xref: /openbmc/linux/security/device_cgroup.c (revision 67e306fd)
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 #define DEVCG_ACC_MKNOD 1
19 #define DEVCG_ACC_READ  2
20 #define DEVCG_ACC_WRITE 4
21 #define DEVCG_ACC_MASK (DEVCG_ACC_MKNOD | DEVCG_ACC_READ | DEVCG_ACC_WRITE)
22 
23 #define DEVCG_DEV_BLOCK 1
24 #define DEVCG_DEV_CHAR  2
25 #define DEVCG_DEV_ALL   4  /* this represents all devices */
26 
27 static DEFINE_MUTEX(devcgroup_mutex);
28 
29 enum devcg_behavior {
30 	DEVCG_DEFAULT_NONE,
31 	DEVCG_DEFAULT_ALLOW,
32 	DEVCG_DEFAULT_DENY,
33 };
34 
35 /*
36  * exception list locking rules:
37  * hold devcgroup_mutex for update/read.
38  * hold rcu_read_lock() for read.
39  */
40 
41 struct dev_exception_item {
42 	u32 major, minor;
43 	short type;
44 	short access;
45 	struct list_head list;
46 	struct rcu_head rcu;
47 };
48 
49 struct dev_cgroup {
50 	struct cgroup_subsys_state css;
51 	struct list_head exceptions;
52 	enum devcg_behavior behavior;
53 };
54 
55 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
56 {
57 	return s ? container_of(s, struct dev_cgroup, css) : NULL;
58 }
59 
60 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
61 {
62 	return css_to_devcgroup(task_css(task, devices_cgrp_id));
63 }
64 
65 /*
66  * called under devcgroup_mutex
67  */
68 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
69 {
70 	struct dev_exception_item *ex, *tmp, *new;
71 
72 	lockdep_assert_held(&devcgroup_mutex);
73 
74 	list_for_each_entry(ex, orig, list) {
75 		new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
76 		if (!new)
77 			goto free_and_exit;
78 		list_add_tail(&new->list, dest);
79 	}
80 
81 	return 0;
82 
83 free_and_exit:
84 	list_for_each_entry_safe(ex, tmp, dest, list) {
85 		list_del(&ex->list);
86 		kfree(ex);
87 	}
88 	return -ENOMEM;
89 }
90 
91 /*
92  * called under devcgroup_mutex
93  */
94 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
95 			     struct dev_exception_item *ex)
96 {
97 	struct dev_exception_item *excopy, *walk;
98 
99 	lockdep_assert_held(&devcgroup_mutex);
100 
101 	excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
102 	if (!excopy)
103 		return -ENOMEM;
104 
105 	list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
106 		if (walk->type != ex->type)
107 			continue;
108 		if (walk->major != ex->major)
109 			continue;
110 		if (walk->minor != ex->minor)
111 			continue;
112 
113 		walk->access |= ex->access;
114 		kfree(excopy);
115 		excopy = NULL;
116 	}
117 
118 	if (excopy != NULL)
119 		list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
120 	return 0;
121 }
122 
123 /*
124  * called under devcgroup_mutex
125  */
126 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
127 			     struct dev_exception_item *ex)
128 {
129 	struct dev_exception_item *walk, *tmp;
130 
131 	lockdep_assert_held(&devcgroup_mutex);
132 
133 	list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
134 		if (walk->type != ex->type)
135 			continue;
136 		if (walk->major != ex->major)
137 			continue;
138 		if (walk->minor != ex->minor)
139 			continue;
140 
141 		walk->access &= ~ex->access;
142 		if (!walk->access) {
143 			list_del_rcu(&walk->list);
144 			kfree_rcu(walk, rcu);
145 		}
146 	}
147 }
148 
149 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
150 {
151 	struct dev_exception_item *ex, *tmp;
152 
153 	list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
154 		list_del_rcu(&ex->list);
155 		kfree_rcu(ex, rcu);
156 	}
157 }
158 
159 /**
160  * dev_exception_clean - frees all entries of the exception list
161  * @dev_cgroup: dev_cgroup with the exception list to be cleaned
162  *
163  * called under devcgroup_mutex
164  */
165 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
166 {
167 	lockdep_assert_held(&devcgroup_mutex);
168 
169 	__dev_exception_clean(dev_cgroup);
170 }
171 
172 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
173 {
174 	return (devcg->behavior != DEVCG_DEFAULT_NONE);
175 }
176 
177 /**
178  * devcgroup_online - initializes devcgroup's behavior and exceptions based on
179  * 		      parent's
180  * @css: css getting online
181  * returns 0 in case of success, error code otherwise
182  */
183 static int devcgroup_online(struct cgroup_subsys_state *css)
184 {
185 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
186 	struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
187 	int ret = 0;
188 
189 	mutex_lock(&devcgroup_mutex);
190 
191 	if (parent_dev_cgroup == NULL)
192 		dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
193 	else {
194 		ret = dev_exceptions_copy(&dev_cgroup->exceptions,
195 					  &parent_dev_cgroup->exceptions);
196 		if (!ret)
197 			dev_cgroup->behavior = parent_dev_cgroup->behavior;
198 	}
199 	mutex_unlock(&devcgroup_mutex);
200 
201 	return ret;
202 }
203 
204 static void devcgroup_offline(struct cgroup_subsys_state *css)
205 {
206 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
207 
208 	mutex_lock(&devcgroup_mutex);
209 	dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
210 	mutex_unlock(&devcgroup_mutex);
211 }
212 
213 /*
214  * called from kernel/cgroup.c with cgroup_lock() held.
