xref: /openbmc/linux/kernel/bpf/syscall.c (revision 96de25060d192523fa3c75110dc6348df47fa078)
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 (ptr)
723 			memcpy(value, ptr, value_size);
724 		rcu_read_unlock();
725 		err = ptr ? 0 : -ENOENT;
726 	}
727 
728 	if (err)
729 		goto free_value;
730 
731 	err = -EFAULT;
732 	if (copy_to_user(uvalue, value, value_size) != 0)
733 		goto free_value;
734 
735 	err = 0;
736 
737 free_value:
738 	kfree(value);
739 free_key:
740 	kfree(key);
741 err_put:
742 	fdput(f);
743 	return err;
744 }
745 
746 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
747 
748 static int map_update_elem(union bpf_attr *attr)
749 {
750 	void __user *ukey = u64_to_user_ptr(attr->key);
751 	void __user *uvalue = u64_to_user_ptr(attr->value);
752 	int ufd = attr->map_fd;
753 	struct bpf_map *map;
754 	void *key, *value;
755 	u32 value_size;
756 	struct fd f;
757 	int err;
758 
759 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
760 		return -EINVAL;
761 
762 	f = fdget(ufd);
763 	map = __bpf_map_get(f);
764 	if (IS_ERR(map))
765 		return PTR_ERR(map);
766 
767 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
768 		err = -EPERM;
769 		goto err_put;
770 	}
771 
772 	key = memdup_user(ukey, map->key_size);
773 	if (IS_ERR(key)) {
774 		err = PTR_ERR(key);
775 		goto err_put;
776 	}
777 
778 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
779 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
780 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
781 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
782 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
783 	else
784 		value_size = map->value_size;
785 
786 	err = -ENOMEM;
787 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
788 	if (!value)
789 		goto free_key;
790 
791 	err = -EFAULT;
792 	if (copy_from_user(value, uvalue, value_size) != 0)
793 		goto free_value;
794 
795 	/* Need to create a kthread, thus must support schedule */
796 	if (bpf_map_is_dev_bound(map)) {
797 		err = bpf_map_offload_update_elem(map, key, value, attr->flags);
798 		goto out;
799 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
800 		   map->map_type == BPF_MAP_TYPE_SOCKHASH ||
801 		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
802 		err = map->ops->map_update_elem(map, key, value, attr->flags);
803 		goto out;
804 	}
805 
806 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
807 	 * inside bpf map update or delete otherwise deadlocks are possible
808 	 */
809 	preempt_disable();
810 	__this_cpu_inc(bpf_prog_active);
811 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
812 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
813 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
814 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
815 		err = bpf_percpu_array_update(map, key, value, attr->flags);
816 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
817 		err = bpf_percpu_cgroup_storage_update(map, key, value,
818 						       attr->flags);
819 	} else if (IS_FD_ARRAY(map)) {
820 		rcu_read_lock();
821 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
822 						   attr->flags);
823 		rcu_read_unlock();
824 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
825 		rcu_read_lock();
826 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
827 						  attr->flags);
828 		rcu_read_unlock();
829 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
830 		/* rcu_read_lock() is not needed */
831 		err = bpf_fd_reuseport_array_update_elem(map, key, value,
832 							 attr->flags);
833 	} else {
834 		rcu_read_lock();
835 		err = map->ops->map_update_elem(map, key, value, attr->flags);
836 		rcu_read_unlock();
837 	}
838 	__this_cpu_dec(bpf_prog_active);
839 	preempt_enable();
840 out:
841 free_value:
842 	kfree(value);
843 free_key:
844 	kfree(key);
845 err_put:
846 	fdput(f);
847 	return err;
848 }
849 
850 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
851 
852 static int map_delete_elem(union bpf_attr *attr)
853 {
854 	void __user *ukey = u64_to_user_ptr(attr->key);
855 	int ufd = attr->map_fd;
856 	struct bpf_map *map;
857 	struct fd f;
858 	void *key;
859 	int err;
860 
861 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
862 		return -EINVAL;
863 
864 	f = fdget(ufd);
865 	map = __bpf_map_get(f);
866 	if (IS_ERR(map))
867 		return PTR_ERR(map);
868 
869 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
870 		err = -EPERM;
871 		goto err_put;
872 	}
873 
874 	key = memdup_user(ukey, map->key_size);
875 	if (IS_ERR(key)) {
876 		err = PTR_ERR(key);
877 		goto err_put;
878 	}
879 
880 	if (bpf_map_is_dev_bound(map)) {
881 		err = bpf_map_offload_delete_elem(map, key);
882 		goto out;
883 	}
884 
885 	preempt_disable();
886 	__this_cpu_inc(bpf_prog_active);
887 	rcu_read_lock();
888 	err = map->ops->map_delete_elem(map, key);
889 	rcu_read_unlock();
890 	__this_cpu_dec(bpf_prog_active);
891 	preempt_enable();
892 out:
893 	kfree(key);
894 err_put:
895 	fdput(f);
896 	return err;
897 }
898 
899 /* last field in 'union bpf_attr' used by this command */
900 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
901 
902 static int map_get_next_key(union bpf_attr *attr)
903 {
904 	void __user *ukey = u64_to_user_ptr(attr->key);
905 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
906 	int ufd = attr->map_fd;
907 	struct bpf_map *map;
908 	void *key, *next_key;
909 	struct fd f;
910 	int err;
911 
912 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
913 		return -EINVAL;
914 
915 	f = fdget(ufd);
916 	map = __bpf_map_get(f);
917 	if (IS_ERR(map))
918 		return PTR_ERR(map);
919 
920 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
921 		err = -EPERM;
922 		goto err_put;
923 	}
924 
925 	if (ukey) {
926 		key = memdup_user(ukey, map->key_size);
927 		if (IS_ERR(key)) {
928 			err = PTR_ERR(key);
929 			goto err_put;
930 		}
931 	} else {
932 		key = NULL;
933 	}
934 
935 	err = -ENOMEM;
936 	next_key = kmalloc(map->key_size, GFP_USER);
937 	if (!next_key)
938 		goto free_key;
939 
940 	if (bpf_map_is_dev_bound(map)) {
941 		err = bpf_map_offload_get_next_key(map, key, next_key);
942 		goto out;
943 	}
944 
945 	rcu_read_lock();
946 	err = map->ops->map_get_next_key(map, key, next_key);
947 	rcu_read_unlock();
948 out:
949 	if (err)
950 		goto free_next_key;
951 
952 	err = -EFAULT;
953 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
954 		goto free_next_key;
955 
956 	err = 0;
957 
958 free_next_key:
959 	kfree(next_key);
960 free_key:
961 	kfree(key);
962 err_put:
963 	fdput(f);
964 	return err;
965 }
966 
967 static const struct bpf_prog_ops * const bpf_prog_types[] = {
968 #define BPF_PROG_TYPE(_id, _name) \
969 	[_id] = & _name ## _prog_ops,
970 #define BPF_MAP_TYPE(_id, _ops)
971 #include <linux/bpf_types.h>
972 #undef BPF_PROG_TYPE
973 #undef BPF_MAP_TYPE
974 };
975 
976 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
977 {
978 	const struct bpf_prog_ops *ops;
979 
980 	if (type >= ARRAY_SIZE(bpf_prog_types))
981 		return -EINVAL;
982 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
983 	ops = bpf_prog_types[type];
984 	if (!ops)
985 		return -EINVAL;
986 
987 	if (!bpf_prog_is_dev_bound(prog->aux))
988 		prog->aux->ops = ops;
989 	else
990 		prog->aux->ops = &bpf_offload_prog_ops;
991 	prog->type = type;
992 	return 0;
993 }
994 
995 /* drop refcnt on maps used by eBPF program and free auxilary data */
996 static void free_used_maps(struct bpf_prog_aux *aux)
997 {
998 	enum bpf_cgroup_storage_type stype;
999 	int i;
1000 
1001 	for_each_cgroup_storage_type(stype) {
1002 		if (!aux->cgroup_storage[stype])
1003 			continue;
1004 		bpf_cgroup_storage_release(aux->prog,
1005 					   aux->cgroup_storage[stype]);
1006 	}
1007 
1008 	for (i = 0; i < aux->used_map_cnt; i++)
1009 		bpf_map_put(aux->used_maps[i]);
1010 
1011 	kfree(aux->used_maps);
1012 }
1013 
1014 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1015 {
1016 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1017 	unsigned long user_bufs;
1018 
1019 	if (user) {
1020 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1021 		if (user_bufs > memlock_limit) {
1022 			atomic_long_sub(pages, &user->locked_vm);
1023 			return -EPERM;
1024 		}
1025 	}
1026 
1027 	return 0;
1028 }
1029 
1030 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1031 {
1032 	if (user)
1033 		atomic_long_sub(pages, &user->locked_vm);
1034 }
1035 
1036 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1037 {
1038 	struct user_struct *user = get_current_user();
1039 	int ret;
1040 
1041 	ret = __bpf_prog_charge(user, prog->pages);
1042 	if (ret) {
1043 		free_uid(user);
1044 		return ret;
1045 	}
1046 
1047 	prog->aux->user = user;
1048 	return 0;
1049 }
1050 
1051 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1052 {
1053 	struct user_struct *user = prog->aux->user;
1054 
1055 	__bpf_prog_uncharge(user, prog->pages);
1056 	free_uid(user);
1057 }
1058 
1059 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1060 {
1061 	int id;
1062 
1063 	idr_preload(GFP_KERNEL);
1064 	spin_lock_bh(&prog_idr_lock);
1065 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1066 	if (id > 0)
1067 		prog->aux->id = id;
1068 	spin_unlock_bh(&prog_idr_lock);
1069 	idr_preload_end();
1070 
1071 	/* id is in [1, INT_MAX) */
1072 	if (WARN_ON_ONCE(!id))
1073 		return -ENOSPC;
1074 
1075 	return id > 0 ? 0 : id;
1076 }
1077 
1078 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1079 {
1080 	/* cBPF to eBPF migrations are currently not in the idr store.
1081 	 * Offloaded programs are removed from the store when their device
1082 	 * disappears - even if someone grabs an fd to them they are unusable,
1083 	 * simply waiting for refcnt to drop to be freed.
1084 	 */
1085 	if (!prog->aux->id)
1086 		return;
1087 
1088 	if (do_idr_lock)
1089 		spin_lock_bh(&prog_idr_lock);
1090 	else
1091 		__acquire(&prog_idr_lock);
1092 
1093 	idr_remove(&prog_idr, prog->aux->id);
1094 	prog->aux->id = 0;
1095 
1096 	if (do_idr_lock)
1097 		spin_unlock_bh(&prog_idr_lock);
1098 	else
1099 		__release(&prog_idr_lock);
1100 }
1101 
1102 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1103 {
1104 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1105 
1106 	free_used_maps(aux);
1107 	bpf_prog_uncharge_memlock(aux->prog);
1108 	security_bpf_prog_free(aux);
1109 	bpf_prog_free(aux->prog);
1110 }
1111 
1112 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1113 {
1114 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
1115 		/* bpf_prog_free_id() must be called first */
1116 		bpf_prog_free_id(prog, do_idr_lock);
1117 		bpf_prog_kallsyms_del_all(prog);
1118 
1119 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1120 	}
1121 }
1122 
1123 void bpf_prog_put(struct bpf_prog *prog)
1124 {
1125 	__bpf_prog_put(prog, true);
1126 }
1127 EXPORT_SYMBOL_GPL(bpf_prog_put);
1128 
1129 static int bpf_prog_release(struct inode *inode, struct file *filp)
1130 {
1131 	struct bpf_prog *prog = filp->private_data;
1132 
1133 	bpf_prog_put(prog);
1134 	return 0;
1135 }
1136 
1137 #ifdef CONFIG_PROC_FS
1138 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1139 {
1140 	const struct bpf_prog *prog = filp->private_data;
1141 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1142 
1143 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1144 	seq_printf(m,
1145 		   "prog_type:\t%u\n"
1146 		   "prog_jited:\t%u\n"
1147 		   "prog_tag:\t%s\n"
1148 		   "memlock:\t%llu\n"
1149 		   "prog_id:\t%u\n",
1150 		   prog->type,
1151 		   prog->jited,
1152 		   prog_tag,
1153 		   prog->pages * 1ULL << PAGE_SHIFT,
1154 		   prog->aux->id);
1155 }
1156 #endif
1157 
1158 const struct file_operations bpf_prog_fops = {
1159 #ifdef CONFIG_PROC_FS
1160 	.show_fdinfo	= bpf_prog_show_fdinfo,
1161 #endif
1162 	.release	= bpf_prog_release,
1163 	.read		= bpf_dummy_read,
1164 	.write		= bpf_dummy_write,
1165 };
1166 
1167 int bpf_prog_new_fd(struct bpf_prog *prog)
1168 {
1169 	int ret;
1170 
1171 	ret = security_bpf_prog(prog);
1172 	if (ret < 0)
1173 		return ret;
1174 
1175 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1176 				O_RDWR | O_CLOEXEC);
1177 }
1178 
1179 static struct bpf_prog *____bpf_prog_get(struct fd f)
1180 {
1181 	if (!f.file)
1182 		return ERR_PTR(-EBADF);
1183 	if (f.file->f_op != &bpf_prog_fops) {
1184 		fdput(f);
1185 		return ERR_PTR(-EINVAL);
1186 	}
1187 
1188 	return f.file->private_data;
1189 }
1190 
1191 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1192 {
1193 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1194 		atomic_sub(i, &prog->aux->refcnt);
1195 		return ERR_PTR(-EBUSY);
1196 	}
1197 	return prog;
1198 }
1199 EXPORT_SYMBOL_GPL(bpf_prog_add);
1200 
1201 void bpf_prog_sub(struct bpf_prog *prog, int i)
1202 {
1203 	/* Only to be used for undoing previous bpf_prog_add() in some
1204 	 * error path. We still know that another entity in our call
1205 	 * path holds a reference to the program, thus atomic_sub() can
1206 	 * be safely used in such cases!
1207 	 */
1208 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1209 }
1210 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1211 
1212 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1213 {
1214 	return bpf_prog_add(prog, 1);
1215 }
1216 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1217 
1218 /* prog_idr_lock should have been held */
1219 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1220 {
1221 	int refold;
1222 
1223 	refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1224 
1225 	if (refold >= BPF_MAX_REFCNT) {
1226 		__bpf_prog_put(prog, false);
1227 		return ERR_PTR(-EBUSY);
1228 	}
1229 
1230 	if (!refold)
1231 		return ERR_PTR(-ENOENT);
1232 
1233 	return prog;
1234 }
1235 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1236 
1237 bool bpf_prog_get_ok(struct bpf_prog *prog,
1238 			    enum bpf_prog_type *attach_type, bool attach_drv)
1239 {
1240 	/* not an attachment, just a refcount inc, always allow */
1241 	if (!attach_type)
1242 		return true;
1243 
1244 	if (prog->type != *attach_type)
1245 		return false;
1246 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1247 		return false;
1248 
1249 	return true;
1250 }
1251 
1252 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1253 				       bool attach_drv)
1254 {
1255 	struct fd f = fdget(ufd);
1256 	struct bpf_prog *prog;
1257 
1258 	prog = ____bpf_prog_get(f);
1259 	if (IS_ERR(prog))
1260 		return prog;
1261 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1262 		prog = ERR_PTR(-EINVAL);
1263 		goto out;
1264 	}
1265 
1266 	prog = bpf_prog_inc(prog);
1267 out:
1268 	fdput(f);
1269 	return prog;
1270 }
1271 
1272 struct bpf_prog *bpf_prog_get(u32 ufd)
1273 {
1274 	return __bpf_prog_get(ufd, NULL, false);
1275 }
1276 
1277 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1278 				       bool attach_drv)
1279 {
1280 	return __bpf_prog_get(ufd, &type, attach_drv);
1281 }
1282 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1283 
1284 /* Initially all BPF programs could be loaded w/o specifying
1285  * expected_attach_type. Later for some of them specifying expected_attach_type
1286  * at load time became required so that program could be validated properly.
1287  * Programs of types that are allowed to be loaded both w/ and w/o (for
1288  * backward compatibility) expected_attach_type, should have the default attach
1289  * type assigned to expected_attach_type for the latter case, so that it can be
1290  * validated later at attach time.
1291  *
1292  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1293  * prog type requires it but has some attach types that have to be backward
1294  * compatible.
1295  */
1296 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1297 {
1298 	switch (attr->prog_type) {
1299 	case BPF_PROG_TYPE_CGROUP_SOCK:
1300 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1301 		 * exist so checking for non-zero is the way to go here.
1302 		 */
1303 		if (!attr->expected_attach_type)
1304 			attr->expected_attach_type =
1305 				BPF_CGROUP_INET_SOCK_CREATE;
1306 		break;
1307 	}
1308 }
1309 
1310 static int
1311 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1312 				enum bpf_attach_type expected_attach_type)
1313 {
1314 	switch (prog_type) {
1315 	case BPF_PROG_TYPE_CGROUP_SOCK:
1316 		switch (expected_attach_type) {
1317 		case BPF_CGROUP_INET_SOCK_CREATE:
1318 		case BPF_CGROUP_INET4_POST_BIND:
1319 		case BPF_CGROUP_INET6_POST_BIND:
1320 			return 0;
1321 		default:
1322 			return -EINVAL;
1323 		}
1324 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1325 		switch (expected_attach_type) {
1326 		case BPF_CGROUP_INET4_BIND:
1327 		case BPF_CGROUP_INET6_BIND:
1328 		case BPF_CGROUP_INET4_CONNECT:
1329 		case BPF_CGROUP_INET6_CONNECT:
1330 		case BPF_CGROUP_UDP4_SENDMSG:
1331 		case BPF_CGROUP_UDP6_SENDMSG:
1332 			return 0;
1333 		default:
1334 			return -EINVAL;
1335 		}
1336 	default:
1337 		return 0;
1338 	}
1339 }
1340 
1341 /* last field in 'union bpf_attr' used by this command */
1342 #define	BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1343 
1344 static int bpf_prog_load(union bpf_attr *attr)
1345 {
1346 	enum bpf_prog_type type = attr->prog_type;
1347 	struct bpf_prog *prog;
1348 	int err;
1349 	char license[128];
1350 	bool is_gpl;
1351 
1352 	if (CHECK_ATTR(BPF_PROG_LOAD))
1353 		return -EINVAL;
1354 
1355 	if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1356 		return -EINVAL;
1357 
1358 	/* copy eBPF program license from user space */
1359 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1360 			      sizeof(license) - 1) < 0)
1361 		return -EFAULT;
1362 	license[sizeof(license) - 1] = 0;
1363 
1364 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1365 	is_gpl = license_is_gpl_compatible(license);
1366 
1367 	if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1368 		return -E2BIG;
1369 
1370 	if (type == BPF_PROG_TYPE_KPROBE &&
1371 	    attr->kern_version != LINUX_VERSION_CODE)
1372 		return -EINVAL;
1373 
1374 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1375 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
1376 	    !capable(CAP_SYS_ADMIN))
1377 		return -EPERM;
1378 
1379 	bpf_prog_load_fixup_attach_type(attr);
1380 	if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1381 		return -EINVAL;
1382 
1383 	/* plain bpf_prog allocation */
1384 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1385 	if (!prog)
1386 		return -ENOMEM;
1387 
1388 	prog->expected_attach_type = attr->expected_attach_type;
1389 
1390 	prog->aux->offload_requested = !!attr->prog_ifindex;
1391 
1392 	err = security_bpf_prog_alloc(prog->aux);
1393 	if (err)
1394 		goto free_prog_nouncharge;
1395 
1396 	err = bpf_prog_charge_memlock(prog);
1397 	if (err)
1398 		goto free_prog_sec;
1399 
1400 	prog->len = attr->insn_cnt;
1401 
1402 	err = -EFAULT;
1403 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1404 			   bpf_prog_insn_size(prog)) != 0)
1405 		goto free_prog;
1406 
1407 	prog->orig_prog = NULL;
1408 	prog->jited = 0;
1409 
1410 	atomic_set(&prog->aux->refcnt, 1);
1411 	prog->gpl_compatible = is_gpl ? 1 : 0;
1412 
1413 	if (bpf_prog_is_dev_bound(prog->aux)) {
1414 		err = bpf_prog_offload_init(prog, attr);
1415 		if (err)
1416 			goto free_prog;
1417 	}
1418 
1419 	/* find program type: socket_filter vs tracing_filter */
1420 	err = find_prog_type(type, prog);
1421 	if (err < 0)
1422 		goto free_prog;
1423 
1424 	prog->aux->load_time = ktime_get_boot_ns();
1425 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1426 	if (err)
1427 		goto free_prog;
1428 
1429 	/* run eBPF verifier */
1430 	err = bpf_check(&prog, attr);
1431 	if (err < 0)
1432 		goto free_used_maps;
1433 
1434 	prog = bpf_prog_select_runtime(prog, &err);
1435 	if (err < 0)
1436 		goto free_used_maps;
1437 
1438 	err = bpf_prog_alloc_id(prog);
1439 	if (err)
1440 		goto free_used_maps;
1441 
1442 	err = bpf_prog_new_fd(prog);
1443 	if (err < 0) {
1444 		/* failed to allocate fd.
1445 		 * bpf_prog_put() is needed because the above
1446 		 * bpf_prog_alloc_id() has published the prog
1447 		 * to the userspace and the userspace may
1448 		 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1449 		 */
1450 		bpf_prog_put(prog);
1451 		return err;
1452 	}
1453 
1454 	bpf_prog_kallsyms_add(prog);
1455 	return err;
1456 
1457 free_used_maps:
1458 	bpf_prog_kallsyms_del_subprogs(prog);
1459 	free_used_maps(prog->aux);
1460 free_prog:
1461 	bpf_prog_uncharge_memlock(prog);
1462 free_prog_sec:
1463 	security_bpf_prog_free(prog->aux);
1464 free_prog_nouncharge:
1465 	bpf_prog_free(prog);
1466 	return err;
1467 }
1468 
1469 #define BPF_OBJ_LAST_FIELD file_flags
1470 
1471 static int bpf_obj_pin(const union bpf_attr *attr)
1472 {
1473 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1474 		return -EINVAL;
1475 
1476 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1477 }
1478 
1479 static int bpf_obj_get(const union bpf_attr *attr)
1480 {
1481 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1482 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1483 		return -EINVAL;
1484 
1485 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1486 				attr->file_flags);
1487 }
1488 
1489 struct bpf_raw_tracepoint {
1490 	struct bpf_raw_event_map *btp;
1491 	struct bpf_prog *prog;
1492 };
1493 
1494 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1495 {
1496 	struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1497 
1498 	if (raw_tp->prog) {
1499 		bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1500 		bpf_prog_put(raw_tp->prog);
1501 	}
1502 	kfree(raw_tp);
1503 	return 0;
1504 }
1505 
1506 static const struct file_operations bpf_raw_tp_fops = {
1507 	.release	= bpf_raw_tracepoint_release,
1508 	.read		= bpf_dummy_read,
1509 	.write		= bpf_dummy_write,
1510 };
1511 
1512 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1513 
1514 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1515 {
1516 	struct bpf_raw_tracepoint *raw_tp;
1517 	struct bpf_raw_event_map *btp;
1518 	struct bpf_prog *prog;
1519 	char tp_name[128];
1520 	int tp_fd, err;
1521 
1522 	if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1523 			      sizeof(tp_name) - 1) < 0)
1524 		return -EFAULT;
1525 	tp_name[sizeof(tp_name) - 1] = 0;
1526 
1527 	btp = bpf_find_raw_tracepoint(tp_name);
1528 	if (!btp)
1529 		return -ENOENT;
1530 
1531 	raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1532 	if (!raw_tp)
1533 		return -ENOMEM;
1534 	raw_tp->btp = btp;
1535 
1536 	prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1537 				 BPF_PROG_TYPE_RAW_TRACEPOINT);
1538 	if (IS_ERR(prog)) {
1539 		err = PTR_ERR(prog);
1540 		goto out_free_tp;
1541 	}
1542 
1543 	err = bpf_probe_register(raw_tp->btp, prog);
1544 	if (err)
1545 		goto out_put_prog;
1546 
1547 	raw_tp->prog = prog;
1548 	tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1549 				 O_CLOEXEC);
1550 	if (tp_fd < 0) {
1551 		bpf_probe_unregister(raw_tp->btp, prog);
1552 		err = tp_fd;
1553 		goto out_put_prog;
1554 	}
1555 	return tp_fd;
1556 
1557 out_put_prog:
1558 	bpf_prog_put(prog);
1559 out_free_tp:
1560 	kfree(raw_tp);
1561 	return err;
1562 }
1563 
1564 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1565 					     enum bpf_attach_type attach_type)
1566 {
1567 	switch (prog->type) {
1568 	case BPF_PROG_TYPE_CGROUP_SOCK:
1569 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1570 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1571 	default:
1572 		return 0;
1573 	}
1574 }
1575 
1576 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1577 
1578 #define BPF_F_ATTACH_MASK \
1579 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1580 
1581 static int bpf_prog_attach(const union bpf_attr *attr)
1582 {
1583 	enum bpf_prog_type ptype;
1584 	struct bpf_prog *prog;
1585 	int ret;
1586 
1587 	if (!capable(CAP_NET_ADMIN))
1588 		return -EPERM;
1589 
1590 	if (CHECK_ATTR(BPF_PROG_ATTACH))
1591 		return -EINVAL;
1592 
1593 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1594 		return -EINVAL;
1595 
1596 	switch (attr->attach_type) {
1597 	case BPF_CGROUP_INET_INGRESS:
1598 	case BPF_CGROUP_INET_EGRESS:
1599 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1600 		break;
1601 	case BPF_CGROUP_INET_SOCK_CREATE:
1602 	case BPF_CGROUP_INET4_POST_BIND:
1603 	case BPF_CGROUP_INET6_POST_BIND:
1604 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1605 		break;
1606 	case BPF_CGROUP_INET4_BIND:
1607 	case BPF_CGROUP_INET6_BIND:
1608 	case BPF_CGROUP_INET4_CONNECT:
1609 	case BPF_CGROUP_INET6_CONNECT:
1610 	case BPF_CGROUP_UDP4_SENDMSG:
1611 	case BPF_CGROUP_UDP6_SENDMSG:
1612 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1613 		break;
1614 	case BPF_CGROUP_SOCK_OPS:
1615 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1616 		break;
1617 	case BPF_CGROUP_DEVICE:
1618 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1619 		break;
1620 	case BPF_SK_MSG_VERDICT:
1621 		ptype = BPF_PROG_TYPE_SK_MSG;
1622 		break;
1623 	case BPF_SK_SKB_STREAM_PARSER:
1624 	case BPF_SK_SKB_STREAM_VERDICT:
1625 		ptype = BPF_PROG_TYPE_SK_SKB;
1626 		break;
1627 	case BPF_LIRC_MODE2:
1628 		ptype = BPF_PROG_TYPE_LIRC_MODE2;
1629 		break;
1630 	case BPF_FLOW_DISSECTOR:
1631 		ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1632 		break;
1633 	default:
1634 		return -EINVAL;
1635 	}
1636 
