fs.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Landlock - Filesystem management and hooks
   4 *
   5 * Copyright © 2016-2020 Mickaël Salaün <mic@digikod.net>
   6 * Copyright © 2018-2020 ANSSI
   7 * Copyright © 2021-2025 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_audit.h>
  27#include <linux/lsm_hooks.h>
  28#include <linux/mount.h>
  29#include <linux/namei.h>
  30#include <linux/path.h>
  31#include <linux/pid.h>
  32#include <linux/rcupdate.h>
  33#include <linux/sched/signal.h>
  34#include <linux/spinlock.h>
  35#include <linux/stat.h>
  36#include <linux/types.h>
  37#include <linux/wait_bit.h>
  38#include <linux/workqueue.h>
  39#include <uapi/linux/fiemap.h>
  40#include <uapi/linux/landlock.h>
  41
  42#include "access.h"
  43#include "audit.h"
  44#include "common.h"
  45#include "cred.h"
  46#include "domain.h"
  47#include "fs.h"
  48#include "limits.h"
  49#include "object.h"
  50#include "ruleset.h"
  51#include "setup.h"
  52
  53/* Underlying object management */
  54
  55static void release_inode(struct landlock_object *const object)
  56	__releases(object->lock)
  57{
  58	struct inode *const inode = object->underobj;
  59	struct super_block *sb;
  60
  61	if (!inode) {
  62		spin_unlock(&object->lock);
  63		return;
  64	}
  65
  66	/*
  67	 * Protects against concurrent use by hook_sb_delete() of the reference
  68	 * to the underlying inode.
  69	 */
  70	object->underobj = NULL;
  71	/*
  72	 * Makes sure that if the filesystem is concurrently unmounted,
  73	 * hook_sb_delete() will wait for us to finish iput().
  74	 */
  75	sb = inode->i_sb;
  76	atomic_long_inc(&landlock_superblock(sb)->inode_refs);
  77	spin_unlock(&object->lock);
  78	/*
  79	 * Because object->underobj was not NULL, hook_sb_delete() and
  80	 * get_inode_object() guarantee that it is safe to reset
  81	 * landlock_inode(inode)->object while it is not NULL.  It is therefore
  82	 * not necessary to lock inode->i_lock.
  83	 */
  84	rcu_assign_pointer(landlock_inode(inode)->object, NULL);
  85	/*
  86	 * Now, new rules can safely be tied to @inode with get_inode_object().
  87	 */
  88
  89	iput(inode);
  90	if (atomic_long_dec_and_test(&landlock_superblock(sb)->inode_refs))
  91		wake_up_var(&landlock_superblock(sb)->inode_refs);
  92}
  93
  94static const struct landlock_object_underops landlock_fs_underops = {
  95	.release = release_inode
  96};
  97
  98/* IOCTL helpers */
  99
 100/**
 101 * is_masked_device_ioctl - Determine whether an IOCTL command is always
 102 * permitted with Landlock for device files.  These commands can not be
 103 * restricted on device files by enforcing a Landlock policy.
 104 *
 105 * @cmd: The IOCTL command that is supposed to be run.
 106 *
 107 * By default, any IOCTL on a device file requires the
 108 * LANDLOCK_ACCESS_FS_IOCTL_DEV right.  However, we blanket-permit some
 109 * commands, if:
 110 *
 111 * 1. The command is implemented in fs/ioctl.c's do_vfs_ioctl(),
 112 *    not in f_ops->unlocked_ioctl() or f_ops->compat_ioctl().
 113 *
 114 * 2. The command is harmless when invoked on devices.
 115 *
 116 * We also permit commands that do not make sense for devices, but where the
 117 * do_vfs_ioctl() implementation returns a more conventional error code.
 118 *
 119 * Any new IOCTL commands that are implemented in fs/ioctl.c's do_vfs_ioctl()
 120 * should be considered for inclusion here.
 121 *
 122 * Returns: true if the IOCTL @cmd can not be restricted with Landlock for
 123 * device files.
 124 */
 125static __attribute_const__ bool is_masked_device_ioctl(const unsigned int cmd)
 126{
 127	switch (cmd) {
 128	/*
 129	 * FIOCLEX, FIONCLEX, FIONBIO and FIOASYNC manipulate the FD's
 130	 * close-on-exec and the file's buffered-IO and async flags.  These
 131	 * operations are also available through fcntl(2), and are
 132	 * unconditionally permitted in Landlock.
 133	 */
 134	case FIOCLEX:
 135	case FIONCLEX:
 136	case FIONBIO:
 137	case FIOASYNC:
 138	/*
 139	 * FIOQSIZE queries the size of a regular file, directory, or link.
 140	 *
 141	 * We still permit it, because it always returns -ENOTTY for
 142	 * other file types.
 143	 */
 144	case FIOQSIZE:
 145	/*
 146	 * FIFREEZE and FITHAW freeze and thaw the file system which the
 147	 * given file belongs to.  Requires CAP_SYS_ADMIN.
 148	 *
 149	 * These commands operate on the file system's superblock rather
 150	 * than on the file itself.  The same operations can also be
 151	 * done through any other file or directory on the same file
 152	 * system, so it is safe to permit these.
 153	 */
 154	case FIFREEZE:
 155	case FITHAW:
 156	/*
 157	 * FS_IOC_FIEMAP queries information about the allocation of
 158	 * blocks within a file.
 159	 *
 160	 * This IOCTL command only makes sense for regular files and is
 161	 * not implemented by devices. It is harmless to permit.
 162	 */
 163	case FS_IOC_FIEMAP:
 164	/*
 165	 * FIGETBSZ queries the file system's block size for a file or
 166	 * directory.
 167	 *
 168	 * This command operates on the file system's superblock rather
 169	 * than on the file itself.  The same operation can also be done
 170	 * through any other file or directory on the same file system,
 171	 * so it is safe to permit it.
 172	 */
 173	case FIGETBSZ:
 174	/*
 175	 * FICLONE, FICLONERANGE and FIDEDUPERANGE make files share
 176	 * their underlying storage ("reflink") between source and
 177	 * destination FDs, on file systems which support that.
 178	 *
 179	 * These IOCTL commands only apply to regular files
 180	 * and are harmless to permit for device files.
 181	 */
 182	case FICLONE:
 183	case FICLONERANGE:
 184	case FIDEDUPERANGE:
 185	/*
 186	 * FS_IOC_GETFSUUID and FS_IOC_GETFSSYSFSPATH both operate on
 187	 * the file system superblock, not on the specific file, so
 188	 * these operations are available through any other file on the
 189	 * same file system as well.
 190	 */
 191	case FS_IOC_GETFSUUID:
 192	case FS_IOC_GETFSSYSFSPATH:
 193		return true;
 194
 195	/*
 196	 * FIONREAD, FS_IOC_GETFLAGS, FS_IOC_SETFLAGS, FS_IOC_FSGETXATTR and
 197	 * FS_IOC_FSSETXATTR are forwarded to device implementations.
 198	 */
 199
 200	/*
 201	 * file_ioctl() commands (FIBMAP, FS_IOC_RESVSP, FS_IOC_RESVSP64,
 202	 * FS_IOC_UNRESVSP, FS_IOC_UNRESVSP64 and FS_IOC_ZERO_RANGE) are
 203	 * forwarded to device implementations, so not permitted.
 204	 */
 205
 206	/* Other commands are guarded by the access right. */
 207	default:
 208		return false;
 209	}
 210}
 211
 212/*
 213 * is_masked_device_ioctl_compat - same as the helper above, but checking the
 214 * "compat" IOCTL commands.
 215 *
 216 * The IOCTL commands with special handling in compat-mode should behave the
 217 * same as their non-compat counterparts.
 218 */
 219static __attribute_const__ bool
 220is_masked_device_ioctl_compat(const unsigned int cmd)
 221{
 222	switch (cmd) {
 223	/* FICLONE is permitted, same as in the non-compat variant. */
 224	case FICLONE:
 225		return true;
 226
 227#if defined(CONFIG_X86_64)
 228	/*
 229	 * FS_IOC_RESVSP_32, FS_IOC_RESVSP64_32, FS_IOC_UNRESVSP_32,
 230	 * FS_IOC_UNRESVSP64_32, FS_IOC_ZERO_RANGE_32: not blanket-permitted,
 231	 * for consistency with their non-compat variants.
 232	 */
 233	case FS_IOC_RESVSP_32:
 234	case FS_IOC_RESVSP64_32:
 235	case FS_IOC_UNRESVSP_32:
 236	case FS_IOC_UNRESVSP64_32:
 237	case FS_IOC_ZERO_RANGE_32:
 238#endif
 239
 240	/*
 241	 * FS_IOC32_GETFLAGS, FS_IOC32_SETFLAGS are forwarded to their device
 242	 * implementations.
 243	 */
 244	case FS_IOC32_GETFLAGS:
 245	case FS_IOC32_SETFLAGS:
 246		return false;
 247	default:
 248		return is_masked_device_ioctl(cmd);
 249	}
 250}
 251
 252/* Ruleset management */
 253
 254static struct landlock_object *get_inode_object(struct inode *const inode)
 255{
 256	struct landlock_object *object, *new_object;
 257	struct landlock_inode_security *inode_sec = landlock_inode(inode);
 258
 259	rcu_read_lock();
 260retry:
 261	object = rcu_dereference(inode_sec->object);
 262	if (object) {
 263		if (likely(refcount_inc_not_zero(&object->usage))) {
 264			rcu_read_unlock();
 265			return object;
 266		}
 267		/*
 268		 * We are racing with release_inode(), the object is going
 269		 * away.  Wait for release_inode(), then retry.
 270		 */
 271		spin_lock(&object->lock);
 272		spin_unlock(&object->lock);
 273		goto retry;
 274	}
 275	rcu_read_unlock();
 276
 277	/*
 278	 * If there is no object tied to @inode, then create a new one (without
 279	 * holding any locks).
