squid on Linux corrupting filesystem

From: ing. AG van Loo <eddy@dont-contact.us>
Date: Fri, 26 Sep 1997 15:50:10 +0200

Hello,

I'm experiencing a nasty problem with our primary proxy server,
squid-1.1.15 on a Linux system (www-proxy1.HZeeland.nl).

The squid process is corrupting the filesystem that is used as the cache-dir.
This happens not at once, but only after the server has been running for
at least a week or more.

Then certain objects in the cache suddenly cannot be retrieved
anymore and error messages start to appear in the squid access.log
like this:
875272390.420 4057 145.19.52.59 ERR_READ_ERROR/400 1003 GET http://194.93.128.253/ShellGrad/IMAGES/formhome.GIF - DIRECT/194.93.128.253 -
875272407.365 4151 145.19.52.59 ERR_READ_ERROR/400 1001 GET http://194.93.128.253/ShellGrad/IMAGES/formany.GIF - DIRECT/194.93.128.253 -
875272504.748 10139 145.19.52.59 ERR_READ_ERROR/400 1001 GET http://194.93.128.253/ShellGrad/IMAGES/helpus5.GIF - DIRECT/194.93.128.253 -

In the /var/log/syslog, messages like this appear:
Sep 26 07:58:28 www-proxy1 kernel: EXT2-fs warning (device 03:04): ext2_free_blocks: bit already cleared for block 836390
Sep 26 07:58:28 www-proxy1 kernel: EXT2-fs warning (device 03:04): ext2_free_inode: bit already cleared for inode 206058

When the filesystem is checked (after the squid process is stopped and the
filesystem unmounted) an enormous amount of inconsistencies are reported
and fixed, but after a while the errors start appearing again.
Even when the filesystem is re-created (mke2fs) the problems come back after
a while.

All other filesystems (multiple EIDE and SCSI disks are in that system)
don't suffer these problems, only the filesystem where the cache-dir
points to.

First I assumed a defect in the harddrive on the filesystem used for squid,
so I've installed an extra 1.3GB disk (on the SCSI bus) used exclusively
for the cache-dir filesystem
But this gives the same problems, now on the filesystem on the new disk.

The system consists of a Intel Zappa board (Intel Advanced/ZP) with a
P100 processor, 256kB pipeline burst cache, 64MB EDO RAM.
Two EIDE disks are on the on-board IDE busses,
A Ultra-SCSI disk drive, a older SCSI-II drive, a Archive 2GB DAT drive
are on a Adaptec AHA2940UW (PCI) Ultra-Wide SCSI controller.
The OS is Linux (Slackware 3.1) with kernel 2.0.29

My secondary proxy server, a similar Linux box but without a SCSI controller
does not have this problem.

Is there a known problem with Linux, squid and Adaptec 2940UW ?

TIA, Eddy van Loo

-- 
Network Management, Unix Support, Surfnet-IC/ICP, Postmaster, Webmaster
at Hogeschool Zeeland, Vlissingen / Zeeland Polytechnic, Flushing, Holland.
Email: eddy@HZeeland.nl  Phone: +31-(0)118-489275  Fax: +31-(0)118-489200
WWW: http://www.HZeeland.nl/~eddy/
Received on Fri Sep 26 1997 - 07:07:30 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:10 MST