d5c15a7c67c3b5cf6a0d26a93e4c63072d8b85dc
[linux-2.6.git] / security / device_cgroup.c
1 /*
2  * dev_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
14 #define ACC_MKNOD 1
15 #define ACC_READ  2
16 #define ACC_WRITE 4
17 #define ACC_MASK (ACC_MKNOD | ACC_READ | ACC_WRITE)
18
19 #define DEV_BLOCK 1
20 #define DEV_CHAR  2
21 #define DEV_ALL   4  /* this represents all devices */
22
23 /*
24  * whitelist locking rules:
25  * cgroup_lock() cannot be taken under dev_cgroup->lock.
26  * dev_cgroup->lock can be taken with or without cgroup_lock().
27  *
28  * modifications always require cgroup_lock
29  * modifications to a list which is visible require the
30  *   dev_cgroup->lock *and* cgroup_lock()
31  * walking the list requires dev_cgroup->lock or cgroup_lock().
32  *
33  * reasoning: dev_whitelist_copy() needs to kmalloc, so needs
34  *   a mutex, which the cgroup_lock() is.  Since modifying
35  *   a visible list requires both locks, either lock can be
36  *   taken for walking the list.
37  */
38
39 struct dev_whitelist_item {
40         u32 major, minor;
41         short type;
42         short access;
43         struct list_head list;
44         struct rcu_head rcu;
45 };
46
47 struct dev_cgroup {
48         struct cgroup_subsys_state css;
49         struct list_head whitelist;
50         spinlock_t lock;
51 };
52
53 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
54 {
55         return container_of(s, struct dev_cgroup, css);
56 }
57
58 static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup)
59 {
60         return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id));
61 }
62
63 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
64 {
65         return css_to_devcgroup(task_subsys_state(task, devices_subsys_id));
66 }
67
68 struct cgroup_subsys devices_subsys;
69
70 static int devcgroup_can_attach(struct cgroup_subsys *ss,
71                 struct cgroup *new_cgroup, struct task_struct *task)
72 {
73         if (current != task && !capable(CAP_SYS_ADMIN))
74                         return -EPERM;
75
76         return 0;
77 }
78
79 /*
80  * called under cgroup_lock()
81  */
82 static int dev_whitelist_copy(struct list_head *dest, struct list_head *orig)
83 {
84         struct dev_whitelist_item *wh, *tmp, *new;
85
86         list_for_each_entry(wh, orig, list) {
87                 new = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
88                 if (!new)
89                         goto free_and_exit;
90                 list_add_tail(&new->list, dest);
91         }
92
93         return 0;
94
95 free_and_exit:
96         list_for_each_entry_safe(wh, tmp, dest, list) {
97                 list_del(&wh->list);
98                 kfree(wh);
99         }
100         return -ENOMEM;
101 }
102
103 /* Stupid prototype - don't bother combining existing entries */
104 /*
105  * called under cgroup_lock()
106  * since the list is visible to other tasks, we need the spinlock also
107  */
108 static int dev_whitelist_add(struct dev_cgroup *dev_cgroup,
109                         struct dev_whitelist_item *wh)
110 {
111         struct dev_whitelist_item *whcopy, *walk;
112
113         whcopy = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
114         if (!whcopy)
115                 return -ENOMEM;
116
117         spin_lock(&dev_cgroup->lock);
118         list_for_each_entry(walk, &dev_cgroup->whitelist, list) {
119                 if (walk->type != wh->type)
120                         continue;
121                 if (walk->major != wh->major)
122                         continue;
123                 if (walk->minor != wh->minor)
124                         continue;
125
126                 walk->access |= wh->access;
127                 kfree(whcopy);
128                 whcopy = NULL;
129         }
130
131         if (whcopy != NULL)
132                 list_add_tail_rcu(&whcopy->list, &dev_cgroup->whitelist);
133         spin_unlock(&dev_cgroup->lock);
134         return 0;
135 }
136
137 static void whitelist_item_free(struct rcu_head *rcu)
138 {
139         struct dev_whitelist_item *item;
140
141         item = container_of(rcu, struct dev_whitelist_item, rcu);
142         kfree(item);
143 }
144
145 /*
146  * called under cgroup_lock()
147  * since the list is visible to other tasks, we need the spinlock also
148  */
149 static void dev_whitelist_rm(struct dev_cgroup *dev_cgroup,
150                         struct dev_whitelist_item *wh)
151 {
152         struct dev_whitelist_item *walk, *tmp;
153
154         spin_lock(&dev_cgroup->lock);
155         list_for_each_entry_safe(walk, tmp, &dev_cgroup->whitelist, list) {
156                 if (walk->type == DEV_ALL)
157                         goto remove;
158                 if (walk->type != wh->type)
159                         continue;
160                 if (walk->major != ~0 && walk->major != wh->major)
161                         continue;
162                 if (walk->minor != ~0 && walk->minor != wh->minor)
163                         continue;
164
165 remove:
166                 walk->access &= ~wh->access;
167                 if (!walk->access) {
168                         list_del_rcu(&walk->list);
169                         call_rcu(&walk->rcu, whitelist_item_free);
170                 }
171         }
172         spin_unlock(&dev_cgroup->lock);
173 }
174
175 /*
176  * called from kernel/cgroup.c with cgroup_lock() held.