215  */
216 static struct cgroup_subsys_state *
217 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
218 {
219 	struct dev_cgroup *dev_cgroup;
220 
221 	dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
222 	if (!dev_cgroup)
223 		return ERR_PTR(-ENOMEM);
224 	INIT_LIST_HEAD(&dev_cgroup->exceptions);
225 	dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
226 
227 	return &dev_cgroup->css;
228 }
229 
230 static void devcgroup_css_free(struct cgroup_subsys_state *css)
231 {
232 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
233 
234 	__dev_exception_clean(dev_cgroup);
235 	kfree(dev_cgroup);
236 }
237 
238 #define DEVCG_ALLOW 1
239 #define DEVCG_DENY 2
240 #define DEVCG_LIST 3
241 
242 #define MAJMINLEN 13
243 #define ACCLEN 4
244 
245 static void set_access(char *acc, short access)
246 {
247 	int idx = 0;
248 	memset(acc, 0, ACCLEN);
249 	if (access & DEVCG_ACC_READ)
250 		acc[idx++] = 'r';
251 	if (access & DEVCG_ACC_WRITE)
252 		acc[idx++] = 'w';
253 	if (access & DEVCG_ACC_MKNOD)
254 		acc[idx++] = 'm';
255 }
256 
257 static char type_to_char(short type)
258 {
259 	if (type == DEVCG_DEV_ALL)
260 		return 'a';
261 	if (type == DEVCG_DEV_CHAR)
262 		return 'c';
263 	if (type == DEVCG_DEV_BLOCK)
264 		return 'b';
265 	return 'X';
266 }
267 
268 static void set_majmin(char *str, unsigned m)
269 {
270 	if (m == ~0)
271 		strcpy(str, "*");
272 	else
273 		sprintf(str, "%u", m);
274 }
275 
276 static int devcgroup_seq_show(struct seq_file *m, void *v)
277 {
278 	struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
279 	struct dev_exception_item *ex;
280 	char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
281 
282 	rcu_read_lock();
283 	/*
284 	 * To preserve the compatibility:
285 	 * - Only show the "all devices" when the default policy is to allow
286 	 * - List the exceptions in case the default policy is to deny
287 	 * This way, the file remains as a "whitelist of devices"
288 	 */
289 	if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
290 		set_access(acc, DEVCG_ACC_MASK);
291 		set_majmin(maj, ~0);
292 		set_majmin(min, ~0);
293 		seq_printf(m, "%c %s:%s %s\n", type_to_char(DEVCG_DEV_ALL),
294 			   maj, min, acc);
295 	} else {
296 		list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
297 			set_access(acc, ex->access);
298 			set_majmin(maj, ex->major);
299 			set_majmin(min, ex->minor);
300 			seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
301 				   maj, min, acc);
302 		}
303 	}
304 	rcu_read_unlock();
305 
306 	return 0;
307 }
308 
309 /**
310  * match_exception	- iterates the exception list trying to find a complete match
311  * @exceptions: list of exceptions
312  * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
313  * @major: device file major number, ~0 to match all
314  * @minor: device file minor number, ~0 to match all
315  * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
316  *
317  * It is considered a complete match if an exception is found that will
318  * contain the entire range of provided parameters.
