Loading...
1/*
2 * device_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>
12#include <linux/seq_file.h>
13#include <linux/slab.h>
14#include <linux/rcupdate.h>
15#include <linux/mutex.h>
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
26static DEFINE_MUTEX(devcgroup_mutex);
27
28/*
29 * whitelist locking rules:
30 * hold devcgroup_mutex for update/read.
31 * hold rcu_read_lock() for read.
32 */
33
34struct dev_whitelist_item {
35 u32 major, minor;
36 short type;
37 short access;
38 struct list_head list;
39 struct rcu_head rcu;
40};
41
42struct dev_cgroup {
43 struct cgroup_subsys_state css;
44 struct list_head whitelist;
45};
46
47static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
48{
49 return container_of(s, struct dev_cgroup, css);
50}
51
52static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup)
53{
54 return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id));
55}
56
57static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
58{
59 return css_to_devcgroup(task_subsys_state(task, devices_subsys_id));
60}
61
62struct cgroup_subsys devices_subsys;
63
64static int devcgroup_can_attach(struct cgroup *new_cgrp,
65 struct cgroup_taskset *set)
66{
67 struct task_struct *task = cgroup_taskset_first(set);
68
69 if (current != task && !capable(CAP_SYS_ADMIN))
70 return -EPERM;
71 return 0;
72}
73
74/*
75 * called under devcgroup_mutex
76 */
77static int dev_whitelist_copy(struct list_head *dest, struct list_head *orig)
78{
79 struct dev_whitelist_item *wh, *tmp, *new;
80
81 list_for_each_entry(wh, orig, list) {
82 new = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
83 if (!new)
84 goto free_and_exit;
85 list_add_tail(&new->list, dest);
86 }
87
88 return 0;
89
90free_and_exit:
91 list_for_each_entry_safe(wh, tmp, dest, list) {
92 list_del(&wh->list);
93 kfree(wh);
94 }
95 return -ENOMEM;
96}
97
98/* Stupid prototype - don't bother combining existing entries */
99/*
100 * called under devcgroup_mutex
101 */
102static int dev_whitelist_add(struct dev_cgroup *dev_cgroup,
103 struct dev_whitelist_item *wh)
104{
105 struct dev_whitelist_item *whcopy, *walk;
106
107 whcopy = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
108 if (!whcopy)
109 return -ENOMEM;
110
111 list_for_each_entry(walk, &dev_cgroup->whitelist, list) {
112 if (walk->type != wh->type)
113 continue;
114 if (walk->major != wh->major)
115 continue;
116 if (walk->minor != wh->minor)
117 continue;
118
119 walk->access |= wh->access;
120 kfree(whcopy);
121 whcopy = NULL;
122 }
123
124 if (whcopy != NULL)
125 list_add_tail_rcu(&whcopy->list, &dev_cgroup->whitelist);
126 return 0;
127}
128
129/*
130 * called under devcgroup_mutex
131 */
132static void dev_whitelist_rm(struct dev_cgroup *dev_cgroup,
133 struct dev_whitelist_item *wh)
134{
135 struct dev_whitelist_item *walk, *tmp;
136
137 list_for_each_entry_safe(walk, tmp, &dev_cgroup->whitelist, list) {
138 if (walk->type == DEV_ALL)
139 goto remove;
140 if (walk->type != wh->type)
141 continue;
142 if (walk->major != ~0 && walk->major != wh->major)
143 continue;
144 if (walk->minor != ~0 && walk->minor != wh->minor)
145 continue;
146
147remove:
148 walk->access &= ~wh->access;
149 if (!walk->access) {
150 list_del_rcu(&walk->list);
151 kfree_rcu(walk, rcu);
152 }
153 }
154}
155
156/*
157 * called from kernel/cgroup.c with cgroup_lock() held.
