Loading...
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
28enum devcg_behavior {
29 DEVCG_DEFAULT_NONE,
30 DEVCG_DEFAULT_ALLOW,
31 DEVCG_DEFAULT_DENY,
32};
33
34/*
35 * exception list locking rules:
36 * hold devcgroup_mutex for update/read.
37 * hold rcu_read_lock() for read.
38 */
39
40struct dev_exception_item {
41 u32 major, minor;
42 short type;
43 short access;
44 struct list_head list;
45 struct rcu_head rcu;
46};
47
48struct dev_cgroup {
49 struct cgroup_subsys_state css;
50 struct list_head exceptions;
51 enum devcg_behavior behavior;
52};
53
54static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
55{
56 return s ? container_of(s, struct dev_cgroup, css) : NULL;
57}
58
59static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
60{
61 return css_to_devcgroup(task_css(task, devices_cgrp_id));
62}
63
64/*
65 * called under devcgroup_mutex
66 */
67static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
68{
69 struct dev_exception_item *ex, *tmp, *new;
70
71 lockdep_assert_held(&devcgroup_mutex);
72
73 list_for_each_entry(ex, orig, list) {
74 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
75 if (!new)
76 goto free_and_exit;
77 list_add_tail(&new->list, dest);
78 }
79
80 return 0;
81
82free_and_exit:
83 list_for_each_entry_safe(ex, tmp, dest, list) {
84 list_del(&ex->list);
85 kfree(ex);
86 }
87 return -ENOMEM;
88}
89
90/*
91 * called under devcgroup_mutex
92 */
93static int dev_exception_add(struct dev_cgroup *dev_cgroup,
94 struct dev_exception_item *ex)
95{
96 struct dev_exception_item *excopy, *walk;
97
98 lockdep_assert_held(&devcgroup_mutex);
99
100 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
101 if (!excopy)
102 return -ENOMEM;
103
104 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
105 if (walk->type != ex->type)
106 continue;
107 if (walk->major != ex->major)
108 continue;
109 if (walk->minor != ex->minor)
110 continue;
111
112 walk->access |= ex->access;
113 kfree(excopy);
114 excopy = NULL;
115 }
116
117 if (excopy != NULL)
118 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
119 return 0;
120}
121
122/*
123 * called under devcgroup_mutex
124 */
125static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
126 struct dev_exception_item *ex)
127{
128 struct dev_exception_item *walk, *tmp;
129
130 lockdep_assert_held(&devcgroup_mutex);
131
132 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
133 if (walk->type != ex->type)
134 continue;
135 if (walk->major != ex->major)
136 continue;
137 if (walk->minor != ex->minor)
138 continue;
139
140 walk->access &= ~ex->access;
141 if (!walk->access) {
142 list_del_rcu(&walk->list);
143 kfree_rcu(walk, rcu);
144 }
145 }
146}
147
148static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
149{
150 struct dev_exception_item *ex, *tmp;
151
152 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
153 list_del_rcu(&ex->list);
154 kfree_rcu(ex, rcu);
155 }
156}
157
158/**
159 * dev_exception_clean - frees all entries of the exception list
160 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
161 *
162 * called under devcgroup_mutex
163 */
164static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
165{
166 lockdep_assert_held(&devcgroup_mutex);
167
168 __dev_exception_clean(dev_cgroup);
169}
170
171static inline bool is_devcg_online(const struct dev_cgroup *devcg)
172{
173 return (devcg->behavior != DEVCG_DEFAULT_NONE);
174}
175
176/**
177 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
178 * parent's
179 * @css: css getting online
180 * returns 0 in case of success, error code otherwise
181 */
182static int devcgroup_online(struct cgroup_subsys_state *css)
183{
184 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
185 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css_parent(css));
186 int ret = 0;
187
188 mutex_lock(&devcgroup_mutex);
189
190 if (parent_dev_cgroup == NULL)
191 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
192 else {
193 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
194 &parent_dev_cgroup->exceptions);
195 if (!ret)
196 dev_cgroup->behavior = parent_dev_cgroup->behavior;
197 }
198 mutex_unlock(&devcgroup_mutex);
199
200 return ret;
201}
202
203static void devcgroup_offline(struct cgroup_subsys_state *css)
204{
205 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
206
207 mutex_lock(&devcgroup_mutex);
208 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
209 mutex_unlock(&devcgroup_mutex);
210}
211
212/*
213 * called from kernel/cgroup.c with cgroup_lock() held.
