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 container_of(s, struct dev_cgroup, css);
59 static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup)
61 return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id));
64 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
66 return css_to_devcgroup(task_subsys_state(task, devices_subsys_id));
69 struct cgroup_subsys devices_subsys;
71 static int devcgroup_can_attach(struct cgroup *new_cgrp,
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);
189 * called from kernel/cgroup.c with cgroup_lock() held.
191 static struct cgroup_subsys_state *devcgroup_css_alloc(struct cgroup *cgroup)
193 struct dev_cgroup *dev_cgroup, *parent_dev_cgroup;
194 struct cgroup *parent_cgroup;
197 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
199 return ERR_PTR(-ENOMEM);
200 INIT_LIST_HEAD(&dev_cgroup->exceptions);
201 parent_cgroup = cgroup->parent;
203 if (parent_cgroup == NULL)
204 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
206 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
207 mutex_lock(&devcgroup_mutex);
208 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
209 &parent_dev_cgroup->exceptions);
210 dev_cgroup->behavior = parent_dev_cgroup->behavior;
211 mutex_unlock(&devcgroup_mutex);
218 return &dev_cgroup->css;
221 static void devcgroup_css_free(struct cgroup *cgroup)
223 struct dev_cgroup *dev_cgroup;
225 dev_cgroup = cgroup_to_devcgroup(cgroup);
226 __dev_exception_clean(dev_cgroup);
230 #define DEVCG_ALLOW 1
237 static void set_access(char *acc, short access)
240 memset(acc, 0, ACCLEN);
241 if (access & ACC_READ)
243 if (access & ACC_WRITE)
245 if (access & ACC_MKNOD)
249 static char type_to_char(short type)
253 if (type == DEV_CHAR)
255 if (type == DEV_BLOCK)
260 static void set_majmin(char *str, unsigned m)
265 sprintf(str, "%u", m);
268 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
271 struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
272 struct dev_exception_item *ex;
273 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
277 * To preserve the compatibility:
278 * - Only show the "all devices" when the default policy is to allow
279 * - List the exceptions in case the default policy is to deny
280 * This way, the file remains as a "whitelist of devices"
282 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
283 set_access(acc, ACC_MASK);
286 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
289 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
290 set_access(acc, ex->access);
291 set_majmin(maj, ex->major);
292 set_majmin(min, ex->minor);
293 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
303 * may_access - verifies if a new exception is part of what is allowed
304 * by a dev cgroup based on the default policy +
305 * exceptions. This is used to make sure a child cgroup
306 * won't have more privileges than its parent or to
307 * verify if a certain access is allowed.
308 * @dev_cgroup: dev cgroup to be tested against
309 * @refex: new exception
310 * @behavior: behavior of the exception
312 static bool may_access(struct dev_cgroup *dev_cgroup,
313 struct dev_exception_item *refex,
314 enum devcg_behavior behavior)
316 struct dev_exception_item *ex;
319 rcu_lockdep_assert(rcu_read_lock_held() ||
320 lockdep_is_held(&devcgroup_mutex),
321 "device_cgroup::may_access() called without proper synchronization");
323 list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
324 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
326 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
328 if (ex->major != ~0 && ex->major != refex->major)
330 if (ex->minor != ~0 && ex->minor != refex->minor)
332 if (refex->access & (~ex->access))
338 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
339 if (behavior == DEVCG_DEFAULT_ALLOW) {
340 /* the exception will deny access to certain devices */
343 /* the exception will allow access to certain devices */
346 * a new exception allowing access shouldn't
347 * match an parent's exception
353 /* only behavior == DEVCG_DEFAULT_DENY allowed here */
355 /* parent has an exception that matches the proposed */
365 * when adding a new allow rule to a device exception list, the rule
366 * must be allowed in the parent device
368 static int parent_has_perm(struct dev_cgroup *childcg,
369 struct dev_exception_item *ex)
371 struct cgroup *pcg = childcg->css.cgroup->parent;
372 struct dev_cgroup *parent;
376 parent = cgroup_to_devcgroup(pcg);
377 return may_access(parent, ex, childcg->behavior);
381 * may_allow_all - checks if it's possible to change the behavior to
382 * allow based on parent's rules.
383 * @parent: device cgroup's parent
384 * returns: != 0 in case it's allowed, 0 otherwise
386 static inline int may_allow_all(struct dev_cgroup *parent)
390 return parent->behavior == DEVCG_DEFAULT_ALLOW;
394 * Modify the exception list using allow/deny rules.
395 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
396 * so we can give a container CAP_MKNOD to let it create devices but not
397 * modify the exception list.
398 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
399 * us to also grant CAP_SYS_ADMIN to containers without giving away the
400 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
402 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
403 * new access is only allowed if you're in the top-level cgroup, or your
404 * parent cgroup has the access you're asking for.
406 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
407 int filetype, const char *buffer)
410 char temp[12]; /* 11 + 1 characters needed for a u32 */
412 struct dev_exception_item ex;
413 struct cgroup *p = devcgroup->css.cgroup;
414 struct dev_cgroup *parent = NULL;
416 if (!capable(CAP_SYS_ADMIN))
420 parent = cgroup_to_devcgroup(p->parent);
422 memset(&ex, 0, sizeof(ex));
429 if (!may_allow_all(parent))
431 dev_exception_clean(devcgroup);
432 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
436 rc = dev_exceptions_copy(&devcgroup->exceptions,
437 &parent->exceptions);
442 dev_exception_clean(devcgroup);
443 devcgroup->behavior = DEVCG_DEFAULT_DENY;
465 } else if (isdigit(*b)) {
466 memset(temp, 0, sizeof(temp));
467 for (count = 0; count < sizeof(temp) - 1; count++) {
473 rc = kstrtou32(temp, 10, &ex.major);
487 } else if (isdigit(*b)) {
488 memset(temp, 0, sizeof(temp));
489 for (count = 0; count < sizeof(temp) - 1; count++) {
495 rc = kstrtou32(temp, 10, &ex.minor);
503 for (b++, count = 0; count < 3; count++, b++) {
506 ex.access |= ACC_READ;
509 ex.access |= ACC_WRITE;
512 ex.access |= ACC_MKNOD;
525 if (!parent_has_perm(devcgroup, &ex))
528 * If the default policy is to allow by default, try to remove
529 * an matching exception instead. And be silent about it: we
530 * don't want to break compatibility
532 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
533 dev_exception_rm(devcgroup, &ex);
536 return dev_exception_add(devcgroup, &ex);
539 * If the default policy is to deny by default, try to remove
540 * an matching exception instead. And be silent about it: we
541 * don't want to break compatibility
543 if (devcgroup->behavior == DEVCG_DEFAULT_DENY) {
544 dev_exception_rm(devcgroup, &ex);
547 return dev_exception_add(devcgroup, &ex);
554 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
559 mutex_lock(&devcgroup_mutex);
560 retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
561 cft->private, buffer);
562 mutex_unlock(&devcgroup_mutex);
566 static struct cftype dev_cgroup_files[] = {
569 .write_string = devcgroup_access_write,
570 .private = DEVCG_ALLOW,
574 .write_string = devcgroup_access_write,
575 .private = DEVCG_DENY,
579 .read_seq_string = devcgroup_seq_read,
580 .private = DEVCG_LIST,
585 struct cgroup_subsys devices_subsys = {
587 .can_attach = devcgroup_can_attach,
588 .css_alloc = devcgroup_css_alloc,
589 .css_free = devcgroup_css_free,
590 .subsys_id = devices_subsys_id,
591 .base_cftypes = dev_cgroup_files,
594 * While devices cgroup has the rudimentary hierarchy support which
595 * checks the parent's restriction, it doesn't properly propagates
596 * config changes in ancestors to their descendents. A child
597 * should only be allowed to add more restrictions to the parent's
598 * configuration. Fix it and remove the following.
600 .broken_hierarchy = true,
604 * __devcgroup_check_permission - checks if an inode operation is permitted
605 * @dev_cgroup: the dev cgroup to be tested against
607 * @major: device major number
608 * @minor: device minor number
609 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
611 * returns 0 on success, -EPERM case the operation is not permitted
613 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
616 struct dev_cgroup *dev_cgroup;
617 struct dev_exception_item ex;
620 memset(&ex, 0, sizeof(ex));
627 dev_cgroup = task_devcgroup(current);
628 rc = may_access(dev_cgroup, &ex, dev_cgroup->behavior);
637 int __devcgroup_inode_permission(struct inode *inode, int mask)
639 short type, access = 0;
641 if (S_ISBLK(inode->i_mode))
643 if (S_ISCHR(inode->i_mode))
645 if (mask & MAY_WRITE)
650 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
654 int devcgroup_inode_mknod(int mode, dev_t dev)
658 if (!S_ISBLK(mode) && !S_ISCHR(mode))
666 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),