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  * whitelist locking rules:
  30  * hold devcgroup_mutex for update/read.
  31  * hold rcu_read_lock() for read.
  32  */
  33
  34 struct dev_whitelist_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 whitelist;
  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_whitelist_copy(struct list_head *dest, struct list_head *orig)
  79 {
  80         struct dev_whitelist_item *wh, *tmp, *new;
  81
  82         list_for_each_entry(wh, orig, list) {
  83                 new = kmemdup(wh, sizeof(*wh), 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(wh, tmp, dest, list) {
  93                 list_del(&wh->list);
  94                 kfree(wh);
  95         }
  96         return -ENOMEM;
  97 }
  98
  99 /* Stupid prototype - don't bother combining existing entries */
 100 /*
 101  * called under devcgroup_mutex
 102  */
 103 static int dev_whitelist_add(struct dev_cgroup *dev_cgroup,
 104                         struct dev_whitelist_item *wh)
 105 {
 106         struct dev_whitelist_item *whcopy, *walk;
 107
 108         whcopy = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
 109         if (!whcopy)
 110                 return -ENOMEM;
 111
 112         list_for_each_entry(walk, &dev_cgroup->whitelist, list) {
 113                 if (walk->type != wh->type)
 114                         continue;
 115                 if (walk->major != wh->major)
 116                         continue;
 117                 if (walk->minor != wh->minor)
 118                         continue;
 119
 120                 walk->access |= wh->access;
 121                 kfree(whcopy);
 122                 whcopy = NULL;
 123         }
 124
 125         if (whcopy != NULL)
 126                 list_add_tail_rcu(&whcopy->list, &dev_cgroup->whitelist);
 127         return 0;
 128 }
 129
 130 /*
 131  * called under devcgroup_mutex
 132  */
 133 static void dev_whitelist_rm(struct dev_cgroup *dev_cgroup,
 134                         struct dev_whitelist_item *wh)
 135 {
 136         struct dev_whitelist_item *walk, *tmp;
 137
 138         list_for_each_entry_safe(walk, tmp, &dev_cgroup->whitelist, list) {
 139                 if (walk->type == DEV_ALL)
 140                         goto remove;
 141                 if (walk->type != wh->type)
 142                         continue;
 143                 if (walk->major != ~0 && walk->major != wh->major)
 144                         continue;
 145                 if (walk->minor != ~0 && walk->minor != wh->minor)
 146                         continue;
 147
 148 remove:
 149                 walk->access &= ~wh->access;
 150                 if (!walk->access) {
 151                         list_del_rcu(&walk->list);
 152                         kfree_rcu(walk, rcu);
 153                 }
 154         }
 155 }
 156
 157 /*
 158  * called from kernel/cgroup.c with cgroup_lock() held.
 159  */
 160 static struct cgroup_subsys_state *devcgroup_create(struct cgroup *cgroup)
 161 {
 162         struct dev_cgroup *dev_cgroup, *parent_dev_cgroup;
 163         struct cgroup *parent_cgroup;
 164         int ret;
 165
 166         dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
 167         if (!dev_cgroup)
 168                 return ERR_PTR(-ENOMEM);
 169         INIT_LIST_HEAD(&dev_cgroup->whitelist);
 170         parent_cgroup = cgroup->parent;
 171
 172         if (parent_cgroup == NULL) {
 173                 struct dev_whitelist_item *wh;
 174                 wh = kmalloc(sizeof(*wh), GFP_KERNEL);
 175                 if (!wh) {
 176                         kfree(dev_cgroup);
 177                         return ERR_PTR(-ENOMEM);
 178                 }
 179                 wh->minor = wh->major = ~0;
 180                 wh->type = DEV_ALL;
 181                 wh->access = ACC_MASK;
 182                 dev_cgroup->deny_all = false;
 183                 list_add(&wh->list, &dev_cgroup->whitelist);
 184         } else {
 185                 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
 186                 mutex_lock(&devcgroup_mutex);
 187                 ret = dev_whitelist_copy(&dev_cgroup->whitelist,
 188                                 &parent_dev_cgroup->whitelist);
 189                 dev_cgroup->deny_all = parent_dev_cgroup->deny_all;
 190                 mutex_unlock(&devcgroup_mutex);
 191                 if (ret) {
 192                         kfree(dev_cgroup);
 193                         return ERR_PTR(ret);
 194                 }
 195         }
 196
 197         return &dev_cgroup->css;
 198 }
 199
 200 static void devcgroup_destroy(struct cgroup *cgroup)
 201 {
 202         struct dev_cgroup *dev_cgroup;
 203         struct dev_whitelist_item *wh, *tmp;
 204
 205         dev_cgroup = cgroup_to_devcgroup(cgroup);
 206         list_for_each_entry_safe(wh, tmp, &dev_cgroup->whitelist, list) {
 207                 list_del(&wh->list);
 208                 kfree(wh);
 209         }
 210         kfree(dev_cgroup);
 211 }
 212
 213 #define DEVCG_ALLOW 1
 214 #define DEVCG_DENY 2
 215 #define DEVCG_LIST 3
 216
 217 #define MAJMINLEN 13
 218 #define ACCLEN 4
 219
 220 static void set_access(char *acc, short access)
 221 {
 222         int idx = 0;
 223         memset(acc, 0, ACCLEN);
 224         if (access & ACC_READ)
 225                 acc[idx++] = 'r';
 226         if (access & ACC_WRITE)
 227                 acc[idx++] = 'w';
 228         if (access & ACC_MKNOD)
 229                 acc[idx++] = 'm';
 230 }
 231
 232 static char type_to_char(short type)
 233 {
 234         if (type == DEV_ALL)
 235                 return 'a';
 236         if (type == DEV_CHAR)
 237                 return 'c';
 238         if (type == DEV_BLOCK)
 239                 return 'b';
 240         return 'X';
 241 }
 242
 243 static void set_majmin(char *str, unsigned m)
 244 {
 245         if (m == ~0)
 246                 strcpy(str, "*");
 247         else
 248                 sprintf(str, "%u", m);
 249 }
 250
 251 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
 252                                 struct seq_file *m)
 253 {
 254         struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
 255         struct dev_whitelist_item *wh;
 256         char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
 257
 258         rcu_read_lock();
 259         list_for_each_entry_rcu(wh, &devcgroup->whitelist, list) {
 260                 set_access(acc, wh->access);
 261                 set_majmin(maj, wh->major);
 262                 set_majmin(min, wh->minor);
 263                 seq_printf(m, "%c %s:%s %s\n", type_to_char(wh->type),
 264                            maj, min, acc);
 265         }
 266         rcu_read_unlock();
 267
 268         return 0;
 269 }
 270
 271 /*
 272  * may_access_whitelist:
 273  * does the access granted to dev_cgroup c contain the access
 274  * requested in whitelist item refwh.
 275  * return 1 if yes, 0 if no.
 276  * call with devcgroup_mutex held
 277  */
 278 static int may_access_whitelist(struct dev_cgroup *c,
 279                                        struct dev_whitelist_item *refwh)
 280 {
 281         struct dev_whitelist_item *whitem;
 282
 283         list_for_each_entry(whitem, &c->whitelist, list) {
 284                 if (whitem->type & DEV_ALL)
 285                         return 1;
 286                 if ((refwh->type & DEV_BLOCK) && !(whitem->type & DEV_BLOCK))
 287                         continue;
 288                 if ((refwh->type & DEV_CHAR) && !(whitem->type & DEV_CHAR))
 289                         continue;
 290                 if (whitem->major != ~0 && whitem->major != refwh->major)
 291                         continue;
 292                 if (whitem->minor != ~0 && whitem->minor != refwh->minor)
 293                         continue;
 294                 if (refwh->access & (~whitem->access))
 295                         continue;
 296                 return 1;
 297         }
 298         return 0;
 299 }
 300
 301 /*
 302  * parent_has_perm:
 303  * when adding a new allow rule to a device whitelist, the rule
 304  * must be allowed in the parent device
 305  */
 306 static int parent_has_perm(struct dev_cgroup *childcg,
 307                                   struct dev_whitelist_item *wh)
 308 {
 309         struct cgroup *pcg = childcg->css.cgroup->parent;
 310         struct dev_cgroup *parent;
 311
 312         if (!pcg)
 313                 return 1;
 314         parent = cgroup_to_devcgroup(pcg);
 315         return may_access_whitelist(parent, wh);
 316 }
 317
 318 /*
 319  * Modify the whitelist using allow/deny rules.
