xref: /linux/kernel/bpf/syscall.c (revision 13a370b9d275959ac75e92dc14e43eeae75804f8)
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->map_type,
332 		   map->key_size,
333 		   map->value_size,
334 		   map->max_entries,
335 		   map->map_flags,
336 		   map->pages * 1ULL << PAGE_SHIFT);
337 
338 	if (owner_prog_type) {
339 		seq_printf(m, "owner_prog_type:\t%u\n",
340 			   owner_prog_type);
341 		seq_printf(m, "owner_jited:\t%u\n",
342 			   owner_jited);
343 	}
344 }
345 #endif
346 
347 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
348 			      loff_t *ppos)
349 {
350 	/* We need this handler such that alloc_file() enables
351 	 * f_mode with FMODE_CAN_READ.
352 	 */
353 	return -EINVAL;
354 }
355 
356 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
357 			       size_t siz, loff_t *ppos)
358 {
359 	/* We need this handler such that alloc_file() enables
360 	 * f_mode with FMODE_CAN_WRITE.
361 	 */
362 	return -EINVAL;
363 }
364 
365 const struct file_operations bpf_map_fops = {
366 #ifdef CONFIG_PROC_FS
367 	.show_fdinfo	= bpf_map_show_fdinfo,
368 #endif
369 	.release	= bpf_map_release,
370 	.read		= bpf_dummy_read,
371 	.write		= bpf_dummy_write,
372 };
373 
374 int bpf_map_new_fd(struct bpf_map *map, int flags)
375 {
376 	int ret;
377 
378 	ret = security_bpf_map(map, OPEN_FMODE(flags));
379 	if (ret < 0)
380 		return ret;
381 
382 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
383 				flags | O_CLOEXEC);
384 }
385 
386 int bpf_get_file_flag(int flags)
387 {
388 	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
389 		return -EINVAL;
390 	if (flags & BPF_F_RDONLY)
391 		return O_RDONLY;
392 	if (flags & BPF_F_WRONLY)
393 		return O_WRONLY;
394 	return O_RDWR;
395 }
396 
397 /* helper macro to check that unused fields 'union bpf_attr' are zero */
398 #define CHECK_ATTR(CMD) \
399 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
400 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
401 		   sizeof(*attr) - \
402 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
403 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
404 
405 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
406  * Return 0 on success and < 0 on error.
407  */
408 static int bpf_obj_name_cpy(char *dst, const char *src)
409 {
410 	const char *end = src + BPF_OBJ_NAME_LEN;
411 
412 	memset(dst, 0, BPF_OBJ_NAME_LEN);
413 
414 	/* Copy all isalnum() and '_' char */
415 	while (src < end && *src) {
416 		if (!isalnum(*src) && *src != '_')
417 			return -EINVAL;
418 		*dst++ = *src++;
419 	}
420 
421 	/* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
422 	if (src == end)
423 		return -EINVAL;
424 
425 	return 0;
426 }
427 
428 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
429 /* called via syscall */
430 static int map_create(union bpf_attr *attr)
431 {
432 	int numa_node = bpf_map_attr_numa_node(attr);
433 	struct bpf_map *map;
434 	int f_flags;
435 	int err;
436 
437 	err = CHECK_ATTR(BPF_MAP_CREATE);
438 	if (err)
439 		return -EINVAL;
440 
441 	f_flags = bpf_get_file_flag(attr->map_flags);
442 	if (f_flags < 0)
443 		return f_flags;
444 
445 	if (numa_node != NUMA_NO_NODE &&
446 	    ((unsigned int)numa_node >= nr_node_ids ||
447 	     !node_online(numa_node)))
448 		return -EINVAL;
449 
450 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
451 	map = find_and_alloc_map(attr);
452 	if (IS_ERR(map))
453 		return PTR_ERR(map);
454 
455 	err = bpf_obj_name_cpy(map->name, attr->map_name);
456 	if (err)
457 		goto free_map_nouncharge;
458 
459 	atomic_set(&map->refcnt, 1);
460 	atomic_set(&map->usercnt, 1);
461 
462 	if (bpf_map_support_seq_show(map) &&
463 	    (attr->btf_key_type_id || attr->btf_value_type_id)) {
464 		struct btf *btf;
465 
466 		if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
467 			err = -EINVAL;
468 			goto free_map_nouncharge;
469 		}
470 
471 		btf = btf_get_by_fd(attr->btf_fd);
472 		if (IS_ERR(btf)) {
473 			err = PTR_ERR(btf);
474 			goto free_map_nouncharge;
475 		}
476 
477 		err = map->ops->map_check_btf(map, btf, attr->btf_key_type_id,
478 					      attr->btf_value_type_id);
479 		if (err) {
480 			btf_put(btf);
481 			goto free_map_nouncharge;
482 		}
483 
484 		map->btf = btf;
485 		map->btf_key_type_id = attr->btf_key_type_id;
486 		map->btf_value_type_id = attr->btf_value_type_id;
487 	}
488 
489 	err = security_bpf_map_alloc(map);
490 	if (err)
491 		goto free_map_nouncharge;
492 
493 	err = bpf_map_charge_memlock(map);
494 	if (err)
495 		goto free_map_sec;
496 
497 	err = bpf_map_alloc_id(map);
498 	if (err)
499 		goto free_map;
500 
501 	err = bpf_map_new_fd(map, f_flags);
502 	if (err < 0) {
503 		/* failed to allocate fd.
504 		 * bpf_map_put() is needed because the above
505 		 * bpf_map_alloc_id() has published the map
506 		 * to the userspace and the userspace may
507 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
508 		 */
509 		bpf_map_put(map);
510 		return err;
511 	}
512 
513 	return err;
514 
515 free_map:
516 	bpf_map_uncharge_memlock(map);
517 free_map_sec:
518 	security_bpf_map_free(map);
519 free_map_nouncharge:
520 	btf_put(map->btf);
521 	map->ops->map_free(map);
522 	return err;
523 }
524 
525 /* if error is returned, fd is released.
