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 list_for_each_entry(ex, orig, list) {
86 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
89 list_add_tail(&new->list, dest);
95 list_for_each_entry_safe(ex, tmp, dest, list) {
103 * called under devcgroup_mutex
105 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
106 struct dev_exception_item *ex)
108 struct dev_exception_item *excopy, *walk;
110 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
114 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
115 if (walk->type != ex->type)
117 if (walk->major != ex->major)
119 if (walk->minor != ex->minor)
122 walk->access |= ex->access;
128 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
133 * called under devcgroup_mutex
135 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
136 struct dev_exception_item *ex)
138 struct dev_exception_item *walk, *tmp;
140 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
141 if (walk->type != ex->type)
143 if (walk->major != ex->major)
145 if (walk->minor != ex->minor)
148 walk->access &= ~ex->access;
150 list_del_rcu(&walk->list);
151 kfree_rcu(walk, rcu);
157 * dev_exception_clean - frees all entries of the exception list
158 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
160 * called under devcgroup_mutex
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) {
173 * called from kernel/cgroup.c with cgroup_lock() held.
175 static struct cgroup_subsys_state *devcgroup_create(struct cgroup *cgroup)
177 struct dev_cgroup *dev_cgroup, *parent_dev_cgroup;
178 struct cgroup *parent_cgroup;
181 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
183 return ERR_PTR(-ENOMEM);
184 INIT_LIST_HEAD(&dev_cgroup->exceptions);
185 parent_cgroup = cgroup->parent;
187 if (parent_cgroup == NULL)
188 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
190 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
191 mutex_lock(&devcgroup_mutex);
192 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
193 &parent_dev_cgroup->exceptions);
194 dev_cgroup->behavior = parent_dev_cgroup->behavior;
195 mutex_unlock(&devcgroup_mutex);
202 return &dev_cgroup->css;
205 static void devcgroup_destroy(struct cgroup *cgroup)
207 struct dev_cgroup *dev_cgroup;
209 dev_cgroup = cgroup_to_devcgroup(cgroup);
210 dev_exception_clean(dev_cgroup);
214 #define DEVCG_ALLOW 1
221 static void set_access(char *acc, short access)
224 memset(acc, 0, ACCLEN);
225 if (access & ACC_READ)
227 if (access & ACC_WRITE)
229 if (access & ACC_MKNOD)
233 static char type_to_char(short type)
237 if (type == DEV_CHAR)
239 if (type == DEV_BLOCK)
244 static void set_majmin(char *str, unsigned m)
249 sprintf(str, "%u", m);
252 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
255 struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
256 struct dev_exception_item *ex;
257 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
261 * To preserve the compatibility:
262 * - Only show the "all devices" when the default policy is to allow
263 * - List the exceptions in case the default policy is to deny
264 * This way, the file remains as a "whitelist of devices"
266 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
267 set_access(acc, ACC_MASK);
270 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
273 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
274 set_access(acc, ex->access);
275 set_majmin(maj, ex->major);
276 set_majmin(min, ex->minor);
277 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
287 * may_access - verifies if a new exception is part of what is allowed
288 * by a dev cgroup based on the default policy +
289 * exceptions. This is used to make sure a child cgroup
290 * won't have more privileges than its parent or to
291 * verify if a certain access is allowed.
292 * @dev_cgroup: dev cgroup to be tested against
293 * @refex: new exception
295 static int may_access(struct dev_cgroup *dev_cgroup,
296 struct dev_exception_item *refex)
298 struct dev_exception_item *ex;
301 list_for_each_entry(ex, &dev_cgroup->exceptions, list) {
302 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
304 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
306 if (ex->major != ~0 && ex->major != refex->major)
308 if (ex->minor != ~0 && ex->minor != refex->minor)
310 if (refex->access & (~ex->access))
317 * In two cases we'll consider this new exception valid:
318 * - the dev cgroup has its default policy to allow + exception list:
319 * the new exception should *not* match any of the exceptions
320 * (behavior == DEVCG_DEFAULT_ALLOW, !match)
321 * - the dev cgroup has its default policy to deny + exception list:
322 * the new exception *should* match the exceptions
323 * (behavior == DEVCG_DEFAULT_DENY, match)
325 if ((dev_cgroup->behavior == DEVCG_DEFAULT_DENY) == match)
332 * when adding a new allow rule to a device exception list, the rule
333 * must be allowed in the parent device
335 static int parent_has_perm(struct dev_cgroup *childcg,
336 struct dev_exception_item *ex)
338 struct cgroup *pcg = childcg->css.cgroup->parent;
339 struct dev_cgroup *parent;
343 parent = cgroup_to_devcgroup(pcg);
344 return may_access(parent, ex);
348 * Modify the exception list using allow/deny rules.
