[PATCH v6 0/6] landlock: Restrict whiteout object creation

["Günther Noack" <gnoack@google.com>]

Hello!

As discussed in [1], the renameat2() syscall's RENAME_WHITEOUT flag allows
the creation of chardev directory entries with major=minor=0 as "whiteout
objects" in the location of the rename source file [2].

This functionality is available even without having any OverlayFS mounted
and can be invoked with the regular renameat2(2) syscall [3].

In V1 [5], it was discussed that whiteout objects are not the same as character
devices, and should therefore be treated differently.  After considering a new
access right in V2 and V3, we eventually settled on guarding it with the
existing LANDLOCK_ACCESS_FS_MAKE_REG access right and introducing a new erratum
[6].

V6 addresses the remaining review feedback: the V4 feedback on the selftests
in patches 3/5 and 4/5, and the V5 remarks on the get_dentry_access() helper.

Motivation
==========

The RENAME_WHITEOUT flag side-steps all of the existing Landlock access
rights, which are designed to restrict the creation of directory entries.
It is desirable to restrict that.

This patch set fixes that by adding a check in Landlock's path_rename and
path_mknod hooks.  Where we previously treated whiteout creation like the
creation of a character device, we now guard it with
LANDLOCK_ACCESS_FS_MAKE_REG and add an erratum to probe for that.


[1] https://lore.kernel.org/all/adUBCQXrt7kmgqJT@google.com/
[2] https://docs.kernel.org/filesystems/overlayfs.html#whiteouts-and-opaque-directories
[3] https://man7.org/linux/man-pages/man2/renameat2.2.html#DESCRIPTION
[4] https://codesearch.debian.net/search?q=rename.*RENAME_WHITEOUT&literal=0
[5] https://lore.kernel.org/all/20260411090944.3131168-2-gnoack@google.com/
[6] https://lore.kernel.org/all/20260720.chow9ohYie5b@digikod.net/

Changelog
=========

v6:
 - Smaller code changes:
   - get_dentry_access(): Drop the incorrect __attribute_const__ and use a
     local inode variable
 - Fix a LANDLOCK_ACCESS_FS_MAKE_REG typo in the commit message of patch 2/5
 - Tests, addressing the remaining v4 feedback on patch 3/5:
   - Add rename_whiteout_allowed: RENAME_WHITEOUT works when MAKE_REG is
     granted, and the whiteout object shows up in the source location
   - Add rename_whiteout_reparenting: the MAKE_REG right for the created
     whiteout object is required in the source directory of the rename
   - Add rename_whiteout_exchange: RENAME_EXCHANGE of an existing whiteout
     object requires MAKE_REG in the directory which the whiteout moves into
   - Add audit test make_whiteout: denied whiteout creation is logged with
     blockers=fs.make_reg instead of fs.make_char
 - Tests, addressing the v4 feedback on patch 4/5:
   - Rework the OverlayFS rename test to rename a FIFO which originates
     from the lower layer, instead of preparing it in the upper layer with
     unlink() and mknod().  The unlink() had already created the whiteout
     in the upper layer before Landlock was enforced
   - Check that the whiteout object in the upper layer did not exist before
     the rename
 - Verified that the new tests fail on a kernel which lacks the fix from
   patch 2/5

v5:
 - Erratum: Clarify that this is about mknod(2) only
 - Add Cc stable@vger.kernel.org and Depends-on lines
 - Smaller code changes
   - Use dev_t where appropriate
   - Introduce get_dentry_access() helper
   - Use get_mode_access(S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV)
     to calculate the required access right for the created whiteout object
 - Tests:
   - rename_whiteout_denied: move a FIFO
     - otherwise, the test succeeded without the change
   - make_whiteout: drop set_cap(CAP_MKNOD) (not needed)
 - This is an intermediary state: It incorporates feedback to 2/5 and some 3/5
   feedback from v4, but not the 4/5 feedback yet.
 - https://lore.kernel.org/all/20260731154353.1790268-1-gnoack@google.com/

v4:
 - Guard it with LANDLOCK_ACCESS_FS_MAKE_REG as discussed.
 - Selftests and documentation.
 - https://lore.kernel.org/all/20260724161004.2360749-1-gnoack@google.com/

v3:
 - Do LANDLOCK_ACCESS_FS_MAKE_WHITEOUT check as part of
   current_check_refer_path().
 - https://lore.kernel.org/all/20260610092318.3868884-1-gnoack@google.com/

v2:
 - Introduce LANDLOCK_ACCESS_FS_MAKE_WHITEOUT access right
   and guard it with that.
 - Bump ABI version
 - https://lore.kernel.org/all/20260513160552.4022649-1-gnoack@google.com/

v1:
 - initial version
   https://lore.kernel.org/all/20260411090944.3131168-2-gnoack@google.com/


Günther Noack (6):
  selftests/landlock: Use an actual chardev for MAKE_CHAR audit test
  landlock: Require LANDLOCK_ACCESS_FS_MAKE_REG for whiteout creation
  selftests/landlock: Add tests for whiteout object creation
  selftests/landlock: Add audit test for whiteout object creation
  selftests/landlock: Test whiteout object behaviour in OverlayFS
    renames
  landlock: Link the erratum documentation for whiteout objects

 Documentation/userspace-api/landlock.rst   |   3 +
 include/uapi/linux/landlock.h              |   1 +
 security/landlock/errata/abi-1.h           |  23 ++
 security/landlock/fs.c                     |  41 +++-
 tools/testing/selftests/landlock/fs_test.c | 253 ++++++++++++++++++++-
 5 files changed, 309 insertions(+), 12 deletions(-)

Range-diff against v5:
1:  d1a8a84d92e0 = 1:  49a6fa0831e8 selftests/landlock: Use an actual chardev for MAKE_CHAR audit test
2:  9fe2edf17413 ! 2:  a29c7919a412 landlock: Require LANDLOCK_ACCESS_FS_MAKE_REG for whiteout creation
    @@ Commit message
         For the mknod(2) case, introduce a Landlock erratum.  The creation of
         whiteout objects through mknod(2) was previously guarded using
         LANDLOCK_ACCESS_FS_MAKE_CHAR, and it is now guarded using
    -    LANDLOCK_ACCESS_MAKE_REG.
    +    LANDLOCK_ACCESS_FS_MAKE_REG.
     
         For the renameat2(2) case, fix a bug: Before this commit, renameat2(2)
         with RENAME_WHITEOUT would create a directory entry even when all
    @@ security/landlock/fs.c: static __attribute_const__ access_mask_t get_mode_access
      	}
      }
      
    -+static __attribute_const__ access_mask_t
    -+get_dentry_access(const struct dentry *const dentry)
    ++static access_mask_t get_dentry_access(const struct dentry *const dentry)
     +{
    -+	return get_mode_access(d_backing_inode(dentry)->i_mode,
    -+			       d_backing_inode(dentry)->i_rdev);
    ++	const struct inode *const inode = d_backing_inode(dentry);
    ++
    ++	return get_mode_access(inode->i_mode, inode->i_rdev);
     +}
     +
      static access_mask_t maybe_remove(const struct dentry *const dentry)
3:  d23ea62b3325 ! 3:  0711d9e709fb selftests/landlock: Add tests for whiteout object creation
    @@ tools/testing/selftests/landlock/fs_test.c: TEST_F_FORK(layout1, rename_file)
     +				TMP_DIR "/s3d1/s3d2/s3d3/f2", RENAME_WHITEOUT));
     +	EXPECT_EQ(EACCES, errno);
     +}
    ++
    ++static bool is_whiteout(const char *const path)
    ++{
    ++	struct stat st;
    ++
    ++	if (stat(path, &st) == -1)
    ++		return false;
    ++
    ++	return S_ISCHR(st.st_mode) && st.st_rdev == makedev(0, 0);
    ++}
    ++
    ++static bool is_fifo(const char *const path)
    ++{
    ++	struct stat st;
    ++
    ++	return stat(path, &st) == 0 && S_ISFIFO(st.st_mode);
    ++}
    ++
    ++TEST_F_FORK(layout1, rename_whiteout_allowed)
    ++{
    ++	const struct rule rules[] = {
    ++		{
    ++			.path = dir_s3d3,
    ++			.access = LANDLOCK_ACCESS_FS_MAKE_REG,
    ++		},
    ++		{},
    ++	};
    ++
    ++	/* The affected file is a FIFO. */
    ++	ASSERT_EQ(0, unlink(file1_s3d3));
    ++	ASSERT_EQ(0, mknod(file1_s3d3, S_IFIFO | 0600, 0));
    ++
    ++	/* Allow MAKE_REG below dir_s3d3. */
    ++	enforce_fs(_metadata, LANDLOCK_ACCESS_FS_MAKE_REG, rules);
    ++
    ++	/*
    ++	 * Rename a file with RENAME_WHITEOUT within the same directory.
    ++	 * Allowed, because MAKE_REG is granted for the whiteout object which
    ++	 * gets created in the source location.
    ++	 */
    ++	EXPECT_EQ(0, renameat2(AT_FDCWD, file1_s3d3, AT_FDCWD,
    ++			       TMP_DIR "/s3d1/s3d2/s3d3/f2", RENAME_WHITEOUT));
    ++
    ++	/* A whiteout object took the place of the moved FIFO. */
    ++	EXPECT_TRUE(is_whiteout(file1_s3d3));
    ++	EXPECT_TRUE(is_fifo(TMP_DIR "/s3d1/s3d2/s3d3/f2"));
    ++}
    ++
    ++TEST_F_FORK(layout1, rename_whiteout_reparenting)
    ++{
    ++	const struct rule rules[] = {
    ++		{
    ++			.path = dir_s3d2,
    ++			.access = LANDLOCK_ACCESS_FS_REFER,
    ++		},
    ++		{
    ++			.path = dir_s3d3,
    ++			.access = LANDLOCK_ACCESS_FS_MAKE_REG,
    ++		},
    ++		{},
    ++	};
    ++
    ++	/* The moved files are FIFOs. */
    ++	ASSERT_EQ(0, unlink(file1_s3d3));
    ++	ASSERT_EQ(0, mknod(file1_s3d3, S_IFIFO | 0600, 0));
    ++	ASSERT_EQ(0, unlink(file1_s3d4));
    ++	ASSERT_EQ(0, mknod(file1_s3d4, S_IFIFO | 0600, 0));
    ++
    ++	/* Allow REFER below dir_s3d2, but MAKE_REG only below dir_s3d3. */
    ++	enforce_fs(_metadata,
    ++		   LANDLOCK_ACCESS_FS_MAKE_REG | LANDLOCK_ACCESS_FS_REFER,
    ++		   rules);
    ++
    ++	/*
    ++	 * The whiteout object is created in the source directory: Moving the
    ++	 * FIFO out of dir_s3d4 is denied because MAKE_REG is not granted
    ++	 * there, even though it is granted in the destination directory
    ++	 * dir_s3d3.
    ++	 */
    ++	EXPECT_EQ(-1, renameat2(AT_FDCWD, file1_s3d4, AT_FDCWD,
    ++				TMP_DIR "/s3d1/s3d2/s3d3/f2", RENAME_WHITEOUT));
    ++	EXPECT_EQ(EACCES, errno);
    ++
    ++	/*
    ++	 * Moving the FIFO out of dir_s3d3 is allowed, because MAKE_REG is
    ++	 * granted there for the created whiteout object.
    ++	 */
    ++	EXPECT_EQ(0, renameat2(AT_FDCWD, file1_s3d3, AT_FDCWD,
    ++			       TMP_DIR "/s3d1/s3d2/s3d4/f2", RENAME_WHITEOUT));
    ++
    ++	/* A whiteout object took the place of the moved FIFO. */
    ++	EXPECT_TRUE(is_whiteout(file1_s3d3));
    ++	EXPECT_TRUE(is_fifo(TMP_DIR "/s3d1/s3d2/s3d4/f2"));
    ++}
    ++
    ++TEST_F_FORK(layout1, rename_whiteout_exchange)
    ++{
    ++	const char *const whiteout_s3d3 = TMP_DIR "/s3d1/s3d2/s3d3/f2";
    ++	const struct rule rules[] = {
    ++		{
    ++			.path = dir_s3d2,
    ++			.access = LANDLOCK_ACCESS_FS_REFER,
    ++		},
    ++		{
    ++			.path = dir_s3d3,
    ++			.access = LANDLOCK_ACCESS_FS_MAKE_REG,
    ++		},
    ++		{},
    ++	};
    ++
    ++	/* The exchanged files are FIFOs and an existing whiteout object. */
    ++	ASSERT_EQ(0, unlink(file1_s3d3));
    ++	ASSERT_EQ(0, mknod(file1_s3d3, S_IFIFO | 0600, 0));
    ++	ASSERT_EQ(0, mknod(whiteout_s3d3, S_IFCHR | 0600, makedev(0, 0)));
    ++	ASSERT_EQ(0, unlink(file1_s3d4));
    ++	ASSERT_EQ(0, mknod(file1_s3d4, S_IFIFO | 0600, 0));
    ++
    ++	/* Allow REFER below dir_s3d2, but MAKE_REG only below dir_s3d3. */
    ++	enforce_fs(_metadata,
    ++		   LANDLOCK_ACCESS_FS_MAKE_REG | LANDLOCK_ACCESS_FS_REFER,
    ++		   rules);
    ++
    ++	/*
    ++	 * With RENAME_EXCHANGE, the whiteout object moves into the source
    ++	 * directory of the rename: Exchanging the FIFO in dir_s3d4 with the
    ++	 * whiteout object is denied because MAKE_REG is not granted in
    ++	 * dir_s3d4, even though it is granted in the whiteout object's own
    ++	 * directory dir_s3d3.
    ++	 */
    ++	EXPECT_EQ(-1, renameat2(AT_FDCWD, file1_s3d4, AT_FDCWD, whiteout_s3d3,
    ++				RENAME_EXCHANGE));
    ++	EXPECT_EQ(EACCES, errno);
    ++
    ++	/*
    ++	 * Exchanging the FIFO in dir_s3d3 with the whiteout object is
    ++	 * allowed, because MAKE_REG is granted in the directory into which
    ++	 * the whiteout object moves.
    ++	 */
    ++	EXPECT_EQ(0, renameat2(AT_FDCWD, file1_s3d3, AT_FDCWD, whiteout_s3d3,
    ++			       RENAME_EXCHANGE));
    ++
    ++	/* The FIFO and the whiteout object swapped places. */
    ++	EXPECT_TRUE(is_whiteout(file1_s3d3));
    ++	EXPECT_TRUE(is_fifo(whiteout_s3d3));
    ++}
     +
      TEST_F_FORK(layout1, rename_dir)
      {
    @@ tools/testing/selftests/landlock/fs_test.c: TEST_F_FORK(layout1, make_char)
      
     +TEST_F_FORK(layout1, make_whiteout)
     +{
    -+	/* Creates a whiteout object (creation guarded by MAKE_REG). */
    ++	/*
    ++	 * Creates a whiteout object (creation guarded by MAKE_REG).
    ++	 *
    ++	 * Contrary to the other character devices, this does not require
    ++	 * CAP_MKNOD, cf. vfs_mknod().
    ++	 */
     +	test_make_file(_metadata, LANDLOCK_ACCESS_FS_MAKE_REG, S_IFCHR,
     +		       makedev(0, 0));
     +}
-:  ------------ > 4:  f61b6ddb635e selftests/landlock: Add audit test for whiteout object creation
4:  10ed1d74636d ! 5:  4862aead03f2 selftests/landlock: Test whiteout object behaviour in OverlayFS renames
    @@ Metadata
      ## Commit message ##
         selftests/landlock: Test whiteout object behaviour in OverlayFS renames
     
    -    Even though OverlayFS uses vfs_rename() with RENAME_WHITEOUT, and even
    -    though RENAME_WHITEOUT requires LANDLOCK_ACCESS_FS_MAKE_REG, a process
    -    that renames non-regular files in an OverlayFS can do so without
    -    having the LANDLOCK_ACCESS_FS_MAKE_REG right in that location.
    +    Even though OverlayFS uses vfs_rename() with RENAME_WHITEOUT on its backing
    +    directories, and even though RENAME_WHITEOUT requires
    +    LANDLOCK_ACCESS_FS_MAKE_REG, a process that renames non-regular files in an
    +    OverlayFS can do so without having the LANDLOCK_ACCESS_FS_MAKE_REG right in
    +    that location.
     
    -    This works, and is supposed to work, because OverlayFS uses the
    -    credentials determined at mount time for the internal vfs_rename()
    -    operation.  The rename happens with the credentials of the user who
    -    mounted the OverlayFS.
    +    This works, and is supposed to work, because the changes to the backing
    +    directories are done by OverlayFS, not by the originator task that did the
    +    original rename() on the OverlayFS mount.  Therefore, the changes done to
    +    backing directories are not subject to the originator task's credentials.
     
         Signed-off-by: Günther Noack <gnoack@google.com>
     
      ## tools/testing/selftests/landlock/fs_test.c ##
    +@@ tools/testing/selftests/landlock/fs_test.c: static bool is_fifo(const char *const path)
    + 	return stat(path, &st) == 0 && S_ISFIFO(st.st_mode);
    + }
    + 
    ++static bool is_missing(const char *const path)
    ++{
    ++	struct stat st;
    ++
    ++	return stat(path, &st) == -1 && errno == ENOENT;
    ++}
    ++
    + TEST_F_FORK(layout1, rename_whiteout_allowed)
    + {
    + 	const struct rule rules[] = {
    +@@ tools/testing/selftests/landlock/fs_test.c: static const char lower_fo1[] = LOWER_DATA "/fo1";
    + static const char lower_do1[] = LOWER_DATA "/do1";
    + static const char lower_do1_fo2[] = LOWER_DATA "/do1/fo2";
    + static const char lower_do1_fl3[] = LOWER_DATA "/do1/fl3";
    ++/* lower_pl1 is a FIFO and is deliberately not in the lists below. */
    ++static const char lower_pl1[] = LOWER_DATA "/pl1";
    + 
    + static const char (*lower_base_files[])[] = {
    + 	&lower_fl1,
    +@@ tools/testing/selftests/landlock/fs_test.c: static const char (*upper_sub_files[])[] = {
    + #define MERGE_BASE TMP_DIR "/merge"
    + #define MERGE_DATA MERGE_BASE "/data"
    + static const char merge_fl1[] = MERGE_DATA "/fl1";
    ++/* merge_pl1 is a FIFO and is deliberately not in the lists below. */
    ++static const char merge_pl1[] = MERGE_DATA "/pl1";
    + static const char merge_dl1[] = MERGE_DATA "/dl1";
    + static const char merge_dl1_fl2[] = MERGE_DATA "/dl1/fl2";
    + static const char merge_fu1[] = MERGE_DATA "/fu1";
    +@@ tools/testing/selftests/landlock/fs_test.c: static const char (*merge_sub_files[])[] = {
    +  * │       │   ├── fl3
    +  * │       │   └── fo2
    +  * │       ├── fl1
    +- * │       └── fo1
    ++ * │       ├── fo1
    ++ * │       └── pl1 [FIFO]
    +  * ├── merge
    +  * │   └── data
    +  * │       ├── dl1
    +@@ tools/testing/selftests/landlock/fs_test.c: static const char (*merge_sub_files[])[] = {
    +  * │       │   └── fu2
    +  * │       ├── fl1
    +  * │       ├── fo1
    +- * │       └── fu1
    ++ * │       ├── fu1
    ++ * │       └── pl1 [FIFO]
    +  * └── upper
    +  *     ├── data
    +  *     │   ├── do1
    +@@ tools/testing/selftests/landlock/fs_test.c: FIXTURE_SETUP(layout2_overlay)
    + 	create_file(_metadata, lower_fo1);
    + 	create_file(_metadata, lower_do1_fo2);
    + 	create_file(_metadata, lower_do1_fl3);
    ++	ASSERT_EQ(0, mknod(lower_pl1, S_IFIFO | 0600, 0));
    + 
    + 	create_directory(_metadata, UPPER_BASE);
    + 	set_cap(_metadata, CAP_SYS_ADMIN);
    +@@ tools/testing/selftests/landlock/fs_test.c: FIXTURE_TEARDOWN_PARENT(layout2_overlay)
    + 	EXPECT_EQ(0, remove_path(lower_fl1));
    + 	EXPECT_EQ(0, remove_path(lower_do1_fo2));
    + 	EXPECT_EQ(0, remove_path(lower_fo1));
    ++	EXPECT_EQ(0, remove_path(lower_pl1));
    + 
    + 	/* umount(LOWER_BASE)) is handled by namespace lifetime. */
    + 	EXPECT_EQ(0, remove_path(LOWER_BASE));
     @@ tools/testing/selftests/landlock/fs_test.c: TEST_F_FORK(layout2_overlay, same_content_different_file)
      	}
      }
      
