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         struct cgroup *parent_cgroup;
 240
 241         dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
 242         if (!dev_cgroup)
 243                 return ERR_PTR(-ENOMEM);
 244         INIT_LIST_HEAD(&dev_cgroup->exceptions);
 245         INIT_LIST_HEAD(&dev_cgroup->propagate_pending);
 246         dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
 247         parent_cgroup = cgroup->parent;
 248
 249         return &dev_cgroup->css;
 250 }
 251
 252 static void devcgroup_css_free(struct cgroup *cgroup)
 253 {
 254         struct dev_cgroup *dev_cgroup;
 255
 256         dev_cgroup = cgroup_to_devcgroup(cgroup);
 257         __dev_exception_clean(dev_cgroup);
 258         kfree(dev_cgroup);
 259 }
 260
 261 #define DEVCG_ALLOW 1
 262 #define DEVCG_DENY 2
 263 #define DEVCG_LIST 3
 264
 265 #define MAJMINLEN 13
 266 #define ACCLEN 4
 267
 268 static void set_access(char *acc, short access)
 269 {
 270         int idx = 0;
 271         memset(acc, 0, ACCLEN);
 272         if (access & ACC_READ)
 273                 acc[idx++] = 'r';
 274         if (access & ACC_WRITE)
 275                 acc[idx++] = 'w';
 276         if (access & ACC_MKNOD)
 277                 acc[idx++] = 'm';
 278 }
 279
 280 static char type_to_char(short type)
 281 {
 282         if (type == DEV_ALL)
 283                 return 'a';
 284         if (type == DEV_CHAR)
 285                 return 'c';
 286         if (type == DEV_BLOCK)
 287                 return 'b';
 288         return 'X';
 289 }
 290
 291 static void set_majmin(char *str, unsigned m)
 292 {
 293         if (m == ~0)
 294                 strcpy(str, "*");
 295         else
 296                 sprintf(str, "%u", m);
 297 }
 298
 299 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
 300                                 struct seq_file *m)
 301 {
 302         struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
 303         struct dev_exception_item *ex;
 304         char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
 305
 306         rcu_read_lock();
 307         /*
 308          * To preserve the compatibility:
 309          * - Only show the "all devices" when the default policy is to allow
 310          * - List the exceptions in case the default policy is to deny
 311          * This way, the file remains as a "whitelist of devices"
 312          */
 313         if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 314                 set_access(acc, ACC_MASK);
 315                 set_majmin(maj, ~0);
 316                 set_majmin(min, ~0);
 317                 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
 318                            maj, min, acc);
 319         } else {
 320                 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
 321                         set_access(acc, ex->access);
 322                         set_majmin(maj, ex->major);
 323                         set_majmin(min, ex->minor);
 324                         seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
 325                                    maj, min, acc);
 326                 }
 327         }
 328         rcu_read_unlock();
 329
 330         return 0;
 331 }
 332
 333 /**
 334  * may_access - verifies if a new exception is part of what is allowed
 335  *              by a dev cgroup based on the default policy +
 336  *              exceptions. This is used to make sure a child cgroup
 337  *              won't have more privileges than its parent or to
 338  *              verify if a certain access is allowed.
 339  * @dev_cgroup: dev cgroup to be tested against
 340  * @refex: new exception
 341  * @behavior: behavior of the exception
 342  */
 343 static bool may_access(struct dev_cgroup *dev_cgroup,
 344                        struct dev_exception_item *refex,
 345                        enum devcg_behavior behavior)
 346 {
 347         struct dev_exception_item *ex;
 348         bool match = false;
 349
 350         rcu_lockdep_assert(rcu_read_lock_held() ||
 351                            lockdep_is_held(&devcgroup_mutex),
 352                            "device_cgroup::may_access() called without proper synchronization");
 353
 354         list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
 355                 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
 356                         continue;
 357                 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
 358                         continue;
 359                 if (ex->major != ~0 && ex->major != refex->major)
 360                         continue;
 361                 if (ex->minor != ~0 && ex->minor != refex->minor)
 362                         continue;
 363                 if (refex->access & (~ex->access))
 364                         continue;
 365                 match = true;
 366                 break;
 367         }
 368
 369         if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 370                 if (behavior == DEVCG_DEFAULT_ALLOW) {
 371                         /* the exception will deny access to certain devices */
 372                         return true;
 373                 } else {
 374                         /* the exception will allow access to certain devices */
 375                         if (match)
 376                                 /*
 377                                  * a new exception allowing access shouldn't
 378                                  * match an parent's exception
 379                                  */
 380                                 return false;
 381                         return true;
 382                 }
 383         } else {
 384                 /* only behavior == DEVCG_DEFAULT_DENY allowed here */
 385                 if (match)
 386                         /* parent has an exception that matches the proposed */
 387                         return true;
 388                 else
 389                         return false;
 390         }
 391         return false;
 392 }
 393
 394 /*
 395  * parent_has_perm:
 396  * when adding a new allow rule to a device exception list, the rule
 397  * must be allowed in the parent device
 398  */
 399 static int parent_has_perm(struct dev_cgroup *childcg,
 400                                   struct dev_exception_item *ex)
 401 {
 402         struct cgroup *pcg = childcg->css.cgroup->parent;
 403         struct dev_cgroup *parent;
 404
 405         if (!pcg)
 406                 return 1;
 407         parent = cgroup_to_devcgroup(pcg);
 408         return may_access(parent, ex, childcg->behavior);
 409 }
 410
 411 /**
 412  * may_allow_all - checks if it's possible to change the behavior to
 413  *                 allow based on parent's rules.
