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 static void __dev_exception_clean(struct dev_cgroup *dev_cgroup)
 163 {
 164         struct dev_exception_item *ex, *tmp;
 165
 166         list_for_each_entry_safe(ex, tmp, &dev_cgroup->exceptions, list) {
 167                 list_del_rcu(&ex->list);
 168                 kfree_rcu(ex, rcu);
 169         }
 170 }
 171
 172 /**
 173  * dev_exception_clean - frees all entries of the exception list
 174  * @dev_cgroup: dev_cgroup with the exception list to be cleaned
 175  *
 176  * called under devcgroup_mutex
 177  */
 178 static void dev_exception_clean(struct dev_cgroup *dev_cgroup)
 179 {
 180         lockdep_assert_held(&devcgroup_mutex);
 181
 182         __dev_exception_clean(dev_cgroup);
 183 }
 184
 185 /*
 186  * called from kernel/cgroup.c with cgroup_lock() held.
 187  */
 188 static struct cgroup_subsys_state *devcgroup_css_alloc(struct cgroup *cgroup)
 189 {
 190         struct dev_cgroup *dev_cgroup, *parent_dev_cgroup;
 191         struct cgroup *parent_cgroup;
 192         int ret;
 193
 194         dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
 195         if (!dev_cgroup)
 196                 return ERR_PTR(-ENOMEM);
 197         INIT_LIST_HEAD(&dev_cgroup->exceptions);
 198         parent_cgroup = cgroup->parent;
 199
 200         if (parent_cgroup == NULL)
 201                 dev_cgroup->behavior = DEVCG_DEFAULT_ALLOW;
 202         else {
 203                 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
 204                 mutex_lock(&devcgroup_mutex);
 205                 ret = dev_exceptions_copy(&dev_cgroup->exceptions,
 206                                           &parent_dev_cgroup->exceptions);
 207                 dev_cgroup->behavior = parent_dev_cgroup->behavior;
 208                 mutex_unlock(&devcgroup_mutex);
 209                 if (ret) {
 210                         kfree(dev_cgroup);
 211                         return ERR_PTR(ret);
 212                 }
 213         }
 214
 215         return &dev_cgroup->css;
 216 }
 217
 218 static void devcgroup_css_free(struct cgroup *cgroup)
 219 {
 220         struct dev_cgroup *dev_cgroup;
 221
 222         dev_cgroup = cgroup_to_devcgroup(cgroup);
 223         __dev_exception_clean(dev_cgroup);
 224         kfree(dev_cgroup);
 225 }
 226
 227 #define DEVCG_ALLOW 1
 228 #define DEVCG_DENY 2
 229 #define DEVCG_LIST 3
 230
 231 #define MAJMINLEN 13
 232 #define ACCLEN 4
 233
 234 static void set_access(char *acc, short access)
 235 {
 236         int idx = 0;
 237         memset(acc, 0, ACCLEN);
 238         if (access & ACC_READ)
 239                 acc[idx++] = 'r';
 240         if (access & ACC_WRITE)
 241                 acc[idx++] = 'w';
 242         if (access & ACC_MKNOD)
 243                 acc[idx++] = 'm';
 244 }
 245
 246 static char type_to_char(short type)
 247 {
 248         if (type == DEV_ALL)
 249                 return 'a';
 250         if (type == DEV_CHAR)
 251                 return 'c';
 252         if (type == DEV_BLOCK)
 253                 return 'b';
 254         return 'X';
 255 }
 256
 257 static void set_majmin(char *str, unsigned m)
 258 {
 259         if (m == ~0)
 260                 strcpy(str, "*");
 261         else
 262                 sprintf(str, "%u", m);
 263 }
 264
 265 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
 266                                 struct seq_file *m)
 267 {
 268         struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
 269         struct dev_exception_item *ex;
 270         char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
 271
 272         rcu_read_lock();
 273         /*
 274          * To preserve the compatibility:
 275          * - Only show the "all devices" when the default policy is to allow
 276          * - List the exceptions in case the default policy is to deny
 277          * This way, the file remains as a "whitelist of devices"
 278          */
 279         if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 280                 set_access(acc, ACC_MASK);
 281                 set_majmin(maj, ~0);
 282                 set_majmin(min, ~0);
 283                 seq_printf(m, "%c %s:%s %s\n", type_to_char(DEV_ALL),
 284                            maj, min, acc);
 285         } else {
 286                 list_for_each_entry_rcu(ex, &devcgroup->exceptions, list) {
 287                         set_access(acc, ex->access);
 288                         set_majmin(maj, ex->major);
 289                         set_majmin(min, ex->minor);
 290                         seq_printf(m, "%c %s:%s %s\n", type_to_char(ex->type),
 291                                    maj, min, acc);
 292                 }
 293         }
 294         rcu_read_unlock();
 295
 296         return 0;
 297 }
 298
 299 /**
 300  * may_access - verifies if a new exception is part of what is allowed
 301  *              by a dev cgroup based on the default policy +
 302  *              exceptions. This is used to make sure a child cgroup
 303  *              won't have more privileges than its parent or to
 304  *              verify if a certain access is allowed.
 305  * @dev_cgroup: dev cgroup to be tested against
 306  * @refex: new exception
 307  */
 308 static int may_access(struct dev_cgroup *dev_cgroup,
 309                       struct dev_exception_item *refex)
 310 {
 311         struct dev_exception_item *ex;
 312         bool match = false;
 313
 314         rcu_lockdep_assert(rcu_read_lock_held() ||
 315                            lockdep_is_held(&devcgroup_mutex),
 316                            "device_cgroup::may_access() called without proper synchronization");
 317
 318         list_for_each_entry_rcu(ex, &dev_cgroup->exceptions, list) {
 319                 if ((refex->type & DEV_BLOCK) && !(ex->type & DEV_BLOCK))
 320                         continue;
 321                 if ((refex->type & DEV_CHAR) && !(ex->type & DEV_CHAR))
 322                         continue;
 323                 if (ex->major != ~0 && ex->major != refex->major)
 324                         continue;
 325                 if (ex->minor != ~0 && ex->minor != refex->minor)
 326                         continue;
 327                 if (refex->access & (~ex->access))
 328                         continue;
 329                 match = true;
 330                 break;
 331         }
 332
 333         /*
 334          * In two cases we'll consider this new exception valid:
 335          * - the dev cgroup has its default policy to allow + exception list:
 336          *   the new exception should *not* match any of the exceptions
 337          *   (behavior == DEVCG_DEFAULT_ALLOW, !match)
 338          * - the dev cgroup has its default policy to deny + exception list:
 339          *   the new exception *should* match the exceptions
 340          *   (behavior == DEVCG_DEFAULT_DENY, match)
 341          */
 342         if ((dev_cgroup->behavior == DEVCG_DEFAULT_DENY) == match)
 343                 return 1;
 344         return 0;
 345 }
 346
 347 /*
 348  * parent_has_perm:
 349  * when adding a new allow rule to a device exception list, the rule
 350  * must be allowed in the parent device
 351  */
 352 static int parent_has_perm(struct dev_cgroup *childcg,
 353                                   struct dev_exception_item *ex)
 354 {
 355         struct cgroup *pcg = childcg->css.cgroup->parent;
 356         struct dev_cgroup *parent;
 357
 358         if (!pcg)
 359                 return 1;
 360         parent = cgroup_to_devcgroup(pcg);
 361         return may_access(parent, ex);
 362 }
 363
 364 /**
 365  * may_allow_all - checks if it's possible to change the behavior to
 366  *                 allow based on parent's rules.
 367  * @parent: device cgroup's parent
 368  * returns: != 0 in case it's allowed, 0 otherwise
 369  */
 370 static inline int may_allow_all(struct dev_cgroup *parent)
 371 {
 372         if (!parent)
 373                 return 1;
 374         return parent->behavior == DEVCG_DEFAULT_ALLOW;
 375 }
 376
 377 /*
 378  * Modify the exception list using allow/deny rules.
 379  * CAP_SYS_ADMIN is needed for this.  It's at least separate from CAP_MKNOD
 380  * so we can give a container CAP_MKNOD to let it create devices but not
 381  * modify the exception list.
 382  * It seems likely we'll want to add a CAP_CONTAINER capability to allow
 383  * us to also grant CAP_SYS_ADMIN to containers without giving away the
 384  * device exception list controls, but for now we'll stick with CAP_SYS_ADMIN
 385  *
 386  * Taking rules away is always allowed (given CAP_SYS_ADMIN).  Granting
 387  * new access is only allowed if you're in the top-level cgroup, or your
 388  * parent cgroup has the access you're asking for.
