Linux 2.4.21-ac3

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

Thijs on Wed 25/06 22:29 +0200:

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=<[email protected]>, 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=<[email protected]>
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.

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.

I can confirm this problem. I use postfix 2.0.12 on a Debian woody installation, all mails get corrupted. Postfix queue in on reiserfs, so I think the bug is filesystem independent.

Geller Sandor <[email protected]>