kernel page fault with nfs

Tobias C. Berner tcberner at gmail.com
Tue Oct 21 09:10:17 UTC 2014


Hi Marcelo

The following ist the current fstab-line which seems to run smoothly:
odo.firefly:/storage/multimedia       /multimedia          nfs 
readahead=4,soft,intr,rw,tcp,wsize=32768,rsize=32768,late 0 0 

nfsstat -m: 
odo.firefly:/storage/multimedia on /multimedia
nfsv3,tcp,resvport,soft,intr,cto,lockd,sec=sys,acdirmin=3,acdirmax=60,acregmin=5,a
cregmax=60,nametimeo=60,negnametimeo=60,rsize=32768,wsize=32768,readdirsi
ze=32768,readahead=4,wcommitsize=2798255,timeout=120,retrans=2




Now the bad line (no different appart from the typo)
odo.firefly:/storage/multimedia       /multimedia          nfs 
readahead=4,soft,intr,rw,tcp,wsize=32767,rsize=32767,late 0 0 
which leads to the page-faults.
And as you said wsize/rsize gets rounded down to the multiple of 512:
odo.firefly:/storage/multimedia on /multimedia
nfsv3,tcp,resvport,soft,intr,cto,lockd,sec=sys,acdirmin=3,acdirmax=60,acregmin=5,a
cregmax=60,nametimeo=60,negnametimeo=60,rsize=32256,wsize=32256,readdirsi
ze=32256,readahead=4,wcommitsize=2798255,timeout=120,retrans=2


I can easily reproduce the pagefault by letting for example multimedia/gpodder write 
to the nfs.



hope this helps,

mfg Tobias




On Tuesday 21 October 2014 15.45:24 Marcelo Araujo wrote:
> Hello Tobias,
> 
> That sounds good, at least you don't have any crash so far.
> I agree with you, seems a bug, I'm gonna take a look on that.
> 
> Could you share with me your testbed or how you can reproduce the issue?
> 
> Best Regards,
> 
> 2014-10-21 15:36 GMT+08:00 T.C.Berner <tcberner at gmail.com>:
> > The system now has an uptime of >24h using NFS heavily.
> > 
> > So wsize/rsize=2^15-1 seems to have been the problem.... which is imho a
> > bug therefore.
> > 
> > 
> > mfg Tobias
> > 
> > 2014-10-21 5:11 GMT+02:00 Marcelo Araujo <araujobsdport at gmail.com>:
> >> Hello Tibias,
> >> 
> >> Any news?
> >> 
> >> 
> >> Best Regards,
> >> 
> >> 2014-10-20 20:55 GMT+08:00 Rick Macklem <rmacklem at uoguelph.ca>:
> >>> Tobias C. Berner wrote:
> >>> > Now that I posted it, 32767 should of course be 2^15=32768. Let me
> >>> > recheck if it still
> >>> > hangs with the correct value.
> >>> > 
> >>> > On Monday 20 October 2014 09.15:39 Tobias C. Berner wrote:
> >>> > > Hi Marcelo
> >>> > > 
> >>> > > Yes, I'm using readahead:
> >>> > > The mountoptions are
> >>> > > "readahead=4,soft,intr,rw,tcp,wsize=32767,rsize=32767,late"
> >>> 
> >>> If you type "nfsstat -m", you will see what is actually getting used.
> >>> (I suspect the above rsize/wsize got clipped to 32256 or something like
> >>> 
> >>>  that. I think it clips it to a multiple of 512.)
> >>> 
> >>> If rsize/wsize are not a power of 2, there are issues, although I've
> >>> never
> >>> been able to see why it is broken. Maybe it should clip it to the power
> >>> of
> >>> 2 below the value, since it causes unexplained problems otherwise.
> >>> 
> >>> rick
> >>> 
> >>> > > mfg Tobias
> >>> > > 
> >>> > > On Monday 20 October 2014 10.41:30 Marcelo Araujo wrote:
> >>> > > > Hello Tobias,
> >>> > > > 
> >>> > > > Could you show how you are mount the NFS share?
> >>> > > > Are you using 'readahead' option?
> >>> > > > 
> >>> > > > Best Regards,
> >>> > > > 
> >>> > > > 2014-10-19 17:40 GMT+08:00 Tobias C. Berner 
<tcberner at gmail.com>:
> >>> > > > > both are at 1100038.
> >>> > > > > 
> >>> > > > > On Sunday 19 October 2014 11.12:36 Marcelo Araujo wrote:
> >>> > > > > > It is still strange, could you do what Allan said and send us
> >>> > > > > > the
> >>> > > > > > result
> >>> > > > > 
> >>> > > > > in
> >>> > > > > 
> >>> > > > > > case you are not sure you have world and kernel in the same
> >>> > > > > > revision!
> >>> > > > > > 
> >>> > > > > > On Oct 19, 2014 6:48 AM, "Tobias C. Berner"
> >>> > > > > > 
> >>> > > > > > <tcberner at gmail.com> wrote:
> >>> > > > > > >  Hi
> >>> > > > > > > 
> >>> > > > > > > World ist from october 16, installed world and kernel then.
> >>> > > > > > > 
> >>> > > > > > > Kernel was later rebuilt with debug-options.
> >>> > > > > > > 
> >>> > > > > > > 
> >>> > > > > > > 
> >>> > > > > > > 
> >>> > > > > > > 
> >>> > > > > > > Is the following more sensible?
> >>> > 
> >>> > 
##################################################
> >>> > 
> >>> > > > > > > # kgdb NOXON/kernel.debug vmcore.1


More information about the freebsd-current mailing list