path: root/main/openssl/openssl-0.9.8k-padlock-sha.patch

#
# OpenSSL patch to support VIA C7 hash engine
# Written by: Timo Teras <timo.teras@iki.fi>
# based on patch by: Michal Ludvig <michal@logix.cz>
#                    http://www.logix.cz/michal/devel/padlock
#
Index: openssl-0.9.8k/crypto/engine/eng_padlock.c
===================================================================
--- openssl-0.9.8k.orig/crypto/engine/eng_padlock.c	2009-07-27 16:18:20.000000000 +0300
+++ openssl-0.9.8k/crypto/engine/eng_padlock.c	2009-07-30 22:02:54.000000000 +0300
@@ -1,10 +1,13 @@
-/* 
+/*
  * Support for VIA PadLock Advanced Cryptography Engine (ACE)
  * Written by Michal Ludvig <michal@logix.cz>
  *            http://www.logix.cz/michal
  *
- * Big thanks to Andy Polyakov for a help with optimization, 
- * assembler fixes, port to MS Windows and a lot of other 
+ * SHA support by Timo Teras <timo.teras@iki.fi> based on code
+ * originally by Michal Ludvig.
+ *
+ * Big thanks to Andy Polyakov for a help with optimization,
+ * assembler fixes, port to MS Windows and a lot of other
  * valuable work on this engine!
  */
 
@@ -66,6 +69,13 @@
 #include <stdio.h>
 #include <string.h>
 
+#include <signal.h>
+#include <stdint.h>
+#include <unistd.h>
+#include <sys/mman.h>
+#include <sys/ucontext.h>
+#include <arpa/inet.h>
+
 #include <openssl/opensslconf.h>
 #include <openssl/crypto.h>
 #include <openssl/dso.h>
@@ -74,12 +84,23 @@
 #ifndef OPENSSL_NO_AES
 #include <openssl/aes.h>
 #endif
+#ifndef OPENSSL_NO_SHA
+#include <openssl/sha.h>
+#endif
 #include <openssl/rand.h>
 #include <openssl/err.h>
 
 #ifndef OPENSSL_NO_HW
 #ifndef OPENSSL_NO_HW_PADLOCK
 
+/* PadLock RNG is disabled by default */
+#define	PADLOCK_NO_RNG	1
+
+/* No ASM routines for SHA in MSC yet */
+#ifdef _MSC_VER
+#define OPENSSL_NO_SHA
+#endif
+
 /* Attempt to have a single source for both 0.9.7 and 0.9.8 :-) */
 #if (OPENSSL_VERSION_NUMBER >= 0x00908000L)
 #  ifndef OPENSSL_NO_DYNAMIC_ENGINE
@@ -96,7 +117,7 @@
 /* VIA PadLock AES is available *ONLY* on some x86 CPUs.
    Not only that it doesn't exist elsewhere, but it
    even can't be compiled on other platforms!
- 
+
    In addition, because of the heavy use of inline assembler,
    compiler choice is limited to GCC and Microsoft C. */
 #undef COMPILE_HW_PADLOCK
@@ -138,20 +159,42 @@
 static int padlock_init(ENGINE *e);
 
 /* RNG Stuff */
+#ifndef PADLOCK_NO_RNG
 static RAND_METHOD padlock_rand;
+#endif
 
 /* Cipher Stuff */
 #ifndef OPENSSL_NO_AES
 static int padlock_ciphers(ENGINE *e, const EVP_CIPHER **cipher, const int **nids, int nid);
 #endif
 
+/* Digest Stuff */
+#ifndef OPENSSL_NO_SHA
+static int padlock_digests(ENGINE *e, const EVP_MD **digest, const int **nids, int nid);
+static volatile void *padlock_cached_sha_buffer = NULL;
+#endif
+
 /* Engine names */
 static const char *padlock_id = "padlock";
 static char padlock_name[100];
 
