]>
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 | ||
b66862f7 PE |
52 | static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s) |
53 | { | |
54 | return container_of(s, struct dev_cgroup, css); | |
55 | } | |
56 | ||
08ce5f16 SH |
57 | static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup) |
58 | { | |
b66862f7 | 59 | return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id)); |
08ce5f16 SH |
60 | } |
61 | ||
f92523e3 PM |
62 | static inline struct dev_cgroup *task_devcgroup(struct task_struct *task) |
63 | { | |
64 | return css_to_devcgroup(task_subsys_state(task, devices_subsys_id)); | |
65 | } | |
66 | ||
08ce5f16 SH |
67 | struct cgroup_subsys devices_subsys; |
68 | ||
69 | static int devcgroup_can_attach(struct cgroup_subsys *ss, | |
70 | struct cgroup *new_cgroup, struct task_struct *task) | |
71 | { | |
72 | if (current != task && !capable(CAP_SYS_ADMIN)) | |
73 | return -EPERM; | |
74 | ||
75 | return 0; | |
76 | } | |
77 | ||
78 | /* | |
79 | * called under cgroup_lock() | |
80 | */ | |
81 | static int dev_whitelist_copy(struct list_head *dest, struct list_head *orig) | |
82 | { | |
83 | struct dev_whitelist_item *wh, *tmp, *new; | |
84 | ||
85 | list_for_each_entry(wh, orig, list) { | |
86 | new = kmalloc(sizeof(*wh), GFP_KERNEL); | |
87 | if (!new) | |
88 | goto free_and_exit; | |
89 | new->major = wh->major; | |
90 | new->minor = wh->minor; | |
91 | new->type = wh->type; | |
92 | new->access = wh->access; | |
93 | list_add_tail(&new->list, dest); | |
94 | } | |
95 | ||
96 | return 0; | |
97 | ||
98 | free_and_exit: | |
99 | list_for_each_entry_safe(wh, tmp, dest, list) { | |
100 | list_del(&wh->list); | |
101 | kfree(wh); | |
102 | } | |
103 | return -ENOMEM; | |
104 | } | |
105 | ||
106 | /* Stupid prototype - don't bother combining existing entries */ | |
107 | /* | |
108 | * called under cgroup_lock() | |
109 | * since the list is visible to other tasks, we need the spinlock also | |
110 | */ | |
111 | static int dev_whitelist_add(struct dev_cgroup *dev_cgroup, | |
112 | struct dev_whitelist_item *wh) | |
113 | { | |
d1ee2971 | 114 | struct dev_whitelist_item *whcopy, *walk; |
08ce5f16 SH |
115 | |
116 | whcopy = kmalloc(sizeof(*whcopy), GFP_KERNEL); | |
117 | if (!whcopy) | |
118 | return -ENOMEM; | |
119 | ||
120 | memcpy(whcopy, wh, sizeof(*whcopy)); | |
121 | spin_lock(&dev_cgroup->lock); | |
d1ee2971 PE |
122 | list_for_each_entry(walk, &dev_cgroup->whitelist, list) { |
123 | if (walk->type != wh->type) | |
124 | continue; | |
125 | if (walk->major != wh->major) | |
126 | continue; | |
127 | if (walk->minor != wh->minor) | |
128 | continue; | |
129 | ||
130 | walk->access |= wh->access; | |
131 | kfree(whcopy); | |
132 | whcopy = NULL; | |
133 | } | |
134 | ||
135 | if (whcopy != NULL) | |
136 | list_add_tail(&whcopy->list, &dev_cgroup->whitelist); | |
08ce5f16 SH |
137 | spin_unlock(&dev_cgroup->lock); |
138 | return 0; | |
139 | } | |
140 | ||
141 | /* | |
142 | * called under cgroup_lock() | |
143 | * since the list is visible to other tasks, we need the spinlock also | |
144 | */ | |
145 | static void dev_whitelist_rm(struct dev_cgroup *dev_cgroup, | |
146 | struct dev_whitelist_item *wh) | |
147 | { | |
148 | struct dev_whitelist_item *walk, *tmp; | |
149 | ||
150 | spin_lock(&dev_cgroup->lock); | |
151 | list_for_each_entry_safe(walk, tmp, &dev_cgroup->whitelist, list) { | |
152 | if (walk->type == DEV_ALL) | |
153 | goto remove; | |
154 | if (walk->type != wh->type) | |
155 | continue; | |
156 | if (walk->major != ~0 && walk->major != wh->major) | |
157 | continue; | |
158 | if (walk->minor != ~0 && walk->minor != wh->minor) | |
159 | continue; | |
160 | ||
161 | remove: | |
162 | walk->access &= ~wh->access; | |
163 | if (!walk->access) { | |
164 | list_del(&walk->list); | |
165 | kfree(walk); | |
166 | } | |
167 | } | |
168 | spin_unlock(&dev_cgroup->lock); | |
169 | } | |
170 | ||
171 | /* | |
172 | * called from kernel/cgroup.c with cgroup_lock() held. | |
173 | */ | |
174 | static struct cgroup_subsys_state *devcgroup_create(struct cgroup_subsys *ss, | |
175 | struct cgroup *cgroup) | |
176 | { | |
177 | struct dev_cgroup *dev_cgroup, *parent_dev_cgroup; | |
178 | struct cgroup *parent_cgroup; | |
179 | int ret; | |
180 | ||
181 | dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL); | |
182 | if (!dev_cgroup) | |
183 | return ERR_PTR(-ENOMEM); | |
184 | INIT_LIST_HEAD(&dev_cgroup->whitelist); | |
185 | parent_cgroup = cgroup->parent; | |
186 | ||
187 | if (parent_cgroup == NULL) { | |
188 | struct dev_whitelist_item *wh; | |
189 | wh = kmalloc(sizeof(*wh), GFP_KERNEL); | |
190 | if (!wh) { | |
191 | kfree(dev_cgroup); | |
192 | return ERR_PTR(-ENOMEM); | |
193 | } | |
194 | wh->minor = wh->major = ~0; | |
195 | wh->type = DEV_ALL; | |
196 | wh->access = ACC_MKNOD | ACC_READ | ACC_WRITE; | |
197 | list_add(&wh->list, &dev_cgroup->whitelist); | |
198 | } else { | |
199 | parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup); | |
200 | ret = dev_whitelist_copy(&dev_cgroup->whitelist, | |
201 | &parent_dev_cgroup->whitelist); | |
202 | if (ret) { | |
203 | kfree(dev_cgroup); | |
204 | return ERR_PTR(ret); | |
205 | } | |
206 | } | |
207 | ||
208 | spin_lock_init(&dev_cgroup->lock); | |
209 | return &dev_cgroup->css; | |
210 | } | |
211 | ||
212 | static void devcgroup_destroy(struct cgroup_subsys *ss, | |
213 | struct cgroup *cgroup) | |
214 | { | |
215 | struct dev_cgroup *dev_cgroup; | |
216 | struct dev_whitelist_item *wh, *tmp; | |
217 | ||
218 | dev_cgroup = cgroup_to_devcgroup(cgroup); | |
219 | list_for_each_entry_safe(wh, tmp, &dev_cgroup->whitelist, list) { | |
220 | list_del(&wh->list); | |
221 | kfree(wh); | |
222 | } | |
223 | kfree(dev_cgroup); | |
224 | } | |
225 | ||
226 | #define DEVCG_ALLOW 1 | |
227 | #define DEVCG_DENY 2 | |
29486df3 SH |
228 | #define DEVCG_LIST 3 |
229 | ||
17d213f8 | 230 | #define MAJMINLEN 13 |
29486df3 | 231 | #define ACCLEN 4 |
08ce5f16 SH |
232 | |
233 | static void set_access(char *acc, short access) | |
234 | { | |
235 | int idx = 0; | |
29486df3 | 236 | memset(acc, 0, ACCLEN); |
08ce5f16 SH |
237 | if (access & ACC_READ) |
238 | acc[idx++] = 'r'; | |
239 | if (access & ACC_WRITE) | |
240 | acc[idx++] = 'w'; | |
241 | if (access & ACC_MKNOD) | |
242 | acc[idx++] = 'm'; | |
243 | } | |
244 | ||
245 | static char type_to_char(short type) | |
246 | { | |
247 | if (type == DEV_ALL) | |
248 | return 'a'; | |
249 | if (type == DEV_CHAR) | |
250 | return 'c'; | |
251 | if (type == DEV_BLOCK) | |
252 | return 'b'; | |
253 | return 'X'; | |
254 | } | |
255 | ||
29486df3 | 256 | static void set_majmin(char *str, unsigned m) |
08ce5f16 | 257 | { |
29486df3 | 258 | memset(str, 0, MAJMINLEN); |
08ce5f16 SH |
259 | if (m == ~0) |
260 | sprintf(str, "*"); | |
261 | else | |
17d213f8 | 262 | snprintf(str, MAJMINLEN, "%u", m); |
08ce5f16 SH |
263 | } |
264 | ||
29486df3 SH |
265 | static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft, |
266 | struct seq_file *m) | |
08ce5f16 | 267 | { |
29486df3 | 268 | struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup); |
08ce5f16 | 269 | struct dev_whitelist_item *wh; |
29486df3 | 270 | char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN]; |
08ce5f16 SH |
271 | |
272 | spin_lock(&devcgroup->lock); | |
273 | list_for_each_entry(wh, &devcgroup->whitelist, list) { | |
274 | set_access(acc, wh->access); | |
29486df3 SH |
275 | set_majmin(maj, wh->major); |
276 | set_majmin(min, wh->minor); | |
277 | seq_printf(m, "%c %s:%s %s\n", type_to_char(wh->type), | |
278 | maj, min, acc); | |
08ce5f16 SH |
279 | } |
280 | spin_unlock(&devcgroup->lock); | |
281 | ||
29486df3 | 282 | return 0; |
08ce5f16 SH |
283 | } |
284 | ||
285 | /* | |
286 | * may_access_whitelist: | |
287 | * does the access granted to dev_cgroup c contain the access | |
288 | * requested in whitelist item refwh. | |
289 | * return 1 if yes, 0 if no. | |
290 | * call with c->lock held | |
291 | */ | |
292 | static int may_access_whitelist(struct dev_cgroup *c, | |
293 | struct dev_whitelist_item *refwh) | |
294 | { | |
295 | struct dev_whitelist_item *whitem; | |
296 | ||
297 | list_for_each_entry(whitem, &c->whitelist, list) { | |
298 | if (whitem->type & DEV_ALL) | |
299 | return 1; | |
300 | if ((refwh->type & DEV_BLOCK) && !(whitem->type & DEV_BLOCK)) | |
301 | continue; | |
302 | if ((refwh->type & DEV_CHAR) && !(whitem->type & DEV_CHAR)) | |
303 | continue; | |
304 | if (whitem->major != ~0 && whitem->major != refwh->major) | |
305 | continue; | |
306 | if (whitem->minor != ~0 && whitem->minor != refwh->minor) | |
307 | continue; | |
ec229e83 | 308 | if (refwh->access & (~whitem->access)) |
08ce5f16 SH |
309 | continue; |
310 | return 1; | |
311 | } | |
312 | return 0; | |
313 | } | |
314 | ||
315 | /* | |
316 | * parent_has_perm: | |
317 | * when adding a new allow rule to a device whitelist, the rule | |
318 | * must be allowed in the parent device | |
319 | */ | |
f92523e3 | 320 | static int parent_has_perm(struct dev_cgroup *childcg, |
08ce5f16 SH |
321 | struct dev_whitelist_item *wh) |
322 | { | |
f92523e3 | 323 | struct cgroup *pcg = childcg->css.cgroup->parent; |
08ce5f16 SH |
324 | struct dev_cgroup *parent; |
325 | int ret; | |
326 | ||
327 | if (!pcg) | |
328 | return 1; | |
329 | parent = cgroup_to_devcgroup(pcg); | |
330 | spin_lock(&parent->lock); | |
331 | ret = may_access_whitelist(parent, wh); | |
332 | spin_unlock(&parent->lock); | |
333 | return ret; | |
334 | } | |
335 | ||
336 | /* | |
337 | * Modify the whitelist using allow/deny rules. | |
338 | * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD | |
339 | * so we can give a container CAP_MKNOD to let it create devices but not | |
340 | * modify the whitelist. | |
341 | * It seems likely we'll want to add a CAP_CONTAINER capability to allow | |
342 | * us to also grant CAP_SYS_ADMIN to containers without giving away the | |
343 | * device whitelist controls, but for now we'll stick with CAP_SYS_ADMIN | |
344 | * | |
345 | * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting | |
346 | * new access is only allowed if you're in the top-level cgroup, or your | |
347 | * parent cgroup has the access you're asking for. | |
