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 *cgroup_to_devcgroup(struct cgroup *cgroup)
61 return css_to_devcgroup(cgroup_css(cgroup, devices_subsys_id));
64 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
66 return css_to_devcgroup(task_css(task, devices_subsys_id));
69 struct cgroup_subsys devices_subsys;
71 static int devcgroup_can_attach(struct cgroup_subsys_state *new_css,
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 * @css: css getting online
197 * returns 0 in case of success, error code otherwise
199 static int devcgroup_online(struct cgroup_subsys_state *css)
201 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
202 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css_parent(css));
205 mutex_lock(&devcgroup_mutex);
207 if (parent_dev_cgroup == NULL)
208 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
210 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
211 &parent_dev_cgroup->exceptions);
213 dev_cgroup->behavior = parent_dev_cgroup->behavior;
215 mutex_unlock(&devcgroup_mutex);
220 static void devcgroup_offline(struct cgroup_subsys_state *css)
222 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
224 mutex_lock(&devcgroup_mutex);
225 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
226 mutex_unlock(&devcgroup_mutex);
230 * called from kernel/cgroup.c with cgroup_lock() held.
232 static struct cgroup_subsys_state *
233 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
235 struct dev_cgroup *dev_cgroup;
237 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
239 return ERR_PTR(-ENOMEM);
240 INIT_LIST_HEAD(&dev_cgroup->exceptions);
241 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
243 return &dev_cgroup->css;
246 static void devcgroup_css_free(struct cgroup_subsys_state *css)
248 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
250 __dev_exception_clean(dev_cgroup);
254 #define DEVCG_ALLOW 1
261 static void set_access(char *acc, short access)
264 memset(acc, 0, ACCLEN);
265 if (access & ACC_READ)
267 if (access & ACC_WRITE)
269 if (access & ACC_MKNOD)
273 static char type_to_char(short type)
277 if (type == DEV_CHAR)
279 if (type == DEV_BLOCK)
284 static void set_majmin(char *str, unsigned m)
289 sprintf(str, "%u", m);
292 static int devcgroup_seq_read(struct cgroup_subsys_state *css,
293 struct cftype *cft, struct seq_file *m)
295 struct dev_cgroup *devcgroup = css_to_devcgroup(css);
296 struct dev_exception_item *ex;
297 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
301 * To preserve the compatibility:
302 * - Only show the "all devices" when the default policy is to allow
303 * - List the exceptions in case the default policy is to deny
304 * This way, the file remains as a "whitelist of devices"
306 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
307 set_access(acc, ACC_MASK);
310 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
313 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
314 set_access(acc, ex->access);
315 set_majmin(maj, ex->major);
316 set_majmin(min, ex->minor);
317 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
327 * may_access - verifies if a new exception is part of what is allowed
328 * by a dev cgroup based on the default policy +
329 * exceptions. This is used to make sure a child cgroup
330 * won't have more privileges than its parent or to
331 * verify if a certain access is allowed.
332 * @dev_cgroup: dev cgroup to be tested against
333 * @refex: new exception
334 * @behavior: behavior of the exception
336 static bool may_access(struct dev_cgroup *dev_cgroup,
337 struct dev_exception_item *refex,
338 enum devcg_behavior behavior)
340 struct dev_exception_item *ex;
343 rcu_lockdep_assert(rcu_read_lock_held() ||
344 lockdep_is_held(&devcgroup_mutex),
345 "device_cgroup::may_access() called without proper synchronization");
347 list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
348 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
350 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
352 if (ex->major != ~0 && ex->major != refex->major)
354 if (ex->minor != ~0 && ex->minor != refex->minor)
356 if (refex->access & (~ex->access))
362 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
363 if (behavior == DEVCG_DEFAULT_ALLOW) {
364 /* the exception will deny access to certain devices */
367 /* the exception will allow access to certain devices */
370 * a new exception allowing access shouldn't
371 * match an parent's exception
377 /* only behavior == DEVCG_DEFAULT_DENY allowed here */
379 /* parent has an exception that matches the proposed */
389 * when adding a new allow rule to a device exception list, the rule
390 * must be allowed in the parent device
392 static int parent_has_perm(struct dev_cgroup *childcg,
393 struct dev_exception_item *ex)
395 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css));
399 return may_access(parent, ex, childcg->behavior);
403 * may_allow_all - checks if it's possible to change the behavior to
404 * allow based on parent's rules.
405 * @parent: device cgroup's parent
406 * returns: != 0 in case it's allowed, 0 otherwise
408 static inline int may_allow_all(struct dev_cgroup *parent)
412 return parent->behavior == DEVCG_DEFAULT_ALLOW;
416 * revalidate_active_exceptions - walks through the active exception list and
417 * revalidates the exceptions based on parent's
418 * behavior and exceptions. The exceptions that
419 * are no longer valid will be removed.
420 * Called with devcgroup_mutex held.
421 * @devcg: cgroup which exceptions will be checked
423 * This is one of the three key functions for hierarchy implementation.
424 * This function is responsible for re-evaluating all the cgroup's active
425 * exceptions due to a parent's exception change.
426 * Refer to Documentation/cgroups/devices.txt for more details.
428 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
430 struct dev_exception_item *ex;
431 struct list_head *this, *tmp;
433 list_for_each_safe(this, tmp, &devcg->exceptions) {
434 ex = container_of(this, struct dev_exception_item, list);
435 if (!parent_has_perm(devcg, ex))
436 dev_exception_rm(devcg, ex);
441 * propagate_exception - propagates a new exception to the children
442 * @devcg_root: device cgroup that added a new exception
443 * @ex: new exception to be propagated
445 * returns: 0 in case of success, != 0 in case of error
447 static int propagate_exception(struct dev_cgroup *devcg_root,
448 struct dev_exception_item *ex)
450 struct cgroup *root = devcg_root->css.cgroup, *pos;
455 cgroup_for_each_descendant_pre(pos, root) {
456 struct dev_cgroup *devcg = cgroup_to_devcgroup(pos);
459 * Because devcgroup_mutex is held, no devcg will become
460 * online or offline during the tree walk (see on/offline
461 * methods), and online ones are safe to access outside RCU
462 * read lock without bumping refcnt.
464 if (!is_devcg_online(devcg))
470 * in case both root's behavior and devcg is allow, a new
471 * restriction means adding to the exception list
473 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
474 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
475 rc = dev_exception_add(devcg, ex);
480 * in the other possible cases:
481 * root's behavior: allow, devcg's: deny
482 * root's behavior: deny, devcg's: deny
483 * the exception will be removed
485 dev_exception_rm(devcg, ex);
487 revalidate_active_exceptions(devcg);
496 static inline bool has_children(struct dev_cgroup *devcgroup)
498 struct cgroup *cgrp = devcgroup->css.cgroup;
500 return !list_empty(&cgrp->children);
504 * Modify the exception list using allow/deny rules.
505 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
506 * so we can give a container CAP_MKNOD to let it create devices but not
507 * modify the exception list.
508 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
509 * us to also grant CAP_SYS_ADMIN to containers without giving away the
510 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
512 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
513 * new access is only allowed if you're in the top-level cgroup, or your
514 * parent cgroup has the access you're asking for.
516 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
517 int filetype, const char *buffer)
520 char temp[12]; /* 11 + 1 characters needed for a u32 */
522 struct dev_exception_item ex;
523 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&devcgroup->css));
525 if (!capable(CAP_SYS_ADMIN))
528 memset(&ex, 0, sizeof(ex));
535 if (has_children(devcgroup))
538 if (!may_allow_all(parent))
540 dev_exception_clean(devcgroup);
541 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
545 rc = dev_exceptions_copy(&devcgroup->exceptions,
546 &parent->exceptions);
551 if (has_children(devcgroup))
554 dev_exception_clean(devcgroup);
555 devcgroup->behavior = DEVCG_DEFAULT_DENY;
577 } else if (isdigit(*b)) {
578 memset(temp, 0, sizeof(temp));
579 for (count = 0; count < sizeof(temp) - 1; count++) {
585 rc = kstrtou32(temp, 10, &ex.major);
599 } else if (isdigit(*b)) {
600 memset(temp, 0, sizeof(temp));
601 for (count = 0; count < sizeof(temp) - 1; count++) {
607 rc = kstrtou32(temp, 10, &ex.minor);
615 for (b++, count = 0; count < 3; count++, b++) {
618 ex.access |= ACC_READ;
621 ex.access |= ACC_WRITE;
624 ex.access |= ACC_MKNOD;
637 if (!parent_has_perm(devcgroup, &ex))
640 * If the default policy is to allow by default, try to remove
641 * an matching exception instead. And be silent about it: we
642 * don't want to break compatibility
644 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
645 dev_exception_rm(devcgroup, &ex);
648 rc = dev_exception_add(devcgroup, &ex);
652 * If the default policy is to deny by default, try to remove
653 * an matching exception instead. And be silent about it: we
654 * don't want to break compatibility
656 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
657 dev_exception_rm(devcgroup, &ex);
659 rc = dev_exception_add(devcgroup, &ex);
663 /* we only propagate new restrictions */
664 rc = propagate_exception(devcgroup, &ex);
672 static int devcgroup_access_write(struct cgroup_subsys_state *css,
673 struct cftype *cft, const char *buffer)
677 mutex_lock(&devcgroup_mutex);
678 retval = devcgroup_update_access(css_to_devcgroup(css),
679 cft->private, buffer);
680 mutex_unlock(&devcgroup_mutex);
684 static struct cftype dev_cgroup_files[] = {
687 .write_string = devcgroup_access_write,
688 .private = DEVCG_ALLOW,
692 .write_string = devcgroup_access_write,
693 .private = DEVCG_DENY,
697 .read_seq_string = devcgroup_seq_read,
698 .private = DEVCG_LIST,
703 struct cgroup_subsys devices_subsys = {
705 .can_attach = devcgroup_can_attach,
706 .css_alloc = devcgroup_css_alloc,
707 .css_free = devcgroup_css_free,
708 .css_online = devcgroup_online,
709 .css_offline = devcgroup_offline,
710 .subsys_id = devices_subsys_id,
711 .base_cftypes = dev_cgroup_files,
715 * __devcgroup_check_permission - checks if an inode operation is permitted
716 * @dev_cgroup: the dev cgroup to be tested against
718 * @major: device major number
719 * @minor: device minor number
720 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
722 * returns 0 on success, -EPERM case the operation is not permitted
724 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
727 struct dev_cgroup *dev_cgroup;
728 struct dev_exception_item ex;
731 memset(&ex, 0, sizeof(ex));
738 dev_cgroup = task_devcgroup(current);
739 rc = may_access(dev_cgroup, &ex, dev_cgroup->behavior);
748 int __devcgroup_inode_permission(struct inode *inode, int mask)
750 short type, access = 0;
752 if (S_ISBLK(inode->i_mode))
754 if (S_ISCHR(inode->i_mode))
756 if (mask & MAY_WRITE)
761 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
765 int devcgroup_inode_mknod(int mode, dev_t dev)
769 if (!S_ISBLK(mode) && !S_ISCHR(mode))
777 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),