From: Andrei Botila <[email protected]>
Through this RFC we try to standardize the way input lengths equal to 0
are handled in all skcipher algorithms. Currently, in xts when an input
has a length smaller than XTS_BLOCK_SIZE it returns -EINVAL while the
other algorithms return 0 for input lengths equal to zero.
The algorithms that implement this check are CBC, ARC4, CFB, OFB, SALSA20,
CTR, ECB and PCBC, XTS being the outlier here. All of them call
skcipher_walk_virt() which returns 0 if skcipher_walk_skcipher() finds
that input length is equal to 0.
This case was discovered when fuzz testing was enabled since it generates
this input length.
This RFC wants to find out if the approach is ok before updating the
other xts implementations.
Signed-off-by: Andrei Botila <[email protected]>
---
crypto/xts.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/crypto/xts.c b/crypto/xts.c
index 29efa15f1495..51eaf08603af 100644
--- a/crypto/xts.c
+++ b/crypto/xts.c
@@ -258,6 +258,9 @@ static int encrypt(struct skcipher_request *req)
struct skcipher_request *subreq = &rctx->subreq;
int err;
+ if (!req->cryptlen)
+ return 0;
+
err = init_crypt(req, encrypt_done) ?:
xor_tweak_pre(req, true) ?:
crypto_skcipher_encrypt(subreq) ?:
@@ -275,6 +278,9 @@ static int decrypt(struct skcipher_request *req)
struct skcipher_request *subreq = &rctx->subreq;
int err;
+ if (!req->cryptlen)
+ return 0;
+
err = init_crypt(req, decrypt_done) ?:
xor_tweak_pre(req, false) ?:
crypto_skcipher_decrypt(subreq) ?:
--
2.17.1
Am Donnerstag, 5. M?rz 2020, 11:22:55 CET schrieb Andrei Botila:
Hi Andrei,
> From: Andrei Botila <[email protected]>
>
> Through this RFC we try to standardize the way input lengths equal to 0
> are handled in all skcipher algorithms. Currently, in xts when an input
> has a length smaller than XTS_BLOCK_SIZE it returns -EINVAL while the
> other algorithms return 0 for input lengths equal to zero.
> The algorithms that implement this check are CBC, ARC4, CFB, OFB, SALSA20,
> CTR, ECB and PCBC, XTS being the outlier here. All of them call
> skcipher_walk_virt() which returns 0 if skcipher_walk_skcipher() finds
> that input length is equal to 0.
> This case was discovered when fuzz testing was enabled since it generates
> this input length.
> This RFC wants to find out if the approach is ok before updating the
> other xts implementations.
It may be a good idea to consolidate that. However, changing only one
implementation is not good.
All XTS implementations would need to be converted then.
Ciao
Stephan