xref: /linux/kernel/bpf/syscall.c (revision 1c2dd16add7ebd7fc6a47f21a42f502cfed5ec2f)
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 
26 DEFINE_PER_CPU(int, bpf_prog_active);
27 
28 int sysctl_unprivileged_bpf_disabled __read_mostly;
29 
30 static const struct bpf_map_ops * const bpf_map_types[] = {
31 #define BPF_PROG_TYPE(_id, _ops)
32 #define BPF_MAP_TYPE(_id, _ops) \
33 	[_id] = &_ops,
34 #include <linux/bpf_types.h>
35 #undef BPF_PROG_TYPE
36 #undef BPF_MAP_TYPE
37 };
38 
39 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
40 {
41 	struct bpf_map *map;
42 
43 	if (attr->map_type >= ARRAY_SIZE(bpf_map_types) ||
44 	    !bpf_map_types[attr->map_type])
45 		return ERR_PTR(-EINVAL);
46 
47 	map = bpf_map_types[attr->map_type]->map_alloc(attr);
48 	if (IS_ERR(map))
49 		return map;
50 	map->ops = bpf_map_types[attr->map_type];
51 	map->map_type = attr->map_type;
52 	return map;
53 }
54 
55 void *bpf_map_area_alloc(size_t size)
56 {
57 	/* We definitely need __GFP_NORETRY, so OOM killer doesn't
58 	 * trigger under memory pressure as we really just want to
59 	 * fail instead.
60 	 */
61 	const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
62 	void *area;
63 
64 	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
65 		area = kmalloc(size, GFP_USER | flags);
66 		if (area != NULL)
67 			return area;
68 	}
69 
70 	return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | flags,
71 			 PAGE_KERNEL);
72 }
73 
74 void bpf_map_area_free(void *area)
75 {
76 	kvfree(area);
77 }
78 
79 int bpf_map_precharge_memlock(u32 pages)
80 {
81 	struct user_struct *user = get_current_user();
82 	unsigned long memlock_limit, cur;
83 
84 	memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
85 	cur = atomic_long_read(&user->locked_vm);
86 	free_uid(user);
87 	if (cur + pages > memlock_limit)
88 		return -EPERM;
89 	return 0;
90 }
91 
92 static int bpf_map_charge_memlock(struct bpf_map *map)
93 {
94 	struct user_struct *user = get_current_user();
95 	unsigned long memlock_limit;
96 
97 	memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
98 
99 	atomic_long_add(map->pages, &user->locked_vm);
100 
101 	if (atomic_long_read(&user->locked_vm) > memlock_limit) {
102 		atomic_long_sub(map->pages, &user->locked_vm);
103 		free_uid(user);
104 		return -EPERM;
105 	}
106 	map->user = user;
107 	return 0;
108 }
109 
110 static void bpf_map_uncharge_memlock(struct bpf_map *map)
111 {
112 	struct user_struct *user = map->user;
113 
114 	atomic_long_sub(map->pages, &user->locked_vm);
115 	free_uid(user);
116 }
117 
118 /* called from workqueue */
119 static void bpf_map_free_deferred(struct work_struct *work)
120 {
121 	struct bpf_map *map = container_of(work, struct bpf_map, work);
122 
123 	bpf_map_uncharge_memlock(map);
124 	/* implementation dependent freeing */
125 	map->ops->map_free(map);
126 }
127 
128 static void bpf_map_put_uref(struct bpf_map *map)
129 {
130 	if (atomic_dec_and_test(&map->usercnt)) {
131 		if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
132 			bpf_fd_array_map_clear(map);
133 	}
134 }
135 
136 /* decrement map refcnt and schedule it for freeing via workqueue
