drivers/md/dm-crypt.c

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