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 | */ |
53 | struct 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 | */ |
73 | struct 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 | |
92 | struct 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 | |
99 | struct crypt_config; |
100 | |
101 | struct 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 | |
113 | struct iv_benbi_private { |
114 | int shift; |
115 | }; |
116 | |
117 | #define LMK_SEED_SIZE 64 /* hash + 0 */ |
118 | struct iv_lmk_private { |
119 | struct crypto_shash *hash_tfm; |
120 | u8 *seed; |
121 | }; |
122 | |
123 | #define TCW_WHITENING_SIZE 16 |
124 | struct 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 |
131 | struct 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 | */ |
139 | enum 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 | |
144 | enum 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 | */ |
153 | struct 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 ; /* 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 | |
236 | static DEFINE_SPINLOCK(dm_crypt_clients_lock); |
237 | static unsigned int dm_crypt_clients_n; |
238 | static 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 | |
242 | static void crypt_endio(struct bio *clone); |
243 | static void kcryptd_queue_crypt(struct dm_crypt_io *io); |
244 | static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc, |
245 | struct scatterlist *sg); |
246 | |
247 | static bool crypt_integrity_aead(struct crypt_config *cc); |
248 | |
249 | /* |
250 | * Use this to access cipher attributes that are independent of the key. |
251 | */ |
252 | static struct crypto_skcipher *any_tfm(struct crypt_config *cc) |
253 | { |
254 | return cc->cipher_tfm.tfms[0]; |
255 | } |
256 | |
257 | static 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 | |
318 | static 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 | |
327 | static 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 | |
336 | static 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 | |
346 | static 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 | |
359 | static 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(tfm: any_tfm_aead(cc)); |
367 | else |
368 | bs = crypto_skcipher_blocksize(tfm: 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 | |
390 | static void crypt_iv_benbi_dtr(struct crypt_config *cc) |
391 | { |
392 | } |
393 | |
394 | static 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 | |
407 | static 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 | |
415 | static 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(ptr: lmk->hash_tfm)) |
420 | crypto_free_shash(tfm: lmk->hash_tfm); |
421 | lmk->hash_tfm = NULL; |
422 | |
423 | kfree_sensitive(objp: lmk->seed); |
424 | lmk->seed = NULL; |
425 | } |
426 | |
427 | static 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(alg_name: "md5" , type: 0, |
438 | CRYPTO_ALG_ALLOCATES_MEMORY); |
439 | if (IS_ERR(ptr: lmk->hash_tfm)) { |
440 | ti->error = "Error initializing LMK hash" ; |
441 | return PTR_ERR(ptr: 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 | |
460 | static 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 | |
473 | static 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 | |
483 | static 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, data: 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: data + 16, len: 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, data: (u8 *)buf, len: sizeof(buf)); |
516 | if (r) |
517 | return r; |
518 | |
519 | /* No MD5 padding here */ |
520 | r = crypto_shash_export(desc, out: &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 | |
531 | static 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, sg: dmreq->sg_in); |
540 | src = kmap_local_page(page: sg_page(sg)); |
541 | r = crypt_iv_lmk_one(cc, iv, dmreq, data: src + sg->offset); |
542 | kunmap_local(src); |
543 | } else |
544 | memset(iv, 0, cc->iv_size); |
545 | |
546 | return r; |
547 | } |
548 | |
549 | static 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, sg: dmreq->sg_out); |
560 | dst = kmap_local_page(page: sg_page(sg)); |
561 | r = crypt_iv_lmk_one(cc, iv, dmreq, data: dst + sg->offset); |
562 | |
563 | /* Tweak the first block of plaintext sector */ |
564 | if (!r) |
565 | crypto_xor(dst: dst + sg->offset, src: iv, size: cc->iv_size); |
566 | |
567 | kunmap_local(dst); |
568 | return r; |
569 | } |
570 | |
571 | static 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(objp: tcw->iv_seed); |
576 | tcw->iv_seed = NULL; |
577 | kfree_sensitive(objp: tcw->whitening); |
578 | tcw->whitening = NULL; |
579 | |
580 | if (tcw->crc32_tfm && !IS_ERR(ptr: tcw->crc32_tfm)) |
581 | crypto_free_shash(tfm: tcw->crc32_tfm); |
582 | tcw->crc32_tfm = NULL; |
583 | } |
584 | |
585 | static 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(alg_name: "crc32" , type: 0, |
601 | CRYPTO_ALG_ALLOCATES_MEMORY); |
602 | if (IS_ERR(ptr: tcw->crc32_tfm)) { |
603 | ti->error = "Error initializing CRC32 in TCW" ; |
604 | return PTR_ERR(ptr: tcw->crc32_tfm); |
605 | } |
606 | |
607 | tcw->iv_seed = kzalloc(size: 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 | |
618 | static 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 | |
630 | static 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 | |
640 | static 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(dst: buf, src1: tcw->whitening, src2: (u8 *)§or, size: 8); |
652 | crypto_xor_cpy(dst: &buf[8], src1: tcw->whitening + 8, src2: (u8 *)§or, size: 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, data: &buf[i * 4], len: 4, out: &buf[i * 4]); |
658 | if (r) |
659 | goto out; |
660 | } |
661 | crypto_xor(dst: &buf[0], src: &buf[12], size: 4); |
662 | crypto_xor(dst: &buf[4], src: &buf[8], size: 4); |
663 | |
664 | /* apply whitening (8 bytes) to whole sector */ |
665 | for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++) |
666 | crypto_xor(dst: data + i * 8, src: buf, size: 8); |
667 | out: |
668 | memzero_explicit(s: buf, count: sizeof(buf)); |
669 | return r; |
670 | } |
671 | |
672 | static 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, sg: dmreq->sg_in); |
684 | src = kmap_local_page(page: sg_page(sg)); |
685 | r = crypt_iv_tcw_whitening(cc, dmreq, data: src + sg->offset); |
686 | kunmap_local(src); |
687 | } |
688 | |
689 | /* Calculate IV */ |
690 | crypto_xor_cpy(dst: iv, src1: tcw->iv_seed, src2: (u8 *)§or, size: 8); |
691 | if (cc->iv_size > 8) |
692 | crypto_xor_cpy(dst: &iv[8], src1: tcw->iv_seed + 8, src2: (u8 *)§or, |
693 | size: cc->iv_size - 8); |
694 | |
695 | return r; |
696 | } |
697 | |
698 | static 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, sg: dmreq->sg_out); |
710 | dst = kmap_local_page(page: sg_page(sg)); |
711 | r = crypt_iv_tcw_whitening(cc, dmreq, data: dst + sg->offset); |
712 | kunmap_local(dst); |
713 | |
714 | return r; |
715 | } |
716 | |
717 | static 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(buf: iv, len: cc->iv_size); |
722 | return 0; |
723 | } |
724 | |
725 | static 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(tfm: 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 | |
741 | static 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(size: 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: &src, dst: &dst, cryptlen: cc->iv_size, iv: buf); |
768 | skcipher_request_set_callback(req, flags: 0, compl: crypto_req_done, data: &wait); |
769 | err = crypto_wait_req(err: crypto_skcipher_encrypt(req), wait: &wait); |
770 | kfree_sensitive(objp: req); |
771 | |
772 | return err; |
773 | } |
774 | |
775 | static 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(tfm: elephant->tfm); |
780 | elephant->tfm = NULL; |
781 | } |
782 | |
783 | static 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(alg_name: "ecb(aes)" , type: 0, |
790 | CRYPTO_ALG_ALLOCATES_MEMORY); |
791 | if (IS_ERR(ptr: elephant->tfm)) { |
792 | r = PTR_ERR(ptr: 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 | |
803 | static 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 | |
813 | static 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 | |
823 | static 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 | |
854 | static 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 | |
885 | static 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 | |
916 | static 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 | |
947 | static 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(tfm: elephant->tfm, GFP_NOIO); |
957 | es = kzalloc(size: 16, GFP_NOIO); /* Key for AES */ |
958 | ks = kzalloc(size: 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: &src, dst: &dst, cryptlen: 16, NULL); |
971 | skcipher_request_set_callback(req, flags: 0, compl: crypto_req_done, data: &wait); |
972 | r = crypto_wait_req(err: crypto_skcipher_encrypt(req), wait: &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(err: crypto_skcipher_encrypt(req), wait: &wait); |
980 | if (r) |
981 | goto out; |
982 | |
983 | sg = crypt_get_sg_data(cc, sg: dmreq->sg_out); |
984 | data = kmap_local_page(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, sg: dmreq->sg_in); |
990 | data2 = kmap_local_page(page: sg_page(sg: 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(d: (u32 *)data_offset, n: cc->sector_size / sizeof(u32)); |
997 | diffuser_b_decrypt(d: (u32 *)data_offset, n: cc->sector_size / sizeof(u32)); |
998 | diffuser_a_decrypt(d: (u32 *)data_offset, n: cc->sector_size / sizeof(u32)); |
999 | diffuser_cpu_to_disk(d: (__le32 *)data_offset, n: cc->sector_size / sizeof(u32)); |
1000 | } |
1001 | |
1002 | for (i = 0; i < (cc->sector_size / 32); i++) |
1003 | crypto_xor(dst: data_offset + i * 32, src: ks, size: 32); |
1004 | |
1005 | if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) { |
1006 | diffuser_disk_to_cpu(d: (u32 *)data_offset, n: cc->sector_size / sizeof(u32)); |
1007 | diffuser_a_encrypt(d: (u32 *)data_offset, n: cc->sector_size / sizeof(u32)); |
1008 | diffuser_b_encrypt(d: (u32 *)data_offset, n: cc->sector_size / sizeof(u32)); |
1009 | diffuser_cpu_to_disk(d: (__le32 *)data_offset, n: cc->sector_size / sizeof(u32)); |
1010 | } |
1011 | |
1012 | kunmap_local(data); |
