xref: /linux/kernel/bpf/syscall.c (revision 9bacbced0e32204deb8b9d011279f9beddd8c2ef)
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/btf.h>
34 #include <linux/nospec.h>
35 
36 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
37 			   (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
38 			   (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
39 			   (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
40 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
41 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
42 
43 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
44 
45 DEFINE_PER_CPU(int, bpf_prog_active);
46 static DEFINE_IDR(prog_idr);
47 static DEFINE_SPINLOCK(prog_idr_lock);
48 static DEFINE_IDR(map_idr);
49 static DEFINE_SPINLOCK(map_idr_lock);
50 
51 int sysctl_unprivileged_bpf_disabled __read_mostly;
52 
53 static const struct bpf_map_ops * const bpf_map_types[] = {
54 #define BPF_PROG_TYPE(_id, _ops)
55 #define BPF_MAP_TYPE(_id, _ops) \
56 	[_id] = &_ops,
57 #include <linux/bpf_types.h>
58 #undef BPF_PROG_TYPE
59 #undef BPF_MAP_TYPE
60 };
61 
62 /*
63  * If we're handed a bigger struct than we know of, ensure all the unknown bits
64  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
65  * we don't know about yet.
66  *
67  * There is a ToCToU between this function call and the following
68  * copy_from_user() call. However, this is not a concern since this function is
69  * meant to be a future-proofing of bits.
70  */
71 int bpf_check_uarg_tail_zero(void __user *uaddr,
72 			     size_t expected_size,
73 			     size_t actual_size)
74 {
75 	unsigned char __user *addr;
76 	unsigned char __user *end;
77 	unsigned char val;
78 	int err;
79 
80 	if (unlikely(actual_size > PAGE_SIZE))	/* silly large */
81 		return -E2BIG;
82 
83 	if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
84 		return -EFAULT;
85 
86 	if (actual_size <= expected_size)
87 		return 0;
88 
89 	addr = uaddr + expected_size;
90 	end  = uaddr + actual_size;
91 
92 	for (; addr < end; addr++) {
93 		err = get_user(val, addr);
94 		if (err)
95 			return err;
96 		if (val)
97 			return -E2BIG;
98 	}
99 
100 	return 0;
101 }
102 
103 const struct bpf_map_ops bpf_map_offload_ops = {
104 	.map_alloc = bpf_map_offload_map_alloc,
105 	.map_free = bpf_map_offload_map_free,
106 };
107 
108 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
109 {
110 	const struct bpf_map_ops *ops;
111 	u32 type = attr->map_type;
112 	struct bpf_map *map;
113 	int err;
114 
115 	if (type >= ARRAY_SIZE(bpf_map_types))
116 		return ERR_PTR(-EINVAL);
117 	type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
118 	ops = bpf_map_types[type];
119 	if (!ops)
120 		return ERR_PTR(-EINVAL);
121 
122 	if (ops->map_alloc_check) {
123 		err = ops->map_alloc_check(attr);
124 		if (err)
125 			return ERR_PTR(err);
126 	}
127 	if (attr->map_ifindex)
128 		ops = &bpf_map_offload_ops;
129 	map = ops->map_alloc(attr);
130 	if (IS_ERR(map))
131 		return map;
132 	map->ops = ops;
133 	map->map_type = type;
134 	return map;
135 }
136 
137 void *bpf_map_area_alloc(size_t size, int numa_node)
138 {
139 	/* We definitely need __GFP_NORETRY, so OOM killer doesn't
140 	 * trigger under memory pressure as we really just want to
141 	 * fail instead.
142 	 */
143 	const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
144 	void *area;
145 
146 	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
147 		area = kmalloc_node(size, GFP_USER | flags, numa_node);
148 		if (area != NULL)
149 			return area;
150 	}
151 
152 	return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
153 					   __builtin_return_address(0));
154 }
155 
156 void bpf_map_area_free(void *area)
157 {
158 	kvfree(area);
159 }
160 
161 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
162 {
163 	map->map_type = attr->map_type;
164 	map->key_size = attr->key_size;
165 	map->value_size = attr->value_size;
166 	map->max_entries = attr->max_entries;
167 	map->map_flags = attr->map_flags;
168 	map->numa_node = bpf_map_attr_numa_node(attr);
169 }
170 
171 int bpf_map_precharge_memlock(u32 pages)
172 {
173 	struct user_struct *user = get_current_user();
174 	unsigned long memlock_limit, cur;
175 
176 	memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
177 	cur = atomic_long_read(&user->locked_vm);
178 	free_uid(user);
179 	if (cur + pages > memlock_limit)
180 		return -EPERM;
181 	return 0;
182 }
183 
184 static int bpf_map_charge_memlock(struct bpf_map *map)
185 {
186 	struct user_struct *user = get_current_user();
187 	unsigned long memlock_limit;
188 
189 	memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
190 
191 	atomic_long_add(map->pages, &user->locked_vm);
192 
193 	if (atomic_long_read(&user->locked_vm) > memlock_limit) {
194 		atomic_long_sub(map->pages, &user->locked_vm);
195 		free_uid(user);
196 		return -EPERM;
197 	}
198 	map->user = user;
199 	return 0;
200 }
201 
202 static void bpf_map_uncharge_memlock(struct bpf_map *map)
203 {
204 	struct user_struct *user = map->user;
205 
206 	atomic_long_sub(map->pages, &user->locked_vm);
207 	free_uid(user);
208 }
209 
210 static int bpf_map_alloc_id(struct bpf_map *map)
211 {
212 	int id;
213 
214 	idr_preload(GFP_KERNEL);
215 	spin_lock_bh(&map_idr_lock);
216 	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
217 	if (id > 0)
218 		map->id = id;
219 	spin_unlock_bh(&map_idr_lock);
220 	idr_preload_end();
221 
222 	if (WARN_ON_ONCE(!id))
223 		return -ENOSPC;
224 
225 	return id > 0 ? 0 : id;
226 }
227 
228 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
229 {
230 	unsigned long flags;
231 
232 	/* Offloaded maps are removed from the IDR store when their device
233 	 * disappears - even if someone holds an fd to them they are unusable,
234 	 * the memory is gone, all ops will fail; they are simply waiting for
235 	 * refcnt to drop to be freed.
236 	 */
237 	if (!map->id)
238 		return;
239 
240 	if (do_idr_lock)
241 		spin_lock_irqsave(&map_idr_lock, flags);
242 	else
243 		__acquire(&map_idr_lock);
244 
245 	idr_remove(&map_idr, map->id);
246 	map->id = 0;
247 
248 	if (do_idr_lock)
249 		spin_unlock_irqrestore(&map_idr_lock, flags);
250 	else
251 		__release(&map_idr_lock);
252 }
253 
254 /* called from workqueue */
255 static void bpf_map_free_deferred(struct work_struct *work)
256 {
257 	struct bpf_map *map = container_of(work, struct bpf_map, work);
258 
259 	bpf_map_uncharge_memlock(map);
260 	security_bpf_map_free(map);
261 	/* implementation dependent freeing */
262 	map->ops->map_free(map);
263 }
264 
265 static void bpf_map_put_uref(struct bpf_map *map)
266 {
267 	if (atomic_dec_and_test(&map->usercnt)) {
268 		if (map->ops->map_release_uref)
269 			map->ops->map_release_uref(map);
270 	}
271 }
272 
273 /* decrement map refcnt and schedule it for freeing via workqueue
274  * (unrelying map implementation ops->map_free() might sleep)
275  */
276 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
277 {
278 	if (atomic_dec_and_test(&map->refcnt)) {
279 		/* bpf_map_free_id() must be called first */
280 		bpf_map_free_id(map, do_idr_lock);
281 		btf_put(map->btf);
282 		INIT_WORK(&map->work, bpf_map_free_deferred);
283 		schedule_work(&map->work);
284 	}
285 }
286 
287 void bpf_map_put(struct bpf_map *map)
288 {
289 	__bpf_map_put(map, true);
290 }
291 EXPORT_SYMBOL_GPL(bpf_map_put);
292 
293 void bpf_map_put_with_uref(struct bpf_map *map)
294 {
295 	bpf_map_put_uref(map);
296 	bpf_map_put(map);
297 }
298 
299 static int bpf_map_release(struct inode *inode, struct file *filp)
300 {
301 	struct bpf_map *map = filp->private_data;
302 
303 	if (map->ops->map_release)
304 		map->ops->map_release(map, filp);
305 
306 	bpf_map_put_with_uref(map);
307 	return 0;
308 }
309 
310 #ifdef CONFIG_PROC_FS
311 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
312 {
313 	const struct bpf_map *map = filp->private_data;
314 	const struct bpf_array *array;
315 	u32 owner_prog_type = 0;
316 	u32 owner_jited = 0;
317 
318 	if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
319 		array = container_of(map, struct bpf_array, map);
320 		owner_prog_type = array->owner_prog_type;
321 		owner_jited = array->owner_jited;
322 	}
323 
324 	seq_printf(m,
325 		   "map_type:\t%u\n"
326 		   "key_size:\t%u\n"
327 		   "value_size:\t%u\n"
328 		   "max_entries:\t%u\n"
329 		   "map_flags:\t%#x\n"
330 		   "memlock:\t%llu\n"
331 		   "map_id:\t%u\n",
332 		   map->map_type,
333 		   map->key_size,
334 		   map->value_size,
335 		   map->max_entries,
336 		   map->map_flags,
337 		   map->pages * 1ULL << PAGE_SHIFT,
338 		   map->id);
339 
340 	if (owner_prog_type) {
341 		seq_printf(m, "owner_prog_type:\t%u\n",
342 			   owner_prog_type);
343 		seq_printf(m, "owner_jited:\t%u\n",
344 			   owner_jited);
345 	}
346 }
347 #endif
348 
349 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
350 			      loff_t *ppos)
351 {
352 	/* We need this handler such that alloc_file() enables
353 	 * f_mode with FMODE_CAN_READ.
354 	 */
355 	return -EINVAL;
356 }
357 
358 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
359 			       size_t siz, loff_t *ppos)
360 {
361 	/* We need this handler such that alloc_file() enables
362 	 * f_mode with FMODE_CAN_WRITE.
363 	 */
364 	return -EINVAL;
365 }
366 
367 const struct file_operations bpf_map_fops = {
368 #ifdef CONFIG_PROC_FS
369 	.show_fdinfo	= bpf_map_show_fdinfo,
370 #endif
371 	.release	= bpf_map_release,
372 	.read		= bpf_dummy_read,
373 	.write		= bpf_dummy_write,
374 };
375 
376 int bpf_map_new_fd(struct bpf_map *map, int flags)
377 {
378 	int ret;
379 
380 	ret = security_bpf_map(map, OPEN_FMODE(flags));
381 	if (ret < 0)
382 		return ret;
383 
384 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
385 				flags | O_CLOEXEC);
386 }
387 
388 int bpf_get_file_flag(int flags)
389 {
390 	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
391 		return -EINVAL;
392 	if (flags & BPF_F_RDONLY)
393 		return O_RDONLY;
394 	if (flags & BPF_F_WRONLY)
395 		return O_WRONLY;
396 	return O_RDWR;
397 }
398 
399 /* helper macro to check that unused fields 'union bpf_attr' are zero */
400 #define CHECK_ATTR(CMD) \
401 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
402 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
403 		   sizeof(*attr) - \
404 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
405 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
406 
407 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
408  * Return 0 on success and < 0 on error.
409  */
410 static int bpf_obj_name_cpy(char *dst, const char *src)
411 {
412 	const char *end = src + BPF_OBJ_NAME_LEN;
413 
414 	memset(dst, 0, BPF_OBJ_NAME_LEN);
415 
416 	/* Copy all isalnum() and '_' char */
417 	while (src < end && *src) {
418 		if (!isalnum(*src) && *src != '_')
419 			return -EINVAL;
420 		*dst++ = *src++;
421 	}
422 
423 	/* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
424 	if (src == end)
425 		return -EINVAL;
426 
427 	return 0;
428 }
429 
430 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
431 /* called via syscall */
432 static int map_create(union bpf_attr *attr)
433 {
434 	int numa_node = bpf_map_attr_numa_node(attr);
435 	struct bpf_map *map;
436 	int f_flags;
437 	int err;
438 
439 	err = CHECK_ATTR(BPF_MAP_CREATE);
440 	if (err)
441 		return -EINVAL;
442 
443 	f_flags = bpf_get_file_flag(attr->map_flags);
444 	if (f_flags < 0)
445 		return f_flags;
446 
447 	if (numa_node != NUMA_NO_NODE &&
448 	    ((unsigned int)numa_node >= nr_node_ids ||
449 	     !node_online(numa_node)))
450 		return -EINVAL;
451 
452 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
453 	map = find_and_alloc_map(attr);
454 	if (IS_ERR(map))
455 		return PTR_ERR(map);
456 
457 	err = bpf_obj_name_cpy(map->name, attr->map_name);
458 	if (err)
459 		goto free_map_nouncharge;
460 
461 	atomic_set(&map->refcnt, 1);
462 	atomic_set(&map->usercnt, 1);
463 
464 	if (bpf_map_support_seq_show(map) &&
465 	    (attr->btf_key_type_id || attr->btf_value_type_id)) {
466 		struct btf *btf;
467 
468 		if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
469 			err = -EINVAL;
470 			goto free_map_nouncharge;
471 		}
472 
473 		btf = btf_get_by_fd(attr->btf_fd);
474 		if (IS_ERR(btf)) {
475 			err = PTR_ERR(btf);
476 			goto free_map_nouncharge;
477 		}
478 
479 		err = map->ops->map_check_btf(map, btf, attr->btf_key_type_id,
480 					      attr->btf_value_type_id);
481 		if (err) {
482 			btf_put(btf);
483 			goto free_map_nouncharge;
484 		}
485 
486 		map->btf = btf;
487 		map->btf_key_type_id = attr->btf_key_type_id;
488 		map->btf_value_type_id = attr->btf_value_type_id;
489 	}
490 
491 	err = security_bpf_map_alloc(map);
492 	if (err)
493 		goto free_map_nouncharge;
494 
495 	err = bpf_map_charge_memlock(map);
496 	if (err)
497 		goto free_map_sec;
498 
499 	err = bpf_map_alloc_id(map);
500 	if (err)
501 		goto free_map;
502 
503 	err = bpf_map_new_fd(map, f_flags);
504 	if (err < 0) {
505 		/* failed to allocate fd.
506 		 * bpf_map_put() is needed because the above
507 		 * bpf_map_alloc_id() has published the map
508 		 * to the userspace and the userspace may
509 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
510 		 */
511 		bpf_map_put(map);
512 		return err;
513 	}
514 
515 	return err;
516 
517 free_map:
518 	bpf_map_uncharge_memlock(map);
519 free_map_sec:
520 	security_bpf_map_free(map);
521 free_map_nouncharge:
522 	btf_put(map->btf);
523 	map->ops->map_free(map);
524 	return err;
525 }
526 
527 /* if error is returned, fd is released.
528  * On success caller should complete fd access with matching fdput()
529  */
530 struct bpf_map *__bpf_map_get(struct fd f)
531 {
532 	if (!f.file)
533 		return ERR_PTR(-EBADF);
534 	if (f.file->f_op != &bpf_map_fops) {
535 		fdput(f);
536 		return ERR_PTR(-EINVAL);
537 	}
538 
539 	return f.file->private_data;
540 }
541 
542 /* prog's and map's refcnt limit */
543 #define BPF_MAX_REFCNT 32768
544 
545 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
546 {
547 	if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
548 		atomic_dec(&map->refcnt);
549 		return ERR_PTR(-EBUSY);
550 	}
551 	if (uref)
552 		atomic_inc(&map->usercnt);
553 	return map;
554 }
555 EXPORT_SYMBOL_GPL(bpf_map_inc);
556 
557 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
558 {
559 	struct fd f = fdget(ufd);
560 	struct bpf_map *map;
561 
562 	map = __bpf_map_get(f);
563 	if (IS_ERR(map))
564 		return map;
565 
566 	map = bpf_map_inc(map, true);
567 	fdput(f);
568 
569 	return map;
570 }
571 
572 /* map_idr_lock should have been held */
573 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
574 					    bool uref)
575 {
576 	int refold;
577 
578 	refold = __atomic_add_unless(&map->refcnt, 1, 0);
579 
580 	if (refold >= BPF_MAX_REFCNT) {
581 		__bpf_map_put(map, false);
582 		return ERR_PTR(-EBUSY);
583 	}
584 
585 	if (!refold)
586 		return ERR_PTR(-ENOENT);
587 
588 	if (uref)
589 		atomic_inc(&map->usercnt);
590 
591 	return map;
592 }
593 
594 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
595 {
596 	return -ENOTSUPP;
597 }
598 
599 /* last field in 'union bpf_attr' used by this command */
600 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
601 
602 static int map_lookup_elem(union bpf_attr *attr)
603 {
604 	void __user *ukey = u64_to_user_ptr(attr->key);
605 	void __user *uvalue = u64_to_user_ptr(attr->value);
606 	int ufd = attr->map_fd;
607 	struct bpf_map *map;
608 	void *key, *value, *ptr;
609 	u32 value_size;
610 	struct fd f;
611 	int err;
612 
613 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
614 		return -EINVAL;
615 
616 	f = fdget(ufd);
617 	map = __bpf_map_get(f);
618 	if (IS_ERR(map))
619 		return PTR_ERR(map);
620 
621 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
622 		err = -EPERM;
623 		goto err_put;
624 	}
625 
626 	key = memdup_user(ukey, map->key_size);
627 	if (IS_ERR(key)) {
628 		err = PTR_ERR(key);
629 		goto err_put;
630 	}
631 
632 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
633 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
634 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
635 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
636 	else if (IS_FD_MAP(map))
637 		value_size = sizeof(u32);
638 	else
639 		value_size = map->value_size;
640 
641 	err = -ENOMEM;
642 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
643 	if (!value)
644 		goto free_key;
645 
646 	if (bpf_map_is_dev_bound(map)) {
647 		err = bpf_map_offload_lookup_elem(map, key, value);
648 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
649 		   map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
650 		err = bpf_percpu_hash_copy(map, key, value);
651 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
652 		err = bpf_percpu_array_copy(map, key, value);
653 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
654 		err = bpf_stackmap_copy(map, key, value);
655 	} else if (IS_FD_ARRAY(map)) {
656 		err = bpf_fd_array_map_lookup_elem(map, key, value);
657 	} else if (IS_FD_HASH(map)) {
658 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
659 	} else {
660 		rcu_read_lock();
661 		ptr = map->ops->map_lookup_elem(map, key);
662 		if (ptr)
663 			memcpy(value, ptr, value_size);
664 		rcu_read_unlock();
665 		err = ptr ? 0 : -ENOENT;
666 	}
667 
668 	if (err)
669 		goto free_value;
670 
671 	err = -EFAULT;
672 	if (copy_to_user(uvalue, value, value_size) != 0)
