xref: /linux/kernel/bpf/syscall.c (revision a3a4a816b4b194c45d0217e8b9e08b2639802cda)
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 map_flags
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 {
356 		rcu_read_lock();
357 		ptr = map->ops->map_lookup_elem(map, key);
358 		if (ptr)
359 			memcpy(value, ptr, value_size);
360 		rcu_read_unlock();
361 		err = ptr ? 0 : -ENOENT;
362 	}
363 
364 	if (err)
365 		goto free_value;
366 
367 	err = -EFAULT;
368 	if (copy_to_user(uvalue, value, value_size) != 0)
369 		goto free_value;
370 
371 	trace_bpf_map_lookup_elem(map, ufd, key, value);
372 	err = 0;
373 
374 free_value:
375 	kfree(value);
376 free_key:
377 	kfree(key);
378 err_put:
379 	fdput(f);
380 	return err;
381 }
382 
383 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
384 
385 static int map_update_elem(union bpf_attr *attr)
386 {
387 	void __user *ukey = u64_to_user_ptr(attr->key);
388 	void __user *uvalue = u64_to_user_ptr(attr->value);
389 	int ufd = attr->map_fd;
390 	struct bpf_map *map;
391 	void *key, *value;
392 	u32 value_size;
393 	struct fd f;
394 	int err;
395 
396 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
397 		return -EINVAL;
398 
399 	f = fdget(ufd);
400 	map = __bpf_map_get(f);
401 	if (IS_ERR(map))
402 		return PTR_ERR(map);
403 
404 	err = -ENOMEM;
405 	key = kmalloc(map->key_size, GFP_USER);
406 	if (!key)
407 		goto err_put;
408 
409 	err = -EFAULT;
410 	if (copy_from_user(key, ukey, map->key_size) != 0)
411 		goto free_key;
412 
413 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
414 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
415 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
416 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
417 	else
418 		value_size = map->value_size;
419 
420 	err = -ENOMEM;
421 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
422 	if (!value)
423 		goto free_key;
424 
425 	err = -EFAULT;
426 	if (copy_from_user(value, uvalue, value_size) != 0)
427 		goto free_value;
428 
429 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
430 	 * inside bpf map update or delete otherwise deadlocks are possible
431 	 */
432 	preempt_disable();
433 	__this_cpu_inc(bpf_prog_active);
434 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
435 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
436 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
437 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
438 		err = bpf_percpu_array_update(map, key, value, attr->flags);
439 	} else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
440 		   map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
441 		   map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY) {
442 		rcu_read_lock();
443 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
444 						   attr->flags);
445 		rcu_read_unlock();
446 	} else {
447 		rcu_read_lock();
448 		err = map->ops->map_update_elem(map, key, value, attr->flags);
449 		rcu_read_unlock();
450 	}
451 	__this_cpu_dec(bpf_prog_active);
452 	preempt_enable();
453 
454 	if (!err)
455 		trace_bpf_map_update_elem(map, ufd, key, value);
456 free_value:
457 	kfree(value);
458 free_key:
459 	kfree(key);
460 err_put:
461 	fdput(f);
462 	return err;
463 }
464 
465 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
466 
467 static int map_delete_elem(union bpf_attr *attr)
468 {
469 	void __user *ukey = u64_to_user_ptr(attr->key);
470 	int ufd = attr->map_fd;
471 	struct bpf_map *map;
472 	struct fd f;
473 	void *key;
474 	int err;
475 
476 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
477 		return -EINVAL;
478 
479 	f = fdget(ufd);
480 	map = __bpf_map_get(f);
481 	if (IS_ERR(map))
482 		return PTR_ERR(map);
483 
484 	err = -ENOMEM;
485 	key = kmalloc(map->key_size, GFP_USER);
486 	if (!key)
487 		goto err_put;
488 
489 	err = -EFAULT;
490 	if (copy_from_user(key, ukey, map->key_size) != 0)
491 		goto free_key;
492 
493 	preempt_disable();
494 	__this_cpu_inc(bpf_prog_active);
495 	rcu_read_lock();
496 	err = map->ops->map_delete_elem(map, key);
497 	rcu_read_unlock();
498 	__this_cpu_dec(bpf_prog_active);
499 	preempt_enable();
500 
501 	if (!err)
502 		trace_bpf_map_delete_elem(map, ufd, key);
503 free_key:
504 	kfree(key);
505 err_put:
506 	fdput(f);
507 	return err;
508 }
509 
510 /* last field in 'union bpf_attr' used by this command */
511 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
512 
513 static int map_get_next_key(union bpf_attr *attr)
514 {
515 	void __user *ukey = u64_to_user_ptr(attr->key);
516 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
517 	int ufd = attr->map_fd;
518 	struct bpf_map *map;
519 	void *key, *next_key;
520 	struct fd f;
521 	int err;
522 
523 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
524 		return -EINVAL;
525 
526 	f = fdget(ufd);
527 	map = __bpf_map_get(f);
528 	if (IS_ERR(map))
529 		return PTR_ERR(map);
530 
531 	err = -ENOMEM;
532 	key = kmalloc(map->key_size, GFP_USER);
533 	if (!key)
534 		goto err_put;
535 
536 	err = -EFAULT;
537 	if (copy_from_user(key, ukey, map->key_size) != 0)
538 		goto free_key;
539 
540 	err = -ENOMEM;
541 	next_key = kmalloc(map->key_size, GFP_USER);
542 	if (!next_key)
543 		goto free_key;
544 
545 	rcu_read_lock();
546 	err = map->ops->map_get_next_key(map, key, next_key);
547 	rcu_read_unlock();
548 	if (err)
549 		goto free_next_key;
550 
551 	err = -EFAULT;
552 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
553 		goto free_next_key;
554 
555 	trace_bpf_map_next_key(map, ufd, key, next_key);
556 	err = 0;
557 
558 free_next_key:
559 	kfree(next_key);
560 free_key:
561 	kfree(key);
562 err_put:
563 	fdput(f);
564 	return err;
565 }
566 
567 static LIST_HEAD(bpf_prog_types);
568 
569 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
570 {
571 	struct bpf_prog_type_list *tl;
572 
573 	list_for_each_entry(tl, &bpf_prog_types, list_node) {
574 		if (tl->type == type) {
575 			prog->aux->ops = tl->ops;
576 			prog->type = type;
577 			return 0;
578 		}
579 	}
580 
581 	return -EINVAL;
582 }
583 
584 void bpf_register_prog_type(struct bpf_prog_type_list *tl)
585 {
586 	list_add(&tl->list_node, &bpf_prog_types);
587 }
588 
589 /* fixup insn->imm field of bpf_call instructions:
590  * if (insn->imm == BPF_FUNC_map_lookup_elem)
591  *      insn->imm = bpf_map_lookup_elem - __bpf_call_base;
592  * else if (insn->imm == BPF_FUNC_map_update_elem)
593  *      insn->imm = bpf_map_update_elem - __bpf_call_base;
594  * else ...
595  *
596  * this function is called after eBPF program passed verification
597  */
598 static void fixup_bpf_calls(struct bpf_prog *prog)
599 {
600 	const struct bpf_func_proto *fn;
601 	int i;
602 
603 	for (i = 0; i < prog->len; i++) {
604 		struct bpf_insn *insn = &prog->insnsi[i];
605 
606 		if (insn->code == (BPF_JMP | BPF_CALL)) {
607 			/* we reach here when program has bpf_call instructions
608 			 * and it passed bpf_check(), means that
609 			 * ops->get_func_proto must have been supplied, check it
610 			 */
611 			BUG_ON(!prog->aux->ops->get_func_proto);
612 
613 			if (insn->imm == BPF_FUNC_get_route_realm)
614 				prog->dst_needed = 1;
615 			if (insn->imm == BPF_FUNC_get_prandom_u32)
616 				bpf_user_rnd_init_once();
617 			if (insn->imm == BPF_FUNC_xdp_adjust_head)
618 				prog->xdp_adjust_head = 1;
619 			if (insn->imm == BPF_FUNC_tail_call) {
620 				/* mark bpf_tail_call as different opcode
621 				 * to avoid conditional branch in
622 				 * interpeter for every normal call
623 				 * and to prevent accidental JITing by
624 				 * JIT compiler that doesn't support
625 				 * bpf_tail_call yet
626 				 */
627 				insn->imm = 0;
628 				insn->code |= BPF_X;
629 				continue;
630 			}
631 
632 			fn = prog->aux->ops->get_func_proto(insn->imm);
633 			/* all functions that have prototype and verifier allowed
