[00:31:34] --- Simon Wilkinson has become available
[01:01:04] --- Simon Wilkinson has left
[01:03:47] --- kaj has become available
[02:57:16] --- abo has left
[03:23:36] --- haba has become available
[03:36:55] --- Simon Wilkinson has become available
[03:47:07] --- jaltman has left: Replaced by new connection
[03:47:08] --- jaltman has become available
[03:52:18] --- Simon Wilkinson has left
[06:24:18] --- haba has left
[07:17:46] --- RedBear has left
[07:23:36] --- deason has become available
[07:42:13] --- meffie has become available
[08:22:31] --- reuteras has left
[09:11:17] --- jaltman has left: Disconnected
[09:11:44] --- jaltman has become available
[09:25:26] --- abo has become available
[09:44:47] --- haba has become available
[10:14:28] --- haba has left
[10:21:09] --- meffie has left
[10:21:48] --- meffie has become available
[10:22:20] --- meffie has left
[10:32:03] --- kaj has left
[10:44:51] --- RedBear has become available
[11:32:14] <jaltman> OpenAFS was accepted into GSoC 2010
[11:54:44] <kula> huzzah
[12:38:50] --- haba has become available
[13:23:59] --- kaj has become available
[13:24:55] --- Simon Wilkinson has become available
[13:25:36] --- Simon Wilkinson has left
[13:25:57] --- Simon Wilkinson has become available
[13:36:00] --- Russ has become available
[14:10:15] <kaj> Congratz!  What's the project?
[14:13:02] <kaduk@mit.edu/barnowl> Hm, it appears that I was sloppy in de-validating #1340 (xdr: stop the madness);
the build had previously been broken, but I had only been working on a 1.5.72
tarball previously.
[14:15:48] <kaduk@mit.edu/barnowl> Do I change my review to neutral?
[14:16:13] <Simon Wilkinson> What's the issue?
[14:16:23] <Simon Wilkinson> That the build is currently broken, but #1340 doesn't fix it?
[14:16:28] <kaduk@mit.edu/barnowl> Yes.
[14:16:42] <kaduk@mit.edu/barnowl> (INT_MAX not defined)
[14:16:43] <Simon Wilkinson> And you just need an appropriately applied limits.h?
[14:17:10] <kaduk@mit.edu/barnowl> It seems so.  I just threw it in sysincludes.h at random, and things
compiled.
[14:17:39] <kaduk@mit.edu/barnowl> Hm, or maybe not.
[14:17:55] <kaduk@mit.edu/barnowl> (not "build formerly broken, and 1340 doesn't fix", that is)
[14:18:07] <kaduk@mit.edu/barnowl> I should probably actually try it instead of trying to tell from reading code.
[14:18:45] <Simon Wilkinson> If you can let me know the error you're seeing that would be great ...
[14:20:10] <Simon Wilkinson> But, I suspect if you are broken, you were broken due to my change to add xdr_mem.c to the Unix kernel.
[14:20:17] <kaduk@mit.edu/barnowl> So, I have only tried building (a patched) 1.5.72, and master+1340
I haven't tried building just master.
master+1340 fails with 
xdr_mem.c:97: error: 'INT_MAX' undeclared (first use in this function)
[14:20:20] <Simon Wilkinson> And yes, XBSD doesn't include limits.h
[14:20:26] <kaduk@mit.edu/barnowl> Ah, yes, that would do it.
[14:20:53] <Simon Wilkinson> The question is whether its needed for XBSD, or just for FBSD
[14:21:39] <kaduk@mit.edu/barnowl> I'm not sure that I can answer that
[14:22:39] <Simon Wilkinson> sysincludes.h needs to be taken out and given a stern talking to.
[14:23:56] <Russ> The whole way we deal with headers in kernel space always confuses the crap out of me, although I'm sure it was for a good reason and it's probably hard to fix.
[14:25:09] <Simon Wilkinson> Too many people have used too many different styles in that file. It looks like we started with a single header list, and just added and removed things from that list on a per header, per OS basis. But then OBSD and LINUX came along and just added their own list of headers. And then we started using both "h/" and <sys/>, and then my mind started hurting.
[14:25:25] <Simon Wilkinson> kaduk: Do you need #included <sys/limits.h> ?
[14:28:28] <kaduk@mit.edu/barnowl> Let me check what I actually compiled with ...
[14:32:35] <kaduk@mit.edu/barnowl> I compiled with this:
diff --git a/src/afs/sysincludes.h b/src/afs/sysincludes.h
@ -340,6 +340,8 @ MALLOC_DECLARE(M_AFS);
 # include "h/proc.h"
 # if !defined(AFS_FBSD_ENV)
 #  include "h/ioctl.h"
+# else
+#  include "limits.h"
 # endif /* AFS_FBSD_ENV */

