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