634 			 * programs to call them, must be real in-kernel functions
635 			 */
636 			BUG_ON(!fn->func);
637 			insn->imm = fn->func - __bpf_call_base;
638 		}
639 	}
640 }
641 
642 /* drop refcnt on maps used by eBPF program and free auxilary data */
643 static void free_used_maps(struct bpf_prog_aux *aux)
644 {
645 	int i;
646 
647 	for (i = 0; i < aux->used_map_cnt; i++)
648 		bpf_map_put(aux->used_maps[i]);
649 
650 	kfree(aux->used_maps);
651 }
652 
653 int __bpf_prog_charge(struct user_struct *user, u32 pages)
654 {
655 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
656 	unsigned long user_bufs;
657 
658 	if (user) {
659 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
660 		if (user_bufs > memlock_limit) {
661 			atomic_long_sub(pages, &user->locked_vm);
662 			return -EPERM;
663 		}
664 	}
665 
666 	return 0;
667 }
668 
669 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
670 {
671 	if (user)
672 		atomic_long_sub(pages, &user->locked_vm);
673 }
674 
675 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
676 {
677 	struct user_struct *user = get_current_user();
678 	int ret;
679 
680 	ret = __bpf_prog_charge(user, prog->pages);
681 	if (ret) {
682 		free_uid(user);
683 		return ret;
684 	}
685 
686 	prog->aux->user = user;
687 	return 0;
688 }
689 
690 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
691 {
692 	struct user_struct *user = prog->aux->user;
693 
694 	__bpf_prog_uncharge(user, prog->pages);
695 	free_uid(user);
696 }
697 
698 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
699 {
700 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
701 
702 	free_used_maps(aux);
703 	bpf_prog_uncharge_memlock(aux->prog);
704 	bpf_prog_free(aux->prog);
705 }
706 
707 void bpf_prog_put(struct bpf_prog *prog)
708 {
709 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
710 		trace_bpf_prog_put_rcu(prog);
711 		bpf_prog_kallsyms_del(prog);
712 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
713 	}
714 }
715 EXPORT_SYMBOL_GPL(bpf_prog_put);
716 
717 static int bpf_prog_release(struct inode *inode, struct file *filp)
718 {
719 	struct bpf_prog *prog = filp->private_data;
720 
721 	bpf_prog_put(prog);
722 	return 0;
723 }
724 
725 #ifdef CONFIG_PROC_FS
726 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
727 {
728 	const struct bpf_prog *prog = filp->private_data;
729 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
730 
731 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
732 	seq_printf(m,
733 		   "prog_type:\t%u\n"
734 		   "prog_jited:\t%u\n"
735 		   "prog_tag:\t%s\n"
736 		   "memlock:\t%llu\n",
737 		   prog->type,
738 		   prog->jited,
739 		   prog_tag,
740 		   prog->pages * 1ULL << PAGE_SHIFT);
741 }
742 #endif
743 
744 static const struct file_operations bpf_prog_fops = {
745 #ifdef CONFIG_PROC_FS
746 	.show_fdinfo	= bpf_prog_show_fdinfo,
747 #endif
748 	.release	= bpf_prog_release,
749 };
750 
751 int bpf_prog_new_fd(struct bpf_prog *prog)
752 {
753 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
754 				O_RDWR | O_CLOEXEC);
755 }
756 
757 static struct bpf_prog *____bpf_prog_get(struct fd f)
758 {
759 	if (!f.file)
760 		return ERR_PTR(-EBADF);
761 	if (f.file->f_op != &bpf_prog_fops) {
762 		fdput(f);
763 		return ERR_PTR(-EINVAL);
764 	}
765 
766 	return f.file->private_data;
767 }
768 
769 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
770 {
771 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
772 		atomic_sub(i, &prog->aux->refcnt);
773 		return ERR_PTR(-EBUSY);
774 	}
775 	return prog;
776 }
777 EXPORT_SYMBOL_GPL(bpf_prog_add);
778 
779 void bpf_prog_sub(struct bpf_prog *prog, int i)
780 {
781 	/* Only to be used for undoing previous bpf_prog_add() in some
782 	 * error path. We still know that another entity in our call
783 	 * path holds a reference to the program, thus atomic_sub() can
784 	 * be safely used in such cases!
785 	 */
786 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
787 }
788 EXPORT_SYMBOL_GPL(bpf_prog_sub);
789 
790 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
791 {
792 	return bpf_prog_add(prog, 1);
793 }
794 EXPORT_SYMBOL_GPL(bpf_prog_inc);
795 
796 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type)
797 {
798 	struct fd f = fdget(ufd);
799 	struct bpf_prog *prog;
800 
801 	prog = ____bpf_prog_get(f);
802 	if (IS_ERR(prog))
803 		return prog;
804 	if (type && prog->type != *type) {
805 		prog = ERR_PTR(-EINVAL);
806 		goto out;
807 	}
808 
809 	prog = bpf_prog_inc(prog);
810 out:
811 	fdput(f);
812 	return prog;
813 }
814 
815 struct bpf_prog *bpf_prog_get(u32 ufd)
816 {
817 	return __bpf_prog_get(ufd, NULL);
818 }
819 
820 struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type)
821 {
822 	struct bpf_prog *prog = __bpf_prog_get(ufd, &type);
823 
824 	if (!IS_ERR(prog))
825 		trace_bpf_prog_get_type(prog);
826 	return prog;
827 }
828 EXPORT_SYMBOL_GPL(bpf_prog_get_type);
829 
830 /* last field in 'union bpf_attr' used by this command */
831 #define	BPF_PROG_LOAD_LAST_FIELD kern_version
832 
833 static int bpf_prog_load(union bpf_attr *attr)
834 {
835 	enum bpf_prog_type type = attr->prog_type;
836 	struct bpf_prog *prog;
837 	int err;
838 	char license[128];
839 	bool is_gpl;
840 
841 	if (CHECK_ATTR(BPF_PROG_LOAD))
842 		return -EINVAL;
843 
844 	/* copy eBPF program license from user space */
845 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
846 			      sizeof(license) - 1) < 0)
847 		return -EFAULT;
848 	license[sizeof(license) - 1] = 0;
849 
850 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
851 	is_gpl = license_is_gpl_compatible(license);
852 
853 	if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
854 		return -E2BIG;
855 
856 	if (type == BPF_PROG_TYPE_KPROBE &&
857 	    attr->kern_version != LINUX_VERSION_CODE)
858 		return -EINVAL;
859 
860 	if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN))
861 		return -EPERM;
862 
863 	/* plain bpf_prog allocation */
864 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
865 	if (!prog)
866 		return -ENOMEM;
867 
868 	err = bpf_prog_charge_memlock(prog);
869 	if (err)
870 		goto free_prog_nouncharge;
871 
872 	prog->len = attr->insn_cnt;
873 
874 	err = -EFAULT;
875 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
876 			   bpf_prog_insn_size(prog)) != 0)
877 		goto free_prog;
878 
879 	prog->orig_prog = NULL;
880 	prog->jited = 0;
881 
882 	atomic_set(&prog->aux->refcnt, 1);
883 	prog->gpl_compatible = is_gpl ? 1 : 0;
884 
885 	/* find program type: socket_filter vs tracing_filter */
886 	err = find_prog_type(type, prog);
887 	if (err < 0)
888 		goto free_prog;
889 
890 	/* run eBPF verifier */
891 	err = bpf_check(&prog, attr);
892 	if (err < 0)
893 		goto free_used_maps;
894 
895 	/* fixup BPF_CALL->imm field */
896 	fixup_bpf_calls(prog);
897 
898 	/* eBPF program is ready to be JITed */
899 	prog = bpf_prog_select_runtime(prog, &err);
900 	if (err < 0)
901 		goto free_used_maps;
902 
903 	err = bpf_prog_new_fd(prog);
904 	if (err < 0)
905 		/* failed to allocate fd */
906 		goto free_used_maps;
907 
908 	bpf_prog_kallsyms_add(prog);
909 	trace_bpf_prog_load(prog, err);
910 	return err;
911 
912 free_used_maps:
913 	free_used_maps(prog->aux);
914 free_prog:
915 	bpf_prog_uncharge_memlock(prog);
916 free_prog_nouncharge:
917 	bpf_prog_free(prog);
918 	return err;
919 }
920 
921 #define BPF_OBJ_LAST_FIELD bpf_fd
922 
923 static int bpf_obj_pin(const union bpf_attr *attr)
924 {
925 	if (CHECK_ATTR(BPF_OBJ))
926 		return -EINVAL;
927 
928 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
929 }
930 
931 static int bpf_obj_get(const union bpf_attr *attr)
932 {
933 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0)
934 		return -EINVAL;
935 
936 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname));
937 }
938 
939 #ifdef CONFIG_CGROUP_BPF
940 
941 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
942 
943 static int bpf_prog_attach(const union bpf_attr *attr)
944 {
945 	enum bpf_prog_type ptype;
946 	struct bpf_prog *prog;
947 	struct cgroup *cgrp;
948 	int ret;
949 
950 	if (!capable(CAP_NET_ADMIN))
951 		return -EPERM;
952 
953 	if (CHECK_ATTR(BPF_PROG_ATTACH))
954 		return -EINVAL;
955 
956 	if (attr->attach_flags & ~BPF_F_ALLOW_OVERRIDE)
957 		return -EINVAL;
958 
959 	switch (attr->attach_type) {
960 	case BPF_CGROUP_INET_INGRESS:
961 	case BPF_CGROUP_INET_EGRESS:
962 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
963 		break;
964 	case BPF_CGROUP_INET_SOCK_CREATE:
965 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
966 		break;
967 	default:
968 		return -EINVAL;
969 	}
970 
971 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
972 	if (IS_ERR(prog))
973 		return PTR_ERR(prog);
974 
975 	cgrp = cgroup_get_from_fd(attr->target_fd);
976 	if (IS_ERR(cgrp)) {
977 		bpf_prog_put(prog);
978 		return PTR_ERR(cgrp);
979 	}
980 
981 	ret = cgroup_bpf_update(cgrp, prog, attr->attach_type,
982 				attr->attach_flags & BPF_F_ALLOW_OVERRIDE);
983 	if (ret)
984 		bpf_prog_put(prog);
985 	cgroup_put(cgrp);
986 
987 	return ret;
988 }
989 
990 #define BPF_PROG_DETACH_LAST_FIELD attach_type
991 
992 static int bpf_prog_detach(const union bpf_attr *attr)
993 {
994 	struct cgroup *cgrp;
995 	int ret;
996 
997 	if (!capable(CAP_NET_ADMIN))
998 		return -EPERM;
999 
1000 	if (CHECK_ATTR(BPF_PROG_DETACH))
1001 		return -EINVAL;
1002 
1003 	switch (attr->attach_type) {
1004 	case BPF_CGROUP_INET_INGRESS:
1005 	case BPF_CGROUP_INET_EGRESS:
1006 	case BPF_CGROUP_INET_SOCK_CREATE:
1007 		cgrp = cgroup_get_from_fd(attr->target_fd);
1008 		if (IS_ERR(cgrp))
1009 			return PTR_ERR(cgrp);
1010 
1011 		ret = cgroup_bpf_update(cgrp, NULL, attr->attach_type, false);
1012 		cgroup_put(cgrp);
1013 		break;
1014 
1015 	default:
1016 		return -EINVAL;
1017 	}
1018 
1019 	return ret;
1020 }
1021 #endif /* CONFIG_CGROUP_BPF */
1022 
1023 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
1024 {
1025 	union bpf_attr attr = {};
1026 	int err;
1027 
1028 	if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
1029 		return -EPERM;
1030 
1031 	if (!access_ok(VERIFY_READ, uattr, 1))
1032 		return -EFAULT;
1033 
1034 	if (size > PAGE_SIZE)	/* silly large */
1035 		return -E2BIG;
1036 
1037 	/* If we're handed a bigger struct than we know of,
1038 	 * ensure all the unknown bits are 0 - i.e. new
1039 	 * user-space does not rely on any kernel feature
1040 	 * extensions we dont know about yet.
1041 	 */
1042 	if (size > sizeof(attr)) {
1043 		unsigned char __user *addr;
1044 		unsigned char __user *end;
1045 		unsigned char val;
1046 
1047 		addr = (void __user *)uattr + sizeof(attr);
1048 		end  = (void __user *)uattr + size;
1049 
1050 		for (; addr < end; addr++) {
1051 			err = get_user(val, addr);
1052 			if (err)
1053 				return err;
1054 			if (val)
1055 				return -E2BIG;
1056 		}
1057 		size = sizeof(attr);
1058 	}
1059 
1060 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
1061 	if (copy_from_user(&attr, uattr, size) != 0)
1062 		return -EFAULT;
1063 
1064 	switch (cmd) {
1065 	case BPF_MAP_CREATE:
1066 		err = map_create(&attr);
1067 		break;
1068 	case BPF_MAP_LOOKUP_ELEM:
1069 		err = map_lookup_elem(&attr);
1070 		break;
1071 	case BPF_MAP_UPDATE_ELEM:
1072 		err = map_update_elem(&attr);
1073 		break;
1074 	case BPF_MAP_DELETE_ELEM:
1075 		err = map_delete_elem(&attr);
1076 		break;
1077 	case BPF_MAP_GET_NEXT_KEY:
1078 		err = map_get_next_key(&attr);
1079 		break;
1080 	case BPF_PROG_LOAD:
1081 		err = bpf_prog_load(&attr);
1082 		break;
1083 	case BPF_OBJ_PIN:
1084 		err = bpf_obj_pin(&attr);
1085 		break;
1086 	case BPF_OBJ_GET:
1087 		err = bpf_obj_get(&attr);
1088 		break;
1089 
1090 #ifdef CONFIG_CGROUP_BPF
1091 	case BPF_PROG_ATTACH:
1092 		err = bpf_prog_attach(&attr);
1093 		break;
1094 	case BPF_PROG_DETACH:
1095 		err = bpf_prog_detach(&attr);
1096 		break;
1097 #endif
1098 
1099 	default:
1100 		err = -EINVAL;
1101 		break;
1102 	}
1103 
1104 	return err;
1105 }
1106