 280	 */
 281	new_object = landlock_create_object(&landlock_fs_underops, inode);
 282	if (IS_ERR(new_object))
 283		return new_object;
 284
 285	/*
 286	 * Protects against concurrent calls to get_inode_object() or
 287	 * hook_sb_delete().
 288	 */
 289	spin_lock(&inode->i_lock);
 290	if (unlikely(rcu_access_pointer(inode_sec->object))) {
 291		/* Someone else just created the object, bail out and retry. */
 292		spin_unlock(&inode->i_lock);
 293		kfree(new_object);
 294
 295		rcu_read_lock();
 296		goto retry;
 297	}
 298
 299	/*
 300	 * @inode will be released by hook_sb_delete() on its superblock
 301	 * shutdown, or by release_inode() when no more ruleset references the
 302	 * related object.
 303	 */
 304	ihold(inode);
 305	rcu_assign_pointer(inode_sec->object, new_object);
 306	spin_unlock(&inode->i_lock);
 307	return new_object;
 308}
 309
 310/* All access rights that can be tied to files. */
 311/* clang-format off */
 312#define ACCESS_FILE ( \
 313	LANDLOCK_ACCESS_FS_EXECUTE | \
 314	LANDLOCK_ACCESS_FS_WRITE_FILE | \
 315	LANDLOCK_ACCESS_FS_READ_FILE | \
 316	LANDLOCK_ACCESS_FS_TRUNCATE | \
 317	LANDLOCK_ACCESS_FS_IOCTL_DEV)
 318/* clang-format on */
 319
 320/*
 321 * @path: Should have been checked by get_path_from_fd().
 322 */
 323int landlock_append_fs_rule(struct landlock_ruleset *const ruleset,
 324			    const struct path *const path,
 325			    access_mask_t access_rights)
 326{
 327	int err;
 328	struct landlock_id id = {
 329		.type = LANDLOCK_KEY_INODE,
 330	};
 331
 332	/* Files only get access rights that make sense. */
 333	if (!d_is_dir(path->dentry) &&
 334	    (access_rights | ACCESS_FILE) != ACCESS_FILE)
 335		return -EINVAL;
 336	if (WARN_ON_ONCE(ruleset->num_layers != 1))
 337		return -EINVAL;
 338
 339	/* Transforms relative access rights to absolute ones. */
 340	access_rights |= LANDLOCK_MASK_ACCESS_FS &
 341			 ~landlock_get_fs_access_mask(ruleset, 0);
 342	id.key.object = get_inode_object(d_backing_inode(path->dentry));
 343	if (IS_ERR(id.key.object))
 344		return PTR_ERR(id.key.object);
 345	mutex_lock(&ruleset->lock);
 346	err = landlock_insert_rule(ruleset, id, access_rights);
 347	mutex_unlock(&ruleset->lock);
 348	/*
 349	 * No need to check for an error because landlock_insert_rule()
 350	 * increments the refcount for the new object if needed.
 351	 */
 352	landlock_put_object(id.key.object);
 353	return err;
 354}
 355
 356/* Access-control management */
 357
 358/*
 359 * The lifetime of the returned rule is tied to @domain.
 360 *
 361 * Returns NULL if no rule is found or if @dentry is negative.
 362 */
 363static const struct landlock_rule *
 364find_rule(const struct landlock_ruleset *const domain,
 365	  const struct dentry *const dentry)
 366{
 367	const struct landlock_rule *rule;
 368	const struct inode *inode;
 369	struct landlock_id id = {
 370		.type = LANDLOCK_KEY_INODE,
 371	};
 372
 373	/* Ignores nonexistent leafs. */
 374	if (d_is_negative(dentry))
 375		return NULL;
 376
 377	inode = d_backing_inode(dentry);
 378	rcu_read_lock();
 379	id.key.object = rcu_dereference(landlock_inode(inode)->object);
 380	rule = landlock_find_rule(domain, id);
 381	rcu_read_unlock();
 382	return rule;
 383}
 384
 385/*
 386 * Allows access to pseudo filesystems that will never be mountable (e.g.
 387 * sockfs, pipefs), but can still be reachable through
 388 * /proc/<pid>/fd/<file-descriptor>
 389 */
 390static bool is_nouser_or_private(const struct dentry *dentry)
 391{
 392	return (dentry->d_sb->s_flags & SB_NOUSER) ||
 393	       (d_is_positive(dentry) &&
 394		unlikely(IS_PRIVATE(d_backing_inode(dentry))));
 395}
 396
 397static const struct access_masks any_fs = {
 398	.fs = ~0,
 399};
 400
 401/*
 402 * Check that a destination file hierarchy has more restrictions than a source
 403 * file hierarchy.  This is only used for link and rename actions.
 404 *
 405 * @layer_masks_child2: Optional child masks.
 406 */
 407static bool no_more_access(
 408	const layer_mask_t (*const layer_masks_parent1)[LANDLOCK_NUM_ACCESS_FS],
 409	const layer_mask_t (*const layer_masks_child1)[LANDLOCK_NUM_ACCESS_FS],
 410	const bool child1_is_directory,
 411	const layer_mask_t (*const layer_masks_parent2)[LANDLOCK_NUM_ACCESS_FS],
 412	const layer_mask_t (*const layer_masks_child2)[LANDLOCK_NUM_ACCESS_FS],
 413	const bool child2_is_directory)
 414{
 415	unsigned long access_bit;
 416
 417	for (access_bit = 0; access_bit < ARRAY_SIZE(*layer_masks_parent2);
 418	     access_bit++) {
 419		/* Ignores accesses that only make sense for directories. */
 420		const bool is_file_access =
 421			!!(BIT_ULL(access_bit) & ACCESS_FILE);
 422
 423		if (child1_is_directory || is_file_access) {
 424			/*
 425			 * Checks if the destination restrictions are a
 426			 * superset of the source ones (i.e. inherited access
 427			 * rights without child exceptions):
 428			 * restrictions(parent2) >= restrictions(child1)
 429			 */
 430			if ((((*layer_masks_parent1)[access_bit] &
 431			      (*layer_masks_child1)[access_bit]) |
 432			     (*layer_masks_parent2)[access_bit]) !=
 433			    (*layer_masks_parent2)[access_bit])
 434				return false;
 435		}
 436
 437		if (!layer_masks_child2)
 438			continue;
 439		if (child2_is_directory || is_file_access) {
 440			/*
 441			 * Checks inverted restrictions for RENAME_EXCHANGE:
 442			 * restrictions(parent1) >= restrictions(child2)
 443			 */
 444			if ((((*layer_masks_parent2)[access_bit] &
 445			      (*layer_masks_child2)[access_bit]) |
 446			     (*layer_masks_parent1)[access_bit]) !=
 447			    (*layer_masks_parent1)[access_bit])
 448				return false;
 449		}
 450	}
 451	return true;
 452}
 453
 454#define NMA_TRUE(...) KUNIT_EXPECT_TRUE(test, no_more_access(__VA_ARGS__))
 455#define NMA_FALSE(...) KUNIT_EXPECT_FALSE(test, no_more_access(__VA_ARGS__))
 456
 457#ifdef CONFIG_SECURITY_LANDLOCK_KUNIT_TEST
 458
 459static void test_no_more_access(struct kunit *const test)
 460{
 461	const layer_mask_t rx0[LANDLOCK_NUM_ACCESS_FS] = {
 462		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(0),
 463		[BIT_INDEX(LANDLOCK_ACCESS_FS_READ_FILE)] = BIT_ULL(0),
 464	};
 465	const layer_mask_t mx0[LANDLOCK_NUM_ACCESS_FS] = {
 466		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(0),
 467		[BIT_INDEX(LANDLOCK_ACCESS_FS_MAKE_REG)] = BIT_ULL(0),
 468	};
 469	const layer_mask_t x0[LANDLOCK_NUM_ACCESS_FS] = {
 470		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(0),
 471	};
 472	const layer_mask_t x1[LANDLOCK_NUM_ACCESS_FS] = {
 473		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(1),
 474	};
 475	const layer_mask_t x01[LANDLOCK_NUM_ACCESS_FS] = {
 476		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(0) |
 477							  BIT_ULL(1),
 478	};
 479	const layer_mask_t allows_all[LANDLOCK_NUM_ACCESS_FS] = {};
 480
 481	/* Checks without restriction. */
 482	NMA_TRUE(&x0, &allows_all, false, &allows_all, NULL, false);
 483	NMA_TRUE(&allows_all, &x0, false, &allows_all, NULL, false);
 484	NMA_FALSE(&x0, &x0, false, &allows_all, NULL, false);
 485
 486	/*
 487	 * Checks that we can only refer a file if no more access could be
 488	 * inherited.
