xref: /linux/kernel/bpf/syscall.c (revision 4f139972b489f8bc2c821aa25ac65018d92af3f7)
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 LIST_HEAD(bpf_map_types);
31 
32 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
33 {
34 	struct bpf_map_type_list *tl;
35 	struct bpf_map *map;
36 
37 	list_for_each_entry(tl, &bpf_map_types, list_node) {
38 		if (tl->type == attr->map_type) {
39 			map = tl->ops->map_alloc(attr);
40 			if (IS_ERR(map))
41 				return map;
42 			map->ops = tl->ops;
43 			map->map_type = attr->map_type;
44 			return map;
45 		}
46 	}
47 	return ERR_PTR(-EINVAL);
48 }
49 
50 /* boot time registration of different map implementations */
51 void bpf_register_map_type(struct bpf_map_type_list *tl)
52 {
53 	list_add(&tl->list_node, &bpf_map_types);
54 }
55 
56 void *bpf_map_area_alloc(size_t size)
57 {
58 	/* We definitely need __GFP_NORETRY, so OOM killer doesn't
59 	 * trigger under memory pressure as we really just want to
60 	 * fail instead.
61 	 */
62 	const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
63 	void *area;
64 
65 	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
66 		area = kmalloc(size, GFP_USER | flags);
67 		if (area != NULL)
68 			return area;
69 	}
70 
71 	return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | flags,
72 			 PAGE_KERNEL);
73 }
74 
75 void bpf_map_area_free(void *area)
76 {
77 	kvfree(area);
78 }
79 
80 int bpf_map_precharge_memlock(u32 pages)
81 {
82 	struct user_struct *user = get_current_user();
83 	unsigned long memlock_limit, cur;
84 
85 	memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
86 	cur = atomic_long_read(&user->locked_vm);
87 	free_uid(user);
88 	if (cur + pages > memlock_limit)
89 		return -EPERM;
90 	return 0;
91 }
92 
93 static int bpf_map_charge_memlock(struct bpf_map *map)
94 {
95 	struct user_struct *user = get_current_user();
96 	unsigned long memlock_limit;
97 
98 	memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
99 
100 	atomic_long_add(map->pages, &user->locked_vm);
101 
102 	if (atomic_long_read(&user->locked_vm) > memlock_limit) {
103 		atomic_long_sub(map->pages, &user->locked_vm);
104 		free_uid(user);
105 		return -EPERM;
106 	}
107 	map->user = user;
108 	return 0;
109 }
110 
111 static void bpf_map_uncharge_memlock(struct bpf_map *map)
112 {
113 	struct user_struct *user = map->user;
114 
115 	atomic_long_sub(map->pages, &user->locked_vm);
116 	free_uid(user);
117 }
118 
119 /* called from workqueue */
120 static void bpf_map_free_deferred(struct work_struct *work)
121 {
122 	struct bpf_map *map = container_of(work, struct bpf_map, work);
123 
124 	bpf_map_uncharge_memlock(map);
125 	/* implementation dependent freeing */
126 	map->ops->map_free(map);
127 }
128 
129 static void bpf_map_put_uref(struct bpf_map *map)
130 {
131 	if (atomic_dec_and_test(&map->usercnt)) {
132 		if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
133 			bpf_fd_array_map_clear(map);
134 	}
135 }
136 
137 /* decrement map refcnt and schedule it for freeing via workqueue
138  * (unrelying map implementation ops->map_free() might sleep)
139  */
140 void bpf_map_put(struct bpf_map *map)
141 {
142 	if (atomic_dec_and_test(&map->refcnt)) {
143 		INIT_WORK(&map->work, bpf_map_free_deferred);
144 		schedule_work(&map->work);
145 	}
146 }
147 
148 void bpf_map_put_with_uref(struct bpf_map *map)
149 {
150 	bpf_map_put_uref(map);
151 	bpf_map_put(map);
152 }
153 
154 static int bpf_map_release(struct inode *inode, struct file *filp)
155 {
156 	struct bpf_map *map = filp->private_data;
157 
158 	if (map->ops->map_release)
159 		map->ops->map_release(map, filp);
160 
161 	bpf_map_put_with_uref(map);
162 	return 0;
163 }
164 
165 #ifdef CONFIG_PROC_FS
166 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
167 {
168 	const struct bpf_map *map = filp->private_data;
169 	const struct bpf_array *array;
170 	u32 owner_prog_type = 0;
171 
172 	if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
173 		array = container_of(map, struct bpf_array, map);
174 		owner_prog_type = array->owner_prog_type;
175 	}
176 
177 	seq_printf(m,
178 		   "map_type:\t%u\n"
179 		   "key_size:\t%u\n"
180 		   "value_size:\t%u\n"
181 		   "max_entries:\t%u\n"
182 		   "map_flags:\t%#x\n"
183 		   "memlock:\t%llu\n",
184 		   map->map_type,
185 		   map->key_size,
186 		   map->value_size,
187 		   map->max_entries,
188 		   map->map_flags,
189 		   map->pages * 1ULL << PAGE_SHIFT);
190 
191 	if (owner_prog_type)
192 		seq_printf(m, "owner_prog_type:\t%u\n",
193 			   owner_prog_type);
194 }
195 #endif
196 
197 static const struct file_operations bpf_map_fops = {
198 #ifdef CONFIG_PROC_FS
199 	.show_fdinfo	= bpf_map_show_fdinfo,
200 #endif
201 	.release	= bpf_map_release,
202 };
203 
204 int bpf_map_new_fd(struct bpf_map *map)
205 {
206 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
207 				O_RDWR | O_CLOEXEC);
208 }
209 
210 /* helper macro to check that unused fields 'union bpf_attr' are zero */
211 #define CHECK_ATTR(CMD) \
212 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
213 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
214 		   sizeof(*attr) - \
215 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
216 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
217 
218 #define BPF_MAP_CREATE_LAST_FIELD inner_map_fd
219 /* called via syscall */
220 static int map_create(union bpf_attr *attr)
221 {
222 	struct bpf_map *map;
223 	int err;
224 
225 	err = CHECK_ATTR(BPF_MAP_CREATE);
226 	if (err)
227 		return -EINVAL;
228 
229 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
230 	map = find_and_alloc_map(attr);
231 	if (IS_ERR(map))
232 		return PTR_ERR(map);
233 
234 	atomic_set(&map->refcnt, 1);
235 	atomic_set(&map->usercnt, 1);
236 
237 	err = bpf_map_charge_memlock(map);
238 	if (err)
239 		goto free_map_nouncharge;
240 
241 	err = bpf_map_new_fd(map);
242 	if (err < 0)
243 		/* failed to allocate fd */
244 		goto free_map;
245 
246 	trace_bpf_map_create(map, err);
247 	return err;
248 
249 free_map:
250 	bpf_map_uncharge_memlock(map);
251 free_map_nouncharge:
252 	map->ops->map_free(map);
253 	return err;
254 }
255 
256 /* if error is returned, fd is released.
257  * On success caller should complete fd access with matching fdput()
258  */
259 struct bpf_map *__bpf_map_get(struct fd f)
260 {
261 	if (!f.file)
262 		return ERR_PTR(-EBADF);
263 	if (f.file->f_op != &bpf_map_fops) {
264 		fdput(f);
265 		return ERR_PTR(-EINVAL);
266 	}
267 
268 	return f.file->private_data;
269 }
270 
271 /* prog's and map's refcnt limit */
272 #define BPF_MAX_REFCNT 32768
273 
274 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
275 {
276 	if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
277 		atomic_dec(&map->refcnt);
278 		return ERR_PTR(-EBUSY);
279 	}
280 	if (uref)
281 		atomic_inc(&map->usercnt);
282 	return map;
283 }
284 
285 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
286 {
287 	struct fd f = fdget(ufd);
288 	struct bpf_map *map;
289 
290 	map = __bpf_map_get(f);
291 	if (IS_ERR(map))
292 		return map;
293 
294 	map = bpf_map_inc(map, true);
295 	fdput(f);
296 
297 	return map;
298 }
299 
300 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
301 {
302 	return -ENOTSUPP;
303 }
304 
305 /* last field in 'union bpf_attr' used by this command */
306 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
307 
308 static int map_lookup_elem(union bpf_attr *attr)
309 {
310 	void __user *ukey = u64_to_user_ptr(attr->key);
311 	void __user *uvalue = u64_to_user_ptr(attr->value);
312 	int ufd = attr->map_fd;
313 	struct bpf_map *map;
314 	void *key, *value, *ptr;
315 	u32 value_size;
316 	struct fd f;
317 	int err;
318 
319 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
320 		return -EINVAL;
321 
322 	f = fdget(ufd);
323 	map = __bpf_map_get(f);
324 	if (IS_ERR(map))
325 		return PTR_ERR(map);
326 
327 	err = -ENOMEM;
328 	key = kmalloc(map->key_size, GFP_USER);
329 	if (!key)
330 		goto err_put;
331 
332 	err = -EFAULT;
333 	if (copy_from_user(key, ukey, map->key_size) != 0)
334 		goto free_key;
335 
336 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
337 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
338 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
339 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
340 	else
341 		value_size = map->value_size;
342 
343 	err = -ENOMEM;
344 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
345 	if (!value)
346 		goto free_key;
347 
348 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
349 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
350 		err = bpf_percpu_hash_copy(map, key, value);
351 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
352 		err = bpf_percpu_array_copy(map, key, value);
353 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
354 		err = bpf_stackmap_copy(map, key, value);
355 	} else if (map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
356 		   map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
357 		err = -ENOTSUPP;
358 	} else {
359 		rcu_read_lock();
360 		ptr = map->ops->map_lookup_elem(map, key);
361 		if (ptr)
362 			memcpy(value, ptr, value_size);
363 		rcu_read_unlock();
364 		err = ptr ? 0 : -ENOENT;
365 	}
366 
367 	if (err)
368 		goto free_value;
369 
370 	err = -EFAULT;
371 	if (copy_to_user(uvalue, value, value_size) != 0)
372 		goto free_value;
373 
374 	trace_bpf_map_lookup_elem(map, ufd, key, value);
375 	err = 0;
376 
377 free_value:
378 	kfree(value);
379 free_key:
380 	kfree(key);
381 err_put:
382 	fdput(f);
383 	return err;
384 }
385 
386 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
387 
388 static int map_update_elem(union bpf_attr *attr)
389 {
390 	void __user *ukey = u64_to_user_ptr(attr->key);
391 	void __user *uvalue = u64_to_user_ptr(attr->value);
392 	int ufd = attr->map_fd;
393 	struct bpf_map *map;
394 	void *key, *value;
395 	u32 value_size;
396 	struct fd f;
397 	int err;
398 
399 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
400 		return -EINVAL;
401 
402 	f = fdget(ufd);
403 	map = __bpf_map_get(f);
404 	if (IS_ERR(map))
405 		return PTR_ERR(map);
406 
407 	err = -ENOMEM;
408 	key = kmalloc(map->key_size, GFP_USER);
409 	if (!key)
410 		goto err_put;
411 
412 	err = -EFAULT;
413 	if (copy_from_user(key, ukey, map->key_size) != 0)
414 		goto free_key;
415 
416 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
417 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
418 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
419 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
420 	else
421 		value_size = map->value_size;
422 
423 	err = -ENOMEM;
424 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
425 	if (!value)
426 		goto free_key;
427 
428 	err = -EFAULT;
429 	if (copy_from_user(value, uvalue, value_size) != 0)
430 		goto free_value;
431 
432 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
433 	 * inside bpf map update or delete otherwise deadlocks are possible
434 	 */
435 	preempt_disable();
436 	__this_cpu_inc(bpf_prog_active);
437 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
438 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
439 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
440 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
441 		err = bpf_percpu_array_update(map, key, value, attr->flags);
442 	} else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
443 		   map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
444 		   map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY ||
445 		   map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) {
446 		rcu_read_lock();
447 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
448 						   attr->flags);
449 		rcu_read_unlock();
450 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
451 		rcu_read_lock();
452 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
453 						  attr->flags);
454 		rcu_read_unlock();
455 	} else {
456 		rcu_read_lock();
457 		err = map->ops->map_update_elem(map, key, value, attr->flags);
458 		rcu_read_unlock();
459 	}
460 	__this_cpu_dec(bpf_prog_active);
461 	preempt_enable();
462 
463 	if (!err)
464 		trace_bpf_map_update_elem(map, ufd, key, value);
465 free_value:
466 	kfree(value);
467 free_key:
468 	kfree(key);
469 err_put:
470 	fdput(f);
471 	return err;
472 }
473 
474 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
475 
476 static int map_delete_elem(union bpf_attr *attr)
477 {
478 	void __user *ukey = u64_to_user_ptr(attr->key);
479 	int ufd = attr->map_fd;
480 	struct bpf_map *map;
481 	struct fd f;
482 	void *key;
483 	int err;
484 
485 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
486 		return -EINVAL;
487 
488 	f = fdget(ufd);
489 	map = __bpf_map_get(f);
490 	if (IS_ERR(map))
491 		return PTR_ERR(map);
492 
493 	err = -ENOMEM;
494 	key = kmalloc(map->key_size, GFP_USER);
495 	if (!key)
496 		goto err_put;
497 
498 	err = -EFAULT;
499 	if (copy_from_user(key, ukey, map->key_size) != 0)
500 		goto free_key;
501 
502 	preempt_disable();
503 	__this_cpu_inc(bpf_prog_active);
504 	rcu_read_lock();
505 	err = map->ops->map_delete_elem(map, key);
506 	rcu_read_unlock();
507 	__this_cpu_dec(bpf_prog_active);
508 	preempt_enable();
509 
510 	if (!err)
511 		trace_bpf_map_delete_elem(map, ufd, key);
512 free_key:
513 	kfree(key);
514 err_put:
515 	fdput(f);
516 	return err;
517 }
518 
519 /* last field in 'union bpf_attr' used by this command */
520 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
521 
522 static int map_get_next_key(union bpf_attr *attr)
523 {
524 	void __user *ukey = u64_to_user_ptr(attr->key);
525 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
526 	int ufd = attr->map_fd;
527 	struct bpf_map *map;
528 	void *key, *next_key;
529 	struct fd f;
530 	int err;
531 
532 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
533 		return -EINVAL;
534 
535 	f = fdget(ufd);
536 	map = __bpf_map_get(f);
537 	if (IS_ERR(map))
538 		return PTR_ERR(map);
539 
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 
549 	err = -ENOMEM;
550 	next_key = kmalloc(map->key_size, GFP_USER);
551 	if (!next_key)
552 		goto free_key;
553 
554 	rcu_read_lock();
555 	err = map->ops->map_get_next_key(map, key, next_key);
556 	rcu_read_unlock();
557 	if (err)
558 		goto free_next_key;
559 
560 	err = -EFAULT;
561 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
562 		goto free_next_key;
563 
564 	trace_bpf_map_next_key(map, ufd, key, next_key);
565 	err = 0;
566 
567 free_next_key:
568 	kfree(next_key);
569 free_key:
570 	kfree(key);
571 err_put:
572 	fdput(f);
573 	return err;
574 }
575 
576 static LIST_HEAD(bpf_prog_types);
577 
578 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
579 {
580 	struct bpf_prog_type_list *tl;
581 
582 	list_for_each_entry(tl, &bpf_prog_types, list_node) {
583 		if (tl->type == type) {
584 			prog->aux->ops = tl->ops;
585 			prog->type = type;
586 			return 0;
587 		}
588 	}
589 
590 	return -EINVAL;
591 }
592 
593 void bpf_register_prog_type(struct bpf_prog_type_list *tl)
594 {
595 	list_add(&tl->list_node, &bpf_prog_types);
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