Re: More than 1 NIC - Coda Server
Hi > I had trouble under a previous version (4.6.5 I think) with a > multi-homed server. The problem seemed to be that when I did > venus-setup on a client with a non-primary address of the server the > client's venus didn't end up with proper communication with the > server. I believe the venus's server-bound packets got delivered, but > the reply packets had an ip src addr of the server's primary address, > so they were ignored by the venus. > ... Sorry Jan, i think was wrong with my Port binding assumtion, and everybody excuse me for sending this long mail with LOG files to a mailing list ... :) My Problem sounds more like the one describeda by Greg: if I put on node on the first NIC (eth0) everything is fine. if i just change the cable and put the same node on eth1 - venus tells me that the server is down - and exits. tcpdump reveals that communication is done - in both directions - in both cases... With turned on debugging, The venus logs on are identically until on the machine it works : [ W(14) : : 17:01:29 ] userent::Connect: RPC2_NewBinding(81ce) [ W(14) : : 17:01:29 ] userent::Connect: RPC2_NewBinding -> RPC2_SUCCESS [ W(14) : : 17:01:29 ] userent::Connect: NewConnectFS(81ce) on the machine that doesnt work : [ W(14) : : 16:51:45 ] userent::Connect: RPC2_NewBinding -> RPC2_NOBINDING (F) [ W(14) : : 16:51:45 ] srvent::Reset: host = 81ce [ W(14) : : 16:51:45 ] srvent::Reset: RPC2_Unbind -> RPC2_NOCONNECTION (F) [ W(14) : : 16:51:45 ] vsgdb::DownEvent: host = 81ce [ W(14) : : 16:51:45 ] vsr::~vsr: uid = 0 [ W(14) : : 16:51:45 ] NotifyUserOfServerDownEvent ... [ W(14) : : 16:51:45 ] Getattr : returns Connection timed out, elapsed = 1. 0 msec [ X(00) : : 16:51:45 ] CHILD: mount system call failed. Killing parent. I will later try to put the RPC2 into debugging mode, parse some tcpdumps, and or try to locate this Problem - if I get more Information I will report. Greg : How did you solve that Problem? Did you ? robin --- Robin Gareus Universitaetsrechenzentrum Im Neuenheimer Feld 293 69120 Heidelberg Email : [EMAIL PROTECTED] Phone : 06221/54 4599 ---
Re: More than 1 NIC - Coda Server
| I had trouble under a previous version (4.6.5 I think) with a | multi-homed server. | ... | I believe the venus's server-bound packets got delivered, but | the reply packets had an ip src addr of the server's primary address, | so they were ignored by the venus. Yes, the clients used to talk back to whatever address the server put in the header of the rpc2 packet. At some point, right before 5.0.0 as I recall, I made modifications to use the sender-address as returned by recvfrom, as far as I know this should be the preferred address that can be used to send replies back to the peer. Jan
Re: More than 1 NIC - Coda Server
I had trouble under a previous version (4.6.5 I think) with a multi-homed server. The problem seemed to be that when I did venus-setup on a client with a non-primary address of the server the client's venus didn't end up with proper communication with the server. I believe the venus's server-bound packets got delivered, but the reply packets had an ip src addr of the server's primary address, so they were ignored by the venus. I'm fuzzy on the details, but would suggest looking with tcpdump. (You didn't indicate if you tried this, which addresses you used in the various venuses, and whether you got ICMP port unreachables back, or something else.) Greg Troxel <[EMAIL PROTECTED]>
Re: More than 1 NIC - Coda Server
On Wed, Jun 16, 1999 at 04:26:44PM +0200, Robin Gareus wrote: > Hi! > I just set up a Coda SCM on a (linux) Machine with 3 NICs, > each in a different subnet. The Problem is, that the Coda Server only > sets up the Ports for the first NIC. Requests from the other subnets are > not replied. > Do I have to run 2 additional Servers on that Machine ? - How do I set > them up ? (different config files ??) > > Thanks > > robin Hi Robin, It shouldn't do that. Following is the output from 'netstat -a' and as you can see, the coda server does not bind to any specific address: udp0 0 *:codasrv *:* udp0 0 *:codasrv-se*:* Compare this to the nameserver which does bind itself to interfaces: udp0 0 localhost:domain*:* udp0 0 verdi.coda.cs.cm:domain *:* udp0 0 *:domain*:* So, my guess is that it is a routing/setup issue. But, I don't have any machines with multiple NIC's/networks here, so I can't tell you what is wrong or how to fix it, it could be some wrong assumption or bug in the rpc2 code. Turning rpc2-debug messages up to level 10 will give a _lot_ of information in /vice/srv/SrvErr, which might give clues. I think the easiest way of increasing the rpc2 debuglevel is attaching gdb, and changing the corresponding variable: # gdb codasrv `pidof codasrv` gdb> p RPC2_DebugLevel = 10 gdb> c Jan