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(unsigned int __nocast 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, 0);
 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         if (essiv_tfm == NULL) {
 177                 ti->error = PFX "Error allocating crypto tfm for ESSIV";
 178                 kfree(salt);
 179                 return -EINVAL;
 180         }
 181         if (crypto_tfm_alg_blocksize(essiv_tfm)
 182             != crypto_tfm_alg_ivsize(cc->tfm)) {
 183                 ti->error = PFX "Block size of ESSIV cipher does "
 184                                 "not match IV size of block cipher";
 185                 crypto_free_tfm(essiv_tfm);
 186                 kfree(salt);
 187                 return -EINVAL;
 188         }
 189         if (crypto_cipher_setkey(essiv_tfm, salt, saltsize) < 0) {
 190                 ti->error = PFX "Failed to set key for ESSIV cipher";
 191                 crypto_free_tfm(essiv_tfm);
 192                 kfree(salt);
 193                 return -EINVAL;
 194         }
 195         kfree(salt);
 196
 197         cc->iv_gen_private = (void *)essiv_tfm;
 198         return 0;
 199 }
 200
 201 static void crypt_iv_essiv_dtr(struct crypt_config *cc)
 202 {
 203         crypto_free_tfm((struct crypto_tfm *)cc->iv_gen_private);
 204         cc->iv_gen_private = NULL;
 205 }
 206
 207 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
 208 {
 209         struct scatterlist sg = { NULL, };
 210
 211         memset(iv, 0, cc->iv_size);
 212         *(u64 *)iv = cpu_to_le64(sector);
 213
 214         sg.page = virt_to_page(iv);
 215         sg.offset = offset_in_page(iv);
 216         sg.length = cc->iv_size;
 217         crypto_cipher_encrypt((struct crypto_tfm *)cc->iv_gen_private,
 218                               &sg, &sg, cc->iv_size);
 219
 220         return 0;
 221 }
 222
 223 static struct crypt_iv_operations crypt_iv_plain_ops = {
 224         .generator = crypt_iv_plain_gen
 225 };
 226
 227 static struct crypt_iv_operations crypt_iv_essiv_ops = {
 228         .ctr       = crypt_iv_essiv_ctr,
 229         .dtr       = crypt_iv_essiv_dtr,
 230         .generator = crypt_iv_essiv_gen
 231 };
 232
 233
 234 static inline int
 235 crypt_convert_scatterlist(struct crypt_config *cc, struct scatterlist *out,
 236                           struct scatterlist *in, unsigned int length,
 237                           int write, sector_t sector)
 238 {
 239         u8 iv[cc->iv_size];
 240         int r;
 241
 242         if (cc->iv_gen_ops) {
 243                 r = cc->iv_gen_ops->generator(cc, iv, sector);
 244                 if (r < 0)
 245                         return r;
 246
 247                 if (write)
 248                         r = crypto_cipher_encrypt_iv(cc->tfm, out, in, length, iv);
 249                 else
 250                         r = crypto_cipher_decrypt_iv(cc->tfm, out, in, length, iv);
 251         } else {
 252                 if (write)
 253                         r = crypto_cipher_encrypt(cc->tfm, out, in, length);
 254                 else
 255                         r = crypto_cipher_decrypt(cc->tfm, out, in, length);
 256         }
 257
 258         return r;
 259 }
 260
 261 static void
 262 crypt_convert_init(struct crypt_config *cc, struct convert_context *ctx,
 263                    struct bio *bio_out, struct bio *bio_in,
 264                    sector_t sector, int write)
 265 {
 266         ctx->bio_in = bio_in;
 267         ctx->bio_out = bio_out;
 268         ctx->offset_in = 0;
 269         ctx->offset_out = 0;
 270         ctx->idx_in = bio_in ? bio_in->bi_idx : 0;
 271         ctx->idx_out = bio_out ? bio_out->bi_idx : 0;
 272         ctx->sector = sector + cc->iv_offset;
 273         ctx->write = write;
 274 }
 275
 276 /*
 277  * Encrypt / decrypt data from one bio to another one (can be the same one)
 278  */
 279 static int crypt_convert(struct crypt_config *cc,
 280                          struct convert_context *ctx)
 281 {
 282         int r = 0;
 283
 284         while(ctx->idx_in < ctx->bio_in->bi_vcnt &&
 285               ctx->idx_out < ctx->bio_out->bi_vcnt) {
 286                 struct bio_vec *bv_in = bio_iovec_idx(ctx->bio_in, ctx->idx_in);
 287                 struct bio_vec *bv_out = bio_iovec_idx(ctx->bio_out, ctx->idx_out);
 288                 struct scatterlist sg_in = {
 289                         .page = bv_in->bv_page,
 290                         .offset = bv_in->bv_offset + ctx->offset_in,
 291                         .length = 1 << SECTOR_SHIFT
 292                 };
 293                 struct scatterlist sg_out = {
 294                         .page = bv_out->bv_page,
 295                         .offset = bv_out->bv_offset + ctx->offset_out,
 296                         .length = 1 << SECTOR_SHIFT
 297                 };
 298
 299                 ctx->offset_in += sg_in.length;
 300                 if (ctx->offset_in >= bv_in->bv_len) {
 301                         ctx->offset_in = 0;
 302                         ctx->idx_in++;
 303                 }
 304
 305                 ctx->offset_out += sg_out.length;
 306                 if (ctx->offset_out >= bv_out->bv_len) {
 307                         ctx->offset_out = 0;
 308                         ctx->idx_out++;
 309                 }
 310
 311                 r = crypt_convert_scatterlist(cc, &sg_out, &sg_in, sg_in.length,
 312                                               ctx->write, ctx->sector);
 313                 if (r < 0)
 314                         break;
 315
 316                 ctx->sector++;
 317         }
 318
 319         return r;
 320 }
 321
 322 /*
 323  * Generate a new unfragmented bio with the given size
 324  * This should never violate the device limitations
 325  * May return a smaller bio when running out of pages
 326  */
 327 static struct bio *
 328 crypt_alloc_buffer(struct crypt_config *cc, unsigned int size,
 329                    struct bio *base_bio, unsigned int *bio_vec_idx)
 330 {
 331         struct bio *bio;
 332         unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 333         int gfp_mask = GFP_NOIO | __GFP_HIGHMEM;
 334         unsigned long flags = current->flags;
 335         unsigned int i;
 336
 337         /*
 338          * Tell VM to act less aggressively and fail earlier.
 339          * This is not necessary but increases throughput.
 340          * FIXME: Is this really intelligent?
 341          */
 342         current->flags &= ~PF_MEMALLOC;
 343
 344         if (base_bio)
 345                 bio = bio_clone(base_bio, GFP_NOIO);
 346         else
 347                 bio = bio_alloc(GFP_NOIO, nr_iovecs);
 348         if (!bio) {
 349                 if (flags & PF_MEMALLOC)
 350                         current->flags |= PF_MEMALLOC;
 351                 return NULL;
 352         }
 353
 354         /* if the last bio was not complete, continue where that one ended */
 355         bio->bi_idx = *bio_vec_idx;
 356         bio->bi_vcnt = *bio_vec_idx;
 357         bio->bi_size = 0;
 358         bio->bi_flags &= ~(1 << BIO_SEG_VALID);
 359
 360         /* bio->bi_idx pages have already been allocated */
 361         size -= bio->bi_idx * PAGE_SIZE;
 362
 363         for(i = bio->bi_idx; i < nr_iovecs; i++) {
 364                 struct bio_vec *bv = bio_iovec_idx(bio, i);
 365
 366                 bv->bv_page = mempool_alloc(cc->page_pool, gfp_mask);
 367                 if (!bv->bv_page)
 368                         break;
 369
 370                 /*
 371                  * if additional pages cannot be allocated without waiting,
 372                  * return a partially allocated bio, the caller will then try
 373                  * to allocate additional bios while submitting this partial bio
 374                  */
 375                 if ((i - bio->bi_idx) == (MIN_BIO_PAGES - 1))
 376                         gfp_mask = (gfp_mask | __GFP_NOWARN) & ~__GFP_WAIT;
 377
 378                 bv->bv_offset = 0;
 379                 if (size > PAGE_SIZE)
 380                         bv->bv_len = PAGE_SIZE;
 381                 else
 382                         bv->bv_len = size;
 383
 384                 bio->bi_size += bv->bv_len;
 385                 bio->bi_vcnt++;
 386                 size -= bv->bv_len;
 387         }
 388
 389         if (flags & PF_MEMALLOC)
 390                 current->flags |= PF_MEMALLOC;
 391
 392         if (!bio->bi_size) {
 393                 bio_put(bio);
 394                 return NULL;
 395         }
 396
 397         /*
 398          * Remember the last bio_vec allocated to be able
 399          * to correctly continue after the splitting.
 400          */
 401         *bio_vec_idx = bio->bi_vcnt;
 402
 403         return bio;
 404 }
 405
 406 static void crypt_free_buffer_pages(struct crypt_config *cc,
 407                                     struct bio *bio, unsigned int bytes)
 408 {
 409         unsigned int i, start, end;
 410         struct bio_vec *bv;
 411
 412         /*
 413          * This is ugly, but Jens Axboe thinks that using bi_idx in the
 414          * endio function is too dangerous at the moment, so I calculate the
 415          * correct position using bi_vcnt and bi_size.
 416          * The bv_offset and bv_len fields might already be modified but we
 417          * know that we always allocated whole pages.
 418          * A fix to the bi_idx issue in the kernel is in the works, so
 419          * we will hopefully be able to revert to the cleaner solution soon.
 420          */
 421         i = bio->bi_vcnt - 1;
 422         bv = bio_iovec_idx(bio, i);
 423         end = (i << PAGE_SHIFT) + (bv->bv_offset + bv->bv_len) - bio->bi_size;
 424         start = end - bytes;
 425
 426         start >>= PAGE_SHIFT;
 427         if (!bio->bi_size)
 428                 end = bio->bi_vcnt;
 429         else
 430                 end >>= PAGE_SHIFT;
 431
 432         for(i = start; i < end; i++) {
 433                 bv = bio_iovec_idx(bio, i);
 434                 BUG_ON(!bv->bv_page);
 435                 mempool_free(bv->bv_page, cc->page_pool);
 436                 bv->bv_page = NULL;
 437         }
 438 }
 439
 440 /*
 441  * One of the bios was finished. Check for completion of
 442  * the whole request and correctly clean up the buffer.
 443  */
 444 static void dec_pending(struct crypt_io *io, int error)
 445 {
 446         struct crypt_config *cc = (struct crypt_config *) io->target->private;
 447
 448         if (error < 0)
 449                 io->error = error;
 450
 451         if (!atomic_dec_and_test(&io->pending))
 452                 return;
 453
 454         if (io->first_clone)
 455                 bio_put(io->first_clone);
 456
 457         bio_endio(io->bio, io->bio->bi_size, io->error);
 458
 459         mempool_free(io, cc->io_pool);
 460 }
 461
 462 /*
 463  * kcryptd:
 464  *
 465  * Needed because it would be very unwise to do decryption in an
 466  * interrupt context, so bios returning from read requests get
 467  * queued here.
 468  */
 469 static struct workqueue_struct *_kcryptd_workqueue;
 470
 471 static void kcryptd_do_work(void *data)
 472 {
 473         struct crypt_io *io = (struct crypt_io *) data;
 474         struct crypt_config *cc = (struct crypt_config *) io->target->private;
 475         struct convert_context ctx;
 476         int r;
 477
 478         crypt_convert_init(cc, &ctx, io->bio, io->bio,
 479                            io->bio->bi_sector - io->target->begin, 0);
 480         r = crypt_convert(cc, &ctx);
 481
 482         dec_pending(io, r);
 483 }
 484
 485 static void kcryptd_queue_io(struct crypt_io *io)
 486 {
 487         INIT_WORK(&io->work, kcryptd_do_work, io);
 488         queue_work(_kcryptd_workqueue, &io->work);
 489 }
 490
 491 /*
 492  * Decode key from its hex representation
 493  */
 494 static int crypt_decode_key(u8 *key, char *hex, unsigned int size)
 495 {
 496         char buffer[3];
 497         char *endp;
 498         unsigned int i;
 499
 500         buffer[2] = '\0';
 501
 502         for(i = 0; i < size; i++) {
 503                 buffer[0] = *hex++;
 504                 buffer[1] = *hex++;
 505
 506                 key[i] = (u8)simple_strtoul(buffer, &endp, 16);
 507
 508                 if (endp != &buffer[2])
 509                         return -EINVAL;
 510         }
 511
 512         if (*hex != '\0')
 513                 return -EINVAL;
 514
 515         return 0;
 516 }
 517
 518 /*
 519  * Encode key into its hex representation
 520  */
 521 static void crypt_encode_key(char *hex, u8 *key, unsigned int size)
 522 {
 523         unsigned int i;
 524
 525         for(i = 0; i < size; i++) {
 526                 sprintf(hex, "%02x", *key);
 527                 hex += 2;
 528                 key++;
 529         }
 530 }
 531
 532 /*
 533  * Construct an encryption mapping:
 534  * <cipher> <key> <iv_offset> <dev_path> <start>
 535  */
 536 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 537 {
 538         struct crypt_config *cc;
 539         struct crypto_tfm *tfm;
 540         char *tmp;
 541         char *cipher;
 542         char *chainmode;
 543         char *ivmode;
 544         char *ivopts;
 545         unsigned int crypto_flags;
 546         unsigned int key_size;
 547
 548         if (argc != 5) {
 549                 ti->error = PFX "Not enough arguments";
 550                 return -EINVAL;
 551         }
 552
 553         tmp = argv[0];
 554         cipher = strsep(&tmp, "-");
 555         chainmode = strsep(&tmp, "-");
 556         ivopts = strsep(&tmp, "-");
 557         ivmode = strsep(&ivopts, ":");
 558
 559         if (tmp)
 560                 DMWARN(PFX "Unexpected additional cipher options");
 561
 562         key_size = strlen(argv[1]) >> 1;
 563
 564         cc = kmalloc(sizeof(*cc) + key_size * sizeof(u8), GFP_KERNEL);
 565         if (cc == NULL) {
 566                 ti->error =
 567                         PFX "Cannot allocate transparent encryption context";
 568                 return -ENOMEM;
 569         }
 570
 571         cc->key_size = key_size;
 572         if ((!key_size && strcmp(argv[1], "-") != 0) ||
 573             (key_size && crypt_decode_key(cc->key, argv[1], key_size) < 0)) {
 574                 ti->error = PFX "Error decoding key";
 575                 goto bad1;
 576         }
 577
 578         /* Compatiblity mode for old dm-crypt cipher strings */
 579         if (!chainmode || (strcmp(chainmode, "plain") == 0 && !ivmode)) {
 580                 chainmode = "cbc";
 581                 ivmode = "plain";
 582         }
 583
 584         /* Choose crypto_flags according to chainmode */
 585         if (strcmp(chainmode, "cbc") == 0)
 586                 crypto_flags = CRYPTO_TFM_MODE_CBC;
 587         else if (strcmp(chainmode, "ecb") == 0)
 588                 crypto_flags = CRYPTO_TFM_MODE_ECB;
 589         else {
 590                 ti->error = PFX "Unknown chaining mode";
 591                 goto bad1;
 592         }
 593
 594         if (crypto_flags != CRYPTO_TFM_MODE_ECB && !ivmode) {
 595                 ti->error = PFX "This chaining mode requires an IV mechanism";
 596                 goto bad1;
 597         }
 598
 599         tfm = crypto_alloc_tfm(cipher, crypto_flags);
 600         if (!tfm) {
 601                 ti->error = PFX "Error allocating crypto tfm";
 602                 goto bad1;
 603         }
 604         if (crypto_tfm_alg_type(tfm) != CRYPTO_ALG_TYPE_CIPHER) {
 605                 ti->error = PFX "Expected cipher algorithm";
 606                 goto bad2;
 607         }
 608
 609         cc->tfm = tfm;
 610
 611         /*
 612          * Choose ivmode. Valid modes: "plain", "essiv:<esshash>".
 613          * See comments at iv code
 614          */
 615
 616         if (ivmode == NULL)
 617                 cc->iv_gen_ops = NULL;
 618         else if (strcmp(ivmode, "plain") == 0)
 619                 cc->iv_gen_ops = &crypt_iv_plain_ops;
 620         else if (strcmp(ivmode, "essiv") == 0)
 621                 cc->iv_gen_ops = &crypt_iv_essiv_ops;
 622         else {
 623                 ti->error = PFX "Invalid IV mode";
 624                 goto bad2;
 625         }
 626
 627         if (cc->iv_gen_ops && cc->iv_gen_ops->ctr &&
 628             cc->iv_gen_ops->ctr(cc, ti, ivopts) < 0)
 629                 goto bad2;
 630
 631         if (tfm->crt_cipher.cit_decrypt_iv && tfm->crt_cipher.cit_encrypt_iv)
 632                 /* at least a 64 bit sector number should fit in our buffer */
 633                 cc->iv_size = max(crypto_tfm_alg_ivsize(tfm),
 634                                   (unsigned int)(sizeof(u64) / sizeof(u8)));
 635         else {
 636                 cc->iv_size = 0;
 637                 if (cc->iv_gen_ops) {
 638                         DMWARN(PFX "Selected cipher does not support IVs");
 639                         if (cc->iv_gen_ops->dtr)
 640                                 cc->iv_gen_ops->dtr(cc);
 641                         cc->iv_gen_ops = NULL;
 642                 }
 643         }
 644
 645         cc->io_pool = mempool_create(MIN_IOS, mempool_alloc_slab,
 646                                      mempool_free_slab, _crypt_io_pool);
 647         if (!cc->io_pool) {
 648                 ti->error = PFX "Cannot allocate crypt io mempool";
 649                 goto bad3;
 650         }
 651
 652         cc->page_pool = mempool_create(MIN_POOL_PAGES, mempool_alloc_page,
 653                                        mempool_free_page, NULL);
 654         if (!cc->page_pool) {
 655                 ti->error = PFX "Cannot allocate page mempool";
 656                 goto bad4;
 657         }
 658
 659         if (tfm->crt_cipher.cit_setkey(tfm, cc->key, key_size) < 0) {
 660                 ti->error = PFX "Error setting key";
 661                 goto bad5;
 662         }
 663
 664         if (sscanf(argv[2], SECTOR_FORMAT, &cc->iv_offset) != 1) {
 665                 ti->error = PFX "Invalid iv_offset sector";
 666                 goto bad5;
 667         }
 668
 669         if (sscanf(argv[4], SECTOR_FORMAT, &cc->start) != 1) {
 670                 ti->error = PFX "Invalid device sector";
 671                 goto bad5;
 672         }
 673
 674         if (dm_get_device(ti, argv[3], cc->start, ti->len,
 675                           dm_table_get_mode(ti->table), &cc->dev)) {
 676                 ti->error = PFX "Device lookup failed";
 677                 goto bad5;
 678         }
 679
 680         if (ivmode && cc->iv_gen_ops) {
 681                 if (ivopts)
 682                         *(ivopts - 1) = ':';
 683                 cc->iv_mode = kmalloc(strlen(ivmode) + 1, GFP_KERNEL);
 684                 if (!cc->iv_mode) {
 685                         ti->error = PFX "Error kmallocing iv_mode string";
 686                         goto bad5;
 687                 }
 688                 strcpy(cc->iv_mode, ivmode);
 689         } else
 690                 cc->iv_mode = NULL;
 691
 692         ti->private = cc;
 693         return 0;
 694
 695 bad5:
 696         mempool_destroy(cc->page_pool);
 697 bad4:
 698         mempool_destroy(cc->io_pool);
 699 bad3:
 700         if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
 701                 cc->iv_gen_ops->dtr(cc);
 702 bad2:
 703         crypto_free_tfm(tfm);
 704 bad1:
 705         kfree(cc);
 706         return -EINVAL;
 707 }
 708
 709 static void crypt_dtr(struct dm_target *ti)
 710 {
 711         struct crypt_config *cc = (struct crypt_config *) ti->private;
 712
 713         mempool_destroy(cc->page_pool);
 714         mempool_destroy(cc->io_pool);
 715
 716         if (cc->iv_mode)
 717                 kfree(cc->iv_mode);
 718         if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
 719                 cc->iv_gen_ops->dtr(cc);