1013 | out: |
1014 | kfree_sensitive(objp: ks); |
1015 | kfree_sensitive(objp: es); |
1016 | skcipher_request_free(req); |
1017 | return r; |
1018 | } |
1019 | |
1020 | static 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 | |
1034 | static 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 | |
1043 | static 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(tfm: elephant->tfm, key: &cc->key[key_offset], keylen: cc->key_extra_size); |
1049 | } |
1050 | |
1051 | static 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(tfm: elephant->tfm, key, keylen: cc->key_extra_size); |
1058 | } |
1059 | |
1060 | static const struct crypt_iv_operations crypt_iv_plain_ops = { |
1061 | .generator = crypt_iv_plain_gen |
1062 | }; |
1063 | |
1064 | static const struct crypt_iv_operations crypt_iv_plain64_ops = { |
1065 | .generator = crypt_iv_plain64_gen |
1066 | }; |
1067 | |
1068 | static const struct crypt_iv_operations crypt_iv_plain64be_ops = { |
1069 | .generator = crypt_iv_plain64be_gen |
1070 | }; |
1071 | |
1072 | static const struct crypt_iv_operations crypt_iv_essiv_ops = { |
1073 | .generator = crypt_iv_essiv_gen |
1074 | }; |
1075 | |
1076 | static 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 | |
1082 | static const struct crypt_iv_operations crypt_iv_null_ops = { |
1083 | .generator = crypt_iv_null_gen |
1084 | }; |
1085 | |
1086 | static 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 | |
1095 | static 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 | |
1104 | static const struct crypt_iv_operations crypt_iv_random_ops = { |
1105 | .generator = crypt_iv_random_gen |
1106 | }; |
1107 | |
1108 | static const struct crypt_iv_operations crypt_iv_eboiv_ops = { |
1109 | .ctr = crypt_iv_eboiv_ctr, |
1110 | .generator = crypt_iv_eboiv_gen |
1111 | }; |
1112 | |
1113 | static 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 | */ |
1125 | static bool crypt_integrity_aead(struct crypt_config *cc) |
1126 | { |
1127 | return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags); |
1128 | } |
1129 | |
1130 | static 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 */ |
1136 | static 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 | |
1145 | static 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(ptr: bip)) |
1156 | return PTR_ERR(ptr: 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 | |
1170 | static 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(disk: cc->dev->bdev->bd_disk); |
1174 | struct mapped_device *md = dm_table_get_md(t: ti->table); |
1175 | |
1176 | /* From now we require underlying device with our integrity profile */ |
1177 | if (!bi || strcasecmp(s1: bi->profile->name, s2: "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(tfm: any_tfm_aead(cc), authsize: 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 | |
1217 | static 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(x: &ctx->restart); |
1230 | } |
1231 | |
1232 | static 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 | |
1238 | static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq) |
1239 | { |
1240 | return (void *)((char *)dmreq - cc->dmreq_start); |
1241 | } |
1242 | |
1243 | static 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 | |
1254 | static 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 | |
1260 | static __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 | |
1268 | static 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 | |
1277 | static 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 | |
1287 | static 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 | |
1293 | static 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(sg: &dmreq->sg_in[0], buf: sector, buflen: sizeof(uint64_t)); |
1334 | sg_set_buf(sg: &dmreq->sg_in[1], buf: org_iv, buflen: cc->iv_size); |
1335 | sg_set_page(sg: &dmreq->sg_in[2], page: bv_in.bv_page, len: cc->sector_size, offset: bv_in.bv_offset); |
1336 | sg_set_buf(sg: &dmreq->sg_in[3], buf: tag, buflen: cc->integrity_tag_size); |
1337 | |
1338 | sg_init_table(dmreq->sg_out, 4); |
1339 | sg_set_buf(sg: &dmreq->sg_out[0], buf: sector, buflen: sizeof(uint64_t)); |
1340 | sg_set_buf(sg: &dmreq->sg_out[1], buf: org_iv, buflen: cc->iv_size); |
1341 | sg_set_page(sg: &dmreq->sg_out[2], page: bv_out.bv_page, len: cc->sector_size, offset: bv_out.bv_offset); |
1342 | sg_set_buf(sg: &dmreq->sg_out[3], buf: tag, buflen: 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, assoclen: sizeof(uint64_t) + cc->iv_size); |
1361 | if (bio_data_dir(ctx->bio_in) == WRITE) { |
1362 | aead_request_set_crypt(req, src: dmreq->sg_in, dst: dmreq->sg_out, |
1363 | cryptlen: 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, src: dmreq->sg_in, dst: dmreq->sg_out, |
1370 | cryptlen: 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, op: "integrity-aead" , |
1382 | bio: ctx->bio_in, sector: s, result: 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(bio: ctx->bio_in, iter: &ctx->iter_in, bytes: cc->sector_size); |
1390 | bio_advance_iter(bio: ctx->bio_out, iter: &ctx->iter_out, bytes: cc->sector_size); |
1391 | |
1392 | return r; |
1393 | } |
1394 | |
1395 | static 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: sg_in, page: bv_in.bv_page, len: cc->sector_size, offset: bv_in.bv_offset); |
1433 | |
1434 | sg_init_table(sg_out, 1); |
1435 | sg_set_page(sg: sg_out, page: bv_out.bv_page, len: cc->sector_size, offset: 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, src: sg_in, dst: sg_out, cryptlen: 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(bio: ctx->bio_in, iter: &ctx->iter_in, bytes: cc->sector_size); |
1467 | bio_advance_iter(bio: ctx->bio_out, iter: &ctx->iter_out, bytes: cc->sector_size); |
1468 | |
1469 | return r; |
1470 | } |
1471 | |
1472 | static void kcryptd_async_done(void *async_req, int error); |
1473 | |
1474 | static 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(pool: &cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO); |
1481 | if (!ctx->r.req) |
1482 | return -ENOMEM; |
1483 | } |
1484 | |
1485 | skcipher_request_set_tfm(req: ctx->r.req, tfm: 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(req: ctx->r.req, |
1492 | CRYPTO_TFM_REQ_MAY_BACKLOG, |
1493 | compl: kcryptd_async_done, data: dmreq_of_req(cc, req: ctx->r.req)); |
1494 | |
1495 | return 0; |
1496 | } |
1497 | |
1498 | static 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(pool: &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(req: ctx->r.req_aead, tfm: 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(req: ctx->r.req_aead, |
1514 | CRYPTO_TFM_REQ_MAY_BACKLOG, |
1515 | compl: kcryptd_async_done, data: dmreq_of_req(cc, req: ctx->r.req_aead)); |
1516 | |
1517 | return 0; |
1518 | } |
1519 | |
1520 | static 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 | |
1529 | static 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(bio: base_bio, data_size: cc->per_bio_data_size); |
1533 | |
1534 | if ((struct skcipher_request *)(io + 1) != req) |
1535 | mempool_free(element: req, pool: &cc->req_pool); |
1536 | } |
1537 | |
1538 | static 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(bio: base_bio, data_size: cc->per_bio_data_size); |
1542 | |
1543 | if ((struct aead_request *)(io + 1) != req) |
1544 | mempool_free(element: req, pool: &cc->req_pool); |
1545 | } |
1546 | |
1547 | static 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 | */ |
1558 | static 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(v: &ctx->cc_pending, i: 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(v: &ctx->cc_pending); |
1582 | |
1583 | if (crypt_integrity_aead(cc)) |
1584 | r = crypt_convert_block_aead(cc, ctx, req: ctx->r.req_aead, tag_offset); |
1585 | else |
1586 | r = crypt_convert_block_skcipher(cc, ctx, req: 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(x: &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(x: &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(v: &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(v: &ctx->cc_pending); |
1639 | return BLK_STS_PROTECTION; |
1640 | /* |
1641 | * There was an error while processing the request. |
1642 | */ |
1643 | default: |
1644 | atomic_dec(v: &ctx->cc_pending); |
1645 | return BLK_STS_IOERR; |
1646 | } |
1647 | } |
1648 | |
1649 | return 0; |
1650 | } |
1651 | |
1652 | static 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 | */ |
1674 | static 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 | |
1683 | retry: |
1684 | if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM)) |
1685 | mutex_lock(&cc->bio_alloc_lock); |
1686 | |
1687 | clone = bio_alloc_bioset(bdev: cc->dev->bdev, nr_vecs: nr_iovecs, opf: io->base_bio->bi_opf, |
1688 | GFP_NOIO, bs: &cc->bs); |
1689 | clone->bi_private = io; |
1690 | clone->bi_end_io = crypt_endio; |
1691 | clone->bi_ioprio = io->base_bio->bi_ioprio; |
1692 | |
1693 | remaining_size = size; |
1694 | |
1695 | while (remaining_size) { |
1696 | struct page *pages; |
1697 | unsigned size_to_add; |
1698 | unsigned remaining_order = __fls(word: (remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT); |
1699 | order = min(order, remaining_order); |
1700 | |
1701 | while (order > 0) { |
1702 | if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) + |
1703 | (1 << order) > dm_crypt_pages_per_client)) |
1704 | goto decrease_order; |
1705 | pages = alloc_pages(gfp: gfp_mask |
1706 | | __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP, |
1707 | order); |
1708 | if (likely(pages != NULL)) { |
1709 | percpu_counter_add(fbc: &cc->n_allocated_pages, amount: 1 << order); |
1710 | goto have_pages; |
1711 | } |
1712 | decrease_order: |
1713 | order--; |
1714 | } |
1715 | |
1716 | pages = mempool_alloc(pool: &cc->page_pool, gfp_mask); |
1717 | if (!pages) { |
1718 | crypt_free_buffer_pages(cc, clone); |
1719 | bio_put(clone); |
1720 | gfp_mask |= __GFP_DIRECT_RECLAIM; |
1721 | order = 0; |
1722 | goto retry; |
1723 | } |
1724 | |
1725 | have_pages: |
1726 | size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size); |
1727 | __bio_add_page(bio: clone, page: pages, len: size_to_add, off: 0); |
1728 | remaining_size -= size_to_add; |
1729 | } |
1730 | |
1731 | /* Allocate space for integrity tags */ |
