Loading...
1/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2 *
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.
6 *
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.
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#include <linux/cred.h>
27#include <linux/timekeeping.h>
28#include <linux/ctype.h>
29#include <linux/nospec.h>
30
31#define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
32 (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
33 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
34 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
35#define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
36#define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
37
38#define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
39
40DEFINE_PER_CPU(int, bpf_prog_active);
41static DEFINE_IDR(prog_idr);
42static DEFINE_SPINLOCK(prog_idr_lock);
43static DEFINE_IDR(map_idr);
44static DEFINE_SPINLOCK(map_idr_lock);
45
46int sysctl_unprivileged_bpf_disabled __read_mostly;
47
48static const struct bpf_map_ops * const bpf_map_types[] = {
49#define BPF_PROG_TYPE(_id, _ops)
50#define BPF_MAP_TYPE(_id, _ops) \
51 [_id] = &_ops,
52#include <linux/bpf_types.h>
53#undef BPF_PROG_TYPE
54#undef BPF_MAP_TYPE
55};
56
57/*
58 * If we're handed a bigger struct than we know of, ensure all the unknown bits
59 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
60 * we don't know about yet.
61 *
62 * There is a ToCToU between this function call and the following
63 * copy_from_user() call. However, this is not a concern since this function is
64 * meant to be a future-proofing of bits.
65 */
66static int check_uarg_tail_zero(void __user *uaddr,
67 size_t expected_size,
68 size_t actual_size)
69{
70 unsigned char __user *addr;
71 unsigned char __user *end;
72 unsigned char val;
73 int err;
74
75 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
76 return -E2BIG;
77
78 if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
79 return -EFAULT;
80
81 if (actual_size <= expected_size)
82 return 0;
83
84 addr = uaddr + expected_size;
85 end = uaddr + actual_size;
86
87 for (; addr < end; addr++) {
88 err = get_user(val, addr);
89 if (err)
90 return err;
91 if (val)
92 return -E2BIG;
93 }
94
95 return 0;
96}
97
98const struct bpf_map_ops bpf_map_offload_ops = {
99 .map_alloc = bpf_map_offload_map_alloc,
100 .map_free = bpf_map_offload_map_free,
101};
102
103static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
104{
105 const struct bpf_map_ops *ops;
106 u32 type = attr->map_type;
107 struct bpf_map *map;
108 int err;
109
110 if (type >= ARRAY_SIZE(bpf_map_types))
111 return ERR_PTR(-EINVAL);
112 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
113 ops = bpf_map_types[type];
114 if (!ops)
115 return ERR_PTR(-EINVAL);
116
117 if (ops->map_alloc_check) {
118 err = ops->map_alloc_check(attr);
119 if (err)
120 return ERR_PTR(err);
121 }
122 if (attr->map_ifindex)
123 ops = &bpf_map_offload_ops;
124 map = ops->map_alloc(attr);
125 if (IS_ERR(map))
126 return map;
127 map->ops = ops;
128 map->map_type = type;
129 return map;
130}
131
132void *bpf_map_area_alloc(size_t size, int numa_node)
133{
134 /* We definitely need __GFP_NORETRY, so OOM killer doesn't
135 * trigger under memory pressure as we really just want to
136 * fail instead.
137 */
138 const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
139 void *area;
140
141 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
142 area = kmalloc_node(size, GFP_USER | flags, numa_node);
143 if (area != NULL)
144 return area;
145 }
146
147 return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
148 __builtin_return_address(0));
149}
150
151void bpf_map_area_free(void *area)
152{
153 kvfree(area);
154}
155
156void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
157{
158 map->map_type = attr->map_type;
159 map->key_size = attr->key_size;
160 map->value_size = attr->value_size;
161 map->max_entries = attr->max_entries;
162 map->map_flags = attr->map_flags;
163 map->numa_node = bpf_map_attr_numa_node(attr);
164}
165
166int bpf_map_precharge_memlock(u32 pages)
167{
168 struct user_struct *user = get_current_user();
169 unsigned long memlock_limit, cur;
170
171 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
172 cur = atomic_long_read(&user->locked_vm);
173 free_uid(user);
174 if (cur + pages > memlock_limit)
175 return -EPERM;
176 return 0;
177}
178
179static int bpf_map_charge_memlock(struct bpf_map *map)
180{
181 struct user_struct *user = get_current_user();
182 unsigned long memlock_limit;
183
184 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
185
186 atomic_long_add(map->pages, &user->locked_vm);
187
188 if (atomic_long_read(&user->locked_vm) > memlock_limit) {
189 atomic_long_sub(map->pages, &user->locked_vm);
190 free_uid(user);
191 return -EPERM;
192 }
193 map->user = user;
194 return 0;
195}
196
197static void bpf_map_uncharge_memlock(struct bpf_map *map)
198{
199 struct user_struct *user = map->user;
200
201 atomic_long_sub(map->pages, &user->locked_vm);
202 free_uid(user);
203}
204
205static int bpf_map_alloc_id(struct bpf_map *map)
206{
207 int id;
208
209 idr_preload(GFP_KERNEL);
210 spin_lock_bh(&map_idr_lock);
211 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
212 if (id > 0)
213 map->id = id;
214 spin_unlock_bh(&map_idr_lock);
215 idr_preload_end();
216
217 if (WARN_ON_ONCE(!id))
218 return -ENOSPC;
219
220 return id > 0 ? 0 : id;
221}
222
223void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
224{
225 unsigned long flags;
226
227 /* Offloaded maps are removed from the IDR store when their device
228 * disappears - even if someone holds an fd to them they are unusable,
229 * the memory is gone, all ops will fail; they are simply waiting for
230 * refcnt to drop to be freed.
231 */
232 if (!map->id)
233 return;
234
235 if (do_idr_lock)
236 spin_lock_irqsave(&map_idr_lock, flags);
237 else
238 __acquire(&map_idr_lock);
239
240 idr_remove(&map_idr, map->id);
241 map->id = 0;
242
243 if (do_idr_lock)
244 spin_unlock_irqrestore(&map_idr_lock, flags);
245 else
246 __release(&map_idr_lock);
247}
248
249/* called from workqueue */
250static void bpf_map_free_deferred(struct work_struct *work)
251{
252 struct bpf_map *map = container_of(work, struct bpf_map, work);
253
254 bpf_map_uncharge_memlock(map);
255 security_bpf_map_free(map);
256 /* implementation dependent freeing */
257 map->ops->map_free(map);
258}
259
260static void bpf_map_put_uref(struct bpf_map *map)
261{
262 if (atomic_dec_and_test(&map->usercnt)) {
263 if (map->ops->map_release_uref)
264 map->ops->map_release_uref(map);
265 }
266}
267
268/* decrement map refcnt and schedule it for freeing via workqueue
269 * (unrelying map implementation ops->map_free() might sleep)
270 */
271static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
272{
273 if (atomic_dec_and_test(&map->refcnt)) {
274 /* bpf_map_free_id() must be called first */
275 bpf_map_free_id(map, do_idr_lock);
276 INIT_WORK(&map->work, bpf_map_free_deferred);
277 schedule_work(&map->work);
278 }
279}
280
281void bpf_map_put(struct bpf_map *map)
282{
283 __bpf_map_put(map, true);
284}
285
286void bpf_map_put_with_uref(struct bpf_map *map)
287{
288 bpf_map_put_uref(map);
289 bpf_map_put(map);
290}
291
292static int bpf_map_release(struct inode *inode, struct file *filp)
293{
294 struct bpf_map *map = filp->private_data;
295
296 if (map->ops->map_release)
297 map->ops->map_release(map, filp);
298
299 bpf_map_put_with_uref(map);
300 return 0;
301}
302
303#ifdef CONFIG_PROC_FS
304static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
305{
306 const struct bpf_map *map = filp->private_data;
307 const struct bpf_array *array;
308 u32 owner_prog_type = 0;
309 u32 owner_jited = 0;
310
311 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
312 array = container_of(map, struct bpf_array, map);
313 owner_prog_type = array->owner_prog_type;
314 owner_jited = array->owner_jited;
315 }
316
317 seq_printf(m,
318 "map_type:\t%u\n"
319 "key_size:\t%u\n"
320 "value_size:\t%u\n"
321 "max_entries:\t%u\n"
322 "map_flags:\t%#x\n"
323 "memlock:\t%llu\n",
324 map->map_type,
325 map->key_size,
326 map->value_size,
327 map->max_entries,
328 map->map_flags,
329 map->pages * 1ULL << PAGE_SHIFT);
330
331 if (owner_prog_type) {
332 seq_printf(m, "owner_prog_type:\t%u\n",
333 owner_prog_type);
334 seq_printf(m, "owner_jited:\t%u\n",
335 owner_jited);
336 }
337}
338#endif
339
340static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
341 loff_t *ppos)
342{
343 /* We need this handler such that alloc_file() enables
344 * f_mode with FMODE_CAN_READ.
345 */
346 return -EINVAL;
347}
348
349static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
350 size_t siz, loff_t *ppos)
351{
352 /* We need this handler such that alloc_file() enables
353 * f_mode with FMODE_CAN_WRITE.
354 */
355 return -EINVAL;
356}
357
358const struct file_operations bpf_map_fops = {
359#ifdef CONFIG_PROC_FS
360 .show_fdinfo = bpf_map_show_fdinfo,
361#endif
362 .release = bpf_map_release,
363 .read = bpf_dummy_read,
364 .write = bpf_dummy_write,
365};
366
367int bpf_map_new_fd(struct bpf_map *map, int flags)
368{
369 int ret;
370
371 ret = security_bpf_map(map, OPEN_FMODE(flags));
372 if (ret < 0)
373 return ret;
374
375 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
376 flags | O_CLOEXEC);
377}
378
379int bpf_get_file_flag(int flags)
380{
381 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
382 return -EINVAL;
383 if (flags & BPF_F_RDONLY)
384 return O_RDONLY;
385 if (flags & BPF_F_WRONLY)
386 return O_WRONLY;
387 return O_RDWR;
388}
389
390/* helper macro to check that unused fields 'union bpf_attr' are zero */
391#define CHECK_ATTR(CMD) \
392 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
393 sizeof(attr->CMD##_LAST_FIELD), 0, \
394 sizeof(*attr) - \
395 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
396 sizeof(attr->CMD##_LAST_FIELD)) != NULL
397
398/* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
399 * Return 0 on success and < 0 on error.
400 */
401static int bpf_obj_name_cpy(char *dst, const char *src)
402{
403 const char *end = src + BPF_OBJ_NAME_LEN;
404
405 memset(dst, 0, BPF_OBJ_NAME_LEN);
406
407 /* Copy all isalnum() and '_' char */
408 while (src < end && *src) {
409 if (!isalnum(*src) && *src != '_')
410 return -EINVAL;
411 *dst++ = *src++;
412 }
413
414 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
415 if (src == end)
416 return -EINVAL;
417
418 return 0;
419}
420
421#define BPF_MAP_CREATE_LAST_FIELD map_ifindex
422/* called via syscall */
423static int map_create(union bpf_attr *attr)
424{
425 int numa_node = bpf_map_attr_numa_node(attr);
426 struct bpf_map *map;
427 int f_flags;
428 int err;
429
430 err = CHECK_ATTR(BPF_MAP_CREATE);
431 if (err)
432 return -EINVAL;
433
434 f_flags = bpf_get_file_flag(attr->map_flags);
435 if (f_flags < 0)
436 return f_flags;
437
438 if (numa_node != NUMA_NO_NODE &&
439 ((unsigned int)numa_node >= nr_node_ids ||
440 !node_online(numa_node)))
441 return -EINVAL;
442
443 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
444 map = find_and_alloc_map(attr);
445 if (IS_ERR(map))
446 return PTR_ERR(map);
447
448 err = bpf_obj_name_cpy(map->name, attr->map_name);
449 if (err)
450 goto free_map_nouncharge;
451
452 atomic_set(&map->refcnt, 1);
453 atomic_set(&map->usercnt, 1);
454
455 err = security_bpf_map_alloc(map);
456 if (err)
457 goto free_map_nouncharge;
458
459 err = bpf_map_charge_memlock(map);
460 if (err)
461 goto free_map_sec;
462
463 err = bpf_map_alloc_id(map);
464 if (err)
465 goto free_map;
466
467 err = bpf_map_new_fd(map, f_flags);
468 if (err < 0) {
469 /* failed to allocate fd.
470 * bpf_map_put() is needed because the above
471 * bpf_map_alloc_id() has published the map
472 * to the userspace and the userspace may
473 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
474 */
475 bpf_map_put(map);
476 return err;
477 }
478
479 trace_bpf_map_create(map, err);
480 return err;
481
482free_map:
483 bpf_map_uncharge_memlock(map);
484free_map_sec:
485 security_bpf_map_free(map);
486free_map_nouncharge:
487 map->ops->map_free(map);
488 return err;
489}
490
491/* if error is returned, fd is released.
492 * On success caller should complete fd access with matching fdput()
493 */
494struct bpf_map *__bpf_map_get(struct fd f)
495{
496 if (!f.file)
497 return ERR_PTR(-EBADF);
498 if (f.file->f_op != &bpf_map_fops) {
499 fdput(f);
500 return ERR_PTR(-EINVAL);
501 }
502
503 return f.file->private_data;
504}
505
506/* prog's and map's refcnt limit */
507#define BPF_MAX_REFCNT 32768
508
509struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
510{
511 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
512 atomic_dec(&map->refcnt);
513 return ERR_PTR(-EBUSY);
514 }
515 if (uref)
516 atomic_inc(&map->usercnt);
517 return map;
518}
519
520struct bpf_map *bpf_map_get_with_uref(u32 ufd)
521{
522 struct fd f = fdget(ufd);
523 struct bpf_map *map;
524
525 map = __bpf_map_get(f);
526 if (IS_ERR(map))
527 return map;
528
529 map = bpf_map_inc(map, true);
530 fdput(f);
531
532 return map;
533}
534
535/* map_idr_lock should have been held */
536static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
537 bool uref)
538{
539 int refold;
540
541 refold = __atomic_add_unless(&map->refcnt, 1, 0);
542
543 if (refold >= BPF_MAX_REFCNT) {
544 __bpf_map_put(map, false);
545 return ERR_PTR(-EBUSY);
546 }
547
548 if (!refold)
549 return ERR_PTR(-ENOENT);
550
551 if (uref)
552 atomic_inc(&map->usercnt);
553
554 return map;
555}
556
557int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
558{
559 return -ENOTSUPP;
560}
561
562/* last field in 'union bpf_attr' used by this command */
563#define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
564
565static int map_lookup_elem(union bpf_attr *attr)
566{
567 void __user *ukey = u64_to_user_ptr(attr->key);
568 void __user *uvalue = u64_to_user_ptr(attr->value);
569 int ufd = attr->map_fd;
570 struct bpf_map *map;
571 void *key, *value, *ptr;
572 u32 value_size;
573 struct fd f;
574 int err;
575
576 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
577 return -EINVAL;
578
579 f = fdget(ufd);
580 map = __bpf_map_get(f);
581 if (IS_ERR(map))
582 return PTR_ERR(map);
583
584 if (!(f.file->f_mode & FMODE_CAN_READ)) {
585 err = -EPERM;
586 goto err_put;
587 }
588
589 key = memdup_user(ukey, map->key_size);
590 if (IS_ERR(key)) {
591 err = PTR_ERR(key);
592 goto err_put;
593 }
594
595 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
596 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
597 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
598 value_size = round_up(map->value_size, 8) * num_possible_cpus();
599 else if (IS_FD_MAP(map))
600 value_size = sizeof(u32);
601 else
602 value_size = map->value_size;
603
604 err = -ENOMEM;
605 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
606 if (!value)
607 goto free_key;
608
609 if (bpf_map_is_dev_bound(map)) {
610 err = bpf_map_offload_lookup_elem(map, key, value);
611 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
612 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
613 err = bpf_percpu_hash_copy(map, key, value);
614 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
615 err = bpf_percpu_array_copy(map, key, value);
616 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
617 err = bpf_stackmap_copy(map, key, value);
618 } else if (IS_FD_ARRAY(map)) {
619 err = bpf_fd_array_map_lookup_elem(map, key, value);
620 } else if (IS_FD_HASH(map)) {
621 err = bpf_fd_htab_map_lookup_elem(map, key, value);
622 } else {
623 rcu_read_lock();
624 ptr = map->ops->map_lookup_elem(map, key);
625 if (ptr)
626 memcpy(value, ptr, value_size);
627 rcu_read_unlock();
628 err = ptr ? 0 : -ENOENT;
629 }
630
631 if (err)
632 goto free_value;
633
634 err = -EFAULT;
635 if (copy_to_user(uvalue, value, value_size) != 0)
636 goto free_value;
637
638 trace_bpf_map_lookup_elem(map, ufd, key, value);
639 err = 0;
640
641free_value:
642 kfree(value);
643free_key:
644 kfree(key);
645err_put:
646 fdput(f);
647 return err;
648}
649
650#define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
651
652static int map_update_elem(union bpf_attr *attr)
653{
654 void __user *ukey = u64_to_user_ptr(attr->key);
655 void __user *uvalue = u64_to_user_ptr(attr->value);
656 int ufd = attr->map_fd;
657 struct bpf_map *map;
658 void *key, *value;
659 u32 value_size;
660 struct fd f;
661 int err;
662
663 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
664 return -EINVAL;
665
666 f = fdget(ufd);
667 map = __bpf_map_get(f);
668 if (IS_ERR(map))
669 return PTR_ERR(map);
670
671 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
672 err = -EPERM;
673 goto err_put;
674 }
675
676 key = memdup_user(ukey, map->key_size);
677 if (IS_ERR(key)) {
678 err = PTR_ERR(key);
679 goto err_put;
680 }
681
682 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
683 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
684 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
685 value_size = round_up(map->value_size, 8) * num_possible_cpus();
686 else
687 value_size = map->value_size;
688
689 err = -ENOMEM;
690 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
691 if (!value)
692 goto free_key;
693
694 err = -EFAULT;
695 if (copy_from_user(value, uvalue, value_size) != 0)
696 goto free_value;
697
698 /* Need to create a kthread, thus must support schedule */
699 if (bpf_map_is_dev_bound(map)) {
700 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
701 goto out;
702 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP) {
703 err = map->ops->map_update_elem(map, key, value, attr->flags);
704 goto out;
705 }
706
707 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
708 * inside bpf map update or delete otherwise deadlocks are possible
709 */
710 preempt_disable();
711 __this_cpu_inc(bpf_prog_active);
712 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
713 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
714 err = bpf_percpu_hash_update(map, key, value, attr->flags);
715 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
716 err = bpf_percpu_array_update(map, key, value, attr->flags);
717 } else if (IS_FD_ARRAY(map)) {
718 rcu_read_lock();
719 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
720 attr->flags);
721 rcu_read_unlock();
722 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
723 rcu_read_lock();
724 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
725 attr->flags);
726 rcu_read_unlock();
727 } else {
728 rcu_read_lock();
729 err = map->ops->map_update_elem(map, key, value, attr->flags);
730 rcu_read_unlock();
731 }
732 __this_cpu_dec(bpf_prog_active);
733 preempt_enable();
734out:
735 if (!err)
736 trace_bpf_map_update_elem(map, ufd, key, value);
737free_value:
738 kfree(value);
739free_key:
740 kfree(key);
741err_put:
742 fdput(f);
743 return err;
744}
745
746#define BPF_MAP_DELETE_ELEM_LAST_FIELD key
747
748static int map_delete_elem(union bpf_attr *attr)
749{
750 void __user *ukey = u64_to_user_ptr(attr->key);
751 int ufd = attr->map_fd;
752 struct bpf_map *map;
753 struct fd f;
754 void *key;
755 int err;
756
757 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
758 return -EINVAL;
759
760 f = fdget(ufd);
761 map = __bpf_map_get(f);
762 if (IS_ERR(map))
763 return PTR_ERR(map);
764
765 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
766 err = -EPERM;
767 goto err_put;
768 }
769
770 key = memdup_user(ukey, map->key_size);
771 if (IS_ERR(key)) {
772 err = PTR_ERR(key);
773 goto err_put;
774 }
775
776 if (bpf_map_is_dev_bound(map)) {
777 err = bpf_map_offload_delete_elem(map, key);
778 goto out;
779 }
780
781 preempt_disable();
782 __this_cpu_inc(bpf_prog_active);
783 rcu_read_lock();
784 err = map->ops->map_delete_elem(map, key);
785 rcu_read_unlock();
786 __this_cpu_dec(bpf_prog_active);
787 preempt_enable();
788out:
789 if (!err)
790 trace_bpf_map_delete_elem(map, ufd, key);
791 kfree(key);
792err_put:
793 fdput(f);
794 return err;
795}
796
797/* last field in 'union bpf_attr' used by this command */
798#define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
799
800static int map_get_next_key(union bpf_attr *attr)
801{
802 void __user *ukey = u64_to_user_ptr(attr->key);
803 void __user *unext_key = u64_to_user_ptr(attr->next_key);
804 int ufd = attr->map_fd;
805 struct bpf_map *map;
806 void *key, *next_key;
807 struct fd f;
808 int err;
809
810 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
811 return -EINVAL;
812
813 f = fdget(ufd);
814 map = __bpf_map_get(f);
815 if (IS_ERR(map))
816 return PTR_ERR(map);
817
818 if (!(f.file->f_mode & FMODE_CAN_READ)) {
819 err = -EPERM;
820 goto err_put;
821 }
822
823 if (ukey) {
824 key = memdup_user(ukey, map->key_size);
825 if (IS_ERR(key)) {
826 err = PTR_ERR(key);
827 goto err_put;
828 }
829 } else {
830 key = NULL;
831 }
832
833 err = -ENOMEM;
834 next_key = kmalloc(map->key_size, GFP_USER);
835 if (!next_key)
836 goto free_key;
837
838 if (bpf_map_is_dev_bound(map)) {
839 err = bpf_map_offload_get_next_key(map, key, next_key);
840 goto out;
841 }
842
843 rcu_read_lock();
844 err = map->ops->map_get_next_key(map, key, next_key);
845 rcu_read_unlock();
846out:
847 if (err)
848 goto free_next_key;
849
850 err = -EFAULT;
851 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
852 goto free_next_key;
853
854 trace_bpf_map_next_key(map, ufd, key, next_key);
855 err = 0;
856
857free_next_key:
858 kfree(next_key);
859free_key:
860 kfree(key);
861err_put:
862 fdput(f);
863 return err;
864}
865
866static const struct bpf_prog_ops * const bpf_prog_types[] = {
867#define BPF_PROG_TYPE(_id, _name) \
868 [_id] = & _name ## _prog_ops,
869#define BPF_MAP_TYPE(_id, _ops)
870#include <linux/bpf_types.h>
871#undef BPF_PROG_TYPE
872#undef BPF_MAP_TYPE
873};
874
875static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
876{
877 const struct bpf_prog_ops *ops;
878
879 if (type >= ARRAY_SIZE(bpf_prog_types))
880 return -EINVAL;
881 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
882 ops = bpf_prog_types[type];
883 if (!ops)
884 return -EINVAL;
885
886 if (!bpf_prog_is_dev_bound(prog->aux))
887 prog->aux->ops = ops;
888 else
889 prog->aux->ops = &bpf_offload_prog_ops;
890 prog->type = type;
891 return 0;
892}
893
894/* drop refcnt on maps used by eBPF program and free auxilary data */
895static void free_used_maps(struct bpf_prog_aux *aux)
896{
897 int i;
898
899 for (i = 0; i < aux->used_map_cnt; i++)
900 bpf_map_put(aux->used_maps[i]);
901
902 kfree(aux->used_maps);
903}
904
905int __bpf_prog_charge(struct user_struct *user, u32 pages)
906{
907 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
908 unsigned long user_bufs;
909
910 if (user) {
911 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
912 if (user_bufs > memlock_limit) {
913 atomic_long_sub(pages, &user->locked_vm);
914 return -EPERM;
915 }
916 }
917
918 return 0;
919}
920
921void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
922{
923 if (user)
924 atomic_long_sub(pages, &user->locked_vm);
925}
926
927static int bpf_prog_charge_memlock(struct bpf_prog *prog)
928{
929 struct user_struct *user = get_current_user();
930 int ret;
931
932 ret = __bpf_prog_charge(user, prog->pages);
933 if (ret) {
934 free_uid(user);
935 return ret;
936 }
937
938 prog->aux->user = user;
939 return 0;
940}
941
942static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
943{
944 struct user_struct *user = prog->aux->user;
945
946 __bpf_prog_uncharge(user, prog->pages);
947 free_uid(user);
948}
949
950static int bpf_prog_alloc_id(struct bpf_prog *prog)
951{
952 int id;
953
954 idr_preload(GFP_KERNEL);
955 spin_lock_bh(&prog_idr_lock);
956 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
957 if (id > 0)
958 prog->aux->id = id;
959 spin_unlock_bh(&prog_idr_lock);
960 idr_preload_end();
961
962 /* id is in [1, INT_MAX) */
963 if (WARN_ON_ONCE(!id))
964 return -ENOSPC;
965
966 return id > 0 ? 0 : id;
967}
968
969void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
970{
971 /* cBPF to eBPF migrations are currently not in the idr store.
972 * Offloaded programs are removed from the store when their device
973 * disappears - even if someone grabs an fd to them they are unusable,
974 * simply waiting for refcnt to drop to be freed.
975 */
976 if (!prog->aux->id)
977 return;
978
979 if (do_idr_lock)
980 spin_lock_bh(&prog_idr_lock);
981 else
982 __acquire(&prog_idr_lock);
983
984 idr_remove(&prog_idr, prog->aux->id);
985 prog->aux->id = 0;
986
987 if (do_idr_lock)
988 spin_unlock_bh(&prog_idr_lock);
989 else
990 __release(&prog_idr_lock);
991}
992
993static void __bpf_prog_put_rcu(struct rcu_head *rcu)
994{
995 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
996
997 free_used_maps(aux);
998 bpf_prog_uncharge_memlock(aux->prog);
999 security_bpf_prog_free(aux);
1000 bpf_prog_free(aux->prog);
1001}
1002
1003static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1004{
1005 if (atomic_dec_and_test(&prog->aux->refcnt)) {
1006 int i;
1007
1008 trace_bpf_prog_put_rcu(prog);
1009 /* bpf_prog_free_id() must be called first */
1010 bpf_prog_free_id(prog, do_idr_lock);
1011
1012 for (i = 0; i < prog->aux->func_cnt; i++)
1013 bpf_prog_kallsyms_del(prog->aux->func[i]);
1014 bpf_prog_kallsyms_del(prog);
1015
1016 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1017 }
1018}
1019
1020void bpf_prog_put(struct bpf_prog *prog)
1021{
1022 __bpf_prog_put(prog, true);
1023}
1024EXPORT_SYMBOL_GPL(bpf_prog_put);
1025
1026static int bpf_prog_release(struct inode *inode, struct file *filp)
1027{
1028 struct bpf_prog *prog = filp->private_data;
1029
1030 bpf_prog_put(prog);
1031 return 0;
1032}
1033
1034#ifdef CONFIG_PROC_FS
1035static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1036{
1037 const struct bpf_prog *prog = filp->private_data;
1038 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1039
1040 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1041 seq_printf(m,
1042 "prog_type:\t%u\n"
1043 "prog_jited:\t%u\n"
1044 "prog_tag:\t%s\n"
1045 "memlock:\t%llu\n",
1046 prog->type,
1047 prog->jited,
1048 prog_tag,
1049 prog->pages * 1ULL << PAGE_SHIFT);
1050}
1051#endif
1052
1053const struct file_operations bpf_prog_fops = {
1054#ifdef CONFIG_PROC_FS
1055 .show_fdinfo = bpf_prog_show_fdinfo,
1056#endif
1057 .release = bpf_prog_release,
1058 .read = bpf_dummy_read,
1059 .write = bpf_dummy_write,
1060};
1061
1062int bpf_prog_new_fd(struct bpf_prog *prog)
1063{
1064 int ret;
1065
1066 ret = security_bpf_prog(prog);
1067 if (ret < 0)
1068 return ret;
1069
1070 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1071 O_RDWR | O_CLOEXEC);
1072}
1073
1074static struct bpf_prog *____bpf_prog_get(struct fd f)
1075{
1076 if (!f.file)
1077 return ERR_PTR(-EBADF);
1078 if (f.file->f_op != &bpf_prog_fops) {
1079 fdput(f);
1080 return ERR_PTR(-EINVAL);
1081 }
1082
1083 return f.file->private_data;
1084}
1085
1086struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1087{
1088 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1089 atomic_sub(i, &prog->aux->refcnt);
1090 return ERR_PTR(-EBUSY);
1091 }
1092 return prog;
1093}
1094EXPORT_SYMBOL_GPL(bpf_prog_add);
1095
1096void bpf_prog_sub(struct bpf_prog *prog, int i)
1097{
1098 /* Only to be used for undoing previous bpf_prog_add() in some
1099 * error path. We still know that another entity in our call
1100 * path holds a reference to the program, thus atomic_sub() can
1101 * be safely used in such cases!
1102 */
1103 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1104}
1105EXPORT_SYMBOL_GPL(bpf_prog_sub);
1106
1107struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1108{
1109 return bpf_prog_add(prog, 1);
1110}
1111EXPORT_SYMBOL_GPL(bpf_prog_inc);
1112
1113/* prog_idr_lock should have been held */
1114struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1115{
1116 int refold;
1117
1118 refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
1119
1120 if (refold >= BPF_MAX_REFCNT) {
1121 __bpf_prog_put(prog, false);
1122 return ERR_PTR(-EBUSY);
1123 }
1124
1125 if (!refold)
1126 return ERR_PTR(-ENOENT);
1127
1128 return prog;
1129}
1130EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1131
1132bool bpf_prog_get_ok(struct bpf_prog *prog,
1133 enum bpf_prog_type *attach_type, bool attach_drv)
1134{
1135 /* not an attachment, just a refcount inc, always allow */
1136 if (!attach_type)
1137 return true;
1138
1139 if (prog->type != *attach_type)
1140 return false;
1141 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1142 return false;
1143
1144 return true;
1145}
1146
1147static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1148 bool attach_drv)
1149{
1150 struct fd f = fdget(ufd);
1151 struct bpf_prog *prog;
1152
1153 prog = ____bpf_prog_get(f);
1154 if (IS_ERR(prog))
1155 return prog;
1156 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1157 prog = ERR_PTR(-EINVAL);
1158 goto out;
1159 }
1160
1161 prog = bpf_prog_inc(prog);
1162out:
1163 fdput(f);
1164 return prog;
1165}
1166
1167struct bpf_prog *bpf_prog_get(u32 ufd)
1168{
1169 return __bpf_prog_get(ufd, NULL, false);
1170}
1171
1172struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1173 bool attach_drv)
1174{
1175 struct bpf_prog *prog = __bpf_prog_get(ufd, &type, attach_drv);
1176
1177 if (!IS_ERR(prog))
1178 trace_bpf_prog_get_type(prog);
1179 return prog;
1180}
1181EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1182
1183/* Initially all BPF programs could be loaded w/o specifying
1184 * expected_attach_type. Later for some of them specifying expected_attach_type
1185 * at load time became required so that program could be validated properly.
1186 * Programs of types that are allowed to be loaded both w/ and w/o (for
1187 * backward compatibility) expected_attach_type, should have the default attach
1188 * type assigned to expected_attach_type for the latter case, so that it can be
1189 * validated later at attach time.
1190 *
1191 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1192 * prog type requires it but has some attach types that have to be backward
1193 * compatible.
1194 */
1195static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1196{
1197 switch (attr->prog_type) {
1198 case BPF_PROG_TYPE_CGROUP_SOCK:
1199 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1200 * exist so checking for non-zero is the way to go here.
