xref: /openbmc/linux/security/device_cgroup.c (revision 2d96b44f)
1 /*
2  * device_cgroup.c - device cgroup subsystem
3  *
4  * Copyright 2007 IBM Corp
5  */
6 
7 #include <linux/device_cgroup.h>
8 #include <linux/cgroup.h>
9 #include <linux/ctype.h>
10 #include <linux/list.h>
11 #include <linux/uaccess.h>
12 #include <linux/seq_file.h>
13 #include <linux/slab.h>
14 #include <linux/rcupdate.h>
15 #include <linux/mutex.h>
16 
17 #define ACC_MKNOD 1
18 #define ACC_READ  2
19 #define ACC_WRITE 4
20 #define ACC_MASK (ACC_MKNOD | ACC_READ | ACC_WRITE)
21 
22 #define DEV_BLOCK 1
23 #define DEV_CHAR  2
24 #define DEV_ALL   4  /* this represents all devices */
25 
26 static DEFINE_MUTEX(devcgroup_mutex);
27 
28 enum devcg_behavior {
29 	DEVCG_DEFAULT_NONE,
30 	DEVCG_DEFAULT_ALLOW,
31 	DEVCG_DEFAULT_DENY,
32 };
33 
34 /*
35  * exception list locking rules:
36  * hold devcgroup_mutex for update/read.
37  * hold rcu_read_lock() for read.
38  */
39 
40 struct dev_exception_item {
41 	u32 major, minor;
42 	short type;
43 	short access;
44 	struct list_head list;
45 	struct rcu_head rcu;
46 };
47 
48 struct dev_cgroup {
49 	struct cgroup_subsys_state css;
50 	struct list_head exceptions;
51 	enum devcg_behavior behavior;
52 };
53 
54 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
55 {
56 	return s ? container_of(s, struct dev_cgroup, css) : NULL;
57 }
58 
59 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
60 {
61 	return css_to_devcgroup(task_css(task, devices_cgrp_id));
62 }
63 
64 /*
65  * called under devcgroup_mutex
66  */
67 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
68 {
69 	struct dev_exception_item *ex, *tmp, *new;
70 
71 	lockdep_assert_held(&devcgroup_mutex);
72 
73 	list_for_each_entry(ex, orig, list) {
74 		new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
75 		if (!new)
76 			goto free_and_exit;
77 		list_add_tail(&new->list, dest);
78 	}
79 
80 	return 0;
81 
82 free_and_exit:
83 	list_for_each_entry_safe(ex, tmp, dest, list) {
84 		list_del(&ex->list);
85 		kfree(ex);
86 	}
87 	return -ENOMEM;
88 }
89 
90 /*
91  * called under devcgroup_mutex
92  */
93 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
94 			     struct dev_exception_item *ex)
95 {
96 	struct dev_exception_item *excopy, *walk;
97 
98 	lockdep_assert_held(&devcgroup_mutex);
99 
100 	excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
101 	if (!excopy)
102 		return -ENOMEM;
103 
104 	list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
105 		if (walk->type != ex->type)
106 			continue;
107 		if (walk->major != ex->major)
108 			continue;
109 		if (walk->minor != ex->minor)
110 			continue;
111 
112 		walk->access |= ex->access;
113 		kfree(excopy);
114 		excopy = NULL;
115 	}
116 
117 	if (excopy != NULL)
118 		list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
119 	return 0;
120 }
121 
122 /*
123  * called under devcgroup_mutex
124  */
125 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
126 			     struct dev_exception_item *ex)
127 {
128 	struct dev_exception_item *walk, *tmp;
129 
130 	lockdep_assert_held(&devcgroup_mutex);
131 
132 	list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
133 		if (walk->type != ex->type)
134 			continue;
135 		if (walk->major != ex->major)
136 			continue;
137 		if (walk->minor != ex->minor)
138 			continue;
139 
140 		walk->access &= ~ex->access;
141 		if (!walk->access) {
142 			list_del_rcu(&walk->list);
143 			kfree_rcu(walk, rcu);
144 		}
145 	}
146 }
147 
148 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
149 {
150 	struct dev_exception_item *ex, *tmp;
151 
152 	list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
153 		list_del_rcu(&ex->list);
154 		kfree_rcu(ex, rcu);
155 	}
156 }
157 
158 /**
159  * dev_exception_clean - frees all entries of the exception list
160  * @dev_cgroup: dev_cgroup with the exception list to be cleaned
161  *
162  * called under devcgroup_mutex
163  */
164 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
165 {
166 	lockdep_assert_held(&devcgroup_mutex);
167 
168 	__dev_exception_clean(dev_cgroup);
169 }
170 
171 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
172 {
173 	return (devcg->behavior != DEVCG_DEFAULT_NONE);
174 }
175 
176 /**
177  * devcgroup_online - initializes devcgroup's behavior and exceptions based on
178  * 		      parent's
179  * @css: css getting online
180  * returns 0 in case of success, error code otherwise
181  */
182 static int devcgroup_online(struct cgroup_subsys_state *css)
183 {
184 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
185 	struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
186 	int ret = 0;
187 
188 	mutex_lock(&devcgroup_mutex);
189 
190 	if (parent_dev_cgroup == NULL)
191 		dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
192 	else {
193 		ret = dev_exceptions_copy(&dev_cgroup->exceptions,
194 					  &parent_dev_cgroup->exceptions);
195 		if (!ret)
196 			dev_cgroup->behavior = parent_dev_cgroup->behavior;
197 	}
198 	mutex_unlock(&devcgroup_mutex);
199 
200 	return ret;
201 }
202 
203 static void devcgroup_offline(struct cgroup_subsys_state *css)
204 {
205 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
206 
207 	mutex_lock(&devcgroup_mutex);
208 	dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
209 	mutex_unlock(&devcgroup_mutex);
210 }
211 
212 /*
213  * called from kernel/cgroup.c with cgroup_lock() held.
214  */
215 static struct cgroup_subsys_state *
216 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
217 {
218 	struct dev_cgroup *dev_cgroup;
219 
220 	dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
221 	if (!dev_cgroup)
222 		return ERR_PTR(-ENOMEM);
223 	INIT_LIST_HEAD(&dev_cgroup->exceptions);
224 	dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
225 
226 	return &dev_cgroup->css;
227 }
228 
229 static void devcgroup_css_free(struct cgroup_subsys_state *css)
230 {
231 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
232 
233 	__dev_exception_clean(dev_cgroup);
234 	kfree(dev_cgroup);
235 }
236 
237 #define DEVCG_ALLOW 1
238 #define DEVCG_DENY 2
239 #define DEVCG_LIST 3
240 
241 #define MAJMINLEN 13
242 #define ACCLEN 4
243 
244 static void set_access(char *acc, short access)
245 {
246 	int idx = 0;
247 	memset(acc, 0, ACCLEN);
248 	if (access & ACC_READ)
249 		acc[idx++] = 'r';
250 	if (access & ACC_WRITE)
251 		acc[idx++] = 'w';
252 	if (access & ACC_MKNOD)
253 		acc[idx++] = 'm';
254 }
255 
256 static char type_to_char(short type)
257 {
258 	if (type == DEV_ALL)
259 		return 'a';
260 	if (type == DEV_CHAR)
261 		return 'c';
262 	if (type == DEV_BLOCK)
263 		return 'b';
264 	return 'X';
265 }
266 
267 static void set_majmin(char *str, unsigned m)
268 {
269 	if (m == ~0)
270 		strcpy(str, "*");
271 	else
272 		sprintf(str, "%u", m);
273 }
274 
275 static int devcgroup_seq_show(struct seq_file *m, void *v)
276 {
277 	struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
278 	struct dev_exception_item *ex;
279 	char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
280 
281 	rcu_read_lock();
282 	/*
283 	 * To preserve the compatibility:
284 	 * - Only show the "all devices" when the default policy is to allow
285 	 * - List the exceptions in case the default policy is to deny
286 	 * This way, the file remains as a "whitelist of devices"
287 	 */
288 	if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
289 		set_access(acc, ACC_MASK);
290 		set_majmin(maj, ~0);
291 		set_majmin(min, ~0);
292 		seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
293 			   maj, min, acc);
294 	} else {
295 		list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
296 			set_access(acc, ex->access);
297 			set_majmin(maj, ex->major);
298 			set_majmin(min, ex->minor);
299 			seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
300 				   maj, min, acc);
301 		}
302 	}
303 	rcu_read_unlock();
304 
305 	return 0;
306 }
307 
308 /**
309  * match_exception	- iterates the exception list trying to find a complete match
310  * @exceptions: list of exceptions
311  * @type: device type (DEV_BLOCK or DEV_CHAR)
312  * @major: device file major number, ~0 to match all
313  * @minor: device file minor number, ~0 to match all
314  * @access: permission mask (ACC_READ, ACC_WRITE, ACC_MKNOD)
315  *
316  * It is considered a complete match if an exception is found that will
317  * contain the entire range of provided parameters.
318  *
319  * Return: true in case it matches an exception completely
320  */
321 static bool match_exception(struct list_head *exceptions, short type,
322 			    u32 major, u32 minor, short access)
323 {
324 	struct dev_exception_item *ex;
325 
326 	list_for_each_entry_rcu(ex, exceptions, list) {
327 		if ((type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
328 			continue;
329 		if ((type & DEV_CHAR) && !(ex->type & DEV_CHAR))
330 			continue;
331 		if (ex->major != ~0 && ex->major != major)
332 			continue;
333 		if (ex->minor != ~0 && ex->minor != minor)
334 			continue;
335 		/* provided access cannot have more than the exception rule */
336 		if (access & (~ex->access))
337 			continue;
338 		return true;
339 	}
340 	return false;
341 }
342 
343 /**
344  * match_exception_partial - iterates the exception list trying to find a partial match
345  * @exceptions: list of exceptions
346  * @type: device type (DEV_BLOCK or DEV_CHAR)
347  * @major: device file major number, ~0 to match all
348  * @minor: device file minor number, ~0 to match all
349  * @access: permission mask (ACC_READ, ACC_WRITE, ACC_MKNOD)
350  *
351  * It is considered a partial match if an exception's range is found to
352  * contain *any* of the devices specified by provided parameters. This is
353  * used to make sure no extra access is being granted that is forbidden by
354  * any of the exception list.
355  *
356  * Return: true in case the provided range mat matches an exception completely
357  */
358 static bool match_exception_partial(struct list_head *exceptions, short type,
359 				    u32 major, u32 minor, short access)
360 {
361 	struct dev_exception_item *ex;
362 
363 	list_for_each_entry_rcu(ex, exceptions, list) {
364 		if ((type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
365 			continue;
366 		if ((type & DEV_CHAR) && !(ex->type & DEV_CHAR))
367 			continue;
368 		/*
369 		 * We must be sure that both the exception and the provided
370 		 * range aren't masking all devices
371 		 */
372 		if (ex->major != ~0 && major != ~0 && ex->major != major)
373 			continue;
374 		if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
375 			continue;
376 		/*
377 		 * In order to make sure the provided range isn't matching
378 		 * an exception, all its access bits shouldn't match the
379 		 * exception's access bits
380 		 */
381 		if (!(access & ex->access))
382 			continue;
383 		return true;
384 	}
385 	return false;
386 }
387 
388 /**
389  * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
390  * @dev_cgroup: dev cgroup to be tested against
391  * @refex: new exception
392  * @behavior: behavior of the exception's dev_cgroup
393  *
394  * This is used to make sure a child cgroup won't have more privileges
395  * than its parent
396  */
397 static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
398 		          struct dev_exception_item *refex,
399 		          enum devcg_behavior behavior)
400 {
401 	bool match = false;
402 
403 	RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
404 			 !lockdep_is_held(&devcgroup_mutex),
405 			 "device_cgroup:verify_new_ex called without proper synchronization");
406 
407 	if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
408 		if (behavior == DEVCG_DEFAULT_ALLOW) {
409 			/*
410 			 * new exception in the child doesn't matter, only
411 			 * adding extra restrictions
412 			 */
413 			return true;
414 		} else {
415 			/*
416 			 * new exception in the child will add more devices
417 			 * that can be acessed, so it can't match any of
418 			 * parent's exceptions, even slightly
419 			 */
420 			match = match_exception_partial(&dev_cgroup->exceptions,
421 							refex->type,
422 							refex->major,
423 							refex->minor,
424 							refex->access);
425 
426 			if (match)
427 				return false;
428 			return true;
429 		}
430 	} else {
431 		/*
432 		 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
433 		 * the new exception will add access to more devices and must
434 		 * be contained completely in an parent's exception to be
435 		 * allowed
436 		 */
437 		match = match_exception(&dev_cgroup->exceptions, refex->type,
438 					refex->major, refex->minor,
439 					refex->access);
440 
441 		if (match)
442 			/* parent has an exception that matches the proposed */
443 			return true;
444 		else
445 			return false;
446 	}
447 	return false;
448 }
449 
450 /*
451  * parent_has_perm:
452  * when adding a new allow rule to a device exception list, the rule
453  * must be allowed in the parent device
454  */
455 static int parent_has_perm(struct dev_cgroup *childcg,
456 				  struct dev_exception_item *ex)
457 {
458 	struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
459 
460 	if (!parent)
461 		return 1;
462 	return verify_new_ex(parent, ex, childcg->behavior);
463 }
464 
465 /**
466  * parent_allows_removal - verify if it's ok to remove an exception
467  * @childcg: child cgroup from where the exception will be removed
468  * @ex: exception being removed
469  *
470  * When removing an exception in cgroups with default ALLOW policy, it must
471  * be checked if removing it will give the child cgroup more access than the
472  * parent.
473  *
474  * Return: true if it's ok to remove exception, false otherwise
475  */
476 static bool parent_allows_removal(struct dev_cgroup *childcg,
477 				  struct dev_exception_item *ex)
478 {
479 	struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
480 
481 	if (!parent)
482 		return true;
483 
484 	/* It's always allowed to remove access to devices */
485 	if (childcg->behavior == DEVCG_DEFAULT_DENY)
486 		return true;
487 
488 	/*
489 	 * Make sure you're not removing part or a whole exception existing in
490 	 * the parent cgroup
491 	 */
492 	return !match_exception_partial(&parent->exceptions, ex->type,
493 					ex->major, ex->minor, ex->access);
494 }
495 
496 /**
497  * may_allow_all - checks if it's possible to change the behavior to
498  *		   allow based on parent's rules.
499  * @parent: device cgroup's parent
500  * returns: != 0 in case it's allowed, 0 otherwise
501  */
502 static inline int may_allow_all(struct dev_cgroup *parent)
503 {
504 	if (!parent)
505 		return 1;
506 	return parent->behavior == DEVCG_DEFAULT_ALLOW;
507 }
508 
509 /**
510  * revalidate_active_exceptions - walks through the active exception list and
511  * 				  revalidates the exceptions based on parent's
512  * 				  behavior and exceptions. The exceptions that
513  * 				  are no longer valid will be removed.
514  * 				  Called with devcgroup_mutex held.
515  * @devcg: cgroup which exceptions will be checked
516  *
517  * This is one of the three key functions for hierarchy implementation.
518  * This function is responsible for re-evaluating all the cgroup's active
519  * exceptions due to a parent's exception change.
520  * Refer to Documentation/cgroups/devices.txt for more details.
521  */
522 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
523 {
524 	struct dev_exception_item *ex;
525 	struct list_head *this, *tmp;
526 
527 	list_for_each_safe(this, tmp, &devcg->exceptions) {
528 		ex = container_of(this, struct dev_exception_item, list);
529 		if (!parent_has_perm(devcg, ex))
530 			dev_exception_rm(devcg, ex);
531 	}
532 }
533 
534 /**
535  * propagate_exception - propagates a new exception to the children
536  * @devcg_root: device cgroup that added a new exception
537  * @ex: new exception to be propagated
538  *
539  * returns: 0 in case of success, != 0 in case of error
540  */
541 static int propagate_exception(struct dev_cgroup *devcg_root,
542 			       struct dev_exception_item *ex)
543 {
544 	struct cgroup_subsys_state *pos;
545 	int rc = 0;
546 
547 	rcu_read_lock();
548 
549 	css_for_each_descendant_pre(pos, &devcg_root->css) {
550 		struct dev_cgroup *devcg = css_to_devcgroup(pos);
551 
552 		/*
553 		 * Because devcgroup_mutex is held, no devcg will become
554 		 * online or offline during the tree walk (see on/offline
555 		 * methods), and online ones are safe to access outside RCU
556 		 * read lock without bumping refcnt.
557 		 */
558 		if (pos == &devcg_root->css || !is_devcg_online(devcg))
559 			continue;
560 
561 		rcu_read_unlock();
562 
563 		/*
564 		 * in case both root's behavior and devcg is allow, a new
565 		 * restriction means adding to the exception list
566 		 */
567 		if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
568 		    devcg->behavior == DEVCG_DEFAULT_ALLOW) {
569 			rc = dev_exception_add(devcg, ex);
570 			if (rc)
571 				break;
572 		} else {
573 			/*
574 			 * in the other possible cases:
575 			 * root's behavior: allow, devcg's: deny
576 			 * root's behavior: deny, devcg's: deny
577 			 * the exception will be removed
578 			 */
579 			dev_exception_rm(devcg, ex);
580 		}
581 		revalidate_active_exceptions(devcg);
582 
583 		rcu_read_lock();
584 	}
585 
586 	rcu_read_unlock();
587 	return rc;
588 }
589 
590 /*
591  * Modify the exception list using allow/deny rules.
592  * CAP_SYS_ADMIN is needed for this.  It's at least separate from CAP_MKNOD
593  * so we can give a container CAP_MKNOD to let it create devices but not
594  * modify the exception list.
595  * It seems likely we'll want to add a CAP_CONTAINER capability to allow
596  * us to also grant CAP_SYS_ADMIN to containers without giving away the
597  * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
598  *
599  * Taking rules away is always allowed (given CAP_SYS_ADMIN).  Granting
600  * new access is only allowed if you're in the top-level cgroup, or your
601  * parent cgroup has the access you're asking for.
602  */
603 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
604 				   int filetype, char *buffer)
605 {
606 	const char *b;
607 	char temp[12];		/* 11 + 1 characters needed for a u32 */
608 	int count, rc = 0;
609 	struct dev_exception_item ex;
610 	struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
611 
612 	if (!capable(CAP_SYS_ADMIN))
613 		return -EPERM;
614 
615 	memset(&ex, 0, sizeof(ex));
616 	b = buffer;
617 
618 	switch (*b) {
619 	case 'a':
620 		switch (filetype) {
621 		case DEVCG_ALLOW:
622 			if (css_has_online_children(&devcgroup->css))
623 				return -EINVAL;
624 
625 			if (!may_allow_all(parent))
626 				return -EPERM;
627 			dev_exception_clean(devcgroup);
628 			devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
629 			if (!parent)
630 				break;
631 
632 			rc = dev_exceptions_copy(&devcgroup->exceptions,
633 						 &parent->exceptions);
634 			if (rc)
635 				return rc;
636 			break;
637 		case DEVCG_DENY:
638 			if (css_has_online_children(&devcgroup->css))
639 				return -EINVAL;
640 
641 			dev_exception_clean(devcgroup);
642 			devcgroup->behavior = DEVCG_DEFAULT_DENY;
643 			break;
644 		default:
645 			return -EINVAL;
646 		}
647 		return 0;
648 	case 'b':
649 		ex.type = DEV_BLOCK;
650 		break;
651 	case 'c':
652 		ex.type = DEV_CHAR;
653 		break;
654 	default:
655 		return -EINVAL;
656 	}
657 	b++;
658 	if (!isspace(*b))
659 		return -EINVAL;
660 	b++;
661 	if (*b == '*') {
662 		ex.major = ~0;
663 		b++;
664 	} else if (isdigit(*b)) {
665 		memset(temp, 0, sizeof(temp));
666 		for (count = 0; count < sizeof(temp) - 1; count++) {
667 			temp[count] = *b;
668 			b++;
669 			if (!isdigit(*b))
670 				break;
671 		}
672 		rc = kstrtou32(temp, 10, &ex.major);
673 		if (rc)
674 			return -EINVAL;
675 	} else {
676 		return -EINVAL;
677 	}
678 	if (*b != ':')
679 		return -EINVAL;
680 	b++;
681 
682 	/* read minor */
683 	if (*b == '*') {
684 		ex.minor = ~0;
685 		b++;
686 	} else if (isdigit(*b)) {
687 		memset(temp, 0, sizeof(temp));
688 		for (count = 0; count < sizeof(temp) - 1; count++) {
689 			temp[count] = *b;
690 			b++;
691 			if (!isdigit(*b))
692 				break;
693 		}
694 		rc = kstrtou32(temp, 10, &ex.minor);
695 		if (rc)
696 			return -EINVAL;
697 	} else {
698 		return -EINVAL;
699 	}
700 	if (!isspace(*b))
701 		return -EINVAL;
702 	for (b++, count = 0; count < 3; count++, b++) {
703 		switch (*b) {
704 		case 'r':
705 			ex.access |= ACC_READ;
706 			break;
707 		case 'w':
708 			ex.access |= ACC_WRITE;
709 			break;
710 		case 'm':
711 			ex.access |= ACC_MKNOD;
712 			break;
713 		case '\n':
714 		case '\0':
715 			count = 3;
716 			break;
717 		default:
718 			return -EINVAL;
719 		}
720 	}
721 
722 	switch (filetype) {
723 	case DEVCG_ALLOW:
724 		/*
725 		 * If the default policy is to allow by default, try to remove
726 		 * an matching exception instead. And be silent about it: we
727 		 * don't want to break compatibility
728 		 */
729 		if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
730 			/* Check if the parent allows removing it first */
