Re: filesystem detection (was: Re: mkntfs -Q and libparted)

[Szakacsits Szabolcs]

        [ CC'd to the most relevant mailing lists ]
Hello,

On Thu, 25 Nov 2004, B.Hakvoort wrote:

> i did some testing and can confirm the error. It only occurs when you
> create a ntfs filesystem on a partition which contains a reiserfs
> filesystem.

Good caught! 

Actually this is a know problem and every tool, kernel has it: if there
are several filesystem signatures on a partition then the detection
depends on the order how the filesystem is being tried, or if they
continue to try and there are several fs signatures then they fail to
detect any because the result is ambiguous.

> I _guess_ it's a parted bug.

For parted it's impossible (or it would be quite difficult) to decide what
the real filesystem is if there are several signatures detected. Probably
this is the reason it reports "unknown" -- IMHO correctly. Andrew can
confirm this (or the source).

The solution is, _all_ mkfs.* utility must wipe _all_ other filesystem
signatures on the partition. This is quite though and basically quite
impossible to do efficiently _but_ even _trying_ could improve the
situation a lot and it wouldn't be hard to do.

In short, mkntfs should look for and wipe at least the most common
filesystem signatures. This is what's happening, implicitely, when 
the -Q option isn't provided that's why it worked before. 

        Szaka

> Ok, steps to reproduce:
> - create partition with fdisk (5000M)
> - create reiserfs filesystem in this partition (mkreiserfs /dev/hdc1)
> - create ntfs filesystem in this partition (mkntfs /dev/hdc1 -Q)
> - according to parted the filesystem is unknown, but according to
> ntfsresize --info it's OK
> (Dropping the '-Q' 'fixes' this)
> 
> So it seems you were right and the filesystem is correctly created, but
> somehow parted cannot detect it.