERR(map))
461 return PTR_ERR(map);
462
463 key = memdup_user(ukey, map->key_size);
464 if (IS_ERR(key)) {
465 err = PTR_ERR(key);
466 goto err_put;
467 }
468
469 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
470 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
471 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
472 value_size = round_up(map->value_size, 8) * num_possible_cpus();
473 else if (IS_FD_MAP(map))
474 value_size = sizeof(u32);
475 else
476 value_size = map->value_size;
477
478 err = -ENOMEM;
479 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
480 if (!value)
481 goto free_key;
482
483 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
484 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
485 err = bpf_percpu_hash_copy(map, key, value);
486 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
487 err = bpf_percpu_array_copy(map, key, value);
488 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
489 err = bpf_stackmap_copy(map, key, value);
490 } else if (IS_FD_ARRAY(map)) {
491 err = bpf_fd_array_map_lookup_elem(map, key, value);
492 } else if (IS_FD_HASH(map)) {
493 err = bpf_fd_htab_map_lookup_elem(map, key, value);
494 } else {
495 rcu_read_lock();
496 ptr = map->ops->map_lookup_elem(map, key);
497 if (ptr)
498 memcpy(value, ptr, value_size);
499 rcu_read_unlock();
500 err = ptr ? 0 : -ENOENT;
501 }
502
503 if (err)
504 goto free_value;
505
506 err = -EFAULT;
507 if (copy_to_user(uvalue, value, value_size) != 0)
508 goto free_value;
509
510 trace_bpf_map_lookup_elem(map, ufd, key, value);
511 err = 0;
512
513free_value:
514 kfree(value);
515free_key:
516 kfree(key);
517err_put:
518 fdput(f);
519 return err;
520}
521
522#define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
523
524static int map_update_elem(union bpf_attr *attr)
525{
526 void __user *ukey = u64_to_user_ptr(attr->key);
527 void __user *uvalue = u64_to_user_ptr(attr->value);
528 int ufd = attr->map_fd;
529 struct bpf_map *map;
530 void *key, *value;
531 u32 value_size;
532 struct fd f;
533 int err;
534
535 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
536 return -EINVAL;
537
538 f = fdget(ufd);
539 map = __bpf_map_get(f);
540 if (IS_ERR(map))
541 return PTR_ERR(map);
542
543 key = memdup_user(ukey, map->key_size);
544 if (IS_ERR(key)) {
545 err = PTR_ERR(key);
546 goto err_put;
547 }
548
549 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
550 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
551 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
552 value_size = round_up(map->value_size, 8) * num_possible_cpus();
553 else
554 value_size = map->value_size;
555
556 err = -ENOMEM;
557 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
558 if (!value)
559 goto free_key;
560
561 err = -EFAULT;
562 if (copy_from_user(value, uvalue, value_size) != 0)
563 goto free_value;
564
565
566
567
568 preempt_disable();
569 __this_cpu_inc(bpf_prog_active);
570 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
571 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
572 err = bpf_percpu_hash_update(map, key, value, attr->flags);
573 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
574 err = bpf_percpu_array_update(map, key, value, attr->flags);
575 } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
576 map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
577 map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY ||
578 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) {
579 rcu_read_lock();
580 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
581 attr->flags);
582 rcu_read_unlock();
583 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
584 rcu_read_lock();
585 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
586 attr->flags);
587 rcu_read_unlock();
588 } else {
589 rcu_read_lock();
590 err = map->ops->map_update_elem(map, key, value, attr->flags);
591 rcu_read_unlock();
592 }
593 __this_cpu_dec(bpf_prog_active);
594 preempt_enable();
595
596 if (!err)
597 trace_bpf_map_update_elem(map, ufd, key, value);
598free_value:
599 kfree(value);
600free_key:
601 kfree(key);
602err_put:
603 fdput(f);
604 return err;
605}
606
607#define BPF_MAP_DELETE_ELEM_LAST_FIELD key
608
609static int map_delete_elem(union bpf_attr *attr)
610{
611 void __user *ukey = u64_to_user_ptr(attr->key);
612 int ufd = attr->map_fd;
613 struct bpf_map *map;
614 struct fd f;
615 void *key;
616 int err;
617
618 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
619 return -EINVAL;
620
621 f = fdget(ufd);
622 map = __bpf_map_get(f);
623 if (IS_ERR(map))
624 return PTR_ERR(map);
625
626 key = memdup_user(ukey, map->key_size);
627 if (IS_ERR(key)) {
628 err = PTR_ERR(key);
629 goto err_put;
630 }
631
632 preempt_disable();
633 __this_cpu_inc(bpf_prog_active);
634 rcu_read_lock();
635 err = map->ops->map_delete_elem(map, key);
636 rcu_read_unlock();
637 __this_cpu_dec(bpf_prog_active);
638 preempt_enable();
639
640 if (!err)
641 trace_bpf_map_delete_elem(map, ufd, key);
642 kfree(key);
643err_put:
644 fdput(f);
645 return err;
646}
647
648
649#define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
650
651static int map_get_next_key(union bpf_attr *attr)
652{
653 void __user *ukey = u64_to_user_ptr(attr->key);
654 void __user *unext_key = u64_to_user_ptr(attr->next_key);
655 int ufd = attr->map_fd;
656 struct bpf_map *map;
657 void *key, *next_key;
658 struct fd f;
659 int err;
660
661 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
662 return -EINVAL;
663
664 f = fdget(ufd);
665 map = __bpf_map_get(f);
666 if (IS_ERR(map))
667 return PTR_ERR(map);
668
669 if (ukey) {
670 key = memdup_user(ukey, map->key_size);
671 if (IS_ERR(key)) {
672 err = PTR_ERR(key);
673 goto err_put;
674 }
675 } else {
676 key = NULL;
677 }
678
679 err = -ENOMEM;
680 next_key = kmalloc(map->key_size, GFP_USER);
681 if (!next_key)
682 goto free_key;
683
684 rcu_read_lock();
685 err = map->ops->map_get_next_key(map, key, next_key);
686 rcu_read_unlock();
687 if (err)
688 goto free_next_key;
689
690 err = -EFAULT;
691 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
692 goto free_next_key;
693
694 trace_bpf_map_next_key(map, ufd, key, next_key);
695 err = 0;
696
697free_next_key:
698 kfree(next_key);
699free_key:
700 kfree(key);
701err_put:
702 fdput(f);
703 return err;
704}
705
706static const struct bpf_verifier_ops * const bpf_prog_types[] = {
707#define BPF_PROG_TYPE(_id, _ops) \
708 [_id] = &_ops,
709#define BPF_MAP_TYPE(_id, _ops)
710#include <linux/bpf_types.h>
711#undef BPF_PROG_TYPE
712#undef BPF_MAP_TYPE
713};
714
715static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
716{
717 if (type >= ARRAY_SIZE(bpf_prog_types) || !bpf_prog_types[type])
718 return -EINVAL;
719
720 prog->aux->ops = bpf_prog_types[type];
721 prog->type = type;
722 return 0;
723}
724
725
726static void free_used_maps(struct bpf_prog_aux *aux)
727{
728 int i;
729
730 for (i = 0; i < aux->used_map_cnt; i++)
731 bpf_map_put(aux->used_maps[i]);
732
733 kfree(aux->used_maps);
734}
735
736int __bpf_prog_charge(struct user_struct *user, u32 pages)
737{
738 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
739 unsigned long user_bufs;
740
741 if (user) {
742 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
743 if (user_bufs > memlock_limit) {
744 atomic_long_sub(pages, &user->locked_vm);
745 return -EPERM;
746 }
747 }
748
749 return 0;
750}
751
752void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
753{
754 if (user)
755 atomic_long_sub(pages, &user->locked_vm);
756}
757
758static int bpf_prog_charge_memlock(struct bpf_prog *prog)
759{
760 struct user_struct *user = get_current_user();
761 int ret;
762
763 ret = __bpf_prog_charge(user, prog->pages);
764 if (ret) {
765 free_uid(user);
766 return ret;
767 }
768
769 prog->aux->user = user;
770 return 0;
771}
772
773static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
774{
775 struct user_struct *user = prog->aux->user;
776
777 __bpf_prog_uncharge(user, prog->pages);
778 free_uid(user);
779}
780
781static int bpf_prog_alloc_id(struct bpf_prog *prog)
782{
783 int id;
784
785 spin_lock_bh(&prog_idr_lock);
786 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
787 if (id > 0)
788 prog->aux->id = id;
789 spin_unlock_bh(&prog_idr_lock);
790
791
792 if (WARN_ON_ONCE(!id))
793 return -ENOSPC;
794
795 return id > 0 ? 0 : id;
796}
797
798static void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
799{
800
801 if (!prog->aux->id)
802 return;
803
804 if (do_idr_lock)
805 spin_lock_bh(&prog_idr_lock);
806 else
807 __acquire(&prog_idr_lock);
808
809 idr_remove(&prog_idr, prog->aux->id);
810
811 if (do_idr_lock)
812 spin_unlock_bh(&prog_idr_lock);
813 else
814 __release(&prog_idr_lock);
815}
816
817static void __bpf_prog_put_rcu(struct rcu_head *rcu)
818{
819 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
820
821 free_used_maps(aux);
822 bpf_prog_uncharge_memlock(aux->prog);
823 bpf_prog_free(aux->prog);
824}
825
826static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
827{
828 if (atomic_dec_and_test(&prog->aux->refcnt)) {
829 trace_bpf_prog_put_rcu(prog);
830
831 bpf_prog_free_id(prog, do_idr_lock);
832 bpf_prog_kallsyms_del(prog);
833 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
834 }
835}
836
837void bpf_prog_put(struct bpf_prog *prog)
838{
839 __bpf_prog_put(prog, true);
840}
841EXPORT_SYMBOL_GPL(bpf_prog_put);
842
843static int bpf_prog_release(struct inode *inode, struct file *filp)
844{
845 struct bpf_prog *prog = filp->private_data;
846
847 bpf_prog_put(prog);
848 return 0;
849}
850
851#ifdef CONFIG_PROC_FS
852static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
853{
854 const struct bpf_prog *prog = filp->private_data;
855 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
856
857 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
858 seq_printf(m,
859 "prog_type:\t%u\n"
860 "prog_jited:\t%u\n"
861 "prog_tag:\t%s\n"
862 "memlock:\t%llu\n",
863 prog->type,
864 prog->jited,
865 prog_tag,
866 prog->pages * 1ULL << PAGE_SHIFT);
867}
868#endif
869
870static const struct file_operations bpf_prog_fops = {
871#ifdef CONFIG_PROC_FS
872 .show_fdinfo = bpf_prog_show_fdinfo,
873#endif
874 .release = bpf_prog_release,
875};
876
877int bpf_prog_new_fd(struct bpf_prog *prog)
878{
879 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
880 O_RDWR | O_CLOEXEC);
881}
882
883static struct bpf_prog *____bpf_prog_get(struct fd f)
884{
885 if (!f.file)
886 return ERR_PTR(-EBADF);
887 if (f.file->f_op != &bpf_prog_fops) {
888 fdput(f);
889 return ERR_PTR(-EINVAL);
890 }
891
892 return f.file->private_data;
893}
894
895struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
896{
897 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
898 atomic_sub(i, &prog->aux->refcnt);
899 return ERR_PTR(-EBUSY);
900 }
901 return prog;
902}
903EXPORT_SYMBOL_GPL(bpf_prog_add);
904
905void bpf_prog_sub(struct bpf_prog *prog, int i)
906{
907
908
909
910
911
912 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
913}
914EXPORT_SYMBOL_GPL(bpf_prog_sub);
915
916struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
917{
918 return bpf_prog_add(prog, 1);
919}
920EXPORT_SYMBOL_GPL(bpf_prog_inc);
921
922
923struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
924{
925 int refold;
926
927 refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
928
929 if (refold >= BPF_MAX_REFCNT) {
930 __bpf_prog_put(prog, false);
931 return ERR_PTR(-EBUSY);
932 }
933
934 if (!refold)
935 return ERR_PTR(-ENOENT);
936
937 return prog;
938}
939EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
940
941static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type)
942{
943 struct fd f = fdget(ufd);
944 struct bpf_prog *prog;
945
946 prog = ____bpf_prog_get(f);
947 if (IS_ERR(prog))
948 return prog;
949 if (type && prog->type != *type) {
950 prog = ERR_PTR(-EINVAL);
951 goto out;
952 }
953
954 prog = bpf_prog_inc(prog);
955out:
956 fdput(f);
957 return prog;
958}
959
960struct bpf_prog *bpf_prog_get(u32 ufd)
961{
962 return __bpf_prog_get(ufd, NULL);
963}
964
965struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type)
966{
967 struct bpf_prog *prog = __bpf_prog_get(ufd, &type);
968
969 if (!IS_ERR(prog))
970 trace_bpf_prog_get_type(prog);
971 return prog;
972}
973EXPORT_SYMBOL_GPL(bpf_prog_get_type);
974
975
976#define BPF_PROG_LOAD_LAST_FIELD prog_flags
977
978static int bpf_prog_load(union bpf_attr *attr)
979{
980 enum bpf_prog_type type = attr->prog_type;
981 struct bpf_prog *prog;
982 int err;
983 char license[128];
984 bool is_gpl;
985
986 if (CHECK_ATTR(BPF_PROG_LOAD))
987 return -EINVAL;
988
989 if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
990 return -EINVAL;
991
992
993 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
994 sizeof(license) - 1) < 0)
995 return -EFAULT;
996 license[sizeof(license) - 1] = 0;
997
998
999 is_gpl = license_is_gpl_compatible(license);
1000
1001 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1002 return -E2BIG;
1003
1004 if (type == BPF_PROG_TYPE_KPROBE &&
1005 attr->kern_version != LINUX_VERSION_CODE)
1006 return -EINVAL;
1007
1008 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1009 type != BPF_PROG_TYPE_CGROUP_SKB &&
1010 !capable(CAP_SYS_ADMIN))
1011 return -EPERM;
1012
1013
1014 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1015 if (!prog)
1016 return -ENOMEM;
1017
1018 err = bpf_prog_charge_memlock(prog);
1019 if (err)
1020 goto free_prog_nouncharge;
1021
1022 prog->len = attr->insn_cnt;
1023
1024 err = -EFAULT;
1025 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1026 bpf_prog_insn_size(prog)) != 0)
1027 goto free_prog;
1028
1029 prog->orig_prog = NULL;
1030 prog->jited = 0;
1031
1032 atomic_set(&prog->aux->refcnt, 1);
1033 prog->gpl_compatible = is_gpl ? 1 : 0;
1034
1035
1036 err = find_prog_type(type, prog);
1037 if (err < 0)
1038 goto free_prog;
1039
1040
1041 err = bpf_check(&prog, attr);
1042 if (err < 0)
1043 goto free_used_maps;
1044
1045
1046 prog = bpf_prog_select_runtime(prog, &err);
1047 if (err < 0)
1048 goto free_used_maps;
1049
1050 err = bpf_prog_alloc_id(prog);
1051 if (err)
1052 goto free_used_maps;
1053
1054 err = bpf_prog_new_fd(prog);
1055 if (err < 0) {
1056
1057
1058
1059
1060
1061
1062 bpf_prog_put(prog);
1063 return err;
1064 }
1065
1066 bpf_prog_kallsyms_add(prog);
1067 trace_bpf_prog_load(prog, err);
1068 return err;
1069
1070free_used_maps:
1071 free_used_maps(prog->aux);
1072free_prog:
1073 bpf_prog_uncharge_memlock(prog);
1074free_prog_nouncharge:
1075 bpf_prog_free(prog);
1076 return err;
1077}
1078
1079#define BPF_OBJ_LAST_FIELD bpf_fd
1080
1081static int bpf_obj_pin(const union bpf_attr *attr)
1082{
1083 if (CHECK_ATTR(BPF_OBJ))
1084 return -EINVAL;
1085
1086 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1087}
1088
1089static int bpf_obj_get(const union bpf_attr *attr)
1090{
1091 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0)
1092 return -EINVAL;
1093
1094 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname));
1095}
1096
1097#ifdef CONFIG_CGROUP_BPF
1098
1099#define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1100
1101static int sockmap_get_from_fd(const union bpf_attr *attr, bool attach)
1102{
1103 struct bpf_prog *prog = NULL;
1104 int ufd = attr->target_fd;
1105 struct bpf_map *map;
1106 struct fd f;
1107 int err;
1108
1109 f = fdget(ufd);
1110 map = __bpf_map_get(f);
1111 if (IS_ERR(map))
1112 return PTR_ERR(map);
1113
1114 if (attach) {
1115 prog = bpf_prog_get_type(attr->attach_bpf_fd,
1116 BPF_PROG_TYPE_SK_SKB);
1117 if (IS_ERR(prog)) {
1118 fdput(f);
1119 return PTR_ERR(prog);
1120 }
1121 }
1122
1123 err = sock_map_prog(map, prog, attr->attach_type);
1124 if (err) {
1125 fdput(f);
1126 if (prog)
1127 bpf_prog_put(prog);
1128 return err;
1129 }
1130
1131 fdput(f);
1132 return 0;
1133}
1134
1135static int bpf_prog_attach(const union bpf_attr *attr)
1136{
1137 enum bpf_prog_type ptype;
1138 struct bpf_prog *prog;
1139 struct cgroup *cgrp;
1140 int ret;
1141
1142 if (!capable(CAP_NET_ADMIN))
1143 return -EPERM;
1144
1145 if (CHECK_ATTR(BPF_PROG_ATTACH))
1146 return -EINVAL;
1147
1148 if (attr->attach_flags & ~BPF_F_ALLOW_OVERRIDE)
1149 return -EINVAL;
1150
1151 switch (attr->attach_type) {
1152 case BPF_CGROUP_INET_INGRESS:
1153 case BPF_CGROUP_INET_EGRESS:
1154 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1155 break;
1156 case BPF_CGROUP_INET_SOCK_CREATE:
1157 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1158 break;
1159 case BPF_CGROUP_SOCK_OPS:
1160 ptype = BPF_PROG_TYPE_SOCK_OPS;
1161 break;
1162 case BPF_SK_SKB_STREAM_PARSER:
1163 case BPF_SK_SKB_STREAM_VERDICT:
1164 return sockmap_get_from_fd(attr, true);
1165 default:
1166 return -EINVAL;
1167 }
1168
1169 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1170 if (IS_ERR(prog))
1171 return PTR_ERR(prog);
1172
1173 cgrp = cgroup_get_from_fd(attr->target_fd);
1174 if (IS_ERR(cgrp)) {
1175 bpf_prog_put(prog);
1176 return PTR_ERR(cgrp);
1177 }
1178
1179 ret = cgroup_bpf_update(cgrp, prog, attr->attach_type,
1180 attr->attach_flags & BPF_F_ALLOW_OVERRIDE);
1181 if (ret)
1182 bpf_prog_put(prog);
1183 cgroup_put(cgrp);
1184
1185 return ret;
1186}
1187
1188#define BPF_PROG_DETACH_LAST_FIELD attach_type
1189
1190static int bpf_prog_detach(const union bpf_attr *attr)
1191{
1192 struct cgroup *cgrp;
1193 int ret;
1194
1195 if (!capable(CAP_NET_ADMIN))
1196 return -EPERM;
1197
1198 if (CHECK_ATTR(BPF_PROG_DETACH))
1199 return -EINVAL;
1200
1201 switch (attr->attach_type) {
1202 case BPF_CGROUP_INET_INGRESS:
1203 case BPF_CGROUP_INET_EGRESS:
1204 case BPF_CGROUP_INET_SOCK_CREATE:
1205 case BPF_CGROUP_SOCK_OPS:
1206 cgrp = cgroup_get_from_fd(attr->target_fd);
1207 if (IS_ERR(cgrp))
1208 return PTR_ERR(cgrp);
1209
1210 ret = cgroup_bpf_update(cgrp, NULL, attr->attach_type, false);
1211 cgroup_put(cgrp);
1212 break;
1213 case BPF_SK_SKB_STREAM_PARSER:
1214 case BPF_SK_SKB_STREAM_VERDICT:
1215 ret = sockmap_get_from_fd(attr, false);
1216 break;
1217 default:
1218 return -EINVAL;
1219 }
1220
1221 return ret;
1222}
1223
1224#endif
1225
1226#define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1227
1228static int bpf_prog_test_run(const union bpf_attr *attr,
1229 union bpf_attr __user *uattr)
1230{
1231 struct bpf_prog *prog;
1232 int ret = -ENOTSUPP;
1233
1234 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1235 return -EINVAL;
