Partial update of bootscripts

Bryan Kadzban bryan at kadzban.is-a-geek.net
Tue Jul 12 20:09:30 PDT 2011


Bruce Dubbs wrote:
> Bryan Kadzban wrote:
>> Commit the udev_retry update as a separate revision in svn.  It's
>> pretty independent of the other changes being proposed.  Maybe that
>> was the plan already though.
> 
> I wasn't planning that, but I suppose I could.  However, I think I'll
>  have a complete LFS bootscript rewrite ready in a week or two.

Either way.

>> For CD burners, I'd *expect* (but don't know for sure) that it
>> would be best these days to use the /dev/disk/by-* symlinks anyway,
>> so those may become less of a problem.  But it's been a long time
>> since I tried to burn a CD; it's possible that the burning software
>> these days requires a device node, not a symlink.  Or something
>> else equally dumb.
> 
> Symlinks like /dev/cdrom->sr0 work fine for me.

Good that they work today.  But that particular symlink won't exist when
(if?) the udev developers rip out the write_cd_rules and write_net_rules
stuff.

(Which are, AFAIK, the only two things that require a writable rules
directory at early boot time, so they're the only two things that need
to be copied out of udev's tmpfs, wherever that ends up being put.)

OTOH if /dev/cdrom works today, it's likely that /dev/disk/by-* will
work as well, meaning that at least write_cd_rules is no longer strictly
needed.

>> If alsactl is in /usr/sbin, and /usr is not on the root fs, then
>> the controlC* rule that we have will fail, since it won't be able
>> to run "alsactl restore".  Similarly for any other rules that get
>> installed that require binaries or other configuration (...usb-db,
>> for instance, although we don't install that) in /usr.
> 
> Got point.  I've made the point before that there needs to be a
> writable analog to /run that is a top level directory, not a
> mountpoint, that is persistent across boots for this type of thing.
> Until that's done, I don't see a way to make / read only.

Well, the writability doesn't actually matter for these rules.  The only
thing that matters is whether the executable (or data file, or whatever)
is available when udev runs the rule.  If not, the RUN will fail, and
the rule will need a retry.

(Of course, / will still need to be writable when installing software no
matter how the system is set up, otherwise /bin, /sbin, /lib, /lib64,
and friends can't be changed.  That could be made a flippable switch
though, I suppose.)

But yes, until these executables (or data files) find a way to be added
to the rootfs, or be created by something before udev runs, we can't get
rid of the retry script.  (In other words, until no RUN can possibly
fail at coldplug time.)

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 261 bytes
Desc: OpenPGP digital signature
URL: <http://lists.linuxfromscratch.org/pipermail/lfs-dev/attachments/20110712/9f9b38c5/attachment.sig>


More information about the lfs-dev mailing list