177  */
178 static struct cgroup_subsys_state *devcgroup_create(struct cgroup_subsys *ss,
179                                                 struct cgroup *cgroup)
180 {
181         struct dev_cgroup *dev_cgroup, *parent_dev_cgroup;
182         struct cgroup *parent_cgroup;
183         int ret;
184
185         dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
186         if (!dev_cgroup)
187                 return ERR_PTR(-ENOMEM);
188         INIT_LIST_HEAD(&dev_cgroup->whitelist);
189         parent_cgroup = cgroup->parent;
190
191         if (parent_cgroup == NULL) {
192                 struct dev_whitelist_item *wh;
193                 wh = kmalloc(sizeof(*wh), GFP_KERNEL);
194                 if (!wh) {
195                         kfree(dev_cgroup);
196                         return ERR_PTR(-ENOMEM);
197                 }
198                 wh->minor = wh->major = ~0;
199                 wh->type = DEV_ALL;
200                 wh->access = ACC_MASK;
201                 list_add(&wh->list, &dev_cgroup->whitelist);
202         } else {
203                 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
204                 ret = dev_whitelist_copy(&dev_cgroup->whitelist,
205                                 &parent_dev_cgroup->whitelist);
206                 if (ret) {
207                         kfree(dev_cgroup);
208                         return ERR_PTR(ret);
209                 }
210         }
211
212         spin_lock_init(&dev_cgroup->lock);
213         return &dev_cgroup->css;
214 }
215
216 static void devcgroup_destroy(struct cgroup_subsys *ss,
217                         struct cgroup *cgroup)
218 {
219         struct dev_cgroup *dev_cgroup;
220         struct dev_whitelist_item *wh, *tmp;
221
222         dev_cgroup = cgroup_to_devcgroup(cgroup);
223         list_for_each_entry_safe(wh, tmp, &dev_cgroup->whitelist, list) {
224                 list_del(&wh->list);
225                 kfree(wh);
226         }
227         kfree(dev_cgroup);
228 }
229
230 #define DEVCG_ALLOW 1
231 #define DEVCG_DENY 2
232 #define DEVCG_LIST 3
233
234 #define MAJMINLEN 13
235 #define ACCLEN 4
236
237 static void set_access(char *acc, short access)
238 {
239         int idx = 0;
240         memset(acc, 0, ACCLEN);
241         if (access & ACC_READ)
242                 acc[idx++] = 'r';
243         if (access & ACC_WRITE)
244                 acc[idx++] = 'w';
245         if (access & ACC_MKNOD)
246                 acc[idx++] = 'm';
247 }
248
249 static char type_to_char(short type)
250 {
251         if (type == DEV_ALL)
252                 return 'a';
253         if (type == DEV_CHAR)
254                 return 'c';
255         if (type == DEV_BLOCK)
256                 return 'b';
257         return 'X';
258 }
259
260 static void set_majmin(char *str, unsigned m)
261 {
262         if (m == ~0)
263                 strcpy(str, "*");
264         else
265                 sprintf(str, "%u", m);
266 }
267
268 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
269                                 struct seq_file *m)
270 {
271         struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
272         struct dev_whitelist_item *wh;
273         char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
274
275         rcu_read_lock();
276         list_for_each_entry_rcu(wh, &devcgroup->whitelist, list) {
277                 set_access(acc, wh->access);
278                 set_majmin(maj, wh->major);
279                 set_majmin(min, wh->minor);
280                 seq_printf(m, "%c %s:%s %s\n", type_to_char(wh->type),
281                            maj, min, acc);
282         }
283         rcu_read_unlock();
284
285         return 0;
286 }
287
288 /*
289  * may_access_whitelist:
290  * does the access granted to dev_cgroup c contain the access
291  * requested in whitelist item refwh.
292  * return 1 if yes, 0 if no.
293  * call with c->lock held
294  */
295 static int may_access_whitelist(struct dev_cgroup *c,
296                                        struct dev_whitelist_item *refwh)
297 {
298         struct dev_whitelist_item *whitem;
299
300         list_for_each_entry(whitem, &c->whitelist, list) {
301                 if (whitem->type & DEV_ALL)
302                         return 1;
303                 if ((refwh->type & DEV_BLOCK) && !(whitem->type & DEV_BLOCK))
304                         continue;
305                 if ((refwh->type & DEV_CHAR) && !(whitem->type & DEV_CHAR))
306                         continue;
307                 if (whitem->major != ~0 && whitem->major != refwh->major)
308                         continue;
309                 if (whitem->minor != ~0 && whitem->minor != refwh->minor)
310                         continue;
311                 if (refwh->access & (~whitem->access))
312                         continue;
313                 return 1;
314         }
315         return 0;
316 }
317
318 /*
319  * parent_has_perm:
320  * when adding a new allow rule to a device whitelist, the rule
321  * must be allowed in the parent device
322  */
323 static int parent_has_perm(struct dev_cgroup *childcg,
324                                   struct dev_whitelist_item *wh)
325 {
326         struct cgroup *pcg = childcg->css.cgroup->parent;
327         struct dev_cgroup *parent;
328         int ret;
329
330         if (!pcg)
331                 return 1;
332         parent = cgroup_to_devcgroup(pcg);
333         spin_lock(&parent->lock);
334         ret = may_access_whitelist(parent, wh);
335         spin_unlock(&parent->lock);
336         return ret;
337 }
338
339 /*
340  * Modify the whitelist using allow/deny rules.
341  * CAP_SYS_ADMIN is needed for this.  It's at least separate from CAP_MKNOD
342  * so we can give a container CAP_MKNOD to let it create devices but not
343  * modify the whitelist.
344  * It seems likely we'll want to add a CAP_CONTAINER capability to allow
345  * us to also grant CAP_SYS_ADMIN to containers without giving away the
346  * device whitelist controls, but for now we'll stick with CAP_SYS_ADMIN
347  *
348  * Taking rules away is always allowed (given CAP_SYS_ADMIN).  Granting
349  * new access is only allowed if you're in the top-level cgroup, or your
350  * parent cgroup has the access you're asking for.
351  */
352 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
353                                    int filetype, const char *buffer)
354 {
355         struct dev_cgroup *cur_devcgroup;
356         const char *b;
357         char *endp;
358         int count;
359         struct dev_whitelist_item wh;
360
361         if (!capable(CAP_SYS_ADMIN))
362                 return -EPERM;
363
364         cur_devcgroup = task_devcgroup(current);
365
366         memset(&wh, 0, sizeof(wh));
367         b = buffer;
368
369         switch (*b) {
370         case 'a':
371                 wh.type = DEV_ALL;
372                 wh.access = ACC_MASK;
373                 wh.major = ~0;
374                 wh.minor = ~0;
375                 goto handle;
376         case 'b':
377                 wh.type = DEV_BLOCK;
378                 break;
379         case 'c':
380                 wh.type = DEV_CHAR;
381                 break;
382         default:
383                 return -EINVAL;
384         }
385         b++;
386         if (!isspace(*b))
387                 return -EINVAL;
388         b++;
389         if (*b == '*') {
390                 wh.major = ~0;
391                 b++;
392         } else if (isdigit(*b)) {
393                 wh.major = simple_strtoul(b, &endp, 10);
394                 b = endp;
395         } else {
396                 return -EINVAL;
397         }
398         if (*b != ':')
399                 return -EINVAL;
400         b++;
401
402         /* read minor */
403         if (*b == '*') {
404                 wh.minor = ~0;
405                 b++;
406         } else if (isdigit(*b)) {
407                 wh.minor = simple_strtoul(b, &endp, 10);
408                 b = endp;
409         } else {
410                 return -EINVAL;
411         }
412         if (!isspace(*b))
413                 return -EINVAL;
414         for (b++, count = 0; count < 3; count++, b++) {
415                 switch (*b) {
416                 case 'r':
417                         wh.access |= ACC_READ;
418                         break;
419                 case 'w':
420                         wh.access |= ACC_WRITE;
421                         break;
422                 case 'm':
423                         wh.access |= ACC_MKNOD;
424                         break;
425                 case '\n':
426                 case '\0':
427                         count = 3;
428                         break;
429                 default:
430                         return -EINVAL;
431                 }
432         }
433
434 handle:
435         switch (filetype) {
436         case DEVCG_ALLOW:
437                 if (!parent_has_perm(devcgroup, &wh))
438                         return -EPERM;
439                 return dev_whitelist_add(devcgroup, &wh);
440         case DEVCG_DENY:
441                 dev_whitelist_rm(devcgroup, &wh);
442                 break;
443         default:
444                 return -EINVAL;
445         }
446         return 0;
447 }
448
449 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
450                                   const char *buffer)
451 {
452         int retval;
453         if (!cgroup_lock_live_group(cgrp))
454                 return -ENODEV;
455         retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
456                                          cft->private, buffer);
457         cgroup_unlock();
458         return retval;
459 }
460
461 static struct cftype dev_cgroup_files[] = {
462         {
463                 .name = "allow",
464                 .write_string  = devcgroup_access_write,
465                 .private = DEVCG_ALLOW,
466         },
467         {
468                 .name = "deny",
469                 .write_string = devcgroup_access_write,
470                 .private = DEVCG_DENY,
471         },
472         {
473                 .name = "list",
474                 .read_seq_string = devcgroup_seq_read,
475                 .private = DEVCG_LIST,
476         },
477 };
478
479 static int devcgroup_populate(struct cgroup_subsys *ss,
480                                 struct cgroup *cgroup)
481 {
482         return cgroup_add_files(cgroup, ss, dev_cgroup_files,
483                                         ARRAY_SIZE(dev_cgroup_files));
484 }
485
486 struct cgroup_subsys devices_subsys = {
487         .name = "devices",
488         .can_attach = devcgroup_can_attach,
489         .create = devcgroup_create,
490         .destroy  = devcgroup_destroy,
491         .populate = devcgroup_populate,
492         .subsys_id = devices_subsys_id,
493 };
494
495 int devcgroup_inode_permission(struct inode *inode, int mask)
496 {
497         struct dev_cgroup *dev_cgroup;
498         struct dev_whitelist_item *wh;
499
500         dev_t device = inode->i_rdev;
501         if (!device)
502                 return 0;
503         if (!S_ISBLK(inode->i_mode) && !S_ISCHR(inode->i_mode))
504                 return 0;
505
506         rcu_read_lock();
507
508         dev_cgroup = task_devcgroup(current);
509
510         list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
511                 if (wh->type & DEV_ALL)
512                         goto acc_check;
513                 if ((wh->type & DEV_BLOCK) && !S_ISBLK(inode->i_mode))
514                         continue;
515                 if ((wh->type & DEV_CHAR) && !S_ISCHR(inode->i_mode))
516                         continue;
517                 if (wh->major != ~0 && wh->major != imajor(inode))
518                         continue;
519                 if (wh->minor != ~0 && wh->minor != iminor(inode))
520                         continue;
521 acc_check:
522                 if ((mask & MAY_WRITE) && !(wh->access & ACC_WRITE))
523                         continue;
524                 if ((mask & MAY_READ) && !(wh->access & ACC_READ))
525                         continue;
526                 rcu_read_unlock();
527                 return 0;
528         }
529
530         rcu_read_unlock();
531
532         return -EPERM;
533 }
534
535 int devcgroup_inode_mknod(int mode, dev_t dev)
536 {
537         struct dev_cgroup *dev_cgroup;
538         struct dev_whitelist_item *wh;
539
540         rcu_read_lock();
541
542         dev_cgroup = task_devcgroup(current);
543
544         list_for_each_entry(wh, &dev_cgroup->whitelist, list) {
545                 if (wh->type & DEV_ALL)
546                         goto acc_check;
547                 if ((wh->type & DEV_BLOCK) && !S_ISBLK(mode))
548                         continue;
549                 if ((wh->type & DEV_CHAR) && !S_ISCHR(mode))
550                         continue;
551                 if (wh->major != ~0 && wh->major != MAJOR(dev))
552                         continue;
553                 if (wh->minor != ~0 && wh->minor != MINOR(dev))
554                         continue;
555 acc_check:
556                 if (!(wh->access & ACC_MKNOD))
557                         continue;
558                 rcu_read_unlock();
559                 return 0;
560         }
561
562         rcu_read_unlock();
563
564         return -EPERM;
565 }