319  *
320  * Return: true in case it matches an exception completely
321  */
322 static bool match_exception(struct list_head *exceptions, short type,
323 			    u32 major, u32 minor, short access)
324 {
325 	struct dev_exception_item *ex;
326 
327 	list_for_each_entry_rcu(ex, exceptions, list) {
328 		if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
329 			continue;
330 		if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
331 			continue;
332 		if (ex->major != ~0 && ex->major != major)
333 			continue;
334 		if (ex->minor != ~0 && ex->minor != minor)
335 			continue;
336 		/* provided access cannot have more than the exception rule */
337 		if (access & (~ex->access))
338 			continue;
339 		return true;
340 	}
341 	return false;
342 }
343 
344 /**
345  * match_exception_partial - iterates the exception list trying to find a partial match
346  * @exceptions: list of exceptions
347  * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
348  * @major: device file major number, ~0 to match all
349  * @minor: device file minor number, ~0 to match all
350  * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
351  *
352  * It is considered a partial match if an exception's range is found to
353  * contain *any* of the devices specified by provided parameters. This is
354  * used to make sure no extra access is being granted that is forbidden by
355  * any of the exception list.
356  *
357  * Return: true in case the provided range mat matches an exception completely
358  */
359 static bool match_exception_partial(struct list_head *exceptions, short type,
360 				    u32 major, u32 minor, short access)
361 {
362 	struct dev_exception_item *ex;
363 
364 	list_for_each_entry_rcu(ex, exceptions, list) {
365 		if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
366 			continue;
367 		if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
368 			continue;
369 		/*
370 		 * We must be sure that both the exception and the provided
371 		 * range aren't masking all devices
372 		 */
373 		if (ex->major != ~0 && major != ~0 && ex->major != major)
374 			continue;
375 		if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
376 			continue;
377 		/*
378 		 * In order to make sure the provided range isn't matching
379 		 * an exception, all its access bits shouldn't match the
380 		 * exception's access bits
381 		 */
382 		if (!(access & ex->access))
383 			continue;
384 		return true;
385 	}
386 	return false;
387 }
388 
389 /**
390  * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
391  * @dev_cgroup: dev cgroup to be tested against
392  * @refex: new exception
393  * @behavior: behavior of the exception's dev_cgroup
394  *
395  * This is used to make sure a child cgroup won't have more privileges
396  * than its parent
397  */
398 static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
399 		          struct dev_exception_item *refex,
400 		          enum devcg_behavior behavior)
401 {
402 	bool match = false;
403 
404 	RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
405 			 !lockdep_is_held(&devcgroup_mutex),
406 			 "device_cgroup:verify_new_ex called without proper synchronization");
407 
408 	if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
409 		if (behavior == DEVCG_DEFAULT_ALLOW) {
410 			/*
411 			 * new exception in the child doesn't matter, only
412 			 * adding extra restrictions
413 			 */
414 			return true;
415 		} else {
416 			/*
417 			 * new exception in the child will add more devices
418 			 * that can be acessed, so it can't match any of
419 			 * parent's exceptions, even slightly
420 			 */
421 			match = match_exception_partial(&dev_cgroup->exceptions,
422 							refex->type,
423 							refex->major,
424 							refex->minor,
425 							refex->access);
426 
427 			if (match)
428 				return false;
429 			return true;
430 		}
431 	} else {
432 		/*
433 		 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
434 		 * the new exception will add access to more devices and must
435 		 * be contained completely in an parent's exception to be
436 		 * allowed
437 		 */
438 		match = match_exception(&dev_cgroup->exceptions, refex->type,
439 					refex->major, refex->minor,
440 					refex->access);
441 
442 		if (match)
443 			/* parent has an exception that matches the proposed */
444 			return true;
445 		else
446 			return false;
447 	}
448 	return false;
449 }
450 
451 /*
452  * parent_has_perm:
453  * when adding a new allow rule to a device exception list, the rule
454  * must be allowed in the parent device
455  */
456 static int parent_has_perm(struct dev_cgroup *childcg,
457 				  struct dev_exception_item *ex)
458 {
459 	struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
460 
461 	if (!parent)
462 		return 1;
463 	return verify_new_ex(parent, ex, childcg->behavior);
464 }
465 
466 /**
467  * parent_allows_removal - verify if it's ok to remove an exception
468  * @childcg: child cgroup from where the exception will be removed
469  * @ex: exception being removed
470  *
471  * When removing an exception in cgroups with default ALLOW policy, it must
472  * be checked if removing it will give the child cgroup more access than the
473  * parent.
474  *
475  * Return: true if it's ok to remove exception, false otherwise
476  */
477 static bool parent_allows_removal(struct dev_cgroup *childcg,
478 				  struct dev_exception_item *ex)
479 {
480 	struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
481 
482 	if (!parent)
483 		return true;
484 
485 	/* It's always allowed to remove access to devices */
486 	if (childcg->behavior == DEVCG_DEFAULT_DENY)
487 		return true;
488 
489 	/*
490 	 * Make sure you're not removing part or a whole exception existing in
491 	 * the parent cgroup
492 	 */
493 	return !match_exception_partial(&parent->exceptions, ex->type,
494 					ex->major, ex->minor, ex->access);
495 }
496 
497 /**
498  * may_allow_all - checks if it's possible to change the behavior to
499  *		   allow based on parent's rules.
500  * @parent: device cgroup's parent
501  * returns: != 0 in case it's allowed, 0 otherwise
502  */
503 static inline int may_allow_all(struct dev_cgroup *parent)
504 {
505 	if (!parent)
506 		return 1;
507 	return parent->behavior == DEVCG_DEFAULT_ALLOW;
508 }
509 
510 /**
511  * revalidate_active_exceptions - walks through the active exception list and
512  * 				  revalidates the exceptions based on parent's
513  * 				  behavior and exceptions. The exceptions that
514  * 				  are no longer valid will be removed.
515  * 				  Called with devcgroup_mutex held.
516  * @devcg: cgroup which exceptions will be checked
517  *
518  * This is one of the three key functions for hierarchy implementation.
519  * This function is responsible for re-evaluating all the cgroup's active
520  * exceptions due to a parent's exception change.
521  * Refer to Documentation/cgroups/devices.txt for more details.
522  */
523 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
524 {
525 	struct dev_exception_item *ex;
526 	struct list_head *this, *tmp;
527 
528 	list_for_each_safe(this, tmp, &devcg->exceptions) {
529 		ex = container_of(this, struct dev_exception_item, list);
530 		if (!parent_has_perm(devcg, ex))
531 			dev_exception_rm(devcg, ex);
532 	}
533 }
534 
535 /**
536  * propagate_exception - propagates a new exception to the children
537  * @devcg_root: device cgroup that added a new exception
538  * @ex: new exception to be propagated
539  *
540  * returns: 0 in case of success, != 0 in case of error
541  */
542 static int propagate_exception(struct dev_cgroup *devcg_root,
543 			       struct dev_exception_item *ex)
544 {
545 	struct cgroup_subsys_state *pos;
546 	int rc = 0;
547 
548 	rcu_read_lock();
549 
550 	css_for_each_descendant_pre(pos, &devcg_root->css) {
551 		struct dev_cgroup *devcg = css_to_devcgroup(pos);
552 
553 		/*
554 		 * Because devcgroup_mutex is held, no devcg will become
555 		 * online or offline during the tree walk (see on/offline
556 		 * methods), and online ones are safe to access outside RCU
557 		 * read lock without bumping refcnt.
558 		 */
559 		if (pos == &devcg_root->css || !is_devcg_online(devcg))
560 			continue;
561 
562 		rcu_read_unlock();
563 
564 		/*
565 		 * in case both root's behavior and devcg is allow, a new
566 		 * restriction means adding to the exception list
567 		 */
568 		if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
569 		    devcg->behavior == DEVCG_DEFAULT_ALLOW) {
570 			rc = dev_exception_add(devcg, ex);
571 			if (rc)
572 				break;
573 		} else {
574 			/*
575 			 * in the other possible cases:
576 			 * root's behavior: allow, devcg's: deny
577 			 * root's behavior: deny, devcg's: deny
578 			 * the exception will be removed
579 			 */
580 			dev_exception_rm(devcg, ex);
581 		}
582 		revalidate_active_exceptions(devcg);
583 
584 		rcu_read_lock();
585 	}
586 
587 	rcu_read_unlock();
588 	return rc;
589 }
590 
591 /*
592  * Modify the exception list using allow/deny rules.
