block-qcow.c

   1 /*
   2  * Block driver for the QCOW format
   3  *
   4  * Copyright (c) 2004 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 "vl.h"
  25 #include "block_int.h"
  26 #include <zlib.h>
  27 #include "aes.h"
  28
  29 /**************************************************************/
  30 /* QEMU COW block driver with compression and encryption support */
  31
  32 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
  33 #define QCOW_VERSION 1
  34
  35 #define QCOW_CRYPT_NONE 0
  36 #define QCOW_CRYPT_AES  1
  37
  38 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
  39
  40 typedef struct QCowHeader {
  41     uint32_t magic;
  42     uint32_t version;
  43     uint64_t backing_file_offset;
  44     uint32_t backing_file_size;
  45     uint32_t mtime;
  46     uint64_t size; /* in bytes */
  47     uint8_t cluster_bits;
  48     uint8_t l2_bits;
  49     uint32_t crypt_method;
  50     uint64_t l1_table_offset;
  51 } QCowHeader;
  52
  53 #define L2_CACHE_SIZE 16
  54
  55 typedef struct BDRVQcowState {
  56     int fd;
  57     int cluster_bits;
  58     int cluster_size;
  59     int cluster_sectors;
  60     int l2_bits;
  61     int l2_size;
  62     int l1_size;
  63     uint64_t cluster_offset_mask;
  64     uint64_t l1_table_offset;
  65     uint64_t *l1_table;
  66     uint64_t *l2_cache;
  67     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
  68     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  69     uint8_t *cluster_cache;
  70     uint8_t *cluster_data;
  71     uint64_t cluster_cache_offset;
  72     uint32_t crypt_method; /* current crypt method, 0 if no key yet */
  73     uint32_t crypt_method_header;
  74     AES_KEY aes_encrypt_key;
  75     AES_KEY aes_decrypt_key;
  76 } BDRVQcowState;
  77
  78 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
  79
  80 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
  81 {
  82     const QCowHeader *cow_header = (const void *)buf;
  83
  84     if (be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
  85         be32_to_cpu(cow_header->version) == QCOW_VERSION)
  86         return 100;
  87     else
  88         return 0;
  89 }
  90
  91 static int qcow_open(BlockDriverState *bs, const char *filename)
  92 {
  93     BDRVQcowState *s = bs->opaque;
  94     int fd, len, i, shift;
  95     QCowHeader header;
  96
  97     fd = open(filename, O_RDWR | O_BINARY | O_LARGEFILE);
  98     if (fd < 0) {
  99         fd = open(filename, O_RDONLY | O_BINARY | O_LARGEFILE);
 100         if (fd < 0)
 101             return -1;
 102     }
 103     s->fd = fd;
 104     if (read(fd, &header, sizeof(header)) != sizeof(header))
 105         goto fail;
 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 || header.version != QCOW_VERSION)
 116         goto fail;
 117     if (header.size <= 1 || header.cluster_bits < 9)
 118         goto fail;
 119     if (header.crypt_method > QCOW_CRYPT_AES)
 120         goto fail;
 121     s->crypt_method_header = header.crypt_method;
 122     if (s->crypt_method_header)
 123         bs->encrypted = 1;
 124     s->cluster_bits = header.cluster_bits;
 125     s->cluster_size = 1 << s->cluster_bits;
 126     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 127     s->l2_bits = header.l2_bits;
 128     s->l2_size = 1 << s->l2_bits;
 129     bs->total_sectors = header.size / 512;
 130     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
 131
 132     /* read the level 1 table */
 133     shift = s->cluster_bits + s->l2_bits;
 134     s->l1_size = (header.size + (1LL << shift) - 1) >> shift;
 135
 136     s->l1_table_offset = header.l1_table_offset;
 137     s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
 138     if (!s->l1_table)
 139         goto fail;
 140     lseek(fd, s->l1_table_offset, SEEK_SET);
 141     if (read(fd, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
 142         s->l1_size * sizeof(uint64_t))
 143         goto fail;
 144     for(i = 0;i < s->l1_size; i++) {
 145         be64_to_cpus(&s->l1_table[i]);
 146     }
 147     /* alloc L2 cache */
 148     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 149     if (!s->l2_cache)
 150         goto fail;
 151     s->cluster_cache = qemu_malloc(s->cluster_size);
 152     if (!s->cluster_cache)
 153         goto fail;
 154     s->cluster_data = qemu_malloc(s->cluster_size);
 155     if (!s->cluster_data)
 156         goto fail;
 157     s->cluster_cache_offset = -1;
 158
 159     /* read the backing file name */
 160     if (header.backing_file_offset != 0) {
 161         len = header.backing_file_size;
 162         if (len > 1023)
 163             len = 1023;
