segfault with data= option in fstab

Binary package hint: mdadm

Reported last night, cannot find in launchpad (!?). If 9/5 version appears, use THIS report. Included more information.

Other mdadm segfaults seem to be different conditions, many are closed, some long ago. Hence new report.

9.04 desktop CD.

When I include "data=journal" on /boot in fstab, or "data=ordered" in / in fstab, mdadm gives:

Sep 5 21:02:29 bagend kernel: [ 14.835579] mdadm[1259]: segfault at 102 ip b7e995eb sp bfab6110 error 4 in libc-2.9.so[b7e22000+15c000]

Note that my md array is NOT / or /boot. These are simple partitions. Details are below. This is especially disturbing because a condition which has nothing to do with md is causing md to segfault. To me this suggests corruption of kernel memory, which could be any code in the kernel, affecting anything! For this reason I view this bug as "critical". I would value your assessment, since I need to decide whether to go ahead or drop back to 8.10. Backups get overwritten, I don't have the resources to put a backup on the shelf while continuing regular backups. PLEASE give me your opinion! <email address hidden>

System is P3-800 w/768MB. Has been running 8.10 with same HW same exact disk/partition configuration 3/2009 to Sept3, 2009 without issue. System (with different physical drives) ran various versions of Suse for 6-8 years previously using md and lvm2 without issue, albiet with version 0.9 md Superblocks.

I erased the disks (Including zeroing the MD array disks) Sept 3 and did a clean 9.04 desktop CD install. sda is an 80GB PATA on the motherboard IDE0 (Intel 440BX chipset). sda1=/boot, sda5=/, sda6=swap. All partitions ext3 and reformatted.

Ran "Update" and applied everything offered.

sdb and sdc are 500GB PATA on the motherboard IDE0 and IDE1. Used mdadm to configure md0 out of sdb and sdc (Entire device, no partition table) using superblock 1.2 (Same as 8.10 system). Created lvm2 volume group and logical volume using the entire disks, created ext3 file system. Specify data=ordered in fstab.

Restored file system on md0/vg/lv from backup using "cp -rp", did verify (diff -r), waited for resync to complete (40 hours, diff and resync were concurrent).

Configured various things like automounter, NFS, etc. Very simple one liners.

Rebooted. System would not reassemble md array. Did NOT run any mdadm commands that modify anything, just "mdadm --examine". Figured out that the UUID of my array did not match the UUID in /etc/mdadm/mdadm.conf. It matched the UUID of my 8.10 system. I have never been aware of mdadm.conf previously. It must have been created in 8.10 by something automatically.

Did a "mdadm --examine --scan >> /etc/mdadm/mdadm.conf" and removed the incorrect line. Now system boots and assembles.

I like to explicitly list all options in fstab, not have them default.

When I added data=ordered to / and data=journal to /boot, mdadm segfaults (as shown above). Same segfault occurs with either one alone. Take them both out (as in fstab attached) and system runs fine.

Attached is a gzipped tar file with dmesg, fstab, mdadm.examine.sdb (and sdc), messages, syslog, "tune2fs -l" of all filesystems.

Tell me what else to send.

--Ray

mdadm:
  Installed: 2.6.7.1-1ubuntu8
  Candidate: 2.6.7.1-1ubuntu8
  Version table:
 *** 2.6.7.1-1ubuntu8 0
        500 http://us.archive.ubuntu.com jaunty/main Packages
        100 /var/lib/dpkg/status