137  * (unrelying map implementation ops->map_free() might sleep)
138  */
139 void bpf_map_put(struct bpf_map *map)
140 {
141 	if (atomic_dec_and_test(&map->refcnt)) {
142 		INIT_WORK(&map->work, bpf_map_free_deferred);
143 		schedule_work(&map->work);
144 	}
145 }
146 
147 void bpf_map_put_with_uref(struct bpf_map *map)
148 {
149 	bpf_map_put_uref(map);
150 	bpf_map_put(map);
151 }
152 
153 static int bpf_map_release(struct inode *inode, struct file *filp)
154 {
155 	struct bpf_map *map = filp->private_data;
156 
157 	if (map->ops->map_release)
158 		map->ops->map_release(map, filp);
159 
160 	bpf_map_put_with_uref(map);
161 	return 0;
162 }
163 
164 #ifdef CONFIG_PROC_FS
165 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
166 {
167 	const struct bpf_map *map = filp->private_data;
168 	const struct bpf_array *array;
169 	u32 owner_prog_type = 0;
170 
171 	if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
172 		array = container_of(map, struct bpf_array, map);
173 		owner_prog_type = array->owner_prog_type;
174 	}
175 
176 	seq_printf(m,
177 		   "map_type:\t%u\n"
178 		   "key_size:\t%u\n"
179 		   "value_size:\t%u\n"
180 		   "max_entries:\t%u\n"
181 		   "map_flags:\t%#x\n"
182 		   "memlock:\t%llu\n",
183 		   map->map_type,
184 		   map->key_size,
185 		   map->value_size,
186 		   map->max_entries,
187 		   map->map_flags,
188 		   map->pages * 1ULL << PAGE_SHIFT);
189 
190 	if (owner_prog_type)
191 		seq_printf(m, "owner_prog_type:\t%u\n",
192 			   owner_prog_type);
193 }
194 #endif
195 
196 static const struct file_operations bpf_map_fops = {
197 #ifdef CONFIG_PROC_FS
198 	.show_fdinfo	= bpf_map_show_fdinfo,
199 #endif
200 	.release	= bpf_map_release,
201 };
202 
203 int bpf_map_new_fd(struct bpf_map *map)
204 {
205 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
206 				O_RDWR | O_CLOEXEC);
207 }
208 
209 /* helper macro to check that unused fields 'union bpf_attr' are zero */
210 #define CHECK_ATTR(CMD) \
211 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
212 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
213 		   sizeof(*attr) - \
214 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
215 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
216 
217 #define BPF_MAP_CREATE_LAST_FIELD inner_map_fd
218 /* called via syscall */
219 static int map_create(union bpf_attr *attr)
220 {
221 	struct bpf_map *map;
222 	int err;
223 
224 	err = CHECK_ATTR(BPF_MAP_CREATE);
225 	if (err)
226 		return -EINVAL;
227 
228 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
229 	map = find_and_alloc_map(attr);
230 	if (IS_ERR(map))
231 		return PTR_ERR(map);
232 
233 	atomic_set(&map->refcnt, 1);
234 	atomic_set(&map->usercnt, 1);
235 
236 	err = bpf_map_charge_memlock(map);
237 	if (err)
238 		goto free_map_nouncharge;
239 
240 	err = bpf_map_new_fd(map);
241 	if (err < 0)
242 		/* failed to allocate fd */
243 		goto free_map;
244 
245 	trace_bpf_map_create(map, err);
246 	return err;
247 
248 free_map:
249 	bpf_map_uncharge_memlock(map);
250 free_map_nouncharge:
251 	map->ops->map_free(map);
252 	return err;
253 }
254 
255 /* if error is returned, fd is released.
