[02/11] block/io: add bdrv_co_write_compressed
diff mbox

Message ID 1464686130-12265-3-git-send-email-den@openvz.org
State New
Headers show

Commit Message

Denis V. Lunev May 31, 2016, 9:15 a.m. UTC
From: Pavel Butsykin <pbutsykin@virtuozzo.com>

This patch just adds the interface to the bdrv_co_pwritev_compressed,
which is currently not used but will be useful for safe implementation of the
bdrv_co_write_compressed callback in format drivers.

Signed-off-by: Pavel Butsykin <pbutsykin@virtuozzo.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
CC: Jeff Cody <jcody@redhat.com>
CC: Markus Armbruster <armbru@redhat.com>
CC: Eric Blake <eblake@redhat.com>
CC: John Snow <jsnow@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
CC: Kevin Wolf <kwolf@redhat.com>
---
 block/io.c                | 78 +++++++++++++++++++++++++++++++++++++++++++----
 include/block/block_int.h |  5 +++
 qemu-img.c                |  2 +-
 3 files changed, 78 insertions(+), 7 deletions(-)

Comments

Stefan Hajnoczi June 13, 2016, 1:18 p.m. UTC | #1
On Tue, May 31, 2016 at 12:15:21PM +0300, Denis V. Lunev wrote:
> From: Pavel Butsykin <pbutsykin@virtuozzo.com>
> 
> This patch just adds the interface to the bdrv_co_pwritev_compressed,
> which is currently not used but will be useful for safe implementation of the
> bdrv_co_write_compressed callback in format drivers.
> 
> Signed-off-by: Pavel Butsykin <pbutsykin@virtuozzo.com>
> Signed-off-by: Denis V. Lunev <den@openvz.org>
> CC: Jeff Cody <jcody@redhat.com>
> CC: Markus Armbruster <armbru@redhat.com>
> CC: Eric Blake <eblake@redhat.com>
> CC: John Snow <jsnow@redhat.com>
> CC: Stefan Hajnoczi <stefanha@redhat.com>
> CC: Kevin Wolf <kwolf@redhat.com>
> ---
>  block/io.c                | 78 +++++++++++++++++++++++++++++++++++++++++++----
>  include/block/block_int.h |  5 +++
>  qemu-img.c                |  2 +-
>  3 files changed, 78 insertions(+), 7 deletions(-)

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Eric Blake June 13, 2016, 2:32 p.m. UTC | #2
On 05/31/2016 03:15 AM, Denis V. Lunev wrote:
> From: Pavel Butsykin <pbutsykin@virtuozzo.com>
> 
> This patch just adds the interface to the bdrv_co_pwritev_compressed,
> which is currently not used but will be useful for safe implementation of the
> bdrv_co_write_compressed callback in format drivers.
> 
> Signed-off-by: Pavel Butsykin <pbutsykin@virtuozzo.com>
> Signed-off-by: Denis V. Lunev <den@openvz.org>
> CC: Jeff Cody <jcody@redhat.com>
> CC: Markus Armbruster <armbru@redhat.com>
> CC: Eric Blake <eblake@redhat.com>
> CC: John Snow <jsnow@redhat.com>
> CC: Stefan Hajnoczi <stefanha@redhat.com>
> CC: Kevin Wolf <kwolf@redhat.com>
> ---
>  block/io.c                | 78 +++++++++++++++++++++++++++++++++++++++++++----
>  include/block/block_int.h |  5 +++
>  qemu-img.c                |  2 +-
>  3 files changed, 78 insertions(+), 7 deletions(-)
> 
> diff --git a/block/io.c b/block/io.c
> index c5bb6ae..54cd9a4 100644
> --- a/block/io.c
> +++ b/block/io.c
> @@ -1779,8 +1779,8 @@ int bdrv_is_allocated_above(BlockDriverState *top,
>      return 0;
>  }
>  
> -int bdrv_pwrite_compressed(BlockDriverState *bs, int64_t offset,
> -                           const void *buf, int count)
> +int coroutine_fn bdrv_co_pwritev_compressed(BlockDriverState *bs,
> +    int64_t offset, unsigned int bytes, QEMUIOVector *qiov)

Why the rename s/count/bytes/?  Would it be better to get the name right
in 1/11?

>  {
>      BlockDriver *drv = bs->drv;
>      int ret;
> @@ -1788,18 +1788,84 @@ int bdrv_pwrite_compressed(BlockDriverState *bs, int64_t offset,
>      if (!drv) {
>          return -ENOMEDIUM;
>      }
> -    if (!drv->bdrv_write_compressed) {
> +
> +    if (!drv->bdrv_co_write_compressed) {
>          return -ENOTSUP;

This is a (temporary) regression - since none of the drivers have
.bdrv_co_write_compressed yet, you will always fail.  Rather, the
transition period should support both interfaces at once...

>      }
> -    ret = bdrv_check_byte_request(bs, offset, count);
> +
> +    ret = bdrv_check_byte_request(bs, offset, bytes);
>      if (ret < 0) {
>          return ret;
>      }
>  
>      assert(QLIST_EMPTY(&bs->dirty_bitmaps));
> +    assert(qemu_in_coroutine());
> +
> +    return drv->bdrv_co_write_compressed(bs, offset >> BDRV_SECTOR_BITS,
> +                                         bytes >> BDRV_SECTOR_BITS, qiov);

...and call into either the old or the new interface according to what
is present.


> +int bdrv_pwrite_compressed(BlockDriverState *bs, int64_t offset,
> +                           const void *buf, int count)
> +{
> +    BdrvWriteCompressedCo data;
> +    QEMUIOVector qiov;
> +    BlockDriver *drv = bs->drv;
> +    struct iovec iov = {
> +        .iov_base = (void *)buf,
> +        .iov_len = count,
> +    };
> +    qemu_iovec_init_external(&qiov, &iov, 1);
>  
> -    return drv->bdrv_write_compressed(bs, offset >> BDRV_SECTOR_BITS, buf,
> -                                      count >> BDRV_SECTOR_BITS);
> +    data = (BdrvWriteCompressedCo) {
> +        .bs     = bs,
> +        .offset = offset,
> +        .qiov   = &qiov,
> +        .ret    = -EINPROGRESS,
> +    };
> +
> +    if (!drv) {
> +        return -ENOMEDIUM;
> +    }
> +
> +    if (drv->bdrv_write_compressed) {
> +        int ret = bdrv_check_byte_request(bs, offset, count);
> +        if (ret < 0) {
> +            return ret;
> +        }
> +        assert(QLIST_EMPTY(&bs->dirty_bitmaps));
> +        return drv->bdrv_write_compressed(bs, offset >> BDRV_SECTOR_BITS, buf,
> +                                          count >> BDRV_SECTOR_BITS);
> +    }

Oh, you're catering to the old code up front, without a coroutine, and
only the new code gets coroutine treatment.  Maybe it's okay after all.

> +
> +    if (qemu_in_coroutine()) {
> +        /* Fast-path if already in coroutine context */
> +        bdrv_write_compressed_co_entry(&data);
> +    } else {
> +        AioContext *aio_context = bdrv_get_aio_context(bs);
> +
> +        Coroutine *co = qemu_coroutine_create(bdrv_write_compressed_co_entry);
> +        qemu_coroutine_enter(co, &data);
> +        while (data.ret == -EINPROGRESS) {
> +            aio_poll(aio_context, true);
> +        }
> +    }
> +    return data.ret;
>  }
>  
>  int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
> diff --git a/include/block/block_int.h b/include/block/block_int.h
> index 30a9717..ccba9c9 100644
> --- a/include/block/block_int.h
> +++ b/include/block/block_int.h
> @@ -207,6 +207,9 @@ struct BlockDriver {
>      int (*bdrv_write_compressed)(BlockDriverState *bs, int64_t sector_num,
>                                   const uint8_t *buf, int nb_sectors);
>  
> +    int coroutine_fn (*bdrv_co_write_compressed)(BlockDriverState *bs,

Might be better to name this bdrv_co_pwrite_compressed if we want to
make it byte-based...

> +        int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);

...it seems odd that you have to add a new sector-based interface given
that you are trying to convert to byte-based.

> +
>      int (*bdrv_snapshot_create)(BlockDriverState *bs,
>                                  QEMUSnapshotInfo *sn_info);
>      int (*bdrv_snapshot_goto)(BlockDriverState *bs,
> @@ -535,6 +538,8 @@ int coroutine_fn bdrv_co_preadv(BlockDriverState *bs,
>  int coroutine_fn bdrv_co_pwritev(BlockDriverState *bs,
>      int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
>      BdrvRequestFlags flags);
> +int coroutine_fn bdrv_co_pwritev_compressed(BlockDriverState *bs,
> +    int64_t offset, unsigned int bytes, QEMUIOVector *qiov);
>  
>  int get_tmp_filename(char *filename, int size);
>  BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
> diff --git a/qemu-img.c b/qemu-img.c
> index eb744d4..ab54027 100644
> --- a/qemu-img.c
> +++ b/qemu-img.c
> @@ -2024,7 +2024,7 @@ static int img_convert(int argc, char **argv)
>          const char *preallocation =
>              qemu_opt_get(opts, BLOCK_OPT_PREALLOC);
>  
> -        if (!drv->bdrv_write_compressed) {
> +        if (!drv->bdrv_write_compressed && !drv->bdrv_co_write_compressed) {
>              error_report("Compression not supported for this file format");
>              ret = -1;
>              goto out;
>
Pavel Butsykin June 22, 2016, 12:26 p.m. UTC | #3
On 13.06.2016 17:32, Eric Blake wrote:
> On 05/31/2016 03:15 AM, Denis V. Lunev wrote:
>> From: Pavel Butsykin <pbutsykin@virtuozzo.com>
>>
>> This patch just adds the interface to the bdrv_co_pwritev_compressed,
>> which is currently not used but will be useful for safe implementation of the
>> bdrv_co_write_compressed callback in format drivers.
>>
>> Signed-off-by: Pavel Butsykin <pbutsykin@virtuozzo.com>
>> Signed-off-by: Denis V. Lunev <den@openvz.org>
>> CC: Jeff Cody <jcody@redhat.com>
>> CC: Markus Armbruster <armbru@redhat.com>
>> CC: Eric Blake <eblake@redhat.com>
>> CC: John Snow <jsnow@redhat.com>
>> CC: Stefan Hajnoczi <stefanha@redhat.com>
>> CC: Kevin Wolf <kwolf@redhat.com>
>> ---
>>   block/io.c                | 78 +++++++++++++++++++++++++++++++++++++++++++----
>>   include/block/block_int.h |  5 +++
>>   qemu-img.c                |  2 +-
>>   3 files changed, 78 insertions(+), 7 deletions(-)
>>
>> diff --git a/block/io.c b/block/io.c
>> index c5bb6ae..54cd9a4 100644
>> --- a/block/io.c
>> +++ b/block/io.c
>> @@ -1779,8 +1779,8 @@ int bdrv_is_allocated_above(BlockDriverState *top,
>>       return 0;
>>   }
>>
>> -int bdrv_pwrite_compressed(BlockDriverState *bs, int64_t offset,
>> -                           const void *buf, int count)
>> +int coroutine_fn bdrv_co_pwritev_compressed(BlockDriverState *bs,
>> +    int64_t offset, unsigned int bytes, QEMUIOVector *qiov)
>
> Why the rename s/count/bytes/?  Would it be better to get the name right
> in 1/11?

It made like the bdrv_co_pwritev interface. For the
bdrv_pwrite_compressed() func also need rename s/count/bytes/ in 1/11.

>>   {
>>       BlockDriver *drv = bs->drv;
>>       int ret;
>> @@ -1788,18 +1788,84 @@ int bdrv_pwrite_compressed(BlockDriverState *bs, int64_t offset,
>>       if (!drv) {
>>           return -ENOMEDIUM;
>>       }
>> -    if (!drv->bdrv_write_compressed) {
>> +
>> +    if (!drv->bdrv_co_write_compressed) {
>>           return -ENOTSUP;
>
> This is a (temporary) regression - since none of the drivers have
> .bdrv_co_write_compressed yet, you will always fail.  Rather, the
> transition period should support both interfaces at once...
>
>>       }
>> -    ret = bdrv_check_byte_request(bs, offset, count);
>> +
>> +    ret = bdrv_check_byte_request(bs, offset, bytes);
>>       if (ret < 0) {
>>           return ret;
>>       }
>>
>>       assert(QLIST_EMPTY(&bs->dirty_bitmaps));
>> +    assert(qemu_in_coroutine());
>> +
>> +    return drv->bdrv_co_write_compressed(bs, offset >> BDRV_SECTOR_BITS,
>> +                                         bytes >> BDRV_SECTOR_BITS, qiov);
>
> ...and call into either the old or the new interface according to what
> is present.
>
>
>> +int bdrv_pwrite_compressed(BlockDriverState *bs, int64_t offset,
>> +                           const void *buf, int count)
>> +{
>> +    BdrvWriteCompressedCo data;
>> +    QEMUIOVector qiov;
>> +    BlockDriver *drv = bs->drv;
>> +    struct iovec iov = {
>> +        .iov_base = (void *)buf,
>> +        .iov_len = count,
>> +    };
>> +    qemu_iovec_init_external(&qiov, &iov, 1);
>>
>> -    return drv->bdrv_write_compressed(bs, offset >> BDRV_SECTOR_BITS, buf,
>> -                                      count >> BDRV_SECTOR_BITS);
>> +    data = (BdrvWriteCompressedCo) {
>> +        .bs     = bs,
>> +        .offset = offset,
>> +        .qiov   = &qiov,
>> +        .ret    = -EINPROGRESS,
>> +    };
>> +
>> +    if (!drv) {
>> +        return -ENOMEDIUM;
>> +    }
>> +
>> +    if (drv->bdrv_write_compressed) {
>> +        int ret = bdrv_check_byte_request(bs, offset, count);
>> +        if (ret < 0) {
>> +            return ret;
>> +        }
>> +        assert(QLIST_EMPTY(&bs->dirty_bitmaps));
>> +        return drv->bdrv_write_compressed(bs, offset >> BDRV_SECTOR_BITS, buf,
>> +                                          count >> BDRV_SECTOR_BITS);
>> +    }
>
> Oh, you're catering to the old code up front, without a coroutine, and
> only the new code gets coroutine treatment.  Maybe it's okay after all.

Yes, it is made for a gradual transition from
drv->bdrv_write_compressed to drv->bdrv_co_write_compressed.

>> +
>> +    if (qemu_in_coroutine()) {
>> +        /* Fast-path if already in coroutine context */
>> +        bdrv_write_compressed_co_entry(&data);
>> +    } else {
>> +        AioContext *aio_context = bdrv_get_aio_context(bs);
>> +
>> +        Coroutine *co = qemu_coroutine_create(bdrv_write_compressed_co_entry);
>> +        qemu_coroutine_enter(co, &data);
>> +        while (data.ret == -EINPROGRESS) {
>> +            aio_poll(aio_context, true);
>> +        }
>> +    }
>> +    return data.ret;
>>   }
>>
>>   int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
>> diff --git a/include/block/block_int.h b/include/block/block_int.h
>> index 30a9717..ccba9c9 100644
>> --- a/include/block/block_int.h
>> +++ b/include/block/block_int.h
>> @@ -207,6 +207,9 @@ struct BlockDriver {
>>       int (*bdrv_write_compressed)(BlockDriverState *bs, int64_t sector_num,
>>                                    const uint8_t *buf, int nb_sectors);
>>
>> +    int coroutine_fn (*bdrv_co_write_compressed)(BlockDriverState *bs,
>
> Might be better to name this bdrv_co_pwrite_compressed if we want to
> make it byte-based...
>
>> +        int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
>
> ...it seems odd that you have to add a new sector-based interface given
> that you are trying to convert to byte-based.
>

When I sent this patch, Kevin had not yet converted the sector-based
interfaces to byte-based for format drivers :)

>> +
>>       int (*bdrv_snapshot_create)(BlockDriverState *bs,
>>                                   QEMUSnapshotInfo *sn_info);
>>       int (*bdrv_snapshot_goto)(BlockDriverState *bs,
>> @@ -535,6 +538,8 @@ int coroutine_fn bdrv_co_preadv(BlockDriverState *bs,
>>   int coroutine_fn bdrv_co_pwritev(BlockDriverState *bs,
>>       int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
>>       BdrvRequestFlags flags);
>> +int coroutine_fn bdrv_co_pwritev_compressed(BlockDriverState *bs,
>> +    int64_t offset, unsigned int bytes, QEMUIOVector *qiov);
>>
>>   int get_tmp_filename(char *filename, int size);
>>   BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
>> diff --git a/qemu-img.c b/qemu-img.c
>> index eb744d4..ab54027 100644
>> --- a/qemu-img.c
>> +++ b/qemu-img.c
>> @@ -2024,7 +2024,7 @@ static int img_convert(int argc, char **argv)
>>           const char *preallocation =
>>               qemu_opt_get(opts, BLOCK_OPT_PREALLOC);
>>
>> -        if (!drv->bdrv_write_compressed) {
>> +        if (!drv->bdrv_write_compressed && !drv->bdrv_co_write_compressed) {
>>               error_report("Compression not supported for this file format");
>>               ret = -1;
>>               goto out;
>>
>
Kevin Wolf June 28, 2016, 11:09 a.m. UTC | #4
Am 31.05.2016 um 11:15 hat Denis V. Lunev geschrieben:
> From: Pavel Butsykin <pbutsykin@virtuozzo.com>
> 
> This patch just adds the interface to the bdrv_co_pwritev_compressed,
> which is currently not used but will be useful for safe implementation of the
> bdrv_co_write_compressed callback in format drivers.
> 
> Signed-off-by: Pavel Butsykin <pbutsykin@virtuozzo.com>
> Signed-off-by: Denis V. Lunev <den@openvz.org>
> CC: Jeff Cody <jcody@redhat.com>
> CC: Markus Armbruster <armbru@redhat.com>
> CC: Eric Blake <eblake@redhat.com>
> CC: John Snow <jsnow@redhat.com>
> CC: Stefan Hajnoczi <stefanha@redhat.com>
> CC: Kevin Wolf <kwolf@redhat.com>
> ---
>  block/io.c                | 78 +++++++++++++++++++++++++++++++++++++++++++----
>  include/block/block_int.h |  5 +++
>  qemu-img.c                |  2 +-
>  3 files changed, 78 insertions(+), 7 deletions(-)
> 
> diff --git a/block/io.c b/block/io.c
> index c5bb6ae..54cd9a4 100644
> --- a/block/io.c
> +++ b/block/io.c
> @@ -1779,8 +1779,8 @@ int bdrv_is_allocated_above(BlockDriverState *top,
>      return 0;
>  }
>  
> -int bdrv_pwrite_compressed(BlockDriverState *bs, int64_t offset,
> -                           const void *buf, int count)
> +int coroutine_fn bdrv_co_pwritev_compressed(BlockDriverState *bs,
> +    int64_t offset, unsigned int bytes, QEMUIOVector *qiov)
>  {
>      BlockDriver *drv = bs->drv;
>      int ret;
> @@ -1788,18 +1788,84 @@ int bdrv_pwrite_compressed(BlockDriverState *bs, int64_t offset,
>      if (!drv) {
>          return -ENOMEDIUM;
>      }
> -    if (!drv->bdrv_write_compressed) {
> +
> +    if (!drv->bdrv_co_write_compressed) {
>          return -ENOTSUP;
>      }
> -    ret = bdrv_check_byte_request(bs, offset, count);
> +
> +    ret = bdrv_check_byte_request(bs, offset, bytes);
>      if (ret < 0) {
>          return ret;
>      }
>  
>      assert(QLIST_EMPTY(&bs->dirty_bitmaps));
> +    assert(qemu_in_coroutine());
> +
> +    return drv->bdrv_co_write_compressed(bs, offset >> BDRV_SECTOR_BITS,
> +                                         bytes >> BDRV_SECTOR_BITS, qiov);
> +}
> +
> +typedef struct BdrvWriteCompressedCo {
> +    BlockDriverState *bs;
> +    int64_t offset;
> +    QEMUIOVector *qiov;
> +    int ret;
> +} BdrvWriteCompressedCo;

