1 // SPDX-License-Identifier: GPL-2.0
3 * device_cgroup.c - device cgroup subsystem
5 * Copyright 2007 IBM Corp
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>
18 #ifdef CONFIG_CGROUP_DEVICE
20 static DEFINE_MUTEX(devcgroup_mutex);
29 * exception list locking rules:
30 * hold devcgroup_mutex for update/read.
31 * hold rcu_read_lock() for read.
34 struct dev_exception_item {
38 struct list_head list;
43 struct cgroup_subsys_state css;
44 struct list_head exceptions;
45 enum devcg_behavior behavior;
48 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
50 return s ? container_of(s, struct dev_cgroup, css) : NULL;
53 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
55 return css_to_devcgroup(task_css(task, devices_cgrp_id));
59 * called under devcgroup_mutex
61 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
63 struct dev_exception_item *ex, *tmp, *new;
65 lockdep_assert_held(&devcgroup_mutex);
67 list_for_each_entry(ex, orig, list) {
68 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
71 list_add_tail(&new->list, dest);
77 list_for_each_entry_safe(ex, tmp, dest, list) {
85 * called under devcgroup_mutex
87 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
88 struct dev_exception_item *ex)
90 struct dev_exception_item *excopy, *walk;
92 lockdep_assert_held(&devcgroup_mutex);
94 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
98 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
99 if (walk->type != ex->type)
101 if (walk->major != ex->major)
103 if (walk->minor != ex->minor)
106 walk->access |= ex->access;
112 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
117 * called under devcgroup_mutex
119 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
120 struct dev_exception_item *ex)
122 struct dev_exception_item *walk, *tmp;
124 lockdep_assert_held(&devcgroup_mutex);
126 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
127 if (walk->type != ex->type)
129 if (walk->major != ex->major)
131 if (walk->minor != ex->minor)
134 walk->access &= ~ex->access;
136 list_del_rcu(&walk->list);
137 kfree_rcu(walk, rcu);
142 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
144 struct dev_exception_item *ex, *tmp;
146 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
147 list_del_rcu(&ex->list);
153 * dev_exception_clean - frees all entries of the exception list
154 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
156 * called under devcgroup_mutex
158 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
160 lockdep_assert_held(&devcgroup_mutex);
162 __dev_exception_clean(dev_cgroup);
165 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
167 return (devcg->behavior != DEVCG_DEFAULT_NONE);
171 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
173 * @css: css getting online
174 * returns 0 in case of success, error code otherwise
176 static int devcgroup_online(struct cgroup_subsys_state *css)
178 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
179 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
182 mutex_lock(&devcgroup_mutex);
184 if (parent_dev_cgroup == NULL)
185 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
187 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
188 &parent_dev_cgroup->exceptions);
190 dev_cgroup->behavior = parent_dev_cgroup->behavior;
192 mutex_unlock(&devcgroup_mutex);
197 static void devcgroup_offline(struct cgroup_subsys_state *css)
199 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
201 mutex_lock(&devcgroup_mutex);
202 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
203 mutex_unlock(&devcgroup_mutex);
207 * called from kernel/cgroup.c with cgroup_lock() held.
