xref: /linux/kernel/bpf/syscall.c (revision 995231c820e3bd3633cb38bf4ea6f2541e1da331)
1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2  *
3  * This program is free software; you can redistribute it and/or
4  * modify it under the terms of version 2 of the GNU General Public
5  * License as published by the Free Software Foundation.
6  *
7  * This program is distributed in the hope that it will be useful, but
8  * WITHOUT ANY WARRANTY; without even the implied warranty of
9  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
10  * General Public License for more details.
11  */
12 #include <linux/bpf.h>
13 #include <linux/bpf_trace.h>
14 #include <linux/syscalls.h>
15 #include <linux/slab.h>
16 #include <linux/sched/signal.h>
17 #include <linux/vmalloc.h>
18 #include <linux/mmzone.h>
19 #include <linux/anon_inodes.h>
20 #include <linux/file.h>
21 #include <linux/license.h>
22 #include <linux/filter.h>
23 #include <linux/version.h>
24 #include <linux/kernel.h>
25 #include <linux/idr.h>
26 #include <linux/cred.h>
27 #include <linux/timekeeping.h>
28 #include <linux/ctype.h>
29 
30 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
31 			   (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
32 			   (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
33 			   (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
34 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
35 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
36 
37 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
38 
39 DEFINE_PER_CPU(int, bpf_prog_active);
40 static DEFINE_IDR(prog_idr);
41 static DEFINE_SPINLOCK(prog_idr_lock);
42 static DEFINE_IDR(map_idr);
43 static DEFINE_SPINLOCK(map_idr_lock);
44 
45 int sysctl_unprivileged_bpf_disabled __read_mostly;
46 
47 static const struct bpf_map_ops * const bpf_map_types[] = {
48 #define BPF_PROG_TYPE(_id, _ops)
49 #define BPF_MAP_TYPE(_id, _ops) \
50 	[_id] = &_ops,
51 #include <linux/bpf_types.h>
52 #undef BPF_PROG_TYPE
53 #undef BPF_MAP_TYPE
54 };
55 
56 /*
57  * If we're handed a bigger struct than we know of, ensure all the unknown bits
58  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
59  * we don't know about yet.
60  *
61  * There is a ToCToU between this function call and the following
62  * copy_from_user() call. However, this is not a concern since this function is
63  * meant to be a future-proofing of bits.
64  */
65 static int check_uarg_tail_zero(void __user *uaddr,
66 				size_t expected_size,
67 				size_t actual_size)
68 {
69 	unsigned char __user *addr;
70 	unsigned char __user *end;
71 	unsigned char val;
72 	int err;
73 
74 	if (unlikely(actual_size > PAGE_SIZE))	/* silly large */
75 		return -E2BIG;
76 
77 	if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
78 		return -EFAULT;
79 
80 	if (actual_size <= expected_size)
81 		return 0;
82 
83 	addr = uaddr + expected_size;
84 	end  = uaddr + actual_size;
85 
86 	for (; addr < end; addr++) {
87 		err = get_user(val, addr);
88 		if (err)
89 			return err;
90 		if (val)
91 			return -E2BIG;
92 	}
93 
94 	return 0;
95 }
96 
97 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
98 {
99 	struct bpf_map *map;
100 
101 	if (attr->map_type >= ARRAY_SIZE(bpf_map_types) ||
102 	    !bpf_map_types[attr->map_type])
103 		return ERR_PTR(-EINVAL);
104 
105 	map = bpf_map_types[attr->map_type]->map_alloc(attr);
106 	if (IS_ERR(map))
107 		return map;
108 	map->ops = bpf_map_types[attr->map_type];
109 	map->map_type = attr->map_type;
110 	return map;
111 }
112 
113 void *bpf_map_area_alloc(size_t size, int numa_node)
114 {
115 	/* We definitely need __GFP_NORETRY, so OOM killer doesn't
116 	 * trigger under memory pressure as we really just want to
117 	 * fail instead.
118 	 */
119 	const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
120 	void *area;
121 
122 	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
123 		area = kmalloc_node(size, GFP_USER | flags, numa_node);
124 		if (area != NULL)
125 			return area;
126 	}
127 
128 	return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
129 					   __builtin_return_address(0));
130 }
131 
132 void bpf_map_area_free(void *area)
133 {
134 	kvfree(area);
135 }
136 
137 int bpf_map_precharge_memlock(u32 pages)
138 {
139 	struct user_struct *user = get_current_user();
140 	unsigned long memlock_limit, cur;
141 
142 	memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
143 	cur = atomic_long_read(&user->locked_vm);
144 	free_uid(user);
145 	if (cur + pages > memlock_limit)
146 		return -EPERM;
147 	return 0;
148 }
149 
150 static int bpf_map_charge_memlock(struct bpf_map *map)
151 {
152 	struct user_struct *user = get_current_user();
153 	unsigned long memlock_limit;
154 
155 	memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
156 
157 	atomic_long_add(map->pages, &user->locked_vm);
158 
159 	if (atomic_long_read(&user->locked_vm) > memlock_limit) {
160 		atomic_long_sub(map->pages, &user->locked_vm);
161 		free_uid(user);
162 		return -EPERM;
163 	}
164 	map->user = user;
165 	return 0;
166 }
167 
168 static void bpf_map_uncharge_memlock(struct bpf_map *map)
169 {
170 	struct user_struct *user = map->user;
171 
172 	atomic_long_sub(map->pages, &user->locked_vm);
173 	free_uid(user);
174 }
175 
176 static int bpf_map_alloc_id(struct bpf_map *map)
177 {
178 	int id;
179 
180 	spin_lock_bh(&map_idr_lock);
181 	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
182 	if (id > 0)
183 		map->id = id;
184 	spin_unlock_bh(&map_idr_lock);
185 
186 	if (WARN_ON_ONCE(!id))
187 		return -ENOSPC;
188 
189 	return id > 0 ? 0 : id;
190 }
191 
192 static void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
193 {
194 	unsigned long flags;
195 
196 	if (do_idr_lock)
197 		spin_lock_irqsave(&map_idr_lock, flags);
198 	else
199 		__acquire(&map_idr_lock);
200 
201 	idr_remove(&map_idr, map->id);
202 
203 	if (do_idr_lock)
204 		spin_unlock_irqrestore(&map_idr_lock, flags);
205 	else
206 		__release(&map_idr_lock);
207 }
208 
209 /* called from workqueue */
210 static void bpf_map_free_deferred(struct work_struct *work)
211 {
212 	struct bpf_map *map = container_of(work, struct bpf_map, work);
213 
214 	bpf_map_uncharge_memlock(map);
215 	security_bpf_map_free(map);
216 	/* implementation dependent freeing */
217 	map->ops->map_free(map);
218 }
219 
220 static void bpf_map_put_uref(struct bpf_map *map)
221 {
222 	if (atomic_dec_and_test(&map->usercnt)) {
223 		if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
224 			bpf_fd_array_map_clear(map);
225 	}
226 }
227 
228 /* decrement map refcnt and schedule it for freeing via workqueue
229  * (unrelying map implementation ops->map_free() might sleep)
230  */
231 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
232 {
233 	if (atomic_dec_and_test(&map->refcnt)) {
234 		/* bpf_map_free_id() must be called first */
235 		bpf_map_free_id(map, do_idr_lock);
236 		INIT_WORK(&map->work, bpf_map_free_deferred);
237 		schedule_work(&map->work);
238 	}
239 }
240 
241 void bpf_map_put(struct bpf_map *map)
242 {
243 	__bpf_map_put(map, true);
244 }
245 
246 void bpf_map_put_with_uref(struct bpf_map *map)
247 {
248 	bpf_map_put_uref(map);
249 	bpf_map_put(map);
250 }
251 
252 static int bpf_map_release(struct inode *inode, struct file *filp)
253 {
254 	struct bpf_map *map = filp->private_data;
255 
256 	if (map->ops->map_release)
257 		map->ops->map_release(map, filp);
258 
259 	bpf_map_put_with_uref(map);
260 	return 0;
261 }
262 
263 #ifdef CONFIG_PROC_FS
264 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
265 {
266 	const struct bpf_map *map = filp->private_data;
267 	const struct bpf_array *array;
268 	u32 owner_prog_type = 0;
269 	u32 owner_jited = 0;
270 
271 	if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
272 		array = container_of(map, struct bpf_array, map);
273 		owner_prog_type = array->owner_prog_type;
274 		owner_jited = array->owner_jited;
275 	}
276 
277 	seq_printf(m,
278 		   "map_type:\t%u\n"
279 		   "key_size:\t%u\n"
280 		   "value_size:\t%u\n"
281 		   "max_entries:\t%u\n"
282 		   "map_flags:\t%#x\n"
283 		   "memlock:\t%llu\n",
284 		   map->map_type,
285 		   map->key_size,
286 		   map->value_size,
287 		   map->max_entries,
288 		   map->map_flags,
289 		   map->pages * 1ULL << PAGE_SHIFT);
290 
291 	if (owner_prog_type) {
292 		seq_printf(m, "owner_prog_type:\t%u\n",
293 			   owner_prog_type);
294 		seq_printf(m, "owner_jited:\t%u\n",
295 			   owner_jited);
296 	}
297 }
298 #endif
299 
300 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
301 			      loff_t *ppos)
302 {
303 	/* We need this handler such that alloc_file() enables
304 	 * f_mode with FMODE_CAN_READ.
305 	 */
306 	return -EINVAL;
307 }
308 
309 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
310 			       size_t siz, loff_t *ppos)
311 {
312 	/* We need this handler such that alloc_file() enables
313 	 * f_mode with FMODE_CAN_WRITE.
314 	 */
315 	return -EINVAL;
316 }
317 
318 const struct file_operations bpf_map_fops = {
319 #ifdef CONFIG_PROC_FS
320 	.show_fdinfo	= bpf_map_show_fdinfo,
321 #endif
322 	.release	= bpf_map_release,
323 	.read		= bpf_dummy_read,
324 	.write		= bpf_dummy_write,
325 };
326 
327 int bpf_map_new_fd(struct bpf_map *map, int flags)
328 {
329 	int ret;
330 
331 	ret = security_bpf_map(map, OPEN_FMODE(flags));
332 	if (ret < 0)
333 		return ret;
334 
335 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
336 				flags | O_CLOEXEC);
337 }
338 
339 int bpf_get_file_flag(int flags)
340 {
341 	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
342 		return -EINVAL;
343 	if (flags & BPF_F_RDONLY)
344 		return O_RDONLY;
345 	if (flags & BPF_F_WRONLY)
346 		return O_WRONLY;
347 	return O_RDWR;
348 }
349 
350 /* helper macro to check that unused fields 'union bpf_attr' are zero */
351 #define CHECK_ATTR(CMD) \
352 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
353 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
354 		   sizeof(*attr) - \
355 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
356 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
357 
358 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
359  * Return 0 on success and < 0 on error.
360  */
361 static int bpf_obj_name_cpy(char *dst, const char *src)
362 {
363 	const char *end = src + BPF_OBJ_NAME_LEN;
364 
365 	memset(dst, 0, BPF_OBJ_NAME_LEN);
366 
367 	/* Copy all isalnum() and '_' char */
368 	while (src < end && *src) {
369 		if (!isalnum(*src) && *src != '_')
370 			return -EINVAL;
371 		*dst++ = *src++;
372 	}
373 
374 	/* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
375 	if (src == end)
376 		return -EINVAL;
377 
378 	return 0;
379 }
380 
381 #define BPF_MAP_CREATE_LAST_FIELD map_name
382 /* called via syscall */
383 static int map_create(union bpf_attr *attr)
384 {
385 	int numa_node = bpf_map_attr_numa_node(attr);
386 	struct bpf_map *map;
387 	int f_flags;
388 	int err;
389 
390 	err = CHECK_ATTR(BPF_MAP_CREATE);
391 	if (err)
392 		return -EINVAL;
393 
394 	f_flags = bpf_get_file_flag(attr->map_flags);
395 	if (f_flags < 0)
396 		return f_flags;
397 
398 	if (numa_node != NUMA_NO_NODE &&
399 	    ((unsigned int)numa_node >= nr_node_ids ||
400 	     !node_online(numa_node)))
401 		return -EINVAL;
402 
403 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
404 	map = find_and_alloc_map(attr);
405 	if (IS_ERR(map))
406 		return PTR_ERR(map);
407 
408 	err = bpf_obj_name_cpy(map->name, attr->map_name);
409 	if (err)
410 		goto free_map_nouncharge;
411 
412 	atomic_set(&map->refcnt, 1);
413 	atomic_set(&map->usercnt, 1);
414 
415 	err = security_bpf_map_alloc(map);
416 	if (err)
417 		goto free_map_nouncharge;
418 
419 	err = bpf_map_charge_memlock(map);
420 	if (err)
421 		goto free_map_sec;
422 
423 	err = bpf_map_alloc_id(map);
424 	if (err)
425 		goto free_map;
426 
427 	err = bpf_map_new_fd(map, f_flags);
428 	if (err < 0) {
429 		/* failed to allocate fd.
430 		 * bpf_map_put() is needed because the above
431 		 * bpf_map_alloc_id() has published the map
432 		 * to the userspace and the userspace may
433 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
434 		 */
435 		bpf_map_put(map);
436 		return err;
437 	}
438 
439 	trace_bpf_map_create(map, err);
440 	return err;
441 
442 free_map:
443 	bpf_map_uncharge_memlock(map);
444 free_map_sec:
445 	security_bpf_map_free(map);
446 free_map_nouncharge:
447 	map->ops->map_free(map);
448 	return err;
449 }
450 
451 /* if error is returned, fd is released.
452  * On success caller should complete fd access with matching fdput()
453  */
454 struct bpf_map *__bpf_map_get(struct fd f)
455 {
456 	if (!f.file)
457 		return ERR_PTR(-EBADF);
458 	if (f.file->f_op != &bpf_map_fops) {
459 		fdput(f);
460 		return ERR_PTR(-EINVAL);
461 	}
462 
463 	return f.file->private_data;
464 }
465 
466 /* prog's and map's refcnt limit */
467 #define BPF_MAX_REFCNT 32768
468 
469 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
470 {
471 	if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
472 		atomic_dec(&map->refcnt);
473 		return ERR_PTR(-EBUSY);
474 	}
475 	if (uref)
476 		atomic_inc(&map->usercnt);
477 	return map;
478 }
479 
480 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
481 {
482 	struct fd f = fdget(ufd);
483 	struct bpf_map *map;
484 
485 	map = __bpf_map_get(f);
486 	if (IS_ERR(map))
487 		return map;
488 
489 	map = bpf_map_inc(map, true);
490 	fdput(f);
491 
492 	return map;
493 }
494 
495 /* map_idr_lock should have been held */
496 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
497 					    bool uref)
498 {
499 	int refold;
500 
501 	refold = __atomic_add_unless(&map->refcnt, 1, 0);
502 
503 	if (refold >= BPF_MAX_REFCNT) {
504 		__bpf_map_put(map, false);
505 		return ERR_PTR(-EBUSY);
506 	}
507 
508 	if (!refold)
509 		return ERR_PTR(-ENOENT);
510 
511 	if (uref)
512 		atomic_inc(&map->usercnt);
513 
514 	return map;
515 }
516 
517 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
518 {
519 	return -ENOTSUPP;
520 }
521 
522 /* last field in 'union bpf_attr' used by this command */
523 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
524 
525 static int map_lookup_elem(union bpf_attr *attr)
526 {
527 	void __user *ukey = u64_to_user_ptr(attr->key);
528 	void __user *uvalue = u64_to_user_ptr(attr->value);
529 	int ufd = attr->map_fd;
530 	struct bpf_map *map;
531 	void *key, *value, *ptr;
532 	u32 value_size;
533 	struct fd f;
534 	int err;
535 
536 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
537 		return -EINVAL;
538 
539 	f = fdget(ufd);
540 	map = __bpf_map_get(f);
541 	if (IS_ERR(map))
542 		return PTR_ERR(map);
543 
544 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
545 		err = -EPERM;
546 		goto err_put;
547 	}
548 
549 	key = memdup_user(ukey, map->key_size);
550 	if (IS_ERR(key)) {
551 		err = PTR_ERR(key);
552 		goto err_put;
553 	}
554 
555 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
556 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
557 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
558 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
559 	else if (IS_FD_MAP(map))
560 		value_size = sizeof(u32);
561 	else
562 		value_size = map->value_size;
563 
564 	err = -ENOMEM;
565 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
566 	if (!value)
567 		goto free_key;
568 
569 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
570 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
571 		err = bpf_percpu_hash_copy(map, key, value);
572 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
573 		err = bpf_percpu_array_copy(map, key, value);
574 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
575 		err = bpf_stackmap_copy(map, key, value);
576 	} else if (IS_FD_ARRAY(map)) {
577 		err = bpf_fd_array_map_lookup_elem(map, key, value);
578 	} else if (IS_FD_HASH(map)) {
579 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
580 	} else {
581 		rcu_read_lock();
582 		ptr = map->ops->map_lookup_elem(map, key);
583 		if (ptr)
584 			memcpy(value, ptr, value_size);
585 		rcu_read_unlock();
586 		err = ptr ? 0 : -ENOENT;
587 	}
588 
589 	if (err)
590 		goto free_value;
591 
592 	err = -EFAULT;
593 	if (copy_to_user(uvalue, value, value_size) != 0)
594 		goto free_value;
595 
596 	trace_bpf_map_lookup_elem(map, ufd, key, value);
597 	err = 0;
598 
599 free_value:
600 	kfree(value);
601 free_key:
602 	kfree(key);
603 err_put:
604 	fdput(f);
605 	return err;
606 }
607 
608 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
609 
610 static int map_update_elem(union bpf_attr *attr)
611 {
612 	void __user *ukey = u64_to_user_ptr(attr->key);
613 	void __user *uvalue = u64_to_user_ptr(attr->value);
614 	int ufd = attr->map_fd;
615 	struct bpf_map *map;
616 	void *key, *value;
617 	u32 value_size;
618 	struct fd f;
619 	int err;
620 
621 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
622 		return -EINVAL;
623 
624 	f = fdget(ufd);
625 	map = __bpf_map_get(f);
626 	if (IS_ERR(map))
627 		return PTR_ERR(map);
628 
629 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
630 		err = -EPERM;
631 		goto err_put;
632 	}
633 
634 	key = memdup_user(ukey, map->key_size);
635 	if (IS_ERR(key)) {
636 		err = PTR_ERR(key);
637 		goto err_put;
638 	}
639 
640 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
641 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
642 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
643 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
644 	else
645 		value_size = map->value_size;
646 
647 	err = -ENOMEM;
648 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
649 	if (!value)
650 		goto free_key;
651 
652 	err = -EFAULT;
653 	if (copy_from_user(value, uvalue, value_size) != 0)
654 		goto free_value;
655 
656 	/* Need to create a kthread, thus must support schedule */
657 	if (map->map_type == BPF_MAP_TYPE_CPUMAP) {
658 		err = map->ops->map_update_elem(map, key, value, attr->flags);
659 		goto out;
660 	}
661 
662 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
663 	 * inside bpf map update or delete otherwise deadlocks are possible
664 	 */
665 	preempt_disable();
666 	__this_cpu_inc(bpf_prog_active);
667 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
668 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
669 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
670 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
671 		err = bpf_percpu_array_update(map, key, value, attr->flags);
672 	} else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
673 		   map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
674 		   map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY ||
675 		   map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) {
676 		rcu_read_lock();
677 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
678 						   attr->flags);
679 		rcu_read_unlock();
680 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
681 		rcu_read_lock();
682 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
683 						  attr->flags);
684 		rcu_read_unlock();
685 	} else {
686 		rcu_read_lock();
687 		err = map->ops->map_update_elem(map, key, value, attr->flags);
688 		rcu_read_unlock();
689 	}
690 	__this_cpu_dec(bpf_prog_active);
691 	preempt_enable();
692 out:
693 	if (!err)
694 		trace_bpf_map_update_elem(map, ufd, key, value);
695 free_value:
696 	kfree(value);
697 free_key:
698 	kfree(key);
699 err_put:
700 	fdput(f);
701 	return err;
702 }
703 
704 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
705 
706 static int map_delete_elem(union bpf_attr *attr)
707 {
708 	void __user *ukey = u64_to_user_ptr(attr->key);
709 	int ufd = attr->map_fd;
710 	struct bpf_map *map;
711 	struct fd f;
712 	void *key;
713 	int err;
714 
715 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
716 		return -EINVAL;
717 
718 	f = fdget(ufd);
719 	map = __bpf_map_get(f);
720 	if (IS_ERR(map))
721 		return PTR_ERR(map);
722 
723 	if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
724 		err = -EPERM;
725 		goto err_put;
726 	}
727 
728 	key = memdup_user(ukey, map->key_size);
729 	if (IS_ERR(key)) {
730 		err = PTR_ERR(key);
731 		goto err_put;
732 	}
733 
734 	preempt_disable();
735 	__this_cpu_inc(bpf_prog_active);
736 	rcu_read_lock();
737 	err = map->ops->map_delete_elem(map, key);
738 	rcu_read_unlock();
739 	__this_cpu_dec(bpf_prog_active);
740 	preempt_enable();
741 
742 	if (!err)
743 		trace_bpf_map_delete_elem(map, ufd, key);
744 	kfree(key);
745 err_put:
746 	fdput(f);
747 	return err;
748 }
749 
750 /* last field in 'union bpf_attr' used by this command */
751 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
752 
753 static int map_get_next_key(union bpf_attr *attr)
754 {
755 	void __user *ukey = u64_to_user_ptr(attr->key);
756 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
757 	int ufd = attr->map_fd;
758 	struct bpf_map *map;
759 	void *key, *next_key;
760 	struct fd f;
761 	int err;
762 
763 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
764 		return -EINVAL;
765 
766 	f = fdget(ufd);
767 	map = __bpf_map_get(f);
768 	if (IS_ERR(map))
769 		return PTR_ERR(map);
770 
771 	if (!(f.file->f_mode & FMODE_CAN_READ)) {
772 		err = -EPERM;
773 		goto err_put;
774 	}
775 
776 	if (ukey) {
777 		key = memdup_user(ukey, map->key_size);
778 		if (IS_ERR(key)) {
779 			err = PTR_ERR(key);
780 			goto err_put;
781 		}
782 	} else {
783 		key = NULL;
784 	}
785 
786 	err = -ENOMEM;
787 	next_key = kmalloc(map->key_size, GFP_USER);
788 	if (!next_key)
789 		goto free_key;
790 
791 	rcu_read_lock();
792 	err = map->ops->map_get_next_key(map, key, next_key);
793 	rcu_read_unlock();
794 	if (err)
795 		goto free_next_key;
796 
797 	err = -EFAULT;
798 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
799 		goto free_next_key;
800 
801 	trace_bpf_map_next_key(map, ufd, key, next_key);
802 	err = 0;
803 
804 free_next_key:
805 	kfree(next_key);
806 free_key:
807 	kfree(key);
808 err_put:
809 	fdput(f);
810 	return err;
811 }
812 
813 static const struct bpf_prog_ops * const bpf_prog_types[] = {
814 #define BPF_PROG_TYPE(_id, _name) \
815 	[_id] = & _name ## _prog_ops,
816 #define BPF_MAP_TYPE(_id, _ops)
817 #include <linux/bpf_types.h>
818 #undef BPF_PROG_TYPE
819 #undef BPF_MAP_TYPE
820 };
821 
822 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
823 {
824 	if (type >= ARRAY_SIZE(bpf_prog_types) || !bpf_prog_types[type])
825 		return -EINVAL;
826 
827 	prog->aux->ops = bpf_prog_types[type];
828 	prog->type = type;
829 	return 0;
830 }
831 
832 /* drop refcnt on maps used by eBPF program and free auxilary data */
833 static void free_used_maps(struct bpf_prog_aux *aux)
834 {
835 	int i;
836 
837 	for (i = 0; i < aux->used_map_cnt; i++)
838 		bpf_map_put(aux->used_maps[i]);
839 
840 	kfree(aux->used_maps);
841 }
842 
843 int __bpf_prog_charge(struct user_struct *user, u32 pages)
844 {
845 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
846 	unsigned long user_bufs;
847 
848 	if (user) {
849 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
850 		if (user_bufs > memlock_limit) {
851 			atomic_long_sub(pages, &user->locked_vm);
852 			return -EPERM;
853 		}
854 	}
855 
856 	return 0;
857 }
858 
859 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
860 {
861 	if (user)
862 		atomic_long_sub(pages, &user->locked_vm);
863 }
864 
865 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
866 {
867 	struct user_struct *user = get_current_user();
868 	int ret;
869 
870 	ret = __bpf_prog_charge(user, prog->pages);
871 	if (ret) {
872 		free_uid(user);
873 		return ret;
874 	}
875 
876 	prog->aux->user = user;
877 	return 0;
878 }
879 
880 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
881 {
882 	struct user_struct *user = prog->aux->user;
883 
884 	__bpf_prog_uncharge(user, prog->pages);
885 	free_uid(user);
886 }
887 
888 static int bpf_prog_alloc_id(struct bpf_prog *prog)
889 {
890 	int id;
891 
892 	spin_lock_bh(&prog_idr_lock);
893 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
894 	if (id > 0)
895 		prog->aux->id = id;
896 	spin_unlock_bh(&prog_idr_lock);
897 
898 	/* id is in [1, INT_MAX) */
899 	if (WARN_ON_ONCE(!id))
900 		return -ENOSPC;
901 
902 	return id > 0 ? 0 : id;
903 }
904 
905 static void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
906 {
907 	/* cBPF to eBPF migrations are currently not in the idr store. */
908 	if (!prog->aux->id)
909 		return;
910 
911 	if (do_idr_lock)
912 		spin_lock_bh(&prog_idr_lock);
913 	else
914 		__acquire(&prog_idr_lock);
915 
916 	idr_remove(&prog_idr, prog->aux->id);
917 
918 	if (do_idr_lock)
919 		spin_unlock_bh(&prog_idr_lock);
920 	else
921 		__release(&prog_idr_lock);
922 }
923 
924 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
925 {
926 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
927 
928 	free_used_maps(aux);
929 	bpf_prog_uncharge_memlock(aux->prog);
930 	security_bpf_prog_free(aux);
931 	bpf_prog_free(aux->prog);
932 }
933 
934 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
935 {
936 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
937 		trace_bpf_prog_put_rcu(prog);
