Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 25 Oct 95 16:43:08 +0000
Cc:, David Greenman <davidg@Root.COM>
Subject:   Re(2): panic: free vnode isn't
Message-ID:  <"MAC-951025164255-1E49*/G=Andrew/S=Gordon/O=NET-TEL Computer Systems Ltd/PRMD=NET-TEL/ADMD=Gold 400/C=GB/"@MHS>
In-Reply-To: <"SunOS:857-951025122108-11BB*/DD.RFC-822=owner-hackers(a) 400/C=GB/"@MHS>

Next in thread | Previous in thread | Raw E-Mail | Index | Archive | Help
>     Nope.  I switched back to the non-debugging kernel today and it
> only ran for about a day before it locked up.  Again, no syslog
> messages or any indication that something went wrong (except that
> everything is frozen).  We left the office at 9:20pm for dinner, and
> it died at 9:30pm (figures).
> ...
> too much of a load on the server?  The FTP server is chroot'd to a
> local directory, but anything beneath ~ftp/pub is NFS-mounted.  All
> user home directories are also NFS-mounted.

This sounds a little like a problem we have been having - which I haven't
reported previously as I have not had time to characterise it properly.
However, it goes somthing like this...

Server is running 2.0.5R (with more recent SCSI drivers from -STABLE a few weeks
ago, probably not important).  It is configured as both NFS server and client,
and also runs SAMBA to serve files to Windows machines.  Some of the Windows
users mount partitions through SAMBA which are in turn NFS mounted by the server
from a third machine.  All this worked fine for some months, although with very
occasional freezes - usually when one of the DOS machines had been crashed and
rebooted, but otherwise inexplicable.  More recently, Windows 95 has appeared on
the scene, and when a Win95 machine accesses files on one of these SAMBA->NFS
mounted partitions, the freeze happens consistently.

The nature of the freeze is some kind of deadlock in the filesystem - if you
catch it just after the freeze, terminal/telnet sessions are normally still
alive, but as soon as they touch the filesystem they also block forever - and
things like mail check in the shell etc. means that most processes end up frozen
after a few minutes.

The problem has completely gone away since we moved all the files used by the
Windows users onto local discs on the server, so the problem would appear to lie
in the NFS client code.  My best guess is that this relates to the fact that
SAMBA appears to do non-blocking I/O on files, in order to serve multiple
request in parallel from a given client (there is one SAMBA process per client)
- presumably Win3.11 never makes multiple requests in parallel apart from the
special case of crashing between submitting a request and getting the result,
wheras Win95 requests in parallel as a matter of course???   Changing compile
options on SAMBA to USE_MMAP=1 appeared to have a beneficial effect, though I
can't afford to run these sort of tests on our main fileserver to be sure.

[Apologies if this is a known bug or not relevant to the problem in hand - I had
been meaning to set up a spare machine with -STABLE and reproduce the problem
there before posting about it].


Want to link to this message? Use this URL: <"MAC-951025164255-1E49*/G=Andrew/S=Gordon/O=NET-TEL Computer Systems Ltd/PRMD=NET-TEL/ADMD=Gold 400/C=GB/">