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/block_int.h"
  26 #include "qemu/module.h"
  27 #include <zlib.h>
  28 #include "qemu/aes.h"
  29 #include "block/qcow2.h"
  30 #include "qemu/error-report.h"
  31 #include "qapi/qmp/qerror.h"
  32 #include "qapi/qmp/qbool.h"
  33 #include "qapi/util.h"
  34 #include "qapi/qmp/types.h"
  35 #include "qapi-event.h"
  36 #include "trace.h"
  37 #include "qemu/option_int.h"
  38
  39 /*
  40   Differences with QCOW:
  41
  42   - Support for multiple incremental snapshots.
  43   - Memory management by reference counts.
  44   - Clusters which have a reference count of one have the bit
  45     QCOW_OFLAG_COPIED to optimize write performance.
  46   - Size of compressed clusters is stored in sectors to reduce bit usage
  47     in the cluster offsets.
  48   - Support for storing additional data (such as the VM state) in the
  49     snapshots.
  50   - If a backing store is used, the cluster size is not constrained
  51     (could be backported to QCOW).
  52   - L2 tables have always a size of one cluster.
  53 */
  54
  55
  56 typedef struct {
  57     uint32_t magic;
  58     uint32_t len;
  59 } QEMU_PACKED QCowExtension;
  60
  61 #define  QCOW2_EXT_MAGIC_END 0
  62 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
  63 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
  64
  65 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
  66 {
  67     const QCowHeader *cow_header = (const void *)buf;
  68
  69     if (buf_size >= sizeof(QCowHeader) &&
  70         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
  71         be32_to_cpu(cow_header->version) >= 2)
  72         return 100;
  73     else
  74         return 0;
  75 }
  76
  77
  78 /*
  79  * read qcow2 extension and fill bs
  80  * start reading from start_offset
  81  * finish reading upon magic of value 0 or when end_offset reached
  82  * unknown magic is skipped (future extension this version knows nothing about)
  83  * return 0 upon success, non-0 otherwise
  84  */
  85 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
  86                                  uint64_t end_offset, void **p_feature_table,
  87                                  Error **errp)
  88 {
  89     BDRVQcowState *s = bs->opaque;
  90     QCowExtension ext;
  91     uint64_t offset;
  92     int ret;
  93
  94 #ifdef DEBUG_EXT
  95     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
  96 #endif
  97     offset = start_offset;
  98     while (offset < end_offset) {
  99
 100 #ifdef DEBUG_EXT
 101         /* Sanity check */
 102         if (offset > s->cluster_size)
 103             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
 104
 105         printf("attempting to read extended header in offset %lu\n", offset);
 106 #endif
 107
 108         ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
 109         if (ret < 0) {
 110             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
 111                              "pread fail from offset %" PRIu64, offset);
 112             return 1;
 113         }
 114         be32_to_cpus(&ext.magic);
 115         be32_to_cpus(&ext.len);
 116         offset += sizeof(ext);
 117 #ifdef DEBUG_EXT
 118         printf("ext.magic = 0x%x\n", ext.magic);
 119 #endif
 120         if (ext.len > end_offset - offset) {
 121             error_setg(errp, "Header extension too large");
 122             return -EINVAL;
 123         }
 124
 125         switch (ext.magic) {
 126         case QCOW2_EXT_MAGIC_END:
 127             return 0;
 128
 129         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
 130             if (ext.len >= sizeof(bs->backing_format)) {
 131                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
 132                            " too large (>=%zu)", ext.len,
 133                            sizeof(bs->backing_format));
 134                 return 2;
 135             }
 136             ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len);
 137             if (ret < 0) {
 138                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
 139                                  "Could not read format name");
 140                 return 3;
 141             }
 142             bs->backing_format[ext.len] = '\0';
 143 #ifdef DEBUG_EXT
 144             printf("Qcow2: Got format extension %s\n", bs->backing_format);
 145 #endif
 146             break;
 147
 148         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
 149             if (p_feature_table != NULL) {
 150                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
 151                 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
 152                 if (ret < 0) {
 153                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
 154                                      "Could not read table");
 155                     return ret;
 156                 }
 157
 158                 *p_feature_table = feature_table;
 159             }
 160             break;
 161
 162         default:
 163             /* unknown magic - save it in case we need to rewrite the header */
 164             {
 165                 Qcow2UnknownHeaderExtension *uext;
 166
 167                 uext = g_malloc0(sizeof(*uext)  + ext.len);
 168                 uext->magic = ext.magic;
 169                 uext->len = ext.len;
 170                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
 171
 172                 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
 173                 if (ret < 0) {
 174                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
 175                                      "Could not read data");
 176                     return ret;
 177                 }
 178             }
 179             break;
 180         }
 181
 182         offset += ((ext.len + 7) & ~7);
 183     }
 184
 185     return 0;
 186 }
 187
 188 static void cleanup_unknown_header_ext(BlockDriverState *bs)
 189 {
 190     BDRVQcowState *s = bs->opaque;
 191     Qcow2UnknownHeaderExtension *uext, *next;
 192
 193     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
 194         QLIST_REMOVE(uext, next);
 195         g_free(uext);
 196     }
 197 }
 198
 199 static void GCC_FMT_ATTR(3, 4) report_unsupported(BlockDriverState *bs,
 200     Error **errp, const char *fmt, ...)
 201 {
 202     char msg[64];
 203     va_list ap;
 204
 205     va_start(ap, fmt);
 206     vsnprintf(msg, sizeof(msg), fmt, ap);
 207     va_end(ap);
 208
 209     error_set(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
 210               bdrv_get_device_name(bs), "qcow2", msg);
 211 }
 212
 213 static void report_unsupported_feature(BlockDriverState *bs,
 214     Error **errp, Qcow2Feature *table, uint64_t mask)
 215 {
 216     char *features = g_strdup("");
 217     char *old;
 218
 219     while (table && table->name[0] != '\0') {
 220         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
 221             if (mask & (1ULL << table->bit)) {
 222                 old = features;
 223                 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
 224                                            table->name);
 225                 g_free(old);
 226                 mask &= ~(1ULL << table->bit);
 227             }
 228         }
 229         table++;
 230     }
 231
 232     if (mask) {
 233         old = features;
 234         features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
 235                                    old, *old ? ", " : "", mask);
 236         g_free(old);
 237     }
 238
 239     report_unsupported(bs, errp, "%s", features);
 240     g_free(features);
 241 }
 242
 243 /*
 244  * Sets the dirty bit and flushes afterwards if necessary.
 245  *
 246  * The incompatible_features bit is only set if the image file header was
 247  * updated successfully.  Therefore it is not required to check the return
 248  * value of this function.
 249  */
 250 int qcow2_mark_dirty(BlockDriverState *bs)
 251 {
 252     BDRVQcowState *s = bs->opaque;
 253     uint64_t val;
 254     int ret;
 255
 256     assert(s->qcow_version >= 3);
 257
 258     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
 259         return 0; /* already dirty */
 260     }
 261
 262     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
 263     ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
 264                       &val, sizeof(val));
 265     if (ret < 0) {
 266         return ret;
 267     }
 268     ret = bdrv_flush(bs->file);
 269     if (ret < 0) {
 270         return ret;
 271     }
 272
 273     /* Only treat image as dirty if the header was updated successfully */
 274     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
 275     return 0;
 276 }
 277
 278 /*
 279  * Clears the dirty bit and flushes before if necessary.  Only call this
 280  * function when there are no pending requests, it does not guard against
 281  * concurrent requests dirtying the image.
 282  */
 283 static int qcow2_mark_clean(BlockDriverState *bs)
 284 {
 285     BDRVQcowState *s = bs->opaque;
 286
 287     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
 288         int ret;
 289
 290         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
 291
 292         ret = bdrv_flush(bs);
 293         if (ret < 0) {
 294             return ret;
 295         }
 296
 297         return qcow2_update_header(bs);
 298     }
 299     return 0;
 300 }
 301
 302 /*
 303  * Marks the image as corrupt.
 304  */
 305 int qcow2_mark_corrupt(BlockDriverState *bs)
 306 {
 307     BDRVQcowState *s = bs->opaque;
 308
 309     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
 310     return qcow2_update_header(bs);
 311 }
 312
 313 /*
 314  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
 315  * before if necessary.
 316  */
 317 int qcow2_mark_consistent(BlockDriverState *bs)
 318 {
 319     BDRVQcowState *s = bs->opaque;
 320
 321     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
 322         int ret = bdrv_flush(bs);
 323         if (ret < 0) {
 324             return ret;
 325         }
 326
 327         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
 328         return qcow2_update_header(bs);
 329     }
 330     return 0;
 331 }
 332
 333 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
 334                        BdrvCheckMode fix)
 335 {
 336     int ret = qcow2_check_refcounts(bs, result, fix);
 337     if (ret < 0) {
 338         return ret;
 339     }
 340
 341     if (fix && result->check_errors == 0 && result->corruptions == 0) {
 342         ret = qcow2_mark_clean(bs);
 343         if (ret < 0) {
 344             return ret;
 345         }
 346         return qcow2_mark_consistent(bs);
 347     }
 348     return ret;
 349 }
 350
 351 static int validate_table_offset(BlockDriverState *bs, uint64_t offset,
 352                                  uint64_t entries, size_t entry_len)
 353 {
 354     BDRVQcowState *s = bs->opaque;
 355     uint64_t size;
 356
 357     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
 358      * because values will be passed to qemu functions taking int64_t. */
 359     if (entries > INT64_MAX / entry_len) {
 360         return -EINVAL;
 361     }
 362
 363     size = entries * entry_len;
 364
 365     if (INT64_MAX - size < offset) {
 366         return -EINVAL;
 367     }
 368
 369     /* Tables must be cluster aligned */
 370     if (offset & (s->cluster_size - 1)) {
 371         return -EINVAL;
 372     }
 373
 374     return 0;
 375 }
 376
 377 static QemuOptsList qcow2_runtime_opts = {
 378     .name = "qcow2",
 379     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
 380     .desc = {
 381         {
 382             .name = QCOW2_OPT_LAZY_REFCOUNTS,
 383             .type = QEMU_OPT_BOOL,
 384             .help = "Postpone refcount updates",
 385         },
 386         {
 387             .name = QCOW2_OPT_DISCARD_REQUEST,
 388             .type = QEMU_OPT_BOOL,
 389             .help = "Pass guest discard requests to the layer below",
 390         },
 391         {
 392             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
 393             .type = QEMU_OPT_BOOL,
 394             .help = "Generate discard requests when snapshot related space "
 395                     "is freed",
 396         },
 397         {
 398             .name = QCOW2_OPT_DISCARD_OTHER,
 399             .type = QEMU_OPT_BOOL,
 400             .help = "Generate discard requests when other clusters are freed",
 401         },
 402         {
 403             .name = QCOW2_OPT_OVERLAP,
 404             .type = QEMU_OPT_STRING,
 405             .help = "Selects which overlap checks to perform from a range of "
 406                     "templates (none, constant, cached, all)",
 407         },
 408         {
 409             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
 410             .type = QEMU_OPT_STRING,
 411             .help = "Selects which overlap checks to perform from a range of "
 412                     "templates (none, constant, cached, all)",
 413         },
 414         {
 415             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
 416             .type = QEMU_OPT_BOOL,
 417             .help = "Check for unintended writes into the main qcow2 header",
 418         },
 419         {
 420             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
 421             .type = QEMU_OPT_BOOL,
 422             .help = "Check for unintended writes into the active L1 table",
 423         },
 424         {
 425             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
 426             .type = QEMU_OPT_BOOL,
 427             .help = "Check for unintended writes into an active L2 table",
 428         },
 429         {
 430             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
 431             .type = QEMU_OPT_BOOL,
 432             .help = "Check for unintended writes into the refcount table",
 433         },
 434         {
 435             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
 436             .type = QEMU_OPT_BOOL,
 437             .help = "Check for unintended writes into a refcount block",
 438         },
 439         {
 440             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
 441             .type = QEMU_OPT_BOOL,
 442             .help = "Check for unintended writes into the snapshot table",
 443         },
 444         {
 445             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
 446             .type = QEMU_OPT_BOOL,
 447             .help = "Check for unintended writes into an inactive L1 table",
 448         },
 449         {
 450             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
 451             .type = QEMU_OPT_BOOL,
 452             .help = "Check for unintended writes into an inactive L2 table",
 453         },
 454         {
 455             .name = QCOW2_OPT_CACHE_SIZE,
 456             .type = QEMU_OPT_SIZE,
 457             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
 458                     "cache size",
 459         },
 460         {
 461             .name = QCOW2_OPT_L2_CACHE_SIZE,
 462             .type = QEMU_OPT_SIZE,
 463             .help = "Maximum L2 table cache size",
 464         },
 465         {
 466             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
 467             .type = QEMU_OPT_SIZE,
 468             .help = "Maximum refcount block cache size",
 469         },
 470         { /* end of list */ }
 471     },
 472 };
 473
 474 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
 475     [QCOW2_OL_MAIN_HEADER_BITNR]    = QCOW2_OPT_OVERLAP_MAIN_HEADER,
 476     [QCOW2_OL_ACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L1,
 477     [QCOW2_OL_ACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L2,
 478     [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
 479     [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
 480     [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
 481     [QCOW2_OL_INACTIVE_L1_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L1,
 482     [QCOW2_OL_INACTIVE_L2_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L2,
 483 };
 484
 485 static void read_cache_sizes(QemuOpts *opts, uint64_t *l2_cache_size,
 486                              uint64_t *refcount_cache_size, Error **errp)
 487 {
 488     uint64_t combined_cache_size;
 489     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
 490
 491     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
 492     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
 493     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
 494
 495     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
 496     *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0);
 497     *refcount_cache_size = qemu_opt_get_size(opts,
 498                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
 499
 500     if (combined_cache_size_set) {
 501         if (l2_cache_size_set && refcount_cache_size_set) {
 502             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
 503                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
 504                        "the same time");
 505             return;
 506         } else if (*l2_cache_size > combined_cache_size) {
 507             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
 508                        QCOW2_OPT_CACHE_SIZE);
 509             return;
 510         } else if (*refcount_cache_size > combined_cache_size) {
 511             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
 512                        QCOW2_OPT_CACHE_SIZE);
 513             return;
 514         }
 515
 516         if (l2_cache_size_set) {
 517             *refcount_cache_size = combined_cache_size - *l2_cache_size;
 518         } else if (refcount_cache_size_set) {
 519             *l2_cache_size = combined_cache_size - *refcount_cache_size;
 520         } else {
 521             *refcount_cache_size = combined_cache_size
 522                                  / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1);
 523             *l2_cache_size = combined_cache_size - *refcount_cache_size;
 524         }
 525     } else {
 526         if (!l2_cache_size_set && !refcount_cache_size_set) {
 527             *l2_cache_size = DEFAULT_L2_CACHE_BYTE_SIZE;
 528             *refcount_cache_size = *l2_cache_size
 529                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 530         } else if (!l2_cache_size_set) {
 531             *l2_cache_size = *refcount_cache_size
 532                            * DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 533         } else if (!refcount_cache_size_set) {
 534             *refcount_cache_size = *l2_cache_size
