xref: /openbmc/linux/kernel/bpf/syscall.c (revision 3557b3fd)
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 really just want to fail instead of triggering OOM killer
140 	 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
141 	 * which is used for lower order allocation requests.
142 	 *
143 	 * It has been observed that higher order allocation requests done by
144 	 * vmalloc with __GFP_NORETRY being set might fail due to not trying
145 	 * to reclaim memory from the page cache, thus we set
146 	 * __GFP_RETRY_MAYFAIL to avoid such situations.
147 	 */
148 
149 	const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
150 	void *area;
151 
152 	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
153 		area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
154 				    numa_node);
155 		if (area != NULL)
156 			return area;
157 	}
158 
159 	return __vmalloc_node_flags_caller(size, numa_node,
160 					   GFP_KERNEL | __GFP_RETRY_MAYFAIL |
161 					   flags, __builtin_return_address(0));
162 }
163 
164 void bpf_map_area_free(void *area)
165 {
166 	kvfree(area);
167 }
168 
169 static u32 bpf_map_flags_retain_permanent(u32 flags)
170 {
171 	/* Some map creation flags are not tied to the map object but
172 	 * rather to the map fd instead, so they have no meaning upon
173 	 * map object inspection since multiple file descriptors with
174 	 * different (access) properties can exist here. Thus, given
175 	 * this has zero meaning for the map itself, lets clear these
176 	 * from here.
177 	 */
178 	return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
179 }
180 
181 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
182 {
183 	map->map_type = attr->map_type;
184 	map->key_size = attr->key_size;
185 	map->value_size = attr->value_size;
186 	map->max_entries = attr->max_entries;
187 	map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
188 	map->numa_node = bpf_map_attr_numa_node(attr);
189 }
190 
191 int bpf_map_precharge_memlock(u32 pages)
192 {
193 	struct user_struct *user = get_current_user();
194 	unsigned long memlock_limit, cur;
195 
196 	memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
197 	cur = atomic_long_read(&user->locked_vm);
198 	free_uid(user);
199 	if (cur + pages > memlock_limit)
200 		return -EPERM;
201 	return 0;
202 }
203 
204 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
205 {
206 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
207 
208 	if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
209 		atomic_long_sub(pages, &user->locked_vm);
210 		return -EPERM;
211 	}
212 	return 0;
213 }
214 
215 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
216 {
217 	atomic_long_sub(pages, &user->locked_vm);
218 }
219 
220 static int bpf_map_init_memlock(struct bpf_map *map)
221 {
222 	struct user_struct *user = get_current_user();
223 	int ret;
224 
225 	ret = bpf_charge_memlock(user, map->pages);
226 	if (ret) {
227 		free_uid(user);
228 		return ret;
229 	}
230 	map->user = user;
231 	return ret;
232 }
233 
234 static void bpf_map_release_memlock(struct bpf_map *map)
235 {
236 	struct user_struct *user = map->user;
237 	bpf_uncharge_memlock(user, map->pages);
238 	free_uid(user);
239 }
240 
241 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
242 {
243 	int ret;
244 
245 	ret = bpf_charge_memlock(map->user, pages);
246 	if (ret)
247 		return ret;
248 	map->pages += pages;
249 	return ret;
250 }
251 
252 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
253 {
254 	bpf_uncharge_memlock(map->user, pages);
255 	map->pages -= pages;
256 }
257 
258 static int bpf_map_alloc_id(struct bpf_map *map)
259 {
260 	int id;
261 
262 	idr_preload(GFP_KERNEL);
263 	spin_lock_bh(&map_idr_lock);
264 	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
265 	if (id > 0)
266 		map->id = id;
267 	spin_unlock_bh(&map_idr_lock);
268 	idr_preload_end();
269 
270 	if (WARN_ON_ONCE(!id))
271 		return -ENOSPC;
272 
273 	return id > 0 ? 0 : id;
274 }
275 
276 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
277 {
278 	unsigned long flags;
279 
280 	/* Offloaded maps are removed from the IDR store when their device
281 	 * disappears - even if someone holds an fd to them they are unusable,
282 	 * the memory is gone, all ops will fail; they are simply waiting for
283 	 * refcnt to drop to be freed.
284 	 */
285 	if (!map->id)
286 		return;
287 
288 	if (do_idr_lock)
289 		spin_lock_irqsave(&map_idr_lock, flags);
290 	else
291 		__acquire(&map_idr_lock);
292 
293 	idr_remove(&map_idr, map->id);
294 	map->id = 0;
295 
296 	if (do_idr_lock)
297 		spin_unlock_irqrestore(&map_idr_lock, flags);
298 	else
299 		__release(&map_idr_lock);
300 }
301 
302 /* called from workqueue */
303 static void bpf_map_free_deferred(struct work_struct *work)
304 {
305 	struct bpf_map *map = container_of(work, struct bpf_map, work);
306 
307 	bpf_map_release_memlock(map);
308 	security_bpf_map_free(map);
309 	/* implementation dependent freeing */
310 	map->ops->map_free(map);
311 }
312 
313 static void bpf_map_put_uref(struct bpf_map *map)
314 {
315 	if (atomic_dec_and_test(&map->usercnt)) {
316 		if (map->ops->map_release_uref)
317 			map->ops->map_release_uref(map);
318 	}
319 }
320 
321 /* decrement map refcnt and schedule it for freeing via workqueue
322  * (unrelying map implementation ops->map_free() might sleep)
323  */
324 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
325 {
326 	if (atomic_dec_and_test(&map->refcnt)) {
327 		/* bpf_map_free_id() must be called first */
328 		bpf_map_free_id(map, do_idr_lock);
329 		btf_put(map->btf);
330 		INIT_WORK(&map->work, bpf_map_free_deferred);
331 		schedule_work(&map->work);
332 	}
333 }
334 
335 void bpf_map_put(struct bpf_map *map)
336 {
337 	__bpf_map_put(map, true);
338 }
339 EXPORT_SYMBOL_GPL(bpf_map_put);
340 
341 void bpf_map_put_with_uref(struct bpf_map *map)
342 {
343 	bpf_map_put_uref(map);
344 	bpf_map_put(map);
345 }
346 
347 static int bpf_map_release(struct inode *inode, struct file *filp)
348 {
349 	struct bpf_map *map = filp->private_data;
350 
351 	if (map->ops->map_release)
352 		map->ops->map_release(map, filp);
353 
354 	bpf_map_put_with_uref(map);
355 	return 0;
356 }
357 
358 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
359 {
360 	fmode_t mode = f.file->f_mode;
361 
362 	/* Our file permissions may have been overridden by global
363 	 * map permissions facing syscall side.
364 	 */
365 	if (READ_ONCE(map->frozen))
366 		mode &= ~FMODE_CAN_WRITE;
367 	return mode;
368 }
369 
370 #ifdef CONFIG_PROC_FS
371 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
372 {
373 	const struct bpf_map *map = filp->private_data;
374 	const struct bpf_array *array;
375 	u32 owner_prog_type = 0;
376 	u32 owner_jited = 0;
377 
378 	if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
379 		array = container_of(map, struct bpf_array, map);
380 		owner_prog_type = array->owner_prog_type;
381 		owner_jited = array->owner_jited;
382 	}
383 
384 	seq_printf(m,
385 		   "map_type:\t%u\n"
386 		   "key_size:\t%u\n"
387 		   "value_size:\t%u\n"
388 		   "max_entries:\t%u\n"
389 		   "map_flags:\t%#x\n"
390 		   "memlock:\t%llu\n"
391 		   "map_id:\t%u\n"
392 		   "frozen:\t%u\n",
393 		   map->map_type,
394 		   map->key_size,
395 		   map->value_size,
396 		   map->max_entries,
397 		   map->map_flags,
398 		   map->pages * 1ULL << PAGE_SHIFT,
399 		   map->id,
400 		   READ_ONCE(map->frozen));
401 
402 	if (owner_prog_type) {
403 		seq_printf(m, "owner_prog_type:\t%u\n",
404 			   owner_prog_type);
405 		seq_printf(m, "owner_jited:\t%u\n",
406 			   owner_jited);
407 	}
408 }
409 #endif
410 
411 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
412 			      loff_t *ppos)
413 {
414 	/* We need this handler such that alloc_file() enables
415 	 * f_mode with FMODE_CAN_READ.
416 	 */
417 	return -EINVAL;
418 }
419 
420 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
421 			       size_t siz, loff_t *ppos)
422 {
423 	/* We need this handler such that alloc_file() enables
424 	 * f_mode with FMODE_CAN_WRITE.
425 	 */
426 	return -EINVAL;
427 }
428 
429 const struct file_operations bpf_map_fops = {
430 #ifdef CONFIG_PROC_FS
431 	.show_fdinfo	= bpf_map_show_fdinfo,
432 #endif
433 	.release	= bpf_map_release,
434 	.read		= bpf_dummy_read,
435 	.write		= bpf_dummy_write,
436 };
437 
438 int bpf_map_new_fd(struct bpf_map *map, int flags)
439 {
440 	int ret;
441 
442 	ret = security_bpf_map(map, OPEN_FMODE(flags));
443 	if (ret < 0)
444 		return ret;
445 
446 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
447 				flags | O_CLOEXEC);
448 }
449 
450 int bpf_get_file_flag(int flags)
451 {
452 	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
453 		return -EINVAL;
454 	if (flags & BPF_F_RDONLY)
455 		return O_RDONLY;
456 	if (flags & BPF_F_WRONLY)
457 		return O_WRONLY;
458 	return O_RDWR;
459 }
460 
461 /* helper macro to check that unused fields 'union bpf_attr' are zero */
462 #define CHECK_ATTR(CMD) \
463 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
464 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
465 		   sizeof(*attr) - \
466 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
467 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
468 
469 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
470  * Return 0 on success and < 0 on error.
471  */
472 static int bpf_obj_name_cpy(char *dst, const char *src)
473 {
474 	const char *end = src + BPF_OBJ_NAME_LEN;
475 
476 	memset(dst, 0, BPF_OBJ_NAME_LEN);
477 	/* Copy all isalnum(), '_' and '.' chars. */
478 	while (src < end && *src) {
479 		if (!isalnum(*src) &&
480 		    *src != '_' && *src != '.')
