Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 2003 Jana Saout <jana@saout.de>
4 * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
5 * Copyright (C) 2006-2020 Red Hat, Inc. All rights reserved.
6 * Copyright (C) 2013-2020 Milan Broz <gmazyland@gmail.com>
7 *
8 * This file is released under the GPL.
9 */
10
11#include <linux/completion.h>
12#include <linux/err.h>
13#include <linux/module.h>
14#include <linux/init.h>
15#include <linux/kernel.h>
16#include <linux/key.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/blk-integrity.h>
20#include <linux/mempool.h>
21#include <linux/slab.h>
22#include <linux/crypto.h>
23#include <linux/workqueue.h>
24#include <linux/kthread.h>
25#include <linux/backing-dev.h>
26#include <linux/atomic.h>
27#include <linux/scatterlist.h>
28#include <linux/rbtree.h>
29#include <linux/ctype.h>
30#include <asm/page.h>
31#include <asm/unaligned.h>
32#include <crypto/hash.h>
33#include <crypto/md5.h>
34#include <crypto/skcipher.h>
35#include <crypto/aead.h>
36#include <crypto/authenc.h>
37#include <crypto/utils.h>
38#include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
39#include <linux/key-type.h>
40#include <keys/user-type.h>
41#include <keys/encrypted-type.h>
42#include <keys/trusted-type.h>
43
44#include <linux/device-mapper.h>
45
46#include "dm-audit.h"
47
48#define DM_MSG_PREFIX "crypt"
49
50/*
51 * context holding the current state of a multi-part conversion
52 */
53struct convert_context {
54 struct completion restart;
55 struct bio *bio_in;
56 struct bvec_iter iter_in;
57 struct bio *bio_out;
58 struct bvec_iter iter_out;
59 atomic_t cc_pending;
60 u64 cc_sector;
61 union {
62 struct skcipher_request *req;
63 struct aead_request *req_aead;
64 } r;
65 bool aead_recheck;
66 bool aead_failed;
67
68};
69
70/*
71 * per bio private data
72 */
73struct dm_crypt_io {
74 struct crypt_config *cc;
75 struct bio *base_bio;
76 u8 *integrity_metadata;
77 bool integrity_metadata_from_pool:1;
78
79 struct work_struct work;
80
81 struct convert_context ctx;
82
83 atomic_t io_pending;
84 blk_status_t error;
85 sector_t sector;
86
87 struct bvec_iter saved_bi_iter;
88
89 struct rb_node rb_node;
90} CRYPTO_MINALIGN_ATTR;
91
92struct dm_crypt_request {
93 struct convert_context *ctx;
94 struct scatterlist sg_in[4];
95 struct scatterlist sg_out[4];
96 u64 iv_sector;
97};
98
99struct crypt_config;
100
101struct crypt_iv_operations {
102 int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
103 const char *opts);
104 void (*dtr)(struct crypt_config *cc);
105 int (*init)(struct crypt_config *cc);
106 int (*wipe)(struct crypt_config *cc);
107 int (*generator)(struct crypt_config *cc, u8 *iv,
108 struct dm_crypt_request *dmreq);
109 int (*post)(struct crypt_config *cc, u8 *iv,
110 struct dm_crypt_request *dmreq);
111};
112
113struct iv_benbi_private {
114 int shift;
115};
116
117#define LMK_SEED_SIZE 64 /* hash + 0 */
118struct iv_lmk_private {
119 struct crypto_shash *hash_tfm;
120 u8 *seed;
121};
122
123#define TCW_WHITENING_SIZE 16
124struct iv_tcw_private {
125 struct crypto_shash *crc32_tfm;
126 u8 *iv_seed;
127 u8 *whitening;
128};
129
130#define ELEPHANT_MAX_KEY_SIZE 32
131struct iv_elephant_private {
132 struct crypto_skcipher *tfm;
133};
134
135/*
136 * Crypt: maps a linear range of a block device
137 * and encrypts / decrypts at the same time.
138 */
139enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
140 DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
141 DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE,
142 DM_CRYPT_WRITE_INLINE };
143
144enum cipher_flags {
145 CRYPT_MODE_INTEGRITY_AEAD, /* Use authenticated mode for cipher */
146 CRYPT_IV_LARGE_SECTORS, /* Calculate IV from sector_size, not 512B sectors */
147 CRYPT_ENCRYPT_PREPROCESS, /* Must preprocess data for encryption (elephant) */
148};
149
150/*
151 * The fields in here must be read only after initialization.
152 */
153struct crypt_config {
154 struct dm_dev *dev;
155 sector_t start;
156
157 struct percpu_counter n_allocated_pages;
158
159 struct workqueue_struct *io_queue;
160 struct workqueue_struct *crypt_queue;
161
162 spinlock_t write_thread_lock;
163 struct task_struct *write_thread;
164 struct rb_root write_tree;
165
166 char *cipher_string;
167 char *cipher_auth;
168 char *key_string;
169
170 const struct crypt_iv_operations *iv_gen_ops;
171 union {
172 struct iv_benbi_private benbi;
173 struct iv_lmk_private lmk;
174 struct iv_tcw_private tcw;
175 struct iv_elephant_private elephant;
176 } iv_gen_private;
177 u64 iv_offset;
178 unsigned int iv_size;
179 unsigned short sector_size;
180 unsigned char sector_shift;
181
182 union {
183 struct crypto_skcipher **tfms;
184 struct crypto_aead **tfms_aead;
185 } cipher_tfm;
186 unsigned int tfms_count;
187 unsigned long cipher_flags;
188
189 /*
190 * Layout of each crypto request:
191 *
192 * struct skcipher_request
193 * context
194 * padding
195 * struct dm_crypt_request
196 * padding
197 * IV
198 *
199 * The padding is added so that dm_crypt_request and the IV are
200 * correctly aligned.
201 */
202 unsigned int dmreq_start;
203
204 unsigned int per_bio_data_size;
205
206 unsigned long flags;
207 unsigned int key_size;
208 unsigned int key_parts; /* independent parts in key buffer */
209 unsigned int key_extra_size; /* additional keys length */
210 unsigned int key_mac_size; /* MAC key size for authenc(...) */
211
212 unsigned int integrity_tag_size;
213 unsigned int integrity_iv_size;
214 unsigned int on_disk_tag_size;
215
216 /*
217 * pool for per bio private data, crypto requests,
218 * encryption requeusts/buffer pages and integrity tags
219 */
220 unsigned int tag_pool_max_sectors;
221 mempool_t tag_pool;
222 mempool_t req_pool;
223 mempool_t page_pool;
224
225 struct bio_set bs;
226 struct mutex bio_alloc_lock;
227
228 u8 *authenc_key; /* space for keys in authenc() format (if used) */
229 u8 key[] __counted_by(key_size);
230};
231
232#define MIN_IOS 64
233#define MAX_TAG_SIZE 480
234#define POOL_ENTRY_SIZE 512
235
236static DEFINE_SPINLOCK(dm_crypt_clients_lock);
237static unsigned int dm_crypt_clients_n;
238static volatile unsigned long dm_crypt_pages_per_client;
239#define DM_CRYPT_MEMORY_PERCENT 2
240#define DM_CRYPT_MIN_PAGES_PER_CLIENT (BIO_MAX_VECS * 16)
241
242static void crypt_endio(struct bio *clone);
243static void kcryptd_queue_crypt(struct dm_crypt_io *io);
244static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
245 struct scatterlist *sg);
246
247static bool crypt_integrity_aead(struct crypt_config *cc);
248
249/*
250 * Use this to access cipher attributes that are independent of the key.
251 */
252static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
253{
254 return cc->cipher_tfm.tfms[0];
255}
256
257static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
258{
259 return cc->cipher_tfm.tfms_aead[0];
260}
261
262/*
263 * Different IV generation algorithms:
264 *
265 * plain: the initial vector is the 32-bit little-endian version of the sector
266 * number, padded with zeros if necessary.
267 *
268 * plain64: the initial vector is the 64-bit little-endian version of the sector
269 * number, padded with zeros if necessary.
270 *
271 * plain64be: the initial vector is the 64-bit big-endian version of the sector
272 * number, padded with zeros if necessary.
273 *
274 * essiv: "encrypted sector|salt initial vector", the sector number is
275 * encrypted with the bulk cipher using a salt as key. The salt
276 * should be derived from the bulk cipher's key via hashing.
277 *
278 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
279 * (needed for LRW-32-AES and possible other narrow block modes)
280 *
281 * null: the initial vector is always zero. Provides compatibility with
282 * obsolete loop_fish2 devices. Do not use for new devices.
283 *
284 * lmk: Compatible implementation of the block chaining mode used
285 * by the Loop-AES block device encryption system
286 * designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
287 * It operates on full 512 byte sectors and uses CBC
288 * with an IV derived from the sector number, the data and
289 * optionally extra IV seed.
290 * This means that after decryption the first block
291 * of sector must be tweaked according to decrypted data.
292 * Loop-AES can use three encryption schemes:
293 * version 1: is plain aes-cbc mode
294 * version 2: uses 64 multikey scheme with lmk IV generator
295 * version 3: the same as version 2 with additional IV seed
296 * (it uses 65 keys, last key is used as IV seed)
297 *
298 * tcw: Compatible implementation of the block chaining mode used
299 * by the TrueCrypt device encryption system (prior to version 4.1).
300 * For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
301 * It operates on full 512 byte sectors and uses CBC
302 * with an IV derived from initial key and the sector number.
303 * In addition, whitening value is applied on every sector, whitening
304 * is calculated from initial key, sector number and mixed using CRC32.
305 * Note that this encryption scheme is vulnerable to watermarking attacks
306 * and should be used for old compatible containers access only.
307 *
308 * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
309 * The IV is encrypted little-endian byte-offset (with the same key
310 * and cipher as the volume).
311 *
312 * elephant: The extended version of eboiv with additional Elephant diffuser
313 * used with Bitlocker CBC mode.
314 * This mode was used in older Windows systems
315 * https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
316 */
317
318static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
319 struct dm_crypt_request *dmreq)
320{
321 memset(iv, 0, cc->iv_size);
322 *(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
323
324 return 0;
325}
326
327static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
328 struct dm_crypt_request *dmreq)
329{
330 memset(iv, 0, cc->iv_size);
331 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
332
333 return 0;
334}
335
336static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
337 struct dm_crypt_request *dmreq)
338{
339 memset(iv, 0, cc->iv_size);
340 /* iv_size is at least of size u64; usually it is 16 bytes */
341 *(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
342
343 return 0;
344}
345
346static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
347 struct dm_crypt_request *dmreq)
348{
349 /*
350 * ESSIV encryption of the IV is now handled by the crypto API,
351 * so just pass the plain sector number here.
352 */
353 memset(iv, 0, cc->iv_size);
354 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
355
356 return 0;
357}
358
359static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
360 const char *opts)
361{
362 unsigned int bs;
363 int log;
364
365 if (crypt_integrity_aead(cc))
366 bs = crypto_aead_blocksize(any_tfm_aead(cc));
367 else
368 bs = crypto_skcipher_blocksize(any_tfm(cc));
369 log = ilog2(bs);
370
371 /*
372 * We need to calculate how far we must shift the sector count
373 * to get the cipher block count, we use this shift in _gen.
374 */
375 if (1 << log != bs) {
376 ti->error = "cypher blocksize is not a power of 2";
377 return -EINVAL;
378 }
379
380 if (log > 9) {
381 ti->error = "cypher blocksize is > 512";
382 return -EINVAL;
383 }
384
385 cc->iv_gen_private.benbi.shift = 9 - log;
386
387 return 0;
388}
389
390static void crypt_iv_benbi_dtr(struct crypt_config *cc)
391{
392}
393
394static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
395 struct dm_crypt_request *dmreq)
396{
397 __be64 val;
398
399 memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
400
401 val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
402 put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
403
404 return 0;
405}
406
407static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
408 struct dm_crypt_request *dmreq)
409{
410 memset(iv, 0, cc->iv_size);
411
412 return 0;
413}
414
415static void crypt_iv_lmk_dtr(struct crypt_config *cc)
416{
417 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
418
419 if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
420 crypto_free_shash(lmk->hash_tfm);
421 lmk->hash_tfm = NULL;
422
423 kfree_sensitive(lmk->seed);
424 lmk->seed = NULL;
425}
426
427static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
428 const char *opts)
429{
430 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
431
432 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
433 ti->error = "Unsupported sector size for LMK";
434 return -EINVAL;
435 }
436
437 lmk->hash_tfm = crypto_alloc_shash("md5", 0,
438 CRYPTO_ALG_ALLOCATES_MEMORY);
439 if (IS_ERR(lmk->hash_tfm)) {
440 ti->error = "Error initializing LMK hash";
441 return PTR_ERR(lmk->hash_tfm);
442 }
443
444 /* No seed in LMK version 2 */
445 if (cc->key_parts == cc->tfms_count) {
446 lmk->seed = NULL;
447 return 0;
448 }
449
450 lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
451 if (!lmk->seed) {
452 crypt_iv_lmk_dtr(cc);
453 ti->error = "Error kmallocing seed storage in LMK";
454 return -ENOMEM;
455 }
456
457 return 0;
458}
459
460static int crypt_iv_lmk_init(struct crypt_config *cc)
461{
462 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
463 int subkey_size = cc->key_size / cc->key_parts;
464
465 /* LMK seed is on the position of LMK_KEYS + 1 key */
466 if (lmk->seed)
467 memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
468 crypto_shash_digestsize(lmk->hash_tfm));
469
470 return 0;
471}
472
473static int crypt_iv_lmk_wipe(struct crypt_config *cc)
474{
475 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
476
477 if (lmk->seed)
478 memset(lmk->seed, 0, LMK_SEED_SIZE);
479
480 return 0;
481}
482
483static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
484 struct dm_crypt_request *dmreq,
485 u8 *data)
486{
487 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
488 SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
489 struct md5_state md5state;
490 __le32 buf[4];
491 int i, r;
492
493 desc->tfm = lmk->hash_tfm;
494
495 r = crypto_shash_init(desc);
496 if (r)
497 return r;
498
499 if (lmk->seed) {
500 r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
501 if (r)
502 return r;
503 }
504
505 /* Sector is always 512B, block size 16, add data of blocks 1-31 */
506 r = crypto_shash_update(desc, data + 16, 16 * 31);
507 if (r)
508 return r;
509
510 /* Sector is cropped to 56 bits here */
511 buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
512 buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
513 buf[2] = cpu_to_le32(4024);
514 buf[3] = 0;
515 r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
516 if (r)
517 return r;
518
519 /* No MD5 padding here */
520 r = crypto_shash_export(desc, &md5state);
521 if (r)
522 return r;
523
524 for (i = 0; i < MD5_HASH_WORDS; i++)
525 __cpu_to_le32s(&md5state.hash[i]);
526 memcpy(iv, &md5state.hash, cc->iv_size);
527
528 return 0;
529}
530
531static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
532 struct dm_crypt_request *dmreq)
533{
534 struct scatterlist *sg;
535 u8 *src;
536 int r = 0;
537
538 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
539 sg = crypt_get_sg_data(cc, dmreq->sg_in);
540 src = kmap_local_page(sg_page(sg));
541 r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
542 kunmap_local(src);
543 } else
544 memset(iv, 0, cc->iv_size);
545
546 return r;
547}
548
549static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
550 struct dm_crypt_request *dmreq)
551{
552 struct scatterlist *sg;
553 u8 *dst;
554 int r;
555
556 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
557 return 0;
558
559 sg = crypt_get_sg_data(cc, dmreq->sg_out);
560 dst = kmap_local_page(sg_page(sg));
561 r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
562
563 /* Tweak the first block of plaintext sector */
564 if (!r)
565 crypto_xor(dst + sg->offset, iv, cc->iv_size);
566
567 kunmap_local(dst);
568 return r;
569}
570
571static void crypt_iv_tcw_dtr(struct crypt_config *cc)
572{
573 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
574
575 kfree_sensitive(tcw->iv_seed);
576 tcw->iv_seed = NULL;
577 kfree_sensitive(tcw->whitening);
578 tcw->whitening = NULL;
579
580 if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
581 crypto_free_shash(tcw->crc32_tfm);
582 tcw->crc32_tfm = NULL;
583}
584
585static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
586 const char *opts)
587{
588 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
589
590 if (cc->sector_size != (1 << SECTOR_SHIFT)) {
591 ti->error = "Unsupported sector size for TCW";
592 return -EINVAL;
593 }
594
595 if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
596 ti->error = "Wrong key size for TCW";
597 return -EINVAL;
598 }
599
600 tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
601 CRYPTO_ALG_ALLOCATES_MEMORY);
602 if (IS_ERR(tcw->crc32_tfm)) {
603 ti->error = "Error initializing CRC32 in TCW";
604 return PTR_ERR(tcw->crc32_tfm);
605 }
606
607 tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
608 tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
609 if (!tcw->iv_seed || !tcw->whitening) {
610 crypt_iv_tcw_dtr(cc);
611 ti->error = "Error allocating seed storage in TCW";
612 return -ENOMEM;
613 }
614
615 return 0;
616}
617
618static int crypt_iv_tcw_init(struct crypt_config *cc)
619{
620 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
621 int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
622
623 memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
624 memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
625 TCW_WHITENING_SIZE);
626
627 return 0;
628}
629
630static int crypt_iv_tcw_wipe(struct crypt_config *cc)
631{
632 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
633
634 memset(tcw->iv_seed, 0, cc->iv_size);
635 memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
636
637 return 0;
638}
639
640static int crypt_iv_tcw_whitening(struct crypt_config *cc,
641 struct dm_crypt_request *dmreq,
642 u8 *data)
643{
644 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
645 __le64 sector = cpu_to_le64(dmreq->iv_sector);
646 u8 buf[TCW_WHITENING_SIZE];
647 SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
648 int i, r;
649
650 /* xor whitening with sector number */
651 crypto_xor_cpy(buf, tcw->whitening, (u8 *)§or, 8);
652 crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)§or, 8);
653
654 /* calculate crc32 for every 32bit part and xor it */
655 desc->tfm = tcw->crc32_tfm;
656 for (i = 0; i < 4; i++) {
657 r = crypto_shash_digest(desc, &buf[i * 4], 4, &buf[i * 4]);
658 if (r)
659 goto out;
660 }
661 crypto_xor(&buf[0], &buf[12], 4);
662 crypto_xor(&buf[4], &buf[8], 4);
663
664 /* apply whitening (8 bytes) to whole sector */
665 for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
666 crypto_xor(data + i * 8, buf, 8);
667out:
668 memzero_explicit(buf, sizeof(buf));
669 return r;
670}
671
672static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
673 struct dm_crypt_request *dmreq)
674{
675 struct scatterlist *sg;
676 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
677 __le64 sector = cpu_to_le64(dmreq->iv_sector);
678 u8 *src;
679 int r = 0;
680
681 /* Remove whitening from ciphertext */
682 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
683 sg = crypt_get_sg_data(cc, dmreq->sg_in);
684 src = kmap_local_page(sg_page(sg));
685 r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
686 kunmap_local(src);
687 }
688
689 /* Calculate IV */
690 crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)§or, 8);
691 if (cc->iv_size > 8)
692 crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)§or,
693 cc->iv_size - 8);
694
695 return r;
696}
697
698static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
699 struct dm_crypt_request *dmreq)
700{
701 struct scatterlist *sg;
702 u8 *dst;
703 int r;
704
705 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
706 return 0;
707
708 /* Apply whitening on ciphertext */
709 sg = crypt_get_sg_data(cc, dmreq->sg_out);
710 dst = kmap_local_page(sg_page(sg));
711 r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
712 kunmap_local(dst);
713
714 return r;
715}
716
717static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
718 struct dm_crypt_request *dmreq)
719{
720 /* Used only for writes, there must be an additional space to store IV */
721 get_random_bytes(iv, cc->iv_size);
722 return 0;
723}
724
725static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
726 const char *opts)
727{
728 if (crypt_integrity_aead(cc)) {
729 ti->error = "AEAD transforms not supported for EBOIV";
730 return -EINVAL;
731 }
732
733 if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
734 ti->error = "Block size of EBOIV cipher does not match IV size of block cipher";
735 return -EINVAL;
736 }
737
738 return 0;
739}
740
741static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
742 struct dm_crypt_request *dmreq)
743{
744 struct crypto_skcipher *tfm = any_tfm(cc);
745 struct skcipher_request *req;
746 struct scatterlist src, dst;
747 DECLARE_CRYPTO_WAIT(wait);
748 unsigned int reqsize;
749 int err;
750 u8 *buf;
751
752 reqsize = sizeof(*req) + crypto_skcipher_reqsize(tfm);
753 reqsize = ALIGN(reqsize, __alignof__(__le64));
754
755 req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
756 if (!req)
757 return -ENOMEM;
758
759 skcipher_request_set_tfm(req, tfm);
760
761 buf = (u8 *)req + reqsize;
762 memset(buf, 0, cc->iv_size);
763 *(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
764
765 sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
766 sg_init_one(&dst, iv, cc->iv_size);
767 skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
768 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
769 err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
770 kfree_sensitive(req);
771
772 return err;
773}
774
775static void crypt_iv_elephant_dtr(struct crypt_config *cc)
776{
777 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
778
779 crypto_free_skcipher(elephant->tfm);
780 elephant->tfm = NULL;
781}
782
783static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
784 const char *opts)
785{
786 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
787 int r;
788
789 elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
790 CRYPTO_ALG_ALLOCATES_MEMORY);
791 if (IS_ERR(elephant->tfm)) {
792 r = PTR_ERR(elephant->tfm);
793 elephant->tfm = NULL;
794 return r;
795 }
796
797 r = crypt_iv_eboiv_ctr(cc, ti, NULL);
798 if (r)
799 crypt_iv_elephant_dtr(cc);
800 return r;
801}
802
803static void diffuser_disk_to_cpu(u32 *d, size_t n)
804{
805#ifndef __LITTLE_ENDIAN
806 int i;
807
808 for (i = 0; i < n; i++)
809 d[i] = le32_to_cpu((__le32)d[i]);
810#endif
811}
812
813static void diffuser_cpu_to_disk(__le32 *d, size_t n)
814{
815#ifndef __LITTLE_ENDIAN
816 int i;
817
818 for (i = 0; i < n; i++)
819 d[i] = cpu_to_le32((u32)d[i]);
820#endif
821}
822
823static void diffuser_a_decrypt(u32 *d, size_t n)
824{
825 int i, i1, i2, i3;
826
827 for (i = 0; i < 5; i++) {
828 i1 = 0;
829 i2 = n - 2;
830 i3 = n - 5;
831
832 while (i1 < (n - 1)) {
833 d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
834 i1++; i2++; i3++;
835
836 if (i3 >= n)
837 i3 -= n;
838
839 d[i1] += d[i2] ^ d[i3];
840 i1++; i2++; i3++;
841
842 if (i2 >= n)
843 i2 -= n;
844
845 d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
846 i1++; i2++; i3++;
847
848 d[i1] += d[i2] ^ d[i3];
849 i1++; i2++; i3++;
850 }
851 }
852}
853
854static void diffuser_a_encrypt(u32 *d, size_t n)
855{
856 int i, i1, i2, i3;
857
858 for (i = 0; i < 5; i++) {
859 i1 = n - 1;
860 i2 = n - 2 - 1;
861 i3 = n - 5 - 1;
862
863 while (i1 > 0) {
864 d[i1] -= d[i2] ^ d[i3];
865 i1--; i2--; i3--;
866
867 d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
868 i1--; i2--; i3--;
869
870 if (i2 < 0)
871 i2 += n;
872
873 d[i1] -= d[i2] ^ d[i3];
874 i1--; i2--; i3--;
875
876 if (i3 < 0)
877 i3 += n;
878
879 d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
880 i1--; i2--; i3--;
881 }
882 }
883}
884
885static void diffuser_b_decrypt(u32 *d, size_t n)
886{
887 int i, i1, i2, i3;
888
889 for (i = 0; i < 3; i++) {
890 i1 = 0;
891 i2 = 2;
892 i3 = 5;
893
894 while (i1 < (n - 1)) {
895 d[i1] += d[i2] ^ d[i3];
896 i1++; i2++; i3++;
897
898 d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
899 i1++; i2++; i3++;
900
901 if (i2 >= n)
902 i2 -= n;
903
904 d[i1] += d[i2] ^ d[i3];
905 i1++; i2++; i3++;
906
907 if (i3 >= n)
908 i3 -= n;
909
910 d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
911 i1++; i2++; i3++;
912 }
913 }
914}
915
916static void diffuser_b_encrypt(u32 *d, size_t n)
917{
918 int i, i1, i2, i3;
919
920 for (i = 0; i < 3; i++) {
921 i1 = n - 1;
922 i2 = 2 - 1;
923 i3 = 5 - 1;
924
925 while (i1 > 0) {
926 d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
927 i1--; i2--; i3--;
928
929 if (i3 < 0)
930 i3 += n;
931
932 d[i1] -= d[i2] ^ d[i3];
933 i1--; i2--; i3--;
934
935 if (i2 < 0)
936 i2 += n;
937
938 d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
939 i1--; i2--; i3--;
940
941 d[i1] -= d[i2] ^ d[i3];
942 i1--; i2--; i3--;
943 }
944 }
945}
946
947static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
948{
949 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
950 u8 *es, *ks, *data, *data2, *data_offset;
951 struct skcipher_request *req;
952 struct scatterlist *sg, *sg2, src, dst;
953 DECLARE_CRYPTO_WAIT(wait);
954 int i, r;
955
956 req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
957 es = kzalloc(16, GFP_NOIO); /* Key for AES */
958 ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
959
960 if (!req || !es || !ks) {
961 r = -ENOMEM;
962 goto out;
963 }
964
965 *(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
966
967 /* E(Ks, e(s)) */
968 sg_init_one(&src, es, 16);
969 sg_init_one(&dst, ks, 16);
970 skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
971 skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
972 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
973 if (r)
974 goto out;
975
976 /* E(Ks, e'(s)) */
977 es[15] = 0x80;
978 sg_init_one(&dst, &ks[16], 16);
979 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
980 if (r)
981 goto out;
982
983 sg = crypt_get_sg_data(cc, dmreq->sg_out);
984 data = kmap_local_page(sg_page(sg));
985 data_offset = data + sg->offset;
986
987 /* Cannot modify original bio, copy to sg_out and apply Elephant to it */
988 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
989 sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
990 data2 = kmap_local_page(sg_page(sg2));
991 memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
992 kunmap_local(data2);
993 }
994
995 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
996 diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
997 diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
998 diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
999 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1000 }
1001
1002 for (i = 0; i < (cc->sector_size / 32); i++)
1003 crypto_xor(data_offset + i * 32, ks, 32);
1004
1005 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1006 diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32));
1007 diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1008 diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32));
1009 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32));
1010 }
1011
1012 kunmap_local(data);
1013out:
1014 kfree_sensitive(ks);
1015 kfree_sensitive(es);
1016 skcipher_request_free(req);
1017 return r;
1018}
1019
1020static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1021 struct dm_crypt_request *dmreq)
1022{
1023 int r;
1024
1025 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1026 r = crypt_iv_elephant(cc, dmreq);
1027 if (r)
1028 return r;
1029 }
1030
1031 return crypt_iv_eboiv_gen(cc, iv, dmreq);
1032}
1033
1034static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1035 struct dm_crypt_request *dmreq)
1036{
1037 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1038 return crypt_iv_elephant(cc, dmreq);
1039
1040 return 0;
1041}
1042
1043static int crypt_iv_elephant_init(struct crypt_config *cc)
1044{
1045 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1046 int key_offset = cc->key_size - cc->key_extra_size;
1047
1048 return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1049}
1050
1051static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1052{
1053 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1054 u8 key[ELEPHANT_MAX_KEY_SIZE];
1055
1056 memset(key, 0, cc->key_extra_size);
1057 return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1058}
1059
1060static const struct crypt_iv_operations crypt_iv_plain_ops = {
1061 .generator = crypt_iv_plain_gen
1062};
1063
1064static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1065 .generator = crypt_iv_plain64_gen
1066};
1067
1068static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1069 .generator = crypt_iv_plain64be_gen
1070};
1071
1072static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1073 .generator = crypt_iv_essiv_gen
1074};
1075
1076static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1077 .ctr = crypt_iv_benbi_ctr,
1078 .dtr = crypt_iv_benbi_dtr,
1079 .generator = crypt_iv_benbi_gen
1080};
1081
1082static const struct crypt_iv_operations crypt_iv_null_ops = {
1083 .generator = crypt_iv_null_gen
1084};
1085
1086static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1087 .ctr = crypt_iv_lmk_ctr,
1088 .dtr = crypt_iv_lmk_dtr,
1089 .init = crypt_iv_lmk_init,
1090 .wipe = crypt_iv_lmk_wipe,
1091 .generator = crypt_iv_lmk_gen,
1092 .post = crypt_iv_lmk_post
1093};
1094
1095static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1096 .ctr = crypt_iv_tcw_ctr,
1097 .dtr = crypt_iv_tcw_dtr,
1098 .init = crypt_iv_tcw_init,
1099 .wipe = crypt_iv_tcw_wipe,
1100 .generator = crypt_iv_tcw_gen,
1101 .post = crypt_iv_tcw_post
1102};
1103
1104static const struct crypt_iv_operations crypt_iv_random_ops = {
1105 .generator = crypt_iv_random_gen
1106};
1107
1108static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1109 .ctr = crypt_iv_eboiv_ctr,
1110 .generator = crypt_iv_eboiv_gen
1111};
1112
1113static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1114 .ctr = crypt_iv_elephant_ctr,
1115 .dtr = crypt_iv_elephant_dtr,
1116 .init = crypt_iv_elephant_init,
1117 .wipe = crypt_iv_elephant_wipe,
1118 .generator = crypt_iv_elephant_gen,
1119 .post = crypt_iv_elephant_post
1120};
1121
1122/*
1123 * Integrity extensions
1124 */
1125static bool crypt_integrity_aead(struct crypt_config *cc)
1126{
1127 return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1128}
1129
1130static bool crypt_integrity_hmac(struct crypt_config *cc)
1131{
1132 return crypt_integrity_aead(cc) && cc->key_mac_size;
1133}
1134
1135/* Get sg containing data */
1136static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1137 struct scatterlist *sg)
1138{
1139 if (unlikely(crypt_integrity_aead(cc)))
1140 return &sg[2];
1141
1142 return sg;
1143}
1144
1145static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1146{
1147 struct bio_integrity_payload *bip;
1148 unsigned int tag_len;
1149 int ret;
1150
1151 if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1152 return 0;
1153
1154 bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1155 if (IS_ERR(bip))
1156 return PTR_ERR(bip);
1157
1158 tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1159
1160 bip->bip_iter.bi_sector = io->cc->start + io->sector;
1161
1162 ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1163 tag_len, offset_in_page(io->integrity_metadata));
1164 if (unlikely(ret != tag_len))
1165 return -ENOMEM;
1166
1167 return 0;
1168}
1169
1170static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1171{
1172#ifdef CONFIG_BLK_DEV_INTEGRITY
1173 struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1174 struct mapped_device *md = dm_table_get_md(ti->table);
1175
1176 /* From now we require underlying device with our integrity profile */
1177 if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1178 ti->error = "Integrity profile not supported.";
1179 return -EINVAL;
1180 }
1181
1182 if (bi->tag_size != cc->on_disk_tag_size ||
1183 bi->tuple_size != cc->on_disk_tag_size) {
1184 ti->error = "Integrity profile tag size mismatch.";
1185 return -EINVAL;
1186 }
1187 if (1 << bi->interval_exp != cc->sector_size) {
1188 ti->error = "Integrity profile sector size mismatch.";
1189 return -EINVAL;
1190 }
1191
1192 if (crypt_integrity_aead(cc)) {
1193 cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1194 DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1195 cc->integrity_tag_size, cc->integrity_iv_size);
1196
1197 if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1198 ti->error = "Integrity AEAD auth tag size is not supported.";
1199 return -EINVAL;
1200 }
1201 } else if (cc->integrity_iv_size)
1202 DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1203 cc->integrity_iv_size);
1204
1205 if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1206 ti->error = "Not enough space for integrity tag in the profile.";
1207 return -EINVAL;
1208 }
1209
1210 return 0;
1211#else
1212 ti->error = "Integrity profile not supported.";
1213 return -EINVAL;
1214#endif
1215}
1216
1217static void crypt_convert_init(struct crypt_config *cc,
1218 struct convert_context *ctx,
1219 struct bio *bio_out, struct bio *bio_in,
1220 sector_t sector)
1221{
1222 ctx->bio_in = bio_in;
1223 ctx->bio_out = bio_out;
1224 if (bio_in)
1225 ctx->iter_in = bio_in->bi_iter;
1226 if (bio_out)
1227 ctx->iter_out = bio_out->bi_iter;
1228 ctx->cc_sector = sector + cc->iv_offset;
1229 init_completion(&ctx->restart);
1230}
1231
1232static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1233 void *req)
1234{
1235 return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1236}
1237
1238static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1239{
1240 return (void *)((char *)dmreq - cc->dmreq_start);
1241}
1242
1243static u8 *iv_of_dmreq(struct crypt_config *cc,
1244 struct dm_crypt_request *dmreq)
1245{
1246 if (crypt_integrity_aead(cc))
1247 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1248 crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1249 else
1250 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1251 crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1252}
1253
1254static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1255 struct dm_crypt_request *dmreq)
1256{
1257 return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1258}
1259
1260static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1261 struct dm_crypt_request *dmreq)
1262{
1263 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1264
1265 return (__le64 *) ptr;
1266}
1267
1268static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1269 struct dm_crypt_request *dmreq)
1270{
1271 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1272 cc->iv_size + sizeof(uint64_t);
1273
1274 return (unsigned int *)ptr;
1275}
1276
1277static void *tag_from_dmreq(struct crypt_config *cc,
1278 struct dm_crypt_request *dmreq)
1279{
1280 struct convert_context *ctx = dmreq->ctx;
1281 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1282
1283 return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1284 cc->on_disk_tag_size];
1285}
1286
1287static void *iv_tag_from_dmreq(struct crypt_config *cc,
1288 struct dm_crypt_request *dmreq)
1289{
1290 return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1291}
1292
1293static int crypt_convert_block_aead(struct crypt_config *cc,
1294 struct convert_context *ctx,
1295 struct aead_request *req,
1296 unsigned int tag_offset)
1297{
1298 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1299 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1300 struct dm_crypt_request *dmreq;
1301 u8 *iv, *org_iv, *tag_iv, *tag;
1302 __le64 *sector;
1303 int r = 0;
1304
1305 BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1306
1307 /* Reject unexpected unaligned bio. */
1308 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1309 return -EIO;
1310
1311 dmreq = dmreq_of_req(cc, req);
1312 dmreq->iv_sector = ctx->cc_sector;
1313 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1314 dmreq->iv_sector >>= cc->sector_shift;
1315 dmreq->ctx = ctx;
1316
1317 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1318
1319 sector = org_sector_of_dmreq(cc, dmreq);
1320 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1321
1322 iv = iv_of_dmreq(cc, dmreq);
1323 org_iv = org_iv_of_dmreq(cc, dmreq);
1324 tag = tag_from_dmreq(cc, dmreq);
1325 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1326
1327 /* AEAD request:
1328 * |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1329 * | (authenticated) | (auth+encryption) | |
1330 * | sector_LE | IV | sector in/out | tag in/out |
1331 */
1332 sg_init_table(dmreq->sg_in, 4);
1333 sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1334 sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1335 sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1336 sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1337
1338 sg_init_table(dmreq->sg_out, 4);
1339 sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1340 sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1341 sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1342 sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1343
1344 if (cc->iv_gen_ops) {
1345 /* For READs use IV stored in integrity metadata */
1346 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1347 memcpy(org_iv, tag_iv, cc->iv_size);
1348 } else {
1349 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1350 if (r < 0)
1351 return r;
1352 /* Store generated IV in integrity metadata */
1353 if (cc->integrity_iv_size)
1354 memcpy(tag_iv, org_iv, cc->iv_size);
1355 }
1356 /* Working copy of IV, to be modified in crypto API */
1357 memcpy(iv, org_iv, cc->iv_size);
1358 }
1359
1360 aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1361 if (bio_data_dir(ctx->bio_in) == WRITE) {
1362 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1363 cc->sector_size, iv);
1364 r = crypto_aead_encrypt(req);
1365 if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1366 memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1367 cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1368 } else {
1369 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1370 cc->sector_size + cc->integrity_tag_size, iv);
1371 r = crypto_aead_decrypt(req);
1372 }
1373
1374 if (r == -EBADMSG) {
1375 sector_t s = le64_to_cpu(*sector);
1376
1377 ctx->aead_failed = true;
1378 if (ctx->aead_recheck) {
1379 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1380 ctx->bio_in->bi_bdev, s);
1381 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1382 ctx->bio_in, s, 0);
1383 }
1384 }
1385
1386 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1387 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1388
1389 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1390 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1391
1392 return r;
1393}
1394
1395static int crypt_convert_block_skcipher(struct crypt_config *cc,
1396 struct convert_context *ctx,
1397 struct skcipher_request *req,
1398 unsigned int tag_offset)
1399{
1400 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1401 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1402 struct scatterlist *sg_in, *sg_out;
1403 struct dm_crypt_request *dmreq;
1404 u8 *iv, *org_iv, *tag_iv;
1405 __le64 *sector;
1406 int r = 0;
1407
1408 /* Reject unexpected unaligned bio. */
1409 if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1410 return -EIO;
1411
1412 dmreq = dmreq_of_req(cc, req);
1413 dmreq->iv_sector = ctx->cc_sector;
1414 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1415 dmreq->iv_sector >>= cc->sector_shift;
1416 dmreq->ctx = ctx;
1417
1418 *org_tag_of_dmreq(cc, dmreq) = tag_offset;
1419
1420 iv = iv_of_dmreq(cc, dmreq);
1421 org_iv = org_iv_of_dmreq(cc, dmreq);
1422 tag_iv = iv_tag_from_dmreq(cc, dmreq);
1423
1424 sector = org_sector_of_dmreq(cc, dmreq);
1425 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1426
1427 /* For skcipher we use only the first sg item */
1428 sg_in = &dmreq->sg_in[0];
1429 sg_out = &dmreq->sg_out[0];
1430
1431 sg_init_table(sg_in, 1);
1432 sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1433
1434 sg_init_table(sg_out, 1);
1435 sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1436
1437 if (cc->iv_gen_ops) {
1438 /* For READs use IV stored in integrity metadata */
1439 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1440 memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1441 } else {
1442 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1443 if (r < 0)
1444 return r;
1445 /* Data can be already preprocessed in generator */
1446 if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1447 sg_in = sg_out;
1448 /* Store generated IV in integrity metadata */
1449 if (cc->integrity_iv_size)
1450 memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1451 }
1452 /* Working copy of IV, to be modified in crypto API */
1453 memcpy(iv, org_iv, cc->iv_size);
1454 }
1455
1456 skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1457
1458 if (bio_data_dir(ctx->bio_in) == WRITE)
1459 r = crypto_skcipher_encrypt(req);
1460 else
1461 r = crypto_skcipher_decrypt(req);
1462
1463 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1464 r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1465
1466 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1467 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1468
1469 return r;
1470}
1471
1472static void kcryptd_async_done(void *async_req, int error);
1473
1474static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1475 struct convert_context *ctx)
1476{
1477 unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1478
1479 if (!ctx->r.req) {
1480 ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1481 if (!ctx->r.req)
1482 return -ENOMEM;
1483 }
1484
1485 skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1486
1487 /*
1488 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1489 * requests if driver request queue is full.
1490 */
1491 skcipher_request_set_callback(ctx->r.req,
1492 CRYPTO_TFM_REQ_MAY_BACKLOG,
1493 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1494
1495 return 0;
1496}
1497
1498static int crypt_alloc_req_aead(struct crypt_config *cc,
1499 struct convert_context *ctx)
1500{
1501 if (!ctx->r.req_aead) {
1502 ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1503 if (!ctx->r.req_aead)
1504 return -ENOMEM;
1505 }
1506
1507 aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1508
1509 /*
1510 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1511 * requests if driver request queue is full.
1512 */
1513 aead_request_set_callback(ctx->r.req_aead,
1514 CRYPTO_TFM_REQ_MAY_BACKLOG,
1515 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1516
1517 return 0;
1518}
1519
1520static int crypt_alloc_req(struct crypt_config *cc,
1521 struct convert_context *ctx)
1522{
1523 if (crypt_integrity_aead(cc))
1524 return crypt_alloc_req_aead(cc, ctx);
1525 else
1526 return crypt_alloc_req_skcipher(cc, ctx);
1527}
1528
1529static void crypt_free_req_skcipher(struct crypt_config *cc,
1530 struct skcipher_request *req, struct bio *base_bio)
1531{
1532 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1533
1534 if ((struct skcipher_request *)(io + 1) != req)
1535 mempool_free(req, &cc->req_pool);
1536}
1537
1538static void crypt_free_req_aead(struct crypt_config *cc,
1539 struct aead_request *req, struct bio *base_bio)
1540{
1541 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1542
1543 if ((struct aead_request *)(io + 1) != req)
1544 mempool_free(req, &cc->req_pool);
1545}
1546
1547static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1548{
1549 if (crypt_integrity_aead(cc))
1550 crypt_free_req_aead(cc, req, base_bio);
1551 else
1552 crypt_free_req_skcipher(cc, req, base_bio);
1553}
1554
1555/*
1556 * Encrypt / decrypt data from one bio to another one (can be the same one)
1557 */
1558static blk_status_t crypt_convert(struct crypt_config *cc,
1559 struct convert_context *ctx, bool atomic, bool reset_pending)
1560{
1561 unsigned int tag_offset = 0;
1562 unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1563 int r;
1564
1565 /*
1566 * if reset_pending is set we are dealing with the bio for the first time,
1567 * else we're continuing to work on the previous bio, so don't mess with
1568 * the cc_pending counter
1569 */
1570 if (reset_pending)
1571 atomic_set(&ctx->cc_pending, 1);
1572
1573 while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1574
1575 r = crypt_alloc_req(cc, ctx);
1576 if (r) {
1577 complete(&ctx->restart);
1578 return BLK_STS_DEV_RESOURCE;
1579 }
1580
1581 atomic_inc(&ctx->cc_pending);
1582
1583 if (crypt_integrity_aead(cc))
1584 r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1585 else
1586 r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1587
1588 switch (r) {
1589 /*
1590 * The request was queued by a crypto driver
1591 * but the driver request queue is full, let's wait.
1592 */
1593 case -EBUSY:
1594 if (in_interrupt()) {
1595 if (try_wait_for_completion(&ctx->restart)) {
1596 /*
1597 * we don't have to block to wait for completion,
1598 * so proceed
1599 */
1600 } else {
1601 /*
1602 * we can't wait for completion without blocking
1603 * exit and continue processing in a workqueue
1604 */
1605 ctx->r.req = NULL;
1606 ctx->cc_sector += sector_step;
1607 tag_offset++;
1608 return BLK_STS_DEV_RESOURCE;
1609 }
1610 } else {
1611 wait_for_completion(&ctx->restart);
1612 }
1613 reinit_completion(&ctx->restart);
1614 fallthrough;
1615 /*
1616 * The request is queued and processed asynchronously,
1617 * completion function kcryptd_async_done() will be called.
1618 */
1619 case -EINPROGRESS:
1620 ctx->r.req = NULL;
1621 ctx->cc_sector += sector_step;
1622 tag_offset++;
1623 continue;
1624 /*
1625 * The request was already processed (synchronously).
1626 */
1627 case 0:
1628 atomic_dec(&ctx->cc_pending);
1629 ctx->cc_sector += sector_step;
1630 tag_offset++;
1631 if (!atomic)
1632 cond_resched();
1633 continue;
1634 /*
1635 * There was a data integrity error.
1636 */
1637 case -EBADMSG:
1638 atomic_dec(&ctx->cc_pending);
1639 return BLK_STS_PROTECTION;
1640 /*
1641 * There was an error while processing the request.
1642 */
1643 default:
1644 atomic_dec(&ctx->cc_pending);
1645 return BLK_STS_IOERR;
1646 }
1647 }
1648
1649 return 0;
1650}
1651
1652static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1653
1654/*
1655 * Generate a new unfragmented bio with the given size
1656 * This should never violate the device limitations (but only because
1657 * max_segment_size is being constrained to PAGE_SIZE).
1658 *
1659 * This function may be called concurrently. If we allocate from the mempool
1660 * concurrently, there is a possibility of deadlock. For example, if we have
1661 * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1662 * the mempool concurrently, it may deadlock in a situation where both processes
1663 * have allocated 128 pages and the mempool is exhausted.
1664 *
1665 * In order to avoid this scenario we allocate the pages under a mutex.
1666 *
1667 * In order to not degrade performance with excessive locking, we try
1668 * non-blocking allocations without a mutex first but on failure we fallback
1669 * to blocking allocations with a mutex.
1670 *
1671 * In order to reduce allocation overhead, we try to allocate compound pages in
1672 * the first pass. If they are not available, we fall back to the mempool.
1673 */
1674static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1675{
1676 struct crypt_config *cc = io->cc;
1677 struct bio *clone;
1678 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1679 gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1680 unsigned int remaining_size;
1681 unsigned int order = MAX_PAGE_ORDER;
1682
1683retry:
1684 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1685 mutex_lock(&cc->bio_alloc_lock);
1686
1687 clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1688 GFP_NOIO, &cc->bs);
1689 clone->bi_private = io;
1690 clone->bi_end_io = crypt_endio;
1691
1692 remaining_size = size;
1693
1694 while (remaining_size) {
1695 struct page *pages;
1696 unsigned size_to_add;
1697 unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT);
1698 order = min(order, remaining_order);
1699
1700 while (order > 0) {
1701 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) +
1702 (1 << order) > dm_crypt_pages_per_client))
1703 goto decrease_order;
1704 pages = alloc_pages(gfp_mask
1705 | __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1706 order);
1707 if (likely(pages != NULL)) {
1708 percpu_counter_add(&cc->n_allocated_pages, 1 << order);
1709 goto have_pages;
1710 }
1711decrease_order:
1712 order--;
1713 }
1714
1715 pages = mempool_alloc(&cc->page_pool, gfp_mask);
1716 if (!pages) {
1717 crypt_free_buffer_pages(cc, clone);
1718 bio_put(clone);
1719 gfp_mask |= __GFP_DIRECT_RECLAIM;
1720 order = 0;
1721 goto retry;
1722 }
1723
1724have_pages:
1725 size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1726 __bio_add_page(clone, pages, size_to_add, 0);
1727 remaining_size -= size_to_add;
1728 }
1729
1730 /* Allocate space for integrity tags */
1731 if (dm_crypt_integrity_io_alloc(io, clone)) {
1732 crypt_free_buffer_pages(cc, clone);
1733 bio_put(clone);
1734 clone = NULL;
1735 }
1736
1737 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1738 mutex_unlock(&cc->bio_alloc_lock);
1739
1740 return clone;
1741}
1742
1743static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1744{
1745 struct folio_iter fi;
1746
1747 if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1748 bio_for_each_folio_all(fi, clone) {
1749 if (folio_test_large(fi.folio)) {
1750 percpu_counter_sub(&cc->n_allocated_pages,
1751 1 << folio_order(fi.folio));
1752 folio_put(fi.folio);
1753 } else {
1754 mempool_free(&fi.folio->page, &cc->page_pool);
1755 }
1756 }
1757 }
1758}
1759
1760static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1761 struct bio *bio, sector_t sector)
1762{
1763 io->cc = cc;
1764 io->base_bio = bio;
1765 io->sector = sector;
1766 io->error = 0;
1767 io->ctx.aead_recheck = false;
1768 io->ctx.aead_failed = false;
1769 io->ctx.r.req = NULL;
1770 io->integrity_metadata = NULL;
1771 io->integrity_metadata_from_pool = false;
1772 atomic_set(&io->io_pending, 0);
1773}
1774
1775static void crypt_inc_pending(struct dm_crypt_io *io)
1776{
1777 atomic_inc(&io->io_pending);
1778}
1779
1780static void kcryptd_queue_read(struct dm_crypt_io *io);
1781
1782/*
1783 * One of the bios was finished. Check for completion of
1784 * the whole request and correctly clean up the buffer.
1785 */
1786static void crypt_dec_pending(struct dm_crypt_io *io)
1787{
1788 struct crypt_config *cc = io->cc;
1789 struct bio *base_bio = io->base_bio;
1790 blk_status_t error = io->error;
1791
1792 if (!atomic_dec_and_test(&io->io_pending))
1793 return;
1794
1795 if (likely(!io->ctx.aead_recheck) && unlikely(io->ctx.aead_failed) &&
1796 cc->on_disk_tag_size && bio_data_dir(base_bio) == READ) {
1797 io->ctx.aead_recheck = true;
1798 io->ctx.aead_failed = false;
1799 io->error = 0;
1800 kcryptd_queue_read(io);
1801 return;
1802 }
1803
1804 if (io->ctx.r.req)
1805 crypt_free_req(cc, io->ctx.r.req, base_bio);
1806
1807 if (unlikely(io->integrity_metadata_from_pool))
1808 mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1809 else
1810 kfree(io->integrity_metadata);
1811
1812 base_bio->bi_status = error;
1813
1814 bio_endio(base_bio);
1815}
1816
1817/*
1818 * kcryptd/kcryptd_io:
1819 *
1820 * Needed because it would be very unwise to do decryption in an
1821 * interrupt context.
1822 *
1823 * kcryptd performs the actual encryption or decryption.
1824 *
1825 * kcryptd_io performs the IO submission.
1826 *
1827 * They must be separated as otherwise the final stages could be
1828 * starved by new requests which can block in the first stages due
1829 * to memory allocation.
1830 *
1831 * The work is done per CPU global for all dm-crypt instances.
1832 * They should not depend on each other and do not block.
1833 */
1834static void crypt_endio(struct bio *clone)
1835{
1836 struct dm_crypt_io *io = clone->bi_private;
1837 struct crypt_config *cc = io->cc;
1838 unsigned int rw = bio_data_dir(clone);
1839 blk_status_t error = clone->bi_status;
1840
1841 if (io->ctx.aead_recheck && !error) {
1842 kcryptd_queue_crypt(io);
1843 return;
1844 }
1845
1846 /*
1847 * free the processed pages
1848 */
1849 if (rw == WRITE || io->ctx.aead_recheck)
1850 crypt_free_buffer_pages(cc, clone);
1851
1852 bio_put(clone);
1853
1854 if (rw == READ && !error) {
1855 kcryptd_queue_crypt(io);
1856 return;
1857 }
1858
1859 if (unlikely(error))
1860 io->error = error;
1861
1862 crypt_dec_pending(io);
1863}
1864
1865#define CRYPT_MAP_READ_GFP GFP_NOWAIT
1866
1867static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1868{
1869 struct crypt_config *cc = io->cc;
1870 struct bio *clone;
1871
1872 if (io->ctx.aead_recheck) {
1873 if (!(gfp & __GFP_DIRECT_RECLAIM))
1874 return 1;
1875 crypt_inc_pending(io);
1876 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1877 if (unlikely(!clone)) {
1878 crypt_dec_pending(io);
1879 return 1;
1880 }
1881 clone->bi_iter.bi_sector = cc->start + io->sector;
1882 crypt_convert_init(cc, &io->ctx, clone, clone, io->sector);
1883 io->saved_bi_iter = clone->bi_iter;
1884 dm_submit_bio_remap(io->base_bio, clone);
1885 return 0;
1886 }
1887
1888 /*
1889 * We need the original biovec array in order to decrypt the whole bio
1890 * data *afterwards* -- thanks to immutable biovecs we don't need to
1891 * worry about the block layer modifying the biovec array; so leverage
1892 * bio_alloc_clone().
1893 */
1894 clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1895 if (!clone)
1896 return 1;
1897 clone->bi_private = io;
1898 clone->bi_end_io = crypt_endio;
1899
1900 crypt_inc_pending(io);
1901
1902 clone->bi_iter.bi_sector = cc->start + io->sector;
1903
1904 if (dm_crypt_integrity_io_alloc(io, clone)) {
1905 crypt_dec_pending(io);
1906 bio_put(clone);
1907 return 1;
1908 }
1909
1910 dm_submit_bio_remap(io->base_bio, clone);
1911 return 0;
1912}
1913
1914static void kcryptd_io_read_work(struct work_struct *work)
1915{
1916 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1917
1918 crypt_inc_pending(io);
1919 if (kcryptd_io_read(io, GFP_NOIO))
1920 io->error = BLK_STS_RESOURCE;
1921 crypt_dec_pending(io);
1922}
1923
1924static void kcryptd_queue_read(struct dm_crypt_io *io)
1925{
1926 struct crypt_config *cc = io->cc;
1927
1928 INIT_WORK(&io->work, kcryptd_io_read_work);
1929 queue_work(cc->io_queue, &io->work);
1930}
1931
1932static void kcryptd_io_write(struct dm_crypt_io *io)
1933{
1934 struct bio *clone = io->ctx.bio_out;
1935
1936 dm_submit_bio_remap(io->base_bio, clone);
1937}
1938
1939#define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1940
1941static int dmcrypt_write(void *data)
1942{
1943 struct crypt_config *cc = data;
1944 struct dm_crypt_io *io;
1945
1946 while (1) {
1947 struct rb_root write_tree;
1948 struct blk_plug plug;
1949
1950 spin_lock_irq(&cc->write_thread_lock);
1951continue_locked:
1952
1953 if (!RB_EMPTY_ROOT(&cc->write_tree))
1954 goto pop_from_list;
1955
1956 set_current_state(TASK_INTERRUPTIBLE);
1957
1958 spin_unlock_irq(&cc->write_thread_lock);
1959
1960 if (unlikely(kthread_should_stop())) {
1961 set_current_state(TASK_RUNNING);
1962 break;
1963 }
1964
1965 schedule();
1966
1967 set_current_state(TASK_RUNNING);
1968 spin_lock_irq(&cc->write_thread_lock);
1969 goto continue_locked;
1970
1971pop_from_list:
1972 write_tree = cc->write_tree;
1973 cc->write_tree = RB_ROOT;
1974 spin_unlock_irq(&cc->write_thread_lock);
1975
1976 BUG_ON(rb_parent(write_tree.rb_node));
1977
1978 /*
1979 * Note: we cannot walk the tree here with rb_next because
1980 * the structures may be freed when kcryptd_io_write is called.
1981 */
1982 blk_start_plug(&plug);
1983 do {
1984 io = crypt_io_from_node(rb_first(&write_tree));
1985 rb_erase(&io->rb_node, &write_tree);
1986 kcryptd_io_write(io);
1987 cond_resched();
1988 } while (!RB_EMPTY_ROOT(&write_tree));
1989 blk_finish_plug(&plug);
1990 }
1991 return 0;
1992}
1993
1994static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1995{
1996 struct bio *clone = io->ctx.bio_out;
1997 struct crypt_config *cc = io->cc;
1998 unsigned long flags;
1999 sector_t sector;
2000 struct rb_node **rbp, *parent;
2001
2002 if (unlikely(io->error)) {
2003 crypt_free_buffer_pages(cc, clone);
2004 bio_put(clone);
2005 crypt_dec_pending(io);
2006 return;
2007 }
2008
2009 /* crypt_convert should have filled the clone bio */
2010 BUG_ON(io->ctx.iter_out.bi_size);
2011
2012 clone->bi_iter.bi_sector = cc->start + io->sector;
2013
2014 if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
2015 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
2016 dm_submit_bio_remap(io->base_bio, clone);
2017 return;
2018 }
2019
2020 spin_lock_irqsave(&cc->write_thread_lock, flags);
2021 if (RB_EMPTY_ROOT(&cc->write_tree))
2022 wake_up_process(cc->write_thread);
2023 rbp = &cc->write_tree.rb_node;
2024 parent = NULL;
2025 sector = io->sector;
2026 while (*rbp) {
2027 parent = *rbp;
2028 if (sector < crypt_io_from_node(parent)->sector)
2029 rbp = &(*rbp)->rb_left;
2030 else
2031 rbp = &(*rbp)->rb_right;
2032 }
2033 rb_link_node(&io->rb_node, parent, rbp);
2034 rb_insert_color(&io->rb_node, &cc->write_tree);
2035 spin_unlock_irqrestore(&cc->write_thread_lock, flags);
2036}
2037
2038static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
2039 struct convert_context *ctx)
2040
2041{
2042 if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2043 return false;
2044
2045 /*
2046 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2047 * constraints so they do not need to be issued inline by
2048 * kcryptd_crypt_write_convert().
2049 */
2050 switch (bio_op(ctx->bio_in)) {
2051 case REQ_OP_WRITE:
2052 case REQ_OP_WRITE_ZEROES:
2053 return true;
2054 default:
2055 return false;
2056 }
2057}
2058
2059static void kcryptd_crypt_write_continue(struct work_struct *work)
2060{
2061 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2062 struct crypt_config *cc = io->cc;
2063 struct convert_context *ctx = &io->ctx;
2064 int crypt_finished;
2065 sector_t sector = io->sector;
2066 blk_status_t r;
2067
2068 wait_for_completion(&ctx->restart);
2069 reinit_completion(&ctx->restart);
2070
2071 r = crypt_convert(cc, &io->ctx, true, false);
2072 if (r)
2073 io->error = r;
2074 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2075 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2076 /* Wait for completion signaled by kcryptd_async_done() */
2077 wait_for_completion(&ctx->restart);
2078 crypt_finished = 1;
2079 }
2080
2081 /* Encryption was already finished, submit io now */
2082 if (crypt_finished) {
2083 kcryptd_crypt_write_io_submit(io, 0);
2084 io->sector = sector;
2085 }
2086
2087 crypt_dec_pending(io);
2088}
2089
2090static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2091{
2092 struct crypt_config *cc = io->cc;
2093 struct convert_context *ctx = &io->ctx;
2094 struct bio *clone;
2095 int crypt_finished;
2096 sector_t sector = io->sector;
2097 blk_status_t r;
2098
2099 /*
2100 * Prevent io from disappearing until this function completes.
2101 */
2102 crypt_inc_pending(io);
2103 crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2104
2105 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2106 if (unlikely(!clone)) {
2107 io->error = BLK_STS_IOERR;
2108 goto dec;
2109 }
2110
2111 io->ctx.bio_out = clone;
2112 io->ctx.iter_out = clone->bi_iter;
2113
2114 if (crypt_integrity_aead(cc)) {
2115 bio_copy_data(clone, io->base_bio);
2116 io->ctx.bio_in = clone;
2117 io->ctx.iter_in = clone->bi_iter;
2118 }
2119
2120 sector += bio_sectors(clone);
2121
2122 crypt_inc_pending(io);
2123 r = crypt_convert(cc, ctx,
2124 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2125 /*
2126 * Crypto API backlogged the request, because its queue was full
2127 * and we're in softirq context, so continue from a workqueue
2128 * (TODO: is it actually possible to be in softirq in the write path?)
2129 */
2130 if (r == BLK_STS_DEV_RESOURCE) {
2131 INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2132 queue_work(cc->crypt_queue, &io->work);
2133 return;
2134 }
2135 if (r)
2136 io->error = r;
2137 crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2138 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2139 /* Wait for completion signaled by kcryptd_async_done() */
2140 wait_for_completion(&ctx->restart);
2141 crypt_finished = 1;
2142 }
2143
2144 /* Encryption was already finished, submit io now */
2145 if (crypt_finished) {
2146 kcryptd_crypt_write_io_submit(io, 0);
2147 io->sector = sector;
2148 }
2149
2150dec:
2151 crypt_dec_pending(io);
2152}
2153
2154static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2155{
2156 if (io->ctx.aead_recheck) {
2157 if (!io->error) {
2158 io->ctx.bio_in->bi_iter = io->saved_bi_iter;
2159 bio_copy_data(io->base_bio, io->ctx.bio_in);
2160 }
2161 crypt_free_buffer_pages(io->cc, io->ctx.bio_in);
2162 bio_put(io->ctx.bio_in);
2163 }
2164 crypt_dec_pending(io);
2165}
2166
2167static void kcryptd_crypt_read_continue(struct work_struct *work)
2168{
2169 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2170 struct crypt_config *cc = io->cc;
2171 blk_status_t r;
2172
2173 wait_for_completion(&io->ctx.restart);
2174 reinit_completion(&io->ctx.restart);
2175
2176 r = crypt_convert(cc, &io->ctx, true, false);
2177 if (r)
2178 io->error = r;
2179
2180 if (atomic_dec_and_test(&io->ctx.cc_pending))
2181 kcryptd_crypt_read_done(io);
2182
2183 crypt_dec_pending(io);
2184}
2185
2186static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2187{
2188 struct crypt_config *cc = io->cc;
2189 blk_status_t r;
2190
2191 crypt_inc_pending(io);
2192
2193 if (io->ctx.aead_recheck) {
2194 io->ctx.cc_sector = io->sector + cc->iv_offset;
2195 r = crypt_convert(cc, &io->ctx,
2196 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2197 } else {
2198 crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2199 io->sector);
2200
2201 r = crypt_convert(cc, &io->ctx,
2202 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2203 }
2204 /*
2205 * Crypto API backlogged the request, because its queue was full
2206 * and we're in softirq context, so continue from a workqueue
2207 */
2208 if (r == BLK_STS_DEV_RESOURCE) {
2209 INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2210 queue_work(cc->crypt_queue, &io->work);
2211 return;
2212 }
2213 if (r)
2214 io->error = r;
2215
2216 if (atomic_dec_and_test(&io->ctx.cc_pending))
2217 kcryptd_crypt_read_done(io);
2218
2219 crypt_dec_pending(io);
2220}
2221
2222static void kcryptd_async_done(void *data, int error)
2223{
2224 struct dm_crypt_request *dmreq = data;
2225 struct convert_context *ctx = dmreq->ctx;
2226 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2227 struct crypt_config *cc = io->cc;
2228
2229 /*
2230 * A request from crypto driver backlog is going to be processed now,
2231 * finish the completion and continue in crypt_convert().
2232 * (Callback will be called for the second time for this request.)
2233 */
2234 if (error == -EINPROGRESS) {
2235 complete(&ctx->restart);
2236 return;
2237 }
2238
2239 if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2240 error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2241
2242 if (error == -EBADMSG) {
2243 sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2244
2245 ctx->aead_failed = true;
2246 if (ctx->aead_recheck) {
2247 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2248 ctx->bio_in->bi_bdev, s);
2249 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2250 ctx->bio_in, s, 0);
2251 }
2252 io->error = BLK_STS_PROTECTION;
2253 } else if (error < 0)
2254 io->error = BLK_STS_IOERR;
2255
2256 crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2257
2258 if (!atomic_dec_and_test(&ctx->cc_pending))
2259 return;
2260
2261 /*
2262 * The request is fully completed: for inline writes, let
2263 * kcryptd_crypt_write_convert() do the IO submission.
2264 */
2265 if (bio_data_dir(io->base_bio) == READ) {
2266 kcryptd_crypt_read_done(io);
2267 return;
2268 }
2269
2270 if (kcryptd_crypt_write_inline(cc, ctx)) {
2271 complete(&ctx->restart);
2272 return;
2273 }
2274
2275 kcryptd_crypt_write_io_submit(io, 1);
2276}
2277
2278static void kcryptd_crypt(struct work_struct *work)
2279{
2280 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2281
2282 if (bio_data_dir(io->base_bio) == READ)
2283 kcryptd_crypt_read_convert(io);
2284 else
2285 kcryptd_crypt_write_convert(io);
2286}
2287
2288static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2289{
2290 struct crypt_config *cc = io->cc;
2291
2292 if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2293 (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2294 /*
2295 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2296 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2297 * it is being executed with irqs disabled.
2298 */
2299 if (!(in_hardirq() || irqs_disabled())) {
2300 kcryptd_crypt(&io->work);
2301 return;
2302 }
2303 }
2304
2305 INIT_WORK(&io->work, kcryptd_crypt);
2306 queue_work(cc->crypt_queue, &io->work);
2307}
2308
2309static void crypt_free_tfms_aead(struct crypt_config *cc)
2310{
2311 if (!cc->cipher_tfm.tfms_aead)
2312 return;
2313
2314 if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2315 crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2316 cc->cipher_tfm.tfms_aead[0] = NULL;
2317 }
2318
2319 kfree(cc->cipher_tfm.tfms_aead);
2320 cc->cipher_tfm.tfms_aead = NULL;
2321}
2322
2323static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2324{
2325 unsigned int i;
2326
2327 if (!cc->cipher_tfm.tfms)
2328 return;
2329
2330 for (i = 0; i < cc->tfms_count; i++)
2331 if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2332 crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2333 cc->cipher_tfm.tfms[i] = NULL;
2334 }
2335
2336 kfree(cc->cipher_tfm.tfms);
2337 cc->cipher_tfm.tfms = NULL;
2338}
2339
2340static void crypt_free_tfms(struct crypt_config *cc)
2341{
2342 if (crypt_integrity_aead(cc))
2343 crypt_free_tfms_aead(cc);
2344 else
2345 crypt_free_tfms_skcipher(cc);
2346}
2347
2348static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2349{
2350 unsigned int i;
2351 int err;
2352
2353 cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2354 sizeof(struct crypto_skcipher *),
2355 GFP_KERNEL);
2356 if (!cc->cipher_tfm.tfms)
2357 return -ENOMEM;
2358
2359 for (i = 0; i < cc->tfms_count; i++) {
2360 cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2361 CRYPTO_ALG_ALLOCATES_MEMORY);
2362 if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2363 err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2364 crypt_free_tfms(cc);
2365 return err;
2366 }
2367 }
2368
2369 /*
2370 * dm-crypt performance can vary greatly depending on which crypto
2371 * algorithm implementation is used. Help people debug performance
2372 * problems by logging the ->cra_driver_name.
2373 */
2374 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2375 crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2376 return 0;
2377}
2378
2379static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2380{
2381 int err;
2382
2383 cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2384 if (!cc->cipher_tfm.tfms)
2385 return -ENOMEM;
2386
2387 cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2388 CRYPTO_ALG_ALLOCATES_MEMORY);
2389 if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2390 err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2391 crypt_free_tfms(cc);
2392 return err;
2393 }
2394
2395 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2396 crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2397 return 0;
2398}
2399
2400static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2401{
2402 if (crypt_integrity_aead(cc))
2403 return crypt_alloc_tfms_aead(cc, ciphermode);
2404 else
2405 return crypt_alloc_tfms_skcipher(cc, ciphermode);
2406}
2407
2408static unsigned int crypt_subkey_size(struct crypt_config *cc)
2409{
2410 return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2411}
2412
2413static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2414{
2415 return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2416}
2417
2418/*
2419 * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2420 * the key must be for some reason in special format.
2421 * This funcion converts cc->key to this special format.
2422 */
2423static void crypt_copy_authenckey(char *p, const void *key,
2424 unsigned int enckeylen, unsigned int authkeylen)
2425{
2426 struct crypto_authenc_key_param *param;
2427 struct rtattr *rta;
2428
2429 rta = (struct rtattr *)p;
2430 param = RTA_DATA(rta);
2431 param->enckeylen = cpu_to_be32(enckeylen);
2432 rta->rta_len = RTA_LENGTH(sizeof(*param));
2433 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2434 p += RTA_SPACE(sizeof(*param));
2435 memcpy(p, key + enckeylen, authkeylen);
2436 p += authkeylen;
2437 memcpy(p, key, enckeylen);
2438}
2439
2440static int crypt_setkey(struct crypt_config *cc)
2441{
2442 unsigned int subkey_size;
2443 int err = 0, i, r;
2444
2445 /* Ignore extra keys (which are used for IV etc) */
2446 subkey_size = crypt_subkey_size(cc);
2447
2448 if (crypt_integrity_hmac(cc)) {
2449 if (subkey_size < cc->key_mac_size)
2450 return -EINVAL;
2451
2452 crypt_copy_authenckey(cc->authenc_key, cc->key,
2453 subkey_size - cc->key_mac_size,
2454 cc->key_mac_size);
2455 }
2456
2457 for (i = 0; i < cc->tfms_count; i++) {
2458 if (crypt_integrity_hmac(cc))
2459 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2460 cc->authenc_key, crypt_authenckey_size(cc));
2461 else if (crypt_integrity_aead(cc))
2462 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2463 cc->key + (i * subkey_size),
2464 subkey_size);
2465 else
2466 r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2467 cc->key + (i * subkey_size),
2468 subkey_size);
2469 if (r)
2470 err = r;
2471 }
2472
2473 if (crypt_integrity_hmac(cc))
2474 memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2475
2476 return err;
2477}
2478
2479#ifdef CONFIG_KEYS
2480
2481static bool contains_whitespace(const char *str)
2482{
2483 while (*str)
2484 if (isspace(*str++))
2485 return true;
2486 return false;
2487}
2488
2489static int set_key_user(struct crypt_config *cc, struct key *key)
2490{
2491 const struct user_key_payload *ukp;
2492
2493 ukp = user_key_payload_locked(key);
2494 if (!ukp)
2495 return -EKEYREVOKED;
2496
2497 if (cc->key_size != ukp->datalen)
2498 return -EINVAL;
2499
2500 memcpy(cc->key, ukp->data, cc->key_size);
2501
2502 return 0;
2503}
2504
2505static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2506{
2507 const struct encrypted_key_payload *ekp;
2508
2509 ekp = key->payload.data[0];
2510 if (!ekp)
2511 return -EKEYREVOKED;
2512
2513 if (cc->key_size != ekp->decrypted_datalen)
2514 return -EINVAL;
2515
2516 memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2517
2518 return 0;
2519}
2520
2521static int set_key_trusted(struct crypt_config *cc, struct key *key)
2522{
2523 const struct trusted_key_payload *tkp;
2524
2525 tkp = key->payload.data[0];
2526 if (!tkp)
2527 return -EKEYREVOKED;
2528
2529 if (cc->key_size != tkp->key_len)
2530 return -EINVAL;
2531
2532 memcpy(cc->key, tkp->key, cc->key_size);
2533
2534 return 0;
2535}
2536
2537static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2538{
2539 char *new_key_string, *key_desc;
2540 int ret;
2541 struct key_type *type;
2542 struct key *key;
2543 int (*set_key)(struct crypt_config *cc, struct key *key);
2544
2545 /*
2546 * Reject key_string with whitespace. dm core currently lacks code for
2547 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2548 */
2549 if (contains_whitespace(key_string)) {
2550 DMERR("whitespace chars not allowed in key string");
2551 return -EINVAL;
2552 }
2553
2554 /* look for next ':' separating key_type from key_description */
2555 key_desc = strchr(key_string, ':');
2556 if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2557 return -EINVAL;
2558
2559 if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2560 type = &key_type_logon;
2561 set_key = set_key_user;
2562 } else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2563 type = &key_type_user;
2564 set_key = set_key_user;
2565 } else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2566 !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2567 type = &key_type_encrypted;
2568 set_key = set_key_encrypted;
2569 } else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2570 !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2571 type = &key_type_trusted;
2572 set_key = set_key_trusted;
2573 } else {
2574 return -EINVAL;
2575 }
2576
2577 new_key_string = kstrdup(key_string, GFP_KERNEL);
2578 if (!new_key_string)
2579 return -ENOMEM;
2580
2581 key = request_key(type, key_desc + 1, NULL);
2582 if (IS_ERR(key)) {
2583 kfree_sensitive(new_key_string);
2584 return PTR_ERR(key);
2585 }
2586
2587 down_read(&key->sem);
2588
2589 ret = set_key(cc, key);
2590 if (ret < 0) {
2591 up_read(&key->sem);
2592 key_put(key);
2593 kfree_sensitive(new_key_string);
2594 return ret;
2595 }
2596
2597 up_read(&key->sem);
2598 key_put(key);
2599
2600 /* clear the flag since following operations may invalidate previously valid key */
2601 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2602
2603 ret = crypt_setkey(cc);
2604
2605 if (!ret) {
2606 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2607 kfree_sensitive(cc->key_string);
2608 cc->key_string = new_key_string;
2609 } else
2610 kfree_sensitive(new_key_string);
2611
2612 return ret;
2613}
2614
2615static int get_key_size(char **key_string)
2616{
2617 char *colon, dummy;
2618 int ret;
2619
2620 if (*key_string[0] != ':')
2621 return strlen(*key_string) >> 1;
2622
2623 /* look for next ':' in key string */
2624 colon = strpbrk(*key_string + 1, ":");
2625 if (!colon)
2626 return -EINVAL;
2627
2628 if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2629 return -EINVAL;
2630
2631 *key_string = colon;
2632
2633 /* remaining key string should be :<logon|user>:<key_desc> */
2634
2635 return ret;
2636}
2637
2638#else
2639
2640static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2641{
2642 return -EINVAL;
2643}
2644
2645static int get_key_size(char **key_string)
2646{
2647 return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2648}
2649
2650#endif /* CONFIG_KEYS */
2651
2652static int crypt_set_key(struct crypt_config *cc, char *key)
2653{
2654 int r = -EINVAL;
2655 int key_string_len = strlen(key);
2656
2657 /* Hyphen (which gives a key_size of zero) means there is no key. */
2658 if (!cc->key_size && strcmp(key, "-"))
2659 goto out;
2660
2661 /* ':' means the key is in kernel keyring, short-circuit normal key processing */
2662 if (key[0] == ':') {
2663 r = crypt_set_keyring_key(cc, key + 1);
2664 goto out;
2665 }
2666
2667 /* clear the flag since following operations may invalidate previously valid key */
2668 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2669
2670 /* wipe references to any kernel keyring key */
2671 kfree_sensitive(cc->key_string);
2672 cc->key_string = NULL;
2673
2674 /* Decode key from its hex representation. */
2675 if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2676 goto out;
2677
2678 r = crypt_setkey(cc);
2679 if (!r)
2680 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2681
2682out:
2683 /* Hex key string not needed after here, so wipe it. */
2684 memset(key, '0', key_string_len);
2685
2686 return r;
2687}
2688
2689static int crypt_wipe_key(struct crypt_config *cc)
2690{
2691 int r;
2692
2693 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2694 get_random_bytes(&cc->key, cc->key_size);
2695
2696 /* Wipe IV private keys */
2697 if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2698 r = cc->iv_gen_ops->wipe(cc);
2699 if (r)
2700 return r;
2701 }
2702
2703 kfree_sensitive(cc->key_string);
2704 cc->key_string = NULL;
2705 r = crypt_setkey(cc);
2706 memset(&cc->key, 0, cc->key_size * sizeof(u8));
2707
2708 return r;
2709}
2710
2711static void crypt_calculate_pages_per_client(void)
2712{
2713 unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2714
2715 if (!dm_crypt_clients_n)
2716 return;
2717
2718 pages /= dm_crypt_clients_n;
2719 if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2720 pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2721 dm_crypt_pages_per_client = pages;
2722}
2723
2724static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2725{
2726 struct crypt_config *cc = pool_data;
2727 struct page *page;
2728
2729 /*
2730 * Note, percpu_counter_read_positive() may over (and under) estimate
2731 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2732 * but avoids potential spinlock contention of an exact result.
2733 */
2734 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2735 likely(gfp_mask & __GFP_NORETRY))
2736 return NULL;
2737
2738 page = alloc_page(gfp_mask);
2739 if (likely(page != NULL))
2740 percpu_counter_add(&cc->n_allocated_pages, 1);
2741
2742 return page;
2743}
2744
2745static void crypt_page_free(void *page, void *pool_data)
2746{
2747 struct crypt_config *cc = pool_data;
2748
2749 __free_page(page);
2750 percpu_counter_sub(&cc->n_allocated_pages, 1);
2751}
2752
2753static void crypt_dtr(struct dm_target *ti)
2754{
2755 struct crypt_config *cc = ti->private;
2756
2757 ti->private = NULL;
2758
2759 if (!cc)
2760 return;
2761
2762 if (cc->write_thread)
2763 kthread_stop(cc->write_thread);
2764
2765 if (cc->io_queue)
2766 destroy_workqueue(cc->io_queue);
2767 if (cc->crypt_queue)
2768 destroy_workqueue(cc->crypt_queue);
2769
2770 crypt_free_tfms(cc);
2771
2772 bioset_exit(&cc->bs);
2773
2774 mempool_exit(&cc->page_pool);
2775 mempool_exit(&cc->req_pool);
2776 mempool_exit(&cc->tag_pool);
2777
2778 WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2779 percpu_counter_destroy(&cc->n_allocated_pages);
2780
2781 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2782 cc->iv_gen_ops->dtr(cc);
2783
2784 if (cc->dev)
2785 dm_put_device(ti, cc->dev);
2786
2787 kfree_sensitive(cc->cipher_string);
2788 kfree_sensitive(cc->key_string);
2789 kfree_sensitive(cc->cipher_auth);
2790 kfree_sensitive(cc->authenc_key);
2791
2792 mutex_destroy(&cc->bio_alloc_lock);
2793
2794 /* Must zero key material before freeing */
2795 kfree_sensitive(cc);
2796
2797 spin_lock(&dm_crypt_clients_lock);
2798 WARN_ON(!dm_crypt_clients_n);
2799 dm_crypt_clients_n--;
2800 crypt_calculate_pages_per_client();
2801 spin_unlock(&dm_crypt_clients_lock);
2802
2803 dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2804}
2805
2806static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2807{
2808 struct crypt_config *cc = ti->private;
2809
2810 if (crypt_integrity_aead(cc))
2811 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2812 else
2813 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2814
2815 if (cc->iv_size)
2816 /* at least a 64 bit sector number should fit in our buffer */
2817 cc->iv_size = max(cc->iv_size,
2818 (unsigned int)(sizeof(u64) / sizeof(u8)));
2819 else if (ivmode) {
2820 DMWARN("Selected cipher does not support IVs");
2821 ivmode = NULL;
2822 }
2823
2824 /* Choose ivmode, see comments at iv code. */
2825 if (ivmode == NULL)
2826 cc->iv_gen_ops = NULL;
2827 else if (strcmp(ivmode, "plain") == 0)
2828 cc->iv_gen_ops = &crypt_iv_plain_ops;
2829 else if (strcmp(ivmode, "plain64") == 0)
2830 cc->iv_gen_ops = &crypt_iv_plain64_ops;
2831 else if (strcmp(ivmode, "plain64be") == 0)
2832 cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2833 else if (strcmp(ivmode, "essiv") == 0)
2834 cc->iv_gen_ops = &crypt_iv_essiv_ops;
2835 else if (strcmp(ivmode, "benbi") == 0)
2836 cc->iv_gen_ops = &crypt_iv_benbi_ops;
2837 else if (strcmp(ivmode, "null") == 0)
2838 cc->iv_gen_ops = &crypt_iv_null_ops;
2839 else if (strcmp(ivmode, "eboiv") == 0)
2840 cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2841 else if (strcmp(ivmode, "elephant") == 0) {
2842 cc->iv_gen_ops = &crypt_iv_elephant_ops;
2843 cc->key_parts = 2;
2844 cc->key_extra_size = cc->key_size / 2;
2845 if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2846 return -EINVAL;
2847 set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2848 } else if (strcmp(ivmode, "lmk") == 0) {
2849 cc->iv_gen_ops = &crypt_iv_lmk_ops;
2850 /*
2851 * Version 2 and 3 is recognised according
2852 * to length of provided multi-key string.
2853 * If present (version 3), last key is used as IV seed.
2854 * All keys (including IV seed) are always the same size.
2855 */
2856 if (cc->key_size % cc->key_parts) {
2857 cc->key_parts++;
2858 cc->key_extra_size = cc->key_size / cc->key_parts;
2859 }
2860 } else if (strcmp(ivmode, "tcw") == 0) {
2861 cc->iv_gen_ops = &crypt_iv_tcw_ops;
2862 cc->key_parts += 2; /* IV + whitening */
2863 cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2864 } else if (strcmp(ivmode, "random") == 0) {
2865 cc->iv_gen_ops = &crypt_iv_random_ops;
2866 /* Need storage space in integrity fields. */
2867 cc->integrity_iv_size = cc->iv_size;
2868 } else {
2869 ti->error = "Invalid IV mode";
2870 return -EINVAL;
2871 }
2872
2873 return 0;
2874}
2875
2876/*
2877 * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2878 * The HMAC is needed to calculate tag size (HMAC digest size).
2879 * This should be probably done by crypto-api calls (once available...)
2880 */
2881static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2882{
2883 char *start, *end, *mac_alg = NULL;
2884 struct crypto_ahash *mac;
2885
2886 if (!strstarts(cipher_api, "authenc("))
2887 return 0;
2888
2889 start = strchr(cipher_api, '(');
2890 end = strchr(cipher_api, ',');
2891 if (!start || !end || ++start > end)
2892 return -EINVAL;
2893
2894 mac_alg = kmemdup_nul(start, end - start, GFP_KERNEL);
2895 if (!mac_alg)
2896 return -ENOMEM;
2897
2898 mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2899 kfree(mac_alg);
2900
2901 if (IS_ERR(mac))
2902 return PTR_ERR(mac);
2903
2904 cc->key_mac_size = crypto_ahash_digestsize(mac);
2905 crypto_free_ahash(mac);
2906
2907 cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2908 if (!cc->authenc_key)
2909 return -ENOMEM;
2910
2911 return 0;
2912}
2913
2914static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2915 char **ivmode, char **ivopts)
2916{
2917 struct crypt_config *cc = ti->private;
2918 char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2919 int ret = -EINVAL;
2920
2921 cc->tfms_count = 1;
2922
2923 /*
2924 * New format (capi: prefix)
2925 * capi:cipher_api_spec-iv:ivopts
2926 */
2927 tmp = &cipher_in[strlen("capi:")];
2928
2929 /* Separate IV options if present, it can contain another '-' in hash name */
2930 *ivopts = strrchr(tmp, ':');
2931 if (*ivopts) {
2932 **ivopts = '\0';
2933 (*ivopts)++;
2934 }
2935 /* Parse IV mode */
2936 *ivmode = strrchr(tmp, '-');
2937 if (*ivmode) {
2938 **ivmode = '\0';
2939 (*ivmode)++;
2940 }
2941 /* The rest is crypto API spec */
2942 cipher_api = tmp;
2943
2944 /* Alloc AEAD, can be used only in new format. */
2945 if (crypt_integrity_aead(cc)) {
2946 ret = crypt_ctr_auth_cipher(cc, cipher_api);
2947 if (ret < 0) {
2948 ti->error = "Invalid AEAD cipher spec";
2949 return ret;
2950 }
2951 }
2952
2953 if (*ivmode && !strcmp(*ivmode, "lmk"))
2954 cc->tfms_count = 64;
2955
2956 if (*ivmode && !strcmp(*ivmode, "essiv")) {
2957 if (!*ivopts) {
2958 ti->error = "Digest algorithm missing for ESSIV mode";
2959 return -EINVAL;
2960 }
2961 ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2962 cipher_api, *ivopts);
2963 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2964 ti->error = "Cannot allocate cipher string";
2965 return -ENOMEM;
2966 }
2967 cipher_api = buf;
2968 }
2969
2970 cc->key_parts = cc->tfms_count;
2971
2972 /* Allocate cipher */
2973 ret = crypt_alloc_tfms(cc, cipher_api);
2974 if (ret < 0) {
2975 ti->error = "Error allocating crypto tfm";
2976 return ret;
2977 }
2978
2979 if (crypt_integrity_aead(cc))
2980 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2981 else
2982 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2983
2984 return 0;
2985}
2986
2987static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2988 char **ivmode, char **ivopts)
2989{
2990 struct crypt_config *cc = ti->private;
2991 char *tmp, *cipher, *chainmode, *keycount;
2992 char *cipher_api = NULL;
2993 int ret = -EINVAL;
2994 char dummy;
2995
2996 if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2997 ti->error = "Bad cipher specification";
2998 return -EINVAL;
2999 }
3000
3001 /*
3002 * Legacy dm-crypt cipher specification
3003 * cipher[:keycount]-mode-iv:ivopts
3004 */
3005 tmp = cipher_in;
3006 keycount = strsep(&tmp, "-");
3007 cipher = strsep(&keycount, ":");
3008
3009 if (!keycount)
3010 cc->tfms_count = 1;
3011 else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
3012 !is_power_of_2(cc->tfms_count)) {
3013 ti->error = "Bad cipher key count specification";
3014 return -EINVAL;
3015 }
3016 cc->key_parts = cc->tfms_count;
3017
3018 chainmode = strsep(&tmp, "-");
3019 *ivmode = strsep(&tmp, ":");
3020 *ivopts = tmp;
3021
3022 /*
3023 * For compatibility with the original dm-crypt mapping format, if
3024 * only the cipher name is supplied, use cbc-plain.
3025 */
3026 if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
3027 chainmode = "cbc";
3028 *ivmode = "plain";
3029 }
3030
3031 if (strcmp(chainmode, "ecb") && !*ivmode) {
3032 ti->error = "IV mechanism required";
3033 return -EINVAL;
3034 }
3035
3036 cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
3037 if (!cipher_api)
3038 goto bad_mem;
3039
3040 if (*ivmode && !strcmp(*ivmode, "essiv")) {
3041 if (!*ivopts) {
3042 ti->error = "Digest algorithm missing for ESSIV mode";
3043 kfree(cipher_api);
3044 return -EINVAL;
3045 }
3046 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3047 "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
3048 } else {
3049 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3050 "%s(%s)", chainmode, cipher);
3051 }
3052 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3053 kfree(cipher_api);
3054 goto bad_mem;
3055 }
3056
3057 /* Allocate cipher */
3058 ret = crypt_alloc_tfms(cc, cipher_api);
3059 if (ret < 0) {
3060 ti->error = "Error allocating crypto tfm";
3061 kfree(cipher_api);
3062 return ret;
3063 }
3064 kfree(cipher_api);
3065
3066 return 0;
3067bad_mem:
3068 ti->error = "Cannot allocate cipher strings";
3069 return -ENOMEM;
3070}
3071
3072static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3073{
3074 struct crypt_config *cc = ti->private;
3075 char *ivmode = NULL, *ivopts = NULL;
3076 int ret;
3077
3078 cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3079 if (!cc->cipher_string) {
3080 ti->error = "Cannot allocate cipher strings";
3081 return -ENOMEM;
3082 }
3083
3084 if (strstarts(cipher_in, "capi:"))
3085 ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3086 else
3087 ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3088 if (ret)
3089 return ret;
3090
3091 /* Initialize IV */
3092 ret = crypt_ctr_ivmode(ti, ivmode);
3093 if (ret < 0)
3094 return ret;
3095
3096 /* Initialize and set key */
3097 ret = crypt_set_key(cc, key);
3098 if (ret < 0) {
3099 ti->error = "Error decoding and setting key";
3100 return ret;
3101 }
3102
3103 /* Allocate IV */
3104 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3105 ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3106 if (ret < 0) {
3107 ti->error = "Error creating IV";
3108 return ret;
3109 }
3110 }
3111
3112 /* Initialize IV (set keys for ESSIV etc) */
3113 if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3114 ret = cc->iv_gen_ops->init(cc);
3115 if (ret < 0) {
3116 ti->error = "Error initialising IV";
3117 return ret;
3118 }
3119 }
3120
3121 /* wipe the kernel key payload copy */
3122 if (cc->key_string)
3123 memset(cc->key, 0, cc->key_size * sizeof(u8));
3124
3125 return ret;
3126}
3127
3128static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3129{
3130 struct crypt_config *cc = ti->private;
3131 struct dm_arg_set as;
3132 static const struct dm_arg _args[] = {
3133 {0, 8, "Invalid number of feature args"},
3134 };
3135 unsigned int opt_params, val;
3136 const char *opt_string, *sval;
3137 char dummy;
3138 int ret;
3139
3140 /* Optional parameters */
3141 as.argc = argc;
3142 as.argv = argv;
3143
3144 ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3145 if (ret)
3146 return ret;
3147
3148 while (opt_params--) {
3149 opt_string = dm_shift_arg(&as);
3150 if (!opt_string) {
3151 ti->error = "Not enough feature arguments";
3152 return -EINVAL;
3153 }
3154
3155 if (!strcasecmp(opt_string, "allow_discards"))
3156 ti->num_discard_bios = 1;
3157
3158 else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3159 set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3160
3161 else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3162 set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3163 else if (!strcasecmp(opt_string, "no_read_workqueue"))
3164 set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3165 else if (!strcasecmp(opt_string, "no_write_workqueue"))
3166 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3167 else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3168 if (val == 0 || val > MAX_TAG_SIZE) {
3169 ti->error = "Invalid integrity arguments";
3170 return -EINVAL;
3171 }
3172 cc->on_disk_tag_size = val;
3173 sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3174 if (!strcasecmp(sval, "aead")) {
3175 set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3176 } else if (strcasecmp(sval, "none")) {
3177 ti->error = "Unknown integrity profile";
3178 return -EINVAL;
3179 }
3180
3181 cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3182 if (!cc->cipher_auth)
3183 return -ENOMEM;
3184 } else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3185 if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3186 cc->sector_size > 4096 ||
3187 (cc->sector_size & (cc->sector_size - 1))) {
3188 ti->error = "Invalid feature value for sector_size";
3189 return -EINVAL;
3190 }
3191 if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3192 ti->error = "Device size is not multiple of sector_size feature";
3193 return -EINVAL;
3194 }
3195 cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3196 } else if (!strcasecmp(opt_string, "iv_large_sectors"))
3197 set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3198 else {
3199 ti->error = "Invalid feature arguments";
3200 return -EINVAL;
3201 }
3202 }
3203
3204 return 0;
3205}
3206
3207#ifdef CONFIG_BLK_DEV_ZONED
3208static int crypt_report_zones(struct dm_target *ti,
3209 struct dm_report_zones_args *args, unsigned int nr_zones)
3210{
3211 struct crypt_config *cc = ti->private;
3212
3213 return dm_report_zones(cc->dev->bdev, cc->start,
3214 cc->start + dm_target_offset(ti, args->next_sector),
3215 args, nr_zones);
3216}
3217#else
3218#define crypt_report_zones NULL
3219#endif
3220
3221/*
3222 * Construct an encryption mapping:
3223 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3224 */
3225static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3226{
3227 struct crypt_config *cc;
3228 const char *devname = dm_table_device_name(ti->table);
3229 int key_size;
3230 unsigned int align_mask;
3231 unsigned long long tmpll;
3232 int ret;
3233 size_t iv_size_padding, additional_req_size;
3234 char dummy;
3235
3236 if (argc < 5) {
3237 ti->error = "Not enough arguments";
3238 return -EINVAL;
3239 }
3240
3241 key_size = get_key_size(&argv[1]);
3242 if (key_size < 0) {
3243 ti->error = "Cannot parse key size";
3244 return -EINVAL;
3245 }
3246
3247 cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3248 if (!cc) {
3249 ti->error = "Cannot allocate encryption context";
3250 return -ENOMEM;
3251 }
3252 cc->key_size = key_size;
3253 cc->sector_size = (1 << SECTOR_SHIFT);
3254 cc->sector_shift = 0;
3255
3256 ti->private = cc;
3257
3258 spin_lock(&dm_crypt_clients_lock);
3259 dm_crypt_clients_n++;
3260 crypt_calculate_pages_per_client();
3261 spin_unlock(&dm_crypt_clients_lock);
3262
3263 ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3264 if (ret < 0)
3265 goto bad;
3266
3267 /* Optional parameters need to be read before cipher constructor */
3268 if (argc > 5) {
3269 ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3270 if (ret)
3271 goto bad;
3272 }
3273
3274 ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3275 if (ret < 0)
3276 goto bad;
3277
3278 if (crypt_integrity_aead(cc)) {
3279 cc->dmreq_start = sizeof(struct aead_request);
3280 cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3281 align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3282 } else {
3283 cc->dmreq_start = sizeof(struct skcipher_request);
3284 cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3285 align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3286 }
3287 cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3288
3289 if (align_mask < CRYPTO_MINALIGN) {
3290 /* Allocate the padding exactly */
3291 iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3292 & align_mask;
3293 } else {
3294 /*
3295 * If the cipher requires greater alignment than kmalloc
3296 * alignment, we don't know the exact position of the
3297 * initialization vector. We must assume worst case.
3298 */
3299 iv_size_padding = align_mask;
3300 }
3301
3302 /* ...| IV + padding | original IV | original sec. number | bio tag offset | */
3303 additional_req_size = sizeof(struct dm_crypt_request) +
3304 iv_size_padding + cc->iv_size +
3305 cc->iv_size +
3306 sizeof(uint64_t) +
3307 sizeof(unsigned int);
3308
3309 ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3310 if (ret) {
3311 ti->error = "Cannot allocate crypt request mempool";
3312 goto bad;
3313 }
3314
3315 cc->per_bio_data_size = ti->per_io_data_size =
3316 ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3317 ARCH_DMA_MINALIGN);
3318
3319 ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3320 if (ret) {
3321 ti->error = "Cannot allocate page mempool";
3322 goto bad;
3323 }
3324
3325 ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3326 if (ret) {
3327 ti->error = "Cannot allocate crypt bioset";
3328 goto bad;
3329 }
3330
3331 mutex_init(&cc->bio_alloc_lock);
3332
3333 ret = -EINVAL;
3334 if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3335 (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3336 ti->error = "Invalid iv_offset sector";
3337 goto bad;
3338 }
3339 cc->iv_offset = tmpll;
3340
3341 ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3342 if (ret) {
3343 ti->error = "Device lookup failed";
3344 goto bad;
3345 }
3346
3347 ret = -EINVAL;
3348 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3349 ti->error = "Invalid device sector";
3350 goto bad;
3351 }
3352 cc->start = tmpll;
3353
3354 if (bdev_is_zoned(cc->dev->bdev)) {
3355 /*
3356 * For zoned block devices, we need to preserve the issuer write
3357 * ordering. To do so, disable write workqueues and force inline
3358 * encryption completion.
3359 */
3360 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3361 set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3362
3363 /*
3364 * All zone append writes to a zone of a zoned block device will
3365 * have the same BIO sector, the start of the zone. When the
3366 * cypher IV mode uses sector values, all data targeting a
3367 * zone will be encrypted using the first sector numbers of the
3368 * zone. This will not result in write errors but will
3369 * cause most reads to fail as reads will use the sector values
3370 * for the actual data locations, resulting in IV mismatch.
3371 * To avoid this problem, ask DM core to emulate zone append
3372 * operations with regular writes.
3373 */
3374 DMDEBUG("Zone append operations will be emulated");
3375 ti->emulate_zone_append = true;
3376 }
3377
3378 if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3379 ret = crypt_integrity_ctr(cc, ti);
3380 if (ret)
3381 goto bad;
3382
3383 cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3384 if (!cc->tag_pool_max_sectors)
3385 cc->tag_pool_max_sectors = 1;
3386
3387 ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3388 cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3389 if (ret) {
3390 ti->error = "Cannot allocate integrity tags mempool";
3391 goto bad;
3392 }
3393
3394 cc->tag_pool_max_sectors <<= cc->sector_shift;
3395 }
3396
3397 ret = -ENOMEM;
3398 cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3399 if (!cc->io_queue) {
3400 ti->error = "Couldn't create kcryptd io queue";
3401 goto bad;
3402 }
3403
3404 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3405 cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3406 1, devname);
3407 else
3408 cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3409 WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3410 num_online_cpus(), devname);
3411 if (!cc->crypt_queue) {
3412 ti->error = "Couldn't create kcryptd queue";
3413 goto bad;
3414 }
3415
3416 spin_lock_init(&cc->write_thread_lock);
3417 cc->write_tree = RB_ROOT;
3418
3419 cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3420 if (IS_ERR(cc->write_thread)) {
3421 ret = PTR_ERR(cc->write_thread);
3422 cc->write_thread = NULL;
3423 ti->error = "Couldn't spawn write thread";
3424 goto bad;
3425 }
3426
3427 ti->num_flush_bios = 1;
3428 ti->limit_swap_bios = true;
3429 ti->accounts_remapped_io = true;
3430
3431 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3432 return 0;
3433
3434bad:
3435 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3436 crypt_dtr(ti);
3437 return ret;
3438}
3439
3440static int crypt_map(struct dm_target *ti, struct bio *bio)
3441{
3442 struct dm_crypt_io *io;
3443 struct crypt_config *cc = ti->private;
3444
3445 /*
3446 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3447 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3448 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3449 */
3450 if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3451 bio_op(bio) == REQ_OP_DISCARD)) {
3452 bio_set_dev(bio, cc->dev->bdev);
3453 if (bio_sectors(bio))
3454 bio->bi_iter.bi_sector = cc->start +
3455 dm_target_offset(ti, bio->bi_iter.bi_sector);
3456 return DM_MAPIO_REMAPPED;
3457 }
3458
3459 /*
3460 * Check if bio is too large, split as needed.
3461 */
3462 if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) &&
3463 (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3464 dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT));
3465
3466 /*
3467 * Ensure that bio is a multiple of internal sector encryption size
3468 * and is aligned to this size as defined in IO hints.
3469 */
3470 if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3471 return DM_MAPIO_KILL;
3472
3473 if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3474 return DM_MAPIO_KILL;
3475
3476 io = dm_per_bio_data(bio, cc->per_bio_data_size);
3477 crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3478
3479 if (cc->on_disk_tag_size) {
3480 unsigned int tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3481
3482 if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3483 io->integrity_metadata = NULL;
3484 else
3485 io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3486
3487 if (unlikely(!io->integrity_metadata)) {
3488 if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3489 dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3490 io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3491 io->integrity_metadata_from_pool = true;
3492 }
3493 }
3494
3495 if (crypt_integrity_aead(cc))
3496 io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3497 else
3498 io->ctx.r.req = (struct skcipher_request *)(io + 1);
3499
3500 if (bio_data_dir(io->base_bio) == READ) {
3501 if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3502 kcryptd_queue_read(io);
3503 } else
3504 kcryptd_queue_crypt(io);
3505
3506 return DM_MAPIO_SUBMITTED;
3507}
3508
3509static char hex2asc(unsigned char c)
3510{
3511 return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3512}
3513
3514static void crypt_status(struct dm_target *ti, status_type_t type,
3515 unsigned int status_flags, char *result, unsigned int maxlen)
3516{
3517 struct crypt_config *cc = ti->private;
3518 unsigned int i, sz = 0;
3519 int num_feature_args = 0;
3520
3521 switch (type) {
3522 case STATUSTYPE_INFO:
3523 result[0] = '\0';
3524 break;
3525
3526 case STATUSTYPE_TABLE:
3527 DMEMIT("%s ", cc->cipher_string);
3528
3529 if (cc->key_size > 0) {
3530 if (cc->key_string)
3531 DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3532 else {
3533 for (i = 0; i < cc->key_size; i++) {
3534 DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3535 hex2asc(cc->key[i] & 0xf));
3536 }
3537 }
3538 } else
3539 DMEMIT("-");
3540
3541 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3542 cc->dev->name, (unsigned long long)cc->start);
3543
3544 num_feature_args += !!ti->num_discard_bios;
3545 num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3546 num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3547 num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3548 num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3549 num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3550 num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3551 if (cc->on_disk_tag_size)
3552 num_feature_args++;
3553 if (num_feature_args) {
3554 DMEMIT(" %d", num_feature_args);
3555 if (ti->num_discard_bios)
3556 DMEMIT(" allow_discards");
3557 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3558 DMEMIT(" same_cpu_crypt");
3559 if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3560 DMEMIT(" submit_from_crypt_cpus");
3561 if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3562 DMEMIT(" no_read_workqueue");
3563 if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3564 DMEMIT(" no_write_workqueue");
3565 if (cc->on_disk_tag_size)
3566 DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3567 if (cc->sector_size != (1 << SECTOR_SHIFT))
3568 DMEMIT(" sector_size:%d", cc->sector_size);
3569 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3570 DMEMIT(" iv_large_sectors");
3571 }
3572 break;
3573
3574 case STATUSTYPE_IMA:
3575 DMEMIT_TARGET_NAME_VERSION(ti->type);
3576 DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3577 DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3578 DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3579 'y' : 'n');
3580 DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3581 'y' : 'n');
3582 DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3583 'y' : 'n');
3584 DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3585 'y' : 'n');
3586
3587 if (cc->on_disk_tag_size)
3588 DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3589 cc->on_disk_tag_size, cc->cipher_auth);
3590 if (cc->sector_size != (1 << SECTOR_SHIFT))
3591 DMEMIT(",sector_size=%d", cc->sector_size);
3592 if (cc->cipher_string)
3593 DMEMIT(",cipher_string=%s", cc->cipher_string);
3594
3595 DMEMIT(",key_size=%u", cc->key_size);
3596 DMEMIT(",key_parts=%u", cc->key_parts);
3597 DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3598 DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3599 DMEMIT(";");
3600 break;
3601 }
3602}
3603
3604static void crypt_postsuspend(struct dm_target *ti)
3605{
3606 struct crypt_config *cc = ti->private;
3607
3608 set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3609}
3610
3611static int crypt_preresume(struct dm_target *ti)
3612{
3613 struct crypt_config *cc = ti->private;
3614
3615 if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3616 DMERR("aborting resume - crypt key is not set.");
3617 return -EAGAIN;
3618 }
3619
3620 return 0;
3621}
3622
3623static void crypt_resume(struct dm_target *ti)
3624{
3625 struct crypt_config *cc = ti->private;
3626
3627 clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3628}
3629
3630/* Message interface
3631 * key set <key>
3632 * key wipe
3633 */
3634static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3635 char *result, unsigned int maxlen)
3636{
3637 struct crypt_config *cc = ti->private;
3638 int key_size, ret = -EINVAL;
3639
3640 if (argc < 2)
3641 goto error;
3642
3643 if (!strcasecmp(argv[0], "key")) {
3644 if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3645 DMWARN("not suspended during key manipulation.");
3646 return -EINVAL;
3647 }
3648 if (argc == 3 && !strcasecmp(argv[1], "set")) {
3649 /* The key size may not be changed. */
3650 key_size = get_key_size(&argv[2]);
3651 if (key_size < 0 || cc->key_size != key_size) {
3652 memset(argv[2], '0', strlen(argv[2]));
3653 return -EINVAL;
3654 }
3655
3656 ret = crypt_set_key(cc, argv[2]);
3657 if (ret)
3658 return ret;
3659 if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3660 ret = cc->iv_gen_ops->init(cc);
3661 /* wipe the kernel key payload copy */
3662 if (cc->key_string)
3663 memset(cc->key, 0, cc->key_size * sizeof(u8));
3664 return ret;
3665 }
3666 if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3667 return crypt_wipe_key(cc);
3668 }
3669
3670error:
3671 DMWARN("unrecognised message received.");
3672 return -EINVAL;
3673}
3674
3675static int crypt_iterate_devices(struct dm_target *ti,
3676 iterate_devices_callout_fn fn, void *data)
3677{
3678 struct crypt_config *cc = ti->private;
3679
3680 return fn(ti, cc->dev, cc->start, ti->len, data);
3681}
3682
3683static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3684{
3685 struct crypt_config *cc = ti->private;
3686
3687 /*
3688 * Unfortunate constraint that is required to avoid the potential
3689 * for exceeding underlying device's max_segments limits -- due to
3690 * crypt_alloc_buffer() possibly allocating pages for the encryption
3691 * bio that are not as physically contiguous as the original bio.
3692 */
3693 limits->max_segment_size = PAGE_SIZE;
3694
3695 limits->logical_block_size =
3696 max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3697 limits->physical_block_size =
3698 max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3699 limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3700 limits->dma_alignment = limits->logical_block_size - 1;
3701}
3702
3703static struct target_type crypt_target = {
3704 .name = "crypt",
3705 .version = {1, 25, 0},
3706 .module = THIS_MODULE,
3707 .ctr = crypt_ctr,
3708 .dtr = crypt_dtr,
3709 .features = DM_TARGET_ZONED_HM,
3710 .report_zones = crypt_report_zones,
3711 .map = crypt_map,
3712 .status = crypt_status,
3713 .postsuspend = crypt_postsuspend,
3714 .preresume = crypt_preresume,
3715 .resume = crypt_resume,
3716 .message = crypt_message,
3717 .iterate_devices = crypt_iterate_devices,
3718 .io_hints = crypt_io_hints,
3719};
3720module_dm(crypt);
3721
3722MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3723MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3724MODULE_LICENSE("GPL");
1/*
2 * Copyright (C) 2003 Jana Saout <jana@saout.de>
3 * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
4 * Copyright (C) 2006-2015 Red Hat, Inc. All rights reserved.
5 * Copyright (C) 2013 Milan Broz <gmazyland@gmail.com>
6 *
7 * This file is released under the GPL.
8 */
9
10#include <linux/completion.h>
11#include <linux/err.h>
12#include <linux/module.h>
13#include <linux/init.h>
14#include <linux/kernel.h>
15#include <linux/bio.h>
16#include <linux/blkdev.h>
17#include <linux/mempool.h>
18#include <linux/slab.h>
19#include <linux/crypto.h>
20#include <linux/workqueue.h>
21#include <linux/kthread.h>
22#include <linux/backing-dev.h>
23#include <linux/atomic.h>
24#include <linux/scatterlist.h>
25#include <linux/rbtree.h>
26#include <asm/page.h>
27#include <asm/unaligned.h>
28#include <crypto/hash.h>
29#include <crypto/md5.h>
30#include <crypto/algapi.h>
31#include <crypto/skcipher.h>
32
33#include <linux/device-mapper.h>
34
35#define DM_MSG_PREFIX "crypt"
36
37/*
38 * context holding the current state of a multi-part conversion
39 */
40struct convert_context {
41 struct completion restart;
42 struct bio *bio_in;
43 struct bio *bio_out;
44 struct bvec_iter iter_in;
45 struct bvec_iter iter_out;
46 sector_t cc_sector;
47 atomic_t cc_pending;
48 struct skcipher_request *req;
49};
50
51/*
52 * per bio private data
53 */
54struct dm_crypt_io {
55 struct crypt_config *cc;
56 struct bio *base_bio;
57 struct work_struct work;
58
59 struct convert_context ctx;
60
61 atomic_t io_pending;
62 int error;
63 sector_t sector;
64
65 struct rb_node rb_node;
66} CRYPTO_MINALIGN_ATTR;
67
68struct dm_crypt_request {
69 struct convert_context *ctx;
70 struct scatterlist sg_in;
71 struct scatterlist sg_out;
72 sector_t iv_sector;
73};
74
75struct crypt_config;
76
77struct crypt_iv_operations {
78 int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
79 const char *opts);
80 void (*dtr)(struct crypt_config *cc);
81 int (*init)(struct crypt_config *cc);
82 int (*wipe)(struct crypt_config *cc);
83 int (*generator)(struct crypt_config *cc, u8 *iv,
84 struct dm_crypt_request *dmreq);
85 int (*post)(struct crypt_config *cc, u8 *iv,
86 struct dm_crypt_request *dmreq);
87};
88
89struct iv_essiv_private {
90 struct crypto_ahash *hash_tfm;
91 u8 *salt;
92};
93
94struct iv_benbi_private {
95 int shift;
96};
97
98#define LMK_SEED_SIZE 64 /* hash + 0 */
99struct iv_lmk_private {
100 struct crypto_shash *hash_tfm;
101 u8 *seed;
102};
103
104#define TCW_WHITENING_SIZE 16
105struct iv_tcw_private {
106 struct crypto_shash *crc32_tfm;
107 u8 *iv_seed;
108 u8 *whitening;
109};
110
111/*
112 * Crypt: maps a linear range of a block device
113 * and encrypts / decrypts at the same time.
114 */
115enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
116 DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
117 DM_CRYPT_EXIT_THREAD};
118
119/*
120 * The fields in here must be read only after initialization.
121 */
122struct crypt_config {
123 struct dm_dev *dev;
124 sector_t start;
125
126 /*
127 * pool for per bio private data, crypto requests and
128 * encryption requeusts/buffer pages
129 */
130 mempool_t *req_pool;
131 mempool_t *page_pool;
132 struct bio_set *bs;
133 struct mutex bio_alloc_lock;
134
135 struct workqueue_struct *io_queue;
136 struct workqueue_struct *crypt_queue;
137
138 struct task_struct *write_thread;
139 wait_queue_head_t write_thread_wait;
140 struct rb_root write_tree;
141
142 char *cipher;
143 char *cipher_string;
144
145 struct crypt_iv_operations *iv_gen_ops;
146 union {
147 struct iv_essiv_private essiv;
148 struct iv_benbi_private benbi;
149 struct iv_lmk_private lmk;
150 struct iv_tcw_private tcw;
151 } iv_gen_private;
152 sector_t iv_offset;
153 unsigned int iv_size;
154
155 /* ESSIV: struct crypto_cipher *essiv_tfm */
156 void *iv_private;
157 struct crypto_skcipher **tfms;
158 unsigned tfms_count;
159
160 /*
161 * Layout of each crypto request:
162 *
163 * struct skcipher_request
164 * context
165 * padding
166 * struct dm_crypt_request
167 * padding
168 * IV
169 *
170 * The padding is added so that dm_crypt_request and the IV are
171 * correctly aligned.
172 */
173 unsigned int dmreq_start;
174
175 unsigned int per_bio_data_size;
176
177 unsigned long flags;
178 unsigned int key_size;
179 unsigned int key_parts; /* independent parts in key buffer */
180 unsigned int key_extra_size; /* additional keys length */
181 u8 key[0];
182};
183
184#define MIN_IOS 16
185
186static void clone_init(struct dm_crypt_io *, struct bio *);
187static void kcryptd_queue_crypt(struct dm_crypt_io *io);
188static u8 *iv_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq);
189
190/*
191 * Use this to access cipher attributes that are the same for each CPU.
192 */
193static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
194{
195 return cc->tfms[0];
196}
197
198/*
199 * Different IV generation algorithms:
200 *
201 * plain: the initial vector is the 32-bit little-endian version of the sector
202 * number, padded with zeros if necessary.
203 *
204 * plain64: the initial vector is the 64-bit little-endian version of the sector
205 * number, padded with zeros if necessary.
206 *
207 * essiv: "encrypted sector|salt initial vector", the sector number is
208 * encrypted with the bulk cipher using a salt as key. The salt
209 * should be derived from the bulk cipher's key via hashing.
210 *
211 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
212 * (needed for LRW-32-AES and possible other narrow block modes)
213 *
214 * null: the initial vector is always zero. Provides compatibility with
215 * obsolete loop_fish2 devices. Do not use for new devices.
216 *
217 * lmk: Compatible implementation of the block chaining mode used
218 * by the Loop-AES block device encryption system
219 * designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
220 * It operates on full 512 byte sectors and uses CBC
221 * with an IV derived from the sector number, the data and
222 * optionally extra IV seed.
223 * This means that after decryption the first block
224 * of sector must be tweaked according to decrypted data.
225 * Loop-AES can use three encryption schemes:
226 * version 1: is plain aes-cbc mode
227 * version 2: uses 64 multikey scheme with lmk IV generator
228 * version 3: the same as version 2 with additional IV seed
229 * (it uses 65 keys, last key is used as IV seed)
230 *
231 * tcw: Compatible implementation of the block chaining mode used
232 * by the TrueCrypt device encryption system (prior to version 4.1).
233 * For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
234 * It operates on full 512 byte sectors and uses CBC
235 * with an IV derived from initial key and the sector number.
236 * In addition, whitening value is applied on every sector, whitening
237 * is calculated from initial key, sector number and mixed using CRC32.
238 * Note that this encryption scheme is vulnerable to watermarking attacks
239 * and should be used for old compatible containers access only.
240 *
241 * plumb: unimplemented, see:
242 * http://article.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt/454
243 */
244
245static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
246 struct dm_crypt_request *dmreq)
247{
248 memset(iv, 0, cc->iv_size);
249 *(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
250
251 return 0;
252}
253
254static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
255 struct dm_crypt_request *dmreq)
256{
257 memset(iv, 0, cc->iv_size);
258 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
259
260 return 0;
261}
262
263/* Initialise ESSIV - compute salt but no local memory allocations */
264static int crypt_iv_essiv_init(struct crypt_config *cc)
265{
266 struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv;
267 AHASH_REQUEST_ON_STACK(req, essiv->hash_tfm);
268 struct scatterlist sg;
269 struct crypto_cipher *essiv_tfm;
270 int err;
271
272 sg_init_one(&sg, cc->key, cc->key_size);
273 ahash_request_set_tfm(req, essiv->hash_tfm);
274 ahash_request_set_callback(req, CRYPTO_TFM_REQ_MAY_SLEEP, NULL, NULL);
275 ahash_request_set_crypt(req, &sg, essiv->salt, cc->key_size);
276
277 err = crypto_ahash_digest(req);
278 ahash_request_zero(req);
279 if (err)
280 return err;
281
282 essiv_tfm = cc->iv_private;
283
284 err = crypto_cipher_setkey(essiv_tfm, essiv->salt,
285 crypto_ahash_digestsize(essiv->hash_tfm));
286 if (err)
287 return err;
288
289 return 0;
290}
291
292/* Wipe salt and reset key derived from volume key */
293static int crypt_iv_essiv_wipe(struct crypt_config *cc)
294{
295 struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv;
296 unsigned salt_size = crypto_ahash_digestsize(essiv->hash_tfm);
297 struct crypto_cipher *essiv_tfm;
298 int r, err = 0;
299
300 memset(essiv->salt, 0, salt_size);
301
302 essiv_tfm = cc->iv_private;
303 r = crypto_cipher_setkey(essiv_tfm, essiv->salt, salt_size);
304 if (r)
305 err = r;
306
307 return err;
308}
309
310/* Set up per cpu cipher state */
311static struct crypto_cipher *setup_essiv_cpu(struct crypt_config *cc,
312 struct dm_target *ti,
313 u8 *salt, unsigned saltsize)
314{
315 struct crypto_cipher *essiv_tfm;
316 int err;
317
318 /* Setup the essiv_tfm with the given salt */
319 essiv_tfm = crypto_alloc_cipher(cc->cipher, 0, CRYPTO_ALG_ASYNC);
320 if (IS_ERR(essiv_tfm)) {
321 ti->error = "Error allocating crypto tfm for ESSIV";
322 return essiv_tfm;
323 }
324
325 if (crypto_cipher_blocksize(essiv_tfm) !=
326 crypto_skcipher_ivsize(any_tfm(cc))) {
327 ti->error = "Block size of ESSIV cipher does "
328 "not match IV size of block cipher";
329 crypto_free_cipher(essiv_tfm);
330 return ERR_PTR(-EINVAL);
331 }
332
333 err = crypto_cipher_setkey(essiv_tfm, salt, saltsize);
334 if (err) {
335 ti->error = "Failed to set key for ESSIV cipher";
336 crypto_free_cipher(essiv_tfm);
337 return ERR_PTR(err);
338 }
339
340 return essiv_tfm;
341}
342
343static void crypt_iv_essiv_dtr(struct crypt_config *cc)
344{
345 struct crypto_cipher *essiv_tfm;
346 struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv;
347
348 crypto_free_ahash(essiv->hash_tfm);
349 essiv->hash_tfm = NULL;
350
351 kzfree(essiv->salt);
352 essiv->salt = NULL;
353
354 essiv_tfm = cc->iv_private;
355
356 if (essiv_tfm)
357 crypto_free_cipher(essiv_tfm);
358
359 cc->iv_private = NULL;
360}
361
362static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti,
363 const char *opts)
364{
365 struct crypto_cipher *essiv_tfm = NULL;
366 struct crypto_ahash *hash_tfm = NULL;
367 u8 *salt = NULL;
368 int err;
369
370 if (!opts) {
371 ti->error = "Digest algorithm missing for ESSIV mode";
372 return -EINVAL;
373 }
374
375 /* Allocate hash algorithm */
376 hash_tfm = crypto_alloc_ahash(opts, 0, CRYPTO_ALG_ASYNC);
377 if (IS_ERR(hash_tfm)) {
378 ti->error = "Error initializing ESSIV hash";
379 err = PTR_ERR(hash_tfm);
380 goto bad;
381 }
382
383 salt = kzalloc(crypto_ahash_digestsize(hash_tfm), GFP_KERNEL);
384 if (!salt) {
385 ti->error = "Error kmallocing salt storage in ESSIV";
386 err = -ENOMEM;
387 goto bad;
388 }
389
390 cc->iv_gen_private.essiv.salt = salt;
391 cc->iv_gen_private.essiv.hash_tfm = hash_tfm;
392
393 essiv_tfm = setup_essiv_cpu(cc, ti, salt,
394 crypto_ahash_digestsize(hash_tfm));
395 if (IS_ERR(essiv_tfm)) {
396 crypt_iv_essiv_dtr(cc);
397 return PTR_ERR(essiv_tfm);
398 }
399 cc->iv_private = essiv_tfm;
400
401 return 0;
402
403bad:
404 if (hash_tfm && !IS_ERR(hash_tfm))
405 crypto_free_ahash(hash_tfm);
406 kfree(salt);
407 return err;
408}
409
410static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
411 struct dm_crypt_request *dmreq)
412{
413 struct crypto_cipher *essiv_tfm = cc->iv_private;
414
415 memset(iv, 0, cc->iv_size);
416 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
417 crypto_cipher_encrypt_one(essiv_tfm, iv, iv);
418
419 return 0;
420}
421
422static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
423 const char *opts)
424{
425 unsigned bs = crypto_skcipher_blocksize(any_tfm(cc));
426 int log = ilog2(bs);
427
428 /* we need to calculate how far we must shift the sector count
429 * to get the cipher block count, we use this shift in _gen */
430
431 if (1 << log != bs) {
432 ti->error = "cypher blocksize is not a power of 2";
433 return -EINVAL;
434 }
435
436 if (log > 9) {
437 ti->error = "cypher blocksize is > 512";
438 return -EINVAL;
439 }
440
441 cc->iv_gen_private.benbi.shift = 9 - log;
442
443 return 0;
444}
445
446static void crypt_iv_benbi_dtr(struct crypt_config *cc)
447{
448}
449
450static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
451 struct dm_crypt_request *dmreq)
452{
453 __be64 val;
454
455 memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
456
457 val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
458 put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
459
460 return 0;
461}
462
463static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
464 struct dm_crypt_request *dmreq)
465{
466 memset(iv, 0, cc->iv_size);
467
468 return 0;
469}
470
471static void crypt_iv_lmk_dtr(struct crypt_config *cc)
472{
473 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
474
475 if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
476 crypto_free_shash(lmk->hash_tfm);
477 lmk->hash_tfm = NULL;
478
479 kzfree(lmk->seed);
480 lmk->seed = NULL;
481}
482
483static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
484 const char *opts)
485{
486 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
487
488 lmk->hash_tfm = crypto_alloc_shash("md5", 0, 0);
489 if (IS_ERR(lmk->hash_tfm)) {
490 ti->error = "Error initializing LMK hash";
491 return PTR_ERR(lmk->hash_tfm);
492 }
493
494 /* No seed in LMK version 2 */
495 if (cc->key_parts == cc->tfms_count) {
496 lmk->seed = NULL;
497 return 0;
498 }
499
500 lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
501 if (!lmk->seed) {
502 crypt_iv_lmk_dtr(cc);
503 ti->error = "Error kmallocing seed storage in LMK";
504 return -ENOMEM;
505 }
506
507 return 0;
508}
509
510static int crypt_iv_lmk_init(struct crypt_config *cc)
511{
512 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
513 int subkey_size = cc->key_size / cc->key_parts;
514
515 /* LMK seed is on the position of LMK_KEYS + 1 key */
516 if (lmk->seed)
517 memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
518 crypto_shash_digestsize(lmk->hash_tfm));
519
520 return 0;
521}
522
523static int crypt_iv_lmk_wipe(struct crypt_config *cc)
524{
525 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
526
527 if (lmk->seed)
528 memset(lmk->seed, 0, LMK_SEED_SIZE);
529
530 return 0;
531}
532
533static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
534 struct dm_crypt_request *dmreq,
535 u8 *data)
536{
537 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
538 SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
539 struct md5_state md5state;
540 __le32 buf[4];
541 int i, r;
542
543 desc->tfm = lmk->hash_tfm;
544 desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
545
546 r = crypto_shash_init(desc);
547 if (r)
548 return r;
549
550 if (lmk->seed) {
551 r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
552 if (r)
553 return r;
554 }
555
556 /* Sector is always 512B, block size 16, add data of blocks 1-31 */
557 r = crypto_shash_update(desc, data + 16, 16 * 31);
558 if (r)
559 return r;
560
561 /* Sector is cropped to 56 bits here */
562 buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
563 buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
564 buf[2] = cpu_to_le32(4024);
565 buf[3] = 0;
566 r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
567 if (r)
568 return r;
569
570 /* No MD5 padding here */
571 r = crypto_shash_export(desc, &md5state);
572 if (r)
573 return r;
574
575 for (i = 0; i < MD5_HASH_WORDS; i++)
576 __cpu_to_le32s(&md5state.hash[i]);
577 memcpy(iv, &md5state.hash, cc->iv_size);
578
579 return 0;
580}
581
582static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
583 struct dm_crypt_request *dmreq)
584{
585 u8 *src;
586 int r = 0;
587
588 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
589 src = kmap_atomic(sg_page(&dmreq->sg_in));
590 r = crypt_iv_lmk_one(cc, iv, dmreq, src + dmreq->sg_in.offset);
591 kunmap_atomic(src);
592 } else
593 memset(iv, 0, cc->iv_size);
594
595 return r;
596}
597
598static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
599 struct dm_crypt_request *dmreq)
600{
601 u8 *dst;
602 int r;
603
604 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
605 return 0;
606
607 dst = kmap_atomic(sg_page(&dmreq->sg_out));
608 r = crypt_iv_lmk_one(cc, iv, dmreq, dst + dmreq->sg_out.offset);
609
610 /* Tweak the first block of plaintext sector */
611 if (!r)
612 crypto_xor(dst + dmreq->sg_out.offset, iv, cc->iv_size);
613
614 kunmap_atomic(dst);
615 return r;
616}
617
618static void crypt_iv_tcw_dtr(struct crypt_config *cc)
619{
620 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
621
622 kzfree(tcw->iv_seed);
623 tcw->iv_seed = NULL;
624 kzfree(tcw->whitening);
625 tcw->whitening = NULL;
626
627 if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
628 crypto_free_shash(tcw->crc32_tfm);
629 tcw->crc32_tfm = NULL;
630}
631
632static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
633 const char *opts)
634{
635 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
636
637 if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
638 ti->error = "Wrong key size for TCW";
639 return -EINVAL;
640 }
641
642 tcw->crc32_tfm = crypto_alloc_shash("crc32", 0, 0);
643 if (IS_ERR(tcw->crc32_tfm)) {
644 ti->error = "Error initializing CRC32 in TCW";
645 return PTR_ERR(tcw->crc32_tfm);
646 }
647
648 tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
649 tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
650 if (!tcw->iv_seed || !tcw->whitening) {
651 crypt_iv_tcw_dtr(cc);
652 ti->error = "Error allocating seed storage in TCW";
653 return -ENOMEM;
654 }
655
656 return 0;
657}
658
659static int crypt_iv_tcw_init(struct crypt_config *cc)
660{
661 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
662 int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
663
664 memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
665 memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
666 TCW_WHITENING_SIZE);
667
668 return 0;
669}
670
671static int crypt_iv_tcw_wipe(struct crypt_config *cc)
672{
673 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
674
675 memset(tcw->iv_seed, 0, cc->iv_size);
676 memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
677
678 return 0;
679}
680
681static int crypt_iv_tcw_whitening(struct crypt_config *cc,
682 struct dm_crypt_request *dmreq,
683 u8 *data)
684{
685 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
686 u64 sector = cpu_to_le64((u64)dmreq->iv_sector);
687 u8 buf[TCW_WHITENING_SIZE];
688 SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
689 int i, r;
690
691 /* xor whitening with sector number */
692 memcpy(buf, tcw->whitening, TCW_WHITENING_SIZE);
693 crypto_xor(buf, (u8 *)§or, 8);
694 crypto_xor(&buf[8], (u8 *)§or, 8);
695
696 /* calculate crc32 for every 32bit part and xor it */
697 desc->tfm = tcw->crc32_tfm;
698 desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
699 for (i = 0; i < 4; i++) {
700 r = crypto_shash_init(desc);
701 if (r)
702 goto out;
703 r = crypto_shash_update(desc, &buf[i * 4], 4);
704 if (r)
705 goto out;
706 r = crypto_shash_final(desc, &buf[i * 4]);
707 if (r)
708 goto out;
709 }
710 crypto_xor(&buf[0], &buf[12], 4);
711 crypto_xor(&buf[4], &buf[8], 4);
712
713 /* apply whitening (8 bytes) to whole sector */
714 for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
715 crypto_xor(data + i * 8, buf, 8);
716out:
717 memzero_explicit(buf, sizeof(buf));
718 return r;
719}
720
721static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
722 struct dm_crypt_request *dmreq)
723{
724 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
725 u64 sector = cpu_to_le64((u64)dmreq->iv_sector);
726 u8 *src;
727 int r = 0;
728
729 /* Remove whitening from ciphertext */
730 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
731 src = kmap_atomic(sg_page(&dmreq->sg_in));
732 r = crypt_iv_tcw_whitening(cc, dmreq, src + dmreq->sg_in.offset);
733 kunmap_atomic(src);
734 }
735
736 /* Calculate IV */
737 memcpy(iv, tcw->iv_seed, cc->iv_size);
738 crypto_xor(iv, (u8 *)§or, 8);
739 if (cc->iv_size > 8)
740 crypto_xor(&iv[8], (u8 *)§or, cc->iv_size - 8);
741
742 return r;
743}
744
745static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
746 struct dm_crypt_request *dmreq)
747{
748 u8 *dst;
749 int r;
750
751 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
752 return 0;
753
754 /* Apply whitening on ciphertext */
755 dst = kmap_atomic(sg_page(&dmreq->sg_out));
756 r = crypt_iv_tcw_whitening(cc, dmreq, dst + dmreq->sg_out.offset);
757 kunmap_atomic(dst);
758
759 return r;
760}
761
762static struct crypt_iv_operations crypt_iv_plain_ops = {
763 .generator = crypt_iv_plain_gen
764};
765
766static struct crypt_iv_operations crypt_iv_plain64_ops = {
767 .generator = crypt_iv_plain64_gen
768};
769
770static struct crypt_iv_operations crypt_iv_essiv_ops = {
771 .ctr = crypt_iv_essiv_ctr,
772 .dtr = crypt_iv_essiv_dtr,
773 .init = crypt_iv_essiv_init,
774 .wipe = crypt_iv_essiv_wipe,
775 .generator = crypt_iv_essiv_gen
776};
777
778static struct crypt_iv_operations crypt_iv_benbi_ops = {
779 .ctr = crypt_iv_benbi_ctr,
780 .dtr = crypt_iv_benbi_dtr,
781 .generator = crypt_iv_benbi_gen
782};
783
784static struct crypt_iv_operations crypt_iv_null_ops = {
785 .generator = crypt_iv_null_gen
786};
787
788static struct crypt_iv_operations crypt_iv_lmk_ops = {
789 .ctr = crypt_iv_lmk_ctr,
790 .dtr = crypt_iv_lmk_dtr,
791 .init = crypt_iv_lmk_init,
792 .wipe = crypt_iv_lmk_wipe,
793 .generator = crypt_iv_lmk_gen,
794 .post = crypt_iv_lmk_post
795};
796
797static struct crypt_iv_operations crypt_iv_tcw_ops = {
798 .ctr = crypt_iv_tcw_ctr,
799 .dtr = crypt_iv_tcw_dtr,
800 .init = crypt_iv_tcw_init,
801 .wipe = crypt_iv_tcw_wipe,
802 .generator = crypt_iv_tcw_gen,
803 .post = crypt_iv_tcw_post
804};
805
806static void crypt_convert_init(struct crypt_config *cc,
807 struct convert_context *ctx,
808 struct bio *bio_out, struct bio *bio_in,
809 sector_t sector)
810{
811 ctx->bio_in = bio_in;
812 ctx->bio_out = bio_out;
813 if (bio_in)
814 ctx->iter_in = bio_in->bi_iter;
815 if (bio_out)
816 ctx->iter_out = bio_out->bi_iter;
817 ctx->cc_sector = sector + cc->iv_offset;
818 init_completion(&ctx->restart);
819}
820
821static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
822 struct skcipher_request *req)
823{
824 return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
825}
826
827static struct skcipher_request *req_of_dmreq(struct crypt_config *cc,
828 struct dm_crypt_request *dmreq)
829{
830 return (struct skcipher_request *)((char *)dmreq - cc->dmreq_start);
831}
832
833static u8 *iv_of_dmreq(struct crypt_config *cc,
834 struct dm_crypt_request *dmreq)
835{
836 return (u8 *)ALIGN((unsigned long)(dmreq + 1),
837 crypto_skcipher_alignmask(any_tfm(cc)) + 1);
838}
839
840static int crypt_convert_block(struct crypt_config *cc,
841 struct convert_context *ctx,
842 struct skcipher_request *req)
843{
844 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
845 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
846 struct dm_crypt_request *dmreq;
847 u8 *iv;
848 int r;
849
850 dmreq = dmreq_of_req(cc, req);
851 iv = iv_of_dmreq(cc, dmreq);
852
853 dmreq->iv_sector = ctx->cc_sector;
854 dmreq->ctx = ctx;
855 sg_init_table(&dmreq->sg_in, 1);
856 sg_set_page(&dmreq->sg_in, bv_in.bv_page, 1 << SECTOR_SHIFT,
857 bv_in.bv_offset);
858
859 sg_init_table(&dmreq->sg_out, 1);
860 sg_set_page(&dmreq->sg_out, bv_out.bv_page, 1 << SECTOR_SHIFT,
861 bv_out.bv_offset);
862
863 bio_advance_iter(ctx->bio_in, &ctx->iter_in, 1 << SECTOR_SHIFT);
864 bio_advance_iter(ctx->bio_out, &ctx->iter_out, 1 << SECTOR_SHIFT);
865
866 if (cc->iv_gen_ops) {
867 r = cc->iv_gen_ops->generator(cc, iv, dmreq);
868 if (r < 0)
869 return r;
870 }
871
872 skcipher_request_set_crypt(req, &dmreq->sg_in, &dmreq->sg_out,
873 1 << SECTOR_SHIFT, iv);
874
875 if (bio_data_dir(ctx->bio_in) == WRITE)
876 r = crypto_skcipher_encrypt(req);
877 else
878 r = crypto_skcipher_decrypt(req);
879
880 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
881 r = cc->iv_gen_ops->post(cc, iv, dmreq);
882
883 return r;
884}
885
886static void kcryptd_async_done(struct crypto_async_request *async_req,
887 int error);
888
889static void crypt_alloc_req(struct crypt_config *cc,
890 struct convert_context *ctx)
891{
892 unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1);
893
894 if (!ctx->req)
895 ctx->req = mempool_alloc(cc->req_pool, GFP_NOIO);
896
897 skcipher_request_set_tfm(ctx->req, cc->tfms[key_index]);
898
899 /*
900 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
901 * requests if driver request queue is full.
902 */
903 skcipher_request_set_callback(ctx->req,
904 CRYPTO_TFM_REQ_MAY_BACKLOG | CRYPTO_TFM_REQ_MAY_SLEEP,
905 kcryptd_async_done, dmreq_of_req(cc, ctx->req));
906}
907
908static void crypt_free_req(struct crypt_config *cc,
909 struct skcipher_request *req, struct bio *base_bio)
910{
911 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
912
913 if ((struct skcipher_request *)(io + 1) != req)
914 mempool_free(req, cc->req_pool);
915}
916
917/*
918 * Encrypt / decrypt data from one bio to another one (can be the same one)
919 */
920static int crypt_convert(struct crypt_config *cc,
921 struct convert_context *ctx)
922{
923 int r;
924
925 atomic_set(&ctx->cc_pending, 1);
926
927 while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
928
929 crypt_alloc_req(cc, ctx);
930
931 atomic_inc(&ctx->cc_pending);
932
933 r = crypt_convert_block(cc, ctx, ctx->req);
934
935 switch (r) {
936 /*
937 * The request was queued by a crypto driver
938 * but the driver request queue is full, let's wait.
939 */
940 case -EBUSY:
941 wait_for_completion(&ctx->restart);
942 reinit_completion(&ctx->restart);
943 /* fall through */
944 /*
945 * The request is queued and processed asynchronously,
946 * completion function kcryptd_async_done() will be called.
947 */
948 case -EINPROGRESS:
949 ctx->req = NULL;
950 ctx->cc_sector++;
951 continue;
952 /*
953 * The request was already processed (synchronously).
954 */
955 case 0:
956 atomic_dec(&ctx->cc_pending);
957 ctx->cc_sector++;
958 cond_resched();
959 continue;
960
961 /* There was an error while processing the request. */
962 default:
963 atomic_dec(&ctx->cc_pending);
964 return r;
965 }
966 }
967
968 return 0;
969}
970
971static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
972
973/*
974 * Generate a new unfragmented bio with the given size
975 * This should never violate the device limitations (but only because
976 * max_segment_size is being constrained to PAGE_SIZE).
977 *
978 * This function may be called concurrently. If we allocate from the mempool
979 * concurrently, there is a possibility of deadlock. For example, if we have
980 * mempool of 256 pages, two processes, each wanting 256, pages allocate from
981 * the mempool concurrently, it may deadlock in a situation where both processes
982 * have allocated 128 pages and the mempool is exhausted.
983 *
984 * In order to avoid this scenario we allocate the pages under a mutex.
985 *
986 * In order to not degrade performance with excessive locking, we try
987 * non-blocking allocations without a mutex first but on failure we fallback
988 * to blocking allocations with a mutex.
989 */
990static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size)
991{
992 struct crypt_config *cc = io->cc;
993 struct bio *clone;
994 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
995 gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
996 unsigned i, len, remaining_size;
997 struct page *page;
998 struct bio_vec *bvec;
999
1000retry:
1001 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1002 mutex_lock(&cc->bio_alloc_lock);
1003
1004 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, cc->bs);
1005 if (!clone)
1006 goto return_clone;
1007
1008 clone_init(io, clone);
1009
1010 remaining_size = size;
1011
1012 for (i = 0; i < nr_iovecs; i++) {
1013 page = mempool_alloc(cc->page_pool, gfp_mask);
1014 if (!page) {
1015 crypt_free_buffer_pages(cc, clone);
1016 bio_put(clone);
1017 gfp_mask |= __GFP_DIRECT_RECLAIM;
1018 goto retry;
1019 }
1020
1021 len = (remaining_size > PAGE_SIZE) ? PAGE_SIZE : remaining_size;
1022
1023 bvec = &clone->bi_io_vec[clone->bi_vcnt++];
1024 bvec->bv_page = page;
1025 bvec->bv_len = len;
1026 bvec->bv_offset = 0;
1027
1028 clone->bi_iter.bi_size += len;
1029
1030 remaining_size -= len;
1031 }
1032
1033return_clone:
1034 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1035 mutex_unlock(&cc->bio_alloc_lock);
1036
1037 return clone;
1038}
1039
1040static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1041{
1042 unsigned int i;
1043 struct bio_vec *bv;
1044
1045 bio_for_each_segment_all(bv, clone, i) {
1046 BUG_ON(!bv->bv_page);
1047 mempool_free(bv->bv_page, cc->page_pool);
1048 bv->bv_page = NULL;
1049 }
1050}
1051
1052static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1053 struct bio *bio, sector_t sector)
1054{
1055 io->cc = cc;
1056 io->base_bio = bio;
1057 io->sector = sector;
1058 io->error = 0;
1059 io->ctx.req = NULL;
1060 atomic_set(&io->io_pending, 0);
1061}
1062
1063static void crypt_inc_pending(struct dm_crypt_io *io)
1064{
1065 atomic_inc(&io->io_pending);
1066}
1067
1068/*
1069 * One of the bios was finished. Check for completion of
1070 * the whole request and correctly clean up the buffer.
1071 */
1072static void crypt_dec_pending(struct dm_crypt_io *io)
1073{
1074 struct crypt_config *cc = io->cc;
1075 struct bio *base_bio = io->base_bio;
1076 int error = io->error;
1077
1078 if (!atomic_dec_and_test(&io->io_pending))
1079 return;
1080
1081 if (io->ctx.req)
1082 crypt_free_req(cc, io->ctx.req, base_bio);
1083
1084 base_bio->bi_error = error;
1085 bio_endio(base_bio);
1086}
1087
1088/*
1089 * kcryptd/kcryptd_io:
1090 *
1091 * Needed because it would be very unwise to do decryption in an
1092 * interrupt context.
1093 *
1094 * kcryptd performs the actual encryption or decryption.
1095 *
1096 * kcryptd_io performs the IO submission.
1097 *
1098 * They must be separated as otherwise the final stages could be
1099 * starved by new requests which can block in the first stages due
1100 * to memory allocation.
1101 *
1102 * The work is done per CPU global for all dm-crypt instances.
1103 * They should not depend on each other and do not block.
1104 */
1105static void crypt_endio(struct bio *clone)
1106{
1107 struct dm_crypt_io *io = clone->bi_private;
1108 struct crypt_config *cc = io->cc;
1109 unsigned rw = bio_data_dir(clone);
1110 int error;
1111
1112 /*
1113 * free the processed pages
1114 */
1115 if (rw == WRITE)
1116 crypt_free_buffer_pages(cc, clone);
1117
1118 error = clone->bi_error;
1119 bio_put(clone);
1120
1121 if (rw == READ && !error) {
1122 kcryptd_queue_crypt(io);
1123 return;
1124 }
1125
1126 if (unlikely(error))
1127 io->error = error;
1128
1129 crypt_dec_pending(io);
1130}
1131
1132static void clone_init(struct dm_crypt_io *io, struct bio *clone)
1133{
1134 struct crypt_config *cc = io->cc;
1135
1136 clone->bi_private = io;
1137 clone->bi_end_io = crypt_endio;
1138 clone->bi_bdev = cc->dev->bdev;
1139 clone->bi_rw = io->base_bio->bi_rw;
1140}
1141
1142static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1143{
1144 struct crypt_config *cc = io->cc;
1145 struct bio *clone;
1146
1147 /*
1148 * We need the original biovec array in order to decrypt
1149 * the whole bio data *afterwards* -- thanks to immutable
1150 * biovecs we don't need to worry about the block layer
1151 * modifying the biovec array; so leverage bio_clone_fast().
1152 */
1153 clone = bio_clone_fast(io->base_bio, gfp, cc->bs);
1154 if (!clone)
1155 return 1;
1156
1157 crypt_inc_pending(io);
1158
1159 clone_init(io, clone);
1160 clone->bi_iter.bi_sector = cc->start + io->sector;
1161
1162 generic_make_request(clone);
1163 return 0;
1164}
1165
1166static void kcryptd_io_read_work(struct work_struct *work)
1167{
1168 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1169
1170 crypt_inc_pending(io);
1171 if (kcryptd_io_read(io, GFP_NOIO))
1172 io->error = -ENOMEM;
1173 crypt_dec_pending(io);
1174}
1175
1176static void kcryptd_queue_read(struct dm_crypt_io *io)
1177{
1178 struct crypt_config *cc = io->cc;
1179
1180 INIT_WORK(&io->work, kcryptd_io_read_work);
1181 queue_work(cc->io_queue, &io->work);
1182}
1183
1184static void kcryptd_io_write(struct dm_crypt_io *io)
1185{
1186 struct bio *clone = io->ctx.bio_out;
1187
1188 generic_make_request(clone);
1189}
1190
1191#define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1192
1193static int dmcrypt_write(void *data)
1194{
1195 struct crypt_config *cc = data;
1196 struct dm_crypt_io *io;
1197
1198 while (1) {
1199 struct rb_root write_tree;
1200 struct blk_plug plug;
1201
1202 DECLARE_WAITQUEUE(wait, current);
1203
1204 spin_lock_irq(&cc->write_thread_wait.lock);
1205continue_locked:
1206
1207 if (!RB_EMPTY_ROOT(&cc->write_tree))
1208 goto pop_from_list;
1209
1210 if (unlikely(test_bit(DM_CRYPT_EXIT_THREAD, &cc->flags))) {
1211 spin_unlock_irq(&cc->write_thread_wait.lock);
1212 break;
1213 }
1214
1215 __set_current_state(TASK_INTERRUPTIBLE);
1216 __add_wait_queue(&cc->write_thread_wait, &wait);
1217
1218 spin_unlock_irq(&cc->write_thread_wait.lock);
1219
1220 schedule();
1221
1222 spin_lock_irq(&cc->write_thread_wait.lock);
1223 __remove_wait_queue(&cc->write_thread_wait, &wait);
1224 goto continue_locked;
1225
1226pop_from_list:
1227 write_tree = cc->write_tree;
1228 cc->write_tree = RB_ROOT;
1229 spin_unlock_irq(&cc->write_thread_wait.lock);
1230
1231 BUG_ON(rb_parent(write_tree.rb_node));
1232
1233 /*
1234 * Note: we cannot walk the tree here with rb_next because
1235 * the structures may be freed when kcryptd_io_write is called.
1236 */
1237 blk_start_plug(&plug);
1238 do {
1239 io = crypt_io_from_node(rb_first(&write_tree));
1240 rb_erase(&io->rb_node, &write_tree);
1241 kcryptd_io_write(io);
1242 } while (!RB_EMPTY_ROOT(&write_tree));
1243 blk_finish_plug(&plug);
1244 }
1245 return 0;
1246}
1247
1248static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1249{
1250 struct bio *clone = io->ctx.bio_out;
1251 struct crypt_config *cc = io->cc;
1252 unsigned long flags;
1253 sector_t sector;
1254 struct rb_node **rbp, *parent;
1255
1256 if (unlikely(io->error < 0)) {
1257 crypt_free_buffer_pages(cc, clone);
1258 bio_put(clone);
1259 crypt_dec_pending(io);
1260 return;
1261 }
1262
1263 /* crypt_convert should have filled the clone bio */
1264 BUG_ON(io->ctx.iter_out.bi_size);
1265
1266 clone->bi_iter.bi_sector = cc->start + io->sector;
1267
1268 if (likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) {
1269 generic_make_request(clone);
1270 return;
1271 }
1272
1273 spin_lock_irqsave(&cc->write_thread_wait.lock, flags);
1274 rbp = &cc->write_tree.rb_node;
1275 parent = NULL;
1276 sector = io->sector;
1277 while (*rbp) {
1278 parent = *rbp;
1279 if (sector < crypt_io_from_node(parent)->sector)
1280 rbp = &(*rbp)->rb_left;
1281 else
1282 rbp = &(*rbp)->rb_right;
1283 }
1284 rb_link_node(&io->rb_node, parent, rbp);
1285 rb_insert_color(&io->rb_node, &cc->write_tree);
1286
1287 wake_up_locked(&cc->write_thread_wait);
1288 spin_unlock_irqrestore(&cc->write_thread_wait.lock, flags);
1289}
1290
1291static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
1292{
1293 struct crypt_config *cc = io->cc;
1294 struct bio *clone;
1295 int crypt_finished;
1296 sector_t sector = io->sector;
1297 int r;
1298
1299 /*
1300 * Prevent io from disappearing until this function completes.
1301 */
1302 crypt_inc_pending(io);
1303 crypt_convert_init(cc, &io->ctx, NULL, io->base_bio, sector);
1304
1305 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1306 if (unlikely(!clone)) {
1307 io->error = -EIO;
1308 goto dec;
1309 }
1310
1311 io->ctx.bio_out = clone;
1312 io->ctx.iter_out = clone->bi_iter;
1313
1314 sector += bio_sectors(clone);
1315
1316 crypt_inc_pending(io);
1317 r = crypt_convert(cc, &io->ctx);
1318 if (r)
1319 io->error = -EIO;
1320 crypt_finished = atomic_dec_and_test(&io->ctx.cc_pending);
1321
1322 /* Encryption was already finished, submit io now */
1323 if (crypt_finished) {
1324 kcryptd_crypt_write_io_submit(io, 0);
1325 io->sector = sector;
1326 }
1327
1328dec:
1329 crypt_dec_pending(io);
1330}
1331
1332static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
1333{
1334 crypt_dec_pending(io);
1335}
1336
1337static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
1338{
1339 struct crypt_config *cc = io->cc;
1340 int r = 0;
1341
1342 crypt_inc_pending(io);
1343
1344 crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
1345 io->sector);
1346
1347 r = crypt_convert(cc, &io->ctx);
1348 if (r < 0)
1349 io->error = -EIO;
1350
1351 if (atomic_dec_and_test(&io->ctx.cc_pending))
1352 kcryptd_crypt_read_done(io);
1353
1354 crypt_dec_pending(io);
1355}
1356
1357static void kcryptd_async_done(struct crypto_async_request *async_req,
1358 int error)
1359{
1360 struct dm_crypt_request *dmreq = async_req->data;
1361 struct convert_context *ctx = dmreq->ctx;
1362 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1363 struct crypt_config *cc = io->cc;
1364
1365 /*
1366 * A request from crypto driver backlog is going to be processed now,
1367 * finish the completion and continue in crypt_convert().
1368 * (Callback will be called for the second time for this request.)
