Skip to contentSkip to conversation

NFS client problem with kernel 2.6 and SGI IRIX 6.5

DISCUSSION James Pearson 2 participants 2 messages Oct 17, 2005

lore
Share & tools
lore
#0 lore raw
Quoted 31 lines

The summary is as follows: I do have problems with the 2.6 series kernel, which do not occur with a 2.4 series kernel (and an other- wise unchanged system). I discovered it with Mathematica version 5.0, but do think that other programs are also involved (e.g. OpenOffice 1.1.4, that doesn't find its default (or any other) printer any longer). The symptom is, that certain ressources are reported missing, that are definitively there and which lie somewhere within the application-tree, that tree lying within a hierarchie being nfs-auto-mounted from the SGI system to the (Intel architec- ture) Linux client. File contents (or whole files?) seems to get 'lost' somehow.

It doesn't seem to be the MSBit Problem of the 32bit nfs cookies (alone) - the branch is exported with the IRIX '32bitclients' option, to avoid the 64bit cookies, that led to a similar problem with the printer in OpenOffice under the 2.4 series kernels, and vanished with the 32bit-option. The reason for me to state this is, that when I applied a 32bit-'SGI-IRIX-induced'-patch for (early) 2.6 kernels (Debians 2.6.8) the problem didn't go away, and it also still occurs when using the 2.6.12-kernel, where some kernel-version ago (2.6.10 or 11?) that part of the cookie problem was solved via a translation table (once and for all, I hope).

The problem occurs when requesting nfs v2 as well as nfs v3 protocol. An LD_ASSUME_KERNEL does not seem to help, as it does with other problems.

When testing or compiling kernels, I always used the 'debianized' versions, but to my understanding, they are nearly unaltered compared to the 'plain' kernels (see Debian changelogs).

Quoted 7 lines

The problem is severe to us, as the same configuration also exports our home-directories, which are, of course, writeable, contrary to the application-tree, which is read-only. Thus any help will be welcome.

I'm willing to try whatever I can do to resolve the problem, but I need guidance in what to do and what (else) you need to know.

Is this similar to the issue in the following thread? :

http://marc.theaimsgroup.com/?l=linux-kernel&m=108741268200839&w=2

James Pearson

#1 lore raw
Quoted 3 lines

Is this similar to the issue in the following thread? :

http://marc.theaimsgroup.com/?l=linux-kernel&m=108741268200839&w=2

I still have to investigate Trond Myklebust's suggestions (strace on 'df' with unmodified 2.6 kernel and 2.6.12 and the tcpdump on the malfunctioning nfs, hopefully can do it tomorrow, but from memory, I think the following holds:

When 'your' thread above, which leads to the URL
http://www.fys.uio.no/~trondmy/src/2.4.18/linux-2.4.18-seekdir.dif,
that isn't available any longer (at least here), is similar to

http://www.ussg.iu.edu/hypermail/linux/kernel/0502.1/0506.html
http://kerneltrap.org/mailarchive/1/message/19372/thread

that is the patch, that I applied to some kernels earlier than 2.6.11, then the behaviour is, that with this patch or a 'recent' kernel (I do believe it starts with 2.6.11), the 'find'-problem goes away - I'll re-check that -, but the 'other' problems (Mathematica and OpenOffice not finding certain 'resources') stay.

Nevertheless, 'seekdir' sounds very promising as a cause for the problem(s), so if 'linux-2.4.18-seekdir.dif' is handling a problem different from '0506.html', then it may be worth investigating.

Since I can't access linux-2.4.18-seekdir.dif, could you please either have a look if it's the same thing else send me that patch?

Thank you very much,
Ruediger Oberhage