 389  */
 390 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
 391                                    int filetype, const char *buffer)
 392 {
 393         const char *b;
 394         char temp[12];          /* 11 + 1 characters needed for a u32 */
 395         int count, rc;
 396         struct dev_exception_item ex;
 397         struct cgroup *p = devcgroup->css.cgroup;
 398         struct dev_cgroup *parent = NULL;
 399
 400         if (!capable(CAP_SYS_ADMIN))
 401                 return -EPERM;
 402
 403         if (p->parent)
 404                 parent = cgroup_to_devcgroup(p->parent);
 405
 406         memset(&ex, 0, sizeof(ex));
 407         b = buffer;
 408
 409         switch (*b) {
 410         case 'a':
 411                 switch (filetype) {
 412                 case DEVCG_ALLOW:
 413                         if (!may_allow_all(parent))
 414                                 return -EPERM;
 415                         dev_exception_clean(devcgroup);
 416                         devcgroup->behavior = DEVCG_DEFAULT_ALLOW;
 417                         if (!parent)
 418                                 break;
 419
 420                         rc = dev_exceptions_copy(&devcgroup->exceptions,
 421                                                  &parent->exceptions);
 422                         if (rc)
 423                                 return rc;
 424                         break;
 425                 case DEVCG_DENY:
 426                         dev_exception_clean(devcgroup);
 427                         devcgroup->behavior = DEVCG_DEFAULT_DENY;
 428                         break;
 429                 default:
 430                         return -EINVAL;
 431                 }
 432                 return 0;
 433         case 'b':
 434                 ex.type = DEV_BLOCK;
 435                 break;
 436         case 'c':
 437                 ex.type = DEV_CHAR;
 438                 break;
 439         default:
 440                 return -EINVAL;
 441         }
 442         b++;
 443         if (!isspace(*b))
 444                 return -EINVAL;
 445         b++;
 446         if (*b == '*') {
 447                 ex.major = ~0;
 448                 b++;
 449         } else if (isdigit(*b)) {
 450                 memset(temp, 0, sizeof(temp));
 451                 for (count = 0; count < sizeof(temp) - 1; count++) {
 452                         temp[count] = *b;
 453                         b++;
 454                         if (!isdigit(*b))
 455                                 break;
 456                 }
 457                 rc = kstrtou32(temp, 10, &ex.major);
 458                 if (rc)
 459                         return -EINVAL;
 460         } else {
 461                 return -EINVAL;
 462         }
 463         if (*b != ':')
 464                 return -EINVAL;
 465         b++;
 466
 467         /* read minor */
 468         if (*b == '*') {
 469                 ex.minor = ~0;
 470                 b++;
 471         } else if (isdigit(*b)) {
 472                 memset(temp, 0, sizeof(temp));
 473                 for (count = 0; count < sizeof(temp) - 1; count++) {
 474                         temp[count] = *b;
 475                         b++;
 476                         if (!isdigit(*b))
 477                                 break;
 478                 }
 479                 rc = kstrtou32(temp, 10, &ex.minor);
 480                 if (rc)
 481                         return -EINVAL;
 482         } else {
 483                 return -EINVAL;
 484         }
 485         if (!isspace(*b))
 486                 return -EINVAL;
 487         for (b++, count = 0; count < 3; count++, b++) {
 488                 switch (*b) {
 489                 case 'r':
 490                         ex.access |= ACC_READ;
 491                         break;
 492                 case 'w':
 493                         ex.access |= ACC_WRITE;
 494                         break;
 495                 case 'm':
 496                         ex.access |= ACC_MKNOD;
 497                         break;
 498                 case '\n':
 499                 case '\0':
 500                         count = 3;
 501                         break;
 502                 default:
 503                         return -EINVAL;
 504                 }
 505         }
 506
 507         switch (filetype) {
 508         case DEVCG_ALLOW:
 509                 if (!parent_has_perm(devcgroup, &ex))
 510                         return -EPERM;
 511                 /*
 512                  * If the default policy is to allow by default, try to remove
 513                  * an matching exception instead. And be silent about it: we
 514                  * don't want to break compatibility
 515                  */