214 */
215static struct cgroup_subsys_state *
216devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
217{
218 struct dev_cgroup *dev_cgroup;
219
220 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
221 if (!dev_cgroup)
222 return ERR_PTR(-ENOMEM);
223 INIT_LIST_HEAD(&dev_cgroup->exceptions);
224 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
225
226 return &dev_cgroup->css;
227}
228
229static void devcgroup_css_free(struct cgroup_subsys_state *css)
230{
231 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
232
233 __dev_exception_clean(dev_cgroup);
234 kfree(dev_cgroup);
235}
236
237#define DEVCG_ALLOW 1
238#define DEVCG_DENY 2
239#define DEVCG_LIST 3
240
241#define MAJMINLEN 13
242#define ACCLEN 4
243
244static void set_access(char *acc, short access)
245{
246 int idx = 0;
247 memset(acc, 0, ACCLEN);
248 if (access & ACC_READ)
249 acc[idx++] = 'r';
250 if (access & ACC_WRITE)
251 acc[idx++] = 'w';
252 if (access & ACC_MKNOD)
253 acc[idx++] = 'm';
254}
255
256static char type_to_char(short type)
257{
258 if (type == DEV_ALL)
259 return 'a';
260 if (type == DEV_CHAR)
261 return 'c';
262 if (type == DEV_BLOCK)
263 return 'b';
264 return 'X';
265}
266
267static void set_majmin(char *str, unsigned m)
268{
269 if (m == ~0)
270 strcpy(str, "*");
271 else
272 sprintf(str, "%u", m);
273}
274
275static int devcgroup_seq_show(struct seq_file *m, void *v)
276{
277 struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
278 struct dev_exception_item *ex;
279 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
280
281 rcu_read_lock();
282 /*
283 * To preserve the compatibility:
284 * - Only show the "all devices" when the default policy is to allow
285 * - List the exceptions in case the default policy is to deny
286 * This way, the file remains as a "whitelist of devices"
287 */
288 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
289 set_access(acc, ACC_MASK);
290 set_majmin(maj, ~0);
291 set_majmin(min, ~0);
292 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
293 maj, min, acc);
294 } else {
295 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
296 set_access(acc, ex->access);
297 set_majmin(maj, ex->major);
298 set_majmin(min, ex->minor);
299 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
300 maj, min, acc);
301 }
302 }
303 rcu_read_unlock();
304
305 return 0;
306}
307
308/**
309 * match_exception - iterates the exception list trying to find a complete match
310 * @exceptions: list of exceptions
311 * @type: device type (DEV_BLOCK or DEV_CHAR)
312 * @major: device file major number, ~0 to match all
313 * @minor: device file minor number, ~0 to match all
314 * @access: permission mask (ACC_READ, ACC_WRITE, ACC_MKNOD)
315 *
316 * It is considered a complete match if an exception is found that will
317 * contain the entire range of provided parameters.
318 *
319 * Return: true in case it matches an exception completely
320 */
321static bool match_exception(struct list_head *exceptions, short type,
322 u32 major, u32 minor, short access)
323{
324 struct dev_exception_item *ex;
325
326 list_for_each_entry_rcu(ex, exceptions, list) {
327 if ((type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
328 continue;
329 if ((type & DEV_CHAR) && !(ex->type & DEV_CHAR))
330 continue;
331 if (ex->major != ~0 && ex->major != major)
332 continue;
333 if (ex->minor != ~0 && ex->minor != minor)
334 continue;
335 /* provided access cannot have more than the exception rule */
336 if (access & (~ex->access))
337 continue;
338 return true;
339 }
340 return false;
341}
342
343/**
344 * match_exception_partial - iterates the exception list trying to find a partial match
345 * @exceptions: list of exceptions
346 * @type: device type (DEV_BLOCK or DEV_CHAR)
347 * @major: device file major number, ~0 to match all
348 * @minor: device file minor number, ~0 to match all
349 * @access: permission mask (ACC_READ, ACC_WRITE, ACC_MKNOD)
350 *
351 * It is considered a partial match if an exception's range is found to
352 * contain *any* of the devices specified by provided parameters. This is
353 * used to make sure no extra access is being granted that is forbidden by
354 * any of the exception list.
355 *
356 * Return: true in case the provided range mat matches an exception completely
357 */
358static bool match_exception_partial(struct list_head *exceptions, short type,
359 u32 major, u32 minor, short access)
360{
361 struct dev_exception_item *ex;
362
363 list_for_each_entry_rcu(ex, exceptions, list) {
364 if ((type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
365 continue;
366 if ((type & DEV_CHAR) && !(ex->type & DEV_CHAR))
367 continue;
368 /*
369 * We must be sure that both the exception and the provided
370 * range aren't masking all devices
371 */
372 if (ex->major != ~0 && major != ~0 && ex->major != major)
373 continue;
374 if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
375 continue;
376 /*
377 * In order to make sure the provided range isn't matching
378 * an exception, all its access bits shouldn't match the
379 * exception's access bits
380 */
381 if (!(access & ex->access))
382 continue;
383 return true;
384 }
385 return false;
386}
387
388/**
389 * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
390 * @dev_cgroup: dev cgroup to be tested against
391 * @refex: new exception
392 * @behavior: behavior of the exception's dev_cgroup
393 *
394 * This is used to make sure a child cgroup won't have more privileges
395 * than its parent
396 */
397static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
398 struct dev_exception_item *refex,
399 enum devcg_behavior behavior)
400{
401 bool match = false;
402
403 rcu_lockdep_assert(rcu_read_lock_held() ||
404 lockdep_is_held(&devcgroup_mutex),
405 "device_cgroup:verify_new_ex called without proper synchronization");
406
407 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
408 if (behavior == DEVCG_DEFAULT_ALLOW) {
409 /*
410 * new exception in the child doesn't matter, only
411 * adding extra restrictions
412 */
413 return true;
414 } else {
415 /*
416 * new exception in the child will add more devices
417 * that can be acessed, so it can't match any of
418 * parent's exceptions, even slightly
419 */
420 match = match_exception_partial(&dev_cgroup->exceptions,
421 refex->type,
422 refex->major,
423 refex->minor,
424 refex->access);
425
426 if (match)
427 return false;
428 return true;
429 }
430 } else {
431 /*
432 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
433 * the new exception will add access to more devices and must
434 * be contained completely in an parent's exception to be
435 * allowed
436 */
437 match = match_exception(&dev_cgroup->exceptions, refex->type,
438 refex->major, refex->minor,
439 refex->access);
440
441 if (match)
442 /* parent has an exception that matches the proposed */
443 return true;
444 else
445 return false;
446 }
447 return false;
448}
449
450/*
451 * parent_has_perm:
452 * when adding a new allow rule to a device exception list, the rule
453 * must be allowed in the parent device
454 */
455static int parent_has_perm(struct dev_cgroup *childcg,
456 struct dev_exception_item *ex)
457{
458 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css));
459
460 if (!parent)
461 return 1;
462 return verify_new_ex(parent, ex, childcg->behavior);
463}
464
465/**
466 * parent_allows_removal - verify if it's ok to remove an exception
467 * @childcg: child cgroup from where the exception will be removed
468 * @ex: exception being removed
469 *
470 * When removing an exception in cgroups with default ALLOW policy, it must
471 * be checked if removing it will give the child cgroup more access than the
472 * parent.
473 *
474 * Return: true if it's ok to remove exception, false otherwise
475 */
476static bool parent_allows_removal(struct dev_cgroup *childcg,
477 struct dev_exception_item *ex)
478{
479 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css));
480
481 if (!parent)
482 return true;
483
484 /* It's always allowed to remove access to devices */
485 if (childcg->behavior == DEVCG_DEFAULT_DENY)
486 return true;
487
488 /*
489 * Make sure you're not removing part or a whole exception existing in
490 * the parent cgroup
491 */
492 return !match_exception_partial(&parent->exceptions, ex->type,
493 ex->major, ex->minor, ex->access);
494}
495
496/**
497 * may_allow_all - checks if it's possible to change the behavior to
498 * allow based on parent's rules.
499 * @parent: device cgroup's parent
500 * returns: != 0 in case it's allowed, 0 otherwise
501 */
502static inline int may_allow_all(struct dev_cgroup *parent)
503{
504 if (!parent)
505 return 1;
506 return parent->behavior == DEVCG_DEFAULT_ALLOW;
507}
508
509/**
510 * revalidate_active_exceptions - walks through the active exception list and
511 * revalidates the exceptions based on parent's
512 * behavior and exceptions. The exceptions that
513 * are no longer valid will be removed.
514 * Called with devcgroup_mutex held.
515 * @devcg: cgroup which exceptions will be checked
516 *
517 * This is one of the three key functions for hierarchy implementation.
518 * This function is responsible for re-evaluating all the cgroup's active
519 * exceptions due to a parent's exception change.
520 * Refer to Documentation/cgroups/devices.txt for more details.
521 */
522static void revalidate_active_exceptions(struct dev_cgroup *devcg)
523{
524 struct dev_exception_item *ex;
525 struct list_head *this, *tmp;
526
527 list_for_each_safe(this, tmp, &devcg->exceptions) {
528 ex = container_of(this, struct dev_exception_item, list);
529 if (!parent_has_perm(devcg, ex))
530 dev_exception_rm(devcg, ex);
531 }
532}
533
534/**
535 * propagate_exception - propagates a new exception to the children
536 * @devcg_root: device cgroup that added a new exception
537 * @ex: new exception to be propagated
538 *
539 * returns: 0 in case of success, != 0 in case of error
540 */
541static int propagate_exception(struct dev_cgroup *devcg_root,
542 struct dev_exception_item *ex)
543{
544 struct cgroup_subsys_state *pos;
545 int rc = 0;
546
547 rcu_read_lock();
548
549 css_for_each_descendant_pre(pos, &devcg_root->css) {
550 struct dev_cgroup *devcg = css_to_devcgroup(pos);
551
552 /*
553 * Because devcgroup_mutex is held, no devcg will become
554 * online or offline during the tree walk (see on/offline
555 * methods), and online ones are safe to access outside RCU
556 * read lock without bumping refcnt.
557 */
558 if (pos == &devcg_root->css || !is_devcg_online(devcg))
559 continue;
560
561 rcu_read_unlock();
562
563 /*
564 * in case both root's behavior and devcg is allow, a new
565 * restriction means adding to the exception list
566 */
567 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
568 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
569 rc = dev_exception_add(devcg, ex);
570 if (rc)
571 break;
572 } else {
573 /*
574 * in the other possible cases:
575 * root's behavior: allow, devcg's: deny
576 * root's behavior: deny, devcg's: deny
577 * the exception will be removed
578 */
579 dev_exception_rm(devcg, ex);
580 }
581 revalidate_active_exceptions(devcg);
582
583 rcu_read_lock();
584 }
585
586 rcu_read_unlock();
587 return rc;
588}
589
590static inline bool has_children(struct dev_cgroup *devcgroup)
591{
592 struct cgroup *cgrp = devcgroup->css.cgroup;
593
594 return !list_empty(&cgrp->children);
595}
596
597/*
598 * Modify the exception list using allow/deny rules.