 /* Available features */
-static int padlock_use_ace = 0;	/* Advanced Cryptography Engine */
-static int padlock_use_rng = 0;	/* Random Number Generator */
+enum padlock_flags {
+	PADLOCK_RNG  = 0x01,
+	PADLOCK_ACE  = 0x02,
+	PADLOCK_ACE2 = 0x04,
+	PADLOCK_PHE  = 0x08,
+	PADLOCK_PMM  = 0x10
+};
+enum padlock_flags padlock_flags;
+
+#define PADLOCK_HAVE_RNG  (padlock_flags & PADLOCK_RNG)
+#define PADLOCK_HAVE_ACE  (padlock_flags & (PADLOCK_ACE|PADLOCK_ACE2))
+#define PADLOCK_HAVE_ACE1 (padlock_flags & PADLOCK_ACE)
+#define PADLOCK_HAVE_ACE2 (padlock_flags & PADLOCK_ACE2)
+#define PADLOCK_HAVE_PHE  (padlock_flags & PADLOCK_PHE)
+#define PADLOCK_HAVE_PMM  (padlock_flags & PADLOCK_PMM)
+
 #ifndef OPENSSL_NO_AES
 static int padlock_aes_align_required = 1;
 #endif
@@ -165,25 +208,30 @@
 	/* Check available features */
 	padlock_available();
 
-#if 1	/* disable RNG for now, see commentary in vicinity of RNG code */
-	padlock_use_rng=0;
-#endif
-
 	/* Generate a nice engine name with available features */
 	BIO_snprintf(padlock_name, sizeof(padlock_name),
-		"VIA PadLock (%s, %s)", 
-		 padlock_use_rng ? "RNG" : "no-RNG",
-		 padlock_use_ace ? "ACE" : "no-ACE");
+		"VIA PadLock: %s%s%s%s%s",
+		 padlock_flags ? "" : "not supported",
+		 PADLOCK_HAVE_RNG ? "RNG " : "",
+		 PADLOCK_HAVE_ACE ? (PADLOCK_HAVE_ACE2 ? "ACE2 " : "ACE ") : "",
+		 PADLOCK_HAVE_PHE ? "PHE " : "",
+		 PADLOCK_HAVE_PMM ? "PMM " : "");
 
-	/* Register everything or return with an error */ 
+	/* Register everything or return with an error */
 	if (!ENGINE_set_id(e, padlock_id) ||
 	    !ENGINE_set_name(e, padlock_name) ||
 
-	    !ENGINE_set_init_function(e, padlock_init) ||
+	    !ENGINE_set_init_function(e, padlock_init)
 #ifndef OPENSSL_NO_AES
-	    (padlock_use_ace && !ENGINE_set_ciphers (e, padlock_ciphers)) ||
+	    || (PADLOCK_HAVE_ACE && !ENGINE_set_ciphers (e, padlock_ciphers))
 #endif
-	    (padlock_use_rng && !ENGINE_set_RAND (e, &padlock_rand))) {
+#ifndef OPENSSL_NO_SHA
+	    || (PADLOCK_HAVE_PHE && !ENGINE_set_digests (e, padlock_digests))
+#endif
+#ifndef PADLOCK_NO_RNG
+	    || (PADLOCK_HAVE_RNG && !ENGINE_set_RAND (e, &padlock_rand))
+#endif
+	    ) {
 		return 0;
 	}
 
@@ -213,7 +261,7 @@
 static int
 padlock_init(ENGINE *e)
 {
-	return (padlock_use_rng || padlock_use_ace);
+	return (padlock_flags);
 }
 
 /* This stuff is needed if this ENGINE is being compiled into a self-contained
@@ -247,7 +295,7 @@
 #define AES_KEY_SIZE_192	24
 #define AES_KEY_SIZE_256	32
 
-/* Here we store the status information relevant to the 
+/* Here we store the status information relevant to the
    current context. */
 /* BIG FAT WARNING:
  * 	Inline assembler in PADLOCK_XCRYPT_ASM()
@@ -306,7 +354,7 @@
 {
 	int result = -1;
 
-	/* We're checking if the bit #21 of EFLAGS 
+	/* We're checking if the bit #21 of EFLAGS
 	   can be toggled. If yes = CPUID is available. */
 	asm volatile (
 		"pushf\n"
@@ -322,7 +370,7 @@
 		"xorl %%eax, %%ecx\n"
 		"movl %%ecx, %0\n"
 		: "=r" (result) : : "eax", "ecx");
-	
+
 	return (result == 0);
 }
 
@@ -365,10 +413,22 @@
 		: "+a"(eax), "=d"(edx) : : "ecx");
 
 	/* Fill up some flags */
-	padlock_use_ace = ((edx & (0x3<<6)) == (0x3<<6));
-	padlock_use_rng = ((edx & (0x3<<2)) == (0x3<<2));
+	padlock_flags |= ((edx & (0x3<<3)) ? PADLOCK_RNG : 0);
+	padlock_flags |= ((edx & (0x3<<7)) ? PADLOCK_ACE : 0);
+	padlock_flags |= ((edx & (0x3<<9)) ? PADLOCK_ACE2 : 0);
+	padlock_flags |= ((edx & (0x3<<11)) ? PADLOCK_PHE : 0);
+	padlock_flags |= ((edx & (0x3<<13)) ? PADLOCK_PMM : 0);
 
-	return padlock_use_ace + padlock_use_rng;
+	return padlock_flags;
+}
+
+static inline void
+padlock_htonl_block(uint32_t *data, size_t count)
+{
+	while (count--) {
+		asm volatile ("bswapl %0" : "+r"(*data));
+		data++;
+	}
 }
 
 #ifndef OPENSSL_NO_AES
@@ -377,17 +437,14 @@
 padlock_bswapl(AES_KEY *ks)
 {
 	size_t i = sizeof(ks->rd_key)/sizeof(ks->rd_key[0]);
-	unsigned int *key = ks->rd_key;
+	uint32_t *key = (uint32_t*) ks->rd_key;
 
-	while (i--) {
-		asm volatile ("bswapl %0" : "+r"(*key));
-		key++;
-	}
+	padlock_htonl_block(key, i);
 }
 #endif
 
 /* Force key reload from memory to the CPU microcode.
-   Loading EFLAGS from the stack clears EFLAGS[30] 
+   Loading EFLAGS from the stack clears EFLAGS[30]
    which does the trick. */
 static inline void
 padlock_reload_key(void)
@@ -423,7 +480,7 @@
 }
 
 /* Template for padlock_xcrypt_* modes */
-/* BIG FAT WARNING: 
+/* BIG FAT WARNING:
  * 	The offsets used with 'leal' instructions
  * 	describe items of the 'padlock_cipher_data'
  * 	structure.
@@ -475,7 +532,7 @@
  * In case you wonder 'rep xcrypt*' instructions above are *not*
  * affected by the Direction Flag and pointers advance toward
  * larger addresses unconditionally.
- */ 
+ */
 static inline unsigned char *
 padlock_memcpy(void *dst,const void *src,size_t n)
 {
@@ -501,7 +558,7 @@
 	_asm _emit 0x0f _asm _emit 0xa7	\
 	_asm _emit code
 
-/* BIG FAT WARNING: 
+/* BIG FAT WARNING:
  * 	The offsets used with 'lea' instructions
  * 	describe items of the 'padlock_cipher_data'
  * 	structure.
@@ -840,7 +897,7 @@
 	return 1;
 }
 
-/* 
+/*
  * Simplified version of padlock_aes_cipher() used when
  * 1) both input and output buffers are at aligned addresses.
  * or when
@@ -895,7 +952,7 @@
 # error "insane PADLOCK_CHUNK..."
 #endif
 
-/* Re-align the arguments to 16-Bytes boundaries and run the 
+/* Re-align the arguments to 16-Bytes boundaries and run the
    encryption function itself. This function is not AES-specific. */
 static int
 padlock_aes_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out_arg,
@@ -1157,6 +1214,514 @@
 
 #endif /* OPENSSL_NO_AES */
 
+#ifndef OPENSSL_NO_SHA
+
+#define DIGEST_DATA(ctx)	((struct padlock_digest_data *)(ctx->md_data))
+#define PADLOCK_SHA_ALIGN(dd)	(uint32_t*)(((uintptr_t)(dd) + 15) & ~15)
+#define PADLOCK_SHA_PAGES	14
+#define PADLOCK_SHA_BUFFER	(512 - sizeof(size_t) - 4*sizeof(void*))
+#define PADLOCK_SHA_INITVECTOR_SIZE (8 * sizeof(uint32_t))
+
+struct padlock_digest_data {
+	union {
+		unsigned char		smallbuffer[PADLOCK_SHA_BUFFER];
+		struct {
+			unsigned char	padlockctx[128+16];
+			unsigned char	*buffer;
+			size_t		mmap_size;
+			uint64_t	total;
+		};
+	};
+	void	*initvector;
+	size_t	used;
+	void	(*hash)(void *padlockctx, const void *buf, size_t len);
+	int	(*update)(EVP_MD_CTX *ctx, const void *buffer, size_t len);
+	int	(*final)(EVP_MD_CTX *ctx, unsigned char *buffer);
+};
+
+static inline void *
+padlock_atomic_xchg(volatile void **mem, void *fixed)
+{
+	/* No lock prefix due the xchg asserts it anyway, and the
+	 * funny unsigned long* cast is required to workaround some gcc
+	 * problems if compiling in PIC mode */
+	asm volatile (
+		"xchg %0, %1"
+		: "=r"(fixed)
+		: "m"(*(unsigned long*)mem), "0"(fixed)
+		: "memory");
+	return fixed;
+}
+
+static void
+padlock_do_sha1(void *padlockctx, const void *buf, size_t len)
+{
+	asm volatile (
+		"xsha1"
+		: "+S"(buf), "+D"(padlockctx)
+		: "c"(len), "a"(0));
+}
+
+static void
+padlock_do_sha256(void *padlockctx, const void *buf, size_t len)
+{
+	asm volatile (
+		"xsha256"
+		: "+S"(buf), "+D"(padlockctx)
+		: "c"(len), "a"(0));
+}
+
+static void
+handle_sigsegv(int sig, siginfo_t *info, void *uctxp)
+{
+	ucontext_t *uctx = uctxp;
+	uctx->uc_mcontext.gregs[14] += 4;
+}
+
+static void
+padlock_sha_nonfinalizing(struct padlock_digest_data *data)
+{
+	struct sigaction act, oldact;
+	size_t bofs = 0;
+
+	if (data->used != data->mmap_size) {
+		bofs = data->mmap_size - data->used;
+		memmove(&data->buffer[bofs], data->buffer, data->used);
+	}
+
+	memset(&act, 0, sizeof(act));
+	act.sa_sigaction = handle_sigsegv;
+	act.sa_flags = SA_SIGINFO;
+	sigaction(SIGSEGV, &act, &oldact);
+	data->hash(PADLOCK_SHA_ALIGN(data->padlockctx),
+		   &data->buffer[bofs], data->used + 64);
+	sigaction(SIGSEGV, &oldact, NULL);
+}
+
+static void
+padlock_free_buffer(void *buf)
+{
+	buf = padlock_atomic_xchg(&padlock_cached_sha_buffer, buf);
+	if (buf != NULL) {
+		munmap(buf, (PADLOCK_SHA_PAGES + 1) * getpagesize());
+	}
+}
+
+static void *
+padlock_allocate_buffer(size_t *maxsize)
+{
+	void *buf;
+	size_t size, page;
+
+	page = getpagesize();
+	buf = padlock_atomic_xchg(&padlock_cached_sha_buffer, NULL);
+	if (buf != NULL)
+		goto ret;
+
+	size = (PADLOCK_SHA_PAGES + 1) * page;
+	buf = mmap(0, size, PROT_READ | PROT_WRITE,
+		   MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
+	if (buf == NULL)
+		return NULL;
+
+	/* Try locking the pages to avoid swapping, but don't fail if
+	 * we are over quota. */
+	mlock(buf, size);
+
+	if (mprotect(buf + PADLOCK_SHA_PAGES * page, page, PROT_NONE) < 0) {
+		munmap(buf, size);
+		return NULL;
+	}
+
+ret:
+	*maxsize = PADLOCK_SHA_PAGES * page - 64;
+
+	return buf;
+}
+
+static int
+padlock_multi_update(EVP_MD_CTX *ctx, const void *data, size_t len)
+{
+	struct padlock_digest_data *ddata = DIGEST_DATA(ctx);
+	size_t chunk_size;
+
+	if (ddata->buffer == NULL)
+		ddata->buffer = padlock_allocate_buffer(&ddata->mmap_size);
+
+	while (len) {
+		if (ddata->used + len < ddata->mmap_size) {
+			memcpy(&ddata->buffer[ddata->used], data, len);
+			ddata->used += len;
+			ddata->total += len;
+			return 1;
+		}
+
+		chunk_size = ddata->mmap_size - ddata->used;
+		memcpy(&ddata->buffer[ddata->used], data, chunk_size);
+
+		data += chunk_size;
+		len -= chunk_size;
+		ddata->used = ddata->mmap_size;
+		ddata->total += chunk_size;
+		padlock_sha_nonfinalizing(ddata);
+		ddata->used = 0;
+	}
+
+	return 1;
+}
+
+static int
+padlock_oneshot_final(EVP_MD_CTX *ctx, unsigned char *md)
+{
+	struct padlock_digest_data *ddata = DIGEST_DATA(ctx);
+	size_t size = EVP_MD_CTX_size(ctx);
+
+	memcpy(md, PADLOCK_SHA_ALIGN(ddata->padlockctx), size);
+	return 1;
+}
+
+static int
+padlock_copy_final(EVP_MD_CTX *ctx, unsigned char *md)
+{
+	struct padlock_digest_data *ddata = DIGEST_DATA(ctx);
+	char padlockctx[128+16];
+	void *aligned = PADLOCK_SHA_ALIGN(padlockctx);
+	size_t size = EVP_MD_CTX_size(ctx);
+
+	memcpy(aligned, ddata->initvector, PADLOCK_SHA_INITVECTOR_SIZE);
+	ddata->hash(aligned, ddata->smallbuffer, ddata->used);
+	padlock_htonl_block(aligned, size / sizeof(uint32_t));
+	memcpy(md, aligned, size);
+
+	return 1;
+}
+
+static int
+padlock_multi_final(EVP_MD_CTX *ctx, unsigned char *md)
+{
+	static const char padding[64] = { 0x80, };
+	struct padlock_digest_data *ddata = DIGEST_DATA(ctx);
+	size_t mdsize = EVP_MD_CTX_size(ctx);
+	void *aligned = PADLOCK_SHA_ALIGN(ddata->padlockctx);
+
+	if (ddata->used == ddata->total) {
+		/* Sweet, everything fits in one buffer. */
+		ddata->hash(aligned, ddata->buffer, ddata->used);
+	} else {
+		/* Hardware already hashed some buffers.
+		 * Do finalizing manually */
+		union {
+			uint64_t u64;
+			uint32_t u32[2];
+		} bits_le, bits;
+		size_t lastblocklen, padlen;
+
+		/* BigEndianise the length. */
+		bits_le.u64 = ddata->total * 8;
+		bits.u32[1] = htonl(bits_le.u32[0]);
+		bits.u32[0] = htonl(bits_le.u32[1]);
+
+		/* Append padding, leave space for length. */
+		lastblocklen = ddata->total & 63;
+		padlen = (lastblocklen < 56) ? (56 - lastblocklen) : ((64+56) - lastblocklen);
+		padlock_multi_update(ctx, padding, padlen);
+
+		/* Length in BigEndian64 */
+		padlock_multi_update(ctx, (const char *) &bits, sizeof(bits));
+
+		/* And finally calculate it */
+		padlock_sha_nonfinalizing(ddata);
+	}
+	padlock_htonl_block(aligned, mdsize / sizeof(uint32_t));
+	memcpy(md, aligned, mdsize);
+
+	return 1;
+}
+
+static int
+padlock_copy_update(EVP_MD_CTX *ctx, const void *data, size_t len)
+{
+	struct padlock_digest_data *ddata = DIGEST_DATA(ctx);
+
+	if (ddata->used + len > sizeof(ddata->smallbuffer)) {
+		ddata->update = padlock_multi_update;
+		ddata->final  = padlock_multi_final;
+
+		if (ddata->used != 0) {
+			void *buffer;
+			size_t mmap_size;
+
+			buffer = padlock_allocate_buffer(&mmap_size);
+			memcpy(buffer, ddata->smallbuffer, ddata->used);
+			ddata->buffer = buffer;
+			ddata->total = ddata->used;
+			ddata->mmap_size = mmap_size;
+		} else {
+			ddata->buffer = NULL;
+			ddata->total = 0;
+		}
+
+		memcpy(PADLOCK_SHA_ALIGN(ddata->padlockctx), ddata->initvector,
+		       PADLOCK_SHA_INITVECTOR_SIZE);
+
+		return padlock_multi_update(ctx, data, len);
+	}
+
+	memcpy(&ddata->smallbuffer[ddata->used], data, len);
+	ddata->used += len;
+
+	return 1;
+}
+
+static int
+padlock_oneshot_update(EVP_MD_CTX *ctx, const void *data, size_t len)
+{
+	struct padlock_digest_data *ddata = DIGEST_DATA(ctx);
+	void *aligned = PADLOCK_SHA_ALIGN(ddata->padlockctx);
+	size_t mdsize = EVP_MD_CTX_size(ctx);
+
+	/* Oneshot update is only possible if context flags indicate so */
+	if (!(ctx->flags & EVP_MD_CTX_FLAG_ONESHOT)) {
+		ddata->update = padlock_copy_update;
+		ddata->final  = padlock_copy_final;
+		return padlock_copy_update(ctx, data, len);
+	}
+
+	memcpy(aligned, ddata->initvector, PADLOCK_SHA_INITVECTOR_SIZE);
+	ddata->hash(aligned, data, len);
+	padlock_htonl_block(aligned, mdsize / sizeof(uint32_t));
+	ddata->used += len;
+
+	return 1;
+}
+
+static int
+padlock_sha_init(struct padlock_digest_data *ddata)
+{
+	ddata->used      = 0;
+	ddata->update    = padlock_oneshot_update;
+	ddata->final     = padlock_oneshot_final;
+
+	return 1;
+}
+
+static int
+padlock_sha1_init(EVP_MD_CTX *ctx)
+{
+	static uint32_t sha1_initvector[8] = {
+		0x67452301, 0xEFCDAB89, 0x98BADCFE, 0x10325476,
+		0xC3D2E1F0
+	};
+	struct padlock_digest_data *ddata = DIGEST_DATA(ctx);
+
+	ddata->hash = padlock_do_sha1;
+	ddata->initvector = sha1_initvector;
+	return padlock_sha_init(ddata);
+}
+
+static int
+padlock_sha224_init(EVP_MD_CTX *ctx)
+{
+	static uint32_t sha224_initvector[] = {
+		0xC1059ED8, 0x367CD507, 0x3070DD17, 0xF70E5939,
+		0xFFC00B31, 0x68581511, 0x64F98FA7, 0xBEFA4FA4,
+	};
+	struct padlock_digest_data *ddata = DIGEST_DATA(ctx);
+
+	ddata->hash = padlock_do_sha256;
+	ddata->initvector = sha224_initvector;
+	return padlock_sha_init(ddata);
+}
+
+static int
+padlock_sha256_init(EVP_MD_CTX *ctx)
+{
+	static uint32_t sha256_initvector[] = {
+		0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A,
+		0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19
+	};
+	struct padlock_digest_data *ddata = DIGEST_DATA(ctx);
+
+	ddata->hash = padlock_do_sha256;
+	ddata->initvector = sha256_initvector;
+	return padlock_sha_init(ddata);
+}
+
+static int
+padlock_sha_update(EVP_MD_CTX *ctx, const void *data, size_t length)
+{
+	return DIGEST_DATA(ctx)->update(ctx, data, length);
+}
+
+static int
+padlock_sha_final(EVP_MD_CTX *ctx, unsigned char *md)
+{
+	return DIGEST_DATA(ctx)->final(ctx, md);
+}
+
+static int
+padlock_sha_copy(EVP_MD_CTX *to, const EVP_MD_CTX *from)
+{
+	struct padlock_digest_data *dfrom = DIGEST_DATA(from);
+	struct padlock_digest_data *dto = DIGEST_DATA(to);
+
+	/* When we get here, dto is already a memcpied from dfrom,
+	 * it's ok for all other cases except when data is on a separate
+	 * mmapped area. It would be nice if we had a flag, if this is
+	 * a "finalization copy", so we could do finalizing SHA here and
+	 * store the result to *to precalculated. But there's no such
+	 * flag as to is reset on copy. */
+
+	if (dfrom->update != padlock_copy_update) {
+		/* Recopy the context, as they might have different alignment */
+		memcpy(PADLOCK_SHA_ALIGN(dto->padlockctx),
+		       PADLOCK_SHA_ALIGN(dfrom->padlockctx),
+		       PADLOCK_SHA_INITVECTOR_SIZE);
+	}
+
+	if (dfrom->update == padlock_multi_update) {
+		/* Update total, and copy the buffer */
+		dto->total     = dfrom->total - dfrom->used;
+		dto->buffer    = NULL;
+		dto->used      = 0;
+		dto->mmap_size = 0;
+		if (dfrom->used != 0)
+			padlock_sha_update(to, dfrom->buffer, dfrom->used);
+	}
+
+	return 1;
+}
+
+static int
+padlock_sha_cleanup(EVP_MD_CTX *ctx)
+{
+	struct padlock_digest_data *ddata = DIGEST_DATA(ctx);
+
+	if (ddata->update == padlock_multi_update && ddata->buffer != NULL)
+		padlock_free_buffer(ddata->buffer);
+
+	return 1;
+}
+
+static const EVP_MD padlock_sha1_md = {
+	NID_sha1,
+	NID_sha1WithRSAEncryption,
+	SHA_DIGEST_LENGTH,
+	0,
+	padlock_sha1_init,
+	padlock_sha_update,
+	padlock_sha_final,
+	padlock_sha_copy,
+	padlock_sha_cleanup,
+	EVP_PKEY_RSA_method,
+	SHA_CBLOCK,
+	sizeof(struct padlock_digest_data),
+};
+
+static const EVP_MD padlock_dss1_md = {
+	NID_dsa,
+	NID_dsaWithSHA1,
+	SHA_DIGEST_LENGTH,
+	0,
+	padlock_sha1_init,
+	padlock_sha_update,
+	padlock_sha_final,
+	padlock_sha_copy,
+	padlock_sha_cleanup,
+	EVP_PKEY_DSA_method,
+	SHA_CBLOCK,
+	sizeof(struct padlock_digest_data),
+};
+
+static const EVP_MD padlock_sha224_md = {
+	NID_sha224,
+	NID_sha224WithRSAEncryption,
+	SHA224_DIGEST_LENGTH,
+	0,
+	padlock_sha224_init,
+	padlock_sha_update,
+	padlock_sha_final,
+	padlock_sha_copy,
+	padlock_sha_cleanup,
+	EVP_PKEY_RSA_method,
+	SHA_CBLOCK,
+	sizeof(struct padlock_digest_data),
+};
+
+static const EVP_MD padlock_sha256_md = {
+	NID_sha256,
+	NID_sha256WithRSAEncryption,
+	SHA256_DIGEST_LENGTH,
+	0,
+	padlock_sha256_init,
+	padlock_sha_update,
+	padlock_sha_final,
+	padlock_sha_copy,
+	padlock_sha_cleanup,
+	EVP_PKEY_RSA_method,
+	SHA_CBLOCK,
+	sizeof(struct padlock_digest_data),
+};
+
+static int padlock_digest_nids[] = {
+#if !defined(OPENSSL_NO_SHA)
+	NID_sha1,
+	NID_dsa,
+#endif
+#if !defined(OPENSSL_NO_SHA256)
+#if !defined(OPENSSL_NO_SHA224)
+	NID_sha224,
+#endif
+	NID_sha256,
+#endif
+};
+
+static int padlock_digest_nids_num = sizeof(padlock_digest_nids)/sizeof(padlock_digest_nids[0]);
+
+static int
+padlock_digests (ENGINE *e, const EVP_MD **digest, const int **nids, int nid)
+{
+	/* No specific digest => return a list of supported nids ... */
+	if (!digest) {
+		*nids = padlock_digest_nids;
+		return padlock_digest_nids_num;
+	}
+
+	/* ... or the requested "digest" otherwise */
+	switch (nid) {
+#if !defined(OPENSSL_NO_SHA)
+	  case NID_sha1:
+	    *digest = &padlock_sha1_md;
+	    break;
+	  case NID_dsa:
+	    *digest = &padlock_dss1_md;
+	    break;
+#endif
+
+#if !defined(OPENSSL_NO_SHA256)
+#if !defined(OPENSSL_NO_SHA224)
+	  case NID_sha224:
+	    *digest = &padlock_sha224_md;
+	    break;
+#endif	/* OPENSSL_NO_SHA224 */
+
+	  case NID_sha256:
+	    *digest = &padlock_sha256_md;
+	    break;
+#endif	/* OPENSSL_NO_SHA256 */
+
+	  default:
+	    /* Sorry, we don't support this NID */
+	    *digest = NULL;
+	    return 0;
+	}
+
+	return 1;
+}
+
+#endif /* OPENSSL_NO_SHA */
+
+#ifndef PADLOCK_NO_RNG
 /* ===== Random Number Generator ===== */
 /*
  * This code is not engaged. The reason is that it does not comply
@@ -1164,7 +1729,7 @@
  * (posted at http://www.via.com.tw/en/viac3/c3.jsp) nor does it
  * provide meaningful error control...
  */
-/* Wrapper that provides an interface between the API and 
+/* Wrapper that provides an interface between the API and
    the raw PadLock RNG */
 static int
 padlock_rand_bytes(unsigned char *output, int count)
@@ -1212,6 +1777,7 @@
 	padlock_rand_bytes,	/* pseudorand */
 	padlock_rand_status,	/* rand status */
 };
+#endif /* PADLOCK_NO_RNG */
 
 #endif /* COMPILE_HW_PADLOCK */