 489	 */
 490	NMA_TRUE(&x0, &x0, false, &rx0, NULL, false);
 491	NMA_TRUE(&rx0, &rx0, false, &rx0, NULL, false);
 492	NMA_FALSE(&rx0, &rx0, false, &x0, NULL, false);
 493	NMA_FALSE(&rx0, &rx0, false, &x1, NULL, false);
 494
 495	/* Checks allowed referring with different nested domains. */
 496	NMA_TRUE(&x0, &x1, false, &x0, NULL, false);
 497	NMA_TRUE(&x1, &x0, false, &x0, NULL, false);
 498	NMA_TRUE(&x0, &x01, false, &x0, NULL, false);
 499	NMA_TRUE(&x0, &x01, false, &rx0, NULL, false);
 500	NMA_TRUE(&x01, &x0, false, &x0, NULL, false);
 501	NMA_TRUE(&x01, &x0, false, &rx0, NULL, false);
 502	NMA_FALSE(&x01, &x01, false, &x0, NULL, false);
 503
 504	/* Checks that file access rights are also enforced for a directory. */
 505	NMA_FALSE(&rx0, &rx0, true, &x0, NULL, false);
 506
 507	/* Checks that directory access rights don't impact file referring... */
 508	NMA_TRUE(&mx0, &mx0, false, &x0, NULL, false);
 509	/* ...but only directory referring. */
 510	NMA_FALSE(&mx0, &mx0, true, &x0, NULL, false);
 511
 512	/* Checks directory exchange. */
 513	NMA_TRUE(&mx0, &mx0, true, &mx0, &mx0, true);
 514	NMA_TRUE(&mx0, &mx0, true, &mx0, &x0, true);
 515	NMA_FALSE(&mx0, &mx0, true, &x0, &mx0, true);
 516	NMA_FALSE(&mx0, &mx0, true, &x0, &x0, true);
 517	NMA_FALSE(&mx0, &mx0, true, &x1, &x1, true);
 518
 519	/* Checks file exchange with directory access rights... */
 520	NMA_TRUE(&mx0, &mx0, false, &mx0, &mx0, false);
 521	NMA_TRUE(&mx0, &mx0, false, &mx0, &x0, false);
 522	NMA_TRUE(&mx0, &mx0, false, &x0, &mx0, false);
 523	NMA_TRUE(&mx0, &mx0, false, &x0, &x0, false);
 524	/* ...and with file access rights. */
 525	NMA_TRUE(&rx0, &rx0, false, &rx0, &rx0, false);
 526	NMA_TRUE(&rx0, &rx0, false, &rx0, &x0, false);
 527	NMA_FALSE(&rx0, &rx0, false, &x0, &rx0, false);
 528	NMA_FALSE(&rx0, &rx0, false, &x0, &x0, false);
 529	NMA_FALSE(&rx0, &rx0, false, &x1, &x1, false);
 530
 531	/*
 532	 * Allowing the following requests should not be a security risk
 533	 * because domain 0 denies execute access, and domain 1 is always
 534	 * nested with domain 0.  However, adding an exception for this case
 535	 * would mean to check all nested domains to make sure none can get
 536	 * more privileges (e.g. processes only sandboxed by domain 0).
 537	 * Moreover, this behavior (i.e. composition of N domains) could then
 538	 * be inconsistent compared to domain 1's ruleset alone (e.g. it might
 539	 * be denied to link/rename with domain 1's ruleset, whereas it would
 540	 * be allowed if nested on top of domain 0).  Another drawback would be
 541	 * to create a cover channel that could enable sandboxed processes to
 542	 * infer most of the filesystem restrictions from their domain.  To
 543	 * make it simple, efficient, safe, and more consistent, this case is
 544	 * always denied.
 545	 */
 546	NMA_FALSE(&x1, &x1, false, &x0, NULL, false);
 547	NMA_FALSE(&x1, &x1, false, &rx0, NULL, false);
 548	NMA_FALSE(&x1, &x1, true, &x0, NULL, false);
 549	NMA_FALSE(&x1, &x1, true, &rx0, NULL, false);
 550
 551	/* Checks the same case of exclusive domains with a file... */
 552	NMA_TRUE(&x1, &x1, false, &x01, NULL, false);
 553	NMA_FALSE(&x1, &x1, false, &x01, &x0, false);
 554	NMA_FALSE(&x1, &x1, false, &x01, &x01, false);
 555	NMA_FALSE(&x1, &x1, false, &x0, &x0, false);
 556	/* ...and with a directory. */
 557	NMA_FALSE(&x1, &x1, false, &x0, &x0, true);
 558	NMA_FALSE(&x1, &x1, true, &x0, &x0, false);
 559	NMA_FALSE(&x1, &x1, true, &x0, &x0, true);
 560}
 561
 562#endif /* CONFIG_SECURITY_LANDLOCK_KUNIT_TEST */
 563
 564#undef NMA_TRUE
 565#undef NMA_FALSE
 566
 567static bool is_layer_masks_allowed(
 568	layer_mask_t (*const layer_masks)[LANDLOCK_NUM_ACCESS_FS])
 569{
 570	return !memchr_inv(layer_masks, 0, sizeof(*layer_masks));
 571}
 572
 573/*
 574 * Removes @layer_masks accesses that are not requested.
 575 *
 576 * Returns true if the request is allowed, false otherwise.
 577 */
 578static bool
 579scope_to_request(const access_mask_t access_request,
 580		 layer_mask_t (*const layer_masks)[LANDLOCK_NUM_ACCESS_FS])
 581{
 582	const unsigned long access_req = access_request;
 583	unsigned long access_bit;
 584
 585	if (WARN_ON_ONCE(!layer_masks))
 586		return true;
 587
 588	for_each_clear_bit(access_bit, &access_req, ARRAY_SIZE(*layer_masks))
 589		(*layer_masks)[access_bit] = 0;
 590
 591	return is_layer_masks_allowed(layer_masks);
 592}
 593
 594#ifdef CONFIG_SECURITY_LANDLOCK_KUNIT_TEST
 595
 596static void test_scope_to_request_with_exec_none(struct kunit *const test)
 597{
 598	/* Allows everything. */
 599	layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {};
 600
 601	/* Checks and scopes with execute. */
 602	KUNIT_EXPECT_TRUE(test, scope_to_request(LANDLOCK_ACCESS_FS_EXECUTE,
 603						 &layer_masks));
 604	KUNIT_EXPECT_EQ(test, 0,
 605			layer_masks[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)]);
 606	KUNIT_EXPECT_EQ(test, 0,
 607			layer_masks[BIT_INDEX(LANDLOCK_ACCESS_FS_WRITE_FILE)]);
 608}
 609
 610static void test_scope_to_request_with_exec_some(struct kunit *const test)
 611{
 612	/* Denies execute and write. */
 613	layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {
 614		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(0),
 615		[BIT_INDEX(LANDLOCK_ACCESS_FS_WRITE_FILE)] = BIT_ULL(1),
 616	};
 617
 618	/* Checks and scopes with execute. */
 619	KUNIT_EXPECT_FALSE(test, scope_to_request(LANDLOCK_ACCESS_FS_EXECUTE,
 620						  &layer_masks));
 621	KUNIT_EXPECT_EQ(test, BIT_ULL(0),
 622			layer_masks[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)]);
 623	KUNIT_EXPECT_EQ(test, 0,
 624			layer_masks[BIT_INDEX(LANDLOCK_ACCESS_FS_WRITE_FILE)]);
 625}
 626
 627static void test_scope_to_request_without_access(struct kunit *const test)
 628{
 629	/* Denies execute and write. */
 630	layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {
 631		[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)] = BIT_ULL(0),
 632		[BIT_INDEX(LANDLOCK_ACCESS_FS_WRITE_FILE)] = BIT_ULL(1),
 633	};
 634
 635	/* Checks and scopes without access request. */
 636	KUNIT_EXPECT_TRUE(test, scope_to_request(0, &layer_masks));
 637	KUNIT_EXPECT_EQ(test, 0,
 638			layer_masks[BIT_INDEX(LANDLOCK_ACCESS_FS_EXECUTE)]);
 639	KUNIT_EXPECT_EQ(test, 0,
 640			layer_masks[BIT_INDEX(LANDLOCK_ACCESS_FS_WRITE_FILE)]);
 641}
 642
 643#endif /* CONFIG_SECURITY_LANDLOCK_KUNIT_TEST */
 644
 645/*
 646 * Returns true if there is at least one access right different than
 647 * LANDLOCK_ACCESS_FS_REFER.
 648 */
 649static bool
 650is_eacces(const layer_mask_t (*const layer_masks)[LANDLOCK_NUM_ACCESS_FS],
 651	  const access_mask_t access_request)
 652{
 653	unsigned long access_bit;
 654	/* LANDLOCK_ACCESS_FS_REFER alone must return -EXDEV. */
 655	const unsigned long access_check = access_request &
 656					   ~LANDLOCK_ACCESS_FS_REFER;
 657
 658	if (!layer_masks)
 659		return false;
 660
 661	for_each_set_bit(access_bit, &access_check, ARRAY_SIZE(*layer_masks)) {
 662		if ((*layer_masks)[access_bit])
 663			return true;
 664	}
 665	return false;
 666}
 667
 668#define IE_TRUE(...) KUNIT_EXPECT_TRUE(test, is_eacces(__VA_ARGS__))
 669#define IE_FALSE(...) KUNIT_EXPECT_FALSE(test, is_eacces(__VA_ARGS__))
 670
 671#ifdef CONFIG_SECURITY_LANDLOCK_KUNIT_TEST
 672
 673static void test_is_eacces_with_none(struct kunit *const test)
 674{
 675	const layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {};
 676
 677	IE_FALSE(&layer_masks, 0);
 678	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_REFER);
 679	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_EXECUTE);
 680	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_WRITE_FILE);
 681}
 682
 683static void test_is_eacces_with_refer(struct kunit *const test)
 684{
 685	const layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {
 686		[BIT_INDEX(LANDLOCK_ACCESS_FS_REFER)] = BIT_ULL(0),
 687	};
 688
 689	IE_FALSE(&layer_masks, 0);
 690	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_REFER);
 691	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_EXECUTE);
 692	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_WRITE_FILE);
 693}
 694
 695static void test_is_eacces_with_write(struct kunit *const test)
 696{
 697	const layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {
 698		[BIT_INDEX(LANDLOCK_ACCESS_FS_WRITE_FILE)] = BIT_ULL(0),
 699	};
 700
 701	IE_FALSE(&layer_masks, 0);
 702	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_REFER);
 703	IE_FALSE(&layer_masks, LANDLOCK_ACCESS_FS_EXECUTE);
 704
 705	IE_TRUE(&layer_masks, LANDLOCK_ACCESS_FS_WRITE_FILE);
 706}
 707
 708#endif /* CONFIG_SECURITY_LANDLOCK_KUNIT_TEST */
 709
 710#undef IE_TRUE
 711#undef IE_FALSE
 712
 713/**
 714 * is_access_to_paths_allowed - Check accesses for requests with a common path
 715 *
 716 * @domain: Domain to check against.
