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 {
  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 = -EMEDIUMTYPE;
 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 int64_t coroutine_fn qcow_co_get_block_status(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     if (!cluster_offset) {
 414         return 0;
 415     }
 416     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypt_method) {
 417         return BDRV_BLOCK_DATA;
 418     }
 419     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
 420     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
 421 }
 422
 423 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
 424                              const uint8_t *buf, int buf_size)
 425 {
 426     z_stream strm1, *strm = &strm1;
 427     int ret, out_len;
 428
 429     memset(strm, 0, sizeof(*strm));
 430
 431     strm->next_in = (uint8_t *)buf;
 432     strm->avail_in = buf_size;
 433     strm->next_out = out_buf;
 434     strm->avail_out = out_buf_size;
 435
 436     ret = inflateInit2(strm, -12);
 437     if (ret != Z_OK)
 438         return -1;
 439     ret = inflate(strm, Z_FINISH);
 440     out_len = strm->next_out - out_buf;
 441     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
 442         out_len != out_buf_size) {
 443         inflateEnd(strm);
 444         return -1;
 445     }
 446     inflateEnd(strm);
 447     return 0;
 448 }
 449
 450 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
 451 {
 452     BDRVQcowState *s = bs->opaque;
 453     int ret, csize;
 454     uint64_t coffset;
 455
 456     coffset = cluster_offset & s->cluster_offset_mask;
 457     if (s->cluster_cache_offset != coffset) {
 458         csize = cluster_offset >> (63 - s->cluster_bits);
 459         csize &= (s->cluster_size - 1);
 460         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
 461         if (ret != csize)
 462             return -1;
 463         if (decompress_buffer(s->cluster_cache, s->cluster_size,
 464                               s->cluster_data, csize) < 0) {
 465             return -1;
 466         }
 467         s->cluster_cache_offset = coffset;
 468     }
 469     return 0;
 470 }
 471
 472 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
 473                          int nb_sectors, QEMUIOVector *qiov)
 474 {
 475     BDRVQcowState *s = bs->opaque;
 476     int index_in_cluster;
 477     int ret = 0, n;
 478     uint64_t cluster_offset;
 479     struct iovec hd_iov;
 480     QEMUIOVector hd_qiov;
 481     uint8_t *buf;
 482     void *orig_buf;
 483
 484     if (qiov->niov > 1) {
 485         buf = orig_buf = qemu_blockalign(bs, qiov->size);
 486     } else {
 487         orig_buf = NULL;
 488         buf = (uint8_t *)qiov->iov->iov_base;
 489     }
 490
 491     qemu_co_mutex_lock(&s->lock);
 492
 493     while (nb_sectors != 0) {
 494         /* prepare next request */
 495         cluster_offset = get_cluster_offset(bs, sector_num << 9,
 496                                                  0, 0, 0, 0);
 497         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 498         n = s->cluster_sectors - index_in_cluster;
 499         if (n > nb_sectors) {
 500             n = nb_sectors;
 501         }
 502
 503         if (!cluster_offset) {
 504             if (bs->backing_hd) {
 505                 /* read from the base image */
 506                 hd_iov.iov_base = (void *)buf;
 507                 hd_iov.iov_len = n * 512;
 508                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 509                 qemu_co_mutex_unlock(&s->lock);
 510                 ret = bdrv_co_readv(bs->backing_hd, sector_num,
 511                                     n, &hd_qiov);
 512                 qemu_co_mutex_lock(&s->lock);
 513                 if (ret < 0) {
 514                     goto fail;
 515                 }
 516             } else {
 517                 /* Note: in this case, no need to wait */
 518                 memset(buf, 0, 512 * n);
 519             }
 520         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 521             /* add AIO support for compressed blocks ? */
 522             if (decompress_cluster(bs, cluster_offset) < 0) {
 523                 goto fail;
 524             }
 525             memcpy(buf,
 526                    s->cluster_cache + index_in_cluster * 512, 512 * n);
 527         } else {
 528             if ((cluster_offset & 511) != 0) {
 529                 goto fail;
 530             }
 531             hd_iov.iov_base = (void *)buf;
 532             hd_iov.iov_len = n * 512;
 533             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 534             qemu_co_mutex_unlock(&s->lock);
 535             ret = bdrv_co_readv(bs->file,
 536                                 (cluster_offset >> 9) + index_in_cluster,
 537                                 n, &hd_qiov);
 538             qemu_co_mutex_lock(&s->lock);
 539             if (ret < 0) {
 540                 break;
 541             }
 542             if (s->crypt_method) {
 543                 encrypt_sectors(s, sector_num, buf, buf,
 544                                 n, 0,
 545                                 &s->aes_decrypt_key);
 546             }
 547         }
 548         ret = 0;
 549
 550         nb_sectors -= n;
 551         sector_num += n;
 552         buf += n * 512;
 553     }
 554
 555 done:
 556     qemu_co_mutex_unlock(&s->lock);
 557
 558     if (qiov->niov > 1) {
 559         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
 560         qemu_vfree(orig_buf);
 561     }
 562
 563     return ret;
 564
 565 fail:
 566     ret = -EIO;
 567     goto done;
 568 }
 569
 570 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
 571                           int nb_sectors, QEMUIOVector *qiov)
 572 {
 573     BDRVQcowState *s = bs->opaque;
 574     int index_in_cluster;
 575     uint64_t cluster_offset;
 576     const uint8_t *src_buf;
 577     int ret = 0, n;
 578     uint8_t *cluster_data = NULL;
 579     struct iovec hd_iov;
 580     QEMUIOVector hd_qiov;
 581     uint8_t *buf;
 582     void *orig_buf;
 583
 584     s->cluster_cache_offset = -1; /* disable compressed cache */
 585
 586     if (qiov->niov > 1) {
 587         buf = orig_buf = qemu_blockalign(bs, qiov->size);
 588         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
 589     } else {
 590         orig_buf = NULL;
 591         buf = (uint8_t *)qiov->iov->iov_base;
 592     }
 593
 594     qemu_co_mutex_lock(&s->lock);
 595
 596     while (nb_sectors != 0) {
 597
 598         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 599         n = s->cluster_sectors - index_in_cluster;
 600         if (n > nb_sectors) {
 601             n = nb_sectors;
 602         }
 603         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
 604                                             index_in_cluster,
 605                                             index_in_cluster + n);
 606         if (!cluster_offset || (cluster_offset & 511) != 0) {
 607             ret = -EIO;
 608             break;
 609         }
 610         if (s->crypt_method) {
 611             if (!cluster_data) {
 612                 cluster_data = g_malloc0(s->cluster_size);
 613             }
 614             encrypt_sectors(s, sector_num, cluster_data, buf,
 615                             n, 1, &s->aes_encrypt_key);
 616             src_buf = cluster_data;
 617         } else {
 618             src_buf = buf;
 619         }
 620
 621         hd_iov.iov_base = (void *)src_buf;
 622         hd_iov.iov_len = n * 512;
 623         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 624         qemu_co_mutex_unlock(&s->lock);
 625         ret = bdrv_co_writev(bs->file,
 626                              (cluster_offset >> 9) + index_in_cluster,
 627                              n, &hd_qiov);
 628         qemu_co_mutex_lock(&s->lock);
 629         if (ret < 0) {
 630             break;
 631         }
 632         ret = 0;
 633
 634         nb_sectors -= n;
 635         sector_num += n;
 636         buf += n * 512;
 637     }
 638     qemu_co_mutex_unlock(&s->lock);
 639
 640     if (qiov->niov > 1) {
 641         qemu_vfree(orig_buf);
 642     }
 643     g_free(cluster_data);
 644
 645     return ret;
 646 }
 647
 648 static void qcow_close(BlockDriverState *bs)
 649 {
 650     BDRVQcowState *s = bs->opaque;
 651
 652     g_free(s->l1_table);
 653     g_free(s->l2_cache);
 654     g_free(s->cluster_cache);
 655     g_free(s->cluster_data);
 656
 657     migrate_del_blocker(s->migration_blocker);
 658     error_free(s->migration_blocker);
 659 }
 660
 661 static int qcow_create(const char *filename, QEMUOptionParameter *options)
 662 {
