xref: /linux/kernel/bpf/syscall.c (revision 544029862cbb1d7903e19f2e58f48d4884e1201b)
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(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 *btf,
460 		     const struct btf_type *key_type,
461 		     const struct btf_type *value_type)
462 {
463 	return -ENOTSUPP;
464 }
465 
466 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
467 			 u32 btf_key_id, u32 btf_value_id)
468 {
469 	const struct btf_type *key_type, *value_type;
470 	u32 key_size, value_size;
471 	int ret = 0;
472 
473 	key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
474 	if (!key_type || key_size != map->key_size)
475 		return -EINVAL;
476 
477 	value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
478 	if (!value_type || value_size != map->value_size)
479 		return -EINVAL;
480 
481 	map->spin_lock_off = btf_find_spin_lock(btf, value_type);
482 
483 	if (map_value_has_spin_lock(map)) {
484 		if (map->map_type != BPF_MAP_TYPE_HASH &&
485 		    map->map_type != BPF_MAP_TYPE_ARRAY &&
486 		    map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE)
487 			return -ENOTSUPP;
488 		if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
489 		    map->value_size) {
490 			WARN_ONCE(1,
491 				  "verifier bug spin_lock_off %d value_size %d\n",
492 				  map->spin_lock_off, map->value_size);
493 			return -EFAULT;
494 		}
495 	}
496 
497 	if (map->ops->map_check_btf)
498 		ret = map->ops->map_check_btf(map, btf, key_type, value_type);
499 
500 	return ret;
501 }
502 
503 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
504 /* called via syscall */
505 static int map_create(union bpf_attr *attr)
506 {
507 	int numa_node = bpf_map_attr_numa_node(attr);
508 	struct bpf_map *map;
509 	int f_flags;
510 	int err;
511 
512 	err = CHECK_ATTR(BPF_MAP_CREATE);
513 	if (err)
514 		return -EINVAL;
515 
516 	f_flags = bpf_get_file_flag(attr->map_flags);
517 	if (f_flags < 0)
518 		return f_flags;
519 
520 	if (numa_node != NUMA_NO_NODE &&
521 	    ((unsigned int)numa_node >= nr_node_ids ||
522 	     !node_online(numa_node)))
523 		return -EINVAL;
524 
525 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
526 	map = find_and_alloc_map(attr);
527 	if (IS_ERR(map))
528 		return PTR_ERR(map);
529 
530 	err = bpf_obj_name_cpy(map->name, attr->map_name);
531 	if (err)
532 		goto free_map_nouncharge;
533 
534 	atomic_set(&map->refcnt, 1);
535 	atomic_set(&map->usercnt, 1);
536 
537 	if (attr->btf_key_type_id || attr->btf_value_type_id) {
538 		struct btf *btf;
539 
540 		if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
541 			err = -EINVAL;
542 			goto free_map_nouncharge;
543 		}
544 
545 		btf = btf_get_by_fd(attr->btf_fd);
546 		if (IS_ERR(btf)) {
547 			err = PTR_ERR(btf);
548 			goto free_map_nouncharge;
549 		}
550 
551 		err = map_check_btf(map, btf, attr->btf_key_type_id,
552 				    attr->btf_value_type_id);
553 		if (err) {
554 			btf_put(btf);
555 			goto free_map_nouncharge;
556 		}
557 
558 		map->btf = btf;
559 		map->btf_key_type_id = attr->btf_key_type_id;
560 		map->btf_value_type_id = attr->btf_value_type_id;
561 	} else {
562 		map->spin_lock_off = -EINVAL;
563 	}
564 
565 	err = security_bpf_map_alloc(map);
566 	if (err)
567 		goto free_map_nouncharge;
568 
569 	err = bpf_map_init_memlock(map);
570 	if (err)
571 		goto free_map_sec;
572 
573 	err = bpf_map_alloc_id(map);
574 	if (err)
575 		goto free_map;
576 
577 	err = bpf_map_new_fd(map, f_flags);
578 	if (err < 0) {
579 		/* failed to allocate fd.
580 		 * bpf_map_put_with_uref() is needed because the above
581 		 * bpf_map_alloc_id() has published the map
582 		 * to the userspace and the userspace may
583 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
584 		 */
585 		bpf_map_put_with_uref(map);
586 		return err;
587 	}
588 
589 	return err;
590 
591 free_map:
592 	bpf_map_release_memlock(map);
593 free_map_sec:
594 	security_bpf_map_free(map);
595 free_map_nouncharge:
596 	btf_put(map->btf);
597 	map->ops->map_free(map);
598 	return err;
599 }
600 
601 /* if error is returned, fd is released.
602  * On success caller should complete fd access with matching fdput()
603  */
604 struct bpf_map *__bpf_map_get(struct fd f)
605 {
606 	if (!f.file)
607 		return ERR_PTR(-EBADF);
608 	if (f.file->f_op != &bpf_map_fops) {
609 		fdput(f);
610 		return ERR_PTR(-EINVAL);
611 	}
612 
613 	return f.file->private_data;
614 }
615 
616 /* prog's and map's refcnt limit */
617 #define BPF_MAX_REFCNT 32768
618 
619 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
620 {
621 	if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
622 		atomic_dec(&map->refcnt);
623 		return ERR_PTR(-EBUSY);
624 	}
625 	if (uref)
626 		atomic_inc(&map->usercnt);
627 	return map;
628 }
629 EXPORT_SYMBOL_GPL(bpf_map_inc);
630 
631 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
632 {
633 	struct fd f = fdget(ufd);
634 	struct bpf_map *map;
635 
636 	map = __bpf_map_get(f);
637 	if (IS_ERR(map))
638 		return map;
639 
640 	map = bpf_map_inc(map, true);
641 	fdput(f);
642 
643 	return map;
644 }
645 
646 /* map_idr_lock should have been held */
647 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
648 					    bool uref)
649 {
650 	int refold;
651 
652 	refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
653 
654 	if (refold >= BPF_MAX_REFCNT) {
655 		__bpf_map_put(map, false);
656 		return ERR_PTR(-EBUSY);
657 	}
658 
659 	if (!refold)
660 		return ERR_PTR(-ENOENT);
661 
662 	if (uref)
663 		atomic_inc(&map->usercnt);
664 
665 	return map;
666 }
667 
668 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
669 {
670 	return -ENOTSUPP;
671 }
672 
673 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
674 {
675 	if (key_size)
676 		return memdup_user(ukey, key_size);
677 
678 	if (ukey)
679 		return ERR_PTR(-EINVAL);
680 
681 	return NULL;
682 }
683 
684 /* last field in 'union bpf_attr' used by this command */
685 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
686 
687 static int map_lookup_elem(union bpf_attr *attr)
688 {
689 	void __user *ukey = u64_to_user_ptr(attr->key);
690 	void __user *uvalue = u64_to_user_ptr(attr->value);
691 	int ufd = attr->map_fd;
692 	struct bpf_map *map;
693 	void *key, *value, *ptr;
694 	u32 value_size;
695 	struct fd f;
696 	int err;
697 
698 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
699 		return -EINVAL;
700 
701 	if (attr->flags & ~BPF_F_LOCK)
702 		return -EINVAL;
703 
704 	f = fdget(ufd);
705 	map = __bpf_map_get(f);
706 	if (IS_ERR(map))
707 		return PTR_ERR(map);
708 
709 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
710 		err = -EPERM;
711 		goto err_put;
712 	}
713 
714 	if ((attr->flags & BPF_F_LOCK) &&
715 	    !map_value_has_spin_lock(map)) {
716 		err = -EINVAL;
717 		goto err_put;
718 	}
719 
720 	key = __bpf_copy_key(ukey, map->key_size);
721 	if (IS_ERR(key)) {
722 		err = PTR_ERR(key);
723 		goto err_put;
724 	}
725 
726 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
727 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
728 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
729 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
730 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
731 	else if (IS_FD_MAP(map))
732 		value_size = sizeof(u32);
733 	else
734 		value_size = map->value_size;
735 
736 	err = -ENOMEM;
737 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
738 	if (!value)
739 		goto free_key;
740 
741 	if (bpf_map_is_dev_bound(map)) {
742 		err = bpf_map_offload_lookup_elem(map, key, value);
743 		goto done;
744 	}
745 
746 	preempt_disable();
747 	this_cpu_inc(bpf_prog_active);
748 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
749 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
750 		err = bpf_percpu_hash_copy(map, key, value);
751 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
752 		err = bpf_percpu_array_copy(map, key, value);
753 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
754 		err = bpf_percpu_cgroup_storage_copy(map, key, value);
755 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
756 		err = bpf_stackmap_copy(map, key, value);
757 	} else if (IS_FD_ARRAY(map)) {
758 		err = bpf_fd_array_map_lookup_elem(map, key, value);
759 	} else if (IS_FD_HASH(map)) {
760 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
761 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
762 		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
763 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
764 		   map->map_type == BPF_MAP_TYPE_STACK) {
765 		err = map->ops->map_peek_elem(map, value);
766 	} else {
767 		rcu_read_lock();
768 		ptr = map->ops->map_lookup_elem(map, key);
769 		if (IS_ERR(ptr)) {
770 			err = PTR_ERR(ptr);
771 		} else if (!ptr) {
772 			err = -ENOENT;
773 		} else {
774 			err = 0;
775 			if (attr->flags & BPF_F_LOCK)
776 				/* lock 'ptr' and copy everything but lock */
777 				copy_map_value_locked(map, value, ptr, true);
778 			else
779 				copy_map_value(map, value, ptr);
780 			/* mask lock, since value wasn't zero inited */
781 			check_and_init_map_lock(map, value);
782 		}
783 		rcu_read_unlock();
784 	}
785 	this_cpu_dec(bpf_prog_active);
786 	preempt_enable();
787 
788 done:
789 	if (err)
790 		goto free_value;
791 
792 	err = -EFAULT;
793 	if (copy_to_user(uvalue, value, value_size) != 0)
794 		goto free_value;
795 
796 	err = 0;
797 
798 free_value:
799 	kfree(value);
800 free_key:
801 	kfree(key);
802 err_put:
803 	fdput(f);
804 	return err;
805 }
806 
807 static void maybe_wait_bpf_programs(struct bpf_map *map)
808 {
809 	/* Wait for any running BPF programs to complete so that
810 	 * userspace, when we return to it, knows that all programs
811 	 * that could be running use the new map value.
812 	 */
813 	if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
814 	    map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
815 		synchronize_rcu();
816 }
817 
818 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
819 
820 static int map_update_elem(union bpf_attr *attr)
821 {
822 	void __user *ukey = u64_to_user_ptr(attr->key);
823 	void __user *uvalue = u64_to_user_ptr(attr->value);
824 	int ufd = attr->map_fd;
825 	struct bpf_map *map;
826 	void *key, *value;
827 	u32 value_size;
828 	struct fd f;
829 	int err;
830 
831 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
832 		return -EINVAL;
833 
834 	f = fdget(ufd);
835 	map = __bpf_map_get(f);
836 	if (IS_ERR(map))
837 		return PTR_ERR(map);
838 
839 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
840 		err = -EPERM;
841 		goto err_put;
842 	}
843 
844 	if ((attr->flags & BPF_F_LOCK) &&
845 	    !map_value_has_spin_lock(map)) {
846 		err = -EINVAL;
847 		goto err_put;
848 	}
849 
850 	key = __bpf_copy_key(ukey, map->key_size);
851 	if (IS_ERR(key)) {
852 		err = PTR_ERR(key);
853 		goto err_put;
854 	}
855 
856 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
857 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
858 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
859 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
860 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
861 	else
862 		value_size = map->value_size;
863 
864 	err = -ENOMEM;
865 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
866 	if (!value)
867 		goto free_key;
868 
869 	err = -EFAULT;
870 	if (copy_from_user(value, uvalue, value_size) != 0)
871 		goto free_value;
872 
873 	/* Need to create a kthread, thus must support schedule */
874 	if (bpf_map_is_dev_bound(map)) {
875 		err = bpf_map_offload_update_elem(map, key, value, attr->flags);
876 		goto out;
877 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
878 		   map->map_type == BPF_MAP_TYPE_SOCKHASH ||
879 		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
880 		err = map->ops->map_update_elem(map, key, value, attr->flags);
881 		goto out;
882 	}
883 
884 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
885 	 * inside bpf map update or delete otherwise deadlocks are possible
886 	 */
887 	preempt_disable();
888 	__this_cpu_inc(bpf_prog_active);
889 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
890 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
891 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
892 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
893 		err = bpf_percpu_array_update(map, key, value, attr->flags);
894 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
895 		err = bpf_percpu_cgroup_storage_update(map, key, value,
896 						       attr->flags);
897 	} else if (IS_FD_ARRAY(map)) {
898 		rcu_read_lock();
899 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
900 						   attr->flags);
901 		rcu_read_unlock();
902 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
903 		rcu_read_lock();
904 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
905 						  attr->flags);
906 		rcu_read_unlock();
907 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
908 		/* rcu_read_lock() is not needed */
909 		err = bpf_fd_reuseport_array_update_elem(map, key, value,
910 							 attr->flags);
911 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
912 		   map->map_type == BPF_MAP_TYPE_STACK) {
913 		err = map->ops->map_push_elem(map, value, attr->flags);
914 	} else {
915 		rcu_read_lock();
916 		err = map->ops->map_update_elem(map, key, value, attr->flags);
917 		rcu_read_unlock();
918 	}
919 	__this_cpu_dec(bpf_prog_active);
920 	preempt_enable();
921 	maybe_wait_bpf_programs(map);
922 out:
923 free_value:
924 	kfree(value);
925 free_key:
926 	kfree(key);
927 err_put:
928 	fdput(f);
929 	return err;
930 }
931 
932 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
933 
934 static int map_delete_elem(union bpf_attr *attr)
935 {
936 	void __user *ukey = u64_to_user_ptr(attr->key);
937 	int ufd = attr->map_fd;
938 	struct bpf_map *map;
939 	struct fd f;
940 	void *key;
941 	int err;
942 
943 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
944 		return -EINVAL;
945 
946 	f = fdget(ufd);
947 	map = __bpf_map_get(f);
948 	if (IS_ERR(map))
949 		return PTR_ERR(map);
950 
951 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
952 		err = -EPERM;
953 		goto err_put;
954 	}
955 
956 	key = __bpf_copy_key(ukey, map->key_size);
957 	if (IS_ERR(key)) {
958 		err = PTR_ERR(key);
959 		goto err_put;
960 	}
961 
962 	if (bpf_map_is_dev_bound(map)) {
963 		err = bpf_map_offload_delete_elem(map, key);
964 		goto out;
965 	}
966 
967 	preempt_disable();
968 	__this_cpu_inc(bpf_prog_active);
969 	rcu_read_lock();
970 	err = map->ops->map_delete_elem(map, key);
971 	rcu_read_unlock();
972 	__this_cpu_dec(bpf_prog_active);
973 	preempt_enable();
974 	maybe_wait_bpf_programs(map);
975 out:
976 	kfree(key);
977 err_put:
978 	fdput(f);
979 	return err;
980 }
981 
982 /* last field in 'union bpf_attr' used by this command */
983 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
984 
985 static int map_get_next_key(union bpf_attr *attr)
986 {
987 	void __user *ukey = u64_to_user_ptr(attr->key);
988 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
989 	int ufd = attr->map_fd;
990 	struct bpf_map *map;
991 	void *key, *next_key;
992 	struct fd f;
993 	int err;
994 
995 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
996 		return -EINVAL;
997 
998 	f = fdget(ufd);
999 	map = __bpf_map_get(f);
1000 	if (IS_ERR(map))
1001 		return PTR_ERR(map);
1002 
1003 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
1004 		err = -EPERM;
1005 		goto err_put;
1006 	}
1007 
1008 	if (ukey) {
1009 		key = __bpf_copy_key(ukey, map->key_size);
1010 		if (IS_ERR(key)) {
1011 			err = PTR_ERR(key);
1012 			goto err_put;
1013 		}
1014 	} else {
1015 		key = NULL;
1016 	}
1017 
1018 	err = -ENOMEM;
1019 	next_key = kmalloc(map->key_size, GFP_USER);
1020 	if (!next_key)
1021 		goto free_key;
1022 
1023 	if (bpf_map_is_dev_bound(map)) {
1024 		err = bpf_map_offload_get_next_key(map, key, next_key);
1025 		goto out;
1026 	}
1027 
1028 	rcu_read_lock();
1029 	err = map->ops->map_get_next_key(map, key, next_key);
1030 	rcu_read_unlock();
1031 out:
1032 	if (err)
1033 		goto free_next_key;
1034 
1035 	err = -EFAULT;
1036 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1037 		goto free_next_key;
1038 
1039 	err = 0;
1040 
1041 free_next_key:
1042 	kfree(next_key);
1043 free_key:
1044 	kfree(key);
1045 err_put:
1046 	fdput(f);
1047 	return err;
1048 }
1049 
1050 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1051 
1052 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1053 {
1054 	void __user *ukey = u64_to_user_ptr(attr->key);
1055 	void __user *uvalue = u64_to_user_ptr(attr->value);
1056 	int ufd = attr->map_fd;
1057 	struct bpf_map *map;
1058 	void *key, *value;
1059 	u32 value_size;
1060 	struct fd f;
1061 	int err;
1062 
1063 	if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1064 		return -EINVAL;
1065 
1066 	f = fdget(ufd);
1067 	map = __bpf_map_get(f);
1068 	if (IS_ERR(map))
1069 		return PTR_ERR(map);
1070 
1071 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
1072 		err = -EPERM;
1073 		goto err_put;
1074 	}
1075 
1076 	key = __bpf_copy_key(ukey, map->key_size);
1077 	if (IS_ERR(key)) {
1078 		err = PTR_ERR(key);
1079 		goto err_put;
1080 	}
1081 
1082 	value_size = map->value_size;
1083 
1084 	err = -ENOMEM;
1085 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1086 	if (!value)
1087 		goto free_key;
1088 
1089 	if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1090 	    map->map_type == BPF_MAP_TYPE_STACK) {
1091 		err = map->ops->map_pop_elem(map, value);
1092 	} else {
1093 		err = -ENOTSUPP;
1094 	}
1095 
1096 	if (err)
1097 		goto free_value;
1098 
1099 	if (copy_to_user(uvalue, value, value_size) != 0)
1100 		goto free_value;
1101 
1102 	err = 0;
1103 
1104 free_value:
1105 	kfree(value);
1106 free_key:
1107 	kfree(key);
1108 err_put:
1109 	fdput(f);
1110 	return err;
1111 }
1112 
1113 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1114 #define BPF_PROG_TYPE(_id, _name) \
1115 	[_id] = & _name ## _prog_ops,
1116 #define BPF_MAP_TYPE(_id, _ops)
1117 #include <linux/bpf_types.h>
1118 #undef BPF_PROG_TYPE
1119 #undef BPF_MAP_TYPE
1120 };
1121 
1122 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1123 {
1124 	const struct bpf_prog_ops *ops;
1125 
1126 	if (type >= ARRAY_SIZE(bpf_prog_types))
1127 		return -EINVAL;
1128 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1129 	ops = bpf_prog_types[type];
1130 	if (!ops)
1131 		return -EINVAL;
1132 
1133 	if (!bpf_prog_is_dev_bound(prog->aux))
1134 		prog->aux->ops = ops;
1135 	else
1136 		prog->aux->ops = &bpf_offload_prog_ops;
1137 	prog->type = type;
1138 	return 0;
1139 }
1140 
1141 /* drop refcnt on maps used by eBPF program and free auxilary data */
1142 static void free_used_maps(struct bpf_prog_aux *aux)
1143 {
1144 	enum bpf_cgroup_storage_type stype;
1145 	int i;
1146 
1147 	for_each_cgroup_storage_type(stype) {
1148 		if (!aux->cgroup_storage[stype])
1149 			continue;
1150 		bpf_cgroup_storage_release(aux->prog,
1151 					   aux->cgroup_storage[stype]);
1152 	}
1153 
1154 	for (i = 0; i < aux->used_map_cnt; i++)
1155 		bpf_map_put(aux->used_maps[i]);
1156 
1157 	kfree(aux->used_maps);
1158 }
1159 
1160 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1161 {
1162 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1163 	unsigned long user_bufs;
1164 
1165 	if (user) {
1166 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1167 		if (user_bufs > memlock_limit) {
1168 			atomic_long_sub(pages, &user->locked_vm);
1169 			return -EPERM;
1170 		}
1171 	}
1172 
1173 	return 0;
1174 }
1175 
1176 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1177 {
1178 	if (user)
1179 		atomic_long_sub(pages, &user->locked_vm);
1180 }
1181 
1182 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1183 {
1184 	struct user_struct *user = get_current_user();
1185 	int ret;
1186 
1187 	ret = __bpf_prog_charge(user, prog->pages);
1188 	if (ret) {
1189 		free_uid(user);
1190 		return ret;
1191 	}
1192 
1193 	prog->aux->user = user;
1194 	return 0;
1195 }
1196 
1197 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1198 {
1199 	struct user_struct *user = prog->aux->user;
1200 
1201 	__bpf_prog_uncharge(user, prog->pages);
1202 	free_uid(user);
1203 }
1204 
1205 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1206 {
1207 	int id;
1208 
1209 	idr_preload(GFP_KERNEL);
1210 	spin_lock_bh(&prog_idr_lock);
1211 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1212 	if (id > 0)
1213 		prog->aux->id = id;
1214 	spin_unlock_bh(&prog_idr_lock);
1215 	idr_preload_end();
1216 
1217 	/* id is in [1, INT_MAX) */
1218 	if (WARN_ON_ONCE(!id))
1219 		return -ENOSPC;
1220 
1221 	return id > 0 ? 0 : id;
1222 }
1223 
1224 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1225 {
1226 	/* cBPF to eBPF migrations are currently not in the idr store.
1227 	 * Offloaded programs are removed from the store when their device
1228 	 * disappears - even if someone grabs an fd to them they are unusable,
1229 	 * simply waiting for refcnt to drop to be freed.
1230 	 */
1231 	if (!prog->aux->id)
1232 		return;
1233 
1234 	if (do_idr_lock)
1235 		spin_lock_bh(&prog_idr_lock);
1236 	else
1237 		__acquire(&prog_idr_lock);
1238 
1239 	idr_remove(&prog_idr, prog->aux->id);
1240 	prog->aux->id = 0;
1241 
1242 	if (do_idr_lock)
1243 		spin_unlock_bh(&prog_idr_lock);
1244 	else
1245 		__release(&prog_idr_lock);
1246 }
1247 
1248 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1249 {
1250 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1251 
1252 	free_used_maps(aux);
1253 	bpf_prog_uncharge_memlock(aux->prog);
1254 	security_bpf_prog_free(aux);
1255 	bpf_prog_free(aux->prog);
1256 }
1257 
1258 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1259 {
1260 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
1261 		/* bpf_prog_free_id() must be called first */
1262 		bpf_prog_free_id(prog, do_idr_lock);
1263 		bpf_prog_kallsyms_del_all(prog);
1264 		btf_put(prog->aux->btf);
1265 		kvfree(prog->aux->func_info);
1266 		bpf_prog_free_linfo(prog);
1267 
1268 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1269 	}
1270 }
1271 
1272 void bpf_prog_put(struct bpf_prog *prog)
1273 {
1274 	__bpf_prog_put(prog, true);
1275 }
1276 EXPORT_SYMBOL_GPL(bpf_prog_put);
1277 
1278 static int bpf_prog_release(struct inode *inode, struct file *filp)
1279 {
1280 	struct bpf_prog *prog = filp->private_data;
1281 
1282 	bpf_prog_put(prog);
1283 	return 0;
1284 }
1285 
1286 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1287 			       struct bpf_prog_stats *stats)
1288 {
1289 	u64 nsecs = 0, cnt = 0;
1290 	int cpu;
1291 
1292 	for_each_possible_cpu(cpu) {
1293 		const struct bpf_prog_stats *st;
1294 		unsigned int start;
1295 		u64 tnsecs, tcnt;
1296 
1297 		st = per_cpu_ptr(prog->aux->stats, cpu);
1298 		do {
1299 			start = u64_stats_fetch_begin_irq(&st->syncp);
1300 			tnsecs = st->nsecs;
1301 			tcnt = st->cnt;
1302 		} while (u64_stats_fetch_retry_irq(&st->syncp, start));
1303 		nsecs += tnsecs;
1304 		cnt += tcnt;
1305 	}
1306 	stats->nsecs = nsecs;
1307 	stats->cnt = cnt;
1308 }
1309 
1310 #ifdef CONFIG_PROC_FS
1311 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1312 {
1313 	const struct bpf_prog *prog = filp->private_data;
1314 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1315 	struct bpf_prog_stats stats;
1316 
1317 	bpf_prog_get_stats(prog, &stats);
1318 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1319 	seq_printf(m,
1320 		   "prog_type:\t%u\n"
1321 		   "prog_jited:\t%u\n"
1322 		   "prog_tag:\t%s\n"
1323 		   "memlock:\t%llu\n"
1324 		   "prog_id:\t%u\n"
1325 		   "run_time_ns:\t%llu\n"
1326 		   "run_cnt:\t%llu\n",
1327 		   prog->type,
1328 		   prog->jited,
1329 		   prog_tag,
1330 		   prog->pages * 1ULL << PAGE_SHIFT,
1331 		   prog->aux->id,
1332 		   stats.nsecs,
1333 		   stats.cnt);
1334 }
1335 #endif
1336 
1337 const struct file_operations bpf_prog_fops = {
1338 #ifdef CONFIG_PROC_FS
1339 	.show_fdinfo	= bpf_prog_show_fdinfo,
1340 #endif
1341 	.release	= bpf_prog_release,
1342 	.read		= bpf_dummy_read,
1343 	.write		= bpf_dummy_write,
1344 };
1345 
1346 int bpf_prog_new_fd(struct bpf_prog *prog)
1347 {
1348 	int ret;
1349 
1350 	ret = security_bpf_prog(prog);
1351 	if (ret < 0)
1352 		return ret;
1353 
1354 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1355 				O_RDWR | O_CLOEXEC);
1356 }
1357 
1358 static struct bpf_prog *____bpf_prog_get(struct fd f)
1359 {
1360 	if (!f.file)
1361 		return ERR_PTR(-EBADF);
1362 	if (f.file->f_op != &bpf_prog_fops) {
1363 		fdput(f);
1364 		return ERR_PTR(-EINVAL);
1365 	}
1366 
1367 	return f.file->private_data;
1368 }
1369 
1370 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1371 {
1372 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1373 		atomic_sub(i, &prog->aux->refcnt);
1374 		return ERR_PTR(-EBUSY);
1375 	}
1376 	return prog;
1377 }
1378 EXPORT_SYMBOL_GPL(bpf_prog_add);
1379 
1380 void bpf_prog_sub(struct bpf_prog *prog, int i)
1381 {
1382 	/* Only to be used for undoing previous bpf_prog_add() in some
1383 	 * error path. We still know that another entity in our call
1384 	 * path holds a reference to the program, thus atomic_sub() can
1385 	 * be safely used in such cases!
1386 	 */
1387 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1388 }
1389 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1390 
1391 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1392 {
1393 	return bpf_prog_add(prog, 1);
1394 }
1395 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1396 
1397 /* prog_idr_lock should have been held */
1398 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1399 {
1400 	int refold;
1401 
1402 	refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1403 
1404 	if (refold >= BPF_MAX_REFCNT) {
1405 		__bpf_prog_put(prog, false);
1406 		return ERR_PTR(-EBUSY);
1407 	}
1408 
1409 	if (!refold)
1410 		return ERR_PTR(-ENOENT);
1411 
1412 	return prog;
1413 }
1414 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1415 
1416 bool bpf_prog_get_ok(struct bpf_prog *prog,
1417 			    enum bpf_prog_type *attach_type, bool attach_drv)
1418 {
1419 	/* not an attachment, just a refcount inc, always allow */
1420 	if (!attach_type)
1421 		return true;
1422 
1423 	if (prog->type != *attach_type)
1424 		return false;
1425 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1426 		return false;
1427 
1428 	return true;
1429 }
1430 
1431 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1432 				       bool attach_drv)
1433 {
1434 	struct fd f = fdget(ufd);
1435 	struct bpf_prog *prog;
1436 
1437 	prog = ____bpf_prog_get(f);
1438 	if (IS_ERR(prog))
1439 		return prog;
1440 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1441 		prog = ERR_PTR(-EINVAL);
1442 		goto out;
1443 	}
1444 
1445 	prog = bpf_prog_inc(prog);
1446 out:
1447 	fdput(f);
1448 	return prog;
1449 }
1450 
1451 struct bpf_prog *bpf_prog_get(u32 ufd)
1452 {
1453 	return __bpf_prog_get(ufd, NULL, false);
1454 }
1455 
1456 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1457 				       bool attach_drv)
1458 {
1459 	return __bpf_prog_get(ufd, &type, attach_drv);
1460 }
1461 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1462 
1463 /* Initially all BPF programs could be loaded w/o specifying
1464  * expected_attach_type. Later for some of them specifying expected_attach_type
1465  * at load time became required so that program could be validated properly.
1466  * Programs of types that are allowed to be loaded both w/ and w/o (for
1467  * backward compatibility) expected_attach_type, should have the default attach
1468  * type assigned to expected_attach_type for the latter case, so that it can be
1469  * validated later at attach time.
1470  *
1471  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1472  * prog type requires it but has some attach types that have to be backward
1473  * compatible.
1474  */
1475 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1476 {
1477 	switch (attr->prog_type) {
1478 	case BPF_PROG_TYPE_CGROUP_SOCK:
1479 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1480 		 * exist so checking for non-zero is the way to go here.
1481 		 */
1482 		if (!attr->expected_attach_type)
1483 			attr->expected_attach_type =
1484 				BPF_CGROUP_INET_SOCK_CREATE;
1485 		break;
1486 	}
1487 }
1488 
1489 static int
1490 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1491 				enum bpf_attach_type expected_attach_type)
1492 {
1493 	switch (prog_type) {
1494 	case BPF_PROG_TYPE_CGROUP_SOCK:
1495 		switch (expected_attach_type) {
1496 		case BPF_CGROUP_INET_SOCK_CREATE:
1497 		case BPF_CGROUP_INET4_POST_BIND:
1498 		case BPF_CGROUP_INET6_POST_BIND:
1499 			return 0;
1500 		default:
1501 			return -EINVAL;
1502 		}
1503 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1504 		switch (expected_attach_type) {
1505 		case BPF_CGROUP_INET4_BIND:
1506 		case BPF_CGROUP_INET6_BIND:
1507 		case BPF_CGROUP_INET4_CONNECT:
1508 		case BPF_CGROUP_INET6_CONNECT:
1509 		case BPF_CGROUP_UDP4_SENDMSG:
1510 		case BPF_CGROUP_UDP6_SENDMSG:
1511 			return 0;
1512 		default:
1513 			return -EINVAL;
1514 		}
1515 	default:
1516 		return 0;
1517 	}
1518 }
1519 
1520 /* last field in 'union bpf_attr' used by this command */
1521 #define	BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1522 
1523 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1524 {
1525 	enum bpf_prog_type type = attr->prog_type;
1526 	struct bpf_prog *prog;
1527 	int err;
1528 	char license[128];
1529 	bool is_gpl;
1530 
1531 	if (CHECK_ATTR(BPF_PROG_LOAD))
1532 		return -EINVAL;
1533 
1534 	if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT))
1535 		return -EINVAL;
1536 
1537 	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1538 	    (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1539 	    !capable(CAP_SYS_ADMIN))
1540 		return -EPERM;
1541 
1542 	/* copy eBPF program license from user space */
1543 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1544 			      sizeof(license) - 1) < 0)
1545 		return -EFAULT;
1546 	license[sizeof(license) - 1] = 0;
1547 
1548 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1549 	is_gpl = license_is_gpl_compatible(license);
1550 
1551 	if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1552 		return -E2BIG;
1553 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1554 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
1555 	    !capable(CAP_SYS_ADMIN))
1556 		return -EPERM;
1557 
1558 	bpf_prog_load_fixup_attach_type(attr);
1559 	if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1560 		return -EINVAL;
1561 
1562 	/* plain bpf_prog allocation */
1563 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1564 	if (!prog)
1565 		return -ENOMEM;
1566 
1567 	prog->expected_attach_type = attr->expected_attach_type;
1568 
1569 	prog->aux->offload_requested = !!attr->prog_ifindex;
1570 
1571 	err = security_bpf_prog_alloc(prog->aux);
1572 	if (err)
1573 		goto free_prog_nouncharge;
1574 
1575 	err = bpf_prog_charge_memlock(prog);
1576 	if (err)
1577 		goto free_prog_sec;
1578 
1579 	prog->len = attr->insn_cnt;
1580 
1581 	err = -EFAULT;
1582 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1583 			   bpf_prog_insn_size(prog)) != 0)
1584 		goto free_prog;
1585 
1586 	prog->orig_prog = NULL;
1587 	prog->jited = 0;
1588 
1589 	atomic_set(&prog->aux->refcnt, 1);
1590 	prog->gpl_compatible = is_gpl ? 1 : 0;
1591 
1592 	if (bpf_prog_is_dev_bound(prog->aux)) {
1593 		err = bpf_prog_offload_init(prog, attr);
1594 		if (err)
1595 			goto free_prog;
1596 	}
1597 
1598 	/* find program type: socket_filter vs tracing_filter */
1599 	err = find_prog_type(type, prog);
1600 	if (err < 0)
1601 		goto free_prog;
1602 
1603 	prog->aux->load_time = ktime_get_boot_ns();
1604 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1605 	if (err)
1606 		goto free_prog;
1607 
1608 	/* run eBPF verifier */
1609 	err = bpf_check(&prog, attr, uattr);
1610 	if (err < 0)
1611 		goto free_used_maps;
1612 
1613 	prog = bpf_prog_select_runtime(prog, &err);
1614 	if (err < 0)
1615 		goto free_used_maps;
1616 
1617 	err = bpf_prog_alloc_id(prog);
1618 	if (err)
1619 		goto free_used_maps;
1620 
1621 	err = bpf_prog_new_fd(prog);
1622 	if (err < 0) {
1623 		/* failed to allocate fd.