1201 */
1202 if (!attr->expected_attach_type)
1203 attr->expected_attach_type =
1204 BPF_CGROUP_INET_SOCK_CREATE;
1205 break;
1206 }
1207}
1208
1209static int
1210bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1211 enum bpf_attach_type expected_attach_type)
1212{
1213 switch (prog_type) {
1214 case BPF_PROG_TYPE_CGROUP_SOCK:
1215 switch (expected_attach_type) {
1216 case BPF_CGROUP_INET_SOCK_CREATE:
1217 case BPF_CGROUP_INET4_POST_BIND:
1218 case BPF_CGROUP_INET6_POST_BIND:
1219 return 0;
1220 default:
1221 return -EINVAL;
1222 }
1223 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1224 switch (expected_attach_type) {
1225 case BPF_CGROUP_INET4_BIND:
1226 case BPF_CGROUP_INET6_BIND:
1227 case BPF_CGROUP_INET4_CONNECT:
1228 case BPF_CGROUP_INET6_CONNECT:
1229 return 0;
1230 default:
1231 return -EINVAL;
1232 }
1233 default:
1234 return 0;
1235 }
1236}
1237
1238/* last field in 'union bpf_attr' used by this command */
1239#define BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1240
1241static int bpf_prog_load(union bpf_attr *attr)
1242{
1243 enum bpf_prog_type type = attr->prog_type;
1244 struct bpf_prog *prog;
1245 int err;
1246 char license[128];
1247 bool is_gpl;
1248
1249 if (CHECK_ATTR(BPF_PROG_LOAD))
1250 return -EINVAL;
1251
1252 if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1253 return -EINVAL;
1254
1255 /* copy eBPF program license from user space */
1256 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1257 sizeof(license) - 1) < 0)
1258 return -EFAULT;
1259 license[sizeof(license) - 1] = 0;
1260
1261 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1262 is_gpl = license_is_gpl_compatible(license);
1263
1264 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1265 return -E2BIG;
1266
1267 if (type == BPF_PROG_TYPE_KPROBE &&
1268 attr->kern_version != LINUX_VERSION_CODE)
1269 return -EINVAL;
1270
1271 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1272 type != BPF_PROG_TYPE_CGROUP_SKB &&
1273 !capable(CAP_SYS_ADMIN))
1274 return -EPERM;
1275
1276 bpf_prog_load_fixup_attach_type(attr);
1277 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1278 return -EINVAL;
1279
1280 /* plain bpf_prog allocation */
1281 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1282 if (!prog)
1283 return -ENOMEM;
1284
1285 prog->expected_attach_type = attr->expected_attach_type;
1286
1287 prog->aux->offload_requested = !!attr->prog_ifindex;
1288
1289 err = security_bpf_prog_alloc(prog->aux);
1290 if (err)
1291 goto free_prog_nouncharge;
1292
1293 err = bpf_prog_charge_memlock(prog);
1294 if (err)
1295 goto free_prog_sec;
1296
1297 prog->len = attr->insn_cnt;
1298
1299 err = -EFAULT;
1300 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1301 bpf_prog_insn_size(prog)) != 0)
1302 goto free_prog;
1303
1304 prog->orig_prog = NULL;
1305 prog->jited = 0;
1306
1307 atomic_set(&prog->aux->refcnt, 1);
1308 prog->gpl_compatible = is_gpl ? 1 : 0;
1309
1310 if (bpf_prog_is_dev_bound(prog->aux)) {
1311 err = bpf_prog_offload_init(prog, attr);
1312 if (err)
1313 goto free_prog;
1314 }
1315
1316 /* find program type: socket_filter vs tracing_filter */
1317 err = find_prog_type(type, prog);
1318 if (err < 0)
1319 goto free_prog;
1320
1321 prog->aux->load_time = ktime_get_boot_ns();
1322 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1323 if (err)
1324 goto free_prog;
1325
1326 /* run eBPF verifier */
1327 err = bpf_check(&prog, attr);
1328 if (err < 0)
1329 goto free_used_maps;
1330
1331 /* eBPF program is ready to be JITed */
1332 if (!prog->bpf_func)
1333 prog = bpf_prog_select_runtime(prog, &err);
1334 if (err < 0)
1335 goto free_used_maps;
1336
1337 err = bpf_prog_alloc_id(prog);
1338 if (err)
1339 goto free_used_maps;
1340
1341 err = bpf_prog_new_fd(prog);
1342 if (err < 0) {
1343 /* failed to allocate fd.
1344 * bpf_prog_put() is needed because the above
1345 * bpf_prog_alloc_id() has published the prog
1346 * to the userspace and the userspace may
1347 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1348 */
1349 bpf_prog_put(prog);
1350 return err;
1351 }
1352
1353 bpf_prog_kallsyms_add(prog);
1354 trace_bpf_prog_load(prog, err);
1355 return err;
1356
1357free_used_maps:
1358 free_used_maps(prog->aux);
1359free_prog:
1360 bpf_prog_uncharge_memlock(prog);
1361free_prog_sec:
1362 security_bpf_prog_free(prog->aux);
1363free_prog_nouncharge:
1364 bpf_prog_free(prog);
1365 return err;
1366}
1367
1368#define BPF_OBJ_LAST_FIELD file_flags
1369
1370static int bpf_obj_pin(const union bpf_attr *attr)
1371{
1372 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1373 return -EINVAL;
1374
1375 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1376}
1377
1378static int bpf_obj_get(const union bpf_attr *attr)
1379{
1380 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1381 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1382 return -EINVAL;
1383
1384 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1385 attr->file_flags);
1386}
1387
1388struct bpf_raw_tracepoint {
1389 struct bpf_raw_event_map *btp;
1390 struct bpf_prog *prog;
1391};
1392
1393static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1394{
1395 struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1396
1397 if (raw_tp->prog) {
1398 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1399 bpf_prog_put(raw_tp->prog);
1400 }
1401 kfree(raw_tp);
1402 return 0;
1403}
1404
1405static const struct file_operations bpf_raw_tp_fops = {
1406 .release = bpf_raw_tracepoint_release,
1407 .read = bpf_dummy_read,
1408 .write = bpf_dummy_write,
1409};
1410
1411#define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1412
1413static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1414{
1415 struct bpf_raw_tracepoint *raw_tp;
1416 struct bpf_raw_event_map *btp;
1417 struct bpf_prog *prog;
1418 char tp_name[128];
1419 int tp_fd, err;
1420
1421 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1422 sizeof(tp_name) - 1) < 0)
1423 return -EFAULT;
1424 tp_name[sizeof(tp_name) - 1] = 0;
1425
1426 btp = bpf_find_raw_tracepoint(tp_name);
1427 if (!btp)
1428 return -ENOENT;
1429
1430 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1431 if (!raw_tp)
1432 return -ENOMEM;
1433 raw_tp->btp = btp;
1434
1435 prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1436 BPF_PROG_TYPE_RAW_TRACEPOINT);
1437 if (IS_ERR(prog)) {
1438 err = PTR_ERR(prog);
1439 goto out_free_tp;
1440 }
1441
1442 err = bpf_probe_register(raw_tp->btp, prog);
1443 if (err)
1444 goto out_put_prog;
1445
1446 raw_tp->prog = prog;
1447 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1448 O_CLOEXEC);
1449 if (tp_fd < 0) {
1450 bpf_probe_unregister(raw_tp->btp, prog);
1451 err = tp_fd;
1452 goto out_put_prog;
1453 }
1454 return tp_fd;
1455
1456out_put_prog:
1457 bpf_prog_put(prog);
1458out_free_tp:
1459 kfree(raw_tp);
1460 return err;
1461}
1462
1463#ifdef CONFIG_CGROUP_BPF
1464
1465static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1466 enum bpf_attach_type attach_type)
1467{
1468 switch (prog->type) {
1469 case BPF_PROG_TYPE_CGROUP_SOCK:
1470 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1471 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1472 default:
1473 return 0;
1474 }
1475}
1476
1477#define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1478
1479static int sockmap_get_from_fd(const union bpf_attr *attr,
1480 int type, bool attach)
1481{
1482 struct bpf_prog *prog = NULL;
1483 int ufd = attr->target_fd;
1484 struct bpf_map *map;
1485 struct fd f;
1486 int err;
1487
1488 f = fdget(ufd);
1489 map = __bpf_map_get(f);
1490 if (IS_ERR(map))
1491 return PTR_ERR(map);
1492
1493 if (attach) {
1494 prog = bpf_prog_get_type(attr->attach_bpf_fd, type);
1495 if (IS_ERR(prog)) {
1496 fdput(f);
1497 return PTR_ERR(prog);
1498 }
1499 }
1500
1501 err = sock_map_prog(map, prog, attr->attach_type);
1502 if (err) {
1503 fdput(f);
1504 if (prog)
1505 bpf_prog_put(prog);
1506 return err;
1507 }
1508
1509 fdput(f);
1510 return 0;
1511}
1512
1513#define BPF_F_ATTACH_MASK \
1514 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1515
1516static int bpf_prog_attach(const union bpf_attr *attr)
1517{
1518 enum bpf_prog_type ptype;
1519 struct bpf_prog *prog;
1520 struct cgroup *cgrp;
1521 int ret;
1522
1523 if (!capable(CAP_NET_ADMIN))
1524 return -EPERM;
1525
1526 if (CHECK_ATTR(BPF_PROG_ATTACH))
1527 return -EINVAL;
1528
1529 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1530 return -EINVAL;
1531
1532 switch (attr->attach_type) {
1533 case BPF_CGROUP_INET_INGRESS:
1534 case BPF_CGROUP_INET_EGRESS:
1535 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1536 break;
1537 case BPF_CGROUP_INET_SOCK_CREATE:
1538 case BPF_CGROUP_INET4_POST_BIND:
1539 case BPF_CGROUP_INET6_POST_BIND:
1540 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1541 break;
1542 case BPF_CGROUP_INET4_BIND:
1543 case BPF_CGROUP_INET6_BIND:
1544 case BPF_CGROUP_INET4_CONNECT:
1545 case BPF_CGROUP_INET6_CONNECT:
1546 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1547 break;
1548 case BPF_CGROUP_SOCK_OPS:
1549 ptype = BPF_PROG_TYPE_SOCK_OPS;
1550 break;
1551 case BPF_CGROUP_DEVICE:
1552 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1553 break;
1554 case BPF_SK_MSG_VERDICT:
1555 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, true);
1556 case BPF_SK_SKB_STREAM_PARSER:
1557 case BPF_SK_SKB_STREAM_VERDICT:
1558 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, true);
1559 default:
1560 return -EINVAL;
1561 }
1562
1563 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1564 if (IS_ERR(prog))
1565 return PTR_ERR(prog);
1566
1567 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1568 bpf_prog_put(prog);
1569 return -EINVAL;
1570 }
1571
1572 cgrp = cgroup_get_from_fd(attr->target_fd);
1573 if (IS_ERR(cgrp)) {
1574 bpf_prog_put(prog);
1575 return PTR_ERR(cgrp);
1576 }
1577
1578 ret = cgroup_bpf_attach(cgrp, prog, attr->attach_type,
1579 attr->attach_flags);
1580 if (ret)
1581 bpf_prog_put(prog);
1582 cgroup_put(cgrp);
1583
1584 return ret;
1585}
1586
1587#define BPF_PROG_DETACH_LAST_FIELD attach_type
1588
1589static int bpf_prog_detach(const union bpf_attr *attr)
1590{
1591 enum bpf_prog_type ptype;
1592 struct bpf_prog *prog;
1593 struct cgroup *cgrp;
1594 int ret;
1595
1596 if (!capable(CAP_NET_ADMIN))
1597 return -EPERM;
1598
1599 if (CHECK_ATTR(BPF_PROG_DETACH))
1600 return -EINVAL;
1601
1602 switch (attr->attach_type) {
1603 case BPF_CGROUP_INET_INGRESS:
1604 case BPF_CGROUP_INET_EGRESS:
1605 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1606 break;
1607 case BPF_CGROUP_INET_SOCK_CREATE:
1608 case BPF_CGROUP_INET4_POST_BIND:
1609 case BPF_CGROUP_INET6_POST_BIND:
1610 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1611 break;
1612 case BPF_CGROUP_INET4_BIND:
1613 case BPF_CGROUP_INET6_BIND:
1614 case BPF_CGROUP_INET4_CONNECT:
1615 case BPF_CGROUP_INET6_CONNECT:
1616 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1617 break;
1618 case BPF_CGROUP_SOCK_OPS:
1619 ptype = BPF_PROG_TYPE_SOCK_OPS;
1620 break;
1621 case BPF_CGROUP_DEVICE:
1622 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1623 break;
1624 case BPF_SK_MSG_VERDICT:
1625 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, false);
1626 case BPF_SK_SKB_STREAM_PARSER:
1627 case BPF_SK_SKB_STREAM_VERDICT:
1628 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, false);
1629 default:
1630 return -EINVAL;
1631 }
1632
1633 cgrp = cgroup_get_from_fd(attr->target_fd);
1634 if (IS_ERR(cgrp))
1635 return PTR_ERR(cgrp);
1636
1637 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1638 if (IS_ERR(prog))
1639 prog = NULL;
1640
1641 ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type, 0);
1642 if (prog)
1643 bpf_prog_put(prog);
1644 cgroup_put(cgrp);
1645 return ret;
1646}
1647
1648#define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1649
1650static int bpf_prog_query(const union bpf_attr *attr,
1651 union bpf_attr __user *uattr)
1652{
1653 struct cgroup *cgrp;
1654 int ret;
1655
1656 if (!capable(CAP_NET_ADMIN))
1657 return -EPERM;
1658 if (CHECK_ATTR(BPF_PROG_QUERY))
1659 return -EINVAL;
1660 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1661 return -EINVAL;
1662
1663 switch (attr->query.attach_type) {
1664 case BPF_CGROUP_INET_INGRESS:
1665 case BPF_CGROUP_INET_EGRESS:
1666 case BPF_CGROUP_INET_SOCK_CREATE:
1667 case BPF_CGROUP_INET4_BIND:
1668 case BPF_CGROUP_INET6_BIND:
1669 case BPF_CGROUP_INET4_POST_BIND:
1670 case BPF_CGROUP_INET6_POST_BIND:
1671 case BPF_CGROUP_INET4_CONNECT:
1672 case BPF_CGROUP_INET6_CONNECT:
1673 case BPF_CGROUP_SOCK_OPS:
1674 case BPF_CGROUP_DEVICE:
1675 break;
1676 default:
1677 return -EINVAL;
1678 }
1679 cgrp = cgroup_get_from_fd(attr->query.target_fd);
1680 if (IS_ERR(cgrp))
1681 return PTR_ERR(cgrp);
1682 ret = cgroup_bpf_query(cgrp, attr, uattr);
1683 cgroup_put(cgrp);
1684 return ret;
1685}
1686#endif /* CONFIG_CGROUP_BPF */
1687
1688#define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1689
1690static int bpf_prog_test_run(const union bpf_attr *attr,
1691 union bpf_attr __user *uattr)
1692{
1693 struct bpf_prog *prog;
1694 int ret = -ENOTSUPP;
1695
1696 if (!capable(CAP_SYS_ADMIN))
1697 return -EPERM;
1698 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1699 return -EINVAL;
1700
1701 prog = bpf_prog_get(attr->test.prog_fd);
1702 if (IS_ERR(prog))
1703 return PTR_ERR(prog);
1704
1705 if (prog->aux->ops->test_run)
1706 ret = prog->aux->ops->test_run(prog, attr, uattr);
1707
1708 bpf_prog_put(prog);
1709 return ret;
1710}
1711
1712#define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1713
1714static int bpf_obj_get_next_id(const union bpf_attr *attr,
1715 union bpf_attr __user *uattr,
1716 struct idr *idr,
1717 spinlock_t *lock)
1718{
1719 u32 next_id = attr->start_id;
1720 int err = 0;
1721
1722 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1723 return -EINVAL;
1724
1725 if (!capable(CAP_SYS_ADMIN))
1726 return -EPERM;
1727
1728 next_id++;
1729 spin_lock_bh(lock);
1730 if (!idr_get_next(idr, &next_id))
1731 err = -ENOENT;
1732 spin_unlock_bh(lock);
1733
1734 if (!err)
1735 err = put_user(next_id, &uattr->next_id);
1736
1737 return err;
1738}
1739
1740#define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1741
1742static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1743{
1744 struct bpf_prog *prog;
1745 u32 id = attr->prog_id;
1746 int fd;
1747
1748 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1749 return -EINVAL;
1750
1751 if (!capable(CAP_SYS_ADMIN))
1752 return -EPERM;
1753
1754 spin_lock_bh(&prog_idr_lock);
1755 prog = idr_find(&prog_idr, id);
1756 if (prog)
1757 prog = bpf_prog_inc_not_zero(prog);
1758 else
1759 prog = ERR_PTR(-ENOENT);
1760 spin_unlock_bh(&prog_idr_lock);
1761
1762 if (IS_ERR(prog))
1763 return PTR_ERR(prog);
1764
1765 fd = bpf_prog_new_fd(prog);
1766 if (fd < 0)
1767 bpf_prog_put(prog);
1768
1769 return fd;
1770}
1771
1772#define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1773
1774static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1775{
1776 struct bpf_map *map;
1777 u32 id = attr->map_id;
1778 int f_flags;
1779 int fd;
1780
1781 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1782 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1783 return -EINVAL;
1784
1785 if (!capable(CAP_SYS_ADMIN))
1786 return -EPERM;
1787
1788 f_flags = bpf_get_file_flag(attr->open_flags);
1789 if (f_flags < 0)
1790 return f_flags;
1791
1792 spin_lock_bh(&map_idr_lock);
1793 map = idr_find(&map_idr, id);
1794 if (map)
1795 map = bpf_map_inc_not_zero(map, true);
1796 else
1797 map = ERR_PTR(-ENOENT);
1798 spin_unlock_bh(&map_idr_lock);
1799
1800 if (IS_ERR(map))
1801 return PTR_ERR(map);
1802
1803 fd = bpf_map_new_fd(map, f_flags);
1804 if (fd < 0)
1805 bpf_map_put(map);
1806
1807 return fd;
1808}
1809
1810static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1811 unsigned long addr)
1812{
1813 int i;
1814
1815 for (i = 0; i < prog->aux->used_map_cnt; i++)
1816 if (prog->aux->used_maps[i] == (void *)addr)
1817 return prog->aux->used_maps[i];
1818 return NULL;
1819}
1820
1821static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1822{
1823 const struct bpf_map *map;
1824 struct bpf_insn *insns;
1825 u64 imm;
1826 int i;
1827
1828 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1829 GFP_USER);
1830 if (!insns)
1831 return insns;
1832
1833 for (i = 0; i < prog->len; i++) {
1834 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1835 insns[i].code = BPF_JMP | BPF_CALL;
1836 insns[i].imm = BPF_FUNC_tail_call;
1837 /* fall-through */
1838 }
1839 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1840 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1841 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1842 insns[i].code = BPF_JMP | BPF_CALL;
1843 if (!bpf_dump_raw_ok())
1844 insns[i].imm = 0;
1845 continue;
1846 }
1847
1848 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1849 continue;
1850
1851 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1852 map = bpf_map_from_imm(prog, imm);
1853 if (map) {
1854 insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1855 insns[i].imm = map->id;
1856 insns[i + 1].imm = 0;
1857 continue;
1858 }
1859
1860 if (!bpf_dump_raw_ok() &&
1861 imm == (unsigned long)prog->aux) {
1862 insns[i].imm = 0;
1863 insns[i + 1].imm = 0;
1864 continue;
1865 }
1866 }
1867
1868 return insns;
1869}
1870
1871static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1872 const union bpf_attr *attr,
1873 union bpf_attr __user *uattr)
1874{
1875 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1876 struct bpf_prog_info info = {};
1877 u32 info_len = attr->info.info_len;
1878 char __user *uinsns;
1879 u32 ulen;
1880 int err;
1881
1882 err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1883 if (err)
1884 return err;
1885 info_len = min_t(u32, sizeof(info), info_len);
1886
1887 if (copy_from_user(&info, uinfo, info_len))
1888 return -EFAULT;
1889
1890 info.type = prog->type;
1891 info.id = prog->aux->id;
1892 info.load_time = prog->aux->load_time;
1893 info.created_by_uid = from_kuid_munged(current_user_ns(),
1894 prog->aux->user->uid);
1895
1896 memcpy(info.tag, prog->tag, sizeof(prog->tag));
1897 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1898
1899 ulen = info.nr_map_ids;
1900 info.nr_map_ids = prog->aux->used_map_cnt;
1901 ulen = min_t(u32, info.nr_map_ids, ulen);
1902 if (ulen) {
1903 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1904 u32 i;
1905
1906 for (i = 0; i < ulen; i++)
1907 if (put_user(prog->aux->used_maps[i]->id,
1908 &user_map_ids[i]))
1909 return -EFAULT;
1910 }
1911
1912 if (!capable(CAP_SYS_ADMIN)) {
1913 info.jited_prog_len = 0;
1914 info.xlated_prog_len = 0;
1915 goto done;
1916 }
1917
1918 ulen = info.xlated_prog_len;
1919 info.xlated_prog_len = bpf_prog_insn_size(prog);
1920 if (info.xlated_prog_len && ulen) {
1921 struct bpf_insn *insns_sanitized;
1922 bool fault;
1923
1924 if (prog->blinded && !bpf_dump_raw_ok()) {
1925 info.xlated_prog_insns = 0;
1926 goto done;
1927 }
1928 insns_sanitized = bpf_insn_prepare_dump(prog);
1929 if (!insns_sanitized)
1930 return -ENOMEM;
1931 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1932 ulen = min_t(u32, info.xlated_prog_len, ulen);
1933 fault = copy_to_user(uinsns, insns_sanitized, ulen);
1934 kfree(insns_sanitized);
1935 if (fault)
1936 return -EFAULT;
1937 }
1938
1939 if (bpf_prog_is_dev_bound(prog->aux)) {
1940 err = bpf_prog_offload_info_fill(&info, prog);
1941 if (err)
1942 return err;
1943 goto done;
1944 }
1945
1946 /* NOTE: the following code is supposed to be skipped for offload.
1947 * bpf_prog_offload_info_fill() is the place to fill similar fields
1948 * for offload.
1949 */
1950 ulen = info.jited_prog_len;
1951 info.jited_prog_len = prog->jited_len;
1952 if (info.jited_prog_len && ulen) {
1953 if (bpf_dump_raw_ok()) {
1954 uinsns = u64_to_user_ptr(info.jited_prog_insns);
1955 ulen = min_t(u32, info.jited_prog_len, ulen);
1956 if (copy_to_user(uinsns, prog->bpf_func, ulen))
1957 return -EFAULT;
1958 } else {
1959 info.jited_prog_insns = 0;
1960 }
1961 }
1962
1963done:
1964 if (copy_to_user(uinfo, &info, info_len) ||
1965 put_user(info_len, &uattr->info.info_len))
1966 return -EFAULT;
1967
1968 return 0;
1969}
1970
1971static int bpf_map_get_info_by_fd(struct bpf_map *map,
1972 const union bpf_attr *attr,
1973 union bpf_attr __user *uattr)
1974{
1975 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1976 struct bpf_map_info info = {};
1977 u32 info_len = attr->info.info_len;
1978 int err;
1979
1980 err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1981 if (err)
1982 return err;
1983 info_len = min_t(u32, sizeof(info), info_len);
1984
1985 info.type = map->map_type;
1986 info.id = map->id;
1987 info.key_size = map->key_size;
1988 info.value_size = map->value_size;
1989 info.max_entries = map->max_entries;
1990 info.map_flags = map->map_flags;
1991 memcpy(info.name, map->name, sizeof(map->name));
1992
1993 if (bpf_map_is_dev_bound(map)) {
1994 err = bpf_map_offload_info_fill(&info, map);
1995 if (err)
1996 return err;
1997 }
1998
1999 if (copy_to_user(uinfo, &info, info_len) ||
2000 put_user(info_len, &uattr->info.info_len))
2001 return -EFAULT;
2002
2003 return 0;
2004}
2005
2006#define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2007
2008static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2009 union bpf_attr __user *uattr)
2010{
2011 int ufd = attr->info.bpf_fd;
2012 struct fd f;
2013 int err;
2014
2015 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2016 return -EINVAL;
2017
2018 f = fdget(ufd);
2019 if (!f.file)
2020 return -EBADFD;
2021
2022 if (f.file->f_op == &bpf_prog_fops)
2023 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2024 uattr);
2025 else if (f.file->f_op == &bpf_map_fops)
2026 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2027 uattr);
2028 else
2029 err = -EINVAL;
2030
2031 fdput(f);
2032 return err;
2033}
2034
2035SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2036{
2037 union bpf_attr attr = {};
2038 int err;
2039
2040 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2041 return -EPERM;
2042
2043 err = check_uarg_tail_zero(uattr, sizeof(attr), size);
2044 if (err)
2045 return err;
2046 size = min_t(u32, size, sizeof(attr));
2047
2048 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2049 if (copy_from_user(&attr, uattr, size) != 0)
2050 return -EFAULT;
2051
2052 err = security_bpf(cmd, &attr, size);
2053 if (err < 0)
2054 return err;
2055
2056 switch (cmd) {
2057 case BPF_MAP_CREATE:
2058 err = map_create(&attr);
2059 break;
2060 case BPF_MAP_LOOKUP_ELEM:
2061 err = map_lookup_elem(&attr);
2062 break;
2063 case BPF_MAP_UPDATE_ELEM:
2064 err = map_update_elem(&attr);
2065 break;
2066 case BPF_MAP_DELETE_ELEM:
2067 err = map_delete_elem(&attr);
2068 break;
2069 case BPF_MAP_GET_NEXT_KEY:
2070 err = map_get_next_key(&attr);
2071 break;
2072 case BPF_PROG_LOAD:
2073 err = bpf_prog_load(&attr);
2074 break;
2075 case BPF_OBJ_PIN:
2076 err = bpf_obj_pin(&attr);
2077 break;
2078 case BPF_OBJ_GET:
2079 err = bpf_obj_get(&attr);
2080 break;
2081#ifdef CONFIG_CGROUP_BPF
2082 case BPF_PROG_ATTACH:
2083 err = bpf_prog_attach(&attr);
2084 break;
2085 case BPF_PROG_DETACH:
2086 err = bpf_prog_detach(&attr);
2087 break;
2088 case BPF_PROG_QUERY:
2089 err = bpf_prog_query(&attr, uattr);
2090 break;
2091#endif
2092 case BPF_PROG_TEST_RUN:
2093 err = bpf_prog_test_run(&attr, uattr);
2094 break;
2095 case BPF_PROG_GET_NEXT_ID:
2096 err = bpf_obj_get_next_id(&attr, uattr,
2097 &prog_idr, &prog_idr_lock);
2098 break;
2099 case BPF_MAP_GET_NEXT_ID:
2100 err = bpf_obj_get_next_id(&attr, uattr,
2101 &map_idr, &map_idr_lock);
2102 break;
2103 case BPF_PROG_GET_FD_BY_ID:
2104 err = bpf_prog_get_fd_by_id(&attr);
2105 break;
2106 case BPF_MAP_GET_FD_BY_ID:
2107 err = bpf_map_get_fd_by_id(&attr);
2108 break;
2109 case BPF_OBJ_GET_INFO_BY_FD:
2110 err = bpf_obj_get_info_by_fd(&attr, uattr);
2111 break;
2112 case BPF_RAW_TRACEPOINT_OPEN:
2113 err = bpf_raw_tracepoint_open(&attr);
2114 break;
2115 default:
2116 err = -EINVAL;
2117 break;
2118 }
2119
2120 return err;
2121}
1// SPDX-License-Identifier: GPL-2.0-only
2/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 */
4#include <linux/bpf.h>
5#include <linux/bpf-cgroup.h>
6#include <linux/bpf_trace.h>
7#include <linux/bpf_lirc.h>
8#include <linux/bpf_verifier.h>
9#include <linux/bsearch.h>
10#include <linux/btf.h>
11#include <linux/syscalls.h>
12#include <linux/slab.h>
13#include <linux/sched/signal.h>
14#include <linux/vmalloc.h>
15#include <linux/mmzone.h>
16#include <linux/anon_inodes.h>
17#include <linux/fdtable.h>
18#include <linux/file.h>
19#include <linux/fs.h>
20#include <linux/license.h>
21#include <linux/filter.h>
22#include <linux/kernel.h>
23#include <linux/idr.h>
24#include <linux/cred.h>
25#include <linux/timekeeping.h>
26#include <linux/ctype.h>
27#include <linux/nospec.h>
28#include <linux/audit.h>
29#include <uapi/linux/btf.h>
30#include <linux/pgtable.h>
31#include <linux/bpf_lsm.h>
32#include <linux/poll.h>
33#include <linux/sort.h>
34#include <linux/bpf-netns.h>
35#include <linux/rcupdate_trace.h>
36#include <linux/memcontrol.h>
37#include <linux/trace_events.h>
38#include <linux/tracepoint.h>
39
40#include <net/netfilter/nf_bpf_link.h>
41#include <net/netkit.h>
42#include <net/tcx.h>
43
44#define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
45 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
46 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
47#define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
48#define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
49#define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
50 IS_FD_HASH(map))
51
52#define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
53
54DEFINE_PER_CPU(int, bpf_prog_active);
55static DEFINE_IDR(prog_idr);
56static DEFINE_SPINLOCK(prog_idr_lock);
57static DEFINE_IDR(map_idr);
58static DEFINE_SPINLOCK(map_idr_lock);
59static DEFINE_IDR(link_idr);
60static DEFINE_SPINLOCK(link_idr_lock);
61
62int sysctl_unprivileged_bpf_disabled __read_mostly =
63 IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
64
65static const struct bpf_map_ops * const bpf_map_types[] = {
66#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
67#define BPF_MAP_TYPE(_id, _ops) \
68 [_id] = &_ops,
69#define BPF_LINK_TYPE(_id, _name)
70#include <linux/bpf_types.h>
71#undef BPF_PROG_TYPE
72#undef BPF_MAP_TYPE
73#undef BPF_LINK_TYPE
74};
75
76/*
77 * If we're handed a bigger struct than we know of, ensure all the unknown bits
78 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
79 * we don't know about yet.
80 *
81 * There is a ToCToU between this function call and the following
82 * copy_from_user() call. However, this is not a concern since this function is
83 * meant to be a future-proofing of bits.
84 */
85int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
86 size_t expected_size,
87 size_t actual_size)
88{
89 int res;
90
91 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
92 return -E2BIG;
93
94 if (actual_size <= expected_size)
95 return 0;
96
97 if (uaddr.is_kernel)
98 res = memchr_inv(uaddr.kernel + expected_size, 0,
99 actual_size - expected_size) == NULL;
100 else
101 res = check_zeroed_user(uaddr.user + expected_size,
102 actual_size - expected_size);
103 if (res < 0)
104 return res;
105 return res ? 0 : -E2BIG;
106}
107
108const struct bpf_map_ops bpf_map_offload_ops = {
109 .map_meta_equal = bpf_map_meta_equal,
110 .map_alloc = bpf_map_offload_map_alloc,
111 .map_free = bpf_map_offload_map_free,
112 .map_check_btf = map_check_no_btf,
113 .map_mem_usage = bpf_map_offload_map_mem_usage,
114};
115
116static void bpf_map_write_active_inc(struct bpf_map *map)
117{
118 atomic64_inc(&map->writecnt);
119}
120
121static void bpf_map_write_active_dec(struct bpf_map *map)
122{
123 atomic64_dec(&map->writecnt);
124}
125
126bool bpf_map_write_active(const struct bpf_map *map)
127{
128 return atomic64_read(&map->writecnt) != 0;
129}
130
131static u32 bpf_map_value_size(const struct bpf_map *map)
132{
133 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
134 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
135 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
136 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
137 return round_up(map->value_size, 8) * num_possible_cpus();
138 else if (IS_FD_MAP(map))
139 return sizeof(u32);
140 else
141 return map->value_size;
142}
143
144static void maybe_wait_bpf_programs(struct bpf_map *map)
145{
146 /* Wait for any running non-sleepable BPF programs to complete so that
147 * userspace, when we return to it, knows that all non-sleepable
148 * programs that could be running use the new map value. For sleepable
149 * BPF programs, synchronize_rcu_tasks_trace() should be used to wait
150 * for the completions of these programs, but considering the waiting
151 * time can be very long and userspace may think it will hang forever,
152 * so don't handle sleepable BPF programs now.
153 */
154 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
155 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
156 synchronize_rcu();
157}
158
159static void unpin_uptr_kaddr(void *kaddr)
160{
161 if (kaddr)
162 unpin_user_page(virt_to_page(kaddr));
163}
164
165static void __bpf_obj_unpin_uptrs(struct btf_record *rec, u32 cnt, void *obj)
166{
167 const struct btf_field *field;
168 void **uptr_addr;
169 int i;
170
171 for (i = 0, field = rec->fields; i < cnt; i++, field++) {
172 if (field->type != BPF_UPTR)
173 continue;
174
175 uptr_addr = obj + field->offset;
176 unpin_uptr_kaddr(*uptr_addr);
177 }
178}
179
180static void bpf_obj_unpin_uptrs(struct btf_record *rec, void *obj)
181{
182 if (!btf_record_has_field(rec, BPF_UPTR))
183 return;
184
185 __bpf_obj_unpin_uptrs(rec, rec->cnt, obj);
186}
187
188static int bpf_obj_pin_uptrs(struct btf_record *rec, void *obj)
189{
190 const struct btf_field *field;
191 const struct btf_type *t;
192 unsigned long start, end;
193 struct page *page;
194 void **uptr_addr;
195 int i, err;
196
197 if (!btf_record_has_field(rec, BPF_UPTR))
198 return 0;
199
200 for (i = 0, field = rec->fields; i < rec->cnt; i++, field++) {
201 if (field->type != BPF_UPTR)
202 continue;
203
204 uptr_addr = obj + field->offset;
205 start = *(unsigned long *)uptr_addr;
206 if (!start)
207 continue;
208
209 t = btf_type_by_id(field->kptr.btf, field->kptr.btf_id);
210 /* t->size was checked for zero before */
211 if (check_add_overflow(start, t->size - 1, &end)) {
212 err = -EFAULT;
213 goto unpin_all;
214 }
215
216 /* The uptr's struct cannot span across two pages */
217 if ((start & PAGE_MASK) != (end & PAGE_MASK)) {
218 err = -EOPNOTSUPP;
219 goto unpin_all;
220 }
221
222 err = pin_user_pages_fast(start, 1, FOLL_LONGTERM | FOLL_WRITE, &page);
223 if (err != 1)
224 goto unpin_all;
225
226 if (PageHighMem(page)) {
227 err = -EOPNOTSUPP;
228 unpin_user_page(page);
229 goto unpin_all;
230 }
231
232 *uptr_addr = page_address(page) + offset_in_page(start);
233 }
234
235 return 0;
236
237unpin_all:
238 __bpf_obj_unpin_uptrs(rec, i, obj);
239 return err;
240}
241
242static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
243 void *key, void *value, __u64 flags)
244{
245 int err;
246
247 /* Need to create a kthread, thus must support schedule */
248 if (bpf_map_is_offloaded(map)) {
249 return bpf_map_offload_update_elem(map, key, value, flags);
250 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
251 map->map_type == BPF_MAP_TYPE_ARENA ||
252 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
253 return map->ops->map_update_elem(map, key, value, flags);
254 } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
255 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
256 return sock_map_update_elem_sys(map, key, value, flags);
257 } else if (IS_FD_PROG_ARRAY(map)) {
258 return bpf_fd_array_map_update_elem(map, map_file, key, value,
259 flags);
260 }
261
262 bpf_disable_instrumentation();
263 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
264 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
265 err = bpf_percpu_hash_update(map, key, value, flags);
266 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
267 err = bpf_percpu_array_update(map, key, value, flags);
268 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
269 err = bpf_percpu_cgroup_storage_update(map, key, value,
270 flags);
271 } else if (IS_FD_ARRAY(map)) {
272 err = bpf_fd_array_map_update_elem(map, map_file, key, value,
273 flags);
274 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
275 err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
276 flags);
277 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
278 /* rcu_read_lock() is not needed */
279 err = bpf_fd_reuseport_array_update_elem(map, key, value,
280 flags);
281 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
282 map->map_type == BPF_MAP_TYPE_STACK ||
283 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
284 err = map->ops->map_push_elem(map, value, flags);
285 } else {
286 err = bpf_obj_pin_uptrs(map->record, value);
287 if (!err) {
288 rcu_read_lock();
289 err = map->ops->map_update_elem(map, key, value, flags);
290 rcu_read_unlock();
291 if (err)
292 bpf_obj_unpin_uptrs(map->record, value);
293 }
294 }
295 bpf_enable_instrumentation();
296
297 return err;
298}
299
300static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
301 __u64 flags)
302{
303 void *ptr;
304 int err;
305
306 if (bpf_map_is_offloaded(map))
307 return bpf_map_offload_lookup_elem(map, key, value);
308
309 bpf_disable_instrumentation();
310 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
311 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
312 err = bpf_percpu_hash_copy(map, key, value);
313 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
314 err = bpf_percpu_array_copy(map, key, value);
315 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
316 err = bpf_percpu_cgroup_storage_copy(map, key, value);
317 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
318 err = bpf_stackmap_copy(map, key, value);
319 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
320 err = bpf_fd_array_map_lookup_elem(map, key, value);
321 } else if (IS_FD_HASH(map)) {
322 err = bpf_fd_htab_map_lookup_elem(map, key, value);
323 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
324 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
325 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
326 map->map_type == BPF_MAP_TYPE_STACK ||
327 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
328 err = map->ops->map_peek_elem(map, value);
329 } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
330 /* struct_ops map requires directly updating "value" */
331 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
332 } else {
333 rcu_read_lock();
334 if (map->ops->map_lookup_elem_sys_only)
335 ptr = map->ops->map_lookup_elem_sys_only(map, key);
336 else
337 ptr = map->ops->map_lookup_elem(map, key);
338 if (IS_ERR(ptr)) {
339 err = PTR_ERR(ptr);
340 } else if (!ptr) {
341 err = -ENOENT;
342 } else {
343 err = 0;
344 if (flags & BPF_F_LOCK)
345 /* lock 'ptr' and copy everything but lock */
346 copy_map_value_locked(map, value, ptr, true);
347 else
348 copy_map_value(map, value, ptr);
349 /* mask lock and timer, since value wasn't zero inited */
350 check_and_init_map_value(map, value);
351 }
352 rcu_read_unlock();
353 }
354
355 bpf_enable_instrumentation();
356
357 return err;
358}
359
360/* Please, do not use this function outside from the map creation path
361 * (e.g. in map update path) without taking care of setting the active
362 * memory cgroup (see at bpf_map_kmalloc_node() for example).
363 */
364static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
365{
366 /* We really just want to fail instead of triggering OOM killer
367 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
368 * which is used for lower order allocation requests.
369 *
370 * It has been observed that higher order allocation requests done by
371 * vmalloc with __GFP_NORETRY being set might fail due to not trying
372 * to reclaim memory from the page cache, thus we set
373 * __GFP_RETRY_MAYFAIL to avoid such situations.
374 */
375
376 gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
377 unsigned int flags = 0;
378 unsigned long align = 1;
379 void *area;
380
381 if (size >= SIZE_MAX)
382 return NULL;
383
384 /* kmalloc()'ed memory can't be mmap()'ed */
385 if (mmapable) {
386 BUG_ON(!PAGE_ALIGNED(size));
387 align = SHMLBA;
388 flags = VM_USERMAP;
389 } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
390 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
391 numa_node);
392 if (area != NULL)
393 return area;
394 }
395
396 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
397 gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
398 flags, numa_node, __builtin_return_address(0));
399}
400
401void *bpf_map_area_alloc(u64 size, int numa_node)
402{
403 return __bpf_map_area_alloc(size, numa_node, false);
404}
405
406void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
407{
408 return __bpf_map_area_alloc(size, numa_node, true);
409}
410
411void bpf_map_area_free(void *area)
412{
413 kvfree(area);
414}
415
416static u32 bpf_map_flags_retain_permanent(u32 flags)
417{
418 /* Some map creation flags are not tied to the map object but
419 * rather to the map fd instead, so they have no meaning upon
420 * map object inspection since multiple file descriptors with
421 * different (access) properties can exist here. Thus, given
422 * this has zero meaning for the map itself, lets clear these
423 * from here.
424 */
425 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
426}
427
428void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
429{
430 map->map_type = attr->map_type;
431 map->key_size = attr->key_size;
432 map->value_size = attr->value_size;
433 map->max_entries = attr->max_entries;
434 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
435 map->numa_node = bpf_map_attr_numa_node(attr);
436 map->map_extra = attr->map_extra;
437}
438
439static int bpf_map_alloc_id(struct bpf_map *map)
440{
441 int id;
442
443 idr_preload(GFP_KERNEL);
444 spin_lock_bh(&map_idr_lock);
445 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
446 if (id > 0)
447 map->id = id;
448 spin_unlock_bh(&map_idr_lock);
449 idr_preload_end();
450
451 if (WARN_ON_ONCE(!id))
452 return -ENOSPC;
453
454 return id > 0 ? 0 : id;
455}
456
457void bpf_map_free_id(struct bpf_map *map)
458{
459 unsigned long flags;
460
461 /* Offloaded maps are removed from the IDR store when their device
462 * disappears - even if someone holds an fd to them they are unusable,
463 * the memory is gone, all ops will fail; they are simply waiting for
464 * refcnt to drop to be freed.
465 */
466 if (!map->id)
467 return;
468
469 spin_lock_irqsave(&map_idr_lock, flags);
470
471 idr_remove(&map_idr, map->id);
472 map->id = 0;
473
474 spin_unlock_irqrestore(&map_idr_lock, flags);
475}
476
477#ifdef CONFIG_MEMCG
478static void bpf_map_save_memcg(struct bpf_map *map)
479{
480 /* Currently if a map is created by a process belonging to the root
481 * memory cgroup, get_obj_cgroup_from_current() will return NULL.
482 * So we have to check map->objcg for being NULL each time it's
483 * being used.
484 */
485 if (memcg_bpf_enabled())
486 map->objcg = get_obj_cgroup_from_current();
487}
488
489static void bpf_map_release_memcg(struct bpf_map *map)
490{
491 if (map->objcg)
492 obj_cgroup_put(map->objcg);
493}
494
495static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
496{
497 if (map->objcg)
498 return get_mem_cgroup_from_objcg(map->objcg);
499
500 return root_mem_cgroup;
501}
502
503void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
504 int node)
505{
506 struct mem_cgroup *memcg, *old_memcg;
507 void *ptr;
508
509 memcg = bpf_map_get_memcg(map);
510 old_memcg = set_active_memcg(memcg);
511 ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
512 set_active_memcg(old_memcg);
513 mem_cgroup_put(memcg);
514
515 return ptr;
516}
517
518void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
519{
520 struct mem_cgroup *memcg, *old_memcg;
521 void *ptr;
522
523 memcg = bpf_map_get_memcg(map);
524 old_memcg = set_active_memcg(memcg);
525 ptr = kzalloc(size, flags | __GFP_ACCOUNT);
526 set_active_memcg(old_memcg);
527 mem_cgroup_put(memcg);
528
529 return ptr;
530}
531
532void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
533 gfp_t flags)
534{
535 struct mem_cgroup *memcg, *old_memcg;
536 void *ptr;
537
538 memcg = bpf_map_get_memcg(map);
539 old_memcg = set_active_memcg(memcg);
540 ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
541 set_active_memcg(old_memcg);
542 mem_cgroup_put(memcg);
543
544 return ptr;
545}
546
547void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
548 size_t align, gfp_t flags)
549{
550 struct mem_cgroup *memcg, *old_memcg;
551 void __percpu *ptr;
552
553 memcg = bpf_map_get_memcg(map);
554 old_memcg = set_active_memcg(memcg);
555 ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
556 set_active_memcg(old_memcg);
557 mem_cgroup_put(memcg);
558
559 return ptr;
560}
561
562#else
563static void bpf_map_save_memcg(struct bpf_map *map)
564{
565}
566
567static void bpf_map_release_memcg(struct bpf_map *map)
568{
569}
570#endif
571
572int bpf_map_alloc_pages(const struct bpf_map *map, gfp_t gfp, int nid,
573 unsigned long nr_pages, struct page **pages)
574{
575 unsigned long i, j;
576 struct page *pg;
577 int ret = 0;
578#ifdef CONFIG_MEMCG
579 struct mem_cgroup *memcg, *old_memcg;
580
581 memcg = bpf_map_get_memcg(map);
582 old_memcg = set_active_memcg(memcg);
583#endif
584 for (i = 0; i < nr_pages; i++) {
585 pg = alloc_pages_node(nid, gfp | __GFP_ACCOUNT, 0);
586
587 if (pg) {
588 pages[i] = pg;
589 continue;
590 }
591 for (j = 0; j < i; j++)
592 __free_page(pages[j]);
593 ret = -ENOMEM;
594 break;
595 }
596
597#ifdef CONFIG_MEMCG
598 set_active_memcg(old_memcg);
599 mem_cgroup_put(memcg);
600#endif
601 return ret;
602}
603
604
605static int btf_field_cmp(const void *a, const void *b)
606{
607 const struct btf_field *f1 = a, *f2 = b;
608
609 if (f1->offset < f2->offset)
610 return -1;
611 else if (f1->offset > f2->offset)
612 return 1;
613 return 0;
614}
615
616struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
617 u32 field_mask)
618{
619 struct btf_field *field;
620
621 if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & field_mask))
622 return NULL;
623 field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
624 if (!field || !(field->type & field_mask))
625 return NULL;
626 return field;
627}
628
629void btf_record_free(struct btf_record *rec)
630{
631 int i;
632
633 if (IS_ERR_OR_NULL(rec))
634 return;
635 for (i = 0; i < rec->cnt; i++) {
636 switch (rec->fields[i].type) {
637 case BPF_KPTR_UNREF:
638 case BPF_KPTR_REF:
639 case BPF_KPTR_PERCPU:
640 case BPF_UPTR:
641 if (rec->fields[i].kptr.module)
642 module_put(rec->fields[i].kptr.module);
643 if (btf_is_kernel(rec->fields[i].kptr.btf))
644 btf_put(rec->fields[i].kptr.btf);
645 break;
646 case BPF_LIST_HEAD:
647 case BPF_LIST_NODE:
648 case BPF_RB_ROOT:
649 case BPF_RB_NODE:
650 case BPF_SPIN_LOCK:
651 case BPF_TIMER:
652 case BPF_REFCOUNT:
653 case BPF_WORKQUEUE:
654 /* Nothing to release */
655 break;
656 default:
657 WARN_ON_ONCE(1);
658 continue;
659 }
660 }
661 kfree(rec);
662}
663
664void bpf_map_free_record(struct bpf_map *map)
665{
666 btf_record_free(map->record);
667 map->record = NULL;
668}
669
670struct btf_record *btf_record_dup(const struct btf_record *rec)
671{
672 const struct btf_field *fields;
673 struct btf_record *new_rec;
674 int ret, size, i;
675
676 if (IS_ERR_OR_NULL(rec))
677 return NULL;
678 size = offsetof(struct btf_record, fields[rec->cnt]);
679 new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
680 if (!new_rec)
681 return ERR_PTR(-ENOMEM);
682 /* Do a deep copy of the btf_record */
683 fields = rec->fields;
684 new_rec->cnt = 0;
685 for (i = 0; i < rec->cnt; i++) {
686 switch (fields[i].type) {
687 case BPF_KPTR_UNREF:
688 case BPF_KPTR_REF:
689 case BPF_KPTR_PERCPU:
690 case BPF_UPTR:
691 if (btf_is_kernel(fields[i].kptr.btf))
692 btf_get(fields[i].kptr.btf);
693 if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
694 ret = -ENXIO;
695 goto free;
696 }
697 break;
698 case BPF_LIST_HEAD:
699 case BPF_LIST_NODE:
700 case BPF_RB_ROOT:
701 case BPF_RB_NODE:
702 case BPF_SPIN_LOCK:
703 case BPF_TIMER:
704 case BPF_REFCOUNT:
705 case BPF_WORKQUEUE:
706 /* Nothing to acquire */
707 break;
708 default:
709 ret = -EFAULT;
710 WARN_ON_ONCE(1);
711 goto free;
712 }
713 new_rec->cnt++;
714 }
715 return new_rec;
716free:
717 btf_record_free(new_rec);
718 return ERR_PTR(ret);
719}
720
721bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
722{
723 bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
724 int size;
725
726 if (!a_has_fields && !b_has_fields)
727 return true;
728 if (a_has_fields != b_has_fields)
729 return false;
730 if (rec_a->cnt != rec_b->cnt)
731 return false;
732 size = offsetof(struct btf_record, fields[rec_a->cnt]);
733 /* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
734 * members are zeroed out. So memcmp is safe to do without worrying
735 * about padding/unused fields.
736 *
737 * While spin_lock, timer, and kptr have no relation to map BTF,
738 * list_head metadata is specific to map BTF, the btf and value_rec
739 * members in particular. btf is the map BTF, while value_rec points to
740 * btf_record in that map BTF.
741 *
742 * So while by default, we don't rely on the map BTF (which the records
743 * were parsed from) matching for both records, which is not backwards
744 * compatible, in case list_head is part of it, we implicitly rely on
745 * that by way of depending on memcmp succeeding for it.
746 */
747 return !memcmp(rec_a, rec_b, size);
748}
749
750void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
751{
752 if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
753 return;
754 bpf_timer_cancel_and_free(obj + rec->timer_off);
755}
756
757void bpf_obj_free_workqueue(const struct btf_record *rec, void *obj)
758{
759 if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_WORKQUEUE)))
760 return;
761 bpf_wq_cancel_and_free(obj + rec->wq_off);
762}
763
764void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
765{
766 const struct btf_field *fields;
767 int i;
768
769 if (IS_ERR_OR_NULL(rec))
770 return;
771 fields = rec->fields;
772 for (i = 0; i < rec->cnt; i++) {
773 struct btf_struct_meta *pointee_struct_meta;
774 const struct btf_field *field = &fields[i];
775 void *field_ptr = obj + field->offset;
776 void *xchgd_field;
777
778 switch (fields[i].type) {
779 case BPF_SPIN_LOCK:
780 break;
781 case BPF_TIMER:
782 bpf_timer_cancel_and_free(field_ptr);
783 break;
784 case BPF_WORKQUEUE:
785 bpf_wq_cancel_and_free(field_ptr);
786 break;
787 case BPF_KPTR_UNREF:
788 WRITE_ONCE(*(u64 *)field_ptr, 0);
789 break;
790 case BPF_KPTR_REF:
791 case BPF_KPTR_PERCPU:
792 xchgd_field = (void *)xchg((unsigned long *)field_ptr, 0);
793 if (!xchgd_field)
794 break;
795
796 if (!btf_is_kernel(field->kptr.btf)) {
797 pointee_struct_meta = btf_find_struct_meta(field->kptr.btf,
798 field->kptr.btf_id);
799 migrate_disable();
800 __bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
801 pointee_struct_meta->record : NULL,
802 fields[i].type == BPF_KPTR_PERCPU);
803 migrate_enable();
804 } else {
805 field->kptr.dtor(xchgd_field);
806 }
807 break;
808 case BPF_UPTR:
809 /* The caller ensured that no one is using the uptr */
810 unpin_uptr_kaddr(*(void **)field_ptr);
811 break;
812 case BPF_LIST_HEAD:
813 if (WARN_ON_ONCE(rec->spin_lock_off < 0))
814 continue;
815 bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
816 break;
817 case BPF_RB_ROOT:
818 if (WARN_ON_ONCE(rec->spin_lock_off < 0))
819 continue;
820 bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
821 break;
822 case BPF_LIST_NODE:
823 case BPF_RB_NODE:
824 case BPF_REFCOUNT:
825 break;
826 default:
827 WARN_ON_ONCE(1);
828 continue;
829 }
830 }
831}
832
833static void bpf_map_free(struct bpf_map *map)
834{
835 struct btf_record *rec = map->record;
836 struct btf *btf = map->btf;
837
838 /* implementation dependent freeing */
839 map->ops->map_free(map);
840 /* Delay freeing of btf_record for maps, as map_free
841 * callback usually needs access to them. It is better to do it here
842 * than require each callback to do the free itself manually.
843 *
844 * Note that the btf_record stashed in map->inner_map_meta->record was
845 * already freed using the map_free callback for map in map case which
846 * eventually calls bpf_map_free_meta, since inner_map_meta is only a
847 * template bpf_map struct used during verification.
848 */
849 btf_record_free(rec);
850 /* Delay freeing of btf for maps, as map_free callback may need
851 * struct_meta info which will be freed with btf_put().
852 */
853 btf_put(btf);
854}
855
856/* called from workqueue */
857static void bpf_map_free_deferred(struct work_struct *work)
858{
859 struct bpf_map *map = container_of(work, struct bpf_map, work);
860
861 security_bpf_map_free(map);
862 bpf_map_release_memcg(map);
863 bpf_map_free(map);
864}
865
866static void bpf_map_put_uref(struct bpf_map *map)
867{
868 if (atomic64_dec_and_test(&map->usercnt)) {
869 if (map->ops->map_release_uref)
870 map->ops->map_release_uref(map);
871 }
872}
873
874static void bpf_map_free_in_work(struct bpf_map *map)
875{
876 INIT_WORK(&map->work, bpf_map_free_deferred);
877 /* Avoid spawning kworkers, since they all might contend
878 * for the same mutex like slab_mutex.
879 */
880 queue_work(system_unbound_wq, &map->work);
881}
882
883static void bpf_map_free_rcu_gp(struct rcu_head *rcu)
884{
885 bpf_map_free_in_work(container_of(rcu, struct bpf_map, rcu));
886}
887
888static void bpf_map_free_mult_rcu_gp(struct rcu_head *rcu)
889{
890 if (rcu_trace_implies_rcu_gp())
891 bpf_map_free_rcu_gp(rcu);
892 else
893 call_rcu(rcu, bpf_map_free_rcu_gp);
894}
895
896/* decrement map refcnt and schedule it for freeing via workqueue
897 * (underlying map implementation ops->map_free() might sleep)
898 */
899void bpf_map_put(struct bpf_map *map)
900{
901 if (atomic64_dec_and_test(&map->refcnt)) {
902 /* bpf_map_free_id() must be called first */
903 bpf_map_free_id(map);
904
905 WARN_ON_ONCE(atomic64_read(&map->sleepable_refcnt));
906 if (READ_ONCE(map->free_after_mult_rcu_gp))
907 call_rcu_tasks_trace(&map->rcu, bpf_map_free_mult_rcu_gp);
908 else if (READ_ONCE(map->free_after_rcu_gp))
909 call_rcu(&map->rcu, bpf_map_free_rcu_gp);
910 else
911 bpf_map_free_in_work(map);
912 }
913}
914EXPORT_SYMBOL_GPL(bpf_map_put);
915
916void bpf_map_put_with_uref(struct bpf_map *map)
917{
918 bpf_map_put_uref(map);
919 bpf_map_put(map);
920}
921
922static int bpf_map_release(struct inode *inode, struct file *filp)
923{
924 struct bpf_map *map = filp->private_data;
925
926 if (map->ops->map_release)
927 map->ops->map_release(map, filp);
928
929 bpf_map_put_with_uref(map);
930 return 0;
931}
932
933static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
934{
935 fmode_t mode = fd_file(f)->f_mode;
936
937 /* Our file permissions may have been overridden by global
938 * map permissions facing syscall side.
939 */
940 if (READ_ONCE(map->frozen))
941 mode &= ~FMODE_CAN_WRITE;
942 return mode;
943}
944
945#ifdef CONFIG_PROC_FS
946/* Show the memory usage of a bpf map */
947static u64 bpf_map_memory_usage(const struct bpf_map *map)
948{
949 return map->ops->map_mem_usage(map);
950}
951
952static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
953{
954 struct bpf_map *map = filp->private_data;
955 u32 type = 0, jited = 0;
956
957 if (map_type_contains_progs(map)) {
958 spin_lock(&map->owner.lock);
959 type = map->owner.type;
960 jited = map->owner.jited;
961 spin_unlock(&map->owner.lock);
962 }
963
964 seq_printf(m,
965 "map_type:\t%u\n"
966 "key_size:\t%u\n"
967 "value_size:\t%u\n"
968 "max_entries:\t%u\n"
969 "map_flags:\t%#x\n"
970 "map_extra:\t%#llx\n"
971 "memlock:\t%llu\n"
972 "map_id:\t%u\n"
973 "frozen:\t%u\n",
974 map->map_type,
975 map->key_size,
976 map->value_size,
977 map->max_entries,
978 map->map_flags,
979 (unsigned long long)map->map_extra,
980 bpf_map_memory_usage(map),
981 map->id,
982 READ_ONCE(map->frozen));
983 if (type) {
984 seq_printf(m, "owner_prog_type:\t%u\n", type);
985 seq_printf(m, "owner_jited:\t%u\n", jited);
986 }
987}
988#endif
989
990static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
991 loff_t *ppos)
992{
993 /* We need this handler such that alloc_file() enables
994 * f_mode with FMODE_CAN_READ.
995 */
996 return -EINVAL;
997}
998
999static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
1000 size_t siz, loff_t *ppos)
1001{
1002 /* We need this handler such that alloc_file() enables
1003 * f_mode with FMODE_CAN_WRITE.
1004 */
1005 return -EINVAL;
1006}
1007
1008/* called for any extra memory-mapped regions (except initial) */
1009static void bpf_map_mmap_open(struct vm_area_struct *vma)
1010{
1011 struct bpf_map *map = vma->vm_file->private_data;
1012
1013 if (vma->vm_flags & VM_MAYWRITE)
1014 bpf_map_write_active_inc(map);
1015}
1016
1017/* called for all unmapped memory region (including initial) */
1018static void bpf_map_mmap_close(struct vm_area_struct *vma)
1019{
1020 struct bpf_map *map = vma->vm_file->private_data;
1021
1022 if (vma->vm_flags & VM_MAYWRITE)
1023 bpf_map_write_active_dec(map);
1024}
1025
1026static const struct vm_operations_struct bpf_map_default_vmops = {
1027 .open = bpf_map_mmap_open,
1028 .close = bpf_map_mmap_close,
1029};
1030
1031static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
1032{
1033 struct bpf_map *map = filp->private_data;
1034 int err = 0;
1035
1036 if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
1037 return -ENOTSUPP;
1038
1039 if (!(vma->vm_flags & VM_SHARED))
1040 return -EINVAL;
1041
1042 mutex_lock(&map->freeze_mutex);
1043
1044 if (vma->vm_flags & VM_WRITE) {
1045 if (map->frozen) {
1046 err = -EPERM;
1047 goto out;
1048 }
1049 /* map is meant to be read-only, so do not allow mapping as
1050 * writable, because it's possible to leak a writable page
1051 * reference and allows user-space to still modify it after
1052 * freezing, while verifier will assume contents do not change
1053 */
1054 if (map->map_flags & BPF_F_RDONLY_PROG) {
1055 err = -EACCES;
1056 goto out;
1057 }
1058 bpf_map_write_active_inc(map);
1059 }
1060out:
1061 mutex_unlock(&map->freeze_mutex);
1062 if (err)
1063 return err;
1064
1065 /* set default open/close callbacks */
1066 vma->vm_ops = &bpf_map_default_vmops;
1067 vma->vm_private_data = map;
1068 vm_flags_clear(vma, VM_MAYEXEC);
1069 /* If mapping is read-only, then disallow potentially re-mapping with
1070 * PROT_WRITE by dropping VM_MAYWRITE flag. This VM_MAYWRITE clearing
1071 * means that as far as BPF map's memory-mapped VMAs are concerned,
1072 * VM_WRITE and VM_MAYWRITE and equivalent, if one of them is set,
1073 * both should be set, so we can forget about VM_MAYWRITE and always
1074 * check just VM_WRITE
1075 */
1076 if (!(vma->vm_flags & VM_WRITE))
1077 vm_flags_clear(vma, VM_MAYWRITE);
1078
1079 err = map->ops->map_mmap(map, vma);
1080 if (err) {
1081 if (vma->vm_flags & VM_WRITE)
1082 bpf_map_write_active_dec(map);
1083 }
1084
1085 return err;
1086}
1087
1088static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
1089{
1090 struct bpf_map *map = filp->private_data;
1091
1092 if (map->ops->map_poll)
1093 return map->ops->map_poll(map, filp, pts);
1094
1095 return EPOLLERR;
1096}
1097
1098static unsigned long bpf_get_unmapped_area(struct file *filp, unsigned long addr,
1099 unsigned long len, unsigned long pgoff,
1100 unsigned long flags)
1101{
1102 struct bpf_map *map = filp->private_data;
1103
1104 if (map->ops->map_get_unmapped_area)
1105 return map->ops->map_get_unmapped_area(filp, addr, len, pgoff, flags);
1106#ifdef CONFIG_MMU
1107 return mm_get_unmapped_area(current->mm, filp, addr, len, pgoff, flags);
1108#else
1109 return addr;
1110#endif
1111}
1112
1113const struct file_operations bpf_map_fops = {
1114#ifdef CONFIG_PROC_FS
1115 .show_fdinfo = bpf_map_show_fdinfo,
1116#endif
1117 .release = bpf_map_release,
1118 .read = bpf_dummy_read,
1119 .write = bpf_dummy_write,
1120 .mmap = bpf_map_mmap,
1121 .poll = bpf_map_poll,
1122 .get_unmapped_area = bpf_get_unmapped_area,
1123};
1124
1125int bpf_map_new_fd(struct bpf_map *map, int flags)
1126{
1127 int ret;
1128
1129 ret = security_bpf_map(map, OPEN_FMODE(flags));
1130 if (ret < 0)
1131 return ret;
1132
1133 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
1134 flags | O_CLOEXEC);
1135}
1136
1137int bpf_get_file_flag(int flags)
1138{
1139 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
1140 return -EINVAL;
1141 if (flags & BPF_F_RDONLY)
1142 return O_RDONLY;
1143 if (flags & BPF_F_WRONLY)
1144 return O_WRONLY;
1145 return O_RDWR;
1146}
1147
1148/* helper macro to check that unused fields 'union bpf_attr' are zero */
1149#define CHECK_ATTR(CMD) \
1150 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
1151 sizeof(attr->CMD##_LAST_FIELD), 0, \
1152 sizeof(*attr) - \
1153 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
1154 sizeof(attr->CMD##_LAST_FIELD)) != NULL
1155
1156/* dst and src must have at least "size" number of bytes.
1157 * Return strlen on success and < 0 on error.
1158 */
1159int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
1160{
1161 const char *end = src + size;
1162 const char *orig_src = src;
1163
1164 memset(dst, 0, size);
1165 /* Copy all isalnum(), '_' and '.' chars. */
1166 while (src < end && *src) {
1167 if (!isalnum(*src) &&
1168 *src != '_' && *src != '.')
