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 *task_devcgroup(struct task_struct *task)
  60 {
  61         return css_to_devcgroup(task_css(task, devices_subsys_id));
  62 }
  63
  64 struct cgroup_subsys devices_subsys;
  65
  66 /*
  67  * called under devcgroup_mutex
  68  */
  69 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
  70 {
  71         struct dev_exception_item *ex, *tmp, *new;
  72
  73         lockdep_assert_held(&devcgroup_mutex);
  74
  75         list_for_each_entry(ex, orig, list) {
  76                 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
  77                 if (!new)
  78                         goto free_and_exit;
  79                 list_add_tail(&new->list, dest);
  80         }
  81
  82         return 0;
  83
  84 free_and_exit:
  85         list_for_each_entry_safe(ex, tmp, dest, list) {
  86                 list_del(&ex->list);
  87                 kfree(ex);
  88         }
  89         return -ENOMEM;
  90 }
  91
  92 /*
  93  * called under devcgroup_mutex
  94  */
  95 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
  96                              struct dev_exception_item *ex)
  97 {
  98         struct dev_exception_item *excopy, *walk;
  99
 100         lockdep_assert_held(&devcgroup_mutex);
 101
 102         excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
 103         if (!excopy)
 104                 return -ENOMEM;
 105
 106         list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
 107                 if (walk->type != ex->type)
 108                         continue;
 109                 if (walk->major != ex->major)
 110                         continue;
 111                 if (walk->minor != ex->minor)
 112                         continue;
 113
 114                 walk->access |= ex->access;
 115                 kfree(excopy);
 116                 excopy = NULL;
 117         }
 118
 119         if (excopy != NULL)
 120                 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
 121         return 0;
 122 }
 123
 124 /*
 125  * called under devcgroup_mutex
 126  */
 127 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
 128                              struct dev_exception_item *ex)
 129 {
 130         struct dev_exception_item *walk, *tmp;
 131
 132         lockdep_assert_held(&devcgroup_mutex);
 133
 134         list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
 135                 if (walk->type != ex->type)
 136                         continue;
 137                 if (walk->major != ex->major)
 138                         continue;
 139                 if (walk->minor != ex->minor)
 140                         continue;
 141
 142                 walk->access &= ~ex->access;
 143                 if (!walk->access) {
 144                         list_del_rcu(&walk->list);
 145                         kfree_rcu(walk, rcu);
 146                 }
 147         }
 148 }
 149
 150 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
 151 {
 152         struct dev_exception_item *ex, *tmp;
 153
 154         list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
 155                 list_del_rcu(&ex->list);
 156                 kfree_rcu(ex, rcu);
 157         }
 158 }
 159
 160 /**
 161  * dev_exception_clean - frees all entries of the exception list
 162  * @dev_cgroup: dev_cgroup with the exception list to be cleaned
 163  *
 164  * called under devcgroup_mutex
 165  */
 166 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
 167 {
 168         lockdep_assert_held(&devcgroup_mutex);
 169
 170         __dev_exception_clean(dev_cgroup);
 171 }
 172
 173 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
 174 {
 175         return (devcg->behavior != DEVCG_DEFAULT_NONE);
 176 }
 177
 178 /**
 179  * devcgroup_online - initializes devcgroup's behavior and exceptions based on
 180  *                    parent's
 181  * @css: css getting online
 182  * returns 0 in case of success, error code otherwise
 183  */
 184 static int devcgroup_online(struct cgroup_subsys_state *css)
 185 {
 186         struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
 187         struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css_parent(css));
 188         int ret = 0;
 189
 190         mutex_lock(&devcgroup_mutex);
 191
 192         if (parent_dev_cgroup == NULL)
 193                 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
 194         else {
 195                 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
 196                                           &parent_dev_cgroup->exceptions);
 197                 if (!ret)
 198                         dev_cgroup->behavior = parent_dev_cgroup->behavior;
 199         }
 200         mutex_unlock(&devcgroup_mutex);
 201
 202         return ret;
 203 }
 204
 205 static void devcgroup_offline(struct cgroup_subsys_state *css)
 206 {
 207         struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
 208
 209         mutex_lock(&devcgroup_mutex);
 210         dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
 211         mutex_unlock(&devcgroup_mutex);
 212 }
 213
 214 /*
 215  * called from kernel/cgroup.c with cgroup_lock() held.
 216  */
 217 static struct cgroup_subsys_state *
 218 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
 219 {
 220         struct dev_cgroup *dev_cgroup;
 221
 222         dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
 223         if (!dev_cgroup)
 224                 return ERR_PTR(-ENOMEM);
 225         INIT_LIST_HEAD(&dev_cgroup->exceptions);
 226         dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
 227
 228         return &dev_cgroup->css;
 229 }
 230
 231 static void devcgroup_css_free(struct cgroup_subsys_state *css)
 232 {
 233         struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
 234
 235         __dev_exception_clean(dev_cgroup);
 236         kfree(dev_cgroup);
 237 }
 238
 239 #define DEVCG_ALLOW 1
 240 #define DEVCG_DENY 2
 241 #define DEVCG_LIST 3
 242
 243 #define MAJMINLEN 13
 244 #define ACCLEN 4
 245
 246 static void set_access(char *acc, short access)
 247 {
 248         int idx = 0;
 249         memset(acc, 0, ACCLEN);
 250         if (access & ACC_READ)
 251                 acc[idx++] = 'r';
 252         if (access & ACC_WRITE)
 253                 acc[idx++] = 'w';
 254         if (access & ACC_MKNOD)
 255                 acc[idx++] = 'm';
 256 }
 257
 258 static char type_to_char(short type)
 259 {
 260         if (type == DEV_ALL)
 261                 return 'a';
 262         if (type == DEV_CHAR)
 263                 return 'c';
 264         if (type == DEV_BLOCK)
 265                 return 'b';
 266         return 'X';
 267 }
 268
 269 static void set_majmin(char *str, unsigned m)
 270 {
 271         if (m == ~0)
 272                 strcpy(str, "*");
 273         else
 274                 sprintf(str, "%u", m);
 275 }
 276
 277 static int devcgroup_seq_show(struct seq_file *m, void *v)
 278 {
 279         struct dev_cgroup *devcgroup = css_to_devcgroup(seq_css(m));
 280         struct dev_exception_item *ex;
 281         char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
 282
 283         rcu_read_lock();
 284         /*
 285          * To preserve the compatibility:
 286          * - Only show the "all devices" when the default policy is to allow
 287          * - List the exceptions in case the default policy is to deny
 288          * This way, the file remains as a "whitelist of devices"
 289          */
 290         if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 291                 set_access(acc, ACC_MASK);
 292                 set_majmin(maj, ~0);
 293                 set_majmin(min, ~0);
 294                 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
 295                            maj, min, acc);
 296         } else {
 297                 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
 298                         set_access(acc, ex->access);
 299                         set_majmin(maj, ex->major);
 300                         set_majmin(min, ex->minor);
 301                         seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
 302                                    maj, min, acc);
 303                 }
 304         }
 305         rcu_read_unlock();
 306
 307         return 0;
 308 }
 309
 310 /**
 311  * match_exception      - iterates the exception list trying to match a rule
 312  *                        based on type, major, minor and access type. It is
 313  *                        considered a match if an exception is found that
 314  *                        will contain the entire range of provided parameters.
 315  * @exceptions: list of exceptions
 316  * @type: device type (DEV_BLOCK or DEV_CHAR)
 317  * @major: device file major number, ~0 to match all
 318  * @minor: device file minor number, ~0 to match all
 319  * @access: permission mask (ACC_READ, ACC_WRITE, ACC_MKNOD)
 320  *
 321  * returns: true in case it matches an exception completely
 322  */
 323 static bool match_exception(struct list_head *exceptions, short type,
 324                             u32 major, u32 minor, short access)
 325 {
 326         struct dev_exception_item *ex;
 327
 328         list_for_each_entry_rcu(ex, exceptions, list) {
 329                 if ((type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
 330                         continue;
 331                 if ((type & DEV_CHAR) && !(ex->type & DEV_CHAR))
 332                         continue;
 333                 if (ex->major != ~0 && ex->major != major)
 334                         continue;
 335                 if (ex->minor != ~0 && ex->minor != minor)
 336                         continue;
 337                 /* provided access cannot have more than the exception rule */
 338                 if (access & (~ex->access))
 339                         continue;
 340                 return true;
 341         }
 342         return false;
 343 }
 344
 345 /**
 346  * match_exception_partial - iterates the exception list trying to match a rule
 347  *                           based on type, major, minor and access type. It is
 348  *                           considered a match if an exception's range is
 349  *                           found to contain *any* of the devices specified by
 350  *                           provided parameters. This is used to make sure no
 351  *                           extra access is being granted that is forbidden by
 352  *                           any of the exception list.
