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 static int devcgroup_can_attach(struct cgroup_subsys_state *new_css,
  67                                 struct cgroup_taskset *set)
  68 {
  69         struct task_struct *task = cgroup_taskset_first(set);
  70
  71         if (current != task && !capable(CAP_SYS_ADMIN))
  72                 return -EPERM;
  73         return 0;
  74 }
  75
  76 /*
  77  * called under devcgroup_mutex
  78  */
  79 static int dev_exceptions_copy(struct list_head *dest, struct list_head *orig)
  80 {
  81         struct dev_exception_item *ex, *tmp, *new;
  82
  83         lockdep_assert_held(&devcgroup_mutex);
  84
  85         list_for_each_entry(ex, orig, list) {
  86                 new = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
  87                 if (!new)
  88                         goto free_and_exit;
  89                 list_add_tail(&new->list, dest);
  90         }
  91
  92         return 0;
  93
  94 free_and_exit:
  95         list_for_each_entry_safe(ex, tmp, dest, list) {
  96                 list_del(&ex->list);
  97                 kfree(ex);
  98         }
  99         return -ENOMEM;
 100 }
 101
 102 /*
 103  * called under devcgroup_mutex
 104  */
 105 static int dev_exception_add(struct dev_cgroup *dev_cgroup,
 106                              struct dev_exception_item *ex)
 107 {
 108         struct dev_exception_item *excopy, *walk;
 109
 110         lockdep_assert_held(&devcgroup_mutex);
 111
 112         excopy = kmemdup(ex, sizeof(*ex), GFP_KERNEL);
 113         if (!excopy)
 114                 return -ENOMEM;
 115
 116         list_for_each_entry(walk, &dev_cgroup->exceptions, list) {
 117                 if (walk->type != ex->type)
 118                         continue;
 119                 if (walk->major != ex->major)
 120                         continue;
 121                 if (walk->minor != ex->minor)
 122                         continue;
 123
 124                 walk->access |= ex->access;
 125                 kfree(excopy);
 126                 excopy = NULL;
 127         }
 128
 129         if (excopy != NULL)
 130                 list_add_tail_rcu(&excopy->list, &dev_cgroup->exceptions);
 131         return 0;
 132 }
 133
 134 /*
 135  * called under devcgroup_mutex
 136  */
 137 static void dev_exception_rm(struct dev_cgroup *dev_cgroup,
 138                              struct dev_exception_item *ex)
 139 {
 140         struct dev_exception_item *walk, *tmp;
 141
 142         lockdep_assert_held(&devcgroup_mutex);
 143
 144         list_for_each_entry_safe(walk, tmp, &dev_cgroup->exceptions, list) {
 145                 if (walk->type != ex->type)
 146                         continue;
 147                 if (walk->major != ex->major)
 148                         continue;
 149                 if (walk->minor != ex->minor)
 150                         continue;
 151
 152                 walk->access &= ~ex->access;
 153                 if (!walk->access) {
 154                         list_del_rcu(&walk->list);
 155                         kfree_rcu(walk, rcu);
 156                 }
 157         }
 158 }
 159
 160 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
 161 {
 162         struct dev_exception_item *ex, *tmp;
 163
 164         list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
 165                 list_del_rcu(&ex->list);
 166                 kfree_rcu(ex, rcu);
 167         }
 168 }
 169
 170 /**
 171  * dev_exception_clean - frees all entries of the exception list
 172  * @dev_cgroup: dev_cgroup with the exception list to be cleaned
 173  *
 174  * called under devcgroup_mutex
 175  */
 176 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
 177 {
 178         lockdep_assert_held(&devcgroup_mutex);
 179
 180         __dev_exception_clean(dev_cgroup);
 181 }
 182
 183 static inline bool is_devcg_online(const struct dev_cgroup *devcg)
 184 {
 185         return (devcg->behavior != DEVCG_DEFAULT_NONE);
 186 }
 187
 188 /**
 189  * devcgroup_online - initializes devcgroup's behavior and exceptions based on
 190  *                    parent's
 191  * @css: css getting online
 192  * returns 0 in case of success, error code otherwise
 193  */
 194 static int devcgroup_online(struct cgroup_subsys_state *css)
 195 {
 196         struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
 197         struct dev_cgroup *parent_dev_cgroup = css_to_devcgroup(css_parent(css));
 198         int ret = 0;
 199
 200         mutex_lock(&devcgroup_mutex);
 201
 202         if (parent_dev_cgroup == NULL)
 203                 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
 204         else {
 205                 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
 206                                           &parent_dev_cgroup->exceptions);
 207                 if (!ret)
 208                         dev_cgroup->behavior = parent_dev_cgroup->behavior;
 209         }
 210         mutex_unlock(&devcgroup_mutex);
 211
 212         return ret;
 213 }
 214
 215 static void devcgroup_offline(struct cgroup_subsys_state *css)
 216 {
 217         struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
 218
 219         mutex_lock(&devcgroup_mutex);
 220         dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
 221         mutex_unlock(&devcgroup_mutex);
 222 }
 223
 224 /*
 225  * called from kernel/cgroup.c with cgroup_lock() held.
