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;
66 static int devcgroup_can_attach(struct cgroup_subsys_state *new_css,
67 struct cgroup_taskset *set)
69 struct task_struct *task = cgroup_taskset_first(set);
71 if (current != task && !capable(CAP_SYS_ADMIN))
77 * called under devcgroup_mutex
79 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
81 struct dev_exception_item *ex, *tmp, *new;
83 lockdep_assert_held(&devcgroup_mutex);
85 list_for_each_entry(ex, orig, list) {
86 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
89 list_add_tail(&new->list, dest);
95 list_for_each_entry_safe(ex, tmp, dest, list) {
103 * called under devcgroup_mutex
105 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
106 struct dev_exception_item *ex)
108 struct dev_exception_item *excopy, *walk;
110 lockdep_assert_held(&devcgroup_mutex);
112 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
116 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
117 if (walk->type != ex->type)
119 if (walk->major != ex->major)
121 if (walk->minor != ex->minor)
124 walk->access |= ex->access;
130 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
135 * called under devcgroup_mutex
137 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
138 struct dev_exception_item *ex)
140 struct dev_exception_item *walk, *tmp;
142 lockdep_assert_held(&devcgroup_mutex);
144 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
145 if (walk->type != ex->type)
147 if (walk->major != ex->major)
149 if (walk->minor != ex->minor)
152 walk->access &= ~ex->access;
154 list_del_rcu(&walk->list);
155 kfree_rcu(walk, rcu);
160 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
162 struct dev_exception_item *ex, *tmp;
164 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
165 list_del_rcu(&ex->list);
171 * dev_exception_clean - frees all entries of the exception list
172 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
174 * called under devcgroup_mutex
176 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
178 lockdep_assert_held(&devcgroup_mutex);
180 __dev_exception_clean(dev_cgroup);
183 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
185 return (devcg->behavior != DEVCG_DEFAULT_NONE);
189 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
191 * @css: css getting online
192 * returns 0 in case of success, error code otherwise
194 static int devcgroup_online(struct cgroup_subsys_state *css)
196 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
197 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css_parent(css));
200 mutex_lock(&devcgroup_mutex);
202 if (parent_dev_cgroup == NULL)
203 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
205 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
206 &parent_dev_cgroup->exceptions);
208 dev_cgroup->behavior = parent_dev_cgroup->behavior;
210 mutex_unlock(&devcgroup_mutex);
215 static void devcgroup_offline(struct cgroup_subsys_state *css)
217 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
219 mutex_lock(&devcgroup_mutex);
220 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
221 mutex_unlock(&devcgroup_mutex);
225 * called from kernel/cgroup.c with cgroup_lock() held.
227 static struct cgroup_subsys_state *
228 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
230 struct dev_cgroup *dev_cgroup;
232 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
234 return ERR_PTR(-ENOMEM);
235 INIT_LIST_HEAD(&dev_cgroup->exceptions);
236 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
238 return &dev_cgroup->css;
241 static void devcgroup_css_free(struct cgroup_subsys_state *css)
243 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
245 __dev_exception_clean(dev_cgroup);
249 #define DEVCG_ALLOW 1
256 static void set_access(char *acc, short access)
259 memset(acc, 0, ACCLEN);
260 if (access & ACC_READ)
262 if (access & ACC_WRITE)
264 if (access & ACC_MKNOD)
268 static char type_to_char(short type)
272 if (type == DEV_CHAR)
274 if (type == DEV_BLOCK)
279 static void set_majmin(char *str, unsigned m)
284 sprintf(str, "%u", m);
287 static int devcgroup_seq_read(struct cgroup_subsys_state *css,
288 struct cftype *cft, struct seq_file *m)
290 struct dev_cgroup *devcgroup = css_to_devcgroup(css);
291 struct dev_exception_item *ex;
292 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
296 * To preserve the compatibility:
297 * - Only show the "all devices" when the default policy is to allow
298 * - List the exceptions in case the default policy is to deny
299 * This way, the file remains as a "whitelist of devices"
301 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
302 set_access(acc, ACC_MASK);
305 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
308 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
309 set_access(acc, ex->access);
310 set_majmin(maj, ex->major);
311 set_majmin(min, ex->minor);
312 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
322 * may_access - verifies if a new exception is part of what is allowed
323 * by a dev cgroup based on the default policy +
324 * exceptions. This is used to make sure a child cgroup
325 * won't have more privileges than its parent or to
326 * verify if a certain access is allowed.
327 * @dev_cgroup: dev cgroup to be tested against
328 * @refex: new exception
329 * @behavior: behavior of the exception
331 static bool may_access(struct dev_cgroup *dev_cgroup,
332 struct dev_exception_item *refex,
333 enum devcg_behavior behavior)
335 struct dev_exception_item *ex;
338 rcu_lockdep_assert(rcu_read_lock_held() ||
339 lockdep_is_held(&devcgroup_mutex),
340 "device_cgroup::may_access() called without proper synchronization");
342 list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
343 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
345 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
347 if (ex->major != ~0 && ex->major != refex->major)
349 if (ex->minor != ~0 && ex->minor != refex->minor)
351 if (refex->access & (~ex->access))
357 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
358 if (behavior == DEVCG_DEFAULT_ALLOW) {
359 /* the exception will deny access to certain devices */
362 /* the exception will allow access to certain devices */
365 * a new exception allowing access shouldn't
366 * match an parent's exception
372 /* only behavior == DEVCG_DEFAULT_DENY allowed here */
374 /* parent has an exception that matches the proposed */
384 * when adding a new allow rule to a device exception list, the rule
385 * must be allowed in the parent device
387 static int parent_has_perm(struct dev_cgroup *childcg,
388 struct dev_exception_item *ex)
390 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css));
394 return may_access(parent, ex, childcg->behavior);
398 * may_allow_all - checks if it's possible to change the behavior to
399 * allow based on parent's rules.
400 * @parent: device cgroup's parent
401 * returns: != 0 in case it's allowed, 0 otherwise
403 static inline int may_allow_all(struct dev_cgroup *parent)
407 return parent->behavior == DEVCG_DEFAULT_ALLOW;
411 * revalidate_active_exceptions - walks through the active exception list and
412 * revalidates the exceptions based on parent's
413 * behavior and exceptions. The exceptions that
414 * are no longer valid will be removed.
415 * Called with devcgroup_mutex held.
416 * @devcg: cgroup which exceptions will be checked
418 * This is one of the three key functions for hierarchy implementation.
419 * This function is responsible for re-evaluating all the cgroup's active
420 * exceptions due to a parent's exception change.
421 * Refer to Documentation/cgroups/devices.txt for more details.
423 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
425 struct dev_exception_item *ex;
426 struct list_head *this, *tmp;
428 list_for_each_safe(this, tmp, &devcg->exceptions) {
429 ex = container_of(this, struct dev_exception_item, list);
430 if (!parent_has_perm(devcg, ex))
431 dev_exception_rm(devcg, ex);
436 * propagate_exception - propagates a new exception to the children
437 * @devcg_root: device cgroup that added a new exception
438 * @ex: new exception to be propagated
440 * returns: 0 in case of success, != 0 in case of error
442 static int propagate_exception(struct dev_cgroup *devcg_root,
443 struct dev_exception_item *ex)
445 struct cgroup_subsys_state *pos;
450 css_for_each_descendant_pre(pos, &devcg_root->css) {
451 struct dev_cgroup *devcg = css_to_devcgroup(pos);
454 * Because devcgroup_mutex is held, no devcg will become
455 * online or offline during the tree walk (see on/offline
456 * methods), and online ones are safe to access outside RCU
457 * read lock without bumping refcnt.
459 if (pos == &devcg_root->css || !is_devcg_online(devcg))
465 * in case both root's behavior and devcg is allow, a new
466 * restriction means adding to the exception list
468 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
469 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
470 rc = dev_exception_add(devcg, ex);
475 * in the other possible cases:
476 * root's behavior: allow, devcg's: deny
477 * root's behavior: deny, devcg's: deny
478 * the exception will be removed
480 dev_exception_rm(devcg, ex);
482 revalidate_active_exceptions(devcg);
491 static inline bool has_children(struct dev_cgroup *devcgroup)
493 struct cgroup *cgrp = devcgroup->css.cgroup;
495 return !list_empty(&cgrp->children);
499 * Modify the exception list using allow/deny rules.
500 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
501 * so we can give a container CAP_MKNOD to let it create devices but not
502 * modify the exception list.
503 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
504 * us to also grant CAP_SYS_ADMIN to containers without giving away the
505 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
507 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
508 * new access is only allowed if you're in the top-level cgroup, or your
509 * parent cgroup has the access you're asking for.
511 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
512 int filetype, const char *buffer)
515 char temp[12]; /* 11 + 1 characters needed for a u32 */
517 struct dev_exception_item ex;
518 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&devcgroup->css));
520 if (!capable(CAP_SYS_ADMIN))
523 memset(&ex, 0, sizeof(ex));
530 if (has_children(devcgroup))
533 if (!may_allow_all(parent))
535 dev_exception_clean(devcgroup);
536 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
540 rc = dev_exceptions_copy(&devcgroup->exceptions,
541 &parent->exceptions);
546 if (has_children(devcgroup))
549 dev_exception_clean(devcgroup);
550 devcgroup->behavior = DEVCG_DEFAULT_DENY;
572 } else if (isdigit(*b)) {
573 memset(temp, 0, sizeof(temp));
574 for (count = 0; count < sizeof(temp) - 1; count++) {
580 rc = kstrtou32(temp, 10, &ex.major);
594 } else if (isdigit(*b)) {
595 memset(temp, 0, sizeof(temp));
596 for (count = 0; count < sizeof(temp) - 1; count++) {
602 rc = kstrtou32(temp, 10, &ex.minor);
610 for (b++, count = 0; count < 3; count++, b++) {
613 ex.access |= ACC_READ;
616 ex.access |= ACC_WRITE;
619 ex.access |= ACC_MKNOD;
632 if (!parent_has_perm(devcgroup, &ex))
635 * If the default policy is to allow by default, try to remove
636 * an matching exception instead. And be silent about it: we
637 * don't want to break compatibility
639 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
640 dev_exception_rm(devcgroup, &ex);
643 rc = dev_exception_add(devcgroup, &ex);
647 * If the default policy is to deny by default, try to remove
648 * an matching exception instead. And be silent about it: we
649 * don't want to break compatibility
651 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
652 dev_exception_rm(devcgroup, &ex);
654 rc = dev_exception_add(devcgroup, &ex);
658 /* we only propagate new restrictions */
659 rc = propagate_exception(devcgroup, &ex);
667 static int devcgroup_access_write(struct cgroup_subsys_state *css,
668 struct cftype *cft, const char *buffer)
672 mutex_lock(&devcgroup_mutex);
673 retval = devcgroup_update_access(css_to_devcgroup(css),
674 cft->private, buffer);
675 mutex_unlock(&devcgroup_mutex);
679 static struct cftype dev_cgroup_files[] = {
682 .write_string = devcgroup_access_write,
683 .private = DEVCG_ALLOW,
687 .write_string = devcgroup_access_write,
688 .private = DEVCG_DENY,
692 .read_seq_string = devcgroup_seq_read,
693 .private = DEVCG_LIST,
698 struct cgroup_subsys devices_subsys = {
700 .can_attach = devcgroup_can_attach,
701 .css_alloc = devcgroup_css_alloc,
702 .css_free = devcgroup_css_free,
703 .css_online = devcgroup_online,
704 .css_offline = devcgroup_offline,
705 .subsys_id = devices_subsys_id,
706 .base_cftypes = dev_cgroup_files,
710 * __devcgroup_check_permission - checks if an inode operation is permitted
711 * @dev_cgroup: the dev cgroup to be tested against
713 * @major: device major number
714 * @minor: device minor number
715 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
717 * returns 0 on success, -EPERM case the operation is not permitted
719 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
722 struct dev_cgroup *dev_cgroup;
723 struct dev_exception_item ex;
726 memset(&ex, 0, sizeof(ex));
733 dev_cgroup = task_devcgroup(current);
734 rc = may_access(dev_cgroup, &ex, dev_cgroup->behavior);
743 int __devcgroup_inode_permission(struct inode *inode, int mask)
745 short type, access = 0;
747 if (S_ISBLK(inode->i_mode))
749 if (S_ISCHR(inode->i_mode))
751 if (mask & MAY_WRITE)
756 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
760 int devcgroup_inode_mknod(int mode, dev_t dev)
764 if (!S_ISBLK(mode) && !S_ISCHR(mode))
772 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),