938 		/* bpf_prog_free_id() must be called first */
939 		bpf_prog_free_id(prog, do_idr_lock);
940 		bpf_prog_kallsyms_del(prog);
941 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
942 	}
943 }
944 
945 void bpf_prog_put(struct bpf_prog *prog)
946 {
947 	__bpf_prog_put(prog, true);
948 }
949 EXPORT_SYMBOL_GPL(bpf_prog_put);
950 
951 static int bpf_prog_release(struct inode *inode, struct file *filp)
952 {
953 	struct bpf_prog *prog = filp->private_data;
954 
955 	bpf_prog_put(prog);
956 	return 0;
957 }
958 
959 #ifdef CONFIG_PROC_FS
960 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
961 {
962 	const struct bpf_prog *prog = filp->private_data;
963 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
964 
965 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
966 	seq_printf(m,
967 		   "prog_type:\t%u\n"
968 		   "prog_jited:\t%u\n"
969 		   "prog_tag:\t%s\n"
970 		   "memlock:\t%llu\n",
971 		   prog->type,
972 		   prog->jited,
973 		   prog_tag,
974 		   prog->pages * 1ULL << PAGE_SHIFT);
975 }
976 #endif
977 
978 const struct file_operations bpf_prog_fops = {
979 #ifdef CONFIG_PROC_FS
980 	.show_fdinfo	= bpf_prog_show_fdinfo,
981 #endif
982 	.release	= bpf_prog_release,
983 	.read		= bpf_dummy_read,
984 	.write		= bpf_dummy_write,
985 };
986 
987 int bpf_prog_new_fd(struct bpf_prog *prog)
988 {
989 	int ret;
990 
991 	ret = security_bpf_prog(prog);
992 	if (ret < 0)
993 		return ret;
994 
995 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
996 				O_RDWR | O_CLOEXEC);
997 }
998 
999 static struct bpf_prog *____bpf_prog_get(struct fd f)
1000 {
1001 	if (!f.file)
1002 		return ERR_PTR(-EBADF);
1003 	if (f.file->f_op != &bpf_prog_fops) {
1004 		fdput(f);
1005 		return ERR_PTR(-EINVAL);
1006 	}
1007 
1008 	return f.file->private_data;
1009 }
1010 
1011 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1012 {
1013 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1014 		atomic_sub(i, &prog->aux->refcnt);
1015 		return ERR_PTR(-EBUSY);
1016 	}
1017 	return prog;
1018 }
1019 EXPORT_SYMBOL_GPL(bpf_prog_add);
1020 
1021 void bpf_prog_sub(struct bpf_prog *prog, int i)
1022 {
1023 	/* Only to be used for undoing previous bpf_prog_add() in some
1024 	 * error path. We still know that another entity in our call
1025 	 * path holds a reference to the program, thus atomic_sub() can
1026 	 * be safely used in such cases!
1027 	 */
1028 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1029 }
1030 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1031 
1032 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1033 {
1034 	return bpf_prog_add(prog, 1);
1035 }
1036 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1037 
1038 /* prog_idr_lock should have been held */
1039 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1040 {
1041 	int refold;
1042 
1043 	refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
1044 
1045 	if (refold >= BPF_MAX_REFCNT) {
1046 		__bpf_prog_put(prog, false);
1047 		return ERR_PTR(-EBUSY);
1048 	}
1049 
1050 	if (!refold)
1051 		return ERR_PTR(-ENOENT);
1052 
1053 	return prog;
1054 }
1055 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1056 
1057 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type)
1058 {
1059 	struct fd f = fdget(ufd);
1060 	struct bpf_prog *prog;
1061 
1062 	prog = ____bpf_prog_get(f);
1063 	if (IS_ERR(prog))
1064 		return prog;
1065 	if (type && prog->type != *type) {
1066 		prog = ERR_PTR(-EINVAL);
1067 		goto out;
1068 	}
1069 
1070 	prog = bpf_prog_inc(prog);
1071 out:
1072 	fdput(f);
1073 	return prog;
1074 }
1075 
1076 struct bpf_prog *bpf_prog_get(u32 ufd)
1077 {
1078 	return __bpf_prog_get(ufd, NULL);
1079 }
1080 
1081 struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type)
1082 {
1083 	struct bpf_prog *prog = __bpf_prog_get(ufd, &type);
1084 
1085 	if (!IS_ERR(prog))
1086 		trace_bpf_prog_get_type(prog);
1087 	return prog;
1088 }
1089 EXPORT_SYMBOL_GPL(bpf_prog_get_type);
1090 
1091 /* last field in 'union bpf_attr' used by this command */
1092 #define	BPF_PROG_LOAD_LAST_FIELD prog_name
1093 
1094 static int bpf_prog_load(union bpf_attr *attr)
1095 {
1096 	enum bpf_prog_type type = attr->prog_type;
1097 	struct bpf_prog *prog;
1098 	int err;
1099 	char license[128];
1100 	bool is_gpl;
1101 
1102 	if (CHECK_ATTR(BPF_PROG_LOAD))
1103 		return -EINVAL;
1104 
1105 	if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1106 		return -EINVAL;
1107 
1108 	/* copy eBPF program license from user space */
1109 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1110 			      sizeof(license) - 1) < 0)
1111 		return -EFAULT;
1112 	license[sizeof(license) - 1] = 0;
1113 
1114 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1115 	is_gpl = license_is_gpl_compatible(license);
1116 
1117 	if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1118 		return -E2BIG;
1119 
1120 	if (type == BPF_PROG_TYPE_KPROBE &&
1121 	    attr->kern_version != LINUX_VERSION_CODE)
1122 		return -EINVAL;
1123 
1124 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1125 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
1126 	    !capable(CAP_SYS_ADMIN))
1127 		return -EPERM;
1128 
1129 	/* plain bpf_prog allocation */
1130 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1131 	if (!prog)
1132 		return -ENOMEM;
1133 
1134 	err = security_bpf_prog_alloc(prog->aux);
1135 	if (err)
1136 		goto free_prog_nouncharge;
1137 
1138 	err = bpf_prog_charge_memlock(prog);
1139 	if (err)
1140 		goto free_prog_sec;
1141 
1142 	prog->len = attr->insn_cnt;
1143 
1144 	err = -EFAULT;
1145 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1146 			   bpf_prog_insn_size(prog)) != 0)
1147 		goto free_prog;
1148 
1149 	prog->orig_prog = NULL;
1150 	prog->jited = 0;
1151 
1152 	atomic_set(&prog->aux->refcnt, 1);
1153 	prog->gpl_compatible = is_gpl ? 1 : 0;
1154 
1155 	/* find program type: socket_filter vs tracing_filter */
1156 	err = find_prog_type(type, prog);
1157 	if (err < 0)
1158 		goto free_prog;
1159 
1160 	prog->aux->load_time = ktime_get_boot_ns();
1161 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1162 	if (err)
1163 		goto free_prog;
1164 
1165 	/* run eBPF verifier */
1166 	err = bpf_check(&prog, attr);
1167 	if (err < 0)
1168 		goto free_used_maps;
1169 
1170 	/* eBPF program is ready to be JITed */
1171 	prog = bpf_prog_select_runtime(prog, &err);
1172 	if (err < 0)
1173 		goto free_used_maps;
1174 
1175 	err = bpf_prog_alloc_id(prog);
1176 	if (err)
1177 		goto free_used_maps;
1178 
1179 	err = bpf_prog_new_fd(prog);
1180 	if (err < 0) {
1181 		/* failed to allocate fd.