 717 * @path: File hierarchy to walk through.  For refer checks, this would be
 718 *     the common mountpoint.
 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 * @log_request_parent1: Audit request to fill if the related access is denied.
 732 * @dentry_child1: Dentry to the initial child of the parent1 path.  This
 733 *     pointer must be NULL for non-refer actions (i.e. not link nor rename).
 734 * @access_request_parent2: Similar to @access_request_parent1 but for a
 735 *     request involving a source and a destination.  This refers to the
 736 *     destination, except in case of RENAME_EXCHANGE where it also refers to
 737 *     the source.  Must be set to 0 when using a simple path request.
 738 * @layer_masks_parent2: Similar to @layer_masks_parent1 but for a refer
 739 *     action.  This must be NULL otherwise.
 740 * @log_request_parent2: Audit request to fill if the related access is denied.
 741 * @dentry_child2: Dentry to the initial child of the parent2 path.  This
 742 *     pointer is only set for RENAME_EXCHANGE actions and must be NULL
 743 *     otherwise.
 744 *
 745 * This helper first checks that the destination has a superset of restrictions
 746 * compared to the source (if any) for a common path.  Because of
 747 * RENAME_EXCHANGE actions, source and destinations may be swapped.  It then
 748 * checks that the collected accesses and the remaining ones are enough to
 749 * allow the request.
 750 *
 751 * Returns:
 752 * - true if the access request is granted;
 753 * - false otherwise.
 754 */
 755static bool is_access_to_paths_allowed(
 756	const struct landlock_ruleset *const domain,
 757	const struct path *const path,
 758	const access_mask_t access_request_parent1,
 759	layer_mask_t (*const layer_masks_parent1)[LANDLOCK_NUM_ACCESS_FS],
 760	struct landlock_request *const log_request_parent1,
 761	struct dentry *const dentry_child1,
 762	const access_mask_t access_request_parent2,
 763	layer_mask_t (*const layer_masks_parent2)[LANDLOCK_NUM_ACCESS_FS],
 764	struct landlock_request *const log_request_parent2,
 765	struct dentry *const dentry_child2)
 766{
 767	bool allowed_parent1 = false, allowed_parent2 = false, is_dom_check,
 768	     child1_is_directory = true, child2_is_directory = true;
 769	struct path walker_path;
 770	access_mask_t access_masked_parent1, access_masked_parent2;
 771	layer_mask_t _layer_masks_child1[LANDLOCK_NUM_ACCESS_FS],
 772		_layer_masks_child2[LANDLOCK_NUM_ACCESS_FS];
 773	layer_mask_t(*layer_masks_child1)[LANDLOCK_NUM_ACCESS_FS] = NULL,
 774	(*layer_masks_child2)[LANDLOCK_NUM_ACCESS_FS] = NULL;
 775
 776	if (!access_request_parent1 && !access_request_parent2)
 777		return true;
 778
 779	if (WARN_ON_ONCE(!path))
 780		return true;
 781
 782	if (is_nouser_or_private(path->dentry))
 783		return true;
 784
 785	if (WARN_ON_ONCE(!layer_masks_parent1))
 786		return false;
 787
 788	allowed_parent1 = is_layer_masks_allowed(layer_masks_parent1);
 789
 790	if (unlikely(layer_masks_parent2)) {
 791		if (WARN_ON_ONCE(!dentry_child1))
 792			return false;
 793
 794		allowed_parent2 = is_layer_masks_allowed(layer_masks_parent2);
 795
 796		/*
 797		 * For a double request, first check for potential privilege
 798		 * escalation by looking at domain handled accesses (which are
 799		 * a superset of the meaningful requested accesses).
 800		 */
 801		access_masked_parent1 = access_masked_parent2 =
 802			landlock_union_access_masks(domain).fs;
 803		is_dom_check = true;
 804	} else {
 805		if (WARN_ON_ONCE(dentry_child1 || dentry_child2))
 806			return false;
 807		/* For a simple request, only check for requested accesses. */
 808		access_masked_parent1 = access_request_parent1;
 809		access_masked_parent2 = access_request_parent2;
 810		is_dom_check = false;
 811	}
 812
 813	if (unlikely(dentry_child1)) {
 814		landlock_unmask_layers(
 815			find_rule(domain, dentry_child1),
 816			landlock_init_layer_masks(
 817				domain, LANDLOCK_MASK_ACCESS_FS,
 818				&_layer_masks_child1, LANDLOCK_KEY_INODE),
 819			&_layer_masks_child1, ARRAY_SIZE(_layer_masks_child1));
 820		layer_masks_child1 = &_layer_masks_child1;
 821		child1_is_directory = d_is_dir(dentry_child1);
 822	}
 823	if (unlikely(dentry_child2)) {
 824		landlock_unmask_layers(
 825			find_rule(domain, dentry_child2),
 826			landlock_init_layer_masks(
 827				domain, LANDLOCK_MASK_ACCESS_FS,
 828				&_layer_masks_child2, LANDLOCK_KEY_INODE),
 829			&_layer_masks_child2, ARRAY_SIZE(_layer_masks_child2));
 830		layer_masks_child2 = &_layer_masks_child2;
 831		child2_is_directory = d_is_dir(dentry_child2);
 832	}
 833
 834	walker_path = *path;
 835	path_get(&walker_path);
 836	/*
 837	 * We need to walk through all the hierarchy to not miss any relevant
 838	 * restriction.
 839	 */
 840	while (true) {
 841		const struct landlock_rule *rule;
 842
 843		/*
 844		 * If at least all accesses allowed on the destination are
 845		 * already allowed on the source, respectively if there is at
 846		 * least as much as restrictions on the destination than on the
 847		 * source, then we can safely refer files from the source to
 848		 * the destination without risking a privilege escalation.
 849		 * This also applies in the case of RENAME_EXCHANGE, which
 850		 * implies checks on both direction.  This is crucial for
 851		 * standalone multilayered security policies.  Furthermore,
 852		 * this helps avoid policy writers to shoot themselves in the
 853		 * foot.
 854		 */
 855		if (unlikely(is_dom_check &&
 856			     no_more_access(
 857				     layer_masks_parent1, layer_masks_child1,
 858				     child1_is_directory, layer_masks_parent2,
 859				     layer_masks_child2,
 860				     child2_is_directory))) {
 861			/*
 862			 * Now, downgrades the remaining checks from domain
 863			 * handled accesses to requested accesses.
 864			 */
 865			is_dom_check = false;
 866			access_masked_parent1 = access_request_parent1;
 867			access_masked_parent2 = access_request_parent2;
 868
 869			allowed_parent1 =
 870				allowed_parent1 ||
 871				scope_to_request(access_masked_parent1,
 872						 layer_masks_parent1);
 873			allowed_parent2 =
 874				allowed_parent2 ||
 875				scope_to_request(access_masked_parent2,
 876						 layer_masks_parent2);
 877
 878			/* Stops when all accesses are granted. */
 879			if (allowed_parent1 && allowed_parent2)
 880				break;
 881		}
 882
 883		rule = find_rule(domain, walker_path.dentry);
 884		allowed_parent1 = allowed_parent1 ||
 885				  landlock_unmask_layers(
 886					  rule, access_masked_parent1,
 887					  layer_masks_parent1,
 888					  ARRAY_SIZE(*layer_masks_parent1));
 889		allowed_parent2 = allowed_parent2 ||
 890				  landlock_unmask_layers(
 891					  rule, access_masked_parent2,
 892					  layer_masks_parent2,
 893					  ARRAY_SIZE(*layer_masks_parent2));
 894
 895		/* Stops when a rule from each layer grants access. */
 896		if (allowed_parent1 && allowed_parent2)
 897			break;
 898
 899jump_up:
 900		if (walker_path.dentry == walker_path.mnt->mnt_root) {
 901			if (follow_up(&walker_path)) {
 902				/* Ignores hidden mount points. */
 903				goto jump_up;
 904			} else {
 905				/*
 906				 * Stops at the real root.  Denies access
 907				 * because not all layers have granted access.
 908				 */
 909				break;
 910			}
 911		}
 912
 913		if (unlikely(IS_ROOT(walker_path.dentry))) {
 914			if (likely(walker_path.mnt->mnt_flags & MNT_INTERNAL)) {
 915				/*
 916				 * Stops and allows access when reaching disconnected root
 917				 * directories that are part of internal filesystems (e.g. nsfs,
 918				 * which is reachable through /proc/<pid>/ns/<namespace>).
 919				 */
 920				allowed_parent1 = true;
 921				allowed_parent2 = true;
 922				break;
 923			}
 924
 925			/*
 926			 * We reached a disconnected root directory from a bind mount.
 927			 * Let's continue the walk with the mount point we missed.
 928			 */
 929			dput(walker_path.dentry);
 930			walker_path.dentry = walker_path.mnt->mnt_root;
 931			dget(walker_path.dentry);
 932		} else {
 933			struct dentry *const parent_dentry =
 934				dget_parent(walker_path.dentry);
 935
 936			dput(walker_path.dentry);
 937			walker_path.dentry = parent_dentry;
 938		}
 939	}
 940	path_put(&walker_path);
 941
 942	/*
 943	 * Check CONFIG_AUDIT to enable elision of log_request_parent* and
 944	 * associated caller's stack variables thanks to dead code elimination.