673 		goto free_value;
674 
675 	err = 0;
676 
677 free_value:
678 	kfree(value);
679 free_key:
680 	kfree(key);
681 err_put:
682 	fdput(f);
683 	return err;
684 }
685 
686 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
687 
688 static int map_update_elem(union bpf_attr *attr)
689 {
690 	void __user *ukey = u64_to_user_ptr(attr->key);
691 	void __user *uvalue = u64_to_user_ptr(attr->value);
692 	int ufd = attr->map_fd;
693 	struct bpf_map *map;
694 	void *key, *value;
695 	u32 value_size;
696 	struct fd f;
697 	int err;
698 
699 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
700 		return -EINVAL;
701 
702 	f = fdget(ufd);
703 	map = __bpf_map_get(f);
704 	if (IS_ERR(map))
705 		return PTR_ERR(map);
706 
707 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
708 		err = -EPERM;
709 		goto err_put;
710 	}
711 
712 	key = memdup_user(ukey, map->key_size);
713 	if (IS_ERR(key)) {
714 		err = PTR_ERR(key);
715 		goto err_put;
716 	}
717 
718 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
719 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
720 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
721 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
722 	else
723 		value_size = map->value_size;
724 
725 	err = -ENOMEM;
726 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
727 	if (!value)
728 		goto free_key;
729 
730 	err = -EFAULT;
731 	if (copy_from_user(value, uvalue, value_size) != 0)
732 		goto free_value;
733 
734 	/* Need to create a kthread, thus must support schedule */
735 	if (bpf_map_is_dev_bound(map)) {
736 		err = bpf_map_offload_update_elem(map, key, value, attr->flags);
737 		goto out;
738 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
739 		   map->map_type == BPF_MAP_TYPE_SOCKHASH ||
740 		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
741 		err = map->ops->map_update_elem(map, key, value, attr->flags);
742 		goto out;
743 	}
744 
745 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
746 	 * inside bpf map update or delete otherwise deadlocks are possible
747 	 */
748 	preempt_disable();
749 	__this_cpu_inc(bpf_prog_active);
750 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
751 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
752 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
753 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
754 		err = bpf_percpu_array_update(map, key, value, attr->flags);
755 	} else if (IS_FD_ARRAY(map)) {
756 		rcu_read_lock();
757 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
758 						   attr->flags);
759 		rcu_read_unlock();
760 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
761 		rcu_read_lock();
762 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
763 						  attr->flags);
764 		rcu_read_unlock();
765 	} else {
766 		rcu_read_lock();
767 		err = map->ops->map_update_elem(map, key, value, attr->flags);
768 		rcu_read_unlock();
769 	}
770 	__this_cpu_dec(bpf_prog_active);
771 	preempt_enable();
772 out:
773 free_value:
774 	kfree(value);
775 free_key:
776 	kfree(key);
777 err_put:
778 	fdput(f);
779 	return err;
780 }
781 
782 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
783 
784 static int map_delete_elem(union bpf_attr *attr)
785 {
786 	void __user *ukey = u64_to_user_ptr(attr->key);
787 	int ufd = attr->map_fd;
788 	struct bpf_map *map;
789 	struct fd f;
790 	void *key;
791 	int err;
792 
793 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
794 		return -EINVAL;
795 
796 	f = fdget(ufd);
797 	map = __bpf_map_get(f);
798 	if (IS_ERR(map))
799 		return PTR_ERR(map);
800 
801 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
802 		err = -EPERM;
803 		goto err_put;
804 	}
805 
806 	key = memdup_user(ukey, map->key_size);
807 	if (IS_ERR(key)) {
808 		err = PTR_ERR(key);
809 		goto err_put;
810 	}
811 
812 	if (bpf_map_is_dev_bound(map)) {
813 		err = bpf_map_offload_delete_elem(map, key);
814 		goto out;
815 	}
816 
817 	preempt_disable();
818 	__this_cpu_inc(bpf_prog_active);
819 	rcu_read_lock();
820 	err = map->ops->map_delete_elem(map, key);
821 	rcu_read_unlock();
822 	__this_cpu_dec(bpf_prog_active);
823 	preempt_enable();
824 out:
825 	kfree(key);
826 err_put:
827 	fdput(f);
828 	return err;
829 }
830 
831 /* last field in 'union bpf_attr' used by this command */
832 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
833 
834 static int map_get_next_key(union bpf_attr *attr)
835 {
836 	void __user *ukey = u64_to_user_ptr(attr->key);
837 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
838 	int ufd = attr->map_fd;
839 	struct bpf_map *map;
840 	void *key, *next_key;
841 	struct fd f;
842 	int err;
843 
844 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
845 		return -EINVAL;
846 
847 	f = fdget(ufd);
848 	map = __bpf_map_get(f);
849 	if (IS_ERR(map))
850 		return PTR_ERR(map);
851 
852 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
853 		err = -EPERM;
854 		goto err_put;
855 	}
856 
857 	if (ukey) {
858 		key = memdup_user(ukey, map->key_size);
859 		if (IS_ERR(key)) {
860 			err = PTR_ERR(key);
861 			goto err_put;
862 		}
863 	} else {
864 		key = NULL;
865 	}
866 
867 	err = -ENOMEM;
868 	next_key = kmalloc(map->key_size, GFP_USER);
869 	if (!next_key)
870 		goto free_key;
871 
872 	if (bpf_map_is_dev_bound(map)) {
873 		err = bpf_map_offload_get_next_key(map, key, next_key);
874 		goto out;
875 	}
876 
877 	rcu_read_lock();
878 	err = map->ops->map_get_next_key(map, key, next_key);
879 	rcu_read_unlock();
880 out:
881 	if (err)
882 		goto free_next_key;
883 
884 	err = -EFAULT;
885 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
886 		goto free_next_key;
887 
888 	err = 0;
889 
890 free_next_key:
891 	kfree(next_key);
892 free_key:
893 	kfree(key);
894 err_put:
895 	fdput(f);
896 	return err;
897 }
898 
899 static const struct bpf_prog_ops * const bpf_prog_types[] = {
900 #define BPF_PROG_TYPE(_id, _name) \
901 	[_id] = & _name ## _prog_ops,
902 #define BPF_MAP_TYPE(_id, _ops)
903 #include <linux/bpf_types.h>
904 #undef BPF_PROG_TYPE
905 #undef BPF_MAP_TYPE
906 };
907 
908 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
909 {
910 	const struct bpf_prog_ops *ops;
911 
912 	if (type >= ARRAY_SIZE(bpf_prog_types))
913 		return -EINVAL;
914 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
915 	ops = bpf_prog_types[type];
916 	if (!ops)
917 		return -EINVAL;
918 
919 	if (!bpf_prog_is_dev_bound(prog->aux))
920 		prog->aux->ops = ops;
921 	else
922 		prog->aux->ops = &bpf_offload_prog_ops;
923 	prog->type = type;
924 	return 0;
925 }
926 
927 /* drop refcnt on maps used by eBPF program and free auxilary data */
928 static void free_used_maps(struct bpf_prog_aux *aux)
929 {
930 	int i;
931 
932 	for (i = 0; i < aux->used_map_cnt; i++)
933 		bpf_map_put(aux->used_maps[i]);
934 
935 	kfree(aux->used_maps);
936 }
937 
938 int __bpf_prog_charge(struct user_struct *user, u32 pages)
939 {
940 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
941 	unsigned long user_bufs;
942 
943 	if (user) {
944 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
945 		if (user_bufs > memlock_limit) {
946 			atomic_long_sub(pages, &user->locked_vm);
947 			return -EPERM;
948 		}
949 	}
950 
951 	return 0;
952 }
953 
954 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
955 {
956 	if (user)
957 		atomic_long_sub(pages, &user->locked_vm);
958 }
959 
960 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
961 {
962 	struct user_struct *user = get_current_user();
963 	int ret;
964 
965 	ret = __bpf_prog_charge(user, prog->pages);
966 	if (ret) {
967 		free_uid(user);
968 		return ret;
969 	}
970 
971 	prog->aux->user = user;
972 	return 0;
973 }
974 
975 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
976 {
977 	struct user_struct *user = prog->aux->user;
978 
979 	__bpf_prog_uncharge(user, prog->pages);
980 	free_uid(user);
981 }
982 
983 static int bpf_prog_alloc_id(struct bpf_prog *prog)
984 {
985 	int id;
986 
987 	idr_preload(GFP_KERNEL);
988 	spin_lock_bh(&prog_idr_lock);
989 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
990 	if (id > 0)
991 		prog->aux->id = id;
992 	spin_unlock_bh(&prog_idr_lock);
993 	idr_preload_end();
994 
995 	/* id is in [1, INT_MAX) */
996 	if (WARN_ON_ONCE(!id))
997 		return -ENOSPC;
998 
999 	return id > 0 ? 0 : id;
1000 }
1001 
1002 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1003 {
1004 	/* cBPF to eBPF migrations are currently not in the idr store.
