2 * device_cgroup.c - device cgroup subsystem
4 * Copyright 2007 IBM Corp
7 #include <linux/device_cgroup.h>
8 #include <linux/cgroup.h>
9 #include <linux/ctype.h>
10 #include <linux/list.h>
11 #include <linux/uaccess.h>
12 #include <linux/seq_file.h>
13 #include <linux/slab.h>
14 #include <linux/rcupdate.h>
15 #include <linux/mutex.h>
20 #define ACC_MASK (ACC_MKNOD | ACC_READ | ACC_WRITE)
24 #define DEV_ALL 4 /* this represents all devices */
26 static DEFINE_MUTEX(devcgroup_mutex
);
35 * exception list locking rules:
36 * hold devcgroup_mutex for update/read.
37 * hold rcu_read_lock() for read.
40 struct dev_exception_item
{
44 struct list_head list
;
49 struct cgroup_subsys_state css
;
50 struct list_head exceptions
;
51 enum devcg_behavior behavior
;
54 static inline struct dev_cgroup
*css_to_devcgroup(struct cgroup_subsys_state
*s
)
56 return s
? container_of(s
, struct dev_cgroup
, css
) : NULL
;
59 static inline struct dev_cgroup
*task_devcgroup(struct task_struct
*task
)
61 return css_to_devcgroup(task_css(task
, devices_cgrp_id
));
65 * called under devcgroup_mutex
67 static int dev_exceptions_copy(struct list_head
*dest
, struct list_head
*orig
)
69 struct dev_exception_item
*ex
, *tmp
, *new;
71 lockdep_assert_held(&devcgroup_mutex
);
73 list_for_each_entry(ex
, orig
, list
) {
74 new = kmemdup(ex
, sizeof(*ex
), GFP_KERNEL
);
77 list_add_tail(&new->list
, dest
);
83 list_for_each_entry_safe(ex
, tmp
, dest
, list
) {
91 * called under devcgroup_mutex
93 static int dev_exception_add(struct dev_cgroup
*dev_cgroup
,
94 struct dev_exception_item
*ex
)
96 struct dev_exception_item
*excopy
, *walk
;
98 lockdep_assert_held(&devcgroup_mutex
);
100 excopy
= kmemdup(ex
, sizeof(*ex
), GFP_KERNEL
);
104 list_for_each_entry(walk
, &dev_cgroup
->exceptions
, list
) {
105 if (walk
->type
!= ex
->type
)
107 if (walk
->major
!= ex
->major
)
109 if (walk
->minor
!= ex
->minor
)
112 walk
->access
|= ex
->access
;
118 list_add_tail_rcu(&excopy
->list
, &dev_cgroup
->exceptions
);
123 * called under devcgroup_mutex
125 static void dev_exception_rm(struct dev_cgroup
*dev_cgroup
,
126 struct dev_exception_item
*ex
)
128 struct dev_exception_item
*walk
, *tmp
;
130 lockdep_assert_held(&devcgroup_mutex
);
132 list_for_each_entry_safe(walk
, tmp
, &dev_cgroup
->exceptions
, list
) {
133 if (walk
->type
!= ex
->type
)
135 if (walk
->major
!= ex
->major
)
137 if (walk
->minor
!= ex
->minor
)
140 walk
->access
&= ~ex
->access
;
142 list_del_rcu(&walk
->list
);
143 kfree_rcu(walk
, rcu
);
148 static void __dev_exception_clean(struct dev_cgroup
*dev_cgroup
)
150 struct dev_exception_item
*ex
, *tmp
;
152 list_for_each_entry_safe(ex
, tmp
, &dev_cgroup
->exceptions
, list
) {
153 list_del_rcu(&ex
->list
);
159 * dev_exception_clean - frees all entries of the exception list
160 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
162 * called under devcgroup_mutex
164 static void dev_exception_clean(struct dev_cgroup
*dev_cgroup
)
166 lockdep_assert_held(&devcgroup_mutex
);
168 __dev_exception_clean(dev_cgroup
);
171 static inline bool is_devcg_online(const struct dev_cgroup
*devcg
)
173 return (devcg
->behavior
!= DEVCG_DEFAULT_NONE
);
177 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
179 * @css: css getting online
180 * returns 0 in case of success, error code otherwise
182 static int devcgroup_online(struct cgroup_subsys_state
*css
)
184 struct dev_cgroup
*dev_cgroup
= css_to_devcgroup(css
);
185 struct dev_cgroup
*parent_dev_cgroup
= css_to_devcgroup(css_parent(css
));
188 mutex_lock(&devcgroup_mutex
);
190 if (parent_dev_cgroup
== NULL
)
191 dev_cgroup
->behavior
= DEVCG_DEFAULT_ALLOW
;
193 ret
= dev_exceptions_copy(&dev_cgroup
->exceptions
,
194 &parent_dev_cgroup
->exceptions
);
196 dev_cgroup
->behavior
= parent_dev_cgroup
->behavior
;
198 mutex_unlock(&devcgroup_mutex
);
203 static void devcgroup_offline(struct cgroup_subsys_state
*css
)
205 struct dev_cgroup
*dev_cgroup
= css_to_devcgroup(css
);
207 mutex_lock(&devcgroup_mutex
);
208 dev_cgroup
->behavior
= DEVCG_DEFAULT_NONE
;
209 mutex_unlock(&devcgroup_mutex
);
213 * called from kernel/cgroup.c with cgroup_lock() held.
215 static struct cgroup_subsys_state
*
216 devcgroup_css_alloc(struct cgroup_subsys_state
*parent_css
)
218 struct dev_cgroup
*dev_cgroup
;
220 dev_cgroup
= kzalloc(sizeof(*dev_cgroup
), GFP_KERNEL
);
222 return ERR_PTR(-ENOMEM
);
223 INIT_LIST_HEAD(&dev_cgroup
->exceptions
);
224 dev_cgroup
->behavior
= DEVCG_DEFAULT_NONE
;
226 return &dev_cgroup
->css
;
229 static void devcgroup_css_free(struct cgroup_subsys_state
*css
)
231 struct dev_cgroup
*dev_cgroup
= css_to_devcgroup(css
);
233 __dev_exception_clean(dev_cgroup
);
237 #define DEVCG_ALLOW 1
244 static void set_access(char *acc
, short access
)
247 memset(acc
, 0, ACCLEN
);
248 if (access
& ACC_READ
)
250 if (access
& ACC_WRITE
)
252 if (access
& ACC_MKNOD
)
256 static char type_to_char(short type
)
260 if (type
== DEV_CHAR
)
262 if (type
== DEV_BLOCK
)
267 static void set_majmin(char *str
, unsigned m
)
272 sprintf(str
, "%u", m
);
275 static int devcgroup_seq_show(struct seq_file
*m
, void *v
)
277 struct dev_cgroup
*devcgroup
= css_to_devcgroup(seq_css(m
));
278 struct dev_exception_item
*ex
;
279 char maj
[MAJMINLEN
], min
[MAJMINLEN
], acc
[ACCLEN
];
283 * To preserve the compatibility:
284 * - Only show the "all devices" when the default policy is to allow
285 * - List the exceptions in case the default policy is to deny
286 * This way, the file remains as a "whitelist of devices"
288 if (devcgroup
->behavior
== DEVCG_DEFAULT_ALLOW
) {
289 set_access(acc
, ACC_MASK
);
292 seq_printf(m
, "%c %s:%s %s\n", type_to_char(DEV_ALL
),
295 list_for_each_entry_rcu(ex
, &devcgroup
->exceptions
, list
) {
296 set_access(acc
, ex
->access
);
297 set_majmin(maj
, ex
->major
);
298 set_majmin(min
, ex
->minor
);
299 seq_printf(m
, "%c %s:%s %s\n", type_to_char(ex
->type
),
309 * match_exception - iterates the exception list trying to find a complete match
310 * @exceptions: list of exceptions
311 * @type: device type (DEV_BLOCK or DEV_CHAR)
312 * @major: device file major number, ~0 to match all
313 * @minor: device file minor number, ~0 to match all
314 * @access: permission mask (ACC_READ, ACC_WRITE, ACC_MKNOD)
316 * It is considered a complete match if an exception is found that will
317 * contain the entire range of provided parameters.
319 * Return: true in case it matches an exception completely
321 static bool match_exception(struct list_head
*exceptions
, short type
,
322 u32 major
, u32 minor
, short access
)
324 struct dev_exception_item
*ex
;
326 list_for_each_entry_rcu(ex
, exceptions
, list
) {
327 if ((type
& DEV_BLOCK
) && !(ex
->type
& DEV_BLOCK
))
329 if ((type
& DEV_CHAR
) && !(ex
->type
& DEV_CHAR
))
331 if (ex
->major
!= ~0 && ex
->major
!= major
)
333 if (ex
->minor
!= ~0 && ex
->minor
!= minor
)
335 /* provided access cannot have more than the exception rule */
336 if (access
& (~ex
->access
))
344 * match_exception_partial - iterates the exception list trying to find a partial match
345 * @exceptions: list of exceptions
346 * @type: device type (DEV_BLOCK or DEV_CHAR)
347 * @major: device file major number, ~0 to match all
348 * @minor: device file minor number, ~0 to match all
349 * @access: permission mask (ACC_READ, ACC_WRITE, ACC_MKNOD)
351 * It is considered a partial match if an exception's range is found to
352 * contain *any* of the devices specified by provided parameters. This is
353 * used to make sure no extra access is being granted that is forbidden by
354 * any of the exception list.
356 * Return: true in case the provided range mat matches an exception completely
358 static bool match_exception_partial(struct list_head
*exceptions
, short type
,
359 u32 major
, u32 minor
, short access
)
361 struct dev_exception_item
*ex
;
363 list_for_each_entry_rcu(ex
, exceptions
, list
) {
364 if ((type
& DEV_BLOCK
) && !(ex
->type
& DEV_BLOCK
))
366 if ((type
& DEV_CHAR
) && !(ex
->type
& DEV_CHAR
))
369 * We must be sure that both the exception and the provided
370 * range aren't masking all devices
372 if (ex
->major
!= ~0 && major
!= ~0 && ex
->major
!= major
)
374 if (ex
->minor
!= ~0 && minor
!= ~0 && ex
->minor
!= minor
)
377 * In order to make sure the provided range isn't matching
378 * an exception, all its access bits shouldn't match the
379 * exception's access bits
381 if (!(access
& ex
->access
))
389 * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
390 * @dev_cgroup: dev cgroup to be tested against
391 * @refex: new exception
392 * @behavior: behavior of the exception's dev_cgroup
394 * This is used to make sure a child cgroup won't have more privileges
397 static bool verify_new_ex(struct dev_cgroup
*dev_cgroup
,
398 struct dev_exception_item
*refex
,
399 enum devcg_behavior behavior
)
403 rcu_lockdep_assert(rcu_read_lock_held() ||
404 lockdep_is_held(&devcgroup_mutex
),
405 "device_cgroup:verify_new_ex called without proper synchronization");
407 if (dev_cgroup
->behavior
== DEVCG_DEFAULT_ALLOW
) {
408 if (behavior
== DEVCG_DEFAULT_ALLOW
) {
410 * new exception in the child doesn't matter, only
411 * adding extra restrictions
416 * new exception in the child will add more devices
417 * that can be acessed, so it can't match any of
418 * parent's exceptions, even slightly
420 match
= match_exception_partial(&dev_cgroup
->exceptions
,
432 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
433 * the new exception will add access to more devices and must
434 * be contained completely in an parent's exception to be
437 match
= match_exception(&dev_cgroup
->exceptions
, refex
->type
,
438 refex
->major
, refex
->minor
,
442 /* parent has an exception that matches the proposed */
452 * when adding a new allow rule to a device exception list, the rule
453 * must be allowed in the parent device
455 static int parent_has_perm(struct dev_cgroup
*childcg
,
456 struct dev_exception_item
*ex
)
458 struct dev_cgroup
*parent
= css_to_devcgroup(css_parent(&childcg
->css
));
462 return verify_new_ex(parent
, ex
, childcg
->behavior
);
466 * may_allow_all - checks if it's possible to change the behavior to
467 * allow based on parent's rules.
