Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
[deliverable/linux.git] / security / device_cgroup.c
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
26 static DEFINE_MUTEX(devcgroup_mutex);
27
28 enum 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
40 struct 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
48 struct dev_cgroup {
49 struct cgroup_subsys_state css;
50 struct list_head exceptions;
51 enum devcg_behavior behavior;
52 };
53
54 static 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
59 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
60 {
61 return css_to_devcgroup(task_css(task, devices_subsys_id));
62 }
63
64 struct cgroup_subsys devices_subsys;
65
66 /*
67 * called under devcgroup_mutex
68 */
69 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
70 {
71 struct dev_exception_item *ex, *tmp, *new;
72
73 lockdep_assert_held(&devcgroup_mutex);
74
75 list_for_each_entry(ex, orig, list) {
76 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
77 if (!new)
78 goto free_and_exit;
79 list_add_tail(&new->list, dest);
80 }
81
82 return 0;
83
84 free_and_exit:
85 list_for_each_entry_safe(ex, tmp, dest, list) {
86 list_del(&ex->list);
87 kfree(ex);
88 }
89 return -ENOMEM;
90 }
91
92 /*
93 * called under devcgroup_mutex
94 */
95 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
96 struct dev_exception_item *ex)
97 {
98 struct dev_exception_item *excopy, *walk;
99
100 lockdep_assert_held(&devcgroup_mutex);
101
102 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
103 if (!excopy)
104 return -ENOMEM;
105
106 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
107 if (walk->type != ex->type)
108 continue;
109 if (walk->major != ex->major)
110 continue;
111 if (walk->minor != ex->minor)
112 continue;
113
114 walk->access |= ex->access;
115 kfree(excopy);
116 excopy = NULL;
117 }
118
119 if (excopy != NULL)
120 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
121 return 0;
122 }
123
124 /*
125 * called under devcgroup_mutex
126 */
127 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
128 struct dev_exception_item *ex)
129 {
130 struct dev_exception_item *walk, *tmp;
131
132 lockdep_assert_held(&devcgroup_mutex);
133
134 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
135 if (walk->type != ex->type)
136 continue;
137 if (walk->major != ex->major)
138 continue;
139 if (walk->minor != ex->minor)
140 continue;
141
142 walk->access &= ~ex->access;
143 if (!walk->access) {
144 list_del_rcu(&walk->list);
145 kfree_rcu(walk, rcu);
146 }
147 }
148 }
149
150 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
151 {
152 struct dev_exception_item *ex, *tmp;
153
154 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
155 list_del_rcu(&ex->list);
156 kfree_rcu(ex, rcu);
157 }
158 }
159
160 /**
161 * dev_exception_clean - frees all entries of the exception list
162 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
163 *
164 * called under devcgroup_mutex
165 */
166 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
167 {
168 lockdep_assert_held(&devcgroup_mutex);
169
170 __dev_exception_clean(dev_cgroup);
171 }
172
173 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
174 {
175 return (devcg->behavior != DEVCG_DEFAULT_NONE);
176 }
177
178 /**
179 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
180 * parent's
181 * @css: css getting online
182 * returns 0 in case of success, error code otherwise
183 */
184 static int devcgroup_online(struct cgroup_subsys_state *css)
185 {
186 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
187 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css_parent(css));
188 int ret = 0;
189
190 mutex_lock(&devcgroup_mutex);
191
192 if (parent_dev_cgroup == NULL)
193 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
194 else {
195 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
196 &parent_dev_cgroup->exceptions);
197 if (!ret)
198 dev_cgroup->behavior = parent_dev_cgroup->behavior;
199 }
200 mutex_unlock(&devcgroup_mutex);
201
202 return ret;
203 }
204
205 static void devcgroup_offline(struct cgroup_subsys_state *css)
206 {
207 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
208
209 mutex_lock(&devcgroup_mutex);
210 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
211 mutex_unlock(&devcgroup_mutex);
212 }
213
214 /*
215 * called from kernel/cgroup.c with cgroup_lock() held.
216 */
217 static struct cgroup_subsys_state *
218 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
219 {
220 struct dev_cgroup *dev_cgroup;
221
222 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
223 if (!dev_cgroup)
224 return ERR_PTR(-ENOMEM);
225 INIT_LIST_HEAD(&dev_cgroup->exceptions);
226 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
227
228 return &dev_cgroup->css;
229 }
230
231 static void devcgroup_css_free(struct cgroup_subsys_state *css)
232 {
233 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
234
235 __dev_exception_clean(dev_cgroup);
236 kfree(dev_cgroup);
237 }
238
239 #define DEVCG_ALLOW 1
240 #define DEVCG_DENY 2
241 #define DEVCG_LIST 3
242
243 #define MAJMINLEN 13
244 #define ACCLEN 4
245
246 static void set_access(char *acc, short access)
247 {
248 int idx = 0;
249 memset(acc, 0, ACCLEN);
250 if (access & ACC_READ)
251 acc[idx++] = 'r';
252 if (access & ACC_WRITE)
253 acc[idx++] = 'w';
254 if (access & ACC_MKNOD)
255 acc[idx++] = 'm';
256 }
257
258 static char type_to_char(short type)
259 {
260 if (type == DEV_ALL)
261 return 'a';
262 if (type == DEV_CHAR)
263 return 'c';
264 if (type == DEV_BLOCK)
265 return 'b';
266 return 'X';
267 }
268
269 static void set_majmin(char *str, unsigned m)
270 {
271 if (m == ~0)
272 strcpy(str, "*");
273 else
274 sprintf(str, "%u", m);
275 }
276
277 static int devcgroup_seq_show(struct seq_file *m, void *v)
278 {
279 struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
280 struct dev_exception_item *ex;
281 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
282
283 rcu_read_lock();
284 /*
285 * To preserve the compatibility:
286 * - Only show the "all devices" when the default policy is to allow
287 * - List the exceptions in case the default policy is to deny
288 * This way, the file remains as a "whitelist of devices"
289 */
290 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
291 set_access(acc, ACC_MASK);
292 set_majmin(maj, ~0);
293 set_majmin(min, ~0);
294 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
295 maj, min, acc);
296 } else {
297 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
298 set_access(acc, ex->access);
299 set_majmin(maj, ex->major);
300 set_majmin(min, ex->minor);
301 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
302 maj, min, acc);
303 }
304 }
305 rcu_read_unlock();
306
307 return 0;
308 }
309
310 /**
311 * may_access - verifies if a new exception is part of what is allowed
312 * by a dev cgroup based on the default policy +
313 * exceptions. This is used to make sure a child cgroup
314 * won't have more privileges than its parent or to
315 * verify if a certain access is allowed.
316 * @dev_cgroup: dev cgroup to be tested against
317 * @refex: new exception
318 * @behavior: behavior of the exception
319 */
320 static bool may_access(struct dev_cgroup *dev_cgroup,
321 struct dev_exception_item *refex,
322 enum devcg_behavior behavior)
323 {
324 struct dev_exception_item *ex;
325 bool match = false;
326
327 rcu_lockdep_assert(rcu_read_lock_held() ||
328 lockdep_is_held(&devcgroup_mutex),
329 "device_cgroup::may_access() called without proper synchronization");
330
331 list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
332 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
333 continue;
334 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
335 continue;
336 if (ex->major != ~0 && ex->major != refex->major)
337 continue;
338 if (ex->minor != ~0 && ex->minor != refex->minor)
339 continue;
340 if (refex->access & (~ex->access))
341 continue;
342 match = true;
343 break;
344 }
345
346 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
347 if (behavior == DEVCG_DEFAULT_ALLOW) {
348 /* the exception will deny access to certain devices */
349 return true;
350 } else {
351 /* the exception will allow access to certain devices */
352 if (match)
353 /*
354 * a new exception allowing access shouldn't
355 * match an parent's exception
356 */
357 return false;
358 return true;
359 }
360 } else {
361 /* only behavior == DEVCG_DEFAULT_DENY allowed here */
362 if (match)
363 /* parent has an exception that matches the proposed */
364 return true;
365 else
366 return false;
367 }
368 return false;
369 }
370
371 /*
372 * parent_has_perm:
373 * when adding a new allow rule to a device exception list, the rule
374 * must be allowed in the parent device
375 */
376 static int parent_has_perm(struct dev_cgroup *childcg,
377 struct dev_exception_item *ex)
378 {
379 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css));
380
381 if (!parent)
382 return 1;
383 return may_access(parent, ex, childcg->behavior);
384 }
385
386 /**
387 * may_allow_all - checks if it's possible to change the behavior to
388 * allow based on parent's rules.
389 * @parent: device cgroup's parent
390 * returns: != 0 in case it's allowed, 0 otherwise
391 */
392 static inline int may_allow_all(struct dev_cgroup *parent)
393 {
394 if (!parent)
395 return 1;
396 return parent->behavior == DEVCG_DEFAULT_ALLOW;
397 }
398
399 /**
400 * revalidate_active_exceptions - walks through the active exception list and
401 * revalidates the exceptions based on parent's
402 * behavior and exceptions. The exceptions that
403 * are no longer valid will be removed.
404 * Called with devcgroup_mutex held.
405 * @devcg: cgroup which exceptions will be checked
406 *
407 * This is one of the three key functions for hierarchy implementation.
408 * This function is responsible for re-evaluating all the cgroup's active
409 * exceptions due to a parent's exception change.
410 * Refer to Documentation/cgroups/devices.txt for more details.
411 */
412 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
413 {
414 struct dev_exception_item *ex;
415 struct list_head *this, *tmp;
416
417 list_for_each_safe(this, tmp, &devcg->exceptions) {
418 ex = container_of(this, struct dev_exception_item, list);
419 if (!parent_has_perm(devcg, ex))
420 dev_exception_rm(devcg, ex);
421 }
422 }
423
424 /**
425 * propagate_exception - propagates a new exception to the children
426 * @devcg_root: device cgroup that added a new exception
427 * @ex: new exception to be propagated
428 *
429 * returns: 0 in case of success, != 0 in case of error
430 */
431 static int propagate_exception(struct dev_cgroup *devcg_root,
432 struct dev_exception_item *ex)
433 {
434 struct cgroup_subsys_state *pos;
435 int rc = 0;
436
437 rcu_read_lock();
438
439 css_for_each_descendant_pre(pos, &devcg_root->css) {
440 struct dev_cgroup *devcg = css_to_devcgroup(pos);
441
442 /*
443 * Because devcgroup_mutex is held, no devcg will become
444 * online or offline during the tree walk (see on/offline
445 * methods), and online ones are safe to access outside RCU
446 * read lock without bumping refcnt.
447 */
448 if (pos == &devcg_root->css || !is_devcg_online(devcg))
449 continue;
450
451 rcu_read_unlock();
452
453 /*
454 * in case both root's behavior and devcg is allow, a new
455 * restriction means adding to the exception list
456 */
457 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
458 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
459 rc = dev_exception_add(devcg, ex);
460 if (rc)
461 break;
462 } else {
463 /*
464 * in the other possible cases:
465 * root's behavior: allow, devcg's: deny
466 * root's behavior: deny, devcg's: deny
467 * the exception will be removed
468 */
469 dev_exception_rm(devcg, ex);
470 }
471 revalidate_active_exceptions(devcg);
472
473 rcu_read_lock();
474 }
475
476 rcu_read_unlock();
477 return rc;
478 }
479
480 static inline bool has_children(struct dev_cgroup *devcgroup)
481 {
482 struct cgroup *cgrp = devcgroup->css.cgroup;
483
484 return !list_empty(&cgrp->children);
485 }
486
487 /*
488 * Modify the exception list using allow/deny rules.
