device_cgroup.c

  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/bpf-cgroup.h>
  9#include <linux/device_cgroup.h>
 10#include <linux/cgroup.h>
 11#include <linux/ctype.h>
 12#include <linux/list.h>
 13#include <linux/uaccess.h>
 14#include <linux/seq_file.h>
 15#include <linux/slab.h>
 16#include <linux/rcupdate.h>
 17#include <linux/mutex.h>
 18
 19#ifdef CONFIG_CGROUP_DEVICE
 20
 21static DEFINE_MUTEX(devcgroup_mutex);
 22
 23enum devcg_behavior {
 24	DEVCG_DEFAULT_NONE,
 25	DEVCG_DEFAULT_ALLOW,
 26	DEVCG_DEFAULT_DENY,
 27};
 28
 29/*
 30 * exception list locking rules:
 31 * hold devcgroup_mutex for update/read.
 32 * hold rcu_read_lock() for read.
 33 */
 34
 35struct dev_exception_item {
 36	u32 major, minor;
 37	short type;
 38	short access;
 39	struct list_head list;
 40	struct rcu_head rcu;
 41};
 42
 43struct dev_cgroup {
 44	struct cgroup_subsys_state css;
 45	struct list_head exceptions;
 46	enum devcg_behavior behavior;
 47};
 48
 49static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
 50{
 51	return s ? container_of(s, struct dev_cgroup, css) : NULL;
 52}
 53
 54static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
 55{
 56	return css_to_devcgroup(task_css(task, devices_cgrp_id));
 57}
 58
 59/*
 60 * called under devcgroup_mutex
 61 */
 62static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
 63{
 64	struct dev_exception_item *ex, *tmp, *new;
 65
 66	lockdep_assert_held(&devcgroup_mutex);
 67
 68	list_for_each_entry(ex, orig, list) {
 69		new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
 70		if (!new)
 71			goto free_and_exit;
 72		list_add_tail(&new->list, dest);
 73	}
 74
 75	return 0;
 76
 77free_and_exit:
 78	list_for_each_entry_safe(ex, tmp, dest, list) {
 79		list_del(&ex->list);
 80		kfree(ex);
 81	}
 82	return -ENOMEM;
 83}
 84
 85static void dev_exceptions_move(struct list_head *dest, struct list_head *orig)
 86{
 87	struct dev_exception_item *ex, *tmp;
 88
 89	lockdep_assert_held(&devcgroup_mutex);
 90
 91	list_for_each_entry_safe(ex, tmp, orig, list) {
 92		list_move_tail(&ex->list, dest);
 93	}
 94}
 95
 96/*
 97 * called under devcgroup_mutex
 98 */
 99static int dev_exception_add(struct dev_cgroup *dev_cgroup,
100			     struct dev_exception_item *ex)
101{
102	struct dev_exception_item *excopy, *walk;
103
104	lockdep_assert_held(&devcgroup_mutex);
105
106	excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
107	if (!excopy)
108		return -ENOMEM;
109
110	list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
111		if (walk->type != ex->type)
112			continue;
113		if (walk->major != ex->major)
114			continue;
115		if (walk->minor != ex->minor)
116			continue;
117
118		walk->access |= ex->access;
119		kfree(excopy);
120		excopy = NULL;
121	}
122
123	if (excopy != NULL)
124		list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
125	return 0;
126}
127
128/*
129 * called under devcgroup_mutex
130 */
131static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
132			     struct dev_exception_item *ex)
133{
134	struct dev_exception_item *walk, *tmp;
135
136	lockdep_assert_held(&devcgroup_mutex);
137
138	list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
139		if (walk->type != ex->type)
140			continue;
141		if (walk->major != ex->major)
142			continue;
143		if (walk->minor != ex->minor)
144			continue;
145
146		walk->access &= ~ex->access;
147		if (!walk->access) {
148			list_del_rcu(&walk->list);
149			kfree_rcu(walk, rcu);
150		}
151	}
152}
153
154static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
155{
156	struct dev_exception_item *ex, *tmp;
157
158	list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
159		list_del_rcu(&ex->list);
160		kfree_rcu(ex, rcu);
161	}
162}
163
164/**
165 * dev_exception_clean - frees all entries of the exception list
166 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
167 *
168 * called under devcgroup_mutex
169 */
170static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
171{
172	lockdep_assert_held(&devcgroup_mutex);
173
174	__dev_exception_clean(dev_cgroup);
175}
176
177static inline bool is_devcg_online(const struct dev_cgroup *devcg)
178{
179	return (devcg->behavior != DEVCG_DEFAULT_NONE);
180}
181
182/**
183 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
184 * 		      parent's
185 * @css: css getting online
186 * returns 0 in case of success, error code otherwise
187 */
188static int devcgroup_online(struct cgroup_subsys_state *css)
189{
190	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
191	struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
192	int ret = 0;
193
194	mutex_lock(&devcgroup_mutex);
195
196	if (parent_dev_cgroup == NULL)
197		dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
198	else {
199		ret = dev_exceptions_copy(&dev_cgroup->exceptions,
200					  &parent_dev_cgroup->exceptions);
201		if (!ret)
202			dev_cgroup->behavior = parent_dev_cgroup->behavior;
203	}
204	mutex_unlock(&devcgroup_mutex);
205
206	return ret;
207}
208
209static void devcgroup_offline(struct cgroup_subsys_state *css)
210{
211	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
212
213	mutex_lock(&devcgroup_mutex);
214	dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
215	mutex_unlock(&devcgroup_mutex);
216}
217
218/*
219 * called from kernel/cgroup/cgroup.c with cgroup_lock() held.
220 */
221static struct cgroup_subsys_state *
222devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
223{
224	struct dev_cgroup *dev_cgroup;
225
226	dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
227	if (!dev_cgroup)
228		return ERR_PTR(-ENOMEM);
229	INIT_LIST_HEAD(&dev_cgroup->exceptions);
230	dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
231
232	return &dev_cgroup->css;
233}
234
235static void devcgroup_css_free(struct cgroup_subsys_state *css)
236{
237	struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
238
239	__dev_exception_clean(dev_cgroup);
240	kfree(dev_cgroup);
241}
242
243#define DEVCG_ALLOW 1
244#define DEVCG_DENY 2
245#define DEVCG_LIST 3
246
247static void seq_putaccess(struct seq_file *m, short access)
248{
249	if (access & DEVCG_ACC_READ)
250		seq_putc(m, 'r');
251	if (access & DEVCG_ACC_WRITE)
252		seq_putc(m, 'w');
253	if (access & DEVCG_ACC_MKNOD)
254		seq_putc(m, 'm');
255}
256
257static void seq_puttype(struct seq_file *m, short type)
258{
259	if (type == DEVCG_DEV_ALL)
260		seq_putc(m, 'a');
261	else if (type == DEVCG_DEV_CHAR)
262		seq_putc(m, 'c');
263	else if (type == DEVCG_DEV_BLOCK)
264		seq_putc(m, 'b');
265	else
266		seq_putc(m, 'X');
267}
268
269static void seq_putversion(struct seq_file *m, unsigned int version)
270{
271	if (version == ~0)
272		seq_putc(m, '*');
273	else
274		seq_printf(m, "%u", version);
275}
276
277static int devcgroup_seq_show(struct seq_file *m, void *v)
278{
279	struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
280	struct dev_exception_item *ex;
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		seq_puts(m, "a *:* rwm\n");
291	} else {
292		list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
293			seq_puttype(m, ex->type);
294			seq_putc(m, ' ');
295			seq_putversion(m, ex->major);
296			seq_putc(m, ':');
297			seq_putversion(m, ex->minor);
298			seq_putc(m, ' ');
299			seq_putaccess(m, ex->access);
300			seq_putc(m, '\n');
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 (DEVCG_DEV_BLOCK or DEVCG_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 (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_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 */
321static 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 & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
328			continue;
329		if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_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 (DEVCG_DEV_BLOCK or DEVCG_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 (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_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 */
358static 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				lockdep_is_held(&devcgroup_mutex)) {
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 */
398static 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 accessed, 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 */
456static 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 */
477static 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 */
503static 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/admin-guide/cgroup-v1/devices.rst for more details.
522 */
523static 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 */
542static 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				return rc;
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 */
604static 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	struct dev_cgroup tmp_devcgrp;
613
614	if (!capable(CAP_SYS_ADMIN))
615		return -EPERM;
616
617	memset(&ex, 0, sizeof(ex));
618	memset(&tmp_devcgrp, 0, sizeof(tmp_devcgrp));
619	b = buffer;
620
621	switch (*b) {
622	case 'a':
623		switch (filetype) {
624		case DEVCG_ALLOW:
625			if (css_has_online_children(&devcgroup->css))
626				return -EINVAL;
627
628			if (!may_allow_all(parent))
629				return -EPERM;
630			if (!parent) {
631				devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
632				dev_exception_clean(devcgroup);
633				break;
634			}
635
636			INIT_LIST_HEAD(&tmp_devcgrp.exceptions);
637			rc = dev_exceptions_copy(&tmp_devcgrp.exceptions,
638						 &devcgroup->exceptions);
639			if (rc)
640				return rc;
641			dev_exception_clean(devcgroup);
642			rc = dev_exceptions_copy(&devcgroup->exceptions,
643						 &parent->exceptions);
644			if (rc) {
645				dev_exceptions_move(&devcgroup->exceptions,
646						    &tmp_devcgrp.exceptions);
647				return rc;
648			}
649			devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
650			dev_exception_clean(&tmp_devcgrp);
651			break;
652		case DEVCG_DENY:
653			if (css_has_online_children(&devcgroup->css))
654				return -EINVAL;
655
656			dev_exception_clean(devcgroup);
657			devcgroup->behavior = DEVCG_DEFAULT_DENY;
658			break;
659		default:
660			return -EINVAL;
661		}
662		return 0;
663	case 'b':
664		ex.type = DEVCG_DEV_BLOCK;
665		break;
666	case 'c':
667		ex.type = DEVCG_DEV_CHAR;
668		break;
669	default:
670		return -EINVAL;
671	}
672	b++;
673	if (!isspace(*b))
674		return -EINVAL;
675	b++;
676	if (*b == '*') {
677		ex.major = ~0;
678		b++;
679	} else if (isdigit(*b)) {
680		memset(temp, 0, sizeof(temp));
681		for (count = 0; count < sizeof(temp) - 1; count++) {
682			temp[count] = *b;
683			b++;
684			if (!isdigit(*b))
685				break;
686		}
687		rc = kstrtou32(temp, 10, &ex.major);
688		if (rc)
689			return -EINVAL;
690	} else {
691		return -EINVAL;
692	}
693	if (*b != ':')
694		return -EINVAL;
695	b++;
696
697	/* read minor */
698	if (*b == '*') {
699		ex.minor = ~0;
700		b++;
701	} else if (isdigit(*b)) {
702		memset(temp, 0, sizeof(temp));
703		for (count = 0; count < sizeof(temp) - 1; count++) {
704			temp[count] = *b;
705			b++;
706			if (!isdigit(*b))
707				break;
708		}
709		rc = kstrtou32(temp, 10, &ex.minor);
710		if (rc)
711			return -EINVAL;
712	} else {
713		return -EINVAL;
714	}
715	if (!isspace(*b))
716		return -EINVAL;
717	for (b++, count = 0; count < 3; count++, b++) {
718		switch (*b) {
719		case 'r':
720			ex.access |= DEVCG_ACC_READ;
721			break;
722		case 'w':
723			ex.access |= DEVCG_ACC_WRITE;
724			break;
725		case 'm':
726			ex.access |= DEVCG_ACC_MKNOD;
727			break;
728		case '\n':
729		case '\0':
730			count = 3;
731			break;
732		default:
733			return -EINVAL;
734		}
735	}
736
737	switch (filetype) {
738	case DEVCG_ALLOW:
739		/*
740		 * If the default policy is to allow by default, try to remove
741		 * an matching exception instead. And be silent about it: we
742		 * don't want to break compatibility
743		 */
744		if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
745			/* Check if the parent allows removing it first */
746			if (!parent_allows_removal(devcgroup, &ex))
747				return -EPERM;
748			dev_exception_rm(devcgroup, &ex);
749			break;
750		}
751
752		if (!parent_has_perm(devcgroup, &ex))
753			return -EPERM;
754		rc = dev_exception_add(devcgroup, &ex);
755		break;
756	case DEVCG_DENY:
757		/*
758		 * If the default policy is to deny by default, try to remove
759		 * an matching exception instead. And be silent about it: we
760		 * don't want to break compatibility
761		 */
762		if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
763			dev_exception_rm(devcgroup, &ex);
764		else
765			rc = dev_exception_add(devcgroup, &ex);
766
767		if (rc)
768			break;
769		/* we only propagate new restrictions */
770		rc = propagate_exception(devcgroup, &ex);
771		break;
772	default:
773		rc = -EINVAL;
774	}
775	return rc;
776}
777
778static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
779				      char *buf, size_t nbytes, loff_t off)
780{
781	int retval;
782
783	mutex_lock(&devcgroup_mutex);
784	retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
785					 of_cft(of)->private, strstrip(buf));
786	mutex_unlock(&devcgroup_mutex);
787	return retval ?: nbytes;
788}
789
790static struct cftype dev_cgroup_files[] = {
791	{
792		.name = "allow",
793		.write = devcgroup_access_write,
794		.private = DEVCG_ALLOW,
795	},
796	{
797		.name = "deny",
798		.write = devcgroup_access_write,
799		.private = DEVCG_DENY,
800	},
801	{
802		.name = "list",
803		.seq_show = devcgroup_seq_show,
804		.private = DEVCG_LIST,
805	},
806	{ }	/* terminate */
807};
808
809struct cgroup_subsys devices_cgrp_subsys = {
810	.css_alloc = devcgroup_css_alloc,
811	.css_free = devcgroup_css_free,
812	.css_online = devcgroup_online,
813	.css_offline = devcgroup_offline,
814	.legacy_cftypes = dev_cgroup_files,
815};
816
817/**
818 * devcgroup_legacy_check_permission - checks if an inode operation is permitted
819 * @type: device type
820 * @major: device major number
821 * @minor: device minor number
822 * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
823 *
824 * returns 0 on success, -EPERM case the operation is not permitted
825 */
826static int devcgroup_legacy_check_permission(short type, u32 major, u32 minor,
827					short access)
828{
829	struct dev_cgroup *dev_cgroup;
830	bool rc;
831
832	rcu_read_lock();
833	dev_cgroup = task_devcgroup(current);
834	if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
835		/* Can't match any of the exceptions, even partially */
836		rc = !match_exception_partial(&dev_cgroup->exceptions,
837					      type, major, minor, access);
838	else
839		/* Need to match completely one exception to be allowed */
840		rc = match_exception(&dev_cgroup->exceptions, type, major,
841				     minor, access);
842	rcu_read_unlock();
843
844	if (!rc)
845		return -EPERM;
846
847	return 0;
848}
849
850#endif /* CONFIG_CGROUP_DEVICE */
851
852#if defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF)
853
854int devcgroup_check_permission(short type, u32 major, u32 minor, short access)
855{
856	int rc = BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type, major, minor, access);
857
858	if (rc)
859		return rc;
860
861	#ifdef CONFIG_CGROUP_DEVICE
862	return devcgroup_legacy_check_permission(type, major, minor, access);
863
864	#else /* CONFIG_CGROUP_DEVICE */
865	return 0;
866
867	#endif /* CONFIG_CGROUP_DEVICE */
868}
869EXPORT_SYMBOL(devcgroup_check_permission);
870#endif /* defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF) */