209 static struct cgroup_subsys_state *
210 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
212 struct dev_cgroup *dev_cgroup;
214 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
216 return ERR_PTR(-ENOMEM);
217 INIT_LIST_HEAD(&dev_cgroup->exceptions);
218 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
220 return &dev_cgroup->css;
223 static void devcgroup_css_free(struct cgroup_subsys_state *css)
225 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
227 __dev_exception_clean(dev_cgroup);
231 #define DEVCG_ALLOW 1
238 static void set_access(char *acc, short access)
241 memset(acc, 0, ACCLEN);
242 if (access & DEVCG_ACC_READ)
244 if (access & DEVCG_ACC_WRITE)
246 if (access & DEVCG_ACC_MKNOD)
250 static char type_to_char(short type)
252 if (type == DEVCG_DEV_ALL)
254 if (type == DEVCG_DEV_CHAR)
256 if (type == DEVCG_DEV_BLOCK)
261 static void set_majmin(char *str, unsigned m)
266 sprintf(str, "%u", m);
269 static int devcgroup_seq_show(struct seq_file *m, void *v)
271 struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
272 struct dev_exception_item *ex;
273 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
277 * To preserve the compatibility:
278 * - Only show the "all devices" when the default policy is to allow
279 * - List the exceptions in case the default policy is to deny
280 * This way, the file remains as a "whitelist of devices"
282 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
283 set_access(acc, DEVCG_ACC_MASK);
286 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEVCG_DEV_ALL),
289 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
290 set_access(acc, ex->access);
291 set_majmin(maj, ex->major);
292 set_majmin(min, ex->minor);
293 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
303 * match_exception - iterates the exception list trying to find a complete match
304 * @exceptions: list of exceptions
305 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
306 * @major: device file major number, ~0 to match all
307 * @minor: device file minor number, ~0 to match all
308 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
310 * It is considered a complete match if an exception is found that will
311 * contain the entire range of provided parameters.
313 * Return: true in case it matches an exception completely
315 static bool match_exception(struct list_head *exceptions, short type,
316 u32 major, u32 minor, short access)
318 struct dev_exception_item *ex;
320 list_for_each_entry_rcu(ex, exceptions, list) {
321 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
323 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
325 if (ex->major != ~0 && ex->major != major)
327 if (ex->minor != ~0 && ex->minor != minor)
329 /* provided access cannot have more than the exception rule */
330 if (access & (~ex->access))
338 * match_exception_partial - iterates the exception list trying to find a partial match
339 * @exceptions: list of exceptions
340 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
341 * @major: device file major number, ~0 to match all
342 * @minor: device file minor number, ~0 to match all
343 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
345 * It is considered a partial match if an exception's range is found to
346 * contain *any* of the devices specified by provided parameters. This is
347 * used to make sure no extra access is being granted that is forbidden by
348 * any of the exception list.
350 * Return: true in case the provided range mat matches an exception completely
352 static bool match_exception_partial(struct list_head *exceptions, short type,
353 u32 major, u32 minor, short access)
355 struct dev_exception_item *ex;
357 list_for_each_entry_rcu(ex, exceptions, list,
358 lockdep_is_held(&devcgroup_mutex)) {
359 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
361 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
364 * We must be sure that both the exception and the provided
365 * range aren't masking all devices
367 if (ex->major != ~0 && major != ~0 && ex->major != major)
369 if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
372 * In order to make sure the provided range isn't matching
373 * an exception, all its access bits shouldn't match the
374 * exception's access bits
376 if (!(access & ex->access))
384 * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
385 * @dev_cgroup: dev cgroup to be tested against
386 * @refex: new exception
387 * @behavior: behavior of the exception's dev_cgroup
389 * This is used to make sure a child cgroup won't have more privileges
392 static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
393 struct dev_exception_item *refex,
394 enum devcg_behavior behavior)
398 RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
399 !lockdep_is_held(&devcgroup_mutex),
400 "device_cgroup:verify_new_ex called without proper synchronization");
402 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
403 if (behavior == DEVCG_DEFAULT_ALLOW) {
405 * new exception in the child doesn't matter, only
406 * adding extra restrictions
411 * new exception in the child will add more devices
412 * that can be acessed, so it can't match any of
413 * parent's exceptions, even slightly
415 match = match_exception_partial(&dev_cgroup->exceptions,
427 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
428 * the new exception will add access to more devices and must
429 * be contained completely in an parent's exception to be
432 match = match_exception(&dev_cgroup->exceptions, refex->type,
433 refex->major, refex->minor,
437 /* parent has an exception that matches the proposed */
447 * when adding a new allow rule to a device exception list, the rule
448 * must be allowed in the parent device
450 static int parent_has_perm(struct dev_cgroup *childcg,
451 struct dev_exception_item *ex)
453 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
457 return verify_new_ex(parent, ex, childcg->behavior);
461 * parent_allows_removal - verify if it's ok to remove an exception
462 * @childcg: child cgroup from where the exception will be removed
463 * @ex: exception being removed
465 * When removing an exception in cgroups with default ALLOW policy, it must
466 * be checked if removing it will give the child cgroup more access than the
469 * Return: true if it's ok to remove exception, false otherwise
471 static bool parent_allows_removal(struct dev_cgroup *childcg,
472 struct dev_exception_item *ex)
474 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
479 /* It's always allowed to remove access to devices */
480 if (childcg->behavior == DEVCG_DEFAULT_DENY)
484 * Make sure you're not removing part or a whole exception existing in
487 return !match_exception_partial(&parent->exceptions, ex->type,
488 ex->major, ex->minor, ex->access);
492 * may_allow_all - checks if it's possible to change the behavior to
493 * allow based on parent's rules.
494 * @parent: device cgroup's parent
495 * returns: != 0 in case it's allowed, 0 otherwise
497 static inline int may_allow_all(struct dev_cgroup *parent)
501 return parent->behavior == DEVCG_DEFAULT_ALLOW;
505 * revalidate_active_exceptions - walks through the active exception list and
506 * revalidates the exceptions based on parent's
507 * behavior and exceptions. The exceptions that
508 * are no longer valid will be removed.
509 * Called with devcgroup_mutex held.
510 * @devcg: cgroup which exceptions will be checked
512 * This is one of the three key functions for hierarchy implementation.
513 * This function is responsible for re-evaluating all the cgroup's active
514 * exceptions due to a parent's exception change.
515 * Refer to Documentation/admin-guide/cgroup-v1/devices.rst for more details.
517 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
519 struct dev_exception_item *ex;
520 struct list_head *this, *tmp;
522 list_for_each_safe(this, tmp, &devcg->exceptions) {
523 ex = container_of(this, struct dev_exception_item, list);
524 if (!parent_has_perm(devcg, ex))
525 dev_exception_rm(devcg, ex);
530 * propagate_exception - propagates a new exception to the children
531 * @devcg_root: device cgroup that added a new exception
532 * @ex: new exception to be propagated
534 * returns: 0 in case of success, != 0 in case of error
536 static int propagate_exception(struct dev_cgroup *devcg_root,
537 struct dev_exception_item *ex)
539 struct cgroup_subsys_state *pos;
544 css_for_each_descendant_pre(pos, &devcg_root->css) {
545 struct dev_cgroup *devcg = css_to_devcgroup(pos);
548 * Because devcgroup_mutex is held, no devcg will become
549 * online or offline during the tree walk (see on/offline
550 * methods), and online ones are safe to access outside RCU
551 * read lock without bumping refcnt.
553 if (pos == &devcg_root->css || !is_devcg_online(devcg))
559 * in case both root's behavior and devcg is allow, a new
560 * restriction means adding to the exception list
562 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
563 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
564 rc = dev_exception_add(devcg, ex);
569 * in the other possible cases:
570 * root's behavior: allow, devcg's: deny
571 * root's behavior: deny, devcg's: deny
572 * the exception will be removed
574 dev_exception_rm(devcg, ex);
576 revalidate_active_exceptions(devcg);
586 * Modify the exception list using allow/deny rules.
587 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
588 * so we can give a container CAP_MKNOD to let it create devices but not
589 * modify the exception list.
590 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
591 * us to also grant CAP_SYS_ADMIN to containers without giving away the
592 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
594 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
595 * new access is only allowed if you're in the top-level cgroup, or your
596 * parent cgroup has the access you're asking for.
