block/qcow.c

   1 /*
   2  * Block driver for the QCOW format
   3  *
   4  * Copyright (c) 2004-2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu-common.h"
  25 #include "block_int.h"
  26 #include "module.h"
  27 #include <zlib.h>
  28 #include "aes.h"
  29 #include "migration.h"
  30
  31 /**************************************************************/
  32 /* QEMU COW block driver with compression and encryption support */
  33
  34 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
  35 #define QCOW_VERSION 1
  36
  37 #define QCOW_CRYPT_NONE 0
  38 #define QCOW_CRYPT_AES  1
  39
  40 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
  41
  42 typedef struct QCowHeader {
  43     uint32_t magic;
  44     uint32_t version;
  45     uint64_t backing_file_offset;
  46     uint32_t backing_file_size;
  47     uint32_t mtime;
  48     uint64_t size; /* in bytes */
  49     uint8_t cluster_bits;
  50     uint8_t l2_bits;
  51     uint32_t crypt_method;
  52     uint64_t l1_table_offset;
  53 } QCowHeader;
  54
  55 #define L2_CACHE_SIZE 16
  56
  57 typedef struct BDRVQcowState {
  58     int cluster_bits;
  59     int cluster_size;
  60     int cluster_sectors;
  61     int l2_bits;
  62     int l2_size;
  63     int l1_size;
  64     uint64_t cluster_offset_mask;
  65     uint64_t l1_table_offset;
  66     uint64_t *l1_table;
  67     uint64_t *l2_cache;
  68     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
  69     uint32_t l2_cache_counts[L2_CACHE_SIZE];
  70     uint8_t *cluster_cache;
  71     uint8_t *cluster_data;
  72     uint64_t cluster_cache_offset;
  73     uint32_t crypt_method; /* current crypt method, 0 if no key yet */
  74     uint32_t crypt_method_header;
  75     AES_KEY aes_encrypt_key;
  76     AES_KEY aes_decrypt_key;
  77     CoMutex lock;
  78     Error *migration_blocker;
  79 } BDRVQcowState;
  80
  81 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
  82
  83 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
  84 {
  85     const QCowHeader *cow_header = (const void *)buf;
  86
  87     if (buf_size >= sizeof(QCowHeader) &&
  88         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
  89         be32_to_cpu(cow_header->version) == QCOW_VERSION)
  90         return 100;
  91     else
  92         return 0;
  93 }
  94
  95 static int qcow_open(BlockDriverState *bs, int flags)
  96 {
  97     BDRVQcowState *s = bs->opaque;
  98     int len, i, shift;
  99     QCowHeader header;
 100
 101     if (bdrv_pread(bs->file, 0, &header, sizeof(header)) != sizeof(header))
 102         goto fail;
 103     be32_to_cpus(&header.magic);
 104     be32_to_cpus(&header.version);
 105     be64_to_cpus(&header.backing_file_offset);
 106     be32_to_cpus(&header.backing_file_size);
 107     be32_to_cpus(&header.mtime);
 108     be64_to_cpus(&header.size);
 109     be32_to_cpus(&header.crypt_method);
 110     be64_to_cpus(&header.l1_table_offset);
 111
 112     if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
 113         goto fail;
 114     if (header.size <= 1 || header.cluster_bits < 9)
 115         goto fail;
 116     if (header.crypt_method > QCOW_CRYPT_AES)
 117         goto fail;
 118     s->crypt_method_header = header.crypt_method;
 119     if (s->crypt_method_header)
 120         bs->encrypted = 1;
 121     s->cluster_bits = header.cluster_bits;
 122     s->cluster_size = 1 << s->cluster_bits;
 123     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 124     s->l2_bits = header.l2_bits;
 125     s->l2_size = 1 << s->l2_bits;
 126     bs->total_sectors = header.size / 512;
 127     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
 128
 129     /* read the level 1 table */
 130     shift = s->cluster_bits + s->l2_bits;
 131     s->l1_size = (header.size + (1LL << shift) - 1) >> shift;
 132
 133     s->l1_table_offset = header.l1_table_offset;
 134     s->l1_table = g_malloc(s->l1_size * sizeof(uint64_t));
 135     if (!s->l1_table)
 136         goto fail;
 137     if (bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
 138         s->l1_size * sizeof(uint64_t))
 139         goto fail;
 140     for(i = 0;i < s->l1_size; i++) {
 141         be64_to_cpus(&s->l1_table[i]);
 142     }
 143     /* alloc L2 cache */
 144     s->l2_cache = g_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 145     if (!s->l2_cache)
 146         goto fail;
 147     s->cluster_cache = g_malloc(s->cluster_size);
 148     if (!s->cluster_cache)
 149         goto fail;
 150     s->cluster_data = g_malloc(s->cluster_size);
 151     if (!s->cluster_data)
 152         goto fail;
 153     s->cluster_cache_offset = -1;
 154
 155     /* read the backing file name */
 156     if (header.backing_file_offset != 0) {
 157         len = header.backing_file_size;
 158         if (len > 1023)
 159             len = 1023;
 160         if (bdrv_pread(bs->file, header.backing_file_offset, bs->backing_file, len) != len)
 161             goto fail;
 162         bs->backing_file[len] = '\0';
 163     }
 164
 165     /* Disable migration when qcow images are used */
 166     error_set(&s->migration_blocker,
 167               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
 168               "qcow", bs->device_name, "live migration");
 169     migrate_add_blocker(s->migration_blocker);
 170
 171     qemu_co_mutex_init(&s->lock);
 172     return 0;
 173
 174  fail:
 175     g_free(s->l1_table);
 176     g_free(s->l2_cache);
 177     g_free(s->cluster_cache);
 178     g_free(s->cluster_data);
 179     return -1;
 180 }
 181
 182 static int qcow_set_key(BlockDriverState *bs, const char *key)
 183 {
 184     BDRVQcowState *s = bs->opaque;
 185     uint8_t keybuf[16];
 186     int len, i;
 187
 188     memset(keybuf, 0, 16);
 189     len = strlen(key);
 190     if (len > 16)
 191         len = 16;
 192     /* XXX: we could compress the chars to 7 bits to increase
 193        entropy */
 194     for(i = 0;i < len;i++) {
 195         keybuf[i] = key[i];
 196     }
 197     s->crypt_method = s->crypt_method_header;
 198
 199     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
 200         return -1;
 201     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
 202         return -1;
 203     return 0;
 204 }
 205
 206 /* The crypt function is compatible with the linux cryptoloop
 207    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
 208    supported */
 209 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
 210                             uint8_t *out_buf, const uint8_t *in_buf,
 211                             int nb_sectors, int enc,
 212                             const AES_KEY *key)
 213 {
 214     union {
 215         uint64_t ll[2];
 216         uint8_t b[16];
 217     } ivec;
 218     int i;
 219
 220     for(i = 0; i < nb_sectors; i++) {
 221         ivec.ll[0] = cpu_to_le64(sector_num);
 222         ivec.ll[1] = 0;
 223         AES_cbc_encrypt(in_buf, out_buf, 512, key,
 224                         ivec.b, enc);
 225         sector_num++;
 226         in_buf += 512;
 227         out_buf += 512;
 228     }
 229 }
 230
 231 /* 'allocate' is:
 232  *
 233  * 0 to not allocate.
 234  *
 235  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
 236  * 'n_end')
 237  *
 238  * 2 to allocate a compressed cluster of size
 239  * 'compressed_size'. 'compressed_size' must be > 0 and <
 240  * cluster_size
 241  *
 242  * return 0 if not allocated.
