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/*
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);
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_WARN(!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(childcg->css.parent);
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(childcg->css.parent);
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
590/*
591 * Modify the exception list using allow/deny rules.
592 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
593 * so we can give a container CAP_MKNOD to let it create devices but not
594 * modify the exception list.
595 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
596 * us to also grant CAP_SYS_ADMIN to containers without giving away the
597 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
598 *
599 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
600 * new access is only allowed if you're in the top-level cgroup, or your
601 * parent cgroup has the access you're asking for.
602 */
603static int devcgroup_update_access(struct dev_cgroup *devcgroup,
604 int filetype, char *buffer)
605{
606 const char *b;
607 char temp[12]; /* 11 + 1 characters needed for a u32 */
608 int count, rc = 0;
609 struct dev_exception_item ex;
610 struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
611
612 if (!capable(CAP_SYS_ADMIN))
613 return -EPERM;
614
615 memset(&ex, 0, sizeof(ex));
616 b = buffer;
617
618 switch (*b) {
619 case 'a':
620 switch (filetype) {
621 case DEVCG_ALLOW:
622 if (css_has_online_children(&devcgroup->css))
623 return -EINVAL;
624
625 if (!may_allow_all(parent))
626 return -EPERM;
627 dev_exception_clean(devcgroup);
628 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
629 if (!parent)
630 break;
631
632 rc = dev_exceptions_copy(&devcgroup->exceptions,
633 &parent->exceptions);
634 if (rc)
635 return rc;
636 break;
637 case DEVCG_DENY:
638 if (css_has_online_children(&devcgroup->css))
639 return -EINVAL;
640
641 dev_exception_clean(devcgroup);
642 devcgroup->behavior = DEVCG_DEFAULT_DENY;
643 break;
644 default:
645 return -EINVAL;
646 }
647 return 0;
648 case 'b':
649 ex.type = DEV_BLOCK;
650 break;
651 case 'c':
652 ex.type = DEV_CHAR;
653 break;
654 default:
655 return -EINVAL;
656 }
657 b++;
658 if (!isspace(*b))
659 return -EINVAL;
660 b++;
661 if (*b == '*') {
662 ex.major = ~0;
663 b++;
664 } else if (isdigit(*b)) {
665 memset(temp, 0, sizeof(temp));
666 for (count = 0; count < sizeof(temp) - 1; count++) {
667 temp[count] = *b;
668 b++;
669 if (!isdigit(*b))
670 break;
671 }
672 rc = kstrtou32(temp, 10, &ex.major);
673 if (rc)
674 return -EINVAL;
675 } else {
676 return -EINVAL;
677 }
678 if (*b != ':')
679 return -EINVAL;
680 b++;
681
682 /* read minor */
683 if (*b == '*') {
684 ex.minor = ~0;
685 b++;
686 } else if (isdigit(*b)) {
687 memset(temp, 0, sizeof(temp));
688 for (count = 0; count < sizeof(temp) - 1; count++) {
689 temp[count] = *b;
690 b++;
691 if (!isdigit(*b))
692 break;
693 }
694 rc = kstrtou32(temp, 10, &ex.minor);
695 if (rc)
696 return -EINVAL;
697 } else {
698 return -EINVAL;
699 }
700 if (!isspace(*b))
701 return -EINVAL;
702 for (b++, count = 0; count < 3; count++, b++) {
703 switch (*b) {
704 case 'r':
705 ex.access |= ACC_READ;
706 break;
707 case 'w':
708 ex.access |= ACC_WRITE;
709 break;
710 case 'm':
711 ex.access |= ACC_MKNOD;
712 break;
713 case '\n':
714 case '\0':
715 count = 3;
716 break;
717 default:
718 return -EINVAL;
719 }
720 }
721
722 switch (filetype) {
723 case DEVCG_ALLOW:
724 /*
725 * If the default policy is to allow by default, try to remove
726 * an matching exception instead. And be silent about it: we
727 * don't want to break compatibility
728 */
729 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
730 /* Check if the parent allows removing it first */
731 if (!parent_allows_removal(devcgroup, &ex))
732 return -EPERM;
733 dev_exception_rm(devcgroup, &ex);
734 break;
735 }
736
737 if (!parent_has_perm(devcgroup, &ex))
738 return -EPERM;
739 rc = dev_exception_add(devcgroup, &ex);
740 break;
741 case DEVCG_DENY:
742 /*
743 * If the default policy is to deny by default, try to remove
744 * an matching exception instead. And be silent about it: we
745 * don't want to break compatibility
746 */
747 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
748 dev_exception_rm(devcgroup, &ex);
749 else
750 rc = dev_exception_add(devcgroup, &ex);
751
752 if (rc)
753 break;
754 /* we only propagate new restrictions */
755 rc = propagate_exception(devcgroup, &ex);
756 break;
757 default:
758 rc = -EINVAL;
759 }
760 return rc;
761}
762
763static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
764 char *buf, size_t nbytes, loff_t off)
765{
766 int retval;
767
768 mutex_lock(&devcgroup_mutex);
769 retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
770 of_cft(of)->private, strstrip(buf));
771 mutex_unlock(&devcgroup_mutex);
772 return retval ?: nbytes;
773}
774
775static struct cftype dev_cgroup_files[] = {
776 {
777 .name = "allow",
778 .write = devcgroup_access_write,
779 .private = DEVCG_ALLOW,
780 },
781 {
782 .name = "deny",
783 .write = devcgroup_access_write,
784 .private = DEVCG_DENY,
785 },
786 {
787 .name = "list",
788 .seq_show = devcgroup_seq_show,
789 .private = DEVCG_LIST,
790 },
791 { } /* terminate */
792};
793
794struct cgroup_subsys devices_cgrp_subsys = {
795 .css_alloc = devcgroup_css_alloc,
796 .css_free = devcgroup_css_free,
797 .css_online = devcgroup_online,
798 .css_offline = devcgroup_offline,
799 .legacy_cftypes = dev_cgroup_files,
800};
801
802/**
803 * __devcgroup_check_permission - checks if an inode operation is permitted
804 * @dev_cgroup: the dev cgroup to be tested against
805 * @type: device type
806 * @major: device major number
807 * @minor: device minor number
808 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
809 *
810 * returns 0 on success, -EPERM case the operation is not permitted
811 */
812static int __devcgroup_check_permission(short type, u32 major, u32 minor,
813 short access)
814{
815 struct dev_cgroup *dev_cgroup;
816 bool rc;
817
818 rcu_read_lock();
819 dev_cgroup = task_devcgroup(current);
820 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
821 /* Can't match any of the exceptions, even partially */
822 rc = !match_exception_partial(&dev_cgroup->exceptions,
823 type, major, minor, access);
824 else
825 /* Need to match completely one exception to be allowed */
826 rc = match_exception(&dev_cgroup->exceptions, type, major,
827 minor, access);
828 rcu_read_unlock();
829
830 if (!rc)
831 return -EPERM;
832
833 return 0;
834}
835
836int __devcgroup_inode_permission(struct inode *inode, int mask)
837{
838 short type, access = 0;
839
840 if (S_ISBLK(inode->i_mode))
841 type = DEV_BLOCK;
842 if (S_ISCHR(inode->i_mode))
843 type = DEV_CHAR;
844 if (mask & MAY_WRITE)
845 access |= ACC_WRITE;
846 if (mask & MAY_READ)
847 access |= ACC_READ;
848
849 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
850 access);
851}
852
853int devcgroup_inode_mknod(int mode, dev_t dev)
854{
855 short type;
856
857 if (!S_ISBLK(mode) && !S_ISCHR(mode))
858 return 0;
859
860 if (S_ISBLK(mode))
861 type = DEV_BLOCK;
862 else
863 type = DEV_CHAR;
864
865 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),
866 ACC_MKNOD);
867
868}