xref: /linux/drivers/md/dm-crypt.c (revision ed3174d93c342b8b2eeba6bbd124707d55304a7b)
1 /*
2  * Copyright (C) 2003 Christophe Saout <christophe@saout.de>
3  * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
4  * Copyright (C) 2006-2007 Red Hat, Inc. All rights reserved.
5  *
6  * This file is released under the GPL.
7  */
8 
9 #include <linux/completion.h>
10 #include <linux/err.h>
11 #include <linux/module.h>
12 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/bio.h>
15 #include <linux/blkdev.h>
16 #include <linux/mempool.h>
17 #include <linux/slab.h>
18 #include <linux/crypto.h>
19 #include <linux/workqueue.h>
20 #include <linux/backing-dev.h>
21 #include <asm/atomic.h>
22 #include <linux/scatterlist.h>
23 #include <asm/page.h>
24 #include <asm/unaligned.h>
25 
26 #include "dm.h"
27 
28 #define DM_MSG_PREFIX "crypt"
29 #define MESG_STR(x) x, sizeof(x)
30 
31 /*
32  * context holding the current state of a multi-part conversion
33  */
34 struct convert_context {
35 	struct completion restart;
36 	struct bio *bio_in;
37 	struct bio *bio_out;
38 	unsigned int offset_in;
39 	unsigned int offset_out;
40 	unsigned int idx_in;
41 	unsigned int idx_out;
42 	sector_t sector;
43 	atomic_t pending;
44 };
45 
46 /*
47  * per bio private data
48  */
49 struct dm_crypt_io {
50 	struct dm_target *target;
51 	struct bio *base_bio;
52 	struct work_struct work;
53 
54 	struct convert_context ctx;
55 
56 	atomic_t pending;
57 	int error;
58 	sector_t sector;
59 };
60 
61 struct dm_crypt_request {
62 	struct scatterlist sg_in;
63 	struct scatterlist sg_out;
64 };
65 
66 struct crypt_config;
67 
68 struct crypt_iv_operations {
69 	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
70 		   const char *opts);
71 	void (*dtr)(struct crypt_config *cc);
72 	const char *(*status)(struct crypt_config *cc);
73 	int (*generator)(struct crypt_config *cc, u8 *iv, sector_t sector);
74 };
75 
76 /*
77  * Crypt: maps a linear range of a block device
78  * and encrypts / decrypts at the same time.
79  */
80 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID };
81 struct crypt_config {
82 	struct dm_dev *dev;
83 	sector_t start;
84 
85 	/*
86 	 * pool for per bio private data, crypto requests and
87 	 * encryption requeusts/buffer pages
88 	 */
89 	mempool_t *io_pool;
90 	mempool_t *req_pool;
91 	mempool_t *page_pool;
92 	struct bio_set *bs;
93 
94 	struct workqueue_struct *io_queue;
95 	struct workqueue_struct *crypt_queue;
96 	/*
97 	 * crypto related data
98 	 */
99 	struct crypt_iv_operations *iv_gen_ops;
100 	char *iv_mode;
101 	union {
102 		struct crypto_cipher *essiv_tfm;
103 		int benbi_shift;
104 	} iv_gen_private;
105 	sector_t iv_offset;
106 	unsigned int iv_size;
107 
108 	/*
109 	 * Layout of each crypto request:
110 	 *
111 	 *   struct ablkcipher_request
112 	 *      context
113 	 *      padding
114 	 *   struct dm_crypt_request
115 	 *      padding
116 	 *   IV
117 	 *
118 	 * The padding is added so that dm_crypt_request and the IV are
119 	 * correctly aligned.
120 	 */
121 	unsigned int dmreq_start;
122 	struct ablkcipher_request *req;
123 
124 	char cipher[CRYPTO_MAX_ALG_NAME];
125 	char chainmode[CRYPTO_MAX_ALG_NAME];
126 	struct crypto_ablkcipher *tfm;
127 	unsigned long flags;
128 	unsigned int key_size;
129 	u8 key[0];
130 };
131 
132 #define MIN_IOS        16
133 #define MIN_POOL_PAGES 32
134 #define MIN_BIO_PAGES  8
135 
136 static struct kmem_cache *_crypt_io_pool;
137 
138 static void clone_init(struct dm_crypt_io *, struct bio *);
139 static void kcryptd_queue_crypt(struct dm_crypt_io *io);
140 
141 /*
142  * Different IV generation algorithms:
143  *
144  * plain: the initial vector is the 32-bit little-endian version of the sector
145  *        number, padded with zeros if necessary.
146  *
147  * essiv: "encrypted sector|salt initial vector", the sector number is
148  *        encrypted with the bulk cipher using a salt as key. The salt
149  *        should be derived from the bulk cipher's key via hashing.
150  *
151  * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
152  *        (needed for LRW-32-AES and possible other narrow block modes)
153  *
154  * null: the initial vector is always zero.  Provides compatibility with
155  *       obsolete loop_fish2 devices.  Do not use for new devices.
156  *
157  * plumb: unimplemented, see:
158  * http://article.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt/454
159  */
160 
161 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
162 {
163 	memset(iv, 0, cc->iv_size);
164 	*(u32 *)iv = cpu_to_le32(sector & 0xffffffff);
165 
166 	return 0;
167 }
168 
169 static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti,
170 			      const char *opts)
171 {
172 	struct crypto_cipher *essiv_tfm;
173 	struct crypto_hash *hash_tfm;
174 	struct hash_desc desc;
175 	struct scatterlist sg;
176 	unsigned int saltsize;
177 	u8 *salt;
178 	int err;
179 
180 	if (opts == NULL) {
181 		ti->error = "Digest algorithm missing for ESSIV mode";
182 		return -EINVAL;
183 	}
184 
185 	/* Hash the cipher key with the given hash algorithm */
186 	hash_tfm = crypto_alloc_hash(opts, 0, CRYPTO_ALG_ASYNC);
187 	if (IS_ERR(hash_tfm)) {
188 		ti->error = "Error initializing ESSIV hash";
189 		return PTR_ERR(hash_tfm);
190 	}
191 
192 	saltsize = crypto_hash_digestsize(hash_tfm);
193 	salt = kmalloc(saltsize, GFP_KERNEL);
194 	if (salt == NULL) {
195 		ti->error = "Error kmallocing salt storage in ESSIV";
196 		crypto_free_hash(hash_tfm);
197 		return -ENOMEM;
198 	}
199 
200 	sg_init_one(&sg, cc->key, cc->key_size);
201 	desc.tfm = hash_tfm;
202 	desc.flags = CRYPTO_TFM_REQ_MAY_SLEEP;
203 	err = crypto_hash_digest(&desc, &sg, cc->key_size, salt);
204 	crypto_free_hash(hash_tfm);
205 
206 	if (err) {
207 		ti->error = "Error calculating hash in ESSIV";
208 		kfree(salt);
209 		return err;
210 	}
211 
212 	/* Setup the essiv_tfm with the given salt */
213 	essiv_tfm = crypto_alloc_cipher(cc->cipher, 0, CRYPTO_ALG_ASYNC);
214 	if (IS_ERR(essiv_tfm)) {
215 		ti->error = "Error allocating crypto tfm for ESSIV";
216 		kfree(salt);
217 		return PTR_ERR(essiv_tfm);
218 	}
219 	if (crypto_cipher_blocksize(essiv_tfm) !=
220 	    crypto_ablkcipher_ivsize(cc->tfm)) {
221 		ti->error = "Block size of ESSIV cipher does "
222 			    "not match IV size of block cipher";
223 		crypto_free_cipher(essiv_tfm);
224 		kfree(salt);
225 		return -EINVAL;
226 	}
227 	err = crypto_cipher_setkey(essiv_tfm, salt, saltsize);
228 	if (err) {
229 		ti->error = "Failed to set key for ESSIV cipher";
230 		crypto_free_cipher(essiv_tfm);
231 		kfree(salt);
232 		return err;
233 	}
234 	kfree(salt);
235 
236 	cc->iv_gen_private.essiv_tfm = essiv_tfm;
237 	return 0;
238 }
239 
240 static void crypt_iv_essiv_dtr(struct crypt_config *cc)
241 {
242 	crypto_free_cipher(cc->iv_gen_private.essiv_tfm);
243 	cc->iv_gen_private.essiv_tfm = NULL;
244 }
245 
246 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
247 {
248 	memset(iv, 0, cc->iv_size);
249 	*(u64 *)iv = cpu_to_le64(sector);
250 	crypto_cipher_encrypt_one(cc->iv_gen_private.essiv_tfm, iv, iv);
251 	return 0;
252 }
253 
254 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
255 			      const char *opts)
256 {
257 	unsigned bs = crypto_ablkcipher_blocksize(cc->tfm);
258 	int log = ilog2(bs);
259 
260 	/* we need to calculate how far we must shift the sector count
261 	 * to get the cipher block count, we use this shift in _gen */
262 
263 	if (1 << log != bs) {
264 		ti->error = "cypher blocksize is not a power of 2";
265 		return -EINVAL;
266 	}
267 
268 	if (log > 9) {
269 		ti->error = "cypher blocksize is > 512";
270 		return -EINVAL;
271 	}
272 
273 	cc->iv_gen_private.benbi_shift = 9 - log;
274 
275 	return 0;
276 }
277 
278 static void crypt_iv_benbi_dtr(struct crypt_config *cc)
279 {
280 }
281 
282 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
283 {
284 	__be64 val;
285 
286 	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
287 
288 	val = cpu_to_be64(((u64)sector << cc->iv_gen_private.benbi_shift) + 1);
289 	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
290 
291 	return 0;
292 }
293 
294 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
295 {
296 	memset(iv, 0, cc->iv_size);
297 
298 	return 0;
299 }
300 
301 static struct crypt_iv_operations crypt_iv_plain_ops = {
302 	.generator = crypt_iv_plain_gen
303 };
304 
305 static struct crypt_iv_operations crypt_iv_essiv_ops = {
306 	.ctr       = crypt_iv_essiv_ctr,
307 	.dtr       = crypt_iv_essiv_dtr,
308 	.generator = crypt_iv_essiv_gen
309 };
310 
311 static struct crypt_iv_operations crypt_iv_benbi_ops = {
312 	.ctr	   = crypt_iv_benbi_ctr,
313 	.dtr	   = crypt_iv_benbi_dtr,
314 	.generator = crypt_iv_benbi_gen
315 };
316 
317 static struct crypt_iv_operations crypt_iv_null_ops = {
318 	.generator = crypt_iv_null_gen
319 };
320 
321 static void crypt_convert_init(struct crypt_config *cc,
322 			       struct convert_context *ctx,
323 			       struct bio *bio_out, struct bio *bio_in,
324 			       sector_t sector)
325 {
326 	ctx->bio_in = bio_in;
327 	ctx->bio_out = bio_out;
328 	ctx->offset_in = 0;
329 	ctx->offset_out = 0;
330 	ctx->idx_in = bio_in ? bio_in->bi_idx : 0;
331 	ctx->idx_out = bio_out ? bio_out->bi_idx : 0;
332 	ctx->sector = sector + cc->iv_offset;
333 	init_completion(&ctx->restart);
334 	/*
335 	 * Crypto operation can be asynchronous,
336 	 * ctx->pending is increased after request submission.
337 	 * We need to ensure that we don't call the crypt finish
338 	 * operation before pending got incremented
339 	 * (dependent on crypt submission return code).
340 	 */
341 	atomic_set(&ctx->pending, 2);
342 }
343 
344 static int crypt_convert_block(struct crypt_config *cc,
345 			       struct convert_context *ctx,
346 			       struct ablkcipher_request *req)
347 {
348 	struct bio_vec *bv_in = bio_iovec_idx(ctx->bio_in, ctx->idx_in);
349 	struct bio_vec *bv_out = bio_iovec_idx(ctx->bio_out, ctx->idx_out);
350 	struct dm_crypt_request *dmreq;
351 	u8 *iv;
352 	int r = 0;
353 
354 	dmreq = (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
355 	iv = (u8 *)ALIGN((unsigned long)(dmreq + 1),
356 			 crypto_ablkcipher_alignmask(cc->tfm) + 1);
357 
358 	sg_init_table(&dmreq->sg_in, 1);
359 	sg_set_page(&dmreq->sg_in, bv_in->bv_page, 1 << SECTOR_SHIFT,
360 		    bv_in->bv_offset + ctx->offset_in);
361 
362 	sg_init_table(&dmreq->sg_out, 1);
363 	sg_set_page(&dmreq->sg_out, bv_out->bv_page, 1 << SECTOR_SHIFT,
364 		    bv_out->bv_offset + ctx->offset_out);
365 
366 	ctx->offset_in += 1 << SECTOR_SHIFT;
367 	if (ctx->offset_in >= bv_in->bv_len) {
368 		ctx->offset_in = 0;
369 		ctx->idx_in++;
370 	}
371 
372 	ctx->offset_out += 1 << SECTOR_SHIFT;
373 	if (ctx->offset_out >= bv_out->bv_len) {
374 		ctx->offset_out = 0;
375 		ctx->idx_out++;
376 	}
377 
378 	if (cc->iv_gen_ops) {
379 		r = cc->iv_gen_ops->generator(cc, iv, ctx->sector);
380 		if (r < 0)
381 			return r;
382 	}
383 
384 	ablkcipher_request_set_crypt(req, &dmreq->sg_in, &dmreq->sg_out,
385 				     1 << SECTOR_SHIFT, iv);
386 
387 	if (bio_data_dir(ctx->bio_in) == WRITE)
388 		r = crypto_ablkcipher_encrypt(req);
389 	else
390 		r = crypto_ablkcipher_decrypt(req);
391 
392 	return r;
393 }
394 
395 static void kcryptd_async_done(struct crypto_async_request *async_req,
396 			       int error);
397 static void crypt_alloc_req(struct crypt_config *cc,
398 			    struct convert_context *ctx)
399 {
400 	if (!cc->req)
401 		cc->req = mempool_alloc(cc->req_pool, GFP_NOIO);
402 	ablkcipher_request_set_tfm(cc->req, cc->tfm);
403 	ablkcipher_request_set_callback(cc->req, CRYPTO_TFM_REQ_MAY_BACKLOG |
404 					     CRYPTO_TFM_REQ_MAY_SLEEP,
405 					     kcryptd_async_done, ctx);
406 }
407 
408 /*
409  * Encrypt / decrypt data from one bio to another one (can be the same one)
410  */
411 static int crypt_convert(struct crypt_config *cc,
412 			 struct convert_context *ctx)
413 {
414 	int r = 0;
415 
416 	while(ctx->idx_in < ctx->bio_in->bi_vcnt &&
417 	      ctx->idx_out < ctx->bio_out->bi_vcnt) {
418 
419 		crypt_alloc_req(cc, ctx);
420 
421 		r = crypt_convert_block(cc, ctx, cc->req);
422 
423 		switch (r) {
424 		case -EBUSY:
425 			wait_for_completion(&ctx->restart);
426 			INIT_COMPLETION(ctx->restart);
427 			/* fall through*/
428 		case -EINPROGRESS:
429 			atomic_inc(&ctx->pending);
430 			cc->req = NULL;
431 			r = 0;
432 			/* fall through*/
433 		case 0:
434 			ctx->sector++;
435 			continue;
436 		}
437 
438 		break;
439 	}
440 
441 	/*
442 	 * If there are pending crypto operation run async
443 	 * code. Otherwise process return code synchronously.
444 	 * The step of 2 ensures that async finish doesn't
445 	 * call crypto finish too early.
446 	 */
447 	if (atomic_sub_return(2, &ctx->pending))
448 		return -EINPROGRESS;
449 
450 	return r;
451 }
452 
453 static void dm_crypt_bio_destructor(struct bio *bio)
454 {
455 	struct dm_crypt_io *io = bio->bi_private;
456 	struct crypt_config *cc = io->target->private;
457 
458 	bio_free(bio, cc->bs);
459 }
460 
461 /*
462  * Generate a new unfragmented bio with the given size
463  * This should never violate the device limitations
464  * May return a smaller bio when running out of pages
465  */
466 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size)
467 {
468 	struct crypt_config *cc = io->target->private;
469 	struct bio *clone;
470 	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
471 	gfp_t gfp_mask = GFP_NOIO | __GFP_HIGHMEM;
472 	unsigned i, len;
473 	struct page *page;
474 
475 	clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, cc->bs);
476 	if (!clone)
477 		return NULL;
478 
479 	clone_init(io, clone);
480 
481 	for (i = 0; i < nr_iovecs; i++) {
482 		page = mempool_alloc(cc->page_pool, gfp_mask);
483 		if (!page)
484 			break;
485 
486 		/*
487 		 * if additional pages cannot be allocated without waiting,
488 		 * return a partially allocated bio, the caller will then try
489 		 * to allocate additional bios while submitting this partial bio
490 		 */
491 		if (i == (MIN_BIO_PAGES - 1))
492 			gfp_mask = (gfp_mask | __GFP_NOWARN) & ~__GFP_WAIT;
493 
494 		len = (size > PAGE_SIZE) ? PAGE_SIZE : size;
495 
496 		if (!bio_add_page(clone, page, len, 0)) {
497 			mempool_free(page, cc->page_pool);
498 			break;
499 		}
500 
501 		size -= len;
502 	}
503 
504 	if (!clone->bi_size) {
505 		bio_put(clone);
506 		return NULL;
507 	}
508 
509 	return clone;
510 }
511 
512 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
513 {
514 	unsigned int i;
515 	struct bio_vec *bv;
516 
517 	for (i = 0; i < clone->bi_vcnt; i++) {
518 		bv = bio_iovec_idx(clone, i);
519 		BUG_ON(!bv->bv_page);
520 		mempool_free(bv->bv_page, cc->page_pool);
521 		bv->bv_page = NULL;
522 	}
523 }
524 
525 /*
526  * One of the bios was finished. Check for completion of
527  * the whole request and correctly clean up the buffer.
528  */
529 static void crypt_dec_pending(struct dm_crypt_io *io)
530 {
531 	struct crypt_config *cc = io->target->private;
532 
533 	if (!atomic_dec_and_test(&io->pending))
534 		return;
535 
536 	bio_endio(io->base_bio, io->error);
537 	mempool_free(io, cc->io_pool);
538 }
539 
540 /*
541  * kcryptd/kcryptd_io:
542  *
543  * Needed because it would be very unwise to do decryption in an
544  * interrupt context.
545  *
546  * kcryptd performs the actual encryption or decryption.
547  *
548  * kcryptd_io performs the IO submission.
549  *
550  * They must be separated as otherwise the final stages could be
551  * starved by new requests which can block in the first stages due
552  * to memory allocation.
553  */
554 static void crypt_endio(struct bio *clone, int error)
555 {
556 	struct dm_crypt_io *io = clone->bi_private;
557 	struct crypt_config *cc = io->target->private;
558 	unsigned rw = bio_data_dir(clone);
559 
560 	if (unlikely(!bio_flagged(clone, BIO_UPTODATE) && !error))
561 		error = -EIO;
562 
563 	/*
564 	 * free the processed pages
565 	 */
566 	if (rw == WRITE)
567 		crypt_free_buffer_pages(cc, clone);
568 
569 	bio_put(clone);
570 
571 	if (rw == READ && !error) {
572 		kcryptd_queue_crypt(io);
573 		return;
574 	}
575 
576 	if (unlikely(error))
577 		io->error = error;
578 
579 	crypt_dec_pending(io);
580 }
581 
582 static void clone_init(struct dm_crypt_io *io, struct bio *clone)
583 {
584 	struct crypt_config *cc = io->target->private;
585 
586 	clone->bi_private = io;
587 	clone->bi_end_io  = crypt_endio;
588 	clone->bi_bdev    = cc->dev->bdev;
589 	clone->bi_rw      = io->base_bio->bi_rw;
590 	clone->bi_destructor = dm_crypt_bio_destructor;
591 }
592 
593 static void kcryptd_io_read(struct dm_crypt_io *io)
594 {
595 	struct crypt_config *cc = io->target->private;
596 	struct bio *base_bio = io->base_bio;
597 	struct bio *clone;
598 
599 	atomic_inc(&io->pending);
600 
601 	/*
602 	 * The block layer might modify the bvec array, so always
603 	 * copy the required bvecs because we need the original
604 	 * one in order to decrypt the whole bio data *afterwards*.
