Re: [PATCH bpf-next v3 2/6] crypto: Add BPF hash algorithm type registration module

[Daniel Hodges]

On Wed, Dec 17, 2025 at 06:45:14AM +0900, Song Liu wrote:
> On Sun, Dec 7, 2025 at 7:01 PM Daniel Hodges <g...@danielhodges.dev> wrote:
> >
> > Add bpf_crypto_shash module that registers a hash type with the BPF
> > crypto infrastructure, enabling BPF programs to access kernel hash
> > algorithms through a unified interface.
> >
> > Update the bpf_crypto_type interface with hash-specific callbacks:
> >    - alloc_tfm: Allocates crypto_shash context with proper descriptor size
> >    - free_tfm: Releases hash transform and context memory
> >    - has_algo: Checks algorithm availability via crypto_has_shash()
> >    - hash: Performs single-shot hashing via crypto_shash_digest()
> >    - digestsize: Returns the output size for the hash algorithm
> >    - get_flags: Exposes transform flags to BPF programs
> >
> > Update bpf_shash_ctx to contain crypto_shash transform and shash_desc
> > descriptor to accommodate algorithm-specific descriptor requirements.
> >
> > Signed-off-by: Daniel Hodges <g...@danielhodges.dev>
> > ---
> >  crypto/Makefile           |  3 ++
> >  crypto/bpf_crypto_shash.c | 95 +++++++++++++++++++++++++++++++++++++++
> >  2 files changed, 98 insertions(+)
> >  create mode 100644 crypto/bpf_crypto_shash.c
> >
> > diff --git a/crypto/Makefile b/crypto/Makefile
> > index 16a35649dd91..853dff375906 100644
> > --- a/crypto/Makefile
> > +++ b/crypto/Makefile
> > @@ -30,6 +30,9 @@ obj-$(CONFIG_CRYPTO_ECHAINIV) += echainiv.o
> >  crypto_hash-y += ahash.o
> >  crypto_hash-y += shash.o
> >  obj-$(CONFIG_CRYPTO_HASH2) += crypto_hash.o
> > +ifeq ($(CONFIG_BPF_SYSCALL),y)
> > +obj-$(CONFIG_CRYPTO_HASH2) += bpf_crypto_shash.o
> > +endif
> >
> >  obj-$(CONFIG_CRYPTO_AKCIPHER2) += akcipher.o
> >  obj-$(CONFIG_CRYPTO_SIG2) += sig.o
> > diff --git a/crypto/bpf_crypto_shash.c b/crypto/bpf_crypto_shash.c
> > new file mode 100644
> > index 000000000000..95c178ec0ce8
> > --- /dev/null
> > +++ b/crypto/bpf_crypto_shash.c
> > @@ -0,0 +1,95 @@
> > +// SPDX-License-Identifier: GPL-2.0-only
> > +/* Copyright (c) 2025 Meta Platforms, Inc. and affiliates. */
> > +#include <linux/types.h>
> > +#include <linux/module.h>
> > +#include <linux/bpf_crypto.h>
> > +#include <crypto/hash.h>
> > +
> > +struct bpf_shash_ctx {
> > +       struct crypto_shash *tfm;
> > +       struct shash_desc desc;
> > +};
> > +
> > +static void *bpf_crypto_shash_alloc_tfm(const char *algo)
> > +{
> > +       struct bpf_shash_ctx *ctx;
> > +       struct crypto_shash *tfm;
> > +
> > +       tfm = crypto_alloc_shash(algo, 0, 0);
> > +       if (IS_ERR(tfm))
> > +               return tfm;
> > +
> > +       ctx = kzalloc(sizeof(*ctx) + crypto_shash_descsize(tfm), 
> > GFP_KERNEL);
> > +       if (!ctx) {
> > +               crypto_free_shash(tfm);
> > +               return ERR_PTR(-ENOMEM);
> > +       }
> > +
> > +       ctx->tfm = tfm;
> > +       ctx->desc.tfm = tfm;
> > +
> > +       return ctx;
> > +}
> 
> What if we let bpf_crypto_shash_alloc_tfm() return a "struct shash_desc"?
> shash_desc->tfm is already struct crypto_shash. This way, we don't need
> bpf_shash_ctx any more. Would this work?
> 
> Thanks,
> Song
Yeah, that's much cleaner than the void * return.

-Daniel