xref: /freebsd/crypto/openssl/crypto/modes/gcm128.c (revision b077aed33b7b6aefca7b17ddb250cf521f938613)
1 /*
2  * Copyright 2010-2021 The OpenSSL Project Authors. All Rights Reserved.
3  *
4  * Licensed under the Apache License 2.0 (the "License").  You may not use
5  * this file except in compliance with the License.  You can obtain a copy
6  * in the file LICENSE in the source distribution or at
7  * https://www.openssl.org/source/license.html
8  */
9 
10 #include <string.h>
11 #include <openssl/crypto.h>
12 #include "internal/cryptlib.h"
13 #include "internal/endian.h"
14 #include "crypto/modes.h"
15 
16 #if defined(__GNUC__) && !defined(STRICT_ALIGNMENT)
17 typedef size_t size_t_aX __attribute((__aligned__(1)));
18 #else
19 typedef size_t size_t_aX;
20 #endif
21 
22 #if defined(BSWAP4) && defined(STRICT_ALIGNMENT)
23 /* redefine, because alignment is ensured */
24 # undef  GETU32
25 # define GETU32(p)       BSWAP4(*(const u32 *)(p))
26 # undef  PUTU32
27 # define PUTU32(p,v)     *(u32 *)(p) = BSWAP4(v)
28 #endif
29 
30 #define PACK(s)         ((size_t)(s)<<(sizeof(size_t)*8-16))
31 #define REDUCE1BIT(V)   do { \
32         if (sizeof(size_t)==8) { \
33                 u64 T = U64(0xe100000000000000) & (0-(V.lo&1)); \
34                 V.lo  = (V.hi<<63)|(V.lo>>1); \
35                 V.hi  = (V.hi>>1 )^T; \
36         } \
37         else { \
38                 u32 T = 0xe1000000U & (0-(u32)(V.lo&1)); \
39                 V.lo  = (V.hi<<63)|(V.lo>>1); \
40                 V.hi  = (V.hi>>1 )^((u64)T<<32); \
41         } \
42 } while(0)
43 
44 /*-
45  * Even though permitted values for TABLE_BITS are 8, 4 and 1, it should
46  * never be set to 8. 8 is effectively reserved for testing purposes.
47  * TABLE_BITS>1 are lookup-table-driven implementations referred to as
48  * "Shoup's" in GCM specification. In other words OpenSSL does not cover
49  * whole spectrum of possible table driven implementations. Why? In
50  * non-"Shoup's" case memory access pattern is segmented in such manner,
51  * that it's trivial to see that cache timing information can reveal
52  * fair portion of intermediate hash value. Given that ciphertext is
53  * always available to attacker, it's possible for him to attempt to
54  * deduce secret parameter H and if successful, tamper with messages
55  * [which is nothing but trivial in CTR mode]. In "Shoup's" case it's
56  * not as trivial, but there is no reason to believe that it's resistant
57  * to cache-timing attack. And the thing about "8-bit" implementation is
58  * that it consumes 16 (sixteen) times more memory, 4KB per individual
59  * key + 1KB shared. Well, on pros side it should be twice as fast as
60  * "4-bit" version. And for gcc-generated x86[_64] code, "8-bit" version
61  * was observed to run ~75% faster, closer to 100% for commercial
62  * compilers... Yet "4-bit" procedure is preferred, because it's
63  * believed to provide better security-performance balance and adequate
64  * all-round performance. "All-round" refers to things like:
65  *
66  * - shorter setup time effectively improves overall timing for
67  *   handling short messages;
68  * - larger table allocation can become unbearable because of VM
69  *   subsystem penalties (for example on Windows large enough free
70  *   results in VM working set trimming, meaning that consequent
71  *   malloc would immediately incur working set expansion);
72  * - larger table has larger cache footprint, which can affect
73  *   performance of other code paths (not necessarily even from same
74  *   thread in Hyper-Threading world);
75  *
76  * Value of 1 is not appropriate for performance reasons.
77  */
78 #if     TABLE_BITS==8
79 
gcm_init_8bit(u128 Htable[256],u64 H[2])80 static void gcm_init_8bit(u128 Htable[256], u64 H[2])
81 {
82     int i, j;
83     u128 V;
84 
85     Htable[0].hi = 0;
86     Htable[0].lo = 0;
87     V.hi = H[0];
88     V.lo = H[1];
89 
90     for (Htable[128] = V, i = 64; i > 0; i >>= 1) {
91         REDUCE1BIT(V);
92         Htable[i] = V;
93     }
94 
95     for (i = 2; i < 256; i <<= 1) {
96         u128 *Hi = Htable + i, H0 = *Hi;
97         for (j = 1; j < i; ++j) {
98             Hi[j].hi = H0.hi ^ Htable[j].hi;
99             Hi[j].lo = H0.lo ^ Htable[j].lo;
100         }
101     }
102 }
103 
gcm_gmult_8bit(u64 Xi[2],const u128 Htable[256])104 static void gcm_gmult_8bit(u64 Xi[2], const u128 Htable[256])
105 {
106     u128 Z = { 0, 0 };
107     const u8 *xi = (const u8 *)Xi + 15;
108     size_t rem, n = *xi;
109     DECLARE_IS_ENDIAN;
110     static const size_t rem_8bit[256] = {
111         PACK(0x0000), PACK(0x01C2), PACK(0x0384), PACK(0x0246),
112         PACK(0x0708), PACK(0x06CA), PACK(0x048C), PACK(0x054E),
113         PACK(0x0E10), PACK(0x0FD2), PACK(0x0D94), PACK(0x0C56),
114         PACK(0x0918), PACK(0x08DA), PACK(0x0A9C), PACK(0x0B5E),
115         PACK(0x1C20), PACK(0x1DE2), PACK(0x1FA4), PACK(0x1E66),
116         PACK(0x1B28), PACK(0x1AEA), PACK(0x18AC), PACK(0x196E),
117         PACK(0x1230), PACK(0x13F2), PACK(0x11B4), PACK(0x1076),
118         PACK(0x1538), PACK(0x14FA), PACK(0x16BC), PACK(0x177E),
119         PACK(0x3840), PACK(0x3982), PACK(0x3BC4), PACK(0x3A06),
120         PACK(0x3F48), PACK(0x3E8A), PACK(0x3CCC), PACK(0x3D0E),
121         PACK(0x3650), PACK(0x3792), PACK(0x35D4), PACK(0x3416),
122         PACK(0x3158), PACK(0x309A), PACK(0x32DC), PACK(0x331E),
123         PACK(0x2460), PACK(0x25A2), PACK(0x27E4), PACK(0x2626),
124         PACK(0x2368), PACK(0x22AA), PACK(0x20EC), PACK(0x212E),
125         PACK(0x2A70), PACK(0x2BB2), PACK(0x29F4), PACK(0x2836),
126         PACK(0x2D78), PACK(0x2CBA), PACK(0x2EFC), PACK(0x2F3E),
127         PACK(0x7080), PACK(0x7142), PACK(0x7304), PACK(0x72C6),
128         PACK(0x7788), PACK(0x764A), PACK(0x740C), PACK(0x75CE),
129         PACK(0x7E90), PACK(0x7F52), PACK(0x7D14), PACK(0x7CD6),
130         PACK(0x7998), PACK(0x785A), PACK(0x7A1C), PACK(0x7BDE),
131         PACK(0x6CA0), PACK(0x6D62), PACK(0x6F24), PACK(0x6EE6),
132         PACK(0x6BA8), PACK(0x6A6A), PACK(0x682C), PACK(0x69EE),
133         PACK(0x62B0), PACK(0x6372), PACK(0x6134), PACK(0x60F6),
134         PACK(0x65B8), PACK(0x647A), PACK(0x663C), PACK(0x67FE),
135         PACK(0x48C0), PACK(0x4902), PACK(0x4B44), PACK(0x4A86),
136         PACK(0x4FC8), PACK(0x4E0A), PACK(0x4C4C), PACK(0x4D8E),
137         PACK(0x46D0), PACK(0x4712), PACK(0x4554), PACK(0x4496),
138         PACK(0x41D8), PACK(0x401A), PACK(0x425C), PACK(0x439E),
139         PACK(0x54E0), PACK(0x5522), PACK(0x5764), PACK(0x56A6),
140         PACK(0x53E8), PACK(0x522A), PACK(0x506C), PACK(0x51AE),
141         PACK(0x5AF0), PACK(0x5B32), PACK(0x5974), PACK(0x58B6),
142         PACK(0x5DF8), PACK(0x5C3A), PACK(0x5E7C), PACK(0x5FBE),
143         PACK(0xE100), PACK(0xE0C2), PACK(0xE284), PACK(0xE346),
144         PACK(0xE608), PACK(0xE7CA), PACK(0xE58C), PACK(0xE44E),
145         PACK(0xEF10), PACK(0xEED2), PACK(0xEC94), PACK(0xED56),
146         PACK(0xE818), PACK(0xE9DA), PACK(0xEB9C), PACK(0xEA5E),
147         PACK(0xFD20), PACK(0xFCE2), PACK(0xFEA4), PACK(0xFF66),
148         PACK(0xFA28), PACK(0xFBEA), PACK(0xF9AC), PACK(0xF86E),
149         PACK(0xF330), PACK(0xF2F2), PACK(0xF0B4), PACK(0xF176),
150         PACK(0xF438), PACK(0xF5FA), PACK(0xF7BC), PACK(0xF67E),
151         PACK(0xD940), PACK(0xD882), PACK(0xDAC4), PACK(0xDB06),
152         PACK(0xDE48), PACK(0xDF8A), PACK(0xDDCC), PACK(0xDC0E),
153         PACK(0xD750), PACK(0xD692), PACK(0xD4D4), PACK(0xD516),
154         PACK(0xD058), PACK(0xD19A), PACK(0xD3DC), PACK(0xD21E),
155         PACK(0xC560), PACK(0xC4A2), PACK(0xC6E4), PACK(0xC726),
156         PACK(0xC268), PACK(0xC3AA), PACK(0xC1EC), PACK(0xC02E),
157         PACK(0xCB70), PACK(0xCAB2), PACK(0xC8F4), PACK(0xC936),
158         PACK(0xCC78), PACK(0xCDBA), PACK(0xCFFC), PACK(0xCE3E),
159         PACK(0x9180), PACK(0x9042), PACK(0x9204), PACK(0x93C6),
160         PACK(0x9688), PACK(0x974A), PACK(0x950C), PACK(0x94CE),
161         PACK(0x9F90), PACK(0x9E52), PACK(0x9C14), PACK(0x9DD6),
162         PACK(0x9898), PACK(0x995A), PACK(0x9B1C), PACK(0x9ADE),
163         PACK(0x8DA0), PACK(0x8C62), PACK(0x8E24), PACK(0x8FE6),
164         PACK(0x8AA8), PACK(0x8B6A), PACK(0x892C), PACK(0x88EE),
165         PACK(0x83B0), PACK(0x8272), PACK(0x8034), PACK(0x81F6),
166         PACK(0x84B8), PACK(0x857A), PACK(0x873C), PACK(0x86FE),
167         PACK(0xA9C0), PACK(0xA802), PACK(0xAA44), PACK(0xAB86),
168         PACK(0xAEC8), PACK(0xAF0A), PACK(0xAD4C), PACK(0xAC8E),
169         PACK(0xA7D0), PACK(0xA612), PACK(0xA454), PACK(0xA596),
170         PACK(0xA0D8), PACK(0xA11A), PACK(0xA35C), PACK(0xA29E),
171         PACK(0xB5E0), PACK(0xB422), PACK(0xB664), PACK(0xB7A6),
172         PACK(0xB2E8), PACK(0xB32A), PACK(0xB16C), PACK(0xB0AE),
173         PACK(0xBBF0), PACK(0xBA32), PACK(0xB874), PACK(0xB9B6),
174         PACK(0xBCF8), PACK(0xBD3A), PACK(0xBF7C), PACK(0xBEBE)
175     };
176 
177     while (1) {
178         Z.hi ^= Htable[n].hi;
179         Z.lo ^= Htable[n].lo;
180 
181         if ((u8 *)Xi == xi)
182             break;
183 
184         n = *(--xi);
185 
186         rem = (size_t)Z.lo & 0xff;
187         Z.lo = (Z.hi << 56) | (Z.lo >> 8);
188         Z.hi = (Z.hi >> 8);
189         if (sizeof(size_t) == 8)
190             Z.hi ^= rem_8bit[rem];
191         else
192             Z.hi ^= (u64)rem_8bit[rem] << 32;
193     }
194 
195     if (IS_LITTLE_ENDIAN) {
196 # ifdef BSWAP8
197         Xi[0] = BSWAP8(Z.hi);
198         Xi[1] = BSWAP8(Z.lo);
199 # else
200         u8 *p = (u8 *)Xi;
201         u32 v;
202         v = (u32)(Z.hi >> 32);
203         PUTU32(p, v);
204         v = (u32)(Z.hi);
205         PUTU32(p + 4, v);
206         v = (u32)(Z.lo >> 32);
207         PUTU32(p + 8, v);
208         v = (u32)(Z.lo);
209         PUTU32(p + 12, v);
210 # endif
211     } else {
212         Xi[0] = Z.hi;
213         Xi[1] = Z.lo;
214     }
215 }
216 
217 # define GCM_MUL(ctx)      gcm_gmult_8bit(ctx->Xi.u,ctx->Htable)
218 
219 #elif   TABLE_BITS==4
220 
gcm_init_4bit(u128 Htable[16],u64 H[2])221 static void gcm_init_4bit(u128 Htable[16], u64 H[2])
222 {
223     u128 V;
224 # if defined(OPENSSL_SMALL_FOOTPRINT)
225     int i;
226 # endif
227 
228     Htable[0].hi = 0;
229     Htable[0].lo = 0;
230     V.hi = H[0];
231     V.lo = H[1];
232 
233 # if defined(OPENSSL_SMALL_FOOTPRINT)
234     for (Htable[8] = V, i = 4; i > 0; i >>= 1) {
235         REDUCE1BIT(V);
236         Htable[i] = V;
237     }
238 
239     for (i = 2; i < 16; i <<= 1) {
240         u128 *Hi = Htable + i;
241         int j;
242         for (V = *Hi, j = 1; j < i; ++j) {
243             Hi[j].hi = V.hi ^ Htable[j].hi;
244             Hi[j].lo = V.lo ^ Htable[j].lo;
245         }
246     }
247 # else
248     Htable[8] = V;
249     REDUCE1BIT(V);
250     Htable[4] = V;
251     REDUCE1BIT(V);
252     Htable[2] = V;
253     REDUCE1BIT(V);
254     Htable[1] = V;
255     Htable[3].hi = V.hi ^ Htable[2].hi, Htable[3].lo = V.lo ^ Htable[2].lo;
256     V = Htable[4];
257     Htable[5].hi = V.hi ^ Htable[1].hi, Htable[5].lo = V.lo ^ Htable[1].lo;
258     Htable[6].hi = V.hi ^ Htable[2].hi, Htable[6].lo = V.lo ^ Htable[2].lo;
259     Htable[7].hi = V.hi ^ Htable[3].hi, Htable[7].lo = V.lo ^ Htable[3].lo;
260     V = Htable[8];
261     Htable[9].hi = V.hi ^ Htable[1].hi, Htable[9].lo = V.lo ^ Htable[1].lo;
262     Htable[10].hi = V.hi ^ Htable[2].hi, Htable[10].lo = V.lo ^ Htable[2].lo;
263     Htable[11].hi = V.hi ^ Htable[3].hi, Htable[11].lo = V.lo ^ Htable[3].lo;
264     Htable[12].hi = V.hi ^ Htable[4].hi, Htable[12].lo = V.lo ^ Htable[4].lo;
265     Htable[13].hi = V.hi ^ Htable[5].hi, Htable[13].lo = V.lo ^ Htable[5].lo;
266     Htable[14].hi = V.hi ^ Htable[6].hi, Htable[14].lo = V.lo ^ Htable[6].lo;
267     Htable[15].hi = V.hi ^ Htable[7].hi, Htable[15].lo = V.lo ^ Htable[7].lo;
268 # endif
269 # if defined(GHASH_ASM) && (defined(__arm__) || defined(__arm))
270     /*
271      * ARM assembler expects specific dword order in Htable.
272      */
273     {
274         int j;
275         DECLARE_IS_ENDIAN;
276 
277         if (IS_LITTLE_ENDIAN)
278             for (j = 0; j < 16; ++j) {
279                 V = Htable[j];
280                 Htable[j].hi = V.lo;
281                 Htable[j].lo = V.hi;
282         } else
283             for (j = 0; j < 16; ++j) {
284                 V = Htable[j];
285                 Htable[j].hi = V.lo << 32 | V.lo >> 32;
286                 Htable[j].lo = V.hi << 32 | V.hi >> 32;
287             }
288     }
289 # endif
290 }
291 
292 # ifndef GHASH_ASM
293 static const size_t rem_4bit[16] = {
294     PACK(0x0000), PACK(0x1C20), PACK(0x3840), PACK(0x2460),
295     PACK(0x7080), PACK(0x6CA0), PACK(0x48C0), PACK(0x54E0),
296     PACK(0xE100), PACK(0xFD20), PACK(0xD940), PACK(0xC560),
297     PACK(0x9180), PACK(0x8DA0), PACK(0xA9C0), PACK(0xB5E0)
298 };
299 
gcm_gmult_4bit(u64 Xi[2],const u128 Htable[16])300 static void gcm_gmult_4bit(u64 Xi[2], const u128 Htable[16])
301 {
302     u128 Z;
303     int cnt = 15;
304     size_t rem, nlo, nhi;
305     DECLARE_IS_ENDIAN;
306 
307     nlo = ((const u8 *)Xi)[15];
308     nhi = nlo >> 4;
309     nlo &= 0xf;
310 
311     Z.hi = Htable[nlo].hi;
312     Z.lo = Htable[nlo].lo;
313 
314     while (1) {
315         rem = (size_t)Z.lo & 0xf;
316         Z.lo = (Z.hi << 60) | (Z.lo >> 4);
317         Z.hi = (Z.hi >> 4);
318         if (sizeof(size_t) == 8)
319             Z.hi ^= rem_4bit[rem];
320         else
321             Z.hi ^= (u64)rem_4bit[rem] << 32;
322 
323         Z.hi ^= Htable[nhi].hi;
324         Z.lo ^= Htable[nhi].lo;
325 
326         if (--cnt < 0)
327             break;
328 
329         nlo = ((const u8 *)Xi)[cnt];
330         nhi = nlo >> 4;
331         nlo &= 0xf;
332 
333         rem = (size_t)Z.lo & 0xf;
334         Z.lo = (Z.hi << 60) | (Z.lo >> 4);
335         Z.hi = (Z.hi >> 4);
336         if (sizeof(size_t) == 8)
337             Z.hi ^= rem_4bit[rem];
338         else
339             Z.hi ^= (u64)rem_4bit[rem] << 32;
340 
341         Z.hi ^= Htable[nlo].hi;
342         Z.lo ^= Htable[nlo].lo;
343     }
344 
345     if (IS_LITTLE_ENDIAN) {
346 #  ifdef BSWAP8
347         Xi[0] = BSWAP8(Z.hi);
348         Xi[1] = BSWAP8(Z.lo);
349 #  else
350         u8 *p = (u8 *)Xi;
351         u32 v;
352         v = (u32)(Z.hi >> 32);
353         PUTU32(p, v);
354         v = (u32)(Z.hi);
355         PUTU32(p + 4, v);
356         v = (u32)(Z.lo >> 32);
357         PUTU32(p + 8, v);
358         v = (u32)(Z.lo);
359         PUTU32(p + 12, v);
360 #  endif
361     } else {
362         Xi[0] = Z.hi;
363         Xi[1] = Z.lo;
364     }
365 }
366 
367 #  if !defined(OPENSSL_SMALL_FOOTPRINT)
368 /*
369  * Streamed gcm_mult_4bit, see CRYPTO_gcm128_[en|de]crypt for
370  * details... Compiler-generated code doesn't seem to give any
371  * performance improvement, at least not on x86[_64]. It's here
372  * mostly as reference and a placeholder for possible future
373  * non-trivial optimization[s]...
374  */
gcm_ghash_4bit(u64 Xi[2],const u128 Htable[16],const u8 * inp,size_t len)375 static void gcm_ghash_4bit(u64 Xi[2], const u128 Htable[16],
376                            const u8 *inp, size_t len)
377 {
378     u128 Z;
379     int cnt;
380     size_t rem, nlo, nhi;
381     DECLARE_IS_ENDIAN;
382 
383 #   if 1
384     do {
385         cnt = 15;
386         nlo = ((const u8 *)Xi)[15];
387         nlo ^= inp[15];
388         nhi = nlo >> 4;
389         nlo &= 0xf;
390 
391         Z.hi = Htable[nlo].hi;
392         Z.lo = Htable[nlo].lo;
393 
394         while (1) {
395             rem = (size_t)Z.lo & 0xf;
396             Z.lo = (Z.hi << 60) | (Z.lo >> 4);
397             Z.hi = (Z.hi >> 4);
398             if (sizeof(size_t) == 8)
399                 Z.hi ^= rem_4bit[rem];
400             else
401                 Z.hi ^= (u64)rem_4bit[rem] << 32;
402 
403             Z.hi ^= Htable[nhi].hi;
404             Z.lo ^= Htable[nhi].lo;
405 
406             if (--cnt < 0)
407                 break;
408 
409             nlo = ((const u8 *)Xi)[cnt];
410             nlo ^= inp[cnt];
411             nhi = nlo >> 4;
412             nlo &= 0xf;
413 
414             rem = (size_t)Z.lo & 0xf;
415             Z.lo = (Z.hi << 60) | (Z.lo >> 4);
416             Z.hi = (Z.hi >> 4);
417             if (sizeof(size_t) == 8)
418                 Z.hi ^= rem_4bit[rem];
419             else
420                 Z.hi ^= (u64)rem_4bit[rem] << 32;
421 
422             Z.hi ^= Htable[nlo].hi;
423             Z.lo ^= Htable[nlo].lo;
424         }
425 #   else
426     /*
427      * Extra 256+16 bytes per-key plus 512 bytes shared tables
428      * [should] give ~50% improvement... One could have PACK()-ed
429      * the rem_8bit even here, but the priority is to minimize
430      * cache footprint...
431      */
432     u128 Hshr4[16];             /* Htable shifted right by 4 bits */
433     u8 Hshl4[16];               /* Htable shifted left by 4 bits */
434     static const unsigned short rem_8bit[256] = {
435         0x0000, 0x01C2, 0x0384, 0x0246, 0x0708, 0x06CA, 0x048C, 0x054E,
436         0x0E10, 0x0FD2, 0x0D94, 0x0C56, 0x0918, 0x08DA, 0x0A9C, 0x0B5E,
437         0x1C20, 0x1DE2, 0x1FA4, 0x1E66, 0x1B28, 0x1AEA, 0x18AC, 0x196E,
438         0x1230, 0x13F2, 0x11B4, 0x1076, 0x1538, 0x14FA, 0x16BC, 0x177E,
439         0x3840, 0x3982, 0x3BC4, 0x3A06, 0x3F48, 0x3E8A, 0x3CCC, 0x3D0E,
440         0x3650, 0x3792, 0x35D4, 0x3416, 0x3158, 0x309A, 0x32DC, 0x331E,
441         0x2460, 0x25A2, 0x27E4, 0x2626, 0x2368, 0x22AA, 0x20EC, 0x212E,
442         0x2A70, 0x2BB2, 0x29F4, 0x2836, 0x2D78, 0x2CBA, 0x2EFC, 0x2F3E,
443         0x7080, 0x7142, 0x7304, 0x72C6, 0x7788, 0x764A, 0x740C, 0x75CE,
444         0x7E90, 0x7F52, 0x7D14, 0x7CD6, 0x7998, 0x785A, 0x7A1C, 0x7BDE,
445         0x6CA0, 0x6D62, 0x6F24, 0x6EE6, 0x6BA8, 0x6A6A, 0x682C, 0x69EE,
446         0x62B0, 0x6372, 0x6134, 0x60F6, 0x65B8, 0x647A, 0x663C, 0x67FE,
447         0x48C0, 0x4902, 0x4B44, 0x4A86, 0x4FC8, 0x4E0A, 0x4C4C, 0x4D8E,
448         0x46D0, 0x4712, 0x4554, 0x4496, 0x41D8, 0x401A, 0x425C, 0x439E,
449         0x54E0, 0x5522, 0x5764, 0x56A6, 0x53E8, 0x522A, 0x506C, 0x51AE,
450         0x5AF0, 0x5B32, 0x5974, 0x58B6, 0x5DF8, 0x5C3A, 0x5E7C, 0x5FBE,
451         0xE100, 0xE0C2, 0xE284, 0xE346, 0xE608, 0xE7CA, 0xE58C, 0xE44E,
452         0xEF10, 0xEED2, 0xEC94, 0xED56, 0xE818, 0xE9DA, 0xEB9C, 0xEA5E,
453         0xFD20, 0xFCE2, 0xFEA4, 0xFF66, 0xFA28, 0xFBEA, 0xF9AC, 0xF86E,
454         0xF330, 0xF2F2, 0xF0B4, 0xF176, 0xF438, 0xF5FA, 0xF7BC, 0xF67E,
455         0xD940, 0xD882, 0xDAC4, 0xDB06, 0xDE48, 0xDF8A, 0xDDCC, 0xDC0E,
456         0xD750, 0xD692, 0xD4D4, 0xD516, 0xD058, 0xD19A, 0xD3DC, 0xD21E,
457         0xC560, 0xC4A2, 0xC6E4, 0xC726, 0xC268, 0xC3AA, 0xC1EC, 0xC02E,
458         0xCB70, 0xCAB2, 0xC8F4, 0xC936, 0xCC78, 0xCDBA, 0xCFFC, 0xCE3E,
459         0x9180, 0x9042, 0x9204, 0x93C6, 0x9688, 0x974A, 0x950C, 0x94CE,
460         0x9F90, 0x9E52, 0x9C14, 0x9DD6, 0x9898, 0x995A, 0x9B1C, 0x9ADE,
461         0x8DA0, 0x8C62, 0x8E24, 0x8FE6, 0x8AA8, 0x8B6A, 0x892C, 0x88EE,
462         0x83B0, 0x8272, 0x8034, 0x81F6, 0x84B8, 0x857A, 0x873C, 0x86FE,
463         0xA9C0, 0xA802, 0xAA44, 0xAB86, 0xAEC8, 0xAF0A, 0xAD4C, 0xAC8E,
464         0xA7D0, 0xA612, 0xA454, 0xA596, 0xA0D8, 0xA11A, 0xA35C, 0xA29E,
465         0xB5E0, 0xB422, 0xB664, 0xB7A6, 0xB2E8, 0xB32A, 0xB16C, 0xB0AE,
466         0xBBF0, 0xBA32, 0xB874, 0xB9B6, 0xBCF8, 0xBD3A, 0xBF7C, 0xBEBE
467     };
468     /*
469      * This pre-processing phase slows down procedure by approximately
470      * same time as it makes each loop spin faster. In other words
471      * single block performance is approximately same as straightforward
472      * "4-bit" implementation, and then it goes only faster...
473      */
474     for (cnt = 0; cnt < 16; ++cnt) {
475         Z.hi = Htable[cnt].hi;
476         Z.lo = Htable[cnt].lo;
477         Hshr4[cnt].lo = (Z.hi << 60) | (Z.lo >> 4);
478         Hshr4[cnt].hi = (Z.hi >> 4);
479         Hshl4[cnt] = (u8)(Z.lo << 4);
480     }
481 
482     do {
483         for (Z.lo = 0, Z.hi = 0, cnt = 15; cnt; --cnt) {
484             nlo = ((const u8 *)Xi)[cnt];
485             nlo ^= inp[cnt];
486             nhi = nlo >> 4;
487             nlo &= 0xf;
488 
489             Z.hi ^= Htable[nlo].hi;
490             Z.lo ^= Htable[nlo].lo;
491 
492             rem = (size_t)Z.lo & 0xff;
493 
494             Z.lo = (Z.hi << 56) | (Z.lo >> 8);
495             Z.hi = (Z.hi >> 8);
496 
497             Z.hi ^= Hshr4[nhi].hi;
498             Z.lo ^= Hshr4[nhi].lo;
499             Z.hi ^= (u64)rem_8bit[rem ^ Hshl4[nhi]] << 48;
500         }
501 
502         nlo = ((const u8 *)Xi)[0];
503         nlo ^= inp[0];
504         nhi = nlo >> 4;
505         nlo &= 0xf;
506 
507         Z.hi ^= Htable[nlo].hi;
508         Z.lo ^= Htable[nlo].lo;
509 
510         rem = (size_t)Z.lo & 0xf;
511 
512         Z.lo = (Z.hi << 60) | (Z.lo >> 4);
513         Z.hi = (Z.hi >> 4);
514 
515         Z.hi ^= Htable[nhi].hi;
516         Z.lo ^= Htable[nhi].lo;
517         Z.hi ^= ((u64)rem_8bit[rem << 4]) << 48;
518 #   endif
519 
520         if (IS_LITTLE_ENDIAN) {
521 #   ifdef BSWAP8
522             Xi[0] = BSWAP8(Z.hi);
523             Xi[1] = BSWAP8(Z.lo);
524 #   else
525             u8 *p = (u8 *)Xi;
526             u32 v;
527             v = (u32)(Z.hi >> 32);
528             PUTU32(p, v);
529             v = (u32)(Z.hi);
530             PUTU32(p + 4, v);
531             v = (u32)(Z.lo >> 32);
532             PUTU32(p + 8, v);
533             v = (u32)(Z.lo);
534             PUTU32(p + 12, v);
535 #   endif
536         } else {
537             Xi[0] = Z.hi;
538             Xi[1] = Z.lo;
539         }
540     } while (inp += 16, len -= 16);
541 }
542 #  endif
543 # else
544 void gcm_gmult_4bit(u64 Xi[2], const u128 Htable[16]);
545 void gcm_ghash_4bit(u64 Xi[2], const u128 Htable[16], const u8 *inp,
546                     size_t len);
547 # endif
548 
549 # define GCM_MUL(ctx)      gcm_gmult_4bit(ctx->Xi.u,ctx->Htable)
550 # if defined(GHASH_ASM) || !defined(OPENSSL_SMALL_FOOTPRINT)
551 #  define GHASH(ctx,in,len) gcm_ghash_4bit((ctx)->Xi.u,(ctx)->Htable,in,len)
552 /*
553  * GHASH_CHUNK is "stride parameter" missioned to mitigate cache trashing
554  * effect. In other words idea is to hash data while it's still in L1 cache
555  * after encryption pass...
556  */
557 #  define GHASH_CHUNK       (3*1024)
558 # endif
559 
560 #else                           /* TABLE_BITS */
561 
562 static void gcm_gmult_1bit(u64 Xi[2], const u64 H[2])
563 {
564     u128 V, Z = { 0, 0 };
565     long X;
566     int i, j;
567     const long *xi = (const long *)Xi;
568     DECLARE_IS_ENDIAN;
569 
570     V.hi = H[0];                /* H is in host byte order, no byte swapping */
571     V.lo = H[1];
572 
573     for (j = 0; j < 16 / sizeof(long); ++j) {
574         if (IS_LITTLE_ENDIAN) {
575             if (sizeof(long) == 8) {
576 # ifdef BSWAP8
577                 X = (long)(BSWAP8(xi[j]));
578 # else
579                 const u8 *p = (const u8 *)(xi + j);
580                 X = (long)((u64)GETU32(p) << 32 | GETU32(p + 4));
581 # endif
582             } else {
583                 const u8 *p = (const u8 *)(xi + j);
584                 X = (long)GETU32(p);
585             }
586         } else
587             X = xi[j];
588 
589         for (i = 0; i < 8 * sizeof(long); ++i, X <<= 1) {
590             u64 M = (u64)(X >> (8 * sizeof(long) - 1));
591             Z.hi ^= V.hi & M;
592             Z.lo ^= V.lo & M;
593 
594             REDUCE1BIT(V);
595         }
596     }
597 
598     if (IS_LITTLE_ENDIAN) {
599 # ifdef BSWAP8
600         Xi[0] = BSWAP8(Z.hi);
601         Xi[1] = BSWAP8(Z.lo);
602 # else
603         u8 *p = (u8 *)Xi;
604         u32 v;
605         v = (u32)(Z.hi >> 32);
606         PUTU32(p, v);
607         v = (u32)(Z.hi);
608         PUTU32(p + 4, v);
609         v = (u32)(Z.lo >> 32);
610         PUTU32(p + 8, v);
611         v = (u32)(Z.lo);
612         PUTU32(p + 12, v);
613 # endif
614     } else {
615         Xi[0] = Z.hi;
616         Xi[1] = Z.lo;
617     }
618 }
619 
620 # define GCM_MUL(ctx)      gcm_gmult_1bit(ctx->Xi.u,ctx->H.u)
621 
622 #endif
623 
624 #if     TABLE_BITS==4 && (defined(GHASH_ASM) || defined(OPENSSL_CPUID_OBJ))
625 # if    !defined(I386_ONLY) && \
626         (defined(__i386)        || defined(__i386__)    || \
627          defined(__x86_64)      || defined(__x86_64__)  || \
628          defined(_M_IX86)       || defined(_M_AMD64)    || defined(_M_X64))
629 #  define GHASH_ASM_X86_OR_64
630 #  define GCM_FUNCREF_4BIT
631 
632 void gcm_init_clmul(u128 Htable[16], const u64 Xi[2]);
633 void gcm_gmult_clmul(u64 Xi[2], const u128 Htable[16]);
634 void gcm_ghash_clmul(u64 Xi[2], const u128 Htable[16], const u8 *inp,
635                      size_t len);
636 
637 #  if defined(__i386) || defined(__i386__) || defined(_M_IX86)
638 #   define gcm_init_avx   gcm_init_clmul
639 #   define gcm_gmult_avx  gcm_gmult_clmul
640 #   define gcm_ghash_avx  gcm_ghash_clmul
641 #  else
642 void gcm_init_avx(u128 Htable[16], const u64 Xi[2]);
643 void gcm_gmult_avx(u64 Xi[2], const u128 Htable[16]);
644 void gcm_ghash_avx(u64 Xi[2], const u128 Htable[16], const u8 *inp,
645                    size_t len);
646 #  endif
647 
648 #  if   defined(__i386) || defined(__i386__) || defined(_M_IX86)
649 #   define GHASH_ASM_X86
650 void gcm_gmult_4bit_mmx(u64 Xi[2], const u128 Htable[16]);
651 void gcm_ghash_4bit_mmx(u64 Xi[2], const u128 Htable[16], const u8 *inp,
652                         size_t len);
653 
654 void gcm_gmult_4bit_x86(u64 Xi[2], const u128 Htable[16]);
655 void gcm_ghash_4bit_x86(u64 Xi[2], const u128 Htable[16], const u8 *inp,
656                         size_t len);
657 #  endif
658 # elif defined(__arm__) || defined(__arm) || defined(__aarch64__)
659 #  include "arm_arch.h"
660 #  if __ARM_MAX_ARCH__>=7
661 #   define GHASH_ASM_ARM
662 #   define GCM_FUNCREF_4BIT
663 #   define PMULL_CAPABLE        (OPENSSL_armcap_P & ARMV8_PMULL)
664 #   if defined(__arm__) || defined(__arm)
665 #    define NEON_CAPABLE        (OPENSSL_armcap_P & ARMV7_NEON)
666 #   endif
667 void gcm_init_neon(u128 Htable[16], const u64 Xi[2]);
668 void gcm_gmult_neon(u64 Xi[2], const u128 Htable[16]);
669 void gcm_ghash_neon(u64 Xi[2], const u128 Htable[16], const u8 *inp,
670                     size_t len);
671 void gcm_init_v8(u128 Htable[16], const u64 Xi[2]);
672 void gcm_gmult_v8(u64 Xi[2], const u128 Htable[16]);
673 void gcm_ghash_v8(u64 Xi[2], const u128 Htable[16], const u8 *inp,
674                   size_t len);
675 #  endif
676 # elif defined(__sparc__) || defined(__sparc)
677 #  include "crypto/sparc_arch.h"
678 #  define GHASH_ASM_SPARC
679 #  define GCM_FUNCREF_4BIT
680 void gcm_init_vis3(u128 Htable[16], const u64 Xi[2]);
681 void gcm_gmult_vis3(u64 Xi[2], const u128 Htable[16]);
682 void gcm_ghash_vis3(u64 Xi[2], const u128 Htable[16], const u8 *inp,
683                     size_t len);
684 # elif defined(OPENSSL_CPUID_OBJ) && (defined(__powerpc__) || defined(__ppc__) || defined(_ARCH_PPC))
685 #  include "crypto/ppc_arch.h"
686 #  define GHASH_ASM_PPC
687 #  define GCM_FUNCREF_4BIT
688 void gcm_init_p8(u128 Htable[16], const u64 Xi[2]);
689 void gcm_gmult_p8(u64 Xi[2], const u128 Htable[16]);
690 void gcm_ghash_p8(u64 Xi[2], const u128 Htable[16], const u8 *inp,
691                   size_t len);
692 # endif
693 #endif
694 
695 #ifdef GCM_FUNCREF_4BIT
696 # undef  GCM_MUL
697 # define GCM_MUL(ctx)           (*gcm_gmult_p)(ctx->Xi.u,ctx->Htable)
698 # ifdef GHASH
699 #  undef  GHASH
700 #  define GHASH(ctx,in,len)     (*gcm_ghash_p)(ctx->Xi.u,ctx->Htable,in,len)
701 # endif
702 #endif
703 
704 void CRYPTO_gcm128_init(GCM128_CONTEXT *ctx, void *key, block128_f block)
705 {
706     DECLARE_IS_ENDIAN;
707 
708     memset(ctx, 0, sizeof(*ctx));
709     ctx->block = block;
710     ctx->key = key;
711 
712     (*block) (ctx->H.c, ctx->H.c, key);
713 
714     if (IS_LITTLE_ENDIAN) {
715         /* H is stored in host byte order */
716 #ifdef BSWAP8
717         ctx->H.u[0] = BSWAP8(ctx->H.u[0]);
718         ctx->H.u[1] = BSWAP8(ctx->H.u[1]);
719 #else
720         u8 *p = ctx->H.c;
721         u64 hi, lo;
722         hi = (u64)GETU32(p) << 32 | GETU32(p + 4);
723         lo = (u64)GETU32(p + 8) << 32 | GETU32(p + 12);
724         ctx->H.u[0] = hi;
725         ctx->H.u[1] = lo;
726 #endif
727     }
728 #if     TABLE_BITS==8
729     gcm_init_8bit(ctx->Htable, ctx->H.u);
730 #elif   TABLE_BITS==4
731 # if    defined(GHASH)
732 #  define CTX__GHASH(f) (ctx->ghash = (f))
733 # else
734 #  define CTX__GHASH(f) (ctx->ghash = NULL)
735 # endif
736 # if    defined(GHASH_ASM_X86_OR_64)
737 #  if   !defined(GHASH_ASM_X86) || defined(OPENSSL_IA32_SSE2)
738     if (OPENSSL_ia32cap_P[1] & (1 << 1)) { /* check PCLMULQDQ bit */
739         if (((OPENSSL_ia32cap_P[1] >> 22) & 0x41) == 0x41) { /* AVX+MOVBE */
740             gcm_init_avx(ctx->Htable, ctx->H.u);
741             ctx->gmult = gcm_gmult_avx;
742             CTX__GHASH(gcm_ghash_avx);
743         } else {
744             gcm_init_clmul(ctx->Htable, ctx->H.u);
745             ctx->gmult = gcm_gmult_clmul;
746             CTX__GHASH(gcm_ghash_clmul);
747         }
748         return;
749     }
750 #  endif
751     gcm_init_4bit(ctx->Htable, ctx->H.u);
752 #  if   defined(GHASH_ASM_X86)  /* x86 only */
753 #   if  defined(OPENSSL_IA32_SSE2)
754     if (OPENSSL_ia32cap_P[0] & (1 << 25)) { /* check SSE bit */
755 #   else
756     if (OPENSSL_ia32cap_P[0] & (1 << 23)) { /* check MMX bit */
757 #   endif
758         ctx->gmult = gcm_gmult_4bit_mmx;
759         CTX__GHASH(gcm_ghash_4bit_mmx);
760     } else {
761         ctx->gmult = gcm_gmult_4bit_x86;
762         CTX__GHASH(gcm_ghash_4bit_x86);
763     }
764 #  else
765     ctx->gmult = gcm_gmult_4bit;
766     CTX__GHASH(gcm_ghash_4bit);
767 #  endif
768 # elif  defined(GHASH_ASM_ARM)
769 #  ifdef PMULL_CAPABLE
770     if (PMULL_CAPABLE) {
771         gcm_init_v8(ctx->Htable, ctx->H.u);
772         ctx->gmult = gcm_gmult_v8;
773         CTX__GHASH(gcm_ghash_v8);
774     } else
775 #  endif
776 #  ifdef NEON_CAPABLE
777     if (NEON_CAPABLE) {
778         gcm_init_neon(ctx->Htable, ctx->H.u);
779         ctx->gmult = gcm_gmult_neon;
780         CTX__GHASH(gcm_ghash_neon);
781     } else
782 #  endif
783     {
784         gcm_init_4bit(ctx->Htable, ctx->H.u);
785         ctx->gmult = gcm_gmult_4bit;
786         CTX__GHASH(gcm_ghash_4bit);
787     }
788 # elif  defined(GHASH_ASM_SPARC)
789     if (OPENSSL_sparcv9cap_P[0] & SPARCV9_VIS3) {
790         gcm_init_vis3(ctx->Htable, ctx->H.u);
