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/rcupdate.h> 14 #include <linux/mutex.h> 15 16 #define ACC_MKNOD 1 17 #define ACC_READ 2 18 #define ACC_WRITE 4 19 #define ACC_MASK (ACC_MKNOD | ACC_READ | ACC_WRITE) 20 21 #define DEV_BLOCK 1 22 #define DEV_CHAR 2 23 #define DEV_ALL 4 /* this represents all devices */ 24 25 static DEFINE_MUTEX(devcgroup_mutex); 26 27 /* 28 * whitelist locking rules: 29 * hold devcgroup_mutex for update/read. 30 * hold rcu_read_lock() for read. 31 */ 32 33 struct dev_whitelist_item { 34 u32 major, minor; 35 short type; 36 short access; 37 struct list_head list; 38 struct rcu_head rcu; 39 }; 40 41 struct dev_cgroup { 42 struct cgroup_subsys_state css; 43 struct list_head whitelist; 44 }; 45 46 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s) 47 { 48 return container_of(s, struct dev_cgroup, css); 49 } 50 51 static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup) 52 { 53 return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id)); 54 } 55 56 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task) 57 { 58 return css_to_devcgroup(task_subsys_state(task, devices_subsys_id)); 59 } 60 61 struct cgroup_subsys devices_subsys; 62 63 static int devcgroup_can_attach(struct cgroup_subsys *ss, 64 struct cgroup *new_cgroup, struct task_struct *task, 65 bool threadgroup) 66 { 67 if (current != task && !capable(CAP_SYS_ADMIN)) 68 return -EPERM; 69 70 return 0; 71 } 72 73 /* 74 * called under devcgroup_mutex 75 */ 76 static int dev_whitelist_copy(struct list_head *dest, struct list_head *orig) 77 { 78 struct dev_whitelist_item *wh, *tmp, *new; 79 80 list_for_each_entry(wh, orig, list) { 81 new = kmemdup(wh, sizeof(*wh), GFP_KERNEL); 82 if (!new) 83 goto free_and_exit; 84 list_add_tail(&new->list, dest); 85 } 86 87 return 0; 88 89 free_and_exit: 90 list_for_each_entry_safe(wh, tmp, dest, list) { 91 list_del(&wh->list); 92 kfree(wh); 93 } 94 return -ENOMEM; 95 } 96 97 /* Stupid prototype - don't bother combining existing entries */ 98 /* 99 * called under devcgroup_mutex 100 */ 101 static int dev_whitelist_add(struct dev_cgroup *dev_cgroup, 102 struct dev_whitelist_item *wh) 103 { 104 struct dev_whitelist_item *whcopy, *walk; 105 106 whcopy = kmemdup(wh, sizeof(*wh), GFP_KERNEL); 107 if (!whcopy) 108 return -ENOMEM; 109 110 list_for_each_entry(walk, &dev_cgroup->whitelist, list) { 111 if (walk->type != wh->type) 112 continue; 113 if (walk->major != wh->major) 114 continue; 115 if (walk->minor != wh->minor) 116 continue; 117 118 walk->access |= wh->access; 119 kfree(whcopy); 120 whcopy = NULL; 121 } 122 123 if (whcopy != NULL) 124 list_add_tail_rcu(&whcopy->list, &dev_cgroup->whitelist); 125 return 0; 126 } 127 128 static void whitelist_item_free(struct rcu_head *rcu) 129 { 130 struct dev_whitelist_item *item; 131 132 item = container_of(rcu, struct dev_whitelist_item, rcu); 133 kfree(item); 134 } 135 136 /* 137 * called under devcgroup_mutex 138 */ 139 static void dev_whitelist_rm(struct dev_cgroup *dev_cgroup, 140 struct dev_whitelist_item *wh) 141 { 142 struct dev_whitelist_item *walk, *tmp; 143 144 list_for_each_entry_safe(walk, tmp, &dev_cgroup->whitelist, list) { 145 if (walk->type == DEV_ALL) 146 goto remove; 147 if (walk->type != wh->type) 148 continue; 149 if (walk->major != ~0 && walk->major != wh->major) 150 continue; 151 if (walk->minor != ~0 && walk->minor != wh->minor) 152 continue; 153 154 remove: 155 walk->access &= ~wh->access; 156 if (!walk->access) { 157 list_del_rcu(&walk->list); 158 call_rcu(&walk->rcu, whitelist_item_free); 159 } 160 } 161 } 162 163 /* 164 * called from kernel/cgroup.c with cgroup_lock() held. 165 */ 166 static struct cgroup_subsys_state *devcgroup_create(struct cgroup_subsys *ss, 167 struct cgroup *cgroup) 168 { 169 struct dev_cgroup *dev_cgroup, *parent_dev_cgroup; 170 struct cgroup *parent_cgroup; 171 int ret; 172 173 dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL); 174 if (!dev_cgroup) 175 return ERR_PTR(-ENOMEM); 176 INIT_LIST_HEAD(&dev_cgroup->whitelist); 177 parent_cgroup = cgroup->parent; 178 179 if (parent_cgroup == NULL) { 180 struct dev_whitelist_item *wh; 181 wh = kmalloc(sizeof(*wh), GFP_KERNEL); 182 if (!wh) { 183 kfree(dev_cgroup); 184 return ERR_PTR(-ENOMEM); 185 } 186 wh->minor = wh->major = ~0; 187 wh->type = DEV_ALL; 188 wh->access = ACC_MASK; 189 list_add(&wh->list, &dev_cgroup->whitelist); 190 } else { 191 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup); 192 mutex_lock(&devcgroup_mutex); 193 ret = dev_whitelist_copy(&dev_cgroup->whitelist, 194 &parent_dev_cgroup->whitelist); 195 mutex_unlock(&devcgroup_mutex); 196 if (ret) { 197 kfree(dev_cgroup); 198 return ERR_PTR(ret); 199 } 200 } 201 202 return &dev_cgroup->css; 203 } 204 205 static void devcgroup_destroy(struct cgroup_subsys *ss, 206 struct cgroup *cgroup) 207 { 208 struct dev_cgroup *dev_cgroup; 209 struct dev_whitelist_item *wh, *tmp; 210 211 dev_cgroup = cgroup_to_devcgroup(cgroup); 212 list_for_each_entry_safe(wh, tmp, &dev_cgroup->whitelist, list) { 213 list_del(&wh->list); 214 kfree(wh); 215 } 216 kfree(dev_cgroup); 217 } 218 219 #define DEVCG_ALLOW 1 220 #define DEVCG_DENY 2 221 #define DEVCG_LIST 3 222 223 #define MAJMINLEN 13 224 #define ACCLEN 4 225 226 static void set_access(char *acc, short access) 227 { 228 int idx = 0; 229 memset(acc, 0, ACCLEN); 230 if (access & ACC_READ) 231 acc[idx++] = 'r'; 232 if (access & ACC_WRITE) 233 acc[idx++] = 'w'; 234 if (access & ACC_MKNOD) 235 acc[idx++] = 'm'; 236 } 237 238 static char type_to_char(short type) 239 { 240 if (type == DEV_ALL) 241 return 'a'; 242 if (type == DEV_CHAR) 243 return 'c'; 244 if (type == DEV_BLOCK) 245 return 'b'; 246 return 'X'; 247 } 248 249 static void set_majmin(char *str, unsigned m) 250 { 251 if (m == ~0) 252 strcpy(str, "*"); 253 else 254 sprintf(str, "%u", m); 255 } 256 257 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft, 258 struct seq_file *m) 259 { 260 struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup); 261 struct dev_whitelist_item *wh; 262 char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN]; 263 264 rcu_read_lock(); 265 list_for_each_entry_rcu(wh, &devcgroup->whitelist, list) { 266 set_access(acc, wh->access); 267 set_majmin(maj, wh->major); 268 set_majmin(min, wh->minor); 269 seq_printf(m, "%c %s:%s %s\n", type_to_char(wh->type), 270 maj, min, acc); 271 } 272 rcu_read_unlock(); 273 274 return 0; 275 } 276 277 /* 278 * may_access_whitelist: 279 * does the access granted to dev_cgroup c contain the access 280 * requested in whitelist item refwh. 281 * return 1 if yes, 0 if no. 282 * call with devcgroup_mutex held 283 */ 284 static int may_access_whitelist(struct dev_cgroup *c, 285 struct dev_whitelist_item *refwh) 286 { 287 struct dev_whitelist_item *whitem; 288 289 list_for_each_entry(whitem, &c->whitelist, list) { 290 if (whitem->type & DEV_ALL) 291 return 1; 292 if ((refwh->type & DEV_BLOCK) && !(whitem->type & DEV_BLOCK)) 293 continue; 294 if ((refwh->type & DEV_CHAR) && !(whitem->type & DEV_CHAR)) 295 continue; 296 if (whitem->major != ~0 && whitem->major != refwh->major) 297 continue; 298 if (whitem->minor != ~0 && whitem->minor != refwh->minor) 299 continue; 300 if (refwh->access & (~whitem->access)) 301 continue; 302 return 1; 303 } 304 return 0; 305 } 306 307 /* 308 * parent_has_perm: 309 * when adding a new allow rule to a device whitelist, the rule 310 * must be allowed in the parent device 311 */ 312 static int parent_has_perm(struct dev_cgroup *childcg, 313 struct dev_whitelist_item *wh) 314 { 315 struct cgroup *pcg = childcg->css.cgroup->parent; 316 struct dev_cgroup *parent; 317 318 if (!pcg) 319 return 1; 320 parent = cgroup_to_devcgroup(pcg); 321 return may_access_whitelist(parent, wh); 322 } 323 324 /* 325 * Modify the whitelist using allow/deny rules. 326 * CAP_SYS_ADMIN is needed for this. It's at least separate from CAP_MKNOD 327 * so we can give a container CAP_MKNOD to let it create devices but not 328 * modify the whitelist. 329 * It seems likely we'll want to add a CAP_CONTAINER capability to allow 330 * us to also grant CAP_SYS_ADMIN to containers without giving away the 331 * device whitelist controls, but for now we'll stick with CAP_SYS_ADMIN 332 * 333 * Taking rules away is always allowed (given CAP_SYS_ADMIN). Granting 334 * new access is only allowed if you're in the top-level cgroup, or your 335 * parent cgroup has the access you're asking for. 336 */ 337 static int devcgroup_update_access(struct dev_cgroup *devcgroup, 338 int filetype, const char *buffer) 339 { 340 const char *b; 341 char *endp; 342 int count; 343 struct dev_whitelist_item wh; 344 345 if (!capable(CAP_SYS_ADMIN)) 346 return -EPERM; 347 348 memset(&wh, 0, sizeof(wh)); 349 b = buffer; 350 351 switch (*b) { 352 case 'a': 353 wh.type = DEV_ALL; 354 wh.access = ACC_MASK; 355 wh.major = ~0; 356 wh.minor = ~0; 357 goto handle; 358 case 'b': 359 wh.type = DEV_BLOCK; 360 break; 361 case 'c': 362 wh.type = DEV_CHAR; 363 break; 364 default: 365 return -EINVAL; 366 } 367 b++; 368 if (!isspace(*b)) 369 return -EINVAL; 370 b++; 371 if (*b == '*') { 372 wh.major = ~0; 373 b++; 374 } else if (isdigit(*b)) { 375 wh.major = simple_strtoul(b, &endp, 10); 376 b = endp; 377 } else { 378 return -EINVAL; 379 } 380 if (*b != ':') 381 return -EINVAL; 382 b++; 383 384 /* read minor */ 385 if (*b == '*') { 386 wh.minor = ~0; 387 b++; 388 } else if (isdigit(*b)) { 389 wh.minor = simple_strtoul(b, &endp, 10); 390 b = endp; 391 } else { 392 return -EINVAL; 393 } 394 if (!isspace(*b)) 395 return -EINVAL; 396 for (b++, count = 0; count < 3; count++, b++) { 397 switch (*b) { 398 case 'r': 399 wh.access |= ACC_READ; 400 break; 401 case 'w': 402 wh.access |= ACC_WRITE; 403 break; 404 case 'm': 405 wh.access |= ACC_MKNOD; 406 break; 407 case '\n': 408 case '\0': 409 count = 3; 410 break; 411 