Experience testing recent -mm ZFS patches: WARNING: releng-8.2-zfsv28-20110301 will eat your cookie

[Jeffrey Baitis]

Caveat: I have been using FreeBSD for five days now (shun the noob).

That stated, I built world using
http://people.freebsd.org/~mm/patches/zfs/v28/releng-8.2-zfsv28-20110301-nopython.patch.xz
and http://people.freebsd.org/~mm/patches/zfs/v28/stable-8-zfsv28-20110301-nopython.patch.xz.

I am running a single-disk ZFS root filesystem (version 15). My
make.conf contains CPUTYPE?=athlon64, machine has 4GB RAM.

After extremely extensive testing, the RELENG_8 kernel is working well
thus far (uptime 4:36), cvsup performed last night.

RELENG_8_2 caused processes on the system to grind to an unrecoverable halt.

As soon as I noticed my skype application taking 100% CPU time, my
next attempt was to launch firefox.

truss -f firefox revealed that the process hung after a stat call to a
specific file:
67894: stat("/usr/local/etc/fonts/fonts.conf",{ mode=-rwxr-xr-x
,inode=240214,size=5223,blksize=5632 }) = 0 (0x0)

As an experiment, I performed the following:
ls /usr/local/etc/fonts/fonts.conf

This command hung immediately and, like my other processes, was
un-killable. Any attempts to stat the inode in question resulted in
further cookie-eating.

In order to rectify the issue, the machine required reboot; of course,
rebooting was impossible without pulling the plug.

Please let me know if further information / test cases executed on the
RELENG_8_2 patched kernel.

Regards,
Jeff