xref: /openbmc/linux/drivers/md/dm-crypt.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
1 /*
2  * Copyright (C) 2003 Christophe Saout <christophe@saout.de>
3  * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
4  *
5  * This file is released under the GPL.
6  */
7 
8 #include <linux/module.h>
9 #include <linux/init.h>
10 #include <linux/kernel.h>
11 #include <linux/bio.h>
12 #include <linux/blkdev.h>
13 #include <linux/mempool.h>
14 #include <linux/slab.h>
15 #include <linux/crypto.h>
16 #include <linux/workqueue.h>
17 #include <asm/atomic.h>
18 #include <linux/scatterlist.h>
19 #include <asm/page.h>
20 
21 #include "dm.h"
22 
23 #define PFX	"crypt: "
24 
25 /*
26  * per bio private data
27  */
28 struct crypt_io {
29 	struct dm_target *target;
30 	struct bio *bio;
31 	struct bio *first_clone;
32 	struct work_struct work;
33 	atomic_t pending;
34 	int error;
35 };
36 
37 /*
38  * context holding the current state of a multi-part conversion
39  */
40 struct convert_context {
41 	struct bio *bio_in;
42 	struct bio *bio_out;
43 	unsigned int offset_in;
44 	unsigned int offset_out;
45 	unsigned int idx_in;
46 	unsigned int idx_out;
47 	sector_t sector;
48 	int write;
49 };
50 
51 struct crypt_config;
52 
53 struct crypt_iv_operations {
54 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
55 	           const char *opts);
56 	void (*dtr)(struct crypt_config *cc);
57 	const char *(*status)(struct crypt_config *cc);
58 	int (*generator)(struct crypt_config *cc, u8 *iv, sector_t sector);
59 };
60 
61 /*
62  * Crypt: maps a linear range of a block device
63  * and encrypts / decrypts at the same time.
64  */
65 struct crypt_config {
66 	struct dm_dev *dev;
67 	sector_t start;
68 
69 	/*
70 	 * pool for per bio private data and
71 	 * for encryption buffer pages
72 	 */
73 	mempool_t *io_pool;
74 	mempool_t *page_pool;
75 
76 	/*
77 	 * crypto related data
78 	 */
79 	struct crypt_iv_operations *iv_gen_ops;
80 	char *iv_mode;
81 	void *iv_gen_private;
82 	sector_t iv_offset;
83 	unsigned int iv_size;
84 
85 	struct crypto_tfm *tfm;
86 	unsigned int key_size;
87 	u8 key[0];
88 };
89 
90 #define MIN_IOS        256
91 #define MIN_POOL_PAGES 32
92 #define MIN_BIO_PAGES  8
93 
94 static kmem_cache_t *_crypt_io_pool;
95 
96 /*
97  * Mempool alloc and free functions for the page
98  */
99 static void *mempool_alloc_page(gfp_t gfp_mask, void *data)
100 {
101 	return alloc_page(gfp_mask);
102 }
103 
104 static void mempool_free_page(void *page, void *data)
105 {
106 	__free_page(page);
107 }
108 
109 
110 /*
111  * Different IV generation algorithms:
112  *
113  * plain: the initial vector is the 32-bit low-endian version of the sector
114  *        number, padded with zeros if neccessary.
115  *
116  * ess_iv: "encrypted sector|salt initial vector", the sector number is
117  *         encrypted with the bulk cipher using a salt as key. The salt
118  *         should be derived from the bulk cipher's key via hashing.
119  *
120  * plumb: unimplemented, see:
121  * http://article.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt/454
122  */
123 
124 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
125 {
126 	memset(iv, 0, cc->iv_size);
127 	*(u32 *)iv = cpu_to_le32(sector & 0xffffffff);
128 
129 	return 0;
130 }
131 
132 static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti,
133 	                      const char *opts)
134 {
135 	struct crypto_tfm *essiv_tfm;
136 	struct crypto_tfm *hash_tfm;
137 	struct scatterlist sg;
138 	unsigned int saltsize;
139 	u8 *salt;
140 
141 	if (opts == NULL) {
142 		ti->error = PFX "Digest algorithm missing for ESSIV mode";
143 		return -EINVAL;
144 	}
145 
146 	/* Hash the cipher key with the given hash algorithm */
147 	hash_tfm = crypto_alloc_tfm(opts, CRYPTO_TFM_REQ_MAY_SLEEP);
148 	if (hash_tfm == NULL) {
149 		ti->error = PFX "Error initializing ESSIV hash";
150 		return -EINVAL;
151 	}
152 
153 	if (crypto_tfm_alg_type(hash_tfm) != CRYPTO_ALG_TYPE_DIGEST) {
154 		ti->error = PFX "Expected digest algorithm for ESSIV hash";
155 		crypto_free_tfm(hash_tfm);
156 		return -EINVAL;
157 	}
158 
159 	saltsize = crypto_tfm_alg_digestsize(hash_tfm);
160 	salt = kmalloc(saltsize, GFP_KERNEL);
161 	if (salt == NULL) {
162 		ti->error = PFX "Error kmallocing salt storage in ESSIV";
163 		crypto_free_tfm(hash_tfm);
164 		return -ENOMEM;
165 	}
166 
167 	sg_set_buf(&sg, cc->key, cc->key_size);
168 	crypto_digest_digest(hash_tfm, &sg, 1, salt);
169 	crypto_free_tfm(hash_tfm);
170 
171 	/* Setup the essiv_tfm with the given salt */
172 	essiv_tfm = crypto_alloc_tfm(crypto_tfm_alg_name(cc->tfm),
173 	                             CRYPTO_TFM_MODE_ECB |
174 	                             CRYPTO_TFM_REQ_MAY_SLEEP);
175 	if (essiv_tfm == NULL) {
176 		ti->error = PFX "Error allocating crypto tfm for ESSIV";
177 		kfree(salt);
178 		return -EINVAL;
179 	}
180 	if (crypto_tfm_alg_blocksize(essiv_tfm)
181 	    != crypto_tfm_alg_ivsize(cc->tfm)) {
182 		ti->error = PFX "Block size of ESSIV cipher does "
183 			        "not match IV size of block cipher";
184 		crypto_free_tfm(essiv_tfm);
185 		kfree(salt);
186 		return -EINVAL;
187 	}
188 	if (crypto_cipher_setkey(essiv_tfm, salt, saltsize) < 0) {
189 		ti->error = PFX "Failed to set key for ESSIV cipher";
190 		crypto_free_tfm(essiv_tfm);
191 		kfree(salt);
192 		return -EINVAL;
193 	}
194 	kfree(salt);
195 
196 	cc->iv_gen_private = (void *)essiv_tfm;
197 	return 0;
198 }
199 
200 static void crypt_iv_essiv_dtr(struct crypt_config *cc)
201 {
202 	crypto_free_tfm((struct crypto_tfm *)cc->iv_gen_private);
203 	cc->iv_gen_private = NULL;
204 }
205 
206 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
207 {
208 	struct scatterlist sg;
209 
210 	memset(iv, 0, cc->iv_size);
211 	*(u64 *)iv = cpu_to_le64(sector);
212 
213 	sg_set_buf(&sg, iv, cc->iv_size);
214 	crypto_cipher_encrypt((struct crypto_tfm *)cc->iv_gen_private,
215 	                      &sg, &sg, cc->iv_size);
216 
217 	return 0;
218 }
219 
220 static struct crypt_iv_operations crypt_iv_plain_ops = {
221 	.generator = crypt_iv_plain_gen
222 };
223 
224 static struct crypt_iv_operations crypt_iv_essiv_ops = {
225 	.ctr       = crypt_iv_essiv_ctr,
226 	.dtr       = crypt_iv_essiv_dtr,
227 	.generator = crypt_iv_essiv_gen
228 };
229 
230 
231 static inline int
232 crypt_convert_scatterlist(struct crypt_config *cc, struct scatterlist *out,
233                           struct scatterlist *in, unsigned int length,
234                           int write, sector_t sector)
235 {
236 	u8 iv[cc->iv_size];
237 	int r;
238 
239 	if (cc->iv_gen_ops) {
240 		r = cc->iv_gen_ops->generator(cc, iv, sector);
241 		if (r < 0)
242 			return r;
243 
244 		if (write)
245 			r = crypto_cipher_encrypt_iv(cc->tfm, out, in, length, iv);
246 		else
247 			r = crypto_cipher_decrypt_iv(cc->tfm, out, in, length, iv);
248 	} else {
249 		if (write)
250 			r = crypto_cipher_encrypt(cc->tfm, out, in, length);
251 		else
252 			r = crypto_cipher_decrypt(cc->tfm, out, in, length);
253 	}
254 
255 	return r;
256 }
257 
258 static void
259 crypt_convert_init(struct crypt_config *cc, struct convert_context *ctx,
260                    struct bio *bio_out, struct bio *bio_in,
261                    sector_t sector, int write)
262 {
263 	ctx->bio_in = bio_in;
264 	ctx->bio_out = bio_out;
265 	ctx->offset_in = 0;
266 	ctx->offset_out = 0;
267 	ctx->idx_in = bio_in ? bio_in->bi_idx : 0;
268 	ctx->idx_out = bio_out ? bio_out->bi_idx : 0;
269 	ctx->sector = sector + cc->iv_offset;
270 	ctx->write = write;
271 }
272 
273 /*
274  * Encrypt / decrypt data from one bio to another one (can be the same one)
275  */
276 static int crypt_convert(struct crypt_config *cc,
277                          struct convert_context *ctx)
278 {
279 	int r = 0;
280 
281 	while(ctx->idx_in < ctx->bio_in->bi_vcnt &&
282 	      ctx->idx_out < ctx->bio_out->bi_vcnt) {
283 		struct bio_vec *bv_in = bio_iovec_idx(ctx->bio_in, ctx->idx_in);
284 		struct bio_vec *bv_out = bio_iovec_idx(ctx->bio_out, ctx->idx_out);
285 		struct scatterlist sg_in = {
286 			.page = bv_in->bv_page,
287 			.offset = bv_in->bv_offset + ctx->offset_in,
288 			.length = 1 << SECTOR_SHIFT
289 		};
290 		struct scatterlist sg_out = {
291 			.page = bv_out->bv_page,
292 			.offset = bv_out->bv_offset + ctx->offset_out,
293 			.length = 1 << SECTOR_SHIFT
294 		};
295 
296 		ctx->offset_in += sg_in.length;
297 		if (ctx->offset_in >= bv_in->bv_len) {
298 			ctx->offset_in = 0;
299 			ctx->idx_in++;
300 		}
301 
302 		ctx->offset_out += sg_out.length;
303 		if (ctx->offset_out >= bv_out->bv_len) {
304 			ctx->offset_out = 0;
305 			ctx->idx_out++;
306 		}
307 
308 		r = crypt_convert_scatterlist(cc, &sg_out, &sg_in, sg_in.length,
309 		                              ctx->write, ctx->sector);
310 		if (r < 0)
311 			break;
312 
313 		ctx->sector++;
314 	}
315 
316 	return r;
317 }
318 
319 /*
320  * Generate a new unfragmented bio with the given size
321  * This should never violate the device limitations
322  * May return a smaller bio when running out of pages
323  */
324 static struct bio *
325 crypt_alloc_buffer(struct crypt_config *cc, unsigned int size,
326                    struct bio *base_bio, unsigned int *bio_vec_idx)
327 {
328 	struct bio *bio;
329 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
330 	gfp_t gfp_mask = GFP_NOIO | __GFP_HIGHMEM;
331 	unsigned int i;
332 
333 	/*
334 	 * Use __GFP_NOMEMALLOC to tell the VM to act less aggressively and
335 	 * to fail earlier.  This is not necessary but increases throughput.
336 	 * FIXME: Is this really intelligent?
337 	 */
338 	if (base_bio)
339 		bio = bio_clone(base_bio, GFP_NOIO|__GFP_NOMEMALLOC);
340 	else
341 		bio = bio_alloc(GFP_NOIO|__GFP_NOMEMALLOC, nr_iovecs);
342 	if (!bio)
343 		return NULL;
344 
345 	/* if the last bio was not complete, continue where that one ended */
346 	bio->bi_idx = *bio_vec_idx;
347 	bio->bi_vcnt = *bio_vec_idx;
348 	bio->bi_size = 0;
349 	bio->bi_flags &= ~(1 << BIO_SEG_VALID);
350 
351 	/* bio->bi_idx pages have already been allocated */
352 	size -= bio->bi_idx * PAGE_SIZE;
353 
354 	for(i = bio->bi_idx; i < nr_iovecs; i++) {
355 		struct bio_vec *bv = bio_iovec_idx(bio, i);
356 
357 		bv->bv_page = mempool_alloc(cc->page_pool, gfp_mask);
358 		if (!bv->bv_page)
359 			break;
360 
361 		/*
362 		 * if additional pages cannot be allocated without waiting,
363 		 * return a partially allocated bio, the caller will then try
364 		 * to allocate additional bios while submitting this partial bio
365 		 */
366 		if ((i - bio->bi_idx) == (MIN_BIO_PAGES - 1))
367 			gfp_mask = (gfp_mask | __GFP_NOWARN) & ~__GFP_WAIT;
368 
369 		bv->bv_offset = 0;
370 		if (size > PAGE_SIZE)
371 			bv->bv_len = PAGE_SIZE;
372 		else
373 			bv->bv_len = size;
374 
375 		bio->bi_size += bv->bv_len;
376 		bio->bi_vcnt++;
377 		size -= bv->bv_len;
378 	}
379 
380 	if (!bio->bi_size) {
381 		bio_put(bio);
382 		return NULL;
383 	}
384 
385 	/*
386 	 * Remember the last bio_vec allocated to be able
387 	 * to correctly continue after the splitting.
388 	 */
389 	*bio_vec_idx = bio->bi_vcnt;
390 
391 	return bio;
392 }
393 
394 static void crypt_free_buffer_pages(struct crypt_config *cc,
395                                     struct bio *bio, unsigned int bytes)
396 {
397 	unsigned int i, start, end;
398 	struct bio_vec *bv;
399 
400 	/*
401 	 * This is ugly, but Jens Axboe thinks that using bi_idx in the
402 	 * endio function is too dangerous at the moment, so I calculate the
403 	 * correct position using bi_vcnt and bi_size.
404 	 * The bv_offset and bv_len fields might already be modified but we
405 	 * know that we always allocated whole pages.
406 	 * A fix to the bi_idx issue in the kernel is in the works, so
407 	 * we will hopefully be able to revert to the cleaner solution soon.
408 	 */
409 	i = bio->bi_vcnt - 1;
410 	bv = bio_iovec_idx(bio, i);
411 	end = (i << PAGE_SHIFT) + (bv->bv_offset + bv->bv_len) - bio->bi_size;
412 	start = end - bytes;
413 
414 	start >>= PAGE_SHIFT;
415 	if (!bio->bi_size)
416 		end = bio->bi_vcnt;
417 	else
418 		end >>= PAGE_SHIFT;
419 
420 	for(i = start; i < end; i++) {
421 		bv = bio_iovec_idx(bio, i);
422 		BUG_ON(!bv->bv_page);
423 		mempool_free(bv->bv_page, cc->page_pool);
424 		bv->bv_page = NULL;
425 	}
426 }
427 
428 /*
429  * One of the bios was finished. Check for completion of
430  * the whole request and correctly clean up the buffer.
431  */
432 static void dec_pending(struct crypt_io *io, int error)
433 {
434 	struct crypt_config *cc = (struct crypt_config *) io->target->private;
435 
436 	if (error < 0)
437 		io->error = error;
438 
439 	if (!atomic_dec_and_test(&io->pending))
440 		return;
441 
442 	if (io->first_clone)
443 		bio_put(io->first_clone);
444 
445 	bio_endio(io->bio, io->bio->bi_size, io->error);
446 
447 	mempool_free(io, cc->io_pool);
448 }
449 
450 /*
451  * kcryptd:
452  *
453  * Needed because it would be very unwise to do decryption in an
454  * interrupt context, so bios returning from read requests get
455  * queued here.
456  */
457 static struct workqueue_struct *_kcryptd_workqueue;
458 
459 static void kcryptd_do_work(void *data)
460 {
461 	struct crypt_io *io = (struct crypt_io *) data;
462 	struct crypt_config *cc = (struct crypt_config *) io->target->private;
463 	struct convert_context ctx;
464 	int r;
465 
466 	crypt_convert_init(cc, &ctx, io->bio, io->bio,
467 	                   io->bio->bi_sector - io->target->begin, 0);
468 	r = crypt_convert(cc, &ctx);
469 
470 	dec_pending(io, r);
471 }
472 
473 static void kcryptd_queue_io(struct crypt_io *io)
474 {
475 	INIT_WORK(&io->work, kcryptd_do_work, io);
476 	queue_work(_kcryptd_workqueue, &io->work);
477 }
478 
479 /*
480  * Decode key from its hex representation
481  */
482 static int crypt_decode_key(u8 *key, char *hex, unsigned int size)
483 {
484 	char buffer[3];
485 	char *endp;
486 	unsigned int i;
487 
488 	buffer[2] = '\0';
489 
490 	for(i = 0; i < size; i++) {
491 		buffer[0] = *hex++;
492 		buffer[1] = *hex++;
493 
494 		key[i] = (u8)simple_strtoul(buffer, &endp, 16);
495 
496 		if (endp != &buffer[2])
497 			return -EINVAL;
498 	}
499 
500 	if (*hex != '\0')
501 		return -EINVAL;
502 
503 	return 0;
504 }
505 
506 /*
507  * Encode key into its hex representation
508  */
509 static void crypt_encode_key(char *hex, u8 *key, unsigned int size)
510 {
511 	unsigned int i;
512 
513 	for(i = 0; i < size; i++) {
514 		sprintf(hex, "%02x", *key);
515 		hex += 2;
516 		key++;
517 	}
518 }
519 
520 /*
521  * Construct an encryption mapping:
522  * <cipher> <key> <iv_offset> <dev_path> <start>
523  */
524 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
525 {
526 	struct crypt_config *cc;
527 	struct crypto_tfm *tfm;
528 	char *tmp;
529 	char *cipher;
530 	char *chainmode;
531 	char *ivmode;
532 	char *ivopts;
533 	unsigned int crypto_flags;
534 	unsigned int key_size;
535 
536 	if (argc != 5) {
537 		ti->error = PFX "Not enough arguments";
538 		return -EINVAL;
539 	}
540 
541 	tmp = argv[0];
542 	cipher = strsep(&tmp, "-");
543 	chainmode = strsep(&tmp, "-");
544 	ivopts = strsep(&tmp, "-");
545 	ivmode = strsep(&ivopts, ":");
546 
547 	if (tmp)
548 		DMWARN(PFX "Unexpected additional cipher options");
549 
550 	key_size = strlen(argv[1]) >> 1;
551 
552 	cc = kmalloc(sizeof(*cc) + key_size * sizeof(u8), GFP_KERNEL);
553 	if (cc == NULL) {
554 		ti->error =
555 			PFX "Cannot allocate transparent encryption context";
556 		return -ENOMEM;
557 	}
558 
559 	cc->key_size = key_size;
560 	if ((!key_size && strcmp(argv[1], "-") != 0) ||
561 	    (key_size && crypt_decode_key(cc->key, argv[1], key_size) < 0)) {
562 		ti->error = PFX "Error decoding key";
563 		goto bad1;
564 	}
565 
566 	/* Compatiblity mode for old dm-crypt cipher strings */
567 	if (!chainmode || (strcmp(chainmode, "plain") == 0 && !ivmode)) {
568 		chainmode = "cbc";
569 		ivmode = "plain";
570 	}
571 
572 	/* Choose crypto_flags according to chainmode */
573 	if (strcmp(chainmode, "cbc") == 0)
574 		crypto_flags = CRYPTO_TFM_MODE_CBC;
575 	else if (strcmp(chainmode, "ecb") == 0)
576 		crypto_flags = CRYPTO_TFM_MODE_ECB;
577 	else {
578 		ti->error = PFX "Unknown chaining mode";
579 		goto bad1;
580 	}
581 
582 	if (crypto_flags != CRYPTO_TFM_MODE_ECB && !ivmode) {
583 		ti->error = PFX "This chaining mode requires an IV mechanism";
584 		goto bad1;
585 	}
586 
587 	tfm = crypto_alloc_tfm(cipher, crypto_flags | CRYPTO_TFM_REQ_MAY_SLEEP);
588 	if (!tfm) {
589 		ti->error = PFX "Error allocating crypto tfm";
590 		goto bad1;
591 	}
592 	if (crypto_tfm_alg_type(tfm) != CRYPTO_ALG_TYPE_CIPHER) {
593 		ti->error = PFX "Expected cipher algorithm";
594 		goto bad2;
595 	}
596 
597 	cc->tfm = tfm;
598 
599 	/*
600 	 * Choose ivmode. Valid modes: "plain", "essiv:<esshash>".
