1 // SPDX-License-Identifier: GPL-2.0-only
3 * Landlock LSM - Filesystem management and hooks
5 * Copyright © 2016-2020 Mickaël Salaün <mic@digikod.net>
6 * Copyright © 2018-2020 ANSSI
7 * Copyright © 2021-2022 Microsoft Corporation
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>
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>
41 /* Underlying object management */
43 static void release_inode(struct landlock_object *const object)
44 __releases(object->lock)
46 struct inode *const inode = object->underobj;
47 struct super_block *sb;
50 spin_unlock(&object->lock);
55 * Protects against concurrent use by hook_sb_delete() of the reference
56 * to the underlying inode.
58 object->underobj = NULL;
60 * Makes sure that if the filesystem is concurrently unmounted,
61 * hook_sb_delete() will wait for us to finish iput().
64 atomic_long_inc(&landlock_superblock(sb)->inode_refs);
65 spin_unlock(&object->lock);
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.
72 rcu_assign_pointer(landlock_inode(inode)->object, NULL);
74 * Now, new rules can safely be tied to @inode with get_inode_object().
78 if (atomic_long_dec_and_test(&landlock_superblock(sb)->inode_refs))
79 wake_up_var(&landlock_superblock(sb)->inode_refs);
82 static const struct landlock_object_underops landlock_fs_underops = {
83 .release = release_inode
86 /* Ruleset management */
88 static struct landlock_object *get_inode_object(struct inode *const inode)
90 struct landlock_object *object, *new_object;
91 struct landlock_inode_security *inode_sec = landlock_inode(inode);
95 object = rcu_dereference(inode_sec->object);
97 if (likely(refcount_inc_not_zero(&object->usage))) {
102 * We are racing with release_inode(), the object is going
103 * away. Wait for release_inode(), then retry.
105 spin_lock(&object->lock);
106 spin_unlock(&object->lock);
112 * If there is no object tied to @inode, then create a new one (without
113 * holding any locks).
115 new_object = landlock_create_object(&landlock_fs_underops, inode);
116 if (IS_ERR(new_object))
120 * Protects against concurrent calls to get_inode_object() or
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);
134 * @inode will be released by hook_sb_delete() on its superblock
135 * shutdown, or by release_inode() when no more ruleset references the
139 rcu_assign_pointer(inode_sec->object, new_object);
140 spin_unlock(&inode->i_lock);
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 */
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.
157 /* clang-format off */
158 #define ACCESS_INITIALLY_DENIED ( \
159 LANDLOCK_ACCESS_FS_REFER)
160 /* clang-format on */
163 * @path: Should have been checked by get_path_from_fd().
165 int landlock_append_fs_rule(struct landlock_ruleset *const ruleset,
166 const struct path *const path,
167 access_mask_t access_rights)
170 struct landlock_object *object;
172 /* Files only get access rights that make sense. */
173 if (!d_is_dir(path->dentry) &&
174 (access_rights | ACCESS_FILE) != ACCESS_FILE)
176 if (WARN_ON_ONCE(ruleset->num_layers != 1))
179 /* Transforms relative access rights to absolute ones. */
181 LANDLOCK_MASK_ACCESS_FS &
182 ~(ruleset->fs_access_masks[0] | ACCESS_INITIALLY_DENIED);
183 object = get_inode_object(d_backing_inode(path->dentry));
185 return PTR_ERR(object);
186 mutex_lock(&ruleset->lock);
187 err = landlock_insert_rule(ruleset, object, access_rights);
188 mutex_unlock(&ruleset->lock);
190 * No need to check for an error because landlock_insert_rule()
191 * increments the refcount for the new object if needed.
193 landlock_put_object(object);
197 /* Access-control management */
200 * The lifetime of the returned rule is tied to @domain.
202 * Returns NULL if no rule is found or if @dentry is negative.
204 static inline const struct landlock_rule *
205 find_rule(const struct landlock_ruleset *const domain,
206 const struct dentry *const dentry)
208 const struct landlock_rule *rule;
209 const struct inode *inode;
211 /* Ignores nonexistent leafs. */
212 if (d_is_negative(dentry))
215 inode = d_backing_inode(dentry);
217 rule = landlock_find_rule(
218 domain, rcu_dereference(landlock_inode(inode)->object));
224 * @layer_masks is read and may be updated according to the access request and
227 * Returns true if the request is allowed (i.e. relevant layer masks for the
228 * request are empty).
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])
237 if (!access_request || !layer_masks)
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>
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;
261 * Records in @layer_masks which layer grants access to each
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];
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>
282 static inline bool is_nouser_or_private(const struct dentry *dentry)
284 return (dentry->d_sb->s_flags & SB_NOUSER) ||
285 (d_is_positive(dentry) &&
286 unlikely(IS_PRIVATE(d_backing_inode(dentry))));
289 static inline access_mask_t
290 get_handled_accesses(const struct landlock_ruleset *const domain)
292 access_mask_t access_dom = ACCESS_INITIALLY_DENIED;
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;
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])
305 access_mask_t handled_accesses = 0;
308 memset(layer_masks, 0, sizeof(*layer_masks));
309 /* An empty access request can happen because of O_WRONLY | O_RDWR. */
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;
318 for_each_set_bit(access_bit, &access_req,
319 ARRAY_SIZE(*layer_masks)) {
321 * Artificially handles all initially denied by default
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);
333 return handled_accesses;
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.
340 * @layer_masks_child2: Optional child masks.
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)
350 unsigned long access_bit;
352 for (access_bit = 0; access_bit < ARRAY_SIZE(*layer_masks_parent2);
354 /* Ignores accesses that only make sense for directories. */
355 const bool is_file_access =
356 !!(BIT_ULL(access_bit) & ACCESS_FILE);
358 if (child1_is_directory || is_file_access) {
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)
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])
372 if (!layer_masks_child2)
374 if (child2_is_directory || is_file_access) {
376 * Checks inverted restrictions for RENAME_EXCHANGE:
377 * restrictions(parent1) >= restrictions(child2)
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])
390 * Removes @layer_masks accesses that are not requested.
392 * Returns true if the request is allowed, false otherwise.
395 scope_to_request(const access_mask_t access_request,
396 layer_mask_t (*const layer_masks)[LANDLOCK_NUM_ACCESS_FS])
398 const unsigned long access_req = access_request;
399 unsigned long access_bit;
401 if (WARN_ON_ONCE(!layer_masks))
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));
410 * Returns true if there is at least one access right different than
411 * LANDLOCK_ACCESS_FS_REFER.
414 is_eacces(const layer_mask_t (*const layer_masks)[LANDLOCK_NUM_ACCESS_FS],
415 const access_mask_t access_request)
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;
425 for_each_set_bit(access_bit, &access_check, ARRAY_SIZE(*layer_masks)) {
426 if ((*layer_masks)[access_bit])
433 * check_access_path_dual - Check accesses for requests with a common path
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
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
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
468 * - 0 if the access request is granted;
469 * - -EACCES if it is denied because of access right other than
470 * LANDLOCK_ACCESS_FS_REFER;
471 * - -EXDEV if the renaming or linking would be a privileged escalation
472 * (according to each layered policies), or if LANDLOCK_ACCESS_FS_REFER is
473 * not allowed by the source or the destination.
475 static int check_access_path_dual(
476 const struct landlock_ruleset *const domain,
477 const struct path *const path,
478 const access_mask_t access_request_parent1,
479 layer_mask_t (*const layer_masks_parent1)[LANDLOCK_NUM_ACCESS_FS],
480 const struct dentry *const dentry_child1,
481 const access_mask_t access_request_parent2,
482 layer_mask_t (*const layer_masks_parent2)[LANDLOCK_NUM_ACCESS_FS],
483 const struct dentry *const dentry_child2)
485 bool allowed_parent1 = false, allowed_parent2 = false, is_dom_check,
486 child1_is_directory = true, child2_is_directory = true;
487 struct path walker_path;
488 access_mask_t access_masked_parent1, access_masked_parent2;
489 layer_mask_t _layer_masks_child1[LANDLOCK_NUM_ACCESS_FS],
490 _layer_masks_child2[LANDLOCK_NUM_ACCESS_FS];
491 layer_mask_t(*layer_masks_child1)[LANDLOCK_NUM_ACCESS_FS] = NULL,
492 (*layer_masks_child2)[LANDLOCK_NUM_ACCESS_FS] = NULL;
494 if (!access_request_parent1 && !access_request_parent2)
496 if (WARN_ON_ONCE(!domain || !path))
498 if (is_nouser_or_private(path->dentry))
500 if (WARN_ON_ONCE(domain->num_layers < 1 || !layer_masks_parent1))
503 if (unlikely(layer_masks_parent2)) {
504 if (WARN_ON_ONCE(!dentry_child1))
507 * For a double request, first check for potential privilege
508 * escalation by looking at domain handled accesses (which are
509 * a superset of the meaningful requested accesses).
511 access_masked_parent1 = access_masked_parent2 =
512 get_handled_accesses(domain);
515 if (WARN_ON_ONCE(dentry_child1 || dentry_child2))
517 /* For a simple request, only check for requested accesses. */
518 access_masked_parent1 = access_request_parent1;
519 access_masked_parent2 = access_request_parent2;
520 is_dom_check = false;
523 if (unlikely(dentry_child1)) {
524 unmask_layers(find_rule(domain, dentry_child1),
525 init_layer_masks(domain, LANDLOCK_MASK_ACCESS_FS,
526 &_layer_masks_child1),
527 &_layer_masks_child1);
528 layer_masks_child1 = &_layer_masks_child1;
529 child1_is_directory = d_is_dir(dentry_child1);
531 if (unlikely(dentry_child2)) {
532 unmask_layers(find_rule(domain, dentry_child2),
533 init_layer_masks(domain, LANDLOCK_MASK_ACCESS_FS,
534 &_layer_masks_child2),
535 &_layer_masks_child2);
536 layer_masks_child2 = &_layer_masks_child2;
537 child2_is_directory = d_is_dir(dentry_child2);
541 path_get(&walker_path);
543 * We need to walk through all the hierarchy to not miss any relevant
547 struct dentry *parent_dentry;
548 const struct landlock_rule *rule;
551 * If at least all accesses allowed on the destination are
552 * already allowed on the source, respectively if there is at
553 * least as much as restrictions on the destination than on the
554 * source, then we can safely refer files from the source to
555 * the destination without risking a privilege escalation.
556 * This also applies in the case of RENAME_EXCHANGE, which
557 * implies checks on both direction. This is crucial for
558 * standalone multilayered security policies. Furthermore,
559 * this helps avoid policy writers to shoot themselves in the
562 if (unlikely(is_dom_check &&
564 layer_masks_parent1, layer_masks_child1,
565 child1_is_directory, layer_masks_parent2,
567 child2_is_directory))) {
568 allowed_parent1 = scope_to_request(
569 access_request_parent1, layer_masks_parent1);
570 allowed_parent2 = scope_to_request(
571 access_request_parent2, layer_masks_parent2);
573 /* Stops when all accesses are granted. */
574 if (allowed_parent1 && allowed_parent2)
578 * Now, downgrades the remaining checks from domain
579 * handled accesses to requested accesses.
581 is_dom_check = false;
582 access_masked_parent1 = access_request_parent1;
583 access_masked_parent2 = access_request_parent2;
586 rule = find_rule(domain, walker_path.dentry);
587 allowed_parent1 = unmask_layers(rule, access_masked_parent1,
588 layer_masks_parent1);
589 allowed_parent2 = unmask_layers(rule, access_masked_parent2,
590 layer_masks_parent2);
592 /* Stops when a rule from each layer grants access. */
593 if (allowed_parent1 && allowed_parent2)
597 if (walker_path.dentry == walker_path.mnt->mnt_root) {
598 if (follow_up(&walker_path)) {
599 /* Ignores hidden mount points. */
603 * Stops at the real root. Denies access
604 * because not all layers have granted access.
609 if (unlikely(IS_ROOT(walker_path.dentry))) {
611 * Stops at disconnected root directories. Only allows
612 * access to internal filesystems (e.g. nsfs, which is
613 * reachable through /proc/<pid>/ns/<namespace>).
615 allowed_parent1 = allowed_parent2 =
616 !!(walker_path.mnt->mnt_flags & MNT_INTERNAL);
619 parent_dentry = dget_parent(walker_path.dentry);
620 dput(walker_path.dentry);
621 walker_path.dentry = parent_dentry;
623 path_put(&walker_path);
625 if (allowed_parent1 && allowed_parent2)
629 * This prioritizes EACCES over EXDEV for all actions, including
630 * renames with RENAME_EXCHANGE.
632 if (likely(is_eacces(layer_masks_parent1, access_request_parent1) ||
633 is_eacces(layer_masks_parent2, access_request_parent2)))
637 * Gracefully forbids reparenting if the destination directory
638 * hierarchy is not a superset of restrictions of the source directory
639 * hierarchy, or if LANDLOCK_ACCESS_FS_REFER is not allowed by the
640 * source or the destination.
645 static inline int check_access_path(const struct landlock_ruleset *const domain,
646 const struct path *const path,
647 access_mask_t access_request)
649 layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {};
651 access_request = init_layer_masks(domain, access_request, &layer_masks);
652 return check_access_path_dual(domain, path, access_request,
653 &layer_masks, NULL, 0, NULL, NULL);
656 static inline int current_check_access_path(const struct path *const path,
657 const access_mask_t access_request)
659 const struct landlock_ruleset *const dom =
660 landlock_get_current_domain();
664 return check_access_path(dom, path, access_request);
667 static inline access_mask_t get_mode_access(const umode_t mode)
669 switch (mode & S_IFMT) {
671 return LANDLOCK_ACCESS_FS_MAKE_SYM;
673 /* A zero mode translates to S_IFREG. */
675 return LANDLOCK_ACCESS_FS_MAKE_REG;
677 return LANDLOCK_ACCESS_FS_MAKE_DIR;
679 return LANDLOCK_ACCESS_FS_MAKE_CHAR;
681 return LANDLOCK_ACCESS_FS_MAKE_BLOCK;
683 return LANDLOCK_ACCESS_FS_MAKE_FIFO;
685 return LANDLOCK_ACCESS_FS_MAKE_SOCK;
692 static inline access_mask_t maybe_remove(const struct dentry *const dentry)
694 if (d_is_negative(dentry))
696 return d_is_dir(dentry) ? LANDLOCK_ACCESS_FS_REMOVE_DIR :
697 LANDLOCK_ACCESS_FS_REMOVE_FILE;
701 * collect_domain_accesses - Walk through a file path and collect accesses
703 * @domain: Domain to check against.
704 * @mnt_root: Last directory to check.
705 * @dir: Directory to start the walk from.
706 * @layer_masks_dom: Where to store the collected accesses.
708 * This helper is useful to begin a path walk from the @dir directory to a
709 * @mnt_root directory used as a mount point. This mount point is the common
710 * ancestor between the source and the destination of a renamed and linked
711 * file. While walking from @dir to @mnt_root, we record all the domain's
712 * allowed accesses in @layer_masks_dom.
714 * This is similar to check_access_path_dual() but much simpler because it only
715 * handles walking on the same mount point and only checks one set of accesses.
718 * - true if all the domain access rights are allowed for @dir;
719 * - false if the walk reached @mnt_root.