 164         lseek(fd, header.backing_file_offset, SEEK_SET);
 165         if (read(fd, bs->backing_file, len) != len)
 166             goto fail;
 167         bs->backing_file[len] = '\0';
 168     }
 169     return 0;
 170
 171  fail:
 172     qemu_free(s->l1_table);
 173     qemu_free(s->l2_cache);
 174     qemu_free(s->cluster_cache);
 175     qemu_free(s->cluster_data);
 176     close(fd);
 177     return -1;
 178 }
 179
 180 static int qcow_set_key(BlockDriverState *bs, const char *key)
 181 {
 182     BDRVQcowState *s = bs->opaque;
 183     uint8_t keybuf[16];
 184     int len, i;
 185
 186     memset(keybuf, 0, 16);
 187     len = strlen(key);
 188     if (len > 16)
 189         len = 16;
 190     /* XXX: we could compress the chars to 7 bits to increase
 191        entropy */
 192     for(i = 0;i < len;i++) {
 193         keybuf[i] = key[i];
 194     }
 195     s->crypt_method = s->crypt_method_header;
 196
 197     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
 198         return -1;
 199     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
 200         return -1;
 201 #if 0
 202     /* test */
 203     {
 204         uint8_t in[16];
 205         uint8_t out[16];
 206         uint8_t tmp[16];
 207         for(i=0;i<16;i++)
 208             in[i] = i;
 209         AES_encrypt(in, tmp, &s->aes_encrypt_key);
 210         AES_decrypt(tmp, out, &s->aes_decrypt_key);
 211         for(i = 0; i < 16; i++)
 212             printf(" %02x", tmp[i]);
 213         printf("\n");
 214         for(i = 0; i < 16; i++)
 215             printf(" %02x", out[i]);
 216         printf("\n");
 217     }
 218 #endif
 219     return 0;
 220 }
 221
 222 /* The crypt function is compatible with the linux cryptoloop
 223    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
 224    supported */
 225 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
 226                             uint8_t *out_buf, const uint8_t *in_buf,
 227                             int nb_sectors, int enc,
 228                             const AES_KEY *key)
 229 {
 230     union {
 231         uint64_t ll[2];
 232         uint8_t b[16];
 233     } ivec;
 234     int i;
 235
 236     for(i = 0; i < nb_sectors; i++) {
 237         ivec.ll[0] = cpu_to_le64(sector_num);
 238         ivec.ll[1] = 0;
 239         AES_cbc_encrypt(in_buf, out_buf, 512, key,
 240                         ivec.b, enc);
 241         sector_num++;
 242         in_buf += 512;
 243         out_buf += 512;
 244     }
 245 }
 246
 247 /* 'allocate' is:
 248  *
 249  * 0 to not allocate.
 250  *
 251  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
 252  * 'n_end')
 253  *
 254  * 2 to allocate a compressed cluster of size
 255  * 'compressed_size'. 'compressed_size' must be > 0 and <
 256  * cluster_size
 257  *
 258  * return 0 if not allocated.
 259  */
 260 static uint64_t get_cluster_offset(BlockDriverState *bs,
 261                                    uint64_t offset, int allocate,
 262                                    int compressed_size,
 263                                    int n_start, int n_end)
 264 {
 265     BDRVQcowState *s = bs->opaque;
 266     int min_index, i, j, l1_index, l2_index;
 267     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
 268     uint32_t min_count;
 269     int new_l2_table;
 270
 271     l1_index = offset >> (s->l2_bits + s->cluster_bits);
 272     l2_offset = s->l1_table[l1_index];
 273     new_l2_table = 0;
 274     if (!l2_offset) {
 275         if (!allocate)
 276             return 0;
 277         /* allocate a new l2 entry */
 278         l2_offset = lseek(s->fd, 0, SEEK_END);
 279         /* round to cluster size */
 280         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
 281         /* update the L1 entry */
 282         s->l1_table[l1_index] = l2_offset;
 283         tmp = cpu_to_be64(l2_offset);
 284         lseek(s->fd, s->l1_table_offset + l1_index * sizeof(tmp), SEEK_SET);
 285         if (write(s->fd, &tmp, sizeof(tmp)) != sizeof(tmp))
 286             return 0;
 287         new_l2_table = 1;
 288     }
 289     for(i = 0; i < L2_CACHE_SIZE; i++) {
 290         if (l2_offset == s->l2_cache_offsets[i]) {
 291             /* increment the hit count */
 292             if (++s->l2_cache_counts[i] == 0xffffffff) {
 293                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 294                     s->l2_cache_counts[j] >>= 1;
 295                 }
 296             }
 297             l2_table = s->l2_cache + (i << s->l2_bits);
 298             goto found;
 299         }
 300     }
 301     /* not found: load a new entry in the least used one */
 302     min_index = 0;
 303     min_count = 0xffffffff;
 304     for(i = 0; i < L2_CACHE_SIZE; i++) {
 305         if (s->l2_cache_counts[i] < min_count) {
 306             min_count = s->l2_cache_counts[i];
 307             min_index = i;
 308         }
 309     }
 310     l2_table = s->l2_cache + (min_index << s->l2_bits);
 311     lseek(s->fd, l2_offset, SEEK_SET);
 312     if (new_l2_table) {
 313         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
 314         if (write(s->fd, l2_table, s->l2_size * sizeof(uint64_t)) !=
 315             s->l2_size * sizeof(uint64_t))
 316             return 0;
 317     } else {
 318         if (read(s->fd, l2_table, s->l2_size * sizeof(uint64_t)) !=
 319             s->l2_size * sizeof(uint64_t))
 320             return 0;
 321     }
 322     s->l2_cache_offsets[min_index] = l2_offset;
 323     s->l2_cache_counts[min_index] = 1;
 324  found:
 325     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
 326     cluster_offset = be64_to_cpu(l2_table[l2_index]);
 327     if (!cluster_offset ||
 328         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
 329         if (!allocate)
 330             return 0;
 331         /* allocate a new cluster */
 332         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
 333             (n_end - n_start) < s->cluster_sectors) {
 334             /* if the cluster is already compressed, we must
 335                decompress it in the case it is not completely
 336                overwritten */
 337             if (decompress_cluster(s, cluster_offset) < 0)
 338                 return 0;
 339             cluster_offset = lseek(s->fd, 0, SEEK_END);
 340             cluster_offset = (cluster_offset + s->cluster_size - 1) &
 341                 ~(s->cluster_size - 1);
 342             /* write the cluster content */
 343             lseek(s->fd, cluster_offset, SEEK_SET);
 344             if (write(s->fd, s->cluster_cache, s->cluster_size) !=
 345                 s->cluster_size)
 346                 return -1;
 347         } else {
 348             cluster_offset = lseek(s->fd, 0, SEEK_END);
 349             if (allocate == 1) {
 350                 /* round to cluster size */
 351                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
 352                     ~(s->cluster_size - 1);
 353                 ftruncate(s->fd, cluster_offset + s->cluster_size);
 354                 /* if encrypted, we must initialize the cluster
 355                    content which won't be written */
 356                 if (s->crypt_method &&
 357                     (n_end - n_start) < s->cluster_sectors) {
 358                     uint64_t start_sect;
 359                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
 360                     memset(s->cluster_data + 512, 0xaa, 512);
 361                     for(i = 0; i < s->cluster_sectors; i++) {
 362                         if (i < n_start || i >= n_end) {
 363                             encrypt_sectors(s, start_sect + i,
 364                                             s->cluster_data,
 365                                             s->cluster_data + 512, 1, 1,
 366                                             &s->aes_encrypt_key);
 367                             lseek(s->fd, cluster_offset + i * 512, SEEK_SET);
 368                             if (write(s->fd, s->cluster_data, 512) != 512)
 369                                 return -1;
 370                         }
 371                     }
 372                 }
 373             } else {
 374                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
 375                     (uint64_t)compressed_size << (63 - s->cluster_bits);
 376             }
 377         }
 378         /* update L2 table */
 379         tmp = cpu_to_be64(cluster_offset);
 380         l2_table[l2_index] = tmp;
 381         lseek(s->fd, l2_offset + l2_index * sizeof(tmp), SEEK_SET);
 382         if (write(s->fd, &tmp, sizeof(tmp)) != sizeof(tmp))
 383             return 0;
 384     }
 385     return cluster_offset;
 386 }
 387
 388 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
 389                              int nb_sectors, int *pnum)
 390 {
 391     BDRVQcowState *s = bs->opaque;
 392     int index_in_cluster, n;
 393     uint64_t cluster_offset;
 394
 395     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
 396     index_in_cluster = sector_num & (s->cluster_sectors - 1);
 397     n = s->cluster_sectors - index_in_cluster;
 398     if (n > nb_sectors)
 399         n = nb_sectors;
 400     *pnum = n;
 401     return (cluster_offset != 0);
 402 }
 403
 404 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
