Re: weird postfix mailspool corruption with 2.4.21-ac2+ (was Re: Linux 2.4.21-ac3)

Re: weird postfix mailspool corruption with 2.4.21-ac2+ (was Re: Linux 2.4.21-ac3)

[Thijs]

Scott McDermott wrote:
> 
> does changing the unix_dgram_ops `poll' op from `dgram_poll'
> to `datagram_poll' in net/unix/af_unix.c change anything for
> you? I can't test this myself until later this week.  Also I
> don't know what other bug the unix_peer_get() addition is
> supposed to fix, so...

Tried it, but doesn't fix it, unfortunately...
guess it's time for the next suspect ;)

regards,

--Thijs Welman

Re: weird postfix mailspool corruption with 2.4.21-ac2+ (was Re: Linux 2.4.21-ac3)

[Jonathan Hudson]

In article <fa.h3c32fv.r5m12l@ifi.uio.no>,
	Scott McDermott <vaxerdec@frontiernet.net> writes:
SM> Thijs on Wed 25/06 22:29 +0200:
>> Since 2.4.21-ac2 i'm experiencing problems with Postfix on
>> Debian Stable. Messages get corrupted while being handled
>> by Postfix.
SM> 
SM> I thought I was crazy, I'm seeing this also.  I had to
SM> switch my kernel back on our mail hub because postfix kept
SM> doing this every time I boot the new kernel.  I tried to
SM> figure out why but could not; the system *appears* to run
SM> normally otherwise.  I was moving from 2.4.21-pre4-ac1
SM> +gibbs-aic7xxx patches from that time, to 2.4.21-ac2
SM> +gibbs-aic7xxx from 20030603, and I thought that might be it
SM> (the AIC patches); are you running AIC adapters on your
SM> system with the gibbs patches by any chance?
SM> 

No AIC or any kind here. Bring on the next suspect.

Re: weird postfix mailspool corruption with 2.4.21-ac2+ (was Re: Linux 2.4.21-ac3)

[Scott McDermott]

Jonathan Hudson on Wed 25/06 23:34 +0100:
> No AIC or any kind here. Bring on the next suspect.

does changing the unix_dgram_ops `poll' op from `dgram_poll'
to `datagram_poll' in net/unix/af_unix.c change anything for
you? I can't test this myself until later this week.  Also I
don't know what other bug the unix_peer_get() addition is
supposed to fix, so...

Re: weird postfix mailspool corruption with 2.4.21-ac2+ (was Re: Linux 2.4.21-ac3)

[Jonathan Hudson]

On Thu, 2003-06-26 at 00:19, Scott McDermott wrote:
> Jonathan Hudson on Wed 25/06 23:34 +0100:
> > No AIC or any kind here. Bring on the next suspect.
> 
> does changing the unix_dgram_ops `poll' op from `dgram_poll'
> to `datagram_poll' in net/unix/af_unix.c change anything for
> you? I can't test this myself until later this week.  Also I
> don't know what other bug the unix_peer_get() addition is
> supposed to fix, so...

I'll see if I can test this out, once I've build -ac3 on a box where I
don't drop real messages.

-jonathan

Re: Linux 2.4.21-ac3

[Thijs]

Hi,

Since 2.4.21-ac2 i'm experiencing problems with Postfix on Debian 
Stable. Messages get corrupted while being handled by Postfix.

Vanilla 2.4.21 and 2.4.21-ac1 are fine, but 2.4.21-ac2/3 causes 
problems. Going back to ac1 resolves the issue. I tried kernels on 
several Debian servers, but all have the same problem. Could be it's 
something in postfix that emerges with this specific patch, but it's at 
least curious. I'm not too familiar with this matter unfortunately.

The only logentries i see are:

postfix/qmgr[399]: warning: active/0/4/04E5B17E3F: too many length bits, 
record type 255
postfix/qmgr[399]: warning: 04E5B17E3F: envelope records out of order
postfix/qmgr[399]: warning: saving corrupt file "04E5B17E3F" from queue 
"active" to queue "corrupt"
...or just...
postfix/smtp[536]: warning: corrupted queue file: active/C/7/C7AC317E3F

Tested on: Intel PPro, Intel Celeron, AMD Duron
Tested on: ext2 and ext3

All other programms seem to work fine, no other strange messages 
whatsoever...

Regards,

--Thijs Welman
Delft University of Technology, the Netherlands

weird postfix mailspool corruption with 2.4.21-ac2+ (was Re: Linux 2.4.21-ac3)

[Scott McDermott]

Thijs on Wed 25/06 22:29 +0200:
> Since 2.4.21-ac2 i'm experiencing problems with Postfix on
> Debian Stable. Messages get corrupted while being handled
> by Postfix.

I thought I was crazy, I'm seeing this also.  I had to
switch my kernel back on our mail hub because postfix kept
doing this every time I boot the new kernel.  I tried to
figure out why but could not; the system *appears* to run
normally otherwise.  I was moving from 2.4.21-pre4-ac1
+gibbs-aic7xxx patches from that time, to 2.4.21-ac2
+gibbs-aic7xxx from 20030603, and I thought that might be it
(the AIC patches); are you running AIC adapters on your
system with the gibbs patches by any chance?

here's what I'm seeing on my mail hub getting a test message
from one of the exchangers:

Jun 24 05:07:54 lujuria postfix/smtpd[997]: connect from mx1.mydomain.tld[1.2.3.4]
Jun 24 05:07:54 lujuria postfix/nqmgr[983]: 7017833812: from=<someone@someother.tld>, size=347, nrcpt=1 (queue active)
Jun 24 05:07:54 lujuria postfix/nqmgr[983]: warning: active/7/0/7017833812: too many length bits, record type 255
Jun 24 05:07:54 lujuria postfix/nqmgr[983]: warning: 7017833812: envelope records out of order
Jun 24 05:07:54 lujuria postfix/nqmgr[983]: warning: saving corrupt file "7017833812" from queue "active" to queue "corrupt"
Jun 24 05:07:54 lujuria postfix/smtpd[997]: A6CF3339AA: client=mx1.mydomain.tld[1.2.3.4]
Jun 24 05:07:54 lujuria postfix/cleanup[991]: A6CF3339AA: message-id=<q$28m4t24a-frqgm---0t-84$dx$z@gq1.5axh>
Jun 24 05:07:54 lujuria postfix/smtpd[997]: disconnect from mx1.mydomain.tld[1.2.3.4]

this is system is RH8+updates with the mentioned kernels.
Very bizarre.

> All other programms seem to work fine, no other strange
> messages whatsoever...

yeah that's the weird part that had me scratching my head as
well.  The other poster said it happens for him on reiserfs,
and I'm on ext3, so that would seem to rule out
filesystems...

hmm looking at -ac2 log, looks like the following might be
possible:

    - Fix AF_UNIX dgram select problem
       + postfix uses lots of these, this sounds like the
         most likely candidate for sure

    - Revert problematic scsi change (fixes ppa/imm)
       + does it touch generic code?

    - Fix data direction for start stop scsi command
       + have no idea what this is

    - Add more unexpected apic filtering
       + I'm on SMP...don't know if this is related??

I don't see how it could be any of the other changes.