On Mon, Mar 31, 2008 at 04:58:06PM +0200, Valerie Clement wrote:

Aneesh Kumar K.V wrote:

quoted

I looked at the code. Nothing appears straight. I am now running tests.
meanwhile you can dump the block ee_pblock in the ext4_ext_zeroout
and see if we are zeroing some wrong blocks that would be great.

I didn't see anything incorrect here for the moment, but adding traces
in the code often changes the behavior so that I can't reproduce the problem
running the same test.

quoted

quoted

How I reproduce the problem:
# mkfs.ext3 -I256 -E test_fs -b 1024 /dev/sdc1
# mount -t ext4dev /dev/sdc1 /mnt/test
# fsstress -d /mnt/test -n1000 -p1000

Does the fsstress you are using have fallocate support ?. If so can you 
send me the patch so that i can run the same test. 

No, the fsstress I'm using doesn't support fallocate.

That means it is not due to ext4_ext_zeroout. Which implies we have a
generic file system corruption.

quoted

Also can you disable delalloc and try.

OK, done. The tests are still running.

quoted

ENOSPC handling with delalloc is not yet done

I often got the problem when the disk is filled to 10% of its capacity.

I actually added fallocate to fsstress and created the file filesystem
as you suggested. I am able to reproduce the problem once. Currently
doing a code audit. Will let you know if i make any progress.


-aneesh