[v2,2/5] btrfs: zoned: don't clear dirty flag of extent buffer

Commit Message

Johannes Thumshirn Nov. 23, 2023, 3:47 p.m. UTC

One a zoned filesystem, never clear the dirty flag of an extent buffer,
but instead mark it as zeroout.

On writeout, when encountering a marked extent_buffer, zero it out.

Reviewed-by: Christoph Hellwig <[e-mail]>
Reviewed-by: Josef Bacik <[e-mail]>
Signed-off-by: Johannes Thumshirn <[e-mail]>
---
 fs/btrfs/disk-io.c   |  7 ++++++-
 fs/btrfs/extent_io.c | 16 ++++++++++++++--
 fs/btrfs/zoned.c     |  3 ++-
 3 files changed, 22 insertions(+), 4 deletions(-)

Patch

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 460b88526f56..9c09062d3d0a 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -254,8 +254,13 @@  blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
 	if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
 		return BLK_STS_IOERR;
 
+	/*
+	 * If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
+	 * checksum it but zero-out its content. This is done to preserve
+	 * ordering of I/O without unnecessarily writing out data.
+	 */
 	if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
-		WARN_ON_ONCE(found_start != 0);
+		memzero_extent_buffer(eb, 0, eb->len);
 		return BLK_STS_OK;
 	}
 
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 49514ef829fb..c378094b5cc8 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -3748,6 +3748,20 @@  void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
 	if (trans && btrfs_header_generation(eb) != trans->transid)
 		return;
 
+	/*
+	 * Instead of clearing the dirty flag off of the buffer, mark it as
+	 * EXTENT_BUFFER_ZONED_ZEROOUT. This allows us to preserve
+	 * write-ordering in zoned mode, without the need to later re-dirty
+	 * the extent_buffer.
+	 *
+	 * The actual zeroout of the buffer will happen later in
+	 * btree_csum_one_bio.
+	 */
+	if (btrfs_is_zoned(fs_info)) {
+		set_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags);
+		return;
+	}
+
 	if (!test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags))
 		return;
 
@@ -4139,8 +4153,6 @@  static void __write_extent_buffer(const struct extent_buffer *eb,
 	/* For unmapped (dummy) ebs, no need to check their uptodate status. */
 	const bool check_uptodate = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
 
-	WARN_ON(test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags));
-
 	if (check_eb_range(eb, start, len))
 		return;
 
diff --git a/fs/btrfs/zoned.c b/fs/btrfs/zoned.c
index b9bfde6fb929..ed8e002b33e7 100644
--- a/fs/btrfs/zoned.c
+++ b/fs/btrfs/zoned.c
@@ -1722,7 +1722,8 @@  void btrfs_redirty_list_add(struct btrfs_transaction *trans,
 	    btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN))
 		return;
 
-	ASSERT(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
+	ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
+	ASSERT(test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags));
 
 	memzero_extent_buffer(eb, 0, eb->len);
 	set_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags);