1 // SPDX-License-Identifier: GPL-2.0 1
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
21 static DEFINE_MUTEX(devcgroup_mutex);
22
23 enum 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
35 struct 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
43 struct dev_cgroup {
44 struct cgroup_subsys_state css;
45 struct list_head exceptions;
46 enum devcg_behavior behavior;
47 };
48
49 static inline struct dev_cgroup *css_to_devcgr
50 {
51 return s ? container_of(s, struct dev_
52 }
53
54 static inline struct dev_cgroup *task_devcgrou
55 {
56 return css_to_devcgroup(task_css(task,
57 }
58
59 /*
60 * called under devcgroup_mutex
61 */
62 static int dev_exceptions_copy(struct list_hea
63 {
64 struct dev_exception_item *ex, *tmp, *
65
66 lockdep_assert_held(&devcgroup_mutex);
67
68 list_for_each_entry(ex, orig, list) {
69 new = kmemdup(ex, sizeof(*ex),
70 if (!new)
71 goto free_and_exit;
72 list_add_tail(&new->list, dest
73 }
74
75 return 0;
76
77 free_and_exit:
78 list_for_each_entry_safe(ex, tmp, dest
79 list_del(&ex->list);
80 kfree(ex);
81 }
82 return -ENOMEM;
83 }
84
85 static void dev_exceptions_move(struct list_he
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
92 list_move_tail(&ex->list, dest
93 }
94 }
95
96 /*
97 * called under devcgroup_mutex
98 */
99 static int dev_exception_add(struct dev_cgroup
100 struct dev_except
101 {
102 struct dev_exception_item *excopy, *wa
103
104 lockdep_assert_held(&devcgroup_mutex);
105
106 excopy = kmemdup(ex, sizeof(*ex), GFP_
107 if (!excopy)
108 return -ENOMEM;
109
110 list_for_each_entry(walk, &dev_cgroup-
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->lis
125 return 0;
126 }
127
128 /*
129 * called under devcgroup_mutex
130 */
131 static void dev_exception_rm(struct dev_cgroup
132 struct dev_except
133 {
134 struct dev_exception_item *walk, *tmp;
135
136 lockdep_assert_held(&devcgroup_mutex);
137
138 list_for_each_entry_safe(walk, tmp, &d
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->li
149 kfree_rcu(walk, rcu);
150 }
151 }
152 }
153
154 static void __dev_exception_clean(struct dev_c
155 {
156 struct dev_exception_item *ex, *tmp;
157
158 list_for_each_entry_safe(ex, tmp, &dev
159 list_del_rcu(&ex->list);
160 kfree_rcu(ex, rcu);
161 }
162 }
163
164 /**
165 * dev_exception_clean - frees all entries of
166 * @dev_cgroup: dev_cgroup with the exception
167 *
168 * called under devcgroup_mutex
169 */
170 static void dev_exception_clean(struct dev_cgr
171 {
172 lockdep_assert_held(&devcgroup_mutex);
173
174 __dev_exception_clean(dev_cgroup);
175 }
176
177 static inline bool is_devcg_online(const struc
178 {
179 return (devcg->behavior != DEVCG_DEFAU
180 }
181
182 /**
183 * devcgroup_online - initializes devcgroup's
184 * parent's
185 * @css: css getting online
186 * returns 0 in case of success, error code ot
187 */
188 static int devcgroup_online(struct cgroup_subs
189 {
190 struct dev_cgroup *dev_cgroup = css_to
191 struct dev_cgroup *parent_dev_cgroup =
192 int ret = 0;
193
194 mutex_lock(&devcgroup_mutex);
195
196 if (parent_dev_cgroup == NULL)
197 dev_cgroup->behavior = DEVCG_D
198 else {
199 ret = dev_exceptions_copy(&dev
200 &par
201 if (!ret)
202 dev_cgroup->behavior =
203 }
204 mutex_unlock(&devcgroup_mutex);
205
206 return ret;
207 }
208
209 static void devcgroup_offline(struct cgroup_su
210 {
211 struct dev_cgroup *dev_cgroup = css_to
212
213 mutex_lock(&devcgroup_mutex);
214 dev_cgroup->behavior = DEVCG_DEFAULT_N
215 mutex_unlock(&devcgroup_mutex);
216 }
217
218 /*
219 * called from kernel/cgroup/cgroup.c with cgr
220 */
221 static struct cgroup_subsys_state *
222 devcgroup_css_alloc(struct cgroup_subsys_state
223 {
224 struct dev_cgroup *dev_cgroup;
225
226 dev_cgroup = kzalloc(sizeof(*dev_cgrou
227 if (!dev_cgroup)
228 return ERR_PTR(-ENOMEM);
229 INIT_LIST_HEAD(&dev_cgroup->exceptions
230 dev_cgroup->behavior = DEVCG_DEFAULT_N
231
232 return &dev_cgroup->css;
233 }
234
235 static void devcgroup_css_free(struct cgroup_s
236 {
237 struct dev_cgroup *dev_cgroup = css_to
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
250 static 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
262 static 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
273 static void set_majmin(char *str, unsigned m)
274 {
275 if (m == ~0)
276 strcpy(str, "*");
277 else
278 sprintf(str, "%u", m);
279 }
280
281 static int devcgroup_seq_show(struct seq_file
282 {
283 struct dev_cgroup *devcgroup = css_to_
284 struct dev_exception_item *ex;
285 char maj[MAJMINLEN], min[MAJMINLEN], a
286
287 rcu_read_lock();
288 /*
289 * To preserve the compatibility:
290 * - Only show the "all devices" when
291 * - List the exceptions in case the d
292 * This way, the file remains as a "wh
293 */
294 if (devcgroup->behavior == DEVCG_DEFAU
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",
299 maj, min, acc);
300 } else {
301 list_for_each_entry_rcu(ex, &d
302 set_access(acc, ex->ac
303 set_majmin(maj, ex->ma
304 set_majmin(min, ex->mi
305 seq_printf(m, "%c %s:%
306 maj, min, a
307 }
308 }
309 rcu_read_unlock();
310
311 return 0;
312 }
313
314 /**
315 * match_exception - iterates the excepti
316 * @exceptions: list of exceptions
317 * @type: device type (DEVCG_DEV_BLOCK or DEVC
318 * @major: device file major number, ~0 to mat
319 * @minor: device file minor number, ~0 to mat
320 * @access: permission mask (DEVCG_ACC_READ, D
321 *
322 * It is considered a complete match if an exc
323 * contain the entire range of provided parame
324 *
325 * Return: true in case it matches an exceptio
326 */
327 static bool match_exception(struct list_head *
328 u32 major, u32 min
329 {
330 struct dev_exception_item *ex;
331
332 list_for_each_entry_rcu(ex, exceptions
333 if ((type & DEVCG_DEV_BLOCK) &
334 continue;
335 if ((type & DEVCG_DEV_CHAR) &&
336 continue;
337 if (ex->major != ~0 && ex->maj
338 continue;
339 if (ex->minor != ~0 && ex->min
340 continue;
341 /* provided access cannot have
342 if (access & (~ex->access))
343 continue;
344 return true;
345 }
346 return false;
347 }
348
349 /**
350 * match_exception_partial - iterates the exce
351 * @exceptions: list of exceptions
352 * @type: device type (DEVCG_DEV_BLOCK or DEVC
353 * @major: device file major number, ~0 to mat
354 * @minor: device file minor number, ~0 to mat
355 * @access: permission mask (DEVCG_ACC_READ, D
356 *
357 * It is considered a partial match if an exce
358 * contain *any* of the devices specified by p
359 * used to make sure no extra access is being
360 * any of the exception list.
361 *
362 * Return: true in case the provided range mat
363 */
364 static bool match_exception_partial(struct lis
365 u32 major,
366 {
367 struct dev_exception_item *ex;
368
369 list_for_each_entry_rcu(ex, exceptions
370 lockdep_is_hel
371 if ((type & DEVCG_DEV_BLOCK) &
372 continue;
373 if ((type & DEVCG_DEV_CHAR) &&
374 continue;
375 /*
376 * We must be sure that both t
377 * range aren't masking all de
378 */
379 if (ex->major != ~0 && major !
380 continue;
381 if (ex->minor != ~0 && minor !
