Udev in b6_0: to be or not to be

Nathan Coulson conathan at conet.dyndns.org
Tue Jun 1 21:56:08 PDT 2004


> Nathan Coulson wrote:
>
>> I can yank the check for now, but still if you have a readwrite root
>> filesystem at bootup, you'll get a few garbage devices underneath the
>> ramfs you mounted on /dev.  I dont think that is acceptable in the
>> longterm.
>
> OK, I hadn't thought it through that way. In that case, I change my
> opinion and agree that the kernel patch is the way to go until such time
> as we can build an initramfs that does this stuff before the root
> filesystem is mounted.
>
> Also, this begs the question of why the bootscripts want to have the
> root filesystem mounted read-only in the first place... in the past this
> was because it was best to always run an fsck on the root filesystem,
> but with people using journalling filesystems now (in my case XFS) that
> is becoming less useful and more of an annoyance than anything else.

technically, since we dont use the harddrive, it should work in ro, or rw
mode.  [I have not tested to see what mount / -o remount,rw does, but I
assume it wont give anyone any grief (part of mountfs)].

In this one case, someone somehow got a hotplug event occuring before
rcsysinit.d/S20udev.  It calls /sbin/hotplug like it always does, which
created /dev/.udev.tdb, since it can write to the harddrive.  In the case
that the filesystem is readonly, it probably wont create /dev/.udev.tdb. 
(Info everyone already has, just saying this is the exception to the above
mention).

checkfs, I'm going to assume causes problems on mounted readwrite
filesystems, but I dont know any specifics.  [just that everyone else says
that].

I dont think anything else would cause us grief before S40mountfs...



More information about the lfs-dev mailing list