 243  */
 244 static uint64_t get_cluster_offset(BlockDriverState *bs,
 245                                    uint64_t offset, int allocate,
 246                                    int compressed_size,
 247                                    int n_start, int n_end)
 248 {
 249     BDRVQcowState *s = bs->opaque;
 250     int min_index, i, j, l1_index, l2_index;
 251     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
 252     uint32_t min_count;
 253     int new_l2_table;
 254
 255     l1_index = offset >> (s->l2_bits + s->cluster_bits);
 256     l2_offset = s->l1_table[l1_index];
 257     new_l2_table = 0;
 258     if (!l2_offset) {
 259         if (!allocate)
 260             return 0;
 261         /* allocate a new l2 entry */
 262         l2_offset = bdrv_getlength(bs->file);
 263         /* round to cluster size */
 264         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
 265         /* update the L1 entry */
 266         s->l1_table[l1_index] = l2_offset;
 267         tmp = cpu_to_be64(l2_offset);
 268         if (bdrv_pwrite_sync(bs->file,
 269                 s->l1_table_offset + l1_index * sizeof(tmp),
 270                 &tmp, sizeof(tmp)) < 0)
 271             return 0;
 272         new_l2_table = 1;
 273     }
 274     for(i = 0; i < L2_CACHE_SIZE; i++) {
 275         if (l2_offset == s->l2_cache_offsets[i]) {
 276             /* increment the hit count */
 277             if (++s->l2_cache_counts[i] == 0xffffffff) {
 278                 for(j = 0; j < L2_CACHE_SIZE; j++) {
 279                     s->l2_cache_counts[j] >>= 1;
 280                 }
 281             }
 282             l2_table = s->l2_cache + (i << s->l2_bits);
 283             goto found;
 284         }
 285     }
 286     /* not found: load a new entry in the least used one */
 287     min_index = 0;
 288     min_count = 0xffffffff;
 289     for(i = 0; i < L2_CACHE_SIZE; i++) {
 290         if (s->l2_cache_counts[i] < min_count) {
 291             min_count = s->l2_cache_counts[i];
 292             min_index = i;
 293         }
 294     }
 295     l2_table = s->l2_cache + (min_index << s->l2_bits);
 296     if (new_l2_table) {
 297         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
 298         if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
 299                 s->l2_size * sizeof(uint64_t)) < 0)
 300             return 0;
 301     } else {
 302         if (bdrv_pread(bs->file, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
 303             s->l2_size * sizeof(uint64_t))
 304             return 0;
 305     }
 306     s->l2_cache_offsets[min_index] = l2_offset;
 307     s->l2_cache_counts[min_index] = 1;
 308  found:
 309     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
 310     cluster_offset = be64_to_cpu(l2_table[l2_index]);
 311     if (!cluster_offset ||
 312         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
 313         if (!allocate)
 314             return 0;
 315         /* allocate a new cluster */
 316         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
 317             (n_end - n_start) < s->cluster_sectors) {
 318             /* if the cluster is already compressed, we must
 319                decompress it in the case it is not completely
 320                overwritten */
 321             if (decompress_cluster(bs, cluster_offset) < 0)
 322                 return 0;
 323             cluster_offset = bdrv_getlength(bs->file);
 324             cluster_offset = (cluster_offset + s->cluster_size - 1) &
 325                 ~(s->cluster_size - 1);
 326             /* write the cluster content */
 327             if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache, s->cluster_size) !=
 328                 s->cluster_size)
 329                 return -1;
 330         } else {
 331             cluster_offset = bdrv_getlength(bs->file);
 332             if (allocate == 1) {
 333                 /* round to cluster size */
 334                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
 335                     ~(s->cluster_size - 1);
 336                 bdrv_truncate(bs->file, cluster_offset + s->cluster_size);
 337                 /* if encrypted, we must initialize the cluster
 338                    content which won't be written */
 339                 if (s->crypt_method &&
 340                     (n_end - n_start) < s->cluster_sectors) {
 341                     uint64_t start_sect;
 342                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
 343                     memset(s->cluster_data + 512, 0x00, 512);
 344                     for(i = 0; i < s->cluster_sectors; i++) {
 345                         if (i < n_start || i >= n_end) {
 346                             encrypt_sectors(s, start_sect + i,
 347                                             s->cluster_data,
 348                                             s->cluster_data + 512, 1, 1,
 349                                             &s->aes_encrypt_key);
 350                             if (bdrv_pwrite(bs->file, cluster_offset + i * 512,
 351                                             s->cluster_data, 512) != 512)
 352                                 return -1;
 353                         }
 354                     }
 355                 }
 356             } else if (allocate == 2) {
 357                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
 358                     (uint64_t)compressed_size << (63 - s->cluster_bits);
 359             }
 360         }
 361         /* update L2 table */
 362         tmp = cpu_to_be64(cluster_offset);
 363         l2_table[l2_index] = tmp;
 364         if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
 365                 &tmp, sizeof(tmp)) < 0)
 366             return 0;
 367     }
 368     return cluster_offset;
 369 }
 370
 371 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
 372                              int nb_sectors, int *pnum)
 373 {
 374     BDRVQcowState *s = bs->opaque;
 375     int index_in_cluster, n;
 376     uint64_t cluster_offset;
 377
 378     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
 379     index_in_cluster = sector_num & (s->cluster_sectors - 1);
 380     n = s->cluster_sectors - index_in_cluster;
 381     if (n > nb_sectors)
 382         n = nb_sectors;
 383     *pnum = n;
 384     return (cluster_offset != 0);
 385 }
 386
 387 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
 388                              const uint8_t *buf, int buf_size)
 389 {
 390     z_stream strm1, *strm = &strm1;
 391     int ret, out_len;
 392
 393     memset(strm, 0, sizeof(*strm));
 394
 395     strm->next_in = (uint8_t *)buf;
 396     strm->avail_in = buf_size;
 397     strm->next_out = out_buf;
 398     strm->avail_out = out_buf_size;
 399
 400     ret = inflateInit2(strm, -12);
 401     if (ret != Z_OK)
 402         return -1;
 403     ret = inflate(strm, Z_FINISH);
 404     out_len = strm->next_out - out_buf;
 405     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
 406         out_len != out_buf_size) {
 407         inflateEnd(strm);
 408         return -1;
 409     }
 410     inflateEnd(strm);
 411     return 0;
 412 }
 413
 414 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
 415 {
 416     BDRVQcowState *s = bs->opaque;
 417     int ret, csize;
 418     uint64_t coffset;
 419
 420     coffset = cluster_offset & s->cluster_offset_mask;
 421     if (s->cluster_cache_offset != coffset) {
 422         csize = cluster_offset >> (63 - s->cluster_bits);
 423         csize &= (s->cluster_size - 1);
 424         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
 425         if (ret != csize)
 426             return -1;
 427         if (decompress_buffer(s->cluster_cache, s->cluster_size,
 428                               s->cluster_data, csize) < 0) {
 429             return -1;
 430         }
 431         s->cluster_cache_offset = coffset;
 432     }
 433     return 0;
 434 }
 435
 436 static int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
 437                          int nb_sectors, QEMUIOVector *qiov)
 438 {
 439     BDRVQcowState *s = bs->opaque;
 440     int index_in_cluster;
 441     int ret = 0, n;
 442     uint64_t cluster_offset;
 443     struct iovec hd_iov;
 444     QEMUIOVector hd_qiov;
 445     uint8_t *buf;
 446     void *orig_buf;
 447
 448     if (qiov->niov > 1) {
 449         buf = orig_buf = qemu_blockalign(bs, qiov->size);
 450     } else {
 451         orig_buf = NULL;
 452         buf = (uint8_t *)qiov->iov->iov_base;
 453     }
 454
 455     qemu_co_mutex_lock(&s->lock);
 456
 457     while (nb_sectors != 0) {
 458         /* prepare next request */
 459         cluster_offset = get_cluster_offset(bs, sector_num << 9,
 460                                                  0, 0, 0, 0);
 461         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 462         n = s->cluster_sectors - index_in_cluster;
 463         if (n > nb_sectors) {
 464             n = nb_sectors;
 465         }
 466
 467         if (!cluster_offset) {
 468             if (bs->backing_hd) {
 469                 /* read from the base image */
 470                 hd_iov.iov_base = (void *)buf;
 471                 hd_iov.iov_len = n * 512;
 472                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 473                 qemu_co_mutex_unlock(&s->lock);
 474                 ret = bdrv_co_readv(bs->backing_hd, sector_num,
 475                                     n, &hd_qiov);
 476                 qemu_co_mutex_lock(&s->lock);
 477                 if (ret < 0) {
 478                     goto fail;
 479                 }
 480             } else {
 481                 /* Note: in this case, no need to wait */
 482                 memset(buf, 0, 512 * n);
 483             }
 484         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
 485             /* add AIO support for compressed blocks ? */
 486             if (decompress_cluster(bs, cluster_offset) < 0) {
 487                 goto fail;
 488             }
 489             memcpy(buf,
 490                    s->cluster_cache + index_in_cluster * 512, 512 * n);
 491         } else {
 492             if ((cluster_offset & 511) != 0) {
 493                 goto fail;
 494             }
 495             hd_iov.iov_base = (void *)buf;
 496             hd_iov.iov_len = n * 512;
 497             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 498             qemu_co_mutex_unlock(&s->lock);
 499             ret = bdrv_co_readv(bs->file,
 500                                 (cluster_offset >> 9) + index_in_cluster,
 501                                 n, &hd_qiov);
 502             qemu_co_mutex_lock(&s->lock);
 503             if (ret < 0) {
 504                 break;
 505             }
 506             if (s->crypt_method) {
 507                 encrypt_sectors(s, sector_num, buf, buf,
 508                                 n, 0,
 509                                 &s->aes_decrypt_key);
 510             }
 511         }
 512         ret = 0;
 513
 514         nb_sectors -= n;
 515         sector_num += n;
 516         buf += n * 512;
 517     }
 518
 519 done:
 520     qemu_co_mutex_unlock(&s->lock);
 521
 522     if (qiov->niov > 1) {
 523         qemu_iovec_from_buffer(qiov, orig_buf, qiov->size);
 524         qemu_vfree(orig_buf);
 525     }
 526
 527     return ret;
 528
 529 fail:
 530     ret = -EIO;
 531     goto done;
 532 }
 533
 534 static int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
 535                           int nb_sectors, QEMUIOVector *qiov)
 536 {
 537     BDRVQcowState *s = bs->opaque;
 538     int index_in_cluster;
 539     uint64_t cluster_offset;
 540     const uint8_t *src_buf;
 541     int ret = 0, n;
 542     uint8_t *cluster_data = NULL;
 543     struct iovec hd_iov;
 544     QEMUIOVector hd_qiov;
 545     uint8_t *buf;
 546     void *orig_buf;
 547
 548     s->cluster_cache_offset = -1; /* disable compressed cache */
 549
 550     if (qiov->niov > 1) {
 551         buf = orig_buf = qemu_blockalign(bs, qiov->size);
 552         qemu_iovec_to_buffer(qiov, buf);
 553     } else {
 554         orig_buf = NULL;
 555         buf = (uint8_t *)qiov->iov->iov_base;
 556     }
 557
 558     qemu_co_mutex_lock(&s->lock);
 559
 560     while (nb_sectors != 0) {
 561
 562         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 563         n = s->cluster_sectors - index_in_cluster;
 564         if (n > nb_sectors) {
 565             n = nb_sectors;
 566         }
 567         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
 568                                             index_in_cluster,
 569                                             index_in_cluster + n);
 570         if (!cluster_offset || (cluster_offset & 511) != 0) {
 571             ret = -EIO;
 572             break;
 573         }
 574         if (s->crypt_method) {
 575             if (!cluster_data) {
 576                 cluster_data = g_malloc0(s->cluster_size);
 577             }
 578             encrypt_sectors(s, sector_num, cluster_data, buf,
 579                             n, 1, &s->aes_encrypt_key);
 580             src_buf = cluster_data;
 581         } else {
 582             src_buf = buf;
 583         }
 584
 585         hd_iov.iov_base = (void *)src_buf;
 586         hd_iov.iov_len = n * 512;
