read only filesystem tests

read only filesystem tests

[Radoslav Kolev]

Hello,

I am interested in read only filesystem tests, and more specifically 
squashfs. I saw that ~2 year ago there was a patch set submitted, but 
not accepted that added support for it with some feedback, that
the tests are copied and double the maintenance burden, but sadly there 
is no easy way to avoid that.

What is the current mood about read only filesystem tests and fstests?
Is this the right project to fill this need and is it worth it trying to 
address some of the problems with the old squashfs patches?

Best regards,
Radoslav

Re: read only filesystem tests

[Eryu Guan]

On Mon, Nov 15, 2021 at 01:18:25PM +0200, Radoslav Kolev wrote:
> Hello,
> 
> I am interested in read only filesystem tests, and more specifically
> squashfs. I saw that ~2 year ago there was a patch set submitted, but not
> accepted that added support for it with some feedback, that
> the tests are copied and double the maintenance burden, but sadly there is
> no easy way to avoid that.
> 
> What is the current mood about read only filesystem tests and fstests?

I'd like to see tests for readonly filesystems, squashfs is definitely
one target, and other readonly filesystems like erofs could benefit from
the tests as well.

> Is this the right project to fill this need and is it worth it trying to
> address some of the problems with the old squashfs patches?

I think so, at least I'm happy to see readonly tests in fstests
personally. But, to be honest, I still didn't come out with a good way
to share existing generic tests for RO filesystems.. And I think it's a
good opportunity to use wisdom from the community :)

Thanks,
Eryu

Re: read only filesystem tests

[Dave Chinner]

On Mon, Nov 15, 2021 at 01:18:25PM +0200, Radoslav Kolev wrote:
> Hello,
> 
> I am interested in read only filesystem tests, and more specifically
> squashfs. I saw that ~2 year ago there was a patch set submitted, but not
> accepted that added support for it with some feedback, that
> the tests are copied and double the maintenance burden, but sadly there is
> no easy way to avoid that.
> 
> What is the current mood about read only filesystem tests and fstests?
> Is this the right project to fill this need and is it worth it trying to
> address some of the problems with the old squashfs patches?

What sort of read only tests are you talking about? i.e. how does
the initial filesystem state that the test is going to run on get
generated? AFAICT you can't reuse any of the existing tests because
they all require writing to the filesystem to create the initial
test conditions. You can tell fsx/fsstress not do do modifying
operations from the command line, so you can still run stress tests
based on these on read-only filesystems.

There are definitely limits on what fstests can realistically
exercise w.r.t. read-only filesystems.  That said, there is
definitely scope for fstests to support read-only tests - just
create a tests/read-only directory, a common/read-only include file
and build the tests and infrastructure out from that. If the RO
tests have their own test group definitions (e.g. 'check -g
ro-auto'), then it won't impact on any of the existing read-write
filesystem tests or test groupings.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com