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

[bot+bpf-ci@kernel.org]

[-- Attachment #1: Type: text/plain, Size: 2862 bytes --]

> 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 @@
[ ... ]
> +
> +static const struct bpf_crypto_type bpf_crypto_shash_type = {
> +	.alloc_tfm	= bpf_crypto_shash_alloc_tfm,
> +	.free_tfm	= bpf_crypto_shash_free_tfm,
> +	.has_algo	= bpf_crypto_shash_has_algo,
> +	.hash		= bpf_crypto_shash_hash,
> +	.digestsize	= bpf_crypto_shash_digestsize,
> +	.get_flags	= bpf_crypto_shash_get_flags,
> +	.owner		= THIS_MODULE,
> +	.name		= "hash",
> +};

The bpf_crypto_type implementation leaves several function pointers as
NULL (setkey, setauthsize, encrypt, decrypt, ivsize, statesize), which is
appropriate for plain hash algorithms. However, this creates an interface
contract mismatch with bpf_crypto_ctx_create() in kernel/bpf/crypto.c.

Looking at the caller bpf_crypto_ctx_create():

  Line 174-177:
    if (!params->key_len || params->key_len > sizeof(params->key)) {
        *err = -EINVAL;
        goto err_module_put;
    }

Can hash contexts be created? This check requires key_len to be non-zero,
but plain hash algorithms like SHA256 don't use keys. This prevents
creating hash contexts entirely.

  Line 186-189:
    ctx->tfm = type->alloc_tfm(params->algo);
    if (IS_ERR(ctx->tfm)) {
        *err = PTR_ERR(ctx->tfm);
        goto err_free_tfm;

Can bpf_crypto_shash_free_tfm handle ERR_PTR? When alloc_tfm fails and
returns ERR_PTR, the code stores it in ctx->tfm then jumps to err_free_tfm
which calls type->free_tfm(ctx->tfm). In bpf_crypto_shash_free_tfm(), the
code dereferences ctx->tfm without checking for ERR_PTR, which will crash:

    static void bpf_crypto_shash_free_tfm(void *tfm)
    {
        struct bpf_shash_ctx *ctx = tfm;
        crypto_free_shash(ctx->tfm);  /* dereferences ctx */
        kfree(ctx);
    }

  Line 198:
    *err = type->setkey(ctx->tfm, params->key, params->key_len);

Does this dereference NULL? The hash implementation doesn't provide setkey
(it's NULL in bpf_crypto_shash_type), but this line calls it
unconditionally.

  Line 207:
    ctx->siv_len = type->ivsize(ctx->tfm) + type->statesize(ctx->tfm);

Do these dereference NULL? The hash implementation doesn't provide ivsize
or statesize (both NULL), but this line calls them unconditionally.

Note: These issues are fixed in commit 14cc32f873e0 ("bpf: Add SHA hash
kfunc for cryptographic hashing") which makes the setkey, ivsize, and
statesize calls conditional, and fixes the ERR_PTR handling by changing
the error path from "goto err_free_tfm" to "goto err_free_ctx".


---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

CI run summary: https://github.com/kernel-patches/bpf/actions/runs/20015426139