Root/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
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_subsys_id));
62}
63
64struct cgroup_subsys devices_subsys;
65
66/*
67 * called under devcgroup_mutex
68 */
69static 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
84free_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 */
95static 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 */
127static 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
150static 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 */
166static 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
173static 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 */
184static 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
205static 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 */
217static struct cgroup_subsys_state *
218devcgroup_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
231static 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
246static 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
258static 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
269static void set_majmin(char *str, unsigned m)
270{
271    if (m == ~0)
272        strcpy(str, "*");
273    else
274        sprintf(str, "%u", m);
275}
276
277static int devcgroup_seq_read(struct cgroup_subsys_state *css,
278                  struct cftype *cft, struct seq_file *m)
279{
280    struct dev_cgroup *devcgroup = css_to_devcgroup(css);
281    struct dev_exception_item *ex;
282    char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
283
284    rcu_read_lock();
285    /*
286     * To preserve the compatibility:
287     * - Only show the "all devices" when the default policy is to allow
288     * - List the exceptions in case the default policy is to deny
289     * This way, the file remains as a "whitelist of devices"
290     */
291    if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
292        set_access(acc, ACC_MASK);
293        set_majmin(maj, ~0);
294        set_majmin(min, ~0);
295        seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
296               maj, min, acc);
297    } else {
298        list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
299            set_access(acc, ex->access);
300            set_majmin(maj, ex->major);
301            set_majmin(min, ex->minor);
302            seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
303                   maj, min, acc);
304        }
305    }
306    rcu_read_unlock();
307
308    return 0;
309}
310
311/**
312 * may_access - verifies if a new exception is part of what is allowed
313 * by a dev cgroup based on the default policy +
314 * exceptions. This is used to make sure a child cgroup
315 * won't have more privileges than its parent or to
316 * verify if a certain access is allowed.
317 * @dev_cgroup: dev cgroup to be tested against
318 * @refex: new exception
319 * @behavior: behavior of the exception
320 */
321static bool may_access(struct dev_cgroup *dev_cgroup,
322               struct dev_exception_item *refex,
323               enum devcg_behavior behavior)
324{
325    struct dev_exception_item *ex;
326    bool match = false;
327
328    rcu_lockdep_assert(rcu_read_lock_held() ||
329               lockdep_is_held(&devcgroup_mutex),
330               "device_cgroup::may_access() called without proper synchronization");
331
332    list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
333        if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
334            continue;
335        if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
336            continue;
337        if (ex->major != ~0 && ex->major != refex->major)
338            continue;
339        if (ex->minor != ~0 && ex->minor != refex->minor)
340            continue;
341        if (refex->access & (~ex->access))
342            continue;
343        match = true;
344        break;
345    }
346
347    if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
348        if (behavior == DEVCG_DEFAULT_ALLOW) {
349            /* the exception will deny access to certain devices */
350            return true;
351        } else {
352            /* the exception will allow access to certain devices */
353            if (match)
354                /*
355                 * a new exception allowing access shouldn't
356                 * match an parent's exception
357                 */
358                return false;
359            return true;
360        }
361    } else {
362        /* only behavior == DEVCG_DEFAULT_DENY allowed here */
363        if (match)
364            /* parent has an exception that matches the proposed */
365            return true;
366        else
367            return false;
368    }
369    return false;
370}
371
372/*
373 * parent_has_perm:
374 * when adding a new allow rule to a device exception list, the rule
375 * must be allowed in the parent device
376 */
377static int parent_has_perm(struct dev_cgroup *childcg,
378                  struct dev_exception_item *ex)
379{
380    struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css));
381
382    if (!parent)
383        return 1;
384    return may_access(parent, ex, childcg->behavior);
385}
386
387/**
388 * may_allow_all - checks if it's possible to change the behavior to
389 * allow based on parent's rules.
390 * @parent: device cgroup's parent
391 * returns: != 0 in case it's allowed, 0 otherwise
392 */
393static inline int may_allow_all(struct dev_cgroup *parent)
394{
395    if (!parent)
396        return 1;
397    return parent->behavior == DEVCG_DEFAULT_ALLOW;
398}
399
400/**
401 * revalidate_active_exceptions - walks through the active exception list and
402 * revalidates the exceptions based on parent's
403 * behavior and exceptions. The exceptions that
404 * are no longer valid will be removed.
405 * Called with devcgroup_mutex held.
406 * @devcg: cgroup which exceptions will be checked
407 *
408 * This is one of the three key functions for hierarchy implementation.
409 * This function is responsible for re-evaluating all the cgroup's active
410 * exceptions due to a parent's exception change.
411 * Refer to Documentation/cgroups/devices.txt for more details.
412 */
413static void revalidate_active_exceptions(struct dev_cgroup *devcg)
414{
415    struct dev_exception_item *ex;
416    struct list_head *this, *tmp;
417
418    list_for_each_safe(this, tmp, &devcg->exceptions) {
419        ex = container_of(this, struct dev_exception_item, list);
420        if (!parent_has_perm(devcg, ex))
421            dev_exception_rm(devcg, ex);
422    }
423}
424
425/**
426 * propagate_exception - propagates a new exception to the children
427 * @devcg_root: device cgroup that added a new exception
428 * @ex: new exception to be propagated
429 *
430 * returns: 0 in case of success, != 0 in case of error
431 */
432static int propagate_exception(struct dev_cgroup *devcg_root,
433                   struct dev_exception_item *ex)
434{
435    struct cgroup_subsys_state *pos;
436    int rc = 0;
437
438    rcu_read_lock();
439
440    css_for_each_descendant_pre(pos, &devcg_root->css) {
441        struct dev_cgroup *devcg = css_to_devcgroup(pos);
442
443        /*
444         * Because devcgroup_mutex is held, no devcg will become
445         * online or offline during the tree walk (see on/offline
446         * methods), and online ones are safe to access outside RCU
447         * read lock without bumping refcnt.
448         */
449        if (pos == &devcg_root->css || !is_devcg_online(devcg))
450            continue;
451
452        rcu_read_unlock();
453
454        /*
455         * in case both root's behavior and devcg is allow, a new
456         * restriction means adding to the exception list
457         */
458        if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
459            devcg->behavior == DEVCG_DEFAULT_ALLOW) {
460            rc = dev_exception_add(devcg, ex);
461            if (rc)
462                break;
463        } else {
464            /*
465             * in the other possible cases:
466             * root's behavior: allow, devcg's: deny
467             * root's behavior: deny, devcg's: deny
468             * the exception will be removed
469             */
470            dev_exception_rm(devcg, ex);
471        }
472        revalidate_active_exceptions(devcg);
473
474        rcu_read_lock();
475    }
476
477    rcu_read_unlock();
478    return rc;
479}
480
481static inline bool has_children(struct dev_cgroup *devcgroup)
482{
483    struct cgroup *cgrp = devcgroup->css.cgroup;
484
485    return !list_empty(&cgrp->children);
486}
487
488/*
489 * Modify the exception list using allow/deny rules.
490 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
491 * so we can give a container CAP_MKNOD to let it create devices but not
492 * modify the exception list.
493 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
494 * us to also grant CAP_SYS_ADMIN to containers without giving away the
495 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
496 *
497 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
498 * new access is only allowed if you're in the top-level cgroup, or your
499 * parent cgroup has the access you're asking for.
500 */
501static int devcgroup_update_access(struct dev_cgroup *devcgroup,
502                   int filetype, const char *buffer)
503{
504    const char *b;
505    char temp[12]; /* 11 + 1 characters needed for a u32 */
506    int count, rc = 0;
507    struct dev_exception_item ex;
508    struct dev_cgroup *parent = css_to_devcgroup(css_parent(&devcgroup->css));
509
510    if (!capable(CAP_SYS_ADMIN))
511        return -EPERM;
512
513    memset(&ex, 0, sizeof(ex));
514    b = buffer;
515
516    switch (*b) {
517    case 'a':
518        switch (filetype) {
519        case DEVCG_ALLOW:
520            if (has_children(devcgroup))
521                return -EINVAL;
522
523            if (!may_allow_all(parent))
524                return -EPERM;
525            dev_exception_clean(devcgroup);
526            devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
527            if (!parent)
528                break;
529
530            rc = dev_exceptions_copy(&devcgroup->exceptions,
531                         &parent->exceptions);
532            if (rc)
533                return rc;
534            break;
535        case DEVCG_DENY:
536            if (has_children(devcgroup))
537                return -EINVAL;
538
539            dev_exception_clean(devcgroup);
540            devcgroup->behavior = DEVCG_DEFAULT_DENY;
541            break;
542        default:
543            return -EINVAL;
544        }
545        return 0;
546    case 'b':
547        ex.type = DEV_BLOCK;
548        break;
549    case 'c':
550        ex.type = DEV_CHAR;
551        break;
552    default:
553        return -EINVAL;
554    }
555    b++;
556    if (!isspace(*b))
557        return -EINVAL;
558    b++;
559    if (*b == '*') {
560        ex.major = ~0;
561        b++;
562    } else if (isdigit(*b)) {
563        memset(temp, 0, sizeof(temp));
564        for (count = 0; count < sizeof(temp) - 1; count++) {
565            temp[count] = *b;
566            b++;
567            if (!isdigit(*b))
568                break;
569        }
570        rc = kstrtou32(temp, 10, &ex.major);
571        if (rc)
572            return -EINVAL;
573    } else {
574        return -EINVAL;
575    }
576    if (*b != ':')
577        return -EINVAL;
578    b++;
579
580    /* read minor */
581    if (*b == '*') {
582        ex.minor = ~0;
583        b++;
584    } else if (isdigit(*b)) {
585        memset(temp, 0, sizeof(temp));
586        for (count = 0; count < sizeof(temp) - 1; count++) {
587            temp[count] = *b;
588            b++;
589            if (!isdigit(*b))
590                break;
591        }
592        rc = kstrtou32(temp, 10, &ex.minor);
593        if (rc)
594            return -EINVAL;
595    } else {
596        return -EINVAL;
597    }
598    if (!isspace(*b))
599        return -EINVAL;
600    for (b++, count = 0; count < 3; count++, b++) {
601        switch (*b) {
602        case 'r':
603            ex.access |= ACC_READ;
604            break;
605        case 'w':
606            ex.access |= ACC_WRITE;
607            break;
608        case 'm':
609            ex.access |= ACC_MKNOD;
610            break;
611        case '\n':
612        case '\0':
613            count = 3;
614            break;
615        default:
616            return -EINVAL;
617        }
618    }
619
620    switch (filetype) {
621    case DEVCG_ALLOW:
622        if (!parent_has_perm(devcgroup, &ex))
623            return -EPERM;
624        /*
625         * If the default policy is to allow by default, try to remove
626         * an matching exception instead. And be silent about it: we
627         * don't want to break compatibility
628         */
629        if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
630            dev_exception_rm(devcgroup, &ex);
631            return 0;
632        }
633        rc = dev_exception_add(devcgroup, &ex);
634        break;
635    case DEVCG_DENY:
636        /*
637         * If the default policy is to deny by default, try to remove
638         * an matching exception instead. And be silent about it: we
639         * don't want to break compatibility
640         */
641        if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
642            dev_exception_rm(devcgroup, &ex);
643        else
644            rc = dev_exception_add(devcgroup, &ex);
645
646        if (rc)
647            break;
648        /* we only propagate new restrictions */
649        rc = propagate_exception(devcgroup, &ex);
650        break;
651    default:
652        rc = -EINVAL;
653    }
654    return rc;
655}
656
657static int devcgroup_access_write(struct cgroup_subsys_state *css,
658                  struct cftype *cft, const char *buffer)
659{
660    int retval;
661
662    mutex_lock(&devcgroup_mutex);
663    retval = devcgroup_update_access(css_to_devcgroup(css),
664                     cft->private, buffer);
665    mutex_unlock(&devcgroup_mutex);
666    return retval;
667}
668
669static struct cftype dev_cgroup_files[] = {
670    {
671        .name = "allow",
672        .write_string = devcgroup_access_write,
673        .private = DEVCG_ALLOW,
674    },
675    {
676        .name = "deny",
677        .write_string = devcgroup_access_write,
678        .private = DEVCG_DENY,
679    },
680    {
681        .name = "list",
682        .read_seq_string = devcgroup_seq_read,
683        .private = DEVCG_LIST,
684    },
685    { } /* terminate */
686};
687
688struct cgroup_subsys devices_subsys = {
689    .name = "devices",
690    .css_alloc = devcgroup_css_alloc,
691    .css_free = devcgroup_css_free,
692    .css_online = devcgroup_online,
693    .css_offline = devcgroup_offline,
694    .subsys_id = devices_subsys_id,
695    .base_cftypes = dev_cgroup_files,
696};
697
698/**
699 * __devcgroup_check_permission - checks if an inode operation is permitted
700 * @dev_cgroup: the dev cgroup to be tested against
701 * @type: device type
702 * @major: device major number
703 * @minor: device minor number
704 * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
705 *
706 * returns 0 on success, -EPERM case the operation is not permitted
707 */
708static int __devcgroup_check_permission(short type, u32 major, u32 minor,
709                        short access)
710{
711    struct dev_cgroup *dev_cgroup;
712    struct dev_exception_item ex;
713    int rc;
714
715    memset(&ex, 0, sizeof(ex));
716    ex.type = type;
717    ex.major = major;
718    ex.minor = minor;
719    ex.access = access;
720
721    rcu_read_lock();
722    dev_cgroup = task_devcgroup(current);
723    rc = may_access(dev_cgroup, &ex, dev_cgroup->behavior);
724    rcu_read_unlock();
725
726    if (!rc)
727        return -EPERM;
728
729    return 0;
730}
731
732int __devcgroup_inode_permission(struct inode *inode, int mask)
733{
734    short type, access = 0;
735
736    if (S_ISBLK(inode->i_mode))
737        type = DEV_BLOCK;
738    if (S_ISCHR(inode->i_mode))
739        type = DEV_CHAR;
740    if (mask & MAY_WRITE)
741        access |= ACC_WRITE;
742    if (mask & MAY_READ)
743        access |= ACC_READ;
744
745    return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
746            access);
747}
748
749int devcgroup_inode_mknod(int mode, dev_t dev)
750{
751    short type;
752
753    if (!S_ISBLK(mode) && !S_ISCHR(mode))
754        return 0;
755
756    if (S_ISBLK(mode))
757        type = DEV_BLOCK;
758    else
759        type = DEV_CHAR;
760
761    return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),
762            ACC_MKNOD);
763
764}
765

Archive Download this file



interactive