605 	 */
606 	clone = bio_alloc_bioset(GFP_NOIO, bio_segments(base_bio), cc->bs);
607 	if (unlikely(!clone)) {
608 		io->error = -ENOMEM;
609 		crypt_dec_pending(io);
610 		return;
611 	}
612 
613 	clone_init(io, clone);
614 	clone->bi_idx = 0;
615 	clone->bi_vcnt = bio_segments(base_bio);
616 	clone->bi_size = base_bio->bi_size;
617 	clone->bi_sector = cc->start + io->sector;
618 	memcpy(clone->bi_io_vec, bio_iovec(base_bio),
619 	       sizeof(struct bio_vec) * clone->bi_vcnt);
620 
621 	generic_make_request(clone);
622 }
623 
624 static void kcryptd_io_write(struct dm_crypt_io *io)
625 {
626 	struct bio *clone = io->ctx.bio_out;
627 
628 	generic_make_request(clone);
629 }
630 
631 static void kcryptd_io(struct work_struct *work)
632 {
633 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
634 
635 	if (bio_data_dir(io->base_bio) == READ)
636 		kcryptd_io_read(io);
637 	else
638 		kcryptd_io_write(io);
639 }
640 
641 static void kcryptd_queue_io(struct dm_crypt_io *io)
642 {
643 	struct crypt_config *cc = io->target->private;
644 
645 	INIT_WORK(&io->work, kcryptd_io);
646 	queue_work(cc->io_queue, &io->work);
647 }
648 
649 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io,
650 					  int error, int async)
651 {
652 	struct bio *clone = io->ctx.bio_out;
653 	struct crypt_config *cc = io->target->private;
654 
655 	if (unlikely(error < 0)) {
656 		crypt_free_buffer_pages(cc, clone);
657 		bio_put(clone);
658 		io->error = -EIO;
659 		return;
660 	}
661 
662 	/* crypt_convert should have filled the clone bio */
663 	BUG_ON(io->ctx.idx_out < clone->bi_vcnt);
664 
665 	clone->bi_sector = cc->start + io->sector;
666 	io->sector += bio_sectors(clone);
667 
668 	if (async)
669 		kcryptd_queue_io(io);
670 	else {
671 		atomic_inc(&io->pending);
672 		generic_make_request(clone);
673 	}
674 }
675 
676 static void kcryptd_crypt_write_convert_loop(struct dm_crypt_io *io)
677 {
678 	struct crypt_config *cc = io->target->private;
679 	struct bio *clone;
680 	unsigned remaining = io->base_bio->bi_size;
681 	int r;
682 
683 	/*
684 	 * The allocated buffers can be smaller than the whole bio,
685 	 * so repeat the whole process until all the data can be handled.
686 	 */
687 	while (remaining) {
688 		clone = crypt_alloc_buffer(io, remaining);
689 		if (unlikely(!clone)) {
690 			io->error = -ENOMEM;
691 			return;
692 		}
693 
694 		io->ctx.bio_out = clone;
695 		io->ctx.idx_out = 0;
696 
697 		remaining -= clone->bi_size;
698 
699 		r = crypt_convert(cc, &io->ctx);
700 
701 		if (r != -EINPROGRESS) {
702 			kcryptd_crypt_write_io_submit(io, r, 0);
703 			if (unlikely(r < 0))
704 				return;
705 		} else
706 			atomic_inc(&io->pending);
707 
708 		/* out of memory -> run queues */
709 		if (unlikely(remaining))
710 			congestion_wait(WRITE, HZ/100);
711 	}
712 }
713 
714 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
715 {
716 	struct crypt_config *cc = io->target->private;
717 
718 	/*
719 	 * Prevent io from disappearing until this function completes.
720 	 */
721 	atomic_inc(&io->pending);
722 
723 	crypt_convert_init(cc, &io->ctx, NULL, io->base_bio, io->sector);
724 	kcryptd_crypt_write_convert_loop(io);
725 
726 	crypt_dec_pending(io);
727 }
728 
729 static void kcryptd_crypt_read_done(struct dm_crypt_io *io, int error)
730 {
731 	if (unlikely(error < 0))
732 		io->error = -EIO;
733 
734 	crypt_dec_pending(io);
735 }
736 
737 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
738 {
739 	struct crypt_config *cc = io->target->private;
740 	int r = 0;
741 
742 	atomic_inc(&io->pending);
743 
744 	crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
745 			   io->sector);
746 
747 	r = crypt_convert(cc, &io->ctx);
748 
749 	if (r != -EINPROGRESS)
750 		kcryptd_crypt_read_done(io, r);
751 
752 	crypt_dec_pending(io);
753 }
754 
755 static void kcryptd_async_done(struct crypto_async_request *async_req,
756 			       int error)
757 {
758 	struct convert_context *ctx = async_req->data;
759 	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
760 	struct crypt_config *cc = io->target->private;
761 
762 	if (error == -EINPROGRESS) {
763 		complete(&ctx->restart);
764 		return;
765 	}
766 
767 	mempool_free(ablkcipher_request_cast(async_req), cc->req_pool);
