1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Landlock LSM - Filesystem management and hooks 4 * 5 * Copyright © 2016-2020 Mickaël Salaün <mic@digikod.net> 6 * Copyright © 2018-2020 ANSSI 7 * Copyright © 2021-2022 Microsoft Corporation 8 */ 9 10 #include <linux/atomic.h> 11 #include <linux/bitops.h> 12 #include <linux/bits.h> 13 #include <linux/compiler_types.h> 14 #include <linux/dcache.h> 15 #include <linux/err.h> 16 #include <linux/fs.h> 17 #include <linux/init.h> 18 #include <linux/kernel.h> 19 #include <linux/limits.h> 20 #include <linux/list.h> 21 #include <linux/lsm_hooks.h> 22 #include <linux/mount.h> 23 #include <linux/namei.h> 24 #include <linux/path.h> 25 #include <linux/rcupdate.h> 26 #include <linux/spinlock.h> 27 #include <linux/stat.h> 28 #include <linux/types.h> 29 #include <linux/wait_bit.h> 30 #include <linux/workqueue.h> 31 #include <uapi/linux/landlock.h> 32 33 #include "common.h" 34 #include "cred.h" 35 #include "fs.h" 36 #include "limits.h" 37 #include "object.h" 38 #include "ruleset.h" 39 #include "setup.h" 40 41 /* Underlying object management */ 42 43 static void release_inode(struct landlock_object *const object) 44 __releases(object->lock) 45 { 46 struct inode *const inode = object->underobj; 47 struct super_block *sb; 48 49 if (!inode) { 50 spin_unlock(&object->lock); 51 return; 52 } 53 54 /* 55 * Protects against concurrent use by hook_sb_delete() of the reference 56 * to the underlying inode. 57 */ 58 object->underobj = NULL; 59 /* 60 * Makes sure that if the filesystem is concurrently unmounted, 61 * hook_sb_delete() will wait for us to finish iput(). 62 */ 63 sb = inode->i_sb; 64 atomic_long_inc(&landlock_superblock(sb)->inode_refs); 65 spin_unlock(&object->lock); 66 /* 67 * Because object->underobj was not NULL, hook_sb_delete() and 68 * get_inode_object() guarantee that it is safe to reset 69 * landlock_inode(inode)->object while it is not NULL. It is therefore 70 * not necessary to lock inode->i_lock. 71 */ 72 rcu_assign_pointer(landlock_inode(inode)->object, NULL); 73 /* 74 * Now, new rules can safely be tied to @inode with get_inode_object(). 75 */ 76 77 iput(inode); 78 if (atomic_long_dec_and_test(&landlock_superblock(sb)->inode_refs)) 79 wake_up_var(&landlock_superblock(sb)->inode_refs); 80 } 81 82 static const struct landlock_object_underops landlock_fs_underops = { 83 .release = release_inode 84 }; 85 86 /* Ruleset management */ 87 88 static struct landlock_object *get_inode_object(struct inode *const inode) 89 { 90 struct landlock_object *object, *new_object; 91 struct landlock_inode_security *inode_sec = landlock_inode(inode); 92 93 rcu_read_lock(); 94 retry: 95 object = rcu_dereference(inode_sec->object); 96 if (object) { 97 if (likely(refcount_inc_not_zero(&object->usage))) { 98 rcu_read_unlock(); 99 return object; 100 } 101 /* 102 * We are racing with release_inode(), the object is going 103 * away. Wait for release_inode(), then retry. 104 */ 105 spin_lock(&object->lock); 106 spin_unlock(&object->lock); 107 goto retry; 108 } 109 rcu_read_unlock(); 110 111 /* 112 * If there is no object tied to @inode, then create a new one (without 113 * holding any locks). 114 */ 115 new_object = landlock_create_object(&landlock_fs_underops, inode); 116 if (IS_ERR(new_object)) 117 return new_object; 118 119 /* 120 * Protects against concurrent calls to get_inode_object() or 121 * hook_sb_delete(). 122 */ 123 spin_lock(&inode->i_lock); 124 if (unlikely(rcu_access_pointer(inode_sec->object))) { 125 /* Someone else just created the object, bail out and retry. */ 126 spin_unlock(&inode->i_lock); 127 kfree(new_object); 128 129 rcu_read_lock(); 130 goto retry; 131 } 132 133 /* 134 * @inode will be released by hook_sb_delete() on its superblock 135 * shutdown, or by release_inode() when no more ruleset references the 136 * related object. 137 */ 138 ihold(inode); 139 rcu_assign_pointer(inode_sec->object, new_object); 140 spin_unlock(&inode->i_lock); 141 return new_object; 142 } 143 144 /* All access rights that can be tied to files. */ 145 /* clang-format off */ 146 #define ACCESS_FILE ( \ 147 LANDLOCK_ACCESS_FS_EXECUTE | \ 148 LANDLOCK_ACCESS_FS_WRITE_FILE | \ 149 LANDLOCK_ACCESS_FS_READ_FILE) 150 /* clang-format on */ 151 152 /* 153 * All access rights that are denied by default whether they are handled or not 154 * by a ruleset/layer. This must be ORed with all ruleset->fs_access_masks[] 155 * entries when we need to get the absolute handled access masks. 156 */ 157 /* clang-format off */ 158 #define ACCESS_INITIALLY_DENIED ( \ 159 LANDLOCK_ACCESS_FS_REFER) 160 /* clang-format on */ 161 162 /* 163 * @path: Should have been checked by get_path_from_fd(). 164 */ 165 int landlock_append_fs_rule(struct landlock_ruleset *const ruleset, 166 const struct path *const path, 167 access_mask_t access_rights) 168 { 169 int err; 170 struct landlock_object *object; 171 172 /* Files only get access rights that make sense. */ 173 if (!d_is_dir(path->dentry) && 174 (access_rights | ACCESS_FILE) != ACCESS_FILE) 175 return -EINVAL; 176 if (WARN_ON_ONCE(ruleset->num_layers != 1)) 177 return -EINVAL; 178 179 /* Transforms relative access rights to absolute ones. */ 180 access_rights |= 181 LANDLOCK_MASK_ACCESS_FS & 182 ~(ruleset->fs_access_masks[0] | ACCESS_INITIALLY_DENIED); 183 object = get_inode_object(d_backing_inode(path->dentry)); 184 if (IS_ERR(object)) 185 return PTR_ERR(object); 186 mutex_lock(&ruleset->lock); 187 err = landlock_insert_rule(ruleset, object, access_rights); 188 mutex_unlock(&ruleset->lock); 189 /* 190 * No need to check for an error because landlock_insert_rule() 191 * increments the refcount for the new object if needed. 