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         /* temporary list for pending propagation operations */
  53         struct list_head propagate_pending;
  54 };
  55
  56 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
  57 {
  58         return container_of(s, struct dev_cgroup, css);
  59 }
  60
  61 static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup)
  62 {
  63         return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id));
  64 }
  65
  66 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
  67 {
  68         return css_to_devcgroup(task_subsys_state(task, devices_subsys_id));
  69 }
  70
  71 struct cgroup_subsys devices_subsys;
  72
  73 static int devcgroup_can_attach(struct cgroup *new_cgrp,
  74                                 struct cgroup_taskset *set)
  75 {
  76         struct task_struct *task = cgroup_taskset_first(set);
  77
  78         if (current != task && !capable(CAP_SYS_ADMIN))
  79                 return -EPERM;
  80         return 0;
  81 }
  82
  83 /*
  84  * called under devcgroup_mutex
  85  */
  86 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
  87 {
  88         struct dev_exception_item *ex, *tmp, *new;
  89
  90         lockdep_assert_held(&devcgroup_mutex);
  91
  92         list_for_each_entry(ex, orig, list) {
  93                 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
  94                 if (!new)
  95                         goto free_and_exit;
  96                 list_add_tail(&new->list, dest);
  97         }
  98
  99         return 0;
 100
 101 free_and_exit:
 102         list_for_each_entry_safe(ex, tmp, dest, list) {
 103                 list_del(&ex->list);
 104                 kfree(ex);
 105         }
 106         return -ENOMEM;
 107 }
 108
 109 /*
 110  * called under devcgroup_mutex
 111  */
 112 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
 113                              struct dev_exception_item *ex)
 114 {
 115         struct dev_exception_item *excopy, *walk;
 116
 117         lockdep_assert_held(&devcgroup_mutex);
 118
 119         excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
 120         if (!excopy)
 121                 return -ENOMEM;
 122
 123         list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
 124                 if (walk->type != ex->type)
 125                         continue;
 126                 if (walk->major != ex->major)
 127                         continue;
 128                 if (walk->minor != ex->minor)
 129                         continue;
 130
 131                 walk->access |= ex->access;
 132                 kfree(excopy);
 133                 excopy = NULL;
 134         }
 135
 136         if (excopy != NULL)
 137                 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
 138         return 0;
 139 }
 140
 141 /*
 142  * called under devcgroup_mutex
 143  */
 144 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
 145                              struct dev_exception_item *ex)
 146 {
 147         struct dev_exception_item *walk, *tmp;
 148
 149         lockdep_assert_held(&devcgroup_mutex);
 150
 151         list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
 152                 if (walk->type != ex->type)
 153                         continue;
 154                 if (walk->major != ex->major)
 155                         continue;
 156                 if (walk->minor != ex->minor)
 157                         continue;
 158
 159                 walk->access &= ~ex->access;
 160                 if (!walk->access) {
 161                         list_del_rcu(&walk->list);
 162                         kfree_rcu(walk, rcu);
 163                 }
 164         }
 165 }
 166
 167 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
 168 {
 169         struct dev_exception_item *ex, *tmp;
 170
 171         list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
 172                 list_del_rcu(&ex->list);
 173                 kfree_rcu(ex, rcu);
 174         }
 175 }
 176
 177 /**
 178  * dev_exception_clean - frees all entries of the exception list
 179  * @dev_cgroup: dev_cgroup with the exception list to be cleaned
 180  *
 181  * called under devcgroup_mutex
 182  */
 183 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
 184 {
 185         lockdep_assert_held(&devcgroup_mutex);
 186
 187         __dev_exception_clean(dev_cgroup);
 188 }
 189
 190 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
 191 {
 192         return (devcg->behavior != DEVCG_DEFAULT_NONE);
 193 }
 194
 195 /**
 196  * devcgroup_online - initializes devcgroup's behavior and exceptions based on
 197  *                    parent's
 198  * @cgroup: cgroup getting online
 199  * returns 0 in case of success, error code otherwise
 200  */
 201 static int devcgroup_online(struct cgroup *cgroup)
 202 {
 203         struct dev_cgroup *dev_cgroup, *parent_dev_cgroup = NULL;
 204         int ret = 0;
 205
 206         mutex_lock(&devcgroup_mutex);
 207         dev_cgroup = cgroup_to_devcgroup(cgroup);
 208         if (cgroup->parent)
 209                 parent_dev_cgroup = cgroup_to_devcgroup(cgroup->parent);
 210
 211         if (parent_dev_cgroup == NULL)
 212                 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
 213         else {
 214                 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
 215                                           &parent_dev_cgroup->exceptions);
 216                 if (!ret)
 217                         dev_cgroup->behavior = parent_dev_cgroup->behavior;
 218         }
 219         mutex_unlock(&devcgroup_mutex);
 220
 221         return ret;
 222 }
 223
 224 static void devcgroup_offline(struct cgroup *cgroup)
 225 {
 226         struct dev_cgroup *dev_cgroup = cgroup_to_devcgroup(cgroup);
 227
 228         mutex_lock(&devcgroup_mutex);
 229         dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
 230         mutex_unlock(&devcgroup_mutex);
 231 }
 232
 233 /*
 234  * called from kernel/cgroup.c with cgroup_lock() held.
 235  */
 236 static struct cgroup_subsys_state *devcgroup_css_alloc(struct cgroup *cgroup)
 237 {
 238         struct dev_cgroup *dev_cgroup;
 239
 240         dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
 241         if (!dev_cgroup)
 242                 return ERR_PTR(-ENOMEM);
 243         INIT_LIST_HEAD(&dev_cgroup->exceptions);
 244         INIT_LIST_HEAD(&dev_cgroup->propagate_pending);
 245         dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
 246
 247         return &dev_cgroup->css;
 248 }
 249
 250 static void devcgroup_css_free(struct cgroup *cgroup)
 251 {
 252         struct dev_cgroup *dev_cgroup;
 253
 254         dev_cgroup = cgroup_to_devcgroup(cgroup);
 255         __dev_exception_clean(dev_cgroup);
 256         kfree(dev_cgroup);
 257 }
 258
 259 #define DEVCG_ALLOW 1
 260 #define DEVCG_DENY 2
 261 #define DEVCG_LIST 3
 262
 263 #define MAJMINLEN 13
 264 #define ACCLEN 4
 265
 266 static void set_access(char *acc, short access)
 267 {
 268         int idx = 0;
 269         memset(acc, 0, ACCLEN);
 270         if (access & ACC_READ)
 271                 acc[idx++] = 'r';
 272         if (access & ACC_WRITE)
 273                 acc[idx++] = 'w';
 274         if (access & ACC_MKNOD)
 275                 acc[idx++] = 'm';
 276 }
 277
 278 static char type_to_char(short type)
 279 {
 280         if (type == DEV_ALL)
 281                 return 'a';
 282         if (type == DEV_CHAR)
 283                 return 'c';
 284         if (type == DEV_BLOCK)
 285                 return 'b';
 286         return 'X';
 287 }
 288
 289 static void set_majmin(char *str, unsigned m)
 290 {
 291         if (m == ~0)
 292                 strcpy(str, "*");
 293         else
 294                 sprintf(str, "%u", m);
 295 }
 296
 297 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
 298                                 struct seq_file *m)
 299 {
 300         struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
 301         struct dev_exception_item *ex;
 302         char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
 303
 304         rcu_read_lock();
 305         /*
 306          * To preserve the compatibility:
 307          * - Only show the "all devices" when the default policy is to allow
 308          * - List the exceptions in case the default policy is to deny
 309          * This way, the file remains as a "whitelist of devices"
 310          */
 311         if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 312                 set_access(acc, ACC_MASK);
 313                 set_majmin(maj, ~0);
 314                 set_majmin(min, ~0);
 315                 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
 316                            maj, min, acc);
 317         } else {
 318                 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
 319                         set_access(acc, ex->access);
 320                         set_majmin(maj, ex->major);
 321                         set_majmin(min, ex->minor);
 322                         seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
 323                                    maj, min, acc);
 324                 }
 325         }
 326         rcu_read_unlock();
 327
 328         return 0;
 329 }
 330
 331 /**
 332  * may_access - verifies if a new exception is part of what is allowed
 333  *              by a dev cgroup based on the default policy +
 334  *              exceptions. This is used to make sure a child cgroup
 335  *              won't have more privileges than its parent or to
 336  *              verify if a certain access is allowed.
 337  * @dev_cgroup: dev cgroup to be tested against
 338  * @refex: new exception
 339  * @behavior: behavior of the exception
 340  */
 341 static bool may_access(struct dev_cgroup *dev_cgroup,
 342                        struct dev_exception_item *refex,
 343                        enum devcg_behavior behavior)
 344 {
 345         struct dev_exception_item *ex;
 346         bool match = false;
 347
 348         rcu_lockdep_assert(rcu_read_lock_held() ||
 349                            lockdep_is_held(&devcgroup_mutex),
 350                            "device_cgroup::may_access() called without proper synchronization");
 351
 352         list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
 353                 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
 354                         continue;
 355                 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
 356                         continue;
 357                 if (ex->major != ~0 && ex->major != refex->major)
 358                         continue;
 359                 if (ex->minor != ~0 && ex->minor != refex->minor)
 360                         continue;
 361                 if (refex->access & (~ex->access))
 362                         continue;
 363                 match = true;
 364                 break;
 365         }
 366
 367         if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 368                 if (behavior == DEVCG_DEFAULT_ALLOW) {
 369                         /* the exception will deny access to certain devices */
 370                         return true;
 371                 } else {
 372                         /* the exception will allow access to certain devices */
 373                         if (match)
 374                                 /*
 375                                  * a new exception allowing access shouldn't
 376                                  * match an parent's exception
 377                                  */
 378                                 return false;
 379                         return true;
 380                 }
 381         } else {
 382                 /* only behavior == DEVCG_DEFAULT_DENY allowed here */
 383                 if (match)
 384                         /* parent has an exception that matches the proposed */
 385                         return true;
 386                 else
 387                         return false;
 388         }
 389         return false;
 390 }
 391
 392 /*
 393  * parent_has_perm:
 394  * when adding a new allow rule to a device exception list, the rule
 395  * must be allowed in the parent device
 396  */
 397 static int parent_has_perm(struct dev_cgroup *childcg,
 398                                   struct dev_exception_item *ex)
 399 {
 400         struct cgroup *pcg = childcg->css.cgroup->parent;
 401         struct dev_cgroup *parent;
 402
 403         if (!pcg)
 404                 return 1;
 405         parent = cgroup_to_devcgroup(pcg);
 406         return may_access(parent, ex, childcg->behavior);
 407 }
 408
 409 /**
 410  * may_allow_all - checks if it's possible to change the behavior to
 411  *                 allow based on parent's rules.
 412  * @parent: device cgroup's parent
 413  * returns: != 0 in case it's allowed, 0 otherwise
 414  */
 415 static inline int may_allow_all(struct dev_cgroup *parent)
 416 {
 417         if (!parent)
 418                 return 1;
 419         return parent->behavior == DEVCG_DEFAULT_ALLOW;
 420 }
 421
 422 /**
 423  * revalidate_active_exceptions - walks through the active exception list and
 424  *                                revalidates the exceptions based on parent's
 425  *                                behavior and exceptions. The exceptions that
 426  *                                are no longer valid will be removed.
 427  *                                Called with devcgroup_mutex held.
 428  * @devcg: cgroup which exceptions will be checked
 429  *
 430  * This is one of the three key functions for hierarchy implementation.
 431  * This function is responsible for re-evaluating all the cgroup's active
 432  * exceptions due to a parent's exception change.
 433  * Refer to Documentation/cgroups/devices.txt for more details.