 720         crypto_free_tfm(cc->tfm);
 721         dm_put_device(ti, cc->dev);
 722         kfree(cc);
 723 }
 724
 725 static int crypt_endio(struct bio *bio, unsigned int done, int error)
 726 {
 727         struct crypt_io *io = (struct crypt_io *) bio->bi_private;
 728         struct crypt_config *cc = (struct crypt_config *) io->target->private;
 729
 730         if (bio_data_dir(bio) == WRITE) {
 731                 /*
 732                  * free the processed pages, even if
 733                  * it's only a partially completed write
 734                  */
 735                 crypt_free_buffer_pages(cc, bio, done);
 736         }
 737
 738         if (bio->bi_size)
 739                 return 1;
 740
 741         bio_put(bio);
 742
 743         /*
 744          * successful reads are decrypted by the worker thread
 745          */
 746         if ((bio_data_dir(bio) == READ)
 747             && bio_flagged(bio, BIO_UPTODATE)) {
 748                 kcryptd_queue_io(io);
 749                 return 0;
 750         }
 751
 752         dec_pending(io, error);
 753         return error;
 754 }
 755
 756 static inline struct bio *
 757 crypt_clone(struct crypt_config *cc, struct crypt_io *io, struct bio *bio,
 758             sector_t sector, unsigned int *bvec_idx,
 759             struct convert_context *ctx)
 760 {
 761         struct bio *clone;
 762
 763         if (bio_data_dir(bio) == WRITE) {
 764                 clone = crypt_alloc_buffer(cc, bio->bi_size,
 765                                  io->first_clone, bvec_idx);
 766                 if (clone) {
 767                         ctx->bio_out = clone;
 768                         if (crypt_convert(cc, ctx) < 0) {
 769                                 crypt_free_buffer_pages(cc, clone,
 770                                                         clone->bi_size);
 771                                 bio_put(clone);
 772                                 return NULL;
 773                         }
 774                 }
 775         } else {
 776                 /*
 777                  * The block layer might modify the bvec array, so always
 778                  * copy the required bvecs because we need the original
 779                  * one in order to decrypt the whole bio data *afterwards*.
 780                  */
 781                 clone = bio_alloc(GFP_NOIO, bio_segments(bio));
 782                 if (clone) {
 783                         clone->bi_idx = 0;
 784                         clone->bi_vcnt = bio_segments(bio);
 785                         clone->bi_size = bio->bi_size;
 786                         memcpy(clone->bi_io_vec, bio_iovec(bio),
 787                                sizeof(struct bio_vec) * clone->bi_vcnt);
 788                 }
 789         }
 790
 791         if (!clone)
 792                 return NULL;
 793
 794         clone->bi_private = io;
 795         clone->bi_end_io = crypt_endio;
 796         clone->bi_bdev = cc->dev->bdev;
 797         clone->bi_sector = cc->start + sector;
 798         clone->bi_rw = bio->bi_rw;
 799
 800         return clone;
 801 }
 802
 803 static int crypt_map(struct dm_target *ti, struct bio *bio,
 804                      union map_info *map_context)
 805 {
 806         struct crypt_config *cc = (struct crypt_config *) ti->private;
 807         struct crypt_io *io = mempool_alloc(cc->io_pool, GFP_NOIO);
 808         struct convert_context ctx;
 809         struct bio *clone;
 810         unsigned int remaining = bio->bi_size;
 811         sector_t sector = bio->bi_sector - ti->begin;
 812         unsigned int bvec_idx = 0;
 813
 814         io->target = ti;
 815         io->bio = bio;
 816         io->first_clone = NULL;
 817         io->error = 0;
 818         atomic_set(&io->pending, 1); /* hold a reference */
 819
 820         if (bio_data_dir(bio) == WRITE)
 821                 crypt_convert_init(cc, &ctx, NULL, bio, sector, 1);
 822
 823         /*
 824          * The allocated buffers can be smaller than the whole bio,
 825          * so repeat the whole process until all the data can be handled.
 826          */
 827         while (remaining) {
 828                 clone = crypt_clone(cc, io, bio, sector, &bvec_idx, &ctx);
 829                 if (!clone)
 830                         goto cleanup;
 831
 832                 if (!io->first_clone) {
 833                         /*
 834                          * hold a reference to the first clone, because it
 835                          * holds the bio_vec array and that can't be freed
 836                          * before all other clones are released
 837                          */
 838                         bio_get(clone);
 839                         io->first_clone = clone;
 840                 }
 841                 atomic_inc(&io->pending);
 842
 843                 remaining -= clone->bi_size;
 844                 sector += bio_sectors(clone);
 845
 846                 generic_make_request(clone);
 847
 848                 /* out of memory -> run queues */
 849                 if (remaining)
 850                         blk_congestion_wait(bio_data_dir(clone), HZ/100);
 851         }
 852
 853         /* drop reference, clones could have returned before we reach this */
 854         dec_pending(io, 0);
 855         return 0;
 856
 857 cleanup:
 858         if (io->first_clone) {
 859                 dec_pending(io, -ENOMEM);
 860                 return 0;
 861         }
 862
 863         /* if no bio has been dispatched yet, we can directly return the error */
 864         mempool_free(io, cc->io_pool);
 865         return -ENOMEM;
 866 }
 867
 868 static int crypt_status(struct dm_target *ti, status_type_t type,
 869                         char *result, unsigned int maxlen)
 870 {
 871         struct crypt_config *cc = (struct crypt_config *) ti->private;
 872         const char *cipher;
 873         const char *chainmode = NULL;
 874         unsigned int sz = 0;
 875
 876         switch (type) {
 877         case STATUSTYPE_INFO:
 878                 result[0] = '\0';
 879                 break;
 880
 881         case STATUSTYPE_TABLE:
 882                 cipher = crypto_tfm_alg_name(cc->tfm);
 883
 884                 switch(cc->tfm->crt_cipher.cit_mode) {
 885                 case CRYPTO_TFM_MODE_CBC:
 886                         chainmode = "cbc";
 887                         break;
 888                 case CRYPTO_TFM_MODE_ECB:
 889                         chainmode = "ecb";
 890                         break;
 891                 default:
 892                         BUG();
 893                 }
 894
 895                 if (cc->iv_mode)
 896                         DMEMIT("%s-%s-%s ", cipher, chainmode, cc->iv_mode);
 897                 else
 898                         DMEMIT("%s-%s ", cipher, chainmode);
 899
 900                 if (cc->key_size > 0) {
 901                         if ((maxlen - sz) < ((cc->key_size << 1) + 1))
 902                                 return -ENOMEM;
 903
 904                         crypt_encode_key(result + sz, cc->key, cc->key_size);
 905                         sz += cc->key_size << 1;
 906                 } else {
 907                         if (sz >= maxlen)
 908                                 return -ENOMEM;
 909                         result[sz++] = '-';
 910                 }
 911
 912                 DMEMIT(" " SECTOR_FORMAT " %s " SECTOR_FORMAT,
 913                        cc->iv_offset, cc->dev->name, cc->start);
 914                 break;
 915         }
 916         return 0;
 917 }
 918
 919 static struct target_type crypt_target = {
 920         .name   = "crypt",
 921         .version= {1, 1, 0},
 922         .module = THIS_MODULE,
 923         .ctr    = crypt_ctr,
 924         .dtr    = crypt_dtr,
 925         .map    = crypt_map,
 926         .status = crypt_status,
 927 };
 928
 929 static int __init dm_crypt_init(void)
 930 {
 931         int r;
 932
 933         _crypt_io_pool = kmem_cache_create("dm-crypt_io",
 934                                            sizeof(struct crypt_io),
 935                                            0, 0, NULL, NULL);
 936         if (!_crypt_io_pool)
 937                 return -ENOMEM;
 938
 939         _kcryptd_workqueue = create_workqueue("kcryptd");
 940         if (!_kcryptd_workqueue) {
 941                 r = -ENOMEM;
 942                 DMERR(PFX "couldn't create kcryptd");
 943                 goto bad1;
 944         }
 945
 946         r = dm_register_target(&crypt_target);
 947         if (r < 0) {
 948                 DMERR(PFX "register failed %d", r);
 949                 goto bad2;
 950         }
 951
 952         return 0;
 953
 954 bad2:
 955         destroy_workqueue(_kcryptd_workqueue);
 956 bad1:
 957         kmem_cache_destroy(_crypt_io_pool);
 958         return r;
 959 }
 960
 961 static void __exit dm_crypt_exit(void)
 962 {
 963         int r = dm_unregister_target(&crypt_target);
 964
 965         if (r < 0)
 966                 DMERR(PFX "unregister failed %d", r);
 967
 968         destroy_workqueue(_kcryptd_workqueue);
 969         kmem_cache_destroy(_crypt_io_pool);
 970 }
 971
 972 module_init(dm_crypt_init);
 973 module_exit(dm_crypt_exit);
 974
 975 MODULE_AUTHOR("Christophe Saout <christophe@saout.de>");
 976 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
 977 MODULE_LICENSE("GPL");