xref: /openbmc/linux/security/device_cgroup.c (revision 0e17c50f)
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 static DEFINE_MUTEX(devcgroup_mutex);
19 
20 enum devcg_behavior {
21 	DEVCG_DEFAULT_NONE,
22 	DEVCG_DEFAULT_ALLOW,
23 	DEVCG_DEFAULT_DENY,
24 };
25 
26 /*
27  * exception list locking rules:
28  * hold devcgroup_mutex for update/read.
29  * hold rcu_read_lock() for read.
30  */
31 
32 struct dev_exception_item {
33 	u32 major, minor;
34 	short type;
35 	short access;
36 	struct list_head list;
37 	struct rcu_head rcu;
38 };
39 
40 struct dev_cgroup {
41 	struct cgroup_subsys_state css;
42 	struct list_head exceptions;
43 	enum devcg_behavior behavior;
44 };
45 
46 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
47 {
48 	return s ? container_of(s, struct dev_cgroup, css) : NULL;
49 }
50 
51 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
52 {
53 	return css_to_devcgroup(task_css(task, devices_cgrp_id));
54 }
55 
56 /*
57  * called under devcgroup_mutex
58  */
59 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
60 {
61 	struct dev_exception_item *ex, *tmp, *new;
62 
63 	lockdep_assert_held(&devcgroup_mutex);
64 
65 	list_for_each_entry(ex, orig, list) {
66 		new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
67 		if (!new)
68 			goto free_and_exit;
69 		list_add_tail(&new->list, dest);
70 	}
71 
72 	return 0;
73 
74 free_and_exit:
75 	list_for_each_entry_safe(ex, tmp, dest, list) {
76 		list_del(&ex->list);
77 		kfree(ex);
78 	}
79 	return -ENOMEM;
80 }
81 
82 /*
83  * called under devcgroup_mutex
84  */
85 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
86 			     struct dev_exception_item *ex)
87 {
88 	struct dev_exception_item *excopy, *walk;
89 
90 	lockdep_assert_held(&devcgroup_mutex);
91 
92 	excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
93 	if (!excopy)
94 		return -ENOMEM;
95 
96 	list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
97 		if (walk->type != ex->type)
98 			continue;
99 		if (walk->major != ex->major)
100 			continue;
101 		if (walk->minor != ex->minor)
102 			continue;
103 
104 		walk->access |= ex->access;
105 		kfree(excopy);
106 		excopy = NULL;
107 	}
108 
109 	if (excopy != NULL)
110 		list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
111 	return 0;
112 }
113 
114 /*
115  * called under devcgroup_mutex
116  */
117 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
118 			     struct dev_exception_item *ex)
119 {
120 	struct dev_exception_item *walk, *tmp;
121 
122 	lockdep_assert_held(&devcgroup_mutex);
123 
124 	list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
125 		if (walk->type != ex->type)
126 			continue;
127 		if (walk->major != ex->major)
128 			continue;
129 		if (walk->minor != ex->minor)
130 			continue;
131 
132 		walk->access &= ~ex->access;
133 		if (!walk->access) {
134 			list_del_rcu(&walk->list);
135 			kfree_rcu(walk, rcu);
136 		}
137 	}
138 }
139 
140 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
141 {
142 	struct dev_exception_item *ex, *tmp;
143 
144 	list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
145 		list_del_rcu(&ex->list);
146 		kfree_rcu(ex, rcu);
147 	}
148 }
149 
150 /**
151  * dev_exception_clean - frees all entries of the exception list
152  * @dev_cgroup: dev_cgroup with the exception list to be cleaned
153  *
154  * called under devcgroup_mutex
155  */
156 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
157 {
158 	lockdep_assert_held(&devcgroup_mutex);
159 
160 	__dev_exception_clean(dev_cgroup);
161 }
162 
163 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
164 {
165 	return (devcg->behavior != DEVCG_DEFAULT_NONE);
166 }
167 
168 /**
169  * devcgroup_online - initializes devcgroup's behavior and exceptions based on
170  * 		      parent's
171  * @css: css getting online
172  * returns 0 in case of success, error code otherwise
173  */
174 static int devcgroup_online(struct cgroup_subsys_state *css)
175 {
176 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
177 	struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
178 	int ret = 0;
179 
180 	mutex_lock(&devcgroup_mutex);
181 
182 	if (parent_dev_cgroup == NULL)
183 		dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
184 	else {
185 		ret = dev_exceptions_copy(&dev_cgroup->exceptions,
186 					  &parent_dev_cgroup->exceptions);
187 		if (!ret)
188 			dev_cgroup->behavior = parent_dev_cgroup->behavior;
189 	}
190 	mutex_unlock(&devcgroup_mutex);
191 
192 	return ret;
193 }
194 
195 static void devcgroup_offline(struct cgroup_subsys_state *css)
196 {
197 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
198 
199 	mutex_lock(&devcgroup_mutex);
200 	dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
201 	mutex_unlock(&devcgroup_mutex);
202 }
203 
204 /*
205  * called from kernel/cgroup.c with cgroup_lock() held.
206  */
207 static struct cgroup_subsys_state *
208 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
209 {
210 	struct dev_cgroup *dev_cgroup;
211 
212 	dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
213 	if (!dev_cgroup)
214 		return ERR_PTR(-ENOMEM);
215 	INIT_LIST_HEAD(&dev_cgroup->exceptions);
216 	dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
217 
218 	return &dev_cgroup->css;
219 }
220 
221 static void devcgroup_css_free(struct cgroup_subsys_state *css)
222 {
223 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
224 
225 	__dev_exception_clean(dev_cgroup);
226 	kfree(dev_cgroup);
227 }
228 
229 #define DEVCG_ALLOW 1
230 #define DEVCG_DENY 2
231 #define DEVCG_LIST 3
232 
233 #define MAJMINLEN 13
234 #define ACCLEN 4
235 
236 static void set_access(char *acc, short access)
237 {
238 	int idx = 0;
239 	memset(acc, 0, ACCLEN);
240 	if (access & DEVCG_ACC_READ)
241 		acc[idx++] = 'r';
242 	if (access & DEVCG_ACC_WRITE)
243 		acc[idx++] = 'w';
244 	if (access & DEVCG_ACC_MKNOD)
245 		acc[idx++] = 'm';
246 }
247 
248 static char type_to_char(short type)
249 {
250 	if (type == DEVCG_DEV_ALL)
251 		return 'a';
252 	if (type == DEVCG_DEV_CHAR)
253 		return 'c';
254 	if (type == DEVCG_DEV_BLOCK)
255 		return 'b';
256 	return 'X';
257 }
258 
259 static void set_majmin(char *str, unsigned m)
260 {
261 	if (m == ~0)
262 		strcpy(str, "*");
263 	else
264 		sprintf(str, "%u", m);
265 }
266 
267 static int devcgroup_seq_show(struct seq_file *m, void *v)
268 {
269 	struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
270 	struct dev_exception_item *ex;
271 	char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
272 
273 	rcu_read_lock();
274 	/*
275 	 * To preserve the compatibility:
276 	 * - Only show the "all devices" when the default policy is to allow
277 	 * - List the exceptions in case the default policy is to deny
278 	 * This way, the file remains as a "whitelist of devices"
279 	 */
280 	if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
281 		set_access(acc, DEVCG_ACC_MASK);
282 		set_majmin(maj, ~0);
283 		set_majmin(min, ~0);
284 		seq_printf(m, "%c %s:%s %s\n", type_to_char(DEVCG_DEV_ALL),
285 			   maj, min, acc);
286 	} else {
287 		list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
288 			set_access(acc, ex->access);
289 			set_majmin(maj, ex->major);
290 			set_majmin(min, ex->minor);
291 			seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
292 				   maj, min, acc);
293 		}
294 	}
295 	rcu_read_unlock();
296 
297 	return 0;
298 }
299 
300 /**
301  * match_exception	- iterates the exception list trying to find a complete match
302  * @exceptions: list of exceptions
303  * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
304  * @major: device file major number, ~0 to match all
305  * @minor: device file minor number, ~0 to match all
306  * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
307  *
308  * It is considered a complete match if an exception is found that will
309  * contain the entire range of provided parameters.