721 static bool collect_domain_accesses(
722 const struct landlock_ruleset *const domain,
723 const struct dentry *const mnt_root, struct dentry *dir,
724 layer_mask_t (*const layer_masks_dom)[LANDLOCK_NUM_ACCESS_FS])
726 unsigned long access_dom;
729 if (WARN_ON_ONCE(!domain || !mnt_root || !dir || !layer_masks_dom))
731 if (is_nouser_or_private(dir))
734 access_dom = init_layer_masks(domain, LANDLOCK_MASK_ACCESS_FS,
739 struct dentry *parent_dentry;
741 /* Gets all layers allowing all domain accesses. */
742 if (unmask_layers(find_rule(domain, dir), access_dom,
745 * Stops when all handled accesses are allowed by at
746 * least one rule in each layer.
752 /* We should not reach a root other than @mnt_root. */
753 if (dir == mnt_root || WARN_ON_ONCE(IS_ROOT(dir)))
756 parent_dentry = dget_parent(dir);
765 * current_check_refer_path - Check if a rename or link action is allowed
767 * @old_dentry: File or directory requested to be moved or linked.
768 * @new_dir: Destination parent directory.
769 * @new_dentry: Destination file or directory.
770 * @removable: Sets to true if it is a rename operation.
771 * @exchange: Sets to true if it is a rename operation with RENAME_EXCHANGE.
773 * Because of its unprivileged constraints, Landlock relies on file hierarchies
774 * (and not only inodes) to tie access rights to files. Being able to link or
775 * rename a file hierarchy brings some challenges. Indeed, moving or linking a
776 * file (i.e. creating a new reference to an inode) can have an impact on the
777 * actions allowed for a set of files if it would change its parent directory
778 * (i.e. reparenting).
780 * To avoid trivial access right bypasses, Landlock first checks if the file or
781 * directory requested to be moved would gain new access rights inherited from
782 * its new hierarchy. Before returning any error, Landlock then checks that
783 * the parent source hierarchy and the destination hierarchy would allow the
784 * link or rename action. If it is not the case, an error with EACCES is
785 * returned to inform user space that there is no way to remove or create the
786 * requested source file type. If it should be allowed but the new inherited
787 * access rights would be greater than the source access rights, then the
788 * kernel returns an error with EXDEV. Prioritizing EACCES over EXDEV enables
789 * user space to abort the whole operation if there is no way to do it, or to
790 * manually copy the source to the destination if this remains allowed, e.g.
791 * because file creation is allowed on the destination directory but not direct
794 * To achieve this goal, the kernel needs to compare two file hierarchies: the
795 * one identifying the source file or directory (including itself), and the
796 * destination one. This can be seen as a multilayer partial ordering problem.
797 * The kernel walks through these paths and collects in a matrix the access
798 * rights that are denied per layer. These matrices are then compared to see
799 * if the destination one has more (or the same) restrictions as the source
800 * one. If this is the case, the requested action will not return EXDEV, which
801 * doesn't mean the action is allowed. The parent hierarchy of the source
802 * (i.e. parent directory), and the destination hierarchy must also be checked
803 * to verify that they explicitly allow such action (i.e. referencing,
804 * creation and potentially removal rights). The kernel implementation is then
805 * required to rely on potentially four matrices of access rights: one for the
806 * source file or directory (i.e. the child), a potentially other one for the
807 * other source/destination (in case of RENAME_EXCHANGE), one for the source
808 * parent hierarchy and a last one for the destination hierarchy. These
809 * ephemeral matrices take some space on the stack, which limits the number of
810 * layers to a deemed reasonable number: 16.
813 * - 0 if access is allowed;
814 * - -EXDEV if @old_dentry would inherit new access rights from @new_dir;
815 * - -EACCES if file removal or creation is denied.
817 static int current_check_refer_path(struct dentry *const old_dentry,
818 const struct path *const new_dir,
819 struct dentry *const new_dentry,
820 const bool removable, const bool exchange)
822 const struct landlock_ruleset *const dom =
823 landlock_get_current_domain();
824 bool allow_parent1, allow_parent2;
825 access_mask_t access_request_parent1, access_request_parent2;
827 layer_mask_t layer_masks_parent1[LANDLOCK_NUM_ACCESS_FS],
828 layer_masks_parent2[LANDLOCK_NUM_ACCESS_FS];
832 if (WARN_ON_ONCE(dom->num_layers < 1))
834 if (unlikely(d_is_negative(old_dentry)))
837 if (unlikely(d_is_negative(new_dentry)))
839 access_request_parent1 =
840 get_mode_access(d_backing_inode(new_dentry)->i_mode);
842 access_request_parent1 = 0;
844 access_request_parent2 =
845 get_mode_access(d_backing_inode(old_dentry)->i_mode);
847 access_request_parent1 |= maybe_remove(old_dentry);
848 access_request_parent2 |= maybe_remove(new_dentry);
851 /* The mount points are the same for old and new paths, cf. EXDEV. */
852 if (old_dentry->d_parent == new_dir->dentry) {
854 * The LANDLOCK_ACCESS_FS_REFER access right is not required
855 * for same-directory referer (i.e. no reparenting).
857 access_request_parent1 = init_layer_masks(
858 dom, access_request_parent1 | access_request_parent2,
859 &layer_masks_parent1);
860 return check_access_path_dual(dom, new_dir,
861 access_request_parent1,
862 &layer_masks_parent1, NULL, 0,
866 access_request_parent1 |= LANDLOCK_ACCESS_FS_REFER;
867 access_request_parent2 |= LANDLOCK_ACCESS_FS_REFER;
869 /* Saves the common mount point. */
870 mnt_dir.mnt = new_dir->mnt;
871 mnt_dir.dentry = new_dir->mnt->mnt_root;
873 /* new_dir->dentry is equal to new_dentry->d_parent */
874 allow_parent1 = collect_domain_accesses(dom, mnt_dir.dentry,
875 old_dentry->d_parent,
876 &layer_masks_parent1);
877 allow_parent2 = collect_domain_accesses(
878 dom, mnt_dir.dentry, new_dir->dentry, &layer_masks_parent2);
880 if (allow_parent1 && allow_parent2)
884 * To be able to compare source and destination domain access rights,
885 * take into account the @old_dentry access rights aggregated with its
886 * parent access rights. This will be useful to compare with the
887 * destination parent access rights.
889 return check_access_path_dual(dom, &mnt_dir, access_request_parent1,
890 &layer_masks_parent1, old_dentry,
891 access_request_parent2,
892 &layer_masks_parent2,
893 exchange ? new_dentry : NULL);
898 static void hook_inode_free_security(struct inode *const inode)
901 * All inodes must already have been untied from their object by
902 * release_inode() or hook_sb_delete().
904 WARN_ON_ONCE(landlock_inode(inode)->object);
907 /* Super-block hooks */
910 * Release the inodes used in a security policy.
912 * Cf. fsnotify_unmount_inodes() and invalidate_inodes()
914 static void hook_sb_delete(struct super_block *const sb)
916 struct inode *inode, *prev_inode = NULL;
918 if (!landlock_initialized)
921 spin_lock(&sb->s_inode_list_lock);
922 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
923 struct landlock_object *object;
925 /* Only handles referenced inodes. */
926 if (!atomic_read(&inode->i_count))
930 * Protects against concurrent modification of inode (e.g.
931 * from get_inode_object()).
933 spin_lock(&inode->i_lock);
935 * Checks I_FREEING and I_WILL_FREE to protect against a race
936 * condition when release_inode() just called iput(), which
937 * could lead to a NULL dereference of inode->security or a
938 * second call to iput() for the same Landlock object. Also
939 * checks I_NEW because such inode cannot be tied to an object.
941 if (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW)) {
942 spin_unlock(&inode->i_lock);
947 object = rcu_dereference(landlock_inode(inode)->object);
950 spin_unlock(&inode->i_lock);
953 /* Keeps a reference to this inode until the next loop walk. */
955 spin_unlock(&inode->i_lock);
958 * If there is no concurrent release_inode() ongoing, then we
959 * are in charge of calling iput() on this inode, otherwise we
960 * will just wait for it to finish.
962 spin_lock(&object->lock);
963 if (object->underobj == inode) {
964 object->underobj = NULL;
965 spin_unlock(&object->lock);
969 * Because object->underobj was not NULL,
970 * release_inode() and get_inode_object() guarantee
971 * that it is safe to reset
972 * landlock_inode(inode)->object while it is not NULL.
973 * It is therefore not necessary to lock inode->i_lock.
975 rcu_assign_pointer(landlock_inode(inode)->object, NULL);
977 * At this point, we own the ihold() reference that was
978 * originally set up by get_inode_object() and the
979 * __iget() reference that we just set in this loop
980 * walk. Therefore the following call to iput() will
981 * not sleep nor drop the inode because there is now at
982 * least two references to it.
986 spin_unlock(&object->lock);
992 * At this point, we still own the __iget() reference
993 * that we just set in this loop walk. Therefore we
994 * can drop the list lock and know that the inode won't
995 * disappear from under us until the next loop walk.
997 spin_unlock(&sb->s_inode_list_lock);
999 * We can now actually put the inode reference from the
1000 * previous loop walk, which is not needed anymore.
1004 spin_lock(&sb->s_inode_list_lock);
1008 spin_unlock(&sb->s_inode_list_lock);
1010 /* Puts the inode reference from the last loop walk, if any. */
1013 /* Waits for pending iput() in release_inode(). */
1014 wait_var_event(&landlock_superblock(sb)->inode_refs,
1015 !atomic_long_read(&landlock_superblock(sb)->inode_refs));
1019 * Because a Landlock security policy is defined according to the filesystem
1020 * topology (i.e. the mount namespace), changing it may grant access to files
1021 * not previously allowed.
1023 * To make it simple, deny any filesystem topology modification by landlocked
1024 * processes. Non-landlocked processes may still change the namespace of a
1025 * landlocked process, but this kind of threat must be handled by a system-wide
1026 * access-control security policy.
1028 * This could be lifted in the future if Landlock can safely handle mount
1029 * namespace updates requested by a landlocked process. Indeed, we could
1030 * update the current domain (which is currently read-only) by taking into
1031 * account the accesses of the source and the destination of a new mount point.
1032 * However, it would also require to make all the child domains dynamically
1033 * inherit these new constraints. Anyway, for backward compatibility reasons,
1034 * a dedicated user space option would be required (e.g. as a ruleset flag).
1036 static int hook_sb_mount(const char *const dev_name,
1037 const struct path *const path, const char *const type,
1038 const unsigned long flags, void *const data)
1040 if (!landlock_get_current_domain())
1045 static int hook_move_mount(const struct path *const from_path,
1046 const struct path *const to_path)
1048 if (!landlock_get_current_domain())
1054 * Removing a mount point may reveal a previously hidden file hierarchy, which
1055 * may then grant access to files, which may have previously been forbidden.
1057 static int hook_sb_umount(struct vfsmount *const mnt, const int flags)
1059 if (!landlock_get_current_domain())
1064 static int hook_sb_remount(struct super_block *const sb, void *const mnt_opts)
1066 if (!landlock_get_current_domain())
1072 * pivot_root(2), like mount(2), changes the current mount namespace. It must
1073 * then be forbidden for a landlocked process.
1075 * However, chroot(2) may be allowed because it only changes the relative root
1076 * directory of the current process. Moreover, it can be used to restrict the
1077 * view of the filesystem.
