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 | ||
c39a2a30 AR |
28 | enum devcg_behavior { |
29 | DEVCG_DEFAULT_NONE, | |
30 | DEVCG_DEFAULT_ALLOW, | |
31 | DEVCG_DEFAULT_DENY, | |
32 | }; | |
33 | ||
08ce5f16 | 34 | /* |
db9aeca9 | 35 | * exception list locking rules: |
b4046f00 | 36 | * hold devcgroup_mutex for update/read. |
47c59803 | 37 | * hold rcu_read_lock() for read. |
08ce5f16 SH |
38 | */ |
39 | ||
db9aeca9 | 40 | struct dev_exception_item { |
08ce5f16 SH |
41 | u32 major, minor; |
42 | short type; | |
43 | short access; | |
44 | struct list_head list; | |
4efd1a1b | 45 | struct rcu_head rcu; |
08ce5f16 SH |
46 | }; |
47 | ||
48 | struct dev_cgroup { | |
49 | struct cgroup_subsys_state css; | |
db9aeca9 | 50 | struct list_head exceptions; |
c39a2a30 | 51 | enum devcg_behavior behavior; |
08ce5f16 SH |
52 | }; |
53 | ||
b66862f7 PE |
54 | static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s) |
55 | { | |
a7c6d554 | 56 | return s ? container_of(s, struct dev_cgroup, css) : NULL; |
b66862f7 PE |
57 | } |
58 | ||
f92523e3 PM |
59 | static inline struct dev_cgroup *task_devcgroup(struct task_struct *task) |
60 | { | |
8af01f56 | 61 | return css_to_devcgroup(task_css(task, devices_subsys_id)); |
f92523e3 PM |
62 | } |
63 | ||
08ce5f16 SH |
64 | struct cgroup_subsys devices_subsys; |
65 | ||
eb95419b | 66 | static int devcgroup_can_attach(struct cgroup_subsys_state *new_css, |
761b3ef5 | 67 | struct cgroup_taskset *set) |
08ce5f16 | 68 | { |
2f7ee569 | 69 | struct task_struct *task = cgroup_taskset_first(set); |
08ce5f16 | 70 | |
2f7ee569 TH |
71 | if (current != task && !capable(CAP_SYS_ADMIN)) |
72 | return -EPERM; | |
08ce5f16 SH |
73 | return 0; |
74 | } | |
75 | ||
76 | /* | |
b4046f00 | 77 | * called under devcgroup_mutex |
08ce5f16 | 78 | */ |
db9aeca9 | 79 | static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig) |
08ce5f16 | 80 | { |
db9aeca9 | 81 | struct dev_exception_item *ex, *tmp, *new; |
08ce5f16 | 82 | |
4b1c7840 TH |
83 | lockdep_assert_held(&devcgroup_mutex); |
84 | ||
db9aeca9 AR |
85 | list_for_each_entry(ex, orig, list) { |
86 | new = kmemdup(ex, sizeof(*ex), GFP_KERNEL); | |
08ce5f16 SH |
87 | if (!new) |
88 | goto free_and_exit; | |
08ce5f16 SH |
89 | list_add_tail(&new->list, dest); |
90 | } | |
91 | ||
92 | return 0; | |
93 | ||
94 | free_and_exit: | |
db9aeca9 AR |
95 | list_for_each_entry_safe(ex, tmp, dest, list) { |
96 | list_del(&ex->list); | |
97 | kfree(ex); | |
08ce5f16 SH |
98 | } |
99 | return -ENOMEM; | |
100 | } | |
101 | ||
08ce5f16 | 102 | /* |
b4046f00 | 103 | * called under devcgroup_mutex |
08ce5f16 | 104 | */ |
db9aeca9 AR |
105 | static int dev_exception_add(struct dev_cgroup *dev_cgroup, |
106 | struct dev_exception_item *ex) | |
08ce5f16 | 107 | { |
db9aeca9 | 108 | struct dev_exception_item *excopy, *walk; |
08ce5f16 | 109 | |
4b1c7840 TH |
110 | lockdep_assert_held(&devcgroup_mutex); |
111 | ||
db9aeca9 AR |
112 | excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL); |
113 | if (!excopy) | |
08ce5f16 SH |
114 | return -ENOMEM; |
115 | ||
db9aeca9 AR |
116 | list_for_each_entry(walk, &dev_cgroup->exceptions, list) { |
117 | if (walk->type != ex->type) | |
d1ee2971 | 118 | continue; |
db9aeca9 | 119 | if (walk->major != ex->major) |
d1ee2971 | 120 | continue; |
db9aeca9 | 121 | if (walk->minor != ex->minor) |
d1ee2971 PE |
122 | continue; |
123 | ||
db9aeca9 AR |
124 | walk->access |= ex->access; |