481 			return -EINVAL;
482 		*dst++ = *src++;
483 	}
484 
485 	/* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
486 	if (src == end)
487 		return -EINVAL;
488 
489 	return 0;
490 }
491 
492 int map_check_no_btf(const struct bpf_map *map,
493 		     const struct btf *btf,
494 		     const struct btf_type *key_type,
495 		     const struct btf_type *value_type)
496 {
497 	return -ENOTSUPP;
498 }
499 
500 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
501 			 u32 btf_key_id, u32 btf_value_id)
502 {
503 	const struct btf_type *key_type, *value_type;
504 	u32 key_size, value_size;
505 	int ret = 0;
506 
507 	/* Some maps allow key to be unspecified. */
508 	if (btf_key_id) {
509 		key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
510 		if (!key_type || key_size != map->key_size)
511 			return -EINVAL;
512 	} else {
513 		key_type = btf_type_by_id(btf, 0);
514 		if (!map->ops->map_check_btf)
515 			return -EINVAL;
516 	}
517 
518 	value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
519 	if (!value_type || value_size != map->value_size)
520 		return -EINVAL;
521 
522 	map->spin_lock_off = btf_find_spin_lock(btf, value_type);
523 
524 	if (map_value_has_spin_lock(map)) {
525 		if (map->map_flags & BPF_F_RDONLY_PROG)
526 			return -EACCES;
527 		if (map->map_type != BPF_MAP_TYPE_HASH &&
528 		    map->map_type != BPF_MAP_TYPE_ARRAY &&
529 		    map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE)
530 			return -ENOTSUPP;
531 		if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
532 		    map->value_size) {
533 			WARN_ONCE(1,
534 				  "verifier bug spin_lock_off %d value_size %d\n",
535 				  map->spin_lock_off, map->value_size);
536 			return -EFAULT;
537 		}
538 	}
539 
540 	if (map->ops->map_check_btf)
541 		ret = map->ops->map_check_btf(map, btf, key_type, value_type);
542 
543 	return ret;
544 }
545 
546 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
547 /* called via syscall */
548 static int map_create(union bpf_attr *attr)
549 {
550 	int numa_node = bpf_map_attr_numa_node(attr);
551 	struct bpf_map *map;
552 	int f_flags;
553 	int err;
554 
555 	err = CHECK_ATTR(BPF_MAP_CREATE);
556 	if (err)
557 		return -EINVAL;
558 
559 	f_flags = bpf_get_file_flag(attr->map_flags);
560 	if (f_flags < 0)
561 		return f_flags;
562 
563 	if (numa_node != NUMA_NO_NODE &&
564 	    ((unsigned int)numa_node >= nr_node_ids ||
565 	     !node_online(numa_node)))
566 		return -EINVAL;
567 
568 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
569 	map = find_and_alloc_map(attr);
570 	if (IS_ERR(map))
571 		return PTR_ERR(map);
572 
573 	err = bpf_obj_name_cpy(map->name, attr->map_name);
574 	if (err)
575 		goto free_map_nouncharge;
576 
577 	atomic_set(&map->refcnt, 1);
578 	atomic_set(&map->usercnt, 1);
579 
580 	if (attr->btf_key_type_id || attr->btf_value_type_id) {
581 		struct btf *btf;
582 
583 		if (!attr->btf_value_type_id) {
584 			err = -EINVAL;
585 			goto free_map_nouncharge;
586 		}
587 
588 		btf = btf_get_by_fd(attr->btf_fd);
589 		if (IS_ERR(btf)) {
590 			err = PTR_ERR(btf);
591 			goto free_map_nouncharge;
592 		}
593 
594 		err = map_check_btf(map, btf, attr->btf_key_type_id,
595 				    attr->btf_value_type_id);
596 		if (err) {
597 			btf_put(btf);
598 			goto free_map_nouncharge;
599 		}
600 
601 		map->btf = btf;
602 		map->btf_key_type_id = attr->btf_key_type_id;
603 		map->btf_value_type_id = attr->btf_value_type_id;
604 	} else {
605 		map->spin_lock_off = -EINVAL;
606 	}
607 
608 	err = security_bpf_map_alloc(map);
609 	if (err)
610 		goto free_map_nouncharge;
611 
612 	err = bpf_map_init_memlock(map);
613 	if (err)
614 		goto free_map_sec;
615 
616 	err = bpf_map_alloc_id(map);
617 	if (err)
618 		goto free_map;
619 
620 	err = bpf_map_new_fd(map, f_flags);
621 	if (err < 0) {
622 		/* failed to allocate fd.
623 		 * bpf_map_put_with_uref() is needed because the above
624 		 * bpf_map_alloc_id() has published the map
625 		 * to the userspace and the userspace may
626 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
627 		 */
628 		bpf_map_put_with_uref(map);
629 		return err;
630 	}
631 
632 	return err;
633 
634 free_map:
635 	bpf_map_release_memlock(map);
636 free_map_sec:
637 	security_bpf_map_free(map);
638 free_map_nouncharge:
639 	btf_put(map->btf);
640 	map->ops->map_free(map);
641 	return err;
642 }
643 
644 /* if error is returned, fd is released.
645  * On success caller should complete fd access with matching fdput()
646  */
647 struct bpf_map *__bpf_map_get(struct fd f)
648 {
649 	if (!f.file)
650 		return ERR_PTR(-EBADF);
651 	if (f.file->f_op != &bpf_map_fops) {
652 		fdput(f);
653 		return ERR_PTR(-EINVAL);
654 	}
655 
656 	return f.file->private_data;
657 }
658 
659 /* prog's and map's refcnt limit */
660 #define BPF_MAX_REFCNT 32768
661 
662 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
663 {
664 	if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
665 		atomic_dec(&map->refcnt);
666 		return ERR_PTR(-EBUSY);
667 	}
668 	if (uref)
669 		atomic_inc(&map->usercnt);
670 	return map;
671 }
672 EXPORT_SYMBOL_GPL(bpf_map_inc);
673 
674 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
675 {
676 	struct fd f = fdget(ufd);
677 	struct bpf_map *map;
678 
679 	map = __bpf_map_get(f);
680 	if (IS_ERR(map))
681 		return map;
682 
683 	map = bpf_map_inc(map, true);
684 	fdput(f);
685 
686 	return map;
687 }
688 
689 /* map_idr_lock should have been held */
690 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
691 					    bool uref)
692 {
693 	int refold;
694 
695 	refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
696 
697 	if (refold >= BPF_MAX_REFCNT) {
698 		__bpf_map_put(map, false);
699 		return ERR_PTR(-EBUSY);
700 	}
701 
702 	if (!refold)
703 		return ERR_PTR(-ENOENT);
704 
705 	if (uref)
706 		atomic_inc(&map->usercnt);
707 
708 	return map;
709 }
710 
711 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
712 {
713 	return -ENOTSUPP;
714 }
715 
716 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
717 {
718 	if (key_size)
719 		return memdup_user(ukey, key_size);
720 
721 	if (ukey)
722 		return ERR_PTR(-EINVAL);
723 
724 	return NULL;
725 }
726 
727 /* last field in 'union bpf_attr' used by this command */
728 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
729 
730 static int map_lookup_elem(union bpf_attr *attr)
731 {
732 	void __user *ukey = u64_to_user_ptr(attr->key);
733 	void __user *uvalue = u64_to_user_ptr(attr->value);
734 	int ufd = attr->map_fd;
735 	struct bpf_map *map;
736 	void *key, *value, *ptr;
737 	u32 value_size;
738 	struct fd f;
739 	int err;
740 
741 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
742 		return -EINVAL;
743 
744 	if (attr->flags & ~BPF_F_LOCK)
745 		return -EINVAL;
746 
747 	f = fdget(ufd);
748 	map = __bpf_map_get(f);
749 	if (IS_ERR(map))
750 		return PTR_ERR(map);
751 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
752 		err = -EPERM;
753 		goto err_put;
754 	}
755 
756 	if ((attr->flags & BPF_F_LOCK) &&
757 	    !map_value_has_spin_lock(map)) {
758 		err = -EINVAL;
759 		goto err_put;
760 	}
761 
762 	key = __bpf_copy_key(ukey, map->key_size);
763 	if (IS_ERR(key)) {
764 		err = PTR_ERR(key);
765 		goto err_put;
766 	}
767 
768 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
769 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
770 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
771 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
772 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
773 	else if (IS_FD_MAP(map))
774 		value_size = sizeof(u32);
775 	else
776 		value_size = map->value_size;
777 
778 	err = -ENOMEM;
779 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
780 	if (!value)
781 		goto free_key;
782 
783 	if (bpf_map_is_dev_bound(map)) {
784 		err = bpf_map_offload_lookup_elem(map, key, value);
785 		goto done;
786 	}
787 
788 	preempt_disable();
789 	this_cpu_inc(bpf_prog_active);
790 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
791 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
792 		err = bpf_percpu_hash_copy(map, key, value);
793 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
794 		err = bpf_percpu_array_copy(map, key, value);
795 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
796 		err = bpf_percpu_cgroup_storage_copy(map, key, value);
797 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
798 		err = bpf_stackmap_copy(map, key, value);
799 	} else if (IS_FD_ARRAY(map)) {
800 		err = bpf_fd_array_map_lookup_elem(map, key, value);
801 	} else if (IS_FD_HASH(map)) {
802 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
803 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
804 		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
805 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
806 		   map->map_type == BPF_MAP_TYPE_STACK) {
807 		err = map->ops->map_peek_elem(map, value);
808 	} else {
809 		rcu_read_lock();
810 		ptr = map->ops->map_lookup_elem(map, key);
811 		if (IS_ERR(ptr)) {
812 			err = PTR_ERR(ptr);
813 		} else if (!ptr) {
814 			err = -ENOENT;
815 		} else {
816 			err = 0;
817 			if (attr->flags & BPF_F_LOCK)
818 				/* lock 'ptr' and copy everything but lock */
819 				copy_map_value_locked(map, value, ptr, true);
820 			else
821 				copy_map_value(map, value, ptr);
822 			/* mask lock, since value wasn't zero inited */
823 			check_and_init_map_lock(map, value);
824 		}
825 		rcu_read_unlock();
826 	}
827 	this_cpu_dec(bpf_prog_active);
828 	preempt_enable();
829 
830 done:
831 	if (err)
832 		goto free_value;
833 
834 	err = -EFAULT;
835 	if (copy_to_user(uvalue, value, value_size) != 0)
836 		goto free_value;
837 
838 	err = 0;
839 
840 free_value:
841 	kfree(value);
842 free_key:
843 	kfree(key);
844 err_put:
845 	fdput(f);
846 	return err;
847 }
848 
849 static void maybe_wait_bpf_programs(struct bpf_map *map)
850 {
851 	/* Wait for any running BPF programs to complete so that
852 	 * userspace, when we return to it, knows that all programs
853 	 * that could be running use the new map value.
