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) {
82 static void dev_exceptions_move(struct list_head *dest, struct list_head *orig)
84 struct dev_exception_item *ex, *tmp;
86 lockdep_assert_held(&devcgroup_mutex);
88 list_for_each_entry_safe(ex, tmp, orig, list) {
89 list_move_tail(&ex->list, dest);
94 * called under devcgroup_mutex
96 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
97 struct dev_exception_item *ex)
99 struct dev_exception_item *excopy, *walk;
101 lockdep_assert_held(&devcgroup_mutex);
103 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
107 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
108 if (walk->type != ex->type)
110 if (walk->major != ex->major)
112 if (walk->minor != ex->minor)
115 walk->access |= ex->access;
121 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
126 * called under devcgroup_mutex
128 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
129 struct dev_exception_item *ex)
131 struct dev_exception_item *walk, *tmp;
133 lockdep_assert_held(&devcgroup_mutex);
135 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
136 if (walk->type != ex->type)
138 if (walk->major != ex->major)
140 if (walk->minor != ex->minor)
143 walk->access &= ~ex->access;
145 list_del_rcu(&walk->list);
146 kfree_rcu(walk, rcu);
151 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
153 struct dev_exception_item *ex, *tmp;
155 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
156 list_del_rcu(&ex->list);
162 * dev_exception_clean - frees all entries of the exception list
163 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
165 * called under devcgroup_mutex
167 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
169 lockdep_assert_held(&devcgroup_mutex);
171 __dev_exception_clean(dev_cgroup);
174 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
176 return (devcg->behavior != DEVCG_DEFAULT_NONE);
180 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
182 * @css: css getting online
183 * returns 0 in case of success, error code otherwise
185 static int devcgroup_online(struct cgroup_subsys_state *css)
187 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
188 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
191 mutex_lock(&devcgroup_mutex);
193 if (parent_dev_cgroup == NULL)
194 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
196 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
197 &parent_dev_cgroup->exceptions);
199 dev_cgroup->behavior = parent_dev_cgroup->behavior;
201 mutex_unlock(&devcgroup_mutex);
206 static void devcgroup_offline(struct cgroup_subsys_state *css)
208 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
210 mutex_lock(&devcgroup_mutex);
211 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
212 mutex_unlock(&devcgroup_mutex);
216 * called from kernel/cgroup.c with cgroup_lock() held.
218 static struct cgroup_subsys_state *
219 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
221 struct dev_cgroup *dev_cgroup;
223 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
225 return ERR_PTR(-ENOMEM);
226 INIT_LIST_HEAD(&dev_cgroup->exceptions);
227 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
229 return &dev_cgroup->css;
232 static void devcgroup_css_free(struct cgroup_subsys_state *css)
234 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
236 __dev_exception_clean(dev_cgroup);
240 #define DEVCG_ALLOW 1
247 static void set_access(char *acc, short access)
250 memset(acc, 0, ACCLEN);
251 if (access & DEVCG_ACC_READ)
253 if (access & DEVCG_ACC_WRITE)
255 if (access & DEVCG_ACC_MKNOD)
259 static char type_to_char(short type)
261 if (type == DEVCG_DEV_ALL)
263 if (type == DEVCG_DEV_CHAR)
265 if (type == DEVCG_DEV_BLOCK)
270 static void set_majmin(char *str, unsigned m)
275 sprintf(str, "%u", m);
278 static int devcgroup_seq_show(struct seq_file *m, void *v)
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];
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"
291 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
292 set_access(acc, DEVCG_ACC_MASK);
295 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEVCG_DEV_ALL),
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),
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)
319 * It is considered a complete match if an exception is found that will
320 * contain the entire range of provided parameters.
322 * Return: true in case it matches an exception completely
324 static bool match_exception(struct list_head *exceptions, short type,
325 u32 major, u32 minor, short access)
327 struct dev_exception_item *ex;
329 list_for_each_entry_rcu(ex, exceptions, list) {
330 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
332 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
334 if (ex->major != ~0 && ex->major != major)
336 if (ex->minor != ~0 && ex->minor != minor)
338 /* provided access cannot have more than the exception rule */
339 if (access & (~ex->access))
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)
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.
359 * Return: true in case the provided range mat matches an exception completely
361 static bool match_exception_partial(struct list_head *exceptions, short type,
362 u32 major, u32 minor, short access)
364 struct dev_exception_item *ex;
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))
370 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
373 * We must be sure that both the exception and the provided
374 * range aren't masking all devices
376 if (ex->major != ~0 && major != ~0 && ex->major != major)
378 if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
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
385 if (!(access & ex->access))
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
398 * This is used to make sure a child cgroup won't have more privileges
401 static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
402 struct dev_exception_item *refex,
403 enum devcg_behavior behavior)
407 RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
408 !lockdep_is_held(&devcgroup_mutex),
409 "device_cgroup:verify_new_ex called without proper synchronization");
411 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
412 if (behavior == DEVCG_DEFAULT_ALLOW) {
414 * new exception in the child doesn't matter, only
415 * adding extra restrictions
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
424 match = match_exception_partial(&dev_cgroup->exceptions,
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
441 match = match_exception(&dev_cgroup->exceptions, refex->type,
442 refex->major, refex->minor,
446 /* parent has an exception that matches the proposed */
456 * when adding a new allow rule to a device exception list, the rule
457 * must be allowed in the parent device
459 static int parent_has_perm(struct dev_cgroup *childcg,
460 struct dev_exception_item *ex)
462 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
466 return verify_new_ex(parent, ex, childcg->behavior);
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
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
478 * Return: true if it's ok to remove exception, false otherwise
480 static bool parent_allows_removal(struct dev_cgroup *childcg,
481 struct dev_exception_item *ex)
483 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
488 /* It's always allowed to remove access to devices */
489 if (childcg->behavior == DEVCG_DEFAULT_DENY)
493 * Make sure you're not removing part or a whole exception existing in
496 return !match_exception_partial(&parent->exceptions, ex->type,
497 ex->major, ex->minor, ex->access);
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
506 static inline int may_allow_all(struct dev_cgroup *parent)
510 return parent->behavior == DEVCG_DEFAULT_ALLOW;
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
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.
