[02:07:03] --- squinney has become available
[02:13:21] --- Stephan Wiesand has become available
[04:56:53] --- jaltman/FrogsLeap has left: Disconnected
[05:01:12] --- jaltman/FrogsLeap has become available
[05:10:08] --- abo has left
[05:15:07] --- abo has become available
[05:30:08] --- mvita has become available
[06:25:47] --- ballbery has become available
[06:27:32] --- meffie has become available
[06:31:55] --- mvita has left
[06:40:18] --- mvita has become available
[07:30:33] --- deason has become available
[08:33:01] <meffie> > (Instead we'd argue about what options we tell people to use :)
ha!
[08:35:21] --- squinney has left
[09:40:32] --- Stephan Wiesand has left
[09:42:44] <kaduk@mit.edu/barnowl> Hmm, getting hcrypt's prng in the kernel on irix could be more
annoying than anticipated ... hacking stubs to get rand-fortuna.c
built in the kernel reveals that (aside from needing to pull in
aliases for rx_kmutex functionality and tweak it) the fortuna code
wants getpid, gettimeofday, abort, open, rk_cloexec, read, close, and
getuid.
[09:50:35] <Simon Wilkinson> Nah, it doesn't.
[09:50:42] <Simon Wilkinson> They're all just there to seed the RNG
[09:52:01] <kaduk@mit.edu/barnowl> It checks for pid changes to reseed after fork.
[09:52:15] <kaduk@mit.edu/barnowl> I'm stubbing them out in config.h for the moment, we'll see how it
looks.
[09:52:42] <Simon Wilkinson> Love is quite happy to take changes to make Heimdal work in kernel, providing they're suitably protected from the rest of the code.
[09:54:00] <kaduk@mit.edu/barnowl> I'm not sure what that sort of protection would look like -- are there
examples to look at?
[09:54:22] <Simon Wilkinson> #ifdef KERNEL
[09:54:35] <kaduk@mit.edu/barnowl> Ah, that's easy enough :)
[09:54:59] <kaduk@mit.edu/barnowl> We would have to take those as new imports from upstream, though, not
patching things in external/ manually?
[09:55:09] <Simon Wilkinson> Yeah.
[09:55:18] <kaduk@mit.edu/barnowl> Does he like github pull requests or something else?
[09:55:50] <Simon Wilkinson> github pull requests, or git format-patches
[09:56:00] <kaduk@mit.edu/barnowl> Okay.
[09:56:25] <Simon Wilkinson> There's a bunch of us with commit access - it may be quicker to poll one of us (me, Nico, Roland, …) than going directly to Love.
[09:56:47] <kaduk@mit.edu/barnowl> I guess I should clone heimdal, then.  I could certainly put something
up on my fork and mention it if that works, too.
[09:57:11] <Simon Wilkinson> I tend start by modifying src/external, and then working from there.
[09:57:49] <kaduk@mit.edu/barnowl> That was the plan.  (Well, get one or two more tweaks in locally and
then revert it and do it again in external/.)
[10:23:59] --- meffie has left
[10:29:26] --- stephan.wiesand has become available
[10:48:58] --- mvita has left
[10:53:07] --- mvita has become available
[11:49:29] --- mvita has left
[11:49:42] --- mvita has become available
[11:50:10] --- mvita has left
[11:56:03] --- mvita has become available
[11:56:05] --- mvita has left
[11:59:30] --- mvita has become available
[12:13:14] <kaduk@mit.edu/barnowl> Whoops, kernel panic corrupted my repository, that was exciting.
Anyway, rand-fortuna wants to just initialize fortuna_mutex to a
constant value, but rx_kumtex_t only provides a MUTEX_INIT macro that
must be called with the mutex as an argument.
[12:48:13] <kaduk@mit.edu/barnowl> I suppose I could borrow the GLOCK...
[12:58:24] <Simon Wilkinson> Not in RX
[13:00:18] <kaduk@mit.edu/barnowl> Any suggestions other than making a new global lock and initializing
it very early?
[13:03:09] <Simon Wilkinson> Depends on the problem that you are trying to solve...
[13:03:33] <Simon Wilkinson> Is this just going to be used for providing epoch and cid values, or is it a general kernel PRNG?
[13:04:49] <kaduk@mit.edu/barnowl> I'm still operating on the presumption that it will be a general
kernel PRNG.  If it was just to seed epoch/cid, there are much simpler
solutions.
[13:05:36] <Simon Wilkinson> Well, you're going to have to seed it before you can use it.
[13:05:44] <kaduk@mit.edu/barnowl> (rxgk will need some entropy as well.)
[13:05:52] <Simon Wilkinson> Why not just initialise the mutex in the call you use to seed it?
[13:06:15] <kaduk@mit.edu/barnowl> Right, and there's already code in afsd to "seed the kernel rng"
[13:06:49] <Simon Wilkinson> You've seen afsd's attempt at an RNG, right?
[13:07:39] <kaduk@mit.edu/barnowl> The simple way to do such initialization at seed time would make the
syscall to pass entropy in from userspace a one-time-only offering.
That might be fine, but I don't know that I want to make that decision
without explicit consideration.
[13:08:50] <kaduk@mit.edu/barnowl> > afsd's attempt at an RNG
I'm not sure I know what you mean.
I see the bits where we call AFSOP_ADVISEADDR and the kernel uses that
and some timestamps and such.
[13:09:41] <Simon Wilkinson> Just use an atomic to protect the initialisation
[13:10:56] <kaduk@mit.edu/barnowl> Okay.
[13:42:52] --- mvita has left
[13:47:19] <kaduk@mit.edu/barnowl> Actually, am I even supposed to use the afs_kmutex_t type outside of
rx?
[13:49:50] <Simon Wilkinson> For kernel code, yes.
[13:50:54] <Simon Wilkinson> What are you using for locking in rxgk?
[13:51:12] <kaduk@mit.edu/barnowl> Nothing, yet :)
[14:16:08] --- ballbery has left
[14:46:52] --- mvita has become available
[14:49:25] --- mvita has left
[16:10:28] --- stephan.wiesand has left
[16:39:10] <kaduk@mit.edu/barnowl> Bah, clang is unhappy about the use of panic() in defining osi_Panic()
for JUAFS on FreeBSD.
[17:09:53] --- deason has left
[20:12:54] --- Brandon Allbery has left
[23:59:45] --- Simon Wilkinson has left