Re: [RFC 0/4] fs: provide per-filesystem options to disable fscrypt

[Eric Biggers <ebiggers@kernel.org>]

On Thu, Nov 10, 2022 at 03:12:21PM +0100, Niels de Vos wrote:
> While more filesystems are getting support for fscrypt, it is useful to
> be able to disable fscrypt for a selection of filesystems, while
> enabling it for others.
> 
> The new USE_FS_ENCRYPTION define gets picked up in
> include/linux/fscrypt.h. This allows filesystems to choose to use the
> empty function definitions, or the functional ones when fscrypt is to be
> used with the filesystem.
> 
> Using USE_FS_ENCRYPTION is a relatively clean approach, and requires
> minimal changes to the filesystems supporting fscrypt. This RFC is
> mostly for checking the acceptance of this solution, or if an other
> direction is preferred.
> 
> ---
> 
> Niels de Vos (4):
>   fscrypt: introduce USE_FS_ENCRYPTION
>   fs: make fscrypt support an ext4 config option
>   fs: make fscrypt support a f2fs config option
>   fs: make fscrypt support a UBIFS config option

So as others have pointed out, it doesn't seem worth the complexity to do this.

For a bit of historical context, before Linux v5.1, we did have per-filesystem
options for this: CONFIG_EXT4_ENCRYPTION, CONFIG_F2FS_FS_ENCRYPTION, and
CONFIG_UBIFS_FS_ENCRYPTION.  If you enabled one of these, it selected
CONFIG_FS_ENCRYPTION to get the code in fs/crypto/.  CONFIG_FS_ENCRYPTION was a
tristate, so the code in fs/crypto/ could be built as a loadable module if it
was only needed by filesystems that were loadable modules themselves.

Having fs/crypto/ possibly be a loadable module was problematic, though, because
it made it impossible to call into fs/crypto/ from built-in code such as
fs/buffer.c, fs/ioctl.c, fs/libfs.c, fs/super.c, fs/iomap/direct-io.c, etc.  So
that's why we made CONFIG_FS_ENCRYPTION into a bool.  At the same time, we
decided to simplify the kconfig options by removing the per-filesystem options
so that it worked like CONFIG_QUOTA, CONFIG_FS_DAX, CONFIG_FS_POSIX_ACL, etc.

I suppose we *could* have *just* changed CONFIG_FS_ENCRYPTION to a bool to solve
the first problem, and kept the per-filesystem options.  I think that wouldn't
have made a lot of sense, though, for the reasons that Ted has already covered.

A further point, beyond what Ted has already covered, is that
non-filesystem-specific code can't honor filesystem-specific options.  So e.g.
if you had a filesystem with encryption disabled by kconfig, that then called
into fs/iomap/direct-io.c to process an I/O request, it could potentially still
call into fs/crypto/ to enable encryption on that I/O request, since
fs/iomap/direct-io.c would think that encryption support is enabled.

Granted, that *should* never actually happen, because this would only make a
difference on encrypted files, and the filesystem shouldn't have allowed an
encrypted file to be opened if it doesn't have encryption support enabled.  But
it does seem a bit odd, given that it would go against the goal of compiling out
all encryption code for a filesystem.

- Eric