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