1169 return -EINVAL;
1170 *dst++ = *src++;
1171 }
1172
1173 /* No '\0' found in "size" number of bytes */
1174 if (src == end)
1175 return -EINVAL;
1176
1177 return src - orig_src;
1178}
1179
1180int map_check_no_btf(const struct bpf_map *map,
1181 const struct btf *btf,
1182 const struct btf_type *key_type,
1183 const struct btf_type *value_type)
1184{
1185 return -ENOTSUPP;
1186}
1187
1188static int map_check_btf(struct bpf_map *map, struct bpf_token *token,
1189 const struct btf *btf, u32 btf_key_id, u32 btf_value_id)
1190{
1191 const struct btf_type *key_type, *value_type;
1192 u32 key_size, value_size;
1193 int ret = 0;
1194
1195 /* Some maps allow key to be unspecified. */
1196 if (btf_key_id) {
1197 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
1198 if (!key_type || key_size != map->key_size)
1199 return -EINVAL;
1200 } else {
1201 key_type = btf_type_by_id(btf, 0);
1202 if (!map->ops->map_check_btf)
1203 return -EINVAL;
1204 }
1205
1206 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1207 if (!value_type || value_size != map->value_size)
1208 return -EINVAL;
1209
1210 map->record = btf_parse_fields(btf, value_type,
1211 BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1212 BPF_RB_ROOT | BPF_REFCOUNT | BPF_WORKQUEUE | BPF_UPTR,
1213 map->value_size);
1214 if (!IS_ERR_OR_NULL(map->record)) {
1215 int i;
1216
1217 if (!bpf_token_capable(token, CAP_BPF)) {
1218 ret = -EPERM;
1219 goto free_map_tab;
1220 }
1221 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1222 ret = -EACCES;
1223 goto free_map_tab;
1224 }
1225 for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1226 switch (map->record->field_mask & (1 << i)) {
1227 case 0:
1228 continue;
1229 case BPF_SPIN_LOCK:
1230 if (map->map_type != BPF_MAP_TYPE_HASH &&
1231 map->map_type != BPF_MAP_TYPE_ARRAY &&
1232 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1233 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1234 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1235 map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1236 map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1237 ret = -EOPNOTSUPP;
1238 goto free_map_tab;
1239 }
1240 break;
1241 case BPF_TIMER:
1242 case BPF_WORKQUEUE:
1243 if (map->map_type != BPF_MAP_TYPE_HASH &&
1244 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1245 map->map_type != BPF_MAP_TYPE_ARRAY) {
1246 ret = -EOPNOTSUPP;
1247 goto free_map_tab;
1248 }
1249 break;
1250 case BPF_KPTR_UNREF:
1251 case BPF_KPTR_REF:
1252 case BPF_KPTR_PERCPU:
1253 case BPF_REFCOUNT:
1254 if (map->map_type != BPF_MAP_TYPE_HASH &&
1255 map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
1256 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1257 map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH &&
1258 map->map_type != BPF_MAP_TYPE_ARRAY &&
1259 map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY &&
1260 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1261 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1262 map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1263 map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1264 ret = -EOPNOTSUPP;
1265 goto free_map_tab;
1266 }
1267 break;
1268 case BPF_UPTR:
1269 if (map->map_type != BPF_MAP_TYPE_TASK_STORAGE) {
1270 ret = -EOPNOTSUPP;
1271 goto free_map_tab;
1272 }
1273 break;
1274 case BPF_LIST_HEAD:
1275 case BPF_RB_ROOT:
1276 if (map->map_type != BPF_MAP_TYPE_HASH &&
1277 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1278 map->map_type != BPF_MAP_TYPE_ARRAY) {
1279 ret = -EOPNOTSUPP;
1280 goto free_map_tab;
1281 }
1282 break;
1283 default:
1284 /* Fail if map_type checks are missing for a field type */
1285 ret = -EOPNOTSUPP;
1286 goto free_map_tab;
1287 }
1288 }
1289 }
1290
1291 ret = btf_check_and_fixup_fields(btf, map->record);
1292 if (ret < 0)
1293 goto free_map_tab;
1294
1295 if (map->ops->map_check_btf) {
1296 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1297 if (ret < 0)
1298 goto free_map_tab;
1299 }
1300
1301 return ret;
1302free_map_tab:
1303 bpf_map_free_record(map);
1304 return ret;
1305}
1306
1307static bool bpf_net_capable(void)
1308{
1309 return capable(CAP_NET_ADMIN) || capable(CAP_SYS_ADMIN);
1310}
1311
1312#define BPF_MAP_CREATE_LAST_FIELD map_token_fd
1313/* called via syscall */
1314static int map_create(union bpf_attr *attr)
1315{
1316 const struct bpf_map_ops *ops;
1317 struct bpf_token *token = NULL;
1318 int numa_node = bpf_map_attr_numa_node(attr);
1319 u32 map_type = attr->map_type;
1320 struct bpf_map *map;
1321 bool token_flag;
1322 int f_flags;
1323 int err;
1324
1325 err = CHECK_ATTR(BPF_MAP_CREATE);
1326 if (err)
1327 return -EINVAL;
1328
1329 /* check BPF_F_TOKEN_FD flag, remember if it's set, and then clear it
1330 * to avoid per-map type checks tripping on unknown flag
1331 */
1332 token_flag = attr->map_flags & BPF_F_TOKEN_FD;
1333 attr->map_flags &= ~BPF_F_TOKEN_FD;
1334
1335 if (attr->btf_vmlinux_value_type_id) {
1336 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1337 attr->btf_key_type_id || attr->btf_value_type_id)
1338 return -EINVAL;
1339 } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1340 return -EINVAL;
1341 }
1342
1343 if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1344 attr->map_type != BPF_MAP_TYPE_ARENA &&
1345 attr->map_extra != 0)
1346 return -EINVAL;
1347
1348 f_flags = bpf_get_file_flag(attr->map_flags);
1349 if (f_flags < 0)
1350 return f_flags;
1351
1352 if (numa_node != NUMA_NO_NODE &&
1353 ((unsigned int)numa_node >= nr_node_ids ||
1354 !node_online(numa_node)))
1355 return -EINVAL;
1356
1357 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1358 map_type = attr->map_type;
1359 if (map_type >= ARRAY_SIZE(bpf_map_types))
1360 return -EINVAL;
1361 map_type = array_index_nospec(map_type, ARRAY_SIZE(bpf_map_types));
1362 ops = bpf_map_types[map_type];
1363 if (!ops)
1364 return -EINVAL;
1365
1366 if (ops->map_alloc_check) {
1367 err = ops->map_alloc_check(attr);
1368 if (err)
1369 return err;
1370 }
1371 if (attr->map_ifindex)
1372 ops = &bpf_map_offload_ops;
1373 if (!ops->map_mem_usage)
1374 return -EINVAL;
1375
1376 if (token_flag) {
1377 token = bpf_token_get_from_fd(attr->map_token_fd);
1378 if (IS_ERR(token))
1379 return PTR_ERR(token);
1380
1381 /* if current token doesn't grant map creation permissions,
1382 * then we can't use this token, so ignore it and rely on
1383 * system-wide capabilities checks
1384 */
1385 if (!bpf_token_allow_cmd(token, BPF_MAP_CREATE) ||
1386 !bpf_token_allow_map_type(token, attr->map_type)) {
1387 bpf_token_put(token);
1388 token = NULL;
1389 }
1390 }
1391
1392 err = -EPERM;
1393
1394 /* Intent here is for unprivileged_bpf_disabled to block BPF map
1395 * creation for unprivileged users; other actions depend
1396 * on fd availability and access to bpffs, so are dependent on
1397 * object creation success. Even with unprivileged BPF disabled,
1398 * capability checks are still carried out.
1399 */
1400 if (sysctl_unprivileged_bpf_disabled && !bpf_token_capable(token, CAP_BPF))
1401 goto put_token;
1402
1403 /* check privileged map type permissions */
1404 switch (map_type) {
1405 case BPF_MAP_TYPE_ARRAY:
1406 case BPF_MAP_TYPE_PERCPU_ARRAY:
1407 case BPF_MAP_TYPE_PROG_ARRAY:
1408 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1409 case BPF_MAP_TYPE_CGROUP_ARRAY:
1410 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1411 case BPF_MAP_TYPE_HASH:
1412 case BPF_MAP_TYPE_PERCPU_HASH:
1413 case BPF_MAP_TYPE_HASH_OF_MAPS:
1414 case BPF_MAP_TYPE_RINGBUF:
1415 case BPF_MAP_TYPE_USER_RINGBUF:
1416 case BPF_MAP_TYPE_CGROUP_STORAGE:
1417 case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
1418 /* unprivileged */
1419 break;
1420 case BPF_MAP_TYPE_SK_STORAGE:
1421 case BPF_MAP_TYPE_INODE_STORAGE:
1422 case BPF_MAP_TYPE_TASK_STORAGE:
1423 case BPF_MAP_TYPE_CGRP_STORAGE:
1424 case BPF_MAP_TYPE_BLOOM_FILTER:
1425 case BPF_MAP_TYPE_LPM_TRIE:
1426 case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
1427 case BPF_MAP_TYPE_STACK_TRACE:
1428 case BPF_MAP_TYPE_QUEUE:
1429 case BPF_MAP_TYPE_STACK:
1430 case BPF_MAP_TYPE_LRU_HASH:
1431 case BPF_MAP_TYPE_LRU_PERCPU_HASH:
1432 case BPF_MAP_TYPE_STRUCT_OPS:
1433 case BPF_MAP_TYPE_CPUMAP:
1434 case BPF_MAP_TYPE_ARENA:
1435 if (!bpf_token_capable(token, CAP_BPF))
1436 goto put_token;
1437 break;
1438 case BPF_MAP_TYPE_SOCKMAP:
1439 case BPF_MAP_TYPE_SOCKHASH:
1440 case BPF_MAP_TYPE_DEVMAP:
1441 case BPF_MAP_TYPE_DEVMAP_HASH:
1442 case BPF_MAP_TYPE_XSKMAP:
1443 if (!bpf_token_capable(token, CAP_NET_ADMIN))
1444 goto put_token;
1445 break;
1446 default:
1447 WARN(1, "unsupported map type %d", map_type);
1448 goto put_token;
1449 }
1450
1451 map = ops->map_alloc(attr);
1452 if (IS_ERR(map)) {
1453 err = PTR_ERR(map);
1454 goto put_token;
1455 }
1456 map->ops = ops;
1457 map->map_type = map_type;
1458
1459 err = bpf_obj_name_cpy(map->name, attr->map_name,
1460 sizeof(attr->map_name));
1461 if (err < 0)
1462 goto free_map;
1463
1464 atomic64_set(&map->refcnt, 1);
1465 atomic64_set(&map->usercnt, 1);
1466 mutex_init(&map->freeze_mutex);
1467 spin_lock_init(&map->owner.lock);
1468
1469 if (attr->btf_key_type_id || attr->btf_value_type_id ||
1470 /* Even the map's value is a kernel's struct,
1471 * the bpf_prog.o must have BTF to begin with
1472 * to figure out the corresponding kernel's
1473 * counter part. Thus, attr->btf_fd has
1474 * to be valid also.
1475 */
1476 attr->btf_vmlinux_value_type_id) {
1477 struct btf *btf;
1478
1479 btf = btf_get_by_fd(attr->btf_fd);
1480 if (IS_ERR(btf)) {
1481 err = PTR_ERR(btf);
1482 goto free_map;
1483 }
1484 if (btf_is_kernel(btf)) {
1485 btf_put(btf);
1486 err = -EACCES;
1487 goto free_map;
1488 }
1489 map->btf = btf;
1490
1491 if (attr->btf_value_type_id) {
1492 err = map_check_btf(map, token, btf, attr->btf_key_type_id,
1493 attr->btf_value_type_id);
1494 if (err)
1495 goto free_map;
1496 }
1497
1498 map->btf_key_type_id = attr->btf_key_type_id;
1499 map->btf_value_type_id = attr->btf_value_type_id;
1500 map->btf_vmlinux_value_type_id =
1501 attr->btf_vmlinux_value_type_id;
1502 }
1503
1504 err = security_bpf_map_create(map, attr, token);
1505 if (err)
1506 goto free_map_sec;
1507
1508 err = bpf_map_alloc_id(map);
1509 if (err)
1510 goto free_map_sec;
1511
1512 bpf_map_save_memcg(map);
1513 bpf_token_put(token);
1514
1515 err = bpf_map_new_fd(map, f_flags);
1516 if (err < 0) {
1517 /* failed to allocate fd.
1518 * bpf_map_put_with_uref() is needed because the above
1519 * bpf_map_alloc_id() has published the map
1520 * to the userspace and the userspace may
1521 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1522 */
1523 bpf_map_put_with_uref(map);
1524 return err;
1525 }
1526
1527 return err;
1528
1529free_map_sec:
1530 security_bpf_map_free(map);
1531free_map:
1532 bpf_map_free(map);
1533put_token:
1534 bpf_token_put(token);
1535 return err;
1536}
1537
1538void bpf_map_inc(struct bpf_map *map)
1539{
1540 atomic64_inc(&map->refcnt);
1541}
1542EXPORT_SYMBOL_GPL(bpf_map_inc);
1543
1544void bpf_map_inc_with_uref(struct bpf_map *map)
1545{
1546 atomic64_inc(&map->refcnt);
1547 atomic64_inc(&map->usercnt);
1548}
1549EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1550
1551struct bpf_map *bpf_map_get(u32 ufd)
1552{
1553 CLASS(fd, f)(ufd);
1554 struct bpf_map *map = __bpf_map_get(f);
1555
1556 if (!IS_ERR(map))
1557 bpf_map_inc(map);
1558
1559 return map;
1560}
1561EXPORT_SYMBOL(bpf_map_get);
1562
1563struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1564{
1565 CLASS(fd, f)(ufd);
1566 struct bpf_map *map = __bpf_map_get(f);
1567
1568 if (!IS_ERR(map))
1569 bpf_map_inc_with_uref(map);
1570
1571 return map;
1572}
1573
1574/* map_idr_lock should have been held or the map should have been
1575 * protected by rcu read lock.
1576 */
1577struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1578{
1579 int refold;
1580
1581 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1582 if (!refold)
1583 return ERR_PTR(-ENOENT);
1584 if (uref)
1585 atomic64_inc(&map->usercnt);
1586
1587 return map;
1588}
1589
1590struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1591{
1592 spin_lock_bh(&map_idr_lock);
1593 map = __bpf_map_inc_not_zero(map, false);
1594 spin_unlock_bh(&map_idr_lock);
1595
1596 return map;
1597}
1598EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1599
1600int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1601{
1602 return -ENOTSUPP;
1603}
1604
1605static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1606{
1607 if (key_size)
1608 return vmemdup_user(ukey, key_size);
1609
1610 if (ukey)
1611 return ERR_PTR(-EINVAL);
1612
1613 return NULL;
1614}
1615
1616static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1617{
1618 if (key_size)
1619 return kvmemdup_bpfptr(ukey, key_size);
1620
1621 if (!bpfptr_is_null(ukey))
1622 return ERR_PTR(-EINVAL);
1623
1624 return NULL;
1625}
1626
1627/* last field in 'union bpf_attr' used by this command */
1628#define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1629
1630static int map_lookup_elem(union bpf_attr *attr)
1631{
1632 void __user *ukey = u64_to_user_ptr(attr->key);
1633 void __user *uvalue = u64_to_user_ptr(attr->value);
1634 struct bpf_map *map;
1635 void *key, *value;
1636 u32 value_size;
1637 int err;
1638
1639 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1640 return -EINVAL;
1641
1642 if (attr->flags & ~BPF_F_LOCK)
1643 return -EINVAL;
1644
1645 CLASS(fd, f)(attr->map_fd);
1646 map = __bpf_map_get(f);
1647 if (IS_ERR(map))
1648 return PTR_ERR(map);
1649 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ))
1650 return -EPERM;
1651
1652 if ((attr->flags & BPF_F_LOCK) &&
1653 !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1654 return -EINVAL;
1655
1656 key = __bpf_copy_key(ukey, map->key_size);
1657 if (IS_ERR(key))
1658 return PTR_ERR(key);
1659
1660 value_size = bpf_map_value_size(map);
1661
1662 err = -ENOMEM;
1663 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1664 if (!value)
1665 goto free_key;
1666
1667 if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1668 if (copy_from_user(value, uvalue, value_size))
1669 err = -EFAULT;
1670 else
1671 err = bpf_map_copy_value(map, key, value, attr->flags);
1672 goto free_value;
1673 }
1674
1675 err = bpf_map_copy_value(map, key, value, attr->flags);
1676 if (err)
1677 goto free_value;
1678
1679 err = -EFAULT;
1680 if (copy_to_user(uvalue, value, value_size) != 0)
1681 goto free_value;
1682
1683 err = 0;
1684
1685free_value:
1686 kvfree(value);
1687free_key:
1688 kvfree(key);
1689 return err;
1690}
1691
1692
1693#define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1694
1695static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1696{
1697 bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1698 bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1699 struct bpf_map *map;
1700 void *key, *value;
1701 u32 value_size;
1702 int err;
1703
1704 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1705 return -EINVAL;
1706
1707 CLASS(fd, f)(attr->map_fd);
1708 map = __bpf_map_get(f);
1709 if (IS_ERR(map))
1710 return PTR_ERR(map);
1711 bpf_map_write_active_inc(map);
1712 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1713 err = -EPERM;
1714 goto err_put;
1715 }
1716
1717 if ((attr->flags & BPF_F_LOCK) &&
1718 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1719 err = -EINVAL;
1720 goto err_put;
1721 }
1722
1723 key = ___bpf_copy_key(ukey, map->key_size);
1724 if (IS_ERR(key)) {
1725 err = PTR_ERR(key);
1726 goto err_put;
1727 }
1728
1729 value_size = bpf_map_value_size(map);
1730 value = kvmemdup_bpfptr(uvalue, value_size);
1731 if (IS_ERR(value)) {
1732 err = PTR_ERR(value);
1733 goto free_key;
1734 }
1735
1736 err = bpf_map_update_value(map, fd_file(f), key, value, attr->flags);
1737 if (!err)
1738 maybe_wait_bpf_programs(map);
1739
1740 kvfree(value);
1741free_key:
1742 kvfree(key);
1743err_put:
1744 bpf_map_write_active_dec(map);
1745 return err;
1746}
1747
1748#define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1749
1750static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1751{
1752 bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1753 struct bpf_map *map;
1754 void *key;
1755 int err;
1756
1757 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1758 return -EINVAL;
1759
1760 CLASS(fd, f)(attr->map_fd);
1761 map = __bpf_map_get(f);
1762 if (IS_ERR(map))
1763 return PTR_ERR(map);
1764 bpf_map_write_active_inc(map);
1765 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1766 err = -EPERM;
1767 goto err_put;
1768 }
1769
1770 key = ___bpf_copy_key(ukey, map->key_size);
1771 if (IS_ERR(key)) {
1772 err = PTR_ERR(key);
1773 goto err_put;
1774 }
1775
1776 if (bpf_map_is_offloaded(map)) {
1777 err = bpf_map_offload_delete_elem(map, key);
1778 goto out;
1779 } else if (IS_FD_PROG_ARRAY(map) ||
1780 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1781 /* These maps require sleepable context */
1782 err = map->ops->map_delete_elem(map, key);
1783 goto out;
1784 }
1785
1786 bpf_disable_instrumentation();
1787 rcu_read_lock();
1788 err = map->ops->map_delete_elem(map, key);
1789 rcu_read_unlock();
1790 bpf_enable_instrumentation();
1791 if (!err)
1792 maybe_wait_bpf_programs(map);
1793out:
1794 kvfree(key);
1795err_put:
1796 bpf_map_write_active_dec(map);
1797 return err;
1798}
1799
1800/* last field in 'union bpf_attr' used by this command */
1801#define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1802
1803static int map_get_next_key(union bpf_attr *attr)
1804{
1805 void __user *ukey = u64_to_user_ptr(attr->key);
1806 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1807 struct bpf_map *map;
1808 void *key, *next_key;
1809 int err;
1810
1811 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1812 return -EINVAL;
1813
1814 CLASS(fd, f)(attr->map_fd);
1815 map = __bpf_map_get(f);
1816 if (IS_ERR(map))
1817 return PTR_ERR(map);
1818 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ))
1819 return -EPERM;
1820
1821 if (ukey) {
1822 key = __bpf_copy_key(ukey, map->key_size);
1823 if (IS_ERR(key))
1824 return PTR_ERR(key);
1825 } else {
1826 key = NULL;
1827 }
1828
1829 err = -ENOMEM;
1830 next_key = kvmalloc(map->key_size, GFP_USER);
1831 if (!next_key)
1832 goto free_key;
1833
1834 if (bpf_map_is_offloaded(map)) {
1835 err = bpf_map_offload_get_next_key(map, key, next_key);
1836 goto out;
1837 }
1838
1839 rcu_read_lock();
1840 err = map->ops->map_get_next_key(map, key, next_key);
1841 rcu_read_unlock();
1842out:
1843 if (err)
1844 goto free_next_key;
1845
1846 err = -EFAULT;
1847 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1848 goto free_next_key;
1849
1850 err = 0;
1851
1852free_next_key:
1853 kvfree(next_key);
1854free_key:
1855 kvfree(key);
1856 return err;
1857}
1858
1859int generic_map_delete_batch(struct bpf_map *map,
1860 const union bpf_attr *attr,
1861 union bpf_attr __user *uattr)
1862{
1863 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1864 u32 cp, max_count;
1865 int err = 0;
1866 void *key;
1867
1868 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1869 return -EINVAL;
1870
1871 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1872 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1873 return -EINVAL;
1874 }
1875
1876 max_count = attr->batch.count;
1877 if (!max_count)
1878 return 0;
1879
1880 if (put_user(0, &uattr->batch.count))
1881 return -EFAULT;
1882
1883 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1884 if (!key)
1885 return -ENOMEM;
1886
1887 for (cp = 0; cp < max_count; cp++) {
1888 err = -EFAULT;
1889 if (copy_from_user(key, keys + cp * map->key_size,
1890 map->key_size))
1891 break;
1892
1893 if (bpf_map_is_offloaded(map)) {
1894 err = bpf_map_offload_delete_elem(map, key);
1895 break;
1896 }
1897
1898 bpf_disable_instrumentation();
1899 rcu_read_lock();
1900 err = map->ops->map_delete_elem(map, key);
1901 rcu_read_unlock();
1902 bpf_enable_instrumentation();
1903 if (err)
1904 break;
1905 cond_resched();
1906 }
1907 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1908 err = -EFAULT;
1909
1910 kvfree(key);
1911
1912 return err;
1913}
1914
1915int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1916 const union bpf_attr *attr,
1917 union bpf_attr __user *uattr)
1918{
1919 void __user *values = u64_to_user_ptr(attr->batch.values);
1920 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1921 u32 value_size, cp, max_count;
1922 void *key, *value;
1923 int err = 0;
1924
1925 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1926 return -EINVAL;
1927
1928 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1929 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1930 return -EINVAL;
1931 }
1932
1933 value_size = bpf_map_value_size(map);
1934
1935 max_count = attr->batch.count;
1936 if (!max_count)
1937 return 0;
1938
1939 if (put_user(0, &uattr->batch.count))
1940 return -EFAULT;
1941
1942 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1943 if (!key)
1944 return -ENOMEM;
1945
1946 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1947 if (!value) {
1948 kvfree(key);
1949 return -ENOMEM;
1950 }
1951
1952 for (cp = 0; cp < max_count; cp++) {
1953 err = -EFAULT;
1954 if (copy_from_user(key, keys + cp * map->key_size,
1955 map->key_size) ||
1956 copy_from_user(value, values + cp * value_size, value_size))
1957 break;
1958
1959 err = bpf_map_update_value(map, map_file, key, value,
1960 attr->batch.elem_flags);
1961
1962 if (err)
1963 break;
1964 cond_resched();
1965 }
1966
1967 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1968 err = -EFAULT;
1969
1970 kvfree(value);
1971 kvfree(key);
1972
1973 return err;
1974}
1975
1976int generic_map_lookup_batch(struct bpf_map *map,
1977 const union bpf_attr *attr,
1978 union bpf_attr __user *uattr)
1979{
1980 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1981 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1982 void __user *values = u64_to_user_ptr(attr->batch.values);
1983 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1984 void *buf, *buf_prevkey, *prev_key, *key, *value;
1985 u32 value_size, cp, max_count;
1986 int err;
1987
1988 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1989 return -EINVAL;
1990
1991 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1992 !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1993 return -EINVAL;
1994
1995 value_size = bpf_map_value_size(map);
1996
1997 max_count = attr->batch.count;
1998 if (!max_count)
1999 return 0;
2000
2001 if (put_user(0, &uattr->batch.count))
2002 return -EFAULT;
2003
2004 buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
2005 if (!buf_prevkey)
2006 return -ENOMEM;
2007
2008 buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
2009 if (!buf) {
2010 kvfree(buf_prevkey);
2011 return -ENOMEM;
2012 }
2013
2014 err = -EFAULT;
2015 prev_key = NULL;
2016 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
2017 goto free_buf;
2018 key = buf;
2019 value = key + map->key_size;
2020 if (ubatch)
2021 prev_key = buf_prevkey;
2022
2023 for (cp = 0; cp < max_count;) {
2024 rcu_read_lock();
2025 err = map->ops->map_get_next_key(map, prev_key, key);
2026 rcu_read_unlock();
2027 if (err)
2028 break;
2029 err = bpf_map_copy_value(map, key, value,
2030 attr->batch.elem_flags);
2031
2032 if (err == -ENOENT)
2033 goto next_key;
2034
2035 if (err)
2036 goto free_buf;
2037
2038 if (copy_to_user(keys + cp * map->key_size, key,
2039 map->key_size)) {
2040 err = -EFAULT;
2041 goto free_buf;
2042 }
2043 if (copy_to_user(values + cp * value_size, value, value_size)) {
2044 err = -EFAULT;
2045 goto free_buf;
2046 }
2047
2048 cp++;
2049next_key:
2050 if (!prev_key)
2051 prev_key = buf_prevkey;
2052
2053 swap(prev_key, key);
2054 cond_resched();
2055 }
2056
2057 if (err == -EFAULT)
2058 goto free_buf;
2059
2060 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
2061 (cp && copy_to_user(uobatch, prev_key, map->key_size))))
2062 err = -EFAULT;
2063
2064free_buf:
2065 kvfree(buf_prevkey);
2066 kvfree(buf);
2067 return err;
2068}
2069
2070#define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
2071
2072static int map_lookup_and_delete_elem(union bpf_attr *attr)
2073{
2074 void __user *ukey = u64_to_user_ptr(attr->key);
2075 void __user *uvalue = u64_to_user_ptr(attr->value);
2076 struct bpf_map *map;
2077 void *key, *value;
2078 u32 value_size;
2079 int err;
2080
2081 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
2082 return -EINVAL;
2083
2084 if (attr->flags & ~BPF_F_LOCK)
2085 return -EINVAL;
2086
2087 CLASS(fd, f)(attr->map_fd);
2088 map = __bpf_map_get(f);
2089 if (IS_ERR(map))
2090 return PTR_ERR(map);
2091 bpf_map_write_active_inc(map);
2092 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
2093 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
2094 err = -EPERM;
2095 goto err_put;
2096 }
2097
2098 if (attr->flags &&
2099 (map->map_type == BPF_MAP_TYPE_QUEUE ||
2100 map->map_type == BPF_MAP_TYPE_STACK)) {
2101 err = -EINVAL;
2102 goto err_put;
2103 }
2104
2105 if ((attr->flags & BPF_F_LOCK) &&
2106 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
2107 err = -EINVAL;
2108 goto err_put;
2109 }
2110
2111 key = __bpf_copy_key(ukey, map->key_size);
2112 if (IS_ERR(key)) {
2113 err = PTR_ERR(key);
2114 goto err_put;
2115 }
2116
2117 value_size = bpf_map_value_size(map);
2118
2119 err = -ENOMEM;
2120 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
2121 if (!value)
2122 goto free_key;
2123
2124 err = -ENOTSUPP;
2125 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
2126 map->map_type == BPF_MAP_TYPE_STACK) {
2127 err = map->ops->map_pop_elem(map, value);
2128 } else if (map->map_type == BPF_MAP_TYPE_HASH ||
2129 map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
2130 map->map_type == BPF_MAP_TYPE_LRU_HASH ||
2131 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
2132 if (!bpf_map_is_offloaded(map)) {
2133 bpf_disable_instrumentation();
2134 rcu_read_lock();
2135 err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
2136 rcu_read_unlock();
2137 bpf_enable_instrumentation();
2138 }
2139 }
2140
2141 if (err)
2142 goto free_value;
2143
2144 if (copy_to_user(uvalue, value, value_size) != 0) {
2145 err = -EFAULT;
2146 goto free_value;
2147 }
2148
2149 err = 0;
2150
2151free_value:
2152 kvfree(value);
2153free_key:
2154 kvfree(key);
2155err_put:
2156 bpf_map_write_active_dec(map);
2157 return err;
2158}
2159
2160#define BPF_MAP_FREEZE_LAST_FIELD map_fd
2161
2162static int map_freeze(const union bpf_attr *attr)
2163{
2164 int err = 0;
2165 struct bpf_map *map;
2166
2167 if (CHECK_ATTR(BPF_MAP_FREEZE))
2168 return -EINVAL;
2169
2170 CLASS(fd, f)(attr->map_fd);
2171 map = __bpf_map_get(f);
2172 if (IS_ERR(map))
2173 return PTR_ERR(map);
2174
2175 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record))
2176 return -ENOTSUPP;
2177
2178 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE))
2179 return -EPERM;
2180
2181 mutex_lock(&map->freeze_mutex);
2182 if (bpf_map_write_active(map)) {
2183 err = -EBUSY;
2184 goto err_put;
2185 }
2186 if (READ_ONCE(map->frozen)) {
2187 err = -EBUSY;
2188 goto err_put;
2189 }
2190
2191 WRITE_ONCE(map->frozen, true);
2192err_put:
2193 mutex_unlock(&map->freeze_mutex);
2194 return err;
2195}
2196
2197static const struct bpf_prog_ops * const bpf_prog_types[] = {
2198#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
2199 [_id] = & _name ## _prog_ops,
2200#define BPF_MAP_TYPE(_id, _ops)
2201#define BPF_LINK_TYPE(_id, _name)
2202#include <linux/bpf_types.h>
2203#undef BPF_PROG_TYPE
2204#undef BPF_MAP_TYPE
2205#undef BPF_LINK_TYPE
2206};
2207
2208static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
2209{
2210 const struct bpf_prog_ops *ops;
2211
2212 if (type >= ARRAY_SIZE(bpf_prog_types))
2213 return -EINVAL;
2214 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
2215 ops = bpf_prog_types[type];
2216 if (!ops)
2217 return -EINVAL;
2218
2219 if (!bpf_prog_is_offloaded(prog->aux))
2220 prog->aux->ops = ops;
2221 else
2222 prog->aux->ops = &bpf_offload_prog_ops;
2223 prog->type = type;
2224 return 0;
2225}
2226
2227enum bpf_audit {
2228 BPF_AUDIT_LOAD,
2229 BPF_AUDIT_UNLOAD,
2230 BPF_AUDIT_MAX,
2231};
2232
2233static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
2234 [BPF_AUDIT_LOAD] = "LOAD",
2235 [BPF_AUDIT_UNLOAD] = "UNLOAD",
2236};
2237
2238static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
2239{
2240 struct audit_context *ctx = NULL;
2241 struct audit_buffer *ab;
2242
2243 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2244 return;
2245 if (audit_enabled == AUDIT_OFF)
2246 return;
2247 if (!in_irq() && !irqs_disabled())
2248 ctx = audit_context();
2249 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2250 if (unlikely(!ab))
2251 return;
2252 audit_log_format(ab, "prog-id=%u op=%s",
2253 prog->aux->id, bpf_audit_str[op]);
2254 audit_log_end(ab);
2255}
2256
2257static int bpf_prog_alloc_id(struct bpf_prog *prog)
2258{
2259 int id;
2260
2261 idr_preload(GFP_KERNEL);
2262 spin_lock_bh(&prog_idr_lock);
2263 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2264 if (id > 0)
2265 prog->aux->id = id;
2266 spin_unlock_bh(&prog_idr_lock);
2267 idr_preload_end();
2268
2269 /* id is in [1, INT_MAX) */
2270 if (WARN_ON_ONCE(!id))
2271 return -ENOSPC;
2272
2273 return id > 0 ? 0 : id;
2274}
2275
2276void bpf_prog_free_id(struct bpf_prog *prog)
2277{
2278 unsigned long flags;
2279
2280 /* cBPF to eBPF migrations are currently not in the idr store.
2281 * Offloaded programs are removed from the store when their device
2282 * disappears - even if someone grabs an fd to them they are unusable,
2283 * simply waiting for refcnt to drop to be freed.
2284 */
2285 if (!prog->aux->id)
2286 return;
2287
2288 spin_lock_irqsave(&prog_idr_lock, flags);
2289 idr_remove(&prog_idr, prog->aux->id);
2290 prog->aux->id = 0;
2291 spin_unlock_irqrestore(&prog_idr_lock, flags);
2292}
2293
2294static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2295{
2296 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2297
2298 kvfree(aux->func_info);
2299 kfree(aux->func_info_aux);
2300 free_uid(aux->user);
2301 security_bpf_prog_free(aux->prog);
2302 bpf_prog_free(aux->prog);
2303}
2304
2305static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2306{
2307 bpf_prog_kallsyms_del_all(prog);
2308 btf_put(prog->aux->btf);
2309 module_put(prog->aux->mod);
2310 kvfree(prog->aux->jited_linfo);
2311 kvfree(prog->aux->linfo);
2312 kfree(prog->aux->kfunc_tab);
2313 if (prog->aux->attach_btf)
2314 btf_put(prog->aux->attach_btf);
2315
2316 if (deferred) {
2317 if (prog->sleepable)
2318 call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2319 else
2320 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2321 } else {
2322 __bpf_prog_put_rcu(&prog->aux->rcu);
2323 }
2324}
2325
2326static void bpf_prog_put_deferred(struct work_struct *work)
2327{
2328 struct bpf_prog_aux *aux;
2329 struct bpf_prog *prog;
2330
2331 aux = container_of(work, struct bpf_prog_aux, work);
2332 prog = aux->prog;
2333 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2334 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2335 bpf_prog_free_id(prog);
2336 __bpf_prog_put_noref(prog, true);
2337}
2338
2339static void __bpf_prog_put(struct bpf_prog *prog)
2340{
2341 struct bpf_prog_aux *aux = prog->aux;
2342
2343 if (atomic64_dec_and_test(&aux->refcnt)) {
2344 if (in_irq() || irqs_disabled()) {
2345 INIT_WORK(&aux->work, bpf_prog_put_deferred);
2346 schedule_work(&aux->work);
2347 } else {
2348 bpf_prog_put_deferred(&aux->work);
2349 }
2350 }
2351}
2352
2353void bpf_prog_put(struct bpf_prog *prog)
2354{
2355 __bpf_prog_put(prog);
2356}
2357EXPORT_SYMBOL_GPL(bpf_prog_put);
2358
2359static int bpf_prog_release(struct inode *inode, struct file *filp)
2360{
2361 struct bpf_prog *prog = filp->private_data;
2362
2363 bpf_prog_put(prog);
2364 return 0;
2365}
2366
2367struct bpf_prog_kstats {
2368 u64 nsecs;
2369 u64 cnt;
2370 u64 misses;
2371};
2372
2373void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2374{
2375 struct bpf_prog_stats *stats;
2376 unsigned int flags;
2377
2378 stats = this_cpu_ptr(prog->stats);
2379 flags = u64_stats_update_begin_irqsave(&stats->syncp);
2380 u64_stats_inc(&stats->misses);
2381 u64_stats_update_end_irqrestore(&stats->syncp, flags);
2382}
2383
2384static void bpf_prog_get_stats(const struct bpf_prog *prog,
2385 struct bpf_prog_kstats *stats)
2386{
2387 u64 nsecs = 0, cnt = 0, misses = 0;
2388 int cpu;
2389
2390 for_each_possible_cpu(cpu) {
2391 const struct bpf_prog_stats *st;
2392 unsigned int start;
2393 u64 tnsecs, tcnt, tmisses;
2394
2395 st = per_cpu_ptr(prog->stats, cpu);
2396 do {
2397 start = u64_stats_fetch_begin(&st->syncp);
2398 tnsecs = u64_stats_read(&st->nsecs);
2399 tcnt = u64_stats_read(&st->cnt);
2400 tmisses = u64_stats_read(&st->misses);
2401 } while (u64_stats_fetch_retry(&st->syncp, start));
2402 nsecs += tnsecs;
2403 cnt += tcnt;
2404 misses += tmisses;
2405 }
2406 stats->nsecs = nsecs;
2407 stats->cnt = cnt;
2408 stats->misses = misses;
2409}
2410
2411#ifdef CONFIG_PROC_FS
2412static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2413{
2414 const struct bpf_prog *prog = filp->private_data;
2415 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2416 struct bpf_prog_kstats stats;
2417
2418 bpf_prog_get_stats(prog, &stats);
2419 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2420 seq_printf(m,
2421 "prog_type:\t%u\n"
2422 "prog_jited:\t%u\n"
2423 "prog_tag:\t%s\n"
2424 "memlock:\t%llu\n"
2425 "prog_id:\t%u\n"
2426 "run_time_ns:\t%llu\n"
2427 "run_cnt:\t%llu\n"
2428 "recursion_misses:\t%llu\n"
2429 "verified_insns:\t%u\n",
2430 prog->type,
2431 prog->jited,
2432 prog_tag,
2433 prog->pages * 1ULL << PAGE_SHIFT,
2434 prog->aux->id,
2435 stats.nsecs,
2436 stats.cnt,
2437 stats.misses,
2438 prog->aux->verified_insns);
2439}
2440#endif
2441
2442const struct file_operations bpf_prog_fops = {
2443#ifdef CONFIG_PROC_FS
2444 .show_fdinfo = bpf_prog_show_fdinfo,
2445#endif
2446 .release = bpf_prog_release,
2447 .read = bpf_dummy_read,
2448 .write = bpf_dummy_write,
2449};
2450
2451int bpf_prog_new_fd(struct bpf_prog *prog)
2452{
2453 int ret;
2454
2455 ret = security_bpf_prog(prog);
2456 if (ret < 0)
2457 return ret;
2458
2459 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2460 O_RDWR | O_CLOEXEC);
2461}
2462
2463void bpf_prog_add(struct bpf_prog *prog, int i)
2464{
2465 atomic64_add(i, &prog->aux->refcnt);
2466}
2467EXPORT_SYMBOL_GPL(bpf_prog_add);
2468
2469void bpf_prog_sub(struct bpf_prog *prog, int i)
2470{
2471 /* Only to be used for undoing previous bpf_prog_add() in some
2472 * error path. We still know that another entity in our call
2473 * path holds a reference to the program, thus atomic_sub() can
2474 * be safely used in such cases!
2475 */
2476 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2477}
2478EXPORT_SYMBOL_GPL(bpf_prog_sub);
2479
2480void bpf_prog_inc(struct bpf_prog *prog)
2481{
2482 atomic64_inc(&prog->aux->refcnt);
2483}
2484EXPORT_SYMBOL_GPL(bpf_prog_inc);
2485
2486/* prog_idr_lock should have been held */
2487struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2488{
2489 int refold;
2490
2491 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2492
2493 if (!refold)
2494 return ERR_PTR(-ENOENT);
2495
2496 return prog;
2497}
2498EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2499
2500bool bpf_prog_get_ok(struct bpf_prog *prog,
2501 enum bpf_prog_type *attach_type, bool attach_drv)
2502{
2503 /* not an attachment, just a refcount inc, always allow */
2504 if (!attach_type)
2505 return true;
2506
2507 if (prog->type != *attach_type)
2508 return false;
2509 if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2510 return false;
2511
2512 return true;
2513}
2514
2515static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2516 bool attach_drv)
2517{
2518 CLASS(fd, f)(ufd);
2519 struct bpf_prog *prog;
2520
2521 if (fd_empty(f))
2522 return ERR_PTR(-EBADF);
2523 if (fd_file(f)->f_op != &bpf_prog_fops)
2524 return ERR_PTR(-EINVAL);
2525
2526 prog = fd_file(f)->private_data;
2527 if (!bpf_prog_get_ok(prog, attach_type, attach_drv))
2528 return ERR_PTR(-EINVAL);
2529
2530 bpf_prog_inc(prog);
2531 return prog;
2532}
2533
2534struct bpf_prog *bpf_prog_get(u32 ufd)
2535{
2536 return __bpf_prog_get(ufd, NULL, false);
2537}
2538
2539struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2540 bool attach_drv)
2541{
2542 return __bpf_prog_get(ufd, &type, attach_drv);
2543}
2544EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2545
2546/* Initially all BPF programs could be loaded w/o specifying
2547 * expected_attach_type. Later for some of them specifying expected_attach_type
2548 * at load time became required so that program could be validated properly.
2549 * Programs of types that are allowed to be loaded both w/ and w/o (for
2550 * backward compatibility) expected_attach_type, should have the default attach
2551 * type assigned to expected_attach_type for the latter case, so that it can be
2552 * validated later at attach time.
2553 *
2554 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2555 * prog type requires it but has some attach types that have to be backward
2556 * compatible.
2557 */
2558static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2559{
2560 switch (attr->prog_type) {
2561 case BPF_PROG_TYPE_CGROUP_SOCK:
2562 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2563 * exist so checking for non-zero is the way to go here.
2564 */
2565 if (!attr->expected_attach_type)
2566 attr->expected_attach_type =
2567 BPF_CGROUP_INET_SOCK_CREATE;
2568 break;
2569 case BPF_PROG_TYPE_SK_REUSEPORT:
2570 if (!attr->expected_attach_type)
2571 attr->expected_attach_type =
2572 BPF_SK_REUSEPORT_SELECT;
2573 break;
2574 }
2575}
2576
2577static int
2578bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2579 enum bpf_attach_type expected_attach_type,
2580 struct btf *attach_btf, u32 btf_id,
2581 struct bpf_prog *dst_prog)
2582{
2583 if (btf_id) {
2584 if (btf_id > BTF_MAX_TYPE)
2585 return -EINVAL;
2586
2587 if (!attach_btf && !dst_prog)
2588 return -EINVAL;
2589
2590 switch (prog_type) {
2591 case BPF_PROG_TYPE_TRACING:
2592 case BPF_PROG_TYPE_LSM:
2593 case BPF_PROG_TYPE_STRUCT_OPS:
2594 case BPF_PROG_TYPE_EXT:
2595 break;
2596 default:
2597 return -EINVAL;
2598 }
2599 }
2600
2601 if (attach_btf && (!btf_id || dst_prog))
2602 return -EINVAL;
2603
2604 if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2605 prog_type != BPF_PROG_TYPE_EXT)
2606 return -EINVAL;
2607
2608 switch (prog_type) {
2609 case BPF_PROG_TYPE_CGROUP_SOCK:
2610 switch (expected_attach_type) {
2611 case BPF_CGROUP_INET_SOCK_CREATE:
2612 case BPF_CGROUP_INET_SOCK_RELEASE:
2613 case BPF_CGROUP_INET4_POST_BIND:
2614 case BPF_CGROUP_INET6_POST_BIND:
2615 return 0;
2616 default:
2617 return -EINVAL;
2618 }
2619 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2620 switch (expected_attach_type) {
2621 case BPF_CGROUP_INET4_BIND:
2622 case BPF_CGROUP_INET6_BIND:
2623 case BPF_CGROUP_INET4_CONNECT:
2624 case BPF_CGROUP_INET6_CONNECT:
2625 case BPF_CGROUP_UNIX_CONNECT:
2626 case BPF_CGROUP_INET4_GETPEERNAME:
2627 case BPF_CGROUP_INET6_GETPEERNAME:
2628 case BPF_CGROUP_UNIX_GETPEERNAME:
2629 case BPF_CGROUP_INET4_GETSOCKNAME:
2630 case BPF_CGROUP_INET6_GETSOCKNAME:
2631 case BPF_CGROUP_UNIX_GETSOCKNAME:
2632 case BPF_CGROUP_UDP4_SENDMSG:
2633 case BPF_CGROUP_UDP6_SENDMSG:
2634 case BPF_CGROUP_UNIX_SENDMSG:
2635 case BPF_CGROUP_UDP4_RECVMSG:
2636 case BPF_CGROUP_UDP6_RECVMSG:
2637 case BPF_CGROUP_UNIX_RECVMSG:
2638 return 0;
2639 default:
2640 return -EINVAL;
2641 }
2642 case BPF_PROG_TYPE_CGROUP_SKB:
2643 switch (expected_attach_type) {
2644 case BPF_CGROUP_INET_INGRESS:
2645 case BPF_CGROUP_INET_EGRESS:
2646 return 0;
2647 default:
2648 return -EINVAL;
2649 }
2650 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2651 switch (expected_attach_type) {
2652 case BPF_CGROUP_SETSOCKOPT:
2653 case BPF_CGROUP_GETSOCKOPT:
2654 return 0;
2655 default:
2656 return -EINVAL;
2657 }
2658 case BPF_PROG_TYPE_SK_LOOKUP:
2659 if (expected_attach_type == BPF_SK_LOOKUP)
2660 return 0;
2661 return -EINVAL;
2662 case BPF_PROG_TYPE_SK_REUSEPORT:
2663 switch (expected_attach_type) {
2664 case BPF_SK_REUSEPORT_SELECT:
2665 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2666 return 0;
2667 default:
2668 return -EINVAL;
2669 }
2670 case BPF_PROG_TYPE_NETFILTER:
2671 if (expected_attach_type == BPF_NETFILTER)
2672 return 0;
2673 return -EINVAL;
2674 case BPF_PROG_TYPE_SYSCALL:
2675 case BPF_PROG_TYPE_EXT:
2676 if (expected_attach_type)
2677 return -EINVAL;
2678 fallthrough;
2679 default:
2680 return 0;
2681 }
2682}
2683
2684static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2685{
2686 switch (prog_type) {
2687 case BPF_PROG_TYPE_SCHED_CLS:
2688 case BPF_PROG_TYPE_SCHED_ACT:
2689 case BPF_PROG_TYPE_XDP:
2690 case BPF_PROG_TYPE_LWT_IN:
2691 case BPF_PROG_TYPE_LWT_OUT:
2692 case BPF_PROG_TYPE_LWT_XMIT:
2693 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2694 case BPF_PROG_TYPE_SK_SKB:
2695 case BPF_PROG_TYPE_SK_MSG:
2696 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2697 case BPF_PROG_TYPE_CGROUP_DEVICE:
2698 case BPF_PROG_TYPE_CGROUP_SOCK:
2699 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2700 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2701 case BPF_PROG_TYPE_CGROUP_SYSCTL:
2702 case BPF_PROG_TYPE_SOCK_OPS:
2703 case BPF_PROG_TYPE_EXT: /* extends any prog */
2704 case BPF_PROG_TYPE_NETFILTER:
2705 return true;
2706 case BPF_PROG_TYPE_CGROUP_SKB:
2707 /* always unpriv */
2708 case BPF_PROG_TYPE_SK_REUSEPORT:
2709 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2710 default:
2711 return false;
2712 }
2713}
2714
2715static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2716{
2717 switch (prog_type) {
2718 case BPF_PROG_TYPE_KPROBE:
2719 case BPF_PROG_TYPE_TRACEPOINT:
2720 case BPF_PROG_TYPE_PERF_EVENT:
2721 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2722 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2723 case BPF_PROG_TYPE_TRACING:
2724 case BPF_PROG_TYPE_LSM:
2725 case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2726 case BPF_PROG_TYPE_EXT: /* extends any prog */
2727 return true;
2728 default:
2729 return false;
2730 }
2731}
2732
2733/* last field in 'union bpf_attr' used by this command */
2734#define BPF_PROG_LOAD_LAST_FIELD prog_token_fd
2735
2736static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
2737{
2738 enum bpf_prog_type type = attr->prog_type;
2739 struct bpf_prog *prog, *dst_prog = NULL;
2740 struct btf *attach_btf = NULL;
2741 struct bpf_token *token = NULL;
2742 bool bpf_cap;
2743 int err;
2744 char license[128];
2745
2746 if (CHECK_ATTR(BPF_PROG_LOAD))
2747 return -EINVAL;
2748
2749 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2750 BPF_F_ANY_ALIGNMENT |
2751 BPF_F_TEST_STATE_FREQ |
2752 BPF_F_SLEEPABLE |
2753 BPF_F_TEST_RND_HI32 |
2754 BPF_F_XDP_HAS_FRAGS |
2755 BPF_F_XDP_DEV_BOUND_ONLY |
2756 BPF_F_TEST_REG_INVARIANTS |
2757 BPF_F_TOKEN_FD))
2758 return -EINVAL;
2759
2760 bpf_prog_load_fixup_attach_type(attr);
2761
2762 if (attr->prog_flags & BPF_F_TOKEN_FD) {
2763 token = bpf_token_get_from_fd(attr->prog_token_fd);
2764 if (IS_ERR(token))
2765 return PTR_ERR(token);
2766 /* if current token doesn't grant prog loading permissions,
2767 * then we can't use this token, so ignore it and rely on
2768 * system-wide capabilities checks
2769 */
2770 if (!bpf_token_allow_cmd(token, BPF_PROG_LOAD) ||
2771 !bpf_token_allow_prog_type(token, attr->prog_type,
2772 attr->expected_attach_type)) {
2773 bpf_token_put(token);
2774 token = NULL;
2775 }
2776 }
2777
2778 bpf_cap = bpf_token_capable(token, CAP_BPF);
2779 err = -EPERM;
2780
2781 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2782 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2783 !bpf_cap)
2784 goto put_token;
2785
2786 /* Intent here is for unprivileged_bpf_disabled to block BPF program
2787 * creation for unprivileged users; other actions depend
2788 * on fd availability and access to bpffs, so are dependent on
2789 * object creation success. Even with unprivileged BPF disabled,
2790 * capability checks are still carried out for these
2791 * and other operations.
2792 */
2793 if (sysctl_unprivileged_bpf_disabled && !bpf_cap)
2794 goto put_token;
2795
2796 if (attr->insn_cnt == 0 ||
2797 attr->insn_cnt > (bpf_cap ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS)) {
2798 err = -E2BIG;
2799 goto put_token;
2800 }
2801 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2802 type != BPF_PROG_TYPE_CGROUP_SKB &&
2803 !bpf_cap)
2804 goto put_token;
2805
2806 if (is_net_admin_prog_type(type) && !bpf_token_capable(token, CAP_NET_ADMIN))
2807 goto put_token;
2808 if (is_perfmon_prog_type(type) && !bpf_token_capable(token, CAP_PERFMON))
2809 goto put_token;
2810
2811 /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2812 * or btf, we need to check which one it is
2813 */
2814 if (attr->attach_prog_fd) {
2815 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2816 if (IS_ERR(dst_prog)) {
2817 dst_prog = NULL;
2818 attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2819 if (IS_ERR(attach_btf)) {
2820 err = -EINVAL;
2821 goto put_token;
2822 }
2823 if (!btf_is_kernel(attach_btf)) {
2824 /* attaching through specifying bpf_prog's BTF
2825 * objects directly might be supported eventually
2826 */
2827 btf_put(attach_btf);
2828 err = -ENOTSUPP;
2829 goto put_token;
2830 }
2831 }
2832 } else if (attr->attach_btf_id) {
2833 /* fall back to vmlinux BTF, if BTF type ID is specified */
2834 attach_btf = bpf_get_btf_vmlinux();
2835 if (IS_ERR(attach_btf)) {
2836 err = PTR_ERR(attach_btf);
2837 goto put_token;
2838 }
2839 if (!attach_btf) {
2840 err = -EINVAL;
2841 goto put_token;
2842 }
2843 btf_get(attach_btf);
2844 }
2845
2846 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2847 attach_btf, attr->attach_btf_id,
2848 dst_prog)) {
2849 if (dst_prog)
2850 bpf_prog_put(dst_prog);
2851 if (attach_btf)
2852 btf_put(attach_btf);
2853 err = -EINVAL;
2854 goto put_token;
2855 }
2856
2857 /* plain bpf_prog allocation */
2858 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2859 if (!prog) {
2860 if (dst_prog)
2861 bpf_prog_put(dst_prog);
2862 if (attach_btf)
2863 btf_put(attach_btf);
2864 err = -EINVAL;
2865 goto put_token;
2866 }
2867
2868 prog->expected_attach_type = attr->expected_attach_type;
2869 prog->sleepable = !!(attr->prog_flags & BPF_F_SLEEPABLE);
2870 prog->aux->attach_btf = attach_btf;
2871 prog->aux->attach_btf_id = attr->attach_btf_id;
2872 prog->aux->dst_prog = dst_prog;
2873 prog->aux->dev_bound = !!attr->prog_ifindex;
2874 prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2875
2876 /* move token into prog->aux, reuse taken refcnt */
2877 prog->aux->token = token;
2878 token = NULL;
2879
2880 prog->aux->user = get_current_user();
2881 prog->len = attr->insn_cnt;
2882
2883 err = -EFAULT;
2884 if (copy_from_bpfptr(prog->insns,
2885 make_bpfptr(attr->insns, uattr.is_kernel),
2886 bpf_prog_insn_size(prog)) != 0)
2887 goto free_prog;
2888 /* copy eBPF program license from user space */
2889 if (strncpy_from_bpfptr(license,
2890 make_bpfptr(attr->license, uattr.is_kernel),
2891 sizeof(license) - 1) < 0)
2892 goto free_prog;
2893 license[sizeof(license) - 1] = 0;
2894
2895 /* eBPF programs must be GPL compatible to use GPL-ed functions */
2896 prog->gpl_compatible = license_is_gpl_compatible(license) ? 1 : 0;
2897
2898 prog->orig_prog = NULL;
2899 prog->jited = 0;
2900
2901 atomic64_set(&prog->aux->refcnt, 1);
2902
2903 if (bpf_prog_is_dev_bound(prog->aux)) {
2904 err = bpf_prog_dev_bound_init(prog, attr);
2905 if (err)
2906 goto free_prog;
2907 }
2908
2909 if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2910 bpf_prog_is_dev_bound(dst_prog->aux)) {
2911 err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2912 if (err)
2913 goto free_prog;
2914 }
2915
2916 /*
2917 * Bookkeeping for managing the program attachment chain.
2918 *
2919 * It might be tempting to set attach_tracing_prog flag at the attachment
2920 * time, but this will not prevent from loading bunch of tracing prog
2921 * first, then attach them one to another.
2922 *
2923 * The flag attach_tracing_prog is set for the whole program lifecycle, and
2924 * doesn't have to be cleared in bpf_tracing_link_release, since tracing
2925 * programs cannot change attachment target.
2926 */
2927 if (type == BPF_PROG_TYPE_TRACING && dst_prog &&
2928 dst_prog->type == BPF_PROG_TYPE_TRACING) {
2929 prog->aux->attach_tracing_prog = true;
2930 }
2931
2932 /* find program type: socket_filter vs tracing_filter */
2933 err = find_prog_type(type, prog);
2934 if (err < 0)
2935 goto free_prog;
2936
2937 prog->aux->load_time = ktime_get_boottime_ns();
2938 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2939 sizeof(attr->prog_name));
2940 if (err < 0)
2941 goto free_prog;
2942
2943 err = security_bpf_prog_load(prog, attr, token);
2944 if (err)
2945 goto free_prog_sec;
2946
2947 /* run eBPF verifier */
2948 err = bpf_check(&prog, attr, uattr, uattr_size);
2949 if (err < 0)
2950 goto free_used_maps;
2951
2952 prog = bpf_prog_select_runtime(prog, &err);
2953 if (err < 0)
2954 goto free_used_maps;
2955
2956 err = bpf_prog_alloc_id(prog);
2957 if (err)
2958 goto free_used_maps;
2959
2960 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2961 * effectively publicly exposed. However, retrieving via
2962 * bpf_prog_get_fd_by_id() will take another reference,
2963 * therefore it cannot be gone underneath us.
2964 *
2965 * Only for the time /after/ successful bpf_prog_new_fd()
2966 * and before returning to userspace, we might just hold
2967 * one reference and any parallel close on that fd could
2968 * rip everything out. Hence, below notifications must
2969 * happen before bpf_prog_new_fd().
2970 *
2971 * Also, any failure handling from this point onwards must
2972 * be using bpf_prog_put() given the program is exposed.
2973 */
2974 bpf_prog_kallsyms_add(prog);
2975 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2976 bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2977
2978 err = bpf_prog_new_fd(prog);
2979 if (err < 0)
2980 bpf_prog_put(prog);
2981 return err;
2982
2983free_used_maps:
2984 /* In case we have subprogs, we need to wait for a grace
2985 * period before we can tear down JIT memory since symbols
2986 * are already exposed under kallsyms.
2987 */
2988 __bpf_prog_put_noref(prog, prog->aux->real_func_cnt);
2989 return err;
2990
2991free_prog_sec:
2992 security_bpf_prog_free(prog);
2993free_prog:
2994 free_uid(prog->aux->user);
2995 if (prog->aux->attach_btf)
2996 btf_put(prog->aux->attach_btf);
2997 bpf_prog_free(prog);
2998put_token:
2999 bpf_token_put(token);
3000 return err;
3001}
3002
3003#define BPF_OBJ_LAST_FIELD path_fd
3004
3005static int bpf_obj_pin(const union bpf_attr *attr)
3006{
3007 int path_fd;
3008
3009 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags & ~BPF_F_PATH_FD)
3010 return -EINVAL;
3011
3012 /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
3013 if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
3014 return -EINVAL;
3015
3016 path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
3017 return bpf_obj_pin_user(attr->bpf_fd, path_fd,
3018 u64_to_user_ptr(attr->pathname));
3019}
3020
3021static int bpf_obj_get(const union bpf_attr *attr)
3022{
3023 int path_fd;
3024
3025 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
3026 attr->file_flags & ~(BPF_OBJ_FLAG_MASK | BPF_F_PATH_FD))
3027 return -EINVAL;
3028
3029 /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
3030 if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
3031 return -EINVAL;
3032
3033 path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
3034 return bpf_obj_get_user(path_fd, u64_to_user_ptr(attr->pathname),
3035 attr->file_flags);
3036}
3037
3038/* bpf_link_init_sleepable() allows to specify whether BPF link itself has
3039 * "sleepable" semantics, which normally would mean that BPF link's attach
3040 * hook can dereference link or link's underlying program for some time after
3041 * detachment due to RCU Tasks Trace-based lifetime protection scheme.
3042 * BPF program itself can be non-sleepable, yet, because it's transitively
3043 * reachable through BPF link, its freeing has to be delayed until after RCU
3044 * Tasks Trace GP.
3045 */
3046void bpf_link_init_sleepable(struct bpf_link *link, enum bpf_link_type type,
3047 const struct bpf_link_ops *ops, struct bpf_prog *prog,
3048 bool sleepable)
3049{
3050 WARN_ON(ops->dealloc && ops->dealloc_deferred);
3051 atomic64_set(&link->refcnt, 1);
3052 link->type = type;
3053 link->sleepable = sleepable;
3054 link->id = 0;
3055 link->ops = ops;
3056 link->prog = prog;
3057}
3058
3059void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
3060 const struct bpf_link_ops *ops, struct bpf_prog *prog)
3061{
3062 bpf_link_init_sleepable(link, type, ops, prog, false);
3063}
3064
3065static void bpf_link_free_id(int id)
3066{
3067 if (!id)
3068 return;
3069
3070 spin_lock_bh(&link_idr_lock);
3071 idr_remove(&link_idr, id);
3072 spin_unlock_bh(&link_idr_lock);
3073}
3074
3075/* Clean up bpf_link and corresponding anon_inode file and FD. After
3076 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
3077 * anon_inode's release() call. This helper marks bpf_link as
3078 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
3079 * is not decremented, it's the responsibility of a calling code that failed
3080 * to complete bpf_link initialization.
3081 * This helper eventually calls link's dealloc callback, but does not call
3082 * link's release callback.
3083 */
3084void bpf_link_cleanup(struct bpf_link_primer *primer)
3085{
3086 primer->link->prog = NULL;
3087 bpf_link_free_id(primer->id);
3088 fput(primer->file);
3089 put_unused_fd(primer->fd);
3090}
3091
3092void bpf_link_inc(struct bpf_link *link)
3093{
3094 atomic64_inc(&link->refcnt);
3095}
3096
3097static void bpf_link_dealloc(struct bpf_link *link)
3098{
3099 /* now that we know that bpf_link itself can't be reached, put underlying BPF program */
3100 if (link->prog)
3101 bpf_prog_put(link->prog);
3102
3103 /* free bpf_link and its containing memory */
3104 if (link->ops->dealloc_deferred)
3105 link->ops->dealloc_deferred(link);
3106 else
3107 link->ops->dealloc(link);
3108}
3109
3110static void bpf_link_defer_dealloc_rcu_gp(struct rcu_head *rcu)
3111{
3112 struct bpf_link *link = container_of(rcu, struct bpf_link, rcu);
3113
3114 bpf_link_dealloc(link);
3115}
3116
3117static void bpf_link_defer_dealloc_mult_rcu_gp(struct rcu_head *rcu)
3118{
3119 if (rcu_trace_implies_rcu_gp())
3120 bpf_link_defer_dealloc_rcu_gp(rcu);
3121 else
3122 call_rcu(rcu, bpf_link_defer_dealloc_rcu_gp);
3123}
3124
3125/* bpf_link_free is guaranteed to be called from process context */
3126static void bpf_link_free(struct bpf_link *link)
3127{
3128 const struct bpf_link_ops *ops = link->ops;
3129
3130 bpf_link_free_id(link->id);
3131 /* detach BPF program, clean up used resources */
3132 if (link->prog)
3133 ops->release(link);
3134 if (ops->dealloc_deferred) {
3135 /* Schedule BPF link deallocation, which will only then
3136 * trigger putting BPF program refcount.
3137 * If underlying BPF program is sleepable or BPF link's target
3138 * attach hookpoint is sleepable or otherwise requires RCU GPs
3139 * to ensure link and its underlying BPF program is not
3140 * reachable anymore, we need to first wait for RCU tasks
3141 * trace sync, and then go through "classic" RCU grace period
3142 */
3143 if (link->sleepable || (link->prog && link->prog->sleepable))
3144 call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp);
3145 else
3146 call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
3147 } else if (ops->dealloc) {
3148 bpf_link_dealloc(link);
3149 }
3150}
3151
3152static void bpf_link_put_deferred(struct work_struct *work)
3153{
3154 struct bpf_link *link = container_of(work, struct bpf_link, work);
3155
3156 bpf_link_free(link);
3157}
3158
3159/* bpf_link_put might be called from atomic context. It needs to be called
3160 * from sleepable context in order to acquire sleeping locks during the process.
3161 */
3162void bpf_link_put(struct bpf_link *link)
3163{
3164 if (!atomic64_dec_and_test(&link->refcnt))
3165 return;
3166
3167 INIT_WORK(&link->work, bpf_link_put_deferred);
3168 schedule_work(&link->work);
3169}
3170EXPORT_SYMBOL(bpf_link_put);
3171
3172static void bpf_link_put_direct(struct bpf_link *link)
3173{
3174 if (!atomic64_dec_and_test(&link->refcnt))
3175 return;
3176 bpf_link_free(link);
3177}
3178
3179static int bpf_link_release(struct inode *inode, struct file *filp)
3180{
3181 struct bpf_link *link = filp->private_data;
3182
3183 bpf_link_put_direct(link);
3184 return 0;
3185}
3186
3187#ifdef CONFIG_PROC_FS
3188#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
3189#define BPF_MAP_TYPE(_id, _ops)
3190#define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
3191static const char *bpf_link_type_strs[] = {
3192 [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
3193#include <linux/bpf_types.h>
3194};
3195#undef BPF_PROG_TYPE
3196#undef BPF_MAP_TYPE
3197#undef BPF_LINK_TYPE
3198
3199static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
3200{
3201 const struct bpf_link *link = filp->private_data;
3202 const struct bpf_prog *prog = link->prog;
3203 enum bpf_link_type type = link->type;
3204 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
3205
3206 if (type < ARRAY_SIZE(bpf_link_type_strs) && bpf_link_type_strs[type]) {
3207 seq_printf(m, "link_type:\t%s\n", bpf_link_type_strs[type]);
3208 } else {
3209 WARN_ONCE(1, "missing BPF_LINK_TYPE(...) for link type %u\n", type);
3210 seq_printf(m, "link_type:\t<%u>\n", type);
3211 }
3212 seq_printf(m, "link_id:\t%u\n", link->id);
3213
3214 if (prog) {
3215 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
3216 seq_printf(m,
3217 "prog_tag:\t%s\n"
3218 "prog_id:\t%u\n",
3219 prog_tag,
3220 prog->aux->id);
3221 }
3222 if (link->ops->show_fdinfo)
3223 link->ops->show_fdinfo(link, m);
3224}
3225#endif
3226
3227static __poll_t bpf_link_poll(struct file *file, struct poll_table_struct *pts)
3228{
3229 struct bpf_link *link = file->private_data;
3230
3231 return link->ops->poll(file, pts);
3232}
3233
3234static const struct file_operations bpf_link_fops = {
3235#ifdef CONFIG_PROC_FS
3236 .show_fdinfo = bpf_link_show_fdinfo,
3237#endif
3238 .release = bpf_link_release,
3239 .read = bpf_dummy_read,
3240 .write = bpf_dummy_write,
3241};
3242
3243static const struct file_operations bpf_link_fops_poll = {
3244#ifdef CONFIG_PROC_FS
3245 .show_fdinfo = bpf_link_show_fdinfo,
3246#endif
3247 .release = bpf_link_release,
3248 .read = bpf_dummy_read,
3249 .write = bpf_dummy_write,
3250 .poll = bpf_link_poll,
3251};
3252
3253static int bpf_link_alloc_id(struct bpf_link *link)
3254{
3255 int id;
3256
3257 idr_preload(GFP_KERNEL);
3258 spin_lock_bh(&link_idr_lock);
3259 id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
3260 spin_unlock_bh(&link_idr_lock);
3261 idr_preload_end();
3262
3263 return id;
3264}
3265
3266/* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
3267 * reserving unused FD and allocating ID from link_idr. This is to be paired
3268 * with bpf_link_settle() to install FD and ID and expose bpf_link to
3269 * user-space, if bpf_link is successfully attached. If not, bpf_link and
3270 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
3271 * transient state is passed around in struct bpf_link_primer.
3272 * This is preferred way to create and initialize bpf_link, especially when
3273 * there are complicated and expensive operations in between creating bpf_link
3274 * itself and attaching it to BPF hook. By using bpf_link_prime() and
3275 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
3276 * expensive (and potentially failing) roll back operations in a rare case
3277 * that file, FD, or ID can't be allocated.
