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_cgrp_id));
65 * called under devcgroup_mutex
67 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
69 struct dev_exception_item *ex, *tmp, *new;
71 lockdep_assert_held(&devcgroup_mutex);
73 list_for_each_entry(ex, orig, list) {
74 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
77 list_add_tail(&new->list, dest);
83 list_for_each_entry_safe(ex, tmp, dest, list) {
91 * called under devcgroup_mutex
93 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
94 struct dev_exception_item *ex)
96 struct dev_exception_item *excopy, *walk;
98 lockdep_assert_held(&devcgroup_mutex);
100 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
104 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
105 if (walk->type != ex->type)
107 if (walk->major != ex->major)
109 if (walk->minor != ex->minor)
112 walk->access |= ex->access;
118 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
123 * called under devcgroup_mutex
125 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
126 struct dev_exception_item *ex)
128 struct dev_exception_item *walk, *tmp;
130 lockdep_assert_held(&devcgroup_mutex);
132 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
133 if (walk->type != ex->type)
135 if (walk->major != ex->major)
137 if (walk->minor != ex->minor)
140 walk->access &= ~ex->access;
142 list_del_rcu(&walk->list);
143 kfree_rcu(walk, rcu);
148 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
150 struct dev_exception_item *ex, *tmp;
152 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
153 list_del_rcu(&ex->list);
159 * dev_exception_clean - frees all entries of the exception list
160 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
162 * called under devcgroup_mutex
164 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
166 lockdep_assert_held(&devcgroup_mutex);
168 __dev_exception_clean(dev_cgroup);
171 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
173 return (devcg->behavior != DEVCG_DEFAULT_NONE);
177 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
179 * @css: css getting online
180 * returns 0 in case of success, error code otherwise
182 static int devcgroup_online(struct cgroup_subsys_state *css)
184 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
185 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css_parent(css));
188 mutex_lock(&devcgroup_mutex);
190 if (parent_dev_cgroup == NULL)
191 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
193 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
194 &parent_dev_cgroup->exceptions);
196 dev_cgroup->behavior = parent_dev_cgroup->behavior;
198 mutex_unlock(&devcgroup_mutex);
203 static void devcgroup_offline(struct cgroup_subsys_state *css)
205 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
207 mutex_lock(&devcgroup_mutex);
208 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
209 mutex_unlock(&devcgroup_mutex);
213 * called from kernel/cgroup.c with cgroup_lock() held.
215 static struct cgroup_subsys_state *
216 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
218 struct dev_cgroup *dev_cgroup;
220 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
222 return ERR_PTR(-ENOMEM);
223 INIT_LIST_HEAD(&dev_cgroup->exceptions);
224 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
226 return &dev_cgroup->css;
229 static void devcgroup_css_free(struct cgroup_subsys_state *css)
231 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
233 __dev_exception_clean(dev_cgroup);
237 #define DEVCG_ALLOW 1
244 static void set_access(char *acc, short access)
247 memset(acc, 0, ACCLEN);
248 if (access & ACC_READ)
250 if (access & ACC_WRITE)
252 if (access & ACC_MKNOD)
256 static char type_to_char(short type)
260 if (type == DEV_CHAR)
262 if (type == DEV_BLOCK)
267 static void set_majmin(char *str, unsigned m)
272 sprintf(str, "%u", m);
275 static int devcgroup_seq_show(struct seq_file *m, void *v)
277 struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
278 struct dev_exception_item *ex;
279 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
283 * To preserve the compatibility:
284 * - Only show the "all devices" when the default policy is to allow
285 * - List the exceptions in case the default policy is to deny
286 * This way, the file remains as a "whitelist of devices"
288 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
289 set_access(acc, ACC_MASK);
292 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
295 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
296 set_access(acc, ex->access);
297 set_majmin(maj, ex->major);
298 set_majmin(min, ex->minor);
299 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
309 * may_access - verifies if a new exception is part of what is allowed
310 * by a dev cgroup based on the default policy +
311 * exceptions. This is used to make sure a child cgroup
312 * won't have more privileges than its parent or to
313 * verify if a certain access is allowed.
314 * @dev_cgroup: dev cgroup to be tested against
315 * @refex: new exception
316 * @behavior: behavior of the exception
318 static bool may_access(struct dev_cgroup *dev_cgroup,
319 struct dev_exception_item *refex,
320 enum devcg_behavior behavior)
322 struct dev_exception_item *ex;
325 rcu_lockdep_assert(rcu_read_lock_held() ||
326 lockdep_is_held(&devcgroup_mutex),
327 "device_cgroup::may_access() called without proper synchronization");
329 list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
330 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
332 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
334 if (ex->major != ~0 && ex->major != refex->major)
336 if (ex->minor != ~0 && ex->minor != refex->minor)
338 if (refex->access & (~ex->access))
344 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
345 if (behavior == DEVCG_DEFAULT_ALLOW) {
346 /* the exception will deny access to certain devices */
349 /* the exception will allow access to certain devices */
352 * a new exception allowing access shouldn't
353 * match an parent's exception
359 /* only behavior == DEVCG_DEFAULT_DENY allowed here */
361 /* parent has an exception that matches the proposed */
371 * when adding a new allow rule to a device exception list, the rule
372 * must be allowed in the parent device
374 static int parent_has_perm(struct dev_cgroup *childcg,
375 struct dev_exception_item *ex)
377 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css));
381 return may_access(parent, ex, childcg->behavior);
385 * may_allow_all - checks if it's possible to change the behavior to
386 * allow based on parent's rules.
387 * @parent: device cgroup's parent
388 * returns: != 0 in case it's allowed, 0 otherwise
390 static inline int may_allow_all(struct dev_cgroup *parent)
394 return parent->behavior == DEVCG_DEFAULT_ALLOW;
398 * revalidate_active_exceptions - walks through the active exception list and
399 * revalidates the exceptions based on parent's
400 * behavior and exceptions. The exceptions that
401 * are no longer valid will be removed.
402 * Called with devcgroup_mutex held.
403 * @devcg: cgroup which exceptions will be checked
405 * This is one of the three key functions for hierarchy implementation.
406 * This function is responsible for re-evaluating all the cgroup's active
407 * exceptions due to a parent's exception change.
408 * Refer to Documentation/cgroups/devices.txt for more details.
410 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
412 struct dev_exception_item *ex;
413 struct list_head *this, *tmp;
415 list_for_each_safe(this, tmp, &devcg->exceptions) {
416 ex = container_of(this, struct dev_exception_item, list);
417 if (!parent_has_perm(devcg, ex))
418 dev_exception_rm(devcg, ex);
423 * propagate_exception - propagates a new exception to the children
424 * @devcg_root: device cgroup that added a new exception
425 * @ex: new exception to be propagated
427 * returns: 0 in case of success, != 0 in case of error
429 static int propagate_exception(struct dev_cgroup *devcg_root,
430 struct dev_exception_item *ex)
432 struct cgroup_subsys_state *pos;
437 css_for_each_descendant_pre(pos, &devcg_root->css) {
438 struct dev_cgroup *devcg = css_to_devcgroup(pos);
441 * Because devcgroup_mutex is held, no devcg will become
442 * online or offline during the tree walk (see on/offline
443 * methods), and online ones are safe to access outside RCU
444 * read lock without bumping refcnt.
446 if (pos == &devcg_root->css || !is_devcg_online(devcg))
452 * in case both root's behavior and devcg is allow, a new
453 * restriction means adding to the exception list
455 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
456 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
457 rc = dev_exception_add(devcg, ex);
462 * in the other possible cases:
463 * root's behavior: allow, devcg's: deny
464 * root's behavior: deny, devcg's: deny
465 * the exception will be removed
467 dev_exception_rm(devcg, ex);
469 revalidate_active_exceptions(devcg);
478 static inline bool has_children(struct dev_cgroup *devcgroup)
480 struct cgroup *cgrp = devcgroup->css.cgroup;
482 return !list_empty(&cgrp->children);
486 * Modify the exception list using allow/deny rules.
487 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
488 * so we can give a container CAP_MKNOD to let it create devices but not
489 * modify the exception list.
