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;
67 * called under devcgroup_mutex
69 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
71 struct dev_exception_item *ex, *tmp, *new;
73 lockdep_assert_held(&devcgroup_mutex);
75 list_for_each_entry(ex, orig, list) {
76 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
79 list_add_tail(&new->list, dest);
85 list_for_each_entry_safe(ex, tmp, dest, list) {
93 * called under devcgroup_mutex
95 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
96 struct dev_exception_item *ex)
98 struct dev_exception_item *excopy, *walk;
100 lockdep_assert_held(&devcgroup_mutex);
102 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
106 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
107 if (walk->type != ex->type)
109 if (walk->major != ex->major)
111 if (walk->minor != ex->minor)
114 walk->access |= ex->access;
120 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
125 * called under devcgroup_mutex
127 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
128 struct dev_exception_item *ex)
130 struct dev_exception_item *walk, *tmp;
132 lockdep_assert_held(&devcgroup_mutex);
134 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
135 if (walk->type != ex->type)
137 if (walk->major != ex->major)
139 if (walk->minor != ex->minor)
142 walk->access &= ~ex->access;
144 list_del_rcu(&walk->list);
145 kfree_rcu(walk, rcu);
150 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
152 struct dev_exception_item *ex, *tmp;
154 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
155 list_del_rcu(&ex->list);
161 * dev_exception_clean - frees all entries of the exception list
162 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
164 * called under devcgroup_mutex
166 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
168 lockdep_assert_held(&devcgroup_mutex);
170 __dev_exception_clean(dev_cgroup);
173 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
175 return (devcg->behavior != DEVCG_DEFAULT_NONE);
179 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
181 * @css: css getting online
182 * returns 0 in case of success, error code otherwise
184 static int devcgroup_online(struct cgroup_subsys_state *css)
186 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
187 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css_parent(css));
190 mutex_lock(&devcgroup_mutex);
192 if (parent_dev_cgroup == NULL)
193 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
195 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
196 &parent_dev_cgroup->exceptions);
198 dev_cgroup->behavior = parent_dev_cgroup->behavior;
200 mutex_unlock(&devcgroup_mutex);
205 static void devcgroup_offline(struct cgroup_subsys_state *css)
207 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
209 mutex_lock(&devcgroup_mutex);
210 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
211 mutex_unlock(&devcgroup_mutex);
215 * called from kernel/cgroup.c with cgroup_lock() held.
217 static struct cgroup_subsys_state *
218 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
220 struct dev_cgroup *dev_cgroup;
222 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
224 return ERR_PTR(-ENOMEM);
225 INIT_LIST_HEAD(&dev_cgroup->exceptions);
226 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
228 return &dev_cgroup->css;
231 static void devcgroup_css_free(struct cgroup_subsys_state *css)
233 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
235 __dev_exception_clean(dev_cgroup);
239 #define DEVCG_ALLOW 1
246 static void set_access(char *acc, short access)
249 memset(acc, 0, ACCLEN);
250 if (access & ACC_READ)
252 if (access & ACC_WRITE)
254 if (access & ACC_MKNOD)
258 static char type_to_char(short type)
262 if (type == DEV_CHAR)
264 if (type == DEV_BLOCK)
269 static void set_majmin(char *str, unsigned m)
274 sprintf(str, "%u", m);
277 static int devcgroup_seq_read(struct cgroup_subsys_state *css,
278 struct cftype *cft, struct seq_file *m)
280 struct dev_cgroup *devcgroup = css_to_devcgroup(css);
281 struct dev_exception_item *ex;
282 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
286 * To preserve the compatibility:
287 * - Only show the "all devices" when the default policy is to allow
288 * - List the exceptions in case the default policy is to deny
289 * This way, the file remains as a "whitelist of devices"
291 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
292 set_access(acc, ACC_MASK);
295 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
298 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
299 set_access(acc, ex->access);
300 set_majmin(maj, ex->major);
301 set_majmin(min, ex->minor);
302 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
312 * may_access - verifies if a new exception is part of what is allowed
313 * by a dev cgroup based on the default policy +
314 * exceptions. This is used to make sure a child cgroup
315 * won't have more privileges than its parent or to
316 * verify if a certain access is allowed.
