Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * device_cgroup.c - device cgroup subsystem
4 *
5 * Copyright 2007 IBM Corp
6 */
7
8#include <linux/device_cgroup.h>
9#include <linux/cgroup.h>
10#include <linux/ctype.h>
11#include <linux/list.h>
12#include <linux/uaccess.h>
13#include <linux/seq_file.h>
14#include <linux/slab.h>
15#include <linux/rcupdate.h>
16#include <linux/mutex.h>
17
18static DEFINE_MUTEX(devcgroup_mutex);
19
20enum devcg_behavior {
21 DEVCG_DEFAULT_NONE,
22 DEVCG_DEFAULT_ALLOW,
23 DEVCG_DEFAULT_DENY,
24};
25
26/*
27 * exception list locking rules:
28 * hold devcgroup_mutex for update/read.
29 * hold rcu_read_lock() for read.
30 */
31
32struct dev_exception_item {
33 u32 major, minor;
34 short type;
35 short access;
36 struct list_head list;
37 struct rcu_head rcu;
38};
39
40struct dev_cgroup {
41 struct cgroup_subsys_state css;
42 struct list_head exceptions;
43 enum devcg_behavior behavior;
44};
45
46static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
47{
48 return s ? container_of(s, struct dev_cgroup, css) : NULL;
49}
50
51static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
52{
53 return css_to_devcgroup(task_css(task, devices_cgrp_id));
54}
55
56/*
57 * called under devcgroup_mutex
58 */
59static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
60{
61 struct dev_exception_item *ex, *tmp, *new;
62
63 lockdep_assert_held(&devcgroup_mutex);
64
65 list_for_each_entry(ex, orig, list) {
66 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
67 if (!new)
68 goto free_and_exit;
69 list_add_tail(&new->list, dest);
70 }
71
72 return 0;
73
74free_and_exit:
75 list_for_each_entry_safe(ex, tmp, dest, list) {
76 list_del(&ex->list);
77 kfree(ex);
78 }
79 return -ENOMEM;
80}
81
82/*
83 * called under devcgroup_mutex
84 */
85static int dev_exception_add(struct dev_cgroup *dev_cgroup,
86 struct dev_exception_item *ex)
87{
88 struct dev_exception_item *excopy, *walk;
89
90 lockdep_assert_held(&devcgroup_mutex);
91
92 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
93 if (!excopy)
94 return -ENOMEM;
95
96 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
97 if (walk->type != ex->type)
98 continue;
99 if (walk->major != ex->major)
100 continue;
101 if (walk->minor != ex->minor)
102 continue;
103
104 walk->access |= ex->access;
105 kfree(excopy);
106 excopy = NULL;
107 }
108
109 if (excopy != NULL)
110 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
111 return 0;
112}
113
114/*
115 * called under devcgroup_mutex
116 */
117static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
118 struct dev_exception_item *ex)
119{
120 struct dev_exception_item *walk, *tmp;
121
122 lockdep_assert_held(&devcgroup_mutex);
123
124 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
125 if (walk->type != ex->type)
126 continue;
127 if (walk->major != ex->major)
128 continue;
129 if (walk->minor != ex->minor)
130 continue;
131
132 walk->access &= ~ex->access;
133 if (!walk->access) {
134 list_del_rcu(&walk->list);
135 kfree_rcu(walk, rcu);
136 }
137 }
138}
139
140static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
141{
142 struct dev_exception_item *ex, *tmp;
143
144 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
145 list_del_rcu(&ex->list);
146 kfree_rcu(ex, rcu);
147 }
148}
149
150/**
151 * dev_exception_clean - frees all entries of the exception list
152 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
153 *
154 * called under devcgroup_mutex
155 */
156static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
157{
158 lockdep_assert_held(&devcgroup_mutex);
159
160 __dev_exception_clean(dev_cgroup);
161}
162
163static inline bool is_devcg_online(const struct dev_cgroup *devcg)
164{
165 return (devcg->behavior != DEVCG_DEFAULT_NONE);
166}
167
168/**
169 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
170 * parent's
171 * @css: css getting online
172 * returns 0 in case of success, error code otherwise
173 */
174static int devcgroup_online(struct cgroup_subsys_state *css)
175{
176 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
177 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
178 int ret = 0;
179
180 mutex_lock(&devcgroup_mutex);
181
182 if (parent_dev_cgroup == NULL)
183 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
184 else {
185 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
186 &parent_dev_cgroup->exceptions);
187 if (!ret)
188 dev_cgroup->behavior = parent_dev_cgroup->behavior;
189 }
190 mutex_unlock(&devcgroup_mutex);
191
192 return ret;
193}
194
195static void devcgroup_offline(struct cgroup_subsys_state *css)
196{
197 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
198
199 mutex_lock(&devcgroup_mutex);
200 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
201 mutex_unlock(&devcgroup_mutex);
202}
203
204/*
205 * called from kernel/cgroup.c with cgroup_lock() held.
206 */
207static struct cgroup_subsys_state *
208devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
209{
210 struct dev_cgroup *dev_cgroup;
211
212 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
213 if (!dev_cgroup)
214 return ERR_PTR(-ENOMEM);
215 INIT_LIST_HEAD(&dev_cgroup->exceptions);
216 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
217
218 return &dev_cgroup->css;
219}
220
221static void devcgroup_css_free(struct cgroup_subsys_state *css)
222{
223 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
224
225 __dev_exception_clean(dev_cgroup);
226 kfree(dev_cgroup);
227}
228
229#define DEVCG_ALLOW 1
230#define DEVCG_DENY 2
231#define DEVCG_LIST 3
232
233#define MAJMINLEN 13
234#define ACCLEN 4
235
236static void set_access(char *acc, short access)
237{
238 int idx = 0;
239 memset(acc, 0, ACCLEN);
240 if (access & DEVCG_ACC_READ)
241 acc[idx++] = 'r';
242 if (access & DEVCG_ACC_WRITE)
243 acc[idx++] = 'w';
244 if (access & DEVCG_ACC_MKNOD)
245 acc[idx++] = 'm';
246}
247
248static char type_to_char(short type)
249{
250 if (type == DEVCG_DEV_ALL)
251 return 'a';
252 if (type == DEVCG_DEV_CHAR)
253 return 'c';
254 if (type == DEVCG_DEV_BLOCK)
255 return 'b';
256 return 'X';
257}
258
259static void set_majmin(char *str, unsigned m)
260{
261 if (m == ~0)
262 strcpy(str, "*");
263 else
264 sprintf(str, "%u", m);
265}
266
267static int devcgroup_seq_show(struct seq_file *m, void *v)
268{
269 struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
270 struct dev_exception_item *ex;
271 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
272
273 rcu_read_lock();
274 /*
275 * To preserve the compatibility:
276 * - Only show the "all devices" when the default policy is to allow
277 * - List the exceptions in case the default policy is to deny
278 * This way, the file remains as a "whitelist of devices"
279 */
280 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
281 set_access(acc, DEVCG_ACC_MASK);
282 set_majmin(maj, ~0);
283 set_majmin(min, ~0);
284 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEVCG_DEV_ALL),
285 maj, min, acc);
286 } else {
287 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
288 set_access(acc, ex->access);
289 set_majmin(maj, ex->major);
290 set_majmin(min, ex->minor);
291 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
292 maj, min, acc);
293 }
294 }
295 rcu_read_unlock();
296
297 return 0;
298}
299
300/**
301 * match_exception - iterates the exception list trying to find a complete match
302 * @exceptions: list of exceptions
303 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
304 * @major: device file major number, ~0 to match all
305 * @minor: device file minor number, ~0 to match all
306 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
307 *
308 * It is considered a complete match if an exception is found that will
309 * contain the entire range of provided parameters.
310 *
311 * Return: true in case it matches an exception completely
312 */
313static bool match_exception(struct list_head *exceptions, short type,
314 u32 major, u32 minor, short access)
315{
316 struct dev_exception_item *ex;
317
318 list_for_each_entry_rcu(ex, exceptions, list) {
319 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
320 continue;
321 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
322 continue;
323 if (ex->major != ~0 && ex->major != major)
324 continue;
325 if (ex->minor != ~0 && ex->minor != minor)
326 continue;
327 /* provided access cannot have more than the exception rule */
328 if (access & (~ex->access))
329 continue;
330 return true;
331 }
332 return false;
333}
334
335/**
336 * match_exception_partial - iterates the exception list trying to find a partial match
337 * @exceptions: list of exceptions
338 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
339 * @major: device file major number, ~0 to match all
340 * @minor: device file minor number, ~0 to match all
341 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
342 *
343 * It is considered a partial match if an exception's range is found to
344 * contain *any* of the devices specified by provided parameters. This is
345 * used to make sure no extra access is being granted that is forbidden by
346 * any of the exception list.
