On Nov 5, 2008, at 4:50 PM, J. Bruce Fields wrote:
On Wed, Nov 05, 2008 at 04:34:01PM -0500, Chuck Lever wrote:
On Nov 5, 2008, at 3:45 PM, J. Bruce Fields wrote:
On Wed, Nov 05, 2008 at 02:49:55PM -0500, bfields wrote:
By the way, I did some miscellaneous nsm cleanup a few months ago
and
never got around to sending it out. I'll pass that along in a
moment.
For which I meant to send out a cover letter and forgot, sorry!
These look similar to some prototype patches I've played with to
split
the host cache, except for the elimination of h_server and sm_sticky,
which would be dandy if we can get away with it. I also went to the
trouble of adding a SLAB for the nlm_host structures.
Basically they look OK to me, except I would rename expire_hosts() as
nlm_expire_hosts().
Would you like me to port these up to 2.6.28 and add an nlm_host
SLAB?
That would be fine. Or use your patches instead in any places where
you
think there's an advantage. The versions I sent were generated
against
2.6.28-rc3, so should be recent enough.
OK, I can do that. I'll apply most of your patches after the NSM IPv6
patches and the SLAB-ify patches. There might be some additional
optimizations, and I haven't figured out what to do about the
"redundant RPC client shutdown" question you raised.
There's a couple places (e.g. sm_sticky removal) where we need to
audit
and ensure nothing's broken, and leave notes in the commit message
sufficient to reconstruct the audit.
I might postpone these (either drop them, or put them at the end) so
we can have some more time to think carefully about what needs to be
done here. As I said before, IMO it would be cleaner to go without
h_server and sm_sticky, so I'm in favor of making these patches work
if possible.
And it'd be nice to figure out how to test nsm changes.
Hah! ;-)
--b.
Anyway, the first four patches are just small cleanup and
bugfixing in
the same area. The remaining patches split up the client's and
server's
host lists into two separate tables.
They could all use another look-over and some testing; just thought
you
might be interested as long as you're in the neighborhood.
Also todo in vaguely the same area:
- nlmsvc_invalidate_all should be done on nfsd shutdown, but
we're doing shutdown per transport for some reason, which
seems dumb. So fix that reference counting so there's just
one count for the whole server.
- it would be nice to be able to shutdown the server
independently of the client; currently there's no way to do
that, as there's no way for sm-notify to distinguish client
from server hosts. (There may be no solution in the case
where a single host has both client and server relationships
with us--but people probably shouldn't be doing that anyway
(it seems deadlock prone.)
- the debugging code in nlm_shutdown_hosts() may be overkill.
--b.
--
Chuck Lever
chuck[dot]lever[at]oracle[dot]com
--
Chuck Lever
chuck[dot]lever[at]oracle[dot]com
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