256  * On success caller should complete fd access with matching fdput()
257  */
258 struct bpf_map *__bpf_map_get(struct fd f)
259 {
260 	if (!f.file)
261 		return ERR_PTR(-EBADF);
262 	if (f.file->f_op != &bpf_map_fops) {
263 		fdput(f);
264 		return ERR_PTR(-EINVAL);
265 	}
266 
267 	return f.file->private_data;
268 }
269 
270 /* prog's and map's refcnt limit */
271 #define BPF_MAX_REFCNT 32768
272 
273 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
274 {
275 	if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
276 		atomic_dec(&map->refcnt);
277 		return ERR_PTR(-EBUSY);
278 	}
279 	if (uref)
280 		atomic_inc(&map->usercnt);
281 	return map;
282 }
283 
284 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
285 {
286 	struct fd f = fdget(ufd);
287 	struct bpf_map *map;
288 
289 	map = __bpf_map_get(f);
290 	if (IS_ERR(map))
291 		return map;
292 
293 	map = bpf_map_inc(map, true);
294 	fdput(f);
295 
296 	return map;
297 }
298 
299 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
300 {
301 	return -ENOTSUPP;
302 }
303 
304 /* last field in 'union bpf_attr' used by this command */
305 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
306 
307 static int map_lookup_elem(union bpf_attr *attr)
308 {
309 	void __user *ukey = u64_to_user_ptr(attr->key);
310 	void __user *uvalue = u64_to_user_ptr(attr->value);
311 	int ufd = attr->map_fd;
312 	struct bpf_map *map;
313 	void *key, *value, *ptr;
314 	u32 value_size;
315 	struct fd f;
316 	int err;
317 
318 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
319 		return -EINVAL;
320 
321 	f = fdget(ufd);
322 	map = __bpf_map_get(f);
323 	if (IS_ERR(map))
324 		return PTR_ERR(map);
325 
326 	err = -ENOMEM;
327 	key = kmalloc(map->key_size, GFP_USER);
328 	if (!key)
329 		goto err_put;
330 
331 	err = -EFAULT;
332 	if (copy_from_user(key, ukey, map->key_size) != 0)
333 		goto free_key;
334 
335 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
336 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
337 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
338 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
339 	else
340 		value_size = map->value_size;
341 
342 	err = -ENOMEM;
343 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
344 	if (!value)
345 		goto free_key;
346 
347 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
348 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
349 		err = bpf_percpu_hash_copy(map, key, value);
350 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
351 		err = bpf_percpu_array_copy(map, key, value);
352 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
353 		err = bpf_stackmap_copy(map, key, value);
354 	} else if (map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
355 		   map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
356 		err = -ENOTSUPP;
357 	} else {
358 		rcu_read_lock();
359 		ptr = map->ops->map_lookup_elem(map, key);
360 		if (ptr)
361 			memcpy(value, ptr, value_size);
362 		rcu_read_unlock();
363 		err = ptr ? 0 : -ENOENT;
364 	}
365 
366 	if (err)
367 		goto free_value;
368 
369 	err = -EFAULT;
370 	if (copy_to_user(uvalue, value, value_size) != 0)
371 		goto free_value;
372 
373 	trace_bpf_map_lookup_elem(map, ufd, key, value);
374 	err = 0;
375 
376 free_value:
377 	kfree(value);
378 free_key:
379 	kfree(key);
380 err_put:
381 	fdput(f);
382 	return err;
383 }
384 
385 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
386 
387 static int map_update_elem(union bpf_attr *attr)
388 {
389 	void __user *ukey = u64_to_user_ptr(attr->key);
390 	void __user *uvalue = u64_to_user_ptr(attr->value);
391 	int ufd = attr->map_fd;
392 	struct bpf_map *map;
393 	void *key, *value;
394 	u32 value_size;
395 	struct fd f;
396 	int err;
397 
398 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
399 		return -EINVAL;
400 
401 	f = fdget(ufd);
402 	map = __bpf_map_get(f);
403 	if (IS_ERR(map))
404 		return PTR_ERR(map);
405 
406 	err = -ENOMEM;
407 	key = kmalloc(map->key_size, GFP_USER);
408 	if (!key)
409 		goto err_put;
410 
411 	err = -EFAULT;
412 	if (copy_from_user(key, ukey, map->key_size) != 0)
413 		goto free_key;
414 
415 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
416 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
417 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
418 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
419 	else
420 		value_size = map->value_size;
421 
422 	err = -ENOMEM;
423 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
424 	if (!value)
425 		goto free_key;
426 
427 	err = -EFAULT;
428 	if (copy_from_user(value, uvalue, value_size) != 0)
429 		goto free_value;
430 
431 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
432 	 * inside bpf map update or delete otherwise deadlocks are possible
433 	 */
434 	preempt_disable();
435 	__this_cpu_inc(bpf_prog_active);
436 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
437 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
438 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
439 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
440 		err = bpf_percpu_array_update(map, key, value, attr->flags);
441 	} else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
442 		   map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
443 		   map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY ||
444 		   map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) {
445 		rcu_read_lock();
446 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
447 						   attr->flags);
448 		rcu_read_unlock();
449 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
450 		rcu_read_lock();
451 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
452 						  attr->flags);
453 		rcu_read_unlock();
454 	} else {
455 		rcu_read_lock();
456 		err = map->ops->map_update_elem(map, key, value, attr->flags);
457 		rcu_read_unlock();
458 	}
459 	__this_cpu_dec(bpf_prog_active);
460 	preempt_enable();
461 
462 	if (!err)
463 		trace_bpf_map_update_elem(map, ufd, key, value);
464 free_value:
465 	kfree(value);
466 free_key:
467 	kfree(key);
468 err_put:
469 	fdput(f);
470 	return err;
471 }
472 
473 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
474 
475 static int map_delete_elem(union bpf_attr *attr)
476 {
477 	void __user *ukey = u64_to_user_ptr(attr->key);
478 	int ufd = attr->map_fd;
479 	struct bpf_map *map;
480 	struct fd f;
481 	void *key;
482 	int err;
483 
484 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
485 		return -EINVAL;
486 
487 	f = fdget(ufd);
488 	map = __bpf_map_get(f);
489 	if (IS_ERR(map))
490 		return PTR_ERR(map);
491 
492 	err = -ENOMEM;
493 	key = kmalloc(map->key_size, GFP_USER);
494 	if (!key)
495 		goto err_put;
496 
497 	err = -EFAULT;
498 	if (copy_from_user(key, ukey, map->key_size) != 0)
499 		goto free_key;
500 
501 	preempt_disable();
502 	__this_cpu_inc(bpf_prog_active);
503 	rcu_read_lock();
504 	err = map->ops->map_delete_elem(map, key);
505 	rcu_read_unlock();
506 	__this_cpu_dec(bpf_prog_active);
507 	preempt_enable();
508 
509 	if (!err)
510 		trace_bpf_map_delete_elem(map, ufd, key);
511 free_key:
512 	kfree(key);
513 err_put:
514 	fdput(f);
515 	return err;
516 }
517 
518 /* last field in 'union bpf_attr' used by this command */
519 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
520 
521 static int map_get_next_key(union bpf_attr *attr)
522 {
523 	void __user *ukey = u64_to_user_ptr(attr->key);
524 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
525 	int ufd = attr->map_fd;
526 	struct bpf_map *map;
527 	void *key, *next_key;
528 	struct fd f;
529 	int err;
530 
531 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
532 		return -EINVAL;
533 
534 	f = fdget(ufd);
535 	map = __bpf_map_get(f);
536 	if (IS_ERR(map))
537 		return PTR_ERR(map);
538 
539 	if (ukey) {
540 		err = -ENOMEM;
541 		key = kmalloc(map->key_size, GFP_USER);
542 		if (!key)
543 			goto err_put;
544 
545 		err = -EFAULT;
546 		if (copy_from_user(key, ukey, map->key_size) != 0)
547 			goto free_key;
548 	} else {
549 		key = NULL;
550 	}
551 
552 	err = -ENOMEM;
553 	next_key = kmalloc(map->key_size, GFP_USER);
554 	if (!next_key)
555 		goto free_key;
556 
557 	rcu_read_lock();
558 	err = map->ops->map_get_next_key(map, key, next_key);
559 	rcu_read_unlock();
560 	if (err)
561 		goto free_next_key;
562 
563 	err = -EFAULT;
564 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
565 		goto free_next_key;
566 
567 	trace_bpf_map_next_key(map, ufd, key, next_key);
568 	err = 0;
569 
570 free_next_key:
571 	kfree(next_key);
572 free_key:
573 	kfree(key);
574 err_put:
575 	fdput(f);
576 	return err;
577 }
578 
579 static const struct bpf_verifier_ops * const bpf_prog_types[] = {
580 #define BPF_PROG_TYPE(_id, _ops) \
581 	[_id] = &_ops,
582 #define BPF_MAP_TYPE(_id, _ops)
583 #include <linux/bpf_types.h>
584 #undef BPF_PROG_TYPE
585 #undef BPF_MAP_TYPE
586 };
587 
588 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
589 {
590 	if (type >= ARRAY_SIZE(bpf_prog_types) || !bpf_prog_types[type])
591 		return -EINVAL;
592 
593 	prog->aux->ops = bpf_prog_types[type];
594 	prog->type = type;
595 	return 0;
596 }
597 
598 /* drop refcnt on maps used by eBPF program and free auxilary data */
599 static void free_used_maps(struct bpf_prog_aux *aux)
600 {
601 	int i;
602 
603 	for (i = 0; i < aux->used_map_cnt; i++)
604 		bpf_map_put(aux->used_maps[i]);
605 
606 	kfree(aux->used_maps);
607 }
608 
609 int __bpf_prog_charge(struct user_struct *user, u32 pages)
610 {
611 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
612 	unsigned long user_bufs;
613 
614 	if (user) {
615 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
616 		if (user_bufs > memlock_limit) {
617 			atomic_long_sub(pages, &user->locked_vm);
618 			return -EPERM;
619 		}
620 	}
621 
622 	return 0;
623 }
624 
625 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
626 {
627 	if (user)
628 		atomic_long_sub(pages, &user->locked_vm);
629 }
630 
631 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
632 {
633 	struct user_struct *user = get_current_user();
634 	int ret;
635 
636 	ret = __bpf_prog_charge(user, prog->pages);
637 	if (ret) {
638 		free_uid(user);
639 		return ret;
640 	}
641 
642 	prog->aux->user = user;
643 	return 0;
644 }
645 
646 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
647 {
648 	struct user_struct *user = prog->aux->user;
649 
650 	__bpf_prog_uncharge(user, prog->pages);
651 	free_uid(user);
652 }
653 
654 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
655 {
656 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
657 
658 	free_used_maps(aux);
659 	bpf_prog_uncharge_memlock(aux->prog);
660 	bpf_prog_free(aux->prog);
661 }
662 
663 void bpf_prog_put(struct bpf_prog *prog)
664 {
665 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
666 		trace_bpf_prog_put_rcu(prog);
667 		bpf_prog_kallsyms_del(prog);
668 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
669 	}
670 }
671 EXPORT_SYMBOL_GPL(bpf_prog_put);
672 
673 static int bpf_prog_release(struct inode *inode, struct file *filp)
674 {
675 	struct bpf_prog *prog = filp->private_data;
676 
677 	bpf_prog_put(prog);
678 	return 0;
679 }
680 
681 #ifdef CONFIG_PROC_FS
682 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
683 {
684 	const struct bpf_prog *prog = filp->private_data;
685 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
686 
687 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
688 	seq_printf(m,
689 		   "prog_type:\t%u\n"
690 		   "prog_jited:\t%u\n"
691 		   "prog_tag:\t%s\n"
692 		   "memlock:\t%llu\n",
693 		   prog->type,
694 		   prog->jited,
695 		   prog_tag,
696 		   prog->pages * 1ULL << PAGE_SHIFT);
697 }
698 #endif
699 
700 static const struct file_operations bpf_prog_fops = {
701 #ifdef CONFIG_PROC_FS
702 	.show_fdinfo	= bpf_prog_show_fdinfo,
703 #endif
704 	.release	= bpf_prog_release,
705 };
706 
707 int bpf_prog_new_fd(struct bpf_prog *prog)
708 {
709 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
710 				O_RDWR | O_CLOEXEC);
711 }
712 
713 static struct bpf_prog *____bpf_prog_get(struct fd f)
714 {
715 	if (!f.file)
716 		return ERR_PTR(-EBADF);
717 	if (f.file->f_op != &bpf_prog_fops) {
718 		fdput(f);
719 		return ERR_PTR(-EINVAL);
720 	}
721 
722 	return f.file->private_data;
723 }
724 
725 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
726 {
727 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
728 		atomic_sub(i, &prog->aux->refcnt);
729 		return ERR_PTR(-EBUSY);
730 	}
731 	return prog;
732 }
733 EXPORT_SYMBOL_GPL(bpf_prog_add);
734 
735 void bpf_prog_sub(struct bpf_prog *prog, int i)
736 {
737 	/* Only to be used for undoing previous bpf_prog_add() in some
738 	 * error path. We still know that another entity in our call
739 	 * path holds a reference to the program, thus atomic_sub() can
740 	 * be safely used in such cases!