348 | */ | |
f92523e3 PM |
349 | static int devcgroup_update_access(struct dev_cgroup *devcgroup, |
350 | int filetype, const char *buffer) | |
08ce5f16 | 351 | { |
f92523e3 PM |
352 | struct dev_cgroup *cur_devcgroup; |
353 | const char *b; | |
08ce5f16 SH |
354 | int retval = 0, count; |
355 | struct dev_whitelist_item wh; | |
356 | ||
357 | if (!capable(CAP_SYS_ADMIN)) | |
358 | return -EPERM; | |
359 | ||
f92523e3 | 360 | cur_devcgroup = task_devcgroup(current); |
08ce5f16 SH |
361 | |
362 | memset(&wh, 0, sizeof(wh)); | |
363 | b = buffer; | |
364 | ||
365 | switch (*b) { | |
366 | case 'a': | |
367 | wh.type = DEV_ALL; | |
368 | wh.access = ACC_MASK; | |
d823f6bf LZ |
369 | wh.major = ~0; |
370 | wh.minor = ~0; | |
08ce5f16 SH |
371 | goto handle; |
372 | case 'b': | |
373 | wh.type = DEV_BLOCK; | |
374 | break; | |
375 | case 'c': | |
376 | wh.type = DEV_CHAR; | |
377 | break; | |
378 | default: | |
f92523e3 | 379 | return -EINVAL; |
08ce5f16 SH |
380 | } |
381 | b++; | |
f92523e3 PM |
382 | if (!isspace(*b)) |
383 | return -EINVAL; | |
08ce5f16 SH |
384 | b++; |
385 | if (*b == '*') { | |
386 | wh.major = ~0; | |
387 | b++; | |
388 | } else if (isdigit(*b)) { | |
389 | wh.major = 0; | |
390 | while (isdigit(*b)) { | |
391 | wh.major = wh.major*10+(*b-'0'); | |
392 | b++; | |
393 | } | |
394 | } else { | |
f92523e3 | 395 | return -EINVAL; |
08ce5f16 | 396 | } |
f92523e3 PM |
397 | if (*b != ':') |
398 | return -EINVAL; | |
08ce5f16 SH |
399 | b++; |
400 | ||
401 | /* read minor */ | |
402 | if (*b == '*') { | |
403 | wh.minor = ~0; | |
404 | b++; | |
405 | } else if (isdigit(*b)) { | |
406 | wh.minor = 0; | |
407 | while (isdigit(*b)) { | |
408 | wh.minor = wh.minor*10+(*b-'0'); | |
409 | b++; | |
410 | } | |
411 | } else { | |
f92523e3 | 412 | return -EINVAL; |
08ce5f16 | 413 | } |
f92523e3 PM |
414 | if (!isspace(*b)) |
415 | return -EINVAL; | |
08ce5f16 SH |
416 | for (b++, count = 0; count < 3; count++, b++) { |
417 | switch (*b) { | |
418 | case 'r': | |
419 | wh.access |= ACC_READ; | |
420 | break; | |
421 | case 'w': | |
422 | wh.access |= ACC_WRITE; | |
423 | break; | |
424 | case 'm': | |
425 | wh.access |= ACC_MKNOD; | |
426 | break; | |
427 | case '\n': | |
428 | case '\0': | |
429 | count = 3; | |
430 | break; | |
431 | default: | |
f92523e3 | 432 | return -EINVAL; |
08ce5f16 SH |
433 | } |
434 | } | |
435 | ||
436 | handle: | |
437 | retval = 0; | |
438 | switch (filetype) { | |
439 | case DEVCG_ALLOW: | |
f92523e3 PM |
440 | if (!parent_has_perm(devcgroup, &wh)) |
441 | return -EPERM; | |
442 | return dev_whitelist_add(devcgroup, &wh); | |
08ce5f16 SH |
443 | case DEVCG_DENY: |
444 | dev_whitelist_rm(devcgroup, &wh); | |
445 | break; | |
446 | default: | |
f92523e3 | 447 | return -EINVAL; |
08ce5f16 | 448 | } |
f92523e3 PM |
449 | return 0; |
450 | } | |
08ce5f16 | 451 | |
f92523e3 PM |
452 | static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft, |
453 | const char *buffer) | |
454 | { | |
455 | int retval; | |
456 | if (!cgroup_lock_live_group(cgrp)) | |
457 | return -ENODEV; | |
458 | retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp), | |
459 | cft->private, buffer); | |
08ce5f16 | 460 | cgroup_unlock(); |
08ce5f16 SH |
461 | return retval; |
462 | } | |
463 | ||
464 | static struct cftype dev_cgroup_files[] = { | |
465 | { | |
466 | .name = "allow", | |
f92523e3 | 467 | .write_string = devcgroup_access_write, |
