aes192.asm: bugfix

No more buffer overruns.

2 files changed, 27 insertions, 28 deletions

diff --git a/include/aesni/data.h b/include/aesni/data.h
index b44b1d6..0cee7c6 100644
--- a/include/aesni/data.h
+++ b/include/aesni/data.h
@@ -57,7 +57,6 @@ Aes128KeySchedule;
 typedef struct
 {
     AesBlock128 keys[13];
-    char fillers[8];
 }
 Aes192KeySchedule;
 
diff --git a/src/aes192.asm b/src/aes192.asm
index f3a56a8..6851ae1 100644
--- a/src/aes192.asm
+++ b/src/aes192.asm
@@ -99,30 +99,44 @@
     ; * xmm0[63:32]  == w[1],
     ; * xmm0[31:0]   == w[0].
 
-    movdqa [ecx], xmm0          ; sets w[0], w[1], w[2], w[3]
-    movdqa [ecx + 10h], xmm1    ; sets w[4], w[5]
-    add ecx, 18h                ; ecx = &w[6]
+    movdqa [ecx], xmm0                  ; sets w[0], w[1], w[2], w[3]
+    movq qword ptr [ecx + 10h], xmm1    ; sets w[4], w[5]
 
     aeskeygenassist xmm7, xmm1, 1      ; xmm7[63:32] = RotWord(SubWord(w[5]))^Rcon,
-    call gen_round_key                 ; sets w[6], w[7], w[8], w[9], w[10], w[11]
+    call aes192_keygen_assist
+    movdqu [ecx + 18h], xmm0
+    movq qword ptr [ecx + 28h], xmm1
     aeskeygenassist xmm7, xmm1, 2      ; xmm7[63:32] = RotWord(SubWord(w[11]))^Rcon
-    call gen_round_key                 ; sets w[12], w[13], w[14], w[15], w[16], w[17]
+    call aes192_keygen_assist
+    movdqa [ecx + 30h], xmm0
+    movq qword ptr [ecx + 40h], xmm1
     aeskeygenassist xmm7, xmm1, 4      ; xmm7[63:32] = RotWord(SubWord(w[17]))^Rcon
-    call gen_round_key                 ; sets w[18], w[19], w[20], w[21], w[22], w[23]
+    call aes192_keygen_assist
+    movdqu [ecx + 48h], xmm0
+    movq qword ptr [ecx + 58h], xmm1
     aeskeygenassist xmm7, xmm1, 8      ; xmm7[63:32] = RotWord(SubWord(w[23]))^Rcon
-    call gen_round_key                 ; sets w[24], w[25], w[26], w[27], w[28], w[29]
+    call aes192_keygen_assist
+    movdqa [ecx + 60h], xmm0
+    movq qword ptr [ecx + 70h], xmm1
     aeskeygenassist xmm7, xmm1, 10h    ; xmm7[63:32] = RotWord(SubWord(w[29]))^Rcon
-    call gen_round_key                 ; sets w[30], w[31], w[32], w[33], w[34], w[35]
+    call aes192_keygen_assist
+    movdqu [ecx + 78h], xmm0
+    movq qword ptr [ecx + 88h], xmm1
     aeskeygenassist xmm7, xmm1, 20h    ; xmm7[63:32] = RotWord(SubWord(w[35]))^Rcon
-    call gen_round_key                 ; sets w[36], w[37], w[38], w[39], w[40], w[41]
+    call aes192_keygen_assist
+    movdqa [ecx + 90h], xmm0
+    movq qword ptr [ecx + 0a0h], xmm1
     aeskeygenassist xmm7, xmm1, 40h    ; xmm7[63:32] = RotWord(SubWord(w[41]))^Rcon
-    call gen_round_key                 ; sets w[42], w[43], w[44], w[45], w[46], w[47]
+    call aes192_keygen_assist
+    movdqu [ecx + 0a8h], xmm0
+    movq qword ptr [ecx + 0b8h], xmm1
     aeskeygenassist xmm7, xmm1, 80h    ; xmm7[63:32] = RotWord(SubWord(w[49]))^Rcon
-    call gen_round_key                 ; sets w[48], w[49], w[50], w[51], w[52], w[53]
+    call aes192_keygen_assist
+    movdqa [ecx + 0c0h], xmm0
 
