freebsd stable - Aug 2016 - unionfs bugs, a partial patch and some comments [Was: Re: 1-BETA3 Panic: __lockmgr_args: downgrade a recursed lockmgr nfs @ /usr/local/share/deploy-tools/RELENG_11/src/sys/fs/unionfs/union_vnops.c:1905]

On Mon, Aug 08, 2016 at 10:02:00AM +0200, Harry Schmalzbauer
wrote:
> Bez?glich Rick Macklem's Nachricht vom 07.08.2016 23:34 (localtime):
> > Harry Schmalzbauer wrote:
> >> Hello,
> >>
> >> I had another crash which I'm quite sure was triggered by
mount_unionfs:
> > Just in case you are not already aware, unionfs is always broken. Read
> > the BUGS
> > section at the end of "man mount_unionfs". If it were easy
to fix,
> > someone would
> > have done so long ago. Yes, some use it successfully, but if not...
> > 
> > Sorry, but I suspect that is how it will remain, rick
> 
> Thanks for the hint, not happy to hear that, but I was not aware of that
> explicit warning in man 8 mount_unionfs :-(
> 
> This feature is utterly important for me (all my productive machines
> have "/" read-only mounted and "/etc" is an union to a
writable, synch
> mounted separate fs), so back in 2012, after a lot of locking redesign
> has been done in 9-current, I got Attilio Raos attention and he gave out
> some test patches for 9.0.
> He was aware of missing locking adjustments, but patches addressing the
> majority of them didn't work.
> Since then I'm draging a minimal patch which prevents at least the
> kernel panics for me.
> Unfortunately I don't have the skills to continue Attilio Raos work.
> 
> Just for anybody else needing unionfs:
> https://people.freebsd.org/~attilio/unionfs_missing_insmntque_lock.patch
> 
> This patch still applies and I'm successfully using this (unmodified)
up
> to FreeBSD-10.3 and never had any panic in all these years.
Having spent some time looking at unionfs, I'm a bit skeptical that this
patch will address the panic you reported earlier, though I'd be
interested to know if it does. Reading the code, I think it will just
address an INVARIANTS-only assertion in insmntque1().

Unfortunately, unionfs is quite difficult to fix within the current
constraints of FreeBSD's VFS. unionfs_readdir() is a particularly good
demonstration of this fact: some callers of VOP_READDIR expect the
cookies returned by the FS to be monotonically increasing, but unionfs
has no straightforward way to make this guarantee.
> 
> I will continue using it for FreeBSD-11 and I guess it will also prevent
> my last reported panics.
> But I wanted to take part in the BETA test without local modifications
> at first.
> 
> Another very importend usage scenario of unionfs for me is for my build
> host(s). I'm (nfs4-)sharing a svn-checked out read-only portstree. My
> inofficial "ports/inofficial" directory perfectly shows up by
> unionfs-mounting it below the unaltered portstree :-)
> 
> For me, unionfs is as important as ZFS (and nullfs) is in FreeBSD.
> 
> First thing to do for me, after I won in lottery, was to find someone
> who can be sponsored fixing unionfs ;-) And bringing MNAMELEN into 21st
> century state, matching ZFS needs:
>
https://lists.freebsd.org/pipermail/freebsd-hackers/2015-November/048640.html
> This is another patch I'm carrying for a very long time which solves
> tremendous limitations for me. Without that, I couldn't use ZFS
> snapshots in real world, along with a human-friendly dataset naming :-)
> 
> -Harry
> 
> _______________________________________________
> freebsd-stable at freebsd.org mailing list
> https://lists.freebsd.org/mailman/listinfo/freebsd-stable
> To unsubscribe, send any mail to "freebsd-stable-unsubscribe at
freebsd.org"

Bez?glich Mark Johnston's Nachricht vom 09.08.2016 08:02 (localtime):
?
>>
>> Just for anybody else needing unionfs:
>>
https://people.freebsd.org/~attilio/unionfs_missing_insmntque_lock.patch
>>
>> This patch still applies and I'm successfully using this
(unmodified) up
>> to FreeBSD-10.3 and never had any panic in all these years.
> 
> Having spent some time looking at unionfs, I'm a bit skeptical that
this
> patch will address the panic you reported earlier, though I'd be
> interested to know if it does. 
Thanks for your attention.
I can confirm that it has prevented panics for more than 4 years
(9.0-10.3) and it seems to be still "good enough" to also prevent
panics
in 11-BETA4.
I updated my build host (stable/11, this time with the
unionfs_missing_insmntque_lock.patch), where the recent panics happened
and unionfs gets much more utilized than usually in my setups: No panic
with that patch anymore.
Just one message like "prevented resource deadlock" occured.
> Reading the code, I think it will just
> address an INVARIANTS-only assertion in insmntque1().
> 
> Unfortunately, unionfs is quite difficult to fix within the current
> constraints of FreeBSD's VFS. unionfs_readdir() is a particularly good
> demonstration of this fact: some callers of VOP_READDIR expect the
> cookies returned by the FS to be monotonically increasing, but unionfs
> has no straightforward way to make this guarantee.
I'm sorry, I can't provide help here. My skills would require a huge
ammount of lerning-time to get into that matter. I'd love to do that,
but I can't afford :-(

Thanks,

-Harry