amd64/167067: ZFS panics the server

[Aldo]

>Number:         167067
>Category:       amd64
>Synopsis:       ZFS panics the server
>Confidential:   no
>Severity:       critical
>Priority:       high
>Responsible:    freebsd-amd64
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Wed Apr 18 16:40:02 UTC 2012
>Closed-Date:
>Last-Modified:
>Originator:     Aldo
>Release:        FreeBSD 9.0-RELEASE #0. AMD64
>Organization:
>Environment:
FreeBSD -snip- 9.0-RELEASE FreeBSD 9.0-RELEASE #0. AMD64
>Description:
Installed FreeBSD. Added the patch to not panic ZFS on "ls -la" (sorry, don't remember number/link, damn, it was related to something assuming FreeBSD in some ondisk structure if I remember) on Solaris created volumes, and recompiled zfs.ko.

It mounts my pool and dataset ok.
Installed Samba 3.6.4. Also tried 3.6.1, compiled with variuos options with clang or gcc 4.2. Doesn't change.

If I run stress tests with scripts it runs fine. When my users try to access the Samba share, the server panics (see attach).
It seems something related to concurrency. I moved 200Gb with no problems, when two people open a word document the server is gone.
The panic reports it was running smbd in the moment.

Pool v28 upgraded from Opensolaris v22.
The picture of the dump is here http://forums.freebsd.org/showthread.php?t=31447
>How-To-Repeat:
Don't really know, it surely needs several people accessing the share.
Files were heavily manipulated on their permissions (setfacl).

>Fix:
After a full working day trying any kind of sysctl (memory, zil, limits) I decommissioned the server, since it was in production and needed to function.

>Release-Note:
>Audit-Trail:
>Unformatted: