xref: /openbmc/linux/security/device_cgroup.c (revision 95298d63)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * device_cgroup.c - device cgroup subsystem
4  *
5  * Copyright 2007 IBM Corp
6  */
7 
8 #include <linux/device_cgroup.h>
9 #include <linux/cgroup.h>
10 #include <linux/ctype.h>
11 #include <linux/list.h>
12 #include <linux/uaccess.h>
13 #include <linux/seq_file.h>
14 #include <linux/slab.h>
15 #include <linux/rcupdate.h>
16 #include <linux/mutex.h>
17 
18 #ifdef CONFIG_CGROUP_DEVICE
19 
20 static DEFINE_MUTEX(devcgroup_mutex);
21 
22 enum devcg_behavior {
23 	DEVCG_DEFAULT_NONE,
24 	DEVCG_DEFAULT_ALLOW,
25 	DEVCG_DEFAULT_DENY,
26 };
27 
28 /*
29  * exception list locking rules:
30  * hold devcgroup_mutex for update/read.
31  * hold rcu_read_lock() for read.
32  */
33 
34 struct dev_exception_item {
35 	u32 major, minor;
36 	short type;
37 	short access;
38 	struct list_head list;
39 	struct rcu_head rcu;
40 };
41 
42 struct dev_cgroup {
43 	struct cgroup_subsys_state css;
44 	struct list_head exceptions;
45 	enum devcg_behavior behavior;
46 };
47 
48 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
49 {
50 	return s ? container_of(s, struct dev_cgroup, css) : NULL;
51 }
52 
53 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
54 {
55 	return css_to_devcgroup(task_css(task, devices_cgrp_id));
56 }
57 
58 /*
59  * called under devcgroup_mutex
60  */
61 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
62 {
63 	struct dev_exception_item *ex, *tmp, *new;
64 
65 	lockdep_assert_held(&devcgroup_mutex);
66 
67 	list_for_each_entry(ex, orig, list) {
68 		new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
69 		if (!new)
70 			goto free_and_exit;
71 		list_add_tail(&new->list, dest);
72 	}
73 
74 	return 0;
75 
76 free_and_exit:
77 	list_for_each_entry_safe(ex, tmp, dest, list) {
78 		list_del(&ex->list);
79 		kfree(ex);
80 	}
81 	return -ENOMEM;
82 }
83 
84 /*
85  * called under devcgroup_mutex
86  */
87 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
88 			     struct dev_exception_item *ex)
89 {
90 	struct dev_exception_item *excopy, *walk;
91 
92 	lockdep_assert_held(&devcgroup_mutex);
93 
94 	excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
95 	if (!excopy)
96 		return -ENOMEM;
97 
98 	list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
99 		if (walk->type != ex->type)
100 			continue;
101 		if (walk->major != ex->major)
102 			continue;
103 		if (walk->minor != ex->minor)
104 			continue;
105 
106 		walk->access |= ex->access;
107 		kfree(excopy);
108 		excopy = NULL;
109 	}
110 
111 	if (excopy != NULL)
112 		list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
113 	return 0;
114 }
115 
116 /*
117  * called under devcgroup_mutex
118  */
119 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
120 			     struct dev_exception_item *ex)
121 {
122 	struct dev_exception_item *walk, *tmp;
123 
124 	lockdep_assert_held(&devcgroup_mutex);
125 
126 	list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
127 		if (walk->type != ex->type)
128 			continue;
129 		if (walk->major != ex->major)
130 			continue;
131 		if (walk->minor != ex->minor)
132 			continue;
133 
134 		walk->access &= ~ex->access;
135 		if (!walk->access) {
136 			list_del_rcu(&walk->list);
137 			kfree_rcu(walk, rcu);
138 		}
139 	}
140 }
141 
142 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
143 {
144 	struct dev_exception_item *ex, *tmp;
145 
146 	list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
147 		list_del_rcu(&ex->list);
148 		kfree_rcu(ex, rcu);
149 	}
150 }
151 
152 /**
153  * dev_exception_clean - frees all entries of the exception list
154  * @dev_cgroup: dev_cgroup with the exception list to be cleaned
155  *
156  * called under devcgroup_mutex
157  */
158 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
159 {
160 	lockdep_assert_held(&devcgroup_mutex);
161 
162 	__dev_exception_clean(dev_cgroup);
163 }
164 
165 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
166 {
167 	return (devcg->behavior != DEVCG_DEFAULT_NONE);
168 }
169 
170 /**
171  * devcgroup_online - initializes devcgroup's behavior and exceptions based on
172  * 		      parent's
173  * @css: css getting online
174  * returns 0 in case of success, error code otherwise
175  */
176 static int devcgroup_online(struct cgroup_subsys_state *css)
177 {
178 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
179 	struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
180 	int ret = 0;
181 
182 	mutex_lock(&devcgroup_mutex);
183 
184 	if (parent_dev_cgroup == NULL)
185 		dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
186 	else {
187 		ret = dev_exceptions_copy(&dev_cgroup->exceptions,
188 					  &parent_dev_cgroup->exceptions);
189 		if (!ret)
190 			dev_cgroup->behavior = parent_dev_cgroup->behavior;
191 	}
192 	mutex_unlock(&devcgroup_mutex);
193 
194 	return ret;
195 }
196 
197 static void devcgroup_offline(struct cgroup_subsys_state *css)
198 {
199 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
200 
201 	mutex_lock(&devcgroup_mutex);
202 	dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
203 	mutex_unlock(&devcgroup_mutex);
204 }
205 
206 /*
207  * called from kernel/cgroup.c with cgroup_lock() held.
208  */
209 static struct cgroup_subsys_state *
210 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
211 {
212 	struct dev_cgroup *dev_cgroup;
213 
214 	dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
215 	if (!dev_cgroup)
216 		return ERR_PTR(-ENOMEM);
217 	INIT_LIST_HEAD(&dev_cgroup->exceptions);
218 	dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
219 
220 	return &dev_cgroup->css;
221 }
222 
223 static void devcgroup_css_free(struct cgroup_subsys_state *css)
224 {
225 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
226 
227 	__dev_exception_clean(dev_cgroup);
228 	kfree(dev_cgroup);
229 }
230 
231 #define DEVCG_ALLOW 1
232 #define DEVCG_DENY 2
233 #define DEVCG_LIST 3
234 
235 #define MAJMINLEN 13
236 #define ACCLEN 4
237 
238 static void set_access(char *acc, short access)
239 {
240 	int idx = 0;
241 	memset(acc, 0, ACCLEN);
242 	if (access & DEVCG_ACC_READ)
243 		acc[idx++] = 'r';
244 	if (access & DEVCG_ACC_WRITE)
245 		acc[idx++] = 'w';
246 	if (access & DEVCG_ACC_MKNOD)
247 		acc[idx++] = 'm';
248 }
249 
250 static char type_to_char(short type)
251 {
252 	if (type == DEVCG_DEV_ALL)
253 		return 'a';
254 	if (type == DEVCG_DEV_CHAR)
255 		return 'c';
256 	if (type == DEVCG_DEV_BLOCK)
257 		return 'b';
258 	return 'X';
259 }
260 
261 static void set_majmin(char *str, unsigned m)
262 {
263 	if (m == ~0)
264 		strcpy(str, "*");
265 	else
266 		sprintf(str, "%u", m);
267 }
268 
269 static int devcgroup_seq_show(struct seq_file *m, void *v)
270 {
271 	struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
272 	struct dev_exception_item *ex;
273 	char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
274 
275 	rcu_read_lock();
276 	/*
277 	 * To preserve the compatibility:
278 	 * - Only show the "all devices" when the default policy is to allow
279 	 * - List the exceptions in case the default policy is to deny
280 	 * This way, the file remains as a "whitelist of devices"
281 	 */
282 	if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
283 		set_access(acc, DEVCG_ACC_MASK);
284 		set_majmin(maj, ~0);
285 		set_majmin(min, ~0);
286 		seq_printf(m, "%c %s:%s %s\n", type_to_char(DEVCG_DEV_ALL),
287 			   maj, min, acc);
288 	} else {
289 		list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
290 			set_access(acc, ex->access);
291 			set_majmin(maj, ex->major);
292 			set_majmin(min, ex->minor);
293 			seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
294 				   maj, min, acc);
295 		}
296 	}
297 	rcu_read_unlock();
298 
299 	return 0;
300 }
301 
302 /**
303  * match_exception	- iterates the exception list trying to find a complete match
304  * @exceptions: list of exceptions
305  * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
306  * @major: device file major number, ~0 to match all
307  * @minor: device file minor number, ~0 to match all
308  * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
309  *
310  * It is considered a complete match if an exception is found that will
311  * contain the entire range of provided parameters.
