Re: udev still stinks

"Thomas \"fake\" Jakobi" <fake-o+BlaNWM+K2Y/[email protected]> Tue, 14 Mar 2006 21:43:07 +0100
Newsgroups gmane.linux.distributions.rock.devel
Message-ID <[email protected]>
hi,

Tobias Hintze schrieb:
> On Tue, Mar 14, 2006 at 11:11:33AM +0100, Benjamin Schieder wrote:
> [...]
>> > > 
>> > > mount --bind works for me.
>> > 
>> > i think that's by far the worst solution.
>> > 
>> > quoting from udev/README:
>> > =================
>> >   - It is recommended to use the /lib/udev/devices directory to place
>> >     device nodes and symlinks in, which are copied to /dev at every boot.
>> >     That way, nodes for broken subsystems or devices which can't be
>> >     detected automatically by the kernel, will always be available.
>> > =================
>> > 
>> > that should always be the way how we create or /dev/fd -> /proc/self/fd
>> > symlink.
>> 
>> Yes, but that's not the problem we're talking about here.
>> The problem here is the chroot after installation. So:
>> Boot CD -> partition harddisk -> install gems -> a bit of magic ->
>> -> chroot into installed system -> some more magic
>>    ^^^
>> This has nothing to do with /lib/udev/devices
>> 
> 
> oh yes it has!
> chrooting into that installed system is like booting into the system.
> so instead of bind-mounting one should mount tmpfs and copy from
> /lib/udev/devices.
> that way you get exactly what you would get when booting.

the bind mount is an alternative to re-replicating udev events for newly
loaded modules or plugged-in hardware which won't happen just by doing a
 chroot or killing and re-starting udevd.

i chose a bind-mount instead of re-replicating the udev events because
the contents of /sys can't change when chrooting.

/lib/udev/devices is for static device entries that a udev event won't
happen for ever.

greets,

fake