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);
162 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
164 struct dev_exception_item *ex, *tmp;
166 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
167 list_del_rcu(&ex->list);
173 * dev_exception_clean - frees all entries of the exception list
174 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
176 * called under devcgroup_mutex
178 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
180 lockdep_assert_held(&devcgroup_mutex);
182 __dev_exception_clean(dev_cgroup);
186 * called from kernel/cgroup.c with cgroup_lock() held.
188 static struct cgroup_subsys_state *devcgroup_css_alloc(struct cgroup *cgroup)
190 struct dev_cgroup *dev_cgroup, *parent_dev_cgroup;
191 struct cgroup *parent_cgroup;
194 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
196 return ERR_PTR(-ENOMEM);
197 INIT_LIST_HEAD(&dev_cgroup->exceptions);
198 parent_cgroup = cgroup->parent;
200 if (parent_cgroup == NULL)
201 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
203 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
204 mutex_lock(&devcgroup_mutex);
205 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
206 &parent_dev_cgroup->exceptions);
207 dev_cgroup->behavior = parent_dev_cgroup->behavior;
208 mutex_unlock(&devcgroup_mutex);
215 return &dev_cgroup->css;
218 static void devcgroup_css_free(struct cgroup *cgroup)
220 struct dev_cgroup *dev_cgroup;
222 dev_cgroup = cgroup_to_devcgroup(cgroup);
223 __dev_exception_clean(dev_cgroup);
227 #define DEVCG_ALLOW 1
234 static void set_access(char *acc, short access)
237 memset(acc, 0, ACCLEN);
238 if (access & ACC_READ)
240 if (access & ACC_WRITE)
242 if (access & ACC_MKNOD)
246 static char type_to_char(short type)
250 if (type == DEV_CHAR)
252 if (type == DEV_BLOCK)
257 static void set_majmin(char *str, unsigned m)
262 sprintf(str, "%u", m);
265 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
268 struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
269 struct dev_exception_item *ex;
270 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
274 * To preserve the compatibility:
275 * - Only show the "all devices" when the default policy is to allow
276 * - List the exceptions in case the default policy is to deny
277 * This way, the file remains as a "whitelist of devices"
279 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
280 set_access(acc, ACC_MASK);
283 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
286 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
287 set_access(acc, ex->access);
288 set_majmin(maj, ex->major);
289 set_majmin(min, ex->minor);
290 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
300 * may_access - verifies if a new exception is part of what is allowed
301 * by a dev cgroup based on the default policy +
302 * exceptions. This is used to make sure a child cgroup
303 * won't have more privileges than its parent or to
304 * verify if a certain access is allowed.
305 * @dev_cgroup: dev cgroup to be tested against
306 * @refex: new exception
308 static int may_access(struct dev_cgroup *dev_cgroup,
309 struct dev_exception_item *refex)
311 struct dev_exception_item *ex;
314 rcu_lockdep_assert(rcu_read_lock_held() ||
315 lockdep_is_held(&devcgroup_mutex),
316 "device_cgroup::may_access() called without proper synchronization");
318 list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
319 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
321 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
323 if (ex->major != ~0 && ex->major != refex->major)
325 if (ex->minor != ~0 && ex->minor != refex->minor)
327 if (refex->access & (~ex->access))
334 * In two cases we'll consider this new exception valid:
335 * - the dev cgroup has its default policy to allow + exception list:
336 * the new exception should *not* match any of the exceptions
337 * (behavior == DEVCG_DEFAULT_ALLOW, !match)
338 * - the dev cgroup has its default policy to deny + exception list:
339 * the new exception *should* match the exceptions
340 * (behavior == DEVCG_DEFAULT_DENY, match)
342 if ((dev_cgroup->behavior == DEVCG_DEFAULT_DENY) == match)
349 * when adding a new allow rule to a device exception list, the rule
350 * must be allowed in the parent device
352 static int parent_has_perm(struct dev_cgroup *childcg,
353 struct dev_exception_item *ex)
355 struct cgroup *pcg = childcg->css.cgroup->parent;
356 struct dev_cgroup *parent;
360 parent = cgroup_to_devcgroup(pcg);
361 return may_access(parent, ex);
365 * may_allow_all - checks if it's possible to change the behavior to
366 * allow based on parent's rules.
367 * @parent: device cgroup's parent
368 * returns: != 0 in case it's allowed, 0 otherwise
370 static inline int may_allow_all(struct dev_cgroup *parent)
374 return parent->behavior == DEVCG_DEFAULT_ALLOW;
378 * Modify the exception list using allow/deny rules.
