RE: [Lksctp-developers] connect(2) won't return when ICMP unreach is received

[Forissier, J. <jer...@hp...>]

Sridhar Samudrala wrote:
> On Tue, 30 Nov 2004, Forissier, Jerome wrote:
>=20
>> Hi,
>>=20
>> I'm playing with lkcstp (2.6 kernel, RedHat Fedora Core 2), and I've
>> noticed the following: the connect(2) syscall won't return when an
>> ICMP Destination Unreachable packet is received. For instance, I'm
>> trying to connect to a Linux host that does not have SCTP loaded.
>=20
> Just like TCP, SCTP does retries if it doesn't get any response to
> INITs.

But in this case it _does_ get a response (ICMP destination
unreachable). IMHO this should cause connect(2) to return immediately.

> By default the no. of retries is set to 8(max_init_retries)
> each tried with an exponential backoff starting at an intial value of
> 3 secs(rto_initial) and a maximum backoff of 60 secs(rto_max). So you
> will see INITs sent at 0, 3, 9, 21, 45, 93, 153, 213 secs and finally
> the connect() will return with ETIMEDOUT at 273secs. You can
> access/change these defaults using the /proc interface under
> /proc/sys/net/sctp      =20

Indeed, I observed that INITs are retransmitted as you state. But (this
is another issue) connect(2) _won't_ timeout. My test program remained
blocked in connect(2) for 2+ hours... until I finally ^C'ed it...

--=20
Jerome