468 * @parent: device cgroup's parent
469 * returns: != 0 in case it's allowed, 0 otherwise
471 static inline int may_allow_all(struct dev_cgroup
*parent
)
475 return parent
->behavior
== DEVCG_DEFAULT_ALLOW
;
479 * revalidate_active_exceptions - walks through the active exception list and
480 * revalidates the exceptions based on parent's
481 * behavior and exceptions. The exceptions that
482 * are no longer valid will be removed.
483 * Called with devcgroup_mutex held.
484 * @devcg: cgroup which exceptions will be checked
486 * This is one of the three key functions for hierarchy implementation.
487 * This function is responsible for re-evaluating all the cgroup's active
488 * exceptions due to a parent's exception change.
489 * Refer to Documentation/cgroups/devices.txt for more details.
491 static void revalidate_active_exceptions(struct dev_cgroup
*devcg
)
493 struct dev_exception_item
*ex
;
494 struct list_head
*this, *tmp
;
496 list_for_each_safe(this, tmp
, &devcg
->exceptions
) {
497 ex
= container_of(this, struct dev_exception_item
, list
);
498 if (!parent_has_perm(devcg
, ex
))
499 dev_exception_rm(devcg
, ex
);
504 * propagate_exception - propagates a new exception to the children
505 * @devcg_root: device cgroup that added a new exception
506 * @ex: new exception to be propagated
508 * returns: 0 in case of success, != 0 in case of error
510 static int propagate_exception(struct dev_cgroup
*devcg_root
,
511 struct dev_exception_item
*ex
)
513 struct cgroup_subsys_state
*pos
;
518 css_for_each_descendant_pre(pos
, &devcg_root
->css
) {
519 struct dev_cgroup
*devcg
= css_to_devcgroup(pos
);
522 * Because devcgroup_mutex is held, no devcg will become
523 * online or offline during the tree walk (see on/offline
524 * methods), and online ones are safe to access outside RCU
525 * read lock without bumping refcnt.
527 if (pos
== &devcg_root
->css
|| !is_devcg_online(devcg
))
533 * in case both root's behavior and devcg is allow, a new
534 * restriction means adding to the exception list
536 if (devcg_root
->behavior
== DEVCG_DEFAULT_ALLOW
&&
537 devcg
->behavior
== DEVCG_DEFAULT_ALLOW
) {
538 rc
= dev_exception_add(devcg
, ex
);
543 * in the other possible cases:
544 * root's behavior: allow, devcg's: deny
545 * root's behavior: deny, devcg's: deny
546 * the exception will be removed
548 dev_exception_rm(devcg
, ex
);
550 revalidate_active_exceptions(devcg
);
559 static inline bool has_children(struct dev_cgroup
*devcgroup
)
561 struct cgroup
*cgrp
= devcgroup
->css
.cgroup
;
563 return !list_empty(&cgrp
->children
);
567 * Modify the exception list using allow/deny rules.
568 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
569 * so we can give a container CAP_MKNOD to let it create devices but not
570 * modify the exception list.
571 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
572 * us to also grant CAP_SYS_ADMIN to containers without giving away the
573 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
575 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
576 * new access is only allowed if you're in the top-level cgroup, or your
577 * parent cgroup has the access you're asking for.