1182 		 * bpf_prog_put() is needed because the above
1183 		 * bpf_prog_alloc_id() has published the prog
1184 		 * to the userspace and the userspace may
1185 		 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1186 		 */
1187 		bpf_prog_put(prog);
1188 		return err;
1189 	}
1190 
1191 	bpf_prog_kallsyms_add(prog);
1192 	trace_bpf_prog_load(prog, err);
1193 	return err;
1194 
1195 free_used_maps:
1196 	free_used_maps(prog->aux);
1197 free_prog:
1198 	bpf_prog_uncharge_memlock(prog);
1199 free_prog_sec:
1200 	security_bpf_prog_free(prog->aux);
1201 free_prog_nouncharge:
1202 	bpf_prog_free(prog);
1203 	return err;
1204 }
1205 
1206 #define BPF_OBJ_LAST_FIELD file_flags
1207 
1208 static int bpf_obj_pin(const union bpf_attr *attr)
1209 {
1210 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1211 		return -EINVAL;
1212 
1213 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1214 }
1215 
1216 static int bpf_obj_get(const union bpf_attr *attr)
1217 {
1218 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1219 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1220 		return -EINVAL;
1221 
1222 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1223 				attr->file_flags);
1224 }
1225 
1226 #ifdef CONFIG_CGROUP_BPF
1227 
1228 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1229 
1230 static int sockmap_get_from_fd(const union bpf_attr *attr, bool attach)
1231 {
1232 	struct bpf_prog *prog = NULL;
1233 	int ufd = attr->target_fd;
1234 	struct bpf_map *map;
1235 	struct fd f;
1236 	int err;
1237 
1238 	f = fdget(ufd);
1239 	map = __bpf_map_get(f);
1240 	if (IS_ERR(map))
1241 		return PTR_ERR(map);
1242 
1243 	if (attach) {
1244 		prog = bpf_prog_get_type(attr->attach_bpf_fd,
1245 					 BPF_PROG_TYPE_SK_SKB);
1246 		if (IS_ERR(prog)) {
1247 			fdput(f);
1248 			return PTR_ERR(prog);
1249 		}
1250 	}
1251 
1252 	err = sock_map_prog(map, prog, attr->attach_type);
1253 	if (err) {
1254 		fdput(f);
1255 		if (prog)
1256 			bpf_prog_put(prog);
1257 		return err;
1258 	}
1259 
1260 	fdput(f);
1261 	return 0;
1262 }
1263 
1264 #define BPF_F_ATTACH_MASK \
1265 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1266 
1267 static int bpf_prog_attach(const union bpf_attr *attr)
1268 {
1269 	enum bpf_prog_type ptype;
1270 	struct bpf_prog *prog;
1271 	struct cgroup *cgrp;
1272 	int ret;
1273 
1274 	if (!capable(CAP_NET_ADMIN))
1275 		return -EPERM;
1276 
1277 	if (CHECK_ATTR(BPF_PROG_ATTACH))
1278 		return -EINVAL;
1279 
1280 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1281 		return -EINVAL;
1282 
1283 	switch (attr->attach_type) {
1284 	case BPF_CGROUP_INET_INGRESS:
1285 	case BPF_CGROUP_INET_EGRESS:
1286 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1287 		break;
1288 	case BPF_CGROUP_INET_SOCK_CREATE:
1289 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1290 		break;
1291 	case BPF_CGROUP_SOCK_OPS:
1292 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1293 		break;
1294 	case BPF_SK_SKB_STREAM_PARSER:
1295 	case BPF_SK_SKB_STREAM_VERDICT:
1296 		return sockmap_get_from_fd(attr, true);
1297 	default:
1298 		return -EINVAL;
1299 	}
1300 
1301 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1302 	if (IS_ERR(prog))
1303 		return PTR_ERR(prog);
1304 
1305 	cgrp = cgroup_get_from_fd(attr->target_fd);
1306 	if (IS_ERR(cgrp)) {
1307 		bpf_prog_put(prog);
1308 		return PTR_ERR(cgrp);
1309 	}
1310 
1311 	ret = cgroup_bpf_attach(cgrp, prog, attr->attach_type,
1312 				attr->attach_flags);
1313 	if (ret)
1314 		bpf_prog_put(prog);
1315 	cgroup_put(cgrp);
1316 
1317 	return ret;
1318 }
1319 
1320 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1321 
1322 static int bpf_prog_detach(const union bpf_attr *attr)
1323 {
1324 	enum bpf_prog_type ptype;
1325 	struct bpf_prog *prog;
1326 	struct cgroup *cgrp;
1327 	int ret;
1328 
1329 	if (!capable(CAP_NET_ADMIN))
1330 		return -EPERM;
1331 
1332 	if (CHECK_ATTR(BPF_PROG_DETACH))
1333 		return -EINVAL;
1334 
1335 	switch (attr->attach_type) {
1336 	case BPF_CGROUP_INET_INGRESS:
1337 	case BPF_CGROUP_INET_EGRESS:
1338 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1339 		break;
1340 	case BPF_CGROUP_INET_SOCK_CREATE:
1341 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1342 		break;
1343 	case BPF_CGROUP_SOCK_OPS:
1344 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1345 		break;
1346 	case BPF_SK_SKB_STREAM_PARSER:
1347 	case BPF_SK_SKB_STREAM_VERDICT:
1348 		return sockmap_get_from_fd(attr, false);
1349 	default:
1350 		return -EINVAL;
1351 	}
1352 
1353 	cgrp = cgroup_get_from_fd(attr->target_fd);
1354 	if (IS_ERR(cgrp))
1355 		return PTR_ERR(cgrp);
1356 
1357 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1358 	if (IS_ERR(prog))
1359 		prog = NULL;
1360 
1361 	ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type, 0);
1362 	if (prog)
1363 		bpf_prog_put(prog);
1364 	cgroup_put(cgrp);
1365 	return ret;
1366 }
1367 
1368 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1369 
1370 static int bpf_prog_query(const union bpf_attr *attr,
1371 			  union bpf_attr __user *uattr)
1372 {
1373 	struct cgroup *cgrp;
1374 	int ret;
1375 
1376 	if (!capable(CAP_NET_ADMIN))
1377 		return -EPERM;
1378 	if (CHECK_ATTR(BPF_PROG_QUERY))
1379 		return -EINVAL;
1380 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1381 		return -EINVAL;
1382 
1383 	switch (attr->query.attach_type) {
1384 	case BPF_CGROUP_INET_INGRESS:
1385 	case BPF_CGROUP_INET_EGRESS:
1386 	case BPF_CGROUP_INET_SOCK_CREATE:
1387 	case BPF_CGROUP_SOCK_OPS:
1388 		break;
1389 	default:
1390 		return -EINVAL;
1391 	}
1392 	cgrp = cgroup_get_from_fd(attr->query.target_fd);
1393 	if (IS_ERR(cgrp))
1394 		return PTR_ERR(cgrp);
1395 	ret = cgroup_bpf_query(cgrp, attr, uattr);
1396 	cgroup_put(cgrp);
1397 	return ret;
1398 }
1399 #endif /* CONFIG_CGROUP_BPF */
1400 
1401 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1402 
1403 static int bpf_prog_test_run(const union bpf_attr *attr,
1404 			     union bpf_attr __user *uattr)
1405 {
1406 	struct bpf_prog *prog;
1407 	int ret = -ENOTSUPP;
1408 
1409 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1410 		return -EINVAL;
1411 
1412 	prog = bpf_prog_get(attr->test.prog_fd);
1413 	if (IS_ERR(prog))
1414 		return PTR_ERR(prog);
1415 
1416 	if (prog->aux->ops->test_run)
1417 		ret = prog->aux->ops->test_run(prog, attr, uattr);
1418 
1419 	bpf_prog_put(prog);
1420 	return ret;
1421 }
1422 
1423 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1424 
1425 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1426 			       union bpf_attr __user *uattr,
1427 			       struct idr *idr,
1428 			       spinlock_t *lock)
1429 {
1430 	u32 next_id = attr->start_id;
1431 	int err = 0;
1432 
1433 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1434 		return -EINVAL;
1435 
1436 	if (!capable(CAP_SYS_ADMIN))
1437 		return -EPERM;
1438 
1439 	next_id++;
1440 	spin_lock_bh(lock);
1441 	if (!idr_get_next(idr, &next_id))
1442 		err = -ENOENT;
1443 	spin_unlock_bh(lock);
1444 
1445 	if (!err)
1446 		err = put_user(next_id, &uattr->next_id);
1447 
1448 	return err;
1449 }
1450 
1451 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1452 
1453 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1454 {
1455 	struct bpf_prog *prog;
1456 	u32 id = attr->prog_id;
1457 	int fd;
1458 
1459 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1460 		return -EINVAL;
1461 
1462 	if (!capable(CAP_SYS_ADMIN))
1463 		return -EPERM;
1464 
1465 	spin_lock_bh(&prog_idr_lock);
1466 	prog = idr_find(&prog_idr, id);
1467 	if (prog)
1468 		prog = bpf_prog_inc_not_zero(prog);
1469 	else
1470 		prog = ERR_PTR(-ENOENT);
1471 	spin_unlock_bh(&prog_idr_lock);
1472 
1473 	if (IS_ERR(prog))
1474 		return PTR_ERR(prog);
1475 
1476 	fd = bpf_prog_new_fd(prog);
1477 	if (fd < 0)
1478 		bpf_prog_put(prog);
1479 
1480 	return fd;
1481 }
1482 
1483 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1484 
1485 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1486 {
1487 	struct bpf_map *map;
1488 	u32 id = attr->map_id;
1489 	int f_flags;
1490 	int fd;
1491 
1492 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1493 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1494 		return -EINVAL;
1495 
1496 	if (!capable(CAP_SYS_ADMIN))
1497 		return -EPERM;
1498 
1499 	f_flags = bpf_get_file_flag(attr->open_flags);
1500 	if (f_flags < 0)
1501 		return f_flags;
1502 
1503 	spin_lock_bh(&map_idr_lock);
1504 	map = idr_find(&map_idr, id);
1505 	if (map)
1506 		map = bpf_map_inc_not_zero(map, true);
1507 	else
1508 		map = ERR_PTR(-ENOENT);
1509 	spin_unlock_bh(&map_idr_lock);
1510 
1511 	if (IS_ERR(map))
1512 		return PTR_ERR(map);
1513 
1514 	fd = bpf_map_new_fd(map, f_flags);
1515 	if (fd < 0)
1516 		bpf_map_put(map);
1517 
1518 	return fd;
1519 }
1520 
1521 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1522 				   const union bpf_attr *attr,
1523 				   union bpf_attr __user *uattr)
1524 {
1525 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1526 	struct bpf_prog_info info = {};
1527 	u32 info_len = attr->info.info_len;
1528 	char __user *uinsns;
1529 	u32 ulen;
1530 	int err;
1531 
1532 	err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1533 	if (err)
1534 		return err;
1535 	info_len = min_t(u32, sizeof(info), info_len);
1536 
1537 	if (copy_from_user(&info, uinfo, info_len))
1538 		return -EFAULT;
1539 
1540 	info.type = prog->type;
1541 	info.id = prog->aux->id;
1542 	info.load_time = prog->aux->load_time;
1543 	info.created_by_uid = from_kuid_munged(current_user_ns(),
1544 					       prog->aux->user->uid);
1545 
1546 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
1547 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1548 
1549 	ulen = info.nr_map_ids;
1550 	info.nr_map_ids = prog->aux->used_map_cnt;
1551 	ulen = min_t(u32, info.nr_map_ids, ulen);
1552 	if (ulen) {
1553 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1554 		u32 i;
1555 
1556 		for (i = 0; i < ulen; i++)