 320  * CAP_SYS_ADMIN is needed for this.  It's at least separate from CAP_MKNOD
 321  * so we can give a container CAP_MKNOD to let it create devices but not
 322  * modify the whitelist.
 323  * It seems likely we'll want to add a CAP_CONTAINER capability to allow
 324  * us to also grant CAP_SYS_ADMIN to containers without giving away the
 325  * device whitelist controls, but for now we'll stick with CAP_SYS_ADMIN
 326  *
 327  * Taking rules away is always allowed (given CAP_SYS_ADMIN).  Granting
 328  * new access is only allowed if you're in the top-level cgroup, or your
 329  * parent cgroup has the access you're asking for.
 330  */
 331 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
 332                                    int filetype, const char *buffer)
 333 {
 334         const char *b;
 335         char *endp;
 336         int count;
 337         struct dev_whitelist_item wh;
 338
 339         if (!capable(CAP_SYS_ADMIN))
 340                 return -EPERM;
 341
 342         memset(&wh, 0, sizeof(wh));
 343         b = buffer;
 344
 345         switch (*b) {
 346         case 'a':
 347                 wh.type = DEV_ALL;
 348                 wh.access = ACC_MASK;
 349                 wh.major = ~0;
 350                 wh.minor = ~0;
 351                 goto handle;
 352         case 'b':
 353                 wh.type = DEV_BLOCK;
 354                 break;
 355         case 'c':
 356                 wh.type = DEV_CHAR;
 357                 break;
 358         default:
 359                 return -EINVAL;
 360         }
 361         b++;
 362         if (!isspace(*b))
 363                 return -EINVAL;
 364         b++;
 365         if (*b == '*') {
 366                 wh.major = ~0;
 367                 b++;
 368         } else if (isdigit(*b)) {
 369                 wh.major = simple_strtoul(b, &endp, 10);
 370                 b = endp;
 371         } else {
 372                 return -EINVAL;
 373         }
 374         if (*b != ':')
 375                 return -EINVAL;
 376         b++;
 377
 378         /* read minor */
 379         if (*b == '*') {
 380                 wh.minor = ~0;
 381                 b++;
 382         } else if (isdigit(*b)) {
 383                 wh.minor = simple_strtoul(b, &endp, 10);
 384                 b = endp;
 385         } else {
 386                 return -EINVAL;
 387         }
 388         if (!isspace(*b))
 389                 return -EINVAL;
 390         for (b++, count = 0; count < 3; count++, b++) {
 391                 switch (*b) {
 392                 case 'r':
 393                         wh.access |= ACC_READ;
 394                         break;
 395                 case 'w':
 396                         wh.access |= ACC_WRITE;
 397                         break;
 398                 case 'm':
 399                         wh.access |= ACC_MKNOD;
 400                         break;
 401                 case '\n':
 402                 case '\0':
 403                         count = 3;
 404                         break;
 405                 default:
 406                         return -EINVAL;
 407                 }
 408         }
 409
 410 handle:
 411         switch (filetype) {
 412         case DEVCG_ALLOW:
 413                 if (!parent_has_perm(devcgroup, &wh))
 414                         return -EPERM;
 415                 devcgroup->deny_all = false;
 416                 return dev_whitelist_add(devcgroup, &wh);
 417         case DEVCG_DENY:
 418                 dev_whitelist_rm(devcgroup, &wh);
 419                 devcgroup->deny_all = true;
 420                 break;
 421         default:
 422                 return -EINVAL;
 423         }
 424         return 0;
 425 }
 426
 427 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
 428                                   const char *buffer)
 429 {
 430         int retval;
 431
 432         mutex_lock(&devcgroup_mutex);
 433         retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
