better description of problem:
Client is using MS Frontpage to upload his web site
to our server. His machine crashes (surprise), presumably
while holding a lock on my server. On my www servers the
following happens:
# cd /www [this is the NFS mounted filesystem]
# ls
blah.com
blah.com.au
blah.au.com
broken-site.com.au
etc.com
.
.
# cd broken-site.com.au
# ls
[error something like this]
no such file or directory: .
IE: I can cd to the directory, it shows up as a child
of its parent, but attempting to list it's contents
doesn't work (on one client), with a very strange error.
The directory was listable on both the NFS server, and
also on a different machine that had the same filesystem
NFS mounted! Restarting the broken machine didn't fix
the problem, neither did moving, renaming or softlinking
to the directory. Stopping the nfs server didn't fix it
either, and the first reboot of all NFS clients and the
server didn't fix the problem. 2nd reboot of whole cluster
did fix it.
Responses interspersed, thanks for help.
> -----Original Message-----
> From: Gus [mailto:[EMAIL PROTECTED]]
> Sent: Wednesday, 26 April 2000 3:16 PM
> To: Richard Lane
> Subject: Re: [SLUG] NFS / Lilo / lockd / kernel problems
>
>
> On Wed, Apr 19, 2000 at 08:18:26PM +1000, Richard Lane wrote:
> > NFS won't start on one of my systems, reboot doesn't
> > fix the problem. (In fact I have to toggle the power to
> > get it to reboot, it appears to be wedging on shutdown.
>
> any error messages? logs? does it bomb out with an error,
> or just hang? at which point does it hang?
No error messages, no logs, nothing (not even a signal to
the shell I'm running the shutdown command from). Just
a second or so's pause, then WEDGE. Don't know if anything
is on the console.
> whats the network topology?
flat, switched 100 Mbit ethernet, unrouted IPs.
> what other OS's are involved?
Various linuxes of 2.2.x kernel, _should_ be all RH 6.0
Nothing else.
> what did you have (NFS) mounted at the time?
/www (a couple of thousand virtual sites)
> what was exported?
nothing
> since you believe its a problem with lockd,
> was anything holding a lock at the time?
Almost certainly. Bug was tickled by a client's
Frontpage crashing (he was uploading to his site).
We're running the MS FP extensions.
> was statd running?
yes
> is a firewall or tcp wrappers preventing any rpc traffic?
no (ARRRGGG!)
> what did a tcpdump of the traffic show?
can't do one (easily)
> did it ever work?
It usually works, it just gets stuck a 0 - 4 times a week,
and I have to shut down all the NFS clients, and the NFS server,
to unstick it. This time even that didn't work (first time).
> > Checking the kernel version shows it is a little old, so
> > I want to try an upgrade (I also think the kernel might be
> > corrupt: why else would would a reboot not fix a problem).
>
> which kernel version?
2.2.10
> if the kernel image was corrupt, i'd
> expect something a little more catastrophic than hanging
> on shutdown..
agreed, but can't work out why it's wedging, or why a lockd
bug would show up as the error described at top.
> the 2.0 kernels used a user-space nfs, which worked fine, had
> a few features that are much easier to do in userspace
> (crossing mounts, re-exporting nfs mounted volumes, etc)
> but couldn't do locking (thanks to user-space)
>
> the 2.2 kernels can still use the old nfsd, but there's now
> a better (in the sense that it supports locking) kernel-space
> implementation. its still pretty "experimental" - you really
> want to apply the sgi nlm4 patches, and dhiggins' merge
> of trond's and neilb's client/server patches for any real
> (esp. heterogenous) use. you will need newer user-space
> tools (hpa's nfs-utils, new (or patched) mount) to use the
> new nfs patches. all of these will be in 2.2.16, in case
> you were wondering.
HOORAY!! 2.2.16 here I come! (when it's released)
Is 2.2.15 out, or still 2.2.15-pre??
> the LDP's NAG documents how to setup an NFS server, and there
> are man pages for each of the nfs-utils programs, as well as
> nfs(5) and exports(5), not to mention umpteen mail archives
> accessable via any web search, and of course, the source code
> itself - what in particular wasn't documented to your
> satisfaction?
lockd (even reading the source proved unenlightening)
Of course, IMNSHO I don't believe I should have to
read the source, I should be able to type "man lockd"
and get the info I want. man -k lockd, at worst.
> > Assuming I build and install a new kernel, is there any
> > way to get Linux to boot the new kernel ONCE, then go back
> > the normal kernel? IE something like:
> >
> > # shutdown -r0 --kernel=2.2.14
>
> man lilo, particularly the bit about "-R"
Ah Ha! I'll go look.
> --
> - Gus
Thanks,
Richard Lane
IT Manager
___________________________________________
NetRegistry http://www.netregistry.com.au
Tel: +61 2 9699 6099 | Fax: +61 2 9699 6088
<[EMAIL PROTECTED]>
--
SLUG - Sydney Linux Users Group Mailing List - http://www.slug.org.au
To unsubscribe send email to [EMAIL PROTECTED] with
unsubscribe in the text
