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     uint32_t crypt_method;
  52     uint64_t l1_table_offset;
  53 } QCowHeader;
  54
  55 #define L2_CACHE_SIZE 16
  56
  57 typedef struct BDRVQcowState {
  58     int cluster_bits;
  59     int cluster_size;
  60     int cluster_sectors;
  61     int l2_bits;
  62     int l2_size;
  63     int l1_size;
  64     uint64_t cluster_offset_mask;
  65     uint64_t l1_table_offset;
  66     uint64_t *l1_table;
  67     uint64_t *l2_cache;
  68     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
  69     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  70     uint8_t *cluster_cache;
  71     uint8_t *cluster_data;
  72     uint64_t cluster_cache_offset;
  73     uint32_t crypt_method; /* current crypt method, 0 if no key yet */
  74     uint32_t crypt_method_header;
  75     AES_KEY aes_encrypt_key;
  76     AES_KEY aes_decrypt_key;
  77     CoMutex lock;
  78     Error *migration_blocker;
  79 } BDRVQcowState;
  80
  81 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
  82
  83 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
  84 {
  85     const QCowHeader *cow_header = (const void *)buf;
  86
  87     if (buf_size >= sizeof(QCowHeader) &&
  88         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
  89         be32_to_cpu(cow_header->version) == QCOW_VERSION)
  90         return 100;
  91     else
  92         return 0;
  93 }
  94
  95 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
  96                      Error **errp)
  97 {
  98     BDRVQcowState *s = bs->opaque;
  99     int len, i, shift, ret;
 100     QCowHeader header;
 101
 102     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
 103     if (ret < 0) {
 104         goto fail;
 105     }
 106     be32_to_cpus(&header.magic);
 107     be32_to_cpus(&header.version);
 108     be64_to_cpus(&header.backing_file_offset);
 109     be32_to_cpus(&header.backing_file_size);
 110     be32_to_cpus(&header.mtime);
 111     be64_to_cpus(&header.size);
 112     be32_to_cpus(&header.crypt_method);
 113     be64_to_cpus(&header.l1_table_offset);
 114
 115     if (header.magic != QCOW_MAGIC) {
 116         ret = -EMEDIUMTYPE;
 117         goto fail;
 118     }
 119     if (header.version != QCOW_VERSION) {
 120         char version[64];
 121         snprintf(version, sizeof(version), "QCOW version %d", header.version);
 122         qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
 123             bs->device_name, "qcow", version);
 124         ret = -ENOTSUP;
 125         goto fail;
 126     }
 127
 128     if (header.size <= 1 || header.cluster_bits < 9) {
 129         ret = -EINVAL;
 130         goto fail;
 131     }
 132     if (header.crypt_method > QCOW_CRYPT_AES) {
 133         ret = -EINVAL;
 134         goto fail;
 135     }
 136     s->crypt_method_header = header.crypt_method;
 137     if (s->crypt_method_header) {
 138         bs->encrypted = 1;
 139     }
 140     s->cluster_bits = header.cluster_bits;
 141     s->cluster_size = 1 << s->cluster_bits;
 142     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 143     s->l2_bits = header.l2_bits;
 144     s->l2_size = 1 << s->l2_bits;
 145     bs->total_sectors = header.size / 512;
 146     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
 147
 148     /* read the level 1 table */
 149     shift = s->cluster_bits + s->l2_bits;
 150     s->l1_size = (header.size + (1LL << shift) - 1) >> shift;
 151
 152     s->l1_table_offset = header.l1_table_offset;
 153     s->l1_table = g_malloc(s->l1_size * sizeof(uint64_t));
 154
 155     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
 156                s->l1_size * sizeof(uint64_t));
 157     if (ret < 0) {
 158         goto fail;
 159     }
 160
 161     for(i = 0;i < s->l1_size; i++) {
 162         be64_to_cpus(&s->l1_table[i]);
 163     }
 164     /* alloc L2 cache */
 165     s->l2_cache = g_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 166     s->cluster_cache = g_malloc(s->cluster_size);
 167     s->cluster_data = g_malloc(s->cluster_size);
 168     s->cluster_cache_offset = -1;
 169
 170     /* read the backing file name */
 171     if (header.backing_file_offset != 0) {
 172         len = header.backing_file_size;
 173         if (len > 1023) {
 174             len = 1023;
 175         }
 176         ret = bdrv_pread(bs->file, header.backing_file_offset,
 177                    bs->backing_file, len);
 178         if (ret < 0) {
 179             goto fail;
 180         }
 181         bs->backing_file[len] = '\0';
 182     }
 183
 184     /* Disable migration when qcow images are used */
 185     error_set(&s->migration_blocker,
 186               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
 187               "qcow", bs->device_name, "live migration");
 188     migrate_add_blocker(s->migration_blocker);
 189
 190     qemu_co_mutex_init(&s->lock);
 191     return 0;
 192
 193  fail:
 194     g_free(s->l1_table);
 195     g_free(s->l2_cache);
 196     g_free(s->cluster_cache);
 197     g_free(s->cluster_data);
 198     return ret;
 199 }
 200
 201
 202 /* We have nothing to do for QCOW reopen, stubs just return
 203  * success */
 204 static int qcow_reopen_prepare(BDRVReopenState *state,
 205                                BlockReopenQueue *queue, Error **errp)
 206 {
 207     return 0;
 208 }
 209
 210 static int qcow_set_key(BlockDriverState *bs, const char *key)
 211 {
 212     BDRVQcowState *s = bs->opaque;
 213     uint8_t keybuf[16];
 214     int len, i;
 215
 216     memset(keybuf, 0, 16);
 217     len = strlen(key);
 218     if (len > 16)
 219         len = 16;
 220     /* XXX: we could compress the chars to 7 bits to increase
 221        entropy */
 222     for(i = 0;i < len;i++) {
 223         keybuf[i] = key[i];
 224     }
 225     s->crypt_method = s->crypt_method_header;
 226
 227     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
 228         return -1;
 229     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
 230         return -1;
 231     return 0;
 232 }
 233
 234 /* The crypt function is compatible with the linux cryptoloop
 235    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
 236    supported */
 237 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
 238                             uint8_t *out_buf, const uint8_t *in_buf,
 239                             int nb_sectors, int enc,
 240                             const AES_KEY *key)
 241 {
 242     union {
 243         uint64_t ll[2];
 244         uint8_t b[16];
 245     } ivec;
 246     int i;
 247
 248     for(i = 0; i < nb_sectors; i++) {
 249         ivec.ll[0] = cpu_to_le64(sector_num);
 250         ivec.ll[1] = 0;
 251         AES_cbc_encrypt(in_buf, out_buf, 512, key,
 252                         ivec.b, enc);
 253         sector_num++;
 254         in_buf += 512;
 255         out_buf += 512;
 256     }
 257 }
 258
 259 /* 'allocate' is:
 260  *
 261  * 0 to not allocate.
 262  *
 263  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
 264  * 'n_end')
 265  *
 266  * 2 to allocate a compressed cluster of size
 267  * 'compressed_size'. 'compressed_size' must be > 0 and <
 268  * cluster_size
 269  *
 270  * return 0 if not allocated.
 271  */
 272 static uint64_t get_cluster_offset(BlockDriverState *bs,
 273                                    uint64_t offset, int allocate,
 274                                    int compressed_size,
 275                                    int n_start, int n_end)
 276 {
 277     BDRVQcowState *s = bs->opaque;
 278     int min_index, i, j, l1_index, l2_index;
 279     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
 280     uint32_t min_count;
 281     int new_l2_table;
 282
 283     l1_index = offset >> (s->l2_bits + s->cluster_bits);
 284     l2_offset = s->l1_table[l1_index];
 285     new_l2_table = 0;
 286     if (!l2_offset) {
 287         if (!allocate)
 288             return 0;
 289         /* allocate a new l2 entry */
 290         l2_offset = bdrv_getlength(bs->file);
 291         /* round to cluster size */
 292         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
 293         /* update the L1 entry */
 294         s->l1_table[l1_index] = l2_offset;
 295         tmp = cpu_to_be64(l2_offset);
 296         if (bdrv_pwrite_sync(bs->file,
 297                 s->l1_table_offset + l1_index * sizeof(tmp),
 298                 &tmp, sizeof(tmp)) < 0)
 299             return 0;
 300         new_l2_table = 1;
 301     }
 302     for(i = 0; i < L2_CACHE_SIZE; i++) {
 303         if (l2_offset == s->l2_cache_offsets[i]) {
 304             /* increment the hit count */
 305             if (++s->l2_cache_counts[i] == 0xffffffff) {
 306                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 307                     s->l2_cache_counts[j] >>= 1;
 308                 }
 309             }
 310             l2_table = s->l2_cache + (i << s->l2_bits);
 311             goto found;
 312         }
 313     }
 314     /* not found: load a new entry in the least used one */
 315     min_index = 0;
 316     min_count = 0xffffffff;
 317     for(i = 0; i < L2_CACHE_SIZE; i++) {
 318         if (s->l2_cache_counts[i] < min_count) {
 319             min_count = s->l2_cache_counts[i];
 320             min_index = i;
 321         }
 322     }
 323     l2_table = s->l2_cache + (min_index << s->l2_bits);
 324     if (new_l2_table) {
 325         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
 326         if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
 327                 s->l2_size * sizeof(uint64_t)) < 0)
 328             return 0;
 329     } else {
 330         if (bdrv_pread(bs->file, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
 331             s->l2_size * sizeof(uint64_t))
 332             return 0;
 333     }
 334     s->l2_cache_offsets[min_index] = l2_offset;
 335     s->l2_cache_counts[min_index] = 1;
 336  found:
 337     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
 338     cluster_offset = be64_to_cpu(l2_table[l2_index]);
 339     if (!cluster_offset ||
 340         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
 341         if (!allocate)
 342             return 0;
 343         /* allocate a new cluster */
 344         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
 345             (n_end - n_start) < s->cluster_sectors) {
 346             /* if the cluster is already compressed, we must
 347                decompress it in the case it is not completely
 348                overwritten */
 349             if (decompress_cluster(bs, cluster_offset) < 0)
 350                 return 0;
 351             cluster_offset = bdrv_getlength(bs->file);
 352             cluster_offset = (cluster_offset + s->cluster_size - 1) &
 353                 ~(s->cluster_size - 1);
 354             /* write the cluster content */
 355             if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache, s->cluster_size) !=
 356                 s->cluster_size)
 357                 return -1;
 358         } else {
 359             cluster_offset = bdrv_getlength(bs->file);
 360             if (allocate == 1) {
 361                 /* round to cluster size */
 362                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
 363                     ~(s->cluster_size - 1);
 364                 bdrv_truncate(bs->file, cluster_offset + s->cluster_size);
 365                 /* if encrypted, we must initialize the cluster
 366                    content which won't be written */
 367                 if (s->crypt_method &&
 368                     (n_end - n_start) < s->cluster_sectors) {
 369                     uint64_t start_sect;
 370                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
 371                     memset(s->cluster_data + 512, 0x00, 512);
 372                     for(i = 0; i < s->cluster_sectors; i++) {
 373                         if (i < n_start || i >= n_end) {
 374                             encrypt_sectors(s, start_sect + i,
 375                                             s->cluster_data,
 376                                             s->cluster_data + 512, 1, 1,
 377                                             &s->aes_encrypt_key);
 378                             if (bdrv_pwrite(bs->file, cluster_offset + i * 512,
 379                                             s->cluster_data, 512) != 512)
 380                                 return -1;
 381                         }
 382                     }
 383                 }
 384             } else if (allocate == 2) {
 385                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
 386                     (uint64_t)compressed_size << (63 - s->cluster_bits);
 387             }
 388         }
 389         /* update L2 table */
 390         tmp = cpu_to_be64(cluster_offset);
 391         l2_table[l2_index] = tmp;
 392         if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
 393                 &tmp, sizeof(tmp)) < 0)
 394             return 0;
 395     }
 396     return cluster_offset;
 397 }
 398
 399 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
 400         int64_t sector_num, int nb_sectors, int *pnum)
 401 {
 402     BDRVQcowState *s = bs->opaque;
 403     int index_in_cluster, n;
 404     uint64_t cluster_offset;
 405
 406     qemu_co_mutex_lock(&s->lock);
 407     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
 408     qemu_co_mutex_unlock(&s->lock);
 409     index_in_cluster = sector_num & (s->cluster_sectors - 1);
 410     n = s->cluster_sectors - index_in_cluster;
 411     if (n > nb_sectors)
 412         n = nb_sectors;
 413     *pnum = n;
 414     if (!cluster_offset) {
 415         return 0;
 416     }
 417     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypt_method) {
 418         return BDRV_BLOCK_DATA;
 419     }
 420     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
 421     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
 422 }
 423
 424 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
 425                              const uint8_t *buf, int buf_size)
 426 {
 427     z_stream strm1, *strm = &strm1;
 428     int ret, out_len;
 429
 430     memset(strm, 0, sizeof(*strm));
 431
 432     strm->next_in = (uint8_t *)buf;
 433     strm->avail_in = buf_size;
 434     strm->next_out = out_buf;
 435     strm->avail_out = out_buf_size;
 436
 437     ret = inflateInit2(strm, -12);
 438     if (ret != Z_OK)
 439         return -1;
 440     ret = inflate(strm, Z_FINISH);
 441     out_len = strm->next_out - out_buf;
 442     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
 443         out_len != out_buf_size) {
 444         inflateEnd(strm);
 445         return -1;
 446     }
 447     inflateEnd(strm);
 448     return 0;
 449 }
 450
 451 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
 452 {
 453     BDRVQcowState *s = bs->opaque;
 454     int ret, csize;
 455     uint64_t coffset;
 456
 457     coffset = cluster_offset & s->cluster_offset_mask;
 458     if (s->cluster_cache_offset != coffset) {
 459         csize = cluster_offset >> (63 - s->cluster_bits);
 460         csize &= (s->cluster_size - 1);
 461         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
 462         if (ret != csize)
 463             return -1;
 464         if (decompress_buffer(s->cluster_cache, s->cluster_size,
 465                               s->cluster_data, csize) < 0) {
 466             return -1;
 467         }
 468         s->cluster_cache_offset = coffset;
 469     }
 470     return 0;
 471 }
 472
 473 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
 474                          int nb_sectors, QEMUIOVector *qiov)
 475 {
 476     BDRVQcowState *s = bs->opaque;
 477     int index_in_cluster;
 478     int ret = 0, n;
 479     uint64_t cluster_offset;
 480     struct iovec hd_iov;
 481     QEMUIOVector hd_qiov;
 482     uint8_t *buf;
 483     void *orig_buf;
 484
 485     if (qiov->niov > 1) {
 486         buf = orig_buf = qemu_blockalign(bs, qiov->size);
 487     } else {
 488         orig_buf = NULL;
 489         buf = (uint8_t *)qiov->iov->iov_base;
 490     }
 491
 492     qemu_co_mutex_lock(&s->lock);
 493
 494     while (nb_sectors != 0) {
 495         /* prepare next request */
 496         cluster_offset = get_cluster_offset(bs, sector_num << 9,
 497                                                  0, 0, 0, 0);
 498         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 499         n = s->cluster_sectors - index_in_cluster;
 500         if (n > nb_sectors) {
 501             n = nb_sectors;
 502         }
 503
 504         if (!cluster_offset) {
 505             if (bs->backing_hd) {
 506                 /* read from the base image */
 507                 hd_iov.iov_base = (void *)buf;
 508                 hd_iov.iov_len = n * 512;
 509                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 510                 qemu_co_mutex_unlock(&s->lock);
 511                 ret = bdrv_co_readv(bs->backing_hd, sector_num,
 512                                     n, &hd_qiov);
 513                 qemu_co_mutex_lock(&s->lock);
 514                 if (ret < 0) {
 515                     goto fail;
 516                 }
 517             } else {
 518                 /* Note: in this case, no need to wait */
 519                 memset(buf, 0, 512 * n);
 520             }
 521         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 522             /* add AIO support for compressed blocks ? */
 523             if (decompress_cluster(bs, cluster_offset) < 0) {
 524                 goto fail;
 525             }
 526             memcpy(buf,
 527                    s->cluster_cache + index_in_cluster * 512, 512 * n);
 528         } else {
 529             if ((cluster_offset & 511) != 0) {
 530                 goto fail;
 531             }
 532             hd_iov.iov_base = (void *)buf;
 533             hd_iov.iov_len = n * 512;
 534             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 535             qemu_co_mutex_unlock(&s->lock);
 536             ret = bdrv_co_readv(bs->file,
 537                                 (cluster_offset >> 9) + index_in_cluster,
 538                                 n, &hd_qiov);
 539             qemu_co_mutex_lock(&s->lock);
 540             if (ret < 0) {
 541                 break;
 542             }
 543             if (s->crypt_method) {
 544                 encrypt_sectors(s, sector_num, buf, buf,
 545                                 n, 0,
 546                                 &s->aes_decrypt_key);
 547             }
 548         }
 549         ret = 0;
 550
 551         nb_sectors -= n;
 552         sector_num += n;
 553         buf += n * 512;
 554     }
 555
 556 done:
 557     qemu_co_mutex_unlock(&s->lock);
 558
 559     if (qiov->niov > 1) {
 560         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
 561         qemu_vfree(orig_buf);
 562     }
 563
 564     return ret;
 565
 566 fail:
 567     ret = -EIO;
 568     goto done;
 569 }
 570
 571 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
 572                           int nb_sectors, QEMUIOVector *qiov)
 573 {
 574     BDRVQcowState *s = bs->opaque;
 575     int index_in_cluster;
 576     uint64_t cluster_offset;
 577     const uint8_t *src_buf;
 578     int ret = 0, n;
 579     uint8_t *cluster_data = NULL;
 580     struct iovec hd_iov;
 581     QEMUIOVector hd_qiov;
 582     uint8_t *buf;
 583     void *orig_buf;
 584
 585     s->cluster_cache_offset = -1; /* disable compressed cache */
 586
 587     if (qiov->niov > 1) {
 588         buf = orig_buf = qemu_blockalign(bs, qiov->size);
 589         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
 590     } else {
 591         orig_buf = NULL;
 592         buf = (uint8_t *)qiov->iov->iov_base;
 593     }
 594
 595     qemu_co_mutex_lock(&s->lock);
 596
 597     while (nb_sectors != 0) {
 598
 599         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 600         n = s->cluster_sectors - index_in_cluster;
 601         if (n > nb_sectors) {
 602             n = nb_sectors;
 603         }
 604         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
 605                                             index_in_cluster,
 606                                             index_in_cluster + n);
 607         if (!cluster_offset || (cluster_offset & 511) != 0) {
 608             ret = -EIO;
 609             break;
 610         }
 611         if (s->crypt_method) {
 612             if (!cluster_data) {
 613                 cluster_data = g_malloc0(s->cluster_size);
 614             }
 615             encrypt_sectors(s, sector_num, cluster_data, buf,
 616                             n, 1, &s->aes_encrypt_key);
 617             src_buf = cluster_data;
 618         } else {
 619             src_buf = buf;
 620         }
 621
 622         hd_iov.iov_base = (void *)src_buf;
 623         hd_iov.iov_len = n * 512;
 624         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 625         qemu_co_mutex_unlock(&s->lock);
 626         ret = bdrv_co_writev(bs->file,
 627                              (cluster_offset >> 9) + index_in_cluster,
 628                              n, &hd_qiov);
 629         qemu_co_mutex_lock(&s->lock);
 630         if (ret < 0) {
 631             break;
 632         }
 633         ret = 0;
 634
 635         nb_sectors -= n;
 636         sector_num += n;
 637         buf += n * 512;
 638     }
 639     qemu_co_mutex_unlock(&s->lock);
 640
 641     if (qiov->niov > 1) {
 642         qemu_vfree(orig_buf);
 643     }
 644     g_free(cluster_data);
 645
 646     return ret;
 647 }
 648
 649 static void qcow_close(BlockDriverState *bs)
 650 {
 651     BDRVQcowState *s = bs->opaque;
 652
 653     g_free(s->l1_table);
 654     g_free(s->l2_cache);
 655     g_free(s->cluster_cache);
 656     g_free(s->cluster_data);
 657
 658     migrate_del_blocker(s->migration_blocker);
 659     error_free(s->migration_blocker);
 660 }
 661
 662 static int qcow_create(const char *filename, QEMUOptionParameter *options,
 663                        Error **errp)
 664 {
 665     int header_size, backing_filename_len, l1_size, shift, i;
 666     QCowHeader header;
 667     uint8_t *tmp;
 668     int64_t total_size = 0;