 434                                          cft->private, buffer);
 435         mutex_unlock(&devcgroup_mutex);
 436         return retval;
 437 }
 438
 439 static struct cftype dev_cgroup_files[] = {
 440         {
 441                 .name = "allow",
 442                 .write_string  = devcgroup_access_write,
 443                 .private = DEVCG_ALLOW,
 444         },
 445         {
 446                 .name = "deny",
 447                 .write_string = devcgroup_access_write,
 448                 .private = DEVCG_DENY,
 449         },
 450         {
 451                 .name = "list",
 452                 .read_seq_string = devcgroup_seq_read,
 453                 .private = DEVCG_LIST,
 454         },
 455         { }     /* terminate */
 456 };
 457
 458 struct cgroup_subsys devices_subsys = {
 459         .name = "devices",
 460         .can_attach = devcgroup_can_attach,
 461         .create = devcgroup_create,
 462         .destroy = devcgroup_destroy,
 463         .subsys_id = devices_subsys_id,
 464         .base_cftypes = dev_cgroup_files,
 465
 466         /*
 467          * While devices cgroup has the rudimentary hierarchy support which
 468          * checks the parent's restriction, it doesn't properly propagates
 469          * config changes in ancestors to their descendents.  A child
 470          * should only be allowed to add more restrictions to the parent's
 471          * configuration.  Fix it and remove the following.
 472          */
 473         .broken_hierarchy = true,
 474 };
 475
 476 int __devcgroup_inode_permission(struct inode *inode, int mask)
 477 {
 478         struct dev_cgroup *dev_cgroup;
 479         struct dev_whitelist_item *wh;
 480
 481         rcu_read_lock();
 482
 483         dev_cgroup = task_devcgroup(current);
 484
 485         list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
 486                 if (wh->type & DEV_ALL)
 487                         goto found;
 488                 if ((wh->type & DEV_BLOCK) && !S_ISBLK(inode->i_mode))
 489                         continue;
 490                 if ((wh->type & DEV_CHAR) && !S_ISCHR(inode->i_mode))
 491                         continue;
 492                 if (wh->major != ~0 && wh->major != imajor(inode))
 493                         continue;
 494                 if (wh->minor != ~0 && wh->minor != iminor(inode))
 495                         continue;
 496
 497                 if ((mask & MAY_WRITE) && !(wh->access & ACC_WRITE))
 498                         continue;
 499                 if ((mask & MAY_READ) && !(wh->access & ACC_READ))
 500                         continue;
 501 found:
 502                 rcu_read_unlock();
 503                 return 0;
 504         }
 505
 506         rcu_read_unlock();
 507
 508         return -EPERM;
 509 }
 510
 511 int devcgroup_inode_mknod(int mode, dev_t dev)
 512 {
 513         struct dev_cgroup *dev_cgroup;
 514         struct dev_whitelist_item *wh;
 515
 516         if (!S_ISBLK(mode) && !S_ISCHR(mode))
 517                 return 0;
 518
 519         rcu_read_lock();
 520
 521         dev_cgroup = task_devcgroup(current);
 522
 523         list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
 524                 if (wh->type & DEV_ALL)
 525                         goto found;
 526                 if ((wh->type & DEV_BLOCK) && !S_ISBLK(mode))
 527                         continue;
 528                 if ((wh->type & DEV_CHAR) && !S_ISCHR(mode))
 529                         continue;
 530                 if (wh->major != ~0 && wh->major != MAJOR(dev))
 531                         continue;
 532                 if (wh->minor != ~0 && wh->minor != MINOR(dev))
 533                         continue;
 534
 535                 if (!(wh->access & ACC_MKNOD))
 536                         continue;
 537 found:
 538                 rcu_read_unlock();
 539                 return 0;
 540         }
 541
 542         rcu_read_unlock();
 543
 544         return -EPERM;
 545 }