[Cluster-devel] [PATCH 0/8] GFS2: Pre-pull patch posting (merge window)

[Cluster-devel] [PATCH 0/8] GFS2: Pre-pull patch posting (merge window)

[Bob Peterson]

Hi,

We've got eight GFS2 patches for this merge window:

1. Andy Price submitted a patch to make gfs2_write_full_page a
   static function.
2. Dan Carpenter submitted a patch to fix a ERR_PTR thinko.

I've also got a few patches, three of which fix bugs related to
deleting very large files, which cause GFS2 to run out of
journal space:

3. The first one prevents GFS2 delete operation from requesting too
   much journal space.
4. The second one fixes a problem whereby GFS2 can hang because it
   wasn't taking journal space demand into its calculations.
5. The third one wakes up IO waiters when a flush is done to restart
   processes stuck waiting for journal space to become available.

The other three patches are a performance improvement related to
spin_lock contention between multiple writers:

6. The "tr_touched" variable was switched to a flag to be more atomic
   and eliminate the possibility of some races. 
7. Function meta_lo_add was moved inline with its only caller to make
   the code more readable and efficient.
8. Contention on the gfs2_log_lock spinlock was greatly reduced by
   avoiding the lock altogether in cases where we don't really need
   it: buffers that already appear in the appropriate metadata list
   for the journal. Many thanks to Steve Whitehouse for the ideas and
   principles behind these patches.

Regards,

Bob Peterson
---
Andrew Price (1):
  gfs2: Make gfs2_write_full_page static

Bob Peterson (6):
  GFS2: Limit number of transaction blocks requested for truncates
  GFS2: Made logd daemon take into account log demand
  GFS2: Wake up io waiters whenever a flush is done
  GFS2: Switch tr_touched to flag in transaction
  GFS2: Inline function meta_lo_add
  GFS2: Reduce contention on gfs2_log_lock

Dan Carpenter (1):
  GFS2: Fix reference to ERR_PTR in gfs2_glock_iter_next

 fs/gfs2/aops.c       |  4 +--
 fs/gfs2/bmap.c       | 29 +++++++++++++++++--
 fs/gfs2/glock.c      | 12 ++++----
 fs/gfs2/incore.h     | 11 +++++--
 fs/gfs2/log.c        | 21 ++++++++++----
 fs/gfs2/meta_io.c    |  6 ++--
 fs/gfs2/ops_fstype.c |  1 +
 fs/gfs2/trans.c      | 81 ++++++++++++++++++++++++++++------------------------
 8 files changed, 105 insertions(+), 60 deletions(-)

-- 
2.9.3

[Cluster-devel] [PATCH 1/8] GFS2: Fix reference to ERR_PTR in gfs2_glock_iter_next

[Bob Peterson]

From: Dan Carpenter <dan.carpenter@oracle.com>

This patch fixes a place where function gfs2_glock_iter_next can
reference an invalid error pointer.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Bob Peterson <rpeterso@redhat.com>
---
 fs/gfs2/glock.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
index 14cbf60..2928f12 100644
--- a/fs/gfs2/glock.c
+++ b/fs/gfs2/glock.c
@@ -1802,16 +1802,18 @@ void gfs2_glock_exit(void)
 
 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
 {
-	do {
-		gi->gl = rhashtable_walk_next(&gi->hti);
+	while ((gi->gl = rhashtable_walk_next(&gi->hti))) {
 		if (IS_ERR(gi->gl)) {
 			if (PTR_ERR(gi->gl) == -EAGAIN)
 				continue;
 			gi->gl = NULL;
+			return;
 		}
-	/* Skip entries for other sb and dead entries */
-	} while ((gi->gl) && ((gi->sdp != gi->gl->gl_name.ln_sbd) ||
-			      __lockref_is_dead(&gi->gl->gl_lockref)));
+		/* Skip entries for other sb and dead entries */
+		if (gi->sdp == gi->gl->gl_name.ln_sbd &&
+		    !__lockref_is_dead(&gi->gl->gl_lockref))
+			return;
+	}
 }
 
 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
-- 
2.9.3

[Cluster-devel] [PATCH 2/8] GFS2: Limit number of transaction blocks requested for truncates

[Bob Peterson]

This patch limits the number of transaction blocks requested during
file truncates. If we have very large multi-terabyte files, and want
to delete or truncate them, they might span so many resource groups
that we overflow the journal blocks, and cause an assert failure.
By limiting the number of blocks in the transaction, we prevent this
overflow and give other running processes time to do transactions.

The limiting factor I chose is sd_log_thresh2 which is currently
set to 4/5ths of the journal. This same ratio is used in function
gfs2_ail_flush_reqd to determine when a log flush is required.
If we make the maximum value less than this, we can get into a
infinite hang whereby the log stops moving because the number of
used blocks is less than the threshold and the iterative loop
needs more, but since we're under the threshold, the log daemon
never starts any IO on the log.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
---
 fs/gfs2/bmap.c | 29 ++++++++++++++++++++++++++---
 1 file changed, 26 insertions(+), 3 deletions(-)

diff --git a/fs/gfs2/bmap.c b/fs/gfs2/bmap.c
index 645721f..cbad0ef 100644
--- a/fs/gfs2/bmap.c
+++ b/fs/gfs2/bmap.c
@@ -720,6 +720,7 @@ static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 {
 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 	struct gfs2_rgrp_list rlist;
+	struct gfs2_trans *tr;
 	u64 bn, bstart;
 	u32 blen, btotal;
 	__be64 *p;
@@ -728,6 +729,7 @@ static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 	unsigned int revokes = 0;
 	int x;
 	int error;
+	int jblocks_rqsted;
 
 	error = gfs2_rindex_update(sdp);
 	if (error)
@@ -791,12 +793,17 @@ static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 	if (gfs2_rs_active(&ip->i_res)) /* needs to be done with the rgrp glock held */
 		gfs2_rs_deltree(&ip->i_res);
 
-	error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE +
-				 RES_INDIRECT + RES_STATFS + RES_QUOTA,
-				 revokes);
+restart:
+	jblocks_rqsted = rg_blocks + RES_DINODE +
+		RES_INDIRECT + RES_STATFS + RES_QUOTA +
+		gfs2_struct2blk(sdp, revokes, sizeof(u64));
+	if (jblocks_rqsted > atomic_read(&sdp->sd_log_thresh2))
+		jblocks_rqsted = atomic_read(&sdp->sd_log_thresh2);
+	error = gfs2_trans_begin(sdp, jblocks_rqsted, revokes);
 	if (error)
 		goto out_rg_gunlock;
 
+	tr = current->journal_info;
 	down_write(&ip->i_rw_mutex);
 
 	gfs2_trans_add_meta(ip->i_gl, dibh);
@@ -810,6 +817,16 @@ static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 		if (!*p)
 			continue;
 
+		/* check for max reasonable journal transaction blocks */
+		if (tr->tr_num_buf_new + RES_STATFS +
+		    RES_QUOTA >= atomic_read(&sdp->sd_log_thresh2)) {
+			if (rg_blocks >= tr->tr_num_buf_new)
+				rg_blocks -= tr->tr_num_buf_new;
+			else
+				rg_blocks = 0;
+			break;
+		}
+
 		bn = be64_to_cpu(*p);
 
 		if (bstart + blen == bn)
@@ -827,6 +844,9 @@ static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 		*p = 0;
 		gfs2_add_inode_blocks(&ip->i_inode, -1);
 	}
+	if (p == bottom)
+		rg_blocks = 0;
+
 	if (bstart) {
 		__gfs2_free_blocks(ip, bstart, blen, metadata);
 		btotal += blen;
@@ -844,6 +864,9 @@ static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 
 	gfs2_trans_end(sdp);
 
+	if (rg_blocks)
+		goto restart;
+
 out_rg_gunlock:
 	gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
 out_rlist:
-- 
2.9.3

[Cluster-devel] [PATCH 3/8] GFS2: Made logd daemon take into account log demand

[Bob Peterson]

Before this patch, the logd daemon only tried to flush things when
the log blocks pinned exceeded a certain threshold. But when we're
deleting very large files, it may require a huge number of journal
blocks, and that, in turn, may exceed the threshold. This patch
factors that into account.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
---
 fs/gfs2/incore.h     | 1 +
 fs/gfs2/log.c        | 9 +++++++--
 fs/gfs2/ops_fstype.c | 1 +
 3 files changed, 9 insertions(+), 2 deletions(-)

diff --git a/fs/gfs2/incore.h b/fs/gfs2/incore.h
index a6a3389..00d8dc2 100644
--- a/fs/gfs2/incore.h
+++ b/fs/gfs2/incore.h
@@ -794,6 +794,7 @@ struct gfs2_sbd {
 	atomic_t sd_log_thresh1;
 	atomic_t sd_log_thresh2;
 	atomic_t sd_log_blks_free;
+	atomic_t sd_log_blks_needed;
 	wait_queue_head_t sd_log_waitq;
 	wait_queue_head_t sd_logd_waitq;
 
diff --git a/fs/gfs2/log.c b/fs/gfs2/log.c
index e58ccef0..4df349c 100644
--- a/fs/gfs2/log.c
+++ b/fs/gfs2/log.c
@@ -349,6 +349,7 @@ int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
 	if (gfs2_assert_warn(sdp, blks) ||
 	    gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
 		return -EINVAL;
+	atomic_add(blks, &sdp->sd_log_blks_needed);
 retry:
 	free_blocks = atomic_read(&sdp->sd_log_blks_free);
 	if (unlikely(free_blocks <= wanted)) {
@@ -370,6 +371,7 @@ retry:
 			wake_up(&sdp->sd_reserving_log_wait);
 		goto retry;
 	}
+	atomic_sub(blks, &sdp->sd_log_blks_needed);
 	trace_gfs2_log_blocks(sdp, -blks);
 
 	/*
@@ -891,13 +893,16 @@ void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 
 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
 {
-	return (atomic_read(&sdp->sd_log_pinned) >= atomic_read(&sdp->sd_log_thresh1));
+	return (atomic_read(&sdp->sd_log_pinned) +
+		atomic_read(&sdp->sd_log_blks_needed) >=
+		atomic_read(&sdp->sd_log_thresh1));
 }
 
 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
 {
 	unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
-	return used_blocks >= atomic_read(&sdp->sd_log_thresh2);
+	return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
+		atomic_read(&sdp->sd_log_thresh2);
 }
 
 /**
diff --git a/fs/gfs2/ops_fstype.c b/fs/gfs2/ops_fstype.c
index ff72ac6..86281a9 100644
--- a/fs/gfs2/ops_fstype.c
+++ b/fs/gfs2/ops_fstype.c
@@ -683,6 +683,7 @@ static int init_journal(struct gfs2_sbd *sdp, int undo)
 		goto fail_jindex;
 	}
 
+	atomic_set(&sdp->sd_log_blks_needed, 0);
 	if (sdp->sd_args.ar_spectator) {
 		sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 		atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
-- 
2.9.3

[Cluster-devel] [PATCH 4/8] GFS2: Wake up io waiters whenever a flush is done

[Bob Peterson]

Before this patch, if a process called function gfs2_log_reserve to
reserve some journal blocks, but the journal not enough blocks were
free, it would call io_schedule. However, in the log flush daemon,
it woke up the waiters only if an gfs2_ail_flush was no longer
required. This resulted in situations where processes would wait
forever because the number of blocks required was so high that it
pushed the journal into a perpetual state of flush being required.

This patch changes the logd daemon so that it wakes up io waiters
every time the log is actually flushed.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
---
 fs/gfs2/log.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/fs/gfs2/log.c b/fs/gfs2/log.c
index 4df349c..5028a9d 100644
--- a/fs/gfs2/log.c
+++ b/fs/gfs2/log.c
@@ -918,12 +918,15 @@ int gfs2_logd(void *data)
 	struct gfs2_sbd *sdp = data;
 	unsigned long t = 1;
 	DEFINE_WAIT(wait);
+	bool did_flush;
 
 	while (!kthread_should_stop()) {
 
+		did_flush = false;
 		if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
 			gfs2_ail1_empty(sdp);
 			gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
+			did_flush = true;
 		}
 
 		if (gfs2_ail_flush_reqd(sdp)) {
@@ -931,9 +934,10 @@ int gfs2_logd(void *data)
 			gfs2_ail1_wait(sdp);
 			gfs2_ail1_empty(sdp);
 			gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
+			did_flush = true;
 		}
 
-		if (!gfs2_ail_flush_reqd(sdp))
+		if (!gfs2_ail_flush_reqd(sdp) || did_flush)
 			wake_up(&sdp->sd_log_waitq);
 
 		t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
-- 
2.9.3

[Cluster-devel] [PATCH 5/8] GFS2: Switch tr_touched to flag in transaction

[Bob Peterson]

