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);
29 * exception list locking rules:
30 * hold devcgroup_mutex for update/read.
31 * hold rcu_read_lock() for read.
34 struct dev_exception_item {
38 struct list_head list;
43 struct cgroup_subsys_state css;
44 struct list_head exceptions;
51 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
53 return container_of(s, struct dev_cgroup, css);
56 static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup)
58 return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id));
61 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
63 return css_to_devcgroup(task_subsys_state(task, devices_subsys_id));
66 struct cgroup_subsys devices_subsys;
68 static int devcgroup_can_attach(struct cgroup *new_cgrp,
69 struct cgroup_taskset *set)
71 struct task_struct *task = cgroup_taskset_first(set);
73 if (current != task && !capable(CAP_SYS_ADMIN))
79 * called under devcgroup_mutex
81 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
83 struct dev_exception_item *ex, *tmp, *new;
85 lockdep_assert_held(&devcgroup_mutex);
87 list_for_each_entry(ex, orig, list) {
88 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
91 list_add_tail(&new->list, dest);
97 list_for_each_entry_safe(ex, tmp, dest, list) {
105 * called under devcgroup_mutex
107 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
108 struct dev_exception_item *ex)
110 struct dev_exception_item *excopy, *walk;
112 lockdep_assert_held(&devcgroup_mutex);
114 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
118 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
119 if (walk->type != ex->type)
121 if (walk->major != ex->major)
123 if (walk->minor != ex->minor)
126 walk->access |= ex->access;
132 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
137 * called under devcgroup_mutex
139 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
140 struct dev_exception_item *ex)
142 struct dev_exception_item *walk, *tmp;
144 lockdep_assert_held(&devcgroup_mutex);
146 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
147 if (walk->type != ex->type)
149 if (walk->major != ex->major)
151 if (walk->minor != ex->minor)
154 walk->access &= ~ex->access;
156 list_del_rcu(&walk->list);
157 kfree_rcu(walk, rcu);
163 * dev_exception_clean - frees all entries of the exception list
164 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
166 * called under devcgroup_mutex
168 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
170 struct dev_exception_item *ex, *tmp;
172 lockdep_assert_held(&devcgroup_mutex);
174 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
175 list_del_rcu(&ex->list);
181 * called from kernel/cgroup.c with cgroup_lock() held.
183 static struct cgroup_subsys_state *devcgroup_css_alloc(struct cgroup *cgroup)
185 struct dev_cgroup *dev_cgroup, *parent_dev_cgroup;
186 struct cgroup *parent_cgroup;
189 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
191 return ERR_PTR(-ENOMEM);
192 INIT_LIST_HEAD(&dev_cgroup->exceptions);
193 parent_cgroup = cgroup->parent;
195 if (parent_cgroup == NULL)
196 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
198 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
199 mutex_lock(&devcgroup_mutex);
200 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
201 &parent_dev_cgroup->exceptions);
202 dev_cgroup->behavior = parent_dev_cgroup->behavior;
203 mutex_unlock(&devcgroup_mutex);
210 return &dev_cgroup->css;
213 static void devcgroup_css_free(struct cgroup *cgroup)
215 struct dev_cgroup *dev_cgroup;
217 dev_cgroup = cgroup_to_devcgroup(cgroup);
218 mutex_lock(&devcgroup_mutex);
219 dev_exception_clean(dev_cgroup);
220 mutex_unlock(&devcgroup_mutex);
224 #define DEVCG_ALLOW 1
231 static void set_access(char *acc, short access)
234 memset(acc, 0, ACCLEN);
235 if (access & ACC_READ)
237 if (access & ACC_WRITE)
239 if (access & ACC_MKNOD)
243 static char type_to_char(short type)
247 if (type == DEV_CHAR)
249 if (type == DEV_BLOCK)
254 static void set_majmin(char *str, unsigned m)
259 sprintf(str, "%u", m);
262 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
265 struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
266 struct dev_exception_item *ex;
267 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
271 * To preserve the compatibility:
272 * - Only show the "all devices" when the default policy is to allow
273 * - List the exceptions in case the default policy is to deny
274 * This way, the file remains as a "whitelist of devices"
276 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
277 set_access(acc, ACC_MASK);
280 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
283 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
284 set_access(acc, ex->access);
285 set_majmin(maj, ex->major);
286 set_majmin(min, ex->minor);
287 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
297 * may_access - verifies if a new exception is part of what is allowed
298 * by a dev cgroup based on the default policy +
299 * exceptions. This is used to make sure a child cgroup
300 * won't have more privileges than its parent or to
301 * verify if a certain access is allowed.
302 * @dev_cgroup: dev cgroup to be tested against
303 * @refex: new exception
305 static int may_access(struct dev_cgroup *dev_cgroup,
306 struct dev_exception_item *refex)
308 struct dev_exception_item *ex;
311 rcu_lockdep_assert(rcu_read_lock_held() ||
312 lockdep_is_held(&devcgroup_mutex),
313 "device_cgroup::may_access() called without proper synchronization");
315 list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
316 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
318 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
320 if (ex->major != ~0 && ex->major != refex->major)
322 if (ex->minor != ~0 && ex->minor != refex->minor)
324 if (refex->access & (~ex->access))
331 * In two cases we'll consider this new exception valid:
332 * - the dev cgroup has its default policy to allow + exception list:
333 * the new exception should *not* match any of the exceptions
334 * (behavior == DEVCG_DEFAULT_ALLOW, !match)
335 * - the dev cgroup has its default policy to deny + exception list:
336 * the new exception *should* match the exceptions
337 * (behavior == DEVCG_DEFAULT_DENY, match)
339 if ((dev_cgroup->behavior == DEVCG_DEFAULT_DENY) == match)
346 * when adding a new allow rule to a device exception list, the rule
347 * must be allowed in the parent device
349 static int parent_has_perm(struct dev_cgroup *childcg,
350 struct dev_exception_item *ex)
352 struct cgroup *pcg = childcg->css.cgroup->parent;
353 struct dev_cgroup *parent;
357 parent = cgroup_to_devcgroup(pcg);
358 return may_access(parent, ex);
362 * may_allow_all - checks if it's possible to change the behavior to
363 * allow based on parent's rules.
364 * @parent: device cgroup's parent
365 * returns: != 0 in case it's allowed, 0 otherwise
367 static inline int may_allow_all(struct dev_cgroup *parent)
371 return parent->behavior == DEVCG_DEFAULT_ALLOW;
375 * Modify the exception list using allow/deny rules.
376 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
377 * so we can give a container CAP_MKNOD to let it create devices but not
378 * modify the exception list.
379 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
380 * us to also grant CAP_SYS_ADMIN to containers without giving away the
381 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
383 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
384 * new access is only allowed if you're in the top-level cgroup, or your
385 * parent cgroup has the access you're asking for.
387 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
388 int filetype, const char *buffer)
391 char temp[12]; /* 11 + 1 characters needed for a u32 */
393 struct dev_exception_item ex;
394 struct cgroup *p = devcgroup->css.cgroup;
395 struct dev_cgroup *parent = NULL;
397 if (!capable(CAP_SYS_ADMIN))
401 parent = cgroup_to_devcgroup(p->parent);
403 memset(&ex, 0, sizeof(ex));
410 if (!may_allow_all(parent))
412 dev_exception_clean(devcgroup);
413 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
417 rc = dev_exceptions_copy(&devcgroup->exceptions,
418 &parent->exceptions);
423 dev_exception_clean(devcgroup);
424 devcgroup->behavior = DEVCG_DEFAULT_DENY;
446 } else if (isdigit(*b)) {
447 memset(temp, 0, sizeof(temp));
448 for (count = 0; count < sizeof(temp) - 1; count++) {
454 rc = kstrtou32(temp, 10, &ex.major);
468 } else if (isdigit(*b)) {
469 memset(temp, 0, sizeof(temp));
470 for (count = 0; count < sizeof(temp) - 1; count++) {
476 rc = kstrtou32(temp, 10, &ex.minor);
484 for (b++, count = 0; count < 3; count++, b++) {
487 ex.access |= ACC_READ;
490 ex.access |= ACC_WRITE;
493 ex.access |= ACC_MKNOD;
506 if (!parent_has_perm(devcgroup, &ex))
509 * If the default policy is to allow by default, try to remove
510 * an matching exception instead. And be silent about it: we
511 * don't want to break compatibility
513 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
514 dev_exception_rm(devcgroup, &ex);
517 return dev_exception_add(devcgroup, &ex);
520 * If the default policy is to deny by default, try to remove
521 * an matching exception instead. And be silent about it: we
522 * don't want to break compatibility
524 if (devcgroup->behavior == DEVCG_DEFAULT_DENY) {
525 dev_exception_rm(devcgroup, &ex);
528 return dev_exception_add(devcgroup, &ex);
535 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
540 mutex_lock(&devcgroup_mutex);
541 retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
542 cft->private, buffer);
543 mutex_unlock(&devcgroup_mutex);
547 static struct cftype dev_cgroup_files[] = {
550 .write_string = devcgroup_access_write,
551 .private = DEVCG_ALLOW,
555 .write_string = devcgroup_access_write,
556 .private = DEVCG_DENY,
560 .read_seq_string = devcgroup_seq_read,
561 .private = DEVCG_LIST,
566 struct cgroup_subsys devices_subsys = {
568 .can_attach = devcgroup_can_attach,
569 .css_alloc = devcgroup_css_alloc,
570 .css_free = devcgroup_css_free,
571 .subsys_id = devices_subsys_id,
572 .base_cftypes = dev_cgroup_files,
575 * While devices cgroup has the rudimentary hierarchy support which
576 * checks the parent's restriction, it doesn't properly propagates
577 * config changes in ancestors to their descendents. A child
578 * should only be allowed to add more restrictions to the parent's
579 * configuration. Fix it and remove the following.
581 .broken_hierarchy = true,
585 * __devcgroup_check_permission - checks if an inode operation is permitted
586 * @dev_cgroup: the dev cgroup to be tested against
588 * @major: device major number
589 * @minor: device minor number
590 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
592 * returns 0 on success, -EPERM case the operation is not permitted
594 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
597 struct dev_cgroup *dev_cgroup;
598 struct dev_exception_item ex;
601 memset(&ex, 0, sizeof(ex));
608 dev_cgroup = task_devcgroup(current);
609 rc = may_access(dev_cgroup, &ex);
618 int __devcgroup_inode_permission(struct inode *inode, int mask)
620 short type, access = 0;
622 if (S_ISBLK(inode->i_mode))
624 if (S_ISCHR(inode->i_mode))
626 if (mask & MAY_WRITE)
631 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
635 int devcgroup_inode_mknod(int mode, dev_t dev)
639 if (!S_ISBLK(mode) && !S_ISCHR(mode))
647 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),