 663     int header_size, backing_filename_len, l1_size, shift, i;
 664     QCowHeader header;
 665     uint8_t *tmp;
 666     int64_t total_size = 0;
 667     const char *backing_file = NULL;
 668     int flags = 0;
 669     int ret;
 670     BlockDriverState *qcow_bs;
 671
 672     /* Read out options */
 673     while (options && options->name) {
 674         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 675             total_size = options->value.n / 512;
 676         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
 677             backing_file = options->value.s;
 678         } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
 679             flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
 680         }
 681         options++;
 682     }
 683
 684     ret = bdrv_create_file(filename, options);
 685     if (ret < 0) {
 686         return ret;
 687     }
 688
 689     ret = bdrv_file_open(&qcow_bs, filename, NULL, BDRV_O_RDWR);
 690     if (ret < 0) {
 691         return ret;
 692     }
 693
 694     ret = bdrv_truncate(qcow_bs, 0);
 695     if (ret < 0) {
 696         goto exit;
 697     }
 698
 699     memset(&header, 0, sizeof(header));
 700     header.magic = cpu_to_be32(QCOW_MAGIC);
 701     header.version = cpu_to_be32(QCOW_VERSION);
 702     header.size = cpu_to_be64(total_size * 512);
 703     header_size = sizeof(header);
 704     backing_filename_len = 0;
 705     if (backing_file) {
 706         if (strcmp(backing_file, "fat:")) {
 707             header.backing_file_offset = cpu_to_be64(header_size);
 708             backing_filename_len = strlen(backing_file);
 709             header.backing_file_size = cpu_to_be32(backing_filename_len);
 710             header_size += backing_filename_len;
 711         } else {
 712             /* special backing file for vvfat */
 713             backing_file = NULL;
 714         }
 715         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
 716                                     unmodifyed sectors */
 717         header.l2_bits = 12; /* 32 KB L2 tables */
 718     } else {
 719         header.cluster_bits = 12; /* 4 KB clusters */
 720         header.l2_bits = 9; /* 4 KB L2 tables */
 721     }
 722     header_size = (header_size + 7) & ~7;
 723     shift = header.cluster_bits + header.l2_bits;
 724     l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
 725
 726     header.l1_table_offset = cpu_to_be64(header_size);
 727     if (flags & BLOCK_FLAG_ENCRYPT) {
 728         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
 729     } else {
 730         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
 731     }
 732
 733     /* write all the data */
 734     ret = bdrv_pwrite(qcow_bs, 0, &header, sizeof(header));
 735     if (ret != sizeof(header)) {
 736         goto exit;
 737     }
 738
 739     if (backing_file) {
 740         ret = bdrv_pwrite(qcow_bs, sizeof(header),
 741             backing_file, backing_filename_len);
 742         if (ret != backing_filename_len) {
 743             goto exit;
 744         }
 745     }
 746
 747     tmp = g_malloc0(BDRV_SECTOR_SIZE);
 748     for (i = 0; i < ((sizeof(uint64_t)*l1_size + BDRV_SECTOR_SIZE - 1)/
 749         BDRV_SECTOR_SIZE); i++) {
 750         ret = bdrv_pwrite(qcow_bs, header_size +
 751             BDRV_SECTOR_SIZE*i, tmp, BDRV_SECTOR_SIZE);
 752         if (ret != BDRV_SECTOR_SIZE) {
 753             g_free(tmp);
 754             goto exit;
 755         }
 756     }
 757
 758     g_free(tmp);
 759     ret = 0;
 760 exit:
 761     bdrv_unref(qcow_bs);
 762     return ret;
 763 }
 764
 765 static int qcow_make_empty(BlockDriverState *bs)
 766 {
 767     BDRVQcowState *s = bs->opaque;
 768     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
 769     int ret;
 770
 771     memset(s->l1_table, 0, l1_length);
 772     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
 773             l1_length) < 0)
 774         return -1;
 775     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
 776     if (ret < 0)
 777         return ret;
 778
 779     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 780     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
 781     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
 782
 783     return 0;
 784 }
 785
 786 /* XXX: put compressed sectors first, then all the cluster aligned
 787    tables to avoid losing bytes in alignment */
 788 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
 789                                  const uint8_t *buf, int nb_sectors)
 790 {
 791     BDRVQcowState *s = bs->opaque;
 792     z_stream strm;
 793     int ret, out_len;
 794     uint8_t *out_buf;
 795     uint64_t cluster_offset;
 796
 797     if (nb_sectors != s->cluster_sectors) {
 798         ret = -EINVAL;
 799
 800         /* Zero-pad last write if image size is not cluster aligned */
 801         if (sector_num + nb_sectors == bs->total_sectors &&
 802             nb_sectors < s->cluster_sectors) {
 803             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
 804             memset(pad_buf, 0, s->cluster_size);
 805             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
 806             ret = qcow_write_compressed(bs, sector_num,
 807                                         pad_buf, s->cluster_sectors);
 808             qemu_vfree(pad_buf);
 809         }
 810         return ret;
 811     }
 812
 813     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
 814
 815     /* best compression, small window, no zlib header */
 816     memset(&strm, 0, sizeof(strm));
 817     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
 818                        Z_DEFLATED, -12,
 819                        9, Z_DEFAULT_STRATEGY);
 820     if (ret != 0) {
 821         ret = -EINVAL;
 822         goto fail;
 823     }
 824
 825     strm.avail_in = s->cluster_size;
 826     strm.next_in = (uint8_t *)buf;
 827     strm.avail_out = s->cluster_size;
 828     strm.next_out = out_buf;
 829
 830     ret = deflate(&strm, Z_FINISH);
 831     if (ret != Z_STREAM_END && ret != Z_OK) {
 832         deflateEnd(&strm);
 833         ret = -EINVAL;
 834         goto fail;
 835     }
 836     out_len = strm.next_out - out_buf;
 837
 838     deflateEnd(&strm);
 839
 840     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
 841         /* could not compress: write normal cluster */
 842         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
 843         if (ret < 0) {
 844             goto fail;
 845         }
 846     } else {
 847         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
 848                                             out_len, 0, 0);
 849         if (cluster_offset == 0) {
 850             ret = -EIO;
 851             goto fail;
 852         }
 853
 854         cluster_offset &= s->cluster_offset_mask;
 855         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
 856         if (ret < 0) {
 857             goto fail;
 858         }
 859     }
 860
 861     ret = 0;
 862 fail:
 863     g_free(out_buf);
 864     return ret;
 865 }
 866
 867 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 868 {
 869     BDRVQcowState *s = bs->opaque;
 870     bdi->cluster_size = s->cluster_size;
 871     return 0;
 872 }
 873
 874
 875 static QEMUOptionParameter qcow_create_options[] = {
 876     {
 877         .name = BLOCK_OPT_SIZE,
 878         .type = OPT_SIZE,
 879         .help = "Virtual disk size"
 880     },
 881     {
 882         .name = BLOCK_OPT_BACKING_FILE,
 883         .type = OPT_STRING,
 884         .help = "File name of a base image"
 885     },
 886     {
 887         .name = BLOCK_OPT_ENCRYPT,
 888         .type = OPT_FLAG,
 889         .help = "Encrypt the image"
 890     },
 891     { NULL }
 892 };
 893
 894 static BlockDriver bdrv_qcow = {
 895     .format_name        = "qcow",
 896     .instance_size      = sizeof(BDRVQcowState),
 897     .bdrv_probe         = qcow_probe,
 898     .bdrv_open          = qcow_open,
 899     .bdrv_close         = qcow_close,
 900     .bdrv_reopen_prepare = qcow_reopen_prepare,
 901     .bdrv_create        = qcow_create,
 902     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
 903
 904     .bdrv_co_readv          = qcow_co_readv,
 905     .bdrv_co_writev         = qcow_co_writev,
 906     .bdrv_co_get_block_status   = qcow_co_get_block_status,
 907
 908     .bdrv_set_key           = qcow_set_key,
 909     .bdrv_make_empty        = qcow_make_empty,
 910     .bdrv_write_compressed  = qcow_write_compressed,
 911     .bdrv_get_info          = qcow_get_info,
 912
 913     .create_options = qcow_create_options,
 914 };
 915
 916 static void bdrv_qcow_init(void)
 917 {
 918     bdrv_register(&bdrv_qcow);
 919 }
 920
 921 block_init(bdrv_qcow_init);