crypto: vmac - Make VMAC work when blocks aren't aligned

VMAC implementation, as it is, does not work with blocks that
are not multiples of 128-bytes.  Furthermore, this is a problem
when using the implementation on scatterlists, even
when the complete plain text is 128-byte multiple, as the pieces
that get passed to vmac_update can be pretty much any size.

I also added test cases for unaligned blocks.

Signed-off-by: Salman Qazi <sqazi@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
diff --git a/crypto/vmac.c b/crypto/vmac.c
index f2338ca..2eb11a3 100644
--- a/crypto/vmac.c
+++ b/crypto/vmac.c
@@ -375,6 +375,11 @@
 	u64 pkh = ctx->polykey[0];
 	u64 pkl = ctx->polykey[1];
 
+	if (!mbytes)
+		return;
+
+	BUG_ON(mbytes % VMAC_NHBYTES);
+
 	mptr = (u64 *)m;
 	i = mbytes / VMAC_NHBYTES;  /* Must be non-zero */
 
@@ -454,7 +459,7 @@
 }
 
 static u64 vmac(unsigned char m[], unsigned int mbytes,
-			unsigned char n[16], u64 *tagl,
+			const unsigned char n[16], u64 *tagl,
 			struct vmac_ctx_t *ctx)
 {
 	u64 *in_n, *out_p;
@@ -559,8 +564,33 @@
 {
 	struct crypto_shash *parent = pdesc->tfm;
 	struct vmac_ctx_t *ctx = crypto_shash_ctx(parent);
+	int expand;
+	int min;
 
-	vhash_update(p, len, &ctx->__vmac_ctx);
+	expand = VMAC_NHBYTES - ctx->partial_size > 0 ?
+			VMAC_NHBYTES - ctx->partial_size : 0;
+
+	min = len < expand ? len : expand;
+
+	memcpy(ctx->partial + ctx->partial_size, p, min);
+	ctx->partial_size += min;
+
+	if (len < expand)
+		return 0;
+
+	vhash_update(ctx->partial, VMAC_NHBYTES, &ctx->__vmac_ctx);
+	ctx->partial_size = 0;
+
+	len -= expand;
+	p += expand;
+
+	if (len % VMAC_NHBYTES) {
+		memcpy(ctx->partial, p + len - (len % VMAC_NHBYTES),
+			len % VMAC_NHBYTES);
+		ctx->partial_size = len % VMAC_NHBYTES;
+	}
+
+	vhash_update(p, len - len % VMAC_NHBYTES, &ctx->__vmac_ctx);
 
 	return 0;
 }
@@ -572,10 +602,20 @@
 	vmac_t mac;
 	u8 nonce[16] = {};
 
-	mac = vmac(NULL, 0, nonce, NULL, ctx);
+	/* vmac() ends up accessing outside the array bounds that
+	 * we specify.  In appears to access up to the next 2-word
+	 * boundary.  We'll just be uber cautious and zero the
+	 * unwritten bytes in the buffer.
+	 */
+	if (ctx->partial_size) {
+		memset(ctx->partial + ctx->partial_size, 0,
+			VMAC_NHBYTES - ctx->partial_size);
+	}
+	mac = vmac(ctx->partial, ctx->partial_size, nonce, NULL, ctx);
 	memcpy(out, &mac, sizeof(vmac_t));
 	memset(&mac, 0, sizeof(vmac_t));
 	memset(&ctx->__vmac_ctx, 0, sizeof(struct vmac_ctx));
+	ctx->partial_size = 0;
 	return 0;
 }
 
@@ -673,4 +713,3 @@
 
 MODULE_LICENSE("GPL");
 MODULE_DESCRIPTION("VMAC hash algorithm");
-