Re: implicit mounting of the root partition for rw

[U.Mutlu]

Jordan Uggla wrote, On 2011-11-11 01:31:
> On Thu, Nov 10, 2011 at 2:31 PM, U.Mutlu<address@hidden>  wrote:
> > grub2 on my system (Debian 6) seems to implicitly mount the
> > root partition for rw. But in /etc/fstab it is mounted again.
> > Isn't that additional mounting not harmful for the HD ?
> > Can one safely disable the mount for / in /etc/fstab ?
> Grub needs to be able to read files from /boot/ to be able to load the
> kernel. This is unavoidable, and does absolutely no harm to the hard
> drive. In addition, with the single possible exception of writing to
> /boot/grub/grubenv (not done by default in Debian's grub) grub never
> writes to the hard drive at all. Once grub loads the linux kernel and
> jumps to it, almost nothing that grub has done is applicable to the OS
> being booted. Grub doesn't, and can't, "mount" the root filesystem for
> the OS it loads, the kernel and initrd scripts need to do that on
> their own from the ground up after grub is finished with its part and
> is long gone. This is simply how booting works. In addition to that,
> "mounting" is not something that affects hardware, you can read from
> and write to a device through /dev/ without mounting it. What mounting
> does is tell the kernel to interpret the bits that it is already able
> to read from the hardware so that you can work with the abstraction
> that is files and paths.
>
> In short, nothing can or should be done differently, your hardware is fine.

My hardware (HD) unfortunately isn't fine. Here's the story:

I have a HD with 4 primary partitions. On the first partition
the OS was installed (ie. the root partition incl. /root and boot).
Recently SMART reported HD errors in that partition.
A badblocks test revealed many bad blocks on that partition
(the SATA-controller gets exception errors and these get written
to syslog):
  Checking blocks 0 to 30274460
  Checking for bad blocks (read-only test): done
  Pass completed, 1175 bad blocks found.

But the funny thing is that all the 3 other partitions of
that same HD have not even a single bad block!
So I wonder how this is possible, even mathematically...

Using ddrescue I then copied the partition to one of the other partitions.
Fortunately less than 2 MB data loss, mainly some of the log files
(in my case the log files are frequently written with some debug outputs).
So, my system is up again, but I wonder why that happened at all,
and why it happened so? :-)  I mean, one would expect that the
bad blocks should normally be distributed among all partitions,
but in this case only the root partition was affected...

I have only this speculative explanation: since the root partition
is (somehow) implicitly mounted during kernel startup and again
in /etc/fstab then maybe that partition 'aged' much earlier than
the other partitions, due to the double mounting... (?)
If someone has a better theory let me know please.

BTW, I now disabled the (unneccessary?) mounting of the root partition
in /etc/fstab, and the system still works normal.

BTW2, using "mke2fs -v -c -t ext3 -j /dev/sda1" one can do a "low-level"-format
and exclude such bad blocks for allocation. I would recommend to format
a partition only that way. It takes much longer, but is IMO much safer.