xref: /openbmc/linux/kernel/bpf/syscall.c (revision 5ef12cb4a3a78ffb331c03a795a15eea4ae35155)
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/syscalls.h>
15 #include <linux/slab.h>
16 #include <linux/sched/signal.h>
17 #include <linux/vmalloc.h>
18 #include <linux/mmzone.h>
19 #include <linux/anon_inodes.h>
20 #include <linux/file.h>
21 #include <linux/license.h>
22 #include <linux/filter.h>
23 #include <linux/version.h>
24 #include <linux/kernel.h>
25 #include <linux/idr.h>
26 #include <linux/cred.h>
27 #include <linux/timekeeping.h>
28 #include <linux/ctype.h>
29 #include <linux/nospec.h>
30 
31 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
32 			   (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
33 			   (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
34 			   (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
35 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
36 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
37 
38 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
39 
40 DEFINE_PER_CPU(int, bpf_prog_active);
41 static DEFINE_IDR(prog_idr);
42 static DEFINE_SPINLOCK(prog_idr_lock);
43 static DEFINE_IDR(map_idr);
44 static DEFINE_SPINLOCK(map_idr_lock);
45 
46 int sysctl_unprivileged_bpf_disabled __read_mostly;
47 
48 static const struct bpf_map_ops * const bpf_map_types[] = {
49 #define BPF_PROG_TYPE(_id, _ops)
50 #define BPF_MAP_TYPE(_id, _ops) \
51 	[_id] = &_ops,
52 #include <linux/bpf_types.h>
53 #undef BPF_PROG_TYPE
54 #undef BPF_MAP_TYPE
55 };
56 
57 /*
58  * If we're handed a bigger struct than we know of, ensure all the unknown bits
59  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
60  * we don't know about yet.
61  *
62  * There is a ToCToU between this function call and the following
63  * copy_from_user() call. However, this is not a concern since this function is
64  * meant to be a future-proofing of bits.
65  */
66 static int check_uarg_tail_zero(void __user *uaddr,
67 				size_t expected_size,
68 				size_t actual_size)
69 {
70 	unsigned char __user *addr;
71 	unsigned char __user *end;
72 	unsigned char val;
73 	int err;
74 
75 	if (unlikely(actual_size > PAGE_SIZE))	/* silly large */
76 		return -E2BIG;
77 
78 	if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
79 		return -EFAULT;
80 
81 	if (actual_size <= expected_size)
82 		return 0;
83 
84 	addr = uaddr + expected_size;
85 	end  = uaddr + actual_size;
86 
87 	for (; addr < end; addr++) {
88 		err = get_user(val, addr);
89 		if (err)
90 			return err;
91 		if (val)
92 			return -E2BIG;
93 	}
94 
95 	return 0;
96 }
97 
98 const struct bpf_map_ops bpf_map_offload_ops = {
99 	.map_alloc = bpf_map_offload_map_alloc,
100 	.map_free = bpf_map_offload_map_free,
101 };
102 
103 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
104 {
105 	const struct bpf_map_ops *ops;
106 	u32 type = attr->map_type;
107 	struct bpf_map *map;
108 	int err;
109 
110 	if (type >= ARRAY_SIZE(bpf_map_types))
111 		return ERR_PTR(-EINVAL);
112 	type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
113 	ops = bpf_map_types[type];
114 	if (!ops)
115 		return ERR_PTR(-EINVAL);
116 
117 	if (ops->map_alloc_check) {
118 		err = ops->map_alloc_check(attr);
119 		if (err)
120 			return ERR_PTR(err);
121 	}
122 	if (attr->map_ifindex)
123 		ops = &bpf_map_offload_ops;
124 	map = ops->map_alloc(attr);
125 	if (IS_ERR(map))
126 		return map;
127 	map->ops = ops;
128 	map->map_type = type;
129 	return map;
130 }
131 
132 void *bpf_map_area_alloc(size_t size, int numa_node)
133 {
134 	/* We definitely need __GFP_NORETRY, so OOM killer doesn't
135 	 * trigger under memory pressure as we really just want to
136 	 * fail instead.
137 	 */
138 	const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
139 	void *area;
140 
141 	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
142 		area = kmalloc_node(size, GFP_USER | flags, numa_node);
143 		if (area != NULL)
144 			return area;
145 	}
146 
147 	return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
148 					   __builtin_return_address(0));
149 }
150 
151 void bpf_map_area_free(void *area)
152 {
153 	kvfree(area);
154 }
155 
156 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
157 {
158 	map->map_type = attr->map_type;
159 	map->key_size = attr->key_size;
160 	map->value_size = attr->value_size;
161 	map->max_entries = attr->max_entries;
162 	map->map_flags = attr->map_flags;
163 	map->numa_node = bpf_map_attr_numa_node(attr);
164 }
165 
166 int bpf_map_precharge_memlock(u32 pages)
167 {
168 	struct user_struct *user = get_current_user();
169 	unsigned long memlock_limit, cur;
170 
171 	memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
172 	cur = atomic_long_read(&user->locked_vm);
173 	free_uid(user);
174 	if (cur + pages > memlock_limit)
175 		return -EPERM;
176 	return 0;
177 }
178 
179 static int bpf_map_charge_memlock(struct bpf_map *map)
180 {
181 	struct user_struct *user = get_current_user();
182 	unsigned long memlock_limit;
183 
184 	memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
185 
186 	atomic_long_add(map->pages, &user->locked_vm);
187 
188 	if (atomic_long_read(&user->locked_vm) > memlock_limit) {
189 		atomic_long_sub(map->pages, &user->locked_vm);
190 		free_uid(user);
191 		return -EPERM;
192 	}
193 	map->user = user;
194 	return 0;
195 }
196 
197 static void bpf_map_uncharge_memlock(struct bpf_map *map)
198 {
199 	struct user_struct *user = map->user;
200 
201 	atomic_long_sub(map->pages, &user->locked_vm);
202 	free_uid(user);
203 }
204 
205 static int bpf_map_alloc_id(struct bpf_map *map)
206 {
207 	int id;
208 
209 	idr_preload(GFP_KERNEL);
210 	spin_lock_bh(&map_idr_lock);
211 	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
212 	if (id > 0)
213 		map->id = id;
214 	spin_unlock_bh(&map_idr_lock);
215 	idr_preload_end();
216 
217 	if (WARN_ON_ONCE(!id))
218 		return -ENOSPC;
219 
220 	return id > 0 ? 0 : id;
221 }
222 
223 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
224 {
225 	unsigned long flags;
226 
227 	/* Offloaded maps are removed from the IDR store when their device
228 	 * disappears - even if someone holds an fd to them they are unusable,
229 	 * the memory is gone, all ops will fail; they are simply waiting for
230 	 * refcnt to drop to be freed.
231 	 */
232 	if (!map->id)
233 		return;
234 
235 	if (do_idr_lock)
236 		spin_lock_irqsave(&map_idr_lock, flags);
237 	else
238 		__acquire(&map_idr_lock);
239 
240 	idr_remove(&map_idr, map->id);
241 	map->id = 0;
242 
243 	if (do_idr_lock)
244 		spin_unlock_irqrestore(&map_idr_lock, flags);
245 	else
246 		__release(&map_idr_lock);
247 }
248 
249 /* called from workqueue */
250 static void bpf_map_free_deferred(struct work_struct *work)
251 {
252 	struct bpf_map *map = container_of(work, struct bpf_map, work);
253 
254 	bpf_map_uncharge_memlock(map);
255 	security_bpf_map_free(map);
256 	/* implementation dependent freeing */
257 	map->ops->map_free(map);
258 }
259 
260 static void bpf_map_put_uref(struct bpf_map *map)
261 {
262 	if (atomic_dec_and_test(&map->usercnt)) {
263 		if (map->ops->map_release_uref)
264 			map->ops->map_release_uref(map);
265 	}
266 }
267 
268 /* decrement map refcnt and schedule it for freeing via workqueue
269  * (unrelying map implementation ops->map_free() might sleep)
270  */
271 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
272 {
273 	if (atomic_dec_and_test(&map->refcnt)) {
274 		/* bpf_map_free_id() must be called first */
275 		bpf_map_free_id(map, do_idr_lock);
276 		INIT_WORK(&map->work, bpf_map_free_deferred);
277 		schedule_work(&map->work);
278 	}
279 }
280 
281 void bpf_map_put(struct bpf_map *map)
282 {
283 	__bpf_map_put(map, true);
284 }
285 
286 void bpf_map_put_with_uref(struct bpf_map *map)
287 {
288 	bpf_map_put_uref(map);
289 	bpf_map_put(map);
290 }
291 
292 static int bpf_map_release(struct inode *inode, struct file *filp)
293 {
294 	struct bpf_map *map = filp->private_data;
295 
296 	if (map->ops->map_release)
297 		map->ops->map_release(map, filp);
298 
299 	bpf_map_put_with_uref(map);
300 	return 0;
301 }
302 
303 #ifdef CONFIG_PROC_FS
304 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
305 {
306 	const struct bpf_map *map = filp->private_data;
307 	const struct bpf_array *array;
308 	u32 owner_prog_type = 0;
309 	u32 owner_jited = 0;
310 
311 	if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
312 		array = container_of(map, struct bpf_array, map);
313 		owner_prog_type = array->owner_prog_type;
314 		owner_jited = array->owner_jited;
315 	}
316 
317 	seq_printf(m,
318 		   "map_type:\t%u\n"
319 		   "key_size:\t%u\n"
320 		   "value_size:\t%u\n"
321 		   "max_entries:\t%u\n"
322 		   "map_flags:\t%#x\n"
323 		   "memlock:\t%llu\n",
324 		   map->map_type,
325 		   map->key_size,
326 		   map->value_size,
327 		   map->max_entries,
328 		   map->map_flags,
329 		   map->pages * 1ULL << PAGE_SHIFT);
330 
331 	if (owner_prog_type) {
332 		seq_printf(m, "owner_prog_type:\t%u\n",
333 			   owner_prog_type);
334 		seq_printf(m, "owner_jited:\t%u\n",
335 			   owner_jited);
336 	}
337 }
338 #endif
339 
340 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
341 			      loff_t *ppos)
342 {
343 	/* We need this handler such that alloc_file() enables
344 	 * f_mode with FMODE_CAN_READ.
345 	 */
346 	return -EINVAL;
347 }
348 
349 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
350 			       size_t siz, loff_t *ppos)
351 {
352 	/* We need this handler such that alloc_file() enables
353 	 * f_mode with FMODE_CAN_WRITE.
354 	 */
355 	return -EINVAL;
356 }
357 
358 const struct file_operations bpf_map_fops = {
359 #ifdef CONFIG_PROC_FS
360 	.show_fdinfo	= bpf_map_show_fdinfo,
361 #endif
362 	.release	= bpf_map_release,
363 	.read		= bpf_dummy_read,
364 	.write		= bpf_dummy_write,
365 };
366 
367 int bpf_map_new_fd(struct bpf_map *map, int flags)
368 {
369 	int ret;
370 
371 	ret = security_bpf_map(map, OPEN_FMODE(flags));
372 	if (ret < 0)
373 		return ret;
374 
375 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
376 				flags | O_CLOEXEC);
377 }
378 
379 int bpf_get_file_flag(int flags)
380 {
381 	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
382 		return -EINVAL;
383 	if (flags & BPF_F_RDONLY)
384 		return O_RDONLY;
385 	if (flags & BPF_F_WRONLY)
386 		return O_WRONLY;
387 	return O_RDWR;
388 }
389 
390 /* helper macro to check that unused fields 'union bpf_attr' are zero */
391 #define CHECK_ATTR(CMD) \
392 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
393 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
394 		   sizeof(*attr) - \
395 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
396 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
397 
398 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
399  * Return 0 on success and < 0 on error.
400  */
401 static int bpf_obj_name_cpy(char *dst, const char *src)
402 {
403 	const char *end = src + BPF_OBJ_NAME_LEN;
404 
405 	memset(dst, 0, BPF_OBJ_NAME_LEN);
406 
407 	/* Copy all isalnum() and '_' char */
408 	while (src < end && *src) {
409 		if (!isalnum(*src) && *src != '_')
410 			return -EINVAL;
411 		*dst++ = *src++;
412 	}
413 
414 	/* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
415 	if (src == end)
416 		return -EINVAL;
417 
418 	return 0;
419 }
420 
421 #define BPF_MAP_CREATE_LAST_FIELD map_ifindex
422 /* called via syscall */
423 static int map_create(union bpf_attr *attr)
424 {
425 	int numa_node = bpf_map_attr_numa_node(attr);
426 	struct bpf_map *map;
427 	int f_flags;
428 	int err;
429 
430 	err = CHECK_ATTR(BPF_MAP_CREATE);
431 	if (err)
432 		return -EINVAL;
433 
434 	f_flags = bpf_get_file_flag(attr->map_flags);
435 	if (f_flags < 0)
436 		return f_flags;
437 
438 	if (numa_node != NUMA_NO_NODE &&
439 	    ((unsigned int)numa_node >= nr_node_ids ||
440 	     !node_online(numa_node)))
441 		return -EINVAL;
442 
443 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
444 	map = find_and_alloc_map(attr);
445 	if (IS_ERR(map))
446 		return PTR_ERR(map);
447 
448 	err = bpf_obj_name_cpy(map->name, attr->map_name);
449 	if (err)
450 		goto free_map_nouncharge;
451 
452 	atomic_set(&map->refcnt, 1);
453 	atomic_set(&map->usercnt, 1);
454 
455 	err = security_bpf_map_alloc(map);
456 	if (err)
457 		goto free_map_nouncharge;
458 
459 	err = bpf_map_charge_memlock(map);
460 	if (err)
461 		goto free_map_sec;
462 
463 	err = bpf_map_alloc_id(map);
464 	if (err)
465 		goto free_map;
466 
467 	err = bpf_map_new_fd(map, f_flags);
468 	if (err < 0) {
469 		/* failed to allocate fd.
470 		 * bpf_map_put() is needed because the above
471 		 * bpf_map_alloc_id() has published the map
472 		 * to the userspace and the userspace may
473 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
474 		 */
475 		bpf_map_put(map);
476 		return err;
477 	}
478 
479 	trace_bpf_map_create(map, err);
480 	return err;
481 
482 free_map:
483 	bpf_map_uncharge_memlock(map);
484 free_map_sec:
485 	security_bpf_map_free(map);
486 free_map_nouncharge:
487 	map->ops->map_free(map);
488 	return err;
489 }
490 
491 /* if error is returned, fd is released.
492  * On success caller should complete fd access with matching fdput()
493  */
494 struct bpf_map *__bpf_map_get(struct fd f)
495 {
496 	if (!f.file)
497 		return ERR_PTR(-EBADF);
498 	if (f.file->f_op != &bpf_map_fops) {
499 		fdput(f);
500 		return ERR_PTR(-EINVAL);
501 	}
502 
503 	return f.file->private_data;
504 }
505 
506 /* prog's and map's refcnt limit */
507 #define BPF_MAX_REFCNT 32768
508 
509 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
510 {
511 	if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
512 		atomic_dec(&map->refcnt);
513 		return ERR_PTR(-EBUSY);
514 	}
515 	if (uref)
516 		atomic_inc(&map->usercnt);
517 	return map;
518 }
519 
520 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
521 {
522 	struct fd f = fdget(ufd);
523 	struct bpf_map *map;
524 
525 	map = __bpf_map_get(f);
526 	if (IS_ERR(map))
527 		return map;
528 
529 	map = bpf_map_inc(map, true);
530 	fdput(f);
531 
532 	return map;
533 }
534 
535 /* map_idr_lock should have been held */
536 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
537 					    bool uref)
538 {
539 	int refold;
540 
541 	refold = __atomic_add_unless(&map->refcnt, 1, 0);
542 
543 	if (refold >= BPF_MAX_REFCNT) {
544 		__bpf_map_put(map, false);
545 		return ERR_PTR(-EBUSY);
546 	}
547 
548 	if (!refold)
549 		return ERR_PTR(-ENOENT);
550 
551 	if (uref)
552 		atomic_inc(&map->usercnt);
553 
554 	return map;
555 }
556 
557 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
558 {
559 	return -ENOTSUPP;
560 }
561 
562 /* last field in 'union bpf_attr' used by this command */
563 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
564 
565 static int map_lookup_elem(union bpf_attr *attr)
566 {
567 	void __user *ukey = u64_to_user_ptr(attr->key);
568 	void __user *uvalue = u64_to_user_ptr(attr->value);
569 	int ufd = attr->map_fd;
570 	struct bpf_map *map;
571 	void *key, *value, *ptr;
572 	u32 value_size;
573 	struct fd f;
574 	int err;
575 
576 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
577 		return -EINVAL;
578 
579 	f = fdget(ufd);
580 	map = __bpf_map_get(f);
581 	if (IS_ERR(map))
582 		return PTR_ERR(map);
583 
584 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
585 		err = -EPERM;
586 		goto err_put;
587 	}
588 
589 	key = memdup_user(ukey, map->key_size);
590 	if (IS_ERR(key)) {
591 		err = PTR_ERR(key);
592 		goto err_put;
593 	}
594 
595 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
596 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
597 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
598 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
599 	else if (IS_FD_MAP(map))
600 		value_size = sizeof(u32);
601 	else
602 		value_size = map->value_size;
603 
604 	err = -ENOMEM;
605 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
606 	if (!value)
607 		goto free_key;
608 
609 	if (bpf_map_is_dev_bound(map)) {
610 		err = bpf_map_offload_lookup_elem(map, key, value);
611 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
612 		   map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
613 		err = bpf_percpu_hash_copy(map, key, value);
614 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
615 		err = bpf_percpu_array_copy(map, key, value);
616 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
617 		err = bpf_stackmap_copy(map, key, value);
618 	} else if (IS_FD_ARRAY(map)) {
619 		err = bpf_fd_array_map_lookup_elem(map, key, value);
620 	} else if (IS_FD_HASH(map)) {
621 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
622 	} else {
623 		rcu_read_lock();
624 		ptr = map->ops->map_lookup_elem(map, key);
625 		if (ptr)
626 			memcpy(value, ptr, value_size);
627 		rcu_read_unlock();
628 		err = ptr ? 0 : -ENOENT;
629 	}
630 
631 	if (err)
632 		goto free_value;
633 
634 	err = -EFAULT;
635 	if (copy_to_user(uvalue, value, value_size) != 0)
636 		goto free_value;
637 
638 	trace_bpf_map_lookup_elem(map, ufd, key, value);
639 	err = 0;
640 
641 free_value:
642 	kfree(value);
643 free_key:
644 	kfree(key);
645 err_put:
646 	fdput(f);
647 	return err;
648 }
649 
650 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
651 
652 static int map_update_elem(union bpf_attr *attr)
653 {
654 	void __user *ukey = u64_to_user_ptr(attr->key);
655 	void __user *uvalue = u64_to_user_ptr(attr->value);
656 	int ufd = attr->map_fd;
657 	struct bpf_map *map;
658 	void *key, *value;
659 	u32 value_size;
660 	struct fd f;
661 	int err;
662 
663 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
664 		return -EINVAL;
665 
666 	f = fdget(ufd);
667 	map = __bpf_map_get(f);
668 	if (IS_ERR(map))
669 		return PTR_ERR(map);
670 
671 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
672 		err = -EPERM;
673 		goto err_put;
674 	}
675 
676 	key = memdup_user(ukey, map->key_size);
677 	if (IS_ERR(key)) {
678 		err = PTR_ERR(key);
679 		goto err_put;
680 	}
681 
682 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
683 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
684 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
685 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
686 	else
687 		value_size = map->value_size;
688 
689 	err = -ENOMEM;
690 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
691 	if (!value)
692 		goto free_key;
693 
694 	err = -EFAULT;
695 	if (copy_from_user(value, uvalue, value_size) != 0)
696 		goto free_value;
697 
698 	/* Need to create a kthread, thus must support schedule */
699 	if (bpf_map_is_dev_bound(map)) {
700 		err = bpf_map_offload_update_elem(map, key, value, attr->flags);
701 		goto out;
702 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP) {
703 		err = map->ops->map_update_elem(map, key, value, attr->flags);
704 		goto out;
705 	}
706 
707 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
708 	 * inside bpf map update or delete otherwise deadlocks are possible
709 	 */
710 	preempt_disable();
711 	__this_cpu_inc(bpf_prog_active);
712 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
713 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
714 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
715 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
716 		err = bpf_percpu_array_update(map, key, value, attr->flags);
717 	} else if (IS_FD_ARRAY(map)) {
718 		rcu_read_lock();
719 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
720 						   attr->flags);
721 		rcu_read_unlock();
722 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
723 		rcu_read_lock();
724 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
725 						  attr->flags);
726 		rcu_read_unlock();
727 	} else {
728 		rcu_read_lock();
729 		err = map->ops->map_update_elem(map, key, value, attr->flags);
730 		rcu_read_unlock();
731 	}
732 	__this_cpu_dec(bpf_prog_active);
733 	preempt_enable();
734 out:
735 	if (!err)
736 		trace_bpf_map_update_elem(map, ufd, key, value);
737 free_value:
738 	kfree(value);
739 free_key:
740 	kfree(key);
741 err_put:
742 	fdput(f);
743 	return err;
744 }
745 
746 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
747 
748 static int map_delete_elem(union bpf_attr *attr)
749 {
750 	void __user *ukey = u64_to_user_ptr(attr->key);
751 	int ufd = attr->map_fd;
752 	struct bpf_map *map;
753 	struct fd f;
754 	void *key;
755 	int err;
756 
757 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
758 		return -EINVAL;
759 
760 	f = fdget(ufd);
761 	map = __bpf_map_get(f);
762 	if (IS_ERR(map))
763 		return PTR_ERR(map);
764 
765 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
766 		err = -EPERM;
767 		goto err_put;
768 	}
769 
770 	key = memdup_user(ukey, map->key_size);
771 	if (IS_ERR(key)) {
772 		err = PTR_ERR(key);
773 		goto err_put;
774 	}
775 
776 	if (bpf_map_is_dev_bound(map)) {
777 		err = bpf_map_offload_delete_elem(map, key);
778 		goto out;
779 	}
780 
781 	preempt_disable();
782 	__this_cpu_inc(bpf_prog_active);
783 	rcu_read_lock();
784 	err = map->ops->map_delete_elem(map, key);
785 	rcu_read_unlock();
786 	__this_cpu_dec(bpf_prog_active);
787 	preempt_enable();
788 out:
789 	if (!err)
790 		trace_bpf_map_delete_elem(map, ufd, key);
791 	kfree(key);
792 err_put:
793 	fdput(f);
794 	return err;
795 }
796 
797 /* last field in 'union bpf_attr' used by this command */
798 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
799 
800 static int map_get_next_key(union bpf_attr *attr)
801 {
802 	void __user *ukey = u64_to_user_ptr(attr->key);
803 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
804 	int ufd = attr->map_fd;
805 	struct bpf_map *map;
806 	void *key, *next_key;
807 	struct fd f;
808 	int err;
809 
810 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
811 		return -EINVAL;
812 
813 	f = fdget(ufd);
814 	map = __bpf_map_get(f);
815 	if (IS_ERR(map))
816 		return PTR_ERR(map);
817 
818 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
819 		err = -EPERM;
820 		goto err_put;
821 	}
822 
823 	if (ukey) {
824 		key = memdup_user(ukey, map->key_size);
825 		if (IS_ERR(key)) {
826 			err = PTR_ERR(key);
827 			goto err_put;
828 		}
829 	} else {
830 		key = NULL;
831 	}
832 
833 	err = -ENOMEM;
834 	next_key = kmalloc(map->key_size, GFP_USER);
835 	if (!next_key)
836 		goto free_key;
837 
838 	if (bpf_map_is_dev_bound(map)) {
839 		err = bpf_map_offload_get_next_key(map, key, next_key);
840 		goto out;
841 	}
842 
843 	rcu_read_lock();
844 	err = map->ops->map_get_next_key(map, key, next_key);
845 	rcu_read_unlock();
846 out:
847 	if (err)
848 		goto free_next_key;
849 
850 	err = -EFAULT;
851 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
852 		goto free_next_key;
853 
854 	trace_bpf_map_next_key(map, ufd, key, next_key);
855 	err = 0;
856 
857 free_next_key:
858 	kfree(next_key);
859 free_key:
860 	kfree(key);
861 err_put:
862 	fdput(f);
863 	return err;
864 }
865 
866 static const struct bpf_prog_ops * const bpf_prog_types[] = {
867 #define BPF_PROG_TYPE(_id, _name) \
868 	[_id] = & _name ## _prog_ops,
869 #define BPF_MAP_TYPE(_id, _ops)
870 #include <linux/bpf_types.h>
871 #undef BPF_PROG_TYPE
872 #undef BPF_MAP_TYPE
873 };
874 
875 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
876 {
877 	const struct bpf_prog_ops *ops;
878 
879 	if (type >= ARRAY_SIZE(bpf_prog_types))
880 		return -EINVAL;
881 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
882 	ops = bpf_prog_types[type];
883 	if (!ops)
884 		return -EINVAL;
885 
886 	if (!bpf_prog_is_dev_bound(prog->aux))
887 		prog->aux->ops = ops;
888 	else
889 		prog->aux->ops = &bpf_offload_prog_ops;
890 	prog->type = type;
891 	return 0;
892 }
893 
894 /* drop refcnt on maps used by eBPF program and free auxilary data */
895 static void free_used_maps(struct bpf_prog_aux *aux)
896 {
897 	int i;
898 
899 	for (i = 0; i < aux->used_map_cnt; i++)
900 		bpf_map_put(aux->used_maps[i]);
901 
902 	kfree(aux->used_maps);
903 }
904 
905 int __bpf_prog_charge(struct user_struct *user, u32 pages)
906 {
907 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
908 	unsigned long user_bufs;
909 
910 	if (user) {
911 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
912 		if (user_bufs > memlock_limit) {
913 			atomic_long_sub(pages, &user->locked_vm);
914 			return -EPERM;
915 		}
916 	}
917 
918 	return 0;
919 }
920 
921 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
922 {
923 	if (user)
924 		atomic_long_sub(pages, &user->locked_vm);
925 }
926 
927 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
928 {
929 	struct user_struct *user = get_current_user();
930 	int ret;
931 
932 	ret = __bpf_prog_charge(user, prog->pages);
933 	if (ret) {
934 		free_uid(user);
935 		return ret;
936 	}
937 
938 	prog->aux->user = user;
939 	return 0;
940 }
941 
942 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
943 {
944 	struct user_struct *user = prog->aux->user;
945 
946 	__bpf_prog_uncharge(user, prog->pages);
947 	free_uid(user);
948 }
949 
950 static int bpf_prog_alloc_id(struct bpf_prog *prog)
951 {
952 	int id;
953 
954 	idr_preload(GFP_KERNEL);
955 	spin_lock_bh(&prog_idr_lock);
956 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
957 	if (id > 0)
958 		prog->aux->id = id;
959 	spin_unlock_bh(&prog_idr_lock);
960 	idr_preload_end();
961 
962 	/* id is in [1, INT_MAX) */
963 	if (WARN_ON_ONCE(!id))
964 		return -ENOSPC;
965 
966 	return id > 0 ? 0 : id;
967 }
968 
969 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
970 {
971 	/* cBPF to eBPF migrations are currently not in the idr store.
972 	 * Offloaded programs are removed from the store when their device
973 	 * disappears - even if someone grabs an fd to them they are unusable,
974 	 * simply waiting for refcnt to drop to be freed.
975 	 */
976 	if (!prog->aux->id)
977 		return;
978 
979 	if (do_idr_lock)
980 		spin_lock_bh(&prog_idr_lock);
981 	else
982 		__acquire(&prog_idr_lock);
983 
984 	idr_remove(&prog_idr, prog->aux->id);
985 	prog->aux->id = 0;
986 
987 	if (do_idr_lock)
988 		spin_unlock_bh(&prog_idr_lock);
989 	else
990 		__release(&prog_idr_lock);
991 }
992 
993 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
994 {
995 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
996 
997 	free_used_maps(aux);
998 	bpf_prog_uncharge_memlock(aux->prog);
999 	security_bpf_prog_free(aux);
1000 	bpf_prog_free(aux->prog);
1001 }
1002 
1003 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1004 {
1005 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
1006 		int i;
1007 
1008 		trace_bpf_prog_put_rcu(prog);
1009 		/* bpf_prog_free_id() must be called first */
1010 		bpf_prog_free_id(prog, do_idr_lock);
1011 
1012 		for (i = 0; i < prog->aux->func_cnt; i++)
1013 			bpf_prog_kallsyms_del(prog->aux->func[i]);
1014 		bpf_prog_kallsyms_del(prog);
1015 
1016 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1017 	}
1018 }
1019 
1020 void bpf_prog_put(struct bpf_prog *prog)
1021 {
1022 	__bpf_prog_put(prog, true);
1023 }
1024 EXPORT_SYMBOL_GPL(bpf_prog_put);
1025 
1026 static int bpf_prog_release(struct inode *inode, struct file *filp)
1027 {
1028 	struct bpf_prog *prog = filp->private_data;
1029 
1030 	bpf_prog_put(prog);
1031 	return 0;
1032 }
1033 
1034 #ifdef CONFIG_PROC_FS
1035 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1036 {
1037 	const struct bpf_prog *prog = filp->private_data;
1038 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1039 
1040 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1041 	seq_printf(m,
1042 		   "prog_type:\t%u\n"
1043 		   "prog_jited:\t%u\n"
1044 		   "prog_tag:\t%s\n"
1045 		   "memlock:\t%llu\n",
1046 		   prog->type,
1047 		   prog->jited,
1048 		   prog_tag,
1049 		   prog->pages * 1ULL << PAGE_SHIFT);
1050 }
1051 #endif
1052 
1053 const struct file_operations bpf_prog_fops = {
1054 #ifdef CONFIG_PROC_FS
1055 	.show_fdinfo	= bpf_prog_show_fdinfo,
1056 #endif
1057 	.release	= bpf_prog_release,
1058 	.read		= bpf_dummy_read,
1059 	.write		= bpf_dummy_write,
1060 };
1061 
1062 int bpf_prog_new_fd(struct bpf_prog *prog)
1063 {
1064 	int ret;
1065 
1066 	ret = security_bpf_prog(prog);
1067 	if (ret < 0)
1068 		return ret;
1069 
1070 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1071 				O_RDWR | O_CLOEXEC);
1072 }
1073 
1074 static struct bpf_prog *____bpf_prog_get(struct fd f)
1075 {
1076 	if (!f.file)
1077 		return ERR_PTR(-EBADF);
1078 	if (f.file->f_op != &bpf_prog_fops) {
1079 		fdput(f);
1080 		return ERR_PTR(-EINVAL);
1081 	}
1082 
1083 	return f.file->private_data;
1084 }
1085 
1086 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1087 {
1088 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1089 		atomic_sub(i, &prog->aux->refcnt);
1090 		return ERR_PTR(-EBUSY);
1091 	}
1092 	return prog;
1093 }
1094 EXPORT_SYMBOL_GPL(bpf_prog_add);
1095 
1096 void bpf_prog_sub(struct bpf_prog *prog, int i)
1097 {
1098 	/* Only to be used for undoing previous bpf_prog_add() in some
1099 	 * error path. We still know that another entity in our call
1100 	 * path holds a reference to the program, thus atomic_sub() can
1101 	 * be safely used in such cases!
1102 	 */
1103 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1104 }
1105 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1106 
1107 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1108 {
1109 	return bpf_prog_add(prog, 1);
1110 }
1111 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1112 
1113 /* prog_idr_lock should have been held */
1114 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1115 {
1116 	int refold;
1117 
1118 	refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
1119 
1120 	if (refold >= BPF_MAX_REFCNT) {
1121 		__bpf_prog_put(prog, false);
1122 		return ERR_PTR(-EBUSY);
1123 	}
1124 
1125 	if (!refold)
1126 		return ERR_PTR(-ENOENT);
1127 
1128 	return prog;
1129 }
1130 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1131 
1132 bool bpf_prog_get_ok(struct bpf_prog *prog,
1133 			    enum bpf_prog_type *attach_type, bool attach_drv)
1134 {
1135 	/* not an attachment, just a refcount inc, always allow */
1136 	if (!attach_type)
1137 		return true;
1138 
1139 	if (prog->type != *attach_type)
1140 		return false;
1141 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1142 		return false;
1143 
1144 	return true;
1145 }
1146 
1147 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1148 				       bool attach_drv)
1149 {
1150 	struct fd f = fdget(ufd);
1151 	struct bpf_prog *prog;
1152 
1153 	prog = ____bpf_prog_get(f);
1154 	if (IS_ERR(prog))
1155 		return prog;
1156 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1157 		prog = ERR_PTR(-EINVAL);
1158 		goto out;
1159 	}
1160 
1161 	prog = bpf_prog_inc(prog);
1162 out:
1163 	fdput(f);
1164 	return prog;
1165 }
1166 
1167 struct bpf_prog *bpf_prog_get(u32 ufd)
1168 {
1169 	return __bpf_prog_get(ufd, NULL, false);
1170 }
1171 
1172 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1173 				       bool attach_drv)
1174 {
1175 	struct bpf_prog *prog = __bpf_prog_get(ufd, &type, attach_drv);
1176 
1177 	if (!IS_ERR(prog))
1178 		trace_bpf_prog_get_type(prog);
1179 	return prog;
1180 }
1181 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1182 
1183 /* Initially all BPF programs could be loaded w/o specifying
1184  * expected_attach_type. Later for some of them specifying expected_attach_type
1185  * at load time became required so that program could be validated properly.
1186  * Programs of types that are allowed to be loaded both w/ and w/o (for
1187  * backward compatibility) expected_attach_type, should have the default attach
1188  * type assigned to expected_attach_type for the latter case, so that it can be
1189  * validated later at attach time.
1190  *
1191  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1192  * prog type requires it but has some attach types that have to be backward
1193  * compatible.
1194  */
1195 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1196 {
1197 	switch (attr->prog_type) {
1198 	case BPF_PROG_TYPE_CGROUP_SOCK:
1199 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1200 		 * exist so checking for non-zero is the way to go here.
1201 		 */
1202 		if (!attr->expected_attach_type)
1203 			attr->expected_attach_type =
1204 				BPF_CGROUP_INET_SOCK_CREATE;
1205 		break;
1206 	}
1207 }
1208 
1209 static int
1210 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1211 				enum bpf_attach_type expected_attach_type)
1212 {
1213 	switch (prog_type) {
1214 	case BPF_PROG_TYPE_CGROUP_SOCK:
1215 		switch (expected_attach_type) {
1216 		case BPF_CGROUP_INET_SOCK_CREATE:
1217 		case BPF_CGROUP_INET4_POST_BIND:
1218 		case BPF_CGROUP_INET6_POST_BIND:
1219 			return 0;
1220 		default:
1221 			return -EINVAL;
1222 		}
1223 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1224 		switch (expected_attach_type) {
1225 		case BPF_CGROUP_INET4_BIND:
1226 		case BPF_CGROUP_INET6_BIND:
1227 		case BPF_CGROUP_INET4_CONNECT:
1228 		case BPF_CGROUP_INET6_CONNECT:
1229 			return 0;
1230 		default:
1231 			return -EINVAL;
1232 		}
1233 	default:
1234 		return 0;
1235 	}
1236 }
1237 
1238 /* last field in 'union bpf_attr' used by this command */
1239 #define	BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1240 
1241 static int bpf_prog_load(union bpf_attr *attr)
1242 {
1243 	enum bpf_prog_type type = attr->prog_type;
1244 	struct bpf_prog *prog;
1245 	int err;
1246 	char license[128];
1247 	bool is_gpl;
1248 
1249 	if (CHECK_ATTR(BPF_PROG_LOAD))
1250 		return -EINVAL;
1251 
1252 	if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1253 		return -EINVAL;
1254 
1255 	/* copy eBPF program license from user space */
1256 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1257 			      sizeof(license) - 1) < 0)
1258 		return -EFAULT;
1259 	license[sizeof(license) - 1] = 0;
1260 
1261 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1262 	is_gpl = license_is_gpl_compatible(license);
1263 
1264 	if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1265 		return -E2BIG;
1266 
1267 	if (type == BPF_PROG_TYPE_KPROBE &&
1268 	    attr->kern_version != LINUX_VERSION_CODE)
1269 		return -EINVAL;
1270 
1271 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1272 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
1273 	    !capable(CAP_SYS_ADMIN))
1274 		return -EPERM;
1275 
1276 	bpf_prog_load_fixup_attach_type(attr);
1277 	if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1278 		return -EINVAL;
1279 
1280 	/* plain bpf_prog allocation */
1281 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1282 	if (!prog)
1283 		return -ENOMEM;
1284 
1285 	prog->expected_attach_type = attr->expected_attach_type;
1286 
1287 	prog->aux->offload_requested = !!attr->prog_ifindex;
1288 
1289 	err = security_bpf_prog_alloc(prog->aux);
1290 	if (err)
1291 		goto free_prog_nouncharge;
1292 
1293 	err = bpf_prog_charge_memlock(prog);
1294 	if (err)
1295 		goto free_prog_sec;
1296 
1297 	prog->len = attr->insn_cnt;
1298 
1299 	err = -EFAULT;
1300 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1301 			   bpf_prog_insn_size(prog)) != 0)
1302 		goto free_prog;
1303 
1304 	prog->orig_prog = NULL;
1305 	prog->jited = 0;
1306 
1307 	atomic_set(&prog->aux->refcnt, 1);
1308 	prog->gpl_compatible = is_gpl ? 1 : 0;
1309 
1310 	if (bpf_prog_is_dev_bound(prog->aux)) {
1311 		err = bpf_prog_offload_init(prog, attr);
1312 		if (err)
1313 			goto free_prog;
1314 	}
1315 
1316 	/* find program type: socket_filter vs tracing_filter */
1317 	err = find_prog_type(type, prog);
1318 	if (err < 0)
1319 		goto free_prog;
1320 
1321 	prog->aux->load_time = ktime_get_boot_ns();
1322 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1323 	if (err)
1324 		goto free_prog;
1325 
1326 	/* run eBPF verifier */
1327 	err = bpf_check(&prog, attr);
1328 	if (err < 0)
1329 		goto free_used_maps;
1330 
1331 	/* eBPF program is ready to be JITed */
1332 	if (!prog->bpf_func)
1333 		prog = bpf_prog_select_runtime(prog, &err);
1334 	if (err < 0)
1335 		goto free_used_maps;
1336 
1337 	err = bpf_prog_alloc_id(prog);
1338 	if (err)
1339 		goto free_used_maps;
1340 
1341 	err = bpf_prog_new_fd(prog);
1342 	if (err < 0) {
1343 		/* failed to allocate fd.
1344 		 * bpf_prog_put() is needed because the above
1345 		 * bpf_prog_alloc_id() has published the prog
1346 		 * to the userspace and the userspace may
1347 		 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1348 		 */
1349 		bpf_prog_put(prog);
1350 		return err;
1351 	}
1352 
1353 	bpf_prog_kallsyms_add(prog);
1354 	trace_bpf_prog_load(prog, err);
1355 	return err;
1356 
1357 free_used_maps:
1358 	free_used_maps(prog->aux);
1359 free_prog:
1360 	bpf_prog_uncharge_memlock(prog);
1361 free_prog_sec:
1362 	security_bpf_prog_free(prog->aux);
1363 free_prog_nouncharge:
1364 	bpf_prog_free(prog);
1365 	return err;
1366 }
1367 
1368 #define BPF_OBJ_LAST_FIELD file_flags
1369 
1370 static int bpf_obj_pin(const union bpf_attr *attr)
1371 {
1372 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1373 		return -EINVAL;
1374 
1375 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1376 }
1377 
1378 static int bpf_obj_get(const union bpf_attr *attr)
1379 {
1380 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1381 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1382 		return -EINVAL;
1383 
1384 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1385 				attr->file_flags);
1386 }
1387 
1388 struct bpf_raw_tracepoint {
1389 	struct bpf_raw_event_map *btp;
1390 	struct bpf_prog *prog;
1391 };
1392 
1393 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1394 {
1395 	struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1396 
1397 	if (raw_tp->prog) {
1398 		bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1399 		bpf_prog_put(raw_tp->prog);
1400 	}
1401 	kfree(raw_tp);
1402 	return 0;
1403 }
1404 
1405 static const struct file_operations bpf_raw_tp_fops = {
1406 	.release	= bpf_raw_tracepoint_release,
1407 	.read		= bpf_dummy_read,
1408 	.write		= bpf_dummy_write,
1409 };
1410 
1411 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1412 
1413 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1414 {
1415 	struct bpf_raw_tracepoint *raw_tp;
1416 	struct bpf_raw_event_map *btp;
1417 	struct bpf_prog *prog;
1418 	char tp_name[128];
1419 	int tp_fd, err;
1420 
1421 	if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1422 			      sizeof(tp_name) - 1) < 0)
1423 		return -EFAULT;
1424 	tp_name[sizeof(tp_name) - 1] = 0;
1425 
1426 	btp = bpf_find_raw_tracepoint(tp_name);
1427 	if (!btp)
1428 		return -ENOENT;
1429 
1430 	raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1431 	if (!raw_tp)
1432 		return -ENOMEM;
1433 	raw_tp->btp = btp;
1434 
1435 	prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1436 				 BPF_PROG_TYPE_RAW_TRACEPOINT);
1437 	if (IS_ERR(prog)) {
1438 		err = PTR_ERR(prog);
1439 		goto out_free_tp;
1440 	}
1441 
1442 	err = bpf_probe_register(raw_tp->btp, prog);
1443 	if (err)
1444 		goto out_put_prog;
1445 
1446 	raw_tp->prog = prog;
1447 	tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1448 				 O_CLOEXEC);
1449 	if (tp_fd < 0) {
1450 		bpf_probe_unregister(raw_tp->btp, prog);
1451 		err = tp_fd;
1452 		goto out_put_prog;
1453 	}
1454 	return tp_fd;
1455 
1456 out_put_prog:
1457 	bpf_prog_put(prog);
1458 out_free_tp:
1459 	kfree(raw_tp);
1460 	return err;
1461 }
1462 
1463 #ifdef CONFIG_CGROUP_BPF
1464 
1465 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1466 					     enum bpf_attach_type attach_type)
1467 {
1468 	switch (prog->type) {
1469 	case BPF_PROG_TYPE_CGROUP_SOCK:
1470 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1471 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1472 	default:
1473 		return 0;
1474 	}
1475 }
1476 
1477 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1478 
1479 static int sockmap_get_from_fd(const union bpf_attr *attr,
1480 			       int type, bool attach)
1481 {
1482 	struct bpf_prog *prog = NULL;
1483 	int ufd = attr->target_fd;
1484 	struct bpf_map *map;
1485 	struct fd f;
1486 	int err;
1487 
1488 	f = fdget(ufd);
1489 	map = __bpf_map_get(f);
1490 	if (IS_ERR(map))
1491 		return PTR_ERR(map);
1492 
1493 	if (attach) {
1494 		prog = bpf_prog_get_type(attr->attach_bpf_fd, type);
1495 		if (IS_ERR(prog)) {
1496 			fdput(f);
1497 			return PTR_ERR(prog);
1498 		}
1499 	}
1500 
1501 	err = sock_map_prog(map, prog, attr->attach_type);
1502 	if (err) {
1503 		fdput(f);
1504 		if (prog)
1505 			bpf_prog_put(prog);
1506 		return err;
1507 	}
1508 
1509 	fdput(f);
1510 	return 0;
1511 }
1512 
1513 #define BPF_F_ATTACH_MASK \
1514 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1515 
1516 static int bpf_prog_attach(const union bpf_attr *attr)
1517 {
1518 	enum bpf_prog_type ptype;
1519 	struct bpf_prog *prog;
1520 	struct cgroup *cgrp;
1521 	int ret;
1522 
1523 	if (!capable(CAP_NET_ADMIN))
1524 		return -EPERM;
1525 
1526 	if (CHECK_ATTR(BPF_PROG_ATTACH))
1527 		return -EINVAL;
1528 
1529 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1530 		return -EINVAL;
1531 
1532 	switch (attr->attach_type) {
1533 	case BPF_CGROUP_INET_INGRESS:
1534 	case BPF_CGROUP_INET_EGRESS:
1535 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1536 		break;
1537 	case BPF_CGROUP_INET_SOCK_CREATE:
1538 	case BPF_CGROUP_INET4_POST_BIND:
1539 	case BPF_CGROUP_INET6_POST_BIND:
1540 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1541 		break;
1542 	case BPF_CGROUP_INET4_BIND:
1543 	case BPF_CGROUP_INET6_BIND:
1544 	case BPF_CGROUP_INET4_CONNECT:
1545 	case BPF_CGROUP_INET6_CONNECT:
1546 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1547 		break;
1548 	case BPF_CGROUP_SOCK_OPS:
1549 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1550 		break;
1551 	case BPF_CGROUP_DEVICE:
1552 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1553 		break;
1554 	case BPF_SK_MSG_VERDICT:
1555 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, true);
1556 	case BPF_SK_SKB_STREAM_PARSER:
1557 	case BPF_SK_SKB_STREAM_VERDICT:
1558 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, true);
1559 	default:
1560 		return -EINVAL;
1561 	}
1562 
1563 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1564 	if (IS_ERR(prog))
1565 		return PTR_ERR(prog);
1566 
1567 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1568 		bpf_prog_put(prog);
1569 		return -EINVAL;
1570 	}
1571 
1572 	cgrp = cgroup_get_from_fd(attr->target_fd);
1573 	if (IS_ERR(cgrp)) {
1574 		bpf_prog_put(prog);
1575 		return PTR_ERR(cgrp);
1576 	}
1577 
1578 	ret = cgroup_bpf_attach(cgrp, prog, attr->attach_type,
1579 				attr->attach_flags);
1580 	if (ret)
1581 		bpf_prog_put(prog);
1582 	cgroup_put(cgrp);
1583 
1584 	return ret;
1585 }
1586 
1587 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1588 
1589 static int bpf_prog_detach(const union bpf_attr *attr)
1590 {
1591 	enum bpf_prog_type ptype;
1592 	struct bpf_prog *prog;
1593 	struct cgroup *cgrp;
1594 	int ret;
1595 
1596 	if (!capable(CAP_NET_ADMIN))
1597 		return -EPERM;
1598 
1599 	if (CHECK_ATTR(BPF_PROG_DETACH))
1600 		return -EINVAL;
1601 
1602 	switch (attr->attach_type) {
1603 	case BPF_CGROUP_INET_INGRESS:
1604 	case BPF_CGROUP_INET_EGRESS:
1605 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1606 		break;
1607 	case BPF_CGROUP_INET_SOCK_CREATE:
1608 	case BPF_CGROUP_INET4_POST_BIND:
1609 	case BPF_CGROUP_INET6_POST_BIND:
1610 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1611 		break;
1612 	case BPF_CGROUP_INET4_BIND:
1613 	case BPF_CGROUP_INET6_BIND:
1614 	case BPF_CGROUP_INET4_CONNECT:
1615 	case BPF_CGROUP_INET6_CONNECT:
1616 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1617 		break;
1618 	case BPF_CGROUP_SOCK_OPS:
1619 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1620 		break;
1621 	case BPF_CGROUP_DEVICE:
1622 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1623 		break;
1624 	case BPF_SK_MSG_VERDICT:
1625 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, false);
1626 	case BPF_SK_SKB_STREAM_PARSER:
1627 	case BPF_SK_SKB_STREAM_VERDICT:
1628 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, false);
1629 	default:
1630 		return -EINVAL;
1631 	}
1632 
1633 	cgrp = cgroup_get_from_fd(attr->target_fd);
1634 	if (IS_ERR(cgrp))
1635 		return PTR_ERR(cgrp);
1636 
1637 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1638 	if (IS_ERR(prog))
1639 		prog = NULL;
1640 
1641 	ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type, 0);
1642 	if (prog)
1643 		bpf_prog_put(prog);