1732 | if (dm_crypt_integrity_io_alloc(io, bio: clone)) { |
1733 | crypt_free_buffer_pages(cc, clone); |
1734 | bio_put(clone); |
1735 | clone = NULL; |
1736 | } |
1737 | |
1738 | if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM)) |
1739 | mutex_unlock(lock: &cc->bio_alloc_lock); |
1740 | |
1741 | return clone; |
1742 | } |
1743 | |
1744 | static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone) |
1745 | { |
1746 | struct folio_iter fi; |
1747 | |
1748 | if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */ |
1749 | bio_for_each_folio_all(fi, clone) { |
1750 | if (folio_test_large(folio: fi.folio)) { |
1751 | percpu_counter_sub(fbc: &cc->n_allocated_pages, |
1752 | amount: 1 << folio_order(folio: fi.folio)); |
1753 | folio_put(folio: fi.folio); |
1754 | } else { |
1755 | mempool_free(element: &fi.folio->page, pool: &cc->page_pool); |
1756 | } |
1757 | } |
1758 | } |
1759 | } |
1760 | |
1761 | static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc, |
1762 | struct bio *bio, sector_t sector) |
1763 | { |
1764 | io->cc = cc; |
1765 | io->base_bio = bio; |
1766 | io->sector = sector; |
1767 | io->error = 0; |
1768 | io->ctx.aead_recheck = false; |
1769 | io->ctx.aead_failed = false; |
1770 | io->ctx.r.req = NULL; |
1771 | io->integrity_metadata = NULL; |
1772 | io->integrity_metadata_from_pool = false; |
1773 | atomic_set(v: &io->io_pending, i: 0); |
1774 | } |
1775 | |
1776 | static void crypt_inc_pending(struct dm_crypt_io *io) |
1777 | { |
1778 | atomic_inc(v: &io->io_pending); |
1779 | } |
1780 | |
1781 | static void kcryptd_queue_read(struct dm_crypt_io *io); |
1782 | |
1783 | /* |
1784 | * One of the bios was finished. Check for completion of |
1785 | * the whole request and correctly clean up the buffer. |
1786 | */ |
1787 | static void crypt_dec_pending(struct dm_crypt_io *io) |
1788 | { |
1789 | struct crypt_config *cc = io->cc; |
1790 | struct bio *base_bio = io->base_bio; |
1791 | blk_status_t error = io->error; |
1792 | |
1793 | if (!atomic_dec_and_test(v: &io->io_pending)) |
1794 | return; |
1795 | |
1796 | if (likely(!io->ctx.aead_recheck) && unlikely(io->ctx.aead_failed) && |
1797 | cc->on_disk_tag_size && bio_data_dir(base_bio) == READ) { |
1798 | io->ctx.aead_recheck = true; |
1799 | io->ctx.aead_failed = false; |
1800 | io->error = 0; |
1801 | kcryptd_queue_read(io); |
1802 | return; |
1803 | } |
1804 | |
1805 | if (io->ctx.r.req) |
1806 | crypt_free_req(cc, req: io->ctx.r.req, base_bio); |
1807 | |
1808 | if (unlikely(io->integrity_metadata_from_pool)) |
1809 | mempool_free(element: io->integrity_metadata, pool: &io->cc->tag_pool); |
1810 | else |
1811 | kfree(objp: io->integrity_metadata); |
1812 | |
1813 | base_bio->bi_status = error; |
1814 | |
1815 | bio_endio(base_bio); |
1816 | } |
1817 | |
1818 | /* |
1819 | * kcryptd/kcryptd_io: |
1820 | * |
1821 | * Needed because it would be very unwise to do decryption in an |
1822 | * interrupt context. |
1823 | * |
1824 | * kcryptd performs the actual encryption or decryption. |
1825 | * |
1826 | * kcryptd_io performs the IO submission. |
1827 | * |
1828 | * They must be separated as otherwise the final stages could be |
1829 | * starved by new requests which can block in the first stages due |
1830 | * to memory allocation. |
1831 | * |
1832 | * The work is done per CPU global for all dm-crypt instances. |
1833 | * They should not depend on each other and do not block. |
1834 | */ |
1835 | static void crypt_endio(struct bio *clone) |
1836 | { |
1837 | struct dm_crypt_io *io = clone->bi_private; |
1838 | struct crypt_config *cc = io->cc; |
1839 | unsigned int rw = bio_data_dir(clone); |
1840 | blk_status_t error = clone->bi_status; |
1841 | |
1842 | if (io->ctx.aead_recheck && !error) { |
1843 | kcryptd_queue_crypt(io); |
1844 | return; |
1845 | } |
1846 | |
1847 | /* |
1848 | * free the processed pages |
1849 | */ |
1850 | if (rw == WRITE || io->ctx.aead_recheck) |
1851 | crypt_free_buffer_pages(cc, clone); |
1852 | |
1853 | bio_put(clone); |
1854 | |
1855 | if (rw == READ && !error) { |
1856 | kcryptd_queue_crypt(io); |
1857 | return; |
1858 | } |
1859 | |
1860 | if (unlikely(error)) |
1861 | io->error = error; |
1862 | |
1863 | crypt_dec_pending(io); |
1864 | } |
1865 | |
1866 | #define CRYPT_MAP_READ_GFP GFP_NOWAIT |
1867 | |
1868 | static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp) |
1869 | { |
1870 | struct crypt_config *cc = io->cc; |
1871 | struct bio *clone; |
1872 | |
1873 | if (io->ctx.aead_recheck) { |
1874 | if (!(gfp & __GFP_DIRECT_RECLAIM)) |
1875 | return 1; |
1876 | crypt_inc_pending(io); |
1877 | clone = crypt_alloc_buffer(io, size: io->base_bio->bi_iter.bi_size); |
1878 | if (unlikely(!clone)) { |
1879 | crypt_dec_pending(io); |
1880 | return 1; |
1881 | } |
1882 | clone->bi_iter.bi_sector = cc->start + io->sector; |
1883 | crypt_convert_init(cc, ctx: &io->ctx, bio_out: clone, bio_in: clone, sector: io->sector); |
1884 | io->saved_bi_iter = clone->bi_iter; |
1885 | dm_submit_bio_remap(clone: io->base_bio, tgt_clone: clone); |
1886 | return 0; |
1887 | } |
1888 | |
1889 | /* |
1890 | * We need the original biovec array in order to decrypt the whole bio |
1891 | * data *afterwards* -- thanks to immutable biovecs we don't need to |
1892 | * worry about the block layer modifying the biovec array; so leverage |
1893 | * bio_alloc_clone(). |
1894 | */ |
1895 | clone = bio_alloc_clone(bdev: cc->dev->bdev, bio_src: io->base_bio, gfp, bs: &cc->bs); |
1896 | if (!clone) |
1897 | return 1; |
1898 | clone->bi_private = io; |
1899 | clone->bi_end_io = crypt_endio; |
1900 | |
1901 | crypt_inc_pending(io); |
1902 | |
1903 | clone->bi_iter.bi_sector = cc->start + io->sector; |
1904 | |
1905 | if (dm_crypt_integrity_io_alloc(io, bio: clone)) { |
1906 | crypt_dec_pending(io); |
1907 | bio_put(clone); |
1908 | return 1; |
1909 | } |
1910 | |
1911 | dm_submit_bio_remap(clone: io->base_bio, tgt_clone: clone); |
1912 | return 0; |
1913 | } |
1914 | |
1915 | static void kcryptd_io_read_work(struct work_struct *work) |
1916 | { |
1917 | struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); |
1918 | |
1919 | crypt_inc_pending(io); |
1920 | if (kcryptd_io_read(io, GFP_NOIO)) |
1921 | io->error = BLK_STS_RESOURCE; |
1922 | crypt_dec_pending(io); |
1923 | } |
1924 | |
1925 | static void kcryptd_queue_read(struct dm_crypt_io *io) |
1926 | { |
1927 | struct crypt_config *cc = io->cc; |
1928 | |
1929 | INIT_WORK(&io->work, kcryptd_io_read_work); |
1930 | queue_work(wq: cc->io_queue, work: &io->work); |
1931 | } |
1932 | |
1933 | static void kcryptd_io_write(struct dm_crypt_io *io) |
1934 | { |
1935 | struct bio *clone = io->ctx.bio_out; |
1936 | |
1937 | dm_submit_bio_remap(clone: io->base_bio, tgt_clone: clone); |
1938 | } |
1939 | |
1940 | #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node) |
1941 | |
1942 | static int dmcrypt_write(void *data) |
1943 | { |
1944 | struct crypt_config *cc = data; |
1945 | struct dm_crypt_io *io; |
1946 | |
1947 | while (1) { |
1948 | struct rb_root write_tree; |
1949 | struct blk_plug plug; |
1950 | |
1951 | spin_lock_irq(lock: &cc->write_thread_lock); |
1952 | continue_locked: |
1953 | |
1954 | if (!RB_EMPTY_ROOT(&cc->write_tree)) |
1955 | goto pop_from_list; |
1956 | |
1957 | set_current_state(TASK_INTERRUPTIBLE); |
1958 | |
1959 | spin_unlock_irq(lock: &cc->write_thread_lock); |
1960 | |
1961 | if (unlikely(kthread_should_stop())) { |
1962 | set_current_state(TASK_RUNNING); |
1963 | break; |
1964 | } |
1965 | |
1966 | schedule(); |
1967 | |
1968 | spin_lock_irq(lock: &cc->write_thread_lock); |
1969 | goto continue_locked; |
1970 | |
1971 | pop_from_list: |
1972 | write_tree = cc->write_tree; |
1973 | cc->write_tree = RB_ROOT; |
1974 | spin_unlock_irq(lock: &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 | |
1994 | static 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(clone: io->base_bio, tgt_clone: 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(tsk: 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(node: &io->rb_node, parent, rb_link: rbp); |
2034 | rb_insert_color(&io->rb_node, &cc->write_tree); |
2035 | spin_unlock_irqrestore(lock: &cc->write_thread_lock, flags); |
2036 | } |
2037 | |
2038 | static 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(bio: 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 | |
2059 | static 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(x: &ctx->restart); |
2070 | |
2071 | r = crypt_convert(cc, ctx: &io->ctx, atomic: true, reset_pending: false); |
2072 | if (r) |
2073 | io->error = r; |
2074 | crypt_finished = atomic_dec_and_test(v: &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, async: 0); |
2084 | io->sector = sector; |
2085 | } |
2086 | |
2087 | crypt_dec_pending(io); |
2088 | } |
2089 | |
2090 | static 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, bio_in: io->base_bio, sector); |
2104 | |
2105 | clone = crypt_alloc_buffer(io, size: 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(dst: clone, src: 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), reset_pending: 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(wq: cc->crypt_queue, work: &io->work); |
2133 | return; |
2134 | } |
2135 | if (r) |
2136 | io->error = r; |
2137 | crypt_finished = atomic_dec_and_test(v: &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, async: 0); |
2147 | io->sector = sector; |
2148 | } |
2149 | |
2150 | dec: |
2151 | crypt_dec_pending(io); |
2152 | } |
2153 | |
2154 | static 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(dst: io->base_bio, src: io->ctx.bio_in); |
2160 | } |
2161 | crypt_free_buffer_pages(cc: io->cc, clone: io->ctx.bio_in); |
2162 | bio_put(io->ctx.bio_in); |
2163 | } |
2164 | crypt_dec_pending(io); |
2165 | } |
2166 | |
2167 | static 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(x: &io->ctx.restart); |
2175 | |
2176 | r = crypt_convert(cc, ctx: &io->ctx, atomic: true, reset_pending: false); |
2177 | if (r) |
2178 | io->error = r; |
2179 | |
2180 | if (atomic_dec_and_test(v: &io->ctx.cc_pending)) |
2181 | kcryptd_crypt_read_done(io); |
2182 | |
2183 | crypt_dec_pending(io); |
2184 | } |
2185 | |
2186 | static 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, ctx: &io->ctx, |