 353  * @exceptions: list of exceptions
 354  * @type: device type (DEV_BLOCK or DEV_CHAR)
 355  * @major: device file major number, ~0 to match all
 356  * @minor: device file minor number, ~0 to match all
 357  * @access: permission mask (ACC_READ, ACC_WRITE, ACC_MKNOD)
 358  *
 359  * returns: true in case the provided range mat matches an exception completely
 360  */
 361 static bool match_exception_partial(struct list_head *exceptions, short type,
 362                                     u32 major, u32 minor, short access)
 363 {
 364         struct dev_exception_item *ex;
 365
 366         list_for_each_entry_rcu(ex, exceptions, list) {
 367                 if ((type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
 368                         continue;
 369                 if ((type & DEV_CHAR) && !(ex->type & DEV_CHAR))
 370                         continue;
 371                 /*
 372                  * We must be sure that both the exception and the provided
 373                  * range aren't masking all devices
 374                  */
 375                 if (ex->major != ~0 && major != ~0 && ex->major != major)
 376                         continue;
 377                 if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
 378                         continue;
 379                 /*
 380                  * In order to make sure the provided range isn't matching
 381                  * an exception, all its access bits shouldn't match the
 382                  * exception's access bits
 383                  */
 384                 if (!(access & ex->access))
 385                         continue;
 386                 return true;
 387         }
 388         return false;
 389 }
 390
 391 /**
 392  * verify_new_ex - verifies if a new exception is part of what is allowed
 393  *                 by a dev cgroup based on the default policy +
 394  *                 exceptions. This is used to make sure a child cgroup
 395  *                 won't have more privileges than its parent
 396  * @dev_cgroup: dev cgroup to be tested against
 397  * @refex: new exception
 398  * @behavior: behavior of the exception's dev_cgroup
 399  */
 400 static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
 401                           struct dev_exception_item *refex,
 402                           enum devcg_behavior behavior)
 403 {
 404         bool match = false;
 405
 406         rcu_lockdep_assert(rcu_read_lock_held() ||
 407                            lockdep_is_held(&devcgroup_mutex),
 408                            "device_cgroup:verify_new_ex called without proper synchronization");
 409
 410         if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 411                 if (behavior == DEVCG_DEFAULT_ALLOW) {
 412                         /*
 413                          * new exception in the child doesn't matter, only
 414                          * adding extra restrictions
 415                          */
 416                         return true;
 417                 } else {
 418                         /*
 419                          * new exception in the child will add more devices
 420                          * that can be acessed, so it can't match any of
 421                          * parent's exceptions, even slightly
 422                          */
 423                         match = match_exception_partial(&dev_cgroup->exceptions,
 424                                                         refex->type,
 425                                                         refex->major,
 426                                                         refex->minor,
 427                                                         refex->access);
 428
 429                         if (match)
 430                                 return false;
 431                         return true;
 432                 }
 433         } else {
 434                 /*
 435                  * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
 436                  * the new exception will add access to more devices and must
 437                  * be contained completely in an parent's exception to be
 438                  * allowed
 439                  */
 440                 match = match_exception(&dev_cgroup->exceptions, refex->type,
 441                                         refex->major, refex->minor,
 442                                         refex->access);
 443
 444                 if (match)
 445                         /* parent has an exception that matches the proposed */
 446                         return true;
 447                 else
 448                         return false;
 449         }
 450         return false;
 451 }
 452
 453 /*
 454  * parent_has_perm:
 455  * when adding a new allow rule to a device exception list, the rule
 456  * must be allowed in the parent device
 457  */
 458 static int parent_has_perm(struct dev_cgroup *childcg,
 459                                   struct dev_exception_item *ex)
 460 {
 461         struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css));
 462
 463         if (!parent)
 464                 return 1;
 465         return verify_new_ex(parent, ex, childcg->behavior);
 466 }
 467
 468 /**
 469  * parent_allows_removal - verify if it's ok to remove an exception
 470  * @childcg: child cgroup from where the exception will be removed
 471  * @ex: exception being removed
 472  *
 473  * When removing an exception in cgroups with default ALLOW policy, it must
 474  * be checked if removing it will give the child cgroup more access than the
 475  * parent.
 476  *
 477  * Return: true if it's ok to remove exception, false otherwise
 478  */
 479 static bool parent_allows_removal(struct dev_cgroup *childcg,
 480                                   struct dev_exception_item *ex)
 481 {
 482         struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css));
 483
 484         if (!parent)
 485                 return true;
 486
 487         /* It's always allowed to remove access to devices */
 488         if (childcg->behavior == DEVCG_DEFAULT_DENY)
 489                 return true;
 490
 491         /*
 492          * Make sure you're not removing part or a whole exception existing in
 493          * the parent cgroup
 494          */
 495         return !match_exception_partial(&parent->exceptions, ex->type,
 496                                         ex->major, ex->minor, ex->access);
 497 }
 498
 499 /**
 500  * may_allow_all - checks if it's possible to change the behavior to
 501  *                 allow based on parent's rules.
 502  * @parent: device cgroup's parent
 503  * returns: != 0 in case it's allowed, 0 otherwise
 504  */
 505 static inline int may_allow_all(struct dev_cgroup *parent)
 506 {
 507         if (!parent)
 508                 return 1;
 509         return parent->behavior == DEVCG_DEFAULT_ALLOW;
 510 }
 511
 512 /**
 513  * revalidate_active_exceptions - walks through the active exception list and
 514  *                                revalidates the exceptions based on parent's
 515  *                                behavior and exceptions. The exceptions that
 516  *                                are no longer valid will be removed.
 517  *                                Called with devcgroup_mutex held.
 518  * @devcg: cgroup which exceptions will be checked
 519  *
 520  * This is one of the three key functions for hierarchy implementation.
 521  * This function is responsible for re-evaluating all the cgroup's active
 522  * exceptions due to a parent's exception change.
 523  * Refer to Documentation/cgroups/devices.txt for more details.
 524  */
 525 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
 526 {
 527         struct dev_exception_item *ex;
 528         struct list_head *this, *tmp;
 529
 530         list_for_each_safe(this, tmp, &devcg->exceptions) {
 531                 ex = container_of(this, struct dev_exception_item, list);
 532                 if (!parent_has_perm(devcg, ex))
 533                         dev_exception_rm(devcg, ex);
 534         }
 535 }
 536
 537 /**
 538  * propagate_exception - propagates a new exception to the children
 539  * @devcg_root: device cgroup that added a new exception
 540  * @ex: new exception to be propagated
 541  *
 542  * returns: 0 in case of success, != 0 in case of error
 543  */
 544 static int propagate_exception(struct dev_cgroup *devcg_root,
 545                                struct dev_exception_item *ex)
 546 {
 547         struct cgroup_subsys_state *pos;
 548         int rc = 0;
 549
 550         rcu_read_lock();
 551
 552         css_for_each_descendant_pre(pos, &devcg_root->css) {
 553                 struct dev_cgroup *devcg = css_to_devcgroup(pos);
 554
 555                 /*
 556                  * Because devcgroup_mutex is held, no devcg will become
 557                  * online or offline during the tree walk (see on/offline
 558                  * methods), and online ones are safe to access outside RCU
 559                  * read lock without bumping refcnt.
 560                  */
 561                 if (pos == &devcg_root->css || !is_devcg_online(devcg))
 562                         continue;
 563
 564                 rcu_read_unlock();
 565
 566                 /*
 567                  * in case both root's behavior and devcg is allow, a new
 568                  * restriction means adding to the exception list
 569                  */
 570                 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
 571                     devcg->behavior == DEVCG_DEFAULT_ALLOW) {
 572                         rc = dev_exception_add(devcg, ex);
 573                         if (rc)
 574                                 break;
 575                 } else {
 576                         /*
 577                          * in the other possible cases:
 578                          * root's behavior: allow, devcg's: deny
 579                          * root's behavior: deny, devcg's: deny
 580                          * the exception will be removed
 581                          */
 582                         dev_exception_rm(devcg, ex);
 583                 }
 584                 revalidate_active_exceptions(devcg);
 585
 586                 rcu_read_lock();
 587         }
 588
 589         rcu_read_unlock();
 590         return rc;
 591 }
 592
 593 static inline bool has_children(struct dev_cgroup *devcgroup)
 594 {
 595         struct cgroup *cgrp = devcgroup->css.cgroup;
 596
 597         return !list_empty(&cgrp->children);
 598 }
 599
 600 /*
 601  * Modify the exception list using allow/deny rules.
 602  * CAP_SYS_ADMIN is needed for this.  It's at least separate from CAP_MKNOD
 603  * so we can give a container CAP_MKNOD to let it create devices but not
 604  * modify the exception list.
