flock LOCK_EX hangs

flock LOCK_EX hangs

[Baurzhan Ismagulov]

Hello all,

I'm experiencing problems with the kernel NFS server. The client hangs
on the flock call. This is Debian etch, Linux 2.6.16-1-686-smp, the
server and the client is the same machine. The directory is exported
with rw, no mount options. The lockd kernel thread is present, rpcinfo
-p reports nlockmgr registered, the firewall passes all the traffic
through.

I've googled and searched the list archive, the postings are usually
quite old (only two were from 2006 and referred to Linux 2.6), and I
couldn't see how I could further analyze the problem. I've also tried
mounting with nolock, but this apparently makes flock / fcntl to fail,
since the tools I use (automake and debootstrap, although I checked this
only with the latter) exited with an error code, and I rely upon it.

How can I fix the problem?

Thanks in advance,
-- 
Baurzhan Ismagulov
http://www.kz-easy.com/

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

Re: flock LOCK_EX hangs

[J. Bruce Fields]

On Wed, May 16, 2007 at 03:12:53PM +0200, Baurzhan Ismagulov wrote:
> I'm experiencing problems with the kernel NFS server. The client hangs
> on the flock call. This is Debian etch, Linux 2.6.16-1-686-smp, the
> server and the client is the same machine. The directory is exported
> with rw, no mount options. The lockd kernel thread is present, rpcinfo
> -p reports nlockmgr registered, the firewall passes all the traffic
> through.
> 
> I've googled and searched the list archive, the postings are usually
> quite old (only two were from 2006 and referred to Linux 2.6), and I
> couldn't see how I could further analyze the problem. I've also tried
> mounting with nolock, but this apparently makes flock / fcntl to fail,
> since the tools I use (automake and debootstrap, although I checked this
> only with the latter) exited with an error code, and I rely upon it.

Might be fun to watch the network traffic as this is happening.  I'd
probably do something like

	wireshark -Slk -ilo

then try to get a lock, and try to understand what's happening on the
"network": is the client sending requests that the server doesn't
answer?  Or are there errors coming back?  Etc.

Or you could run

	tcpdump -s0 -ilo -wtmp.pcap

and mail a copy of tmp.pcap and somebody could take a look.

There's also been a lot fixed since 2.6.16, so it might be worth trying
something newer.

--b.

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

Re: flock LOCK_EX hangs

[Baurzhan Ismagulov]

[-- Attachment #1: Type: text/plain, Size: 841 bytes --]

Hello Bruce,

On Wed, May 16, 2007 at 01:30:08PM -0400, J. Bruce Fields wrote:
> Might be fun to watch the network traffic as this is happening.

Eh, I've hoped someone would have a silver bullet for me, so that I
don't have to do that :) .

There is one LOCK request and one LOCK reply with the NLM DENIED GRACE
PERIOD error. Does this say anything?

This is still under 2.6.16, I'm planning to upgrade on Friday, unless
the problem can be solved with the current version. Besides, is it
possible to turn on logging? Also, I've seen the following messages in
the kernel log: "do_vfs_lock: VFS is out of sync with lock manager!"
This happened only when I dumped the traffic the last time, not within
the day, when I tried many times (the logs were not rotated).

Any ideas?

With kind regards,
-- 
Baurzhan Ismagulov
http://www.kz-easy.com/

[-- Attachment #2: flock01.tcpdump.gz --]
[-- Type: application/octet-stream, Size: 2055 bytes --]

[-- Attachment #3: Type: text/plain, Size: 286 bytes --]

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

[-- Attachment #4: Type: text/plain, Size: 140 bytes --]

_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

Re: flock LOCK_EX hangs

[Baurzhan Ismagulov]

Hello Bruce,

On Wed, May 16, 2007 at 01:30:08PM -0400, J. Bruce Fields wrote:
> > I'm experiencing problems with the kernel NFS server. The client hangs
> > on the flock call.
...
> There's also been a lot fixed since 2.6.16, so it might be worth trying
> something newer.

I've tried 2.6.20, and it worked. It's amazing how long time it has been
since I had kernel bugs, I haven't even thought in that direction.
Thanks much!

With kind regards,
-- 
Baurzhan Ismagulov
http://www.kz-easy.com/

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs