block/qcow.c

   1 /*
   2  * Block driver for the QCOW format
   3  *
   4  * Copyright (c) 2004-2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu/osdep.h"
  25 #include "qemu-common.h"
  26 #include "block/block_int.h"
  27 #include "qemu/module.h"
  28 #include <zlib.h>
  29 #include "qapi/qmp/qerror.h"
  30 #include "crypto/cipher.h"
  31 #include "migration/migration.h"
  32
  33 /**************************************************************/
  34 /* QEMU COW block driver with compression and encryption support */
  35
  36 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
  37 #define QCOW_VERSION 1
  38
  39 #define QCOW_CRYPT_NONE 0
  40 #define QCOW_CRYPT_AES  1
  41
  42 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
  43
  44 typedef struct QCowHeader {
  45     uint32_t magic;
  46     uint32_t version;
  47     uint64_t backing_file_offset;
  48     uint32_t backing_file_size;
  49     uint32_t mtime;
  50     uint64_t size; /* in bytes */
  51     uint8_t cluster_bits;
  52     uint8_t l2_bits;
  53     uint16_t padding;
  54     uint32_t crypt_method;
  55     uint64_t l1_table_offset;
  56 } QEMU_PACKED QCowHeader;
  57
  58 #define L2_CACHE_SIZE 16
  59
  60 typedef struct BDRVQcowState {
  61     int cluster_bits;
  62     int cluster_size;
  63     int cluster_sectors;
  64     int l2_bits;
  65     int l2_size;
  66     unsigned int l1_size;
  67     uint64_t cluster_offset_mask;
  68     uint64_t l1_table_offset;
  69     uint64_t *l1_table;
  70     uint64_t *l2_cache;
  71     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
  72     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  73     uint8_t *cluster_cache;
  74     uint8_t *cluster_data;
  75     uint64_t cluster_cache_offset;
  76     QCryptoCipher *cipher; /* NULL if no key yet */
  77     uint32_t crypt_method_header;
  78     CoMutex lock;
  79     Error *migration_blocker;
  80 } BDRVQcowState;
  81
  82 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
  83
  84 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
  85 {
  86     const QCowHeader *cow_header = (const void *)buf;
  87
  88     if (buf_size >= sizeof(QCowHeader) &&
  89         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
  90         be32_to_cpu(cow_header->version) == QCOW_VERSION)
  91         return 100;
  92     else
  93         return 0;
  94 }
  95
  96 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
  97                      Error **errp)
  98 {
  99     BDRVQcowState *s = bs->opaque;
 100     unsigned int len, i, shift;
 101     int ret;
 102     QCowHeader header;
 103
 104     ret = bdrv_pread(bs->file->bs, 0, &header, sizeof(header));
 105     if (ret < 0) {
 106         goto fail;
 107     }
 108     be32_to_cpus(&header.magic);
 109     be32_to_cpus(&header.version);
 110     be64_to_cpus(&header.backing_file_offset);
 111     be32_to_cpus(&header.backing_file_size);
 112     be32_to_cpus(&header.mtime);
 113     be64_to_cpus(&header.size);
 114     be32_to_cpus(&header.crypt_method);
 115     be64_to_cpus(&header.l1_table_offset);
 116
 117     if (header.magic != QCOW_MAGIC) {
 118         error_setg(errp, "Image not in qcow format");
 119         ret = -EINVAL;
 120         goto fail;
 121     }
 122     if (header.version != QCOW_VERSION) {
 123         char version[64];
 124         snprintf(version, sizeof(version), "QCOW version %" PRIu32,
 125                  header.version);
 126         error_setg(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
 127                    bdrv_get_device_or_node_name(bs), "qcow", version);
 128         ret = -ENOTSUP;
 129         goto fail;
 130     }
 131
 132     if (header.size <= 1) {
 133         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
 134         ret = -EINVAL;
 135         goto fail;
 136     }
 137     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
 138         error_setg(errp, "Cluster size must be between 512 and 64k");
 139         ret = -EINVAL;
 140         goto fail;
 141     }
 142
 143     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
 144      * so bytes = num_entries << 3. */
 145     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
 146         error_setg(errp, "L2 table size must be between 512 and 64k");
 147         ret = -EINVAL;
 148         goto fail;
 149     }
 150
 151     if (header.crypt_method > QCOW_CRYPT_AES) {
 152         error_setg(errp, "invalid encryption method in qcow header");
 153         ret = -EINVAL;
 154         goto fail;
 155     }
