xref: /openbmc/linux/kernel/bpf/syscall.c (revision 6d99a79c)
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/nospec.h>
34 
35 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
36 			   (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
37 			   (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
38 			   (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
39 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
40 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
41 
42 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
43 
44 DEFINE_PER_CPU(int, bpf_prog_active);
45 static DEFINE_IDR(prog_idr);
46 static DEFINE_SPINLOCK(prog_idr_lock);
47 static DEFINE_IDR(map_idr);
48 static DEFINE_SPINLOCK(map_idr_lock);
49 
50 int sysctl_unprivileged_bpf_disabled __read_mostly;
51 
52 static const struct bpf_map_ops * const bpf_map_types[] = {
53 #define BPF_PROG_TYPE(_id, _ops)
54 #define BPF_MAP_TYPE(_id, _ops) \
55 	[_id] = &_ops,
56 #include <linux/bpf_types.h>
57 #undef BPF_PROG_TYPE
58 #undef BPF_MAP_TYPE
59 };
60 
61 /*
62  * If we're handed a bigger struct than we know of, ensure all the unknown bits
63  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
64  * we don't know about yet.
65  *
66  * There is a ToCToU between this function call and the following
67  * copy_from_user() call. However, this is not a concern since this function is
68  * meant to be a future-proofing of bits.
69  */
70 int bpf_check_uarg_tail_zero(void __user *uaddr,
71 			     size_t expected_size,
72 			     size_t actual_size)
73 {
74 	unsigned char __user *addr;
75 	unsigned char __user *end;
76 	unsigned char val;
77 	int err;
78 
79 	if (unlikely(actual_size > PAGE_SIZE))	/* silly large */
80 		return -E2BIG;
81 
82 	if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
83 		return -EFAULT;
84 
85 	if (actual_size <= expected_size)
86 		return 0;
87 
88 	addr = uaddr + expected_size;
89 	end  = uaddr + actual_size;
90 
91 	for (; addr < end; addr++) {
92 		err = get_user(val, addr);
93 		if (err)
94 			return err;
95 		if (val)
96 			return -E2BIG;
97 	}
98 
99 	return 0;
100 }
101 
102 const struct bpf_map_ops bpf_map_offload_ops = {
103 	.map_alloc = bpf_map_offload_map_alloc,
104 	.map_free = bpf_map_offload_map_free,
105 	.map_check_btf = map_check_no_btf,
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 int map_check_no_btf(const struct bpf_map *map,
459 		     const struct btf_type *key_type,
460 		     const struct btf_type *value_type)
461 {
462 	return -ENOTSUPP;
463 }
464 
465 static int map_check_btf(const struct bpf_map *map, const struct btf *btf,
466 			 u32 btf_key_id, u32 btf_value_id)
467 {
468 	const struct btf_type *key_type, *value_type;
469 	u32 key_size, value_size;
470 	int ret = 0;
471 
472 	key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
473 	if (!key_type || key_size != map->key_size)
474 		return -EINVAL;
475 
476 	value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
477 	if (!value_type || value_size != map->value_size)
478 		return -EINVAL;
479 
480 	if (map->ops->map_check_btf)
481 		ret = map->ops->map_check_btf(map, key_type, value_type);
482 
483 	return ret;
484 }
485 
486 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
487 /* called via syscall */
488 static int map_create(union bpf_attr *attr)
489 {
490 	int numa_node = bpf_map_attr_numa_node(attr);
491 	struct bpf_map *map;
492 	int f_flags;
493 	int err;
494 
495 	err = CHECK_ATTR(BPF_MAP_CREATE);
496 	if (err)
497 		return -EINVAL;
498 
499 	f_flags = bpf_get_file_flag(attr->map_flags);
500 	if (f_flags < 0)
501 		return f_flags;
502 
503 	if (numa_node != NUMA_NO_NODE &&
504 	    ((unsigned int)numa_node >= nr_node_ids ||
505 	     !node_online(numa_node)))
506 		return -EINVAL;
507 
508 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
509 	map = find_and_alloc_map(attr);
510 	if (IS_ERR(map))
511 		return PTR_ERR(map);
512 
513 	err = bpf_obj_name_cpy(map->name, attr->map_name);
514 	if (err)
515 		goto free_map_nouncharge;
516 
517 	atomic_set(&map->refcnt, 1);
518 	atomic_set(&map->usercnt, 1);
519 
520 	if (attr->btf_key_type_id || attr->btf_value_type_id) {
521 		struct btf *btf;
522 
523 		if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
524 			err = -EINVAL;
525 			goto free_map_nouncharge;
526 		}
527 
528 		btf = btf_get_by_fd(attr->btf_fd);
529 		if (IS_ERR(btf)) {
530 			err = PTR_ERR(btf);
531 			goto free_map_nouncharge;
532 		}
533 
534 		err = map_check_btf(map, btf, attr->btf_key_type_id,
535 				    attr->btf_value_type_id);
536 		if (err) {
537 			btf_put(btf);
538 			goto free_map_nouncharge;
539 		}
540 
541 		map->btf = btf;
542 		map->btf_key_type_id = attr->btf_key_type_id;
543 		map->btf_value_type_id = attr->btf_value_type_id;
544 	}
545 
546 	err = security_bpf_map_alloc(map);
547 	if (err)
548 		goto free_map_nouncharge;
549 
550 	err = bpf_map_init_memlock(map);
551 	if (err)
552 		goto free_map_sec;
553 
554 	err = bpf_map_alloc_id(map);
555 	if (err)
556 		goto free_map;
557 
558 	err = bpf_map_new_fd(map, f_flags);
559 	if (err < 0) {
560 		/* failed to allocate fd.
561 		 * bpf_map_put() is needed because the above
562 		 * bpf_map_alloc_id() has published the map
563 		 * to the userspace and the userspace may
564 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
565 		 */
566 		bpf_map_put(map);
567 		return err;
568 	}
569 
570 	return err;
571 
572 free_map:
573 	bpf_map_release_memlock(map);
574 free_map_sec:
575 	security_bpf_map_free(map);
576 free_map_nouncharge:
577 	btf_put(map->btf);
578 	map->ops->map_free(map);
579 	return err;
580 }
581 
582 /* if error is returned, fd is released.
583  * On success caller should complete fd access with matching fdput()
584  */
585 struct bpf_map *__bpf_map_get(struct fd f)
586 {
587 	if (!f.file)
588 		return ERR_PTR(-EBADF);
589 	if (f.file->f_op != &bpf_map_fops) {
590 		fdput(f);
591 		return ERR_PTR(-EINVAL);
592 	}
593 
594 	return f.file->private_data;
595 }
596 
597 /* prog's and map's refcnt limit */
598 #define BPF_MAX_REFCNT 32768
599 
600 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
601 {
602 	if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
603 		atomic_dec(&map->refcnt);
604 		return ERR_PTR(-EBUSY);
605 	}
606 	if (uref)
607 		atomic_inc(&map->usercnt);
608 	return map;
609 }
610 EXPORT_SYMBOL_GPL(bpf_map_inc);
611 
612 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
613 {
614 	struct fd f = fdget(ufd);
615 	struct bpf_map *map;
616 
617 	map = __bpf_map_get(f);
618 	if (IS_ERR(map))
619 		return map;
620 
621 	map = bpf_map_inc(map, true);
622 	fdput(f);
623 
624 	return map;
625 }
626 
627 /* map_idr_lock should have been held */
628 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
629 					    bool uref)
630 {
631 	int refold;
632 
633 	refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
634 
635 	if (refold >= BPF_MAX_REFCNT) {
636 		__bpf_map_put(map, false);
637 		return ERR_PTR(-EBUSY);
638 	}
639 
640 	if (!refold)
641 		return ERR_PTR(-ENOENT);
642 
643 	if (uref)
644 		atomic_inc(&map->usercnt);
645 
646 	return map;
647 }
648 
649 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
650 {
651 	return -ENOTSUPP;
652 }
653 
654 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
655 {
656 	if (key_size)
657 		return memdup_user(ukey, key_size);
658 
659 	if (ukey)
660 		return ERR_PTR(-EINVAL);
661 
662 	return NULL;
663 }
664 
665 /* last field in 'union bpf_attr' used by this command */
666 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
667 
668 static int map_lookup_elem(union bpf_attr *attr)
669 {
670 	void __user *ukey = u64_to_user_ptr(attr->key);
671 	void __user *uvalue = u64_to_user_ptr(attr->value);
672 	int ufd = attr->map_fd;
673 	struct bpf_map *map;
674 	void *key, *value, *ptr;
675 	u32 value_size;
676 	struct fd f;
677 	int err;
678 
679 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
680 		return -EINVAL;
681 
682 	f = fdget(ufd);
683 	map = __bpf_map_get(f);
684 	if (IS_ERR(map))
685 		return PTR_ERR(map);
686 
687 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
688 		err = -EPERM;
689 		goto err_put;
690 	}
691 
692 	key = __bpf_copy_key(ukey, map->key_size);
693 	if (IS_ERR(key)) {
694 		err = PTR_ERR(key);
695 		goto err_put;
696 	}
697 
698 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
699 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
700 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
701 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
702 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
703 	else if (IS_FD_MAP(map))
704 		value_size = sizeof(u32);
705 	else
706 		value_size = map->value_size;
707 
708 	err = -ENOMEM;
709 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
710 	if (!value)
711 		goto free_key;
712 
713 	if (bpf_map_is_dev_bound(map)) {
714 		err = bpf_map_offload_lookup_elem(map, key, value);
715 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
716 		   map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
717 		err = bpf_percpu_hash_copy(map, key, value);
718 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
719 		err = bpf_percpu_array_copy(map, key, value);
720 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
721 		err = bpf_percpu_cgroup_storage_copy(map, key, value);
722 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
723 		err = bpf_stackmap_copy(map, key, value);
724 	} else if (IS_FD_ARRAY(map)) {
725 		err = bpf_fd_array_map_lookup_elem(map, key, value);
726 	} else if (IS_FD_HASH(map)) {
727 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
728 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
729 		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
730 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
731 		   map->map_type == BPF_MAP_TYPE_STACK) {
732 		err = map->ops->map_peek_elem(map, value);
733 	} else {
734 		rcu_read_lock();
735 		ptr = map->ops->map_lookup_elem(map, key);
736 		if (IS_ERR(ptr)) {
737 			err = PTR_ERR(ptr);
738 		} else if (!ptr) {
739 			err = -ENOENT;
740 		} else {
741 			err = 0;
742 			memcpy(value, ptr, value_size);
743 		}
744 		rcu_read_unlock();
745 	}
746 
747 	if (err)
748 		goto free_value;
749 
750 	err = -EFAULT;
751 	if (copy_to_user(uvalue, value, value_size) != 0)
752 		goto free_value;
753 
754 	err = 0;
755 
756 free_value:
757 	kfree(value);
758 free_key:
759 	kfree(key);
760 err_put:
761 	fdput(f);
762 	return err;
763 }
764 
765 static void maybe_wait_bpf_programs(struct bpf_map *map)
766 {
767 	/* Wait for any running BPF programs to complete so that
768 	 * userspace, when we return to it, knows that all programs
769 	 * that could be running use the new map value.
770 	 */
771 	if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
772 	    map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
773 		synchronize_rcu();
774 }
775 
776 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
777 
778 static int map_update_elem(union bpf_attr *attr)
779 {
780 	void __user *ukey = u64_to_user_ptr(attr->key);
781 	void __user *uvalue = u64_to_user_ptr(attr->value);
782 	int ufd = attr->map_fd;
783 	struct bpf_map *map;
784 	void *key, *value;
785 	u32 value_size;
786 	struct fd f;
787 	int err;
788 
789 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
790 		return -EINVAL;
791 
792 	f = fdget(ufd);
793 	map = __bpf_map_get(f);
794 	if (IS_ERR(map))
795 		return PTR_ERR(map);
796 
797 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
798 		err = -EPERM;
799 		goto err_put;
800 	}
801 
802 	key = __bpf_copy_key(ukey, map->key_size);
803 	if (IS_ERR(key)) {
804 		err = PTR_ERR(key);
805 		goto err_put;
806 	}
807 
808 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
809 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
810 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
811 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
812 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
813 	else
814 		value_size = map->value_size;
815 
816 	err = -ENOMEM;
817 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
818 	if (!value)
819 		goto free_key;
820 
821 	err = -EFAULT;
822 	if (copy_from_user(value, uvalue, value_size) != 0)
823 		goto free_value;
824 
825 	/* Need to create a kthread, thus must support schedule */
826 	if (bpf_map_is_dev_bound(map)) {
827 		err = bpf_map_offload_update_elem(map, key, value, attr->flags);
828 		goto out;
829 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
830 		   map->map_type == BPF_MAP_TYPE_SOCKHASH ||
831 		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
832 		err = map->ops->map_update_elem(map, key, value, attr->flags);
833 		goto out;
834 	}
835 
836 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
837 	 * inside bpf map update or delete otherwise deadlocks are possible
838 	 */
839 	preempt_disable();
840 	__this_cpu_inc(bpf_prog_active);
841 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
842 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
843 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
844 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
845 		err = bpf_percpu_array_update(map, key, value, attr->flags);
846 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
847 		err = bpf_percpu_cgroup_storage_update(map, key, value,
848 						       attr->flags);
849 	} else if (IS_FD_ARRAY(map)) {
850 		rcu_read_lock();
851 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
852 						   attr->flags);
853 		rcu_read_unlock();
854 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
855 		rcu_read_lock();
856 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
857 						  attr->flags);
858 		rcu_read_unlock();
859 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
860 		/* rcu_read_lock() is not needed */
861 		err = bpf_fd_reuseport_array_update_elem(map, key, value,
862 							 attr->flags);
863 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
864 		   map->map_type == BPF_MAP_TYPE_STACK) {
865 		err = map->ops->map_push_elem(map, value, attr->flags);
866 	} else {
867 		rcu_read_lock();
868 		err = map->ops->map_update_elem(map, key, value, attr->flags);
869 		rcu_read_unlock();
870 	}
871 	__this_cpu_dec(bpf_prog_active);
872 	preempt_enable();
873 	maybe_wait_bpf_programs(map);
874 out:
875 free_value:
876 	kfree(value);
877 free_key:
878 	kfree(key);
879 err_put:
880 	fdput(f);
881 	return err;
882 }
883 
884 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
885 
886 static int map_delete_elem(union bpf_attr *attr)
887 {
888 	void __user *ukey = u64_to_user_ptr(attr->key);
889 	int ufd = attr->map_fd;
890 	struct bpf_map *map;
891 	struct fd f;
892 	void *key;
893 	int err;
894 
895 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
896 		return -EINVAL;
897 
898 	f = fdget(ufd);
899 	map = __bpf_map_get(f);
900 	if (IS_ERR(map))
901 		return PTR_ERR(map);
902 
903 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
904 		err = -EPERM;
905 		goto err_put;
906 	}
907 
908 	key = __bpf_copy_key(ukey, map->key_size);
909 	if (IS_ERR(key)) {
910 		err = PTR_ERR(key);
911 		goto err_put;
912 	}
913 
914 	if (bpf_map_is_dev_bound(map)) {
915 		err = bpf_map_offload_delete_elem(map, key);
916 		goto out;
917 	}
918 
919 	preempt_disable();
920 	__this_cpu_inc(bpf_prog_active);
921 	rcu_read_lock();
922 	err = map->ops->map_delete_elem(map, key);
