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 
28 void clmul_ghash_mul(char *dst, const be128 *shash);
29 
30 void clmul_ghash_update(char *dst, const char *src, unsigned int srclen,
31 			const be128 *shash);
32 
33 void clmul_ghash_setkey(be128 *shash, const u8 *key);
34 
35 struct ghash_async_ctx {
36 	struct cryptd_ahash *cryptd_tfm;
37 };
38 
39 struct ghash_ctx {
40 	be128 shash;
41 };
42 
43 struct ghash_desc_ctx {
44 	u8 buffer[GHASH_BLOCK_SIZE];
45 	u32 bytes;
46 };
47 
48 static int ghash_init(struct shash_desc *desc)
49 {
50 	struct ghash_desc_ctx *dctx = shash_desc_ctx(desc);
51 
52 	memset(dctx, 0, sizeof(*dctx));
53 
54 	return 0;
55 }
56 
57 static int ghash_setkey(struct crypto_shash *tfm,
58 			const u8 *key, unsigned int keylen)
59 {
60 	struct ghash_ctx *ctx = crypto_shash_ctx(tfm);
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 	clmul_ghash_setkey(&ctx->shash, key);
68 
69 	return 0;
70 }
71 
72 static int ghash_update(struct shash_desc *desc,
73 			 const u8 *src, unsigned int srclen)
74 {
75 	struct ghash_desc_ctx *dctx = shash_desc_ctx(desc);
76 	struct ghash_ctx *ctx = crypto_shash_ctx(desc->tfm);
77 	u8 *dst = dctx->buffer;
78 
79 	kernel_fpu_begin();
80 	if (dctx->bytes) {
81 		int n = min(srclen, dctx->bytes);
82 		u8 *pos = dst + (GHASH_BLOCK_SIZE - dctx->bytes);
83 
84 		dctx->bytes -= n;
85 		srclen -= n;
86 
87 		while (n--)
88 			*pos++ ^= *src++;
89 
90 		if (!dctx->bytes)
91 			clmul_ghash_mul(dst, &ctx->shash);
92 	}
93 
94 	clmul_ghash_update(dst, src, srclen, &ctx->shash);
95 	kernel_fpu_end();
96 
97 	if (srclen & 0xf) {
98 		src += srclen - (srclen & 0xf);
99 		srclen &= 0xf;
100 		dctx->bytes = GHASH_BLOCK_SIZE - srclen;
101 		while (srclen--)
102 			*dst++ ^= *src++;
103 	}
104 
105 	return 0;
106 }
107 
108 static void ghash_flush(struct ghash_ctx *ctx, struct ghash_desc_ctx *dctx)
109 {
110 	u8 *dst = dctx->buffer;
111 
112 	if (dctx->bytes) {
113 		u8 *tmp = dst + (GHASH_BLOCK_SIZE - dctx->bytes);
114 
115 		while (dctx->bytes--)
116 			*tmp++ ^= 0;
117 
118 		kernel_fpu_begin();
119 		clmul_ghash_mul(dst, &ctx->shash);
120 		kernel_fpu_end();
121 	}
122 
123 	dctx->bytes = 0;
124 }
125 
126 static int ghash_final(struct shash_desc *desc, u8 *dst)
127 {
128 	struct ghash_desc_ctx *dctx = shash_desc_ctx(desc);
129 	struct ghash_ctx *ctx = crypto_shash_ctx(desc->tfm);
130 	u8 *buf = dctx->buffer;
131 
132 	ghash_flush(ctx, dctx);
133 	memcpy(dst, buf, GHASH_BLOCK_SIZE);
134 
135 	return 0;
136 }
137 
138 static struct shash_alg ghash_alg = {
139 	.digestsize	= GHASH_DIGEST_SIZE,
140 	.init		= ghash_init,
141 	.update		= ghash_update,
142 	.final		= ghash_final,
143 	.setkey		= ghash_setkey,
144 	.descsize	= sizeof(struct ghash_desc_ctx),
145 	.base		= {
146 		.cra_name		= "__ghash",
147 		.cra_driver_name	= "__ghash-pclmulqdqni",
148 		.cra_priority		= 0,
149 		.cra_flags		= CRYPTO_ALG_TYPE_SHASH,
150 		.cra_blocksize		= GHASH_BLOCK_SIZE,
151 		.cra_ctxsize		= sizeof(struct ghash_ctx),
152 		.cra_module		= THIS_MODULE,
153 	},
154 };
155 
156 static int ghash_async_init(struct ahash_request *req)
157 {
158 	struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
159 	struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm);
160 	struct ahash_request *cryptd_req = ahash_request_ctx(req);
161 	struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm;
162 
163 	if (!irq_fpu_usable()) {
164 		memcpy(cryptd_req, req, sizeof(*req));
165 		ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base);
166 		return crypto_ahash_init(cryptd_req);
167 	} else {
168 		struct shash_desc *desc = cryptd_shash_desc(cryptd_req);
169 		struct crypto_shash *child = cryptd_ahash_child(cryptd_tfm);
170 
171 		desc->tfm = child;
172 		desc->flags = req->base.flags;
173 		return crypto_shash_init(desc);
174 	}
175 }
176 
177 static int ghash_async_update(struct ahash_request *req)
178 {
179 	struct ahash_request *cryptd_req = ahash_request_ctx(req);
180 
181 	if (!irq_fpu_usable()) {
182 		struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
183 		struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm);
184 		struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm;
185 
186 		memcpy(cryptd_req, req, sizeof(*req));
187 		ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base);
188 		return crypto_ahash_update(cryptd_req);
189 	} else {
190 		struct shash_desc *desc = cryptd_shash_desc(cryptd_req);
191 		return shash_ahash_update(req, desc);
192 	}
193 }
194 
195 static int ghash_async_final(struct ahash_request *req)
196 {
197 	struct ahash_request *cryptd_req = ahash_request_ctx(req);
198 
199 	if (!irq_fpu_usable()) {
200 		struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
201 		struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm);
202 		struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm;
203 
204 		memcpy(cryptd_req, req, sizeof(*req));
205 		ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base);
206 		return crypto_ahash_final(cryptd_req);
207 	} else {
208 		struct shash_desc *desc = cryptd_shash_desc(cryptd_req);
209 		return crypto_shash_final(desc, req->result);
210 	}
211 }
212 
213 static int ghash_async_digest(struct ahash_request *req)
214 {
215 	struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
216 	struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm);
217 	struct ahash_request *cryptd_req = ahash_request_ctx(req);
218 	struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm;
219 
220 	if (!irq_fpu_usable()) {
221 		memcpy(cryptd_req, req, sizeof(*req));
222 		ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base);
223 		return crypto_ahash_digest(cryptd_req);
224 	} else {
225 		struct shash_desc *desc = cryptd_shash_desc(cryptd_req);
226 		struct crypto_shash *child = cryptd_ahash_child(cryptd_tfm);
227 
228 		desc->tfm = child;
229 		desc->flags = req->base.flags;
230 		return shash_ahash_digest(req, desc);
231 	}
232 }
233 
234 static int ghash_async_setkey(struct crypto_ahash *tfm, const u8 *key,
235 			      unsigned int keylen)
236 {
237 	struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm);
238 	struct crypto_ahash *child = &ctx->cryptd_tfm->base;
239 	int err;
240 
241 	crypto_ahash_clear_flags(child, CRYPTO_TFM_REQ_MASK);
242 	crypto_ahash_set_flags(child, crypto_ahash_get_flags(tfm)
243 			       & CRYPTO_TFM_REQ_MASK);
244 	err = crypto_ahash_setkey(child, key, keylen);
245 	crypto_ahash_set_flags(tfm, crypto_ahash_get_flags(child)
246 			       & CRYPTO_TFM_RES_MASK);
247 
248 	return err;
249 }
250 
251 static int ghash_async_init_tfm(struct crypto_tfm *tfm)
252 {
253 	struct cryptd_ahash *cryptd_tfm;
254 	struct ghash_async_ctx *ctx = crypto_tfm_ctx(tfm);
255 
256 	cryptd_tfm = cryptd_alloc_ahash("__ghash-pclmulqdqni", 0, 0);
257 	if (IS_ERR(cryptd_tfm))
258 		return PTR_ERR(cryptd_tfm);
259 	ctx->cryptd_tfm = cryptd_tfm;
260 	crypto_ahash_set_reqsize(__crypto_ahash_cast(tfm),
261 				 sizeof(struct ahash_request) +
262 				 crypto_ahash_reqsize(&cryptd_tfm->base));
263 
264 	return 0;
265 }
266 
267 static void ghash_async_exit_tfm(struct crypto_tfm *tfm)
268 {
269 	struct ghash_async_ctx *ctx = crypto_tfm_ctx(tfm);
270 
271 	cryptd_free_ahash(ctx->cryptd_tfm);
272 }
273 
274 static struct ahash_alg ghash_async_alg = {
275 	.init		= ghash_async_init,
276 	.update		= ghash_async_update,
277 	.final		= ghash_async_final,
278 	.setkey		= ghash_async_setkey,
279 	.digest		= ghash_async_digest,
280 	.halg = {
281 		.digestsize	= GHASH_DIGEST_SIZE,
282 		.base = {
283 			.cra_name		= "ghash",
284 			.cra_driver_name	= "ghash-clmulni",
285 			.cra_priority		= 400,
286 			.cra_flags		= CRYPTO_ALG_TYPE_AHASH | CRYPTO_ALG_ASYNC,
287 			.cra_blocksize		= GHASH_BLOCK_SIZE,
288 			.cra_type		= &crypto_ahash_type,
289 			.cra_module		= THIS_MODULE,
290 			.cra_init		= ghash_async_init_tfm,
291 			.cra_exit		= ghash_async_exit_tfm,
292 		},
293 	},
294 };
295 
296 static const struct x86_cpu_id pcmul_cpu_id[] = {
297 	X86_FEATURE_MATCH(X86_FEATURE_PCLMULQDQ), /* Pickle-Mickle-Duck */
298 	{}
299 };
300 MODULE_DEVICE_TABLE(x86cpu, pcmul_cpu_id);
301 
302 static int __init ghash_pclmulqdqni_mod_init(void)
303 {
304 	int err;
305 
306 	if (!x86_match_cpu(pcmul_cpu_id))
307 		return -ENODEV;
308 
309 	err = crypto_register_shash(&ghash_alg);
310 	if (err)
311 		goto err_out;
312 	err = crypto_register_ahash(&ghash_async_alg);
313 	if (err)
314 		goto err_shash;
315 
316 	return 0;
317 
318 err_shash:
319 	crypto_unregister_shash(&ghash_alg);
320 err_out:
321 	return err;
322 }
323 
324 static void __exit ghash_pclmulqdqni_mod_exit(void)
325 {
326 	crypto_unregister_ahash(&ghash_async_alg);
327 	crypto_unregister_shash(&ghash_alg);
328 }
329 
330 module_init(ghash_pclmulqdqni_mod_init);
331 module_exit(ghash_pclmulqdqni_mod_exit);
332 
333 MODULE_LICENSE("GPL");
334 MODULE_DESCRIPTION("GHASH Message Digest Algorithm, "
335 		   "acclerated by PCLMULQDQ-NI");
336 MODULE_ALIAS("ghash");
337