Btrfs: incompatible format change to remove hole extents

Btrfs has always had these filler extent data items for holes in inodes.  This
has made somethings very easy, like logging hole punches and sending hole
punches.  However for large holey files these extent data items are pure
overhead.  So add an incompatible feature to no longer add hole extents to
reduce the amount of metadata used by these sort of files.  This has a few
changes for logging and send obviously since they will need to detect holes and
log/send the holes if there are any.  I've tested this thoroughly with xfstests
and it doesn't cause any issues with and without the incompat format set.
Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <clm@fb.com>

1 files changed, 162 insertions, 10 deletions

diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
index 9f7fc51ca334..e7d7a837512a 100644
--- a/fs/btrfs/tree-log.c
+++ b/fs/btrfs/tree-log.c
@@ -3194,7 +3194,7 @@ static int log_inode_item(struct btrfs_trans_handle *trans,
 static noinline int copy_items(struct btrfs_trans_handle *trans,
 			       struct inode *inode,
 			       struct btrfs_path *dst_path,
-			       struct extent_buffer *src,
+			       struct btrfs_path *src_path, u64 *last_extent,
 			       int start_slot, int nr, int inode_only)
 {
 	unsigned long src_offset;
@@ -3202,6 +3202,8 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 	struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
 	struct btrfs_file_extent_item *extent;
 	struct btrfs_inode_item *inode_item;
+	struct extent_buffer *src = src_path->nodes[0];
+	struct btrfs_key first_key, last_key, key;
 	int ret;
 	struct btrfs_key *ins_keys;
 	u32 *ins_sizes;
@@ -3209,6 +3211,9 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 	int i;
 	struct list_head ordered_sums;
 	int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
+	bool has_extents = false;
+	bool need_find_last_extent = (*last_extent == 0);
+	bool done = false;
 
 	INIT_LIST_HEAD(&ordered_sums);
 
@@ -3217,6 +3222,8 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 	if (!ins_data)
 		return -ENOMEM;
 
+	first_key.objectid = (u64)-1;
+
 	ins_sizes = (u32 *)ins_data;
 	ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
 
@@ -3237,6 +3244,9 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 
 		src_offset = btrfs_item_ptr_offset(src, start_slot + i);
 
+		if ((i == (nr - 1)))
+			last_key = ins_keys[i];
+
 		if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
 			inode_item = btrfs_item_ptr(dst_path->nodes[0],
 						    dst_path->slots[0],
@@ -3248,6 +3258,21 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 					   src_offset, ins_sizes[i]);
 		}
 
+		/*
+		 * We set need_find_last_extent here in case we know we were
+		 * processing other items and then walk into the first extent in
+		 * the inode.  If we don't hit an extent then nothing changes,
+		 * we'll do the last search the next time around.
+		 */
+		if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
+			has_extents = true;
+			if (need_find_last_extent &&
+			    first_key.objectid == (u64)-1)
+				first_key = ins_keys[i];
+		} else {
+			need_find_last_extent = false;
+		}
+
 		/* take a reference on file data extents so that truncates
 		 * or deletes of this inode don't have to relog the inode
 		 * again
@@ -3312,6 +3337,126 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 		list_del(&sums->list);
 		kfree(sums);
 	}
+
+	if (!has_extents)
+		return ret;
+
+	/*
+	 * Because we use btrfs_search_forward we could skip leaves that were
+	 * not modified and then assume *last_extent is valid when it really
+	 * isn't.  So back up to the previous leaf and read the end of the last
+	 * extent before we go and fill in holes.
+	 */
+	if (need_find_last_extent) {
+		u64 len;
+
+		ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path);
+		if (ret < 0)
+			return ret;
+		if (ret)
+			goto fill_holes;
+		if (src_path->slots[0])
+			src_path->slots[0]--;
+		src = src_path->nodes[0];
+		btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
+		if (key.objectid != btrfs_ino(inode) ||
+		    key.type != BTRFS_EXTENT_DATA_KEY)
+			goto fill_holes;
+		extent = btrfs_item_ptr(src, src_path->slots[0],
+					struct btrfs_file_extent_item);
+		if (btrfs_file_extent_type(src, extent) ==
+		    BTRFS_FILE_EXTENT_INLINE) {
+			len = btrfs_file_extent_inline_len(src, extent);
+			*last_extent = ALIGN(key.offset + len,
+					     log->sectorsize);
+		} else {
+			len = btrfs_file_extent_num_bytes(src, extent);
+			*last_extent = key.offset + len;
+		}
+	}
+fill_holes:
+	/* So we did prev_leaf, now we need to move to the next leaf, but a few
+	 * things could have happened
+	 *
+	 * 1) A merge could have happened, so we could currently be on a leaf
+	 * that holds what we were copying in the first place.
+	 * 2) A split could have happened, and now not all of the items we want
+	 * are on the same leaf.
+	 *
+	 * So we need to adjust how we search for holes, we need to drop the
+	 * path and re-search for the first extent key we found, and then walk
+	 * forward until we hit the last one we copied.
+	 */
+	if (need_find_last_extent) {
+		/* btrfs_prev_leaf could return 1 without releasing the path */
+		btrfs_release_path(src_path);
+		ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key,
+					src_path, 0, 0);
+		if (ret < 0)
+			return ret;
+		ASSERT(ret == 0);
+		src = src_path->nodes[0];
+		i = src_path->slots[0];
+	} else {
+		i = start_slot;
+	}
+
+	/*
+	 * Ok so here we need to go through and fill in any holes we may have
+	 * to make sure that holes are punched for those areas in case they had
+	 * extents previously.
+	 */
+	while (!done) {
+		u64 offset, len;
+		u64 extent_end;
+
+		if (i >= btrfs_header_nritems(src_path->nodes[0])) {
+			ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path);
+			if (ret < 0)
+				return ret;
+			ASSERT(ret == 0);
+			src = src_path->nodes[0];
+			i = 0;
+		}
+
+		btrfs_item_key_to_cpu(src, &key, i);
+		if (!btrfs_comp_cpu_keys(&key, &last_key))
+			done = true;
+		if (key.objectid != btrfs_ino(inode) ||
+		    key.type != BTRFS_EXTENT_DATA_KEY) {
+			i++;
+			continue;
+		}
+		extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
+		if (btrfs_file_extent_type(src, extent) ==
+		    BTRFS_FILE_EXTENT_INLINE) {
+			len = btrfs_file_extent_inline_len(src, extent);
+			extent_end = ALIGN(key.offset + len, log->sectorsize);
+		} else {
+			len = btrfs_file_extent_num_bytes(src, extent);
+			extent_end = key.offset + len;
+		}
+		i++;
+
+		if (*last_extent == key.offset) {
+			*last_extent = extent_end;
+			continue;
+		}
+		offset = *last_extent;
+		len = key.offset - *last_extent;
+		ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
+					       offset, 0, 0, len, 0, len, 0,
+					       0, 0);
+		if (ret)
+			break;
+		*last_extent = offset + len;
+	}
+	/*
+	 * Need to let the callers know we dropped the path so they should
+	 * re-search.
+	 */
+	if (!ret && need_find_last_extent)
+		ret = 1;
 	return ret;
 }
 
@@ -3630,6 +3775,7 @@ static int btrfs_log_inode(struct btrfs_trans_handle *trans,
 	struct btrfs_key max_key;
 	struct btrfs_root *log = root->log_root;
 	struct extent_buffer *src = NULL;
+	u64 last_extent = 0;
 	int err = 0;
 	int ret;
 	int nritems;
@@ -3745,11 +3891,15 @@ again:
 			goto next_slot;
 		}
 
-		ret = copy_items(trans, inode, dst_path, src, ins_start_slot,
-				 ins_nr, inode_only);
-		if (ret) {
+		ret = copy_items(trans, inode, dst_path, path, &last_extent,
+				 ins_start_slot, ins_nr, inode_only);
+		if (ret < 0) {
 			err = ret;
 			goto out_unlock;
+		} if (ret) {
+			ins_nr = 0;
+			btrfs_release_path(path);
+			continue;
 		}
 		ins_nr = 1;
 		ins_start_slot = path->slots[0];
@@ -3763,13 +3913,14 @@ next_slot:
 			goto again;
 		}
 		if (ins_nr) {
-			ret = copy_items(trans, inode, dst_path, src,
-					 ins_start_slot,
+			ret = copy_items(trans, inode, dst_path, path,
+					 &last_extent, ins_start_slot,
 					 ins_nr, inode_only);
-			if (ret) {
+			if (ret < 0) {
 				err = ret;
 				goto out_unlock;
 			}
+			ret = 0;
 			ins_nr = 0;
 		}
 		btrfs_release_path(path);
@@ -3784,12 +3935,13 @@ next_slot:
 		}
 	}
 	if (ins_nr) {
-		ret = copy_items(trans, inode, dst_path, src, ins_start_slot,
-				 ins_nr, inode_only);
-		if (ret) {
+		ret = copy_items(trans, inode, dst_path, path, &last_extent,
+				 ins_start_slot, ins_nr, inode_only);
+		if (ret < 0) {
 			err = ret;
 			goto out_unlock;
 		}
+		ret = 0;
 		ins_nr = 0;
 	}