block/qcow.c

   1 /*
   2  * Block driver for the QCOW format
   3  *
   4  * Copyright (c) 2004-2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu-common.h"
  25 #include "block/block_int.h"
  26 #include "qemu/module.h"
  27 #include <zlib.h>
  28 #include "qemu/aes.h"
  29 #include "migration/migration.h"
  30
  31 /**************************************************************/
  32 /* QEMU COW block driver with compression and encryption support */
  33
  34 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
  35 #define QCOW_VERSION 1
  36
  37 #define QCOW_CRYPT_NONE 0
  38 #define QCOW_CRYPT_AES  1
  39
  40 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
  41
  42 typedef struct QCowHeader {
  43     uint32_t magic;
  44     uint32_t version;
  45     uint64_t backing_file_offset;
  46     uint32_t backing_file_size;
  47     uint32_t mtime;
  48     uint64_t size; /* in bytes */
  49     uint8_t cluster_bits;
  50     uint8_t l2_bits;
  51     uint32_t crypt_method;
  52     uint64_t l1_table_offset;
  53 } QCowHeader;
  54
  55 #define L2_CACHE_SIZE 16
  56
  57 typedef struct BDRVQcowState {
  58     int cluster_bits;
  59     int cluster_size;
  60     int cluster_sectors;
  61     int l2_bits;
  62     int l2_size;
  63     int l1_size;
  64     uint64_t cluster_offset_mask;
  65     uint64_t l1_table_offset;
  66     uint64_t *l1_table;
  67     uint64_t *l2_cache;
  68     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
  69     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  70     uint8_t *cluster_cache;
  71     uint8_t *cluster_data;
  72     uint64_t cluster_cache_offset;
  73     uint32_t crypt_method; /* current crypt method, 0 if no key yet */
  74     uint32_t crypt_method_header;
  75     AES_KEY aes_encrypt_key;
  76     AES_KEY aes_decrypt_key;
  77     CoMutex lock;
  78     Error *migration_blocker;
  79 } BDRVQcowState;
  80
  81 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
  82
  83 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
  84 {
  85     const QCowHeader *cow_header = (const void *)buf;
  86
  87     if (buf_size >= sizeof(QCowHeader) &&
  88         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
  89         be32_to_cpu(cow_header->version) == QCOW_VERSION)
  90         return 100;
  91     else
  92         return 0;
  93 }
  94
  95 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
  96                      Error **errp)
  97 {
  98     BDRVQcowState *s = bs->opaque;
  99     int len, i, shift, ret;
 100     QCowHeader header;
 101
 102     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
 103     if (ret < 0) {
 104         goto fail;
 105     }
 106     be32_to_cpus(&header.magic);
 107     be32_to_cpus(&header.version);
 108     be64_to_cpus(&header.backing_file_offset);
 109     be32_to_cpus(&header.backing_file_size);
 110     be32_to_cpus(&header.mtime);
 111     be64_to_cpus(&header.size);
 112     be32_to_cpus(&header.crypt_method);
 113     be64_to_cpus(&header.l1_table_offset);
 114
 115     if (header.magic != QCOW_MAGIC) {
 116         error_setg(errp, "Image not in qcow format");
 117         ret = -EINVAL;
 118         goto fail;
 119     }
 120     if (header.version != QCOW_VERSION) {
 121         char version[64];
 122         snprintf(version, sizeof(version), "QCOW version %" PRIu32,
 123                  header.version);
 124         error_set(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
 125                   bs->device_name, "qcow", version);
 126         ret = -ENOTSUP;
 127         goto fail;
 128     }
 129
 130     if (header.size <= 1 || header.cluster_bits < 9) {
 131         error_setg(errp, "invalid value in qcow header");
 132         ret = -EINVAL;
 133         goto fail;
 134     }
 135     if (header.crypt_method > QCOW_CRYPT_AES) {
 136         error_setg(errp, "invalid encryption method in qcow header");
 137         ret = -EINVAL;
 138         goto fail;
 139     }
 140     s->crypt_method_header = header.crypt_method;
 141     if (s->crypt_method_header) {
 142         bs->encrypted = 1;
 143     }
 144     s->cluster_bits = header.cluster_bits;
 145     s->cluster_size = 1 << s->cluster_bits;
 146     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 147     s->l2_bits = header.l2_bits;
 148     s->l2_size = 1 << s->l2_bits;
 149     bs->total_sectors = header.size / 512;
 150     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
 151
 152     /* read the level 1 table */
 153     shift = s->cluster_bits + s->l2_bits;
 154     s->l1_size = (header.size + (1LL << shift) - 1) >> shift;
 155
 156     s->l1_table_offset = header.l1_table_offset;
 157     s->l1_table = g_malloc(s->l1_size * sizeof(uint64_t));
 158
 159     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
 160                s->l1_size * sizeof(uint64_t));
 161     if (ret < 0) {
 162         goto fail;
 163     }
 164
 165     for(i = 0;i < s->l1_size; i++) {
 166         be64_to_cpus(&s->l1_table[i]);
 167     }
 168     /* alloc L2 cache */
 169     s->l2_cache = g_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 170     s->cluster_cache = g_malloc(s->cluster_size);
 171     s->cluster_data = g_malloc(s->cluster_size);
 172     s->cluster_cache_offset = -1;
 173
 174     /* read the backing file name */
 175     if (header.backing_file_offset != 0) {
 176         len = header.backing_file_size;
 177         if (len > 1023) {
 178             len = 1023;
 179         }
 180         ret = bdrv_pread(bs->file, header.backing_file_offset,
 181                    bs->backing_file, len);
 182         if (ret < 0) {
 183             goto fail;
 184         }
 185         bs->backing_file[len] = '\0';
 186     }
 187
 188     /* Disable migration when qcow images are used */
 189     error_set(&s->migration_blocker,
 190               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
 191               "qcow", bs->device_name, "live migration");
 192     migrate_add_blocker(s->migration_blocker);
 193
 194     qemu_co_mutex_init(&s->lock);
 195     return 0;
 196
 197  fail:
 198     g_free(s->l1_table);
 199     g_free(s->l2_cache);
 200     g_free(s->cluster_cache);
 201     g_free(s->cluster_data);
 202     return ret;
 203 }
 204
 205
 206 /* We have nothing to do for QCOW reopen, stubs just return
 207  * success */
 208 static int qcow_reopen_prepare(BDRVReopenState *state,
 209                                BlockReopenQueue *queue, Error **errp)
 210 {
 211     return 0;
 212 }
 213
 214 static int qcow_set_key(BlockDriverState *bs, const char *key)
 215 {
 216     BDRVQcowState *s = bs->opaque;
 217     uint8_t keybuf[16];
 218     int len, i;
 219
 220     memset(keybuf, 0, 16);
 221     len = strlen(key);
 222     if (len > 16)
 223         len = 16;
 224     /* XXX: we could compress the chars to 7 bits to increase
 225        entropy */
 226     for(i = 0;i < len;i++) {
 227         keybuf[i] = key[i];
 228     }
 229     s->crypt_method = s->crypt_method_header;
 230
 231     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
 232         return -1;
 233     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
 234         return -1;
 235     return 0;
 236 }
 237
 238 /* The crypt function is compatible with the linux cryptoloop
 239    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
 240    supported */
 241 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
 242                             uint8_t *out_buf, const uint8_t *in_buf,
 243                             int nb_sectors, int enc,
 244                             const AES_KEY *key)
 245 {
 246     union {
 247         uint64_t ll[2];
 248         uint8_t b[16];
 249     } ivec;
 250     int i;
 251
 252     for(i = 0; i < nb_sectors; i++) {
 253         ivec.ll[0] = cpu_to_le64(sector_num);
 254         ivec.ll[1] = 0;
 255         AES_cbc_encrypt(in_buf, out_buf, 512, key,
 256                         ivec.b, enc);
 257         sector_num++;
 258         in_buf += 512;
 259         out_buf += 512;
 260     }
 261 }
 262
 263 /* 'allocate' is:
 264  *
 265  * 0 to not allocate.
 266  *
 267  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
 268  * 'n_end')
 269  *
 270  * 2 to allocate a compressed cluster of size
 271  * 'compressed_size'. 'compressed_size' must be > 0 and <
 272  * cluster_size
 273  *
 274  * return 0 if not allocated.
 275  */
 276 static uint64_t get_cluster_offset(BlockDriverState *bs,
 277                                    uint64_t offset, int allocate,
 278                                    int compressed_size,
 279                                    int n_start, int n_end)
 280 {
 281     BDRVQcowState *s = bs->opaque;
 282     int min_index, i, j, l1_index, l2_index;
 283     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
 284     uint32_t min_count;
 285     int new_l2_table;
 286
 287     l1_index = offset >> (s->l2_bits + s->cluster_bits);
 288     l2_offset = s->l1_table[l1_index];
 289     new_l2_table = 0;
 290     if (!l2_offset) {
 291         if (!allocate)
 292             return 0;
 293         /* allocate a new l2 entry */
 294         l2_offset = bdrv_getlength(bs->file);
 295         /* round to cluster size */
 296         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
 297         /* update the L1 entry */
 298         s->l1_table[l1_index] = l2_offset;
 299         tmp = cpu_to_be64(l2_offset);
 300         if (bdrv_pwrite_sync(bs->file,
 301                 s->l1_table_offset + l1_index * sizeof(tmp),
 302                 &tmp, sizeof(tmp)) < 0)
 303             return 0;
 304         new_l2_table = 1;
 305     }
 306     for(i = 0; i < L2_CACHE_SIZE; i++) {
 307         if (l2_offset == s->l2_cache_offsets[i]) {
 308             /* increment the hit count */
 309             if (++s->l2_cache_counts[i] == 0xffffffff) {
 310                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 311                     s->l2_cache_counts[j] >>= 1;
 312                 }
 313             }
 314             l2_table = s->l2_cache + (i << s->l2_bits);
 315             goto found;
 316         }
 317     }
 318     /* not found: load a new entry in the least used one */
 319     min_index = 0;
 320     min_count = 0xffffffff;
 321     for(i = 0; i < L2_CACHE_SIZE; i++) {
 322         if (s->l2_cache_counts[i] < min_count) {
 323             min_count = s->l2_cache_counts[i];
 324             min_index = i;
 325         }
 326     }
 327     l2_table = s->l2_cache + (min_index << s->l2_bits);
 328     if (new_l2_table) {
 329         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
 330         if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
 331                 s->l2_size * sizeof(uint64_t)) < 0)
 332             return 0;
 333     } else {
 334         if (bdrv_pread(bs->file, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
 335             s->l2_size * sizeof(uint64_t))
 336             return 0;
 337     }
 338     s->l2_cache_offsets[min_index] = l2_offset;
 339     s->l2_cache_counts[min_index] = 1;
 340  found:
 341     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
 342     cluster_offset = be64_to_cpu(l2_table[l2_index]);
 343     if (!cluster_offset ||
 344         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
 345         if (!allocate)
 346             return 0;
 347         /* allocate a new cluster */
 348         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
 349             (n_end - n_start) < s->cluster_sectors) {
 350             /* if the cluster is already compressed, we must
 351                decompress it in the case it is not completely
 352                overwritten */
 353             if (decompress_cluster(bs, cluster_offset) < 0)
 354                 return 0;
 355             cluster_offset = bdrv_getlength(bs->file);
 356             cluster_offset = (cluster_offset + s->cluster_size - 1) &
 357                 ~(s->cluster_size - 1);
 358             /* write the cluster content */
 359             if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache, s->cluster_size) !=
 360                 s->cluster_size)
 361                 return -1;
 362         } else {
 363             cluster_offset = bdrv_getlength(bs->file);
 364             if (allocate == 1) {
 365                 /* round to cluster size */
 366                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
 367                     ~(s->cluster_size - 1);
 368                 bdrv_truncate(bs->file, cluster_offset + s->cluster_size);
 369                 /* if encrypted, we must initialize the cluster
 370                    content which won't be written */
 371                 if (s->crypt_method &&
 372                     (n_end - n_start) < s->cluster_sectors) {
 373                     uint64_t start_sect;
 374                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
 375                     memset(s->cluster_data + 512, 0x00, 512);
 376                     for(i = 0; i < s->cluster_sectors; i++) {
 377                         if (i < n_start || i >= n_end) {
 378                             encrypt_sectors(s, start_sect + i,
 379                                             s->cluster_data,
 380                                             s->cluster_data + 512, 1, 1,
 381                                             &s->aes_encrypt_key);
 382                             if (bdrv_pwrite(bs->file, cluster_offset + i * 512,
 383                                             s->cluster_data, 512) != 512)
 384                                 return -1;
 385                         }
 386                     }
 387                 }
 388             } else if (allocate == 2) {
 389                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
 390                     (uint64_t)compressed_size << (63 - s->cluster_bits);
 391             }
 392         }
 393         /* update L2 table */
 394         tmp = cpu_to_be64(cluster_offset);
 395         l2_table[l2_index] = tmp;
 396         if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
 397                 &tmp, sizeof(tmp)) < 0)
 398             return 0;
 399     }
 400     return cluster_offset;
 401 }
 402
 403 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
 404         int64_t sector_num, int nb_sectors, int *pnum)
 405 {
 406     BDRVQcowState *s = bs->opaque;
 407     int index_in_cluster, n;
 408     uint64_t cluster_offset;
 409
 410     qemu_co_mutex_lock(&s->lock);
 411     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
 412     qemu_co_mutex_unlock(&s->lock);
 413     index_in_cluster = sector_num & (s->cluster_sectors - 1);
 414     n = s->cluster_sectors - index_in_cluster;
 415     if (n > nb_sectors)
 416         n = nb_sectors;
 417     *pnum = n;
 418     if (!cluster_offset) {
 419         return 0;
 420     }
 421     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypt_method) {
 422         return BDRV_BLOCK_DATA;
 423     }
 424     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
 425     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
 426 }
 427
 428 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
 429                              const uint8_t *buf, int buf_size)
 430 {
 431     z_stream strm1, *strm = &strm1;
 432     int ret, out_len;
 433
 434     memset(strm, 0, sizeof(*strm));
 435
 436     strm->next_in = (uint8_t *)buf;
 437     strm->avail_in = buf_size;
 438     strm->next_out = out_buf;
 439     strm->avail_out = out_buf_size;
 440
 441     ret = inflateInit2(strm, -12);
 442     if (ret != Z_OK)
 443         return -1;
 444     ret = inflate(strm, Z_FINISH);
 445     out_len = strm->next_out - out_buf;
 446     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
 447         out_len != out_buf_size) {
 448         inflateEnd(strm);
 449         return -1;
 450     }
 451     inflateEnd(strm);
 452     return 0;
 453 }
 454
 455 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
 456 {
 457     BDRVQcowState *s = bs->opaque;
 458     int ret, csize;
 459     uint64_t coffset;
 460
 461     coffset = cluster_offset & s->cluster_offset_mask;
 462     if (s->cluster_cache_offset != coffset) {
 463         csize = cluster_offset >> (63 - s->cluster_bits);
 464         csize &= (s->cluster_size - 1);
 465         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
 466         if (ret != csize)
 467             return -1;
 468         if (decompress_buffer(s->cluster_cache, s->cluster_size,
 469                               s->cluster_data, csize) < 0) {
 470             return -1;
 471         }
 472         s->cluster_cache_offset = coffset;
 473     }
 474     return 0;
 475 }
 476
 477 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
 478                          int nb_sectors, QEMUIOVector *qiov)
 479 {
 480     BDRVQcowState *s = bs->opaque;
 481     int index_in_cluster;
 482     int ret = 0, n;
 483     uint64_t cluster_offset;
 484     struct iovec hd_iov;
 485     QEMUIOVector hd_qiov;
 486     uint8_t *buf;
 487     void *orig_buf;
 488
 489     if (qiov->niov > 1) {
 490         buf = orig_buf = qemu_blockalign(bs, qiov->size);
 491     } else {
 492         orig_buf = NULL;
 493         buf = (uint8_t *)qiov->iov->iov_base;
 494     }
 495
 496     qemu_co_mutex_lock(&s->lock);
 497
 498     while (nb_sectors != 0) {
 499         /* prepare next request */
 500         cluster_offset = get_cluster_offset(bs, sector_num << 9,
 501                                                  0, 0, 0, 0);
 502         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 503         n = s->cluster_sectors - index_in_cluster;
 504         if (n > nb_sectors) {
 505             n = nb_sectors;
 506         }
 507
 508         if (!cluster_offset) {
 509             if (bs->backing_hd) {
 510                 /* read from the base image */
 511                 hd_iov.iov_base = (void *)buf;
 512                 hd_iov.iov_len = n * 512;
 513                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 514                 qemu_co_mutex_unlock(&s->lock);
 515                 ret = bdrv_co_readv(bs->backing_hd, sector_num,
 516                                     n, &hd_qiov);
 517                 qemu_co_mutex_lock(&s->lock);
 518                 if (ret < 0) {
 519                     goto fail;
 520                 }
 521             } else {
 522                 /* Note: in this case, no need to wait */
 523                 memset(buf, 0, 512 * n);
 524             }
 525         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 526             /* add AIO support for compressed blocks ? */
 527             if (decompress_cluster(bs, cluster_offset) < 0) {
 528                 goto fail;
 529             }
 530             memcpy(buf,
 531                    s->cluster_cache + index_in_cluster * 512, 512 * n);
 532         } else {
 533             if ((cluster_offset & 511) != 0) {
 534                 goto fail;
 535             }
 536             hd_iov.iov_base = (void *)buf;
 537             hd_iov.iov_len = n * 512;
 538             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 539             qemu_co_mutex_unlock(&s->lock);
 540             ret = bdrv_co_readv(bs->file,
 541                                 (cluster_offset >> 9) + index_in_cluster,
 542                                 n, &hd_qiov);
 543             qemu_co_mutex_lock(&s->lock);
 544             if (ret < 0) {
 545                 break;
 546             }
 547             if (s->crypt_method) {
 548                 encrypt_sectors(s, sector_num, buf, buf,
 549                                 n, 0,
 550                                 &s->aes_decrypt_key);
 551             }
 552         }
 553         ret = 0;
 554
 555         nb_sectors -= n;
 556         sector_num += n;
 557         buf += n * 512;
 558     }
 559
 560 done:
 561     qemu_co_mutex_unlock(&s->lock);
 562
 563     if (qiov->niov > 1) {
 564         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
 565         qemu_vfree(orig_buf);
 566     }
 567
 568     return ret;
 569
 570 fail:
 571     ret = -EIO;
 572     goto done;
 573 }
 574
 575 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
 576                           int nb_sectors, QEMUIOVector *qiov)
 577 {
 578     BDRVQcowState *s = bs->opaque;
 579     int index_in_cluster;
 580     uint64_t cluster_offset;
 581     const uint8_t *src_buf;
 582     int ret = 0, n;
 583     uint8_t *cluster_data = NULL;
 584     struct iovec hd_iov;
 585     QEMUIOVector hd_qiov;
 586     uint8_t *buf;
 587     void *orig_buf;
 588
 589     s->cluster_cache_offset = -1; /* disable compressed cache */
 590
 591     if (qiov->niov > 1) {
 592         buf = orig_buf = qemu_blockalign(bs, qiov->size);
 593         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
 594     } else {
 595         orig_buf = NULL;
 596         buf = (uint8_t *)qiov->iov->iov_base;
 597     }
 598
 599     qemu_co_mutex_lock(&s->lock);
 600
 601     while (nb_sectors != 0) {
 602
 603         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 604         n = s->cluster_sectors - index_in_cluster;
 605         if (n > nb_sectors) {
 606             n = nb_sectors;
 607         }
 608         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
 609                                             index_in_cluster,
 610                                             index_in_cluster + n);
 611         if (!cluster_offset || (cluster_offset & 511) != 0) {
 612             ret = -EIO;
 613             break;
 614         }
 615         if (s->crypt_method) {
 616             if (!cluster_data) {
 617                 cluster_data = g_malloc0(s->cluster_size);
 618             }
 619             encrypt_sectors(s, sector_num, cluster_data, buf,
 620                             n, 1, &s->aes_encrypt_key);
 621             src_buf = cluster_data;
 622         } else {
 623             src_buf = buf;
 624         }
 625
 626         hd_iov.iov_base = (void *)src_buf;
 627         hd_iov.iov_len = n * 512;
 628         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 629         qemu_co_mutex_unlock(&s->lock);
 630         ret = bdrv_co_writev(bs->file,
 631                              (cluster_offset >> 9) + index_in_cluster,
 632                              n, &hd_qiov);
 633         qemu_co_mutex_lock(&s->lock);
 634         if (ret < 0) {
 635             break;
 636         }
 637         ret = 0;
 638
 639         nb_sectors -= n;
 640         sector_num += n;
 641         buf += n * 512;
 642     }
 643     qemu_co_mutex_unlock(&s->lock);
 644
 645     if (qiov->niov > 1) {
 646         qemu_vfree(orig_buf);
 647     }
 648     g_free(cluster_data);
 649
 650     return ret;
 651 }
 652
 653 static void qcow_close(BlockDriverState *bs)
 654 {
 655     BDRVQcowState *s = bs->opaque;
 656
 657     g_free(s->l1_table);
 658     g_free(s->l2_cache);
 659     g_free(s->cluster_cache);
 660     g_free(s->cluster_data);
 661
 662     migrate_del_blocker(s->migration_blocker);
 663     error_free(s->migration_blocker);
 664 }
 665
 666 static int qcow_create(const char *filename, QEMUOptionParameter *options,
 667                        Error **errp)
 668 {
 669     int header_size, backing_filename_len, l1_size, shift, i;
 670     QCowHeader header;
 671     uint8_t *tmp;
 672     int64_t total_size = 0;
 673     const char *backing_file = NULL;
 674     int flags = 0;
 675     Error *local_err = NULL;
 676     int ret;
 677     BlockDriverState *qcow_bs;
 678
 679     /* Read out options */
 680     while (options && options->name) {
 681         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 682             total_size = options->value.n / 512;
 683         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
 684             backing_file = options->value.s;
 685         } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
 686             flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
 687         }
 688         options++;
 689     }
 690
 691     ret = bdrv_create_file(filename, options, &local_err);
 692     if (ret < 0) {
 693         error_propagate(errp, local_err);
 694         return ret;
 695     }
 696
 697     qcow_bs = NULL;
 698     ret = bdrv_open(&qcow_bs, filename, NULL, NULL,
 699                     BDRV_O_RDWR | BDRV_O_PROTOCOL, NULL, &local_err);
 700     if (ret < 0) {
 701         error_propagate(errp, local_err);
 702         return ret;
 703     }
 704
 705     ret = bdrv_truncate(qcow_bs, 0);
 706     if (ret < 0) {
 707         goto exit;
 708     }
 709
 710     memset(&header, 0, sizeof(header));
 711     header.magic = cpu_to_be32(QCOW_MAGIC);
 712     header.version = cpu_to_be32(QCOW_VERSION);
 713     header.size = cpu_to_be64(total_size * 512);
 714     header_size = sizeof(header);
 715     backing_filename_len = 0;
 716     if (backing_file) {
 717         if (strcmp(backing_file, "fat:")) {
 718             header.backing_file_offset = cpu_to_be64(header_size);
 719             backing_filename_len = strlen(backing_file);
 720             header.backing_file_size = cpu_to_be32(backing_filename_len);
 721             header_size += backing_filename_len;
 722         } else {
 723             /* special backing file for vvfat */
 724             backing_file = NULL;
 725         }
 726         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
 727                                     unmodified sectors */
 728         header.l2_bits = 12; /* 32 KB L2 tables */
 729     } else {
 730         header.cluster_bits = 12; /* 4 KB clusters */
 731         header.l2_bits = 9; /* 4 KB L2 tables */
 732     }
 733     header_size = (header_size + 7) & ~7;
 734     shift = header.cluster_bits + header.l2_bits;
 735     l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
 736
 737     header.l1_table_offset = cpu_to_be64(header_size);
 738     if (flags & BLOCK_FLAG_ENCRYPT) {
 739         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
 740     } else {
 741         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
 742     }
 743
 744     /* write all the data */
 745     ret = bdrv_pwrite(qcow_bs, 0, &header, sizeof(header));
 746     if (ret != sizeof(header)) {
 747         goto exit;
 748     }
 749
 750     if (backing_file) {
 751         ret = bdrv_pwrite(qcow_bs, sizeof(header),
 752             backing_file, backing_filename_len);
 753         if (ret != backing_filename_len) {
 754             goto exit;
 755         }
 756     }
 757
 758     tmp = g_malloc0(BDRV_SECTOR_SIZE);
 759     for (i = 0; i < ((sizeof(uint64_t)*l1_size + BDRV_SECTOR_SIZE - 1)/
 760         BDRV_SECTOR_SIZE); i++) {
 761         ret = bdrv_pwrite(qcow_bs, header_size +
 762             BDRV_SECTOR_SIZE*i, tmp, BDRV_SECTOR_SIZE);
 763         if (ret != BDRV_SECTOR_SIZE) {
 764             g_free(tmp);
 765             goto exit;
 766         }
 767     }
 768
 769     g_free(tmp);
 770     ret = 0;
 771 exit:
 772     bdrv_unref(qcow_bs);
 773     return ret;
 774 }
 775
 776 static int qcow_make_empty(BlockDriverState *bs)
 777 {
 778     BDRVQcowState *s = bs->opaque;
 779     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
 780     int ret;
 781
 782     memset(s->l1_table, 0, l1_length);
 783     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
 784             l1_length) < 0)
 785         return -1;
 786     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
 787     if (ret < 0)
 788         return ret;
 789
 790     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 791     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
 792     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
 793
 794     return 0;
 795 }
 796
 797 /* XXX: put compressed sectors first, then all the cluster aligned
 798    tables to avoid losing bytes in alignment */
 799 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
 800                                  const uint8_t *buf, int nb_sectors)
 801 {
 802     BDRVQcowState *s = bs->opaque;
 803     z_stream strm;
 804     int ret, out_len;
 805     uint8_t *out_buf;
 806     uint64_t cluster_offset;
 807
 808     if (nb_sectors != s->cluster_sectors) {
 809         ret = -EINVAL;
 810
 811         /* Zero-pad last write if image size is not cluster aligned */
 812         if (sector_num + nb_sectors == bs->total_sectors &&
 813             nb_sectors < s->cluster_sectors) {
 814             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
 815             memset(pad_buf, 0, s->cluster_size);
 816             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
 817             ret = qcow_write_compressed(bs, sector_num,
 818                                         pad_buf, s->cluster_sectors);
 819             qemu_vfree(pad_buf);
 820         }
 821         return ret;
 822     }
 823
 824     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
 825
 826     /* best compression, small window, no zlib header */
 827     memset(&strm, 0, sizeof(strm));
 828     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
 829                        Z_DEFLATED, -12,
 830                        9, Z_DEFAULT_STRATEGY);
 831     if (ret != 0) {
 832         ret = -EINVAL;
 833         goto fail;
 834     }
 835
 836     strm.avail_in = s->cluster_size;
 837     strm.next_in = (uint8_t *)buf;
 838     strm.avail_out = s->cluster_size;
 839     strm.next_out = out_buf;
 840
 841     ret = deflate(&strm, Z_FINISH);
 842     if (ret != Z_STREAM_END && ret != Z_OK) {
 843         deflateEnd(&strm);
 844         ret = -EINVAL;
 845         goto fail;
 846     }
 847     out_len = strm.next_out - out_buf;
 848
 849     deflateEnd(&strm);
 850
 851     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
 852         /* could not compress: write normal cluster */
 853         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
 854         if (ret < 0) {
 855             goto fail;
 856         }
 857     } else {
 858         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
 859                                             out_len, 0, 0);
 860         if (cluster_offset == 0) {
 861             ret = -EIO;
 862             goto fail;
 863         }
 864
 865         cluster_offset &= s->cluster_offset_mask;
 866         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
 867         if (ret < 0) {
 868             goto fail;
 869         }
 870     }
 871
 872     ret = 0;
 873 fail:
 874     g_free(out_buf);
 875     return ret;
 876 }
 877
 878 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 879 {
 880     BDRVQcowState *s = bs->opaque;
 881     bdi->cluster_size = s->cluster_size;
 882     return 0;
 883 }
 884
 885
 886 static QEMUOptionParameter qcow_create_options[] = {
 887     {
 888         .name = BLOCK_OPT_SIZE,
 889         .type = OPT_SIZE,
 890         .help = "Virtual disk size"
 891     },
 892     {
 893         .name = BLOCK_OPT_BACKING_FILE,
 894         .type = OPT_STRING,
 895         .help = "File name of a base image"
 896     },
 897     {
 898         .name = BLOCK_OPT_ENCRYPT,
 899         .type = OPT_FLAG,
 900         .help = "Encrypt the image"
 901     },
 902     { NULL }
 903 };
 904
 905 static BlockDriver bdrv_qcow = {
 906     .format_name        = "qcow",
 907     .instance_size      = sizeof(BDRVQcowState),
 908     .bdrv_probe         = qcow_probe,
 909     .bdrv_open          = qcow_open,
 910     .bdrv_close         = qcow_close,
 911     .bdrv_reopen_prepare = qcow_reopen_prepare,
 912     .bdrv_create        = qcow_create,
 913     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
 914
 915     .bdrv_co_readv          = qcow_co_readv,
 916     .bdrv_co_writev         = qcow_co_writev,
 917     .bdrv_co_get_block_status   = qcow_co_get_block_status,
 918
 919     .bdrv_set_key           = qcow_set_key,
 920     .bdrv_make_empty        = qcow_make_empty,
 921     .bdrv_write_compressed  = qcow_write_compressed,
 922     .bdrv_get_info          = qcow_get_info,
 923
 924     .create_options = qcow_create_options,
 925 };
 926
 927 static void bdrv_qcow_init(void)
 928 {
 929     bdrv_register(&bdrv_qcow);
 930 }
 931
 932 block_init(bdrv_qcow_init);