On 9/29/26 17:51, Denis V. Lunev wrote:
> This email originated from an IP that might not be authorized by the domain
> it was sent from.
> Do not click links or open attachments unless it is an email you expected to
> receive.
> Backup and commit re-query the source block status for every task and
> size a task by the copy buffer, so a long run of zeroes turns into a
> crowd of small write-zeroes requests. This series reuses what the
> up-front scan already learned and lets one write-zeroes task cover a
> whole run.
>
> Backup of a 16G qcow2 image holding 1G of data, the rest never
> allocated, sync=full to a raw target on ext4:
>
> tasks time
> before 16384 1.6s
> after 1088 1.0s
>
> Such an image is the normal case rather than a corner one: a guest with
> discard enabled on a disk which is mostly free leaves exactly this
> shape behind.
>
> v3, all from Andrey's review of v2:
> - 3/9: the commit message says which permission stops whom. A device or
> an export asks for BLK_PERM_CONSISTENT_READ along with BLK_PERM_WRITE
> and the job withholds CONSISTENT_READ above base_overlay, having to
> share WRITE or it would block its own writes to base; a job target
> asks for WRITE alone and is stopped by the op blocker instead.
> - 4/9: assertNotIn() rather than a bare assert, and test_bitmap_straddle
> checks the target map, not only the content.
> - 9/9: zero_widen is a bool again. A task widened before another task's
> request was refused reached the write loop with the chunk still at its
> full widened size, so it wrote the whole range as one request with no
> BDRV_REQ_NO_FALLBACK, against a target which had just said it cannot
> zero by metadata. The chunk is now bounded whichever way the flag was
> read, and a request carries NO_FALLBACK whenever the target is still
> believed to oblige.
> block_copy_chunk_size() is called under s->lock, as its own comment
> asks for.
> The commit message no longer says a refused request wrote nothing:
> bdrv_co_do_pwrite_zeroes() fragments by bl.max_pwrite_zeroes and a
> driver may refuse a later fragment after an earlier one landed. The
> range is rewritten whole, which is safe because zeroes over zeroes
> change nothing.
> - Reviewed-by tags collected on 1-8.
> - rebased on master
>
> v2, all from Andrey's review:
> - 2/9: holes spelled out in both layouts, and zero runs added, so the
> write-zeroes path of 3/9 is covered too
> - 3/9: COMMIT_ZERO_CHUNK has a comment of its own saying what bounds it;
> the cache check drops its dead half and asserts instead
> - 4/9: a Case namedtuple pairs each size with its layout, and
> create_image(), write_layout(), dirty_layout() and backup_and_check()
> take out the duplication
> - 8/9: g_assert_not_reached() for the sync mode which cannot reach there
> - 9/9: a widened write-zeroes request only pays off where the target
> zeroes by metadata. supported_zero_flags rules out the targets which
> cannot, and the first widened request asks for BDRV_REQ_NO_FALLBACK to
> settle the rest, since a driver may advertise it and only learn better
> from a failing call. A target which would write the zeroes out fails
> that request without writing, and the run keeps its requests at the
> buffer chunk size from there on. With the size question settled that
> way the cap became BDRV_REQUEST_MAX_BYTES rather than 256M.
> - rebased on master
>
> Signed-off-by: Denis V. Lunev <[email protected]>
> CC: Vladimir Sementsov-Ogievskiy <[email protected]>
> CC: John Snow <[email protected]>
> CC: Andrey Drobyshev <[email protected]>
>
> Denis V. Lunev (9):
> block/commit: pass BDRV_WANT_PRECISE to block-status
> iotests/040: cover large and fragmented commit runs
> block/commit: batch block-status queries
> iotests/124: cover backup of zero clusters and holes
> block/block-copy: don't reserve memory for zero tasks
> block/block-copy: extract block_copy_set_task_method()
> block/block-copy: track known-zero source clusters
> block/backup: pre-fill zero_bitmap for full/bitmap
> block/block-copy: coalesce write-zeroes tasks
>
> block/backup.c | 93 ++++++----
> block/block-copy.c | 300 ++++++++++++++++++++++++++++----
> block/commit.c | 58 +++++--
> include/block/block-copy.h | 5 +
> tests/qemu-iotests/040 | 102 ++++++++++-
> tests/qemu-iotests/040.out | 4 +-
> tests/qemu-iotests/124 | 347 ++++++++++++++++++++++++++++++++++++-
> tests/qemu-iotests/124.out | 4 +-
> 8 files changed, 825 insertions(+), 88 deletions(-)
>
>
> base-commit: f8296b816fabd370307cd22b0270b610fc0fa279
ping
