xref: /linux/kernel/bpf/syscall.c (revision b7d3826c2ed6c3e626e7ae796c5df2c0d2551c6a)
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 /* last field in 'union bpf_attr' used by this command */
655 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
656 
657 static int map_lookup_elem(union bpf_attr *attr)
658 {
659 	void __user *ukey = u64_to_user_ptr(attr->key);
660 	void __user *uvalue = u64_to_user_ptr(attr->value);
661 	int ufd = attr->map_fd;
662 	struct bpf_map *map;
663 	void *key, *value, *ptr;
664 	u32 value_size;
665 	struct fd f;
666 	int err;
667 
668 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
669 		return -EINVAL;
670 
671 	f = fdget(ufd);
672 	map = __bpf_map_get(f);
673 	if (IS_ERR(map))
674 		return PTR_ERR(map);
675 
676 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
677 		err = -EPERM;
678 		goto err_put;
679 	}
680 
681 	key = memdup_user(ukey, map->key_size);
682 	if (IS_ERR(key)) {
683 		err = PTR_ERR(key);
684 		goto err_put;
685 	}
686 
687 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
688 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
689 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
690 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
691 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
692 	else if (IS_FD_MAP(map))
693 		value_size = sizeof(u32);
694 	else
695 		value_size = map->value_size;
696 
697 	err = -ENOMEM;
698 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
699 	if (!value)
700 		goto free_key;
701 
702 	if (bpf_map_is_dev_bound(map)) {
703 		err = bpf_map_offload_lookup_elem(map, key, value);
704 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
705 		   map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
706 		err = bpf_percpu_hash_copy(map, key, value);
707 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
708 		err = bpf_percpu_array_copy(map, key, value);
709 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
710 		err = bpf_percpu_cgroup_storage_copy(map, key, value);
711 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
712 		err = bpf_stackmap_copy(map, key, value);
713 	} else if (IS_FD_ARRAY(map)) {
714 		err = bpf_fd_array_map_lookup_elem(map, key, value);
715 	} else if (IS_FD_HASH(map)) {
716 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
717 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
718 		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
719 	} else {
720 		rcu_read_lock();
721 		ptr = map->ops->map_lookup_elem(map, key);
722 		if (IS_ERR(ptr)) {
723 			err = PTR_ERR(ptr);
724 		} else if (!ptr) {
725 			err = -ENOENT;
726 		} else {
727 			err = 0;
728 			memcpy(value, ptr, value_size);
729 		}
730 		rcu_read_unlock();
731 	}
732 
733 	if (err)
734 		goto free_value;
735 
736 	err = -EFAULT;
737 	if (copy_to_user(uvalue, value, value_size) != 0)
738 		goto free_value;
739 
740 	err = 0;
741 
742 free_value:
743 	kfree(value);
744 free_key:
745 	kfree(key);
746 err_put:
747 	fdput(f);
748 	return err;
749 }
750 
751 static void maybe_wait_bpf_programs(struct bpf_map *map)
752 {
753 	/* Wait for any running BPF programs to complete so that
754 	 * userspace, when we return to it, knows that all programs
755 	 * that could be running use the new map value.
756 	 */
757 	if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
758 	    map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
759 		synchronize_rcu();
760 }
761 
762 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
763 
764 static int map_update_elem(union bpf_attr *attr)
765 {
766 	void __user *ukey = u64_to_user_ptr(attr->key);
767 	void __user *uvalue = u64_to_user_ptr(attr->value);
768 	int ufd = attr->map_fd;
769 	struct bpf_map *map;
770 	void *key, *value;
771 	u32 value_size;
772 	struct fd f;
773 	int err;
774 
775 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
776 		return -EINVAL;
777 
778 	f = fdget(ufd);
779 	map = __bpf_map_get(f);
780 	if (IS_ERR(map))
781 		return PTR_ERR(map);
782 
783 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
784 		err = -EPERM;
785 		goto err_put;
786 	}
787 
788 	key = memdup_user(ukey, map->key_size);
789 	if (IS_ERR(key)) {
790 		err = PTR_ERR(key);
791 		goto err_put;
792 	}
793 
794 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
795 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
796 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
797 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
798 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
799 	else
800 		value_size = map->value_size;
801 
802 	err = -ENOMEM;
803 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
804 	if (!value)
805 		goto free_key;
806 
807 	err = -EFAULT;
808 	if (copy_from_user(value, uvalue, value_size) != 0)
809 		goto free_value;
810 
811 	/* Need to create a kthread, thus must support schedule */
812 	if (bpf_map_is_dev_bound(map)) {
813 		err = bpf_map_offload_update_elem(map, key, value, attr->flags);
814 		goto out;
815 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
816 		   map->map_type == BPF_MAP_TYPE_SOCKHASH ||
817 		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
818 		err = map->ops->map_update_elem(map, key, value, attr->flags);
819 		goto out;
820 	}
821 
822 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
823 	 * inside bpf map update or delete otherwise deadlocks are possible
824 	 */
825 	preempt_disable();
826 	__this_cpu_inc(bpf_prog_active);
827 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
828 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
829 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
830 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
831 		err = bpf_percpu_array_update(map, key, value, attr->flags);
832 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
833 		err = bpf_percpu_cgroup_storage_update(map, key, value,
834 						       attr->flags);
835 	} else if (IS_FD_ARRAY(map)) {
836 		rcu_read_lock();
837 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
838 						   attr->flags);
839 		rcu_read_unlock();
840 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
841 		rcu_read_lock();
842 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
843 						  attr->flags);
844 		rcu_read_unlock();
845 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
846 		/* rcu_read_lock() is not needed */
847 		err = bpf_fd_reuseport_array_update_elem(map, key, value,
848 							 attr->flags);
849 	} else {
850 		rcu_read_lock();
851 		err = map->ops->map_update_elem(map, key, value, attr->flags);
852 		rcu_read_unlock();
853 	}
854 	__this_cpu_dec(bpf_prog_active);
855 	preempt_enable();
856 	maybe_wait_bpf_programs(map);
857 out:
858 free_value:
859 	kfree(value);
860 free_key:
861 	kfree(key);
862 err_put:
863 	fdput(f);
864 	return err;
865 }
866 
867 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
868 
869 static int map_delete_elem(union bpf_attr *attr)
870 {
871 	void __user *ukey = u64_to_user_ptr(attr->key);
872 	int ufd = attr->map_fd;
873 	struct bpf_map *map;
874 	struct fd f;
875 	void *key;
876 	int err;
877 
878 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
879 		return -EINVAL;
880 
881 	f = fdget(ufd);
882 	map = __bpf_map_get(f);
883 	if (IS_ERR(map))
884 		return PTR_ERR(map);
885 
886 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
887 		err = -EPERM;
888 		goto err_put;
889 	}
890 
891 	key = memdup_user(ukey, map->key_size);
892 	if (IS_ERR(key)) {
893 		err = PTR_ERR(key);
894 		goto err_put;
895 	}
896 
897 	if (bpf_map_is_dev_bound(map)) {
898 		err = bpf_map_offload_delete_elem(map, key);
899 		goto out;
900 	}
901 
902 	preempt_disable();
903 	__this_cpu_inc(bpf_prog_active);
904 	rcu_read_lock();
905 	err = map->ops->map_delete_elem(map, key);
906 	rcu_read_unlock();
907 	__this_cpu_dec(bpf_prog_active);
908 	preempt_enable();
909 	maybe_wait_bpf_programs(map);
910 out:
911 	kfree(key);
912 err_put:
913 	fdput(f);
914 	return err;
915 }
916 
917 /* last field in 'union bpf_attr' used by this command */
918 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
919 
920 static int map_get_next_key(union bpf_attr *attr)
921 {
922 	void __user *ukey = u64_to_user_ptr(attr->key);
923 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
924 	int ufd = attr->map_fd;
925 	struct bpf_map *map;
926 	void *key, *next_key;
927 	struct fd f;
928 	int err;
929 
930 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
931 		return -EINVAL;
932 
933 	f = fdget(ufd);
934 	map = __bpf_map_get(f);
935 	if (IS_ERR(map))
936 		return PTR_ERR(map);
937 
938 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
939 		err = -EPERM;
940 		goto err_put;
941 	}
942 
943 	if (ukey) {
944 		key = memdup_user(ukey, map->key_size);
945 		if (IS_ERR(key)) {
946 			err = PTR_ERR(key);
947 			goto err_put;
948 		}
949 	} else {
950 		key = NULL;
951 	}
952 
953 	err = -ENOMEM;
954 	next_key = kmalloc(map->key_size, GFP_USER);
955 	if (!next_key)
956 		goto free_key;
957 
958 	if (bpf_map_is_dev_bound(map)) {
959 		err = bpf_map_offload_get_next_key(map, key, next_key);
960 		goto out;
961 	}
962 
963 	rcu_read_lock();
964 	err = map->ops->map_get_next_key(map, key, next_key);
965 	rcu_read_unlock();
966 out:
967 	if (err)
968 		goto free_next_key;
969 
970 	err = -EFAULT;
971 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
972 		goto free_next_key;
973 
974 	err = 0;
975 
976 free_next_key:
977 	kfree(next_key);
978 free_key:
979 	kfree(key);
980 err_put:
981 	fdput(f);
982 	return err;
983 }
984 
985 static const struct bpf_prog_ops * const bpf_prog_types[] = {
986 #define BPF_PROG_TYPE(_id, _name) \
987 	[_id] = & _name ## _prog_ops,
988 #define BPF_MAP_TYPE(_id, _ops)
989 #include <linux/bpf_types.h>
990 #undef BPF_PROG_TYPE
991 #undef BPF_MAP_TYPE
992 };
993 
994 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
995 {
996 	const struct bpf_prog_ops *ops;
997 
998 	if (type >= ARRAY_SIZE(bpf_prog_types))
999 		return -EINVAL;
1000 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1001 	ops = bpf_prog_types[type];
1002 	if (!ops)
1003 		return -EINVAL;
1004 
1005 	if (!bpf_prog_is_dev_bound(prog->aux))
1006 		prog->aux->ops = ops;
1007 	else
1008 		prog->aux->ops = &bpf_offload_prog_ops;
1009 	prog->type = type;
1010 	return 0;
1011 }
1012 
1013 /* drop refcnt on maps used by eBPF program and free auxilary data */
1014 static void free_used_maps(struct bpf_prog_aux *aux)
1015 {
1016 	enum bpf_cgroup_storage_type stype;
1017 	int i;
1018 
1019 	for_each_cgroup_storage_type(stype) {
1020 		if (!aux->cgroup_storage[stype])
1021 			continue;
1022 		bpf_cgroup_storage_release(aux->prog,
1023 					   aux->cgroup_storage[stype]);
1024 	}
1025 
1026 	for (i = 0; i < aux->used_map_cnt; i++)
1027 		bpf_map_put(aux->used_maps[i]);
1028 
1029 	kfree(aux->used_maps);
1030 }
1031 
1032 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1033 {
1034 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1035 	unsigned long user_bufs;
1036 
1037 	if (user) {
1038 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1039 		if (user_bufs > memlock_limit) {
1040 			atomic_long_sub(pages, &user->locked_vm);
1041 			return -EPERM;
1042 		}
1043 	}
1044 
1045 	return 0;
1046 }
1047 
1048 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1049 {
1050 	if (user)
1051 		atomic_long_sub(pages, &user->locked_vm);
1052 }
1053 
1054 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1055 {
1056 	struct user_struct *user = get_current_user();
1057 	int ret;
1058 
1059 	ret = __bpf_prog_charge(user, prog->pages);
1060 	if (ret) {
1061 		free_uid(user);
1062 		return ret;
1063 	}
1064 
1065 	prog->aux->user = user;
1066 	return 0;
1067 }
1068 
1069 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1070 {
1071 	struct user_struct *user = prog->aux->user;
1072 
1073 	__bpf_prog_uncharge(user, prog->pages);
1074 	free_uid(user);
1075 }
1076 
1077 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1078 {
1079 	int id;
1080 
1081 	idr_preload(GFP_KERNEL);
1082 	spin_lock_bh(&prog_idr_lock);
1083 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1084 	if (id > 0)
1085 		prog->aux->id = id;
1086 	spin_unlock_bh(&prog_idr_lock);
1087 	idr_preload_end();
1088 
1089 	/* id is in [1, INT_MAX) */
1090 	if (WARN_ON_ONCE(!id))
1091 		return -ENOSPC;
1092 
1093 	return id > 0 ? 0 : id;
1094 }
1095 
1096 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1097 {
1098 	/* cBPF to eBPF migrations are currently not in the idr store.
1099 	 * Offloaded programs are removed from the store when their device
1100 	 * disappears - even if someone grabs an fd to them they are unusable,
1101 	 * simply waiting for refcnt to drop to be freed.
1102 	 */
1103 	if (!prog->aux->id)
1104 		return;
1105 
1106 	if (do_idr_lock)
1107 		spin_lock_bh(&prog_idr_lock);
1108 	else
1109 		__acquire(&prog_idr_lock);
1110 
1111 	idr_remove(&prog_idr, prog->aux->id);
1112 	prog->aux->id = 0;
1113 
1114 	if (do_idr_lock)
1115 		spin_unlock_bh(&prog_idr_lock);
1116 	else
1117 		__release(&prog_idr_lock);
1118 }
1119 
1120 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1121 {
1122 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1123 
1124 	free_used_maps(aux);
1125 	bpf_prog_uncharge_memlock(aux->prog);
1126 	security_bpf_prog_free(aux);
1127 	bpf_prog_free(aux->prog);
1128 }
1129 
1130 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1131 {
1132 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
1133 		/* bpf_prog_free_id() must be called first */
1134 		bpf_prog_free_id(prog, do_idr_lock);
1135 		bpf_prog_kallsyms_del_all(prog);
1136 
1137 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1138 	}
1139 }
1140 
1141 void bpf_prog_put(struct bpf_prog *prog)
1142 {
1143 	__bpf_prog_put(prog, true);
1144 }
1145 EXPORT_SYMBOL_GPL(bpf_prog_put);
1146 
1147 static int bpf_prog_release(struct inode *inode, struct file *filp)
1148 {
1149 	struct bpf_prog *prog = filp->private_data;
1150 
1151 	bpf_prog_put(prog);
1152 	return 0;
1153 }
1154 
1155 #ifdef CONFIG_PROC_FS
1156 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1157 {
1158 	const struct bpf_prog *prog = filp->private_data;
1159 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1160 
1161 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1162 	seq_printf(m,
1163 		   "prog_type:\t%u\n"
1164 		   "prog_jited:\t%u\n"
1165 		   "prog_tag:\t%s\n"
1166 		   "memlock:\t%llu\n"
1167 		   "prog_id:\t%u\n",
1168 		   prog->type,
1169 		   prog->jited,
1170 		   prog_tag,
1171 		   prog->pages * 1ULL << PAGE_SHIFT,
1172 		   prog->aux->id);
1173 }
1174 #endif
1175 
1176 const struct file_operations bpf_prog_fops = {
1177 #ifdef CONFIG_PROC_FS
1178 	.show_fdinfo	= bpf_prog_show_fdinfo,
1179 #endif
1180 	.release	= bpf_prog_release,
1181 	.read		= bpf_dummy_read,
1182 	.write		= bpf_dummy_write,
1183 };
1184 
1185 int bpf_prog_new_fd(struct bpf_prog *prog)
1186 {
1187 	int ret;
1188 
1189 	ret = security_bpf_prog(prog);
1190 	if (ret < 0)
1191 		return ret;
1192 
1193 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1194 				O_RDWR | O_CLOEXEC);
1195 }
1196 
1197 static struct bpf_prog *____bpf_prog_get(struct fd f)
1198 {
1199 	if (!f.file)
1200 		return ERR_PTR(-EBADF);
1201 	if (f.file->f_op != &bpf_prog_fops) {
1202 		fdput(f);
1203 		return ERR_PTR(-EINVAL);
1204 	}
1205 
1206 	return f.file->private_data;
1207 }
1208 
1209 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1210 {
1211 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1212 		atomic_sub(i, &prog->aux->refcnt);
1213 		return ERR_PTR(-EBUSY);
1214 	}
1215 	return prog;
1216 }
1217 EXPORT_SYMBOL_GPL(bpf_prog_add);
1218 
1219 void bpf_prog_sub(struct bpf_prog *prog, int i)
1220 {
1221 	/* Only to be used for undoing previous bpf_prog_add() in some
1222 	 * error path. We still know that another entity in our call
1223 	 * path holds a reference to the program, thus atomic_sub() can
1224 	 * be safely used in such cases!
1225 	 */
1226 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1227 }
1228 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1229 
1230 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1231 {
1232 	return bpf_prog_add(prog, 1);
1233 }
1234 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1235 
1236 /* prog_idr_lock should have been held */
1237 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1238 {
1239 	int refold;
1240 
1241 	refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1242 
1243 	if (refold >= BPF_MAX_REFCNT) {
1244 		__bpf_prog_put(prog, false);
1245 		return ERR_PTR(-EBUSY);
1246 	}
1247 
1248 	if (!refold)
1249 		return ERR_PTR(-ENOENT);
1250 
1251 	return prog;
1252 }
1253 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1254 
1255 bool bpf_prog_get_ok(struct bpf_prog *prog,
1256 			    enum bpf_prog_type *attach_type, bool attach_drv)
1257 {
1258 	/* not an attachment, just a refcount inc, always allow */
1259 	if (!attach_type)
1260 		return true;
1261 
1262 	if (prog->type != *attach_type)
1263 		return false;
1264 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1265 		return false;
1266 
1267 	return true;
1268 }
1269 
1270 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1271 				       bool attach_drv)
1272 {
1273 	struct fd f = fdget(ufd);
1274 	struct bpf_prog *prog;
1275 
1276 	prog = ____bpf_prog_get(f);
1277 	if (IS_ERR(prog))
1278 		return prog;
1279 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1280 		prog = ERR_PTR(-EINVAL);
1281 		goto out;
1282 	}
1283 
1284 	prog = bpf_prog_inc(prog);
1285 out:
1286 	fdput(f);
1287 	return prog;
1288 }
1289 
1290 struct bpf_prog *bpf_prog_get(u32 ufd)
1291 {
1292 	return __bpf_prog_get(ufd, NULL, false);
1293 }
1294 
1295 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1296 				       bool attach_drv)
1297 {
1298 	return __bpf_prog_get(ufd, &type, attach_drv);
1299 }
1300 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1301 
1302 /* Initially all BPF programs could be loaded w/o specifying
1303  * expected_attach_type. Later for some of them specifying expected_attach_type
1304  * at load time became required so that program could be validated properly.
1305  * Programs of types that are allowed to be loaded both w/ and w/o (for
1306  * backward compatibility) expected_attach_type, should have the default attach
1307  * type assigned to expected_attach_type for the latter case, so that it can be
1308  * validated later at attach time.
1309  *
1310  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1311  * prog type requires it but has some attach types that have to be backward
1312  * compatible.
1313  */
1314 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1315 {
1316 	switch (attr->prog_type) {
1317 	case BPF_PROG_TYPE_CGROUP_SOCK:
1318 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1319 		 * exist so checking for non-zero is the way to go here.
1320 		 */
1321 		if (!attr->expected_attach_type)
1322 			attr->expected_attach_type =
1323 				BPF_CGROUP_INET_SOCK_CREATE;
1324 		break;
1325 	}
1326 }
1327 
1328 static int
1329 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1330 				enum bpf_attach_type expected_attach_type)
1331 {
1332 	switch (prog_type) {
1333 	case BPF_PROG_TYPE_CGROUP_SOCK:
1334 		switch (expected_attach_type) {
1335 		case BPF_CGROUP_INET_SOCK_CREATE:
1336 		case BPF_CGROUP_INET4_POST_BIND:
1337 		case BPF_CGROUP_INET6_POST_BIND:
1338 			return 0;
1339 		default:
1340 			return -EINVAL;
1341 		}
1342 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1343 		switch (expected_attach_type) {
1344 		case BPF_CGROUP_INET4_BIND:
1345 		case BPF_CGROUP_INET6_BIND:
1346 		case BPF_CGROUP_INET4_CONNECT:
1347 		case BPF_CGROUP_INET6_CONNECT:
1348 		case BPF_CGROUP_UDP4_SENDMSG:
1349 		case BPF_CGROUP_UDP6_SENDMSG:
1350 			return 0;
1351 		default:
1352 			return -EINVAL;
1353 		}
1354 	default:
1355 		return 0;
1356 	}
1357 }
1358 
1359 /* last field in 'union bpf_attr' used by this command */
1360 #define	BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1361 
1362 static int bpf_prog_load(union bpf_attr *attr)
1363 {
1364 	enum bpf_prog_type type = attr->prog_type;
1365 	struct bpf_prog *prog;
1366 	int err;
1367 	char license[128];
1368 	bool is_gpl;
1369 
1370 	if (CHECK_ATTR(BPF_PROG_LOAD))
1371 		return -EINVAL;
1372 
1373 	if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1374 		return -EINVAL;
1375 
1376 	/* copy eBPF program license from user space */
1377 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1378 			      sizeof(license) - 1) < 0)
1379 		return -EFAULT;
1380 	license[sizeof(license) - 1] = 0;
1381 
1382 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1383 	is_gpl = license_is_gpl_compatible(license);
1384 
1385 	if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1386 		return -E2BIG;
1387 
1388 	if (type == BPF_PROG_TYPE_KPROBE &&
1389 	    attr->kern_version != LINUX_VERSION_CODE)
1390 		return -EINVAL;
1391 
1392 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1393 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
1394 	    !capable(CAP_SYS_ADMIN))
1395 		return -EPERM;
1396 
1397 	bpf_prog_load_fixup_attach_type(attr);
1398 	if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1399 		return -EINVAL;
1400 
1401 	/* plain bpf_prog allocation */
1402 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1403 	if (!prog)
1404 		return -ENOMEM;
1405 
1406 	prog->expected_attach_type = attr->expected_attach_type;
1407 
1408 	prog->aux->offload_requested = !!attr->prog_ifindex;
1409 
1410 	err = security_bpf_prog_alloc(prog->aux);
1411 	if (err)
1412 		goto free_prog_nouncharge;
1413 
1414 	err = bpf_prog_charge_memlock(prog);
1415 	if (err)
1416 		goto free_prog_sec;
1417 
1418 	prog->len = attr->insn_cnt;
1419 
1420 	err = -EFAULT;
1421 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1422 			   bpf_prog_insn_size(prog)) != 0)
1423 		goto free_prog;
1424 
1425 	prog->orig_prog = NULL;
1426 	prog->jited = 0;
1427 
1428 	atomic_set(&prog->aux->refcnt, 1);
1429 	prog->gpl_compatible = is_gpl ? 1 : 0;
1430 
1431 	if (bpf_prog_is_dev_bound(prog->aux)) {
1432 		err = bpf_prog_offload_init(prog, attr);
1433 		if (err)
1434 			goto free_prog;
1435 	}
1436 
1437 	/* find program type: socket_filter vs tracing_filter */
1438 	err = find_prog_type(type, prog);
1439 	if (err < 0)
1440 		goto free_prog;
1441 
1442 	prog->aux->load_time = ktime_get_boot_ns();
1443 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1444 	if (err)
1445 		goto free_prog;
1446 
1447 	/* run eBPF verifier */
1448 	err = bpf_check(&prog, attr);
1449 	if (err < 0)
1450 		goto free_used_maps;
1451 
1452 	prog = bpf_prog_select_runtime(prog, &err);
1453 	if (err < 0)
1454 		goto free_used_maps;
1455 
1456 	err = bpf_prog_alloc_id(prog);
1457 	if (err)
1458 		goto free_used_maps;
1459 
1460 	err = bpf_prog_new_fd(prog);
1461 	if (err < 0) {
1462 		/* failed to allocate fd.
1463 		 * bpf_prog_put() is needed because the above
1464 		 * bpf_prog_alloc_id() has published the prog
1465 		 * to the userspace and the userspace may
1466 		 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1467 		 */
1468 		bpf_prog_put(prog);
1469 		return err;
1470 	}
1471 
1472 	bpf_prog_kallsyms_add(prog);
1473 	return err;
1474 
1475 free_used_maps:
1476 	bpf_prog_kallsyms_del_subprogs(prog);
1477 	free_used_maps(prog->aux);
1478 free_prog:
1479 	bpf_prog_uncharge_memlock(prog);
1480 free_prog_sec:
1481 	security_bpf_prog_free(prog->aux);
1482 free_prog_nouncharge:
1483 	bpf_prog_free(prog);
1484 	return err;
1485 }
1486 
1487 #define BPF_OBJ_LAST_FIELD file_flags
1488 
1489 static int bpf_obj_pin(const union bpf_attr *attr)
1490 {
1491 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1492 		return -EINVAL;
1493 
1494 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1495 }
1496 
1497 static int bpf_obj_get(const union bpf_attr *attr)
1498 {
1499 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1500 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1501 		return -EINVAL;
1502 
1503 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1504 				attr->file_flags);
1505 }
1506 
1507 struct bpf_raw_tracepoint {
1508 	struct bpf_raw_event_map *btp;
1509 	struct bpf_prog *prog;
1510 };
1511 
1512 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1513 {
1514 	struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1515 
1516 	if (raw_tp->prog) {
1517 		bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1518 		bpf_prog_put(raw_tp->prog);
1519 	}
1520 	kfree(raw_tp);
1521 	return 0;
1522 }
1523 
1524 static const struct file_operations bpf_raw_tp_fops = {
1525 	.release	= bpf_raw_tracepoint_release,
1526 	.read		= bpf_dummy_read,
1527 	.write		= bpf_dummy_write,
1528 };
1529 
1530 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1531 
1532 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1533 {
1534 	struct bpf_raw_tracepoint *raw_tp;
1535 	struct bpf_raw_event_map *btp;
1536 	struct bpf_prog *prog;
1537 	char tp_name[128];
1538 	int tp_fd, err;
1539 
1540 	if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1541 			      sizeof(tp_name) - 1) < 0)
1542 		return -EFAULT;
1543 	tp_name[sizeof(tp_name) - 1] = 0;
1544 
1545 	btp = bpf_find_raw_tracepoint(tp_name);
1546 	if (!btp)
1547 		return -ENOENT;
1548 
1549 	raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1550 	if (!raw_tp)
1551 		return -ENOMEM;
1552 	raw_tp->btp = btp;
1553 
1554 	prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1555 				 BPF_PROG_TYPE_RAW_TRACEPOINT);
1556 	if (IS_ERR(prog)) {
1557 		err = PTR_ERR(prog);
1558 		goto out_free_tp;
1559 	}
1560 
1561 	err = bpf_probe_register(raw_tp->btp, prog);
1562 	if (err)
1563 		goto out_put_prog;
1564 
1565 	raw_tp->prog = prog;
1566 	tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1567 				 O_CLOEXEC);
1568 	if (tp_fd < 0) {
1569 		bpf_probe_unregister(raw_tp->btp, prog);
1570 		err = tp_fd;
1571 		goto out_put_prog;
1572 	}
1573 	return tp_fd;
1574 
1575 out_put_prog:
1576 	bpf_prog_put(prog);
1577 out_free_tp:
1578 	kfree(raw_tp);
1579 	return err;
1580 }
1581 
1582 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1583 					     enum bpf_attach_type attach_type)
1584 {
1585 	switch (prog->type) {
1586 	case BPF_PROG_TYPE_CGROUP_SOCK:
1587 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1588 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1589 	default:
1590 		return 0;
1591 	}
1592 }
1593 
1594 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1595 
1596 #define BPF_F_ATTACH_MASK \
1597 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1598 
1599 static int bpf_prog_attach(const union bpf_attr *attr)
1600 {
1601 	enum bpf_prog_type ptype;
1602 	struct bpf_prog *prog;
1603 	int ret;
1604 
1605 	if (!capable(CAP_NET_ADMIN))
1606 		return -EPERM;
1607 
1608 	if (CHECK_ATTR(BPF_PROG_ATTACH))
1609 		return -EINVAL;
1610 
1611 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1612 		return -EINVAL;
1613 
1614 	switch (attr->attach_type) {
1615 	case BPF_CGROUP_INET_INGRESS:
1616 	case BPF_CGROUP_INET_EGRESS:
1617 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1618 		break;
1619 	case BPF_CGROUP_INET_SOCK_CREATE:
1620 	case BPF_CGROUP_INET4_POST_BIND:
1621 	case BPF_CGROUP_INET6_POST_BIND:
1622 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1623 		break;
1624 	case BPF_CGROUP_INET4_BIND:
1625 	case BPF_CGROUP_INET6_BIND:
1626 	case BPF_CGROUP_INET4_CONNECT:
1627 	case BPF_CGROUP_INET6_CONNECT:
1628 	case BPF_CGROUP_UDP4_SENDMSG:
1629 	case BPF_CGROUP_UDP6_SENDMSG:
1630 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1631 		break;
1632 	case BPF_CGROUP_SOCK_OPS:
1633 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1634 		break;
1635 	case BPF_CGROUP_DEVICE:
1636 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1637 		break;
1638 	case BPF_SK_MSG_VERDICT:
1639 		ptype = BPF_PROG_TYPE_SK_MSG;
1640 		break;
1641 	case BPF_SK_SKB_STREAM_PARSER:
1642 	case BPF_SK_SKB_STREAM_VERDICT:
1643 		ptype = BPF_PROG_TYPE_SK_SKB;
1644 		break;
1645 	case BPF_LIRC_MODE2:
1646 		ptype = BPF_PROG_TYPE_LIRC_MODE2;
1647 		break;
1648 	case BPF_FLOW_DISSECTOR:
1649 		ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1650 		break;
1651 	default:
1652 		return -EINVAL;
1653 	}
1654 
1655 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1656 	if (IS_ERR(prog))
1657 		return PTR_ERR(prog);
1658 
1659 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1660 		bpf_prog_put(prog);
1661 		return -EINVAL;
1662 	}
1663 
1664 	switch (ptype) {
1665 	case BPF_PROG_TYPE_SK_SKB:
1666 	case BPF_PROG_TYPE_SK_MSG:
1667 		ret = sock_map_get_from_fd(attr, prog);
1668 		break;
1669 	case BPF_PROG_TYPE_LIRC_MODE2:
1670 		ret = lirc_prog_attach(attr, prog);
1671 		break;
1672 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
1673 		ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1674 		break;
1675 	default:
1676 		ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1677 	}
1678 
1679 	if (ret)
1680 		bpf_prog_put(prog);
1681 	return ret;
1682 }
1683 
1684 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1685 
1686 static int bpf_prog_detach(const union bpf_attr *attr)
1687 {
1688 	enum bpf_prog_type ptype;
1689 
1690 	if (!capable(CAP_NET_ADMIN))
1691 		return -EPERM;
1692 
1693 	if (CHECK_ATTR(BPF_PROG_DETACH))
1694 		return -EINVAL;
1695 
1696 	switch (attr->attach_type) {
1697 	case BPF_CGROUP_INET_INGRESS:
1698 	case BPF_CGROUP_INET_EGRESS:
1699 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1700 		break;
1701 	case BPF_CGROUP_INET_SOCK_CREATE:
1702 	case BPF_CGROUP_INET4_POST_BIND:
1703 	case BPF_CGROUP_INET6_POST_BIND:
1704 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1705 		break;
1706 	case BPF_CGROUP_INET4_BIND:
1707 	case BPF_CGROUP_INET6_BIND:
1708 	case BPF_CGROUP_INET4_CONNECT:
1709 	case BPF_CGROUP_INET6_CONNECT:
1710 	case BPF_CGROUP_UDP4_SENDMSG:
1711 	case BPF_CGROUP_UDP6_SENDMSG:
1712 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1713 		break;
1714 	case BPF_CGROUP_SOCK_OPS:
1715 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1716 		break;
1717 	case BPF_CGROUP_DEVICE:
1718 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1719 		break;
1720 	case BPF_SK_MSG_VERDICT:
1721 		return sock_map_get_from_fd(attr, NULL);
1722 	case BPF_SK_SKB_STREAM_PARSER:
1723 	case BPF_SK_SKB_STREAM_VERDICT:
1724 		return sock_map_get_from_fd(attr, NULL);
1725 	case BPF_LIRC_MODE2:
1726 		return lirc_prog_detach(attr);
1727 	case BPF_FLOW_DISSECTOR:
1728 		return skb_flow_dissector_bpf_prog_detach(attr);
1729 	default:
1730 		return -EINVAL;
1731 	}
1732 
1733 	return cgroup_bpf_prog_detach(attr, ptype);
1734 }
1735 
1736 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1737 
1738 static int bpf_prog_query(const union bpf_attr *attr,
1739 			  union bpf_attr __user *uattr)
1740 {
1741 	if (!capable(CAP_NET_ADMIN))
1742 		return -EPERM;
1743 	if (CHECK_ATTR(BPF_PROG_QUERY))
1744 		return -EINVAL;
1745 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1746 		return -EINVAL;
1747 
1748 	switch (attr->query.attach_type) {
1749 	case BPF_CGROUP_INET_INGRESS:
1750 	case BPF_CGROUP_INET_EGRESS:
1751 	case BPF_CGROUP_INET_SOCK_CREATE:
1752 	case BPF_CGROUP_INET4_BIND:
1753 	case BPF_CGROUP_INET6_BIND:
1754 	case BPF_CGROUP_INET4_POST_BIND:
1755 	case BPF_CGROUP_INET6_POST_BIND:
1756 	case BPF_CGROUP_INET4_CONNECT:
1757 	case BPF_CGROUP_INET6_CONNECT:
1758 	case BPF_CGROUP_UDP4_SENDMSG:
1759 	case BPF_CGROUP_UDP6_SENDMSG:
1760 	case BPF_CGROUP_SOCK_OPS:
1761 	case BPF_CGROUP_DEVICE:
1762 		break;
1763 	case BPF_LIRC_MODE2:
1764 		return lirc_prog_query(attr, uattr);
1765 	default:
1766 		return -EINVAL;
1767 	}
1768 
1769 	return cgroup_bpf_prog_query(attr, uattr);
1770 }
1771 
1772 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1773 
1774 static int bpf_prog_test_run(const union bpf_attr *attr,
1775 			     union bpf_attr __user *uattr)
1776 {
1777 	struct bpf_prog *prog;
1778 	int ret = -ENOTSUPP;
1779 
1780 	if (!capable(CAP_SYS_ADMIN))
1781 		return -EPERM;
1782 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1783 		return -EINVAL;
1784 
1785 	prog = bpf_prog_get(attr->test.prog_fd);
1786 	if (IS_ERR(prog))
1787 		return PTR_ERR(prog);
