Commit | Line | Data |
---|---|---|
08ce5f16 | 1 | /* |
47c59803 | 2 | * device_cgroup.c - device cgroup subsystem |
08ce5f16 SH |
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> |
5a0e3ad6 | 13 | #include <linux/slab.h> |
47c59803 | 14 | #include <linux/rcupdate.h> |
b4046f00 | 15 | #include <linux/mutex.h> |
08ce5f16 SH |
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 | ||
b4046f00 LZ |
26 | static DEFINE_MUTEX(devcgroup_mutex); |
27 | ||
08ce5f16 | 28 | /* |
db9aeca9 | 29 | * exception list locking rules: |
b4046f00 | 30 | * hold devcgroup_mutex for update/read. |
47c59803 | 31 | * hold rcu_read_lock() for read. |
08ce5f16 SH |
32 | */ |
33 | ||
db9aeca9 | 34 | struct dev_exception_item { |
08ce5f16 SH |
35 | u32 major, minor; |
36 | short type; | |
37 | short access; | |
38 | struct list_head list; | |
4efd1a1b | 39 | struct rcu_head rcu; |
08ce5f16 SH |
40 | }; |
41 | ||
42 | struct dev_cgroup { | |
43 | struct cgroup_subsys_state css; | |
db9aeca9 | 44 | struct list_head exceptions; |
5b7aa7d5 AR |
45 | enum { |
46 | DEVCG_DEFAULT_ALLOW, | |
47 | DEVCG_DEFAULT_DENY, | |
48 | } behavior; | |
08ce5f16 SH |
49 | }; |
50 | ||
b66862f7 PE |
51 | static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s) |
52 | { | |
53 | return container_of(s, struct dev_cgroup, css); | |
54 | } | |
55 | ||
08ce5f16 SH |
56 | static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup) |
57 | { | |
b66862f7 | 58 | return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id)); |
08ce5f16 SH |
59 | } |
60 | ||
f92523e3 PM |
61 | static inline struct dev_cgroup *task_devcgroup(struct task_struct *task) |
62 | { | |
63 | return css_to_devcgroup(task_subsys_state(task, devices_subsys_id)); | |
64 | } | |
65 | ||
08ce5f16 SH |
66 | struct cgroup_subsys devices_subsys; |
67 | ||
761b3ef5 LZ |
68 | static int devcgroup_can_attach(struct cgroup *new_cgrp, |
69 | struct cgroup_taskset *set) | |
08ce5f16 | 70 | { |
2f7ee569 | 71 | struct task_struct *task = cgroup_taskset_first(set); |
08ce5f16 | 72 | |
2f7ee569 TH |
73 | if (current != task && !capable(CAP_SYS_ADMIN)) |
74 | return -EPERM; | |
08ce5f16 SH |
75 | return 0; |
76 | } | |
77 | ||
78 | /* | |
b4046f00 | 79 | * called under devcgroup_mutex |
08ce5f16 | 80 | */ |
db9aeca9 | 81 | static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig) |
08ce5f16 | 82 | { |
db9aeca9 | 83 | struct dev_exception_item *ex, *tmp, *new; |
08ce5f16 | 84 | |
4b1c7840 TH |
85 | lockdep_assert_held(&devcgroup_mutex); |
86 | ||
db9aeca9 AR |
87 | list_for_each_entry(ex, orig, list) { |
88 | new = kmemdup(ex, sizeof(*ex), GFP_KERNEL); | |
08ce5f16 SH |
89 | if (!new) |
90 | goto free_and_exit; | |
08ce5f16 SH |
91 | list_add_tail(&new->list, dest); |
92 | } | |
93 | ||
94 | return 0; | |
95 | ||
96 | free_and_exit: | |
db9aeca9 AR |
97 | list_for_each_entry_safe(ex, tmp, dest, list) { |
98 | list_del(&ex->list); | |
99 | kfree(ex); | |
08ce5f16 SH |
100 | } |
101 | return -ENOMEM; | |
102 | } | |
103 | ||
08ce5f16 | 104 | /* |
b4046f00 | 105 | * called under devcgroup_mutex |
08ce5f16 | 106 | */ |
db9aeca9 AR |
107 | static int dev_exception_add(struct dev_cgroup *dev_cgroup, |
108 | struct dev_exception_item *ex) | |
08ce5f16 | 109 | { |
db9aeca9 | 110 | struct dev_exception_item *excopy, *walk; |
08ce5f16 | 111 | |
4b1c7840 TH |
112 | lockdep_assert_held(&devcgroup_mutex); |
113 | ||
db9aeca9 AR |
114 | excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL); |
115 | if (!excopy) | |
08ce5f16 SH |
116 | return -ENOMEM; |
117 | ||
db9aeca9 AR |
118 | list_for_each_entry(walk, &dev_cgroup->exceptions, list) { |
119 | if (walk->type != ex->type) | |
d1ee2971 | 120 | continue; |
db9aeca9 | 121 | if (walk->major != ex->major) |
d1ee2971 | 122 | continue; |
db9aeca9 | 123 | if (walk->minor != ex->minor) |
d1ee2971 PE |
124 | continue; |
125 | ||
db9aeca9 AR |
126 | walk->access |= ex->access; |
127 | kfree(excopy); | |
128 | excopy = NULL; | |
d1ee2971 PE |
129 | } |
130 | ||
db9aeca9 AR |
131 | if (excopy != NULL) |
132 | list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions); | |
08ce5f16 SH |
133 | return 0; |
134 | } | |
135 | ||
136 | /* | |
b4046f00 | 137 | * called under devcgroup_mutex |
08ce5f16 | 138 | */ |
db9aeca9 AR |
139 | static void dev_exception_rm(struct dev_cgroup *dev_cgroup, |
140 | struct dev_exception_item *ex) | |
08ce5f16 | 141 | { |
db9aeca9 | 142 | struct dev_exception_item *walk, *tmp; |
08ce5f16 | 143 | |
4b1c7840 TH |
144 | lockdep_assert_held(&devcgroup_mutex); |
145 | ||
db9aeca9 AR |
146 | list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) { |
147 | if (walk->type != ex->type) | |
08ce5f16 | 148 | continue; |
db9aeca9 | 149 | if (walk->major != ex->major) |
08ce5f16 | 150 | continue; |
db9aeca9 | 151 | if (walk->minor != ex->minor) |
08ce5f16 SH |
152 | continue; |
153 | ||
db9aeca9 | 154 | walk->access &= ~ex->access; |
08ce5f16 | 155 | if (!walk->access) { |
4efd1a1b | 156 | list_del_rcu(&walk->list); |
6034f7e6 | 157 | kfree_rcu(walk, rcu); |
08ce5f16 SH |
158 | } |
159 | } | |
08ce5f16 SH |
160 | } |
161 | ||
868539a3 | 162 | /** |
db9aeca9 AR |
163 | * dev_exception_clean - frees all entries of the exception list |
164 | * @dev_cgroup: dev_cgroup with the exception list to be cleaned | |
868539a3 AR |
165 | * |
166 | * called under devcgroup_mutex | |
167 | */ | |
db9aeca9 | 168 | static void dev_exception_clean(struct dev_cgroup *dev_cgroup) |
868539a3 | 169 | { |
db9aeca9 | 170 | struct dev_exception_item *ex, *tmp; |
868539a3 | 171 | |
4b1c7840 TH |
172 | lockdep_assert_held(&devcgroup_mutex); |
173 | ||
db9aeca9 | 174 | list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) { |
201e72ac TH |
175 | list_del_rcu(&ex->list); |
176 | kfree_rcu(ex, rcu); | |
868539a3 AR |
177 | } |
178 | } | |
179 | ||
08ce5f16 SH |
180 | /* |
181 | * called from kernel/cgroup.c with cgroup_lock() held. | |
182 | */ | |
92fb9748 | 183 | static struct cgroup_subsys_state *devcgroup_css_alloc(struct cgroup *cgroup) |
08ce5f16 SH |
184 | { |
185 | struct dev_cgroup *dev_cgroup, *parent_dev_cgroup; | |
186 | struct cgroup *parent_cgroup; | |
187 | int ret; | |
188 | ||
189 | dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL); | |
190 | if (!dev_cgroup) | |
191 | return ERR_PTR(-ENOMEM); | |
db9aeca9 | 192 | INIT_LIST_HEAD(&dev_cgroup->exceptions); |
08ce5f16 SH |
193 | parent_cgroup = cgroup->parent; |
194 | ||
ad676077 | 195 | if (parent_cgroup == NULL) |
5b7aa7d5 | 196 | dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW; |
ad676077 | 197 | else { |
08ce5f16 | 198 | parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup); |
b4046f00 | 199 | mutex_lock(&devcgroup_mutex); |
db9aeca9 AR |
200 | ret = dev_exceptions_copy(&dev_cgroup->exceptions, |
201 | &parent_dev_cgroup->exceptions); | |
5b7aa7d5 | 202 | dev_cgroup->behavior = parent_dev_cgroup->behavior; |
b4046f00 | 203 | mutex_unlock(&devcgroup_mutex); |
08ce5f16 SH |
204 | if (ret) { |
205 | kfree(dev_cgroup); | |
206 | return ERR_PTR(ret); | |
207 | } | |
208 | } | |
209 | ||
08ce5f16 SH |
210 | return &dev_cgroup->css; |
211 | } | |
212 | ||
92fb9748 | 213 | static void devcgroup_css_free(struct cgroup *cgroup) |
08ce5f16 SH |
214 | { |
215 | struct dev_cgroup *dev_cgroup; | |
08ce5f16 SH |
216 | |
217 | dev_cgroup = cgroup_to_devcgroup(cgroup); | |
103a197c | 218 | mutex_lock(&devcgroup_mutex); |
db9aeca9 | 219 | dev_exception_clean(dev_cgroup); |
103a197c | 220 | mutex_unlock(&devcgroup_mutex); |
08ce5f16 SH |
221 | kfree(dev_cgroup); |
222 | } | |
223 | ||
224 | #define DEVCG_ALLOW 1 | |
225 | #define DEVCG_DENY 2 | |
29486df3 SH |
226 | #define DEVCG_LIST 3 |
227 | ||
17d213f8 | 228 | #define MAJMINLEN 13 |
29486df3 | 229 | #define ACCLEN 4 |
08ce5f16 SH |
230 | |
231 | static void set_access(char *acc, short access) | |
232 | { | |
233 | int idx = 0; | |
29486df3 | 234 | memset(acc, 0, ACCLEN); |
08ce5f16 SH |
235 | if (access & ACC_READ) |
236 | acc[idx++] = 'r'; | |
237 | if (access & ACC_WRITE) | |
238 | acc[idx++] = 'w'; | |
239 | if (access & ACC_MKNOD) | |
240 | acc[idx++] = 'm'; | |
241 | } | |
242 | ||
243 | static char type_to_char(short type) | |
244 | { | |
245 | if (type == DEV_ALL) | |
246 | return 'a'; | |
247 | if (type == DEV_CHAR) | |
248 | return 'c'; | |
249 | if (type == DEV_BLOCK) | |
250 | return 'b'; | |
251 | return 'X'; | |
252 | } | |
253 | ||
29486df3 | 254 | static void set_majmin(char *str, unsigned m) |
08ce5f16 | 255 | { |
08ce5f16 | 256 | if (m == ~0) |
7759fc9d | 257 | strcpy(str, "*"); |
08ce5f16 | 258 | else |
7759fc9d | 259 | sprintf(str, "%u", m); |
08ce5f16 SH |
260 | } |
261 | ||
29486df3 SH |
262 | static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft, |
263 | struct seq_file *m) | |
08ce5f16 | 264 | { |
29486df3 | 265 | struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup); |
db9aeca9 | 266 | struct dev_exception_item *ex; |
29486df3 | 267 | char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN]; |
08ce5f16 | 268 | |
4efd1a1b | 269 | rcu_read_lock(); |
ad676077 AR |
270 | /* |
271 | * To preserve the compatibility: | |
272 | * - Only show the "all devices" when the default policy is to allow | |
273 | * - List the exceptions in case the default policy is to deny | |
274 | * This way, the file remains as a "whitelist of devices" | |
275 | */ | |
5b7aa7d5 | 276 | if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) { |
ad676077 AR |
277 | set_access(acc, ACC_MASK); |
278 | set_majmin(maj, ~0); | |
279 | set_majmin(min, ~0); | |
280 | seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL), | |
29486df3 | 281 | maj, min, acc); |
ad676077 | 282 | } else { |
db9aeca9 AR |
283 | list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) { |
284 | set_access(acc, ex->access); | |
285 | set_majmin(maj, ex->major); | |
286 | set_majmin(min, ex->minor); | |
287 | seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type), | |
ad676077 AR |
288 | maj, min, acc); |
289 | } | |
08ce5f16 | 290 | } |
4efd1a1b | 291 | rcu_read_unlock(); |
08ce5f16 | 292 | |
29486df3 | 293 | return 0; |
08ce5f16 SH |
294 | } |
295 | ||
ad676077 | 296 | /** |
db9aeca9 AR |
297 | * may_access - verifies if a new exception is part of what is allowed |
298 | * by a dev cgroup based on the default policy + | |
299 | * exceptions. This is used to make sure a child cgroup | |
300 | * won't have more privileges than its parent or to | |
301 | * verify if a certain access is allowed. | |
ad676077 | 302 | * @dev_cgroup: dev cgroup to be tested against |
db9aeca9 | 303 | * @refex: new exception |
08ce5f16 | 304 | */ |
db9aeca9 AR |
305 | static int may_access(struct dev_cgroup *dev_cgroup, |
306 | struct dev_exception_item *refex) | |
08ce5f16 | 307 | { |
db9aeca9 | 308 | struct dev_exception_item *ex; |
ad676077 | 309 | bool match = false; |
08ce5f16 | 310 | |
4b1c7840 TH |
311 | rcu_lockdep_assert(rcu_read_lock_held() || |
312 | lockdep_is_held(&devcgroup_mutex), | |
313 | "device_cgroup::may_access() called without proper synchronization"); | |
314 | ||
201e72ac | 315 | list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) { |
db9aeca9 | 316 | if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK)) |
08ce5f16 | 317 | continue; |
db9aeca9 | 318 | if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR)) |
08ce5f16 | 319 | continue; |
db9aeca9 | 320 | if (ex->major != ~0 && ex->major != refex->major) |
08ce5f16 | 321 | continue; |
db9aeca9 | 322 | if (ex->minor != ~0 && ex->minor != refex->minor) |
08ce5f16 | 323 | continue; |
db9aeca9 | 324 | if (refex->access & (~ex->access)) |
08ce5f16 | 325 | continue; |
ad676077 AR |
326 | match = true; |
327 | break; | |
08ce5f16 | 328 | } |
ad676077 AR |
329 | |
330 | /* | |
db9aeca9 | 331 | * In two cases we'll consider this new exception valid: |
ad676077 | 332 | * - the dev cgroup has its default policy to allow + exception list: |
db9aeca9 | 333 | * the new exception should *not* match any of the exceptions |
5b7aa7d5 | 334 | * (behavior == DEVCG_DEFAULT_ALLOW, !match) |
ad676077 | 335 | * - the dev cgroup has its default policy to deny + exception list: |
db9aeca9 | 336 | * the new exception *should* match the exceptions |
5b7aa7d5 | 337 | * (behavior == DEVCG_DEFAULT_DENY, match) |
ad676077 | 338 | */ |
5b7aa7d5 | 339 | if ((dev_cgroup->behavior == DEVCG_DEFAULT_DENY) == match) |
ad676077 | 340 | return 1; |
08ce5f16 SH |
341 | return 0; |
342 | } | |
343 | ||
344 | /* | |
345 | * parent_has_perm: | |
db9aeca9 | 346 | * when adding a new allow rule to a device exception list, the rule |
08ce5f16 SH |
347 | * must be allowed in the parent device |
348 | */ | |
f92523e3 | 349 | static int parent_has_perm(struct dev_cgroup *childcg, |
db9aeca9 | 350 | struct dev_exception_item *ex) |
08ce5f16 | 351 | { |
f92523e3 | 352 | struct cgroup *pcg = childcg->css.cgroup->parent; |
08ce5f16 | 353 | struct dev_cgroup *parent; |
08ce5f16 SH |
354 | |
355 | if (!pcg) | |
356 | return 1; | |
357 | parent = cgroup_to_devcgroup(pcg); | |
db9aeca9 | 358 | return may_access(parent, ex); |
08ce5f16 SH |
359 | } |
360 | ||
4cef7299 AR |
361 | /** |
362 | * may_allow_all - checks if it's possible to change the behavior to | |
363 | * allow based on parent's rules. | |
364 | * @parent: device cgroup's parent | |
365 | * returns: != 0 in case it's allowed, 0 otherwise | |
366 | */ | |
367 | static inline int may_allow_all(struct dev_cgroup *parent) | |
368 | { | |
64e10477 AR |
369 | if (!parent) |
370 | return 1; | |
4cef7299 AR |
371 | return parent->behavior == DEVCG_DEFAULT_ALLOW; |
372 | } | |
373 | ||
08ce5f16 | 374 | /* |
db9aeca9 | 375 | * Modify the exception list using allow/deny rules. |
08ce5f16 SH |
376 | * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD |
377 | * so we can give a container CAP_MKNOD to let it create devices but not | |
db9aeca9 | 378 | * modify the exception list. |
08ce5f16 SH |
379 | * It seems likely we'll want to add a CAP_CONTAINER capability to allow |
380 | * us to also grant CAP_SYS_ADMIN to containers without giving away the | |
db9aeca9 | 381 | * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN |
08ce5f16 SH |
382 | * |
383 | * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting | |
384 | * new access is only allowed if you're in the top-level cgroup, or your | |
385 | * parent cgroup has the access you're asking for. | |
386 | */ | |
f92523e3 PM |
387 | static int devcgroup_update_access(struct dev_cgroup *devcgroup, |
388 | int filetype, const char *buffer) | |
08ce5f16 | 389 | { |
f92523e3 | 390 | const char *b; |
26fd8405 AR |
391 | char temp[12]; /* 11 + 1 characters needed for a u32 */ |
392 | int count, rc; | |
db9aeca9 | 393 | struct dev_exception_item ex; |
4cef7299 | 394 | struct cgroup *p = devcgroup->css.cgroup; |
64e10477 | 395 | struct dev_cgroup *parent = NULL; |
08ce5f16 SH |
396 | |
397 | if (!capable(CAP_SYS_ADMIN)) | |
398 | return -EPERM; | |
399 | ||
64e10477 AR |
400 | if (p->parent) |
401 | parent = cgroup_to_devcgroup(p->parent); | |
402 | ||
db9aeca9 | 403 | memset(&ex, 0, sizeof(ex)); |
08ce5f16 SH |
404 | b = buffer; |
405 | ||
406 | switch (*b) { | |
407 | case 'a': | |
ad676077 AR |
408 | switch (filetype) { |
409 | case DEVCG_ALLOW: | |
4cef7299 | 410 | if (!may_allow_all(parent)) |
ad676077 | 411 | return -EPERM; |
db9aeca9 | 412 | dev_exception_clean(devcgroup); |
64e10477 AR |
413 | devcgroup->behavior = DEVCG_DEFAULT_ALLOW; |
414 | if (!parent) | |
415 | break; | |
416 | ||
4cef7299 AR |
417 | rc = dev_exceptions_copy(&devcgroup->exceptions, |
418 | &parent->exceptions); | |
419 | if (rc) | |
420 | return rc; | |
ad676077 AR |
421 | break; |
422 | case DEVCG_DENY: | |
db9aeca9 | 423 | dev_exception_clean(devcgroup); |
5b7aa7d5 | 424 | devcgroup->behavior = DEVCG_DEFAULT_DENY; |
ad676077 AR |
425 | break; |
426 | default: | |
427 | return -EINVAL; | |
428 | } | |
429 | return 0; | |
08ce5f16 | 430 | case 'b': |
db9aeca9 | 431 | ex.type = DEV_BLOCK; |
08ce5f16 SH |
432 | break; |
433 | case 'c': | |
db9aeca9 | 434 | ex.type = DEV_CHAR; |
08ce5f16 SH |
435 | break; |
436 | default: | |
f92523e3 | 437 | return -EINVAL; |
08ce5f16 SH |
438 | } |
439 | b++; | |
f92523e3 PM |
440 | if (!isspace(*b)) |
441 | return -EINVAL; | |
08ce5f16 SH |
442 | b++; |
443 | if (*b == '*') { | |
db9aeca9 | 444 | ex.major = ~0; |
08ce5f16 SH |
445 | b++; |
446 | } else if (isdigit(*b)) { | |
26fd8405 AR |
447 | memset(temp, 0, sizeof(temp)); |
448 | for (count = 0; count < sizeof(temp) - 1; count++) { | |
449 | temp[count] = *b; | |
450 | b++; | |
451 | if (!isdigit(*b)) | |
452 | break; | |
453 | } | |
454 | rc = kstrtou32(temp, 10, &ex.major); | |
455 | if (rc) | |
456 | return -EINVAL; | |
08ce5f16 | 457 | } else { |
f92523e3 | 458 | return -EINVAL; |
08ce5f16 | 459 | } |
f92523e3 PM |
460 | if (*b != ':') |
461 | return -EINVAL; | |
08ce5f16 SH |
462 | b++; |
463 | ||
464 | /* read minor */ | |
465 | if (*b == '*') { | |
db9aeca9 | 466 | ex.minor = ~0; |
08ce5f16 SH |
467 | b++; |
468 | } else if (isdigit(*b)) { | |
26fd8405 AR |
469 | memset(temp, 0, sizeof(temp)); |
470 | for (count = 0; count < sizeof(temp) - 1; count++) { | |
471 | temp[count] = *b; | |
472 | b++; | |
473 | if (!isdigit(*b)) | |
474 | break; | |
475 | } | |
476 | rc = kstrtou32(temp, 10, &ex.minor); | |
477 | if (rc) | |
478 | return -EINVAL; | |
08ce5f16 | 479 | } else { |
f92523e3 | 480 | return -EINVAL; |
08ce5f16 | 481 | } |
f92523e3 PM |
482 | if (!isspace(*b)) |
483 | return -EINVAL; | |
08ce5f16 SH |
484 | for (b++, count = 0; count < 3; count++, b++) { |
485 | switch (*b) { | |
486 | case 'r': | |
db9aeca9 | 487 | ex.access |= ACC_READ; |
08ce5f16 SH |
488 | break; |
489 | case 'w': | |
db9aeca9 | 490 | ex.access |= ACC_WRITE; |
08ce5f16 SH |
491 | break; |
492 | case 'm': | |
db9aeca9 | 493 | ex.access |= ACC_MKNOD; |
08ce5f16 SH |
494 | break; |
495 | case '\n': | |
496 | case '\0': | |
497 | count = 3; | |
498 | break; | |
499 | default: | |
f92523e3 | 500 | return -EINVAL; |
08ce5f16 SH |
501 | } |
502 | } | |
503 | ||
08ce5f16 SH |
504 | switch (filetype) { |
505 | case DEVCG_ALLOW: | |
db9aeca9 | 506 | if (!parent_has_perm(devcgroup, &ex)) |
f92523e3 | 507 | return -EPERM; |
ad676077 AR |
508 | /* |
509 | * If the default policy is to allow by default, try to remove | |
510 | * an matching exception instead. And be silent about it: we | |
511 | * don't want to break compatibility | |
512 | */ | |
5b7aa7d5 | 513 | if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) { |
db9aeca9 | 514 | dev_exception_rm(devcgroup, &ex); |
ad676077 AR |
515 | return 0; |
516 | } | |
db9aeca9 | 517 | return dev_exception_add(devcgroup, &ex); |
08ce5f16 | 518 | case DEVCG_DENY: |
ad676077 AR |
519 | /* |
520 | * If the default policy is to deny by default, try to remove | |
521 | * an matching exception instead. And be silent about it: we | |
522 | * don't want to break compatibility | |
523 | */ | |
5b7aa7d5 | 524 | if (devcgroup->behavior == DEVCG_DEFAULT_DENY) { |
db9aeca9 | 525 | dev_exception_rm(devcgroup, &ex); |
ad676077 AR |
526 | return 0; |
527 | } | |
db9aeca9 | 528 | return dev_exception_add(devcgroup, &ex); |
08ce5f16 | 529 | default: |
f92523e3 | 530 | return -EINVAL; |
08ce5f16 | 531 | } |
f92523e3 PM |
532 | return 0; |
533 | } | |
08ce5f16 | 534 | |
f92523e3 PM |
535 | static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft, |
536 | const char *buffer) | |
537 | { | |
538 | int retval; | |
b4046f00 LZ |
539 | |
540 | mutex_lock(&devcgroup_mutex); | |
f92523e3 PM |
541 | retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp), |
542 | cft->private, buffer); | |
b4046f00 | 543 | mutex_unlock(&devcgroup_mutex); |
08ce5f16 SH |
544 | return retval; |
545 | } | |
546 | ||
547 | static struct cftype dev_cgroup_files[] = { | |
548 | { | |
549 | .name = "allow", | |
f92523e3 | 550 | .write_string = devcgroup_access_write, |
08ce5f16 SH |
551 | .private = DEVCG_ALLOW, |
552 | }, | |
553 | { | |
554 | .name = "deny", | |
f92523e3 | 555 | .write_string = devcgroup_access_write, |
08ce5f16 SH |
556 | .private = DEVCG_DENY, |
557 | }, | |
