Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * device_cgroup.c - device cgroup subsystem
4 *
5 * Copyright 2007 IBM Corp
6 */
7
8#include <linux/bpf-cgroup.h>
9#include <linux/device_cgroup.h>
10#include <linux/cgroup.h>
11#include <linux/ctype.h>
12#include <linux/list.h>
13#include <linux/uaccess.h>
14#include <linux/seq_file.h>
15#include <linux/slab.h>
16#include <linux/rcupdate.h>
17#include <linux/mutex.h>
18
19#ifdef CONFIG_CGROUP_DEVICE
20
21static DEFINE_MUTEX(devcgroup_mutex);
22
23enum devcg_behavior {
24 DEVCG_DEFAULT_NONE,
25 DEVCG_DEFAULT_ALLOW,
26 DEVCG_DEFAULT_DENY,
27};
28
29/*
30 * exception list locking rules:
31 * hold devcgroup_mutex for update/read.
32 * hold rcu_read_lock() for read.
33 */
34
35struct dev_exception_item {
36 u32 major, minor;
37 short type;
38 short access;
39 struct list_head list;
40 struct rcu_head rcu;
41};
42
43struct dev_cgroup {
44 struct cgroup_subsys_state css;
45 struct list_head exceptions;
46 enum devcg_behavior behavior;
47};
48
49static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
50{
51 return s ? container_of(s, struct dev_cgroup, css) : NULL;
52}
53
54static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
55{
56 return css_to_devcgroup(task_css(task, devices_cgrp_id));
57}
58
59/*
60 * called under devcgroup_mutex
61 */
62static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
63{
64 struct dev_exception_item *ex, *tmp, *new;
65
66 lockdep_assert_held(&devcgroup_mutex);
67
68 list_for_each_entry(ex, orig, list) {
69 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
70 if (!new)
71 goto free_and_exit;
72 list_add_tail(&new->list, dest);
73 }
74
75 return 0;
76
77free_and_exit:
78 list_for_each_entry_safe(ex, tmp, dest, list) {
79 list_del(&ex->list);
80 kfree(ex);
81 }
82 return -ENOMEM;
83}
84
85static void dev_exceptions_move(struct list_head *dest, struct list_head *orig)
86{
87 struct dev_exception_item *ex, *tmp;
88
89 lockdep_assert_held(&devcgroup_mutex);
90
91 list_for_each_entry_safe(ex, tmp, orig, list) {
92 list_move_tail(&ex->list, dest);
93 }
94}
95
96/*
97 * called under devcgroup_mutex
98 */
99static int dev_exception_add(struct dev_cgroup *dev_cgroup,
100 struct dev_exception_item *ex)
101{
102 struct dev_exception_item *excopy, *walk;
103
104 lockdep_assert_held(&devcgroup_mutex);
105
106 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
107 if (!excopy)
108 return -ENOMEM;
109
110 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
111 if (walk->type != ex->type)
112 continue;
113 if (walk->major != ex->major)
114 continue;
115 if (walk->minor != ex->minor)
116 continue;
117
118 walk->access |= ex->access;
119 kfree(excopy);
120 excopy = NULL;
121 }
122
123 if (excopy != NULL)
124 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
125 return 0;
126}
127
128/*
129 * called under devcgroup_mutex
130 */
131static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
132 struct dev_exception_item *ex)
133{
134 struct dev_exception_item *walk, *tmp;
135
136 lockdep_assert_held(&devcgroup_mutex);
137
138 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
139 if (walk->type != ex->type)
140 continue;
141 if (walk->major != ex->major)
142 continue;
143 if (walk->minor != ex->minor)
144 continue;
145
146 walk->access &= ~ex->access;
147 if (!walk->access) {
148 list_del_rcu(&walk->list);
149 kfree_rcu(walk, rcu);
150 }
151 }
152}
153
154static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
155{
156 struct dev_exception_item *ex, *tmp;
157
158 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
159 list_del_rcu(&ex->list);
160 kfree_rcu(ex, rcu);
161 }
162}
163
164/**
165 * dev_exception_clean - frees all entries of the exception list
166 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
167 *
168 * called under devcgroup_mutex
169 */
170static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
171{
172 lockdep_assert_held(&devcgroup_mutex);
173
174 __dev_exception_clean(dev_cgroup);
175}
176
177static inline bool is_devcg_online(const struct dev_cgroup *devcg)
178{
179 return (devcg->behavior != DEVCG_DEFAULT_NONE);
180}
181
182/**
183 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
184 * parent's
185 * @css: css getting online
186 * returns 0 in case of success, error code otherwise
187 */
188static int devcgroup_online(struct cgroup_subsys_state *css)
189{
190 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
191 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
192 int ret = 0;
193
194 mutex_lock(&devcgroup_mutex);
195
196 if (parent_dev_cgroup == NULL)
197 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
198 else {
199 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
200 &parent_dev_cgroup->exceptions);
201 if (!ret)
202 dev_cgroup->behavior = parent_dev_cgroup->behavior;
203 }
204 mutex_unlock(&devcgroup_mutex);
205
206 return ret;
207}
208
209static void devcgroup_offline(struct cgroup_subsys_state *css)
210{
211 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
212
213 mutex_lock(&devcgroup_mutex);
214 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
215 mutex_unlock(&devcgroup_mutex);
216}
217
218/*
219 * called from kernel/cgroup/cgroup.c with cgroup_lock() held.
220 */
221static struct cgroup_subsys_state *
222devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
223{
224 struct dev_cgroup *dev_cgroup;
225
226 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
227 if (!dev_cgroup)
228 return ERR_PTR(-ENOMEM);
229 INIT_LIST_HEAD(&dev_cgroup->exceptions);
230 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
231
232 return &dev_cgroup->css;
233}
234
235static void devcgroup_css_free(struct cgroup_subsys_state *css)
236{
237 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
238
239 __dev_exception_clean(dev_cgroup);
240 kfree(dev_cgroup);
241}
242
243#define DEVCG_ALLOW 1
244#define DEVCG_DENY 2
245#define DEVCG_LIST 3
246
247#define MAJMINLEN 13
248#define ACCLEN 4
249
250static void set_access(char *acc, short access)
251{
252 int idx = 0;
253 memset(acc, 0, ACCLEN);
254 if (access & DEVCG_ACC_READ)
255 acc[idx++] = 'r';
256 if (access & DEVCG_ACC_WRITE)
257 acc[idx++] = 'w';
258 if (access & DEVCG_ACC_MKNOD)
259 acc[idx++] = 'm';
260}
261
262static char type_to_char(short type)
263{
264 if (type == DEVCG_DEV_ALL)
265 return 'a';
266 if (type == DEVCG_DEV_CHAR)
267 return 'c';
268 if (type == DEVCG_DEV_BLOCK)
269 return 'b';
270 return 'X';
271}
272
273static void set_majmin(char *str, unsigned m)
274{
275 if (m == ~0)
276 strcpy(str, "*");
277 else
278 sprintf(str, "%u", m);
279}
280
281static int devcgroup_seq_show(struct seq_file *m, void *v)
282{
283 struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
284 struct dev_exception_item *ex;
285 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
286
287 rcu_read_lock();
288 /*
289 * To preserve the compatibility:
290 * - Only show the "all devices" when the default policy is to allow
291 * - List the exceptions in case the default policy is to deny
292 * This way, the file remains as a "whitelist of devices"
293 */
294 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
295 set_access(acc, DEVCG_ACC_MASK);
296 set_majmin(maj, ~0);
297 set_majmin(min, ~0);
298 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEVCG_DEV_ALL),
299 maj, min, acc);
300 } else {
301 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
302 set_access(acc, ex->access);
303 set_majmin(maj, ex->major);
304 set_majmin(min, ex->minor);
305 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
306 maj, min, acc);
307 }
308 }
309 rcu_read_unlock();
310
311 return 0;
312}
313
314/**
315 * match_exception - iterates the exception list trying to find a complete match
316 * @exceptions: list of exceptions
317 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
318 * @major: device file major number, ~0 to match all
319 * @minor: device file minor number, ~0 to match all
320 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
321 *
322 * It is considered a complete match if an exception is found that will
323 * contain the entire range of provided parameters.
