FD_CLOSE on socket error

[Dan Loch]

Hello,

I have run into an intermittent problem with a TCP/IP interface.  It occurs most nights in the wee hours of the morning.  Only occurs one time per day, the rest of the day all is well with the interface.

I am receiving the error:

07/18/2005 11:41:00 [pti :sele:ERR /0:qsight_adt_cmd] Network error on FD_CLOSE on socket 2028: 10053

07/18/2005 11:41:00 [pti :sele:ERR /0:qsight_adt_cmd] Network error on FD_CLOSE on socket 2028: 10053

07/27/2005 06:21:05 [pdl :PDL :ERR /0:to_qsight_adt] write of 468 bytes failed: Unknown error

07/27/2005 06:21:05 [pdl :PDL :ERR /0:to_qsight_adt] PDL signaled exception: code 1, msg write failure

I found the following in the archives.

View Enclosure: Problem

Defect: HUBad08811 Added 030115 by rabbott

---

The normal way to shutdown a socket is for the remote end to send an a

tcp packet with FIN+ACK flags set. Some brain-dead applications will

send RST instead, which is a “hard shutdown” of the socket.

If PDL receives a RST, the engine becomes very unstable. You’ll see

the following errors in the log file:

[pti :sele:ERR /0: pdl_cmd] Network error on FD_CLOSE on socket

808: 10053

[pti :sele:ERR /0: pdl_cmd] Network error on FD_CLOSE on socket

808: 10053

Note that these errors are coming from pti, not pdl. The engine will

then become unstable, sometimes consuming 100% cpu, sometimes panicing

if a hcicmd is sent, etc.

The length-encoded tcp driver seems to behave normally.

This problem was discovered at a customer site with Cloverleaf 3.7.1P

on Win2K. There are no known workarounds!

---

The remote vendor I am sending to claims there have been no abnormal shutdowns and we are at a loss to determine the cause of this problem.

Has anyone else seen this and have you been able to determine the cause and fix for this issue?  Is this an issue with the Cloverleaf software?

This interface is our only outbound VPN interface.

Currently running Cloverleaf 3.7.1P on Win 2000.

I have implemented 5.3 with HA failover on new hardware (still Win 2000) but have not activated production interfaces there yet.

Thanks for any info or help.

Dan Loch

Crittenton Hospital

Rochester, MI

Topic

[Kevin Scantlan]

So, it this is a bug, has it been fixed?  If so, what version and what release?

Thanks.

[Rob Abbott]

Yes, this was a bug and was fixed in 3.8.1P rev4.  It only appeared on Windows, and only if a RST packet is received by a PDL thread.

If you need to trace your network to determine whether RST packets are being received, you can use the Ethereal packet sniffer from http://www.ethereal.com

Rob Abbott
Cloverleaf Emeritus

[Kevin Scantlan]

Is there an equivalent patch for versions greater than 5.0?

[Rob Abbott]

The following 5.x releases have the fix:

5.1 Rev 2

5.2.1 and above (includes 5.3, etc.)

Rob Abbott
Cloverleaf Emeritus

[Author]

Replies