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
18#ifdef CONFIG_CGROUP_DEVICE
19
20static DEFINE_MUTEX(devcgroup_mutex);
21
22enum devcg_behavior {
23 DEVCG_DEFAULT_NONE,
24 DEVCG_DEFAULT_ALLOW,
25 DEVCG_DEFAULT_DENY,
26};
27
28/*
29 * exception list locking rules:
30 * hold devcgroup_mutex for update/read.
31 * hold rcu_read_lock() for read.
32 */
33
34struct dev_exception_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 exceptions;
45 enum devcg_behavior behavior;
46};
47
48static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
49{
50 return s ? container_of(s, struct dev_cgroup, css) : NULL;
51}
52
53static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
54{
55 return css_to_devcgroup(task_css(task, devices_cgrp_id));
56}
57
58/*
59 * called under devcgroup_mutex
60 */
61static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
62{
63 struct dev_exception_item *ex, *tmp, *new;
64
65 lockdep_assert_held(&devcgroup_mutex);
66
67 list_for_each_entry(ex, orig, list) {
68 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
69 if (!new)
70 goto free_and_exit;
71 list_add_tail(&new->list, dest);
72 }
73
74 return 0;
75
76free_and_exit:
77 list_for_each_entry_safe(ex, tmp, dest, list) {
78 list_del(&ex->list);
79 kfree(ex);
80 }
81 return -ENOMEM;
82}
83
84/*
85 * called under devcgroup_mutex
86 */
87static int dev_exception_add(struct dev_cgroup *dev_cgroup,
88 struct dev_exception_item *ex)
89{
90 struct dev_exception_item *excopy, *walk;
91
92 lockdep_assert_held(&devcgroup_mutex);
93
94 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
95 if (!excopy)
96 return -ENOMEM;
97
98 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
99 if (walk->type != ex->type)
100 continue;
101 if (walk->major != ex->major)
102 continue;
103 if (walk->minor != ex->minor)
104 continue;
105
106 walk->access |= ex->access;
107 kfree(excopy);
108 excopy = NULL;
109 }
110
111 if (excopy != NULL)
112 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
113 return 0;
114}
115
116/*
117 * called under devcgroup_mutex
118 */
119static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
120 struct dev_exception_item *ex)
121{
122 struct dev_exception_item *walk, *tmp;
123
124 lockdep_assert_held(&devcgroup_mutex);
125
126 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
127 if (walk->type != ex->type)
128 continue;
129 if (walk->major != ex->major)
130 continue;
131 if (walk->minor != ex->minor)
132 continue;
133
134 walk->access &= ~ex->access;
135 if (!walk->access) {
136 list_del_rcu(&walk->list);
137 kfree_rcu(walk, rcu);
138 }
139 }
140}
141
142static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
143{
144 struct dev_exception_item *ex, *tmp;
145
146 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
147 list_del_rcu(&ex->list);
148 kfree_rcu(ex, rcu);
149 }
150}
151
152/**
153 * dev_exception_clean - frees all entries of the exception list
154 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
155 *
156 * called under devcgroup_mutex
157 */
158static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
159{
160 lockdep_assert_held(&devcgroup_mutex);
161
162 __dev_exception_clean(dev_cgroup);
163}
164
165static inline bool is_devcg_online(const struct dev_cgroup *devcg)
166{
167 return (devcg->behavior != DEVCG_DEFAULT_NONE);
168}
169
170/**
171 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
172 * parent's
173 * @css: css getting online
174 * returns 0 in case of success, error code otherwise
175 */
176static int devcgroup_online(struct cgroup_subsys_state *css)
177{
178 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
179 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
180 int ret = 0;
181
182 mutex_lock(&devcgroup_mutex);
183
184 if (parent_dev_cgroup == NULL)
185 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
186 else {
187 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
188 &parent_dev_cgroup->exceptions);
189 if (!ret)
190 dev_cgroup->behavior = parent_dev_cgroup->behavior;
191 }
192 mutex_unlock(&devcgroup_mutex);
193
194 return ret;
195}
196
197static void devcgroup_offline(struct cgroup_subsys_state *css)
198{
199 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
200
201 mutex_lock(&devcgroup_mutex);
202 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
203 mutex_unlock(&devcgroup_mutex);
204}
205
206/*
207 * called from kernel/cgroup.c with cgroup_lock() held.
208 */
209static struct cgroup_subsys_state *
210devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
211{
212 struct dev_cgroup *dev_cgroup;
213
214 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
215 if (!dev_cgroup)
216 return ERR_PTR(-ENOMEM);
217 INIT_LIST_HEAD(&dev_cgroup->exceptions);
218 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
219
220 return &dev_cgroup->css;
221}
222
223static void devcgroup_css_free(struct cgroup_subsys_state *css)
224{
225 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
226
227 __dev_exception_clean(dev_cgroup);
228 kfree(dev_cgroup);
229}
230
231#define DEVCG_ALLOW 1
232#define DEVCG_DENY 2
233#define DEVCG_LIST 3
234
235#define MAJMINLEN 13
236#define ACCLEN 4
237
238static void set_access(char *acc, short access)
239{
240 int idx = 0;
241 memset(acc, 0, ACCLEN);
242 if (access & DEVCG_ACC_READ)
243 acc[idx++] = 'r';
244 if (access & DEVCG_ACC_WRITE)
245 acc[idx++] = 'w';
246 if (access & DEVCG_ACC_MKNOD)
247 acc[idx++] = 'm';
248}
249
250static char type_to_char(short type)
251{
252 if (type == DEVCG_DEV_ALL)
253 return 'a';
254 if (type == DEVCG_DEV_CHAR)
255 return 'c';
256 if (type == DEVCG_DEV_BLOCK)
257 return 'b';
258 return 'X';
259}
260
261static void set_majmin(char *str, unsigned m)
262{
263 if (m == ~0)
264 strcpy(str, "*");
265 else
266 sprintf(str, "%u", m);
267}
268
269static int devcgroup_seq_show(struct seq_file *m, void *v)
270{
271 struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
272 struct dev_exception_item *ex;
273 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
274
275 rcu_read_lock();
276 /*
277 * To preserve the compatibility:
278 * - Only show the "all devices" when the default policy is to allow
279 * - List the exceptions in case the default policy is to deny
280 * This way, the file remains as a "whitelist of devices"
281 */
282 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
283 set_access(acc, DEVCG_ACC_MASK);
284 set_majmin(maj, ~0);
285 set_majmin(min, ~0);
286 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEVCG_DEV_ALL),
287 maj, min, acc);
288 } else {
289 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
290 set_access(acc, ex->access);
291 set_majmin(maj, ex->major);
292 set_majmin(min, ex->minor);
293 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
294 maj, min, acc);
295 }
296 }
297 rcu_read_unlock();
298
299 return 0;
300}
301
302/**
303 * match_exception - iterates the exception list trying to find a complete match
304 * @exceptions: list of exceptions
305 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
306 * @major: device file major number, ~0 to match all
307 * @minor: device file minor number, ~0 to match all
308 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
309 *
310 * It is considered a complete match if an exception is found that will
311 * contain the entire range of provided parameters.
312 *
313 * Return: true in case it matches an exception completely
314 */
315static bool match_exception(struct list_head *exceptions, short type,
316 u32 major, u32 minor, short access)
317{
318 struct dev_exception_item *ex;
319
320 list_for_each_entry_rcu(ex, exceptions, list) {
321 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
322 continue;
323 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
324 continue;
325 if (ex->major != ~0 && ex->major != major)
326 continue;
327 if (ex->minor != ~0 && ex->minor != minor)
328 continue;
329 /* provided access cannot have more than the exception rule */
330 if (access & (~ex->access))
331 continue;
332 return true;
333 }
334 return false;
335}
336
337/**
338 * match_exception_partial - iterates the exception list trying to find a partial match
339 * @exceptions: list of exceptions
340 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
341 * @major: device file major number, ~0 to match all
342 * @minor: device file minor number, ~0 to match all
343 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
344 *
345 * It is considered a partial match if an exception's range is found to
346 * contain *any* of the devices specified by provided parameters. This is
347 * used to make sure no extra access is being granted that is forbidden by
348 * any of the exception list.
349 *
350 * Return: true in case the provided range mat matches an exception completely
351 */
352static bool match_exception_partial(struct list_head *exceptions, short type,
353 u32 major, u32 minor, short access)
354{
355 struct dev_exception_item *ex;
356
357 list_for_each_entry_rcu(ex, exceptions, list,
358 lockdep_is_held(&devcgroup_mutex)) {
359 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
360 continue;
361 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
362 continue;
363 /*
364 * We must be sure that both the exception and the provided
365 * range aren't masking all devices
366 */
367 if (ex->major != ~0 && major != ~0 && ex->major != major)
368 continue;
369 if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
370 continue;
371 /*
372 * In order to make sure the provided range isn't matching
373 * an exception, all its access bits shouldn't match the
374 * exception's access bits
375 */
376 if (!(access & ex->access))
377 continue;
378 return true;
379 }
380 return false;
381}
382
383/**
384 * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
385 * @dev_cgroup: dev cgroup to be tested against
386 * @refex: new exception
387 * @behavior: behavior of the exception's dev_cgroup
388 *
389 * This is used to make sure a child cgroup won't have more privileges
390 * than its parent
391 */
392static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
393 struct dev_exception_item *refex,
394 enum devcg_behavior behavior)
395{
396 bool match = false;
397
398 RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
399 !lockdep_is_held(&devcgroup_mutex),
400 "device_cgroup:verify_new_ex called without proper synchronization");
401
402 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
403 if (behavior == DEVCG_DEFAULT_ALLOW) {
404 /*
405 * new exception in the child doesn't matter, only
406 * adding extra restrictions
407 */
408 return true;
409 } else {
410 /*
411 * new exception in the child will add more devices
412 * that can be acessed, so it can't match any of
413 * parent's exceptions, even slightly
414 */
415 match = match_exception_partial(&dev_cgroup->exceptions,
416 refex->type,
417 refex->major,
418 refex->minor,
419 refex->access);
420
421 if (match)
422 return false;
423 return true;
424 }
425 } else {
426 /*
427 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
428 * the new exception will add access to more devices and must
429 * be contained completely in an parent's exception to be
430 * allowed
431 */
432 match = match_exception(&dev_cgroup->exceptions, refex->type,
433 refex->major, refex->minor,
434 refex->access);
435
436 if (match)
437 /* parent has an exception that matches the proposed */
438 return true;
439 else
440 return false;
441 }
442 return false;
443}
444
445/*
446 * parent_has_perm:
447 * when adding a new allow rule to a device exception list, the rule
448 * must be allowed in the parent device
449 */
450static int parent_has_perm(struct dev_cgroup *childcg,
451 struct dev_exception_item *ex)
452{
453 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
454
455 if (!parent)
456 return 1;
457 return verify_new_ex(parent, ex, childcg->behavior);
458}
459
460/**
461 * parent_allows_removal - verify if it's ok to remove an exception
462 * @childcg: child cgroup from where the exception will be removed
463 * @ex: exception being removed
464 *
465 * When removing an exception in cgroups with default ALLOW policy, it must
466 * be checked if removing it will give the child cgroup more access than the
467 * parent.
468 *
469 * Return: true if it's ok to remove exception, false otherwise
470 */
471static bool parent_allows_removal(struct dev_cgroup *childcg,
472 struct dev_exception_item *ex)
473{
474 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
475
476 if (!parent)
477 return true;
478
479 /* It's always allowed to remove access to devices */
480 if (childcg->behavior == DEVCG_DEFAULT_DENY)
481 return true;
482
483 /*
484 * Make sure you're not removing part or a whole exception existing in
485 * the parent cgroup
486 */
487 return !match_exception_partial(&parent->exceptions, ex->type,
488 ex->major, ex->minor, ex->access);
489}
490
491/**
492 * may_allow_all - checks if it's possible to change the behavior to
493 * allow based on parent's rules.
494 * @parent: device cgroup's parent
495 * returns: != 0 in case it's allowed, 0 otherwise
496 */
497static inline int may_allow_all(struct dev_cgroup *parent)
498{
499 if (!parent)
500 return 1;
501 return parent->behavior == DEVCG_DEFAULT_ALLOW;
502}
503
504/**
505 * revalidate_active_exceptions - walks through the active exception list and
506 * revalidates the exceptions based on parent's
507 * behavior and exceptions. The exceptions that
508 * are no longer valid will be removed.
509 * Called with devcgroup_mutex held.
510 * @devcg: cgroup which exceptions will be checked
511 *
512 * This is one of the three key functions for hierarchy implementation.
513 * This function is responsible for re-evaluating all the cgroup's active
514 * exceptions due to a parent's exception change.
515 * Refer to Documentation/admin-guide/cgroup-v1/devices.rst for more details.
