dyer@spdcc.COM (Steve Dyer) (11/10/90)
Configuration: Sun SLC w/local disk running SunOS 4.1 and RU-CAP 2.0 with the enet packet filter installed in the running /vmunix. /etc/atalk.local contains the single line: "enet0 spdcc-etalk 1025". There is a single MacIIfx (an eventual Aufs client) running ethertalk phase one on the same ethernet. No appletalk routers involved. CAP entries in /etc/services (no YP) have been added. I manually start aarpd and atis. Atistest times out and asks if atis is running. Atlook just returns after looking for =:=@*. Aufs reports that SrvrRegister fails. Running debugging on the atis daemon doesn't tell me too much--it seems to receive and drop on the floor a ddpRTMP request packet which the Mac sends on reboot or network restart. I guess that's an indication that the ethernet filter is at least returning SOME data. This is about as absolutely a vanilla ethertalk CAP build that I can imagine. I just ran Configure and gen.makes, and used all the default settings. It's not clear to me why it shouldn't be working. Has anyone tried this successfully? Do you have any words of wisdom to offer? -- Steve Dyer dyer@ursa-major.spdcc.com aka {ima,harvard,rayssd,linus,m2c}!spdcc!dyer dyer@arktouros.mit.edu, dyer@hstbme.mit.edu
dyer@spdcc.COM (Steve Dyer) (11/10/90)
In article <4836@spdcc.SPDCC.COM> dyer@ursa-major.spdcc.com (Steve Dyer) writes: >I manually start aarpd and atis. Atistest times out and asks if >atis is running. Atlook just returns after looking for =:=@*. >Aufs reports that SrvrRegister fails. After digging around in the guts of the code, it seems that nis_node and nis_net are not getting initialized when ethertalk support using /etc/cap.ether.shared is used. When it isn't used, nis_node and nis_net are initialized from this_node and this_net. Patching this seemed to get me past this hurdle (i.e. I could run lwsrv.) This seems like a rather egregious bug that anyone who tried to bring up CAP/ethertalk would hit immediately. -- Steve Dyer dyer@ursa-major.spdcc.com aka {ima,harvard,rayssd,linus,m2c}!spdcc!dyer dyer@arktouros.mit.edu, dyer@hstbme.mit.edu
hedrick@athos.rutgers.edu (Charles Hedrick) (11/13/90)
nis_node and nis_net are set correctly if you let atis figure out the network number by listening to RTMP broadcasts from the bridges. Sounds like insufficient testing of the code for running on networks without bridges. The reason it might work in a simple test is that nis_node and nis_net are mapped into locations in /etc/cap.ether.shared. Suppose the code was debugged on a network that has bridges. They would have gotten correctly set. If we shifted to a configuration that didn't listen to RTMP, but didn't zero /etc/cap.ether.shared, the correct values would have remained in the file from the other configuration. Thus the fact that they didn't get set would not show up. We're planning a new release of CAP/ethertalk anyway, to fix various reliability issues that we discovered over some period of use. This sounds easy enough to fix.
brad@CAYMAN.COM (11/18/90)
>> nis_node and nis_net are set correctly if you let atis figure out the >> network number by listening to RTMP broadcasts from the bridges. >> Sounds like insufficient testing of the code for running on networks >> without bridges. The reason it might work in a simple test is that Sorry if this is old; running ecap, it does not seem work if there are no bridges (nis_node is never set) I have a simple fix for this (and I think the diff below is right, but don't quote me) andros% diff atis.c atis.c~ 50d49 < #include "../lib/cap/ethertalk/sharedpage.h" 113d111 < u_char *sharedpage; 403,404d400 < < nis_node = this_node; andros% -brad