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
;
47 struct cgroup_subsys_state css
;
48 struct list_head whitelist
;
52 static inline struct dev_cgroup
*css_to_devcgroup(struct cgroup_subsys_state
*s
)
54 return container_of(s
, struct dev_cgroup
, css
);
57 static inline struct dev_cgroup
*cgroup_to_devcgroup(struct cgroup
*cgroup
)
59 return css_to_devcgroup(cgroup_subsys_state(cgroup
, devices_subsys_id
));
62 struct cgroup_subsys devices_subsys
;
64 static int devcgroup_can_attach(struct cgroup_subsys
*ss
,
65 struct cgroup
*new_cgroup
, struct task_struct
*task
)
67 if (current
!= task
&& !capable(CAP_SYS_ADMIN
))
74 * called under cgroup_lock()
76 static int dev_whitelist_copy(struct list_head
*dest
, struct list_head
*orig
)
78 struct dev_whitelist_item
*wh
, *tmp
, *new;
80 list_for_each_entry(wh
, orig
, list
) {
81 new = kmalloc(sizeof(*wh
), GFP_KERNEL
);
84 new->major
= wh
->major
;
85 new->minor
= wh
->minor
;
87 new->access
= wh
->access
;
88 list_add_tail(&new->list
, dest
);
94 list_for_each_entry_safe(wh
, tmp
, dest
, list
) {
101 /* Stupid prototype - don't bother combining existing entries */
103 * called under cgroup_lock()
104 * since the list is visible to other tasks, we need the spinlock also
106 static int dev_whitelist_add(struct dev_cgroup
*dev_cgroup
,
107 struct dev_whitelist_item
*wh
)
109 struct dev_whitelist_item
*whcopy
;
111 whcopy
= kmalloc(sizeof(*whcopy
), GFP_KERNEL
);
115 memcpy(whcopy
, wh
, sizeof(*whcopy
));
116 spin_lock(&dev_cgroup
->lock
);
117 list_add_tail(&whcopy
->list
, &dev_cgroup
->whitelist
);
118 spin_unlock(&dev_cgroup
->lock
);
123 * called under cgroup_lock()
124 * since the list is visible to other tasks, we need the spinlock also
126 static void dev_whitelist_rm(struct dev_cgroup
*dev_cgroup
,
127 struct dev_whitelist_item
*wh
)
129 struct dev_whitelist_item
*walk
, *tmp
;
131 spin_lock(&dev_cgroup
->lock
);
132 list_for_each_entry_safe(walk
, tmp
, &dev_cgroup
->whitelist
, list
) {
133 if (walk
->type
== DEV_ALL
)
135 if (walk
->type
!= wh
->type
)
137 if (walk
->major
!= ~0 && walk
->major
!= wh
->major
)
139 if (walk
->minor
!= ~0 && walk
->minor
!= wh
->minor
)
143 walk
->access
&= ~wh
->access
;
145 list_del(&walk
->list
);
149 spin_unlock(&dev_cgroup
->lock
);
153 * called from kernel/cgroup.c with cgroup_lock() held.
155 static struct cgroup_subsys_state
*devcgroup_create(struct cgroup_subsys
*ss
,
156 struct cgroup
*cgroup
)
158 struct dev_cgroup
*dev_cgroup
, *parent_dev_cgroup
;
159 struct cgroup
*parent_cgroup
;
162 dev_cgroup
= kzalloc(sizeof(*dev_cgroup
), GFP_KERNEL
);
164 return ERR_PTR(-ENOMEM
);
165 INIT_LIST_HEAD(&dev_cgroup
->whitelist
);
166 parent_cgroup
= cgroup
->parent
;
168 if (parent_cgroup
== NULL
) {
169 struct dev_whitelist_item
*wh
;
170 wh
= kmalloc(sizeof(*wh
), GFP_KERNEL
);
173 return ERR_PTR(-ENOMEM
);
175 wh
->minor
= wh
->major
= ~0;
177 wh
->access
= ACC_MKNOD
| ACC_READ
| ACC_WRITE
;
178 list_add(&wh
->list
, &dev_cgroup
->whitelist
);
180 parent_dev_cgroup
= cgroup_to_devcgroup(parent_cgroup
);
181 ret
= dev_whitelist_copy(&dev_cgroup
->whitelist
,
182 &parent_dev_cgroup
