crypto: x86/xts - implement support for ciphertext stealing

Align the x86 code with the generic XTS template, which now supports
ciphertext stealing as described by the IEEE XTS-AES spec P1619.
Tested-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

arch/x86/crypto/aesni-intel_glue.c

@@ -609,7 +609,8 @@ static int xts_encrypt(struct skcipher_request *req)
 	return glue_xts_req_128bit(&aesni_enc_xts, req,
 				   XTS_TWEAK_CAST(aesni_xts_tweak),
 				   aes_ctx(ctx->raw_tweak_ctx),
-				   aes_ctx(ctx->raw_crypt_ctx));
+				   aes_ctx(ctx->raw_crypt_ctx),
+				   false);
 }
 
 static int xts_decrypt(struct skcipher_request *req)
@@ -620,7 +621,8 @@ static int xts_decrypt(struct skcipher_request *req)
 	return glue_xts_req_128bit(&aesni_dec_xts, req,
 				   XTS_TWEAK_CAST(aesni_xts_tweak),
 				   aes_ctx(ctx->raw_tweak_ctx),
-				   aes_ctx(ctx->raw_crypt_ctx));
+				   aes_ctx(ctx->raw_crypt_ctx),
+				   true);
 }
 
 static int

arch/x86/crypto/camellia_aesni_avx2_glue.c

@@ -182,7 +182,7 @@ static int xts_encrypt(struct skcipher_request *req)
 
 	return glue_xts_req_128bit(&camellia_enc_xts, req,
 				   XTS_TWEAK_CAST(camellia_enc_blk),
-				   &ctx->tweak_ctx, &ctx->crypt_ctx);
+				   &ctx->tweak_ctx, &ctx->crypt_ctx, false);
 }
 
 static int xts_decrypt(struct skcipher_request *req)
@@ -192,7 +192,7 @@ static int xts_decrypt(struct skcipher_request *req)
 
 	return glue_xts_req_128bit(&camellia_dec_xts, req,
 				   XTS_TWEAK_CAST(camellia_enc_blk),
-				   &ctx->tweak_ctx, &ctx->crypt_ctx);
+				   &ctx->tweak_ctx, &ctx->crypt_ctx, true);
 }
 
 static struct skcipher_alg camellia_algs[] = {

arch/x86/crypto/camellia_aesni_avx_glue.c

@@ -208,7 +208,7 @@ static int xts_encrypt(struct skcipher_request *req)
 
 	return glue_xts_req_128bit(&camellia_enc_xts, req,
 				   XTS_TWEAK_CAST(camellia_enc_blk),
-				   &ctx->tweak_ctx, &ctx->crypt_ctx);
+				   &ctx->tweak_ctx, &ctx->crypt_ctx, false);
 }
 
 static int xts_decrypt(struct skcipher_request *req)
@@ -218,7 +218,7 @@ static int xts_decrypt(struct skcipher_request *req)
 
 	return glue_xts_req_128bit(&camellia_dec_xts, req,
 				   XTS_TWEAK_CAST(camellia_enc_blk),
-				   &ctx->tweak_ctx, &ctx->crypt_ctx);
+				   &ctx->tweak_ctx, &ctx->crypt_ctx, true);
 }
 
 static struct skcipher_alg camellia_algs[] = {

arch/x86/crypto/cast6_avx_glue.c

@@ -201,7 +201,7 @@ static int xts_encrypt(struct skcipher_request *req)
 
 	return glue_xts_req_128bit(&cast6_enc_xts, req,
 				   XTS_TWEAK_CAST(__cast6_encrypt),
-				   &ctx->tweak_ctx, &ctx->crypt_ctx);
+				   &ctx->tweak_ctx, &ctx->crypt_ctx, false);
 }
 
 static int xts_decrypt(struct skcipher_request *req)
@@ -211,7 +211,7 @@ static int xts_decrypt(struct skcipher_request *req)
 