599 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
600 * so we can give a container CAP_MKNOD to let it create devices but not
601 * modify the exception list.
602 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
603 * us to also grant CAP_SYS_ADMIN to containers without giving away the
604 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
605 *
606 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
607 * new access is only allowed if you're in the top-level cgroup, or your
608 * parent cgroup has the access you're asking for.
609 */
610static int devcgroup_update_access(struct dev_cgroup *devcgroup,
611 int filetype, char *buffer)
612{
613 const char *b;
614 char temp[12]; /* 11 + 1 characters needed for a u32 */
615 int count, rc = 0;
616 struct dev_exception_item ex;
617 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&devcgroup->css));
618
619 if (!capable(CAP_SYS_ADMIN))
620 return -EPERM;
621
622 memset(&ex, 0, sizeof(ex));
623 b = buffer;
624
625 switch (*b) {
626 case 'a':
627 switch (filetype) {
628 case DEVCG_ALLOW:
629 if (has_children(devcgroup))
630 return -EINVAL;
631
632 if (!may_allow_all(parent))
633 return -EPERM;
634 dev_exception_clean(devcgroup);
635 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
636 if (!parent)
637 break;
638
639 rc = dev_exceptions_copy(&devcgroup->exceptions,
640 &parent->exceptions);
641 if (rc)
642 return rc;
643 break;
644 case DEVCG_DENY:
645 if (has_children(devcgroup))
646 return -EINVAL;
647
648 dev_exception_clean(devcgroup);
649 devcgroup->behavior = DEVCG_DEFAULT_DENY;
650 break;
651 default:
652 return -EINVAL;
653 }
654 return 0;
655 case 'b':
656 ex.type = DEV_BLOCK;
657 break;
658 case 'c':
659 ex.type = DEV_CHAR;
660 break;
661 default:
662 return -EINVAL;
663 }
664 b++;
665 if (!isspace(*b))
666 return -EINVAL;
667 b++;
668 if (*b == '*') {
669 ex.major = ~0;
670 b++;
671 } else if (isdigit(*b)) {
672 memset(temp, 0, sizeof(temp));
673 for (count = 0; count < sizeof(temp) - 1; count++) {
674 temp[count] = *b;
675 b++;
676 if (!isdigit(*b))
677 break;
678 }
679 rc = kstrtou32(temp, 10, &ex.major);
680 if (rc)
681 return -EINVAL;
682 } else {
683 return -EINVAL;
684 }
685 if (*b != ':')
686 return -EINVAL;
687 b++;
688
689 /* read minor */
690 if (*b == '*') {
691 ex.minor = ~0;
692 b++;
693 } else if (isdigit(*b)) {
694 memset(temp, 0, sizeof(temp));
695 for (count = 0; count < sizeof(temp) - 1; count++) {
696 temp[count] = *b;
697 b++;
698 if (!isdigit(*b))
699 break;
700 }
701 rc = kstrtou32(temp, 10, &ex.minor);
702 if (rc)
703 return -EINVAL;
704 } else {
705 return -EINVAL;
706 }
707 if (!isspace(*b))
708 return -EINVAL;
709 for (b++, count = 0; count < 3; count++, b++) {
710 switch (*b) {
711 case 'r':
712 ex.access |= ACC_READ;
713 break;
714 case 'w':
715 ex.access |= ACC_WRITE;
716 break;
717 case 'm':
718 ex.access |= ACC_MKNOD;
719 break;
720 case '\n':
721 case '\0':
722 count = 3;
723 break;
724 default:
725 return -EINVAL;
726 }
727 }
728
729 switch (filetype) {
730 case DEVCG_ALLOW:
731 /*
732 * If the default policy is to allow by default, try to remove
733 * an matching exception instead. And be silent about it: we
734 * don't want to break compatibility
735 */
736 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
737 /* Check if the parent allows removing it first */
738 if (!parent_allows_removal(devcgroup, &ex))
739 return -EPERM;
740 dev_exception_rm(devcgroup, &ex);
