Date: Wed, 24 Jul 1996 01:43:33 -0400 (EDT) From: Dmitry Kohmanyuk <dk@dog.farm.org> To: phk@critter.tfs.com (Poul-Henning Kamp) Cc: dk+@ua.net, freebsd-hackers@freebsd.org Subject: Re: lp booting: first cut Message-ID: <199607240543.BAA00994@dog.farm.org> In-Reply-To: <406.838160103@critter.tfs.com> from Poul-Henning Kamp at "Jul 24, 96 00:15:03 am"
next in thread | previous in thread | raw e-mail | index | archive | help
Quoting Poul-Henning Kamp:
> >Now if somebody would explain me how to trace this mysterious hang by DDB.
> >I try to boot -s, and still get a hang. ctrl-alt-esc just throws me into
> >debugger with stack trace down to idleloop() or something. `ps' in ddb
> >doesn't show anything but a basic crowd (swapper init pagedaemon vmdaemon
> >update). I would try to boot -d now.
>
> Try to run tcpdump on the server and see what comes your way.
as I said, the tcpdump in 2.2-960501-SNAP doesn't work with lp0.
I have carefully looked at BPF stuff in lpt.c and cannot understand why.
I would be grateful for a patch.
also, it appears that portmapper is broken at least twice in this snap:
first, it cannot get list of all addresses (including aliases) - I have
resorted to hack on from_local.c:from_local() to return TRUE always ;-(
second, mountd doesn't work if hostname is bound to alias or to nearest
end of slip interface, e.g.:
working configuration:
!ifconfig -au
lp0: flags=8851<UP,POINTOPOINT,RUNNING,SIMPLEX,MULTICAST> mtu 1500
inet 193.124.48.195 --> 193.124.48.198 netmask 0xffffff00
^^my hostname^^ ^^^diskless^^^
tun0: flags=8051<UP,POINTOPOINT,RUNNING,MULTICAST> mtu 1500
inet 10.0.2.15 --> 10.0.2.2 netmask 0xff000000
lo0: flags=8009<UP,LOOPBACK,MULTICAST> mtu 16384
inet 127.0.0.1 netmask 0xff000000
inet 193.124.48.200 netmask 0xffffff00
inet 193.124.48.201 netmask 0xffffff00
inet 193.124.48.202 netmask 0xffffff00
if dog.farm.org (193.124.48.195) is bound as alias to lo0 instead
(and I use other address for closer end of lp0), or if I bound it
to closer end of sl0, mountd and nfsd cannot register with portmapper:
not-working configuration:
changed lp0:
lp0: flags=8851<UP,POINTOPOINT,RUNNING,SIMPLEX,MULTICAST> mtu 1500
inet 193.124.48.197 --> 193.124.48.198 netmask 0xffffff00
added sl0:
sl1: flags=c010<POINTOPOINT,LINK2,MULTICAST> mtu 552
inet 193.124.48.195 --> 0.0.0.1 netmask 0xffffff00
I understand that binding addresses to separate slip interface is a kludge,
but it is somewhat more reliable kludge than using interface alias.
(and I have used to do it on SCO(tm) Unix(tm) 3.2.4.2 <horror>,
which doesn't have aliases.)
Also, after a couple of hours spent with DDB (btw, why a manual page
mentions x/m which doesn't work??), I figured that the hang is inside
infinite loop in nfs/nfs_socket.c:nfs_reply(). I have then changed
the place where I put my hostname (e.g., changed from non-working
to working configuration ;-)), and it started to work! So it
seems that this is a multi-homed setup I have which causes this trouble.
(probably nfs client sends packets to .195, and expects answers coming
back from this address, but they arrive with .197. )
Now I managed to get it single-user (flags s in bootp config file)
until `Enter pathname of shell or RETURN for sh:' and `Erase is...'
line after that. Then, it hangs. (the only tool I have on the remote
side is netstat -I lp0 4, which is pretty enough to see when the interface
dies; I cannot ping anymore, I cannot even break to DDB on the client.)
At this point, I start to think that my DTR-1 (Cyrix 486SLC, running
at 8MHz since running it at 25MHz causes reboot just after loading
kernel) is somewhat fundamentally broken and non-compatible with PLIP.
I have already tweaked LPMAXSPIN[12].
Please somebody test my patches on more standard machine and report
whether they work or not. I am 99% sure that they would be absolutely OK.
Dmitry.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199607240543.BAA00994>