[v5,0/6] fscrypt preparations for encryption+casefolding

This is a cleaned up and fixed version of the fscrypt patches to prepare
for directories that are both encrypted and casefolded.

Patches 1-3 start deriving a SipHash key for the new dirhash method that
will be used by encrypted+casefolded directories.  To avoid unnecessary
overhead, we only do this if the directory is actually casefolded.

Patch 4 fixes a bug in UBIFS where it didn't gracefully handle invalid
hash values in fscrypt no-key names.  This is an existing bug, but the
new fscrypt no-key name format (patch 6) made it much easier to trigger;
it started being hit by 'kvm-xfstests -c ubifs -g encrypt'.

Patch 5 updates UBIFS to make it ready for the new fscrypt no-key name
format that always includes the dirhash.

Patch 6 modifies the fscrypt no-key names to always include the dirhash,
since with the new dirhash method the dirhash will no longer be
computable from the ciphertext filename without the key.  It also fixes
a longstanding issue where there could be collisions in the no-key
names, due to not using a proper cryptographic hash to abbreviate names.

For more information see the main patch series, which includes the
filesystem-specific changes:
https://lkml.kernel.org/linux-fscrypt/20200117214246.235591-1-drosen@google.com/T/#u

This applies to fscrypt.git#master.

Changed v4 => v5:
  - Fixed UBIFS encryption to work with the new no-key name format.

Daniel Rosenberg (3):
  fscrypt: don't allow v1 policies with casefolding
  fscrypt: derive dirhash key for casefolded directories
  fscrypt: improve format of no-key names

Eric Biggers (3):
  fscrypt: clarify what is meant by a per-file key
  ubifs: don't trigger assertion on invalid no-key filename
  ubifs: allow both hash and disk name to be provided in no-key names

 Documentation/filesystems/fscrypt.rst |  40 +++--
 fs/crypto/Kconfig                     |   1 +
 fs/crypto/fname.c                     | 239 ++++++++++++++++++++------
 fs/crypto/fscrypt_private.h           |  19 +-
 fs/crypto/hooks.c                     |  44 +++++
 fs/crypto/keysetup.c                  |  81 ++++++---
 fs/crypto/keysetup_v1.c               |   4 +-
 fs/crypto/policy.c                    |   7 +
 fs/inode.c                            |   3 +-
 fs/ubifs/dir.c                        |   6 +-
 fs/ubifs/journal.c                    |   4 +-
 fs/ubifs/key.h                        |   1 -
 include/linux/fscrypt.h               |  94 +++-------
 13 files changed, 365 insertions(+), 178 deletions(-)

On Mon, Jan 20, 2020 at 02:31:55PM -0800, Eric Biggers wrote:
> This is a cleaned up and fixed version of the fscrypt patches to prepare
> for directories that are both encrypted and casefolded.
> 
> Patches 1-3 start deriving a SipHash key for the new dirhash method that
> will be used by encrypted+casefolded directories.  To avoid unnecessary
> overhead, we only do this if the directory is actually casefolded.
> 
> Patch 4 fixes a bug in UBIFS where it didn't gracefully handle invalid
> hash values in fscrypt no-key names.  This is an existing bug, but the
> new fscrypt no-key name format (patch 6) made it much easier to trigger;
> it started being hit by 'kvm-xfstests -c ubifs -g encrypt'.
> 
> Patch 5 updates UBIFS to make it ready for the new fscrypt no-key name
> format that always includes the dirhash.
> 
> Patch 6 modifies the fscrypt no-key names to always include the dirhash,
> since with the new dirhash method the dirhash will no longer be
> computable from the ciphertext filename without the key.  It also fixes
> a longstanding issue where there could be collisions in the no-key
> names, due to not using a proper cryptographic hash to abbreviate names.
> 
> For more information see the main patch series, which includes the
> filesystem-specific changes:
> https://lkml.kernel.org/linux-fscrypt/20200117214246.235591-1-drosen@google.com/T/#u
> 
> This applies to fscrypt.git#master.
> 
> Changed v4 => v5:
>   - Fixed UBIFS encryption to work with the new no-key name format.

I've applied this series to fscrypt.git#master; however I'd still like Acked-bys
from the UBIFS maintainers on the two UBIFS patches, as well as more
Reviewed-bys from anyone interested.  If I don't hear anything from anyone, I
might drop these to give more time, especially if there isn't an v5.5-rc8.

- Eric

On Wed, Jan 22, 2020 at 3:06 PM Eric Biggers <ebiggers@kernel.org> wrote:
>
> I've applied this series to fscrypt.git#master; however I'd still like Acked-bys
> from the UBIFS maintainers on the two UBIFS patches, as well as more
> Reviewed-bys from anyone interested.  If I don't hear anything from anyone, I
> might drop these to give more time, especially if there isn't an v5.5-rc8.
>
> - Eric

The patches look good to me. Thanks for the fixups.
-Daniel