]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
08ce5f16 | 2 | /* |
47c59803 | 3 | * device_cgroup.c - device cgroup subsystem |
08ce5f16 SH |
4 | * |
5 | * Copyright 2007 IBM Corp | |
6 | */ | |
7 | ||
8 | #include <linux/device_cgroup.h> | |
9 | #include <linux/cgroup.h> | |
10 | #include <linux/ctype.h> | |
11 | #include <linux/list.h> | |
12 | #include <linux/uaccess.h> | |
29486df3 | 13 | #include <linux/seq_file.h> |
5a0e3ad6 | 14 | #include <linux/slab.h> |
47c59803 | 15 | #include <linux/rcupdate.h> |
b4046f00 | 16 | #include <linux/mutex.h> |
08ce5f16 | 17 | |
eec8fd02 OU |
18 | #ifdef CONFIG_CGROUP_DEVICE |
19 | ||
b4046f00 LZ |
20 | static DEFINE_MUTEX(devcgroup_mutex); |
21 | ||
c39a2a30 AR |
22 | enum devcg_behavior { |
23 | DEVCG_DEFAULT_NONE, | |
24 | DEVCG_DEFAULT_ALLOW, | |
25 | DEVCG_DEFAULT_DENY, | |
26 | }; | |
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; |
c39a2a30 | 45 | enum devcg_behavior behavior; |
08ce5f16 SH |
46 | }; |
47 | ||
b66862f7 PE |
48 | static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s) |
49 | { | |
a7c6d554 | 50 | return s ? container_of(s, struct dev_cgroup, css) : NULL; |
b66862f7 PE |
51 | } |
52 | ||
f92523e3 PM |
53 | static inline struct dev_cgroup *task_devcgroup(struct task_struct *task) |
54 | { | |
073219e9 | 55 | return css_to_devcgroup(task_css(task, devices_cgrp_id)); |
f92523e3 PM |
56 | } |
57 | ||
08ce5f16 | 58 | /* |
b4046f00 | 59 | * called under devcgroup_mutex |
08ce5f16 | 60 | */ |
db9aeca9 | 61 | static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig) |
08ce5f16 | 62 | { |
db9aeca9 | 63 | struct dev_exception_item *ex, *tmp, *new; |
08ce5f16 | 64 | |
4b1c7840 TH |
65 | lockdep_assert_held(&devcgroup_mutex); |
66 | ||
db9aeca9 AR |
67 | list_for_each_entry(ex, orig, list) { |
68 | new = kmemdup(ex, sizeof(*ex), GFP_KERNEL); | |
08ce5f16 SH |
69 | if (!new) |
70 | goto free_and_exit; | |
08ce5f16 SH |
71 | list_add_tail(&new->list, dest); |
72 | } | |
73 | ||
74 | return 0; | |
75 | ||
76 | free_and_exit: | |
db9aeca9 AR |
77 | list_for_each_entry_safe(ex, tmp, dest, list) { |
78 | list_del(&ex->list); | |
79 | kfree(ex); | |
08ce5f16 SH |
80 | } |
81 | return -ENOMEM; | |
82 | } | |
83 | ||
08ce5f16 | 84 | /* |
b4046f00 | 85 | * called under devcgroup_mutex |
08ce5f16 | 86 | */ |
db9aeca9 AR |
87 | static int dev_exception_add(struct dev_cgroup *dev_cgroup, |
88 | struct dev_exception_item *ex) | |
08ce5f16 | 89 | { |
db9aeca9 | 90 | struct dev_exception_item *excopy, *walk; |
08ce5f16 | 91 | |
4b1c7840 TH |
92 | lockdep_assert_held(&devcgroup_mutex); |
93 | ||
db9aeca9 AR |
94 | excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL); |
95 | if (!excopy) | |
08ce5f16 SH |
96 | return -ENOMEM; |
97 | ||
db9aeca9 AR |
98 | list_for_each_entry(walk, &dev_cgroup->exceptions, list) { |
99 | if (walk->type != ex->type) | |
d1ee2971 | 100 | continue; |
db9aeca9 | 101 | if (walk->major != ex->major) |
d1ee2971 | 102 | continue; |
db9aeca9 | 103 | if (walk->minor != ex->minor) |
d1ee2971 PE |
104 | continue; |
105 | ||
db9aeca9 AR |
106 | walk->access |= ex->access; |
107 | kfree(excopy); | |
108 | excopy = NULL; | |
d1ee2971 PE |
109 | } |
110 | ||
db9aeca9 AR |
111 | if (excopy != NULL) |
112 | list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions); | |
08ce5f16 SH |
113 | return 0; |
114 | } | |
115 | ||
116 | /* | |
b4046f00 | 117 | * called under devcgroup_mutex |
08ce5f16 | 118 | */ |
db9aeca9 AR |
119 | static void dev_exception_rm(struct dev_cgroup *dev_cgroup, |
120 | struct dev_exception_item *ex) | |
08ce5f16 | 121 | { |
db9aeca9 | 122 | struct dev_exception_item *walk, *tmp; |
08ce5f16 | 123 | |
4b1c7840 TH |
124 | lockdep_assert_held(&devcgroup_mutex); |
125 | ||
db9aeca9 AR |
126 | list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) { |
127 | if (walk->type != ex->type) | |
08ce5f16 | 128 | continue; |
db9aeca9 | 129 | if (walk->major != ex->major) |
08ce5f16 | 130 | continue; |
db9aeca9 | 131 | if (walk->minor != ex->minor) |
08ce5f16 SH |
132 | continue; |
133 | ||
db9aeca9 | 134 | walk->access &= ~ex->access; |
08ce5f16 | 135 | if (!walk->access) { |
4efd1a1b | 136 | list_del_rcu(&walk->list); |
6034f7e6 | 137 | kfree_rcu(walk, rcu); |
08ce5f16 SH |
138 | } |
139 | } | |
08ce5f16 SH |
140 | } |
141 | ||
53eb8c82 JS |
142 | static void __dev_exception_clean(struct dev_cgroup *dev_cgroup) |
143 | { | |
144 | struct dev_exception_item *ex, *tmp; | |
145 | ||
146 | list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) { | |
147 | list_del_rcu(&ex->list); | |
148 | kfree_rcu(ex, rcu); | |
149 | } | |
150 | } | |
151 | ||
868539a3 | 152 | /** |
db9aeca9 AR |
153 | * dev_exception_clean - frees all entries of the exception list |
154 | * @dev_cgroup: dev_cgroup with the exception list to be cleaned | |
868539a3 AR |
155 | * |
156 | * called under devcgroup_mutex | |
157 | */ | |
db9aeca9 | 158 | static void dev_exception_clean(struct dev_cgroup *dev_cgroup) |
868539a3 | 159 | { |
4b1c7840 TH |
160 | lockdep_assert_held(&devcgroup_mutex); |
161 | ||
53eb8c82 | 162 | __dev_exception_clean(dev_cgroup); |
868539a3 AR |
163 | } |
164 | ||
bd2953eb AR |
165 | static inline bool is_devcg_online(const struct dev_cgroup *devcg) |
166 | { | |
167 | return (devcg->behavior != DEVCG_DEFAULT_NONE); | |
168 | } | |
169 | ||
1909554c AR |
170 | /** |
171 | * devcgroup_online - initializes devcgroup's behavior and exceptions based on | |
172 | * parent's | |
eb95419b | 173 | * @css: css getting online |
1909554c AR |
174 | * returns 0 in case of success, error code otherwise |
175 | */ | |
eb95419b | 176 | static int devcgroup_online(struct cgroup_subsys_state *css) |
1909554c | 177 | { |
eb95419b | 178 | struct dev_cgroup *dev_cgroup = css_to_devcgroup(css); |
5c9d535b | 179 | struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent); |
1909554c AR |
180 | int ret = 0; |
181 | ||
182 | mutex_lock(&devcgroup_mutex); | |
1909554c AR |
183 | |
184 | if (parent_dev_cgroup == NULL) | |
185 | dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW; | |
186 | else { | |
187 | ret = dev_exceptions_copy(&dev_cgroup->exceptions, | |
188 | &parent_dev_cgroup->exceptions); | |
189 | if (!ret) | |
190 | dev_cgroup->behavior = parent_dev_cgroup->behavior; | |
191 | } | |
192 | mutex_unlock(&devcgroup_mutex); | |
193 | ||
194 | return ret; | |
195 | } | |
196 | ||
eb95419b | 197 | static void devcgroup_offline(struct cgroup_subsys_state *css) |
1909554c | 198 | { |
eb95419b | 199 | struct dev_cgroup *dev_cgroup = css_to_devcgroup(css); |
1909554c AR |
200 | |
201 | mutex_lock(&devcgroup_mutex); | |
202 | dev_cgroup->behavior = DEVCG_DEFAULT_NONE; | |
203 | mutex_unlock(&devcgroup_mutex); | |
204 | } | |
205 | ||
08ce5f16 SH |
206 | /* |
207 | * called from kernel/cgroup.c with cgroup_lock() held. | |
208 | */ | |
eb95419b TH |
209 | static struct cgroup_subsys_state * |
210 | devcgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
08ce5f16 | 211 | { |
1909554c | 212 | struct dev_cgroup *dev_cgroup; |
08ce5f16 SH |
213 | |
214 | dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL); | |
215 | if (!dev_cgroup) | |
216 | return ERR_PTR(-ENOMEM); | |
db9aeca9 | 217 | INIT_LIST_HEAD(&dev_cgroup->exceptions); |
1909554c | 218 | dev_cgroup->behavior = DEVCG_DEFAULT_NONE; |
08ce5f16 | 219 | |
08ce5f16 SH |
220 | return &dev_cgroup->css; |
221 | } | |
222 | ||
eb95419b | 223 | static void devcgroup_css_free(struct cgroup_subsys_state *css) |
08ce5f16 | 224 | { |
eb95419b | 225 | struct dev_cgroup *dev_cgroup = css_to_devcgroup(css); |
08ce5f16 | 226 | |
53eb8c82 | 227 | __dev_exception_clean(dev_cgroup); |
08ce5f16 SH |
228 | kfree(dev_cgroup); |
229 | } | |
230 | ||
231 | #define DEVCG_ALLOW 1 | |
232 | #define DEVCG_DENY 2 | |
29486df3 SH |
233 | #define DEVCG_LIST 3 |
234 | ||
17d213f8 | 235 | #define MAJMINLEN 13 |
29486df3 | 236 | #define ACCLEN 4 |
08ce5f16 SH |
237 | |
238 | static void set_access(char *acc, short access) | |
239 | { | |
240 | int idx = 0; | |
