| 1 |
On Mon, 24 Dec 2012 16:06:27 +0800 |
| 2 |
Mark David Dumlao <madumlao@×××××.com> wrote: |
| 3 |
|
| 4 |
> On Mon, Dec 24, 2012 at 9:27 AM, Walter Dnes <waltdnes@××××××××.org> |
| 5 |
> wrote: |
| 6 |
> > On Sun, Dec 23, 2012 at 08:39:41PM +0000, Neil Bothwick wrote |
| 7 |
> > |
| 8 |
> >> You are only considering the case of /usr being on a plain hard |
| 9 |
> >> disk partition, what if it in on an LVM volume, or encrypted (or |
| 10 |
> >> both) of mounted over the network? All of these require something |
| 11 |
> >> to be run before they can be mounted, and if that cannot be run |
| 12 |
> >> until udev has started, we have been painted into a corner. |
| 13 |
> > |
| 14 |
> > I agree that there will always be a small number of corner-cases |
| 15 |
> > where an initr* is required. What annoys me, and probably a lot of |
| 16 |
> > other people, is the-dog-in-the-manger attitude |
| 17 |
> > http://en.wikipedia.org/wiki/The_Dog_in_the_Manger where some people |
| 18 |
> > seem to say "If my weirdo, corner-case system can't boot a |
| 19 |
> > separate /usr without an initr* then, by-golly, I'll see to it that |
| 20 |
> > *NOBODY* can boot a separate /usr without an initr*". |
| 21 |
> |
| 22 |
> This is misleading in two ways. |
| 23 |
> |
| 24 |
> 1) You're talking as if having a functionally merged /usr and / system |
| 25 |
> (i.e., many programs needed by the sysad to fix a non-booting system |
| 26 |
> are in /usr, and programs in /usr will break if /usr is not in sync |
| 27 |
> with /) is a weirdo corner case. It is NOT. It is very likely how the |
| 28 |
> vast majority of Linux systems on the planet work. Separate /usr is |
| 29 |
> itself the weirdo corner case. It was in fact a weirdo corner case |
| 30 |
> since day 1. |
| 31 |
> 2) You're talking as if Lennart or whoever is breaking into your |
| 32 |
> systems and actively preventing you from customizing it to boot a |
| 33 |
> separate /usr. If this is the case you _really_ need to change your |
| 34 |
> ssh keys, they wiped that vulnerability a couple years ago. |
| 35 |
> |
| 36 |
> Nobody's preventing you from building a custom system that cleanly |
| 37 |
> separates / and /usr. But hey, don't pretend that even Gentoo does it |
| 38 |
> correctly. Besides the equery tests in this thread, I've never |
| 39 |
> personally confirmed that any other distro does - and Fedora cleanly |
| 40 |
> admits that they don't. |
| 41 |
> |
| 42 |
|
| 43 |
The ultimate weird corner case is having a separate / and /usr so the |
| 44 |
either of these two thing can happen: |
| 45 |
|
| 46 |
a. there's enough $STUFF in / to fix large-scale errors |
| 47 |
b. there's enough $STUFF in / to mount /usr ro over NFS (as in for a |
| 48 |
terminal server) |
| 49 |
|
| 50 |
|
| 51 |
a. is fixed by just using what all sysadmins use anyway - a proper |
| 52 |
rescue disk built for that specific purposes (instead of trying to get |
| 53 |
half a system to do it for you) |
| 54 |
|
| 55 |
b. is resolved by mounting /, not /usr. It's a terminal server, so the |
| 56 |
only thing not under full user control is ~. There is no point in |
| 57 |
having half the system local and the rest of it remote, just mount |
| 58 |
everything remotely. And if it's a terminal server, it will have a real |
| 59 |
sysadmin, someone who can maintain the code needed to get NFS up at |
| 60 |
boot time. If the mount of / breaks, the solution is a. |
| 61 |
|
| 62 |
Are there any other cases, apart from emotional attachment based on |
| 63 |
inertia, where a separate / and /usr are desirable? As I see it, there |
| 64 |
is only the system, and it is an atomic unit. |
| 65 |
|
| 66 |
-- |
| 67 |
Alan McKinnon |
| 68 |
alan.mckinnon@×××××.com |
Archives