854 	 */
855 	if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
856 	    map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
857 		synchronize_rcu();
858 }
859 
860 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
861 
862 static int map_update_elem(union bpf_attr *attr)
863 {
864 	void __user *ukey = u64_to_user_ptr(attr->key);
865 	void __user *uvalue = u64_to_user_ptr(attr->value);
866 	int ufd = attr->map_fd;
867 	struct bpf_map *map;
868 	void *key, *value;
869 	u32 value_size;
870 	struct fd f;
871 	int err;
872 
873 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
874 		return -EINVAL;
875 
876 	f = fdget(ufd);
877 	map = __bpf_map_get(f);
878 	if (IS_ERR(map))
879 		return PTR_ERR(map);
880 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
881 		err = -EPERM;
882 		goto err_put;
883 	}
884 
885 	if ((attr->flags & BPF_F_LOCK) &&
886 	    !map_value_has_spin_lock(map)) {
887 		err = -EINVAL;
888 		goto err_put;
889 	}
890 
891 	key = __bpf_copy_key(ukey, map->key_size);
892 	if (IS_ERR(key)) {
893 		err = PTR_ERR(key);
894 		goto err_put;
895 	}
896 
897 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
898 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
899 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
900 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
901 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
902 	else
903 		value_size = map->value_size;
904 
905 	err = -ENOMEM;
906 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
907 	if (!value)
908 		goto free_key;
909 
910 	err = -EFAULT;
911 	if (copy_from_user(value, uvalue, value_size) != 0)
912 		goto free_value;
913 
914 	/* Need to create a kthread, thus must support schedule */
915 	if (bpf_map_is_dev_bound(map)) {
916 		err = bpf_map_offload_update_elem(map, key, value, attr->flags);
917 		goto out;
918 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
919 		   map->map_type == BPF_MAP_TYPE_SOCKHASH ||
920 		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
921 		err = map->ops->map_update_elem(map, key, value, attr->flags);
922 		goto out;
923 	}
924 
925 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
926 	 * inside bpf map update or delete otherwise deadlocks are possible
927 	 */
928 	preempt_disable();
929 	__this_cpu_inc(bpf_prog_active);
930 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
931 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
932 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
933 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
934 		err = bpf_percpu_array_update(map, key, value, attr->flags);
935 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
936 		err = bpf_percpu_cgroup_storage_update(map, key, value,
937 						       attr->flags);
938 	} else if (IS_FD_ARRAY(map)) {
939 		rcu_read_lock();
940 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
941 						   attr->flags);
942 		rcu_read_unlock();
943 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
944 		rcu_read_lock();
945 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
946 						  attr->flags);
947 		rcu_read_unlock();
948 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
949 		/* rcu_read_lock() is not needed */
950 		err = bpf_fd_reuseport_array_update_elem(map, key, value,
951 							 attr->flags);
952 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
953 		   map->map_type == BPF_MAP_TYPE_STACK) {
954 		err = map->ops->map_push_elem(map, value, attr->flags);
955 	} else {
956 		rcu_read_lock();
957 		err = map->ops->map_update_elem(map, key, value, attr->flags);
958 		rcu_read_unlock();
959 	}
960 	__this_cpu_dec(bpf_prog_active);
961 	preempt_enable();
962 	maybe_wait_bpf_programs(map);
963 out:
964 free_value:
965 	kfree(value);
966 free_key:
967 	kfree(key);
968 err_put:
969 	fdput(f);
970 	return err;
971 }
972 
973 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
974 
975 static int map_delete_elem(union bpf_attr *attr)
976 {
977 	void __user *ukey = u64_to_user_ptr(attr->key);
978 	int ufd = attr->map_fd;
979 	struct bpf_map *map;
980 	struct fd f;
981 	void *key;
982 	int err;
983 
984 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
985 		return -EINVAL;
986 
987 	f = fdget(ufd);
988 	map = __bpf_map_get(f);
989 	if (IS_ERR(map))
990 		return PTR_ERR(map);
991 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
992 		err = -EPERM;
993 		goto err_put;
994 	}
995 
996 	key = __bpf_copy_key(ukey, map->key_size);
997 	if (IS_ERR(key)) {
998 		err = PTR_ERR(key);
999 		goto err_put;
1000 	}
1001 
1002 	if (bpf_map_is_dev_bound(map)) {
1003 		err = bpf_map_offload_delete_elem(map, key);
1004 		goto out;
1005 	}
1006 
1007 	preempt_disable();
1008 	__this_cpu_inc(bpf_prog_active);
1009 	rcu_read_lock();
1010 	err = map->ops->map_delete_elem(map, key);
1011 	rcu_read_unlock();
1012 	__this_cpu_dec(bpf_prog_active);
1013 	preempt_enable();
1014 	maybe_wait_bpf_programs(map);
1015 out:
1016 	kfree(key);
1017 err_put:
1018 	fdput(f);
1019 	return err;
1020 }
1021 
1022 /* last field in 'union bpf_attr' used by this command */
1023 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1024 
1025 static int map_get_next_key(union bpf_attr *attr)
1026 {
1027 	void __user *ukey = u64_to_user_ptr(attr->key);
1028 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
1029 	int ufd = attr->map_fd;
1030 	struct bpf_map *map;
1031 	void *key, *next_key;
1032 	struct fd f;
1033 	int err;
1034 
1035 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1036 		return -EINVAL;
1037 
1038 	f = fdget(ufd);
1039 	map = __bpf_map_get(f);
1040 	if (IS_ERR(map))
1041 		return PTR_ERR(map);
1042 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1043 		err = -EPERM;
1044 		goto err_put;
1045 	}
1046 
1047 	if (ukey) {
1048 		key = __bpf_copy_key(ukey, map->key_size);
1049 		if (IS_ERR(key)) {
1050 			err = PTR_ERR(key);
1051 			goto err_put;
1052 		}
1053 	} else {
1054 		key = NULL;
1055 	}
1056 
1057 	err = -ENOMEM;
1058 	next_key = kmalloc(map->key_size, GFP_USER);
1059 	if (!next_key)
1060 		goto free_key;
1061 
1062 	if (bpf_map_is_dev_bound(map)) {
1063 		err = bpf_map_offload_get_next_key(map, key, next_key);
1064 		goto out;
1065 	}
1066 
1067 	rcu_read_lock();
1068 	err = map->ops->map_get_next_key(map, key, next_key);
1069 	rcu_read_unlock();
1070 out:
1071 	if (err)
1072 		goto free_next_key;
1073 
1074 	err = -EFAULT;
1075 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1076 		goto free_next_key;
1077 
1078 	err = 0;
1079 
1080 free_next_key:
1081 	kfree(next_key);
1082 free_key:
1083 	kfree(key);
1084 err_put:
1085 	fdput(f);
1086 	return err;
1087 }
1088 
1089 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1090 
1091 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1092 {
1093 	void __user *ukey = u64_to_user_ptr(attr->key);
1094 	void __user *uvalue = u64_to_user_ptr(attr->value);
1095 	int ufd = attr->map_fd;
1096 	struct bpf_map *map;
1097 	void *key, *value;
1098 	u32 value_size;
1099 	struct fd f;
1100 	int err;
1101 
1102 	if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1103 		return -EINVAL;
1104 
1105 	f = fdget(ufd);
1106 	map = __bpf_map_get(f);
1107 	if (IS_ERR(map))
1108 		return PTR_ERR(map);
1109 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1110 		err = -EPERM;
1111 		goto err_put;
1112 	}
1113 
1114 	key = __bpf_copy_key(ukey, map->key_size);
1115 	if (IS_ERR(key)) {
1116 		err = PTR_ERR(key);
1117 		goto err_put;
1118 	}
1119 
1120 	value_size = map->value_size;
1121 
1122 	err = -ENOMEM;
1123 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1124 	if (!value)
1125 		goto free_key;
1126 
1127 	if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1128 	    map->map_type == BPF_MAP_TYPE_STACK) {
1129 		err = map->ops->map_pop_elem(map, value);
1130 	} else {
1131 		err = -ENOTSUPP;
1132 	}
1133 
1134 	if (err)
1135 		goto free_value;
1136 
1137 	if (copy_to_user(uvalue, value, value_size) != 0)
1138 		goto free_value;
1139 
1140 	err = 0;
1141 
1142 free_value:
1143 	kfree(value);
1144 free_key:
1145 	kfree(key);
1146 err_put:
1147 	fdput(f);
1148 	return err;
1149 }
1150 
1151 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1152 
1153 static int map_freeze(const union bpf_attr *attr)
1154 {
1155 	int err = 0, ufd = attr->map_fd;
1156 	struct bpf_map *map;
1157 	struct fd f;
1158 
1159 	if (CHECK_ATTR(BPF_MAP_FREEZE))
1160 		return -EINVAL;
1161 
1162 	f = fdget(ufd);
1163 	map = __bpf_map_get(f);
1164 	if (IS_ERR(map))
1165 		return PTR_ERR(map);
1166 	if (READ_ONCE(map->frozen)) {
1167 		err = -EBUSY;
1168 		goto err_put;
1169 	}
1170 	if (!capable(CAP_SYS_ADMIN)) {
1171 		err = -EPERM;
1172 		goto err_put;
1173 	}
1174 
1175 	WRITE_ONCE(map->frozen, true);
1176 err_put:
1177 	fdput(f);
1178 	return err;
1179 }
1180 
1181 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1182 #define BPF_PROG_TYPE(_id, _name) \
1183 	[_id] = & _name ## _prog_ops,
1184 #define BPF_MAP_TYPE(_id, _ops)
1185 #include <linux/bpf_types.h>
1186 #undef BPF_PROG_TYPE
1187 #undef BPF_MAP_TYPE
1188 };
1189 
1190 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1191 {
1192 	const struct bpf_prog_ops *ops;
1193 
1194 	if (type >= ARRAY_SIZE(bpf_prog_types))
1195 		return -EINVAL;
1196 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1197 	ops = bpf_prog_types[type];
1198 	if (!ops)
1199 		return -EINVAL;
1200 
1201 	if (!bpf_prog_is_dev_bound(prog->aux))
1202 		prog->aux->ops = ops;
1203 	else
1204 		prog->aux->ops = &bpf_offload_prog_ops;
1205 	prog->type = type;
1206 	return 0;
1207 }
1208 
1209 /* drop refcnt on maps used by eBPF program and free auxilary data */
1210 static void free_used_maps(struct bpf_prog_aux *aux)
1211 {
1212 	enum bpf_cgroup_storage_type stype;
1213 	int i;
1214 
1215 	for_each_cgroup_storage_type(stype) {
1216 		if (!aux->cgroup_storage[stype])
1217 			continue;
1218 		bpf_cgroup_storage_release(aux->prog,
1219 					   aux->cgroup_storage[stype]);
1220 	}
1221 
1222 	for (i = 0; i < aux->used_map_cnt; i++)
1223 		bpf_map_put(aux->used_maps[i]);
1224 
1225 	kfree(aux->used_maps);
1226 }
1227 
1228 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1229 {
1230 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1231 	unsigned long user_bufs;
1232 
1233 	if (user) {
1234 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1235 		if (user_bufs > memlock_limit) {
1236 			atomic_long_sub(pages, &user->locked_vm);
1237 			return -EPERM;
1238 		}
1239 	}
1240 
1241 	return 0;
1242 }
1243 
1244 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1245 {
1246 	if (user)
1247 		atomic_long_sub(pages, &user->locked_vm);
1248 }
1249 
1250 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1251 {
1252 	struct user_struct *user = get_current_user();
1253 	int ret;
1254 
1255 	ret = __bpf_prog_charge(user, prog->pages);
1256 	if (ret) {
1257 		free_uid(user);
1258 		return ret;
1259 	}
1260 
1261 	prog->aux->user = user;
1262 	return 0;
1263 }
1264 
1265 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1266 {
1267 	struct user_struct *user = prog->aux->user;
1268 
1269 	__bpf_prog_uncharge(user, prog->pages);
1270 	free_uid(user);
1271 }
1272 
1273 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1274 {
1275 	int id;
1276 
1277 	idr_preload(GFP_KERNEL);
1278 	spin_lock_bh(&prog_idr_lock);
1279 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1280 	if (id > 0)
1281 		prog->aux->id = id;
1282 	spin_unlock_bh(&prog_idr_lock);
1283 	idr_preload_end();
1284 
1285 	/* id is in [1, INT_MAX) */
1286 	if (WARN_ON_ONCE(!id))
1287 		return -ENOSPC;
1288 
1289 	return id > 0 ? 0 : id;
1290 }
1291 
1292 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1293 {
1294 	/* cBPF to eBPF migrations are currently not in the idr store.
1295 	 * Offloaded programs are removed from the store when their device
1296 	 * disappears - even if someone grabs an fd to them they are unusable,
1297 	 * simply waiting for refcnt to drop to be freed.
1298 	 */
1299 	if (!prog->aux->id)
1300 		return;
1301 
1302 	if (do_idr_lock)
1303 		spin_lock_bh(&prog_idr_lock);
1304 	else
1305 		__acquire(&prog_idr_lock);
1306 
1307 	idr_remove(&prog_idr, prog->aux->id);
1308 	prog->aux->id = 0;
1309 
1310 	if (do_idr_lock)
1311 		spin_unlock_bh(&prog_idr_lock);
1312 	else
1313 		__release(&prog_idr_lock);
1314 }
1315 
1316 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1317 {
1318 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1319 
1320 	free_used_maps(aux);
1321 	bpf_prog_uncharge_memlock(aux->prog);
1322 	security_bpf_prog_free(aux);
1323 	bpf_prog_free(aux->prog);
1324 }
1325 
1326 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1327 {
1328 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
1329 		perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1330 		/* bpf_prog_free_id() must be called first */
1331 		bpf_prog_free_id(prog, do_idr_lock);
1332 		bpf_prog_kallsyms_del_all(prog);
1333 		btf_put(prog->aux->btf);
1334 		kvfree(prog->aux->func_info);
1335 		bpf_prog_free_linfo(prog);
1336 
1337 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1338 	}
1339 }
1340 
1341 void bpf_prog_put(struct bpf_prog *prog)
1342 {
1343 	__bpf_prog_put(prog, true);
1344 }
1345 EXPORT_SYMBOL_GPL(bpf_prog_put);
1346 
1347 static int bpf_prog_release(struct inode *inode, struct file *filp)
1348 {
1349 	struct bpf_prog *prog = filp->private_data;
1350 
1351 	bpf_prog_put(prog);
1352 	return 0;
1353 }
1354 
1355 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1356 			       struct bpf_prog_stats *stats)
1357 {
1358 	u64 nsecs = 0, cnt = 0;
1359 	int cpu;
1360 
1361 	for_each_possible_cpu(cpu) {
1362 		const struct bpf_prog_stats *st;
1363 		unsigned int start;
1364 		u64 tnsecs, tcnt;
1365 
1366 		st = per_cpu_ptr(prog->aux->stats, cpu);
1367 		do {
1368 			start = u64_stats_fetch_begin_irq(&st->syncp);
1369 			tnsecs = st->nsecs;
1370 			tcnt = st->cnt;
1371 		} while (u64_stats_fetch_retry_irq(&st->syncp, start));
1372 		nsecs += tnsecs;
1373 		cnt += tcnt;
1374 	}
1375 	stats->nsecs = nsecs;
1376 	stats->cnt = cnt;
1377 }
1378 
1379 #ifdef CONFIG_PROC_FS
1380 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1381 {
1382 	const struct bpf_prog *prog = filp->private_data;
1383 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1384 	struct bpf_prog_stats stats;
1385 
1386 	bpf_prog_get_stats(prog, &stats);
1387 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1388 	seq_printf(m,
1389 		   "prog_type:\t%u\n"
1390 		   "prog_jited:\t%u\n"
1391 		   "prog_tag:\t%s\n"
1392 		   "memlock:\t%llu\n"
1393 		   "prog_id:\t%u\n"
1394 		   "run_time_ns:\t%llu\n"
1395 		   "run_cnt:\t%llu\n",
1396 		   prog->type,
1397 		   prog->jited,
1398 		   prog_tag,
1399 		   prog->pages * 1ULL << PAGE_SHIFT,
1400 		   prog->aux->id,
1401 		   stats.nsecs,
1402 		   stats.cnt);
1403 }
1404 #endif
1405 
1406 const struct file_operations bpf_prog_fops = {
1407 #ifdef CONFIG_PROC_FS
1408 	.show_fdinfo	= bpf_prog_show_fdinfo,
1409 #endif
1410 	.release	= bpf_prog_release,
1411 	.read		= bpf_dummy_read,
1412 	.write		= bpf_dummy_write,
1413 };
1414 
1415 int bpf_prog_new_fd(struct bpf_prog *prog)
1416 {
1417 	int ret;
1418 
1419 	ret = security_bpf_prog(prog);
1420 	if (ret < 0)
1421 		return ret;
1422 
1423 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1424 				O_RDWR | O_CLOEXEC);
1425 }
1426 
1427 static struct bpf_prog *____bpf_prog_get(struct fd f)
1428 {
1429 	if (!f.file)
1430 		return ERR_PTR(-EBADF);
1431 	if (f.file->f_op != &bpf_prog_fops) {
1432 		fdput(f);
1433 		return ERR_PTR(-EINVAL);
1434 	}
1435 
1436 	return f.file->private_data;
1437 }
1438 
1439 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1440 {
1441 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1442 		atomic_sub(i, &prog->aux->refcnt);
1443 		return ERR_PTR(-EBUSY);
1444 	}
1445 	return prog;
1446 }
1447 EXPORT_SYMBOL_GPL(bpf_prog_add);
1448 
1449 void bpf_prog_sub(struct bpf_prog *prog, int i)
1450 {
1451 	/* Only to be used for undoing previous bpf_prog_add() in some
1452 	 * error path. We still know that another entity in our call
1453 	 * path holds a reference to the program, thus atomic_sub() can
1454 	 * be safely used in such cases!
1455 	 */
1456 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1457 }
1458 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1459 
1460 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1461 {
1462 	return bpf_prog_add(prog, 1);
1463 }
1464 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1465 
1466 /* prog_idr_lock should have been held */
1467 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1468 {
1469 	int refold;
1470 
1471 	refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1472 
1473 	if (refold >= BPF_MAX_REFCNT) {
1474 		__bpf_prog_put(prog, false);
1475 		return ERR_PTR(-EBUSY);
1476 	}
1477 
1478 	if (!refold)
1479 		return ERR_PTR(-ENOENT);
1480 
1481 	return prog;
1482 }
1483 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1484 
1485 bool bpf_prog_get_ok(struct bpf_prog *prog,
1486 			    enum bpf_prog_type *attach_type, bool attach_drv)
1487 {
1488 	/* not an attachment, just a refcount inc, always allow */
1489 	if (!attach_type)
1490 		return true;
1491 
1492 	if (prog->type != *attach_type)
1493 		return false;
1494 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1495 		return false;
1496 
1497 	return true;
1498 }
1499 
1500 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1501 				       bool attach_drv)
1502 {
1503 	struct fd f = fdget(ufd);
1504 	struct bpf_prog *prog;
1505 
1506 	prog = ____bpf_prog_get(f);
1507 	if (IS_ERR(prog))
1508 		return prog;
1509 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1510 		prog = ERR_PTR(-EINVAL);
1511 		goto out;
1512 	}
1513 
1514 	prog = bpf_prog_inc(prog);
1515 out:
1516 	fdput(f);
1517 	return prog;
1518 }
1519 
1520 struct bpf_prog *bpf_prog_get(u32 ufd)
1521 {
1522 	return __bpf_prog_get(ufd, NULL, false);
1523 }
1524 
1525 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1526 				       bool attach_drv)
1527 {
1528 	return __bpf_prog_get(ufd, &type, attach_drv);
1529 }
1530 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1531 
1532 /* Initially all BPF programs could be loaded w/o specifying
1533  * expected_attach_type. Later for some of them specifying expected_attach_type
1534  * at load time became required so that program could be validated properly.
1535  * Programs of types that are allowed to be loaded both w/ and w/o (for
1536  * backward compatibility) expected_attach_type, should have the default attach
1537  * type assigned to expected_attach_type for the latter case, so that it can be
1538  * validated later at attach time.
1539  *
1540  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1541  * prog type requires it but has some attach types that have to be backward
1542  * compatible.
1543  */
1544 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1545 {
1546 	switch (attr->prog_type) {
1547 	case BPF_PROG_TYPE_CGROUP_SOCK:
1548 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1549 		 * exist so checking for non-zero is the way to go here.
1550 		 */
1551 		if (!attr->expected_attach_type)
1552 			attr->expected_attach_type =
1553 				BPF_CGROUP_INET_SOCK_CREATE;
1554 		break;
1555 	}
1556 }
1557 
1558 static int
1559 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1560 				enum bpf_attach_type expected_attach_type)
1561 {
1562 	switch (prog_type) {
1563 	case BPF_PROG_TYPE_CGROUP_SOCK:
1564 		switch (expected_attach_type) {
1565 		case BPF_CGROUP_INET_SOCK_CREATE:
1566 		case BPF_CGROUP_INET4_POST_BIND:
1567 		case BPF_CGROUP_INET6_POST_BIND:
1568 			return 0;
1569 		default:
1570 			return -EINVAL;
1571 		}
1572 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1573 		switch (expected_attach_type) {
1574 		case BPF_CGROUP_INET4_BIND:
1575 		case BPF_CGROUP_INET6_BIND:
1576 		case BPF_CGROUP_INET4_CONNECT:
1577 		case BPF_CGROUP_INET6_CONNECT:
1578 		case BPF_CGROUP_UDP4_SENDMSG:
1579 		case BPF_CGROUP_UDP6_SENDMSG:
1580 			return 0;
1581 		default:
1582 			return -EINVAL;
1583 		}
1584 	default:
1585 		return 0;
1586 	}
1587 }
1588 
1589 /* last field in 'union bpf_attr' used by this command */
1590 #define	BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1591 
1592 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1593 {
1594 	enum bpf_prog_type type = attr->prog_type;
1595 	struct bpf_prog *prog;
1596 	int err;
1597 	char license[128];
1598 	bool is_gpl;
1599 
1600 	if (CHECK_ATTR(BPF_PROG_LOAD))
1601 		return -EINVAL;
1602 
1603 	if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT))
1604 		return -EINVAL;
1605 
1606 	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1607 	    (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1608 	    !capable(CAP_SYS_ADMIN))
1609 		return -EPERM;
1610 
1611 	/* copy eBPF program license from user space */
1612 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1613 			      sizeof(license) - 1) < 0)
1614 		return -EFAULT;
1615 	license[sizeof(license) - 1] = 0;
1616 
1617 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1618 	is_gpl = license_is_gpl_compatible(license);
1619 
1620 	if (attr->insn_cnt == 0 ||
1621 	    attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1622 		return -E2BIG;
1623 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1624 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
1625 	    !capable(CAP_SYS_ADMIN))
1626 		return -EPERM;
1627 
1628 	bpf_prog_load_fixup_attach_type(attr);
1629 	if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1630 		return -EINVAL;
1631 
1632 	/* plain bpf_prog allocation */
1633 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1634 	if (!prog)
1635 		return -ENOMEM;
1636 
1637 	prog->expected_attach_type = attr->expected_attach_type;
1638 
1639 	prog->aux->offload_requested = !!attr->prog_ifindex;
1640 
1641 	err = security_bpf_prog_alloc(prog->aux);
1642 	if (err)
1643 		goto free_prog_nouncharge;
1644 
1645 	err = bpf_prog_charge_memlock(prog);
1646 	if (err)
1647 		goto free_prog_sec;
1648 
1649 	prog->len = attr->insn_cnt;
1650 
1651 	err = -EFAULT;
1652 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1653 			   bpf_prog_insn_size(prog)) != 0)
1654 		goto free_prog;
1655 
1656 	prog->orig_prog = NULL;
1657 	prog->jited = 0;
1658 
1659 	atomic_set(&prog->aux->refcnt, 1);
1660 	prog->gpl_compatible = is_gpl ? 1 : 0;
1661 
1662 	if (bpf_prog_is_dev_bound(prog->aux)) {
1663 		err = bpf_prog_offload_init(prog, attr);
1664 		if (err)
1665 			goto free_prog;
1666 	}
1667 
1668 	/* find program type: socket_filter vs tracing_filter */
1669 	err = find_prog_type(type, prog);
1670 	if (err < 0)
1671 		goto free_prog;
1672 
1673 	prog->aux->load_time = ktime_get_boot_ns();
1674 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1675 	if (err)
1676 		goto free_prog;
1677 
1678 	/* run eBPF verifier */
1679 	err = bpf_check(&prog, attr, uattr);
1680 	if (err < 0)
1681 		goto free_used_maps;
1682 
1683 	prog = bpf_prog_select_runtime(prog, &err);
1684 	if (err < 0)
1685 		goto free_used_maps;
1686 
1687 	err = bpf_prog_alloc_id(prog);
1688 	if (err)
1689 		goto free_used_maps;
1690 
1691 	err = bpf_prog_new_fd(prog);
1692 	if (err < 0) {
1693 		/* failed to allocate fd.
1694 		 * bpf_prog_put() is needed because the above
1695 		 * bpf_prog_alloc_id() has published the prog
1696 		 * to the userspace and the userspace may
1697 		 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1698 		 */
1699 		bpf_prog_put(prog);
1700 		return err;
1701 	}
1702 
1703 	bpf_prog_kallsyms_add(prog);
1704 	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1705 	return err;
1706 
1707 free_used_maps:
1708 	bpf_prog_free_linfo(prog);
1709 	kvfree(prog->aux->func_info);
1710 	btf_put(prog->aux->btf);
1711 	bpf_prog_kallsyms_del_subprogs(prog);
1712 	free_used_maps(prog->aux);
1713 free_prog:
1714 	bpf_prog_uncharge_memlock(prog);
1715 free_prog_sec:
1716 	security_bpf_prog_free(prog->aux);
1717 free_prog_nouncharge:
1718 	bpf_prog_free(prog);
1719 	return err;
1720 }
1721 
1722 #define BPF_OBJ_LAST_FIELD file_flags
1723 
1724 static int bpf_obj_pin(const union bpf_attr *attr)
1725 {
1726 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1727 		return -EINVAL;
1728 
1729 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1730 }
1731 
1732 static int bpf_obj_get(const union bpf_attr *attr)
1733 {
1734 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1735 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1736 		return -EINVAL;
1737 
1738 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1739 				attr->file_flags);
1740 }
1741 
1742 struct bpf_raw_tracepoint {
1743 	struct bpf_raw_event_map *btp;
1744 	struct bpf_prog *prog;
1745 };
1746 
1747 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1748 {
1749 	struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1750 
1751 	if (raw_tp->prog) {
1752 		bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1753 		bpf_prog_put(raw_tp->prog);
1754 	}
1755 	bpf_put_raw_tracepoint(raw_tp->btp);
1756 	kfree(raw_tp);
1757 	return 0;
1758 }
1759 
1760 static const struct file_operations bpf_raw_tp_fops = {
1761 	.release	= bpf_raw_tracepoint_release,
1762 	.read		= bpf_dummy_read,
1763 	.write		= bpf_dummy_write,
1764 };
1765 
1766 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1767 
1768 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1769 {
1770 	struct bpf_raw_tracepoint *raw_tp;
1771 	struct bpf_raw_event_map *btp;
1772 	struct bpf_prog *prog;
1773 	char tp_name[128];
1774 	int tp_fd, err;
1775 
1776 	if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1777 			      sizeof(tp_name) - 1) < 0)
1778 		return -EFAULT;
1779 	tp_name[sizeof(tp_name) - 1] = 0;
1780 
1781 	btp = bpf_get_raw_tracepoint(tp_name);
1782 	if (!btp)
1783 		return -ENOENT;
1784 
1785 	raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1786 	if (!raw_tp) {
1787 		err = -ENOMEM;
1788 		goto out_put_btp;
1789 	}
1790 	raw_tp->btp = btp;
1791 
1792 	prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1793 				 BPF_PROG_TYPE_RAW_TRACEPOINT);
1794 	if (IS_ERR(prog)) {
1795 		err = PTR_ERR(prog);
1796 		goto out_free_tp;
1797 	}
1798 
1799 	err = bpf_probe_register(raw_tp->btp, prog);
1800 	if (err)
1801 		goto out_put_prog;
1802 
1803 	raw_tp->prog = prog;
1804 	tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1805 				 O_CLOEXEC);
1806 	if (tp_fd < 0) {
1807 		bpf_probe_unregister(raw_tp->btp, prog);
1808 		err = tp_fd;
1809 		goto out_put_prog;
1810 	}
1811 	return tp_fd;
1812 
1813 out_put_prog:
1814 	bpf_prog_put(prog);
1815 out_free_tp:
1816 	kfree(raw_tp);
1817 out_put_btp:
1818 	bpf_put_raw_tracepoint(btp);
1819 	return err;
1820 }
1821 
1822 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1823 					     enum bpf_attach_type attach_type)
1824 {
1825 	switch (prog->type) {
1826 	case BPF_PROG_TYPE_CGROUP_SOCK:
1827 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1828 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1829 	default:
1830 		return 0;
1831 	}
1832 }
1833 
1834 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1835 
1836 #define BPF_F_ATTACH_MASK \
1837 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1838 
1839 static int bpf_prog_attach(const union bpf_attr *attr)
1840 {
1841 	enum bpf_prog_type ptype;
1842 	struct bpf_prog *prog;
1843 	int ret;
1844 
1845 	if (!capable(CAP_NET_ADMIN))
1846 		return -EPERM;
1847 
1848 	if (CHECK_ATTR(BPF_PROG_ATTACH))
1849 		return -EINVAL;
1850 
1851 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1852 		return -EINVAL;
1853 
1854 	switch (attr->attach_type) {
1855 	case BPF_CGROUP_INET_INGRESS:
1856 	case BPF_CGROUP_INET_EGRESS:
1857 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1858 		break;
1859 	case BPF_CGROUP_INET_SOCK_CREATE:
1860 	case BPF_CGROUP_INET4_POST_BIND:
1861 	case BPF_CGROUP_INET6_POST_BIND:
1862 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1863 		break;
1864 	case BPF_CGROUP_INET4_BIND:
1865 	case BPF_CGROUP_INET6_BIND:
1866 	case BPF_CGROUP_INET4_CONNECT:
1867 	case BPF_CGROUP_INET6_CONNECT:
1868 	case BPF_CGROUP_UDP4_SENDMSG:
1869 	case BPF_CGROUP_UDP6_SENDMSG:
1870 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1871 		break;
1872 	case BPF_CGROUP_SOCK_OPS:
1873 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1874 		break;
1875 	case BPF_CGROUP_DEVICE:
1876 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1877 		break;
1878 	case BPF_SK_MSG_VERDICT:
1879 		ptype = BPF_PROG_TYPE_SK_MSG;
1880 		break;
1881 	case BPF_SK_SKB_STREAM_PARSER:
1882 	case BPF_SK_SKB_STREAM_VERDICT:
1883 		ptype = BPF_PROG_TYPE_SK_SKB;
1884 		break;
1885 	case BPF_LIRC_MODE2:
1886 		ptype = BPF_PROG_TYPE_LIRC_MODE2;
1887 		break;
1888 	case BPF_FLOW_DISSECTOR:
1889 		ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1890 		break;
1891 	case BPF_CGROUP_SYSCTL:
1892 		ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1893 		break;
1894 	default:
1895 		return -EINVAL;
1896 	}
1897 
1898 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1899 	if (IS_ERR(prog))
1900 		return PTR_ERR(prog);
1901 
1902 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1903 		bpf_prog_put(prog);
1904 		return -EINVAL;
1905 	}
1906 
1907 	switch (ptype) {
1908 	case BPF_PROG_TYPE_SK_SKB:
1909 	case BPF_PROG_TYPE_SK_MSG:
1910 		ret = sock_map_get_from_fd(attr, prog);
1911 		break;
1912 	case BPF_PROG_TYPE_LIRC_MODE2:
1913 		ret = lirc_prog_attach(attr, prog);
1914 		break;
1915 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
1916 		ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1917 		break;
1918 	default:
1919 		ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1920 	}
1921 
1922 	if (ret)
1923 		bpf_prog_put(prog);
1924 	return ret;
1925 }
1926 
1927 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1928 
1929 static int bpf_prog_detach(const union bpf_attr *attr)
1930 {
1931 	enum bpf_prog_type ptype;
1932 
1933 	if (!capable(CAP_NET_ADMIN))
1934 		return -EPERM;
1935 
1936 	if (CHECK_ATTR(BPF_PROG_DETACH))
1937 		return -EINVAL;
1938 
1939 	switch (attr->attach_type) {
1940 	case BPF_CGROUP_INET_INGRESS:
1941 	case BPF_CGROUP_INET_EGRESS:
1942 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1943 		break;
1944 	case BPF_CGROUP_INET_SOCK_CREATE:
1945 	case BPF_CGROUP_INET4_POST_BIND:
1946 	case BPF_CGROUP_INET6_POST_BIND:
1947 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1948 		break;
1949 	case BPF_CGROUP_INET4_BIND:
1950 	case BPF_CGROUP_INET6_BIND:
1951 	case BPF_CGROUP_INET4_CONNECT:
1952 	case BPF_CGROUP_INET6_CONNECT:
1953 	case BPF_CGROUP_UDP4_SENDMSG:
1954 	case BPF_CGROUP_UDP6_SENDMSG:
1955 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1956 		break;
1957 	case BPF_CGROUP_SOCK_OPS:
1958 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1959 		break;
1960 	case BPF_CGROUP_DEVICE:
1961 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1962 		break;
1963 	case BPF_SK_MSG_VERDICT:
1964 		return sock_map_get_from_fd(attr, NULL);
1965 	case BPF_SK_SKB_STREAM_PARSER:
1966 	case BPF_SK_SKB_STREAM_VERDICT:
1967 		return sock_map_get_from_fd(attr, NULL);
1968 	case BPF_LIRC_MODE2:
1969 		return lirc_prog_detach(attr);
1970 	case BPF_FLOW_DISSECTOR:
1971 		return skb_flow_dissector_bpf_prog_detach(attr);
1972 	case BPF_CGROUP_SYSCTL:
1973 		ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1974 		break;
1975 	default:
1976 		return -EINVAL;
1977 	}
1978 
1979 	return cgroup_bpf_prog_detach(attr, ptype);
1980 }
1981 
1982 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1983 
1984 static int bpf_prog_query(const union bpf_attr *attr,
1985 			  union bpf_attr __user *uattr)
1986 {
1987 	if (!capable(CAP_NET_ADMIN))
1988 		return -EPERM;
1989 	if (CHECK_ATTR(BPF_PROG_QUERY))
1990 		return -EINVAL;
1991 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1992 		return -EINVAL;
1993 
1994 	switch (attr->query.attach_type) {
1995 	case BPF_CGROUP_INET_INGRESS:
1996 	case BPF_CGROUP_INET_EGRESS:
1997 	case BPF_CGROUP_INET_SOCK_CREATE:
1998 	case BPF_CGROUP_INET4_BIND:
1999 	case BPF_CGROUP_INET6_BIND:
2000 	case BPF_CGROUP_INET4_POST_BIND:
2001 	case BPF_CGROUP_INET6_POST_BIND:
2002 	case BPF_CGROUP_INET4_CONNECT:
2003 	case BPF_CGROUP_INET6_CONNECT:
2004 	case BPF_CGROUP_UDP4_SENDMSG:
2005 	case BPF_CGROUP_UDP6_SENDMSG:
2006 	case BPF_CGROUP_SOCK_OPS:
2007 	case BPF_CGROUP_DEVICE:
2008 	case BPF_CGROUP_SYSCTL:
2009 		break;
2010 	case BPF_LIRC_MODE2:
2011 		return lirc_prog_query(attr, uattr);
2012 	default:
2013 		return -EINVAL;
2014 	}
2015 
2016 	return cgroup_bpf_prog_query(attr, uattr);
2017 }
2018 
2019 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2020 
2021 static int bpf_prog_test_run(const union bpf_attr *attr,
2022 			     union bpf_attr __user *uattr)
2023 {
2024 	struct bpf_prog *prog;
2025 	int ret = -ENOTSUPP;
2026 
2027 	if (!capable(CAP_SYS_ADMIN))
2028 		return -EPERM;
2029 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2030 		return -EINVAL;
2031 
2032 	if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2033 	    (!attr->test.ctx_size_in && attr->test.ctx_in))
2034 		return -EINVAL;
2035 
2036 	if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2037 	    (!attr->test.ctx_size_out && attr->test.ctx_out))
2038 		return -EINVAL;
2039 
2040 	prog = bpf_prog_get(attr->test.prog_fd);
2041 	if (IS_ERR(prog))
2042 		return PTR_ERR(prog);
2043 
2044 	if (prog->aux->ops->test_run)
2045 		ret = prog->aux->ops->test_run(prog, attr, uattr);
2046 
2047 	bpf_prog_put(prog);
2048 	return ret;
2049 }
2050 
2051 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2052 
2053 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2054 			       union bpf_attr __user *uattr,
2055 			       struct idr *idr,
2056 			       spinlock_t *lock)
2057 {
2058 	u32 next_id = attr->start_id;
2059 	int err = 0;
2060 
2061 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2062 		return -EINVAL;
2063 
2064 	if (!capable(CAP_SYS_ADMIN))
2065 		return -EPERM;
2066 
2067 	next_id++;
2068 	spin_lock_bh(lock);
2069 	if (!idr_get_next(idr, &next_id))
2070 		err = -ENOENT;
2071 	spin_unlock_bh(lock);
2072 
2073 	if (!err)
2074 		err = put_user(next_id, &uattr->next_id);
2075 
2076 	return err;
2077 }
2078 
2079 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2080 
2081 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2082 {
2083 	struct bpf_prog *prog;
2084 	u32 id = attr->prog_id;
2085 	int fd;
2086 
2087 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2088 		return -EINVAL;
2089 
2090 	if (!capable(CAP_SYS_ADMIN))
2091 		return -EPERM;
2092 
2093 	spin_lock_bh(&prog_idr_lock);
2094 	prog = idr_find(&prog_idr, id);
2095 	if (prog)
2096 		prog = bpf_prog_inc_not_zero(prog);
2097 	else
2098 		prog = ERR_PTR(-ENOENT);
2099 	spin_unlock_bh(&prog_idr_lock);
2100 
2101 	if (IS_ERR(prog))
2102 		return PTR_ERR(prog);
2103 
2104 	fd = bpf_prog_new_fd(prog);
2105 	if (fd < 0)
2106 		bpf_prog_put(prog);
2107 
2108 	return fd;
2109 }
2110 
2111 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2112 
2113 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2114 {
2115 	struct bpf_map *map;
2116 	u32 id = attr->map_id;
2117 	int f_flags;
2118 	int fd;
2119 
2120 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2121 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2122 		return -EINVAL;
2123 
2124 	if (!capable(CAP_SYS_ADMIN))
2125 		return -EPERM;
2126 
2127 	f_flags = bpf_get_file_flag(attr->open_flags);
2128 	if (f_flags < 0)
2129 		return f_flags;
2130 
2131 	spin_lock_bh(&map_idr_lock);
2132 	map = idr_find(&map_idr, id);
2133 	if (map)
2134 		map = bpf_map_inc_not_zero(map, true);
2135 	else
2136 		map = ERR_PTR(-ENOENT);
2137 	spin_unlock_bh(&map_idr_lock);
2138 
2139 	if (IS_ERR(map))
2140 		return PTR_ERR(map);
2141 
2142 	fd = bpf_map_new_fd(map, f_flags);
2143 	if (fd < 0)
2144 		bpf_map_put_with_uref(map);
2145 
2146 	return fd;
2147 }
2148 
2149 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2150 					      unsigned long addr, u32 *off,
2151 					      u32 *type)
2152 {
2153 	const struct bpf_map *map;
2154 	int i;
2155 
2156 	for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2157 		map = prog->aux->used_maps[i];
2158 		if (map == (void *)addr) {
2159 			*type = BPF_PSEUDO_MAP_FD;
2160 			return map;
2161 		}
2162 		if (!map->ops->map_direct_value_meta)
2163 			continue;
2164 		if (!map->ops->map_direct_value_meta(map, addr, off)) {
2165 			*type = BPF_PSEUDO_MAP_VALUE;
2166 			return map;
2167 		}
2168 	}
2169 
2170 	return NULL;
2171 }
2172 
2173 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2174 {
2175 	const struct bpf_map *map;
2176 	struct bpf_insn *insns;
2177 	u32 off, type;
2178 	u64 imm;
2179 	int i;
2180 
2181 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2182 			GFP_USER);
2183 	if (!insns)
2184 		return insns;
2185 
2186 	for (i = 0; i < prog->len; i++) {
2187 		if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2188 			insns[i].code = BPF_JMP | BPF_CALL;
2189 			insns[i].imm = BPF_FUNC_tail_call;
2190 			/* fall-through */
2191 		}
2192 		if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2193 		    insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2194 			if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2195 				insns[i].code = BPF_JMP | BPF_CALL;
2196 			if (!bpf_dump_raw_ok())
2197 				insns[i].imm = 0;
2198 			continue;
2199 		}
2200 
2201 		if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2202 			continue;
2203 
2204 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2205 		map = bpf_map_from_imm(prog, imm, &off, &type);
2206 		if (map) {
2207 			insns[i].src_reg = type;
2208 			insns[i].imm = map->id;
2209 			insns[i + 1].imm = off;
2210 			continue;
2211 		}
2212 	}
2213 
2214 	return insns;
2215 }
2216 
2217 static int set_info_rec_size(struct bpf_prog_info *info)
2218 {
2219 	/*
2220 	 * Ensure info.*_rec_size is the same as kernel expected size
2221 	 *
2222 	 * or
2223 	 *
2224 	 * Only allow zero *_rec_size if both _rec_size and _cnt are
2225 	 * zero.  In this case, the kernel will set the expected
2226 	 * _rec_size back to the info.
