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