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) {
84 static void dev_exceptions_move(struct list_head *dest, struct list_head *orig)
86 struct dev_exception_item *ex, *tmp;
88 lockdep_assert_held(&devcgroup_mutex);
90 list_for_each_entry_safe(ex, tmp, orig, list) {
91 list_move_tail(&ex->list, dest);
96 * called under devcgroup_mutex
98 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
99 struct dev_exception_item *ex)
101 struct dev_exception_item *excopy, *walk;
103 lockdep_assert_held(&devcgroup_mutex);
105 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
109 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
110 if (walk->type != ex->type)
112 if (walk->major != ex->major)
114 if (walk->minor != ex->minor)
117 walk->access |= ex->access;
123 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
128 * called under devcgroup_mutex
130 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
131 struct dev_exception_item *ex)
133 struct dev_exception_item *walk, *tmp;
135 lockdep_assert_held(&devcgroup_mutex);
137 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
138 if (walk->type != ex->type)
140 if (walk->major != ex->major)
142 if (walk->minor != ex->minor)
145 walk->access &= ~ex->access;
147 list_del_rcu(&walk->list);
148 kfree_rcu(walk, rcu);
153 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
155 struct dev_exception_item *ex, *tmp;
157 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
158 list_del_rcu(&ex->list);
164 * dev_exception_clean - frees all entries of the exception list
165 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
167 * called under devcgroup_mutex
169 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
171 lockdep_assert_held(&devcgroup_mutex);
173 __dev_exception_clean(dev_cgroup);
176 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
178 return (devcg->behavior != DEVCG_DEFAULT_NONE);
182 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
184 * @css: css getting online
185 * returns 0 in case of success, error code otherwise
187 static int devcgroup_online(struct cgroup_subsys_state *css)
189 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
190 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
193 mutex_lock(&devcgroup_mutex);
195 if (parent_dev_cgroup == NULL)
196 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
198 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
199 &parent_dev_cgroup->exceptions);
201 dev_cgroup->behavior = parent_dev_cgroup->behavior;
203 mutex_unlock(&devcgroup_mutex);
208 static void devcgroup_offline(struct cgroup_subsys_state *css)
210 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
212 mutex_lock(&devcgroup_mutex);
213 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
214 mutex_unlock(&devcgroup_mutex);
218 * called from kernel/cgroup.c with cgroup_lock() held.
220 static struct cgroup_subsys_state *
221 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
223 struct dev_cgroup *dev_cgroup;
225 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
227 return ERR_PTR(-ENOMEM);
228 INIT_LIST_HEAD(&dev_cgroup->exceptions);
229 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
231 return &dev_cgroup->css;
234 static void devcgroup_css_free(struct cgroup_subsys_state *css)
236 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
238 __dev_exception_clean(dev_cgroup);
242 #define DEVCG_ALLOW 1
249 static void set_access(char *acc, short access)
252 memset(acc, 0, ACCLEN);
253 if (access & DEVCG_ACC_READ)
255 if (access & DEVCG_ACC_WRITE)
257 if (access & DEVCG_ACC_MKNOD)
261 static char type_to_char(short type)
263 if (type == DEVCG_DEV_ALL)
265 if (type == DEVCG_DEV_CHAR)
267 if (type == DEVCG_DEV_BLOCK)
272 static void set_majmin(char *str, unsigned m)
277 sprintf(str, "%u", m);
280 static int devcgroup_seq_show(struct seq_file *m, void *v)
282 struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
283 struct dev_exception_item *ex;
284 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
288 * To preserve the compatibility:
289 * - Only show the "all devices" when the default policy is to allow
290 * - List the exceptions in case the default policy is to deny
291 * This way, the file remains as a "whitelist of devices"
293 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
294 set_access(acc, DEVCG_ACC_MASK);
297 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEVCG_DEV_ALL),
300 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
301 set_access(acc, ex->access);
302 set_majmin(maj, ex->major);
303 set_majmin(min, ex->minor);
304 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
314 * match_exception - iterates the exception list trying to find a complete match
315 * @exceptions: list of exceptions
316 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
317 * @major: device file major number, ~0 to match all
318 * @minor: device file minor number, ~0 to match all
319 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
321 * It is considered a complete match if an exception is found that will
322 * contain the entire range of provided parameters.
324 * Return: true in case it matches an exception completely
326 static bool match_exception(struct list_head *exceptions, short type,
327 u32 major, u32 minor, short access)
329 struct dev_exception_item *ex;
331 list_for_each_entry_rcu(ex, exceptions, list) {
332 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
334 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
336 if (ex->major != ~0 && ex->major != major)
338 if (ex->minor != ~0 && ex->minor != minor)
340 /* provided access cannot have more than the exception rule */
341 if (access & (~ex->access))
349 * match_exception_partial - iterates the exception list trying to find a partial match
350 * @exceptions: list of exceptions
351 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
352 * @major: device file major number, ~0 to match all
353 * @minor: device file minor number, ~0 to match all
354 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
356 * It is considered a partial match if an exception's range is found to
357 * contain *any* of the devices specified by provided parameters. This is
358 * used to make sure no extra access is being granted that is forbidden by
359 * any of the exception list.
361 * Return: true in case the provided range mat matches an exception completely
363 static bool match_exception_partial(struct list_head *exceptions, short type,
364 u32 major, u32 minor, short access)
366 struct dev_exception_item *ex;
368 list_for_each_entry_rcu(ex, exceptions, list,
369 lockdep_is_held(&devcgroup_mutex)) {
370 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
372 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
375 * We must be sure that both the exception and the provided
376 * range aren't masking all devices
378 if (ex->major != ~0 && major != ~0 && ex->major != major)
380 if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
383 * In order to make sure the provided range isn't matching
384 * an exception, all its access bits shouldn't match the
385 * exception's access bits
387 if (!(access & ex->access))
395 * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
396 * @dev_cgroup: dev cgroup to be tested against
397 * @refex: new exception
398 * @behavior: behavior of the exception's dev_cgroup
400 * This is used to make sure a child cgroup won't have more privileges
403 static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
404 struct dev_exception_item *refex,
405 enum devcg_behavior behavior)
409 RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
410 !lockdep_is_held(&devcgroup_mutex),
411 "device_cgroup:verify_new_ex called without proper synchronization");
413 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
414 if (behavior == DEVCG_DEFAULT_ALLOW) {
416 * new exception in the child doesn't matter, only
417 * adding extra restrictions
422 * new exception in the child will add more devices
423 * that can be acessed, so it can't match any of
424 * parent's exceptions, even slightly
426 match = match_exception_partial(&dev_cgroup->exceptions,
438 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
439 * the new exception will add access to more devices and must
440 * be contained completely in an parent's exception to be
443 match = match_exception(&dev_cgroup->exceptions, refex->type,
444 refex->major, refex->minor,
448 /* parent has an exception that matches the proposed */
458 * when adding a new allow rule to a device exception list, the rule
459 * must be allowed in the parent device
461 static int parent_has_perm(struct dev_cgroup *childcg,
462 struct dev_exception_item *ex)
464 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
468 return verify_new_ex(parent, ex, childcg->behavior);
472 * parent_allows_removal - verify if it's ok to remove an exception
473 * @childcg: child cgroup from where the exception will be removed
474 * @ex: exception being removed
476 * When removing an exception in cgroups with default ALLOW policy, it must
477 * be checked if removing it will give the child cgroup more access than the
480 * Return: true if it's ok to remove exception, false otherwise
482 static bool parent_allows_removal(struct dev_cgroup *childcg,
483 struct dev_exception_item *ex)
485 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
490 /* It's always allowed to remove access to devices */
491 if (childcg->behavior == DEVCG_DEFAULT_DENY)
495 * Make sure you're not removing part or a whole exception existing in
498 return !match_exception_partial(&parent->exceptions, ex->type,
499 ex->major, ex->minor, ex->access);
503 * may_allow_all - checks if it's possible to change the behavior to
504 * allow based on parent's rules.
505 * @parent: device cgroup's parent
506 * returns: != 0 in case it's allowed, 0 otherwise
508 static inline int may_allow_all(struct dev_cgroup *parent)
512 return parent->behavior == DEVCG_DEFAULT_ALLOW;
516 * revalidate_active_exceptions - walks through the active exception list and
517 * revalidates the exceptions based on parent's
518 * behavior and exceptions. The exceptions that
519 * are no longer valid will be removed.
520 * Called with devcgroup_mutex held.
521 * @devcg: cgroup which exceptions will be checked
523 * This is one of the three key functions for hierarchy implementation.
524 * This function is responsible for re-evaluating all the cgroup's active
525 * exceptions due to a parent's exception change.
526 * Refer to Documentation/admin-guide/cgroup-v1/devices.rst for more details.
528 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
530 struct dev_exception_item *ex;
531 struct list_head *this, *tmp;
533 list_for_each_safe(this, tmp, &devcg->exceptions) {
534 ex = container_of(this, struct dev_exception_item, list);
535 if (!parent_has_perm(devcg, ex))
536 dev_exception_rm(devcg, ex);
541 * propagate_exception - propagates a new exception to the children
542 * @devcg_root: device cgroup that added a new exception
543 * @ex: new exception to be propagated
545 * returns: 0 in case of success, != 0 in case of error
547 static int propagate_exception(struct dev_cgroup *devcg_root,
548 struct dev_exception_item *ex)
550 struct cgroup_subsys_state *pos;
555 css_for_each_descendant_pre(pos, &devcg_root->css) {
556 struct dev_cgroup *devcg = css_to_devcgroup(pos);
559 * Because devcgroup_mutex is held, no devcg will become
560 * online or offline during the tree walk (see on/offline
561 * methods), and online ones are safe to access outside RCU
562 * read lock without bumping refcnt.
564 if (pos == &devcg_root->css || !is_devcg_online(devcg))
570 * in case both root's behavior and devcg is allow, a new
571 * restriction means adding to the exception list
573 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
574 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
575 rc = dev_exception_add(devcg, ex);
580 * in the other possible cases:
581 * root's behavior: allow, devcg's: deny
582 * root's behavior: deny, devcg's: deny
583 * the exception will be removed
585 dev_exception_rm(devcg, ex);
587 revalidate_active_exceptions(devcg);
597 * Modify the exception list using allow/deny rules.
598 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
599 * so we can give a container CAP_MKNOD to let it create devices but not
600 * modify the exception list.
601 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
602 * us to also grant CAP_SYS_ADMIN to containers without giving away the
603 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
605 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
606 * new access is only allowed if you're in the top-level cgroup, or your
607 * parent cgroup has the access you're asking for.
609 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
610 int filetype, char *buffer)
613 char temp[12]; /* 11 + 1 characters needed for a u32 */
615 struct dev_exception_item ex;
616 struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
617 struct dev_cgroup tmp_devcgrp;
619 if (!capable(CAP_SYS_ADMIN))
622 memset(&ex, 0, sizeof(ex));
623 memset(&tmp_devcgrp, 0, sizeof(tmp_devcgrp));
630 if (css_has_online_children(&devcgroup->css))
633 if (!may_allow_all(parent))
636 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
637 dev_exception_clean(devcgroup);
641 INIT_LIST_HEAD(&tmp_devcgrp.exceptions);
642 rc = dev_exceptions_copy(&tmp_devcgrp.exceptions,
643 &devcgroup->exceptions);
646 dev_exception_clean(devcgroup);
647 rc = dev_exceptions_copy(&devcgroup->exceptions,
648 &parent->exceptions);
650 dev_exceptions_move(&devcgroup->exceptions,
651 &tmp_devcgrp.exceptions);
654 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
655 dev_exception_clean(&tmp_devcgrp);
658 if (css_has_online_children(&devcgroup->css))
661 dev_exception_clean(devcgroup);
662 devcgroup->behavior = DEVCG_DEFAULT_DENY;
669 ex.type = DEVCG_DEV_BLOCK;
672 ex.type = DEVCG_DEV_CHAR;
684 } else if (isdigit(*b)) {
685 memset(temp, 0, sizeof(temp));
686 for (count = 0; count < sizeof(temp) - 1; count++) {
692 rc = kstrtou32(temp, 10, &ex.major);
706 } else if (isdigit(*b)) {
707 memset(temp, 0, sizeof(temp));
708 for (count = 0; count < sizeof(temp) - 1; count++) {
714 rc = kstrtou32(temp, 10, &ex.minor);
722 for (b++, count = 0; count < 3; count++, b++) {
725 ex.access |= DEVCG_ACC_READ;
728 ex.access |= DEVCG_ACC_WRITE;
731 ex.access |= DEVCG_ACC_MKNOD;
745 * If the default policy is to allow by default, try to remove
746 * an matching exception instead. And be silent about it: we
747 * don't want to break compatibility
749 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
750 /* Check if the parent allows removing it first */
751 if (!parent_allows_removal(devcgroup, &ex))
753 dev_exception_rm(devcgroup, &ex);
757 if (!parent_has_perm(devcgroup, &ex))
759 rc = dev_exception_add(devcgroup, &ex);
763 * If the default policy is to deny by default, try to remove
764 * an matching exception instead. And be silent about it: we
765 * don't want to break compatibility
767 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
768 dev_exception_rm(devcgroup, &ex);
770 rc = dev_exception_add(devcgroup, &ex);
774 /* we only propagate new restrictions */
775 rc = propagate_exception(devcgroup, &ex);
783 static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
784 char *buf, size_t nbytes, loff_t off)
788 mutex_lock(&devcgroup_mutex);
789 retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
790 of_cft(of)->private, strstrip(buf));
791 mutex_unlock(&devcgroup_mutex);
792 return retval ?: nbytes;
795 static struct cftype dev_cgroup_files[] = {
798 .write = devcgroup_access_write,
799 .private = DEVCG_ALLOW,
803 .write = devcgroup_access_write,
804 .private = DEVCG_DENY,
808 .seq_show = devcgroup_seq_show,
809 .private = DEVCG_LIST,
814 struct cgroup_subsys devices_cgrp_subsys = {
815 .css_alloc = devcgroup_css_alloc,
816 .css_free = devcgroup_css_free,
817 .css_online = devcgroup_online,
818 .css_offline = devcgroup_offline,
819 .legacy_cftypes = dev_cgroup_files,
823 * devcgroup_legacy_check_permission - checks if an inode operation is permitted
824 * @dev_cgroup: the dev cgroup to be tested against
826 * @major: device major number
827 * @minor: device minor number
828 * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
830 * returns 0 on success, -EPERM case the operation is not permitted
832 static int devcgroup_legacy_check_permission(short type, u32 major, u32 minor,
835 struct dev_cgroup *dev_cgroup;
839 dev_cgroup = task_devcgroup(current);
840 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
841 /* Can't match any of the exceptions, even partially */
842 rc = !match_exception_partial(&dev_cgroup->exceptions,
843 type, major, minor, access);
845 /* Need to match completely one exception to be allowed */
846 rc = match_exception(&dev_cgroup->exceptions, type, major,
856 #endif /* CONFIG_CGROUP_DEVICE */
858 #if defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF)
860 int devcgroup_check_permission(short type, u32 major, u32 minor, short access)
862 int rc = BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type, major, minor, access);
867 #ifdef CONFIG_CGROUP_DEVICE
868 return devcgroup_legacy_check_permission(type, major, minor, access);
870 #else /* CONFIG_CGROUP_DEVICE */
873 #endif /* CONFIG_CGROUP_DEVICE */
875 EXPORT_SYMBOL(devcgroup_check_permission);
876 #endif /* defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF) */