29486df3 | 241 | memset(acc, 0, ACCLEN); |
67e306fd | 242 | if (access & DEVCG_ACC_READ) |
08ce5f16 | 243 | acc[idx++] = 'r'; |
67e306fd | 244 | if (access & DEVCG_ACC_WRITE) |
08ce5f16 | 245 | acc[idx++] = 'w'; |
67e306fd | 246 | if (access & DEVCG_ACC_MKNOD) |
08ce5f16 SH |
247 | acc[idx++] = 'm'; |
248 | } | |
249 | ||
250 | static char type_to_char(short type) | |
251 | { | |
67e306fd | 252 | if (type == DEVCG_DEV_ALL) |
08ce5f16 | 253 | return 'a'; |
67e306fd | 254 | if (type == DEVCG_DEV_CHAR) |
08ce5f16 | 255 | return 'c'; |
67e306fd | 256 | if (type == DEVCG_DEV_BLOCK) |
08ce5f16 SH |
257 | return 'b'; |
258 | return 'X'; | |
259 | } | |
260 | ||
29486df3 | 261 | static void set_majmin(char *str, unsigned m) |
08ce5f16 | 262 | { |
08ce5f16 | 263 | if (m == ~0) |
7759fc9d | 264 | strcpy(str, "*"); |
08ce5f16 | 265 | else |
7759fc9d | 266 | sprintf(str, "%u", m); |
08ce5f16 SH |
267 | } |
268 | ||
2da8ca82 | 269 | static int devcgroup_seq_show(struct seq_file *m, void *v) |
08ce5f16 | 270 | { |
2da8ca82 | 271 | struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m)); |
db9aeca9 | 272 | struct dev_exception_item *ex; |
29486df3 | 273 | char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN]; |
08ce5f16 | 274 | |
4efd1a1b | 275 | rcu_read_lock(); |
ad676077 AR |
276 | /* |
277 | * To preserve the compatibility: | |
278 | * - Only show the "all devices" when the default policy is to allow | |
279 | * - List the exceptions in case the default policy is to deny | |
280 | * This way, the file remains as a "whitelist of devices" | |
281 | */ | |
5b7aa7d5 | 282 | if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) { |
67e306fd | 283 | set_access(acc, DEVCG_ACC_MASK); |
ad676077 AR |
284 | set_majmin(maj, ~0); |
285 | set_majmin(min, ~0); | |
67e306fd | 286 | seq_printf(m, "%c %s:%s %s\n", type_to_char(DEVCG_DEV_ALL), |
29486df3 | 287 | maj, min, acc); |
ad676077 | 288 | } else { |
db9aeca9 AR |
289 | list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) { |
290 | set_access(acc, ex->access); | |
291 | set_majmin(maj, ex->major); | |
292 | set_majmin(min, ex->minor); | |
293 | seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type), | |
ad676077 AR |
294 | maj, min, acc); |
295 | } | |
08ce5f16 | 296 | } |
4efd1a1b | 297 | rcu_read_unlock(); |
08ce5f16 | 298 | |
29486df3 | 299 | return 0; |
08ce5f16 SH |
300 | } |
301 | ||
ad676077 | 302 | /** |
f5f3cf6f | 303 | * match_exception - iterates the exception list trying to find a complete match |
79d71974 | 304 | * @exceptions: list of exceptions |
67e306fd | 305 | * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR) |
79d71974 AR |
306 | * @major: device file major number, ~0 to match all |
307 | * @minor: device file minor number, ~0 to match all | |
67e306fd | 308 | * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD) |
79d71974 | 309 | * |
f5f3cf6f AR |
310 | * It is considered a complete match if an exception is found that will |
311 | * contain the entire range of provided parameters. | |
312 | * | |
313 | * Return: true in case it matches an exception completely | |
08ce5f16 | 314 | */ |
79d71974 AR |
315 | static bool match_exception(struct list_head *exceptions, short type, |
316 | u32 major, u32 minor, short access) | |
08ce5f16 | 317 | { |
db9aeca9 | 318 | struct dev_exception_item *ex; |
08ce5f16 | 319 | |
79d71974 | 320 | list_for_each_entry_rcu(ex, exceptions, list) { |
67e306fd | 321 | if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK)) |
79d71974 | 322 | continue; |
67e306fd | 323 | if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR)) |
79d71974 AR |
324 | continue; |
325 | if (ex->major != ~0 && ex->major != major) | |
326 | continue; | |
327 | if (ex->minor != ~0 && ex->minor != minor) | |
328 | continue; | |
329 | /* provided access cannot have more than the exception rule */ | |
330 | if (access & (~ex->access)) | |
331 | continue; | |
332 | return true; | |
333 | } | |
334 | return false; | |
335 | } | |
336 | ||
