RE:(2) [PATCH 2/2] ext4: track zeroed out blocks of unwritten extents for fast commit

[Daejun Park]

Hi Jan,

On Wed, Oct 07, 2026 at 15:41:17 +0200, Jan Kara wrote:
> Also putting this into ext4_split_extent_zeroout() looks a bit too easy to
> miss. In fact I think placing ext4_fc_track_range() into
> ext4_issue_zeroout() would make sense because that is where the writing of
> "data" really happens. This will fix the use in
> ext4_split_extent_zeroout() as well as ext4_ext_convert_to_initialized().
> And it will also fix the same class of problem which I think we have in
> ext4_alloc_file_blocks()...

Thanks, that makes sense. In v2, ext4_issue_zeroout() and
ext4_ext_zeroout() take the handle, and the range is tracked there once
the zeroout has succeeded.

ext4_alloc_file_blocks() zeroes out without a handle, so it passes NULL.
As far as I can tell there is no gap there:
ext4_convert_unwritten_extents() right after it tracks the blocks it
converts through ext4_map_blocks(), and blocks that were already written
keep their mapping. I also tried FALLOC_FL_WRITE_ZEROES on scsi_debug
(lbpws=1 lbprz=1) over an unwritten extent, over one block in the middle
of one, and over a hole, each followed by an in-place write, fsync and
EXT4_IOC_SHUTDOWN. The written block survives replay on the base kernel
as well, and the tracepoint shows the range tracked twice, at allocation
and at conversion. If you had a different path in mind, please let me
know.

I'll send v2 shortly.

Daejun