 434  */
 435 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
 436 {
 437         struct dev_exception_item *ex;
 438         struct list_head *this, *tmp;
 439
 440         list_for_each_safe(this, tmp, &devcg->exceptions) {
 441                 ex = container_of(this, struct dev_exception_item, list);
 442                 if (!parent_has_perm(devcg, ex))
 443                         dev_exception_rm(devcg, ex);
 444         }
 445 }
 446
 447 /**
 448  * get_online_devcg - walks the cgroup tree and fills a list with the online
 449  *                    groups
 450  * @root: cgroup used as starting point
 451  * @online: list that will be filled with online groups
 452  *
 453  * Must be called with devcgroup_mutex held. Grabs RCU lock.
 454  * Because devcgroup_mutex is held, no devcg will become online or offline
 455  * during the tree walk (see devcgroup_online, devcgroup_offline)
 456  * A separated list is needed because propagate_behavior() and
 457  * propagate_exception() need to allocate memory and can block.
 458  */
 459 static void get_online_devcg(struct cgroup *root, struct list_head *online)
 460 {
 461         struct cgroup *pos;
 462         struct dev_cgroup *devcg;
 463
 464         lockdep_assert_held(&devcgroup_mutex);
 465
 466         rcu_read_lock();
 467         cgroup_for_each_descendant_pre(pos, root) {
 468                 devcg = cgroup_to_devcgroup(pos);
 469                 if (is_devcg_online(devcg))
 470                         list_add_tail(&devcg->propagate_pending, online);
 471         }
 472         rcu_read_unlock();
 473 }
 474
 475 /**
 476  * propagate_exception - propagates a new exception to the children
 477  * @devcg_root: device cgroup that added a new exception
 478  * @ex: new exception to be propagated
 479  *
 480  * returns: 0 in case of success, != 0 in case of error
 481  */
 482 static int propagate_exception(struct dev_cgroup *devcg_root,
 483                                struct dev_exception_item *ex)
 484 {
 485         struct cgroup *root = devcg_root->css.cgroup;
 486         struct dev_cgroup *devcg, *parent, *tmp;
 487         int rc = 0;
 488         LIST_HEAD(pending);
 489
 490         get_online_devcg(root, &pending);
 491
 492         list_for_each_entry_safe(devcg, tmp, &pending, propagate_pending) {
 493                 parent = cgroup_to_devcgroup(devcg->css.cgroup->parent);
 494
 495                 /*
 496                  * in case both root's behavior and devcg is allow, a new
 497                  * restriction means adding to the exception list
 498                  */
 499                 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
 500                     devcg->behavior == DEVCG_DEFAULT_ALLOW) {
 501                         rc = dev_exception_add(devcg, ex);
 502                         if (rc)
 503                                 break;
 504                 } else {
 505                         /*
 506                          * in the other possible cases:
 507                          * root's behavior: allow, devcg's: deny
 508                          * root's behavior: deny, devcg's: deny
 509                          * the exception will be removed
 510                          */
 511                         dev_exception_rm(devcg, ex);
 512                 }
 513                 revalidate_active_exceptions(devcg);
 514
 515                 list_del_init(&devcg->propagate_pending);
 516         }
 517         return rc;
 518 }
 519
 520 static inline bool has_children(struct dev_cgroup *devcgroup)
 521 {
 522         struct cgroup *cgrp = devcgroup->css.cgroup;
 523
 524         return !list_empty(&cgrp->children);
 525 }
 526
 527 /*
 528  * Modify the exception list using allow/deny rules.
 529  * CAP_SYS_ADMIN is needed for this.  It's at least separate from CAP_MKNOD
 530  * so we can give a container CAP_MKNOD to let it create devices but not
 531  * modify the exception list.
 532  * It seems likely we'll want to add a CAP_CONTAINER capability to allow
 533  * us to also grant CAP_SYS_ADMIN to containers without giving away the
 534  * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
 535  *
 536  * Taking rules away is always allowed (given CAP_SYS_ADMIN).  Granting
 537  * new access is only allowed if you're in the top-level cgroup, or your
 538  * parent cgroup has the access you're asking for.
 539  */
 540 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
 541                                    int filetype, const char *buffer)
 542 {
 543         const char *b;
 544         char temp[12];          /* 11 + 1 characters needed for a u32 */
 545         int count, rc = 0;
 546         struct dev_exception_item ex;
 547         struct cgroup *p = devcgroup->css.cgroup;
 548         struct dev_cgroup *parent = NULL;
 549
 550         if (!capable(CAP_SYS_ADMIN))
 551                 return -EPERM;
 552
 553         if (p->parent)
 554                 parent = cgroup_to_devcgroup(p->parent);
 555
 556         memset(&ex, 0, sizeof(ex));
 557         b = buffer;
 558
 559         switch (*b) {
 560         case 'a':
 561                 switch (filetype) {
 562                 case DEVCG_ALLOW:
 563                         if (has_children(devcgroup))
 564                                 return -EINVAL;
 565
 566                         if (!may_allow_all(parent))
 567                                 return -EPERM;
 568                         dev_exception_clean(devcgroup);
 569                         devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
 570                         if (!parent)
 571                                 break;
 572
 573                         rc = dev_exceptions_copy(&devcgroup->exceptions,
 574                                                  &parent->exceptions);
 575                         if (rc)
 576                                 return rc;
 577                         break;
 578                 case DEVCG_DENY:
 579                         if (has_children(devcgroup))
 580                                 return -EINVAL;
 581
 582                         dev_exception_clean(devcgroup);
 583                         devcgroup->behavior = DEVCG_DEFAULT_DENY;
 584                         break;
 585                 default:
 586                         return -EINVAL;
 587                 }
 588                 return 0;
 589         case 'b':
 590                 ex.type = DEV_BLOCK;
 591                 break;
 592         case 'c':
 593                 ex.type = DEV_CHAR;
 594                 break;
 595         default:
 596                 return -EINVAL;
 597         }
 598         b++;
 599         if (!isspace(*b))
 600                 return -EINVAL;
 601         b++;
 602         if (*b == '*') {
 603                 ex.major = ~0;
 604                 b++;
 605         } else if (isdigit(*b)) {
 606                 memset(temp, 0, sizeof(temp));
 607                 for (count = 0; count < sizeof(temp) - 1; count++) {
 608                         temp[count] = *b;
 609                         b++;
 610                         if (!isdigit(*b))
 611                                 break;
 612                 }
 613                 rc = kstrtou32(temp, 10, &ex.major);
 614                 if (rc)
 615                         return -EINVAL;
 616         } else {
 617                 return -EINVAL;
 618         }
 619         if (*b != ':')
 620                 return -EINVAL;
 621         b++;
 622
 623         /* read minor */
 624         if (*b == '*') {
 625                 ex.minor = ~0;
 626                 b++;
 627         } else if (isdigit(*b)) {
 628                 memset(temp, 0, sizeof(temp));
 629                 for (count = 0; count < sizeof(temp) - 1; count++) {
 630                         temp[count] = *b;
 631                         b++;
 632                         if (!isdigit(*b))
 633                                 break;
 634                 }
 635                 rc = kstrtou32(temp, 10, &ex.minor);
 636                 if (rc)
 637                         return -EINVAL;
 638         } else {
 639                 return -EINVAL;
 640         }
 641         if (!isspace(*b))
 642                 return -EINVAL;
 643         for (b++, count = 0; count < 3; count++, b++) {
 644                 switch (*b) {
 645                 case 'r':
 646                         ex.access |= ACC_READ;
 647                         break;
 648                 case 'w':
 649                         ex.access |= ACC_WRITE;
 650                         break;
 651                 case 'm':
 652                         ex.access |= ACC_MKNOD;
 653                         break;
 654                 case '\n':
 655                 case '\0':
 656                         count = 3;
 657                         break;
 658                 default:
 659                         return -EINVAL;
 660                 }
 661         }
 662
 663         switch (filetype) {
 664         case DEVCG_ALLOW:
 665                 if (!parent_has_perm(devcgroup, &ex))
 666                         return -EPERM;
 667                 /*
 668                  * If the default policy is to allow by default, try to remove
 669                  * an matching exception instead. And be silent about it: we
 670                  * don't want to break compatibility
 671                  */
 672                 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 673                         dev_exception_rm(devcgroup, &ex);
 674                         return 0;
 675                 }
 676                 rc = dev_exception_add(devcgroup, &ex);
 677                 break;
 678         case DEVCG_DENY:
 679                 /*
 680                  * If the default policy is to deny by default, try to remove
 681                  * an matching exception instead. And be silent about it: we
 682                  * don't want to break compatibility
 683                  */
 684                 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
 685                         dev_exception_rm(devcgroup, &ex);
 686                 else
 687                         rc = dev_exception_add(devcgroup, &ex);
 688
 689                 if (rc)
 690                         break;
 691                 /* we only propagate new restrictions */
 692                 rc = propagate_exception(devcgroup, &ex);
 693                 break;
 694         default:
 695                 rc = -EINVAL;
 696         }
 697         return rc;
 698 }
 699
 700 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
 701                                   const char *buffer)
 702 {
 703         int retval;
 704
 705         mutex_lock(&devcgroup_mutex);
 706         retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
 707                                          cft->private, buffer);
 708         mutex_unlock(&devcgroup_mutex);
 709         return retval;
 710 }
 711
 712 static struct cftype dev_cgroup_files[] = {
 713         {
 714                 .name = "allow",
 715                 .write_string  = devcgroup_access_write,
 716                 .private = DEVCG_ALLOW,
 717         },
 718         {
 719                 .name = "deny",
 720                 .write_string = devcgroup_access_write,
 721                 .private = DEVCG_DENY,
 722         },
 723         {
 724                 .name = "list",
 725                 .read_seq_string = devcgroup_seq_read,
 726                 .private = DEVCG_LIST,
 727         },
 728         { }     /* terminate */
 729 };
 730
 731 struct cgroup_subsys devices_subsys = {
 732         .name = "devices",
 733         .can_attach = devcgroup_can_attach,
 734         .css_alloc = devcgroup_css_alloc,
 735         .css_free = devcgroup_css_free,
 736         .css_online = devcgroup_online,
 737         .css_offline = devcgroup_offline,
 738         .subsys_id = devices_subsys_id,
 739         .base_cftypes = dev_cgroup_files,
 740 };
 741
 742 /**
 743  * __devcgroup_check_permission - checks if an inode operation is permitted
 744  * @dev_cgroup: the dev cgroup to be tested against
 745  * @type: device type
 746  * @major: device major number
 747  * @minor: device minor number
 748  * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
 749  *
 750  * returns 0 on success, -EPERM case the operation is not permitted
 751  */
 752 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
 753                                         short access)
 754 {
 755         struct dev_cgroup *dev_cgroup;
 756         struct dev_exception_item ex;
 757         int rc;
 758
 759         memset(&ex, 0, sizeof(ex));
 760         ex.type = type;
 761         ex.major = major;
 762         ex.minor = minor;
 763         ex.access = access;
 764
 765         rcu_read_lock();
 766         dev_cgroup = task_devcgroup(current);
 767         rc = may_access(dev_cgroup, &ex, dev_cgroup->behavior);
 768         rcu_read_unlock();
 769
 770         if (!rc)
 771                 return -EPERM;
 772
 773         return 0;
 774 }
 775
 776 int __devcgroup_inode_permission(struct inode *inode, int mask)
 777 {
 778         short type, access = 0;
 779
 780         if (S_ISBLK(inode->i_mode))
 781                 type = DEV_BLOCK;
 782         if (S_ISCHR(inode->i_mode))
 783                 type = DEV_CHAR;
 784         if (mask & MAY_WRITE)
 785                 access |= ACC_WRITE;
 786         if (mask & MAY_READ)
 787                 access |= ACC_READ;
 788
 789         return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
 790                         access);
 791 }
 792
 793 int devcgroup_inode_mknod(int mode, dev_t dev)
 794 {
 795         short type;
 796
 797         if (!S_ISBLK(mode) && !S_ISCHR(mode))
 798                 return 0;
 799
 800         if (S_ISBLK(mode))
 801                 type = DEV_BLOCK;
 802         else
 803                 type = DEV_CHAR;
 804
 805         return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),
 806                         ACC_MKNOD);
 807
 808 }