731 			if (!parent_allows_removal(devcgroup, &ex))
732 				return -EPERM;
733 			dev_exception_rm(devcgroup, &ex);
734 			break;
735 		}
736 
737 		if (!parent_has_perm(devcgroup, &ex))
738 			return -EPERM;
739 		rc = dev_exception_add(devcgroup, &ex);
740 		break;
741 	case DEVCG_DENY:
742 		/*
743 		 * If the default policy is to deny by default, try to remove
744 		 * an matching exception instead. And be silent about it: we
745 		 * don't want to break compatibility
746 		 */
747 		if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
748 			dev_exception_rm(devcgroup, &ex);
749 		else
750 			rc = dev_exception_add(devcgroup, &ex);
751 
752 		if (rc)
753 			break;
754 		/* we only propagate new restrictions */
755 		rc = propagate_exception(devcgroup, &ex);
756 		break;
757 	default:
758 		rc = -EINVAL;
759 	}
760 	return rc;
761 }
762 
763 static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
764 				      char *buf, size_t nbytes, loff_t off)
765 {
766 	int retval;
767 
768 	mutex_lock(&devcgroup_mutex);
769 	retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
770 					 of_cft(of)->private, strstrip(buf));
771 	mutex_unlock(&devcgroup_mutex);
772 	return retval ?: nbytes;
773 }
774 
775 static struct cftype dev_cgroup_files[] = {
776 	{
777 		.name = "allow",
778 		.write = devcgroup_access_write,
779 		.private = DEVCG_ALLOW,
780 	},
781 	{
782 		.name = "deny",
783 		.write = devcgroup_access_write,
784 		.private = DEVCG_DENY,
785 	},
786 	{
787 		.name = "list",
788 		.seq_show = devcgroup_seq_show,
789 		.private = DEVCG_LIST,
790 	},
791 	{ }	/* terminate */
792 };
793 
794 struct cgroup_subsys devices_cgrp_subsys = {
795 	.css_alloc = devcgroup_css_alloc,
796 	.css_free = devcgroup_css_free,
797 	.css_online = devcgroup_online,
798 	.css_offline = devcgroup_offline,
799 	.legacy_cftypes = dev_cgroup_files,
800 };
801 
802 /**
803  * __devcgroup_check_permission - checks if an inode operation is permitted
804  * @dev_cgroup: the dev cgroup to be tested against
805  * @type: device type
806  * @major: device major number
807  * @minor: device minor number
808  * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
809  *
810  * returns 0 on success, -EPERM case the operation is not permitted
811  */
812 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
813 				        short access)
814 {
815 	struct dev_cgroup *dev_cgroup;
816 	bool rc;
817 
818 	rcu_read_lock();
819 	dev_cgroup = task_devcgroup(current);
820 	if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
821 		/* Can't match any of the exceptions, even partially */
822 		rc = !match_exception_partial(&dev_cgroup->exceptions,
823 					      type, major, minor, access);
824 	else
825 		/* Need to match completely one exception to be allowed */
826 		rc = match_exception(&dev_cgroup->exceptions, type, major,
827 				     minor, access);
828 	rcu_read_unlock();
829 
830 	if (!rc)
831 		return -EPERM;
832 
833 	return 0;
834 }
835 
836 int __devcgroup_inode_permission(struct inode *inode, int mask)
837 {
838 	short type, access = 0;
839 
840 	if (S_ISBLK(inode->i_mode))
841 		type = DEV_BLOCK;
842 	if (S_ISCHR(inode->i_mode))
843 		type = DEV_CHAR;
844 	if (mask & MAY_WRITE)
845 		access |= ACC_WRITE;
846 	if (mask & MAY_READ)
847 		access |= ACC_READ;
848 
849 	return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
850 			access);
851 }
852 
853 int devcgroup_inode_mknod(int mode, dev_t dev)
854 {
855 	short type;
856 
857 	if (!S_ISBLK(mode) && !S_ISCHR(mode))
858 		return 0;
859 
860 	if (S_ISBLK(mode))
861 		type = DEV_BLOCK;
862 	else
863 		type = DEV_CHAR;
864 
865 	return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),
866 			ACC_MKNOD);
867 
868 }
869