310  *
311  * Return: true in case it matches an exception completely
312  */
313 static bool match_exception(struct list_head *exceptions, short type,
314 			    u32 major, u32 minor, short access)
315 {
316 	struct dev_exception_item *ex;
317 
318 	list_for_each_entry_rcu(ex, exceptions, list) {
319 		if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
320 			continue;
321 		if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
322 			continue;
323 		if (ex->major != ~0 && ex->major != major)
324 			continue;
325 		if (ex->minor != ~0 && ex->minor != minor)
326 			continue;
327 		/* provided access cannot have more than the exception rule */
328 		if (access & (~ex->access))
329 			continue;
330 		return true;
331 	}
332 	return false;
333 }
334 
335 /**
336  * match_exception_partial - iterates the exception list trying to find a partial match
337  * @exceptions: list of exceptions
338  * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
339  * @major: device file major number, ~0 to match all
340  * @minor: device file minor number, ~0 to match all
341  * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
342  *
343  * It is considered a partial match if an exception's range is found to
344  * contain *any* of the devices specified by provided parameters. This is
345  * used to make sure no extra access is being granted that is forbidden by
346  * any of the exception list.
347  *
348  * Return: true in case the provided range mat matches an exception completely
349  */
350 static bool match_exception_partial(struct list_head *exceptions, short type,
351 				    u32 major, u32 minor, short access)
352 {
353 	struct dev_exception_item *ex;
354 
355 	list_for_each_entry_rcu(ex, exceptions, list) {
356 		if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
357 			continue;
358 		if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
359 			continue;
360 		/*
361 		 * We must be sure that both the exception and the provided
362 		 * range aren't masking all devices
363 		 */
364 		if (ex->major != ~0 && major != ~0 && ex->major != major)
365 			continue;
366 		if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
367 			continue;
368 		/*
369 		 * In order to make sure the provided range isn't matching
370 		 * an exception, all its access bits shouldn't match the
371 		 * exception's access bits
372 		 */
373 		if (!(access & ex->access))
374 			continue;
375 		return true;
376 	}
377 	return false;
378 }
379 
380 /**
381  * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
382  * @dev_cgroup: dev cgroup to be tested against
383  * @refex: new exception
384  * @behavior: behavior of the exception's dev_cgroup
385  *
386  * This is used to make sure a child cgroup won't have more privileges
387  * than its parent
388  */
389 static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
390 		          struct dev_exception_item *refex,
391 		          enum devcg_behavior behavior)
392 {
393 	bool match = false;
394 
395 	RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
396 			 !lockdep_is_held(&devcgroup_mutex),
397 			 "device_cgroup:verify_new_ex called without proper synchronization");
398 
399 	if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
400 		if (behavior == DEVCG_DEFAULT_ALLOW) {
401 			/*
402 			 * new exception in the child doesn't matter, only
403 			 * adding extra restrictions
404 			 */
405 			return true;
406 		} else {
407 			/*
408 			 * new exception in the child will add more devices
409 			 * that can be acessed, so it can't match any of
410 			 * parent's exceptions, even slightly
411 			 */
412 			match = match_exception_partial(&dev_cgroup->exceptions,
413 							refex->type,
414 							refex->major,
415 							refex->minor,
416 							refex->access);
417 
418 			if (match)
419 				return false;
420 			return true;
421 		}
422 	} else {
423 		/*
424 		 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
425 		 * the new exception will add access to more devices and must
426 		 * be contained completely in an parent's exception to be
427 		 * allowed
428 		 */
429 		match = match_exception(&dev_cgroup->exceptions, refex->type,
430 					refex->major, refex->minor,
431 					refex->access);
432 
433 		if (match)
434 			/* parent has an exception that matches the proposed */
435 			return true;
436 		else
437 			return false;
438 	}
439 	return false;
440 }
441 
442 /*
443  * parent_has_perm:
444  * when adding a new allow rule to a device exception list, the rule
445  * must be allowed in the parent device
446  */
447 static int parent_has_perm(struct dev_cgroup *childcg,
448 				  struct dev_exception_item *ex)
449 {
450 	struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
451 
452 	if (!parent)
453 		return 1;
454 	return verify_new_ex(parent, ex, childcg->behavior);
455 }
456 
457 /**
458  * parent_allows_removal - verify if it's ok to remove an exception
459  * @childcg: child cgroup from where the exception will be removed
460  * @ex: exception being removed
461  *
462  * When removing an exception in cgroups with default ALLOW policy, it must
463  * be checked if removing it will give the child cgroup more access than the
464  * parent.
465  *
466  * Return: true if it's ok to remove exception, false otherwise
467  */
468 static bool parent_allows_removal(struct dev_cgroup *childcg,
469 				  struct dev_exception_item *ex)
470 {
471 	struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
472 
473 	if (!parent)
474 		return true;
475 
476 	/* It's always allowed to remove access to devices */
477 	if (childcg->behavior == DEVCG_DEFAULT_DENY)
478 		return true;
479 
480 	/*
481 	 * Make sure you're not removing part or a whole exception existing in
482 	 * the parent cgroup
483 	 */
484 	return !match_exception_partial(&parent->exceptions, ex->type,
485 					ex->major, ex->minor, ex->access);
486 }
487 
488 /**
489  * may_allow_all - checks if it's possible to change the behavior to
490  *		   allow based on parent's rules.
491  * @parent: device cgroup's parent
492  * returns: != 0 in case it's allowed, 0 otherwise
493  */
494 static inline int may_allow_all(struct dev_cgroup *parent)
495 {
496 	if (!parent)
497 		return 1;
498 	return parent->behavior == DEVCG_DEFAULT_ALLOW;
499 }
500 
501 /**
502  * revalidate_active_exceptions - walks through the active exception list and
503  * 				  revalidates the exceptions based on parent's
504  * 				  behavior and exceptions. The exceptions that
505  * 				  are no longer valid will be removed.
506  * 				  Called with devcgroup_mutex held.
507  * @devcg: cgroup which exceptions will be checked
508  *
509  * This is one of the three key functions for hierarchy implementation.
510  * This function is responsible for re-evaluating all the cgroup's active
511  * exceptions due to a parent's exception change.
512  * Refer to Documentation/admin-guide/cgroup-v1/devices.rst for more details.
513  */
514 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
515 {
516 	struct dev_exception_item *ex;
517 	struct list_head *this, *tmp;
518 
519 	list_for_each_safe(this, tmp, &devcg->exceptions) {
520 		ex = container_of(this, struct dev_exception_item, list);
521 		if (!parent_has_perm(devcg, ex))
522 			dev_exception_rm(devcg, ex);
523 	}
524 }
525 
526 /**
527  * propagate_exception - propagates a new exception to the children
528  * @devcg_root: device cgroup that added a new exception
529  * @ex: new exception to be propagated
530  *
531  * returns: 0 in case of success, != 0 in case of error
532  */
533 static int propagate_exception(struct dev_cgroup *devcg_root,
534 			       struct dev_exception_item *ex)
535 {
536 	struct cgroup_subsys_state *pos;
537 	int rc = 0;
538 
539 	rcu_read_lock();
540 
541 	css_for_each_descendant_pre(pos, &devcg_root->css) {
542 		struct dev_cgroup *devcg = css_to_devcgroup(pos);
543 
544 		/*
545 		 * Because devcgroup_mutex is held, no devcg will become
546 		 * online or offline during the tree walk (see on/offline
547 		 * methods), and online ones are safe to access outside RCU
548 		 * read lock without bumping refcnt.
549 		 */
550 		if (pos == &devcg_root->css || !is_devcg_online(devcg))
551 			continue;
552 
553 		rcu_read_unlock();
554 
555 		/*
556 		 * in case both root's behavior and devcg is allow, a new
557 		 * restriction means adding to the exception list
558 		 */
559 		if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
560 		    devcg->behavior == DEVCG_DEFAULT_ALLOW) {
561 			rc = dev_exception_add(devcg, ex);
562 			if (rc)
563 				return rc;
564 		} else {
565 			/*
566 			 * in the other possible cases:
567 			 * root's behavior: allow, devcg's: deny
568 			 * root's behavior: deny, devcg's: deny
569 			 * the exception will be removed
570 			 */
571 			dev_exception_rm(devcg, ex);
572 		}
573 		revalidate_active_exceptions(devcg);
574 
575 		rcu_read_lock();
576 	}
577 
578 	rcu_read_unlock();
579 	return rc;
580 }
581 
582 /*
583  * Modify the exception list using allow/deny rules.
584  * CAP_SYS_ADMIN is needed for this.  It's at least separate from CAP_MKNOD
585  * so we can give a container CAP_MKNOD to let it create devices but not
586  * modify the exception list.
587  * It seems likely we'll want to add a CAP_CONTAINER capability to allow
588  * us to also grant CAP_SYS_ADMIN to containers without giving away the
589  * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
590  *
591  * Taking rules away is always allowed (given CAP_SYS_ADMIN).  Granting
592  * new access is only allowed if you're in the top-level cgroup, or your
593  * parent cgroup has the access you're asking for.
594  */
595 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
596 				   int filetype, char *buffer)
597 {
598 	const char *b;
599 	char temp[12];		/* 11 + 1 characters needed for a u32 */
600 	int count, rc = 0;
601 	struct dev_exception_item ex;
602 	struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
603 
604 	if (!capable(CAP_SYS_ADMIN))
605 		return -EPERM;
606 
607 	memset(&ex, 0, sizeof(ex));
608 	b = buffer;
609 
610 	switch (*b) {
611 	case 'a':
612 		switch (filetype) {
613 		case DEVCG_ALLOW:
614 			if (css_has_online_children(&devcgroup->css))
615 				return -EINVAL;
616 
617 			if (!may_allow_all(parent))
618 				return -EPERM;
619 			dev_exception_clean(devcgroup);
620 			devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
621 			if (!parent)
622 				break;
623 
624 			rc = dev_exceptions_copy(&devcgroup->exceptions,
625 						 &parent->exceptions);
626 			if (rc)
627 				return rc;
628 			break;
629 		case DEVCG_DENY:
630 			if (css_has_online_children(&devcgroup->css))
631 				return -EINVAL;
632 
633 			dev_exception_clean(devcgroup);
634 			devcgroup->behavior = DEVCG_DEFAULT_DENY;
635 			break;
636 		default:
637 			return -EINVAL;
638 		}
639 		return 0;
640 	case 'b':
641 		ex.type = DEVCG_DEV_BLOCK;
642 		break;
643 	case 'c':
644 		ex.type = DEVCG_DEV_CHAR;
645 		break;
646 	default:
647 		return -EINVAL;
648 	}
649 	b++;
650 	if (!isspace(*b))
651 		return -EINVAL;
652 	b++;
653 	if (*b == '*') {
654 		ex.major = ~0;
655 		b++;
656 	} else if (isdigit(*b)) {
657 		memset(temp, 0, sizeof(temp));
658 		for (count = 0; count < sizeof(temp) - 1; count++) {
659 			temp[count] = *b;
660 			b++;
661 			if (!isdigit(*b))
662 				break;
663 		}
664 		rc = kstrtou32(temp, 10, &ex.major);
665 		if (rc)
666 			return -EINVAL;
667 	} else {
668 		return -EINVAL;
669 	}
670 	if (*b != ':')
671 		return -EINVAL;
672 	b++;
673 
674 	/* read minor */
675 	if (*b == '*') {
676 		ex.minor = ~0;
677 		b++;
678 	} else if (isdigit(*b)) {
679 		memset(temp, 0, sizeof(temp));
680 		for (count = 0; count < sizeof(temp) - 1; count++) {
681 			temp[count] = *b;
682 			b++;
683 			if (!isdigit(*b))
684 				break;
685 		}
686 		rc = kstrtou32(temp, 10, &ex.minor);
687 		if (rc)
688 			return -EINVAL;
689 	} else {
690 		return -EINVAL;
691 	}
692 	if (!isspace(*b))
693 		return -EINVAL;
694 	for (b++, count = 0; count < 3; count++, b++) {
695 		switch (*b) {
696 		case 'r':
697 			ex.access |= DEVCG_ACC_READ;
698 			break;
699 		case 'w':
700 			ex.access |= DEVCG_ACC_WRITE;
701 			break;
702 		case 'm':
703 			ex.access |= DEVCG_ACC_MKNOD;
704 			break;
705 		case '\n':
706 		case '\0':
707 			count = 3;
708 			break;
709 		default:
710 			return -EINVAL;
711 		}
712 	}
713 
714 	switch (filetype) {
715 	case DEVCG_ALLOW:
716 		/*
717 		 * If the default policy is to allow by default, try to remove
718 		 * an matching exception instead. And be silent about it: we