593  * CAP_SYS_ADMIN is needed for this.  It's at least separate from CAP_MKNOD
594  * so we can give a container CAP_MKNOD to let it create devices but not
595  * modify the exception list.
596  * It seems likely we'll want to add a CAP_CONTAINER capability to allow
597  * us to also grant CAP_SYS_ADMIN to containers without giving away the
598  * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
599  *
600  * Taking rules away is always allowed (given CAP_SYS_ADMIN).  Granting
601  * new access is only allowed if you're in the top-level cgroup, or your
602  * parent cgroup has the access you're asking for.
603  */
604 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
605 				   int filetype, char *buffer)
606 {
607 	const char *b;
608 	char temp[12];		/* 11 + 1 characters needed for a u32 */
609 	int count, rc = 0;
610 	struct dev_exception_item ex;
611 	struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
612 
613 	if (!capable(CAP_SYS_ADMIN))
614 		return -EPERM;
615 
616 	memset(&ex, 0, sizeof(ex));
617 	b = buffer;
618 
619 	switch (*b) {
620 	case 'a':
621 		switch (filetype) {
622 		case DEVCG_ALLOW:
623 			if (css_has_online_children(&devcgroup->css))
624 				return -EINVAL;
625 
626 			if (!may_allow_all(parent))
627 				return -EPERM;
628 			dev_exception_clean(devcgroup);
629 			devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
630 			if (!parent)
631 				break;
632 
633 			rc = dev_exceptions_copy(&devcgroup->exceptions,
634 						 &parent->exceptions);
635 			if (rc)
636 				return rc;
637 			break;
638 		case DEVCG_DENY:
639 			if (css_has_online_children(&devcgroup->css))
640 				return -EINVAL;
641 
642 			dev_exception_clean(devcgroup);
643 			devcgroup->behavior = DEVCG_DEFAULT_DENY;
644 			break;
645 		default:
646 			return -EINVAL;
647 		}
648 		return 0;
649 	case 'b':
650 		ex.type = DEVCG_DEV_BLOCK;
651 		break;
652 	case 'c':
653 		ex.type = DEVCG_DEV_CHAR;
654 		break;
655 	default:
656 		return -EINVAL;
657 	}
658 	b++;
659 	if (!isspace(*b))
660 		return -EINVAL;
661 	b++;
662 	if (*b == '*') {
663 		ex.major = ~0;
664 		b++;
665 	} else if (isdigit(*b)) {
666 		memset(temp, 0, sizeof(temp));
667 		for (count = 0; count < sizeof(temp) - 1; count++) {
668 			temp[count] = *b;
669 			b++;
670 			if (!isdigit(*b))
671 				break;
672 		}
673 		rc = kstrtou32(temp, 10, &ex.major);
674 		if (rc)
675 			return -EINVAL;
676 	} else {
677 		return -EINVAL;
678 	}
679 	if (*b != ':')
680 		return -EINVAL;
681 	b++;
682 
683 	/* read minor */
684 	if (*b == '*') {
685 		ex.minor = ~0;
686 		b++;
687 	} else if (isdigit(*b)) {
688 		memset(temp, 0, sizeof(temp));
689 		for (count = 0; count < sizeof(temp) - 1; count++) {
690 			temp[count] = *b;
691 			b++;
692 			if (!isdigit(*b))
693 				break;
694 		}
695 		rc = kstrtou32(temp, 10, &ex.minor);
696 		if (rc)
697 			return -EINVAL;
698 	} else {
699 		return -EINVAL;
700 	}
701 	if (!isspace(*b))
702 		return -EINVAL;
703 	for (b++, count = 0; count < 3; count++, b++) {
704 		switch (*b) {
705 		case 'r':
706 			ex.access |= DEVCG_ACC_READ;
707 			break;
708 		case 'w':
709 			ex.access |= DEVCG_ACC_WRITE;
710 			break;
711 		case 'm':
712 			ex.access |= DEVCG_ACC_MKNOD;
713 			break;
714 		case '\n':
715 		case '\0':
716 			count = 3;
717 			break;
718 		default:
719 			return -EINVAL;
720 		}
721 	}
722 
723 	switch (filetype) {
724 	case DEVCG_ALLOW:
725 		/*
726 		 * If the default policy is to allow by default, try to remove
727 		 * an matching exception instead. And be silent about it: we
728 		 * don't want to break compatibility
729 		 */
730 		if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
731 			/* Check if the parent allows removing it first */
