Changement de la concaténation des chaînes de bits

L'implémentation précédente n'était pas  cohérente. Étant données deux
chaînes X et Y de longueur x et y, et Z=X∥Y de longueur z=x+y,

- pad10* et la construction des tweaks fonctionnaient selon la logique
  "indices faibles = LSB", donc

    Z[0] = Y[0]
    Z[z-1] = X[x-1]

- le découpage  de M, C  et A en  blocs fonctionnait selon  la logique
  "indices faibles = premiers blocs", donc

    Z[0]     = X[0]
    Z[z-1] = Y[y-1]

En conséquence, la  façon dont M, C et A  étaient paddés n'avait aucun
sens, e.g. pour un message M de taille 35, pad10*(M*) donnait :

    { M[34], M[33], M[32], 0b10000000, 0, … }

Les deux seules façons logiques de padder M* sont

    { M[32], M[33], M[34], 0b10000000, 0, … }

ou

    { M[2], M[1], M[0], 0b10000000, 0, … }

Après revue d'autres implémentations de  ΘCB3 et SCT-2, j'ai choisi de
suivre la convention MSB. En conséquence, quand la spécification dit

    Z = X∥Y

L'implémentation traduira :

    Z[] = { X[0], … X[x-1], Y[0], … Y[y-1] }

Dans  la même  logique,  les  compteurs de  blocs  seront insérés  MSB
d'abord et paddés en conséquence, e.g.

    j=0x01020304 ≡ J[] = { 0, …, 0x01, 0x02, 0x03, 0x04 }

3 files changed, 87 insertions, 82 deletions

diff --git a/src/ref/lilliput-ae-utils.h b/src/ref/lilliput-ae-utils.h
index 0efb776..a66b75c 100644
--- a/src/ref/lilliput-ae-utils.h
+++ b/src/ref/lilliput-ae-utils.h
@@ -28,16 +28,6 @@ This file provides functions used by both authenticated encryption modes.
 #include "constants.h"
 
 
-static inline uint8_t upper_nibble(uint8_t i)
-{
-    return i >> 4;
-}
-
-static inline uint8_t lower_nibble(uint8_t i)
-{
-    return i & 0x0f;
-}
-
 static inline void encrypt(const uint8_t K[KEY_BYTES],
                            const uint8_t T[TWEAK_BYTES],
                            const uint8_t M[BLOCK_BYTES],
@@ -68,35 +58,41 @@ static inline void xor_arrays(size_t len, uint8_t out[len], const uint8_t a[len]
 
 static inline void pad10(size_t X_len, const uint8_t X[X_len], uint8_t padded[BLOCK_BYTES])
 {
-    /* pad10*(X) = X || 1 || 0^{n-|X|-1} */
-
-    /* For example, with uint8_t X[3] = { [0]=0x01, [1]=0x02, [2]=0x03 }
+    /* Assuming 0 < |X| < n:
+     *
+     * pad10*(X) = X || 1 || 0^{n-|X|-1}
+     *
+     * For example, with uint8_t X[3] = { [0]=0x01, [1]=0x02, [2]=0x03 }
      *
      * pad10*(X) =
-     *       X[2]     X[1]     X[0]   1 0*
-     *     00000011 00000010 00000001 1 0000000 00000000...
+     *       X[0]     X[1]     X[2]   1 0*
+     *     00000001 00000010 00000011 1 0000000 00000000...
      *
-     * - padded[0, 11]:  zeroes
-     * - padded[12]:     10000000
-     * - padded[13, 15]: X[0, 2]
+     * - padded[0, 2]:  X[0, 2]
+     * - padded[3]:     10000000
+     * - padded[4, 15]: zeroes
      */
 
-    /* Assume that X_len<BLOCK_BYTES. */
+    memcpy(padded, X, X_len);
+    padded[X_len] = 0x80;
 
-    size_t pad_len = BLOCK_BYTES-X_len;
+    /* memset(&padded[BLOCK_BYTES], 0, 0) may or may not constitute
+     * undefined behaviour; use a straight loop instead. */
 
-    memset(padded, 0, pad_len-1);
-    padded[pad_len-1] = 0x80;
-    memcpy(padded+pad_len, X, X_len);
+    for (size_t i=X_len+1; i<BLOCK_BYTES; i++)
+    {
+        padded[i] = 0;
+    }
 }
 
 static inline void copy_block_index(size_t index, uint8_t tweak[TWEAK_BYTES])
 {
-    /* NB: little-endian architectures can simply use:
-     *     memcpy(tweak, &index, sizeof(index)); */
-    for (size_t i=0; i<sizeof(index); i++)
+    size_t s = sizeof(index);
+    uint8_t *dest = &tweak[TWEAK_BYTES-s];
+
+    for (size_t i=0; i<s; i++)
     {
-        tweak[i] = index >> 8*i & 0xff;
+        dest[i] = index >> 8*(s-1-i);
     }
 }
 
