[Cluster-devel] GFS2: Clean up fsync()

[Steven Whitehouse <swhiteho@redhat.com>]

This patch (still being tested) is designed to clean up GFS2's fsync
implementation and ensure that it really does get everything on
disk. Since ->write_inode() has been updated, we can call that
via the vfs library function sync_inode_metadata() and the only
remaining thing that has to be done is to ensure that we get
revoke records in the log after the inode has been written back.

This is just a "heads up" at this stage so you can see what I'm
planning.

diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index e483108..23eab47 100644
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -545,18 +545,10 @@ static int gfs2_close(struct inode *inode, struct file *file)
 /**
  * gfs2_fsync - sync the dirty data for a file (across the cluster)
  * @file: the file that points to the dentry (we ignore this)
- * @dentry: the dentry that points to the inode to sync
+ * @datasync: set if we can ignore timestamp changes
  *
- * The VFS will flush "normal" data for us. We only need to worry
- * about metadata here. For journaled data, we just do a log flush
- * as we can't avoid it. Otherwise we can just bale out if datasync
- * is set. For stuffed inodes we must flush the log in order to
- * ensure that all data is on disk.
- *
- * The call to write_inode_now() is there to write back metadata and
- * the inode itself. It does also try and write the data, but thats
- * (hopefully) a no-op due to the VFS having already called filemap_fdatawrite()
- * for us.
+ * The VFS will flush data for us. We only need to worry
+ * about metadata here.
  *
  * Returns: errno
  */
@@ -565,22 +557,20 @@ static int gfs2_fsync(struct file *file, int datasync)
 {
 	struct inode *inode = file->f_mapping->host;
 	int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
-	int ret = 0;
-
-	if (gfs2_is_jdata(GFS2_I(inode))) {
-		gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
-		return 0;
-	}
+	struct gfs2_inode *ip = GFS2_I(inode);
+	int ret;
 
-	if (sync_state != 0) {
-		if (!datasync)
-			ret = write_inode_now(inode, 0);
+	if (datasync)
+		sync_state &= ~I_DIRTY_SYNC;
 
-		if (gfs2_is_stuffed(GFS2_I(inode)))
-			gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
+	if (sync_state) {
+		ret = sync_inode_metadata(inode, 1);
+		if (ret)
+			return ret;
+		gfs2_ail_flush(ip->i_gl);
 	}
 
-	return ret;
+	return 0;
 }
 
 /**
diff --git a/fs/gfs2/glops.c b/fs/gfs2/glops.c
index 25eeb2b..7c1b08f 100644
--- a/fs/gfs2/glops.c
+++ b/fs/gfs2/glops.c
@@ -28,33 +28,18 @@
 #include "trans.h"
 
 /**
- * ail_empty_gl - remove all buffers for a given lock from the AIL
+ * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
  * @gl: the glock
  *
  * None of the buffers should be dirty, locked, or pinned.
  */
 
-static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
+static void __gfs2_ail_flush(struct gfs2_glock *gl)
 {
 	struct gfs2_sbd *sdp = gl->gl_sbd;
 	struct list_head *head = &gl->gl_ail_list;
 	struct gfs2_bufdata *bd;
 	struct buffer_head *bh;
-	struct gfs2_trans tr;
-
-	memset(&tr, 0, sizeof(tr));
-	tr.tr_revokes = atomic_read(&gl->gl_ail_count);
-
-	if (!tr.tr_revokes)
-		return;
-
-	/* A shortened, inline version of gfs2_trans_begin() */
-	tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
-	tr.tr_ip = (unsigned long)__builtin_return_address(0);
-	INIT_LIST_HEAD(&tr.tr_list_buf);
-	gfs2_log_reserve(sdp, tr.tr_reserved);
-	BUG_ON(current->journal_info);
-	current->journal_info = &tr;
 
 	spin_lock(&sdp->sd_ail_lock);
 	while (!list_empty(head)) {
@@ -76,7 +61,47 @@ static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
 	}
 	gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
 	spin_unlock(&sdp->sd_ail_lock);
+}
+
+
+static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
+{
+	struct gfs2_sbd *sdp = gl->gl_sbd;
+	struct gfs2_trans tr;
+
+	memset(&tr, 0, sizeof(tr));
+	tr.tr_revokes = atomic_read(&gl->gl_ail_count);
+
+	if (!tr.tr_revokes)
+		return;
+
+	/* A shortened, inline version of gfs2_trans_begin() */
+	tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
+	tr.tr_ip = (unsigned long)__builtin_return_address(0);
+	INIT_LIST_HEAD(&tr.tr_list_buf);
+	gfs2_log_reserve(sdp, tr.tr_reserved);
+	BUG_ON(current->journal_info);
+	current->journal_info = &tr;
+
+	__gfs2_ail_flush(gl);
+
+	gfs2_trans_end(sdp);
+	gfs2_log_flush(sdp, NULL);
+}
 
+void gfs2_ail_flush(struct gfs2_glock *gl)
+{
+	struct gfs2_sbd *sdp = gl->gl_sbd;
+	unsigned int revokes = atomic_read(&gl->gl_ail_count);
+	int ret;
+
+	if (!revokes)
+		return;
+
+	ret = gfs2_trans_begin(sdp, 0, revokes);
+	if (ret)
+		return;
+	__gfs2_ail_flush(gl);
 	gfs2_trans_end(sdp);
 	gfs2_log_flush(sdp, NULL);
 }
diff --git a/fs/gfs2/glops.h b/fs/gfs2/glops.h
index b3aa2e3..6fce409 100644
--- a/fs/gfs2/glops.h
+++ b/fs/gfs2/glops.h
@@ -23,4 +23,6 @@ extern const struct gfs2_glock_operations gfs2_quota_glops;
 extern const struct gfs2_glock_operations gfs2_journal_glops;
 extern const struct gfs2_glock_operations *gfs2_glops_list[];
 
+extern void gfs2_ail_flush(struct gfs2_glock *gl);
+
 #endif /* __GLOPS_DOT_H__ */