1369 */
1370 if (error == -EINPROGRESS) {
1371 complete(&ctx->restart);
1372 return;
1373 }
1374
1375 if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
1376 error = cc->iv_gen_ops->post(cc, iv_of_dmreq(cc, dmreq), dmreq);
1377
1378 if (error < 0)
1379 io->error = -EIO;
1380
1381 crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
1382
1383 if (!atomic_dec_and_test(&ctx->cc_pending))
1384 return;
1385
1386 if (bio_data_dir(io->base_bio) == READ)
1387 kcryptd_crypt_read_done(io);
1388 else
1389 kcryptd_crypt_write_io_submit(io, 1);
1390}
1391
1392static void kcryptd_crypt(struct work_struct *work)
1393{
1394 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1395
1396 if (bio_data_dir(io->base_bio) == READ)
1397 kcryptd_crypt_read_convert(io);
1398 else
1399 kcryptd_crypt_write_convert(io);
1400}
1401
1402static void kcryptd_queue_crypt(struct dm_crypt_io *io)
1403{
1404 struct crypt_config *cc = io->cc;
1405
1406 INIT_WORK(&io->work, kcryptd_crypt);
1407 queue_work(cc->crypt_queue, &io->work);
1408}
1409
1410/*
1411 * Decode key from its hex representation
1412 */
1413static int crypt_decode_key(u8 *key, char *hex, unsigned int size)
1414{
1415 char buffer[3];
1416 unsigned int i;
1417
1418 buffer[2] = '\0';
1419
1420 for (i = 0; i < size; i++) {
1421 buffer[0] = *hex++;
1422 buffer[1] = *hex++;
1423
1424 if (kstrtou8(buffer, 16, &key[i]))
1425 return -EINVAL;
1426 }
1427
1428 if (*hex != '\0')
1429 return -EINVAL;
1430
1431 return 0;
1432}
1433
1434static void crypt_free_tfms(struct crypt_config *cc)
1435{
1436 unsigned i;
1437
1438 if (!cc->tfms)
1439 return;
1440
1441 for (i = 0; i < cc->tfms_count; i++)
1442 if (cc->tfms[i] && !IS_ERR(cc->tfms[i])) {
1443 crypto_free_skcipher(cc->tfms[i]);
1444 cc->tfms[i] = NULL;
1445 }
1446
1447 kfree(cc->tfms);
1448 cc->tfms = NULL;
1449}
1450
1451static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
1452{
1453 unsigned i;
1454 int err;
1455
1456 cc->tfms = kmalloc(cc->tfms_count * sizeof(struct crypto_skcipher *),
1457 GFP_KERNEL);
1458 if (!cc->tfms)
1459 return -ENOMEM;
1460
1461 for (i = 0; i < cc->tfms_count; i++) {
1462 cc->tfms[i] = crypto_alloc_skcipher(ciphermode, 0, 0);
1463 if (IS_ERR(cc->tfms[i])) {
1464 err = PTR_ERR(cc->tfms[i]);
1465 crypt_free_tfms(cc);
1466 return err;
1467 }
1468 }
1469
1470 return 0;
1471}
1472
1473static int crypt_setkey_allcpus(struct crypt_config *cc)
1474{
1475 unsigned subkey_size;
1476 int err = 0, i, r;
1477
1478 /* Ignore extra keys (which are used for IV etc) */
1479 subkey_size = (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
1480
1481 for (i = 0; i < cc->tfms_count; i++) {
1482 r = crypto_skcipher_setkey(cc->tfms[i],
1483 cc->key + (i * subkey_size),
1484 subkey_size);
1485 if (r)
1486 err = r;
1487 }
1488
1489 return err;
1490}
1491
1492static int crypt_set_key(struct crypt_config *cc, char *key)
1493{
1494 int r = -EINVAL;
1495 int key_string_len = strlen(key);
1496
1497 /* The key size may not be changed. */
1498 if (cc->key_size != (key_string_len >> 1))
1499 goto out;
1500
1501 /* Hyphen (which gives a key_size of zero) means there is no key. */
1502 if (!cc->key_size && strcmp(key, "-"))
1503 goto out;
1504
1505 if (cc->key_size && crypt_decode_key(cc->key, key, cc->key_size) < 0)
1506 goto out;
1507
1508 set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
1509
1510 r = crypt_setkey_allcpus(cc);
1511
1512out:
1513 /* Hex key string not needed after here, so wipe it. */
1514 memset(key, '0', key_string_len);
1515
1516 return r;
1517}
1518
1519static int crypt_wipe_key(struct crypt_config *cc)
1520{
1521 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
1522 memset(&cc->key, 0, cc->key_size * sizeof(u8));
1523
1524 return crypt_setkey_allcpus(cc);
1525}
1526
1527static void crypt_dtr(struct dm_target *ti)
1528{
1529 struct crypt_config *cc = ti->private;
1530
1531 ti->private = NULL;
1532
1533 if (!cc)
1534 return;
1535
1536 if (cc->write_thread) {
1537 spin_lock_irq(&cc->write_thread_wait.lock);
1538 set_bit(DM_CRYPT_EXIT_THREAD, &cc->flags);
1539 wake_up_locked(&cc->write_thread_wait);
1540 spin_unlock_irq(&cc->write_thread_wait.lock);
1541 kthread_stop(cc->write_thread);
1542 }
1543
1544 if (cc->io_queue)
1545 destroy_workqueue(cc->io_queue);
1546 if (cc->crypt_queue)
1547 destroy_workqueue(cc->crypt_queue);
1548
1549 crypt_free_tfms(cc);
1550
1551 if (cc->bs)
1552 bioset_free(cc->bs);
1553
1554 mempool_destroy(cc->page_pool);
1555 mempool_destroy(cc->req_pool);
1556
1557 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
1558 cc->iv_gen_ops->dtr(cc);
1559
1560 if (cc->dev)
1561 dm_put_device(ti, cc->dev);
1562
1563 kzfree(cc->cipher);
1564 kzfree(cc->cipher_string);
1565
1566 /* Must zero key material before freeing */
1567 kzfree(cc);
1568}
1569
1570static int crypt_ctr_cipher(struct dm_target *ti,
1571 char *cipher_in, char *key)
1572{
1573 struct crypt_config *cc = ti->private;
1574 char *tmp, *cipher, *chainmode, *ivmode, *ivopts, *keycount;
1575 char *cipher_api = NULL;
1576 int ret = -EINVAL;
1577 char dummy;
1578
1579 /* Convert to crypto api definition? */
1580 if (strchr(cipher_in, '(')) {
1581 ti->error = "Bad cipher specification";
1582 return -EINVAL;
1583 }
1584
1585 cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
1586 if (!cc->cipher_string)
1587 goto bad_mem;
1588
1589 /*
1590 * Legacy dm-crypt cipher specification
1591 * cipher[:keycount]-mode-iv:ivopts
1592 */
1593 tmp = cipher_in;
1594 keycount = strsep(&tmp, "-");
1595 cipher = strsep(&keycount, ":");
1596
1597 if (!keycount)
1598 cc->tfms_count = 1;
1599 else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
1600 !is_power_of_2(cc->tfms_count)) {
1601 ti->error = "Bad cipher key count specification";
1602 return -EINVAL;
1603 }
1604 cc->key_parts = cc->tfms_count;
1605 cc->key_extra_size = 0;
1606
1607 cc->cipher = kstrdup(cipher, GFP_KERNEL);
1608 if (!cc->cipher)
1609 goto bad_mem;
1610
1611 chainmode = strsep(&tmp, "-");
1612 ivopts = strsep(&tmp, "-");
1613 ivmode = strsep(&ivopts, ":");
1614
1615 if (tmp)
1616 DMWARN("Ignoring unexpected additional cipher options");
1617
1618 /*
1619 * For compatibility with the original dm-crypt mapping format, if
1620 * only the cipher name is supplied, use cbc-plain.
1621 */
1622 if (!chainmode || (!strcmp(chainmode, "plain") && !ivmode)) {
1623 chainmode = "cbc";
1624 ivmode = "plain";
1625 }
1626
1627 if (strcmp(chainmode, "ecb") && !ivmode) {
1628 ti->error = "IV mechanism required";
1629 return -EINVAL;
1630 }
1631
1632 cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
1633 if (!cipher_api)
1634 goto bad_mem;
1635
1636 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
1637 "%s(%s)", chainmode, cipher);
1638 if (ret < 0) {
1639 kfree(cipher_api);
1640 goto bad_mem;
1641 }
1642
1643 /* Allocate cipher */
1644 ret = crypt_alloc_tfms(cc, cipher_api);
1645 if (ret < 0) {
1646 ti->error = "Error allocating crypto tfm";
1647 goto bad;
1648 }
1649
1650 /* Initialize IV */
1651 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
1652 if (cc->iv_size)
1653 /* at least a 64 bit sector number should fit in our buffer */
1654 cc->iv_size = max(cc->iv_size,
1655 (unsigned int)(sizeof(u64) / sizeof(u8)));
1656 else if (ivmode) {
1657 DMWARN("Selected cipher does not support IVs");
1658 ivmode = NULL;
1659 }
1660
1661 /* Choose ivmode, see comments at iv code. */
1662 if (ivmode == NULL)
1663 cc->iv_gen_ops = NULL;
1664 else if (strcmp(ivmode, "plain") == 0)
1665 cc->iv_gen_ops = &crypt_iv_plain_ops;
1666 else if (strcmp(ivmode, "plain64") == 0)
1667 cc->iv_gen_ops = &crypt_iv_plain64_ops;
1668 else if (strcmp(ivmode, "essiv") == 0)
1669 cc->iv_gen_ops = &crypt_iv_essiv_ops;
1670 else if (strcmp(ivmode, "benbi") == 0)
1671 cc->iv_gen_ops = &crypt_iv_benbi_ops;
1672 else if (strcmp(ivmode, "null") == 0)
1673 cc->iv_gen_ops = &crypt_iv_null_ops;
1674 else if (strcmp(ivmode, "lmk") == 0) {
1675 cc->iv_gen_ops = &crypt_iv_lmk_ops;
1676 /*
1677 * Version 2 and 3 is recognised according
1678 * to length of provided multi-key string.
1679 * If present (version 3), last key is used as IV seed.
1680 * All keys (including IV seed) are always the same size.
1681 */
1682 if (cc->key_size % cc->key_parts) {
1683 cc->key_parts++;
1684 cc->key_extra_size = cc->key_size / cc->key_parts;
1685 }
1686 } else if (strcmp(ivmode, "tcw") == 0) {
1687 cc->iv_gen_ops = &crypt_iv_tcw_ops;
1688 cc->key_parts += 2; /* IV + whitening */
1689 cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
1690 } else {
1691 ret = -EINVAL;
1692 ti->error = "Invalid IV mode";
1693 goto bad;
1694 }
1695
1696 /* Initialize and set key */
1697 ret = crypt_set_key(cc, key);
1698 if (ret < 0) {
1699 ti->error = "Error decoding and setting key";
1700 goto bad;
1701 }
1702
1703 /* Allocate IV */
1704 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
1705 ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
1706 if (ret < 0) {
1707 ti->error = "Error creating IV";
1708 goto bad;
1709 }
1710 }
1711
1712 /* Initialize IV (set keys for ESSIV etc) */
1713 if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
1714 ret = cc->iv_gen_ops->init(cc);
1715 if (ret < 0) {
1716 ti->error = "Error initialising IV";
1717 goto bad;
1718 }
1719 }
1720
1721 ret = 0;
1722bad:
1723 kfree(cipher_api);
1724 return ret;
1725
1726bad_mem:
1727 ti->error = "Cannot allocate cipher strings";
1728 return -ENOMEM;
1729}
1730
1731/*
1732 * Construct an encryption mapping:
1733 * <cipher> <key> <iv_offset> <dev_path> <start>
1734 */
1735static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1736{
1737 struct crypt_config *cc;
1738 unsigned int key_size, opt_params;
1739 unsigned long long tmpll;
1740 int ret;
1741 size_t iv_size_padding;
1742 struct dm_arg_set as;
1743 const char *opt_string;
1744 char dummy;
1745
1746 static struct dm_arg _args[] = {
1747 {0, 3, "Invalid number of feature args"},
1748 };
1749
1750 if (argc < 5) {
1751 ti->error = "Not enough arguments";
1752 return -EINVAL;
1753 }
1754
1755 key_size = strlen(argv[1]) >> 1;
1756
1757 cc = kzalloc(sizeof(*cc) + key_size * sizeof(u8), GFP_KERNEL);
1758 if (!cc) {
1759 ti->error = "Cannot allocate encryption context";
1760 return -ENOMEM;
1761 }
1762 cc->key_size = key_size;
1763
1764 ti->private = cc;
1765 ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
1766 if (ret < 0)
1767 goto bad;
1768
1769 cc->dmreq_start = sizeof(struct skcipher_request);
1770 cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
1771 cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
1772
1773 if (crypto_skcipher_alignmask(any_tfm(cc)) < CRYPTO_MINALIGN) {
1774 /* Allocate the padding exactly */
1775 iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
1776 & crypto_skcipher_alignmask(any_tfm(cc));
1777 } else {
1778 /*
1779 * If the cipher requires greater alignment than kmalloc
1780 * alignment, we don't know the exact position of the
1781 * initialization vector. We must assume worst case.
1782 */
1783 iv_size_padding = crypto_skcipher_alignmask(any_tfm(cc));
1784 }
1785
1786 ret = -ENOMEM;
1787 cc->req_pool = mempool_create_kmalloc_pool(MIN_IOS, cc->dmreq_start +
1788 sizeof(struct dm_crypt_request) + iv_size_padding + cc->iv_size);
1789 if (!cc->req_pool) {
1790 ti->error = "Cannot allocate crypt request mempool";
1791 goto bad;
1792 }
1793
1794 cc->per_bio_data_size = ti->per_io_data_size =
1795 ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start +
1796 sizeof(struct dm_crypt_request) + iv_size_padding + cc->iv_size,
1797 ARCH_KMALLOC_MINALIGN);
1798
1799 cc->page_pool = mempool_create_page_pool(BIO_MAX_PAGES, 0);
1800 if (!cc->page_pool) {
1801 ti->error = "Cannot allocate page mempool";
1802 goto bad;
1803 }
1804
1805 cc->bs = bioset_create(MIN_IOS, 0);
1806 if (!cc->bs) {
1807 ti->error = "Cannot allocate crypt bioset";
1808 goto bad;
1809 }
1810
1811 mutex_init(&cc->bio_alloc_lock);
1812
1813 ret = -EINVAL;
1814 if (sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) {
1815 ti->error = "Invalid iv_offset sector";
1816 goto bad;
1817 }
1818 cc->iv_offset = tmpll;
1819
1820 ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
1821 if (ret) {
1822 ti->error = "Device lookup failed";
1823 goto bad;
1824 }
1825
1826 ret = -EINVAL;
1827 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1) {
1828 ti->error = "Invalid device sector";
1829 goto bad;
1830 }
1831 cc->start = tmpll;
1832
1833 argv += 5;
1834 argc -= 5;
1835
1836 /* Optional parameters */
1837 if (argc) {
1838 as.argc = argc;
1839 as.argv = argv;
1840
1841 ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
1842 if (ret)
1843 goto bad;
1844
1845 ret = -EINVAL;
1846 while (opt_params--) {
1847 opt_string = dm_shift_arg(&as);
1848 if (!opt_string) {
1849 ti->error = "Not enough feature arguments";
1850 goto bad;
1851 }
1852
1853 if (!strcasecmp(opt_string, "allow_discards"))
1854 ti->num_discard_bios = 1;
1855
1856 else if (!strcasecmp(opt_string, "same_cpu_crypt"))
1857 set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
1858
1859 else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
1860 set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
1861
1862 else {
1863 ti->error = "Invalid feature arguments";
1864 goto bad;
1865 }
1866 }
1867 }
1868
1869 ret = -ENOMEM;
1870 cc->io_queue = alloc_workqueue("kcryptd_io", WQ_MEM_RECLAIM, 1);
1871 if (!cc->io_queue) {
1872 ti->error = "Couldn't create kcryptd io queue";
1873 goto bad;
1874 }
1875
1876 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
1877 cc->crypt_queue = alloc_workqueue("kcryptd", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM, 1);
1878 else
1879 cc->crypt_queue = alloc_workqueue("kcryptd", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
1880 num_online_cpus());
1881 if (!cc->crypt_queue) {
1882 ti->error = "Couldn't create kcryptd queue";
1883 goto bad;
1884 }
1885
1886 init_waitqueue_head(&cc->write_thread_wait);
1887 cc->write_tree = RB_ROOT;
1888
1889 cc->write_thread = kthread_create(dmcrypt_write, cc, "dmcrypt_write");
1890 if (IS_ERR(cc->write_thread)) {
1891 ret = PTR_ERR(cc->write_thread);
1892 cc->write_thread = NULL;
1893 ti->error = "Couldn't spawn write thread";
1894 goto bad;
1895 }
1896 wake_up_process(cc->write_thread);
1897
1898 ti->num_flush_bios = 1;
1899 ti->discard_zeroes_data_unsupported = true;
1900
1901 return 0;
1902
1903bad:
1904 crypt_dtr(ti);
1905 return ret;
1906}
1907
1908static int crypt_map(struct dm_target *ti, struct bio *bio)
1909{
1910 struct dm_crypt_io *io;
1911 struct crypt_config *cc = ti->private;
1912
1913 /*
1914 * If bio is REQ_FLUSH or REQ_DISCARD, just bypass crypt queues.
1915 * - for REQ_FLUSH device-mapper core ensures that no IO is in-flight
1916 * - for REQ_DISCARD caller must use flush if IO ordering matters
1917 */
1918 if (unlikely(bio->bi_rw & (REQ_FLUSH | REQ_DISCARD))) {
1919 bio->bi_bdev = cc->dev->bdev;
1920 if (bio_sectors(bio))
1921 bio->bi_iter.bi_sector = cc->start +
1922 dm_target_offset(ti, bio->bi_iter.bi_sector);
1923 return DM_MAPIO_REMAPPED;
1924 }
1925
1926 io = dm_per_bio_data(bio, cc->per_bio_data_size);
1927 crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
1928 io->ctx.req = (struct skcipher_request *)(io + 1);
1929
1930 if (bio_data_dir(io->base_bio) == READ) {
1931 if (kcryptd_io_read(io, GFP_NOWAIT))
1932 kcryptd_queue_read(io);
1933 } else
1934 kcryptd_queue_crypt(io);
1935
1936 return DM_MAPIO_SUBMITTED;
1937}
1938
1939static void crypt_status(struct dm_target *ti, status_type_t type,
1940 unsigned status_flags, char *result, unsigned maxlen)
1941{
1942 struct crypt_config *cc = ti->private;
1943 unsigned i, sz = 0;
1944 int num_feature_args = 0;
1945
1946 switch (type) {
1947 case STATUSTYPE_INFO:
1948 result[0] = '\0';
1949 break;
1950
1951 case STATUSTYPE_TABLE:
1952 DMEMIT("%s ", cc->cipher_string);
1953
1954 if (cc->key_size > 0)
1955 for (i = 0; i < cc->key_size; i++)
1956 DMEMIT("%02x", cc->key[i]);
1957 else
1958 DMEMIT("-");
1959
1960 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
1961 cc->dev->name, (unsigned long long)cc->start);
1962
1963 num_feature_args += !!ti->num_discard_bios;
1964 num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
1965 num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
1966 if (num_feature_args) {
1967 DMEMIT(" %d", num_feature_args);
1968 if (ti->num_discard_bios)
1969 DMEMIT(" allow_discards");
1970 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
1971 DMEMIT(" same_cpu_crypt");
1972 if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
1973 DMEMIT(" submit_from_crypt_cpus");
1974 }
1975
1976 break;
1977 }
1978}
1979
1980static void crypt_postsuspend(struct dm_target *ti)
1981{
1982 struct crypt_config *cc = ti->private;
1983
1984 set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
1985}
1986
1987static int crypt_preresume(struct dm_target *ti)
1988{
1989 struct crypt_config *cc = ti->private;
1990
1991 if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
1992 DMERR("aborting resume - crypt key is not set.");
1993 return -EAGAIN;
1994 }
1995
1996 return 0;
1997}
1998
1999static void crypt_resume(struct dm_target *ti)
2000{
2001 struct crypt_config *cc = ti->private;
2002
2003 clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
2004}
2005
2006/* Message interface
2007 * key set <key>
2008 * key wipe
2009 */
2010static int crypt_message(struct dm_target *ti, unsigned argc, char **argv)
2011{
2012 struct crypt_config *cc = ti->private;
2013 int ret = -EINVAL;
2014
2015 if (argc < 2)
2016 goto error;
2017
2018 if (!strcasecmp(argv[0], "key")) {
2019 if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
2020 DMWARN("not suspended during key manipulation.");
2021 return -EINVAL;
2022 }
2023 if (argc == 3 && !strcasecmp(argv[1], "set")) {
2024 ret = crypt_set_key(cc, argv[2]);
2025 if (ret)
2026 return ret;
2027 if (cc->iv_gen_ops && cc->iv_gen_ops->init)
2028 ret = cc->iv_gen_ops->init(cc);
2029 return ret;
2030 }
2031 if (argc == 2 && !strcasecmp(argv[1], "wipe")) {
2032 if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2033 ret = cc->iv_gen_ops->wipe(cc);
2034 if (ret)
2035 return ret;
2036 }
2037 return crypt_wipe_key(cc);
2038 }
2039 }
2040
2041error:
2042 DMWARN("unrecognised message received.");
2043 return -EINVAL;
2044}
2045
2046static int crypt_iterate_devices(struct dm_target *ti,
2047 iterate_devices_callout_fn fn, void *data)
2048{
2049 struct crypt_config *cc = ti->private;
2050
2051 return fn(ti, cc->dev, cc->start, ti->len, data);
2052}
2053
2054static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
2055{
2056 /*
2057 * Unfortunate constraint that is required to avoid the potential
2058 * for exceeding underlying device's max_segments limits -- due to
2059 * crypt_alloc_buffer() possibly allocating pages for the encryption
2060 * bio that are not as physically contiguous as the original bio.
2061 */
2062 limits->max_segment_size = PAGE_SIZE;
2063}
2064
2065static struct target_type crypt_target = {
2066 .name = "crypt",
2067 .version = {1, 14, 1},
2068 .module = THIS_MODULE,
2069 .ctr = crypt_ctr,
2070 .dtr = crypt_dtr,
2071 .map = crypt_map,
2072 .status = crypt_status,
2073 .postsuspend = crypt_postsuspend,
2074 .preresume = crypt_preresume,
2075 .resume = crypt_resume,
2076 .message = crypt_message,
2077 .iterate_devices = crypt_iterate_devices,
2078 .io_hints = crypt_io_hints,
2079};
2080
2081static int __init dm_crypt_init(void)
2082{
2083 int r;
2084
2085 r = dm_register_target(&crypt_target);
2086 if (r < 0)
2087 DMERR("register failed %d", r);
2088
2089 return r;
2090}
2091
2092static void __exit dm_crypt_exit(void)
2093{
2094 dm_unregister_target(&crypt_target);
2095}
2096
2097module_init(dm_crypt_init);
2098module_exit(dm_crypt_exit);
2099
2100MODULE_AUTHOR("Jana Saout <jana@saout.de>");
2101MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
2102MODULE_LICENSE("GPL");