 535                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 536         }
 537     }
 538 }
 539
 540 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
 541                       Error **errp)
 542 {
 543     BDRVQcowState *s = bs->opaque;
 544     unsigned int len, i;
 545     int ret = 0;
 546     QCowHeader header;
 547     QemuOpts *opts = NULL;
 548     Error *local_err = NULL;
 549     uint64_t ext_end;
 550     uint64_t l1_vm_state_index;
 551     const char *opt_overlap_check, *opt_overlap_check_template;
 552     int overlap_check_template = 0;
 553     uint64_t l2_cache_size, refcount_cache_size;
 554
 555     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
 556     if (ret < 0) {
 557         error_setg_errno(errp, -ret, "Could not read qcow2 header");
 558         goto fail;
 559     }
 560     be32_to_cpus(&header.magic);
 561     be32_to_cpus(&header.version);
 562     be64_to_cpus(&header.backing_file_offset);
 563     be32_to_cpus(&header.backing_file_size);
 564     be64_to_cpus(&header.size);
 565     be32_to_cpus(&header.cluster_bits);
 566     be32_to_cpus(&header.crypt_method);
 567     be64_to_cpus(&header.l1_table_offset);
 568     be32_to_cpus(&header.l1_size);
 569     be64_to_cpus(&header.refcount_table_offset);
 570     be32_to_cpus(&header.refcount_table_clusters);
 571     be64_to_cpus(&header.snapshots_offset);
 572     be32_to_cpus(&header.nb_snapshots);
 573
 574     if (header.magic != QCOW_MAGIC) {
 575         error_setg(errp, "Image is not in qcow2 format");
 576         ret = -EINVAL;
 577         goto fail;
 578     }
 579     if (header.version < 2 || header.version > 3) {
 580         report_unsupported(bs, errp, "QCOW version %" PRIu32, header.version);
 581         ret = -ENOTSUP;
 582         goto fail;
 583     }
 584
 585     s->qcow_version = header.version;
 586
 587     /* Initialise cluster size */
 588     if (header.cluster_bits < MIN_CLUSTER_BITS ||
 589         header.cluster_bits > MAX_CLUSTER_BITS) {
 590         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
 591                    header.cluster_bits);
 592         ret = -EINVAL;
 593         goto fail;
 594     }
 595
 596     s->cluster_bits = header.cluster_bits;
 597     s->cluster_size = 1 << s->cluster_bits;
 598     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 599
 600     /* Initialise version 3 header fields */
 601     if (header.version == 2) {
 602         header.incompatible_features    = 0;
 603         header.compatible_features      = 0;
 604         header.autoclear_features       = 0;
 605         header.refcount_order           = 4;
 606         header.header_length            = 72;
 607     } else {
 608         be64_to_cpus(&header.incompatible_features);
 609         be64_to_cpus(&header.compatible_features);
 610         be64_to_cpus(&header.autoclear_features);
 611         be32_to_cpus(&header.refcount_order);
 612         be32_to_cpus(&header.header_length);
 613
 614         if (header.header_length < 104) {
 615             error_setg(errp, "qcow2 header too short");
 616             ret = -EINVAL;
 617             goto fail;
 618         }
 619     }
 620
 621     if (header.header_length > s->cluster_size) {
 622         error_setg(errp, "qcow2 header exceeds cluster size");
 623         ret = -EINVAL;
 624         goto fail;
 625     }
 626
 627     if (header.header_length > sizeof(header)) {
 628         s->unknown_header_fields_size = header.header_length - sizeof(header);
 629         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
 630         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
 631                          s->unknown_header_fields_size);
 632         if (ret < 0) {
 633             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
 634                              "fields");
 635             goto fail;
 636         }
 637     }
 638
 639     if (header.backing_file_offset > s->cluster_size) {
 640         error_setg(errp, "Invalid backing file offset");
 641         ret = -EINVAL;
 642         goto fail;
 643     }
 644
 645     if (header.backing_file_offset) {
 646         ext_end = header.backing_file_offset;
 647     } else {
 648         ext_end = 1 << header.cluster_bits;
 649     }
 650
 651     /* Handle feature bits */
 652     s->incompatible_features    = header.incompatible_features;
 653     s->compatible_features      = header.compatible_features;
 654     s->autoclear_features       = header.autoclear_features;
 655
 656     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
 657         void *feature_table = NULL;
 658         qcow2_read_extensions(bs, header.header_length, ext_end,
 659                               &feature_table, NULL);
 660         report_unsupported_feature(bs, errp, feature_table,
 661                                    s->incompatible_features &
 662                                    ~QCOW2_INCOMPAT_MASK);
 663         ret = -ENOTSUP;
 664         g_free(feature_table);
 665         goto fail;
 666     }
 667
 668     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
 669         /* Corrupt images may not be written to unless they are being repaired
 670          */
 671         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
 672             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
 673                        "read/write");
 674             ret = -EACCES;
 675             goto fail;
 676         }
 677     }
 678
 679     /* Check support for various header values */
 680     if (header.refcount_order != 4) {
 681         report_unsupported(bs, errp, "%d bit reference counts",
 682                            1 << header.refcount_order);
 683         ret = -ENOTSUP;
 684         goto fail;
 685     }
 686     s->refcount_order = header.refcount_order;
 687
 688     if (header.crypt_method > QCOW_CRYPT_AES) {
 689         error_setg(errp, "Unsupported encryption method: %" PRIu32,
 690                    header.crypt_method);
 691         ret = -EINVAL;
 692         goto fail;
 693     }
 694     s->crypt_method_header = header.crypt_method;
 695     if (s->crypt_method_header) {
 696         bs->encrypted = 1;
 697     }
 698
 699     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
 700     s->l2_size = 1 << s->l2_bits;
 701     bs->total_sectors = header.size / 512;
 702     s->csize_shift = (62 - (s->cluster_bits - 8));
 703     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
 704     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
 705
 706     s->refcount_table_offset = header.refcount_table_offset;
 707     s->refcount_table_size =
 708         header.refcount_table_clusters << (s->cluster_bits - 3);
 709
 710     if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
 711         error_setg(errp, "Reference count table too large");
 712         ret = -EINVAL;
 713         goto fail;
 714     }
 715
 716     ret = validate_table_offset(bs, s->refcount_table_offset,
 717                                 s->refcount_table_size, sizeof(uint64_t));
 718     if (ret < 0) {
 719         error_setg(errp, "Invalid reference count table offset");
 720         goto fail;
 721     }
 722
 723     /* Snapshot table offset/length */
 724     if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
 725         error_setg(errp, "Too many snapshots");
 726         ret = -EINVAL;
 727         goto fail;
 728     }
 729
 730     ret = validate_table_offset(bs, header.snapshots_offset,
 731                                 header.nb_snapshots,
 732                                 sizeof(QCowSnapshotHeader));
 733     if (ret < 0) {
 734         error_setg(errp, "Invalid snapshot table offset");
 735         goto fail;
 736     }
 737
 738     /* read the level 1 table */
 739     if (header.l1_size > QCOW_MAX_L1_SIZE) {
 740         error_setg(errp, "Active L1 table too large");
 741         ret = -EFBIG;
 742         goto fail;
 743     }
 744     s->l1_size = header.l1_size;
 745
 746     l1_vm_state_index = size_to_l1(s, header.size);
 747     if (l1_vm_state_index > INT_MAX) {
 748         error_setg(errp, "Image is too big");
 749         ret = -EFBIG;
 750         goto fail;
 751     }
 752     s->l1_vm_state_index = l1_vm_state_index;
 753
 754     /* the L1 table must contain at least enough entries to put
 755        header.size bytes */
 756     if (s->l1_size < s->l1_vm_state_index) {
 757         error_setg(errp, "L1 table is too small");
 758         ret = -EINVAL;
 759         goto fail;
 760     }
 761
 762     ret = validate_table_offset(bs, header.l1_table_offset,
 763                                 header.l1_size, sizeof(uint64_t));
 764     if (ret < 0) {
 765         error_setg(errp, "Invalid L1 table offset");
 766         goto fail;
 767     }
 768     s->l1_table_offset = header.l1_table_offset;
 769
 770
 771     if (s->l1_size > 0) {
 772         s->l1_table = qemu_try_blockalign(bs->file,
 773             align_offset(s->l1_size * sizeof(uint64_t), 512));
 774         if (s->l1_table == NULL) {
 775             error_setg(errp, "Could not allocate L1 table");
 776             ret = -ENOMEM;
 777             goto fail;
 778         }
 779         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
 780                          s->l1_size * sizeof(uint64_t));
 781         if (ret < 0) {
 782             error_setg_errno(errp, -ret, "Could not read L1 table");
 783             goto fail;
 784         }
 785         for(i = 0;i < s->l1_size; i++) {
 786             be64_to_cpus(&s->l1_table[i]);
 787         }
 788     }
 789
 790     /* get L2 table/refcount block cache size from command line options */
 791     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
 792     qemu_opts_absorb_qdict(opts, options, &local_err);
 793     if (local_err) {
 794         error_propagate(errp, local_err);
 795         ret = -EINVAL;
 796         goto fail;
 797     }
 798
 799     read_cache_sizes(opts, &l2_cache_size, &refcount_cache_size, &local_err);
 800     if (local_err) {
 801         error_propagate(errp, local_err);
 802         ret = -EINVAL;
 803         goto fail;
 804     }
 805
 806     l2_cache_size /= s->cluster_size;
 807     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
 808         l2_cache_size = MIN_L2_CACHE_SIZE;
 809     }
 810     if (l2_cache_size > INT_MAX) {
 811         error_setg(errp, "L2 cache size too big");
 812         ret = -EINVAL;
 813         goto fail;
 814     }
 815
 816     refcount_cache_size /= s->cluster_size;
 817     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
 818         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
 819     }
 820     if (refcount_cache_size > INT_MAX) {
 821         error_setg(errp, "Refcount cache size too big");
 822         ret = -EINVAL;
 823         goto fail;
 824     }
 825
 826     /* alloc L2 table/refcount block cache */
 827     s->l2_table_cache = qcow2_cache_create(bs, l2_cache_size);
 828     s->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size);
 829     if (s->l2_table_cache == NULL || s->refcount_block_cache == NULL) {
 830         error_setg(errp, "Could not allocate metadata caches");
 831         ret = -ENOMEM;
 832         goto fail;
 833     }
 834
 835     s->cluster_cache = g_malloc(s->cluster_size);
 836     /* one more sector for decompressed data alignment */
 837     s->cluster_data = qemu_try_blockalign(bs->file, QCOW_MAX_CRYPT_CLUSTERS
 838                                                     * s->cluster_size + 512);
 839     if (s->cluster_data == NULL) {
 840         error_setg(errp, "Could not allocate temporary cluster buffer");
 841         ret = -ENOMEM;
 842         goto fail;
 843     }
 844
 845     s->cluster_cache_offset = -1;
 846     s->flags = flags;
 847
 848     ret = qcow2_refcount_init(bs);
 849     if (ret != 0) {
 850         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
 851         goto fail;
 852     }
 853
 854     QLIST_INIT(&s->cluster_allocs);
 855     QTAILQ_INIT(&s->discards);
 856
 857     /* read qcow2 extensions */
 858     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
 859         &local_err)) {
 860         error_propagate(errp, local_err);
 861         ret = -EINVAL;
 862         goto fail;
 863     }
 864
 865     /* read the backing file name */
 866     if (header.backing_file_offset != 0) {
 867         len = header.backing_file_size;
 868         if (len > MIN(1023, s->cluster_size - header.backing_file_offset)) {
 869             error_setg(errp, "Backing file name too long");
 870             ret = -EINVAL;
 871             goto fail;
 872         }
 873         ret = bdrv_pread(bs->file, header.backing_file_offset,
 874                          bs->backing_file, len);
 875         if (ret < 0) {
 876             error_setg_errno(errp, -ret, "Could not read backing file name");
 877             goto fail;
 878         }
 879         bs->backing_file[len] = '\0';
 880     }
 881
 882     /* Internal snapshots */
 883     s->snapshots_offset = header.snapshots_offset;
 884     s->nb_snapshots = header.nb_snapshots;
 885
 886     ret = qcow2_read_snapshots(bs);
 887     if (ret < 0) {
 888         error_setg_errno(errp, -ret, "Could not read snapshots");
 889         goto fail;
 890     }
 891
 892     /* Clear unknown autoclear feature bits */
 893     if (!bs->read_only && !(flags & BDRV_O_INCOMING) && s->autoclear_features) {
 894         s->autoclear_features = 0;
 895         ret = qcow2_update_header(bs);
 896         if (ret < 0) {
 897             error_setg_errno(errp, -ret, "Could not update qcow2 header");
 898             goto fail;
 899         }
 900     }
 901
 902     /* Initialise locks */
 903     qemu_co_mutex_init(&s->lock);
 904
 905     /* Repair image if dirty */
 906     if (!(flags & (BDRV_O_CHECK | BDRV_O_INCOMING)) && !bs->read_only &&
 907         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
 908         BdrvCheckResult result = {0};
 909
 910         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS);
 911         if (ret < 0) {
 912             error_setg_errno(errp, -ret, "Could not repair dirty image");
 913             goto fail;
 914         }
 915     }
 916
 917     /* Enable lazy_refcounts according to image and command line options */
 918     s->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
 919         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
 920
 921     s->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
 922     s->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
 923     s->discard_passthrough[QCOW2_DISCARD_REQUEST] =
 924         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
 925                           flags & BDRV_O_UNMAP);
 926     s->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
 927         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
 928     s->discard_passthrough[QCOW2_DISCARD_OTHER] =
 929         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
 930
 931     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
 932     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
 933     if (opt_overlap_check_template && opt_overlap_check &&
 934         strcmp(opt_overlap_check_template, opt_overlap_check))
 935     {
 936         error_setg(errp, "Conflicting values for qcow2 options '"
 937                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
 938                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
 939         ret = -EINVAL;
 940         goto fail;
 941     }
 942     if (!opt_overlap_check) {
 943         opt_overlap_check = opt_overlap_check_template ?: "cached";
 944     }
 945
 946     if (!strcmp(opt_overlap_check, "none")) {
 947         overlap_check_template = 0;
 948     } else if (!strcmp(opt_overlap_check, "constant")) {
 949         overlap_check_template = QCOW2_OL_CONSTANT;
 950     } else if (!strcmp(opt_overlap_check, "cached")) {
 951         overlap_check_template = QCOW2_OL_CACHED;
 952     } else if (!strcmp(opt_overlap_check, "all")) {
 953         overlap_check_template = QCOW2_OL_ALL;
 954     } else {
 955         error_setg(errp, "Unsupported value '%s' for qcow2 option "
 956                    "'overlap-check'. Allowed are either of the following: "
 957                    "none, constant, cached, all", opt_overlap_check);
 958         ret = -EINVAL;
 959         goto fail;
 960     }
 961
 962     s->overlap_check = 0;
 963     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
 964         /* overlap-check defines a template bitmask, but every flag may be
 965          * overwritten through the associated boolean option */
 966         s->overlap_check |=
 967             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
 968                               overlap_check_template & (1 << i)) << i;
 969     }
 970
 971     qemu_opts_del(opts);
 972     opts = NULL;
 973
 974     if (s->use_lazy_refcounts && s->qcow_version < 3) {
 975         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
 976                    "qemu 1.1 compatibility level");
 977         ret = -EINVAL;
 978         goto fail;
 979     }
 980
 981 #ifdef DEBUG_ALLOC
 982     {
 983         BdrvCheckResult result = {0};
 984         qcow2_check_refcounts(bs, &result, 0);
 985     }
 986 #endif
 987     return ret;
 988
 989  fail:
 990     qemu_opts_del(opts);
 991     g_free(s->unknown_header_fields);
 992     cleanup_unknown_header_ext(bs);
 993     qcow2_free_snapshots(bs);
 994     qcow2_refcount_close(bs);
 995     qemu_vfree(s->l1_table);
 996     /* else pre-write overlap checks in cache_destroy may crash */
 997     s->l1_table = NULL;
 998     if (s->l2_table_cache) {
 999         qcow2_cache_destroy(bs, s->l2_table_cache);
1000     }
1001     if (s->refcount_block_cache) {
1002         qcow2_cache_destroy(bs, s->refcount_block_cache);
1003     }
1004     g_free(s->cluster_cache);
1005     qemu_vfree(s->cluster_data);
1006     return ret;
1007 }
1008
1009 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1010 {
1011     BDRVQcowState *s = bs->opaque;
1012
1013     bs->bl.write_zeroes_alignment = s->cluster_sectors;
1014 }
1015
1016 static int qcow2_set_key(BlockDriverState *bs, const char *key)
1017 {
1018     BDRVQcowState *s = bs->opaque;
1019     uint8_t keybuf[16];
1020     int len, i;
1021
1022     memset(keybuf, 0, 16);
1023     len = strlen(key);
1024     if (len > 16)
1025         len = 16;
1026     /* XXX: we could compress the chars to 7 bits to increase
1027        entropy */
1028     for(i = 0;i < len;i++) {
1029         keybuf[i] = key[i];
1030     }
1031     s->crypt_method = s->crypt_method_header;
1032
1033     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
1034         return -1;
1035     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
1036         return -1;
1037 #if 0
1038     /* test */
1039     {
1040         uint8_t in[16];
1041         uint8_t out[16];
1042         uint8_t tmp[16];
1043         for(i=0;i<16;i++)
1044             in[i] = i;
1045         AES_encrypt(in, tmp, &s->aes_encrypt_key);
1046         AES_decrypt(tmp, out, &s->aes_decrypt_key);
1047         for(i = 0; i < 16; i++)
1048             printf(" %02x", tmp[i]);
1049         printf("\n");
1050         for(i = 0; i < 16; i++)
1051             printf(" %02x", out[i]);
1052         printf("\n");
1053     }
1054 #endif
1055     return 0;
1056 }
1057
1058 /* We have no actual commit/abort logic for qcow2, but we need to write out any
1059  * unwritten data if we reopen read-only. */
1060 static int qcow2_reopen_prepare(BDRVReopenState *state,
1061                                 BlockReopenQueue *queue, Error **errp)
1062 {
1063     int ret;
1064
1065     if ((state->flags & BDRV_O_RDWR) == 0) {
1066         ret = bdrv_flush(state->bs);
1067         if (ret < 0) {
1068             return ret;
1069         }
1070
1071         ret = qcow2_mark_clean(state->bs);
1072         if (ret < 0) {
1073             return ret;
1074         }
1075     }
1076
1077     return 0;
1078 }
1079
1080 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1081         int64_t sector_num, int nb_sectors, int *pnum)
1082 {
1083     BDRVQcowState *s = bs->opaque;
1084     uint64_t cluster_offset;
1085     int index_in_cluster, ret;
1086     int64_t status = 0;
1087
1088     *pnum = nb_sectors;
1089     qemu_co_mutex_lock(&s->lock);
1090     ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
1091     qemu_co_mutex_unlock(&s->lock);
1092     if (ret < 0) {
1093         return ret;
1094     }
1095
1096     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1097         !s->crypt_method) {
1098         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1099         cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1100         status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1101     }
1102     if (ret == QCOW2_CLUSTER_ZERO) {
1103         status |= BDRV_BLOCK_ZERO;
1104     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1105         status |= BDRV_BLOCK_DATA;
1106     }
1107     return status;
1108 }
1109
1110 /* handle reading after the end of the backing file */
1111 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
1112                   int64_t sector_num, int nb_sectors)
1113 {
1114     int n1;
1115     if ((sector_num + nb_sectors) <= bs->total_sectors)
1116         return nb_sectors;
1117     if (sector_num >= bs->total_sectors)
1118         n1 = 0;
1119     else
1120         n1 = bs->total_sectors - sector_num;
1121
1122     qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1));
1123
1124     return n1;
1125 }
1126
1127 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
1128                           int remaining_sectors, QEMUIOVector *qiov)
1129 {
1130     BDRVQcowState *s = bs->opaque;
1131     int index_in_cluster, n1;
1132     int ret;
1133     int cur_nr_sectors; /* number of sectors in current iteration */
1134     uint64_t cluster_offset = 0;
1135     uint64_t bytes_done = 0;
1136     QEMUIOVector hd_qiov;
1137     uint8_t *cluster_data = NULL;
1138
1139     qemu_iovec_init(&hd_qiov, qiov->niov);
1140
1141     qemu_co_mutex_lock(&s->lock);
1142
1143     while (remaining_sectors != 0) {
1144
1145         /* prepare next request */
1146         cur_nr_sectors = remaining_sectors;
1147         if (s->crypt_method) {
1148             cur_nr_sectors = MIN(cur_nr_sectors,
1149                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1150         }
1151
1152         ret = qcow2_get_cluster_offset(bs, sector_num << 9,
1153             &cur_nr_sectors, &cluster_offset);
1154         if (ret < 0) {
1155             goto fail;
1156         }
1157
1158         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1159
1160         qemu_iovec_reset(&hd_qiov);
1161         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1162             cur_nr_sectors * 512);
1163
1164         switch (ret) {
1165         case QCOW2_CLUSTER_UNALLOCATED:
1166
1167             if (bs->backing_hd) {
1168                 /* read from the base image */
1169                 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov,
1170                     sector_num, cur_nr_sectors);
1171                 if (n1 > 0) {
1172                     QEMUIOVector local_qiov;
1173
1174                     qemu_iovec_init(&local_qiov, hd_qiov.niov);
1175                     qemu_iovec_concat(&local_qiov, &hd_qiov, 0,
1176                                       n1 * BDRV_SECTOR_SIZE);
1177
1178                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1179                     qemu_co_mutex_unlock(&s->lock);
1180                     ret = bdrv_co_readv(bs->backing_hd, sector_num,
1181                                         n1, &local_qiov);
1182                     qemu_co_mutex_lock(&s->lock);
1183
1184                     qemu_iovec_destroy(&local_qiov);
1185
1186                     if (ret < 0) {
1187                         goto fail;
1188                     }
1189                 }
1190             } else {
1191                 /* Note: in this case, no need to wait */
1192                 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1193             }
1194             break;
1195
1196         case QCOW2_CLUSTER_ZERO:
1197             qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1198             break;
1199
1200         case QCOW2_CLUSTER_COMPRESSED:
1201             /* add AIO support for compressed blocks ? */
1202             ret = qcow2_decompress_cluster(bs, cluster_offset);
1203             if (ret < 0) {
1204                 goto fail;
1205             }
1206
1207             qemu_iovec_from_buf(&hd_qiov, 0,
1208                 s->cluster_cache + index_in_cluster * 512,
1209                 512 * cur_nr_sectors);
1210             break;
1211
1212         case QCOW2_CLUSTER_NORMAL:
1213             if ((cluster_offset & 511) != 0) {
1214                 ret = -EIO;
1215                 goto fail;
1216             }
1217
1218             if (s->crypt_method) {
1219                 /*
1220                  * For encrypted images, read everything into a temporary
1221                  * contiguous buffer on which the AES functions can work.
1222                  */
1223                 if (!cluster_data) {
1224                     cluster_data =
1225                         qemu_try_blockalign(bs->file, QCOW_MAX_CRYPT_CLUSTERS
1226                                                       * s->cluster_size);
1227                     if (cluster_data == NULL) {
1228                         ret = -ENOMEM;
1229                         goto fail;
1230                     }
1231                 }
1232
1233                 assert(cur_nr_sectors <=
1234                     QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1235                 qemu_iovec_reset(&hd_qiov);
1236                 qemu_iovec_add(&hd_qiov, cluster_data,
1237                     512 * cur_nr_sectors);
1238             }
1239
1240             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1241             qemu_co_mutex_unlock(&s->lock);
1242             ret = bdrv_co_readv(bs->file,
1243                                 (cluster_offset >> 9) + index_in_cluster,
1244                                 cur_nr_sectors, &hd_qiov);
1245             qemu_co_mutex_lock(&s->lock);
1246             if (ret < 0) {
1247                 goto fail;
1248             }
1249             if (s->crypt_method) {
1250                 qcow2_encrypt_sectors(s, sector_num,  cluster_data,
1251                     cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
1252                 qemu_iovec_from_buf(qiov, bytes_done,
1253                     cluster_data, 512 * cur_nr_sectors);
1254             }
1255             break;
1256
1257         default:
1258             g_assert_not_reached();
1259             ret = -EIO;
1260             goto fail;
1261         }
1262
1263         remaining_sectors -= cur_nr_sectors;
1264         sector_num += cur_nr_sectors;
1265         bytes_done += cur_nr_sectors * 512;
1266     }
1267     ret = 0;
1268
1269 fail:
1270     qemu_co_mutex_unlock(&s->lock);
1271
1272     qemu_iovec_destroy(&hd_qiov);
1273     qemu_vfree(cluster_data);
1274
1275     return ret;
1276 }
1277
1278 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
1279                            int64_t sector_num,
1280                            int remaining_sectors,
1281                            QEMUIOVector *qiov)
1282 {
1283     BDRVQcowState *s = bs->opaque;
1284     int index_in_cluster;
1285     int ret;
1286     int cur_nr_sectors; /* number of sectors in current iteration */
1287     uint64_t cluster_offset;
1288     QEMUIOVector hd_qiov;
1289     uint64_t bytes_done = 0;
1290     uint8_t *cluster_data = NULL;
1291     QCowL2Meta *l2meta = NULL;
1292
1293     trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
1294                                  remaining_sectors);
1295
1296     qemu_iovec_init(&hd_qiov, qiov->niov);
1297
1298     s->cluster_cache_offset = -1; /* disable compressed cache */
1299
1300     qemu_co_mutex_lock(&s->lock);
1301
1302     while (remaining_sectors != 0) {
1303
1304         l2meta = NULL;
1305
1306         trace_qcow2_writev_start_part(qemu_coroutine_self());
1307         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1308         cur_nr_sectors = remaining_sectors;
1309         if (s->crypt_method &&
1310             cur_nr_sectors >
1311             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster) {
1312             cur_nr_sectors =
1313                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster;
1314         }
1315
1316         ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
1317             &cur_nr_sectors, &cluster_offset, &l2meta);
1318         if (ret < 0) {
1319             goto fail;
1320         }
1321
1322         assert((cluster_offset & 511) == 0);
1323
1324         qemu_iovec_reset(&hd_qiov);
1325         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1326             cur_nr_sectors * 512);
1327
1328         if (s->crypt_method) {
1329             if (!cluster_data) {
1330                 cluster_data = qemu_try_blockalign(bs->file,
1331                                                    QCOW_MAX_CRYPT_CLUSTERS
1332                                                    * s->cluster_size);
1333                 if (cluster_data == NULL) {
1334                     ret = -ENOMEM;
1335                     goto fail;
1336                 }
1337             }
1338
1339             assert(hd_qiov.size <=
1340                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1341             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1342
1343             qcow2_encrypt_sectors(s, sector_num, cluster_data,
1344                 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
1345
1346             qemu_iovec_reset(&hd_qiov);
1347             qemu_iovec_add(&hd_qiov, cluster_data,
1348                 cur_nr_sectors * 512);
1349         }
1350
1351         ret = qcow2_pre_write_overlap_check(bs, 0,
1352                 cluster_offset + index_in_cluster * BDRV_SECTOR_SIZE,
1353                 cur_nr_sectors * BDRV_SECTOR_SIZE);
1354         if (ret < 0) {
1355             goto fail;
1356         }
1357
1358         qemu_co_mutex_unlock(&s->lock);
1359         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1360         trace_qcow2_writev_data(qemu_coroutine_self(),
1361                                 (cluster_offset >> 9) + index_in_cluster);
1362         ret = bdrv_co_writev(bs->file,
1363                              (cluster_offset >> 9) + index_in_cluster,
1364                              cur_nr_sectors, &hd_qiov);
1365         qemu_co_mutex_lock(&s->lock);
1366         if (ret < 0) {
1367             goto fail;
1368         }
1369
1370         while (l2meta != NULL) {
1371             QCowL2Meta *next;
1372
1373             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1374             if (ret < 0) {
1375                 goto fail;
1376             }
1377
1378             /* Take the request off the list of running requests */
1379             if (l2meta->nb_clusters != 0) {
1380                 QLIST_REMOVE(l2meta, next_in_flight);
1381             }
1382
1383             qemu_co_queue_restart_all(&l2meta->dependent_requests);
1384
1385             next = l2meta->next;
1386             g_free(l2meta);
1387             l2meta = next;
1388         }
1389
1390         remaining_sectors -= cur_nr_sectors;
1391         sector_num += cur_nr_sectors;
1392         bytes_done += cur_nr_sectors * 512;
1393         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
1394     }
1395     ret = 0;
1396
1397 fail:
1398     qemu_co_mutex_unlock(&s->lock);
1399
1400     while (l2meta != NULL) {
1401         QCowL2Meta *next;
1402
1403         if (l2meta->nb_clusters != 0) {
1404             QLIST_REMOVE(l2meta, next_in_flight);
1405         }
1406         qemu_co_queue_restart_all(&l2meta->dependent_requests);
1407
1408         next = l2meta->next;
1409         g_free(l2meta);
1410         l2meta = next;
1411     }
1412
1413     qemu_iovec_destroy(&hd_qiov);
1414     qemu_vfree(cluster_data);
1415     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
1416
1417     return ret;
1418 }
1419
1420 static void qcow2_close(BlockDriverState *bs)
1421 {
1422     BDRVQcowState *s = bs->opaque;
1423     qemu_vfree(s->l1_table);
1424     /* else pre-write overlap checks in cache_destroy may crash */
1425     s->l1_table = NULL;
1426
1427     if (!(bs->open_flags & BDRV_O_INCOMING)) {
1428         qcow2_cache_flush(bs, s->l2_table_cache);
1429         qcow2_cache_flush(bs, s->refcount_block_cache);
1430
1431         qcow2_mark_clean(bs);
1432     }
1433
1434     qcow2_cache_destroy(bs, s->l2_table_cache);
1435     qcow2_cache_destroy(bs, s->refcount_block_cache);
1436
1437     g_free(s->unknown_header_fields);
1438     cleanup_unknown_header_ext(bs);
1439
1440     g_free(s->cluster_cache);
1441     qemu_vfree(s->cluster_data);
1442     qcow2_refcount_close(bs);
1443     qcow2_free_snapshots(bs);
1444 }
1445
1446 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
1447 {
1448     BDRVQcowState *s = bs->opaque;
1449     int flags = s->flags;
1450     AES_KEY aes_encrypt_key;
1451     AES_KEY aes_decrypt_key;
1452     uint32_t crypt_method = 0;
1453     QDict *options;
1454     Error *local_err = NULL;
1455     int ret;
1456
1457     /*
1458      * Backing files are read-only which makes all of their metadata immutable,
1459      * that means we don't have to worry about reopening them here.
1460      */
1461
1462     if (s->crypt_method) {
1463         crypt_method = s->crypt_method;
1464         memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key));
1465         memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key));
1466     }
1467
1468     qcow2_close(bs);
1469
1470     bdrv_invalidate_cache(bs->file, &local_err);
1471     if (local_err) {
1472         error_propagate(errp, local_err);
1473         return;
1474     }
1475
1476     memset(s, 0, sizeof(BDRVQcowState));
1477     options = qdict_clone_shallow(bs->options);
1478
1479     ret = qcow2_open(bs, options, flags, &local_err);
1480     QDECREF(options);
1481     if (local_err) {
1482         error_setg(errp, "Could not reopen qcow2 layer: %s",
1483                    error_get_pretty(local_err));
1484         error_free(local_err);
1485         return;
1486     } else if (ret < 0) {
1487         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
1488         return;
1489     }
1490
1491     if (crypt_method) {
1492         s->crypt_method = crypt_method;
1493         memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key));
1494         memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key));
1495     }
1496 }
1497
1498 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
1499     size_t len, size_t buflen)
1500 {
1501     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
1502     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
1503
1504     if (buflen < ext_len) {
1505         return -ENOSPC;
1506     }
1507
1508     *ext_backing_fmt = (QCowExtension) {
1509         .magic  = cpu_to_be32(magic),
1510         .len    = cpu_to_be32(len),
1511     };
1512     memcpy(buf + sizeof(QCowExtension), s, len);
1513
1514     return ext_len;
1515 }
1516
1517 /*
1518  * Updates the qcow2 header, including the variable length parts of it, i.e.
1519  * the backing file name and all extensions. qcow2 was not designed to allow
1520  * such changes, so if we run out of space (we can only use the first cluster)
1521  * this function may fail.
1522  *
1523  * Returns 0 on success, -errno in error cases.
1524  */
1525 int qcow2_update_header(BlockDriverState *bs)
1526 {
1527     BDRVQcowState *s = bs->opaque;
1528     QCowHeader *header;
1529     char *buf;
1530     size_t buflen = s->cluster_size;
1531     int ret;
1532     uint64_t total_size;
1533     uint32_t refcount_table_clusters;
1534     size_t header_length;
1535     Qcow2UnknownHeaderExtension *uext;
1536
1537     buf = qemu_blockalign(bs, buflen);
1538
1539     /* Header structure */
1540     header = (QCowHeader*) buf;
1541
1542     if (buflen < sizeof(*header)) {
1543         ret = -ENOSPC;
1544         goto fail;
1545     }
1546
1547     header_length = sizeof(*header) + s->unknown_header_fields_size;
1548     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1549     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1550
1551     *header = (QCowHeader) {
1552         /* Version 2 fields */
1553         .magic                  = cpu_to_be32(QCOW_MAGIC),
1554         .version                = cpu_to_be32(s->qcow_version),
1555         .backing_file_offset    = 0,
1556         .backing_file_size      = 0,
1557         .cluster_bits           = cpu_to_be32(s->cluster_bits),
1558         .size                   = cpu_to_be64(total_size),
1559         .crypt_method           = cpu_to_be32(s->crypt_method_header),
1560         .l1_size                = cpu_to_be32(s->l1_size),
1561         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
1562         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
1563         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1564         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
1565         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
1566
1567         /* Version 3 fields */
1568         .incompatible_features  = cpu_to_be64(s->incompatible_features),
1569         .compatible_features    = cpu_to_be64(s->compatible_features),
1570         .autoclear_features     = cpu_to_be64(s->autoclear_features),
1571         .refcount_order         = cpu_to_be32(s->refcount_order),
1572         .header_length          = cpu_to_be32(header_length),
1573     };
1574
1575     /* For older versions, write a shorter header */
1576     switch (s->qcow_version) {
1577     case 2:
1578         ret = offsetof(QCowHeader, incompatible_features);
1579         break;
1580     case 3:
1581         ret = sizeof(*header);
1582         break;
1583     default:
1584         ret = -EINVAL;
1585         goto fail;
1586     }
1587
1588     buf += ret;
1589     buflen -= ret;
1590     memset(buf, 0, buflen);
1591
1592     /* Preserve any unknown field in the header */
1593     if (s->unknown_header_fields_size) {
1594         if (buflen < s->unknown_header_fields_size) {
1595             ret = -ENOSPC;
1596             goto fail;
1597         }
1598
1599         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1600         buf += s->unknown_header_fields_size;
1601         buflen -= s->unknown_header_fields_size;
1602     }
1603
1604     /* Backing file format header extension */
1605     if (*bs->backing_format) {
1606         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1607                              bs->backing_format, strlen(bs->backing_format),
1608                              buflen);
1609         if (ret < 0) {
1610             goto fail;
1611         }
1612
1613         buf += ret;
1614         buflen -= ret;
1615     }
1616
1617     /* Feature table */
1618     Qcow2Feature features[] = {
1619         {
1620             .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1621             .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
1622             .name = "dirty bit",
1623         },
1624         {
1625             .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1626             .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
1627             .name = "corrupt bit",
1628         },
1629         {
1630             .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1631             .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1632             .name = "lazy refcounts",
1633         },
1634     };
1635
1636     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1637                          features, sizeof(features), buflen);
1638     if (ret < 0) {
1639         goto fail;
1640     }
1641     buf += ret;
1642     buflen -= ret;
1643
1644     /* Keep unknown header extensions */
1645     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1646         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1647         if (ret < 0) {
1648             goto fail;
1649         }
1650
1651         buf += ret;
1652         buflen -= ret;
1653     }
1654
1655     /* End of header extensions */
1656     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1657     if (ret < 0) {
1658         goto fail;
1659     }
1660
1661     buf += ret;
1662     buflen -= ret;
1663
1664     /* Backing file name */
1665     if (*bs->backing_file) {
1666         size_t backing_file_len = strlen(bs->backing_file);
1667
1668         if (buflen < backing_file_len) {
1669             ret = -ENOSPC;
1670             goto fail;
1671         }
1672
1673         /* Using strncpy is ok here, since buf is not NUL-terminated. */
1674         strncpy(buf, bs->backing_file, buflen);
1675
1676         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1677         header->backing_file_size   = cpu_to_be32(backing_file_len);
1678     }
1679
1680     /* Write the new header */
1681     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
1682     if (ret < 0) {
1683         goto fail;
1684     }
1685
1686     ret = 0;
1687 fail:
1688     qemu_vfree(header);
1689     return ret;
1690 }
1691
1692 static int qcow2_change_backing_file(BlockDriverState *bs,
1693     const char *backing_file, const char *backing_fmt)
1694 {
1695     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1696     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1697
1698     return qcow2_update_header(bs);
1699 }
1700
1701 static int preallocate(BlockDriverState *bs)
1702 {
1703     uint64_t nb_sectors;
1704     uint64_t offset;
1705     uint64_t host_offset = 0;
1706     int num;
1707     int ret;
1708     QCowL2Meta *meta;
1709
1710     nb_sectors = bdrv_nb_sectors(bs);
1711     offset = 0;
1712
1713     while (nb_sectors) {
1714         num = MIN(nb_sectors, INT_MAX >> BDRV_SECTOR_BITS);
1715         ret = qcow2_alloc_cluster_offset(bs, offset, &num,
1716                                          &host_offset, &meta);
1717         if (ret < 0) {
1718             return ret;
1719         }
1720
1721         while (meta) {
1722             QCowL2Meta *next = meta->next;
1723
1724             ret = qcow2_alloc_cluster_link_l2(bs, meta);
1725             if (ret < 0) {
1726                 qcow2_free_any_clusters(bs, meta->alloc_offset,
1727                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
1728                 return ret;
1729             }
1730
1731             /* There are no dependent requests, but we need to remove our
1732              * request from the list of in-flight requests */
1733             QLIST_REMOVE(meta, next_in_flight);
1734
1735             g_free(meta);
1736             meta = next;
1737         }
1738
1739         /* TODO Preallocate data if requested */
1740
1741         nb_sectors -= num;
1742         offset += num << BDRV_SECTOR_BITS;
1743     }
1744
1745     /*
1746      * It is expected that the image file is large enough to actually contain
1747      * all of the allocated clusters (otherwise we get failing reads after
1748      * EOF). Extend the image to the last allocated sector.
1749      */
1750     if (host_offset != 0) {
1751         uint8_t buf[BDRV_SECTOR_SIZE];
1752         memset(buf, 0, BDRV_SECTOR_SIZE);
1753         ret = bdrv_write(bs->file, (host_offset >> BDRV_SECTOR_BITS) + num - 1,
1754                          buf, 1);
1755         if (ret < 0) {
1756             return ret;
1757         }
1758     }
1759
1760     return 0;
1761 }
1762
1763 static int qcow2_create2(const char *filename, int64_t total_size,
1764                          const char *backing_file, const char *backing_format,
1765                          int flags, size_t cluster_size, PreallocMode prealloc,
1766                          QemuOpts *opts, int version,
1767                          Error **errp)
1768 {
1769     /* Calculate cluster_bits */
1770     int cluster_bits;
1771     cluster_bits = ffs(cluster_size) - 1;
1772     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
1773         (1 << cluster_bits) != cluster_size)
1774     {
1775         error_setg(errp, "Cluster size must be a power of two between %d and "
1776                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
1777         return -EINVAL;
1778     }
1779
1780     /*
1781      * Open the image file and write a minimal qcow2 header.
1782      *
1783      * We keep things simple and start with a zero-sized image. We also
1784      * do without refcount blocks or a L1 table for now. We'll fix the
1785      * inconsistency later.
1786      *
1787      * We do need a refcount table because growing the refcount table means
1788      * allocating two new refcount blocks - the seconds of which would be at
1789      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
1790      * size for any qcow2 image.
1791      */
1792     BlockDriverState* bs;
1793     QCowHeader *header;
1794     uint64_t* refcount_table;
1795     Error *local_err = NULL;
1796     int ret;
1797
1798     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
1799         int64_t meta_size = 0;
1800         uint64_t nreftablee, nrefblocke, nl1e, nl2e;
1801         int64_t aligned_total_size = align_offset(total_size, cluster_size);
1802
1803         /* header: 1 cluster */
1804         meta_size += cluster_size;
1805
1806         /* total size of L2 tables */
1807         nl2e = aligned_total_size / cluster_size;
1808         nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
1809         meta_size += nl2e * sizeof(uint64_t);
1810
1811         /* total size of L1 tables */
1812         nl1e = nl2e * sizeof(uint64_t) / cluster_size;
1813         nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
1814         meta_size += nl1e * sizeof(uint64_t);
1815
1816         /* total size of refcount blocks
1817          *
1818          * note: every host cluster is reference-counted, including metadata
1819          * (even refcount blocks are recursively included).