923 	rcu_read_unlock();
924 	__this_cpu_dec(bpf_prog_active);
925 	preempt_enable();
926 	maybe_wait_bpf_programs(map);
927 out:
928 	kfree(key);
929 err_put:
930 	fdput(f);
931 	return err;
932 }
933 
934 /* last field in 'union bpf_attr' used by this command */
935 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
936 
937 static int map_get_next_key(union bpf_attr *attr)
938 {
939 	void __user *ukey = u64_to_user_ptr(attr->key);
940 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
941 	int ufd = attr->map_fd;
942 	struct bpf_map *map;
943 	void *key, *next_key;
944 	struct fd f;
945 	int err;
946 
947 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
948 		return -EINVAL;
949 
950 	f = fdget(ufd);
951 	map = __bpf_map_get(f);
952 	if (IS_ERR(map))
953 		return PTR_ERR(map);
954 
955 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
956 		err = -EPERM;
957 		goto err_put;
958 	}
959 
960 	if (ukey) {
961 		key = __bpf_copy_key(ukey, map->key_size);
962 		if (IS_ERR(key)) {
963 			err = PTR_ERR(key);
964 			goto err_put;
965 		}
966 	} else {
967 		key = NULL;
968 	}
969 
970 	err = -ENOMEM;
971 	next_key = kmalloc(map->key_size, GFP_USER);
972 	if (!next_key)
973 		goto free_key;
974 
975 	if (bpf_map_is_dev_bound(map)) {
976 		err = bpf_map_offload_get_next_key(map, key, next_key);
977 		goto out;
978 	}
979 
980 	rcu_read_lock();
981 	err = map->ops->map_get_next_key(map, key, next_key);
982 	rcu_read_unlock();
983 out:
984 	if (err)
985 		goto free_next_key;
986 
987 	err = -EFAULT;
988 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
989 		goto free_next_key;
990 
991 	err = 0;
992 
993 free_next_key:
994 	kfree(next_key);
995 free_key:
996 	kfree(key);
997 err_put:
998 	fdput(f);
999 	return err;
1000 }
1001 
1002 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1003 
1004 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1005 {
1006 	void __user *ukey = u64_to_user_ptr(attr->key);
1007 	void __user *uvalue = u64_to_user_ptr(attr->value);
1008 	int ufd = attr->map_fd;
1009 	struct bpf_map *map;
1010 	void *key, *value;
1011 	u32 value_size;
1012 	struct fd f;
1013 	int err;
1014 
1015 	if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1016 		return -EINVAL;
1017 
1018 	f = fdget(ufd);
1019 	map = __bpf_map_get(f);
1020 	if (IS_ERR(map))
1021 		return PTR_ERR(map);
1022 
1023 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
1024 		err = -EPERM;
1025 		goto err_put;
1026 	}
1027 
1028 	key = __bpf_copy_key(ukey, map->key_size);
1029 	if (IS_ERR(key)) {
1030 		err = PTR_ERR(key);
1031 		goto err_put;
1032 	}
1033 
1034 	value_size = map->value_size;
1035 
1036 	err = -ENOMEM;
1037 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1038 	if (!value)
1039 		goto free_key;
1040 
1041 	if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1042 	    map->map_type == BPF_MAP_TYPE_STACK) {
1043 		err = map->ops->map_pop_elem(map, value);
1044 	} else {
1045 		err = -ENOTSUPP;
1046 	}
1047 
1048 	if (err)
1049 		goto free_value;
1050 
1051 	if (copy_to_user(uvalue, value, value_size) != 0)
1052 		goto free_value;
1053 
1054 	err = 0;
1055 
1056 free_value:
1057 	kfree(value);
1058 free_key:
1059 	kfree(key);
1060 err_put:
1061 	fdput(f);
1062 	return err;
1063 }
1064 
1065 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1066 #define BPF_PROG_TYPE(_id, _name) \
1067 	[_id] = & _name ## _prog_ops,
1068 #define BPF_MAP_TYPE(_id, _ops)
1069 #include <linux/bpf_types.h>
1070 #undef BPF_PROG_TYPE
1071 #undef BPF_MAP_TYPE
1072 };
1073 
1074 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1075 {
1076 	const struct bpf_prog_ops *ops;
1077 
1078 	if (type >= ARRAY_SIZE(bpf_prog_types))
1079 		return -EINVAL;
1080 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1081 	ops = bpf_prog_types[type];
1082 	if (!ops)
1083 		return -EINVAL;
1084 
1085 	if (!bpf_prog_is_dev_bound(prog->aux))
1086 		prog->aux->ops = ops;
1087 	else
1088 		prog->aux->ops = &bpf_offload_prog_ops;
1089 	prog->type = type;
1090 	return 0;
1091 }
1092 
1093 /* drop refcnt on maps used by eBPF program and free auxilary data */
1094 static void free_used_maps(struct bpf_prog_aux *aux)
1095 {
1096 	enum bpf_cgroup_storage_type stype;
1097 	int i;
1098 
1099 	for_each_cgroup_storage_type(stype) {
1100 		if (!aux->cgroup_storage[stype])
1101 			continue;
1102 		bpf_cgroup_storage_release(aux->prog,
1103 					   aux->cgroup_storage[stype]);
1104 	}
1105 
1106 	for (i = 0; i < aux->used_map_cnt; i++)
1107 		bpf_map_put(aux->used_maps[i]);
1108 
1109 	kfree(aux->used_maps);
1110 }
1111 
1112 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1113 {
1114 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1115 	unsigned long user_bufs;
1116 
1117 	if (user) {
1118 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1119 		if (user_bufs > memlock_limit) {
1120 			atomic_long_sub(pages, &user->locked_vm);
1121 			return -EPERM;
1122 		}
1123 	}
1124 
1125 	return 0;
1126 }
1127 
1128 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1129 {
1130 	if (user)
1131 		atomic_long_sub(pages, &user->locked_vm);
1132 }
1133 
1134 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1135 {
1136 	struct user_struct *user = get_current_user();
1137 	int ret;
1138 
1139 	ret = __bpf_prog_charge(user, prog->pages);
1140 	if (ret) {
1141 		free_uid(user);
1142 		return ret;
1143 	}
1144 
1145 	prog->aux->user = user;
1146 	return 0;
1147 }
1148 
1149 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1150 {
1151 	struct user_struct *user = prog->aux->user;
1152 
1153 	__bpf_prog_uncharge(user, prog->pages);
1154 	free_uid(user);
1155 }
1156 
1157 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1158 {
1159 	int id;
1160 
1161 	idr_preload(GFP_KERNEL);
1162 	spin_lock_bh(&prog_idr_lock);
1163 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1164 	if (id > 0)
1165 		prog->aux->id = id;
1166 	spin_unlock_bh(&prog_idr_lock);
1167 	idr_preload_end();
1168 
1169 	/* id is in [1, INT_MAX) */
1170 	if (WARN_ON_ONCE(!id))
1171 		return -ENOSPC;
1172 
1173 	return id > 0 ? 0 : id;
1174 }
1175 
1176 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1177 {
1178 	/* cBPF to eBPF migrations are currently not in the idr store.
1179 	 * Offloaded programs are removed from the store when their device
1180 	 * disappears - even if someone grabs an fd to them they are unusable,
1181 	 * simply waiting for refcnt to drop to be freed.
1182 	 */
1183 	if (!prog->aux->id)
1184 		return;
1185 
1186 	if (do_idr_lock)
1187 		spin_lock_bh(&prog_idr_lock);
1188 	else
1189 		__acquire(&prog_idr_lock);
1190 
1191 	idr_remove(&prog_idr, prog->aux->id);
1192 	prog->aux->id = 0;
1193 
1194 	if (do_idr_lock)
1195 		spin_unlock_bh(&prog_idr_lock);
1196 	else
1197 		__release(&prog_idr_lock);
1198 }
1199 
1200 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1201 {
1202 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1203 
1204 	free_used_maps(aux);
1205 	bpf_prog_uncharge_memlock(aux->prog);
1206 	security_bpf_prog_free(aux);
1207 	bpf_prog_free(aux->prog);
1208 }
1209 
1210 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1211 {
1212 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
1213 		/* bpf_prog_free_id() must be called first */
1214 		bpf_prog_free_id(prog, do_idr_lock);
1215 		bpf_prog_kallsyms_del_all(prog);
1216 
1217 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1218 	}
1219 }
1220 
1221 void bpf_prog_put(struct bpf_prog *prog)
1222 {
1223 	__bpf_prog_put(prog, true);
1224 }
1225 EXPORT_SYMBOL_GPL(bpf_prog_put);
1226 
1227 static int bpf_prog_release(struct inode *inode, struct file *filp)
1228 {
1229 	struct bpf_prog *prog = filp->private_data;
1230 
1231 	bpf_prog_put(prog);
1232 	return 0;
1233 }
1234 
1235 #ifdef CONFIG_PROC_FS
1236 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1237 {
1238 	const struct bpf_prog *prog = filp->private_data;
1239 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1240 
1241 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1242 	seq_printf(m,
1243 		   "prog_type:\t%u\n"
1244 		   "prog_jited:\t%u\n"
1245 		   "prog_tag:\t%s\n"
1246 		   "memlock:\t%llu\n"
1247 		   "prog_id:\t%u\n",
1248 		   prog->type,
1249 		   prog->jited,
1250 		   prog_tag,
1251 		   prog->pages * 1ULL << PAGE_SHIFT,
1252 		   prog->aux->id);
1253 }
1254 #endif
1255 
1256 const struct file_operations bpf_prog_fops = {
1257 #ifdef CONFIG_PROC_FS
1258 	.show_fdinfo	= bpf_prog_show_fdinfo,
1259 #endif
1260 	.release	= bpf_prog_release,
1261 	.read		= bpf_dummy_read,
1262 	.write		= bpf_dummy_write,
1263 };
1264 
1265 int bpf_prog_new_fd(struct bpf_prog *prog)
1266 {
1267 	int ret;
1268 
1269 	ret = security_bpf_prog(prog);
1270 	if (ret < 0)
1271 		return ret;
1272 
1273 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1274 				O_RDWR | O_CLOEXEC);
1275 }
1276 
1277 static struct bpf_prog *____bpf_prog_get(struct fd f)
1278 {
1279 	if (!f.file)
1280 		return ERR_PTR(-EBADF);
1281 	if (f.file->f_op != &bpf_prog_fops) {
1282 		fdput(f);
1283 		return ERR_PTR(-EINVAL);
1284 	}
1285 
1286 	return f.file->private_data;
1287 }
1288 
1289 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1290 {
1291 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1292 		atomic_sub(i, &prog->aux->refcnt);
1293 		return ERR_PTR(-EBUSY);
1294 	}
1295 	return prog;
1296 }
1297 EXPORT_SYMBOL_GPL(bpf_prog_add);
1298 
1299 void bpf_prog_sub(struct bpf_prog *prog, int i)
1300 {
1301 	/* Only to be used for undoing previous bpf_prog_add() in some
1302 	 * error path. We still know that another entity in our call
1303 	 * path holds a reference to the program, thus atomic_sub() can
1304 	 * be safely used in such cases!