@@ -108,17 +104,17 @@ static inline void fill_index_tweak(
 {
     /* With an s-bit block index, the t-bit tweak is filled as follows:
      *
-     * - bits [  1, t-4]: block index
-     *        [  1,   s]: actual block index
-     *        [s+1, t-4]: 0-padding
-     * - bits [t-3,   t]: 4-bit prefix
+     * [     1,   4]: 4-bit prefix
+     * [     5,   t]: block index
+     *  [    5, t-s]: 0-padding
+     *  [t-s+1,   t]: actual block index, from MSB to LSB
      */
 
-    copy_block_index(block_index, tweak);
+    tweak[0] = prefix<<4;
 
     /* Assume padding bytes have already been set to 0. */
 
-    tweak[TWEAK_BYTES-1] |= prefix << 4;
+    copy_block_index(block_index, tweak);
 }
 
 static void process_associated_data(
diff --git a/src/ref/lilliput-i.c b/src/ref/lilliput-i.c
index 6f869c3..fb06237 100644
--- a/src/ref/lilliput-i.c
+++ b/src/ref/lilliput-i.c
@@ -32,58 +32,67 @@ static const uint8_t _0n[BLOCK_BYTES] = {
 };
 
 
+static uint8_t _upper_nibble(uint8_t i)
+{
+    return i >> 4;
+}
+
+static uint8_t _lower_nibble(uint8_t i)
+{
+    return i & 0x0f;
+}
+
 static void _init_msg_tweak(const uint8_t N[NONCE_BYTES], uint8_t tweak[TWEAK_BYTES])
 {
     /* With an s-bit block index, the t-bit tweak is filled as follows:
      *
-     * - bits [      1, t-|N|-4]: block index
-     *        [      1,       s]: actual block index
-     *        [    s+1, t-|N|-4]: 0-padding
-     * - bits [t-|N|-3,     t-4]: nonce
-     * - bits [    t-3,       t]: 4-bit prefix
+     * [     1,     4]: 4-bit prefix
+     * [     5, |N|+4]: nonce
+     * [ |N|+5,     t]: block index
+     *  [|N|+5,   t-s]: 0-padding
+     *  [t-s+1,     t]: actual block index, from MSB to LSB
      *
-     * This function sets bits s+1 to t-4 once and for all.
+     * This function sets bits 5 to t-s once and for all.
      */
 
-    size_t N_start = TWEAK_BYTES - NONCE_BYTES - 1;
-
-    for (size_t i=sizeof(size_t); i<N_start; i++)
-    {
-        tweak[i] = 0;
-    }
-
-    tweak[N_start] = lower_nibble(N[0]) << 4;
+    tweak[0] = _upper_nibble(N[0]);
 
     for (size_t i=1; i<NONCE_BYTES; i++)
     {
-        tweak[N_start+i] = lower_nibble(N[i]) << 4 ^ upper_nibble(N[i-1]);
+        tweak[i] = _lower_nibble(N[i-1]) << 4 ^ _upper_nibble(N[i]);
     }
 
-    tweak[TWEAK_BYTES-1] = upper_nibble(N[NONCE_BYTES-1]);
+    tweak[NONCE_BYTES] = _lower_nibble(N[NONCE_BYTES-1]) << 4;
+
+    /* The number of bits we need to zero out is:
+     *     t - |N| - s - 4        - 4
+     *                   (prefix)   (zeroed out by previous assignment)
+     */
+    memset(&tweak[NONCE_BYTES+1], 0, TWEAK_BYTES-NONCE_BYTES-sizeof(size_t)-1);
 }
 
 static void _fill_msg_tweak(
-    uint8_t       prefix,
-    size_t        block_index,
-    uint8_t       tweak[TWEAK_BYTES]
+    uint8_t prefix,
+    size_t  block_index,
+    uint8_t tweak[TWEAK_BYTES]
 )
 {
     /* With an s-bit block index, the t-bit tweak is filled as follows:
      *
-     * - bits [      1, t-|N|-4]: block index
-     *        [      1,       s]: actual block index
-     *        [    s+1, t-|N|-4]: 0-padding
-     * - bits [t-|N|-3,     t-4]: nonce
-     * - bits [    t-3,       t]: 4-bit prefix
+     * [     1,     4]: 4-bit prefix
+     * [     5, |N|+4]: nonce
+     * [ |N|+5,     t]: block index
+     *  [|N|+5,   t-s]: 0-padding
+     *  [t-s+1,     t]: actual block index, from MSB to LSB
      *
-     * This function assumes bits s+1 to t-3 have already been set,
-     * and only sets bits 1 to s and t-3 to t.
+     * This function assumes bits 5 to t-s have already been set, and
+     * only sets bits 1 to 4 and t-s+1 to t.
      */
 
