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_subsys_id
));
64 struct cgroup_subsys devices_subsys
;
67 * called under devcgroup_mutex
69 static int dev_exceptions_copy(struct list_head
*dest
, struct list_head
*orig
)
71 struct dev_exception_item
*ex
, *tmp
, *new;
73 lockdep_assert_held(&devcgroup_mutex
);
75 list_for_each_entry(ex
, orig
, list
) {
76 new = kmemdup(ex
, sizeof(*ex
), GFP_KERNEL
);
79 list_add_tail(&new->list
, dest
);
85 list_for_each_entry_safe(ex
, tmp
, dest
, list
) {
93 * called under devcgroup_mutex
95 static int dev_exception_add(struct dev_cgroup
*dev_cgroup
,
96 struct dev_exception_item
*ex
)
98 struct dev_exception_item
*excopy
, *walk
;
100 lockdep_assert_held(&devcgroup_mutex
);
102 excopy
= kmemdup(ex
, sizeof(*ex
), GFP_KERNEL
);
106 list_for_each_entry(walk
, &dev_cgroup
->exceptions
, list
) {
107 if (walk
->type
!= ex
->type
)
109 if (walk
->major
!= ex
->major
)
111 if (walk
->minor
!= ex
->minor
)
114 walk
->access
|= ex
->access
;
120 list_add_tail_rcu(&excopy
->list
, &dev_cgroup
->exceptions
);
125 * called under devcgroup_mutex
127 static void dev_exception_rm(struct dev_cgroup
*dev_cgroup
,
128 struct dev_exception_item
*ex
)
130 struct dev_exception_item
*walk
, *tmp
;
132 lockdep_assert_held(&devcgroup_mutex
);
134 list_for_each_entry_safe(walk
, tmp
, &dev_cgroup
->exceptions
, list
) {
135 if (walk
->type
!= ex
->type
)
137 if (walk
->major
!= ex
->major
)
139 if (walk
->minor
!= ex
->minor
)
142 walk
->access
&= ~ex
->access
;
144 list_del_rcu(&walk
->list
);
145 kfree_rcu(walk
, rcu
);
150 static void __dev_exception_clean(struct dev_cgroup
*dev_cgroup
)
152 struct dev_exception_item
*ex
, *tmp
;
154 list_for_each_entry_safe(ex
, tmp
, &dev_cgroup
->exceptions
, list
) {
155 list_del_rcu(&ex
->list
);
161 * dev_exception_clean - frees all entries of the exception list
162 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
164 * called under devcgroup_mutex
166 static void dev_exception_clean(struct dev_cgroup
*dev_cgroup
)
168 lockdep_assert_held(&devcgroup_mutex
);
170 __dev_exception_clean(dev_cgroup
);
173 static inline bool is_devcg_online(const struct dev_cgroup
*devcg
)
175 return (devcg
->behavior
!= DEVCG_DEFAULT_NONE
);
179 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
181 * @css: css getting online
182 * returns 0 in case of success, error code otherwise
184 static int devcgroup_online(struct cgroup_subsys_state
*css
)
186 struct dev_cgroup
*dev_cgroup
= css_to_devcgroup(css
);
187 struct dev_cgroup
*parent_dev_cgroup
= css_to_devcgroup(css_parent(css
));
190 mutex_lock(&devcgroup_mutex
);
192 if (parent_dev_cgroup
== NULL
)
193 dev_cgroup
->behavior
= DEVCG_DEFAULT_ALLOW
;
195 ret
= dev_exceptions_copy(&dev_cgroup
->exceptions
,
196 &parent_dev_cgroup
->exceptions
);
198 dev_cgroup
->behavior
= parent_dev_cgroup
->behavior
;
200 mutex_unlock(&devcgroup_mutex
);
205 static void devcgroup_offline(struct cgroup_subsys_state
*css
)
207 struct dev_cgroup
*dev_cgroup
= css_to_devcgroup(css
);
209 mutex_lock(&devcgroup_mutex
);
210 dev_cgroup
->behavior
= DEVCG_DEFAULT_NONE
;
211 mutex_unlock(&devcgroup_mutex
);
215 * called from kernel/cgroup.c with cgroup_lock() held.
217 static struct cgroup_subsys_state
*
218 devcgroup_css_alloc(struct cgroup_subsys_state
*parent_css
)
220 struct dev_cgroup
*dev_cgroup
;
222 dev_cgroup
= kzalloc(sizeof(*dev_cgroup
), GFP_KERNEL
);
224 return ERR_PTR(-ENOMEM
);
225 INIT_LIST_HEAD(&dev_cgroup
->exceptions
);
226 dev_cgroup
->behavior
= DEVCG_DEFAULT_NONE
;
228 return &dev_cgroup
->css
;
231 static void devcgroup_css_free(struct cgroup_subsys_state
*css
)
233 struct dev_cgroup
*dev_cgroup
= css_to_devcgroup(css
);
235 __dev_exception_clean(dev_cgroup
);
239 #define DEVCG_ALLOW 1
246 static void set_access(char *acc
, short access
)
249 memset(acc
, 0, ACCLEN
);
250 if (access
& ACC_READ
)
252 if (access
& ACC_WRITE
)
254 if (access
& ACC_MKNOD
)
258 static char type_to_char(short type
)
262 if (type
== DEV_CHAR
)
264 if (type
== DEV_BLOCK
)
269 static void set_majmin(char *str
, unsigned m
)
274 sprintf(str
, "%u", m
);
277 static int devcgroup_seq_show(struct seq_file
*m
, void *v
)
279 struct dev_cgroup
*devcgroup
= css_to_devcgroup(seq_css(m
));
280 struct dev_exception_item
*ex
;
281 char maj
[MAJMINLEN
], min
[MAJMINLEN
], acc
[ACCLEN
];
285 * To preserve the compatibility:
286 * - Only show the "all devices" when the default policy is to allow
287 * - List the exceptions in case the default policy is to deny
288 * This way, the file remains as a "whitelist of devices"
290 if (devcgroup
->behavior
== DEVCG_DEFAULT_ALLOW
) {
291 set_access(acc
, ACC_MASK
);
294 seq_printf(m
, "%c %s:%s %s\n", type_to_char(DEV_ALL
),
297 list_for_each_entry_rcu(ex
, &devcgroup
->exceptions
, list
) {
298 set_access(acc
, ex
->access
);
299 set_majmin(maj
, ex
->major
);
300 set_majmin(min
, ex
->minor
);
301 seq_printf(m
, "%c %s:%s %s\n", type_to_char(ex
->type
),
311 * match_exception - iterates the exception list trying to match a rule
312 * based on type, major, minor and access type. It is
313 * considered a match if an exception is found that
314 * will contain the entire range of provided parameters.
315 * @exceptions: list of exceptions
316 * @type: device type (DEV_BLOCK or 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 (ACC_READ, ACC_WRITE, ACC_MKNOD)
321 * returns: true in case it matches an exception completely
323 static bool match_exception(struct list_head
*exceptions
, short type
,
324 u32 major
, u32 minor
, short access
)
326 struct dev_exception_item
*ex
;
328 list_for_each_entry_rcu(ex
, exceptions
, list
) {
329 if ((type
& DEV_BLOCK
) && !(ex
->type
& DEV_BLOCK
))
331 if ((type
& DEV_CHAR
) && !(ex
->type
& DEV_CHAR
))
333 if (ex
->major
!= ~0 && ex
->major
!= major
)
335 if (ex
->minor
!= ~0 && ex
->minor
!= minor
)
337 /* provided access cannot have more than the exception rule */
338 if (access
& (~ex
->access
))
346 * match_exception_partial - iterates the exception list trying to match a rule
347 * based on type, major, minor and access type. It is
348 * considered a match if an exception's range is
349 * found to contain *any* of the devices specified by
350 * provided parameters. This is used to make sure no
351 * extra access is being granted that is forbidden by
352 * any of the exception list.
353 * @exceptions: list of exceptions
354 * @type: device type (DEV_BLOCK or DEV_CHAR)
355 * @major: device file major number, ~0 to match all
356 * @minor: device file minor number, ~0 to match all
357 * @access: permission mask (ACC_READ, ACC_WRITE, ACC_MKNOD)
359 * returns: 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
& DEV_BLOCK
) && !(ex
->type
& DEV_BLOCK
))
369 if ((type
& DEV_CHAR
) && !(ex
->type
& 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 part of what is allowed
393 * by a dev cgroup based on the default policy +
394 * exceptions. This is used to make sure a child cgroup
395 * won't have more privileges than its parent
396 * @dev_cgroup: dev cgroup to be tested against
397 * @refex: new exception
398 * @behavior: behavior of the exception's dev_cgroup
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_assert(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(css_parent(&childcg
->css
));
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(css_parent(&childcg
->css
));
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/cgroups/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
);
593 static inline bool has_children(struct dev_cgroup
*devcgroup
)
595 struct cgroup
*cgrp
= devcgroup
->css
.cgroup
;
597 return !list_empty(&cgrp
->children
);
601 * Modify the exception list using allow/deny rules.
602 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
603 * so we can give a container CAP_MKNOD to let it create devices but not
604 * modify the exception list.
605 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
606 * us to also grant CAP_SYS_ADMIN to containers without giving away the
607 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
609 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
610 * new access is only allowed if you're in the top-level cgroup, or your
611 * parent cgroup has the access you're asking for.
