Merge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[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_cgrp_id));
62 }
63
64 /*
65 * called under devcgroup_mutex
66 */
67 static 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
82 free_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 */
93 static 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 */
125 static 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
148 static 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 */
164 static 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
171 static 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 */
182 static int devcgroup_online(struct cgroup_subsys_state *css)
183 {
184 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
185 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css_parent(css));
186 int ret = 0;
187
188 mutex_lock(&devcgroup_mutex);
189
190 if (parent_dev_cgroup == NULL)
191 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
192 else {
193 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
194 &parent_dev_cgroup->exceptions);
195 if (!ret)
196 dev_cgroup->behavior = parent_dev_cgroup->behavior;
197 }
198 mutex_unlock(&devcgroup_mutex);
199
200 return ret;
201 }
202
203 static 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 */
215 static struct cgroup_subsys_state *
216 devcgroup_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
229 static 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
244 static 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
256 static 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
267 static void set_majmin(char *str, unsigned m)
268 {
269 if (m == ~0)
270 strcpy(str, "*");
271 else
272 sprintf(str, "%u", m);
273 }
274
275 static 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 * may_access - verifies if a new exception is part of what is allowed
310 * by a dev cgroup based on the default policy +
311 * exceptions. This is used to make sure a child cgroup
312 * won't have more privileges than its parent or to
313 * verify if a certain access is allowed.
314 * @dev_cgroup: dev cgroup to be tested against
315 * @refex: new exception
316 * @behavior: behavior of the exception
317 */
318 static bool may_access(struct dev_cgroup *dev_cgroup,
319 struct dev_exception_item *refex,
320 enum devcg_behavior behavior)
321 {
322 struct dev_exception_item *ex;
323 bool match = false;
324
325 rcu_lockdep_assert(rcu_read_lock_held() ||
326 lockdep_is_held(&devcgroup_mutex),
327 "device_cgroup::may_access() called without proper synchronization");
328
329 list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
330 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
331 continue;
332 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
333 continue;
334 if (ex->major != ~0 && ex->major != refex->major)
335 continue;
336 if (ex->minor != ~0 && ex->minor != refex->minor)
337 continue;
338 if (refex->access & (~ex->access))
339 continue;
340 match = true;
341 break;
342 }
343
344 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
345 if (behavior == DEVCG_DEFAULT_ALLOW) {
346 /* the exception will deny access to certain devices */
347 return true;
348 } else {
349 /* the exception will allow access to certain devices */
350 if (match)
351 /*
352 * a new exception allowing access shouldn't
353 * match an parent's exception
354 */
355 return false;
356 return true;
357 }
358 } else {
359 /* only behavior == DEVCG_DEFAULT_DENY allowed here */
360 if (match)
361 /* parent has an exception that matches the proposed */
362 return true;
363 else
364 return false;
365 }
366 return false;
367 }
368
369 /*
370 * parent_has_perm:
371 * when adding a new allow rule to a device exception list, the rule
372 * must be allowed in the parent device
373 */
374 static int parent_has_perm(struct dev_cgroup *childcg,
375 struct dev_exception_item *ex)
376 {
377 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css));
378
379 if (!parent)
380 return 1;
381 return may_access(parent, ex, childcg->behavior);
382 }
383
384 /**
385 * may_allow_all - checks if it's possible to change the behavior to
386 * allow based on parent's rules.
387 * @parent: device cgroup's parent
388 * returns: != 0 in case it's allowed, 0 otherwise
389 */
390 static inline int may_allow_all(struct dev_cgroup *parent)
391 {
392 if (!parent)
393 return 1;
394 return parent->behavior == DEVCG_DEFAULT_ALLOW;
395 }
396
397 /**
398 * revalidate_active_exceptions - walks through the active exception list and
399 * revalidates the exceptions based on parent's
400 * behavior and exceptions. The exceptions that
401 * are no longer valid will be removed.
402 * Called with devcgroup_mutex held.
403 * @devcg: cgroup which exceptions will be checked
404 *
405 * This is one of the three key functions for hierarchy implementation.
406 * This function is responsible for re-evaluating all the cgroup's active
407 * exceptions due to a parent's exception change.
408 * Refer to Documentation/cgroups/devices.txt for more details.