324 *
325 * Return: true in case it matches an exception completely
326 */
327static bool match_exception(struct list_head *exceptions, short type,
328 u32 major, u32 minor, short access)
329{
330 struct dev_exception_item *ex;
331
332 list_for_each_entry_rcu(ex, exceptions, list) {
333 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
334 continue;
335 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
336 continue;
337 if (ex->major != ~0 && ex->major != major)
338 continue;
339 if (ex->minor != ~0 && ex->minor != minor)
340 continue;
341 /* provided access cannot have more than the exception rule */
342 if (access & (~ex->access))
343 continue;
344 return true;
345 }
346 return false;
347}
348
349/**
350 * match_exception_partial - iterates the exception list trying to find a partial match
351 * @exceptions: list of exceptions
352 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
353 * @major: device file major number, ~0 to match all
354 * @minor: device file minor number, ~0 to match all
355 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
356 *
357 * It is considered a partial match if an exception's range is found to
358 * contain *any* of the devices specified by provided parameters. This is
359 * used to make sure no extra access is being granted that is forbidden by
360 * any of the exception list.
361 *
362 * Return: true in case the provided range mat matches an exception completely
363 */
364static bool match_exception_partial(struct list_head *exceptions, short type,
365 u32 major, u32 minor, short access)
366{
367 struct dev_exception_item *ex;
368
369 list_for_each_entry_rcu(ex, exceptions, list,
370 lockdep_is_held(&devcgroup_mutex)) {
371 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
372 continue;
373 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
374 continue;
375 /*
376 * We must be sure that both the exception and the provided
377 * range aren't masking all devices
378 */
379 if (ex->major != ~0 && major != ~0 && ex->major != major)
380 continue;
381 if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
382 continue;
383 /*
384 * In order to make sure the provided range isn't matching
385 * an exception, all its access bits shouldn't match the
386 * exception's access bits
387 */
388 if (!(access & ex->access))
389 continue;
390 return true;
391 }
392 return false;
393}
394
395/**
396 * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
397 * @dev_cgroup: dev cgroup to be tested against
398 * @refex: new exception
399 * @behavior: behavior of the exception's dev_cgroup
400 *
401 * This is used to make sure a child cgroup won't have more privileges
402 * than its parent
403 */
404static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
405 struct dev_exception_item *refex,
406 enum devcg_behavior behavior)
407{
408 bool match = false;
409
410 RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
411 !lockdep_is_held(&devcgroup_mutex),
412 "device_cgroup:verify_new_ex called without proper synchronization");
413
414 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
415 if (behavior == DEVCG_DEFAULT_ALLOW) {
416 /*
417 * new exception in the child doesn't matter, only
418 * adding extra restrictions
419 */
420 return true;
421 } else {
422 /*
423 * new exception in the child will add more devices
424 * that can be accessed, so it can't match any of
425 * parent's exceptions, even slightly
426 */
427 match = match_exception_partial(&dev_cgroup->exceptions,
428 refex->type,
429 refex->major,
430 refex->minor,
431 refex->access);
432
433 if (match)
434 return false;
435 return true;
436 }
437 } else {
438 /*
439 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
440 * the new exception will add access to more devices and must
441 * be contained completely in an parent's exception to be
442 * allowed
443 */
444 match = match_exception(&dev_cgroup->exceptions, refex->type,
445 refex->major, refex->minor,
446 refex->access);
447
448 if (match)
449 /* parent has an exception that matches the proposed */
450 return true;
451 else
452 return false;
453 }
454 return false;
455}
456
457/*
458 * parent_has_perm:
459 * when adding a new allow rule to a device exception list, the rule
460 * must be allowed in the parent device
461 */
462static int parent_has_perm(struct dev_cgroup *childcg,
463 struct dev_exception_item *ex)
464{
465 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
466
467 if (!parent)
468 return 1;
469 return verify_new_ex(parent, ex, childcg->behavior);
470}
471
472/**
473 * parent_allows_removal - verify if it's ok to remove an exception
474 * @childcg: child cgroup from where the exception will be removed
475 * @ex: exception being removed
476 *
477 * When removing an exception in cgroups with default ALLOW policy, it must
478 * be checked if removing it will give the child cgroup more access than the
479 * parent.