29486df3 SH |
558 | { |
559 | .name = "list", | |
560 | .read_seq_string = devcgroup_seq_read, | |
561 | .private = DEVCG_LIST, | |
562 | }, | |
4baf6e33 | 563 | { } /* terminate */ |
08ce5f16 SH |
564 | }; |
565 | ||
08ce5f16 SH |
566 | struct cgroup_subsys devices_subsys = { |
567 | .name = "devices", | |
568 | .can_attach = devcgroup_can_attach, | |
92fb9748 TH |
569 | .css_alloc = devcgroup_css_alloc, |
570 | .css_free = devcgroup_css_free, | |
08ce5f16 | 571 | .subsys_id = devices_subsys_id, |
4baf6e33 | 572 | .base_cftypes = dev_cgroup_files, |
8c7f6edb TH |
573 | |
574 | /* | |
575 | * While devices cgroup has the rudimentary hierarchy support which | |
576 | * checks the parent's restriction, it doesn't properly propagates | |
577 | * config changes in ancestors to their descendents. A child | |
578 | * should only be allowed to add more restrictions to the parent's | |
579 | * configuration. Fix it and remove the following. | |
580 | */ | |
581 | .broken_hierarchy = true, | |
08ce5f16 SH |
582 | }; |
583 | ||
ad676077 AR |
584 | /** |
585 | * __devcgroup_check_permission - checks if an inode operation is permitted | |
586 | * @dev_cgroup: the dev cgroup to be tested against | |
587 | * @type: device type | |
588 | * @major: device major number | |
589 | * @minor: device minor number | |
590 | * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD | |
591 | * | |
592 | * returns 0 on success, -EPERM case the operation is not permitted | |
593 | */ | |
8c9506d1 | 594 | static int __devcgroup_check_permission(short type, u32 major, u32 minor, |
ad676077 | 595 | short access) |
08ce5f16 | 596 | { |
8c9506d1 | 597 | struct dev_cgroup *dev_cgroup; |
db9aeca9 | 598 | struct dev_exception_item ex; |
ad676077 | 599 | int rc; |
36fd71d2 | 600 | |
db9aeca9 AR |
601 | memset(&ex, 0, sizeof(ex)); |
602 | ex.type = type; | |
603 | ex.major = major; | |
604 | ex.minor = minor; | |
605 | ex.access = access; | |
36fd71d2 | 606 | |
ad676077 | 607 | rcu_read_lock(); |
8c9506d1 | 608 | dev_cgroup = task_devcgroup(current); |
db9aeca9 | 609 | rc = may_access(dev_cgroup, &ex); |
ad676077 | 610 | rcu_read_unlock(); |
cd500819 | 611 | |
ad676077 AR |
612 | if (!rc) |
613 | return -EPERM; | |
36fd71d2 | 614 | |
ad676077 AR |
615 | return 0; |
616 | } | |
08ce5f16 | 617 | |
ad676077 AR |
618 | int __devcgroup_inode_permission(struct inode *inode, int mask) |
619 | { | |
ad676077 AR |
620 | short type, access = 0; |
621 | ||
622 | if (S_ISBLK(inode->i_mode)) | |
623 | type = DEV_BLOCK; | |
624 | if (S_ISCHR(inode->i_mode)) | |
625 | type = DEV_CHAR; | |
626 | if (mask & MAY_WRITE) | |
627 | access |= ACC_WRITE; | |
628 | if (mask & MAY_READ) | |
629 | access |= ACC_READ; | |
630 | ||
8c9506d1 JS |
631 | return __devcgroup_check_permission(type, imajor(inode), iminor(inode), |
632 | access); | |
08ce5f16 SH |
633 | } |
634 | ||
635 | int devcgroup_inode_mknod(int mode, dev_t dev) | |
636 | { | |
ad676077 | 637 | short type; |
08ce5f16 | 638 | |
0b82ac37 SH |
639 | if (!S_ISBLK(mode) && !S_ISCHR(mode)) |
640 | return 0; | |
641 | ||
ad676077 AR |
642 | if (S_ISBLK(mode)) |
643 | type = DEV_BLOCK; | |
644 | else | |
645 | type = DEV_CHAR; | |
36fd71d2 | 646 | |
8c9506d1 JS |
647 | return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev), |
648 | ACC_MKNOD); | |
36fd71d2 | 649 | |
08ce5f16 | 650 | } |