as a crude hack
[14:33:28] <Simon Wilkinson> I think you probably want h/limits.h
[14:34:41] <kaduk@mit.edu/barnowl> I can look at it more carefully later tonight.
[14:39:06] <Simon Wilkinson> I think we should probably try:
[14:39:08] <Simon Wilkinson> diff --git a/src/afs/sysincludes.h b/src/afs/sysincludes.h
index b914956..54b434c 100644
--- a/src/afs/sysincludes.h
+++ b/src/afs/sysincludes.h
@@ -230,6 +230,10 @@ typedef unsigned short etap_event_t;
 # include "h/socketvar.h"
 # include "h/protosw.h"
 
+#if defined(AFS_FBSD_ENV)
+# include "limits.h"
+#endif
+
 # if defined(AFS_SGI_ENV) || defined(AFS_HPUX_ENV) || defined(AFS_SUN5_ENV) || 
 #  include "h/dirent.h"
 #  ifdef       AFS_SUN5_ENV

[14:42:52] <kaduk@mit.edu/barnowl> Okay; I'll confirm that that builds, after rehearsal.
[15:16:47] --- summatusmentis has become available
[15:52:07] <Simon Wilkinson> For those that care, there's nothing in the gerrit 2.1.2.1 update that we care about, so we won't be installing it...
[15:55:16] --- deason has left
[16:05:53] --- kaj has left
[16:56:22] --- jaltman has left: Disconnected
[17:08:03] --- jaltman has become available
[17:38:49] --- Russ has left: Disconnected
[18:13:47] --- Russ has become available
[18:30:12] --- Russ has left: Replaced by new connection
[18:30:12] --- Russ has become available
[19:27:15] --- deason has become available
[19:45:40] --- Born Fool has become available
[20:47:30] --- Russ has left: Disconnected
[20:50:24] --- Russ has become available
[21:06:49] <kaduk@mit.edu/barnowl> Hm, I did a git reset --hard; ./regen.sh in my checkout, and
I am still getting the 'INT_MAX undeclared' build failure.
I guess that means that the build was broken before the xdr patch?
[21:08:36] <shadow@gmail.com/owl06BCDF48> quite possibly.
[21:08:51] <shadow@gmail.com/owl06BCDF48> a previous patch broke it the same way on irix, and i assume here too
[21:10:05] <kaduk@mit.edu/barnowl> Now trying Simon's proposed patch (with whitespace change).
Oh, and the build just finished.
Now to see if aklog still segfaults ...
[21:10:16] <shadow@gmail.com/owl06BCDF48> i assume it will
[21:16:04] <kaduk@mit.edu/barnowl> Indeed, aklog does still dump core.
[21:16:40] <kaduk@mit.edu/barnowl> Did you want anything else from that right now, or should I move on
to debugging shutdown hangs?
[21:16:54] <shadow@gmail.com/owl06BCDF48> did aklog work before that patch?
[21:17:04] <shadow@gmail.com/owl06BCDF48> or is this without it?
[21:17:09] <shadow@gmail.com/owl06BCDF48> (the xdr patch, i mean)
[21:17:40] <kaduk@mit.edu/barnowl> This is without the xdr patch
[21:17:47] <kaduk@mit.edu/barnowl> Should be a stock master from last night.
[21:18:13] <shadow@gmail.com/owl06BCDF48> ok. then you can move on. i wonder what the aklog issue is, but we
won't get it tonight
[21:18:28] <kaduk@mit.edu/barnowl> Okay.
[21:19:19] <kaduk@mit.edu/barnowl> Now, do I want to revert to my patched 1.5.72 before debugging in
earnest...
[21:19:32] <shadow@gmail.com/owl06BCDF48> up to you
[21:20:10] <shadow@gmail.com/owl06BCDF48> i have to drive 80 miles round trip in the morning to bring someone
back to town, so at some point i will fall over. but i have a couple
more things to do first
[21:20:28] <kaduk@mit.edu/barnowl> Fun.
[21:25:46] <kaduk@mit.edu/barnowl> Looks like we have threads in:
osi_NetReceive->soreceive_dgram->_sleep
WaitV->afs_osi_TimedSleep->_sleep
afsdb_req->afs_osi_Sleep->_sleep
afs_brsDaemons->afs_osi_Sleep->_sleep
afs_CacheTruncateDaemon->afs_osi_Sleep->_sleep

I'm not seeing anything else that looks terribly interesting.
[21:27:44] <shadow@gmail.com/owl06BCDF48> > WaitV->afs_osi_TimedSleep->_sleep

it's conceivable i broke this somehow
[21:28:04] <kaduk@mit.edu/barnowl> Yeah ... or maybe I got the unit conversion wrong.
[21:28:26] <shadow@gmail.com/owl06BCDF48> msec/usec?
[21:29:01] <kaduk@mit.edu/barnowl> I'll check once I reboot
[21:33:31] --- Born Fool has left
[21:35:44] --- deason has left
[21:51:15] <kaduk@mit.edu/barnowl> Things seem to only be sleeping for 500 ms, so I don't think I messed up.
[21:58:37] <shadow@gmail.com/owl06BCDF48> ok. then it implies one of the things which is supposed to change
termState and wake up is either not changing, or not waking up
[21:58:58] <shadow@gmail.com/owl06BCDF48> (afs_termState)
[22:04:04] <kaduk@mit.edu/barnowl> I guess I could get a coredump and look at the value of afs_termState ...
[22:04:59] <shadow@gmail.com/owl06BCDF48> with full thread backtraces i could probably tell you without a full
dump
[22:06:02] <kaduk@mit.edu/barnowl> "how full is 'full'?"
[22:06:33] <shadow@gmail.com/owl06BCDF48> um. for the threads you listed, at least the fraes back up to syscall
[22:06:36] <shadow@gmail.com/owl06BCDF48> frames
[22:07:11] <kaduk@mit.edu/barnowl> Not sure if that's available in the live debugger; we'll see.
[22:18:41] <kaduk@mit.edu/barnowl> I am trying to multitask between fscks, and looking at my inability
to write/do fs operations troubles as well.
I can create files in a system:anyuser write directory, which show
up as owned by user 32766.  Is there something special about that
pts ID (athena cell)?  I should be either 24729 or 33554683 ...

Also, is fs setcrypt/fs getcrypt trustworthy for whether encryption
is actually being used, or do I need to tcpdump to be sure?
[22:20:00] <shadow@gmail.com/owl06BCDF48> 32766 is anonymous
[22:20:13] <kaduk@mit.edu/barnowl> Rather as I thought ...
[22:20:21] <shadow@gmail.com/owl06BCDF48> and fs setcrypt applies to connections made after it's set.
[22:20:39] <shadow@gmail.com/owl06BCDF48> and only works if you have keying material. 32766 probably does not
[22:21:16] <shadow@gmail.com/owl06BCDF48> making another wild guess, something in afs_user.c is not matching your
userid with the one with the tokens. 

[22:21:37] <shadow@gmail.com/owl06BCDF48> nah, perhaps not that
[22:22:38] <kaduk@mit.edu/barnowl> I'm perfectly happy to print out the key material, but I haven't
yet found where I would need to hook to do so
[22:25:39] <kaduk@mit.edu/barnowl> ... WTF.
I can read /afs/dev.mit.edu/system/README, but trying to read
/afs/athena.mit.edu/contrib/bitbucket/unos.txt hangs
Hm, I wonder if this is related to setcrypt?
[22:28:56] <kaduk@mit.edu/barnowl> Huh.  And now that I have this 'less' process hung in afsslp,
when I go to shutdown, all of the afsd processes have terminated.
[22:29:35] <shadow@gmail.com/owl06BCDF48> yeah, i think there's something wacky with wakeups
[22:41:55] <kaduk@mit.edu/barnowl> Yeah, if I setcrypt, I can read anyuser-readable stuff, but not my own.
(No hang on reading the anyuser-readable stuff being the notable part)
[22:42:28] <kaduk@mit.edu/barnowl> So maybe it is afs_user.c
[22:50:20] <kaduk@mit.edu/barnowl> This time on shutdown, all afsd's were done, and nothing in afsslp,
but shutdown (reboot) still hung.  The only thing that looked odd
was that init was in state mntref.
[23:05:40] <kaduk@mit.edu/barnowl> That mntwait init is:
[...]
_sleep
vfs_mount_destroy
dounmount
vfs_unmountall
boot
reboot
syscall
[23:13:13] <kaduk@mit.edu/barnowl> vfs_mount_destroy msleep()s on the mount point with the mount point's
mutex; it does seem consistent with a wakeup getting lost, and racing
for where that wakeup gets lost.
[23:17:27] --- reuteras has become available