[PATCHSET v3 0/2] xfs_repair: enable upgrading to nrext64

[PATCHSET v3 0/2] xfs_repair: enable upgrading to nrext64

[Darrick J. Wong]

Hi all,

This patchset enables sysadmins to upgrade an existing filesystem to use
large extent counts.

v3: don't allow upgrades for filesystems that are more than 90% full
    because we just dont want to go there
v2: pull in patches from chandan/dave

If you're going to start using this mess, you probably ought to just
pull from my git trees, which are linked below.

This is an extraordinary way to destroy everything.  Enjoy!
Comments and questions are, as always, welcome.

--D

xfsprogs git tree:
https://git.kernel.org/cgit/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=nrext64-upgrade
---
 include/libxfs.h         |    1 
 include/xfs_mount.h      |    1 
 libxfs/init.c            |   24 +++--
 libxfs/libxfs_api_defs.h |    3 +
 man/man8/xfs_admin.8     |    7 +
 repair/globals.c         |    1 
 repair/globals.h         |    1 
 repair/phase2.c          |  234 ++++++++++++++++++++++++++++++++++++++++++++--
 repair/xfs_repair.c      |   11 ++
 9 files changed, 266 insertions(+), 17 deletions(-)

[PATCH 1/2] xfs_repair: check filesystem geometry before allowing upgrades

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

Currently, the two V5 feature upgrades permitted by xfs_repair do not
affect filesystem space usage, so we haven't needed to verify the
geometry.

However, this will change once we start to allow the sysadmin to add new
metadata indexes to existing filesystems.  Add all the infrastructure we
need to ensure that the log will still be large enough, that there's
enough space for metadata space reservations, and the root inode will
still be where we expect it to be after the upgrade.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandan.babu@oracle.com>
[david: Recompute transaction reservation values; Exit with error if upgrade fails]
Signed-off-by: Dave Chinner <david@fromorbit.com>
[djwong: Refuse to upgrade if any part of the fs has < 10% free]
---
 include/libxfs.h         |    1 
 include/xfs_mount.h      |    1 
 libxfs/init.c            |   24 ++++-
 libxfs/libxfs_api_defs.h |    3 +
 repair/phase2.c          |  210 ++++++++++++++++++++++++++++++++++++++++++++--
 5 files changed, 222 insertions(+), 17 deletions(-)


diff --git a/include/libxfs.h b/include/libxfs.h
index 915bf511..7d6e9a33 100644
--- a/include/libxfs.h
+++ b/include/libxfs.h
@@ -77,6 +77,7 @@ struct iomap;
 #include "xfs_refcount_btree.h"
 #include "xfs_refcount.h"
 #include "xfs_btree_staging.h"
+#include "xfs_ag_resv.h"
 
 #ifndef ARRAY_SIZE
 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
diff --git a/include/xfs_mount.h b/include/xfs_mount.h
index ba80aa79..24b1d873 100644
--- a/include/xfs_mount.h
+++ b/include/xfs_mount.h
@@ -259,6 +259,7 @@ __XFS_UNSUPP_OPSTATE(shutdown)
 
 #define LIBXFS_BHASHSIZE(sbp) 		(1<<10)
 
+void libxfs_compute_all_maxlevels(struct xfs_mount *mp);
 struct xfs_mount *libxfs_mount(struct xfs_mount *mp, struct xfs_sb *sb,
 		dev_t dev, dev_t logdev, dev_t rtdev, unsigned int flags);
 int libxfs_flush_mount(struct xfs_mount *mp);
diff --git a/libxfs/init.c b/libxfs/init.c
index a01a41b2..15052696 100644
--- a/libxfs/init.c
+++ b/libxfs/init.c
@@ -728,6 +728,21 @@ xfs_agbtree_compute_maxlevels(
 	mp->m_agbtree_maxlevels = max(levels, mp->m_refc_maxlevels);
 }
 
+/* Compute maximum possible height of all btrees. */
+void
+libxfs_compute_all_maxlevels(
+	struct xfs_mount	*mp)
+{
+	xfs_alloc_compute_maxlevels(mp);
+	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
+	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
+	xfs_ialloc_setup_geometry(mp);
+	xfs_rmapbt_compute_maxlevels(mp);
+	xfs_refcountbt_compute_maxlevels(mp);
+
+	xfs_agbtree_compute_maxlevels(mp);
+}
+
 /*
  * Mount structure initialization, provides a filled-in xfs_mount_t
  * such that the numerous XFS_* macros can be used.  If dev is zero,
@@ -772,14 +787,7 @@ libxfs_mount(
 		mp->m_swidth = sbp->sb_width;
 	}
 
-	xfs_alloc_compute_maxlevels(mp);
-	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
-	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
-	xfs_ialloc_setup_geometry(mp);
-	xfs_rmapbt_compute_maxlevels(mp);
-	xfs_refcountbt_compute_maxlevels(mp);
-
-	xfs_agbtree_compute_maxlevels(mp);
+	libxfs_compute_all_maxlevels(mp);
 
 	/*
 	 * Check that the data (and log if separate) are an ok size.
diff --git a/libxfs/libxfs_api_defs.h b/libxfs/libxfs_api_defs.h
index 370ad8b3..824f2c4d 100644
--- a/libxfs/libxfs_api_defs.h
+++ b/libxfs/libxfs_api_defs.h
@@ -21,6 +21,8 @@
 
 #define xfs_ag_init_headers		libxfs_ag_init_headers
 #define xfs_ag_block_count		libxfs_ag_block_count
+#define xfs_ag_resv_init		libxfs_ag_resv_init
+#define xfs_ag_resv_free		libxfs_ag_resv_free
 
 #define xfs_alloc_ag_max_usable		libxfs_alloc_ag_max_usable
 #define xfs_allocbt_maxlevels_ondisk	libxfs_allocbt_maxlevels_ondisk
@@ -112,6 +114,7 @@
 #define xfs_highbit64			libxfs_highbit64
 #define xfs_ialloc_calc_rootino		libxfs_ialloc_calc_rootino
 #define xfs_iallocbt_maxlevels_ondisk	libxfs_iallocbt_maxlevels_ondisk
+#define xfs_ialloc_read_agi		libxfs_ialloc_read_agi
 #define xfs_idata_realloc		libxfs_idata_realloc
 #define xfs_idestroy_fork		libxfs_idestroy_fork
 #define xfs_iext_lookup_extent		libxfs_iext_lookup_extent
diff --git a/repair/phase2.c b/repair/phase2.c
index 13832701..a0fd1e9c 100644
--- a/repair/phase2.c
+++ b/repair/phase2.c
@@ -133,7 +133,8 @@ zero_log(
 
 static bool
 set_inobtcount(
-	struct xfs_mount	*mp)
+	struct xfs_mount	*mp,
+	struct xfs_sb		*new_sb)
 {
 	if (!xfs_has_crc(mp)) {
 		printf(
@@ -153,14 +154,15 @@ set_inobtcount(
 	}
 
 	printf(_("Adding inode btree counts to filesystem.\n"));
-	mp->m_sb.sb_features_ro_compat |= XFS_SB_FEAT_RO_COMPAT_INOBTCNT;
-	mp->m_sb.sb_features_incompat |= XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR;
+	new_sb->sb_features_ro_compat |= XFS_SB_FEAT_RO_COMPAT_INOBTCNT;
+	new_sb->sb_features_incompat |= XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR;
 	return true;
 }
 
 static bool
 set_bigtime(
-	struct xfs_mount	*mp)
+	struct xfs_mount	*mp,
+	struct xfs_sb		*new_sb)
 {
 	if (!xfs_has_crc(mp)) {
 		printf(
@@ -174,28 +176,218 @@ set_bigtime(
 	}
 
 	printf(_("Adding large timestamp support to filesystem.\n"));
-	mp->m_sb.sb_features_incompat |= (XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR |
-					  XFS_SB_FEAT_INCOMPAT_BIGTIME);
+	new_sb->sb_features_incompat |= (XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR |
+					 XFS_SB_FEAT_INCOMPAT_BIGTIME);
 	return true;
 }
 
+struct check_state {
+	struct xfs_sb		sb;
+	uint64_t		features;
+	bool			finobt_nores;
+};
+
+static inline void
+capture_old_state(
+	struct check_state	*old_state,
+	const struct xfs_mount	*mp)
+{
+	memcpy(&old_state->sb, &mp->m_sb, sizeof(struct xfs_sb));
+	old_state->finobt_nores = mp->m_finobt_nores;
+	old_state->features = mp->m_features;
+}
+
+static inline void
+restore_old_state(
+	struct xfs_mount		*mp,
+	const struct check_state	*old_state)
+{
+	memcpy(&mp->m_sb, &old_state->sb, sizeof(struct xfs_sb));
+	mp->m_finobt_nores = old_state->finobt_nores;
+	mp->m_features = old_state->features;
+	libxfs_compute_all_maxlevels(mp);
+	libxfs_trans_init(mp);
+}
+
+static inline void
+install_new_state(
+	struct xfs_mount	*mp,
+	struct xfs_sb		*new_sb)
+{
+	memcpy(&mp->m_sb, new_sb, sizeof(struct xfs_sb));
+	mp->m_features |= libxfs_sb_version_to_features(new_sb);
+	libxfs_compute_all_maxlevels(mp);
+	libxfs_trans_init(mp);
+}
+
+/*
+ * Make sure we can actually upgrade this (v5) filesystem without running afoul
+ * of root inode or log size requirements that would prevent us from mounting
+ * the filesystem.  If everything checks out, commit the new geometry.
+ */
+static void
+install_new_geometry(
+	struct xfs_mount	*mp,
+	struct xfs_sb		*new_sb)
+{
+	struct check_state	old;
+	struct xfs_perag	*pag;
+	xfs_ino_t		rootino;
+	xfs_agnumber_t		agno;
+	int			min_logblocks;
+	int			error;
+
+	capture_old_state(&old, mp);
+	install_new_state(mp, new_sb);
+
+	/*
+	 * The existing log must be large enough to satisfy the new minimum log
+	 * size requirements.
+	 */
+	min_logblocks = libxfs_log_calc_minimum_size(mp);
+	if (old.sb.sb_logblocks < min_logblocks) {
+		printf(
+	_("Filesystem log too small to upgrade filesystem; need %u blocks, have %u.\n"),
+				min_logblocks, old.sb.sb_logblocks);
+		exit(1);
+	}
+
+	/*
+	 * The root inode must be where xfs_repair will expect it to be with
+	 * the new geometry.
+	 */
+	rootino = libxfs_ialloc_calc_rootino(mp, new_sb->sb_unit);
+	if (old.sb.sb_rootino != rootino) {
+		printf(
+	_("Cannot upgrade filesystem, root inode (%llu) cannot be moved to %llu.\n"),
+				(unsigned long long)old.sb.sb_rootino,
+				(unsigned long long)rootino);
+		exit(1);
+	}
+
+	/* Make sure we have enough space for per-AG reservations. */
+	for_each_perag(mp, agno, pag) {
+		struct xfs_trans	*tp;
+		struct xfs_agf		*agf;
+		struct xfs_buf		*agi_bp, *agf_bp;
+		unsigned int		avail, agblocks;
+
+		/* Put back the old super so that we can read AG headers. */
+		restore_old_state(mp, &old);
+
+		/*
+		 * Create a dummy transaction so that we can load the AGI and
+		 * AGF buffers in memory with the old fs geometry and pin them
+		 * there while we try to make a per-AG reservation with the new
+		 * geometry.
+		 */
+		error = -libxfs_trans_alloc_empty(mp, &tp);
+		if (error)
+			do_error(
+	_("Cannot reserve resources for upgrade check, err=%d.\n"),
+					error);
+
+		error = -libxfs_ialloc_read_agi(mp, tp, pag->pag_agno,
+				&agi_bp);
+		if (error)
+			do_error(
+	_("Cannot read AGI %u for upgrade check, err=%d.\n"),
+					pag->pag_agno, error);
+
+		error = -libxfs_alloc_read_agf(mp, tp, pag->pag_agno, 0,
+				&agf_bp);
+		if (error)
+			do_error(
+	_("Cannot read AGF %u for upgrade check, err=%d.\n"),
+					pag->pag_agno, error);
+		agf = agf_bp->b_addr;
+		agblocks = be32_to_cpu(agf->agf_length);
+
+		/*
+		 * Install the new superblock and try to make a per-AG space
+		 * reservation with the new geometry.  We pinned the AG header
+		 * buffers to the transaction, so we shouldn't hit any
+		 * corruption errors on account of the new geometry.
+		 */
+		install_new_state(mp, new_sb);
+
+		error = -libxfs_ag_resv_init(pag, tp);
+		if (error == ENOSPC) {
+			printf(
+	_("Not enough free space would remain in AG %u for metadata.\n"),
+					pag->pag_agno);
+			exit(1);
+		}
+		if (error)
+			do_error(
+	_("Error %d while checking AG %u space reservation.\n"),
+					error, pag->pag_agno);
+
+		/*
+		 * Would we have at least 10% free space in this AG after
+		 * making per-AG reservations?
+		 */
+		avail = pag->pagf_freeblks + pag->pagf_flcount;
+		avail -= pag->pag_meta_resv.ar_reserved;
+		avail -= pag->pag_rmapbt_resv.ar_asked;
+		if (avail < agblocks / 10) {
+			printf(
+	_("AG %u will be low on space after upgrade.\n"),
+					pag->pag_agno);
+			exit(1);
+		}
+		libxfs_trans_cancel(tp);
+	}
+
+	/*
+	 * Would we have at least 10% free space in the data device after all
+	 * the upgrades?
+	 */
+	if (mp->m_sb.sb_fdblocks < mp->m_sb.sb_dblocks / 10) {
+		printf(_("Filesystem will be low on space after upgrade.\n"));
+		exit(1);
+	}
+
+	/*
+	 * Release the per-AG reservations and mark the per-AG structure as
+	 * uninitialized so that we don't trip over stale cached counters
+	 * after the upgrade/
+	 */
+	for_each_perag(mp, agno, pag) {
+		libxfs_ag_resv_free(pag);
+		pag->pagf_init = 0;
+		pag->pagi_init = 0;
+	}
+
+	/*
+	 * Restore the old state to get everything back to a clean state,
+	 * upgrade the featureset one more time, and recompute the btree max
+	 * levels for this filesystem.
+	 */
+	restore_old_state(mp, &old);
+	install_new_state(mp, new_sb);
+}
+
 /* Perform the user's requested upgrades on filesystem. */
 static void
 upgrade_filesystem(
 	struct xfs_mount	*mp)
 {
+	struct xfs_sb		new_sb;
 	struct xfs_buf		*bp;
 	bool			dirty = false;
 	int			error;
 
