• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
css_to_devcgroup(struct cgroup_subsys_state * s)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 
task_devcgroup(struct task_struct * task)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  */
dev_exceptions_copy(struct list_head * dest,struct list_head * orig)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 
dev_exceptions_move(struct list_head * dest,struct list_head * orig)82 static void dev_exceptions_move(struct list_head *dest, struct list_head *orig)
83 {
84 	struct dev_exception_item *ex, *tmp;
85 
86 	lockdep_assert_held(&devcgroup_mutex);
87 
88 	list_for_each_entry_safe(ex, tmp, orig, list) {
89 		list_move_tail(&ex->list, dest);
90 	}
91 }
92 
93 /*
94  * called under devcgroup_mutex
95  */
dev_exception_add(struct dev_cgroup * dev_cgroup,struct dev_exception_item * ex)96 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
97 			     struct dev_exception_item *ex)
98 {
99 	struct dev_exception_item *excopy, *walk;
100 
101 	lockdep_assert_held(&devcgroup_mutex);
102 
103 	excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
104 	if (!excopy)
105 		return -ENOMEM;
106 
107 	list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
108 		if (walk->type != ex->type)
109 			continue;
110 		if (walk->major != ex->major)
111 			continue;
112 		if (walk->minor != ex->minor)
113 			continue;
114 
115 		walk->access |= ex->access;
116 		kfree(excopy);
117 		excopy = NULL;
118 	}
119 
120 	if (excopy != NULL)
121 		list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
122 	return 0;
123 }
124 
125 /*
126  * called under devcgroup_mutex
127  */
dev_exception_rm(struct dev_cgroup * dev_cgroup,struct dev_exception_item * ex)128 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
129 			     struct dev_exception_item *ex)
130 {
131 	struct dev_exception_item *walk, *tmp;
132 
133 	lockdep_assert_held(&devcgroup_mutex);
134 
135 	list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
136 		if (walk->type != ex->type)
137 			continue;
138 		if (walk->major != ex->major)
139 			continue;
140 		if (walk->minor != ex->minor)
141 			continue;
142 
143 		walk->access &= ~ex->access;
144 		if (!walk->access) {
145 			list_del_rcu(&walk->list);
146 			kfree_rcu(walk, rcu);
147 		}
148 	}
149 }
150 
__dev_exception_clean(struct dev_cgroup * dev_cgroup)151 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
152 {
153 	struct dev_exception_item *ex, *tmp;
154 
155 	list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
156 		list_del_rcu(&ex->list);
157 		kfree_rcu(ex, rcu);
158 	}
159 }
160 
161 /**
162  * dev_exception_clean - frees all entries of the exception list
163  * @dev_cgroup: dev_cgroup with the exception list to be cleaned
164  *
165  * called under devcgroup_mutex
166  */
dev_exception_clean(struct dev_cgroup * dev_cgroup)167 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
168 {
169 	lockdep_assert_held(&devcgroup_mutex);
170 
171 	__dev_exception_clean(dev_cgroup);
172 }
173 
is_devcg_online(const struct dev_cgroup * devcg)174 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
175 {
176 	return (devcg->behavior != DEVCG_DEFAULT_NONE);
177 }
178 
179 /**
180  * devcgroup_online - initializes devcgroup's behavior and exceptions based on
181  * 		      parent's
182  * @css: css getting online
183  * returns 0 in case of success, error code otherwise
184  */
devcgroup_online(struct cgroup_subsys_state * css)185 static int devcgroup_online(struct cgroup_subsys_state *css)
186 {
187 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
188 	struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
189 	int ret = 0;
190 
191 	mutex_lock(&devcgroup_mutex);
192 
193 	if (parent_dev_cgroup == NULL)
194 		dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
195 	else {
196 		ret = dev_exceptions_copy(&dev_cgroup->exceptions,
197 					  &parent_dev_cgroup->exceptions);
198 		if (!ret)
199 			dev_cgroup->behavior = parent_dev_cgroup->behavior;
200 	}
201 	mutex_unlock(&devcgroup_mutex);
202 
203 	return ret;
204 }
205 
devcgroup_offline(struct cgroup_subsys_state * css)206 static void devcgroup_offline(struct cgroup_subsys_state *css)
207 {
208 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
209 
210 	mutex_lock(&devcgroup_mutex);
211 	dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
212 	mutex_unlock(&devcgroup_mutex);
213 }
214 
215 /*
216  * called from kernel/cgroup.c with cgroup_lock() held.
217  */
218 static struct cgroup_subsys_state *
devcgroup_css_alloc(struct cgroup_subsys_state * parent_css)219 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
220 {
221 	struct dev_cgroup *dev_cgroup;
222 
223 	dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
224 	if (!dev_cgroup)
225 		return ERR_PTR(-ENOMEM);
226 	INIT_LIST_HEAD(&dev_cgroup->exceptions);
227 	dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
228 
229 	return &dev_cgroup->css;
230 }
231 
devcgroup_css_free(struct cgroup_subsys_state * css)232 static void devcgroup_css_free(struct cgroup_subsys_state *css)
233 {
234 	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
235 
236 	__dev_exception_clean(dev_cgroup);
237 	kfree(dev_cgroup);
238 }
239 
240 #define DEVCG_ALLOW 1
241 #define DEVCG_DENY 2
242 #define DEVCG_LIST 3
243 
244 #define MAJMINLEN 13
245 #define ACCLEN 4
246 
set_access(char * acc,short access)247 static void set_access(char *acc, short access)
248 {
249 	int idx = 0;
250 	memset(acc, 0, ACCLEN);
251 	if (access & DEVCG_ACC_READ)
252 		acc[idx++] = 'r';
253 	if (access & DEVCG_ACC_WRITE)
254 		acc[idx++] = 'w';
255 	if (access & DEVCG_ACC_MKNOD)
256 		acc[idx++] = 'm';
257 }
258 
type_to_char(short type)259 static char type_to_char(short type)
260 {
261 	if (type == DEVCG_DEV_ALL)
262 		return 'a';
263 	if (type == DEVCG_DEV_CHAR)
264 		return 'c';
265 	if (type == DEVCG_DEV_BLOCK)
266 		return 'b';
267 	return 'X';
268 }
269 
set_majmin(char * str,unsigned m)270 static void set_majmin(char *str, unsigned m)
271 {
272 	if (m == ~0)
273 		strcpy(str, "*");
274 	else
275 		sprintf(str, "%u", m);
276 }
277 
devcgroup_seq_show(struct seq_file * m,void * v)278 static int devcgroup_seq_show(struct seq_file *m, void *v)
279 {
280 	struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
281 	struct dev_exception_item *ex;
282 	char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
283 
284 	rcu_read_lock();
285 	/*
286 	 * To preserve the compatibility:
287 	 * - Only show the "all devices" when the default policy is to allow
288 	 * - List the exceptions in case the default policy is to deny
289 	 * This way, the file remains as a "whitelist of devices"
290 	 */
291 	if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
292 		set_access(acc, DEVCG_ACC_MASK);
293 		set_majmin(maj, ~0);
294 		set_majmin(min, ~0);
295 		seq_printf(m, "%c %s:%s %s\n", type_to_char(DEVCG_DEV_ALL),
296 			   maj, min, acc);
297 	} else {
298 		list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
299 			set_access(acc, ex->access);
300 			set_majmin(maj, ex->major);
301 			set_majmin(min, ex->minor);
302 			seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
303 				   maj, min, acc);
304 		}
305 	}
306 	rcu_read_unlock();
307 
308 	return 0;
309 }
310 
311 /**
312  * match_exception	- iterates the exception list trying to find a complete match
313  * @exceptions: list of exceptions
314  * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
315  * @major: device file major number, ~0 to match all
316  * @minor: device file minor number, ~0 to match all
317  * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
318  *
319  * It is considered a complete match if an exception is found that will
320  * contain the entire range of provided parameters.
321  *
322  * Return: true in case it matches an exception completely
323  */
match_exception(struct list_head * exceptions,short type,u32 major,u32 minor,short access)324 static bool match_exception(struct list_head *exceptions, short type,
325 			    u32 major, u32 minor, short access)
326 {
327 	struct dev_exception_item *ex;
328 
329 	list_for_each_entry_rcu(ex, exceptions, list) {
330 		if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
331 			continue;
332 		if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
333 			continue;
334 		if (ex->major != ~0 && ex->major != major)
335 			continue;
336 		if (ex->minor != ~0 && ex->minor != minor)
337 			continue;
338 		/* provided access cannot have more than the exception rule */
339 		if (access & (~ex->access))
340 			continue;
341 		return true;
342 	}
343 	return false;
344 }
345 
346 /**
347  * match_exception_partial - iterates the exception list trying to find a partial match
348  * @exceptions: list of exceptions
349  * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
350  * @major: device file major number, ~0 to match all
351  * @minor: device file minor number, ~0 to match all
352  * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
353  *
354  * It is considered a partial match if an exception's range is found to
355  * contain *any* of the devices specified by provided parameters. This is
356  * used to make sure no extra access is being granted that is forbidden by
357  * any of the exception list.
358  *
359  * Return: true in case the provided range mat matches an exception completely
360  */
match_exception_partial(struct list_head * exceptions,short type,u32 major,u32 minor,short access)361 static bool match_exception_partial(struct list_head *exceptions, short type,
362 				    u32 major, u32 minor, short access)
363 {
364 	struct dev_exception_item *ex;
365 
366 	list_for_each_entry_rcu(ex, exceptions, list,
367 				lockdep_is_held(&devcgroup_mutex)) {
368 		if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
369 			continue;
370 		if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
371 			continue;
372 		/*
373 		 * We must be sure that both the exception and the provided
374 		 * range aren't masking all devices
375 		 */
376 		if (ex->major != ~0 && major != ~0 && ex->major != major)
377 			continue;
378 		if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
379 			continue;
380 		/*
381 		 * In order to make sure the provided range isn't matching
382 		 * an exception, all its access bits shouldn't match the
383 		 * exception's access bits
384 		 */
385 		if (!(access & ex->access))
386 			continue;
387 		return true;
388 	}
389 	return false;
390 }
391 
392 /**
393  * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
394  * @dev_cgroup: dev cgroup to be tested against
395  * @refex: new exception
396  * @behavior: behavior of the exception's dev_cgroup
397  *
398  * This is used to make sure a child cgroup won't have more privileges
399  * than its parent
400  */
verify_new_ex(struct dev_cgroup * dev_cgroup,struct dev_exception_item * refex,enum devcg_behavior behavior)401 static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
402 		          struct dev_exception_item *refex,
403 		          enum devcg_behavior behavior)
404 {
405 	bool match = false;
406 
407 	RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
408 			 !lockdep_is_held(&devcgroup_mutex),
409 			 "device_cgroup:verify_new_ex called without proper synchronization");
410 
411 	if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
412 		if (behavior == DEVCG_DEFAULT_ALLOW) {
413 			/*
414 			 * new exception in the child doesn't matter, only
415 			 * adding extra restrictions
416 			 */
417 			return true;
418 		} else {
419 			/*
420 			 * new exception in the child will add more devices
421 			 * that can be acessed, so it can't match any of
422 			 * parent's exceptions, even slightly
423 			 */
424 			match = match_exception_partial(&dev_cgroup->exceptions,
425 							refex->type,
426 							refex->major,
427 							refex->minor,
428 							refex->access);
429 
430 			if (match)
431 				return false;
432 			return true;
433 		}
434 	} else {
435 		/*
436 		 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
437 		 * the new exception will add access to more devices and must
438 		 * be contained completely in an parent's exception to be
439 		 * allowed
440 		 */
441 		match = match_exception(&dev_cgroup->exceptions, refex->type,
442 					refex->major, refex->minor,
443 					refex->access);
444 
445 		if (match)
446 			/* parent has an exception that matches the proposed */
447 			return true;
448 		else
449 			return false;
450 	}
451 	return false;
452 }
453 
454 /*
455  * parent_has_perm:
456  * when adding a new allow rule to a device exception list, the rule
457  * must be allowed in the parent device
458  */
parent_has_perm(struct dev_cgroup * childcg,struct dev_exception_item * ex)459 static int parent_has_perm(struct dev_cgroup *childcg,
460 				  struct dev_exception_item *ex)
461 {
462 	struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
463 
464 	if (!parent)
465 		return 1;
466 	return verify_new_ex(parent, ex, childcg->behavior);
467 }
468 
469 /**
470  * parent_allows_removal - verify if it's ok to remove an exception
471  * @childcg: child cgroup from where the exception will be removed
472  * @ex: exception being removed
473  *
474  * When removing an exception in cgroups with default ALLOW policy, it must
475  * be checked if removing it will give the child cgroup more access than the
476  * parent.
