btrfs: improve btree readahead for full send operations

Currently a full send operation uses the standard btree readahead when
iterating over the subvolume/snapshot btree, which despite bringing good
performance benefits, it could be improved in a few aspects for use cases
such as full send operations, which are guaranteed to visit every node
and leaf of a btree, in ascending and sequential order. The limitations
of that standard btree readahead implementation are the following:

1) It only triggers readahead for leaves that are physically close
   to the leaf being read, within a 64K range;

2) It only triggers readahead for the next or previous leaves if the
   leaf being read is not currently in memory;

3) It never triggers readahead for nodes.

So add a new readahead mode that addresses all these points and use it
for full send operations.

The following test script was used to measure the improvement on a box
using an average, consumer grade, spinning disk and with 16GiB of RAM:

  $ cat test.sh
  #!/bin/bash

  DEV=/dev/sdj
  MNT=/mnt/sdj
  MKFS_OPTIONS="--nodesize 16384"     # default, just to be explicit
  MOUNT_OPTIONS="-o max_inline=2048"  # default, just to be explicit

  mkfs.btrfs -f $MKFS_OPTIONS $DEV > /dev/null
  mount $MOUNT_OPTIONS $DEV $MNT

  # Create files with inline data to make it easier and faster to create
  # large btrees.
  add_files()
  {
      local total=$1
      local start_offset=$2
      local number_jobs=$3
      local total_per_job=$(($total / $number_jobs))

      echo "Creating $total new files using $number_jobs jobs"
      for ((n = 0; n < $number_jobs; n++)); do
          (
              local start_num=$(($start_offset + $n * $total_per_job))
              for ((i = 1; i <= $total_per_job; i++)); do
                  local file_num=$((start_num + $i))
                  local file_path="$MNT/file_${file_num}"
                  xfs_io -f -c "pwrite -S 0xab 0 2000" $file_path > /dev/null
                  if [ $? -ne 0 ]; then
                      echo "Failed creating file $file_path"
                      break
                  fi
              done
          ) &
          worker_pids[$n]=$!
      done

      wait ${worker_pids[@]}

      sync
      echo
      echo "btree node/leaf count: $(btrfs inspect-internal dump-tree -t 5 $DEV | egrep '^(node|leaf) ' | wc -l)"
  }

  initial_file_count=500000
  add_files $initial_file_count 0 4

  echo
  echo "Creating first snapshot..."
  btrfs subvolume snapshot -r $MNT $MNT/snap1

  echo
  echo "Adding more files..."
  add_files $((initial_file_count / 4)) $initial_file_count 4

  echo
  echo "Updating 1/50th of the initial files..."
  for ((i = 1; i < $initial_file_count; i += 50)); do
      xfs_io -c "pwrite -S 0xcd 0 20" $MNT/file_$i > /dev/null
  done

  echo
  echo "Creating second snapshot..."
  btrfs subvolume snapshot -r $MNT $MNT/snap2

  umount $MNT

  echo 3 > /proc/sys/vm/drop_caches
  blockdev --flushbufs $DEV &> /dev/null
  hdparm -F $DEV &> /dev/null

  mount $MOUNT_OPTIONS $DEV $MNT

  echo
  echo "Testing full send..."
  start=$(date +%s)
  btrfs send $MNT/snap1 > /dev/null
  end=$(date +%s)
  echo
  echo "Full send took $((end - start)) seconds"

  umount $MNT

The durations of the full send operation in seconds were the following:

Before this change:  217 seconds
After this change:   205 seconds (-5.7%)

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

1 files changed, 24 insertions, 4 deletions

diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index 26c2d50ea2db..a484fb72a01f 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -1279,12 +1279,13 @@ static void reada_for_search(struct btrfs_fs_info *fs_info,
 	u64 search;
 	u64 target;
 	u64 nread = 0;
+	u64 nread_max;
 	struct extent_buffer *eb;
 	u32 nr;
 	u32 blocksize;
 	u32 nscan = 0;
 
-	if (level != 1)
+	if (level != 1 && path->reada != READA_FORWARD_ALWAYS)
 		return;
 
 	if (!path->nodes[level])
@@ -1292,6 +1293,20 @@ static void reada_for_search(struct btrfs_fs_info *fs_info,
 
 	node = path->nodes[level];
 
+	/*
+	 * Since the time between visiting leaves is much shorter than the time
+	 * between visiting nodes, limit read ahead of nodes to 1, to avoid too
+	 * much IO at once (possibly random).
+	 */
+	if (path->reada == READA_FORWARD_ALWAYS) {
+		if (level > 1)
+			nread_max = node->fs_info->nodesize;
+		else
+			nread_max = SZ_128K;
+	} else {
+		nread_max = SZ_64K;
+	}
+
 	search = btrfs_node_blockptr(node, slot);
 	blocksize = fs_info->nodesize;
 	eb = find_extent_buffer(fs_info, search);
@@ -1310,7 +1325,8 @@ static void reada_for_search(struct btrfs_fs_info *fs_info,
 			if (nr == 0)
 				break;
 			nr--;
-		} else if (path->reada == READA_FORWARD) {
+		} else if (path->reada == READA_FORWARD ||
+			   path->reada == READA_FORWARD_ALWAYS) {
 			nr++;
 			if (nr >= nritems)
 				break;
@@ -1321,13 +1337,14 @@ static void reada_for_search(struct btrfs_fs_info *fs_info,
 				break;
 		}
 		search = btrfs_node_blockptr(node, nr);
-		if ((search <= target && target - search <= 65536) ||
+		if (path->reada == READA_FORWARD_ALWAYS ||
+		    (search <= target && target - search <= 65536) ||
 		    (search > target && search - target <= 65536)) {
 			btrfs_readahead_node_child(node, nr);
 			nread += blocksize;
 		}
 		nscan++;
-		if ((nread > 65536 || nscan > 32))
+		if (nread > nread_max || nscan > 32)
 			break;
 	}
 }
@@ -1436,6 +1453,9 @@ read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
 
 	tmp = find_extent_buffer(fs_info, blocknr);
 	if (tmp) {
+		if (p->reada == READA_FORWARD_ALWAYS)
+			reada_for_search(fs_info, p, level, slot, key->objectid);
+
 		/* first we do an atomic uptodate check */
 		if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
 			/*