1 |
On Fri, 14 Dec 2012 08:53:35 -0800 |
2 |
Mark Knecht <markknecht@×××××.com> wrote: |
3 |
|
4 |
> I guess the other question that's lurking here for me is why do you |
5 |
> have /usr on a separate partition? What's the usage model that drives |
6 |
> a person to do that? The most I've ever done is move /usr/portage and |
7 |
> /usr/src to other places. My /usr never has all that much in it beyond |
8 |
> those two directories, along with maybe /usr/share. Would it not be |
9 |
> easier for you in the long run to move /usr back to / and not have to |
10 |
> deal with this question at all? |
11 |
|
12 |
It should be moving in the other direction for stability reasons and |
13 |
busybox is no full answer. |
14 |
|
15 |
On OpenBSD which has the benefit of userland being part of it. All the |
16 |
critical single user binaries are in root and built statically as much |
17 |
as possible, maximising system reliability no matter the custom |
18 |
requirements or packages. |
19 |
|
20 |
The way I have it on OpenBSD |
21 |
|
22 |
/ ro |
23 |
|
24 |
100 megabytes and I never need to fsck and can reliably fix all |
25 |
but the most likely problem and snapshot quickly, though there is no |
26 |
need as the kernel is rock solid. |
27 |
|
28 |
/usr ro,nodev |
29 |
~600 megabytes that I almost never need to fsck even when I pull the |
30 |
plug |
31 |
|
32 |
/usr/local ro,nodev,nosuid |
33 |
All installed packages go here and I can give users the ability to only |
34 |
mount writeable this location. There are other plusses I won't bother |
35 |
going into. |
36 |
|
37 |
|
38 |
All the BSDs and debian stable (old and initramfs) still get's this |
39 |
right with debian suggesting a seperate /usr during install in |
40 |
compliance with the filesystem hiearchical standard and the upcoming |
41 |
draft/version 3, which states the real technical and uptime benefits of |
42 |
a seperate /usr. |
43 |
|
44 |
https://wiki.linuxfoundation.org/en/FHS |
45 |
|
46 |
Unfortunately stability and security often only get's noticed and |
47 |
chosen over other function when it's completely obliterated and has |
48 |
stopped functioning alltogether. |
49 |
|
50 |
When hard worked (including rusty russel) documents like this get |
51 |
ignored when freedesktop.org is given so much credence even though |
52 |
freedesktop.org is actually simply stating opinion without having |
53 |
debate/comments on it's site and in contrast a combined root/usr has no |
54 |
technical benefit not addressed elsewhere (grub etc..) and when the |
55 |
issues in userland are far from insurmountable it is quite worrying and |
56 |
I am grateful to those who have stood up against this and the trend |
57 |
of added complexity into pid1/systemd and early boot. |
58 |
|
59 |
What is also worrying is the recent trends of the kits, udisks |
60 |
dropping features for months to get multiseat and dbus getting |
61 |
everywhere like Windows and RPC. |
62 |
|
63 |
I can take spread out documentation compared to OpenBSD but some of |
64 |
these issues are quite rediculous, I just wish OpenBSD had more devs for |
65 |
KMS and stable updates as it is perhaps due to being a smaller project |
66 |
involving both core userland and kernel and with hard fast goals, far |
67 |
better managed. |