 669     const char *backing_file = NULL;
 670     int flags = 0;
 671     Error *local_err = NULL;
 672     int ret;
 673     BlockDriverState *qcow_bs;
 674
 675     /* Read out options */
 676     while (options && options->name) {
 677         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 678             total_size = options->value.n / 512;
 679         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
 680             backing_file = options->value.s;
 681         } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
 682             flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
 683         }
 684         options++;
 685     }
 686
 687     ret = bdrv_create_file(filename, options, &local_err);
 688     if (ret < 0) {
 689         qerror_report_err(local_err);
 690         error_free(local_err);
 691         return ret;
 692     }
 693
 694     qcow_bs = NULL;
 695     ret = bdrv_open(&qcow_bs, filename, NULL, NULL,
 696                     BDRV_O_RDWR | BDRV_O_PROTOCOL, NULL, &local_err);
 697     if (ret < 0) {
 698         qerror_report_err(local_err);
 699         error_free(local_err);
 700         return ret;
 701     }
 702
 703     ret = bdrv_truncate(qcow_bs, 0);
 704     if (ret < 0) {
 705         goto exit;
 706     }
 707
 708     memset(&header, 0, sizeof(header));
 709     header.magic = cpu_to_be32(QCOW_MAGIC);
 710     header.version = cpu_to_be32(QCOW_VERSION);
 711     header.size = cpu_to_be64(total_size * 512);
 712     header_size = sizeof(header);
 713     backing_filename_len = 0;
 714     if (backing_file) {
 715         if (strcmp(backing_file, "fat:")) {
 716             header.backing_file_offset = cpu_to_be64(header_size);
 717             backing_filename_len = strlen(backing_file);
 718             header.backing_file_size = cpu_to_be32(backing_filename_len);
 719             header_size += backing_filename_len;
 720         } else {
 721             /* special backing file for vvfat */
 722             backing_file = NULL;
 723         }
 724         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
 725                                     unmodifyed sectors */
 726         header.l2_bits = 12; /* 32 KB L2 tables */
 727     } else {
 728         header.cluster_bits = 12; /* 4 KB clusters */
 729         header.l2_bits = 9; /* 4 KB L2 tables */
 730     }
 731     header_size = (header_size + 7) & ~7;
 732     shift = header.cluster_bits + header.l2_bits;
 733     l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
 734
 735     header.l1_table_offset = cpu_to_be64(header_size);
 736     if (flags & BLOCK_FLAG_ENCRYPT) {
 737         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
 738     } else {
 739         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
 740     }
 741
 742     /* write all the data */
 743     ret = bdrv_pwrite(qcow_bs, 0, &header, sizeof(header));
 744     if (ret != sizeof(header)) {
 745         goto exit;
 746     }
 747
 748     if (backing_file) {
 749         ret = bdrv_pwrite(qcow_bs, sizeof(header),
 750             backing_file, backing_filename_len);
 751         if (ret != backing_filename_len) {
 752             goto exit;
 753         }
 754     }
 755
 756     tmp = g_malloc0(BDRV_SECTOR_SIZE);
 757     for (i = 0; i < ((sizeof(uint64_t)*l1_size + BDRV_SECTOR_SIZE - 1)/
 758         BDRV_SECTOR_SIZE); i++) {
 759         ret = bdrv_pwrite(qcow_bs, header_size +
 760             BDRV_SECTOR_SIZE*i, tmp, BDRV_SECTOR_SIZE);
 761         if (ret != BDRV_SECTOR_SIZE) {
 762             g_free(tmp);
 763             goto exit;
 764         }
 765     }
 766
 767     g_free(tmp);
 768     ret = 0;
 769 exit:
 770     bdrv_unref(qcow_bs);
 771     return ret;
 772 }
 773
 774 static int qcow_make_empty(BlockDriverState *bs)
 775 {
 776     BDRVQcowState *s = bs->opaque;
 777     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
 778     int ret;
 779
 780     memset(s->l1_table, 0, l1_length);
 781     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
 782             l1_length) < 0)
 783         return -1;
 784     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
 785     if (ret < 0)
 786         return ret;
 787
 788     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 789     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
 790     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
 791
 792     return 0;
 793 }
 794
