?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
bpf: add ability to charge bpf maps memory dynamically
[linux/fpc-iii.git] / kernel / bpf / syscall.c
blob7958252a4d29627e2fb8256914561682077285be
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/bpf_lirc.h>
15 #include <linux/btf.h>
16 #include <linux/syscalls.h>
17 #include <linux/slab.h>
18 #include <linux/sched/signal.h>
19 #include <linux/vmalloc.h>
20 #include <linux/mmzone.h>
21 #include <linux/anon_inodes.h>
22 #include <linux/fdtable.h>
23 #include <linux/file.h>
24 #include <linux/fs.h>
25 #include <linux/license.h>
26 #include <linux/filter.h>
27 #include <linux/version.h>
28 #include <linux/kernel.h>
29 #include <linux/idr.h>
30 #include <linux/cred.h>
31 #include <linux/timekeeping.h>
32 #include <linux/ctype.h>
33 #include <linux/btf.h>
34 #include <linux/nospec.h>
35
36 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
37 (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
38 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
39 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
40 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
41 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
42
43 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
44
45 DEFINE_PER_CPU(int, bpf_prog_active);
46 static DEFINE_IDR(prog_idr);
47 static DEFINE_SPINLOCK(prog_idr_lock);
48 static DEFINE_IDR(map_idr);
49 static DEFINE_SPINLOCK(map_idr_lock);
50
51 int sysctl_unprivileged_bpf_disabled __read_mostly;
52
53 static const struct bpf_map_ops * const bpf_map_types[] = {
54 #define BPF_PROG_TYPE(_id, _ops)
55 #define BPF_MAP_TYPE(_id, _ops) \
56 [_id] = &_ops,
57 #include <linux/bpf_types.h>
58 #undef BPF_PROG_TYPE
59 #undef BPF_MAP_TYPE
60 };
61
62 /*
63 * If we're handed a bigger struct than we know of, ensure all the unknown bits
64 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
65 * we don't know about yet.
66 *
67 * There is a ToCToU between this function call and the following
68 * copy_from_user() call. However, this is not a concern since this function is
69 * meant to be a future-proofing of bits.
70 */
71 int bpf_check_uarg_tail_zero(void __user *uaddr,
72 size_t expected_size,
73 size_t actual_size)
74 {
75 unsigned char __user *addr;
76 unsigned char __user *end;
77 unsigned char val;
78 int err;
79
80 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
81 return -E2BIG;
82
83 if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
84 return -EFAULT;
85
86 if (actual_size <= expected_size)
87 return 0;
88
89 addr = uaddr + expected_size;
90 end = uaddr + actual_size;
91
92 for (; addr < end; addr++) {
93 err = get_user(val, addr);
94 if (err)
95 return err;
96 if (val)
97 return -E2BIG;
98 }
99
100 return 0;
101 }
102
103 const struct bpf_map_ops bpf_map_offload_ops = {
104 .map_alloc = bpf_map_offload_map_alloc,
105 .map_free = bpf_map_offload_map_free,
106 };
107
108 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
109 {
110 const struct bpf_map_ops *ops;
111 u32 type = attr->map_type;
112 struct bpf_map *map;
113 int err;
114
115 if (type >= ARRAY_SIZE(bpf_map_types))
116 return ERR_PTR(-EINVAL);
117 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
118 ops = bpf_map_types[type];
119 if (!ops)
120 return ERR_PTR(-EINVAL);
121
122 if (ops->map_alloc_check) {
123 err = ops->map_alloc_check(attr);
124 if (err)
125 return ERR_PTR(err);
126 }
127 if (attr->map_ifindex)
128 ops = &bpf_map_offload_ops;
129 map = ops->map_alloc(attr);
130 if (IS_ERR(map))
131 return map;
132 map->ops = ops;
133 map->map_type = type;
134 return map;
135 }
136
137 void *bpf_map_area_alloc(size_t size, int numa_node)
138 {
139 /* We definitely need __GFP_NORETRY, so OOM killer doesn't
140 * trigger under memory pressure as we really just want to
141 * fail instead.
142 */
143 const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
144 void *area;
145
146 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
147 area = kmalloc_node(size, GFP_USER | flags, numa_node);
148 if (area != NULL)
149 return area;
150 }
151
152 return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
153 __builtin_return_address(0));
154 }
155
156 void bpf_map_area_free(void *area)
157 {
158 kvfree(area);
159 }
160
161 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
162 {
163 map->map_type = attr->map_type;
164 map->key_size = attr->key_size;
165 map->value_size = attr->value_size;
166 map->max_entries = attr->max_entries;
167 map->map_flags = attr->map_flags;
168 map->numa_node = bpf_map_attr_numa_node(attr);
169 }
170
171 int bpf_map_precharge_memlock(u32 pages)
172 {
173 struct user_struct *user = get_current_user();
174 unsigned long memlock_limit, cur;
175
176 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
177 cur = atomic_long_read(&user->locked_vm);
178 free_uid(user);
179 if (cur + pages > memlock_limit)
180 return -EPERM;
181 return 0;
182 }
183
184 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
185 {
186 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
187
188 if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
189 atomic_long_sub(pages, &user->locked_vm);
190 return -EPERM;
191 }
192 return 0;
193 }
194
195 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
196 {
197 atomic_long_sub(pages, &user->locked_vm);
198 }
199
200 static int bpf_map_init_memlock(struct bpf_map *map)
201 {
202 struct user_struct *user = get_current_user();
203 int ret;
204
205 ret = bpf_charge_memlock(user, map->pages);
206 if (ret) {
207 free_uid(user);
208 return ret;
209 }
210 map->user = user;
211 return ret;
212 }
213
214 static void bpf_map_release_memlock(struct bpf_map *map)
215 {
216 struct user_struct *user = map->user;
217 bpf_uncharge_memlock(user, map->pages);
218 free_uid(user);
219 }
220
221 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
222 {
223 int ret;
224
225 ret = bpf_charge_memlock(map->user, pages);
226 if (ret)
227 return ret;
228 map->pages += pages;
229 return ret;
230 }
231
232 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
233 {
234 bpf_uncharge_memlock(map->user, pages);
235 map->pages -= pages;
236 }
237
238 static int bpf_map_alloc_id(struct bpf_map *map)
239 {
240 int id;
241
242 idr_preload(GFP_KERNEL);
243 spin_lock_bh(&map_idr_lock);
244 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
245 if (id > 0)
246 map->id = id;
247 spin_unlock_bh(&map_idr_lock);
248 idr_preload_end();
249
250 if (WARN_ON_ONCE(!id))
251 return -ENOSPC;
252
253 return id > 0 ? 0 : id;
254 }
255
256 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
257 {
258 unsigned long flags;
259
260 /* Offloaded maps are removed from the IDR store when their device
261 * disappears - even if someone holds an fd to them they are unusable,
262 * the memory is gone, all ops will fail; they are simply waiting for
263 * refcnt to drop to be freed.
264 */
265 if (!map->id)
266 return;
267
268 if (do_idr_lock)
269 spin_lock_irqsave(&map_idr_lock, flags);
270 else
271 __acquire(&map_idr_lock);
272
273 idr_remove(&map_idr, map->id);
274 map->id = 0;
275
276 if (do_idr_lock)
277 spin_unlock_irqrestore(&map_idr_lock, flags);
278 else
279 __release(&map_idr_lock);
280 }
281
282 /* called from workqueue */
283 static void bpf_map_free_deferred(struct work_struct *work)
284 {
285 struct bpf_map *map = container_of(work, struct bpf_map, work);
286
287 bpf_map_release_memlock(map);
288 security_bpf_map_free(map);
289 /* implementation dependent freeing */
290 map->ops->map_free(map);
291 }
292
293 static void bpf_map_put_uref(struct bpf_map *map)
294 {
295 if (atomic_dec_and_test(&map->usercnt)) {
296 if (map->ops->map_release_uref)
297 map->ops->map_release_uref(map);
298 }
299 }
300
301 /* decrement map refcnt and schedule it for freeing via workqueue
302 * (unrelying map implementation ops->map_free() might sleep)
303 */
304 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
305 {
306 if (atomic_dec_and_test(&map->refcnt)) {
307 /* bpf_map_free_id() must be called first */
308 bpf_map_free_id(map, do_idr_lock);
309 btf_put(map->btf);
310 INIT_WORK(&map->work, bpf_map_free_deferred);
311 schedule_work(&map->work);
312 }
313 }
314
315 void bpf_map_put(struct bpf_map *map)
316 {
317 __bpf_map_put(map, true);
318 }
319 EXPORT_SYMBOL_GPL(bpf_map_put);
320
321 void bpf_map_put_with_uref(struct bpf_map *map)
322 {
323 bpf_map_put_uref(map);
324 bpf_map_put(map);
325 }
326
327 static int bpf_map_release(struct inode *inode, struct file *filp)
328 {
329 struct bpf_map *map = filp->private_data;
330
331 if (map->ops->map_release)
332 map->ops->map_release(map, filp);
333
334 bpf_map_put_with_uref(map);
335 return 0;
336 }
337
338 #ifdef CONFIG_PROC_FS
339 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
340 {
341 const struct bpf_map *map = filp->private_data;
342 const struct bpf_array *array;
343 u32 owner_prog_type = 0;
344 u32 owner_jited = 0;
345
346 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
347 array = container_of(map, struct bpf_array, map);
348 owner_prog_type = array->owner_prog_type;
349 owner_jited = array->owner_jited;
350 }
351
352 seq_printf(m,
353 "map_type:\t%u\n"
354 "key_size:\t%u\n"
355 "value_size:\t%u\n"
356 "max_entries:\t%u\n"
357 "map_flags:\t%#x\n"
358 "memlock:\t%llu\n"
359 "map_id:\t%u\n",
360 map->map_type,
361 map->key_size,
362 map->value_size,
363 map->max_entries,
364 map->map_flags,
365 map->pages * 1ULL << PAGE_SHIFT,
366 map->id);
367
368 if (owner_prog_type) {
369 seq_printf(m, "owner_prog_type:\t%u\n",
370 owner_prog_type);
371 seq_printf(m, "owner_jited:\t%u\n",
372 owner_jited);
373 }
374 }
375 #endif
376
377 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
378 loff_t *ppos)
379 {
380 /* We need this handler such that alloc_file() enables
381 * f_mode with FMODE_CAN_READ.
382 */
383 return -EINVAL;
384 }
385
386 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
387 size_t siz, loff_t *ppos)
388 {
389 /* We need this handler such that alloc_file() enables
390 * f_mode with FMODE_CAN_WRITE.
391 */
392 return -EINVAL;
393 }
394
395 const struct file_operations bpf_map_fops = {
396 #ifdef CONFIG_PROC_FS
397 .show_fdinfo = bpf_map_show_fdinfo,
398 #endif
399 .release = bpf_map_release,
400 .read = bpf_dummy_read,
401 .write = bpf_dummy_write,
402 };
403
404 int bpf_map_new_fd(struct bpf_map *map, int flags)
405 {
406 int ret;
407
408 ret = security_bpf_map(map, OPEN_FMODE(flags));
409 if (ret < 0)
410 return ret;
411
412 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
413 flags | O_CLOEXEC);
414 }
415
416 int bpf_get_file_flag(int flags)
417 {
418 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
419 return -EINVAL;
420 if (flags & BPF_F_RDONLY)
421 return O_RDONLY;
422 if (flags & BPF_F_WRONLY)
423 return O_WRONLY;
424 return O_RDWR;
425 }
426
427 /* helper macro to check that unused fields 'union bpf_attr' are zero */
428 #define CHECK_ATTR(CMD) \
429 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
430 sizeof(attr->CMD##_LAST_FIELD), 0, \
431 sizeof(*attr) - \
432 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
433 sizeof(attr->CMD##_LAST_FIELD)) != NULL
434
435 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
436 * Return 0 on success and < 0 on error.
437 */
438 static int bpf_obj_name_cpy(char *dst, const char *src)
439 {
440 const char *end = src + BPF_OBJ_NAME_LEN;
441
442 memset(dst, 0, BPF_OBJ_NAME_LEN);
443
444 /* Copy all isalnum() and '_' char */
445 while (src < end && *src) {
446 if (!isalnum(*src) && *src != '_')
447 return -EINVAL;
448 *dst++ = *src++;
449 }
450
451 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
452 if (src == end)
453 return -EINVAL;
454
455 return 0;
456 }
457
458 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
459 /* called via syscall */
460 static int map_create(union bpf_attr *attr)
461 {
462 int numa_node = bpf_map_attr_numa_node(attr);
463 struct bpf_map *map;
464 int f_flags;
465 int err;
466
467 err = CHECK_ATTR(BPF_MAP_CREATE);
468 if (err)
469 return -EINVAL;
470
471 f_flags = bpf_get_file_flag(attr->map_flags);
472 if (f_flags < 0)
473 return f_flags;
474
475 if (numa_node != NUMA_NO_NODE &&
476 ((unsigned int)numa_node >= nr_node_ids ||
477 !node_online(numa_node)))
478 return -EINVAL;
479
480 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
481 map = find_and_alloc_map(attr);
482 if (IS_ERR(map))
483 return PTR_ERR(map);
484
485 err = bpf_obj_name_cpy(map->name, attr->map_name);
486 if (err)
487 goto free_map_nouncharge;
488
489 atomic_set(&map->refcnt, 1);
490 atomic_set(&map->usercnt, 1);
491
492 if (bpf_map_support_seq_show(map) &&
493 (attr->btf_key_type_id || attr->btf_value_type_id)) {
494 struct btf *btf;
495
496 if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
497 err = -EINVAL;
498 goto free_map_nouncharge;
499 }
500
501 btf = btf_get_by_fd(attr->btf_fd);
502 if (IS_ERR(btf)) {
503 err = PTR_ERR(btf);
504 goto free_map_nouncharge;
505 }
506
507 err = map->ops->map_check_btf(map, btf, attr->btf_key_type_id,
508 attr->btf_value_type_id);
509 if (err) {
510 btf_put(btf);
511 goto free_map_nouncharge;
512 }
513
514 map->btf = btf;
515 map->btf_key_type_id = attr->btf_key_type_id;
516 map->btf_value_type_id = attr->btf_value_type_id;
517 }
518
519 err = security_bpf_map_alloc(map);
520 if (err)
521 goto free_map_nouncharge;
522
523 err = bpf_map_init_memlock(map);
524 if (err)
525 goto free_map_sec;
526
527 err = bpf_map_alloc_id(map);
528 if (err)
529 goto free_map;
530
531 err = bpf_map_new_fd(map, f_flags);
532 if (err < 0) {
533 /* failed to allocate fd.
534 * bpf_map_put() is needed because the above
535 * bpf_map_alloc_id() has published the map
536 * to the userspace and the userspace may
537 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
538 */
539 bpf_map_put(map);
540 return err;
541 }
542
543 return err;
544
545 free_map:
546 bpf_map_release_memlock(map);
547 free_map_sec:
548 security_bpf_map_free(map);
549 free_map_nouncharge:
550 btf_put(map->btf);
551 map->ops->map_free(map);
552 return err;
553 }
554
555 /* if error is returned, fd is released.
556 * On success caller should complete fd access with matching fdput()
557 */
558 struct bpf_map *__bpf_map_get(struct fd f)
559 {
560 if (!f.file)
561 return ERR_PTR(-EBADF);
562 if (f.file->f_op != &bpf_map_fops) {
563 fdput(f);
564 return ERR_PTR(-EINVAL);
565 }
566
567 return f.file->private_data;
568 }
569
570 /* prog's and map's refcnt limit */
571 #define BPF_MAX_REFCNT 32768
572
573 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
574 {
575 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
576 atomic_dec(&map->refcnt);
577 return ERR_PTR(-EBUSY);
578 }
579 if (uref)
580 atomic_inc(&map->usercnt);
581 return map;
582 }
583 EXPORT_SYMBOL_GPL(bpf_map_inc);
584
585 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
586 {
587 struct fd f = fdget(ufd);
588 struct bpf_map *map;
589
590 map = __bpf_map_get(f);
591 if (IS_ERR(map))
592 return map;
593
594 map = bpf_map_inc(map, true);
595 fdput(f);
596
597 return map;
598 }
599
600 /* map_idr_lock should have been held */
601 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
602 bool uref)
603 {
604 int refold;
605
606 refold = __atomic_add_unless(&map->refcnt, 1, 0);
607
608 if (refold >= BPF_MAX_REFCNT) {
609 __bpf_map_put(map, false);
610 return ERR_PTR(-EBUSY);
611 }
612
613 if (!refold)
614 return ERR_PTR(-ENOENT);
615
616 if (uref)
617 atomic_inc(&map->usercnt);
618
619 return map;
620 }
621
622 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
623 {
624 return -ENOTSUPP;
625 }
626
627 /* last field in 'union bpf_attr' used by this command */
628 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
629
630 static int map_lookup_elem(union bpf_attr *attr)
631 {
632 void __user *ukey = u64_to_user_ptr(attr->key);
633 void __user *uvalue = u64_to_user_ptr(attr->value);
634 int ufd = attr->map_fd;
635 struct bpf_map *map;
636 void *key, *value, *ptr;
637 u32 value_size;
638 struct fd f;
639 int err;
640
641 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
642 return -EINVAL;
643
644 f = fdget(ufd);
645 map = __bpf_map_get(f);
646 if (IS_ERR(map))
647 return PTR_ERR(map);
648
649 if (!(f.file->f_mode & FMODE_CAN_READ)) {
650 err = -EPERM;
651 goto err_put;
652 }
653
654 key = memdup_user(ukey, map->key_size);
655 if (IS_ERR(key)) {
656 err = PTR_ERR(key);
657 goto err_put;
658 }
659
660 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
661 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
662 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
663 value_size = round_up(map->value_size, 8) * num_possible_cpus();
664 else if (IS_FD_MAP(map))
665 value_size = sizeof(u32);
666 else
667 value_size = map->value_size;
668
669 err = -ENOMEM;
670 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
671 if (!value)
672 goto free_key;
673
674 if (bpf_map_is_dev_bound(map)) {
675 err = bpf_map_offload_lookup_elem(map, key, value);
676 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
677 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
678 err = bpf_percpu_hash_copy(map, key, value);
679 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
680 err = bpf_percpu_array_copy(map, key, value);
681 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
682 err = bpf_stackmap_copy(map, key, value);
683 } else if (IS_FD_ARRAY(map)) {
684 err = bpf_fd_array_map_lookup_elem(map, key, value);
685 } else if (IS_FD_HASH(map)) {
686 err = bpf_fd_htab_map_lookup_elem(map, key, value);
687 } else {
688 rcu_read_lock();
689 ptr = map->ops->map_lookup_elem(map, key);
690 if (ptr)
691 memcpy(value, ptr, value_size);
692 rcu_read_unlock();
693 err = ptr ? 0 : -ENOENT;
694 }
695
696 if (err)
697 goto free_value;
698
699 err = -EFAULT;
700 if (copy_to_user(uvalue, value, value_size) != 0)
701 goto free_value;
702
703 err = 0;
704
705 free_value:
706 kfree(value);
707 free_key:
708 kfree(key);
709 err_put:
710 fdput(f);
711 return err;
712 }
713
714 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
715
716 static int map_update_elem(union bpf_attr *attr)
717 {
718 void __user *ukey = u64_to_user_ptr(attr->key);
719 void __user *uvalue = u64_to_user_ptr(attr->value);
720 int ufd = attr->map_fd;
721 struct bpf_map *map;
722 void *key, *value;
723 u32 value_size;
724 struct fd f;
725 int err;
726
727 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
728 return -EINVAL;
729
730 f = fdget(ufd);
731 map = __bpf_map_get(f);
732 if (IS_ERR(map))
733 return PTR_ERR(map);
734
735 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
736 err = -EPERM;
737 goto err_put;
738 }
739
740 key = memdup_user(ukey, map->key_size);
741 if (IS_ERR(key)) {
742 err = PTR_ERR(key);
743 goto err_put;
744 }
745
746 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
747 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
748 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
749 value_size = round_up(map->value_size, 8) * num_possible_cpus();
750 else
751 value_size = map->value_size;
752
753 err = -ENOMEM;
754 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
755 if (!value)
756 goto free_key;
757
758 err = -EFAULT;
759 if (copy_from_user(value, uvalue, value_size) != 0)
760 goto free_value;
761
762 /* Need to create a kthread, thus must support schedule */
763 if (bpf_map_is_dev_bound(map)) {
764 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
765 goto out;
766 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
767 map->map_type == BPF_MAP_TYPE_SOCKHASH ||
768 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
769 err = map->ops->map_update_elem(map, key, value, attr->flags);
770 goto out;
771 }
772
773 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
774 * inside bpf map update or delete otherwise deadlocks are possible
775 */
776 preempt_disable();
777 __this_cpu_inc(bpf_prog_active);
778 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
779 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
780 err = bpf_percpu_hash_update(map, key, value, attr->flags);
781 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
782 err = bpf_percpu_array_update(map, key, value, attr->flags);
783 } else if (IS_FD_ARRAY(map)) {
784 rcu_read_lock();
785 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
786 attr->flags);
787 rcu_read_unlock();
788 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
789 rcu_read_lock();
790 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
791 attr->flags);
792 rcu_read_unlock();
793 } else {
794 rcu_read_lock();
795 err = map->ops->map_update_elem(map, key, value, attr->flags);
796 rcu_read_unlock();
797 }
798 __this_cpu_dec(bpf_prog_active);
799 preempt_enable();
800 out:
801 free_value:
802 kfree(value);
803 free_key:
804 kfree(key);
805 err_put:
806 fdput(f);
807 return err;
808 }
809
810 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
811
812 static int map_delete_elem(union bpf_attr *attr)
813 {
814 void __user *ukey = u64_to_user_ptr(attr->key);
815 int ufd = attr->map_fd;
816 struct bpf_map *map;
817 struct fd f;
818 void *key;
819 int err;
820
821 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
822 return -EINVAL;
823
824 f = fdget(ufd);
825 map = __bpf_map_get(f);
826 if (IS_ERR(map))
827 return PTR_ERR(map);
828
829 if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
830 err = -EPERM;
831 goto err_put;
832 }
833
834 key = memdup_user(ukey, map->key_size);
835 if (IS_ERR(key)) {
836 err = PTR_ERR(key);
837 goto err_put;
838 }
839
840 if (bpf_map_is_dev_bound(map)) {
841 err = bpf_map_offload_delete_elem(map, key);
842 goto out;
843 }
844
845 preempt_disable();
846 __this_cpu_inc(bpf_prog_active);
847 rcu_read_lock();
848 err = map->ops->map_delete_elem(map, key);
849 rcu_read_unlock();
850 __this_cpu_dec(bpf_prog_active);
851 preempt_enable();
852 out:
853 kfree(key);
854 err_put:
855 fdput(f);
856 return err;
857 }
858
859 /* last field in 'union bpf_attr' used by this command */
860 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
861
862 static int map_get_next_key(union bpf_attr *attr)
863 {
864 void __user *ukey = u64_to_user_ptr(attr->key);
865 void __user *unext_key = u64_to_user_ptr(attr->next_key);
866 int ufd = attr->map_fd;
867 struct bpf_map *map;
868 void *key, *next_key;
869 struct fd f;
870 int err;
871
872 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
873 return -EINVAL;
874
875 f = fdget(ufd);
876 map = __bpf_map_get(f);
877 if (IS_ERR(map))
878 return PTR_ERR(map);
879
880 if (!(f.file->f_mode & FMODE_CAN_READ)) {
881 err = -EPERM;
882 goto err_put;
883 }
884
885 if (ukey) {
886 key = memdup_user(ukey, map->key_size);
887 if (IS_ERR(key)) {
888 err = PTR_ERR(key);
889 goto err_put;
890 }
891 } else {
892 key = NULL;
893 }
894
895 err = -ENOMEM;
896 next_key = kmalloc(map->key_size, GFP_USER);
897 if (!next_key)
898 goto free_key;
899
900 if (bpf_map_is_dev_bound(map)) {
901 err = bpf_map_offload_get_next_key(map, key, next_key);
902 goto out;
903 }
904
905 rcu_read_lock();
906 err = map->ops->map_get_next_key(map, key, next_key);
907 rcu_read_unlock();
908 out:
909 if (err)
910 goto free_next_key;
911
912 err = -EFAULT;
913 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
914 goto free_next_key;
915
916 err = 0;
917
918 free_next_key:
919 kfree(next_key);
920 free_key:
921 kfree(key);
922 err_put:
923 fdput(f);
924 return err;
925 }
926
927 static const struct bpf_prog_ops * const bpf_prog_types[] = {
928 #define BPF_PROG_TYPE(_id, _name) \
929 [_id] = & _name ## _prog_ops,
930 #define BPF_MAP_TYPE(_id, _ops)
931 #include <linux/bpf_types.h>
932 #undef BPF_PROG_TYPE
933 #undef BPF_MAP_TYPE
934 };
935
936 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
937 {
938 const struct bpf_prog_ops *ops;
939
940 if (type >= ARRAY_SIZE(bpf_prog_types))
941 return -EINVAL;
942 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
943 ops = bpf_prog_types[type];
944 if (!ops)
945 return -EINVAL;
946
947 if (!bpf_prog_is_dev_bound(prog->aux))
948 prog->aux->ops = ops;
949 else
950 prog->aux->ops = &bpf_offload_prog_ops;
951 prog->type = type;
952 return 0;
953 }
954
955 /* drop refcnt on maps used by eBPF program and free auxilary data */
956 static void free_used_maps(struct bpf_prog_aux *aux)
957 {
958 int i;
959
960 for (i = 0; i < aux->used_map_cnt; i++)
961 bpf_map_put(aux->used_maps[i]);
962
963 kfree(aux->used_maps);
964 }
965
966 int __bpf_prog_charge(struct user_struct *user, u32 pages)
967 {
968 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
969 unsigned long user_bufs;
970
971 if (user) {
972 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
973 if (user_bufs > memlock_limit) {
974 atomic_long_sub(pages, &user->locked_vm);
975 return -EPERM;
976 }
977 }
978
979 return 0;
980 }
981
982 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
983 {
984 if (user)
985 atomic_long_sub(pages, &user->locked_vm);
986 }
987
988 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
989 {
990 struct user_struct *user = get_current_user();
991 int ret;
992
993 ret = __bpf_prog_charge(user, prog->pages);
994 if (ret) {
995 free_uid(user);
996 return ret;
997 }
998
999 prog->aux->user = user;
1000 return 0;
1001 }
1002
1003 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1004 {
1005 struct user_struct *user = prog->aux->user;
1006
1007 __bpf_prog_uncharge(user, prog->pages);
1008 free_uid(user);
1009 }
1010
1011 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1012 {
1013 int id;
1014
1015 idr_preload(GFP_KERNEL);
1016 spin_lock_bh(&prog_idr_lock);
1017 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1018 if (id > 0)
1019 prog->aux->id = id;
1020 spin_unlock_bh(&prog_idr_lock);
1021 idr_preload_end();
1022
1023 /* id is in [1, INT_MAX) */
1024 if (WARN_ON_ONCE(!id))
1025 return -ENOSPC;
1026
1027 return id > 0 ? 0 : id;
1028 }
1029
1030 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1031 {
1032 /* cBPF to eBPF migrations are currently not in the idr store.