1305 	 */
1306 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1307 }
1308 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1309 
1310 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1311 {
1312 	return bpf_prog_add(prog, 1);
1313 }
1314 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1315 
1316 /* prog_idr_lock should have been held */
1317 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1318 {
1319 	int refold;
1320 
1321 	refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1322 
1323 	if (refold >= BPF_MAX_REFCNT) {
1324 		__bpf_prog_put(prog, false);
1325 		return ERR_PTR(-EBUSY);
1326 	}
1327 
1328 	if (!refold)
1329 		return ERR_PTR(-ENOENT);
1330 
1331 	return prog;
1332 }
1333 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1334 
1335 bool bpf_prog_get_ok(struct bpf_prog *prog,
1336 			    enum bpf_prog_type *attach_type, bool attach_drv)
1337 {
1338 	/* not an attachment, just a refcount inc, always allow */
1339 	if (!attach_type)
1340 		return true;
1341 
1342 	if (prog->type != *attach_type)
1343 		return false;
1344 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1345 		return false;
1346 
1347 	return true;
1348 }
1349 
1350 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1351 				       bool attach_drv)
1352 {
1353 	struct fd f = fdget(ufd);
1354 	struct bpf_prog *prog;
1355 
1356 	prog = ____bpf_prog_get(f);
1357 	if (IS_ERR(prog))
1358 		return prog;
1359 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1360 		prog = ERR_PTR(-EINVAL);
1361 		goto out;
1362 	}
1363 
1364 	prog = bpf_prog_inc(prog);
1365 out:
1366 	fdput(f);
1367 	return prog;
1368 }
1369 
1370 struct bpf_prog *bpf_prog_get(u32 ufd)
1371 {
1372 	return __bpf_prog_get(ufd, NULL, false);
1373 }
1374 
1375 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1376 				       bool attach_drv)
1377 {
1378 	return __bpf_prog_get(ufd, &type, attach_drv);
1379 }
1380 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1381 
1382 /* Initially all BPF programs could be loaded w/o specifying
1383  * expected_attach_type. Later for some of them specifying expected_attach_type
1384  * at load time became required so that program could be validated properly.
1385  * Programs of types that are allowed to be loaded both w/ and w/o (for
1386  * backward compatibility) expected_attach_type, should have the default attach
1387  * type assigned to expected_attach_type for the latter case, so that it can be
1388  * validated later at attach time.
1389  *
1390  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1391  * prog type requires it but has some attach types that have to be backward
1392  * compatible.
1393  */
1394 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1395 {
1396 	switch (attr->prog_type) {
1397 	case BPF_PROG_TYPE_CGROUP_SOCK:
1398 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1399 		 * exist so checking for non-zero is the way to go here.
1400 		 */
1401 		if (!attr->expected_attach_type)
1402 			attr->expected_attach_type =
1403 				BPF_CGROUP_INET_SOCK_CREATE;
1404 		break;
1405 	}
1406 }
1407 
1408 static int
1409 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1410 				enum bpf_attach_type expected_attach_type)
1411 {
1412 	switch (prog_type) {
1413 	case BPF_PROG_TYPE_CGROUP_SOCK:
1414 		switch (expected_attach_type) {
1415 		case BPF_CGROUP_INET_SOCK_CREATE:
1416 		case BPF_CGROUP_INET4_POST_BIND:
1417 		case BPF_CGROUP_INET6_POST_BIND:
1418 			return 0;
1419 		default:
1420 			return -EINVAL;
1421 		}
1422 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1423 		switch (expected_attach_type) {
1424 		case BPF_CGROUP_INET4_BIND:
1425 		case BPF_CGROUP_INET6_BIND:
1426 		case BPF_CGROUP_INET4_CONNECT:
1427 		case BPF_CGROUP_INET6_CONNECT:
1428 		case BPF_CGROUP_UDP4_SENDMSG:
1429 		case BPF_CGROUP_UDP6_SENDMSG:
1430 			return 0;
1431 		default:
1432 			return -EINVAL;
1433 		}
1434 	default:
1435 		return 0;
1436 	}
1437 }
1438 
1439 /* last field in 'union bpf_attr' used by this command */
1440 #define	BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1441 
1442 static int bpf_prog_load(union bpf_attr *attr)
1443 {
1444 	enum bpf_prog_type type = attr->prog_type;
1445 	struct bpf_prog *prog;
1446 	int err;
1447 	char license[128];
1448 	bool is_gpl;
1449 
1450 	if (CHECK_ATTR(BPF_PROG_LOAD))
1451 		return -EINVAL;
1452 
1453 	if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1454 		return -EINVAL;
1455 
1456 	/* copy eBPF program license from user space */
1457 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1458 			      sizeof(license) - 1) < 0)
1459 		return -EFAULT;
1460 	license[sizeof(license) - 1] = 0;
1461 
1462 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1463 	is_gpl = license_is_gpl_compatible(license);
1464 
1465 	if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1466 		return -E2BIG;
1467 
1468 	if (type == BPF_PROG_TYPE_KPROBE &&
1469 	    attr->kern_version != LINUX_VERSION_CODE)
1470 		return -EINVAL;
1471 
1472 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1473 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
1474 	    !capable(CAP_SYS_ADMIN))
1475 		return -EPERM;
1476 
1477 	bpf_prog_load_fixup_attach_type(attr);
1478 	if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1479 		return -EINVAL;
1480 
1481 	/* plain bpf_prog allocation */
1482 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1483 	if (!prog)
1484 		return -ENOMEM;
1485 
1486 	prog->expected_attach_type = attr->expected_attach_type;
1487 
1488 	prog->aux->offload_requested = !!attr->prog_ifindex;
1489 
1490 	err = security_bpf_prog_alloc(prog->aux);
1491 	if (err)
1492 		goto free_prog_nouncharge;
1493 
1494 	err = bpf_prog_charge_memlock(prog);
1495 	if (err)
1496 		goto free_prog_sec;
1497 
1498 	prog->len = attr->insn_cnt;
1499 
1500 	err = -EFAULT;
1501 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1502 			   bpf_prog_insn_size(prog)) != 0)
1503 		goto free_prog;
1504 
1505 	prog->orig_prog = NULL;
1506 	prog->jited = 0;
1507 
1508 	atomic_set(&prog->aux->refcnt, 1);
1509 	prog->gpl_compatible = is_gpl ? 1 : 0;
1510 
1511 	if (bpf_prog_is_dev_bound(prog->aux)) {
1512 		err = bpf_prog_offload_init(prog, attr);
1513 		if (err)
1514 			goto free_prog;
1515 	}
1516 
1517 	/* find program type: socket_filter vs tracing_filter */
1518 	err = find_prog_type(type, prog);
1519 	if (err < 0)
1520 		goto free_prog;
1521 
1522 	prog->aux->load_time = ktime_get_boot_ns();
1523 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1524 	if (err)
1525 		goto free_prog;
1526 
1527 	/* run eBPF verifier */
1528 	err = bpf_check(&prog, attr);
1529 	if (err < 0)
1530 		goto free_used_maps;
1531 
1532 	prog = bpf_prog_select_runtime(prog, &err);
1533 	if (err < 0)
1534 		goto free_used_maps;
1535 
1536 	err = bpf_prog_alloc_id(prog);
1537 	if (err)
1538 		goto free_used_maps;
1539 
1540 	err = bpf_prog_new_fd(prog);
1541 	if (err < 0) {
1542 		/* failed to allocate fd.
