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