526  * On success caller should complete fd access with matching fdput()
527  */
528 struct bpf_map *__bpf_map_get(struct fd f)
529 {
530 	if (!f.file)
531 		return ERR_PTR(-EBADF);
532 	if (f.file->f_op != &bpf_map_fops) {
533 		fdput(f);
534 		return ERR_PTR(-EINVAL);
535 	}
536 
537 	return f.file->private_data;
538 }
539 
540 /* prog's and map's refcnt limit */
541 #define BPF_MAX_REFCNT 32768
542 
543 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
544 {
545 	if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
546 		atomic_dec(&map->refcnt);
547 		return ERR_PTR(-EBUSY);
548 	}
549 	if (uref)
550 		atomic_inc(&map->usercnt);
551 	return map;
552 }
553 EXPORT_SYMBOL_GPL(bpf_map_inc);
554 
555 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
556 {
557 	struct fd f = fdget(ufd);
558 	struct bpf_map *map;
559 
560 	map = __bpf_map_get(f);
561 	if (IS_ERR(map))
562 		return map;
563 
564 	map = bpf_map_inc(map, true);
565 	fdput(f);
566 
567 	return map;
568 }
569 
570 /* map_idr_lock should have been held */
571 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
572 					    bool uref)
573 {
574 	int refold;
575 
576 	refold = __atomic_add_unless(&map->refcnt, 1, 0);
577 
578 	if (refold >= BPF_MAX_REFCNT) {
579 		__bpf_map_put(map, false);
580 		return ERR_PTR(-EBUSY);
581 	}
582 
583 	if (!refold)
584 		return ERR_PTR(-ENOENT);
585 
586 	if (uref)
587 		atomic_inc(&map->usercnt);
588 
589 	return map;
590 }
591 
592 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
593 {
594 	return -ENOTSUPP;
595 }
596 
597 /* last field in 'union bpf_attr' used by this command */
598 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
599 
600 static int map_lookup_elem(union bpf_attr *attr)
601 {
602 	void __user *ukey = u64_to_user_ptr(attr->key);
603 	void __user *uvalue = u64_to_user_ptr(attr->value);
604 	int ufd = attr->map_fd;
605 	struct bpf_map *map;
606 	void *key, *value, *ptr;
607 	u32 value_size;
608 	struct fd f;
609 	int err;
610 
611 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
612 		return -EINVAL;
613 
614 	f = fdget(ufd);
615 	map = __bpf_map_get(f);
616 	if (IS_ERR(map))
617 		return PTR_ERR(map);
618 
619 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
620 		err = -EPERM;
621 		goto err_put;
622 	}
623 
624 	key = memdup_user(ukey, map->key_size);
625 	if (IS_ERR(key)) {
626 		err = PTR_ERR(key);
627 		goto err_put;
628 	}
629 
630 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
631 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
632 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
633 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
634 	else if (IS_FD_MAP(map))
635 		value_size = sizeof(u32);
636 	else
637 		value_size = map->value_size;
638 
639 	err = -ENOMEM;
640 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
641 	if (!value)
642 		goto free_key;
643 
644 	if (bpf_map_is_dev_bound(map)) {
645 		err = bpf_map_offload_lookup_elem(map, key, value);
646 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
647 		   map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
648 		err = bpf_percpu_hash_copy(map, key, value);
649 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
650 		err = bpf_percpu_array_copy(map, key, value);
651 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
652 		err = bpf_stackmap_copy(map, key, value);
653 	} else if (IS_FD_ARRAY(map)) {
654 		err = bpf_fd_array_map_lookup_elem(map, key, value);
655 	} else if (IS_FD_HASH(map)) {
656 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
657 	} else {
658 		rcu_read_lock();
659 		ptr = map->ops->map_lookup_elem(map, key);
660 		if (ptr)
661 			memcpy(value, ptr, value_size);
662 		rcu_read_unlock();
663 		err = ptr ? 0 : -ENOENT;
664 	}
665 
666 	if (err)
667 		goto free_value;
668 
669 	err = -EFAULT;
670 	if (copy_to_user(uvalue, value, value_size) != 0)
671 		goto free_value;
672 
673 	err = 0;
674 
675 free_value:
676 	kfree(value);
677 free_key:
678 	kfree(key);
679 err_put:
680 	fdput(f);
681 	return err;
682 }
683 
684 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
685 
686 static int map_update_elem(union bpf_attr *attr)
687 {
688 	void __user *ukey = u64_to_user_ptr(attr->key);
689 	void __user *uvalue = u64_to_user_ptr(attr->value);
690 	int ufd = attr->map_fd;
691 	struct bpf_map *map;
692 	void *key, *value;
693 	u32 value_size;
694 	struct fd f;
695 	int err;
696 
697 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
698 		return -EINVAL;
699 
700 	f = fdget(ufd);
701 	map = __bpf_map_get(f);
702 	if (IS_ERR(map))
703 		return PTR_ERR(map);
704 
705 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
706 		err = -EPERM;
707 		goto err_put;
708 	}
709 
710 	key = memdup_user(ukey, map->key_size);
711 	if (IS_ERR(key)) {
712 		err = PTR_ERR(key);
713 		goto err_put;
714 	}
715 
716 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
717 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
718 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
719 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
720 	else
721 		value_size = map->value_size;
722 
723 	err = -ENOMEM;
724 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
725 	if (!value)
726 		goto free_key;
727 
728 	err = -EFAULT;
729 	if (copy_from_user(value, uvalue, value_size) != 0)
730 		goto free_value;
731 
732 	/* Need to create a kthread, thus must support schedule */
733 	if (bpf_map_is_dev_bound(map)) {
734 		err = bpf_map_offload_update_elem(map, key, value, attr->flags);
735 		goto out;
736 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP) {
737 		err = map->ops->map_update_elem(map, key, value, attr->flags);
738 		goto out;
739 	}
740 
741 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
742 	 * inside bpf map update or delete otherwise deadlocks are possible
743 	 */
744 	preempt_disable();
745 	__this_cpu_inc(bpf_prog_active);
746 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
747 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
748 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
749 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
750 		err = bpf_percpu_array_update(map, key, value, attr->flags);
751 	} else if (IS_FD_ARRAY(map)) {
752 		rcu_read_lock();
753 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
754 						   attr->flags);
755 		rcu_read_unlock();
756 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
757 		rcu_read_lock();
758 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
759 						  attr->flags);
760 		rcu_read_unlock();
761 	} else {
762 		rcu_read_lock();
763 		err = map->ops->map_update_elem(map, key, value, attr->flags);
764 		rcu_read_unlock();
765 	}
766 	__this_cpu_dec(bpf_prog_active);
767 	preempt_enable();
768 out:
769 free_value:
770 	kfree(value);
771 free_key:
772 	kfree(key);
773 err_put:
774 	fdput(f);
775 	return err;
776 }
777 
778 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
779 
780 static int map_delete_elem(union bpf_attr *attr)
781 {
782 	void __user *ukey = u64_to_user_ptr(attr->key);
783 	int ufd = attr->map_fd;
784 	struct bpf_map *map;
785 	struct fd f;
786 	void *key;
787 	int err;
788 
789 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
790 		return -EINVAL;
791 
792 	f = fdget(ufd);
793 	map = __bpf_map_get(f);
794 	if (IS_ERR(map))
795 		return PTR_ERR(map);
796 
797 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
798 		err = -EPERM;
799 		goto err_put;
800 	}
801 
802 	key = memdup_user(ukey, map->key_size);
803 	if (IS_ERR(key)) {
804 		err = PTR_ERR(key);
805 		goto err_put;
806 	}
807 
808 	if (bpf_map_is_dev_bound(map)) {
809 		err = bpf_map_offload_delete_elem(map, key);
810 		goto out;
811 	}
812 
813 	preempt_disable();
814 	__this_cpu_inc(bpf_prog_active);
815 	rcu_read_lock();
816 	err = map->ops->map_delete_elem(map, key);
817 	rcu_read_unlock();
818 	__this_cpu_dec(bpf_prog_active);
819 	preempt_enable();
820 out:
821 	kfree(key);
822 err_put:
823 	fdput(f);
824 	return err;
825 }
826 
827 /* last field in 'union bpf_attr' used by this command */
828 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
829 
830 static int map_get_next_key(union bpf_attr *attr)
831 {
832 	void __user *ukey = u64_to_user_ptr(attr->key);
833 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
834 	int ufd = attr->map_fd;
835 	struct bpf_map *map;
836 	void *key, *next_key;
837 	struct fd f;
838 	int err;
839 
840 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
841 		return -EINVAL;
842 
843 	f = fdget(ufd);
844 	map = __bpf_map_get(f);
845 	if (IS_ERR(map))
846 		return PTR_ERR(map);
847 
848 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
849 		err = -EPERM;
850 		goto err_put;
851 	}
852 
853 	if (ukey) {
854 		key = memdup_user(ukey, map->key_size);
855 		if (IS_ERR(key)) {
856 			err = PTR_ERR(key);
857 			goto err_put;
858 		}
859 	} else {
860 		key = NULL;
861 	}
862 
863 	err = -ENOMEM;
864 	next_key = kmalloc(map->key_size, GFP_USER);
865 	if (!next_key)
866 		goto free_key;
867 
868 	if (bpf_map_is_dev_bound(map)) {
869 		err = bpf_map_offload_get_next_key(map, key, next_key);
870 		goto out;
871 	}
872 
873 	rcu_read_lock();
874 	err = map->ops->map_get_next_key(map, key, next_key);
875 	rcu_read_unlock();
876 out:
877 	if (err)
878 		goto free_next_key;
879 
880 	err = -EFAULT;
881 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
882 		goto free_next_key;
883 
884 	err = 0;
885 
886 free_next_key:
887 	kfree(next_key);
888 free_key:
889 	kfree(key);
890 err_put:
891 	fdput(f);
892 	return err;
893 }
894 
895 static const struct bpf_prog_ops * const bpf_prog_types[] = {
896 #define BPF_PROG_TYPE(_id, _name) \
897 	[_id] = & _name ## _prog_ops,
898 #define BPF_MAP_TYPE(_id, _ops)
899 #include <linux/bpf_types.h>
900 #undef BPF_PROG_TYPE
901 #undef BPF_MAP_TYPE
902 };
903 
904 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
905 {
906 	const struct bpf_prog_ops *ops;
907 
908 	if (type >= ARRAY_SIZE(bpf_prog_types))
909 		return -EINVAL;
910 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
911 	ops = bpf_prog_types[type];
912 	if (!ops)
913 		return -EINVAL;
914 
915 	if (!bpf_prog_is_dev_bound(prog->aux))
916 		prog->aux->ops = ops;
917 	else
918 		prog->aux->ops = &bpf_offload_prog_ops;
919 	prog->type = type;
920 	return 0;
921 }
922 
923 /* drop refcnt on maps used by eBPF program and free auxilary data */
924 static void free_used_maps(struct bpf_prog_aux *aux)
925 {
926 	int i;
927 
928 	for (i = 0; i < aux->used_map_cnt; i++)
929 		bpf_map_put(aux->used_maps[i]);
930 
931 	kfree(aux->used_maps);
932 }
933 
934 int __bpf_prog_charge(struct user_struct *user, u32 pages)
935 {
936 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
937 	unsigned long user_bufs;
938 
939 	if (user) {
940 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
941 		if (user_bufs > memlock_limit) {
942 			atomic_long_sub(pages, &user->locked_vm);
943 			return -EPERM;
944 		}
945 	}
946 
947 	return 0;
948 }
949 
950 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
951 {
952 	if (user)
953 		atomic_long_sub(pages, &user->locked_vm);
954 }
955 
956 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
957 {
958 	struct user_struct *user = get_current_user();
959 	int ret;
960 
961 	ret = __bpf_prog_charge(user, prog->pages);
962 	if (ret) {
963 		free_uid(user);
964 		return ret;
965 	}
966 
967 	prog->aux->user = user;
968 	return 0;
969 }
970 
971 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
972 {
973 	struct user_struct *user = prog->aux->user;
974 
975 	__bpf_prog_uncharge(user, prog->pages);
976 	free_uid(user);
977 }
978 
979 static int bpf_prog_alloc_id(struct bpf_prog *prog)
980 {
981 	int id;
982 
983 	idr_preload(GFP_KERNEL);
984 	spin_lock_bh(&prog_idr_lock);
985 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
986 	if (id > 0)
987 		prog->aux->id = id;
988 	spin_unlock_bh(&prog_idr_lock);
989 	idr_preload_end();
990 
991 	/* id is in [1, INT_MAX) */
992 	if (WARN_ON_ONCE(!id))
993 		return -ENOSPC;
994 
995 	return id > 0 ? 0 : id;
996 }
997 
998 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
999 {
1000 	/* cBPF to eBPF migrations are currently not in the idr store.
1001 	 * Offloaded programs are removed from the store when their device
1002 	 * disappears - even if someone grabs an fd to them they are unusable,
1003 	 * simply waiting for refcnt to drop to be freed.
1004 	 */
1005 	if (!prog->aux->id)
1006 		return;
1007 
1008 	if (do_idr_lock)
1009 		spin_lock_bh(&prog_idr_lock);
1010 	else
1011 		__acquire(&prog_idr_lock);
1012 
1013 	idr_remove(&prog_idr, prog->aux->id);
1014 	prog->aux->id = 0;
1015 
1016 	if (do_idr_lock)
1017 		spin_unlock_bh(&prog_idr_lock);
1018 	else
1019 		__release(&prog_idr_lock);
1020 }
1021 
1022 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1023 {
1024 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1025 
1026 	free_used_maps(aux);
1027 	bpf_prog_uncharge_memlock(aux->prog);
1028 	security_bpf_prog_free(aux);
1029 	bpf_prog_free(aux->prog);
1030 }
1031 
1032 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1033 {
1034 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
1035 		int i;
1036 
1037 		/* bpf_prog_free_id() must be called first */
1038 		bpf_prog_free_id(prog, do_idr_lock);
1039 
1040 		for (i = 0; i < prog->aux->func_cnt; i++)
1041 			bpf_prog_kallsyms_del(prog->aux->func[i]);
1042 		bpf_prog_kallsyms_del(prog);
1043 
1044 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1045 	}
1046 }
1047 
1048 void bpf_prog_put(struct bpf_prog *prog)
1049 {
1050 	__bpf_prog_put(prog, true);
1051 }
1052 EXPORT_SYMBOL_GPL(bpf_prog_put);
1053 
1054 static int bpf_prog_release(struct inode *inode, struct file *filp)
1055 {
1056 	struct bpf_prog *prog = filp->private_data;
1057 
1058 	bpf_prog_put(prog);
1059 	return 0;
1060 }
1061 
1062 #ifdef CONFIG_PROC_FS
1063 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1064 {
1065 	const struct bpf_prog *prog = filp->private_data;
1066 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1067 
1068 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1069 	seq_printf(m,
1070 		   "prog_type:\t%u\n"
1071 		   "prog_jited:\t%u\n"
1072 		   "prog_tag:\t%s\n"
1073 		   "memlock:\t%llu\n",
1074 		   prog->type,
1075 		   prog->jited,
1076 		   prog_tag,
1077 		   prog->pages * 1ULL << PAGE_SHIFT);
1078 }
1079 #endif
1080 
1081 const struct file_operations bpf_prog_fops = {
1082 #ifdef CONFIG_PROC_FS
1083 	.show_fdinfo	= bpf_prog_show_fdinfo,
1084 #endif
1085 	.release	= bpf_prog_release,
1086 	.read		= bpf_dummy_read,
1087 	.write		= bpf_dummy_write,
1088 };
1089 
1090 int bpf_prog_new_fd(struct bpf_prog *prog)
1091 {
1092 	int ret;
1093 
1094 	ret = security_bpf_prog(prog);
1095 	if (ret < 0)
1096 		return ret;
1097 
1098 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1099 				O_RDWR | O_CLOEXEC);
1100 }
1101 
1102 static struct bpf_prog *____bpf_prog_get(struct fd f)
1103 {
1104 	if (!f.file)
1105 		return ERR_PTR(-EBADF);
1106 	if (f.file->f_op != &bpf_prog_fops) {
1107 		fdput(f);
1108 		return ERR_PTR(-EINVAL);
1109 	}
1110 
1111 	return f.file->private_data;
1112 }
1113 
1114 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1115 {
1116 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1117 		atomic_sub(i, &prog->aux->refcnt);
1118 		return ERR_PTR(-EBUSY);
1119 	}
1120 	return prog;
1121 }
1122 EXPORT_SYMBOL_GPL(bpf_prog_add);
1123 
1124 void bpf_prog_sub(struct bpf_prog *prog, int i)
1125 {
1126 	/* Only to be used for undoing previous bpf_prog_add() in some
1127 	 * error path. We still know that another entity in our call
1128 	 * path holds a reference to the program, thus atomic_sub() can
1129 	 * be safely used in such cases!
1130 	 */
1131 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1132 }
1133 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1134 
1135 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1136 {
1137 	return bpf_prog_add(prog, 1);
1138 }
1139 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1140 
1141 /* prog_idr_lock should have been held */
1142 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1143 {
1144 	int refold;
1145 
1146 	refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
1147 
1148 	if (refold >= BPF_MAX_REFCNT) {
1149 		__bpf_prog_put(prog, false);
1150 		return ERR_PTR(-EBUSY);
1151 	}
1152 
1153 	if (!refold)
1154 		return ERR_PTR(-ENOENT);
1155 
1156 	return prog;
1157 }
1158 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1159 
1160 bool bpf_prog_get_ok(struct bpf_prog *prog,
1161 			    enum bpf_prog_type *attach_type, bool attach_drv)
1162 {
1163 	/* not an attachment, just a refcount inc, always allow */
1164 	if (!attach_type)
1165 		return true;
1166 
1167 	if (prog->type != *attach_type)
1168 		return false;
1169 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1170 		return false;
1171 
1172 	return true;
1173 }
1174 
1175 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1176 				       bool attach_drv)
1177 {
1178 	struct fd f = fdget(ufd);
1179 	struct bpf_prog *prog;
1180 
1181 	prog = ____bpf_prog_get(f);
1182 	if (IS_ERR(prog))
1183 		return prog;
1184 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1185 		prog = ERR_PTR(-EINVAL);
1186 		goto out;
1187 	}
1188 
1189 	prog = bpf_prog_inc(prog);
1190 out:
1191 	fdput(f);
1192 	return prog;
1193 }
1194 
1195 struct bpf_prog *bpf_prog_get(u32 ufd)
1196 {
1197 	return __bpf_prog_get(ufd, NULL, false);
1198 }
1199 
1200 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1201 				       bool attach_drv)
1202 {
1203 	return __bpf_prog_get(ufd, &type, attach_drv);
1204 }
1205 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1206 
1207 /* Initially all BPF programs could be loaded w/o specifying
1208  * expected_attach_type. Later for some of them specifying expected_attach_type
1209  * at load time became required so that program could be validated properly.
1210  * Programs of types that are allowed to be loaded both w/ and w/o (for
1211  * backward compatibility) expected_attach_type, should have the default attach
1212  * type assigned to expected_attach_type for the latter case, so that it can be
1213  * validated later at attach time.
1214  *
1215  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1216  * prog type requires it but has some attach types that have to be backward
1217  * compatible.
1218  */
1219 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1220 {
1221 	switch (attr->prog_type) {
1222 	case BPF_PROG_TYPE_CGROUP_SOCK:
1223 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1224 		 * exist so checking for non-zero is the way to go here.
1225 		 */
1226 		if (!attr->expected_attach_type)
1227 			attr->expected_attach_type =
1228 				BPF_CGROUP_INET_SOCK_CREATE;
1229 		break;
1230 	}
1231 }
1232 
1233 static int
1234 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1235 				enum bpf_attach_type expected_attach_type)
1236 {
1237 	switch (prog_type) {
1238 	case BPF_PROG_TYPE_CGROUP_SOCK:
1239 		switch (expected_attach_type) {
1240 		case BPF_CGROUP_INET_SOCK_CREATE:
1241 		case BPF_CGROUP_INET4_POST_BIND:
1242 		case BPF_CGROUP_INET6_POST_BIND:
1243 			return 0;
1244 		default:
1245 			return -EINVAL;
1246 		}
1247 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1248 		switch (expected_attach_type) {
1249 		case BPF_CGROUP_INET4_BIND:
1250 		case BPF_CGROUP_INET6_BIND:
1251 		case BPF_CGROUP_INET4_CONNECT:
1252 		case BPF_CGROUP_INET6_CONNECT:
1253 		case BPF_CGROUP_UDP4_SENDMSG:
1254 		case BPF_CGROUP_UDP6_SENDMSG:
1255 			return 0;
1256 		default:
1257 			return -EINVAL;
1258 		}
1259 	default:
1260 		return 0;
1261 	}
1262 }
1263 
1264 /* last field in 'union bpf_attr' used by this command */
1265 #define	BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1266 
1267 static int bpf_prog_load(union bpf_attr *attr)
1268 {
1269 	enum bpf_prog_type type = attr->prog_type;
1270 	struct bpf_prog *prog;
1271 	int err;
1272 	char license[128];
1273 	bool is_gpl;
1274 
1275 	if (CHECK_ATTR(BPF_PROG_LOAD))
1276 		return -EINVAL;
1277 
1278 	if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1279 		return -EINVAL;
1280 
1281 	/* copy eBPF program license from user space */
1282 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1283 			      sizeof(license) - 1) < 0)
1284 		return -EFAULT;
1285 	license[sizeof(license) - 1] = 0;
1286 
1287 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1288 	is_gpl = license_is_gpl_compatible(license);
1289 
1290 	if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1291 		return -E2BIG;
1292 
1293 	if (type == BPF_PROG_TYPE_KPROBE &&
1294 	    attr->kern_version != LINUX_VERSION_CODE)
1295 		return -EINVAL;
1296 
1297 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1298 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
1299 	    !capable(CAP_SYS_ADMIN))
1300 		return -EPERM;
1301 
1302 	bpf_prog_load_fixup_attach_type(attr);
1303 	if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1304 		return -EINVAL;
1305 
1306 	/* plain bpf_prog allocation */
1307 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1308 	if (!prog)
1309 		return -ENOMEM;
1310 
1311 	prog->expected_attach_type = attr->expected_attach_type;
1312 
1313 	prog->aux->offload_requested = !!attr->prog_ifindex;
1314 
1315 	err = security_bpf_prog_alloc(prog->aux);
1316 	if (err)
1317 		goto free_prog_nouncharge;
1318 
1319 	err = bpf_prog_charge_memlock(prog);
1320 	if (err)
1321 		goto free_prog_sec;
1322 
1323 	prog->len = attr->insn_cnt;
1324 
1325 	err = -EFAULT;
1326 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1327 			   bpf_prog_insn_size(prog)) != 0)
1328 		goto free_prog;
1329 
1330 	prog->orig_prog = NULL;
1331 	prog->jited = 0;
1332 
1333 	atomic_set(&prog->aux->refcnt, 1);
1334 	prog->gpl_compatible = is_gpl ? 1 : 0;
1335 
1336 	if (bpf_prog_is_dev_bound(prog->aux)) {
1337 		err = bpf_prog_offload_init(prog, attr);
1338 		if (err)
1339 			goto free_prog;
1340 	}
1341 
1342 	/* find program type: socket_filter vs tracing_filter */
1343 	err = find_prog_type(type, prog);
1344 	if (err < 0)
1345 		goto free_prog;
1346 
1347 	prog->aux->load_time = ktime_get_boot_ns();
1348 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1349 	if (err)
1350 		goto free_prog;
1351 
1352 	/* run eBPF verifier */
1353 	err = bpf_check(&prog, attr);
1354 	if (err < 0)
1355 		goto free_used_maps;
1356 
1357 	/* eBPF program is ready to be JITed */
1358 	if (!prog->bpf_func)
1359 		prog = bpf_prog_select_runtime(prog, &err);
1360 	if (err < 0)
1361 		goto free_used_maps;
1362 
1363 	err = bpf_prog_alloc_id(prog);
1364 	if (err)
1365 		goto free_used_maps;
1366 
1367 	err = bpf_prog_new_fd(prog);
1368 	if (err < 0) {
1369 		/* failed to allocate fd.
1370 		 * bpf_prog_put() is needed because the above
1371 		 * bpf_prog_alloc_id() has published the prog
1372 		 * to the userspace and the userspace may
1373 		 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1374 		 */
1375 		bpf_prog_put(prog);
1376 		return err;
1377 	}
1378 
1379 	bpf_prog_kallsyms_add(prog);
1380 	return err;
1381 
1382 free_used_maps:
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 #ifdef CONFIG_CGROUP_BPF
1489 
1490 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1491 					     enum bpf_attach_type attach_type)
1492 {
1493 	switch (prog->type) {
1494 	case BPF_PROG_TYPE_CGROUP_SOCK:
1495 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1496 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1497 	default:
1498 		return 0;
1499 	}
1500 }
1501 
1502 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1503 
1504 static int sockmap_get_from_fd(const union bpf_attr *attr,
1505 			       int type, bool attach)
1506 {
1507 	struct bpf_prog *prog = NULL;
1508 	int ufd = attr->target_fd;
1509 	struct bpf_map *map;
1510 	struct fd f;
1511 	int err;
1512 
1513 	f = fdget(ufd);
1514 	map = __bpf_map_get(f);
1515 	if (IS_ERR(map))
1516 		return PTR_ERR(map);
1517 
1518 	if (attach) {
1519 		prog = bpf_prog_get_type(attr->attach_bpf_fd, type);
1520 		if (IS_ERR(prog)) {
1521 			fdput(f);
1522 			return PTR_ERR(prog);
1523 		}
1524 	}
1525 
1526 	err = sock_map_prog(map, prog, attr->attach_type);
1527 	if (err) {
1528 		fdput(f);
1529 		if (prog)
1530 			bpf_prog_put(prog);
1531 		return err;
1532 	}
1533 
1534 	fdput(f);
1535 	return 0;
1536 }
1537 
1538 #define BPF_F_ATTACH_MASK \
1539 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1540 
1541 static int bpf_prog_attach(const union bpf_attr *attr)
1542 {
1543 	enum bpf_prog_type ptype;
1544 	struct bpf_prog *prog;
1545 	struct cgroup *cgrp;
1546 	int ret;
1547 
1548 	if (!capable(CAP_NET_ADMIN))
1549 		return -EPERM;
1550 
1551 	if (CHECK_ATTR(BPF_PROG_ATTACH))
1552 		return -EINVAL;
1553 
1554 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1555 		return -EINVAL;
1556 
1557 	switch (attr->attach_type) {
1558 	case BPF_CGROUP_INET_INGRESS:
1559 	case BPF_CGROUP_INET_EGRESS:
1560 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1561 		break;
1562 	case BPF_CGROUP_INET_SOCK_CREATE:
1563 	case BPF_CGROUP_INET4_POST_BIND:
1564 	case BPF_CGROUP_INET6_POST_BIND:
1565 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1566 		break;
1567 	case BPF_CGROUP_INET4_BIND:
1568 	case BPF_CGROUP_INET6_BIND:
1569 	case BPF_CGROUP_INET4_CONNECT:
1570 	case BPF_CGROUP_INET6_CONNECT:
1571 	case BPF_CGROUP_UDP4_SENDMSG:
1572 	case BPF_CGROUP_UDP6_SENDMSG:
1573 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1574 		break;
1575 	case BPF_CGROUP_SOCK_OPS:
1576 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1577 		break;
1578 	case BPF_CGROUP_DEVICE:
1579 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1580 		break;
1581 	case BPF_SK_MSG_VERDICT:
1582 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, true);
1583 	case BPF_SK_SKB_STREAM_PARSER:
1584 	case BPF_SK_SKB_STREAM_VERDICT:
1585 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, true);
1586 	case BPF_LIRC_MODE2:
1587 		return lirc_prog_attach(attr);
1588 	default:
1589 		return -EINVAL;
1590 	}
1591 
1592 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1593 	if (IS_ERR(prog))
1594 		return PTR_ERR(prog);
1595 
1596 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1597 		bpf_prog_put(prog);
1598 		return -EINVAL;
1599 	}
1600 
1601 	cgrp = cgroup_get_from_fd(attr->target_fd);
1602 	if (IS_ERR(cgrp)) {
1603 		bpf_prog_put(prog);
1604 		return PTR_ERR(cgrp);
1605 	}
1606 
1607 	ret = cgroup_bpf_attach(cgrp, prog, attr->attach_type,
1608 				attr->attach_flags);
1609 	if (ret)
1610 		bpf_prog_put(prog);
1611 	cgroup_put(cgrp);
1612 
1613 	return ret;
1614 }
1615 
1616 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1617 
1618 static int bpf_prog_detach(const union bpf_attr *attr)
1619 {
1620 	enum bpf_prog_type ptype;
1621 	struct bpf_prog *prog;
1622 	struct cgroup *cgrp;
1623 	int ret;
1624 
1625 	if (!capable(CAP_NET_ADMIN))
1626 		return -EPERM;
1627 
1628 	if (CHECK_ATTR(BPF_PROG_DETACH))
1629 		return -EINVAL;
1630 
1631 	switch (attr->attach_type) {
1632 	case BPF_CGROUP_INET_INGRESS:
1633 	case BPF_CGROUP_INET_EGRESS:
1634 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1635 		break;
1636 	case BPF_CGROUP_INET_SOCK_CREATE:
1637 	case BPF_CGROUP_INET4_POST_BIND:
1638 	case BPF_CGROUP_INET6_POST_BIND:
1639 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1640 		break;
1641 	case BPF_CGROUP_INET4_BIND:
1642 	case BPF_CGROUP_INET6_BIND:
1643 	case BPF_CGROUP_INET4_CONNECT:
1644 	case BPF_CGROUP_INET6_CONNECT:
1645 	case BPF_CGROUP_UDP4_SENDMSG:
1646 	case BPF_CGROUP_UDP6_SENDMSG:
1647 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1648 		break;
1649 	case BPF_CGROUP_SOCK_OPS:
1650 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1651 		break;
1652 	case BPF_CGROUP_DEVICE:
1653 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1654 		break;
1655 	case BPF_SK_MSG_VERDICT:
1656 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, false);
1657 	case BPF_SK_SKB_STREAM_PARSER:
1658 	case BPF_SK_SKB_STREAM_VERDICT:
1659 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, false);
1660 	case BPF_LIRC_MODE2:
1661 		return lirc_prog_detach(attr);
1662 	default:
1663 		return -EINVAL;
1664 	}
1665 
1666 	cgrp = cgroup_get_from_fd(attr->target_fd);
1667 	if (IS_ERR(cgrp))
1668 		return PTR_ERR(cgrp);
1669 
1670 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1671 	if (IS_ERR(prog))
1672 		prog = NULL;
1673 
1674 	ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type, 0);
1675 	if (prog)
1676 		bpf_prog_put(prog);
1677 	cgroup_put(cgrp);
1678 	return ret;
1679 }
1680 
1681 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1682 
1683 static int bpf_prog_query(const union bpf_attr *attr,
1684 			  union bpf_attr __user *uattr)
1685 {
1686 	struct cgroup *cgrp;
1687 	int ret;
1688 
1689 	if (!capable(CAP_NET_ADMIN))
1690 		return -EPERM;
1691 	if (CHECK_ATTR(BPF_PROG_QUERY))
1692 		return -EINVAL;
1693 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1694 		return -EINVAL;
1695 
1696 	switch (attr->query.attach_type) {
1697 	case BPF_CGROUP_INET_INGRESS:
1698 	case BPF_CGROUP_INET_EGRESS:
1699 	case BPF_CGROUP_INET_SOCK_CREATE:
1700 	case BPF_CGROUP_INET4_BIND:
1701 	case BPF_CGROUP_INET6_BIND:
1702 	case BPF_CGROUP_INET4_POST_BIND:
1703 	case BPF_CGROUP_INET6_POST_BIND:
1704 	case BPF_CGROUP_INET4_CONNECT:
1705 	case BPF_CGROUP_INET6_CONNECT:
1706 	case BPF_CGROUP_UDP4_SENDMSG:
1707 	case BPF_CGROUP_UDP6_SENDMSG:
1708 	case BPF_CGROUP_SOCK_OPS:
1709 	case BPF_CGROUP_DEVICE:
1710 		break;
1711 	case BPF_LIRC_MODE2:
1712 		return lirc_prog_query(attr, uattr);
1713 	default:
1714 		return -EINVAL;
1715 	}
1716 	cgrp = cgroup_get_from_fd(attr->query.target_fd);
1717 	if (IS_ERR(cgrp))
1718 		return PTR_ERR(cgrp);
1719 	ret = cgroup_bpf_query(cgrp, attr, uattr);
1720 	cgroup_put(cgrp);
1721 	return ret;
1722 }
1723 #endif /* CONFIG_CGROUP_BPF */
