[dm-crypt] Filesystem unreadabe after resizing LUKS

Arno Wagner arno at wagner.name
Fri Jan 31 04:17:05 CET 2014


On Fri, Jan 31, 2014 at 02:36:33 CET, Robert Nichols wrote:
> On 01/30/2014 03:01 PM, Arno Wagner wrote:
> >On Thu, Jan 30, 2014 at 16:24:27 CET, Pawel Chojnacki wrote:
> >>Hello,
> >>
> >>Two days ago I tried to shrink my LUKS-encrypted /dev/sdb5 partition from
> >>119 to 110 GB according to
> >>http://ubuntuforums.org/showthread.php?t=726724. Each step worked
> >>properly, but in the end I'm not able to mount the
> >>partition. I wanted to ask if you have any idea what may have caused that
> >>and how to fix it:
> >>
> >># cryptsetup luksOpen /dev/sdb5 neuro
> >>
> >>accepts the proper password, doesn't recognize any other, and fails to
> >>mount the partition inside:
> >>
> >># mount /dev/mapper/neuro /mnt/
> >>mount: unknown filesystem type 'LVM2_member'
> >
> >This means LUKS is fine, just the resize failed.
> >I should pount out the clear "It should go without
> >saying, resizing your crypt may result in data loss
> >Be sure to BACK UP your data first." given in that posting.
> 
> The thing that bugs me about the procedure in that thread is that it
> includes a totally unnecessary and complex ("I had to do this by trial
> and error") step of "cryptsetup ... resize". Nowhere in the LUKS header
> is there any field that holds the size of the container. Each time the
> LUKS container is opened it takes on the size of whatever is holding it
> (physical partition, LVM logical volume, dmsetup mapping, ... whatever).
> 
> The "resize" operation in cryptsetup is useful only in the fairly rare
> circumstance that you actually want to _use_ an encrypted area _while_
> it is smaller than whatever is holding it. That is similar to having
> a filesystem that is smaller than its partition, except that the
> filesystem _does_ have persistent knowledge of its size whereas a LUKS
> container does _not_.

Indeed. I missed that. In fact, I was unaware  that cryptsetup
even had a resize operation (well, I must have been aware at 
some time, because I revised the man-page, but apparently 
forgot about it immediately), possibly because I know that there
is no size-field anywhere in the header and that it is completely
unnecessary to have one.

I guess there is some massive misunderstanding somewhere in that
set of instructions....

Arno

--- 
Arno Wagner,     Dr. sc. techn., Dipl. Inform.,    Email: arno at wagner.name
GnuPG: ID: CB5D9718  FP: 12D6 C03B 1B30 33BB 13CF  B774 E35C 5FA1 CB5D 9718
----
A good decision is based on knowledge and not on numbers. -  Plato


More information about the dm-crypt mailing list