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 static inline struct dev_cgroup
*task_devcgroup(struct task_struct
*task
)
64 return css_to_devcgroup(task_subsys_state(task
, devices_subsys_id
));
67 struct cgroup_subsys devices_subsys
;
69 static int devcgroup_can_attach(struct cgroup_subsys
*ss
,
70 struct cgroup
*new_cgroup
, struct task_struct
*task
)
72 if (current
!= task
&& !capable(CAP_SYS_ADMIN
))
79 * called under cgroup_lock()
81 static int dev_whitelist_copy(struct list_head
*dest
, struct list_head
*orig
)
83 struct dev_whitelist_item
*wh
, *tmp
, *new;
85 list_for_each_entry(wh
, orig
, list
) {
86 new = kmalloc(sizeof(*wh
), GFP_KERNEL
);
89 new->major
= wh
->major
;
90 new->minor
= wh
->minor
;
92 new->access
= wh
->access
;
93 list_add_tail(&new->list
, dest
);
99 list_for_each_entry_safe(wh
, tmp
, dest
, list
) {
106 /* Stupid prototype - don't bother combining existing entries */
108 * called under cgroup_lock()
109 * since the list is visible to other tasks, we need the spinlock also
111 static int dev_whitelist_add(struct dev_cgroup
*dev_cgroup
,
112 struct dev_whitelist_item
*wh
)
114 struct dev_whitelist_item
*whcopy
, *walk
;
116 whcopy
= kmalloc(sizeof(*whcopy
), GFP_KERNEL
);
120 memcpy(whcopy
, wh
, sizeof(*whcopy
));
121 spin_lock(&dev_cgroup
->lock
);
122 list_for_each_entry(walk
, &dev_cgroup
->whitelist
, list
) {
123 if (walk
->type
!= wh
->type
)
125 if (walk
->major
!= wh
->major
)
127 if (walk
->minor
!= wh
->minor
)
130 walk
->access
|= wh
->access
;
136 list_add_tail(&whcopy
->list
, &dev_cgroup
->whitelist
);
137 spin_unlock(&dev_cgroup
->lock
);
142 * called under cgroup_lock()
143 * since the list is visible to other tasks, we need the spinlock also
145 static void dev_whitelist_rm(struct dev_cgroup
*dev_cgroup
,
146 struct dev_whitelist_item
*wh
)
148 struct dev_whitelist_item
*walk
, *tmp
;
150 spin_lock(&dev_cgroup
->lock
);
151 list_for_each_entry_safe(walk
, tmp
, &dev_cgroup
->whitelist
, list
) {
152 if (walk
->type
== DEV_ALL
)
154 if (walk
->type
!= wh
->type
)
156 if (walk
->major
!= ~0 && walk
->major
!= wh
->major
)
158 if (walk
->minor
!= ~0 && walk
->minor
!= wh
->minor
)
162 walk
->access
&= ~wh
->access
;
164 list_del(&walk
->list
);
168 spin_unlock(&dev_cgroup
->lock
);
172 * called from kernel/cgroup.c with cgroup_lock() held.
