NFS in 2.4.9

NFS in 2.4.9

[volodya]

I have upgraded to 2.4.9 and NFS no longer works for me. I get
---------------------------------------------------------------
NFS: NFSv3 not supported.
nfs warning: mount version older than kernel
---------------------------------------------------------------
even though I upgraded to the most recent version of util-linux 
(2.11h) and when reading certain files programs lock up and the kernel
prints out the following messages:

nfs: server node4 not responding, still trying
nfs: server node4 not responding, still trying

However, node4 is fine (I can telnet in it) and seems to work ok. 
(node4 is running 2.4.7, with knfsd).

No oopses, etc..

                       Vladimir Dergachev

Re: NFS in 2.4.9

[Trond Myklebust]

>>>>> " " == volodya  <volodya@mindspring.com> writes:

     > I have upgraded to 2.4.9 and NFS no longer works for me. I get
     > ---------------------------------------------------------------
     > NFS: NFSv3 not supported.  nfs warning: mount version older
     > than kernel

You forgot to enable NFSv3 support in your 2.4.9 kernel.

     > ---------------------------------------------------------------
     > even though I upgraded to the most recent version of util-linux
     > (2.11h) and when reading certain files programs lock up and the
     > kernel prints out the following messages:

     > nfs: server node4 not responding, still trying nfs: server
     > node4 not responding, still trying

     > However, node4 is fine (I can telnet in it) and seems to work
     > ok. (node4 is running 2.4.7, with knfsd).

In 99.999% of cases this is due to a network configuration
error. Usually it's things like running full-duplex against a
half-duplex capable switch etc.
TCP is less sensitive to this sort of thing than UDP is, so you won't
see it using telnet.

If you can't resolve where the problem lies, try setting rsize and
wsize manually to some smaller value.

Cheers,
  Trond

Re: NFS in 2.4.9

[volodya]

On 28 Aug 2001, Trond Myklebust wrote:

> >>>>> " " == volodya  <volodya@mindspring.com> writes:
> 
>      > I have upgraded to 2.4.9 and NFS no longer works for me. I get
>      > ---------------------------------------------------------------
>      > NFS: NFSv3 not supported.  nfs warning: mount version older
>      > than kernel
> 
> You forgot to enable NFSv3 support in your 2.4.9 kernel.

I did.. I'll make clean and recompile the kernel again, just in case. 
But still: why complain about mount ? Why allow to mount a partition ?
I do not think this is the cause (also see below).

> 
>      > ---------------------------------------------------------------
>      > even though I upgraded to the most recent version of util-linux
>      > (2.11h) and when reading certain files programs lock up and the
>      > kernel prints out the following messages:
> 
>      > nfs: server node4 not responding, still trying nfs: server
>      > node4 not responding, still trying
> 
>      > However, node4 is fine (I can telnet in it) and seems to work
>      > ok. (node4 is running 2.4.7, with knfsd).
> 
> In 99.999% of cases this is due to a network configuration
> error. Usually it's things like running full-duplex against a
> half-duplex capable switch etc.
> TCP is less sensitive to this sort of thing than UDP is, so you won't
> see it using telnet.

You were right about networking. It turns out that I can ping just fine 
when the packet size is 5524 and under (with header 5532). When I ping 
with size 5525 and above (with header 5533) no packets (at all !) get
thru. Pinging localhost goes fine. None of the machines on my local
network can ping with sizes about 5525. Using tcpdump shows that packets
do arrive on the interface. Setting wsize and rsize to 5000 helped - thank
you very much ! I guess that between 2.4.7 and 2.4.9 the default rsize and
wsize went up.

Now I am wondering what is wrong with large udp packets ? (my mtu is 
1500 and they do get chopped up in smaller packets, but so do size
5524). Any ideas ?

                  thanks !

                       Vladimir Dergachev

> 
> If you can't resolve where the problem lies, try setting rsize and
> wsize manually to some smaller value.
> 
> Cheers,
>   Trond
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
> 

Re: NFS in 2.4.9

[Trond Myklebust]

>>>>> " " == volodya  <volodya@mindspring.com> writes:


    >> > NFS: NFSv3 not supported.  nfs warning: mount version older
    >> > than kernel
    >>
    >> You forgot to enable NFSv3 support in your 2.4.9 kernel.

     > I did.. I'll make clean and recompile the kernel again, just in
     > case. But still: why complain about mount ? Why allow to mount
     > a partition ?  I do not think this is the cause (also see
     > below).

The mount format changed when we added NFSv3. The new format is
backward compatible, but we still print out a message in order to
ensure people get the combination NFS + mount correct.

     > the interface. Setting wsize and rsize to 5000 helped - thank
     > you very much ! I guess that between 2.4.7 and 2.4.9 the
     > default rsize and wsize went up.

No. The default should still be the same. I changed the latter last in
the 2.4.0 pre series. You can check what the actual value is using
'/proc/mounts'.

     > Now I am wondering what is wrong with large udp packets ? (my
     > mtu is 1500 and they do get chopped up in smaller packets, but
     > so do size 5524). Any ideas ?

You're probably seeing fragments disappearing. UDP has no mechanism
detect this, so when it happens, the RPC layer first has to time out
and then resend the entire request. In TCP connections, only the
missing fragment needs to be resent.
Now as to the reason why fragments are disappearing, I suggest you
look at your network layout and try to measure (with the aid of a
packet sniffer such as tcpdump) on which segment the loss is occuring.

Cheers,
  Trond