08ce5f16 SH |
468 | .private = DEVCG_ALLOW, |
469 | }, | |
470 | { | |
471 | .name = "deny", | |
f92523e3 | 472 | .write_string = devcgroup_access_write, |
08ce5f16 SH |
473 | .private = DEVCG_DENY, |
474 | }, | |
29486df3 SH |
475 | { |
476 | .name = "list", | |
477 | .read_seq_string = devcgroup_seq_read, | |
478 | .private = DEVCG_LIST, | |
479 | }, | |
08ce5f16 SH |
480 | }; |
481 | ||
482 | static int devcgroup_populate(struct cgroup_subsys *ss, | |
483 | struct cgroup *cgroup) | |
484 | { | |
485 | return cgroup_add_files(cgroup, ss, dev_cgroup_files, | |
486 | ARRAY_SIZE(dev_cgroup_files)); | |
487 | } | |
488 | ||
489 | struct cgroup_subsys devices_subsys = { | |
490 | .name = "devices", | |
491 | .can_attach = devcgroup_can_attach, | |
492 | .create = devcgroup_create, | |
493 | .destroy = devcgroup_destroy, | |
494 | .populate = devcgroup_populate, | |
495 | .subsys_id = devices_subsys_id, | |
496 | }; | |
497 | ||
498 | int devcgroup_inode_permission(struct inode *inode, int mask) | |
499 | { | |
08ce5f16 SH |
500 | struct dev_cgroup *dev_cgroup; |
501 | struct dev_whitelist_item *wh; | |
502 | ||
503 | dev_t device = inode->i_rdev; | |
504 | if (!device) | |
505 | return 0; | |
506 | if (!S_ISBLK(inode->i_mode) && !S_ISCHR(inode->i_mode)) | |
507 | return 0; | |
cc9cb219 PE |
508 | dev_cgroup = css_to_devcgroup(task_subsys_state(current, |
509 | devices_subsys_id)); | |
08ce5f16 SH |
510 | if (!dev_cgroup) |
511 | return 0; | |
512 | ||
513 | spin_lock(&dev_cgroup->lock); | |
514 | list_for_each_entry(wh, &dev_cgroup->whitelist, list) { | |
515 | if (wh->type & DEV_ALL) | |
516 | goto acc_check; | |
517 | if ((wh->type & DEV_BLOCK) && !S_ISBLK(inode->i_mode)) | |
518 | continue; | |
519 | if ((wh->type & DEV_CHAR) && !S_ISCHR(inode->i_mode)) | |
520 | continue; | |
521 | if (wh->major != ~0 && wh->major != imajor(inode)) | |
522 | continue; | |
523 | if (wh->minor != ~0 && wh->minor != iminor(inode)) | |
524 | continue; | |
525 | acc_check: | |
526 | if ((mask & MAY_WRITE) && !(wh->access & ACC_WRITE)) | |
527 | continue; | |
528 | if ((mask & MAY_READ) && !(wh->access & ACC_READ)) | |
529 | continue; | |
530 | spin_unlock(&dev_cgroup->lock); | |
531 | return 0; | |
532 | } | |
533 | spin_unlock(&dev_cgroup->lock); | |
534 | ||
535 | return -EPERM; | |
536 | } | |
537 | ||
538 | int devcgroup_inode_mknod(int mode, dev_t dev) | |
539 | { | |
08ce5f16 SH |
540 | struct dev_cgroup *dev_cgroup; |
541 | struct dev_whitelist_item *wh; | |
542 | ||
cc9cb219 PE |
543 | dev_cgroup = css_to_devcgroup(task_subsys_state(current, |
544 | devices_subsys_id)); | |
08ce5f16 SH |
545 | if (!dev_cgroup) |
546 | return 0; | |
547 | ||
548 | spin_lock(&dev_cgroup->lock); | |
549 | list_for_each_entry(wh, &dev_cgroup->whitelist, list) { | |
550 | if (wh->type & DEV_ALL) | |
551 | goto acc_check; | |
552 | if ((wh->type & DEV_BLOCK) && !S_ISBLK(mode)) | |
553 | continue; | |
554 | if ((wh->type & DEV_CHAR) && !S_ISCHR(mode)) | |
555 | continue; | |
556 | if (wh->major != ~0 && wh->major != MAJOR(dev)) | |
557 | continue; | |
558 | if (wh->minor != ~0 && wh->minor != MINOR(dev)) | |
559 | continue; | |
560 | acc_check: | |
561 | if (!(wh->access & ACC_MKNOD)) | |
562 | continue; | |
563 | spin_unlock(&dev_cgroup->lock); | |
564 | return 0; | |
565 | } | |
566 | spin_unlock(&dev_cgroup->lock); | |
567 | return -EPERM; | |
568 | } |