409 */
410 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
411 {
412 struct dev_exception_item *ex;
413 struct list_head *this, *tmp;
414
415 list_for_each_safe(this, tmp, &devcg->exceptions) {
416 ex = container_of(this, struct dev_exception_item, list);
417 if (!parent_has_perm(devcg, ex))
418 dev_exception_rm(devcg, ex);
419 }
420 }
421
422 /**
423 * propagate_exception - propagates a new exception to the children
424 * @devcg_root: device cgroup that added a new exception
425 * @ex: new exception to be propagated
426 *
427 * returns: 0 in case of success, != 0 in case of error
428 */
429 static int propagate_exception(struct dev_cgroup *devcg_root,
430 struct dev_exception_item *ex)
431 {
432 struct cgroup_subsys_state *pos;
433 int rc = 0;
434
435 rcu_read_lock();
436
437 css_for_each_descendant_pre(pos, &devcg_root->css) {
438 struct dev_cgroup *devcg = css_to_devcgroup(pos);
439
440 /*
441 * Because devcgroup_mutex is held, no devcg will become
442 * online or offline during the tree walk (see on/offline
443 * methods), and online ones are safe to access outside RCU
444 * read lock without bumping refcnt.
445 */
446 if (pos == &devcg_root->css || !is_devcg_online(devcg))
447 continue;
448
449 rcu_read_unlock();
450
451 /*
452 * in case both root's behavior and devcg is allow, a new
453 * restriction means adding to the exception list
454 */
455 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
456 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
457 rc = dev_exception_add(devcg, ex);
458 if (rc)
459 break;
460 } else {
461 /*
462 * in the other possible cases:
463 * root's behavior: allow, devcg's: deny
464 * root's behavior: deny, devcg's: deny
465 * the exception will be removed
466 */
467 dev_exception_rm(devcg, ex);
468 }
469 revalidate_active_exceptions(devcg);
470
471 rcu_read_lock();
472 }
473
474 rcu_read_unlock();
475 return rc;
476 }
477
478 static inline bool has_children(struct dev_cgroup *devcgroup)
479 {
480 struct cgroup *cgrp = devcgroup->css.cgroup;
481
482 return !list_empty(&cgrp->children);
483 }
484
485 /*
486 * Modify the exception list using allow/deny rules.
487 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
488 * so we can give a container CAP_MKNOD to let it create devices but not
489 * modify the exception list.
490 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
491 * us to also grant CAP_SYS_ADMIN to containers without giving away the
492 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
493 *
494 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
495 * new access is only allowed if you're in the top-level cgroup, or your
496 * parent cgroup has the access you're asking for.
497 */
498 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
499 int filetype, char *buffer)
500 {
501 const char *b;
502 char temp[12]; /* 11 + 1 characters needed for a u32 */
503 int count, rc = 0;
504 struct dev_exception_item ex;
505 struct dev_cgroup *parent = css_to_devcgroup(css_parent(&devcgroup->css));
506
507 if (!capable(CAP_SYS_ADMIN))
508 return -EPERM;
509
510 memset(&ex, 0, sizeof(ex));
511 b = buffer;
512
513 switch (*b) {
514 case 'a':
515 switch (filetype) {
516 case DEVCG_ALLOW:
517 if (has_children(devcgroup))
518 return -EINVAL;
519
520 if (!may_allow_all(parent))
521 return -EPERM;
522 dev_exception_clean(devcgroup);
523 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
524 if (!parent)
525 break;
526
527 rc = dev_exceptions_copy(&devcgroup->exceptions,
528 &parent->exceptions);
529 if (rc)
530 return rc;
531 break;
532 case DEVCG_DENY:
533 if (has_children(devcgroup))
534 return -EINVAL;
535
536 dev_exception_clean(devcgroup);
537 devcgroup->behavior = DEVCG_DEFAULT_DENY;
538 break;
539 default:
540 return -EINVAL;
541 }
542 return 0;
543 case 'b':
544 ex.type = DEV_BLOCK;
545 break;
546 case 'c':
547 ex.type = DEV_CHAR;
548 break;
549 default:
550 return -EINVAL;
551 }
552 b++;
553 if (!isspace(*b))
554 return -EINVAL;
555 b++;
556 if (*b == '*') {
557 ex.major = ~0;
558 b++;
559 } else if (isdigit(*b)) {
560 memset(temp, 0, sizeof(temp));
561 for (count = 0; count < sizeof(temp) - 1; count++) {
562 temp[count] = *b;
563 b++;
564 if (!isdigit(*b))
565 break;
566 }
567 rc = kstrtou32(temp, 10, &ex.major);
568 if (rc)
569 return -EINVAL;
570 } else {
571 return -EINVAL;
572 }
573 if (*b != ':')
574 return -EINVAL;
575 b++;
576
577 /* read minor */
578 if (*b == '*') {
579 ex.minor = ~0;
580 b++;
581 } else if (isdigit(*b)) {
582 memset(temp, 0, sizeof(temp));
583 for (count = 0; count < sizeof(temp) - 1; count++) {
584 temp[count] = *b;
585 b++;
586 if (!isdigit(*b))
587 break;
588 }
589 rc = kstrtou32(temp, 10, &ex.minor);
590 if (rc)
591 return -EINVAL;
592 } else {
593 return -EINVAL;
594 }
595 if (!isspace(*b))
596 return -EINVAL;
597 for (b++, count = 0; count < 3; count++, b++) {
598 switch (*b) {
599 case 'r':
600 ex.access |= ACC_READ;
601 break;
602 case 'w':
603 ex.access |= ACC_WRITE;
604 break;
605 case 'm':
606 ex.access |= ACC_MKNOD;
607 break;
608 case '\n':
609 case '\0':
610 count = 3;
611 break;
612 default:
613 return -EINVAL;
614 }
615 }
616
617 switch (filetype) {
618 case DEVCG_ALLOW:
619 if (!parent_has_perm(devcgroup, &ex))
620 return -EPERM;
621 /*
622 * If the default policy is to allow by default, try to remove
623 * an matching exception instead. And be silent about it: we
624 * don't want to break compatibility
625 */
626 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
627 dev_exception_rm(devcgroup, &ex);
628 return 0;
629 }
630 rc = dev_exception_add(devcgroup, &ex);
631 break;
632 case DEVCG_DENY:
633 /*
634 * If the default policy is to deny by default, try to remove
635 * an matching exception instead. And be silent about it: we
636 * don't want to break compatibility
637 */
638 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
639 dev_exception_rm(devcgroup, &ex);
640 else
641 rc = dev_exception_add(devcgroup, &ex);
642
643 if (rc)
644 break;
645 /* we only propagate new restrictions */
646 rc = propagate_exception(devcgroup, &ex);
647 break;
648 default:
649 rc = -EINVAL;
650 }
651 return rc;
652 }
653
654 static int devcgroup_access_write(struct cgroup_subsys_state *css,
655 struct cftype *cft, char *buffer)
656 {
657 int retval;
658
659 mutex_lock(&devcgroup_mutex);
660 retval = devcgroup_update_access(css_to_devcgroup(css),
661 cft->private, buffer);
662 mutex_unlock(&devcgroup_mutex);
663 return retval;
664 }
665
666 static struct cftype dev_cgroup_files[] = {
667 {
668 .name = "allow",
669 .write_string = devcgroup_access_write,
670 .private = DEVCG_ALLOW,
671 },
672 {
673 .name = "deny",
674 .write_string = devcgroup_access_write,
675 .private = DEVCG_DENY,
676 },
677 {
678 .name = "list",
679 .seq_show = devcgroup_seq_show,
680 .private = DEVCG_LIST,
681 },
682 { } /* terminate */
683 };
684
685 struct cgroup_subsys devices_cgrp_subsys = {
686 .css_alloc = devcgroup_css_alloc,
687 .css_free = devcgroup_css_free,
688 .css_online = devcgroup_online,
689 .css_offline = devcgroup_offline,
690 .base_cftypes = dev_cgroup_files,
691 };
692
693 /**
694 * __devcgroup_check_permission - checks if an inode operation is permitted
695 * @dev_cgroup: the dev cgroup to be tested against
696 * @type: device type
697 * @major: device major number
698 * @minor: device minor number
699 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
700 *
701 * returns 0 on success, -EPERM case the operation is not permitted
702 */
703 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
704 short access)
705 {
706 struct dev_cgroup *dev_cgroup;
707 struct dev_exception_item ex;
708 int rc;
709
710 memset(&ex, 0, sizeof(ex));
711 ex.type = type;
712 ex.major = major;
713 ex.minor = minor;
714 ex.access = access;
715
716 rcu_read_lock();
717 dev_cgroup = task_devcgroup(current);
718 rc = may_access(dev_cgroup, &ex, dev_cgroup->behavior);
719 rcu_read_unlock();
720
721 if (!rc)
722 return -EPERM;
723
724 return 0;
725 }
726
727 int __devcgroup_inode_permission(struct inode *inode, int mask)
728 {
729 short type, access = 0;
730
731 if (S_ISBLK(inode->i_mode))
732 type = DEV_BLOCK;
733 if (S_ISCHR(inode->i_mode))
734 type = DEV_CHAR;
735 if (mask & MAY_WRITE)
736 access |= ACC_WRITE;
737 if (mask & MAY_READ)
738 access |= ACC_READ;
739
740 return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
741 access);
742 }
743
744 int devcgroup_inode_mknod(int mode, dev_t dev)
745 {
746 short type;
747
748 if (!S_ISBLK(mode) && !S_ISCHR(mode))
749 return 0;
750
751 if (S_ISBLK(mode))
752 type = DEV_BLOCK;
753 else
754 type = DEV_CHAR;
755
756 return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),
757 ACC_MKNOD);
758
759 }
This page took 0.046094 seconds and 5 git commands to generate.