SUMMARY: Problem with NFS freezing with "svcktcp_reply: mbuf_send returned 32" in the kern syslog

From: <rhaskins_at_shiva.com>
Date: Fri, 13 Feb 1998 09:28:58 -0500

Thanks to all the folks who replied (in the order in which they arrived in
my inbox):

Knut Hellebo Knut.Hellebo_at_nho.hydro.com
Michael Matthews matthewm_at_voyager.sgate.com
Craig T. Biggerstaff craig_at_blkbox.com
Phil Farrell farrell_at_pangea.stanford.edu
gege_at_cal.enst.fr

A summary of the responses I received:

1) ensure that we are running the latest patched version of 4.0a (we are)
2) the problem is being caused by a route problem causing buffers to
overflow
3) use nestat -a and netstat -m to try to figure out what processes are
running out of buffers

After some discussion with Digital, they indicated that this pointed to a
problem with NSR and/or cron. Cron has not been changed, but we about a
week(!) prior to this problem, we did change our NSR configuration
slightly. I have difficulty seeing how this relates to a tcp buffer
problem, as the machine only runs NSR locally and is not an NSR server.

The problem simply disappeared, and has not returned since. It remains a
mystery as to what caused this, and further, what corrected it.

Bob Haskins
Senior Unix Administrator
Shiva Corporation
rhaskins_at_shiva.com


The original message is below:


Good morning. We are experiencing a severe problem with NFS freezing with
the following error message in the kern syslog:
Feb 11 05:59:47 alpha vmunix: svcktcp_reply: mbuf_send returned 32
Feb 11 05:59:47 alpha vmunix: rfs_dispatch: sendreply failed
Feb 11 05:59:47 alpha vmunix: svcktcp_reply: mbuf_send returned 32
Feb 11 05:59:47 alpha vmunix: rfs_dispatch: sendreply failed
I've checked the archives for this problem, and there is one entry about
this exact problem but no summary. There are other entries with
"svcktcp_reply" and "rfs_dispatch" but they appear to deal with things
besides NFS and are several years old, again with no summaries. I can find
no documentation on svcktcp_reply, mbuf_send, rfs_dispatch or sendreply in
the man pages.
The configuration of the machine is as follows (from the last reboot):
Feb 10 18:42:58 alpha vmunix: Digital UNIX V4.0A (Rev. 464); Sat Jan 24
10:06:35 EST 1998
Feb 10 18:42:58 alpha vmunix: physical memory = 512.00 megabytes.
Feb 10 18:42:58 alpha vmunix: available memory = 497.57 megabytes.
Feb 10 18:42:58 alpha vmunix: using 1958 buffers containing 15.29 megabytes
 of memory
Feb 10 18:42:58 alpha vmunix: Digital AlphaPC 64 274 MHz system
Feb 10 18:42:58 alpha vmunix: DECchip 21072
Feb 10 18:42:58 alpha vmunix: Firmware revision: 4.8
Feb 10 18:42:58 alpha vmunix: OSF/1 PALcode version 1.45
Has anyone seen this? The NFS freeze-ups seem to be getting longer over
time, to the point that NFS is totally unusable. A reboot corrects the
situation for a short period of time (20 minutes or so), but over the
course of 4 hours or so NFS gets totally unusable. I *will* post a summary,
*if* I receive any replies.
Received on Fri Feb 13 1998 - 15:32:38 NZDT

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:37 NZDT