1724 
1725 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1726 
1727 static int bpf_prog_test_run(const union bpf_attr *attr,
1728 			     union bpf_attr __user *uattr)
1729 {
1730 	struct bpf_prog *prog;
1731 	int ret = -ENOTSUPP;
1732 
1733 	if (!capable(CAP_SYS_ADMIN))
1734 		return -EPERM;
1735 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1736 		return -EINVAL;
1737 
1738 	prog = bpf_prog_get(attr->test.prog_fd);
1739 	if (IS_ERR(prog))
1740 		return PTR_ERR(prog);
1741 
1742 	if (prog->aux->ops->test_run)
1743 		ret = prog->aux->ops->test_run(prog, attr, uattr);
1744 
1745 	bpf_prog_put(prog);
1746 	return ret;
1747 }
1748 
1749 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1750 
1751 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1752 			       union bpf_attr __user *uattr,
1753 			       struct idr *idr,
1754 			       spinlock_t *lock)
1755 {
1756 	u32 next_id = attr->start_id;
1757 	int err = 0;
1758 
1759 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1760 		return -EINVAL;
1761 
1762 	if (!capable(CAP_SYS_ADMIN))
1763 		return -EPERM;
1764 
1765 	next_id++;
1766 	spin_lock_bh(lock);
1767 	if (!idr_get_next(idr, &next_id))
1768 		err = -ENOENT;
1769 	spin_unlock_bh(lock);
1770 
1771 	if (!err)
1772 		err = put_user(next_id, &uattr->next_id);
1773 
1774 	return err;
1775 }
1776 
1777 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1778 
1779 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1780 {
1781 	struct bpf_prog *prog;
1782 	u32 id = attr->prog_id;
1783 	int fd;
1784 
1785 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1786 		return -EINVAL;
1787 
1788 	if (!capable(CAP_SYS_ADMIN))
1789 		return -EPERM;
1790 
1791 	spin_lock_bh(&prog_idr_lock);
1792 	prog = idr_find(&prog_idr, id);
1793 	if (prog)
1794 		prog = bpf_prog_inc_not_zero(prog);
1795 	else
1796 		prog = ERR_PTR(-ENOENT);
1797 	spin_unlock_bh(&prog_idr_lock);
1798 
1799 	if (IS_ERR(prog))
1800 		return PTR_ERR(prog);
1801 
1802 	fd = bpf_prog_new_fd(prog);
1803 	if (fd < 0)
1804 		bpf_prog_put(prog);
1805 
1806 	return fd;
1807 }
1808 
1809 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1810 
1811 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1812 {
1813 	struct bpf_map *map;
1814 	u32 id = attr->map_id;
1815 	int f_flags;
1816 	int fd;
1817 
1818 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1819 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1820 		return -EINVAL;
1821 
1822 	if (!capable(CAP_SYS_ADMIN))
1823 		return -EPERM;
1824 
1825 	f_flags = bpf_get_file_flag(attr->open_flags);
1826 	if (f_flags < 0)
1827 		return f_flags;
1828 
1829 	spin_lock_bh(&map_idr_lock);
1830 	map = idr_find(&map_idr, id);
1831 	if (map)
1832 		map = bpf_map_inc_not_zero(map, true);
1833 	else
1834 		map = ERR_PTR(-ENOENT);
1835 	spin_unlock_bh(&map_idr_lock);
1836 
1837 	if (IS_ERR(map))
1838 		return PTR_ERR(map);
1839 
1840 	fd = bpf_map_new_fd(map, f_flags);
1841 	if (fd < 0)
1842 		bpf_map_put(map);
1843 
1844 	return fd;
1845 }
1846 
1847 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1848 					      unsigned long addr)
1849 {
1850 	int i;
1851 
1852 	for (i = 0; i < prog->aux->used_map_cnt; i++)
1853 		if (prog->aux->used_maps[i] == (void *)addr)
1854 			return prog->aux->used_maps[i];
1855 	return NULL;
1856 }
1857 
1858 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1859 {
1860 	const struct bpf_map *map;
1861 	struct bpf_insn *insns;
1862 	u64 imm;
1863 	int i;
1864 
1865 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1866 			GFP_USER);
1867 	if (!insns)
1868 		return insns;
1869 
1870 	for (i = 0; i < prog->len; i++) {
1871 		if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1872 			insns[i].code = BPF_JMP | BPF_CALL;
1873 			insns[i].imm = BPF_FUNC_tail_call;
1874 			/* fall-through */
1875 		}
1876 		if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1877 		    insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1878 			if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1879 				insns[i].code = BPF_JMP | BPF_CALL;
1880 			if (!bpf_dump_raw_ok())
1881 				insns[i].imm = 0;
1882 			continue;
1883 		}
1884 
1885 		if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1886 			continue;
1887 
1888 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1889 		map = bpf_map_from_imm(prog, imm);
1890 		if (map) {
1891 			insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1892 			insns[i].imm = map->id;
1893 			insns[i + 1].imm = 0;
1894 			continue;
1895 		}
1896 
1897 		if (!bpf_dump_raw_ok() &&
1898 		    imm == (unsigned long)prog->aux) {
1899 			insns[i].imm = 0;
1900 			insns[i + 1].imm = 0;
1901 			continue;
1902 		}
1903 	}
1904 
1905 	return insns;
1906 }
1907 
1908 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1909 				   const union bpf_attr *attr,
1910 				   union bpf_attr __user *uattr)
1911 {
1912 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1913 	struct bpf_prog_info info = {};
1914 	u32 info_len = attr->info.info_len;
1915 	char __user *uinsns;
1916 	u32 ulen;
1917 	int err;
1918 
1919 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1920 	if (err)
1921 		return err;
1922 	info_len = min_t(u32, sizeof(info), info_len);
1923 
1924 	if (copy_from_user(&info, uinfo, info_len))
1925 		return -EFAULT;
1926 
1927 	info.type = prog->type;
1928 	info.id = prog->aux->id;
1929 	info.load_time = prog->aux->load_time;
1930 	info.created_by_uid = from_kuid_munged(current_user_ns(),
1931 					       prog->aux->user->uid);
1932 	info.gpl_compatible = prog->gpl_compatible;
1933 
1934 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
1935 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1936 
1937 	ulen = info.nr_map_ids;
1938 	info.nr_map_ids = prog->aux->used_map_cnt;
1939 	ulen = min_t(u32, info.nr_map_ids, ulen);
1940 	if (ulen) {
1941 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1942 		u32 i;
1943 
1944 		for (i = 0; i < ulen; i++)
1945 			if (put_user(prog->aux->used_maps[i]->id,
1946 				     &user_map_ids[i]))
1947 				return -EFAULT;
1948 	}
1949 
1950 	if (!capable(CAP_SYS_ADMIN)) {
1951 		info.jited_prog_len = 0;
1952 		info.xlated_prog_len = 0;
1953 		info.nr_jited_ksyms = 0;
1954 		goto done;
1955 	}
1956 
1957 	ulen = info.xlated_prog_len;
1958 	info.xlated_prog_len = bpf_prog_insn_size(prog);
1959 	if (info.xlated_prog_len && ulen) {
1960 		struct bpf_insn *insns_sanitized;
1961 		bool fault;
1962 
1963 		if (prog->blinded && !bpf_dump_raw_ok()) {
1964 			info.xlated_prog_insns = 0;
1965 			goto done;
1966 		}
1967 		insns_sanitized = bpf_insn_prepare_dump(prog);
1968 		if (!insns_sanitized)
1969 			return -ENOMEM;
1970 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1971 		ulen = min_t(u32, info.xlated_prog_len, ulen);
1972 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
1973 		kfree(insns_sanitized);
1974 		if (fault)
1975 			return -EFAULT;
1976 	}
1977 
1978 	if (bpf_prog_is_dev_bound(prog->aux)) {
1979 		err = bpf_prog_offload_info_fill(&info, prog);
1980 		if (err)
1981 			return err;
1982 		goto done;
1983 	}
1984 
1985 	/* NOTE: the following code is supposed to be skipped for offload.
1986 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
1987 	 * for offload.
1988 	 */
1989 	ulen = info.jited_prog_len;
1990 	if (prog->aux->func_cnt) {
1991 		u32 i;
1992 
1993 		info.jited_prog_len = 0;
1994 		for (i = 0; i < prog->aux->func_cnt; i++)
1995 			info.jited_prog_len += prog->aux->func[i]->jited_len;
1996 	} else {
1997 		info.jited_prog_len = prog->jited_len;
1998 	}
1999 
2000 	if (info.jited_prog_len && ulen) {
2001 		if (bpf_dump_raw_ok()) {
2002 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
2003 			ulen = min_t(u32, info.jited_prog_len, ulen);
2004 
2005 			/* for multi-function programs, copy the JITed
2006 			 * instructions for all the functions
2007 			 */
2008 			if (prog->aux->func_cnt) {
2009 				u32 len, free, i;
2010 				u8 *img;
2011 
2012 				free = ulen;
2013 				for (i = 0; i < prog->aux->func_cnt; i++) {
2014 					len = prog->aux->func[i]->jited_len;
2015 					len = min_t(u32, len, free);
2016 					img = (u8 *) prog->aux->func[i]->bpf_func;
2017 					if (copy_to_user(uinsns, img, len))
2018 						return -EFAULT;
2019 					uinsns += len;
2020 					free -= len;
2021 					if (!free)
2022 						break;
2023 				}
2024 			} else {
2025 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
2026 					return -EFAULT;
2027 			}
2028 		} else {
2029 			info.jited_prog_insns = 0;
2030 		}
2031 	}
2032 
2033 	ulen = info.nr_jited_ksyms;
2034 	info.nr_jited_ksyms = prog->aux->func_cnt;
2035 	if (info.nr_jited_ksyms && ulen) {
2036 		if (bpf_dump_raw_ok()) {
2037 			u64 __user *user_ksyms;
2038 			ulong ksym_addr;
2039 			u32 i;
2040 
2041 			/* copy the address of the kernel symbol
2042 			 * corresponding to each function
2043 			 */
2044 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2045 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2046 			for (i = 0; i < ulen; i++) {
2047 				ksym_addr = (ulong) prog->aux->func[i]->bpf_func;
2048 				ksym_addr &= PAGE_MASK;
2049 				if (put_user((u64) ksym_addr, &user_ksyms[i]))
2050 					return -EFAULT;
2051 			}
2052 		} else {
2053 			info.jited_ksyms = 0;
2054 		}
2055 	}
2056 
2057 	ulen = info.nr_jited_func_lens;
2058 	info.nr_jited_func_lens = prog->aux->func_cnt;
2059 	if (info.nr_jited_func_lens && ulen) {
2060 		if (bpf_dump_raw_ok()) {
2061 			u32 __user *user_lens;
2062 			u32 func_len, i;
2063 
2064 			/* copy the JITed image lengths for each function */
2065 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2066 			user_lens = u64_to_user_ptr(info.jited_func_lens);
2067 			for (i = 0; i < ulen; i++) {
2068 				func_len = prog->aux->func[i]->jited_len;
2069 				if (put_user(func_len, &user_lens[i]))
2070 					return -EFAULT;
2071 			}
2072 		} else {
2073 			info.jited_func_lens = 0;
2074 		}
2075 	}
2076 
2077 done:
2078 	if (copy_to_user(uinfo, &info, info_len) ||
2079 	    put_user(info_len, &uattr->info.info_len))
2080 		return -EFAULT;
2081 
2082 	return 0;
2083 }
2084 
2085 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2086 				  const union bpf_attr *attr,
2087 				  union bpf_attr __user *uattr)
2088 {
2089 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2090 	struct bpf_map_info info = {};
2091 	u32 info_len = attr->info.info_len;
2092 	int err;
2093 
2094 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2095 	if (err)
2096 		return err;
2097 	info_len = min_t(u32, sizeof(info), info_len);
2098 
2099 	info.type = map->map_type;
2100 	info.id = map->id;
2101 	info.key_size = map->key_size;
2102 	info.value_size = map->value_size;
2103 	info.max_entries = map->max_entries;
2104 	info.map_flags = map->map_flags;
2105 	memcpy(info.name, map->name, sizeof(map->name));
2106 
2107 	if (map->btf) {
2108 		info.btf_id = btf_id(map->btf);
2109 		info.btf_key_type_id = map->btf_key_type_id;
2110 		info.btf_value_type_id = map->btf_value_type_id;
2111 	}
2112 
2113 	if (bpf_map_is_dev_bound(map)) {
2114 		err = bpf_map_offload_info_fill(&info, map);
2115 		if (err)
2116 			return err;
2117 	}
2118 
2119 	if (copy_to_user(uinfo, &info, info_len) ||
2120 	    put_user(info_len, &uattr->info.info_len))
2121 		return -EFAULT;
2122 
2123 	return 0;
2124 }
2125 
2126 static int bpf_btf_get_info_by_fd(struct btf *btf,
2127 				  const union bpf_attr *attr,
2128 				  union bpf_attr __user *uattr)
2129 {
2130 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2131 	u32 info_len = attr->info.info_len;
2132 	int err;
2133 
2134 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2135 	if (err)
2136 		return err;
2137 
2138 	return btf_get_info_by_fd(btf, attr, uattr);
2139 }
2140 
2141 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2142 
2143 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2144 				  union bpf_attr __user *uattr)
2145 {
2146 	int ufd = attr->info.bpf_fd;
2147 	struct fd f;
2148 	int err;
2149 
2150 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2151 		return -EINVAL;
2152 
2153 	f = fdget(ufd);
2154 	if (!f.file)
2155 		return -EBADFD;
2156 
2157 	if (f.file->f_op == &bpf_prog_fops)
2158 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2159 					      uattr);
2160 	else if (f.file->f_op == &bpf_map_fops)
2161 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2162 					     uattr);
2163 	else if (f.file->f_op == &btf_fops)
2164 		err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2165 	else
2166 		err = -EINVAL;
2167 
2168 	fdput(f);
2169 	return err;
2170 }
2171 
2172 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2173 
2174 static int bpf_btf_load(const union bpf_attr *attr)
2175 {
2176 	if (CHECK_ATTR(BPF_BTF_LOAD))
2177 		return -EINVAL;
2178 
2179 	if (!capable(CAP_SYS_ADMIN))
2180 		return -EPERM;
2181 
2182 	return btf_new_fd(attr);
2183 }
2184 
2185 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2186 
2187 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2188 {
2189 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2190 		return -EINVAL;
2191 
2192 	if (!capable(CAP_SYS_ADMIN))
2193 		return -EPERM;
2194 
2195 	return btf_get_fd_by_id(attr->btf_id);
2196 }
2197 
2198 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2199 				    union bpf_attr __user *uattr,
2200 				    u32 prog_id, u32 fd_type,
2201 				    const char *buf, u64 probe_offset,
2202 				    u64 probe_addr)
2203 {
2204 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2205 	u32 len = buf ? strlen(buf) : 0, input_len;
2206 	int err = 0;
2207 
2208 	if (put_user(len, &uattr->task_fd_query.buf_len))
2209 		return -EFAULT;
2210 	input_len = attr->task_fd_query.buf_len;
2211 	if (input_len && ubuf) {
2212 		if (!len) {
2213 			/* nothing to copy, just make ubuf NULL terminated */
2214 			char zero = '\0';
2215 
2216 			if (put_user(zero, ubuf))
2217 				return -EFAULT;
2218 		} else if (input_len >= len + 1) {
2219 			/* ubuf can hold the string with NULL terminator */
2220 			if (copy_to_user(ubuf, buf, len + 1))
2221 				return -EFAULT;
2222 		} else {
2223 			/* ubuf cannot hold the string with NULL terminator,
2224 			 * do a partial copy with NULL terminator.
2225 			 */
2226 			char zero = '\0';
2227 
2228 			err = -ENOSPC;
2229 			if (copy_to_user(ubuf, buf, input_len - 1))
2230 				return -EFAULT;
2231 			if (put_user(zero, ubuf + input_len - 1))
2232 				return -EFAULT;
2233 		}
2234 	}
2235 
2236 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2237 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2238 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2239 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2240 		return -EFAULT;
2241 
2242 	return err;
2243 }
2244 
2245 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2246 
2247 static int bpf_task_fd_query(const union bpf_attr *attr,
2248 			     union bpf_attr __user *uattr)
2249 {
2250 	pid_t pid = attr->task_fd_query.pid;
2251 	u32 fd = attr->task_fd_query.fd;
2252 	const struct perf_event *event;
2253 	struct files_struct *files;
2254 	struct task_struct *task;
2255 	struct file *file;
2256 	int err;
2257 
2258 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2259 		return -EINVAL;
2260 
2261 	if (!capable(CAP_SYS_ADMIN))
2262 		return -EPERM;
2263 
2264 	if (attr->task_fd_query.flags != 0)
2265 		return -EINVAL;
2266 
2267 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2268 	if (!task)
2269 		return -ENOENT;
2270 
2271 	files = get_files_struct(task);
2272 	put_task_struct(task);
2273 	if (!files)
2274 		return -ENOENT;
2275 
2276 	err = 0;
2277 	spin_lock(&files->file_lock);
2278 	file = fcheck_files(files, fd);
2279 	if (!file)
2280 		err = -EBADF;
2281 	else
2282 		get_file(file);
2283 	spin_unlock(&files->file_lock);
2284 	put_files_struct(files);
2285 
2286 	if (err)
2287 		goto out;
2288 
2289 	if (file->f_op == &bpf_raw_tp_fops) {
2290 		struct bpf_raw_tracepoint *raw_tp = file->private_data;
2291 		struct bpf_raw_event_map *btp = raw_tp->btp;
2292 
2293 		err = bpf_task_fd_query_copy(attr, uattr,
2294 					     raw_tp->prog->aux->id,
2295 					     BPF_FD_TYPE_RAW_TRACEPOINT,
2296 					     btp->tp->name, 0, 0);
2297 		goto put_file;
2298 	}
2299 
2300 	event = perf_get_event(file);
2301 	if (!IS_ERR(event)) {
2302 		u64 probe_offset, probe_addr;
2303 		u32 prog_id, fd_type;
2304 		const char *buf;
2305 
2306 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2307 					      &buf, &probe_offset,
2308 					      &probe_addr);
2309 		if (!err)
2310 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2311 						     fd_type, buf,
2312 						     probe_offset,
2313 						     probe_addr);
2314 		goto put_file;
2315 	}
2316 
2317 	err = -ENOTSUPP;
2318 put_file:
2319 	fput(file);
2320 out:
2321 	return err;
2322 }
2323 
2324 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2325 {
2326 	union bpf_attr attr = {};
2327 	int err;
2328 
2329 	if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2330 		return -EPERM;
2331 
2332 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2333 	if (err)
2334 		return err;
2335 	size = min_t(u32, size, sizeof(attr));
2336 
2337 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
2338 	if (copy_from_user(&attr, uattr, size) != 0)
2339 		return -EFAULT;
2340 
2341 	err = security_bpf(cmd, &attr, size);
2342 	if (err < 0)
2343 		return err;
2344 
2345 	switch (cmd) {
2346 	case BPF_MAP_CREATE:
2347 		err = map_create(&attr);
2348 		break;
2349 	case BPF_MAP_LOOKUP_ELEM:
2350 		err = map_lookup_elem(&attr);
2351 		break;
2352 	case BPF_MAP_UPDATE_ELEM:
2353 		err = map_update_elem(&attr);
2354 		break;
2355 	case BPF_MAP_DELETE_ELEM:
2356 		err = map_delete_elem(&attr);
2357 		break;
2358 	case BPF_MAP_GET_NEXT_KEY:
2359 		err = map_get_next_key(&attr);
2360 		break;
2361 	case BPF_PROG_LOAD:
2362 		err = bpf_prog_load(&attr);
2363 		break;
2364 	case BPF_OBJ_PIN:
2365 		err = bpf_obj_pin(&attr);
2366 		break;
2367 	case BPF_OBJ_GET:
2368 		err = bpf_obj_get(&attr);
2369 		break;
2370 #ifdef CONFIG_CGROUP_BPF
2371 	case BPF_PROG_ATTACH:
2372 		err = bpf_prog_attach(&attr);
2373 		break;
2374 	case BPF_PROG_DETACH:
2375 		err = bpf_prog_detach(&attr);
2376 		break;
2377 	case BPF_PROG_QUERY:
2378 		err = bpf_prog_query(&attr, uattr);
2379 		break;
2380 #endif
2381 	case BPF_PROG_TEST_RUN:
2382 		err = bpf_prog_test_run(&attr, uattr);
2383 		break;
2384 	case BPF_PROG_GET_NEXT_ID:
2385 		err = bpf_obj_get_next_id(&attr, uattr,
2386 					  &prog_idr, &prog_idr_lock);
2387 		break;
2388 	case BPF_MAP_GET_NEXT_ID:
2389 		err = bpf_obj_get_next_id(&attr, uattr,
2390 					  &map_idr, &map_idr_lock);
2391 		break;
2392 	case BPF_PROG_GET_FD_BY_ID:
2393 		err = bpf_prog_get_fd_by_id(&attr);
2394 		break;
2395 	case BPF_MAP_GET_FD_BY_ID:
2396 		err = bpf_map_get_fd_by_id(&attr);
2397 		break;
2398 	case BPF_OBJ_GET_INFO_BY_FD:
2399 		err = bpf_obj_get_info_by_fd(&attr, uattr);
2400 		break;
2401 	case BPF_RAW_TRACEPOINT_OPEN:
2402 		err = bpf_raw_tracepoint_open(&attr);
2403 		break;
2404 	case BPF_BTF_LOAD:
2405 		err = bpf_btf_load(&attr);
2406 		break;
2407 	case BPF_BTF_GET_FD_BY_ID:
2408 		err = bpf_btf_get_fd_by_id(&attr);
2409 		break;
2410 	case BPF_TASK_FD_QUERY:
2411 		err = bpf_task_fd_query(&attr, uattr);
2412 		break;
2413 	default:
2414 		err = -EINVAL;
2415 		break;
2416 	}
2417 
2418 	return err;
2419 }
2420