[9fans] Weird problem.

[9fans] Weird problem.

[Dan Cross]

I'm looking at a Plan 9 machine where ps(1) keeps hanging; in particular,
it hangs while trying to open one of the status files under /proc/*. (eg,
open("/proc/$somepid/status", OREAD); hangs).  Looking at the source code
to devproc.c, I see that one of the first things procopen() does is lock
p->debug once it finds the proctable entry for the process that particular
file is associated with.  My guess, based only on vague feeling and nothing
more substantial, is that there exists some sort of deadlock with p->debug.

Is there a way to see what locks a process holds, short of instrumenting
the kernel to do so?  I'd hate to simply reboot the machine, since I think
that the error state would be difficult to recreate, and I'd really like
to understand what's going on.  Alternatively, if anyone's seen this before
and knows what's up, that'd be great.

	- Dan C.

Re: [9fans] Weird problem.

[Russ Cox]

If you run acid -k, you can get a kernel stack trace
for the process doing the open.  That will tell you
what exactly is blocking.  p->debug is a QLock, and
unfortunately they don't keep track of who is currently
holding it.  The kernel trace might suggest where to
go next though.

grep /proc/ /proc/*/fd might be useful too.