579 static int devcgroup_update_access(struct dev_cgroup
*devcgroup
,
580 int filetype
, char *buffer
)
583 char temp
[12]; /* 11 + 1 characters needed for a u32 */
585 struct dev_exception_item ex
;
586 struct dev_cgroup
*parent
= css_to_devcgroup(css_parent(&devcgroup
->css
));
588 if (!capable(CAP_SYS_ADMIN
))
591 memset(&ex
, 0, sizeof(ex
));
598 if (has_children(devcgroup
))
601 if (!may_allow_all(parent
))
603 dev_exception_clean(devcgroup
);
604 devcgroup
->behavior
= DEVCG_DEFAULT_ALLOW
;
608 rc
= dev_exceptions_copy(&devcgroup
->exceptions
,
609 &parent
->exceptions
);
614 if (has_children(devcgroup
))
617 dev_exception_clean(devcgroup
);
618 devcgroup
->behavior
= DEVCG_DEFAULT_DENY
;
640 } else if (isdigit(*b
)) {
641 memset(temp
, 0, sizeof(temp
));
642 for (count
= 0; count
< sizeof(temp
) - 1; count
++) {
648 rc
= kstrtou32(temp
, 10, &ex
.major
);
662 } else if (isdigit(*b
)) {
663 memset(temp
, 0, sizeof(temp
));
664 for (count
= 0; count
< sizeof(temp
) - 1; count
++) {
670 rc
= kstrtou32(temp
, 10, &ex
.minor
);
678 for (b
++, count
= 0; count
< 3; count
++, b
++) {
681 ex
.access
|= ACC_READ
;
684 ex
.access
|= ACC_WRITE
;
687 ex
.access
|= ACC_MKNOD
;
700 if (!parent_has_perm(devcgroup
, &ex
))
703 * If the default policy is to allow by default, try to remove
704 * an matching exception instead. And be silent about it: we
705 * don't want to break compatibility
707 if (devcgroup
->behavior
== DEVCG_DEFAULT_ALLOW
) {
708 dev_exception_rm(devcgroup
, &ex
);
711 rc
= dev_exception_add(devcgroup
, &ex
);
715 * If the default policy is to deny by default, try to remove
716 * an matching exception instead. And be silent about it: we
717 * don't want to break compatibility
719 if (devcgroup
->behavior
== DEVCG_DEFAULT_DENY
)
720 dev_exception_rm(devcgroup
, &ex
);
722 rc
= dev_exception_add(devcgroup
, &ex
);
726 /* we only propagate new restrictions */
727 rc
= propagate_exception(devcgroup
, &ex
);
735 static int devcgroup_access_write(struct cgroup_subsys_state
*css
,
736 struct cftype
*cft
, char *buffer
)
740 mutex_lock(&devcgroup_mutex
);
741 retval
= devcgroup_update_access(css_to_devcgroup(css
),
742 cft
->private, buffer
);
743 mutex_unlock(&devcgroup_mutex
);
747 static struct cftype dev_cgroup_files
[] = {
750 .write_string
= devcgroup_access_write
,
751 .private = DEVCG_ALLOW
,
755 .write_string
= devcgroup_access_write
,
756 .private = DEVCG_DENY
,
760 .seq_show
= devcgroup_seq_show
,
761 .private = DEVCG_LIST
,
766 struct cgroup_subsys devices_cgrp_subsys
= {
767 .css_alloc
= devcgroup_css_alloc
,
768 .css_free
= devcgroup_css_free
,
769 .css_online
= devcgroup_online
,
770 .css_offline
= devcgroup_offline
,
771 .base_cftypes
= dev_cgroup_files
,
775 * __devcgroup_check_permission - checks if an inode operation is permitted
776 * @dev_cgroup: the dev cgroup to be tested against
778 * @major: device major number
779 * @minor: device minor number
780 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
782 * returns 0 on success, -EPERM case the operation is not permitted
784 static int __devcgroup_check_permission(short type
, u32 major
, u32 minor
,
787 struct dev_cgroup
*dev_cgroup
;
791 dev_cgroup
= task_devcgroup(current
);
792 if (dev_cgroup
->behavior
== DEVCG_DEFAULT_ALLOW
)
793 /* Can't match any of the exceptions, even partially */
794 rc
= !match_exception_partial(&dev_cgroup
->exceptions
,
795 type
, major
, minor
, access
);
797 /* Need to match completely one exception to be allowed */
798 rc
= match_exception(&dev_cgroup
->exceptions
, type
, major
,
808 int __devcgroup_inode_permission(struct inode
*inode
, int mask
)
810 short type
, access
= 0;
812 if (S_ISBLK(inode
->i_mode
))
814 if (S_ISCHR(inode
->i_mode
))
816 if (mask
& MAY_WRITE
)
821 return __devcgroup_check_permission(type
, imajor(inode
), iminor(inode
),
825 int devcgroup_inode_mknod(int mode
, dev_t dev
)
829 if (!S_ISBLK(mode
) && !S_ISCHR(mode
))
837 return __devcgroup_check_permission(type
, MAJOR(dev
), MINOR(dev
),