480 *
481 * Return: true if it's ok to remove exception, false otherwise
482 */
483static bool parent_allows_removal(struct dev_cgroup *childcg,
484 struct dev_exception_item *ex)
485{
486 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
487
488 if (!parent)
489 return true;
490
491 /* It's always allowed to remove access to devices */
492 if (childcg->behavior == DEVCG_DEFAULT_DENY)
493 return true;
494
495 /*
496 * Make sure you're not removing part or a whole exception existing in
497 * the parent cgroup
498 */
499 return !match_exception_partial(&parent->exceptions, ex->type,
500 ex->major, ex->minor, ex->access);
501}
502
503/**
504 * may_allow_all - checks if it's possible to change the behavior to
505 * allow based on parent's rules.
506 * @parent: device cgroup's parent
507 * returns: != 0 in case it's allowed, 0 otherwise
508 */
509static inline int may_allow_all(struct dev_cgroup *parent)
510{
511 if (!parent)
512 return 1;
513 return parent->behavior == DEVCG_DEFAULT_ALLOW;
514}
515
516/**
517 * revalidate_active_exceptions - walks through the active exception list and
518 * revalidates the exceptions based on parent's
519 * behavior and exceptions. The exceptions that
520 * are no longer valid will be removed.
521 * Called with devcgroup_mutex held.
522 * @devcg: cgroup which exceptions will be checked
523 *
524 * This is one of the three key functions for hierarchy implementation.
525 * This function is responsible for re-evaluating all the cgroup's active
526 * exceptions due to a parent's exception change.
527 * Refer to Documentation/admin-guide/cgroup-v1/devices.rst for more details.
528 */
529static void revalidate_active_exceptions(struct dev_cgroup *devcg)
530{
531 struct dev_exception_item *ex;
532 struct list_head *this, *tmp;
533
534 list_for_each_safe(this, tmp, &devcg->exceptions) {
535 ex = container_of(this, struct dev_exception_item, list);
536 if (!parent_has_perm(devcg, ex))
537 dev_exception_rm(devcg, ex);
538 }
539}
540
541/**
542 * propagate_exception - propagates a new exception to the children
543 * @devcg_root: device cgroup that added a new exception
544 * @ex: new exception to be propagated
545 *
546 * returns: 0 in case of success, != 0 in case of error
547 */
548static int propagate_exception(struct dev_cgroup *devcg_root,
549 struct dev_exception_item *ex)
550{
551 struct cgroup_subsys_state *pos;
552 int rc = 0;
553
554 rcu_read_lock();
555
556 css_for_each_descendant_pre(pos, &devcg_root->css) {
557 struct dev_cgroup *devcg = css_to_devcgroup(pos);
558
559 /*
560 * Because devcgroup_mutex is held, no devcg will become
561 * online or offline during the tree walk (see on/offline
562 * methods), and online ones are safe to access outside RCU
563 * read lock without bumping refcnt.