 945	 */
 946#ifdef CONFIG_AUDIT
 947	if (!allowed_parent1 && log_request_parent1) {
 948		log_request_parent1->type = LANDLOCK_REQUEST_FS_ACCESS;
 949		log_request_parent1->audit.type = LSM_AUDIT_DATA_PATH;
 950		log_request_parent1->audit.u.path = *path;
 951		log_request_parent1->access = access_masked_parent1;
 952		log_request_parent1->layer_masks = layer_masks_parent1;
 953		log_request_parent1->layer_masks_size =
 954			ARRAY_SIZE(*layer_masks_parent1);
 955	}
 956
 957	if (!allowed_parent2 && log_request_parent2) {
 958		log_request_parent2->type = LANDLOCK_REQUEST_FS_ACCESS;
 959		log_request_parent2->audit.type = LSM_AUDIT_DATA_PATH;
 960		log_request_parent2->audit.u.path = *path;
 961		log_request_parent2->access = access_masked_parent2;
 962		log_request_parent2->layer_masks = layer_masks_parent2;
 963		log_request_parent2->layer_masks_size =
 964			ARRAY_SIZE(*layer_masks_parent2);
 965	}
 966#endif /* CONFIG_AUDIT */
 967
 968	return allowed_parent1 && allowed_parent2;
 969}
 970
 971static int current_check_access_path(const struct path *const path,
 972				     access_mask_t access_request)
 973{
 974	const struct access_masks masks = {
 975		.fs = access_request,
 976	};
 977	const struct landlock_cred_security *const subject =
 978		landlock_get_applicable_subject(current_cred(), masks, NULL);
 979	layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {};
 980	struct landlock_request request = {};
 981
 982	if (!subject)
 983		return 0;
 984
 985	access_request = landlock_init_layer_masks(subject->domain,
 986						   access_request, &layer_masks,
 987						   LANDLOCK_KEY_INODE);
 988	if (is_access_to_paths_allowed(subject->domain, path, access_request,
 989				       &layer_masks, &request, NULL, 0, NULL,
 990				       NULL, NULL))
 991		return 0;
 992
 993	landlock_log_denial(subject, &request);
 994	return -EACCES;
 995}
 996
 997static __attribute_const__ access_mask_t get_mode_access(const umode_t mode)
 998{
 999	switch (mode & S_IFMT) {
1000	case S_IFLNK:
1001		return LANDLOCK_ACCESS_FS_MAKE_SYM;
1002	case S_IFDIR:
1003		return LANDLOCK_ACCESS_FS_MAKE_DIR;
1004	case S_IFCHR:
1005		return LANDLOCK_ACCESS_FS_MAKE_CHAR;
1006	case S_IFBLK:
1007		return LANDLOCK_ACCESS_FS_MAKE_BLOCK;
1008	case S_IFIFO:
1009		return LANDLOCK_ACCESS_FS_MAKE_FIFO;
1010	case S_IFSOCK:
1011		return LANDLOCK_ACCESS_FS_MAKE_SOCK;
1012	case S_IFREG:
1013	case 0:
1014		/* A zero mode translates to S_IFREG. */
1015	default:
1016		/* Treats weird files as regular files. */
1017		return LANDLOCK_ACCESS_FS_MAKE_REG;
1018	}
1019}
1020
1021static access_mask_t maybe_remove(const struct dentry *const dentry)
1022{
1023	if (d_is_negative(dentry))
1024		return 0;
1025	return d_is_dir(dentry) ? LANDLOCK_ACCESS_FS_REMOVE_DIR :
1026				  LANDLOCK_ACCESS_FS_REMOVE_FILE;
1027}
1028
1029/**
1030 * collect_domain_accesses - Walk through a file path and collect accesses
1031 *
1032 * @domain: Domain to check against.
1033 * @mnt_root: Last directory to check.
1034 * @dir: Directory to start the walk from.
1035 * @layer_masks_dom: Where to store the collected accesses.
1036 *
1037 * This helper is useful to begin a path walk from the @dir directory to a
1038 * @mnt_root directory used as a mount point.  This mount point is the common
1039 * ancestor between the source and the destination of a renamed and linked
1040 * file.  While walking from @dir to @mnt_root, we record all the domain's
1041 * allowed accesses in @layer_masks_dom.
1042 *
1043 * Because of disconnected directories, this walk may not reach @mnt_dir.  In
1044 * this case, the walk will continue to @mnt_dir after this call.
1045 *
1046 * This is similar to is_access_to_paths_allowed() but much simpler because it
1047 * only handles walking on the same mount point and only checks one set of
1048 * accesses.
1049 *
1050 * Returns:
1051 * - true if all the domain access rights are allowed for @dir;
1052 * - false if the walk reached @mnt_root.
1053 */
1054static bool collect_domain_accesses(
1055	const struct landlock_ruleset *const domain,
1056	const struct dentry *const mnt_root, struct dentry *dir,
1057	layer_mask_t (*const layer_masks_dom)[LANDLOCK_NUM_ACCESS_FS])
1058{
1059	unsigned long access_dom;
1060	bool ret = false;
1061
1062	if (WARN_ON_ONCE(!domain || !mnt_root || !dir || !layer_masks_dom))
1063		return true;
1064	if (is_nouser_or_private(dir))
1065		return true;
1066
1067	access_dom = landlock_init_layer_masks(domain, LANDLOCK_MASK_ACCESS_FS,
1068					       layer_masks_dom,
1069					       LANDLOCK_KEY_INODE);
1070
1071	dget(dir);
1072	while (true) {
1073		struct dentry *parent_dentry;
1074
1075		/* Gets all layers allowing all domain accesses. */
1076		if (landlock_unmask_layers(find_rule(domain, dir), access_dom,
1077					   layer_masks_dom,
1078					   ARRAY_SIZE(*layer_masks_dom))) {
1079			/*
1080			 * Stops when all handled accesses are allowed by at
1081			 * least one rule in each layer.
1082			 */
1083			ret = true;
1084			break;
1085		}
1086
1087		/*
1088		 * Stops at the mount point or the filesystem root for a disconnected
1089		 * directory.
1090		 */
1091		if (dir == mnt_root || unlikely(IS_ROOT(dir)))
1092			break;
1093
1094		parent_dentry = dget_parent(dir);
1095		dput(dir);
1096		dir = parent_dentry;
1097	}
1098	dput(dir);
1099	return ret;
1100}
1101
1102/**
1103 * current_check_refer_path - Check if a rename or link action is allowed
1104 *
1105 * @old_dentry: File or directory requested to be moved or linked.
1106 * @new_dir: Destination parent directory.
1107 * @new_dentry: Destination file or directory.
1108 * @removable: Sets to true if it is a rename operation.
1109 * @exchange: Sets to true if it is a rename operation with RENAME_EXCHANGE.
1110 *
1111 * Because of its unprivileged constraints, Landlock relies on file hierarchies
1112 * (and not only inodes) to tie access rights to files.  Being able to link or
1113 * rename a file hierarchy brings some challenges.  Indeed, moving or linking a
1114 * file (i.e. creating a new reference to an inode) can have an impact on the
1115 * actions allowed for a set of files if it would change its parent directory
1116 * (i.e. reparenting).
1117 *
1118 * To avoid trivial access right bypasses, Landlock first checks if the file or
1119 * directory requested to be moved would gain new access rights inherited from
1120 * its new hierarchy.  Before returning any error, Landlock then checks that
1121 * the parent source hierarchy and the destination hierarchy would allow the
1122 * link or rename action.  If it is not the case, an error with EACCES is
1123 * returned to inform user space that there is no way to remove or create the
1124 * requested source file type.  If it should be allowed but the new inherited
1125 * access rights would be greater than the source access rights, then the
1126 * kernel returns an error with EXDEV.  Prioritizing EACCES over EXDEV enables
1127 * user space to abort the whole operation if there is no way to do it, or to
1128 * manually copy the source to the destination if this remains allowed, e.g.
1129 * because file creation is allowed on the destination directory but not direct
1130 * linking.
1131 *
1132 * To achieve this goal, the kernel needs to compare two file hierarchies: the
1133 * one identifying the source file or directory (including itself), and the
1134 * destination one.  This can be seen as a multilayer partial ordering problem.
1135 * The kernel walks through these paths and collects in a matrix the access
1136 * rights that are denied per layer.  These matrices are then compared to see
1137 * if the destination one has more (or the same) restrictions as the source
1138 * one.  If this is the case, the requested action will not return EXDEV, which
1139 * doesn't mean the action is allowed.  The parent hierarchy of the source
1140 * (i.e. parent directory), and the destination hierarchy must also be checked
1141 * to verify that they explicitly allow such action (i.e.  referencing,
1142 * creation and potentially removal rights).  The kernel implementation is then
1143 * required to rely on potentially four matrices of access rights: one for the
1144 * source file or directory (i.e. the child), a potentially other one for the
1145 * other source/destination (in case of RENAME_EXCHANGE), one for the source
1146 * parent hierarchy and a last one for the destination hierarchy.  These
1147 * ephemeral matrices take some space on the stack, which limits the number of
1148 * layers to a deemed reasonable number: 16.
1149 *
1150 * Returns:
1151 * - 0 if access is allowed;
1152 * - -EXDEV if @old_dentry would inherit new access rights from @new_dir;
1153 * - -EACCES if file removal or creation is denied.