 605  * It seems likely we'll want to add a CAP_CONTAINER capability to allow
 606  * us to also grant CAP_SYS_ADMIN to containers without giving away the
 607  * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
 608  *
 609  * Taking rules away is always allowed (given CAP_SYS_ADMIN).  Granting
 610  * new access is only allowed if you're in the top-level cgroup, or your
 611  * parent cgroup has the access you're asking for.
 612  */
 613 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
 614                                    int filetype, const char *buffer)
 615 {
 616         const char *b;
 617         char temp[12];          /* 11 + 1 characters needed for a u32 */
 618         int count, rc = 0;
 619         struct dev_exception_item ex;
 620         struct dev_cgroup *parent = css_to_devcgroup(css_parent(&devcgroup->css));
 621
 622         if (!capable(CAP_SYS_ADMIN))
 623                 return -EPERM;
 624
 625         memset(&ex, 0, sizeof(ex));
 626         b = buffer;
 627
 628         switch (*b) {
 629         case 'a':
 630                 switch (filetype) {
 631                 case DEVCG_ALLOW:
 632                         if (has_children(devcgroup))
 633                                 return -EINVAL;
 634
 635                         if (!may_allow_all(parent))
 636                                 return -EPERM;
 637                         dev_exception_clean(devcgroup);
 638                         devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
 639                         if (!parent)
 640                                 break;
 641
 642                         rc = dev_exceptions_copy(&devcgroup->exceptions,
 643                                                  &parent->exceptions);
 644                         if (rc)
 645                                 return rc;
 646                         break;
 647                 case DEVCG_DENY:
 648                         if (has_children(devcgroup))
 649                                 return -EINVAL;
 650
 651                         dev_exception_clean(devcgroup);
 652                         devcgroup->behavior = DEVCG_DEFAULT_DENY;
 653                         break;
 654                 default:
 655                         return -EINVAL;
 656                 }
 657                 return 0;
 658         case 'b':
 659                 ex.type = DEV_BLOCK;
 660                 break;
 661         case 'c':
 662                 ex.type = DEV_CHAR;
 663                 break;
 664         default:
 665                 return -EINVAL;
 666         }
 667         b++;
 668         if (!isspace(*b))
 669                 return -EINVAL;
 670         b++;
 671         if (*b == '*') {
 672                 ex.major = ~0;
 673                 b++;
 674         } else if (isdigit(*b)) {
 675                 memset(temp, 0, sizeof(temp));
 676                 for (count = 0; count < sizeof(temp) - 1; count++) {
 677                         temp[count] = *b;
 678                         b++;
 679                         if (!isdigit(*b))
 680                                 break;
 681                 }
 682                 rc = kstrtou32(temp, 10, &ex.major);
 683                 if (rc)
 684                         return -EINVAL;
 685         } else {
 686                 return -EINVAL;
 687         }
 688         if (*b != ':')
 689                 return -EINVAL;
 690         b++;
 691
 692         /* read minor */
 693         if (*b == '*') {
 694                 ex.minor = ~0;
 695                 b++;
 696         } else if (isdigit(*b)) {
 697                 memset(temp, 0, sizeof(temp));
 698                 for (count = 0; count < sizeof(temp) - 1; count++) {
 699                         temp[count] = *b;
 700                         b++;
 701                         if (!isdigit(*b))
 702                                 break;
 703                 }
 704                 rc = kstrtou32(temp, 10, &ex.minor);
 705                 if (rc)
 706                         return -EINVAL;
 707         } else {
 708                 return -EINVAL;
 709         }
 710         if (!isspace(*b))
 711                 return -EINVAL;
 712         for (b++, count = 0; count < 3; count++, b++) {
 713                 switch (*b) {
 714                 case 'r':
 715                         ex.access |= ACC_READ;
 716                         break;
 717                 case 'w':
 718                         ex.access |= ACC_WRITE;
 719                         break;
 720                 case 'm':
 721                         ex.access |= ACC_MKNOD;
 722                         break;
 723                 case '\n':
 724                 case '\0':
 725                         count = 3;
 726                         break;
 727                 default:
 728                         return -EINVAL;
 729                 }
 730         }
 731
 732         switch (filetype) {
 733         case DEVCG_ALLOW:
 734                 /*
 735                  * If the default policy is to allow by default, try to remove
 736                  * an matching exception instead. And be silent about it: we
 737                  * don't want to break compatibility
 738                  */
 739                 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 740                         /* Check if the parent allows removing it first */
 741                         if (!parent_allows_removal(devcgroup, &ex))
 742                                 return -EPERM;
 743                         dev_exception_rm(devcgroup, &ex);
 744                         break;
 745                 }
 746
 747                 if (!parent_has_perm(devcgroup, &ex))
 748                         return -EPERM;
 749                 rc = dev_exception_add(devcgroup, &ex);
 750                 break;
 751         case DEVCG_DENY:
 752                 /*
 753                  * If the default policy is to deny by default, try to remove
 754                  * an matching exception instead. And be silent about it: we
 755                  * don't want to break compatibility
 756                  */
 757                 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
 758                         dev_exception_rm(devcgroup, &ex);
 759                 else
 760                         rc = dev_exception_add(devcgroup, &ex);
 761
 762                 if (rc)
 763                         break;
 764                 /* we only propagate new restrictions */
 765                 rc = propagate_exception(devcgroup, &ex);
 766                 break;
 767         default:
 768                 rc = -EINVAL;
 769         }
 770         return rc;
 771 }
 772
 773 static int devcgroup_access_write(struct cgroup_subsys_state *css,
 774                                   struct cftype *cft, const char *buffer)
 775 {
 776         int retval;
 777
 778         mutex_lock(&devcgroup_mutex);
 779         retval = devcgroup_update_access(css_to_devcgroup(css),
 780                                          cft->private, buffer);
 781         mutex_unlock(&devcgroup_mutex);
 782         return retval;
 783 }
 784
 785 static struct cftype dev_cgroup_files[] = {
 786         {
 787                 .name = "allow",
 788                 .write_string  = devcgroup_access_write,
 789                 .private = DEVCG_ALLOW,
 790         },
 791         {
 792                 .name = "deny",
 793                 .write_string = devcgroup_access_write,
 794                 .private = DEVCG_DENY,
 795         },
 796         {
 797                 .name = "list",
 798                 .seq_show = devcgroup_seq_show,
 799                 .private = DEVCG_LIST,
 800         },
 801         { }     /* terminate */
 802 };
 803
 804 struct cgroup_subsys devices_subsys = {
 805         .name = "devices",
 806         .css_alloc = devcgroup_css_alloc,
 807         .css_free = devcgroup_css_free,
 808         .css_online = devcgroup_online,
 809         .css_offline = devcgroup_offline,
 810         .subsys_id = devices_subsys_id,
 811         .base_cftypes = dev_cgroup_files,
 812 };
 813
 814 /**
 815  * __devcgroup_check_permission - checks if an inode operation is permitted
 816  * @dev_cgroup: the dev cgroup to be tested against
 817  * @type: device type
 818  * @major: device major number
 819  * @minor: device minor number
 820  * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
 821  *
 822  * returns 0 on success, -EPERM case the operation is not permitted
 823  */
 824 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
 825                                         short access)
 826 {
 827         struct dev_cgroup *dev_cgroup;
 828         bool rc;
 829
 830         rcu_read_lock();
 831         dev_cgroup = task_devcgroup(current);
 832         if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
 833                 /* Can't match any of the exceptions, even partially */
 834                 rc = !match_exception_partial(&dev_cgroup->exceptions,
 835                                               type, major, minor, access);
 836         else
 837                 /* Need to match completely one exception to be allowed */
 838                 rc = match_exception(&dev_cgroup->exceptions, type, major,
 839                                      minor, access);
 840         rcu_read_unlock();
 841
 842         if (!rc)
 843                 return -EPERM;
 844
 845         return 0;
 846 }
 847
 848 int __devcgroup_inode_permission(struct inode *inode, int mask)
 849 {
 850         short type, access = 0;
 851
 852         if (S_ISBLK(inode->i_mode))
 853                 type = DEV_BLOCK;
 854         if (S_ISCHR(inode->i_mode))
 855                 type = DEV_CHAR;
 856         if (mask & MAY_WRITE)
 857                 access |= ACC_WRITE;
 858         if (mask & MAY_READ)
 859                 access |= ACC_READ;
 860
 861         return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
 862                         access);
 863 }
 864
 865 int devcgroup_inode_mknod(int mode, dev_t dev)
 866 {
 867         short type;
 868
 869         if (!S_ISBLK(mode) && !S_ISCHR(mode))
 870                 return 0;
 871
 872         if (S_ISBLK(mode))
 873                 type = DEV_BLOCK;
 874         else
 875                 type = DEV_CHAR;
 876
 877         return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),
 878                         ACC_MKNOD);
 879
 880 }