Re: [PATCH v4 2/2] dm-integrity: introduce ahash support for the internal hash

[Harald Freudenberger <freude@linux.ibm.com>]

On 2025-02-05 21:23, Mikulas Patocka wrote:
> Introduce ahash support for the "internal hash" algorithm.
> 
> Rework the dm-integrity code to be able to run the "internal hash"
> either with a synchronous ("shash") or asynchronous ("ahash") hash
> algorithm implementation.
> 
> The get_mac() function now tries to decide which of the digest
> implemenations to use if there is a choice:
> - If an ahash and shash tfm is available and both are backed by the
>   same driver name it is assumed that the shash is the faster
>   implementation and thus the shash tfm is delivered to the caller.
> - If an ahash and shash tfm is available but the backing device driver
>   divers (different driver names) it is assumed that the ahash
>   implementation is a "better" hardware based implementation and thus
>   the ahash tfm is delivered to the caller.
> - If there is no choice, for example only an ahash or an shash
>   implementation is available then this tfm is delivered to the
>   caller. Especially in cases where only an ahash implementation is
>   available this is now used instead of failing.
> - The caller can steer this choice by passing a NULL to the ahash or
>   shash parameter, thus enforcing to only allocate an algorithm of the
>   remaining possibility.
> 
> The function integrity_sector_checksum() is now only a dispatcher
> function calling one of the two new functions
> integrity_ahash_sector_checksum() or integrity_shash_sector_checksum()
> based on which tfm is allocated based on the two new fields
> internal_shash and internal_ahash in struct dm_integrity_c.
> 
> Together with this comes some slight rework around availability and
> digest size of the internal hash in use.
> 
> Signed-off-by: Harald Freudenberger <freude@linux.ibm.com>
> Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
> ---
>  drivers/md/dm-integrity.c |  350 
> +++++++++++++++++++++++++++++++++-------------
>  1 file changed, 252 insertions(+), 98 deletions(-)
> 
> Index: linux-2.6/drivers/md/dm-integrity.c
> ===================================================================

...

>  static struct target_type integrity_target = {
>  	.name			= "integrity",
> -	.version		= {1, 13, 0},
> +	.version		= {1, 14, 0},
>  	.module			= THIS_MODULE,
>  	.features		= DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
>  	.ctr			= dm_integrity_ctr,

Mikulas, I am still working.
All the straight forward tests where successful. Even my new phma-shaxxx
was running fine. The integrity-compat-test also found no issues.
However, as you wrote, I enforced using ahashes in get_mac() and
catched some weird kernel crashes but had no time to investigate 
further.
I'll continue on this as soon I have some spare time and will report
the results.

kind regards
Harald Freudenberger