dm-verity: remove support for asynchronous hashes
authorEric Biggers <ebiggers@kernel.org>
Wed, 9 Jul 2025 19:09:02 +0000 (12:09 -0700)
committerMikulas Patocka <mpatocka@redhat.com>
Thu, 31 Jul 2025 13:39:55 +0000 (15:39 +0200)
commitbdf253d580d7d30e7620844c63a5013fe7ba3f87
treeb249a8be3f19e38b235a517b329f39e1c850bc3f
parent6e11952a6abc4641dc8ae63f01b318b31b44e8db
dm-verity: remove support for asynchronous hashes

The support for asynchronous hashes in dm-verity has outlived its
usefulness.  It adds significant code complexity and opportunity for
bugs.  I don't know of anyone using it in practice.  (The original
submitter of the code possibly was, but that was 8 years ago.)  Data I
recently collected for en/decryption shows that using off-CPU crypto
"accelerators" is consistently much slower than the CPU
(https://lore.kernel.org/r/20250704070322.20692-1-ebiggers@kernel.org/),
even on CPUs that lack dedicated cryptographic instructions.  Similar
results are likely to be seen for hashing.

I already removed support for asynchronous hashes from fsverity two
years ago, and no one ever complained.

Moreover, neither dm-verity, fsverity, nor fscrypt has ever actually
used the asynchronous crypto algorithms in a truly asynchronous manner.
The lack of interest in such optimizations provides further evidence
that it's only the CPU-based crypto that actually matters.

Historically, it's also been common for people to forget to enable the
optimized SHA-256 code, which could contribute to an off-CPU crypto
engine being perceived as more useful than it really is.  In 6.16 I
fixed that: the optimized SHA-256 code is now enabled by default.

Therefore, let's drop the support for asynchronous hashes in dm-verity.

Tested with verity-compat-test.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
drivers/md/dm-verity-fec.c
drivers/md/dm-verity-target.c
drivers/md/dm-verity.h