2 * dev_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>
17 #define ACC_MASK (ACC_MKNOD | ACC_READ | ACC_WRITE)
21 #define DEV_ALL 4 /* this represents all devices */
24 * whitelist locking rules:
25 * cgroup_lock() cannot be taken under dev_cgroup->lock.
26 * dev_cgroup->lock can be taken with or without cgroup_lock().
28 * modifications always require cgroup_lock
29 * modifications to a list which is visible require the
30 * dev_cgroup->lock *and* cgroup_lock()
31 * walking the list requires dev_cgroup->lock or cgroup_lock().
33 * reasoning: dev_whitelist_copy() needs to kmalloc, so needs
34 * a mutex, which the cgroup_lock() is. Since modifying
35 * a visible list requires both locks, either lock can be
36 * taken for walking the list.
39 struct dev_whitelist_item {
43 struct list_head list;
48 struct cgroup_subsys_state css;
49 struct list_head whitelist;
53 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
55 return container_of(s, struct dev_cgroup, css);
58 static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup)
60 return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id));
63 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
65 return css_to_devcgroup(task_subsys_state(task, devices_subsys_id));
68 struct cgroup_subsys devices_subsys;
70 static int devcgroup_can_attach(struct cgroup_subsys *ss,
71 struct cgroup *new_cgroup, struct task_struct *task)
73 if (current != task && !capable(CAP_SYS_ADMIN))
80 * called under cgroup_lock()
82 static int dev_whitelist_copy(struct list_head *dest, struct list_head *orig)
84 struct dev_whitelist_item *wh, *tmp, *new;
86 list_for_each_entry(wh, orig, list) {
87 new = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
90 list_add_tail(&new->list, dest);
96 list_for_each_entry_safe(wh, tmp, dest, list) {
103 /* Stupid prototype - don't bother combining existing entries */
105 * called under cgroup_lock()
106 * since the list is visible to other tasks, we need the spinlock also
108 static int dev_whitelist_add(struct dev_cgroup *dev_cgroup,
109 struct dev_whitelist_item *wh)
111 struct dev_whitelist_item *whcopy, *walk;
113 whcopy = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
117 spin_lock(&dev_cgroup->lock);
118 list_for_each_entry(walk, &dev_cgroup->whitelist, list) {
119 if (walk->type != wh->type)
121 if (walk->major != wh->major)
123 if (walk->minor != wh->minor)
126 walk->access |= wh->access;
132 list_add_tail_rcu(&whcopy->list, &dev_cgroup->whitelist);
133 spin_unlock(&dev_cgroup->lock);
137 static void whitelist_item_free(struct rcu_head *rcu)
139 struct dev_whitelist_item *item;
141 item = container_of(rcu, struct dev_whitelist_item, rcu);
146 * called under cgroup_lock()
147 * since the list is visible to other tasks, we need the spinlock also
149 static void dev_whitelist_rm(struct dev_cgroup *dev_cgroup,
150 struct dev_whitelist_item *wh)
152 struct dev_whitelist_item *walk, *tmp;
154 spin_lock(&dev_cgroup->lock);
155 list_for_each_entry_safe(walk, tmp, &dev_cgroup->whitelist, list) {
156 if (walk->type == DEV_ALL)
158 if (walk->type != wh->type)
160 if (walk->major != ~0 && walk->major != wh->major)
162 if (walk->minor != ~0 && walk->minor != wh->minor)
166 walk->access &= ~wh->access;
168 list_del_rcu(&walk->list);
169 call_rcu(&walk->rcu, whitelist_item_free);
172 spin_unlock(&dev_cgroup->lock);
176 * called from kernel/cgroup.c with cgroup_lock() held.
178 static struct cgroup_subsys_state *devcgroup_create(struct cgroup_subsys *ss,
179 struct cgroup *cgroup)
181 struct dev_cgroup *dev_cgroup, *parent_dev_cgroup;
182 struct cgroup *parent_cgroup;
185 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
187 return ERR_PTR(-ENOMEM);
188 INIT_LIST_HEAD(&dev_cgroup->whitelist);
189 parent_cgroup = cgroup->parent;
191 if (parent_cgroup == NULL) {
192 struct dev_whitelist_item *wh;
193 wh = kmalloc(sizeof(*wh), GFP_KERNEL);
196 return ERR_PTR(-ENOMEM);
198 wh->minor = wh->major = ~0;
200 wh->access = ACC_MASK;
201 list_add(&wh->list, &dev_cgroup->whitelist);
203 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
204 ret = dev_whitelist_copy(&dev_cgroup->whitelist,
205 &parent_dev_cgroup->whitelist);
212 spin_lock_init(&dev_cgroup->lock);
213 return &dev_cgroup->css;
216 static void devcgroup_destroy(struct cgroup_subsys *ss,
217 struct cgroup *cgroup)
219 struct dev_cgroup *dev_cgroup;
220 struct dev_whitelist_item *wh, *tmp;
222 dev_cgroup = cgroup_to_devcgroup(cgroup);
223 list_for_each_entry_safe(wh, tmp, &dev_cgroup->whitelist, list) {
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_whitelist_item *wh;
273 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
276 list_for_each_entry_rcu(wh, &devcgroup->whitelist, list) {
277 set_access(acc, wh->access);
278 set_majmin(maj, wh->major);
279 set_majmin(min, wh->minor);
280 seq_printf(m, "%c %s:%s %s\n", type_to_char(wh->type),
289 * may_access_whitelist:
290 * does the access granted to dev_cgroup c contain the access
291 * requested in whitelist item refwh.
292 * return 1 if yes, 0 if no.
293 * call with c->lock held
295 static int may_access_whitelist(struct dev_cgroup *c,
296 struct dev_whitelist_item *refwh)
298 struct dev_whitelist_item *whitem;
300 list_for_each_entry(whitem, &c->whitelist, list) {
301 if (whitem->type & DEV_ALL)
303 if ((refwh->type & DEV_BLOCK) && !(whitem->type & DEV_BLOCK))
305 if ((refwh->type & DEV_CHAR) && !(whitem->type & DEV_CHAR))
307 if (whitem->major != ~0 && whitem->major != refwh->major)
309 if (whitem->minor != ~0 && whitem->minor != refwh->minor)
311 if (refwh->access & (~whitem->access))
320 * when adding a new allow rule to a device whitelist, the rule
321 * must be allowed in the parent device
323 static int parent_has_perm(struct dev_cgroup *childcg,
324 struct dev_whitelist_item *wh)
326 struct cgroup *pcg = childcg->css.cgroup->parent;
327 struct dev_cgroup *parent;
332 parent = cgroup_to_devcgroup(pcg);
333 spin_lock(&parent->lock);
334 ret = may_access_whitelist(parent, wh);
335 spin_unlock(&parent->lock);
340 * Modify the whitelist using allow/deny rules.
341 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
342 * so we can give a container CAP_MKNOD to let it create devices but not
343 * modify the whitelist.
344 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
345 * us to also grant CAP_SYS_ADMIN to containers without giving away the
346 * device whitelist controls, but for now we'll stick with CAP_SYS_ADMIN
348 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
349 * new access is only allowed if you're in the top-level cgroup, or your
350 * parent cgroup has the access you're asking for.
352 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
353 int filetype, const char *buffer)
355 struct dev_cgroup *cur_devcgroup;
358 int retval = 0, count;
359 struct dev_whitelist_item wh;
361 if (!capable(CAP_SYS_ADMIN))
364 cur_devcgroup = task_devcgroup(current);
366 memset(&wh, 0, sizeof(wh));
372 wh.access = ACC_MASK;
392 } else if (isdigit(*b)) {
393 wh.major = simple_strtoul(b, &endp, 10);
406 } else if (isdigit(*b)) {
407 wh.minor = simple_strtoul(b, &endp, 10);
414 for (b++, count = 0; count < 3; count++, b++) {
417 wh.access |= ACC_READ;
420 wh.access |= ACC_WRITE;
423 wh.access |= ACC_MKNOD;
438 if (!parent_has_perm(devcgroup, &wh))
440 return dev_whitelist_add(devcgroup, &wh);
442 dev_whitelist_rm(devcgroup, &wh);
450 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
454 if (!cgroup_lock_live_group(cgrp))
456 retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
457 cft->private, buffer);
462 static struct cftype dev_cgroup_files[] = {
465 .write_string = devcgroup_access_write,
466 .private = DEVCG_ALLOW,
470 .write_string = devcgroup_access_write,
471 .private = DEVCG_DENY,
475 .read_seq_string = devcgroup_seq_read,
476 .private = DEVCG_LIST,
480 static int devcgroup_populate(struct cgroup_subsys *ss,
481 struct cgroup *cgroup)
483 return cgroup_add_files(cgroup, ss, dev_cgroup_files,
484 ARRAY_SIZE(dev_cgroup_files));
487 struct cgroup_subsys devices_subsys = {
489 .can_attach = devcgroup_can_attach,
490 .create = devcgroup_create,
491 .destroy = devcgroup_destroy,
492 .populate = devcgroup_populate,
493 .subsys_id = devices_subsys_id,
496 int devcgroup_inode_permission(struct inode *inode, int mask)
498 struct dev_cgroup *dev_cgroup;
499 struct dev_whitelist_item *wh;
501 dev_t device = inode->i_rdev;
504 if (!S_ISBLK(inode->i_mode) && !S_ISCHR(inode->i_mode))
509 dev_cgroup = task_devcgroup(current);
511 list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
512 if (wh->type & DEV_ALL)
514 if ((wh->type & DEV_BLOCK) && !S_ISBLK(inode->i_mode))
516 if ((wh->type & DEV_CHAR) && !S_ISCHR(inode->i_mode))
518 if (wh->major != ~0 && wh->major != imajor(inode))
520 if (wh->minor != ~0 && wh->minor != iminor(inode))
523 if ((mask & MAY_WRITE) && !(wh->access & ACC_WRITE))
525 if ((mask & MAY_READ) && !(wh->access & ACC_READ))
536 int devcgroup_inode_mknod(int mode, dev_t dev)
538 struct dev_cgroup *dev_cgroup;
539 struct dev_whitelist_item *wh;
543 dev_cgroup = task_devcgroup(current);
545 list_for_each_entry(wh, &dev_cgroup->whitelist, list) {
546 if (wh->type & DEV_ALL)
548 if ((wh->type & DEV_BLOCK) && !S_ISBLK(mode))
550 if ((wh->type & DEV_CHAR) && !S_ISCHR(mode))
552 if (wh->major != ~0 && wh->major != MAJOR(dev))
554 if (wh->minor != ~0 && wh->minor != MINOR(dev))
557 if (!(wh->access & ACC_MKNOD))