192 */ 193 landlock_put_object(object); 194 return err; 195 } 196 197 /* Access-control management */ 198 199 /* 200 * The lifetime of the returned rule is tied to @domain. 201 * 202 * Returns NULL if no rule is found or if @dentry is negative. 203 */ 204 static inline const struct landlock_rule * 205 find_rule(const struct landlock_ruleset *const domain, 206 const struct dentry *const dentry) 207 { 208 const struct landlock_rule *rule; 209 const struct inode *inode; 210 211 /* Ignores nonexistent leafs. */ 212 if (d_is_negative(dentry)) 213 return NULL; 214 215 inode = d_backing_inode(dentry); 216 rcu_read_lock(); 217 rule = landlock_find_rule( 218 domain, rcu_dereference(landlock_inode(inode)->object)); 219 rcu_read_unlock(); 220 return rule; 221 } 222 223 /* 224 * @layer_masks is read and may be updated according to the access request and 225 * the matching rule. 226 * 227 * Returns true if the request is allowed (i.e. relevant layer masks for the 228 * request are empty). 229 */ 230 static inline bool 231 unmask_layers(const struct landlock_rule *const rule, 232 const access_mask_t access_request, 233 layer_mask_t (*const layer_masks)[LANDLOCK_NUM_ACCESS_FS]) 234 { 235 size_t layer_level; 236 237 if (!access_request || !layer_masks) 238 return true; 239 if (!rule) 240 return false; 241 242 /* 243 * An access is granted if, for each policy layer, at least one rule 244 * encountered on the pathwalk grants the requested access, 245 * regardless of its position in the layer stack. We must then check 246 * the remaining layers for each inode, from the first added layer to 247 * the last one. When there is multiple requested accesses, for each 248 * policy layer, the full set of requested accesses may not be granted 249 * by only one rule, but by the union (binary OR) of multiple rules. 250 * E.g. /a/b <execute> + /a <read> => /a/b <execute + read> 251 */ 252 for (layer_level = 0; layer_level < rule->num_layers; layer_level++) { 253 const struct landlock_layer *const layer = 254 &rule->layers[layer_level]; 255 const layer_mask_t layer_bit = BIT_ULL(layer->level - 1); 256 const unsigned long access_req = access_request; 257 unsigned long access_bit; 258 bool is_empty; 259 260 /* 261 * Records in @layer_masks which layer grants access to each 262 * requested access. 263 */ 264 is_empty = true; 265 for_each_set_bit(access_bit, &access_req, 266 ARRAY_SIZE(*layer_masks)) { 267 if (layer->access & BIT_ULL(access_bit)) 268 (*layer_masks)[access_bit] &= ~layer_bit; 269 is_empty = is_empty && !(*layer_masks)[access_bit]; 270 } 271 if (is_empty) 272 return true; 273 } 274 return false; 275 } 276 277 /* 278 * Allows access to pseudo filesystems that will never be mountable (e.g. 279 * sockfs, pipefs), but can still be reachable through 280 * /proc/<pid>/fd/<file-descriptor> 281 */ 282 static inline bool is_nouser_or_private(const struct dentry *dentry) 283 { 284 return (dentry->d_sb->s_flags & SB_NOUSER) || 285 (d_is_positive(dentry) && 286 unlikely(IS_PRIVATE(d_backing_inode(dentry)))); 287 } 288 289 static inline access_mask_t 290 get_handled_accesses(const struct landlock_ruleset *const domain) 291 { 292 access_mask_t access_dom = ACCESS_INITIALLY_DENIED; 293 size_t layer_level; 294 295 for (layer_level = 0; layer_level < domain->num_layers; layer_level++) 296 access_dom |= domain->fs_access_masks[layer_level]; 297 return access_dom & LANDLOCK_MASK_ACCESS_FS; 298 } 299 300 static inline access_mask_t 301 init_layer_masks(const struct landlock_ruleset *const domain, 302 const access_mask_t access_request, 303 layer_mask_t (*const layer_masks)[LANDLOCK_NUM_ACCESS_FS]) 304 { 305 access_mask_t handled_accesses = 0; 306 size_t layer_level; 307 308 memset(layer_masks, 0, sizeof(*layer_masks)); 309 /* An empty access request can happen because of O_WRONLY | O_RDWR. */ 310 if (!access_request) 311 return 0; 312 313 /* Saves all handled accesses per layer. */ 314 for (layer_level = 0; layer_level < domain->num_layers; layer_level++) { 315 const unsigned long access_req = access_request; 316 unsigned long access_bit; 317 318 for_each_set_bit(access_bit, &access_req, 319 ARRAY_SIZE(*layer_masks)) { 320 /* 321 * Artificially handles all initially denied by default 322 * access rights. 323 */ 324 if (BIT_ULL(access_bit) & 325 (domain->fs_access_masks[layer_level] | 326 ACCESS_INITIALLY_DENIED)) { 327 (*layer_masks)[access_bit] |= 328 BIT_ULL(layer_level); 329 handled_accesses |= BIT_ULL(access_bit); 330 } 331 } 332 } 333 return handled_accesses; 334 } 335 336 /* 337 * Check that a destination file hierarchy has more restrictions than a source 338 * file hierarchy. This is only used for link and rename actions. 339 * 340 * @layer_masks_child2: Optional child masks. 