AES-NI backport from HEAD. Note that e_aes.c doesn't implement all modes
[openssl.git] / crypto / modes / ctr128.c
index 932037f..181614a 100644 (file)
@@ -62,10 +62,10 @@ typedef unsigned int u32;
 typedef unsigned char u8;
 
 #define STRICT_ALIGNMENT
-#if defined(__i386)    || defined(__i386__)    || \
-    defined(__x86_64)  || defined(__x86_64__)  || \
-    defined(_M_IX86)   || defined(_M_AMD64)    || defined(_M_X64) || \
-    defined(__s390__)  || defined(__s390x__)
+#if defined(__i386)    || defined(__i386__)    || \
+    defined(__x86_64)  || defined(__x86_64__)  || \
+    defined(_M_IX86)   || defined(_M_AMD64)    || defined(_M_X64) || \
+    defined(__s390__)  || defined(__s390x__)
 #  undef STRICT_ALIGNMENT
 #endif
 
@@ -182,3 +182,84 @@ void CRYPTO_ctr128_encrypt(const unsigned char *in, unsigned char *out,
 
        *num=n;
 }
+
+#define GETU32(p)      ((u32)(p)[0]<<24|(u32)(p)[1]<<16|(u32)(p)[2]<<8|(u32)(p)[3])
+#define PUTU32(p,v)    ((p)[0]=(u8)((v)>>24),(p)[1]=(u8)((v)>>16),(p)[2]=(u8)((v)>>8),(p)[3]=(u8)(v))
+
+/* increment upper 96 bits of 128-bit counter by 1 */
+static void ctr96_inc(unsigned char *counter) {
+       u32 n=12;
+       u8  c;
+
+       do {
+               --n;
+               c = counter[n];
+               ++c;
+               counter[n] = c;
+               if (c) return;
+       } while (n);
+}
+
+void CRYPTO_ctr128_encrypt_ctr32(const unsigned char *in, unsigned char *out,
+                       size_t len, const void *key,
+                       unsigned char ivec[16], unsigned char ecount_buf[16],
+                       unsigned int *num, ctr128_f func)
+{
+       unsigned int n,ctr32;
+
+       assert(in && out && key && ecount_buf && num);
+       assert(*num < 16);
+
+       n = *num;
+
+       while (n && len) {
+               *(out++) = *(in++) ^ ecount_buf[n];
+               --len;
+               n = (n+1) % 16;
+       }
+
+       ctr32 = GETU32(ivec+12);
+       while (len>=16) {
+               size_t blocks = len/16;
+               /*
+                * 1<<28 is just a not-so-small yet not-so-large number...
+                * Below condition is practically never met, but it has to
+                * be checked for code correctness.
+                */
+               if (sizeof(size_t)>sizeof(unsigned int) && blocks>(1U<<28))
+                       blocks = (1U<<28);
+               /*
+                * As (*func) operates on 32-bit counter, caller
+                * has to handle overflow. 'if' below detects the
+                * overflow, which is then handled by limiting the
+                * amount of blocks to the exact overflow point...
+                */
+               ctr32 += (u32)blocks;
+               if (ctr32 < blocks) {
+                       blocks -= ctr32;
+                       ctr32   = 0;
+               }
+               (*func)(in,out,blocks,key,ivec);
+               /* (*ctr) does not update ivec, caller does: */
+               PUTU32(ivec+12,ctr32);
+               /* ... overflow was detected, propogate carry. */
+               if (ctr32 == 0) ctr96_inc(ivec);
+               blocks *= 16;
+               len -= blocks;
+               out += blocks;
+               in  += blocks;
+       }
+       if (len) {
+               memset(ecount_buf,0,16);
+               (*func)(ecount_buf,ecount_buf,1,key,ivec);
+               ++ctr32;
+               PUTU32(ivec+12,ctr32);
+               if (ctr32 == 0) ctr96_inc(ivec);
+               while (len--) {
+                       out[n] = in[n] ^ ecount_buf[n];
+                       ++n;
+               }
+       }
+
+       *num=n;
+}