fsck.vfat detects errors on pristine mkfs.vfat filesystem

[Ralph Corderoy]

Ubuntu 8.04, dosfstools 2.11-2.3ubuntu1.

mkfs.vfat creates a filesystem which initially passes fsck.vfat's
checks.

    $ dd if=/dev/zero of=fat.fs bs=1k count=1k
    1024+0 records in
    1024+0 records out
    1048576 bytes (1.0 MB) copied, 0.0404714 s, 25.9 MB/s
    $
    $ mkfs -t vfat -F 32 -v fat.fs
    mkfs.vfat 2.11 (12 Mar 2005)
    fat.fs has 64 heads and 32 sectors per track,
    logical sector size is 512,
    using 0xf8 media descriptor, with 2048 sectors;
    file system has 2 32-bit FATs and 1 sector per cluster.
    FAT size is 16 sectors, and provides 1984 clusters.
    Volume ID is 498c32bb, no volume label.
    $
    $ fsck.vfat -n fat.fs
    dosfsck 2.11, 12 Mar 2005, FAT32, LFN
    fat.fs: 0 files, 1/1984 clusters
    $

However, then put a file on the filesystem and fsck.vfat thinks there's
a problem.

    $ mkdir mnt
    $ sudo mount -o loop fat.fs mnt
    $ sudo dd if=/dev/urandom of=mnt/foo bs=1k count=100
    100+0 records in
    100+0 records out
    102400 bytes (102 kB) copied, 0.0760562 s, 1.3 MB/s
    $ sudo umount mnt
    $
    $ fsck.vfat -n fat.fs
    dosfsck 2.11, 12 Mar 2005, FAT32, LFN
    Free cluster summary wrong (1983 vs. really 1783)
      Auto-correcting.
    Leaving file system unchanged.
    fat.fs: 1 files, 201/1984 clusters
    $

At least one of the kernel's filesystem code or dosfstool's fsck is
wrong. For a LTS release I don't think it's acceptable. I think there
was probably nothing wrong with my operational FS, but I let fsck.vfat
try and fix it, which didn't go well, and ended up loosing it. (I had
backups.)

Trying the above with 8.10's dosfstools 2.11-5 doesn't exhibit the same
problem; it appears fixed.

[Onno Benschop (onno-itmaze)]

I am unable to test this on a hardy machine at this time, but can confirm that it has the same behaviour on gutsy and that it has been fixed on intrepid.

Gutsy: 2.11-2.2ubuntu1 - confirmed
Hardy: 2.11-2.3ubuntu1 - confirmed
Intrepid: 2.11-5 - does not have a problem

[Onno Benschop (onno-itmaze)]

Hmm, I did in fact test it on a hardy machine, but forgot to edit the message before hitting send :)