1033 * Offloaded programs are removed from the store when their device
1034 * disappears - even if someone grabs an fd to them they are unusable,
1035 * simply waiting for refcnt to drop to be freed.
1036 */
1037 if (!prog->aux->id)
1038 return;
1039
1040 if (do_idr_lock)
1041 spin_lock_bh(&prog_idr_lock);
1042 else
1043 __acquire(&prog_idr_lock);
1044
1045 idr_remove(&prog_idr, prog->aux->id);
1046 prog->aux->id = 0;
1047
1048 if (do_idr_lock)
1049 spin_unlock_bh(&prog_idr_lock);
1050 else
1051 __release(&prog_idr_lock);
1052 }
1053
1054 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1055 {
1056 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1057
1058 free_used_maps(aux);
1059 bpf_prog_uncharge_memlock(aux->prog);
1060 security_bpf_prog_free(aux);
1061 bpf_prog_free(aux->prog);
1062 }
1063
1064 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1065 {
1066 if (atomic_dec_and_test(&prog->aux->refcnt)) {
1067 /* bpf_prog_free_id() must be called first */
1068 bpf_prog_free_id(prog, do_idr_lock);
1069 bpf_prog_kallsyms_del_all(prog);
1070
1071 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1072 }
1073 }
1074
1075 void bpf_prog_put(struct bpf_prog *prog)
1076 {
1077 __bpf_prog_put(prog, true);
1078 }
1079 EXPORT_SYMBOL_GPL(bpf_prog_put);
1080
1081 static int bpf_prog_release(struct inode *inode, struct file *filp)
1082 {
1083 struct bpf_prog *prog = filp->private_data;
1084
1085 bpf_prog_put(prog);
1086 return 0;
1087 }
1088
1089 #ifdef CONFIG_PROC_FS
1090 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1091 {
1092 const struct bpf_prog *prog = filp->private_data;
1093 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1094
1095 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1096 seq_printf(m,
1097 "prog_type:\t%u\n"
1098 "prog_jited:\t%u\n"
1099 "prog_tag:\t%s\n"
1100 "memlock:\t%llu\n"
1101 "prog_id:\t%u\n",
1102 prog->type,
1103 prog->jited,
1104 prog_tag,
1105 prog->pages * 1ULL << PAGE_SHIFT,
1106 prog->aux->id);
1107 }
1108 #endif
1109
1110 const struct file_operations bpf_prog_fops = {
1111 #ifdef CONFIG_PROC_FS
1112 .show_fdinfo = bpf_prog_show_fdinfo,
1113 #endif
1114 .release = bpf_prog_release,
1115 .read = bpf_dummy_read,
1116 .write = bpf_dummy_write,
1117 };
1118
1119 int bpf_prog_new_fd(struct bpf_prog *prog)
1120 {
1121 int ret;
1122
1123 ret = security_bpf_prog(prog);
1124 if (ret < 0)
1125 return ret;
1126
1127 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1128 O_RDWR | O_CLOEXEC);
1129 }
1130
1131 static struct bpf_prog *____bpf_prog_get(struct fd f)
1132 {
1133 if (!f.file)
1134 return ERR_PTR(-EBADF);
1135 if (f.file->f_op != &bpf_prog_fops) {
1136 fdput(f);
1137 return ERR_PTR(-EINVAL);
1138 }
1139
1140 return f.file->private_data;
1141 }
1142
1143 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1144 {
1145 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1146 atomic_sub(i, &prog->aux->refcnt);
1147 return ERR_PTR(-EBUSY);
1148 }
1149 return prog;
1150 }
1151 EXPORT_SYMBOL_GPL(bpf_prog_add);
1152
1153 void bpf_prog_sub(struct bpf_prog *prog, int i)
1154 {
1155 /* Only to be used for undoing previous bpf_prog_add() in some
1156 * error path. We still know that another entity in our call
1157 * path holds a reference to the program, thus atomic_sub() can
1158 * be safely used in such cases!
1159 */
1160 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1161 }
1162 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1163
1164 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1165 {
1166 return bpf_prog_add(prog, 1);
1167 }
1168 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1169
1170 /* prog_idr_lock should have been held */
1171 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1172 {
1173 int refold;
1174
1175 refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
1176
1177 if (refold >= BPF_MAX_REFCNT) {
1178 __bpf_prog_put(prog, false);
1179 return ERR_PTR(-EBUSY);
1180 }
1181
1182 if (!refold)
1183 return ERR_PTR(-ENOENT);
1184
1185 return prog;
1186 }
1187 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1188
1189 bool bpf_prog_get_ok(struct bpf_prog *prog,
1190 enum bpf_prog_type *attach_type, bool attach_drv)
1191 {
1192 /* not an attachment, just a refcount inc, always allow */
1193 if (!attach_type)
1194 return true;
1195
1196 if (prog->type != *attach_type)
1197 return false;
1198 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1199 return false;
1200
1201 return true;
1202 }
1203
1204 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1205 bool attach_drv)
1206 {
1207 struct fd f = fdget(ufd);
1208 struct bpf_prog *prog;
1209
1210 prog = ____bpf_prog_get(f);
1211 if (IS_ERR(prog))
1212 return prog;
1213 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1214 prog = ERR_PTR(-EINVAL);
1215 goto out;
1216 }
1217
1218 prog = bpf_prog_inc(prog);
1219 out:
1220 fdput(f);
1221 return prog;
1222 }
1223
1224 struct bpf_prog *bpf_prog_get(u32 ufd)
1225 {
1226 return __bpf_prog_get(ufd, NULL, false);
1227 }
1228
1229 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1230 bool attach_drv)
1231 {
1232 return __bpf_prog_get(ufd, &type, attach_drv);
1233 }
1234 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1235
1236 /* Initially all BPF programs could be loaded w/o specifying
1237 * expected_attach_type. Later for some of them specifying expected_attach_type
1238 * at load time became required so that program could be validated properly.
1239 * Programs of types that are allowed to be loaded both w/ and w/o (for
1240 * backward compatibility) expected_attach_type, should have the default attach
1241 * type assigned to expected_attach_type for the latter case, so that it can be
1242 * validated later at attach time.
1243 *
1244 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1245 * prog type requires it but has some attach types that have to be backward
1246 * compatible.
1247 */
1248 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1249 {
1250 switch (attr->prog_type) {
1251 case BPF_PROG_TYPE_CGROUP_SOCK:
1252 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1253 * exist so checking for non-zero is the way to go here.
1254 */
1255 if (!attr->expected_attach_type)
1256 attr->expected_attach_type =
1257 BPF_CGROUP_INET_SOCK_CREATE;
1258 break;
1259 }
1260 }
1261
1262 static int
1263 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1264 enum bpf_attach_type expected_attach_type)
1265 {
1266 switch (prog_type) {
1267 case BPF_PROG_TYPE_CGROUP_SOCK:
1268 switch (expected_attach_type) {
1269 case BPF_CGROUP_INET_SOCK_CREATE:
1270 case BPF_CGROUP_INET4_POST_BIND:
1271 case BPF_CGROUP_INET6_POST_BIND:
1272 return 0;
1273 default:
1274 return -EINVAL;
1275 }
1276 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1277 switch (expected_attach_type) {
1278 case BPF_CGROUP_INET4_BIND:
1279 case BPF_CGROUP_INET6_BIND:
1280 case BPF_CGROUP_INET4_CONNECT:
1281 case BPF_CGROUP_INET6_CONNECT:
1282 case BPF_CGROUP_UDP4_SENDMSG:
1283 case BPF_CGROUP_UDP6_SENDMSG:
1284 return 0;
1285 default:
1286 return -EINVAL;
1287 }
1288 default:
1289 return 0;
1290 }
1291 }
1292
1293 /* last field in 'union bpf_attr' used by this command */
1294 #define BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1295
1296 static int bpf_prog_load(union bpf_attr *attr)
1297 {
1298 enum bpf_prog_type type = attr->prog_type;
1299 struct bpf_prog *prog;
1300 int err;
1301 char license[128];
1302 bool is_gpl;
1303
1304 if (CHECK_ATTR(BPF_PROG_LOAD))
1305 return -EINVAL;
1306
1307 if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1308 return -EINVAL;
1309
1310 /* copy eBPF program license from user space */
1311 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1312 sizeof(license) - 1) < 0)
1313 return -EFAULT;
1314 license[sizeof(license) - 1] = 0;
1315
1316 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1317 is_gpl = license_is_gpl_compatible(license);
1318
1319 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1320 return -E2BIG;
1321
1322 if (type == BPF_PROG_TYPE_KPROBE &&
1323 attr->kern_version != LINUX_VERSION_CODE)
1324 return -EINVAL;
1325
1326 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1327 type != BPF_PROG_TYPE_CGROUP_SKB &&
1328 !capable(CAP_SYS_ADMIN))
1329 return -EPERM;
1330
1331 bpf_prog_load_fixup_attach_type(attr);
1332 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1333 return -EINVAL;
1334
1335 /* plain bpf_prog allocation */
1336 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1337 if (!prog)
1338 return -ENOMEM;
1339
1340 prog->expected_attach_type = attr->expected_attach_type;
1341
1342 prog->aux->offload_requested = !!attr->prog_ifindex;
1343
1344 err = security_bpf_prog_alloc(prog->aux);
1345 if (err)
1346 goto free_prog_nouncharge;
1347
1348 err = bpf_prog_charge_memlock(prog);
1349 if (err)
1350 goto free_prog_sec;
1351
1352 prog->len = attr->insn_cnt;
1353
1354 err = -EFAULT;
1355 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1356 bpf_prog_insn_size(prog)) != 0)
1357 goto free_prog;
1358
1359 prog->orig_prog = NULL;
1360 prog->jited = 0;
1361
1362 atomic_set(&prog->aux->refcnt, 1);
1363 prog->gpl_compatible = is_gpl ? 1 : 0;
1364
1365 if (bpf_prog_is_dev_bound(prog->aux)) {
1366 err = bpf_prog_offload_init(prog, attr);
1367 if (err)
1368 goto free_prog;
1369 }
1370
1371 /* find program type: socket_filter vs tracing_filter */
1372 err = find_prog_type(type, prog);
1373 if (err < 0)
1374 goto free_prog;
1375
1376 prog->aux->load_time = ktime_get_boot_ns();
1377 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1378 if (err)
1379 goto free_prog;
1380
1381 /* run eBPF verifier */
1382 err = bpf_check(&prog, attr);
1383 if (err < 0)
1384 goto free_used_maps;
1385
1386 prog = bpf_prog_select_runtime(prog, &err);
1387 if (err < 0)
1388 goto free_used_maps;
1389
1390 err = bpf_prog_alloc_id(prog);
1391 if (err)
1392 goto free_used_maps;
1393
1394 err = bpf_prog_new_fd(prog);
1395 if (err < 0) {
1396 /* failed to allocate fd.
1397 * bpf_prog_put() is needed because the above
1398 * bpf_prog_alloc_id() has published the prog
1399 * to the userspace and the userspace may
1400 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1401 */
1402 bpf_prog_put(prog);
1403 return err;
1404 }
1405
1406 bpf_prog_kallsyms_add(prog);
1407 return err;
1408
1409 free_used_maps:
1410 bpf_prog_kallsyms_del_subprogs(prog);
1411 free_used_maps(prog->aux);
1412 free_prog:
1413 bpf_prog_uncharge_memlock(prog);
1414 free_prog_sec:
1415 security_bpf_prog_free(prog->aux);
1416 free_prog_nouncharge:
1417 bpf_prog_free(prog);
1418 return err;
1419 }
1420
1421 #define BPF_OBJ_LAST_FIELD file_flags
1422
1423 static int bpf_obj_pin(const union bpf_attr *attr)
1424 {
1425 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1426 return -EINVAL;
1427
1428 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1429 }
1430
1431 static int bpf_obj_get(const union bpf_attr *attr)
1432 {
1433 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1434 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1435 return -EINVAL;
1436
1437 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1438 attr->file_flags);
1439 }
1440
1441 struct bpf_raw_tracepoint {
1442 struct bpf_raw_event_map *btp;
1443 struct bpf_prog *prog;
1444 };
1445
1446 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1447 {
1448 struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1449
1450 if (raw_tp->prog) {
1451 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1452 bpf_prog_put(raw_tp->prog);
1453 }
1454 kfree(raw_tp);
1455 return 0;
1456 }
1457
1458 static const struct file_operations bpf_raw_tp_fops = {
1459 .release = bpf_raw_tracepoint_release,
1460 .read = bpf_dummy_read,
1461 .write = bpf_dummy_write,
1462 };
1463
1464 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1465
1466 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1467 {
1468 struct bpf_raw_tracepoint *raw_tp;
1469 struct bpf_raw_event_map *btp;
1470 struct bpf_prog *prog;
1471 char tp_name[128];
1472 int tp_fd, err;
1473
1474 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1475 sizeof(tp_name) - 1) < 0)
1476 return -EFAULT;
1477 tp_name[sizeof(tp_name) - 1] = 0;
1478
1479 btp = bpf_find_raw_tracepoint(tp_name);
1480 if (!btp)
1481 return -ENOENT;
1482
1483 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1484 if (!raw_tp)
1485 return -ENOMEM;
1486 raw_tp->btp = btp;
1487
1488 prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1489 BPF_PROG_TYPE_RAW_TRACEPOINT);
1490 if (IS_ERR(prog)) {
1491 err = PTR_ERR(prog);
1492 goto out_free_tp;
1493 }
1494
1495 err = bpf_probe_register(raw_tp->btp, prog);
1496 if (err)
1497 goto out_put_prog;
1498
1499 raw_tp->prog = prog;
1500 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1501 O_CLOEXEC);
1502 if (tp_fd < 0) {
1503 bpf_probe_unregister(raw_tp->btp, prog);
1504 err = tp_fd;
1505 goto out_put_prog;
1506 }
1507 return tp_fd;
1508
1509 out_put_prog:
1510 bpf_prog_put(prog);
1511 out_free_tp:
1512 kfree(raw_tp);
1513 return err;
1514 }
1515
1516 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1517 enum bpf_attach_type attach_type)
1518 {
1519 switch (prog->type) {
1520 case BPF_PROG_TYPE_CGROUP_SOCK:
1521 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1522 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1523 default:
1524 return 0;
1525 }
1526 }
1527
1528 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1529
1530 #define BPF_F_ATTACH_MASK \
1531 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1532
1533 static int bpf_prog_attach(const union bpf_attr *attr)
1534 {
1535 enum bpf_prog_type ptype;
1536 struct bpf_prog *prog;
1537 int ret;
1538
1539 if (!capable(CAP_NET_ADMIN))
1540 return -EPERM;
1541
1542 if (CHECK_ATTR(BPF_PROG_ATTACH))
1543 return -EINVAL;
1544
1545 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1546 return -EINVAL;
1547
1548 switch (attr->attach_type) {
1549 case BPF_CGROUP_INET_INGRESS:
1550 case BPF_CGROUP_INET_EGRESS:
1551 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1552 break;
1553 case BPF_CGROUP_INET_SOCK_CREATE:
1554 case BPF_CGROUP_INET4_POST_BIND:
1555 case BPF_CGROUP_INET6_POST_BIND:
1556 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1557 break;
1558 case BPF_CGROUP_INET4_BIND:
1559 case BPF_CGROUP_INET6_BIND:
1560 case BPF_CGROUP_INET4_CONNECT:
1561 case BPF_CGROUP_INET6_CONNECT:
1562 case BPF_CGROUP_UDP4_SENDMSG:
1563 case BPF_CGROUP_UDP6_SENDMSG:
1564 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1565 break;
1566 case BPF_CGROUP_SOCK_OPS:
1567 ptype = BPF_PROG_TYPE_SOCK_OPS;
1568 break;
1569 case BPF_CGROUP_DEVICE:
1570 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1571 break;
1572 case BPF_SK_MSG_VERDICT:
1573 ptype = BPF_PROG_TYPE_SK_MSG;
1574 break;
1575 case BPF_SK_SKB_STREAM_PARSER:
1576 case BPF_SK_SKB_STREAM_VERDICT:
1577 ptype = BPF_PROG_TYPE_SK_SKB;
1578 break;
1579 case BPF_LIRC_MODE2:
1580 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1581 break;
1582 default:
1583 return -EINVAL;
1584 }
1585
1586 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1587 if (IS_ERR(prog))
1588 return PTR_ERR(prog);
1589
1590 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1591 bpf_prog_put(prog);
1592 return -EINVAL;
1593 }
1594
1595 switch (ptype) {
1596 case BPF_PROG_TYPE_SK_SKB:
1597 case BPF_PROG_TYPE_SK_MSG:
1598 ret = sockmap_get_from_fd(attr, ptype, prog);
1599 break;
1600 case BPF_PROG_TYPE_LIRC_MODE2:
1601 ret = lirc_prog_attach(attr, prog);
1602 break;
1603 default:
1604 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1605 }
1606
1607 if (ret)
1608 bpf_prog_put(prog);
1609 return ret;
1610 }
1611
1612 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1613
1614 static int bpf_prog_detach(const union bpf_attr *attr)
1615 {
1616 enum bpf_prog_type ptype;
1617
1618 if (!capable(CAP_NET_ADMIN))
1619 return -EPERM;
1620
1621 if (CHECK_ATTR(BPF_PROG_DETACH))
1622 return -EINVAL;
1623
1624 switch (attr->attach_type) {
1625 case BPF_CGROUP_INET_INGRESS:
1626 case BPF_CGROUP_INET_EGRESS:
1627 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1628 break;
1629 case BPF_CGROUP_INET_SOCK_CREATE:
1630 case BPF_CGROUP_INET4_POST_BIND:
1631 case BPF_CGROUP_INET6_POST_BIND:
1632 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1633 break;
1634 case BPF_CGROUP_INET4_BIND:
1635 case BPF_CGROUP_INET6_BIND:
1636 case BPF_CGROUP_INET4_CONNECT:
1637 case BPF_CGROUP_INET6_CONNECT:
1638 case BPF_CGROUP_UDP4_SENDMSG:
1639 case BPF_CGROUP_UDP6_SENDMSG:
1640 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1641 break;
1642 case BPF_CGROUP_SOCK_OPS:
1643 ptype = BPF_PROG_TYPE_SOCK_OPS;
1644 break;
1645 case BPF_CGROUP_DEVICE:
1646 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1647 break;
1648 case BPF_SK_MSG_VERDICT:
1649 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, NULL);
1650 case BPF_SK_SKB_STREAM_PARSER:
1651 case BPF_SK_SKB_STREAM_VERDICT:
1652 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, NULL);
1653 case BPF_LIRC_MODE2:
1654 return lirc_prog_detach(attr);
1655 default:
1656 return -EINVAL;
1657 }
1658
1659 return cgroup_bpf_prog_detach(attr, ptype);
1660 }
1661
1662 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1663
1664 static int bpf_prog_query(const union bpf_attr *attr,
1665 union bpf_attr __user *uattr)
1666 {
1667 if (!capable(CAP_NET_ADMIN))
1668 return -EPERM;
1669 if (CHECK_ATTR(BPF_PROG_QUERY))
1670 return -EINVAL;
1671 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1672 return -EINVAL;
1673
1674 switch (attr->query.attach_type) {
1675 case BPF_CGROUP_INET_INGRESS:
1676 case BPF_CGROUP_INET_EGRESS:
1677 case BPF_CGROUP_INET_SOCK_CREATE:
1678 case BPF_CGROUP_INET4_BIND:
1679 case BPF_CGROUP_INET6_BIND:
1680 case BPF_CGROUP_INET4_POST_BIND:
1681 case BPF_CGROUP_INET6_POST_BIND:
1682 case BPF_CGROUP_INET4_CONNECT:
1683 case BPF_CGROUP_INET6_CONNECT:
1684 case BPF_CGROUP_UDP4_SENDMSG:
1685 case BPF_CGROUP_UDP6_SENDMSG:
1686 case BPF_CGROUP_SOCK_OPS:
1687 case BPF_CGROUP_DEVICE:
1688 break;
1689 case BPF_LIRC_MODE2:
1690 return lirc_prog_query(attr, uattr);
1691 default:
1692 return -EINVAL;
1693 }
1694
1695 return cgroup_bpf_prog_query(attr, uattr);
1696 }
1697
1698 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1699
1700 static int bpf_prog_test_run(const union bpf_attr *attr,
1701 union bpf_attr __user *uattr)
1702 {
1703 struct bpf_prog *prog;
1704 int ret = -ENOTSUPP;
1705
1706 if (!capable(CAP_SYS_ADMIN))
1707 return -EPERM;
1708 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1709 return -EINVAL;
1710
1711 prog = bpf_prog_get(attr->test.prog_fd);
1712 if (IS_ERR(prog))
1713 return PTR_ERR(prog);
1714
1715 if (prog->aux->ops->test_run)
1716 ret = prog->aux->ops->test_run(prog, attr, uattr);
1717
1718 bpf_prog_put(prog);
1719 return ret;
1720 }
1721
1722 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1723
1724 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1725 union bpf_attr __user *uattr,
1726 struct idr *idr,
1727 spinlock_t *lock)
1728 {
1729 u32 next_id = attr->start_id;
1730 int err = 0;
1731
1732 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1733 return -EINVAL;
1734
1735 if (!capable(CAP_SYS_ADMIN))
1736 return -EPERM;
1737
1738 next_id++;
1739 spin_lock_bh(lock);
1740 if (!idr_get_next(idr, &next_id))
1741 err = -ENOENT;
1742 spin_unlock_bh(lock);
1743
1744 if (!err)
1745 err = put_user(next_id, &uattr->next_id);
1746
1747 return err;
1748 }
1749
1750 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1751
1752 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1753 {
1754 struct bpf_prog *prog;
1755 u32 id = attr->prog_id;
1756 int fd;
1757
1758 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1759 return -EINVAL;
1760
1761 if (!capable(CAP_SYS_ADMIN))
1762 return -EPERM;
1763
1764 spin_lock_bh(&prog_idr_lock);
1765 prog = idr_find(&prog_idr, id);
1766 if (prog)
1767 prog = bpf_prog_inc_not_zero(prog);
1768 else
1769 prog = ERR_PTR(-ENOENT);
1770 spin_unlock_bh(&prog_idr_lock);
1771
1772 if (IS_ERR(prog))
1773 return PTR_ERR(prog);
1774
1775 fd = bpf_prog_new_fd(prog);
1776 if (fd < 0)
1777 bpf_prog_put(prog);
1778
1779 return fd;
1780 }
1781
1782 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1783
1784 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1785 {
1786 struct bpf_map *map;
1787 u32 id = attr->map_id;
1788 int f_flags;
1789 int fd;
1790
1791 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1792 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1793 return -EINVAL;
1794
1795 if (!capable(CAP_SYS_ADMIN))
1796 return -EPERM;
1797
1798 f_flags = bpf_get_file_flag(attr->open_flags);
1799 if (f_flags < 0)
1800 return f_flags;
1801
1802 spin_lock_bh(&map_idr_lock);
1803 map = idr_find(&map_idr, id);
1804 if (map)
1805 map = bpf_map_inc_not_zero(map, true);
1806 else
1807 map = ERR_PTR(-ENOENT);
1808 spin_unlock_bh(&map_idr_lock);
1809
1810 if (IS_ERR(map))
1811 return PTR_ERR(map);
1812
1813 fd = bpf_map_new_fd(map, f_flags);
1814 if (fd < 0)
1815 bpf_map_put(map);
1816
1817 return fd;
1818 }
1819
1820 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1821 unsigned long addr)
1822 {
1823 int i;
1824
1825 for (i = 0; i < prog->aux->used_map_cnt; i++)
1826 if (prog->aux->used_maps[i] == (void *)addr)
1827 return prog->aux->used_maps[i];
1828 return NULL;
1829 }
1830
1831 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1832 {
1833 const struct bpf_map *map;
1834 struct bpf_insn *insns;
1835 u64 imm;
1836 int i;
1837
1838 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1839 GFP_USER);
1840 if (!insns)
1841 return insns;
1842
1843 for (i = 0; i < prog->len; i++) {
1844 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1845 insns[i].code = BPF_JMP | BPF_CALL;
1846 insns[i].imm = BPF_FUNC_tail_call;
1847 /* fall-through */
1848 }
1849 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1850 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1851 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1852 insns[i].code = BPF_JMP | BPF_CALL;
1853 if (!bpf_dump_raw_ok())
1854 insns[i].imm = 0;
1855 continue;
1856 }
1857
1858 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1859 continue;
1860
1861 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1862 map = bpf_map_from_imm(prog, imm);
1863 if (map) {
1864 insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1865 insns[i].imm = map->id;
1866 insns[i + 1].imm = 0;
1867 continue;
1868 }
1869
1870 if (!bpf_dump_raw_ok() &&
1871 imm == (unsigned long)prog->aux) {
1872 insns[i].imm = 0;
1873 insns[i + 1].imm = 0;
1874 continue;
1875 }
1876 }
1877
1878 return insns;
1879 }
1880
1881 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1882 const union bpf_attr *attr,
1883 union bpf_attr __user *uattr)
1884 {
1885 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1886 struct bpf_prog_info info = {};
1887 u32 info_len = attr->info.info_len;
1888 char __user *uinsns;
1889 u32 ulen;
1890 int err;
1891
1892 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1893 if (err)
1894 return err;
1895 info_len = min_t(u32, sizeof(info), info_len);
1896
1897 if (copy_from_user(&info, uinfo, info_len))
1898 return -EFAULT;
1899
1900 info.type = prog->type;
1901 info.id = prog->aux->id;
1902 info.load_time = prog->aux->load_time;
1903 info.created_by_uid = from_kuid_munged(current_user_ns(),
1904 prog->aux->user->uid);
1905 info.gpl_compatible = prog->gpl_compatible;
1906
1907 memcpy(info.tag, prog->tag, sizeof(prog->tag));
1908 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1909
1910 ulen = info.nr_map_ids;
1911 info.nr_map_ids = prog->aux->used_map_cnt;
1912 ulen = min_t(u32, info.nr_map_ids, ulen);
1913 if (ulen) {
1914 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1915 u32 i;
1916
1917 for (i = 0; i < ulen; i++)
1918 if (put_user(prog->aux->used_maps[i]->id,
1919 &user_map_ids[i]))
1920 return -EFAULT;
1921 }
1922
1923 if (!capable(CAP_SYS_ADMIN)) {
1924 info.jited_prog_len = 0;
1925 info.xlated_prog_len = 0;
1926 info.nr_jited_ksyms = 0;
1927 goto done;
1928 }
1929
1930 ulen = info.xlated_prog_len;
1931 info.xlated_prog_len = bpf_prog_insn_size(prog);
1932 if (info.xlated_prog_len && ulen) {
1933 struct bpf_insn *insns_sanitized;
1934 bool fault;
1935
1936 if (prog->blinded && !bpf_dump_raw_ok()) {
1937 info.xlated_prog_insns = 0;
1938 goto done;
1939 }
1940 insns_sanitized = bpf_insn_prepare_dump(prog);
1941 if (!insns_sanitized)
1942 return -ENOMEM;
1943 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1944 ulen = min_t(u32, info.xlated_prog_len, ulen);
1945 fault = copy_to_user(uinsns, insns_sanitized, ulen);
1946 kfree(insns_sanitized);
1947 if (fault)
1948 return -EFAULT;
1949 }
1950
1951 if (bpf_prog_is_dev_bound(prog->aux)) {
1952 err = bpf_prog_offload_info_fill(&info, prog);
1953 if (err)
1954 return err;
1955 goto done;
1956 }
1957
1958 /* NOTE: the following code is supposed to be skipped for offload.
1959 * bpf_prog_offload_info_fill() is the place to fill similar fields
1960 * for offload.
1961 */
1962 ulen = info.jited_prog_len;
1963 if (prog->aux->func_cnt) {
1964 u32 i;
1965
1966 info.jited_prog_len = 0;
1967 for (i = 0; i < prog->aux->func_cnt; i++)
1968 info.jited_prog_len += prog->aux->func[i]->jited_len;
1969 } else {
1970 info.jited_prog_len = prog->jited_len;
1971 }
1972
1973 if (info.jited_prog_len && ulen) {
1974 if (bpf_dump_raw_ok()) {
1975 uinsns = u64_to_user_ptr(info.jited_prog_insns);
1976 ulen = min_t(u32, info.jited_prog_len, ulen);
1977
1978 /* for multi-function programs, copy the JITed
1979 * instructions for all the functions
1980 */
1981 if (prog->aux->func_cnt) {
1982 u32 len, free, i;
1983 u8 *img;
1984
1985 free = ulen;
1986 for (i = 0; i < prog->aux->func_cnt; i++) {
1987 len = prog->aux->func[i]->jited_len;
1988 len = min_t(u32, len, free);
1989 img = (u8 *) prog->aux->func[i]->bpf_func;
1990 if (copy_to_user(uinsns, img, len))
1991 return -EFAULT;
1992 uinsns += len;
1993 free -= len;
1994 if (!free)
1995 break;
1996 }
1997 } else {
1998 if (copy_to_user(uinsns, prog->bpf_func, ulen))
1999 return -EFAULT;
2000 }
2001 } else {
2002 info.jited_prog_insns = 0;
2003 }
2004 }
2005
2006 ulen = info.nr_jited_ksyms;
2007 info.nr_jited_ksyms = prog->aux->func_cnt;
2008 if (info.nr_jited_ksyms && ulen) {
2009 if (bpf_dump_raw_ok()) {
2010 u64 __user *user_ksyms;
2011 ulong ksym_addr;
2012 u32 i;
2013
2014 /* copy the address of the kernel symbol
2015 * corresponding to each function
2016 */
2017 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2018 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2019 for (i = 0; i < ulen; i++) {
2020 ksym_addr = (ulong) prog->aux->func[i]->bpf_func;
2021 ksym_addr &= PAGE_MASK;
2022 if (put_user((u64) ksym_addr, &user_ksyms[i]))
2023 return -EFAULT;
2024 }
2025 } else {
2026 info.jited_ksyms = 0;
2027 }
2028 }
2029
2030 ulen = info.nr_jited_func_lens;
2031 info.nr_jited_func_lens = prog->aux->func_cnt;
2032 if (info.nr_jited_func_lens && ulen) {
2033 if (bpf_dump_raw_ok()) {
2034 u32 __user *user_lens;
2035 u32 func_len, i;
2036
2037 /* copy the JITed image lengths for each function */
2038 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2039 user_lens = u64_to_user_ptr(info.jited_func_lens);
2040 for (i = 0; i < ulen; i++) {
2041 func_len = prog->aux->func[i]->jited_len;
2042 if (put_user(func_len, &user_lens[i]))
2043 return -EFAULT;
2044 }
2045 } else {
2046 info.jited_func_lens = 0;
2047 }
2048 }
2049
2050 done:
2051 if (copy_to_user(uinfo, &info, info_len) ||
2052 put_user(info_len, &uattr->info.info_len))
2053 return -EFAULT;
2054
2055 return 0;
2056 }
2057
2058 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2059 const union bpf_attr *attr,
2060 union bpf_attr __user *uattr)
2061 {
2062 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2063 struct bpf_map_info info = {};
2064 u32 info_len = attr->info.info_len;
2065 int err;
2066
2067 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2068 if (err)
2069 return err;
2070 info_len = min_t(u32, sizeof(info), info_len);
2071
2072 info.type = map->map_type;
2073 info.id = map->id;
2074 info.key_size = map->key_size;
2075 info.value_size = map->value_size;
2076 info.max_entries = map->max_entries;
2077 info.map_flags = map->map_flags;
2078 memcpy(info.name, map->name, sizeof(map->name));
2079
2080 if (map->btf) {
2081 info.btf_id = btf_id(map->btf);
2082 info.btf_key_type_id = map->btf_key_type_id;
2083 info.btf_value_type_id = map->btf_value_type_id;
2084 }
2085
2086 if (bpf_map_is_dev_bound(map)) {
2087 err = bpf_map_offload_info_fill(&info, map);
2088 if (err)
2089 return err;
2090 }
2091
2092 if (copy_to_user(uinfo, &info, info_len) ||
2093 put_user(info_len, &uattr->info.info_len))
2094 return -EFAULT;
2095
2096 return 0;
2097 }
2098
2099 static int bpf_btf_get_info_by_fd(struct btf *btf,
2100 const union bpf_attr *attr,
2101 union bpf_attr __user *uattr)
2102 {
2103 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2104 u32 info_len = attr->info.info_len;
2105 int err;
2106
2107 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2108 if (err)
2109 return err;
2110
2111 return btf_get_info_by_fd(btf, attr, uattr);
2112 }
2113
2114 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2115
2116 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2117 union bpf_attr __user *uattr)
2118 {
2119 int ufd = attr->info.bpf_fd;
2120 struct fd f;
2121 int err;
2122
2123 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2124 return -EINVAL;
2125
2126 f = fdget(ufd);
2127 if (!f.file)
2128 return -EBADFD;
2129
2130 if (f.file->f_op == &bpf_prog_fops)
2131 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2132 uattr);
2133 else if (f.file->f_op == &bpf_map_fops)
2134 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2135 uattr);
2136 else if (f.file->f_op == &btf_fops)
2137 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2138 else
2139 err = -EINVAL;
2140
2141 fdput(f);
2142 return err;
2143 }
2144
2145 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2146
2147 static int bpf_btf_load(const union bpf_attr *attr)
2148 {
2149 if (CHECK_ATTR(BPF_BTF_LOAD))
2150 return -EINVAL;
2151
2152 if (!capable(CAP_SYS_ADMIN))
2153 return -EPERM;
2154
2155 return btf_new_fd(attr);
2156 }
2157
2158 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2159
2160 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2161 {
2162 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2163 return -EINVAL;
2164
2165 if (!capable(CAP_SYS_ADMIN))
2166 return -EPERM;
2167
2168 return btf_get_fd_by_id(attr->btf_id);
2169 }
2170
2171 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2172 union bpf_attr __user *uattr,
2173 u32 prog_id, u32 fd_type,
2174 const char *buf, u64 probe_offset,
2175 u64 probe_addr)
2176 {
2177 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2178 u32 len = buf ? strlen(buf) : 0, input_len;
2179 int err = 0;
2180
2181 if (put_user(len, &uattr->task_fd_query.buf_len))
2182 return -EFAULT;
2183 input_len = attr->task_fd_query.buf_len;
2184 if (input_len && ubuf) {
2185 if (!len) {
2186 /* nothing to copy, just make ubuf NULL terminated */
2187 char zero = '\0';
2188
2189 if (put_user(zero, ubuf))
2190 return -EFAULT;
2191 } else if (input_len >= len + 1) {
2192 /* ubuf can hold the string with NULL terminator */
2193 if (copy_to_user(ubuf, buf, len + 1))
2194 return -EFAULT;
2195 } else {
2196 /* ubuf cannot hold the string with NULL terminator,
2197 * do a partial copy with NULL terminator.
2198 */
2199 char zero = '\0';
2200
2201 err = -ENOSPC;
2202 if (copy_to_user(ubuf, buf, input_len - 1))
2203 return -EFAULT;
2204 if (put_user(zero, ubuf + input_len - 1))
2205 return -EFAULT;
2206 }
2207 }
2208
2209 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2210 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2211 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2212 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2213 return -EFAULT;
2214
2215 return err;
2216 }
2217
2218 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2219
2220 static int bpf_task_fd_query(const union bpf_attr *attr,
2221 union bpf_attr __user *uattr)
2222 {
2223 pid_t pid = attr->task_fd_query.pid;
2224 u32 fd = attr->task_fd_query.fd;
2225 const struct perf_event *event;
2226 struct files_struct *files;
2227 struct task_struct *task;
2228 struct file *file;
2229 int err;
2230
2231 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2232 return -EINVAL;
2233
2234 if (!capable(CAP_SYS_ADMIN))
2235 return -EPERM;
2236
2237 if (attr->task_fd_query.flags != 0)
2238 return -EINVAL;
2239
2240 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2241 if (!task)
2242 return -ENOENT;
2243
2244 files = get_files_struct(task);
2245 put_task_struct(task);
2246 if (!files)
2247 return -ENOENT;
2248
2249 err = 0;
2250 spin_lock(&files->file_lock);
2251 file = fcheck_files(files, fd);
2252 if (!file)
2253 err = -EBADF;
2254 else
2255 get_file(file);
2256 spin_unlock(&files->file_lock);
2257 put_files_struct(files);
2258
2259 if (err)
2260 goto out;
2261
2262 if (file->f_op == &bpf_raw_tp_fops) {
2263 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2264 struct bpf_raw_event_map *btp = raw_tp->btp;
2265
2266 err = bpf_task_fd_query_copy(attr, uattr,
2267 raw_tp->prog->aux->id,
2268 BPF_FD_TYPE_RAW_TRACEPOINT,
2269 btp->tp->name, 0, 0);
2270 goto put_file;
2271 }
2272
2273 event = perf_get_event(file);
2274 if (!IS_ERR(event)) {
2275 u64 probe_offset, probe_addr;
2276 u32 prog_id, fd_type;
2277 const char *buf;
2278
2279 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2280 &buf, &probe_offset,
2281 &probe_addr);
2282 if (!err)
2283 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2284 fd_type, buf,
2285 probe_offset,
2286 probe_addr);
2287 goto put_file;
2288 }
2289
2290 err = -ENOTSUPP;
2291 put_file:
2292 fput(file);
2293 out:
2294 return err;
2295 }
2296
2297 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2298 {
2299 union bpf_attr attr = {};
2300 int err;
2301
2302 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2303 return -EPERM;
2304
2305 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2306 if (err)
2307 return err;
2308 size = min_t(u32, size, sizeof(attr));
2309
2310 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2311 if (copy_from_user(&attr, uattr, size) != 0)
2312 return -EFAULT;
2313
2314 err = security_bpf(cmd, &attr, size);
2315 if (err < 0)
2316 return err;
2317
2318 switch (cmd) {
2319 case BPF_MAP_CREATE:
2320 err = map_create(&attr);
2321 break;
2322 case BPF_MAP_LOOKUP_ELEM:
2323 err = map_lookup_elem(&attr);
2324 break;
2325 case BPF_MAP_UPDATE_ELEM:
2326 err = map_update_elem(&attr);
2327 break;
2328 case BPF_MAP_DELETE_ELEM:
2329 err = map_delete_elem(&attr);
2330 break;
2331 case BPF_MAP_GET_NEXT_KEY:
2332 err = map_get_next_key(&attr);
2333 break;
2334 case BPF_PROG_LOAD:
2335 err = bpf_prog_load(&attr);
2336 break;
2337 case BPF_OBJ_PIN:
2338 err = bpf_obj_pin(&attr);
2339 break;
2340 case BPF_OBJ_GET:
2341 err = bpf_obj_get(&attr);
2342 break;
2343 case BPF_PROG_ATTACH:
2344 err = bpf_prog_attach(&attr);
2345 break;
2346 case BPF_PROG_DETACH:
2347 err = bpf_prog_detach(&attr);
2348 break;
2349 case BPF_PROG_QUERY:
2350 err = bpf_prog_query(&attr, uattr);
2351 break;
2352 case BPF_PROG_TEST_RUN:
2353 err = bpf_prog_test_run(&attr, uattr);
2354 break;
2355 case BPF_PROG_GET_NEXT_ID:
2356 err = bpf_obj_get_next_id(&attr, uattr,
2357 &prog_idr, &prog_idr_lock);
2358 break;
2359 case BPF_MAP_GET_NEXT_ID:
2360 err = bpf_obj_get_next_id(&attr, uattr,
2361 &map_idr, &map_idr_lock);
2362 break;
2363 case BPF_PROG_GET_FD_BY_ID:
2364 err = bpf_prog_get_fd_by_id(&attr);
2365 break;
2366 case BPF_MAP_GET_FD_BY_ID:
2367 err = bpf_map_get_fd_by_id(&attr);
2368 break;
2369 case BPF_OBJ_GET_INFO_BY_FD:
2370 err = bpf_obj_get_info_by_fd(&attr, uattr);
2371 break;
2372 case BPF_RAW_TRACEPOINT_OPEN:
2373 err = bpf_raw_tracepoint_open(&attr);
2374 break;
2375 case BPF_BTF_LOAD:
2376 err = bpf_btf_load(&attr);
2377 break;
2378 case BPF_BTF_GET_FD_BY_ID:
2379 err = bpf_btf_get_fd_by_id(&attr);
2380 break;
2381 case BPF_TASK_FD_QUERY:
2382 err = bpf_task_fd_query(&attr, uattr);
2383 break;
2384 default:
2385 err = -EINVAL;
2386 break;
2387 }
2388
2389 return err;
2390 }
Linux with added FPC-III support