347 *
348 * Return: true in case the provided range mat matches an exception completely
349 */
350static bool match_exception_partial(struct list_head *exceptions, short type,
351 u32 major, u32 minor, short access)
352{
353 struct dev_exception_item *ex;
354
355 list_for_each_entry_rcu(ex, exceptions, list) {
356 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
357 continue;
358 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
359 continue;
360 /*
361 * We must be sure that both the exception and the provided
362 * range aren't masking all devices
363 */
364 if (ex->major != ~0 && major != ~0 && ex->major != major)
365 continue;
366 if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
367 continue;
368 /*
369 * In order to make sure the provided range isn't matching
370 * an exception, all its access bits shouldn't match the
371 * exception's access bits
372 */
373 if (!(access & ex->access))
374 continue;
375 return true;
376 }
377 return false;
378}
379
380/**
381 * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
382 * @dev_cgroup: dev cgroup to be tested against
383 * @refex: new exception
384 * @behavior: behavior of the exception's dev_cgroup
385 *
386 * This is used to make sure a child cgroup won't have more privileges
387 * than its parent
388 */
389static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
390 struct dev_exception_item *refex,
391 enum devcg_behavior behavior)
392{
393 bool match = false;
394
395 RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
396 !lockdep_is_held(&devcgroup_mutex),
397 "device_cgroup:verify_new_ex called without proper synchronization");
398
399 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
400 if (behavior == DEVCG_DEFAULT_ALLOW) {
401 /*
402 * new exception in the child doesn't matter, only
403 * adding extra restrictions
404 */
405 return true;
406 } else {
407 /*
408 * new exception in the child will add more devices
409 * that can be acessed, so it can't match any of
410 * parent's exceptions, even slightly
411 */
412 match = match_exception_partial(&dev_cgroup->exceptions,
413 refex->type,
414 refex->major,
415 refex->minor,
416 refex->access);
417
418 if (match)
419 return false;
420 return true;
421 }
422 } else {
423 /*
424 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
425 * the new exception will add access to more devices and must
426 * be contained completely in an parent's exception to be
427 * allowed
428 */
429 match = match_exception(&dev_cgroup->exceptions, refex->type,
430 refex->major, refex->minor,
431 refex->access);
432
433 if (match)
434 /* parent has an exception that matches the proposed */
435 return true;
436 else
437 return false;
438 }
439 return false;
440}
441
442/*
443 * parent_has_perm:
444 * when adding a new allow rule to a device exception list, the rule
445 * must be allowed in the parent device
446 */
447static int parent_has_perm(struct dev_cgroup *childcg,
448 struct dev_exception_item *ex)
449{
450 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
451
452 if (!parent)
453 return 1;
454 return verify_new_ex(parent, ex, childcg->behavior);
455}
456
457/**
458 * parent_allows_removal - verify if it's ok to remove an exception
459 * @childcg: child cgroup from where the exception will be removed
460 * @ex: exception being removed
461 *
462 * When removing an exception in cgroups with default ALLOW policy, it must
463 * be checked if removing it will give the child cgroup more access than the
464 * parent.
465 *
466 * Return: true if it's ok to remove exception, false otherwise
467 */
468static bool parent_allows_removal(struct dev_cgroup *childcg,
469 struct dev_exception_item *ex)
470{
471 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
472
473 if (!parent)
474 return true;
475
476 /* It's always allowed to remove access to devices */
477 if (childcg->behavior == DEVCG_DEFAULT_DENY)
478 return true;
479
480 /*
481 * Make sure you're not removing part or a whole exception existing in
482 * the parent cgroup
483 */
484 return !match_exception_partial(&parent->exceptions, ex->type,
485 ex->major, ex->minor, ex->access);
486}
487
488/**
489 * may_allow_all - checks if it's possible to change the behavior to
490 * allow based on parent's rules.
491 * @parent: device cgroup's parent
492 * returns: != 0 in case it's allowed, 0 otherwise
493 */
494static inline int may_allow_all(struct dev_cgroup *parent)
495{
496 if (!parent)
497 return 1;
498 return parent->behavior == DEVCG_DEFAULT_ALLOW;
499}
500
501/**
502 * revalidate_active_exceptions - walks through the active exception list and
503 * revalidates the exceptions based on parent's
504 * behavior and exceptions. The exceptions that
505 * are no longer valid will be removed.
506 * Called with devcgroup_mutex held.
507 * @devcg: cgroup which exceptions will be checked
508 *
509 * This is one of the three key functions for hierarchy implementation.
510 * This function is responsible for re-evaluating all the cgroup's active
511 * exceptions due to a parent's exception change.
512 * Refer to Documentation/admin-guide/cgroup-v1/devices.rst for more details.