1543 		 * bpf_prog_put() is needed because the above
1544 		 * bpf_prog_alloc_id() has published the prog
1545 		 * to the userspace and the userspace may
1546 		 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1547 		 */
1548 		bpf_prog_put(prog);
1549 		return err;
1550 	}
1551 
1552 	bpf_prog_kallsyms_add(prog);
1553 	return err;
1554 
1555 free_used_maps:
1556 	bpf_prog_kallsyms_del_subprogs(prog);
1557 	free_used_maps(prog->aux);
1558 free_prog:
1559 	bpf_prog_uncharge_memlock(prog);
1560 free_prog_sec:
1561 	security_bpf_prog_free(prog->aux);
1562 free_prog_nouncharge:
1563 	bpf_prog_free(prog);
1564 	return err;
1565 }
1566 
1567 #define BPF_OBJ_LAST_FIELD file_flags
1568 
1569 static int bpf_obj_pin(const union bpf_attr *attr)
1570 {
1571 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1572 		return -EINVAL;
1573 
1574 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1575 }
1576 
1577 static int bpf_obj_get(const union bpf_attr *attr)
1578 {
1579 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1580 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1581 		return -EINVAL;
1582 
1583 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1584 				attr->file_flags);
1585 }
1586 
1587 struct bpf_raw_tracepoint {
1588 	struct bpf_raw_event_map *btp;
1589 	struct bpf_prog *prog;
1590 };
1591 
1592 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1593 {
1594 	struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1595 
1596 	if (raw_tp->prog) {
1597 		bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1598 		bpf_prog_put(raw_tp->prog);
1599 	}
1600 	kfree(raw_tp);
1601 	return 0;
1602 }
1603 
1604 static const struct file_operations bpf_raw_tp_fops = {
1605 	.release	= bpf_raw_tracepoint_release,
1606 	.read		= bpf_dummy_read,
1607 	.write		= bpf_dummy_write,
1608 };
1609 
1610 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1611 
1612 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1613 {
1614 	struct bpf_raw_tracepoint *raw_tp;
1615 	struct bpf_raw_event_map *btp;
1616 	struct bpf_prog *prog;
1617 	char tp_name[128];
1618 	int tp_fd, err;
1619 
1620 	if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1621 			      sizeof(tp_name) - 1) < 0)
1622 		return -EFAULT;
1623 	tp_name[sizeof(tp_name) - 1] = 0;
1624 
1625 	btp = bpf_find_raw_tracepoint(tp_name);
1626 	if (!btp)
1627 		return -ENOENT;
1628 
1629 	raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1630 	if (!raw_tp)
1631 		return -ENOMEM;
1632 	raw_tp->btp = btp;
1633 
1634 	prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1635 				 BPF_PROG_TYPE_RAW_TRACEPOINT);
1636 	if (IS_ERR(prog)) {
1637 		err = PTR_ERR(prog);
1638 		goto out_free_tp;
1639 	}
1640 
1641 	err = bpf_probe_register(raw_tp->btp, prog);
1642 	if (err)
1643 		goto out_put_prog;
1644 
1645 	raw_tp->prog = prog;
1646 	tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1647 				 O_CLOEXEC);
1648 	if (tp_fd < 0) {
1649 		bpf_probe_unregister(raw_tp->btp, prog);
1650 		err = tp_fd;
1651 		goto out_put_prog;
1652 	}
1653 	return tp_fd;
1654 
1655 out_put_prog:
1656 	bpf_prog_put(prog);
1657 out_free_tp:
1658 	kfree(raw_tp);
1659 	return err;
1660 }
1661 
1662 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1663 					     enum bpf_attach_type attach_type)
1664 {
1665 	switch (prog->type) {
1666 	case BPF_PROG_TYPE_CGROUP_SOCK:
1667 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1668 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1669 	default:
1670 		return 0;
1671 	}
1672 }
1673 
1674 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1675 
1676 #define BPF_F_ATTACH_MASK \
1677 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1678 
1679 static int bpf_prog_attach(const union bpf_attr *attr)
1680 {
1681 	enum bpf_prog_type ptype;
1682 	struct bpf_prog *prog;
1683 	int ret;
1684 
1685 	if (!capable(CAP_NET_ADMIN))
1686 		return -EPERM;
1687 
1688 	if (CHECK_ATTR(BPF_PROG_ATTACH))
1689 		return -EINVAL;
1690 
1691 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1692 		return -EINVAL;
1693 
1694 	switch (attr->attach_type) {
1695 	case BPF_CGROUP_INET_INGRESS:
1696 	case BPF_CGROUP_INET_EGRESS:
1697 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1698 		break;
1699 	case BPF_CGROUP_INET_SOCK_CREATE:
1700 	case BPF_CGROUP_INET4_POST_BIND:
1701 	case BPF_CGROUP_INET6_POST_BIND:
1702 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1703 		break;
1704 	case BPF_CGROUP_INET4_BIND:
1705 	case BPF_CGROUP_INET6_BIND:
1706 	case BPF_CGROUP_INET4_CONNECT:
1707 	case BPF_CGROUP_INET6_CONNECT:
1708 	case BPF_CGROUP_UDP4_SENDMSG:
1709 	case BPF_CGROUP_UDP6_SENDMSG:
1710 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1711 		break;
1712 	case BPF_CGROUP_SOCK_OPS:
1713 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1714 		break;
1715 	case BPF_CGROUP_DEVICE:
1716 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1717 		break;
1718 	case BPF_SK_MSG_VERDICT:
1719 		ptype = BPF_PROG_TYPE_SK_MSG;
1720 		break;
1721 	case BPF_SK_SKB_STREAM_PARSER:
1722 	case BPF_SK_SKB_STREAM_VERDICT:
1723 		ptype = BPF_PROG_TYPE_SK_SKB;
1724 		break;
1725 	case BPF_LIRC_MODE2:
1726 		ptype = BPF_PROG_TYPE_LIRC_MODE2;
1727 		break;
1728 	case BPF_FLOW_DISSECTOR:
1729 		ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1730 		break;
1731 	default:
1732 		return -EINVAL;
1733 	}
1734 
1735 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1736 	if (IS_ERR(prog))
1737 		return PTR_ERR(prog);
1738 
1739 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1740 		bpf_prog_put(prog);
1741 		return -EINVAL;
1742 	}
1743 
1744 	switch (ptype) {
1745 	case BPF_PROG_TYPE_SK_SKB:
1746 	case BPF_PROG_TYPE_SK_MSG:
1747 		ret = sock_map_get_from_fd(attr, prog);
1748 		break;
1749 	case BPF_PROG_TYPE_LIRC_MODE2:
1750 		ret = lirc_prog_attach(attr, prog);
1751 		break;
1752 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
1753 		ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1754 		break;
1755 	default:
1756 		ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1757 	}
1758 
1759 	if (ret)
1760 		bpf_prog_put(prog);
1761 	return ret;
1762 }
1763 
1764 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1765 
1766 static int bpf_prog_detach(const union bpf_attr *attr)
1767 {
1768 	enum bpf_prog_type ptype;
1769 
1770 	if (!capable(CAP_NET_ADMIN))
1771 		return -EPERM;
1772 
1773 	if (CHECK_ATTR(BPF_PROG_DETACH))
1774 		return -EINVAL;
1775 
1776 	switch (attr->attach_type) {
1777 	case BPF_CGROUP_INET_INGRESS:
1778 	case BPF_CGROUP_INET_EGRESS:
1779 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1780 		break;
1781 	case BPF_CGROUP_INET_SOCK_CREATE:
1782 	case BPF_CGROUP_INET4_POST_BIND:
1783 	case BPF_CGROUP_INET6_POST_BIND:
1784 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1785 		break;
1786 	case BPF_CGROUP_INET4_BIND:
1787 	case BPF_CGROUP_INET6_BIND:
1788 	case BPF_CGROUP_INET4_CONNECT:
1789 	case BPF_CGROUP_INET6_CONNECT:
1790 	case BPF_CGROUP_UDP4_SENDMSG:
1791 	case BPF_CGROUP_UDP6_SENDMSG:
1792 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1793 		break;
1794 	case BPF_CGROUP_SOCK_OPS:
1795 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1796 		break;
1797 	case BPF_CGROUP_DEVICE:
1798 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1799 		break;
1800 	case BPF_SK_MSG_VERDICT:
1801 		return sock_map_get_from_fd(attr, NULL);
1802 	case BPF_SK_SKB_STREAM_PARSER:
1803 	case BPF_SK_SKB_STREAM_VERDICT:
1804 		return sock_map_get_from_fd(attr, NULL);
1805 	case BPF_LIRC_MODE2:
1806 		return lirc_prog_detach(attr);
1807 	case BPF_FLOW_DISSECTOR:
1808 		return skb_flow_dissector_bpf_prog_detach(attr);
1809 	default:
1810 		return -EINVAL;
1811 	}
1812 
1813 	return cgroup_bpf_prog_detach(attr, ptype);
1814 }
1815 
1816 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1817 
1818 static int bpf_prog_query(const union bpf_attr *attr,
1819 			  union bpf_attr __user *uattr)
1820 {
1821 	if (!capable(CAP_NET_ADMIN))
1822 		return -EPERM;
1823 	if (CHECK_ATTR(BPF_PROG_QUERY))
1824 		return -EINVAL;
1825 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1826 		return -EINVAL;
1827 
1828 	switch (attr->query.attach_type) {
1829 	case BPF_CGROUP_INET_INGRESS:
1830 	case BPF_CGROUP_INET_EGRESS:
1831 	case BPF_CGROUP_INET_SOCK_CREATE:
1832 	case BPF_CGROUP_INET4_BIND:
1833 	case BPF_CGROUP_INET6_BIND:
1834 	case BPF_CGROUP_INET4_POST_BIND:
1835 	case BPF_CGROUP_INET6_POST_BIND:
1836 	case BPF_CGROUP_INET4_CONNECT:
1837 	case BPF_CGROUP_INET6_CONNECT:
1838 	case BPF_CGROUP_UDP4_SENDMSG:
1839 	case BPF_CGROUP_UDP6_SENDMSG:
1840 	case BPF_CGROUP_SOCK_OPS:
1841 	case BPF_CGROUP_DEVICE:
1842 		break;
1843 	case BPF_LIRC_MODE2:
1844 		return lirc_prog_query(attr, uattr);
1845 	default:
1846 		return -EINVAL;
1847 	}
1848 
1849 	return cgroup_bpf_prog_query(attr, uattr);
1850 }
1851 
1852 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1853 
1854 static int bpf_prog_test_run(const union bpf_attr *attr,
1855 			     union bpf_attr __user *uattr)
1856 {
1857 	struct bpf_prog *prog;
1858 	int ret = -ENOTSUPP;
1859 
1860 	if (!capable(CAP_SYS_ADMIN))
1861 		return -EPERM;
1862 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1863 		return -EINVAL;
1864 
1865 	prog = bpf_prog_get(attr->test.prog_fd);
1866 	if (IS_ERR(prog))
1867 		return PTR_ERR(prog);
1868 
1869 	if (prog->aux->ops->test_run)
1870 		ret = prog->aux->ops->test_run(prog, attr, uattr);
1871 
1872 	bpf_prog_put(prog);
1873 	return ret;
1874 }
1875 
1876 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1877 
1878 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1879 			       union bpf_attr __user *uattr,
1880 			       struct idr *idr,
1881 			       spinlock_t *lock)
1882 {
1883 	u32 next_id = attr->start_id;
1884 	int err = 0;
1885 
1886 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1887 		return -EINVAL;
1888 
1889 	if (!capable(CAP_SYS_ADMIN))
1890 		return -EPERM;
1891 
1892 	next_id++;
1893 	spin_lock_bh(lock);
1894 	if (!idr_get_next(idr, &next_id))
1895 		err = -ENOENT;
1896 	spin_unlock_bh(lock);
1897 
1898 	if (!err)
1899 		err = put_user(next_id, &uattr->next_id);
1900 
1901 	return err;
1902 }
1903 
1904 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1905 
1906 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1907 {
1908 	struct bpf_prog *prog;
1909 	u32 id = attr->prog_id;
1910 	int fd;
1911 
1912 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1913 		return -EINVAL;
1914 
1915 	if (!capable(CAP_SYS_ADMIN))
1916 		return -EPERM;
1917 
1918 	spin_lock_bh(&prog_idr_lock);
1919 	prog = idr_find(&prog_idr, id);
1920 	if (prog)
1921 		prog = bpf_prog_inc_not_zero(prog);
1922 	else
1923 		prog = ERR_PTR(-ENOENT);
1924 	spin_unlock_bh(&prog_idr_lock);
1925 
1926 	if (IS_ERR(prog))
1927 		return PTR_ERR(prog);
1928 
1929 	fd = bpf_prog_new_fd(prog);
1930 	if (fd < 0)
1931 		bpf_prog_put(prog);
1932 
1933 	return fd;
1934 }
1935 
1936 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1937 
1938 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1939 {
1940 	struct bpf_map *map;
1941 	u32 id = attr->map_id;
1942 	int f_flags;
1943 	int fd;
1944 
1945 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1946 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1947 		return -EINVAL;
1948 
1949 	if (!capable(CAP_SYS_ADMIN))
1950 		return -EPERM;
1951 
1952 	f_flags = bpf_get_file_flag(attr->open_flags);
1953 	if (f_flags < 0)
1954 		return f_flags;
1955 
1956 	spin_lock_bh(&map_idr_lock);
1957 	map = idr_find(&map_idr, id);
1958 	if (map)
1959 		map = bpf_map_inc_not_zero(map, true);
1960 	else
1961 		map = ERR_PTR(-ENOENT);
1962 	spin_unlock_bh(&map_idr_lock);
1963 
1964 	if (IS_ERR(map))
1965 		return PTR_ERR(map);
1966 
1967 	fd = bpf_map_new_fd(map, f_flags);
1968 	if (fd < 0)
1969 		bpf_map_put(map);
1970 
1971 	return fd;
1972 }
1973 
1974 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1975 					      unsigned long addr)
1976 {
1977 	int i;
1978 
1979 	for (i = 0; i < prog->aux->used_map_cnt; i++)
1980 		if (prog->aux->used_maps[i] == (void *)addr)
1981 			return prog->aux->used_maps[i];
1982 	return NULL;
1983 }
1984 
1985 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1986 {
1987 	const struct bpf_map *map;
1988 	struct bpf_insn *insns;
1989 	u64 imm;
1990 	int i;
1991 
1992 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1993 			GFP_USER);
1994 	if (!insns)
1995 		return insns;
1996 
1997 	for (i = 0; i < prog->len; i++) {
1998 		if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1999 			insns[i].code = BPF_JMP | BPF_CALL;
2000 			insns[i].imm = BPF_FUNC_tail_call;
2001 			/* fall-through */
2002 		}
2003 		if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2004 		    insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2005 			if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2006 				insns[i].code = BPF_JMP | BPF_CALL;
2007 			if (!bpf_dump_raw_ok())
2008 				insns[i].imm = 0;
2009 			continue;
2010 		}
2011 
2012 		if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2013 			continue;
2014 
2015 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2016 		map = bpf_map_from_imm(prog, imm);
2017 		if (map) {
2018 			insns[i].src_reg = BPF_PSEUDO_MAP_FD;
2019 			insns[i].imm = map->id;
2020 			insns[i + 1].imm = 0;
2021 			continue;
2022 		}
2023 
2024 		if (!bpf_dump_raw_ok() &&
2025 		    imm == (unsigned long)prog->aux) {
2026 			insns[i].imm = 0;
2027 			insns[i + 1].imm = 0;
2028 			continue;
2029 		}
2030 	}
2031 
2032 	return insns;
2033 }
2034 
2035 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2036 				   const union bpf_attr *attr,
2037 				   union bpf_attr __user *uattr)
2038 {
2039 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2040 	struct bpf_prog_info info = {};
2041 	u32 info_len = attr->info.info_len;
2042 	char __user *uinsns;
2043 	u32 ulen;
2044 	int err;
2045 
2046 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2047 	if (err)
2048 		return err;
2049 	info_len = min_t(u32, sizeof(info), info_len);
2050 
2051 	if (copy_from_user(&info, uinfo, info_len))
2052 		return -EFAULT;
2053 
2054 	info.type = prog->type;
2055 	info.id = prog->aux->id;
2056 	info.load_time = prog->aux->load_time;
2057 	info.created_by_uid = from_kuid_munged(current_user_ns(),
2058 					       prog->aux->user->uid);
2059 	info.gpl_compatible = prog->gpl_compatible;
2060 
2061 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
2062 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2063 
2064 	ulen = info.nr_map_ids;
2065 	info.nr_map_ids = prog->aux->used_map_cnt;
2066 	ulen = min_t(u32, info.nr_map_ids, ulen);
2067 	if (ulen) {
2068 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2069 		u32 i;
2070 
2071 		for (i = 0; i < ulen; i++)
2072 			if (put_user(prog->aux->used_maps[i]->id,
2073 				     &user_map_ids[i]))
2074 				return -EFAULT;
2075 	}
2076 
2077 	if (!capable(CAP_SYS_ADMIN)) {
2078 		info.jited_prog_len = 0;
2079 		info.xlated_prog_len = 0;
2080 		info.nr_jited_ksyms = 0;
2081 		info.nr_jited_func_lens = 0;
2082 		goto done;
2083 	}
2084 
2085 	ulen = info.xlated_prog_len;
2086 	info.xlated_prog_len = bpf_prog_insn_size(prog);
2087 	if (info.xlated_prog_len && ulen) {
2088 		struct bpf_insn *insns_sanitized;
2089 		bool fault;
2090 
2091 		if (prog->blinded && !bpf_dump_raw_ok()) {
2092 			info.xlated_prog_insns = 0;
2093 			goto done;
2094 		}
2095 		insns_sanitized = bpf_insn_prepare_dump(prog);
2096 		if (!insns_sanitized)
2097 			return -ENOMEM;
2098 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2099 		ulen = min_t(u32, info.xlated_prog_len, ulen);
2100 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
2101 		kfree(insns_sanitized);
2102 		if (fault)
2103 			return -EFAULT;
2104 	}
2105 
2106 	if (bpf_prog_is_dev_bound(prog->aux)) {
2107 		err = bpf_prog_offload_info_fill(&info, prog);
2108 		if (err)
2109 			return err;
2110 		goto done;
2111 	}
2112 
2113 	/* NOTE: the following code is supposed to be skipped for offload.