This patch eliminates the int variable tr_touched in favor of a
new flag in the transaction. This is a step toward reducing contention
on the gfs2_log_lock spin_lock.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
---
 fs/gfs2/incore.h  | 10 +++++++---
 fs/gfs2/log.c     |  6 +++---
 fs/gfs2/meta_io.c |  6 +++---
 fs/gfs2/trans.c   | 19 ++++++++++---------
 4 files changed, 23 insertions(+), 18 deletions(-)

diff --git a/fs/gfs2/incore.h b/fs/gfs2/incore.h
index 00d8dc2..c45084a 100644
--- a/fs/gfs2/incore.h
+++ b/fs/gfs2/incore.h
@@ -470,15 +470,19 @@ struct gfs2_quota_data {
 	struct rcu_head qd_rcu;
 };
 
+enum {
+	TR_TOUCHED = 1,
+	TR_ATTACHED = 2,
+	TR_ALLOCED = 3,
+};
+
 struct gfs2_trans {
 	unsigned long tr_ip;
 
 	unsigned int tr_blocks;
 	unsigned int tr_revokes;
 	unsigned int tr_reserved;
-	unsigned int tr_touched:1;
-	unsigned int tr_attached:1;
-	unsigned int tr_alloced:1;
+	unsigned long tr_flags;
 
 	unsigned int tr_num_buf_new;
 	unsigned int tr_num_databuf_new;
diff --git a/fs/gfs2/log.c b/fs/gfs2/log.c
index 5028a9d..4fb76c0 100644
--- a/fs/gfs2/log.c
+++ b/fs/gfs2/log.c
@@ -799,7 +799,7 @@ void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl,
 
 static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new)
 {
-	WARN_ON_ONCE(old->tr_attached != 1);
+	WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
 
 	old->tr_num_buf_new	+= new->tr_num_buf_new;
 	old->tr_num_databuf_new	+= new->tr_num_databuf_new;
@@ -823,9 +823,9 @@ static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 	if (sdp->sd_log_tr) {
 		gfs2_merge_trans(sdp->sd_log_tr, tr);
 	} else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
-		gfs2_assert_withdraw(sdp, tr->tr_alloced);
+		gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
 		sdp->sd_log_tr = tr;
-		tr->tr_attached = 1;
+		set_bit(TR_ATTACHED, &tr->tr_flags);
 	}
 
 	sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
diff --git a/fs/gfs2/meta_io.c b/fs/gfs2/meta_io.c
index 373639a5..a88a347 100644
--- a/fs/gfs2/meta_io.c
+++ b/fs/gfs2/meta_io.c
@@ -293,7 +293,7 @@ int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
 	wait_on_buffer(bh);
 	if (unlikely(!buffer_uptodate(bh))) {
 		struct gfs2_trans *tr = current->journal_info;
-		if (tr && tr->tr_touched)
+		if (tr && test_bit(TR_TOUCHED, &tr->tr_flags))
 			gfs2_io_error_bh(sdp, bh);
 		brelse(bh);
 		*bhp = NULL;
@@ -320,7 +320,7 @@ int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh)
 
 	if (!buffer_uptodate(bh)) {
 		struct gfs2_trans *tr = current->journal_info;
-		if (tr && tr->tr_touched)
+		if (tr && test_bit(TR_TOUCHED, &tr->tr_flags))
 			gfs2_io_error_bh(sdp, bh);
 		return -EIO;
 	}
@@ -346,7 +346,7 @@ void gfs2_remove_from_journal(struct buffer_head *bh, int meta)
 			tr->tr_num_buf_rm++;
 		else
 			tr->tr_num_databuf_rm++;
-		tr->tr_touched = 1;
+		set_bit(TR_TOUCHED, &tr->tr_flags);
 		was_pinned = 1;
 		brelse(bh);
 	}
diff --git a/fs/gfs2/trans.c b/fs/gfs2/trans.c
index 0c1bde3..5d7f5b0 100644
--- a/fs/gfs2/trans.c
+++ b/fs/gfs2/trans.c
@@ -48,7 +48,7 @@ int gfs2_trans_begin(struct gfs2_sbd *sdp, unsigned int blocks,
 	tr->tr_blocks = blocks;
 	tr->tr_revokes = revokes;
 	tr->tr_reserved = 1;
-	tr->tr_alloced = 1;
+	set_bit(TR_ALLOCED, &tr->tr_flags);
 	if (blocks)
 		tr->tr_reserved += 6 + blocks;
 	if (revokes)
@@ -78,7 +78,8 @@ static void gfs2_print_trans(const struct gfs2_trans *tr)
 {
 	pr_warn("Transaction created at: %pSR\n", (void *)tr->tr_ip);
 	pr_warn("blocks=%u revokes=%u reserved=%u touched=%u\n",
-		tr->tr_blocks, tr->tr_revokes, tr->tr_reserved, tr->tr_touched);
+		tr->tr_blocks, tr->tr_revokes, tr->tr_reserved,
+		test_bit(TR_TOUCHED, &tr->tr_flags));
 	pr_warn("Buf %u/%u Databuf %u/%u Revoke %u/%u\n",
 		tr->tr_num_buf_new, tr->tr_num_buf_rm,
 		tr->tr_num_databuf_new, tr->tr_num_databuf_rm,
@@ -89,12 +90,12 @@ void gfs2_trans_end(struct gfs2_sbd *sdp)
 {
 	struct gfs2_trans *tr = current->journal_info;
 	s64 nbuf;
-	int alloced = tr->tr_alloced;
+	int alloced = test_bit(TR_ALLOCED, &tr->tr_flags);
 
 	BUG_ON(!tr);
 	current->journal_info = NULL;
 
-	if (!tr->tr_touched) {
+	if (!test_bit(TR_TOUCHED, &tr->tr_flags)) {
 		gfs2_log_release(sdp, tr->tr_reserved);
 		if (alloced) {
 			kfree(tr);
@@ -112,8 +113,8 @@ void gfs2_trans_end(struct gfs2_sbd *sdp)
 		gfs2_print_trans(tr);
 
 	gfs2_log_commit(sdp, tr);
-	if (alloced && !tr->tr_attached)
-			kfree(tr);
+	if (alloced && !test_bit(TR_ATTACHED, &tr->tr_flags))
+		kfree(tr);
 	up_read(&sdp->sd_log_flush_lock);
 
 	if (sdp->sd_vfs->s_flags & MS_SYNCHRONOUS)
@@ -182,7 +183,7 @@ void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh)
 		gfs2_log_lock(sdp);
 	}
 	gfs2_assert(sdp, bd->bd_gl == gl);
-	tr->tr_touched = 1;
+	set_bit(TR_TOUCHED, &tr->tr_flags);
 	if (list_empty(&bd->bd_list)) {
 		set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 		set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
@@ -201,7 +202,7 @@ static void meta_lo_add(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
 	enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 
 	tr = current->journal_info;
-	tr->tr_touched = 1;
+	set_bit(TR_TOUCHED, &tr->tr_flags);
 	if (!list_empty(&bd->bd_list))
 		return;
 	set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
@@ -256,7 +257,7 @@ void gfs2_trans_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
 
 	BUG_ON(!list_empty(&bd->bd_list));
 	gfs2_add_revoke(sdp, bd);
-	tr->tr_touched = 1;
+	set_bit(TR_TOUCHED, &tr->tr_flags);
 	tr->tr_num_revoke++;
 }
 
-- 
2.9.3

[Cluster-devel] [PATCH 6/8] GFS2: Inline function meta_lo_add

[Bob Peterson]

This patch simply combines function meta_lo_add with its only
caller, trans_add_meta. This makes the code easier to read and
will make it easier to reduce contention on gfs2_log_lock.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
---
 fs/gfs2/trans.c | 49 ++++++++++++++++++++++---------------------------
 1 file changed, 22 insertions(+), 27 deletions(-)

diff --git a/fs/gfs2/trans.c b/fs/gfs2/trans.c
index 5d7f5b0..483c5a7 100644
--- a/fs/gfs2/trans.c
+++ b/fs/gfs2/trans.c
@@ -195,16 +195,35 @@ void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh)
 	unlock_buffer(bh);
 }
 
-static void meta_lo_add(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
+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_meta_header *mh;
 	struct gfs2_trans *tr;
 	enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 
+	lock_buffer(bh);
+	gfs2_log_lock(sdp);
+	bd = bh->b_private;
+	if (bd == NULL) {
+		gfs2_log_unlock(sdp);
+		unlock_buffer(bh);
+		lock_page(bh->b_page);
+		if (bh->b_private == NULL)
+			bd = gfs2_alloc_bufdata(gl, bh, &gfs2_buf_lops);
+		else
+			bd = bh->b_private;
+		unlock_page(bh->b_page);
+		lock_buffer(bh);
+		gfs2_log_lock(sdp);
+	}
+	gfs2_assert(sdp, bd->bd_gl == gl);
 	tr = current->journal_info;
 	set_bit(TR_TOUCHED, &tr->tr_flags);
 	if (!list_empty(&bd->bd_list))
-		return;
+		goto out_unlock;
 	set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 	set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
 	mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
@@ -222,31 +241,7 @@ static void meta_lo_add(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
 	mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
 	list_add(&bd->bd_list, &tr->tr_buf);
 	tr->tr_num_buf_new++;
-}
-
-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;
-
-	lock_buffer(bh);
-	gfs2_log_lock(sdp);
-	bd = bh->b_private;
-	if (bd == NULL) {
-		gfs2_log_unlock(sdp);
-		unlock_buffer(bh);
-		lock_page(bh->b_page);
-		if (bh->b_private == NULL)
-			bd = gfs2_alloc_bufdata(gl, bh, &gfs2_buf_lops);
-		else
-			bd = bh->b_private;
-		unlock_page(bh->b_page);
-		lock_buffer(bh);
-		gfs2_log_lock(sdp);
-	}
-	gfs2_assert(sdp, bd->bd_gl == gl);
-	meta_lo_add(sdp, bd);
+out_unlock:
 	gfs2_log_unlock(sdp);
 	unlock_buffer(bh);
 }
-- 
2.9.3

[Cluster-devel] [PATCH 7/8] GFS2: Reduce contention on gfs2_log_lock

[Bob Peterson]

This patch modifies functions gfs2_trans_add_meta and _data so that
they check whether the buffer_head is already in a transaction,
and if so, avoid taking the gfs2_log_lock.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
---
 fs/gfs2/trans.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/fs/gfs2/trans.c b/fs/gfs2/trans.c
index 483c5a7..affef3c 100644
--- a/fs/gfs2/trans.c
+++ b/fs/gfs2/trans.c
@@ -170,6 +170,10 @@ void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh)
 	}
 
 	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) {
@@ -192,6 +196,7 @@ void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh)
 		list_add_tail(&bd->bd_list, &tr->tr_databuf);
 	}
 	gfs2_log_unlock(sdp);
+out:
 	unlock_buffer(bh);
 }
 
@@ -201,10 +206,14 @@ 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_meta_header *mh;
-	struct gfs2_trans *tr;
+	struct gfs2_trans *tr = current->journal_info;
 	enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 
 	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) {
@@ -220,7 +229,6 @@ void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh)
 		gfs2_log_lock(sdp);
 	}
 	gfs2_assert(sdp, bd->bd_gl == gl);
-	tr = current->journal_info;
 	set_bit(TR_TOUCHED, &tr->tr_flags);
 	if (!list_empty(&bd->bd_list))
 		goto out_unlock;
@@ -243,6 +251,7 @@ void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh)
 	tr->tr_num_buf_new++;
 out_unlock:
 	gfs2_log_unlock(sdp);
+out:
 	unlock_buffer(bh);
 }
 
-- 
2.9.3

[Cluster-devel] [PATCH 8/8] gfs2: Make gfs2_write_full_page static

[Bob Peterson]

From: Andrew Price <anprice@redhat.com>

It only gets called from aops.c and doesn't appear in any headers.

Signed-off-by: Andrew Price <anprice@redhat.com>
Signed-off-by: Bob Peterson <rpeterso@redhat.com>
---
 fs/gfs2/aops.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/gfs2/aops.c b/fs/gfs2/aops.c
index 5a6f52e..755b81e 100644
--- a/fs/gfs2/aops.c
+++ b/fs/gfs2/aops.c
@@ -143,8 +143,8 @@ static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
 /* This is the same as calling block_write_full_page, but it also
  * writes pages outside of i_size
  */
-int gfs2_write_full_page(struct page *page, get_block_t *get_block,
-			 struct writeback_control *wbc)
+static int gfs2_write_full_page(struct page *page, get_block_t *get_block,
+				struct writeback_control *wbc)
 {
 	struct inode * const inode = page->mapping->host;
 	loff_t i_size = i_size_read(inode);
-- 
2.9.3