On Thu, 2020-06-18 at 18:12 -0500, Benjamin Marzinski wrote:
> On Thu, Jun 18, 2020 at 08:37:20PM +0000, Martin Wilck wrote:
> >
> > > -int dm_flush_maps (int retries)
> > > +int dm_flush_maps (int need_suspend, int retries)
> > > {
> > > int r = 1;
> > > struct dm_task *dmt;
> > > @@ -1014,7 +1014,10 @@ int dm_flush_maps (int retries)
> > >
> > > r = 0;
> > > do {
> > > - r |= dm_suspend_and_flush_map(names->name, retries);
> > > + if (need_suspend)
> > > + r |= dm_suspend_and_flush_map(names->name,
> > > retries);
> > > + else
> > > + r |= dm_flush_map(names->name);
> >
> > This is fine, but considering the previous discussion, I'd prefer
> > to
> > get rid of need_suspend and dm_suspend_and_flush_map() entirely. It
> > would simplify the _dm_flush_map code path significantly.
> >
> > As we determined that we use the suspend/resume only in multipath
> > anyway, we could add it there in the "non-delegated" code path.
>
> I'm confused. dm_flush_maps() is also called in the non-delegated
> code
> path, where it uses the dm_suspend_and_flush_map(). Are you proposing
> an alternative version of dm_flush_maps() for multipath?
No, just a different error handling in multipath. Instead of calling
just calling dm_flush_maps() again from multipath with need_suspend=1,
we could suspend / resume directly in the error path, and then call
dm_flush_map again. Which would allow us to get rid of need_suspend, as
no callers would set it any more.
IOW, move the handling of this multipath-specific error situation out
of libmultipath, into the multipath-specific code.
Just an idea, I don't insist on it.
Martin
--
Dr. Martin Wilck <mwilck@xxxxxxxx>, Tel. +49 (0)911 74053 2107
SUSE Software Solutions Germany GmbH
HRB 36809, AG Nürnberg GF: Felix
Imendörffer
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
