Deal with padding in one place, in blockwise.

We introduce two functions, to assist with processing
sequences of fixed bytes.  One processes a single byte a bunch of
times, the other does something more complicated.

We use this for all hashes, CMAC and CBCMAC.  This gives a good
performance improvement.

Author: ctz

src/blockwise.c

@@ -126,3 +126,70 @@ void cf_blockwise_xor(uint8_t *partial, size_t *npartial, size_t nblock,
     inb += taken;
   }
 }
+
+void cf_blockwise_acc_byte(uint8_t *partial, size_t *npartial,
+                           size_t nblock,
+                           uint8_t byte, size_t nbytes,
+                           cf_blockwise_in_fn process,
+                           void *ctx)
+{
+  /* only memset the whole of the block once */
+  int filled = 0;
+
+  while (nbytes)
+  {
+    size_t start = *npartial;
+    size_t count = MIN(nbytes, nblock - start);
+
+    if (!filled)
+      memset(partial + start, byte, count);
+
+    if (start == 0 && count == nblock)
+      filled = 1;
+
+    if (start + count == nblock)
+    {
+      process(ctx, partial);
+      *npartial = 0;
+    } else {
+      *npartial += count;
+    }
+
+    nbytes -= count;
+  }
+}
+
+void cf_blockwise_acc_pad(uint8_t *partial, size_t *npartial,
+                          size_t nblock,
+                          uint8_t fbyte, uint8_t mbyte, uint8_t lbyte,
+                          size_t nbytes,
+                          cf_blockwise_in_fn process,
+                          void *ctx)
+{
+
+  switch (nbytes)
+  {
+    case 0: break;
+    case 1: fbyte ^= lbyte;
+            cf_blockwise_accumulate(partial, npartial, nblock, &fbyte, 1, process, ctx);
+            break;
+    case 2:
+            cf_blockwise_accumulate(partial, npartial, nblock, &fbyte, 1, process, ctx);
+            cf_blockwise_accumulate(partial, npartial, nblock, &lbyte, 1, process, ctx);
+            break;
+    default:
+            cf_blockwise_accumulate(partial, npartial, nblock, &fbyte, 1, process, ctx);
+
+            /* If the middle and last bytes differ, then process the last byte separately.
+             * Otherwise, just extend the middle block size. */
+            if (lbyte != mbyte)
+            {
+              cf_blockwise_acc_byte(partial, npartial, nblock, mbyte, nbytes - 2, process, ctx);
+              cf_blockwise_accumulate(partial, npartial, nblock, &lbyte, 1, process, ctx);
+            } else {
+              cf_blockwise_acc_byte(partial, npartial, nblock, mbyte, nbytes - 1, process, ctx);
+            }
+
+            break;
+  }
+}

src/blockwise.h

@@ -92,4 +92,56 @@ void cf_blockwise_xor(uint8_t *partial, size_t *npartial,
                       cf_blockwise_out_fn newblock,
                       void *ctx);
 
+/* This function processes a single byte a number of times. It's useful
+ * for padding, and more efficient than calling cf_blockwise_accumulate
+ * a bunch of times.
+ *
+ * partial is the buffer (maintained by the caller)
+ * on entry, npartial is the currently valid count of used bytes on
+ *   the front of partial.
+ * on exit, npartial is updated to reflect the status of partial.
+ * nblock is the blocksize to accumulate -- partial must be at least
+ *   this long!
+ * process is the processing function, passed ctx and a pointer
+ *   to the data to process (always exactly nblock bytes long!)
+ *   which may not neccessarily be the same as partial.
+ * byte is the byte to process, nbytes times.
+ */
+void cf_blockwise_acc_byte(uint8_t *partial, size_t *npartial,
+                           size_t nblock,
+                           uint8_t byte, size_t nbytes,
+                           cf_blockwise_in_fn process,
+                           void *ctx);
+
+/* This function attempts to process patterns of bytes common in
+ * block cipher padding.
+ *
+ * This takes three bytes:
+ * - a first byte, fbyte,
+ * - a middle byte, mbyte,
+ * - a last byte, lbyte.
+ *
+ * If nbytes is zero, nothing happens.
+ * If nbytes is one, the byte fbyte ^ lbyte is processed.
+ * If nbytes is two, the fbyte then lbyte are processed.
+ * If nbytes is three or more, fbyte, then one or more mbytes, then fbyte
+ *   is processed.
+ *
+ * partial is the buffer (maintained by the caller)
+ * on entry, npartial is the currently valid count of used bytes on
+ *   the front of partial.
+ * on exit, npartial is updated to reflect the status of partial.
+ * nblock is the blocksize to accumulate -- partial must be at least
+ *   this long!
+ * process is the processing function, passed ctx and a pointer
+ *   to the data to process (always exactly nblock bytes long!)
+ *   which may not neccessarily be the same as partial.
+ */
+void cf_blockwise_acc_pad(uint8_t *partial, size_t *npartial,
+                          size_t nblock,
+                          uint8_t fbyte, uint8_t mbyte, uint8_t lbyte,
+                          size_t nbytes,
+                          cf_blockwise_in_fn process,
+                          void *ctx);
+
 #endif

src/cbcmac.c

@@ -53,6 +53,16 @@ void cf_cbcmac_stream_update(cf_cbcmac_stream *ctx, const uint8_t *data, size_t
                           ctx);
 }
 
+void cf_cbcmac_stream_finish_block_zero(cf_cbcmac_stream *ctx)
+{
+  if (ctx->used == 0)
+    return;
+
+  memset(ctx->buffer + ctx->used, 0, ctx->prp->blocksz - ctx->used);
+  cbcmac_process(ctx, ctx->buffer);
+  ctx->used = 0;
+}
+
 void cf_cbcmac_stream_nopad_final(cf_cbcmac_stream *ctx, uint8_t out[CF_MAXBLOCK])
 {
   assert(ctx->used == 0);
@@ -62,7 +72,8 @@ void cf_cbcmac_stream_nopad_final(cf_cbcmac_stream *ctx, uint8_t out[CF_MAXBLOCK
 void cf_cbcmac_stream_pad_final(cf_cbcmac_stream *ctx, uint8_t out[CF_MAXBLOCK])
 {
   uint8_t npad = ctx->prp->blocksz - ctx->used;
-  for (size_t i = 0; i < npad; i++)
-    cf_cbcmac_stream_update(ctx, &npad, 1);
+  cf_blockwise_acc_byte(ctx->buffer, &ctx->used, ctx->prp->blocksz,
+                        npad, npad,
+                        cbcmac_process, ctx);
   cf_cbcmac_stream_nopad_final(ctx, out);
 }

src/ccm.c

@@ -35,9 +35,7 @@ static void write_be(uint8_t *out, size_t value, size_t bytes)
 
 static void zero_pad(cf_cbcmac_stream *cm)
 {
-  const uint8_t zero_byte = 0;
-  while (cm->used != 0)
-    cf_cbcmac_stream_update(cm, &zero_byte, 1);
+  cf_cbcmac_stream_finish_block_zero(cm);
 }
 
 /* nb. block is general workspace. */

src/cmac.c

@@ -132,10 +132,9 @@ void cf_cmac_stream_update(cf_cmac_stream *ctx, const uint8_t *data, size_t len,
   /* Input padding */
   if (needpad)
   {
-    uint8_t pad_block[CF_MAXBLOCK] = { 0x80 };
-    cf_blockwise_accumulate(ctx->buffer, &ctx->used, blocksz,
-                            pad_block, blocksz - ctx->used,
-                            cmac_process_final_pad, ctx);
+    cf_blockwise_acc_pad(ctx->buffer, &ctx->used, blocksz,
+                         0x80, 0x00, 0x00, blocksz - ctx->used,
+                         cmac_process_final_pad, ctx);
   }
 }
 

src/gcm.c

@@ -56,17 +56,20 @@ static void ghash_block(void *vctx, const uint8_t *data)
 static void ghash_add(ghash_ctx *ctx, const uint8_t *buf, size_t n)
 {
   cf_blockwise_accumulate(ctx->buffer, &ctx->buffer_used,
-                          16,
+                          sizeof ctx->buffer,
                           buf, n,
                           ghash_block,
                           ctx);
 }
 
 static void ghash_add_pad(ghash_ctx *ctx)
 {
-  uint8_t byte = 0x00;
-  while (ctx->buffer_used != 0)
-    ghash_add(ctx, &byte, 1);
+  if (ctx->buffer_used == 0)
+    return;
+
+  memset(ctx->buffer + ctx->buffer_used, 0, sizeof(ctx->buffer) - ctx->buffer_used);
+  ghash_block(ctx, ctx->buffer);
+  ctx->buffer_used = 0;
 }
 
 static void ghash_add_aad(ghash_ctx *ctx, const uint8_t *buf, size_t n)

src/modes.h

@@ -194,6 +194,11 @@ void cf_cbcmac_stream_reset(cf_cbcmac_stream *ctx);
  * Process ndata bytes at data. */
 void cf_cbcmac_stream_update(cf_cbcmac_stream *ctx, const uint8_t *data, size_t ndata);
 
