2.6.X File locking on NFS stil broken

DISCUSSION Ruben Puettmann 4 participants 5 messages Jun 9, 2004 – Jun 10, 2004 3 messages not in the archive

Subsystems (2)

NFS client NFS server

lore

Share & tools

lore

Ruben Puettmann

author

2004-06-09 19:17 UTC #0 lore raw

hello,

I have a big iproblem with the nfs file locking on some 2.6 er kernel

Kernel Server: 2.4.9
Kernel Client: 2.6.7-rc3-bk2

               2.6.6-rc3-mm2
               2.6.5

mount options :

10.0.0.149:/export /mnt/server nfs rw,auto,soft,bg,timeo=12,retrans=3,in tr,rsize=8192,wsize=8192,mountvers=3,nfsvers=3,udp 0 0

nfs-utils : 1.0.6

Both Server an Client Debian Sarge. Wiht kernel 2.4.21-pre5 the locking runs very fine against the server.

This little testscript I used ( got it from a friend):

#define _GNU_SOURCE
#include <unistd.h>
#include <stdio.h>
#include <fcntl.h>
#include <string.h>
#include <errno.h>

int main(int c, char **v)

{
    for (++v, --c; c--; ++v) {
        struct flock lck = {F_RDLCK, SEEK_SET, 0, 0}; /* whole file */
        int fd = open(*v, O_RDONLY);
        if ( fd < 0 )
            fprintf(stderr, "%s: %s\n", *v, strerror(errno));
        else {
            if ( fcntl(fd, F_SETLK, &lck) < 0 )
                fprintf(stderr, "%s[%d]: %s\n", *v, lck.l_pid, strerror(errno));
            else
                printf("%s..ok\n", *v);
            close(fd);
        }
    }
    return 0;
}

The programm hangs after executing ./script testfile.

Ruben

Signature · 3 lines

Ruben Puettmann
[email protected]
http://www.puettmann.net

Trond Myklebust

2004-06-09 20:41 UTC #1 lore raw

P=E5 on , 09/06/2004 klokka 15:17, skreiv Ruben Puettmann:

Quoted 4 lines

hello,
=20
=20
I have a big iproblem with the nfs file locking on some 2.6 er kernel=

=20

Quoted 5 lines

=20
Kernel Server: 2.4.9
Kernel Client: 2.6.7-rc3-bk2
               2.6.6-rc3-mm2
               2.6.5

<snip>

Quoted 1 line

This little testscript I used ( got it from a friend):

That script works fine for me: I haven't seen any hangs on locking in any of the 2.6 series kernels so far.

What does 'rpcinfo -p' show on the client and server?

Cheers,
Trond

[email protected]

2004-06-09 20:47 UTC #2 lore raw

Quoted 2 lines · Trond Myklebust

That script works fine for me: I haven't seen any hangs on locking in any of the 2.6 series kernels so far.

If I mount it with nolock it runs fine too ;-). I have not enough C knowledge to see whree the bug is.

Quoted 3 lines

What does 'rpcinfo -p' show on the client and server?

Server :

sslnfs:~# rpcinfo -p 
   Program Vers Proto   Port
    100000    2   tcp    111  portmapper
    100000    2   udp    111  portmapper
    100024    1   udp    759  status
    100024    1   tcp    762  status
    100003    2   udp   2049  nfs
    100003    3   udp   2049  nfs
    100021    1   udp  32779  nlockmgr
    100021    3   udp  32779  nlockmgr
    100021    4   udp  32779  nlockmgr
    100005    1   udp    901  mountd
    100005    1   tcp    904  mountd
    100005    2   udp    901  mountd
    100005    2   tcp    904  mountd
    100005    3   udp    901  mountd
    100005    3   tcp    904  mountd

Client:

root@www10:[~] > rpcinfo -p
Program Vers Proto Port

    100000    2   tcp    111  portmapper
    100000    2   udp    111  portmapper
    100021    1   udp  32768  nlockmgr
    100021    3   udp  32768  nlockmgr
    100021    4   udp  32768  nlockmgr
    100024    1   udp    938  status
    100024    1   tcp    941  status

Ruben

Signature · 3 lines

Ruben Puettmann
[email protected]
http://www.puettmann.net

A message that is not in the archive (sent off-list or not yet received): [email protected]

Ruben Puettmann

author

2004-06-09 21:24 UTC #6 lore raw

Quoted 11 lines · Trond Myklebust

På on , 09/06/2004 klokka 17:03, skreiv Ruben Puettmann:
Quoted 1 line
attached the strace and one tcpdump from the testprogramm.
According to that tcpdump, the server is denying you the lock because it is still in its grace period.
During that period only clients that held locks before the server rebooted are allowed to reclaim those locks. Your client will need to wait until that grace period is over (usually ~ 1 minute or so).

I have done a reboot on teh server ( It was up for over 315 day's ;-( ) now all runs fine seems to be an race condition. I will take a look on it if this happend again.

Thanks for your help !!

Ruben

Signature · 3 lines

Ruben Puettmann
[email protected]
http://www.puettmann.net

Ryan Anderson

2004-06-10 02:59 UTC #7 lore raw

Quoted 8 lines · Ruben Puettmann

Quoted 7 lines · Trond Myklebust
P? on , 09/06/2004 klokka 17:03, skreiv Ruben Puettmann:
Quoted 1 line
attached the strace and one tcpdump from the testprogramm.
According to that tcpdump, the server is denying you the lock because it is still in its grace period.

Quoted 7 lines

Quoted 3 lines
During that period only clients that held locks before the server rebooted are allowed to reclaim those locks. Your client will need to wait until that grace period is over (usually ~ 1 minute or so).
I have done a reboot on teh server ( It was up for over 315 day's ;-( ) now all runs fine seems to be an race condition. I will take a look on it if this happend again.

Did you change libc and/or your DNS configuration without restarting nfs-common? (statd/lockd)

Signature · 1 line

Ryan Anderson