489 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
490 * so we can give a container CAP_MKNOD to let it create devices but not
491 * modify the exception list.
492 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
493 * us to also grant CAP_SYS_ADMIN to containers without giving away the
494 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
495 *
496 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
497 * new access is only allowed if you're in the top-level cgroup, or your
498 * parent cgroup has the access you're asking for.
499 */
500 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
501 int filetype, const char *buffer)
502 {
503 const char *b;
504 char temp[12]; /* 11 + 1 characters needed for a u32 */
505 int count, rc = 0;
506 struct dev_exception_item ex;
507 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&devcgroup->css));
508
509 if (!capable(CAP_SYS_ADMIN))
510 return -EPERM;
511
512 memset(&ex, 0, sizeof(ex));
513 b = buffer;
514
515 switch (*b) {
516 case 'a':
517 switch (filetype) {
518 case DEVCG_ALLOW:
519 if (has_children(devcgroup))
520 return -EINVAL;
521
522 if (!may_allow_all(parent))
523 return -EPERM;
524 dev_exception_clean(devcgroup);
525 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
526 if (!parent)
527 break;
528
529 rc = dev_exceptions_copy(&devcgroup->exceptions,
530 &parent->exceptions);
531 if (rc)
532 return rc;
533 break;
534 case DEVCG_DENY:
535 if (has_children(devcgroup))
536 return -EINVAL;
537
538 dev_exception_clean(devcgroup);
539 devcgroup->behavior = DEVCG_DEFAULT_DENY;
540 break;
541 default:
542 return -EINVAL;
543 }
544 return 0;
545 case 'b':
546 ex.type = DEV_BLOCK;
547 break;
548 case 'c':
549 ex.type = DEV_CHAR;
550 break;
551 default:
552 return -EINVAL;
553 }
554 b++;
555 if (!isspace(*b))
556 return -EINVAL;
557 b++;
558 if (*b == '*') {
559 ex.major = ~0;
560 b++;
561 } else if (isdigit(*b)) {
562 memset(temp, 0, sizeof(temp));
563 for (count = 0; count < sizeof(temp) - 1; count++) {
564 temp[count] = *b;
565 b++;
566 if (!isdigit(*b))
567 break;
568 }
569 rc = kstrtou32(temp, 10, &ex.major);
570 if (rc)
571 return -EINVAL;
572 } else {
573 return -EINVAL;
574 }
575 if (*b != ':')
576 return -EINVAL;
577 b++;
578
579 /* read minor */
580 if (*b == '*') {
581 ex.minor = ~0;
582 b++;
583 } else if (isdigit(*b)) {
584 memset(temp, 0, sizeof(temp));
585 for (count = 0; count < sizeof(temp) - 1; count++) {
586 temp[count] = *b;
587 b++;
588 if (!isdigit(*b))
589 break;
590 }
591 rc = kstrtou32(temp, 10, &ex.minor);
592 if (rc)
593 return -EINVAL;
594 } else {
595 return -EINVAL;
596 }
597 if (!isspace(*b))
598 return -EINVAL;
599 for (b++, count = 0; count < 3; count++, b++) {
600 switch (*b) {
601 case 'r':
602 ex.access |= ACC_READ;
603 break;
604 case 'w':
605 ex.access |= ACC_WRITE;
606 break;
607 case 'm':
608 ex.access |= ACC_MKNOD;
609 break;
610 case '\n':
611 case '\0':
612 count = 3;
613 break;
614 default:
615 return -EINVAL;
616 }
617 }
618
619 switch (filetype) {
620 case DEVCG_ALLOW:
621 if (!parent_has_perm(devcgroup, &ex))
622 return -EPERM;
623 /*
624 * If the default policy is to allow by default, try to remove
625 * an matching exception instead. And be silent about it: we
626 * don't want to break compatibility
627 */
628 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
629 dev_exception_rm(devcgroup, &ex);
630 return 0;
631 }
632 rc = dev_exception_add(devcgroup, &ex);
633 break;
634 case DEVCG_DENY:
635 /*
636 * If the default policy is to deny by default, try to remove
637 * an matching exception instead. And be silent about it: we
638 * don't want to break compatibility
639 */
640 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
641 dev_exception_rm(devcgroup, &ex);
642 else
643 rc = dev_exception_add(devcgroup, &ex);
644
645 if (rc)
646 break;
647 /* we only propagate new restrictions */
648 rc = propagate_exception(devcgroup, &ex);
649 break;
650 default:
651 rc = -EINVAL;
652 }
653 return rc;
654 }
655
656 static int devcgroup_access_write(struct cgroup_subsys_state *css,
657 struct cftype *cft, const char *buffer)
658 {
659 int retval;
660
661 mutex_lock(&devcgroup_mutex);
662 retval = devcgroup_update_access(css_to_devcgroup(css),
663 cft->private, buffer);
664 mutex_unlock(&devcgroup_mutex);
665 return retval;
666 }
667
668 static struct cftype dev_cgroup_files[] = {
669 {
670 .name = "allow",
671 .write_string = devcgroup_access_write,
672 .private = DEVCG_ALLOW,
673 },
674 {
675 .name = "deny",
676 .write_string = devcgroup_access_write,
677 .private = DEVCG_DENY,
678 },
679 {
680 .name = "list",
681 .seq_show = devcgroup_seq_show,
682 .private = DEVCG_LIST,
683 },
684 { } /* terminate */
685 };
686
687 struct cgroup_subsys devices_subsys = {
688 .name = "devices",
689 .css_alloc = devcgroup_css_alloc,
690 .css_free = devcgroup_css_free,
691 .css_online = devcgroup_online,
692 .css_offline = devcgroup_offline,
693 .subsys_id = devices_subsys_id,
694 .base_cftypes = dev_cgroup_files,
695 };
696
697 /**
698 * __devcgroup_check_permission - checks if an inode operation is permitted
699 * @dev_cgroup: the dev cgroup to be tested against
700 * @type: device type
701 * @major: device major number
702 * @minor: device minor number
703 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
704 *
705 * returns 0 on success, -EPERM case the operation is not permitted
706 */
707 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
708 short access)
709 {
710 struct dev_cgroup *dev_cgroup;
711 struct dev_exception_item ex;
712 int rc;
713
714 memset(&ex, 0, sizeof(ex));
715 ex.type = type;
716 ex.major = major;
717 ex.minor = minor;
718 ex.access = access;
719
720 rcu_read_lock();
721 dev_cgroup = task_devcgroup(current);
722 rc = may_access(dev_cgroup, &ex, dev_cgroup->behavior);
723 rcu_read_unlock();
724
725 if (!rc)
726 return -EPERM;
727
728 return 0;
729 }
730
731 int __devcgroup_inode_permission(struct inode *inode, int mask)
732 {
733 short type, access = 0;
734
735 if (S_ISBLK(inode->i_mode))
736 type = DEV_BLOCK;
737 if (S_ISCHR(inode->i_mode))
738 type = DEV_CHAR;
739 if (mask & MAY_WRITE)
740 access |= ACC_WRITE;
741 if (mask & MAY_READ)
742 access |= ACC_READ;
743
744 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
745 access);
746 }
747
748 int devcgroup_inode_mknod(int mode, dev_t dev)
749 {
750 short type;
751
752 if (!S_ISBLK(mode) && !S_ISCHR(mode))
753 return 0;
754
755 if (S_ISBLK(mode))
756 type = DEV_BLOCK;
757 else
758 type = DEV_CHAR;
759
760 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),
761 ACC_MKNOD);
762
763 }
This page took 0.044784 seconds and 6 git commands to generate.