Re: [RFC PATCH 0/2] ceph: safely use 'copy-from' Op on Octopus OSDs

[Luis Henriques <lhenriques@suse.com>]

On Fri, Nov 08, 2019 at 05:22:28PM +0000, Sage Weil wrote:
> On Fri, 8 Nov 2019, Luis Henriques wrote:
> > On Fri, Nov 08, 2019 at 04:57:27PM +0000, Sage Weil wrote:
> > > On Fri, 8 Nov 2019, Luis Henriques wrote:
> > > > On Fri, Nov 08, 2019 at 04:15:35PM +0100, Ilya Dryomov wrote:
> > > > > If the OSD checked for unknown flags, like newer syscalls do, it would
> > > > > be super easy, but it looks like it doesn't.
> > > > > 
> > > > > An obvious solution is to look at require_osd_release in osdmap, but we
> > > > > don't decode that in the kernel because it lives the OSD portion of the
> > > > > osdmap.  We could add that and consider the fact that the client now
> > > > > needs to decode more than just the client portion a design mistake.
> > > > > I'm not sure what can of worms does that open and if copy-from alone is
> > > > > worth it though.  Perhaps that field could be moved to (or a copy of it
> > > > > be replicated in) the client portion of the osdmap starting with
> > > > > octopus?  We seem to be running into it on the client side more and
> > > > > more...
> > > > 
> > > > I can't say I'm thrilled with the idea of going back to hack into the
> > > > OSDs code again, I was hoping to be able to handle this with the
> > > > information we already have on the connection peer_features field.  It
> > > > took me *months* to have the OSD fix merged in so I'm not really
> > > > convinced a change to the osdmap would make it into Octopus :-)
> > > > 
> > > > (But I'll have a look at this and see if I can understand what moving or
> > > > replicating the field in the osdmap would really entail.)
> > > 
> > > Adding a copy of require_osd_release in the client portion of the map is 
> > > an easy thing to do (and probably where it should have gone in the first 
> > > place!).  Let's do that!
> > 
> > Yeah, after sending my reply to Ilya I took a quick look and it _seems_
> > to be as easy as adding a new encode(require_osd_release...) in the
> > OSDMap.  And handle the versions, of course.  Let me spend some time
> > playing with that and I'll try to come up with something during the next
> > few days.
> 
> - You'll need to add it for both OSDMap::Incremental and OSDMap
> - You'll need to make the encoding condition by updating the block like 
> the one below from OSDMap::encode()
> 
>     uint8_t v = 9;
>     if (!HAVE_FEATURE(features, SERVER_LUMINOUS)) {
>       v = 3;
>     } else if (!HAVE_FEATURE(features, SERVER_MIMIC)) {
>       v = 6;
>     } else if (!HAVE_FEATURE(features, SERVER_NAUTILUS)) {
>       v = 7;
>     }
> 
> to include a SERVER_OCTOPUS case too.  Same goes for Incremental::encode()

Awesome, thanks!  I'll give it a try, and test it with the appropriate
kernel client side changes to use this.

Cheers,
--
Luís