 414  * @parent: device cgroup's parent
 415  * returns: != 0 in case it's allowed, 0 otherwise
 416  */
 417 static inline int may_allow_all(struct dev_cgroup *parent)
 418 {
 419         if (!parent)
 420                 return 1;
 421         return parent->behavior == DEVCG_DEFAULT_ALLOW;
 422 }
 423
 424 /**
 425  * revalidate_active_exceptions - walks through the active exception list and
 426  *                                revalidates the exceptions based on parent's
 427  *                                behavior and exceptions. The exceptions that
 428  *                                are no longer valid will be removed.
 429  *                                Called with devcgroup_mutex held.
 430  * @devcg: cgroup which exceptions will be checked
 431  *
 432  * This is one of the three key functions for hierarchy implementation.
 433  * This function is responsible for re-evaluating all the cgroup's active
 434  * exceptions due to a parent's exception change.
 435  * Refer to Documentation/cgroups/devices.txt for more details.
 436  */
 437 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
 438 {
 439         struct dev_exception_item *ex;
 440         struct list_head *this, *tmp;
 441
 442         list_for_each_safe(this, tmp, &devcg->exceptions) {
 443                 ex = container_of(this, struct dev_exception_item, list);
 444                 if (!parent_has_perm(devcg, ex))
 445                         dev_exception_rm(devcg, ex);
 446         }
 447 }
 448
 449 /**
 450  * get_online_devcg - walks the cgroup tree and fills a list with the online
 451  *                    groups
 452  * @root: cgroup used as starting point
 453  * @online: list that will be filled with online groups
 454  *
 455  * Must be called with devcgroup_mutex held. Grabs RCU lock.
 456  * Because devcgroup_mutex is held, no devcg will become online or offline
 457  * during the tree walk (see devcgroup_online, devcgroup_offline)
 458  * A separated list is needed because propagate_behavior() and
 459  * propagate_exception() need to allocate memory and can block.
 460  */
 461 static void get_online_devcg(struct cgroup *root, struct list_head *online)
 462 {
 463         struct cgroup *pos;
 464         struct dev_cgroup *devcg;
 465
 466         lockdep_assert_held(&devcgroup_mutex);
 467
 468         rcu_read_lock();
 469         cgroup_for_each_descendant_pre(pos, root) {
 470                 devcg = cgroup_to_devcgroup(pos);
 471                 if (is_devcg_online(devcg))
 472                         list_add_tail(&devcg->propagate_pending, online);
 473         }
 474         rcu_read_unlock();
 475 }
 476
 477 /**
 478  * propagate_exception - propagates a new exception to the children
 479  * @devcg_root: device cgroup that added a new exception
 480  * @ex: new exception to be propagated
 481  *
 482  * returns: 0 in case of success, != 0 in case of error
 483  */
 484 static int propagate_exception(struct dev_cgroup *devcg_root,
 485                                struct dev_exception_item *ex)
 486 {
 487         struct cgroup *root = devcg_root->css.cgroup;
 488         struct dev_cgroup *devcg, *parent, *tmp;
 489         int rc = 0;
 490         LIST_HEAD(pending);
 491
 492         get_online_devcg(root, &pending);
 493
 494         list_for_each_entry_safe(devcg, tmp, &pending, propagate_pending) {
 495                 parent = cgroup_to_devcgroup(devcg->css.cgroup->parent);
 496
 497                 /*
 498                  * in case both root's behavior and devcg is allow, a new
 499                  * restriction means adding to the exception list
 500                  */
 501                 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
 502                     devcg->behavior == DEVCG_DEFAULT_ALLOW) {
 503                         rc = dev_exception_add(devcg, ex);
 504                         if (rc)
 505                                 break;
 506                 } else {
 507                         /*
 508                          * in the other possible cases:
 509                          * root's behavior: allow, devcg's: deny
 510                          * root's behavior: deny, devcg's: deny
 511                          * the exception will be removed
 512                          */
 513                         dev_exception_rm(devcg, ex);
 514                 }
 515                 revalidate_active_exceptions(devcg);
 516
 517                 list_del_init(&devcg->propagate_pending);
 518         }
 519         return rc;
 520 }
 521
 522 static inline bool has_children(struct dev_cgroup *devcgroup)
 523 {
 524         struct cgroup *cgrp = devcgroup->css.cgroup;
 525
 526         return !list_empty(&cgrp->children);
 527 }
 528
 529 /*
 530  * Modify the exception list using allow/deny rules.