1637 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1638 	if (IS_ERR(prog))
1639 		return PTR_ERR(prog);
1640 
1641 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1642 		bpf_prog_put(prog);
1643 		return -EINVAL;
1644 	}
1645 
1646 	switch (ptype) {
1647 	case BPF_PROG_TYPE_SK_SKB:
1648 	case BPF_PROG_TYPE_SK_MSG:
1649 		ret = sockmap_get_from_fd(attr, ptype, prog);
1650 		break;
1651 	case BPF_PROG_TYPE_LIRC_MODE2:
1652 		ret = lirc_prog_attach(attr, prog);
1653 		break;
1654 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
1655 		ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1656 		break;
1657 	default:
1658 		ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1659 	}
1660 
1661 	if (ret)
1662 		bpf_prog_put(prog);
1663 	return ret;
1664 }
1665 
1666 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1667 
1668 static int bpf_prog_detach(const union bpf_attr *attr)
1669 {
1670 	enum bpf_prog_type ptype;
1671 
1672 	if (!capable(CAP_NET_ADMIN))
1673 		return -EPERM;
1674 
1675 	if (CHECK_ATTR(BPF_PROG_DETACH))
1676 		return -EINVAL;
1677 
1678 	switch (attr->attach_type) {
1679 	case BPF_CGROUP_INET_INGRESS:
1680 	case BPF_CGROUP_INET_EGRESS:
1681 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1682 		break;
1683 	case BPF_CGROUP_INET_SOCK_CREATE:
1684 	case BPF_CGROUP_INET4_POST_BIND:
1685 	case BPF_CGROUP_INET6_POST_BIND:
1686 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1687 		break;
1688 	case BPF_CGROUP_INET4_BIND:
1689 	case BPF_CGROUP_INET6_BIND:
1690 	case BPF_CGROUP_INET4_CONNECT:
1691 	case BPF_CGROUP_INET6_CONNECT:
1692 	case BPF_CGROUP_UDP4_SENDMSG:
1693 	case BPF_CGROUP_UDP6_SENDMSG:
1694 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1695 		break;
1696 	case BPF_CGROUP_SOCK_OPS:
1697 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1698 		break;
1699 	case BPF_CGROUP_DEVICE:
1700 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1701 		break;
1702 	case BPF_SK_MSG_VERDICT:
1703 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, NULL);
1704 	case BPF_SK_SKB_STREAM_PARSER:
1705 	case BPF_SK_SKB_STREAM_VERDICT:
1706 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, NULL);
1707 	case BPF_LIRC_MODE2:
1708 		return lirc_prog_detach(attr);
1709 	case BPF_FLOW_DISSECTOR:
1710 		return skb_flow_dissector_bpf_prog_detach(attr);
1711 	default:
1712 		return -EINVAL;
1713 	}
1714 
1715 	return cgroup_bpf_prog_detach(attr, ptype);
1716 }
1717 
1718 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1719 
1720 static int bpf_prog_query(const union bpf_attr *attr,
1721 			  union bpf_attr __user *uattr)
1722 {
1723 	if (!capable(CAP_NET_ADMIN))
1724 		return -EPERM;
1725 	if (CHECK_ATTR(BPF_PROG_QUERY))
1726 		return -EINVAL;
1727 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1728 		return -EINVAL;
1729 
1730 	switch (attr->query.attach_type) {
1731 	case BPF_CGROUP_INET_INGRESS:
1732 	case BPF_CGROUP_INET_EGRESS:
1733 	case BPF_CGROUP_INET_SOCK_CREATE:
1734 	case BPF_CGROUP_INET4_BIND:
1735 	case BPF_CGROUP_INET6_BIND:
1736 	case BPF_CGROUP_INET4_POST_BIND:
1737 	case BPF_CGROUP_INET6_POST_BIND:
1738 	case BPF_CGROUP_INET4_CONNECT:
1739 	case BPF_CGROUP_INET6_CONNECT:
1740 	case BPF_CGROUP_UDP4_SENDMSG:
1741 	case BPF_CGROUP_UDP6_SENDMSG:
1742 	case BPF_CGROUP_SOCK_OPS:
1743 	case BPF_CGROUP_DEVICE:
1744 		break;
1745 	case BPF_LIRC_MODE2:
1746 		return lirc_prog_query(attr, uattr);
1747 	default:
1748 		return -EINVAL;
1749 	}
1750 
1751 	return cgroup_bpf_prog_query(attr, uattr);
1752 }
1753 
1754 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1755 
1756 static int bpf_prog_test_run(const union bpf_attr *attr,
1757 			     union bpf_attr __user *uattr)
1758 {
1759 	struct bpf_prog *prog;
1760 	int ret = -ENOTSUPP;
1761 
1762 	if (!capable(CAP_SYS_ADMIN))
1763 		return -EPERM;
1764 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1765 		return -EINVAL;
1766 
1767 	prog = bpf_prog_get(attr->test.prog_fd);
1768 	if (IS_ERR(prog))
1769 		return PTR_ERR(prog);
1770 
1771 	if (prog->aux->ops->test_run)
1772 		ret = prog->aux->ops->test_run(prog, attr, uattr);
1773 
1774 	bpf_prog_put(prog);
1775 	return ret;
1776 }
1777 
1778 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1779 
1780 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1781 			       union bpf_attr __user *uattr,
1782 			       struct idr *idr,
1783 			       spinlock_t *lock)
1784 {
1785 	u32 next_id = attr->start_id;
1786 	int err = 0;
1787 
1788 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1789 		return -EINVAL;
1790 
1791 	if (!capable(CAP_SYS_ADMIN))
1792 		return -EPERM;
1793 
1794 	next_id++;
1795 	spin_lock_bh(lock);
1796 	if (!idr_get_next(idr, &next_id))
1797 		err = -ENOENT;
1798 	spin_unlock_bh(lock);
1799 
1800 	if (!err)
1801 		err = put_user(next_id, &uattr->next_id);
1802 
1803 	return err;
1804 }
1805 
1806 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1807 
1808 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1809 {
1810 	struct bpf_prog *prog;
1811 	u32 id = attr->prog_id;
1812 	int fd;
1813 
1814 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1815 		return -EINVAL;
1816 
1817 	if (!capable(CAP_SYS_ADMIN))
1818 		return -EPERM;
1819 
1820 	spin_lock_bh(&prog_idr_lock);
1821 	prog = idr_find(&prog_idr, id);
1822 	if (prog)
1823 		prog = bpf_prog_inc_not_zero(prog);
1824 	else
1825 		prog = ERR_PTR(-ENOENT);
1826 	spin_unlock_bh(&prog_idr_lock);
1827 
1828 	if (IS_ERR(prog))
1829 		return PTR_ERR(prog);
1830 
1831 	fd = bpf_prog_new_fd(prog);
1832 	if (fd < 0)
1833 		bpf_prog_put(prog);
1834 
1835 	return fd;
1836 }
1837 
1838 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1839 
1840 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1841 {
1842 	struct bpf_map *map;
1843 	u32 id = attr->map_id;
1844 	int f_flags;
1845 	int fd;
1846 
1847 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1848 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1849 		return -EINVAL;
1850 
1851 	if (!capable(CAP_SYS_ADMIN))
1852 		return -EPERM;
1853 
1854 	f_flags = bpf_get_file_flag(attr->open_flags);
1855 	if (f_flags < 0)
1856 		return f_flags;
1857 
1858 	spin_lock_bh(&map_idr_lock);
1859 	map = idr_find(&map_idr, id);
1860 	if (map)
1861 		map = bpf_map_inc_not_zero(map, true);
1862 	else
1863 		map = ERR_PTR(-ENOENT);
1864 	spin_unlock_bh(&map_idr_lock);
1865 
1866 	if (IS_ERR(map))
1867 		return PTR_ERR(map);
1868 
1869 	fd = bpf_map_new_fd(map, f_flags);
1870 	if (fd < 0)
1871 		bpf_map_put(map);
1872 
1873 	return fd;
1874 }
1875 
1876 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1877 					      unsigned long addr)
1878 {
1879 	int i;
1880 
1881 	for (i = 0; i < prog->aux->used_map_cnt; i++)
1882 		if (prog->aux->used_maps[i] == (void *)addr)
1883 			return prog->aux->used_maps[i];
1884 	return NULL;
1885 }
1886 
1887 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1888 {
1889 	const struct bpf_map *map;
1890 	struct bpf_insn *insns;
1891 	u64 imm;
1892 	int i;
1893 
1894 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1895 			GFP_USER);
1896 	if (!insns)
1897 		return insns;
1898 
1899 	for (i = 0; i < prog->len; i++) {
1900 		if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1901 			insns[i].code = BPF_JMP | BPF_CALL;
1902 			insns[i].imm = BPF_FUNC_tail_call;
1903 			/* fall-through */
1904 		}
1905 		if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1906 		    insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1907 			if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1908 				insns[i].code = BPF_JMP | BPF_CALL;
1909 			if (!bpf_dump_raw_ok())
1910 				insns[i].imm = 0;
1911 			continue;
1912 		}
1913 
1914 		if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1915 			continue;
1916 
1917 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1918 		map = bpf_map_from_imm(prog, imm);
1919 		if (map) {
1920 			insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1921 			insns[i].imm = map->id;
1922 			insns[i + 1].imm = 0;
1923 			continue;
1924 		}
1925 
1926 		if (!bpf_dump_raw_ok() &&
1927 		    imm == (unsigned long)prog->aux) {
1928 			insns[i].imm = 0;
1929 			insns[i + 1].imm = 0;
1930 			continue;
1931 		}
1932 	}
1933 
1934 	return insns;
1935 }
1936 
1937 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1938 				   const union bpf_attr *attr,
1939 				   union bpf_attr __user *uattr)
1940 {
1941 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1942 	struct bpf_prog_info info = {};
1943 	u32 info_len = attr->info.info_len;
1944 	char __user *uinsns;
1945 	u32 ulen;
1946 	int err;
1947 
1948 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1949 	if (err)
1950 		return err;
1951 	info_len = min_t(u32, sizeof(info), info_len);
1952 
1953 	if (copy_from_user(&info, uinfo, info_len))
1954 		return -EFAULT;
1955 
1956 	info.type = prog->type;
1957 	info.id = prog->aux->id;
1958 	info.load_time = prog->aux->load_time;
1959 	info.created_by_uid = from_kuid_munged(current_user_ns(),
1960 					       prog->aux->user->uid);
1961 	info.gpl_compatible = prog->gpl_compatible;
1962 
1963 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
1964 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1965 
1966 	ulen = info.nr_map_ids;
1967 	info.nr_map_ids = prog->aux->used_map_cnt;
1968 	ulen = min_t(u32, info.nr_map_ids, ulen);
1969 	if (ulen) {
1970 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1971 		u32 i;
1972 
1973 		for (i = 0; i < ulen; i++)
1974 			if (put_user(prog->aux->used_maps[i]->id,
1975 				     &user_map_ids[i]))
1976 				return -EFAULT;
1977 	}
1978 
1979 	if (!capable(CAP_SYS_ADMIN)) {
1980 		info.jited_prog_len = 0;
1981 		info.xlated_prog_len = 0;
1982 		info.nr_jited_ksyms = 0;
1983 		goto done;
1984 	}
1985 
1986 	ulen = info.xlated_prog_len;
1987 	info.xlated_prog_len = bpf_prog_insn_size(prog);
1988 	if (info.xlated_prog_len && ulen) {
1989 		struct bpf_insn *insns_sanitized;
1990 		bool fault;
1991 
1992 		if (prog->blinded && !bpf_dump_raw_ok()) {
1993 			info.xlated_prog_insns = 0;
1994 			goto done;
1995 		}
1996 		insns_sanitized = bpf_insn_prepare_dump(prog);
1997 		if (!insns_sanitized)
1998 			return -ENOMEM;
1999 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2000 		ulen = min_t(u32, info.xlated_prog_len, ulen);
2001 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
2002 		kfree(insns_sanitized);
2003 		if (fault)
2004 			return -EFAULT;
2005 	}
2006 
2007 	if (bpf_prog_is_dev_bound(prog->aux)) {
2008 		err = bpf_prog_offload_info_fill(&info, prog);
2009 		if (err)
2010 			return err;
2011 		goto done;
2012 	}
2013 
2014 	/* NOTE: the following code is supposed to be skipped for offload.
2015 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
2016 	 * for offload.
