I've recently noticed that when using the FreeBSD TCP stack, the
connect() system call ignores the SO_REUSEADDRESS setting and returns
EADDRINUSE when attempting to connect a TCP socket if a connection
with the same (SrcIP,SrcPort,DstIP,DstPort) tuple is still hanging
around in the TIME_WAIT state.
Here's an example sequence of events:
* create a TCP socket
* set SO_REUSEADDRESS and SO_REUSEPORT options
* bind to port 12346
* connect to 1.2.3.4:54321
* send a little data
* close connection
* wait a few seconds
* create a TCP socket
* set SO_REUSEADDRESS and SO_REUSEPORT options
* bind() to port 12346
* connect() to 1.2.3.4:54321
The bind() will succeed due to the SO_REUSEADDR, SO_REUSEPORT option.
The connect() will fail with EADDRINUSE.
If you wait a minute or two between connections it works fine.
There appear to be a certain set of conditions when an existing
TIME_WAIT connection will be discarded and reused -- see line 747
below:
>From tcp_usrreq.c:
733 /*
734 * Cannot simply call in_pcbconnect, because there might be an
735 * earlier incarnation of this same connection still in
736 * TIME_WAIT state, creating an ADDRINUSE error.
737 */
738 error = in_pcbladdr(inp, nam, &ifaddr);
739 if (error)
740 return error;
741 oinp = in_pcblookup_hash(inp->inp_pcbinfo,
742 sin->sin_addr, sin->sin_port,
743 inp->inp_laddr.s_addr != INADDR_ANY ? inp->inp_laddr
744 : ifaddr->sin_addr,
745 inp->inp_lport, 0, NULL);
746 if (oinp) {
747 if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
748 otp->t_state == TCPS_TIME_WAIT &&
749 (ticks - otp->t_starttime) < tcp_msl &&
750 (otp->t_flags & TF_RCVD_CC))
751 otp = tcp_close(otp);
752 else
753 return EADDRINUSE;
754 }
But setting SO_REUSEADDR doesn't affect the behavior of connect().
Should it?
--
Grant Edwards grant.b.edwards Yow! VICARIOUSLY experience
at some reason to LIVE!!
gmail.com
--
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss
