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 container_of(s, struct dev_cgroup, css);
59 static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup)
61 return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id));
64 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
66 return css_to_devcgroup(task_subsys_state(task, devices_subsys_id));
69 struct cgroup_subsys devices_subsys;
71 static int devcgroup_can_attach(struct cgroup *new_cgrp,
72 struct cgroup_taskset *set)
74 struct task_struct *task = cgroup_taskset_first(set);
76 if (current != task && !capable(CAP_SYS_ADMIN))
82 * called under devcgroup_mutex
84 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
86 struct dev_exception_item *ex, *tmp, *new;
88 lockdep_assert_held(&devcgroup_mutex);
90 list_for_each_entry(ex, orig, list) {
91 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
94 list_add_tail(&new->list, dest);
100 list_for_each_entry_safe(ex, tmp, dest, list) {
108 * called under devcgroup_mutex
110 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
111 struct dev_exception_item *ex)
113 struct dev_exception_item *excopy, *walk;
115 lockdep_assert_held(&devcgroup_mutex);
117 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
121 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
122 if (walk->type != ex->type)
124 if (walk->major != ex->major)
126 if (walk->minor != ex->minor)
129 walk->access |= ex->access;
135 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
140 * called under devcgroup_mutex
142 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
143 struct dev_exception_item *ex)
145 struct dev_exception_item *walk, *tmp;
147 lockdep_assert_held(&devcgroup_mutex);
149 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
150 if (walk->type != ex->type)
152 if (walk->major != ex->major)
154 if (walk->minor != ex->minor)
157 walk->access &= ~ex->access;
159 list_del_rcu(&walk->list);
160 kfree_rcu(walk, rcu);
165 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
167 struct dev_exception_item *ex, *tmp;
169 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
170 list_del_rcu(&ex->list);
176 * dev_exception_clean - frees all entries of the exception list
177 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
179 * called under devcgroup_mutex
181 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
183 lockdep_assert_held(&devcgroup_mutex);
185 __dev_exception_clean(dev_cgroup);
188 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
190 return (devcg->behavior != DEVCG_DEFAULT_NONE);
194 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
196 * @cgroup: cgroup getting online
197 * returns 0 in case of success, error code otherwise
199 static int devcgroup_online(struct cgroup *cgroup)
201 struct dev_cgroup *dev_cgroup, *parent_dev_cgroup = NULL;
204 mutex_lock(&devcgroup_mutex);
205 dev_cgroup = cgroup_to_devcgroup(cgroup);
207 parent_dev_cgroup = cgroup_to_devcgroup(cgroup->parent);
209 if (parent_dev_cgroup == NULL)
210 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
212 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
213 &parent_dev_cgroup->exceptions);
215 dev_cgroup->behavior = parent_dev_cgroup->behavior;
217 mutex_unlock(&devcgroup_mutex);
222 static void devcgroup_offline(struct cgroup *cgroup)
224 struct dev_cgroup *dev_cgroup = cgroup_to_devcgroup(cgroup);
226 mutex_lock(&devcgroup_mutex);
227 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
228 mutex_unlock(&devcgroup_mutex);
232 * called from kernel/cgroup.c with cgroup_lock() held.
234 static struct cgroup_subsys_state *devcgroup_css_alloc(struct cgroup *cgroup)
236 struct dev_cgroup *dev_cgroup;
238 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
240 return ERR_PTR(-ENOMEM);
241 INIT_LIST_HEAD(&dev_cgroup->exceptions);
242 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
244 return &dev_cgroup->css;
247 static void devcgroup_css_free(struct cgroup *cgroup)
249 struct dev_cgroup *dev_cgroup;
251 dev_cgroup = cgroup_to_devcgroup(cgroup);
252 __dev_exception_clean(dev_cgroup);
256 #define DEVCG_ALLOW 1
263 static void set_access(char *acc, short access)
266 memset(acc, 0, ACCLEN);
267 if (access & ACC_READ)
269 if (access & ACC_WRITE)
271 if (access & ACC_MKNOD)
275 static char type_to_char(short type)
279 if (type == DEV_CHAR)
281 if (type == DEV_BLOCK)
286 static void set_majmin(char *str, unsigned m)
291 sprintf(str, "%u", m);
294 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
297 struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
298 struct dev_exception_item *ex;
299 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
303 * To preserve the compatibility:
304 * - Only show the "all devices" when the default policy is to allow
305 * - List the exceptions in case the default policy is to deny
306 * This way, the file remains as a "whitelist of devices"
308 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
309 set_access(acc, ACC_MASK);
312 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
315 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
316 set_access(acc, ex->access);
317 set_majmin(maj, ex->major);
318 set_majmin(min, ex->minor);
319 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
329 * may_access - verifies if a new exception is part of what is allowed
330 * by a dev cgroup based on the default policy +
331 * exceptions. This is used to make sure a child cgroup
332 * won't have more privileges than its parent or to
333 * verify if a certain access is allowed.
334 * @dev_cgroup: dev cgroup to be tested against
335 * @refex: new exception
336 * @behavior: behavior of the exception
338 static bool may_access(struct dev_cgroup *dev_cgroup,
339 struct dev_exception_item *refex,
340 enum devcg_behavior behavior)
342 struct dev_exception_item *ex;
345 rcu_lockdep_assert(rcu_read_lock_held() ||
346 lockdep_is_held(&devcgroup_mutex),
347 "device_cgroup::may_access() called without proper synchronization");
349 list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
350 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
352 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
354 if (ex->major != ~0 && ex->major != refex->major)
356 if (ex->minor != ~0 && ex->minor != refex->minor)
358 if (refex->access & (~ex->access))
364 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
365 if (behavior == DEVCG_DEFAULT_ALLOW) {
366 /* the exception will deny access to certain devices */
369 /* the exception will allow access to certain devices */
372 * a new exception allowing access shouldn't
373 * match an parent's exception
379 /* only behavior == DEVCG_DEFAULT_DENY allowed here */
381 /* parent has an exception that matches the proposed */
391 * when adding a new allow rule to a device exception list, the rule
392 * must be allowed in the parent device
394 static int parent_has_perm(struct dev_cgroup *childcg,
395 struct dev_exception_item *ex)
397 struct cgroup *pcg = childcg->css.cgroup->parent;
398 struct dev_cgroup *parent;
402 parent = cgroup_to_devcgroup(pcg);
403 return may_access(parent, ex, childcg->behavior);
407 * may_allow_all - checks if it's possible to change the behavior to
408 * allow based on parent's rules.
409 * @parent: device cgroup's parent
410 * returns: != 0 in case it's allowed, 0 otherwise
412 static inline int may_allow_all(struct dev_cgroup *parent)
416 return parent->behavior == DEVCG_DEFAULT_ALLOW;
420 * revalidate_active_exceptions - walks through the active exception list and
421 * revalidates the exceptions based on parent's
422 * behavior and exceptions. The exceptions that
423 * are no longer valid will be removed.
424 * Called with devcgroup_mutex held.
425 * @devcg: cgroup which exceptions will be checked
427 * This is one of the three key functions for hierarchy implementation.
428 * This function is responsible for re-evaluating all the cgroup's active
429 * exceptions due to a parent's exception change.
430 * Refer to Documentation/cgroups/devices.txt for more details.
432 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
434 struct dev_exception_item *ex;
435 struct list_head *this, *tmp;
437 list_for_each_safe(this, tmp, &devcg->exceptions) {
438 ex = container_of(this, struct dev_exception_item, list);
439 if (!parent_has_perm(devcg, ex))
440 dev_exception_rm(devcg, ex);
445 * propagate_exception - propagates a new exception to the children
446 * @devcg_root: device cgroup that added a new exception
447 * @ex: new exception to be propagated
449 * returns: 0 in case of success, != 0 in case of error
451 static int propagate_exception(struct dev_cgroup *devcg_root,
452 struct dev_exception_item *ex)
454 struct cgroup *root = devcg_root->css.cgroup, *pos;
459 cgroup_for_each_descendant_pre(pos, root) {
460 struct dev_cgroup *devcg = cgroup_to_devcgroup(pos);
463 * Because devcgroup_mutex is held, no devcg will become
464 * online or offline during the tree walk (see on/offline
465 * methods), and online ones are safe to access outside RCU
466 * read lock without bumping refcnt.
468 if (!is_devcg_online(devcg))
474 * in case both root's behavior and devcg is allow, a new
475 * restriction means adding to the exception list
477 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
478 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
479 rc = dev_exception_add(devcg, ex);
484 * in the other possible cases:
485 * root's behavior: allow, devcg's: deny
486 * root's behavior: deny, devcg's: deny
487 * the exception will be removed
489 dev_exception_rm(devcg, ex);
491 revalidate_active_exceptions(devcg);
500 static inline bool has_children(struct dev_cgroup *devcgroup)
502 struct cgroup *cgrp = devcgroup->css.cgroup;
504 return !list_empty(&cgrp->children);
508 * Modify the exception list using allow/deny rules.
509 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
510 * so we can give a container CAP_MKNOD to let it create devices but not
511 * modify the exception list.
512 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
513 * us to also grant CAP_SYS_ADMIN to containers without giving away the
514 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
516 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
517 * new access is only allowed if you're in the top-level cgroup, or your
518 * parent cgroup has the access you're asking for.
520 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
521 int filetype, const char *buffer)
524 char temp[12]; /* 11 + 1 characters needed for a u32 */
526 struct dev_exception_item ex;
527 struct cgroup *p = devcgroup->css.cgroup;
528 struct dev_cgroup *parent = NULL;
530 if (!capable(CAP_SYS_ADMIN))
534 parent = cgroup_to_devcgroup(p->parent);
536 memset(&ex, 0, sizeof(ex));
543 if (has_children(devcgroup))
546 if (!may_allow_all(parent))
548 dev_exception_clean(devcgroup);
549 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
553 rc = dev_exceptions_copy(&devcgroup->exceptions,
554 &parent->exceptions);
559 if (has_children(devcgroup))
562 dev_exception_clean(devcgroup);
563 devcgroup->behavior = DEVCG_DEFAULT_DENY;
585 } else if (isdigit(*b)) {
586 memset(temp, 0, sizeof(temp));
587 for (count = 0; count < sizeof(temp) - 1; count++) {
593 rc = kstrtou32(temp, 10, &ex.major);
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.minor);
623 for (b++, count = 0; count < 3; count++, b++) {
626 ex.access |= ACC_READ;
629 ex.access |= ACC_WRITE;
632 ex.access |= ACC_MKNOD;
645 if (!parent_has_perm(devcgroup, &ex))
648 * If the default policy is to allow by default, try to remove
649 * an matching exception instead. And be silent about it: we
650 * don't want to break compatibility
652 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
653 dev_exception_rm(devcgroup, &ex);
656 rc = dev_exception_add(devcgroup, &ex);
660 * If the default policy is to deny by default, try to remove
661 * an matching exception instead. And be silent about it: we
662 * don't want to break compatibility
664 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
665 dev_exception_rm(devcgroup, &ex);
667 rc = dev_exception_add(devcgroup, &ex);
671 /* we only propagate new restrictions */
672 rc = propagate_exception(devcgroup, &ex);
680 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
685 mutex_lock(&devcgroup_mutex);
686 retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
687 cft->private, buffer);
688 mutex_unlock(&devcgroup_mutex);
692 static struct cftype dev_cgroup_files[] = {
695 .write_string = devcgroup_access_write,
696 .private = DEVCG_ALLOW,
700 .write_string = devcgroup_access_write,
701 .private = DEVCG_DENY,
705 .read_seq_string = devcgroup_seq_read,
706 .private = DEVCG_LIST,
711 struct cgroup_subsys devices_subsys = {
713 .can_attach = devcgroup_can_attach,
714 .css_alloc = devcgroup_css_alloc,
715 .css_free = devcgroup_css_free,
716 .css_online = devcgroup_online,
717 .css_offline = devcgroup_offline,
718 .subsys_id = devices_subsys_id,
719 .base_cftypes = dev_cgroup_files,
723 * __devcgroup_check_permission - checks if an inode operation is permitted
724 * @dev_cgroup: the dev cgroup to be tested against
726 * @major: device major number
727 * @minor: device minor number
728 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
730 * returns 0 on success, -EPERM case the operation is not permitted
732 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
735 struct dev_cgroup *dev_cgroup;
736 struct dev_exception_item ex;
739 memset(&ex, 0, sizeof(ex));
746 dev_cgroup = task_devcgroup(current);
747 rc = may_access(dev_cgroup, &ex, dev_cgroup->behavior);
756 int __devcgroup_inode_permission(struct inode *inode, int mask)
758 short type, access = 0;
760 if (S_ISBLK(inode->i_mode))
762 if (S_ISCHR(inode->i_mode))
764 if (mask & MAY_WRITE)
769 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
773 int devcgroup_inode_mknod(int mode, dev_t dev)
777 if (!S_ISBLK(mode) && !S_ISCHR(mode))
785 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),