526 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
528 struct dev_exception_item *ex;
529 struct list_head *this, *tmp;
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);
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
543 * returns: 0 in case of success, != 0 in case of error
545 static int propagate_exception(struct dev_cgroup *devcg_root,
546 struct dev_exception_item *ex)
548 struct cgroup_subsys_state *pos;
553 css_for_each_descendant_pre(pos, &devcg_root->css) {
554 struct dev_cgroup *devcg = css_to_devcgroup(pos);
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.
562 if (pos == &devcg_root->css || !is_devcg_online(devcg))
568 * in case both root's behavior and devcg is allow, a new
569 * restriction means adding to the exception list
571 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
572 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
573 rc = dev_exception_add(devcg, ex);
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
583 dev_exception_rm(devcg, ex);
585 revalidate_active_exceptions(devcg);
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
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.
607 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
608 int filetype, char *buffer)
611 char temp[12]; /* 11 + 1 characters needed for a u32 */
613 struct dev_exception_item ex;
614 struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
615 struct dev_cgroup tmp_devcgrp;
617 if (!capable(CAP_SYS_ADMIN))
620 memset(&ex, 0, sizeof(ex));
621 memset(&tmp_devcgrp, 0, sizeof(tmp_devcgrp));
628 if (css_has_online_children(&devcgroup->css))
631 if (!may_allow_all(parent))
634 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
635 dev_exception_clean(devcgroup);
639 INIT_LIST_HEAD(&tmp_devcgrp.exceptions);
640 rc = dev_exceptions_copy(&tmp_devcgrp.exceptions,
641 &devcgroup->exceptions);
644 dev_exception_clean(devcgroup);
645 rc = dev_exceptions_copy(&devcgroup->exceptions,
646 &parent->exceptions);
648 dev_exceptions_move(&devcgroup->exceptions,
649 &tmp_devcgrp.exceptions);
652 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
653 dev_exception_clean(&tmp_devcgrp);
656 if (css_has_online_children(&devcgroup->css))
659 dev_exception_clean(devcgroup);
660 devcgroup->behavior = DEVCG_DEFAULT_DENY;
667 ex.type = DEVCG_DEV_BLOCK;
670 ex.type = DEVCG_DEV_CHAR;
682 } else if (isdigit(*b)) {
683 memset(temp, 0, sizeof(temp));
684 for (count = 0; count < sizeof(temp) - 1; count++) {
690 rc = kstrtou32(temp, 10, &ex.major);
704 } else if (isdigit(*b)) {
705 memset(temp, 0, sizeof(temp));
706 for (count = 0; count < sizeof(temp) - 1; count++) {
712 rc = kstrtou32(temp, 10, &ex.minor);
720 for (b++, count = 0; count < 3; count++, b++) {
723 ex.access |= DEVCG_ACC_READ;
726 ex.access |= DEVCG_ACC_WRITE;
729 ex.access |= DEVCG_ACC_MKNOD;
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
747 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
748 /* Check if the parent allows removing it first */
749 if (!parent_allows_removal(devcgroup, &ex))
751 dev_exception_rm(devcgroup, &ex);
755 if (!parent_has_perm(devcgroup, &ex))
757 rc = dev_exception_add(devcgroup, &ex);
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
765 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
766 dev_exception_rm(devcgroup, &ex);
768 rc = dev_exception_add(devcgroup, &ex);
772 /* we only propagate new restrictions */
773 rc = propagate_exception(devcgroup, &ex);
781 static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
782 char *buf, size_t nbytes, loff_t off)
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;
793 static struct cftype dev_cgroup_files[] = {
796 .write = devcgroup_access_write,
797 .private = DEVCG_ALLOW,
801 .write = devcgroup_access_write,
802 .private = DEVCG_DENY,
806 .seq_show = devcgroup_seq_show,
807 .private = DEVCG_LIST,
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,
821 * __devcgroup_check_permission - checks if an inode operation is permitted
822 * @dev_cgroup: the dev cgroup to be tested against
824 * @major: device major number
825 * @minor: device minor number
826 * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
828 * returns 0 on success, -EPERM case the operation is not permitted
830 int __devcgroup_check_permission(short type, u32 major, u32 minor,
833 struct dev_cgroup *dev_cgroup;
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);
843 /* Need to match completely one exception to be allowed */
844 rc = match_exception(&dev_cgroup->exceptions, type, major,