Re: [PATCH 00/18 V2] Repair SWAP-over-NFS

On Sat, 18 Dec 2021, Anna Schumaker wrote:

Hi Neil,

On Thu, Dec 16, 2021 at 7:07 PM NeilBrown [off-list ref] wrote:

quoted

swap-over-NFS currently has a variety of problems.

swap writes call generic_write_checks(), which always fails on a swap
file, so it completely fails.
Even without this, various deadlocks are possible - largely due to
improvements in NFS memory allocation (using NOFS instead of ATOMIC)
which weren't tested against swap-out.

NFS is the only filesystem that has supported fs-based swap IO, and it
hasn't worked for several releases, so now is a convenient time to clean
up the swap-via-filesystem interfaces - we cannot break anything !

So the first few patches here clean up and improve various parts of the
swap-via-filesystem code.  ->activate_swap() is given a cleaner
interface, a new ->swap_rw is introduced instead of burdening
->direct_IO, etc.

Current swap-to-filesystem code only ever submits single-page reads and
writes.  These patches change that to allow multi-page IO when adjacent
requests are submitted.  Writes are also changed to be async rather than
sync.  This substantially speeds up write throughput for swap-over-NFS.

Some of the NFS patches can land independently of the MM patches.  A few
require the MM patches to land first.

Thanks for fixing swap-over-NFS! Looks like it passes all the
swap-related xfstests except for generic/357 on NFS v4.2. This test
checks that we get -EINVAL on a reflinked swapfile, but I'm not sure
if there is a way to check for that on the client side but if you have
any ideas it would be nice to get that test passing while you're at
it!

Thanks for testing!.
I think that testing that swap fails on a reflinked file is bogus.  This
isn't an important part of the API, it is just an internal
implementation detail.
I certainly understand that it could be problematic implementing swap on
a reflinked file within XFS and it is perfectly acceptable to fail such
a request.  But if one day someone decided to implement it - should that
be seen as a regression?

Certainly over NFS there is no reason at all not to swap to a file that
happens to be reflinked on the server.
I don't think it even makes sense to test if the file has holes as the
current nfs_swap_activate() does.  I don't exactly object to the test,
but I think it is misguided and pointless.

Thanks,
NeilBrown