Re: [Qemu-devel] [PATCH v3 1/5] block: split write_zeroes always

[Fam Zheng <famz@redhat.com>]

On Wed, 05/25 21:48, Eric Blake wrote:
> From: "Denis V. Lunev" <den@openvz.org>
> 
> We should split requests even if they are less than write_zeroes_alignment.
> For example we can have the following request:
>   offset 62k
>   size   4k
>   write_zeroes_alignment 64k
> The original code sent 1 request covering 2 qcow2 clusters, and resulted
> in both clusters being allocated. But by splitting the request, we can
> cater to the case where one of the two clusters can be zeroed as a
> whole, for only 1 cluster allocated after the operation.
> 
> Signed-off-by: Denis V. Lunev <den@openvz.org>
> CC: Eric Blake <eblake@redhat.com>
> CC: Kevin Wolf <kwolf@redhat.com>
> Message-Id: <1463476543-3087-2-git-send-email-den@openvz.org>
> 
> [eblake: Avoid exceeding nb_sectors, hoist alignment checks out of
> loop, and update testsuite to show that patch works]
> 
> Signed-off-by: Eric Blake <eblake@redhat.com>
> ---
>  block/io.c                 | 30 +++++++++++++++++-------------
>  tests/qemu-iotests/154.out | 18 ++++++++++++------
>  2 files changed, 29 insertions(+), 19 deletions(-)
> 
> diff --git a/block/io.c b/block/io.c
> index f474b9a..26b5845 100644
> --- a/block/io.c
> +++ b/block/io.c
> @@ -1118,28 +1118,32 @@ static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
>      struct iovec iov = {0};
>      int ret = 0;
>      bool need_flush = false;
> +    int head = 0;
> +    int tail = 0;
> 
>      int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_write_zeroes,
>                                          BDRV_REQUEST_MAX_SECTORS);
> +    if (bs->bl.write_zeroes_alignment) {
> +        assert(is_power_of_2(bs->bl.write_zeroes_alignment));
> +        head = sector_num & (bs->bl.write_zeroes_alignment - 1);
> +        tail = (sector_num + nb_sectors) & (bs->bl.write_zeroes_alignment - 1);
> +        max_write_zeroes &= ~(bs->bl.write_zeroes_alignment - 1);
> +    }
> 
>      while (nb_sectors > 0 && !ret) {
>          int num = nb_sectors;
> 
>          /* Align request.  Block drivers can expect the "bulk" of the request
> -         * to be aligned.
> +         * to be aligned, and that unaligned requests do not cross cluster
> +         * boundaries.
>           */
> -        if (bs->bl.write_zeroes_alignment
> -            && num > bs->bl.write_zeroes_alignment) {
> -            if (sector_num % bs->bl.write_zeroes_alignment != 0) {
> -                /* Make a small request up to the first aligned sector.  */
> -                num = bs->bl.write_zeroes_alignment;
> -                num -= sector_num % bs->bl.write_zeroes_alignment;
> -            } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) {
> -                /* Shorten the request to the last aligned sector.  num cannot
> -                 * underflow because num > bs->bl.write_zeroes_alignment.
> -                 */
> -                num -= (sector_num + num) % bs->bl.write_zeroes_alignment;
> -            }
> +        if (head) {
> +            /* Make a small request up to the first aligned sector.  */
> +            num = MIN(nb_sectors, bs->bl.write_zeroes_alignment - head);
> +            head = 0;
> +        } else if (tail && num > bs->bl.write_zeroes_alignment) {
> +            /* Shorten the request to the last aligned sector.  */
> +            num -= tail;
>          }
> 
>          /* limit request size */
> diff --git a/tests/qemu-iotests/154.out b/tests/qemu-iotests/154.out
> index 8946b73..b9d27c5 100644
> --- a/tests/qemu-iotests/154.out
> +++ b/tests/qemu-iotests/154.out
> @@ -106,11 +106,14 @@ read 1024/1024 bytes at offset 67584
>  read 5120/5120 bytes at offset 68608
>  5 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
>  [{ "start": 0, "length": 32768, "depth": 1, "zero": true, "data": false},
> -{ "start": 32768, "length": 8192, "depth": 0, "zero": false, "data": true, "offset": 20480},
> +{ "start": 32768, "length": 4096, "depth": 0, "zero": false, "data": true, "offset": 20480},
> +{ "start": 36864, "length": 4096, "depth": 0, "zero": true, "data": false},
>  { "start": 40960, "length": 8192, "depth": 1, "zero": true, "data": false},
> -{ "start": 49152, "length": 8192, "depth": 0, "zero": false, "data": true, "offset": 28672},
> +{ "start": 49152, "length": 4096, "depth": 0, "zero": false, "data": true, "offset": 24576},
> +{ "start": 53248, "length": 4096, "depth": 0, "zero": true, "data": false},
>  { "start": 57344, "length": 8192, "depth": 1, "zero": true, "data": false},
> -{ "start": 65536, "length": 8192, "depth": 0, "zero": false, "data": true, "offset": 36864},
> +{ "start": 65536, "length": 4096, "depth": 0, "zero": false, "data": true, "offset": 28672},
> +{ "start": 69632, "length": 4096, "depth": 0, "zero": true, "data": false},
>  { "start": 73728, "length": 134144000, "depth": 1, "zero": true, "data": false}]
> 
>  == spanning two clusters, non-zero after request ==
> @@ -145,11 +148,14 @@ read 7168/7168 bytes at offset 65536
>  read 1024/1024 bytes at offset 72704
>  1 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
>  [{ "start": 0, "length": 32768, "depth": 1, "zero": true, "data": false},
> -{ "start": 32768, "length": 8192, "depth": 0, "zero": false, "data": true, "offset": 20480},
> +{ "start": 32768, "length": 4096, "depth": 0, "zero": true, "data": false},
> +{ "start": 36864, "length": 4096, "depth": 0, "zero": false, "data": true, "offset": 20480},
>  { "start": 40960, "length": 8192, "depth": 1, "zero": true, "data": false},
> -{ "start": 49152, "length": 8192, "depth": 0, "zero": false, "data": true, "offset": 28672},
> +{ "start": 49152, "length": 4096, "depth": 0, "zero": true, "data": false},
> +{ "start": 53248, "length": 4096, "depth": 0, "zero": false, "data": true, "offset": 24576},
>  { "start": 57344, "length": 8192, "depth": 1, "zero": true, "data": false},
> -{ "start": 65536, "length": 8192, "depth": 0, "zero": false, "data": true, "offset": 36864},
> +{ "start": 65536, "length": 4096, "depth": 0, "zero": true, "data": false},
> +{ "start": 69632, "length": 4096, "depth": 0, "zero": false, "data": true, "offset": 28672},
>  { "start": 73728, "length": 134144000, "depth": 1, "zero": true, "data": false}]
> 
>  == spanning two clusters, partially overwriting backing file ==
> -- 
> 2.5.5
> 

Reviewed-by: Fam Zheng <famz@redhat.com>