Just a weekly ping. Any comments appreciated.
On 06/19/2018 04:43 PM, Ari Sundholm wrote:
> This patch series adds a new block driver, blklogwrites, to QEMU. The
> driver is given two block devices: a raw device backed by an image or a
> host block device, and a log device, typically backed by a file, on
> which writes to the raw device are logged.
>
> The logging format used is the same as in the dm-log-writes target of
> the Linux kernel device mapper. The log reflects the writes that have
> been performed on the guest block device and flushed. To be strict, the
> log may contain writes that have not been flushed yet, but they are
> technically outside the bounds of the log until the next flush updates
> the metadata in the log superblock. We believe these semantics to be
> useful even though they may not be completely identical to those of
> dm-log-writes.
>
> This functionality can be used for crash consistency and fs consistency
> testing in filesystem drivers, including on non-Linux guests and older
> guests running Linux versions without support for dm-log-writes. This
> is simple and useful. Admittedly this and more advanced things could
> perhaps be done by extending the quorum driver, but this approach would
> require re-engineering the functionality and involve a more complicated
> setup, so we offer this simple solution which we have found useful
> internally.
>
> In the first patch of this series, two block permission constants are
> moved from block.c to include/block/block.h to make them available
> outside of block.c. The next patch uses these constants.
>
> In patches 2-7, a mechanism to pass the block configuration of a block
> backend to a block driver is introduced, and this mechanism is
> integrated in various hw/block drivers.
>
> In patch 8, the blklogwrites driver is introduced. The driver uses the
> mechanism from the previous patches to align all requests to the
> logical sector size of the underlying block backend of the guest block
> device being logged and use that size as a basis for offsets and sizes
> in the log entries. This means that the sector size in the log metadata
> will be set to the logical sector size, and offsets and sizes of writes
> will be expressed as a multiple of that.
>
> v5:
> - Reorder patches and squash blklogwrites driver patches together
> - Rebase on current upstream master
> - No functional changes
>
> v4:
> - Check return value of snprintf() (flagged by patchew)
> - Don't use C99 for loop initial declaration (flagged by patchew, but
> wasn't QEMU supposed to be C99? Oh well.)
> - Add proper "Since" fields for blklogwrites in block-core.json
> - Rebase on current upstream master
>
> v3:
> - Rebase on current upstream master
>
> v2:
> - Incorporate review feedback
> - Add patches to apply block configurations in more block device
> drivers
>
>
> *** BLURB HERE ***
>
> Aapo Vienamo (1):
> block: Add blklogwrites
>
> Ari Sundholm (7):
> block: Move two block permission constants to the relevant enum
> block: Add a mechanism for passing a block driver a block
> configuration
> hw/scsi/scsi-disk: Always apply block configuration to block driver
> hw/ide/qdev: Always apply block configuration to block driver
> hw/block/virtio-blk: Always apply block configuration to block driver
> hw/block/nvme: Always apply block configuration to block driver
> hw/block/fdc: Always apply block configuration to block driver
>
> MAINTAINERS | 6 +
> block.c | 6 -
> block/Makefile.objs | 1 +
> block/blklogwrites.c | 426 ++++++++++++++++++++++++++++++++++++++++++++++
> block/io.c | 22 +++
> hw/block/block.c | 12 +-
> hw/block/fdc.c | 4 +
> hw/block/nvme.c | 1 +
> hw/block/virtio-blk.c | 1 +
> hw/ide/qdev.c | 2 +
> hw/scsi/scsi-disk.c | 1 +
> include/block/block.h | 17 ++
> include/block/block_int.h | 9 +
> include/hw/block/block.h | 1 +
> qapi/block-core.json | 30 +++-
> 15 files changed, 526 insertions(+), 13 deletions(-)
> create mode 100644 block/blklogwrites.c
>