xref: /openbmc/linux/kernel/bpf/syscall.c (revision 09bae3b6)
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 	.map_check_btf = map_check_no_btf,
107 };
108 
109 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
110 {
111 	const struct bpf_map_ops *ops;
112 	u32 type = attr->map_type;
113 	struct bpf_map *map;
114 	int err;
115 
116 	if (type >= ARRAY_SIZE(bpf_map_types))
117 		return ERR_PTR(-EINVAL);
118 	type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
119 	ops = bpf_map_types[type];
120 	if (!ops)
121 		return ERR_PTR(-EINVAL);
122 
123 	if (ops->map_alloc_check) {
124 		err = ops->map_alloc_check(attr);
125 		if (err)
126 			return ERR_PTR(err);
127 	}
128 	if (attr->map_ifindex)
129 		ops = &bpf_map_offload_ops;
130 	map = ops->map_alloc(attr);
131 	if (IS_ERR(map))
132 		return map;
133 	map->ops = ops;
134 	map->map_type = type;
135 	return map;
136 }
137 
138 void *bpf_map_area_alloc(size_t size, int numa_node)
139 {
140 	/* We definitely need __GFP_NORETRY, so OOM killer doesn't
141 	 * trigger under memory pressure as we really just want to
142 	 * fail instead.
143 	 */
144 	const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
145 	void *area;
146 
147 	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
148 		area = kmalloc_node(size, GFP_USER | flags, numa_node);
149 		if (area != NULL)
150 			return area;
151 	}
152 
153 	return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
154 					   __builtin_return_address(0));
155 }
156 
157 void bpf_map_area_free(void *area)
158 {
159 	kvfree(area);
160 }
161 
162 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
163 {
164 	map->map_type = attr->map_type;
165 	map->key_size = attr->key_size;
166 	map->value_size = attr->value_size;
167 	map->max_entries = attr->max_entries;
168 	map->map_flags = attr->map_flags;
169 	map->numa_node = bpf_map_attr_numa_node(attr);
170 }
171 
172 int bpf_map_precharge_memlock(u32 pages)
173 {
174 	struct user_struct *user = get_current_user();
175 	unsigned long memlock_limit, cur;
176 
177 	memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
178 	cur = atomic_long_read(&user->locked_vm);
179 	free_uid(user);
180 	if (cur + pages > memlock_limit)
181 		return -EPERM;
182 	return 0;
183 }
184 
185 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
186 {
187 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
188 
189 	if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
190 		atomic_long_sub(pages, &user->locked_vm);
191 		return -EPERM;
192 	}
193 	return 0;
194 }
195 
196 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
197 {
198 	atomic_long_sub(pages, &user->locked_vm);
199 }
200 
201 static int bpf_map_init_memlock(struct bpf_map *map)
202 {
203 	struct user_struct *user = get_current_user();
204 	int ret;
205 
206 	ret = bpf_charge_memlock(user, map->pages);
207 	if (ret) {
208 		free_uid(user);
209 		return ret;
210 	}
211 	map->user = user;
212 	return ret;
213 }
214 
215 static void bpf_map_release_memlock(struct bpf_map *map)
216 {
217 	struct user_struct *user = map->user;
218 	bpf_uncharge_memlock(user, map->pages);
219 	free_uid(user);
220 }
221 
222 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
223 {
224 	int ret;
225 
226 	ret = bpf_charge_memlock(map->user, pages);
227 	if (ret)
228 		return ret;
229 	map->pages += pages;
230 	return ret;
231 }
232 
233 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
234 {
235 	bpf_uncharge_memlock(map->user, pages);
236 	map->pages -= pages;
237 }
238 
239 static int bpf_map_alloc_id(struct bpf_map *map)
240 {
241 	int id;
242 
243 	idr_preload(GFP_KERNEL);
244 	spin_lock_bh(&map_idr_lock);
245 	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
246 	if (id > 0)
247 		map->id = id;
248 	spin_unlock_bh(&map_idr_lock);
249 	idr_preload_end();
250 
251 	if (WARN_ON_ONCE(!id))
252 		return -ENOSPC;
253 
254 	return id > 0 ? 0 : id;
255 }
256 
257 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
258 {
259 	unsigned long flags;
260 
261 	/* Offloaded maps are removed from the IDR store when their device
262 	 * disappears - even if someone holds an fd to them they are unusable,
263 	 * the memory is gone, all ops will fail; they are simply waiting for
264 	 * refcnt to drop to be freed.
265 	 */
266 	if (!map->id)
267 		return;
268 
269 	if (do_idr_lock)
270 		spin_lock_irqsave(&map_idr_lock, flags);
271 	else
272 		__acquire(&map_idr_lock);
273 
274 	idr_remove(&map_idr, map->id);
275 	map->id = 0;
276 
277 	if (do_idr_lock)
278 		spin_unlock_irqrestore(&map_idr_lock, flags);
279 	else
280 		__release(&map_idr_lock);
281 }
282 
283 /* called from workqueue */
284 static void bpf_map_free_deferred(struct work_struct *work)
285 {
286 	struct bpf_map *map = container_of(work, struct bpf_map, work);
287 
288 	bpf_map_release_memlock(map);
289 	security_bpf_map_free(map);
290 	/* implementation dependent freeing */
291 	map->ops->map_free(map);
292 }
293 
294 static void bpf_map_put_uref(struct bpf_map *map)
295 {
296 	if (atomic_dec_and_test(&map->usercnt)) {
297 		if (map->ops->map_release_uref)
298 			map->ops->map_release_uref(map);
299 	}
300 }
301 
302 /* decrement map refcnt and schedule it for freeing via workqueue
303  * (unrelying map implementation ops->map_free() might sleep)
304  */
305 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
306 {
307 	if (atomic_dec_and_test(&map->refcnt)) {
308 		/* bpf_map_free_id() must be called first */
309 		bpf_map_free_id(map, do_idr_lock);
310 		btf_put(map->btf);
311 		INIT_WORK(&map->work, bpf_map_free_deferred);
312 		schedule_work(&map->work);
313 	}
314 }
315 
316 void bpf_map_put(struct bpf_map *map)
317 {
318 	__bpf_map_put(map, true);
319 }
320 EXPORT_SYMBOL_GPL(bpf_map_put);
321 
322 void bpf_map_put_with_uref(struct bpf_map *map)
323 {
324 	bpf_map_put_uref(map);
325 	bpf_map_put(map);
326 }
327 
328 static int bpf_map_release(struct inode *inode, struct file *filp)
329 {
330 	struct bpf_map *map = filp->private_data;
331 
332 	if (map->ops->map_release)
333 		map->ops->map_release(map, filp);
334 
335 	bpf_map_put_with_uref(map);
336 	return 0;
337 }
338 
339 #ifdef CONFIG_PROC_FS
340 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
341 {
342 	const struct bpf_map *map = filp->private_data;
343 	const struct bpf_array *array;
344 	u32 owner_prog_type = 0;
345 	u32 owner_jited = 0;
346 
347 	if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
348 		array = container_of(map, struct bpf_array, map);
349 		owner_prog_type = array->owner_prog_type;
350 		owner_jited = array->owner_jited;
351 	}
352 
353 	seq_printf(m,
354 		   "map_type:\t%u\n"
355 		   "key_size:\t%u\n"
356 		   "value_size:\t%u\n"
357 		   "max_entries:\t%u\n"
358 		   "map_flags:\t%#x\n"
359 		   "memlock:\t%llu\n"
360 		   "map_id:\t%u\n",
361 		   map->map_type,
362 		   map->key_size,
363 		   map->value_size,
364 		   map->max_entries,
365 		   map->map_flags,
366 		   map->pages * 1ULL << PAGE_SHIFT,
367 		   map->id);
368 
369 	if (owner_prog_type) {
370 		seq_printf(m, "owner_prog_type:\t%u\n",
371 			   owner_prog_type);
372 		seq_printf(m, "owner_jited:\t%u\n",
373 			   owner_jited);
374 	}
375 }
376 #endif
377 
378 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
379 			      loff_t *ppos)
380 {
381 	/* We need this handler such that alloc_file() enables
382 	 * f_mode with FMODE_CAN_READ.
383 	 */
384 	return -EINVAL;
385 }
386 
387 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
388 			       size_t siz, loff_t *ppos)
389 {
390 	/* We need this handler such that alloc_file() enables
391 	 * f_mode with FMODE_CAN_WRITE.
