udhcp discovery takes minutes

[Denis Vlasenko]

On Monday 14 January 2008 23:40, Danny Sung wrote:
> I'm using udhcpc as packaged in busybox 1.9.0.  For some reason initial 
> discovery can sometimes take several minutes (though once in a while 
> it'll complete in seconds).  I had previously been using busybox 1.6.1, 
> which didn't have any of these problems.
> 
> Glancing at the 1.9.0 code, I should be using the default 
> 'discover_timeout' of 3 seconds.  But I noticed that 'timeout' is 
> initialized as a global to 0, which would cause the select call in the 
> main loop (line 341) to hang forever if it never gets a reply.

        state = INIT_SELECTING;
        for (;;) {
                tv.tv_sec = timeout;
                tv.tv_usec = 0;
                if (listen_mode != LISTEN_NONE && sockfd < 0) {
                        ...creates socket...
                }
                max_fd = udhcp_sp_fd_set(&rfds, sockfd);
                retval = 0; /* If we already timed out, fall through, else... */
                if (tv.tv_sec > 0) {
we do NOT go in here on first pass -
tv.tv_sec == timeout == 0!
                        DEBUG("Waiting on select...");
                        retval = select(max_fd + 1, &rfds, NULL, NULL, &tv);
                }
                if (retval < 0) {...}
                /* If timeout dropped to zero, time to become active:
                 * resend discover/renew/whatever
                 */
                if (retval == 0) {
                        switch (state) {
                        case INIT_SELECTING:
so we have to end up here
                                if (packet_num < discover_retries) {
                                        if (packet_num == 0)
                                                xid = random_xid();
                                        /* send discover packet */
and send discover here
                                        send_discover(xid, requested_ip); /* broadcast */
                                        timeout = discover_timeout;
                                        packet_num++;
                                        continue;
                                }

Can you verify that it indeed happens as expected
(by adding bb_error_msg("DEBUG: we are here") style markers)?

If not, where it goes instead?

> Looking at my DHCP server logs, it seems my server never sees the first 
> discovery request.  But that would explain the hang.  I tried to set 
> timeout to 1 in line 22, but that doesn't appear to solve the problem.
> 
> In fact when I do a network trace, I only see packets coming out when it 
> finally does succeed.  So apparently I'm having strange issues even 
> sending out packets.  You can also tell that the "Sending discover..." 
> message only gets sent once and just hangs, rather than timeout out once 
> a second (or 3 seconds?).  Busybox 1.6.1 appears to be doing the correct 
> thing and sending discovers once a second until it gets something.


--
vda