158 */
159static struct cgroup_subsys_state *devcgroup_create(struct cgroup *cgroup)
160{
161 struct dev_cgroup *dev_cgroup, *parent_dev_cgroup;
162 struct cgroup *parent_cgroup;
163 int ret;
164
165 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
166 if (!dev_cgroup)
167 return ERR_PTR(-ENOMEM);
168 INIT_LIST_HEAD(&dev_cgroup->whitelist);
169 parent_cgroup = cgroup->parent;
170
171 if (parent_cgroup == NULL) {
172 struct dev_whitelist_item *wh;
173 wh = kmalloc(sizeof(*wh), GFP_KERNEL);
174 if (!wh) {
175 kfree(dev_cgroup);
176 return ERR_PTR(-ENOMEM);
177 }
178 wh->minor = wh->major = ~0;
179 wh->type = DEV_ALL;
180 wh->access = ACC_MASK;
181 list_add(&wh->list, &dev_cgroup->whitelist);
182 } else {
183 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
184 mutex_lock(&devcgroup_mutex);
185 ret = dev_whitelist_copy(&dev_cgroup->whitelist,
186 &parent_dev_cgroup->whitelist);
187 mutex_unlock(&devcgroup_mutex);
188 if (ret) {
189 kfree(dev_cgroup);
190 return ERR_PTR(ret);
191 }
192 }
193
194 return &dev_cgroup->css;
195}
196
197static void devcgroup_destroy(struct cgroup *cgroup)
198{
199 struct dev_cgroup *dev_cgroup;
200 struct dev_whitelist_item *wh, *tmp;
201
202 dev_cgroup = cgroup_to_devcgroup(cgroup);
203 list_for_each_entry_safe(wh, tmp, &dev_cgroup->whitelist, list) {
204 list_del(&wh->list);
205 kfree(wh);
206 }
207 kfree(dev_cgroup);
208}
209
210#define DEVCG_ALLOW 1
211#define DEVCG_DENY 2
212#define DEVCG_LIST 3
213
214#define MAJMINLEN 13
215#define ACCLEN 4
216
217static void set_access(char *acc, short access)
218{
219 int idx = 0;
220 memset(acc, 0, ACCLEN);
221 if (access & ACC_READ)
222 acc[idx++] = 'r';
223 if (access & ACC_WRITE)
224 acc[idx++] = 'w';
225 if (access & ACC_MKNOD)
226 acc[idx++] = 'm';
227}
228
229static char type_to_char(short type)
230{
231 if (type == DEV_ALL)
232 return 'a';
233 if (type == DEV_CHAR)
234 return 'c';
235 if (type == DEV_BLOCK)
236 return 'b';
237 return 'X';
238}
239
240static void set_majmin(char *str, unsigned m)
241{
242 if (m == ~0)
243 strcpy(str, "*");
244 else
245 sprintf(str, "%u", m);
246}
247
248static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
249 struct seq_file *m)
250{
251 struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
252 struct dev_whitelist_item *wh;
253 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
254
255 rcu_read_lock();
256 list_for_each_entry_rcu(wh, &devcgroup->whitelist, list) {
257 set_access(acc, wh->access);
258 set_majmin(maj, wh->major);
259 set_majmin(min, wh->minor);
260 seq_printf(m, "%c %s:%s %s\n", type_to_char(wh->type),
261 maj, min, acc);
262 }
263 rcu_read_unlock();
264
265 return 0;
266}
267
268/*
269 * may_access_whitelist:
270 * does the access granted to dev_cgroup c contain the access
271 * requested in whitelist item refwh.
272 * return 1 if yes, 0 if no.
273 * call with devcgroup_mutex held
274 */
275static int may_access_whitelist(struct dev_cgroup *c,
276 struct dev_whitelist_item *refwh)
277{
278 struct dev_whitelist_item *whitem;
279
280 list_for_each_entry(whitem, &c->whitelist, list) {
281 if (whitem->type & DEV_ALL)
282 return 1;
283 if ((refwh->type & DEV_BLOCK) && !(whitem->type & DEV_BLOCK))
284 continue;
285 if ((refwh->type & DEV_CHAR) && !(whitem->type & DEV_CHAR))
286 continue;
287 if (whitem->major != ~0 && whitem->major != refwh->major)
288 continue;
289 if (whitem->minor != ~0 && whitem->minor != refwh->minor)
290 continue;
291 if (refwh->access & (~whitem->access))
292 continue;
293 return 1;
294 }
295 return 0;
296}
297
298/*
299 * parent_has_perm:
300 * when adding a new allow rule to a device whitelist, the rule
301 * must be allowed in the parent device
302 */
303static int parent_has_perm(struct dev_cgroup *childcg,
304 struct dev_whitelist_item *wh)
305{
306 struct cgroup *pcg = childcg->css.cgroup->parent;
307 struct dev_cgroup *parent;
308
309 if (!pcg)
310 return 1;
311 parent = cgroup_to_devcgroup(pcg);
312 return may_access_whitelist(parent, wh);
313}
314
315/*
316 * Modify the whitelist using allow/deny rules.
317 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
318 * so we can give a container CAP_MKNOD to let it create devices but not
319 * modify the whitelist.
320 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
321 * us to also grant CAP_SYS_ADMIN to containers without giving away the
322 * device whitelist controls, but for now we'll stick with CAP_SYS_ADMIN
323 *
324 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
325 * new access is only allowed if you're in the top-level cgroup, or your
326 * parent cgroup has the access you're asking for.
327 */
328static int devcgroup_update_access(struct dev_cgroup *devcgroup,
329 int filetype, const char *buffer)
330{
331 const char *b;
332 char *endp;
333 int count;
334 struct dev_whitelist_item wh;
335
336 if (!capable(CAP_SYS_ADMIN))
337 return -EPERM;
338
339 memset(&wh, 0, sizeof(wh));
340 b = buffer;
341
342 switch (*b) {
343 case 'a':
344 wh.type = DEV_ALL;
345 wh.access = ACC_MASK;
346 wh.major = ~0;
347 wh.minor = ~0;
348 goto handle;
349 case 'b':
350 wh.type = DEV_BLOCK;
351 break;
352 case 'c':
353 wh.type = DEV_CHAR;
354 break;
355 default:
356 return -EINVAL;
357 }
358 b++;
359 if (!isspace(*b))
360 return -EINVAL;
361 b++;
362 if (*b == '*') {
363 wh.major = ~0;
364 b++;
365 } else if (isdigit(*b)) {
366 wh.major = simple_strtoul(b, &endp, 10);
367 b = endp;
368 } else {
369 return -EINVAL;
370 }
371 if (*b != ':')
372 return -EINVAL;
373 b++;
374
375 /* read minor */
376 if (*b == '*') {
377 wh.minor = ~0;
378 b++;
379 } else if (isdigit(*b)) {
380 wh.minor = simple_strtoul(b, &endp, 10);
381 b = endp;
382 } else {
383 return -EINVAL;
384 }
385 if (!isspace(*b))
386 return -EINVAL;
387 for (b++, count = 0; count < 3; count++, b++) {
388 switch (*b) {
389 case 'r':
390 wh.access |= ACC_READ;
391 break;
392 case 'w':
393 wh.access |= ACC_WRITE;
394 break;
395 case 'm':
396 wh.access |= ACC_MKNOD;
397 break;
398 case '\n':
399 case '\0':
400 count = 3;
401 break;
402 default:
403 return -EINVAL;
404 }
405 }
406
407handle:
408 switch (filetype) {
409 case DEVCG_ALLOW:
410 if (!parent_has_perm(devcgroup, &wh))
411 return -EPERM;
412 return dev_whitelist_add(devcgroup, &wh);
413 case DEVCG_DENY:
414 dev_whitelist_rm(devcgroup, &wh);
415 break;
416 default:
417 return -EINVAL;
418 }
419 return 0;
420}
421
422static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
423 const char *buffer)
424{
425 int retval;
426
427 mutex_lock(&devcgroup_mutex);
428 retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
429 cft->private, buffer);
430 mutex_unlock(&devcgroup_mutex);
431 return retval;
432}
433
434static struct cftype dev_cgroup_files[] = {
435 {
436 .name = "allow",
437 .write_string = devcgroup_access_write,
438 .private = DEVCG_ALLOW,
439 },
440 {
441 .name = "deny",
442 .write_string = devcgroup_access_write,
443 .private = DEVCG_DENY,
444 },
445 {
446 .name = "list",
447 .read_seq_string = devcgroup_seq_read,
448 .private = DEVCG_LIST,
449 },
450 { } /* terminate */
451};
452
453struct cgroup_subsys devices_subsys = {
454 .name = "devices",
455 .can_attach = devcgroup_can_attach,
456 .create = devcgroup_create,
457 .destroy = devcgroup_destroy,
458 .subsys_id = devices_subsys_id,
459 .base_cftypes = dev_cgroup_files,
460};
461
462int __devcgroup_inode_permission(struct inode *inode, int mask)
463{
464 struct dev_cgroup *dev_cgroup;
465 struct dev_whitelist_item *wh;
466
467 rcu_read_lock();
468
469 dev_cgroup = task_devcgroup(current);
470
471 list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
472 if (wh->type & DEV_ALL)
473 goto found;
474 if ((wh->type & DEV_BLOCK) && !S_ISBLK(inode->i_mode))
475 continue;
476 if ((wh->type & DEV_CHAR) && !S_ISCHR(inode->i_mode))
477 continue;
478 if (wh->major != ~0 && wh->major != imajor(inode))
479 continue;
480 if (wh->minor != ~0 && wh->minor != iminor(inode))
481 continue;
482
483 if ((mask & MAY_WRITE) && !(wh->access & ACC_WRITE))
484 continue;
485 if ((mask & MAY_READ) && !(wh->access & ACC_READ))
486 continue;
487found:
488 rcu_read_unlock();
489 return 0;
490 }
491
492 rcu_read_unlock();
493
494 return -EPERM;
495}
496
497int devcgroup_inode_mknod(int mode, dev_t dev)
498{
499 struct dev_cgroup *dev_cgroup;
500 struct dev_whitelist_item *wh;
501
502 if (!S_ISBLK(mode) && !S_ISCHR(mode))
503 return 0;
504
505 rcu_read_lock();
506
507 dev_cgroup = task_devcgroup(current);
508
509 list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
510 if (wh->type & DEV_ALL)
511 goto found;
512 if ((wh->type & DEV_BLOCK) && !S_ISBLK(mode))
513 continue;
514 if ((wh->type & DEV_CHAR) && !S_ISCHR(mode))
515 continue;
516 if (wh->major != ~0 && wh->major != MAJOR(dev))
517 continue;
518 if (wh->minor != ~0 && wh->minor != MINOR(dev))
519 continue;
520
521 if (!(wh->access & ACC_MKNOD))
522 continue;
523found:
524 rcu_read_unlock();
525 return 0;
526 }
527
528 rcu_read_unlock();
529
530 return -EPERM;
531}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * device_cgroup.c - device cgroup subsystem
4 *
5 * Copyright 2007 IBM Corp
6 */
7
8#include <linux/bpf-cgroup.h>
9#include <linux/device_cgroup.h>
10#include <linux/cgroup.h>
11#include <linux/ctype.h>
12#include <linux/list.h>
13#include <linux/uaccess.h>
14#include <linux/seq_file.h>
15#include <linux/slab.h>
16#include <linux/rcupdate.h>
17#include <linux/mutex.h>
18
19#ifdef CONFIG_CGROUP_DEVICE
20
21static DEFINE_MUTEX(devcgroup_mutex);
22
23enum devcg_behavior {
24 DEVCG_DEFAULT_NONE,
25 DEVCG_DEFAULT_ALLOW,
26 DEVCG_DEFAULT_DENY,
27};
28
29/*
30 * exception list locking rules:
31 * hold devcgroup_mutex for update/read.
32 * hold rcu_read_lock() for read.
33 */
34
35struct dev_exception_item {
36 u32 major, minor;
37 short type;
38 short access;
39 struct list_head list;
40 struct rcu_head rcu;
41};
42
43struct dev_cgroup {
44 struct cgroup_subsys_state css;
45 struct list_head exceptions;
46 enum devcg_behavior behavior;
47};
48
49static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
50{
51 return s ? container_of(s, struct dev_cgroup, css) : NULL;
52}
53
54static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
55{
56 return css_to_devcgroup(task_css(task, devices_cgrp_id));
57}
58
59/*
60 * called under devcgroup_mutex
61 */
62static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
63{
64 struct dev_exception_item *ex, *tmp, *new;
65
66 lockdep_assert_held(&devcgroup_mutex);
67
68 list_for_each_entry(ex, orig, list) {
69 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
70 if (!new)
71 goto free_and_exit;
72 list_add_tail(&new->list, dest);
73 }
74
75 return 0;
76
77free_and_exit:
78 list_for_each_entry_safe(ex, tmp, dest, list) {
79 list_del(&ex->list);
80 kfree(ex);
81 }
82 return -ENOMEM;
83}
84
85static void dev_exceptions_move(struct list_head *dest, struct list_head *orig)
86{
87 struct dev_exception_item *ex, *tmp;
88
89 lockdep_assert_held(&devcgroup_mutex);
90
91 list_for_each_entry_safe(ex, tmp, orig, list) {
92 list_move_tail(&ex->list, dest);
93 }
94}
95
96/*
97 * called under devcgroup_mutex
98 */
99static int dev_exception_add(struct dev_cgroup *dev_cgroup,
100 struct dev_exception_item *ex)
101{
102 struct dev_exception_item *excopy, *walk;
103
104 lockdep_assert_held(&devcgroup_mutex);
105
106 excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
107 if (!excopy)
108 return -ENOMEM;
109
110 list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
111 if (walk->type != ex->type)
112 continue;
113 if (walk->major != ex->major)
114 continue;
115 if (walk->minor != ex->minor)
116 continue;
117
118 walk->access |= ex->access;
119 kfree(excopy);
120 excopy = NULL;
121 }
122
123 if (excopy != NULL)
124 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
125 return 0;
126}
127
128/*
129 * called under devcgroup_mutex
130 */
131static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
132 struct dev_exception_item *ex)
133{
134 struct dev_exception_item *walk, *tmp;
135
136 lockdep_assert_held(&devcgroup_mutex);
137
138 list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
139 if (walk->type != ex->type)
140 continue;
141 if (walk->major != ex->major)
142 continue;
143 if (walk->minor != ex->minor)
144 continue;
145
146 walk->access &= ~ex->access;
147 if (!walk->access) {
148 list_del_rcu(&walk->list);
149 kfree_rcu(walk, rcu);
150 }
151 }
152}
153
154static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
155{
156 struct dev_exception_item *ex, *tmp;
157
158 list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
159 list_del_rcu(&ex->list);
160 kfree_rcu(ex, rcu);
161 }
162}
163
164/**
165 * dev_exception_clean - frees all entries of the exception list
166 * @dev_cgroup: dev_cgroup with the exception list to be cleaned
167 *
168 * called under devcgroup_mutex
169 */
170static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
171{
172 lockdep_assert_held(&devcgroup_mutex);
173
174 __dev_exception_clean(dev_cgroup);
175}
176
177static inline bool is_devcg_online(const struct dev_cgroup *devcg)
178{
179 return (devcg->behavior != DEVCG_DEFAULT_NONE);
180}
181
182/**
183 * devcgroup_online - initializes devcgroup's behavior and exceptions based on
184 * parent's
185 * @css: css getting online
186 * returns 0 in case of success, error code otherwise
187 */
188static int devcgroup_online(struct cgroup_subsys_state *css)
189{
190 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
191 struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css->parent);
192 int ret = 0;
193
194 mutex_lock(&devcgroup_mutex);
195
196 if (parent_dev_cgroup == NULL)
197 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
198 else {
199 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
200 &parent_dev_cgroup->exceptions);
201 if (!ret)
202 dev_cgroup->behavior = parent_dev_cgroup->behavior;
203 }
204 mutex_unlock(&devcgroup_mutex);
205
206 return ret;
207}
208
209static void devcgroup_offline(struct cgroup_subsys_state *css)
210{
211 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
212
213 mutex_lock(&devcgroup_mutex);
214 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
215 mutex_unlock(&devcgroup_mutex);
216}
217
218/*
219 * called from kernel/cgroup/cgroup.c with cgroup_lock() held.