613 static int devcgroup_update_access(struct dev_cgroup
*devcgroup
,
614 int filetype
, const char *buffer
)
617 char temp
[12]; /* 11 + 1 characters needed for a u32 */
619 struct dev_exception_item ex
;
620 struct dev_cgroup
*parent
= css_to_devcgroup(css_parent(&devcgroup
->css
));
622 if (!capable(CAP_SYS_ADMIN
))
625 memset(&ex
, 0, sizeof(ex
));
632 if (has_children(devcgroup
))
635 if (!may_allow_all(parent
))
637 dev_exception_clean(devcgroup
);
638 devcgroup
->behavior
= DEVCG_DEFAULT_ALLOW
;
642 rc
= dev_exceptions_copy(&devcgroup
->exceptions
,
643 &parent
->exceptions
);
648 if (has_children(devcgroup
))
651 dev_exception_clean(devcgroup
);
652 devcgroup
->behavior
= DEVCG_DEFAULT_DENY
;
674 } else if (isdigit(*b
)) {
675 memset(temp
, 0, sizeof(temp
));
676 for (count
= 0; count
< sizeof(temp
) - 1; count
++) {
682 rc
= kstrtou32(temp
, 10, &ex
.major
);
696 } else if (isdigit(*b
)) {
697 memset(temp
, 0, sizeof(temp
));
698 for (count
= 0; count
< sizeof(temp
) - 1; count
++) {
704 rc
= kstrtou32(temp
, 10, &ex
.minor
);
712 for (b
++, count
= 0; count
< 3; count
++, b
++) {
715 ex
.access
|= ACC_READ
;
718 ex
.access
|= ACC_WRITE
;
721 ex
.access
|= ACC_MKNOD
;
735 * If the default policy is to allow by default, try to remove
736 * an matching exception instead. And be silent about it: we
737 * don't want to break compatibility
739 if (devcgroup
->behavior
== DEVCG_DEFAULT_ALLOW
) {
740 /* Check if the parent allows removing it first */
741 if (!parent_allows_removal(devcgroup
, &ex
))
743 dev_exception_rm(devcgroup
, &ex
);
747 if (!parent_has_perm(devcgroup
, &ex
))
749 rc
= dev_exception_add(devcgroup
, &ex
);
753 * If the default policy is to deny by default, try to remove
754 * an matching exception instead. And be silent about it: we
755 * don't want to break compatibility
757 if (devcgroup
->behavior
== DEVCG_DEFAULT_DENY
)
758 dev_exception_rm(devcgroup
, &ex
);
760 rc
= dev_exception_add(devcgroup
, &ex
);
764 /* we only propagate new restrictions */
765 rc
= propagate_exception(devcgroup
, &ex
);
773 static int devcgroup_access_write(struct cgroup_subsys_state
*css
,
774 struct cftype
*cft
, const char *buffer
)
778 mutex_lock(&devcgroup_mutex
);
779 retval
= devcgroup_update_access(css_to_devcgroup(css
),
780 cft
->private, buffer
);
781 mutex_unlock(&devcgroup_mutex
);
785 static struct cftype dev_cgroup_files
[] = {
788 .write_string
= devcgroup_access_write
,
789 .private = DEVCG_ALLOW
,
793 .write_string
= devcgroup_access_write
,
794 .private = DEVCG_DENY
,
798 .seq_show
= devcgroup_seq_show
,
799 .private = DEVCG_LIST
,
804 struct cgroup_subsys devices_subsys
= {
806 .css_alloc
= devcgroup_css_alloc
,
807 .css_free
= devcgroup_css_free
,
808 .css_online
= devcgroup_online
,
809 .css_offline
= devcgroup_offline
,
810 .subsys_id
= devices_subsys_id
,
811 .base_cftypes
= dev_cgroup_files
,
815 * __devcgroup_check_permission - checks if an inode operation is permitted
816 * @dev_cgroup: the dev cgroup to be tested against
818 * @major: device major number
819 * @minor: device minor number
820 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
822 * returns 0 on success, -EPERM case the operation is not permitted
824 static int __devcgroup_check_permission(short type
, u32 major
, u32 minor
,
827 struct dev_cgroup
*dev_cgroup
;
831 dev_cgroup
= task_devcgroup(current
);
832 if (dev_cgroup
->behavior
== DEVCG_DEFAULT_ALLOW
)
833 /* Can't match any of the exceptions, even partially */
834 rc
= !match_exception_partial(&dev_cgroup
->exceptions
,
835 type
, major
, minor
, access
);
837 /* Need to match completely one exception to be allowed */
838 rc
= match_exception(&dev_cgroup
->exceptions
, type
, major
,
848 int __devcgroup_inode_permission(struct inode
*inode
, int mask
)
850 short type
, access
= 0;
852 if (S_ISBLK(inode
->i_mode
))
854 if (S_ISCHR(inode
->i_mode
))
856 if (mask
& MAY_WRITE
)
861 return __devcgroup_check_permission(type
, imajor(inode
), iminor(inode
),
865 int devcgroup_inode_mknod(int mode
, dev_t dev
)
869 if (!S_ISBLK(mode
) && !S_ISCHR(mode
))
877 return __devcgroup_check_permission(type
, MAJOR(dev
), MINOR(dev
),