Re: [Lksctp-developers] SCTP association restart

[Vlad Y. <vla...@hp...>]

Peter Hanecak wrote:
> Hello,
> 
> I'm trying to trigger SCTP ASSOC CHANGE(SCTP_RESTART) notification on my 
> server application which is using SCTP socket API. Im trying to do that 
> on RHEL 5.1 . The idea is to test the server application whether it 
> correctly handles the reboot of the client machine (those reboots 
> ussualy create new association instead of triggering restart 
> notification thus is not a practical way to test the case).
> 
> I'm trying to do that by initiating new SCTP association and capturing 
> the INIT and COOKIE_ECHO messages my test client is sending to the 
> server application. Then I simply replay those packets again.
> 
> So far I've been unable to trigger this restart event notification.
> 
> 
> 1) identical duplicates:
> 
> If I replay INIT and COOKIE_ECHO packet unchanged, server replies with 
> INIT_ACK (after INIT) and then COOKIE_ACK (after COOKIE_ECHO) but server 
> application does not receive any notification for existing association. 
> And no new association is created instead.
> 
> I would expect the yes, no new association is created but that my server 
> would get "restart" notification from the socket API.

No, a restart is trigger with a different verification tag.
> 
> 
> 2) modified duplicates:
> 
> If I replay modified INIT (different Initiate tag then first INIT), I 
> get INIT_ACK from server. Then I replay COOKIE_ACK (using value of 
> Initiat tag from INIT_ACK to modify Verification tag in duplicate 
> COOKIE_ECHO. But I get no COOKIE_ACK from the server. And server 
> application gets no notification nor is there any message from the kernel.
> 
> I would expect ... well, do not know. :)
> 
> 
> On the other hand, I am able to trigger SCTP ASSOC 
> CHANGE(SCTP_COMM_LOST) by sending ABORT.

You might want to take a look at the SCTP test tool.  You can
get the code from http://sctp.fh-muenster.de/sctp-testtool.html


The idea behind restart is that one end terminated the connection,
while the other end did not.

Another way you can simulate the restart would be this:

1. On the client, set SO_LINGER with l_onoff set to 1 and l_linger set to 0.
This will tigger an ABORT on close().
2. on the client, set ip_table rule to discard an outgoing ABORT.
3. On the client, after you close the socket, reconnect it with the same ports
   and addresses as before.

The above should generate a RESTART notification.

-vlad


> 
> Additinal information:
> 
> # modinfo sctp
> filename:       /lib/modules/2.6.18-53.el5/kernel/net/sctp/sctp.ko
> license:        GPL
> description:    Support for the SCTP protocol (RFC2960)
> author:         Linux Kernel SCTP developers 
> <lks...@li...>
> alias:          net-pf-2-proto-132
> srcversion:     EA347AE19CDDAFFDB5A3CA4
> depends:        ipv6
> vermagic:       2.6.18-53.el5 SMP mod_unload gcc-4.1
> module_sig: 
> 883f35047d39e6836e299a424cb51a112843109e2f64fc6bac16274c9c7feb16ea1f6d43374e6c09f669fbf51166c2dd1e3a393c60983d3444134c0
> 
> 
> So the question is: How precisely do I trigger SCTP association restart 
> notification in my server application? Or is there some testing tool you 
> use for this (and if so, where can I get it and how to use it)?
> 
> Btw, I did found this:
> 
> http://sourceforge.net/mailarchive/message.php?msg_id=4767E7DB.4090203%40hp.com
> 
> "The system reboots yet again and the third INIT is send with the source 
> port of 32768. In this case, the ports and addresses are the same, so 
> this treated as association restart. In this case, the server will 
> re-initialize and re-use the existing association. This is expected 
> behavior."
> 
> This should be what I simulate with [1] except that my server 
> application des not get the restart notification from the socket API.
> 
> 
> Thank you in advance for any feedback.
> 
> Sincerely
> 
> Peter
>