174 static struct cgroup_subsys_state
*devcgroup_create(struct cgroup_subsys
*ss
,
175 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
->whitelist
);
185 parent_cgroup
= cgroup
->parent
;
187 if (parent_cgroup
== NULL
) {
188 struct dev_whitelist_item
*wh
;
189 wh
= kmalloc(sizeof(*wh
), GFP_KERNEL
);
192 return ERR_PTR(-ENOMEM
);
194 wh
->minor
= wh
->major
= ~0;
196 wh
->access
= ACC_MKNOD
| ACC_READ
| ACC_WRITE
;
197 list_add(&wh
->list
, &dev_cgroup
->whitelist
);
199 parent_dev_cgroup
= cgroup_to_devcgroup(parent_cgroup
);
200 ret
= dev_whitelist_copy(&dev_cgroup
->whitelist
,
201 &parent_dev_cgroup
->whitelist
);
208 spin_lock_init(&dev_cgroup
->lock
);
209 return &dev_cgroup
->css
;
212 static void devcgroup_destroy(struct cgroup_subsys
*ss
,
213 struct cgroup
*cgroup
)
215 struct dev_cgroup
*dev_cgroup
;
216 struct dev_whitelist_item
*wh
, *tmp
;
218 dev_cgroup
= cgroup_to_devcgroup(cgroup
);
219 list_for_each_entry_safe(wh
, tmp
, &dev_cgroup
->whitelist
, list
) {
226 #define DEVCG_ALLOW 1
233 static void set_access(char *acc
, short access
)
236 memset(acc
, 0, ACCLEN
);
237 if (access
& ACC_READ
)
239 if (access
& ACC_WRITE
)
241 if (access
& ACC_MKNOD
)
245 static char type_to_char(short type
)
249 if (type
== DEV_CHAR
)
251 if (type
== DEV_BLOCK
)
256 static void set_majmin(char *str
, unsigned m
)
258 memset(str
, 0, MAJMINLEN
);
262 snprintf(str
, MAJMINLEN
, "%u", m
);
265 static int devcgroup_seq_read(struct cgroup
*cgroup
, struct cftype
*cft
,
268 struct dev_cgroup
*devcgroup
= cgroup_to_devcgroup(cgroup
);
269 struct dev_whitelist_item
*wh
;
270 char maj
[MAJMINLEN
], min
[MAJMINLEN
], acc
[ACCLEN
];
272 spin_lock(&devcgroup
->lock
);
273 list_for_each_entry(wh
, &devcgroup
->whitelist
, list
) {
274 set_access(acc
, wh
->access
);
275 set_majmin(maj
, wh
->major
);
276 set_majmin(min
, wh
->minor
);
277 seq_printf(m
, "%c %s:%s %s\n", type_to_char(wh
->type
),
280 spin_unlock(&devcgroup
->lock
);
286 * may_access_whitelist:
287 * does the access granted to dev_cgroup c contain the access
288 * requested in whitelist item refwh.
289 * return 1 if yes, 0 if no.
290 * call with c->lock held
292 static int may_access_whitelist(struct dev_cgroup
*c
,
293 struct dev_whitelist_item
*refwh
)
295 struct dev_whitelist_item
*whitem
;
297 list_for_each_entry(whitem
, &c
->whitelist
, list
) {
298 if (whitem
->type
& DEV_ALL
)
300 if ((refwh
->type
& DEV_BLOCK
) && !(whitem
->type
& DEV_BLOCK
))
302 if ((refwh
->type
& DEV_CHAR
) && !(whitem
->type
& DEV_CHAR
))
304 if (whitem
->major
!= ~0 && whitem
->major
!= refwh
->major
)
306 if (whitem
->minor
!= ~0 && whitem
->minor
!= refwh
->minor
)
308 if (refwh
->access
& (~whitem
->access
))
317 * when adding a new allow rule to a device whitelist, the rule
318 * must be allowed in the parent device
320 static int parent_has_perm(struct dev_cgroup
*childcg
,
321 struct dev_whitelist_item
*wh
)
323 struct cgroup
*pcg
= childcg
->css
.cgroup
->parent
;
324 struct dev_cgroup
*parent
;
329 parent
= cgroup_to_devcgroup(pcg
);
330 spin_lock(&parent
->lock
);
331 ret
= may_access_whitelist(parent
, wh
);
332 spin_unlock(&parent
->lock
);
337 * Modify the whitelist using allow/deny rules.
338 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
339 * so we can give a container CAP_MKNOD to let it create devices but not
340 * modify the whitelist.
341 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
342 * us to also grant CAP_SYS_ADMIN to containers without giving away the
343 * device whitelist controls, but for now we'll stick with CAP_SYS_ADMIN
345 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
346 * new access is only allowed if you're in the top-level cgroup, or your
347 * parent cgroup has the access you're asking for.
349 static int devcgroup_update_access(struct dev_cgroup
*devcgroup
,
350 int filetype
, const char *buffer
)
352 struct dev_cgroup
*cur_devcgroup
;
354 int retval
= 0, count
;
355 struct dev_whitelist_item wh
;
357 if (!capable(CAP_SYS_ADMIN
))
360 cur_devcgroup
= task_devcgroup(current
);
362 memset(&wh
, 0, sizeof(wh
));
368 wh
.access
= ACC_MASK
;
388 } else if (isdigit(*b
)) {
390 while (isdigit(*b
)) {
391 wh
.major
= wh
.major
*10+(*b
-'0');
405 } else if (isdigit(*b
)) {
407 while (isdigit(*b
)) {
408 wh
.minor
= wh
.minor
*10+(*b
-'0');
416 for (b
++, count
= 0; count
< 3; count
++, b
++) {
419 wh
.access
|= ACC_READ
;
422 wh
.access
|= ACC_WRITE
;
425 wh
.access
|= ACC_MKNOD
;
440 if (!parent_has_perm(devcgroup
, &wh
))
442 return dev_whitelist_add(devcgroup
, &wh
);
444 dev_whitelist_rm(devcgroup
, &wh
);
452 static int devcgroup_access_write(struct cgroup
*cgrp
, struct cftype
*cft
,
456 if (!cgroup_lock_live_group(cgrp
))
458 retval
= devcgroup_update_access(cgroup_to_devcgroup(cgrp
),
459 cft
->private, buffer
);
464 static struct cftype dev_cgroup_files
[] = {
467 .write_string
= devcgroup_access_write
,
468 .private = DEVCG_ALLOW
,
472 .write_string
= devcgroup_access_write
,
473 .private = DEVCG_DENY
,
477 .read_seq_string
= devcgroup_seq_read
,
478 .private = DEVCG_LIST
,
482 static int devcgroup_populate(struct cgroup_subsys
*ss
,
483 struct cgroup
*cgroup
)
485 return cgroup_add_files(cgroup
, ss
, dev_cgroup_files
,
486 ARRAY_SIZE(dev_cgroup_files
));
489 struct cgroup_subsys devices_subsys
= {
491 .can_attach
= devcgroup_can_attach
,
492 .create
= devcgroup_create
,
493 .destroy
= devcgroup_destroy
,
494 .populate
= devcgroup_populate
,
495 .subsys_id
= devices_subsys_id
,
498 int devcgroup_inode_permission(struct inode
*inode
, int mask
)
500 struct dev_cgroup
*dev_cgroup
;
501 struct dev_whitelist_item
*wh
;
503 dev_t device
= inode
->i_rdev
;
506 if (!S_ISBLK(inode
->i_mode
) && !S_ISCHR(inode
->i_mode
))
508 dev_cgroup
= css_to_devcgroup(task_subsys_state(current
,
513 spin_lock(&dev_cgroup
->lock
);
514 list_for_each_entry(wh
, &dev_cgroup
->whitelist
, list
) {
515 if (wh
->type
& DEV_ALL
)
517 if ((wh
->type
& DEV_BLOCK
) && !S_ISBLK(inode
->i_mode
))
519 if ((wh
->type
& DEV_CHAR
) && !S_ISCHR(inode
->i_mode
))
521 if (wh
->major
!= ~0 && wh
->major
!= imajor(inode
))
523 if (wh
->minor
!= ~0 && wh
->minor
!= iminor(inode
))
526 if ((mask
& MAY_WRITE
) && !(wh
->access
& ACC_WRITE
))
528 if ((mask
& MAY_READ
) && !(wh
->access
& ACC_READ
))
530 spin_unlock(&dev_cgroup
->lock
);
533 spin_unlock(&dev_cgroup
->lock
);
538 int devcgroup_inode_mknod(int mode
, dev_t dev
)
540 struct dev_cgroup
*dev_cgroup
;
541 struct dev_whitelist_item
*wh
;
543 dev_cgroup
= css_to_devcgroup(task_subsys_state(current
,
548 spin_lock(&dev_cgroup
->lock
);
549 list_for_each_entry(wh
, &dev_cgroup
->whitelist
, list
) {
550 if (wh
->type
& DEV_ALL
)
552 if ((wh
->type
& DEV_BLOCK
) && !S_ISBLK(mode
))
554 if ((wh
->type
& DEV_CHAR
) && !S_ISCHR(mode
))
556 if (wh
->major
!= ~0 && wh
->major
!= MAJOR(dev
))
558 if (wh
->minor
!= ~0 && wh
->minor
!= MINOR(dev
))
561 if (!(wh
->access
& ACC_MKNOD
))
563 spin_unlock(&dev_cgroup
->lock
);
566 spin_unlock(&dev_cgroup
->lock
);