 226  */
 227 static struct cgroup_subsys_state *
 228 devcgroup_css_alloc(struct cgroup_subsys_state *parent_css)
 229 {
 230         struct dev_cgroup *dev_cgroup;
 231
 232         dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
 233         if (!dev_cgroup)
 234                 return ERR_PTR(-ENOMEM);
 235         INIT_LIST_HEAD(&dev_cgroup->exceptions);
 236         dev_cgroup->behavior = DEVCG_DEFAULT_NONE;
 237
 238         return &dev_cgroup->css;
 239 }
 240
 241 static void devcgroup_css_free(struct cgroup_subsys_state *css)
 242 {
 243         struct dev_cgroup *dev_cgroup = css_to_devcgroup(css);
 244
 245         __dev_exception_clean(dev_cgroup);
 246         kfree(dev_cgroup);
 247 }
 248
 249 #define DEVCG_ALLOW 1
 250 #define DEVCG_DENY 2
 251 #define DEVCG_LIST 3
 252
 253 #define MAJMINLEN 13
 254 #define ACCLEN 4
 255
 256 static void set_access(char *acc, short access)
 257 {
 258         int idx = 0;
 259         memset(acc, 0, ACCLEN);
 260         if (access & ACC_READ)
 261                 acc[idx++] = 'r';
 262         if (access & ACC_WRITE)
 263                 acc[idx++] = 'w';
 264         if (access & ACC_MKNOD)
 265                 acc[idx++] = 'm';
 266 }
 267
 268 static char type_to_char(short type)
 269 {
 270         if (type == DEV_ALL)
 271                 return 'a';
 272         if (type == DEV_CHAR)
 273                 return 'c';
 274         if (type == DEV_BLOCK)
 275                 return 'b';
 276         return 'X';
 277 }
 278
 279 static void set_majmin(char *str, unsigned m)
 280 {
 281         if (m == ~0)
 282                 strcpy(str, "*");
 283         else
 284                 sprintf(str, "%u", m);
 285 }
 286
 287 static int devcgroup_seq_read(struct cgroup_subsys_state *css,
 288                               struct cftype *cft, struct seq_file *m)
 289 {
 290         struct dev_cgroup *devcgroup = css_to_devcgroup(css);
 291         struct dev_exception_item *ex;
 292         char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
 293
 294         rcu_read_lock();
 295         /*
 296          * To preserve the compatibility:
 297          * - Only show the "all devices" when the default policy is to allow
 298          * - List the exceptions in case the default policy is to deny
 299          * This way, the file remains as a "whitelist of devices"
 300          */
 301         if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 302                 set_access(acc, ACC_MASK);
 303                 set_majmin(maj, ~0);
 304                 set_majmin(min, ~0);
 305                 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
 306                            maj, min, acc);
 307         } else {
 308                 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
 309                         set_access(acc, ex->access);
 310                         set_majmin(maj, ex->major);
 311                         set_majmin(min, ex->minor);
 312                         seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
 313                                    maj, min, acc);
 314                 }
 315         }
 316         rcu_read_unlock();
 317
 318         return 0;
 319 }
 320
 321 /**
 322  * match_exception      - iterates the exception list trying to match a rule
 323  *                        based on type, major, minor and access type. It is
 324  *                        considered a match if an exception is found that
 325  *                        will contain the entire range of provided parameters.
 326  * @exceptions: list of exceptions
 327  * @type: device type (DEV_BLOCK or DEV_CHAR)
 328  * @major: device file major number, ~0 to match all
 329  * @minor: device file minor number, ~0 to match all
 330  * @access: permission mask (ACC_READ, ACC_WRITE, ACC_MKNOD)
 331  *
 332  * returns: true in case it matches an exception completely
 333  */
 334 static bool match_exception(struct list_head *exceptions, short type,
 335                             u32 major, u32 minor, short access)
 336 {
 337         struct dev_exception_item *ex;
 338
 339         list_for_each_entry_rcu(ex, exceptions, list) {
 340                 if ((type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
 341                         continue;
 342                 if ((type & DEV_CHAR) && !(ex->type & DEV_CHAR))
 343                         continue;
 344                 if (ex->major != ~0 && ex->major != major)
 345                         continue;
 346                 if (ex->minor != ~0 && ex->minor != minor)
 347                         continue;
 348                 /* provided access cannot have more than the exception rule */
 349                 if (access & (~ex->access))
 350                         continue;
 351                 return true;
 352         }
 353         return false;
 354 }
 355
 356 /**
 357  * match_exception_partial - iterates the exception list trying to match a rule
 358  *                           based on type, major, minor and access type. It is
 359  *                           considered a match if an exception's range is
 360  *                           found to contain *any* of the devices specified by
 361  *                           provided parameters. This is used to make sure no
 362  *                           extra access is being granted that is forbidden by
 363  *                           any of the exception list.
 364  * @exceptions: list of exceptions
 365  * @type: device type (DEV_BLOCK or DEV_CHAR)
 366  * @major: device file major number, ~0 to match all
 367  * @minor: device file minor number, ~0 to match all
 368  * @access: permission mask (ACC_READ, ACC_WRITE, ACC_MKNOD)
 369  *
 370  * returns: true in case the provided range mat matches an exception completely
 371  */
 372 static bool match_exception_partial(struct list_head *exceptions, short type,
 373                                     u32 major, u32 minor, short access)
 374 {
 375         struct dev_exception_item *ex;
 376
 377         list_for_each_entry_rcu(ex, exceptions, list) {
 378                 if ((type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
 379                         continue;
 380                 if ((type & DEV_CHAR) && !(ex->type & DEV_CHAR))
 381                         continue;
 382                 /*
 383                  * We must be sure that both the exception and the provided
 384                  * range aren't masking all devices
 385                  */
 386                 if (ex->major != ~0 && major != ~0 && ex->major != major)
 387                         continue;
 388                 if (ex->minor != ~0 && minor != ~0 && ex->minor != minor)
 389                         continue;
 390                 /*
 391                  * In order to make sure the provided range isn't matching
 392                  * an exception, all its access bits shouldn't match the
 393                  * exception's access bits
 394                  */
 395                 if (!(access & ex->access))
 396                         continue;
 397                 return true;
 398         }
 399         return false;
 400 }
 401
 402 /**
 403  * verify_new_ex - verifies if a new exception is part of what is allowed
 404  *                 by a dev cgroup based on the default policy +
 405  *                 exceptions. This is used to make sure a child cgroup
 406  *                 won't have more privileges than its parent
 407  * @dev_cgroup: dev cgroup to be tested against
 408  * @refex: new exception
 409  * @behavior: behavior of the exception's dev_cgroup
 410  */
 411 static bool verify_new_ex(struct dev_cgroup *dev_cgroup,
 412                           struct dev_exception_item *refex,
 413                           enum devcg_behavior behavior)
 414 {
 415         bool match = false;
 416
 417         rcu_lockdep_assert(rcu_read_lock_held() ||
 418                            lockdep_is_held(&devcgroup_mutex),
 419                            "device_cgroup:verify_new_ex called without proper synchronization");
 420
 421         if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 422                 if (behavior == DEVCG_DEFAULT_ALLOW) {
 423                         /*
 424                          * new exception in the child doesn't matter, only
