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