Libvolume incorrectly detects FAT32

Hello,
I'm having one silly partition, which has remnants of previous FAT32 filesystem on it, a LILO bootloader on it and ext2 filesystem spaning all over the partition.

The first 512B look exactly like a valid FAT partition header would, except the magic is "LILO".

I've looked at the sources for FAT detection, and I noted that when a FAT-like filesystem misses magic, it's still considered valid, since as comment says, some old floppies may lack it. I consider this point valid, however, this causes my partition to be detected, incorrectly, as FAT32.

My proposed solution is to compare magic if FAT32 is assumed, and if magic will not match for FAT32, consider it a failure; floppies that old that they lack any magic will most likely not be formatted in FAT32.

Trivial patch is attached, comments/complaints?:)

Quoted 17 lines · iSteve

Hello,
I'm having one silly partition, which has remnants of previous FAT32 filesystem on it, a LILO bootloader on it and ext2 filesystem spaning all over the partition.

The first 512B look exactly like a valid FAT partition header would, except the magic is "LILO".

I've looked at the sources for FAT detection, and I noted that when a FAT-like filesystem misses magic, it's still considered valid, since as comment says, some old floppies may lack it. I consider this point valid, however, this causes my partition to be detected, incorrectly, as FAT32.

My proposed solution is to compare magic if FAT32 is assumed, and if magic will not match for FAT32, consider it a failure; floppies that old that they lack any magic will most likely not be formatted in FAT32.

Quoted 1 line

Trivial patch is attached, comments/complaints?:)

Yes, I fear users of stupid digicams may get hit by this.
The Formating option on cheap digicams may sometimes be buggy implemented. I think you should instead do a
dd if=/dev/zero of=/dev/partition bsQ2 count=1
and reformat your partition.

Quoted 5 lines · Michael Buesch

Yes, I fear users of stupid digicams may get hit by this.
The Formating option on cheap digicams may sometimes be buggy implemented. I think you should instead do a
dd if=/dev/zero of=/dev/partition bsQ2 count=1
and reformat your partition.

Yes, just use a sane formatting program like everything which is libparted based. The usual Linux commandline formatters are horribly broken and I gave up to talk to the ignorant maintainers. Just zero-out the device at the beginning _and_ the end of the device if you reformat it with a new filesystem type. Or use a libparted or similar based program.

Thanks,
Kay

Quoted 7 lines · On Mon, 6 Mar 2006 23:37:20 +0100 Kay Sievers <[email protected]> wrote:

Quoted 5 lines · Michael Buesch

Yes, I fear users of stupid digicams may get hit by this.
The Formating option on cheap digicams may sometimes be buggy implemented. I think you should instead do a
dd if=/dev/zero of=/dev/partition bsQ2 count=1
and reformat your partition.

Quoted 8 lines

Yes, just use a sane formatting program like everything which is libparted based. The usual Linux commandline formatters are horribly broken and I gave up to talk to the ignorant maintainers. Just zero-out the device at the beginning _and_ the end of the device if you reformat it with a new filesystem type. Or use a libparted or similar based program.

Thanks,
Kay

I would have another suggestion then; how about a scoring system?

Probes would be given OPTIONAL scoring system, let's say in percents, eg. in case of fat, lack of magic would make it lose 20%; some other filesystem would return just 0 or 100% hit (ie. just like it is now). Then the probe with greatest score wins.

I understand this design is for working with semibroken partitions, and I fear it won't be welcome warmly; but I still got to propose it:)

I am willing to code it, I've written about ten probes for my own detection system until I noticed libvolumeid, but I want approval of design first.

Quoted 9 lines · iSteve

On Mon, 6 Mar 2006 23:37:20 +0100
Kay Sievers <[email protected]> wrote:
I would have another suggestion then; how about a scoring system?

Probes would be given OPTIONAL scoring system, let's say in percents, eg. in case of fat, lack of magic would make it lose 20%; some other filesystem would return just 0 or 100% hit (ie. just like it is now). Then the probe with greatest score wins.

Quoted 5 lines

I understand this design is for working with semibroken partitions, and I fear it won't be welcome warmly; but I still got to propose it:)

I am willing to code it, I've written about ten probes for my own detection system until I noticed libvolumeid, but I want approval of design first.

I'm all for fixing the stupid formatters. Seems the problems are really rare these days anyway, with Windows doing it right now and almost all distros use HAL today which has the same code and we don't get a lot of error reports anymore.

The right way I think, is to add more checks to the probing code, if the filesytem that is recognized is actually "valid". But that is impossible for some situations like mkswap (which is the silliest formatting tool on earth that I've ever seen) on top of fat. You can even mount the corrupted fat volume with the kernel and write to it.

So, I'm not sure if a score will give us something valuable. You basically push the problem to the user of the lib, which is a slippery road. But if we detect that we can't be sure what kind of filesystem we detect, we could throw an error and don't return any type, that would be at least a safe way to handle that and would fix the real problem, which are in almost all cases the too simple and broken Linux command line tools.

Thanks,
Kay

So the bottom line is, "no, this idea won't make it"? How about, then, let the user of the lib (ie. the app using the lib) specify whether it wants a full match or not?

I think you misunderstand.
The lib (app) is about properly _creating_ the partition with all magic. So basically every formating tool should zero out parts of the partition and _then_ write the new magic there. Obviously some tools fail to zero out the space. This is what needs fixing.
Pushing the problem to kernel-space, by implementing a faulty-by-design scoring system is IMHO not the way to go.

BTW:
I often saw this problem in the past, especially with swap and fat related partitions. So I now always use dd to zero out the first part of a partition before I reformat it.
That's a good workaround the broken format utilities.

To summerize: I don't think there is something broken in the kernel, so I don't think we should go and fix nonbroken code. Fix the broken code in userspace instead.