601 	 * See comments at iv code
602 	 */
603 
604 	if (ivmode == NULL)
605 		cc->iv_gen_ops = NULL;
606 	else if (strcmp(ivmode, "plain") == 0)
607 		cc->iv_gen_ops = &crypt_iv_plain_ops;
608 	else if (strcmp(ivmode, "essiv") == 0)
609 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
610 	else {
611 		ti->error = PFX "Invalid IV mode";
612 		goto bad2;
613 	}
614 
615 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr &&
616 	    cc->iv_gen_ops->ctr(cc, ti, ivopts) < 0)
617 		goto bad2;
618 
619 	if (tfm->crt_cipher.cit_decrypt_iv && tfm->crt_cipher.cit_encrypt_iv)
620 		/* at least a 64 bit sector number should fit in our buffer */
621 		cc->iv_size = max(crypto_tfm_alg_ivsize(tfm),
622 		                  (unsigned int)(sizeof(u64) / sizeof(u8)));
623 	else {
624 		cc->iv_size = 0;
625 		if (cc->iv_gen_ops) {
626 			DMWARN(PFX "Selected cipher does not support IVs");
627 			if (cc->iv_gen_ops->dtr)
628 				cc->iv_gen_ops->dtr(cc);
629 			cc->iv_gen_ops = NULL;
630 		}
631 	}
632 
633 	cc->io_pool = mempool_create(MIN_IOS, mempool_alloc_slab,
634 				     mempool_free_slab, _crypt_io_pool);
635 	if (!cc->io_pool) {
636 		ti->error = PFX "Cannot allocate crypt io mempool";
637 		goto bad3;
638 	}
639 
640 	cc->page_pool = mempool_create(MIN_POOL_PAGES, mempool_alloc_page,
641 				       mempool_free_page, NULL);
642 	if (!cc->page_pool) {
643 		ti->error = PFX "Cannot allocate page mempool";
644 		goto bad4;
645 	}
646 
647 	if (tfm->crt_cipher.cit_setkey(tfm, cc->key, key_size) < 0) {
648 		ti->error = PFX "Error setting key";
649 		goto bad5;
650 	}
651 
652 	if (sscanf(argv[2], SECTOR_FORMAT, &cc->iv_offset) != 1) {
653 		ti->error = PFX "Invalid iv_offset sector";
654 		goto bad5;
655 	}
656 
657 	if (sscanf(argv[4], SECTOR_FORMAT, &cc->start) != 1) {
658 		ti->error = PFX "Invalid device sector";
659 		goto bad5;
660 	}
661 
662 	if (dm_get_device(ti, argv[3], cc->start, ti->len,
663 	                  dm_table_get_mode(ti->table), &cc->dev)) {
664 		ti->error = PFX "Device lookup failed";
665 		goto bad5;
666 	}
667 
668 	if (ivmode && cc->iv_gen_ops) {
669 		if (ivopts)
670 			*(ivopts - 1) = ':';
671 		cc->iv_mode = kmalloc(strlen(ivmode) + 1, GFP_KERNEL);
672 		if (!cc->iv_mode) {
673 			ti->error = PFX "Error kmallocing iv_mode string";
674 			goto bad5;
675 		}
676 		strcpy(cc->iv_mode, ivmode);
677 	} else
678 		cc->iv_mode = NULL;
679 
680 	ti->private = cc;
681 	return 0;
682 
683 bad5:
684 	mempool_destroy(cc->page_pool);
685 bad4:
686 	mempool_destroy(cc->io_pool);
687 bad3:
688 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
689 		cc->iv_gen_ops->dtr(cc);
690 bad2:
691 	crypto_free_tfm(tfm);
692 bad1:
693 	kfree(cc);
694 	return -EINVAL;
695 }
696 
697 static void crypt_dtr(struct dm_target *ti)
698 {
699 	struct crypt_config *cc = (struct crypt_config *) ti->private;
700 
701 	mempool_destroy(cc->page_pool);
702 	mempool_destroy(cc->io_pool);
703 
704 	kfree(cc->iv_mode);
705 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
706 		cc->iv_gen_ops->dtr(cc);
707 	crypto_free_tfm(cc->tfm);