2196 | test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), reset_pending: true); |
2197 | } else { |
2198 | crypt_convert_init(cc, ctx: &io->ctx, bio_out: io->base_bio, bio_in: io->base_bio, |
2199 | sector: io->sector); |
2200 | |
2201 | r = crypt_convert(cc, ctx: &io->ctx, |
2202 | test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), reset_pending: 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(wq: cc->crypt_queue, work: &io->work); |
2211 | return; |
2212 | } |
2213 | if (r) |
2214 | io->error = r; |
2215 | |
2216 | if (atomic_dec_and_test(v: &io->ctx.cc_pending)) |
2217 | kcryptd_crypt_read_done(io); |
2218 | |
2219 | crypt_dec_pending(io); |
2220 | } |
2221 | |
2222 | static 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, op: "integrity-aead" , |
2250 | bio: ctx->bio_in, sector: s, result: 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: req_of_dmreq(cc, dmreq), base_bio: io->base_bio); |
2257 | |
2258 | if (!atomic_dec_and_test(v: &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, async: 1); |
2276 | } |
2277 | |
2278 | static 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 | |
2288 | static 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 | INIT_WORK(&io->work, kcryptd_crypt); |
2301 | queue_work(wq: system_bh_wq, work: &io->work); |
2302 | return; |
2303 | } else { |
2304 | kcryptd_crypt(work: &io->work); |
2305 | return; |
2306 | } |
2307 | } |
2308 | |
2309 | INIT_WORK(&io->work, kcryptd_crypt); |
2310 | queue_work(wq: cc->crypt_queue, work: &io->work); |
2311 | } |
2312 | |
2313 | static void crypt_free_tfms_aead(struct crypt_config *cc) |
2314 | { |
2315 | if (!cc->cipher_tfm.tfms_aead) |
2316 | return; |
2317 | |
2318 | if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(ptr: cc->cipher_tfm.tfms_aead[0])) { |
2319 | crypto_free_aead(tfm: cc->cipher_tfm.tfms_aead[0]); |
2320 | cc->cipher_tfm.tfms_aead[0] = NULL; |
2321 | } |
2322 | |
2323 | kfree(objp: cc->cipher_tfm.tfms_aead); |
2324 | cc->cipher_tfm.tfms_aead = NULL; |
2325 | } |
2326 | |
2327 | static void crypt_free_tfms_skcipher(struct crypt_config *cc) |
2328 | { |
2329 | unsigned int i; |
2330 | |
2331 | if (!cc->cipher_tfm.tfms) |
2332 | return; |
2333 | |
2334 | for (i = 0; i < cc->tfms_count; i++) |
2335 | if (cc->cipher_tfm.tfms[i] && !IS_ERR(ptr: cc->cipher_tfm.tfms[i])) { |
2336 | crypto_free_skcipher(tfm: cc->cipher_tfm.tfms[i]); |
2337 | cc->cipher_tfm.tfms[i] = NULL; |
2338 | } |
2339 | |
2340 | kfree(objp: cc->cipher_tfm.tfms); |
2341 | cc->cipher_tfm.tfms = NULL; |
2342 | } |
2343 | |
2344 | static void crypt_free_tfms(struct crypt_config *cc) |
2345 | { |
2346 | if (crypt_integrity_aead(cc)) |
2347 | crypt_free_tfms_aead(cc); |
2348 | else |
2349 | crypt_free_tfms_skcipher(cc); |
2350 | } |
2351 | |
2352 | static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode) |
2353 | { |
2354 | unsigned int i; |
2355 | int err; |
2356 | |
2357 | cc->cipher_tfm.tfms = kcalloc(n: cc->tfms_count, |
2358 | size: sizeof(struct crypto_skcipher *), |
2359 | GFP_KERNEL); |
2360 | if (!cc->cipher_tfm.tfms) |
2361 | return -ENOMEM; |
2362 | |
2363 | for (i = 0; i < cc->tfms_count; i++) { |
2364 | cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(alg_name: ciphermode, type: 0, |
2365 | CRYPTO_ALG_ALLOCATES_MEMORY); |
2366 | if (IS_ERR(ptr: cc->cipher_tfm.tfms[i])) { |
2367 | err = PTR_ERR(ptr: cc->cipher_tfm.tfms[i]); |
2368 | crypt_free_tfms(cc); |
2369 | return err; |
2370 | } |
2371 | } |
2372 | |
2373 | /* |
2374 | * dm-crypt performance can vary greatly depending on which crypto |
2375 | * algorithm implementation is used. Help people debug performance |
2376 | * problems by logging the ->cra_driver_name. |
2377 | */ |
2378 | DMDEBUG_LIMIT("%s using implementation \"%s\"" , ciphermode, |
2379 | crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name); |
2380 | return 0; |
2381 | } |
2382 | |
2383 | static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode) |
2384 | { |
2385 | int err; |
2386 | |
2387 | cc->cipher_tfm.tfms = kmalloc(size: sizeof(struct crypto_aead *), GFP_KERNEL); |
2388 | if (!cc->cipher_tfm.tfms) |
2389 | return -ENOMEM; |
2390 | |
2391 | cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(alg_name: ciphermode, type: 0, |
2392 | CRYPTO_ALG_ALLOCATES_MEMORY); |
2393 | if (IS_ERR(ptr: cc->cipher_tfm.tfms_aead[0])) { |
2394 | err = PTR_ERR(ptr: cc->cipher_tfm.tfms_aead[0]); |
2395 | crypt_free_tfms(cc); |
2396 | return err; |
2397 | } |
2398 | |
2399 | DMDEBUG_LIMIT("%s using implementation \"%s\"" , ciphermode, |
2400 | crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name); |
2401 | return 0; |
2402 | } |
2403 | |
2404 | static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode) |
2405 | { |
2406 | if (crypt_integrity_aead(cc)) |
2407 | return crypt_alloc_tfms_aead(cc, ciphermode); |
2408 | else |
2409 | return crypt_alloc_tfms_skcipher(cc, ciphermode); |
2410 | } |
2411 | |
2412 | static unsigned int crypt_subkey_size(struct crypt_config *cc) |
2413 | { |
2414 | return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count); |
2415 | } |
2416 | |
2417 | static unsigned int crypt_authenckey_size(struct crypt_config *cc) |
2418 | { |
2419 | return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param)); |
2420 | } |
2421 | |
2422 | /* |
2423 | * If AEAD is composed like authenc(hmac(sha256),xts(aes)), |
2424 | * the key must be for some reason in special format. |
2425 | * This funcion converts cc->key to this special format. |
2426 | */ |
2427 | static void crypt_copy_authenckey(char *p, const void *key, |
2428 | unsigned int enckeylen, unsigned int authkeylen) |
2429 | { |
2430 | struct crypto_authenc_key_param *param; |
2431 | struct rtattr *rta; |
2432 | |
2433 | rta = (struct rtattr *)p; |
2434 | param = RTA_DATA(rta); |
2435 | param->enckeylen = cpu_to_be32(enckeylen); |
2436 | rta->rta_len = RTA_LENGTH(sizeof(*param)); |
2437 | rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM; |
2438 | p += RTA_SPACE(sizeof(*param)); |
2439 | memcpy(p, key + enckeylen, authkeylen); |
2440 | p += authkeylen; |
2441 | memcpy(p, key, enckeylen); |
2442 | } |
2443 | |
2444 | static int crypt_setkey(struct crypt_config *cc) |
2445 | { |
2446 | unsigned int subkey_size; |
2447 | int err = 0, i, r; |
2448 | |
2449 | /* Ignore extra keys (which are used for IV etc) */ |
2450 | subkey_size = crypt_subkey_size(cc); |
2451 | |
2452 | if (crypt_integrity_hmac(cc)) { |
2453 | if (subkey_size < cc->key_mac_size) |
2454 | return -EINVAL; |
2455 | |
2456 | crypt_copy_authenckey(p: cc->authenc_key, key: cc->key, |
2457 | enckeylen: subkey_size - cc->key_mac_size, |
2458 | authkeylen: cc->key_mac_size); |
2459 | } |
2460 | |
2461 | for (i = 0; i < cc->tfms_count; i++) { |
2462 | if (crypt_integrity_hmac(cc)) |
2463 | r = crypto_aead_setkey(tfm: cc->cipher_tfm.tfms_aead[i], |
2464 | key: cc->authenc_key, keylen: crypt_authenckey_size(cc)); |
2465 | else if (crypt_integrity_aead(cc)) |
2466 | r = crypto_aead_setkey(tfm: cc->cipher_tfm.tfms_aead[i], |
2467 | key: cc->key + (i * subkey_size), |
2468 | keylen: subkey_size); |
2469 | else |
2470 | r = crypto_skcipher_setkey(tfm: cc->cipher_tfm.tfms[i], |
2471 | key: cc->key + (i * subkey_size), |
2472 | keylen: subkey_size); |
2473 | if (r) |
2474 | err = r; |
2475 | } |
2476 | |
2477 | if (crypt_integrity_hmac(cc)) |
2478 | memzero_explicit(s: cc->authenc_key, count: crypt_authenckey_size(cc)); |
2479 | |
2480 | return err; |
2481 | } |
2482 | |
2483 | #ifdef CONFIG_KEYS |
2484 | |
2485 | static bool contains_whitespace(const char *str) |
2486 | { |
2487 | while (*str) |
2488 | if (isspace(*str++)) |
2489 | return true; |
2490 | return false; |
2491 | } |
2492 | |
2493 | static int set_key_user(struct crypt_config *cc, struct key *key) |
2494 | { |
2495 | const struct user_key_payload *ukp; |
2496 | |
2497 | ukp = user_key_payload_locked(key); |
2498 | if (!ukp) |
2499 | return -EKEYREVOKED; |
2500 | |
2501 | if (cc->key_size != ukp->datalen) |
2502 | return -EINVAL; |
2503 | |
2504 | memcpy(cc->key, ukp->data, cc->key_size); |
2505 | |
2506 | return 0; |
2507 | } |
2508 | |
2509 | static int set_key_encrypted(struct crypt_config *cc, struct key *key) |
2510 | { |
2511 | const struct encrypted_key_payload *ekp; |
2512 | |
2513 | ekp = key->payload.data[0]; |
2514 | if (!ekp) |
2515 | return -EKEYREVOKED; |
2516 | |
2517 | if (cc->key_size != ekp->decrypted_datalen) |
2518 | return -EINVAL; |
2519 | |
2520 | memcpy(cc->key, ekp->decrypted_data, cc->key_size); |
2521 | |
2522 | return 0; |
2523 | } |
2524 | |
2525 | static int set_key_trusted(struct crypt_config *cc, struct key *key) |
2526 | { |
2527 | const struct trusted_key_payload *tkp; |
2528 | |
2529 | tkp = key->payload.data[0]; |
2530 | if (!tkp) |
2531 | return -EKEYREVOKED; |
2532 | |
2533 | if (cc->key_size != tkp->key_len) |
2534 | return -EINVAL; |
2535 | |
2536 | memcpy(cc->key, tkp->key, cc->key_size); |
2537 | |
2538 | return 0; |
2539 | } |
2540 | |
2541 | static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string) |
2542 | { |
2543 | char *new_key_string, *key_desc; |
2544 | int ret; |
2545 | struct key_type *type; |
2546 | struct key *key; |
2547 | int (*set_key)(struct crypt_config *cc, struct key *key); |
2548 | |
2549 | /* |
2550 | * Reject key_string with whitespace. dm core currently lacks code for |
2551 | * proper whitespace escaping in arguments on DM_TABLE_STATUS path. |
2552 | */ |
2553 | if (contains_whitespace(str: key_string)) { |
2554 | DMERR("whitespace chars not allowed in key string" ); |
2555 | return -EINVAL; |
2556 | } |
2557 | |
2558 | /* look for next ':' separating key_type from key_description */ |
2559 | key_desc = strchr(key_string, ':'); |
2560 | if (!key_desc || key_desc == key_string || !strlen(key_desc + 1)) |
2561 | return -EINVAL; |
2562 | |
2563 | if (!strncmp(key_string, "logon:" , key_desc - key_string + 1)) { |
2564 | type = &key_type_logon; |
2565 | set_key = set_key_user; |
2566 | } else if (!strncmp(key_string, "user:" , key_desc - key_string + 1)) { |
2567 | type = &key_type_user; |
2568 | set_key = set_key_user; |
2569 | } else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) && |
2570 | !strncmp(key_string, "encrypted:" , key_desc - key_string + 1)) { |
