At long last, we're ready to start the Clearview IPMP code review.
The review is divided into two phases:
* Phase 1 covers the removal of the existing Nevada IPMP kernel
implementation. We're starting this now.
* Phase 2 covers the Clearview IPMP kernel implementation, plus all
userland changes. This is the real meat of the project, and will
start in a couple of weeks.
If you're interested in reviewing either or both phases, please let us
know, and please feel free to dig into Phase 1 (details below). Note that
there will only be one integration -- the Phase 2 codebase. However,
because the Clearview IPMP kernel implementation is entirely[1] new, we've
elected to split up the review to ensure Phase 2 reviewers will not be
distracted by meaningless diffs against the old (and irrelevant) Nevada
IPMP kernel implementation.
The remainder of this message discusses Phase 1; we'll provide details for
Phase 2 separately once it's ready.
As stated above, Phase 1 covers the removal of the Nevada IPMP kernel
implementation. While this is a large amount of code change (roughly 9500
lines), the lion's share of it is code removal or comment rework. Indeed,
there are less than 150 new lines of code in this wad. As such, we
believe a week and a half is enough time for this part of the review (we
will be allocating twice as much time for Phase 2).
Please be aware that the IPMP Phase 1 sourcebase is itself based on the
Clearview IP Observability sourcebase, which in turn is based on the
latest Nevada. This has been done for several reasons, but especially
because Phase 2 provides IPMP IP observability support, which is only
possible with the Clearview IP Observability bits. Since IP Observability
is also scheduled to integrate into Nevada before Clearview IPMP, the
code review also matches what we plan to integrate (if this changes, we
will of course provide diffs for review). Also, note that the actual
impact of this is quite small because the changes to the IP sourcebase for
IP Observability are minimal (mostly some changes to the hooks and some
multicast kernel API enhancements).
Because IPMP integrates deep into IP, the removal of IPMP means more
than just weeding out the obviously-unused functions. For instance, the
the IPSQ codepaths have been noticeably simplified since IPSQ<->ill
bindings can no longer change, and all the special-case ifindex handling
for multicast is no longer necessary. In general, we have tried to be
thorough in eradicating all tentacles of complexity that were due to the
Nevada IPMP implementation, but there was a lot of code to sift through,
and we would welcome more eyes. In some corner cases, we have chosen to
leave behind certain aspects of the original design because they were
useful in the new codebase as well, but those are relatively rare and
hopefully will not impede anyone's quest :-)
Finally, Phase 1 also includes a collection of small formatting and
comment fix-ups to the Nevada codebase that proved useful to the Phase 2
codebase but that we didn't want cluttering up the Phase 2 diffs.
With that background, the webrev is available at:
http://cr.opensolaris.org/~meem/clearview-noipmp-cr/
A stable internal Mercurial repo is available at:
/net/zhadum.east/export/ws/clearview/clearview-noipmp-cr
Finally, a stable internal cscope is available at:
/net/zhadum.east/export/ws/clearview/clearview-noipmp-cr/usr/src
Thanks and have fun!
[1] OK, we did reuse a couple of ioctl entrypoints ;-)
--
meem
_______________________________________________
networking-discuss mailing list
[email protected]