 516                 if (devcgroup->behavior == DEVCG_DEFAULT_ALLOW) {
 517                         dev_exception_rm(devcgroup, &ex);
 518                         return 0;
 519                 }
 520                 return dev_exception_add(devcgroup, &ex);
 521         case DEVCG_DENY:
 522                 /*
 523                  * If the default policy is to deny by default, try to remove
 524                  * an matching exception instead. And be silent about it: we
 525                  * don't want to break compatibility
 526                  */
 527                 if (devcgroup->behavior == DEVCG_DEFAULT_DENY) {
 528                         dev_exception_rm(devcgroup, &ex);
 529                         return 0;
 530                 }
 531                 return dev_exception_add(devcgroup, &ex);
 532         default:
 533                 return -EINVAL;
 534         }
 535         return 0;
 536 }
 537
 538 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
 539                                   const char *buffer)
 540 {
 541         int retval;
 542
 543         mutex_lock(&devcgroup_mutex);
 544         retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
 545                                          cft->private, buffer);
 546         mutex_unlock(&devcgroup_mutex);
 547         return retval;
 548 }
 549
 550 static struct cftype dev_cgroup_files[] = {
 551         {
 552                 .name = "allow",
 553                 .write_string  = devcgroup_access_write,
 554                 .private = DEVCG_ALLOW,
 555         },
 556         {
 557                 .name = "deny",
 558                 .write_string = devcgroup_access_write,
 559                 .private = DEVCG_DENY,
 560         },
 561         {
 562                 .name = "list",
 563                 .read_seq_string = devcgroup_seq_read,
 564                 .private = DEVCG_LIST,
 565         },
 566         { }     /* terminate */
 567 };
 568
 569 struct cgroup_subsys devices_subsys = {
 570         .name = "devices",
 571         .can_attach = devcgroup_can_attach,
 572         .css_alloc = devcgroup_css_alloc,
 573         .css_free = devcgroup_css_free,
 574         .subsys_id = devices_subsys_id,
 575         .base_cftypes = dev_cgroup_files,
 576
 577         /*
 578          * While devices cgroup has the rudimentary hierarchy support which
 579          * checks the parent's restriction, it doesn't properly propagates
 580          * config changes in ancestors to their descendents.  A child
 581          * should only be allowed to add more restrictions to the parent's
 582          * configuration.  Fix it and remove the following.
 583          */
 584         .broken_hierarchy = true,
 585 };
 586
 587 /**
 588  * __devcgroup_check_permission - checks if an inode operation is permitted
 589  * @dev_cgroup: the dev cgroup to be tested against
 590  * @type: device type
 591  * @major: device major number
 592  * @minor: device minor number
 593  * @access: combination of ACC_WRITE, ACC_READ and ACC_MKNOD
 594  *
 595  * returns 0 on success, -EPERM case the operation is not permitted
 596  */
 597 static int __devcgroup_check_permission(short type, u32 major, u32 minor,
 598                                         short access)
 599 {
 600         struct dev_cgroup *dev_cgroup;
 601         struct dev_exception_item ex;
 602         int rc;
 603
 604         memset(&ex, 0, sizeof(ex));
 605         ex.type = type;
 606         ex.major = major;
 607         ex.minor = minor;
 608         ex.access = access;
 609
 610         rcu_read_lock();
 611         dev_cgroup = task_devcgroup(current);
 612         rc = may_access(dev_cgroup, &ex);
 613         rcu_read_unlock();
 614
 615         if (!rc)
 616                 return -EPERM;
 617
 618         return 0;
 619 }
 620
 621 int __devcgroup_inode_permission(struct inode *inode, int mask)
 622 {
 623         short type, access = 0;
 624
 625         if (S_ISBLK(inode->i_mode))
 626                 type = DEV_BLOCK;
 627         if (S_ISCHR(inode->i_mode))
 628                 type = DEV_CHAR;
 629         if (mask & MAY_WRITE)
 630                 access |= ACC_WRITE;
 631         if (mask & MAY_READ)
 632                 access |= ACC_READ;
 633
 634         return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
 635                         access);
 636 }
 637
 638 int devcgroup_inode_mknod(int mode, dev_t dev)
 639 {
 640         short type;
 641
 642         if (!S_ISBLK(mode) && !S_ISCHR(mode))
 643                 return 0;
 644
 645         if (S_ISBLK(mode))
 646                 type = DEV_BLOCK;
 647         else
 648                 type = DEV_CHAR;
 649
 650         return __devcgroup_check_permission(type, MAJOR(dev), MINOR(dev),
 651                         ACC_MKNOD);
 652
 653 }