392 	 */
393 	return -EINVAL;
394 }
395 
396 const struct file_operations bpf_map_fops = {
397 #ifdef CONFIG_PROC_FS
398 	.show_fdinfo	= bpf_map_show_fdinfo,
399 #endif
400 	.release	= bpf_map_release,
401 	.read		= bpf_dummy_read,
402 	.write		= bpf_dummy_write,
403 };
404 
405 int bpf_map_new_fd(struct bpf_map *map, int flags)
406 {
407 	int ret;
408 
409 	ret = security_bpf_map(map, OPEN_FMODE(flags));
410 	if (ret < 0)
411 		return ret;
412 
413 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
414 				flags | O_CLOEXEC);
415 }
416 
417 int bpf_get_file_flag(int flags)
418 {
419 	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
420 		return -EINVAL;
421 	if (flags & BPF_F_RDONLY)
422 		return O_RDONLY;
423 	if (flags & BPF_F_WRONLY)
424 		return O_WRONLY;
425 	return O_RDWR;
426 }
427 
428 /* helper macro to check that unused fields 'union bpf_attr' are zero */
429 #define CHECK_ATTR(CMD) \
430 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
431 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
432 		   sizeof(*attr) - \
433 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
434 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
435 
436 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
437  * Return 0 on success and < 0 on error.
438  */
439 static int bpf_obj_name_cpy(char *dst, const char *src)
440 {
441 	const char *end = src + BPF_OBJ_NAME_LEN;
442 
443 	memset(dst, 0, BPF_OBJ_NAME_LEN);
444 
445 	/* Copy all isalnum() and '_' char */
446 	while (src < end && *src) {
447 		if (!isalnum(*src) && *src != '_')
448 			return -EINVAL;
449 		*dst++ = *src++;
450 	}
451 
452 	/* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
453 	if (src == end)
454 		return -EINVAL;
455 
456 	return 0;
457 }
458 
459 int map_check_no_btf(const struct bpf_map *map,
460 		     const struct btf_type *key_type,
461 		     const struct btf_type *value_type)
462 {
463 	return -ENOTSUPP;
464 }
465 
466 static int map_check_btf(const struct bpf_map *map, const struct btf *btf,
467 			 u32 btf_key_id, u32 btf_value_id)
468 {
469 	const struct btf_type *key_type, *value_type;
470 	u32 key_size, value_size;
471 	int ret = 0;
472 
473 	key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
474 	if (!key_type || key_size != map->key_size)
475 		return -EINVAL;
476 
477 	value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
478 	if (!value_type || value_size != map->value_size)
479 		return -EINVAL;
480 
481 	if (map->ops->map_check_btf)
482 		ret = map->ops->map_check_btf(map, key_type, value_type);
483 
484 	return ret;
485 }
486 
487 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
488 /* called via syscall */
489 static int map_create(union bpf_attr *attr)
490 {
491 	int numa_node = bpf_map_attr_numa_node(attr);
492 	struct bpf_map *map;
493 	int f_flags;
494 	int err;
495 
496 	err = CHECK_ATTR(BPF_MAP_CREATE);
497 	if (err)
498 		return -EINVAL;
499 
500 	f_flags = bpf_get_file_flag(attr->map_flags);
501 	if (f_flags < 0)
502 		return f_flags;
503 
504 	if (numa_node != NUMA_NO_NODE &&
505 	    ((unsigned int)numa_node >= nr_node_ids ||
506 	     !node_online(numa_node)))
507 		return -EINVAL;
508 
509 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
510 	map = find_and_alloc_map(attr);
511 	if (IS_ERR(map))
512 		return PTR_ERR(map);
513 
514 	err = bpf_obj_name_cpy(map->name, attr->map_name);
515 	if (err)
516 		goto free_map_nouncharge;
517 
518 	atomic_set(&map->refcnt, 1);
519 	atomic_set(&map->usercnt, 1);
520 
521 	if (attr->btf_key_type_id || attr->btf_value_type_id) {
522 		struct btf *btf;
523 
524 		if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
525 			err = -EINVAL;
526 			goto free_map_nouncharge;
527 		}
528 
529 		btf = btf_get_by_fd(attr->btf_fd);
530 		if (IS_ERR(btf)) {
531 			err = PTR_ERR(btf);
532 			goto free_map_nouncharge;
533 		}
534 
535 		err = map_check_btf(map, btf, attr->btf_key_type_id,
536 				    attr->btf_value_type_id);
537 		if (err) {
538 			btf_put(btf);
539 			goto free_map_nouncharge;
540 		}
541 
542 		map->btf = btf;
543 		map->btf_key_type_id = attr->btf_key_type_id;
544 		map->btf_value_type_id = attr->btf_value_type_id;
545 	}
546 
547 	err = security_bpf_map_alloc(map);
548 	if (err)
549 		goto free_map_nouncharge;
550 
551 	err = bpf_map_init_memlock(map);
552 	if (err)
553 		goto free_map_sec;
554 
555 	err = bpf_map_alloc_id(map);
556 	if (err)
557 		goto free_map;
558 
559 	err = bpf_map_new_fd(map, f_flags);
560 	if (err < 0) {
561 		/* failed to allocate fd.
562 		 * bpf_map_put() is needed because the above
563 		 * bpf_map_alloc_id() has published the map
564 		 * to the userspace and the userspace may
565 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
566 		 */
567 		bpf_map_put(map);
568 		return err;
569 	}
570 
571 	return err;
572 
573 free_map:
574 	bpf_map_release_memlock(map);
575 free_map_sec:
576 	security_bpf_map_free(map);
577 free_map_nouncharge:
578 	btf_put(map->btf);
579 	map->ops->map_free(map);
580 	return err;
581 }
582 
583 /* if error is returned, fd is released.
584  * On success caller should complete fd access with matching fdput()
585  */
586 struct bpf_map *__bpf_map_get(struct fd f)
587 {
588 	if (!f.file)
589 		return ERR_PTR(-EBADF);
590 	if (f.file->f_op != &bpf_map_fops) {
591 		fdput(f);
592 		return ERR_PTR(-EINVAL);
593 	}
594 
595 	return f.file->private_data;
596 }
597 
598 /* prog's and map's refcnt limit */
599 #define BPF_MAX_REFCNT 32768
600 
601 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
602 {
603 	if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
604 		atomic_dec(&map->refcnt);
605 		return ERR_PTR(-EBUSY);
606 	}
607 	if (uref)
608 		atomic_inc(&map->usercnt);
609 	return map;
610 }
611 EXPORT_SYMBOL_GPL(bpf_map_inc);
612 
613 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
614 {
615 	struct fd f = fdget(ufd);
616 	struct bpf_map *map;
617 
618 	map = __bpf_map_get(f);
619 	if (IS_ERR(map))
620 		return map;
621 
622 	map = bpf_map_inc(map, true);
623 	fdput(f);
624 
625 	return map;
626 }
627 
628 /* map_idr_lock should have been held */
629 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
630 					    bool uref)
631 {
632 	int refold;
633 
634 	refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
635 
636 	if (refold >= BPF_MAX_REFCNT) {
637 		__bpf_map_put(map, false);
638 		return ERR_PTR(-EBUSY);
639 	}
640 
641 	if (!refold)
642 		return ERR_PTR(-ENOENT);
643 
644 	if (uref)
645 		atomic_inc(&map->usercnt);
646 
647 	return map;
648 }
649 
650 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
651 {
652 	return -ENOTSUPP;
653 }
654 
655 /* last field in 'union bpf_attr' used by this command */
656 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
657 
658 static int map_lookup_elem(union bpf_attr *attr)
659 {
660 	void __user *ukey = u64_to_user_ptr(attr->key);
661 	void __user *uvalue = u64_to_user_ptr(attr->value);
662 	int ufd = attr->map_fd;
663 	struct bpf_map *map;
664 	void *key, *value, *ptr;
665 	u32 value_size;
666 	struct fd f;
667 	int err;
668 
669 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
670 		return -EINVAL;
671 
672 	f = fdget(ufd);
673 	map = __bpf_map_get(f);
674 	if (IS_ERR(map))
675 		return PTR_ERR(map);
676 
677 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
678 		err = -EPERM;
679 		goto err_put;
680 	}
681 
682 	key = memdup_user(ukey, map->key_size);
683 	if (IS_ERR(key)) {
684 		err = PTR_ERR(key);
685 		goto err_put;
686 	}
687 
688 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
689 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
690 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
691 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
692 	else if (IS_FD_MAP(map))
693 		value_size = sizeof(u32);
694 	else
695 		value_size = map->value_size;
696 
697 	err = -ENOMEM;
698 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
699 	if (!value)
700 		goto free_key;
701 
702 	if (bpf_map_is_dev_bound(map)) {
703 		err = bpf_map_offload_lookup_elem(map, key, value);
704 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
705 		   map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
706 		err = bpf_percpu_hash_copy(map, key, value);
707 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
708 		err = bpf_percpu_array_copy(map, key, value);
709 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
710 		err = bpf_stackmap_copy(map, key, value);
711 	} else if (IS_FD_ARRAY(map)) {
712 		err = bpf_fd_array_map_lookup_elem(map, key, value);
713 	} else if (IS_FD_HASH(map)) {
714 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
715 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
716 		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
717 	} else {
718 		rcu_read_lock();
719 		ptr = map->ops->map_lookup_elem(map, key);
720 		if (ptr)
721 			memcpy(value, ptr, value_size);
722 		rcu_read_unlock();
723 		err = ptr ? 0 : -ENOENT;
724 	}
725 
726 	if (err)
727 		goto free_value;
728 
729 	err = -EFAULT;
730 	if (copy_to_user(uvalue, value, value_size) != 0)
731 		goto free_value;
732 
733 	err = 0;
734 
735 free_value:
736 	kfree(value);
737 free_key:
738 	kfree(key);
739 err_put:
740 	fdput(f);
741 	return err;
742 }
743 
744 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
745 
746 static int map_update_elem(union bpf_attr *attr)
747 {
748 	void __user *ukey = u64_to_user_ptr(attr->key);
749 	void __user *uvalue = u64_to_user_ptr(attr->value);
750 	int ufd = attr->map_fd;
751 	struct bpf_map *map;
752 	void *key, *value;
753 	u32 value_size;
754 	struct fd f;
755 	int err;
756 
757 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
758 		return -EINVAL;
759 
760 	f = fdget(ufd);
761 	map = __bpf_map_get(f);
762 	if (IS_ERR(map))
763 		return PTR_ERR(map);
764 
765 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
766 		err = -EPERM;
767 		goto err_put;
768 	}
769 
770 	key = memdup_user(ukey, map->key_size);
771 	if (IS_ERR(key)) {
772 		err = PTR_ERR(key);
773 		goto err_put;
774 	}
775 
776 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
777 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
778 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
779 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
780 	else
781 		value_size = map->value_size;
782 
783 	err = -ENOMEM;
784 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
785 	if (!value)
786 		goto free_key;
787 
788 	err = -EFAULT;
789 	if (copy_from_user(value, uvalue, value_size) != 0)
790 		goto free_value;
791 
792 	/* Need to create a kthread, thus must support schedule */
793 	if (bpf_map_is_dev_bound(map)) {
794 		err = bpf_map_offload_update_elem(map, key, value, attr->flags);
795 		goto out;
796 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
797 		   map->map_type == BPF_MAP_TYPE_SOCKHASH ||
798 		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
799 		err = map->ops->map_update_elem(map, key, value, attr->flags);
800 		goto out;
801 	}
802 
803 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
804 	 * inside bpf map update or delete otherwise deadlocks are possible
805 	 */
806 	preempt_disable();
807 	__this_cpu_inc(bpf_prog_active);
808 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
809 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
810 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
811 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
812 		err = bpf_percpu_array_update(map, key, value, attr->flags);
813 	} else if (IS_FD_ARRAY(map)) {
814 		rcu_read_lock();
815 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
816 						   attr->flags);
817 		rcu_read_unlock();
818 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
819 		rcu_read_lock();
820 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
821 						  attr->flags);
822 		rcu_read_unlock();
823 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
824 		/* rcu_read_lock() is not needed */
825 		err = bpf_fd_reuseport_array_update_elem(map, key, value,
826 							 attr->flags);
827 	} else {
828 		rcu_read_lock();
829 		err = map->ops->map_update_elem(map, key, value, attr->flags);
830 		rcu_read_unlock();
831 	}
832 	__this_cpu_dec(bpf_prog_active);
833 	preempt_enable();
834 out:
835 free_value:
836 	kfree(value);
837 free_key:
838 	kfree(key);
839 err_put:
840 	fdput(f);
841 	return err;
842 }
843 
844 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
845 
846 static int map_delete_elem(union bpf_attr *attr)
847 {
848 	void __user *ukey = u64_to_user_ptr(attr->key);
849 	int ufd = attr->map_fd;
850 	struct bpf_map *map;
851 	struct fd f;
852 	void *key;
853 	int err;
854 
855 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
856 		return -EINVAL;
857 
858 	f = fdget(ufd);
859 	map = __bpf_map_get(f);
860 	if (IS_ERR(map))
861 		return PTR_ERR(map);
862 
863 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
864 		err = -EPERM;
865 		goto err_put;
866 	}
867 
868 	key = memdup_user(ukey, map->key_size);
869 	if (IS_ERR(key)) {
870 		err = PTR_ERR(key);
871 		goto err_put;
872 	}
873 
874 	if (bpf_map_is_dev_bound(map)) {
875 		err = bpf_map_offload_delete_elem(map, key);
876 		goto out;
877 	}
878 
879 	preempt_disable();
880 	__this_cpu_inc(bpf_prog_active);
881 	rcu_read_lock();
882 	err = map->ops->map_delete_elem(map, key);
883 	rcu_read_unlock();
884 	__this_cpu_dec(bpf_prog_active);
885 	preempt_enable();
886 out:
887 	kfree(key);
888 err_put:
889 	fdput(f);
890 	return err;
891 }
892 
893 /* last field in 'union bpf_attr' used by this command */
894 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
895 
896 static int map_get_next_key(union bpf_attr *attr)
897 {
898 	void __user *ukey = u64_to_user_ptr(attr->key);
899 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
900 	int ufd = attr->map_fd;
901 	struct bpf_map *map;
902 	void *key, *next_key;
903 	struct fd f;
904 	int err;
905 
906 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
907 		return -EINVAL;
908 
909 	f = fdget(ufd);
910 	map = __bpf_map_get(f);
911 	if (IS_ERR(map))
912 		return PTR_ERR(map);
913 
914 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
915 		err = -EPERM;
916 		goto err_put;
917 	}
918 
919 	if (ukey) {
920 		key = memdup_user(ukey, map->key_size);
921 		if (IS_ERR(key)) {
922 			err = PTR_ERR(key);
923 			goto err_put;
924 		}
925 	} else {
926 		key = NULL;
927 	}
928 
929 	err = -ENOMEM;
930 	next_key = kmalloc(map->key_size, GFP_USER);
931 	if (!next_key)
932 		goto free_key;
933 
934 	if (bpf_map_is_dev_bound(map)) {
935 		err = bpf_map_offload_get_next_key(map, key, next_key);
936 		goto out;
937 	}
938 
939 	rcu_read_lock();
940 	err = map->ops->map_get_next_key(map, key, next_key);
941 	rcu_read_unlock();
942 out:
943 	if (err)
944 		goto free_next_key;
945 
946 	err = -EFAULT;
947 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
948 		goto free_next_key;
949 
950 	err = 0;
951 
952 free_next_key:
953 	kfree(next_key);
954 free_key:
955 	kfree(key);
956 err_put:
957 	fdput(f);
958 	return err;
959 }
960 
961 static const struct bpf_prog_ops * const bpf_prog_types[] = {
962 #define BPF_PROG_TYPE(_id, _name) \
963 	[_id] = & _name ## _prog_ops,
964 #define BPF_MAP_TYPE(_id, _ops)
965 #include <linux/bpf_types.h>
966 #undef BPF_PROG_TYPE
967 #undef BPF_MAP_TYPE
968 };
969 
970 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
971 {
972 	const struct bpf_prog_ops *ops;
973 
974 	if (type >= ARRAY_SIZE(bpf_prog_types))
975 		return -EINVAL;
976 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
977 	ops = bpf_prog_types[type];
978 	if (!ops)
979 		return -EINVAL;
980 
981 	if (!bpf_prog_is_dev_bound(prog->aux))
982 		prog->aux->ops = ops;
983 	else
984 		prog->aux->ops = &bpf_offload_prog_ops;
985 	prog->type = type;
986 	return 0;
987 }
988 
989 /* drop refcnt on maps used by eBPF program and free auxilary data */
990 static void free_used_maps(struct bpf_prog_aux *aux)
991 {
992 	int i;
993 
994 	if (aux->cgroup_storage)
995 		bpf_cgroup_storage_release(aux->prog, aux->cgroup_storage);
996 
997 	for (i = 0; i < aux->used_map_cnt; i++)
998 		bpf_map_put(aux->used_maps[i]);
999 
1000 	kfree(aux->used_maps);
1001 }
1002 
1003 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1004 {
1005 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1006 	unsigned long user_bufs;
1007 
1008 	if (user) {
1009 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1010 		if (user_bufs > memlock_limit) {
1011 			atomic_long_sub(pages, &user->locked_vm);
1012 			return -EPERM;
1013 		}
1014 	}
1015 
1016 	return 0;
1017 }
1018 
1019 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1020 {
1021 	if (user)
1022 		atomic_long_sub(pages, &user->locked_vm);
1023 }
1024 
1025 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1026 {
1027 	struct user_struct *user = get_current_user();
1028 	int ret;
1029 
1030 	ret = __bpf_prog_charge(user, prog->pages);
1031 	if (ret) {
1032 		free_uid(user);
1033 		return ret;
1034 	}
1035 
1036 	prog->aux->user = user;
1037 	return 0;
1038 }
1039 
1040 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1041 {
1042 	struct user_struct *user = prog->aux->user;
1043 
1044 	__bpf_prog_uncharge(user, prog->pages);
1045 	free_uid(user);
1046 }
1047 
1048 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1049 {
1050 	int id;
1051 
1052 	idr_preload(GFP_KERNEL);
1053 	spin_lock_bh(&prog_idr_lock);
1054 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1055 	if (id > 0)
1056 		prog->aux->id = id;
1057 	spin_unlock_bh(&prog_idr_lock);
1058 	idr_preload_end();
1059 
1060 	/* id is in [1, INT_MAX) */
1061 	if (WARN_ON_ONCE(!id))
1062 		return -ENOSPC;
1063 
1064 	return id > 0 ? 0 : id;
1065 }
1066 
1067 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1068 {
1069 	/* cBPF to eBPF migrations are currently not in the idr store.
1070 	 * Offloaded programs are removed from the store when their device
1071 	 * disappears - even if someone grabs an fd to them they are unusable,
1072 	 * simply waiting for refcnt to drop to be freed.
1073 	 */
1074 	if (!prog->aux->id)
1075 		return;
1076 
1077 	if (do_idr_lock)
1078 		spin_lock_bh(&prog_idr_lock);
1079 	else
1080 		__acquire(&prog_idr_lock);
1081 
1082 	idr_remove(&prog_idr, prog->aux->id);
1083 	prog->aux->id = 0;
1084 
1085 	if (do_idr_lock)
1086 		spin_unlock_bh(&prog_idr_lock);
1087 	else
1088 		__release(&prog_idr_lock);
1089 }
1090 
1091 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1092 {
1093 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1094 
1095 	free_used_maps(aux);
1096 	bpf_prog_uncharge_memlock(aux->prog);
1097 	security_bpf_prog_free(aux);
1098 	bpf_prog_free(aux->prog);
1099 }
1100 
1101 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1102 {
1103 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
1104 		/* bpf_prog_free_id() must be called first */
1105 		bpf_prog_free_id(prog, do_idr_lock);
1106 		bpf_prog_kallsyms_del_all(prog);
1107 
1108 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1109 	}
1110 }
1111 
1112 void bpf_prog_put(struct bpf_prog *prog)
1113 {
1114 	__bpf_prog_put(prog, true);
1115 }
1116 EXPORT_SYMBOL_GPL(bpf_prog_put);
1117 
1118 static int bpf_prog_release(struct inode *inode, struct file *filp)
1119 {
1120 	struct bpf_prog *prog = filp->private_data;
1121 
1122 	bpf_prog_put(prog);
1123 	return 0;
1124 }
1125 
1126 #ifdef CONFIG_PROC_FS
1127 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1128 {
1129 	const struct bpf_prog *prog = filp->private_data;
1130 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1131 
1132 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1133 	seq_printf(m,
1134 		   "prog_type:\t%u\n"
1135 		   "prog_jited:\t%u\n"
1136 		   "prog_tag:\t%s\n"
1137 		   "memlock:\t%llu\n"
1138 		   "prog_id:\t%u\n",
1139 		   prog->type,
1140 		   prog->jited,
1141 		   prog_tag,
1142 		   prog->pages * 1ULL << PAGE_SHIFT,
1143 		   prog->aux->id);
1144 }
1145 #endif
1146 
1147 const struct file_operations bpf_prog_fops = {
1148 #ifdef CONFIG_PROC_FS
1149 	.show_fdinfo	= bpf_prog_show_fdinfo,
1150 #endif
1151 	.release	= bpf_prog_release,
1152 	.read		= bpf_dummy_read,
1153 	.write		= bpf_dummy_write,
1154 };
1155 
1156 int bpf_prog_new_fd(struct bpf_prog *prog)
1157 {
1158 	int ret;
1159 
1160 	ret = security_bpf_prog(prog);
1161 	if (ret < 0)
1162 		return ret;
1163 
1164 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1165 				O_RDWR | O_CLOEXEC);
1166 }
1167 
1168 static struct bpf_prog *____bpf_prog_get(struct fd f)
1169 {
1170 	if (!f.file)
1171 		return ERR_PTR(-EBADF);
1172 	if (f.file->f_op != &bpf_prog_fops) {
1173 		fdput(f);
1174 		return ERR_PTR(-EINVAL);
1175 	}
1176 
1177 	return f.file->private_data;
1178 }
1179 
1180 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1181 {
1182 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1183 		atomic_sub(i, &prog->aux->refcnt);
1184 		return ERR_PTR(-EBUSY);
1185 	}
1186 	return prog;
1187 }
1188 EXPORT_SYMBOL_GPL(bpf_prog_add);
1189 
1190 void bpf_prog_sub(struct bpf_prog *prog, int i)
1191 {
1192 	/* Only to be used for undoing previous bpf_prog_add() in some
1193 	 * error path. We still know that another entity in our call
1194 	 * path holds a reference to the program, thus atomic_sub() can
1195 	 * be safely used in such cases!
1196 	 */
1197 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1198 }
1199 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1200 
1201 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1202 {
1203 	return bpf_prog_add(prog, 1);
1204 }
1205 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1206 
1207 /* prog_idr_lock should have been held */
1208 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1209 {
1210 	int refold;
1211 
1212 	refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1213 
1214 	if (refold >= BPF_MAX_REFCNT) {
1215 		__bpf_prog_put(prog, false);
1216 		return ERR_PTR(-EBUSY);
1217 	}
1218 
1219 	if (!refold)
1220 		return ERR_PTR(-ENOENT);
1221 
1222 	return prog;
1223 }
1224 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1225 
1226 bool bpf_prog_get_ok(struct bpf_prog *prog,
1227 			    enum bpf_prog_type *attach_type, bool attach_drv)
1228 {
1229 	/* not an attachment, just a refcount inc, always allow */
1230 	if (!attach_type)
1231 		return true;
1232 
1233 	if (prog->type != *attach_type)
1234 		return false;
1235 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1236 		return false;
1237 
1238 	return true;
1239 }
1240 
1241 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1242 				       bool attach_drv)
1243 {
1244 	struct fd f = fdget(ufd);
1245 	struct bpf_prog *prog;
1246 
1247 	prog = ____bpf_prog_get(f);
1248 	if (IS_ERR(prog))
1249 		return prog;
1250 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1251 		prog = ERR_PTR(-EINVAL);
1252 		goto out;
1253 	}
1254 
1255 	prog = bpf_prog_inc(prog);
1256 out:
1257 	fdput(f);
1258 	return prog;
1259 }
1260 
1261 struct bpf_prog *bpf_prog_get(u32 ufd)
1262 {
1263 	return __bpf_prog_get(ufd, NULL, false);
1264 }
1265 
1266 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1267 				       bool attach_drv)
1268 {
1269 	return __bpf_prog_get(ufd, &type, attach_drv);
1270 }
1271 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1272 
1273 /* Initially all BPF programs could be loaded w/o specifying
1274  * expected_attach_type. Later for some of them specifying expected_attach_type
1275  * at load time became required so that program could be validated properly.
1276  * Programs of types that are allowed to be loaded both w/ and w/o (for
1277  * backward compatibility) expected_attach_type, should have the default attach
1278  * type assigned to expected_attach_type for the latter case, so that it can be
1279  * validated later at attach time.
1280  *
1281  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1282  * prog type requires it but has some attach types that have to be backward
1283  * compatible.
1284  */
1285 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1286 {
1287 	switch (attr->prog_type) {
1288 	case BPF_PROG_TYPE_CGROUP_SOCK:
1289 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1290 		 * exist so checking for non-zero is the way to go here.
1291 		 */
1292 		if (!attr->expected_attach_type)
1293 			attr->expected_attach_type =
1294 				BPF_CGROUP_INET_SOCK_CREATE;
1295 		break;
1296 	}
1297 }
1298 
1299 static int
1300 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1301 				enum bpf_attach_type expected_attach_type)
1302 {
1303 	switch (prog_type) {
1304 	case BPF_PROG_TYPE_CGROUP_SOCK:
1305 		switch (expected_attach_type) {
1306 		case BPF_CGROUP_INET_SOCK_CREATE:
1307 		case BPF_CGROUP_INET4_POST_BIND:
1308 		case BPF_CGROUP_INET6_POST_BIND:
1309 			return 0;
1310 		default:
1311 			return -EINVAL;
1312 		}
1313 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1314 		switch (expected_attach_type) {
1315 		case BPF_CGROUP_INET4_BIND:
1316 		case BPF_CGROUP_INET6_BIND:
1317 		case BPF_CGROUP_INET4_CONNECT:
1318 		case BPF_CGROUP_INET6_CONNECT:
1319 		case BPF_CGROUP_UDP4_SENDMSG:
1320 		case BPF_CGROUP_UDP6_SENDMSG:
1321 			return 0;
1322 		default:
1323 			return -EINVAL;
1324 		}
1325 	default:
1326 		return 0;
1327 	}
1328 }
1329 
1330 /* last field in 'union bpf_attr' used by this command */
1331 #define	BPF_PROG_LOAD_LAST_FIELD expected_attach_type
1332 
1333 static int bpf_prog_load(union bpf_attr *attr)
1334 {
1335 	enum bpf_prog_type type = attr->prog_type;
1336 	struct bpf_prog *prog;
1337 	int err;
1338 	char license[128];
1339 	bool is_gpl;
1340 
1341 	if (CHECK_ATTR(BPF_PROG_LOAD))
1342 		return -EINVAL;
1343 
1344 	if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1345 		return -EINVAL;
1346 
1347 	/* copy eBPF program license from user space */
1348 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1349 			      sizeof(license) - 1) < 0)
1350 		return -EFAULT;
1351 	license[sizeof(license) - 1] = 0;
1352 
1353 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1354 	is_gpl = license_is_gpl_compatible(license);
1355 
1356 	if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1357 		return -E2BIG;
1358 
1359 	if (type == BPF_PROG_TYPE_KPROBE &&
1360 	    attr->kern_version != LINUX_VERSION_CODE)
1361 		return -EINVAL;
1362 
1363 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1364 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
1365 	    !capable(CAP_SYS_ADMIN))
1366 		return -EPERM;
1367 
1368 	bpf_prog_load_fixup_attach_type(attr);
1369 	if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1370 		return -EINVAL;
1371 
1372 	/* plain bpf_prog allocation */
1373 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1374 	if (!prog)
1375 		return -ENOMEM;
1376 
1377 	prog->expected_attach_type = attr->expected_attach_type;
1378 
1379 	prog->aux->offload_requested = !!attr->prog_ifindex;
1380 
1381 	err = security_bpf_prog_alloc(prog->aux);
1382 	if (err)
1383 		goto free_prog_nouncharge;
1384 
1385 	err = bpf_prog_charge_memlock(prog);
1386 	if (err)
1387 		goto free_prog_sec;
1388 
1389 	prog->len = attr->insn_cnt;
1390 
1391 	err = -EFAULT;
1392 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1393 			   bpf_prog_insn_size(prog)) != 0)
1394 		goto free_prog;
1395 
1396 	prog->orig_prog = NULL;
1397 	prog->jited = 0;
1398 
1399 	atomic_set(&prog->aux->refcnt, 1);
1400 	prog->gpl_compatible = is_gpl ? 1 : 0;
1401 
1402 	if (bpf_prog_is_dev_bound(prog->aux)) {
1403 		err = bpf_prog_offload_init(prog, attr);
1404 		if (err)
1405 			goto free_prog;
1406 	}
1407 
1408 	/* find program type: socket_filter vs tracing_filter */
1409 	err = find_prog_type(type, prog);
1410 	if (err < 0)
1411 		goto free_prog;
1412 
1413 	prog->aux->load_time = ktime_get_boot_ns();
1414 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1415 	if (err)
1416 		goto free_prog;
1417 
1418 	/* run eBPF verifier */
1419 	err = bpf_check(&prog, attr);
1420 	if (err < 0)
1421 		goto free_used_maps;
1422 
1423 	prog = bpf_prog_select_runtime(prog, &err);
1424 	if (err < 0)
1425 		goto free_used_maps;
1426 
1427 	err = bpf_prog_alloc_id(prog);
1428 	if (err)
1429 		goto free_used_maps;
1430 
1431 	err = bpf_prog_new_fd(prog);
1432 	if (err < 0) {
1433 		/* failed to allocate fd.
1434 		 * bpf_prog_put() is needed because the above
1435 		 * bpf_prog_alloc_id() has published the prog
1436 		 * to the userspace and the userspace may
1437 		 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1438 		 */
1439 		bpf_prog_put(prog);
1440 		return err;
1441 	}
1442 
1443 	bpf_prog_kallsyms_add(prog);
1444 	return err;
1445 
1446 free_used_maps:
1447 	bpf_prog_kallsyms_del_subprogs(prog);
1448 	free_used_maps(prog->aux);
1449 free_prog:
1450 	bpf_prog_uncharge_memlock(prog);
1451 free_prog_sec:
1452 	security_bpf_prog_free(prog->aux);
1453 free_prog_nouncharge:
1454 	bpf_prog_free(prog);
1455 	return err;
1456 }
1457 
1458 #define BPF_OBJ_LAST_FIELD file_flags
1459 
1460 static int bpf_obj_pin(const union bpf_attr *attr)
1461 {
1462 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1463 		return -EINVAL;
1464 
1465 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1466 }
1467 
1468 static int bpf_obj_get(const union bpf_attr *attr)
1469 {
1470 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1471 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1472 		return -EINVAL;
1473 
1474 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1475 				attr->file_flags);
1476 }
1477 
1478 struct bpf_raw_tracepoint {
1479 	struct bpf_raw_event_map *btp;
1480 	struct bpf_prog *prog;
1481 };
1482 
1483 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1484 {
1485 	struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1486 
1487 	if (raw_tp->prog) {
1488 		bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1489 		bpf_prog_put(raw_tp->prog);
1490 	}
1491 	kfree(raw_tp);
1492 	return 0;
1493 }
1494 
1495 static const struct file_operations bpf_raw_tp_fops = {
1496 	.release	= bpf_raw_tracepoint_release,
1497 	.read		= bpf_dummy_read,
1498 	.write		= bpf_dummy_write,
1499 };
1500 
1501 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1502 
1503 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1504 {
1505 	struct bpf_raw_tracepoint *raw_tp;
1506 	struct bpf_raw_event_map *btp;
1507 	struct bpf_prog *prog;
1508 	char tp_name[128];
1509 	int tp_fd, err;
1510 
1511 	if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1512 			      sizeof(tp_name) - 1) < 0)
1513 		return -EFAULT;
1514 	tp_name[sizeof(tp_name) - 1] = 0;
1515 
1516 	btp = bpf_find_raw_tracepoint(tp_name);
1517 	if (!btp)
1518 		return -ENOENT;
1519 
1520 	raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1521 	if (!raw_tp)
1522 		return -ENOMEM;
1523 	raw_tp->btp = btp;
1524 
1525 	prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1526 				 BPF_PROG_TYPE_RAW_TRACEPOINT);
1527 	if (IS_ERR(prog)) {
1528 		err = PTR_ERR(prog);
1529 		goto out_free_tp;
1530 	}
1531 
1532 	err = bpf_probe_register(raw_tp->btp, prog);
1533 	if (err)
1534 		goto out_put_prog;
1535 
1536 	raw_tp->prog = prog;
1537 	tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1538 				 O_CLOEXEC);
1539 	if (tp_fd < 0) {
1540 		bpf_probe_unregister(raw_tp->btp, prog);
1541 		err = tp_fd;
1542 		goto out_put_prog;
1543 	}
1544 	return tp_fd;
1545 
1546 out_put_prog:
1547 	bpf_prog_put(prog);
1548 out_free_tp:
1549 	kfree(raw_tp);
1550 	return err;
1551 }
1552 
1553 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1554 					     enum bpf_attach_type attach_type)
1555 {
1556 	switch (prog->type) {
1557 	case BPF_PROG_TYPE_CGROUP_SOCK:
1558 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1559 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1560 	default:
1561 		return 0;
1562 	}
1563 }
1564 
1565 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1566 
1567 #define BPF_F_ATTACH_MASK \
1568 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1569 
1570 static int bpf_prog_attach(const union bpf_attr *attr)
1571 {
1572 	enum bpf_prog_type ptype;
1573 	struct bpf_prog *prog;
1574 	int ret;
1575 
1576 	if (!capable(CAP_NET_ADMIN))
1577 		return -EPERM;
1578 
1579 	if (CHECK_ATTR(BPF_PROG_ATTACH))
1580 		return -EINVAL;
1581 
1582 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1583 		return -EINVAL;
1584 
1585 	switch (attr->attach_type) {
1586 	case BPF_CGROUP_INET_INGRESS:
1587 	case BPF_CGROUP_INET_EGRESS:
1588 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1589 		break;
1590 	case BPF_CGROUP_INET_SOCK_CREATE:
1591 	case BPF_CGROUP_INET4_POST_BIND:
1592 	case BPF_CGROUP_INET6_POST_BIND:
1593 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1594 		break;
1595 	case BPF_CGROUP_INET4_BIND:
1596 	case BPF_CGROUP_INET6_BIND:
1597 	case BPF_CGROUP_INET4_CONNECT:
1598 	case BPF_CGROUP_INET6_CONNECT:
1599 	case BPF_CGROUP_UDP4_SENDMSG:
1600 	case BPF_CGROUP_UDP6_SENDMSG:
1601 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1602 		break;
1603 	case BPF_CGROUP_SOCK_OPS:
1604 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1605 		break;
1606 	case BPF_CGROUP_DEVICE:
1607 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1608 		break;
1609 	case BPF_SK_MSG_VERDICT:
1610 		ptype = BPF_PROG_TYPE_SK_MSG;
1611 		break;
1612 	case BPF_SK_SKB_STREAM_PARSER:
1613 	case BPF_SK_SKB_STREAM_VERDICT:
1614 		ptype = BPF_PROG_TYPE_SK_SKB;
1615 		break;
1616 	case BPF_LIRC_MODE2:
1617 		ptype = BPF_PROG_TYPE_LIRC_MODE2;
1618 		break;
1619 	default:
1620 		return -EINVAL;
1621 	}
1622 
1623 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1624 	if (IS_ERR(prog))
1625 		return PTR_ERR(prog);
1626 
1627 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1628 		bpf_prog_put(prog);
1629 		return -EINVAL;
1630 	}
1631 
1632 	switch (ptype) {
1633 	case BPF_PROG_TYPE_SK_SKB:
1634 	case BPF_PROG_TYPE_SK_MSG:
1635 		ret = sockmap_get_from_fd(attr, ptype, prog);
1636 		break;
1637 	case BPF_PROG_TYPE_LIRC_MODE2:
1638 		ret = lirc_prog_attach(attr, prog);
1639 		break;
1640 	default:
1641 		ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1642 	}
1643 
1644 	if (ret)
1645 		bpf_prog_put(prog);
1646 	return ret;
1647 }
1648 
1649 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1650 
1651 static int bpf_prog_detach(const union bpf_attr *attr)
1652 {
1653 	enum bpf_prog_type ptype;
1654 
1655 	if (!capable(CAP_NET_ADMIN))
1656 		return -EPERM;
1657 
1658 	if (CHECK_ATTR(BPF_PROG_DETACH))
1659 		return -EINVAL;
1660 
1661 	switch (attr->attach_type) {
1662 	case BPF_CGROUP_INET_INGRESS:
1663 	case BPF_CGROUP_INET_EGRESS:
1664 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1665 		break;
1666 	case BPF_CGROUP_INET_SOCK_CREATE:
1667 	case BPF_CGROUP_INET4_POST_BIND:
1668 	case BPF_CGROUP_INET6_POST_BIND:
1669 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1670 		break;
1671 	case BPF_CGROUP_INET4_BIND:
1672 	case BPF_CGROUP_INET6_BIND:
1673 	case BPF_CGROUP_INET4_CONNECT:
1674 	case BPF_CGROUP_INET6_CONNECT:
1675 	case BPF_CGROUP_UDP4_SENDMSG:
1676 	case BPF_CGROUP_UDP6_SENDMSG:
1677 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1678 		break;
1679 	case BPF_CGROUP_SOCK_OPS:
1680 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1681 		break;
1682 	case BPF_CGROUP_DEVICE:
1683 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1684 		break;
1685 	case BPF_SK_MSG_VERDICT:
1686 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, NULL);
1687 	case BPF_SK_SKB_STREAM_PARSER:
1688 	case BPF_SK_SKB_STREAM_VERDICT:
1689 		return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, NULL);
1690 	case BPF_LIRC_MODE2:
1691 		return lirc_prog_detach(attr);
1692 	default:
1693 		return -EINVAL;
1694 	}
1695 
1696 	return cgroup_bpf_prog_detach(attr, ptype);
1697 }
1698 
1699 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1700 
1701 static int bpf_prog_query(const union bpf_attr *attr,
1702 			  union bpf_attr __user *uattr)
1703 {
1704 	if (!capable(CAP_NET_ADMIN))
1705 		return -EPERM;
1706 	if (CHECK_ATTR(BPF_PROG_QUERY))
1707 		return -EINVAL;
1708 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1709 		return -EINVAL;
1710 
1711 	switch (attr->query.attach_type) {
1712 	case BPF_CGROUP_INET_INGRESS:
1713 	case BPF_CGROUP_INET_EGRESS:
1714 	case BPF_CGROUP_INET_SOCK_CREATE:
1715 	case BPF_CGROUP_INET4_BIND:
1716 	case BPF_CGROUP_INET6_BIND:
1717 	case BPF_CGROUP_INET4_POST_BIND:
1718 	case BPF_CGROUP_INET6_POST_BIND:
1719 	case BPF_CGROUP_INET4_CONNECT:
1720 	case BPF_CGROUP_INET6_CONNECT:
1721 	case BPF_CGROUP_UDP4_SENDMSG:
1722 	case BPF_CGROUP_UDP6_SENDMSG:
1723 	case BPF_CGROUP_SOCK_OPS:
1724 	case BPF_CGROUP_DEVICE:
1725 		break;
1726 	case BPF_LIRC_MODE2:
1727 		return lirc_prog_query(attr, uattr);
1728 	default:
1729 		return -EINVAL;
1730 	}
1731 
1732 	return cgroup_bpf_prog_query(attr, uattr);
1733 }
1734 
1735 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1736 
1737 static int bpf_prog_test_run(const union bpf_attr *attr,
1738 			     union bpf_attr __user *uattr)
1739 {
1740 	struct bpf_prog *prog;
1741 	int ret = -ENOTSUPP;
1742 
1743 	if (!capable(CAP_SYS_ADMIN))
1744 		return -EPERM;
1745 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1746 		return -EINVAL;
1747 
1748 	prog = bpf_prog_get(attr->test.prog_fd);
1749 	if (IS_ERR(prog))
1750 		return PTR_ERR(prog);
1751 
1752 	if (prog->aux->ops->test_run)
1753 		ret = prog->aux->ops->test_run(prog, attr, uattr);
1754 
1755 	bpf_prog_put(prog);
1756 	return ret;
1757 }
1758 
1759 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1760 
1761 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1762 			       union bpf_attr __user *uattr,
1763 			       struct idr *idr,
1764 			       spinlock_t *lock)
1765 {
1766 	u32 next_id = attr->start_id;
1767 	int err = 0;
1768 
1769 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1770 		return -EINVAL;
1771 
1772 	if (!capable(CAP_SYS_ADMIN))
1773 		return -EPERM;
1774 
1775 	next_id++;
1776 	spin_lock_bh(lock);
1777 	if (!idr_get_next(idr, &next_id))
1778 		err = -ENOENT;
1779 	spin_unlock_bh(lock);
1780 
1781 	if (!err)
1782 		err = put_user(next_id, &uattr->next_id);
1783 
1784 	return err;
1785 }
1786 
1787 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1788 
1789 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1790 {
1791 	struct bpf_prog *prog;
1792 	u32 id = attr->prog_id;
1793 	int fd;
1794 
1795 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1796 		return -EINVAL;
1797 
1798 	if (!capable(CAP_SYS_ADMIN))
1799 		return -EPERM;
1800 
1801 	spin_lock_bh(&prog_idr_lock);
1802 	prog = idr_find(&prog_idr, id);
1803 	if (prog)
1804 		prog = bpf_prog_inc_not_zero(prog);
1805 	else
1806 		prog = ERR_PTR(-ENOENT);
1807 	spin_unlock_bh(&prog_idr_lock);
1808 
1809 	if (IS_ERR(prog))
1810 		return PTR_ERR(prog);
1811 
1812 	fd = bpf_prog_new_fd(prog);
1813 	if (fd < 0)
1814 		bpf_prog_put(prog);
1815 
1816 	return fd;
1817 }
1818 
1819 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1820 
1821 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1822 {
1823 	struct bpf_map *map;
1824 	u32 id = attr->map_id;
1825 	int f_flags;
1826 	int fd;
1827 
1828 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1829 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1830 		return -EINVAL;
1831 
1832 	if (!capable(CAP_SYS_ADMIN))
1833 		return -EPERM;
1834 
1835 	f_flags = bpf_get_file_flag(attr->open_flags);
1836 	if (f_flags < 0)
1837 		return f_flags;
1838 
1839 	spin_lock_bh(&map_idr_lock);
1840 	map = idr_find(&map_idr, id);
1841 	if (map)
1842 		map = bpf_map_inc_not_zero(map, true);
1843 	else
1844 		map = ERR_PTR(-ENOENT);
1845 	spin_unlock_bh(&map_idr_lock);
1846 
1847 	if (IS_ERR(map))
1848 		return PTR_ERR(map);
1849 
1850 	fd = bpf_map_new_fd(map, f_flags);
1851 	if (fd < 0)
1852 		bpf_map_put(map);
1853 
1854 	return fd;
1855 }
1856 
1857 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1858 					      unsigned long addr)
1859 {
1860 	int i;
1861 
1862 	for (i = 0; i < prog->aux->used_map_cnt; i++)
1863 		if (prog->aux->used_maps[i] == (void *)addr)
1864 			return prog->aux->used_maps[i];
1865 	return NULL;
1866 }
1867 
1868 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1869 {
1870 	const struct bpf_map *map;
1871 	struct bpf_insn *insns;
1872 	u64 imm;
1873 	int i;
1874 
1875 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1876 			GFP_USER);
1877 	if (!insns)
1878 		return insns;
1879 
1880 	for (i = 0; i < prog->len; i++) {
1881 		if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1882 			insns[i].code = BPF_JMP | BPF_CALL;
1883 			insns[i].imm = BPF_FUNC_tail_call;
1884 			/* fall-through */
1885 		}
1886 		if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1887 		    insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1888 			if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1889 				insns[i].code = BPF_JMP | BPF_CALL;
1890 			if (!bpf_dump_raw_ok())
1891 				insns[i].imm = 0;
1892 			continue;
1893 		}
1894 
1895 		if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1896 			continue;
1897 
1898 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1899 		map = bpf_map_from_imm(prog, imm);
1900 		if (map) {
1901 			insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1902 			insns[i].imm = map->id;
1903 			insns[i + 1].imm = 0;
1904 			continue;
1905 		}
1906 
1907 		if (!bpf_dump_raw_ok() &&
1908 		    imm == (unsigned long)prog->aux) {
1909 			insns[i].imm = 0;
1910 			insns[i + 1].imm = 0;
1911 			continue;
1912 		}
1913 	}
1914 
1915 	return insns;
1916 }
1917 
1918 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1919 				   const union bpf_attr *attr,
1920 				   union bpf_attr __user *uattr)
1921 {
1922 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1923 	struct bpf_prog_info info = {};
1924 	u32 info_len = attr->info.info_len;
1925 	char __user *uinsns;
1926 	u32 ulen;
1927 	int err;
1928 
1929 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1930 	if (err)
1931 		return err;
1932 	info_len = min_t(u32, sizeof(info), info_len);
1933 
1934 	if (copy_from_user(&info, uinfo, info_len))
1935 		return -EFAULT;
1936 
1937 	info.type = prog->type;
1938 	info.id = prog->aux->id;
1939 	info.load_time = prog->aux->load_time;
1940 	info.created_by_uid = from_kuid_munged(current_user_ns(),
1941 					       prog->aux->user->uid);
1942 	info.gpl_compatible = prog->gpl_compatible;
1943 
1944 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
1945 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1946 
1947 	ulen = info.nr_map_ids;
1948 	info.nr_map_ids = prog->aux->used_map_cnt;
1949 	ulen = min_t(u32, info.nr_map_ids, ulen);
1950 	if (ulen) {
1951 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1952 		u32 i;
1953 
1954 		for (i = 0; i < ulen; i++)
1955 			if (put_user(prog->aux->used_maps[i]->id,
1956 				     &user_map_ids[i]))
1957 				return -EFAULT;
1958 	}
1959 
1960 	if (!capable(CAP_SYS_ADMIN)) {
1961 		info.jited_prog_len = 0;
1962 		info.xlated_prog_len = 0;
1963 		info.nr_jited_ksyms = 0;
1964 		goto done;
1965 	}
1966 
1967 	ulen = info.xlated_prog_len;
1968 	info.xlated_prog_len = bpf_prog_insn_size(prog);
1969 	if (info.xlated_prog_len && ulen) {
1970 		struct bpf_insn *insns_sanitized;
1971 		bool fault;
1972 
1973 		if (prog->blinded && !bpf_dump_raw_ok()) {
1974 			info.xlated_prog_insns = 0;
1975 			goto done;
1976 		}
1977 		insns_sanitized = bpf_insn_prepare_dump(prog);
1978 		if (!insns_sanitized)
1979 			return -ENOMEM;
1980 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1981 		ulen = min_t(u32, info.xlated_prog_len, ulen);
1982 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
1983 		kfree(insns_sanitized);
1984 		if (fault)
1985 			return -EFAULT;
1986 	}
1987 
1988 	if (bpf_prog_is_dev_bound(prog->aux)) {
1989 		err = bpf_prog_offload_info_fill(&info, prog);
1990 		if (err)
1991 			return err;
1992 		goto done;
1993 	}
1994 
1995 	/* NOTE: the following code is supposed to be skipped for offload.
1996 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
1997 	 * for offload.
1998 	 */
1999 	ulen = info.jited_prog_len;
2000 	if (prog->aux->func_cnt) {
2001 		u32 i;
2002 
2003 		info.jited_prog_len = 0;
2004 		for (i = 0; i < prog->aux->func_cnt; i++)
2005 			info.jited_prog_len += prog->aux->func[i]->jited_len;
2006 	} else {
2007 		info.jited_prog_len = prog->jited_len;
2008 	}
2009 
2010 	if (info.jited_prog_len && ulen) {
2011 		if (bpf_dump_raw_ok()) {
2012 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
2013 			ulen = min_t(u32, info.jited_prog_len, ulen);
2014 
2015 			/* for multi-function programs, copy the JITed
2016 			 * instructions for all the functions
2017 			 */
2018 			if (prog->aux->func_cnt) {
2019 				u32 len, free, i;
2020 				u8 *img;
2021 
2022 				free = ulen;
2023 				for (i = 0; i < prog->aux->func_cnt; i++) {
2024 					len = prog->aux->func[i]->jited_len;
2025 					len = min_t(u32, len, free);
2026 					img = (u8 *) prog->aux->func[i]->bpf_func;
2027 					if (copy_to_user(uinsns, img, len))
2028 						return -EFAULT;
2029 					uinsns += len;
2030 					free -= len;
2031 					if (!free)
2032 						break;
2033 				}
2034 			} else {
2035 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
2036 					return -EFAULT;
2037 			}
2038 		} else {
2039 			info.jited_prog_insns = 0;
2040 		}
2041 	}
2042 
2043 	ulen = info.nr_jited_ksyms;
2044 	info.nr_jited_ksyms = prog->aux->func_cnt;
2045 	if (info.nr_jited_ksyms && ulen) {
2046 		if (bpf_dump_raw_ok()) {
2047 			u64 __user *user_ksyms;
2048 			ulong ksym_addr;
2049 			u32 i;
2050 
2051 			/* copy the address of the kernel symbol
2052 			 * corresponding to each function
2053 			 */
2054 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2055 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2056 			for (i = 0; i < ulen; i++) {
2057 				ksym_addr = (ulong) prog->aux->func[i]->bpf_func;
2058 				ksym_addr &= PAGE_MASK;
2059 				if (put_user((u64) ksym_addr, &user_ksyms[i]))
2060 					return -EFAULT;
2061 			}
2062 		} else {
2063 			info.jited_ksyms = 0;
2064 		}
2065 	}
2066 
2067 	ulen = info.nr_jited_func_lens;
2068 	info.nr_jited_func_lens = prog->aux->func_cnt;
2069 	if (info.nr_jited_func_lens && ulen) {
2070 		if (bpf_dump_raw_ok()) {
2071 			u32 __user *user_lens;
2072 			u32 func_len, i;
2073 
2074 			/* copy the JITed image lengths for each function */
2075 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2076 			user_lens = u64_to_user_ptr(info.jited_func_lens);
2077 			for (i = 0; i < ulen; i++) {
2078 				func_len = prog->aux->func[i]->jited_len;
2079 				if (put_user(func_len, &user_lens[i]))
2080 					return -EFAULT;
2081 			}
2082 		} else {
2083 			info.jited_func_lens = 0;
2084 		}
2085 	}
2086 
2087 done:
2088 	if (copy_to_user(uinfo, &info, info_len) ||
2089 	    put_user(info_len, &uattr->info.info_len))
2090 		return -EFAULT;
2091 
2092 	return 0;
2093 }
2094 
2095 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2096 				  const union bpf_attr *attr,
2097 				  union bpf_attr __user *uattr)
2098 {
2099 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2100 	struct bpf_map_info info = {};
2101 	u32 info_len = attr->info.info_len;
2102 	int err;
2103 
2104 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2105 	if (err)
2106 		return err;
2107 	info_len = min_t(u32, sizeof(info), info_len);
2108 
2109 	info.type = map->map_type;
2110 	info.id = map->id;
2111 	info.key_size = map->key_size;
2112 	info.value_size = map->value_size;
2113 	info.max_entries = map->max_entries;
2114 	info.map_flags = map->map_flags;
2115 	memcpy(info.name, map->name, sizeof(map->name));
2116 
2117 	if (map->btf) {
2118 		info.btf_id = btf_id(map->btf);
2119 		info.btf_key_type_id = map->btf_key_type_id;
2120 		info.btf_value_type_id = map->btf_value_type_id;
2121 	}
2122 
2123 	if (bpf_map_is_dev_bound(map)) {
2124 		err = bpf_map_offload_info_fill(&info, map);
2125 		if (err)
2126 			return err;
2127 	}
2128 
2129 	if (copy_to_user(uinfo, &info, info_len) ||
2130 	    put_user(info_len, &uattr->info.info_len))
2131 		return -EFAULT;
2132 
2133 	return 0;
2134 }
2135 
2136 static int bpf_btf_get_info_by_fd(struct btf *btf,
2137 				  const union bpf_attr *attr,
2138 				  union bpf_attr __user *uattr)
2139 {
2140 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2141 	u32 info_len = attr->info.info_len;
2142 	int err;
2143 
2144 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2145 	if (err)
2146 		return err;
2147 
2148 	return btf_get_info_by_fd(btf, attr, uattr);
2149 }
2150 
2151 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2152 
2153 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2154 				  union bpf_attr __user *uattr)
2155 {
2156 	int ufd = attr->info.bpf_fd;
2157 	struct fd f;
2158 	int err;
2159 
2160 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2161 		return -EINVAL;
2162 
2163 	f = fdget(ufd);
2164 	if (!f.file)
2165 		return -EBADFD;
2166 
2167 	if (f.file->f_op == &bpf_prog_fops)
2168 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2169 					      uattr);
2170 	else if (f.file->f_op == &bpf_map_fops)
2171 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2172 					     uattr);
2173 	else if (f.file->f_op == &btf_fops)
2174 		err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2175 	else
2176 		err = -EINVAL;
2177 
2178 	fdput(f);
2179 	return err;
2180 }
2181 
2182 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2183 
2184 static int bpf_btf_load(const union bpf_attr *attr)
2185 {
2186 	if (CHECK_ATTR(BPF_BTF_LOAD))
2187 		return -EINVAL;
2188 
2189 	if (!capable(CAP_SYS_ADMIN))
2190 		return -EPERM;
2191 
2192 	return btf_new_fd(attr);
2193 }
2194 
2195 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2196 
2197 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2198 {
2199 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2200 		return -EINVAL;
2201 
2202 	if (!capable(CAP_SYS_ADMIN))
2203 		return -EPERM;
2204 
2205 	return btf_get_fd_by_id(attr->btf_id);
2206 }
2207 
2208 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2209 				    union bpf_attr __user *uattr,
2210 				    u32 prog_id, u32 fd_type,
2211 				    const char *buf, u64 probe_offset,
2212 				    u64 probe_addr)
2213 {
2214 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2215 	u32 len = buf ? strlen(buf) : 0, input_len;
2216 	int err = 0;
2217 
2218 	if (put_user(len, &uattr->task_fd_query.buf_len))
2219 		return -EFAULT;
2220 	input_len = attr->task_fd_query.buf_len;
2221 	if (input_len && ubuf) {
2222 		if (!len) {
2223 			/* nothing to copy, just make ubuf NULL terminated */
2224 			char zero = '\0';
2225 
2226 			if (put_user(zero, ubuf))
2227 				return -EFAULT;
2228 		} else if (input_len >= len + 1) {
2229 			/* ubuf can hold the string with NULL terminator */
2230 			if (copy_to_user(ubuf, buf, len + 1))
2231 				return -EFAULT;
2232 		} else {
2233 			/* ubuf cannot hold the string with NULL terminator,
2234 			 * do a partial copy with NULL terminator.