 531  * CAP_SYS_ADMIN is needed for this.  It's at least separate from CAP_MKNOD
 532  * so we can give a container CAP_MKNOD to let it create devices but not
 533  * modify the exception list.
 534  * It seems likely we'll want to add a CAP_CONTAINER capability to allow
 535  * us to also grant CAP_SYS_ADMIN to containers without giving away the
 536  * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
 537  *
 538  * Taking rules away is always allowed (given CAP_SYS_ADMIN).  Granting
 539  * new access is only allowed if you're in the top-level cgroup, or your
 540  * parent cgroup has the access you're asking for.
 541  */
 542 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
 543                                    int filetype, const char *buffer)
 544 {
 545         const char *b;
 546         char temp[12];          /* 11 + 1 characters needed for a u32 */
 547         int count, rc = 0;
 548         struct dev_exception_item ex;
 549         struct cgroup *p = devcgroup->css.cgroup;
 550         struct dev_cgroup *parent = NULL;
 551
 552         if (!capable(CAP_SYS_ADMIN))
 553                 return -EPERM;
 554
 555         if (p->parent)
 556                 parent = cgroup_to_devcgroup(p->parent);
 557
 558         memset(&ex, 0, sizeof(ex));
 559         b = buffer;
 560
 561         switch (*b) {
 562         case 'a':
 563                 switch (filetype) {
 564                 case DEVCG_ALLOW:
 565                         if (has_children(devcgroup))
 566                                 return -EINVAL;
 567
 568                         if (!may_allow_all(parent))
 569                                 return -EPERM;
 570                         dev_exception_clean(devcgroup);
 571                         devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
 572                         if (!parent)
 573                                 break;
 574
 575                         rc = dev_exceptions_copy(&devcgroup->exceptions,
 576                                                  &parent->exceptions);
 577                         if (rc)
 578                                 return rc;
 579                         break;
 580                 case DEVCG_DENY:
 581                         if (has_children(devcgroup))
 582                                 return -EINVAL;
 583
 584                         dev_exception_clean(devcgroup);
 585                         devcgroup->behavior = DEVCG_DEFAULT_DENY;
 586                         break;
 587                 default:
 588                         return -EINVAL;
 589                 }
 590                 return 0;
 591         case 'b':
 592                 ex.type = DEV_BLOCK;
 593                 break;
 594         case 'c':
 595                 ex.type = DEV_CHAR;
 596                 break;
 597         default:
 598                 return -EINVAL;
 599         }
 600         b++;
 601         if (!isspace(*b))
 602                 return -EINVAL;
 603         b++;
 604         if (*b == '*') {
 605                 ex.major = ~0;
 606                 b++;
 607         } else if (isdigit(*b)) {
 608                 memset(temp, 0, sizeof(temp));
 609                 for (count = 0; count < sizeof(temp) - 1; count++) {
 610                         temp[count] = *b;
 611                         b++;
 612                         if (!isdigit(*b))
 613                                 break;
 614                 }
 615                 rc = kstrtou32(temp, 10, &ex.major);
 616                 if (rc)
 617                         return -EINVAL;
 618         } else {
 619                 return -EINVAL;
 620         }
 621         if (*b != ':')
 622                 return -EINVAL;
 623         b++;
 624
 625         /* read minor */
 626         if (*b == '*') {
 627                 ex.minor = ~0;
 628                 b++;
 629         } else if (isdigit(*b)) {
 630                 memset(temp, 0, sizeof(temp));
 631                 for (count = 0; count < sizeof(temp) - 1; count++) {
 632                         temp[count] = *b;
 633                         b++;
 634                         if (!isdigit(*b))
 635                                 break;
 636                 }
 637                 rc = kstrtou32(temp, 10, &ex.minor);
 638                 if (rc)
 639                         return -EINVAL;
 640         } else {
 641                 return -EINVAL;
 642         }
 643         if (!isspace(*b))
 644                 return -EINVAL;
 645         for (b++, count = 0; count < 3; count++, b++) {
 646                 switch (*b) {
 647                 case 'r':
 648                         ex.access |= ACC_READ;
 649                         break;
 650                 case 'w':
 651                         ex.access |= ACC_WRITE;
 652                         break;
 653                 case 'm':
 654                         ex.access |= ACC_MKNOD;
 655                         break;
 656                 case '\n':
 657                 case '\0':
 658                         count = 3;
 659                         break;
 660                 default:
 661                         return -EINVAL;
 662                 }
 663         }
 664
 665         switch (filetype) {
 666         case DEVCG_ALLOW:
 667                 if (!parent_has_perm(devcgroup, &ex))
 668                         return -EPERM;
 669                 /*
 670                  * If the default policy is to allow by default, try to remove
 671                  * an matching exception instead. And be silent about it: we
 672                  * don't want to break compatibility
 673                  */
 674                 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 675                         dev_exception_rm(devcgroup, &ex);
 676                         return 0;
 677                 }
 678                 rc = dev_exception_add(devcgroup, &ex);
 679                 break;
 680         case DEVCG_DENY:
 681                 /*
 682                  * If the default policy is to deny by default, try to remove
 683                  * an matching exception instead. And be silent about it: we
 684                  * don't want to break compatibility
 685                  */
 686                 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
 687                         dev_exception_rm(devcgroup, &ex);
 688                 else
 689                         rc = dev_exception_add(devcgroup, &ex);
 690
 691                 if (rc)
 692                         break;
 693                 /* we only propagate new restrictions */
 694                 rc = propagate_exception(devcgroup, &ex);
 695                 break;
 696         default:
 697                 rc = -EINVAL;
 698         }
 699         return rc;
 700 }
 701
 702 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
 703                                   const char *buffer)
 704 {
 705         int retval;
 706
 707         mutex_lock(&devcgroup_mutex);
 708         retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
 709                                          cft->private, buffer);
 710         mutex_unlock(&devcgroup_mutex);
 711         return retval;
 712 }
 713
 714 static struct cftype dev_cgroup_files[] = {
 715         {
 716                 .name = "allow",
 717                 .write_string  = devcgroup_access_write,
 718                 .private = DEVCG_ALLOW,
 719         },
 720         {
 721                 .name = "deny",
 722                 .write_string = devcgroup_access_write,
 723                 .private = DEVCG_DENY,
 724         },
 725         {
 726                 .name = "list",
 727                 .read_seq_string = devcgroup_seq_read,
 728                 .private = DEVCG_LIST,
 729         },
 730         { }     /* terminate */
 731 };
 732
 733 struct cgroup_subsys devices_subsys = {
 734         .name = "devices",
 735         .can_attach = devcgroup_can_attach,
 736         .css_alloc = devcgroup_css_alloc,
 737         .css_free = devcgroup_css_free,
 738         .css_online = devcgroup_online,
 739         .css_offline = devcgroup_offline,
 740         .subsys_id = devices_subsys_id,
 741         .base_cftypes = dev_cgroup_files,
 742 };
 743
 744 /**
 745  * __devcgroup_check_permission - checks if an inode operation is permitted
 746  * @dev_cgroup: the dev cgroup to be tested against
 747  * @type: device type
 748  * @major: device major number
 749  * @minor: device minor number
 750  * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
 751  *
 752  * returns 0 on success, -EPERM case the operation is not permitted
 753  */
 754 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
 755                                         short access)
 756 {
 757         struct dev_cgroup *dev_cgroup;
 758         struct dev_exception_item ex;
 759         int rc;
 760
 761         memset(&ex, 0, sizeof(ex));
 762         ex.type = type;
 763         ex.major = major;
 764         ex.minor = minor;
 765         ex.access = access;
 766
 767         rcu_read_lock();
 768         dev_cgroup = task_devcgroup(current);
 769         rc = may_access(dev_cgroup, &ex, dev_cgroup->behavior);
 770         rcu_read_unlock();
 771
 772         if (!rc)
 773                 return -EPERM;
 774
 775         return 0;
 776 }
 777
 778 int __devcgroup_inode_permission(struct inode *inode, int mask)
 779 {
 780         short type, access = 0;
 781
 782         if (S_ISBLK(inode->i_mode))
 783                 type = DEV_BLOCK;
 784         if (S_ISCHR(inode->i_mode))
 785                 type = DEV_CHAR;
 786         if (mask & MAY_WRITE)
 787                 access |= ACC_WRITE;
 788         if (mask & MAY_READ)
 789                 access |= ACC_READ;
 790
 791         return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
 792                         access);
 793 }
 794
 795 int devcgroup_inode_mknod(int mode, dev_t dev)
 796 {
 797         short type;
 798
 799         if (!S_ISBLK(mode) && !S_ISCHR(mode))
 800                 return 0;
 801
 802         if (S_ISBLK(mode))
 803                 type = DEV_BLOCK;
 804         else
 805                 type = DEV_CHAR;
 806
 807         return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),
 808                         ACC_MKNOD);
 809
 810 }