1236
1237 prog = bpf_prog_get(attr->test.prog_fd);
1238 if (IS_ERR(prog))
1239 return PTR_ERR(prog);
1240
1241 if (prog->aux->ops->test_run)
1242 ret = prog->aux->ops->test_run(prog, attr, uattr);
1243
1244 bpf_prog_put(prog);
1245 return ret;
1246}
1247
1248#define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1249
1250static int bpf_obj_get_next_id(const union bpf_attr *attr,
1251 union bpf_attr __user *uattr,
1252 struct idr *idr,
1253 spinlock_t *lock)
1254{
1255 u32 next_id = attr->start_id;
1256 int err = 0;
1257
1258 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1259 return -EINVAL;
1260
1261 if (!capable(CAP_SYS_ADMIN))
1262 return -EPERM;
1263
1264 next_id++;
1265 spin_lock_bh(lock);
1266 if (!idr_get_next(idr, &next_id))
1267 err = -ENOENT;
1268 spin_unlock_bh(lock);
1269
1270 if (!err)
1271 err = put_user(next_id, &uattr->next_id);
1272
1273 return err;
1274}
1275
1276#define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1277
1278static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1279{
1280 struct bpf_prog *prog;
1281 u32 id = attr->prog_id;
1282 int fd;
1283
1284 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1285 return -EINVAL;
1286
1287 if (!capable(CAP_SYS_ADMIN))
1288 return -EPERM;
1289
1290 spin_lock_bh(&prog_idr_lock);
1291 prog = idr_find(&prog_idr, id);
1292 if (prog)
1293 prog = bpf_prog_inc_not_zero(prog);
1294 else
1295 prog = ERR_PTR(-ENOENT);
1296 spin_unlock_bh(&prog_idr_lock);
1297
1298 if (IS_ERR(prog))
1299 return PTR_ERR(prog);
1300
1301 fd = bpf_prog_new_fd(prog);
1302 if (fd < 0)
1303 bpf_prog_put(prog);
1304
1305 return fd;
1306}
1307
1308#define BPF_MAP_GET_FD_BY_ID_LAST_FIELD map_id
1309
1310static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1311{
1312 struct bpf_map *map;
1313 u32 id = attr->map_id;
1314 int fd;
1315
1316 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID))
1317 return -EINVAL;
1318
1319 if (!capable(CAP_SYS_ADMIN))
1320 return -EPERM;
1321
1322 spin_lock_bh(&map_idr_lock);
1323 map = idr_find(&map_idr, id);
1324 if (map)
1325 map = bpf_map_inc_not_zero(map, true);
1326 else
1327 map = ERR_PTR(-ENOENT);
1328 spin_unlock_bh(&map_idr_lock);
1329
1330 if (IS_ERR(map))
1331 return PTR_ERR(map);
1332
1333 fd = bpf_map_new_fd(map);
1334 if (fd < 0)
1335 bpf_map_put(map);
1336
1337 return fd;
1338}
1339
1340static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1341 const union bpf_attr *attr,
1342 union bpf_attr __user *uattr)
1343{
1344 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1345 struct bpf_prog_info info = {};
1346 u32 info_len = attr->info.info_len;
1347 char __user *uinsns;
1348 u32 ulen;
1349 int err;
1350
1351 err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1352 if (err)
1353 return err;
1354 info_len = min_t(u32, sizeof(info), info_len);
1355
1356 if (copy_from_user(&info, uinfo, info_len))
1357 return -EFAULT;
1358
1359 info.type = prog->type;
1360 info.id = prog->aux->id;
1361
1362 memcpy(info.tag, prog->tag, sizeof(prog->tag));
1363
1364 if (!capable(CAP_SYS_ADMIN)) {
1365 info.jited_prog_len = 0;
1366 info.xlated_prog_len = 0;
1367 goto done;
1368 }
1369
1370 ulen = info.jited_prog_len;
1371 info.jited_prog_len = prog->jited_len;
1372 if (info.jited_prog_len && ulen) {
1373 uinsns = u64_to_user_ptr(info.jited_prog_insns);
1374 ulen = min_t(u32, info.jited_prog_len, ulen);
1375 if (copy_to_user(uinsns, prog->bpf_func, ulen))
1376 return -EFAULT;
1377 }
1378
1379 ulen = info.xlated_prog_len;
1380 info.xlated_prog_len = bpf_prog_insn_size(prog);
1381 if (info.xlated_prog_len && ulen) {
1382 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1383 ulen = min_t(u32, info.xlated_prog_len, ulen);