1788 
1789 	if (prog->aux->ops->test_run)
1790 		ret = prog->aux->ops->test_run(prog, attr, uattr);
1791 
1792 	bpf_prog_put(prog);
1793 	return ret;
1794 }
1795 
1796 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1797 
1798 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1799 			       union bpf_attr __user *uattr,
1800 			       struct idr *idr,
1801 			       spinlock_t *lock)
1802 {
1803 	u32 next_id = attr->start_id;
1804 	int err = 0;
1805 
1806 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1807 		return -EINVAL;
1808 
1809 	if (!capable(CAP_SYS_ADMIN))
1810 		return -EPERM;
1811 
1812 	next_id++;
1813 	spin_lock_bh(lock);
1814 	if (!idr_get_next(idr, &next_id))
1815 		err = -ENOENT;
1816 	spin_unlock_bh(lock);
1817 
1818 	if (!err)
1819 		err = put_user(next_id, &uattr->next_id);
1820 
1821 	return err;
1822 }
1823 
1824 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1825 
1826 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1827 {
1828 	struct bpf_prog *prog;
1829 	u32 id = attr->prog_id;
1830 	int fd;
1831 
1832 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1833 		return -EINVAL;
1834 
1835 	if (!capable(CAP_SYS_ADMIN))
1836 		return -EPERM;
1837 
1838 	spin_lock_bh(&prog_idr_lock);
1839 	prog = idr_find(&prog_idr, id);
1840 	if (prog)
1841 		prog = bpf_prog_inc_not_zero(prog);
1842 	else
1843 		prog = ERR_PTR(-ENOENT);
1844 	spin_unlock_bh(&prog_idr_lock);
1845 
1846 	if (IS_ERR(prog))
1847 		return PTR_ERR(prog);
1848 
1849 	fd = bpf_prog_new_fd(prog);
1850 	if (fd < 0)
1851 		bpf_prog_put(prog);
1852 
1853 	return fd;
1854 }
1855 
1856 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1857 
1858 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1859 {
1860 	struct bpf_map *map;
1861 	u32 id = attr->map_id;
1862 	int f_flags;
1863 	int fd;
1864 
1865 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1866 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1867 		return -EINVAL;
1868 
1869 	if (!capable(CAP_SYS_ADMIN))
1870 		return -EPERM;
1871 
1872 	f_flags = bpf_get_file_flag(attr->open_flags);
1873 	if (f_flags < 0)
1874 		return f_flags;
1875 
1876 	spin_lock_bh(&map_idr_lock);
1877 	map = idr_find(&map_idr, id);
1878 	if (map)
1879 		map = bpf_map_inc_not_zero(map, true);
1880 	else
1881 		map = ERR_PTR(-ENOENT);
1882 	spin_unlock_bh(&map_idr_lock);
1883 
1884 	if (IS_ERR(map))
1885 		return PTR_ERR(map);
1886 
1887 	fd = bpf_map_new_fd(map, f_flags);
1888 	if (fd < 0)
1889 		bpf_map_put(map);
1890 
1891 	return fd;
1892 }
1893 
1894 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1895 					      unsigned long addr)
1896 {
1897 	int i;
1898 
1899 	for (i = 0; i < prog->aux->used_map_cnt; i++)
1900 		if (prog->aux->used_maps[i] == (void *)addr)
1901 			return prog->aux->used_maps[i];
1902 	return NULL;
1903 }
1904 
1905 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1906 {
1907 	const struct bpf_map *map;
1908 	struct bpf_insn *insns;
1909 	u64 imm;
1910 	int i;
1911 
1912 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1913 			GFP_USER);
1914 	if (!insns)
1915 		return insns;
1916 
1917 	for (i = 0; i < prog->len; i++) {
1918 		if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1919 			insns[i].code = BPF_JMP | BPF_CALL;
1920 			insns[i].imm = BPF_FUNC_tail_call;
1921 			/* fall-through */
1922 		}
1923 		if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1924 		    insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1925 			if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1926 				insns[i].code = BPF_JMP | BPF_CALL;
1927 			if (!bpf_dump_raw_ok())
1928 				insns[i].imm = 0;
1929 			continue;
1930 		}
1931 
1932 		if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1933 			continue;
1934 
1935 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1936 		map = bpf_map_from_imm(prog, imm);
1937 		if (map) {
1938 			insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1939 			insns[i].imm = map->id;
1940 			insns[i + 1].imm = 0;
1941 			continue;
1942 		}
1943 
1944 		if (!bpf_dump_raw_ok() &&
1945 		    imm == (unsigned long)prog->aux) {
1946 			insns[i].imm = 0;
1947 			insns[i + 1].imm = 0;
1948 			continue;
1949 		}
1950 	}
1951 
1952 	return insns;
1953 }
1954 
1955 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1956 				   const union bpf_attr *attr,
1957 				   union bpf_attr __user *uattr)
1958 {
1959 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1960 	struct bpf_prog_info info = {};
1961 	u32 info_len = attr->info.info_len;
1962 	char __user *uinsns;
1963 	u32 ulen;
1964 	int err;
1965 
1966 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1967 	if (err)
1968 		return err;
1969 	info_len = min_t(u32, sizeof(info), info_len);
1970 
1971 	if (copy_from_user(&info, uinfo, info_len))
1972 		return -EFAULT;
1973 
1974 	info.type = prog->type;
1975 	info.id = prog->aux->id;
1976 	info.load_time = prog->aux->load_time;
1977 	info.created_by_uid = from_kuid_munged(current_user_ns(),
1978 					       prog->aux->user->uid);
1979 	info.gpl_compatible = prog->gpl_compatible;
1980 
1981 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
1982 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1983 
1984 	ulen = info.nr_map_ids;
1985 	info.nr_map_ids = prog->aux->used_map_cnt;
1986 	ulen = min_t(u32, info.nr_map_ids, ulen);
1987 	if (ulen) {
1988 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1989 		u32 i;
1990 
1991 		for (i = 0; i < ulen; i++)
1992 			if (put_user(prog->aux->used_maps[i]->id,
1993 				     &user_map_ids[i]))
1994 				return -EFAULT;
1995 	}
1996 
1997 	if (!capable(CAP_SYS_ADMIN)) {
1998 		info.jited_prog_len = 0;
1999 		info.xlated_prog_len = 0;
2000 		info.nr_jited_ksyms = 0;
2001 		goto done;
2002 	}
2003 
2004 	ulen = info.xlated_prog_len;
2005 	info.xlated_prog_len = bpf_prog_insn_size(prog);
2006 	if (info.xlated_prog_len && ulen) {
2007 		struct bpf_insn *insns_sanitized;
2008 		bool fault;
2009 
2010 		if (prog->blinded && !bpf_dump_raw_ok()) {
2011 			info.xlated_prog_insns = 0;
2012 			goto done;
2013 		}
2014 		insns_sanitized = bpf_insn_prepare_dump(prog);
2015 		if (!insns_sanitized)
2016 			return -ENOMEM;
2017 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2018 		ulen = min_t(u32, info.xlated_prog_len, ulen);
2019 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
2020 		kfree(insns_sanitized);
2021 		if (fault)
2022 			return -EFAULT;
2023 	}
2024 
2025 	if (bpf_prog_is_dev_bound(prog->aux)) {
2026 		err = bpf_prog_offload_info_fill(&info, prog);
2027 		if (err)
2028 			return err;
2029 		goto done;
2030 	}
2031 
2032 	/* NOTE: the following code is supposed to be skipped for offload.
2033 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
2034 	 * for offload.
2035 	 */
2036 	ulen = info.jited_prog_len;
2037 	if (prog->aux->func_cnt) {
2038 		u32 i;
2039 
2040 		info.jited_prog_len = 0;
2041 		for (i = 0; i < prog->aux->func_cnt; i++)
2042 			info.jited_prog_len += prog->aux->func[i]->jited_len;
2043 	} else {
2044 		info.jited_prog_len = prog->jited_len;
2045 	}
2046 
2047 	if (info.jited_prog_len && ulen) {
2048 		if (bpf_dump_raw_ok()) {
2049 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
2050 			ulen = min_t(u32, info.jited_prog_len, ulen);
2051 
2052 			/* for multi-function programs, copy the JITed
2053 			 * instructions for all the functions
2054 			 */
2055 			if (prog->aux->func_cnt) {
2056 				u32 len, free, i;
2057 				u8 *img;
2058 
2059 				free = ulen;
2060 				for (i = 0; i < prog->aux->func_cnt; i++) {
2061 					len = prog->aux->func[i]->jited_len;
2062 					len = min_t(u32, len, free);
2063 					img = (u8 *) prog->aux->func[i]->bpf_func;
2064 					if (copy_to_user(uinsns, img, len))
2065 						return -EFAULT;
2066 					uinsns += len;
2067 					free -= len;
2068 					if (!free)
2069 						break;
2070 				}
2071 			} else {
2072 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
2073 					return -EFAULT;
2074 			}
2075 		} else {
2076 			info.jited_prog_insns = 0;
2077 		}
2078 	}
2079 
2080 	ulen = info.nr_jited_ksyms;
2081 	info.nr_jited_ksyms = prog->aux->func_cnt;
2082 	if (info.nr_jited_ksyms && ulen) {
2083 		if (bpf_dump_raw_ok()) {
2084 			u64 __user *user_ksyms;
2085 			ulong ksym_addr;
2086 			u32 i;
2087 
2088 			/* copy the address of the kernel symbol
2089 			 * corresponding to each function
2090 			 */
2091 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2092 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2093 			for (i = 0; i < ulen; i++) {
2094 				ksym_addr = (ulong) prog->aux->func[i]->bpf_func;
2095 				ksym_addr &= PAGE_MASK;
2096 				if (put_user((u64) ksym_addr, &user_ksyms[i]))
2097 					return -EFAULT;
2098 			}
2099 		} else {
2100 			info.jited_ksyms = 0;
2101 		}
2102 	}
2103 
2104 	ulen = info.nr_jited_func_lens;
2105 	info.nr_jited_func_lens = prog->aux->func_cnt;
2106 	if (info.nr_jited_func_lens && ulen) {
2107 		if (bpf_dump_raw_ok()) {
2108 			u32 __user *user_lens;
2109 			u32 func_len, i;
2110 
2111 			/* copy the JITed image lengths for each function */
2112 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2113 			user_lens = u64_to_user_ptr(info.jited_func_lens);
2114 			for (i = 0; i < ulen; i++) {
2115 				func_len = prog->aux->func[i]->jited_len;
2116 				if (put_user(func_len, &user_lens[i]))
2117 					return -EFAULT;
2118 			}
2119 		} else {
2120 			info.jited_func_lens = 0;
2121 		}
2122 	}
2123 
2124 done:
2125 	if (copy_to_user(uinfo, &info, info_len) ||
2126 	    put_user(info_len, &uattr->info.info_len))
2127 		return -EFAULT;
2128 
2129 	return 0;
2130 }
2131 
2132 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2133 				  const union bpf_attr *attr,
2134 				  union bpf_attr __user *uattr)
2135 {
2136 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2137 	struct bpf_map_info info = {};
2138 	u32 info_len = attr->info.info_len;
2139 	int err;
2140 
2141 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2142 	if (err)
2143 		return err;
2144 	info_len = min_t(u32, sizeof(info), info_len);
2145 
2146 	info.type = map->map_type;
2147 	info.id = map->id;
2148 	info.key_size = map->key_size;
2149 	info.value_size = map->value_size;
2150 	info.max_entries = map->max_entries;
2151 	info.map_flags = map->map_flags;
2152 	memcpy(info.name, map->name, sizeof(map->name));
2153 
2154 	if (map->btf) {
2155 		info.btf_id = btf_id(map->btf);
2156 		info.btf_key_type_id = map->btf_key_type_id;
2157 		info.btf_value_type_id = map->btf_value_type_id;
2158 	}
2159 
2160 	if (bpf_map_is_dev_bound(map)) {
2161 		err = bpf_map_offload_info_fill(&info, map);
2162 		if (err)
2163 			return err;
2164 	}
2165 
2166 	if (copy_to_user(uinfo, &info, info_len) ||
2167 	    put_user(info_len, &uattr->info.info_len))
2168 		return -EFAULT;
2169 
2170 	return 0;
2171 }
2172 
2173 static int bpf_btf_get_info_by_fd(struct btf *btf,
2174 				  const union bpf_attr *attr,
2175 				  union bpf_attr __user *uattr)
2176 {
2177 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2178 	u32 info_len = attr->info.info_len;
2179 	int err;
2180 
2181 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2182 	if (err)
2183 		return err;
2184 
2185 	return btf_get_info_by_fd(btf, attr, uattr);
2186 }
2187 
2188 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2189 
2190 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2191 				  union bpf_attr __user *uattr)
2192 {
2193 	int ufd = attr->info.bpf_fd;
2194 	struct fd f;
2195 	int err;
2196 
2197 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2198 		return -EINVAL;
2199 
2200 	f = fdget(ufd);
2201 	if (!f.file)
2202 		return -EBADFD;
2203 
2204 	if (f.file->f_op == &bpf_prog_fops)
2205 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2206 					      uattr);
2207 	else if (f.file->f_op == &bpf_map_fops)
2208 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2209 					     uattr);
2210 	else if (f.file->f_op == &btf_fops)
2211 		err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2212 	else
2213 		err = -EINVAL;
2214 
2215 	fdput(f);
2216 	return err;
2217 }
2218 
2219 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2220 
2221 static int bpf_btf_load(const union bpf_attr *attr)
2222 {
2223 	if (CHECK_ATTR(BPF_BTF_LOAD))
2224 		return -EINVAL;
2225 
2226 	if (!capable(CAP_SYS_ADMIN))
2227 		return -EPERM;
2228 
2229 	return btf_new_fd(attr);
2230 }
2231 
2232 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2233 
2234 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2235 {
2236 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2237 		return -EINVAL;
2238 
2239 	if (!capable(CAP_SYS_ADMIN))
2240 		return -EPERM;
2241 
2242 	return btf_get_fd_by_id(attr->btf_id);
2243 }
2244 
2245 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2246 				    union bpf_attr __user *uattr,
2247 				    u32 prog_id, u32 fd_type,
2248 				    const char *buf, u64 probe_offset,
2249 				    u64 probe_addr)
2250 {
2251 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2252 	u32 len = buf ? strlen(buf) : 0, input_len;
2253 	int err = 0;
2254 
2255 	if (put_user(len, &uattr->task_fd_query.buf_len))
2256 		return -EFAULT;
2257 	input_len = attr->task_fd_query.buf_len;
2258 	if (input_len && ubuf) {
2259 		if (!len) {
2260 			/* nothing to copy, just make ubuf NULL terminated */
2261 			char zero = '\0';
2262 
2263 			if (put_user(zero, ubuf))
2264 				return -EFAULT;
2265 		} else if (input_len >= len + 1) {
2266 			/* ubuf can hold the string with NULL terminator */
2267 			if (copy_to_user(ubuf, buf, len + 1))
2268 				return -EFAULT;
2269 		} else {
2270 			/* ubuf cannot hold the string with NULL terminator,
2271 			 * do a partial copy with NULL terminator.
