sm3_compress() calls rol32() with shift >= 32, which causes undefined
behavior. This is easily detected by enabling CONFIG_UBSAN.
Explicitly AND with 31 to make the behavior well defined.
Fixes: 4f0fc1600edb ("crypto: sm3 - add OSCCA SM3 secure hash")
Cc: <stable@vger.kernel.org> # v4.15+
Cc: Gilad Ben-Yossef <gilad@benyossef.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
for (i = 0; i <= 63; i++) {
- ss1 = rol32((rol32(a, 12) + e + rol32(t(i), i)), 7);
+ ss1 = rol32((rol32(a, 12) + e + rol32(t(i), i & 31)), 7);
ss2 = ss1 ^ rol32(a, 12);