1 |
On Thu, 2010-07-01 at 18:06 +0200, Nils Larsson wrote: |
2 |
> tor 2010-07-01 klockan 08:49 -0700 skrev Bill Longman: |
3 |
> > On 07/01/2010 08:44 AM, SpaceCake wrote: |
4 |
> > > So, it solves the first problem, identifiying the device, but how can I |
5 |
> > > tell to udev to use always /dev/sds (for example) for this device? |
6 |
> |
7 |
> You need to have the udev rule or the script that it runs look at |
8 |
> something specific(the swaplabel for instance). |
9 |
> |
10 |
> > > I'm thinking how can I instruct udev to turn off swap when the device is |
11 |
> > > removed, but this is another story :) |
12 |
> |
13 |
> I tried doing exactly what you're doing now awhile ago and this is where |
14 |
> I got stuck, swapoff needs the deivce node(path) to still exist, it |
15 |
> can't disable swap without it. I could never get swapoff to run before |
16 |
> udev removed the device node, so I ended up with the system thinking(or |
17 |
> at least reporting) that it had loads more swap than it actually did. |
18 |
|
19 |
This is a bad thing to do. If you have pages swapped out to the device |
20 |
and you remove the device before putting those pages elsewhere then you |
21 |
have effectively hosed your system. If it doesn't fail immediately then |
22 |
as soon as the kernel tries to swap in those pages and finds out the |
23 |
device it's on can't be accessed then you are in for a world of pain. |
24 |
|
25 |
I guess the deeper question (although entirely rhetorical AFAIC) is why |
26 |
would someone want to swap out to a removable device? |