708 	dm_put_device(ti, cc->dev);
709 	kfree(cc);
710 }
711 
712 static int crypt_endio(struct bio *bio, unsigned int done, int error)
713 {
714 	struct crypt_io *io = (struct crypt_io *) bio->bi_private;
715 	struct crypt_config *cc = (struct crypt_config *) io->target->private;
716 
717 	if (bio_data_dir(bio) == WRITE) {
718 		/*
719 		 * free the processed pages, even if
720 		 * it's only a partially completed write
721 		 */
722 		crypt_free_buffer_pages(cc, bio, done);
723 	}
724 
725 	if (bio->bi_size)
726 		return 1;
727 
728 	bio_put(bio);
729 
730 	/*
731 	 * successful reads are decrypted by the worker thread
732 	 */
733 	if ((bio_data_dir(bio) == READ)
734 	    && bio_flagged(bio, BIO_UPTODATE)) {
735 		kcryptd_queue_io(io);
736 		return 0;
737 	}
738 
739 	dec_pending(io, error);
740 	return error;
741 }
742 
743 static inline struct bio *
744 crypt_clone(struct crypt_config *cc, struct crypt_io *io, struct bio *bio,
745             sector_t sector, unsigned int *bvec_idx,
746             struct convert_context *ctx)
747 {
748 	struct bio *clone;
749 
750 	if (bio_data_dir(bio) == WRITE) {
751 		clone = crypt_alloc_buffer(cc, bio->bi_size,
752                                  io->first_clone, bvec_idx);
753 		if (clone) {
754 			ctx->bio_out = clone;
755 			if (crypt_convert(cc, ctx) < 0) {
756 				crypt_free_buffer_pages(cc, clone,
757 				                        clone->bi_size);
758 				bio_put(clone);
759 				return NULL;
760 			}
761 		}
762 	} else {
763 		/*
764 		 * The block layer might modify the bvec array, so always
765 		 * copy the required bvecs because we need the original
766 		 * one in order to decrypt the whole bio data *afterwards*.
767 		 */
768 		clone = bio_alloc(GFP_NOIO, bio_segments(bio));
769 		if (clone) {
770 			clone->bi_idx = 0;
771 			clone->bi_vcnt = bio_segments(bio);
772 			clone->bi_size = bio->bi_size;
773 			memcpy(clone->bi_io_vec, bio_iovec(bio),
774 			       sizeof(struct bio_vec) * clone->bi_vcnt);
775 		}
776 	}
777 
778 	if (!clone)
779 		return NULL;
780 
781 	clone->bi_private = io;
782 	clone->bi_end_io = crypt_endio;
783 	clone->bi_bdev = cc->dev->bdev;
784 	clone->bi_sector = cc->start + sector;
785 	clone->bi_rw = bio->bi_rw;
786 
787 	return clone;
788 }
789 
790 static int crypt_map(struct dm_target *ti, struct bio *bio,
791 		     union map_info *map_context)
792 {
793 	struct crypt_config *cc = (struct crypt_config *) ti->private;
794 	struct crypt_io *io = mempool_alloc(cc->io_pool, GFP_NOIO);
795 	struct convert_context ctx;
796 	struct bio *clone;
797 	unsigned int remaining = bio->bi_size;
798 	sector_t sector = bio->bi_sector - ti->begin;
799 	unsigned int bvec_idx = 0;
800 
801 	io->target = ti;
802 	io->bio = bio;
803 	io->first_clone = NULL;
804 	io->error = 0;
805 	atomic_set(&io->pending, 1); /* hold a reference */
806 
807 	if (bio_data_dir(bio) == WRITE)
808 		crypt_convert_init(cc, &ctx, NULL, bio, sector, 1);
809 
810 	/*
811 	 * The allocated buffers can be smaller than the whole bio,
812 	 * so repeat the whole process until all the data can be handled.
