Quoting karl@??? (karl@???):

> And that works when the root filesystem is on a device with fixed major/
> minor number, e.g. /dev/sda2 /dev/hda1, and even /dev/md1 for md
> devices with the old (0.90) format superblock if they are auto-assebled.
> It doesn't work for devices with dynamic device numbers.

Yeah, I don't use Linux systems where that is an issue. I.e., USB mass
storage gets used only for ancillary removable storage that doesn't have
or need fixed mountpoints. (But if I did want my removable 2TB backup
drive in /etc/fstab, it could be referenced in /etc/fstab by disk label
or UUID. I merely prefer to avoid those ugly solutions, but they do
exist. Currently, it might be /dev/sdd1 or /dev/sde1 depending on
runtime events, but that's not an actual problem: I look in dmesg |
tail, see what it got recognised as, and mount / umount that.)

> Just be shure that e.g. /dev/sda2 points to the right disk if you have
> more than one, and ohh, don't compile in usb-storage and accidentally
> leave an usb-stick connected while rebooting, it could show up as
> /dev/sda.

The latter never, literally never, happens on any system of my
experience where the main storage is SAS or SATA (or PATA or old SCSI,
or ESDI or RLL or MFM...). I keep hearing people warn about that, but
it doesn't appear to happen. USB-attached mass storage gets assigned
device nodes later in the alphabet.

If I ever see it happen, I'll deal with that then, but it's been decades
now since people like Greg Kroah-Hartman (udev guy) and countless others
kept claiming this would be a problem, and I have yet to see it happen
at all, let alone on any system I admin.


> My impression is that the kernel is going in the direction where the
> old-school method is harder to do simply because the kernel devs
> doesn't care about it, but you can counter that by keeping things
> simple, if you wish.

As it happens, I do so wish.