490 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
491 * us to also grant CAP_SYS_ADMIN to containers without giving away the
492 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
494 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
495 * new access is only allowed if you're in the top-level cgroup, or your
496 * parent cgroup has the access you're asking for.
498 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
499 int filetype, char *buffer)
502 char temp[12]; /* 11 + 1 characters needed for a u32 */
504 struct dev_exception_item ex;
505 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&devcgroup->css));
507 if (!capable(CAP_SYS_ADMIN))
510 memset(&ex, 0, sizeof(ex));
517 if (has_children(devcgroup))
520 if (!may_allow_all(parent))
522 dev_exception_clean(devcgroup);
523 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
527 rc = dev_exceptions_copy(&devcgroup->exceptions,
528 &parent->exceptions);
533 if (has_children(devcgroup))
536 dev_exception_clean(devcgroup);
537 devcgroup->behavior = DEVCG_DEFAULT_DENY;
559 } else if (isdigit(*b)) {
560 memset(temp, 0, sizeof(temp));
561 for (count = 0; count < sizeof(temp) - 1; count++) {
567 rc = kstrtou32(temp, 10, &ex.major);
581 } else if (isdigit(*b)) {
582 memset(temp, 0, sizeof(temp));
583 for (count = 0; count < sizeof(temp) - 1; count++) {
589 rc = kstrtou32(temp, 10, &ex.minor);
597 for (b++, count = 0; count < 3; count++, b++) {
600 ex.access |= ACC_READ;
603 ex.access |= ACC_WRITE;
606 ex.access |= ACC_MKNOD;
619 if (!parent_has_perm(devcgroup, &ex))
622 * If the default policy is to allow by default, try to remove
623 * an matching exception instead. And be silent about it: we
624 * don't want to break compatibility
626 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
627 dev_exception_rm(devcgroup, &ex);
630 rc = dev_exception_add(devcgroup, &ex);
634 * If the default policy is to deny by default, try to remove
635 * an matching exception instead. And be silent about it: we
636 * don't want to break compatibility
638 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
639 dev_exception_rm(devcgroup, &ex);
641 rc = dev_exception_add(devcgroup, &ex);
645 /* we only propagate new restrictions */
646 rc = propagate_exception(devcgroup, &ex);
654 static int devcgroup_access_write(struct cgroup_subsys_state *css,
655 struct cftype *cft, char *buffer)
659 mutex_lock(&devcgroup_mutex);
660 retval = devcgroup_update_access(css_to_devcgroup(css),
661 cft->private, buffer);
662 mutex_unlock(&devcgroup_mutex);
666 static struct cftype dev_cgroup_files[] = {
669 .write_string = devcgroup_access_write,
670 .private = DEVCG_ALLOW,
674 .write_string = devcgroup_access_write,
675 .private = DEVCG_DENY,
679 .seq_show = devcgroup_seq_show,
680 .private = DEVCG_LIST,
685 struct cgroup_subsys devices_cgrp_subsys = {
686 .css_alloc = devcgroup_css_alloc,
687 .css_free = devcgroup_css_free,
688 .css_online = devcgroup_online,
689 .css_offline = devcgroup_offline,
690 .base_cftypes = dev_cgroup_files,
694 * __devcgroup_check_permission - checks if an inode operation is permitted
695 * @dev_cgroup: the dev cgroup to be tested against
697 * @major: device major number
698 * @minor: device minor number
699 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
701 * returns 0 on success, -EPERM case the operation is not permitted
703 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
706 struct dev_cgroup *dev_cgroup;
707 struct dev_exception_item ex;
710 memset(&ex, 0, sizeof(ex));
717 dev_cgroup = task_devcgroup(current);
718 rc = may_access(dev_cgroup, &ex, dev_cgroup->behavior);
727 int __devcgroup_inode_permission(struct inode *inode, int mask)
729 short type, access = 0;
731 if (S_ISBLK(inode->i_mode))
733 if (S_ISCHR(inode->i_mode))
735 if (mask & MAY_WRITE)
740 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
744 int devcgroup_inode_mknod(int mode, dev_t dev)
748 if (!S_ISBLK(mode) && !S_ISCHR(mode))
756 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),