[11:20] <kraut> moin
[03:08] <zul> @schedule montreal
[03:08] <ubotu> Schedule for America/Montreal: 18 Sep 12:00: Kernel Team | 19 Sep 08:00: Edubuntu | 19 Sep 16:00: Xubuntu Developers | 20 Sep 08:00: Desktop Team Development | 21 Sep 08:00: MOTU Team | 24 Sep 15:00: Screencast Team
[04:58] <mattwj2005> I have a meeting idea
[04:58] <mattwj2005> my apartment today...job....to clean it...I'll throw in free Ubuntu CDs
[04:58] <mattwj2005> :P
[04:58] <pochu> wi mattwj2005
[04:59] <mattwj2005> :) just kidding
[04:59] <maks_> use robots
[05:00] <mattwj2005> hehe
[05:00] <mattwj2005> I don't I just use beaming technology ;)
[05:00] <mattwj2005> *why don't I
[05:59] <ian_brasil>  @schedule manaus
[05:59] <zul> heylo
[05:59] <ian_brasil> @schedule manaus
[05:59] <ubotu> Schedule for America/Manaus: 18 Sep 12:00: Kernel Team | 19 Sep 08:00: Edubuntu | 19 Sep 16:00: Xubuntu Developers | 20 Sep 08:00: Desktop Team Development | 21 Sep 08:00: MOTU Team | 24 Sep 15:00: Screencast Team
[06:02] <BenC> missing pkl, but he's known to be busy
[06:02] <BenC> Ok, this is the start of the Kernel Team meeting
[06:02] <BenC> agenda hasn't been updated, but it's quite simple
[06:03] <BenC> beta freeze looming, and we are handling as many bugs as possible to make beta release
[06:03] <BenC> The most major issue right now is the unionfs breakage that pkl is working
[06:03] <BenC> pkl_: Can you give a quick one or two lines about your status so you can get back to things?
[06:04] <pkl_> I have identified two serious bugs in Unionfs that the liveCD is hitting
[06:05] <BenC> two...are they both causing the same symptom we've been seeing about refcount backtrace?
[06:05] <ogra> pkl_, i see a lot issues in ltsp as well recently ... if you need any logs or so, just ask ..
[06:06] <pkl_> The first relates to the BUG_ON in fanout.h, called by unionfs_flush (file close).  This is a race condition, not with the actual dentry being closed, but with the parent dentry going bad, with no branches (which is wrong).
[06:07] <pkl_> I believe the problem is related to rsyncing of the parent unionfs directory with the lower filesystems (i.e. ro squashfs, and rw tmpfs).
[06:07] <pkl_> The good thing is whilst I've not discovered the underlying problem, it is easily worked around.
[06:08] <BenC> is that a kernel work around, or a userspace one?
[06:09] <pkl_> kernel unionfs workaround - the only reason why the parent branches are being examined is to resync the mtime.  A simple fix is to check there _are_ branches on the parent dentry, and to skip the mtime resync if there aren't.  This "fixes" the BUG_ON in my tests.
[06:10] <pkl_> It is a workaround that will work while the underlying race is identified.
[06:11] <pkl_> The second bug is the kernel oops in inode.c (unionfs_open).
[06:11] <BenC> ok, that'll make everyone happy happy
[06:11] <BenC> pkl_: did that bug show up after your workaround, or was that known before?
[06:12] <pkl_> The second bug?  That showed up before the workaround.
[06:12] <BenC> ok
[06:12] <pkl_> This is caused because occasionally the vfsmnt point for a lower dentry is NULL.
[06:13] <pkl_> A little bit of background.
[06:13] <pkl_> When a unionfs dentry is opened, unionfs in turn opens all the corresponding dentries on the lower filesystems (the branches).
[06:14] <pkl_> Unionfs stores the correspondng underlying dentries and their vfsmount points.
[06:15] <BenC> pkl_: this all revolves around the vfs changes for apparmor?
[06:15] <pkl_> It is relatef.  apparmor changes the vfs interface so much more vfs functions expect the vfsmount.
[06:16] <pkl_> In this case, however, the vfs dentry_open() function has always expected a vfsmount, and so it is unrelated to the vfs changes.
[06:16] <pkl_> ^unrelated^not caused
[06:17] <BenC> Ok, guess I can't bad mouth kees now :)
[06:17] <pkl_> Again, the underlying reason appears to be a race condition.
[06:17] <pkl_> Still working on the cause.
[06:18] <BenC> pkl_: have you talked to unionfs upstream about this at all?
[06:18] <pkl_> As a temporary workaround, I have put a hack in the unionfs_open code to remember the correct vfsmount point for each branch. In the case of the liveCD there's two branches, and therefore two vfsmount points.
[06:19] <pkl_> if the vfsmount point lookup for a dentry fails, the hack replaces it with the correct vfsmount point previously remembered.
[06:19] <pkl_> This fixes the oops, and it is definately the cause.
[06:20] <BenC> pkl_: Ok, two workarounds sounds like decent progress, so excellent work. When can we get these in lum for an upload?
[06:20] <pkl_> Now I have establised the causes I can email the unionfs guys.
[06:21] <pkl_> I was planning on putting the hacks into lum in a couple of hours (I have to strip them out of my heavily hacked unionfs code).
[06:22] <BenC> pkl_: ok, ping me when you're ready and I'll get an upload out
[06:22] <pkl_> OK
[06:22] <BenC> thanks
[06:24] <BenC> So I sent an email to the kernel-team list explaining what our current focus should be
[06:25] <BenC> At this point in release cycle is where we are usually the busiest
[06:26] <BenC> I don't forsee any problems like we've had in the past, simply because the upstream kernel stabalized very well and we stuck to stock code so strictly
[06:27] <BenC> but we do have some minor usage bugs and some regressions that we should handle in the next few weeks
[06:27] <BenC> and we need to keep a close eye on the incoming bugs (and bugs marked NEW) to make sure we aren't over looking anything major
[06:29] <BenC> amitk, rtg_, kylem: Anything major you guys are working on that needs to get in with beta?
[06:29] <kylem> amt.
[06:29] <rtg_> misc sky2 fixes, perhaps some audio quirks. Dunno for sure yet.
[06:31] <zul> bug #77725 should probably get in
[06:31] <ubotu> Launchpad bug 77725 in linux-source-2.6.17 "No Driver for Attansic Gigabit Ethernet" [Wishlist,In progress]  https://launchpad.net/bugs/77725
[06:31] <kylem> ..
[06:31] <amitk> just the usb-autosuspend-prevention patch from mjg59
[06:31] <kylem> atl1 is in gutsy.
[06:32] <kylem> oh.
[06:32] <kylem> they should file a second bug.
[06:32] <kylem> muppets.
[06:32] <amitk> hmm... I just noticed you cherry-picked it...
[06:33] <BenC> amitk: merged that already
[06:33] <BenC> kylem: yeah, that's what I started to think till I got down to the atl2 part
[06:34] <BenC> ok, atl2 should be put into lum
[06:34] <BenC> anyone up for doing that?
[06:34] <zul> i can since i brought it up
[06:34] <BenC> zul: you sure can...good man :)
[06:35] <BenC> zul: shoot a pull request to #ubuntu-kernel when you can, thanks
[06:35] <zul> no probs
[06:35] <BenC> the other major issue is the header post{inst,rm} scripts that I am working on
[06:35] <BenC> needed for DKMS and feisty compatibility
[06:37] <rtg_> BenC: why is it more complicated then just fixing the perl expressions?
[06:37] <BenC> rtg_: because the scripts from kernel-package need to be munged a bit
[06:37] <BenC> debconf stuff removed/disabled, variable replacements taken into account
[06:38] <BenC> rtg_: fixing the perl replacements in kernel-package doesn't fix our kernel builds, and the scripts are not 1-to-1 compatible with our build system
[06:39] <rtg_> Well, I haven't delved into it in that level of detail.
[06:40] <BenC> Ok, any other business?
[06:41] <dholbach> is somebody on bug 98641?
[06:41] <ubotu> Launchpad bug 98641 in linux-restricted-modules-2.6.22 "[nvidia-glx-new]  Driver is missing libwfb breaking X on 8000 series cards" [High,Confirmed]  https://launchpad.net/bugs/98641
[06:41] <dholbach> it's on the sponsoring bug queue
[06:42] <BenC> dholbach: already fixed
[06:42] <dholbach> rock on - thanks a lot - can it be closed then?
[06:42] <BenC> yeah, definitely
[06:42] <dholbach> ok great
[06:42] <dholbach> gracias
[06:43] <dholbach> thanks for dealing with the sponsoring bugs
[06:43] <BenC> and with that, let's cut this meeting out so we can all get back to work :)
[06:43] <BenC> dholbach: np, thanks for keeping up on them
[06:44] <BenC> thanks everyone
[09:51] <siretart> do meetings using the MootBot need to be registered somewhere?