516 */
517static void revalidate_active_exceptions(struct dev_cgroup *devcg)
518{
519 struct dev_exception_item *ex;
520 struct list_head *this, *tmp;
521
522 list_for_each_safe(this, tmp, &devcg->exceptions) {
523 ex = container_of(this, struct dev_exception_item, list);
524 if (!parent_has_perm(devcg, ex))
525 dev_exception_rm(devcg, ex);
526 }
527}
528
529/**
530 * propagate_exception - propagates a new exception to the children
531 * @devcg_root: device cgroup that added a new exception
532 * @ex: new exception to be propagated
533 *
534 * returns: 0 in case of success, != 0 in case of error
535 */
536static int propagate_exception(struct dev_cgroup *devcg_root,
537 struct dev_exception_item *ex)
538{
539 struct cgroup_subsys_state *pos;
540 int rc = 0;
541
542 rcu_read_lock();
543
544 css_for_each_descendant_pre(pos, &devcg_root->css) {
545 struct dev_cgroup *devcg = css_to_devcgroup(pos);
546
547 /*
548 * Because devcgroup_mutex is held, no devcg will become
549 * online or offline during the tree walk (see on/offline
550 * methods), and online ones are safe to access outside RCU
551 * read lock without bumping refcnt.
552 */
553 if (pos == &devcg_root->css || !is_devcg_online(devcg))
554 continue;
555
556 rcu_read_unlock();
557
558 /*
559 * in case both root's behavior and devcg is allow, a new
560 * restriction means adding to the exception list
561 */
562 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
563 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
564 rc = dev_exception_add(devcg, ex);
565 if (rc)
566 return rc;
567 } else {
568 /*
569 * in the other possible cases:
570 * root's behavior: allow, devcg's: deny
571 * root's behavior: deny, devcg's: deny
572 * the exception will be removed
573 */
574 dev_exception_rm(devcg, ex);
575 }
576 revalidate_active_exceptions(devcg);
577
578 rcu_read_lock();
579 }
580
581 rcu_read_unlock();
582 return rc;
583}
584
585/*
586 * Modify the exception list using allow/deny rules.
587 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
588 * so we can give a container CAP_MKNOD to let it create devices but not
589 * modify the exception list.
590 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
591 * us to also grant CAP_SYS_ADMIN to containers without giving away the
592 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
593 *
594 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
595 * new access is only allowed if you're in the top-level cgroup, or your
596 * parent cgroup has the access you're asking for.
597 */
598static int devcgroup_update_access(struct dev_cgroup *devcgroup,
599 int filetype, char *buffer)
600{
601 const char *b;
602 char temp[12]; /* 11 + 1 characters needed for a u32 */
603 int count, rc = 0;
604 struct dev_exception_item ex;
605 struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
606
607 if (!capable(CAP_SYS_ADMIN))
608 return -EPERM;
609
610 memset(&ex, 0, sizeof(ex));
611 b = buffer;
612
613 switch (*b) {
614 case 'a':
615 switch (filetype) {
616 case DEVCG_ALLOW:
617 if (css_has_online_children(&devcgroup->css))
618 return -EINVAL;
619
620 if (!may_allow_all(parent))
621 return -EPERM;
622 dev_exception_clean(devcgroup);
623 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
624 if (!parent)
625 break;
626
627 rc = dev_exceptions_copy(&devcgroup->exceptions,
628 &parent->exceptions);
629 if (rc)
630 return rc;
631 break;
632 case DEVCG_DENY:
633 if (css_has_online_children(&devcgroup->css))
634 return -EINVAL;
635
636 dev_exception_clean(devcgroup);
637 devcgroup->behavior = DEVCG_DEFAULT_DENY;
638 break;
639 default:
640 return -EINVAL;
641 }
642 return 0;
643 case 'b':
644 ex.type = DEVCG_DEV_BLOCK;
645 break;
646 case 'c':
647 ex.type = DEVCG_DEV_CHAR;
648 break;
649 default:
650 return -EINVAL;
651 }
652 b++;
653 if (!isspace(*b))
654 return -EINVAL;
655 b++;
656 if (*b == '*') {
657 ex.major = ~0;
658 b++;
659 } else if (isdigit(*b)) {
660 memset(temp, 0, sizeof(temp));
661 for (count = 0; count < sizeof(temp) - 1; count++) {
662 temp[count] = *b;
663 b++;
664 if (!isdigit(*b))
665 break;
666 }
667 rc = kstrtou32(temp, 10, &ex.major);
668 if (rc)
669 return -EINVAL;
670 } else {
671 return -EINVAL;
672 }
673 if (*b != ':')
674 return -EINVAL;
675 b++;
676
677 /* read minor */
678 if (*b == '*') {
679 ex.minor = ~0;
680 b++;
681 } else if (isdigit(*b)) {
682 memset(temp, 0, sizeof(temp));
683 for (count = 0; count < sizeof(temp) - 1; count++) {
684 temp[count] = *b;
685 b++;
686 if (!isdigit(*b))
687 break;
688 }
689 rc = kstrtou32(temp, 10, &ex.minor);
690 if (rc)
691 return -EINVAL;
692 } else {
693 return -EINVAL;
694 }
695 if (!isspace(*b))
696 return -EINVAL;
697 for (b++, count = 0; count < 3; count++, b++) {
698 switch (*b) {
699 case 'r':
700 ex.access |= DEVCG_ACC_READ;
701 break;
702 case 'w':
703 ex.access |= DEVCG_ACC_WRITE;
704 break;
705 case 'm':
706 ex.access |= DEVCG_ACC_MKNOD;
707 break;
708 case '\n':
709 case '\0':
710 count = 3;
711 break;
712 default:
713 return -EINVAL;
714 }
715 }
716
717 switch (filetype) {
718 case DEVCG_ALLOW:
719 /*
720 * If the default policy is to allow by default, try to remove
721 * an matching exception instead. And be silent about it: we
722 * don't want to break compatibility
723 */
724 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
725 /* Check if the parent allows removing it first */
726 if (!parent_allows_removal(devcgroup, &ex))
727 return -EPERM;
728 dev_exception_rm(devcgroup, &ex);
729 break;
730 }
731
732 if (!parent_has_perm(devcgroup, &ex))
733 return -EPERM;
734 rc = dev_exception_add(devcgroup, &ex);
735 break;
736 case DEVCG_DENY:
737 /*
738 * If the default policy is to deny by default, try to remove
739 * an matching exception instead. And be silent about it: we
740 * don't want to break compatibility
741 */
742 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
743 dev_exception_rm(devcgroup, &ex);
744 else
745 rc = dev_exception_add(devcgroup, &ex);
746
747 if (rc)
748 break;
749 /* we only propagate new restrictions */
750 rc = propagate_exception(devcgroup, &ex);
751 break;
752 default:
753 rc = -EINVAL;
754 }
755 return rc;
756}
757
758static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
759 char *buf, size_t nbytes, loff_t off)
760{
761 int retval;
762
763 mutex_lock(&devcgroup_mutex);
764 retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
765 of_cft(of)->private, strstrip(buf));
766 mutex_unlock(&devcgroup_mutex);
767 return retval ?: nbytes;
768}
769
770static struct cftype dev_cgroup_files[] = {
771 {
772 .name = "allow",
773 .write = devcgroup_access_write,
774 .private = DEVCG_ALLOW,
775 },
776 {
777 .name = "deny",
778 .write = devcgroup_access_write,
779 .private = DEVCG_DENY,
780 },
781 {
782 .name = "list",
783 .seq_show = devcgroup_seq_show,
784 .private = DEVCG_LIST,
785 },
786 { } /* terminate */
787};
788
789struct cgroup_subsys devices_cgrp_subsys = {
790 .css_alloc = devcgroup_css_alloc,
791 .css_free = devcgroup_css_free,
792 .css_online = devcgroup_online,
793 .css_offline = devcgroup_offline,
794 .legacy_cftypes = dev_cgroup_files,
795};
796
797/**
798 * devcgroup_legacy_check_permission - checks if an inode operation is permitted
799 * @dev_cgroup: the dev cgroup to be tested against
800 * @type: device type
801 * @major: device major number
802 * @minor: device minor number
803 * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
804 *
805 * returns 0 on success, -EPERM case the operation is not permitted
806 */
807static int devcgroup_legacy_check_permission(short type, u32 major, u32 minor,
808 short access)
809{
810 struct dev_cgroup *dev_cgroup;
811 bool rc;
812
813 rcu_read_lock();
814 dev_cgroup = task_devcgroup(current);
815 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
816 /* Can't match any of the exceptions, even partially */
817 rc = !match_exception_partial(&dev_cgroup->exceptions,
818 type, major, minor, access);
819 else
820 /* Need to match completely one exception to be allowed */
821 rc = match_exception(&dev_cgroup->exceptions, type, major,
822 minor, access);
823 rcu_read_unlock();
824
825 if (!rc)
826 return -EPERM;
827
828 return 0;
829}
830
831#endif /* CONFIG_CGROUP_DEVICE */
832
833#if defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF)
834
835int devcgroup_check_permission(short type, u32 major, u32 minor, short access)
836{
837 int rc = BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type, major, minor, access);
838
839 if (rc)
840 return -EPERM;
841
842 #ifdef CONFIG_CGROUP_DEVICE
843 return devcgroup_legacy_check_permission(type, major, minor, access);
844
845 #else /* CONFIG_CGROUP_DEVICE */
846 return 0;
847
848 #endif /* CONFIG_CGROUP_DEVICE */
849}
850EXPORT_SYMBOL(devcgroup_check_permission);
851#endif /* defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF) */