 587         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
 588         qemu_co_mutex_unlock(&s->lock);
 589         ret = bdrv_co_writev(bs->file,
 590                              (cluster_offset >> 9) + index_in_cluster,
 591                              n, &hd_qiov);
 592         qemu_co_mutex_lock(&s->lock);
 593         if (ret < 0) {
 594             break;
 595         }
 596         ret = 0;
 597
 598         nb_sectors -= n;
 599         sector_num += n;
 600         buf += n * 512;
 601     }
 602     qemu_co_mutex_unlock(&s->lock);
 603
 604     if (qiov->niov > 1) {
 605         qemu_vfree(orig_buf);
 606     }
 607     g_free(cluster_data);
 608
 609     return ret;
 610 }
 611
 612 static void qcow_close(BlockDriverState *bs)
 613 {
 614     BDRVQcowState *s = bs->opaque;
 615
 616     g_free(s->l1_table);
 617     g_free(s->l2_cache);
 618     g_free(s->cluster_cache);
 619     g_free(s->cluster_data);
 620
 621     migrate_del_blocker(s->migration_blocker);
 622     error_free(s->migration_blocker);
 623 }
 624
 625 static int qcow_create(const char *filename, QEMUOptionParameter *options)
 626 {
 627     int fd, header_size, backing_filename_len, l1_size, i, shift;
 628     QCowHeader header;
 629     uint64_t tmp;
 630     int64_t total_size = 0;
 631     const char *backing_file = NULL;
 632     int flags = 0;
 633     int ret;
 634
 635     /* Read out options */
 636     while (options && options->name) {
 637         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 638             total_size = options->value.n / 512;
 639         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
 640             backing_file = options->value.s;
 641         } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
 642             flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
 643         }
 644         options++;
 645     }
 646
 647     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
 648     if (fd < 0)
 649         return -errno;
 650     memset(&header, 0, sizeof(header));
 651     header.magic = cpu_to_be32(QCOW_MAGIC);
 652     header.version = cpu_to_be32(QCOW_VERSION);
 653     header.size = cpu_to_be64(total_size * 512);
 654     header_size = sizeof(header);
 655     backing_filename_len = 0;
 656     if (backing_file) {
 657         if (strcmp(backing_file, "fat:")) {
 658             header.backing_file_offset = cpu_to_be64(header_size);
 659             backing_filename_len = strlen(backing_file);
 660             header.backing_file_size = cpu_to_be32(backing_filename_len);
 661             header_size += backing_filename_len;
 662         } else {
 663             /* special backing file for vvfat */
 664             backing_file = NULL;
 665         }
 666         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
 667                                     unmodifyed sectors */
 668         header.l2_bits = 12; /* 32 KB L2 tables */
 669     } else {
 670         header.cluster_bits = 12; /* 4 KB clusters */
 671         header.l2_bits = 9; /* 4 KB L2 tables */
 672     }
 673     header_size = (header_size + 7) & ~7;
 674     shift = header.cluster_bits + header.l2_bits;
 675     l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
 676
 677     header.l1_table_offset = cpu_to_be64(header_size);
 678     if (flags & BLOCK_FLAG_ENCRYPT) {
 679         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
 680     } else {
 681         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
 682     }
 683
 684     /* write all the data */
 685     ret = qemu_write_full(fd, &header, sizeof(header));
 686     if (ret != sizeof(header)) {
 687         ret = -errno;
 688         goto exit;
 689     }
 690
 691     if (backing_file) {
 692         ret = qemu_write_full(fd, backing_file, backing_filename_len);
 693         if (ret != backing_filename_len) {
 694             ret = -errno;
 695             goto exit;
 696         }
 697
 698     }
 699     lseek(fd, header_size, SEEK_SET);
 700     tmp = 0;
 701     for(i = 0;i < l1_size; i++) {
 702         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
 703         if (ret != sizeof(tmp)) {
 704             ret = -errno;
 705             goto exit;
 706         }
 707     }
 708
 709     ret = 0;
 710 exit:
 711     close(fd);
 712     return ret;
 713 }
 714
 715 static int qcow_make_empty(BlockDriverState *bs)
 716 {
 717     BDRVQcowState *s = bs->opaque;
 718     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