382 continue;
383 /*
384 * In order to make sure the p
385 * an exception, all its acces
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
397 * @dev_cgroup: dev cgroup to be tested agains
398 * @refex: new exception
399 * @behavior: behavior of the exception's dev_
400 *
401 * This is used to make sure a child cgroup wo
402 * than its parent
403 */
404 static bool verify_new_ex(struct dev_cgroup *d
405 struct dev_exception
406 enum devcg_behavior
407 {
408 bool match = false;
409
410 RCU_LOCKDEP_WARN(!rcu_read_lock_held()
411 !lockdep_is_held(&dev
412 "device_cgroup:verify
413
414 if (dev_cgroup->behavior == DEVCG_DEFA
415 if (behavior == DEVCG_DEFAULT_
416 /*
417 * new exception in th
418 * adding extra restri
419 */
420 return true;
421 } else {
422 /*
423 * new exception in th
424 * that can be accesse
425 * parent's exceptions
426 */
427 match = match_exceptio
428
429
430
431
432
433 if (match)
434 return false;
435 return true;
436 }
437 } else {
438 /*
439 * Only behavior == DEVCG_DEFA
440 * the new exception will add
441 * be contained completely in
442 * allowed
443 */
444 match = match_exception(&dev_c
445 refex-
446 refex-
447
448 if (match)
449 /* parent has an excep
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 ex
460 * must be allowed in the parent device
461 */
462 static int parent_has_perm(struct dev_cgroup *
463 struct dev_e
464 {
465 struct dev_cgroup *parent = css_to_dev
466
467 if (!parent)
468 return 1;
469 return verify_new_ex(parent, ex, child
470 }
471
472 /**
473 * parent_allows_removal - verify if it's ok t
474 * @childcg: child cgroup from where the excep
475 * @ex: exception being removed
476 *
477 * When removing an exception in cgroups with
478 * be checked if removing it will give the chi
479 * parent.
480 *
481 * Return: true if it's ok to remove exception
482 */
483 static bool parent_allows_removal(struct dev_c
484 struct dev_e
485 {
486 struct dev_cgroup *parent = css_to_dev
487
488 if (!parent)
489 return true;
490
491 /* It's always allowed to remove acces
492 if (childcg->behavior == DEVCG_DEFAULT
493 return true;
494
495 /*
496 * Make sure you're not removing part
497 * the parent cgroup
498 */
499 return !match_exception_partial(&paren
500 ex->ma
501 }
502
503 /**
504 * may_allow_all - checks if it's possible to
505 * allow based on parent's rul
506 * @parent: device cgroup's parent
507 * returns: != 0 in case it's allowed, 0 other
508 */
509 static inline int may_allow_all(struct dev_cgr
510 {
511 if (!parent)
512 return 1;
513 return parent->behavior == DEVCG_DEFAU
514 }
515
516 /**
517 * revalidate_active_exceptions - walks throug
518 * revalidates
519 * behavior and
520 * are no longe
521 * Called with
522 * @devcg: cgroup which exceptions will be che
523 *
524 * This is one of the three key functions for
525 * This function is responsible for re-evaluat
526 * exceptions due to a parent's exception chan
527 * Refer to Documentation/admin-guide/cgroup-v
528 */
529 static void revalidate_active_exceptions(struc
530 {
531 struct dev_exception_item *ex;
532 struct list_head *this, *tmp;
533
534 list_for_each_safe(this, tmp, &devcg->
535 ex = container_of(this, struct
536 if (!parent_has_perm(devcg, ex
537 dev_exception_rm(devcg
538 }
539 }
540
541 /**
542 * propagate_exception - propagates a new exce
543 * @devcg_root: device cgroup that added a new
544 * @ex: new exception to be propagated
545 *
546 * returns: 0 in case of success, != 0 in case
547 */
548 static int propagate_exception(struct dev_cgro
549 struct dev_exce
550 {
551 struct cgroup_subsys_state *pos;
552 int rc = 0;
553
554 rcu_read_lock();
555
556 css_for_each_descendant_pre(pos, &devc
557 struct dev_cgroup *devcg = css
558
559 /*
560 * Because devcgroup_mutex is
561 * online or offline during th
562 * methods), and online ones a
563 * read lock without bumping r
564 */
565 if (pos == &devcg_root->css ||
566 continue;
567
568 rcu_read_unlock();
569
570 /*
571 * in case both root's behavio
572 * restriction means adding to
573 */
574 if (devcg_root->behavior == DE
575 devcg->behavior == DEVCG_D
576 rc = dev_exception_add
577 if (rc)
578 return rc;
579 } else {
580 /*
581 * in the other possib
582 * root's behavior: al
583 * root's behavior: de
584 * the exception will
585 */
586 dev_exception_rm(devcg
587 }
588 revalidate_active_exceptions(d
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
599 * CAP_SYS_ADMIN is needed for this. It's at
600 * so we can give a container CAP_MKNOD to let
601 * modify the exception list.