341 */ 342 static inline bool no_more_access( 343 const layer_mask_t (*const layer_masks_parent1)[LANDLOCK_NUM_ACCESS_FS], 344 const layer_mask_t (*const layer_masks_child1)[LANDLOCK_NUM_ACCESS_FS], 345 const bool child1_is_directory, 346 const layer_mask_t (*const layer_masks_parent2)[LANDLOCK_NUM_ACCESS_FS], 347 const layer_mask_t (*const layer_masks_child2)[LANDLOCK_NUM_ACCESS_FS], 348 const bool child2_is_directory) 349 { 350 unsigned long access_bit; 351 352 for (access_bit = 0; access_bit < ARRAY_SIZE(*layer_masks_parent2); 353 access_bit++) { 354 /* Ignores accesses that only make sense for directories. */ 355 const bool is_file_access = 356 !!(BIT_ULL(access_bit) & ACCESS_FILE); 357 358 if (child1_is_directory || is_file_access) { 359 /* 360 * Checks if the destination restrictions are a 361 * superset of the source ones (i.e. inherited access 362 * rights without child exceptions): 363 * restrictions(parent2) >= restrictions(child1) 364 */ 365 if ((((*layer_masks_parent1)[access_bit] & 366 (*layer_masks_child1)[access_bit]) | 367 (*layer_masks_parent2)[access_bit]) != 368 (*layer_masks_parent2)[access_bit]) 369 return false; 370 } 371 372 if (!layer_masks_child2) 373 continue; 374 if (child2_is_directory || is_file_access) { 375 /* 376 * Checks inverted restrictions for RENAME_EXCHANGE: 377 * restrictions(parent1) >= restrictions(child2) 378 */ 379 if ((((*layer_masks_parent2)[access_bit] & 380 (*layer_masks_child2)[access_bit]) | 381 (*layer_masks_parent1)[access_bit]) != 382 (*layer_masks_parent1)[access_bit]) 383 return false; 384 } 385 } 386 return true; 387 } 388 389 /* 390 * Removes @layer_masks accesses that are not requested. 391 * 392 * Returns true if the request is allowed, false otherwise. 393 */ 394 static inline bool 395 scope_to_request(const access_mask_t access_request, 396 layer_mask_t (*const layer_masks)[LANDLOCK_NUM_ACCESS_FS]) 397 { 398 const unsigned long access_req = access_request; 399 unsigned long access_bit; 400 401 if (WARN_ON_ONCE(!layer_masks)) 402 return true; 403 404 for_each_clear_bit(access_bit, &access_req, ARRAY_SIZE(*layer_masks)) 405 (*layer_masks)[access_bit] = 0; 406 return !memchr_inv(layer_masks, 0, sizeof(*layer_masks)); 407 } 408 409 /* 410 * Returns true if there is at least one access right different than 411 * LANDLOCK_ACCESS_FS_REFER. 412 */ 413 static inline bool 414 is_eacces(const layer_mask_t (*const layer_masks)[LANDLOCK_NUM_ACCESS_FS], 415 const access_mask_t access_request) 416 { 417 unsigned long access_bit; 418 /* LANDLOCK_ACCESS_FS_REFER alone must return -EXDEV. */ 419 const unsigned long access_check = access_request & 420 ~LANDLOCK_ACCESS_FS_REFER; 421 422 if (!layer_masks) 423 return false; 424 425 for_each_set_bit(access_bit, &access_check, ARRAY_SIZE(*layer_masks)) { 426 if ((*layer_masks)[access_bit]) 427 return true; 428 } 429 return false; 430 } 431 432 /** 433 * is_access_to_paths_allowed - Check accesses for requests with a common path 434 * 435 * @domain: Domain to check against. 436 * @path: File hierarchy to walk through. 437 * @access_request_parent1: Accesses to check, once @layer_masks_parent1 is 438 * equal to @layer_masks_parent2 (if any). This is tied to the unique 439 * requested path for most actions, or the source in case of a refer action 440 * (i.e. rename or link), or the source and destination in case of 441 * RENAME_EXCHANGE. 442 * @layer_masks_parent1: Pointer to a matrix of layer masks per access 443 * masks, identifying the layers that forbid a specific access. Bits from 444 * this matrix can be unset according to the @path walk. An empty matrix 445 * means that @domain allows all possible Landlock accesses (i.e. not only 446 * those identified by @access_request_parent1). This matrix can 447 * initially refer to domain layer masks and, when the accesses for the 448 * destination and source are the same, to requested layer masks. 449 * @dentry_child1: Dentry to the initial child of the parent1 path. This 450 * pointer must be NULL for non-refer actions (i.e. not link nor rename). 451 * @access_request_parent2: Similar to @access_request_parent1 but for a 452 * request involving a source and a destination. This refers to the 453 * destination, except in case of RENAME_EXCHANGE where it also refers to 454 * the source. Must be set to 0 when using a simple path request. 455 * @layer_masks_parent2: Similar to @layer_masks_parent1 but for a refer 456 * action. This must be NULL otherwise. 457 * @dentry_child2: Dentry to the initial child of the parent2 path. This 458 * pointer is only set for RENAME_EXCHANGE actions and must be NULL 459 * otherwise. 460 * 461 * This helper first checks that the destination has a superset of restrictions 462 * compared to the source (if any) for a common path. Because of 463 * RENAME_EXCHANGE actions, source and destinations may be swapped. It then 464 * checks that the collected accesses and the remaining ones are enough to 465 * allow the request. 466 * 467 * Returns: 468 * - true if the access request is granted; 469 * - false otherwise. 