->whitelist
);
189 spin_lock_init(&dev_cgroup
->lock
);
190 return &dev_cgroup
->css
;
193 static void devcgroup_destroy(struct cgroup_subsys
*ss
,
194 struct cgroup
*cgroup
)
196 struct dev_cgroup
*dev_cgroup
;
197 struct dev_whitelist_item
*wh
, *tmp
;
199 dev_cgroup
= cgroup_to_devcgroup(cgroup
);
200 list_for_each_entry_safe(wh
, tmp
, &dev_cgroup
->whitelist
, list
) {
207 #define DEVCG_ALLOW 1
214 static void set_access(char *acc
, short access
)
217 memset(acc
, 0, ACCLEN
);
218 if (access
& ACC_READ
)
220 if (access
& ACC_WRITE
)
222 if (access
& ACC_MKNOD
)
226 static char type_to_char(short type
)
230 if (type
== DEV_CHAR
)
232 if (type
== DEV_BLOCK
)
237 static void set_majmin(char *str
, unsigned m
)
239 memset(str
, 0, MAJMINLEN
);
243 snprintf(str
, MAJMINLEN
, "%d", m
);
246 static int devcgroup_seq_read(struct cgroup
*cgroup
, struct cftype
*cft
,
249 struct dev_cgroup
*devcgroup
= cgroup_to_devcgroup(cgroup
);
250 struct dev_whitelist_item
*wh
;
251 char maj
[MAJMINLEN
], min
[MAJMINLEN
], acc
[ACCLEN
];
253 spin_lock(&devcgroup
->lock
);
254 list_for_each_entry(wh
, &devcgroup
->whitelist
, list
) {
255 set_access(acc
, wh
->access
);
256 set_majmin(maj
, wh
->major
);
257 set_majmin(min
, wh
->minor
);
258 seq_printf(m
, "%c %s:%s %s\n", type_to_char(wh
->type
),
261 spin_unlock(&devcgroup
->lock
);
267 * may_access_whitelist:
268 * does the access granted to dev_cgroup c contain the access
269 * requested in whitelist item refwh.
270 * return 1 if yes, 0 if no.
271 * call with c->lock held
273 static int may_access_whitelist(struct dev_cgroup
*c
,
274 struct dev_whitelist_item
*refwh
)
276 struct dev_whitelist_item
*whitem
;
278 list_for_each_entry(whitem
, &c
->whitelist
, list
) {
279 if (whitem
->type
& DEV_ALL
)
281 if ((refwh
->type
& DEV_BLOCK
) && !(whitem
->type
& DEV_BLOCK
))
283 if ((refwh
->type
& DEV_CHAR
) && !(whitem
->type
& DEV_CHAR
))
285 if (whitem
->major
!= ~0 && whitem
->major
!= refwh
->major
)
287 if (whitem
->minor
!= ~0 && whitem
->minor
!= refwh
->minor
)
289 if (refwh
->access
& (~(whitem
->access
| ACC_MASK
)))
298 * when adding a new allow rule to a device whitelist, the rule
299 * must be allowed in the parent device
301 static int parent_has_perm(struct cgroup
*childcg
,
302 struct dev_whitelist_item
*wh
)
304 struct cgroup
*pcg
= childcg
->parent
;
305 struct dev_cgroup
*parent
;
310 parent
= cgroup_to_devcgroup(pcg
);
311 spin_lock(&parent
->lock
);
312 ret
= may_access_whitelist(parent
, wh
);
313 spin_unlock(&parent
->lock
);
318 * Modify the whitelist using allow/deny rules.
319 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
320 * so we can give a container CAP_MKNOD to let it create devices but not
321 * modify the whitelist.
322 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
323 * us to also grant CAP_SYS_ADMIN to containers without giving away the
324 * device whitelist controls, but for now we'll stick with CAP_SYS_ADMIN
326 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
327 * new access is only allowed if you're in the top-level cgroup, or your
328 * parent cgroup has the access you're asking for.
330 static ssize_t
devcgroup_access_write(struct cgroup
*cgroup
, struct cftype
*cft
,
331 struct file
*file
, const char __user
*userbuf
,
332 size_t nbytes
, loff_t
*ppos
)
334 struct cgroup
*cur_cgroup
;
335 struct dev_cgroup
*devcgroup
, *cur_devcgroup
;
336 int filetype
= cft
->private;
338 int retval
= 0, count
;
339 struct dev_whitelist_item wh
;
341 if (!capable(CAP_SYS_ADMIN
))
344 devcgroup
= cgroup_to_devcgroup(cgroup
);
345 cur_cgroup
= task_cgroup(current
, devices_subsys
.subsys_id
);
346 cur_devcgroup
= cgroup_to_devcgroup(cur_cgroup
);
348 buffer
= kmalloc(nbytes
+1, GFP_KERNEL
);
352 if (copy_from_user(buffer
, userbuf
, nbytes
)) {
356 buffer
[nbytes
] = 0; /* nul-terminate */
359 if (cgroup_is_removed(cgroup
)) {
364 memset(&wh
, 0, sizeof(wh
));
370 wh
.access
= ACC_MASK
;
391 } else if (isdigit(*b
)) {
393 while (isdigit(*b
)) {
394 wh
.major
= wh
.major
*10+(*b
-'0');
411 } else if (isdigit(*b
)) {
413 while (isdigit(*b
)) {
414 wh
.minor
= wh
.minor
*10+(*b
-'0');
425 for (b
++, count
= 0; count
< 3; count
++, b
++) {
428 wh
.access
|= ACC_READ
;
431 wh
.access
|= ACC_WRITE
;
434 wh
.access
|= ACC_MKNOD
;
450 if (!parent_has_perm(cgroup
, &wh
))
453 retval
= dev_whitelist_add(devcgroup
, &wh
);
456 dev_whitelist_rm(devcgroup
, &wh
);
473 static struct cftype dev_cgroup_files
[] = {
476 .write
= devcgroup_access_write
,
477 .private = DEVCG_ALLOW
,
481 .write
= devcgroup_access_write
,
482 .private = DEVCG_DENY
,
486 .read_seq_string
= devcgroup_seq_read
,
487 .private = DEVCG_LIST
,
491 static int devcgroup_populate(struct cgroup_subsys
*ss
,
492 struct cgroup
*cgroup
)
494 return cgroup_add_files(cgroup
, ss
, dev_cgroup_files
,
495 ARRAY_SIZE(dev_cgroup_files
));
498 struct cgroup_subsys devices_subsys
= {
500 .can_attach
= devcgroup_can_attach
,
501 .create
= devcgroup_create
,
502 .destroy
= devcgroup_destroy
,
503 .populate
= devcgroup_populate
,
504 .subsys_id
= devices_subsys_id
,
507 int devcgroup_inode_permission(struct inode
*inode
, int mask
)
509 struct dev_cgroup
*dev_cgroup
;
510 struct dev_whitelist_item
*wh
;
512 dev_t device
= inode
->i_rdev
;
515 if (!S_ISBLK(inode
->i_mode
) && !S_ISCHR(inode
->i_mode
))
517 dev_cgroup
= css_to_devcgroup(task_subsys_state(current
,
522 spin_lock(&dev_cgroup
->lock
);
523 list_for_each_entry(wh
, &dev_cgroup
->whitelist
, list
) {
524 if (wh
->type
& DEV_ALL
)
526 if ((wh
->type
& DEV_BLOCK
) && !S_ISBLK(inode
->i_mode
))
528 if ((wh
->type
& DEV_CHAR
) && !S_ISCHR(inode
->i_mode
))
530 if (wh
->major
!= ~0 && wh
->major
!= imajor(inode
))
532 if (wh
->minor
!= ~0 && wh
->minor
!= iminor(inode
))
535 if ((mask
& MAY_WRITE
) && !(wh
->access
& ACC_WRITE
))
537 if ((mask
& MAY_READ
) && !(wh
->access
& ACC_READ
))
539 spin_unlock(&dev_cgroup
->lock
);
542 spin_unlock(&dev_cgroup
->lock
);
547 int devcgroup_inode_mknod(int mode
, dev_t dev
)
549 struct dev_cgroup
*dev_cgroup
;
550 struct dev_whitelist_item
*wh
;
552 dev_cgroup
= css_to_devcgroup(task_subsys_state(current
,
557 spin_lock(&dev_cgroup
->lock
);
558 list_for_each_entry(wh
, &dev_cgroup
->whitelist
, list
) {
559 if (wh
->type
& DEV_ALL
)
561 if ((wh
->type
& DEV_BLOCK
) && !S_ISBLK(mode
))
563 if ((wh
->type
& DEV_CHAR
) && !S_ISCHR(mode
))
565 if (wh
->major
!= ~0 && wh
->major
!= MAJOR(dev
))
567 if (wh
->minor
!= ~0 && wh
->minor
!= MINOR(dev
))
570 if (!(wh
->access
& ACC_MKNOD
))
572 spin_unlock(&dev_cgroup
->lock
);
575 spin_unlock(&dev_cgroup
->lock
);