[PATCH 0/2] btrfs: drop premature generation setting in btrfs_init_new_buffer()

[PATCH 0/2] btrfs: drop premature generation setting in btrfs_init_new_buffer()

[Sun YangKai]

In btrfs_init_new_buffer(), we set the buffer's generation just for the check in
btrfs_clear_buffer_dirty(). So just pass NULL to skip the check and we don't need
to set the generation before erasing the buffer.

No functional change.

See commit messages for details.

Sun YangKai (2):
  btrfs: add comment for btrfs_clear_buffer_dirty
  btrfs: drop premature generation setting in btrfs_init_new_buffer()

 fs/btrfs/extent-tree.c |  6 ++----
 fs/btrfs/extent_io.c   | 20 ++++++++++++++++++++
 2 files changed, 22 insertions(+), 4 deletions(-)

-- 
2.51.2

[PATCH 1/2] btrfs: add comment for btrfs_clear_buffer_dirty

[Sun YangKai]

Signed-off-by: Sun YangKai <sunk67188@gmail.com>
---
 fs/btrfs/extent_io.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)

diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 23273d0e6f22..e80066fc0f02 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -3678,6 +3678,26 @@ static void btree_clear_folio_dirty_tag(struct folio *folio)
 	xa_unlock_irq(&folio->mapping->i_pages);
 }
 
+/*
+ * btrfs_clear_buffer_dirty - Clear the dirty state of an extent buffer
+ * @trans:	Transaction handle, may be NULL.
+ *		If provided, the buffer must belong to the transaction
+ *		(checked via btrfs_header_generation). If the check failed,
+ *		the function returns immediately.
+ * @eb:		The extent buffer to clean.
+ *
+ * This function clears the dirty flag from @eb and updates all accounting
+ * that depends on it (per-CPU counter of dirty metadata bytes, folio dirty
+ * state, address-space tags).
+ *
+ * Special behaviour in zoned mode:
+ *   When the filesystem is zoned (btrfs_is_zoned) the buffer is *not*
+ *   immediately cleaned. Instead the EXTENT_BUFFER_ZONED_ZEROOUT flag is
+ *   set and the buffer remains conceptually dirty.  The physical zero-out
+ *   required by zoned devices is deferred until btree_csum_one_bio(), which
+ *   preserves write-ordering constraints without forcing callers to re-dirty
+ *   the buffer later.
+ */
 void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
 			      struct extent_buffer *eb)
 {
-- 
2.51.2

[PATCH 2/2] btrfs: drop premature generation setting in btrfs_init_new_buffer()

[Sun YangKai]

btrfs_init_new_buffer() sets the header generation only to satisfy
btrfs_clear_buffer_dirty()'s transaction check, then immediately
cleared by memzero_extent_buffer(). This is introduced in
commit cbddcc4fa3443("btrfs: set generation before calling btrfs_clean_tree_block in btrfs_init_new_buffer").

However, after commit c4e54a6571168("btrfs: replace clearing extent buffer dirty bit with btrfs_clean_block")
we can pass NULL instead of the transaction handle to
btrfs_clear_buffer_dirty() so that the generation of the buffer is not
checked. So we can remove the generation assignment that was done only
for that check.

No functional change, just removes a redundant write.

Signed-off-by: Sun YangKai <sunk67188@gmail.com>
---
 fs/btrfs/extent-tree.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index dc4ca98c3780..b22b0aaa99e4 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -5082,9 +5082,6 @@ btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 	    !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state))
 		lockdep_owner = BTRFS_FS_TREE_OBJECTID;
 
-	/* btrfs_clear_buffer_dirty() accesses generation field. */
-	btrfs_set_header_generation(buf, trans->transid);
-
 	/*
 	 * This needs to stay, because we could allocate a freed block from an
 	 * old tree into a new tree, so we need to make sure this new block is
@@ -5093,7 +5090,8 @@ btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 	btrfs_set_buffer_lockdep_class(lockdep_owner, buf, level);
 
 	btrfs_tree_lock_nested(buf, nest);
-	btrfs_clear_buffer_dirty(trans, buf);
+	/* We don't care buf's generation here. */
+	btrfs_clear_buffer_dirty(NULL, buf);
 	clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
 	clear_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &buf->bflags);
 
-- 
2.51.2

Re: [PATCH 0/2] btrfs: drop premature generation setting in btrfs_init_new_buffer()

[Boris Burkov]

On Thu, Nov 20, 2025 at 09:57:02PM +0800, Sun YangKai wrote:
> In btrfs_init_new_buffer(), we set the buffer's generation just for the check in
> btrfs_clear_buffer_dirty(). So just pass NULL to skip the check and we don't need
> to set the generation before erasing the buffer.
> 
> No functional change.
> 
> See commit messages for details.

Looks good, thanks
Reviewed-by: Boris Burkov <boris@bur.io>

> 
> Sun YangKai (2):
>   btrfs: add comment for btrfs_clear_buffer_dirty
>   btrfs: drop premature generation setting in btrfs_init_new_buffer()
> 
>  fs/btrfs/extent-tree.c |  6 ++----
>  fs/btrfs/extent_io.c   | 20 ++++++++++++++++++++
>  2 files changed, 22 insertions(+), 4 deletions(-)
> 
> -- 
> 2.51.2
>