[Drbd-dev] [PATCH 22/23] drbd: implement REQ_OP_WRITE_ZEROES

[Mike Snitzer]

On Thu, Mar 30 2017 at 11:20am -0400,
Martin K. Petersen <martin.petersen at oracle.com> wrote:

> Mike Snitzer <snitzer at redhat.com> writes:
> 
> > I can work on this now.  Only question I have is: should DM thinp take
> > care to zero any misaligned head and tail?  (I assume so but with all
> > the back and forth between Bart, Paolo and Martin I figured I'd ask
> > explicitly).
> 
> Yep, let's make sure our semantics match the hardware ditto.
> 
>  - So write zeroes should behave deterministically and explicitly handle
>    any blocks that can't be cleared via deprovisioning.
> 
>  - And discard can work at the discard granularity in a
>    non-deterministic fashion.

I got pretty far along with implementing the DM thinp support for
WRITE_ZEROES in terms of thinp's DISCARD support (more of an
implementation detail.. or so I thought).

But while discussing this effort with Jeff Moyer he asked: shouldn't the
zeroed blocks be provisioned?  This is a fairly embarassing question not
to be able to answer in the moment.  So I clearly need to learn what the
overall intent of WRITE_ZEROES actually is.

If it is meant as a replacement for WRITE_SAME (as hch switched dm-io
over from WRITE_SAME with a payload of 0 to WRITE_ZEROES) and for the
backing mechanism for blkdev_issue_zeroout() then I guess I have my
answer.  Unless DM thinp can guarantee that the discarded blocks will
always return zeroes (by zeroing before all partial block writes) my
discard based dm-thinp implementation of WRITE_ZEROES is a complete
throw-away (unless block zeroing is enabled.. which it never is because
performance sucks with it).  So if an upper-level of the IO stack
(e.g. ext4) were to assume that a block will _definitely_ have zeroes
then DM thinp would fall short.

This is all to say: I don't see a quick way forward on implementing
performant WRITE_ZEROES support for DM thinp.