Oxenstored short-write issue

Oxenstored short-write issue

[Wei Liu]

Hi Dave

We discovered an issue in oxenstored: when it writes to the ring, it writes
a chunk of contiguous data. When it tries to write across ring boundary,
it returns a short-write. That leads to stalling mini-os's xenbus ring.

Samuel provided a diff which solves the issue:
http://www.gossamer-threads.com/lists/xen/devel/403800#403800

You can find our discussions in that thread. Note that thread was mainly
for a patch to fix an issue in mini-os but we discovered that oxenstored
issue along the way.

Wei.

Re: Oxenstored short-write issue

[David Scott]

Hi,

> On 27 Oct 2015, at 15:37, Wei Liu <wei.liu2@citrix.com> wrote:
> 
> Hi Dave
> 
> We discovered an issue in oxenstored: when it writes to the ring, it writes
> a chunk of contiguous data. When it tries to write across ring boundary,
> it returns a short-write. That leads to stalling mini-os's xenbus ring.
> 
> Samuel provided a diff which solves the issue:
> http://www.gossamer-threads.com/lists/xen/devel/403800#403800
> 
> You can find our discussions in that thread. Note that thread was mainly
> for a patch to fix an issue in mini-os but we discovered that oxenstored
> issue along the way.

Interesting! I always thought it was a bit odd that oxenstored could write a partial packet but I thought it would retry higher up and complete the transmit before sleeping. It’s certainly clearer to complete the write at the lower-level though.

So I think the diff looks ok.

Cheers,
Dave
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: Oxenstored short-write issue

[Wei Liu]

On Tue, Oct 27, 2015 at 03:56:07PM +0000, David Scott wrote:
> Hi,
> 
> > On 27 Oct 2015, at 15:37, Wei Liu <wei.liu2@citrix.com> wrote:
> > 
> > Hi Dave
> > 
> > We discovered an issue in oxenstored: when it writes to the ring, it writes
> > a chunk of contiguous data. When it tries to write across ring boundary,
> > it returns a short-write. That leads to stalling mini-os's xenbus ring.
> > 
> > Samuel provided a diff which solves the issue:
> > http://www.gossamer-threads.com/lists/xen/devel/403800#403800
> > 
> > You can find our discussions in that thread. Note that thread was mainly
> > for a patch to fix an issue in mini-os but we discovered that oxenstored
> > issue along the way.
> 
> Interesting! I always thought it was a bit odd that oxenstored could write a partial packet but I thought it would retry higher up and complete the transmit before sleeping. It’s certainly clearer to complete the write at the lower-level though.
> 
> So I think the diff looks ok.
> 

OK, if you think that hunk is good. I will submit a proper patch for it.

Samuel, can I have your S-o-B on that patch?

Wei.

> Cheers,
> Dave

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: Oxenstored short-write issue

[Samuel Thibault]

Wei Liu, le Tue 27 Oct 2015 15:59:46 +0000, a écrit :
> > Interesting! I always thought it was a bit odd that oxenstored could write a partial packet but I thought it would retry higher up and complete the transmit before sleeping. It’s certainly clearer to complete the write at the lower-level though.
> > 
> > So I think the diff looks ok.
> 
> OK, if you think that hunk is good. I will submit a proper patch for it.

Thanks,

> Samuel, can I have your S-o-B on that patch?

Yep, done so in another mail.

Samuel

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel