block/qcow2.c

   1 /*
   2  * Block driver for the QCOW version 2 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 "block/qcow2.h"
  30
  31 /*
  32   Differences with QCOW:
  33
  34   - Support for multiple incremental snapshots.
  35   - Memory management by reference counts.
  36   - Clusters which have a reference count of one have the bit
  37     QCOW_OFLAG_COPIED to optimize write performance.
  38   - Size of compressed clusters is stored in sectors to reduce bit usage
  39     in the cluster offsets.
  40   - Support for storing additional data (such as the VM state) in the
  41     snapshots.
  42   - If a backing store is used, the cluster size is not constrained
  43     (could be backported to QCOW).
  44   - L2 tables have always a size of one cluster.
  45 */
  46
  47
  48 typedef struct {
  49     uint32_t magic;
  50     uint32_t len;
  51 } QCowExtension;
  52 #define  QCOW_EXT_MAGIC_END 0
  53 #define  QCOW_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
  54
  55
  56
  57 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
  58 {
  59     const QCowHeader *cow_header = (const void *)buf;
  60
  61     if (buf_size >= sizeof(QCowHeader) &&
  62         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
  63         be32_to_cpu(cow_header->version) == QCOW_VERSION)
  64         return 100;
  65     else
  66         return 0;
  67 }
  68
  69
  70 /*
  71  * read qcow2 extension and fill bs
  72  * start reading from start_offset
  73  * finish reading upon magic of value 0 or when end_offset reached
  74  * unknown magic is skipped (future extension this version knows nothing about)
  75  * return 0 upon success, non-0 otherwise
  76  */
  77 static int qcow_read_extensions(BlockDriverState *bs, uint64_t start_offset,
  78                                 uint64_t end_offset)
  79 {
  80     BDRVQcowState *s = bs->opaque;
  81     QCowExtension ext;
  82     uint64_t offset;
  83
  84 #ifdef DEBUG_EXT
  85     printf("qcow_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
  86 #endif
  87     offset = start_offset;
  88     while (offset < end_offset) {
  89
  90 #ifdef DEBUG_EXT
  91         /* Sanity check */
  92         if (offset > s->cluster_size)
  93             printf("qcow_handle_extension: suspicious offset %lu\n", offset);
  94
  95         printf("attemting to read extended header in offset %lu\n", offset);
  96 #endif
  97
  98         if (bdrv_pread(s->hd, offset, &ext, sizeof(ext)) != sizeof(ext)) {
  99             fprintf(stderr, "qcow_handle_extension: ERROR: pread fail from offset %llu\n",
 100                     (unsigned long long)offset);
 101             return 1;
 102         }
 103         be32_to_cpus(&ext.magic);
 104         be32_to_cpus(&ext.len);
 105         offset += sizeof(ext);
 106 #ifdef DEBUG_EXT
 107         printf("ext.magic = 0x%x\n", ext.magic);
 108 #endif
 109         switch (ext.magic) {
 110         case QCOW_EXT_MAGIC_END:
 111             return 0;
 112
 113         case QCOW_EXT_MAGIC_BACKING_FORMAT:
 114             if (ext.len >= sizeof(bs->backing_format)) {
 115                 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
 116                         " (>=%zu)\n",
 117                         ext.len, sizeof(bs->backing_format));
 118                 return 2;
 119             }
 120             if (bdrv_pread(s->hd, offset , bs->backing_format,
 121                            ext.len) != ext.len)
 122                 return 3;
 123             bs->backing_format[ext.len] = '\0';
 124 #ifdef DEBUG_EXT
 125             printf("Qcow2: Got format extension %s\n", bs->backing_format);
 126 #endif
 127             offset += ((ext.len + 7) & ~7);
 128             break;
 129
 130         default:
 131             /* unknown magic -- just skip it */
 132             offset += ((ext.len + 7) & ~7);
 133             break;
 134         }
 135     }
 136
 137     return 0;
 138 }
 139
 140
 141 static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
 142 {
 143     BDRVQcowState *s = bs->opaque;
 144     int len, i, shift, ret;
 145     QCowHeader header;
 146     uint64_t ext_end;
 147
 148     ret = bdrv_file_open(&s->hd, filename, flags);
 149     if (ret < 0)
 150         return ret;
 151     if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
 152         goto fail;
 153     be32_to_cpus(&header.magic);
 154     be32_to_cpus(&header.version);
 155     be64_to_cpus(&header.backing_file_offset);
 156     be32_to_cpus(&header.backing_file_size);
 157     be64_to_cpus(&header.size);
 158     be32_to_cpus(&header.cluster_bits);
 159     be32_to_cpus(&header.crypt_method);
 160     be64_to_cpus(&header.l1_table_offset);
 161     be32_to_cpus(&header.l1_size);
 162     be64_to_cpus(&header.refcount_table_offset);
 163     be32_to_cpus(&header.refcount_table_clusters);
 164     be64_to_cpus(&header.snapshots_offset);
 165     be32_to_cpus(&header.nb_snapshots);
 166
 167     if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
 168         goto fail;
 169     if (header.size <= 1 ||
 170         header.cluster_bits < MIN_CLUSTER_BITS ||
 171         header.cluster_bits > MAX_CLUSTER_BITS)
 172         goto fail;
 173     if (header.crypt_method > QCOW_CRYPT_AES)
 174         goto fail;
 175     s->crypt_method_header = header.crypt_method;
 176     if (s->crypt_method_header)
 177         bs->encrypted = 1;
 178     s->cluster_bits = header.cluster_bits;
 179     s->cluster_size = 1 << s->cluster_bits;
 180     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 181     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
 182     s->l2_size = 1 << s->l2_bits;
 183     bs->total_sectors = header.size / 512;
 184     s->csize_shift = (62 - (s->cluster_bits - 8));
 185     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
 186     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
 187     s->refcount_table_offset = header.refcount_table_offset;
 188     s->refcount_table_size =
 189         header.refcount_table_clusters << (s->cluster_bits - 3);
 190
 191     s->snapshots_offset = header.snapshots_offset;
 192     s->nb_snapshots = header.nb_snapshots;
 193
 194     /* read the level 1 table */
 195     s->l1_size = header.l1_size;
 196     shift = s->cluster_bits + s->l2_bits;
 197     s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
 198     /* the L1 table must contain at least enough entries to put
 199        header.size bytes */
 200     if (s->l1_size < s->l1_vm_state_index)
 201         goto fail;
 202     s->l1_table_offset = header.l1_table_offset;
 203     s->l1_table = qemu_mallocz(
 204         align_offset(s->l1_size * sizeof(uint64_t), 512));
 205     if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
 206         s->l1_size * sizeof(uint64_t))
 207         goto fail;
 208     for(i = 0;i < s->l1_size; i++) {
 209         be64_to_cpus(&s->l1_table[i]);
 210     }
 211     /* alloc L2 cache */
 212     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
 213     s->cluster_cache = qemu_malloc(s->cluster_size);
 214     /* one more sector for decompressed data alignment */
 215     s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
 216                                   + 512);
 217     s->cluster_cache_offset = -1;
 218
 219     if (qcow2_refcount_init(bs) < 0)
 220         goto fail;
 221
 222     /* read qcow2 extensions */
 223     if (header.backing_file_offset)
 224         ext_end = header.backing_file_offset;
 225     else
 226         ext_end = s->cluster_size;
 227     if (qcow_read_extensions(bs, sizeof(header), ext_end))
 228         goto fail;
 229
 230     /* read the backing file name */
 231     if (header.backing_file_offset != 0) {
 232         len = header.backing_file_size;
 233         if (len > 1023)
 234             len = 1023;
 235         if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
 236             goto fail;
 237         bs->backing_file[len] = '\0';
 238     }
 239     if (qcow2_read_snapshots(bs) < 0)
 240         goto fail;
 241
 242 #ifdef DEBUG_ALLOC
 243     qcow2_check_refcounts(bs);
 244 #endif
 245     return 0;
 246
 247  fail:
 248     qcow2_free_snapshots(bs);
 249     qcow2_refcount_close(bs);
 250     qemu_free(s->l1_table);
 251     qemu_free(s->l2_cache);
 252     qemu_free(s->cluster_cache);
 253     qemu_free(s->cluster_data);
 254     bdrv_delete(s->hd);
 255     return -1;
 256 }
 257
 258 static int qcow_set_key(BlockDriverState *bs, const char *key)
 259 {
 260     BDRVQcowState *s = bs->opaque;
 261     uint8_t keybuf[16];
 262     int len, i;
 263
 264     memset(keybuf, 0, 16);
 265     len = strlen(key);
 266     if (len > 16)
 267         len = 16;
 268     /* XXX: we could compress the chars to 7 bits to increase
 269        entropy */
 270     for(i = 0;i < len;i++) {
 271         keybuf[i] = key[i];
 272     }
 273     s->crypt_method = s->crypt_method_header;
 274
 275     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
 276         return -1;
 277     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
 278         return -1;
 279 #if 0
 280     /* test */
 281     {
 282         uint8_t in[16];
 283         uint8_t out[16];
 284         uint8_t tmp[16];
 285         for(i=0;i<16;i++)
 286             in[i] = i;
 287         AES_encrypt(in, tmp, &s->aes_encrypt_key);
 288         AES_decrypt(tmp, out, &s->aes_decrypt_key);
 289         for(i = 0; i < 16; i++)
 290             printf(" %02x", tmp[i]);
 291         printf("\n");
 292         for(i = 0; i < 16; i++)
 293             printf(" %02x", out[i]);
 294         printf("\n");
 295     }
 296 #endif
 297     return 0;
 298 }
 299
 300 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
 301                              int nb_sectors, int *pnum)
 302 {
 303     uint64_t cluster_offset;
 304
 305     *pnum = nb_sectors;
 306     cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, pnum);
 307
 308     return (cluster_offset != 0);
 309 }
 310
 311 /* handle reading after the end of the backing file */
 312 int qcow2_backing_read1(BlockDriverState *bs,
 313                   int64_t sector_num, uint8_t *buf, int nb_sectors)
 314 {
 315     int n1;
 316     if ((sector_num + nb_sectors) <= bs->total_sectors)
 317         return nb_sectors;
 318     if (sector_num >= bs->total_sectors)
 319         n1 = 0;
 320     else
 321         n1 = bs->total_sectors - sector_num;
 322     memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
 323     return n1;
 324 }
 325
 326 typedef struct QCowAIOCB {
 327     BlockDriverAIOCB common;
 328     int64_t sector_num;
 329     QEMUIOVector *qiov;
 330     uint8_t *buf;
 331     void *orig_buf;
 332     int nb_sectors;
 333     int n;
 334     uint64_t cluster_offset;
 335     uint8_t *cluster_data;
 336     BlockDriverAIOCB *hd_aiocb;
 337     struct iovec hd_iov;
 338     QEMUIOVector hd_qiov;
 339     QEMUBH *bh;
 340     QCowL2Meta l2meta;
 341 } QCowAIOCB;
 342
 343 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
 344 {
 345     QCowAIOCB *acb = (QCowAIOCB *)blockacb;
 346     if (acb->hd_aiocb)
 347         bdrv_aio_cancel(acb->hd_aiocb);
 348     qemu_aio_release(acb);
 349 }
 350
 351 static AIOPool qcow_aio_pool = {
 352     .aiocb_size         = sizeof(QCowAIOCB),
 353     .cancel             = qcow_aio_cancel,
 354 };
 355
 356 static void qcow_aio_read_cb(void *opaque, int ret);
 357 static void qcow_aio_read_bh(void *opaque)
 358 {
 359     QCowAIOCB *acb = opaque;
 360     qemu_bh_delete(acb->bh);
 361     acb->bh = NULL;
 362     qcow_aio_read_cb(opaque, 0);
 363 }
 364
 365 static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
 366 {
 367     if (acb->bh)
 368         return -EIO;
 369
 370     acb->bh = qemu_bh_new(cb, acb);
 371     if (!acb->bh)
 372         return -EIO;
 373
 374     qemu_bh_schedule(acb->bh);
 375
 376     return 0;
 377 }
 378
 379 static void qcow_aio_read_cb(void *opaque, int ret)
 380 {
 381     QCowAIOCB *acb = opaque;
 382     BlockDriverState *bs = acb->common.bs;
 383     BDRVQcowState *s = bs->opaque;
 384     int index_in_cluster, n1;
 385
 386     acb->hd_aiocb = NULL;
 387     if (ret < 0)
 388         goto done;
 389
 390     /* post process the read buffer */
 391     if (!acb->cluster_offset) {
 392         /* nothing to do */
 393     } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
 394         /* nothing to do */
 395     } else {
 396         if (s->crypt_method) {
 397             qcow2_encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
 398                             acb->n, 0,
 399                             &s->aes_decrypt_key);
 400         }
 401     }
 402
 403     acb->nb_sectors -= acb->n;
 404     acb->sector_num += acb->n;
 405     acb->buf += acb->n * 512;
 406
 407     if (acb->nb_sectors == 0) {
 408         /* request completed */
 409         ret = 0;
 410         goto done;
 411     }
 412
 413     /* prepare next AIO request */
 414     acb->n = acb->nb_sectors;
 415     acb->cluster_offset =
 416         qcow2_get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
 417     index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
 418
 419     if (!acb->cluster_offset) {
 420         if (bs->backing_hd) {
 421             /* read from the base image */
 422             n1 = qcow2_backing_read1(bs->backing_hd, acb->sector_num,
 423                                acb->buf, acb->n);
 424             if (n1 > 0) {
 425                 acb->hd_iov.iov_base = (void *)acb->buf;
 426                 acb->hd_iov.iov_len = acb->n * 512;
 427                 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
 428                 acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
 429                                     &acb->hd_qiov, acb->n,
 430                                     qcow_aio_read_cb, acb);
 431                 if (acb->hd_aiocb == NULL)
 432                     goto done;
 433             } else {
 434                 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
 435                 if (ret < 0)
 436                     goto done;
 437             }
 438         } else {
 439             /* Note: in this case, no need to wait */
 440             memset(acb->buf, 0, 512 * acb->n);
 441             ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
 442             if (ret < 0)
 443                 goto done;
 444         }
 445     } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
 446         /* add AIO support for compressed blocks ? */
 447         if (qcow2_decompress_cluster(s, acb->cluster_offset) < 0)
 448             goto done;
 449         memcpy(acb->buf,
 450                s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
 451         ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
 452         if (ret < 0)
 453             goto done;
 454     } else {
 455         if ((acb->cluster_offset & 511) != 0) {
 456             ret = -EIO;
 457             goto done;
 458         }
 459
 460         acb->hd_iov.iov_base = (void *)acb->buf;
 461         acb->hd_iov.iov_len = acb->n * 512;
 462         qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
 463         acb->hd_aiocb = bdrv_aio_readv(s->hd,
 464                             (acb->cluster_offset >> 9) + index_in_cluster,
 465                             &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
 466         if (acb->hd_aiocb == NULL)
 467             goto done;
 468     }
 469
 470     return;
 471 done:
 472     if (acb->qiov->niov > 1) {
 473         qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
 474         qemu_vfree(acb->orig_buf);
 475     }
 476     acb->common.cb(acb->common.opaque, ret);
 477     qemu_aio_release(acb);
 478 }
 479
 480 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
 481         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 482         BlockDriverCompletionFunc *cb, void *opaque, int is_write)
 483 {
 484     QCowAIOCB *acb;
 485
 486     acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
 487     if (!acb)
 488         return NULL;
 489     acb->hd_aiocb = NULL;
 490     acb->sector_num = sector_num;
 491     acb->qiov = qiov;
 492     if (qiov->niov > 1) {
 493         acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
 494         if (is_write)
 495             qemu_iovec_to_buffer(qiov, acb->buf);
 496     } else {
 497         acb->buf = (uint8_t *)qiov->iov->iov_base;
 498     }
 499     acb->nb_sectors = nb_sectors;
 500     acb->n = 0;
 501     acb->cluster_offset = 0;
 502     acb->l2meta.nb_clusters = 0;
 503     return acb;
 504 }
 505
 506 static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
 507         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 508         BlockDriverCompletionFunc *cb, void *opaque)
 509 {
 510     QCowAIOCB *acb;
 511
 512     acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
 513     if (!acb)
 514         return NULL;
 515
 516     qcow_aio_read_cb(acb, 0);
 517     return &acb->common;
 518 }
 519
 520 static void qcow_aio_write_cb(void *opaque, int ret)