-    copy_block_index(block_index, tweak);
+    uint8_t *msb = &tweak[0];
+    *msb = prefix<<4 ^ _lower_nibble(*msb);
 
-    uint8_t *msb = &tweak[TWEAK_BYTES-1];
-    *msb = prefix<<4 ^ lower_nibble(*msb);
+    copy_block_index(block_index, tweak);
 }
 
 static void _encrypt_message(
diff --git a/src/ref/lilliput-ii.c b/src/ref/lilliput-ii.c
index 6811d49..9ed17a2 100644
--- a/src/ref/lilliput-ii.c
+++ b/src/ref/lilliput-ii.c
@@ -30,34 +30,34 @@ static void _init_msg_tweak(const uint8_t tag[TAG_BYTES], uint8_t tweak[TWEAK_BY
 {
     /* With an s-bit block index, the t-bit tweak is filled as follows:
      *
-     * - bits [  1, t-1]: tag + block index
-     *        [  1,   s]: tag[1..s] XOR block index
-     *        [s+1, t-1]: tag[s+1..t-1]
-     * - bit t: 1
+     * 1: 1
+     * [     2,   t]: tag[    2,   t] XOR block index
+     *  [    2, t-s]: tag[    2, t-s]
+     *  [t-s+1,   t]: tag[t-s+1,   t] XOR block index
      *
-     * This function sets bits s+1 to t once and for all.
+     * This function sets bits 1 to t-s once and for all.
      */
 
-    memcpy(tweak+sizeof(size_t), tag+sizeof(size_t), TAG_BYTES-sizeof(size_t));
-    tweak[TWEAK_BYTES-1] |= 0x80;
+    memcpy(tweak, tag, TAG_BYTES-sizeof(size_t));
+    tweak[0] |= 0x80;
 }
 
 static void _fill_msg_tweak(const uint8_t tag[TAG_BYTES], size_t block_index, uint8_t tweak[TWEAK_BYTES])
 {
     /* With an s-bit block index, the t-bit tweak is filled as follows:
      *
-     * - bits [  1, t-1]: tag + block index
-     *        [  1,   s]: tag[1..s] XOR block index
-     *        [s+1, t-1]: tag[s+1..t-1]
-     * - bit t: 1
+     * 1: 1
+     * [     2,   t]: tag + block index
+     *  [    2, t-s]: tag[    2, t-s]
+     *  [t-s+1,   t]: tag[t-s+1,   t] XOR block index
      *
-     * This function assumes bits s+1 to t have already been set, and
-     * only sets bits 1 to s.
+     * This function assumes bits 1 to t-s have already been set, and
+     * only sets bits t-s+1 to t.
      */
 
     copy_block_index(block_index, tweak);
 
-    for (size_t i=0; i<sizeof(block_index); i++)
+    for (size_t i=TWEAK_BYTES-sizeof(size_t); i<TWEAK_BYTES; i++)
     {
         tweak[i] ^= tag[i];
     }
@@ -67,12 +67,12 @@ static void _fill_tag_tweak(const uint8_t N[NONCE_BYTES], uint8_t tweak[TWEAK_BY
 {
     /* The t-bit tweak is filled as follows:
      *
-     * - bits [  1, t-7]: N
-     * - bits [t-7,   t]: 0001||0^4
+     * [      1, 8]: 0001||0^4
+     * [t-|N|+1, t]: N
      */
 
-    memcpy(tweak, N, TWEAK_BYTES-1);
-    tweak[TWEAK_BYTES-1] = 0x10;
+    tweak[0] = 0x10;
+    memcpy(&tweak[1], N, TWEAK_BYTES-1);
 }
 
 static void _generate_tag(
@@ -129,8 +129,8 @@ static void _encrypt_message(
     _init_msg_tweak(tag, tweak);
 
     uint8_t padded_N[BLOCK_BYTES];
-    memcpy(padded_N, N, NONCE_BYTES);
-    padded_N[BLOCK_BYTES-1] = 0;
+    padded_N[0] = 0;
+    memcpy(&padded_N[1], N, NONCE_BYTES);
 
     size_t l = M_len / BLOCK_BYTES;
     size_t rest = M_len % BLOCK_BYTES;