1644 	cgroup_put(cgrp);
1645 	return ret;
1646 }
1647 
1648 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1649 
1650 static int bpf_prog_query(const union bpf_attr *attr,
1651 			  union bpf_attr __user *uattr)
1652 {
1653 	struct cgroup *cgrp;
1654 	int ret;
1655 
1656 	if (!capable(CAP_NET_ADMIN))
1657 		return -EPERM;
1658 	if (CHECK_ATTR(BPF_PROG_QUERY))
1659 		return -EINVAL;
1660 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1661 		return -EINVAL;
1662 
1663 	switch (attr->query.attach_type) {
1664 	case BPF_CGROUP_INET_INGRESS:
1665 	case BPF_CGROUP_INET_EGRESS:
1666 	case BPF_CGROUP_INET_SOCK_CREATE:
1667 	case BPF_CGROUP_INET4_BIND:
1668 	case BPF_CGROUP_INET6_BIND:
1669 	case BPF_CGROUP_INET4_POST_BIND:
1670 	case BPF_CGROUP_INET6_POST_BIND:
1671 	case BPF_CGROUP_INET4_CONNECT:
1672 	case BPF_CGROUP_INET6_CONNECT:
1673 	case BPF_CGROUP_SOCK_OPS:
1674 	case BPF_CGROUP_DEVICE:
1675 		break;
1676 	default:
1677 		return -EINVAL;
1678 	}
1679 	cgrp = cgroup_get_from_fd(attr->query.target_fd);
1680 	if (IS_ERR(cgrp))
1681 		return PTR_ERR(cgrp);
1682 	ret = cgroup_bpf_query(cgrp, attr, uattr);
1683 	cgroup_put(cgrp);
1684 	return ret;
1685 }
1686 #endif /* CONFIG_CGROUP_BPF */
1687 
1688 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1689 
1690 static int bpf_prog_test_run(const union bpf_attr *attr,
1691 			     union bpf_attr __user *uattr)
1692 {
1693 	struct bpf_prog *prog;
1694 	int ret = -ENOTSUPP;
1695 
1696 	if (!capable(CAP_SYS_ADMIN))
1697 		return -EPERM;
1698 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1699 		return -EINVAL;
1700 
1701 	prog = bpf_prog_get(attr->test.prog_fd);
1702 	if (IS_ERR(prog))
1703 		return PTR_ERR(prog);
1704 
1705 	if (prog->aux->ops->test_run)
1706 		ret = prog->aux->ops->test_run(prog, attr, uattr);
1707 
1708 	bpf_prog_put(prog);
1709 	return ret;
1710 }
1711 
1712 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1713 
1714 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1715 			       union bpf_attr __user *uattr,
1716 			       struct idr *idr,
1717 			       spinlock_t *lock)
1718 {
1719 	u32 next_id = attr->start_id;
1720 	int err = 0;
1721 
1722 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1723 		return -EINVAL;
1724 
1725 	if (!capable(CAP_SYS_ADMIN))
1726 		return -EPERM;
1727 
1728 	next_id++;
1729 	spin_lock_bh(lock);
1730 	if (!idr_get_next(idr, &next_id))
1731 		err = -ENOENT;
1732 	spin_unlock_bh(lock);
1733 
1734 	if (!err)
1735 		err = put_user(next_id, &uattr->next_id);
1736 
1737 	return err;
1738 }
1739 
1740 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1741 
1742 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1743 {
1744 	struct bpf_prog *prog;
1745 	u32 id = attr->prog_id;
1746 	int fd;
1747 
1748 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1749 		return -EINVAL;
1750 
1751 	if (!capable(CAP_SYS_ADMIN))
1752 		return -EPERM;
1753 
1754 	spin_lock_bh(&prog_idr_lock);
1755 	prog = idr_find(&prog_idr, id);
1756 	if (prog)
1757 		prog = bpf_prog_inc_not_zero(prog);
1758 	else
1759 		prog = ERR_PTR(-ENOENT);
1760 	spin_unlock_bh(&prog_idr_lock);
1761 
1762 	if (IS_ERR(prog))
1763 		return PTR_ERR(prog);
1764 
1765 	fd = bpf_prog_new_fd(prog);
1766 	if (fd < 0)
1767 		bpf_prog_put(prog);
1768 
1769 	return fd;
1770 }
1771 
1772 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1773 
1774 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1775 {
1776 	struct bpf_map *map;
1777 	u32 id = attr->map_id;
1778 	int f_flags;
1779 	int fd;
1780 
1781 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1782 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1783 		return -EINVAL;
1784 
1785 	if (!capable(CAP_SYS_ADMIN))
1786 		return -EPERM;
1787 
1788 	f_flags = bpf_get_file_flag(attr->open_flags);
1789 	if (f_flags < 0)
1790 		return f_flags;
1791 
1792 	spin_lock_bh(&map_idr_lock);
1793 	map = idr_find(&map_idr, id);
1794 	if (map)
1795 		map = bpf_map_inc_not_zero(map, true);
1796 	else
1797 		map = ERR_PTR(-ENOENT);
1798 	spin_unlock_bh(&map_idr_lock);
1799 
1800 	if (IS_ERR(map))
1801 		return PTR_ERR(map);
1802 
1803 	fd = bpf_map_new_fd(map, f_flags);
1804 	if (fd < 0)
1805 		bpf_map_put(map);
1806 
1807 	return fd;
1808 }
1809 
1810 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1811 					      unsigned long addr)
1812 {
1813 	int i;
1814 
1815 	for (i = 0; i < prog->aux->used_map_cnt; i++)
1816 		if (prog->aux->used_maps[i] == (void *)addr)
1817 			return prog->aux->used_maps[i];
1818 	return NULL;
1819 }
1820 
1821 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1822 {
1823 	const struct bpf_map *map;
1824 	struct bpf_insn *insns;
1825 	u64 imm;
1826 	int i;
1827 
1828 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1829 			GFP_USER);
1830 	if (!insns)
1831 		return insns;
1832 
1833 	for (i = 0; i < prog->len; i++) {
1834 		if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1835 			insns[i].code = BPF_JMP | BPF_CALL;
1836 			insns[i].imm = BPF_FUNC_tail_call;
1837 			/* fall-through */
1838 		}
1839 		if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1840 		    insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1841 			if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1842 				insns[i].code = BPF_JMP | BPF_CALL;
1843 			if (!bpf_dump_raw_ok())
1844 				insns[i].imm = 0;
1845 			continue;
1846 		}
1847 
1848 		if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1849 			continue;
1850 
1851 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1852 		map = bpf_map_from_imm(prog, imm);
1853 		if (map) {
1854 			insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1855 			insns[i].imm = map->id;
1856 			insns[i + 1].imm = 0;
1857 			continue;
1858 		}
1859 
1860 		if (!bpf_dump_raw_ok() &&
1861 		    imm == (unsigned long)prog->aux) {
1862 			insns[i].imm = 0;
1863 			insns[i + 1].imm = 0;
1864 			continue;
1865 		}
1866 	}
1867 
1868 	return insns;
1869 }
1870 
1871 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1872 				   const union bpf_attr *attr,
1873 				   union bpf_attr __user *uattr)
1874 {
1875 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1876 	struct bpf_prog_info info = {};
1877 	u32 info_len = attr->info.info_len;
1878 	char __user *uinsns;
1879 	u32 ulen;
1880 	int err;
1881 
1882 	err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1883 	if (err)
1884 		return err;
1885 	info_len = min_t(u32, sizeof(info), info_len);
1886 
1887 	if (copy_from_user(&info, uinfo, info_len))
1888 		return -EFAULT;
1889 
1890 	info.type = prog->type;
1891 	info.id = prog->aux->id;
1892 	info.load_time = prog->aux->load_time;
1893 	info.created_by_uid = from_kuid_munged(current_user_ns(),
1894 					       prog->aux->user->uid);
1895 
1896 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
1897 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1898 
1899 	ulen = info.nr_map_ids;
1900 	info.nr_map_ids = prog->aux->used_map_cnt;
1901 	ulen = min_t(u32, info.nr_map_ids, ulen);
1902 	if (ulen) {
1903 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1904 		u32 i;
1905 
1906 		for (i = 0; i < ulen; i++)
1907 			if (put_user(prog->aux->used_maps[i]->id,
1908 				     &user_map_ids[i]))
1909 				return -EFAULT;
1910 	}
1911 
1912 	if (!capable(CAP_SYS_ADMIN)) {
1913 		info.jited_prog_len = 0;
1914 		info.xlated_prog_len = 0;
1915 		goto done;
1916 	}
1917 
1918 	ulen = info.xlated_prog_len;
1919 	info.xlated_prog_len = bpf_prog_insn_size(prog);
1920 	if (info.xlated_prog_len && ulen) {
1921 		struct bpf_insn *insns_sanitized;
1922 		bool fault;
1923 
1924 		if (prog->blinded && !bpf_dump_raw_ok()) {
1925 			info.xlated_prog_insns = 0;
1926 			goto done;
1927 		}
1928 		insns_sanitized = bpf_insn_prepare_dump(prog);
1929 		if (!insns_sanitized)
1930 			return -ENOMEM;
1931 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1932 		ulen = min_t(u32, info.xlated_prog_len, ulen);
1933 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
1934 		kfree(insns_sanitized);
1935 		if (fault)
1936 			return -EFAULT;
1937 	}
1938 
1939 	if (bpf_prog_is_dev_bound(prog->aux)) {
1940 		err = bpf_prog_offload_info_fill(&info, prog);
1941 		if (err)
1942 			return err;
1943 		goto done;
1944 	}
1945 
1946 	/* NOTE: the following code is supposed to be skipped for offload.