 521 {
 522     QCowAIOCB *acb = opaque;
 523     BlockDriverState *bs = acb->common.bs;
 524     BDRVQcowState *s = bs->opaque;
 525     int index_in_cluster;
 526     const uint8_t *src_buf;
 527     int n_end;
 528
 529     acb->hd_aiocb = NULL;
 530
 531     if (ret < 0)
 532         goto done;
 533
 534     if (qcow2_alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) {
 535         qcow2_free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters);
 536         goto done;
 537     }
 538
 539     acb->nb_sectors -= acb->n;
 540     acb->sector_num += acb->n;
 541     acb->buf += acb->n * 512;
 542
 543     if (acb->nb_sectors == 0) {
 544         /* request completed */
 545         ret = 0;
 546         goto done;
 547     }
 548
 549     index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
 550     n_end = index_in_cluster + acb->nb_sectors;
 551     if (s->crypt_method &&
 552         n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
 553         n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
 554
 555     acb->cluster_offset = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9,
 556                                           index_in_cluster,
 557                                           n_end, &acb->n, &acb->l2meta);
 558     if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
 559         ret = -EIO;
 560         goto done;
 561     }
 562     if (s->crypt_method) {
 563         if (!acb->cluster_data) {
 564             acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
 565                                              s->cluster_size);
 566         }
 567         qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
 568                         acb->n, 1, &s->aes_encrypt_key);
 569         src_buf = acb->cluster_data;
 570     } else {
 571         src_buf = acb->buf;
 572     }
 573     acb->hd_iov.iov_base = (void *)src_buf;
 574     acb->hd_iov.iov_len = acb->n * 512;
 575     qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
 576     acb->hd_aiocb = bdrv_aio_writev(s->hd,
 577                                     (acb->cluster_offset >> 9) + index_in_cluster,
 578                                     &acb->hd_qiov, acb->n,
 579                                     qcow_aio_write_cb, acb);
 580     if (acb->hd_aiocb == NULL)
 581         goto done;
 582
 583     return;
 584
 585 done:
 586     if (acb->qiov->niov > 1)
 587         qemu_vfree(acb->orig_buf);
 588     acb->common.cb(acb->common.opaque, ret);
 589     qemu_aio_release(acb);
 590 }
 591
 592 static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
 593         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 594         BlockDriverCompletionFunc *cb, void *opaque)
 595 {
 596     BDRVQcowState *s = bs->opaque;
 597     QCowAIOCB *acb;
 598
 599     s->cluster_cache_offset = -1; /* disable compressed cache */
 600
 601     acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
 602     if (!acb)
 603         return NULL;
 604
 605     qcow_aio_write_cb(acb, 0);
 606     return &acb->common;
 607 }
 608
 609 static void qcow_close(BlockDriverState *bs)
 610 {
 611     BDRVQcowState *s = bs->opaque;
 612     qemu_free(s->l1_table);
 613     qemu_free(s->l2_cache);
 614     qemu_free(s->cluster_cache);
 615     qemu_free(s->cluster_data);
 616     qcow2_refcount_close(bs);
 617     bdrv_delete(s->hd);
 618 }
 619
 620 static int get_bits_from_size(size_t size)
 621 {
 622     int res = 0;
 623
 624     if (size == 0) {
 625         return -1;
 626     }
 627
 628     while (size != 1) {
 629         /* Not a power of two */
 630         if (size & 1) {
 631             return -1;
 632         }
 633
 634         size >>= 1;
 635         res++;
 636     }
 637
 638     return res;
 639 }
 640
 641 static int qcow_create2(const char *filename, int64_t total_size,
 642                         const char *backing_file, const char *backing_format,
 643                         int flags, size_t cluster_size)
 644 {
 645
 646     int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
 647     int ref_clusters, backing_format_len = 0;
 648     QCowHeader header;
 649     uint64_t tmp, offset;
 650     QCowCreateState s1, *s = &s1;
 651     QCowExtension ext_bf = {0, 0};
 652
 653
 654     memset(s, 0, sizeof(*s));
 655
 656     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
 657     if (fd < 0)
 658         return -1;
 659     memset(&header, 0, sizeof(header));
 660     header.magic = cpu_to_be32(QCOW_MAGIC);
 661     header.version = cpu_to_be32(QCOW_VERSION);
 662     header.size = cpu_to_be64(total_size * 512);
 663     header_size = sizeof(header);
 664     backing_filename_len = 0;
 665     if (backing_file) {
 666         if (backing_format) {
 667             ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
 668             backing_format_len = strlen(backing_format);
 669             ext_bf.len = (backing_format_len + 7) & ~7;
 670             header_size += ((sizeof(ext_bf) + ext_bf.len + 7) & ~7);
 671         }
 672         header.backing_file_offset = cpu_to_be64(header_size);
 673         backing_filename_len = strlen(backing_file);
 674         header.backing_file_size = cpu_to_be32(backing_filename_len);
 675         header_size += backing_filename_len;
 676     }
 677
 678     /* Cluster size */
 679     s->cluster_bits = get_bits_from_size(cluster_size);
 680     if (s->cluster_bits < MIN_CLUSTER_BITS ||
 681         s->cluster_bits > MAX_CLUSTER_BITS)
 682     {
 683         fprintf(stderr, "Cluster size must be a power of two between "
 684             "%d and %dk\n",
 685             1 << MIN_CLUSTER_BITS,
 686             1 << (MAX_CLUSTER_BITS - 10));
 687         return -EINVAL;
 688     }
 689     s->cluster_size = 1 << s->cluster_bits;
 690
 691     header.cluster_bits = cpu_to_be32(s->cluster_bits);
 692     header_size = (header_size + 7) & ~7;
 693     if (flags & BLOCK_FLAG_ENCRYPT) {
 694         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
 695     } else {
 696         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
 697     }
 698     l2_bits = s->cluster_bits - 3;
 699     shift = s->cluster_bits + l2_bits;
 700     l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
 701     offset = align_offset(header_size, s->cluster_size);
 702     s->l1_table_offset = offset;
 703     header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
 704     header.l1_size = cpu_to_be32(l1_size);
 705     offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
 706
 707     s->refcount_table = qemu_mallocz(s->cluster_size);
 708
 709     s->refcount_table_offset = offset;
 710     header.refcount_table_offset = cpu_to_be64(offset);
 711     header.refcount_table_clusters = cpu_to_be32(1);
 712     offset += s->cluster_size;
 713     s->refcount_block_offset = offset;
 714
 715     /* count how many refcount blocks needed */
 716     tmp = offset >> s->cluster_bits;
 717     ref_clusters = (tmp >> (s->cluster_bits - REFCOUNT_SHIFT)) + 1;
 718     for (i=0; i < ref_clusters; i++) {
 719         s->refcount_table[i] = cpu_to_be64(offset);
 720         offset += s->cluster_size;
 721     }
 722
 723     s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
 724
 725     /* update refcounts */
 726     qcow2_create_refcount_update(s, 0, header_size);
 727     qcow2_create_refcount_update(s, s->l1_table_offset,
 728         l1_size * sizeof(uint64_t));
 729     qcow2_create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
 730     qcow2_create_refcount_update(s, s->refcount_block_offset,
 731         ref_clusters * s->cluster_size);
 732
 733     /* write all the data */
 734     write(fd, &header, sizeof(header));
 735     if (backing_file) {
 736         if (backing_format_len) {
 737             char zero[16];
 738             int d = ext_bf.len - backing_format_len;
 739
 740             memset(zero, 0, sizeof(zero));
 741             cpu_to_be32s(&ext_bf.magic);
 742             cpu_to_be32s(&ext_bf.len);
 743             write(fd, &ext_bf, sizeof(ext_bf));
 744             write(fd, backing_format, backing_format_len);
 745             if (d>0) {
 746                 write(fd, zero, d);
 747             }
 748         }
 749         write(fd, backing_file, backing_filename_len);
 750     }
 751     lseek(fd, s->l1_table_offset, SEEK_SET);
 752     tmp = 0;
 753     for(i = 0;i < l1_size; i++) {
 754         write(fd, &tmp, sizeof(tmp));
 755     }
 756     lseek(fd, s->refcount_table_offset, SEEK_SET);
 757     write(fd, s->refcount_table, s->cluster_size);
 758
 759     lseek(fd, s->refcount_block_offset, SEEK_SET);
 760     write(fd, s->refcount_block, ref_clusters * s->cluster_size);
 761
 762     qemu_free(s->refcount_table);
 763     qemu_free(s->refcount_block);
 764     close(fd);
 765     return 0;
 766 }
 767
 768 static int qcow_create(const char *filename, QEMUOptionParameter *options)
 769 {
 770     const char *backing_file = NULL;
 771     const char *backing_fmt = NULL;
 772     uint64_t sectors = 0;
 773     int flags = 0;
 774     size_t cluster_size = 65536;
 775
 776     /* Read out options */
 777     while (options && options->name) {
 778         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 779             sectors = options->value.n / 512;
 780         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
 781             backing_file = options->value.s;
 782         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
 783             backing_fmt = options->value.s;
 784         } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
 785             flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
 786         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
 787             if (options->value.n) {
 788                 cluster_size = options->value.n;
 789             }
 790         }
 791         options++;
 792     }
 793
 794     return qcow_create2(filename, sectors, backing_file, backing_fmt, flags,
 795         cluster_size);
 796 }
 797
 798 static int qcow_make_empty(BlockDriverState *bs)
 799 {
 800 #if 0
 801     /* XXX: not correct */
 802     BDRVQcowState *s = bs->opaque;
 803     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
 804     int ret;
 805
 806     memset(s->l1_table, 0, l1_length);
 807     if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
 808         return -1;
 809     ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
 810     if (ret < 0)
 811         return ret;
 812
 813     l2_cache_reset(bs);
 814 #endif
 815     return 0;
 816 }
 817
 818 /* XXX: put compressed sectors first, then all the cluster aligned
 819    tables to avoid losing bytes in alignment */
 820 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
 821                                  const uint8_t *buf, int nb_sectors)
 822 {
 823     BDRVQcowState *s = bs->opaque;
 824     z_stream strm;
 825     int ret, out_len;
 826     uint8_t *out_buf;
 827     uint64_t cluster_offset;
 828
 829     if (nb_sectors == 0) {
 830         /* align end of file to a sector boundary to ease reading with
 831            sector based I/Os */
 832         cluster_offset = bdrv_getlength(s->hd);
 833         cluster_offset = (cluster_offset + 511) & ~511;
 834         bdrv_truncate(s->hd, cluster_offset);
 835         return 0;
 836     }
 837
 838     if (nb_sectors != s->cluster_sectors)
 839         return -EINVAL;
 840
 841     out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
 842
 843     /* best compression, small window, no zlib header */
 844     memset(&strm, 0, sizeof(strm));
 845     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
 846                        Z_DEFLATED, -12,
 847                        9, Z_DEFAULT_STRATEGY);
 848     if (ret != 0) {
 849         qemu_free(out_buf);
 850         return -1;
 851     }
 852
 853     strm.avail_in = s->cluster_size;
 854     strm.next_in = (uint8_t *)buf;
 855     strm.avail_out = s->cluster_size;
 856     strm.next_out = out_buf;
 857
 858     ret = deflate(&strm, Z_FINISH);
 859     if (ret != Z_STREAM_END && ret != Z_OK) {
 860         qemu_free(out_buf);
 861         deflateEnd(&strm);
 862         return -1;
 863     }
 864     out_len = strm.next_out - out_buf;
 865
 866     deflateEnd(&strm);
 867
 868     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
 869         /* could not compress: write normal cluster */
 870         bdrv_write(bs, sector_num, buf, s->cluster_sectors);
 871     } else {
 872         cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
 873             sector_num << 9, out_len);
 874         if (!cluster_offset)
 875             return -1;
 876         cluster_offset &= s->cluster_offset_mask;
 877         if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
 878             qemu_free(out_buf);
 879             return -1;
 880         }
 881     }
 882
 883     qemu_free(out_buf);
 884     return 0;
 885 }
 886
 887 static void qcow_flush(BlockDriverState *bs)
 888 {
 889     BDRVQcowState *s = bs->opaque;
 890     bdrv_flush(s->hd);
 891 }
 892
 893 static int64_t qcow_vm_state_offset(BDRVQcowState *s)
 894 {
 895         return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
 896 }
 897
 898 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 899 {
 900     BDRVQcowState *s = bs->opaque;
 901     bdi->cluster_size = s->cluster_size;
 902     bdi->vm_state_offset = qcow_vm_state_offset(s);
 903     return 0;
 904 }
 905
 906
 907 static int qcow_check(BlockDriverState *bs)
 908 {
 909     return qcow2_check_refcounts(bs);
 910 }
 911
 912 #if 0
 913 static void dump_refcounts(BlockDriverState *bs)
 914 {
 915     BDRVQcowState *s = bs->opaque;
 916     int64_t nb_clusters, k, k1, size;
 917     int refcount;
 918
 919     size = bdrv_getlength(s->hd);
 920     nb_clusters = size_to_clusters(s, size);
 921     for(k = 0; k < nb_clusters;) {
 922         k1 = k;
 923         refcount = get_refcount(bs, k);
 924         k++;
 925         while (k < nb_clusters && get_refcount(bs, k) == refcount)
 926             k++;
 927         printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
 928     }
 929 }
 930 #endif
 931
 932 static int qcow_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
 933                            int64_t pos, int size)
 934 {
 935     BDRVQcowState *s = bs->opaque;
 936     int growable = bs->growable;
 937
 938     bs->growable = 1;
 939     bdrv_pwrite(bs, qcow_vm_state_offset(s) + pos, buf, size);
 940     bs->growable = growable;
 941
 942     return size;
 943 }
 944
 945 static int qcow_load_vmstate(BlockDriverState *bs, uint8_t *buf,
 946                            int64_t pos, int size)
 947 {
 948     BDRVQcowState *s = bs->opaque;
 949     int growable = bs->growable;
 950     int ret;
 951
 952     bs->growable = 1;
 953     ret = bdrv_pread(bs, qcow_vm_state_offset(s) + pos, buf, size);
 954     bs->growable = growable;
 955
 956     return ret;
 957 }
 958
 959 static QEMUOptionParameter qcow_create_options[] = {
 960     {
 961         .name = BLOCK_OPT_SIZE,
 962         .type = OPT_SIZE,
 963         .help = "Virtual disk size"
 964     },
 965     {
 966         .name = BLOCK_OPT_BACKING_FILE,
 967         .type = OPT_STRING,
 968         .help = "File name of a base image"
 969     },
 970     {
 971         .name = BLOCK_OPT_BACKING_FMT,
 972         .type = OPT_STRING,
 973         .help = "Image format of the base image"
 974     },
 975     {
 976         .name = BLOCK_OPT_ENCRYPT,
 977         .type = OPT_FLAG,
 978         .help = "Encrypt the image"
 979     },
 980     {
 981         .name = BLOCK_OPT_CLUSTER_SIZE,
 982         .type = OPT_SIZE,
 983         .help = "qcow2 cluster size"
 984     },
 985     { NULL }
 986 };
 987
 988 static BlockDriver bdrv_qcow2 = {
 989     .format_name        = "qcow2",
 990     .instance_size      = sizeof(BDRVQcowState),
 991     .bdrv_probe         = qcow_probe,
 992     .bdrv_open          = qcow_open,
 993     .bdrv_close         = qcow_close,
 994     .bdrv_create        = qcow_create,
 995     .bdrv_flush         = qcow_flush,
 996     .bdrv_is_allocated  = qcow_is_allocated,
 997     .bdrv_set_key       = qcow_set_key,
 998     .bdrv_make_empty    = qcow_make_empty,
 999
1000     .bdrv_aio_readv     = qcow_aio_readv,
1001     .bdrv_aio_writev    = qcow_aio_writev,
1002     .bdrv_write_compressed = qcow_write_compressed,
1003
1004     .bdrv_snapshot_create   = qcow2_snapshot_create,
1005     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
1006     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
1007     .bdrv_snapshot_list     = qcow2_snapshot_list,
1008     .bdrv_get_info      = qcow_get_info,
1009
1010     .bdrv_save_vmstate    = qcow_save_vmstate,
1011     .bdrv_load_vmstate    = qcow_load_vmstate,
1012
1013     .create_options = qcow_create_options,
1014     .bdrv_check = qcow_check,
1015 };
1016
1017 static void bdrv_qcow2_init(void)
1018 {
1019     bdrv_register(&bdrv_qcow2);
1020 }
1021
1022 block_init(bdrv_qcow2_init);