[1.2,3/7] block: always open protocol in writeback mode

Formats are entirely in charge of flushes for metadata writes.  For
guest-initiated writes, a writethrough cache is faked in the block layer.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 block.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

Am 22.05.2012 18:17, schrieb Paolo Bonzini:
> Formats are entirely in charge of flushes for metadata writes.  For
> guest-initiated writes, a writethrough cache is faked in the block layer.
> 
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>  block.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
> 
> diff --git a/block.c b/block.c
> index 7add33c..43b3de6 100644
> --- a/block.c
> +++ b/block.c
> @@ -661,7 +661,7 @@ static int bdrv_open_common(BlockDriverState *bs, const char *filename,
>      if (drv->bdrv_file_open) {
>          ret = drv->bdrv_file_open(bs, filename, open_flags);

Why don't you add the writeback flag here? Would the code is raw-posix
that adds O_DSYNC be unused then. and can it still be used in other ways
than format=file/host_* with this patch?

>      } else {
> -        ret = bdrv_file_open(&bs->file, filename, open_flags);
> +        ret = bdrv_file_open(&bs->file, filename, open_flags | BDRV_O_CACHE_WB);
>          if (ret >= 0) {
>              ret = drv->bdrv_open(bs, open_flags);
>          }

Kevin

Il 31/05/2012 13:33, Kevin Wolf ha scritto:
>> > diff --git a/block.c b/block.c
>> > index 7add33c..43b3de6 100644
>> > --- a/block.c
>> > +++ b/block.c
>> > @@ -661,7 +661,7 @@ static int bdrv_open_common(BlockDriverState *bs, const char *filename,
>> >      if (drv->bdrv_file_open) {
>> >          ret = drv->bdrv_file_open(bs, filename, open_flags);
> Why don't you add the writeback flag here? Would the code is raw-posix
> that adds O_DSYNC be unused then. and can it still be used in other ways
> than format=file/host_* with this patch?

Because in this case there is no format over it, so the topmost BDS will
be open with BDRV_O_CACHE_WB and no one will call bdrv_flush.

We could indeed remove the BDRV_O_CACHE_WB handling in all the protocol
drivers (at least those where bdrv_flush_to_disk does something
sensible), and always open in writeback mode.  At that point this patch
could be reverted.

Paolo

>> >      } else {
>> > -        ret = bdrv_file_open(&bs->file, filename, open_flags);
>> > +        ret = bdrv_file_open(&bs->file, filename, open_flags | BDRV_O_CACHE_WB);
>> >          if (ret >= 0) {
>> >              ret = drv->bdrv_open(bs, open_flags);
>> >          }

Am 31.05.2012 13:43, schrieb Paolo Bonzini:
> Il 31/05/2012 13:33, Kevin Wolf ha scritto:
>>>> diff --git a/block.c b/block.c
>>>> index 7add33c..43b3de6 100644
>>>> --- a/block.c
>>>> +++ b/block.c
>>>> @@ -661,7 +661,7 @@ static int bdrv_open_common(BlockDriverState *bs, const char *filename,
>>>>      if (drv->bdrv_file_open) {
>>>>          ret = drv->bdrv_file_open(bs, filename, open_flags);
>> Why don't you add the writeback flag here? Would the code is raw-posix
>> that adds O_DSYNC be unused then. and can it still be used in other ways
>> than format=file/host_* with this patch?
> 
> Because in this case there is no format over it, so the topmost BDS will
> be open with BDRV_O_CACHE_WB and no one will call bdrv_flush.

Doesn't block.c do it now? bs->enable_write_cache still wouldn't be set,
even if .bdrv_file_open() is called with BDRV_O_CACHE_WB.

Kevin

Il 31/05/2012 13:56, Kevin Wolf ha scritto:
>> > Because in this case there is no format over it, so the topmost BDS will
>> > be open with BDRV_O_CACHE_WB and no one will call bdrv_flush.
> Doesn't block.c do it now? bs->enable_write_cache still wouldn't be set,
> even if .bdrv_file_open() is called with BDRV_O_CACHE_WB.

You're right.

Paolo