602 * It seems likely we'll want to add a CAP_CON
603 * us to also grant CAP_SYS_ADMIN to container
604 * device exception list controls, but for now
605 *
606 * Taking rules away is always allowed (given
607 * new access is only allowed if you're in the
608 * parent cgroup has the access you're asking
609 */
610 static int devcgroup_update_access(struct dev_
611 int filetyp
612 {
613 const char *b;
614 char temp[12]; /* 11 + 1 char
615 int count, rc = 0;
616 struct dev_exception_item ex;
617 struct dev_cgroup *parent = css_to_dev
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_dev
625 b = buffer;
626
627 switch (*b) {
628 case 'a':
629 switch (filetype) {
630 case DEVCG_ALLOW:
631 if (css_has_online_chi
632 return -EINVAL
633
634 if (!may_allow_all(par
635 return -EPERM;
636 if (!parent) {
637 devcgroup->beh
638 dev_exception_
639 break;
640 }
641
642 INIT_LIST_HEAD(&tmp_de
643 rc = dev_exceptions_co
644
645 if (rc)
646 return rc;
647 dev_exception_clean(de
648 rc = dev_exceptions_co
649
650 if (rc) {
651 dev_exceptions
652
653 return rc;
654 }
655 devcgroup->behavior =
656 dev_exception_clean(&t
657 break;
658 case DEVCG_DENY:
659 if (css_has_online_chi
660 return -EINVAL
661
662 dev_exception_clean(de
663 devcgroup->behavior =
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
688 temp[count] = *b;
689 b++;
690 if (!isdigit(*b))
691 break;
692 }
693 rc = kstrtou32(temp, 10, &ex.m
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
710 temp[count] = *b;
711 b++;
712 if (!isdigit(*b))
713 break;
714 }
715 rc = kstrtou32(temp, 10, &ex.m
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+
724 switch (*b) {
725 case 'r':
726 ex.access |= DEVCG_ACC
727 break;
728 case 'w':
729 ex.access |= DEVCG_ACC
730 break;
731 case 'm':
732 ex.access |= DEVCG_ACC
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
747 * an matching exception inste
748 * don't want to break compati
749 */
750 if (devcgroup->behavior == DEV
751 /* Check if the parent
752 if (!parent_allows_rem
753 return -EPERM;
754 dev_exception_rm(devcg
755 break;
756 }
757
758 if (!parent_has_perm(devcgroup
759 return -EPERM;
760 rc = dev_exception_add(devcgro
761 break;
762 case DEVCG_DENY:
763 /*
764 * If the default policy is to
765 * an matching exception inste
766 * don't want to break compati
767 */
768 if (devcgroup->behavior == DEV
769 dev_exception_rm(devcg
770 else
771 rc = dev_exception_add
772
773 if (rc)
774 break;
775 /* we only propagate new restr
776 rc = propagate_exception(devcg
777 break;
778 default:
779 rc = -EINVAL;
780 }
781 return rc;
782 }
783
784 static ssize_t devcgroup_access_write(struct k
785 char *bu
786 {
787 int retval;
788
789 mutex_lock(&devcgroup_mutex);
790 retval = devcgroup_update_access(css_t
791 of_cf
792 mutex_unlock(&devcgroup_mutex);
793 return retval ?: nbytes;
794 }
795
796 static struct cftype dev_cgroup_files[] = {
797 {
798 .name = "allow",
799 .write = devcgroup_access_writ
800 .private = DEVCG_ALLOW,
801 },
802 {
803 .name = "deny",
804 .write = devcgroup_access_writ
805 .private = DEVCG_DENY,
806 },
807 {
808 .name = "list",
809 .seq_show = devcgroup_seq_show
810 .private = DEVCG_LIST,
811 },
812 { } /* terminate */
813 };
814
815 struct 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
825 * @type: device type
826 * @major: device major number
827 * @minor: device minor number
828 * @access: combination of DEVCG_ACC_WRITE, DE
829 *
830 * returns 0 on success, -EPERM case the opera
831 */
832 static int devcgroup_legacy_check_permission(s
833 short
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_DEFA
841 /* Can't match any of the exce
842 rc = !match_exception_partial(
843
844 else
845 /* Need to match completely on
846 rc = match_exception(&dev_cgro
847 minor, ac
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(C
859
860 int devcgroup_check_permission(short type, u32
861 {
862 int rc = BPF_CGROUP_RUN_PROG_DEVICE_CG
863
864 if (rc)
865 return rc;
866
867 #ifdef CONFIG_CGROUP_DEVICE
868 return devcgroup_legacy_check_permissi
869
870 #else /* CONFIG_CGROUP_DEVICE */
871 return 0;
872
873 #endif /* CONFIG_CGROUP_DEVICE */
874 }
875 EXPORT_SYMBOL(devcgroup_check_permission);
876 #endif /* defined(CONFIG_CGROUP_DEVICE) || def
877
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.