xref: /linux/security/landlock/fs.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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  * Copyright © 2022 Günther Noack <gnoack3000@gmail.com>
9  * Copyright © 2023-2024 Google LLC
10  */
11 
12 #include <asm/ioctls.h>
13 #include <kunit/test.h>
14 #include <linux/atomic.h>
15 #include <linux/bitops.h>
16 #include <linux/bits.h>
17 #include <linux/compiler_types.h>
18 #include <linux/dcache.h>
19 #include <linux/err.h>
20 #include <linux/falloc.h>
21 #include <linux/fs.h>
22 #include <linux/init.h>
23 #include <linux/kernel.h>
24 #include <linux/limits.h>
25 #include <linux/list.h>
26 #include <linux/lsm_hooks.h>
27 #include <linux/mount.h>
28 #include <linux/namei.h>
29 #include <linux/path.h>
30 #include <linux/rcupdate.h>
31 #include <linux/spinlock.h>
32 #include <linux/stat.h>
33 #include <linux/types.h>
34 #include <linux/wait_bit.h>
35 #include <linux/workqueue.h>
36 #include <uapi/linux/fiemap.h>
37 #include <uapi/linux/landlock.h>
38 
39 #include "common.h"
40 #include "cred.h"
41 #include "fs.h"
42 #include "limits.h"
43 #include "object.h"
44 #include "ruleset.h"
45 #include "setup.h"
46 
47 /* Underlying object management */
48 
49 static void release_inode(struct landlock_object *const object)
50 	__releases(object->lock)
51 {
52 	struct inode *const inode = object->underobj;
53 	struct super_block *sb;
54 
55 	if (!inode) {
56 		spin_unlock(&object->lock);
57 		return;
58 	}
59 
60 	/*
61 	 * Protects against concurrent use by hook_sb_delete() of the reference
62 	 * to the underlying inode.
63 	 */
64 	object->underobj = NULL;
65 	/*
66 	 * Makes sure that if the filesystem is concurrently unmounted,
67 	 * hook_sb_delete() will wait for us to finish iput().
68 	 */
69 	sb = inode->i_sb;
70 	atomic_long_inc(&landlock_superblock(sb)->inode_refs);
71 	spin_unlock(&object->lock);
72 	/*
73 	 * Because object->underobj was not NULL, hook_sb_delete() and
74 	 * get_inode_object() guarantee that it is safe to reset
75 	 * landlock_inode(inode)->object while it is not NULL.  It is therefore
76 	 * not necessary to lock inode->i_lock.
77 	 */
78 	rcu_assign_pointer(landlock_inode(inode)->object, NULL);
79 	/*
80 	 * Now, new rules can safely be tied to @inode with get_inode_object().
81 	 */
82 
83 	iput(inode);
84 	if (atomic_long_dec_and_test(&landlock_superblock(sb)->inode_refs))
85 		wake_up_var(&landlock_superblock(sb)->inode_refs);
86 }
87 
88 static const struct landlock_object_underops landlock_fs_underops = {
89 	.release = release_inode
90 };
91 
92 /* IOCTL helpers */
93 
94 /**
95  * is_masked_device_ioctl - Determine whether an IOCTL command is always
96  * permitted with Landlock for device files.  These commands can not be
97  * restricted on device files by enforcing a Landlock policy.
98  *
99  * @cmd: The IOCTL command that is supposed to be run.
100  *
101  * By default, any IOCTL on a device file requires the
102  * LANDLOCK_ACCESS_FS_IOCTL_DEV right.  However, we blanket-permit some
103  * commands, if:
104  *
105  * 1. The command is implemented in fs/ioctl.c's do_vfs_ioctl(),
106  *    not in f_ops->unlocked_ioctl() or f_ops->compat_ioctl().
107  *
108  * 2. The command is harmless when invoked on devices.
109  *
110  * We also permit commands that do not make sense for devices, but where the
111  * do_vfs_ioctl() implementation returns a more conventional error code.
112  *
113  * Any new IOCTL commands that are implemented in fs/ioctl.c's do_vfs_ioctl()
114  * should be considered for inclusion here.
115  *
116  * Returns: true if the IOCTL @cmd can not be restricted with Landlock for
117  * device files.
118  */
119 static __attribute_const__ bool is_masked_device_ioctl(const unsigned int cmd)
120 {
121 	switch (cmd) {
122 	/*
123 	 * FIOCLEX, FIONCLEX, FIONBIO and FIOASYNC manipulate the FD's
124 	 * close-on-exec and the file's buffered-IO and async flags.  These
125 	 * operations are also available through fcntl(2), and are
126 	 * unconditionally permitted in Landlock.
127 	 */
128 	case FIOCLEX:
129 	case FIONCLEX:
130 	case FIONBIO:
131 	case FIOASYNC:
132 	/*
133 	 * FIOQSIZE queries the size of a regular file, directory, or link.
134 	 *
135 	 * We still permit it, because it always returns -ENOTTY for
136 	 * other file types.
137 	 */
138 	case FIOQSIZE:
139 	/*
140 	 * FIFREEZE and FITHAW freeze and thaw the file system which the
141 	 * given file belongs to.  Requires CAP_SYS_ADMIN.
142 	 *
143 	 * These commands operate on the file system's superblock rather
144 	 * than on the file itself.  The same operations can also be
145 	 * done through any other file or directory on the same file
146 	 * system, so it is safe to permit these.
147 	 */
148 	case FIFREEZE:
149 	case FITHAW:
150 	/*
151 	 * FS_IOC_FIEMAP queries information about the allocation of
152 	 * blocks within a file.
153 	 *
154 	 * This IOCTL command only makes sense for regular files and is
155 	 * not implemented by devices. It is harmless to permit.
156 	 */
157 	case FS_IOC_FIEMAP:
158 	/*
159 	 * FIGETBSZ queries the file system's block size for a file or
160 	 * directory.
161 	 *
162 	 * This command operates on the file system's superblock rather
163 	 * than on the file itself.  The same operation can also be done
164 	 * through any other file or directory on the same file system,
165 	 * so it is safe to permit it.
166 	 */
167 	case FIGETBSZ:
168 	/*
169 	 * FICLONE, FICLONERANGE and FIDEDUPERANGE make files share
170 	 * their underlying storage ("reflink") between source and
171 	 * destination FDs, on file systems which support that.
172 	 *
173 	 * These IOCTL commands only apply to regular files
174 	 * and are harmless to permit for device files.
175 	 */
176 	case FICLONE:
177 	case FICLONERANGE:
178 	case FIDEDUPERANGE:
179 	/*
180 	 * FS_IOC_GETFSUUID and FS_IOC_GETFSSYSFSPATH both operate on
181 	 * the file system superblock, not on the specific file, so
182 	 * these operations are available through any other file on the
183 	 * same file system as well.
184 	 */
185 	case FS_IOC_GETFSUUID:
186 	case FS_IOC_GETFSSYSFSPATH:
187 		return true;
188 
189 	/*
190 	 * FIONREAD, FS_IOC_GETFLAGS, FS_IOC_SETFLAGS, FS_IOC_FSGETXATTR and
191 	 * FS_IOC_FSSETXATTR are forwarded to device implementations.
192 	 */
193 
194 	/*
195 	 * file_ioctl() commands (FIBMAP, FS_IOC_RESVSP, FS_IOC_RESVSP64,
196 	 * FS_IOC_UNRESVSP, FS_IOC_UNRESVSP64 and FS_IOC_ZERO_RANGE) are
197 	 * forwarded to device implementations, so not permitted.
198 	 */
199 
200 	/* Other commands are guarded by the access right. */
201 	default:
202 		return false;
203 	}
204 }
205 
206 /*
207  * is_masked_device_ioctl_compat - same as the helper above, but checking the
208  * "compat" IOCTL commands.
209  *
210  * The IOCTL commands with special handling in compat-mode should behave the
211  * same as their non-compat counterparts.
212  */
213 static __attribute_const__ bool
214 is_masked_device_ioctl_compat(const unsigned int cmd)
215 {
216 	switch (cmd) {
217 	/* FICLONE is permitted, same as in the non-compat variant. */
218 	case FICLONE:
219 		return true;
220 
221 #if defined(CONFIG_X86_64)
222 	/*
223 	 * FS_IOC_RESVSP_32, FS_IOC_RESVSP64_32, FS_IOC_UNRESVSP_32,
224 	 * FS_IOC_UNRESVSP64_32, FS_IOC_ZERO_RANGE_32: not blanket-permitted,
225 	 * for consistency with their non-compat variants.
226 	 */
227 	case FS_IOC_RESVSP_32:
228 	case FS_IOC_RESVSP64_32:
229 	case FS_IOC_UNRESVSP_32:
230 	case FS_IOC_UNRESVSP64_32:
231 	case FS_IOC_ZERO_RANGE_32:
232 #endif
233 
234 	/*
235 	 * FS_IOC32_GETFLAGS, FS_IOC32_SETFLAGS are forwarded to their device
236 	 * implementations.
237 	 */
238 	case FS_IOC32_GETFLAGS:
239 	case FS_IOC32_SETFLAGS:
240 		return false;
241 	default:
242 		return is_masked_device_ioctl(cmd);
243 	}
244 }
245 
246 /* Ruleset management */
247 
248 static struct landlock_object *get_inode_object(struct inode *const inode)
249 {
250 	struct landlock_object *object, *new_object;
251 	struct landlock_inode_security *inode_sec = landlock_inode(inode);
252 
253 	rcu_read_lock();
254 retry:
255 	object = rcu_dereference(inode_sec->object);
256 	if (object) {
257 		if (likely(refcount_inc_not_zero(&object->usage))) {
258 			rcu_read_unlock();
259 			return object;
260 		}
261 		/*
262 		 * We are racing with release_inode(), the object is going
263 		 * away.  Wait for release_inode(), then retry.
264 		 */
265 		spin_lock(&object->lock);
266 		spin_unlock(&object->lock);
267 		goto retry;
268 	}
269 	rcu_read_unlock();
270 
271 	/*
272 	 * If there is no object tied to @inode, then create a new one (without
273 	 * holding any locks).
274 	 */
275 	new_object = landlock_create_object(&landlock_fs_underops, inode);
276 	if (IS_ERR(new_object))
277 		return new_object;
278 
279 	/*
280 	 * Protects against concurrent calls to get_inode_object() or
281 	 * hook_sb_delete().
282 	 */
283 	spin_lock(&inode->i_lock);
284 	if (unlikely(rcu_access_pointer(inode_sec->object))) {
285 		/* Someone else just created the object, bail out and retry. */
286 		spin_unlock(&inode->i_lock);
287 		kfree(new_object);
288 
289 		rcu_read_lock();
290 		goto retry;
291 	}
292 
293 	/*
294 	 * @inode will be released by hook_sb_delete() on its superblock
295 	 * shutdown, or by release_inode() when no more ruleset references the
296 	 * related object.
297 	 */
298 	ihold(inode);
299 	rcu_assign_pointer(inode_sec->object, new_object);
300 	spin_unlock(&inode->i_lock);
301 	return new_object;
302 }
303 
304 /* All access rights that can be tied to files. */
305 /* clang-format off */
306 #define ACCESS_FILE ( \
307 	LANDLOCK_ACCESS_FS_EXECUTE | \
308 	LANDLOCK_ACCESS_FS_WRITE_FILE | \
309 	LANDLOCK_ACCESS_FS_READ_FILE | \
310 	LANDLOCK_ACCESS_FS_TRUNCATE | \
311 	LANDLOCK_ACCESS_FS_IOCTL_DEV)
312 /* clang-format on */
313 
314 /*
315  * @path: Should have been checked by get_path_from_fd().
316  */
317 int landlock_append_fs_rule(struct landlock_ruleset *const ruleset,
318 			    const struct path *const path,
319 			    access_mask_t access_rights)
320 {
321 	int err;
322 	struct landlock_id id = {
323 		.type = LANDLOCK_KEY_INODE,
324 	};
325 
326 	/* Files only get access rights that make sense. */
327 	if (!d_is_dir(path->dentry) &&
328 	    (access_rights | ACCESS_FILE) != ACCESS_FILE)
329 		return -EINVAL;
330 	if (WARN_ON_ONCE(ruleset->num_layers != 1))
331 		return -EINVAL;
332 
333 	/* Transforms relative access rights to absolute ones. */
334 	access_rights |= LANDLOCK_MASK_ACCESS_FS &
335 			 ~landlock_get_fs_access_mask(ruleset, 0);
336 	id.key.object = get_inode_object(d_backing_inode(path->dentry));
337 	if (IS_ERR(id.key.object))
338 		return PTR_ERR(id.key.object);
339 	mutex_lock(&ruleset->lock);
340 	err = landlock_insert_rule(ruleset, id, access_rights);
341 	mutex_unlock(&ruleset->lock);
342 	/*
343 	 * No need to check for an error because landlock_insert_rule()
344 	 * increments the refcount for the new object if needed.
345 	 */
346 	landlock_put_object(id.key.object);
347 	return err;
348 }
349 
350 /* Access-control management */
351 
352 /*
353  * The lifetime of the returned rule is tied to @domain.
354  *
355  * Returns NULL if no rule is found or if @dentry is negative.
356  */
357 static const struct landlock_rule *
358 find_rule(const struct landlock_ruleset *const domain,
359 	  const struct dentry *const dentry)
360 {
361 	const struct landlock_rule *rule;
362 	const struct inode *inode;
363 	struct landlock_id id = {
364 		.type = LANDLOCK_KEY_INODE,
365 	};
366 
367 	/* Ignores nonexistent leafs. */
368 	if (d_is_negative(dentry))
369 		return NULL;
370 
371 	inode = d_backing_inode(dentry);
372 	rcu_read_lock();
373 	id.key.object = rcu_dereference(landlock_inode(inode)->object);
374 	rule = landlock_find_rule(domain, id);
375 	rcu_read_unlock();
376 	return rule;
377 }
378 
379 /*
380  * Allows access to pseudo filesystems that will never be mountable (e.g.
381  * sockfs, pipefs), but can still be reachable through
382  * /proc/<pid>/fd/<file-descriptor>
383  */
384 static bool is_nouser_or_private(const struct dentry *dentry)
385 {
386 	return (dentry->d_sb->s_flags & SB_NOUSER) ||
387 	       (d_is_positive(dentry) &&
388 		unlikely(IS_PRIVATE(d_backing_inode(dentry))));
389 }
390 
391 static access_mask_t
392 get_handled_fs_accesses(const struct landlock_ruleset *const domain)
393 {
394 	/* Handles all initially denied by default access rights. */
395 	return landlock_union_access_masks(domain).fs |
396 	       LANDLOCK_ACCESS_FS_INITIALLY_DENIED;
397 }
398 
399 static const struct access_masks any_fs = {
400 	.fs = ~0,
401 };
402 
403 static const struct landlock_ruleset *get_current_fs_domain(void)
404 {
405 	return landlock_get_applicable_domain(landlock_get_current_domain(),
406 					      any_fs);
407 }
408 
409 /*
410  * Check that a destination file hierarchy has more restrictions than a source
411  * file hierarchy.  This is only used for link and rename actions.
412  *
413  * @layer_masks_child2: Optional child masks.
414  */
415 static bool no_more_access(
416 	const layer_mask_t (*const layer_masks_parent1)[LANDLOCK_NUM_ACCESS_FS],
417 	const layer_mask_t (*const layer_masks_child1)[LANDLOCK_NUM_ACCESS_FS],
418 	const bool child1_is_directory,
419 	const layer_mask_t (*const layer_masks_parent2)[LANDLOCK_NUM_ACCESS_FS],
420 	const layer_mask_t (*const layer_masks_child2)[LANDLOCK_NUM_ACCESS_FS],
421 	const bool child2_is_directory)
422 {
423 	unsigned long access_bit;
424 
425 	for (access_bit = 0; access_bit < ARRAY_SIZE(*layer_masks_parent2);
426 	     access_bit++) {
427 		/* Ignores accesses that only make sense for directories. */
428 		const bool is_file_access =
429 			!!(BIT_ULL(access_bit) & ACCESS_FILE);
430 
431 		if (child1_is_directory || is_file_access) {
432 			/*
433 			 * Checks if the destination restrictions are a
434 			 * superset of the source ones (i.e. inherited access
435 			 * rights without child exceptions):
436 			 * restrictions(parent2) >= restrictions(child1)
437 			 */
438 			if ((((*layer_masks_parent1)[access_bit] &
439 			      (*layer_masks_child1)[access_bit]) |
440 			     (*layer_masks_parent2)[access_bit]) !=
441 			    (*layer_masks_parent2)[access_bit])
442 				return false;
443 		}
444 
445 		if (!layer_masks_child2)
446 			continue;
447 		if (child2_is_directory || is_file_access) {
448 			/*
449 			 * Checks inverted restrictions for RENAME_EXCHANGE:
450 			 * restrictions(parent1) >= restrictions(child2)
451 			 */
452 			if ((((*layer_masks_parent2)[access_bit] &
453 			      (*layer_masks_child2)[access_bit]) |
454 			     (*layer_masks_parent1)[access_bit]) !=
455 			    (*layer_masks_parent1)[access_bit])
456 				return false;
457 		}
458 	}
459 	return true;
460 }
461 
462 #define NMA_TRUE(...) KUNIT_EXPECT_TRUE(test, no_more_access(__VA_ARGS__))
463 #define NMA_FALSE(...) KUNIT_EXPECT_FALSE(test, no_more_access(__VA_ARGS__))
464 
465 #ifdef CONFIG_SECURITY_LANDLOCK_KUNIT_TEST
466 
467 static void test_no_more_access(struct kunit *const test)
468 {
469 	const layer_mask_t rx0[LANDLOCK_NUM_ACCESS_FS] = {
470 		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(0),
471 		[BIT_INDEX(LANDLOCK_ACCESS_FS_READ_FILE)] = BIT_ULL(0),
472 	};
473 	const layer_mask_t mx0[LANDLOCK_NUM_ACCESS_FS] = {
474 		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(0),
475 		[BIT_INDEX(LANDLOCK_ACCESS_FS_MAKE_REG)] = BIT_ULL(0),
476 	};
477 	const layer_mask_t x0[LANDLOCK_NUM_ACCESS_FS] = {
478 		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(0),
479 	};
480 	const layer_mask_t x1[LANDLOCK_NUM_ACCESS_FS] = {
481 		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(1),
482 	};
483 	const layer_mask_t x01[LANDLOCK_NUM_ACCESS_FS] = {
484 		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(0) |
485 							  BIT_ULL(1),
486 	};
487 	const layer_mask_t allows_all[LANDLOCK_NUM_ACCESS_FS] = {};
488 
489 	/* Checks without restriction. */
490 	NMA_TRUE(&x0, &allows_all, false, &allows_all, NULL, false);
491 	NMA_TRUE(&allows_all, &x0, false, &allows_all, NULL, false);
492 	NMA_FALSE(&x0, &x0, false, &allows_all, NULL, false);
493 
494 	/*
495 	 * Checks that we can only refer a file if no more access could be
496 	 * inherited.
497 	 */
498 	NMA_TRUE(&x0, &x0, false, &rx0, NULL, false);
499 	NMA_TRUE(&rx0, &rx0, false, &rx0, NULL, false);
500 	NMA_FALSE(&rx0, &rx0, false, &x0, NULL, false);
501 	NMA_FALSE(&rx0, &rx0, false, &x1, NULL, false);
502 
503 	/* Checks allowed referring with different nested domains. */
504 	NMA_TRUE(&x0, &x1, false, &x0, NULL, false);
505 	NMA_TRUE(&x1, &x0, false, &x0, NULL, false);
506 	NMA_TRUE(&x0, &x01, false, &x0, NULL, false);
507 	NMA_TRUE(&x0, &x01, false, &rx0, NULL, false);
508 	NMA_TRUE(&x01, &x0, false, &x0, NULL, false);
509 	NMA_TRUE(&x01, &x0, false, &rx0, NULL, false);
510 	NMA_FALSE(&x01, &x01, false, &x0, NULL, false);
511 
512 	/* Checks that file access rights are also enforced for a directory. */
513 	NMA_FALSE(&rx0, &rx0, true, &x0, NULL, false);
514 
515 	/* Checks that directory access rights don't impact file referring... */
516 	NMA_TRUE(&mx0, &mx0, false, &x0, NULL, false);
517 	/* ...but only directory referring. */
518 	NMA_FALSE(&mx0, &mx0, true, &x0, NULL, false);
519 
520 	/* Checks directory exchange. */
521 	NMA_TRUE(&mx0, &mx0, true, &mx0, &mx0, true);
522 	NMA_TRUE(&mx0, &mx0, true, &mx0, &x0, true);
523 	NMA_FALSE(&mx0, &mx0, true, &x0, &mx0, true);
524 	NMA_FALSE(&mx0, &mx0, true, &x0, &x0, true);
525 	NMA_FALSE(&mx0, &mx0, true, &x1, &x1, true);
526 
527 	/* Checks file exchange with directory access rights... */
528 	NMA_TRUE(&mx0, &mx0, false, &mx0, &mx0, false);
529 	NMA_TRUE(&mx0, &mx0, false, &mx0, &x0, false);
530 	NMA_TRUE(&mx0, &mx0, false, &x0, &mx0, false);
531 	NMA_TRUE(&mx0, &mx0, false, &x0, &x0, false);
532 	/* ...and with file access rights. */
533 	NMA_TRUE(&rx0, &rx0, false, &rx0, &rx0, false);
534 	NMA_TRUE(&rx0, &rx0, false, &rx0, &x0, false);
535 	NMA_FALSE(&rx0, &rx0, false, &x0, &rx0, false);
536 	NMA_FALSE(&rx0, &rx0, false, &x0, &x0, false);
537 	NMA_FALSE(&rx0, &rx0, false, &x1, &x1, false);
538 
539 	/*
540 	 * Allowing the following requests should not be a security risk
541 	 * because domain 0 denies execute access, and domain 1 is always
542 	 * nested with domain 0.  However, adding an exception for this case
543 	 * would mean to check all nested domains to make sure none can get
544 	 * more privileges (e.g. processes only sandboxed by domain 0).
545 	 * Moreover, this behavior (i.e. composition of N domains) could then
546 	 * be inconsistent compared to domain 1's ruleset alone (e.g. it might
547 	 * be denied to link/rename with domain 1's ruleset, whereas it would
548 	 * be allowed if nested on top of domain 0).  Another drawback would be
549 	 * to create a cover channel that could enable sandboxed processes to
550 	 * infer most of the filesystem restrictions from their domain.  To
551 	 * make it simple, efficient, safe, and more consistent, this case is
552 	 * always denied.
553 	 */
554 	NMA_FALSE(&x1, &x1, false, &x0, NULL, false);
555 	NMA_FALSE(&x1, &x1, false, &rx0, NULL, false);
556 	NMA_FALSE(&x1, &x1, true, &x0, NULL, false);
557 	NMA_FALSE(&x1, &x1, true, &rx0, NULL, false);
558 
559 	/* Checks the same case of exclusive domains with a file... */
560 	NMA_TRUE(&x1, &x1, false, &x01, NULL, false);
561 	NMA_FALSE(&x1, &x1, false, &x01, &x0, false);
562 	NMA_FALSE(&x1, &x1, false, &x01, &x01, false);
563 	NMA_FALSE(&x1, &x1, false, &x0, &x0, false);
564 	/* ...and with a directory. */
565 	NMA_FALSE(&x1, &x1, false, &x0, &x0, true);
566 	NMA_FALSE(&x1, &x1, true, &x0, &x0, false);
567 	NMA_FALSE(&x1, &x1, true, &x0, &x0, true);
568 }
569 
570 #endif /* CONFIG_SECURITY_LANDLOCK_KUNIT_TEST */
571 
572 #undef NMA_TRUE
573 #undef NMA_FALSE
574 
575 /*
576  * Removes @layer_masks accesses that are not requested.
577  *
578  * Returns true if the request is allowed, false otherwise.
579  */
580 static bool
581 scope_to_request(const access_mask_t access_request,
582 		 layer_mask_t (*const layer_masks)[LANDLOCK_NUM_ACCESS_FS])
583 {
584 	const unsigned long access_req = access_request;
585 	unsigned long access_bit;
586 
587 	if (WARN_ON_ONCE(!layer_masks))
588 		return true;
589 
590 	for_each_clear_bit(access_bit, &access_req, ARRAY_SIZE(*layer_masks))
591 		(*layer_masks)[access_bit] = 0;
592 	return !memchr_inv(layer_masks, 0, sizeof(*layer_masks));
593 }
594 
595 #ifdef CONFIG_SECURITY_LANDLOCK_KUNIT_TEST
596 
597 static void test_scope_to_request_with_exec_none(struct kunit *const test)
598 {
599 	/* Allows everything. */
600 	layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {};
601 
602 	/* Checks and scopes with execute. */
603 	KUNIT_EXPECT_TRUE(test, scope_to_request(LANDLOCK_ACCESS_FS_EXECUTE,
604 						 &layer_masks));
605 	KUNIT_EXPECT_EQ(test, 0,
606 			layer_masks[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)]);
607 	KUNIT_EXPECT_EQ(test, 0,
608 			layer_masks[BIT_INDEX(LANDLOCK_ACCESS_FS_WRITE_FILE)]);
609 }
610 
611 static void test_scope_to_request_with_exec_some(struct kunit *const test)
612 {
613 	/* Denies execute and write. */
614 	layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {
615 		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(0),
616 		[BIT_INDEX(LANDLOCK_ACCESS_FS_WRITE_FILE)] = BIT_ULL(1),
617 	};
618 
619 	/* Checks and scopes with execute. */
620 	KUNIT_EXPECT_FALSE(test, scope_to_request(LANDLOCK_ACCESS_FS_EXECUTE,
621 						  &layer_masks));
622 	KUNIT_EXPECT_EQ(test, BIT_ULL(0),
623 			layer_masks[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)]);
624 	KUNIT_EXPECT_EQ(test, 0,
625 			layer_masks[BIT_INDEX(LANDLOCK_ACCESS_FS_WRITE_FILE)]);
626 }
627 
628 static void test_scope_to_request_without_access(struct kunit *const test)
629 {
630 	/* Denies execute and write. */
631 	layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {
632 		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(0),
633 		[BIT_INDEX(LANDLOCK_ACCESS_FS_WRITE_FILE)] = BIT_ULL(1),
634 	};
635 
636 	/* Checks and scopes without access request. */
637 	KUNIT_EXPECT_TRUE(test, scope_to_request(0, &layer_masks));
638 	KUNIT_EXPECT_EQ(test, 0,
639 			layer_masks[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)]);
640 	KUNIT_EXPECT_EQ(test, 0,
641 			layer_masks[BIT_INDEX(LANDLOCK_ACCESS_FS_WRITE_FILE)]);
642 }
643 
644 #endif /* CONFIG_SECURITY_LANDLOCK_KUNIT_TEST */
645 
646 /*
647  * Returns true if there is at least one access right different than
648  * LANDLOCK_ACCESS_FS_REFER.