513 */
514static void revalidate_active_exceptions(struct dev_cgroup *devcg)
515{
516 struct dev_exception_item *ex;
517 struct list_head *this, *tmp;
518
519 list_for_each_safe(this, tmp, &devcg->exceptions) {
520 ex = container_of(this, struct dev_exception_item, list);
521 if (!parent_has_perm(devcg, ex))
522 dev_exception_rm(devcg, ex);
523 }
524}
525
526/**
527 * propagate_exception - propagates a new exception to the children
528 * @devcg_root: device cgroup that added a new exception
529 * @ex: new exception to be propagated
530 *
531 * returns: 0 in case of success, != 0 in case of error
532 */
533static int propagate_exception(struct dev_cgroup *devcg_root,
534 struct dev_exception_item *ex)
535{
536 struct cgroup_subsys_state *pos;
537 int rc = 0;
538
539 rcu_read_lock();
540
541 css_for_each_descendant_pre(pos, &devcg_root->css) {
542 struct dev_cgroup *devcg = css_to_devcgroup(pos);
543
544 /*
545 * Because devcgroup_mutex is held, no devcg will become
546 * online or offline during the tree walk (see on/offline
547 * methods), and online ones are safe to access outside RCU
548 * read lock without bumping refcnt.
549 */
550 if (pos == &devcg_root->css || !is_devcg_online(devcg))
551 continue;
552
553 rcu_read_unlock();
554
555 /*
556 * in case both root's behavior and devcg is allow, a new
557 * restriction means adding to the exception list
558 */
559 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
560 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
561 rc = dev_exception_add(devcg, ex);
562 if (rc)
563 return rc;
564 } else {
565 /*
566 * in the other possible cases:
567 * root's behavior: allow, devcg's: deny
568 * root's behavior: deny, devcg's: deny
569 * the exception will be removed
570 */
571 dev_exception_rm(devcg, ex);
572 }
573 revalidate_active_exceptions(devcg);
574
575 rcu_read_lock();
576 }
577
578 rcu_read_unlock();
579 return rc;
580}
581
582/*
583 * Modify the exception list using allow/deny rules.
584 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
585 * so we can give a container CAP_MKNOD to let it create devices but not
586 * modify the exception list.
587 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
588 * us to also grant CAP_SYS_ADMIN to containers without giving away the
589 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
590 *
591 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
592 * new access is only allowed if you're in the top-level cgroup, or your
593 * parent cgroup has the access you're asking for.
594 */
595static int devcgroup_update_access(struct dev_cgroup *devcgroup,
596 int filetype, char *buffer)
597{
598 const char *b;
599 char temp[12]; /* 11 + 1 characters needed for a u32 */
600 int count, rc = 0;
601 struct dev_exception_item ex;
602 struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
603
604 if (!capable(CAP_SYS_ADMIN))
605 return -EPERM;
606
607 memset(&ex, 0, sizeof(ex));
608 b = buffer;
609
610 switch (*b) {
611 case 'a':
612 switch (filetype) {
613 case DEVCG_ALLOW:
614 if (css_has_online_children(&devcgroup->css))
615 return -EINVAL;
616
617 if (!may_allow_all(parent))
618 return -EPERM;
619 dev_exception_clean(devcgroup);
620 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
621 if (!parent)
622 break;
623
624 rc = dev_exceptions_copy(&devcgroup->exceptions,
625 &parent->exceptions);
626 if (rc)
627 return rc;
628 break;
629 case DEVCG_DENY:
630 if (css_has_online_children(&devcgroup->css))
631 return -EINVAL;
632
633 dev_exception_clean(devcgroup);
634 devcgroup->behavior = DEVCG_DEFAULT_DENY;
635 break;
636 default:
637 return -EINVAL;
638 }
639 return 0;
640 case 'b':
641 ex.type = DEVCG_DEV_BLOCK;
642 break;
643 case 'c':
644 ex.type = DEVCG_DEV_CHAR;
645 break;
646 default:
647 return -EINVAL;
648 }
649 b++;
650 if (!isspace(*b))
651 return -EINVAL;
652 b++;
653 if (*b == '*') {
654 ex.major = ~0;
655 b++;
656 } else if (isdigit(*b)) {
657 memset(temp, 0, sizeof(temp));
658 for (count = 0; count < sizeof(temp) - 1; count++) {
659 temp[count] = *b;
660 b++;
661 if (!isdigit(*b))
662 break;
663 }
664 rc = kstrtou32(temp, 10, &ex.major);
665 if (rc)
666 return -EINVAL;
667 } else {
668 return -EINVAL;
669 }
670 if (*b != ':')
671 return -EINVAL;
672 b++;
673
674 /* read minor */
675 if (*b == '*') {
676 ex.minor = ~0;
677 b++;