470 */ 471 static bool is_access_to_paths_allowed( 472 const struct landlock_ruleset *const domain, 473 const struct path *const path, 474 const access_mask_t access_request_parent1, 475 layer_mask_t (*const layer_masks_parent1)[LANDLOCK_NUM_ACCESS_FS], 476 const struct dentry *const dentry_child1, 477 const access_mask_t access_request_parent2, 478 layer_mask_t (*const layer_masks_parent2)[LANDLOCK_NUM_ACCESS_FS], 479 const struct dentry *const dentry_child2) 480 { 481 bool allowed_parent1 = false, allowed_parent2 = false, is_dom_check, 482 child1_is_directory = true, child2_is_directory = true; 483 struct path walker_path; 484 access_mask_t access_masked_parent1, access_masked_parent2; 485 layer_mask_t _layer_masks_child1[LANDLOCK_NUM_ACCESS_FS], 486 _layer_masks_child2[LANDLOCK_NUM_ACCESS_FS]; 487 layer_mask_t(*layer_masks_child1)[LANDLOCK_NUM_ACCESS_FS] = NULL, 488 (*layer_masks_child2)[LANDLOCK_NUM_ACCESS_FS] = NULL; 489 490 if (!access_request_parent1 && !access_request_parent2) 491 return true; 492 if (WARN_ON_ONCE(!domain || !path)) 493 return true; 494 if (is_nouser_or_private(path->dentry)) 495 return true; 496 if (WARN_ON_ONCE(domain->num_layers < 1 || !layer_masks_parent1)) 497 return false; 498 499 if (unlikely(layer_masks_parent2)) { 500 if (WARN_ON_ONCE(!dentry_child1)) 501 return false; 502 /* 503 * For a double request, first check for potential privilege 504 * escalation by looking at domain handled accesses (which are 505 * a superset of the meaningful requested accesses). 506 */ 507 access_masked_parent1 = access_masked_parent2 = 508 get_handled_accesses(domain); 509 is_dom_check = true; 510 } else { 511 if (WARN_ON_ONCE(dentry_child1 || dentry_child2)) 512 return false; 513 /* For a simple request, only check for requested accesses. */ 514 access_masked_parent1 = access_request_parent1; 515 access_masked_parent2 = access_request_parent2; 516 is_dom_check = false; 517 } 518 519 if (unlikely(dentry_child1)) { 520 unmask_layers(find_rule(domain, dentry_child1), 521 init_layer_masks(domain, LANDLOCK_MASK_ACCESS_FS, 522 &_layer_masks_child1), 523 &_layer_masks_child1); 524 layer_masks_child1 = &_layer_masks_child1; 525 child1_is_directory = d_is_dir(dentry_child1); 526 } 527 if (unlikely(dentry_child2)) { 528 unmask_layers(find_rule(domain, dentry_child2), 529 init_layer_masks(domain, LANDLOCK_MASK_ACCESS_FS, 530 &_layer_masks_child2), 531 &_layer_masks_child2); 532 layer_masks_child2 = &_layer_masks_child2; 533 child2_is_directory = d_is_dir(dentry_child2); 534 } 535 536 walker_path = *path; 537 path_get(&walker_path); 538 /* 539 * We need to walk through all the hierarchy to not miss any relevant 540 * restriction. 541 */ 542 while (true) { 543 struct dentry *parent_dentry; 544 const struct landlock_rule *rule; 545 546 /* 547 * If at least all accesses allowed on the destination are 548 * already allowed on the source, respectively if there is at 549 * least as much as restrictions on the destination than on the 550 * source, then we can safely refer files from the source to 551 * the destination without risking a privilege escalation. 552 * This also applies in the case of RENAME_EXCHANGE, which 553 * implies checks on both direction. This is crucial for 554 * standalone multilayered security policies. Furthermore, 555 * this helps avoid policy writers to shoot themselves in the 556 * foot. 557 */ 558 if (unlikely(is_dom_check && 559 no_more_access( 560 layer_masks_parent1, layer_masks_child1, 561 child1_is_directory, layer_masks_parent2, 562 layer_masks_child2, 563 child2_is_directory))) { 564 allowed_parent1 = scope_to_request( 565 access_request_parent1, layer_masks_parent1); 566 allowed_parent2 = scope_to_request( 567 access_request_parent2, layer_masks_parent2); 568 569 /* Stops when all accesses are granted. */ 570 if (allowed_parent1 && allowed_parent2) 571 break; 572 573 /* 574 * Now, downgrades the remaining checks from domain 575 * handled accesses to requested accesses. 576 */ 577 is_dom_check = false; 578 access_masked_parent1 = access_request_parent1; 579 access_masked_parent2 = access_request_parent2; 580 } 581 582 rule = find_rule(domain, walker_path.dentry); 583 allowed_parent1 = unmask_layers(rule, access_masked_parent1, 584 layer_masks_parent1); 585 allowed_parent2 = unmask_layers(rule, access_masked_parent2, 586 layer_masks_parent2); 587 588 /* Stops when a rule from each layer grants access. */ 589 if (allowed_parent1 && allowed_parent2) 590 break; 591 592 jump_up: 593 if (walker_path.dentry == walker_path.mnt->mnt_root) { 594 if (follow_up(&walker_path)) { 595 /* Ignores hidden mount points. */ 596 goto jump_up; 597 } else { 598 /* 599 * Stops at the real root. Denies access 600 * because not all layers have granted access. 601 */ 602 break; 603 } 604 } 605 if (unlikely(IS_ROOT(walker_path.dentry))) { 606 /* 607 * Stops at disconnected root directories. Only allows 608 * access to internal filesystems (e.g. nsfs, which is 609 * reachable through /proc/<pid>/ns/<namespace>). 610 */ 611 allowed_parent1 = allowed_parent2 = 612 !!(walker_path.mnt->mnt_flags & MNT_INTERNAL); 613 break; 614 } 615 parent_dentry = dget_parent(walker_path.dentry); 616 dput(walker_path.dentry); 617 walker_path.dentry = parent_dentry; 618 } 619 path_put(&walker_path); 620 621 return allowed_parent1 && allowed_parent2; 622 } 623 624 static inline int check_access_path(const struct landlock_ruleset *const domain, 625 const struct path *const path, 626 access_mask_t access_request) 627 { 628 layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {}; 629 630 access_request = init_layer_masks(domain, access_request, &layer_masks); 631 if (is_access_to_paths_allowed(domain, path, access_request, 632 &layer_masks, NULL, 0, NULL, NULL)) 633 return 0; 634 return -EACCES; 635 } 636 637 static inline int current_check_access_path(const struct path *const path, 638 const access_mask_t access_request) 639 { 640 const struct landlock_ruleset *const dom = 641 landlock_get_current_domain(); 642 643 if (!dom) 644 return 0; 645 return check_access_path(dom, path, access_request); 646 } 647 648 static inline access_mask_t get_mode_access(const umode_t mode) 649 { 650 switch (mode & S_IFMT) { 651 case S_IFLNK: 652 return LANDLOCK_ACCESS_FS_MAKE_SYM; 653 case 0: 654 /* A zero mode translates to S_IFREG. */ 655 case S_IFREG: 656 return LANDLOCK_ACCESS_FS_MAKE_REG; 657 case S_IFDIR: 658 return LANDLOCK_ACCESS_FS_MAKE_DIR; 659 case S_IFCHR: 660 return LANDLOCK_ACCESS_FS_MAKE_CHAR; 661 case S_IFBLK: 662 return LANDLOCK_ACCESS_FS_MAKE_BLOCK; 663 case S_IFIFO: 664 return LANDLOCK_ACCESS_FS_MAKE_FIFO; 665 case S_IFSOCK: 666 return LANDLOCK_ACCESS_FS_MAKE_SOCK; 667 default: 668 WARN_ON_ONCE(1); 669 return 0; 670 } 671 } 672 673 static inline access_mask_t maybe_remove(const struct dentry *const dentry) 674 { 675 if (d_is_negative(dentry)) 676 return 0; 677 return d_is_dir(dentry) ? LANDLOCK_ACCESS_FS_REMOVE_DIR : 678 LANDLOCK_ACCESS_FS_REMOVE_FILE; 679 } 680 681 /** 682 * collect_domain_accesses - Walk through a file path and collect accesses 683 * 684 * @domain: Domain to check against. 685 * @mnt_root: Last directory to check. 686 * @dir: Directory to start the walk from. 687 * @layer_masks_dom: Where to store the collected accesses. 688 * 689 * This helper is useful to begin a path walk from the @dir directory to a 690 * @mnt_root directory used as a mount point. This mount point is the common 691 * ancestor between the source and the destination of a renamed and linked 692 * file. While walking from @dir to @mnt_root, we record all the domain's 693 * allowed accesses in @layer_masks_dom. 694 * 695 * This is similar to is_access_to_paths_allowed() but much simpler because it 696 * only handles walking on the same mount point and only checks one set of 697 * accesses. 698 * 699 * Returns: 700 * - true if all the domain access rights are allowed for @dir; 701 * - false if the walk reached @mnt_root. 702 */ 703 static bool collect_domain_accesses( 704 const struct landlock_ruleset *const domain, 705 const struct dentry *const mnt_root, struct dentry *dir, 706 layer_mask_t (*const layer_masks_dom)[LANDLOCK_NUM_ACCESS_FS]) 707 { 708 unsigned long access_dom; 709 bool ret = false; 710 711 if (WARN_ON_ONCE(!domain || !mnt_root || !dir || !layer_masks_dom)) 712 return true; 713 if (is_nouser_or_private(dir)) 714 return true; 715 716 access_dom = init_layer_masks(domain, LANDLOCK_MASK_ACCESS_FS, 717 layer_masks_dom); 718 719 dget(dir); 720 while (true) { 721 struct dentry *parent_dentry; 722 723 /* Gets all layers allowing all domain accesses. */ 724 if (unmask_layers(find_rule(domain, dir), access_dom, 725 layer_masks_dom)) { 726 /* 727 * Stops when all handled accesses are allowed by at 728 * least one rule in each layer. 729 */ 730 ret = true; 731 break; 732 } 733 734 /* We should not reach a root other than @mnt_root. */ 735 if (dir == mnt_root || WARN_ON_ONCE(IS_ROOT(dir))) 736 break; 737 738 parent_dentry = dget_parent(dir); 739 dput(dir); 740 dir = parent_dentry; 741 } 742 dput(dir); 743 return ret; 744 } 745 746 /** 747 * current_check_refer_path - Check if a rename or link action is allowed 748 * 749 * @old_dentry: File or directory requested to be moved or linked. 750 * @new_dir: Destination parent directory. 751 * @new_dentry: Destination file or directory. 752 * @removable: Sets to true if it is a rename operation. 753 * @exchange: Sets to true if it is a rename operation with RENAME_EXCHANGE. 754 * 755 * Because of its unprivileged constraints, Landlock relies on file hierarchies 756 * (and not only inodes) to tie access rights to files. Being able to link or 757 * rename a file hierarchy brings some challenges. Indeed, moving or linking a 758 * file (i.e. creating a new reference to an inode) can have an impact on the 759 * actions allowed for a set of files if it would change its parent directory 760 * (i.e. reparenting). 761 * 762 * To avoid trivial access right bypasses, Landlock first checks if the file or 763 * directory requested to be moved would gain new access rights inherited from 764 * its new hierarchy. Before returning any error, Landlock then checks that 765 * the parent source hierarchy and the destination hierarchy would allow the 766 * link or rename action. If it is not the case, an error with EACCES is 767 * returned to inform user space that there is no way to remove or create the 768 * requested source file type. If it should be allowed but the new inherited 769 * access rights would be greater than the source access rights, then the 770 * kernel returns an error with EXDEV. Prioritizing EACCES over EXDEV enables 771 * user space to abort the whole operation if there is no way to do it, or to 772 * manually copy the source to the destination if this remains allowed, e.g. 773 * because file creation is allowed on the destination directory but not direct 774 * linking. 775 * 776 * To achieve this goal, the kernel needs to compare two file hierarchies: the 777 * one identifying the source file or directory (including itself), and the 778 * destination one. This can be seen as a multilayer partial ordering problem. 779 * The kernel walks through these paths and collects in a matrix the access 780 * rights that are denied per layer. These matrices are then compared to see 781 * if the destination one has more (or the same) restrictions as the source 782 * one. If this is the case, the requested action will not return EXDEV, which 783 * doesn't mean the action is allowed. The parent hierarchy of the source 784 * (i.e. parent directory), and the destination hierarchy must also be checked 785 * to verify that they explicitly allow such action (i.e. referencing, 786 * creation and potentially removal rights). The kernel implementation is then 787 * required to rely on potentially four matrices of access rights: one for the 788 * source file or directory (i.e. the child), a potentially other one for the 789 * other source/destination (in case of RENAME_EXCHANGE), one for the source 790 * parent hierarchy and a last one for the destination hierarchy. These 791 * ephemeral matrices take some space on the stack, which limits the number of 792 * layers to a deemed reasonable number: 16. 793 * 794 * Returns: 795 * - 0 if access is allowed; 796 * - -EXDEV if @old_dentry would inherit new access rights from @new_dir; 797 * - -EACCES if file removal or creation is denied. 798 */ 799 static int current_check_refer_path(struct dentry *const old_dentry, 800 const struct path *const new_dir, 801 struct dentry *const new_dentry, 802 const bool removable, const bool exchange) 803 { 804 const struct landlock_ruleset *const dom = 805 landlock_get_current_domain(); 806 bool allow_parent1, allow_parent2; 807 access_mask_t access_request_parent1, access_request_parent2; 808 struct path mnt_dir; 809 layer_mask_t layer_masks_parent1[LANDLOCK_NUM_ACCESS_FS], 810 layer_masks_parent2[LANDLOCK_NUM_ACCESS_FS]; 811 812 if (!dom) 813 return 0; 814 if (WARN_ON_ONCE(dom->num_layers < 1)) 815 return -EACCES; 816 if (unlikely(d_is_negative(old_dentry))) 817 return -ENOENT; 818 if (exchange) { 819 if (unlikely(d_is_negative(new_dentry))) 820 return -ENOENT; 821 access_request_parent1 = 822 get_mode_access(d_backing_inode(new_dentry)->i_mode); 823 } else { 824 access_request_parent1 = 0; 825 } 826 access_request_parent2 = 827 get_mode_access(d_backing_inode(old_dentry)->i_mode); 828 if (removable) { 829 access_request_parent1 |= maybe_remove(old_dentry); 830 access_request_parent2 |= maybe_remove(new_dentry); 831 } 832 833 /* The mount points are the same for old and new paths, cf. EXDEV. */ 834 if (old_dentry->d_parent == new_dir->dentry) { 835 /* 836 * The LANDLOCK_ACCESS_FS_REFER access right is not required 837 * for same-directory referer (i.e. no reparenting). 838 */ 839 access_request_parent1 = init_layer_masks( 840 dom, access_request_parent1 | access_request_parent2, 841 &layer_masks_parent1); 842 if (is_access_to_paths_allowed( 843 dom, new_dir, access_request_parent1, 844 &layer_masks_parent1, NULL, 0, NULL, NULL)) 845 return 0; 846 return -EACCES; 847 } 848 849 access_request_parent1 |= LANDLOCK_ACCESS_FS_REFER; 850 access_request_parent2 |= LANDLOCK_ACCESS_FS_REFER; 851 852 /* Saves the common mount point. */ 853 mnt_dir.mnt = new_dir->mnt; 854 mnt_dir.dentry = new_dir->mnt->mnt_root; 855 856 /* new_dir->dentry is equal to new_dentry->d_parent */ 857 allow_parent1 = collect_domain_accesses(dom, mnt_dir.dentry, 858 old_dentry->d_parent, 859 &layer_masks_parent1); 860 allow_parent2 = collect_domain_accesses( 861 dom, mnt_dir.dentry, new_dir->dentry, &layer_masks_parent2); 862 863 if (allow_parent1 && allow_parent2) 864 return 0; 865 866 /* 867 * To be able to compare source and destination domain access rights, 868 * take into account the @old_dentry access rights aggregated with its 869 * parent access rights. This will be useful to compare with the 870 * destination parent access rights. 871 */ 872 if (is_access_to_paths_allowed( 873 dom, &mnt_dir, access_request_parent1, &layer_masks_parent1, 874 old_dentry, access_request_parent2, &layer_masks_parent2, 875 exchange ? new_dentry : NULL)) 876 return 0; 877 878 /* 879 * This prioritizes EACCES over EXDEV for all actions, including 880 * renames with RENAME_EXCHANGE. 881 */ 882 if (likely(is_eacces(&layer_masks_parent1, access_request_parent1) || 883 is_eacces(&layer_masks_parent2, access_request_parent2))) 884 return -EACCES; 885 886 /* 887 * Gracefully forbids reparenting if the destination directory 888 * hierarchy is not a superset of restrictions of the source directory 889 * hierarchy, or if LANDLOCK_ACCESS_FS_REFER is not allowed by the 890 * source or the destination. 891 */ 892 return -EXDEV; 893 } 894 895 /* Inode hooks */ 896 897 static void hook_inode_free_security(struct inode *const inode) 898 { 899 /* 900 * All inodes must already have been untied from their object by 901 * release_inode() or hook_sb_delete(). 902 */ 903 WARN_ON_ONCE(landlock_inode(inode)->object); 904 } 905 906 /* Super-block hooks */ 907 908 /* 909 * Release the inodes used in a security policy. 910 * 911 * Cf. fsnotify_unmount_inodes() and invalidate_inodes() 912 */ 913 static void hook_sb_delete(struct super_block *const sb) 914 { 915 struct inode *inode, *prev_inode = NULL; 916 917 if (!landlock_initialized) 918 return; 919 920 spin_lock(&sb->s_inode_list_lock); 921 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 922 struct landlock_object *object; 923 924 /* Only handles referenced inodes. */ 925 if (!atomic_read(&inode->i_count)) 926 continue; 927 928 /* 929 * Protects against concurrent modification of inode (e.g. 930 * from get_inode_object()). 931 */ 932 spin_lock(&inode->i_lock); 933 /* 934 * Checks I_FREEING and I_WILL_FREE to protect against a race 935 * condition when release_inode() just called iput(), which 936 * could lead to a NULL dereference of inode->security or a 937 * second call to iput() for the same Landlock object. Also 938 * checks I_NEW because such inode cannot be tied to an object. 939 */ 940 if (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW)) { 941 spin_unlock(&inode->i_lock); 942 continue; 943 } 944 945 rcu_read_lock(); 946 object = rcu_dereference(landlock_inode(inode)->object); 947 if (!object) { 948 rcu_read_unlock(); 949 spin_unlock(&inode->i_lock); 950 continue; 951 } 952 /* Keeps a reference to this inode until the next loop walk. */ 953 __iget(inode); 954 spin_unlock(&inode->i_lock); 955 956 /* 957 * If there is no concurrent release_inode() ongoing, then we 958 * are in charge of calling iput() on this inode, otherwise we 959 * will just wait for it to finish. 960 */ 961 spin_lock(&object->lock); 962 if (object->underobj == inode) { 963 object->underobj = NULL; 964 spin_unlock(&object->lock); 965 rcu_read_unlock(); 966 967 /* 968 * Because object->underobj was not NULL, 969 * release_inode() and get_inode_object() guarantee 970 * that it is safe to reset 971 * landlock_inode(inode)->object while it is not NULL. 972 * It is therefore not necessary to lock inode->i_lock. 973 */ 974 rcu_assign_pointer(landlock_inode(inode)->object, NULL); 975 /* 976 * At this point, we own the ihold() reference that was 977 * originally set up by get_inode_object() and the 978 * __iget() reference that we just set in this loop 979 * walk. Therefore the following call to iput() will 980 * not sleep nor drop the inode because there is now at 981 * least two references to it. 982 */ 983 iput(inode); 984 } else { 985 spin_unlock(&object->lock); 986 rcu_read_unlock(); 987 } 988 989 if (prev_inode) { 990 /* 991 * At this point, we still own the __iget() reference 992 * that we just set in this loop walk. Therefore we 993 * can drop the list lock and know that the inode won't 994 * disappear from under us until the next loop walk. 995 */ 996 spin_unlock(&sb->s_inode_list_lock); 997 /* 998 * We can now actually put the inode reference from the 999 * previous loop walk, which is not needed anymore. 1000 */ 1001 iput(prev_inode); 1002 cond_resched(); 1003 spin_lock(&sb->s_inode_list_lock); 1004 } 1005 prev_inode = inode; 1006 } 1007 spin_unlock(&sb->s_inode_list_lock); 1008 1009 /* Puts the inode reference from the last loop walk, if any. */ 1010 if (prev_inode) 1011 iput(prev_inode); 1012 /* Waits for pending iput() in release_inode(). */ 1013 wait_var_event(&landlock_superblock(sb)->inode_refs, 1014 !atomic_long_read(&landlock_superblock(sb)->inode_refs)); 1015 } 1016 1017 /* 1018 * Because a Landlock security policy is defined according to the filesystem 1019 * topology (i.e. the mount namespace), changing it may grant access to files 1020 * not previously allowed. 1021 * 1022 * To make it simple, deny any filesystem topology modification by landlocked 1023 * processes. Non-landlocked processes may still change the namespace of a 1024 * landlocked process, but this kind of threat must be handled by a system-wide 1025 * access-control security policy. 1026 * 1027 * This could be lifted in the future if Landlock can safely handle mount 1028 * namespace updates requested by a landlocked process. Indeed, we could 1029 * update the current domain (which is currently read-only) by taking into 1030 * account the accesses of the source and the destination of a new mount point. 1031 * However, it would also require to make all the child domains dynamically 1032 * inherit these new constraints. Anyway, for backward compatibility reasons, 1033 * a dedicated user space option would be required (e.g. as a ruleset flag). 1034 */ 1035 static int hook_sb_mount(const char *const dev_name, 1036 const struct path *const path, const char *const type, 1037 const unsigned long flags, void *const data) 1038 { 1039 if (!landlock_get_current_domain()) 1040 return 0; 1041 return -EPERM; 1042 } 1043 1044 static int hook_move_mount(const struct path *const from_path, 1045 const struct path *const to_path) 1046 { 1047 if (!landlock_get_current_domain()) 1048 return 0; 1049 return -EPERM; 1050 } 1051 1052 /* 1053 * Removing a mount point may reveal a previously hidden file hierarchy, which 1054 * may then grant access to files, which may have previously been forbidden. 1055 */ 1056 static int hook_sb_umount(struct vfsmount *const mnt, const int flags) 1057 { 1058 if (!landlock_get_current_domain()) 1059 return 0; 1060 return -EPERM; 1061 } 1062 1063 static int hook_sb_remount(struct super_block *const sb, void *const mnt_opts) 1064 { 1065 if (!landlock_get_current_domain()) 1066 return 0; 1067 return -EPERM; 1068 } 1069 1070 /* 1071 * pivot_root(2), like mount(2), changes the current mount namespace. It must 1072 * then be forbidden for a landlocked process. 1073 * 1074 * However, chroot(2) may be allowed because it only changes the relative root 1075 * directory of the current process. Moreover, it can be used to restrict the 1076 * view of the filesystem. 1077 */ 1078 static int hook_sb_pivotroot(const struct path *const old_path, 1079 const struct path *const new_path) 1080 { 1081 if (!landlock_get_current_domain()) 1082 return 0; 1083 return -EPERM; 1084 } 1085 1086 /* Path hooks */ 1087 1088 static int hook_path_link(struct dentry *const old_dentry, 1089 const struct path *const new_dir, 1090 struct dentry *const new_dentry) 1091 { 1092 return current_check_refer_path(old_dentry, new_dir, new_dentry, false, 1093 false); 1094 } 1095 1096 static int hook_path_rename(const struct path *const old_dir, 1097 struct dentry *const old_dentry, 1098 const struct path *const new_dir, 1099 struct dentry *const new_dentry, 1100 const unsigned int flags) 1101 { 1102 /* old_dir refers to old_dentry->d_parent and new_dir->mnt */ 1103 return current_check_refer_path(old_dentry, new_dir, new_dentry, true, 1104 !!(flags & RENAME_EXCHANGE)); 1105 } 1106 1107 static int hook_path_mkdir(const struct path *const dir, 1108 struct dentry *const dentry, const umode_t mode) 1109 { 1110 return current_check_access_path(dir, LANDLOCK_ACCESS_FS_MAKE_DIR); 1111 } 1112 1113 static int hook_path_mknod(const struct path *const dir, 1114 struct dentry *const dentry, const umode_t mode, 1115 const unsigned int dev) 1116 { 1117 const struct landlock_ruleset *const dom = 1118 landlock_get_current_domain(); 1119 1120 if (!dom) 1121 return 0; 1122 return check_access_path(dom, dir, get_mode_access(mode)); 1123 } 1124 1125 static int hook_path_symlink(const struct path *const dir, 1126 struct dentry *const dentry, 1127 const char *const old_name) 1128 { 1129 return current_check_access_path(dir, LANDLOCK_ACCESS_FS_MAKE_SYM); 1130 } 1131 1132 static int hook_path_unlink(const struct path *const dir, 1133 struct dentry *const dentry) 1134 { 1135 return current_check_access_path(dir, LANDLOCK_ACCESS_FS_REMOVE_FILE); 1136 } 1137 1138 static int hook_path_rmdir(const struct path *const dir, 1139 struct dentry *const dentry) 1140 { 1141 return current_check_access_path(dir, LANDLOCK_ACCESS_FS_REMOVE_DIR); 1142 } 1143 1144 /* File hooks */ 1145 1146 static inline access_mask_t get_file_access(const struct file *const file) 1147 { 1148 access_mask_t access = 0; 1149 1150 if (file->f_mode & FMODE_READ) { 1151 /* A directory can only be opened in read mode. */ 1152 if (S_ISDIR(file_inode(file)->i_mode)) 1153 return LANDLOCK_ACCESS_FS_READ_DIR; 1154 access = LANDLOCK_ACCESS_FS_READ_FILE; 1155 } 1156 if (file->f_mode & FMODE_WRITE) 1157 access |= LANDLOCK_ACCESS_FS_WRITE_FILE; 1158 /* __FMODE_EXEC is indeed part of f_flags, not f_mode. */ 1159 if (file->f_flags & __FMODE_EXEC) 1160 access |= LANDLOCK_ACCESS_FS_EXECUTE; 1161 return access; 1162 } 1163 1164 static int hook_file_open(struct file *const file) 1165 { 1166 const struct landlock_ruleset *const dom = 1167 landlock_get_current_domain(); 1168 1169 if (!dom) 1170 return 0; 1171 /* 1172 * Because a file may be opened with O_PATH, get_file_access() may 1173 * return 0. This case will be handled with a future Landlock 1174 * evolution. 1175 */ 1176 return check_access_path(dom, &file->f_path, get_file_access(file)); 1177 } 1178 1179 static struct security_hook_list landlock_hooks[] __lsm_ro_after_init = { 1180 LSM_HOOK_INIT(inode_free_security, hook_inode_free_security), 1181 1182 LSM_HOOK_INIT(sb_delete, hook_sb_delete), 1183 LSM_HOOK_INIT(sb_mount, hook_sb_mount), 1184 LSM_HOOK_INIT(move_mount, hook_move_mount), 1185 LSM_HOOK_INIT(sb_umount, hook_sb_umount), 1186 LSM_HOOK_INIT(sb_remount, hook_sb_remount), 1187 LSM_HOOK_INIT(sb_pivotroot, hook_sb_pivotroot), 1188 1189 LSM_HOOK_INIT(path_link, hook_path_link), 1190 LSM_HOOK_INIT(path_rename, hook_path_rename), 1191 LSM_HOOK_INIT(path_mkdir, hook_path_mkdir), 1192 LSM_HOOK_INIT(path_mknod, hook_path_mknod), 1193 LSM_HOOK_INIT(path_symlink, hook_path_symlink), 1194 LSM_HOOK_INIT(path_unlink, hook_path_unlink), 1195 LSM_HOOK_INIT(path_rmdir, hook_path_rmdir), 1196 1197 LSM_HOOK_INIT(file_open, hook_file_open), 1198 }; 1199 1200 __init void landlock_add_fs_hooks(void) 1201 { 1202 security_add_hooks(landlock_hooks, ARRAY_SIZE(landlock_hooks), 1203 LANDLOCK_NAME); 1204 } 1205