312  *
313  * Return: true in case it matches an exception completely
314  */
315 static bool match_exception(struct list_head *exceptions, short type,
316 			    u32 major, u32 minor, short access)
317 {
318 	struct dev_exception_item *ex;
319 
320 	list_for_each_entry_rcu(ex, exceptions, list) {
321 		if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
322 			continue;
323 		if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
324 			continue;
325 		if (ex->major != ~0 && ex->major != major)
326 			continue;
327 		if (ex->minor != ~0 && ex->minor != minor)
328 			continue;
329 		/* provided access cannot have more than the exception rule */
330 		if (access & (~ex->access))
331 			continue;
332 		return true;
333 	}
334 	return false;
335 }
336 
337 /**
338  * match_exception_partial - iterates the exception list trying to find a partial match
339  * @exceptions: list of exceptions
340  * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
341  * @major: device file major number, ~0 to match all
342  * @minor: device file minor number, ~0 to match all
343  * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
344  *
345  * It is considered a partial match if an exception's range is found to
346  * contain *any* of the devices specified by provided parameters. This is
347  * used to make sure no extra access is being granted that is forbidden by
348  * any of the exception list.
349  *
350  * Return: true in case the provided range mat matches an exception completely
351  */
352 static bool match_exception_partial(struct list_head *exceptions, short type,
353 				    u32 major, u32 minor, short access)
354 {
355 	struct dev_exception_item *ex;
356 
357 	list_for_each_entry_rcu(ex, exceptions, list) {
358 		if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
359 			continue;
360 		if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
361 			continue;
362 		/*
363 		 * We must be sure that both the exception and the provided
364 		 * range aren't masking all devices
365 		 */
366 		if (ex->major != ~0 && major != ~0 && ex->major != major)
367 			continue;
368 		if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
369 			continue;
370 		/*
371 		 * In order to make sure the provided range isn't matching
372 		 * an exception, all its access bits shouldn't match the
373 		 * exception's access bits
374 		 */
375 		if (!(access & ex->access))
376 			continue;
377 		return true;
378 	}
379 	return false;
380 }
381 
382 /**
383  * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
384  * @dev_cgroup: dev cgroup to be tested against
385  * @refex: new exception
386  * @behavior: behavior of the exception's dev_cgroup
387  *
388  * This is used to make sure a child cgroup won't have more privileges
389  * than its parent
390  */
391 static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
392 		          struct dev_exception_item *refex,
393 		          enum devcg_behavior behavior)
394 {
395 	bool match = false;
396 
397 	RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
398 			 !lockdep_is_held(&devcgroup_mutex),
399 			 "device_cgroup:verify_new_ex called without proper synchronization");
400 
401 	if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
402 		if (behavior == DEVCG_DEFAULT_ALLOW) {
403 			/*
404 			 * new exception in the child doesn't matter, only
405 			 * adding extra restrictions
406 			 */
407 			return true;
408 		} else {
409 			/*
410 			 * new exception in the child will add more devices
411 			 * that can be acessed, so it can't match any of
412 			 * parent's exceptions, even slightly
413 			 */
414 			match = match_exception_partial(&dev_cgroup->exceptions,
415 							refex->type,
416 							refex->major,
417 							refex->minor,
418 							refex->access);
419 
420 			if (match)
421 				return false;
422 			return true;
423 		}
424 	} else {
425 		/*
426 		 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
427 		 * the new exception will add access to more devices and must
428 		 * be contained completely in an parent's exception to be
429 		 * allowed
430 		 */
431 		match = match_exception(&dev_cgroup->exceptions, refex->type,
432 					refex->major, refex->minor,
433 					refex->access);
434 
435 		if (match)
436 			/* parent has an exception that matches the proposed */
437 			return true;
438 		else
439 			return false;
440 	}
441 	return false;
442 }
443 
444 /*
445  * parent_has_perm:
446  * when adding a new allow rule to a device exception list, the rule
447  * must be allowed in the parent device
448  */
449 static int parent_has_perm(struct dev_cgroup *childcg,
450 				  struct dev_exception_item *ex)
451 {
452 	struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
453 
454 	if (!parent)
455 		return 1;
456 	return verify_new_ex(parent, ex, childcg->behavior);
457 }
458 
459 /**
460  * parent_allows_removal - verify if it's ok to remove an exception
461  * @childcg: child cgroup from where the exception will be removed
462  * @ex: exception being removed
463  *
464  * When removing an exception in cgroups with default ALLOW policy, it must
465  * be checked if removing it will give the child cgroup more access than the
466  * parent.