2235 			 */
2236 			char zero = '\0';
2237 
2238 			err = -ENOSPC;
2239 			if (copy_to_user(ubuf, buf, input_len - 1))
2240 				return -EFAULT;
2241 			if (put_user(zero, ubuf + input_len - 1))
2242 				return -EFAULT;
2243 		}
2244 	}
2245 
2246 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2247 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2248 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2249 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2250 		return -EFAULT;
2251 
2252 	return err;
2253 }
2254 
2255 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2256 
2257 static int bpf_task_fd_query(const union bpf_attr *attr,
2258 			     union bpf_attr __user *uattr)
2259 {
2260 	pid_t pid = attr->task_fd_query.pid;
2261 	u32 fd = attr->task_fd_query.fd;
2262 	const struct perf_event *event;
2263 	struct files_struct *files;
2264 	struct task_struct *task;
2265 	struct file *file;
2266 	int err;
2267 
2268 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2269 		return -EINVAL;
2270 
2271 	if (!capable(CAP_SYS_ADMIN))
2272 		return -EPERM;
2273 
2274 	if (attr->task_fd_query.flags != 0)
2275 		return -EINVAL;
2276 
2277 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2278 	if (!task)
2279 		return -ENOENT;
2280 
2281 	files = get_files_struct(task);
2282 	put_task_struct(task);
2283 	if (!files)
2284 		return -ENOENT;
2285 
2286 	err = 0;
2287 	spin_lock(&files->file_lock);
2288 	file = fcheck_files(files, fd);
2289 	if (!file)
2290 		err = -EBADF;
2291 	else
2292 		get_file(file);
2293 	spin_unlock(&files->file_lock);
2294 	put_files_struct(files);
2295 
2296 	if (err)
2297 		goto out;
2298 
2299 	if (file->f_op == &bpf_raw_tp_fops) {
2300 		struct bpf_raw_tracepoint *raw_tp = file->private_data;
2301 		struct bpf_raw_event_map *btp = raw_tp->btp;
2302 
2303 		err = bpf_task_fd_query_copy(attr, uattr,
2304 					     raw_tp->prog->aux->id,
2305 					     BPF_FD_TYPE_RAW_TRACEPOINT,
2306 					     btp->tp->name, 0, 0);
2307 		goto put_file;
2308 	}
2309 
2310 	event = perf_get_event(file);
2311 	if (!IS_ERR(event)) {
2312 		u64 probe_offset, probe_addr;
2313 		u32 prog_id, fd_type;
2314 		const char *buf;
2315 
2316 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2317 					      &buf, &probe_offset,
2318 					      &probe_addr);
2319 		if (!err)
2320 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2321 						     fd_type, buf,
2322 						     probe_offset,
2323 						     probe_addr);
2324 		goto put_file;
2325 	}
2326 
2327 	err = -ENOTSUPP;
2328 put_file:
2329 	fput(file);
2330 out:
2331 	return err;
2332 }
2333 
2334 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2335 {
2336 	union bpf_attr attr = {};
2337 	int err;
2338 
2339 	if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2340 		return -EPERM;
2341 
2342 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2343 	if (err)
2344 		return err;
2345 	size = min_t(u32, size, sizeof(attr));
2346 
2347 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
2348 	if (copy_from_user(&attr, uattr, size) != 0)
2349 		return -EFAULT;
2350 
2351 	err = security_bpf(cmd, &attr, size);
2352 	if (err < 0)
2353 		return err;
2354 
2355 	switch (cmd) {
2356 	case BPF_MAP_CREATE:
2357 		err = map_create(&attr);
2358 		break;
2359 	case BPF_MAP_LOOKUP_ELEM:
2360 		err = map_lookup_elem(&attr);
2361 		break;
2362 	case BPF_MAP_UPDATE_ELEM:
2363 		err = map_update_elem(&attr);
2364 		break;
2365 	case BPF_MAP_DELETE_ELEM:
2366 		err = map_delete_elem(&attr);
2367 		break;
2368 	case BPF_MAP_GET_NEXT_KEY:
2369 		err = map_get_next_key(&attr);
2370 		break;
2371 	case BPF_PROG_LOAD:
2372 		err = bpf_prog_load(&attr);
2373 		break;
2374 	case BPF_OBJ_PIN:
2375 		err = bpf_obj_pin(&attr);
2376 		break;
2377 	case BPF_OBJ_GET:
2378 		err = bpf_obj_get(&attr);
2379 		break;
2380 	case BPF_PROG_ATTACH:
2381 		err = bpf_prog_attach(&attr);
2382 		break;
2383 	case BPF_PROG_DETACH:
2384 		err = bpf_prog_detach(&attr);
2385 		break;
2386 	case BPF_PROG_QUERY:
2387 		err = bpf_prog_query(&attr, uattr);
2388 		break;
2389 	case BPF_PROG_TEST_RUN:
2390 		err = bpf_prog_test_run(&attr, uattr);
2391 		break;
2392 	case BPF_PROG_GET_NEXT_ID:
2393 		err = bpf_obj_get_next_id(&attr, uattr,
2394 					  &prog_idr, &prog_idr_lock);
2395 		break;
2396 	case BPF_MAP_GET_NEXT_ID:
2397 		err = bpf_obj_get_next_id(&attr, uattr,
2398 					  &map_idr, &map_idr_lock);
2399 		break;
2400 	case BPF_PROG_GET_FD_BY_ID:
2401 		err = bpf_prog_get_fd_by_id(&attr);
2402 		break;
2403 	case BPF_MAP_GET_FD_BY_ID:
2404 		err = bpf_map_get_fd_by_id(&attr);
2405 		break;
2406 	case BPF_OBJ_GET_INFO_BY_FD:
2407 		err = bpf_obj_get_info_by_fd(&attr, uattr);
2408 		break;
2409 	case BPF_RAW_TRACEPOINT_OPEN:
2410 		err = bpf_raw_tracepoint_open(&attr);
2411 		break;
2412 	case BPF_BTF_LOAD:
2413 		err = bpf_btf_load(&attr);
2414 		break;
2415 	case BPF_BTF_GET_FD_BY_ID:
2416 		err = bpf_btf_get_fd_by_id(&attr);
2417 		break;
2418 	case BPF_TASK_FD_QUERY:
2419 		err = bpf_task_fd_query(&attr, uattr);
2420 		break;
2421 	default:
2422 		err = -EINVAL;
2423 		break;
2424 	}
2425 
2426 	return err;
2427 }
2428