125 | kfree(excopy); | |
126 | excopy = NULL; | |
d1ee2971 PE |
127 | } |
128 | ||
db9aeca9 AR |
129 | if (excopy != NULL) |
130 | list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions); | |
08ce5f16 SH |
131 | return 0; |
132 | } | |
133 | ||
134 | /* | |
b4046f00 | 135 | * called under devcgroup_mutex |
08ce5f16 | 136 | */ |
db9aeca9 AR |
137 | static void dev_exception_rm(struct dev_cgroup *dev_cgroup, |
138 | struct dev_exception_item *ex) | |
08ce5f16 | 139 | { |
db9aeca9 | 140 | struct dev_exception_item *walk, *tmp; |
08ce5f16 | 141 | |
4b1c7840 TH |
142 | lockdep_assert_held(&devcgroup_mutex); |
143 | ||
db9aeca9 AR |
144 | list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) { |
145 | if (walk->type != ex->type) | |
08ce5f16 | 146 | continue; |
db9aeca9 | 147 | if (walk->major != ex->major) |
08ce5f16 | 148 | continue; |
db9aeca9 | 149 | if (walk->minor != ex->minor) |
08ce5f16 SH |
150 | continue; |
151 | ||
db9aeca9 | 152 | walk->access &= ~ex->access; |
08ce5f16 | 153 | if (!walk->access) { |
4efd1a1b | 154 | list_del_rcu(&walk->list); |
6034f7e6 | 155 | kfree_rcu(walk, rcu); |
08ce5f16 SH |
156 | } |
157 | } | |
08ce5f16 SH |
158 | } |
159 | ||
53eb8c82 JS |
160 | static void __dev_exception_clean(struct dev_cgroup *dev_cgroup) |
161 | { | |
162 | struct dev_exception_item *ex, *tmp; | |
163 | ||
164 | list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) { | |
165 | list_del_rcu(&ex->list); | |
166 | kfree_rcu(ex, rcu); | |
167 | } | |
168 | } | |
169 | ||
868539a3 | 170 | /** |
db9aeca9 AR |
171 | * dev_exception_clean - frees all entries of the exception list |
172 | * @dev_cgroup: dev_cgroup with the exception list to be cleaned | |
868539a3 AR |
173 | * |
174 | * called under devcgroup_mutex | |
175 | */ | |
db9aeca9 | 176 | static void dev_exception_clean(struct dev_cgroup *dev_cgroup) |
868539a3 | 177 | { |
4b1c7840 TH |
178 | lockdep_assert_held(&devcgroup_mutex); |
179 | ||
53eb8c82 | 180 | __dev_exception_clean(dev_cgroup); |
868539a3 AR |
181 | } |
182 | ||
bd2953eb AR |
183 | static inline bool is_devcg_online(const struct dev_cgroup *devcg) |
184 | { | |
185 | return (devcg->behavior != DEVCG_DEFAULT_NONE); | |
186 | } | |
187 | ||
1909554c AR |
188 | /** |
189 | * devcgroup_online - initializes devcgroup's behavior and exceptions based on | |
190 | * parent's | |
eb95419b | 191 | * @css: css getting online |
1909554c AR |
192 | * returns 0 in case of success, error code otherwise |
193 | */ | |
eb95419b | 194 | static int devcgroup_online(struct cgroup_subsys_state *css) |
1909554c | 195 | { |
eb95419b TH |
196 | struct dev_cgroup *dev_cgroup = css_to_devcgroup(css); |
197 | struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css_parent(css)); | |
1909554c AR |
198 | int ret = 0; |
199 | ||
200 | mutex_lock(&devcgroup_mutex); | |
1909554c AR |
201 | |
202 | if (parent_dev_cgroup == NULL) | |
203 | dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW; | |
204 | else { | |
205 | ret = dev_exceptions_copy(&dev_cgroup->exceptions, | |
206 | &parent_dev_cgroup->exceptions); | |
207 | if (!ret) | |
208 | dev_cgroup->behavior = parent_dev_cgroup->behavior; | |
209 | } | |
210 | mutex_unlock(&devcgroup_mutex); | |
211 | ||
212 | return ret; | |
213 | } | |
214 | ||
eb95419b | 215 | static void devcgroup_offline(struct cgroup_subsys_state *css) |
1909554c | 216 | { |
eb95419b | 217 | struct dev_cgroup *dev_cgroup = css_to_devcgroup(css); |
1909554c AR |
218 | |
219 | mutex_lock(&devcgroup_mutex); | |
220 | dev_cgroup->behavior = DEVCG_DEFAULT_NONE; | |
221 | mutex_unlock(&devcgroup_mutex); | |
222 | } | |
223 | ||
08ce5f16 SH |
224 | /* |
225 | * called from kernel/cgroup.c with cgroup_lock() held. | |
226 | */ | |
eb95419b TH |
227 | static struct cgroup_subsys_state * |
228 | devcgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
08ce5f16 | 229 | { |
1909554c | 230 | struct dev_cgroup *dev_cgroup; |
08ce5f16 SH |
231 | |
232 | dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL); | |
233 | if (!dev_cgroup) | |
234 | return ERR_PTR(-ENOMEM); | |
db9aeca9 | 235 | INIT_LIST_HEAD(&dev_cgroup->exceptions); |
1909554c | 236 | dev_cgroup->behavior = DEVCG_DEFAULT_NONE; |
08ce5f16 | 237 | |
08ce5f16 SH |
238 | return &dev_cgroup->css; |
239 | } | |
240 | ||
eb95419b | 241 | static void devcgroup_css_free(struct cgroup_subsys_state *css) |
08ce5f16 | 242 | { |
eb95419b | 243 | struct dev_cgroup *dev_cgroup = css_to_devcgroup(css); |
08ce5f16 | 244 | |
53eb8c82 | 245 | __dev_exception_clean(dev_cgroup); |
08ce5f16 SH |
246 | kfree(dev_cgroup); |
247 | } | |
248 | ||
249 | #define DEVCG_ALLOW 1 | |
250 | #define DEVCG_DENY 2 | |
29486df3 SH |
251 | #define DEVCG_LIST 3 |
252 | ||
17d213f8 | 253 | #define MAJMINLEN 13 |
29486df3 | 254 | #define ACCLEN 4 |
08ce5f16 SH |
255 | |
256 | static void set_access(char *acc, short access) | |
257 | { | |
258 | int idx = 0; | |
29486df3 | 259 | memset(acc, 0, ACCLEN); |
08ce5f16 SH |
260 | if (access & ACC_READ) |
261 | acc[idx++] = 'r'; | |
262 | if (access & ACC_WRITE) | |
263 | acc[idx++] = 'w'; | |
264 | if (access & ACC_MKNOD) | |
265 | acc[idx++] = 'm'; | |
266 | } | |
267 | ||
268 | static char type_to_char(short type) | |
269 | { | |
270 | if (type == DEV_ALL) | |
271 | return 'a'; | |
272 | if (type == DEV_CHAR) | |
273 | return 'c'; | |
274 | if (type == DEV_BLOCK) | |
275 | return 'b'; | |
276 | return 'X'; | |
277 | } | |
278 | ||
29486df3 | 279 | static void set_majmin(char *str, unsigned m) |
08ce5f16 | 280 | { |
08ce5f16 | 281 | if (m == ~0) |
7759fc9d | 282 | strcpy(str, "*"); |
08ce5f16 | 283 | else |
7759fc9d | 284 | sprintf(str, "%u", m); |
08ce5f16 SH |
285 | } |
286 | ||
182446d0 TH |
287 | static int devcgroup_seq_read(struct cgroup_subsys_state *css, |
288 | struct cftype *cft, struct seq_file *m) | |
08ce5f16 | 289 | { |
182446d0 | 290 | struct dev_cgroup *devcgroup = css_to_devcgroup(css); |
db9aeca9 | 291 | struct dev_exception_item *ex; |
29486df3 | 292 | char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN]; |
08ce5f16 | 293 | |
4efd1a1b | 294 | rcu_read_lock(); |
ad676077 AR |
295 | /* |
296 | * To preserve the compatibility: | |
297 | * - Only show the "all devices" when the default policy is to allow | |
298 | * - List the exceptions in case the default policy is to deny | |
299 | * This way, the file remains as a "whitelist of devices" | |
300 | */ | |
5b7aa7d5 | 301 | if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) { |
ad676077 AR |
302 | set_access(acc, ACC_MASK); |
303 | set_majmin(maj, ~0); | |
304 | set_majmin(min, ~0); | |
305 | seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL), | |
29486df3 | 306 | maj, min, acc); |
ad676077 | 307 | } else { |
db9aeca9 AR |
308 | list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) { |
309 | set_access(acc, ex->access); | |
310 | set_majmin(maj, ex->major); | |
311 | set_majmin(min, ex->minor); | |
312 | seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type), | |
ad676077 AR |
313 | maj, min, acc); |
314 | } | |
08ce5f16 | 315 | } |
4efd1a1b | 316 | rcu_read_unlock(); |
08ce5f16 | 317 | |
29486df3 | 318 | return 0; |
08ce5f16 SH |
319 | } |
320 | ||
ad676077 | 321 | /** |
db9aeca9 AR |
322 | * may_access - verifies if a new exception is part of what is allowed |
323 | * by a dev cgroup based on the default policy + | |
324 | * exceptions. This is used to make sure a child cgroup | |
325 | * won't have more privileges than its parent or to | |
326 | * verify if a certain access is allowed. | |
ad676077 | 327 | * @dev_cgroup: dev cgroup to be tested against |
db9aeca9 | 328 | * @refex: new exception |
c39a2a30 | 329 | * @behavior: behavior of the exception |
08ce5f16 | 330 | */ |
26898fdf | 331 | static bool may_access(struct dev_cgroup *dev_cgroup, |
c39a2a30 AR |
332 | struct dev_exception_item *refex, |
333 | enum devcg_behavior behavior) | |
08ce5f16 | 334 | { |
db9aeca9 | 335 | struct dev_exception_item *ex; |
ad676077 | 336 | bool match = false; |
08ce5f16 | 337 | |
4b1c7840 TH |
338 | rcu_lockdep_assert(rcu_read_lock_held() || |
339 | lockdep_is_held(&devcgroup_mutex), | |
340 | "device_cgroup::may_access() called without proper synchronization"); | |
341 | ||
201e72ac | 342 | list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) { |
db9aeca9 | 343 | if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK)) |
08ce5f16 | 344 | continue; |
db9aeca9 | 345 | if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR)) |
08ce5f16 | 346 | continue; |
db9aeca9 | 347 | if (ex->major != ~0 && ex->major != refex->major) |
08ce5f16 | 348 | continue; |
db9aeca9 | 349 | if (ex->minor != ~0 && ex->minor != refex->minor) |
08ce5f16 | 350 | continue; |
db9aeca9 | 351 | if (refex->access & (~ex->access)) |
08ce5f16 | 352 | continue; |
ad676077 AR |
353 | match = true; |
354 | break; | |
08ce5f16 | 355 | } |
ad676077 | 356 | |
c39a2a30 AR |
357 | if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) { |
358 | if (behavior == DEVCG_DEFAULT_ALLOW) { | |
359 | /* the exception will deny access to certain devices */ | |
360 | return true; | |
361 | } else { | |
362 | /* the exception will allow access to certain devices */ | |
363 | if (match) | |
364 | /* | |
365 | * a new exception allowing access shouldn't | |
366 | * match an parent's exception | |
367 | */ | |
368 | return false; | |
26898fdf | 369 | return true; |
c39a2a30 | 370 | } |
26898fdf | 371 | } else { |
c39a2a30 AR |
372 | /* only behavior == DEVCG_DEFAULT_DENY allowed here */ |
373 | if (match) | |
374 | /* parent has an exception that matches the proposed */ | |
26898fdf | 375 | return true; |
c39a2a30 AR |
376 | else |
377 | return false; | |
26898fdf AR |
378 | } |
379 | return false; | |
08ce5f16 SH |
380 | } |
381 | ||
382 | /* | |
383 | * parent_has_perm: | |
db9aeca9 | 384 | * when adding a new allow rule to a device exception list, the rule |
08ce5f16 SH |
385 | * must be allowed in the parent device |
386 | */ | |
f92523e3 | 387 | static int parent_has_perm(struct dev_cgroup *childcg, |
db9aeca9 | 388 | struct dev_exception_item *ex) |
08ce5f16 | 389 | { |
63876986 | 390 | struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css)); |
08ce5f16 | 391 | |
63876986 | 392 | if (!parent) |
08ce5f16 | 393 | return 1; |
c39a2a30 | 394 | return may_access(parent, ex, childcg->behavior); |
08ce5f16 SH |
395 | } |
396 | ||
4cef7299 AR |
397 | /** |
398 | * may_allow_all - checks if it's possible to change the behavior to | |
399 | * allow based on parent's rules. | |
400 | * @parent: device cgroup's parent | |
401 | * returns: != 0 in case it's allowed, 0 otherwise | |
402 | */ | |
403 | static inline int may_allow_all(struct dev_cgroup *parent) | |
404 | { | |
64e10477 AR |
405 | if (!parent) |
406 | return 1; | |
4cef7299 AR |
407 | return parent->behavior == DEVCG_DEFAULT_ALLOW; |
408 | } | |
409 | ||
bd2953eb AR |
410 | /** |
411 | * revalidate_active_exceptions - walks through the active exception list and | |
412 | * revalidates the exceptions based on parent's | |
413 | * behavior and exceptions. The exceptions that | |
414 | * are no longer valid will be removed. | |
415 | * Called with devcgroup_mutex held. | |
416 | * @devcg: cgroup which exceptions will be checked | |
417 | * | |
418 | * This is one of the three key functions for hierarchy implementation. | |
419 | * This function is responsible for re-evaluating all the cgroup's active | |
420 | * exceptions due to a parent's exception change. | |
421 | * Refer to Documentation/cgroups/devices.txt for more details. | |
422 | */ | |
423 | static void revalidate_active_exceptions(struct dev_cgroup *devcg) | |
424 | { | |
425 | struct dev_exception_item *ex; | |
426 | struct list_head *this, *tmp; | |
427 | ||
428 | list_for_each_safe(this, tmp, &devcg->exceptions) { | |
429 | ex = container_of(this, struct dev_exception_item, list); | |
430 | if (!parent_has_perm(devcg, ex)) | |
431 | dev_exception_rm(devcg, ex); | |
432 | } | |
433 | } | |
434 | ||
bd2953eb AR |
435 | /** |
436 | * propagate_exception - propagates a new exception to the children | |
437 | * @devcg_root: device cgroup that added a new exception | |
438 | * @ex: new exception to be propagated | |
439 | * | |
440 | * returns: 0 in case of success, != 0 in case of error | |
441 | */ | |
442 | static int propagate_exception(struct dev_cgroup *devcg_root, | |
443 | struct dev_exception_item *ex) | |
444 | { | |
492eb21b | 445 | struct cgroup_subsys_state *pos; |
bd2953eb | 446 | int rc = 0; |
bd2953eb | 447 | |
d591fb56 | 448 | rcu_read_lock(); |
bd2953eb | 449 | |
492eb21b TH |
450 | css_for_each_descendant_pre(pos, &devcg_root->css) { |
451 | struct dev_cgroup *devcg = css_to_devcgroup(pos); | |
d591fb56 TH |
452 | |
453 | /* | |
454 | * Because devcgroup_mutex is held, no devcg will become | |
455 | * online or offline during the tree walk (see on/offline | |
456 | * methods), and online ones are safe to access outside RCU | |
457 | * read lock without bumping refcnt. | |
458 | */ | |
bd8815a6 | 459 | if (pos == &devcg_root->css || !is_devcg_online(devcg)) |
d591fb56 TH |
460 | continue; |
461 | ||
462 | rcu_read_unlock(); | |
bd2953eb AR |
463 | |
464 | /* | |
465 | * in case both root's behavior and devcg is allow, a new | |
466 | * restriction means adding to the exception list | |
467 | */ | |
468 | if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW && | |
469 | devcg->behavior == DEVCG_DEFAULT_ALLOW) { | |
470 | rc = dev_exception_add(devcg, ex); | |
471 | if (rc) | |
472 | break; | |
473 | } else { | |
474 | /* | |
475 | * in the other possible cases: | |
476 | * root's behavior: allow, devcg's: deny | |
477 | * root's behavior: deny, devcg's: deny | |
478 | * the exception will be removed | |
479 | */ | |
480 | dev_exception_rm(devcg, ex); | |
481 | } | |
482 | revalidate_active_exceptions(devcg); | |
483 | ||
d591fb56 | 484 | rcu_read_lock(); |
bd2953eb | 485 | } |
d591fb56 TH |
486 | |
487 | rcu_read_unlock(); | |
bd2953eb AR |
488 | return rc; |
489 | } | |
490 | ||
491 | static inline bool has_children(struct dev_cgroup *devcgroup) | |
492 | { | |
493 | struct cgroup *cgrp = devcgroup->css.cgroup; | |
494 | ||
495 | return !list_empty(&cgrp->children); | |
496 | } | |
497 | ||
08ce5f16 | 498 | /* |
db9aeca9 | 499 | * Modify the exception list using allow/deny rules. |
08ce5f16 SH |
500 | * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD |
501 | * so we can give a container CAP_MKNOD to let it create devices but not | |
db9aeca9 | 502 | * modify the exception list. |
08ce5f16 SH |
503 | * It seems likely we'll want to add a CAP_CONTAINER capability to allow |
504 | * us to also grant CAP_SYS_ADMIN to containers without giving away the | |
db9aeca9 | 505 | * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN |
08ce5f16 SH |
506 | * |
507 | * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting | |
508 | * new access is only allowed if you're in the top-level cgroup, or your | |
509 | * parent cgroup has the access you're asking for. | |
510 | */ | |
f92523e3 PM |
511 | static int devcgroup_update_access(struct dev_cgroup *devcgroup, |
512 | int filetype, const char *buffer) | |
08ce5f16 | 513 | { |
f92523e3 | 514 | const char *b; |
26fd8405 | 515 | char temp[12]; /* 11 + 1 characters needed for a u32 */ |
c39a2a30 | 516 | int count, rc = 0; |
db9aeca9 | 517 | struct dev_exception_item ex; |
63876986 | 518 | struct dev_cgroup *parent = css_to_devcgroup(css_parent(&devcgroup->css)); |
08ce5f16 SH |
519 | |
520 | if (!capable(CAP_SYS_ADMIN)) | |
521 | return -EPERM; | |
522 | ||
db9aeca9 | 523 | memset(&ex, 0, sizeof(ex)); |
08ce5f16 SH |
524 | b = buffer; |
525 | ||
526 | switch (*b) { | |
527 | case 'a': | |
ad676077 AR |
528 | switch (filetype) { |
529 | case DEVCG_ALLOW: | |
bd2953eb AR |
530 | if (has_children(devcgroup)) |
531 | return -EINVAL; | |
532 | ||
4cef7299 | 533 | if (!may_allow_all(parent)) |
ad676077 | 534 | return -EPERM; |
db9aeca9 | 535 | dev_exception_clean(devcgroup); |
64e10477 AR |
536 | devcgroup->behavior = DEVCG_DEFAULT_ALLOW; |
537 | if (!parent) | |
538 | break; | |
539 | ||
4cef7299 AR |
540 | rc = dev_exceptions_copy(&devcgroup->exceptions, |
541 | &parent->exceptions); | |
542 | if (rc) | |
543 | return rc; | |
ad676077 AR |
544 | break; |
545 | case DEVCG_DENY: | |
bd2953eb AR |
546 | if (has_children(devcgroup)) |
547 | return -EINVAL; | |
548 | ||
db9aeca9 | 549 | dev_exception_clean(devcgroup); |
5b7aa7d5 | 550 | devcgroup->behavior = DEVCG_DEFAULT_DENY; |
ad676077 AR |
551 | break; |
552 | default: | |
553 | return -EINVAL; | |
554 | } | |
555 | return 0; | |
08ce5f16 | 556 | case 'b': |
db9aeca9 | 557 | ex.type = DEV_BLOCK; |
08ce5f16 SH |
558 | break; |
559 | case 'c': | |
db9aeca9 | 560 | ex.type = DEV_CHAR; |
08ce5f16 SH |
561 | break; |
562 | default: | |
f92523e3 | 563 | return -EINVAL; |
08ce5f16 SH |
564 | } |
565 | b++; | |
f92523e3 PM |
566 | if (!isspace(*b)) |
567 | return -EINVAL; | |
08ce5f16 SH |
568 | b++; |
569 | if (*b == '*') { | |
db9aeca9 | 570 | ex.major = ~0; |
08ce5f16 SH |
571 | b++; |
572 | } else if (isdigit(*b)) { | |
26fd8405 AR |
573 | memset(temp, 0, sizeof(temp)); |
574 | for (count = 0; count < sizeof(temp) - 1; count++) { | |
575 | temp[count] = *b; | |
576 | b++; | |
577 | if (!isdigit(*b)) | |
578 | break; | |
579 | } | |
580 | rc = kstrtou32(temp, 10, &ex.major); | |
581 | if (rc) | |
582 | return -EINVAL; | |
08ce5f16 | 583 | } else { |
f92523e3 | 584 | return -EINVAL; |
08ce5f16 | 585 | } |
f92523e3 PM |
586 | if (*b != ':') |
587 | return -EINVAL; | |
08ce5f16 SH |
588 | b++; |
589 | ||
590 | /* read minor */ | |
591 | if (*b == '*') { | |
db9aeca9 | 592 | ex.minor = ~0; |
08ce5f16 SH |
593 | b++; |
594 | } else if (isdigit(*b)) { | |
26fd8405 AR |
595 | memset(temp, 0, sizeof(temp)); |
596 | for (count = 0; count < sizeof(temp) - 1; count++) { | |
597 | temp[count] = *b; | |
598 | b++; | |
599 | if (!isdigit(*b)) | |
600 | break; | |
601 | } | |
602 | rc = kstrtou32(temp, 10, &ex.minor); | |
603 | if (rc) | |
604 | return -EINVAL; | |
08ce5f16 | 605 | } else { |
f92523e3 | 606 | return -EINVAL; |
08ce5f16 | 607 | } |
f92523e3 PM |
608 | if (!isspace(*b)) |
609 | return -EINVAL; | |
08ce5f16 SH |
610 | for (b++, count = 0; count < 3; count++, b++) { |
611 | switch (*b) { | |
612 | case 'r': | |
db9aeca9 | 613 | ex.access |= ACC_READ; |
08ce5f16 SH |
614 | break; |
615 | case 'w': | |
db9aeca9 | 616 | ex.access |= ACC_WRITE; |
08ce5f16 SH |
617 | break; |
618 | case 'm': | |
db9aeca9 | 619 | ex.access |= ACC_MKNOD; |
08ce5f16 SH |
620 | break; |
621 | case '\n': | |
622 | case '\0': | |
623 | count = 3; | |
624 | break; | |
625 | default: | |
f92523e3 | 626 | return -EINVAL; |
08ce5f16 SH |
627 | } |
628 | } | |
629 | ||
08ce5f16 SH |
630 | switch (filetype) { |
631 | case DEVCG_ALLOW: | |
db9aeca9 | 632 | if (!parent_has_perm(devcgroup, &ex)) |
f92523e3 | 633 | return -EPERM; |
ad676077 AR |
634 | /* |
635 | * If the default policy is to allow by default, try to remove | |
636 | * an matching exception instead. And be silent about it: we | |
637 | * don't want to break compatibility | |
638 | */ | |
5b7aa7d5 | 639 | if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) { |
db9aeca9 | 640 | dev_exception_rm(devcgroup, &ex); |
ad676077 AR |
641 | return 0; |
642 | } | |
bd2953eb AR |
643 | rc = dev_exception_add(devcgroup, &ex); |
644 | break; | |
08ce5f16 | 645 | case DEVCG_DENY: |
ad676077 AR |
646 | /* |
647 | * If the default policy is to deny by default, try to remove | |
648 | * an matching exception instead. And be silent about it: we | |
649 | * don't want to break compatibility | |
650 | */ | |
bd2953eb | 651 | if (devcgroup->behavior == DEVCG_DEFAULT_DENY) |
db9aeca9 | 652 | dev_exception_rm(devcgroup, &ex); |
bd2953eb AR |
653 | else |
654 | rc = dev_exception_add(devcgroup, &ex); | |
655 | ||
656 | if (rc) | |
657 | break; | |
658 | /* we only propagate new restrictions */ | |
659 | rc = propagate_exception(devcgroup, &ex); | |
660 | break; | |
08ce5f16 | 661 | default: |
bd2953eb | 662 | rc = -EINVAL; |
08ce5f16 | 663 | } |
bd2953eb | 664 | return rc; |
f92523e3 | 665 | } |
08ce5f16 | 666 | |
182446d0 TH |
667 | static int devcgroup_access_write(struct cgroup_subsys_state *css, |
668 | struct cftype *cft, const char *buffer) | |
f92523e3 PM |
669 | { |
670 | int retval; | |
b4046f00 LZ |
671 | |
672 | mutex_lock(&devcgroup_mutex); | |
182446d0 | 673 | retval = devcgroup_update_access(css_to_devcgroup(css), |
f92523e3 | 674 | cft->private, buffer); |
b4046f00 | 675 | mutex_unlock(&devcgroup_mutex); |
08ce5f16 SH |
676 | return retval; |
677 | } | |
678 | ||
679 | static struct cftype dev_cgroup_files[] = { | |
680 | { | |
681 | .name = "allow", | |
f92523e3 | 682 | .write_string = devcgroup_access_write, |
08ce5f16 SH |
683 | .private = DEVCG_ALLOW, |
684 | }, | |
685 | { | |
686 | .name = "deny", | |
f92523e3 | 687 | .write_string = devcgroup_access_write, |
08ce5f16 SH |
688 | .private = DEVCG_DENY, |
689 | }, | |
29486df3 SH |
690 | { |
691 | .name = "list", | |
692 | .read_seq_string = devcgroup_seq_read, | |
693 | .private = DEVCG_LIST, | |
694 | }, | |
4baf6e33 | 695 | { } /* terminate */ |
08ce5f16 SH |
696 | }; |
697 | ||
08ce5f16 SH |
698 | struct cgroup_subsys devices_subsys = { |
699 | .name = "devices", | |
700 | .can_attach = devcgroup_can_attach, | |
92fb9748 TH |
701 | .css_alloc = devcgroup_css_alloc, |
702 | .css_free = devcgroup_css_free, | |
1909554c AR |
703 | .css_online = devcgroup_online, |
704 | .css_offline = devcgroup_offline, | |
08ce5f16 | 705 | .subsys_id = devices_subsys_id, |
4baf6e33 | 706 | .base_cftypes = dev_cgroup_files, |
08ce5f16 SH |
707 | }; |
708 | ||
ad676077 AR |
709 | /** |
710 | * __devcgroup_check_permission - checks if an inode operation is permitted | |
711 | * @dev_cgroup: the dev cgroup to be tested against | |
712 | * @type: device type | |
713 | * @major: device major number | |
714 | * @minor: device minor number | |
715 | * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD | |
716 | * | |
717 | * returns 0 on success, -EPERM case the operation is not permitted | |
718 | */ | |
8c9506d1 | 719 | static int __devcgroup_check_permission(short type, u32 major, u32 minor, |
ad676077 | 720 | short access) |
08ce5f16 | 721 | { |
8c9506d1 | 722 | struct dev_cgroup *dev_cgroup; |
db9aeca9 | 723 | struct dev_exception_item ex; |
ad676077 | 724 | int rc; |
36fd71d2 | 725 | |
db9aeca9 AR |
726 | memset(&ex, 0, sizeof(ex)); |
727 | ex.type = type; | |
728 | ex.major = major; | |
729 | ex.minor = minor; | |
730 | ex.access = access; | |
36fd71d2 | 731 | |
ad676077 | 732 | rcu_read_lock(); |
8c9506d1 | 733 | dev_cgroup = task_devcgroup(current); |
c39a2a30 | 734 | rc = may_access(dev_cgroup, &ex, dev_cgroup->behavior); |
ad676077 | 735 | rcu_read_unlock(); |
cd500819 | 736 | |
ad676077 AR |
737 | if (!rc) |
738 | return -EPERM; | |
36fd71d2 | 739 | |
ad676077 AR |
740 | return 0; |
741 | } | |
08ce5f16 | 742 | |
ad676077 AR |
743 | int __devcgroup_inode_permission(struct inode *inode, int mask) |
744 | { | |
ad676077 AR |
745 | short type, access = 0; |
746 | ||
747 | if (S_ISBLK(inode->i_mode)) | |
748 | type = DEV_BLOCK; | |
749 | if (S_ISCHR(inode->i_mode)) | |
750 | type = DEV_CHAR; | |
751 | if (mask & MAY_WRITE) | |
752 | access |= ACC_WRITE; | |
753 | if (mask & MAY_READ) | |
754 | access |= ACC_READ; | |
755 | ||
8c9506d1 JS |
756 | return __devcgroup_check_permission(type, imajor(inode), iminor(inode), |
757 | access); | |
08ce5f16 SH |
758 | } |
759 | ||
760 | int devcgroup_inode_mknod(int mode, dev_t dev) | |
761 | { | |
ad676077 | 762 | short type; |
08ce5f16 | 763 | |
0b82ac37 SH |
764 | if (!S_ISBLK(mode) && !S_ISCHR(mode)) |
765 | return 0; | |
766 | ||
ad676077 AR |
767 | if (S_ISBLK(mode)) |
768 | type = DEV_BLOCK; | |
769 | else | |
770 | type = DEV_CHAR; | |
36fd71d2 | 771 | |
8c9506d1 JS |
772 | return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev), |
773 | ACC_MKNOD); | |
36fd71d2 | 774 | |
08ce5f16 | 775 | } |