Btrfs: improve fsync by filtering extents that we want

This is based on Josef's "Btrfs: turbo charge fsync".

The above Josef's patch performs very good in random sync write test,
because we won't have too much extents to merge.

However, it does not performs good on the test:
dd if=/dev/zero of=foobar bs=4k count=12500 oflag=sync

The reason is when we do sequencial sync write, we need to merge the
current extent just with the previous one, so that we can get accumulated
extents to log:

A(4k) --> AA(8k) --> AAA(12k) --> AAAA(16k) ...

So we'll have to flush more and more checksum into log tree, which is the
bottleneck according to my tests.

But we can avoid this by telling fsync the real extents that are needed
to be logged.

With this, I did the above dd sync write test (size=50m),

         w/o (orig)   w/ (josef's)   w/ (this)
SATA      104KB/s       109KB/s       121KB/s
ramdisk   1.5MB/s       1.5MB/s       10.7MB/s (613%)

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>

1 files changed, 20 insertions, 0 deletions

diff --git a/fs/btrfs/extent_map.c b/fs/btrfs/extent_map.c
index 1fe82cfc1d93..ac606f076eb7 100644
--- a/fs/btrfs/extent_map.c
+++ b/fs/btrfs/extent_map.c
@@ -203,6 +203,8 @@ static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
                         em->block_start = merge->block_start;
                         merge->in_tree = 0;
                         if (merge->generation > em->generation) {
+                                em->mod_start = em->start;
+                                em->mod_len = em->len;
                                 em->generation = merge->generation;
                                 list_move(&em->list, &tree->modified_extents);
                         }
@@ -222,6 +224,7 @@ static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
                 rb_erase(&merge->rb_node, &tree->map);
                 merge->in_tree = 0;
                 if (merge->generation > em->generation) {
+                        em->mod_len = em->len;
                         em->generation = merge->generation;
                         list_move(&em->list, &tree->modified_extents);
                 }
@@ -247,6 +250,7 @@ int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
 {
         int ret = 0;
         struct extent_map *em;
+        bool prealloc = false;
 
         write_lock(&tree->lock);
         em = lookup_extent_mapping(tree, start, len);
@@ -259,8 +263,21 @@ int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
         list_move(&em->list, &tree->modified_extents);
         em->generation = gen;
         clear_bit(EXTENT_FLAG_PINNED, &em->flags);
+        em->mod_start = em->start;
+        em->mod_len = em->len;
+
+        if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
+                prealloc = true;
+                clear_bit(EXTENT_FLAG_PREALLOC, &em->flags);
+        }
 
         try_merge_map(tree, em);
+
+        if (prealloc) {
+                em->mod_start = em->start;
+                em->mod_len = em->len;
+        }
+
         free_extent_map(em);
 out:
         write_unlock(&tree->lock);
@@ -298,6 +315,9 @@ int add_extent_mapping(struct extent_map_tree *tree,
         }
         atomic_inc(&em->refs);
 
+        em->mod_start = em->start;
+        em->mod_len = em->len;
+
         try_merge_map(tree, em);
 out:
         return ret;