 405                              const uint8_t *buf, int buf_size)
 406 {
 407     z_stream strm1, *strm = &strm1;
 408     int ret, out_len;
 409
 410     memset(strm, 0, sizeof(*strm));
 411
 412     strm->next_in = (uint8_t *)buf;
 413     strm->avail_in = buf_size;
 414     strm->next_out = out_buf;
 415     strm->avail_out = out_buf_size;
 416
 417     ret = inflateInit2(strm, -12);
 418     if (ret != Z_OK)
 419         return -1;
 420     ret = inflate(strm, Z_FINISH);
 421     out_len = strm->next_out - out_buf;
 422     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
 423         out_len != out_buf_size) {
 424         inflateEnd(strm);
 425         return -1;
 426     }
 427     inflateEnd(strm);
 428     return 0;
 429 }
 430
 431 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
 432 {
 433     int ret, csize;
 434     uint64_t coffset;
 435
 436     coffset = cluster_offset & s->cluster_offset_mask;
 437     if (s->cluster_cache_offset != coffset) {
 438         csize = cluster_offset >> (63 - s->cluster_bits);
 439         csize &= (s->cluster_size - 1);
 440         lseek(s->fd, coffset, SEEK_SET);
 441         ret = read(s->fd, s->cluster_data, csize);
 442         if (ret != csize)
 443             return -1;
 444         if (decompress_buffer(s->cluster_cache, s->cluster_size,
 445                               s->cluster_data, csize) < 0) {
 446             return -1;
 447         }
 448         s->cluster_cache_offset = coffset;
 449     }
 450     return 0;
 451 }
 452
 453 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
 454                      uint8_t *buf, int nb_sectors)
 455 {
 456     BDRVQcowState *s = bs->opaque;
 457     int ret, index_in_cluster, n;
 458     uint64_t cluster_offset;
 459
 460     while (nb_sectors > 0) {
 461         cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
 462         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 463         n = s->cluster_sectors - index_in_cluster;
 464         if (n > nb_sectors)
 465             n = nb_sectors;
 466         if (!cluster_offset) {
 467             memset(buf, 0, 512 * n);
 468         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 469             if (decompress_cluster(s, cluster_offset) < 0)
 470                 return -1;
 471             memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
 472         } else {
 473             lseek(s->fd, cluster_offset + index_in_cluster * 512, SEEK_SET);
 474             ret = read(s->fd, buf, n * 512);
 475             if (ret != n * 512)
 476                 return -1;
 477             if (s->crypt_method) {
 478                 encrypt_sectors(s, sector_num, buf, buf, n, 0,
 479                                 &s->aes_decrypt_key);
 480             }
 481         }
 482         nb_sectors -= n;
 483         sector_num += n;
 484         buf += n * 512;
 485     }
 486     return 0;
 487 }
 488
 489 static int qcow_write(BlockDriverState *bs, int64_t sector_num,
 490                      const uint8_t *buf, int nb_sectors)
 491 {
 492     BDRVQcowState *s = bs->opaque;
 493     int ret, index_in_cluster, n;
 494     uint64_t cluster_offset;
 495
 496     while (nb_sectors > 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         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
 502                                             index_in_cluster,
 503                                             index_in_cluster + n);
 504         if (!cluster_offset)
 505             return -1;
 506         lseek(s->fd, cluster_offset + index_in_cluster * 512, SEEK_SET);
 507         if (s->crypt_method) {
 508             encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
 509                             &s->aes_encrypt_key);
 510             ret = write(s->fd, s->cluster_data, n * 512);
 511         } else {
 512             ret = write(s->fd, buf, n * 512);
 513         }
 514         if (ret != n * 512)
 515             return -1;
 516         nb_sectors -= n;
 517         sector_num += n;
 518         buf += n * 512;
 519     }
 520     s->cluster_cache_offset = -1; /* disable compressed cache */
 521     return 0;
 522 }
 523
 524 static void qcow_close(BlockDriverState *bs)
 525 {
 526     BDRVQcowState *s = bs->opaque;
 527     qemu_free(s->l1_table);
 528     qemu_free(s->l2_cache);
 529     qemu_free(s->cluster_cache);
 530     qemu_free(s->cluster_data);
 531     close(s->fd);
 532 }
 533
 534 static int qcow_create(const char *filename, int64_t total_size,
 535                       const char *backing_file, int flags)
 536 {
 537     int fd, header_size, backing_filename_len, l1_size, i, shift;
 538     QCowHeader header;
 539     char backing_filename[1024];
 540     uint64_t tmp;
 541     struct stat st;
 542
 543     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
 544               0644);
 545     if (fd < 0)
 546         return -1;
 547     memset(&header, 0, sizeof(header));
 548     header.magic = cpu_to_be32(QCOW_MAGIC);
 549     header.version = cpu_to_be32(QCOW_VERSION);
 550     header.size = cpu_to_be64(total_size * 512);
 551     header_size = sizeof(header);
 552     backing_filename_len = 0;
 553     if (backing_file) {
 554         realpath(backing_file, backing_filename);
 555         if (stat(backing_filename, &st) != 0) {
 556             return -1;
 557         }
 558         header.mtime = cpu_to_be32(st.st_mtime);
 559         header.backing_file_offset = cpu_to_be64(header_size);
 560         backing_filename_len = strlen(backing_filename);
 561         header.backing_file_size = cpu_to_be32(backing_filename_len);
 562         header_size += backing_filename_len;
 563         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
 564                                     unmodifyed sectors */
 565         header.l2_bits = 12; /* 32 KB L2 tables */
 566     } else {
 567         header.cluster_bits = 12; /* 4 KB clusters */
 568         header.l2_bits = 9; /* 4 KB L2 tables */
 569     }
 570     header_size = (header_size + 7) & ~7;
 571     shift = header.cluster_bits + header.l2_bits;
 572     l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
 573
 574     header.l1_table_offset = cpu_to_be64(header_size);
 575     if (flags) {
 576         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
 577     } else {
 578         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
 579     }
 580
 581     /* write all the data */
 582     write(fd, &header, sizeof(header));
 583     if (backing_file) {
 584         write(fd, backing_filename, backing_filename_len);
 585     }
 586     lseek(fd, header_size, SEEK_SET);
 587     tmp = 0;
 588     for(i = 0;i < l1_size; i++) {
 589         write(fd, &tmp, sizeof(tmp));
 590     }
 591     close(fd);
 592     return 0;
 593 }
 594
 595 int qcow_get_cluster_size(BlockDriverState *bs)
 596 {
 597     BDRVQcowState *s = bs->opaque;
 598     if (bs->drv != &bdrv_qcow)
 599         return -1;
 600     return s->cluster_size;
 601 }
 602
 603 /* XXX: put compressed sectors first, then all the cluster aligned
 604    tables to avoid losing bytes in alignment */
 605 int qcow_compress_cluster(BlockDriverState *bs, int64_t sector_num,
 606                           const uint8_t *buf)
 607 {
 608     BDRVQcowState *s = bs->opaque;
 609     z_stream strm;
 610     int ret, out_len;
 611     uint8_t *out_buf;
 612     uint64_t cluster_offset;
 613
 614     if (bs->drv != &bdrv_qcow)
 615         return -1;
 616
 617     out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
 618     if (!out_buf)
 619         return -1;
 620
 621     /* best compression, small window, no zlib header */
 622     memset(&strm, 0, sizeof(strm));
 623     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
 624                        Z_DEFLATED, -12,
 625                        9, Z_DEFAULT_STRATEGY);
 626     if (ret != 0) {
 627         qemu_free(out_buf);
 628         return -1;
 629     }
 630
 631     strm.avail_in = s->cluster_size;
 632     strm.next_in = (uint8_t *)buf;
 633     strm.avail_out = s->cluster_size;
 634     strm.next_out = out_buf;
 635
 636     ret = deflate(&strm, Z_FINISH);
 637     if (ret != Z_STREAM_END && ret != Z_OK) {
 638         qemu_free(out_buf);
 639         deflateEnd(&strm);
 640         return -1;
 641     }
 642     out_len = strm.next_out - out_buf;
 643
 644     deflateEnd(&strm);
 645
 646     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
 647         /* could not compress: write normal cluster */
 648         qcow_write(bs, sector_num, buf, s->cluster_sectors);
 649     } else {
 650         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
 651                                             out_len, 0, 0);
 652         cluster_offset &= s->cluster_offset_mask;
 653         lseek(s->fd, cluster_offset, SEEK_SET);
 654         if (write(s->fd, out_buf, out_len) != out_len) {
 655             qemu_free(out_buf);
 656             return -1;
 657         }
 658     }
 659
 660     qemu_free(out_buf);
 661     return 0;
 662 }
 663
 664 BlockDriver bdrv_qcow = {
 665     "qcow",
 666     sizeof(BDRVQcowState),
 667     qcow_probe,
 668     qcow_open,
 669     qcow_read,
 670     qcow_write,
 671     qcow_close,
 672     qcow_create,
 673     qcow_is_allocated,
 674     qcow_set_key,
 675 };
 676
 677