477  *
478  * Return: true if it's ok to remove exception, false otherwise
479  */
parent_allows_removal(struct dev_cgroup * childcg,struct dev_exception_item * ex)480 static bool parent_allows_removal(struct dev_cgroup *childcg,
481 				  struct dev_exception_item *ex)
482 {
483 	struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
484 
485 	if (!parent)
486 		return true;
487 
488 	/* It's always allowed to remove access to devices */
489 	if (childcg->behavior == DEVCG_DEFAULT_DENY)
490 		return true;
491 
492 	/*
493 	 * Make sure you're not removing part or a whole exception existing in
494 	 * the parent cgroup
495 	 */
496 	return !match_exception_partial(&parent->exceptions, ex->type,
497 					ex->major, ex->minor, ex->access);
498 }
499 
500 /**
501  * may_allow_all - checks if it's possible to change the behavior to
502  *		   allow based on parent's rules.
503  * @parent: device cgroup's parent
504  * returns: != 0 in case it's allowed, 0 otherwise
505  */
may_allow_all(struct dev_cgroup * parent)506 static inline int may_allow_all(struct dev_cgroup *parent)
507 {
508 	if (!parent)
509 		return 1;
510 	return parent->behavior == DEVCG_DEFAULT_ALLOW;
511 }
512 
513 /**
514  * revalidate_active_exceptions - walks through the active exception list and
515  * 				  revalidates the exceptions based on parent's
516  * 				  behavior and exceptions. The exceptions that
517  * 				  are no longer valid will be removed.
518  * 				  Called with devcgroup_mutex held.
519  * @devcg: cgroup which exceptions will be checked
520  *
521  * This is one of the three key functions for hierarchy implementation.
522  * This function is responsible for re-evaluating all the cgroup's active
523  * exceptions due to a parent's exception change.
524  * Refer to Documentation/admin-guide/cgroup-v1/devices.rst for more details.
525  */
revalidate_active_exceptions(struct dev_cgroup * devcg)526 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
527 {
528 	struct dev_exception_item *ex;
529 	struct list_head *this, *tmp;
530 
531 	list_for_each_safe(this, tmp, &devcg->exceptions) {
532 		ex = container_of(this, struct dev_exception_item, list);
533 		if (!parent_has_perm(devcg, ex))
534 			dev_exception_rm(devcg, ex);
535 	}
536 }
537 
538 /**
539  * propagate_exception - propagates a new exception to the children
540  * @devcg_root: device cgroup that added a new exception
541  * @ex: new exception to be propagated
542  *
543  * returns: 0 in case of success, != 0 in case of error
544  */
propagate_exception(struct dev_cgroup * devcg_root,struct dev_exception_item * ex)545 static int propagate_exception(struct dev_cgroup *devcg_root,
546 			       struct dev_exception_item *ex)
547 {
548 	struct cgroup_subsys_state *pos;
549 	int rc = 0;
550 
551 	rcu_read_lock();
552 
553 	css_for_each_descendant_pre(pos, &devcg_root->css) {
554 		struct dev_cgroup *devcg = css_to_devcgroup(pos);
555 
556 		/*
557 		 * Because devcgroup_mutex is held, no devcg will become
558 		 * online or offline during the tree walk (see on/offline
559 		 * methods), and online ones are safe to access outside RCU
560 		 * read lock without bumping refcnt.
561 		 */
562 		if (pos == &devcg_root->css || !is_devcg_online(devcg))
563 			continue;
564 
565 		rcu_read_unlock();
566 
567 		/*
568 		 * in case both root's behavior and devcg is allow, a new
569 		 * restriction means adding to the exception list
570 		 */
571 		if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
572 		    devcg->behavior == DEVCG_DEFAULT_ALLOW) {
573 			rc = dev_exception_add(devcg, ex);
574 			if (rc)
575 				return rc;
576 		} else {
577 			/*
578 			 * in the other possible cases:
579 			 * root's behavior: allow, devcg's: deny
580 			 * root's behavior: deny, devcg's: deny
581 			 * the exception will be removed
582 			 */
583 			dev_exception_rm(devcg, ex);
584 		}
585 		revalidate_active_exceptions(devcg);
586 
587 		rcu_read_lock();
588 	}
589 
590 	rcu_read_unlock();
591 	return rc;
592 }
593 
594 /*
595  * Modify the exception list using allow/deny rules.
