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;
52 /* temporary list for pending propagation operations */
53 struct list_head propagate_pending;
56 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
58 return container_of(s, struct dev_cgroup, css);
61 static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup)
63 return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id));
66 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
68 return css_to_devcgroup(task_subsys_state(task, devices_subsys_id));
71 struct cgroup_subsys devices_subsys;
73 static int devcgroup_can_attach(struct cgroup *new_cgrp,
74 struct cgroup_taskset *set)
76 struct task_struct *task = cgroup_taskset_first(set);
78 if (current != task && !capable(CAP_SYS_ADMIN))
84 * called under devcgroup_mutex
86 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
88 struct dev_exception_item *ex, *tmp, *new;
90 lockdep_assert_held(&devcgroup_mutex);
92 list_for_each_entry(ex, orig, list) {
93 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
96 list_add_tail(&new->list, dest);
102 list_for_each_entry_safe(ex, tmp, dest, list) {
110 * called under devcgroup_mutex
112 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
113 struct dev_exception_item *ex)
115 struct dev_exception_item *excopy, *walk;
117 lockdep_assert_held(&devcgroup_mutex);
119 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
123 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
124 if (walk->type != ex->type)
126 if (walk->major != ex->major)
128 if (walk->minor != ex->minor)
131 walk->access |= ex->access;
137 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
142 * called under devcgroup_mutex
144 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
145 struct dev_exception_item *ex)
147 struct dev_exception_item *walk, *tmp;
149 lockdep_assert_held(&devcgroup_mutex);
151 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
152 if (walk->type != ex->type)
154 if (walk->major != ex->major)
156 if (walk->minor != ex->minor)
159 walk->access &= ~ex->access;
161 list_del_rcu(&walk->list);
162 kfree_rcu(walk, rcu);
167 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
169 struct dev_exception_item *ex, *tmp;
171 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
172 list_del_rcu(&ex->list);
178 * dev_exception_clean - frees all entries of the exception list
179 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
181 * called under devcgroup_mutex
183 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
185 lockdep_assert_held(&devcgroup_mutex);
187 __dev_exception_clean(dev_cgroup);
190 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
192 return (devcg->behavior != DEVCG_DEFAULT_NONE);
196 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
198 * @cgroup: cgroup getting online
199 * returns 0 in case of success, error code otherwise
201 static int devcgroup_online(struct cgroup *cgroup)
203 struct dev_cgroup *dev_cgroup, *parent_dev_cgroup = NULL;
206 mutex_lock(&devcgroup_mutex);
207 dev_cgroup = cgroup_to_devcgroup(cgroup);
209 parent_dev_cgroup = cgroup_to_devcgroup(cgroup->parent);
211 if (parent_dev_cgroup == NULL)
212 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
214 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
215 &parent_dev_cgroup->exceptions);
217 dev_cgroup->behavior = parent_dev_cgroup->behavior;
219 mutex_unlock(&devcgroup_mutex);
224 static void devcgroup_offline(struct cgroup *cgroup)
226 struct dev_cgroup *dev_cgroup = cgroup_to_devcgroup(cgroup);
228 mutex_lock(&devcgroup_mutex);
229 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
230 mutex_unlock(&devcgroup_mutex);
234 * called from kernel/cgroup.c with cgroup_lock() held.
236 static struct cgroup_subsys_state *devcgroup_css_alloc(struct cgroup *cgroup)
238 struct dev_cgroup *dev_cgroup;
239 struct cgroup *parent_cgroup;
241 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
243 return ERR_PTR(-ENOMEM);
244 INIT_LIST_HEAD(&dev_cgroup->exceptions);
245 INIT_LIST_HEAD(&dev_cgroup->propagate_pending);
246 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
247 parent_cgroup = cgroup->parent;
249 return &dev_cgroup->css;
252 static void devcgroup_css_free(struct cgroup *cgroup)
254 struct dev_cgroup *dev_cgroup;
256 dev_cgroup = cgroup_to_devcgroup(cgroup);
257 __dev_exception_clean(dev_cgroup);
261 #define DEVCG_ALLOW 1
268 static void set_access(char *acc, short access)
271 memset(acc, 0, ACCLEN);
272 if (access & ACC_READ)
274 if (access & ACC_WRITE)
276 if (access & ACC_MKNOD)
280 static char type_to_char(short type)
284 if (type == DEV_CHAR)
286 if (type == DEV_BLOCK)
291 static void set_majmin(char *str, unsigned m)
296 sprintf(str, "%u", m);
299 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
302 struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
303 struct dev_exception_item *ex;
304 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
308 * To preserve the compatibility:
309 * - Only show the "all devices" when the default policy is to allow
310 * - List the exceptions in case the default policy is to deny
311 * This way, the file remains as a "whitelist of devices"
313 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
314 set_access(acc, ACC_MASK);
317 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
320 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
321 set_access(acc, ex->access);
322 set_majmin(maj, ex->major);
323 set_majmin(min, ex->minor);
324 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
334 * may_access - verifies if a new exception is part of what is allowed
335 * by a dev cgroup based on the default policy +
336 * exceptions. This is used to make sure a child cgroup
337 * won't have more privileges than its parent or to
338 * verify if a certain access is allowed.
339 * @dev_cgroup: dev cgroup to be tested against
340 * @refex: new exception
341 * @behavior: behavior of the exception
343 static bool may_access(struct dev_cgroup *dev_cgroup,
344 struct dev_exception_item *refex,
345 enum devcg_behavior behavior)
347 struct dev_exception_item *ex;
350 rcu_lockdep_assert(rcu_read_lock_held() ||
351 lockdep_is_held(&devcgroup_mutex),
352 "device_cgroup::may_access() called without proper synchronization");
354 list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
355 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
357 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
359 if (ex->major != ~0 && ex->major != refex->major)
361 if (ex->minor != ~0 && ex->minor != refex->minor)
363 if (refex->access & (~ex->access))
369 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
370 if (behavior == DEVCG_DEFAULT_ALLOW) {
371 /* the exception will deny access to certain devices */
374 /* the exception will allow access to certain devices */
377 * a new exception allowing access shouldn't
378 * match an parent's exception
384 /* only behavior == DEVCG_DEFAULT_DENY allowed here */
386 /* parent has an exception that matches the proposed */
396 * when adding a new allow rule to a device exception list, the rule
397 * must be allowed in the parent device
399 static int parent_has_perm(struct dev_cgroup *childcg,
400 struct dev_exception_item *ex)
402 struct cgroup *pcg = childcg->css.cgroup->parent;
403 struct dev_cgroup *parent;
407 parent = cgroup_to_devcgroup(pcg);
408 return may_access(parent, ex, childcg->behavior);
412 * may_allow_all - checks if it's possible to change the behavior to
413 * allow based on parent's rules.
414 * @parent: device cgroup's parent
415 * returns: != 0 in case it's allowed, 0 otherwise
417 static inline int may_allow_all(struct dev_cgroup *parent)
421 return parent->behavior == DEVCG_DEFAULT_ALLOW;
425 * revalidate_active_exceptions - walks through the active exception list and
426 * revalidates the exceptions based on parent's
427 * behavior and exceptions. The exceptions that
428 * are no longer valid will be removed.
429 * Called with devcgroup_mutex held.
430 * @devcg: cgroup which exceptions will be checked
432 * This is one of the three key functions for hierarchy implementation.
433 * This function is responsible for re-evaluating all the cgroup's active
434 * exceptions due to a parent's exception change.
435 * Refer to Documentation/cgroups/devices.txt for more details.
437 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
439 struct dev_exception_item *ex;
440 struct list_head *this, *tmp;
442 list_for_each_safe(this, tmp, &devcg->exceptions) {
443 ex = container_of(this, struct dev_exception_item, list);
444 if (!parent_has_perm(devcg, ex))
445 dev_exception_rm(devcg, ex);
450 * get_online_devcg - walks the cgroup tree and fills a list with the online
452 * @root: cgroup used as starting point
453 * @online: list that will be filled with online groups
455 * Must be called with devcgroup_mutex held. Grabs RCU lock.
456 * Because devcgroup_mutex is held, no devcg will become online or offline
457 * during the tree walk (see devcgroup_online, devcgroup_offline)
458 * A separated list is needed because propagate_behavior() and
459 * propagate_exception() need to allocate memory and can block.
461 static void get_online_devcg(struct cgroup *root, struct list_head *online)
464 struct dev_cgroup *devcg;
466 lockdep_assert_held(&devcgroup_mutex);
469 cgroup_for_each_descendant_pre(pos, root) {
470 devcg = cgroup_to_devcgroup(pos);
471 if (is_devcg_online(devcg))
472 list_add_tail(&devcg->propagate_pending, online);
478 * propagate_exception - propagates a new exception to the children
479 * @devcg_root: device cgroup that added a new exception
480 * @ex: new exception to be propagated
482 * returns: 0 in case of success, != 0 in case of error
484 static int propagate_exception(struct dev_cgroup *devcg_root,
485 struct dev_exception_item *ex)
487 struct cgroup *root = devcg_root->css.cgroup;
488 struct dev_cgroup *devcg, *parent, *tmp;
492 get_online_devcg(root, &pending);
494 list_for_each_entry_safe(devcg, tmp, &pending, propagate_pending) {
495 parent = cgroup_to_devcgroup(devcg->css.cgroup->parent);
498 * in case both root's behavior and devcg is allow, a new
499 * restriction means adding to the exception list
501 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
502 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
503 rc = dev_exception_add(devcg, ex);
508 * in the other possible cases:
509 * root's behavior: allow, devcg's: deny
510 * root's behavior: deny, devcg's: deny
511 * the exception will be removed
513 dev_exception_rm(devcg, ex);
515 revalidate_active_exceptions(devcg);
517 list_del_init(&devcg->propagate_pending);
522 static inline bool has_children(struct dev_cgroup *devcgroup)
524 struct cgroup *cgrp = devcgroup->css.cgroup;
526 return !list_empty(&cgrp->children);
530 * Modify the exception list using allow/deny rules.
531 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
532 * so we can give a container CAP_MKNOD to let it create devices but not
533 * modify the exception list.
534 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
535 * us to also grant CAP_SYS_ADMIN to containers without giving away the
536 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
538 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
539 * new access is only allowed if you're in the top-level cgroup, or your
540 * parent cgroup has the access you're asking for.
542 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
543 int filetype, const char *buffer)
546 char temp[12]; /* 11 + 1 characters needed for a u32 */
548 struct dev_exception_item ex;
549 struct cgroup *p = devcgroup->css.cgroup;
550 struct dev_cgroup *parent = NULL;
552 if (!capable(CAP_SYS_ADMIN))
556 parent = cgroup_to_devcgroup(p->parent);
558 memset(&ex, 0, sizeof(ex));
565 if (has_children(devcgroup))
568 if (!may_allow_all(parent))
570 dev_exception_clean(devcgroup);
571 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
575 rc = dev_exceptions_copy(&devcgroup->exceptions,
576 &parent->exceptions);
581 if (has_children(devcgroup))
584 dev_exception_clean(devcgroup);
585 devcgroup->behavior = DEVCG_DEFAULT_DENY;
607 } else if (isdigit(*b)) {
608 memset(temp, 0, sizeof(temp));
609 for (count = 0; count < sizeof(temp) - 1; count++) {
615 rc = kstrtou32(temp, 10, &ex.major);
629 } else if (isdigit(*b)) {
630 memset(temp, 0, sizeof(temp));
631 for (count = 0; count < sizeof(temp) - 1; count++) {
637 rc = kstrtou32(temp, 10, &ex.minor);
645 for (b++, count = 0; count < 3; count++, b++) {
648 ex.access |= ACC_READ;
651 ex.access |= ACC_WRITE;
654 ex.access |= ACC_MKNOD;
667 if (!parent_has_perm(devcgroup, &ex))
670 * If the default policy is to allow by default, try to remove
671 * an matching exception instead. And be silent about it: we
672 * don't want to break compatibility
674 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
675 dev_exception_rm(devcgroup, &ex);
678 rc = dev_exception_add(devcgroup, &ex);
682 * If the default policy is to deny by default, try to remove
683 * an matching exception instead. And be silent about it: we
684 * don't want to break compatibility
686 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
687 dev_exception_rm(devcgroup, &ex);
689 rc = dev_exception_add(devcgroup, &ex);
693 /* we only propagate new restrictions */
694 rc = propagate_exception(devcgroup, &ex);
702 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
707 mutex_lock(&devcgroup_mutex);
708 retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
709 cft->private, buffer);
710 mutex_unlock(&devcgroup_mutex);
714 static struct cftype dev_cgroup_files[] = {
717 .write_string = devcgroup_access_write,
718 .private = DEVCG_ALLOW,
722 .write_string = devcgroup_access_write,
723 .private = DEVCG_DENY,
727 .read_seq_string = devcgroup_seq_read,
728 .private = DEVCG_LIST,
733 struct cgroup_subsys devices_subsys = {
735 .can_attach = devcgroup_can_attach,
736 .css_alloc = devcgroup_css_alloc,
737 .css_free = devcgroup_css_free,
738 .css_online = devcgroup_online,
739 .css_offline = devcgroup_offline,
740 .subsys_id = devices_subsys_id,
741 .base_cftypes = dev_cgroup_files,
744 * While devices cgroup has the rudimentary hierarchy support which
745 * checks the parent's restriction, it doesn't properly propagates
746 * config changes in ancestors to their descendents. A child
747 * should only be allowed to add more restrictions to the parent's
748 * configuration. Fix it and remove the following.
750 .broken_hierarchy = true,
754 * __devcgroup_check_permission - checks if an inode operation is permitted
755 * @dev_cgroup: the dev cgroup to be tested against
757 * @major: device major number
758 * @minor: device minor number
759 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
761 * returns 0 on success, -EPERM case the operation is not permitted
763 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
766 struct dev_cgroup *dev_cgroup;
767 struct dev_exception_item ex;
770 memset(&ex, 0, sizeof(ex));
777 dev_cgroup = task_devcgroup(current);
778 rc = may_access(dev_cgroup, &ex, dev_cgroup->behavior);
787 int __devcgroup_inode_permission(struct inode *inode, int mask)
789 short type, access = 0;
791 if (S_ISBLK(inode->i_mode))
793 if (S_ISCHR(inode->i_mode))
795 if (mask & MAY_WRITE)
800 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
804 int devcgroup_inode_mknod(int mode, dev_t dev)
808 if (!S_ISBLK(mode) && !S_ISCHR(mode))
816 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),