732 			if (!parent_allows_removal(devcgroup, &ex))
733 				return -EPERM;
734 			dev_exception_rm(devcgroup, &ex);
735 			break;
736 		}
737 
738 		if (!parent_has_perm(devcgroup, &ex))
739 			return -EPERM;
740 		rc = dev_exception_add(devcgroup, &ex);
741 		break;
742 	case DEVCG_DENY:
743 		/*
744 		 * If the default policy is to deny by default, try to remove
745 		 * an matching exception instead. And be silent about it: we
746 		 * don't want to break compatibility
747 		 */
748 		if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
749 			dev_exception_rm(devcgroup, &ex);
750 		else
751 			rc = dev_exception_add(devcgroup, &ex);
752 
753 		if (rc)
754 			break;
755 		/* we only propagate new restrictions */
756 		rc = propagate_exception(devcgroup, &ex);
757 		break;
758 	default:
759 		rc = -EINVAL;
760 	}
761 	return rc;
762 }
763 
764 static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
765 				      char *buf, size_t nbytes, loff_t off)
766 {
767 	int retval;
768 
769 	mutex_lock(&devcgroup_mutex);
770 	retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
771 					 of_cft(of)->private, strstrip(buf));
772 	mutex_unlock(&devcgroup_mutex);
773 	return retval ?: nbytes;
774 }
775 
776 static struct cftype dev_cgroup_files[] = {
777 	{
778 		.name = "allow",
779 		.write = devcgroup_access_write,
780 		.private = DEVCG_ALLOW,
781 	},
782 	{
783 		.name = "deny",
784 		.write = devcgroup_access_write,
785 		.private = DEVCG_DENY,
786 	},
787 	{
788 		.name = "list",
789 		.seq_show = devcgroup_seq_show,
790 		.private = DEVCG_LIST,
791 	},
792 	{ }	/* terminate */
793 };
794 
795 struct cgroup_subsys devices_cgrp_subsys = {
796 	.css_alloc = devcgroup_css_alloc,
797 	.css_free = devcgroup_css_free,
798 	.css_online = devcgroup_online,
799 	.css_offline = devcgroup_offline,
800 	.legacy_cftypes = dev_cgroup_files,
801 };
802 
803 /**
804  * __devcgroup_check_permission - checks if an inode operation is permitted
805  * @dev_cgroup: the dev cgroup to be tested against
806  * @type: device type
807  * @major: device major number
808  * @minor: device minor number
809  * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
810  *
811  * returns 0 on success, -EPERM case the operation is not permitted
812  */
813 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
814 				        short access)
815 {
816 	struct dev_cgroup *dev_cgroup;
817 	bool rc;
818 
819 	rcu_read_lock();
820 	dev_cgroup = task_devcgroup(current);
821 	if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
822 		/* Can't match any of the exceptions, even partially */
823 		rc = !match_exception_partial(&dev_cgroup->exceptions,
824 					      type, major, minor, access);
825 	else
826 		/* Need to match completely one exception to be allowed */
827 		rc = match_exception(&dev_cgroup->exceptions, type, major,
828 				     minor, access);
829 	rcu_read_unlock();
830 
831 	if (!rc)
832 		return -EPERM;
833 
834 	return 0;
835 }
836 
837 int __devcgroup_inode_permission(struct inode *inode, int mask)
838 {
839 	short type, access = 0;
840 
841 	if (S_ISBLK(inode->i_mode))
842 		type = DEVCG_DEV_BLOCK;
843 	if (S_ISCHR(inode->i_mode))
844 		type = DEVCG_DEV_CHAR;
845 	if (mask & MAY_WRITE)
846 		access |= DEVCG_ACC_WRITE;
847 	if (mask & MAY_READ)
848 		access |= DEVCG_ACC_READ;
849 
850 	return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
851 			access);
852 }
853 
854 int devcgroup_inode_mknod(int mode, dev_t dev)
855 {
856 	short type;
857 
858 	if (!S_ISBLK(mode) && !S_ISCHR(mode))
859 		return 0;
860 
861 	if (S_ISBLK(mode))
862 		type = DEVCG_DEV_BLOCK;
863 	else
864 		type = DEVCG_DEV_CHAR;
865 
866 	return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),
867 			DEVCG_ACC_MKNOD);
868 
869 }
870