[Qemu-devel] iSCSI support for QEMU, update

[Qemu-devel] iSCSI support for QEMU, update

[ronniesahlberg]

iSCSI block driver for QEMU

Please find an updated iSCSI patch.
This patch adds setting FUA on all writes when the bit
BDRV_O_CACHE_WB is not set.


regards
ronnie sahlberg
 

From: Ronnie Sahlberg <ronniesahlberg@gmail.com>
Subject: iSCSI support for QEMU, update
In-Reply-To: 

Re: [Qemu-devel] iSCSI support for QEMU, update

[Christoph Hellwig]

On Fri, Apr 22, 2011 at 06:23:58AM +1000, ronniesahlberg@gmail.com wrote:
> iSCSI block driver for QEMU
> 
> Please find an updated iSCSI patch.
> This patch adds setting FUA on all writes when the bit
> BDRV_O_CACHE_WB is not set.

At this point you need to check for BDRV_O_CACHE_WB and BDRV_O_NOCACHE
as they both have writeback cache semantics.  I submitted a patch to clean
that up, but it didn't get merged yet.

Re: [Qemu-devel] iSCSI support for QEMU, update

[ronnie sahlberg]

Thanks,

On Fri, Apr 22, 2011 at 2:08 PM, Christoph Hellwig <hch@lst.de> wrote:
> On Fri, Apr 22, 2011 at 06:23:58AM +1000, ronniesahlberg@gmail.com wrote:
>> iSCSI block driver for QEMU
>>
>> Please find an updated iSCSI patch.
>> This patch adds setting FUA on all writes when the bit
>> BDRV_O_CACHE_WB is not set.
>
> At this point you need to check for BDRV_O_CACHE_WB and BDRV_O_NOCACHE
> as they both have writeback cache semantics.  I submitted a patch to clean
> that up, but it didn't get merged yet.
>


So I should set FUA when
BDRV_O_CACHE_WB==0  or  BDRV_O_NOCACHE==1
right?



regards
ronnie sahlberg

Re: [Qemu-devel] iSCSI support for QEMU, update

[Christoph Hellwig]

On Fri, Apr 22, 2011 at 04:08:57PM +1000, ronnie sahlberg wrote:
> So I should set FUA when
> BDRV_O_CACHE_WB==0  or  BDRV_O_NOCACHE==1
> right?

At this point you need to set it if either the BDRV_O_CACHE_WB and
BDRV_O_NOCACHE flags are set.  And yes, the naming of the nocache
flag is rather confusing.

Take a look at block.c:bdrv_open_common():

    /*
     * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
     * write cache to the guest.  We do need the fdatasync to flush
     * out transactions for block allocations, and we maybe have a
     * volatile write cache in our backing device to deal with.
     */
    if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
        bs->enable_write_cache = 1;

Re: [Qemu-devel] iSCSI support for QEMU, update

[ronnie sahlberg]

Ok, I am confused

On Fri, Apr 22, 2011 at 4:13 PM, Christoph Hellwig <hch@lst.de> wrote:
> On Fri, Apr 22, 2011 at 04:08:57PM +1000, ronnie sahlberg wrote:
>> So I should set FUA when
>> BDRV_O_CACHE_WB==0  or  BDRV_O_NOCACHE==1
>> right?
>
> At this point you need to set it if either the BDRV_O_CACHE_WB and
> BDRV_O_NOCACHE flags are set.  And yes, the naming of the nocache
> flag is rather confusing.
>
> Take a look at block.c:bdrv_open_common():
>
>    /*
>     * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
>     * write cache to the guest.  We do need the fdatasync to flush
>     * out transactions for block allocations, and we maybe have a
>     * volatile write cache in our backing device to deal with.
>     */
>    if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
>        bs->enable_write_cache = 1;
>

I have to set FUA if the WB(I guessed this meant write-back) flag is set?
Not the opposite (I assume opposite of write-back is write-through) ?



So that would mean
   if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
          caching is NOT ok, write with FUA=1
   else
          caching is ok, write with FUA=0
right ?


regards
ronnie sahlberg

Re: [Qemu-devel] iSCSI support for QEMU, update

[Christoph Hellwig]

On Fri, Apr 22, 2011 at 04:35:11PM +1000, ronnie sahlberg wrote:
> Ok, I am confused

And this time it's my fault, sorry.

> I have to set FUA if the WB(I guessed this meant write-back) flag is set?
> Not the opposite (I assume opposite of write-back is write-through) ?

Sorry, I meant exactly the opposite.  You do not have to set FUA if either
of those two flags is present, so:

	if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
		caching is ok, write with FUA=0
	else
		caching is NOT ok, write with FUA=1

[Qemu-devel] iSCSI support for QEMU, update

[Ronnie Sahlberg]

This is a patch that adds support for iSCSI to QEMU when QEMU is built against libiscsi (https://github.com/sahlberg/libiscsi)

ISCSI devices are specified using a special iSCSI URL:
  -drive file=iscsi://10.1.1.1:3260/iqn.ronnie.test/1


This allows a guest to access iscsi devices without exposing these devices to the host. Which is nice if/when you have a very large number of LUNs that are dedicated for virtual guests.


This patch is updated to set FUA for all writes, unless any of the NOCACHE or CACHE_WB flags are set for the device.


Please review and/or apply.



regards
ronnie sahlberg