741 	 */
742 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
743 }
744 EXPORT_SYMBOL_GPL(bpf_prog_sub);
745 
746 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
747 {
748 	return bpf_prog_add(prog, 1);
749 }
750 EXPORT_SYMBOL_GPL(bpf_prog_inc);
751 
752 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type)
753 {
754 	struct fd f = fdget(ufd);
755 	struct bpf_prog *prog;
756 
757 	prog = ____bpf_prog_get(f);
758 	if (IS_ERR(prog))
759 		return prog;
760 	if (type && prog->type != *type) {
761 		prog = ERR_PTR(-EINVAL);
762 		goto out;
763 	}
764 
765 	prog = bpf_prog_inc(prog);
766 out:
767 	fdput(f);
768 	return prog;
769 }
770 
771 struct bpf_prog *bpf_prog_get(u32 ufd)
772 {
773 	return __bpf_prog_get(ufd, NULL);
774 }
775 
776 struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type)
777 {
778 	struct bpf_prog *prog = __bpf_prog_get(ufd, &type);
779 
780 	if (!IS_ERR(prog))
781 		trace_bpf_prog_get_type(prog);
782 	return prog;
783 }
784 EXPORT_SYMBOL_GPL(bpf_prog_get_type);
785 
786 /* last field in 'union bpf_attr' used by this command */
787 #define	BPF_PROG_LOAD_LAST_FIELD kern_version
788 
789 static int bpf_prog_load(union bpf_attr *attr)
790 {
791 	enum bpf_prog_type type = attr->prog_type;
792 	struct bpf_prog *prog;
793 	int err;
794 	char license[128];
795 	bool is_gpl;
796 
797 	if (CHECK_ATTR(BPF_PROG_LOAD))
798 		return -EINVAL;
799 
800 	/* copy eBPF program license from user space */
801 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
802 			      sizeof(license) - 1) < 0)
803 		return -EFAULT;
804 	license[sizeof(license) - 1] = 0;
805 
806 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
807 	is_gpl = license_is_gpl_compatible(license);
808 
809 	if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
810 		return -E2BIG;
811 
812 	if (type == BPF_PROG_TYPE_KPROBE &&
813 	    attr->kern_version != LINUX_VERSION_CODE)
814 		return -EINVAL;
815 
816 	if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN))
817 		return -EPERM;
818 
819 	/* plain bpf_prog allocation */
820 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
821 	if (!prog)
822 		return -ENOMEM;
823 
824 	err = bpf_prog_charge_memlock(prog);
825 	if (err)
826 		goto free_prog_nouncharge;
827 
828 	prog->len = attr->insn_cnt;
829 
830 	err = -EFAULT;
831 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
832 			   bpf_prog_insn_size(prog)) != 0)
833 		goto free_prog;
834 
835 	prog->orig_prog = NULL;
836 	prog->jited = 0;
837 
838 	atomic_set(&prog->aux->refcnt, 1);
839 	prog->gpl_compatible = is_gpl ? 1 : 0;
840 
841 	/* find program type: socket_filter vs tracing_filter */
842 	err = find_prog_type(type, prog);
843 	if (err < 0)
844 		goto free_prog;
845 
846 	/* run eBPF verifier */
847 	err = bpf_check(&prog, attr);
848 	if (err < 0)
849 		goto free_used_maps;
850 
851 	/* eBPF program is ready to be JITed */
852 	prog = bpf_prog_select_runtime(prog, &err);
853 	if (err < 0)
854 		goto free_used_maps;
855 
856 	err = bpf_prog_new_fd(prog);
857 	if (err < 0)
858 		/* failed to allocate fd */
859 		goto free_used_maps;
860 
861 	bpf_prog_kallsyms_add(prog);
862 	trace_bpf_prog_load(prog, err);
863 	return err;
864 
865 free_used_maps:
866 	free_used_maps(prog->aux);
867 free_prog:
868 	bpf_prog_uncharge_memlock(prog);
869 free_prog_nouncharge:
870 	bpf_prog_free(prog);
871 	return err;
872 }
873 
874 #define BPF_OBJ_LAST_FIELD bpf_fd
875 
876 static int bpf_obj_pin(const union bpf_attr *attr)
877 {
878 	if (CHECK_ATTR(BPF_OBJ))
879 		return -EINVAL;
880 
881 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
882 }
883 
884 static int bpf_obj_get(const union bpf_attr *attr)
885 {
886 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0)
887 		return -EINVAL;
888 
889 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname));
890 }
891 
892 #ifdef CONFIG_CGROUP_BPF
893 
894 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
895 
896 static int bpf_prog_attach(const union bpf_attr *attr)
897 {
898 	enum bpf_prog_type ptype;
899 	struct bpf_prog *prog;
900 	struct cgroup *cgrp;
901 	int ret;
902 
903 	if (!capable(CAP_NET_ADMIN))
904 		return -EPERM;
905 
906 	if (CHECK_ATTR(BPF_PROG_ATTACH))
907 		return -EINVAL;
908 
909 	if (attr->attach_flags & ~BPF_F_ALLOW_OVERRIDE)
910 		return -EINVAL;
911 
912 	switch (attr->attach_type) {
913 	case BPF_CGROUP_INET_INGRESS:
914 	case BPF_CGROUP_INET_EGRESS:
915 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
916 		break;
917 	case BPF_CGROUP_INET_SOCK_CREATE:
918 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
919 		break;
920 	default:
921 		return -EINVAL;
922 	}
923 
924 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
925 	if (IS_ERR(prog))
926 		return PTR_ERR(prog);
927 
928 	cgrp = cgroup_get_from_fd(attr->target_fd);
929 	if (IS_ERR(cgrp)) {
930 		bpf_prog_put(prog);
931 		return PTR_ERR(cgrp);
932 	}
933 
934 	ret = cgroup_bpf_update(cgrp, prog, attr->attach_type,
935 				attr->attach_flags & BPF_F_ALLOW_OVERRIDE);
936 	if (ret)
937 		bpf_prog_put(prog);
938 	cgroup_put(cgrp);
939 
940 	return ret;
941 }
942 
943 #define BPF_PROG_DETACH_LAST_FIELD attach_type
944 
945 static int bpf_prog_detach(const union bpf_attr *attr)
946 {
947 	struct cgroup *cgrp;
948 	int ret;
949 
950 	if (!capable(CAP_NET_ADMIN))
951 		return -EPERM;
952 
953 	if (CHECK_ATTR(BPF_PROG_DETACH))
954 		return -EINVAL;
955 
956 	switch (attr->attach_type) {
957 	case BPF_CGROUP_INET_INGRESS:
958 	case BPF_CGROUP_INET_EGRESS:
959 	case BPF_CGROUP_INET_SOCK_CREATE:
960 		cgrp = cgroup_get_from_fd(attr->target_fd);
961 		if (IS_ERR(cgrp))
962 			return PTR_ERR(cgrp);
963 
964 		ret = cgroup_bpf_update(cgrp, NULL, attr->attach_type, false);
965 		cgroup_put(cgrp);
966 		break;
