From 2c82bcf0eb655679065e5fb2add651e3e6dc3252 Mon Sep 17 00:00:00 2001
From: Chris Mason <chris.mason@oracle.com>
Date: Fri, 7 Nov 2008 18:22:45 -0500
Subject: [PATCH] --- yaml --- r: 129002 b: refs/heads/master c:
 5f2cc086ccab27ac5252b3883ac004347860b4c7 h: refs/heads/master v: v3

---
 [refs]                       |  2 +-
 trunk/fs/btrfs/transaction.c | 30 +++++++++++++++++++++++++++++-
 trunk/fs/btrfs/volumes.c     |  3 ++-
 3 files changed, 32 insertions(+), 3 deletions(-)

diff --git a/[refs] b/[refs]
index 8f7261a8ebda..8a7781adf1b1 100644
--- a/[refs]
+++ b/[refs]
@@ -1,2 +1,2 @@
 ---
-refs/heads/master: 42e70e7a2f9d96fd843723fa46d5121cb3e551d0
+refs/heads/master: 5f2cc086ccab27ac5252b3883ac004347860b4c7
diff --git a/trunk/fs/btrfs/transaction.c b/trunk/fs/btrfs/transaction.c
index e72a013d24bf..202c1b6df4a4 100644
--- a/trunk/fs/btrfs/transaction.c
+++ b/trunk/fs/btrfs/transaction.c
@@ -20,6 +20,7 @@
 #include <linux/sched.h>
 #include <linux/writeback.h>
 #include <linux/pagemap.h>
+#include <linux/blkdev.h>
 #include "ctree.h"
 #include "disk-io.h"
 #include "transaction.h"
@@ -331,6 +332,7 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
 	int werr = 0;
 	struct page *page;
 	struct inode *btree_inode = root->fs_info->btree_inode;
+	struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
 	u64 start = 0;
 	u64 end;
 	unsigned long index;
@@ -371,6 +373,11 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
 			page_cache_release(page);
 		}
 	}
+	/*
+	 * we unplug once and then use the wait_on_extent_bit for
+	 * everything else
+	 */
+	blk_run_address_space(btree_inode->i_mapping);
 	while(1) {
 		ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
 					    EXTENT_DIRTY);
@@ -391,7 +398,28 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
 				if (err)
 					werr = err;
 			}
-			wait_on_page_writeback(page);
+			if (PageWriteback(page)) {
+				/*
+				 * we don't wait on the page writeback bit
+				 * because that triggers a lot of unplugs.
+				 * The extent bits are much nicer to
+				 * the disks, but come with a slightly
+				 * higher latency because we aren't forcing
+				 * unplugs.
+				 */
+				wait_on_extent_writeback(io_tree,
+					 page_offset(page),
+					 page_offset(page) +
+					 PAGE_CACHE_SIZE - 1);
+			}
+			if (PageWriteback(page)) {
+				/*
+				 * the state bits get cleared before the
+				 * page bits, lets add some extra
+				 * paranoia here
+				 */
+				wait_on_page_writeback(page);
+			}
 			page_cache_release(page);
 			cond_resched();
 		}
diff --git a/trunk/fs/btrfs/volumes.c b/trunk/fs/btrfs/volumes.c
index cbb9bb31431d..80a27284dbf1 100644
--- a/trunk/fs/btrfs/volumes.c
+++ b/trunk/fs/btrfs/volumes.c
@@ -200,7 +200,8 @@ static int noinline run_scheduled_bios(struct btrfs_device *device)
 		 * is now congested.  Back off and let other work structs
 		 * run instead
 		 */
-		if (pending && bdi_write_congested(bdi)) {
+		if (pending && bdi_write_congested(bdi) &&
+		    fs_info->fs_devices->open_devices > 1) {
 			struct bio *old_head;
 
 			spin_lock(&device->io_lock);