317 * @dev_cgroup: dev cgroup to be tested against
318 * @refex: new exception
319 * @behavior: behavior of the exception
321 static bool may_access(struct dev_cgroup *dev_cgroup,
322 struct dev_exception_item *refex,
323 enum devcg_behavior behavior)
325 struct dev_exception_item *ex;
328 rcu_lockdep_assert(rcu_read_lock_held() ||
329 lockdep_is_held(&devcgroup_mutex),
330 "device_cgroup::may_access() called without proper synchronization");
332 list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
333 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
335 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
337 if (ex->major != ~0 && ex->major != refex->major)
339 if (ex->minor != ~0 && ex->minor != refex->minor)
341 if (refex->access & (~ex->access))
347 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
348 if (behavior == DEVCG_DEFAULT_ALLOW) {
349 /* the exception will deny access to certain devices */
352 /* the exception will allow access to certain devices */
355 * a new exception allowing access shouldn't
356 * match an parent's exception
362 /* only behavior == DEVCG_DEFAULT_DENY allowed here */
364 /* parent has an exception that matches the proposed */
374 * when adding a new allow rule to a device exception list, the rule
375 * must be allowed in the parent device
377 static int parent_has_perm(struct dev_cgroup *childcg,
378 struct dev_exception_item *ex)
380 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css));
384 return may_access(parent, ex, childcg->behavior);
388 * may_allow_all - checks if it's possible to change the behavior to
389 * allow based on parent's rules.
390 * @parent: device cgroup's parent
391 * returns: != 0 in case it's allowed, 0 otherwise
393 static inline int may_allow_all(struct dev_cgroup *parent)
397 return parent->behavior == DEVCG_DEFAULT_ALLOW;
401 * revalidate_active_exceptions - walks through the active exception list and
402 * revalidates the exceptions based on parent's
403 * behavior and exceptions. The exceptions that
404 * are no longer valid will be removed.
405 * Called with devcgroup_mutex held.
406 * @devcg: cgroup which exceptions will be checked
408 * This is one of the three key functions for hierarchy implementation.
409 * This function is responsible for re-evaluating all the cgroup's active
410 * exceptions due to a parent's exception change.
411 * Refer to Documentation/cgroups/devices.txt for more details.
413 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
415 struct dev_exception_item *ex;
416 struct list_head *this, *tmp;
418 list_for_each_safe(this, tmp, &devcg->exceptions) {
419 ex = container_of(this, struct dev_exception_item, list);
420 if (!parent_has_perm(devcg, ex))
421 dev_exception_rm(devcg, ex);
426 * propagate_exception - propagates a new exception to the children
427 * @devcg_root: device cgroup that added a new exception
428 * @ex: new exception to be propagated
430 * returns: 0 in case of success, != 0 in case of error
432 static int propagate_exception(struct dev_cgroup *devcg_root,
433 struct dev_exception_item *ex)
435 struct cgroup_subsys_state *pos;
440 css_for_each_descendant_pre(pos, &devcg_root->css) {
441 struct dev_cgroup *devcg = css_to_devcgroup(pos);
444 * Because devcgroup_mutex is held, no devcg will become
445 * online or offline during the tree walk (see on/offline
446 * methods), and online ones are safe to access outside RCU
447 * read lock without bumping refcnt.
449 if (pos == &devcg_root->css || !is_devcg_online(devcg))
455 * in case both root's behavior and devcg is allow, a new
456 * restriction means adding to the exception list
458 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
459 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
460 rc = dev_exception_add(devcg, ex);
465 * in the other possible cases:
466 * root's behavior: allow, devcg's: deny
467 * root's behavior: deny, devcg's: deny
468 * the exception will be removed
470 dev_exception_rm(devcg, ex);
472 revalidate_active_exceptions(devcg);
481 static inline bool has_children(struct dev_cgroup *devcgroup)
483 struct cgroup *cgrp = devcgroup->css.cgroup;
485 return !list_empty(&cgrp->children);
489 * Modify the exception list using allow/deny rules.
490 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
491 * so we can give a container CAP_MKNOD to let it create devices but not
492 * modify the exception list.
493 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
494 * us to also grant CAP_SYS_ADMIN to containers without giving away the
495 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
497 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
498 * new access is only allowed if you're in the top-level cgroup, or your
499 * parent cgroup has the access you're asking for.
501 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
502 int filetype, const char *buffer)
505 char temp[12]; /* 11 + 1 characters needed for a u32 */
507 struct dev_exception_item ex;
508 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&devcgroup->css));
510 if (!capable(CAP_SYS_ADMIN))
513 memset(&ex, 0, sizeof(ex));
520 if (has_children(devcgroup))
523 if (!may_allow_all(parent))
525 dev_exception_clean(devcgroup);
526 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
530 rc = dev_exceptions_copy(&devcgroup->exceptions,
531 &parent->exceptions);
536 if (has_children(devcgroup))
539 dev_exception_clean(devcgroup);
540 devcgroup->behavior = DEVCG_DEFAULT_DENY;
562 } else if (isdigit(*b)) {
563 memset(temp, 0, sizeof(temp));
564 for (count = 0; count < sizeof(temp) - 1; count++) {
570 rc = kstrtou32(temp, 10, &ex.major);
584 } else if (isdigit(*b)) {
585 memset(temp, 0, sizeof(temp));
586 for (count = 0; count < sizeof(temp) - 1; count++) {
592 rc = kstrtou32(temp, 10, &ex.minor);
600 for (b++, count = 0; count < 3; count++, b++) {
603 ex.access |= ACC_READ;
606 ex.access |= ACC_WRITE;
609 ex.access |= ACC_MKNOD;
622 if (!parent_has_perm(devcgroup, &ex))
625 * If the default policy is to allow by default, try to remove
626 * an matching exception instead. And be silent about it: we
627 * don't want to break compatibility
629 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
630 dev_exception_rm(devcgroup, &ex);
633 rc = dev_exception_add(devcgroup, &ex);
637 * If the default policy is to deny by default, try to remove
638 * an matching exception instead. And be silent about it: we
639 * don't want to break compatibility
641 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
642 dev_exception_rm(devcgroup, &ex);
644 rc = dev_exception_add(devcgroup, &ex);
648 /* we only propagate new restrictions */
649 rc = propagate_exception(devcgroup, &ex);
657 static int devcgroup_access_write(struct cgroup_subsys_state *css,
658 struct cftype *cft, const char *buffer)
662 mutex_lock(&devcgroup_mutex);
663 retval = devcgroup_update_access(css_to_devcgroup(css),
664 cft->private, buffer);
665 mutex_unlock(&devcgroup_mutex);
669 static struct cftype dev_cgroup_files[] = {
672 .write_string = devcgroup_access_write,
673 .private = DEVCG_ALLOW,
677 .write_string = devcgroup_access_write,
678 .private = DEVCG_DENY,
682 .read_seq_string = devcgroup_seq_read,
683 .private = DEVCG_LIST,
688 struct cgroup_subsys devices_subsys = {
690 .css_alloc = devcgroup_css_alloc,
691 .css_free = devcgroup_css_free,
692 .css_online = devcgroup_online,
693 .css_offline = devcgroup_offline,
694 .subsys_id = devices_subsys_id,
695 .base_cftypes = dev_cgroup_files,
699 * __devcgroup_check_permission - checks if an inode operation is permitted
700 * @dev_cgroup: the dev cgroup to be tested against
702 * @major: device major number
703 * @minor: device minor number
704 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
706 * returns 0 on success, -EPERM case the operation is not permitted
708 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
711 struct dev_cgroup *dev_cgroup;
712 struct dev_exception_item ex;
715 memset(&ex, 0, sizeof(ex));
722 dev_cgroup = task_devcgroup(current);
723 rc = may_access(dev_cgroup, &ex, dev_cgroup->behavior);
732 int __devcgroup_inode_permission(struct inode *inode, int mask)
734 short type, access = 0;
736 if (S_ISBLK(inode->i_mode))
738 if (S_ISCHR(inode->i_mode))
740 if (mask & MAY_WRITE)
745 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
749 int devcgroup_inode_mknod(int mode, dev_t dev)
753 if (!S_ISBLK(mode) && !S_ISCHR(mode))
761 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),