768 
769 	if (!atomic_dec_and_test(&ctx->pending))
770 		return;
771 
772 	if (bio_data_dir(io->base_bio) == READ)
773 		kcryptd_crypt_read_done(io, error);
774 	else
775 		kcryptd_crypt_write_io_submit(io, error, 1);
776 }
777 
778 static void kcryptd_crypt(struct work_struct *work)
779 {
780 	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
781 
782 	if (bio_data_dir(io->base_bio) == READ)
783 		kcryptd_crypt_read_convert(io);
784 	else
785 		kcryptd_crypt_write_convert(io);
786 }
787 
788 static void kcryptd_queue_crypt(struct dm_crypt_io *io)
789 {
790 	struct crypt_config *cc = io->target->private;
791 
792 	INIT_WORK(&io->work, kcryptd_crypt);
793 	queue_work(cc->crypt_queue, &io->work);
794 }
795 
796 /*
797  * Decode key from its hex representation
798  */
799 static int crypt_decode_key(u8 *key, char *hex, unsigned int size)
800 {
801 	char buffer[3];
802 	char *endp;
803 	unsigned int i;
804 
805 	buffer[2] = '\0';
806 
807 	for (i = 0; i < size; i++) {
808 		buffer[0] = *hex++;
809 		buffer[1] = *hex++;
810 
811 		key[i] = (u8)simple_strtoul(buffer, &endp, 16);
812 
813 		if (endp != &buffer[2])
814 			return -EINVAL;
815 	}
816 
817 	if (*hex != '\0')
818 		return -EINVAL;
819 
820 	return 0;
821 }
822 
823 /*
824  * Encode key into its hex representation
825  */
826 static void crypt_encode_key(char *hex, u8 *key, unsigned int size)
827 {
828 	unsigned int i;
829 
830 	for (i = 0; i < size; i++) {
831 		sprintf(hex, "%02x", *key);
832 		hex += 2;
833 		key++;
834 	}
835 }
836 
837 static int crypt_set_key(struct crypt_config *cc, char *key)
838 {
839 	unsigned key_size = strlen(key) >> 1;
840 
841 	if (cc->key_size && cc->key_size != key_size)
842 		return -EINVAL;
843 
844 	cc->key_size = key_size; /* initial settings */
845 
846 	if ((!key_size && strcmp(key, "-")) ||
847 	   (key_size && crypt_decode_key(cc->key, key, key_size) < 0))
848 		return -EINVAL;
849 
850 	set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
851 
852 	return 0;
853 }
854 
855 static int crypt_wipe_key(struct crypt_config *cc)
856 {
857 	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
858 	memset(&cc->key, 0, cc->key_size * sizeof(u8));
859 	return 0;
860 }
861 
862 /*
863  * Construct an encryption mapping:
864  * <cipher> <key> <iv_offset> <dev_path> <start>
865  */
866 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
867 {
868 	struct crypt_config *cc;
869 	struct crypto_ablkcipher *tfm;
870 	char *tmp;
871 	char *cipher;
872 	char *chainmode;
873 	char *ivmode;
874 	char *ivopts;
875 	unsigned int key_size;
876 	unsigned long long tmpll;
877 
878 	if (argc != 5) {
879 		ti->error = "Not enough arguments";
880 		return -EINVAL;
881 	}
882 
883 	tmp = argv[0];
884 	cipher = strsep(&tmp, "-");
885 	chainmode = strsep(&tmp, "-");
886 	ivopts = strsep(&tmp, "-");
887 	ivmode = strsep(&ivopts, ":");
888 
889 	if (tmp)
890 		DMWARN("Unexpected additional cipher options");
891 
892 	key_size = strlen(argv[1]) >> 1;
893 
894  	cc = kzalloc(sizeof(*cc) + key_size * sizeof(u8), GFP_KERNEL);
895 	if (cc == NULL) {
896 		ti->error =
897 			"Cannot allocate transparent encryption context";
898 		return -ENOMEM;
899 	}
900 
901  	if (crypt_set_key(cc, argv[1])) {
902 		ti->error = "Error decoding key";
903 		goto bad_cipher;
904 	}
905 
906 	/* Compatiblity mode for old dm-crypt cipher strings */
907 	if (!chainmode || (strcmp(chainmode, "plain") == 0 && !ivmode)) {
908 		chainmode = "cbc";
909 		ivmode = "plain";
910 	}
911 
912 	if (strcmp(chainmode, "ecb") && !ivmode) {
913 		ti->error = "This chaining mode requires an IV mechanism";
914 		goto bad_cipher;
915 	}
916 
917 	if (snprintf(cc->cipher, CRYPTO_MAX_ALG_NAME, "%s(%s)",
918 		     chainmode, cipher) >= CRYPTO_MAX_ALG_NAME) {
919 		ti->error = "Chain mode + cipher name is too long";
920 		goto bad_cipher;
921 	}
922 
923 	tfm = crypto_alloc_ablkcipher(cc->cipher, 0, 0);
924 	if (IS_ERR(tfm)) {
925 		ti->error = "Error allocating crypto tfm";
926 		goto bad_cipher;
927 	}
928 
929 	strcpy(cc->cipher, cipher);
930 	strcpy(cc->chainmode, chainmode);
931 	cc->tfm = tfm;
932 
933 	/*
934 	 * Choose ivmode. Valid modes: "plain", "essiv:<esshash>", "benbi".
935 	 * See comments at iv code
936 	 */
937 
938 	if (ivmode == NULL)
939 		cc->iv_gen_ops = NULL;
940 	else if (strcmp(ivmode, "plain") == 0)
941 		cc->iv_gen_ops = &crypt_iv_plain_ops;
942 	else if (strcmp(ivmode, "essiv") == 0)
943 		cc->iv_gen_ops = &crypt_iv_essiv_ops;
944 	else if (strcmp(ivmode, "benbi") == 0)
945 		cc->iv_gen_ops = &crypt_iv_benbi_ops;
946 	else if (strcmp(ivmode, "null") == 0)
947 		cc->iv_gen_ops = &crypt_iv_null_ops;
948 	else {
949 		ti->error = "Invalid IV mode";
950 		goto bad_ivmode;
951 	}
952 
953 	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr &&
954 	    cc->iv_gen_ops->ctr(cc, ti, ivopts) < 0)
955 		goto bad_ivmode;
956 
957 	cc->iv_size = crypto_ablkcipher_ivsize(tfm);
958 	if (cc->iv_size)
959 		/* at least a 64 bit sector number should fit in our buffer */
960 		cc->iv_size = max(cc->iv_size,
961 				  (unsigned int)(sizeof(u64) / sizeof(u8)));
962 	else {
963 		if (cc->iv_gen_ops) {
964 			DMWARN("Selected cipher does not support IVs");
965 			if (cc->iv_gen_ops->dtr)
966 				cc->iv_gen_ops->dtr(cc);
967 			cc->iv_gen_ops = NULL;
968 		}
969 	}
970 
971 	cc->io_pool = mempool_create_slab_pool(MIN_IOS, _crypt_io_pool);
972 	if (!cc->io_pool) {
973 		ti->error = "Cannot allocate crypt io mempool";
974 		goto bad_slab_pool;
975 	}
976 
977 	cc->dmreq_start = sizeof(struct ablkcipher_request);
978 	cc->dmreq_start += crypto_ablkcipher_reqsize(tfm);
979 	cc->dmreq_start = ALIGN(cc->dmreq_start, crypto_tfm_ctx_alignment());
980 	cc->dmreq_start += crypto_ablkcipher_alignmask(tfm) &
981 			   ~(crypto_tfm_ctx_alignment() - 1);
982 
983 	cc->req_pool = mempool_create_kmalloc_pool(MIN_IOS, cc->dmreq_start +
984 			sizeof(struct dm_crypt_request) + cc->iv_size);
985 	if (!cc->req_pool) {
986 		ti->error = "Cannot allocate crypt request mempool";
987 		goto bad_req_pool;
988 	}
989 	cc->req = NULL;
990 
991 	cc->page_pool = mempool_create_page_pool(MIN_POOL_PAGES, 0);
992 	if (!cc->page_pool) {
993 		ti->error = "Cannot allocate page mempool";
994 		goto bad_page_pool;
995 	}
996 
997 	cc->bs = bioset_create(MIN_IOS, MIN_IOS);
998 	if (!cc->bs) {
999 		ti->error = "Cannot allocate crypt bioset";
1000 		goto bad_bs;
1001 	}
1002 
1003 	if (crypto_ablkcipher_setkey(tfm, cc->key, key_size) < 0) {
1004 		ti->error = "Error setting key";
1005 		goto bad_device;
1006 	}
1007 
1008 	if (sscanf(argv[2], "%llu", &tmpll) != 1) {
1009 		ti->error = "Invalid iv_offset sector";
1010 		goto bad_device;
1011 	}
1012 	cc->iv_offset = tmpll;
1013 
1014 	if (sscanf(argv[4], "%llu", &tmpll) != 1) {
1015 		ti->error = "Invalid device sector";
1016 		goto bad_device;
1017 	}
1018 	cc->start = tmpll;
1019 
1020 	if (dm_get_device(ti, argv[3], cc->start, ti->len,
1021 			  dm_table_get_mode(ti->table), &cc->dev)) {
1022 		ti->error = "Device lookup failed";
1023 		goto bad_device;
1024 	}
1025 
1026 	if (ivmode && cc->iv_gen_ops) {
1027 		if (ivopts)
1028 			*(ivopts - 1) = ':';
1029 		cc->iv_mode = kmalloc(strlen(ivmode) + 1, GFP_KERNEL);
1030 		if (!cc->iv_mode) {
1031 			ti->error = "Error kmallocing iv_mode string";
1032 			goto bad_ivmode_string;
1033 		}
1034 		strcpy(cc->iv_mode, ivmode);
1035 	} else
1036 		cc->iv_mode = NULL;
1037 
1038 	cc->io_queue = create_singlethread_workqueue("kcryptd_io");
1039 	if (!cc->io_queue) {
1040 		ti->error = "Couldn't create kcryptd io queue";
1041 		goto bad_io_queue;
1042 	}
1043 
1044 	cc->crypt_queue = create_singlethread_workqueue("kcryptd");
1045 	if (!cc->crypt_queue) {
1046 		ti->error = "Couldn't create kcryptd queue";
1047 		goto bad_crypt_queue;
1048 	}
1049 
1050 	ti->private = cc;
1051 	return 0;
1052 
1053 bad_crypt_queue:
1054 	destroy_workqueue(cc->io_queue);
1055 bad_io_queue:
1056 	kfree(cc->iv_mode);
1057 bad_ivmode_string:
1058 	dm_put_device(ti, cc->dev);
1059 bad_device:
1060 	bioset_free(cc->bs);
1061 bad_bs:
1062 	mempool_destroy(cc->page_pool);
1063 bad_page_pool:
1064 	mempool_destroy(cc->req_pool);
1065 bad_req_pool:
1066 	mempool_destroy(cc->io_pool);
1067 bad_slab_pool:
1068 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
1069 		cc->iv_gen_ops->dtr(cc);
1070 bad_ivmode:
1071 	crypto_free_ablkcipher(tfm);
1072 bad_cipher:
1073 	/* Must zero key material before freeing */
1074 	memset(cc, 0, sizeof(*cc) + cc->key_size * sizeof(u8));
1075 	kfree(cc);
1076 	return -EINVAL;
1077 }
1078 
1079 static void crypt_dtr(struct dm_target *ti)
1080 {
1081 	struct crypt_config *cc = (struct crypt_config *) ti->private;
1082 
1083 	destroy_workqueue(cc->io_queue);
1084 	destroy_workqueue(cc->crypt_queue);
1085 
1086 	if (cc->req)
1087 		mempool_free(cc->req, cc->req_pool);
1088 
1089 	bioset_free(cc->bs);
1090 	mempool_destroy(cc->page_pool);
1091 	mempool_destroy(cc->req_pool);
1092 	mempool_destroy(cc->io_pool);
1093 
1094 	kfree(cc->iv_mode);
1095 	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
1096 		cc->iv_gen_ops->dtr(cc);
1097 	crypto_free_ablkcipher(cc->tfm);
1098 	dm_put_device(ti, cc->dev);
1099 
1100 	/* Must zero key material before freeing */
1101 	memset(cc, 0, sizeof(*cc) + cc->key_size * sizeof(u8));
1102 	kfree(cc);
1103 }
1104 
1105 static int crypt_map(struct dm_target *ti, struct bio *bio,
1106 		     union map_info *map_context)
1107 {
1108 	struct crypt_config *cc = ti->private;
1109 	struct dm_crypt_io *io;
1110 
1111 	io = mempool_alloc(cc->io_pool, GFP_NOIO);
1112 	io->target = ti;
1113 	io->base_bio = bio;
1114 	io->sector = bio->bi_sector - ti->begin;
1115 	io->error = 0;
1116 	atomic_set(&io->pending, 0);
1117 
1118 	if (bio_data_dir(io->base_bio) == READ)
1119 		kcryptd_queue_io(io);
1120 	else
1121 		kcryptd_queue_crypt(io);
1122 
1123 	return DM_MAPIO_SUBMITTED;
1124 }
1125 
1126 static int crypt_status(struct dm_target *ti, status_type_t type,
1127 			char *result, unsigned int maxlen)
1128 {
1129 	struct crypt_config *cc = (struct crypt_config *) ti->private;
1130 	unsigned int sz = 0;
1131 
1132 	switch (type) {
1133 	case STATUSTYPE_INFO:
1134 		result[0] = '\0';
1135 		break;
1136 
1137 	case STATUSTYPE_TABLE:
1138 		if (cc->iv_mode)
1139 			DMEMIT("%s-%s-%s ", cc->cipher, cc->chainmode,
1140 			       cc->iv_mode);
1141 		else
1142 			DMEMIT("%s-%s ", cc->cipher, cc->chainmode);
1143 
1144 		if (cc->key_size > 0) {
1145 			if ((maxlen - sz) < ((cc->key_size << 1) + 1))
1146 				return -ENOMEM;
1147 
1148 			crypt_encode_key(result + sz, cc->key, cc->key_size);
1149 			sz += cc->key_size << 1;
1150 		} else {
1151 			if (sz >= maxlen)
1152 				return -ENOMEM;
1153 			result[sz++] = '-';
1154 		}
1155 
1156 		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
1157 				cc->dev->name, (unsigned long long)cc->start);
1158 		break;
1159 	}
1160 	return 0;
1161 }
1162 
1163 static void crypt_postsuspend(struct dm_target *ti)
1164 {
1165 	struct crypt_config *cc = ti->private;
1166 
1167 	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
1168 }
1169 
1170 static int crypt_preresume(struct dm_target *ti)
1171 {
1172 	struct crypt_config *cc = ti->private;
1173 
1174 	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
1175 		DMERR("aborting resume - crypt key is not set.");
1176 		return -EAGAIN;
1177 	}
1178 
1179 	return 0;
1180 }
1181 
1182 static void crypt_resume(struct dm_target *ti)
1183 {
1184 	struct crypt_config *cc = ti->private;
1185 
1186 	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
1187 }
1188 
1189 /* Message interface
1190  *	key set <key>
1191  *	key wipe
1192  */
1193 static int crypt_message(struct dm_target *ti, unsigned argc, char **argv)
1194 {
1195 	struct crypt_config *cc = ti->private;
1196 
1197 	if (argc < 2)
1198 		goto error;
1199 
1200 	if (!strnicmp(argv[0], MESG_STR("key"))) {
1201 		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
1202 			DMWARN("not suspended during key manipulation.");
1203 			return -EINVAL;
1204 		}
1205 		if (argc == 3 && !strnicmp(argv[1], MESG_STR("set")))
1206 			return crypt_set_key(cc, argv[2]);
1207 		if (argc == 2 && !strnicmp(argv[1], MESG_STR("wipe")))
1208 			return crypt_wipe_key(cc);
1209 	}
1210 
1211 error:
1212 	DMWARN("unrecognised message received.");
1213 	return -EINVAL;
1214 }
1215 
1216 static struct target_type crypt_target = {
1217 	.name   = "crypt",
1218 	.version= {1, 5, 0},
1219 	.module = THIS_MODULE,
1220 	.ctr    = crypt_ctr,
1221 	.dtr    = crypt_dtr,
1222 	.map    = crypt_map,
1223 	.status = crypt_status,
1224 	.postsuspend = crypt_postsuspend,
1225 	.preresume = crypt_preresume,
1226 	.resume = crypt_resume,
1227 	.message = crypt_message,
1228 };
1229 
1230 static int __init dm_crypt_init(void)
1231 {
1232 	int r;
1233 
1234 	_crypt_io_pool = KMEM_CACHE(dm_crypt_io, 0);
1235 	if (!_crypt_io_pool)
1236 		return -ENOMEM;
1237 
1238 	r = dm_register_target(&crypt_target);
1239 	if (r < 0) {
1240 		DMERR("register failed %d", r);
1241 		kmem_cache_destroy(_crypt_io_pool);
1242 	}
1243 
1244 	return r;
1245 }
1246 
1247 static void __exit dm_crypt_exit(void)
1248 {
1249 	int r = dm_unregister_target(&crypt_target);
1250 
1251 	if (r < 0)
1252 		DMERR("unregister failed %d", r);
1253 
1254 	kmem_cache_destroy(_crypt_io_pool);
1255 }
1256 
1257 module_init(dm_crypt_init);
1258 module_exit(dm_crypt_exit);
1259 
1260 MODULE_AUTHOR("Christophe Saout <christophe@saout.de>");
1261 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
1262 MODULE_LICENSE("GPL");
1263