 	return glue_xts_req_128bit(&cast6_dec_xts, req,
 				   XTS_TWEAK_CAST(__cast6_encrypt),
-				   &ctx->tweak_ctx, &ctx->crypt_ctx);
+				   &ctx->tweak_ctx, &ctx->crypt_ctx, true);
 }
 
 static struct skcipher_alg cast6_algs[] = {

arch/x86/crypto/glue_helper.c

@@ -14,6 +14,7 @@
 #include <crypto/b128ops.h>
 #include <crypto/gf128mul.h>
 #include <crypto/internal/skcipher.h>
+#include <crypto/scatterwalk.h>
 #include <crypto/xts.h>
 #include <asm/crypto/glue_helper.h>
 
@@ -259,17 +260,36 @@ static unsigned int __glue_xts_req_128bit(const struct common_glue_ctx *gctx,
 int glue_xts_req_128bit(const struct common_glue_ctx *gctx,
 			struct skcipher_request *req,
 			common_glue_func_t tweak_fn, void *tweak_ctx,
-			void *crypt_ctx)
+			void *crypt_ctx, bool decrypt)
 {
+	const bool cts = (req->cryptlen % XTS_BLOCK_SIZE);
 	const unsigned int bsize = 128 / 8;
+	struct skcipher_request subreq;
 	struct skcipher_walk walk;
 	bool fpu_enabled = false;
-	unsigned int nbytes;
+	unsigned int nbytes, tail;
 	int err;
 
+	if (req->cryptlen < XTS_BLOCK_SIZE)
+		return -EINVAL;
+
+	if (unlikely(cts)) {
+		struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+
+		tail = req->cryptlen % XTS_BLOCK_SIZE + XTS_BLOCK_SIZE;
+
+		skcipher_request_set_tfm(&subreq, tfm);
+		skcipher_request_set_callback(&subreq,
+					      crypto_skcipher_get_flags(tfm),
+					      NULL, NULL);
+		skcipher_request_set_crypt(&subreq, req->src, req->dst,
+					   req->cryptlen - tail, req->iv);
+		req = &subreq;
+	}
+
 	err = skcipher_walk_virt(&walk, req, false);
 	nbytes = walk.nbytes;
-	if (!nbytes)
+	if (err)
 		return err;
 
 	/* set minimum length to bsize, for tweak_fn */
@@ -287,6 +307,47 @@ int glue_xts_req_128bit(const struct common_glue_ctx *gctx,
 		nbytes = walk.nbytes;
 	}
 
+	if (unlikely(cts)) {
+		u8 *next_tweak, *final_tweak = req->iv;
+		struct scatterlist *src, *dst;
+		struct scatterlist s[2], d[2];
+		le128 b[2];
+
+		dst = src = scatterwalk_ffwd(s, req->src, req->cryptlen);
+		if (req->dst != req->src)
+			dst = scatterwalk_ffwd(d, req->dst, req->cryptlen);
+
+		if (decrypt) {
+			next_tweak = memcpy(b, req->iv, XTS_BLOCK_SIZE);
+			gf128mul_x_ble(b, b);
+		} else {
+			next_tweak = req->iv;
+		}
+
+		skcipher_request_set_crypt(&subreq, src, dst, XTS_BLOCK_SIZE,
+					   next_tweak);
+
+		err = skcipher_walk_virt(&walk, req, false) ?:
+		      skcipher_walk_done(&walk,
+				__glue_xts_req_128bit(gctx, crypt_ctx, &walk));
+		if (err)
+			goto out;
+
+		scatterwalk_map_and_copy(b, dst, 0, XTS_BLOCK_SIZE, 0);
+		memcpy(b + 1, b, tail - XTS_BLOCK_SIZE);
+		scatterwalk_map_and_copy(b, src, XTS_BLOCK_SIZE,
+					 tail - XTS_BLOCK_SIZE, 0);
+		scatterwalk_map_and_copy(b, dst, 0, tail, 1);
+
+		skcipher_request_set_crypt(&subreq, dst, dst, XTS_BLOCK_SIZE,
+					   final_tweak);
+
+		err = skcipher_walk_virt(&walk, req, false) ?:
+		      skcipher_walk_done(&walk,
+				__glue_xts_req_128bit(gctx, crypt_ctx, &walk));
+	}
+
+out:
 	glue_fpu_end(fpu_enabled);
 
 	return err;

arch/x86/crypto/serpent_avx2_glue.c

@@ -167,7 +167,7 @@ static int xts_encrypt(struct skcipher_request *req)
 
 	return glue_xts_req_128bit(&serpent_enc_xts, req,
 				   XTS_TWEAK_CAST(__serpent_encrypt),
-				   &ctx->tweak_ctx, &ctx->crypt_ctx);
+				   &ctx->tweak_ctx, &ctx->crypt_ctx, false);
 }
 
 static int xts_decrypt(struct skcipher_request *req)
@@ -177,7 +177,7 @@ static int xts_decrypt(struct skcipher_request *req)
 
 	return glue_xts_req_128bit(&serpent_dec_xts, req,
 				   XTS_TWEAK_CAST(__serpent_encrypt),
-				   &ctx->tweak_ctx, &ctx->crypt_ctx);
+				   &ctx->tweak_ctx, &ctx->crypt_ctx, true);
 }
 
 static struct skcipher_alg serpent_algs[] = {

arch/x86/crypto/serpent_avx_glue.c

@@ -207,7 +207,7 @@ static int xts_encrypt(struct skcipher_request *req)
 
 	return glue_xts_req_128bit(&serpent_enc_xts, req,
 				   XTS_TWEAK_CAST(__serpent_encrypt),
-				   &ctx->tweak_ctx, &ctx->crypt_ctx);
+				   &ctx->tweak_ctx, &ctx->crypt_ctx, false);
 }
 
 static int xts_decrypt(struct skcipher_request *req)
@@ -217,7 +217,7 @@ static int xts_decrypt(struct skcipher_request *req)
 
 	return glue_xts_req_128bit(&serpent_dec_xts, req,
 				   XTS_TWEAK_CAST(__serpent_encrypt),
-				   &ctx->tweak_ctx, &ctx->crypt_ctx);
+				   &ctx->tweak_ctx, &ctx->crypt_ctx, true);
 }
 
 static struct skcipher_alg serpent_algs[] = {

arch/x86/crypto/twofish_avx_glue.c

@@ -210,7 +210,7 @@ static int xts_encrypt(struct skcipher_request *req)
 
 	return glue_xts_req_128bit(&twofish_enc_xts, req,
 				   XTS_TWEAK_CAST(twofish_enc_blk),
-				   &ctx->tweak_ctx, &ctx->crypt_ctx);
+				   &ctx->tweak_ctx, &ctx->crypt_ctx, false);
 }
 
 static int xts_decrypt(struct skcipher_request *req)
@@ -220,7 +220,7 @@ static int xts_decrypt(struct skcipher_request *req)
 
 	return glue_xts_req_128bit(&twofish_dec_xts, req,
 				   XTS_TWEAK_CAST(twofish_enc_blk),
-				   &ctx->tweak_ctx, &ctx->crypt_ctx);
+				   &ctx->tweak_ctx, &ctx->crypt_ctx, true);
 }
 
 static struct skcipher_alg twofish_algs[] = {

arch/x86/include/asm/crypto/glue_helper.h

@@ -114,7 +114,7 @@ extern int glue_ctr_req_128bit(const struct common_glue_ctx *gctx,
 extern int glue_xts_req_128bit(const struct common_glue_ctx *gctx,
 			       struct skcipher_request *req,
 			       common_glue_func_t tweak_fn, void *tweak_ctx,
-			       void *crypt_ctx);
+			       void *crypt_ctx, bool decrypt);
 
 extern void glue_xts_crypt_128bit_one(void *ctx, u128 *dst, const u128 *src,
 				      le128 *iv, common_glue_func_t fn);