1079 static int hook_sb_pivotroot(const struct path *const old_path,
1080 const struct path *const new_path)
1082 if (!landlock_get_current_domain())
1089 static int hook_path_link(struct dentry *const old_dentry,
1090 const struct path *const new_dir,
1091 struct dentry *const new_dentry)
1093 return current_check_refer_path(old_dentry, new_dir, new_dentry, false,
1097 static int hook_path_rename(const struct path *const old_dir,
1098 struct dentry *const old_dentry,
1099 const struct path *const new_dir,
1100 struct dentry *const new_dentry,
1101 const unsigned int flags)
1103 /* old_dir refers to old_dentry->d_parent and new_dir->mnt */
1104 return current_check_refer_path(old_dentry, new_dir, new_dentry, true,
1105 !!(flags & RENAME_EXCHANGE));
1108 static int hook_path_mkdir(const struct path *const dir,
1109 struct dentry *const dentry, const umode_t mode)
1111 return current_check_access_path(dir, LANDLOCK_ACCESS_FS_MAKE_DIR);
1114 static int hook_path_mknod(const struct path *const dir,
1115 struct dentry *const dentry, const umode_t mode,
1116 const unsigned int dev)
1118 const struct landlock_ruleset *const dom =
1119 landlock_get_current_domain();
1123 return check_access_path(dom, dir, get_mode_access(mode));
1126 static int hook_path_symlink(const struct path *const dir,
1127 struct dentry *const dentry,
1128 const char *const old_name)
1130 return current_check_access_path(dir, LANDLOCK_ACCESS_FS_MAKE_SYM);
1133 static int hook_path_unlink(const struct path *const dir,
1134 struct dentry *const dentry)
1136 return current_check_access_path(dir, LANDLOCK_ACCESS_FS_REMOVE_FILE);
1139 static int hook_path_rmdir(const struct path *const dir,
1140 struct dentry *const dentry)
1142 return current_check_access_path(dir, LANDLOCK_ACCESS_FS_REMOVE_DIR);
1147 static inline access_mask_t get_file_access(const struct file *const file)
1149 access_mask_t access = 0;
1151 if (file->f_mode & FMODE_READ) {
1152 /* A directory can only be opened in read mode. */
1153 if (S_ISDIR(file_inode(file)->i_mode))
1154 return LANDLOCK_ACCESS_FS_READ_DIR;
1155 access = LANDLOCK_ACCESS_FS_READ_FILE;
1157 if (file->f_mode & FMODE_WRITE)
1158 access |= LANDLOCK_ACCESS_FS_WRITE_FILE;
1159 /* __FMODE_EXEC is indeed part of f_flags, not f_mode. */
1160 if (file->f_flags & __FMODE_EXEC)
1161 access |= LANDLOCK_ACCESS_FS_EXECUTE;
1165 static int hook_file_open(struct file *const file)
1167 const struct landlock_ruleset *const dom =
1168 landlock_get_current_domain();
1173 * Because a file may be opened with O_PATH, get_file_access() may
1174 * return 0. This case will be handled with a future Landlock
1177 return check_access_path(dom, &file->f_path, get_file_access(file));
1180 static struct security_hook_list landlock_hooks[] __lsm_ro_after_init = {
1181 LSM_HOOK_INIT(inode_free_security, hook_inode_free_security),
1183 LSM_HOOK_INIT(sb_delete, hook_sb_delete),
1184 LSM_HOOK_INIT(sb_mount, hook_sb_mount),
1185 LSM_HOOK_INIT(move_mount, hook_move_mount),
1186 LSM_HOOK_INIT(sb_umount, hook_sb_umount),
1187 LSM_HOOK_INIT(sb_remount, hook_sb_remount),
1188 LSM_HOOK_INIT(sb_pivotroot, hook_sb_pivotroot),
1190 LSM_HOOK_INIT(path_link, hook_path_link),
1191 LSM_HOOK_INIT(path_rename, hook_path_rename),
1192 LSM_HOOK_INIT(path_mkdir, hook_path_mkdir),
1193 LSM_HOOK_INIT(path_mknod, hook_path_mknod),
1194 LSM_HOOK_INIT(path_symlink, hook_path_symlink),
1195 LSM_HOOK_INIT(path_unlink, hook_path_unlink),
1196 LSM_HOOK_INIT(path_rmdir, hook_path_rmdir),
1198 LSM_HOOK_INIT(file_open, hook_file_open),
1201 __init void landlock_add_fs_hooks(void)
1203 security_add_hooks(landlock_hooks, ARRAY_SIZE(landlock_hooks),