Lustre discuss - Oct 2007 - lustre e2fsck corrupts files by changing i_size

Hello,

We have an IA64 cluster with 64K kernel page size and we use lustre 
1.4.8.1 and e2fsprogs-1.39cfs8.
We meet the following problem :
1. I Create a Lustre file system, then start this file system and mount 
it on a client node
2. On this client node, I copy 2 files in my /mnt/lustre
3. I umount Lustre file system on this client node then stop Lustre file 
system
4. I use then : /usr/lib/lustre/e2fsck /dev/OSTs-device

And big surprise :  for each files created on the Lustre file system, it 
displays the following :
Pass 1: Checking inodes, blocks, and sizes
Inode 163882, i_size is 199041, should be 262144.  Fix?

The real size of my file is 199041 ! So if I choose "Yes" to the 
question, This file will be corrupted.

Additional info : my device Block size is 4096

I think there is an error in the test code of the phase1. What do you 
think ? Do I need to open a bugzilla for this very serious problem for 
us or do I miss something ?

I tested this case with e2fsprogs-1.39-cfs8 and e2fsprogs-cfs2 and I 
didn''t find a ChangeLog for e2fsprogs provided by CFS. So perhaps this 
problem is corrected in another release ?

Regards,
C?dric Lambert

On Oct 04, 2007  15:12 +0200, C?dric Lambert wrote:
> We have an IA64 cluster with 64K kernel page size and we use lustre 
> 1.4.8.1 and e2fsprogs-1.39cfs8.
> We meet the following problem :
> 1. I Create a Lustre file system, then start this file system and mount 
> it on a client node
> 2. On this client node, I copy 2 files in my /mnt/lustre
> 3. I umount Lustre file system on this client node then stop Lustre file 
> system
> 4. I use then : /usr/lib/lustre/e2fsck /dev/OSTs-device
> 
> And big surprise :  for each files created on the Lustre file system, it 
> displays the following :
> Pass 1: Checking inodes, blocks, and sizes
> Inode 163882, i_size is 199041, should be 262144.  Fix?
> 
> The real size of my file is 199041 ! So if I choose "Yes" to the 
> question, This file will be corrupted.
> 
> Additional info : my device Block size is 4096
> 
> I think there is an error in the test code of the phase1. What do you 
> think ? Do I need to open a bugzilla for this very serious problem for 
> us or do I miss something ?
To follow up on this, bug 13812 was filed and a defect in e2fsck was
found.  This will be fixed in the upcoming e2fsprogs-1.40.2-cfs2 release.

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.