 156     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) {
 157         error_setg(errp, "AES cipher not available");
 158         ret = -EINVAL;
 159         goto fail;
 160     }
 161     s->crypt_method_header = header.crypt_method;
 162     if (s->crypt_method_header) {
 163         bs->encrypted = 1;
 164     }
 165     s->cluster_bits = header.cluster_bits;
 166     s->cluster_size = 1 << s->cluster_bits;
 167     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 168     s->l2_bits = header.l2_bits;
 169     s->l2_size = 1 << s->l2_bits;
 170     bs->total_sectors = header.size / 512;
 171     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
 172
 173     /* read the level 1 table */
 174     shift = s->cluster_bits + s->l2_bits;
 175     if (header.size > UINT64_MAX - (1LL << shift)) {
 176         error_setg(errp, "Image too large");
 177         ret = -EINVAL;
 178         goto fail;
 179     } else {
 180         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
 181         if (l1_size > INT_MAX / sizeof(uint64_t)) {
 182             error_setg(errp, "Image too large");
 183             ret = -EINVAL;
 184             goto fail;
 185         }
 186         s->l1_size = l1_size;
 187     }
 188
 189     s->l1_table_offset = header.l1_table_offset;
 190     s->l1_table = g_try_new(uint64_t, s->l1_size);
 191     if (s->l1_table == NULL) {
 192         error_setg(errp, "Could not allocate memory for L1 table");
 193         ret = -ENOMEM;
 194         goto fail;
 195     }
 196
 197     ret = bdrv_pread(bs->file->bs, s->l1_table_offset, s->l1_table,
 198                s->l1_size * sizeof(uint64_t));
 199     if (ret < 0) {
 200         goto fail;
 201     }
 202
 203     for(i = 0;i < s->l1_size; i++) {
 204         be64_to_cpus(&s->l1_table[i]);
 205     }
 206
 207     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
 208     s->l2_cache =
 209         qemu_try_blockalign(bs->file->bs,
 210                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 211     if (s->l2_cache == NULL) {
 212         error_setg(errp, "Could not allocate L2 table cache");
 213         ret = -ENOMEM;
 214         goto fail;
 215     }
 216     s->cluster_cache = g_malloc(s->cluster_size);
 217     s->cluster_data = g_malloc(s->cluster_size);
 218     s->cluster_cache_offset = -1;
 219
 220     /* read the backing file name */
 221     if (header.backing_file_offset != 0) {
 222         len = header.backing_file_size;
 223         if (len > 1023 || len >= sizeof(bs->backing_file)) {
 224             error_setg(errp, "Backing file name too long");
 225             ret = -EINVAL;
 226             goto fail;
 227         }
 228         ret = bdrv_pread(bs->file->bs, header.backing_file_offset,
 229                    bs->backing_file, len);
 230         if (ret < 0) {
 231             goto fail;
 232         }
 233         bs->backing_file[len] = '\0';
 234     }
 235
 236     /* Disable migration when qcow images are used */
 237     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
 238                "does not support live migration",
 239                bdrv_get_device_or_node_name(bs));
 240     migrate_add_blocker(s->migration_blocker);
 241
 242     qemu_co_mutex_init(&s->lock);
 243     return 0;
 244
 245  fail:
 246     g_free(s->l1_table);
 247     qemu_vfree(s->l2_cache);
 248     g_free(s->cluster_cache);
 249     g_free(s->cluster_data);
 250     return ret;
 251 }
 252
 253
 254 /* We have nothing to do for QCOW reopen, stubs just return
 255  * success */
 256 static int qcow_reopen_prepare(BDRVReopenState *state,
 257                                BlockReopenQueue *queue, Error **errp)
 258 {
 259     return 0;
 260 }
 261
 262 static int qcow_set_key(BlockDriverState *bs, const char *key)
 263 {
 264     BDRVQcowState *s = bs->opaque;
 265     uint8_t keybuf[16];
 266     int len, i;
 267     Error *err;
 268
 269     memset(keybuf, 0, 16);
 270     len = strlen(key);
 271     if (len > 16)
 272         len = 16;
 273     /* XXX: we could compress the chars to 7 bits to increase
 274        entropy */
 275     for(i = 0;i < len;i++) {
 276         keybuf[i] = key[i];
 277     }
 278     assert(bs->encrypted);
 279
 280     qcrypto_cipher_free(s->cipher);
 281     s->cipher = qcrypto_cipher_new(
 282         QCRYPTO_CIPHER_ALG_AES_128,
 283         QCRYPTO_CIPHER_MODE_CBC,
 284         keybuf, G_N_ELEMENTS(keybuf),
 285         &err);
 286
 287     if (!s->cipher) {
 288         /* XXX would be nice if errors in this method could
 289          * be properly propagate to the caller. Would need
 290          * the bdrv_set_key() API signature to be fixed. */
 291         error_free(err);
 292         return -1;
 293     }
 294     return 0;
 295 }
 296
 297 /* The crypt function is compatible with the linux cryptoloop
 298    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
 299    supported */
 300 static int encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
 301                            uint8_t *out_buf, const uint8_t *in_buf,
 302                            int nb_sectors, bool enc, Error **errp)
 303 {
 304     union {
 305         uint64_t ll[2];
 306         uint8_t b[16];
 307     } ivec;
 308     int i;
 309     int ret;
 310
 311     for(i = 0; i < nb_sectors; i++) {
 312         ivec.ll[0] = cpu_to_le64(sector_num);
 313         ivec.ll[1] = 0;
 314         if (qcrypto_cipher_setiv(s->cipher,
 315                                  ivec.b, G_N_ELEMENTS(ivec.b),
 316                                  errp) < 0) {
 317             return -1;
 318         }
 319         if (enc) {
 320             ret = qcrypto_cipher_encrypt(s->cipher,
 321                                          in_buf,
 322                                          out_buf,
 323                                          512,
 324                                          errp);
 325         } else {
 326             ret = qcrypto_cipher_decrypt(s->cipher,
 327                                          in_buf,
 328                                          out_buf,
 329                                          512,
 330                                          errp);
 331         }
 332         if (ret < 0) {
 333             return -1;
 334         }
 335         sector_num++;
 336         in_buf += 512;
 337         out_buf += 512;
 338     }
 339     return 0;
 340 }
 341
 342 /* 'allocate' is:
 343  *
 344  * 0 to not allocate.
 345  *
 346  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
 347  * 'n_end')
 348  *
 349  * 2 to allocate a compressed cluster of size
 350  * 'compressed_size'. 'compressed_size' must be > 0 and <
 351  * cluster_size
 352  *
 353  * return 0 if not allocated.
 354  */
 355 static uint64_t get_cluster_offset(BlockDriverState *bs,
 356                                    uint64_t offset, int allocate,
 357                                    int compressed_size,
 358                                    int n_start, int n_end)
 359 {
 360     BDRVQcowState *s = bs->opaque;
 361     int min_index, i, j, l1_index, l2_index;
 362     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
 363     uint32_t min_count;
 364     int new_l2_table;
 365
 366     l1_index = offset >> (s->l2_bits + s->cluster_bits);
 367     l2_offset = s->l1_table[l1_index];
 368     new_l2_table = 0;
 369     if (!l2_offset) {
 370         if (!allocate)
 371             return 0;
 372         /* allocate a new l2 entry */
 373         l2_offset = bdrv_getlength(bs->file->bs);
 374         /* round to cluster size */
 375         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
 376         /* update the L1 entry */
 377         s->l1_table[l1_index] = l2_offset;
 378         tmp = cpu_to_be64(l2_offset);
 379         if (bdrv_pwrite_sync(bs->file->bs,
 380                 s->l1_table_offset + l1_index * sizeof(tmp),
 381                 &tmp, sizeof(tmp)) < 0)
 382             return 0;
 383         new_l2_table = 1;
 384     }
 385     for(i = 0; i < L2_CACHE_SIZE; i++) {
 386         if (l2_offset == s->l2_cache_offsets[i]) {
 387             /* increment the hit count */
 388             if (++s->l2_cache_counts[i] == 0xffffffff) {
 389                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 390                     s->l2_cache_counts[j] >>= 1;
 391                 }
 392             }
 393             l2_table = s->l2_cache + (i << s->l2_bits);
 394             goto found;
 395         }
 396     }
 397     /* not found: load a new entry in the least used one */
 398     min_index = 0;
 399     min_count = 0xffffffff;
 400     for(i = 0; i < L2_CACHE_SIZE; i++) {
 401         if (s->l2_cache_counts[i] < min_count) {
 402             min_count = s->l2_cache_counts[i];
 403             min_index = i;
 404         }
 405     }
 406     l2_table = s->l2_cache + (min_index << s->l2_bits);
 407     if (new_l2_table) {
 408         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
 409         if (bdrv_pwrite_sync(bs->file->bs, l2_offset, l2_table,
 410                 s->l2_size * sizeof(uint64_t)) < 0)
 411             return 0;
 412     } else {
 413         if (bdrv_pread(bs->file->bs, l2_offset, l2_table,
 414                        s->l2_size * sizeof(uint64_t)) !=
 415             s->l2_size * sizeof(uint64_t))
 416             return 0;
 417     }
 418     s->l2_cache_offsets[min_index] = l2_offset;
 419     s->l2_cache_counts[min_index] = 1;
 420  found:
 421     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
 422     cluster_offset = be64_to_cpu(l2_table[l2_index]);
 423     if (!cluster_offset ||
 424         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
 425         if (!allocate)
 426             return 0;
 427         /* allocate a new cluster */
 428         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
 429             (n_end - n_start) < s->cluster_sectors) {
 430             /* if the cluster is already compressed, we must
 431                decompress it in the case it is not completely
 432                overwritten */
 433             if (decompress_cluster(bs, cluster_offset) < 0)
 434                 return 0;
 435             cluster_offset = bdrv_getlength(bs->file->bs);
 436             cluster_offset = (cluster_offset + s->cluster_size - 1) &
 437                 ~(s->cluster_size - 1);
 438             /* write the cluster content */
 439             if (bdrv_pwrite(bs->file->bs, cluster_offset, s->cluster_cache,
 440                             s->cluster_size) !=
 441                 s->cluster_size)
 442                 return -1;
 443         } else {
 444             cluster_offset = bdrv_getlength(bs->file->bs);
 445             if (allocate == 1) {
 446                 /* round to cluster size */
 447                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
 448                     ~(s->cluster_size - 1);
 449                 bdrv_truncate(bs->file->bs, cluster_offset + s->cluster_size);
 450                 /* if encrypted, we must initialize the cluster
 451                    content which won't be written */
 452                 if (bs->encrypted &&
 453                     (n_end - n_start) < s->cluster_sectors) {
 454                     uint64_t start_sect;
 455                     assert(s->cipher);
 456                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
 457                     memset(s->cluster_data + 512, 0x00, 512);
 458                     for(i = 0; i < s->cluster_sectors; i++) {
 459                         if (i < n_start || i >= n_end) {
 460                             Error *err = NULL;
 461                             if (encrypt_sectors(s, start_sect + i,
 462                                                 s->cluster_data,
 463                                                 s->cluster_data + 512, 1,
 464                                                 true, &err) < 0) {
 465                                 error_free(err);
 466                                 errno = EIO;
 467                                 return -1;
 468                             }
 469                             if (bdrv_pwrite(bs->file->bs,
 470                                             cluster_offset + i * 512,
 471                                             s->cluster_data, 512) != 512)
 472                                 return -1;
 473                         }
 474                     }
 475                 }
 476             } else if (allocate == 2) {
 477                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
 478                     (uint64_t)compressed_size << (63 - s->cluster_bits);
 479             }
 480         }
 481         /* update L2 table */
 482         tmp = cpu_to_be64(cluster_offset);
 483         l2_table[l2_index] = tmp;
 484         if (bdrv_pwrite_sync(bs->file->bs, l2_offset + l2_index * sizeof(tmp),
 485                 &tmp, sizeof(tmp)) < 0)
 486             return 0;
 487     }
 488     return cluster_offset;
 489 }
 490
 491 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
 492         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
 493 {
 494     BDRVQcowState *s = bs->opaque;
 495     int index_in_cluster, n;
 496     uint64_t cluster_offset;
 497
 498     qemu_co_mutex_lock(&s->lock);
 499     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
 500     qemu_co_mutex_unlock(&s->lock);
 501     index_in_cluster = sector_num & (s->cluster_sectors - 1);
 502     n = s->cluster_sectors - index_in_cluster;
 503     if (n > nb_sectors)
 504         n = nb_sectors;
 505     *pnum = n;
 506     if (!cluster_offset) {
 507         return 0;
 508     }
 509     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->cipher) {
 510         return BDRV_BLOCK_DATA;
 511     }
 512     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
 513     *file = bs->file->bs;
 514     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
 515 }
 516
 517 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
 518                              const uint8_t *buf, int buf_size)
 519 {
 520     z_stream strm1, *strm = &strm1;
 521     int ret, out_len;
 522
 523     memset(strm, 0, sizeof(*strm));
 524
 525     strm->next_in = (uint8_t *)buf;
 526     strm->avail_in = buf_size;
 527     strm->next_out = out_buf;
 528     strm->avail_out = out_buf_size;
 529
 530     ret = inflateInit2(strm, -12);
 531     if (ret != Z_OK)
 532         return -1;
 533     ret = inflate(strm, Z_FINISH);
 534     out_len = strm->next_out - out_buf;
 535     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
 536         out_len != out_buf_size) {
 537         inflateEnd(strm);
 538         return -1;
 539     }
 540     inflateEnd(strm);
 541     return 0;
 542 }
 543
 544 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
 545 {
 546     BDRVQcowState *s = bs->opaque;
 547     int ret, csize;
 548     uint64_t coffset;
 549
 550     coffset = cluster_offset & s->cluster_offset_mask;
 551     if (s->cluster_cache_offset != coffset) {
 552         csize = cluster_offset >> (63 - s->cluster_bits);
 553         csize &= (s->cluster_size - 1);
 554         ret = bdrv_pread(bs->file->bs, coffset, s->cluster_data, csize);
 555         if (ret != csize)
 556             return -1;
 557         if (decompress_buffer(s->cluster_cache, s->cluster_size,
 558                               s->cluster_data, csize) < 0) {
 559             return -1;
 560         }
 561         s->cluster_cache_offset = coffset;
 562     }
 563     return 0;
 564 }
 565
 566 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
 567                          int nb_sectors, QEMUIOVector *qiov)
 568 {
 569     BDRVQcowState *s = bs->opaque;
 570     int index_in_cluster;
 571     int ret = 0, n;
 572     uint64_t cluster_offset;
 573     struct iovec hd_iov;
 574     QEMUIOVector hd_qiov;
 575     uint8_t *buf;
 576     void *orig_buf;
 577     Error *err = NULL;
 578
 579     if (qiov->niov > 1) {
 580         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
 581         if (buf == NULL) {
 582             return -ENOMEM;
 583         }
 584     } else {
 585         orig_buf = NULL;
 586         buf = (uint8_t *)qiov->iov->iov_base;
 587     }
 588
 589     qemu_co_mutex_lock(&s->lock);
 590
 591     while (nb_sectors != 0) {
 592         /* prepare next request */
 593         cluster_offset = get_cluster_offset(bs, sector_num << 9,
 594                                                  0, 0, 0, 0);
 595         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 596         n = s->cluster_sectors - index_in_cluster;
 597         if (n > nb_sectors) {
 598             n = nb_sectors;
 599         }
 600
 601         if (!cluster_offset) {
 602             if (bs->backing) {
 603                 /* read from the base image */
 604                 hd_iov.iov_base = (void *)buf;
 605                 hd_iov.iov_len = n * 512;
 606                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 607                 qemu_co_mutex_unlock(&s->lock);
 608                 ret = bdrv_co_readv(bs->backing->bs, sector_num,
 609                                     n, &hd_qiov);
 610                 qemu_co_mutex_lock(&s->lock);
 611                 if (ret < 0) {
 612                     goto fail;
 613                 }
 614             } else {
 615                 /* Note: in this case, no need to wait */
 616                 memset(buf, 0, 512 * n);
 617             }
 618         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 619             /* add AIO support for compressed blocks ? */
 620             if (decompress_cluster(bs, cluster_offset) < 0) {
 621                 goto fail;
 622             }
 623             memcpy(buf,
 624                    s->cluster_cache + index_in_cluster * 512, 512 * n);
 625         } else {
 626             if ((cluster_offset & 511) != 0) {
 627                 goto fail;
 628             }
 629             hd_iov.iov_base = (void *)buf;
 630             hd_iov.iov_len = n * 512;
 631             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 632             qemu_co_mutex_unlock(&s->lock);
 633             ret = bdrv_co_readv(bs->file->bs,
 634                                 (cluster_offset >> 9) + index_in_cluster,
 635                                 n, &hd_qiov);
 636             qemu_co_mutex_lock(&s->lock);
 637             if (ret < 0) {
 638                 break;
 639             }
 640             if (bs->encrypted) {
 641                 assert(s->cipher);
 642                 if (encrypt_sectors(s, sector_num, buf, buf,
 643                                     n, false, &err) < 0) {
 644                     goto fail;
 645                 }
 646             }
 647         }
 648         ret = 0;
 649
 650         nb_sectors -= n;
 651         sector_num += n;
 652         buf += n * 512;
 653     }
 654
 655 done:
 656     qemu_co_mutex_unlock(&s->lock);
 657
 658     if (qiov->niov > 1) {
 659         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
 660         qemu_vfree(orig_buf);
 661     }
 662
 663     return ret;
 664
 665 fail:
 666     error_free(err);
 667     ret = -EIO;
 668     goto done;
 669 }
 670
 671 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
 672                           int nb_sectors, QEMUIOVector *qiov)
 673 {
 674     BDRVQcowState *s = bs->opaque;
 675     int index_in_cluster;
 676     uint64_t cluster_offset;
 677     const uint8_t *src_buf;
 678     int ret = 0, n;
 679     uint8_t *cluster_data = NULL;
 680     struct iovec hd_iov;
 681     QEMUIOVector hd_qiov;
 682     uint8_t *buf;
 683     void *orig_buf;
 684
 685     s->cluster_cache_offset = -1; /* disable compressed cache */
 686
 687     if (qiov->niov > 1) {
 688         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
 689         if (buf == NULL) {
 690             return -ENOMEM;
 691         }
 692         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
 693     } else {
 694         orig_buf = NULL;
 695         buf = (uint8_t *)qiov->iov->iov_base;
 696     }
 697
 698     qemu_co_mutex_lock(&s->lock);
 699
 700     while (nb_sectors != 0) {
 701
 702         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 703         n = s->cluster_sectors - index_in_cluster;
 704         if (n > nb_sectors) {
 705             n = nb_sectors;
 706         }
 707         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
 708                                             index_in_cluster,
 709                                             index_in_cluster + n);
 710         if (!cluster_offset || (cluster_offset & 511) != 0) {
 711             ret = -EIO;
 712             break;
 713         }
 714         if (bs->encrypted) {
 715             Error *err = NULL;
 716             assert(s->cipher);
 717             if (!cluster_data) {
 718                 cluster_data = g_malloc0(s->cluster_size);
 719             }
 720             if (encrypt_sectors(s, sector_num, cluster_data, buf,
 721                                 n, true, &err) < 0) {
 722                 error_free(err);
 723                 ret = -EIO;
 724                 break;
 725             }
 726             src_buf = cluster_data;
 727         } else {
 728             src_buf = buf;
 729         }
 730
 731         hd_iov.iov_base = (void *)src_buf;
 732         hd_iov.iov_len = n * 512;
 733         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 734         qemu_co_mutex_unlock(&s->lock);
 735         ret = bdrv_co_writev(bs->file->bs,
 736                              (cluster_offset >> 9) + index_in_cluster,
 737                              n, &hd_qiov);
 738         qemu_co_mutex_lock(&s->lock);
 739         if (ret < 0) {
 740             break;
 741         }
 742         ret = 0;
 743
 744         nb_sectors -= n;
 745         sector_num += n;
 746         buf += n * 512;
 747     }
 748     qemu_co_mutex_unlock(&s->lock);
 749
 750     if (qiov->niov > 1) {
 751         qemu_vfree(orig_buf);
 752     }
 753     g_free(cluster_data);
 754
 755     return ret;
 756 }
 757
 758 static void qcow_close(BlockDriverState *bs)
 759 {
 760     BDRVQcowState *s = bs->opaque;
 761
 762     qcrypto_cipher_free(s->cipher);
 763     s->cipher = NULL;
 764     g_free(s->l1_table);
 765     qemu_vfree(s->l2_cache);
 766     g_free(s->cluster_cache);
 767     g_free(s->cluster_data);
 768
 769     migrate_del_blocker(s->migration_blocker);
 770     error_free(s->migration_blocker);
 771 }
 772
 773 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
 774 {
 775     int header_size, backing_filename_len, l1_size, shift, i;
 776     QCowHeader header;