564 */
565 if (pos == &devcg_root->css || !is_devcg_online(devcg))
566 continue;
567
568 rcu_read_unlock();
569
570 /*
571 * in case both root's behavior and devcg is allow, a new
572 * restriction means adding to the exception list
573 */
574 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
575 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
576 rc = dev_exception_add(devcg, ex);
577 if (rc)
578 return rc;
579 } else {
580 /*
581 * in the other possible cases:
582 * root's behavior: allow, devcg's: deny
583 * root's behavior: deny, devcg's: deny
584 * the exception will be removed
585 */
586 dev_exception_rm(devcg, ex);
587 }
588 revalidate_active_exceptions(devcg);
589
590 rcu_read_lock();
591 }
592
593 rcu_read_unlock();
594 return rc;
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(devcgroup->css.parent);
618 struct dev_cgroup tmp_devcgrp;
619
620 if (!capable(CAP_SYS_ADMIN))
621 return -EPERM;
622
623 memset(&ex, 0, sizeof(ex));
624 memset(&tmp_devcgrp, 0, sizeof(tmp_devcgrp));
625 b = buffer;
626
627 switch (*b) {
628 case 'a':
629 switch (filetype) {
630 case DEVCG_ALLOW:
631 if (css_has_online_children(&devcgroup->css))
632 return -EINVAL;
633
634 if (!may_allow_all(parent))
635 return -EPERM;
636 if (!parent) {
637 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
638 dev_exception_clean(devcgroup);
639 break;
640 }
641
642 INIT_LIST_HEAD(&tmp_devcgrp.exceptions);
643 rc = dev_exceptions_copy(&tmp_devcgrp.exceptions,
644 &devcgroup->exceptions);
645 if (rc)
646 return rc;
647 dev_exception_clean(devcgroup);
648 rc = dev_exceptions_copy(&devcgroup->exceptions,
649 &parent->exceptions);
650 if (rc) {
651 dev_exceptions_move(&devcgroup->exceptions,
652 &tmp_devcgrp.exceptions);
653 return rc;
654 }
655 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
656 dev_exception_clean(&tmp_devcgrp);
657 break;
658 case DEVCG_DENY:
659 if (css_has_online_children(&devcgroup->css))
660 return -EINVAL;
661
662 dev_exception_clean(devcgroup);
663 devcgroup->behavior = DEVCG_DEFAULT_DENY;
664 break;
665 default:
666 return -EINVAL;
667 }
668 return 0;
669 case 'b':
670 ex.type = DEVCG_DEV_BLOCK;
671 break;
672 case 'c':
673 ex.type = DEVCG_DEV_CHAR;
674 break;
675 default:
676 return -EINVAL;
677 }
678 b++;
679 if (!isspace(*b))
680 return -EINVAL;
681 b++;
682 if (*b == '*') {
683 ex.major = ~0;
684 b++;
685 } else if (isdigit(*b)) {
686 memset(temp, 0, sizeof(temp));
687 for (count = 0; count < sizeof(temp) - 1; count++) {
688 temp[count] = *b;
689 b++;
690 if (!isdigit(*b))
691 break;
692 }
693 rc = kstrtou32(temp, 10, &ex.major);
694 if (rc)
695 return -EINVAL;
696 } else {
697 return -EINVAL;
698 }
699 if (*b != ':')
700 return -EINVAL;
701 b++;
702
703 /* read minor */
704 if (*b == '*') {
705 ex.minor = ~0;
706 b++;
707 } else if (isdigit(*b)) {
708 memset(temp, 0, sizeof(temp));
709 for (count = 0; count < sizeof(temp) - 1; count++) {
710 temp[count] = *b;
711 b++;
712 if (!isdigit(*b))
713 break;
714 }
715 rc = kstrtou32(temp, 10, &ex.minor);
716 if (rc)
717 return -EINVAL;
718 } else {
719 return -EINVAL;
720 }
721 if (!isspace(*b))
722 return -EINVAL;
723 for (b++, count = 0; count < 3; count++, b++) {
724 switch (*b) {
725 case 'r':
726 ex.access |= DEVCG_ACC_READ;
727 break;
728 case 'w':
729 ex.access |= DEVCG_ACC_WRITE;
730 break;
731 case 'm':
732 ex.access |= DEVCG_ACC_MKNOD;
733 break;
734 case '\n':
735 case '\0':
736 count = 3;
737 break;
738 default:
739 return -EINVAL;
740 }
741 }
742
743 switch (filetype) {
744 case DEVCG_ALLOW:
745 /*
746 * If the default policy is to allow by default, try to remove
747 * an matching exception instead. And be silent about it: we
748 * don't want to break compatibility
749 */
750 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
751 /* Check if the parent allows removing it first */
752 if (!parent_allows_removal(devcgroup, &ex))
753 return -EPERM;
754 dev_exception_rm(devcgroup, &ex);
755 break;
756 }
757
758 if (!parent_has_perm(devcgroup, &ex))
759 return -EPERM;
760 rc = dev_exception_add(devcgroup, &ex);
761 break;
762 case DEVCG_DENY:
763 /*
764 * If the default policy is to deny by default, try to remove
765 * an matching exception instead. And be silent about it: we
766 * don't want to break compatibility
767 */
768 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
769 dev_exception_rm(devcgroup, &ex);
770 else
771 rc = dev_exception_add(devcgroup, &ex);
772
773 if (rc)
774 break;
775 /* we only propagate new restrictions */
776 rc = propagate_exception(devcgroup, &ex);
777 break;
778 default:
779 rc = -EINVAL;
780 }
781 return rc;
782}
783
784static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
785 char *buf, size_t nbytes, loff_t off)
786{
787 int retval;
788
789 mutex_lock(&devcgroup_mutex);
790 retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
791 of_cft(of)->private, strstrip(buf));
792 mutex_unlock(&devcgroup_mutex);
793 return retval ?: nbytes;
794}
795
796static struct cftype dev_cgroup_files[] = {
797 {
798 .name = "allow",
799 .write = devcgroup_access_write,
800 .private = DEVCG_ALLOW,
801 },
802 {
803 .name = "deny",
804 .write = devcgroup_access_write,
805 .private = DEVCG_DENY,
806 },
807 {
808 .name = "list",
809 .seq_show = devcgroup_seq_show,
810 .private = DEVCG_LIST,
811 },
812 { } /* terminate */
813};
814
815struct cgroup_subsys devices_cgrp_subsys = {
816 .css_alloc = devcgroup_css_alloc,
817 .css_free = devcgroup_css_free,
818 .css_online = devcgroup_online,
819 .css_offline = devcgroup_offline,
820 .legacy_cftypes = dev_cgroup_files,
821};
822
823/**
824 * devcgroup_legacy_check_permission - checks if an inode operation is permitted
825 * @type: device type
826 * @major: device major number
827 * @minor: device minor number
828 * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
829 *
830 * returns 0 on success, -EPERM case the operation is not permitted
831 */
832static int devcgroup_legacy_check_permission(short type, u32 major, u32 minor,
833 short access)
834{
835 struct dev_cgroup *dev_cgroup;
836 bool rc;
837
838 rcu_read_lock();
839 dev_cgroup = task_devcgroup(current);
840 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
841 /* Can't match any of the exceptions, even partially */
842 rc = !match_exception_partial(&dev_cgroup->exceptions,
843 type, major, minor, access);
844 else
845 /* Need to match completely one exception to be allowed */
846 rc = match_exception(&dev_cgroup->exceptions, type, major,
847 minor, access);
848 rcu_read_unlock();
849
850 if (!rc)
851 return -EPERM;
852
853 return 0;
854}
855
856#endif /* CONFIG_CGROUP_DEVICE */
857
858#if defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF)
859
860int devcgroup_check_permission(short type, u32 major, u32 minor, short access)
861{
862 int rc = BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type, major, minor, access);
863
864 if (rc)
865 return rc;
866
867 #ifdef CONFIG_CGROUP_DEVICE
868 return devcgroup_legacy_check_permission(type, major, minor, access);
869
870 #else /* CONFIG_CGROUP_DEVICE */
871 return 0;
872
873 #endif /* CONFIG_CGROUP_DEVICE */
874}
875EXPORT_SYMBOL(devcgroup_check_permission);
876#endif /* defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF) */
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}