Socket destroy delayed (r=0 w=252)
I've gotten 12 of them so far, spread out over the past 12 hours
or so.
I had never noticed these before, but it seems that I did have
four of them one day (with 1.3.73), and they coincided with my
reconfiguration of named. ... interesting ...
<sherlock mode>
Now I find, in looking at my logs, that these errors seem to
happen mostly either right after a named "Connection refused"
message, e.g.:
Apr 13 22:52:31 luna named[50]: recvfrom: Connection refused
Apr 13 22:52:36 luna kernel: Socket destroy delayed (r=0 w=252)
... or a after an ftp disconnection. The ftp connection is
harder to validate, and may not in fact be related at all. It's
just the only other thing I could find that was going on.
I'm don't feel confident enough with my knowledge of the issues
involved to go debugging this stuff yet, but I did grep around a
bit and found this little comment in af_inet.c:
/* this should never happen. */
/* actually it can if an ack has just been sent. */
printk("Socket destroy delayed (r=%d w=%d)\n",
sk->rmem_alloc, sk->wmem_alloc);
<!-- I've always enjoyed "can't happen" bugs. ;^) -->
Anyway, it would appear to be about the infamous "ack" problem,
(judging from the comment only).
</sherlock mode>
I do have a moderately busy web site on this machine. I'm running
IP Aliasing, ksmbfs, and samba. The machine is a Pentium 100, with
16Mb ram, an IDE 1.2Gb, and a 3c509C.
I'm running Apache 1.0.0, wu-ftp 2.4, named 4.9.3-BETA9.
Hope this helps,
--Bill
+--------------------------------------------------------------------------+
| http://www.bearnet.com/ || http://www.weinman.com/wew/
| Author of The CGI Book -- http://www.cgibook.com/