741 break;
742 }
743
744 if (!parent_has_perm(devcgroup, &ex))
745 return -EPERM;
746 rc = dev_exception_add(devcgroup, &ex);
747 break;
748 case DEVCG_DENY:
749 /*
750 * If the default policy is to deny by default, try to remove
751 * an matching exception instead. And be silent about it: we
752 * don't want to break compatibility
753 */
754 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
755 dev_exception_rm(devcgroup, &ex);
756 else
757 rc = dev_exception_add(devcgroup, &ex);
758
759 if (rc)
760 break;
761 /* we only propagate new restrictions */
762 rc = propagate_exception(devcgroup, &ex);
763 break;
764 default:
765 rc = -EINVAL;
766 }
767 return rc;
768}
769
770static int devcgroup_access_write(struct cgroup_subsys_state *css,
771 struct cftype *cft, char *buffer)
772{
773 int retval;
774
775 mutex_lock(&devcgroup_mutex);
776 retval = devcgroup_update_access(css_to_devcgroup(css),
777 cft->private, buffer);
778 mutex_unlock(&devcgroup_mutex);
779 return retval;
780}
781
782static struct cftype dev_cgroup_files[] = {
783 {
784 .name = "allow",
785 .write_string = devcgroup_access_write,
786 .private = DEVCG_ALLOW,
787 },
788 {
789 .name = "deny",
790 .write_string = devcgroup_access_write,
791 .private = DEVCG_DENY,
792 },
793 {
794 .name = "list",
795 .seq_show = devcgroup_seq_show,
796 .private = DEVCG_LIST,
797 },
798 { } /* terminate */
799};
800
801struct cgroup_subsys devices_cgrp_subsys = {
802 .css_alloc = devcgroup_css_alloc,
803 .css_free = devcgroup_css_free,
804 .css_online = devcgroup_online,
805 .css_offline = devcgroup_offline,
806 .base_cftypes = dev_cgroup_files,
807};
808
809/**
810 * __devcgroup_check_permission - checks if an inode operation is permitted
811 * @dev_cgroup: the dev cgroup to be tested against
812 * @type: device type
813 * @major: device major number
814 * @minor: device minor number
815 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
816 *
817 * returns 0 on success, -EPERM case the operation is not permitted
818 */
819static int __devcgroup_check_permission(short type, u32 major, u32 minor,
820 short access)
821{
822 struct dev_cgroup *dev_cgroup;
823 bool rc;
824
825 rcu_read_lock();
826 dev_cgroup = task_devcgroup(current);
827 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
828 /* Can't match any of the exceptions, even partially */
829 rc = !match_exception_partial(&dev_cgroup->exceptions,
830 type, major, minor, access);
831 else
832 /* Need to match completely one exception to be allowed */
833 rc = match_exception(&dev_cgroup->exceptions, type, major,
834 minor, access);
835 rcu_read_unlock();
836
837 if (!rc)
838 return -EPERM;
839
840 return 0;
841}
842
843int __devcgroup_inode_permission(struct inode *inode, int mask)
844{
845 short type, access = 0;
846
847 if (S_ISBLK(inode->i_mode))
848 type = DEV_BLOCK;
849 if (S_ISCHR(inode->i_mode))
850 type = DEV_CHAR;
851 if (mask & MAY_WRITE)
852 access |= ACC_WRITE;
853 if (mask & MAY_READ)
854 access |= ACC_READ;
855
856 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
857 access);
858}
859
860int devcgroup_inode_mknod(int mode, dev_t dev)
861{
862 short type;
863
864 if (!S_ISBLK(mode) && !S_ISCHR(mode))
865 return 0;
866
867 if (S_ISBLK(mode))
868 type = DEV_BLOCK;
869 else
870 type = DEV_CHAR;
871
872 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),
873 ACC_MKNOD);
874
875}
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}