[systemd-devel] systemd-repart HOWTO + demo code

[Thayne Harbaugh]

Greetings - apologies if this isn't the proper list - please redirect
me.

I'm writing a systemd-repart HOWTO with demo code for various use
cases:

  https://github.com/plastikos/repart_howto

The HOWTO is a typical document while the demos are minimal cases
built with mkosi and available for growing into larger solutions.

Initially it has two use-case demos:

  1. Creating a new partition (xdata) on first boot

  2. Creating a new partition (xdata) with encryption on first boot

Currently #1 works.  It's simple.  I still need to create more details
around growing partitions and other ideas.

The encryption case #2, however, does not work and I'm looking for
some assistance.  While there is significant documentation around the
systemd ecosystem of tools I feel like I have possibly missed some
detail or may not be understanding how to synthesize the information
into a working model.

The encryption demo is identical to case #1 of adding a partition with
the following changes:

  1. The /usr/lib/repart.d/30-xdata.conf file adds the following:

       Encrypt=key-file
       EncryptedVolume=dc-xdata:/run/fscrypt.sock:luks,discard

  2. a /usr/lib/systemd/system/fscrypt.socket file is added along with
     a fscrypt at .service file (see the project source, please).

  3. A /usr/bin/getkey file is added for fscrypt at .service to call.
     getkey provides a simplistic way of producing an encryption key
     for *demonstration* purposes *ONLY*.

  4. A /var/lib/key.bin is added as a binary encryption key that
     getkey retrieves - for *demonstration* purposes *ONLY*.

When built into a disk image by mkosi and started with qemu it fails
to create the new xdata partition.  I see quite a few journal messages
from systemd-repart and related udev events.  I have verified a few
things:

  1. /run/fscrypt.sock does get created

  2. getkey does emit the encryption key to stdout

  3. systemd-repart is started and initiates creating a partition with
     LUKS

There are two notable details that cause me concern and might be where
the problem lies:

  1. There is no evidence that getkey is invoked even when I
     instrument it to generate side-effects.

  2. There are messages in the journal that indicate that
     systemd-repart never exits and that it might be in deadlock waiting
     for child events.

The messages around the possible deadlock of #2 above are the
following:

  Oct 19 12:08:48.510843 localhost systemd-repart[276]: loop0: Acquired exclusive lock.
  Oct 19 12:08:48.512130 localhost systemd-repart[276]: Successfully acquired /dev/loop0, devno=7:0, nr=0, diskseq=11
  Oct 19 12:08:48.516941 localhost systemd-repart[276]: Encrypting future partition 2...
  Oct 19 12:08:48.517077 localhost systemd-repart[276]: Allocating context for crypt device /dev/loop0.
  <SNIP/>
  Oct 19 12:08:49.114118 localhost (udev-worker)[299]: loop0: Processing device (SEQNUM=1996, ACTION=change)
  Oct 19 12:08:49.114972 localhost (udev-worker)[299]: loop0: Failed to flock(/dev/loop0): Resource temporarily unavailable
  Oct 19 12:08:49.114979 localhost (udev-worker)[299]: loop0: Block device is currently locked, requeueing the event.
  <SNIP/>
  Oct 19 12:08:49.328308 localhost (udev-worker)[296]: loop0: Processing device (SEQNUM=1996, ACTION=change)
  Oct 19 12:08:49.328345 localhost (udev-worker)[296]: loop0: Failed to flock(/dev/loop0): Resource temporarily unavailable
  Oct 19 12:08:49.328352 localhost (udev-worker)[296]: loop0: Block device is currently locked, requeueing the event.
  <SNIP/>

Those last three messages about failing to lock repeat until QEMU is
shut down.

There are quite a few additional diagnostic details and logs that may
be of interest but it might be easier to clone the repart_howto
project from github and run it:

  >$ ./demo crypt

Once it executes all of the mkosi configuration files and disk image
are in the build-crypt sub-directory.  All relevant files specific to
partition creation and encryption are in build-crypt/mkosi.extra -
there are not many since I'm trying to keep the demo as simple as
possible.  All relevant journal events and state information can be
extracted from build-crypt/image.raw.

Thanks for either directing me to a more appropriate help resource
and/or providing direct assistance.


P.S. If you want to run demo #1 of partition creation sans encryption
then execute the following from the project source tree:

  >$ ./demo clear