1624 		 * bpf_prog_put() is needed because the above
1625 		 * bpf_prog_alloc_id() has published the prog
1626 		 * to the userspace and the userspace may
1627 		 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1628 		 */
1629 		bpf_prog_put(prog);
1630 		return err;
1631 	}
1632 
1633 	bpf_prog_kallsyms_add(prog);
1634 	return err;
1635 
1636 free_used_maps:
1637 	bpf_prog_free_linfo(prog);
1638 	kvfree(prog->aux->func_info);
1639 	btf_put(prog->aux->btf);
1640 	bpf_prog_kallsyms_del_subprogs(prog);
1641 	free_used_maps(prog->aux);
1642 free_prog:
1643 	bpf_prog_uncharge_memlock(prog);
1644 free_prog_sec:
1645 	security_bpf_prog_free(prog->aux);
1646 free_prog_nouncharge:
1647 	bpf_prog_free(prog);
1648 	return err;
1649 }
1650 
1651 #define BPF_OBJ_LAST_FIELD file_flags
1652 
1653 static int bpf_obj_pin(const union bpf_attr *attr)
1654 {
1655 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1656 		return -EINVAL;
1657 
1658 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1659 }
1660 
1661 static int bpf_obj_get(const union bpf_attr *attr)
1662 {
1663 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1664 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1665 		return -EINVAL;
1666 
1667 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1668 				attr->file_flags);
1669 }
1670 
1671 struct bpf_raw_tracepoint {
1672 	struct bpf_raw_event_map *btp;
1673 	struct bpf_prog *prog;
1674 };
1675 
1676 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1677 {
1678 	struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1679 
1680 	if (raw_tp->prog) {
1681 		bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1682 		bpf_prog_put(raw_tp->prog);
1683 	}
1684 	bpf_put_raw_tracepoint(raw_tp->btp);
1685 	kfree(raw_tp);
1686 	return 0;
1687 }
1688 
1689 static const struct file_operations bpf_raw_tp_fops = {
1690 	.release	= bpf_raw_tracepoint_release,
1691 	.read		= bpf_dummy_read,
1692 	.write		= bpf_dummy_write,
1693 };
1694 
1695 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1696 
1697 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1698 {
1699 	struct bpf_raw_tracepoint *raw_tp;
1700 	struct bpf_raw_event_map *btp;
1701 	struct bpf_prog *prog;
1702 	char tp_name[128];
1703 	int tp_fd, err;
1704 
1705 	if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1706 			      sizeof(tp_name) - 1) < 0)
1707 		return -EFAULT;
1708 	tp_name[sizeof(tp_name) - 1] = 0;
1709 
1710 	btp = bpf_get_raw_tracepoint(tp_name);
1711 	if (!btp)
1712 		return -ENOENT;
1713 
1714 	raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1715 	if (!raw_tp) {
1716 		err = -ENOMEM;
1717 		goto out_put_btp;
1718 	}
1719 	raw_tp->btp = btp;
1720 
1721 	prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1722 				 BPF_PROG_TYPE_RAW_TRACEPOINT);
1723 	if (IS_ERR(prog)) {
1724 		err = PTR_ERR(prog);
1725 		goto out_free_tp;
1726 	}
1727 
1728 	err = bpf_probe_register(raw_tp->btp, prog);
1729 	if (err)
1730 		goto out_put_prog;
1731 
1732 	raw_tp->prog = prog;
1733 	tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1734 				 O_CLOEXEC);
1735 	if (tp_fd < 0) {
1736 		bpf_probe_unregister(raw_tp->btp, prog);
1737 		err = tp_fd;
1738 		goto out_put_prog;
1739 	}
1740 	return tp_fd;
1741 
1742 out_put_prog:
1743 	bpf_prog_put(prog);
1744 out_free_tp:
1745 	kfree(raw_tp);
1746 out_put_btp:
1747 	bpf_put_raw_tracepoint(btp);
1748 	return err;
1749 }
1750 
1751 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1752 					     enum bpf_attach_type attach_type)
1753 {
1754 	switch (prog->type) {
1755 	case BPF_PROG_TYPE_CGROUP_SOCK:
1756 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1757 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1758 	default:
1759 		return 0;
1760 	}
1761 }
1762 
1763 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1764 
1765 #define BPF_F_ATTACH_MASK \
1766 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1767 
1768 static int bpf_prog_attach(const union bpf_attr *attr)
1769 {
1770 	enum bpf_prog_type ptype;
1771 	struct bpf_prog *prog;
1772 	int ret;
1773 
1774 	if (!capable(CAP_NET_ADMIN))
1775 		return -EPERM;
1776 
1777 	if (CHECK_ATTR(BPF_PROG_ATTACH))
1778 		return -EINVAL;
1779 
1780 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1781 		return -EINVAL;
1782 
1783 	switch (attr->attach_type) {
1784 	case BPF_CGROUP_INET_INGRESS:
1785 	case BPF_CGROUP_INET_EGRESS:
1786 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1787 		break;
1788 	case BPF_CGROUP_INET_SOCK_CREATE:
1789 	case BPF_CGROUP_INET4_POST_BIND:
1790 	case BPF_CGROUP_INET6_POST_BIND:
1791 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1792 		break;
1793 	case BPF_CGROUP_INET4_BIND:
1794 	case BPF_CGROUP_INET6_BIND:
1795 	case BPF_CGROUP_INET4_CONNECT:
1796 	case BPF_CGROUP_INET6_CONNECT:
1797 	case BPF_CGROUP_UDP4_SENDMSG:
1798 	case BPF_CGROUP_UDP6_SENDMSG:
1799 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1800 		break;
1801 	case BPF_CGROUP_SOCK_OPS:
1802 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1803 		break;
1804 	case BPF_CGROUP_DEVICE:
1805 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1806 		break;
1807 	case BPF_SK_MSG_VERDICT:
1808 		ptype = BPF_PROG_TYPE_SK_MSG;
1809 		break;
1810 	case BPF_SK_SKB_STREAM_PARSER:
1811 	case BPF_SK_SKB_STREAM_VERDICT:
1812 		ptype = BPF_PROG_TYPE_SK_SKB;
1813 		break;
1814 	case BPF_LIRC_MODE2:
1815 		ptype = BPF_PROG_TYPE_LIRC_MODE2;
1816 		break;
1817 	case BPF_FLOW_DISSECTOR:
1818 		ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1819 		break;
1820 	default:
1821 		return -EINVAL;
1822 	}
1823 
1824 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1825 	if (IS_ERR(prog))
1826 		return PTR_ERR(prog);
1827 
1828 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1829 		bpf_prog_put(prog);
1830 		return -EINVAL;
1831 	}
1832 
1833 	switch (ptype) {
1834 	case BPF_PROG_TYPE_SK_SKB:
1835 	case BPF_PROG_TYPE_SK_MSG:
1836 		ret = sock_map_get_from_fd(attr, prog);
1837 		break;
1838 	case BPF_PROG_TYPE_LIRC_MODE2:
1839 		ret = lirc_prog_attach(attr, prog);
1840 		break;
1841 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
1842 		ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1843 		break;
1844 	default:
1845 		ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1846 	}
1847 
1848 	if (ret)
1849 		bpf_prog_put(prog);
1850 	return ret;
1851 }
1852 
1853 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1854 
1855 static int bpf_prog_detach(const union bpf_attr *attr)
1856 {
1857 	enum bpf_prog_type ptype;
1858 
1859 	if (!capable(CAP_NET_ADMIN))
1860 		return -EPERM;
1861 
1862 	if (CHECK_ATTR(BPF_PROG_DETACH))
1863 		return -EINVAL;
1864 
1865 	switch (attr->attach_type) {
1866 	case BPF_CGROUP_INET_INGRESS:
1867 	case BPF_CGROUP_INET_EGRESS:
1868 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1869 		break;
1870 	case BPF_CGROUP_INET_SOCK_CREATE:
1871 	case BPF_CGROUP_INET4_POST_BIND:
1872 	case BPF_CGROUP_INET6_POST_BIND:
1873 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1874 		break;
1875 	case BPF_CGROUP_INET4_BIND:
1876 	case BPF_CGROUP_INET6_BIND:
1877 	case BPF_CGROUP_INET4_CONNECT:
1878 	case BPF_CGROUP_INET6_CONNECT:
1879 	case BPF_CGROUP_UDP4_SENDMSG:
1880 	case BPF_CGROUP_UDP6_SENDMSG:
1881 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1882 		break;
1883 	case BPF_CGROUP_SOCK_OPS:
1884 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1885 		break;
1886 	case BPF_CGROUP_DEVICE:
1887 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1888 		break;
1889 	case BPF_SK_MSG_VERDICT:
1890 		return sock_map_get_from_fd(attr, NULL);
1891 	case BPF_SK_SKB_STREAM_PARSER:
1892 	case BPF_SK_SKB_STREAM_VERDICT:
1893 		return sock_map_get_from_fd(attr, NULL);
1894 	case BPF_LIRC_MODE2:
1895 		return lirc_prog_detach(attr);
1896 	case BPF_FLOW_DISSECTOR:
1897 		return skb_flow_dissector_bpf_prog_detach(attr);
1898 	default:
1899 		return -EINVAL;
1900 	}
1901 
1902 	return cgroup_bpf_prog_detach(attr, ptype);
1903 }
1904 
1905 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1906 
1907 static int bpf_prog_query(const union bpf_attr *attr,
1908 			  union bpf_attr __user *uattr)
1909 {
1910 	if (!capable(CAP_NET_ADMIN))
1911 		return -EPERM;
1912 	if (CHECK_ATTR(BPF_PROG_QUERY))
1913 		return -EINVAL;
1914 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1915 		return -EINVAL;
1916 
1917 	switch (attr->query.attach_type) {
1918 	case BPF_CGROUP_INET_INGRESS:
1919 	case BPF_CGROUP_INET_EGRESS:
1920 	case BPF_CGROUP_INET_SOCK_CREATE:
1921 	case BPF_CGROUP_INET4_BIND:
1922 	case BPF_CGROUP_INET6_BIND:
1923 	case BPF_CGROUP_INET4_POST_BIND:
1924 	case BPF_CGROUP_INET6_POST_BIND:
1925 	case BPF_CGROUP_INET4_CONNECT:
