xfs
[Top] [All Lists]

Re: nfs errors on exported xfs fs

To: Jason <xfs@xxxxxxxxxxxx>
Subject: Re: nfs errors on exported xfs fs
From: Steve Lord <lord@xxxxxxx>
Date: Thu, 28 Jun 2001 00:06:51 -0500
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: Message from Jason <xfs@xxxxxxxxxxxx> of "Wed, 27 Jun 2001 17:04:52 MDT." <3B3A6694.E967B13B@xxxxxxxxxxxx>
Sender: owner-linux-xfs@xxxxxxxxxxx
Hmmm, your kernel was build on May 21st, but how old is your codebase, 2.4.2
predates this by a while. There are a number of NFS related fixes to XFS
in later kernels. The most recent 2.4.5 patches are here:

ftp://oss.sgi.com/projects/xfs/download/testing/Release-1.0.1-PR3/patches/

        patch-2.4.5-bdev-ioctl  - fixes initial ram disk creation
        patch-2.4.5-xfs-1.0.1-core      - core kernel changes
        patch-2.4.5-xfs-1.0.1-kdb       - kdb debugger
        patch-xfs-1.0.1-only    - xfs itself

Or the cvs tree is also available, just about to go up to 2.4.6-pre6

Steve

> I've been getting several different kinds of errors somewhat
> consistently
> on my nfs mounted xfs partition in my home network.  Here are the
> vitals:
> 
> Network: 100Mbit switched
> 
> NFS server (parsec): 
>  Alpha running Debian GNU/Linux
>  parsec:/var/log# uname -a
> Linux parsec 2.4.2-XFS #4 Mon May 21 17:28:46 MST 2001 alpha unknown
>  Promise UDMA/33 PCI controller
>  80G Maxtor IDE drive
>  30G xfs partition (/gulf) exported via user-space Linux nfs server v2.2
>  second 30G ext2 partition (/memoryhole) also exported
>  linux-2.4-xfs-1.0.patch
>  (yes, I know it's somewhat perverse to be using an IDE hd on a SCSI
>   Alpha with a PC IDE controller to create an SGI partition and then
>   export it via nfs...)
> 
> NFS client (erg):
>  x86 linux (RH 6.2)
> 
> First noticed errors of the form:
> 
> somefile: No such file or directory
> 
> when untarring a 200MB tarball of my /usr/doc directory to the nfs
> partition from the client (erg).  I've also gotten "Input/Output error"s
> several times.
> 
> Untarring locally on parsec wasn't a problem, nor was untarring from erg
> to the ext2 partition.
> 
> This script usually gives me errors, but not always.  Just now, it ran
> fine on directories 0 and 1, but then:
> 
> ...
> 1+0 records in
> 1+0 records out
> 1+0 records in
> 1+0 records out
> 1+0 records in
> 1+0 records out
> dd: 2/0: No such file or directory
> dd: 2/1: No such file or directory
> dd: 2/2: No such file or directory
> dd: 2/3: No such file or directory
> dd: 2/4: No such file or directory
> ...
> 
> The script:
> #!/usr/bin/perl
> 
> #Cause problems on my nfs'ed xfs partition -JH
> 
> for($i=0; $i<500; $i++) {
>     system("mkdir $i");
>     for($j=0; $j<100; $j++) {
>         $size = int(rand(10000));
>         system("dd if=/dev/zero of=$i/$j bs=$size count=1");
> #       system("cp /tmp/qwerty $i/$j");
>     }
> }
> 
> The script runs fine when the cwd is in the exported ext2 partition,
> and sometimes even works on the xfs partition.  The /usr/doc tarball
> seems to always have problems.
> 
> Something screwy seems to be going on with writing to files just after
> creating directories, but changing it to this also produces occasional
> I/O
> errors:
> 
> #!/usr/bin/perl
> 
> for($i=0; $i<500; $i++) {
>         for($j=0; $j<100; $j++) {
>                 $size = int(rand(10000));
>                 system("dd if=/dev/zero of=$j bs=$size count=1");
>         }
> }
> 
> Interestingly, just now I ran that script and then did this:
> 
> [jason@erg] /gulf/test$ rm -rf *
> rm: cannot remove `0': Input/output error
> 
> And now I get:
> [jason@erg] /gulf/test$ ls -la
> ls: 0: Input/output error
> total 8
> drwxrwsr-x    2 jason    nfsusers       14 Jun  8 02:34 .
> drwxrwsr-x   21 root     nfsusers     8192 Jun  8 02:02 ..
> 
> Yet, from parsec locally:
> parsec:/gulf/test# ls -la
> total 8
> drwxrwsr-x    2 jason    nfsusers       14 Jun  8 01:34 .
> drwxrwsr-x   21 root     nfsusers     8192 Jun  8 01:02 ..
> -rw-rw-r--    1 jason    nfsusers        0 Jun  8 01:33 0
> 
> So, something screwy is going on, restricted to nfs accesses of the xfs
> partition.  No troubles locally, and none with the exported ext2 fs.
> Oddly enough, nothing interesting in the logs of either machine, either.
> 
>                                         -J



<Prev in Thread] Current Thread [Next in Thread>