467  *
468  * Return: true if it's ok to remove exception, false otherwise
469  */
470 static bool parent_allows_removal(struct dev_cgroup *childcg,
471 				  struct dev_exception_item *ex)
472 {
473 	struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
474 
475 	if (!parent)
476 		return true;
477 
478 	/* It's always allowed to remove access to devices */
479 	if (childcg->behavior == DEVCG_DEFAULT_DENY)
480 		return true;
481 
482 	/*
483 	 * Make sure you're not removing part or a whole exception existing in
484 	 * the parent cgroup
485 	 */
486 	return !match_exception_partial(&parent->exceptions, ex->type,
487 					ex->major, ex->minor, ex->access);
488 }
489 
490 /**
491  * may_allow_all - checks if it's possible to change the behavior to
492  *		   allow based on parent's rules.
493  * @parent: device cgroup's parent
494  * returns: != 0 in case it's allowed, 0 otherwise
495  */
496 static inline int may_allow_all(struct dev_cgroup *parent)
497 {
498 	if (!parent)
499 		return 1;
500 	return parent->behavior == DEVCG_DEFAULT_ALLOW;
501 }
502 
503 /**
504  * revalidate_active_exceptions - walks through the active exception list and
505  * 				  revalidates the exceptions based on parent's
506  * 				  behavior and exceptions. The exceptions that
507  * 				  are no longer valid will be removed.
508  * 				  Called with devcgroup_mutex held.
509  * @devcg: cgroup which exceptions will be checked
510  *
511  * This is one of the three key functions for hierarchy implementation.
512  * This function is responsible for re-evaluating all the cgroup's active
513  * exceptions due to a parent's exception change.
514  * Refer to Documentation/admin-guide/cgroup-v1/devices.rst for more details.
515  */
516 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
517 {
518 	struct dev_exception_item *ex;
519 	struct list_head *this, *tmp;
520 
521 	list_for_each_safe(this, tmp, &devcg->exceptions) {
522 		ex = container_of(this, struct dev_exception_item, list);
523 		if (!parent_has_perm(devcg, ex))
524 			dev_exception_rm(devcg, ex);
525 	}
526 }
527 
528 /**
529  * propagate_exception - propagates a new exception to the children
530  * @devcg_root: device cgroup that added a new exception
531  * @ex: new exception to be propagated
532  *
533  * returns: 0 in case of success, != 0 in case of error
534  */
535 static int propagate_exception(struct dev_cgroup *devcg_root,
536 			       struct dev_exception_item *ex)
537 {
538 	struct cgroup_subsys_state *pos;
539 	int rc = 0;
540 
541 	rcu_read_lock();
542 
543 	css_for_each_descendant_pre(pos, &devcg_root->css) {
544 		struct dev_cgroup *devcg = css_to_devcgroup(pos);
545 
546 		/*
547 		 * Because devcgroup_mutex is held, no devcg will become
548 		 * online or offline during the tree walk (see on/offline
549 		 * methods), and online ones are safe to access outside RCU
550 		 * read lock without bumping refcnt.
551 		 */
552 		if (pos == &devcg_root->css || !is_devcg_online(devcg))
553 			continue;
554 
555 		rcu_read_unlock();
556 
557 		/*
558 		 * in case both root's behavior and devcg is allow, a new
559 		 * restriction means adding to the exception list
560 		 */
561 		if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
562 		    devcg->behavior == DEVCG_DEFAULT_ALLOW) {
563 			rc = dev_exception_add(devcg, ex);
564 			if (rc)
565 				return rc;
566 		} else {
567 			/*
568 			 * in the other possible cases:
569 			 * root's behavior: allow, devcg's: deny
570 			 * root's behavior: deny, devcg's: deny
571 			 * the exception will be removed
572 			 */
573 			dev_exception_rm(devcg, ex);
574 		}
575 		revalidate_active_exceptions(devcg);
576 
577 		rcu_read_lock();
578 	}
579 
580 	rcu_read_unlock();
581 	return rc;
582 }
583 
584 /*
585  * Modify the exception list using allow/deny rules.
586  * CAP_SYS_ADMIN is needed for this.  It's at least separate from CAP_MKNOD
587  * so we can give a container CAP_MKNOD to let it create devices but not
588  * modify the exception list.
589  * It seems likely we'll want to add a CAP_CONTAINER capability to allow
590  * us to also grant CAP_SYS_ADMIN to containers without giving away the
591  * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
592  *
593  * Taking rules away is always allowed (given CAP_SYS_ADMIN).  Granting
594  * new access is only allowed if you're in the top-level cgroup, or your
595  * parent cgroup has the access you're asking for.
596  */
597 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
598 				   int filetype, char *buffer)
599 {
600 	const char *b;
601 	char temp[12];		/* 11 + 1 characters needed for a u32 */
602 	int count, rc = 0;
603 	struct dev_exception_item ex;
604 	struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
605 
606 	if (!capable(CAP_SYS_ADMIN))
607 		return -EPERM;
608 
609 	memset(&ex, 0, sizeof(ex));
610 	b = buffer;
611 
612 	switch (*b) {
613 	case 'a':
614 		switch (filetype) {
615 		case DEVCG_ALLOW:
616 			if (css_has_online_children(&devcgroup->css))
617 				return -EINVAL;
618 
619 			if (!may_allow_all(parent))
620 				return -EPERM;
621 			dev_exception_clean(devcgroup);
622 			devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
623 			if (!parent)
624 				break;
625 
626 			rc = dev_exceptions_copy(&devcgroup->exceptions,
627 						 &parent->exceptions);
628 			if (rc)
629 				return rc;
630 			break;
631 		case DEVCG_DENY:
632 			if (css_has_online_children(&devcgroup->css))
633 				return -EINVAL;
634 
635 			dev_exception_clean(devcgroup);
636 			devcgroup->behavior = DEVCG_DEFAULT_DENY;
637 			break;
638 		default:
639 			return -EINVAL;
640 		}
641 		return 0;
642 	case 'b':
643 		ex.type = DEVCG_DEV_BLOCK;
644 		break;
645 	case 'c':
646 		ex.type = DEVCG_DEV_CHAR;
647 		break;
648 	default:
649 		return -EINVAL;
650 	}
651 	b++;
652 	if (!isspace(*b))
653 		return -EINVAL;
654 	b++;
655 	if (*b == '*') {
656 		ex.major = ~0;
657 		b++;
658 	} else if (isdigit(*b)) {
659 		memset(temp, 0, sizeof(temp));
660 		for (count = 0; count < sizeof(temp) - 1; count++) {
661 			temp[count] = *b;
662 			b++;
663 			if (!isdigit(*b))
664 				break;
665 		}
666 		rc = kstrtou32(temp, 10, &ex.major);
667 		if (rc)
668 			return -EINVAL;
669 	} else {
670 		return -EINVAL;
671 	}
672 	if (*b != ':')
673 		return -EINVAL;
674 	b++;
675 
676 	/* read minor */
677 	if (*b == '*') {
678 		ex.minor = ~0;
679 		b++;
680 	} else if (isdigit(*b)) {
681 		memset(temp, 0, sizeof(temp));
682 		for (count = 0; count < sizeof(temp) - 1; count++) {
683 			temp[count] = *b;
684 			b++;
685 			if (!isdigit(*b))
686 				break;
687 		}
688 		rc = kstrtou32(temp, 10, &ex.minor);
689 		if (rc)
690 			return -EINVAL;
691 	} else {
692 		return -EINVAL;
693 	}
694 	if (!isspace(*b))
695 		return -EINVAL;
696 	for (b++, count = 0; count < 3; count++, b++) {
697 		switch (*b) {
698 		case 'r':
699 			ex.access |= DEVCG_ACC_READ;
700 			break;
701 		case 'w':
702 			ex.access |= DEVCG_ACC_WRITE;
703 			break;
704 		case 'm':
705 			ex.access |= DEVCG_ACC_MKNOD;
706 			break;
707 		case '\n':
708 		case '\0':
709 			count = 3;
710 			break;
711 		default:
712 			return -EINVAL;
713 		}
714 	}
715 
716 	switch (filetype) {
717 	case DEVCG_ALLOW:
718 		/*
719 		 * If the default policy is to allow by default, try to remove
720 		 * an matching exception instead. And be silent about it: we
721 		 * don't want to break compatibility
722 		 */
723 		if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
724 			/* Check if the parent allows removing it first */
725 			if (!parent_allows_removal(devcgroup, &ex))
726 				return -EPERM;
727 			dev_exception_rm(devcgroup, &ex);
728 			break;
729 		}
730 
731 		if (!parent_has_perm(devcgroup, &ex))
732 			return -EPERM;
733 		rc = dev_exception_add(devcgroup, &ex);
734 		break;
735 	case DEVCG_DENY:
736 		/*
737 		 * If the default policy is to deny by default, try to remove
738 		 * an matching exception instead. And be silent about it: we
739 		 * don't want to break compatibility
740 		 */
741 		if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
742 			dev_exception_rm(devcgroup, &ex);
743 		else
744 			rc = dev_exception_add(devcgroup, &ex);
745 
746 		if (rc)
747 			break;
748 		/* we only propagate new restrictions */
749 		rc = propagate_exception(devcgroup, &ex);
750 		break;
751 	default:
752 		rc = -EINVAL;
753 	}
754 	return rc;
755 }
756 
757 static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
758 				      char *buf, size_t nbytes, loff_t off)
759 {
760 	int retval;
761 
762 	mutex_lock(&devcgroup_mutex);
763 	retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
764 					 of_cft(of)->private, strstrip(buf));
765 	mutex_unlock(&devcgroup_mutex);
766 	return retval ?: nbytes;
767 }
768 
769 static struct cftype dev_cgroup_files[] = {
770 	{
771 		.name = "allow",
772 		.write = devcgroup_access_write,
773 		.private = DEVCG_ALLOW,
774 	},
775 	{
776 		.name = "deny",
777 		.write = devcgroup_access_write,
778 		.private = DEVCG_DENY,
779 	},
780 	{
781 		.name = "list",
782 		.seq_show = devcgroup_seq_show,
783 		.private = DEVCG_LIST,
784 	},
785 	{ }	/* terminate */
786 };
787 
788 struct cgroup_subsys devices_cgrp_subsys = {
789 	.css_alloc = devcgroup_css_alloc,
790 	.css_free = devcgroup_css_free,
791 	.css_online = devcgroup_online,
792 	.css_offline = devcgroup_offline,
793 	.legacy_cftypes = dev_cgroup_files,
794 };
795 
796 /**
797  * devcgroup_legacy_check_permission - checks if an inode operation is permitted
798  * @dev_cgroup: the dev cgroup to be tested against
799  * @type: device type
800  * @major: device major number
801  * @minor: device minor number
802  * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
803  *
804  * returns 0 on success, -EPERM case the operation is not permitted
805  */
806 static int devcgroup_legacy_check_permission(short type, u32 major, u32 minor,
807 					short access)
808 {
809 	struct dev_cgroup *dev_cgroup;
810 	bool rc;
811 
812 	rcu_read_lock();
813 	dev_cgroup = task_devcgroup(current);
814 	if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
815 		/* Can't match any of the exceptions, even partially */
816 		rc = !match_exception_partial(&dev_cgroup->exceptions,
817 					      type, major, minor, access);
818 	else
819 		/* Need to match completely one exception to be allowed */
820 		rc = match_exception(&dev_cgroup->exceptions, type, major,
821 				     minor, access);
822 	rcu_read_unlock();
823 
824 	if (!rc)
825 		return -EPERM;
826 
827 	return 0;
828 }
829 
830 #endif /* CONFIG_CGROUP_DEVICE */
831 
832 #if defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF)
833 
834 int devcgroup_check_permission(short type, u32 major, u32 minor, short access)
835 {
836 	int rc = BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type, major, minor, access);
837 
838 	if (rc)
839 		return -EPERM;
840 
841 	#ifdef CONFIG_CGROUP_DEVICE
842 	return devcgroup_legacy_check_permission(type, major, minor, access);
843 
844 	#else /* CONFIG_CGROUP_DEVICE */
845 	return 0;
846 
847 	#endif /* CONFIG_CGROUP_DEVICE */
848 }
849 EXPORT_SYMBOL(devcgroup_check_permission);
850 #endif /* defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF) */
851