+	memcpy(&new_sb, &mp->m_sb, sizeof(struct xfs_sb));
+
 	if (add_inobtcount)
-		dirty |= set_inobtcount(mp);
+		dirty |= set_inobtcount(mp, &new_sb);
 	if (add_bigtime)
-		dirty |= set_bigtime(mp);
+		dirty |= set_bigtime(mp, &new_sb);
 	if (!dirty)
 		return;
 
-	mp->m_features |= libxfs_sb_version_to_features(&mp->m_sb);
+	install_new_geometry(mp, &new_sb);
 	if (no_modify)
 		return;
 

[PATCH 2/2] xfs_repair: Add support for upgrading to large extent counters

[Darrick J. Wong]

From: Chandan Babu R <chandan.babu@oracle.com>

This commit adds support to xfs_repair to allow upgrading an existing
filesystem to support per-inode large extent counters.

Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandan.babu@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
---
 man/man8/xfs_admin.8 |    7 +++++++
 repair/globals.c     |    1 +
 repair/globals.h     |    1 +
 repair/phase2.c      |   24 ++++++++++++++++++++++++
 repair/xfs_repair.c  |   11 +++++++++++
 5 files changed, 44 insertions(+)


diff --git a/man/man8/xfs_admin.8 b/man/man8/xfs_admin.8
index ad28e0f6..4794d677 100644
--- a/man/man8/xfs_admin.8
+++ b/man/man8/xfs_admin.8
@@ -149,6 +149,13 @@ Upgrade a filesystem to support larger timestamps up to the year 2486.
 The filesystem cannot be downgraded after this feature is enabled.
 Once enabled, the filesystem will not be mountable by older kernels.
 This feature was added to Linux 5.10.
+.TP 0.4i
+.B nrext64
+Upgrade a filesystem to support large per-inode extent counters. The maximum
+data fork extent count will be 2^48 - 1, while the maximum attribute fork
+extent count will be 2^32 - 1. The filesystem cannot be downgraded after this
+feature is enabled. Once enabled, the filesystem will not be mountable by
+older kernels.  This feature was added to Linux 5.19.
 .RE
 .TP
 .BI \-U " uuid"
diff --git a/repair/globals.c b/repair/globals.c
index f8d4f1e4..c4084985 100644
--- a/repair/globals.c
+++ b/repair/globals.c
@@ -51,6 +51,7 @@ int	lazy_count;		/* What to set if to if converting */
 bool	features_changed;	/* did we change superblock feature bits? */
 bool	add_inobtcount;		/* add inode btree counts to AGI */
 bool	add_bigtime;		/* add support for timestamps up to 2486 */
+bool	add_nrext64;
 
 /* misc status variables */
 
diff --git a/repair/globals.h b/repair/globals.h
index 0f98bd2b..b65e4a2d 100644
--- a/repair/globals.h
+++ b/repair/globals.h
@@ -92,6 +92,7 @@ extern int	lazy_count;		/* What to set if to if converting */
 extern bool	features_changed;	/* did we change superblock feature bits? */
 extern bool	add_inobtcount;		/* add inode btree counts to AGI */
 extern bool	add_bigtime;		/* add support for timestamps up to 2486 */
+extern bool	add_nrext64;
 
 /* misc status variables */
 
diff --git a/repair/phase2.c b/repair/phase2.c
index a0fd1e9c..51567b19 100644
--- a/repair/phase2.c
+++ b/repair/phase2.c
@@ -181,6 +181,28 @@ set_bigtime(
 	return true;
 }
 
+static bool
+set_nrext64(
+	struct xfs_mount	*mp,
+	struct xfs_sb		*new_sb)
+{
+	if (!xfs_has_crc(mp)) {
+		printf(
+	_("Nrext64 only supported on V5 filesystems.\n"));
+		exit(0);
+	}
+
+	if (xfs_has_large_extent_counts(mp)) {
+		printf(_("Filesystem already supports nrext64.\n"));
+		exit(0);
+	}
+
+	printf(_("Adding nrext64 to filesystem.\n"));
+	new_sb->sb_features_incompat |= XFS_SB_FEAT_INCOMPAT_NREXT64;
+	new_sb->sb_features_incompat |= XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR;
+	return true;
+}
+
 struct check_state {
 	struct xfs_sb		sb;
 	uint64_t		features;
@@ -384,6 +406,8 @@ upgrade_filesystem(
 		dirty |= set_inobtcount(mp, &new_sb);
 	if (add_bigtime)
 		dirty |= set_bigtime(mp, &new_sb);
+	if (add_nrext64)
+		dirty |= set_nrext64(mp, &new_sb);
 	if (!dirty)
 		return;
 
diff --git a/repair/xfs_repair.c b/repair/xfs_repair.c
index d08b0cec..c94671d8 100644
--- a/repair/xfs_repair.c
+++ b/repair/xfs_repair.c
@@ -67,6 +67,7 @@ enum c_opt_nums {
 	CONVERT_LAZY_COUNT = 0,
 	CONVERT_INOBTCOUNT,
 	CONVERT_BIGTIME,
+	CONVERT_NREXT64,
 	C_MAX_OPTS,
 };
 
@@ -74,6 +75,7 @@ static char *c_opts[] = {
 	[CONVERT_LAZY_COUNT]	= "lazycount",
 	[CONVERT_INOBTCOUNT]	= "inobtcount",
 	[CONVERT_BIGTIME]	= "bigtime",
+	[CONVERT_NREXT64]	= "nrext64",
 	[C_MAX_OPTS]		= NULL,
 };
 
@@ -324,6 +326,15 @@ process_args(int argc, char **argv)
 		_("-c bigtime only supports upgrades\n"));
 					add_bigtime = true;
 					break;
+				case CONVERT_NREXT64:
+					if (!val)
+						do_abort(
+		_("-c nrext64 requires a parameter\n"));
+					if (strtol(val, NULL, 0) != 1)
+						do_abort(
+		_("-c nrext64 only supports upgrades\n"));
+					add_nrext64 = true;
+					break;
 				default:
 					unknown('c', val);
 					break;