220 */
221static struct cgroup_subsys_state *
222devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
223{
224 struct dev_cgroup *dev_cgroup;
225
226 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
227 if (!dev_cgroup)
228 return ERR_PTR(-ENOMEM);
229 INIT_LIST_HEAD(&dev_cgroup->exceptions);
230 dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
231
232 return &dev_cgroup->css;
233}
234
235static void devcgroup_css_free(struct cgroup_subsys_state *css)
236{
237 struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
238
239 __dev_exception_clean(dev_cgroup);
240 kfree(dev_cgroup);
241}
242
243#define DEVCG_ALLOW 1
244#define DEVCG_DENY 2
245#define DEVCG_LIST 3
246
247#define MAJMINLEN 13
248#define ACCLEN 4
249
250static void set_access(char *acc, short access)
251{
252 int idx = 0;
253 memset(acc, 0, ACCLEN);
254 if (access & DEVCG_ACC_READ)
255 acc[idx++] = 'r';
256 if (access & DEVCG_ACC_WRITE)
257 acc[idx++] = 'w';
258 if (access & DEVCG_ACC_MKNOD)
259 acc[idx++] = 'm';
260}
261
262static char type_to_char(short type)
263{
264 if (type == DEVCG_DEV_ALL)
265 return 'a';
266 if (type == DEVCG_DEV_CHAR)
267 return 'c';
268 if (type == DEVCG_DEV_BLOCK)
269 return 'b';
270 return 'X';
271}
272
273static void set_majmin(char *str, unsigned m)
274{
275 if (m == ~0)
276 strcpy(str, "*");
277 else
278 sprintf(str, "%u", m);
279}
280
281static int devcgroup_seq_show(struct seq_file *m, void *v)
282{
283 struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
284 struct dev_exception_item *ex;
285 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
286
287 rcu_read_lock();
288 /*
289 * To preserve the compatibility:
290 * - Only show the "all devices" when the default policy is to allow
291 * - List the exceptions in case the default policy is to deny
292 * This way, the file remains as a "whitelist of devices"
293 */
294 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
295 set_access(acc, DEVCG_ACC_MASK);
296 set_majmin(maj, ~0);
297 set_majmin(min, ~0);
298 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEVCG_DEV_ALL),
299 maj, min, acc);
300 } else {
301 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
302 set_access(acc, ex->access);
303 set_majmin(maj, ex->major);
304 set_majmin(min, ex->minor);
305 seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
306 maj, min, acc);
307 }
308 }
309 rcu_read_unlock();
310
311 return 0;
312}
313
314/**
315 * match_exception - iterates the exception list trying to find a complete match
316 * @exceptions: list of exceptions
317 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
318 * @major: device file major number, ~0 to match all
319 * @minor: device file minor number, ~0 to match all
320 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
321 *
322 * It is considered a complete match if an exception is found that will
323 * contain the entire range of provided parameters.
324 *
325 * Return: true in case it matches an exception completely
326 */
327static bool match_exception(struct list_head *exceptions, short type,
328 u32 major, u32 minor, short access)
329{
330 struct dev_exception_item *ex;
331
332 list_for_each_entry_rcu(ex, exceptions, list) {
333 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
334 continue;
335 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
336 continue;
337 if (ex->major != ~0 && ex->major != major)
338 continue;
339 if (ex->minor != ~0 && ex->minor != minor)
340 continue;
341 /* provided access cannot have more than the exception rule */
342 if (access & (~ex->access))
343 continue;
344 return true;
345 }
346 return false;
347}
348
349/**
350 * match_exception_partial - iterates the exception list trying to find a partial match
351 * @exceptions: list of exceptions
352 * @type: device type (DEVCG_DEV_BLOCK or DEVCG_DEV_CHAR)
353 * @major: device file major number, ~0 to match all
354 * @minor: device file minor number, ~0 to match all
355 * @access: permission mask (DEVCG_ACC_READ, DEVCG_ACC_WRITE, DEVCG_ACC_MKNOD)
356 *
357 * It is considered a partial match if an exception's range is found to
358 * contain *any* of the devices specified by provided parameters. This is
359 * used to make sure no extra access is being granted that is forbidden by
360 * any of the exception list.
361 *
362 * Return: true in case the provided range mat matches an exception completely
363 */
364static bool match_exception_partial(struct list_head *exceptions, short type,
365 u32 major, u32 minor, short access)
366{
367 struct dev_exception_item *ex;
368
369 list_for_each_entry_rcu(ex, exceptions, list,
370 lockdep_is_held(&devcgroup_mutex)) {
371 if ((type & DEVCG_DEV_BLOCK) && !(ex->type & DEVCG_DEV_BLOCK))
372 continue;
373 if ((type & DEVCG_DEV_CHAR) && !(ex->type & DEVCG_DEV_CHAR))
374 continue;
375 /*
376 * We must be sure that both the exception and the provided
377 * range aren't masking all devices
378 */
379 if (ex->major != ~0 && major != ~0 && ex->major != major)
380 continue;
381 if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
382 continue;
383 /*
384 * In order to make sure the provided range isn't matching
385 * an exception, all its access bits shouldn't match the
386 * exception's access bits
387 */
388 if (!(access & ex->access))
389 continue;
390 return true;
391 }
392 return false;
393}
394
395/**
396 * verify_new_ex - verifies if a new exception is allowed by parent cgroup's permissions
397 * @dev_cgroup: dev cgroup to be tested against
398 * @refex: new exception
399 * @behavior: behavior of the exception's dev_cgroup
400 *
401 * This is used to make sure a child cgroup won't have more privileges
402 * than its parent
403 */
404static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
405 struct dev_exception_item *refex,
406 enum devcg_behavior behavior)
407{
408 bool match = false;
409
410 RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
411 !lockdep_is_held(&devcgroup_mutex),
412 "device_cgroup:verify_new_ex called without proper synchronization");
413
414 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
415 if (behavior == DEVCG_DEFAULT_ALLOW) {
416 /*
417 * new exception in the child doesn't matter, only
418 * adding extra restrictions
419 */
420 return true;
421 } else {
422 /*
423 * new exception in the child will add more devices
424 * that can be accessed, so it can't match any of
425 * parent's exceptions, even slightly
426 */
427 match = match_exception_partial(&dev_cgroup->exceptions,
428 refex->type,
429 refex->major,
430 refex->minor,
431 refex->access);
432
433 if (match)
434 return false;
435 return true;
436 }
437 } else {
438 /*
439 * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
440 * the new exception will add access to more devices and must
441 * be contained completely in an parent's exception to be
442 * allowed
443 */
444 match = match_exception(&dev_cgroup->exceptions, refex->type,
445 refex->major, refex->minor,
446 refex->access);
447
448 if (match)
449 /* parent has an exception that matches the proposed */
450 return true;
451 else
452 return false;
453 }
454 return false;
455}
456
457/*
458 * parent_has_perm:
459 * when adding a new allow rule to a device exception list, the rule
460 * must be allowed in the parent device
461 */
462static int parent_has_perm(struct dev_cgroup *childcg,
463 struct dev_exception_item *ex)
464{
465 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
466
467 if (!parent)
468 return 1;
469 return verify_new_ex(parent, ex, childcg->behavior);
470}
471
472/**
473 * parent_allows_removal - verify if it's ok to remove an exception
474 * @childcg: child cgroup from where the exception will be removed
475 * @ex: exception being removed
476 *
477 * When removing an exception in cgroups with default ALLOW policy, it must
478 * be checked if removing it will give the child cgroup more access than the
479 * parent.
480 *
481 * Return: true if it's ok to remove exception, false otherwise
482 */
483static bool parent_allows_removal(struct dev_cgroup *childcg,
484 struct dev_exception_item *ex)
485{
486 struct dev_cgroup *parent = css_to_devcgroup(childcg->css.parent);
487
488 if (!parent)
489 return true;
490
491 /* It's always allowed to remove access to devices */
492 if (childcg->behavior == DEVCG_DEFAULT_DENY)
493 return true;
494
495 /*
496 * Make sure you're not removing part or a whole exception existing in
497 * the parent cgroup
498 */
499 return !match_exception_partial(&parent->exceptions, ex->type,
500 ex->major, ex->minor, ex->access);
501}
502
503/**
504 * may_allow_all - checks if it's possible to change the behavior to
505 * allow based on parent's rules.
506 * @parent: device cgroup's parent
507 * returns: != 0 in case it's allowed, 0 otherwise
508 */
509static inline int may_allow_all(struct dev_cgroup *parent)
510{
511 if (!parent)
512 return 1;
513 return parent->behavior == DEVCG_DEFAULT_ALLOW;
514}
515
516/**
517 * revalidate_active_exceptions - walks through the active exception list and
518 * revalidates the exceptions based on parent's
519 * behavior and exceptions. The exceptions that
520 * are no longer valid will be removed.
521 * Called with devcgroup_mutex held.
522 * @devcg: cgroup which exceptions will be checked
523 *
524 * This is one of the three key functions for hierarchy implementation.
525 * This function is responsible for re-evaluating all the cgroup's active
526 * exceptions due to a parent's exception change.
527 * Refer to Documentation/admin-guide/cgroup-v1/devices.rst for more details.
528 */
529static void revalidate_active_exceptions(struct dev_cgroup *devcg)
530{
531 struct dev_exception_item *ex;
532 struct list_head *this, *tmp;
533
534 list_for_each_safe(this, tmp, &devcg->exceptions) {
535 ex = container_of(this, struct dev_exception_item, list);
536 if (!parent_has_perm(devcg, ex))
537 dev_exception_rm(devcg, ex);
538 }
539}
540
541/**
542 * propagate_exception - propagates a new exception to the children
543 * @devcg_root: device cgroup that added a new exception
544 * @ex: new exception to be propagated
545 *
546 * returns: 0 in case of success, != 0 in case of error
547 */
548static int propagate_exception(struct dev_cgroup *devcg_root,
549 struct dev_exception_item *ex)
550{
551 struct cgroup_subsys_state *pos;
552 int rc = 0;
553
554 rcu_read_lock();
555
556 css_for_each_descendant_pre(pos, &devcg_root->css) {
557 struct dev_cgroup *devcg = css_to_devcgroup(pos);
558
559 /*
560 * Because devcgroup_mutex is held, no devcg will become
561 * online or offline during the tree walk (see on/offline
562 * methods), and online ones are safe to access outside RCU
563 * read lock without bumping refcnt.
564 */
565 if (pos == &devcg_root->css || !is_devcg_online(devcg))
566 continue;
567
568 rcu_read_unlock();
569
570 /*
571 * in case both root's behavior and devcg is allow, a new
572 * restriction means adding to the exception list
573 */
574 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
575 devcg->behavior == DEVCG_DEFAULT_ALLOW) {
576 rc = dev_exception_add(devcg, ex);
577 if (rc)
578 return rc;
579 } else {
580 /*
581 * in the other possible cases:
582 * root's behavior: allow, devcg's: deny
583 * root's behavior: deny, devcg's: deny
584 * the exception will be removed
585 */
586 dev_exception_rm(devcg, ex);
587 }
588 revalidate_active_exceptions(devcg);
589
590 rcu_read_lock();
591 }
592
593 rcu_read_unlock();
594 return rc;
595}
596
597/*
598 * Modify the exception list using allow/deny rules.
599 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD
600 * so we can give a container CAP_MKNOD to let it create devices but not
601 * modify the exception list.
602 * It seems likely we'll want to add a CAP_CONTAINER capability to allow
603 * us to also grant CAP_SYS_ADMIN to containers without giving away the
604 * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
605 *
606 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting
607 * new access is only allowed if you're in the top-level cgroup, or your
608 * parent cgroup has the access you're asking for.
609 */
610static int devcgroup_update_access(struct dev_cgroup *devcgroup,
611 int filetype, char *buffer)
612{
613 const char *b;
614 char temp[12]; /* 11 + 1 characters needed for a u32 */
615 int count, rc = 0;
616 struct dev_exception_item ex;
617 struct dev_cgroup *parent = css_to_devcgroup(devcgroup->css.parent);
618 struct dev_cgroup tmp_devcgrp;
619
620 if (!capable(CAP_SYS_ADMIN))
621 return -EPERM;
622
623 memset(&ex, 0, sizeof(ex));
624 memset(&tmp_devcgrp, 0, sizeof(tmp_devcgrp));
625 b = buffer;
626
627 switch (*b) {
628 case 'a':
629 switch (filetype) {
630 case DEVCG_ALLOW:
631 if (css_has_online_children(&devcgroup->css))
632 return -EINVAL;
633
634 if (!may_allow_all(parent))
635 return -EPERM;
636 if (!parent) {
637 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
638 dev_exception_clean(devcgroup);
639 break;
640 }
641
642 INIT_LIST_HEAD(&tmp_devcgrp.exceptions);
643 rc = dev_exceptions_copy(&tmp_devcgrp.exceptions,
644 &devcgroup->exceptions);
645 if (rc)
646 return rc;
647 dev_exception_clean(devcgroup);
648 rc = dev_exceptions_copy(&devcgroup->exceptions,
649 &parent->exceptions);
650 if (rc) {
651 dev_exceptions_move(&devcgroup->exceptions,
652 &tmp_devcgrp.exceptions);
653 return rc;
654 }
655 devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
656 dev_exception_clean(&tmp_devcgrp);
657 break;
658 case DEVCG_DENY:
659 if (css_has_online_children(&devcgroup->css))
660 return -EINVAL;
661
662 dev_exception_clean(devcgroup);
663 devcgroup->behavior = DEVCG_DEFAULT_DENY;
664 break;
665 default:
666 return -EINVAL;
667 }
668 return 0;
669 case 'b':
670 ex.type = DEVCG_DEV_BLOCK;
671 break;
672 case 'c':
673 ex.type = DEVCG_DEV_CHAR;
674 break;
675 default:
676 return -EINVAL;
677 }
678 b++;
679 if (!isspace(*b))
680 return -EINVAL;
681 b++;
682 if (*b == '*') {
683 ex.major = ~0;
684 b++;
685 } else if (isdigit(*b)) {
686 memset(temp, 0, sizeof(temp));
687 for (count = 0; count < sizeof(temp) - 1; count++) {
688 temp[count] = *b;
689 b++;
690 if (!isdigit(*b))
691 break;
692 }
693 rc = kstrtou32(temp, 10, &ex.major);
694 if (rc)
695 return -EINVAL;
696 } else {
697 return -EINVAL;
698 }
699 if (*b != ':')
700 return -EINVAL;
701 b++;
702
703 /* read minor */
704 if (*b == '*') {
705 ex.minor = ~0;
706 b++;
707 } else if (isdigit(*b)) {
708 memset(temp, 0, sizeof(temp));
709 for (count = 0; count < sizeof(temp) - 1; count++) {
710 temp[count] = *b;
711 b++;
712 if (!isdigit(*b))
713 break;
714 }
715 rc = kstrtou32(temp, 10, &ex.minor);
716 if (rc)
717 return -EINVAL;
718 } else {
719 return -EINVAL;
720 }
721 if (!isspace(*b))
722 return -EINVAL;
723 for (b++, count = 0; count < 3; count++, b++) {
724 switch (*b) {
725 case 'r':
726 ex.access |= DEVCG_ACC_READ;
727 break;
728 case 'w':
729 ex.access |= DEVCG_ACC_WRITE;
730 break;
731 case 'm':
732 ex.access |= DEVCG_ACC_MKNOD;
733 break;
734 case '\n':
735 case '\0':
736 count = 3;
737 break;
738 default:
739 return -EINVAL;
740 }
741 }
742
743 switch (filetype) {
744 case DEVCG_ALLOW:
745 /*
746 * If the default policy is to allow by default, try to remove
747 * an matching exception instead. And be silent about it: we
748 * don't want to break compatibility
749 */
750 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
751 /* Check if the parent allows removing it first */
752 if (!parent_allows_removal(devcgroup, &ex))
753 return -EPERM;
754 dev_exception_rm(devcgroup, &ex);
755 break;
756 }
757
758 if (!parent_has_perm(devcgroup, &ex))
759 return -EPERM;
760 rc = dev_exception_add(devcgroup, &ex);
761 break;
762 case DEVCG_DENY:
763 /*
764 * If the default policy is to deny by default, try to remove
765 * an matching exception instead. And be silent about it: we
766 * don't want to break compatibility
767 */
768 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
769 dev_exception_rm(devcgroup, &ex);
770 else
771 rc = dev_exception_add(devcgroup, &ex);
772
773 if (rc)
774 break;
775 /* we only propagate new restrictions */
776 rc = propagate_exception(devcgroup, &ex);
777 break;
778 default:
779 rc = -EINVAL;
780 }
781 return rc;
782}
783
784static ssize_t devcgroup_access_write(struct kernfs_open_file *of,
785 char *buf, size_t nbytes, loff_t off)
786{
787 int retval;
788
789 mutex_lock(&devcgroup_mutex);
790 retval = devcgroup_update_access(css_to_devcgroup(of_css(of)),
791 of_cft(of)->private, strstrip(buf));
792 mutex_unlock(&devcgroup_mutex);
793 return retval ?: nbytes;
794}
795
796static struct cftype dev_cgroup_files[] = {
797 {
798 .name = "allow",
799 .write = devcgroup_access_write,
800 .private = DEVCG_ALLOW,
801 },
802 {
803 .name = "deny",
804 .write = devcgroup_access_write,
805 .private = DEVCG_DENY,
806 },
807 {
808 .name = "list",
809 .seq_show = devcgroup_seq_show,
810 .private = DEVCG_LIST,
811 },
812 { } /* terminate */
813};
814
815struct cgroup_subsys devices_cgrp_subsys = {
816 .css_alloc = devcgroup_css_alloc,
817 .css_free = devcgroup_css_free,
818 .css_online = devcgroup_online,
819 .css_offline = devcgroup_offline,
820 .legacy_cftypes = dev_cgroup_files,
821};
822
823/**
824 * devcgroup_legacy_check_permission - checks if an inode operation is permitted
825 * @type: device type
826 * @major: device major number
827 * @minor: device minor number
828 * @access: combination of DEVCG_ACC_WRITE, DEVCG_ACC_READ and DEVCG_ACC_MKNOD
829 *
830 * returns 0 on success, -EPERM case the operation is not permitted
831 */
832static int devcgroup_legacy_check_permission(short type, u32 major, u32 minor,
833 short access)
834{
835 struct dev_cgroup *dev_cgroup;
836 bool rc;
837
838 rcu_read_lock();
839 dev_cgroup = task_devcgroup(current);
840 if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
841 /* Can't match any of the exceptions, even partially */
842 rc = !match_exception_partial(&dev_cgroup->exceptions,
843 type, major, minor, access);
844 else
845 /* Need to match completely one exception to be allowed */
846 rc = match_exception(&dev_cgroup->exceptions, type, major,
847 minor, access);
848 rcu_read_unlock();
849
850 if (!rc)
851 return -EPERM;
852
853 return 0;
854}
855
856#endif /* CONFIG_CGROUP_DEVICE */
857
858#if defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF)
859
860int devcgroup_check_permission(short type, u32 major, u32 minor, short access)
861{
862 int rc = BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type, major, minor, access);
863
864 if (rc)
865 return rc;
866
867 #ifdef CONFIG_CGROUP_DEVICE
868 return devcgroup_legacy_check_permission(type, major, minor, access);
869
870 #else /* CONFIG_CGROUP_DEVICE */
871 return 0;
872
873 #endif /* CONFIG_CGROUP_DEVICE */
874}
875EXPORT_SYMBOL(devcgroup_check_permission);
876#endif /* defined(CONFIG_CGROUP_DEVICE) || defined(CONFIG_CGROUP_BPF) */