2017 	 */
2018 	ulen = info.jited_prog_len;
2019 	if (prog->aux->func_cnt) {
2020 		u32 i;
2021 
2022 		info.jited_prog_len = 0;
2023 		for (i = 0; i < prog->aux->func_cnt; i++)
2024 			info.jited_prog_len += prog->aux->func[i]->jited_len;
2025 	} else {
2026 		info.jited_prog_len = prog->jited_len;
2027 	}
2028 
2029 	if (info.jited_prog_len && ulen) {
2030 		if (bpf_dump_raw_ok()) {
2031 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
2032 			ulen = min_t(u32, info.jited_prog_len, ulen);
2033 
2034 			/* for multi-function programs, copy the JITed
2035 			 * instructions for all the functions
2036 			 */
2037 			if (prog->aux->func_cnt) {
2038 				u32 len, free, i;
2039 				u8 *img;
2040 
2041 				free = ulen;
2042 				for (i = 0; i < prog->aux->func_cnt; i++) {
2043 					len = prog->aux->func[i]->jited_len;
2044 					len = min_t(u32, len, free);
2045 					img = (u8 *) prog->aux->func[i]->bpf_func;
2046 					if (copy_to_user(uinsns, img, len))
2047 						return -EFAULT;
2048 					uinsns += len;
2049 					free -= len;
2050 					if (!free)
2051 						break;
2052 				}
2053 			} else {
2054 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
2055 					return -EFAULT;
2056 			}
2057 		} else {
2058 			info.jited_prog_insns = 0;
2059 		}
2060 	}
2061 
2062 	ulen = info.nr_jited_ksyms;
2063 	info.nr_jited_ksyms = prog->aux->func_cnt;
2064 	if (info.nr_jited_ksyms && ulen) {
2065 		if (bpf_dump_raw_ok()) {
2066 			u64 __user *user_ksyms;
2067 			ulong ksym_addr;
2068 			u32 i;
2069 
2070 			/* copy the address of the kernel symbol
2071 			 * corresponding to each function
2072 			 */
2073 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2074 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2075 			for (i = 0; i < ulen; i++) {
2076 				ksym_addr = (ulong) prog->aux->func[i]->bpf_func;
2077 				ksym_addr &= PAGE_MASK;
2078 				if (put_user((u64) ksym_addr, &user_ksyms[i]))
2079 					return -EFAULT;
2080 			}
2081 		} else {
2082 			info.jited_ksyms = 0;
2083 		}
2084 	}
2085 
2086 	ulen = info.nr_jited_func_lens;
2087 	info.nr_jited_func_lens = prog->aux->func_cnt;
2088 	if (info.nr_jited_func_lens && ulen) {
2089 		if (bpf_dump_raw_ok()) {
2090 			u32 __user *user_lens;
2091 			u32 func_len, i;
2092 
2093 			/* copy the JITed image lengths for each function */
2094 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2095 			user_lens = u64_to_user_ptr(info.jited_func_lens);
2096 			for (i = 0; i < ulen; i++) {
2097 				func_len = prog->aux->func[i]->jited_len;
2098 				if (put_user(func_len, &user_lens[i]))
2099 					return -EFAULT;
2100 			}
2101 		} else {
2102 			info.jited_func_lens = 0;
2103 		}
2104 	}
2105 
2106 done:
2107 	if (copy_to_user(uinfo, &info, info_len) ||
2108 	    put_user(info_len, &uattr->info.info_len))
2109 		return -EFAULT;
2110 
2111 	return 0;
2112 }
2113 
2114 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2115 				  const union bpf_attr *attr,
2116 				  union bpf_attr __user *uattr)
2117 {
2118 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2119 	struct bpf_map_info info = {};
2120 	u32 info_len = attr->info.info_len;
2121 	int err;
2122 
2123 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2124 	if (err)
2125 		return err;
2126 	info_len = min_t(u32, sizeof(info), info_len);
2127 
2128 	info.type = map->map_type;
2129 	info.id = map->id;
2130 	info.key_size = map->key_size;
2131 	info.value_size = map->value_size;
2132 	info.max_entries = map->max_entries;
2133 	info.map_flags = map->map_flags;
2134 	memcpy(info.name, map->name, sizeof(map->name));
2135 
2136 	if (map->btf) {
2137 		info.btf_id = btf_id(map->btf);
2138 		info.btf_key_type_id = map->btf_key_type_id;
2139 		info.btf_value_type_id = map->btf_value_type_id;
2140 	}
2141 
2142 	if (bpf_map_is_dev_bound(map)) {
2143 		err = bpf_map_offload_info_fill(&info, map);
2144 		if (err)
2145 			return err;
2146 	}
2147 
2148 	if (copy_to_user(uinfo, &info, info_len) ||
2149 	    put_user(info_len, &uattr->info.info_len))
2150 		return -EFAULT;
2151 
2152 	return 0;
2153 }
2154 
2155 static int bpf_btf_get_info_by_fd(struct btf *btf,
2156 				  const union bpf_attr *attr,
2157 				  union bpf_attr __user *uattr)
2158 {
2159 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2160 	u32 info_len = attr->info.info_len;
2161 	int err;
2162 
2163 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2164 	if (err)
2165 		return err;
2166 
2167 	return btf_get_info_by_fd(btf, attr, uattr);
2168 }
2169 
2170 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2171 
2172 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2173 				  union bpf_attr __user *uattr)
2174 {
2175 	int ufd = attr->info.bpf_fd;
2176 	struct fd f;
2177 	int err;
2178 
2179 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2180 		return -EINVAL;
2181 
2182 	f = fdget(ufd);
2183 	if (!f.file)
2184 		return -EBADFD;
2185 
2186 	if (f.file->f_op == &bpf_prog_fops)
2187 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2188 					      uattr);
2189 	else if (f.file->f_op == &bpf_map_fops)
2190 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2191 					     uattr);
2192 	else if (f.file->f_op == &btf_fops)
2193 		err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2194 	else
2195 		err = -EINVAL;
2196 
2197 	fdput(f);
2198 	return err;
2199 }
2200 
2201 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2202 
2203 static int bpf_btf_load(const union bpf_attr *attr)
2204 {
2205 	if (CHECK_ATTR(BPF_BTF_LOAD))
2206 		return -EINVAL;
2207 
2208 	if (!capable(CAP_SYS_ADMIN))
2209 		return -EPERM;
2210 
2211 	return btf_new_fd(attr);
2212 }
2213 
2214 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2215 
2216 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2217 {
2218 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2219 		return -EINVAL;
2220 
2221 	if (!capable(CAP_SYS_ADMIN))
2222 		return -EPERM;
2223 
2224 	return btf_get_fd_by_id(attr->btf_id);
2225 }
2226 
2227 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2228 				    union bpf_attr __user *uattr,
2229 				    u32 prog_id, u32 fd_type,
2230 				    const char *buf, u64 probe_offset,
2231 				    u64 probe_addr)
2232 {
2233 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2234 	u32 len = buf ? strlen(buf) : 0, input_len;
2235 	int err = 0;
2236 
2237 	if (put_user(len, &uattr->task_fd_query.buf_len))
2238 		return -EFAULT;
2239 	input_len = attr->task_fd_query.buf_len;
2240 	if (input_len && ubuf) {
2241 		if (!len) {
2242 			/* nothing to copy, just make ubuf NULL terminated */
2243 			char zero = '\0';
2244 
2245 			if (put_user(zero, ubuf))
2246 				return -EFAULT;
2247 		} else if (input_len >= len + 1) {
2248 			/* ubuf can hold the string with NULL terminator */
2249 			if (copy_to_user(ubuf, buf, len + 1))
2250 				return -EFAULT;
2251 		} else {
2252 			/* ubuf cannot hold the string with NULL terminator,
2253 			 * do a partial copy with NULL terminator.
2254 			 */
2255 			char zero = '\0';
2256 
2257 			err = -ENOSPC;
2258 			if (copy_to_user(ubuf, buf, input_len - 1))
2259 				return -EFAULT;
2260 			if (put_user(zero, ubuf + input_len - 1))
2261 				return -EFAULT;
2262 		}
2263 	}
2264 
2265 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2266 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2267 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2268 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2269 		return -EFAULT;
2270 
2271 	return err;
2272 }
2273 
2274 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2275 
2276 static int bpf_task_fd_query(const union bpf_attr *attr,
2277 			     union bpf_attr __user *uattr)
2278 {
2279 	pid_t pid = attr->task_fd_query.pid;
2280 	u32 fd = attr->task_fd_query.fd;
2281 	const struct perf_event *event;
2282 	struct files_struct *files;
2283 	struct task_struct *task;
2284 	struct file *file;
2285 	int err;
2286 
2287 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2288 		return -EINVAL;
2289 
2290 	if (!capable(CAP_SYS_ADMIN))
2291 		return -EPERM;
2292 
2293 	if (attr->task_fd_query.flags != 0)
2294 		return -EINVAL;
2295 
2296 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2297 	if (!task)
2298 		return -ENOENT;
2299 
2300 	files = get_files_struct(task);
2301 	put_task_struct(task);
2302 	if (!files)
2303 		return -ENOENT;
2304 
2305 	err = 0;
2306 	spin_lock(&files->file_lock);
2307 	file = fcheck_files(files, fd);
2308 	if (!file)
2309 		err = -EBADF;
2310 	else
2311 		get_file(file);
2312 	spin_unlock(&files->file_lock);
2313 	put_files_struct(files);
2314 
2315 	if (err)
2316 		goto out;
2317 
2318 	if (file->f_op == &bpf_raw_tp_fops) {
2319 		struct bpf_raw_tracepoint *raw_tp = file->private_data;
2320 		struct bpf_raw_event_map *btp = raw_tp->btp;
2321 
2322 		err = bpf_task_fd_query_copy(attr, uattr,
2323 					     raw_tp->prog->aux->id,
2324 					     BPF_FD_TYPE_RAW_TRACEPOINT,
2325 					     btp->tp->name, 0, 0);
2326 		goto put_file;
2327 	}
2328 
2329 	event = perf_get_event(file);
2330 	if (!IS_ERR(event)) {
2331 		u64 probe_offset, probe_addr;
2332 		u32 prog_id, fd_type;
2333 		const char *buf;
2334 
2335 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2336 					      &buf, &probe_offset,
2337 					      &probe_addr);
2338 		if (!err)
2339 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2340 						     fd_type, buf,
2341 						     probe_offset,
2342 						     probe_addr);
2343 		goto put_file;
2344 	}
2345 
2346 	err = -ENOTSUPP;
2347 put_file:
2348 	fput(file);
2349 out:
2350 	return err;
2351 }
2352 
2353 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2354 {
2355 	union bpf_attr attr = {};
2356 	int err;
2357 
2358 	if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2359 		return -EPERM;
2360 
2361 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2362 	if (err)
2363 		return err;
2364 	size = min_t(u32, size, sizeof(attr));
2365 
2366 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
2367 	if (copy_from_user(&attr, uattr, size) != 0)
2368 		return -EFAULT;
2369 
2370 	err = security_bpf(cmd, &attr, size);
2371 	if (err < 0)
2372 		return err;
2373 
2374 	switch (cmd) {
2375 	case BPF_MAP_CREATE:
2376 		err = map_create(&attr);
2377 		break;
2378 	case BPF_MAP_LOOKUP_ELEM:
2379 		err = map_lookup_elem(&attr);
2380 		break;
2381 	case BPF_MAP_UPDATE_ELEM:
2382 		err = map_update_elem(&attr);
2383 		break;
2384 	case BPF_MAP_DELETE_ELEM:
2385 		err = map_delete_elem(&attr);
2386 		break;
2387 	case BPF_MAP_GET_NEXT_KEY:
2388 		err = map_get_next_key(&attr);
2389 		break;
2390 	case BPF_PROG_LOAD:
2391 		err = bpf_prog_load(&attr);
2392 		break;
2393 	case BPF_OBJ_PIN:
2394 		err = bpf_obj_pin(&attr);
2395 		break;
2396 	case BPF_OBJ_GET:
2397 		err = bpf_obj_get(&attr);
2398 		break;
2399 	case BPF_PROG_ATTACH:
2400 		err = bpf_prog_attach(&attr);
2401 		break;
2402 	case BPF_PROG_DETACH:
2403 		err = bpf_prog_detach(&attr);
2404 		break;
2405 	case BPF_PROG_QUERY:
2406 		err = bpf_prog_query(&attr, uattr);
2407 		break;
2408 	case BPF_PROG_TEST_RUN:
2409 		err = bpf_prog_test_run(&attr, uattr);
2410 		break;
2411 	case BPF_PROG_GET_NEXT_ID:
2412 		err = bpf_obj_get_next_id(&attr, uattr,
2413 					  &prog_idr, &prog_idr_lock);
2414 		break;
2415 	case BPF_MAP_GET_NEXT_ID:
2416 		err = bpf_obj_get_next_id(&attr, uattr,
2417 					  &map_idr, &map_idr_lock);
2418 		break;
2419 	case BPF_PROG_GET_FD_BY_ID:
2420 		err = bpf_prog_get_fd_by_id(&attr);
2421 		break;
2422 	case BPF_MAP_GET_FD_BY_ID:
2423 		err = bpf_map_get_fd_by_id(&attr);
2424 		break;
2425 	case BPF_OBJ_GET_INFO_BY_FD:
2426 		err = bpf_obj_get_info_by_fd(&attr, uattr);
2427 		break;
2428 	case BPF_RAW_TRACEPOINT_OPEN:
2429 		err = bpf_raw_tracepoint_open(&attr);
2430 		break;
2431 	case BPF_BTF_LOAD:
2432 		err = bpf_btf_load(&attr);
2433 		break;
2434 	case BPF_BTF_GET_FD_BY_ID:
2435 		err = bpf_btf_get_fd_by_id(&attr);
2436 		break;
2437 	case BPF_TASK_FD_QUERY:
2438 		err = bpf_task_fd_query(&attr, uattr);
2439 		break;
2440 	default:
2441 		err = -EINVAL;
2442 		break;
2443 	}
2444 
2445 	return err;
2446 }
2447