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;
48 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
50 return container_of(s, struct dev_cgroup, css);
53 static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup)
55 return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id));
58 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
60 return css_to_devcgroup(task_subsys_state(task, devices_subsys_id));
63 struct cgroup_subsys devices_subsys;
65 static int devcgroup_can_attach(struct cgroup *new_cgrp,
66 struct cgroup_taskset *set)
68 struct task_struct *task = cgroup_taskset_first(set);
70 if (current != task && !capable(CAP_SYS_ADMIN))
76 * called under devcgroup_mutex
78 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
80 struct dev_exception_item *ex, *tmp, *new;
82 list_for_each_entry(ex, orig, list) {
83 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
86 list_add_tail(&new->list, dest);
92 list_for_each_entry_safe(ex, tmp, dest, list) {
100 * called under devcgroup_mutex
102 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
103 struct dev_exception_item *ex)
105 struct dev_exception_item *excopy, *walk;
107 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
111 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
112 if (walk->type != ex->type)
114 if (walk->major != ex->major)
116 if (walk->minor != ex->minor)
119 walk->access |= ex->access;
125 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
130 * called under devcgroup_mutex
132 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
133 struct dev_exception_item *ex)
135 struct dev_exception_item *walk, *tmp;
137 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
138 if (walk->type != ex->type)
140 if (walk->major != ex->major)
142 if (walk->minor != ex->minor)
145 walk->access &= ~ex->access;
147 list_del_rcu(&walk->list);
148 kfree_rcu(walk, rcu);
154 * dev_exception_clean - frees all entries of the exception list
155 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
157 * called under devcgroup_mutex
159 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
161 struct dev_exception_item *ex, *tmp;
163 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
170 * called from kernel/cgroup.c with cgroup_lock() held.
172 static struct cgroup_subsys_state *devcgroup_create(struct cgroup *cgroup)
174 struct dev_cgroup *dev_cgroup, *parent_dev_cgroup;
175 struct cgroup *parent_cgroup;
178 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
180 return ERR_PTR(-ENOMEM);
181 INIT_LIST_HEAD(&dev_cgroup->exceptions);
182 parent_cgroup = cgroup->parent;
184 if (parent_cgroup == NULL)
185 dev_cgroup->deny_all = false;
187 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
188 mutex_lock(&devcgroup_mutex);
189 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
190 &parent_dev_cgroup->exceptions);
191 dev_cgroup->deny_all = parent_dev_cgroup->deny_all;
192 mutex_unlock(&devcgroup_mutex);
199 return &dev_cgroup->css;
202 static void devcgroup_destroy(struct cgroup *cgroup)
204 struct dev_cgroup *dev_cgroup;
206 dev_cgroup = cgroup_to_devcgroup(cgroup);
207 dev_exception_clean(dev_cgroup);
211 #define DEVCG_ALLOW 1
218 static void set_access(char *acc, short access)
221 memset(acc, 0, ACCLEN);
222 if (access & ACC_READ)
224 if (access & ACC_WRITE)
226 if (access & ACC_MKNOD)
230 static char type_to_char(short type)
234 if (type == DEV_CHAR)
236 if (type == DEV_BLOCK)
241 static void set_majmin(char *str, unsigned m)
246 sprintf(str, "%u", m);
249 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
252 struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
253 struct dev_exception_item *ex;
254 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
258 * To preserve the compatibility:
259 * - Only show the "all devices" when the default policy is to allow
260 * - List the exceptions in case the default policy is to deny
261 * This way, the file remains as a "whitelist of devices"
263 if (devcgroup->deny_all == false) {
264 set_access(acc, ACC_MASK);
267 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
270 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
271 set_access(acc, ex->access);
272 set_majmin(maj, ex->major);
273 set_majmin(min, ex->minor);
274 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
284 * may_access - verifies if a new exception is part of what is allowed
285 * by a dev cgroup based on the default policy +
286 * exceptions. This is used to make sure a child cgroup
287 * won't have more privileges than its parent or to
288 * verify if a certain access is allowed.
289 * @dev_cgroup: dev cgroup to be tested against
290 * @refex: new exception
292 static int may_access(struct dev_cgroup *dev_cgroup,
293 struct dev_exception_item *refex)
295 struct dev_exception_item *ex;
298 list_for_each_entry(ex, &dev_cgroup->exceptions, list) {
299 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
301 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
303 if (ex->major != ~0 && ex->major != refex->major)
305 if (ex->minor != ~0 && ex->minor != refex->minor)
307 if (refex->access & (~ex->access))
314 * In two cases we'll consider this new exception valid:
315 * - the dev cgroup has its default policy to allow + exception list:
316 * the new exception should *not* match any of the exceptions
317 * (!deny_all, !match)
318 * - the dev cgroup has its default policy to deny + exception list:
319 * the new exception *should* match the exceptions
322 if (dev_cgroup->deny_all == match)
329 * when adding a new allow rule to a device exception list, the rule
330 * must be allowed in the parent device
332 static int parent_has_perm(struct dev_cgroup *childcg,
333 struct dev_exception_item *ex)
335 struct cgroup *pcg = childcg->css.cgroup->parent;
336 struct dev_cgroup *parent;
340 parent = cgroup_to_devcgroup(pcg);
341 return may_access(parent, ex);
345 * Modify the exception list using allow/deny rules.
346 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
347 * so we can give a container CAP_MKNOD to let it create devices but not
348 * modify the exception list.
349 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
350 * us to also grant CAP_SYS_ADMIN to containers without giving away the
351 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
353 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
354 * new access is only allowed if you're in the top-level cgroup, or your
355 * parent cgroup has the access you're asking for.
357 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
358 int filetype, const char *buffer)
363 struct dev_exception_item ex;
365 if (!capable(CAP_SYS_ADMIN))
368 memset(&ex, 0, sizeof(ex));
375 if (!parent_has_perm(devcgroup, &ex))
377 dev_exception_clean(devcgroup);
378 devcgroup->deny_all = false;
381 dev_exception_clean(devcgroup);
382 devcgroup->deny_all = true;
404 } else if (isdigit(*b)) {
405 ex.major = simple_strtoul(b, &endp, 10);
418 } else if (isdigit(*b)) {
419 ex.minor = simple_strtoul(b, &endp, 10);
426 for (b++, count = 0; count < 3; count++, b++) {
429 ex.access |= ACC_READ;
432 ex.access |= ACC_WRITE;
435 ex.access |= ACC_MKNOD;
448 if (!parent_has_perm(devcgroup, &ex))
451 * If the default policy is to allow by default, try to remove
452 * an matching exception instead. And be silent about it: we
453 * don't want to break compatibility
455 if (devcgroup->deny_all == false) {
456 dev_exception_rm(devcgroup, &ex);
459 return dev_exception_add(devcgroup, &ex);
462 * If the default policy is to deny by default, try to remove
463 * an matching exception instead. And be silent about it: we
464 * don't want to break compatibility
466 if (devcgroup->deny_all == true) {
467 dev_exception_rm(devcgroup, &ex);
470 return dev_exception_add(devcgroup, &ex);
477 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
482 mutex_lock(&devcgroup_mutex);
483 retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
484 cft->private, buffer);
485 mutex_unlock(&devcgroup_mutex);
489 static struct cftype dev_cgroup_files[] = {
492 .write_string = devcgroup_access_write,
493 .private = DEVCG_ALLOW,
497 .write_string = devcgroup_access_write,
498 .private = DEVCG_DENY,
502 .read_seq_string = devcgroup_seq_read,
503 .private = DEVCG_LIST,
508 struct cgroup_subsys devices_subsys = {
510 .can_attach = devcgroup_can_attach,
511 .create = devcgroup_create,
512 .destroy = devcgroup_destroy,
513 .subsys_id = devices_subsys_id,
514 .base_cftypes = dev_cgroup_files,
517 * While devices cgroup has the rudimentary hierarchy support which
518 * checks the parent's restriction, it doesn't properly propagates
519 * config changes in ancestors to their descendents. A child
520 * should only be allowed to add more restrictions to the parent's
521 * configuration. Fix it and remove the following.
523 .broken_hierarchy = true,
527 * __devcgroup_check_permission - checks if an inode operation is permitted
528 * @dev_cgroup: the dev cgroup to be tested against
530 * @major: device major number
531 * @minor: device minor number
532 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
534 * returns 0 on success, -EPERM case the operation is not permitted
536 static int __devcgroup_check_permission(struct dev_cgroup *dev_cgroup,
537 short type, u32 major, u32 minor,
540 struct dev_exception_item ex;
543 memset(&ex, 0, sizeof(ex));
550 rc = may_access(dev_cgroup, &ex);
559 int __devcgroup_inode_permission(struct inode *inode, int mask)
561 struct dev_cgroup *dev_cgroup = task_devcgroup(current);
562 short type, access = 0;
564 if (S_ISBLK(inode->i_mode))
566 if (S_ISCHR(inode->i_mode))
568 if (mask & MAY_WRITE)
573 return __devcgroup_check_permission(dev_cgroup, type, imajor(inode),
574 iminor(inode), access);
577 int devcgroup_inode_mknod(int mode, dev_t dev)
579 struct dev_cgroup *dev_cgroup = task_devcgroup(current);
582 if (!S_ISBLK(mode) && !S_ISCHR(mode))
590 return __devcgroup_check_permission(dev_cgroup, type, MAJOR(dev),
591 MINOR(dev), ACC_MKNOD);