678 } else if (isdigit(*b)) {
679 memset(temp, 0, sizeof(temp));
680 for (count = 0; count < sizeof(temp) - 1; count++) {
681 temp[count] = *b;
682 b++;
683 if (!isdigit(*b))
684 break;
685 }
686 rc = kstrtou32(temp, 10, &ex.minor);
687 if (rc)
688 return -EINVAL;
689 } else {
690 return -EINVAL;
691 }
692 if (!isspace(*b))
693 return -EINVAL;
694 for (b++, count = 0; count < 3; count++, b++) {
695 switch (*b) {
696 case 'r':
697 ex.access |= DEVCG_ACC_READ;
698 break;
699 case 'w':
700 ex.access |= DEVCG_ACC_WRITE;
701 break;
702 case 'm':
703 ex.access |= DEVCG_ACC_MKNOD;
704 break;
705 case '\n':
706 case '\0':
707 count = 3;
708 break;
709 default:
710 return -EINVAL;
711 }
712 }
713
714 switch (filetype) {
715 case DEVCG_ALLOW:
716 /*
717 * If the default policy is to allow by default, try to remove
718 * an matching exception instead. And be silent about it: we
719 * don't want to break compatibility
720 */
721 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
722 /* Check if the parent allows removing it first */
723 if (!parent_allows_removal(devcgroup, &ex))
724 return -EPERM;
725 dev_exception_rm(devcgroup, &ex);
726 break;
727 }
728
729 if (!parent_has_perm(devcgroup, &ex))
730 return -EPERM;
731 rc = dev_exception_add(devcgroup, &ex);
732 break;
733 case DEVCG_DENY:
734 /*
735 * If the default policy is to deny by default, try to remove
736 * an matching exception instead. And be silent about it: we
737 * don't want to break compatibility
738 */
739 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
740 dev_exception_rm(devcgroup, &ex);
741 else
742 rc = dev_exception_add(devcgroup, &ex);
743
744 if (rc)
745 break;
746 /* we only propagate new restrictions */
747 rc = propagate_exception(devcgroup, &ex);
748 break;
749 default:
750 rc = -EINVAL;
751 }
752 return rc;
753}
754
755static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
756 char *buf, size_t nbytes, loff_t off)
757{
758 int retval;
759
760 mutex_lock(&devcgroup_mutex);
761 retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
762 of_cft(of)->private, strstrip(buf));
763 mutex_unlock(&devcgroup_mutex);
764 return retval ?: nbytes;
765}
766
767static struct cftype dev_cgroup_files[] = {
768 {
769 .name = "allow",
770 .write = devcgroup_access_write,
771 .private = DEVCG_ALLOW,
772 },
773 {
774 .name = "deny",
775 .write = devcgroup_access_write,
776 .private = DEVCG_DENY,
777 },
778 {
779 .name = "list",
780 .seq_show = devcgroup_seq_show,
781 .private = DEVCG_LIST,
782 },
783 { } /* terminate */
784};
785
786struct cgroup_subsys devices_cgrp_subsys = {
787 .css_alloc = devcgroup_css_alloc,
788 .css_free = devcgroup_css_free,
789 .css_online = devcgroup_online,
790 .css_offline = devcgroup_offline,
791 .legacy_cftypes = dev_cgroup_files,
792};
793
794/**
795 * __devcgroup_check_permission - checks if an inode operation is permitted
796 * @dev_cgroup: the dev cgroup to be tested against
797 * @type: device type
798 * @major: device major number
799 * @minor: device minor number
800 * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
801 *
802 * returns 0 on success, -EPERM case the operation is not permitted
803 */
804int __devcgroup_check_permission(short type, u32 major, u32 minor,
805 short access)
806{
807 struct dev_cgroup *dev_cgroup;
808 bool rc;
809
810 rcu_read_lock();
811 dev_cgroup = task_devcgroup(current);
812 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
813 /* Can't match any of the exceptions, even partially */
814 rc = !match_exception_partial(&dev_cgroup->exceptions,
815 type, major, minor, access);
816 else
817 /* Need to match completely one exception to be allowed */
818 rc = match_exception(&dev_cgroup->exceptions, type, major,
819 minor, access);
820 rcu_read_unlock();
821
822 if (!rc)
823 return -EPERM;
824
825 return 0;
826}
1/*
2 * device_cgroup.c - device cgroup subsystem
3 *
4 * Copyright 2007 IBM Corp
5 */
6
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>
16
17#define ACC_MKNOD 1
18#define ACC_READ 2
19#define ACC_WRITE 4
20#define ACC_MASK (ACC_MKNOD | ACC_READ | ACC_WRITE)
21
22#define DEV_BLOCK 1
23#define DEV_CHAR 2
24#define DEV_ALL 4 /* this represents all devices */
25
26static DEFINE_MUTEX(devcgroup_mutex);
27
28/*
29 * whitelist locking rules:
30 * hold devcgroup_mutex for update/read.
31 * hold rcu_read_lock() for read.
32 */
33
34struct dev_whitelist_item {
35 u32 major, minor;
36 short type;
37 short access;
38 struct list_head list;
39 struct rcu_head rcu;
40};
41
42struct dev_cgroup {
43 struct cgroup_subsys_state css;
44 struct list_head whitelist;
45};
46
47static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
48{
49 return container_of(s, struct dev_cgroup, css);
50}
51
52static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup)
53{
54 return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id));
55}
56
57static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
58{
59 return css_to_devcgroup(task_subsys_state(task, devices_subsys_id));
60}
61
62struct cgroup_subsys devices_subsys;
63
64static int devcgroup_can_attach(struct cgroup *new_cgrp,
65 struct cgroup_taskset *set)
66{
67 struct task_struct *task = cgroup_taskset_first(set);
68
69 if (current != task && !capable(CAP_SYS_ADMIN))
70 return -EPERM;
71 return 0;
72}
73
74/*
75 * called under devcgroup_mutex
76 */
77static int dev_whitelist_copy(struct list_head *dest, struct list_head *orig)
78{
79 struct dev_whitelist_item *wh, *tmp, *new;
80
81 list_for_each_entry(wh, orig, list) {
82 new = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
83 if (!new)
84 goto free_and_exit;
85 list_add_tail(&new->list, dest);
86 }
87
88 return 0;
89
90free_and_exit:
91 list_for_each_entry_safe(wh, tmp, dest, list) {
92 list_del(&wh->list);
93 kfree(wh);
94 }
95 return -ENOMEM;
96}
97
98/* Stupid prototype - don't bother combining existing entries */
99/*
100 * called under devcgroup_mutex
101 */
102static int dev_whitelist_add(struct dev_cgroup *dev_cgroup,
103 struct dev_whitelist_item *wh)
104{
105 struct dev_whitelist_item *whcopy, *walk;
106
107 whcopy = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
108 if (!whcopy)
109 return -ENOMEM;
110
111 list_for_each_entry(walk, &dev_cgroup->whitelist, list) {
112 if (walk->type != wh->type)
113 continue;
114 if (walk->major != wh->major)
115 continue;
116 if (walk->minor != wh->minor)
117 continue;
118
119 walk->access |= wh->access;
120 kfree(whcopy);
121 whcopy = NULL;
122 }
123
124 if (whcopy != NULL)
125 list_add_tail_rcu(&whcopy->list, &dev_cgroup->whitelist);
126 return 0;
127}
128
129/*
130 * called under devcgroup_mutex
131 */
132static void dev_whitelist_rm(struct dev_cgroup *dev_cgroup,
133 struct dev_whitelist_item *wh)
134{
135 struct dev_whitelist_item *walk, *tmp;
136
137 list_for_each_entry_safe(walk, tmp, &dev_cgroup->whitelist, list) {
138 if (walk->type == DEV_ALL)
139 goto remove;
140 if (walk->type != wh->type)
141 continue;
142 if (walk->major != ~0 && walk->major != wh->major)
143 continue;
144 if (walk->minor != ~0 && walk->minor != wh->minor)
145 continue;
146
147remove:
148 walk->access &= ~wh->access;
149 if (!walk->access) {
150 list_del_rcu(&walk->list);
151 kfree_rcu(walk, rcu);
152 }
153 }
154}
155
156/*
157 * called from kernel/cgroup.c with cgroup_lock() held.
158 */
159static struct cgroup_subsys_state *devcgroup_create(struct cgroup *cgroup)
160{
161 struct dev_cgroup *dev_cgroup, *parent_dev_cgroup;
162 struct cgroup *parent_cgroup;
163 int ret;
164
165 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
166 if (!dev_cgroup)
167 return ERR_PTR(-ENOMEM);
168 INIT_LIST_HEAD(&dev_cgroup->whitelist);
169 parent_cgroup = cgroup->parent;
170
171 if (parent_cgroup == NULL) {
172 struct dev_whitelist_item *wh;
173 wh = kmalloc(sizeof(*wh), GFP_KERNEL);
174 if (!wh) {
175 kfree(dev_cgroup);
176 return ERR_PTR(-ENOMEM);
177 }
178 wh->minor = wh->major = ~0;
179 wh->type = DEV_ALL;
180 wh->access = ACC_MASK;
181 list_add(&wh->list, &dev_cgroup->whitelist);
182 } else {
183 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
184 mutex_lock(&devcgroup_mutex);
