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