967 
968 	default:
969 		return -EINVAL;
970 	}
971 
972 	return ret;
973 }
974 #endif /* CONFIG_CGROUP_BPF */
975 
976 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
977 
978 static int bpf_prog_test_run(const union bpf_attr *attr,
979 			     union bpf_attr __user *uattr)
980 {
981 	struct bpf_prog *prog;
982 	int ret = -ENOTSUPP;
983 
984 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
985 		return -EINVAL;
986 
987 	prog = bpf_prog_get(attr->test.prog_fd);
988 	if (IS_ERR(prog))
989 		return PTR_ERR(prog);
990 
991 	if (prog->aux->ops->test_run)
992 		ret = prog->aux->ops->test_run(prog, attr, uattr);
993 
994 	bpf_prog_put(prog);
995 	return ret;
996 }
997 
998 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
999 {
1000 	union bpf_attr attr = {};
1001 	int err;
1002 
1003 	if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
1004 		return -EPERM;
1005 
1006 	if (!access_ok(VERIFY_READ, uattr, 1))
1007 		return -EFAULT;
1008 
1009 	if (size > PAGE_SIZE)	/* silly large */
1010 		return -E2BIG;
1011 
1012 	/* If we're handed a bigger struct than we know of,
1013 	 * ensure all the unknown bits are 0 - i.e. new
1014 	 * user-space does not rely on any kernel feature
1015 	 * extensions we dont know about yet.
1016 	 */
1017 	if (size > sizeof(attr)) {
1018 		unsigned char __user *addr;
1019 		unsigned char __user *end;
1020 		unsigned char val;
1021 
1022 		addr = (void __user *)uattr + sizeof(attr);
1023 		end  = (void __user *)uattr + size;
1024 
1025 		for (; addr < end; addr++) {
1026 			err = get_user(val, addr);
1027 			if (err)
1028 				return err;
1029 			if (val)
1030 				return -E2BIG;
1031 		}
1032 		size = sizeof(attr);
1033 	}
1034 
1035 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
1036 	if (copy_from_user(&attr, uattr, size) != 0)
1037 		return -EFAULT;
1038 
1039 	switch (cmd) {
1040 	case BPF_MAP_CREATE:
1041 		err = map_create(&attr);
1042 		break;
1043 	case BPF_MAP_LOOKUP_ELEM:
1044 		err = map_lookup_elem(&attr);
1045 		break;
1046 	case BPF_MAP_UPDATE_ELEM:
1047 		err = map_update_elem(&attr);
1048 		break;
1049 	case BPF_MAP_DELETE_ELEM:
1050 		err = map_delete_elem(&attr);
1051 		break;
1052 	case BPF_MAP_GET_NEXT_KEY:
1053 		err = map_get_next_key(&attr);
1054 		break;
1055 	case BPF_PROG_LOAD:
1056 		err = bpf_prog_load(&attr);
1057 		break;
1058 	case BPF_OBJ_PIN:
1059 		err = bpf_obj_pin(&attr);
1060 		break;
1061 	case BPF_OBJ_GET:
1062 		err = bpf_obj_get(&attr);
1063 		break;
1064 #ifdef CONFIG_CGROUP_BPF
1065 	case BPF_PROG_ATTACH:
1066 		err = bpf_prog_attach(&attr);
1067 		break;
1068 	case BPF_PROG_DETACH:
1069 		err = bpf_prog_detach(&attr);
1070 		break;
1071 #endif
1072 	case BPF_PROG_TEST_RUN:
1073 		err = bpf_prog_test_run(&attr, uattr);
1074 		break;
1075 	default:
1076 		err = -EINVAL;
1077 		break;
1078 	}
1079 
1080 	return err;
1081 }
1082