185 ret = dev_whitelist_copy(&dev_cgroup->whitelist,
186 &parent_dev_cgroup->whitelist);
187 mutex_unlock(&devcgroup_mutex);
188 if (ret) {
189 kfree(dev_cgroup);
190 return ERR_PTR(ret);
191 }
192 }
193
194 return &dev_cgroup->css;
195}
196
197static void devcgroup_destroy(struct cgroup *cgroup)
198{
199 struct dev_cgroup *dev_cgroup;
200 struct dev_whitelist_item *wh, *tmp;
201
202 dev_cgroup = cgroup_to_devcgroup(cgroup);
203 list_for_each_entry_safe(wh, tmp, &dev_cgroup->whitelist, list) {
204 list_del(&wh->list);
205 kfree(wh);
206 }
207 kfree(dev_cgroup);
208}
209
210#define DEVCG_ALLOW 1
211#define DEVCG_DENY 2
212#define DEVCG_LIST 3
213
214#define MAJMINLEN 13
215#define ACCLEN 4
216
217static void set_access(char *acc, short access)
218{
219 int idx = 0;
220 memset(acc, 0, ACCLEN);
221 if (access & ACC_READ)
222 acc[idx++] = 'r';
223 if (access & ACC_WRITE)
224 acc[idx++] = 'w';
225 if (access & ACC_MKNOD)
226 acc[idx++] = 'm';
227}
228
229static char type_to_char(short type)
230{
231 if (type == DEV_ALL)
232 return 'a';
233 if (type == DEV_CHAR)
234 return 'c';
235 if (type == DEV_BLOCK)
236 return 'b';
237 return 'X';
238}
239
240static void set_majmin(char *str, unsigned m)
241{
242 if (m == ~0)
243 strcpy(str, "*");
244 else
245 sprintf(str, "%u", m);
246}
247
248static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
249 struct seq_file *m)
250{
251 struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
252 struct dev_whitelist_item *wh;
253 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
254
255 rcu_read_lock();
256 list_for_each_entry_rcu(wh, &devcgroup->whitelist, list) {
257 set_access(acc, wh->access);
258 set_majmin(maj, wh->major);
259 set_majmin(min, wh->minor);
260 seq_printf(m, "%c %s:%s %s\n", type_to_char(wh->type),
261 maj, min, acc);
262 }
263 rcu_read_unlock();
264
265 return 0;
266}
267
268/*
269 * may_access_whitelist:
270 * does the access granted to dev_cgroup c contain the access
271 * requested in whitelist item refwh.
272 * return 1 if yes, 0 if no.
273 * call with devcgroup_mutex held
274 */
275static int may_access_whitelist(struct dev_cgroup *c,
276 struct dev_whitelist_item *refwh)
277{
278 struct dev_whitelist_item *whitem;
279
280 list_for_each_entry(whitem, &c->whitelist, list) {
281 if (whitem->type & DEV_ALL)
282 return 1;
283 if ((refwh->type & DEV_BLOCK) && !(whitem->type & DEV_BLOCK))
284 continue;
285 if ((refwh->type & DEV_CHAR) && !(whitem->type & DEV_CHAR))
286 continue;
287 if (whitem->major != ~0 && whitem->major != refwh->major)
288 continue;
289 if (whitem->minor != ~0 && whitem->minor != refwh->minor)
290 continue;
291 if (refwh->access & (~whitem->access))
292 continue;
293 return 1;
294 }
295 return 0;
296}
297
298/*
299 * parent_has_perm:
300 * when adding a new allow rule to a device whitelist, the rule
301 * must be allowed in the parent device
302 */
303static int parent_has_perm(struct dev_cgroup *childcg,
304 struct dev_whitelist_item *wh)
305{
306 struct cgroup *pcg = childcg->css.cgroup->parent;
307 struct dev_cgroup *parent;
308
309 if (!pcg)
310 return 1;
311 parent = cgroup_to_devcgroup(pcg);
312 return may_access_whitelist(parent, wh);
313}
314
315/*
316 * Modify the whitelist using allow/deny rules.
317 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
318 * so we can give a container CAP_MKNOD to let it create devices but not
319 * modify the whitelist.
320 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
321 * us to also grant CAP_SYS_ADMIN to containers without giving away the
322 * device whitelist controls, but for now we'll stick with CAP_SYS_ADMIN
323 *
324 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
325 * new access is only allowed if you're in the top-level cgroup, or your
326 * parent cgroup has the access you're asking for.
