security/device_cgroup.c

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