649  */
650 static bool
651 is_eacces(const layer_mask_t (*const layer_masks)[LANDLOCK_NUM_ACCESS_FS],
652 	  const access_mask_t access_request)
653 {
654 	unsigned long access_bit;
655 	/* LANDLOCK_ACCESS_FS_REFER alone must return -EXDEV. */
656 	const unsigned long access_check = access_request &
657 					   ~LANDLOCK_ACCESS_FS_REFER;
658 
659 	if (!layer_masks)
660 		return false;
661 
662 	for_each_set_bit(access_bit, &access_check, ARRAY_SIZE(*layer_masks)) {
663 		if ((*layer_masks)[access_bit])
664 			return true;
665 	}
666 	return false;
667 }
668 
669 #define IE_TRUE(...) KUNIT_EXPECT_TRUE(test, is_eacces(__VA_ARGS__))
670 #define IE_FALSE(...) KUNIT_EXPECT_FALSE(test, is_eacces(__VA_ARGS__))
671 
672 #ifdef CONFIG_SECURITY_LANDLOCK_KUNIT_TEST
673 
674 static void test_is_eacces_with_none(struct kunit *const test)
675 {
676 	const layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {};
677 
678 	IE_FALSE(&layer_masks, 0);
679 	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_REFER);
680 	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_EXECUTE);
681 	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_WRITE_FILE);
682 }
683 
684 static void test_is_eacces_with_refer(struct kunit *const test)
685 {
686 	const layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {
687 		[BIT_INDEX(LANDLOCK_ACCESS_FS_REFER)] = BIT_ULL(0),
688 	};
689 
690 	IE_FALSE(&layer_masks, 0);
691 	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_REFER);
692 	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_EXECUTE);
693 	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_WRITE_FILE);
694 }
695 
696 static void test_is_eacces_with_write(struct kunit *const test)
697 {
698 	const layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {
699 		[BIT_INDEX(LANDLOCK_ACCESS_FS_WRITE_FILE)] = BIT_ULL(0),
700 	};
701 
702 	IE_FALSE(&layer_masks, 0);
703 	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_REFER);
704 	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_EXECUTE);
705 
706 	IE_TRUE(&layer_masks, LANDLOCK_ACCESS_FS_WRITE_FILE);
707 }
708 
709 #endif /* CONFIG_SECURITY_LANDLOCK_KUNIT_TEST */
710 
711 #undef IE_TRUE
712 #undef IE_FALSE
713 
714 /**
715  * is_access_to_paths_allowed - Check accesses for requests with a common path
716  *
717  * @domain: Domain to check against.
718  * @path: File hierarchy to walk through.
719  * @access_request_parent1: Accesses to check, once @layer_masks_parent1 is
720  *     equal to @layer_masks_parent2 (if any).  This is tied to the unique
721  *     requested path for most actions, or the source in case of a refer action
722  *     (i.e. rename or link), or the source and destination in case of
723  *     RENAME_EXCHANGE.
724  * @layer_masks_parent1: Pointer to a matrix of layer masks per access
725  *     masks, identifying the layers that forbid a specific access.  Bits from
726  *     this matrix can be unset according to the @path walk.  An empty matrix
727  *     means that @domain allows all possible Landlock accesses (i.e. not only
728  *     those identified by @access_request_parent1).  This matrix can
729  *     initially refer to domain layer masks and, when the accesses for the
730  *     destination and source are the same, to requested layer masks.
731  * @dentry_child1: Dentry to the initial child of the parent1 path.  This
732  *     pointer must be NULL for non-refer actions (i.e. not link nor rename).
733  * @access_request_parent2: Similar to @access_request_parent1 but for a
734  *     request involving a source and a destination.  This refers to the
735  *     destination, except in case of RENAME_EXCHANGE where it also refers to
736  *     the source.  Must be set to 0 when using a simple path request.
737  * @layer_masks_parent2: Similar to @layer_masks_parent1 but for a refer
738  *     action.  This must be NULL otherwise.
739  * @dentry_child2: Dentry to the initial child of the parent2 path.  This
740  *     pointer is only set for RENAME_EXCHANGE actions and must be NULL
741  *     otherwise.
742  *
743  * This helper first checks that the destination has a superset of restrictions
744  * compared to the source (if any) for a common path.  Because of
745  * RENAME_EXCHANGE actions, source and destinations may be swapped.  It then
746  * checks that the collected accesses and the remaining ones are enough to
747  * allow the request.
748  *
749  * Returns:
750  * - true if the access request is granted;
751  * - false otherwise.
752  */
753 static bool is_access_to_paths_allowed(
754 	const struct landlock_ruleset *const domain,
755 	const struct path *const path,
756 	const access_mask_t access_request_parent1,
757 	layer_mask_t (*const layer_masks_parent1)[LANDLOCK_NUM_ACCESS_FS],
758 	const struct dentry *const dentry_child1,
759 	const access_mask_t access_request_parent2,
760 	layer_mask_t (*const layer_masks_parent2)[LANDLOCK_NUM_ACCESS_FS],
761 	const struct dentry *const dentry_child2)
762 {
763 	bool allowed_parent1 = false, allowed_parent2 = false, is_dom_check,
764 	     child1_is_directory = true, child2_is_directory = true;
765 	struct path walker_path;
766 	access_mask_t access_masked_parent1, access_masked_parent2;
767 	layer_mask_t _layer_masks_child1[LANDLOCK_NUM_ACCESS_FS],
768 		_layer_masks_child2[LANDLOCK_NUM_ACCESS_FS];
769 	layer_mask_t(*layer_masks_child1)[LANDLOCK_NUM_ACCESS_FS] = NULL,
770 	(*layer_masks_child2)[LANDLOCK_NUM_ACCESS_FS] = NULL;
771 
772 	if (!access_request_parent1 && !access_request_parent2)
773 		return true;
774 	if (WARN_ON_ONCE(!domain || !path))
775 		return true;
776 	if (is_nouser_or_private(path->dentry))
777 		return true;
778 	if (WARN_ON_ONCE(domain->num_layers < 1 || !layer_masks_parent1))
779 		return false;
780 
781 	if (unlikely(layer_masks_parent2)) {
782 		if (WARN_ON_ONCE(!dentry_child1))
783 			return false;
784 		/*
785 		 * For a double request, first check for potential privilege
786 		 * escalation by looking at domain handled accesses (which are
787 		 * a superset of the meaningful requested accesses).
788 		 */
789 		access_masked_parent1 = access_masked_parent2 =
790 			get_handled_fs_accesses(domain);
791 		is_dom_check = true;
792 	} else {
793 		if (WARN_ON_ONCE(dentry_child1 || dentry_child2))
794 			return false;
795 		/* For a simple request, only check for requested accesses. */
796 		access_masked_parent1 = access_request_parent1;
797 		access_masked_parent2 = access_request_parent2;
798 		is_dom_check = false;
799 	}
800 
801 	if (unlikely(dentry_child1)) {
802 		landlock_unmask_layers(
803 			find_rule(domain, dentry_child1),
804 			landlock_init_layer_masks(
805 				domain, LANDLOCK_MASK_ACCESS_FS,
806 				&_layer_masks_child1, LANDLOCK_KEY_INODE),
807 			&_layer_masks_child1, ARRAY_SIZE(_layer_masks_child1));
808 		layer_masks_child1 = &_layer_masks_child1;
809 		child1_is_directory = d_is_dir(dentry_child1);
810 	}
811 	if (unlikely(dentry_child2)) {
812 		landlock_unmask_layers(
813 			find_rule(domain, dentry_child2),
814 			landlock_init_layer_masks(
815 				domain, LANDLOCK_MASK_ACCESS_FS,
816 				&_layer_masks_child2, LANDLOCK_KEY_INODE),
817 			&_layer_masks_child2, ARRAY_SIZE(_layer_masks_child2));
818 		layer_masks_child2 = &_layer_masks_child2;
819 		child2_is_directory = d_is_dir(dentry_child2);
820 	}
821 
822 	walker_path = *path;
823 	path_get(&walker_path);
824 	/*
825 	 * We need to walk through all the hierarchy to not miss any relevant
826 	 * restriction.
827 	 */
828 	while (true) {
829 		struct dentry *parent_dentry;
830 		const struct landlock_rule *rule;
831 
832 		/*
833 		 * If at least all accesses allowed on the destination are
834 		 * already allowed on the source, respectively if there is at
835 		 * least as much as restrictions on the destination than on the
836 		 * source, then we can safely refer files from the source to
837 		 * the destination without risking a privilege escalation.
838 		 * This also applies in the case of RENAME_EXCHANGE, which
839 		 * implies checks on both direction.  This is crucial for
840 		 * standalone multilayered security policies.  Furthermore,
841 		 * this helps avoid policy writers to shoot themselves in the
842 		 * foot.
843 		 */
844 		if (unlikely(is_dom_check &&
845 			     no_more_access(
846 				     layer_masks_parent1, layer_masks_child1,
847 				     child1_is_directory, layer_masks_parent2,
848 				     layer_masks_child2,
849 				     child2_is_directory))) {
850 			allowed_parent1 = scope_to_request(
851 				access_request_parent1, layer_masks_parent1);
852 			allowed_parent2 = scope_to_request(
853 				access_request_parent2, layer_masks_parent2);
854 
855 			/* Stops when all accesses are granted. */
856 			if (allowed_parent1 && allowed_parent2)
857 				break;
858 
859 			/*
860 			 * Now, downgrades the remaining checks from domain
861 			 * handled accesses to requested accesses.
862 			 */
863 			is_dom_check = false;
864 			access_masked_parent1 = access_request_parent1;
865 			access_masked_parent2 = access_request_parent2;
866 		}
867 
868 		rule = find_rule(domain, walker_path.dentry);
869 		allowed_parent1 = landlock_unmask_layers(
870 			rule, access_masked_parent1, layer_masks_parent1,
871 			ARRAY_SIZE(*layer_masks_parent1));
872 		allowed_parent2 = landlock_unmask_layers(
873 			rule, access_masked_parent2, layer_masks_parent2,
874 			ARRAY_SIZE(*layer_masks_parent2));
875 
876 		/* Stops when a rule from each layer grants access. */
877 		if (allowed_parent1 && allowed_parent2)
878 			break;
879 jump_up:
880 		if (walker_path.dentry == walker_path.mnt->mnt_root) {
881 			if (follow_up(&walker_path)) {
882 				/* Ignores hidden mount points. */
883 				goto jump_up;
884 			} else {
885 				/*
886 				 * Stops at the real root.  Denies access
887 				 * because not all layers have granted access.
888 				 */
889 				break;
890 			}
891 		}
892 		if (unlikely(IS_ROOT(walker_path.dentry))) {
893 			/*
894 			 * Stops at disconnected root directories.  Only allows
895 			 * access to internal filesystems (e.g. nsfs, which is
896 			 * reachable through /proc/<pid>/ns/<namespace>).
897 			 */
898 			allowed_parent1 = allowed_parent2 =
899 				!!(walker_path.mnt->mnt_flags & MNT_INTERNAL);
900 			break;
901 		}
902 		parent_dentry = dget_parent(walker_path.dentry);
903 		dput(walker_path.dentry);
904 		walker_path.dentry = parent_dentry;
905 	}
906 	path_put(&walker_path);
907 
908 	return allowed_parent1 && allowed_parent2;
909 }
910 
911 static int check_access_path(const struct landlock_ruleset *const domain,
912 			     const struct path *const path,
913 			     access_mask_t access_request)
914 {
915 	layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {};
916 
917 	access_request = landlock_init_layer_masks(
918 		domain, access_request, &layer_masks, LANDLOCK_KEY_INODE);
919 	if (is_access_to_paths_allowed(domain, path, access_request,
920 				       &layer_masks, NULL, 0, NULL, NULL))
921 		return 0;
922 	return -EACCES;
923 }
924 
925 static int current_check_access_path(const struct path *const path,
926 				     const access_mask_t access_request)
927 {
928 	const struct landlock_ruleset *const dom = get_current_fs_domain();
929 
930 	if (!dom)
931 		return 0;
932 	return check_access_path(dom, path, access_request);
933 }
934 
935 static access_mask_t get_mode_access(const umode_t mode)
936 {
937 	switch (mode & S_IFMT) {
938 	case S_IFLNK:
939 		return LANDLOCK_ACCESS_FS_MAKE_SYM;
940 	case 0:
941 		/* A zero mode translates to S_IFREG. */
942 	case S_IFREG:
943 		return LANDLOCK_ACCESS_FS_MAKE_REG;
944 	case S_IFDIR:
945 		return LANDLOCK_ACCESS_FS_MAKE_DIR;
946 	case S_IFCHR:
947 		return LANDLOCK_ACCESS_FS_MAKE_CHAR;
948 	case S_IFBLK:
949 		return LANDLOCK_ACCESS_FS_MAKE_BLOCK;
950 	case S_IFIFO:
951 		return LANDLOCK_ACCESS_FS_MAKE_FIFO;
952 	case S_IFSOCK:
953 		return LANDLOCK_ACCESS_FS_MAKE_SOCK;
954 	default:
955 		WARN_ON_ONCE(1);
956 		return 0;
957 	}
958 }
959 
960 static access_mask_t maybe_remove(const struct dentry *const dentry)
961 {
962 	if (d_is_negative(dentry))
963 		return 0;
964 	return d_is_dir(dentry) ? LANDLOCK_ACCESS_FS_REMOVE_DIR :
965 				  LANDLOCK_ACCESS_FS_REMOVE_FILE;
966 }
967 
968 /**
969  * collect_domain_accesses - Walk through a file path and collect accesses
970  *
971  * @domain: Domain to check against.
972  * @mnt_root: Last directory to check.
973  * @dir: Directory to start the walk from.
974  * @layer_masks_dom: Where to store the collected accesses.
975  *
976  * This helper is useful to begin a path walk from the @dir directory to a
977  * @mnt_root directory used as a mount point.  This mount point is the common
978  * ancestor between the source and the destination of a renamed and linked
979  * file.  While walking from @dir to @mnt_root, we record all the domain's
980  * allowed accesses in @layer_masks_dom.
981  *
982  * This is similar to is_access_to_paths_allowed() but much simpler because it
983  * only handles walking on the same mount point and only checks one set of
984  * accesses.
985  *
986  * Returns:
987  * - true if all the domain access rights are allowed for @dir;
988  * - false if the walk reached @mnt_root.
989  */
990 static bool collect_domain_accesses(
991 	const struct landlock_ruleset *const domain,
992 	const struct dentry *const mnt_root, struct dentry *dir,
993 	layer_mask_t (*const layer_masks_dom)[LANDLOCK_NUM_ACCESS_FS])
994 {
995 	unsigned long access_dom;
996 	bool ret = false;
997 
998 	if (WARN_ON_ONCE(!domain || !mnt_root || !dir || !layer_masks_dom))
999 		return true;
1000 	if (is_nouser_or_private(dir))
1001 		return true;
1002 
1003 	access_dom = landlock_init_layer_masks(domain, LANDLOCK_MASK_ACCESS_FS,
1004 					       layer_masks_dom,
1005 					       LANDLOCK_KEY_INODE);
1006 
1007 	dget(dir);
1008 	while (true) {
1009 		struct dentry *parent_dentry;
1010 
1011 		/* Gets all layers allowing all domain accesses. */
1012 		if (landlock_unmask_layers(find_rule(domain, dir), access_dom,
1013 					   layer_masks_dom,
1014 					   ARRAY_SIZE(*layer_masks_dom))) {
1015 			/*
1016 			 * Stops when all handled accesses are allowed by at
1017 			 * least one rule in each layer.
1018 			 */
1019 			ret = true;
1020 			break;
1021 		}
1022 
1023 		/* We should not reach a root other than @mnt_root. */
1024 		if (dir == mnt_root || WARN_ON_ONCE(IS_ROOT(dir)))
1025 			break;
1026 
1027 		parent_dentry = dget_parent(dir);
1028 		dput(dir);
1029 		dir = parent_dentry;
1030 	}
1031 	dput(dir);
1032 	return ret;
1033 }
1034 
1035 /**
1036  * current_check_refer_path - Check if a rename or link action is allowed
1037  *
1038  * @old_dentry: File or directory requested to be moved or linked.
1039  * @new_dir: Destination parent directory.
1040  * @new_dentry: Destination file or directory.
1041  * @removable: Sets to true if it is a rename operation.
1042  * @exchange: Sets to true if it is a rename operation with RENAME_EXCHANGE.
1043  *
1044  * Because of its unprivileged constraints, Landlock relies on file hierarchies
1045  * (and not only inodes) to tie access rights to files.  Being able to link or
1046  * rename a file hierarchy brings some challenges.  Indeed, moving or linking a
1047  * file (i.e. creating a new reference to an inode) can have an impact on the
1048  * actions allowed for a set of files if it would change its parent directory
1049  * (i.e. reparenting).
1050  *
1051  * To avoid trivial access right bypasses, Landlock first checks if the file or
1052  * directory requested to be moved would gain new access rights inherited from
1053  * its new hierarchy.  Before returning any error, Landlock then checks that
1054  * the parent source hierarchy and the destination hierarchy would allow the
1055  * link or rename action.  If it is not the case, an error with EACCES is
1056  * returned to inform user space that there is no way to remove or create the
1057  * requested source file type.  If it should be allowed but the new inherited
1058  * access rights would be greater than the source access rights, then the
1059  * kernel returns an error with EXDEV.  Prioritizing EACCES over EXDEV enables
1060  * user space to abort the whole operation if there is no way to do it, or to
1061  * manually copy the source to the destination if this remains allowed, e.g.
1062  * because file creation is allowed on the destination directory but not direct
1063  * linking.
1064  *
1065  * To achieve this goal, the kernel needs to compare two file hierarchies: the
1066  * one identifying the source file or directory (including itself), and the
1067  * destination one.  This can be seen as a multilayer partial ordering problem.
1068  * The kernel walks through these paths and collects in a matrix the access
1069  * rights that are denied per layer.  These matrices are then compared to see
1070  * if the destination one has more (or the same) restrictions as the source
1071  * one.  If this is the case, the requested action will not return EXDEV, which
1072  * doesn't mean the action is allowed.  The parent hierarchy of the source
1073  * (i.e. parent directory), and the destination hierarchy must also be checked
1074  * to verify that they explicitly allow such action (i.e.  referencing,
1075  * creation and potentially removal rights).  The kernel implementation is then
1076  * required to rely on potentially four matrices of access rights: one for the
1077  * source file or directory (i.e. the child), a potentially other one for the
1078  * other source/destination (in case of RENAME_EXCHANGE), one for the source
1079  * parent hierarchy and a last one for the destination hierarchy.  These
1080  * ephemeral matrices take some space on the stack, which limits the number of
1081  * layers to a deemed reasonable number: 16.
1082  *
1083  * Returns:
1084  * - 0 if access is allowed;
1085  * - -EXDEV if @old_dentry would inherit new access rights from @new_dir;
1086  * - -EACCES if file removal or creation is denied.
1087  */
1088 static int current_check_refer_path(struct dentry *const old_dentry,
1089 				    const struct path *const new_dir,
1090 				    struct dentry *const new_dentry,
1091 				    const bool removable, const bool exchange)
1092 {
1093 	const struct landlock_ruleset *const dom = get_current_fs_domain();
1094 	bool allow_parent1, allow_parent2;
1095 	access_mask_t access_request_parent1, access_request_parent2;
1096 	struct path mnt_dir;
1097 	struct dentry *old_parent;
1098 	layer_mask_t layer_masks_parent1[LANDLOCK_NUM_ACCESS_FS] = {},
1099 		     layer_masks_parent2[LANDLOCK_NUM_ACCESS_FS] = {};
1100 
1101 	if (!dom)
1102 		return 0;
1103 	if (WARN_ON_ONCE(dom->num_layers < 1))
1104 		return -EACCES;
1105 	if (unlikely(d_is_negative(old_dentry)))
1106 		return -ENOENT;
1107 	if (exchange) {
1108 		if (unlikely(d_is_negative(new_dentry)))
1109 			return -ENOENT;
1110 		access_request_parent1 =
1111 			get_mode_access(d_backing_inode(new_dentry)->i_mode);
1112 	} else {
1113 		access_request_parent1 = 0;
1114 	}
1115 	access_request_parent2 =
1116 		get_mode_access(d_backing_inode(old_dentry)->i_mode);
1117 	if (removable) {
1118 		access_request_parent1 |= maybe_remove(old_dentry);
1119 		access_request_parent2 |= maybe_remove(new_dentry);
1120 	}
1121 
1122 	/* The mount points are the same for old and new paths, cf. EXDEV. */
1123 	if (old_dentry->d_parent == new_dir->dentry) {
1124 		/*
1125 		 * The LANDLOCK_ACCESS_FS_REFER access right is not required
1126 		 * for same-directory referer (i.e. no reparenting).
1127 		 */
1128 		access_request_parent1 = landlock_init_layer_masks(
1129 			dom, access_request_parent1 | access_request_parent2,
1130 			&layer_masks_parent1, LANDLOCK_KEY_INODE);
1131 		if (is_access_to_paths_allowed(
1132 			    dom, new_dir, access_request_parent1,
1133 			    &layer_masks_parent1, NULL, 0, NULL, NULL))
1134 			return 0;
1135 		return -EACCES;
1136 	}
1137 
1138 	access_request_parent1 |= LANDLOCK_ACCESS_FS_REFER;
1139 	access_request_parent2 |= LANDLOCK_ACCESS_FS_REFER;
1140 
1141 	/* Saves the common mount point. */
1142 	mnt_dir.mnt = new_dir->mnt;
1143 	mnt_dir.dentry = new_dir->mnt->mnt_root;
1144 
1145 	/*
1146 	 * old_dentry may be the root of the common mount point and
1147 	 * !IS_ROOT(old_dentry) at the same time (e.g. with open_tree() and
1148 	 * OPEN_TREE_CLONE).  We do not need to call dget(old_parent) because
1149 	 * we keep a reference to old_dentry.
1150 	 */
1151 	old_parent = (old_dentry == mnt_dir.dentry) ? old_dentry :
1152 						      old_dentry->d_parent;
1153 
1154 	/* new_dir->dentry is equal to new_dentry->d_parent */
1155 	allow_parent1 = collect_domain_accesses(dom, mnt_dir.dentry, old_parent,
1156 						&layer_masks_parent1);
1157 	allow_parent2 = collect_domain_accesses(
1158 		dom, mnt_dir.dentry, new_dir->dentry, &layer_masks_parent2);
1159 
1160 	if (allow_parent1 && allow_parent2)
1161 		return 0;
1162 
1163 	/*
1164 	 * To be able to compare source and destination domain access rights,
1165 	 * take into account the @old_dentry access rights aggregated with its
1166 	 * parent access rights.  This will be useful to compare with the
1167 	 * destination parent access rights.
1168 	 */
1169 	if (is_access_to_paths_allowed(
1170 		    dom, &mnt_dir, access_request_parent1, &layer_masks_parent1,
1171 		    old_dentry, access_request_parent2, &layer_masks_parent2,
1172 		    exchange ? new_dentry : NULL))
1173 		return 0;
1174 
1175 	/*
1176 	 * This prioritizes EACCES over EXDEV for all actions, including
1177 	 * renames with RENAME_EXCHANGE.
1178 	 */
1179 	if (likely(is_eacces(&layer_masks_parent1, access_request_parent1) ||
1180 		   is_eacces(&layer_masks_parent2, access_request_parent2)))
1181 		return -EACCES;
1182 
1183 	/*
1184 	 * Gracefully forbids reparenting if the destination directory
1185 	 * hierarchy is not a superset of restrictions of the source directory
1186 	 * hierarchy, or if LANDLOCK_ACCESS_FS_REFER is not allowed by the
1187 	 * source or the destination.
1188 	 */
1189 	return -EXDEV;
1190 }
1191 
1192 /* Inode hooks */
1193 
1194 static void hook_inode_free_security_rcu(void *inode_security)
1195 {
1196 	struct landlock_inode_security *inode_sec;
1197 
1198 	/*
1199 	 * All inodes must already have been untied from their object by
1200 	 * release_inode() or hook_sb_delete().
1201 	 */
1202 	inode_sec = inode_security + landlock_blob_sizes.lbs_inode;
1203 	WARN_ON_ONCE(inode_sec->object);
1204 }
1205 
1206 /* Super-block hooks */
1207 
1208 /*
1209  * Release the inodes used in a security policy.
1210  *
1211  * Cf. fsnotify_unmount_inodes() and invalidate_inodes()
1212  */
1213 static void hook_sb_delete(struct super_block *const sb)
1214 {
1215 	struct inode *inode, *prev_inode = NULL;
1216 
1217 	if (!landlock_initialized)
1218 		return;
1219 
1220 	spin_lock(&sb->s_inode_list_lock);
1221 	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1222 		struct landlock_object *object;
1223 
1224 		/* Only handles referenced inodes. */
1225 		if (!atomic_read(&inode->i_count))
1226 			continue;
1227 
1228 		/*
1229 		 * Protects against concurrent modification of inode (e.g.
1230 		 * from get_inode_object()).
1231 		 */
1232 		spin_lock(&inode->i_lock);
1233 		/*
1234 		 * Checks I_FREEING and I_WILL_FREE  to protect against a race
1235 		 * condition when release_inode() just called iput(), which
1236 		 * could lead to a NULL dereference of inode->security or a
1237 		 * second call to iput() for the same Landlock object.  Also
1238 		 * checks I_NEW because such inode cannot be tied to an object.
1239 		 */
1240 		if (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW)) {
1241 			spin_unlock(&inode->i_lock);
1242 			continue;
1243 		}
1244 
1245 		rcu_read_lock();
1246 		object = rcu_dereference(landlock_inode(inode)->object);
1247 		if (!object) {
1248 			rcu_read_unlock();
1249 			spin_unlock(&inode->i_lock);
1250 			continue;
1251 		}
1252 		/* Keeps a reference to this inode until the next loop walk. */
1253 		__iget(inode);
1254 		spin_unlock(&inode->i_lock);
1255 
1256 		/*
1257 		 * If there is no concurrent release_inode() ongoing, then we
1258 		 * are in charge of calling iput() on this inode, otherwise we
1259 		 * will just wait for it to finish.
1260 		 */
1261 		spin_lock(&object->lock);
1262 		if (object->underobj == inode) {
1263 			object->underobj = NULL;
1264 			spin_unlock(&object->lock);
1265 			rcu_read_unlock();
1266 
1267 			/*
1268 			 * Because object->underobj was not NULL,
1269 			 * release_inode() and get_inode_object() guarantee
1270 			 * that it is safe to reset
1271 			 * landlock_inode(inode)->object while it is not NULL.
1272 			 * It is therefore not necessary to lock inode->i_lock.
1273 			 */
1274 			rcu_assign_pointer(landlock_inode(inode)->object, NULL);
1275 			/*
1276 			 * At this point, we own the ihold() reference that was
1277 			 * originally set up by get_inode_object() and the
1278 			 * __iget() reference that we just set in this loop
1279 			 * walk.  Therefore the following call to iput() will
1280 			 * not sleep nor drop the inode because there is now at
1281 			 * least two references to it.
1282 			 */
1283 			iput(inode);
1284 		} else {
1285 			spin_unlock(&object->lock);
1286 			rcu_read_unlock();
1287 		}
1288 
1289 		if (prev_inode) {
1290 			/*
1291 			 * At this point, we still own the __iget() reference
1292 			 * that we just set in this loop walk.  Therefore we
1293 			 * can drop the list lock and know that the inode won't
1294 			 * disappear from under us until the next loop walk.
1295 			 */
1296 			spin_unlock(&sb->s_inode_list_lock);
1297 			/*
1298 			 * We can now actually put the inode reference from the
1299 			 * previous loop walk, which is not needed anymore.
1300 			 */
1301 			iput(prev_inode);
1302 			cond_resched();
1303 			spin_lock(&sb->s_inode_list_lock);
1304 		}
1305 		prev_inode = inode;
1306 	}
1307 	spin_unlock(&sb->s_inode_list_lock);
1308 
1309 	/* Puts the inode reference from the last loop walk, if any. */
1310 	if (prev_inode)
1311 		iput(prev_inode);
1312 	/* Waits for pending iput() in release_inode(). */
1313 	wait_var_event(&landlock_superblock(sb)->inode_refs,
1314 		       !atomic_long_read(&landlock_superblock(sb)->inode_refs));
1315 }
1316 
1317 /*
1318  * Because a Landlock security policy is defined according to the filesystem
1319  * topology (i.e. the mount namespace), changing it may grant access to files
1320  * not previously allowed.
1321  *
1322  * To make it simple, deny any filesystem topology modification by landlocked
1323  * processes.  Non-landlocked processes may still change the namespace of a
1324  * landlocked process, but this kind of threat must be handled by a system-wide
1325  * access-control security policy.
1326  *
1327  * This could be lifted in the future if Landlock can safely handle mount
1328  * namespace updates requested by a landlocked process.  Indeed, we could
1329  * update the current domain (which is currently read-only) by taking into
1330  * account the accesses of the source and the destination of a new mount point.
1331  * However, it would also require to make all the child domains dynamically
1332  * inherit these new constraints.  Anyway, for backward compatibility reasons,
1333  * a dedicated user space option would be required (e.g. as a ruleset flag).
1334  */
1335 static int hook_sb_mount(const char *const dev_name,
1336 			 const struct path *const path, const char *const type,
1337 			 const unsigned long flags, void *const data)
1338 {
1339 	if (!get_current_fs_domain())
1340 		return 0;
1341 	return -EPERM;
1342 }
1343 
1344 static int hook_move_mount(const struct path *const from_path,
1345 			   const struct path *const to_path)
1346 {
1347 	if (!get_current_fs_domain())
1348 		return 0;
1349 	return -EPERM;
1350 }
1351 
1352 /*
1353  * Removing a mount point may reveal a previously hidden file hierarchy, which
1354  * may then grant access to files, which may have previously been forbidden.
1355  */
1356 static int hook_sb_umount(struct vfsmount *const mnt, const int flags)
1357 {
1358 	if (!get_current_fs_domain())
1359 		return 0;
1360 	return -EPERM;
1361 }
1362 
1363 static int hook_sb_remount(struct super_block *const sb, void *const mnt_opts)
1364 {
1365 	if (!get_current_fs_domain())
1366 		return 0;
1367 	return -EPERM;
1368 }
1369 
1370 /*
1371  * pivot_root(2), like mount(2), changes the current mount namespace.  It must
1372  * then be forbidden for a landlocked process.
1373  *
1374  * However, chroot(2) may be allowed because it only changes the relative root
1375  * directory of the current process.  Moreover, it can be used to restrict the
1376  * view of the filesystem.
1377  */
1378 static int hook_sb_pivotroot(const struct path *const old_path,
1379 			     const struct path *const new_path)
1380 {
1381 	if (!get_current_fs_domain())
1382 		return 0;
1383 	return -EPERM;
1384 }
1385 
1386 /* Path hooks */
1387 
1388 static int hook_path_link(struct dentry *const old_dentry,
1389 			  const struct path *const new_dir,
1390 			  struct dentry *const new_dentry)
1391 {
1392 	return current_check_refer_path(old_dentry, new_dir, new_dentry, false,
1393 					false);
1394 }
1395 
1396 static int hook_path_rename(const struct path *const old_dir,
1397 			    struct dentry *const old_dentry,
1398 			    const struct path *const new_dir,
1399 			    struct dentry *const new_dentry,
1400 			    const unsigned int flags)
1401 {
1402 	/* old_dir refers to old_dentry->d_parent and new_dir->mnt */
1403 	return current_check_refer_path(old_dentry, new_dir, new_dentry, true,
1404 					!!(flags & RENAME_EXCHANGE));
1405 }
1406 
1407 static int hook_path_mkdir(const struct path *const dir,
1408 			   struct dentry *const dentry, const umode_t mode)
1409 {
1410 	return current_check_access_path(dir, LANDLOCK_ACCESS_FS_MAKE_DIR);
1411 }
1412 
1413 static int hook_path_mknod(const struct path *const dir,
1414 			   struct dentry *const dentry, const umode_t mode,
1415 			   const unsigned int dev)
1416 {
1417 	const struct landlock_ruleset *const dom = get_current_fs_domain();
1418 
1419 	if (!dom)
1420 		return 0;
1421 	return check_access_path(dom, dir, get_mode_access(mode));
1422 }
1423 
1424 static int hook_path_symlink(const struct path *const dir,
1425 			     struct dentry *const dentry,
1426 			     const char *const old_name)
1427 {
1428 	return current_check_access_path(dir, LANDLOCK_ACCESS_FS_MAKE_SYM);
1429 }
1430 
1431 static int hook_path_unlink(const struct path *const dir,
1432 			    struct dentry *const dentry)
1433 {
1434 	return current_check_access_path(dir, LANDLOCK_ACCESS_FS_REMOVE_FILE);
1435 }
1436 
1437 static int hook_path_rmdir(const struct path *const dir,
1438 			   struct dentry *const dentry)
1439 {
1440 	return current_check_access_path(dir, LANDLOCK_ACCESS_FS_REMOVE_DIR);
1441 }
1442 
1443 static int hook_path_truncate(const struct path *const path)
1444 {
1445 	return current_check_access_path(path, LANDLOCK_ACCESS_FS_TRUNCATE);
1446 }
1447 
1448 /* File hooks */
1449 
1450 /**
1451  * get_required_file_open_access - Get access needed to open a file
1452  *
1453  * @file: File being opened.
1454  *
1455  * Returns the access rights that are required for opening the given file,
1456  * depending on the file type and open mode.
1457  */
1458 static access_mask_t
1459 get_required_file_open_access(const struct file *const file)
1460 {
1461 	access_mask_t access = 0;
1462 
1463 	if (file->f_mode & FMODE_READ) {
1464 		/* A directory can only be opened in read mode. */
1465 		if (S_ISDIR(file_inode(file)->i_mode))
1466 			return LANDLOCK_ACCESS_FS_READ_DIR;
1467 		access = LANDLOCK_ACCESS_FS_READ_FILE;
1468 	}
1469 	if (file->f_mode & FMODE_WRITE)
1470 		access |= LANDLOCK_ACCESS_FS_WRITE_FILE;
1471 	/* __FMODE_EXEC is indeed part of f_flags, not f_mode. */
1472 	if (file->f_flags & __FMODE_EXEC)
1473 		access |= LANDLOCK_ACCESS_FS_EXECUTE;
1474 	return access;
1475 }
1476 
1477 static int hook_file_alloc_security(struct file *const file)
1478 {
1479 	/*
1480 	 * Grants all access rights, even if most of them are not checked later
1481 	 * on. It is more consistent.
1482 	 *
1483 	 * Notably, file descriptors for regular files can also be acquired
1484 	 * without going through the file_open hook, for example when using
1485 	 * memfd_create(2).