719 		 * don't want to break compatibility
720 		 */
721 		if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
722 			/* Check if the parent allows removing it first */
723 			if (!parent_allows_removal(devcgroup, &ex))
724 				return -EPERM;
725 			dev_exception_rm(devcgroup, &ex);
726 			break;
727 		}
728 
729 		if (!parent_has_perm(devcgroup, &ex))
730 			return -EPERM;
731 		rc = dev_exception_add(devcgroup, &ex);
732 		break;
733 	case DEVCG_DENY:
734 		/*
735 		 * If the default policy is to deny by default, try to remove
736 		 * an matching exception instead. And be silent about it: we
737 		 * don't want to break compatibility
738 		 */
739 		if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
740 			dev_exception_rm(devcgroup, &ex);
741 		else
742 			rc = dev_exception_add(devcgroup, &ex);
743 
744 		if (rc)
745 			break;
746 		/* we only propagate new restrictions */
747 		rc = propagate_exception(devcgroup, &ex);
748 		break;
749 	default:
750 		rc = -EINVAL;
751 	}
752 	return rc;
753 }
754 
755 static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
756 				      char *buf, size_t nbytes, loff_t off)
757 {
758 	int retval;
759 
760 	mutex_lock(&devcgroup_mutex);
761 	retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
762 					 of_cft(of)->private, strstrip(buf));
763 	mutex_unlock(&devcgroup_mutex);
764 	return retval ?: nbytes;
765 }
766 
767 static struct cftype dev_cgroup_files[] = {
768 	{
769 		.name = "allow",
770 		.write = devcgroup_access_write,
771 		.private = DEVCG_ALLOW,
772 	},
773 	{
774 		.name = "deny",
775 		.write = devcgroup_access_write,
776 		.private = DEVCG_DENY,
777 	},
778 	{
779 		.name = "list",
780 		.seq_show = devcgroup_seq_show,
781 		.private = DEVCG_LIST,
782 	},
783 	{ }	/* terminate */
784 };
785 
786 struct cgroup_subsys devices_cgrp_subsys = {
787 	.css_alloc = devcgroup_css_alloc,
788 	.css_free = devcgroup_css_free,
789 	.css_online = devcgroup_online,
790 	.css_offline = devcgroup_offline,
791 	.legacy_cftypes = dev_cgroup_files,
792 };
793 
794 /**
795  * __devcgroup_check_permission - checks if an inode operation is permitted
796  * @dev_cgroup: the dev cgroup to be tested against
797  * @type: device type
798  * @major: device major number
799  * @minor: device minor number
800  * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
801  *
802  * returns 0 on success, -EPERM case the operation is not permitted
803  */
804 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
805 					short access)
806 {
807 	struct dev_cgroup *dev_cgroup;
808 	bool rc;
809 
810 	rcu_read_lock();
811 	dev_cgroup = task_devcgroup(current);
812 	if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
813 		/* Can't match any of the exceptions, even partially */
814 		rc = !match_exception_partial(&dev_cgroup->exceptions,
815 					      type, major, minor, access);
816 	else
817 		/* Need to match completely one exception to be allowed */
818 		rc = match_exception(&dev_cgroup->exceptions, type, major,
819 				     minor, access);
820 	rcu_read_unlock();
821 
822 	if (!rc)
823 		return -EPERM;
824 
825 	return 0;
826 }
827 
828 int devcgroup_check_permission(short type, u32 major, u32 minor, short access)
829 {
830 	int rc = BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type, major, minor, access);
831 
832 	if (rc)
833 		return -EPERM;
834 
835 	return __devcgroup_check_permission(type, major, minor, access);
836 }
837 EXPORT_SYMBOL(devcgroup_check_permission);
838