1947 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
1948 	 * for offload.
1949 	 */
1950 	ulen = info.jited_prog_len;
1951 	info.jited_prog_len = prog->jited_len;
1952 	if (info.jited_prog_len && ulen) {
1953 		if (bpf_dump_raw_ok()) {
1954 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
1955 			ulen = min_t(u32, info.jited_prog_len, ulen);
1956 			if (copy_to_user(uinsns, prog->bpf_func, ulen))
1957 				return -EFAULT;
1958 		} else {
1959 			info.jited_prog_insns = 0;
1960 		}
1961 	}
1962 
1963 done:
1964 	if (copy_to_user(uinfo, &info, info_len) ||
1965 	    put_user(info_len, &uattr->info.info_len))
1966 		return -EFAULT;
1967 
1968 	return 0;
1969 }
1970 
1971 static int bpf_map_get_info_by_fd(struct bpf_map *map,
1972 				  const union bpf_attr *attr,
1973 				  union bpf_attr __user *uattr)
1974 {
1975 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1976 	struct bpf_map_info info = {};
1977 	u32 info_len = attr->info.info_len;
1978 	int err;
1979 
1980 	err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1981 	if (err)
1982 		return err;
1983 	info_len = min_t(u32, sizeof(info), info_len);
1984 
1985 	info.type = map->map_type;
1986 	info.id = map->id;
1987 	info.key_size = map->key_size;
1988 	info.value_size = map->value_size;
1989 	info.max_entries = map->max_entries;
1990 	info.map_flags = map->map_flags;
1991 	memcpy(info.name, map->name, sizeof(map->name));
1992 
1993 	if (bpf_map_is_dev_bound(map)) {
1994 		err = bpf_map_offload_info_fill(&info, map);
1995 		if (err)
1996 			return err;
1997 	}
1998 
1999 	if (copy_to_user(uinfo, &info, info_len) ||
2000 	    put_user(info_len, &uattr->info.info_len))
2001 		return -EFAULT;
2002 
2003 	return 0;
2004 }
2005 
2006 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2007 
2008 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2009 				  union bpf_attr __user *uattr)
2010 {
2011 	int ufd = attr->info.bpf_fd;
2012 	struct fd f;
2013 	int err;
2014 
2015 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2016 		return -EINVAL;
2017 
2018 	f = fdget(ufd);
2019 	if (!f.file)
2020 		return -EBADFD;
2021 
2022 	if (f.file->f_op == &bpf_prog_fops)
2023 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2024 					      uattr);
2025 	else if (f.file->f_op == &bpf_map_fops)
2026 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2027 					     uattr);
2028 	else
2029 		err = -EINVAL;
2030 
2031 	fdput(f);
2032 	return err;
2033 }
2034 
2035 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2036 {
2037 	union bpf_attr attr = {};
2038 	int err;
2039 
2040 	if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2041 		return -EPERM;
2042 
2043 	err = check_uarg_tail_zero(uattr, sizeof(attr), size);
2044 	if (err)
2045 		return err;
2046 	size = min_t(u32, size, sizeof(attr));
2047 
2048 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
2049 	if (copy_from_user(&attr, uattr, size) != 0)
2050 		return -EFAULT;
2051 
2052 	err = security_bpf(cmd, &attr, size);
2053 	if (err < 0)
2054 		return err;
2055 
2056 	switch (cmd) {
2057 	case BPF_MAP_CREATE:
2058 		err = map_create(&attr);
2059 		break;
2060 	case BPF_MAP_LOOKUP_ELEM:
2061 		err = map_lookup_elem(&attr);
2062 		break;
2063 	case BPF_MAP_UPDATE_ELEM:
2064 		err = map_update_elem(&attr);
2065 		break;
2066 	case BPF_MAP_DELETE_ELEM:
2067 		err = map_delete_elem(&attr);
2068 		break;
2069 	case BPF_MAP_GET_NEXT_KEY:
2070 		err = map_get_next_key(&attr);
2071 		break;
2072 	case BPF_PROG_LOAD:
2073 		err = bpf_prog_load(&attr);
2074 		break;
2075 	case BPF_OBJ_PIN:
2076 		err = bpf_obj_pin(&attr);
2077 		break;
2078 	case BPF_OBJ_GET:
2079 		err = bpf_obj_get(&attr);
2080 		break;
2081 #ifdef CONFIG_CGROUP_BPF
2082 	case BPF_PROG_ATTACH:
2083 		err = bpf_prog_attach(&attr);
2084 		break;
2085 	case BPF_PROG_DETACH:
2086 		err = bpf_prog_detach(&attr);
2087 		break;
2088 	case BPF_PROG_QUERY:
2089 		err = bpf_prog_query(&attr, uattr);
2090 		break;
2091 #endif
2092 	case BPF_PROG_TEST_RUN:
2093 		err = bpf_prog_test_run(&attr, uattr);
2094 		break;
2095 	case BPF_PROG_GET_NEXT_ID:
2096 		err = bpf_obj_get_next_id(&attr, uattr,
2097 					  &prog_idr, &prog_idr_lock);
2098 		break;
2099 	case BPF_MAP_GET_NEXT_ID:
2100 		err = bpf_obj_get_next_id(&attr, uattr,
2101 					  &map_idr, &map_idr_lock);
2102 		break;
2103 	case BPF_PROG_GET_FD_BY_ID:
2104 		err = bpf_prog_get_fd_by_id(&attr);
2105 		break;
2106 	case BPF_MAP_GET_FD_BY_ID:
2107 		err = bpf_map_get_fd_by_id(&attr);
2108 		break;
2109 	case BPF_OBJ_GET_INFO_BY_FD:
2110 		err = bpf_obj_get_info_by_fd(&attr, uattr);
2111 		break;
2112 	case BPF_RAW_TRACEPOINT_OPEN:
2113 		err = bpf_raw_tracepoint_open(&attr);
2114 		break;
2115 	default:
2116 		err = -EINVAL;
2117 		break;
2118 	}
2119 
2120 	return err;
2121 }
2122