596  * CAP_SYS_ADMIN is needed for this.  It's at least separate from CAP_MKNOD
597  * so we can give a container CAP_MKNOD to let it create devices but not
598  * modify the exception list.
599  * It seems likely we'll want to add a CAP_CONTAINER capability to allow
600  * us to also grant CAP_SYS_ADMIN to containers without giving away the
601  * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
602  *
603  * Taking rules away is always allowed (given CAP_SYS_ADMIN).  Granting
604  * new access is only allowed if you're in the top-level cgroup, or your
605  * parent cgroup has the access you're asking for.
606  */
devcgroup_update_access(struct dev_cgroup * devcgroup,int filetype,char * buffer)607 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
608 				   int filetype, char *buffer)
609 {
610 	const char *b;
611 	char temp[12];		/* 11 + 1 characters needed for a u32 */
612 	int count, rc = 0;
613 	struct dev_exception_item ex;
614 	struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
615 	struct dev_cgroup tmp_devcgrp;
616 
617 	if (!capable(CAP_SYS_ADMIN))
618 		return -EPERM;
619 
620 	memset(&ex, 0, sizeof(ex));
621 	memset(&tmp_devcgrp, 0, sizeof(tmp_devcgrp));
622 	b = buffer;
623 
624 	switch (*b) {
625 	case 'a':
626 		switch (filetype) {
627 		case DEVCG_ALLOW:
628 			if (css_has_online_children(&devcgroup->css))
629 				return -EINVAL;
630 
631 			if (!may_allow_all(parent))
632 				return -EPERM;
633 			if (!parent) {
634 				devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
635 				dev_exception_clean(devcgroup);
636 				break;
637 			}
638 
639 			INIT_LIST_HEAD(&tmp_devcgrp.exceptions);
640 			rc = dev_exceptions_copy(&tmp_devcgrp.exceptions,
641 						 &devcgroup->exceptions);
642 			if (rc)
643 				return rc;
644 			dev_exception_clean(devcgroup);
645 			rc = dev_exceptions_copy(&devcgroup->exceptions,
646 						 &parent->exceptions);
647 			if (rc) {
648 				dev_exceptions_move(&devcgroup->exceptions,
649 						    &tmp_devcgrp.exceptions);
650 				return rc;
651 			}
652 			devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
653 			dev_exception_clean(&tmp_devcgrp);
654 			break;
655 		case DEVCG_DENY:
656 			if (css_has_online_children(&devcgroup->css))
657 				return -EINVAL;
658 
659 			dev_exception_clean(devcgroup);
660 			devcgroup->behavior = DEVCG_DEFAULT_DENY;
661 			break;
662 		default:
663 			return -EINVAL;
664 		}
665 		return 0;
666 	case 'b':
667 		ex.type = DEVCG_DEV_BLOCK;
668 		break;
669 	case 'c':
670 		ex.type = DEVCG_DEV_CHAR;
671 		break;
672 	default:
673 		return -EINVAL;
674 	}
675 	b++;
676 	if (!isspace(*b))
677 		return -EINVAL;
678 	b++;
679 	if (*b == '*') {
680 		ex.major = ~0;
681 		b++;
682 	} else if (isdigit(*b)) {
683 		memset(temp, 0, sizeof(temp));
684 		for (count = 0; count < sizeof(temp) - 1; count++) {
685 			temp[count] = *b;
686 			b++;
687 			if (!isdigit(*b))
688 				break;
689 		}
690 		rc = kstrtou32(temp, 10, &ex.major);
691 		if (rc)
692 			return -EINVAL;
693 	} else {
694 		return -EINVAL;
695 	}
696 	if (*b != ':')
697 		return -EINVAL;
698 	b++;
699 
700 	/* read minor */
701 	if (*b == '*') {
702 		ex.minor = ~0;
703 		b++;
704 	} else if (isdigit(*b)) {
705 		memset(temp, 0, sizeof(temp));
706 		for (count = 0; count < sizeof(temp) - 1; count++) {
707 			temp[count] = *b;
708 			b++;
709 			if (!isdigit(*b))
710 				break;
711 		}
712 		rc = kstrtou32(temp, 10, &ex.minor);
713 		if (rc)
714 			return -EINVAL;
715 	} else {
716 		return -EINVAL;
717 	}
718 	if (!isspace(*b))
719 		return -EINVAL;
720 	for (b++, count = 0; count < 3; count++, b++) {
721 		switch (*b) {
722 		case 'r':
723 			ex.access |= DEVCG_ACC_READ;
724 			break;
725 		case 'w':
726 			ex.access |= DEVCG_ACC_WRITE;
727 			break;
728 		case 'm':
729 			ex.access |= DEVCG_ACC_MKNOD;
730 			break;
731 		case '\n':
732 		case '\0':
733 			count = 3;
734 			break;
735 		default:
736 			return -EINVAL;
737 		}
738 	}
739 
740 	switch (filetype) {
741 	case DEVCG_ALLOW:
742 		/*
743 		 * If the default policy is to allow 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_ALLOW) {
748 			/* Check if the parent allows removing it first */
749 			if (!parent_allows_removal(devcgroup, &ex))
750 				return -EPERM;
751 			dev_exception_rm(devcgroup, &ex);
752 			break;
753 		}
754 
755 		if (!parent_has_perm(devcgroup, &ex))
756 			return -EPERM;
757 		rc = dev_exception_add(devcgroup, &ex);
758 		break;
759 	case DEVCG_DENY:
760 		/*
761 		 * If the default policy is to deny by default, try to remove
762 		 * an matching exception instead. And be silent about it: we
763 		 * don't want to break compatibility
764 		 */
765 		if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
766 			dev_exception_rm(devcgroup, &ex);
767 		else
768 			rc = dev_exception_add(devcgroup, &ex);
769 
770 		if (rc)
771 			break;
772 		/* we only propagate new restrictions */
773 		rc = propagate_exception(devcgroup, &ex);
774 		break;
775 	default:
776 		rc = -EINVAL;
777 	}
778 	return rc;
779 }
780 
devcgroup_access_write(struct kernfs_open_file * of,char * buf,size_t nbytes,loff_t off)781 static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
782 				      char *buf, size_t nbytes, loff_t off)
783 {
784 	int retval;
785 
786 	mutex_lock(&devcgroup_mutex);
787 	retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
788 					 of_cft(of)->private, strstrip(buf));
789 	mutex_unlock(&devcgroup_mutex);
790 	return retval ?: nbytes;
791 }
792 
793 static struct cftype dev_cgroup_files[] = {
794 	{
795 		.name = "allow",
796 		.write = devcgroup_access_write,
797 		.private = DEVCG_ALLOW,
798 	},
799 	{
800 		.name = "deny",
801 		.write = devcgroup_access_write,
802 		.private = DEVCG_DENY,
803 	},
804 	{
805 		.name = "list",
806 		.seq_show = devcgroup_seq_show,
807 		.private = DEVCG_LIST,
808 	},
809 	{ }	/* terminate */
810 };
811 
812 struct cgroup_subsys devices_cgrp_subsys = {
813 	.css_alloc = devcgroup_css_alloc,
814 	.css_free = devcgroup_css_free,
815 	.css_online = devcgroup_online,
816 	.css_offline = devcgroup_offline,
817 	.legacy_cftypes = dev_cgroup_files,
818 };
819 
820 /**
821  * __devcgroup_check_permission - checks if an inode operation is permitted
822  * @dev_cgroup: the dev cgroup to be tested against
823  * @type: device type
824  * @major: device major number
825  * @minor: device minor number
826  * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
827  *
828  * returns 0 on success, -EPERM case the operation is not permitted
829  */
__devcgroup_check_permission(short type,u32 major,u32 minor,short access)830 int __devcgroup_check_permission(short type, u32 major, u32 minor,
831 				 short access)
832 {
833 	struct dev_cgroup *dev_cgroup;
834 	bool rc;
835 
836 	rcu_read_lock();
837 	dev_cgroup = task_devcgroup(current);
838 	if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
839 		/* Can't match any of the exceptions, even partially */
840 		rc = !match_exception_partial(&dev_cgroup->exceptions,
841 					      type, major, minor, access);
842 	else
843 		/* Need to match completely one exception to be allowed */
844 		rc = match_exception(&dev_cgroup->exceptions, type, major,
845 				     minor, access);
846 	rcu_read_unlock();
847 
848 	if (!rc)
849 		return -EPERM;
850 
851 	return 0;
852 }
853