349 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
350 * so we can give a container CAP_MKNOD to let it create devices but not
351 * modify the exception list.
352 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
353 * us to also grant CAP_SYS_ADMIN to containers without giving away the
354 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
356 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
357 * new access is only allowed if you're in the top-level cgroup, or your
358 * parent cgroup has the access you're asking for.
360 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
361 int filetype, const char *buffer)
366 struct dev_exception_item ex;
368 if (!capable(CAP_SYS_ADMIN))
371 memset(&ex, 0, sizeof(ex));
378 if (!parent_has_perm(devcgroup, &ex))
380 dev_exception_clean(devcgroup);
381 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
384 dev_exception_clean(devcgroup);
385 devcgroup->behavior = DEVCG_DEFAULT_DENY;
407 } else if (isdigit(*b)) {
408 ex.major = simple_strtoul(b, &endp, 10);
421 } else if (isdigit(*b)) {
422 ex.minor = simple_strtoul(b, &endp, 10);
429 for (b++, count = 0; count < 3; count++, b++) {
432 ex.access |= ACC_READ;
435 ex.access |= ACC_WRITE;
438 ex.access |= ACC_MKNOD;
451 if (!parent_has_perm(devcgroup, &ex))
454 * If the default policy is to allow by default, try to remove
455 * an matching exception instead. And be silent about it: we
456 * don't want to break compatibility
458 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
459 dev_exception_rm(devcgroup, &ex);
462 return dev_exception_add(devcgroup, &ex);
465 * If the default policy is to deny by default, try to remove
466 * an matching exception instead. And be silent about it: we
467 * don't want to break compatibility
469 if (devcgroup->behavior == DEVCG_DEFAULT_DENY) {
470 dev_exception_rm(devcgroup, &ex);
473 return dev_exception_add(devcgroup, &ex);
480 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
485 mutex_lock(&devcgroup_mutex);
486 retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
487 cft->private, buffer);
488 mutex_unlock(&devcgroup_mutex);
492 static struct cftype dev_cgroup_files[] = {
495 .write_string = devcgroup_access_write,
496 .private = DEVCG_ALLOW,
500 .write_string = devcgroup_access_write,
501 .private = DEVCG_DENY,
505 .read_seq_string = devcgroup_seq_read,
506 .private = DEVCG_LIST,
511 struct cgroup_subsys devices_subsys = {
513 .can_attach = devcgroup_can_attach,
514 .create = devcgroup_create,
515 .destroy = devcgroup_destroy,
516 .subsys_id = devices_subsys_id,
517 .base_cftypes = dev_cgroup_files,
520 * While devices cgroup has the rudimentary hierarchy support which
521 * checks the parent's restriction, it doesn't properly propagates
522 * config changes in ancestors to their descendents. A child
523 * should only be allowed to add more restrictions to the parent's
524 * configuration. Fix it and remove the following.
526 .broken_hierarchy = true,
530 * __devcgroup_check_permission - checks if an inode operation is permitted
531 * @dev_cgroup: the dev cgroup to be tested against
533 * @major: device major number
534 * @minor: device minor number
535 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
537 * returns 0 on success, -EPERM case the operation is not permitted
539 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
542 struct dev_cgroup *dev_cgroup;
543 struct dev_exception_item ex;
546 memset(&ex, 0, sizeof(ex));
553 dev_cgroup = task_devcgroup(current);
554 rc = may_access(dev_cgroup, &ex);
563 int __devcgroup_inode_permission(struct inode *inode, int mask)
565 short type, access = 0;
567 if (S_ISBLK(inode->i_mode))
569 if (S_ISCHR(inode->i_mode))
571 if (mask & MAY_WRITE)
576 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
580 int devcgroup_inode_mknod(int mode, dev_t dev)
584 if (!S_ISBLK(mode) && !S_ISCHR(mode))
592 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),