[Cluster-devel] [PATCH] gfs2: Allocate bufdata object before taking log lock

[Andreas Gruenbacher]

Before this patch, functions gfs2_trans_add_{meta,data} would take the
log lock, then if they needed to allocate a new bufdata object, they
would drop the lock and reacquire it.  This patch changes it so that if
there's apparently no bufdata object, it allocates a new one before
taking the log lock.  After taking the log lock, it checks for conflicts
and takes measures to resolve the conflict.  This is for performance.

I haven't verified if the page locking in gfs2_trans_add_meta,
originally introduced in commit 18ec7d5c3f434 ("Make journaled data
files identical to normal files on disk"), is actually still needed.

Signed-off-by: Andreas Gruenbacher <agruenba at redhat.com>
---
 fs/gfs2/trans.c | 46 +++++++++++++++++++++++++---------------------
 1 file changed, 25 insertions(+), 21 deletions(-)

diff --git a/fs/gfs2/trans.c b/fs/gfs2/trans.c
index 63fec11ef2ce..c50de22d7cbf 100644
--- a/fs/gfs2/trans.c
+++ b/fs/gfs2/trans.c
@@ -171,7 +171,6 @@ static struct gfs2_bufdata *gfs2_alloc_bufdata(struct gfs2_glock *gl,
 	INIT_LIST_HEAD(&bd->bd_list);
 	INIT_LIST_HEAD(&bd->bd_ail_st_list);
 	INIT_LIST_HEAD(&bd->bd_ail_gl_list);
-	bh->b_private = bd;
 	return bd;
 }
 
@@ -193,24 +192,24 @@ void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh)
 {
 	struct gfs2_trans *tr = current->journal_info;
 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
-	struct gfs2_bufdata *bd;
+	struct gfs2_bufdata *bd = NULL;
 
 	lock_buffer(bh);
 	if (buffer_pinned(bh)) {
 		set_bit(TR_TOUCHED, &tr->tr_flags);
 		goto out;
 	}
-	gfs2_log_lock(sdp);
-	bd = bh->b_private;
-	if (bd == NULL) {
-		gfs2_log_unlock(sdp);
+	if (!bh->b_private) {
 		unlock_buffer(bh);
-		if (bh->b_private == NULL)
-			bd = gfs2_alloc_bufdata(gl, bh);
-		else
-			bd = bh->b_private;
+		bd = gfs2_alloc_bufdata(gl, bh);
 		lock_buffer(bh);
-		gfs2_log_lock(sdp);
+	}
+	gfs2_log_lock(sdp);
+	if (bh->b_private) {
+		kfree(bd);
+		bd = bh->b_private;
+	} else {
+		bh->b_private = bd;
 	}
 	gfs2_assert(sdp, bd->bd_gl == gl);
 	set_bit(TR_TOUCHED, &tr->tr_flags);
@@ -230,7 +229,7 @@ void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh)
 {
 
 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
-	struct gfs2_bufdata *bd;
+	struct gfs2_bufdata *bd = NULL;
 	struct gfs2_meta_header *mh;
 	struct gfs2_trans *tr = current->journal_info;
 	enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
@@ -240,19 +239,24 @@ void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh)
 		set_bit(TR_TOUCHED, &tr->tr_flags);
 		goto out;
 	}
-	gfs2_log_lock(sdp);
-	bd = bh->b_private;
-	if (bd == NULL) {
-		gfs2_log_unlock(sdp);
+	if (!bh->b_private) {
 		unlock_buffer(bh);
+		bd = gfs2_alloc_bufdata(gl, bh);
+		lock_buffer(bh);
+	}
+	gfs2_log_lock(sdp);
+	if (bh->b_private) {
+		kfree(bd);
+		bd = bh->b_private;
+	} else {
 		lock_page(bh->b_page);
-		if (bh->b_private == NULL)
-			bd = gfs2_alloc_bufdata(gl, bh);
-		else
+		if (bh->b_private) {
+			kfree(bd);
 			bd = bh->b_private;
+		} else {
+			bh->b_private = bd;
+		}
 		unlock_page(bh->b_page);
-		lock_buffer(bh);
-		gfs2_log_lock(sdp);
 	}
 	gfs2_assert(sdp, bd->bd_gl == gl);
 	set_bit(TR_TOUCHED, &tr->tr_flags);
-- 
2.26.3