Deadlock in zpool import with degraded pool

[Holger Freyther]

Hi,

Everytime I run zpool import tank, I can see disk activity with gstat,
with top I see the ARC growing and the free memory going down and then
the system seems to lock-up. Adding a swap partition doesn't help. I
have tried the zpool import with 10.1, 10.3 and 11.0-CURRENT.

My pool consists out of two disks and on the last start the second GELI
partition was not attached. When manually doing geli attach + zpool
online it started to resilver. The system still had a broken disk inside
so I took this opportunity to halt and remove it. On reboot the disk
changed from ada2 to ada1. With that the pool was degraded and the only
thing I found was zpool  replace to replace the physical disk with itself.
As it was resilvering before this didn't look like a big loss in terms
of time. GELI is not the fastest and I didn't want to wait too long so I
started zfs destroy of some volumes I will not use anymore. This should
have freed 300GB from 2TB of a ~3TB pool. Before the zfs destroy
completing the system locked-up.

On reboot the system unlocked both GELI partitions but got stuck on
mounting the rootfs. Booting the memstick version of FreeBSD 10.3 and
11.0-CURRENT I do see the deadlock on import (top not updating time, not
responding to switches of the VTY, USB keyboard plug/unplug still
generating messages though).

The system is a AMD64 with 12GB RAM and running FreeBSD 10.1. It has
two 3TB disks, a boot pool and the main pool on top of GELI. The main
pool runs with deduplication enabled

Does this sound familiar? Is there anything I can try before saying
good bye to mybacked up data? If the pool itself can not be repaired,
is there a way to access the volumes with the data?

thanks for your help.