| 1 |
On 19/08/2013 11:31, pk wrote: |
| 2 |
> On 2013-08-19 00:49, Dale wrote: |
| 3 |
> |
| 4 |
>> Picking random message sort of. Isn't eudev still going to support a |
| 5 |
>> separate /usr? That is my understanding. If eudev is not then I may |
| 6 |
>> have to reconsider some things myself here. |
| 7 |
> |
| 8 |
> Yes, that is my understanding as well. But the "decision" to not support |
| 9 |
> a separate /usr lies higher up in the system hierarchy (as I understand |
| 10 |
> it). Gentoo as a system will not support a separate /usr if we are to |
| 11 |
> believe the conversation (I haven't seen any official notice of this |
| 12 |
> though). That is the sad part. The problem I have, as an engineer, is |
| 13 |
> that "everybody" says that a separate /usr is broken, that sysvinit is |
| 14 |
> broken without explaining why. In order to fix a problem you need to |
| 15 |
> know what is broken... The people who claims the brokenness are, imo, |
| 16 |
> hand waving and they've managed to convince higher uppers in the Gentoo |
| 17 |
> infrastructure (as it seems). I guess if you repeat something often |
| 18 |
> enough it becomes a "truth" or said person(s) just agrees to stop the |
| 19 |
> nagging. |
| 20 |
|
| 21 |
|
| 22 |
It's not that separate /usr is broken - it's not. |
| 23 |
|
| 24 |
The issue is a separate /usr without an initramfs. And the issue ONLY |
| 25 |
occurs at early-boot time. |
| 26 |
|
| 27 |
The problem is that with modern hardware much code that was |
| 28 |
traditionally stored in /usr may be needed early in the boot sequence, |
| 29 |
before /usr is mounted. The obvious case is firmware and drivers, and |
| 30 |
the usual example cited is bluetooth keyboards. If you need keyboard |
| 31 |
input at this time, you need to have the bluetooth daemon running, which |
| 32 |
is on /usr, which is not mounted. |
| 33 |
|
| 34 |
The solution is to use an initramfs, and on a technical level it's not |
| 35 |
any different to needing a way to get the ext4 module off disk so you |
| 36 |
can mount /. |
| 37 |
|
| 38 |
Some may argue that bluetooth keyboards are a rarity and that's tough. |
| 39 |
Well, there's Macbook hardware, and phones which have soft keyboards. |
| 40 |
But many scenarios could exist, all due to the fact that hot-pluggable |
| 41 |
hardware can in theory run any arbitrary code to get itself up and |
| 42 |
running, and if that code is on a volume that is not mounted... The |
| 43 |
solution is obvious - all that code should be on / somewhere, or should |
| 44 |
be mountable using an initramfs. |
| 45 |
|
| 46 |
Do you see that although you and I can deal with this with relative |
| 47 |
ease, Aunt Tillie probably couldn't and the junior sysadmins I have to |
| 48 |
deal with certainly can't? |
| 49 |
|
| 50 |
Personally, I think that splitting / and /usr is a daft idea: |
| 51 |
|
| 52 |
a. I have multi-TB hard disks, completely unlike the 5M monsters that |
| 53 |
Thomson had to deal with in the 70s |
| 54 |
b. I haven't had /usr break on me during boot requiring busybox in |
| 55 |
maintenance mode for at least 5 years. Every startup failure in that |
| 56 |
time required a rescue cd anyway, and I always have one of those handy |
| 57 |
c. it IS useful for terminal servers, but those tend to have experienced |
| 58 |
sysadmins, and they really should be OK with an initramfs (or their |
| 59 |
vendor should ship one) |
| 60 |
|
| 61 |
I'm often at the front of the Lennart-bashing parade, and what he says |
| 62 |
often makes sense but only in his narrow view of the world, but in |
| 63 |
*this* case, I can't help but admit he does have a point. |
| 64 |
|
| 65 |
-- |
| 66 |
Alan McKinnon |
| 67 |
alan.mckinnon@×××××.com |
Archives