2227 	 */
2228 
2229 	if ((info->nr_func_info || info->func_info_rec_size) &&
2230 	    info->func_info_rec_size != sizeof(struct bpf_func_info))
2231 		return -EINVAL;
2232 
2233 	if ((info->nr_line_info || info->line_info_rec_size) &&
2234 	    info->line_info_rec_size != sizeof(struct bpf_line_info))
2235 		return -EINVAL;
2236 
2237 	if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2238 	    info->jited_line_info_rec_size != sizeof(__u64))
2239 		return -EINVAL;
2240 
2241 	info->func_info_rec_size = sizeof(struct bpf_func_info);
2242 	info->line_info_rec_size = sizeof(struct bpf_line_info);
2243 	info->jited_line_info_rec_size = sizeof(__u64);
2244 
2245 	return 0;
2246 }
2247 
2248 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2249 				   const union bpf_attr *attr,
2250 				   union bpf_attr __user *uattr)
2251 {
2252 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2253 	struct bpf_prog_info info = {};
2254 	u32 info_len = attr->info.info_len;
2255 	struct bpf_prog_stats stats;
2256 	char __user *uinsns;
2257 	u32 ulen;
2258 	int err;
2259 
2260 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2261 	if (err)
2262 		return err;
2263 	info_len = min_t(u32, sizeof(info), info_len);
2264 
2265 	if (copy_from_user(&info, uinfo, info_len))
2266 		return -EFAULT;
2267 
2268 	info.type = prog->type;
2269 	info.id = prog->aux->id;
2270 	info.load_time = prog->aux->load_time;
2271 	info.created_by_uid = from_kuid_munged(current_user_ns(),
2272 					       prog->aux->user->uid);
2273 	info.gpl_compatible = prog->gpl_compatible;
2274 
2275 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
2276 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2277 
2278 	ulen = info.nr_map_ids;
2279 	info.nr_map_ids = prog->aux->used_map_cnt;
2280 	ulen = min_t(u32, info.nr_map_ids, ulen);
2281 	if (ulen) {
2282 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2283 		u32 i;
2284 
2285 		for (i = 0; i < ulen; i++)
2286 			if (put_user(prog->aux->used_maps[i]->id,
2287 				     &user_map_ids[i]))
2288 				return -EFAULT;
2289 	}
2290 
2291 	err = set_info_rec_size(&info);
2292 	if (err)
2293 		return err;
2294 
2295 	bpf_prog_get_stats(prog, &stats);
2296 	info.run_time_ns = stats.nsecs;
2297 	info.run_cnt = stats.cnt;
2298 
2299 	if (!capable(CAP_SYS_ADMIN)) {
2300 		info.jited_prog_len = 0;
2301 		info.xlated_prog_len = 0;
2302 		info.nr_jited_ksyms = 0;
2303 		info.nr_jited_func_lens = 0;
2304 		info.nr_func_info = 0;
2305 		info.nr_line_info = 0;
2306 		info.nr_jited_line_info = 0;
2307 		goto done;
2308 	}
2309 
2310 	ulen = info.xlated_prog_len;
2311 	info.xlated_prog_len = bpf_prog_insn_size(prog);
2312 	if (info.xlated_prog_len && ulen) {
2313 		struct bpf_insn *insns_sanitized;
2314 		bool fault;
2315 
2316 		if (prog->blinded && !bpf_dump_raw_ok()) {
2317 			info.xlated_prog_insns = 0;
2318 			goto done;
2319 		}
2320 		insns_sanitized = bpf_insn_prepare_dump(prog);
2321 		if (!insns_sanitized)
2322 			return -ENOMEM;
2323 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2324 		ulen = min_t(u32, info.xlated_prog_len, ulen);
2325 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
2326 		kfree(insns_sanitized);
2327 		if (fault)
2328 			return -EFAULT;
2329 	}
2330 
2331 	if (bpf_prog_is_dev_bound(prog->aux)) {
2332 		err = bpf_prog_offload_info_fill(&info, prog);
2333 		if (err)
2334 			return err;
2335 		goto done;
2336 	}
2337 
2338 	/* NOTE: the following code is supposed to be skipped for offload.
2339 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
2340 	 * for offload.
2341 	 */
2342 	ulen = info.jited_prog_len;
2343 	if (prog->aux->func_cnt) {
2344 		u32 i;
2345 
2346 		info.jited_prog_len = 0;
2347 		for (i = 0; i < prog->aux->func_cnt; i++)
2348 			info.jited_prog_len += prog->aux->func[i]->jited_len;
2349 	} else {
2350 		info.jited_prog_len = prog->jited_len;
2351 	}
2352 
2353 	if (info.jited_prog_len && ulen) {
2354 		if (bpf_dump_raw_ok()) {
2355 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
2356 			ulen = min_t(u32, info.jited_prog_len, ulen);
2357 
2358 			/* for multi-function programs, copy the JITed
2359 			 * instructions for all the functions
2360 			 */
2361 			if (prog->aux->func_cnt) {
2362 				u32 len, free, i;
2363 				u8 *img;
2364 
2365 				free = ulen;
2366 				for (i = 0; i < prog->aux->func_cnt; i++) {
2367 					len = prog->aux->func[i]->jited_len;
2368 					len = min_t(u32, len, free);
2369 					img = (u8 *) prog->aux->func[i]->bpf_func;
2370 					if (copy_to_user(uinsns, img, len))
2371 						return -EFAULT;
2372 					uinsns += len;
2373 					free -= len;
2374 					if (!free)
2375 						break;
2376 				}
2377 			} else {
2378 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
2379 					return -EFAULT;
2380 			}
2381 		} else {
2382 			info.jited_prog_insns = 0;
2383 		}
2384 	}
2385 
2386 	ulen = info.nr_jited_ksyms;
2387 	info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2388 	if (ulen) {
2389 		if (bpf_dump_raw_ok()) {
2390 			unsigned long ksym_addr;
2391 			u64 __user *user_ksyms;
2392 			u32 i;
2393 
2394 			/* copy the address of the kernel symbol
2395 			 * corresponding to each function
2396 			 */
2397 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2398 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2399 			if (prog->aux->func_cnt) {
2400 				for (i = 0; i < ulen; i++) {
2401 					ksym_addr = (unsigned long)
2402 						prog->aux->func[i]->bpf_func;
2403 					if (put_user((u64) ksym_addr,
2404 						     &user_ksyms[i]))
2405 						return -EFAULT;
2406 				}
2407 			} else {
2408 				ksym_addr = (unsigned long) prog->bpf_func;
2409 				if (put_user((u64) ksym_addr, &user_ksyms[0]))
2410 					return -EFAULT;
2411 			}
2412 		} else {
2413 			info.jited_ksyms = 0;
2414 		}
2415 	}
2416 
2417 	ulen = info.nr_jited_func_lens;
2418 	info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2419 	if (ulen) {
2420 		if (bpf_dump_raw_ok()) {
2421 			u32 __user *user_lens;
2422 			u32 func_len, i;
2423 
2424 			/* copy the JITed image lengths for each function */
2425 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2426 			user_lens = u64_to_user_ptr(info.jited_func_lens);
2427 			if (prog->aux->func_cnt) {
2428 				for (i = 0; i < ulen; i++) {
2429 					func_len =
2430 						prog->aux->func[i]->jited_len;
2431 					if (put_user(func_len, &user_lens[i]))
2432 						return -EFAULT;
2433 				}
2434 			} else {
2435 				func_len = prog->jited_len;
2436 				if (put_user(func_len, &user_lens[0]))
2437 					return -EFAULT;
2438 			}
2439 		} else {
2440 			info.jited_func_lens = 0;
2441 		}
2442 	}
2443 
2444 	if (prog->aux->btf)
2445 		info.btf_id = btf_id(prog->aux->btf);
2446 
2447 	ulen = info.nr_func_info;
2448 	info.nr_func_info = prog->aux->func_info_cnt;
2449 	if (info.nr_func_info && ulen) {
2450 		char __user *user_finfo;
2451 
2452 		user_finfo = u64_to_user_ptr(info.func_info);
2453 		ulen = min_t(u32, info.nr_func_info, ulen);
2454 		if (copy_to_user(user_finfo, prog->aux->func_info,
2455 				 info.func_info_rec_size * ulen))
2456 			return -EFAULT;
2457 	}
2458 
2459 	ulen = info.nr_line_info;
2460 	info.nr_line_info = prog->aux->nr_linfo;
2461 	if (info.nr_line_info && ulen) {
2462 		__u8 __user *user_linfo;
2463 
2464 		user_linfo = u64_to_user_ptr(info.line_info);
2465 		ulen = min_t(u32, info.nr_line_info, ulen);
2466 		if (copy_to_user(user_linfo, prog->aux->linfo,
2467 				 info.line_info_rec_size * ulen))
2468 			return -EFAULT;
2469 	}
2470 
2471 	ulen = info.nr_jited_line_info;
2472 	if (prog->aux->jited_linfo)
2473 		info.nr_jited_line_info = prog->aux->nr_linfo;
2474 	else
2475 		info.nr_jited_line_info = 0;
2476 	if (info.nr_jited_line_info && ulen) {
2477 		if (bpf_dump_raw_ok()) {
2478 			__u64 __user *user_linfo;
2479 			u32 i;
2480 
2481 			user_linfo = u64_to_user_ptr(info.jited_line_info);
2482 			ulen = min_t(u32, info.nr_jited_line_info, ulen);
2483 			for (i = 0; i < ulen; i++) {
2484 				if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2485 					     &user_linfo[i]))
2486 					return -EFAULT;
2487 			}
2488 		} else {
2489 			info.jited_line_info = 0;
2490 		}
2491 	}
2492 
2493 	ulen = info.nr_prog_tags;
2494 	info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2495 	if (ulen) {
2496 		__u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2497 		u32 i;
2498 
2499 		user_prog_tags = u64_to_user_ptr(info.prog_tags);
2500 		ulen = min_t(u32, info.nr_prog_tags, ulen);
2501 		if (prog->aux->func_cnt) {
2502 			for (i = 0; i < ulen; i++) {
2503 				if (copy_to_user(user_prog_tags[i],
2504 						 prog->aux->func[i]->tag,
2505 						 BPF_TAG_SIZE))
2506 					return -EFAULT;
2507 			}
2508 		} else {
2509 			if (copy_to_user(user_prog_tags[0],
2510 					 prog->tag, BPF_TAG_SIZE))
2511 				return -EFAULT;
2512 		}
2513 	}
2514 
2515 done:
2516 	if (copy_to_user(uinfo, &info, info_len) ||
2517 	    put_user(info_len, &uattr->info.info_len))
2518 		return -EFAULT;
2519 
2520 	return 0;
2521 }
2522 
2523 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2524 				  const union bpf_attr *attr,
2525 				  union bpf_attr __user *uattr)
2526 {
2527 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2528 	struct bpf_map_info info = {};
2529 	u32 info_len = attr->info.info_len;
2530 	int err;
2531 
2532 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2533 	if (err)
2534 		return err;
2535 	info_len = min_t(u32, sizeof(info), info_len);
2536 
2537 	info.type = map->map_type;
2538 	info.id = map->id;
2539 	info.key_size = map->key_size;
2540 	info.value_size = map->value_size;
2541 	info.max_entries = map->max_entries;
2542 	info.map_flags = map->map_flags;
2543 	memcpy(info.name, map->name, sizeof(map->name));
2544 
2545 	if (map->btf) {
2546 		info.btf_id = btf_id(map->btf);
2547 		info.btf_key_type_id = map->btf_key_type_id;
2548 		info.btf_value_type_id = map->btf_value_type_id;
2549 	}
2550 
2551 	if (bpf_map_is_dev_bound(map)) {
2552 		err = bpf_map_offload_info_fill(&info, map);
2553 		if (err)
2554 			return err;
2555 	}
2556 
2557 	if (copy_to_user(uinfo, &info, info_len) ||
2558 	    put_user(info_len, &uattr->info.info_len))
2559 		return -EFAULT;
2560 
2561 	return 0;
2562 }
2563 
2564 static int bpf_btf_get_info_by_fd(struct btf *btf,
2565 				  const union bpf_attr *attr,
2566 				  union bpf_attr __user *uattr)
2567 {
2568 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2569 	u32 info_len = attr->info.info_len;
2570 	int err;
2571 
2572 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2573 	if (err)
2574 		return err;
2575 
2576 	return btf_get_info_by_fd(btf, attr, uattr);
2577 }
2578 
2579 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2580 
2581 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2582 				  union bpf_attr __user *uattr)
2583 {
2584 	int ufd = attr->info.bpf_fd;
2585 	struct fd f;
2586 	int err;
2587 
2588 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2589 		return -EINVAL;
2590 
2591 	f = fdget(ufd);
2592 	if (!f.file)
2593 		return -EBADFD;
2594 
2595 	if (f.file->f_op == &bpf_prog_fops)
2596 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2597 					      uattr);
2598 	else if (f.file->f_op == &bpf_map_fops)
2599 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2600 					     uattr);
2601 	else if (f.file->f_op == &btf_fops)
2602 		err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2603 	else
2604 		err = -EINVAL;
2605 
2606 	fdput(f);
2607 	return err;
2608 }
2609 
2610 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2611 
2612 static int bpf_btf_load(const union bpf_attr *attr)
2613 {
2614 	if (CHECK_ATTR(BPF_BTF_LOAD))
2615 		return -EINVAL;
2616 
2617 	if (!capable(CAP_SYS_ADMIN))
2618 		return -EPERM;
2619 
2620 	return btf_new_fd(attr);
2621 }
2622 
2623 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2624 
2625 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2626 {
2627 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2628 		return -EINVAL;
2629 
2630 	if (!capable(CAP_SYS_ADMIN))
2631 		return -EPERM;
2632 
2633 	return btf_get_fd_by_id(attr->btf_id);
2634 }
2635 
2636 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2637 				    union bpf_attr __user *uattr,
2638 				    u32 prog_id, u32 fd_type,
2639 				    const char *buf, u64 probe_offset,
2640 				    u64 probe_addr)
2641 {
2642 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2643 	u32 len = buf ? strlen(buf) : 0, input_len;
2644 	int err = 0;
2645 
2646 	if (put_user(len, &uattr->task_fd_query.buf_len))
2647 		return -EFAULT;
2648 	input_len = attr->task_fd_query.buf_len;
2649 	if (input_len && ubuf) {
2650 		if (!len) {
2651 			/* nothing to copy, just make ubuf NULL terminated */
2652 			char zero = '\0';
2653 
2654 			if (put_user(zero, ubuf))
2655 				return -EFAULT;
2656 		} else if (input_len >= len + 1) {
2657 			/* ubuf can hold the string with NULL terminator */
2658 			if (copy_to_user(ubuf, buf, len + 1))
2659 				return -EFAULT;
2660 		} else {
2661 			/* ubuf cannot hold the string with NULL terminator,
2662 			 * do a partial copy with NULL terminator.
2663 			 */
2664 			char zero = '\0';
2665 
2666 			err = -ENOSPC;
2667 			if (copy_to_user(ubuf, buf, input_len - 1))
2668 				return -EFAULT;
2669 			if (put_user(zero, ubuf + input_len - 1))
2670 				return -EFAULT;
2671 		}
2672 	}
2673 
2674 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2675 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2676 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2677 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2678 		return -EFAULT;
2679 
2680 	return err;
2681 }
2682 
2683 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2684 
2685 static int bpf_task_fd_query(const union bpf_attr *attr,
2686 			     union bpf_attr __user *uattr)
2687 {
2688 	pid_t pid = attr->task_fd_query.pid;
2689 	u32 fd = attr->task_fd_query.fd;
2690 	const struct perf_event *event;
2691 	struct files_struct *files;
2692 	struct task_struct *task;
2693 	struct file *file;
2694 	int err;
2695 
2696 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2697 		return -EINVAL;
2698 
2699 	if (!capable(CAP_SYS_ADMIN))
2700 		return -EPERM;
2701 
2702 	if (attr->task_fd_query.flags != 0)
2703 		return -EINVAL;
2704 
2705 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2706 	if (!task)
2707 		return -ENOENT;
2708 
2709 	files = get_files_struct(task);
2710 	put_task_struct(task);
2711 	if (!files)
2712 		return -ENOENT;
2713 
2714 	err = 0;
2715 	spin_lock(&files->file_lock);
2716 	file = fcheck_files(files, fd);
2717 	if (!file)
2718 		err = -EBADF;
2719 	else
2720 		get_file(file);
2721 	spin_unlock(&files->file_lock);
2722 	put_files_struct(files);
2723 
2724 	if (err)
2725 		goto out;
2726 
2727 	if (file->f_op == &bpf_raw_tp_fops) {
2728 		struct bpf_raw_tracepoint *raw_tp = file->private_data;
2729 		struct bpf_raw_event_map *btp = raw_tp->btp;
2730 
2731 		err = bpf_task_fd_query_copy(attr, uattr,
2732 					     raw_tp->prog->aux->id,
2733 					     BPF_FD_TYPE_RAW_TRACEPOINT,
2734 					     btp->tp->name, 0, 0);
2735 		goto put_file;
2736 	}
2737 
2738 	event = perf_get_event(file);
2739 	if (!IS_ERR(event)) {
2740 		u64 probe_offset, probe_addr;
2741 		u32 prog_id, fd_type;
2742 		const char *buf;
2743 
2744 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2745 					      &buf, &probe_offset,
2746 					      &probe_addr);
2747 		if (!err)
2748 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2749 						     fd_type, buf,
2750 						     probe_offset,
2751 						     probe_addr);
2752 		goto put_file;
2753 	}
2754 
2755 	err = -ENOTSUPP;
2756 put_file:
2757 	fput(file);
2758 out:
2759 	return err;
2760 }
2761 
2762 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2763 {
2764 	union bpf_attr attr = {};
2765 	int err;
2766 
2767 	if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2768 		return -EPERM;
2769 
2770 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2771 	if (err)
2772 		return err;
2773 	size = min_t(u32, size, sizeof(attr));
2774 
2775 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
2776 	if (copy_from_user(&attr, uattr, size) != 0)
2777 		return -EFAULT;
2778 
2779 	err = security_bpf(cmd, &attr, size);
2780 	if (err < 0)
2781 		return err;
2782 
2783 	switch (cmd) {
2784 	case BPF_MAP_CREATE:
2785 		err = map_create(&attr);
2786 		break;
2787 	case BPF_MAP_LOOKUP_ELEM:
2788 		err = map_lookup_elem(&attr);
2789 		break;
2790 	case BPF_MAP_UPDATE_ELEM:
2791 		err = map_update_elem(&attr);
2792 		break;
2793 	case BPF_MAP_DELETE_ELEM:
2794 		err = map_delete_elem(&attr);
2795 		break;
2796 	case BPF_MAP_GET_NEXT_KEY:
2797 		err = map_get_next_key(&attr);
2798 		break;
2799 	case BPF_MAP_FREEZE:
2800 		err = map_freeze(&attr);
2801 		break;
2802 	case BPF_PROG_LOAD:
2803 		err = bpf_prog_load(&attr, uattr);
2804 		break;
2805 	case BPF_OBJ_PIN:
2806 		err = bpf_obj_pin(&attr);
2807 		break;
2808 	case BPF_OBJ_GET:
2809 		err = bpf_obj_get(&attr);
2810 		break;
2811 	case BPF_PROG_ATTACH:
2812 		err = bpf_prog_attach(&attr);
2813 		break;
2814 	case BPF_PROG_DETACH:
2815 		err = bpf_prog_detach(&attr);
2816 		break;
2817 	case BPF_PROG_QUERY:
2818 		err = bpf_prog_query(&attr, uattr);
2819 		break;
2820 	case BPF_PROG_TEST_RUN:
2821 		err = bpf_prog_test_run(&attr, uattr);
2822 		break;
2823 	case BPF_PROG_GET_NEXT_ID:
2824 		err = bpf_obj_get_next_id(&attr, uattr,
2825 					  &prog_idr, &prog_idr_lock);
2826 		break;
2827 	case BPF_MAP_GET_NEXT_ID:
2828 		err = bpf_obj_get_next_id(&attr, uattr,
2829 					  &map_idr, &map_idr_lock);
2830 		break;
2831 	case BPF_PROG_GET_FD_BY_ID:
2832 		err = bpf_prog_get_fd_by_id(&attr);
2833 		break;
2834 	case BPF_MAP_GET_FD_BY_ID:
2835 		err = bpf_map_get_fd_by_id(&attr);
2836 		break;
2837 	case BPF_OBJ_GET_INFO_BY_FD:
2838 		err = bpf_obj_get_info_by_fd(&attr, uattr);
2839 		break;
2840 	case BPF_RAW_TRACEPOINT_OPEN:
2841 		err = bpf_raw_tracepoint_open(&attr);
2842 		break;
2843 	case BPF_BTF_LOAD:
2844 		err = bpf_btf_load(&attr);
2845 		break;
2846 	case BPF_BTF_GET_FD_BY_ID:
2847 		err = bpf_btf_get_fd_by_id(&attr);
2848 		break;
2849 	case BPF_TASK_FD_QUERY:
2850 		err = bpf_task_fd_query(&attr, uattr);
2851 		break;
2852 	case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2853 		err = map_lookup_and_delete_elem(&attr);
2854 		break;
2855 	default:
2856 		err = -EINVAL;
2857 		break;
2858 	}
2859 
2860 	return err;
2861 }
2862