Unix domain sockets all disappeared on 2.0.21

Unix domain sockets all disappeared on 2.0.21

Post by John Saunde » Tue, 01 Oct 1996 04:00:00



Hi,

I just installed 2.0.21 and it's been running 14 hours so far. I checked
out the daily email only to find problems with news.daily unable to
talk to innd. I checked and innd was running, so I did a netstat -a and
found that _all_ of my unix domain sockets had gone. I checked the kernel
error log and found the following error. If this is caused by, or caused,
the lost sockets or not I don't know. Anyway I re-started innd and I've
got those sockets back. I re-started syslogd and got some /dev/log
sockets back. Has anybody ever seen anything like this before?

Sep 30 21:10:01 nhj kernel: general protection: 0000
Sep 30 21:10:01 nhj kernel: CPU:    0
Sep 30 21:10:01 nhj kernel: EIP:    0010:[<0014d7eb>]
Sep 30 21:10:01 nhj kernel: EFLAGS: 00010246
Sep 30 21:10:01 nhj kernel: eax: f000ef6f   ebx: 00e95808   ecx: 008fef5c   edx: 00000043
Sep 30 21:10:01 nhj kernel: esi: 00000000   edi: 00000000   ebp: 0000002f   esp: 008fef08
Sep 30 21:10:01 nhj kernel: ds: 0018   es: 0018   fs: 002b   gs: 002b   ss: 0018
Sep 30 21:10:01 nhj kernel: Process in.nnrpd (pid: 2178, process nr: 38, stackpage=008fe000)
Sep 30 21:10:01 nhj kernel: Stack: 00e95808 008fef5c 0000002f 00000000 00000000 0000002f 019a32e8 019a3370
Sep 30 21:10:01 nhj kernel:        0806a000 00135512 019a3370 008fef5c 0000002f 00000000 00000000 019a331c
Sep 30 21:10:01 nhj kernel:        0143e6c0 00000000 019a32e8 0806a000 0000002f 00000000 00000000 008fef54
Sep 30 21:10:01 nhj kernel: Call Trace: [<00135512>] [<00120eb2>] [<00128357>] [<0010a4e2>]
Sep 30 21:10:01 nhj kernel: Code: ff d0 83 c4 14 5b 5e 5f 5d c3 8d 76 00 b8 f5 ff ff ff 5b 5e

Sorry no symbols names as in.nnrpd has been stripped. Personally I think
this is caused by the missing sockets and thus only a symptom.

Cheers.
--            +------------------------------------------------------------+

    ,--_|\    |                - http://www.nlc.net.au/              (WWW) |
   /  Oz  \   |                - 018-223-814 or 02-9477-2881       (Phone) |
   \_,--\_/   | NHJ NORTHLINK COMMUNICATIONS - Supplying a professional,   |
         v    | and above all friendly, internet connection service.       |
              +------------------------------------------------------------+

 
 
 

Unix domain sockets all disappeared on 2.0.21

Post by Johann » Thu, 03 Oct 1996 04:00:00



Quote:

>Hi,

>I just installed 2.0.21 and it's been running 14 hours so far. I checked
>out the daily email only to find problems with news.daily unable to
>talk to innd. I checked and innd was running, so I did a netstat -a and
>found that _all_ of my unix domain sockets had gone. I checked the kernel
>error log and found the following error. If this is caused by, or caused,
>the lost sockets or not I don't know. Anyway I re-started innd and I've
>got those sockets back. I re-started syslogd and got some /dev/log
>sockets back. Has anybody ever seen anything like this before?

>Sep 30 21:10:01 nhj kernel: general protection: 0000
>Sep 30 21:10:01 nhj kernel: CPU:    0
>Sep 30 21:10:01 nhj kernel: EIP:    0010:[<0014d7eb>]

[Error removed because of lame more quoted text than new text rule]

Quote:>Sep 30 21:10:01 nhj kernel: Code: ff d0 83 c4 14 5b 5e 5f 5d c3 8d 76 00 b8 f5
ff ff ff 5b 5e

>Sorry no symbols names as in.nnrpd has been stripped. Personally I think
>this is caused by the missing sockets and thus only a symptom.

On a some what related note. I had a 486/120 running Slackware 3.0 running on
it for 6+ months fine. I upgraded to 2.0.0 and then to 2.0.15 right after it
was released. It ran for a couple of months fine also, then one day I started
getting ALOT of those error messages. Practically one for every process and
interrupt handler. It would do that until a) the HD interrupt handler had been
disabled or b) the keyboard interrupt handler had been disabled. Resulting in a
nice reboot since I couldn't do anything. No matter what I did, it would always
give me those exact same errors. I ended wiping the entire HD and resinstalling
RedHat Rembrandt. I tried using the 2.0.15, 2.0.0 and even the 1.2.13 kernel I
still had. Always that error. Could this be in someway related? I had the 486
doing IP masquerading alot of the time for 1-3 machines on our home ethernet
over a 33.6k PPP.

JE

 
 
 

1. UNIX domain sockets vs TCP sockets

Hello all,
 In R. Stevens's Unix Network Programming, second edition Vol. 1 he
states that "On Berkeley-derived implementations Unix domain sockets are
often twice as fast as a TCP socket". Now my question is has anybody
looked into whether this statement is true on non-Berkely
implementations, particularly Solaris?

  I did some testing on my own on Solaris 2.5 and was surprised to see
that unix domain sockets actually performed worse than TCP sockets (when
the client and server reside on the same box). For my test code I found
TCP sockets to be approximately 38% faster than unix domain sockets.

  Now if this conclusion is correct, why would that be the case?. And if
it's wrong, then has anybody studied this issue in some detail?.

Thanx
--
---------------------------
Wael R. Elwasif
Computer Science Department
University of Tennessee, Knoxville
Phone : (423) 974-6852
Fax   : (423) 974-4404

2. Netfilter & Patches

3. Problem with Unix Domain sockets on OSF Unix V4.0

4. fstab missing

5. piggy broken in 2.5.24 build - not piggy; it was BINUTILS

6. LILO 21 barfs w/msg about "pre-21 LILO signature"

7. dhcp does not work with linux 2.2.0-final???

8. PC-9800 additional for 2.5.61-ac1 (21/21) traps

9. Solaris 8 linker patch 109147-21 / 109148-21 breaks runtime linker audit

10. UNIX domain socket weirdness

11. UNIX domain datagram sockets

12. Unix Programming FAQ (v1.21)