327 */
328static int devcgroup_update_access(struct dev_cgroup *devcgroup,
329 int filetype, const char *buffer)
330{
331 const char *b;
332 char *endp;
333 int count;
334 struct dev_whitelist_item wh;
335
336 if (!capable(CAP_SYS_ADMIN))
337 return -EPERM;
338
339 memset(&wh, 0, sizeof(wh));
340 b = buffer;
341
342 switch (*b) {
343 case 'a':
344 wh.type = DEV_ALL;
345 wh.access = ACC_MASK;
346 wh.major = ~0;
347 wh.minor = ~0;
348 goto handle;
349 case 'b':
350 wh.type = DEV_BLOCK;
351 break;
352 case 'c':
353 wh.type = DEV_CHAR;
354 break;
355 default:
356 return -EINVAL;
357 }
358 b++;
359 if (!isspace(*b))
360 return -EINVAL;
361 b++;
362 if (*b == '*') {
363 wh.major = ~0;
364 b++;
365 } else if (isdigit(*b)) {
366 wh.major = simple_strtoul(b, &endp, 10);
367 b = endp;
368 } else {
369 return -EINVAL;
370 }
371 if (*b != ':')
372 return -EINVAL;
373 b++;
374
375 /* read minor */
376 if (*b == '*') {
377 wh.minor = ~0;
378 b++;
379 } else if (isdigit(*b)) {
380 wh.minor = simple_strtoul(b, &endp, 10);
381 b = endp;
382 } else {
383 return -EINVAL;
384 }
385 if (!isspace(*b))
386 return -EINVAL;
387 for (b++, count = 0; count < 3; count++, b++) {
388 switch (*b) {
389 case 'r':
390 wh.access |= ACC_READ;
391 break;
392 case 'w':
393 wh.access |= ACC_WRITE;
394 break;
395 case 'm':
396 wh.access |= ACC_MKNOD;
397 break;
398 case '\n':
399 case '\0':
400 count = 3;
401 break;
402 default:
403 return -EINVAL;
404 }
405 }
406
407handle:
408 switch (filetype) {
409 case DEVCG_ALLOW:
410 if (!parent_has_perm(devcgroup, &wh))
411 return -EPERM;
412 return dev_whitelist_add(devcgroup, &wh);
413 case DEVCG_DENY:
414 dev_whitelist_rm(devcgroup, &wh);
415 break;
416 default:
417 return -EINVAL;
418 }
419 return 0;
420}
421
422static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
423 const char *buffer)
424{
425 int retval;
426
427 mutex_lock(&devcgroup_mutex);
428 retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
429 cft->private, buffer);
430 mutex_unlock(&devcgroup_mutex);
431 return retval;
432}
433
434static struct cftype dev_cgroup_files[] = {
435 {
436 .name = "allow",
437 .write_string = devcgroup_access_write,
438 .private = DEVCG_ALLOW,
439 },
440 {
441 .name = "deny",
442 .write_string = devcgroup_access_write,
443 .private = DEVCG_DENY,
444 },
445 {
446 .name = "list",
447 .read_seq_string = devcgroup_seq_read,
448 .private = DEVCG_LIST,
449 },
450 { } /* terminate */
451};
452
453struct cgroup_subsys devices_subsys = {
454 .name = "devices",
455 .can_attach = devcgroup_can_attach,
456 .create = devcgroup_create,
457 .destroy = devcgroup_destroy,
458 .subsys_id = devices_subsys_id,
459 .base_cftypes = dev_cgroup_files,
460};
461
462int __devcgroup_inode_permission(struct inode *inode, int mask)
463{
464 struct dev_cgroup *dev_cgroup;
465 struct dev_whitelist_item *wh;
466
467 rcu_read_lock();
468
469 dev_cgroup = task_devcgroup(current);
470
471 list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
472 if (wh->type & DEV_ALL)
473 goto found;
474 if ((wh->type & DEV_BLOCK) && !S_ISBLK(inode->i_mode))
475 continue;
476 if ((wh->type & DEV_CHAR) && !S_ISCHR(inode->i_mode))
477 continue;
478 if (wh->major != ~0 && wh->major != imajor(inode))
479 continue;
480 if (wh->minor != ~0 && wh->minor != iminor(inode))
481 continue;
482
483 if ((mask & MAY_WRITE) && !(wh->access & ACC_WRITE))
484 continue;
485 if ((mask & MAY_READ) && !(wh->access & ACC_READ))
486 continue;
487found:
488 rcu_read_unlock();
489 return 0;
490 }
491
492 rcu_read_unlock();
493
494 return -EPERM;
495}
496
497int devcgroup_inode_mknod(int mode, dev_t dev)
498{
499 struct dev_cgroup *dev_cgroup;
500 struct dev_whitelist_item *wh;
501
502 if (!S_ISBLK(mode) && !S_ISCHR(mode))
503 return 0;
504
505 rcu_read_lock();
506
507 dev_cgroup = task_devcgroup(current);
508
509 list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
510 if (wh->type & DEV_ALL)
511 goto found;
512 if ((wh->type & DEV_BLOCK) && !S_ISBLK(mode))
513 continue;
514 if ((wh->type & DEV_CHAR) && !S_ISCHR(mode))
515 continue;
516 if (wh->major != ~0 && wh->major != MAJOR(dev))
517 continue;
518 if (wh->minor != ~0 && wh->minor != MINOR(dev))
519 continue;
520
521 if (!(wh->access & ACC_MKNOD))
522 continue;
523found:
524 rcu_read_unlock();
525 return 0;
526 }
527
528 rcu_read_unlock();
529
530 return -EPERM;
531}