 795 /* XXX: put compressed sectors first, then all the cluster aligned
 796    tables to avoid losing bytes in alignment */
 797 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
 798                                  const uint8_t *buf, int nb_sectors)
 799 {
 800     BDRVQcowState *s = bs->opaque;
 801     z_stream strm;
 802     int ret, out_len;
 803     uint8_t *out_buf;
 804     uint64_t cluster_offset;
 805
 806     if (nb_sectors != s->cluster_sectors) {
 807         ret = -EINVAL;
 808
 809         /* Zero-pad last write if image size is not cluster aligned */
 810         if (sector_num + nb_sectors == bs->total_sectors &&
 811             nb_sectors < s->cluster_sectors) {
 812             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
 813             memset(pad_buf, 0, s->cluster_size);
 814             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
 815             ret = qcow_write_compressed(bs, sector_num,
 816                                         pad_buf, s->cluster_sectors);
 817             qemu_vfree(pad_buf);
 818         }
 819         return ret;
 820     }
 821
 822     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
 823
 824     /* best compression, small window, no zlib header */
 825     memset(&strm, 0, sizeof(strm));
 826     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
 827                        Z_DEFLATED, -12,
 828                        9, Z_DEFAULT_STRATEGY);
 829     if (ret != 0) {
 830         ret = -EINVAL;
 831         goto fail;
 832     }
 833
 834     strm.avail_in = s->cluster_size;
 835     strm.next_in = (uint8_t *)buf;
 836     strm.avail_out = s->cluster_size;
 837     strm.next_out = out_buf;
 838
 839     ret = deflate(&strm, Z_FINISH);
 840     if (ret != Z_STREAM_END && ret != Z_OK) {
 841         deflateEnd(&strm);
 842         ret = -EINVAL;
 843         goto fail;
 844     }
 845     out_len = strm.next_out - out_buf;
 846
 847     deflateEnd(&strm);
 848
 849     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
 850         /* could not compress: write normal cluster */
 851         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
 852         if (ret < 0) {
 853             goto fail;
 854         }
 855     } else {
 856         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
 857                                             out_len, 0, 0);
 858         if (cluster_offset == 0) {
 859             ret = -EIO;
 860             goto fail;
 861         }
 862
 863         cluster_offset &= s->cluster_offset_mask;
 864         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
 865         if (ret < 0) {
 866             goto fail;
 867         }
 868     }
 869
 870     ret = 0;
 871 fail:
 872     g_free(out_buf);
 873     return ret;
 874 }
 875
 876 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 877 {
 878     BDRVQcowState *s = bs->opaque;
 879     bdi->cluster_size = s->cluster_size;
 880     return 0;
 881 }
 882
 883
 884 static QEMUOptionParameter qcow_create_options[] = {
 885     {
 886         .name = BLOCK_OPT_SIZE,
 887         .type = OPT_SIZE,
 888         .help = "Virtual disk size"
 889     },
 890     {
 891         .name = BLOCK_OPT_BACKING_FILE,
 892         .type = OPT_STRING,
 893         .help = "File name of a base image"
 894     },
 895     {
 896         .name = BLOCK_OPT_ENCRYPT,
 897         .type = OPT_FLAG,
 898         .help = "Encrypt the image"
 899     },
 900     { NULL }
 901 };
 902
 903 static BlockDriver bdrv_qcow = {
 904     .format_name        = "qcow",
 905     .instance_size      = sizeof(BDRVQcowState),
 906     .bdrv_probe         = qcow_probe,
 907     .bdrv_open          = qcow_open,
 908     .bdrv_close         = qcow_close,
 909     .bdrv_reopen_prepare = qcow_reopen_prepare,
 910     .bdrv_create        = qcow_create,
 911     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
 912
 913     .bdrv_co_readv          = qcow_co_readv,
 914     .bdrv_co_writev         = qcow_co_writev,
 915     .bdrv_co_get_block_status   = qcow_co_get_block_status,
 916
 917     .bdrv_set_key           = qcow_set_key,
 918     .bdrv_make_empty        = qcow_make_empty,
 919     .bdrv_write_compressed  = qcow_write_compressed,
 920     .bdrv_get_info          = qcow_get_info,
 921
 922     .create_options = qcow_create_options,
 923 };
 924
 925 static void bdrv_qcow_init(void)
 926 {
 927     bdrv_register(&bdrv_qcow);
 928 }
 929
 930 block_init(bdrv_qcow_init);