1/*
2 * Accelerated GHASH implementation with Intel PCLMULQDQ-NI
3 * instructions. This file contains glue code.
4 *
5 * Copyright (c) 2009 Intel Corp.
6 *   Author: Huang Ying <ying.huang@intel.com>
7 *
8 * This program is free software; you can redistribute it and/or modify it
9 * under the terms of the GNU General Public License version 2 as published
10 * by the Free Software Foundation.
11 */
12
13#include <linux/err.h>
14#include <linux/module.h>
15#include <linux/init.h>
16#include <linux/kernel.h>
17#include <linux/crypto.h>
18#include <crypto/algapi.h>
19#include <crypto/cryptd.h>
20#include <crypto/gf128mul.h>
21#include <crypto/internal/hash.h>
22#include <asm/i387.h>
23#include <asm/cpu_device_id.h>
24
25#define GHASH_BLOCK_SIZE	16
26#define GHASH_DIGEST_SIZE	16
27
28void clmul_ghash_mul(char *dst, const u128 *shash);
29
30void clmul_ghash_update(char *dst, const char *src, unsigned int srclen,
31			const u128 *shash);
32
33struct ghash_async_ctx {
34	struct cryptd_ahash *cryptd_tfm;
35};
36
37struct ghash_ctx {
38	u128 shash;
39};
40
41struct ghash_desc_ctx {
42	u8 buffer[GHASH_BLOCK_SIZE];
43	u32 bytes;
44};
45
46static int ghash_init(struct shash_desc *desc)
47{
48	struct ghash_desc_ctx *dctx = shash_desc_ctx(desc);
49
50	memset(dctx, 0, sizeof(*dctx));
51
52	return 0;
53}
54
55static int ghash_setkey(struct crypto_shash *tfm,
56			const u8 *key, unsigned int keylen)
57{
58	struct ghash_ctx *ctx = crypto_shash_ctx(tfm);
59	be128 *x = (be128 *)key;
60	u64 a, b;
61
62	if (keylen != GHASH_BLOCK_SIZE) {
63		crypto_shash_set_flags(tfm, CRYPTO_TFM_RES_BAD_KEY_LEN);
64		return -EINVAL;
65	}
66
67	/* perform multiplication by 'x' in GF(2^128) */
68	a = be64_to_cpu(x->a);
69	b = be64_to_cpu(x->b);
70
71	ctx->shash.a = (b << 1) | (a >> 63);
72	ctx->shash.b = (a << 1) | (b >> 63);
73
74	if (a >> 63)
75		ctx->shash.b ^= ((u64)0xc2) << 56;
76
77	return 0;
78}
79
80static int ghash_update(struct shash_desc *desc,
81			 const u8 *src, unsigned int srclen)
82{
83	struct ghash_desc_ctx *dctx = shash_desc_ctx(desc);
84	struct ghash_ctx *ctx = crypto_shash_ctx(desc->tfm);
85	u8 *dst = dctx->buffer;
86
87	kernel_fpu_begin();
88	if (dctx->bytes) {
89		int n = min(srclen, dctx->bytes);
90		u8 *pos = dst + (GHASH_BLOCK_SIZE - dctx->bytes);
91
92		dctx->bytes -= n;
93		srclen -= n;
94
95		while (n--)
96			*pos++ ^= *src++;
97
98		if (!dctx->bytes)
99			clmul_ghash_mul(dst, &ctx->shash);
100	}
101
102	clmul_ghash_update(dst, src, srclen, &ctx->shash);
103	kernel_fpu_end();
104
105	if (srclen & 0xf) {
106		src += srclen - (srclen & 0xf);
107		srclen &= 0xf;
108		dctx->bytes = GHASH_BLOCK_SIZE - srclen;
109		while (srclen--)
110			*dst++ ^= *src++;
111	}
112
113	return 0;
114}
115
116static void ghash_flush(struct ghash_ctx *ctx, struct ghash_desc_ctx *dctx)
117{
118	u8 *dst = dctx->buffer;
119
120	if (dctx->bytes) {
121		u8 *tmp = dst + (GHASH_BLOCK_SIZE - dctx->bytes);
122
123		while (dctx->bytes--)
124			*tmp++ ^= 0;
125
126		kernel_fpu_begin();
127		clmul_ghash_mul(dst, &ctx->shash);
128		kernel_fpu_end();
129	}
130
131	dctx->bytes = 0;
132}
133
134static int ghash_final(struct shash_desc *desc, u8 *dst)
135{
136	struct ghash_desc_ctx *dctx = shash_desc_ctx(desc);
137	struct ghash_ctx *ctx = crypto_shash_ctx(desc->tfm);
