Date: Mon, 17 Oct 2005 19:18:58 -0400 From: Miles Nordin <carton@Ivy.NET> To: freebsd-sparc64@freebsd.org Subject: Re: Is there a nice diskless HOWTO? Message-ID: <oqd5m3zq8t.fsf@castrovalva.Ivy.NET> In-Reply-To: Bruce O'Neel's message of "Mon, 17 Oct 2005 08:24:39 %2B0000" References: <20051017082439.GA9372@SDF.LONESTAR.ORG> <oqll3uox1n.fsf@castrovalva.Ivy.NET> <20050726161047.GJ62369@funkthat.com> <20050726162755.GA67178@xor.obsecurity.org> <20050726120052.7EFAE16A41F@hub.freebsd.org> <42E7119B.5010101@alaska.net> <200507270515.j6R5Fhj17277@hobbit.sirius.net.au> <oqzms7naaq.fsf_-_@castrovalva.Ivy.NET>
next in thread | previous in thread | raw e-mail | index | archive | help
[-- Attachment #1 --]
>>>>> "bo" == Bruce O'Neel <edoneel@sdf.lonestar.org> writes:
bo> Hi, Is there a nice diskless HOWTO?
AFAICT no.
I believe you can install by hand just as on NetBSD, except while for
NetBSD you just extract the .tgz files, for FreeBSD you have to cat
together the blah.a{b,c,d,e,f} chunks, and then pipe that into gzip |
pax -rpe. or is it bunzip2, i forget.
The three significant differences I found between FreeBSD and NetBSD
nfsroots are:
* On NetBSD, the 'option root-path' option is a path only, not a
server IP, and the IP address of the root NFS server if it differs
from the 'next-server' goes into the 'option swap-server' field.
NetBSD doesn't document this, but that's how it is. On FreeBSD,
the root-path option is "1.2.3.4:/path/to/root" as you'd expect.
* FreeBSD is broken all over the place if the NFS server doesn't
support locking. like, 'vi' doesn't work for example. and AFAICT
it is impossible to disable the default locking-enabled mount
options of an NFS root. NetBSD doesn't use locking anywhere---I
think there is no working NetBSD code written for NFS locking,
client or server.
* NetBSD NFSroot users like to make mfs's and tmpfs's for things like
/dev, /tmp, /var/run. FreeBSD doesn't have mount_mfs---it uses
something else. I couldn't get FreeBSD's mdmfs to work at all, but
FreeBSD has a devfs so it's good there's no incentive to make a
/dev mfs as on NetBSD.
-----8<-----
>>>>> "jg" == John-Mark Gurney <gurney_j@resnet.uoregon.edu> writes:
>>>>> "kk" == Kris Kennaway <kris@obsecurity.org> writes:
>>>>> "rw" == Royce Williams <royce@alaska.net> writes:
>>>>> "a" == alm <alm@sirius.net.au> writes:
jg> http://people.freebsd.org/~murray/sparc64/install.html
kk> Here's a snippet from my dhcpd.conf
rw> http://ezine.daemonnews.org/200301/sparc64-nfsroot.html
a> -Configure a static DHCP lease for it:
a> host netboot {
a> hardware ethernet <mac-addr> ;
a> option host-name "<hostname>" ;
a> fixed-address <IP addr.> ;
a> always-reply-rfc1048 on;
a> filename "/boot/loader";
a> next-server <TFTP/NFS server> ;
a> option root-path "<TFTP/NFS server>:/mnt" ;
a> }
It turned out to be pretty easy just to get the system booted once
hearing that /boot/loader could be fed straight to OpenPROM over tftp.
This:
a> boot net:dhcp
behaved the same on my sparc64 as 'boot net'. The OpenPROM used RARP
(not DHCP) to find the second-stage loader even with 'boot net:dhcp'.
Then it TFTPs to whatever machine replied to the RARP. I wonder if
net:dhcp is a cue to Solaris's inetboot? They have this massively
complicated 'wanboot' framework now, too, including multiple
cryptographic keys and CGI scripts on web servers. <shudder>.
I put /boot/loader in the tftpboot directory and named it C0A8017C to
match the IP of my Sun, 192.168.1.124. /boot/loader then used DHCP
(not bootparams) to get just one option, root-path:
host amber {
fixed-address amber;
hardware ethernet 08:00:20:xx:xx:xx;
option root-path "69.31.131.61:/export0/nbnfs/amber";
}
and /boot/loader loaded loader.conf and whatever else it needed over
NFS rather than TFTP. The only thing loaded over TFTP was
the file C0A8017C (a copy of /boot/loader).
a> lofiadm
I didn't end up using this nor the mfsroot, because I didn't use
sysinstall. I just extracted all the {base,doc,src,..}.{aa,ab,ac...}
files by hand. But it's good to have those instructions written down
because others will probably want to netboot into sysinstall.
Now I have three more problems:
1. The NetBSD NFS server I'm using doesn't support locking, so I
can't run vipw or pwd_mkdb, and vi always says 'UNLOCKED' in the
status bar. I can't 'mount -u -o -L 69.31.131.61:/ex.... /'
because the -L option is documented, obeyed only on initial mount,
not for remount. Is there a way to get this option turned on for
the initial root mount?
2. I tried 'mdmfs -s 32m md /tmp' and it never returned to the
prompt. ^C doesn't work, and I can't get to 'ok' or to 'ddb>' by
pressing BREAK. Is there a sysctl to obey BREAK?
3. Boot hung after ``waiting 15 seconds for SCSI devices to settle''
when I have my Firewire card installed. The Install CD boots
successfully with the card installed and even makes an fwe0. Is
there a way to use the loader to disable a broken device like this
SCSI-over-Firewire? I tried commands like 'disable-module sbp'
but it always says 'sbp not found'.
sorry for all the questions. I'm just getting started.
[-- Attachment #2 --]
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.6 (NetBSD)
iQCVAwUAQ1QxYonCBbTaW/4dAQKS+QP/ZpM3L223fijmWzuS/NpUHl+5Hss7pRWg
eyrwD7s+aKrhPHSI5EC1ho5niQJfP3UeM604rOxHrlVHTmM+9rdjdArPU03eczTH
8X1Irb2g3q96diQ3O9VbD68S/SwluXpmjvhKl3cI5DFWvUKzbYyRcN/dq+JKckiS
R4YOmTpRzVo=
=MqRP
-----END PGP SIGNATURE-----
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?oqd5m3zq8t.fsf>