     ret
 
-gen_round_key:
+aes192_keygen_assist:
     ; Preconditions:
     ; * xmm1[127:96] == 0,
     ; * xmm1[95:64]  == 0,
@@ -132,8 +146,7 @@ gen_round_key:
     ; * xmm0[95:64]  == w[i+2],
     ; * xmm0[63:32]  == w[i+1],
     ; * xmm0[31:0]   == w[i],
-    ; * xmm7[63:32]  == RotWord(SubWord(w[i+5]))^Rcon,
-    ; * ecx == &w[i+6].
+    ; * xmm7[63:32]  == RotWord(SubWord(w[i+5]))^Rcon.
     ;
     ; Postconditions:
     ; * xmm1[127:96] == 0,
@@ -144,7 +157,6 @@ gen_round_key:
     ; * xmm0[95:64]  == w[i+8]  == RotWord(SubWord(w[i+5]))^Rcon^w[i+2]^w[i+1]^w[i],
     ; * xmm0[63:32]  == w[i+7]  == RotWord(SubWord(w[i+5]))^Rcon^w[i+1]^w[i],
     ; * xmm0[31:0]   == w[i+6]  == RotWord(SubWord(w[i+5]))^Rcon^w[i],
-    ; * ecx == &w[i+12],
     ; * the value in xmm6 is also modified.
 
     ; Calculate
@@ -176,13 +188,6 @@ gen_round_key:
                               ; xmm0[63:32]  == w[i+7] == RotWord(SubWord(w[i+5]))^Rcon^w[i+1]^w[i]
                               ; xmm0[31:0]   == w[i+6] == RotWord(SubWord(w[i+5]))^Rcon^w[i]
 
-    ; Set w[i+6], w[i+7], w[i+8] and w[i+9].
-    movdqu [ecx], xmm0    ; w[i+6] = RotWord(SubWord(w[i+5]))^Rcon^w[i]
-                          ; w[i+7] = RotWord(SubWord(w[i+5]))^Rcon^w[i+1]^w[i]
-                          ; w[i+8] = RotWord(SubWord(w[i+5]))^Rcon^w[i+2]^w[i+1]^w[i]
-                          ; w[i+9] = RotWord(SubWord(w[i+5]))^Rcon^w[i+3]^w[i+2]^w[i+1]^w[i]
-    add ecx, 10h          ; ecx = &w[i+10]
-
     ; Calculate
     ;     w[i+5]^w[i+4],
     ;     w[i+4].
@@ -200,11 +205,6 @@ gen_round_key:
                                ; xmm1[63:32] == w[i+11] == RotWord(SubWord(w[i+5]))^Rcon^w[i+5]^w[i+4]^w[i+3]^w[i+2]^w[i+1]^w[i]
                                ; xmm1[31:0]  == w[i+10] == RotWord(SubWord(w[i+5]))^Rcon^w[i+4]^w[i+3]^w[i+2]^w[i+1]^w[i]
 
-    ; Set w[i+10] and w[i+11].
-    movq qword ptr [ecx], xmm1    ; w[i+10] = RotWord(SubWord(w[i+5]))^Rcon^w[i+4]^w[i+3]^w[i+2]^w[i+1]^w[i]
-                                  ; w[i+11] = RotWord(SubWord(w[i+5]))^Rcon^w[i+5]^w[i+4]^w[i+3]^w[i+2]^w[i+1]^w[i]
-    add ecx, 8                    ; ecx = &w[i+12]
-
     ret
 @raw_aes192_expand_key_schedule@36 endp