+/* .. c:function:: $DECL
+ * Finish the current block of data by adding zeroes.  Does nothing if there
+ * are no bytes awaiting processing. */
+void cf_cbcmac_stream_finish_block_zero(cf_cbcmac_stream *ctx);
+
 /* .. c:function:: $DECL
  * Output the MAC to ctx->prp->blocksz bytes at out.
  * ctx->used must be zero: the inputed message must be an exact number of

src/sha1.c

@@ -116,18 +116,15 @@ void cf_sha1_digest_final(cf_sha1_context *ctx, uint8_t hash[CF_SHA1_HASHSZ])
   digested_bytes = digested_bytes * CF_SHA1_BLOCKSZ + ctx->npartial;
   uint64_t digested_bits = digested_bytes * 8;
 
-  size_t zeroes = CF_SHA1_BLOCKSZ - ((digested_bytes + 1 + 8) % CF_SHA1_BLOCKSZ);
+  size_t padbytes = CF_SHA1_BLOCKSZ - ((digested_bytes + 8) % CF_SHA1_BLOCKSZ);
 
   /* Hash 0x80 00 ... block first. */
-  uint8_t buf[8];
-  buf[0] = 0x80;
-  buf[1] = 0x00;
-  cf_sha1_update(ctx, &buf[0], 1);
-
-  while (zeroes--)
-    cf_sha1_update(ctx, &buf[1], 1);
+  cf_blockwise_acc_pad(ctx->partial, &ctx->npartial, sizeof ctx->partial,
+                       0x80, 0x00, 0x00, padbytes,
+                       sha1_update_block, ctx);
 
   /* Now hash length. */
+  uint8_t buf[8];
   write64_be(digested_bits, buf);
   cf_sha1_update(ctx, buf, 8);
 

src/sha256.c

@@ -172,18 +172,15 @@ void cf_sha256_digest_final(cf_sha256_context *ctx, uint8_t hash[CF_SHA256_HASHS
   digested_bytes = digested_bytes * CF_SHA256_BLOCKSZ + ctx->npartial;
   uint64_t digested_bits = digested_bytes * 8;
 
-  size_t zeroes = CF_SHA256_BLOCKSZ - ((digested_bytes + 1 + 8) % CF_SHA256_BLOCKSZ);
+  size_t padbytes = CF_SHA256_BLOCKSZ - ((digested_bytes + 8) % CF_SHA256_BLOCKSZ);
 
   /* Hash 0x80 00 ... block first. */
-  uint8_t buf[8];
-  buf[0] = 0x80;
-  buf[1] = 0x00;
-  cf_sha256_update(ctx, &buf[0], 1);
-
-  while (zeroes--)
-    cf_sha256_update(ctx, &buf[1], 1);
+  cf_blockwise_acc_pad(ctx->partial, &ctx->npartial, sizeof ctx->partial,
+                       0x80, 0x00, 0x00, padbytes,
+                       sha256_update_block, ctx);
 
   /* Now hash length. */
+  uint8_t buf[8];
   write64_be(digested_bits, buf);
   cf_sha256_update(ctx, buf, 8);
 

src/sha3.c

@@ -311,21 +311,12 @@ static void sha3_update(cf_sha3_context *ctx, const void *data, size_t nbytes)
 
 static void pad(cf_sha3_context *ctx, uint8_t domain, size_t npad)
 {
-  uint8_t padding[CF_SHA3_224_BLOCKSZ];
-
   assert(npad >= 1);
 
-  if (npad == 1)
-  {
-    padding[0] = domain | 0x80;
-    sha3_update(ctx, padding, 1);
-    return;
-  }
-
-  memset(padding, 0, npad);
-  padding[0] = domain;
-  padding[npad - 1] = 0x80;
-  sha3_update(ctx, padding, npad);
+  cf_blockwise_acc_pad(ctx->partial, &ctx->npartial, ctx->rate,
+                       domain, 0x00, 0x80,
+                       npad,
+                       sha3_block, ctx);
 }
 
 static void pad_and_squeeze(cf_sha3_context *ctx, uint8_t *out, size_t nout)