On Mon, Aug 24, 2015 at 10:52:31PM +0200, Simon Kitching wrote:
>
> Ken: just one note. A "unified usr" works fine with /usr on a separate
> filesystem. An initrd is necessary, but AIUI in simple cases it can consist
> of two files: busybox + a 4-line shellscript.
>
Bloat ;-) [ busybox *and* 'conventional' full-featureed coreutils etc ]
One of the things I like about LFS is the lack of an initrd : one
less thing to go wrong when I test a new kernel.
> Bruce: AIUI, the main reason for moving /bin/* into /usr/bin rather than
> /usr/bin/* into /bin is that /etc usually needs to be writable by root, ie
> the filesystem on which /etc really resides must be mounted readwrite. But
> for clustered systems it is very desirable for the system binaries to be on
> a filesystem that is mounted read-only. The fedora approach allows a single
> readwrite filesystem (rootfs) and a single readonly filesystem (/usr) which
> contains all binaries and their corresponding libs.
[ nostalgia: ] Ah, the past attempts here to make /etc r/o by my
predecessors such as 'archaic'. I remember them well, and still
like the idea of making things r/o, particularly on a server.
The problem with your description here is that *any* update to
programs or libraries, or installing extra packages, breaks that
ideal. On production systems you could write once to a staging
system and then when completed roll out 1 or more copies to replace
the current production filesystem, but unless I am mistaken that
will need a reboot.
In BLFS we are not particularly great at *pointing* people to
vulnerability fixes, but fedora often have several per day. I am
sure that a r/o /usr will remove some possible vulnerabilities, but
to me it seems to make it harder to update. But: I don't maintain
any production systems, so I might be talking out of my backside.
>
> Just a thought: maybe there is some interest in having "unified usr" in the
> lfs-systemd book, but not in the lfs-sysv one? There is no relation at all
> between systemd and "unified usr" except that they both originated at
> RedHat, but those with an interest in systemd possibly are more interested
> in a less-traditional filesystem layout too.
>
That is up to Douglas.
> Anyway, would it be ok if I send in a patch for the LFS book adding a single
> paragraph just mentioning that "unified usr" is a possible approach, and
> that the mv/ln instructions from the LFS book can be left out in that case
> (with initrd from BLFS needed if /usr is a separate partition)? Maybe
> section 6.5 ("creating directories") would be appropriate?
>
I would prefer to see a hint (at least to start with), explaining
the details of how to do it and also explaining *why* it offers an
advantage - to me, a few 'mv' and 'ln' commands are not a big deal
(and I say that as somebody who forgot to change a hardcoded library
version in one of my own scripts : learning from our mistakes tends
to stick more than just following the instructions, and we are
primarily about learning).
ĸen
--
This one goes up to eleven: but only on a clear day, with the wind in
the right direction.
--
http://lists.linuxfromscratch.org/listinfo/lfs-support
FAQ: http://www.linuxfromscratch.org/blfs/faq.html
Unsubscribe: See the above information page
Do not top post on this list.
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?
A: Top-posting.
Q: What is the most annoying thing in e-mail?
http://en.wikipedia.org/wiki/Posting_style
