rbj@icst-cmr.arpa (Root Boy Jim) (01/14/88)
Several people have mentioned the possibility of read-only roots or
other file systems. I will attempt to describe my solution.
On our VAX, we have three separate parallel filesystems. The first
contains the 4.3BSD tapes (except the source, which is mounted
another place, also read only) as distributed, mounted read only
in a file system called /ro. The second is a merged root and /usr
46M partition on an RM03 with hacked partition tables; the last 33440
sectors are the swap space and the rest is root/usr.
Large sections of the root are symbolically linked to the corresponding
file in the /ro partition. This includes /usr/doc and /usr/man/man?, as
well as quite a few `unused' programs in /usr/bin, such as the plot stuff.
These rarely used programs are available at the cost of a symlink
translation if needed.
The third is /rw, which contains any (or rather, most) changes to the virgin
root/usr file system. I use makefiles and rdist (in a rather nonobvious
way, using *relative* path names) to distribute changes from the /rw
filesystem to the real root/usr.
There are several advantages to this approach, as well as drawbacks.
First, we don't have jillions of backup tapes with /usr/doc on them.
Our root/usr backup fits on one tape.
Second, we have two copys of our precious configuration files, the real
one, and the one in /rw. Old versions are also kept in /rw.
The main drawback is setting everything up, and understanding how to
use it, altho if you want to change /x/y, you cd to /rw/x, change y,
and type make.
Another drawback is that we must have two copys of critical stuff,
such as /bin, /etc, and often used stuff in /usr/{bin,ucb,lib} etc.
Currently, this is about the size of a root partition, or 8M.
Lastly, if one wants to hack something in a symbolically linked directory,
one must remove the link, symbolically link all the files in that
directory except the replacement files, and update the whole mess.
This can get complicated when subdirectorys are involved.
Details upon request. Comments anyone?
(Root Boy) Jim Cottrell <rbj@icst-cmr.arpa>
National Bureau of Standards
Flamer's Hotline: (301) 975-5688
I've read SEVEN MILLION books!!
P.S. Fortunately, none by Jerry Pournelle!steve@edm.UUCP (Stephen Samuel) (01/21/88)
People have been talking about making a r/o root, but most of the
read-only stuff is in /bin, and /usr/bin. These are the prime targets
for romming, since they (theoretically) could be considered stable.
/etc and /lib have lots of r/o type stuff in them too, but both have
enough changable stuff in them that things could start to get messy reall
quick.
No matter what the case, a r/o root is not deadly (as long as /tmp and parts
of /etc are logically writable). Mounting onto a filesystem does not require
a write onto the associated directory.. In fact, I sometimes find some really
weird things in places like /tmp (or even /usr)when I unmount the file system
that was on top of it. THE DIRECTORY IS STILL THERE AND UNTOUCHED.
All that gets stomped on by a mount is some data in an internal system table.
the file system CAN (in theory) remain r/o.
--
-------------
Stephen Samuel Disclaimer: You betcha!
{ihnp4,ubc-vision,seismo!mnetor,vax135}!alberta!edm!steve
BITNET: USERZXCV@UQV-MTS