2272 			 */
2273 			char zero = '\0';
2274 
2275 			err = -ENOSPC;
2276 			if (copy_to_user(ubuf, buf, input_len - 1))
2277 				return -EFAULT;
2278 			if (put_user(zero, ubuf + input_len - 1))
2279 				return -EFAULT;
2280 		}
2281 	}
2282 
2283 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2284 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2285 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2286 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2287 		return -EFAULT;
2288 
2289 	return err;
2290 }
2291 
2292 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2293 
2294 static int bpf_task_fd_query(const union bpf_attr *attr,
2295 			     union bpf_attr __user *uattr)
2296 {
2297 	pid_t pid = attr->task_fd_query.pid;
2298 	u32 fd = attr->task_fd_query.fd;
2299 	const struct perf_event *event;
2300 	struct files_struct *files;
2301 	struct task_struct *task;
2302 	struct file *file;
2303 	int err;
2304 
2305 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2306 		return -EINVAL;
2307 
2308 	if (!capable(CAP_SYS_ADMIN))
2309 		return -EPERM;
2310 
2311 	if (attr->task_fd_query.flags != 0)
2312 		return -EINVAL;
2313 
2314 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2315 	if (!task)
2316 		return -ENOENT;
2317 
2318 	files = get_files_struct(task);
2319 	put_task_struct(task);
2320 	if (!files)
2321 		return -ENOENT;
2322 
2323 	err = 0;
2324 	spin_lock(&files->file_lock);
2325 	file = fcheck_files(files, fd);
2326 	if (!file)
2327 		err = -EBADF;
2328 	else
2329 		get_file(file);
2330 	spin_unlock(&files->file_lock);
2331 	put_files_struct(files);
2332 
2333 	if (err)
2334 		goto out;
2335 
2336 	if (file->f_op == &bpf_raw_tp_fops) {
2337 		struct bpf_raw_tracepoint *raw_tp = file->private_data;
2338 		struct bpf_raw_event_map *btp = raw_tp->btp;
2339 
2340 		err = bpf_task_fd_query_copy(attr, uattr,
2341 					     raw_tp->prog->aux->id,
2342 					     BPF_FD_TYPE_RAW_TRACEPOINT,
2343 					     btp->tp->name, 0, 0);
2344 		goto put_file;
2345 	}
2346 
2347 	event = perf_get_event(file);
2348 	if (!IS_ERR(event)) {
2349 		u64 probe_offset, probe_addr;
2350 		u32 prog_id, fd_type;
2351 		const char *buf;
2352 
2353 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2354 					      &buf, &probe_offset,
2355 					      &probe_addr);
2356 		if (!err)
2357 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2358 						     fd_type, buf,
2359 						     probe_offset,
2360 						     probe_addr);
2361 		goto put_file;
2362 	}
2363 
2364 	err = -ENOTSUPP;
2365 put_file:
2366 	fput(file);
2367 out:
2368 	return err;
2369 }
2370 
2371 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2372 {
2373 	union bpf_attr attr = {};
2374 	int err;
2375 
2376 	if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2377 		return -EPERM;
2378 
2379 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2380 	if (err)
2381 		return err;
2382 	size = min_t(u32, size, sizeof(attr));
2383 
2384 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
2385 	if (copy_from_user(&attr, uattr, size) != 0)
2386 		return -EFAULT;
2387 
2388 	err = security_bpf(cmd, &attr, size);
2389 	if (err < 0)
2390 		return err;
2391 
2392 	switch (cmd) {
2393 	case BPF_MAP_CREATE:
2394 		err = map_create(&attr);
2395 		break;
2396 	case BPF_MAP_LOOKUP_ELEM:
2397 		err = map_lookup_elem(&attr);
2398 		break;
2399 	case BPF_MAP_UPDATE_ELEM:
2400 		err = map_update_elem(&attr);
2401 		break;
2402 	case BPF_MAP_DELETE_ELEM:
2403 		err = map_delete_elem(&attr);
2404 		break;
2405 	case BPF_MAP_GET_NEXT_KEY:
2406 		err = map_get_next_key(&attr);
2407 		break;
2408 	case BPF_PROG_LOAD:
2409 		err = bpf_prog_load(&attr);
2410 		break;
2411 	case BPF_OBJ_PIN:
2412 		err = bpf_obj_pin(&attr);
2413 		break;
2414 	case BPF_OBJ_GET:
2415 		err = bpf_obj_get(&attr);
2416 		break;
2417 	case BPF_PROG_ATTACH:
2418 		err = bpf_prog_attach(&attr);
2419 		break;
2420 	case BPF_PROG_DETACH:
2421 		err = bpf_prog_detach(&attr);
2422 		break;
2423 	case BPF_PROG_QUERY:
2424 		err = bpf_prog_query(&attr, uattr);
2425 		break;
2426 	case BPF_PROG_TEST_RUN:
2427 		err = bpf_prog_test_run(&attr, uattr);
2428 		break;
2429 	case BPF_PROG_GET_NEXT_ID:
2430 		err = bpf_obj_get_next_id(&attr, uattr,
2431 					  &prog_idr, &prog_idr_lock);
2432 		break;
2433 	case BPF_MAP_GET_NEXT_ID:
2434 		err = bpf_obj_get_next_id(&attr, uattr,
2435 					  &map_idr, &map_idr_lock);
2436 		break;
2437 	case BPF_PROG_GET_FD_BY_ID:
2438 		err = bpf_prog_get_fd_by_id(&attr);
2439 		break;
2440 	case BPF_MAP_GET_FD_BY_ID:
2441 		err = bpf_map_get_fd_by_id(&attr);
2442 		break;
2443 	case BPF_OBJ_GET_INFO_BY_FD:
2444 		err = bpf_obj_get_info_by_fd(&attr, uattr);
2445 		break;
2446 	case BPF_RAW_TRACEPOINT_OPEN:
2447 		err = bpf_raw_tracepoint_open(&attr);
2448 		break;
2449 	case BPF_BTF_LOAD:
2450 		err = bpf_btf_load(&attr);
2451 		break;
2452 	case BPF_BTF_GET_FD_BY_ID:
2453 		err = bpf_btf_get_fd_by_id(&attr);
2454 		break;
2455 	case BPF_TASK_FD_QUERY:
2456 		err = bpf_task_fd_query(&attr, uattr);
2457 		break;
2458 	default:
2459 		err = -EINVAL;
2460 		break;
2461 	}
2462 
2463 	return err;
2464 }
2465