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