xfs-masters
[Top] [All Lists]

[xfs-masters] [Bug 417] Reading snapshot gives error in debug.c / xfs_bm

To: xfs-master@xxxxxxxxxxx
Subject: [xfs-masters] [Bug 417] Reading snapshot gives error in debug.c / xfs_bmap_search_extents
From: bugzilla-daemon@xxxxxxxxxxx
Date: Thu, 7 Jul 2005 06:12:40 -0700
Reply-to: xfs-masters@xxxxxxxxxxx
Sender: xfs-masters-bounce@xxxxxxxxxxx
http://oss.sgi.com/bugzilla/show_bug.cgi?id=417





------- Additional Comments From schaffoe@xxxxxxxxxxxxxxxxxxxxxxxx  2005-07-07 
06:12 CST -------
(In reply to comment #1)
> Does it always fail on this same inode?

In general, no. When the same thing, e.g. backup, is done repeatedly (without
rebooting the machine), it is always the same inode. Other operations fail at
other inodes.

> Out of curiosity, are you using the 4KSTACKS config option?  It crosses my
> mind that this -could- be stack corruption, although at the time of the oops,
> there does not seem to be significant stack pressure.

No, 4KSTACKS are not enabled.

> Ideally, some sort of a crashdump would be very helpful, or running this
> on a kernel with KDB to do some poking around on the system... not sure if
> any of this is possible.

I have never done this before; is there a good and not-too-long tutorial 
anywhere?

On the other hand, I have found an absolutely fool-proof way to trigger the bug.
First my setup:

- EasyRaid X12 (12 x 120GiB) with several partitions. One of these partitions is
the (one and only) PV in a LVM2 VG. Three LVs in this LVM2 VG. One formatted
with 200GiB XFS (later mounted on /home and snapshot), two formatted with EXT3
(one 100GiB, mounted on /var and snapshot also).

- 4x Seagate 72GiB SCSI disks with 1 partition each, glued together as PVs in
another LVM2 VG. Two LVs created from this VG, one is 100GiB for the
home-snapshot and the other is 30GiB for the var-snapshot. Snapshot volumes are
created using EVMS 2.5.2 which, I guess, uses dm-snapshot.

Basically, XFS-home and EXT3-var are in the same VG, and snap-home and snap-var
are located in the same, but different VG.

Here is how to trigger the bug:

1) From LTP, run testcases/kernel/fs/fsstress -l 0 -d /home/fsstress -n 1000 -p
50 -r . This is okay for several hours (actually never failed).

2) Activate and mount the EVMS snapshot of home on /snap/home . Still okay.

3) Run: #find /snap/home/fsstress -type f -exec cat '{}' ';' &> /dev/null
   You have the oops.

If you execute these steps with /var (EXT3) instead of /home (XFS), everything
runs fine for several hours and repetitions.

Log-excerpts are attached.



-- 
Configure bugmail: http://oss.sgi.com/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.


<Prev in Thread] Current Thread [Next in Thread>