813 	 */
814 	while (remaining) {
815 		clone = crypt_clone(cc, io, bio, sector, &bvec_idx, &ctx);
816 		if (!clone)
817 			goto cleanup;
818 
819 		if (!io->first_clone) {
820 			/*
821 			 * hold a reference to the first clone, because it
822 			 * holds the bio_vec array and that can't be freed
823 			 * before all other clones are released
824 			 */
825 			bio_get(clone);
826 			io->first_clone = clone;
827 		}
828 		atomic_inc(&io->pending);
829 
830 		remaining -= clone->bi_size;
831 		sector += bio_sectors(clone);
832 
833 		generic_make_request(clone);
834 
835 		/* out of memory -> run queues */
836 		if (remaining)
837 			blk_congestion_wait(bio_data_dir(clone), HZ/100);
838 	}
839 
840 	/* drop reference, clones could have returned before we reach this */
841 	dec_pending(io, 0);
842 	return 0;
843 
844 cleanup:
845 	if (io->first_clone) {
846 		dec_pending(io, -ENOMEM);
847 		return 0;
848 	}
849 
850 	/* if no bio has been dispatched yet, we can directly return the error */
851 	mempool_free(io, cc->io_pool);
852 	return -ENOMEM;
853 }
854 
855 static int crypt_status(struct dm_target *ti, status_type_t type,
856 			char *result, unsigned int maxlen)
857 {
858 	struct crypt_config *cc = (struct crypt_config *) ti->private;
859 	const char *cipher;
860 	const char *chainmode = NULL;
861 	unsigned int sz = 0;
862 
863 	switch (type) {
864 	case STATUSTYPE_INFO:
865 		result[0] = '\0';
866 		break;
867 
868 	case STATUSTYPE_TABLE:
869 		cipher = crypto_tfm_alg_name(cc->tfm);
870 
871 		switch(cc->tfm->crt_cipher.cit_mode) {
872 		case CRYPTO_TFM_MODE_CBC:
873 			chainmode = "cbc";
874 			break;
875 		case CRYPTO_TFM_MODE_ECB:
876 			chainmode = "ecb";
877 			break;
878 		default:
879 			BUG();
880 		}
881 
882 		if (cc->iv_mode)
883 			DMEMIT("%s-%s-%s ", cipher, chainmode, cc->iv_mode);
884 		else
885 			DMEMIT("%s-%s ", cipher, chainmode);
886 
887 		if (cc->key_size > 0) {
888 			if ((maxlen - sz) < ((cc->key_size << 1) + 1))
889 				return -ENOMEM;
890 
891 			crypt_encode_key(result + sz, cc->key, cc->key_size);
892 			sz += cc->key_size << 1;
893 		} else {
894 			if (sz >= maxlen)
895 				return -ENOMEM;
896 			result[sz++] = '-';
897 		}
898 
899 		DMEMIT(" " SECTOR_FORMAT " %s " SECTOR_FORMAT,
900 		       cc->iv_offset, cc->dev->name, cc->start);
901 		break;
902 	}
903 	return 0;
904 }
905 
906 static struct target_type crypt_target = {
907 	.name   = "crypt",
908 	.version= {1, 1, 0},
909 	.module = THIS_MODULE,
910 	.ctr    = crypt_ctr,
911 	.dtr    = crypt_dtr,
912 	.map    = crypt_map,
913 	.status = crypt_status,
914 };
915 
916 static int __init dm_crypt_init(void)
917 {
918 	int r;
919 
920 	_crypt_io_pool = kmem_cache_create("dm-crypt_io",
921 	                                   sizeof(struct crypt_io),
922 	                                   0, 0, NULL, NULL);
923 	if (!_crypt_io_pool)
924 		return -ENOMEM;
925 
926 	_kcryptd_workqueue = create_workqueue("kcryptd");
927 	if (!_kcryptd_workqueue) {
928 		r = -ENOMEM;
929 		DMERR(PFX "couldn't create kcryptd");
930 		goto bad1;
931 	}
932 
933 	r = dm_register_target(&crypt_target);
934 	if (r < 0) {
935 		DMERR(PFX "register failed %d", r);
936 		goto bad2;
937 	}
938 
939 	return 0;
940 
941 bad2:
942 	destroy_workqueue(_kcryptd_workqueue);
943 bad1:
944 	kmem_cache_destroy(_crypt_io_pool);
945 	return r;
946 }
947 
948 static void __exit dm_crypt_exit(void)
949 {
950 	int r = dm_unregister_target(&crypt_target);
951 
952 	if (r < 0)
953 		DMERR(PFX "unregister failed %d", r);
954 
955 	destroy_workqueue(_kcryptd_workqueue);
956 	kmem_cache_destroy(_crypt_io_pool);
957 }
958 
959 module_init(dm_crypt_init);
960 module_exit(dm_crypt_exit);
961 
962 MODULE_AUTHOR("Christophe Saout <christophe@saout.de>");
963 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
964 MODULE_LICENSE("GPL");
965