1384 if (copy_to_user(uinsns, prog->insnsi, ulen))
1385 return -EFAULT;
1386 }
1387
1388done:
1389 if (copy_to_user(uinfo, &info, info_len) ||
1390 put_user(info_len, &uattr->info.info_len))
1391 return -EFAULT;
1392
1393 return 0;
1394}
1395
1396static int bpf_map_get_info_by_fd(struct bpf_map *map,
1397 const union bpf_attr *attr,
1398 union bpf_attr __user *uattr)
1399{
1400 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1401 struct bpf_map_info info = {};
1402 u32 info_len = attr->info.info_len;
1403 int err;
1404
1405 err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1406 if (err)
1407 return err;
1408 info_len = min_t(u32, sizeof(info), info_len);
1409
1410 info.type = map->map_type;
1411 info.id = map->id;
1412 info.key_size = map->key_size;
1413 info.value_size = map->value_size;
1414 info.max_entries = map->max_entries;
1415 info.map_flags = map->map_flags;
1416
1417 if (copy_to_user(uinfo, &info, info_len) ||
1418 put_user(info_len, &uattr->info.info_len))
1419 return -EFAULT;
1420
1421 return 0;
1422}
1423
1424#define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
1425
1426static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
1427 union bpf_attr __user *uattr)
1428{
1429 int ufd = attr->info.bpf_fd;
1430 struct fd f;
1431 int err;
1432
1433 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
1434 return -EINVAL;
1435
1436 f = fdget(ufd);
1437 if (!f.file)
1438 return -EBADFD;
1439
1440 if (f.file->f_op == &bpf_prog_fops)
1441 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
1442 uattr);
1443 else if (f.file->f_op == &bpf_map_fops)
1444 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
1445 uattr);
1446 else
1447 err = -EINVAL;
1448
1449 fdput(f);
1450 return err;
1451}
1452
1453SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
1454{
1455 union bpf_attr attr = {};
1456 int err;
1457
1458 if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
1459 return -EPERM;
1460
1461 err = check_uarg_tail_zero(uattr, sizeof(attr), size);
1462 if (err)
1463 return err;
1464 size = min_t(u32, size, sizeof(attr));
1465
1466
1467 if (copy_from_user(&attr, uattr, size) != 0)
1468 return -EFAULT;
1469
1470 switch (cmd) {
1471 case BPF_MAP_CREATE:
1472 err = map_create(&attr);
1473 break;
1474 case BPF_MAP_LOOKUP_ELEM:
1475 err = map_lookup_elem(&attr);
1476 break;
1477 case BPF_MAP_UPDATE_ELEM:
1478 err = map_update_elem(&attr);
1479 break;
1480 case BPF_MAP_DELETE_ELEM:
1481 err = map_delete_elem(&attr);
1482 break;
1483 case BPF_MAP_GET_NEXT_KEY:
1484 err = map_get_next_key(&attr);
1485 break;
1486 case BPF_PROG_LOAD:
1487 err = bpf_prog_load(&attr);
1488 break;
1489 case BPF_OBJ_PIN:
1490 err = bpf_obj_pin(&attr);
1491 break;
1492 case BPF_OBJ_GET:
1493 err = bpf_obj_get(&attr);
1494 break;
1495#ifdef CONFIG_CGROUP_BPF
1496 case BPF_PROG_ATTACH:
1497 err = bpf_prog_attach(&attr);
1498 break;
1499 case BPF_PROG_DETACH:
1500 err = bpf_prog_detach(&attr);
1501 break;
1502#endif
1503 case BPF_PROG_TEST_RUN:
1504 err = bpf_prog_test_run(&attr, uattr);
1505 break;
1506 case BPF_PROG_GET_NEXT_ID:
1507 err = bpf_obj_get_next_id(&attr, uattr,
1508 &prog_idr, &prog_idr_lock);
1509 break;
1510 case BPF_MAP_GET_NEXT_ID:
1511 err = bpf_obj_get_next_id(&attr, uattr,
1512 &map_idr, &map_idr_lock);
1513 break;
1514 case BPF_PROG_GET_FD_BY_ID:
1515 err = bpf_prog_get_fd_by_id(&attr);
1516 break;
1517 case BPF_MAP_GET_FD_BY_ID:
1518 err = bpf_map_get_fd_by_id(&attr);
1519 break;
1520 case BPF_OBJ_GET_INFO_BY_FD:
1521 err = bpf_obj_get_info_by_fd(&attr, uattr);
1522 break;
1523 default:
1524 err = -EINVAL;
1525 break;
1526 }
1527
1528 return err;
1529}
1530
