strange bug with ip multicasting.

[Wu, Shao \(IT\)]

Summary: a random hanging bug when receiving multicast datagrams on
2.6.x kernels.
Keywards: hang, networking, kernel, multicast, recv, multicast loopback,
poll hang, select hang
Kernel version: 2.6.x
 
Full description of the problem:
-----------------------------------------------
 * This simple program demonstrates a bug in Linux kernel 2.6.x releases
(and
 * possibly 2.4 kernels also). I have tested this code on several 2.6
kernels
 * (RHEL 4.4, 4.5, 5.0 releases) all suffering the same bug. On Sun
Solaris 10,
 * however, it always behave correctly.
 *
 * This bug also affects select/poll's behavior as demonstrated in
CASE2.
 *
 * How to reproduce the bug...
 *
 *    Please make sure the multicast group address and the port number
are not
 *    currently in use on your test system. I use group address
239.255.1.1 and
 *    port number 12345 here.  Here are commands you can use to verify:
 *        netstat -gn
 *        netstat -an | grep 12345
 *    If you don't see port 12345 and group address 239.255.1.1 listed,
that is
 *    good.
 *
 * Test scenarios:
 * 1. Run one copy of this program, if you see it's running and
receiving its
 *    own messages, control-C out of it (because that is how it should
behave).
 *    And try again. If you try that a few times, you will see that this
program
 *    will hang on recv() function!  That is an incorrect behavior. This
program
 *    should always receive its own messages.
 *
 * 2. Follow scenario #1 until you get it hanging, open another window,
and run
 *    a second copy of this program. You will see both copies start
running and
 *    can receive each other's messages. That's a correct behavior --
the first
 *    datagram from the second copy takes the first copy of the process
out of
 *    hanging state.

Test program source file is attached. If you can't find the attached C++
source file, please send email to my personal email account at
swu001@xxxxxxxxx and I'll send the testing code from there.
--------------------------------------------------------

NOTICE: If received in error, please destroy and notify sender. Sender does not intend to waive confidentiality or privilege. Use of this email is prohibited when received in error.

Attachment: simplemcast.cc
Description: simplemcast.cc