379 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
380 * so we can give a container CAP_MKNOD to let it create devices but not
381 * modify the exception list.
382 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
383 * us to also grant CAP_SYS_ADMIN to containers without giving away the
384 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
386 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
387 * new access is only allowed if you're in the top-level cgroup, or your
388 * parent cgroup has the access you're asking for.
390 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
391 int filetype, const char *buffer)
394 char temp[12]; /* 11 + 1 characters needed for a u32 */
396 struct dev_exception_item ex;
397 struct cgroup *p = devcgroup->css.cgroup;
398 struct dev_cgroup *parent = NULL;
400 if (!capable(CAP_SYS_ADMIN))
404 parent = cgroup_to_devcgroup(p->parent);
406 memset(&ex, 0, sizeof(ex));
413 if (!may_allow_all(parent))
415 dev_exception_clean(devcgroup);
416 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
420 rc = dev_exceptions_copy(&devcgroup->exceptions,
421 &parent->exceptions);
426 dev_exception_clean(devcgroup);
427 devcgroup->behavior = DEVCG_DEFAULT_DENY;
449 } else if (isdigit(*b)) {
450 memset(temp, 0, sizeof(temp));
451 for (count = 0; count < sizeof(temp) - 1; count++) {
457 rc = kstrtou32(temp, 10, &ex.major);
471 } else if (isdigit(*b)) {
472 memset(temp, 0, sizeof(temp));
473 for (count = 0; count < sizeof(temp) - 1; count++) {
479 rc = kstrtou32(temp, 10, &ex.minor);
487 for (b++, count = 0; count < 3; count++, b++) {
490 ex.access |= ACC_READ;
493 ex.access |= ACC_WRITE;
496 ex.access |= ACC_MKNOD;
509 if (!parent_has_perm(devcgroup, &ex))
512 * If the default policy is to allow by default, try to remove
513 * an matching exception instead. And be silent about it: we
514 * don't want to break compatibility
516 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
517 dev_exception_rm(devcgroup, &ex);
520 return dev_exception_add(devcgroup, &ex);
523 * If the default policy is to deny by default, try to remove
524 * an matching exception instead. And be silent about it: we
525 * don't want to break compatibility
527 if (devcgroup->behavior == DEVCG_DEFAULT_DENY) {
528 dev_exception_rm(devcgroup, &ex);
531 return dev_exception_add(devcgroup, &ex);
538 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
543 mutex_lock(&devcgroup_mutex);
544 retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
545 cft->private, buffer);
546 mutex_unlock(&devcgroup_mutex);
550 static struct cftype dev_cgroup_files[] = {
553 .write_string = devcgroup_access_write,
554 .private = DEVCG_ALLOW,
558 .write_string = devcgroup_access_write,
559 .private = DEVCG_DENY,
563 .read_seq_string = devcgroup_seq_read,
564 .private = DEVCG_LIST,
569 struct cgroup_subsys devices_subsys = {
571 .can_attach = devcgroup_can_attach,
572 .css_alloc = devcgroup_css_alloc,
573 .css_free = devcgroup_css_free,
574 .subsys_id = devices_subsys_id,
575 .base_cftypes = dev_cgroup_files,
578 * While devices cgroup has the rudimentary hierarchy support which
579 * checks the parent's restriction, it doesn't properly propagates
580 * config changes in ancestors to their descendents. A child
581 * should only be allowed to add more restrictions to the parent's
582 * configuration. Fix it and remove the following.
584 .broken_hierarchy = true,
588 * __devcgroup_check_permission - checks if an inode operation is permitted
589 * @dev_cgroup: the dev cgroup to be tested against
591 * @major: device major number
592 * @minor: device minor number
593 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
595 * returns 0 on success, -EPERM case the operation is not permitted
597 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
600 struct dev_cgroup *dev_cgroup;
601 struct dev_exception_item ex;
604 memset(&ex, 0, sizeof(ex));
611 dev_cgroup = task_devcgroup(current);
612 rc = may_access(dev_cgroup, &ex);
621 int __devcgroup_inode_permission(struct inode *inode, int mask)
623 short type, access = 0;
625 if (S_ISBLK(inode->i_mode))
627 if (S_ISCHR(inode->i_mode))
629 if (mask & MAY_WRITE)
634 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
638 int devcgroup_inode_mknod(int mode, dev_t dev)
642 if (!S_ISBLK(mode) && !S_ISCHR(mode))
650 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),