 719     int ret;
 720
 721     memset(s->l1_table, 0, l1_length);
 722     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
 723             l1_length) < 0)
 724         return -1;
 725     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
 726     if (ret < 0)
 727         return ret;
 728
 729     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 730     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
 731     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
 732
 733     return 0;
 734 }
 735
 736 /* XXX: put compressed sectors first, then all the cluster aligned
 737    tables to avoid losing bytes in alignment */
 738 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
 739                                  const uint8_t *buf, int nb_sectors)
 740 {
 741     BDRVQcowState *s = bs->opaque;
 742     z_stream strm;
 743     int ret, out_len;
 744     uint8_t *out_buf;
 745     uint64_t cluster_offset;
 746
 747     if (nb_sectors != s->cluster_sectors)
 748         return -EINVAL;
 749
 750     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
 751
 752     /* best compression, small window, no zlib header */
 753     memset(&strm, 0, sizeof(strm));
 754     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
 755                        Z_DEFLATED, -12,
 756                        9, Z_DEFAULT_STRATEGY);
 757     if (ret != 0) {
 758         ret = -EINVAL;
 759         goto fail;
 760     }
 761
 762     strm.avail_in = s->cluster_size;
 763     strm.next_in = (uint8_t *)buf;
 764     strm.avail_out = s->cluster_size;
 765     strm.next_out = out_buf;
 766
 767     ret = deflate(&strm, Z_FINISH);
 768     if (ret != Z_STREAM_END && ret != Z_OK) {
 769         deflateEnd(&strm);
 770         ret = -EINVAL;
 771         goto fail;
 772     }
 773     out_len = strm.next_out - out_buf;
 774
 775     deflateEnd(&strm);
 776
 777     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
 778         /* could not compress: write normal cluster */
 779         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
 780         if (ret < 0) {
 781             goto fail;
 782         }
 783     } else {
 784         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
 785                                             out_len, 0, 0);
 786         if (cluster_offset == 0) {
 787             ret = -EIO;
 788             goto fail;
 789         }
 790
 791         cluster_offset &= s->cluster_offset_mask;
 792         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
 793         if (ret < 0) {
 794             goto fail;
 795         }
 796     }
 797
 798     ret = 0;
 799 fail:
 800     g_free(out_buf);
 801     return ret;
 802 }
 803
 804 static coroutine_fn int qcow_co_flush(BlockDriverState *bs)
 805 {
 806     return bdrv_co_flush(bs->file);
 807 }
 808
 809 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 810 {
 811     BDRVQcowState *s = bs->opaque;
 812     bdi->cluster_size = s->cluster_size;
 813     return 0;
 814 }
 815
 816
 817 static QEMUOptionParameter qcow_create_options[] = {
 818     {
 819         .name = BLOCK_OPT_SIZE,
 820         .type = OPT_SIZE,
 821         .help = "Virtual disk size"
 822     },
 823     {
 824         .name = BLOCK_OPT_BACKING_FILE,
 825         .type = OPT_STRING,
 826         .help = "File name of a base image"
 827     },
 828     {
 829         .name = BLOCK_OPT_ENCRYPT,
 830         .type = OPT_FLAG,
 831         .help = "Encrypt the image"
 832     },
 833     { NULL }
 834 };
 835
 836 static BlockDriver bdrv_qcow = {
 837     .format_name        = "qcow",
 838     .instance_size      = sizeof(BDRVQcowState),
 839     .bdrv_probe         = qcow_probe,
 840     .bdrv_open          = qcow_open,
 841     .bdrv_close         = qcow_close,
 842     .bdrv_create        = qcow_create,
 843
 844     .bdrv_co_readv          = qcow_co_readv,
 845     .bdrv_co_writev         = qcow_co_writev,
 846     .bdrv_co_flush_to_disk  = qcow_co_flush,
 847     .bdrv_is_allocated      = qcow_is_allocated,
 848
 849     .bdrv_set_key           = qcow_set_key,
 850     .bdrv_make_empty        = qcow_make_empty,
 851     .bdrv_write_compressed  = qcow_write_compressed,
 852     .bdrv_get_info          = qcow_get_info,
 853
 854     .create_options = qcow_create_options,
 855 };
 856
 857 static void bdrv_qcow_init(void)
 858 {
 859     bdrv_register(&bdrv_qcow);
 860 }
 861
 862 block_init(bdrv_qcow_init);