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 static DEFINE_MUTEX(devcgroup_mutex);
27 * exception list locking rules:
28 * hold devcgroup_mutex for update/read.
29 * hold rcu_read_lock() for read.
32 struct dev_exception_item {
36 struct list_head list;
41 struct cgroup_subsys_state css;
42 struct list_head exceptions;
43 enum devcg_behavior behavior;
46 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
48 return s ? container_of(s, struct dev_cgroup, css) : NULL;
51 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
53 return css_to_devcgroup(task_css(task, devices_cgrp_id));
57 * called under devcgroup_mutex
59 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
61 struct dev_exception_item *ex, *tmp, *new;
63 lockdep_assert_held(&devcgroup_mutex);
65 list_for_each_entry(ex, orig, list) {
66 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
69 list_add_tail(&new->list, dest);
75 list_for_each_entry_safe(ex, tmp, dest, list) {
83 * called under devcgroup_mutex
85 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
86 struct dev_exception_item *ex)
88 struct dev_exception_item *excopy, *walk;
90 lockdep_assert_held(&devcgroup_mutex);
92 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
96 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
97 if (walk->type != ex->type)
99 if (walk->major != ex->major)
101 if (walk->minor != ex->minor)
104 walk->access |= ex->access;
110 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
115 * called under devcgroup_mutex
117 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
118 struct dev_exception_item *ex)
120 struct dev_exception_item *walk, *tmp;
122 lockdep_assert_held(&devcgroup_mutex);
124 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
125 if (walk->type != ex->type)
127 if (walk->major != ex->major)
129 if (walk->minor != ex->minor)
132 walk->access &= ~ex->access;
134 list_del_rcu(&walk->list);
135 kfree_rcu(walk, rcu);
140 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
142 struct dev_exception_item *ex, *tmp;
144 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
145 list_del_rcu(&ex->list);
151 * dev_exception_clean - frees all entries of the exception list
152 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
154 * called under devcgroup_mutex
156 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
158 lockdep_assert_held(&devcgroup_mutex);
160 __dev_exception_clean(dev_cgroup);
163 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
165 return (devcg->behavior != DEVCG_DEFAULT_NONE);
169 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
171 * @css: css getting online
172 * returns 0 in case of success, error code otherwise
174 static int devcgroup_online(struct cgroup_subsys_state *css)
176 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
177 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
180 mutex_lock(&devcgroup_mutex);
182 if (parent_dev_cgroup == NULL)
183 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
185 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
186 &parent_dev_cgroup->exceptions);
188 dev_cgroup->behavior = parent_dev_cgroup->behavior;
190 mutex_unlock(&devcgroup_mutex);
195 static void devcgroup_offline(struct cgroup_subsys_state *css)
197 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
199 mutex_lock(&devcgroup_mutex);
200 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
201 mutex_unlock(&devcgroup_mutex);
205 * called from kernel/cgroup.c with cgroup_lock() held.
207 static struct cgroup_subsys_state *
208 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
210 struct dev_cgroup *dev_cgroup;
212 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
214 return ERR_PTR(-ENOMEM);
215 INIT_LIST_HEAD(&dev_cgroup->exceptions);
216 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
218 return &dev_cgroup->css;
221 static void devcgroup_css_free(struct cgroup_subsys_state *css)
223 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
225 __dev_exception_clean(dev_cgroup);
229 #define DEVCG_ALLOW 1
236 static void set_access(char *acc, short access)
239 memset(acc, 0, ACCLEN);
240 if (access & DEVCG_ACC_READ)
242 if (access & DEVCG_ACC_WRITE)
244 if (access & DEVCG_ACC_MKNOD)
248 static char type_to_char(short type)
250 if (type == DEVCG_DEV_ALL)
252 if (type == DEVCG_DEV_CHAR)
254 if (type == DEVCG_DEV_BLOCK)
259 static void set_majmin(char *str, unsigned m)
264 sprintf(str, "%u", m);
267 static int devcgroup_seq_show(struct seq_file *m, void *v)
269 struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
270 struct dev_exception_item *ex;
271 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
275 * To preserve the compatibility:
276 * - Only show the "all devices" when the default policy is to allow
277 * - List the exceptions in case the default policy is to deny
278 * This way, the file remains as a "whitelist of devices"
280 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
281 set_access(acc, DEVCG_ACC_MASK);
284 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEVCG_DEV_ALL),
287 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
288 set_access(acc, ex->access);
289 set_majmin(maj, ex->major);
290 set_majmin(min, ex->minor);
291 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
301 * match_exception - iterates the exception list trying to find a complete match
302 * @exceptions: list of exceptions
303 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
304 * @major: device file major number, ~0 to match all
305 * @minor: device file minor number, ~0 to match all
306 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
308 * It is considered a complete match if an exception is found that will
309 * contain the entire range of provided parameters.
311 * Return: true in case it matches an exception completely
313 static bool match_exception(struct list_head *exceptions, short type,
314 u32 major, u32 minor, short access)
316 struct dev_exception_item *ex;
318 list_for_each_entry_rcu(ex, exceptions, list) {
319 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
321 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
323 if (ex->major != ~0 && ex->major != major)
325 if (ex->minor != ~0 && ex->minor != minor)
327 /* provided access cannot have more than the exception rule */
328 if (access & (~ex->access))
336 * match_exception_partial - iterates the exception list trying to find a partial match
337 * @exceptions: list of exceptions
338 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
339 * @major: device file major number, ~0 to match all
340 * @minor: device file minor number, ~0 to match all
341 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
343 * It is considered a partial match if an exception's range is found to
344 * contain *any* of the devices specified by provided parameters. This is
345 * used to make sure no extra access is being granted that is forbidden by
346 * any of the exception list.
348 * Return: true in case the provided range mat matches an exception completely
350 static bool match_exception_partial(struct list_head *exceptions, short type,
351 u32 major, u32 minor, short access)
353 struct dev_exception_item *ex;
355 list_for_each_entry_rcu(ex, exceptions, list,
356 lockdep_is_held(&devcgroup_mutex)) {
357 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
359 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
362 * We must be sure that both the exception and the provided
363 * range aren't masking all devices
365 if (ex->major != ~0 && major != ~0 && ex->major != major)
367 if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
370 * In order to make sure the provided range isn't matching
371 * an exception, all its access bits shouldn't match the
372 * exception's access bits
374 if (!(access & ex->access))
382 * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
383 * @dev_cgroup: dev cgroup to be tested against
384 * @refex: new exception
385 * @behavior: behavior of the exception's dev_cgroup
387 * This is used to make sure a child cgroup won't have more privileges
390 static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
391 struct dev_exception_item *refex,
392 enum devcg_behavior behavior)
396 RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
397 !lockdep_is_held(&devcgroup_mutex),
398 "device_cgroup:verify_new_ex called without proper synchronization");
400 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
401 if (behavior == DEVCG_DEFAULT_ALLOW) {
403 * new exception in the child doesn't matter, only
404 * adding extra restrictions
409 * new exception in the child will add more devices
410 * that can be acessed, so it can't match any of
411 * parent's exceptions, even slightly
413 match = match_exception_partial(&dev_cgroup->exceptions,
425 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
426 * the new exception will add access to more devices and must
427 * be contained completely in an parent's exception to be
430 match = match_exception(&dev_cgroup->exceptions, refex->type,
431 refex->major, refex->minor,
435 /* parent has an exception that matches the proposed */
445 * when adding a new allow rule to a device exception list, the rule
446 * must be allowed in the parent device
448 static int parent_has_perm(struct dev_cgroup *childcg,
449 struct dev_exception_item *ex)
451 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
455 return verify_new_ex(parent, ex, childcg->behavior);
459 * parent_allows_removal - verify if it's ok to remove an exception
460 * @childcg: child cgroup from where the exception will be removed
461 * @ex: exception being removed
463 * When removing an exception in cgroups with default ALLOW policy, it must
464 * be checked if removing it will give the child cgroup more access than the
467 * Return: true if it's ok to remove exception, false otherwise
469 static bool parent_allows_removal(struct dev_cgroup *childcg,
470 struct dev_exception_item *ex)
472 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
477 /* It's always allowed to remove access to devices */
478 if (childcg->behavior == DEVCG_DEFAULT_DENY)
482 * Make sure you're not removing part or a whole exception existing in
485 return !match_exception_partial(&parent->exceptions, ex->type,
486 ex->major, ex->minor, ex->access);
490 * may_allow_all - checks if it's possible to change the behavior to
491 * allow based on parent's rules.
492 * @parent: device cgroup's parent
493 * returns: != 0 in case it's allowed, 0 otherwise
495 static inline int may_allow_all(struct dev_cgroup *parent)
499 return parent->behavior == DEVCG_DEFAULT_ALLOW;
503 * revalidate_active_exceptions - walks through the active exception list and
504 * revalidates the exceptions based on parent's
505 * behavior and exceptions. The exceptions that
506 * are no longer valid will be removed.
507 * Called with devcgroup_mutex held.
508 * @devcg: cgroup which exceptions will be checked
510 * This is one of the three key functions for hierarchy implementation.
511 * This function is responsible for re-evaluating all the cgroup's active
512 * exceptions due to a parent's exception change.
513 * Refer to Documentation/admin-guide/cgroup-v1/devices.rst for more details.
515 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
517 struct dev_exception_item *ex;
518 struct list_head *this, *tmp;
520 list_for_each_safe(this, tmp, &devcg->exceptions) {
521 ex = container_of(this, struct dev_exception_item, list);
522 if (!parent_has_perm(devcg, ex))
523 dev_exception_rm(devcg, ex);
528 * propagate_exception - propagates a new exception to the children
529 * @devcg_root: device cgroup that added a new exception
530 * @ex: new exception to be propagated
532 * returns: 0 in case of success, != 0 in case of error
534 static int propagate_exception(struct dev_cgroup *devcg_root,
535 struct dev_exception_item *ex)
537 struct cgroup_subsys_state *pos;
542 css_for_each_descendant_pre(pos, &devcg_root->css) {
543 struct dev_cgroup *devcg = css_to_devcgroup(pos);
546 * Because devcgroup_mutex is held, no devcg will become
547 * online or offline during the tree walk (see on/offline
548 * methods), and online ones are safe to access outside RCU
549 * read lock without bumping refcnt.
551 if (pos == &devcg_root->css || !is_devcg_online(devcg))
557 * in case both root's behavior and devcg is allow, a new
558 * restriction means adding to the exception list
560 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
561 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
562 rc = dev_exception_add(devcg, ex);
567 * in the other possible cases:
568 * root's behavior: allow, devcg's: deny
569 * root's behavior: deny, devcg's: deny
570 * the exception will be removed
572 dev_exception_rm(devcg, ex);
574 revalidate_active_exceptions(devcg);
584 * Modify the exception list using allow/deny rules.
585 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
586 * so we can give a container CAP_MKNOD to let it create devices but not
587 * modify the exception list.
588 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
589 * us to also grant CAP_SYS_ADMIN to containers without giving away the
590 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
592 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
593 * new access is only allowed if you're in the top-level cgroup, or your
594 * parent cgroup has the access you're asking for.
596 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
597 int filetype, char *buffer)
600 char temp[12]; /* 11 + 1 characters needed for a u32 */
602 struct dev_exception_item ex;
603 struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
605 if (!capable(CAP_SYS_ADMIN))
608 memset(&ex, 0, sizeof(ex));
615 if (css_has_online_children(&devcgroup->css))
618 if (!may_allow_all(parent))
620 dev_exception_clean(devcgroup);
621 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
625 rc = dev_exceptions_copy(&devcgroup->exceptions,
626 &parent->exceptions);
631 if (css_has_online_children(&devcgroup->css))
634 dev_exception_clean(devcgroup);
635 devcgroup->behavior = DEVCG_DEFAULT_DENY;
642 ex.type = DEVCG_DEV_BLOCK;
645 ex.type = DEVCG_DEV_CHAR;
657 } else if (isdigit(*b)) {
658 memset(temp, 0, sizeof(temp));
659 for (count = 0; count < sizeof(temp) - 1; count++) {
665 rc = kstrtou32(temp, 10, &ex.major);
679 } else if (isdigit(*b)) {
680 memset(temp, 0, sizeof(temp));
681 for (count = 0; count < sizeof(temp) - 1; count++) {
687 rc = kstrtou32(temp, 10, &ex.minor);
695 for (b++, count = 0; count < 3; count++, b++) {
698 ex.access |= DEVCG_ACC_READ;
701 ex.access |= DEVCG_ACC_WRITE;
704 ex.access |= DEVCG_ACC_MKNOD;
718 * If the default policy is to allow by default, try to remove
719 * an matching exception instead. And be silent about it: we
720 * don't want to break compatibility
722 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
723 /* Check if the parent allows removing it first */
724 if (!parent_allows_removal(devcgroup, &ex))
726 dev_exception_rm(devcgroup, &ex);
730 if (!parent_has_perm(devcgroup, &ex))
732 rc = dev_exception_add(devcgroup, &ex);
736 * If the default policy is to deny by default, try to remove
737 * an matching exception instead. And be silent about it: we
738 * don't want to break compatibility
740 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
741 dev_exception_rm(devcgroup, &ex);
743 rc = dev_exception_add(devcgroup, &ex);
747 /* we only propagate new restrictions */
748 rc = propagate_exception(devcgroup, &ex);
756 static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
757 char *buf, size_t nbytes, loff_t off)
761 mutex_lock(&devcgroup_mutex);
762 retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
763 of_cft(of)->private, strstrip(buf));
764 mutex_unlock(&devcgroup_mutex);
765 return retval ?: nbytes;
768 static struct cftype dev_cgroup_files[] = {
771 .write = devcgroup_access_write,
772 .private = DEVCG_ALLOW,
776 .write = devcgroup_access_write,
777 .private = DEVCG_DENY,
781 .seq_show = devcgroup_seq_show,
782 .private = DEVCG_LIST,
787 struct cgroup_subsys devices_cgrp_subsys = {
788 .css_alloc = devcgroup_css_alloc,
789 .css_free = devcgroup_css_free,
790 .css_online = devcgroup_online,
791 .css_offline = devcgroup_offline,
792 .legacy_cftypes = dev_cgroup_files,
796 * __devcgroup_check_permission - checks if an inode operation is permitted
797 * @dev_cgroup: the dev cgroup to be tested against
799 * @major: device major number
800 * @minor: device minor number
801 * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
803 * returns 0 on success, -EPERM case the operation is not permitted
805 int __devcgroup_check_permission(short type, u32 major, u32 minor,
808 struct dev_cgroup *dev_cgroup;
812 dev_cgroup = task_devcgroup(current);
813 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
814 /* Can't match any of the exceptions, even partially */
815 rc = !match_exception_partial(&dev_cgroup->exceptions,
816 type, major, minor, access);
818 /* Need to match completely one exception to be allowed */
819 rc = match_exception(&dev_cgroup->exceptions, type, major,