 425                          * adding extra restrictions
 426                          */
 427                         return true;
 428                 } else {
 429                         /*
 430                          * new exception in the child will add more devices
 431                          * that can be acessed, so it can't match any of
 432                          * parent's exceptions, even slightly
 433                          */
 434                         match = match_exception_partial(&dev_cgroup->exceptions,
 435                                                         refex->type,
 436                                                         refex->major,
 437                                                         refex->minor,
 438                                                         refex->access);
 439
 440                         if (match)
 441                                 return false;
 442                         return true;
 443                 }
 444         } else {
 445                 /*
 446                  * Only behavior == DEVCG_DEFAULT_DENY allowed here, therefore
 447                  * the new exception will add access to more devices and must
 448                  * be contained completely in an parent's exception to be
 449                  * allowed
 450                  */
 451                 match = match_exception(&dev_cgroup->exceptions, refex->type,
 452                                         refex->major, refex->minor,
 453                                         refex->access);
 454
 455                 if (match)
 456                         /* parent has an exception that matches the proposed */
 457                         return true;
 458                 else
 459                         return false;
 460         }
 461         return false;
 462 }
 463
 464 /*
 465  * parent_has_perm:
 466  * when adding a new allow rule to a device exception list, the rule
 467  * must be allowed in the parent device
 468  */
 469 static int parent_has_perm(struct dev_cgroup *childcg,
 470                                   struct dev_exception_item *ex)
 471 {
 472         struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css));
 473
 474         if (!parent)
 475                 return 1;
 476         return verify_new_ex(parent, ex, childcg->behavior);
 477 }
 478
 479 /**
 480  * parent_allows_removal - verify if it's ok to remove an exception
 481  * @childcg: child cgroup from where the exception will be removed
 482  * @ex: exception being removed
 483  *
 484  * When removing an exception in cgroups with default ALLOW policy, it must
 485  * be checked if removing it will give the child cgroup more access than the
 486  * parent.
 487  *
 488  * Return: true if it's ok to remove exception, false otherwise
 489  */
 490 static bool parent_allows_removal(struct dev_cgroup *childcg,
 491                                   struct dev_exception_item *ex)
 492 {
 493         struct dev_cgroup *parent = css_to_devcgroup(css_parent(&childcg->css));
 494
 495         if (!parent)
 496                 return true;
 497
 498         /* It's always allowed to remove access to devices */
 499         if (childcg->behavior == DEVCG_DEFAULT_DENY)
 500                 return true;
 501
 502         /*
 503          * Make sure you're not removing part or a whole exception existing in
 504          * the parent cgroup
 505          */
 506         return !match_exception_partial(&parent->exceptions, ex->type,
 507                                         ex->major, ex->minor, ex->access);
 508 }
 509
 510 /**
 511  * may_allow_all - checks if it's possible to change the behavior to
 512  *                 allow based on parent's rules.
 513  * @parent: device cgroup's parent
 514  * returns: != 0 in case it's allowed, 0 otherwise
 515  */
 516 static inline int may_allow_all(struct dev_cgroup *parent)
 517 {
 518         if (!parent)
 519                 return 1;
 520         return parent->behavior == DEVCG_DEFAULT_ALLOW;
 521 }
 522
 523 /**
 524  * revalidate_active_exceptions - walks through the active exception list and
 525  *                                revalidates the exceptions based on parent's
 526  *                                behavior and exceptions. The exceptions that
 527  *                                are no longer valid will be removed.
 528  *                                Called with devcgroup_mutex held.
 529  * @devcg: cgroup which exceptions will be checked
 530  *
 531  * This is one of the three key functions for hierarchy implementation.
 532  * This function is responsible for re-evaluating all the cgroup's active
 533  * exceptions due to a parent's exception change.
 534  * Refer to Documentation/cgroups/devices.txt for more details.
 535  */
 536 static void revalidate_active_exceptions(struct dev_cgroup *devcg)
 537 {
 538         struct dev_exception_item *ex;
 539         struct list_head *this, *tmp;
 540
 541         list_for_each_safe(this, tmp, &devcg->exceptions) {
 542                 ex = container_of(this, struct dev_exception_item, list);
 543                 if (!parent_has_perm(devcg, ex))
 544                         dev_exception_rm(devcg, ex);
 545         }
 546 }
 547
 548 /**
 549  * propagate_exception - propagates a new exception to the children
 550  * @devcg_root: device cgroup that added a new exception
 551  * @ex: new exception to be propagated
 552  *
 553  * returns: 0 in case of success, != 0 in case of error
 554  */
 555 static int propagate_exception(struct dev_cgroup *devcg_root,
 556                                struct dev_exception_item *ex)
 557 {
 558         struct cgroup_subsys_state *pos;
 559         int rc = 0;
 560
 561         rcu_read_lock();
 562
 563         css_for_each_descendant_pre(pos, &devcg_root->css) {
 564                 struct dev_cgroup *devcg = css_to_devcgroup(pos);
 565
 566                 /*
 567                  * Because devcgroup_mutex is held, no devcg will become
 568                  * online or offline during the tree walk (see on/offline
 569                  * methods), and online ones are safe to access outside RCU
 570                  * read lock without bumping refcnt.
 571                  */
 572                 if (pos == &devcg_root->css || !is_devcg_online(devcg))
 573                         continue;
 574
 575                 rcu_read_unlock();
 576
 577                 /*
 578                  * in case both root's behavior and devcg is allow, a new
 579                  * restriction means adding to the exception list
 580                  */
 581                 if (devcg_root->behavior == DEVCG_DEFAULT_ALLOW &&
 582                     devcg->behavior == DEVCG_DEFAULT_ALLOW) {
 583                         rc = dev_exception_add(devcg, ex);
 584                         if (rc)
 585                                 break;
 586                 } else {
 587                         /*
 588                          * in the other possible cases:
 589                          * root's behavior: allow, devcg's: deny
 590                          * root's behavior: deny, devcg's: deny
 591                          * the exception will be removed
 592                          */
 593                         dev_exception_rm(devcg, ex);
 594                 }
 595                 revalidate_active_exceptions(devcg);
 596
 597                 rcu_read_lock();
 598         }
 599
 600         rcu_read_unlock();
 601         return rc;
 602 }
 603
 604 static inline bool has_children(struct dev_cgroup *devcgroup)
 605 {
 606         struct cgroup *cgrp = devcgroup->css.cgroup;
 607
 608         return !list_empty(&cgrp->children);
 609 }
 610
 611 /*
 612  * Modify the exception list using allow/deny rules.
 613  * CAP_SYS_ADMIN is needed for this.  It's at least separate from CAP_MKNOD
 614  * so we can give a container CAP_MKNOD to let it create devices but not
 615  * modify the exception list.
 616  * It seems likely we'll want to add a CAP_CONTAINER capability to allow
 617  * us to also grant CAP_SYS_ADMIN to containers without giving away the
 618  * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
 619  *
 620  * Taking rules away is always allowed (given CAP_SYS_ADMIN).  Granting
 621  * new access is only allowed if you're in the top-level cgroup, or your
 622  * parent cgroup has the access you're asking for.
 623  */
 624 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
 625                                    int filetype, const char *buffer)
 626 {
 627         const char *b;
 628         char temp[12];          /* 11 + 1 characters needed for a u32 */
 629         int count, rc = 0;
 630         struct dev_exception_item ex;
 631         struct dev_cgroup *parent = css_to_devcgroup(css_parent(&devcgroup->css));
 632
 633         if (!capable(CAP_SYS_ADMIN))
 634                 return -EPERM;
 635
 636         memset(&ex, 0, sizeof(ex));
 637         b = buffer;
 638
 639         switch (*b) {
 640         case 'a':
 641                 switch (filetype) {
 642                 case DEVCG_ALLOW:
 643                         if (has_children(devcgroup))
 644                                 return -EINVAL;
 645
 646                         if (!may_allow_all(parent))
 647                                 return -EPERM;
 648                         dev_exception_clean(devcgroup);
 649                         devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
 650                         if (!parent)
 651                                 break;
 652
 653                         rc = dev_exceptions_copy(&devcgroup->exceptions,
 654                                                  &parent->exceptions);
 655                         if (rc)
 656                                 return rc;
 657                         break;
 658                 case DEVCG_DENY:
 659                         if (has_children(devcgroup))
 660                                 return -EINVAL;
 661
 662                         dev_exception_clean(devcgroup);
 663                         devcgroup->behavior = DEVCG_DEFAULT_DENY;
 664                         break;
 665                 default:
 666                         return -EINVAL;
 667                 }
 668                 return 0;
 669         case 'b':
 670                 ex.type = DEV_BLOCK;
 671                 break;
 672         case 'c':
 673                 ex.type = DEV_CHAR;
 674                 break;
 675         default:
 676                 return -EINVAL;
 677         }
 678         b++;
 679         if (!isspace(*b))
 680                 return -EINVAL;
 681         b++;
 682         if (*b == '*') {
 683                 ex.major = ~0;
 684                 b++;
 685         } else if (isdigit(*b)) {
 686                 memset(temp, 0, sizeof(temp));
 687                 for (count = 0; count < sizeof(temp) - 1; count++) {
 688                         temp[count] = *b;
 689                         b++;
 690                         if (!isdigit(*b))
 691                                 break;
 692                 }
 693                 rc = kstrtou32(temp, 10, &ex.major);
 694                 if (rc)
 695                         return -EINVAL;
 696         } else {
 697                 return -EINVAL;
 698         }
 699         if (*b != ':')
 700                 return -EINVAL;
 701         b++;
 702
 703         /* read minor */
 704         if (*b == '*') {
 705                 ex.minor = ~0;
 706                 b++;
 707         } else if (isdigit(*b)) {
 708                 memset(temp, 0, sizeof(temp));
 709                 for (count = 0; count < sizeof(temp) - 1; count++) {
 710                         temp[count] = *b;
 711                         b++;
 712                         if (!isdigit(*b))
 713                                 break;
 714                 }
 715                 rc = kstrtou32(temp, 10, &ex.minor);
 716                 if (rc)
 717                         return -EINVAL;
 718         } else {
 719                 return -EINVAL;
 720         }
 721         if (!isspace(*b))
 722                 return -EINVAL;
 723         for (b++, count = 0; count < 3; count++, b++) {
 724                 switch (*b) {
 725                 case 'r':
 726                         ex.access |= ACC_READ;
 727                         break;
 728                 case 'w':
 729                         ex.access |= ACC_WRITE;
 730                         break;
 731                 case 'm':
 732                         ex.access |= ACC_MKNOD;
 733                         break;
 734                 case '\n':
 735                 case '\0':
 736                         count = 3;
 737                         break;
 738                 default:
 739                         return -EINVAL;
 740                 }
 741         }
 742
 743         switch (filetype) {
 744         case DEVCG_ALLOW:
 745                 /*
 746                  * If the default policy is to allow by default, try to remove
 747                  * an matching exception instead. And be silent about it: we
 748                  * don't want to break compatibility
 749                  */
 750                 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 751                         /* Check if the parent allows removing it first */
 752                         if (!parent_allows_removal(devcgroup, &ex))
 753                                 return -EPERM;
 754                         dev_exception_rm(devcgroup, &ex);
 755                         break;
 756                 }
 757
 758                 if (!parent_has_perm(devcgroup, &ex))
 759                         return -EPERM;
 760                 rc = dev_exception_add(devcgroup, &ex);
 761                 break;
 762         case DEVCG_DENY:
 763                 /*
 764                  * If the default policy is to deny by default, try to remove
 765                  * an matching exception instead. And be silent about it: we
 766                  * don't want to break compatibility
 767                  */
 768                 if (devcgroup->behavior == DEVCG_DEFAULT_DENY)
 769                         dev_exception_rm(devcgroup, &ex);
 770                 else
 771                         rc = dev_exception_add(devcgroup, &ex);
 772
 773                 if (rc)
 774                         break;
 775                 /* we only propagate new restrictions */
 776                 rc = propagate_exception(devcgroup, &ex);
 777                 break;
 778         default:
 779                 rc = -EINVAL;
 780         }
 781         return rc;
 782 }
 783
 784 static int devcgroup_access_write(struct cgroup_subsys_state *css,
 785                                   struct cftype *cft, const char *buffer)
 786 {
 787         int retval;
 788
 789         mutex_lock(&devcgroup_mutex);
 790         retval = devcgroup_update_access(css_to_devcgroup(css),
 791                                          cft->private, buffer);
 792         mutex_unlock(&devcgroup_mutex);
 793         return retval;
 794 }
 795
 796 static struct cftype dev_cgroup_files[] = {
 797         {
 798                 .name = "allow",
 799                 .write_string  = devcgroup_access_write,
 800                 .private = DEVCG_ALLOW,
 801         },
 802         {
 803                 .name = "deny",
 804                 .write_string = devcgroup_access_write,
 805                 .private = DEVCG_DENY,
 806         },
 807         {
 808                 .name = "list",
 809                 .read_seq_string = devcgroup_seq_read,
 810                 .private = DEVCG_LIST,
 811         },
 812         { }     /* terminate */
 813 };
 814
 815 struct cgroup_subsys devices_subsys = {
 816         .name = "devices",
 817         .can_attach = devcgroup_can_attach,
 818         .css_alloc = devcgroup_css_alloc,
 819         .css_free = devcgroup_css_free,
 820         .css_online = devcgroup_online,
 821         .css_offline = devcgroup_offline,
 822         .subsys_id = devices_subsys_id,
 823         .base_cftypes = dev_cgroup_files,
 824 };
 825
 826 /**
 827  * __devcgroup_check_permission - checks if an inode operation is permitted
 828  * @dev_cgroup: the dev cgroup to be tested against
 829  * @type: device type
 830  * @major: device major number
 831  * @minor: device minor number
 832  * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
 833  *
 834  * returns 0 on success, -EPERM case the operation is not permitted
 835  */
 836 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
 837                                         short access)
 838 {
 839         struct dev_cgroup *dev_cgroup;
 840         bool rc;
 841
 842         rcu_read_lock();
 843         dev_cgroup = task_devcgroup(current);
 844         if (dev_cgroup->behavior == DEVCG_DEFAULT_ALLOW)
 845                 /* Can't match any of the exceptions, even partially */
 846                 rc = !match_exception_partial(&dev_cgroup->exceptions,
 847                                               type, major, minor, access);
 848         else
 849                 /* Need to match completely one exception to be allowed */
 850                 rc = match_exception(&dev_cgroup->exceptions, type, major,
 851                                      minor, access);
 852         rcu_read_unlock();
 853
 854         if (!rc)
 855                 return -EPERM;
 856
 857         return 0;
 858 }
 859
 860 int __devcgroup_inode_permission(struct inode *inode, int mask)
 861 {
 862         short type, access = 0;
 863
 864         if (S_ISBLK(inode->i_mode))
 865                 type = DEV_BLOCK;
 866         if (S_ISCHR(inode->i_mode))
 867                 type = DEV_CHAR;
 868         if (mask & MAY_WRITE)
 869                 access |= ACC_WRITE;
 870         if (mask & MAY_READ)
 871                 access |= ACC_READ;
 872
 873         return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
 874                         access);
 875 }
 876
 877 int devcgroup_inode_mknod(int mode, dev_t dev)
 878 {
 879         short type;
 880
 881         if (!S_ISBLK(mode) && !S_ISCHR(mode))
 882                 return 0;
 883
 884         if (S_ISBLK(mode))
 885                 type = DEV_BLOCK;
 886         else
 887                 type = DEV_CHAR;
 888
 889         return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),
 890                         ACC_MKNOD);
 891
 892 }