I think you could just reuse RwCo here.

And in fact, this made we wonder whether we can reuse more from the
normal I/O path. If compressed writes could be made just another REQ_*
flag and we went through bdrv_co_pwritev(), we would automatically get
the alignment code if we don't want to convert drivers to byte-based.
And also the existing coroutine wrapper code would be reused instead
implementing it once more.

I'm not completely sure how special compressed writes really are, so
reusing the normal I/O path may or may not work. It might be worth
investigating, though.

> --- a/include/block/block_int.h
> +++ b/include/block/block_int.h
> @@ -207,6 +207,9 @@ struct BlockDriver {
>      int (*bdrv_write_compressed)(BlockDriverState *bs, int64_t sector_num,
>                                   const uint8_t *buf, int nb_sectors);
>  
> +    int coroutine_fn (*bdrv_co_write_compressed)(BlockDriverState *bs,
> +        int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);

This should be called bdrv_co_writev_compressed because it's a vectored
interface.

Kevin
Pavel Butsykin June 28, 2016, 11:35 a.m. UTC | #5
On 28.06.2016 14:09, Kevin Wolf wrote:
> Am 31.05.2016 um 11:15 hat Denis V. Lunev geschrieben:
>> From: Pavel Butsykin <pbutsykin@virtuozzo.com>
>>
>> This patch just adds the interface to the bdrv_co_pwritev_compressed,
>> which is currently not used but will be useful for safe implementation of the
>> bdrv_co_write_compressed callback in format drivers.
>>
>> Signed-off-by: Pavel Butsykin <pbutsykin@virtuozzo.com>
>> Signed-off-by: Denis V. Lunev <den@openvz.org>
>> CC: Jeff Cody <jcody@redhat.com>
>> CC: Markus Armbruster <armbru@redhat.com>
>> CC: Eric Blake <eblake@redhat.com>
>> CC: John Snow <jsnow@redhat.com>
>> CC: Stefan Hajnoczi <stefanha@redhat.com>
>> CC: Kevin Wolf <kwolf@redhat.com>
>> ---
>>   block/io.c                | 78 +++++++++++++++++++++++++++++++++++++++++++----
>>   include/block/block_int.h |  5 +++
>>   qemu-img.c                |  2 +-
>>   3 files changed, 78 insertions(+), 7 deletions(-)
>>
>> diff --git a/block/io.c b/block/io.c
>> index c5bb6ae..54cd9a4 100644
>> --- a/block/io.c
>> +++ b/block/io.c
>> @@ -1779,8 +1779,8 @@ int bdrv_is_allocated_above(BlockDriverState *top,
>>       return 0;
>>   }
>>
>> -int bdrv_pwrite_compressed(BlockDriverState *bs, int64_t offset,
>> -                           const void *buf, int count)
>> +int coroutine_fn bdrv_co_pwritev_compressed(BlockDriverState *bs,
>> +    int64_t offset, unsigned int bytes, QEMUIOVector *qiov)
>>   {
>>       BlockDriver *drv = bs->drv;
>>       int ret;
>> @@ -1788,18 +1788,84 @@ int bdrv_pwrite_compressed(BlockDriverState *bs, int64_t offset,
>>       if (!drv) {
>>           return -ENOMEDIUM;
>>       }
>> -    if (!drv->bdrv_write_compressed) {
>> +
>> +    if (!drv->bdrv_co_write_compressed) {
>>           return -ENOTSUP;
>>       }
>> -    ret = bdrv_check_byte_request(bs, offset, count);
>> +
>> +    ret = bdrv_check_byte_request(bs, offset, bytes);
>>       if (ret < 0) {
>>           return ret;
>>       }
>>
>>       assert(QLIST_EMPTY(&bs->dirty_bitmaps));
>> +    assert(qemu_in_coroutine());
>> +
>> +    return drv->bdrv_co_write_compressed(bs, offset >> BDRV_SECTOR_BITS,
>> +                                         bytes >> BDRV_SECTOR_BITS, qiov);
>> +}
>> +
>> +typedef struct BdrvWriteCompressedCo {
>> +    BlockDriverState *bs;
>> +    int64_t offset;
>> +    QEMUIOVector *qiov;
>> +    int ret;
>> +} BdrvWriteCompressedCo;
>
> I think you could just reuse RwCo here.
>

I reused BlkRwCo, and it's only a temporary change.

> And in fact, this made we wonder whether we can reuse more from the
> normal I/O path. If compressed writes could be made just another REQ_*
> flag and we went through bdrv_co_pwritev(), we would automatically get
> the alignment code if we don't want to convert drivers to byte-based.
> And also the existing coroutine wrapper code would be reused instead
> implementing it once more.
>
> I'm not completely sure how special compressed writes really are, so
> reusing the normal I/O path may or may not work. It might be worth
> investigating, though.
>
>> --- a/include/block/block_int.h
>> +++ b/include/block/block_int.h
>> @@ -207,6 +207,9 @@ struct BlockDriver {
>>       int (*bdrv_write_compressed)(BlockDriverState *bs, int64_t sector_num,
>>                                    const uint8_t *buf, int nb_sectors);
>>
>> +    int coroutine_fn (*bdrv_co_write_compressed)(BlockDriverState *bs,
>> +        int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
>
> This should be called bdrv_co_writev_compressed because it's a vectored
> interface.
>
agree

> Kevin
>

Patch
diff mbox

diff --git a/block/io.c b/block/io.c
index c5bb6ae..54cd9a4 100644
--- a/block/io.c
+++ b/block/io.c
@@ -1779,8 +1779,8 @@  int bdrv_is_allocated_above(BlockDriverState *top,
     return 0;
 }
 
-int bdrv_pwrite_compressed(BlockDriverState *bs, int64_t offset,
-                           const void *buf, int count)
+int coroutine_fn bdrv_co_pwritev_compressed(BlockDriverState *bs,
+    int64_t offset, unsigned int bytes, QEMUIOVector *qiov)
 {
     BlockDriver *drv = bs->drv;
     int ret;
@@ -1788,18 +1788,84 @@  int bdrv_pwrite_compressed(BlockDriverState *bs, int64_t offset,
     if (!drv) {
         return -ENOMEDIUM;
     }
-    if (!drv->bdrv_write_compressed) {
+
+    if (!drv->bdrv_co_write_compressed) {
         return -ENOTSUP;
     }
-    ret = bdrv_check_byte_request(bs, offset, count);
+
+    ret = bdrv_check_byte_request(bs, offset, bytes);
     if (ret < 0) {
         return ret;
     }
 
     assert(QLIST_EMPTY(&bs->dirty_bitmaps));
+    assert(qemu_in_coroutine());
+
+    return drv->bdrv_co_write_compressed(bs, offset >> BDRV_SECTOR_BITS,
+                                         bytes >> BDRV_SECTOR_BITS, qiov);
+}
+
+typedef struct BdrvWriteCompressedCo {
+    BlockDriverState *bs;
+    int64_t offset;
+    QEMUIOVector *qiov;
+    int ret;
+} BdrvWriteCompressedCo;
+
+static void bdrv_write_compressed_co_entry(void *opaque)
+{
+    BdrvWriteCompressedCo *co = opaque;
+
+    co->ret = bdrv_co_pwritev_compressed(co->bs, co->offset, co->qiov->size,
+                                         co->qiov);
+}
+
+int bdrv_pwrite_compressed(BlockDriverState *bs, int64_t offset,
+                           const void *buf, int count)
+{
+    BdrvWriteCompressedCo data;
+    QEMUIOVector qiov;
+    BlockDriver *drv = bs->drv;
+    struct iovec iov = {
+        .iov_base = (void *)buf,
+        .iov_len = count,
+    };
+    qemu_iovec_init_external(&qiov, &iov, 1);
 
-    return drv->bdrv_write_compressed(bs, offset >> BDRV_SECTOR_BITS, buf,
-                                      count >> BDRV_SECTOR_BITS);
+    data = (BdrvWriteCompressedCo) {
+        .bs     = bs,
+        .offset = offset,
+        .qiov   = &qiov,
+        .ret    = -EINPROGRESS,
+    };
+
+    if (!drv) {
+        return -ENOMEDIUM;
+    }
+
+    if (drv->bdrv_write_compressed) {
+        int ret = bdrv_check_byte_request(bs, offset, count);
+        if (ret < 0) {
+            return ret;
+        }
+        assert(QLIST_EMPTY(&bs->dirty_bitmaps));
+        return drv->bdrv_write_compressed(bs, offset >> BDRV_SECTOR_BITS, buf,
+                                          count >> BDRV_SECTOR_BITS);
+    }
+
+    if (qemu_in_coroutine()) {
+        /* Fast-path if already in coroutine context */
+        bdrv_write_compressed_co_entry(&data);
+    } else {
+        AioContext *aio_context = bdrv_get_aio_context(bs);
+
+        Coroutine *co = qemu_coroutine_create(bdrv_write_compressed_co_entry);
+        qemu_coroutine_enter(co, &data);
+        while (data.ret == -EINPROGRESS) {
+            aio_poll(aio_context, true);
+        }
+    }
+    return data.ret;
 }
 
 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
diff --git a/include/block/block_int.h b/include/block/block_int.h
index 30a9717..ccba9c9 100644
--- a/include/block/block_int.h
+++ b/include/block/block_int.h
@@ -207,6 +207,9 @@  struct BlockDriver {
     int (*bdrv_write_compressed)(BlockDriverState *bs, int64_t sector_num,
                                  const uint8_t *buf, int nb_sectors);
 
+    int coroutine_fn (*bdrv_co_write_compressed)(BlockDriverState *bs,
+        int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
+
     int (*bdrv_snapshot_create)(BlockDriverState *bs,
                                 QEMUSnapshotInfo *sn_info);
     int (*bdrv_snapshot_goto)(BlockDriverState *bs,
@@ -535,6 +538,8 @@  int coroutine_fn bdrv_co_preadv(BlockDriverState *bs,
 int coroutine_fn bdrv_co_pwritev(BlockDriverState *bs,
     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
     BdrvRequestFlags flags);
+int coroutine_fn bdrv_co_pwritev_compressed(BlockDriverState *bs,
+    int64_t offset, unsigned int bytes, QEMUIOVector *qiov);
 
 int get_tmp_filename(char *filename, int size);
 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
diff --git a/qemu-img.c b/qemu-img.c
index eb744d4..ab54027 100644
--- a/qemu-img.c
+++ b/qemu-img.c
@@ -2024,7 +2024,7 @@  static int img_convert(int argc, char **argv)
         const char *preallocation =
             qemu_opt_get(opts, BLOCK_OPT_PREALLOC);
 
-        if (!drv->bdrv_write_compressed) {
+        if (!drv->bdrv_write_compressed && !drv->bdrv_co_write_compressed) {
             error_report("Compression not supported for this file format");
             ret = -1;
             goto out;