Home
release-team@conference.openafs.org
Wednesday, May 6, 2015< ^ >
Room Configuration
Room Occupants

GMT+0
[13:24:46] meffie joins the room
[13:35:23] wiesand joins the room
[13:58:23] meffie leaves the room
[13:58:32] meffie joins the room
[14:00:23] <wiesand> Hello
[14:01:14] <wiesand> Looks like this will be a brief meeting, and that’s fine by me ;-)
[14:01:16] Marc Dionne joins the room
[14:01:34] <wiesand> Hi Marc.
[14:01:39] <Marc Dionne> Hi Stephan
[14:02:08] <Marc Dionne> thought I'd drop in to give you an update on the kernel GPL thing
[14:02:19] <wiesand> Great. please do.
[14:03:00] <Marc Dionne> I posted a patch to change the symbol to non GPL, it has been picked up by the x86 maintainers, and included in a pull request this morning
[14:03:14] <Marc Dionne> so all that's left is for Linus to pull it in
[14:03:14] <wiesand> Sounds promising
[14:03:16] kaduk joins the room
[14:03:30] <wiesand> Very good.
[14:03:39] <wiesand> Can you comment on that spinlock issue too?
[14:03:45] <Marc Dionne> so should be fixed in mainline in a few hours
[14:04:38] <Marc Dionne> for the mmap fix?  we can't call osi_Free inside the spinlock, as that may sleep.  there's a simple fix that we already have in our tree, i'll update the change
[14:05:02] <wiesand> Good.
[14:05:26] <wiesand> Finally, Jeffrey gave a heads up regarding EL6.7...
[14:06:33] <wiesand> I understand this becomes relatively easy to sort out if we give up the afs-nfs-translator?
[14:07:12] <Marc Dionne> well that hasn't compiled for linux for quite a while, other GPL symbols.  since 2.6.25 or thereabouts
[14:08:22] <wiesand> So why not ditch it?
[14:08:35] <meffie> it is still used on solaris, iirc
[14:08:42] <Marc Dionne> it works on other platforms, yeah solaris i think
[14:09:36] <Jeffrey Altman> and there are still some sites that build OpenAFS against older kernels to use the nfs export from Linux
[14:09:58] <wiesand> Can we disable it only on kernels where it doesn’t build anyway?
[14:10:28] <wiesand> (EL5 still has a few years to live, and kernel 2.6.18 - it may work there)
[14:10:50] <Jeffrey Altman> I think EL5 is where it is still used
[14:11:00] <Marc Dionne> i thnk we could move away from using d_splice_alias at all if we know that the nfs code can't be used.  (probably existing defines for that)
[14:11:49] <Marc Dionne> but we could leave it for older kernels where the nfs translator might still be used
[14:12:22] <wiesand> Seems more appropriate for a “stable series”.
[14:16:24] <wiesand> So if it’s doable, that’s what I’d prefer.
[14:16:37] <Jeffrey Altman> understood
[14:16:45] <Marc Dionne> i'll try to come up with a patch
[14:17:03] <wiesand> EL6.7 isn’t critically urgent I think.
[14:17:10] <wiesand> Thanks Marc.
[14:17:42] <wiesand> A pre2 with the other Linux changes but not EL6.7 makes sense IMO.
[14:17:59] <Jeffrey Altman> 6.7 has shipped and 6.8 nearing completion
[14:18:25] <kaduk> My only concern about trying to support both would be if there were issues trying to support two different codebases and only one got regularly used by developers, but I will defer to the people actually doing the work for linux support.
[14:18:34] <wiesand> 6.7 *beta* was announced yesterday
[14:18:59] <wiesand> 6.7 will ship in two or three months, from prior experience
[14:20:34] <wiesand> As a partner, you’re quite a bit ahead of the rest of the EL ecosystem...
[14:21:06] <wiesand> Ben: +1.
[14:21:09] <Jeffrey Altman> Red Hat considers 6.7 closed to changes (except for showstoppers)
[14:22:03] <Jeffrey Altman> I think we can move on
[14:22:44] <wiesand> Ok, let’s move on. Marc, thanks for the input and all the work on Linux.
[14:23:49] <wiesand> Most other items were sufficiently discussed via email...
[14:24:28] <wiesand> 11850 is probably close to final, and once that’s available I’d merge all the Yosemite stack.
[14:24:31] <Jeffrey Altman> Ben updated 11850 and I will merge when buildbot finishes
[14:24:40] <wiesand> Thanks.
[14:25:12] <wiesand> A few +1s on 11859 would be nice.
[14:25:32] <Jeffrey Altman> softsig is outside my area of expertise
[14:26:11] <wiesand> [indeed, I think we can move on to “1.8”]
[14:26:27] <Jeffrey Altman> 11859 (yosemite settings) you should merge
[14:27:49] <wiesand> Done.
[14:28:02] <kaduk> w.r.t. softsig, I think we should allow catching FPE
[14:28:23] <kaduk> since we do have things (ab)using it
[14:28:44] <kaduk> It is less clear whether we should also allow PIPE
[14:30:04] <kaduk> I guess there are consumers in vsprocs?
[14:30:37] <meffie> why would we not catch SIGPIPE?
[14:30:56] <kaduk> I don't have a particular reason to not catch it
[14:31:09] <kaduk> consumers in vsprocs, whose handlers manually invoke longjump, fun.
[14:33:17] <kaduk> So, I guess we should make a new rev that allows catching PIPE and FPE, and adds some assertions, and that's probably good?
[14:35:02] <Jeffrey Altman> If Andrew and Chas +1 I will merge that
[14:35:28] <meffie> ok, yes, that makes sense.
[14:35:35] <kaduk> Mike, do you want to make the change?  Should I?
[14:35:35] <Marc Dionne> i recall an issue with the softsig test under rhel5
[14:37:18] <meffie> i can enable SIGPIPE and SIGFPE, i'll look at patchset 4 again.
[14:37:56] <meffie> and ping Andrew for a review
[14:37:56] <Marc Dionne> i commented in the change.  i think the case that's supposed to crash in the softsig-helper test doesn't on rhel5
[14:38:49] <kaduk> Yeah, that crashing code seemed a little suspicious to me when reading it, but it seemed likely to crash in most places.  And it would get noticed if someone ran the tests and they didn't work...
[14:40:03] <kaduk> (It seems to rely on the string literal being placed in the rodata section or similar)
[14:42:13] <kaduk> Perhaps an optimizer would be able to remove the assignment, too, though I'd have to look at it again to double-check.
[14:44:21] <wiesand> this segfaults nicely on EL5: int main(){
  char *bang = "boom";
  *bang = 'X';
  return 0;
}
[14:45:29] <Marc Dionne> this was a while ago, it's possible that gcc updates have changed things
[14:45:38] <wiesand> But indeed not compiled with -O1 or higher.
[14:46:27] <kaduk> I guess we could see if there's a more reliable crashing idiom
[14:47:19] <meffie> its not often you get to write code to crash on purpose.
[14:47:57] <kaduk> indeed
[14:47:58] <Jeffrey Altman> can pragmas be used to disable optimization for the function?
[14:48:36] <meffie> maybe we can just call abort()?
[14:48:53] <kaduk> No, abort() is not good enough
[14:49:02] <Jeffrey Altman> the purpose of the test is to generate specific signals
[14:50:02] <meffie> oops. right, abort() raises a signal.
[14:50:36] <kaduk> The internet is suggesting:
[14:50:39] <kaduk> int *ptr_segv = mmap(NULL, PAGE_SIZE, PROT_NONE, MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS, -1, 0);
[14:52:46] <kaduk> buildbot reported on 11850, but not as verified since the solaris10 builder encountered an exception
[14:53:53] <kaduk> Anyway, it seems that dereferencing NULL is slightly more likely to segfault than writing to a string literal
[14:54:27] <kaduk> I don't think we need to worry too much about this case in the absence of a machine where the current form of things is verified to not work.
[14:54:55] <meffie> ok
[14:56:46] <kaduk> If we can get softsig merged, then we should probably make a pass and convert consumers to use it, then get back to the logging changes
[14:57:07] <meffie> yes
[14:57:32] <kaduk> Any other 1.8 topics for today?
[14:58:14] <meffie> i have nothing.
[14:58:42] <wiesand> The mmap thing segfaults on EL5 with -O*
[14:58:47] <Jeffrey Altman> mike, could you look at the solaris 10 x86 builder?    its failing all build requests at the moment
[14:58:52] <kaduk> In the absence of anything else, it seems like we're done for today
[14:58:53] <meffie> oh. ok.
[14:59:34] <wiesand> Fine, let’s call it a day.
[14:59:42] <wiesand> (here, that is)
[14:59:49] <wiesand> Thanks a lot everyone!
[14:59:54] Marc Dionne leaves the room
[14:59:58] <Jeffrey Altman> take care
[15:00:19] wiesand leaves the room
[15:00:45] <meffie> thanks.
[15:03:56] <meffie> looks like we had another drive failure. the sol10 box  should be back shortly.
[15:06:04] <kaduk> thanks!
[15:06:28] kaduk leaves the room
[15:11:53] <meffie> sol10 bot is back up.
[15:12:26] meffie leaves the room
[15:20:07] shadow@gmail.com/barnowlE5B64A04 leaves the room
[15:20:43] shadow@gmail.com/barnowlE5B64A04 joins the room