1557 			if (put_user(prog->aux->used_maps[i]->id,
1558 				     &user_map_ids[i]))
1559 				return -EFAULT;
1560 	}
1561 
1562 	if (!capable(CAP_SYS_ADMIN)) {
1563 		info.jited_prog_len = 0;
1564 		info.xlated_prog_len = 0;
1565 		goto done;
1566 	}
1567 
1568 	ulen = info.jited_prog_len;
1569 	info.jited_prog_len = prog->jited_len;
1570 	if (info.jited_prog_len && ulen) {
1571 		uinsns = u64_to_user_ptr(info.jited_prog_insns);
1572 		ulen = min_t(u32, info.jited_prog_len, ulen);
1573 		if (copy_to_user(uinsns, prog->bpf_func, ulen))
1574 			return -EFAULT;
1575 	}
1576 
1577 	ulen = info.xlated_prog_len;
1578 	info.xlated_prog_len = bpf_prog_insn_size(prog);
1579 	if (info.xlated_prog_len && ulen) {
1580 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1581 		ulen = min_t(u32, info.xlated_prog_len, ulen);
1582 		if (copy_to_user(uinsns, prog->insnsi, ulen))
1583 			return -EFAULT;
1584 	}
1585 
1586 done:
1587 	if (copy_to_user(uinfo, &info, info_len) ||
1588 	    put_user(info_len, &uattr->info.info_len))
1589 		return -EFAULT;
1590 
1591 	return 0;
1592 }
1593 
1594 static int bpf_map_get_info_by_fd(struct bpf_map *map,
1595 				  const union bpf_attr *attr,
1596 				  union bpf_attr __user *uattr)
1597 {
1598 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1599 	struct bpf_map_info info = {};
1600 	u32 info_len = attr->info.info_len;
1601 	int err;
1602 
1603 	err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1604 	if (err)
1605 		return err;
1606 	info_len = min_t(u32, sizeof(info), info_len);
1607 
1608 	info.type = map->map_type;
1609 	info.id = map->id;
1610 	info.key_size = map->key_size;
1611 	info.value_size = map->value_size;
1612 	info.max_entries = map->max_entries;
1613 	info.map_flags = map->map_flags;
1614 	memcpy(info.name, map->name, sizeof(map->name));
1615 
1616 	if (copy_to_user(uinfo, &info, info_len) ||
1617 	    put_user(info_len, &uattr->info.info_len))
1618 		return -EFAULT;
1619 
1620 	return 0;
1621 }
1622 
1623 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
1624 
1625 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
1626 				  union bpf_attr __user *uattr)
1627 {
1628 	int ufd = attr->info.bpf_fd;
1629 	struct fd f;
1630 	int err;
1631 
1632 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
1633 		return -EINVAL;
1634 
1635 	f = fdget(ufd);
1636 	if (!f.file)
1637 		return -EBADFD;
1638 
1639 	if (f.file->f_op == &bpf_prog_fops)
1640 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
1641 					      uattr);
1642 	else if (f.file->f_op == &bpf_map_fops)
1643 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
1644 					     uattr);
1645 	else
1646 		err = -EINVAL;
1647 
1648 	fdput(f);
1649 	return err;
1650 }
1651 
1652 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
1653 {
1654 	union bpf_attr attr = {};
1655 	int err;
1656 
1657 	if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
1658 		return -EPERM;
1659 
1660 	err = check_uarg_tail_zero(uattr, sizeof(attr), size);
1661 	if (err)
1662 		return err;
1663 	size = min_t(u32, size, sizeof(attr));
1664 
1665 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
1666 	if (copy_from_user(&attr, uattr, size) != 0)
1667 		return -EFAULT;
1668 
1669 	err = security_bpf(cmd, &attr, size);
1670 	if (err < 0)
1671 		return err;
1672 
1673 	switch (cmd) {
1674 	case BPF_MAP_CREATE:
1675 		err = map_create(&attr);
1676 		break;
1677 	case BPF_MAP_LOOKUP_ELEM:
1678 		err = map_lookup_elem(&attr);
1679 		break;
1680 	case BPF_MAP_UPDATE_ELEM:
1681 		err = map_update_elem(&attr);
1682 		break;
1683 	case BPF_MAP_DELETE_ELEM:
1684 		err = map_delete_elem(&attr);
1685 		break;
1686 	case BPF_MAP_GET_NEXT_KEY:
1687 		err = map_get_next_key(&attr);
1688 		break;
1689 	case BPF_PROG_LOAD:
1690 		err = bpf_prog_load(&attr);
1691 		break;
1692 	case BPF_OBJ_PIN:
1693 		err = bpf_obj_pin(&attr);
1694 		break;
1695 	case BPF_OBJ_GET:
1696 		err = bpf_obj_get(&attr);
1697 		break;
1698 #ifdef CONFIG_CGROUP_BPF
1699 	case BPF_PROG_ATTACH:
1700 		err = bpf_prog_attach(&attr);
1701 		break;
1702 	case BPF_PROG_DETACH:
1703 		err = bpf_prog_detach(&attr);
1704 		break;
1705 	case BPF_PROG_QUERY:
1706 		err = bpf_prog_query(&attr, uattr);
1707 		break;
1708 #endif
1709 	case BPF_PROG_TEST_RUN:
1710 		err = bpf_prog_test_run(&attr, uattr);
1711 		break;
1712 	case BPF_PROG_GET_NEXT_ID:
1713 		err = bpf_obj_get_next_id(&attr, uattr,
1714 					  &prog_idr, &prog_idr_lock);
1715 		break;
1716 	case BPF_MAP_GET_NEXT_ID:
1717 		err = bpf_obj_get_next_id(&attr, uattr,
1718 					  &map_idr, &map_idr_lock);
1719 		break;
1720 	case BPF_PROG_GET_FD_BY_ID:
1721 		err = bpf_prog_get_fd_by_id(&attr);
1722 		break;
1723 	case BPF_MAP_GET_FD_BY_ID:
1724 		err = bpf_map_get_fd_by_id(&attr);
1725 		break;
1726 	case BPF_OBJ_GET_INFO_BY_FD:
1727 		err = bpf_obj_get_info_by_fd(&attr, uattr);
1728 		break;
1729 	default:
1730 		err = -EINVAL;
1731 		break;
1732 	}
1733 
1734 	return err;
1735 }
1736