791         ctx->gmult = gcm_gmult_vis3;
792         CTX__GHASH(gcm_ghash_vis3);
793     } else {
794         gcm_init_4bit(ctx->Htable, ctx->H.u);
795         ctx->gmult = gcm_gmult_4bit;
796         CTX__GHASH(gcm_ghash_4bit);
797     }
798 # elif  defined(GHASH_ASM_PPC)
799     if (OPENSSL_ppccap_P & PPC_CRYPTO207) {
800         gcm_init_p8(ctx->Htable, ctx->H.u);
801         ctx->gmult = gcm_gmult_p8;
802         CTX__GHASH(gcm_ghash_p8);
803     } else {
804         gcm_init_4bit(ctx->Htable, ctx->H.u);
805         ctx->gmult = gcm_gmult_4bit;
806         CTX__GHASH(gcm_ghash_4bit);
807     }
808 # else
809     gcm_init_4bit(ctx->Htable, ctx->H.u);
810 # endif
811 # undef CTX__GHASH
812 #endif
813 }
814 
815 void CRYPTO_gcm128_setiv(GCM128_CONTEXT *ctx, const unsigned char *iv,
816                          size_t len)
817 {
818     DECLARE_IS_ENDIAN;
819     unsigned int ctr;
820 #ifdef GCM_FUNCREF_4BIT
821     void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult;
822 #endif
823 
824     ctx->len.u[0] = 0;          /* AAD length */
825     ctx->len.u[1] = 0;          /* message length */
826     ctx->ares = 0;
827     ctx->mres = 0;
828 
829     if (len == 12) {
830         memcpy(ctx->Yi.c, iv, 12);
831         ctx->Yi.c[12] = 0;
832         ctx->Yi.c[13] = 0;
833         ctx->Yi.c[14] = 0;
834         ctx->Yi.c[15] = 1;
835         ctr = 1;
836     } else {
837         size_t i;
838         u64 len0 = len;
839 
840         /* Borrow ctx->Xi to calculate initial Yi */
841         ctx->Xi.u[0] = 0;
842         ctx->Xi.u[1] = 0;
843 
844         while (len >= 16) {
845             for (i = 0; i < 16; ++i)
846                 ctx->Xi.c[i] ^= iv[i];
847             GCM_MUL(ctx);
848             iv += 16;
849             len -= 16;
850         }
851         if (len) {
852             for (i = 0; i < len; ++i)
853                 ctx->Xi.c[i] ^= iv[i];
854             GCM_MUL(ctx);
855         }
856         len0 <<= 3;
857         if (IS_LITTLE_ENDIAN) {
858 #ifdef BSWAP8
859             ctx->Xi.u[1] ^= BSWAP8(len0);
860 #else
861             ctx->Xi.c[8] ^= (u8)(len0 >> 56);
862             ctx->Xi.c[9] ^= (u8)(len0 >> 48);
863             ctx->Xi.c[10] ^= (u8)(len0 >> 40);
864             ctx->Xi.c[11] ^= (u8)(len0 >> 32);
865             ctx->Xi.c[12] ^= (u8)(len0 >> 24);
866             ctx->Xi.c[13] ^= (u8)(len0 >> 16);
867             ctx->Xi.c[14] ^= (u8)(len0 >> 8);
868             ctx->Xi.c[15] ^= (u8)(len0);
869 #endif
870         } else {
871             ctx->Xi.u[1] ^= len0;
872         }
873 
874         GCM_MUL(ctx);
875 
876         if (IS_LITTLE_ENDIAN)
877 #ifdef BSWAP4
878             ctr = BSWAP4(ctx->Xi.d[3]);
879 #else
880             ctr = GETU32(ctx->Xi.c + 12);
881 #endif
882         else
883             ctr = ctx->Xi.d[3];
884 
885         /* Copy borrowed Xi to Yi */
886         ctx->Yi.u[0] = ctx->Xi.u[0];
887         ctx->Yi.u[1] = ctx->Xi.u[1];
888     }
889 
890     ctx->Xi.u[0] = 0;
891     ctx->Xi.u[1] = 0;
892 
893     (*ctx->block) (ctx->Yi.c, ctx->EK0.c, ctx->key);
894     ++ctr;
895     if (IS_LITTLE_ENDIAN)
896 #ifdef BSWAP4
897         ctx->Yi.d[3] = BSWAP4(ctr);
898 #else
899         PUTU32(ctx->Yi.c + 12, ctr);
900 #endif
901     else
902         ctx->Yi.d[3] = ctr;
903 }
904 
905 int CRYPTO_gcm128_aad(GCM128_CONTEXT *ctx, const unsigned char *aad,
906                       size_t len)
907 {
908     size_t i;
909     unsigned int n;
910     u64 alen = ctx->len.u[0];
911 #ifdef GCM_FUNCREF_4BIT
912     void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult;
913 # ifdef GHASH
914     void (*gcm_ghash_p) (u64 Xi[2], const u128 Htable[16],
915                          const u8 *inp, size_t len) = ctx->ghash;
916 # endif
917 #endif
918 
919     if (ctx->len.u[1])
920         return -2;
921 
922     alen += len;
923     if (alen > (U64(1) << 61) || (sizeof(len) == 8 && alen < len))
924         return -1;
925     ctx->len.u[0] = alen;
926 
927     n = ctx->ares;
928     if (n) {
929         while (n && len) {
930             ctx->Xi.c[n] ^= *(aad++);
931             --len;
932             n = (n + 1) % 16;
933         }
934         if (n == 0)
935             GCM_MUL(ctx);
936         else {
937             ctx->ares = n;
938             return 0;
939         }
940     }
941 #ifdef GHASH
942     if ((i = (len & (size_t)-16))) {
943         GHASH(ctx, aad, i);
944         aad += i;
945         len -= i;
946     }
947 #else
948     while (len >= 16) {
949         for (i = 0; i < 16; ++i)
950             ctx->Xi.c[i] ^= aad[i];
951         GCM_MUL(ctx);
952         aad += 16;
953         len -= 16;
954     }
955 #endif
956     if (len) {
957         n = (unsigned int)len;
958         for (i = 0; i < len; ++i)
959             ctx->Xi.c[i] ^= aad[i];
960     }
961 
962     ctx->ares = n;
963     return 0;
964 }
965 
966 int CRYPTO_gcm128_encrypt(GCM128_CONTEXT *ctx,
967                           const unsigned char *in, unsigned char *out,
968                           size_t len)
969 {
970     DECLARE_IS_ENDIAN;
971     unsigned int n, ctr, mres;
972     size_t i;
973     u64 mlen = ctx->len.u[1];
974     block128_f block = ctx->block;
975     void *key = ctx->key;
976 #ifdef GCM_FUNCREF_4BIT
977     void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult;
978 # if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
979     void (*gcm_ghash_p) (u64 Xi[2], const u128 Htable[16],
980                          const u8 *inp, size_t len) = ctx->ghash;
981 # endif
982 #endif
983 
984     mlen += len;
985     if (mlen > ((U64(1) << 36) - 32) || (sizeof(len) == 8 && mlen < len))
986         return -1;
987     ctx->len.u[1] = mlen;
988 
989     mres = ctx->mres;
990 
991     if (ctx->ares) {
992         /* First call to encrypt finalizes GHASH(AAD) */
993 #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
994         if (len == 0) {
995             GCM_MUL(ctx);
996             ctx->ares = 0;
997             return 0;
998         }
999         memcpy(ctx->Xn, ctx->Xi.c, sizeof(ctx->Xi));
1000         ctx->Xi.u[0] = 0;
1001         ctx->Xi.u[1] = 0;
1002         mres = sizeof(ctx->Xi);
1003 #else
1004         GCM_MUL(ctx);
1005 #endif
1006         ctx->ares = 0;
1007     }
1008 
1009     if (IS_LITTLE_ENDIAN)
1010 #ifdef BSWAP4
1011         ctr = BSWAP4(ctx->Yi.d[3]);
1012 #else
1013         ctr = GETU32(ctx->Yi.c + 12);
1014 #endif
1015     else
1016         ctr = ctx->Yi.d[3];
1017 
1018     n = mres % 16;
1019 #if !defined(OPENSSL_SMALL_FOOTPRINT)
1020     if (16 % sizeof(size_t) == 0) { /* always true actually */
1021         do {
1022             if (n) {
1023 # if defined(GHASH)
1024                 while (n && len) {
1025                     ctx->Xn[mres++] = *(out++) = *(in++) ^ ctx->EKi.c[n];
1026                     --len;
1027                     n = (n + 1) % 16;
1028                 }
1029                 if (n == 0) {
1030                     GHASH(ctx, ctx->Xn, mres);
1031                     mres = 0;
1032                 } else {
1033                     ctx->mres = mres;
1034                     return 0;
1035                 }
1036 # else
1037                 while (n && len) {
1038                     ctx->Xi.c[n] ^= *(out++) = *(in++) ^ ctx->EKi.c[n];
1039                     --len;
1040                     n = (n + 1) % 16;
1041                 }
1042                 if (n == 0) {
1043                     GCM_MUL(ctx);
1044                     mres = 0;
1045                 } else {
1046                     ctx->mres = n;
1047                     return 0;
1048                 }
1049 # endif
1050             }
1051 # if defined(STRICT_ALIGNMENT)
1052             if (((size_t)in | (size_t)out) % sizeof(size_t) != 0)
1053                 break;
1054 # endif
1055 # if defined(GHASH)
1056             if (len >= 16 && mres) {
1057                 GHASH(ctx, ctx->Xn, mres);
1058                 mres = 0;
1059             }
1060 #  if defined(GHASH_CHUNK)
1061             while (len >= GHASH_CHUNK) {
1062                 size_t j = GHASH_CHUNK;
1063 
1064                 while (j) {
1065                     size_t_aX *out_t = (size_t_aX *)out;
1066                     const size_t_aX *in_t = (const size_t_aX *)in;
1067 
1068                     (*block) (ctx->Yi.c, ctx->EKi.c, key);
1069                     ++ctr;
1070                     if (IS_LITTLE_ENDIAN)
1071 #   ifdef BSWAP4
1072                         ctx->Yi.d[3] = BSWAP4(ctr);
1073 #   else
1074                         PUTU32(ctx->Yi.c + 12, ctr);
1075 #   endif
1076                     else
1077                         ctx->Yi.d[3] = ctr;
1078                     for (i = 0; i < 16 / sizeof(size_t); ++i)
1079                         out_t[i] = in_t[i] ^ ctx->EKi.t[i];
1080                     out += 16;
1081                     in += 16;
1082                     j -= 16;
1083                 }
1084                 GHASH(ctx, out - GHASH_CHUNK, GHASH_CHUNK);
1085                 len -= GHASH_CHUNK;
1086             }
1087 #  endif
1088             if ((i = (len & (size_t)-16))) {
1089                 size_t j = i;
1090 
1091                 while (len >= 16) {
1092                     size_t_aX *out_t = (size_t_aX *)out;
1093                     const size_t_aX *in_t = (const size_t_aX *)in;
1094 
1095                     (*block) (ctx->Yi.c, ctx->EKi.c, key);
1096                     ++ctr;
1097                     if (IS_LITTLE_ENDIAN)
1098 #  ifdef BSWAP4
1099                         ctx->Yi.d[3] = BSWAP4(ctr);
1100 #  else
1101                         PUTU32(ctx->Yi.c + 12, ctr);
1102 #  endif
1103                     else
1104                         ctx->Yi.d[3] = ctr;
1105                     for (i = 0; i < 16 / sizeof(size_t); ++i)
1106                         out_t[i] = in_t[i] ^ ctx->EKi.t[i];
1107                     out += 16;
1108                     in += 16;
1109                     len -= 16;
1110                 }
1111                 GHASH(ctx, out - j, j);
1112             }
1113 # else
1114             while (len >= 16) {
1115                 size_t *out_t = (size_t *)out;
1116                 const size_t *in_t = (const size_t *)in;
1117 
1118                 (*block) (ctx->Yi.c, ctx->EKi.c, key);
1119                 ++ctr;
1120                 if (IS_LITTLE_ENDIAN)
1121 #  ifdef BSWAP4
1122                     ctx->Yi.d[3] = BSWAP4(ctr);
1123 #  else
1124                     PUTU32(ctx->Yi.c + 12, ctr);
1125 #  endif
1126                 else
1127                     ctx->Yi.d[3] = ctr;
1128                 for (i = 0; i < 16 / sizeof(size_t); ++i)
1129                     ctx->Xi.t[i] ^= out_t[i] = in_t[i] ^ ctx->EKi.t[i];
1130                 GCM_MUL(ctx);
1131                 out += 16;
1132                 in += 16;
1133                 len -= 16;
1134             }
1135 # endif
1136             if (len) {
1137                 (*block) (ctx->Yi.c, ctx->EKi.c, key);
1138                 ++ctr;
1139                 if (IS_LITTLE_ENDIAN)
1140 # ifdef BSWAP4
1141                     ctx->Yi.d[3] = BSWAP4(ctr);
1142 # else
1143                     PUTU32(ctx->Yi.c + 12, ctr);
1144 # endif
1145                 else
1146                     ctx->Yi.d[3] = ctr;
1147 # if defined(GHASH)
1148                 while (len--) {
1149                     ctx->Xn[mres++] = out[n] = in[n] ^ ctx->EKi.c[n];
1150                     ++n;
1151                 }
1152 # else
1153                 while (len--) {
1154                     ctx->Xi.c[n] ^= out[n] = in[n] ^ ctx->EKi.c[n];
1155                     ++n;
1156                 }
1157                 mres = n;
1158 # endif
1159             }
1160 
1161             ctx->mres = mres;
1162             return 0;
1163         } while (0);
1164     }
1165 #endif
1166     for (i = 0; i < len; ++i) {
1167         if (n == 0) {
1168             (*block) (ctx->Yi.c, ctx->EKi.c, key);
1169             ++ctr;
1170             if (IS_LITTLE_ENDIAN)
1171 #ifdef BSWAP4
1172                 ctx->Yi.d[3] = BSWAP4(ctr);
1173 #else
1174                 PUTU32(ctx->Yi.c + 12, ctr);
1175 #endif
1176             else
1177                 ctx->Yi.d[3] = ctr;
1178         }
1179 #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
1180         ctx->Xn[mres++] = out[i] = in[i] ^ ctx->EKi.c[n];
1181         n = (n + 1) % 16;
1182         if (mres == sizeof(ctx->Xn)) {
1183             GHASH(ctx,ctx->Xn,sizeof(ctx->Xn));
1184             mres = 0;
1185         }
1186 #else
1187         ctx->Xi.c[n] ^= out[i] = in[i] ^ ctx->EKi.c[n];
1188         mres = n = (n + 1) % 16;
1189         if (n == 0)
1190             GCM_MUL(ctx);
1191 #endif
1192     }
1193 
1194     ctx->mres = mres;
1195     return 0;
1196 }
1197 
1198 int CRYPTO_gcm128_decrypt(GCM128_CONTEXT *ctx,
1199                           const unsigned char *in, unsigned char *out,
1200                           size_t len)
1201 {
1202     DECLARE_IS_ENDIAN;
1203     unsigned int n, ctr, mres;
1204     size_t i;
1205     u64 mlen = ctx->len.u[1];
1206     block128_f block = ctx->block;
1207     void *key = ctx->key;
1208 #ifdef GCM_FUNCREF_4BIT
1209     void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult;
1210 # if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
1211     void (*gcm_ghash_p) (u64 Xi[2], const u128 Htable[16],
1212                          const u8 *inp, size_t len) = ctx->ghash;
1213 # endif
1214 #endif
1215 
1216     mlen += len;
1217     if (mlen > ((U64(1) << 36) - 32) || (sizeof(len) == 8 && mlen < len))
1218         return -1;
1219     ctx->len.u[1] = mlen;
1220 
1221     mres = ctx->mres;
1222 
1223     if (ctx->ares) {
1224         /* First call to decrypt finalizes GHASH(AAD) */
1225 #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
1226         if (len == 0) {
1227             GCM_MUL(ctx);
1228             ctx->ares = 0;
1229             return 0;
1230         }
1231         memcpy(ctx->Xn, ctx->Xi.c, sizeof(ctx->Xi));
1232         ctx->Xi.u[0] = 0;
1233         ctx->Xi.u[1] = 0;
1234         mres = sizeof(ctx->Xi);
1235 #else
1236         GCM_MUL(ctx);
1237 #endif
1238         ctx->ares = 0;
1239     }
1240 
1241     if (IS_LITTLE_ENDIAN)
1242 #ifdef BSWAP4
1243         ctr = BSWAP4(ctx->Yi.d[3]);
1244 #else
1245         ctr = GETU32(ctx->Yi.c + 12);
1246 #endif
1247     else
1248         ctr = ctx->Yi.d[3];
1249 
1250     n = mres % 16;
1251 #if !defined(OPENSSL_SMALL_FOOTPRINT)
1252     if (16 % sizeof(size_t) == 0) { /* always true actually */
1253         do {
1254             if (n) {
1255 # if defined(GHASH)
1256                 while (n && len) {
1257                     *(out++) = (ctx->Xn[mres++] = *(in++)) ^ ctx->EKi.c[n];
1258                     --len;
1259                     n = (n + 1) % 16;
1260                 }
1261                 if (n == 0) {
1262                     GHASH(ctx, ctx->Xn, mres);
1263                     mres = 0;
1264                 } else {
1265                     ctx->mres = mres;
1266                     return 0;
1267                 }
1268 # else
1269                 while (n && len) {
1270                     u8 c = *(in++);
1271                     *(out++) = c ^ ctx->EKi.c[n];
1272                     ctx->Xi.c[n] ^= c;
1273                     --len;
1274                     n = (n + 1) % 16;
1275                 }
1276                 if (n == 0) {
1277                     GCM_MUL(ctx);
1278                     mres = 0;
1279                 } else {
1280                     ctx->mres = n;
1281                     return 0;
1282                 }
1283 # endif
1284             }
1285 # if defined(STRICT_ALIGNMENT)
1286             if (((size_t)in | (size_t)out) % sizeof(size_t) != 0)
1287                 break;
1288 # endif
1289 # if defined(GHASH)
1290             if (len >= 16 && mres) {
1291                 GHASH(ctx, ctx->Xn, mres);
1292                 mres = 0;
1293             }
1294 #  if defined(GHASH_CHUNK)
1295             while (len >= GHASH_CHUNK) {
1296                 size_t j = GHASH_CHUNK;
1297 
1298                 GHASH(ctx, in, GHASH_CHUNK);
1299                 while (j) {
1300                     size_t_aX *out_t = (size_t_aX *)out;
1301                     const size_t_aX *in_t = (const size_t_aX *)in;
1302 
1303                     (*block) (ctx->Yi.c, ctx->EKi.c, key);
1304                     ++ctr;
1305                     if (IS_LITTLE_ENDIAN)
1306 #   ifdef BSWAP4
1307                         ctx->Yi.d[3] = BSWAP4(ctr);
1308 #   else
1309                         PUTU32(ctx->Yi.c + 12, ctr);
1310 #   endif
1311                     else
1312                         ctx->Yi.d[3] = ctr;
1313                     for (i = 0; i < 16 / sizeof(size_t); ++i)
1314                         out_t[i] = in_t[i] ^ ctx->EKi.t[i];
1315                     out += 16;
1316                     in += 16;
1317                     j -= 16;
1318                 }
1319                 len -= GHASH_CHUNK;
1320             }
1321 #  endif
1322             if ((i = (len & (size_t)-16))) {
1323                 GHASH(ctx, in, i);
1324                 while (len >= 16) {
1325                     size_t_aX *out_t = (size_t_aX *)out;
1326                     const size_t_aX *in_t = (const size_t_aX *)in;
1327 
1328                     (*block) (ctx->Yi.c, ctx->EKi.c, key);
1329                     ++ctr;
1330                     if (IS_LITTLE_ENDIAN)
1331 #  ifdef BSWAP4
1332                         ctx->Yi.d[3] = BSWAP4(ctr);
1333 #  else
1334                         PUTU32(ctx->Yi.c + 12, ctr);
1335 #  endif
1336                     else
1337                         ctx->Yi.d[3] = ctr;
1338                     for (i = 0; i < 16 / sizeof(size_t); ++i)
1339                         out_t[i] = in_t[i] ^ ctx->EKi.t[i];
1340                     out += 16;
1341                     in += 16;
1342                     len -= 16;
1343                 }
1344             }
1345 # else
1346             while (len >= 16) {
1347                 size_t *out_t = (size_t *)out;
1348                 const size_t *in_t = (const size_t *)in;
1349 
1350                 (*block) (ctx->Yi.c, ctx->EKi.c, key);
1351                 ++ctr;
1352                 if (IS_LITTLE_ENDIAN)
1353 #  ifdef BSWAP4
1354                     ctx->Yi.d[3] = BSWAP4(ctr);
1355 #  else
1356                     PUTU32(ctx->Yi.c + 12, ctr);
1357 #  endif
1358                 else
1359                     ctx->Yi.d[3] = ctr;
1360                 for (i = 0; i < 16 / sizeof(size_t); ++i) {
1361                     size_t c = in_t[i];
1362                     out_t[i] = c ^ ctx->EKi.t[i];
1363                     ctx->Xi.t[i] ^= c;
1364                 }
1365                 GCM_MUL(ctx);
1366                 out += 16;
1367                 in += 16;
1368                 len -= 16;
1369             }
1370 # endif
1371             if (len) {
1372                 (*block) (ctx->Yi.c, ctx->EKi.c, key);
1373                 ++ctr;
1374                 if (IS_LITTLE_ENDIAN)
1375 # ifdef BSWAP4
1376                     ctx->Yi.d[3] = BSWAP4(ctr);
1377 # else
1378                     PUTU32(ctx->Yi.c + 12, ctr);
1379 # endif
1380                 else
1381                     ctx->Yi.d[3] = ctr;
1382 # if defined(GHASH)
1383                 while (len--) {
1384                     out[n] = (ctx->Xn[mres++] = in[n]) ^ ctx->EKi.c[n];
1385                     ++n;
1386                 }
1387 # else
1388                 while (len--) {
1389                     u8 c = in[n];
1390                     ctx->Xi.c[n] ^= c;
1391                     out[n] = c ^ ctx->EKi.c[n];
1392                     ++n;
1393                 }
1394                 mres = n;
1395 # endif
1396             }
1397 
1398             ctx->mres = mres;
1399             return 0;
1400         } while (0);
1401     }
1402 #endif
1403     for (i = 0; i < len; ++i) {
1404         u8 c;
1405         if (n == 0) {
1406             (*block) (ctx->Yi.c, ctx->EKi.c, key);
1407             ++ctr;
1408             if (IS_LITTLE_ENDIAN)
1409 #ifdef BSWAP4
1410                 ctx->Yi.d[3] = BSWAP4(ctr);
1411 #else
1412                 PUTU32(ctx->Yi.c + 12, ctr);
1413 #endif
1414             else
1415                 ctx->Yi.d[3] = ctr;
1416         }
1417 #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
1418         out[i] = (ctx->Xn[mres++] = c = in[i]) ^ ctx->EKi.c[n];
1419         n = (n + 1) % 16;
1420         if (mres == sizeof(ctx->Xn)) {
1421             GHASH(ctx,ctx->Xn,sizeof(ctx->Xn));
1422             mres = 0;
1423         }
1424 #else
1425         c = in[i];
1426         out[i] = c ^ ctx->EKi.c[n];
1427         ctx->Xi.c[n] ^= c;
1428         mres = n = (n + 1) % 16;
1429         if (n == 0)
1430             GCM_MUL(ctx);
1431 #endif
1432     }
1433 
1434     ctx->mres = mres;
1435     return 0;
1436 }
1437 
1438 int CRYPTO_gcm128_encrypt_ctr32(GCM128_CONTEXT *ctx,
1439                                 const unsigned char *in, unsigned char *out,
1440                                 size_t len, ctr128_f stream)
1441 {
1442 #if defined(OPENSSL_SMALL_FOOTPRINT)
1443     return CRYPTO_gcm128_encrypt(ctx, in, out, len);
1444 #else
1445     DECLARE_IS_ENDIAN;
1446     unsigned int n, ctr, mres;
1447     size_t i;
1448     u64 mlen = ctx->len.u[1];
1449     void *key = ctx->key;
1450 # ifdef GCM_FUNCREF_4BIT
1451     void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult;
1452 #  ifdef GHASH
1453     void (*gcm_ghash_p) (u64 Xi[2], const u128 Htable[16],
1454                          const u8 *inp, size_t len) = ctx->ghash;
1455 #  endif
1456 # endif
1457 
1458     mlen += len;
1459     if (mlen > ((U64(1) << 36) - 32) || (sizeof(len) == 8 && mlen < len))
1460         return -1;
1461     ctx->len.u[1] = mlen;
1462 
1463     mres = ctx->mres;
1464 
1465     if (ctx->ares) {
1466         /* First call to encrypt finalizes GHASH(AAD) */
1467 #if defined(GHASH)
1468         if (len == 0) {
1469             GCM_MUL(ctx);
1470             ctx->ares = 0;
1471             return 0;
1472         }
1473         memcpy(ctx->Xn, ctx->Xi.c, sizeof(ctx->Xi));
1474         ctx->Xi.u[0] = 0;
1475         ctx->Xi.u[1] = 0;
1476         mres = sizeof(ctx->Xi);
1477 #else
1478         GCM_MUL(ctx);
1479 #endif
1480         ctx->ares = 0;
1481     }
1482 
1483     if (IS_LITTLE_ENDIAN)
1484 # ifdef BSWAP4
1485         ctr = BSWAP4(ctx->Yi.d[3]);
1486 # else
1487         ctr = GETU32(ctx->Yi.c + 12);
1488 # endif
1489     else
1490         ctr = ctx->Yi.d[3];
1491 
1492     n = mres % 16;
1493     if (n) {
1494 # if defined(GHASH)
1495         while (n && len) {
1496             ctx->Xn[mres++] = *(out++) = *(in++) ^ ctx->EKi.c[n];
1497             --len;
1498             n = (n + 1) % 16;
1499         }
1500         if (n == 0) {
1501             GHASH(ctx, ctx->Xn, mres);
1502             mres = 0;
1503         } else {
1504             ctx->mres = mres;
1505             return 0;
1506         }
1507 # else
1508         while (n && len) {
1509             ctx->Xi.c[n] ^= *(out++) = *(in++) ^ ctx->EKi.c[n];
1510             --len;
1511             n = (n + 1) % 16;
1512         }
1513         if (n == 0) {
1514             GCM_MUL(ctx);
1515             mres = 0;
1516         } else {
1517             ctx->mres = n;
1518             return 0;
1519         }
1520 # endif
1521     }
1522 # if defined(GHASH)
1523         if (len >= 16 && mres) {
1524             GHASH(ctx, ctx->Xn, mres);
1525             mres = 0;
1526         }
1527 #  if defined(GHASH_CHUNK)
1528     while (len >= GHASH_CHUNK) {
1529         (*stream) (in, out, GHASH_CHUNK / 16, key, ctx->Yi.c);
1530         ctr += GHASH_CHUNK / 16;
1531         if (IS_LITTLE_ENDIAN)
1532 #   ifdef BSWAP4
1533             ctx->Yi.d[3] = BSWAP4(ctr);
1534 #   else
1535             PUTU32(ctx->Yi.c + 12, ctr);
1536 #   endif
1537         else
1538             ctx->Yi.d[3] = ctr;
1539         GHASH(ctx, out, GHASH_CHUNK);
1540         out += GHASH_CHUNK;
1541         in += GHASH_CHUNK;
1542         len -= GHASH_CHUNK;
1543     }
1544 #  endif
1545 # endif
1546     if ((i = (len & (size_t)-16))) {
1547         size_t j = i / 16;
1548 
1549         (*stream) (in, out, j, key, ctx->Yi.c);
1550         ctr += (unsigned int)j;
1551         if (IS_LITTLE_ENDIAN)
1552 # ifdef BSWAP4
1553             ctx->Yi.d[3] = BSWAP4(ctr);
1554 # else
1555             PUTU32(ctx->Yi.c + 12, ctr);
1556 # endif
1557         else
1558             ctx->Yi.d[3] = ctr;
1559         in += i;
1560         len -= i;
1561 # if defined(GHASH)
1562         GHASH(ctx, out, i);
1563         out += i;
1564 # else
1565         while (j--) {
1566             for (i = 0; i < 16; ++i)
1567                 ctx->Xi.c[i] ^= out[i];
1568             GCM_MUL(ctx);
1569             out += 16;
1570         }
1571 # endif
1572     }
1573     if (len) {
1574         (*ctx->block) (ctx->Yi.c, ctx->EKi.c, key);
1575         ++ctr;
1576         if (IS_LITTLE_ENDIAN)
1577 # ifdef BSWAP4
1578             ctx->Yi.d[3] = BSWAP4(ctr);
1579 # else
1580             PUTU32(ctx->Yi.c + 12, ctr);
1581 # endif
1582         else
1583             ctx->Yi.d[3] = ctr;
1584         while (len--) {
1585 # if defined(GHASH)
1586             ctx->Xn[mres++] = out[n] = in[n] ^ ctx->EKi.c[n];
1587 # else
1588             ctx->Xi.c[mres++] ^= out[n] = in[n] ^ ctx->EKi.c[n];
1589 # endif
1590             ++n;
1591         }
1592     }
1593 
1594     ctx->mres = mres;
1595     return 0;
1596 #endif
1597 }
1598 
1599 int CRYPTO_gcm128_decrypt_ctr32(GCM128_CONTEXT *ctx,
1600                                 const unsigned char *in, unsigned char *out,
1601                                 size_t len, ctr128_f stream)
1602 {
1603 #if defined(OPENSSL_SMALL_FOOTPRINT)
1604     return CRYPTO_gcm128_decrypt(ctx, in, out, len);
1605 #else
1606     DECLARE_IS_ENDIAN;
1607     unsigned int n, ctr, mres;
1608     size_t i;
1609     u64 mlen = ctx->len.u[1];
1610     void *key = ctx->key;
1611 # ifdef GCM_FUNCREF_4BIT
1612     void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult;
1613 #  ifdef GHASH
1614     void (*gcm_ghash_p) (u64 Xi[2], const u128 Htable[16],
1615                          const u8 *inp, size_t len) = ctx->ghash;
1616 #  endif
1617 # endif
1618 
1619     mlen += len;
1620     if (mlen > ((U64(1) << 36) - 32) || (sizeof(len) == 8 && mlen < len))
1621         return -1;
1622     ctx->len.u[1] = mlen;
1623 
1624     mres = ctx->mres;
1625 
1626     if (ctx->ares) {
1627         /* First call to decrypt finalizes GHASH(AAD) */
1628 # if defined(GHASH)
1629         if (len == 0) {
1630             GCM_MUL(ctx);
1631             ctx->ares = 0;
1632             return 0;
1633         }
1634         memcpy(ctx->Xn, ctx->Xi.c, sizeof(ctx->Xi));
1635         ctx->Xi.u[0] = 0;
1636         ctx->Xi.u[1] = 0;
1637         mres = sizeof(ctx->Xi);
1638 # else
1639         GCM_MUL(ctx);
1640 # endif
1641         ctx->ares = 0;
1642     }
1643 
1644     if (IS_LITTLE_ENDIAN)
1645 # ifdef BSWAP4
1646         ctr = BSWAP4(ctx->Yi.d[3]);
1647 # else
1648         ctr = GETU32(ctx->Yi.c + 12);
1649 # endif
1650     else
1651         ctr = ctx->Yi.d[3];
1652 
1653     n = mres % 16;
1654     if (n) {
1655 # if defined(GHASH)
1656         while (n && len) {
1657             *(out++) = (ctx->Xn[mres++] = *(in++)) ^ ctx->EKi.c[n];
1658             --len;
1659             n = (n + 1) % 16;
1660         }
1661         if (n == 0) {
1662             GHASH(ctx, ctx->Xn, mres);
1663             mres = 0;
1664         } else {
1665             ctx->mres = mres;
1666             return 0;
1667         }
1668 # else
1669         while (n && len) {
1670             u8 c = *(in++);
1671             *(out++) = c ^ ctx->EKi.c[n];
1672             ctx->Xi.c[n] ^= c;
1673             --len;
1674             n = (n + 1) % 16;
1675         }
1676         if (n == 0) {
1677             GCM_MUL(ctx);
1678             mres = 0;
1679         } else {
1680             ctx->mres = n;
1681             return 0;
1682         }
1683 # endif
1684     }
1685 # if defined(GHASH)
1686     if (len >= 16 && mres) {
1687         GHASH(ctx, ctx->Xn, mres);
1688         mres = 0;
1689     }
1690 #  if defined(GHASH_CHUNK)
1691     while (len >= GHASH_CHUNK) {
1692         GHASH(ctx, in, GHASH_CHUNK);
1693         (*stream) (in, out, GHASH_CHUNK / 16, key, ctx->Yi.c);
1694         ctr += GHASH_CHUNK / 16;
1695         if (IS_LITTLE_ENDIAN)
1696 #   ifdef BSWAP4
1697             ctx->Yi.d[3] = BSWAP4(ctr);
1698 #   else
1699             PUTU32(ctx->Yi.c + 12, ctr);
1700 #   endif
1701         else
1702             ctx->Yi.d[3] = ctr;
1703         out += GHASH_CHUNK;
1704         in += GHASH_CHUNK;
1705         len -= GHASH_CHUNK;
1706     }
1707 #  endif
1708 # endif
1709     if ((i = (len & (size_t)-16))) {
1710         size_t j = i / 16;
1711 
1712 # if defined(GHASH)
1713         GHASH(ctx, in, i);
1714 # else
1715         while (j--) {
1716             size_t k;
1717             for (k = 0; k < 16; ++k)
1718                 ctx->Xi.c[k] ^= in[k];
1719             GCM_MUL(ctx);
1720             in += 16;
1721         }
1722         j = i / 16;
1723         in -= i;
1724 # endif
1725         (*stream) (in, out, j, key, ctx->Yi.c);
1726         ctr += (unsigned int)j;
1727         if (IS_LITTLE_ENDIAN)
1728 # ifdef BSWAP4
1729             ctx->Yi.d[3] = BSWAP4(ctr);
1730 # else
1731             PUTU32(ctx->Yi.c + 12, ctr);
1732 # endif
1733         else
1734             ctx->Yi.d[3] = ctr;
1735         out += i;
1736         in += i;
1737         len -= i;
1738     }
1739     if (len) {
1740         (*ctx->block) (ctx->Yi.c, ctx->EKi.c, key);
1741         ++ctr;
1742         if (IS_LITTLE_ENDIAN)
1743 # ifdef BSWAP4
1744             ctx->Yi.d[3] = BSWAP4(ctr);
1745 # else
1746             PUTU32(ctx->Yi.c + 12, ctr);
1747 # endif
1748         else
1749             ctx->Yi.d[3] = ctr;
1750         while (len--) {
1751 # if defined(GHASH)
1752             out[n] = (ctx->Xn[mres++] = in[n]) ^ ctx->EKi.c[n];
1753 # else
1754             u8 c = in[n];
1755             ctx->Xi.c[mres++] ^= c;
1756             out[n] = c ^ ctx->EKi.c[n];
1757 # endif
1758             ++n;
1759         }
1760     }
1761 
1762     ctx->mres = mres;
1763     return 0;
1764 #endif
1765 }
1766 
1767 int CRYPTO_gcm128_finish(GCM128_CONTEXT *ctx, const unsigned char *tag,
1768                          size_t len)
1769 {
1770     DECLARE_IS_ENDIAN;
1771     u64 alen = ctx->len.u[0] << 3;
1772     u64 clen = ctx->len.u[1] << 3;
1773 #ifdef GCM_FUNCREF_4BIT
1774     void (*gcm_gmult_p) (u64 Xi[2], const u128 Htable[16]) = ctx->gmult;
1775 # if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
1776     void (*gcm_ghash_p) (u64 Xi[2], const u128 Htable[16],
1777                          const u8 *inp, size_t len) = ctx->ghash;
1778 # endif
1779 #endif
1780 
1781 #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
1782     u128 bitlen;
1783     unsigned int mres = ctx->mres;
1784 
1785     if (mres) {
1786         unsigned blocks = (mres + 15) & -16;
1787 
1788         memset(ctx->Xn + mres, 0, blocks - mres);
1789         mres = blocks;
1790         if (mres == sizeof(ctx->Xn)) {
1791             GHASH(ctx, ctx->Xn, mres);
1792             mres = 0;
1793         }
1794     } else if (ctx->ares) {
1795         GCM_MUL(ctx);
1796     }
1797 #else
1798     if (ctx->mres || ctx->ares)
1799         GCM_MUL(ctx);
1800 #endif
1801 
1802     if (IS_LITTLE_ENDIAN) {
1803 #ifdef BSWAP8
1804         alen = BSWAP8(alen);
1805         clen = BSWAP8(clen);
1806 #else
1807         u8 *p = ctx->len.c;
1808 
1809         ctx->len.u[0] = alen;
1810         ctx->len.u[1] = clen;
1811 
1812         alen = (u64)GETU32(p) << 32 | GETU32(p + 4);
1813         clen = (u64)GETU32(p + 8) << 32 | GETU32(p + 12);
1814 #endif
1815     }
1816 
1817 #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT)
1818     bitlen.hi = alen;
1819     bitlen.lo = clen;
1820     memcpy(ctx->Xn + mres, &bitlen, sizeof(bitlen));
1821     mres += sizeof(bitlen);
1822     GHASH(ctx, ctx->Xn, mres);
1823 #else
1824     ctx->Xi.u[0] ^= alen;
1825     ctx->Xi.u[1] ^= clen;
1826     GCM_MUL(ctx);
1827 #endif
1828 
1829     ctx->Xi.u[0] ^= ctx->EK0.u[0];
1830     ctx->Xi.u[1] ^= ctx->EK0.u[1];
1831 
1832     if (tag && len <= sizeof(ctx->Xi))
1833         return CRYPTO_memcmp(ctx->Xi.c, tag, len);
1834     else
1835         return -1;
1836 }
1837 
1838 void CRYPTO_gcm128_tag(GCM128_CONTEXT *ctx, unsigned char *tag, size_t len)
1839 {
1840     CRYPTO_gcm128_finish(ctx, NULL, 0);
1841     memcpy(tag, ctx->Xi.c,
1842            len <= sizeof(ctx->Xi.c) ? len : sizeof(ctx->Xi.c));
1843 }
1844 
1845 GCM128_CONTEXT *CRYPTO_gcm128_new(void *key, block128_f block)
1846 {
1847     GCM128_CONTEXT *ret;
1848 
1849     if ((ret = OPENSSL_malloc(sizeof(*ret))) != NULL)
1850         CRYPTO_gcm128_init(ret, key, block);
1851 
1852     return ret;
1853 }
1854 
1855 void CRYPTO_gcm128_release(GCM128_CONTEXT *ctx)
1856 {
1857     OPENSSL_clear_free(ctx, sizeof(*ctx));
1858 }
1859