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