     +TEST_F_FORK(layout2_overlay, rename_in_overlay_without_make_reg)
     +{
    -+	struct stat st;
    -+	const char *merge_fl1_renamed = MERGE_DATA "/fl1_renamed";
    ++	const char *const merge_pl1_renamed = MERGE_DATA "/pl1_renamed";
     +
     +	if (self->skip_test)
     +		SKIP(return, "overlayfs is not supported (test)");
     +
     +	/*
    -+	 * In this test, merge_fl1 is a FIFO file.  MAKE_REG is restricted, but
    -+	 * MAKE_FIFO is allowed.  Despite MAKE_REG being restricted, the rename
    -+	 * on the OverlayFS works and creates a whiteout file in the underlying
    -+	 * upper file system.
    ++	 * merge_pl1 is a FIFO which only exists in the lower layer.  Before
    ++	 * the rename, the upper layer has no entry under this name.
     +	 */
    -+	ASSERT_EQ(0, unlink(merge_fl1));
    -+	ASSERT_EQ(0, mknod(merge_fl1, S_IFIFO, 0));
    ++	ASSERT_TRUE(is_fifo(merge_pl1));
    ++	ASSERT_TRUE(is_missing(UPPER_DATA "/pl1"));
    ++
    ++	/* MAKE_REG is restricted, but MAKE_FIFO is not. */
     +	enforce_fs(_metadata, LANDLOCK_ACCESS_FS_MAKE_REG, NULL);
     +
     +	/*
    -+	 * Execute a regular file rename within OverlayFS.
    -+	 * merge_fl1 originates from lower layer, so this triggers a copy-up
    -+	 * and creation of a whiteout in the upper layer.
    ++	 * Rename the FIFO through OverlayFS.  merge_pl1 originates from the
    ++	 * lower layer, so this triggers a copy-up and creates the whiteout in
    ++	 * the upper layer to hide the lower layer FIFO file.  Even though
    ++	 * MAKE_REG is restricted, the rename on the OverlayFS works.
     +	 */
    -+	EXPECT_EQ(0, rename(merge_fl1, merge_fl1_renamed));
    ++	EXPECT_EQ(0, rename(merge_pl1, merge_pl1_renamed));
     +
     +	/* Check that the rename worked. */
    -+	EXPECT_EQ(0, stat(merge_fl1_renamed, &st));
    -+	EXPECT_EQ(-1, stat(merge_fl1, &st));
    -+	EXPECT_EQ(ENOENT, errno);
    ++	EXPECT_TRUE(is_fifo(merge_pl1_renamed));
    ++	EXPECT_TRUE(is_missing(merge_pl1));
     +
     +	/*
    -+	 * Check that the whiteout object on the underlying "upper" filesystem
    -+	 * exists after the rename.  This is OK because it was done with the
    -+	 * credentials of the OverlayFS.
    ++	 * Check that the whiteout object was created on the underlying "upper"
    ++	 * filesystem during the rename.  This is OK because the whiteout object
    ++	 * was created by OverlayFS, not by the calling task.
     +	 */
    -+	EXPECT_EQ(0, stat(UPPER_DATA "/fl1", &st));
    -+	EXPECT_TRUE(S_ISCHR(st.st_mode));
    -+	EXPECT_EQ(0, st.st_rdev);
    ++	EXPECT_TRUE(is_whiteout(UPPER_DATA "/pl1"));
     +}
     +
      FIXTURE(layout3_fs)
5:  35651219395c = 6:  4a53acb99752 landlock: Link the erratum documentation for whiteout objects
-- 
2.55.0.699.gb54405d56f-goog