2571 | type = &key_type_encrypted; |
2572 | set_key = set_key_encrypted; |
2573 | } else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) && |
2574 | !strncmp(key_string, "trusted:" , key_desc - key_string + 1)) { |
2575 | type = &key_type_trusted; |
2576 | set_key = set_key_trusted; |
2577 | } else { |
2578 | return -EINVAL; |
2579 | } |
2580 | |
2581 | new_key_string = kstrdup(s: key_string, GFP_KERNEL); |
2582 | if (!new_key_string) |
2583 | return -ENOMEM; |
2584 | |
2585 | key = request_key(type, description: key_desc + 1, NULL); |
2586 | if (IS_ERR(ptr: key)) { |
2587 | kfree_sensitive(objp: new_key_string); |
2588 | return PTR_ERR(ptr: key); |
2589 | } |
2590 | |
2591 | down_read(sem: &key->sem); |
2592 | |
2593 | ret = set_key(cc, key); |
2594 | if (ret < 0) { |
2595 | up_read(sem: &key->sem); |
2596 | key_put(key); |
2597 | kfree_sensitive(objp: new_key_string); |
2598 | return ret; |
2599 | } |
2600 | |
2601 | up_read(sem: &key->sem); |
2602 | key_put(key); |
2603 | |
2604 | /* clear the flag since following operations may invalidate previously valid key */ |
2605 | clear_bit(nr: DM_CRYPT_KEY_VALID, addr: &cc->flags); |
2606 | |
2607 | ret = crypt_setkey(cc); |
2608 | |
2609 | if (!ret) { |
2610 | set_bit(nr: DM_CRYPT_KEY_VALID, addr: &cc->flags); |
2611 | kfree_sensitive(objp: cc->key_string); |
2612 | cc->key_string = new_key_string; |
2613 | } else |
2614 | kfree_sensitive(objp: new_key_string); |
2615 | |
2616 | return ret; |
2617 | } |
2618 | |
2619 | static int get_key_size(char **key_string) |
2620 | { |
2621 | char *colon, dummy; |
2622 | int ret; |
2623 | |
2624 | if (*key_string[0] != ':') |
2625 | return strlen(*key_string) >> 1; |
2626 | |
2627 | /* look for next ':' in key string */ |
2628 | colon = strpbrk(*key_string + 1, ":" ); |
2629 | if (!colon) |
2630 | return -EINVAL; |
2631 | |
2632 | if (sscanf(*key_string + 1, "%u%c" , &ret, &dummy) != 2 || dummy != ':') |
2633 | return -EINVAL; |
2634 | |
2635 | *key_string = colon; |
2636 | |
2637 | /* remaining key string should be :<logon|user>:<key_desc> */ |
2638 | |
2639 | return ret; |
2640 | } |
2641 | |
2642 | #else |
2643 | |
2644 | static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string) |
2645 | { |
2646 | return -EINVAL; |
2647 | } |
2648 | |
2649 | static int get_key_size(char **key_string) |
2650 | { |
2651 | return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1); |
2652 | } |
2653 | |
2654 | #endif /* CONFIG_KEYS */ |
2655 | |
2656 | static int crypt_set_key(struct crypt_config *cc, char *key) |
2657 | { |
2658 | int r = -EINVAL; |
2659 | int key_string_len = strlen(key); |
2660 | |
2661 | /* Hyphen (which gives a key_size of zero) means there is no key. */ |
2662 | if (!cc->key_size && strcmp(key, "-" )) |
2663 | goto out; |
2664 | |
2665 | /* ':' means the key is in kernel keyring, short-circuit normal key processing */ |
2666 | if (key[0] == ':') { |
2667 | r = crypt_set_keyring_key(cc, key_string: key + 1); |
2668 | goto out; |
2669 | } |
2670 | |
2671 | /* clear the flag since following operations may invalidate previously valid key */ |
2672 | clear_bit(nr: DM_CRYPT_KEY_VALID, addr: &cc->flags); |
2673 | |
2674 | /* wipe references to any kernel keyring key */ |
2675 | kfree_sensitive(objp: cc->key_string); |
2676 | cc->key_string = NULL; |
2677 | |
2678 | /* Decode key from its hex representation. */ |
2679 | if (cc->key_size && hex2bin(dst: cc->key, src: key, count: cc->key_size) < 0) |
2680 | goto out; |
2681 | |
2682 | r = crypt_setkey(cc); |
2683 | if (!r) |
2684 | set_bit(nr: DM_CRYPT_KEY_VALID, addr: &cc->flags); |
2685 | |
2686 | out: |
2687 | /* Hex key string not needed after here, so wipe it. */ |
2688 | memset(key, '0', key_string_len); |
2689 | |
2690 | return r; |
2691 | } |
2692 | |
2693 | static int crypt_wipe_key(struct crypt_config *cc) |
2694 | { |
2695 | int r; |
2696 | |
2697 | clear_bit(nr: DM_CRYPT_KEY_VALID, addr: &cc->flags); |
2698 | get_random_bytes(buf: &cc->key, len: cc->key_size); |
2699 | |
2700 | /* Wipe IV private keys */ |
2701 | if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) { |
2702 | r = cc->iv_gen_ops->wipe(cc); |
2703 | if (r) |
2704 | return r; |
2705 | } |
2706 | |
2707 | kfree_sensitive(objp: cc->key_string); |
2708 | cc->key_string = NULL; |
2709 | r = crypt_setkey(cc); |
2710 | memset(&cc->key, 0, cc->key_size * sizeof(u8)); |
2711 | |
2712 | return r; |
2713 | } |
2714 | |
2715 | static void crypt_calculate_pages_per_client(void) |
2716 | { |
2717 | unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100; |
2718 | |
2719 | if (!dm_crypt_clients_n) |
2720 | return; |
2721 | |
2722 | pages /= dm_crypt_clients_n; |
2723 | if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT) |
2724 | pages = DM_CRYPT_MIN_PAGES_PER_CLIENT; |
2725 | dm_crypt_pages_per_client = pages; |
2726 | } |
2727 | |
2728 | static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data) |
2729 | { |
2730 | struct crypt_config *cc = pool_data; |
2731 | struct page *page; |
2732 | |
2733 | /* |
2734 | * Note, percpu_counter_read_positive() may over (and under) estimate |
2735 | * the current usage by at most (batch - 1) * num_online_cpus() pages, |
2736 | * but avoids potential spinlock contention of an exact result. |
2737 | */ |
2738 | if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) && |
2739 | likely(gfp_mask & __GFP_NORETRY)) |
2740 | return NULL; |
2741 | |
2742 | page = alloc_page(gfp_mask); |
2743 | if (likely(page != NULL)) |
2744 | percpu_counter_add(fbc: &cc->n_allocated_pages, amount: 1); |
2745 | |
2746 | return page; |
2747 | } |
2748 | |
2749 | static void crypt_page_free(void *page, void *pool_data) |
2750 | { |
2751 | struct crypt_config *cc = pool_data; |
2752 | |
2753 | __free_page(page); |
2754 | percpu_counter_sub(fbc: &cc->n_allocated_pages, amount: 1); |
2755 | } |
2756 | |
2757 | static void crypt_dtr(struct dm_target *ti) |
2758 | { |
2759 | struct crypt_config *cc = ti->private; |
2760 | |
2761 | ti->private = NULL; |
2762 | |
2763 | if (!cc) |
2764 | return; |
2765 | |
2766 | if (cc->write_thread) |
2767 | kthread_stop(k: cc->write_thread); |
2768 | |
2769 | if (cc->io_queue) |
2770 | destroy_workqueue(wq: cc->io_queue); |
2771 | if (cc->crypt_queue) |
2772 | destroy_workqueue(wq: cc->crypt_queue); |
2773 | |
2774 | crypt_free_tfms(cc); |
2775 | |
2776 | bioset_exit(&cc->bs); |
2777 | |
2778 | mempool_exit(pool: &cc->page_pool); |
2779 | mempool_exit(pool: &cc->req_pool); |
2780 | mempool_exit(pool: &cc->tag_pool); |
2781 | |
2782 | WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0); |
2783 | percpu_counter_destroy(fbc: &cc->n_allocated_pages); |
2784 | |
2785 | if (cc->iv_gen_ops && cc->iv_gen_ops->dtr) |
2786 | cc->iv_gen_ops->dtr(cc); |
2787 | |
2788 | if (cc->dev) |
2789 | dm_put_device(ti, d: cc->dev); |
2790 | |
2791 | kfree_sensitive(objp: cc->cipher_string); |
2792 | kfree_sensitive(objp: cc->key_string); |
2793 | kfree_sensitive(objp: cc->cipher_auth); |
2794 | kfree_sensitive(objp: cc->authenc_key); |
2795 | |
2796 | mutex_destroy(lock: &cc->bio_alloc_lock); |
2797 | |
2798 | /* Must zero key material before freeing */ |
2799 | kfree_sensitive(objp: cc); |
2800 | |
2801 | spin_lock(lock: &dm_crypt_clients_lock); |
2802 | WARN_ON(!dm_crypt_clients_n); |
2803 | dm_crypt_clients_n--; |
2804 | crypt_calculate_pages_per_client(); |
2805 | spin_unlock(lock: &dm_crypt_clients_lock); |
2806 | |
2807 | dm_audit_log_dtr(DM_MSG_PREFIX, ti, result: 1); |
2808 | } |
2809 | |
2810 | static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode) |
2811 | { |
2812 | struct crypt_config *cc = ti->private; |
2813 | |
2814 | if (crypt_integrity_aead(cc)) |
2815 | cc->iv_size = crypto_aead_ivsize(tfm: any_tfm_aead(cc)); |
2816 | else |
2817 | cc->iv_size = crypto_skcipher_ivsize(tfm: any_tfm(cc)); |
2818 | |
2819 | if (cc->iv_size) |
2820 | /* at least a 64 bit sector number should fit in our buffer */ |
2821 | cc->iv_size = max(cc->iv_size, |
2822 | (unsigned int)(sizeof(u64) / sizeof(u8))); |
2823 | else if (ivmode) { |
2824 | DMWARN("Selected cipher does not support IVs" ); |
2825 | ivmode = NULL; |
2826 | } |
2827 | |
2828 | /* Choose ivmode, see comments at iv code. */ |
2829 | if (ivmode == NULL) |
2830 | cc->iv_gen_ops = NULL; |
2831 | else if (strcmp(ivmode, "plain" ) == 0) |
2832 | cc->iv_gen_ops = &crypt_iv_plain_ops; |
2833 | else if (strcmp(ivmode, "plain64" ) == 0) |
2834 | cc->iv_gen_ops = &crypt_iv_plain64_ops; |
2835 | else if (strcmp(ivmode, "plain64be" ) == 0) |
2836 | cc->iv_gen_ops = &crypt_iv_plain64be_ops; |
2837 | else if (strcmp(ivmode, "essiv" ) == 0) |
2838 | cc->iv_gen_ops = &crypt_iv_essiv_ops; |
2839 | else if (strcmp(ivmode, "benbi" ) == 0) |
2840 | cc->iv_gen_ops = &crypt_iv_benbi_ops; |
2841 | else if (strcmp(ivmode, "null" ) == 0) |
2842 | cc->iv_gen_ops = &crypt_iv_null_ops; |
2843 | else if (strcmp(ivmode, "eboiv" ) == 0) |
2844 | cc->iv_gen_ops = &crypt_iv_eboiv_ops; |
2845 | else if (strcmp(ivmode, "elephant" ) == 0) { |
2846 | cc->iv_gen_ops = &crypt_iv_elephant_ops; |
2847 | cc->key_parts = 2; |
2848 | cc->key_extra_size = cc->key_size / 2; |
2849 | if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE) |
2850 | return -EINVAL; |
2851 | set_bit(nr: CRYPT_ENCRYPT_PREPROCESS, addr: &cc->cipher_flags); |
2852 | } else if (strcmp(ivmode, "lmk" ) == 0) { |
2853 | cc->iv_gen_ops = &crypt_iv_lmk_ops; |
2854 | /* |
2855 | * Version 2 and 3 is recognised according |
2856 | * to length of provided multi-key string. |
2857 | * If present (version 3), last key is used as IV seed. |
2858 | * All keys (including IV seed) are always the same size. |
2859 | */ |
2860 | if (cc->key_size % cc->key_parts) { |
2861 | cc->key_parts++; |
2862 | cc->key_extra_size = cc->key_size / cc->key_parts; |
2863 | } |
2864 | } else if (strcmp(ivmode, "tcw" ) == 0) { |
2865 | cc->iv_gen_ops = &crypt_iv_tcw_ops; |
2866 | cc->key_parts += 2; /* IV + whitening */ |
2867 | cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE; |
2868 | } else if (strcmp(ivmode, "random" ) == 0) { |
2869 | cc->iv_gen_ops = &crypt_iv_random_ops; |
2870 | /* Need storage space in integrity fields. */ |
2871 | cc->integrity_iv_size = cc->iv_size; |
2872 | } else { |
2873 | ti->error = "Invalid IV mode" ; |