1486 	 */
1487 	landlock_file(file)->allowed_access = LANDLOCK_MASK_ACCESS_FS;
1488 	return 0;
1489 }
1490 
1491 static bool is_device(const struct file *const file)
1492 {
1493 	const struct inode *inode = file_inode(file);
1494 
1495 	return S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode);
1496 }
1497 
1498 static int hook_file_open(struct file *const file)
1499 {
1500 	layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {};
1501 	access_mask_t open_access_request, full_access_request, allowed_access,
1502 		optional_access;
1503 	const struct landlock_ruleset *const dom =
1504 		landlock_get_applicable_domain(
1505 			landlock_cred(file->f_cred)->domain, any_fs);
1506 
1507 	if (!dom)
1508 		return 0;
1509 
1510 	/*
1511 	 * Because a file may be opened with O_PATH, get_required_file_open_access()
1512 	 * may return 0.  This case will be handled with a future Landlock
1513 	 * evolution.
1514 	 */
1515 	open_access_request = get_required_file_open_access(file);
1516 
1517 	/*
1518 	 * We look up more access than what we immediately need for open(), so
1519 	 * that we can later authorize operations on opened files.
1520 	 */
1521 	optional_access = LANDLOCK_ACCESS_FS_TRUNCATE;
1522 	if (is_device(file))
1523 		optional_access |= LANDLOCK_ACCESS_FS_IOCTL_DEV;
1524 
1525 	full_access_request = open_access_request | optional_access;
1526 
1527 	if (is_access_to_paths_allowed(
1528 		    dom, &file->f_path,
1529 		    landlock_init_layer_masks(dom, full_access_request,
1530 					      &layer_masks, LANDLOCK_KEY_INODE),
1531 		    &layer_masks, NULL, 0, NULL, NULL)) {
1532 		allowed_access = full_access_request;
1533 	} else {
1534 		unsigned long access_bit;
1535 		const unsigned long access_req = full_access_request;
1536 
1537 		/*
1538 		 * Calculate the actual allowed access rights from layer_masks.
1539 		 * Add each access right to allowed_access which has not been
1540 		 * vetoed by any layer.
1541 		 */
1542 		allowed_access = 0;
1543 		for_each_set_bit(access_bit, &access_req,
1544 				 ARRAY_SIZE(layer_masks)) {
1545 			if (!layer_masks[access_bit])
1546 				allowed_access |= BIT_ULL(access_bit);
1547 		}
1548 	}
1549 
1550 	/*
1551 	 * For operations on already opened files (i.e. ftruncate()), it is the
1552 	 * access rights at the time of open() which decide whether the
1553 	 * operation is permitted. Therefore, we record the relevant subset of
1554 	 * file access rights in the opened struct file.
1555 	 */
1556 	landlock_file(file)->allowed_access = allowed_access;
1557 
1558 	if ((open_access_request & allowed_access) == open_access_request)
1559 		return 0;
1560 
1561 	return -EACCES;
1562 }
1563 
1564 static int hook_file_truncate(struct file *const file)
1565 {
1566 	/*
1567 	 * Allows truncation if the truncate right was available at the time of
1568 	 * opening the file, to get a consistent access check as for read, write
1569 	 * and execute operations.
1570 	 *
1571 	 * Note: For checks done based on the file's Landlock allowed access, we
1572 	 * enforce them independently of whether the current thread is in a
1573 	 * Landlock domain, so that open files passed between independent
1574 	 * processes retain their behaviour.
1575 	 */
1576 	if (landlock_file(file)->allowed_access & LANDLOCK_ACCESS_FS_TRUNCATE)
1577 		return 0;
1578 	return -EACCES;
1579 }
1580 
1581 static int hook_file_ioctl(struct file *file, unsigned int cmd,
1582 			   unsigned long arg)
1583 {
1584 	access_mask_t allowed_access = landlock_file(file)->allowed_access;
1585 
1586 	/*
1587 	 * It is the access rights at the time of opening the file which
1588 	 * determine whether IOCTL can be used on the opened file later.
1589 	 *
1590 	 * The access right is attached to the opened file in hook_file_open().
1591 	 */
1592 	if (allowed_access & LANDLOCK_ACCESS_FS_IOCTL_DEV)
1593 		return 0;
1594 
1595 	if (!is_device(file))
1596 		return 0;
1597 
1598 	if (is_masked_device_ioctl(cmd))
1599 		return 0;
1600 
1601 	return -EACCES;
1602 }
1603 
1604 static int hook_file_ioctl_compat(struct file *file, unsigned int cmd,
1605 				  unsigned long arg)
1606 {
1607 	access_mask_t allowed_access = landlock_file(file)->allowed_access;
1608 
1609 	/*
1610 	 * It is the access rights at the time of opening the file which
1611 	 * determine whether IOCTL can be used on the opened file later.
1612 	 *
1613 	 * The access right is attached to the opened file in hook_file_open().
1614 	 */
1615 	if (allowed_access & LANDLOCK_ACCESS_FS_IOCTL_DEV)
1616 		return 0;
1617 
1618 	if (!is_device(file))
1619 		return 0;
1620 
1621 	if (is_masked_device_ioctl_compat(cmd))
1622 		return 0;
1623 
1624 	return -EACCES;
1625 }
1626 
1627 static void hook_file_set_fowner(struct file *file)
1628 {
1629 	struct landlock_ruleset *new_dom, *prev_dom;
1630 
1631 	/*
1632 	 * Lock already held by __f_setown(), see commit 26f204380a3c ("fs: Fix
1633 	 * file_set_fowner LSM hook inconsistencies").
1634 	 */
1635 	lockdep_assert_held(&file_f_owner(file)->lock);
1636 	new_dom = landlock_get_current_domain();
1637 	landlock_get_ruleset(new_dom);
1638 	prev_dom = landlock_file(file)->fown_domain;
1639 	landlock_file(file)->fown_domain = new_dom;
1640 
1641 	/* Called in an RCU read-side critical section. */
1642 	landlock_put_ruleset_deferred(prev_dom);
1643 }
1644 
1645 static void hook_file_free_security(struct file *file)
1646 {
1647 	landlock_put_ruleset_deferred(landlock_file(file)->fown_domain);
1648 }
1649 
1650 static struct security_hook_list landlock_hooks[] __ro_after_init = {
1651 	LSM_HOOK_INIT(inode_free_security_rcu, hook_inode_free_security_rcu),
1652 
1653 	LSM_HOOK_INIT(sb_delete, hook_sb_delete),
1654 	LSM_HOOK_INIT(sb_mount, hook_sb_mount),
1655 	LSM_HOOK_INIT(move_mount, hook_move_mount),
1656 	LSM_HOOK_INIT(sb_umount, hook_sb_umount),
1657 	LSM_HOOK_INIT(sb_remount, hook_sb_remount),
1658 	LSM_HOOK_INIT(sb_pivotroot, hook_sb_pivotroot),
1659 
1660 	LSM_HOOK_INIT(path_link, hook_path_link),
1661 	LSM_HOOK_INIT(path_rename, hook_path_rename),
1662 	LSM_HOOK_INIT(path_mkdir, hook_path_mkdir),
1663 	LSM_HOOK_INIT(path_mknod, hook_path_mknod),
1664 	LSM_HOOK_INIT(path_symlink, hook_path_symlink),
1665 	LSM_HOOK_INIT(path_unlink, hook_path_unlink),
1666 	LSM_HOOK_INIT(path_rmdir, hook_path_rmdir),
1667 	LSM_HOOK_INIT(path_truncate, hook_path_truncate),
1668 
1669 	LSM_HOOK_INIT(file_alloc_security, hook_file_alloc_security),
1670 	LSM_HOOK_INIT(file_open, hook_file_open),
1671 	LSM_HOOK_INIT(file_truncate, hook_file_truncate),
1672 	LSM_HOOK_INIT(file_ioctl, hook_file_ioctl),
1673 	LSM_HOOK_INIT(file_ioctl_compat, hook_file_ioctl_compat),
1674 	LSM_HOOK_INIT(file_set_fowner, hook_file_set_fowner),
1675 	LSM_HOOK_INIT(file_free_security, hook_file_free_security),
1676 };
1677 
1678 __init void landlock_add_fs_hooks(void)
1679 {
1680 	security_add_hooks(landlock_hooks, ARRAY_SIZE(landlock_hooks),
1681 			   &landlock_lsmid);
1682 }
1683 
1684 #ifdef CONFIG_SECURITY_LANDLOCK_KUNIT_TEST
1685 
1686 /* clang-format off */
1687 static struct kunit_case test_cases[] = {
1688 	KUNIT_CASE(test_no_more_access),
1689 	KUNIT_CASE(test_scope_to_request_with_exec_none),
1690 	KUNIT_CASE(test_scope_to_request_with_exec_some),
1691 	KUNIT_CASE(test_scope_to_request_without_access),
1692 	KUNIT_CASE(test_is_eacces_with_none),
1693 	KUNIT_CASE(test_is_eacces_with_refer),
1694 	KUNIT_CASE(test_is_eacces_with_write),
1695 	{}
1696 };
1697 /* clang-format on */
1698 
1699 static struct kunit_suite test_suite = {
1700 	.name = "landlock_fs",
1701 	.test_cases = test_cases,
1702 };
1703 
1704 kunit_test_suite(test_suite);
1705 
1706 #endif /* CONFIG_SECURITY_LANDLOCK_KUNIT_TEST */
1707