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 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
369 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
372 * We must be sure that both the exception and the provided
373 * range aren't masking all devices
375 if (ex->major != ~0 && major != ~0 && ex->major != major)
377 if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
380 * In order to make sure the provided range isn't matching
381 * an exception, all its access bits shouldn't match the
382 * exception's access bits
384 if (!(access & ex->access))
392 * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
393 * @dev_cgroup: dev cgroup to be tested against
394 * @refex: new exception
395 * @behavior: behavior of the exception's dev_cgroup
397 * This is used to make sure a child cgroup won't have more privileges
400 static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
401 struct dev_exception_item *refex,
402 enum devcg_behavior behavior)
406 RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
407 !lockdep_is_held(&devcgroup_mutex),
408 "device_cgroup:verify_new_ex called without proper synchronization");
410 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
411 if (behavior == DEVCG_DEFAULT_ALLOW) {
413 * new exception in the child doesn't matter, only
414 * adding extra restrictions
419 * new exception in the child will add more devices
420 * that can be acessed, so it can't match any of
421 * parent's exceptions, even slightly
423 match = match_exception_partial(&dev_cgroup->exceptions,
435 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
436 * the new exception will add access to more devices and must
437 * be contained completely in an parent's exception to be
440 match = match_exception(&dev_cgroup->exceptions, refex->type,
441 refex->major, refex->minor,
445 /* parent has an exception that matches the proposed */
455 * when adding a new allow rule to a device exception list, the rule
456 * must be allowed in the parent device
458 static int parent_has_perm(struct dev_cgroup *childcg,
459 struct dev_exception_item *ex)
461 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
465 return verify_new_ex(parent, ex, childcg->behavior);
469 * parent_allows_removal - verify if it's ok to remove an exception
470 * @childcg: child cgroup from where the exception will be removed
471 * @ex: exception being removed
473 * When removing an exception in cgroups with default ALLOW policy, it must
474 * be checked if removing it will give the child cgroup more access than the
477 * Return: true if it's ok to remove exception, false otherwise
479 static bool parent_allows_removal(struct dev_cgroup *childcg,
480 struct dev_exception_item *ex)
482 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
487 /* It's always allowed to remove access to devices */
488 if (childcg->behavior == DEVCG_DEFAULT_DENY)
492 * Make sure you're not removing part or a whole exception existing in
495 return !match_exception_partial(&parent->exceptions, ex->type,
496 ex->major, ex->minor, ex->access);
500 * may_allow_all - checks if it's possible to change the behavior to
501 * allow based on parent's rules.
502 * @parent: device cgroup's parent
503 * returns: != 0 in case it's allowed, 0 otherwise
505 static inline int may_allow_all(struct dev_cgroup *parent)
509 return parent->behavior == DEVCG_DEFAULT_ALLOW;
513 * revalidate_active_exceptions - walks through the active exception list and
514 * revalidates the exceptions based on parent's
515 * behavior and exceptions. The exceptions that
516 * are no longer valid will be removed.
517 * Called with devcgroup_mutex held.
518 * @devcg: cgroup which exceptions will be checked
520 * This is one of the three key functions for hierarchy implementation.
521 * This function is responsible for re-evaluating all the cgroup's active
522 * exceptions due to a parent's exception change.
523 * Refer to Documentation/cgroup-v1/devices.txt for more details.
525 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
527 struct dev_exception_item *ex;
528 struct list_head *this, *tmp;
530 list_for_each_safe(this, tmp, &devcg->exceptions) {
531 ex = container_of(this, struct dev_exception_item, list);
532 if (!parent_has_perm(devcg, ex))
533 dev_exception_rm(devcg, ex);
538 * propagate_exception - propagates a new exception to the children
539 * @devcg_root: device cgroup that added a new exception
540 * @ex: new exception to be propagated
542 * returns: 0 in case of success, != 0 in case of error
544 static int propagate_exception(struct dev_cgroup *devcg_root,
545 struct dev_exception_item *ex)
547 struct cgroup_subsys_state *pos;
552 css_for_each_descendant_pre(pos, &devcg_root->css) {
553 struct dev_cgroup *devcg = css_to_devcgroup(pos);
556 * Because devcgroup_mutex is held, no devcg will become
557 * online or offline during the tree walk (see on/offline
558 * methods), and online ones are safe to access outside RCU
559 * read lock without bumping refcnt.
561 if (pos == &devcg_root->css || !is_devcg_online(devcg))
567 * in case both root's behavior and devcg is allow, a new
568 * restriction means adding to the exception list
570 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
571 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
572 rc = dev_exception_add(devcg, ex);
577 * in the other possible cases:
578 * root's behavior: allow, devcg's: deny
579 * root's behavior: deny, devcg's: deny
580 * the exception will be removed
582 dev_exception_rm(devcg, ex);
584 revalidate_active_exceptions(devcg);
594 * Modify the exception list using allow/deny rules.
595 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
596 * so we can give a container CAP_MKNOD to let it create devices but not
597 * modify the exception list.
598 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
599 * us to also grant CAP_SYS_ADMIN to containers without giving away the
600 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
602 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
603 * new access is only allowed if you're in the top-level cgroup, or your
604 * parent cgroup has the access you're asking for.
606 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
607 int filetype, char *buffer)
610 char temp[12]; /* 11 + 1 characters needed for a u32 */
612 struct dev_exception_item ex;
613 struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
614 struct dev_cgroup tmp_devcgrp;
616 if (!capable(CAP_SYS_ADMIN))
619 memset(&ex, 0, sizeof(ex));
620 memset(&tmp_devcgrp, 0, sizeof(tmp_devcgrp));
627 if (css_has_online_children(&devcgroup->css))
630 if (!may_allow_all(parent))
633 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
634 dev_exception_clean(devcgroup);
638 INIT_LIST_HEAD(&tmp_devcgrp.exceptions);
639 rc = dev_exceptions_copy(&tmp_devcgrp.exceptions,
640 &devcgroup->exceptions);
643 dev_exception_clean(devcgroup);
644 rc = dev_exceptions_copy(&devcgroup->exceptions,
645 &parent->exceptions);
647 dev_exceptions_move(&devcgroup->exceptions,
648 &tmp_devcgrp.exceptions);
651 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
652 dev_exception_clean(&tmp_devcgrp);
655 if (css_has_online_children(&devcgroup->css))
658 dev_exception_clean(devcgroup);
659 devcgroup->behavior = DEVCG_DEFAULT_DENY;
666 ex.type = DEVCG_DEV_BLOCK;
669 ex.type = DEVCG_DEV_CHAR;
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.major);
703 } else if (isdigit(*b)) {
704 memset(temp, 0, sizeof(temp));
705 for (count = 0; count < sizeof(temp) - 1; count++) {
711 rc = kstrtou32(temp, 10, &ex.minor);
719 for (b++, count = 0; count < 3; count++, b++) {
722 ex.access |= DEVCG_ACC_READ;
725 ex.access |= DEVCG_ACC_WRITE;
728 ex.access |= DEVCG_ACC_MKNOD;
742 * If the default policy is to allow by default, try to remove
743 * an matching exception instead. And be silent about it: we
744 * don't want to break compatibility
746 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
747 /* Check if the parent allows removing it first */
748 if (!parent_allows_removal(devcgroup, &ex))
750 dev_exception_rm(devcgroup, &ex);
754 if (!parent_has_perm(devcgroup, &ex))
756 rc = dev_exception_add(devcgroup, &ex);
760 * If the default policy is to deny by default, try to remove
761 * an matching exception instead. And be silent about it: we
762 * don't want to break compatibility
764 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
765 dev_exception_rm(devcgroup, &ex);
767 rc = dev_exception_add(devcgroup, &ex);
771 /* we only propagate new restrictions */
772 rc = propagate_exception(devcgroup, &ex);
780 static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
781 char *buf, size_t nbytes, loff_t off)
785 mutex_lock(&devcgroup_mutex);
786 retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
787 of_cft(of)->private, strstrip(buf));
788 mutex_unlock(&devcgroup_mutex);
789 return retval ?: nbytes;
792 static struct cftype dev_cgroup_files[] = {
795 .write = devcgroup_access_write,
796 .private = DEVCG_ALLOW,
800 .write = devcgroup_access_write,
801 .private = DEVCG_DENY,
805 .seq_show = devcgroup_seq_show,
806 .private = DEVCG_LIST,
811 struct cgroup_subsys devices_cgrp_subsys = {
812 .css_alloc = devcgroup_css_alloc,
813 .css_free = devcgroup_css_free,
814 .css_online = devcgroup_online,
815 .css_offline = devcgroup_offline,
816 .legacy_cftypes = dev_cgroup_files,
820 * __devcgroup_check_permission - checks if an inode operation is permitted
821 * @dev_cgroup: the dev cgroup to be tested against
823 * @major: device major number
824 * @minor: device minor number
825 * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
827 * returns 0 on success, -EPERM case the operation is not permitted
829 int __devcgroup_check_permission(short type, u32 major, u32 minor,
832 struct dev_cgroup *dev_cgroup;
836 dev_cgroup = task_devcgroup(current);
837 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
838 /* Can't match any of the exceptions, even partially */
839 rc = !match_exception_partial(&dev_cgroup->exceptions,
840 type, major, minor, access);
842 /* Need to match completely one exception to be allowed */
843 rc = match_exception(&dev_cgroup->exceptions, type, major,