2114 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
2115 	 * for offload.
2116 	 */
2117 	ulen = info.jited_prog_len;
2118 	if (prog->aux->func_cnt) {
2119 		u32 i;
2120 
2121 		info.jited_prog_len = 0;
2122 		for (i = 0; i < prog->aux->func_cnt; i++)
2123 			info.jited_prog_len += prog->aux->func[i]->jited_len;
2124 	} else {
2125 		info.jited_prog_len = prog->jited_len;
2126 	}
2127 
2128 	if (info.jited_prog_len && ulen) {
2129 		if (bpf_dump_raw_ok()) {
2130 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
2131 			ulen = min_t(u32, info.jited_prog_len, ulen);
2132 
2133 			/* for multi-function programs, copy the JITed
2134 			 * instructions for all the functions
2135 			 */
2136 			if (prog->aux->func_cnt) {
2137 				u32 len, free, i;
2138 				u8 *img;
2139 
2140 				free = ulen;
2141 				for (i = 0; i < prog->aux->func_cnt; i++) {
2142 					len = prog->aux->func[i]->jited_len;
2143 					len = min_t(u32, len, free);
2144 					img = (u8 *) prog->aux->func[i]->bpf_func;
2145 					if (copy_to_user(uinsns, img, len))
2146 						return -EFAULT;
2147 					uinsns += len;
2148 					free -= len;
2149 					if (!free)
2150 						break;
2151 				}
2152 			} else {
2153 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
2154 					return -EFAULT;
2155 			}
2156 		} else {
2157 			info.jited_prog_insns = 0;
2158 		}
2159 	}
2160 
2161 	ulen = info.nr_jited_ksyms;
2162 	info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2163 	if (info.nr_jited_ksyms && ulen) {
2164 		if (bpf_dump_raw_ok()) {
2165 			unsigned long ksym_addr;
2166 			u64 __user *user_ksyms;
2167 			u32 i;
2168 
2169 			/* copy the address of the kernel symbol
2170 			 * corresponding to each function
2171 			 */
2172 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2173 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2174 			if (prog->aux->func_cnt) {
2175 				for (i = 0; i < ulen; i++) {
2176 					ksym_addr = (unsigned long)
2177 						prog->aux->func[i]->bpf_func;
2178 					if (put_user((u64) ksym_addr,
2179 						     &user_ksyms[i]))
2180 						return -EFAULT;
2181 				}
2182 			} else {
2183 				ksym_addr = (unsigned long) prog->bpf_func;
2184 				if (put_user((u64) ksym_addr, &user_ksyms[0]))
2185 					return -EFAULT;
2186 			}
2187 		} else {
2188 			info.jited_ksyms = 0;
2189 		}
2190 	}
2191 
2192 	ulen = info.nr_jited_func_lens;
2193 	info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2194 	if (info.nr_jited_func_lens && ulen) {
2195 		if (bpf_dump_raw_ok()) {
2196 			u32 __user *user_lens;
2197 			u32 func_len, i;
2198 
2199 			/* copy the JITed image lengths for each function */
2200 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2201 			user_lens = u64_to_user_ptr(info.jited_func_lens);
2202 			if (prog->aux->func_cnt) {
2203 				for (i = 0; i < ulen; i++) {
2204 					func_len =
2205 						prog->aux->func[i]->jited_len;
2206 					if (put_user(func_len, &user_lens[i]))
2207 						return -EFAULT;
2208 				}
2209 			} else {
2210 				func_len = prog->jited_len;
2211 				if (put_user(func_len, &user_lens[0]))
2212 					return -EFAULT;
2213 			}
2214 		} else {
2215 			info.jited_func_lens = 0;
2216 		}
2217 	}
2218 
2219 done:
2220 	if (copy_to_user(uinfo, &info, info_len) ||
2221 	    put_user(info_len, &uattr->info.info_len))
2222 		return -EFAULT;
2223 
2224 	return 0;
2225 }
2226 
2227 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2228 				  const union bpf_attr *attr,
2229 				  union bpf_attr __user *uattr)
2230 {
2231 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2232 	struct bpf_map_info info = {};
2233 	u32 info_len = attr->info.info_len;
2234 	int err;
2235 
2236 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2237 	if (err)
2238 		return err;
2239 	info_len = min_t(u32, sizeof(info), info_len);
2240 
2241 	info.type = map->map_type;
2242 	info.id = map->id;
2243 	info.key_size = map->key_size;
2244 	info.value_size = map->value_size;
2245 	info.max_entries = map->max_entries;
2246 	info.map_flags = map->map_flags;
2247 	memcpy(info.name, map->name, sizeof(map->name));
2248 
2249 	if (map->btf) {
2250 		info.btf_id = btf_id(map->btf);
2251 		info.btf_key_type_id = map->btf_key_type_id;
2252 		info.btf_value_type_id = map->btf_value_type_id;
2253 	}
2254 
2255 	if (bpf_map_is_dev_bound(map)) {
2256 		err = bpf_map_offload_info_fill(&info, map);
2257 		if (err)
2258 			return err;
2259 	}
2260 
2261 	if (copy_to_user(uinfo, &info, info_len) ||
2262 	    put_user(info_len, &uattr->info.info_len))
2263 		return -EFAULT;
2264 
2265 	return 0;
2266 }
2267 
2268 static int bpf_btf_get_info_by_fd(struct btf *btf,
2269 				  const union bpf_attr *attr,
2270 				  union bpf_attr __user *uattr)
2271 {
2272 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2273 	u32 info_len = attr->info.info_len;
2274 	int err;
2275 
2276 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2277 	if (err)
2278 		return err;
2279 
2280 	return btf_get_info_by_fd(btf, attr, uattr);
2281 }
2282 
2283 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2284 
2285 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2286 				  union bpf_attr __user *uattr)
2287 {
2288 	int ufd = attr->info.bpf_fd;
2289 	struct fd f;
2290 	int err;
2291 
2292 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2293 		return -EINVAL;
2294 
2295 	f = fdget(ufd);
2296 	if (!f.file)
2297 		return -EBADFD;
2298 
2299 	if (f.file->f_op == &bpf_prog_fops)
2300 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2301 					      uattr);
2302 	else if (f.file->f_op == &bpf_map_fops)
2303 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2304 					     uattr);
2305 	else if (f.file->f_op == &btf_fops)
2306 		err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2307 	else
2308 		err = -EINVAL;
2309 
2310 	fdput(f);
2311 	return err;
2312 }
2313 
2314 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2315 
2316 static int bpf_btf_load(const union bpf_attr *attr)
2317 {
2318 	if (CHECK_ATTR(BPF_BTF_LOAD))
2319 		return -EINVAL;
2320 
2321 	if (!capable(CAP_SYS_ADMIN))
2322 		return -EPERM;
2323 
2324 	return btf_new_fd(attr);
2325 }
2326 
2327 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2328 
2329 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2330 {
2331 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2332 		return -EINVAL;
2333 
2334 	if (!capable(CAP_SYS_ADMIN))
2335 		return -EPERM;
2336 
2337 	return btf_get_fd_by_id(attr->btf_id);
2338 }
2339 
2340 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2341 				    union bpf_attr __user *uattr,
2342 				    u32 prog_id, u32 fd_type,
2343 				    const char *buf, u64 probe_offset,
2344 				    u64 probe_addr)
2345 {
2346 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2347 	u32 len = buf ? strlen(buf) : 0, input_len;
2348 	int err = 0;
2349 
2350 	if (put_user(len, &uattr->task_fd_query.buf_len))
2351 		return -EFAULT;
2352 	input_len = attr->task_fd_query.buf_len;
2353 	if (input_len && ubuf) {
2354 		if (!len) {
2355 			/* nothing to copy, just make ubuf NULL terminated */
2356 			char zero = '\0';
2357 
2358 			if (put_user(zero, ubuf))
2359 				return -EFAULT;
2360 		} else if (input_len >= len + 1) {
2361 			/* ubuf can hold the string with NULL terminator */
2362 			if (copy_to_user(ubuf, buf, len + 1))
2363 				return -EFAULT;
2364 		} else {
2365 			/* ubuf cannot hold the string with NULL terminator,
2366 			 * do a partial copy with NULL terminator.
2367 			 */
2368 			char zero = '\0';
2369 
2370 			err = -ENOSPC;
2371 			if (copy_to_user(ubuf, buf, input_len - 1))
2372 				return -EFAULT;
2373 			if (put_user(zero, ubuf + input_len - 1))
2374 				return -EFAULT;
2375 		}
2376 	}
2377 
2378 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2379 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2380 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2381 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2382 		return -EFAULT;
2383 
2384 	return err;
2385 }
2386 
2387 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2388 
2389 static int bpf_task_fd_query(const union bpf_attr *attr,
2390 			     union bpf_attr __user *uattr)
2391 {
2392 	pid_t pid = attr->task_fd_query.pid;
2393 	u32 fd = attr->task_fd_query.fd;
2394 	const struct perf_event *event;
2395 	struct files_struct *files;
2396 	struct task_struct *task;
2397 	struct file *file;
2398 	int err;
2399 
2400 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2401 		return -EINVAL;
2402 
2403 	if (!capable(CAP_SYS_ADMIN))
2404 		return -EPERM;
2405 
2406 	if (attr->task_fd_query.flags != 0)
2407 		return -EINVAL;
2408 
2409 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2410 	if (!task)
2411 		return -ENOENT;
2412 
2413 	files = get_files_struct(task);
2414 	put_task_struct(task);
2415 	if (!files)
2416 		return -ENOENT;
2417 
2418 	err = 0;
2419 	spin_lock(&files->file_lock);
2420 	file = fcheck_files(files, fd);
2421 	if (!file)
2422 		err = -EBADF;
2423 	else
2424 		get_file(file);
2425 	spin_unlock(&files->file_lock);
2426 	put_files_struct(files);
2427 
2428 	if (err)
2429 		goto out;
2430 
2431 	if (file->f_op == &bpf_raw_tp_fops) {
2432 		struct bpf_raw_tracepoint *raw_tp = file->private_data;
2433 		struct bpf_raw_event_map *btp = raw_tp->btp;
2434 
2435 		err = bpf_task_fd_query_copy(attr, uattr,
2436 					     raw_tp->prog->aux->id,
2437 					     BPF_FD_TYPE_RAW_TRACEPOINT,
2438 					     btp->tp->name, 0, 0);
2439 		goto put_file;
2440 	}
2441 
2442 	event = perf_get_event(file);
2443 	if (!IS_ERR(event)) {
2444 		u64 probe_offset, probe_addr;
2445 		u32 prog_id, fd_type;
2446 		const char *buf;
2447 
2448 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2449 					      &buf, &probe_offset,
2450 					      &probe_addr);
2451 		if (!err)
2452 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2453 						     fd_type, buf,
2454 						     probe_offset,
2455 						     probe_addr);
2456 		goto put_file;
2457 	}
2458 
2459 	err = -ENOTSUPP;
2460 put_file:
2461 	fput(file);
2462 out:
2463 	return err;
2464 }
2465 
2466 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2467 {
2468 	union bpf_attr attr = {};
2469 	int err;
2470 
2471 	if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2472 		return -EPERM;
2473 
2474 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2475 	if (err)
2476 		return err;
2477 	size = min_t(u32, size, sizeof(attr));
2478 
2479 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
2480 	if (copy_from_user(&attr, uattr, size) != 0)
2481 		return -EFAULT;
2482 
2483 	err = security_bpf(cmd, &attr, size);
2484 	if (err < 0)
2485 		return err;
2486 
2487 	switch (cmd) {
2488 	case BPF_MAP_CREATE:
2489 		err = map_create(&attr);
2490 		break;
2491 	case BPF_MAP_LOOKUP_ELEM:
2492 		err = map_lookup_elem(&attr);
2493 		break;
2494 	case BPF_MAP_UPDATE_ELEM:
2495 		err = map_update_elem(&attr);
2496 		break;
2497 	case BPF_MAP_DELETE_ELEM:
2498 		err = map_delete_elem(&attr);
2499 		break;
2500 	case BPF_MAP_GET_NEXT_KEY:
2501 		err = map_get_next_key(&attr);
2502 		break;
2503 	case BPF_PROG_LOAD:
2504 		err = bpf_prog_load(&attr);
2505 		break;
2506 	case BPF_OBJ_PIN:
2507 		err = bpf_obj_pin(&attr);
2508 		break;
2509 	case BPF_OBJ_GET:
2510 		err = bpf_obj_get(&attr);
2511 		break;
2512 	case BPF_PROG_ATTACH:
2513 		err = bpf_prog_attach(&attr);
2514 		break;
2515 	case BPF_PROG_DETACH:
2516 		err = bpf_prog_detach(&attr);
2517 		break;
2518 	case BPF_PROG_QUERY:
2519 		err = bpf_prog_query(&attr, uattr);
2520 		break;
2521 	case BPF_PROG_TEST_RUN:
2522 		err = bpf_prog_test_run(&attr, uattr);
2523 		break;
2524 	case BPF_PROG_GET_NEXT_ID:
2525 		err = bpf_obj_get_next_id(&attr, uattr,
2526 					  &prog_idr, &prog_idr_lock);
2527 		break;
2528 	case BPF_MAP_GET_NEXT_ID:
2529 		err = bpf_obj_get_next_id(&attr, uattr,
2530 					  &map_idr, &map_idr_lock);
2531 		break;
2532 	case BPF_PROG_GET_FD_BY_ID:
2533 		err = bpf_prog_get_fd_by_id(&attr);
2534 		break;
2535 	case BPF_MAP_GET_FD_BY_ID:
2536 		err = bpf_map_get_fd_by_id(&attr);
2537 		break;
2538 	case BPF_OBJ_GET_INFO_BY_FD:
2539 		err = bpf_obj_get_info_by_fd(&attr, uattr);
2540 		break;
2541 	case BPF_RAW_TRACEPOINT_OPEN:
2542 		err = bpf_raw_tracepoint_open(&attr);
2543 		break;
2544 	case BPF_BTF_LOAD:
2545 		err = bpf_btf_load(&attr);
2546 		break;
2547 	case BPF_BTF_GET_FD_BY_ID:
2548 		err = bpf_btf_get_fd_by_id(&attr);
2549 		break;
2550 	case BPF_TASK_FD_QUERY:
2551 		err = bpf_task_fd_query(&attr, uattr);
2552 		break;
2553 	case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2554 		err = map_lookup_and_delete_elem(&attr);
2555 		break;
2556 	default:
2557 		err = -EINVAL;
2558 		break;
2559 	}
2560 
2561 	return err;
2562 }
2563