Possible kafs related regression
markus.suvanto at gmail.com
markus.suvanto at gmail.com
Thu Feb 15 11:01:50 PST 2024
to, 2024-02-15 kello 11:28 -0500, Jeffrey E Altman kirjoitti:
> On 2/15/2024 8:44 AM, Jeffrey E Altman wrote:
> > On 2/15/2024 6:48 AM, markus.suvanto at gmail.com wrote:
> > > Dear all
> > >
> > > Compiling code under /afs filesystem using make causes hang.
> > > For me this is regression and the latest working version was 6.7.2
> > > (at least I never hit this when I was using 6.7.2)
> > > I have not bisected this but first not working version was 6.7.3
> > > and 6.7.4 hangs also.
> > >
> > > I have not tested this using 6.8-rc4
> > >
> > > https://bugzilla.kernel.org/show_bug.cgi?id=218496
> > >
> > > -Markus Suvanto
> >
> > Markus,
> >
> > There is a known bug in all of the 6.7 kernels that breaks AFS
> > communication on various networks.
> >
> > This was discussed by myself and Michael Lass on openafs-devel. A fix
> > was merged into 6.8-rc2 and 6.7.5-rc1.
> >
> > The symptom are FetchData and StoreData calls that stall because the
> > kernel drops the incoming udp packets. A stall when reading a
> > directory would be consistent with that bug.
> >
> > Jeffrey Altman
>
> This commit is the fix for the mentioned issue.
>
> commit fe92f874f09145a6951deacaa4961390238bbe0d
> Author: Michael Lass <bevan at bi-co.net>
> Date: Wed Jan 31 16:52:20 2024 +0100
>
> net: Fix from address in memcpy_to_iter_csum()
>
> While inlining csum_and_memcpy() into memcpy_to_iter_csum(), the from
> address passed to csum_partial_copy_nocheck() was accidentally changed.
> This causes a regression in applications using UDP, as for example
> OpenAFS, causing loss of datagrams.
>
Now I'm using kernel 6.8.0-rc4-00034-g8d3dea210042
and it fixed my problem :)
-Markus
More information about the linux-afs
mailing list