2874 | return -EINVAL; |
2875 | } |
2876 | |
2877 | return 0; |
2878 | } |
2879 | |
2880 | /* |
2881 | * Workaround to parse HMAC algorithm from AEAD crypto API spec. |
2882 | * The HMAC is needed to calculate tag size (HMAC digest size). |
2883 | * This should be probably done by crypto-api calls (once available...) |
2884 | */ |
2885 | static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api) |
2886 | { |
2887 | char *start, *end, *mac_alg = NULL; |
2888 | struct crypto_ahash *mac; |
2889 | |
2890 | if (!strstarts(str: cipher_api, prefix: "authenc(" )) |
2891 | return 0; |
2892 | |
2893 | start = strchr(cipher_api, '('); |
2894 | end = strchr(cipher_api, ','); |
2895 | if (!start || !end || ++start > end) |
2896 | return -EINVAL; |
2897 | |
2898 | mac_alg = kmemdup_nul(s: start, len: end - start, GFP_KERNEL); |
2899 | if (!mac_alg) |
2900 | return -ENOMEM; |
2901 | |
2902 | mac = crypto_alloc_ahash(alg_name: mac_alg, type: 0, CRYPTO_ALG_ALLOCATES_MEMORY); |
2903 | kfree(objp: mac_alg); |
2904 | |
2905 | if (IS_ERR(ptr: mac)) |
2906 | return PTR_ERR(ptr: mac); |
2907 | |
2908 | cc->key_mac_size = crypto_ahash_digestsize(tfm: mac); |
2909 | crypto_free_ahash(tfm: mac); |
2910 | |
2911 | cc->authenc_key = kmalloc(size: crypt_authenckey_size(cc), GFP_KERNEL); |
2912 | if (!cc->authenc_key) |
2913 | return -ENOMEM; |
2914 | |
2915 | return 0; |
2916 | } |
2917 | |
2918 | static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key, |
2919 | char **ivmode, char **ivopts) |
2920 | { |
2921 | struct crypt_config *cc = ti->private; |
2922 | char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME]; |
2923 | int ret = -EINVAL; |
2924 | |
2925 | cc->tfms_count = 1; |
2926 | |
2927 | /* |
2928 | * New format (capi: prefix) |
2929 | * capi:cipher_api_spec-iv:ivopts |
2930 | */ |
2931 | tmp = &cipher_in[strlen("capi:" )]; |
2932 | |
2933 | /* Separate IV options if present, it can contain another '-' in hash name */ |
2934 | *ivopts = strrchr(tmp, ':'); |
2935 | if (*ivopts) { |
2936 | **ivopts = '\0'; |
2937 | (*ivopts)++; |
2938 | } |
2939 | /* Parse IV mode */ |
2940 | *ivmode = strrchr(tmp, '-'); |
2941 | if (*ivmode) { |
2942 | **ivmode = '\0'; |
2943 | (*ivmode)++; |
2944 | } |
2945 | /* The rest is crypto API spec */ |
2946 | cipher_api = tmp; |
2947 | |
2948 | /* Alloc AEAD, can be used only in new format. */ |
2949 | if (crypt_integrity_aead(cc)) { |
2950 | ret = crypt_ctr_auth_cipher(cc, cipher_api); |
2951 | if (ret < 0) { |
2952 | ti->error = "Invalid AEAD cipher spec" ; |
2953 | return ret; |
2954 | } |
2955 | } |
2956 | |
2957 | if (*ivmode && !strcmp(*ivmode, "lmk" )) |
2958 | cc->tfms_count = 64; |
2959 | |
2960 | if (*ivmode && !strcmp(*ivmode, "essiv" )) { |
2961 | if (!*ivopts) { |
2962 | ti->error = "Digest algorithm missing for ESSIV mode" ; |
2963 | return -EINVAL; |
2964 | } |
2965 | ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, fmt: "essiv(%s,%s)" , |
2966 | cipher_api, *ivopts); |
2967 | if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) { |
2968 | ti->error = "Cannot allocate cipher string" ; |
2969 | return -ENOMEM; |
2970 | } |
2971 | cipher_api = buf; |
2972 | } |
2973 | |
2974 | cc->key_parts = cc->tfms_count; |
2975 | |
2976 | /* Allocate cipher */ |
2977 | ret = crypt_alloc_tfms(cc, ciphermode: cipher_api); |
2978 | if (ret < 0) { |
2979 | ti->error = "Error allocating crypto tfm" ; |
2980 | return ret; |
2981 | } |
2982 | |
2983 | if (crypt_integrity_aead(cc)) |
2984 | cc->iv_size = crypto_aead_ivsize(tfm: any_tfm_aead(cc)); |
2985 | else |
2986 | cc->iv_size = crypto_skcipher_ivsize(tfm: any_tfm(cc)); |
2987 | |
2988 | return 0; |
2989 | } |
2990 | |
2991 | static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key, |
2992 | char **ivmode, char **ivopts) |
2993 | { |
2994 | struct crypt_config *cc = ti->private; |
2995 | char *tmp, *cipher, *chainmode, *keycount; |
2996 | char *cipher_api = NULL; |
2997 | int ret = -EINVAL; |
2998 | char dummy; |
2999 | |
3000 | if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) { |
3001 | ti->error = "Bad cipher specification" ; |
3002 | return -EINVAL; |
3003 | } |
3004 | |
3005 | /* |
3006 | * Legacy dm-crypt cipher specification |
3007 | * cipher[:keycount]-mode-iv:ivopts |
3008 | */ |
3009 | tmp = cipher_in; |
3010 | keycount = strsep(&tmp, "-" ); |
3011 | cipher = strsep(&keycount, ":" ); |
3012 | |
3013 | if (!keycount) |
3014 | cc->tfms_count = 1; |
3015 | else if (sscanf(keycount, "%u%c" , &cc->tfms_count, &dummy) != 1 || |
3016 | !is_power_of_2(n: cc->tfms_count)) { |
3017 | ti->error = "Bad cipher key count specification" ; |
3018 | return -EINVAL; |
3019 | } |
3020 | cc->key_parts = cc->tfms_count; |
3021 | |
3022 | chainmode = strsep(&tmp, "-" ); |
3023 | *ivmode = strsep(&tmp, ":" ); |
3024 | *ivopts = tmp; |
3025 | |
3026 | /* |
3027 | * For compatibility with the original dm-crypt mapping format, if |
3028 | * only the cipher name is supplied, use cbc-plain. |
3029 | */ |
3030 | if (!chainmode || (!strcmp(chainmode, "plain" ) && !*ivmode)) { |
3031 | chainmode = "cbc" ; |
3032 | *ivmode = "plain" ; |
3033 | } |
3034 | |
3035 | if (strcmp(chainmode, "ecb" ) && !*ivmode) { |
3036 | ti->error = "IV mechanism required" ; |
3037 | return -EINVAL; |
3038 | } |
3039 | |
3040 | cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL); |
3041 | if (!cipher_api) |
3042 | goto bad_mem; |
3043 | |
3044 | if (*ivmode && !strcmp(*ivmode, "essiv" )) { |
3045 | if (!*ivopts) { |
3046 | ti->error = "Digest algorithm missing for ESSIV mode" ; |
3047 | kfree(objp: cipher_api); |
3048 | return -EINVAL; |
3049 | } |
3050 | ret = snprintf(buf: cipher_api, CRYPTO_MAX_ALG_NAME, |
3051 | fmt: "essiv(%s(%s),%s)" , chainmode, cipher, *ivopts); |
3052 | } else { |
3053 | ret = snprintf(buf: cipher_api, CRYPTO_MAX_ALG_NAME, |
3054 | fmt: "%s(%s)" , chainmode, cipher); |
3055 | } |
3056 | if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) { |
3057 | kfree(objp: cipher_api); |
3058 | goto bad_mem; |
3059 | } |
3060 | |
3061 | /* Allocate cipher */ |
3062 | ret = crypt_alloc_tfms(cc, ciphermode: cipher_api); |
3063 | if (ret < 0) { |
3064 | ti->error = "Error allocating crypto tfm" ; |
3065 | kfree(objp: cipher_api); |
3066 | return ret; |
3067 | } |
3068 | kfree(objp: cipher_api); |
3069 | |
3070 | return 0; |
3071 | bad_mem: |
3072 | ti->error = "Cannot allocate cipher strings" ; |
3073 | return -ENOMEM; |
3074 | } |
3075 | |
3076 | static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key) |
3077 | { |
3078 | struct crypt_config *cc = ti->private; |
3079 | char *ivmode = NULL, *ivopts = NULL; |
3080 | int ret; |
3081 | |
3082 | cc->cipher_string = kstrdup(s: cipher_in, GFP_KERNEL); |
3083 | if (!cc->cipher_string) { |
3084 | ti->error = "Cannot allocate cipher strings" ; |
3085 | return -ENOMEM; |
3086 | } |
3087 | |
3088 | if (strstarts(str: cipher_in, prefix: "capi:" )) |
3089 | ret = crypt_ctr_cipher_new(ti, cipher_in, key, ivmode: &ivmode, ivopts: &ivopts); |
3090 | else |
3091 | ret = crypt_ctr_cipher_old(ti, cipher_in, key, ivmode: &ivmode, ivopts: &ivopts); |
3092 | if (ret) |
3093 | return ret; |
3094 | |
3095 | /* Initialize IV */ |
3096 | ret = crypt_ctr_ivmode(ti, ivmode); |
3097 | if (ret < 0) |
3098 | return ret; |
3099 | |
3100 | /* Initialize and set key */ |
3101 | ret = crypt_set_key(cc, key); |
3102 | if (ret < 0) { |
3103 | ti->error = "Error decoding and setting key" ; |
3104 | return ret; |
3105 | } |
3106 | |
3107 | /* Allocate IV */ |
3108 | if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) { |
3109 | ret = cc->iv_gen_ops->ctr(cc, ti, ivopts); |
3110 | if (ret < 0) { |
3111 | ti->error = "Error creating IV" ; |
3112 | return ret; |
3113 | } |
3114 | } |
3115 | |
3116 | /* Initialize IV (set keys for ESSIV etc) */ |
3117 | if (cc->iv_gen_ops && cc->iv_gen_ops->init) { |
3118 | ret = cc->iv_gen_ops->init(cc); |
3119 | if (ret < 0) { |
3120 | ti->error = "Error initialising IV" ; |
3121 | return ret; |
3122 | } |
3123 | } |
3124 | |
3125 | /* wipe the kernel key payload copy */ |
3126 | if (cc->key_string) |
3127 | memset(cc->key, 0, cc->key_size * sizeof(u8)); |
3128 | |
3129 | return ret; |
3130 | } |
3131 | |
3132 | static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv) |
3133 | { |
3134 | struct crypt_config *cc = ti->private; |
3135 | struct dm_arg_set as; |
3136 | static const struct dm_arg _args[] = { |
3137 | {0, 8, "Invalid number of feature args" }, |
3138 | }; |
3139 | unsigned int opt_params, val; |
3140 | const char *opt_string, *sval; |
3141 | char dummy; |
3142 | int ret; |
3143 | |
3144 | /* Optional parameters */ |
3145 | as.argc = argc; |
3146 | as.argv = argv; |
3147 | |
3148 | ret = dm_read_arg_group(arg: _args, arg_set: &as, num_args: &opt_params, error: &ti->error); |
3149 | if (ret) |
3150 | return ret; |
3151 | |
3152 | while (opt_params--) { |
3153 | opt_string = dm_shift_arg(as: &as); |
3154 | if (!opt_string) { |
3155 | ti->error = "Not enough feature arguments" ; |
3156 | return -EINVAL; |
3157 | } |
3158 | |
3159 | if (!strcasecmp(s1: opt_string, s2: "allow_discards" )) |
3160 | ti->num_discard_bios = 1; |
3161 | |
3162 | else if (!strcasecmp(s1: opt_string, s2: "same_cpu_crypt" )) |
3163 | set_bit(nr: DM_CRYPT_SAME_CPU, addr: &cc->flags); |
3164 | |
3165 | else if (!strcasecmp(s1: opt_string, s2: "submit_from_crypt_cpus" )) |
3166 | set_bit(nr: DM_CRYPT_NO_OFFLOAD, addr: &cc->flags); |
3167 | else if (!strcasecmp(s1: opt_string, s2: "no_read_workqueue" )) |
3168 | set_bit(nr: DM_CRYPT_NO_READ_WORKQUEUE, addr: &cc->flags); |
3169 | else if (!strcasecmp(s1: opt_string, s2: "no_write_workqueue" )) |
3170 | set_bit(nr: DM_CRYPT_NO_WRITE_WORKQUEUE, addr: &cc->flags); |
3171 | else if (sscanf(opt_string, "integrity:%u:" , &val) == 1) { |
3172 | if (val == 0 || val > MAX_TAG_SIZE) { |
3173 | ti->error = "Invalid integrity arguments" ; |
3174 | return -EINVAL; |
3175 | } |
3176 | cc->on_disk_tag_size = val; |
3177 | sval = strchr(opt_string + strlen("integrity:" ), ':') + 1; |
3178 | if (!strcasecmp(s1: sval, s2: "aead" )) { |
3179 | set_bit(nr: CRYPT_MODE_INTEGRITY_AEAD, addr: &cc->cipher_flags); |
3180 | } else if (strcasecmp(s1: sval, s2: "none" )) { |
3181 | ti->error = "Unknown integrity profile" ; |
