Bad news re: new (20080817) ZFS patches and send/recv (broken again)

[Thomas Backman]

On Aug 18, 2009, at 18:11, Stefan Bethke wrote:

> Am 18.08.2009 um 16:20 schrieb Thomas Backman:
>
>> On Aug 17, 2009, at 17:24, Thomas Backman wrote:
>>
>>> On Aug 17, 2009, at 15:25, Thomas Backman wrote:
>>>
>>>> So, I've got myself a source tree almost completely free of  
>>>> patches after today's batch of ZFS patches merged - all that  
>>>> remains is that I uncommented ps -axl from /usr/sbin/crashinfo,  
>>>> since it only coredumps anyway, and added CFLAGS+=-DDEBUG=1 to  
>>>> zfs/Makefile.
>>>>
>>>> One of the changes I didn't already have prior to this must have  
>>>> broken something, though, because this script worked just fine  
>>>> before the merges earlier today.
>>>> The script below is the exact same I linked in http://lists.freebsd.org/pipermail/freebsd-current/2009-July/009174.html 
>>>>  back in July (URL to the script: http://exscape.org/temp/zfs_clone_panic.sh 
>>>>  ) - I made some local changes, thus the name invoked below.
>>>>
>>>> Now that all the patches are merged, you should need nothing but  
>>>> the script, bash, and the ~200MB free space on the partition  
>>>> containing /root/ to reproduce this problem.
>>>> (Note that the "no such pool" in the FIRST script is normal; it  
>>>> simply tries to clean up something that isn't there, without  
>>>> error/sanity checking.)
>>>>
>>>> [...]
>>>> + zpool create -f -R /slave slave ggate666
>>>> ++ date +backup-%Y%m%d-%H%M
>>>> + NOW=backup-20090817-1522
>>>> + echo 'Creating snapshots'
>>>> Creating snapshots
>>>> + zfs snapshot -r tank at backup-20090817-1522
>>>> + echo 'Cloning pool'
>>>> Cloning pool
>>>> + zfs send -R tank at backup-20090817-1522
>>>> + zfs recv -vFd slave
>>>> cannot receive: invalid stream (malformed nvlist)
>>>> warning: cannot send 'tank at backup-20090817-1522': Broken pipe
>>>>
>>>>
>>>> Regards,
>>>> Thomas
>>> This is perhaps more troubling...
>>> [...]
>>> [root at chaos ~]# zpool create testpool ad0s1d
>>> [root at chaos ~]# zpool export testpool
>>> [root at chaos ~]# zpool import testpool
>>> cannot import 'testpool': no such pool available
>>>
>>> Regards,
>>> Thomas
>>
>> OK, I tried to reproduce this in a VM... And I have to say I was a  
>> bit surprised: after doing an installkernel/installworld, but  
>> BEFORE REBOOTING (I install in "multi"-user (one user via ssh),  
>> never had a problem with that), the same issue has appeared, so I'm  
>> guessing zfs.ko can't be to blame here?
>
> It's not the zpool binary by itself: after updating from a two-day  
> old current (make world && reboot):
>
> root at freebsd-current:~# zpool list
> NAME    SIZE   USED  AVAIL    CAP  HEALTH  ALTROOT
> zroot  14.9G   753M  14.1G     4%  ONLINE  -
> root at freebsd-current:~# zpool export zroot
> root at freebsd-current:~# zpool list
> no pools available
> root at freebsd-current:~# zpool import zroot
> cannot import 'zroot': no such pool available
> root at freebsd-current:~# zpool.old import zroot
> cannot import 'zroot': no such pool available
> root at freebsd-current:~# zpool.old import
> root at freebsd-current:~# uname -a
> FreeBSD freebsd-current.lassitu.de 8.0-BETA2 FreeBSD 8.0-BETA2 #1  
> r196359: Tue Aug 18 16:42:41 CEST 2009     root at freebsd-current.lassitu.de 
> :/usr/obj/usr/src/sys/MINIMAL  amd64
>
> I saved zfs and zpool before the installworld. And my root is  
> actually on UFS; this pool was left over from root on zfs raidz  
> experiments.
>
> root at freebsd-current:~# ls -l /sbin/zpool*
> -r-xr-xr-x  1 root  wheel  76752 Aug 18 17:47 /sbin/zpool*
> -r-xr-xr-x  1 root  wheel  76752 Aug 18 17:46 /sbin/zpool.old*
> root at freebsd-current:~# md5 /sbin/zpool*
> MD5 (/sbin/zpool) = 83dcf6343bb0392a38159dd456dcf4c5
> MD5 (/sbin/zpool.old) = 340cb5a383b2fc3c77afbdc881258597

Interestingly, it doesn't appear to be the kernel by itself, either. I  
did a bad thing(tm) and reverted my sources back to r196074 + patches,  
the last rev I had working before yesterday's upgrade, and then did a  
build+installkernel. I did *not* build/install world. Yes, I know that  
this is bound to cause problems, just thought I'd chime in anyway. The  
same problems remained with the old kernel.

Regards,
Thomas