138	u8 *buf = dctx->buffer;
139
140	ghash_flush(ctx, dctx);
141	memcpy(dst, buf, GHASH_BLOCK_SIZE);
142
143	return 0;
144}
145
146static struct shash_alg ghash_alg = {
147	.digestsize	= GHASH_DIGEST_SIZE,
148	.init		= ghash_init,
149	.update		= ghash_update,
150	.final		= ghash_final,
151	.setkey		= ghash_setkey,
152	.descsize	= sizeof(struct ghash_desc_ctx),
153	.base		= {
154		.cra_name		= "__ghash",
155		.cra_driver_name	= "__ghash-pclmulqdqni",
156		.cra_priority		= 0,
157		.cra_flags		= CRYPTO_ALG_TYPE_SHASH |
158					  CRYPTO_ALG_INTERNAL,
159		.cra_blocksize		= GHASH_BLOCK_SIZE,
160		.cra_ctxsize		= sizeof(struct ghash_ctx),
161		.cra_module		= THIS_MODULE,
162	},
163};
164
165static int ghash_async_init(struct ahash_request *req)
166{
167	struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
168	struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm);
169	struct ahash_request *cryptd_req = ahash_request_ctx(req);
170	struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm;
171
172	if (!irq_fpu_usable()) {
173		memcpy(cryptd_req, req, sizeof(*req));
174		ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base);
175		return crypto_ahash_init(cryptd_req);
176	} else {
177		struct shash_desc *desc = cryptd_shash_desc(cryptd_req);
178		struct crypto_shash *child = cryptd_ahash_child(cryptd_tfm);
179
180		desc->tfm = child;
181		desc->flags = req->base.flags;
182		return crypto_shash_init(desc);
183	}
184}
185
186static int ghash_async_update(struct ahash_request *req)
187{
188	struct ahash_request *cryptd_req = ahash_request_ctx(req);
189
190	if (!irq_fpu_usable()) {
191		struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
192		struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm);
193		struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm;
194
195		memcpy(cryptd_req, req, sizeof(*req));
196		ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base);
197		return crypto_ahash_update(cryptd_req);
198	} else {
199		struct shash_desc *desc = cryptd_shash_desc(cryptd_req);
200		return shash_ahash_update(req, desc);
201	}
202}
203
204static int ghash_async_final(struct ahash_request *req)
205{
206	struct ahash_request *cryptd_req = ahash_request_ctx(req);
207
208	if (!irq_fpu_usable()) {
209		struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
210		struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm);
211		struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm;
212
213		memcpy(cryptd_req, req, sizeof(*req));
214		ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base);
215		return crypto_ahash_final(cryptd_req);
216	} else {
217		struct shash_desc *desc = cryptd_shash_desc(cryptd_req);
218		return crypto_shash_final(desc, req->result);
219	}
220}
221
222static int ghash_async_digest(struct ahash_request *req)
223{
224	struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
225	struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm);
226	struct ahash_request *cryptd_req = ahash_request_ctx(req);
227	struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm;
228
229	if (!irq_fpu_usable()) {
230		memcpy(cryptd_req, req, sizeof(*req));
231		ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base);
232		return crypto_ahash_digest(cryptd_req);
233	} else {
234		struct shash_desc *desc = cryptd_shash_desc(cryptd_req);
235		struct crypto_shash *child = cryptd_ahash_child(cryptd_tfm);
236
237		desc->tfm = child;
238		desc->flags = req->base.flags;
239		return shash_ahash_digest(req, desc);
240	}
241}
242
243static int ghash_async_setkey(struct crypto_ahash *tfm, const u8 *key,
244			      unsigned int keylen)
245{
246	struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm);
247	struct crypto_ahash *child = &ctx->cryptd_tfm->base;
248	int err;
249
250	crypto_ahash_clear_flags(child, CRYPTO_TFM_REQ_MASK);
251	crypto_ahash_set_flags(child, crypto_ahash_get_flags(tfm)
252			       & CRYPTO_TFM_REQ_MASK);
253	err = crypto_ahash_setkey(child, key, keylen);
254	crypto_ahash_set_flags(tfm, crypto_ahash_get_flags(child)
255			       & CRYPTO_TFM_RES_MASK);
256
257	return err;
258}
259
260static int ghash_async_init_tfm(struct crypto_tfm *tfm)
261{
262	struct cryptd_ahash *cryptd_tfm;
263	struct ghash_async_ctx *ctx = crypto_tfm_ctx(tfm);
264
265	cryptd_tfm = cryptd_alloc_ahash("__ghash-pclmulqdqni",
266					CRYPTO_ALG_INTERNAL,
267					CRYPTO_ALG_INTERNAL);
268	if (IS_ERR(cryptd_tfm))
269		return PTR_ERR(cryptd_tfm);
270	ctx->cryptd_tfm = cryptd_tfm;
271	crypto_ahash_set_reqsize(__crypto_ahash_cast(tfm),
272				 sizeof(struct ahash_request) +
273				 crypto_ahash_reqsize(&cryptd_tfm->base));
274
275	return 0;
276}
277
278static void ghash_async_exit_tfm(struct crypto_tfm *tfm)
279{
280	struct ghash_async_ctx *ctx = crypto_tfm_ctx(tfm);
281
282	cryptd_free_ahash(ctx->cryptd_tfm);
283}
284
285static struct ahash_alg ghash_async_alg = {
286	.init		= ghash_async_init,
287	.update		= ghash_async_update,
288	.final		= ghash_async_final,
289	.setkey		= ghash_async_setkey,
290	.digest		= ghash_async_digest,
291	.halg = {
292		.digestsize	= GHASH_DIGEST_SIZE,
293		.base = {
294			.cra_name		= "ghash",
295			.cra_driver_name	= "ghash-clmulni",
296			.cra_priority		= 400,
297			.cra_ctxsize		= sizeof(struct ghash_async_ctx),
298			.cra_flags		= CRYPTO_ALG_TYPE_AHASH | CRYPTO_ALG_ASYNC,
299			.cra_blocksize		= GHASH_BLOCK_SIZE,
300			.cra_type		= &crypto_ahash_type,
301			.cra_module		= THIS_MODULE,
302			.cra_init		= ghash_async_init_tfm,
303			.cra_exit		= ghash_async_exit_tfm,
304		},
305	},
306};
307
308static const struct x86_cpu_id pcmul_cpu_id[] = {
309	X86_FEATURE_MATCH(X86_FEATURE_PCLMULQDQ), /* Pickle-Mickle-Duck */
310	{}
311};
312MODULE_DEVICE_TABLE(x86cpu, pcmul_cpu_id);
313
314static int __init ghash_pclmulqdqni_mod_init(void)
315{
316	int err;
317
318	if (!x86_match_cpu(pcmul_cpu_id))
319		return -ENODEV;
320
321	err = crypto_register_shash(&ghash_alg);
322	if (err)
323		goto err_out;
324	err = crypto_register_ahash(&ghash_async_alg);
325	if (err)
326		goto err_shash;
327
328	return 0;
329
330err_shash:
331	crypto_unregister_shash(&ghash_alg);
332err_out:
333	return err;
334}
335
336static void __exit ghash_pclmulqdqni_mod_exit(void)
337{
338	crypto_unregister_ahash(&ghash_async_alg);
339	crypto_unregister_shash(&ghash_alg);
340}
341
342module_init(ghash_pclmulqdqni_mod_init);
343module_exit(ghash_pclmulqdqni_mod_exit);
344
345MODULE_LICENSE("GPL");
346MODULE_DESCRIPTION("GHASH Message Digest Algorithm, "
347		   "acclerated by PCLMULQDQ-NI");
348MODULE_ALIAS_CRYPTO("ghash");
349