1154 */
1155static int current_check_refer_path(struct dentry *const old_dentry,
1156				    const struct path *const new_dir,
1157				    struct dentry *const new_dentry,
1158				    const bool removable, const bool exchange)
1159{
1160	const struct landlock_cred_security *const subject =
1161		landlock_get_applicable_subject(current_cred(), any_fs, NULL);
1162	bool allow_parent1, allow_parent2;
1163	access_mask_t access_request_parent1, access_request_parent2;
1164	struct path mnt_dir;
1165	struct dentry *old_parent;
1166	layer_mask_t layer_masks_parent1[LANDLOCK_NUM_ACCESS_FS] = {},
1167		     layer_masks_parent2[LANDLOCK_NUM_ACCESS_FS] = {};
1168	struct landlock_request request1 = {}, request2 = {};
1169
1170	if (!subject)
1171		return 0;
1172
1173	if (unlikely(d_is_negative(old_dentry)))
1174		return -ENOENT;
1175	if (exchange) {
1176		if (unlikely(d_is_negative(new_dentry)))
1177			return -ENOENT;
1178		access_request_parent1 =
1179			get_mode_access(d_backing_inode(new_dentry)->i_mode);
1180	} else {
1181		access_request_parent1 = 0;
1182	}
1183	access_request_parent2 =
1184		get_mode_access(d_backing_inode(old_dentry)->i_mode);
1185	if (removable) {
1186		access_request_parent1 |= maybe_remove(old_dentry);
1187		access_request_parent2 |= maybe_remove(new_dentry);
1188	}
1189
1190	/* The mount points are the same for old and new paths, cf. EXDEV. */
1191	if (old_dentry->d_parent == new_dir->dentry) {
1192		/*
1193		 * The LANDLOCK_ACCESS_FS_REFER access right is not required
1194		 * for same-directory referer (i.e. no reparenting).
1195		 */
1196		access_request_parent1 = landlock_init_layer_masks(
1197			subject->domain,
1198			access_request_parent1 | access_request_parent2,
1199			&layer_masks_parent1, LANDLOCK_KEY_INODE);
1200		if (is_access_to_paths_allowed(subject->domain, new_dir,
1201					       access_request_parent1,
1202					       &layer_masks_parent1, &request1,
1203					       NULL, 0, NULL, NULL, NULL))
1204			return 0;
1205
1206		landlock_log_denial(subject, &request1);
1207		return -EACCES;
1208	}
1209
1210	access_request_parent1 |= LANDLOCK_ACCESS_FS_REFER;
1211	access_request_parent2 |= LANDLOCK_ACCESS_FS_REFER;
1212
1213	/* Saves the common mount point. */
1214	mnt_dir.mnt = new_dir->mnt;
1215	mnt_dir.dentry = new_dir->mnt->mnt_root;
1216
1217	/*
1218	 * old_dentry may be the root of the common mount point and
1219	 * !IS_ROOT(old_dentry) at the same time (e.g. with open_tree() and
1220	 * OPEN_TREE_CLONE).  We do not need to call dget(old_parent) because
1221	 * we keep a reference to old_dentry.
1222	 */
1223	old_parent = (old_dentry == mnt_dir.dentry) ? old_dentry :
1224						      old_dentry->d_parent;
1225
1226	/* new_dir->dentry is equal to new_dentry->d_parent */
1227	allow_parent1 = collect_domain_accesses(subject->domain, mnt_dir.dentry,
1228						old_parent,
1229						&layer_masks_parent1);
1230	allow_parent2 = collect_domain_accesses(subject->domain, mnt_dir.dentry,
1231						new_dir->dentry,
1232						&layer_masks_parent2);
1233
1234	if (allow_parent1 && allow_parent2)
1235		return 0;
1236
1237	/*
1238	 * To be able to compare source and destination domain access rights,
1239	 * take into account the @old_dentry access rights aggregated with its
1240	 * parent access rights.  This will be useful to compare with the
1241	 * destination parent access rights.
1242	 */
1243	if (is_access_to_paths_allowed(
1244		    subject->domain, &mnt_dir, access_request_parent1,
1245		    &layer_masks_parent1, &request1, old_dentry,
1246		    access_request_parent2, &layer_masks_parent2, &request2,
1247		    exchange ? new_dentry : NULL))
1248		return 0;
1249
1250	if (request1.access) {
1251		request1.audit.u.path.dentry = old_parent;
1252		landlock_log_denial(subject, &request1);
1253	}
1254	if (request2.access) {
1255		request2.audit.u.path.dentry = new_dir->dentry;
1256		landlock_log_denial(subject, &request2);
1257	}
1258
1259	/*
1260	 * This prioritizes EACCES over EXDEV for all actions, including
1261	 * renames with RENAME_EXCHANGE.
1262	 */
1263	if (likely(is_eacces(&layer_masks_parent1, access_request_parent1) ||
1264		   is_eacces(&layer_masks_parent2, access_request_parent2)))
1265		return -EACCES;
1266
1267	/*
1268	 * Gracefully forbids reparenting if the destination directory
1269	 * hierarchy is not a superset of restrictions of the source directory
1270	 * hierarchy, or if LANDLOCK_ACCESS_FS_REFER is not allowed by the
1271	 * source or the destination.
1272	 */
1273	return -EXDEV;
1274}
1275
1276/* Inode hooks */
1277
1278static void hook_inode_free_security_rcu(void *inode_security)
1279{
1280	struct landlock_inode_security *inode_sec;
1281
1282	/*
1283	 * All inodes must already have been untied from their object by
1284	 * release_inode() or hook_sb_delete().
1285	 */
1286	inode_sec = inode_security + landlock_blob_sizes.lbs_inode;
1287	WARN_ON_ONCE(inode_sec->object);
1288}
1289
1290/* Super-block hooks */
1291
1292/*
1293 * Release the inodes used in a security policy.
1294 *
1295 * Cf. fsnotify_unmount_inodes() and evict_inodes()
1296 */
1297static void hook_sb_delete(struct super_block *const sb)
1298{
1299	struct inode *inode, *prev_inode = NULL;
1300
1301	if (!landlock_initialized)
1302		return;
1303
1304	spin_lock(&sb->s_inode_list_lock);
1305	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1306		struct landlock_object *object;
1307
1308		/* Only handles referenced inodes. */
1309		if (!icount_read(inode))
1310			continue;
1311
1312		/*
1313		 * Protects against concurrent modification of inode (e.g.
1314		 * from get_inode_object()).
1315		 */
1316		spin_lock(&inode->i_lock);
1317		/*
1318		 * Checks I_FREEING and I_WILL_FREE  to protect against a race
1319		 * condition when release_inode() just called iput(), which
1320		 * could lead to a NULL dereference of inode->security or a
1321		 * second call to iput() for the same Landlock object.  Also
1322		 * checks I_NEW because such inode cannot be tied to an object.
1323		 */
1324		if (inode_state_read(inode) &
1325		    (I_FREEING | I_WILL_FREE | I_NEW)) {
1326			spin_unlock(&inode->i_lock);
1327			continue;
1328		}
1329
1330		rcu_read_lock();
1331		object = rcu_dereference(landlock_inode(inode)->object);
1332		if (!object) {
1333			rcu_read_unlock();
1334			spin_unlock(&inode->i_lock);
1335			continue;
1336		}
1337		/* Keeps a reference to this inode until the next loop walk. */
1338		__iget(inode);
1339		spin_unlock(&inode->i_lock);
1340
1341		/*
1342		 * If there is no concurrent release_inode() ongoing, then we
1343		 * are in charge of calling iput() on this inode, otherwise we
1344		 * will just wait for it to finish.
1345		 */
1346		spin_lock(&object->lock);
1347		if (object->underobj == inode) {
1348			object->underobj = NULL;
1349			spin_unlock(&object->lock);
1350			rcu_read_unlock();
1351
1352			/*
1353			 * Because object->underobj was not NULL,
1354			 * release_inode() and get_inode_object() guarantee
1355			 * that it is safe to reset
1356			 * landlock_inode(inode)->object while it is not NULL.
1357			 * It is therefore not necessary to lock inode->i_lock.
1358			 */
1359			rcu_assign_pointer(landlock_inode(inode)->object, NULL);
1360			/*
1361			 * At this point, we own the ihold() reference that was
1362			 * originally set up by get_inode_object() and the
1363			 * __iget() reference that we just set in this loop
1364			 * walk.  Therefore there are at least two references
1365			 * on the inode.
1366			 */
1367			iput_not_last(inode);
1368		} else {
1369			spin_unlock(&object->lock);
1370			rcu_read_unlock();
1371		}
1372
1373		if (prev_inode) {
1374			/*
1375			 * At this point, we still own the __iget() reference
1376			 * that we just set in this loop walk.  Therefore we
1377			 * can drop the list lock and know that the inode won't
1378			 * disappear from under us until the next loop walk.
1379			 */
1380			spin_unlock(&sb->s_inode_list_lock);
1381			/*
1382			 * We can now actually put the inode reference from the
1383			 * previous loop walk, which is not needed anymore.
1384			 */
1385			iput(prev_inode);
1386			cond_resched();
1387			spin_lock(&sb->s_inode_list_lock);
1388		}
1389		prev_inode = inode;
1390	}
1391	spin_unlock(&sb->s_inode_list_lock);
1392
1393	/* Puts the inode reference from the last loop walk, if any. */
1394	if (prev_inode)
1395		iput(prev_inode);
1396	/* Waits for pending iput() in release_inode(). */
1397	wait_var_event(&landlock_superblock(sb)->inode_refs,
1398		       !atomic_long_read(&landlock_superblock(sb)->inode_refs));
1399}
1400
1401static void
1402log_fs_change_topology_path(const struct landlock_cred_security *const subject,
1403			    size_t handle_layer, const struct path *const path)
1404{
1405	landlock_log_denial(subject, &(struct landlock_request) {
1406		.type = LANDLOCK_REQUEST_FS_CHANGE_TOPOLOGY,
1407		.audit = {
1408			.type = LSM_AUDIT_DATA_PATH,
1409			.u.path = *path,
1410		},
1411		.layer_plus_one = handle_layer + 1,
1412	});
1413}
1414
1415static void log_fs_change_topology_dentry(
1416	const struct landlock_cred_security *const subject, size_t handle_layer,
1417	struct dentry *const dentry)
1418{
1419	landlock_log_denial(subject, &(struct landlock_request) {
1420		.type = LANDLOCK_REQUEST_FS_CHANGE_TOPOLOGY,
1421		.audit = {
1422			.type = LSM_AUDIT_DATA_DENTRY,
1423			.u.dentry = dentry,
1424		},
1425		.layer_plus_one = handle_layer + 1,
1426	});
1427}
1428
1429/*
1430 * Because a Landlock security policy is defined according to the filesystem
1431 * topology (i.e. the mount namespace), changing it may grant access to files
1432 * not previously allowed.
1433 *
1434 * To make it simple, deny any filesystem topology modification by landlocked
1435 * processes.  Non-landlocked processes may still change the namespace of a
1436 * landlocked process, but this kind of threat must be handled by a system-wide
1437 * access-control security policy.
1438 *
1439 * This could be lifted in the future if Landlock can safely handle mount
1440 * namespace updates requested by a landlocked process.  Indeed, we could
1441 * update the current domain (which is currently read-only) by taking into
1442 * account the accesses of the source and the destination of a new mount point.
1443 * However, it would also require to make all the child domains dynamically
1444 * inherit these new constraints.  Anyway, for backward compatibility reasons,
1445 * a dedicated user space option would be required (e.g. as a ruleset flag).
1446 */
1447static int hook_sb_mount(const char *const dev_name,
1448			 const struct path *const path, const char *const type,
1449			 const unsigned long flags, void *const data)
1450{
1451	size_t handle_layer;
1452	const struct landlock_cred_security *const subject =
1453		landlock_get_applicable_subject(current_cred(), any_fs,
1454						&handle_layer);
1455
1456	if (!subject)
1457		return 0;
1458
1459	log_fs_change_topology_path(subject, handle_layer, path);
1460	return -EPERM;
1461}
1462
1463static int hook_move_mount(const struct path *const from_path,
1464			   const struct path *const to_path)
1465{
1466	size_t handle_layer;
1467	const struct landlock_cred_security *const subject =
1468		landlock_get_applicable_subject(current_cred(), any_fs,
1469						&handle_layer);
1470
1471	if (!subject)
1472		return 0;
1473
1474	log_fs_change_topology_path(subject, handle_layer, to_path);
1475	return -EPERM;
1476}
1477
1478/*
1479 * Removing a mount point may reveal a previously hidden file hierarchy, which
1480 * may then grant access to files, which may have previously been forbidden.
1481 */
1482static int hook_sb_umount(struct vfsmount *const mnt, const int flags)
1483{
1484	size_t handle_layer;
1485	const struct landlock_cred_security *const subject =
1486		landlock_get_applicable_subject(current_cred(), any_fs,
1487						&handle_layer);
1488
1489	if (!subject)
1490		return 0;
1491
1492	log_fs_change_topology_dentry(subject, handle_layer, mnt->mnt_root);
1493	return -EPERM;
1494}
1495
1496static int hook_sb_remount(struct super_block *const sb, void *const mnt_opts)
1497{
1498	size_t handle_layer;
1499	const struct landlock_cred_security *const subject =
1500		landlock_get_applicable_subject(current_cred(), any_fs,
1501						&handle_layer);
1502
1503	if (!subject)
1504		return 0;
1505
1506	log_fs_change_topology_dentry(subject, handle_layer, sb->s_root);
1507	return -EPERM;
1508}
1509
1510/*
1511 * pivot_root(2), like mount(2), changes the current mount namespace.  It must
1512 * then be forbidden for a landlocked process.
1513 *
1514 * However, chroot(2) may be allowed because it only changes the relative root
1515 * directory of the current process.  Moreover, it can be used to restrict the
1516 * view of the filesystem.
1517 */
1518static int hook_sb_pivotroot(const struct path *const old_path,
1519			     const struct path *const new_path)
1520{
1521	size_t handle_layer;
1522	const struct landlock_cred_security *const subject =
1523		landlock_get_applicable_subject(current_cred(), any_fs,
1524						&handle_layer);
1525
1526	if (!subject)
1527		return 0;
1528
1529	log_fs_change_topology_path(subject, handle_layer, new_path);
1530	return -EPERM;
1531}
1532
1533/* Path hooks */
1534
1535static int hook_path_link(struct dentry *const old_dentry,
1536			  const struct path *const new_dir,
1537			  struct dentry *const new_dentry)
1538{
1539	return current_check_refer_path(old_dentry, new_dir, new_dentry, false,
1540					false);
1541}
1542
1543static int hook_path_rename(const struct path *const old_dir,
1544			    struct dentry *const old_dentry,
1545			    const struct path *const new_dir,
1546			    struct dentry *const new_dentry,
1547			    const unsigned int flags)
1548{
1549	/* old_dir refers to old_dentry->d_parent and new_dir->mnt */
1550	return current_check_refer_path(old_dentry, new_dir, new_dentry, true,
1551					!!(flags & RENAME_EXCHANGE));
1552}
1553
1554static int hook_path_mkdir(const struct path *const dir,
1555			   struct dentry *const dentry, const umode_t mode)
1556{
1557	return current_check_access_path(dir, LANDLOCK_ACCESS_FS_MAKE_DIR);
1558}
1559
1560static int hook_path_mknod(const struct path *const dir,
1561			   struct dentry *const dentry, const umode_t mode,
1562			   const unsigned int dev)
1563{
1564	return current_check_access_path(dir, get_mode_access(mode));
1565}
1566
1567static int hook_path_symlink(const struct path *const dir,
1568			     struct dentry *const dentry,
1569			     const char *const old_name)
1570{
1571	return current_check_access_path(dir, LANDLOCK_ACCESS_FS_MAKE_SYM);
1572}
1573
1574static int hook_path_unlink(const struct path *const dir,
1575			    struct dentry *const dentry)
1576{
1577	return current_check_access_path(dir, LANDLOCK_ACCESS_FS_REMOVE_FILE);
1578}
1579
1580static int hook_path_rmdir(const struct path *const dir,
1581			   struct dentry *const dentry)
1582{
1583	return current_check_access_path(dir, LANDLOCK_ACCESS_FS_REMOVE_DIR);
1584}
1585
1586static int hook_path_truncate(const struct path *const path)
1587{
1588	return current_check_access_path(path, LANDLOCK_ACCESS_FS_TRUNCATE);
1589}
1590
1591/* File hooks */
1592
1593/**
1594 * get_required_file_open_access - Get access needed to open a file
1595 *
1596 * @file: File being opened.
1597 *
1598 * Returns the access rights that are required for opening the given file,
1599 * depending on the file type and open mode.
1600 */
1601static access_mask_t
1602get_required_file_open_access(const struct file *const file)
1603{
1604	access_mask_t access = 0;
1605
1606	if (file->f_mode & FMODE_READ) {
1607		/* A directory can only be opened in read mode. */
1608		if (S_ISDIR(file_inode(file)->i_mode))
1609			return LANDLOCK_ACCESS_FS_READ_DIR;
1610		access = LANDLOCK_ACCESS_FS_READ_FILE;
1611	}
1612	if (file->f_mode & FMODE_WRITE)
1613		access |= LANDLOCK_ACCESS_FS_WRITE_FILE;
1614	/* __FMODE_EXEC is indeed part of f_flags, not f_mode. */
1615	if (file->f_flags & __FMODE_EXEC)
1616		access |= LANDLOCK_ACCESS_FS_EXECUTE;
1617	return access;
1618}
1619
1620static int hook_file_alloc_security(struct file *const file)
1621{
1622	/*
1623	 * Grants all access rights, even if most of them are not checked later
1624	 * on. It is more consistent.
1625	 *
1626	 * Notably, file descriptors for regular files can also be acquired
1627	 * without going through the file_open hook, for example when using
1628	 * memfd_create(2).
1629	 */
1630	landlock_file(file)->allowed_access = LANDLOCK_MASK_ACCESS_FS;
1631	return 0;
1632}
1633
1634static bool is_device(const struct file *const file)
1635{
1636	const struct inode *inode = file_inode(file);
1637
1638	return S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode);
1639}
1640
1641static int hook_file_open(struct file *const file)
1642{
1643	layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_FS] = {};
1644	access_mask_t open_access_request, full_access_request, allowed_access,
1645		optional_access;
1646	const struct landlock_cred_security *const subject =
1647		landlock_get_applicable_subject(file->f_cred, any_fs, NULL);
1648	struct landlock_request request = {};
1649
1650	if (!subject)
1651		return 0;
1652
1653	/*
1654	 * Because a file may be opened with O_PATH, get_required_file_open_access()
1655	 * may return 0.  This case will be handled with a future Landlock
1656	 * evolution.
1657	 */
1658	open_access_request = get_required_file_open_access(file);
1659
1660	/*
1661	 * We look up more access than what we immediately need for open(), so
1662	 * that we can later authorize operations on opened files.
1663	 */
1664	optional_access = LANDLOCK_ACCESS_FS_TRUNCATE;
1665	if (is_device(file))
1666		optional_access |= LANDLOCK_ACCESS_FS_IOCTL_DEV;
1667
1668	full_access_request = open_access_request | optional_access;
1669
1670	if (is_access_to_paths_allowed(
1671		    subject->domain, &file->f_path,
1672		    landlock_init_layer_masks(subject->domain,
1673					      full_access_request, &layer_masks,
1674					      LANDLOCK_KEY_INODE),
1675		    &layer_masks, &request, NULL, 0, NULL, NULL, NULL)) {
1676		allowed_access = full_access_request;
1677	} else {
1678		unsigned long access_bit;
1679		const unsigned long access_req = full_access_request;
1680
1681		/*
1682		 * Calculate the actual allowed access rights from layer_masks.
1683		 * Add each access right to allowed_access which has not been
1684		 * vetoed by any layer.
1685		 */
1686		allowed_access = 0;
1687		for_each_set_bit(access_bit, &access_req,
1688				 ARRAY_SIZE(layer_masks)) {
1689			if (!layer_masks[access_bit])
1690				allowed_access |= BIT_ULL(access_bit);
1691		}
1692	}
1693
1694	/*
1695	 * For operations on already opened files (i.e. ftruncate()), it is the
1696	 * access rights at the time of open() which decide whether the
1697	 * operation is permitted. Therefore, we record the relevant subset of
1698	 * file access rights in the opened struct file.
1699	 */
1700	landlock_file(file)->allowed_access = allowed_access;
1701#ifdef CONFIG_AUDIT
1702	landlock_file(file)->deny_masks = landlock_get_deny_masks(
1703		_LANDLOCK_ACCESS_FS_OPTIONAL, optional_access, &layer_masks,
1704		ARRAY_SIZE(layer_masks));
1705#endif /* CONFIG_AUDIT */
1706
1707	if ((open_access_request & allowed_access) == open_access_request)
1708		return 0;
1709
1710	/* Sets access to reflect the actual request. */
1711	request.access = open_access_request;
1712	landlock_log_denial(subject, &request);
1713	return -EACCES;
1714}
1715
1716static int hook_file_truncate(struct file *const file)
1717{
1718	/*
1719	 * Allows truncation if the truncate right was available at the time of
1720	 * opening the file, to get a consistent access check as for read, write
1721	 * and execute operations.
1722	 *
1723	 * Note: For checks done based on the file's Landlock allowed access, we
1724	 * enforce them independently of whether the current thread is in a
1725	 * Landlock domain, so that open files passed between independent
1726	 * processes retain their behaviour.
1727	 */
1728	if (landlock_file(file)->allowed_access & LANDLOCK_ACCESS_FS_TRUNCATE)
1729		return 0;
1730
1731	landlock_log_denial(landlock_cred(file->f_cred), &(struct landlock_request) {
1732		.type = LANDLOCK_REQUEST_FS_ACCESS,
1733		.audit = {
1734			.type = LSM_AUDIT_DATA_FILE,
1735			.u.file = file,
1736		},
1737		.all_existing_optional_access = _LANDLOCK_ACCESS_FS_OPTIONAL,
1738		.access = LANDLOCK_ACCESS_FS_TRUNCATE,
1739#ifdef CONFIG_AUDIT
1740		.deny_masks = landlock_file(file)->deny_masks,
1741#endif /* CONFIG_AUDIT */
1742	});
1743	return -EACCES;
1744}
1745
1746static int hook_file_ioctl_common(const struct file *const file,
1747				  const unsigned int cmd, const bool is_compat)
1748{
1749	access_mask_t allowed_access = landlock_file(file)->allowed_access;
1750
1751	/*
1752	 * It is the access rights at the time of opening the file which
1753	 * determine whether IOCTL can be used on the opened file later.
1754	 *
1755	 * The access right is attached to the opened file in hook_file_open().
1756	 */
1757	if (allowed_access & LANDLOCK_ACCESS_FS_IOCTL_DEV)
1758		return 0;
1759
1760	if (!is_device(file))
1761		return 0;
1762
1763	if (unlikely(is_compat) ? is_masked_device_ioctl_compat(cmd) :
1764				  is_masked_device_ioctl(cmd))
1765		return 0;
1766
1767	landlock_log_denial(landlock_cred(file->f_cred), &(struct landlock_request) {
1768		.type = LANDLOCK_REQUEST_FS_ACCESS,
1769		.audit = {
1770			.type = LSM_AUDIT_DATA_IOCTL_OP,
1771			.u.op = &(struct lsm_ioctlop_audit) {
1772				.path = file->f_path,
1773				.cmd = cmd,
1774			},
1775		},
1776		.all_existing_optional_access = _LANDLOCK_ACCESS_FS_OPTIONAL,
1777		.access = LANDLOCK_ACCESS_FS_IOCTL_DEV,
1778#ifdef CONFIG_AUDIT
1779		.deny_masks = landlock_file(file)->deny_masks,
1780#endif /* CONFIG_AUDIT */
1781	});
1782	return -EACCES;
1783}
1784
1785static int hook_file_ioctl(struct file *file, unsigned int cmd,
1786			   unsigned long arg)
1787{
1788	return hook_file_ioctl_common(file, cmd, false);
1789}
1790
1791static int hook_file_ioctl_compat(struct file *file, unsigned int cmd,
1792				  unsigned long arg)
1793{
1794	return hook_file_ioctl_common(file, cmd, true);
1795}
1796
1797/*
1798 * Always allow sending signals between threads of the same process.  This
1799 * ensures consistency with hook_task_kill().
1800 */
1801static bool control_current_fowner(struct fown_struct *const fown)
1802{
1803	struct task_struct *p;
1804
1805	/*
1806	 * Lock already held by __f_setown(), see commit 26f204380a3c ("fs: Fix
1807	 * file_set_fowner LSM hook inconsistencies").
1808	 */
1809	lockdep_assert_held(&fown->lock);
1810
1811	/*
1812	 * Some callers (e.g. fcntl_dirnotify) may not be in an RCU read-side
1813	 * critical section.
1814	 */
1815	guard(rcu)();
1816	p = pid_task(fown->pid, fown->pid_type);
1817	if (!p)
1818		return true;
1819
1820	return !same_thread_group(p, current);
1821}
1822
1823static void hook_file_set_fowner(struct file *file)
1824{
1825	struct landlock_ruleset *prev_dom;
1826	struct landlock_cred_security fown_subject = {};
1827	size_t fown_layer = 0;
1828
1829	if (control_current_fowner(file_f_owner(file))) {
1830		static const struct access_masks signal_scope = {
1831			.scope = LANDLOCK_SCOPE_SIGNAL,
1832		};
1833		const struct landlock_cred_security *new_subject =
1834			landlock_get_applicable_subject(
1835				current_cred(), signal_scope, &fown_layer);
1836		if (new_subject) {
1837			landlock_get_ruleset(new_subject->domain);
1838			fown_subject = *new_subject;
1839		}
1840	}
1841
1842	prev_dom = landlock_file(file)->fown_subject.domain;
1843	landlock_file(file)->fown_subject = fown_subject;
1844#ifdef CONFIG_AUDIT
1845	landlock_file(file)->fown_layer = fown_layer;
1846#endif /* CONFIG_AUDIT*/
1847
1848	/* May be called in an RCU read-side critical section. */
1849	landlock_put_ruleset_deferred(prev_dom);
1850}
1851
1852static void hook_file_free_security(struct file *file)
1853{
1854	landlock_put_ruleset_deferred(landlock_file(file)->fown_subject.domain);
1855}
1856
1857static struct security_hook_list landlock_hooks[] __ro_after_init = {
1858	LSM_HOOK_INIT(inode_free_security_rcu, hook_inode_free_security_rcu),
1859
1860	LSM_HOOK_INIT(sb_delete, hook_sb_delete),
1861	LSM_HOOK_INIT(sb_mount, hook_sb_mount),
1862	LSM_HOOK_INIT(move_mount, hook_move_mount),
1863	LSM_HOOK_INIT(sb_umount, hook_sb_umount),
1864	LSM_HOOK_INIT(sb_remount, hook_sb_remount),
1865	LSM_HOOK_INIT(sb_pivotroot, hook_sb_pivotroot),
1866
1867	LSM_HOOK_INIT(path_link, hook_path_link),
1868	LSM_HOOK_INIT(path_rename, hook_path_rename),
1869	LSM_HOOK_INIT(path_mkdir, hook_path_mkdir),
1870	LSM_HOOK_INIT(path_mknod, hook_path_mknod),
1871	LSM_HOOK_INIT(path_symlink, hook_path_symlink),
1872	LSM_HOOK_INIT(path_unlink, hook_path_unlink),
1873	LSM_HOOK_INIT(path_rmdir, hook_path_rmdir),
1874	LSM_HOOK_INIT(path_truncate, hook_path_truncate),
1875
1876	LSM_HOOK_INIT(file_alloc_security, hook_file_alloc_security),
1877	LSM_HOOK_INIT(file_open, hook_file_open),
1878	LSM_HOOK_INIT(file_truncate, hook_file_truncate),
1879	LSM_HOOK_INIT(file_ioctl, hook_file_ioctl),
1880	LSM_HOOK_INIT(file_ioctl_compat, hook_file_ioctl_compat),
1881	LSM_HOOK_INIT(file_set_fowner, hook_file_set_fowner),
1882	LSM_HOOK_INIT(file_free_security, hook_file_free_security),
1883};
1884
1885__init void landlock_add_fs_hooks(void)
1886{
1887	security_add_hooks(landlock_hooks, ARRAY_SIZE(landlock_hooks),
1888			   &landlock_lsmid);
1889}
1890
1891#ifdef CONFIG_SECURITY_LANDLOCK_KUNIT_TEST
1892
1893/* clang-format off */
1894static struct kunit_case test_cases[] = {
1895	KUNIT_CASE(test_no_more_access),
1896	KUNIT_CASE(test_scope_to_request_with_exec_none),
1897	KUNIT_CASE(test_scope_to_request_with_exec_some),
1898	KUNIT_CASE(test_scope_to_request_without_access),
1899	KUNIT_CASE(test_is_eacces_with_none),
1900	KUNIT_CASE(test_is_eacces_with_refer),
1901	KUNIT_CASE(test_is_eacces_with_write),
1902	{}
1903};
1904/* clang-format on */
1905
1906static struct kunit_suite test_suite = {
1907	.name = "landlock_fs",
1908	.test_cases = test_cases,
1909};
1910
1911kunit_test_suite(test_suite);
1912
1913#endif /* CONFIG_SECURITY_LANDLOCK_KUNIT_TEST */