598 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
599 int filetype, char *buffer)
602 char temp[12]; /* 11 + 1 characters needed for a u32 */
604 struct dev_exception_item ex;
605 struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
607 if (!capable(CAP_SYS_ADMIN))
610 memset(&ex, 0, sizeof(ex));
617 if (css_has_online_children(&devcgroup->css))
620 if (!may_allow_all(parent))
622 dev_exception_clean(devcgroup);
623 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
627 rc = dev_exceptions_copy(&devcgroup->exceptions,
628 &parent->exceptions);
633 if (css_has_online_children(&devcgroup->css))
636 dev_exception_clean(devcgroup);
637 devcgroup->behavior = DEVCG_DEFAULT_DENY;
644 ex.type = DEVCG_DEV_BLOCK;
647 ex.type = DEVCG_DEV_CHAR;
659 } else if (isdigit(*b)) {
660 memset(temp, 0, sizeof(temp));
661 for (count = 0; count < sizeof(temp) - 1; count++) {
667 rc = kstrtou32(temp, 10, &ex.major);
681 } else if (isdigit(*b)) {
682 memset(temp, 0, sizeof(temp));
683 for (count = 0; count < sizeof(temp) - 1; count++) {
689 rc = kstrtou32(temp, 10, &ex.minor);
697 for (b++, count = 0; count < 3; count++, b++) {
700 ex.access |= DEVCG_ACC_READ;
703 ex.access |= DEVCG_ACC_WRITE;
706 ex.access |= DEVCG_ACC_MKNOD;
720 * If the default policy is to allow by default, try to remove
721 * an matching exception instead. And be silent about it: we
722 * don't want to break compatibility
724 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
725 /* Check if the parent allows removing it first */
726 if (!parent_allows_removal(devcgroup, &ex))
728 dev_exception_rm(devcgroup, &ex);
732 if (!parent_has_perm(devcgroup, &ex))
734 rc = dev_exception_add(devcgroup, &ex);
738 * If the default policy is to deny by default, try to remove
739 * an matching exception instead. And be silent about it: we
740 * don't want to break compatibility
742 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
743 dev_exception_rm(devcgroup, &ex);
745 rc = dev_exception_add(devcgroup, &ex);
749 /* we only propagate new restrictions */
750 rc = propagate_exception(devcgroup, &ex);
758 static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
759 char *buf, size_t nbytes, loff_t off)
763 mutex_lock(&devcgroup_mutex);
764 retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
765 of_cft(of)->private, strstrip(buf));
766 mutex_unlock(&devcgroup_mutex);
767 return retval ?: nbytes;
770 static struct cftype dev_cgroup_files[] = {
773 .write = devcgroup_access_write,
774 .private = DEVCG_ALLOW,
778 .write = devcgroup_access_write,
779 .private = DEVCG_DENY,
783 .seq_show = devcgroup_seq_show,
784 .private = DEVCG_LIST,
789 struct cgroup_subsys devices_cgrp_subsys = {
790 .css_alloc = devcgroup_css_alloc,
791 .css_free = devcgroup_css_free,
792 .css_online = devcgroup_online,
793 .css_offline = devcgroup_offline,
794 .legacy_cftypes = dev_cgroup_files,
798 * devcgroup_legacy_check_permission - checks if an inode operation is permitted
799 * @dev_cgroup: the dev cgroup to be tested against
801 * @major: device major number
802 * @minor: device minor number
803 * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
805 * returns 0 on success, -EPERM case the operation is not permitted
807 static int devcgroup_legacy_check_permission(short type, u32 major, u32 minor,
810 struct dev_cgroup *dev_cgroup;
814 dev_cgroup = task_devcgroup(current);
815 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
816 /* Can't match any of the exceptions, even partially */
817 rc = !match_exception_partial(&dev_cgroup->exceptions,
818 type, major, minor, access);
820 /* Need to match completely one exception to be allowed */
821 rc = match_exception(&dev_cgroup->exceptions, type, major,
831 #endif /* CONFIG_CGROUP_DEVICE */
833 #if defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF)
835 int devcgroup_check_permission(short type, u32 major, u32 minor, short access)
837 int rc = BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type, major, minor, access);
842 #ifdef CONFIG_CGROUP_DEVICE
843 return devcgroup_legacy_check_permission(type, major, minor, access);
845 #else /* CONFIG_CGROUP_DEVICE */
848 #endif /* CONFIG_CGROUP_DEVICE */
850 EXPORT_SYMBOL(devcgroup_check_permission);
851 #endif /* defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF) */