1926 	case BPF_CGROUP_INET6_CONNECT:
1927 	case BPF_CGROUP_UDP4_SENDMSG:
1928 	case BPF_CGROUP_UDP6_SENDMSG:
1929 	case BPF_CGROUP_SOCK_OPS:
1930 	case BPF_CGROUP_DEVICE:
1931 		break;
1932 	case BPF_LIRC_MODE2:
1933 		return lirc_prog_query(attr, uattr);
1934 	default:
1935 		return -EINVAL;
1936 	}
1937 
1938 	return cgroup_bpf_prog_query(attr, uattr);
1939 }
1940 
1941 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1942 
1943 static int bpf_prog_test_run(const union bpf_attr *attr,
1944 			     union bpf_attr __user *uattr)
1945 {
1946 	struct bpf_prog *prog;
1947 	int ret = -ENOTSUPP;
1948 
1949 	if (!capable(CAP_SYS_ADMIN))
1950 		return -EPERM;
1951 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1952 		return -EINVAL;
1953 
1954 	prog = bpf_prog_get(attr->test.prog_fd);
1955 	if (IS_ERR(prog))
1956 		return PTR_ERR(prog);
1957 
1958 	if (prog->aux->ops->test_run)
1959 		ret = prog->aux->ops->test_run(prog, attr, uattr);
1960 
1961 	bpf_prog_put(prog);
1962 	return ret;
1963 }
1964 
1965 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1966 
1967 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1968 			       union bpf_attr __user *uattr,
1969 			       struct idr *idr,
1970 			       spinlock_t *lock)
1971 {
1972 	u32 next_id = attr->start_id;
1973 	int err = 0;
1974 
1975 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1976 		return -EINVAL;
1977 
1978 	if (!capable(CAP_SYS_ADMIN))
1979 		return -EPERM;
1980 
1981 	next_id++;
1982 	spin_lock_bh(lock);
1983 	if (!idr_get_next(idr, &next_id))
1984 		err = -ENOENT;
1985 	spin_unlock_bh(lock);
1986 
1987 	if (!err)
1988 		err = put_user(next_id, &uattr->next_id);
1989 
1990 	return err;
1991 }
1992 
1993 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1994 
1995 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1996 {
1997 	struct bpf_prog *prog;
1998 	u32 id = attr->prog_id;
1999 	int fd;
2000 
2001 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2002 		return -EINVAL;
2003 
2004 	if (!capable(CAP_SYS_ADMIN))
2005 		return -EPERM;
2006 
2007 	spin_lock_bh(&prog_idr_lock);
2008 	prog = idr_find(&prog_idr, id);
2009 	if (prog)
2010 		prog = bpf_prog_inc_not_zero(prog);
2011 	else
2012 		prog = ERR_PTR(-ENOENT);
2013 	spin_unlock_bh(&prog_idr_lock);
2014 
2015 	if (IS_ERR(prog))
2016 		return PTR_ERR(prog);
2017 
2018 	fd = bpf_prog_new_fd(prog);
2019 	if (fd < 0)
2020 		bpf_prog_put(prog);
2021 
2022 	return fd;
2023 }
2024 
2025 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2026 
2027 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2028 {
2029 	struct bpf_map *map;
2030 	u32 id = attr->map_id;
2031 	int f_flags;
2032 	int fd;
2033 
2034 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2035 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2036 		return -EINVAL;
2037 
2038 	if (!capable(CAP_SYS_ADMIN))
2039 		return -EPERM;
2040 
2041 	f_flags = bpf_get_file_flag(attr->open_flags);
2042 	if (f_flags < 0)
2043 		return f_flags;
2044 
2045 	spin_lock_bh(&map_idr_lock);
2046 	map = idr_find(&map_idr, id);
2047 	if (map)
2048 		map = bpf_map_inc_not_zero(map, true);
2049 	else
2050 		map = ERR_PTR(-ENOENT);
2051 	spin_unlock_bh(&map_idr_lock);
2052 
2053 	if (IS_ERR(map))
2054 		return PTR_ERR(map);
2055 
2056 	fd = bpf_map_new_fd(map, f_flags);
2057 	if (fd < 0)
2058 		bpf_map_put_with_uref(map);
2059 
2060 	return fd;
2061 }
2062 
2063 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2064 					      unsigned long addr)
2065 {
2066 	int i;
2067 
2068 	for (i = 0; i < prog->aux->used_map_cnt; i++)
2069 		if (prog->aux->used_maps[i] == (void *)addr)
2070 			return prog->aux->used_maps[i];
2071 	return NULL;
2072 }
2073 
2074 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2075 {
2076 	const struct bpf_map *map;
2077 	struct bpf_insn *insns;
2078 	u64 imm;
2079 	int i;
2080 
2081 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2082 			GFP_USER);
2083 	if (!insns)
2084 		return insns;
2085 
2086 	for (i = 0; i < prog->len; i++) {
2087 		if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2088 			insns[i].code = BPF_JMP | BPF_CALL;
2089 			insns[i].imm = BPF_FUNC_tail_call;
2090 			/* fall-through */
2091 		}
2092 		if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2093 		    insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2094 			if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2095 				insns[i].code = BPF_JMP | BPF_CALL;
2096 			if (!bpf_dump_raw_ok())
2097 				insns[i].imm = 0;
2098 			continue;
2099 		}
2100 
2101 		if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2102 			continue;
2103 
2104 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2105 		map = bpf_map_from_imm(prog, imm);
2106 		if (map) {
2107 			insns[i].src_reg = BPF_PSEUDO_MAP_FD;
2108 			insns[i].imm = map->id;
2109 			insns[i + 1].imm = 0;
2110 			continue;
2111 		}
2112 	}
2113 
2114 	return insns;
2115 }
2116 
2117 static int set_info_rec_size(struct bpf_prog_info *info)
2118 {
2119 	/*
2120 	 * Ensure info.*_rec_size is the same as kernel expected size
2121 	 *
2122 	 * or
2123 	 *
2124 	 * Only allow zero *_rec_size if both _rec_size and _cnt are
2125 	 * zero.  In this case, the kernel will set the expected
2126 	 * _rec_size back to the info.
2127 	 */
2128 
2129 	if ((info->nr_func_info || info->func_info_rec_size) &&
2130 	    info->func_info_rec_size != sizeof(struct bpf_func_info))
2131 		return -EINVAL;
2132 
2133 	if ((info->nr_line_info || info->line_info_rec_size) &&
2134 	    info->line_info_rec_size != sizeof(struct bpf_line_info))
2135 		return -EINVAL;
2136 
2137 	if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2138 	    info->jited_line_info_rec_size != sizeof(__u64))
2139 		return -EINVAL;
2140 
2141 	info->func_info_rec_size = sizeof(struct bpf_func_info);
2142 	info->line_info_rec_size = sizeof(struct bpf_line_info);
2143 	info->jited_line_info_rec_size = sizeof(__u64);
2144 
2145 	return 0;
2146 }
2147 
2148 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2149 				   const union bpf_attr *attr,
2150 				   union bpf_attr __user *uattr)
2151 {
2152 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2153 	struct bpf_prog_info info = {};
2154 	u32 info_len = attr->info.info_len;
2155 	struct bpf_prog_stats stats;
2156 	char __user *uinsns;
2157 	u32 ulen;
2158 	int err;
2159 
2160 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2161 	if (err)
2162 		return err;
2163 	info_len = min_t(u32, sizeof(info), info_len);
2164 
2165 	if (copy_from_user(&info, uinfo, info_len))
2166 		return -EFAULT;
2167 
2168 	info.type = prog->type;
2169 	info.id = prog->aux->id;
2170 	info.load_time = prog->aux->load_time;
2171 	info.created_by_uid = from_kuid_munged(current_user_ns(),
2172 					       prog->aux->user->uid);
2173 	info.gpl_compatible = prog->gpl_compatible;
2174 
2175 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
2176 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2177 
2178 	ulen = info.nr_map_ids;
2179 	info.nr_map_ids = prog->aux->used_map_cnt;
2180 	ulen = min_t(u32, info.nr_map_ids, ulen);
2181 	if (ulen) {
2182 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2183 		u32 i;
2184 
2185 		for (i = 0; i < ulen; i++)
2186 			if (put_user(prog->aux->used_maps[i]->id,
2187 				     &user_map_ids[i]))
2188 				return -EFAULT;
2189 	}
2190 
2191 	err = set_info_rec_size(&info);
2192 	if (err)
2193 		return err;
2194 
2195 	bpf_prog_get_stats(prog, &stats);
2196 	info.run_time_ns = stats.nsecs;
2197 	info.run_cnt = stats.cnt;
2198 
2199 	if (!capable(CAP_SYS_ADMIN)) {
2200 		info.jited_prog_len = 0;
2201 		info.xlated_prog_len = 0;
2202 		info.nr_jited_ksyms = 0;
2203 		info.nr_jited_func_lens = 0;
2204 		info.nr_func_info = 0;
2205 		info.nr_line_info = 0;
2206 		info.nr_jited_line_info = 0;
2207 		goto done;
2208 	}
2209 
2210 	ulen = info.xlated_prog_len;
2211 	info.xlated_prog_len = bpf_prog_insn_size(prog);
2212 	if (info.xlated_prog_len && ulen) {
2213 		struct bpf_insn *insns_sanitized;
2214 		bool fault;
2215 
2216 		if (prog->blinded && !bpf_dump_raw_ok()) {
2217 			info.xlated_prog_insns = 0;
2218 			goto done;
2219 		}
2220 		insns_sanitized = bpf_insn_prepare_dump(prog);
2221 		if (!insns_sanitized)
2222 			return -ENOMEM;
2223 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2224 		ulen = min_t(u32, info.xlated_prog_len, ulen);
2225 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
2226 		kfree(insns_sanitized);
2227 		if (fault)
2228 			return -EFAULT;
2229 	}
2230 
2231 	if (bpf_prog_is_dev_bound(prog->aux)) {
2232 		err = bpf_prog_offload_info_fill(&info, prog);
2233 		if (err)
2234 			return err;
2235 		goto done;
2236 	}
2237 
2238 	/* NOTE: the following code is supposed to be skipped for offload.
2239 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
2240 	 * for offload.
2241 	 */
2242 	ulen = info.jited_prog_len;
2243 	if (prog->aux->func_cnt) {
2244 		u32 i;
2245 
2246 		info.jited_prog_len = 0;
2247 		for (i = 0; i < prog->aux->func_cnt; i++)
2248 			info.jited_prog_len += prog->aux->func[i]->jited_len;
2249 	} else {
2250 		info.jited_prog_len = prog->jited_len;
2251 	}
2252 
2253 	if (info.jited_prog_len && ulen) {
2254 		if (bpf_dump_raw_ok()) {
2255 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
2256 			ulen = min_t(u32, info.jited_prog_len, ulen);
2257 
2258 			/* for multi-function programs, copy the JITed
2259 			 * instructions for all the functions
2260 			 */
2261 			if (prog->aux->func_cnt) {
2262 				u32 len, free, i;
2263 				u8 *img;
2264 
2265 				free = ulen;
2266 				for (i = 0; i < prog->aux->func_cnt; i++) {
2267 					len = prog->aux->func[i]->jited_len;
2268 					len = min_t(u32, len, free);
2269 					img = (u8 *) prog->aux->func[i]->bpf_func;
2270 					if (copy_to_user(uinsns, img, len))
2271 						return -EFAULT;
2272 					uinsns += len;
2273 					free -= len;
2274 					if (!free)
2275 						break;
2276 				}
2277 			} else {
2278 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
2279 					return -EFAULT;
2280 			}
2281 		} else {
2282 			info.jited_prog_insns = 0;
2283 		}
2284 	}
2285 
2286 	ulen = info.nr_jited_ksyms;
2287 	info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2288 	if (ulen) {
2289 		if (bpf_dump_raw_ok()) {
2290 			unsigned long ksym_addr;
2291 			u64 __user *user_ksyms;
2292 			u32 i;
2293 
2294 			/* copy the address of the kernel symbol
2295 			 * corresponding to each function
2296 			 */
2297 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2298 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2299 			if (prog->aux->func_cnt) {
2300 				for (i = 0; i < ulen; i++) {
2301 					ksym_addr = (unsigned long)
2302 						prog->aux->func[i]->bpf_func;
2303 					if (put_user((u64) ksym_addr,
2304 						     &user_ksyms[i]))
2305 						return -EFAULT;
2306 				}
2307 			} else {
2308 				ksym_addr = (unsigned long) prog->bpf_func;
2309 				if (put_user((u64) ksym_addr, &user_ksyms[0]))
2310 					return -EFAULT;
2311 			}
2312 		} else {
2313 			info.jited_ksyms = 0;
2314 		}
2315 	}
2316 
2317 	ulen = info.nr_jited_func_lens;
2318 	info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2319 	if (ulen) {
2320 		if (bpf_dump_raw_ok()) {
2321 			u32 __user *user_lens;
2322 			u32 func_len, i;
2323 
2324 			/* copy the JITed image lengths for each function */
2325 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2326 			user_lens = u64_to_user_ptr(info.jited_func_lens);
2327 			if (prog->aux->func_cnt) {
2328 				for (i = 0; i < ulen; i++) {
2329 					func_len =
2330 						prog->aux->func[i]->jited_len;
2331 					if (put_user(func_len, &user_lens[i]))
2332 						return -EFAULT;
2333 				}
2334 			} else {
2335 				func_len = prog->jited_len;
2336 				if (put_user(func_len, &user_lens[0]))
2337 					return -EFAULT;
2338 			}
2339 		} else {
2340 			info.jited_func_lens = 0;
2341 		}
2342 	}
2343 
2344 	if (prog->aux->btf)
2345 		info.btf_id = btf_id(prog->aux->btf);
2346 
2347 	ulen = info.nr_func_info;
2348 	info.nr_func_info = prog->aux->func_info_cnt;
2349 	if (info.nr_func_info && ulen) {
2350 		char __user *user_finfo;
2351 
2352 		user_finfo = u64_to_user_ptr(info.func_info);
2353 		ulen = min_t(u32, info.nr_func_info, ulen);
2354 		if (copy_to_user(user_finfo, prog->aux->func_info,
2355 				 info.func_info_rec_size * ulen))
2356 			return -EFAULT;
2357 	}
2358 
2359 	ulen = info.nr_line_info;
2360 	info.nr_line_info = prog->aux->nr_linfo;
2361 	if (info.nr_line_info && ulen) {
2362 		__u8 __user *user_linfo;
2363 
2364 		user_linfo = u64_to_user_ptr(info.line_info);
2365 		ulen = min_t(u32, info.nr_line_info, ulen);
2366 		if (copy_to_user(user_linfo, prog->aux->linfo,
2367 				 info.line_info_rec_size * ulen))
2368 			return -EFAULT;
2369 	}
2370 
2371 	ulen = info.nr_jited_line_info;
2372 	if (prog->aux->jited_linfo)
2373 		info.nr_jited_line_info = prog->aux->nr_linfo;
2374 	else
2375 		info.nr_jited_line_info = 0;
2376 	if (info.nr_jited_line_info && ulen) {
2377 		if (bpf_dump_raw_ok()) {
2378 			__u64 __user *user_linfo;
2379 			u32 i;
2380 
2381 			user_linfo = u64_to_user_ptr(info.jited_line_info);
2382 			ulen = min_t(u32, info.nr_jited_line_info, ulen);
2383 			for (i = 0; i < ulen; i++) {
2384 				if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2385 					     &user_linfo[i]))
2386 					return -EFAULT;
2387 			}
2388 		} else {
2389 			info.jited_line_info = 0;
2390 		}
2391 	}
2392 
2393 	ulen = info.nr_prog_tags;
2394 	info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2395 	if (ulen) {
2396 		__u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2397 		u32 i;
2398 
2399 		user_prog_tags = u64_to_user_ptr(info.prog_tags);
2400 		ulen = min_t(u32, info.nr_prog_tags, ulen);
2401 		if (prog->aux->func_cnt) {
2402 			for (i = 0; i < ulen; i++) {
2403 				if (copy_to_user(user_prog_tags[i],
2404 						 prog->aux->func[i]->tag,
2405 						 BPF_TAG_SIZE))
2406 					return -EFAULT;
2407 			}
2408 		} else {
2409 			if (copy_to_user(user_prog_tags[0],
2410 					 prog->tag, BPF_TAG_SIZE))
2411 				return -EFAULT;
2412 		}
2413 	}
2414 
2415 done:
2416 	if (copy_to_user(uinfo, &info, info_len) ||
2417 	    put_user(info_len, &uattr->info.info_len))
2418 		return -EFAULT;
2419 
2420 	return 0;
2421 }
2422 
2423 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2424 				  const union bpf_attr *attr,
2425 				  union bpf_attr __user *uattr)
2426 {
2427 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2428 	struct bpf_map_info info = {};
2429 	u32 info_len = attr->info.info_len;
2430 	int err;
2431 
2432 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2433 	if (err)
2434 		return err;
2435 	info_len = min_t(u32, sizeof(info), info_len);
2436 
2437 	info.type = map->map_type;
2438 	info.id = map->id;
2439 	info.key_size = map->key_size;
2440 	info.value_size = map->value_size;
2441 	info.max_entries = map->max_entries;
2442 	info.map_flags = map->map_flags;
2443 	memcpy(info.name, map->name, sizeof(map->name));
2444 
2445 	if (map->btf) {
2446 		info.btf_id = btf_id(map->btf);
2447 		info.btf_key_type_id = map->btf_key_type_id;
2448 		info.btf_value_type_id = map->btf_value_type_id;
2449 	}
2450 
2451 	if (bpf_map_is_dev_bound(map)) {
2452 		err = bpf_map_offload_info_fill(&info, map);
2453 		if (err)
2454 			return err;
2455 	}
2456 
2457 	if (copy_to_user(uinfo, &info, info_len) ||
2458 	    put_user(info_len, &uattr->info.info_len))
2459 		return -EFAULT;
2460 
2461 	return 0;
2462 }
2463 
2464 static int bpf_btf_get_info_by_fd(struct btf *btf,
2465 				  const union bpf_attr *attr,
2466 				  union bpf_attr __user *uattr)
2467 {
2468 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2469 	u32 info_len = attr->info.info_len;
2470 	int err;
2471 
2472 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2473 	if (err)
2474 		return err;
2475 
2476 	return btf_get_info_by_fd(btf, attr, uattr);
2477 }
2478 
2479 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2480 
2481 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2482 				  union bpf_attr __user *uattr)
2483 {
2484 	int ufd = attr->info.bpf_fd;
2485 	struct fd f;
2486 	int err;
2487 
2488 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2489 		return -EINVAL;
2490 
2491 	f = fdget(ufd);
2492 	if (!f.file)
2493 		return -EBADFD;
2494 
2495 	if (f.file->f_op == &bpf_prog_fops)
2496 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2497 					      uattr);
2498 	else if (f.file->f_op == &bpf_map_fops)
2499 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2500 					     uattr);
2501 	else if (f.file->f_op == &btf_fops)
2502 		err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2503 	else
2504 		err = -EINVAL;
2505 
2506 	fdput(f);
2507 	return err;
2508 }
2509 
2510 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2511 
2512 static int bpf_btf_load(const union bpf_attr *attr)
2513 {
2514 	if (CHECK_ATTR(BPF_BTF_LOAD))
2515 		return -EINVAL;
2516 
2517 	if (!capable(CAP_SYS_ADMIN))
2518 		return -EPERM;
2519 
2520 	return btf_new_fd(attr);
2521 }
2522 
2523 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2524 
2525 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2526 {
2527 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2528 		return -EINVAL;
2529 
2530 	if (!capable(CAP_SYS_ADMIN))
2531 		return -EPERM;
2532 
2533 	return btf_get_fd_by_id(attr->btf_id);
2534 }
2535 
2536 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2537 				    union bpf_attr __user *uattr,
2538 				    u32 prog_id, u32 fd_type,
2539 				    const char *buf, u64 probe_offset,
2540 				    u64 probe_addr)
2541 {
2542 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2543 	u32 len = buf ? strlen(buf) : 0, input_len;
2544 	int err = 0;
2545 
2546 	if (put_user(len, &uattr->task_fd_query.buf_len))
2547 		return -EFAULT;
2548 	input_len = attr->task_fd_query.buf_len;
2549 	if (input_len && ubuf) {
2550 		if (!len) {
2551 			/* nothing to copy, just make ubuf NULL terminated */
2552 			char zero = '\0';
2553 
2554 			if (put_user(zero, ubuf))
2555 				return -EFAULT;
2556 		} else if (input_len >= len + 1) {
2557 			/* ubuf can hold the string with NULL terminator */
2558 			if (copy_to_user(ubuf, buf, len + 1))
2559 				return -EFAULT;
2560 		} else {
2561 			/* ubuf cannot hold the string with NULL terminator,
2562 			 * do a partial copy with NULL terminator.
2563 			 */
2564 			char zero = '\0';
2565 
2566 			err = -ENOSPC;
2567 			if (copy_to_user(ubuf, buf, input_len - 1))
2568 				return -EFAULT;
2569 			if (put_user(zero, ubuf + input_len - 1))
2570 				return -EFAULT;
2571 		}
2572 	}
2573 
2574 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2575 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2576 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2577 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2578 		return -EFAULT;
2579 
2580 	return err;
2581 }
2582 
2583 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2584 
2585 static int bpf_task_fd_query(const union bpf_attr *attr,
2586 			     union bpf_attr __user *uattr)
2587 {
2588 	pid_t pid = attr->task_fd_query.pid;
2589 	u32 fd = attr->task_fd_query.fd;
2590 	const struct perf_event *event;
2591 	struct files_struct *files;
2592 	struct task_struct *task;
2593 	struct file *file;
2594 	int err;
2595 
2596 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2597 		return -EINVAL;
2598 
2599 	if (!capable(CAP_SYS_ADMIN))
2600 		return -EPERM;
2601 
2602 	if (attr->task_fd_query.flags != 0)
2603 		return -EINVAL;
2604 
2605 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2606 	if (!task)
2607 		return -ENOENT;
2608 
2609 	files = get_files_struct(task);
2610 	put_task_struct(task);
2611 	if (!files)
2612 		return -ENOENT;
2613 
2614 	err = 0;
2615 	spin_lock(&files->file_lock);
2616 	file = fcheck_files(files, fd);
2617 	if (!file)
2618 		err = -EBADF;
2619 	else
2620 		get_file(file);
2621 	spin_unlock(&files->file_lock);
2622 	put_files_struct(files);
2623 
2624 	if (err)
2625 		goto out;
2626 
2627 	if (file->f_op == &bpf_raw_tp_fops) {
2628 		struct bpf_raw_tracepoint *raw_tp = file->private_data;
2629 		struct bpf_raw_event_map *btp = raw_tp->btp;
2630 
2631 		err = bpf_task_fd_query_copy(attr, uattr,
2632 					     raw_tp->prog->aux->id,
2633 					     BPF_FD_TYPE_RAW_TRACEPOINT,
2634 					     btp->tp->name, 0, 0);
2635 		goto put_file;
2636 	}
2637 
2638 	event = perf_get_event(file);
2639 	if (!IS_ERR(event)) {
2640 		u64 probe_offset, probe_addr;
2641 		u32 prog_id, fd_type;
2642 		const char *buf;
2643 
2644 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2645 					      &buf, &probe_offset,
2646 					      &probe_addr);
2647 		if (!err)
2648 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2649 						     fd_type, buf,
2650 						     probe_offset,
2651 						     probe_addr);
2652 		goto put_file;
2653 	}
2654 
2655 	err = -ENOTSUPP;
2656 put_file:
2657 	fput(file);
2658 out:
2659 	return err;
2660 }
2661 
2662 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2663 {
2664 	union bpf_attr attr = {};
2665 	int err;
2666 
2667 	if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2668 		return -EPERM;
2669 
2670 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2671 	if (err)
2672 		return err;
2673 	size = min_t(u32, size, sizeof(attr));
2674 
2675 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
2676 	if (copy_from_user(&attr, uattr, size) != 0)
2677 		return -EFAULT;
2678 
2679 	err = security_bpf(cmd, &attr, size);
2680 	if (err < 0)
2681 		return err;
2682 
2683 	switch (cmd) {
2684 	case BPF_MAP_CREATE:
2685 		err = map_create(&attr);
2686 		break;
2687 	case BPF_MAP_LOOKUP_ELEM:
2688 		err = map_lookup_elem(&attr);
2689 		break;
2690 	case BPF_MAP_UPDATE_ELEM:
2691 		err = map_update_elem(&attr);
2692 		break;
2693 	case BPF_MAP_DELETE_ELEM:
2694 		err = map_delete_elem(&attr);
2695 		break;
2696 	case BPF_MAP_GET_NEXT_KEY:
2697 		err = map_get_next_key(&attr);
2698 		break;
2699 	case BPF_PROG_LOAD:
2700 		err = bpf_prog_load(&attr, uattr);
2701 		break;
2702 	case BPF_OBJ_PIN:
2703 		err = bpf_obj_pin(&attr);
2704 		break;
2705 	case BPF_OBJ_GET:
2706 		err = bpf_obj_get(&attr);
2707 		break;
2708 	case BPF_PROG_ATTACH:
2709 		err = bpf_prog_attach(&attr);
2710 		break;
2711 	case BPF_PROG_DETACH:
2712 		err = bpf_prog_detach(&attr);
2713 		break;
2714 	case BPF_PROG_QUERY:
2715 		err = bpf_prog_query(&attr, uattr);
2716 		break;
2717 	case BPF_PROG_TEST_RUN:
2718 		err = bpf_prog_test_run(&attr, uattr);
2719 		break;
2720 	case BPF_PROG_GET_NEXT_ID:
2721 		err = bpf_obj_get_next_id(&attr, uattr,
2722 					  &prog_idr, &prog_idr_lock);
2723 		break;
2724 	case BPF_MAP_GET_NEXT_ID:
2725 		err = bpf_obj_get_next_id(&attr, uattr,
2726 					  &map_idr, &map_idr_lock);
2727 		break;
2728 	case BPF_PROG_GET_FD_BY_ID:
2729 		err = bpf_prog_get_fd_by_id(&attr);
2730 		break;
2731 	case BPF_MAP_GET_FD_BY_ID:
2732 		err = bpf_map_get_fd_by_id(&attr);
2733 		break;
2734 	case BPF_OBJ_GET_INFO_BY_FD:
2735 		err = bpf_obj_get_info_by_fd(&attr, uattr);
2736 		break;
2737 	case BPF_RAW_TRACEPOINT_OPEN:
2738 		err = bpf_raw_tracepoint_open(&attr);
2739 		break;
2740 	case BPF_BTF_LOAD:
2741 		err = bpf_btf_load(&attr);
2742 		break;
2743 	case BPF_BTF_GET_FD_BY_ID:
2744 		err = bpf_btf_get_fd_by_id(&attr);
2745 		break;
2746 	case BPF_TASK_FD_QUERY:
2747 		err = bpf_task_fd_query(&attr, uattr);
2748 		break;
2749 	case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2750 		err = map_lookup_and_delete_elem(&attr);
2751 		break;
2752 	default:
2753 		err = -EINVAL;
2754 		break;
2755 	}
2756 
2757 	return err;
2758 }
2759