Re: should an end user stick to a kernel with an initrd?
On Tue, Oct 08, 2013 at 11:08:50PM +0100, Roger Leigh wrote:
> On Fri, Sep 27, 2013 at 10:28:01PM +0300, Regid Ichira wrote:
> > On Fri, Fri, 27 Sep 2013 13:34:56 -0400, Tom H wrote:
> > > On Fri, Sep 27, 2013 at 3:12 AM, Ralf Mardorf wrote:
> > > > On Thu, 2013-09-26 at 19:07 -0400, Stephen Powell wrote:
> > > >>
> > > >> Traditional device names, such as /dev/sda, /dev/sdb,
> > > >> (and therefore the partitions on those devices, such
> > > >> as /dev/sda1, /dev/sdb1, etc.) are not assigned in a predictable
> > > >> manner anymore. This device name assignment can change from one boot
> > > >> to the next.
> > > >
> > > > This never happened on my machine.
> > >
> > > This won't happen if you have just one disk. ;)
> > >
> > > On a more serious note, do you really think that all the people
> > > maintaining distributions thought "using sdX is far too simple and
> > > easy, let's start using human-non-parsable UUIDs?!"
> >
> > 1. Saying traditional disks names not siigned in a predictable manner
> > seem to contradict the fact that one can write
> > root=/dev/hdd3
>
> You can certainly write that into the fstab, but that won't guarantee
> that the device will be "hdd3"; it might be hdc3, hde3 etc. depending
> upon the presence of other devices and the initialisation order.
For me, the enumeration of devices is guaranteeted. As already
discussed in this thread, with older boards, the enumeration of IDE
and SCSI devices is completely determined by their jumpers and wiring.
I think that does not change with boards that won't boot usb devices.
I can't tell about newer boards.
>
> > in the kernel command line, such as in lilo.
> > 2. I have 2 disks. It never happened to me.
>
> Try plugging in a USB storage device during early boot. On some
> systems this might end up initialised before the physical HDDs
> and then all the hard disks will be renamed and the fstab entries
> will be broken. On most systems the SATA drives will be initialised
> first, but this isn't guaranteed--a lot of this is asynchronous now;
> what if the HDD takes longer to spin up than normal, so gets
> registered later? You want guaranteed reliability, and UUIDs/LABELs
> give you that; the kernel device names might /seem/ stable on a
> given system, but that's really only a result of circumstance, not
> by design.
>
It seems newer hardware is much more problematic in this sense. I
think MS ovecomes this difficulty by somehow attaching a signature for
each device. I don't have the details, don't know the pros and cons.
> > 4. I think that the LABEL mechanism of /etc/fstab is different,
> > predated, and more rigid, from that of a UUID. Again, it seem to
> > me supported by some of the comments in
> > https://lwn.net/Articles/331818/.
>
> Both are handled by udev today, to give you /dev/disk/by-label
> and /dev/disk/by-uuid. I don't think that labels are handled
> specially by the kernel in addition to that, since it can be
> potentially quite complex and filesystem-specific, but I could
> be wrong. Maybe they were in the past, or handled specially
> prior to udev?
>
I don't know either.
In general, I understand that the initrd issue is much more complex
today then it was a few years ago.
Reply to: