akpm at linux-foundation.org
2014-Mar-19 21:10 UTC
[Ocfs2-devel] [patch 4/8] ocfs2: call ocfs2_update_inode_fsync_trans when updating any inode
From: "Darrick J. Wong" <darrick.wong at oracle.com>
Subject: ocfs2: call ocfs2_update_inode_fsync_trans when updating any inode
Ensure that ocfs2_update_inode_fsync_trans() is called any time we touch
an inode in a given transaction. This is a follow-on to the previous
patch to reduce lock contention and deadlocking during an fsync operation.
Signed-off-by: Darrick J. Wong <darrick.wong at oracle.com>
Cc: Mark Fasheh <mfasheh at suse.de>
Cc: Joel Becker <jlbec at evilplan.org>
Cc: Wengang <wen.gang.wang at oracle.com>
Cc: Greg Marsden <greg.marsden at oracle.com>
Cc: Srinivas Eeda <srinivas.eeda at oracle.com>
Signed-off-by: Andrew Morton <akpm at linux-foundation.org>
---
fs/ocfs2/acl.c | 1 +
fs/ocfs2/alloc.c | 2 ++
fs/ocfs2/dir.c | 2 ++
fs/ocfs2/file.c | 7 +++++++
fs/ocfs2/move_extents.c | 2 ++
fs/ocfs2/namei.c | 1 +
fs/ocfs2/suballoc.c | 3 ++-
fs/ocfs2/xattr.c | 3 +++
8 files changed, 20 insertions(+), 1 deletion(-)
diff -puN
fs/ocfs2/acl.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
fs/ocfs2/acl.c
---
a/fs/ocfs2/acl.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
+++ a/fs/ocfs2/acl.c
@@ -205,6 +205,7 @@ static int ocfs2_acl_set_mode(struct ino
di->i_mode = cpu_to_le16(inode->i_mode);
di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+ ocfs2_update_inode_fsync_trans(handle, inode, 0);
ocfs2_journal_dirty(handle, di_bh);
diff -puN
fs/ocfs2/alloc.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
fs/ocfs2/alloc.c
---
a/fs/ocfs2/alloc.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
+++ a/fs/ocfs2/alloc.c
@@ -5728,6 +5728,7 @@ int ocfs2_remove_btree_range(struct inod
}
ocfs2_et_update_clusters(et, -len);
+ ocfs2_update_inode_fsync_trans(handle, inode, 1);
ocfs2_journal_dirty(handle, et->et_root_bh);
@@ -7209,6 +7210,7 @@ int ocfs2_truncate_inline(struct inode *
di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
di->i_ctime_nsec = di->i_mtime_nsec =
cpu_to_le32(inode->i_ctime.tv_nsec);
+ ocfs2_update_inode_fsync_trans(handle, inode, 1);
ocfs2_journal_dirty(handle, di_bh);
out_commit:
diff -puN
fs/ocfs2/dir.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
fs/ocfs2/dir.c
---
a/fs/ocfs2/dir.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
+++ a/fs/ocfs2/dir.c
@@ -3006,6 +3006,7 @@ static int ocfs2_expand_inline_dir(struc
di->i_size = cpu_to_le64(sb->s_blocksize);
di->i_ctime = di->i_mtime = cpu_to_le64(dir->i_ctime.tv_sec);
di->i_ctime_nsec = di->i_mtime_nsec =
cpu_to_le32(dir->i_ctime.tv_nsec);
+ ocfs2_update_inode_fsync_trans(handle, dir, 1);
/*
* This should never fail as our extent list is empty and all
@@ -4405,6 +4406,7 @@ static int ocfs2_dx_dir_remove_index(str
di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
spin_unlock(&OCFS2_I(dir)->ip_lock);
di->i_dx_root = cpu_to_le64(0ULL);
+ ocfs2_update_inode_fsync_trans(handle, dir, 1);
ocfs2_journal_dirty(handle, di_bh);
diff -puN
fs/ocfs2/file.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
fs/ocfs2/file.c
---
a/fs/ocfs2/file.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
+++ a/fs/ocfs2/file.c
@@ -286,6 +286,7 @@ int ocfs2_update_inode_atime(struct inod
inode->i_atime = CURRENT_TIME;
di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
+ ocfs2_update_inode_fsync_trans(handle, inode, 0);
ocfs2_journal_dirty(handle, bh);
out_commit:
@@ -335,6 +336,7 @@ int ocfs2_simple_size_update(struct inod
if (ret < 0)
mlog_errno(ret);
+ ocfs2_update_inode_fsync_trans(handle, inode, 0);
ocfs2_commit_trans(osb, handle);
out:
return ret;
@@ -429,6 +431,7 @@ static int ocfs2_orphan_for_truncate(str
di->i_size = cpu_to_le64(new_i_size);
di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
di->i_ctime_nsec = di->i_mtime_nsec =
cpu_to_le32(inode->i_ctime.tv_nsec);
+ ocfs2_update_inode_fsync_trans(handle, inode, 0);
ocfs2_journal_dirty(handle, fe_bh);
@@ -737,6 +740,7 @@ static handle_t *ocfs2_zero_start_ordere
OCFS2_JOURNAL_ACCESS_WRITE);
if (ret)
mlog_errno(ret);
+ ocfs2_update_inode_fsync_trans(handle, inode, 1);
out:
if (ret) {
@@ -834,6 +838,7 @@ static int ocfs2_write_zero_page(struct
di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
di->i_mtime_nsec = di->i_ctime_nsec;
ocfs2_journal_dirty(handle, di_bh);
+ ocfs2_update_inode_fsync_trans(handle, inode, 1);
ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
}
@@ -1338,6 +1343,7 @@ static int __ocfs2_write_remove_suid(str
di = (struct ocfs2_dinode *) bh->b_data;
di->i_mode = cpu_to_le16(inode->i_mode);
+ ocfs2_update_inode_fsync_trans(handle, inode, 0);
ocfs2_journal_dirty(handle, bh);
@@ -1570,6 +1576,7 @@ static int ocfs2_zero_partial_clusters(s
if (ret)
mlog_errno(ret);
}
+ ocfs2_update_inode_fsync_trans(handle, inode, 1);
ocfs2_commit_trans(osb, handle);
out:
diff -puN
fs/ocfs2/move_extents.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
fs/ocfs2/move_extents.c
---
a/fs/ocfs2/move_extents.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
+++ a/fs/ocfs2/move_extents.c
@@ -151,6 +151,7 @@ static int __ocfs2_move_extent(handle_t
old_blkno, len);
}
+ ocfs2_update_inode_fsync_trans(handle, inode, 0);
out:
ocfs2_free_path(path);
return ret;
@@ -954,6 +955,7 @@ static int ocfs2_move_extents(struct ocf
inode->i_ctime = CURRENT_TIME;
di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+ ocfs2_update_inode_fsync_trans(handle, inode, 0);
ocfs2_journal_dirty(handle, di_bh);
diff -puN
fs/ocfs2/namei.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
fs/ocfs2/namei.c
---
a/fs/ocfs2/namei.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
+++ a/fs/ocfs2/namei.c
@@ -2578,6 +2578,7 @@ int ocfs2_mv_orphaned_inode_to_new(struc
di->i_orphaned_slot = 0;
set_nlink(inode, 1);
ocfs2_set_links_count(di, inode->i_nlink);
+ ocfs2_update_inode_fsync_trans(handle, inode, 1);
ocfs2_journal_dirty(handle, di_bh);
status = ocfs2_add_entry(handle, dentry, inode,
diff -puN
fs/ocfs2/suballoc.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
fs/ocfs2/suballoc.c
---
a/fs/ocfs2/suballoc.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
+++ a/fs/ocfs2/suballoc.c
@@ -771,6 +771,7 @@ static int ocfs2_block_group_alloc(struc
spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
+ ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
status = 0;
@@ -2075,7 +2076,7 @@ int ocfs2_find_new_inode_loc(struct inod
ac->ac_find_loc_priv = res;
*fe_blkno = res->sr_blkno;
-
+ ocfs2_update_inode_fsync_trans(handle, dir, 0);
out:
if (handle)
ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
diff -puN
fs/ocfs2/xattr.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
fs/ocfs2/xattr.c
---
a/fs/ocfs2/xattr.c~ocfs2-call-ocfs2_update_inode_fsync_trans-when-updating-any-inode
+++ a/fs/ocfs2/xattr.c
@@ -2602,6 +2602,7 @@ int ocfs2_xattr_remove(struct inode *ino
oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
spin_unlock(&oi->ip_lock);
+ ocfs2_update_inode_fsync_trans(handle, inode, 0);
ocfs2_journal_dirty(handle, di_bh);
out_commit:
@@ -3614,6 +3615,7 @@ int ocfs2_xattr_set(struct inode *inode,
}
ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs,
&ctxt);
+ ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
ocfs2_commit_trans(osb, ctxt.handle);
@@ -5476,6 +5478,7 @@ static int ocfs2_rm_xattr_cluster(struct
ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
if (ret)
mlog_errno(ret);
+ ocfs2_update_inode_fsync_trans(handle, inode, 0);
out_commit:
ocfs2_commit_trans(osb, handle);
_
Mark Fasheh
2014-Mar-31 02:05 UTC
[Ocfs2-devel] [patch 4/8] ocfs2: call ocfs2_update_inode_fsync_trans when updating any inode
On Wed, Mar 19, 2014 at 02:10:02PM -0700, Andrew Morton wrote:> From: "Darrick J. Wong" <darrick.wong at oracle.com> > Subject: ocfs2: call ocfs2_update_inode_fsync_trans when updating any inode > > Ensure that ocfs2_update_inode_fsync_trans() is called any time we touch > an inode in a given transaction. This is a follow-on to the previous > patch to reduce lock contention and deadlocking during an fsync operation.This looks fine but I have a question - what happens if a future patch adds some disk structure change but forgets to call ocfs2_update_inode_fsync_trans(). Could we wind up skipping some blocks to sync in that case? --Mark -- Mark Fasheh
Darrick J. Wong
2014-Apr-09 22:10 UTC
[Ocfs2-devel] [patch 4/8] ocfs2: call ocfs2_update_inode_fsync_trans when updating any inode
On Sun, Mar 30, 2014 at 07:05:46PM -0700, Mark Fasheh wrote:> On Wed, Mar 19, 2014 at 02:10:02PM -0700, Andrew Morton wrote: > > From: "Darrick J. Wong" <darrick.wong at oracle.com> > > Subject: ocfs2: call ocfs2_update_inode_fsync_trans when updating any inode > > > > Ensure that ocfs2_update_inode_fsync_trans() is called any time we touch > > an inode in a given transaction. This is a follow-on to the previous > > patch to reduce lock contention and deadlocking during an fsync operation. > > This looks fine but I have a question - what happens if a future patch adds > some disk structure change but forgets to call > ocfs2_update_inode_fsync_trans(). Could we wind up skipping some blocks to > sync in that case?Yes, you'd almost certainly miss some blocks if such a programming error were introduced. It was tempting to try to stuff the functionality into ocfs2_journal_dirty() rather than requiring a separate call, but that's not really suitable since not all dirty buffers are tied to inodes. I think the only thing we can really do to ensure at review time that anything calling ocfs2_start_trans() also include a call to ocfs2_update_inode_fsync_trans() somewhere. --D> --Mark > > -- > Mark Fasheh