NFS write got EIO on kernel starting from 2.4.19-pre4 (or -pre3 maybe)

NFS write got EIO on kernel starting from 2.4.19-pre4 (or -pre3 maybe)

[Vladimir Serov]

Hi Trond,
Belive or not, I've got another NFS related problem. I'm getting EIO in 
several programs (dd, make) writing relativly large file (several 
megabytes) over NFS. I've tested several kernels to find out where this 
problem was introdused. Here the list:

Good kernels (doesn't give EIO) : 2.4.18-5asp, 2.4.19, 2.4.20-pre2
Bad kernel (gives EIO): 2.4.20-pre4, 2.4.20-pre6, 2.4.20, 
2.4.21-pre5-ac3, 2.4.21-pre6
2.4.20-pre3 unfortunatly hangs on boot on my hardware.

I'm able to trigger problem by 'dd if=/dev/zero of=test bs=32k count=1024'
I'm using NFS over UDP, and this problem ccured regardless of soft or 
hard mounted fs.
There was no changes in eepro100 driver i'm using for NIC between 
2.4.20-pre2 and 2.4.20-pre4

client hardware : Intel mobo with built-in NIC , P4 2.2Ghz, 512 Mb DDR 
memory
server hardware: MSI i815ept mobo with  3Com 3c905C-TX/TX-M [Tornado] 
PIII 733Mhz , 512Mb SDRAM

client and server are on the same 100Mb hub , client is under ASP Linux 
7.3, server is under RH8.0
Unfortunatly, dd or cp problem is transient , while make (actually ar) 
gets EIO when building openh323 just every time.
I have to stick to the more recent then  2.4.20-pre2 kernel to make 
CD/DVD writer working on the other hand.
Can You  investigate this problem or give me an advise what to do myself ?

By the way , do You have any progress with nfsclient hanging in the D 
state, i've told You previously ?
Cheers, Vladimir.

NFS write got EIO on kernel starting from 2.4.19-pre4 (or -pre3 maybe)

[Trond Myklebust]

>>>>> " " == Vladimir Serov <vserov@infratel.com> writes:

     > Hi Trond, Belive or not, I've got another NFS related
     > problem. I'm getting EIO in several programs (dd, make) writing
     > relativly large file (several megabytes) over NFS. I've tested
     > several kernels to find out where this problem was
     > introdused. Here the list:

     > Good kernels (doesn't give EIO) : 2.4.18-5asp, 2.4.19,
     > 2.4.20-pre2 Bad kernel (gives EIO): 2.4.20-pre4, 2.4.20-pre6,
     > 2.4.20, 2.4.21-pre5-ac3, 2.4.21-pre6 2.4.20-pre3 unfortunatly
     > hangs on boot on my hardware.

Are there any kernel messages to accompany this EIO?

Cheers,
 Trond

NFS write got EIO on kernel starting from 2.4.19-pre4 (or -pre3 maybe)

[Trond Myklebust]

>>>>> " " == Vladimir Serov <vserov@infratel.com> writes:

     > I'm able to trigger problem by 'dd if=/dev/zero of=test bs=32k
     > count=1024' I'm using NFS over UDP, and this problem ccured
     > regardless of soft or hard mounted fs.

You sure about the 'regardless of hard mount' bit? I can write
multi-gigabyte files with 2.4.21-pre6 without ever seeing an error
(just now I ran a 3x3GB read/write using iozone without any trouble).

Cheers,
  Trond

Re: NFS write got EIO on kernel starting from 2.4.19-pre4 (or -pre3 maybe)

[Vladimir Serov]

Trond Myklebust wrote:

>>>>>>" " == Vladimir Serov <vserov@infratel.com> writes:
>>>>>>            
>>>>>>
>
>     > I'm able to trigger problem by 'dd if=/dev/zero of=test bs=32k
>     > count=1024' I'm using NFS over UDP, and this problem ccured
>     > regardless of soft or hard mounted fs.
>
>You sure about the 'regardless of hard mount' bit? I can write
>multi-gigabyte files with 2.4.21-pre6 without ever seeing an error
>(just now I ran a 3x3GB read/write using iozone without any trouble).
>
>Cheers,
>  Trond
>
Hi,Trond
I know myself that NFS is rather reliable for many yers, I was working 
for many years in HEP community (CERN/ALICE,FNAL/D0,JLAB/CLAS) and used 
NFS all the way processing uncounted terrabytes of data. BUT !!! 
Something in my current setup is triggering some hard to exibit problem. 
Mine previous problem with ARM board for example shows on ten physically 
different boards with several PC different in hardware and software. I 
really don't think it's only hardware problem (but it could be hardware 
related, i mean 3COM NIC's can somehow brighten this problem). Well , I 
don't really know what the hell is going on, That's why I'm calling for 
assistance.

Regards, Vladimir

Re: NFS write got EIO on kernel starting from 2.4.19-pre4 (or -pre3 maybe)

[Joe Korty]

On Tue, Apr 01, 2003 at 05:16:24PM +0400, Vladimir Serov wrote:
> Hi Trond,
> Belive or not, I've got another NFS related problem. I'm getting EIO in 
> several programs (dd, make) writing relativly large file (several 
> megabytes) over NFS. I've tested several kernels to find out where this 
> problem was introdused. Here the list:

Hi Vladimir, Everyone,
  Try this patch.  The bug was introduced in 2.4.20-rc3.  It tends to show
up when a slow nfs server is used with a fast client.
Joe


diff -Nur prev/2.4-redhawk/fs/nfs/read.c curr/2.4-redhawk/fs/nfs/read.c
+++ curr/2.4-redhawk/fs/nfs/read.c	2003-04-01 17:04:18.000000000 -0500
+++ curr/2.4-redhawk/fs/nfs/read.c	2003-04-01 17:04:18.000000000 -0500
@@ -424,7 +424,8 @@
 				memset(p + count, 0, PAGE_CACHE_SIZE - count);
 				kunmap(page);
 				count = 0;
-				if (data->res.eof)
+				if (data->res.eof
+				|| (page_index(page) < (PAGE_CACHE_ALIGN(inode->i_size) >> PAGE_CACHE_SHIFT)))
 					SetPageUptodate(page);
 				else
 					SetPageError(page);