[linux-lvm] LVM1 & XFS

[linux-lvm] LVM1 & XFS

[Adrian Head]

[-- Attachment #1: Type: text/plain, Size: 2076 bytes --]

This is just a heads up for others...

Early 2002 I was trying to get LVM1, ext3, resierfs & XFS coexisting.  The 
SGI guys were able to generate a patch that fixed issues with LVM & XFS which 
can be found in the mailing list archives and modified lvm-snap.c.  At that 
time I was able to run a series of tests by hand and everything was OK; 
whereas if I put the cammands in a script it would die a death.  Because of 
other priorities I had to leave this ... so since LVM-1.0.2 was out and the 
XFS CVS keeps marching along and I had a few hours to spare I thought I would 
see if any of the issues have been resolved.

I have got the latest XFS from CVS (2002-02-02) and the latest LVM (1.0.2) 
from CVS and ran through my tests.  The tests outline follows:

# This is to test operational aspects of LVM
#  1.  Create LV
#  2.  mkfs on LV
#  3.  mount LV
#  3a. ls LV
#  4.  copy data onto LV
#  4a  ls LV
#  5.  copy data across LV
#  5a  ls LV
#  6.  copy data off LV
#  6a  ls LV
#  7.  overfill LV
#  7a. ls LV
#  8.  create snapshot
#  9.  mount snapshot
#  10. ls snapshot
#  11. overfill snapshot
#  12. ls snapshot

There still seems to be problems as I can run the tests by hand successfully 
but as soon as I put the commands in a script I get Oopses.

The positive side is that LVM-1.0.2 is way more robust than LVM-1.0.1 as when 
overflowing the snapshot fails; there are no kernel panics on the next reboot 
and no need for rescuing the LV associated with the snapshot.  However, when 
it does Oops it seems to hold the original LV open and therefore, the LV 
cannot be umounted properly - a reboot is the only way of getting the LV back 
to a usable state.

I have tried both vanilla linux-xfs+lvm-1.0.2 as well as 
linux-xfs+lvm-1.0.2+lvm-snap.c.patch and it still dies in the same place but 
the backtraces are different.  

The question I would like to ask any LVM gurus is why would there be a 
difference between running the tests by hand and running them in a script?

Attached is the test script...

-- 
Adrian Head

(Public Key available on request.)

[-- Attachment #2: lvm_test.sh --]
[-- Type: application/x-shellscript, Size: 2024 bytes --]

Re: [linux-lvm] LVM1 & XFS

[Andreas Dilger]

On Feb 03, 2002  15:07 +1000, Adrian Head wrote:
> There still seems to be problems as I can run the tests by hand successfully 
> but as soon as I put the commands in a script I get Oopses.

Well, running the oops through ksymoops and posting it would be a good
start to figuring out where the problems are.

> The question I would like to ask any LVM gurus is why would there be a 
> difference between running the tests by hand and running them in a script?

Timing issues.  Maybe it isaccessing memory that has been freed already.
Maybe it isn't getting a lock on some stuct and if you do it slowly, the
operation has a chance to complete before another one deletes the struct.
Who knows, the ksymoops output will hopefully point us where to look.

Cheers, Andreas
--
Andreas Dilger
http://sourceforge.net/projects/ext2resize/
http://www-mddsp.enel.ucalgary.ca/People/adilger/

Re: [linux-lvm] LVM1 & XFS

[Adrian Head]

Thanks Andreas for answering my questions.

The email was just a heads up to see if other people had seen something 
simular.  Chris Pascoe suggested a few modifications in the XFS code and this 
seems to have fixed this problem for the moment - so I will be giving more 
info to the XFS guys and see what happens.

Thanks Andreas.

-- 
Adrian Head

(Public Key available on request.)