3278 */
3279int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
3280{
3281 struct file *file;
3282 int fd, id;
3283
3284 fd = get_unused_fd_flags(O_CLOEXEC);
3285 if (fd < 0)
3286 return fd;
3287
3288
3289 id = bpf_link_alloc_id(link);
3290 if (id < 0) {
3291 put_unused_fd(fd);
3292 return id;
3293 }
3294
3295 file = anon_inode_getfile("bpf_link",
3296 link->ops->poll ? &bpf_link_fops_poll : &bpf_link_fops,
3297 link, O_CLOEXEC);
3298 if (IS_ERR(file)) {
3299 bpf_link_free_id(id);
3300 put_unused_fd(fd);
3301 return PTR_ERR(file);
3302 }
3303
3304 primer->link = link;
3305 primer->file = file;
3306 primer->fd = fd;
3307 primer->id = id;
3308 return 0;
3309}
3310
3311int bpf_link_settle(struct bpf_link_primer *primer)
3312{
3313 /* make bpf_link fetchable by ID */
3314 spin_lock_bh(&link_idr_lock);
3315 primer->link->id = primer->id;
3316 spin_unlock_bh(&link_idr_lock);
3317 /* make bpf_link fetchable by FD */
3318 fd_install(primer->fd, primer->file);
3319 /* pass through installed FD */
3320 return primer->fd;
3321}
3322
3323int bpf_link_new_fd(struct bpf_link *link)
3324{
3325 return anon_inode_getfd("bpf-link",
3326 link->ops->poll ? &bpf_link_fops_poll : &bpf_link_fops,
3327 link, O_CLOEXEC);
3328}
3329
3330struct bpf_link *bpf_link_get_from_fd(u32 ufd)
3331{
3332 CLASS(fd, f)(ufd);
3333 struct bpf_link *link;
3334
3335 if (fd_empty(f))
3336 return ERR_PTR(-EBADF);
3337 if (fd_file(f)->f_op != &bpf_link_fops && fd_file(f)->f_op != &bpf_link_fops_poll)
3338 return ERR_PTR(-EINVAL);
3339
3340 link = fd_file(f)->private_data;
3341 bpf_link_inc(link);
3342 return link;
3343}
3344EXPORT_SYMBOL(bpf_link_get_from_fd);
3345
3346static void bpf_tracing_link_release(struct bpf_link *link)
3347{
3348 struct bpf_tracing_link *tr_link =
3349 container_of(link, struct bpf_tracing_link, link.link);
3350
3351 WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
3352 tr_link->trampoline,
3353 tr_link->tgt_prog));
3354
3355 bpf_trampoline_put(tr_link->trampoline);
3356
3357 /* tgt_prog is NULL if target is a kernel function */
3358 if (tr_link->tgt_prog)
3359 bpf_prog_put(tr_link->tgt_prog);
3360}
3361
3362static void bpf_tracing_link_dealloc(struct bpf_link *link)
3363{
3364 struct bpf_tracing_link *tr_link =
3365 container_of(link, struct bpf_tracing_link, link.link);
3366
3367 kfree(tr_link);
3368}
3369
3370static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
3371 struct seq_file *seq)
3372{
3373 struct bpf_tracing_link *tr_link =
3374 container_of(link, struct bpf_tracing_link, link.link);
3375 u32 target_btf_id, target_obj_id;
3376
3377 bpf_trampoline_unpack_key(tr_link->trampoline->key,
3378 &target_obj_id, &target_btf_id);
3379 seq_printf(seq,
3380 "attach_type:\t%d\n"
3381 "target_obj_id:\t%u\n"
3382 "target_btf_id:\t%u\n",
3383 tr_link->attach_type,
3384 target_obj_id,
3385 target_btf_id);
3386}
3387
3388static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
3389 struct bpf_link_info *info)
3390{
3391 struct bpf_tracing_link *tr_link =
3392 container_of(link, struct bpf_tracing_link, link.link);
3393
3394 info->tracing.attach_type = tr_link->attach_type;
3395 bpf_trampoline_unpack_key(tr_link->trampoline->key,
3396 &info->tracing.target_obj_id,
3397 &info->tracing.target_btf_id);
3398
3399 return 0;
3400}
3401
3402static const struct bpf_link_ops bpf_tracing_link_lops = {
3403 .release = bpf_tracing_link_release,
3404 .dealloc = bpf_tracing_link_dealloc,
3405 .show_fdinfo = bpf_tracing_link_show_fdinfo,
3406 .fill_link_info = bpf_tracing_link_fill_link_info,
3407};
3408
3409static int bpf_tracing_prog_attach(struct bpf_prog *prog,
3410 int tgt_prog_fd,
3411 u32 btf_id,
3412 u64 bpf_cookie)
3413{
3414 struct bpf_link_primer link_primer;
3415 struct bpf_prog *tgt_prog = NULL;
3416 struct bpf_trampoline *tr = NULL;
3417 struct bpf_tracing_link *link;
3418 u64 key = 0;
3419 int err;
3420
3421 switch (prog->type) {
3422 case BPF_PROG_TYPE_TRACING:
3423 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3424 prog->expected_attach_type != BPF_TRACE_FEXIT &&
3425 prog->expected_attach_type != BPF_MODIFY_RETURN) {
3426 err = -EINVAL;
3427 goto out_put_prog;
3428 }
3429 break;
3430 case BPF_PROG_TYPE_EXT:
3431 if (prog->expected_attach_type != 0) {
3432 err = -EINVAL;
3433 goto out_put_prog;
3434 }
3435 break;
3436 case BPF_PROG_TYPE_LSM:
3437 if (prog->expected_attach_type != BPF_LSM_MAC) {
3438 err = -EINVAL;
3439 goto out_put_prog;
3440 }
3441 break;
3442 default:
3443 err = -EINVAL;
3444 goto out_put_prog;
3445 }
3446
3447 if (!!tgt_prog_fd != !!btf_id) {
3448 err = -EINVAL;
3449 goto out_put_prog;
3450 }
3451
3452 if (tgt_prog_fd) {
3453 /*
3454 * For now we only allow new targets for BPF_PROG_TYPE_EXT. If this
3455 * part would be changed to implement the same for
3456 * BPF_PROG_TYPE_TRACING, do not forget to update the way how
3457 * attach_tracing_prog flag is set.
3458 */
3459 if (prog->type != BPF_PROG_TYPE_EXT) {
3460 err = -EINVAL;
3461 goto out_put_prog;
3462 }
3463
3464 tgt_prog = bpf_prog_get(tgt_prog_fd);
3465 if (IS_ERR(tgt_prog)) {
3466 err = PTR_ERR(tgt_prog);
3467 tgt_prog = NULL;
3468 goto out_put_prog;
3469 }
3470
3471 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3472 }
3473
3474 link = kzalloc(sizeof(*link), GFP_USER);
3475 if (!link) {
3476 err = -ENOMEM;
3477 goto out_put_prog;
3478 }
3479 bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3480 &bpf_tracing_link_lops, prog);
3481 link->attach_type = prog->expected_attach_type;
3482 link->link.cookie = bpf_cookie;
3483
3484 mutex_lock(&prog->aux->dst_mutex);
3485
3486 /* There are a few possible cases here:
3487 *
3488 * - if prog->aux->dst_trampoline is set, the program was just loaded
3489 * and not yet attached to anything, so we can use the values stored
3490 * in prog->aux
3491 *
3492 * - if prog->aux->dst_trampoline is NULL, the program has already been
3493 * attached to a target and its initial target was cleared (below)
3494 *
3495 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3496 * target_btf_id using the link_create API.
3497 *
3498 * - if tgt_prog == NULL when this function was called using the old
3499 * raw_tracepoint_open API, and we need a target from prog->aux
3500 *
3501 * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3502 * was detached and is going for re-attachment.
3503 *
3504 * - if prog->aux->dst_trampoline is NULL and tgt_prog and prog->aux->attach_btf
3505 * are NULL, then program was already attached and user did not provide
3506 * tgt_prog_fd so we have no way to find out or create trampoline
3507 */
3508 if (!prog->aux->dst_trampoline && !tgt_prog) {
3509 /*
3510 * Allow re-attach for TRACING and LSM programs. If it's
3511 * currently linked, bpf_trampoline_link_prog will fail.
3512 * EXT programs need to specify tgt_prog_fd, so they
3513 * re-attach in separate code path.
3514 */
3515 if (prog->type != BPF_PROG_TYPE_TRACING &&
3516 prog->type != BPF_PROG_TYPE_LSM) {
3517 err = -EINVAL;
3518 goto out_unlock;
3519 }
3520 /* We can allow re-attach only if we have valid attach_btf. */
3521 if (!prog->aux->attach_btf) {
3522 err = -EINVAL;
3523 goto out_unlock;
3524 }
3525 btf_id = prog->aux->attach_btf_id;
3526 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3527 }
3528
3529 if (!prog->aux->dst_trampoline ||
3530 (key && key != prog->aux->dst_trampoline->key)) {
3531 /* If there is no saved target, or the specified target is
3532 * different from the destination specified at load time, we
3533 * need a new trampoline and a check for compatibility
3534 */
3535 struct bpf_attach_target_info tgt_info = {};
3536
3537 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3538 &tgt_info);
3539 if (err)
3540 goto out_unlock;
3541
3542 if (tgt_info.tgt_mod) {
3543 module_put(prog->aux->mod);
3544 prog->aux->mod = tgt_info.tgt_mod;
3545 }
3546
3547 tr = bpf_trampoline_get(key, &tgt_info);
3548 if (!tr) {
3549 err = -ENOMEM;
3550 goto out_unlock;
3551 }
3552 } else {
3553 /* The caller didn't specify a target, or the target was the
3554 * same as the destination supplied during program load. This
3555 * means we can reuse the trampoline and reference from program
3556 * load time, and there is no need to allocate a new one. This
3557 * can only happen once for any program, as the saved values in
3558 * prog->aux are cleared below.
3559 */
3560 tr = prog->aux->dst_trampoline;
3561 tgt_prog = prog->aux->dst_prog;
3562 }
3563
3564 err = bpf_link_prime(&link->link.link, &link_primer);
3565 if (err)
3566 goto out_unlock;
3567
3568 err = bpf_trampoline_link_prog(&link->link, tr, tgt_prog);
3569 if (err) {
3570 bpf_link_cleanup(&link_primer);
3571 link = NULL;
3572 goto out_unlock;
3573 }
3574
3575 link->tgt_prog = tgt_prog;
3576 link->trampoline = tr;
3577
3578 /* Always clear the trampoline and target prog from prog->aux to make
3579 * sure the original attach destination is not kept alive after a
3580 * program is (re-)attached to another target.
3581 */
3582 if (prog->aux->dst_prog &&
3583 (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3584 /* got extra prog ref from syscall, or attaching to different prog */
3585 bpf_prog_put(prog->aux->dst_prog);
3586 if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3587 /* we allocated a new trampoline, so free the old one */
3588 bpf_trampoline_put(prog->aux->dst_trampoline);
3589
3590 prog->aux->dst_prog = NULL;
3591 prog->aux->dst_trampoline = NULL;
3592 mutex_unlock(&prog->aux->dst_mutex);
3593
3594 return bpf_link_settle(&link_primer);
3595out_unlock:
3596 if (tr && tr != prog->aux->dst_trampoline)
3597 bpf_trampoline_put(tr);
3598 mutex_unlock(&prog->aux->dst_mutex);
3599 kfree(link);
3600out_put_prog:
3601 if (tgt_prog_fd && tgt_prog)
3602 bpf_prog_put(tgt_prog);
3603 return err;
3604}
3605
3606static void bpf_raw_tp_link_release(struct bpf_link *link)
3607{
3608 struct bpf_raw_tp_link *raw_tp =
3609 container_of(link, struct bpf_raw_tp_link, link);
3610
3611 bpf_probe_unregister(raw_tp->btp, raw_tp);
3612 bpf_put_raw_tracepoint(raw_tp->btp);
3613}
3614
3615static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3616{
3617 struct bpf_raw_tp_link *raw_tp =
3618 container_of(link, struct bpf_raw_tp_link, link);
3619
3620 kfree(raw_tp);
3621}
3622
3623static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3624 struct seq_file *seq)
3625{
3626 struct bpf_raw_tp_link *raw_tp_link =
3627 container_of(link, struct bpf_raw_tp_link, link);
3628
3629 seq_printf(seq,
3630 "tp_name:\t%s\n",
3631 raw_tp_link->btp->tp->name);
3632}
3633
3634static int bpf_copy_to_user(char __user *ubuf, const char *buf, u32 ulen,
3635 u32 len)
3636{
3637 if (ulen >= len + 1) {
3638 if (copy_to_user(ubuf, buf, len + 1))
3639 return -EFAULT;
3640 } else {
3641 char zero = '\0';
3642
3643 if (copy_to_user(ubuf, buf, ulen - 1))
3644 return -EFAULT;
3645 if (put_user(zero, ubuf + ulen - 1))
3646 return -EFAULT;
3647 return -ENOSPC;
3648 }
3649
3650 return 0;
3651}
3652
3653static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3654 struct bpf_link_info *info)
3655{
3656 struct bpf_raw_tp_link *raw_tp_link =
3657 container_of(link, struct bpf_raw_tp_link, link);
3658 char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3659 const char *tp_name = raw_tp_link->btp->tp->name;
3660 u32 ulen = info->raw_tracepoint.tp_name_len;
3661 size_t tp_len = strlen(tp_name);
3662
3663 if (!ulen ^ !ubuf)
3664 return -EINVAL;
3665
3666 info->raw_tracepoint.tp_name_len = tp_len + 1;
3667
3668 if (!ubuf)
3669 return 0;
3670
3671 return bpf_copy_to_user(ubuf, tp_name, ulen, tp_len);
3672}
3673
3674static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3675 .release = bpf_raw_tp_link_release,
3676 .dealloc_deferred = bpf_raw_tp_link_dealloc,
3677 .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3678 .fill_link_info = bpf_raw_tp_link_fill_link_info,
3679};
3680
3681#ifdef CONFIG_PERF_EVENTS
3682struct bpf_perf_link {
3683 struct bpf_link link;
3684 struct file *perf_file;
3685};
3686
3687static void bpf_perf_link_release(struct bpf_link *link)
3688{
3689 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3690 struct perf_event *event = perf_link->perf_file->private_data;
3691
3692 perf_event_free_bpf_prog(event);
3693 fput(perf_link->perf_file);
3694}
3695
3696static void bpf_perf_link_dealloc(struct bpf_link *link)
3697{
3698 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3699
3700 kfree(perf_link);
3701}
3702
3703static int bpf_perf_link_fill_common(const struct perf_event *event,
3704 char __user *uname, u32 *ulenp,
3705 u64 *probe_offset, u64 *probe_addr,
3706 u32 *fd_type, unsigned long *missed)
3707{
3708 const char *buf;
3709 u32 prog_id, ulen;
3710 size_t len;
3711 int err;
3712
3713 ulen = *ulenp;
3714 if (!ulen ^ !uname)
3715 return -EINVAL;
3716
3717 err = bpf_get_perf_event_info(event, &prog_id, fd_type, &buf,
3718 probe_offset, probe_addr, missed);
3719 if (err)
3720 return err;
3721
3722 if (buf) {
3723 len = strlen(buf);
3724 *ulenp = len + 1;
3725 } else {
3726 *ulenp = 1;
3727 }
3728 if (!uname)
3729 return 0;
3730
3731 if (buf) {
3732 err = bpf_copy_to_user(uname, buf, ulen, len);
3733 if (err)
3734 return err;
3735 } else {
3736 char zero = '\0';
3737
3738 if (put_user(zero, uname))
3739 return -EFAULT;
3740 }
3741 return 0;
3742}
3743
3744#ifdef CONFIG_KPROBE_EVENTS
3745static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
3746 struct bpf_link_info *info)
3747{
3748 unsigned long missed;
3749 char __user *uname;
3750 u64 addr, offset;
3751 u32 ulen, type;
3752 int err;
3753
3754 uname = u64_to_user_ptr(info->perf_event.kprobe.func_name);
3755 ulen = info->perf_event.kprobe.name_len;
3756 err = bpf_perf_link_fill_common(event, uname, &ulen, &offset, &addr,
3757 &type, &missed);
3758 if (err)
3759 return err;
3760 if (type == BPF_FD_TYPE_KRETPROBE)
3761 info->perf_event.type = BPF_PERF_EVENT_KRETPROBE;
3762 else
3763 info->perf_event.type = BPF_PERF_EVENT_KPROBE;
3764 info->perf_event.kprobe.name_len = ulen;
3765 info->perf_event.kprobe.offset = offset;
3766 info->perf_event.kprobe.missed = missed;
3767 if (!kallsyms_show_value(current_cred()))
3768 addr = 0;
3769 info->perf_event.kprobe.addr = addr;
3770 info->perf_event.kprobe.cookie = event->bpf_cookie;
3771 return 0;
3772}
3773#endif
3774
3775#ifdef CONFIG_UPROBE_EVENTS
3776static int bpf_perf_link_fill_uprobe(const struct perf_event *event,
3777 struct bpf_link_info *info)
3778{
3779 char __user *uname;
3780 u64 addr, offset;
3781 u32 ulen, type;
3782 int err;
3783
3784 uname = u64_to_user_ptr(info->perf_event.uprobe.file_name);
3785 ulen = info->perf_event.uprobe.name_len;
3786 err = bpf_perf_link_fill_common(event, uname, &ulen, &offset, &addr,
3787 &type, NULL);
3788 if (err)
3789 return err;
3790
3791 if (type == BPF_FD_TYPE_URETPROBE)
3792 info->perf_event.type = BPF_PERF_EVENT_URETPROBE;
3793 else
3794 info->perf_event.type = BPF_PERF_EVENT_UPROBE;
3795 info->perf_event.uprobe.name_len = ulen;
3796 info->perf_event.uprobe.offset = offset;
3797 info->perf_event.uprobe.cookie = event->bpf_cookie;
3798 return 0;
3799}
3800#endif
3801
3802static int bpf_perf_link_fill_probe(const struct perf_event *event,
3803 struct bpf_link_info *info)
3804{
3805#ifdef CONFIG_KPROBE_EVENTS
3806 if (event->tp_event->flags & TRACE_EVENT_FL_KPROBE)
3807 return bpf_perf_link_fill_kprobe(event, info);
3808#endif
3809#ifdef CONFIG_UPROBE_EVENTS
3810 if (event->tp_event->flags & TRACE_EVENT_FL_UPROBE)
3811 return bpf_perf_link_fill_uprobe(event, info);
3812#endif
3813 return -EOPNOTSUPP;
3814}
3815
3816static int bpf_perf_link_fill_tracepoint(const struct perf_event *event,
3817 struct bpf_link_info *info)
3818{
3819 char __user *uname;
3820 u32 ulen;
3821 int err;
3822
3823 uname = u64_to_user_ptr(info->perf_event.tracepoint.tp_name);
3824 ulen = info->perf_event.tracepoint.name_len;
3825 err = bpf_perf_link_fill_common(event, uname, &ulen, NULL, NULL, NULL, NULL);
3826 if (err)
3827 return err;
3828
3829 info->perf_event.type = BPF_PERF_EVENT_TRACEPOINT;
3830 info->perf_event.tracepoint.name_len = ulen;
3831 info->perf_event.tracepoint.cookie = event->bpf_cookie;
3832 return 0;
3833}
3834
3835static int bpf_perf_link_fill_perf_event(const struct perf_event *event,
3836 struct bpf_link_info *info)
3837{
3838 info->perf_event.event.type = event->attr.type;
3839 info->perf_event.event.config = event->attr.config;
3840 info->perf_event.event.cookie = event->bpf_cookie;
3841 info->perf_event.type = BPF_PERF_EVENT_EVENT;
3842 return 0;
3843}
3844
3845static int bpf_perf_link_fill_link_info(const struct bpf_link *link,
3846 struct bpf_link_info *info)
3847{
3848 struct bpf_perf_link *perf_link;
3849 const struct perf_event *event;
3850
3851 perf_link = container_of(link, struct bpf_perf_link, link);
3852 event = perf_get_event(perf_link->perf_file);
3853 if (IS_ERR(event))
3854 return PTR_ERR(event);
3855
3856 switch (event->prog->type) {
3857 case BPF_PROG_TYPE_PERF_EVENT:
3858 return bpf_perf_link_fill_perf_event(event, info);
3859 case BPF_PROG_TYPE_TRACEPOINT:
3860 return bpf_perf_link_fill_tracepoint(event, info);
3861 case BPF_PROG_TYPE_KPROBE:
3862 return bpf_perf_link_fill_probe(event, info);
3863 default:
3864 return -EOPNOTSUPP;
3865 }
3866}
3867
3868static const struct bpf_link_ops bpf_perf_link_lops = {
3869 .release = bpf_perf_link_release,
3870 .dealloc = bpf_perf_link_dealloc,
3871 .fill_link_info = bpf_perf_link_fill_link_info,
3872};
3873
3874static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3875{
3876 struct bpf_link_primer link_primer;
3877 struct bpf_perf_link *link;
3878 struct perf_event *event;
3879 struct file *perf_file;
3880 int err;
3881
3882 if (attr->link_create.flags)
3883 return -EINVAL;
3884
3885 perf_file = perf_event_get(attr->link_create.target_fd);
3886 if (IS_ERR(perf_file))
3887 return PTR_ERR(perf_file);
3888
3889 link = kzalloc(sizeof(*link), GFP_USER);
3890 if (!link) {
3891 err = -ENOMEM;
3892 goto out_put_file;
3893 }
3894 bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3895 link->perf_file = perf_file;
3896
3897 err = bpf_link_prime(&link->link, &link_primer);
3898 if (err) {
3899 kfree(link);
3900 goto out_put_file;
3901 }
3902
3903 event = perf_file->private_data;
3904 err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3905 if (err) {
3906 bpf_link_cleanup(&link_primer);
3907 goto out_put_file;
3908 }
3909 /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3910 bpf_prog_inc(prog);
3911
3912 return bpf_link_settle(&link_primer);
3913
3914out_put_file:
3915 fput(perf_file);
3916 return err;
3917}
3918#else
3919static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3920{
3921 return -EOPNOTSUPP;
3922}
3923#endif /* CONFIG_PERF_EVENTS */
3924
3925static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3926 const char __user *user_tp_name, u64 cookie)
3927{
3928 struct bpf_link_primer link_primer;
3929 struct bpf_raw_tp_link *link;
3930 struct bpf_raw_event_map *btp;
3931 const char *tp_name;
3932 char buf[128];
3933 int err;
3934
3935 switch (prog->type) {
3936 case BPF_PROG_TYPE_TRACING:
3937 case BPF_PROG_TYPE_EXT:
3938 case BPF_PROG_TYPE_LSM:
3939 if (user_tp_name)
3940 /* The attach point for this category of programs
3941 * should be specified via btf_id during program load.
3942 */
3943 return -EINVAL;
3944 if (prog->type == BPF_PROG_TYPE_TRACING &&
3945 prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3946 tp_name = prog->aux->attach_func_name;
3947 break;
3948 }
3949 return bpf_tracing_prog_attach(prog, 0, 0, 0);
3950 case BPF_PROG_TYPE_RAW_TRACEPOINT:
3951 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3952 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3953 return -EFAULT;
3954 buf[sizeof(buf) - 1] = 0;
3955 tp_name = buf;
3956 break;
3957 default:
3958 return -EINVAL;
3959 }
3960
3961 btp = bpf_get_raw_tracepoint(tp_name);
3962 if (!btp)
3963 return -ENOENT;
3964
3965 link = kzalloc(sizeof(*link), GFP_USER);
3966 if (!link) {
3967 err = -ENOMEM;
3968 goto out_put_btp;
3969 }
3970 bpf_link_init_sleepable(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3971 &bpf_raw_tp_link_lops, prog,
3972 tracepoint_is_faultable(btp->tp));
3973 link->btp = btp;
3974 link->cookie = cookie;
3975
3976 err = bpf_link_prime(&link->link, &link_primer);
3977 if (err) {
3978 kfree(link);
3979 goto out_put_btp;
3980 }
3981
3982 err = bpf_probe_register(link->btp, link);
3983 if (err) {
3984 bpf_link_cleanup(&link_primer);
3985 goto out_put_btp;
3986 }
3987
3988 return bpf_link_settle(&link_primer);
3989
3990out_put_btp:
3991 bpf_put_raw_tracepoint(btp);
3992 return err;
3993}
3994
3995#define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.cookie
3996
3997static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3998{
3999 struct bpf_prog *prog;
4000 void __user *tp_name;
4001 __u64 cookie;
4002 int fd;
4003
4004 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
4005 return -EINVAL;
4006
4007 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
4008 if (IS_ERR(prog))
4009 return PTR_ERR(prog);
4010
4011 tp_name = u64_to_user_ptr(attr->raw_tracepoint.name);
4012 cookie = attr->raw_tracepoint.cookie;
4013 fd = bpf_raw_tp_link_attach(prog, tp_name, cookie);
4014 if (fd < 0)
4015 bpf_prog_put(prog);
4016 return fd;
4017}
4018
4019static enum bpf_prog_type
4020attach_type_to_prog_type(enum bpf_attach_type attach_type)
4021{
4022 switch (attach_type) {
4023 case BPF_CGROUP_INET_INGRESS:
4024 case BPF_CGROUP_INET_EGRESS:
4025 return BPF_PROG_TYPE_CGROUP_SKB;
4026 case BPF_CGROUP_INET_SOCK_CREATE:
4027 case BPF_CGROUP_INET_SOCK_RELEASE:
4028 case BPF_CGROUP_INET4_POST_BIND:
4029 case BPF_CGROUP_INET6_POST_BIND:
4030 return BPF_PROG_TYPE_CGROUP_SOCK;
4031 case BPF_CGROUP_INET4_BIND:
4032 case BPF_CGROUP_INET6_BIND:
4033 case BPF_CGROUP_INET4_CONNECT:
4034 case BPF_CGROUP_INET6_CONNECT:
4035 case BPF_CGROUP_UNIX_CONNECT:
4036 case BPF_CGROUP_INET4_GETPEERNAME:
4037 case BPF_CGROUP_INET6_GETPEERNAME:
4038 case BPF_CGROUP_UNIX_GETPEERNAME:
4039 case BPF_CGROUP_INET4_GETSOCKNAME:
4040 case BPF_CGROUP_INET6_GETSOCKNAME:
4041 case BPF_CGROUP_UNIX_GETSOCKNAME:
4042 case BPF_CGROUP_UDP4_SENDMSG:
4043 case BPF_CGROUP_UDP6_SENDMSG:
4044 case BPF_CGROUP_UNIX_SENDMSG:
4045 case BPF_CGROUP_UDP4_RECVMSG:
4046 case BPF_CGROUP_UDP6_RECVMSG:
4047 case BPF_CGROUP_UNIX_RECVMSG:
4048 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
4049 case BPF_CGROUP_SOCK_OPS:
4050 return BPF_PROG_TYPE_SOCK_OPS;
4051 case BPF_CGROUP_DEVICE:
4052 return BPF_PROG_TYPE_CGROUP_DEVICE;
4053 case BPF_SK_MSG_VERDICT:
4054 return BPF_PROG_TYPE_SK_MSG;
4055 case BPF_SK_SKB_STREAM_PARSER:
4056 case BPF_SK_SKB_STREAM_VERDICT:
4057 case BPF_SK_SKB_VERDICT:
4058 return BPF_PROG_TYPE_SK_SKB;
4059 case BPF_LIRC_MODE2:
4060 return BPF_PROG_TYPE_LIRC_MODE2;
4061 case BPF_FLOW_DISSECTOR:
4062 return BPF_PROG_TYPE_FLOW_DISSECTOR;
4063 case BPF_CGROUP_SYSCTL:
4064 return BPF_PROG_TYPE_CGROUP_SYSCTL;
4065 case BPF_CGROUP_GETSOCKOPT:
4066 case BPF_CGROUP_SETSOCKOPT:
4067 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
4068 case BPF_TRACE_ITER:
4069 case BPF_TRACE_RAW_TP:
4070 case BPF_TRACE_FENTRY:
4071 case BPF_TRACE_FEXIT:
4072 case BPF_MODIFY_RETURN:
4073 return BPF_PROG_TYPE_TRACING;
4074 case BPF_LSM_MAC:
4075 return BPF_PROG_TYPE_LSM;
4076 case BPF_SK_LOOKUP:
4077 return BPF_PROG_TYPE_SK_LOOKUP;
4078 case BPF_XDP:
4079 return BPF_PROG_TYPE_XDP;
4080 case BPF_LSM_CGROUP:
4081 return BPF_PROG_TYPE_LSM;
4082 case BPF_TCX_INGRESS:
4083 case BPF_TCX_EGRESS:
4084 case BPF_NETKIT_PRIMARY:
4085 case BPF_NETKIT_PEER:
4086 return BPF_PROG_TYPE_SCHED_CLS;
4087 default:
4088 return BPF_PROG_TYPE_UNSPEC;
4089 }
4090}
4091
4092static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
4093 enum bpf_attach_type attach_type)
4094{
4095 enum bpf_prog_type ptype;
4096
4097 switch (prog->type) {
4098 case BPF_PROG_TYPE_CGROUP_SOCK:
4099 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4100 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4101 case BPF_PROG_TYPE_SK_LOOKUP:
4102 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
4103 case BPF_PROG_TYPE_CGROUP_SKB:
4104 if (!bpf_token_capable(prog->aux->token, CAP_NET_ADMIN))
4105 /* cg-skb progs can be loaded by unpriv user.
4106 * check permissions at attach time.
4107 */
4108 return -EPERM;
4109
4110 ptype = attach_type_to_prog_type(attach_type);
4111 if (prog->type != ptype)
4112 return -EINVAL;
4113
4114 return prog->enforce_expected_attach_type &&
4115 prog->expected_attach_type != attach_type ?
4116 -EINVAL : 0;
4117 case BPF_PROG_TYPE_EXT:
4118 return 0;
4119 case BPF_PROG_TYPE_NETFILTER:
4120 if (attach_type != BPF_NETFILTER)
4121 return -EINVAL;
4122 return 0;
4123 case BPF_PROG_TYPE_PERF_EVENT:
4124 case BPF_PROG_TYPE_TRACEPOINT:
4125 if (attach_type != BPF_PERF_EVENT)
4126 return -EINVAL;
4127 return 0;
4128 case BPF_PROG_TYPE_KPROBE:
4129 if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI &&
4130 attach_type != BPF_TRACE_KPROBE_MULTI)
4131 return -EINVAL;
4132 if (prog->expected_attach_type == BPF_TRACE_KPROBE_SESSION &&
4133 attach_type != BPF_TRACE_KPROBE_SESSION)
4134 return -EINVAL;
4135 if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI &&
4136 attach_type != BPF_TRACE_UPROBE_MULTI)
4137 return -EINVAL;
4138 if (prog->expected_attach_type == BPF_TRACE_UPROBE_SESSION &&
4139 attach_type != BPF_TRACE_UPROBE_SESSION)
4140 return -EINVAL;
4141 if (attach_type != BPF_PERF_EVENT &&
4142 attach_type != BPF_TRACE_KPROBE_MULTI &&
4143 attach_type != BPF_TRACE_KPROBE_SESSION &&
4144 attach_type != BPF_TRACE_UPROBE_MULTI &&
4145 attach_type != BPF_TRACE_UPROBE_SESSION)
4146 return -EINVAL;
4147 return 0;
4148 case BPF_PROG_TYPE_SCHED_CLS:
4149 if (attach_type != BPF_TCX_INGRESS &&
4150 attach_type != BPF_TCX_EGRESS &&
4151 attach_type != BPF_NETKIT_PRIMARY &&
4152 attach_type != BPF_NETKIT_PEER)
4153 return -EINVAL;
4154 return 0;
4155 default:
4156 ptype = attach_type_to_prog_type(attach_type);
4157 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type)
4158 return -EINVAL;
4159 return 0;
4160 }
4161}
4162
4163#define BPF_PROG_ATTACH_LAST_FIELD expected_revision
4164
4165#define BPF_F_ATTACH_MASK_BASE \
4166 (BPF_F_ALLOW_OVERRIDE | \
4167 BPF_F_ALLOW_MULTI | \
4168 BPF_F_REPLACE)
4169
4170#define BPF_F_ATTACH_MASK_MPROG \
4171 (BPF_F_REPLACE | \
4172 BPF_F_BEFORE | \
4173 BPF_F_AFTER | \
4174 BPF_F_ID | \
4175 BPF_F_LINK)
4176
4177static int bpf_prog_attach(const union bpf_attr *attr)
4178{
4179 enum bpf_prog_type ptype;
4180 struct bpf_prog *prog;
4181 int ret;
4182
4183 if (CHECK_ATTR(BPF_PROG_ATTACH))
4184 return -EINVAL;
4185
4186 ptype = attach_type_to_prog_type(attr->attach_type);
4187 if (ptype == BPF_PROG_TYPE_UNSPEC)
4188 return -EINVAL;
4189 if (bpf_mprog_supported(ptype)) {
4190 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4191 return -EINVAL;
4192 } else {
4193 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_BASE)
4194 return -EINVAL;
4195 if (attr->relative_fd ||
4196 attr->expected_revision)
4197 return -EINVAL;
4198 }
4199
4200 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4201 if (IS_ERR(prog))
4202 return PTR_ERR(prog);
4203
4204 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
4205 bpf_prog_put(prog);
4206 return -EINVAL;
4207 }
4208
4209 switch (ptype) {
4210 case BPF_PROG_TYPE_SK_SKB:
4211 case BPF_PROG_TYPE_SK_MSG:
4212 ret = sock_map_get_from_fd(attr, prog);
4213 break;
4214 case BPF_PROG_TYPE_LIRC_MODE2:
4215 ret = lirc_prog_attach(attr, prog);
4216 break;
4217 case BPF_PROG_TYPE_FLOW_DISSECTOR:
4218 ret = netns_bpf_prog_attach(attr, prog);
4219 break;
4220 case BPF_PROG_TYPE_CGROUP_DEVICE:
4221 case BPF_PROG_TYPE_CGROUP_SKB:
4222 case BPF_PROG_TYPE_CGROUP_SOCK:
4223 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4224 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4225 case BPF_PROG_TYPE_CGROUP_SYSCTL:
4226 case BPF_PROG_TYPE_SOCK_OPS:
4227 case BPF_PROG_TYPE_LSM:
4228 if (ptype == BPF_PROG_TYPE_LSM &&
4229 prog->expected_attach_type != BPF_LSM_CGROUP)
4230 ret = -EINVAL;
4231 else
4232 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
4233 break;
4234 case BPF_PROG_TYPE_SCHED_CLS:
4235 if (attr->attach_type == BPF_TCX_INGRESS ||
4236 attr->attach_type == BPF_TCX_EGRESS)
4237 ret = tcx_prog_attach(attr, prog);
4238 else
4239 ret = netkit_prog_attach(attr, prog);
4240 break;
4241 default:
4242 ret = -EINVAL;
4243 }
4244
4245 if (ret)
4246 bpf_prog_put(prog);
4247 return ret;
4248}
4249
4250#define BPF_PROG_DETACH_LAST_FIELD expected_revision
4251
4252static int bpf_prog_detach(const union bpf_attr *attr)
4253{
4254 struct bpf_prog *prog = NULL;
4255 enum bpf_prog_type ptype;
4256 int ret;
4257
4258 if (CHECK_ATTR(BPF_PROG_DETACH))
4259 return -EINVAL;
4260
4261 ptype = attach_type_to_prog_type(attr->attach_type);
4262 if (bpf_mprog_supported(ptype)) {
4263 if (ptype == BPF_PROG_TYPE_UNSPEC)
4264 return -EINVAL;
4265 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4266 return -EINVAL;
4267 if (attr->attach_bpf_fd) {
4268 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4269 if (IS_ERR(prog))
4270 return PTR_ERR(prog);
4271 }
4272 } else if (attr->attach_flags ||
4273 attr->relative_fd ||
4274 attr->expected_revision) {
4275 return -EINVAL;
4276 }
4277
4278 switch (ptype) {
4279 case BPF_PROG_TYPE_SK_MSG:
4280 case BPF_PROG_TYPE_SK_SKB:
4281 ret = sock_map_prog_detach(attr, ptype);
4282 break;
4283 case BPF_PROG_TYPE_LIRC_MODE2:
4284 ret = lirc_prog_detach(attr);
4285 break;
4286 case BPF_PROG_TYPE_FLOW_DISSECTOR:
4287 ret = netns_bpf_prog_detach(attr, ptype);
4288 break;
4289 case BPF_PROG_TYPE_CGROUP_DEVICE:
4290 case BPF_PROG_TYPE_CGROUP_SKB:
4291 case BPF_PROG_TYPE_CGROUP_SOCK:
4292 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4293 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4294 case BPF_PROG_TYPE_CGROUP_SYSCTL:
4295 case BPF_PROG_TYPE_SOCK_OPS:
4296 case BPF_PROG_TYPE_LSM:
4297 ret = cgroup_bpf_prog_detach(attr, ptype);
4298 break;
4299 case BPF_PROG_TYPE_SCHED_CLS:
4300 if (attr->attach_type == BPF_TCX_INGRESS ||
4301 attr->attach_type == BPF_TCX_EGRESS)
4302 ret = tcx_prog_detach(attr, prog);
4303 else
4304 ret = netkit_prog_detach(attr, prog);
4305 break;
4306 default:
4307 ret = -EINVAL;
4308 }
4309
4310 if (prog)
4311 bpf_prog_put(prog);
4312 return ret;
4313}
4314
4315#define BPF_PROG_QUERY_LAST_FIELD query.revision
4316
4317static int bpf_prog_query(const union bpf_attr *attr,
4318 union bpf_attr __user *uattr)
4319{
4320 if (!bpf_net_capable())
4321 return -EPERM;
4322 if (CHECK_ATTR(BPF_PROG_QUERY))
4323 return -EINVAL;
4324 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
4325 return -EINVAL;
4326
4327 switch (attr->query.attach_type) {
4328 case BPF_CGROUP_INET_INGRESS:
4329 case BPF_CGROUP_INET_EGRESS:
4330 case BPF_CGROUP_INET_SOCK_CREATE:
4331 case BPF_CGROUP_INET_SOCK_RELEASE:
4332 case BPF_CGROUP_INET4_BIND:
4333 case BPF_CGROUP_INET6_BIND:
4334 case BPF_CGROUP_INET4_POST_BIND:
4335 case BPF_CGROUP_INET6_POST_BIND:
4336 case BPF_CGROUP_INET4_CONNECT:
4337 case BPF_CGROUP_INET6_CONNECT:
4338 case BPF_CGROUP_UNIX_CONNECT:
4339 case BPF_CGROUP_INET4_GETPEERNAME:
4340 case BPF_CGROUP_INET6_GETPEERNAME:
4341 case BPF_CGROUP_UNIX_GETPEERNAME:
4342 case BPF_CGROUP_INET4_GETSOCKNAME:
4343 case BPF_CGROUP_INET6_GETSOCKNAME:
4344 case BPF_CGROUP_UNIX_GETSOCKNAME:
4345 case BPF_CGROUP_UDP4_SENDMSG:
4346 case BPF_CGROUP_UDP6_SENDMSG:
4347 case BPF_CGROUP_UNIX_SENDMSG:
4348 case BPF_CGROUP_UDP4_RECVMSG:
4349 case BPF_CGROUP_UDP6_RECVMSG:
4350 case BPF_CGROUP_UNIX_RECVMSG:
4351 case BPF_CGROUP_SOCK_OPS:
4352 case BPF_CGROUP_DEVICE:
4353 case BPF_CGROUP_SYSCTL:
4354 case BPF_CGROUP_GETSOCKOPT:
4355 case BPF_CGROUP_SETSOCKOPT:
4356 case BPF_LSM_CGROUP:
4357 return cgroup_bpf_prog_query(attr, uattr);
4358 case BPF_LIRC_MODE2:
4359 return lirc_prog_query(attr, uattr);
4360 case BPF_FLOW_DISSECTOR:
4361 case BPF_SK_LOOKUP:
4362 return netns_bpf_prog_query(attr, uattr);
4363 case BPF_SK_SKB_STREAM_PARSER:
4364 case BPF_SK_SKB_STREAM_VERDICT:
4365 case BPF_SK_MSG_VERDICT:
4366 case BPF_SK_SKB_VERDICT:
4367 return sock_map_bpf_prog_query(attr, uattr);
4368 case BPF_TCX_INGRESS:
4369 case BPF_TCX_EGRESS:
4370 return tcx_prog_query(attr, uattr);
4371 case BPF_NETKIT_PRIMARY:
4372 case BPF_NETKIT_PEER:
4373 return netkit_prog_query(attr, uattr);
4374 default:
4375 return -EINVAL;
4376 }
4377}
4378
4379#define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
4380
4381static int bpf_prog_test_run(const union bpf_attr *attr,
4382 union bpf_attr __user *uattr)
4383{
4384 struct bpf_prog *prog;
4385 int ret = -ENOTSUPP;
4386
4387 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
4388 return -EINVAL;
4389
4390 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
4391 (!attr->test.ctx_size_in && attr->test.ctx_in))
4392 return -EINVAL;
4393
4394 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
4395 (!attr->test.ctx_size_out && attr->test.ctx_out))
4396 return -EINVAL;
4397
4398 prog = bpf_prog_get(attr->test.prog_fd);
4399 if (IS_ERR(prog))
4400 return PTR_ERR(prog);
4401
4402 if (prog->aux->ops->test_run)
4403 ret = prog->aux->ops->test_run(prog, attr, uattr);
4404
4405 bpf_prog_put(prog);
4406 return ret;
4407}
4408
4409#define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
4410
4411static int bpf_obj_get_next_id(const union bpf_attr *attr,
4412 union bpf_attr __user *uattr,
4413 struct idr *idr,
4414 spinlock_t *lock)
4415{
4416 u32 next_id = attr->start_id;
4417 int err = 0;
4418
4419 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
4420 return -EINVAL;
4421
4422 if (!capable(CAP_SYS_ADMIN))
4423 return -EPERM;
4424
4425 next_id++;
4426 spin_lock_bh(lock);
4427 if (!idr_get_next(idr, &next_id))
4428 err = -ENOENT;
4429 spin_unlock_bh(lock);
4430
4431 if (!err)
4432 err = put_user(next_id, &uattr->next_id);
4433
4434 return err;
4435}
4436
4437struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
4438{
4439 struct bpf_map *map;
4440
4441 spin_lock_bh(&map_idr_lock);
4442again:
4443 map = idr_get_next(&map_idr, id);
4444 if (map) {
4445 map = __bpf_map_inc_not_zero(map, false);
4446 if (IS_ERR(map)) {
4447 (*id)++;
4448 goto again;
4449 }
4450 }
4451 spin_unlock_bh(&map_idr_lock);
4452
4453 return map;
4454}
4455
4456struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
4457{
4458 struct bpf_prog *prog;
4459
4460 spin_lock_bh(&prog_idr_lock);
4461again:
4462 prog = idr_get_next(&prog_idr, id);
4463 if (prog) {
4464 prog = bpf_prog_inc_not_zero(prog);
4465 if (IS_ERR(prog)) {
4466 (*id)++;
4467 goto again;
4468 }
4469 }
4470 spin_unlock_bh(&prog_idr_lock);
4471
4472 return prog;
4473}
4474
4475#define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
4476
4477struct bpf_prog *bpf_prog_by_id(u32 id)
4478{
4479 struct bpf_prog *prog;
4480
4481 if (!id)
4482 return ERR_PTR(-ENOENT);
4483
4484 spin_lock_bh(&prog_idr_lock);
4485 prog = idr_find(&prog_idr, id);
4486 if (prog)
4487 prog = bpf_prog_inc_not_zero(prog);
4488 else
4489 prog = ERR_PTR(-ENOENT);
4490 spin_unlock_bh(&prog_idr_lock);
4491 return prog;
4492}
4493
4494static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
4495{
4496 struct bpf_prog *prog;
4497 u32 id = attr->prog_id;
4498 int fd;
4499
4500 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
4501 return -EINVAL;
4502
4503 if (!capable(CAP_SYS_ADMIN))
4504 return -EPERM;
4505
4506 prog = bpf_prog_by_id(id);
4507 if (IS_ERR(prog))
4508 return PTR_ERR(prog);
4509
4510 fd = bpf_prog_new_fd(prog);
4511 if (fd < 0)
4512 bpf_prog_put(prog);
4513
4514 return fd;
4515}
4516
4517#define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
4518
4519static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
4520{
4521 struct bpf_map *map;
4522 u32 id = attr->map_id;
4523 int f_flags;
4524 int fd;
4525
4526 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
4527 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
4528 return -EINVAL;
4529
4530 if (!capable(CAP_SYS_ADMIN))
4531 return -EPERM;
4532
4533 f_flags = bpf_get_file_flag(attr->open_flags);
4534 if (f_flags < 0)
4535 return f_flags;
4536
4537 spin_lock_bh(&map_idr_lock);
4538 map = idr_find(&map_idr, id);
4539 if (map)
4540 map = __bpf_map_inc_not_zero(map, true);
4541 else
4542 map = ERR_PTR(-ENOENT);
4543 spin_unlock_bh(&map_idr_lock);
4544
4545 if (IS_ERR(map))
4546 return PTR_ERR(map);
4547
4548 fd = bpf_map_new_fd(map, f_flags);
4549 if (fd < 0)
4550 bpf_map_put_with_uref(map);
4551
4552 return fd;
4553}
4554
4555static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
4556 unsigned long addr, u32 *off,
4557 u32 *type)
4558{
4559 const struct bpf_map *map;
4560 int i;
4561
4562 mutex_lock(&prog->aux->used_maps_mutex);
4563 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
4564 map = prog->aux->used_maps[i];
4565 if (map == (void *)addr) {
4566 *type = BPF_PSEUDO_MAP_FD;
4567 goto out;
4568 }
4569 if (!map->ops->map_direct_value_meta)
4570 continue;
4571 if (!map->ops->map_direct_value_meta(map, addr, off)) {
4572 *type = BPF_PSEUDO_MAP_VALUE;
4573 goto out;
4574 }
4575 }
4576 map = NULL;
4577
4578out:
4579 mutex_unlock(&prog->aux->used_maps_mutex);
4580 return map;
4581}
4582
4583static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
4584 const struct cred *f_cred)
4585{
4586 const struct bpf_map *map;
4587 struct bpf_insn *insns;
4588 u32 off, type;
4589 u64 imm;
4590 u8 code;
4591 int i;
4592
4593 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
4594 GFP_USER);
4595 if (!insns)
4596 return insns;
4597
4598 for (i = 0; i < prog->len; i++) {
4599 code = insns[i].code;
4600
4601 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
4602 insns[i].code = BPF_JMP | BPF_CALL;
4603 insns[i].imm = BPF_FUNC_tail_call;
4604 /* fall-through */
4605 }
4606 if (code == (BPF_JMP | BPF_CALL) ||
4607 code == (BPF_JMP | BPF_CALL_ARGS)) {
4608 if (code == (BPF_JMP | BPF_CALL_ARGS))
4609 insns[i].code = BPF_JMP | BPF_CALL;
4610 if (!bpf_dump_raw_ok(f_cred))
4611 insns[i].imm = 0;
4612 continue;
4613 }
4614 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
4615 insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
4616 continue;
4617 }
4618
4619 if ((BPF_CLASS(code) == BPF_LDX || BPF_CLASS(code) == BPF_STX ||
4620 BPF_CLASS(code) == BPF_ST) && BPF_MODE(code) == BPF_PROBE_MEM32) {
4621 insns[i].code = BPF_CLASS(code) | BPF_SIZE(code) | BPF_MEM;
4622 continue;
4623 }
4624
4625 if (code != (BPF_LD | BPF_IMM | BPF_DW))
4626 continue;
4627
4628 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
4629 map = bpf_map_from_imm(prog, imm, &off, &type);
4630 if (map) {
4631 insns[i].src_reg = type;
4632 insns[i].imm = map->id;
4633 insns[i + 1].imm = off;
4634 continue;
4635 }
4636 }
4637
4638 return insns;
4639}
4640
4641static int set_info_rec_size(struct bpf_prog_info *info)
4642{
4643 /*
4644 * Ensure info.*_rec_size is the same as kernel expected size
4645 *
4646 * or
4647 *
4648 * Only allow zero *_rec_size if both _rec_size and _cnt are
4649 * zero. In this case, the kernel will set the expected
4650 * _rec_size back to the info.
4651 */
4652
4653 if ((info->nr_func_info || info->func_info_rec_size) &&
4654 info->func_info_rec_size != sizeof(struct bpf_func_info))
4655 return -EINVAL;
4656
4657 if ((info->nr_line_info || info->line_info_rec_size) &&
4658 info->line_info_rec_size != sizeof(struct bpf_line_info))
4659 return -EINVAL;
4660
4661 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
4662 info->jited_line_info_rec_size != sizeof(__u64))
4663 return -EINVAL;
4664
4665 info->func_info_rec_size = sizeof(struct bpf_func_info);
4666 info->line_info_rec_size = sizeof(struct bpf_line_info);
4667 info->jited_line_info_rec_size = sizeof(__u64);
4668
4669 return 0;
4670}
4671
4672static int bpf_prog_get_info_by_fd(struct file *file,
4673 struct bpf_prog *prog,
4674 const union bpf_attr *attr,
4675 union bpf_attr __user *uattr)
4676{
4677 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4678 struct btf *attach_btf = bpf_prog_get_target_btf(prog);
4679 struct bpf_prog_info info;
4680 u32 info_len = attr->info.info_len;
4681 struct bpf_prog_kstats stats;
4682 char __user *uinsns;
4683 u32 ulen;
4684 int err;
4685
4686 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4687 if (err)
4688 return err;
4689 info_len = min_t(u32, sizeof(info), info_len);
4690
4691 memset(&info, 0, sizeof(info));
4692 if (copy_from_user(&info, uinfo, info_len))
4693 return -EFAULT;
4694
4695 info.type = prog->type;
4696 info.id = prog->aux->id;
4697 info.load_time = prog->aux->load_time;
4698 info.created_by_uid = from_kuid_munged(current_user_ns(),
4699 prog->aux->user->uid);
4700 info.gpl_compatible = prog->gpl_compatible;
4701
4702 memcpy(info.tag, prog->tag, sizeof(prog->tag));
4703 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
4704
4705 mutex_lock(&prog->aux->used_maps_mutex);
4706 ulen = info.nr_map_ids;
4707 info.nr_map_ids = prog->aux->used_map_cnt;
4708 ulen = min_t(u32, info.nr_map_ids, ulen);
4709 if (ulen) {
4710 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
4711 u32 i;
4712
4713 for (i = 0; i < ulen; i++)
4714 if (put_user(prog->aux->used_maps[i]->id,
4715 &user_map_ids[i])) {
4716 mutex_unlock(&prog->aux->used_maps_mutex);
4717 return -EFAULT;
4718 }
4719 }
4720 mutex_unlock(&prog->aux->used_maps_mutex);
4721
4722 err = set_info_rec_size(&info);
4723 if (err)
4724 return err;
4725
4726 bpf_prog_get_stats(prog, &stats);
4727 info.run_time_ns = stats.nsecs;
4728 info.run_cnt = stats.cnt;
4729 info.recursion_misses = stats.misses;
4730
4731 info.verified_insns = prog->aux->verified_insns;
4732
4733 if (!bpf_capable()) {
4734 info.jited_prog_len = 0;
4735 info.xlated_prog_len = 0;
4736 info.nr_jited_ksyms = 0;
4737 info.nr_jited_func_lens = 0;
4738 info.nr_func_info = 0;
4739 info.nr_line_info = 0;
4740 info.nr_jited_line_info = 0;
4741 goto done;
4742 }
4743
4744 ulen = info.xlated_prog_len;
4745 info.xlated_prog_len = bpf_prog_insn_size(prog);
4746 if (info.xlated_prog_len && ulen) {
4747 struct bpf_insn *insns_sanitized;
4748 bool fault;
4749
4750 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
4751 info.xlated_prog_insns = 0;
4752 goto done;
4753 }
4754 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4755 if (!insns_sanitized)
4756 return -ENOMEM;
4757 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4758 ulen = min_t(u32, info.xlated_prog_len, ulen);
4759 fault = copy_to_user(uinsns, insns_sanitized, ulen);
4760 kfree(insns_sanitized);
4761 if (fault)
4762 return -EFAULT;
4763 }
4764
4765 if (bpf_prog_is_offloaded(prog->aux)) {
4766 err = bpf_prog_offload_info_fill(&info, prog);
4767 if (err)
4768 return err;
4769 goto done;
4770 }
4771
4772 /* NOTE: the following code is supposed to be skipped for offload.
4773 * bpf_prog_offload_info_fill() is the place to fill similar fields
4774 * for offload.
4775 */
4776 ulen = info.jited_prog_len;
4777 if (prog->aux->func_cnt) {
4778 u32 i;
4779
4780 info.jited_prog_len = 0;
4781 for (i = 0; i < prog->aux->func_cnt; i++)
4782 info.jited_prog_len += prog->aux->func[i]->jited_len;
4783 } else {
4784 info.jited_prog_len = prog->jited_len;
4785 }
4786
4787 if (info.jited_prog_len && ulen) {
4788 if (bpf_dump_raw_ok(file->f_cred)) {
4789 uinsns = u64_to_user_ptr(info.jited_prog_insns);
4790 ulen = min_t(u32, info.jited_prog_len, ulen);
4791
4792 /* for multi-function programs, copy the JITed
4793 * instructions for all the functions
4794 */
4795 if (prog->aux->func_cnt) {
4796 u32 len, free, i;
4797 u8 *img;
4798
4799 free = ulen;
4800 for (i = 0; i < prog->aux->func_cnt; i++) {
4801 len = prog->aux->func[i]->jited_len;
4802 len = min_t(u32, len, free);
4803 img = (u8 *) prog->aux->func[i]->bpf_func;
4804 if (copy_to_user(uinsns, img, len))
4805 return -EFAULT;
4806 uinsns += len;
4807 free -= len;
4808 if (!free)
4809 break;
4810 }
4811 } else {
4812 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4813 return -EFAULT;
4814 }
4815 } else {
4816 info.jited_prog_insns = 0;
4817 }
4818 }
4819
4820 ulen = info.nr_jited_ksyms;
4821 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4822 if (ulen) {
4823 if (bpf_dump_raw_ok(file->f_cred)) {
4824 unsigned long ksym_addr;
4825 u64 __user *user_ksyms;
4826 u32 i;
4827
4828 /* copy the address of the kernel symbol
4829 * corresponding to each function
4830 */
4831 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4832 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4833 if (prog->aux->func_cnt) {
4834 for (i = 0; i < ulen; i++) {
4835 ksym_addr = (unsigned long)
4836 prog->aux->func[i]->bpf_func;
4837 if (put_user((u64) ksym_addr,
4838 &user_ksyms[i]))
4839 return -EFAULT;
4840 }
4841 } else {
4842 ksym_addr = (unsigned long) prog->bpf_func;
4843 if (put_user((u64) ksym_addr, &user_ksyms[0]))
4844 return -EFAULT;
4845 }
4846 } else {
4847 info.jited_ksyms = 0;
4848 }
4849 }
4850
4851 ulen = info.nr_jited_func_lens;
4852 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4853 if (ulen) {
4854 if (bpf_dump_raw_ok(file->f_cred)) {
4855 u32 __user *user_lens;
4856 u32 func_len, i;
4857
4858 /* copy the JITed image lengths for each function */
4859 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4860 user_lens = u64_to_user_ptr(info.jited_func_lens);
4861 if (prog->aux->func_cnt) {
4862 for (i = 0; i < ulen; i++) {
4863 func_len =
4864 prog->aux->func[i]->jited_len;
4865 if (put_user(func_len, &user_lens[i]))
4866 return -EFAULT;
4867 }
4868 } else {
4869 func_len = prog->jited_len;
4870 if (put_user(func_len, &user_lens[0]))
4871 return -EFAULT;
4872 }
4873 } else {
4874 info.jited_func_lens = 0;
4875 }
4876 }
4877
4878 if (prog->aux->btf)
4879 info.btf_id = btf_obj_id(prog->aux->btf);
4880 info.attach_btf_id = prog->aux->attach_btf_id;
4881 if (attach_btf)
4882 info.attach_btf_obj_id = btf_obj_id(attach_btf);
4883
4884 ulen = info.nr_func_info;
4885 info.nr_func_info = prog->aux->func_info_cnt;
4886 if (info.nr_func_info && ulen) {
4887 char __user *user_finfo;
4888
4889 user_finfo = u64_to_user_ptr(info.func_info);
4890 ulen = min_t(u32, info.nr_func_info, ulen);
4891 if (copy_to_user(user_finfo, prog->aux->func_info,
4892 info.func_info_rec_size * ulen))
4893 return -EFAULT;
4894 }
4895
4896 ulen = info.nr_line_info;
4897 info.nr_line_info = prog->aux->nr_linfo;
4898 if (info.nr_line_info && ulen) {
4899 __u8 __user *user_linfo;
4900
4901 user_linfo = u64_to_user_ptr(info.line_info);
4902 ulen = min_t(u32, info.nr_line_info, ulen);
4903 if (copy_to_user(user_linfo, prog->aux->linfo,
4904 info.line_info_rec_size * ulen))
4905 return -EFAULT;
4906 }
4907
4908 ulen = info.nr_jited_line_info;
4909 if (prog->aux->jited_linfo)
4910 info.nr_jited_line_info = prog->aux->nr_linfo;
4911 else
4912 info.nr_jited_line_info = 0;
4913 if (info.nr_jited_line_info && ulen) {
4914 if (bpf_dump_raw_ok(file->f_cred)) {
4915 unsigned long line_addr;
4916 __u64 __user *user_linfo;
4917 u32 i;
4918
4919 user_linfo = u64_to_user_ptr(info.jited_line_info);
4920 ulen = min_t(u32, info.nr_jited_line_info, ulen);
4921 for (i = 0; i < ulen; i++) {
4922 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4923 if (put_user((__u64)line_addr, &user_linfo[i]))
4924 return -EFAULT;
4925 }
4926 } else {
4927 info.jited_line_info = 0;
4928 }
4929 }
4930
4931 ulen = info.nr_prog_tags;
4932 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4933 if (ulen) {
4934 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4935 u32 i;
4936
4937 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4938 ulen = min_t(u32, info.nr_prog_tags, ulen);
4939 if (prog->aux->func_cnt) {
4940 for (i = 0; i < ulen; i++) {
4941 if (copy_to_user(user_prog_tags[i],
4942 prog->aux->func[i]->tag,
4943 BPF_TAG_SIZE))
4944 return -EFAULT;
4945 }
4946 } else {
4947 if (copy_to_user(user_prog_tags[0],
4948 prog->tag, BPF_TAG_SIZE))
4949 return -EFAULT;
4950 }
4951 }
4952
4953done:
4954 if (copy_to_user(uinfo, &info, info_len) ||
4955 put_user(info_len, &uattr->info.info_len))
4956 return -EFAULT;
4957
4958 return 0;
4959}
4960
4961static int bpf_map_get_info_by_fd(struct file *file,
4962 struct bpf_map *map,
4963 const union bpf_attr *attr,
4964 union bpf_attr __user *uattr)
4965{
4966 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4967 struct bpf_map_info info;
4968 u32 info_len = attr->info.info_len;
4969 int err;
4970
4971 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4972 if (err)
4973 return err;
4974 info_len = min_t(u32, sizeof(info), info_len);
4975
4976 memset(&info, 0, sizeof(info));
4977 info.type = map->map_type;
4978 info.id = map->id;
4979 info.key_size = map->key_size;
4980 info.value_size = map->value_size;
4981 info.max_entries = map->max_entries;
4982 info.map_flags = map->map_flags;
4983 info.map_extra = map->map_extra;
4984 memcpy(info.name, map->name, sizeof(map->name));
4985
4986 if (map->btf) {
4987 info.btf_id = btf_obj_id(map->btf);
4988 info.btf_key_type_id = map->btf_key_type_id;
4989 info.btf_value_type_id = map->btf_value_type_id;
4990 }
4991 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4992 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS)
4993 bpf_map_struct_ops_info_fill(&info, map);
4994
4995 if (bpf_map_is_offloaded(map)) {
4996 err = bpf_map_offload_info_fill(&info, map);
4997 if (err)
4998 return err;
4999 }
5000
5001 if (copy_to_user(uinfo, &info, info_len) ||
5002 put_user(info_len, &uattr->info.info_len))
5003 return -EFAULT;
5004
5005 return 0;
5006}
5007
5008static int bpf_btf_get_info_by_fd(struct file *file,
5009 struct btf *btf,
5010 const union bpf_attr *attr,
5011 union bpf_attr __user *uattr)
5012{
5013 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
5014 u32 info_len = attr->info.info_len;
5015 int err;
5016
5017 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
5018 if (err)
5019 return err;
5020
5021 return btf_get_info_by_fd(btf, attr, uattr);
5022}
5023
5024static int bpf_link_get_info_by_fd(struct file *file,
5025 struct bpf_link *link,
5026 const union bpf_attr *attr,
5027 union bpf_attr __user *uattr)
5028{
5029 struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
5030 struct bpf_link_info info;
5031 u32 info_len = attr->info.info_len;
5032 int err;
5033
5034 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
5035 if (err)
5036 return err;
5037 info_len = min_t(u32, sizeof(info), info_len);
5038
5039 memset(&info, 0, sizeof(info));
5040 if (copy_from_user(&info, uinfo, info_len))
5041 return -EFAULT;
5042
5043 info.type = link->type;
5044 info.id = link->id;
5045 if (link->prog)
5046 info.prog_id = link->prog->aux->id;
5047
5048 if (link->ops->fill_link_info) {
5049 err = link->ops->fill_link_info(link, &info);
5050 if (err)
5051 return err;
5052 }
5053
5054 if (copy_to_user(uinfo, &info, info_len) ||
5055 put_user(info_len, &uattr->info.info_len))
5056 return -EFAULT;
5057
5058 return 0;
5059}
5060
5061
5062#define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
5063
5064static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
5065 union bpf_attr __user *uattr)
5066{
5067 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
5068 return -EINVAL;
5069
5070 CLASS(fd, f)(attr->info.bpf_fd);
5071 if (fd_empty(f))
5072 return -EBADFD;
5073
5074 if (fd_file(f)->f_op == &bpf_prog_fops)
5075 return bpf_prog_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr,
5076 uattr);
5077 else if (fd_file(f)->f_op == &bpf_map_fops)
5078 return bpf_map_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr,
5079 uattr);
5080 else if (fd_file(f)->f_op == &btf_fops)
5081 return bpf_btf_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr, uattr);
5082 else if (fd_file(f)->f_op == &bpf_link_fops || fd_file(f)->f_op == &bpf_link_fops_poll)
5083 return bpf_link_get_info_by_fd(fd_file(f), fd_file(f)->private_data,
5084 attr, uattr);
5085 return -EINVAL;
5086}
5087
5088#define BPF_BTF_LOAD_LAST_FIELD btf_token_fd
5089
5090static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
5091{
5092 struct bpf_token *token = NULL;
5093
5094 if (CHECK_ATTR(BPF_BTF_LOAD))
5095 return -EINVAL;
5096
5097 if (attr->btf_flags & ~BPF_F_TOKEN_FD)
5098 return -EINVAL;
5099
5100 if (attr->btf_flags & BPF_F_TOKEN_FD) {
5101 token = bpf_token_get_from_fd(attr->btf_token_fd);
5102 if (IS_ERR(token))
5103 return PTR_ERR(token);
5104 if (!bpf_token_allow_cmd(token, BPF_BTF_LOAD)) {
5105 bpf_token_put(token);
5106 token = NULL;
5107 }
5108 }
5109
5110 if (!bpf_token_capable(token, CAP_BPF)) {
5111 bpf_token_put(token);
5112 return -EPERM;
5113 }
5114
5115 bpf_token_put(token);
5116
5117 return btf_new_fd(attr, uattr, uattr_size);
5118}
5119
5120#define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
5121
5122static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
5123{
5124 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
5125 return -EINVAL;
5126
5127 if (!capable(CAP_SYS_ADMIN))
5128 return -EPERM;
5129
5130 return btf_get_fd_by_id(attr->btf_id);
5131}
5132
5133static int bpf_task_fd_query_copy(const union bpf_attr *attr,
5134 union bpf_attr __user *uattr,
5135 u32 prog_id, u32 fd_type,
5136 const char *buf, u64 probe_offset,
5137 u64 probe_addr)
5138{
5139 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
5140 u32 len = buf ? strlen(buf) : 0, input_len;
5141 int err = 0;
5142
5143 if (put_user(len, &uattr->task_fd_query.buf_len))
5144 return -EFAULT;
5145 input_len = attr->task_fd_query.buf_len;
5146 if (input_len && ubuf) {
5147 if (!len) {
5148 /* nothing to copy, just make ubuf NULL terminated */
5149 char zero = '\0';
5150
5151 if (put_user(zero, ubuf))
5152 return -EFAULT;
5153 } else if (input_len >= len + 1) {
5154 /* ubuf can hold the string with NULL terminator */
5155 if (copy_to_user(ubuf, buf, len + 1))
5156 return -EFAULT;
5157 } else {
5158 /* ubuf cannot hold the string with NULL terminator,
5159 * do a partial copy with NULL terminator.
5160 */
5161 char zero = '\0';
5162
5163 err = -ENOSPC;
5164 if (copy_to_user(ubuf, buf, input_len - 1))
5165 return -EFAULT;
5166 if (put_user(zero, ubuf + input_len - 1))
5167 return -EFAULT;
5168 }
5169 }
5170
5171 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
5172 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
5173 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
5174 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
5175 return -EFAULT;
5176
5177 return err;
5178}
5179
5180#define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
5181
5182static int bpf_task_fd_query(const union bpf_attr *attr,
5183 union bpf_attr __user *uattr)
5184{
5185 pid_t pid = attr->task_fd_query.pid;
5186 u32 fd = attr->task_fd_query.fd;
5187 const struct perf_event *event;
5188 struct task_struct *task;
5189 struct file *file;
5190 int err;
5191
5192 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
5193 return -EINVAL;
5194
5195 if (!capable(CAP_SYS_ADMIN))
5196 return -EPERM;
5197
5198 if (attr->task_fd_query.flags != 0)
5199 return -EINVAL;
5200
5201 rcu_read_lock();
5202 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
5203 rcu_read_unlock();
5204 if (!task)
5205 return -ENOENT;
5206
5207 err = 0;
5208 file = fget_task(task, fd);
5209 put_task_struct(task);
5210 if (!file)
5211 return -EBADF;
5212
5213 if (file->f_op == &bpf_link_fops || file->f_op == &bpf_link_fops_poll) {
5214 struct bpf_link *link = file->private_data;
5215
5216 if (link->ops == &bpf_raw_tp_link_lops) {
5217 struct bpf_raw_tp_link *raw_tp =
5218 container_of(link, struct bpf_raw_tp_link, link);
5219 struct bpf_raw_event_map *btp = raw_tp->btp;
5220
5221 err = bpf_task_fd_query_copy(attr, uattr,
5222 raw_tp->link.prog->aux->id,
5223 BPF_FD_TYPE_RAW_TRACEPOINT,
5224 btp->tp->name, 0, 0);
5225 goto put_file;
5226 }
5227 goto out_not_supp;
5228 }
5229
5230 event = perf_get_event(file);
5231 if (!IS_ERR(event)) {
5232 u64 probe_offset, probe_addr;
5233 u32 prog_id, fd_type;
5234 const char *buf;
5235
5236 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
5237 &buf, &probe_offset,
5238 &probe_addr, NULL);
5239 if (!err)
5240 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
5241 fd_type, buf,
5242 probe_offset,
5243 probe_addr);
5244 goto put_file;
5245 }
5246
5247out_not_supp:
5248 err = -ENOTSUPP;
5249put_file:
5250 fput(file);
5251 return err;
5252}
5253
5254#define BPF_MAP_BATCH_LAST_FIELD batch.flags
5255
5256#define BPF_DO_BATCH(fn, ...) \
5257 do { \
5258 if (!fn) { \
5259 err = -ENOTSUPP; \
5260 goto err_put; \
5261 } \
5262 err = fn(__VA_ARGS__); \
5263 } while (0)
5264
5265static int bpf_map_do_batch(const union bpf_attr *attr,
5266 union bpf_attr __user *uattr,
5267 int cmd)
5268{
5269 bool has_read = cmd == BPF_MAP_LOOKUP_BATCH ||
5270 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
5271 bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
5272 struct bpf_map *map;
5273 int err;
5274
5275 if (CHECK_ATTR(BPF_MAP_BATCH))
5276 return -EINVAL;
5277
5278 CLASS(fd, f)(attr->batch.map_fd);
5279
5280 map = __bpf_map_get(f);
5281 if (IS_ERR(map))
5282 return PTR_ERR(map);
5283 if (has_write)
5284 bpf_map_write_active_inc(map);
5285 if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
5286 err = -EPERM;
5287 goto err_put;
5288 }
5289 if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
5290 err = -EPERM;
5291 goto err_put;
5292 }
5293
5294 if (cmd == BPF_MAP_LOOKUP_BATCH)
5295 BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
5296 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
5297 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
5298 else if (cmd == BPF_MAP_UPDATE_BATCH)
5299 BPF_DO_BATCH(map->ops->map_update_batch, map, fd_file(f), attr, uattr);
5300 else
5301 BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
5302err_put:
5303 if (has_write) {
5304 maybe_wait_bpf_programs(map);
5305 bpf_map_write_active_dec(map);
5306 }
5307 return err;
5308}
5309
5310#define BPF_LINK_CREATE_LAST_FIELD link_create.uprobe_multi.pid
5311static int link_create(union bpf_attr *attr, bpfptr_t uattr)
5312{
5313 struct bpf_prog *prog;
5314 int ret;
5315
5316 if (CHECK_ATTR(BPF_LINK_CREATE))
5317 return -EINVAL;
5318
5319 if (attr->link_create.attach_type == BPF_STRUCT_OPS)
5320 return bpf_struct_ops_link_create(attr);
5321
5322 prog = bpf_prog_get(attr->link_create.prog_fd);
5323 if (IS_ERR(prog))
5324 return PTR_ERR(prog);
5325
5326 ret = bpf_prog_attach_check_attach_type(prog,
5327 attr->link_create.attach_type);
5328 if (ret)
5329 goto out;
5330
5331 switch (prog->type) {
5332 case BPF_PROG_TYPE_CGROUP_SKB:
5333 case BPF_PROG_TYPE_CGROUP_SOCK:
5334 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
5335 case BPF_PROG_TYPE_SOCK_OPS:
5336 case BPF_PROG_TYPE_CGROUP_DEVICE:
5337 case BPF_PROG_TYPE_CGROUP_SYSCTL:
5338 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
5339 ret = cgroup_bpf_link_attach(attr, prog);
5340 break;
5341 case BPF_PROG_TYPE_EXT:
5342 ret = bpf_tracing_prog_attach(prog,
5343 attr->link_create.target_fd,
5344 attr->link_create.target_btf_id,
5345 attr->link_create.tracing.cookie);
5346 break;
5347 case BPF_PROG_TYPE_LSM:
5348 case BPF_PROG_TYPE_TRACING:
5349 if (attr->link_create.attach_type != prog->expected_attach_type) {
5350 ret = -EINVAL;
5351 goto out;
5352 }
5353 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
5354 ret = bpf_raw_tp_link_attach(prog, NULL, attr->link_create.tracing.cookie);
5355 else if (prog->expected_attach_type == BPF_TRACE_ITER)
5356 ret = bpf_iter_link_attach(attr, uattr, prog);
5357 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
5358 ret = cgroup_bpf_link_attach(attr, prog);
5359 else
5360 ret = bpf_tracing_prog_attach(prog,
5361 attr->link_create.target_fd,
5362 attr->link_create.target_btf_id,
5363 attr->link_create.tracing.cookie);
5364 break;
5365 case BPF_PROG_TYPE_FLOW_DISSECTOR:
5366 case BPF_PROG_TYPE_SK_LOOKUP:
5367 ret = netns_bpf_link_create(attr, prog);
5368 break;
5369 case BPF_PROG_TYPE_SK_MSG:
5370 case BPF_PROG_TYPE_SK_SKB:
5371 ret = sock_map_link_create(attr, prog);
5372 break;
5373#ifdef CONFIG_NET
5374 case BPF_PROG_TYPE_XDP:
5375 ret = bpf_xdp_link_attach(attr, prog);
5376 break;
5377 case BPF_PROG_TYPE_SCHED_CLS:
5378 if (attr->link_create.attach_type == BPF_TCX_INGRESS ||
5379 attr->link_create.attach_type == BPF_TCX_EGRESS)
5380 ret = tcx_link_attach(attr, prog);
5381 else
5382 ret = netkit_link_attach(attr, prog);
5383 break;
5384 case BPF_PROG_TYPE_NETFILTER:
5385 ret = bpf_nf_link_attach(attr, prog);
5386 break;
5387#endif
5388 case BPF_PROG_TYPE_PERF_EVENT:
5389 case BPF_PROG_TYPE_TRACEPOINT:
5390 ret = bpf_perf_link_attach(attr, prog);
5391 break;
5392 case BPF_PROG_TYPE_KPROBE:
5393 if (attr->link_create.attach_type == BPF_PERF_EVENT)
5394 ret = bpf_perf_link_attach(attr, prog);
5395 else if (attr->link_create.attach_type == BPF_TRACE_KPROBE_MULTI ||
5396 attr->link_create.attach_type == BPF_TRACE_KPROBE_SESSION)
5397 ret = bpf_kprobe_multi_link_attach(attr, prog);
5398 else if (attr->link_create.attach_type == BPF_TRACE_UPROBE_MULTI ||
5399 attr->link_create.attach_type == BPF_TRACE_UPROBE_SESSION)
5400 ret = bpf_uprobe_multi_link_attach(attr, prog);
5401 break;
5402 default:
5403 ret = -EINVAL;
5404 }
5405
5406out:
5407 if (ret < 0)
5408 bpf_prog_put(prog);
5409 return ret;
5410}
5411
5412static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
5413{
5414 struct bpf_map *new_map, *old_map = NULL;
5415 int ret;
5416
5417 new_map = bpf_map_get(attr->link_update.new_map_fd);
5418 if (IS_ERR(new_map))
5419 return PTR_ERR(new_map);
5420
5421 if (attr->link_update.flags & BPF_F_REPLACE) {
5422 old_map = bpf_map_get(attr->link_update.old_map_fd);
5423 if (IS_ERR(old_map)) {
5424 ret = PTR_ERR(old_map);
5425 goto out_put;
5426 }
5427 } else if (attr->link_update.old_map_fd) {
5428 ret = -EINVAL;
5429 goto out_put;
5430 }
5431
5432 ret = link->ops->update_map(link, new_map, old_map);
5433
5434 if (old_map)
5435 bpf_map_put(old_map);
5436out_put:
5437 bpf_map_put(new_map);
5438 return ret;
5439}
5440
5441#define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
5442
5443static int link_update(union bpf_attr *attr)
5444{
5445 struct bpf_prog *old_prog = NULL, *new_prog;
5446 struct bpf_link *link;
5447 u32 flags;
5448 int ret;
5449
5450 if (CHECK_ATTR(BPF_LINK_UPDATE))
5451 return -EINVAL;
5452
5453 flags = attr->link_update.flags;
5454 if (flags & ~BPF_F_REPLACE)
5455 return -EINVAL;
5456
5457 link = bpf_link_get_from_fd(attr->link_update.link_fd);
5458 if (IS_ERR(link))
5459 return PTR_ERR(link);
5460
5461 if (link->ops->update_map) {
5462 ret = link_update_map(link, attr);
5463 goto out_put_link;
5464 }
5465
5466 new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
5467 if (IS_ERR(new_prog)) {
5468 ret = PTR_ERR(new_prog);
5469 goto out_put_link;
5470 }
5471
5472 if (flags & BPF_F_REPLACE) {
5473 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
5474 if (IS_ERR(old_prog)) {
5475 ret = PTR_ERR(old_prog);
5476 old_prog = NULL;
5477 goto out_put_progs;
5478 }
5479 } else if (attr->link_update.old_prog_fd) {
5480 ret = -EINVAL;
5481 goto out_put_progs;
5482 }
5483
5484 if (link->ops->update_prog)
5485 ret = link->ops->update_prog(link, new_prog, old_prog);
5486 else
5487 ret = -EINVAL;
5488
5489out_put_progs:
5490 if (old_prog)
5491 bpf_prog_put(old_prog);
5492 if (ret)
5493 bpf_prog_put(new_prog);
5494out_put_link:
5495 bpf_link_put_direct(link);
5496 return ret;
5497}
5498
5499#define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
5500
5501static int link_detach(union bpf_attr *attr)
5502{
5503 struct bpf_link *link;
5504 int ret;
5505
5506 if (CHECK_ATTR(BPF_LINK_DETACH))
5507 return -EINVAL;
5508
5509 link = bpf_link_get_from_fd(attr->link_detach.link_fd);
5510 if (IS_ERR(link))
5511 return PTR_ERR(link);
5512
5513 if (link->ops->detach)
5514 ret = link->ops->detach(link);
5515 else
5516 ret = -EOPNOTSUPP;
5517
5518 bpf_link_put_direct(link);
5519 return ret;
5520}
5521
5522struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
5523{
5524 return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
5525}
5526EXPORT_SYMBOL(bpf_link_inc_not_zero);
5527
5528struct bpf_link *bpf_link_by_id(u32 id)
5529{
5530 struct bpf_link *link;
5531
5532 if (!id)
5533 return ERR_PTR(-ENOENT);
5534
5535 spin_lock_bh(&link_idr_lock);
5536 /* before link is "settled", ID is 0, pretend it doesn't exist yet */
5537 link = idr_find(&link_idr, id);
5538 if (link) {
5539 if (link->id)
5540 link = bpf_link_inc_not_zero(link);
5541 else
5542 link = ERR_PTR(-EAGAIN);
5543 } else {
5544 link = ERR_PTR(-ENOENT);
5545 }
5546 spin_unlock_bh(&link_idr_lock);
5547 return link;
5548}
5549
5550struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
5551{
5552 struct bpf_link *link;
5553
5554 spin_lock_bh(&link_idr_lock);
5555again:
5556 link = idr_get_next(&link_idr, id);
5557 if (link) {
5558 link = bpf_link_inc_not_zero(link);
5559 if (IS_ERR(link)) {
5560 (*id)++;
5561 goto again;
5562 }
5563 }
5564 spin_unlock_bh(&link_idr_lock);
5565
5566 return link;
5567}
5568
5569#define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
5570
5571static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
5572{
5573 struct bpf_link *link;
5574 u32 id = attr->link_id;
5575 int fd;
5576
5577 if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
5578 return -EINVAL;
5579
5580 if (!capable(CAP_SYS_ADMIN))
5581 return -EPERM;
5582
5583 link = bpf_link_by_id(id);
5584 if (IS_ERR(link))
5585 return PTR_ERR(link);
5586
5587 fd = bpf_link_new_fd(link);
5588 if (fd < 0)
5589 bpf_link_put_direct(link);
5590
5591 return fd;
5592}
5593
5594DEFINE_MUTEX(bpf_stats_enabled_mutex);
5595
5596static int bpf_stats_release(struct inode *inode, struct file *file)
5597{
5598 mutex_lock(&bpf_stats_enabled_mutex);
5599 static_key_slow_dec(&bpf_stats_enabled_key.key);
5600 mutex_unlock(&bpf_stats_enabled_mutex);
5601 return 0;
5602}
5603
5604static const struct file_operations bpf_stats_fops = {
5605 .release = bpf_stats_release,
5606};
5607
5608static int bpf_enable_runtime_stats(void)
5609{
5610 int fd;
5611
5612 mutex_lock(&bpf_stats_enabled_mutex);
5613
5614 /* Set a very high limit to avoid overflow */
5615 if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
5616 mutex_unlock(&bpf_stats_enabled_mutex);
5617 return -EBUSY;
5618 }
5619
5620 fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
5621 if (fd >= 0)
5622 static_key_slow_inc(&bpf_stats_enabled_key.key);
5623
5624 mutex_unlock(&bpf_stats_enabled_mutex);
5625 return fd;
5626}
5627
5628#define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
5629
5630static int bpf_enable_stats(union bpf_attr *attr)
5631{
5632
5633 if (CHECK_ATTR(BPF_ENABLE_STATS))
5634 return -EINVAL;
5635
5636 if (!capable(CAP_SYS_ADMIN))
5637 return -EPERM;
5638
5639 switch (attr->enable_stats.type) {
5640 case BPF_STATS_RUN_TIME:
5641 return bpf_enable_runtime_stats();
5642 default:
5643 break;
5644 }
5645 return -EINVAL;
5646}
5647
5648#define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
5649
5650static int bpf_iter_create(union bpf_attr *attr)
5651{
5652 struct bpf_link *link;
5653 int err;
5654
5655 if (CHECK_ATTR(BPF_ITER_CREATE))
5656 return -EINVAL;
5657
5658 if (attr->iter_create.flags)
5659 return -EINVAL;
5660
5661 link = bpf_link_get_from_fd(attr->iter_create.link_fd);
5662 if (IS_ERR(link))
5663 return PTR_ERR(link);
5664
5665 err = bpf_iter_new_fd(link);
5666 bpf_link_put_direct(link);
5667
5668 return err;
5669}
5670
5671#define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
5672
5673static int bpf_prog_bind_map(union bpf_attr *attr)
5674{
5675 struct bpf_prog *prog;
5676 struct bpf_map *map;
5677 struct bpf_map **used_maps_old, **used_maps_new;
5678 int i, ret = 0;
5679
5680 if (CHECK_ATTR(BPF_PROG_BIND_MAP))
5681 return -EINVAL;
5682
5683 if (attr->prog_bind_map.flags)
5684 return -EINVAL;
5685
5686 prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
5687 if (IS_ERR(prog))
5688 return PTR_ERR(prog);
5689
5690 map = bpf_map_get(attr->prog_bind_map.map_fd);
5691 if (IS_ERR(map)) {
5692 ret = PTR_ERR(map);
5693 goto out_prog_put;
5694 }
5695
5696 mutex_lock(&prog->aux->used_maps_mutex);
5697
5698 used_maps_old = prog->aux->used_maps;
5699
5700 for (i = 0; i < prog->aux->used_map_cnt; i++)
5701 if (used_maps_old[i] == map) {
5702 bpf_map_put(map);
5703 goto out_unlock;
5704 }
5705
5706 used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
5707 sizeof(used_maps_new[0]),
5708 GFP_KERNEL);
5709 if (!used_maps_new) {
5710 ret = -ENOMEM;
5711 goto out_unlock;
5712 }
5713
5714 /* The bpf program will not access the bpf map, but for the sake of
5715 * simplicity, increase sleepable_refcnt for sleepable program as well.
5716 */
5717 if (prog->sleepable)
5718 atomic64_inc(&map->sleepable_refcnt);
5719 memcpy(used_maps_new, used_maps_old,
5720 sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
5721 used_maps_new[prog->aux->used_map_cnt] = map;
5722
5723 prog->aux->used_map_cnt++;
5724 prog->aux->used_maps = used_maps_new;
5725
5726 kfree(used_maps_old);
5727
5728out_unlock:
5729 mutex_unlock(&prog->aux->used_maps_mutex);
5730
5731 if (ret)
5732 bpf_map_put(map);
5733out_prog_put:
5734 bpf_prog_put(prog);
5735 return ret;
5736}
5737
5738#define BPF_TOKEN_CREATE_LAST_FIELD token_create.bpffs_fd
5739
5740static int token_create(union bpf_attr *attr)
5741{
5742 if (CHECK_ATTR(BPF_TOKEN_CREATE))
5743 return -EINVAL;
5744
5745 /* no flags are supported yet */
5746 if (attr->token_create.flags)
5747 return -EINVAL;
5748
5749 return bpf_token_create(attr);
5750}
5751
5752static int __sys_bpf(enum bpf_cmd cmd, bpfptr_t uattr, unsigned int size)
5753{
5754 union bpf_attr attr;
5755 int err;
5756
5757 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
5758 if (err)
5759 return err;
5760 size = min_t(u32, size, sizeof(attr));
5761
5762 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
5763 memset(&attr, 0, sizeof(attr));
5764 if (copy_from_bpfptr(&attr, uattr, size) != 0)
5765 return -EFAULT;
5766
5767 err = security_bpf(cmd, &attr, size);
5768 if (err < 0)
5769 return err;
5770
5771 switch (cmd) {
5772 case BPF_MAP_CREATE:
5773 err = map_create(&attr);
5774 break;
5775 case BPF_MAP_LOOKUP_ELEM:
5776 err = map_lookup_elem(&attr);
5777 break;
5778 case BPF_MAP_UPDATE_ELEM:
5779 err = map_update_elem(&attr, uattr);
5780 break;
5781 case BPF_MAP_DELETE_ELEM:
5782 err = map_delete_elem(&attr, uattr);
5783 break;
5784 case BPF_MAP_GET_NEXT_KEY:
5785 err = map_get_next_key(&attr);
5786 break;
5787 case BPF_MAP_FREEZE:
5788 err = map_freeze(&attr);
5789 break;
5790 case BPF_PROG_LOAD:
5791 err = bpf_prog_load(&attr, uattr, size);
5792 break;
5793 case BPF_OBJ_PIN:
5794 err = bpf_obj_pin(&attr);
5795 break;
5796 case BPF_OBJ_GET:
5797 err = bpf_obj_get(&attr);
5798 break;
5799 case BPF_PROG_ATTACH:
5800 err = bpf_prog_attach(&attr);
5801 break;
5802 case BPF_PROG_DETACH:
5803 err = bpf_prog_detach(&attr);
5804 break;
5805 case BPF_PROG_QUERY:
5806 err = bpf_prog_query(&attr, uattr.user);
5807 break;
5808 case BPF_PROG_TEST_RUN:
5809 err = bpf_prog_test_run(&attr, uattr.user);
5810 break;
5811 case BPF_PROG_GET_NEXT_ID:
5812 err = bpf_obj_get_next_id(&attr, uattr.user,
5813 &prog_idr, &prog_idr_lock);
5814 break;
5815 case BPF_MAP_GET_NEXT_ID:
5816 err = bpf_obj_get_next_id(&attr, uattr.user,
5817 &map_idr, &map_idr_lock);
5818 break;
5819 case BPF_BTF_GET_NEXT_ID:
5820 err = bpf_obj_get_next_id(&attr, uattr.user,
5821 &btf_idr, &btf_idr_lock);
5822 break;
5823 case BPF_PROG_GET_FD_BY_ID:
5824 err = bpf_prog_get_fd_by_id(&attr);
5825 break;
5826 case BPF_MAP_GET_FD_BY_ID:
5827 err = bpf_map_get_fd_by_id(&attr);
5828 break;
5829 case BPF_OBJ_GET_INFO_BY_FD:
5830 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5831 break;
5832 case BPF_RAW_TRACEPOINT_OPEN:
5833 err = bpf_raw_tracepoint_open(&attr);
5834 break;
5835 case BPF_BTF_LOAD:
5836 err = bpf_btf_load(&attr, uattr, size);
5837 break;
5838 case BPF_BTF_GET_FD_BY_ID:
5839 err = bpf_btf_get_fd_by_id(&attr);
5840 break;
5841 case BPF_TASK_FD_QUERY:
5842 err = bpf_task_fd_query(&attr, uattr.user);
5843 break;
5844 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5845 err = map_lookup_and_delete_elem(&attr);
5846 break;
5847 case BPF_MAP_LOOKUP_BATCH:
5848 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5849 break;
5850 case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5851 err = bpf_map_do_batch(&attr, uattr.user,
5852 BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5853 break;
5854 case BPF_MAP_UPDATE_BATCH:
5855 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5856 break;
5857 case BPF_MAP_DELETE_BATCH:
5858 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5859 break;
5860 case BPF_LINK_CREATE:
5861 err = link_create(&attr, uattr);
5862 break;
5863 case BPF_LINK_UPDATE:
5864 err = link_update(&attr);
5865 break;
5866 case BPF_LINK_GET_FD_BY_ID:
5867 err = bpf_link_get_fd_by_id(&attr);
5868 break;
5869 case BPF_LINK_GET_NEXT_ID:
5870 err = bpf_obj_get_next_id(&attr, uattr.user,
5871 &link_idr, &link_idr_lock);
5872 break;
5873 case BPF_ENABLE_STATS:
5874 err = bpf_enable_stats(&attr);
5875 break;
5876 case BPF_ITER_CREATE:
5877 err = bpf_iter_create(&attr);
5878 break;
5879 case BPF_LINK_DETACH:
5880 err = link_detach(&attr);
5881 break;
5882 case BPF_PROG_BIND_MAP:
5883 err = bpf_prog_bind_map(&attr);
5884 break;
5885 case BPF_TOKEN_CREATE:
5886 err = token_create(&attr);
5887 break;
5888 default:
5889 err = -EINVAL;
5890 break;
5891 }
5892
5893 return err;
5894}
5895
5896SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5897{
5898 return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5899}
5900
5901static bool syscall_prog_is_valid_access(int off, int size,
5902 enum bpf_access_type type,
5903 const struct bpf_prog *prog,
5904 struct bpf_insn_access_aux *info)
5905{
5906 if (off < 0 || off >= U16_MAX)
5907 return false;
5908 if (off % size != 0)
5909 return false;
5910 return true;
5911}
5912
5913BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5914{
5915 switch (cmd) {
5916 case BPF_MAP_CREATE:
5917 case BPF_MAP_DELETE_ELEM:
5918 case BPF_MAP_UPDATE_ELEM:
5919 case BPF_MAP_FREEZE:
5920 case BPF_MAP_GET_FD_BY_ID:
5921 case BPF_PROG_LOAD:
5922 case BPF_BTF_LOAD:
5923 case BPF_LINK_CREATE:
5924 case BPF_RAW_TRACEPOINT_OPEN:
5925 break;
5926 default:
5927 return -EINVAL;
5928 }
5929 return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5930}
5931
5932
5933/* To shut up -Wmissing-prototypes.
5934 * This function is used by the kernel light skeleton
5935 * to load bpf programs when modules are loaded or during kernel boot.
5936 * See tools/lib/bpf/skel_internal.h
5937 */
5938int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5939
5940int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5941{
5942 struct bpf_prog * __maybe_unused prog;
5943 struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5944
5945 switch (cmd) {
5946#ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5947 case BPF_PROG_TEST_RUN:
5948 if (attr->test.data_in || attr->test.data_out ||
5949 attr->test.ctx_out || attr->test.duration ||
5950 attr->test.repeat || attr->test.flags)
5951 return -EINVAL;
5952
5953 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5954 if (IS_ERR(prog))
5955 return PTR_ERR(prog);
5956
5957 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5958 attr->test.ctx_size_in > U16_MAX) {
5959 bpf_prog_put(prog);
5960 return -EINVAL;
5961 }
5962
5963 run_ctx.bpf_cookie = 0;
5964 if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5965 /* recursion detected */
5966 __bpf_prog_exit_sleepable_recur(prog, 0, &run_ctx);
5967 bpf_prog_put(prog);
5968 return -EBUSY;
5969 }
5970 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5971 __bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5972 &run_ctx);
5973 bpf_prog_put(prog);
5974 return 0;
5975#endif
5976 default:
5977 return ____bpf_sys_bpf(cmd, attr, size);
5978 }
5979}
5980EXPORT_SYMBOL(kern_sys_bpf);
5981
5982static const struct bpf_func_proto bpf_sys_bpf_proto = {
5983 .func = bpf_sys_bpf,
5984 .gpl_only = false,
5985 .ret_type = RET_INTEGER,
5986 .arg1_type = ARG_ANYTHING,
5987 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5988 .arg3_type = ARG_CONST_SIZE,
5989};
5990
5991const struct bpf_func_proto * __weak
5992tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5993{
5994 return bpf_base_func_proto(func_id, prog);
5995}
5996
5997BPF_CALL_1(bpf_sys_close, u32, fd)
5998{
5999 /* When bpf program calls this helper there should not be
6000 * an fdget() without matching completed fdput().
6001 * This helper is allowed in the following callchain only:
6002 * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
6003 */
6004 return close_fd(fd);
6005}
6006
6007static const struct bpf_func_proto bpf_sys_close_proto = {
6008 .func = bpf_sys_close,
6009 .gpl_only = false,
6010 .ret_type = RET_INTEGER,
6011 .arg1_type = ARG_ANYTHING,
6012};
6013
6014BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
6015{
6016 *res = 0;
6017 if (flags)
6018 return -EINVAL;
6019
6020 if (name_sz <= 1 || name[name_sz - 1])
6021 return -EINVAL;
6022
6023 if (!bpf_dump_raw_ok(current_cred()))
6024 return -EPERM;
6025
6026 *res = kallsyms_lookup_name(name);
6027 return *res ? 0 : -ENOENT;
6028}
6029
6030static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
6031 .func = bpf_kallsyms_lookup_name,
6032 .gpl_only = false,
6033 .ret_type = RET_INTEGER,
6034 .arg1_type = ARG_PTR_TO_MEM,
6035 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
6036 .arg3_type = ARG_ANYTHING,
6037 .arg4_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_WRITE | MEM_ALIGNED,
6038 .arg4_size = sizeof(u64),
6039};
6040
6041static const struct bpf_func_proto *
6042syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6043{
6044 switch (func_id) {
6045 case BPF_FUNC_sys_bpf:
6046 return !bpf_token_capable(prog->aux->token, CAP_PERFMON)
6047 ? NULL : &bpf_sys_bpf_proto;
6048 case BPF_FUNC_btf_find_by_name_kind:
6049 return &bpf_btf_find_by_name_kind_proto;
6050 case BPF_FUNC_sys_close:
6051 return &bpf_sys_close_proto;
6052 case BPF_FUNC_kallsyms_lookup_name:
6053 return &bpf_kallsyms_lookup_name_proto;
6054 default:
6055 return tracing_prog_func_proto(func_id, prog);
6056 }
6057}
6058
6059const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
6060 .get_func_proto = syscall_prog_func_proto,
6061 .is_valid_access = syscall_prog_is_valid_access,
6062};
6063
6064const struct bpf_prog_ops bpf_syscall_prog_ops = {
6065 .test_run = bpf_prog_test_run_syscall,
6066};
6067
6068#ifdef CONFIG_SYSCTL
6069static int bpf_stats_handler(const struct ctl_table *table, int write,
6070 void *buffer, size_t *lenp, loff_t *ppos)
6071{
6072 struct static_key *key = (struct static_key *)table->data;
6073 static int saved_val;
6074 int val, ret;
6075 struct ctl_table tmp = {
6076 .data = &val,
6077 .maxlen = sizeof(val),
6078 .mode = table->mode,
6079 .extra1 = SYSCTL_ZERO,
6080 .extra2 = SYSCTL_ONE,
6081 };
6082
6083 if (write && !capable(CAP_SYS_ADMIN))
6084 return -EPERM;
6085
6086 mutex_lock(&bpf_stats_enabled_mutex);
6087 val = saved_val;
6088 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
6089 if (write && !ret && val != saved_val) {
6090 if (val)
6091 static_key_slow_inc(key);
6092 else
6093 static_key_slow_dec(key);
6094 saved_val = val;
6095 }
6096 mutex_unlock(&bpf_stats_enabled_mutex);
6097 return ret;
6098}
6099
6100void __weak unpriv_ebpf_notify(int new_state)
6101{
6102}
6103
6104static int bpf_unpriv_handler(const struct ctl_table *table, int write,
6105 void *buffer, size_t *lenp, loff_t *ppos)
6106{
6107 int ret, unpriv_enable = *(int *)table->data;
6108 bool locked_state = unpriv_enable == 1;
6109 struct ctl_table tmp = *table;
6110
6111 if (write && !capable(CAP_SYS_ADMIN))
6112 return -EPERM;
6113
6114 tmp.data = &unpriv_enable;
6115 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
6116 if (write && !ret) {
6117 if (locked_state && unpriv_enable != 1)
6118 return -EPERM;
6119 *(int *)table->data = unpriv_enable;
6120 }
6121
6122 if (write)
6123 unpriv_ebpf_notify(unpriv_enable);
6124
6125 return ret;
6126}
6127
6128static struct ctl_table bpf_syscall_table[] = {
6129 {
6130 .procname = "unprivileged_bpf_disabled",
6131 .data = &sysctl_unprivileged_bpf_disabled,
6132 .maxlen = sizeof(sysctl_unprivileged_bpf_disabled),
6133 .mode = 0644,
6134 .proc_handler = bpf_unpriv_handler,
6135 .extra1 = SYSCTL_ZERO,
6136 .extra2 = SYSCTL_TWO,
6137 },
6138 {
6139 .procname = "bpf_stats_enabled",
6140 .data = &bpf_stats_enabled_key.key,
6141 .mode = 0644,
6142 .proc_handler = bpf_stats_handler,
6143 },
6144};
6145
6146static int __init bpf_syscall_sysctl_init(void)
6147{
6148 register_sysctl_init("kernel", bpf_syscall_table);
6149 return 0;
6150}
6151late_initcall(bpf_syscall_sysctl_init);
6152#endif /* CONFIG_SYSCTL */