 777     uint8_t *tmp;
 778     int64_t total_size = 0;
 779     char *backing_file = NULL;
 780     int flags = 0;
 781     Error *local_err = NULL;
 782     int ret;
 783     BlockDriverState *qcow_bs;
 784
 785     /* Read out options */
 786     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 787                           BDRV_SECTOR_SIZE);
 788     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
 789     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
 790         flags |= BLOCK_FLAG_ENCRYPT;
 791     }
 792
 793     ret = bdrv_create_file(filename, opts, &local_err);
 794     if (ret < 0) {
 795         error_propagate(errp, local_err);
 796         goto cleanup;
 797     }
 798
 799     qcow_bs = NULL;
 800     ret = bdrv_open(&qcow_bs, filename, NULL, NULL,
 801                     BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
 802     if (ret < 0) {
 803         error_propagate(errp, local_err);
 804         goto cleanup;
 805     }
 806
 807     ret = bdrv_truncate(qcow_bs, 0);
 808     if (ret < 0) {
 809         goto exit;
 810     }
 811
 812     memset(&header, 0, sizeof(header));
 813     header.magic = cpu_to_be32(QCOW_MAGIC);
 814     header.version = cpu_to_be32(QCOW_VERSION);
 815     header.size = cpu_to_be64(total_size);
 816     header_size = sizeof(header);
 817     backing_filename_len = 0;
 818     if (backing_file) {
 819         if (strcmp(backing_file, "fat:")) {
 820             header.backing_file_offset = cpu_to_be64(header_size);
 821             backing_filename_len = strlen(backing_file);
 822             header.backing_file_size = cpu_to_be32(backing_filename_len);
 823             header_size += backing_filename_len;
 824         } else {
 825             /* special backing file for vvfat */
 826             backing_file = NULL;
 827         }
 828         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
 829                                     unmodified sectors */
 830         header.l2_bits = 12; /* 32 KB L2 tables */
 831     } else {
 832         header.cluster_bits = 12; /* 4 KB clusters */
 833         header.l2_bits = 9; /* 4 KB L2 tables */
 834     }
 835     header_size = (header_size + 7) & ~7;
 836     shift = header.cluster_bits + header.l2_bits;
 837     l1_size = (total_size + (1LL << shift) - 1) >> shift;
 838
 839     header.l1_table_offset = cpu_to_be64(header_size);
 840     if (flags & BLOCK_FLAG_ENCRYPT) {
 841         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
 842     } else {
 843         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
 844     }
 845
 846     /* write all the data */
 847     ret = bdrv_pwrite(qcow_bs, 0, &header, sizeof(header));
 848     if (ret != sizeof(header)) {
 849         goto exit;
 850     }
 851
 852     if (backing_file) {
 853         ret = bdrv_pwrite(qcow_bs, sizeof(header),
 854             backing_file, backing_filename_len);
 855         if (ret != backing_filename_len) {
 856             goto exit;
 857         }
 858     }
 859
 860     tmp = g_malloc0(BDRV_SECTOR_SIZE);
 861     for (i = 0; i < ((sizeof(uint64_t)*l1_size + BDRV_SECTOR_SIZE - 1)/
 862         BDRV_SECTOR_SIZE); i++) {
 863         ret = bdrv_pwrite(qcow_bs, header_size +
 864             BDRV_SECTOR_SIZE*i, tmp, BDRV_SECTOR_SIZE);
 865         if (ret != BDRV_SECTOR_SIZE) {
 866             g_free(tmp);
 867             goto exit;
 868         }
 869     }
 870
 871     g_free(tmp);
 872     ret = 0;
 873 exit:
 874     bdrv_unref(qcow_bs);
 875 cleanup:
 876     g_free(backing_file);
 877     return ret;
 878 }
 879
 880 static int qcow_make_empty(BlockDriverState *bs)
 881 {
 882     BDRVQcowState *s = bs->opaque;
 883     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
 884     int ret;
 885
 886     memset(s->l1_table, 0, l1_length);
 887     if (bdrv_pwrite_sync(bs->file->bs, s->l1_table_offset, s->l1_table,
 888             l1_length) < 0)
 889         return -1;
 890     ret = bdrv_truncate(bs->file->bs, s->l1_table_offset + l1_length);
 891     if (ret < 0)
 892         return ret;
 893
 894     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 895     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
 896     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
 897
 898     return 0;
 899 }
 900
 901 /* XXX: put compressed sectors first, then all the cluster aligned
 902    tables to avoid losing bytes in alignment */
 903 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
 904                                  const uint8_t *buf, int nb_sectors)
 905 {
 906     BDRVQcowState *s = bs->opaque;
 907     z_stream strm;
 908     int ret, out_len;
 909     uint8_t *out_buf;
 910     uint64_t cluster_offset;
 911
 912     if (nb_sectors != s->cluster_sectors) {
 913         ret = -EINVAL;
 914
 915         /* Zero-pad last write if image size is not cluster aligned */
 916         if (sector_num + nb_sectors == bs->total_sectors &&
 917             nb_sectors < s->cluster_sectors) {
 918             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
 919             memset(pad_buf, 0, s->cluster_size);
 920             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
 921             ret = qcow_write_compressed(bs, sector_num,
 922                                         pad_buf, s->cluster_sectors);
 923             qemu_vfree(pad_buf);
 924         }
 925         return ret;
 926     }
 927
 928     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
 929
 930     /* best compression, small window, no zlib header */
 931     memset(&strm, 0, sizeof(strm));
 932     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
 933                        Z_DEFLATED, -12,
 934                        9, Z_DEFAULT_STRATEGY);
 935     if (ret != 0) {
 936         ret = -EINVAL;
 937         goto fail;
 938     }
 939
 940     strm.avail_in = s->cluster_size;
 941     strm.next_in = (uint8_t *)buf;
 942     strm.avail_out = s->cluster_size;
 943     strm.next_out = out_buf;
 944
 945     ret = deflate(&strm, Z_FINISH);
 946     if (ret != Z_STREAM_END && ret != Z_OK) {
 947         deflateEnd(&strm);
 948         ret = -EINVAL;
 949         goto fail;
 950     }
 951     out_len = strm.next_out - out_buf;
 952
 953     deflateEnd(&strm);
 954
 955     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
 956         /* could not compress: write normal cluster */
 957         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
 958         if (ret < 0) {
 959             goto fail;
 960         }
 961     } else {
 962         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
 963                                             out_len, 0, 0);
 964         if (cluster_offset == 0) {
 965             ret = -EIO;
 966             goto fail;
 967         }
 968
 969         cluster_offset &= s->cluster_offset_mask;
 970         ret = bdrv_pwrite(bs->file->bs, cluster_offset, out_buf, out_len);
 971         if (ret < 0) {
 972             goto fail;
 973         }
 974     }
 975
 976     ret = 0;
 977 fail:
 978     g_free(out_buf);
 979     return ret;
 980 }
 981
 982 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 983 {
 984     BDRVQcowState *s = bs->opaque;
 985     bdi->cluster_size = s->cluster_size;
 986     return 0;
 987 }
 988
 989 static QemuOptsList qcow_create_opts = {
 990     .name = "qcow-create-opts",
 991     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
 992     .desc = {
 993         {
 994             .name = BLOCK_OPT_SIZE,
 995             .type = QEMU_OPT_SIZE,
 996             .help = "Virtual disk size"
 997         },
 998         {
 999             .name = BLOCK_OPT_BACKING_FILE,
1000             .type = QEMU_OPT_STRING,
1001             .help = "File name of a base image"
1002         },
1003         {
1004             .name = BLOCK_OPT_ENCRYPT,
1005             .type = QEMU_OPT_BOOL,
1006             .help = "Encrypt the image",
1007             .def_value_str = "off"
1008         },
1009         { /* end of list */ }
1010     }
1011 };
1012
1013 static BlockDriver bdrv_qcow = {
1014     .format_name        = "qcow",
1015     .instance_size      = sizeof(BDRVQcowState),
1016     .bdrv_probe         = qcow_probe,
1017     .bdrv_open          = qcow_open,
1018     .bdrv_close         = qcow_close,
1019     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1020     .bdrv_create            = qcow_create,
1021     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1022     .supports_backing       = true,
1023
1024     .bdrv_co_readv          = qcow_co_readv,
1025     .bdrv_co_writev         = qcow_co_writev,
1026     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1027
1028     .bdrv_set_key           = qcow_set_key,
1029     .bdrv_make_empty        = qcow_make_empty,
1030     .bdrv_write_compressed  = qcow_write_compressed,
1031     .bdrv_get_info          = qcow_get_info,
1032
1033     .create_opts            = &qcow_create_opts,
1034 };
1035
1036 static void bdrv_qcow_init(void)
1037 {
1038     bdrv_register(&bdrv_qcow);
1039 }
1040
1041 block_init(bdrv_qcow_init);