Commit | Line | Data |
---|---|---|
08ce5f16 SH |
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> | |
29486df3 | 12 | #include <linux/seq_file.h> |
08ce5f16 SH |
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 | }; | |
45 | ||
46 | struct dev_cgroup { | |
47 | struct cgroup_subsys_state css; | |
48 | struct list_head whitelist; | |
49 | spinlock_t lock; | |
50 | }; | |
51 | ||
52 | static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup) | |
53 | { | |
54 | return container_of(cgroup_subsys_state(cgroup, devices_subsys_id), | |
55 | struct dev_cgroup, css); | |
56 | } | |
57 | ||
58 | struct cgroup_subsys devices_subsys; | |
59 | ||
60 | static int devcgroup_can_attach(struct cgroup_subsys *ss, | |
61 | struct cgroup *new_cgroup, struct task_struct *task) | |
62 | { | |
63 | if (current != task && !capable(CAP_SYS_ADMIN)) | |
64 | return -EPERM; | |
65 | ||
66 | return 0; | |
67 | } | |
68 | ||
69 | /* | |
70 | * called under cgroup_lock() | |
71 | */ | |
72 | static int dev_whitelist_copy(struct list_head *dest, struct list_head *orig) | |
73 | { | |
74 | struct dev_whitelist_item *wh, *tmp, *new; | |
75 | ||
76 | list_for_each_entry(wh, orig, list) { | |
77 | new = kmalloc(sizeof(*wh), GFP_KERNEL); | |
78 | if (!new) | |
79 | goto free_and_exit; | |
80 | new->major = wh->major; | |
81 | new->minor = wh->minor; | |
82 | new->type = wh->type; | |
83 | new->access = wh->access; | |
84 | list_add_tail(&new->list, dest); | |
85 | } | |
86 | ||
87 | return 0; | |
88 | ||
89 | free_and_exit: | |
90 | list_for_each_entry_safe(wh, tmp, dest, list) { | |
91 | list_del(&wh->list); | |
92 | kfree(wh); | |
93 | } | |
94 | return -ENOMEM; | |
95 | } | |
96 | ||
97 | /* Stupid prototype - don't bother combining existing entries */ | |
98 | /* | |
99 | * called under cgroup_lock() | |
100 | * since the list is visible to other tasks, we need the spinlock also | |
101 | */ | |
102 | static int dev_whitelist_add(struct dev_cgroup *dev_cgroup, | |
103 | struct dev_whitelist_item *wh) | |
104 | { | |
105 | struct dev_whitelist_item *whcopy; | |
106 | ||
107 | whcopy = kmalloc(sizeof(*whcopy), GFP_KERNEL); | |
108 | if (!whcopy) | |
109 | return -ENOMEM; | |
110 | ||
111 | memcpy(whcopy, wh, sizeof(*whcopy)); | |
112 | spin_lock(&dev_cgroup->lock); | |
113 | list_add_tail(&whcopy->list, &dev_cgroup->whitelist); | |
114 | spin_unlock(&dev_cgroup->lock); | |
115 | return 0; | |
116 | } | |
117 | ||
118 | /* | |
119 | * called under cgroup_lock() | |
120 | * since the list is visible to other tasks, we need the spinlock also | |
121 | */ | |
122 | static void dev_whitelist_rm(struct dev_cgroup *dev_cgroup, | |
123 | struct dev_whitelist_item *wh) | |
124 | { | |
125 | struct dev_whitelist_item *walk, *tmp; | |
126 | ||
127 | spin_lock(&dev_cgroup->lock); | |
128 | list_for_each_entry_safe(walk, tmp, &dev_cgroup->whitelist, list) { | |
129 | if (walk->type == DEV_ALL) | |
130 | goto remove; | |
131 | if (walk->type != wh->type) | |
132 | continue; | |
133 | if (walk->major != ~0 && walk->major != wh->major) | |
134 | continue; | |
135 | if (walk->minor != ~0 && walk->minor != wh->minor) | |
136 | continue; | |
137 | ||
138 | remove: | |
139 | walk->access &= ~wh->access; | |
140 | if (!walk->access) { | |
141 | list_del(&walk->list); | |
142 | kfree(walk); | |
143 | } | |
144 | } | |
145 | spin_unlock(&dev_cgroup->lock); | |
146 | } | |
147 | ||
148 | /* | |
149 | * called from kernel/cgroup.c with cgroup_lock() held. | |
150 | */ | |
151 | static struct cgroup_subsys_state *devcgroup_create(struct cgroup_subsys *ss, | |
152 | struct cgroup *cgroup) | |
153 | { | |
154 | struct dev_cgroup *dev_cgroup, *parent_dev_cgroup; | |
155 | struct cgroup *parent_cgroup; | |
156 | int ret; | |
157 | ||
158 | dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL); | |
159 | if (!dev_cgroup) | |
160 | return ERR_PTR(-ENOMEM); | |
161 | INIT_LIST_HEAD(&dev_cgroup->whitelist); | |
162 | parent_cgroup = cgroup->parent; | |
163 | ||
164 | if (parent_cgroup == NULL) { | |
165 | struct dev_whitelist_item *wh; | |
166 | wh = kmalloc(sizeof(*wh), GFP_KERNEL); | |
167 | if (!wh) { | |
168 | kfree(dev_cgroup); | |
169 | return ERR_PTR(-ENOMEM); | |
170 | } | |
171 | wh->minor = wh->major = ~0; | |
172 | wh->type = DEV_ALL; | |
173 | wh->access = ACC_MKNOD | ACC_READ | ACC_WRITE; | |
174 | list_add(&wh->list, &dev_cgroup->whitelist); | |
175 | } else { | |
176 | parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup); | |
177 | ret = dev_whitelist_copy(&dev_cgroup->whitelist, | |
178 | &parent_dev_cgroup->whitelist); | |
179 | if (ret) { | |
180 | kfree(dev_cgroup); | |
181 | return ERR_PTR(ret); | |
182 | } | |
183 | } | |
184 | ||
185 | spin_lock_init(&dev_cgroup->lock); | |
186 | return &dev_cgroup->css; | |
187 | } | |
188 | ||
189 | static void devcgroup_destroy(struct cgroup_subsys *ss, | |
190 | struct cgroup *cgroup) | |
191 | { | |
192 | struct dev_cgroup *dev_cgroup; | |
193 | struct dev_whitelist_item *wh, *tmp; | |
194 | ||
195 | dev_cgroup = cgroup_to_devcgroup(cgroup); | |
196 | list_for_each_entry_safe(wh, tmp, &dev_cgroup->whitelist, list) { | |
197 | list_del(&wh->list); | |
198 | kfree(wh); | |
199 | } | |
200 | kfree(dev_cgroup); | |
201 | } | |
202 | ||
203 | #define DEVCG_ALLOW 1 | |
204 | #define DEVCG_DENY 2 | |
29486df3 SH |
205 | #define DEVCG_LIST 3 |
206 | ||
207 | #define MAJMINLEN 10 | |
208 | #define ACCLEN 4 | |
08ce5f16 SH |
209 | |
210 | static void set_access(char *acc, short access) | |
211 | { | |
212 | int idx = 0; | |
29486df3 | 213 | memset(acc, 0, ACCLEN); |
08ce5f16 SH |
214 | if (access & ACC_READ) |
215 | acc[idx++] = 'r'; | |
216 | if (access & ACC_WRITE) | |
217 | acc[idx++] = 'w'; | |
218 | if (access & ACC_MKNOD) | |
219 | acc[idx++] = 'm'; | |
220 | } | |
221 | ||
222 | static char type_to_char(short type) | |
223 | { | |
224 | if (type == DEV_ALL) | |
225 | return 'a'; | |
226 | if (type == DEV_CHAR) | |
227 | return 'c'; | |
228 | if (type == DEV_BLOCK) | |
229 | return 'b'; | |
230 | return 'X'; | |
231 | } | |
232 | ||
29486df3 | 233 | static void set_majmin(char *str, unsigned m) |
08ce5f16 | 234 | { |
29486df3 | 235 | memset(str, 0, MAJMINLEN); |
08ce5f16 SH |
236 | if (m == ~0) |
237 | sprintf(str, "*"); | |
238 | else | |
29486df3 | 239 | snprintf(str, MAJMINLEN, "%d", m); |
08ce5f16 SH |
240 | } |
241 | ||
29486df3 SH |
242 | static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft, |
243 | struct seq_file *m) | |
08ce5f16 | 244 | { |
29486df3 | 245 | struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup); |
08ce5f16 | 246 | struct dev_whitelist_item *wh; |
29486df3 | 247 | char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN]; |
08ce5f16 SH |
248 | |
249 | spin_lock(&devcgroup->lock); | |
250 | list_for_each_entry(wh, &devcgroup->whitelist, list) { | |
251 | set_access(acc, wh->access); | |
29486df3 SH |
252 | set_majmin(maj, wh->major); |
253 | set_majmin(min, wh->minor); | |
254 | seq_printf(m, "%c %s:%s %s\n", type_to_char(wh->type), | |
255 | maj, min, acc); | |
08ce5f16 SH |
256 | } |
257 | spin_unlock(&devcgroup->lock); | |
258 | ||
29486df3 | 259 | return 0; |
08ce5f16 SH |
260 | } |
261 | ||
262 | /* | |
263 | * may_access_whitelist: | |
264 | * does the access granted to dev_cgroup c contain the access | |
265 | * requested in whitelist item refwh. | |
266 | * return 1 if yes, 0 if no. | |
267 | * call with c->lock held | |
268 | */ | |
269 | static int may_access_whitelist(struct dev_cgroup *c, | |
270 | struct dev_whitelist_item *refwh) | |
271 | { | |
272 | struct dev_whitelist_item *whitem; | |
273 | ||
274 | list_for_each_entry(whitem, &c->whitelist, list) { | |
275 | if (whitem->type & DEV_ALL) | |
276 | return 1; | |
277 | if ((refwh->type & DEV_BLOCK) && !(whitem->type & DEV_BLOCK)) | |
278 | continue; | |
279 | if ((refwh->type & DEV_CHAR) && !(whitem->type & DEV_CHAR)) | |
280 | continue; | |
281 | if (whitem->major != ~0 && whitem->major != refwh->major) | |
282 | continue; | |
283 | if (whitem->minor != ~0 && whitem->minor != refwh->minor) | |
284 | continue; | |
285 | if (refwh->access & (~(whitem->access | ACC_MASK))) | |
286 | continue; | |
287 | return 1; | |
288 | } | |
289 | return 0; | |
290 | } | |
291 | ||
292 | /* | |
293 | * parent_has_perm: | |
294 | * when adding a new allow rule to a device whitelist, the rule | |
295 | * must be allowed in the parent device | |
296 | */ | |
297 | static int parent_has_perm(struct cgroup *childcg, | |
298 | struct dev_whitelist_item *wh) | |
299 | { | |
300 | struct cgroup *pcg = childcg->parent; | |
301 | struct dev_cgroup *parent; | |
302 | int ret; | |
303 | ||
304 | if (!pcg) | |
305 | return 1; | |
306 | parent = cgroup_to_devcgroup(pcg); | |
307 | spin_lock(&parent->lock); | |
308 | ret = may_access_whitelist(parent, wh); | |
309 | spin_unlock(&parent->lock); | |
310 | return ret; | |
311 | } | |
312 | ||
313 | /* | |
314 | * Modify the whitelist using allow/deny rules. | |
315 | * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD | |
316 | * so we can give a container CAP_MKNOD to let it create devices but not | |
317 | * modify the whitelist. | |
318 | * It seems likely we'll want to add a CAP_CONTAINER capability to allow | |
319 | * us to also grant CAP_SYS_ADMIN to containers without giving away the | |
320 | * device whitelist controls, but for now we'll stick with CAP_SYS_ADMIN | |
321 | * | |
322 | * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting | |
323 | * new access is only allowed if you're in the top-level cgroup, or your | |
324 | * parent cgroup has the access you're asking for. | |
325 | */ | |
326 | static ssize_t devcgroup_access_write(struct cgroup *cgroup, struct cftype *cft, | |
327 | struct file *file, const char __user *userbuf, | |
328 | size_t nbytes, loff_t *ppos) | |
329 | { | |
330 | struct cgroup *cur_cgroup; | |
331 | struct dev_cgroup *devcgroup, *cur_devcgroup; | |
332 | int filetype = cft->private; | |
333 | char *buffer, *b; | |
334 | int retval = 0, count; | |
335 | struct dev_whitelist_item wh; | |
336 | ||
337 | if (!capable(CAP_SYS_ADMIN)) | |
338 | return -EPERM; | |
339 | ||
340 | devcgroup = cgroup_to_devcgroup(cgroup); | |
341 | cur_cgroup = task_cgroup(current, devices_subsys.subsys_id); | |
342 | cur_devcgroup = cgroup_to_devcgroup(cur_cgroup); | |
343 | ||
344 | buffer = kmalloc(nbytes+1, GFP_KERNEL); | |
345 | if (!buffer) | |
346 | return -ENOMEM; | |
347 | ||
348 | if (copy_from_user(buffer, userbuf, nbytes)) { | |
349 | retval = -EFAULT; | |
350 | goto out1; | |
351 | } | |
352 | buffer[nbytes] = 0; /* nul-terminate */ | |
353 | ||
354 | cgroup_lock(); | |
355 | if (cgroup_is_removed(cgroup)) { | |
356 | retval = -ENODEV; | |
357 | goto out2; | |
358 | } | |
359 | ||
360 | memset(&wh, 0, sizeof(wh)); | |
361 | b = buffer; | |
362 | ||
363 | switch (*b) { | |
364 | case 'a': | |
365 | wh.type = DEV_ALL; | |
366 | wh.access = ACC_MASK; | |
367 | goto handle; | |
368 | case 'b': | |
369 | wh.type = DEV_BLOCK; | |
370 | break; | |
371 | case 'c': | |
372 | wh.type = DEV_CHAR; | |
373 | break; | |
374 | default: | |
375 | retval = -EINVAL; | |
376 | goto out2; | |
377 | } | |
378 | b++; | |
379 | if (!isspace(*b)) { | |
380 | retval = -EINVAL; | |
381 | goto out2; | |
382 | } | |
383 | b++; | |
384 | if (*b == '*') { | |
385 | wh.major = ~0; | |
386 | b++; | |
387 | } else if (isdigit(*b)) { | |
388 | wh.major = 0; | |
389 | while (isdigit(*b)) { | |
390 | wh.major = wh.major*10+(*b-'0'); | |
391 | b++; | |
392 | } | |
393 | } else { | |
394 | retval = -EINVAL; | |
395 | goto out2; | |
396 | } | |
397 | if (*b != ':') { | |
398 | retval = -EINVAL; | |
399 | goto out2; | |
400 | } | |
401 | b++; | |
402 | ||
403 | /* read minor */ | |
404 | if (*b == '*') { | |
405 | wh.minor = ~0; | |
406 | b++; | |
407 | } else if (isdigit(*b)) { | |
408 | wh.minor = 0; | |
409 | while (isdigit(*b)) { | |
410 | wh.minor = wh.minor*10+(*b-'0'); | |
411 | b++; | |
412 | } | |
413 | } else { | |
414 | retval = -EINVAL; | |
415 | goto out2; | |
416 | } | |
417 | if (!isspace(*b)) { | |
418 | retval = -EINVAL; | |
419 | goto out2; | |
420 | } | |
421 | for (b++, count = 0; count < 3; count++, b++) { | |
422 | switch (*b) { | |
423 | case 'r': | |
424 | wh.access |= ACC_READ; | |
425 | break; | |
426 | case 'w': | |
427 | wh.access |= ACC_WRITE; | |
428 | break; | |
429 | case 'm': | |
430 | wh.access |= ACC_MKNOD; | |
431 | break; | |
432 | case '\n': | |
433 | case '\0': | |
434 | count = 3; | |
435 | break; | |
436 | default: | |
437 | retval = -EINVAL; | |
438 | goto out2; | |
439 | } | |
440 | } | |
441 | ||
442 | handle: | |
443 | retval = 0; | |
444 | switch (filetype) { | |
445 | case DEVCG_ALLOW: | |
446 | if (!parent_has_perm(cgroup, &wh)) | |
447 | retval = -EPERM; | |
448 | else | |
449 | retval = dev_whitelist_add(devcgroup, &wh); | |
450 | break; | |
451 | case DEVCG_DENY: | |
452 | dev_whitelist_rm(devcgroup, &wh); | |
453 | break; | |
454 | default: | |
455 | retval = -EINVAL; | |
456 | goto out2; | |
457 | } | |
458 | ||
459 | if (retval == 0) | |
460 | retval = nbytes; | |
461 | ||
462 | out2: | |
463 | cgroup_unlock(); | |
464 | out1: | |
465 | kfree(buffer); | |
466 | return retval; | |
467 | } | |
468 | ||
469 | static struct cftype dev_cgroup_files[] = { | |
470 | { | |
471 | .name = "allow", | |
08ce5f16 SH |
472 | .write = devcgroup_access_write, |
473 | .private = DEVCG_ALLOW, | |
474 | }, | |
475 | { | |
476 | .name = "deny", | |
477 | .write = devcgroup_access_write, | |
478 | .private = DEVCG_DENY, | |
479 | }, | |
29486df3 SH |
480 | { |
481 | .name = "list", | |
482 | .read_seq_string = devcgroup_seq_read, | |
483 | .private = DEVCG_LIST, | |
484 | }, | |
08ce5f16 SH |
485 | }; |
486 | ||
487 | static int devcgroup_populate(struct cgroup_subsys *ss, | |
488 | struct cgroup *cgroup) | |
489 | { | |
490 | return cgroup_add_files(cgroup, ss, dev_cgroup_files, | |
491 | ARRAY_SIZE(dev_cgroup_files)); | |
492 | } | |
493 | ||
494 | struct cgroup_subsys devices_subsys = { | |
495 | .name = "devices", | |
496 | .can_attach = devcgroup_can_attach, | |
497 | .create = devcgroup_create, | |
498 | .destroy = devcgroup_destroy, | |
499 | .populate = devcgroup_populate, | |
500 | .subsys_id = devices_subsys_id, | |
501 | }; | |
502 | ||
503 | int devcgroup_inode_permission(struct inode *inode, int mask) | |
504 | { | |
505 | struct cgroup *cgroup; | |
506 | struct dev_cgroup *dev_cgroup; | |
507 | struct dev_whitelist_item *wh; | |
508 | ||
509 | dev_t device = inode->i_rdev; | |
510 | if (!device) | |
511 | return 0; | |
512 | if (!S_ISBLK(inode->i_mode) && !S_ISCHR(inode->i_mode)) | |
513 | return 0; | |
514 | cgroup = task_cgroup(current, devices_subsys.subsys_id); | |
515 | dev_cgroup = cgroup_to_devcgroup(cgroup); | |
516 | if (!dev_cgroup) | |
517 | return 0; | |
518 | ||
519 | spin_lock(&dev_cgroup->lock); | |
520 | list_for_each_entry(wh, &dev_cgroup->whitelist, list) { | |
521 | if (wh->type & DEV_ALL) | |
522 | goto acc_check; | |
523 | if ((wh->type & DEV_BLOCK) && !S_ISBLK(inode->i_mode)) | |
524 | continue; | |
525 | if ((wh->type & DEV_CHAR) && !S_ISCHR(inode->i_mode)) | |
526 | continue; | |
527 | if (wh->major != ~0 && wh->major != imajor(inode)) | |
528 | continue; | |
529 | if (wh->minor != ~0 && wh->minor != iminor(inode)) | |
530 | continue; | |
531 | acc_check: | |
532 | if ((mask & MAY_WRITE) && !(wh->access & ACC_WRITE)) | |
533 | continue; | |
534 | if ((mask & MAY_READ) && !(wh->access & ACC_READ)) | |
535 | continue; | |
536 | spin_unlock(&dev_cgroup->lock); | |
537 | return 0; | |
538 | } | |
539 | spin_unlock(&dev_cgroup->lock); | |
540 | ||
541 | return -EPERM; | |
542 | } | |
543 | ||
544 | int devcgroup_inode_mknod(int mode, dev_t dev) | |
545 | { | |
546 | struct cgroup *cgroup; | |
547 | struct dev_cgroup *dev_cgroup; | |
548 | struct dev_whitelist_item *wh; | |
549 | ||
550 | cgroup = task_cgroup(current, devices_subsys.subsys_id); | |
551 | dev_cgroup = cgroup_to_devcgroup(cgroup); | |
552 | if (!dev_cgroup) | |
553 | return 0; | |
554 | ||
555 | spin_lock(&dev_cgroup->lock); | |
556 | list_for_each_entry(wh, &dev_cgroup->whitelist, list) { | |
557 | if (wh->type & DEV_ALL) | |
558 | goto acc_check; | |
559 | if ((wh->type & DEV_BLOCK) && !S_ISBLK(mode)) | |
560 | continue; | |
561 | if ((wh->type & DEV_CHAR) && !S_ISCHR(mode)) | |
562 | continue; | |
563 | if (wh->major != ~0 && wh->major != MAJOR(dev)) | |
564 | continue; | |
565 | if (wh->minor != ~0 && wh->minor != MINOR(dev)) | |
566 | continue; | |
567 | acc_check: | |
568 | if (!(wh->access & ACC_MKNOD)) | |
569 | continue; | |
570 | spin_unlock(&dev_cgroup->lock); | |
571 | return 0; | |
572 | } | |
573 | spin_unlock(&dev_cgroup->lock); | |
574 | return -EPERM; | |
575 | } |