crypto: s5p - deal gracefully with bogus input sizes

The s5p skcipher driver returns -EINVAL for zero length inputs, which
deviates from the behavior of the generic ECB template, and causes fuzz
tests to fail. In cases where the input is not a multiple of the AES
block size (and the chaining mode is not CTR), it prints an error to
the kernel log, which is a thing we usually try to avoid in response
to situations that can be triggered by unprivileged users.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Acked-by: Kamil Konieczny <k.konieczny@partner.samsung.com>
Reviewed-by: Krzysztof Kozlowski <krzk@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

diff --git a/drivers/crypto/s5p-sss.c b/drivers/crypto/s5p-sss.c
index 9ef2523..ef90c58 100644 (file)
--- a/drivers/crypto/s5p-sss.c
+++ b/drivers/crypto/s5p-sss.c
@@ -2056,9 +2056,12 @@ static int s5p_aes_crypt(struct ablkcipher_request *req, unsigned long mode)
        struct s5p_aes_ctx *ctx = crypto_ablkcipher_ctx(tfm);
        struct s5p_aes_dev *dev = ctx->dev;
 
+       if (!req->nbytes)
+               return 0;
+
        if (!IS_ALIGNED(req->nbytes, AES_BLOCK_SIZE) &&
                        ((mode & FLAGS_AES_MODE_MASK) != FLAGS_AES_CTR)) {
-               dev_err(dev->dev, "request size is not exact amount of AES blocks\n");
+               dev_dbg(dev->dev, "request size is not exact amount of AES blocks\n");
                return -EINVAL;
        }