default: 412 return -EINVAL; 413 } 414 } 415 416 handle: 417 switch (filetype) { 418 case DEVCG_ALLOW: 419 if (!parent_has_perm(devcgroup, &wh)) 420 return -EPERM; 421 return dev_whitelist_add(devcgroup, &wh); 422 case DEVCG_DENY: 423 dev_whitelist_rm(devcgroup, &wh); 424 break; 425 default: 426 return -EINVAL; 427 } 428 return 0; 429 } 430 431 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft, 432 const char *buffer) 433 { 434 int retval; 435 436 mutex_lock(&devcgroup_mutex); 437 retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp), 438 cft->private, buffer); 439 mutex_unlock(&devcgroup_mutex); 440 return retval; 441 } 442 443 static struct cftype dev_cgroup_files[] = { 444 { 445 .name = "allow", 446 .write_string = devcgroup_access_write, 447 .private = DEVCG_ALLOW, 448 }, 449 { 450 .name = "deny", 451 .write_string = devcgroup_access_write, 452 .private = DEVCG_DENY, 453 }, 454 { 455 .name = "list", 456 .read_seq_string = devcgroup_seq_read, 457 .private = DEVCG_LIST, 458 }, 459 }; 460 461 static int devcgroup_populate(struct cgroup_subsys *ss, 462 struct cgroup *cgroup) 463 { 464 return cgroup_add_files(cgroup, ss, dev_cgroup_files, 465 ARRAY_SIZE(dev_cgroup_files)); 466 } 467 468 struct cgroup_subsys devices_subsys = { 469 .name = "devices", 470 .can_attach = devcgroup_can_attach, 471 .create = devcgroup_create, 472 .destroy = devcgroup_destroy, 473 .populate = devcgroup_populate, 474 .subsys_id = devices_subsys_id, 475 }; 476 477 int devcgroup_inode_permission(struct inode *inode, int mask) 478 { 479 struct dev_cgroup *dev_cgroup; 480 struct dev_whitelist_item *wh; 481 482 dev_t device = inode->i_rdev; 483 if (!device) 484 return 0; 485 if (!S_ISBLK(inode->i_mode) && !S_ISCHR(inode->i_mode)) 486 return 0; 487 488 rcu_read_lock(); 489 490 dev_cgroup = task_devcgroup(current); 491 492 list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) { 493 if (wh->type & DEV_ALL) 494 goto found; 495 if ((wh->type & DEV_BLOCK) && !S_ISBLK(inode->i_mode)) 496 continue; 497 if ((wh->type & DEV_CHAR) && !S_ISCHR(inode->i_mode)) 498 continue; 499 if (wh->major != ~0 && wh->major != imajor(inode)) 500 continue; 501 if (wh->minor != ~0 && wh->minor != iminor(inode)) 502 continue; 503 504 if ((mask & MAY_WRITE) && !(wh->access & ACC_WRITE)) 505 continue; 506 if ((mask & MAY_READ) && !(wh->access & ACC_READ)) 507 continue; 508 found: 509 rcu_read_unlock(); 510 return 0; 511 } 512 513 rcu_read_unlock(); 514 515 return -EPERM; 516 } 517 518 int devcgroup_inode_mknod(int mode, dev_t dev) 519 { 520 struct dev_cgroup *dev_cgroup; 521 struct dev_whitelist_item *wh; 522 523 if (!S_ISBLK(mode) && !S_ISCHR(mode)) 524 return 0; 525 526 rcu_read_lock(); 527 528 dev_cgroup = task_devcgroup(current); 529 530 list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) { 531 if (wh->type & DEV_ALL) 532 goto found; 533 if ((wh->type & DEV_BLOCK) && !S_ISBLK(mode)) 534 continue; 535 if ((wh->type & DEV_CHAR) && !S_ISCHR(mode)) 536 continue; 537 if (wh->major != ~0 && wh->major != MAJOR(dev)) 538 continue; 539 if (wh->minor != ~0 && wh->minor != MINOR(dev)) 540 continue; 541 542 if (!(wh->access & ACC_MKNOD)) 543 continue; 544 found: 545 rcu_read_unlock(); 546 return 0; 547 } 548 549 rcu_read_unlock(); 550 551 return -EPERM; 552 } 553