A question on umounting without flushing journal

A question on umounting without flushing journal

[Bo Branten]

Hello,

now I have another question on how to test the ext4 driver we implement 
for another os: At mount time the driver should check the journal and use 
jbd2 to replay if there is anything left in it and I would like to ask if 
there is any other methods to do an unclean umount in linux besides 
pressing reset? I would like to leave a lot of records in the journal that 
our driver can try to process?

Bo Branten

Re: A question on umounting without flushing journal

[Theodore Y. Ts'o]

On Tue, Mar 03, 2020 at 11:26:13AM +0100, Bo Branten wrote:
> 
> now I have another question on how to test the ext4 driver we implement for
> another os: At mount time the driver should check the journal and use jbd2
> to replay if there is anything left in it and I would like to ask if there
> is any other methods to do an unclean umount in linux besides pressing
> reset? I would like to leave a lot of records in the journal that our driver
> can try to process?

The standard way to do this in xfstests is using the dm-flakey
device-mapper device.  We set it up so that all reads and writes are
passed through, and then we start some workload such as fsstress, and
then we reconfigure dm-flakey to drop 100% of all write requests to
the underlying block device.

We then kill the workload, and unmount the file system, and then we
reset the dm-flakey device to pass through 100% of all writes.  This
simulates quite accurately what the block device would look like after
a sudden power failure, but it doesn't require a power-fail rack
(without shortening the lives of the equipment; dropping power tends
to put a lot stress on the hardware).

Cheers,

					- Ted