Using read-only NULLFS leads to panic. gdb output included, e asy to reproduce.

[Tim Robbins]

On Mon, Mar 15, 2004 at 11:05:24AM +0100, Pawel Malachowski wrote:

> On Wed, Mar 10, 2004 at 07:19:33PM +0100, Pawel Malachowski wrote:
> 
> > > Here's the patch against RELENG_4:
> > > http://people.freebsd.org/~tjr/nullfs-4.diff
> > > 
> > > It works, but I'm not going to commit it until I've had time to
> > > check a few things first.
> > 
> > Successfully tested on my 4.9-STABLE box, thanks.
> 
> Well, I shoot myself in the foot.
> 
> After 2 days, my 4.9-RELEASE machine caught o problem again.
> Not panic, new processess trying to access one of filesystems
> were stopping at `inode' state (for example >3k crons). System
> was also not able to shutdown properly.

Next time this happens, break into ddb and save backtraces of a few
processes stuck in the inode state with the "tr <pid>" command.
If you don't have a serial console, just jot down the first few lines of
each backtrace. Try to collect at least 2 different traces - most of them
will be identical. If the bug you're running into is the one I think it is,
reducing the value of kern.maxvnodes may make it easier to reproduce (and
conversely, increasing the value will mask the bug somewhat.)


Tim