[parted-devel] "Error: Can't have a partition outside the disk!"

[Jim Meyering]

Richard W.M. Jones wrote:
> If you take a virtual disk and truncate it, then likely some
> partitions in the partition table will be off the end of the disk.
>
> However parted refuses to start up in this case.  It says:
>
>   Error: Can't have a partition outside the disk!
>
> even if you are trying to do something sensible such as deleting a
> partition which is outside the disk.
>
> I can't really see a "good" way to solve this, since it happens in the
> depths of the constraints/geometry code.  The attached patch just
> deletes the error check, on the basis that parted should confine
> itself to partition tables and not worry about the size of the disk.
> Discuss ...

Rich,

I tried your patch (but retained the warning),
running it against a sample "device" I shrank by one byte...
created like this:

  $ dev=dev-file
  $ dd if=/dev/null of=$dev bs=1 seek=2GB
  0+0 records in
  0+0 records out
  0 bytes (0 B) copied, 0.000256681 s, 0.0 kB/s
  $
  [Exit 1]
  $ parted -s -- $dev mklabel msdos mkpart primary fat32 64s -1s u s p
  Warning: The resulting partition is not properly aligned for best performance.
  Model:  (file)
  Disk /c/parted/parted/dev-file: 3906250s
  Sector size (logical/physical): 512B/512B
  Partition Table: msdos

  Number  Start  End       Size      Type     File system  Flags
   1      64s    3906249s  3906186s  primary               lba

  $ truncate -s -1 $dev
  $ ./parted -s $dev u s p
  Warning: Can't have a partition outside the disk!
  Warning: Can't have a partition outside the disk!
  Warning: Can't have a partition outside the disk!
  ...
  Assertion (part->geom.end < disk->dev->length) at disk.c:1856 in function _partition_check_basic_sanity() failed.


As you can see, this assumption that partitions
fit within the bounds of the device is at least one
assertion deeper...

Commenting out that assertion, we get a little further:

  $ ./parted -s dev-file u s p
  Warning: Can't have a partition outside the disk!
  Warning: Can't have a partition outside the disk!
  Warning: Can't have a partition outside the disk!
  Error: Can't have overlapping partitions.
  [Exit 1]

The bottom line is that the partition-reading
and -validation code is too tightly coupled.
I'd rather be able to read in the entire partition
table description, and then sanitize it in a separate, optional phase.
But no rewrite today... I barely have time to tend to the occasional
Parted bug report.