xfs: turn NLINK feature on by default
mkfs has turned on the XFS_SB_VERSION_NLINKBIT feature bit by
default since November 2007. It's about time we simply made the
kernel code turn it on by default and so always convert v1 inodes to
v2 inodes when reading them in from disk or allocating them. This
This removes needless version checks and modification when bumping
link counts on inodes, and will take code out of a few common code
paths.
text data bss dec hex filename
783251 100867 616 884734 d7ffe fs/xfs/xfs.o.orig
782664 100867 616 884147 d7db3 fs/xfs/xfs.o.patched
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
diff --git a/fs/xfs/xfs_fsops.c b/fs/xfs/xfs_fsops.c
index 02fb943..d49c67a 100644
--- a/fs/xfs/xfs_fsops.c
+++ b/fs/xfs/xfs_fsops.c
@@ -74,11 +74,9 @@
}
if (new_version >= 3) {
geo->version = XFS_FSOP_GEOM_VERSION;
- geo->flags =
+ geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
(xfs_sb_version_hasattr(&mp->m_sb) ?
XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
- (xfs_sb_version_hasnlink(&mp->m_sb) ?
- XFS_FSOP_GEOM_FLAGS_NLINK : 0) |
(xfs_sb_version_hasquota(&mp->m_sb) ?
XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
(xfs_sb_version_hasalign(&mp->m_sb) ?
diff --git a/fs/xfs/xfs_ialloc.c b/fs/xfs/xfs_ialloc.c
index 8f711db..449fa7b 100644
--- a/fs/xfs/xfs_ialloc.c
+++ b/fs/xfs/xfs_ialloc.c
@@ -220,10 +220,8 @@
if (tp)
xfs_icreate_log(tp, agno, agbno, mp->m_ialloc_inos,
mp->m_sb.sb_inodesize, length, gen);
- } else if (xfs_sb_version_hasnlink(&mp->m_sb))
+ } else
version = 2;
- else
- version = 1;
for (j = 0; j < nbufs; j++) {
/*
diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c
index 768087b..9731977 100644
--- a/fs/xfs/xfs_inode.c
+++ b/fs/xfs/xfs_inode.c
@@ -682,6 +682,14 @@
return error;
ASSERT(ip != NULL);
+ /*
+ * We always convert v1 inodes to v2 now - we only support filesystems
+ * with >= v2 inode capability, so there is no reason for ever leaving
+ * an inode in v1 format.
+ */
+ if (ip->i_d.di_version == 1)
+ ip->i_d.di_version = 2;
+
ip->i_d.di_mode = mode;
ip->i_d.di_onlink = 0;
ip->i_d.di_nlink = nlink;
@@ -691,27 +699,6 @@
xfs_set_projid(ip, prid);
memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
- /*
- * If the superblock version is up to where we support new format
- * inodes and this is currently an old format inode, then change
- * the inode version number now. This way we only do the conversion
- * here rather than here and in the flush/logging code.
- */
- if (xfs_sb_version_hasnlink(&mp->m_sb) &&
- ip->i_d.di_version == 1) {
- ip->i_d.di_version = 2;
- /*
- * We've already zeroed the old link count, the projid field,
- * and the pad field.
- */
- }
-
- /*
- * Project ids won't be stored on disk if we are using a version 1 inode.
- */
- if ((prid != 0) && (ip->i_d.di_version == 1))
- xfs_bump_ino_vers2(tp, ip);
-
if (pip && XFS_INHERIT_GID(pip)) {
ip->i_d.di_gid = pip->i_d.di_gid;
if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
@@ -1073,40 +1060,6 @@
}
/*
- * This gets called when the inode's version needs to be changed from 1 to 2.
- * Currently this happens when the nlink field overflows the old 16-bit value
- * or when chproj is called to change the project for the first time.
- * As a side effect the superblock version will also get rev'd
- * to contain the NLINK bit.
- */
-void
-xfs_bump_ino_vers2(
- xfs_trans_t *tp,
- xfs_inode_t *ip)
-{
- xfs_mount_t *mp;
-
- ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
- ASSERT(ip->i_d.di_version == 1);
-
- ip->i_d.di_version = 2;
- ip->i_d.di_onlink = 0;
- memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
- mp = tp->t_mountp;
- if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
- spin_lock(&mp->m_sb_lock);
- if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
- xfs_sb_version_addnlink(&mp->m_sb);
- spin_unlock(&mp->m_sb_lock);
- xfs_mod_sb(tp, XFS_SB_VERSIONNUM);
- } else {
- spin_unlock(&mp->m_sb_lock);
- }
- }
- /* Caller must log the inode */
-}
-
-/*
* Increment the link count on an inode & log the change.
*/
int
@@ -1116,22 +1069,10 @@
{
xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
+ ASSERT(ip->i_d.di_version > 1);
ASSERT(ip->i_d.di_nlink > 0 || (VFS_I(ip)->i_state & I_LINKABLE));
ip->i_d.di_nlink++;
inc_nlink(VFS_I(ip));
- if ((ip->i_d.di_version == 1) &&
- (ip->i_d.di_nlink > XFS_MAXLINK_1)) {
- /*
- * The inode has increased its number of links beyond
- * what can fit in an old format inode. It now needs
- * to be converted to a version 2 inode with a 32 bit
- * link count. If this is the first inode in the file
- * system to do this, then we need to bump the superblock
- * version number as well.
- */
- xfs_bump_ino_vers2(tp, ip);
- }
-
xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
return 0;
}
@@ -3258,6 +3199,7 @@
ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
ASSERT(iip != NULL && iip->ili_fields != 0);
+ ASSERT(ip->i_d.di_version > 1);
/* set *dip = inode's place in the buffer */
dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
@@ -3318,7 +3260,7 @@
}
/*
- * Inode item log recovery for v1/v2 inodes are dependent on the
+ * Inode item log recovery for v2 inodes are dependent on the
* di_flushiter count for correct sequencing. We bump the flush
* iteration count so we can detect flushes which postdate a log record
* during recovery. This is redundant as we now log every change and
@@ -3341,37 +3283,6 @@
if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
ip->i_d.di_flushiter = 0;
- /*
- * If this is really an old format inode and the superblock version
- * has not been updated to support only new format inodes, then
- * convert back to the old inode format. If the superblock version
- * has been updated, then make the conversion permanent.
- */
- ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
- if (ip->i_d.di_version == 1) {
- if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
- /*
- * Convert it back.
- */
- ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
- dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
- } else {
- /*
- * The superblock version has already been bumped,
- * so just make the conversion to the new inode
- * format permanent.
- */
- ip->i_d.di_version = 2;
- dip->di_version = 2;
- ip->i_d.di_onlink = 0;
- dip->di_onlink = 0;
- memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
- memset(&(dip->di_pad[0]), 0,
- sizeof(dip->di_pad));
- ASSERT(xfs_get_projid(ip) == 0);
- }
- }
-
xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
if (XFS_IFORK_Q(ip))
xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
diff --git a/fs/xfs/xfs_inode.h b/fs/xfs/xfs_inode.h
index f2fcde5..d331a0e 100644
--- a/fs/xfs/xfs_inode.h
+++ b/fs/xfs/xfs_inode.h
@@ -379,7 +379,6 @@
struct xfs_inode **, int *);
int xfs_droplink(struct xfs_trans *, struct xfs_inode *);
int xfs_bumplink(struct xfs_trans *, struct xfs_inode *);
-void xfs_bump_ino_vers2(struct xfs_trans *, struct xfs_inode *);
/* from xfs_file.c */
int xfs_zero_eof(struct xfs_inode *, xfs_off_t, xfs_fsize_t);
diff --git a/fs/xfs/xfs_inode_buf.c b/fs/xfs/xfs_inode_buf.c
index 24e9939..cb35ae4 100644
--- a/fs/xfs/xfs_inode_buf.c
+++ b/fs/xfs/xfs_inode_buf.c
@@ -437,17 +437,16 @@
}
/*
- * The inode format changed when we moved the link count and
- * made it 32 bits long. If this is an old format inode,
- * convert it in memory to look like a new one. If it gets
- * flushed to disk we will convert back before flushing or
- * logging it. We zero out the new projid field and the old link
- * count field. We'll handle clearing the pad field (the remains
- * of the old uuid field) when we actually convert the inode to
- * the new format. We don't change the version number so that we
- * can distinguish this from a real new format inode.
+ * Automatically convert version 1 inode formats in memory to version 2
+ * inode format. If the inode is modified, it will get logged and
+ * rewritten as a version 2 inode. We can do this because we set the
+ * superblock feature bit for v2 inodes unconditionally during mount
+ * and it means the reast of the code can assume the inode version is 2
+ * or higher.