337 | /** | |
f5f3cf6f | 338 | * match_exception_partial - iterates the exception list trying to find a partial match |
79d71974 | 339 | * @exceptions: list of exceptions |
67e306fd | 340 | * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR) |
79d71974 AR |
341 | * @major: device file major number, ~0 to match all |
342 | * @minor: device file minor number, ~0 to match all | |
67e306fd | 343 | * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD) |
79d71974 | 344 | * |
f5f3cf6f AR |
345 | * It is considered a partial match if an exception's range is found to |
346 | * contain *any* of the devices specified by provided parameters. This is | |
347 | * used to make sure no extra access is being granted that is forbidden by | |
348 | * any of the exception list. | |
349 | * | |
350 | * Return: true in case the provided range mat matches an exception completely | |
79d71974 AR |
351 | */ |
352 | static bool match_exception_partial(struct list_head *exceptions, short type, | |
353 | u32 major, u32 minor, short access) | |
354 | { | |
355 | struct dev_exception_item *ex; | |
4b1c7840 | 356 | |
bc62d68e AG |
357 | list_for_each_entry_rcu(ex, exceptions, list, |
358 | lockdep_is_held(&devcgroup_mutex)) { | |
67e306fd | 359 | if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK)) |
08ce5f16 | 360 | continue; |
67e306fd | 361 | if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR)) |
08ce5f16 | 362 | continue; |
79d71974 AR |
363 | /* |
364 | * We must be sure that both the exception and the provided | |
365 | * range aren't masking all devices | |
366 | */ | |
367 | if (ex->major != ~0 && major != ~0 && ex->major != major) | |
08ce5f16 | 368 | continue; |
79d71974 | 369 | if (ex->minor != ~0 && minor != ~0 && ex->minor != minor) |
08ce5f16 | 370 | continue; |
79d71974 AR |
371 | /* |
372 | * In order to make sure the provided range isn't matching | |
373 | * an exception, all its access bits shouldn't match the | |
374 | * exception's access bits | |
375 | */ | |
376 | if (!(access & ex->access)) | |
08ce5f16 | 377 | continue; |
79d71974 | 378 | return true; |
08ce5f16 | 379 | } |
79d71974 AR |
380 | return false; |
381 | } | |
382 | ||
383 | /** | |
f5f3cf6f | 384 | * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions |
79d71974 AR |
385 | * @dev_cgroup: dev cgroup to be tested against |
386 | * @refex: new exception | |
387 | * @behavior: behavior of the exception's dev_cgroup | |
f5f3cf6f AR |
388 | * |
389 | * This is used to make sure a child cgroup won't have more privileges | |
390 | * than its parent | |
79d71974 AR |
391 | */ |
392 | static bool verify_new_ex(struct dev_cgroup *dev_cgroup, | |
393 | struct dev_exception_item *refex, | |
394 | enum devcg_behavior behavior) | |
395 | { | |
396 | bool match = false; | |
397 | ||
f78f5b90 | 398 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && |
dc3a04d5 | 399 | !lockdep_is_held(&devcgroup_mutex), |
f78f5b90 | 400 | "device_cgroup:verify_new_ex called without proper synchronization"); |
ad676077 | 401 | |
c39a2a30 AR |
402 | if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) { |
403 | if (behavior == DEVCG_DEFAULT_ALLOW) { | |
79d71974 AR |
404 | /* |
405 | * new exception in the child doesn't matter, only | |
406 | * adding extra restrictions | |
407 | */ | |
c39a2a30 AR |
408 | return true; |
409 | } else { | |
79d71974 AR |
410 | /* |
411 | * new exception in the child will add more devices | |
412 | * that can be acessed, so it can't match any of | |
413 | * parent's exceptions, even slightly | |
414 | */ | |
415 | match = match_exception_partial(&dev_cgroup->exceptions, | |
416 | refex->type, | |
417 | refex->major, | |
418 | refex->minor, | |
419 | refex->access); | |
420 | ||
c39a2a30 | 421 | if (match) |
c39a2a30 | 422 | return false; |
26898fdf | 423 | return true; |
c39a2a30 | 424 | } |
26898fdf | 425 | } else { |
79d71974 AR |
426 | /* |
427 | * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore | |
428 | * the new exception will add access to more devices and must | |
429 | * be contained completely in an parent's exception to be | |
430 | * allowed | |
431 | */ | |
432 | match = match_exception(&dev_cgroup->exceptions, refex->type, | |
433 | refex->major, refex->minor, | |
434 | refex->access); | |
435 | ||
c39a2a30 AR |
436 | if (match) |
437 | /* parent has an exception that matches the proposed */ | |
26898fdf | 438 | return true; |
c39a2a30 AR |
439 | else |
440 | return false; | |
26898fdf AR |
441 | } |
442 | return false; | |
08ce5f16 SH |
443 | } |
444 | ||
445 | /* | |
446 | * parent_has_perm: | |
db9aeca9 | 447 | * when adding a new allow rule to a device exception list, the rule |
08ce5f16 SH |
448 | * must be allowed in the parent device |
449 | */ | |
f92523e3 | 450 | static int parent_has_perm(struct dev_cgroup *childcg, |
db9aeca9 | 451 | struct dev_exception_item *ex) |
08ce5f16 | 452 | { |
5c9d535b | 453 | struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent); |
08ce5f16 | 454 | |
63876986 | 455 | if (!parent) |
08ce5f16 | 456 | return 1; |
79d71974 | 457 | return verify_new_ex(parent, ex, childcg->behavior); |
08ce5f16 SH |
458 | } |
459 | ||
d2c2b11c AR |
460 | /** |
461 | * parent_allows_removal - verify if it's ok to remove an exception | |
462 | * @childcg: child cgroup from where the exception will be removed | |
463 | * @ex: exception being removed | |
464 | * | |
465 | * When removing an exception in cgroups with default ALLOW policy, it must | |
466 | * be checked if removing it will give the child cgroup more access than the | |
467 | * parent. | |
468 | * | |
469 | * Return: true if it's ok to remove exception, false otherwise | |
470 | */ | |
471 | static bool parent_allows_removal(struct dev_cgroup *childcg, | |
472 | struct dev_exception_item *ex) | |
473 | { | |
5c9d535b | 474 | struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent); |
d2c2b11c AR |
475 | |
476 | if (!parent) | |
477 | return true; | |
478 | ||
479 | /* It's always allowed to remove access to devices */ | |
480 | if (childcg->behavior == DEVCG_DEFAULT_DENY) | |
481 | return true; | |
482 | ||
483 | /* | |
484 | * Make sure you're not removing part or a whole exception existing in | |
485 | * the parent cgroup | |
486 | */ | |
487 | return !match_exception_partial(&parent->exceptions, ex->type, | |
488 | ex->major, ex->minor, ex->access); | |
489 | } | |
490 | ||
4cef7299 AR |
491 | /** |
492 | * may_allow_all - checks if it's possible to change the behavior to | |
493 | * allow based on parent's rules. | |
494 | * @parent: device cgroup's parent | |
495 | * returns: != 0 in case it's allowed, 0 otherwise | |
496 | */ | |
497 | static inline int may_allow_all(struct dev_cgroup *parent) | |
498 | { | |
64e10477 AR |
499 | if (!parent) |
500 | return 1; | |
4cef7299 AR |
501 | return parent->behavior == DEVCG_DEFAULT_ALLOW; |
502 | } | |
503 | ||
bd2953eb AR |
504 | /** |
505 | * revalidate_active_exceptions - walks through the active exception list and | |
506 | * revalidates the exceptions based on parent's | |
507 | * behavior and exceptions. The exceptions that | |
508 | * are no longer valid will be removed. | |
509 | * Called with devcgroup_mutex held. | |
510 | * @devcg: cgroup which exceptions will be checked | |
511 | * | |
512 | * This is one of the three key functions for hierarchy implementation. | |
513 | * This function is responsible for re-evaluating all the cgroup's active | |
514 | * exceptions due to a parent's exception change. | |
da82c92f | 515 | * Refer to Documentation/admin-guide/cgroup-v1/devices.rst for more details. |
bd2953eb AR |
516 | */ |
517 | static void revalidate_active_exceptions(struct dev_cgroup *devcg) | |
518 | { | |
519 | struct dev_exception_item *ex; | |
520 | struct list_head *this, *tmp; | |
521 | ||
522 | list_for_each_safe(this, tmp, &devcg->exceptions) { | |
523 | ex = container_of(this, struct dev_exception_item, list); | |
524 | if (!parent_has_perm(devcg, ex)) | |
525 | dev_exception_rm(devcg, ex); | |
526 | } | |
527 | } | |
528 | ||
bd2953eb AR |
529 | /** |
530 | * propagate_exception - propagates a new exception to the children | |
531 | * @devcg_root: device cgroup that added a new exception | |
532 | * @ex: new exception to be propagated | |
533 | * | |
534 | * returns: 0 in case of success, != 0 in case of error | |
535 | */ | |
536 | static int propagate_exception(struct dev_cgroup *devcg_root, | |
537 | struct dev_exception_item *ex) | |
538 | { | |
492eb21b | 539 | struct cgroup_subsys_state *pos; |
bd2953eb | 540 | int rc = 0; |
bd2953eb | 541 | |
d591fb56 | 542 | rcu_read_lock(); |
bd2953eb | 543 | |
492eb21b TH |
544 | css_for_each_descendant_pre(pos, &devcg_root->css) { |
545 | struct dev_cgroup *devcg = css_to_devcgroup(pos); | |
d591fb56 TH |
546 | |
547 | /* | |
548 | * Because devcgroup_mutex is held, no devcg will become | |
549 | * online or offline during the tree walk (see on/offline | |
550 | * methods), and online ones are safe to access outside RCU | |
551 | * read lock without bumping refcnt. | |
552 | */ | |
bd8815a6 | 553 | if (pos == &devcg_root->css || !is_devcg_online(devcg)) |
d591fb56 TH |
554 | continue; |
555 | ||
556 | rcu_read_unlock(); | |
bd2953eb AR |
557 | |
558 | /* | |
559 | * in case both root's behavior and devcg is allow, a new | |
560 | * restriction means adding to the exception list | |
561 | */ | |
562 | if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW && | |
563 | devcg->behavior == DEVCG_DEFAULT_ALLOW) { | |
564 | rc = dev_exception_add(devcg, ex); | |
565 | if (rc) | |
0fcc4c8c | 566 | return rc; |
bd2953eb AR |
567 | } else { |
568 | /* | |
569 | * in the other possible cases: | |
570 | * root's behavior: allow, devcg's: deny | |
571 | * root's behavior: deny, devcg's: deny | |
572 | * the exception will be removed | |
573 | */ | |
574 | dev_exception_rm(devcg, ex); | |
575 | } | |
576 | revalidate_active_exceptions(devcg); | |
577 | ||
d591fb56 | 578 | rcu_read_lock(); |
bd2953eb | 579 | } |
d591fb56 TH |
580 | |
581 | rcu_read_unlock(); | |
bd2953eb AR |
582 | return rc; |
583 | } | |
584 | ||
08ce5f16 | 585 | /* |
db9aeca9 | 586 | * Modify the exception list using allow/deny rules. |
08ce5f16 SH |
587 | * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD |
588 | * so we can give a container CAP_MKNOD to let it create devices but not | |
db9aeca9 | 589 | * modify the exception list. |
08ce5f16 SH |
590 | * It seems likely we'll want to add a CAP_CONTAINER capability to allow |
591 | * us to also grant CAP_SYS_ADMIN to containers without giving away the | |
db9aeca9 | 592 | * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN |
08ce5f16 SH |
593 | * |
594 | * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting | |
595 | * new access is only allowed if you're in the top-level cgroup, or your | |
596 | * parent cgroup has the access you're asking for. | |
597 | */ | |
f92523e3 | 598 | static int devcgroup_update_access(struct dev_cgroup *devcgroup, |
4d3bb511 | 599 | int filetype, char *buffer) |
08ce5f16 | 600 | { |
f92523e3 | 601 | const char *b; |
26fd8405 | 602 | char temp[12]; /* 11 + 1 characters needed for a u32 */ |
c39a2a30 | 603 | int count, rc = 0; |
db9aeca9 | 604 | struct dev_exception_item ex; |
5c9d535b | 605 | struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent); |
08ce5f16 SH |
606 | |
607 | if (!capable(CAP_SYS_ADMIN)) | |
608 | return -EPERM; | |
609 | ||
db9aeca9 | 610 | memset(&ex, 0, sizeof(ex)); |
08ce5f16 SH |
611 | b = buffer; |
612 | ||
613 | switch (*b) { | |
614 | case 'a': | |
ad676077 AR |
615 | switch (filetype) { |
616 | case DEVCG_ALLOW: | |
7a3bb24f | 617 | if (css_has_online_children(&devcgroup->css)) |
bd2953eb AR |
618 | return -EINVAL; |
619 | ||
4cef7299 | 620 | if (!may_allow_all(parent)) |
ad676077 | 621 | return -EPERM; |
db9aeca9 | 622 | dev_exception_clean(devcgroup); |
64e10477 AR |
623 | devcgroup->behavior = DEVCG_DEFAULT_ALLOW; |
624 | if (!parent) | |
625 | break; | |
626 | ||
4cef7299 AR |
