xfs: Fail remount with noattr2 on a v5 xfs with v4 enabled kernel.

Bug: When we compile the kernel with CONFIG_XFS_SUPPORT_V4=y, remount with "-o remount,noattr2" on a v5 XFS does not fail explicitly.

Reproduction:
mkfs.xfs -f /dev/loop0
mount /dev/loop0 /mnt/scratch
mount -o remount,noattr2 /dev/loop0 /mnt/scratch

However, with CONFIG_XFS_SUPPORT_V4=n, the remount correctly fails explicitly. This is because the way the following 2 functions are defined:

static inline bool xfs_has_attr2 (struct xfs_mount *mp)

{
	return !IS_ENABLED(CONFIG_XFS_SUPPORT_V4) ||
		(mp->m_features & XFS_FEAT_ATTR2);
}
static inline bool xfs_has_noattr2 (const struct xfs_mount *mp)
{
	return mp->m_features & XFS_FEAT_NOATTR2;
}

xfs_has_attr2() returns true when CONFIG_XFS_SUPPORT_V4=n and hence, the following if condition in xfs_fs_validate_params() succeeds and returns -EINVAL:

/*
* We have not read the superblock at this point, so only the attr2
* mount option can set the attr2 feature by this stage. */

if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
	xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
	return -EINVAL;
}

With CONFIG_XFS_SUPPORT_V4=y, xfs_has_attr2() always return false and hence no error is returned.

Fix: Check if the existing mount has crc enabled(i.e, of type v5 and has attr2 enabled) and the
remount has noattr2, if yes, return -EINVAL.

I have tested xfs/{189,539} in fstests with v4 and v5 XFS with both CONFIG_XFS_SUPPORT_V4=y/n and they both behave as expected.

This patch also fixes remount from noattr2 -> attr2 (on a v4 xfs).

Related discussion in [1]

[1] https://lore.kernel.org/all/[email protected]/

Signed-off-by Nirjhar Roy (IBM) <[email protected]>

Reviewed-by Christoph Hellwig <[email protected]>

---
 fs/xfs/xfs_super.c | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

fs/xfs/xfs_super.c +0 -0

index b2dd0c0bf509..606a95ac816f 100644

fs/xfs/xfs_super.c +20 -1

+		}
+		mp->m_features &= ~XFS_FEAT_ATTR2;
+		mp->m_features |= XFS_FEAT_NOATTR2;
+	} else if (xfs_has_attr2(new_mp)) {
+		/* noattr2 -> attr2 */
+		mp->m_features &= ~XFS_FEAT_NOATTR2;
+		mp->m_features |= XFS_FEAT_ATTR2;
+	}
+
 	/* inode32 -> inode64 */
 	if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
 		mp->m_features &= ~XFS_FEAT_SMALL_INUMS;

@@ -2126,6 +2141,17 @@ xfs_fs_reconfigure(
 		mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
 	}
 
+	/*
+	 * Now that mp has been modified according to the remount options,
+	 * we do a final option validation with xfs_finish_flags()
+	 * just like it is done during mount. We cannot use
+	 * xfs_finish_flags()on new_mp as it contains only the user
+	 * given options.
+	 */
+	error = xfs_finish_flags(mp);
+	if (error)
+		return error;
+
 	/* ro -> rw */
 	if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
 		error = xfs_remount_rw(mp);
--

2.43.5

Quoted 65 lines · Nirjhar Roy (IBM)

Bug: When we compile the kernel with CONFIG_XFS_SUPPORT_V4=y, remount with "-o remount,noattr2" on a v5 XFS does not fail explicitly.

Reproduction:
mkfs.xfs -f /dev/loop0
mount /dev/loop0 /mnt/scratch
mount -o remount,noattr2 /dev/loop0 /mnt/scratch

However, with CONFIG_XFS_SUPPORT_V4=n, the remount correctly fails explicitly. This is because the way the following 2 functions are defined:

static inline bool xfs_has_attr2 (struct xfs_mount *mp)

{
	return !IS_ENABLED(CONFIG_XFS_SUPPORT_V4) ||
		(mp->m_features & XFS_FEAT_ATTR2);
}
static inline bool xfs_has_noattr2 (const struct xfs_mount *mp)
{
	return mp->m_features & XFS_FEAT_NOATTR2;
}

xfs_has_attr2() returns true when CONFIG_XFS_SUPPORT_V4=n and hence, the following if condition in xfs_fs_validate_params() succeeds and returns -EINVAL:

/*
* We have not read the superblock at this point, so only the attr2
* mount option can set the attr2 feature by this stage. */

if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
	xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
	return -EINVAL;
}

With CONFIG_XFS_SUPPORT_V4=y, xfs_has_attr2() always return false and hence no error is returned.

Fix: Check if the existing mount has crc enabled(i.e, of type v5 and has attr2 enabled) and the
remount has noattr2, if yes, return -EINVAL.

I have tested xfs/{189,539} in fstests with v4 and v5 XFS with both CONFIG_XFS_SUPPORT_V4=y/n and they both behave as expected.

This patch also fixes remount from noattr2 -> attr2 (on a v4 xfs).

Related discussion in [1]

[1] https://lore.kernel.org/all/[email protected]/

Signed-off-by Nirjhar Roy (IBM) <[email protected]>

Reviewed-by Christoph Hellwig <[email protected]>

---
  fs/xfs/xfs_super.c | 26 ++++++++++++++++++++++++++
  1 file changed, 26 insertions(+)

fs/xfs/xfs_super.c +0 -0

index b2dd0c0bf509..606a95ac816f 100644

fs/xfs/xfs_super.c xfs_fs_reconfigure( · line 2122

Quoted 9 lines

fs/xfs/xfs_super.c +6 -0

@@ -2114,6 +2114,21 @@ xfs_fs_reconfigure(
  	if (error)
  		return error;
 
+	/* attr2 -> noattr2 */
+	if (xfs_has_noattr2(new_mp)) {
+		if (xfs_has_crc(mp)) {
+			xfs_warn(mp,
+			"attr2 is always enabled for a V5 filesystem - can't be changed.");
+			return -EINVAL;

This looks good to me now:

Reviewed-by Carlos Maiolino <[email protected]>

I still wish hch's opinion here though before merging it. Giving his was the first RwB, I want to make sure he still keeps his RwB with the above change.

FWIW, for a next patch, there is no need to copy ext4 list for a code change that is totally unrelated to ext4. This just generates unnecessary extra traffic.

Quoted 36 lines

fs/xfs/xfs_super.c +20 -1

+		}
+		mp->m_features &= ~XFS_FEAT_ATTR2;
+		mp->m_features |= XFS_FEAT_NOATTR2;
+	} else if (xfs_has_attr2(new_mp)) {
+		/* noattr2 -> attr2 */
+		mp->m_features &= ~XFS_FEAT_NOATTR2;
+		mp->m_features |= XFS_FEAT_ATTR2;
+	}
+
  	/* inode32 -> inode64 */
  	if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
  		mp->m_features &= ~XFS_FEAT_SMALL_INUMS;

@@ -2126,6 +2141,17 @@ xfs_fs_reconfigure(
  		mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
  	}
 
+	/*
+	 * Now that mp has been modified according to the remount options,
+	 * we do a final option validation with xfs_finish_flags()
+	 * just like it is done during mount. We cannot use
+	 * xfs_finish_flags()on new_mp as it contains only the user
+	 * given options.
+	 */
+	error = xfs_finish_flags(mp);
+	if (error)
+		return error;
+
  	/* ro -> rw */
  	if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
  		error = xfs_remount_rw(mp);
--

2.43.5

This still looks good.

One nit:

fs/xfs/xfs_super.c xfs_fs_reconfigure( · line 2149

Quoted 6 lines

fs/xfs/xfs_super.c +6 -0

+	/*
+	 * Now that mp has been modified according to the remount options,
+	 * we do a final option validation with xfs_finish_flags()
+	 * just like it is done during mount. We cannot use
+	 * xfs_finish_flags()on new_mp as it contains only the user
+	 * given options.

This could use slightly better formatting:

/*
* Now that mp has been modified according to the remount options, we
* do a final option validation with xfs_finish_flags() just like it is
* done during mount. We cannot use xfs_finish_flags() on new_mp as it
* contains only the user given options.
*/