3182 | return -EINVAL; |
3183 | } |
3184 | |
3185 | cc->cipher_auth = kstrdup(s: sval, GFP_KERNEL); |
3186 | if (!cc->cipher_auth) |
3187 | return -ENOMEM; |
3188 | } else if (sscanf(opt_string, "sector_size:%hu%c" , &cc->sector_size, &dummy) == 1) { |
3189 | if (cc->sector_size < (1 << SECTOR_SHIFT) || |
3190 | cc->sector_size > 4096 || |
3191 | (cc->sector_size & (cc->sector_size - 1))) { |
3192 | ti->error = "Invalid feature value for sector_size" ; |
3193 | return -EINVAL; |
3194 | } |
3195 | if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) { |
3196 | ti->error = "Device size is not multiple of sector_size feature" ; |
3197 | return -EINVAL; |
3198 | } |
3199 | cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT; |
3200 | } else if (!strcasecmp(s1: opt_string, s2: "iv_large_sectors" )) |
3201 | set_bit(nr: CRYPT_IV_LARGE_SECTORS, addr: &cc->cipher_flags); |
3202 | else { |
3203 | ti->error = "Invalid feature arguments" ; |
3204 | return -EINVAL; |
3205 | } |
3206 | } |
3207 | |
3208 | return 0; |
3209 | } |
3210 | |
3211 | #ifdef CONFIG_BLK_DEV_ZONED |
3212 | static int crypt_report_zones(struct dm_target *ti, |
3213 | struct dm_report_zones_args *args, unsigned int nr_zones) |
3214 | { |
3215 | struct crypt_config *cc = ti->private; |
3216 | |
3217 | return dm_report_zones(bdev: cc->dev->bdev, start: cc->start, |
3218 | sector: cc->start + dm_target_offset(ti, args->next_sector), |
3219 | args, nr_zones); |
3220 | } |
3221 | #else |
3222 | #define crypt_report_zones NULL |
3223 | #endif |
3224 | |
3225 | /* |
3226 | * Construct an encryption mapping: |
3227 | * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start> |
3228 | */ |
3229 | static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
3230 | { |
3231 | struct crypt_config *cc; |
3232 | const char *devname = dm_table_device_name(t: ti->table); |
3233 | int key_size; |
3234 | unsigned int align_mask; |
3235 | unsigned long long tmpll; |
3236 | int ret; |
3237 | size_t iv_size_padding, additional_req_size; |
3238 | char dummy; |
3239 | |
3240 | if (argc < 5) { |
3241 | ti->error = "Not enough arguments" ; |
3242 | return -EINVAL; |
3243 | } |
3244 | |
3245 | key_size = get_key_size(key_string: &argv[1]); |
3246 | if (key_size < 0) { |
3247 | ti->error = "Cannot parse key size" ; |
3248 | return -EINVAL; |
3249 | } |
3250 | |
3251 | cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL); |
3252 | if (!cc) { |
3253 | ti->error = "Cannot allocate encryption context" ; |
3254 | return -ENOMEM; |
3255 | } |
3256 | cc->key_size = key_size; |
3257 | cc->sector_size = (1 << SECTOR_SHIFT); |
3258 | cc->sector_shift = 0; |
3259 | |
3260 | ti->private = cc; |
3261 | |
3262 | spin_lock(lock: &dm_crypt_clients_lock); |
3263 | dm_crypt_clients_n++; |
3264 | crypt_calculate_pages_per_client(); |
3265 | spin_unlock(lock: &dm_crypt_clients_lock); |
3266 | |
3267 | ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL); |
3268 | if (ret < 0) |
3269 | goto bad; |
3270 | |
3271 | /* Optional parameters need to be read before cipher constructor */ |
3272 | if (argc > 5) { |
3273 | ret = crypt_ctr_optional(ti, argc: argc - 5, argv: &argv[5]); |
3274 | if (ret) |
3275 | goto bad; |
3276 | } |
3277 | |
3278 | ret = crypt_ctr_cipher(ti, cipher_in: argv[0], key: argv[1]); |
3279 | if (ret < 0) |
3280 | goto bad; |
3281 | |
3282 | if (crypt_integrity_aead(cc)) { |
3283 | cc->dmreq_start = sizeof(struct aead_request); |
3284 | cc->dmreq_start += crypto_aead_reqsize(tfm: any_tfm_aead(cc)); |
3285 | align_mask = crypto_aead_alignmask(tfm: any_tfm_aead(cc)); |
3286 | } else { |
3287 | cc->dmreq_start = sizeof(struct skcipher_request); |
3288 | cc->dmreq_start += crypto_skcipher_reqsize(tfm: any_tfm(cc)); |
3289 | align_mask = crypto_skcipher_alignmask(tfm: any_tfm(cc)); |
3290 | } |
3291 | cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request)); |
3292 | |
3293 | if (align_mask < CRYPTO_MINALIGN) { |
3294 | /* Allocate the padding exactly */ |
3295 | iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request)) |
3296 | & align_mask; |
3297 | } else { |
3298 | /* |
3299 | * If the cipher requires greater alignment than kmalloc |
3300 | * alignment, we don't know the exact position of the |
3301 | * initialization vector. We must assume worst case. |
3302 | */ |
3303 | iv_size_padding = align_mask; |
3304 | } |
3305 | |
3306 | /* ...| IV + padding | original IV | original sec. number | bio tag offset | */ |
3307 | additional_req_size = sizeof(struct dm_crypt_request) + |
3308 | iv_size_padding + cc->iv_size + |
3309 | cc->iv_size + |
3310 | sizeof(uint64_t) + |
3311 | sizeof(unsigned int); |
3312 | |
3313 | ret = mempool_init_kmalloc_pool(pool: &cc->req_pool, MIN_IOS, size: cc->dmreq_start + additional_req_size); |
3314 | if (ret) { |
3315 | ti->error = "Cannot allocate crypt request mempool" ; |
3316 | goto bad; |
3317 | } |
3318 | |
3319 | cc->per_bio_data_size = ti->per_io_data_size = |
3320 | ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size, |
3321 | ARCH_DMA_MINALIGN); |
3322 | |
3323 | ret = mempool_init(pool: &cc->page_pool, BIO_MAX_VECS, alloc_fn: crypt_page_alloc, free_fn: crypt_page_free, pool_data: cc); |
3324 | if (ret) { |
3325 | ti->error = "Cannot allocate page mempool" ; |
3326 | goto bad; |
3327 | } |
3328 | |
3329 | ret = bioset_init(&cc->bs, MIN_IOS, 0, flags: BIOSET_NEED_BVECS); |
3330 | if (ret) { |
3331 | ti->error = "Cannot allocate crypt bioset" ; |
3332 | goto bad; |
3333 | } |
3334 | |
3335 | mutex_init(&cc->bio_alloc_lock); |
3336 | |
3337 | ret = -EINVAL; |
3338 | if ((sscanf(argv[2], "%llu%c" , &tmpll, &dummy) != 1) || |
3339 | (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) { |
3340 | ti->error = "Invalid iv_offset sector" ; |
3341 | goto bad; |
3342 | } |
3343 | cc->iv_offset = tmpll; |
3344 | |
3345 | ret = dm_get_device(ti, path: argv[3], mode: dm_table_get_mode(t: ti->table), result: &cc->dev); |
3346 | if (ret) { |
3347 | ti->error = "Device lookup failed" ; |
3348 | goto bad; |
3349 | } |
3350 | |
3351 | ret = -EINVAL; |
3352 | if (sscanf(argv[4], "%llu%c" , &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) { |
3353 | ti->error = "Invalid device sector" ; |
3354 | goto bad; |
3355 | } |
3356 | cc->start = tmpll; |
3357 | |
3358 | if (bdev_is_zoned(bdev: cc->dev->bdev)) { |
3359 | /* |
3360 | * For zoned block devices, we need to preserve the issuer write |
3361 | * ordering. To do so, disable write workqueues and force inline |
3362 | * encryption completion. |
3363 | */ |
3364 | set_bit(nr: DM_CRYPT_NO_WRITE_WORKQUEUE, addr: &cc->flags); |
3365 | set_bit(nr: DM_CRYPT_WRITE_INLINE, addr: &cc->flags); |
3366 | |
3367 | /* |
3368 | * All zone append writes to a zone of a zoned block device will |
3369 | * have the same BIO sector, the start of the zone. When the |
3370 | * cypher IV mode uses sector values, all data targeting a |
3371 | * zone will be encrypted using the first sector numbers of the |
3372 | * zone. This will not result in write errors but will |
3373 | * cause most reads to fail as reads will use the sector values |
3374 | * for the actual data locations, resulting in IV mismatch. |
3375 | * To avoid this problem, ask DM core to emulate zone append |
3376 | * operations with regular writes. |
3377 | */ |
3378 | DMDEBUG("Zone append operations will be emulated" ); |
3379 | ti->emulate_zone_append = true; |
3380 | } |
3381 | |
3382 | if (crypt_integrity_aead(cc) || cc->integrity_iv_size) { |
3383 | ret = crypt_integrity_ctr(cc, ti); |
3384 | if (ret) |
3385 | goto bad; |
3386 | |
3387 | cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size; |
3388 | if (!cc->tag_pool_max_sectors) |
3389 | cc->tag_pool_max_sectors = 1; |
3390 | |
3391 | ret = mempool_init_kmalloc_pool(pool: &cc->tag_pool, MIN_IOS, |
3392 | size: cc->tag_pool_max_sectors * cc->on_disk_tag_size); |
3393 | if (ret) { |
3394 | ti->error = "Cannot allocate integrity tags mempool" ; |
3395 | goto bad; |
3396 | } |
3397 | |
3398 | cc->tag_pool_max_sectors <<= cc->sector_shift; |
3399 | } |
3400 | |
3401 | ret = -ENOMEM; |
3402 | cc->io_queue = alloc_workqueue(fmt: "kcryptd_io/%s" , flags: WQ_MEM_RECLAIM, max_active: 1, devname); |
3403 | if (!cc->io_queue) { |
3404 | ti->error = "Couldn't create kcryptd io queue" ; |
3405 | goto bad; |
3406 | } |
3407 | |
3408 | if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) |
3409 | cc->crypt_queue = alloc_workqueue(fmt: "kcryptd/%s" , flags: WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM, |
3410 | max_active: 1, devname); |
3411 | else |
3412 | cc->crypt_queue = alloc_workqueue(fmt: "kcryptd/%s" , |
3413 | flags: WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND, |
3414 | max_active: num_online_cpus(), devname); |
3415 | if (!cc->crypt_queue) { |
3416 | ti->error = "Couldn't create kcryptd queue" ; |
3417 | goto bad; |
3418 | } |
3419 | |
3420 | spin_lock_init(&cc->write_thread_lock); |
3421 | cc->write_tree = RB_ROOT; |
3422 | |
3423 | cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s" , devname); |
3424 | if (IS_ERR(ptr: cc->write_thread)) { |
3425 | ret = PTR_ERR(ptr: cc->write_thread); |
3426 | cc->write_thread = NULL; |
3427 | ti->error = "Couldn't spawn write thread" ; |
3428 | goto bad; |
3429 | } |
3430 | |
3431 | ti->num_flush_bios = 1; |
3432 | ti->limit_swap_bios = true; |
3433 | ti->accounts_remapped_io = true; |
3434 | |
3435 | dm_audit_log_ctr(DM_MSG_PREFIX, ti, result: 1); |
3436 | return 0; |
3437 | |
3438 | bad: |
3439 | dm_audit_log_ctr(DM_MSG_PREFIX, ti, result: 0); |
3440 | crypt_dtr(ti); |
3441 | return ret; |
3442 | } |
3443 | |
3444 | static int crypt_map(struct dm_target *ti, struct bio *bio) |
3445 | { |
3446 | struct dm_crypt_io *io; |
3447 | struct crypt_config *cc = ti->private; |
3448 | |
3449 | /* |
3450 | * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues. |
3451 | * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight |
3452 | * - for REQ_OP_DISCARD caller must use flush if IO ordering matters |
3453 | */ |
3454 | if (unlikely(bio->bi_opf & REQ_PREFLUSH || |
3455 | bio_op(bio) == REQ_OP_DISCARD)) { |
3456 | bio_set_dev(bio, bdev: cc->dev->bdev); |
3457 | if (bio_sectors(bio)) |
3458 | bio->bi_iter.bi_sector = cc->start + |
3459 | dm_target_offset(ti, bio->bi_iter.bi_sector); |
