HEADS UP: Starting socket locking merge

From: Robert Watson <rwatson_at_FreeBSD.org>
Date: Sat, 12 Jun 2004 17:09:32 -0400 (EDT)
Over the past few days, I've merged locking for the accept path, locking
for UNIX domain sockets, and several pieces of infrastructure necessary to
support locking of the socket layer.  These changes have mostly gone in
without too much disruption (there was one file descriptor leak in the
accept changes).  I'm now beginning a merge of socket locking itself. 
Because I'm breaking this out into a series of function commits and
seeking final review for each step from several developers, this process
will take a week or so to complete.  I will endeavour, at each step along
the way (and ideally after each commit) to leave the system in a
compilable and functional state, but the changes are good that as exposure
for this work increases, so will the discovery of bugs :-).  As such, I'm
sending this e-mail as a warning that there may be some additional
turbulence during the merge. 

For those interested, a change log and patch candidates can be found at: 

    http://www.watson.org/~robert/freebsd/netperf/

I'm doing a selective merge, so the more risky or less complete changes
won't go in immediately, and likewise, a number of the annotations won't
be merged.  For example, I will not be merging the netatalk locking
changes until I've made sure they're complete and run them by the cadre of
volunteers I solicited previously for help in testing those changes. 

Please don't run with debug.mpsafenet turned on until I've sent out a note
detailing when that will (and won't) work.  Per earlier notes to
freebsd-current and the comments in updating, the semantics of
debug.mpsafenet have been modified since it was originally introduced, and
it is only safe to use if you are exercising paths in the stack that are
entirely Giant-safe.

As we've had mbuma merged recently, altq, and several other networking
changes in addition to this, I'd ask people be particularly diligent in
reporting new poor behavior in the network stack.  Because we've also seen
several fixes along the way, I'd also ask you make sure you're up-to-date. 

Thanks,

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert_at_fledge.watson.org      Senior Research Scientist, McAfee Research
Received on Sat Jun 12 2004 - 19:11:19 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:57 UTC