*/
if (ip->i_d.di_version == 1) {
+ ip->i_d.di_version = 2;
+ memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
ip->i_d.di_nlink = ip->i_d.di_onlink;
ip->i_d.di_onlink = 0;
xfs_set_projid(ip, 0);
diff --git a/fs/xfs/xfs_inode_item.c b/fs/xfs/xfs_inode_item.c
index 686889b..a640137 100644
--- a/fs/xfs/xfs_inode_item.c
+++ b/fs/xfs/xfs_inode_item.c
@@ -145,34 +145,6 @@
xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
}
-/*
- * If this is a v1 format inode, then we need to log it as such. This means
- * that we have to copy the link count from the new field to the old. We
- * don't have to worry about the new fields, because nothing trusts them as
- * long as the old inode version number is there.
- */
-STATIC void
-xfs_inode_item_format_v1_inode(
- struct xfs_inode *ip)
-{
- if (!xfs_sb_version_hasnlink(&ip->i_mount->m_sb)) {
- /*
- * Convert it back.
- */
- ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
- ip->i_d.di_onlink = ip->i_d.di_nlink;
- } else {
- /*
- * The superblock version has already been bumped,
- * so just make the conversion to the new inode
- * format permanent.
- */
- ip->i_d.di_version = 2;
- ip->i_d.di_onlink = 0;
- memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
- }
-}
-
STATIC void
xfs_inode_item_format_data_fork(
struct xfs_inode_log_item *iip,
@@ -370,6 +342,8 @@
struct xfs_inode_log_format *ilf;
struct xfs_log_iovec *vecp = NULL;
+ ASSERT(ip->i_d.di_version > 1);
+
ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
ilf->ilf_type = XFS_LI_INODE;
ilf->ilf_ino = ip->i_ino;
@@ -380,8 +354,6 @@
ilf->ilf_size = 2; /* format + core */
xlog_finish_iovec(lv, vecp, sizeof(struct xfs_inode_log_format));
- if (ip->i_d.di_version == 1)
- xfs_inode_item_format_v1_inode(ip);
xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_ICORE,
&ip->i_d,
xfs_icdinode_size(ip->i_d.di_version));
diff --git a/fs/xfs/xfs_ioctl.c b/fs/xfs/xfs_ioctl.c
index 0b18776..f920d42 100644
--- a/fs/xfs/xfs_ioctl.c
+++ b/fs/xfs/xfs_ioctl.c
@@ -1227,15 +1227,8 @@
olddquot = xfs_qm_vop_chown(tp, ip,
&ip->i_pdquot, pdqp);
}
+ ASSERT(ip->i_d.di_version > 1);
xfs_set_projid(ip, fa->fsx_projid);
-
- /*
- * We may have to rev the inode as well as
- * the superblock version number since projids didn't
- * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
- */
- if (ip->i_d.di_version == 1)
- xfs_bump_ino_vers2(tp, ip);
}
}
diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c
index 944f3d9..3f09782 100644
--- a/fs/xfs/xfs_mount.c
+++ b/fs/xfs/xfs_mount.c
@@ -697,6 +697,12 @@
mp->m_update_flags |= XFS_SB_VERSIONNUM;
}
+ /* always use v2 inodes by default now */
+ if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
+ mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
+ mp->m_update_flags |= XFS_SB_VERSIONNUM;
+ }
+
/*
* Check if sb_agblocks is aligned at stripe boundary
* If sb_agblocks is NOT aligned turn off m_dalign since
diff --git a/fs/xfs/xfs_sb.h b/fs/xfs/xfs_sb.h
index a6a76f4..a2826cf 100644
--- a/fs/xfs/xfs_sb.h
+++ b/fs/xfs/xfs_sb.h
@@ -376,16 +376,6 @@
sbp->sb_versionnum |= XFS_SB_VERSION_ATTRBIT;
}
-static inline bool xfs_sb_version_hasnlink(struct xfs_sb *sbp)
-{
- return (sbp->sb_versionnum & XFS_SB_VERSION_NLINKBIT);
-}
-
-static inline void xfs_sb_version_addnlink(struct xfs_sb *sbp)
-{
- sbp->sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
-}
-
static inline bool xfs_sb_version_hasquota(struct xfs_sb *sbp)
{
return (sbp->sb_versionnum & XFS_SB_VERSION_QUOTABIT);