3460 | return DM_MAPIO_REMAPPED; |
3461 | } |
3462 | |
3463 | /* |
3464 | * Check if bio is too large, split as needed. |
3465 | */ |
3466 | if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) && |
3467 | (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size)) |
3468 | dm_accept_partial_bio(bio, n_sectors: ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT)); |
3469 | |
3470 | /* |
3471 | * Ensure that bio is a multiple of internal sector encryption size |
3472 | * and is aligned to this size as defined in IO hints. |
3473 | */ |
3474 | if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0)) |
3475 | return DM_MAPIO_KILL; |
3476 | |
3477 | if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1))) |
3478 | return DM_MAPIO_KILL; |
3479 | |
3480 | io = dm_per_bio_data(bio, data_size: cc->per_bio_data_size); |
3481 | crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector)); |
3482 | |
3483 | if (cc->on_disk_tag_size) { |
3484 | unsigned int tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift); |
3485 | |
3486 | if (unlikely(tag_len > KMALLOC_MAX_SIZE)) |
3487 | io->integrity_metadata = NULL; |
3488 | else |
3489 | io->integrity_metadata = kmalloc(size: tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN); |
3490 | |
3491 | if (unlikely(!io->integrity_metadata)) { |
3492 | if (bio_sectors(bio) > cc->tag_pool_max_sectors) |
3493 | dm_accept_partial_bio(bio, n_sectors: cc->tag_pool_max_sectors); |
3494 | io->integrity_metadata = mempool_alloc(pool: &cc->tag_pool, GFP_NOIO); |
3495 | io->integrity_metadata_from_pool = true; |
3496 | } |
3497 | } |
3498 | |
3499 | if (crypt_integrity_aead(cc)) |
3500 | io->ctx.r.req_aead = (struct aead_request *)(io + 1); |
3501 | else |
3502 | io->ctx.r.req = (struct skcipher_request *)(io + 1); |
3503 | |
3504 | if (bio_data_dir(io->base_bio) == READ) { |
3505 | if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP)) |
3506 | kcryptd_queue_read(io); |
3507 | } else |
3508 | kcryptd_queue_crypt(io); |
3509 | |
3510 | return DM_MAPIO_SUBMITTED; |
3511 | } |
3512 | |
3513 | static char hex2asc(unsigned char c) |
3514 | { |
3515 | return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27); |
3516 | } |
3517 | |
3518 | static void crypt_status(struct dm_target *ti, status_type_t type, |
3519 | unsigned int status_flags, char *result, unsigned int maxlen) |
3520 | { |
3521 | struct crypt_config *cc = ti->private; |
3522 | unsigned int i, sz = 0; |
3523 | int num_feature_args = 0; |
3524 | |
3525 | switch (type) { |
3526 | case STATUSTYPE_INFO: |
3527 | result[0] = '\0'; |
3528 | break; |
3529 | |
3530 | case STATUSTYPE_TABLE: |
3531 | DMEMIT("%s " , cc->cipher_string); |
3532 | |
3533 | if (cc->key_size > 0) { |
3534 | if (cc->key_string) |
3535 | DMEMIT(":%u:%s" , cc->key_size, cc->key_string); |
3536 | else { |
3537 | for (i = 0; i < cc->key_size; i++) { |
3538 | DMEMIT("%c%c" , hex2asc(cc->key[i] >> 4), |
3539 | hex2asc(cc->key[i] & 0xf)); |
3540 | } |
3541 | } |
3542 | } else |
3543 | DMEMIT("-" ); |
3544 | |
3545 | DMEMIT(" %llu %s %llu" , (unsigned long long)cc->iv_offset, |
3546 | cc->dev->name, (unsigned long long)cc->start); |
3547 | |
3548 | num_feature_args += !!ti->num_discard_bios; |
3549 | num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags); |
3550 | num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags); |
3551 | num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags); |
3552 | num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags); |
3553 | num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT); |
3554 | num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags); |
3555 | if (cc->on_disk_tag_size) |
3556 | num_feature_args++; |
3557 | if (num_feature_args) { |
3558 | DMEMIT(" %d" , num_feature_args); |
3559 | if (ti->num_discard_bios) |
3560 | DMEMIT(" allow_discards" ); |
3561 | if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) |
3562 | DMEMIT(" same_cpu_crypt" ); |
3563 | if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) |
3564 | DMEMIT(" submit_from_crypt_cpus" ); |
3565 | if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) |
3566 | DMEMIT(" no_read_workqueue" ); |
3567 | if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) |
3568 | DMEMIT(" no_write_workqueue" ); |
3569 | if (cc->on_disk_tag_size) |
3570 | DMEMIT(" integrity:%u:%s" , cc->on_disk_tag_size, cc->cipher_auth); |
3571 | if (cc->sector_size != (1 << SECTOR_SHIFT)) |
3572 | DMEMIT(" sector_size:%d" , cc->sector_size); |
3573 | if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags)) |
3574 | DMEMIT(" iv_large_sectors" ); |
3575 | } |
3576 | break; |
3577 | |
3578 | case STATUSTYPE_IMA: |
3579 | DMEMIT_TARGET_NAME_VERSION(ti->type); |
3580 | DMEMIT(",allow_discards=%c" , ti->num_discard_bios ? 'y' : 'n'); |
3581 | DMEMIT(",same_cpu_crypt=%c" , test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n'); |
3582 | DMEMIT(",submit_from_crypt_cpus=%c" , test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ? |
3583 | 'y' : 'n'); |
3584 | DMEMIT(",no_read_workqueue=%c" , test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ? |
3585 | 'y' : 'n'); |
3586 | DMEMIT(",no_write_workqueue=%c" , test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ? |
3587 | 'y' : 'n'); |
3588 | DMEMIT(",iv_large_sectors=%c" , test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ? |
3589 | 'y' : 'n'); |
3590 | |
3591 | if (cc->on_disk_tag_size) |
3592 | DMEMIT(",integrity_tag_size=%u,cipher_auth=%s" , |
3593 | cc->on_disk_tag_size, cc->cipher_auth); |
3594 | if (cc->sector_size != (1 << SECTOR_SHIFT)) |
3595 | DMEMIT(",sector_size=%d" , cc->sector_size); |
3596 | if (cc->cipher_string) |
3597 | DMEMIT(",cipher_string=%s" , cc->cipher_string); |
3598 | |
3599 | DMEMIT(",key_size=%u" , cc->key_size); |
3600 | DMEMIT(",key_parts=%u" , cc->key_parts); |
3601 | DMEMIT(",key_extra_size=%u" , cc->key_extra_size); |
3602 | DMEMIT(",key_mac_size=%u" , cc->key_mac_size); |
3603 | DMEMIT(";" ); |
3604 | break; |
3605 | } |
3606 | } |
3607 | |
3608 | static void crypt_postsuspend(struct dm_target *ti) |
3609 | { |
3610 | struct crypt_config *cc = ti->private; |
3611 | |
3612 | set_bit(nr: DM_CRYPT_SUSPENDED, addr: &cc->flags); |
3613 | } |
3614 | |
3615 | static int crypt_preresume(struct dm_target *ti) |
3616 | { |
3617 | struct crypt_config *cc = ti->private; |
3618 | |
3619 | if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) { |
3620 | DMERR("aborting resume - crypt key is not set." ); |
3621 | return -EAGAIN; |
3622 | } |
3623 | |
3624 | return 0; |
3625 | } |
3626 | |
3627 | static void crypt_resume(struct dm_target *ti) |
3628 | { |
3629 | struct crypt_config *cc = ti->private; |
3630 | |
3631 | clear_bit(nr: DM_CRYPT_SUSPENDED, addr: &cc->flags); |
3632 | } |
3633 | |
3634 | /* Message interface |
3635 | * key set <key> |
3636 | * key wipe |
3637 | */ |
3638 | static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv, |
3639 | char *result, unsigned int maxlen) |
3640 | { |
3641 | struct crypt_config *cc = ti->private; |
3642 | int key_size, ret = -EINVAL; |
3643 | |
3644 | if (argc < 2) |
3645 | goto error; |
3646 | |
3647 | if (!strcasecmp(s1: argv[0], s2: "key" )) { |
3648 | if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) { |
3649 | DMWARN("not suspended during key manipulation." ); |
3650 | return -EINVAL; |
3651 | } |
3652 | if (argc == 3 && !strcasecmp(s1: argv[1], s2: "set" )) { |
3653 | /* The key size may not be changed. */ |
3654 | key_size = get_key_size(key_string: &argv[2]); |
3655 | if (key_size < 0 || cc->key_size != key_size) { |
3656 | memset(argv[2], '0', strlen(argv[2])); |
3657 | return -EINVAL; |
3658 | } |
3659 | |
3660 | ret = crypt_set_key(cc, key: argv[2]); |
3661 | if (ret) |
3662 | return ret; |
3663 | if (cc->iv_gen_ops && cc->iv_gen_ops->init) |
3664 | ret = cc->iv_gen_ops->init(cc); |
3665 | /* wipe the kernel key payload copy */ |
3666 | if (cc->key_string) |
3667 | memset(cc->key, 0, cc->key_size * sizeof(u8)); |
3668 | return ret; |
3669 | } |
3670 | if (argc == 2 && !strcasecmp(s1: argv[1], s2: "wipe" )) |
3671 | return crypt_wipe_key(cc); |
3672 | } |
3673 | |
3674 | error: |
3675 | DMWARN("unrecognised message received." ); |
3676 | return -EINVAL; |
3677 | } |
3678 | |
3679 | static int crypt_iterate_devices(struct dm_target *ti, |
3680 | iterate_devices_callout_fn fn, void *data) |
3681 | { |
3682 | struct crypt_config *cc = ti->private; |
3683 | |
3684 | return fn(ti, cc->dev, cc->start, ti->len, data); |
3685 | } |
3686 | |
3687 | static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits) |
3688 | { |
3689 | struct crypt_config *cc = ti->private; |
3690 | |
3691 | /* |
3692 | * Unfortunate constraint that is required to avoid the potential |
3693 | * for exceeding underlying device's max_segments limits -- due to |
3694 | * crypt_alloc_buffer() possibly allocating pages for the encryption |
3695 | * bio that are not as physically contiguous as the original bio. |
3696 | */ |
3697 | limits->max_segment_size = PAGE_SIZE; |
3698 | |
3699 | limits->logical_block_size = |
3700 | max_t(unsigned int, limits->logical_block_size, cc->sector_size); |
3701 | limits->physical_block_size = |
3702 | max_t(unsigned int, limits->physical_block_size, cc->sector_size); |
3703 | limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size); |
3704 | limits->dma_alignment = limits->logical_block_size - 1; |
3705 | } |
3706 | |
3707 | static struct target_type crypt_target = { |
3708 | .name = "crypt" , |
3709 | .version = {1, 25, 0}, |
3710 | .module = THIS_MODULE, |
3711 | .ctr = crypt_ctr, |
3712 | .dtr = crypt_dtr, |
3713 | .features = DM_TARGET_ZONED_HM, |
3714 | .report_zones = crypt_report_zones, |
3715 | .map = crypt_map, |
3716 | .status = crypt_status, |
3717 | .postsuspend = crypt_postsuspend, |
3718 | .preresume = crypt_preresume, |
3719 | .resume = crypt_resume, |
3720 | .message = crypt_message, |
3721 | .iterate_devices = crypt_iterate_devices, |
3722 | .io_hints = crypt_io_hints, |
3723 | }; |
3724 | module_dm(crypt); |
3725 | |
3726 | MODULE_AUTHOR("Jana Saout <jana@saout.de>" ); |
3727 | MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption" ); |
3728 | MODULE_LICENSE("GPL" ); |
3729 | |