1005 	 * Offloaded programs are removed from the store when their device
1006 	 * disappears - even if someone grabs an fd to them they are unusable,
1007 	 * simply waiting for refcnt to drop to be freed.
1008 	 */
1009 	if (!prog->aux->id)
1010 		return;
1011 
1012 	if (do_idr_lock)
1013 		spin_lock_bh(&prog_idr_lock);
1014 	else
1015 		__acquire(&prog_idr_lock);
1016 
1017 	idr_remove(&prog_idr, prog->aux->id);
1018 	prog->aux->id = 0;
1019 
1020 	if (do_idr_lock)
1021 		spin_unlock_bh(&prog_idr_lock);
1022 	else
1023 		__release(&prog_idr_lock);
1024 }
1025 
1026 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1027 {
1028 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1029 
1030 	free_used_maps(aux);
1031 	bpf_prog_uncharge_memlock(aux->prog);
1032 	security_bpf_prog_free(aux);
1033 	bpf_prog_free(aux->prog);
1034 }
1035 
1036 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1037 {
1038 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
1039 		/* bpf_prog_free_id() must be called first */
1040 		bpf_prog_free_id(prog, do_idr_lock);
1041 		bpf_prog_kallsyms_del_all(prog);
1042 
1043 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1044 	}
1045 }
1046 
1047 void bpf_prog_put(struct bpf_prog *prog)
1048 {
1049 	__bpf_prog_put(prog, true);
1050 }
1051 EXPORT_SYMBOL_GPL(bpf_prog_put);
1052 
1053 static int bpf_prog_release(struct inode *inode, struct file *filp)
1054 {
1055 	struct bpf_prog *prog = filp->private_data;
1056 
1057 	bpf_prog_put(prog);
1058 	return 0;
1059 }
1060 
1061 #ifdef CONFIG_PROC_FS
1062 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1063 {
1064 	const struct bpf_prog *prog = filp->private_data;
1065 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1066 
1067 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1068 	seq_printf(m,
1069 		   "prog_type:\t%u\n"
1070 		   "prog_jited:\t%u\n"
1071 		   "prog_tag:\t%s\n"
1072 		   "memlock:\t%llu\n"
1073 		   "prog_id:\t%u\n",
1074 		   prog->type,
1075 		   prog->jited,
1076 		   prog_tag,
1077 		   prog->pages * 1ULL << PAGE_SHIFT,
1078 		   prog->aux->id);
1079 }
1080 #endif
1081 
1082 const struct file_operations bpf_prog_fops = {
1083 #ifdef CONFIG_PROC_FS
1084 	.show_fdinfo	= bpf_prog_show_fdinfo,
1085 #endif
1086 	.release	= bpf_prog_release,
1087 	.read		= bpf_dummy_read,
1088 	.write		= bpf_dummy_write,
1089 };
1090 
1091 int bpf_prog_new_fd(struct bpf_prog *prog)
1092 {
1093 	int ret;
1094 
1095 	ret = security_bpf_prog(prog);
1096 	if (ret < 0)
1097 		return ret;
1098 
1099 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1100 				O_RDWR | O_CLOEXEC);
1101 }
1102 
1103 static struct bpf_prog *____bpf_prog_get(struct fd f)
1104 {
1105 	if (!f.file)
1106 		return ERR_PTR(-EBADF);
1107 	if (f.file->f_op != &bpf_prog_fops) {
1108 		fdput(f);
1109 		return ERR_PTR(-EINVAL);
1110 	}
1111 
1112 	return f.file->private_data;
1113 }
1114 
1115 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1116 {
1117 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1118 		atomic_sub(i, &prog->aux->refcnt);
1119 		return ERR_PTR(-EBUSY);
1120 	}
1121 	return prog;
1122 }
1123 EXPORT_SYMBOL_GPL(bpf_prog_add);
1124 
1125 void bpf_prog_sub(struct bpf_prog *prog, int i)
1126 {
1127 	/* Only to be used for undoing previous bpf_prog_add() in some
1128 	 * error path. We still know that another entity in our call
1129 	 * path holds a reference to the program, thus atomic_sub() can
1130 	 * be safely used in such cases!
1131 	 */
1132 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1133 }
1134 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1135 
1136 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1137 {
1138 	return bpf_prog_add(prog, 1);
1139 }
1140 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1141 
1142 /* prog_idr_lock should have been held */
1143 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1144 {
1145 	int refold;
1146 
1147 	refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
1148 
1149 	if (refold >= BPF_MAX_REFCNT) {
1150 		__bpf_prog_put(prog, false);
1151 		return ERR_PTR(-EBUSY);
1152 	}
1153 
1154 	if (!refold)
1155 		return ERR_PTR(-ENOENT);
1156 
1157 	return prog;
1158 }
1159 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1160 
1161 bool bpf_prog_get_ok(struct bpf_prog *prog,
1162 			    enum bpf_prog_type *attach_type, bool attach_drv)
1163 {
1164 	/* not an attachment, just a refcount inc, always allow */
1165 	if (!attach_type)
1166 		return true;
1167 
1168 	if (prog->type != *attach_type)
1169 		return false;
1170 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1171 		return false;
1172 
1173 	return true;
1174 }
1175 
1176 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1177 				       bool attach_drv)
1178 {
1179 	struct fd f = fdget(ufd);
1180 	struct bpf_prog *prog;
1181 
1182 	prog = ____bpf_prog_get(f);
1183 	if (IS_ERR(prog))
1184 		return prog;
1185 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1186 		prog = ERR_PTR(-EINVAL);
1187 		goto out;
1188 	}
1189 
1190 	prog = bpf_prog_inc(prog);
1191 out:
1192 	fdput(f);
1193 	return prog;
1194 }
1195 
1196 struct bpf_prog *bpf_prog_get(u32 ufd)
1197 {
1198 	return __bpf_prog_get(ufd, NULL, false);
1199 }
1200 
1201 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1202 				       bool attach_drv)
1203 {
1204 	return __bpf_prog_get(ufd, &type, attach_drv);
1205 }
1206 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1207 
1208 /* Initially all BPF programs could be loaded w/o specifying
1209  * expected_attach_type. Later for some of them specifying expected_attach_type
1210  * at load time became required so that program could be validated properly.
1211  * Programs of types that are allowed to be loaded both w/ and w/o (for
1212  * backward compatibility) expected_attach_type, should have the default attach
1213  * type assigned to expected_attach_type for the latter case, so that it can be
1214  * validated later at attach time.
1215  *
1216  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1217  * prog type requires it but has some attach types that have to be backward
1218  * compatible.
1219  */
1220 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1221 {
1222 	switch (attr->prog_type) {
1223 	case BPF_PROG_TYPE_CGROUP_SOCK:
1224 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1225 		 * exist so checking for non-zero is the way to go here.
1226 		 */
1227 		if (!attr->expected_attach_type)
1228 			attr->expected_attach_type =
1229 				BPF_CGROUP_INET_SOCK_CREATE;
1230 		break;
1231 	}
1232 }
1233 
1234 static int
1235 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1236 				enum bpf_attach_type expected_attach_type)
1237 {
1238 	switch (prog_type) {
1239 	case BPF_PROG_TYPE_CGROUP_SOCK:
1240 		switch (expected_attach_type) {
1241 		case BPF_CGROUP_INET_SOCK_CREATE:
1242 		case BPF_CGROUP_INET4_POST_BIND:
1243 		case BPF_CGROUP_INET6_POST_BIND:
1244 			return 0;
1245 		default:
1246 			return -EINVAL;
1247 		}
1248 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1249 		switch (expected_attach_type) {
1250 		case BPF_CGROUP_INET4_BIND:
1251 		case BPF_CGROUP_INET6_BIND:
1252 		case BPF_CGROUP_INET4_CONNECT:
1253 		case BPF_CGROUP_INET6_CONNECT:
1254 		case BPF_CGROUP_UDP4_SENDMSG:
1255 		case BPF_CGROUP_UDP6_SENDMSG:
1256 			return 0;
1257 		default:
1258 			return -EINVAL;
1259 		}
1260 	default:
1261 		return 0;
1262 	}
1263 }
1264 
1265 /* last field in 'union bpf_attr' used by this command */
1266 #define	BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1267 
1268 static int bpf_prog_load(union bpf_attr *attr)
1269 {
1270 	enum bpf_prog_type type = attr->prog_type;
1271 	struct bpf_prog *prog;
1272 	int err;
1273 	char license[128];
1274 	bool is_gpl;
1275 
1276 	if (CHECK_ATTR(BPF_PROG_LOAD))
1277 		return -EINVAL;
1278 
1279 	if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1280 		return -EINVAL;
1281 
1282 	/* copy eBPF program license from user space */
1283 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1284 			      sizeof(license) - 1) < 0)
1285 		return -EFAULT;
1286 	license[sizeof(license) - 1] = 0;
1287 
1288 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1289 	is_gpl = license_is_gpl_compatible(license);
1290 
1291 	if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1292 		return -E2BIG;
1293 
1294 	if (type == BPF_PROG_TYPE_KPROBE &&
1295 	    attr->kern_version != LINUX_VERSION_CODE)
1296 		return -EINVAL;
1297 
1298 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1299 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
1300 	    !capable(CAP_SYS_ADMIN))
1301 		return -EPERM;
1302 
1303 	bpf_prog_load_fixup_attach_type(attr);
1304 	if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1305 		return -EINVAL;
1306 
1307 	/* plain bpf_prog allocation */
1308 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1309 	if (!prog)
1310 		return -ENOMEM;
1311 
1312 	prog->expected_attach_type = attr->expected_attach_type;
1313 
1314 	prog->aux->offload_requested = !!attr->prog_ifindex;
1315 
1316 	err = security_bpf_prog_alloc(prog->aux);
1317 	if (err)
1318 		goto free_prog_nouncharge;
1319 
1320 	err = bpf_prog_charge_memlock(prog);
1321 	if (err)
1322 		goto free_prog_sec;
1323 
1324 	prog->len = attr->insn_cnt;
1325 
1326 	err = -EFAULT;
1327 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1328 			   bpf_prog_insn_size(prog)) != 0)
1329 		goto free_prog;
1330 
1331 	prog->orig_prog = NULL;
1332 	prog->jited = 0;
1333 
1334 	atomic_set(&prog->aux->refcnt, 1);
1335 	prog->gpl_compatible = is_gpl ? 1 : 0;
1336 
1337 	if (bpf_prog_is_dev_bound(prog->aux)) {
1338 		err = bpf_prog_offload_init(prog, attr);
1339 		if (err)
1340 			goto free_prog;
1341 	}
1342 
1343 	/* find program type: socket_filter vs tracing_filter */
1344 	err = find_prog_type(type, prog);
1345 	if (err < 0)
1346 		goto free_prog;
1347 
1348 	prog->aux->load_time = ktime_get_boot_ns();
1349 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1350 	if (err)
1351 		goto free_prog;
1352 
1353 	/* run eBPF verifier */
1354 	err = bpf_check(&prog, attr);
1355 	if (err < 0)
1356 		goto free_used_maps;
1357 
1358 	prog = bpf_prog_select_runtime(prog, &err);
1359 	if (err < 0)
1360 		goto free_used_maps;
1361 
1362 	err = bpf_prog_alloc_id(prog);
1363 	if (err)
1364 		goto free_used_maps;
1365 
1366 	err = bpf_prog_new_fd(prog);
1367 	if (err < 0) {
1368 		/* failed to allocate fd.
1369 		 * bpf_prog_put() is needed because the above
1370 		 * bpf_prog_alloc_id() has published the prog
1371 		 * to the userspace and the userspace may
1372 		 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1373 		 */
1374 		bpf_prog_put(prog);
1375 		return err;
1376 	}
1377 
1378 	bpf_prog_kallsyms_add(prog);
1379 	return err;
1380 
1381 free_used_maps:
1382 	bpf_prog_kallsyms_del_subprogs(prog);
1383 	free_used_maps(prog->aux);
1384 free_prog:
1385 	bpf_prog_uncharge_memlock(prog);
1386 free_prog_sec:
1387 	security_bpf_prog_free(prog->aux);
1388 free_prog_nouncharge:
1389 	bpf_prog_free(prog);
1390 	return err;
1391 }
1392 
1393 #define BPF_OBJ_LAST_FIELD file_flags
1394 
1395 static int bpf_obj_pin(const union bpf_attr *attr)
1396 {
1397 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1398 		return -EINVAL;
1399 
1400 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1401 }
1402 
1403 static int bpf_obj_get(const union bpf_attr *attr)
1404 {
1405 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1406 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1407 		return -EINVAL;
1408 
1409 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1410 				attr->file_flags);
1411 }
1412 
1413 struct bpf_raw_tracepoint {
1414 	struct bpf_raw_event_map *btp;
1415 	struct bpf_prog *prog;
1416 };
1417 
1418 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1419 {
1420 	struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1421 
1422 	if (raw_tp->prog) {
1423 		bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1424 		bpf_prog_put(raw_tp->prog);
1425 	}
1426 	kfree(raw_tp);
1427 	return 0;
1428 }
1429 
1430 static const struct file_operations bpf_raw_tp_fops = {
1431 	.release	= bpf_raw_tracepoint_release,
1432 	.read		= bpf_dummy_read,
1433 	.write		= bpf_dummy_write,
1434 };
1435 
1436 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1437 
1438 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1439 {
1440 	struct bpf_raw_tracepoint *raw_tp;
1441 	struct bpf_raw_event_map *btp;
1442 	struct bpf_prog *prog;
1443 	char tp_name[128];
1444 	int tp_fd, err;
1445 
1446 	if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1447 			      sizeof(tp_name) - 1) < 0)
1448 		return -EFAULT;
1449 	tp_name[sizeof(tp_name) - 1] = 0;
1450 
1451 	btp = bpf_find_raw_tracepoint(tp_name);
1452 	if (!btp)
1453 		return -ENOENT;
1454 
1455 	raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1456 	if (!raw_tp)
1457 		return -ENOMEM;
1458 	raw_tp->btp = btp;
1459 
1460 	prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1461 				 BPF_PROG_TYPE_RAW_TRACEPOINT);
1462 	if (IS_ERR(prog)) {
1463 		err = PTR_ERR(prog);
1464 		goto out_free_tp;
1465 	}
1466 
1467 	err = bpf_probe_register(raw_tp->btp, prog);
1468 	if (err)
1469 		goto out_put_prog;
1470 
1471 	raw_tp->prog = prog;
1472 	tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1473 				 O_CLOEXEC);
1474 	if (tp_fd < 0) {
1475 		bpf_probe_unregister(raw_tp->btp, prog);
1476 		err = tp_fd;
1477 		goto out_put_prog;
1478 	}
1479 	return tp_fd;
1480 
1481 out_put_prog:
1482 	bpf_prog_put(prog);
1483 out_free_tp:
1484 	kfree(raw_tp);
1485 	return err;
1486 }
1487 
1488 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1489 					     enum bpf_attach_type attach_type)
1490 {
1491 	switch (prog->type) {
1492 	case BPF_PROG_TYPE_CGROUP_SOCK:
1493 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1494 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1495 	default:
1496 		return 0;
1497 	}
1498 }
1499 
1500 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1501 
1502 #define BPF_F_ATTACH_MASK \
1503 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1504 
1505 static int bpf_prog_attach(const union bpf_attr *attr)
1506 {
1507 	enum bpf_prog_type ptype;
1508 	struct bpf_prog *prog;
1509 	int ret;
1510 
1511 	if (!capable(CAP_NET_ADMIN))
1512 		return -EPERM;
1513 
1514 	if (CHECK_ATTR(BPF_PROG_ATTACH))
1515 		return -EINVAL;
1516 
1517 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1518 		return -EINVAL;
1519 
1520 	switch (attr->attach_type) {
1521 	case BPF_CGROUP_INET_INGRESS:
1522 	case BPF_CGROUP_INET_EGRESS:
1523 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1524 		break;
1525 	case BPF_CGROUP_INET_SOCK_CREATE:
1526 	case BPF_CGROUP_INET4_POST_BIND:
1527 	case BPF_CGROUP_INET6_POST_BIND:
1528 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1529 		break;
1530 	case BPF_CGROUP_INET4_BIND:
1531 	case BPF_CGROUP_INET6_BIND:
1532 	case BPF_CGROUP_INET4_CONNECT:
1533 	case BPF_CGROUP_INET6_CONNECT:
1534 	case BPF_CGROUP_UDP4_SENDMSG:
1535 	case BPF_CGROUP_UDP6_SENDMSG:
1536 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1537 		break;
1538 	case BPF_CGROUP_SOCK_OPS:
1539 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1540 		break;
1541 	case BPF_CGROUP_DEVICE:
1542 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1543 		break;
1544 	case BPF_SK_MSG_VERDICT:
1545 		ptype = BPF_PROG_TYPE_SK_MSG;
1546 		break;
1547 	case BPF_SK_SKB_STREAM_PARSER:
1548 	case BPF_SK_SKB_STREAM_VERDICT:
1549 		ptype = BPF_PROG_TYPE_SK_SKB;
1550 		break;
1551 	case BPF_LIRC_MODE2:
1552 		ptype = BPF_PROG_TYPE_LIRC_MODE2;
1553 		break;
1554 	default:
1555 		return -EINVAL;
1556 	}
1557 
1558 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1559 	if (IS_ERR(prog))
1560 		return PTR_ERR(prog);
1561 
1562 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1563 		bpf_prog_put(prog);
1564 		return -EINVAL;
1565 	}
1566 
1567 	switch (ptype) {
1568 	case BPF_PROG_TYPE_SK_SKB:
1569 	case BPF_PROG_TYPE_SK_MSG:
1570 		ret = sockmap_get_from_fd(attr, ptype, prog);
1571 		break;
1572 	case BPF_PROG_TYPE_LIRC_MODE2:
1573 		ret = lirc_prog_attach(attr, prog);
1574 		break;
1575 	default:
1576 		ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1577 	}
1578 
1579 	if (ret)
1580 		bpf_prog_put(prog);
1581 	return ret;
1582 }
1583 
1584 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1585 
1586 static int bpf_prog_detach(const union bpf_attr *attr)
1587 {
1588 	enum bpf_prog_type ptype;
1589 
1590 	if (!capable(CAP_NET_ADMIN))
1591 		return -EPERM;
1592 
1593 	if (CHECK_ATTR(BPF_PROG_DETACH))
1594 		return -EINVAL;
1595 
1596 	switch (attr->attach_type) {
1597 	case BPF_CGROUP_INET_INGRESS:
1598 	case BPF_CGROUP_INET_EGRESS:
1599 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1600 		break;
1601 	case BPF_CGROUP_INET_SOCK_CREATE:
1602 	case BPF_CGROUP_INET4_POST_BIND:
1603 	case BPF_CGROUP_INET6_POST_BIND:
1604 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1605 		break;
1606 	case BPF_CGROUP_INET4_BIND:
1607 	case BPF_CGROUP_INET6_BIND:
1608 	case BPF_CGROUP_INET4_CONNECT:
1609 	case BPF_CGROUP_INET6_CONNECT:
1610 	case BPF_CGROUP_UDP4_SENDMSG:
1611 	case BPF_CGROUP_UDP6_SENDMSG:
1612 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1613 		break;
1614 	case BPF_CGROUP_SOCK_OPS:
1615 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1616 		break;
1617 	case BPF_CGROUP_DEVICE:
1618 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1619 		break;
1620 	case BPF_SK_MSG_VERDICT:
1621 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, NULL);
1622 	case BPF_SK_SKB_STREAM_PARSER:
1623 	case BPF_SK_SKB_STREAM_VERDICT:
1624 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, NULL);
1625 	case BPF_LIRC_MODE2:
1626 		return lirc_prog_detach(attr);
1627 	default:
1628 		return -EINVAL;
1629 	}
1630 
1631 	return cgroup_bpf_prog_detach(attr, ptype);
1632 }
1633 
1634 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1635 
1636 static int bpf_prog_query(const union bpf_attr *attr,
1637 			  union bpf_attr __user *uattr)
1638 {
1639 	if (!capable(CAP_NET_ADMIN))
1640 		return -EPERM;
1641 	if (CHECK_ATTR(BPF_PROG_QUERY))
1642 		return -EINVAL;
1643 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1644 		return -EINVAL;
1645 
1646 	switch (attr->query.attach_type) {
1647 	case BPF_CGROUP_INET_INGRESS:
1648 	case BPF_CGROUP_INET_EGRESS:
1649 	case BPF_CGROUP_INET_SOCK_CREATE:
1650 	case BPF_CGROUP_INET4_BIND:
1651 	case BPF_CGROUP_INET6_BIND:
1652 	case BPF_CGROUP_INET4_POST_BIND:
1653 	case BPF_CGROUP_INET6_POST_BIND:
1654 	case BPF_CGROUP_INET4_CONNECT:
1655 	case BPF_CGROUP_INET6_CONNECT:
1656 	case BPF_CGROUP_UDP4_SENDMSG:
1657 	case BPF_CGROUP_UDP6_SENDMSG:
1658 	case BPF_CGROUP_SOCK_OPS:
1659 	case BPF_CGROUP_DEVICE:
1660 		break;
1661 	case BPF_LIRC_MODE2:
1662 		return lirc_prog_query(attr, uattr);
1663 	default:
1664 		return -EINVAL;
1665 	}
1666 
1667 	return cgroup_bpf_prog_query(attr, uattr);
1668 }
1669 
1670 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1671 
1672 static int bpf_prog_test_run(const union bpf_attr *attr,
1673 			     union bpf_attr __user *uattr)
1674 {
1675 	struct bpf_prog *prog;
1676 	int ret = -ENOTSUPP;
1677 
1678 	if (!capable(CAP_SYS_ADMIN))
1679 		return -EPERM;
1680 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1681 		return -EINVAL;
1682 
1683 	prog = bpf_prog_get(attr->test.prog_fd);
1684 	if (IS_ERR(prog))
1685 		return PTR_ERR(prog);
1686 
1687 	if (prog->aux->ops->test_run)
1688 		ret = prog->aux->ops->test_run(prog, attr, uattr);
1689 
1690 	bpf_prog_put(prog);
1691 	return ret;
1692 }
1693 
1694 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1695 
1696 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1697 			       union bpf_attr __user *uattr,
1698 			       struct idr *idr,
1699 			       spinlock_t *lock)
1700 {
1701 	u32 next_id = attr->start_id;
1702 	int err = 0;
1703 
1704 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1705 		return -EINVAL;
1706 
1707 	if (!capable(CAP_SYS_ADMIN))
1708 		return -EPERM;
1709 
1710 	next_id++;
1711 	spin_lock_bh(lock);
1712 	if (!idr_get_next(idr, &next_id))
1713 		err = -ENOENT;
1714 	spin_unlock_bh(lock);
1715 
1716 	if (!err)
1717 		err = put_user(next_id, &uattr->next_id);
1718 
1719 	return err;
1720 }
1721 
1722 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1723 
1724 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1725 {
1726 	struct bpf_prog *prog;
1727 	u32 id = attr->prog_id;
1728 	int fd;
1729 
1730 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1731 		return -EINVAL;
1732 
1733 	if (!capable(CAP_SYS_ADMIN))
1734 		return -EPERM;
1735 
1736 	spin_lock_bh(&prog_idr_lock);
1737 	prog = idr_find(&prog_idr, id);
1738 	if (prog)
1739 		prog = bpf_prog_inc_not_zero(prog);
1740 	else
1741 		prog = ERR_PTR(-ENOENT);
1742 	spin_unlock_bh(&prog_idr_lock);
1743 
1744 	if (IS_ERR(prog))
1745 		return PTR_ERR(prog);
1746 
1747 	fd = bpf_prog_new_fd(prog);
1748 	if (fd < 0)
1749 		bpf_prog_put(prog);
1750 
1751 	return fd;
1752 }
1753 
1754 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1755 
1756 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1757 {
1758 	struct bpf_map *map;
1759 	u32 id = attr->map_id;
1760 	int f_flags;
1761 	int fd;
1762 
1763 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1764 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1765 		return -EINVAL;
1766 
1767 	if (!capable(CAP_SYS_ADMIN))
1768 		return -EPERM;
1769 
1770 	f_flags = bpf_get_file_flag(attr->open_flags);
1771 	if (f_flags < 0)
1772 		return f_flags;
1773 
1774 	spin_lock_bh(&map_idr_lock);
1775 	map = idr_find(&map_idr, id);
1776 	if (map)
1777 		map = bpf_map_inc_not_zero(map, true);
1778 	else
1779 		map = ERR_PTR(-ENOENT);
1780 	spin_unlock_bh(&map_idr_lock);
1781 
1782 	if (IS_ERR(map))
1783 		return PTR_ERR(map);
1784 
1785 	fd = bpf_map_new_fd(map, f_flags);
1786 	if (fd < 0)
1787 		bpf_map_put(map);
1788 
1789 	return fd;
1790 }
1791 
1792 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1793 					      unsigned long addr)
1794 {
1795 	int i;
1796 
1797 	for (i = 0; i < prog->aux->used_map_cnt; i++)
1798 		if (prog->aux->used_maps[i] == (void *)addr)
1799 			return prog->aux->used_maps[i];
1800 	return NULL;
1801 }
1802 
1803 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1804 {
1805 	const struct bpf_map *map;
1806 	struct bpf_insn *insns;
1807 	u64 imm;
1808 	int i;
1809 
1810 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1811 			GFP_USER);
1812 	if (!insns)
1813 		return insns;
1814 
1815 	for (i = 0; i < prog->len; i++) {
1816 		if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1817 			insns[i].code = BPF_JMP | BPF_CALL;
1818 			insns[i].imm = BPF_FUNC_tail_call;
1819 			/* fall-through */
1820 		}
1821 		if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1822 		    insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1823 			if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1824 				insns[i].code = BPF_JMP | BPF_CALL;
1825 			if (!bpf_dump_raw_ok())
1826 				insns[i].imm = 0;
1827 			continue;
1828 		}
1829 
1830 		if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1831 			continue;
1832 
1833 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1834 		map = bpf_map_from_imm(prog, imm);
1835 		if (map) {
1836 			insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1837 			insns[i].imm = map->id;
1838 			insns[i + 1].imm = 0;
1839 			continue;
1840 		}
1841 
1842 		if (!bpf_dump_raw_ok() &&
1843 		    imm == (unsigned long)prog->aux) {
1844 			insns[i].imm = 0;
1845 			insns[i + 1].imm = 0;
1846 			continue;
1847 		}
1848 	}
1849 
1850 	return insns;
1851 }
1852 
1853 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1854 				   const union bpf_attr *attr,
1855 				   union bpf_attr __user *uattr)
1856 {
1857 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1858 	struct bpf_prog_info info = {};
1859 	u32 info_len = attr->info.info_len;
1860 	char __user *uinsns;
1861 	u32 ulen;
1862 	int err;
1863 
1864 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1865 	if (err)
1866 		return err;
1867 	info_len = min_t(u32, sizeof(info), info_len);
1868 
1869 	if (copy_from_user(&info, uinfo, info_len))
1870 		return -EFAULT;
1871 
1872 	info.type = prog->type;
1873 	info.id = prog->aux->id;
1874 	info.load_time = prog->aux->load_time;
1875 	info.created_by_uid = from_kuid_munged(current_user_ns(),
1876 					       prog->aux->user->uid);
1877 	info.gpl_compatible = prog->gpl_compatible;
1878 
1879 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
1880 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1881 
1882 	ulen = info.nr_map_ids;
1883 	info.nr_map_ids = prog->aux->used_map_cnt;
1884 	ulen = min_t(u32, info.nr_map_ids, ulen);
1885 	if (ulen) {
1886 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1887 		u32 i;
1888 
1889 		for (i = 0; i < ulen; i++)
1890 			if (put_user(prog->aux->used_maps[i]->id,
1891 				     &user_map_ids[i]))
1892 				return -EFAULT;
1893 	}
1894 
1895 	if (!capable(CAP_SYS_ADMIN)) {
1896 		info.jited_prog_len = 0;
1897 		info.xlated_prog_len = 0;
1898 		info.nr_jited_ksyms = 0;
1899 		goto done;
1900 	}
1901 
1902 	ulen = info.xlated_prog_len;
1903 	info.xlated_prog_len = bpf_prog_insn_size(prog);
1904 	if (info.xlated_prog_len && ulen) {
1905 		struct bpf_insn *insns_sanitized;
1906 		bool fault;
1907 
1908 		if (prog->blinded && !bpf_dump_raw_ok()) {
1909 			info.xlated_prog_insns = 0;
1910 			goto done;
1911 		}
1912 		insns_sanitized = bpf_insn_prepare_dump(prog);
1913 		if (!insns_sanitized)
1914 			return -ENOMEM;
1915 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1916 		ulen = min_t(u32, info.xlated_prog_len, ulen);
1917 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
1918 		kfree(insns_sanitized);
1919 		if (fault)
1920 			return -EFAULT;
1921 	}
1922 
1923 	if (bpf_prog_is_dev_bound(prog->aux)) {
1924 		err = bpf_prog_offload_info_fill(&info, prog);
1925 		if (err)
1926 			return err;
1927 		goto done;
1928 	}
1929 
1930 	/* NOTE: the following code is supposed to be skipped for offload.
1931 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
1932 	 * for offload.
1933 	 */
1934 	ulen = info.jited_prog_len;
1935 	if (prog->aux->func_cnt) {
1936 		u32 i;
1937 
1938 		info.jited_prog_len = 0;
1939 		for (i = 0; i < prog->aux->func_cnt; i++)
1940 			info.jited_prog_len += prog->aux->func[i]->jited_len;
1941 	} else {
1942 		info.jited_prog_len = prog->jited_len;
1943 	}
1944 
1945 	if (info.jited_prog_len && ulen) {
1946 		if (bpf_dump_raw_ok()) {
1947 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
1948 			ulen = min_t(u32, info.jited_prog_len, ulen);
1949 
1950 			/* for multi-function programs, copy the JITed
1951 			 * instructions for all the functions
1952 			 */
1953 			if (prog->aux->func_cnt) {
1954 				u32 len, free, i;
1955 				u8 *img;
1956 
1957 				free = ulen;
1958 				for (i = 0; i < prog->aux->func_cnt; i++) {
1959 					len = prog->aux->func[i]->jited_len;
1960 					len = min_t(u32, len, free);
1961 					img = (u8 *) prog->aux->func[i]->bpf_func;
1962 					if (copy_to_user(uinsns, img, len))
1963 						return -EFAULT;
1964 					uinsns += len;
1965 					free -= len;
1966 					if (!free)
1967 						break;
1968 				}
1969 			} else {
1970 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
1971 					return -EFAULT;
1972 			}
1973 		} else {
1974 			info.jited_prog_insns = 0;
1975 		}
1976 	}
1977 
1978 	ulen = info.nr_jited_ksyms;
1979 	info.nr_jited_ksyms = prog->aux->func_cnt;
1980 	if (info.nr_jited_ksyms && ulen) {
1981 		if (bpf_dump_raw_ok()) {
1982 			u64 __user *user_ksyms;
1983 			ulong ksym_addr;
1984 			u32 i;
1985 
1986 			/* copy the address of the kernel symbol
1987 			 * corresponding to each function
1988 			 */
1989 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
1990 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
1991 			for (i = 0; i < ulen; i++) {
1992 				ksym_addr = (ulong) prog->aux->func[i]->bpf_func;
1993 				ksym_addr &= PAGE_MASK;
1994 				if (put_user((u64) ksym_addr, &user_ksyms[i]))
1995 					return -EFAULT;
1996 			}
1997 		} else {
1998 			info.jited_ksyms = 0;
1999 		}
2000 	}
2001 
2002 	ulen = info.nr_jited_func_lens;
2003 	info.nr_jited_func_lens = prog->aux->func_cnt;
2004 	if (info.nr_jited_func_lens && ulen) {
2005 		if (bpf_dump_raw_ok()) {
2006 			u32 __user *user_lens;
2007 			u32 func_len, i;
2008 
2009 			/* copy the JITed image lengths for each function */
2010 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2011 			user_lens = u64_to_user_ptr(info.jited_func_lens);
2012 			for (i = 0; i < ulen; i++) {
2013 				func_len = prog->aux->func[i]->jited_len;
2014 				if (put_user(func_len, &user_lens[i]))
2015 					return -EFAULT;
2016 			}
2017 		} else {
2018 			info.jited_func_lens = 0;
2019 		}
2020 	}
2021 
2022 done:
2023 	if (copy_to_user(uinfo, &info, info_len) ||
2024 	    put_user(info_len, &uattr->info.info_len))
2025 		return -EFAULT;
2026 
2027 	return 0;
2028 }
2029 
2030 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2031 				  const union bpf_attr *attr,
2032 				  union bpf_attr __user *uattr)
2033 {
2034 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2035 	struct bpf_map_info info = {};
2036 	u32 info_len = attr->info.info_len;
2037 	int err;
2038 
2039 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2040 	if (err)
2041 		return err;
2042 	info_len = min_t(u32, sizeof(info), info_len);
2043 
2044 	info.type = map->map_type;
2045 	info.id = map->id;
2046 	info.key_size = map->key_size;
2047 	info.value_size = map->value_size;
2048 	info.max_entries = map->max_entries;
2049 	info.map_flags = map->map_flags;
2050 	memcpy(info.name, map->name, sizeof(map->name));
2051 
2052 	if (map->btf) {
2053 		info.btf_id = btf_id(map->btf);
2054 		info.btf_key_type_id = map->btf_key_type_id;
2055 		info.btf_value_type_id = map->btf_value_type_id;
2056 	}
2057 
2058 	if (bpf_map_is_dev_bound(map)) {
2059 		err = bpf_map_offload_info_fill(&info, map);
2060 		if (err)
2061 			return err;
2062 	}
2063 
2064 	if (copy_to_user(uinfo, &info, info_len) ||
2065 	    put_user(info_len, &uattr->info.info_len))
2066 		return -EFAULT;
2067 
2068 	return 0;
2069 }
2070 
2071 static int bpf_btf_get_info_by_fd(struct btf *btf,
2072 				  const union bpf_attr *attr,
2073 				  union bpf_attr __user *uattr)
2074 {
2075 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2076 	u32 info_len = attr->info.info_len;
2077 	int err;
2078 
2079 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2080 	if (err)
2081 		return err;
2082 
2083 	return btf_get_info_by_fd(btf, attr, uattr);
2084 }
2085 
2086 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2087 
2088 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2089 				  union bpf_attr __user *uattr)
2090 {
2091 	int ufd = attr->info.bpf_fd;
2092 	struct fd f;
2093 	int err;
2094 
2095 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2096 		return -EINVAL;
2097 
2098 	f = fdget(ufd);
2099 	if (!f.file)
2100 		return -EBADFD;
2101 
2102 	if (f.file->f_op == &bpf_prog_fops)
2103 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2104 					      uattr);
2105 	else if (f.file->f_op == &bpf_map_fops)
2106 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2107 					     uattr);
2108 	else if (f.file->f_op == &btf_fops)
2109 		err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2110 	else
2111 		err = -EINVAL;
2112 
2113 	fdput(f);
2114 	return err;
2115 }
2116 
2117 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2118 
2119 static int bpf_btf_load(const union bpf_attr *attr)
2120 {
2121 	if (CHECK_ATTR(BPF_BTF_LOAD))
2122 		return -EINVAL;
2123 
2124 	if (!capable(CAP_SYS_ADMIN))
2125 		return -EPERM;
2126 
2127 	return btf_new_fd(attr);
2128 }
2129 
2130 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2131 
2132 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2133 {
2134 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2135 		return -EINVAL;
2136 
2137 	if (!capable(CAP_SYS_ADMIN))
2138 		return -EPERM;
2139 
2140 	return btf_get_fd_by_id(attr->btf_id);
2141 }
2142 
2143 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2144 				    union bpf_attr __user *uattr,
2145 				    u32 prog_id, u32 fd_type,
2146 				    const char *buf, u64 probe_offset,
2147 				    u64 probe_addr)
2148 {
2149 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2150 	u32 len = buf ? strlen(buf) : 0, input_len;
2151 	int err = 0;
2152 
2153 	if (put_user(len, &uattr->task_fd_query.buf_len))
2154 		return -EFAULT;
2155 	input_len = attr->task_fd_query.buf_len;
2156 	if (input_len && ubuf) {
2157 		if (!len) {
2158 			/* nothing to copy, just make ubuf NULL terminated */
2159 			char zero = '\0';
2160 
2161 			if (put_user(zero, ubuf))
2162 				return -EFAULT;
2163 		} else if (input_len >= len + 1) {
2164 			/* ubuf can hold the string with NULL terminator */
2165 			if (copy_to_user(ubuf, buf, len + 1))
2166 				return -EFAULT;
2167 		} else {
2168 			/* ubuf cannot hold the string with NULL terminator,
2169 			 * do a partial copy with NULL terminator.
2170 			 */
2171 			char zero = '\0';
2172 
2173 			err = -ENOSPC;
2174 			if (copy_to_user(ubuf, buf, input_len - 1))
2175 				return -EFAULT;
2176 			if (put_user(zero, ubuf + input_len - 1))
2177 				return -EFAULT;
2178 		}
2179 	}
2180 
2181 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2182 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2183 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2184 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2185 		return -EFAULT;
2186 
2187 	return err;
2188 }
2189 
2190 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2191 
2192 static int bpf_task_fd_query(const union bpf_attr *attr,
2193 			     union bpf_attr __user *uattr)
2194 {
2195 	pid_t pid = attr->task_fd_query.pid;
2196 	u32 fd = attr->task_fd_query.fd;
2197 	const struct perf_event *event;
2198 	struct files_struct *files;
2199 	struct task_struct *task;
2200 	struct file *file;
2201 	int err;
2202 
2203 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2204 		return -EINVAL;
2205 
2206 	if (!capable(CAP_SYS_ADMIN))
2207 		return -EPERM;
2208 
2209 	if (attr->task_fd_query.flags != 0)
2210 		return -EINVAL;
2211 
2212 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2213 	if (!task)
2214 		return -ENOENT;
2215 
2216 	files = get_files_struct(task);
2217 	put_task_struct(task);
2218 	if (!files)
2219 		return -ENOENT;
2220 
2221 	err = 0;
2222 	spin_lock(&files->file_lock);
2223 	file = fcheck_files(files, fd);
2224 	if (!file)
2225 		err = -EBADF;
2226 	else
2227 		get_file(file);
2228 	spin_unlock(&files->file_lock);
2229 	put_files_struct(files);
2230 
2231 	if (err)
2232 		goto out;
2233 
2234 	if (file->f_op == &bpf_raw_tp_fops) {
2235 		struct bpf_raw_tracepoint *raw_tp = file->private_data;
2236 		struct bpf_raw_event_map *btp = raw_tp->btp;
2237 
2238 		err = bpf_task_fd_query_copy(attr, uattr,
2239 					     raw_tp->prog->aux->id,
2240 					     BPF_FD_TYPE_RAW_TRACEPOINT,
2241 					     btp->tp->name, 0, 0);
2242 		goto put_file;
2243 	}
2244 
2245 	event = perf_get_event(file);
2246 	if (!IS_ERR(event)) {
2247 		u64 probe_offset, probe_addr;
2248 		u32 prog_id, fd_type;
2249 		const char *buf;
2250 
2251 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2252 					      &buf, &probe_offset,
2253 					      &probe_addr);
2254 		if (!err)
2255 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2256 						     fd_type, buf,
2257 						     probe_offset,
2258 						     probe_addr);
2259 		goto put_file;
2260 	}
2261 
2262 	err = -ENOTSUPP;
2263 put_file:
2264 	fput(file);
2265 out:
2266 	return err;
2267 }
2268 
2269 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2270 {
2271 	union bpf_attr attr = {};
2272 	int err;
2273 
2274 	if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2275 		return -EPERM;
2276 
2277 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2278 	if (err)
2279 		return err;
2280 	size = min_t(u32, size, sizeof(attr));
2281 
2282 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
2283 	if (copy_from_user(&attr, uattr, size) != 0)
2284 		return -EFAULT;
2285 
2286 	err = security_bpf(cmd, &attr, size);
2287 	if (err < 0)
2288 		return err;
2289 
2290 	switch (cmd) {
2291 	case BPF_MAP_CREATE:
2292 		err = map_create(&attr);
2293 		break;
2294 	case BPF_MAP_LOOKUP_ELEM:
2295 		err = map_lookup_elem(&attr);
2296 		break;
2297 	case BPF_MAP_UPDATE_ELEM:
2298 		err = map_update_elem(&attr);
2299 		break;
2300 	case BPF_MAP_DELETE_ELEM:
2301 		err = map_delete_elem(&attr);
2302 		break;
2303 	case BPF_MAP_GET_NEXT_KEY:
2304 		err = map_get_next_key(&attr);
2305 		break;
2306 	case BPF_PROG_LOAD:
2307 		err = bpf_prog_load(&attr);
2308 		break;
2309 	case BPF_OBJ_PIN:
2310 		err = bpf_obj_pin(&attr);
2311 		break;
2312 	case BPF_OBJ_GET:
2313 		err = bpf_obj_get(&attr);
2314 		break;
2315 	case BPF_PROG_ATTACH:
2316 		err = bpf_prog_attach(&attr);
2317 		break;
2318 	case BPF_PROG_DETACH:
2319 		err = bpf_prog_detach(&attr);
2320 		break;
2321 	case BPF_PROG_QUERY:
2322 		err = bpf_prog_query(&attr, uattr);
2323 		break;
2324 	case BPF_PROG_TEST_RUN:
2325 		err = bpf_prog_test_run(&attr, uattr);
2326 		break;
2327 	case BPF_PROG_GET_NEXT_ID:
2328 		err = bpf_obj_get_next_id(&attr, uattr,
2329 					  &prog_idr, &prog_idr_lock);
2330 		break;
2331 	case BPF_MAP_GET_NEXT_ID:
2332 		err = bpf_obj_get_next_id(&attr, uattr,
2333 					  &map_idr, &map_idr_lock);
2334 		break;
2335 	case BPF_PROG_GET_FD_BY_ID:
2336 		err = bpf_prog_get_fd_by_id(&attr);
2337 		break;
2338 	case BPF_MAP_GET_FD_BY_ID:
2339 		err = bpf_map_get_fd_by_id(&attr);
2340 		break;
2341 	case BPF_OBJ_GET_INFO_BY_FD:
2342 		err = bpf_obj_get_info_by_fd(&attr, uattr);
2343 		break;
2344 	case BPF_RAW_TRACEPOINT_OPEN:
2345 		err = bpf_raw_tracepoint_open(&attr);
2346 		break;
2347 	case BPF_BTF_LOAD:
2348 		err = bpf_btf_load(&attr);
2349 		break;
2350 	case BPF_BTF_GET_FD_BY_ID:
2351 		err = bpf_btf_get_fd_by_id(&attr);
2352 		break;
2353 	case BPF_TASK_FD_QUERY:
2354 		err = bpf_task_fd_query(&attr, uattr);
2355 		break;
2356 	default:
2357 		err = -EINVAL;
2358 		break;
2359 	}
2360 
2361 	return err;
2362 }
2363