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 * whitelist locking rules:
30 * hold devcgroup_mutex for update/read.
31 * hold rcu_read_lock() for read.
34 struct dev_whitelist_item {
38 struct list_head list;
43 struct cgroup_subsys_state css;
44 struct list_head whitelist;
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_whitelist_copy(struct list_head *dest, struct list_head *orig)
80 struct dev_whitelist_item *wh, *tmp, *new;
82 list_for_each_entry(wh, orig, list) {
83 new = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
86 list_add_tail(&new->list, dest);
92 list_for_each_entry_safe(wh, tmp, dest, list) {
99 /* Stupid prototype - don't bother combining existing entries */
101 * called under devcgroup_mutex
103 static int dev_whitelist_add(struct dev_cgroup *dev_cgroup,
104 struct dev_whitelist_item *wh)
106 struct dev_whitelist_item *whcopy, *walk;
108 whcopy = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
112 list_for_each_entry(walk, &dev_cgroup->whitelist, list) {
113 if (walk->type != wh->type)
115 if (walk->major != wh->major)
117 if (walk->minor != wh->minor)
120 walk->access |= wh->access;
126 list_add_tail_rcu(&whcopy->list, &dev_cgroup->whitelist);
131 * called under devcgroup_mutex
133 static void dev_whitelist_rm(struct dev_cgroup *dev_cgroup,
134 struct dev_whitelist_item *wh)
136 struct dev_whitelist_item *walk, *tmp;
138 list_for_each_entry_safe(walk, tmp, &dev_cgroup->whitelist, list) {
139 if (walk->type == DEV_ALL)
141 if (walk->type != wh->type)
143 if (walk->major != ~0 && walk->major != wh->major)
145 if (walk->minor != ~0 && walk->minor != wh->minor)
149 walk->access &= ~wh->access;
151 list_del_rcu(&walk->list);
152 kfree_rcu(walk, rcu);
158 * called from kernel/cgroup.c with cgroup_lock() held.
160 static struct cgroup_subsys_state *devcgroup_create(struct cgroup *cgroup)
162 struct dev_cgroup *dev_cgroup, *parent_dev_cgroup;
163 struct cgroup *parent_cgroup;
166 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
168 return ERR_PTR(-ENOMEM);
169 INIT_LIST_HEAD(&dev_cgroup->whitelist);
170 parent_cgroup = cgroup->parent;
172 if (parent_cgroup == NULL) {
173 struct dev_whitelist_item *wh;
174 wh = kmalloc(sizeof(*wh), GFP_KERNEL);
177 return ERR_PTR(-ENOMEM);
179 wh->minor = wh->major = ~0;
181 wh->access = ACC_MASK;
182 dev_cgroup->deny_all = false;
183 list_add(&wh->list, &dev_cgroup->whitelist);
185 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
186 mutex_lock(&devcgroup_mutex);
187 ret = dev_whitelist_copy(&dev_cgroup->whitelist,
188 &parent_dev_cgroup->whitelist);
189 dev_cgroup->deny_all = parent_dev_cgroup->deny_all;
190 mutex_unlock(&devcgroup_mutex);
197 return &dev_cgroup->css;
200 static void devcgroup_destroy(struct cgroup *cgroup)
202 struct dev_cgroup *dev_cgroup;
203 struct dev_whitelist_item *wh, *tmp;
205 dev_cgroup = cgroup_to_devcgroup(cgroup);
206 list_for_each_entry_safe(wh, tmp, &dev_cgroup->whitelist, list) {
213 #define DEVCG_ALLOW 1
220 static void set_access(char *acc, short access)
223 memset(acc, 0, ACCLEN);
224 if (access & ACC_READ)
226 if (access & ACC_WRITE)
228 if (access & ACC_MKNOD)
232 static char type_to_char(short type)
236 if (type == DEV_CHAR)
238 if (type == DEV_BLOCK)
243 static void set_majmin(char *str, unsigned m)
248 sprintf(str, "%u", m);
251 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
254 struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
255 struct dev_whitelist_item *wh;
256 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
259 list_for_each_entry_rcu(wh, &devcgroup->whitelist, list) {
260 set_access(acc, wh->access);
261 set_majmin(maj, wh->major);
262 set_majmin(min, wh->minor);
263 seq_printf(m, "%c %s:%s %s\n", type_to_char(wh->type),
272 * may_access_whitelist:
273 * does the access granted to dev_cgroup c contain the access
274 * requested in whitelist item refwh.
275 * return 1 if yes, 0 if no.
276 * call with devcgroup_mutex held
278 static int may_access_whitelist(struct dev_cgroup *c,
279 struct dev_whitelist_item *refwh)
281 struct dev_whitelist_item *whitem;
283 list_for_each_entry(whitem, &c->whitelist, list) {
284 if (whitem->type & DEV_ALL)
286 if ((refwh->type & DEV_BLOCK) && !(whitem->type & DEV_BLOCK))
288 if ((refwh->type & DEV_CHAR) && !(whitem->type & DEV_CHAR))
290 if (whitem->major != ~0 && whitem->major != refwh->major)
292 if (whitem->minor != ~0 && whitem->minor != refwh->minor)
294 if (refwh->access & (~whitem->access))
303 * when adding a new allow rule to a device whitelist, the rule
304 * must be allowed in the parent device
306 static int parent_has_perm(struct dev_cgroup *childcg,
307 struct dev_whitelist_item *wh)
309 struct cgroup *pcg = childcg->css.cgroup->parent;
310 struct dev_cgroup *parent;
314 parent = cgroup_to_devcgroup(pcg);
315 return may_access_whitelist(parent, wh);
319 * Modify the whitelist using allow/deny rules.
320 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
321 * so we can give a container CAP_MKNOD to let it create devices but not
322 * modify the whitelist.
323 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
324 * us to also grant CAP_SYS_ADMIN to containers without giving away the
325 * device whitelist controls, but for now we'll stick with CAP_SYS_ADMIN
327 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
328 * new access is only allowed if you're in the top-level cgroup, or your
329 * parent cgroup has the access you're asking for.
331 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
332 int filetype, const char *buffer)
337 struct dev_whitelist_item wh;
339 if (!capable(CAP_SYS_ADMIN))
342 memset(&wh, 0, sizeof(wh));
348 wh.access = ACC_MASK;
368 } else if (isdigit(*b)) {
369 wh.major = simple_strtoul(b, &endp, 10);
382 } else if (isdigit(*b)) {
383 wh.minor = simple_strtoul(b, &endp, 10);
390 for (b++, count = 0; count < 3; count++, b++) {
393 wh.access |= ACC_READ;
396 wh.access |= ACC_WRITE;
399 wh.access |= ACC_MKNOD;
413 if (!parent_has_perm(devcgroup, &wh))
415 devcgroup->deny_all = false;
416 return dev_whitelist_add(devcgroup, &wh);
418 dev_whitelist_rm(devcgroup, &wh);
419 devcgroup->deny_all = true;
427 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
432 mutex_lock(&devcgroup_mutex);
433 retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
434 cft->private, buffer);
435 mutex_unlock(&devcgroup_mutex);
439 static struct cftype dev_cgroup_files[] = {
442 .write_string = devcgroup_access_write,
443 .private = DEVCG_ALLOW,
447 .write_string = devcgroup_access_write,
448 .private = DEVCG_DENY,
452 .read_seq_string = devcgroup_seq_read,
453 .private = DEVCG_LIST,
458 struct cgroup_subsys devices_subsys = {
460 .can_attach = devcgroup_can_attach,
461 .create = devcgroup_create,
462 .destroy = devcgroup_destroy,
463 .subsys_id = devices_subsys_id,
464 .base_cftypes = dev_cgroup_files,
467 * While devices cgroup has the rudimentary hierarchy support which
468 * checks the parent's restriction, it doesn't properly propagates
469 * config changes in ancestors to their descendents. A child
470 * should only be allowed to add more restrictions to the parent's
471 * configuration. Fix it and remove the following.
473 .broken_hierarchy = true,
476 int __devcgroup_inode_permission(struct inode *inode, int mask)
478 struct dev_cgroup *dev_cgroup;
479 struct dev_whitelist_item *wh;
483 dev_cgroup = task_devcgroup(current);
485 list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
486 if (wh->type & DEV_ALL)
488 if ((wh->type & DEV_BLOCK) && !S_ISBLK(inode->i_mode))
490 if ((wh->type & DEV_CHAR) && !S_ISCHR(inode->i_mode))
492 if (wh->major != ~0 && wh->major != imajor(inode))
494 if (wh->minor != ~0 && wh->minor != iminor(inode))
497 if ((mask & MAY_WRITE) && !(wh->access & ACC_WRITE))
499 if ((mask & MAY_READ) && !(wh->access & ACC_READ))
511 int devcgroup_inode_mknod(int mode, dev_t dev)
513 struct dev_cgroup *dev_cgroup;
514 struct dev_whitelist_item *wh;
516 if (!S_ISBLK(mode) && !S_ISCHR(mode))
521 dev_cgroup = task_devcgroup(current);
523 list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
524 if (wh->type & DEV_ALL)
526 if ((wh->type & DEV_BLOCK) && !S_ISBLK(mode))
528 if ((wh->type & DEV_CHAR) && !S_ISCHR(mode))
530 if (wh->major != ~0 && wh->major != MAJOR(dev))
532 if (wh->minor != ~0 && wh->minor != MINOR(dev))
535 if (!(wh->access & ACC_MKNOD))