627 | rc = dev_exceptions_copy(&devcgroup->exceptions, |
628 | &parent->exceptions); | |
629 | if (rc) | |
630 | return rc; | |
ad676077 AR |
631 | break; |
632 | case DEVCG_DENY: | |
7a3bb24f | 633 | if (css_has_online_children(&devcgroup->css)) |
bd2953eb AR |
634 | return -EINVAL; |
635 | ||
db9aeca9 | 636 | dev_exception_clean(devcgroup); |
5b7aa7d5 | 637 | devcgroup->behavior = DEVCG_DEFAULT_DENY; |
ad676077 AR |
638 | break; |
639 | default: | |
640 | return -EINVAL; | |
641 | } | |
642 | return 0; | |
08ce5f16 | 643 | case 'b': |
67e306fd | 644 | ex.type = DEVCG_DEV_BLOCK; |
08ce5f16 SH |
645 | break; |
646 | case 'c': | |
67e306fd | 647 | ex.type = DEVCG_DEV_CHAR; |
08ce5f16 SH |
648 | break; |
649 | default: | |
f92523e3 | 650 | return -EINVAL; |
08ce5f16 SH |
651 | } |
652 | b++; | |
f92523e3 PM |
653 | if (!isspace(*b)) |
654 | return -EINVAL; | |
08ce5f16 SH |
655 | b++; |
656 | if (*b == '*') { | |
db9aeca9 | 657 | ex.major = ~0; |
08ce5f16 SH |
658 | b++; |
659 | } else if (isdigit(*b)) { | |
26fd8405 AR |
660 | memset(temp, 0, sizeof(temp)); |
661 | for (count = 0; count < sizeof(temp) - 1; count++) { | |
662 | temp[count] = *b; | |
663 | b++; | |
664 | if (!isdigit(*b)) | |
665 | break; | |
666 | } | |
667 | rc = kstrtou32(temp, 10, &ex.major); | |
668 | if (rc) | |
669 | return -EINVAL; | |
08ce5f16 | 670 | } else { |
f92523e3 | 671 | return -EINVAL; |
08ce5f16 | 672 | } |
f92523e3 PM |
673 | if (*b != ':') |
674 | return -EINVAL; | |
08ce5f16 SH |
675 | b++; |
676 | ||
677 | /* read minor */ | |
678 | if (*b == '*') { | |
db9aeca9 | 679 | ex.minor = ~0; |
08ce5f16 SH |
680 | b++; |
681 | } else if (isdigit(*b)) { | |
26fd8405 AR |
682 | memset(temp, 0, sizeof(temp)); |
683 | for (count = 0; count < sizeof(temp) - 1; count++) { | |
684 | temp[count] = *b; | |
685 | b++; | |
686 | if (!isdigit(*b)) | |
687 | break; | |
688 | } | |
689 | rc = kstrtou32(temp, 10, &ex.minor); | |
690 | if (rc) | |
691 | return -EINVAL; | |
08ce5f16 | 692 | } else { |
f92523e3 | 693 | return -EINVAL; |
08ce5f16 | 694 | } |
f92523e3 PM |
695 | if (!isspace(*b)) |
696 | return -EINVAL; | |
08ce5f16 SH |
697 | for (b++, count = 0; count < 3; count++, b++) { |
698 | switch (*b) { | |
699 | case 'r': | |
67e306fd | 700 | ex.access |= DEVCG_ACC_READ; |
08ce5f16 SH |
701 | break; |
702 | case 'w': | |
67e306fd | 703 | ex.access |= DEVCG_ACC_WRITE; |
08ce5f16 SH |
704 | break; |
705 | case 'm': | |
67e306fd | 706 | ex.access |= DEVCG_ACC_MKNOD; |
08ce5f16 SH |
707 | break; |
708 | case '\n': | |
709 | case '\0': | |
710 | count = 3; | |
711 | break; | |
712 | default: | |
f92523e3 | 713 | return -EINVAL; |
08ce5f16 SH |
714 | } |
715 | } | |
716 | ||
08ce5f16 SH |
717 | switch (filetype) { |
718 | case DEVCG_ALLOW: | |
ad676077 AR |
719 | /* |
720 | * If the default policy is to allow by default, try to remove | |
721 | * an matching exception instead. And be silent about it: we | |
722 | * don't want to break compatibility | |
723 | */ | |
5b7aa7d5 | 724 | if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) { |
d2c2b11c AR |
725 | /* Check if the parent allows removing it first */ |
726 | if (!parent_allows_removal(devcgroup, &ex)) | |
727 | return -EPERM; | |
db9aeca9 | 728 | dev_exception_rm(devcgroup, &ex); |
d2c2b11c | 729 | break; |
ad676077 | 730 | } |
d2c2b11c AR |
731 | |
732 | if (!parent_has_perm(devcgroup, &ex)) | |
733 | return -EPERM; | |
bd2953eb AR |
734 | rc = dev_exception_add(devcgroup, &ex); |
735 | break; | |
08ce5f16 | 736 | case DEVCG_DENY: |
ad676077 AR |
737 | /* |
738 | * If the default policy is to deny by default, try to remove | |
739 | * an matching exception instead. And be silent about it: we | |
740 | * don't want to break compatibility | |
741 | */ | |
bd2953eb | 742 | if (devcgroup->behavior == DEVCG_DEFAULT_DENY) |
db9aeca9 | 743 | dev_exception_rm(devcgroup, &ex); |
bd2953eb AR |
744 | else |
745 | rc = dev_exception_add(devcgroup, &ex); | |
746 | ||
747 | if (rc) | |
748 | break; | |
749 | /* we only propagate new restrictions */ | |
750 | rc = propagate_exception(devcgroup, &ex); | |
751 | break; | |
08ce5f16 | 752 | default: |
bd2953eb | 753 | rc = -EINVAL; |
08ce5f16 | 754 | } |
bd2953eb | 755 | return rc; |
f92523e3 | 756 | } |
08ce5f16 | 757 | |
451af504 TH |
758 | static ssize_t devcgroup_access_write(struct kernfs_open_file *of, |
759 | char *buf, size_t nbytes, loff_t off) | |
f92523e3 PM |
760 | { |
761 | int retval; | |
b4046f00 LZ |
762 | |
763 | mutex_lock(&devcgroup_mutex); | |
451af504 TH |
764 | retval = devcgroup_update_access(css_to_devcgroup(of_css(of)), |
765 | of_cft(of)->private, strstrip(buf)); | |
b4046f00 | 766 | mutex_unlock(&devcgroup_mutex); |
451af504 | 767 | return retval ?: nbytes; |
08ce5f16 SH |
768 | } |
769 | ||
770 | static struct cftype dev_cgroup_files[] = { | |
771 | { | |
772 | .name = "allow", | |
451af504 | 773 | .write = devcgroup_access_write, |
08ce5f16 SH |
774 | .private = DEVCG_ALLOW, |
775 | }, | |
776 | { | |
777 | .name = "deny", | |
451af504 | 778 | .write = devcgroup_access_write, |
08ce5f16 SH |
779 | .private = DEVCG_DENY, |
780 | }, | |
29486df3 SH |
781 | { |
782 | .name = "list", | |
2da8ca82 | 783 | .seq_show = devcgroup_seq_show, |
29486df3 SH |
784 | .private = DEVCG_LIST, |
785 | }, | |
4baf6e33 | 786 | { } /* terminate */ |
08ce5f16 SH |
787 | }; |
788 | ||
073219e9 | 789 | struct cgroup_subsys devices_cgrp_subsys = { |
92fb9748 TH |
790 | .css_alloc = devcgroup_css_alloc, |
791 | .css_free = devcgroup_css_free, | |
1909554c AR |
792 | .css_online = devcgroup_online, |
793 | .css_offline = devcgroup_offline, | |
5577964e | 794 | .legacy_cftypes = dev_cgroup_files, |
08ce5f16 SH |
795 | }; |
796 | ||
ad676077 | 797 | /** |
eec8fd02 | 798 | * devcgroup_legacy_check_permission - checks if an inode operation is permitted |
ad676077 AR |
799 | * @dev_cgroup: the dev cgroup to be tested against |
800 | * @type: device type | |
801 | * @major: device major number | |
802 | * @minor: device minor number | |
67e306fd | 803 | * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD |
ad676077 AR |
804 | * |
805 | * returns 0 on success, -EPERM case the operation is not permitted | |
806 | */ | |
eec8fd02 | 807 | static int devcgroup_legacy_check_permission(short type, u32 major, u32 minor, |
4b7d4d45 | 808 | short access) |
08ce5f16 | 809 | { |
8c9506d1 | 810 | struct dev_cgroup *dev_cgroup; |
79d71974 | 811 | bool rc; |
36fd71d2 | 812 | |
ad676077 | 813 | rcu_read_lock(); |
8c9506d1 | 814 | dev_cgroup = task_devcgroup(current); |
79d71974 AR |
815 | if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) |
816 | /* Can't match any of the exceptions, even partially */ | |
817 | rc = !match_exception_partial(&dev_cgroup->exceptions, | |
818 | type, major, minor, access); | |
819 | else | |
820 | /* Need to match completely one exception to be allowed */ | |
821 | rc = match_exception(&dev_cgroup->exceptions, type, major, | |
822 | minor, access); | |
ad676077 | 823 | rcu_read_unlock(); |
cd500819 | 824 | |
ad676077 AR |
825 | if (!rc) |
826 | return -EPERM; | |
36fd71d2 | 827 | |
ad676077 AR |
828 | return 0; |
829 | } | |
4b7d4d45 | 830 | |
eec8fd02 OU |
831 | #endif /* CONFIG_CGROUP_DEVICE */ |
832 | ||
833 | #if defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF) | |
834 | ||
4b7d4d45 HK |
835 | int devcgroup_check_permission(short type, u32 major, u32 minor, short access) |
836 | { | |
837 | int rc = BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type, major, minor, access); | |
838 | ||
839 | if (rc) | |
840 | return -EPERM; | |
841 | ||
eec8fd02 OU |
842 | #ifdef CONFIG_CGROUP_DEVICE |
843 | return devcgroup_legacy_check_permission(type, major, minor, access); | |
844 | ||
845 | #else /* CONFIG_CGROUP_DEVICE */ | |
846 | return 0; | |
847 | ||
848 | #endif /* CONFIG_CGROUP_DEVICE */ | |
4b7d4d45 HK |
849 | } |
850 | EXPORT_SYMBOL(devcgroup_check_permission); | |
eec8fd02 | 851 | #endif /* defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF) */ |