1820          * Let:
1821          *   a = total_size (this is the guest disk size)
1822          *   m = meta size not including refcount blocks and refcount tables
1823          *   c = cluster size
1824          *   y1 = number of refcount blocks entries
1825          *   y2 = meta size including everything
1826          * then,
1827          *   y1 = (y2 + a)/c
1828          *   y2 = y1 * sizeof(u16) + y1 * sizeof(u16) * sizeof(u64) / c + m
1829          * we can get y1:
1830          *   y1 = (a + m) / (c - sizeof(u16) - sizeof(u16) * sizeof(u64) / c)
1831          */
1832         nrefblocke = (aligned_total_size + meta_size + cluster_size) /
1833             (cluster_size - sizeof(uint16_t) -
1834              1.0 * sizeof(uint16_t) * sizeof(uint64_t) / cluster_size);
1835         nrefblocke = align_offset(nrefblocke, cluster_size / sizeof(uint16_t));
1836         meta_size += nrefblocke * sizeof(uint16_t);
1837
1838         /* total size of refcount tables */
1839         nreftablee = nrefblocke * sizeof(uint16_t) / cluster_size;
1840         nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t));
1841         meta_size += nreftablee * sizeof(uint64_t);
1842
1843         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
1844                             aligned_total_size + meta_size);
1845         qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc]);
1846     }
1847
1848     ret = bdrv_create_file(filename, opts, &local_err);
1849     if (ret < 0) {
1850         error_propagate(errp, local_err);
1851         return ret;
1852     }
1853
1854     bs = NULL;
1855     ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1856                     NULL, &local_err);
1857     if (ret < 0) {
1858         error_propagate(errp, local_err);
1859         return ret;
1860     }
1861
1862     /* Write the header */
1863     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
1864     header = g_malloc0(cluster_size);
1865     *header = (QCowHeader) {
1866         .magic                      = cpu_to_be32(QCOW_MAGIC),
1867         .version                    = cpu_to_be32(version),
1868         .cluster_bits               = cpu_to_be32(cluster_bits),
1869         .size                       = cpu_to_be64(0),
1870         .l1_table_offset            = cpu_to_be64(0),
1871         .l1_size                    = cpu_to_be32(0),
1872         .refcount_table_offset      = cpu_to_be64(cluster_size),
1873         .refcount_table_clusters    = cpu_to_be32(1),
1874         .refcount_order             = cpu_to_be32(3 + REFCOUNT_SHIFT),
1875         .header_length              = cpu_to_be32(sizeof(*header)),
1876     };
1877
1878     if (flags & BLOCK_FLAG_ENCRYPT) {
1879         header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1880     } else {
1881         header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1882     }
1883
1884     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
1885         header->compatible_features |=
1886             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
1887     }
1888
1889     ret = bdrv_pwrite(bs, 0, header, cluster_size);
1890     g_free(header);
1891     if (ret < 0) {
1892         error_setg_errno(errp, -ret, "Could not write qcow2 header");
1893         goto out;
1894     }
1895
1896     /* Write a refcount table with one refcount block */
1897     refcount_table = g_malloc0(2 * cluster_size);
1898     refcount_table[0] = cpu_to_be64(2 * cluster_size);
1899     ret = bdrv_pwrite(bs, cluster_size, refcount_table, 2 * cluster_size);
1900     g_free(refcount_table);
1901
1902     if (ret < 0) {
1903         error_setg_errno(errp, -ret, "Could not write refcount table");
1904         goto out;
1905     }
1906
1907     bdrv_unref(bs);
1908     bs = NULL;
1909
1910     /*
1911      * And now open the image and make it consistent first (i.e. increase the
1912      * refcount of the cluster that is occupied by the header and the refcount
1913      * table)
1914      */
1915     BlockDriver* drv = bdrv_find_format("qcow2");
1916     assert(drv != NULL);
1917     ret = bdrv_open(&bs, filename, NULL, NULL,
1918         BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv, &local_err);
1919     if (ret < 0) {
1920         error_propagate(errp, local_err);
1921         goto out;
1922     }
1923
1924     ret = qcow2_alloc_clusters(bs, 3 * cluster_size);
1925     if (ret < 0) {
1926         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
1927                          "header and refcount table");
1928         goto out;
1929
1930     } else if (ret != 0) {
1931         error_report("Huh, first cluster in empty image is already in use?");
1932         abort();
1933     }
1934
1935     /* Okay, now that we have a valid image, let's give it the right size */
1936     ret = bdrv_truncate(bs, total_size);
1937     if (ret < 0) {
1938         error_setg_errno(errp, -ret, "Could not resize image");
1939         goto out;
1940     }
1941
1942     /* Want a backing file? There you go.*/
1943     if (backing_file) {
1944         ret = bdrv_change_backing_file(bs, backing_file, backing_format);
1945         if (ret < 0) {
1946             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
1947                              "with format '%s'", backing_file, backing_format);
1948             goto out;
1949         }
1950     }
1951
1952     /* And if we're supposed to preallocate metadata, do that now */
1953     if (prealloc != PREALLOC_MODE_OFF) {
1954         BDRVQcowState *s = bs->opaque;
1955         qemu_co_mutex_lock(&s->lock);
1956         ret = preallocate(bs);
1957         qemu_co_mutex_unlock(&s->lock);
1958         if (ret < 0) {
1959             error_setg_errno(errp, -ret, "Could not preallocate metadata");
1960             goto out;
1961         }
1962     }
1963
1964     bdrv_unref(bs);
1965     bs = NULL;
1966
1967     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
1968     ret = bdrv_open(&bs, filename, NULL, NULL,
1969                     BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_BACKING,
1970                     drv, &local_err);
1971     if (local_err) {
1972         error_propagate(errp, local_err);
1973         goto out;
1974     }
1975
1976     ret = 0;
1977 out:
1978     if (bs) {
1979         bdrv_unref(bs);
1980     }
1981     return ret;
1982 }
1983
1984 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
1985 {
1986     char *backing_file = NULL;
1987     char *backing_fmt = NULL;
1988     char *buf = NULL;
1989     uint64_t size = 0;
1990     int flags = 0;
1991     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
1992     PreallocMode prealloc;
1993     int version = 3;
1994     Error *local_err = NULL;
1995     int ret;
1996
1997     /* Read out options */
1998     size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1999                     BDRV_SECTOR_SIZE);
2000     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2001     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2002     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2003         flags |= BLOCK_FLAG_ENCRYPT;
2004     }
2005     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2006                                          DEFAULT_CLUSTER_SIZE);
2007     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2008     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
2009                                PREALLOC_MODE_MAX, PREALLOC_MODE_OFF,
2010                                &local_err);
2011     if (local_err) {
2012         error_propagate(errp, local_err);
2013         ret = -EINVAL;
2014         goto finish;
2015     }
2016     g_free(buf);
2017     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2018     if (!buf) {
2019         /* keep the default */
2020     } else if (!strcmp(buf, "0.10")) {
2021         version = 2;
2022     } else if (!strcmp(buf, "1.1")) {
2023         version = 3;
2024     } else {
2025         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2026         ret = -EINVAL;
2027         goto finish;
2028     }
2029
2030     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2031         flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2032     }
2033
2034     if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2035         error_setg(errp, "Backing file and preallocation cannot be used at "
2036                    "the same time");
2037         ret = -EINVAL;
2038         goto finish;
2039     }
2040
2041     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2042         error_setg(errp, "Lazy refcounts only supported with compatibility "
2043                    "level 1.1 and above (use compat=1.1 or greater)");
2044         ret = -EINVAL;
2045         goto finish;
2046     }
2047
2048     ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2049                         cluster_size, prealloc, opts, version, &local_err);
2050     if (local_err) {
2051         error_propagate(errp, local_err);
2052     }
2053
2054 finish:
2055     g_free(backing_file);
2056     g_free(backing_fmt);
2057     g_free(buf);
2058     return ret;
2059 }
2060
2061 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
2062     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2063 {
2064     int ret;
2065     BDRVQcowState *s = bs->opaque;
2066
2067     /* Emulate misaligned zero writes */
2068     if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
2069         return -ENOTSUP;
2070     }
2071
2072     /* Whatever is left can use real zero clusters */
2073     qemu_co_mutex_lock(&s->lock);
2074     ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2075         nb_sectors);
2076     qemu_co_mutex_unlock(&s->lock);
2077
2078     return ret;
2079 }
2080
2081 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
2082     int64_t sector_num, int nb_sectors)
2083 {
2084     int ret;
2085     BDRVQcowState *s = bs->opaque;
2086
2087     qemu_co_mutex_lock(&s->lock);
2088     ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2089         nb_sectors, QCOW2_DISCARD_REQUEST);
2090     qemu_co_mutex_unlock(&s->lock);
2091     return ret;
2092 }
2093
2094 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
2095 {
2096     BDRVQcowState *s = bs->opaque;
2097     int64_t new_l1_size;
2098     int ret;
2099
2100     if (offset & 511) {
2101         error_report("The new size must be a multiple of 512");
2102         return -EINVAL;
2103     }
2104
2105     /* cannot proceed if image has snapshots */
2106     if (s->nb_snapshots) {
2107         error_report("Can't resize an image which has snapshots");
2108         return -ENOTSUP;
2109     }
2110
2111     /* shrinking is currently not supported */
2112     if (offset < bs->total_sectors * 512) {
2113         error_report("qcow2 doesn't support shrinking images yet");
2114         return -ENOTSUP;
2115     }
2116
2117     new_l1_size = size_to_l1(s, offset);
2118     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
2119     if (ret < 0) {
2120         return ret;
2121     }
2122
2123     /* write updated header.size */
2124     offset = cpu_to_be64(offset);
2125     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
2126                            &offset, sizeof(uint64_t));
2127     if (ret < 0) {
2128         return ret;
2129     }
2130
2131     s->l1_vm_state_index = new_l1_size;
2132     return 0;
2133 }
2134
2135 /* XXX: put compressed sectors first, then all the cluster aligned
2136    tables to avoid losing bytes in alignment */
2137 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
2138                                   const uint8_t *buf, int nb_sectors)
2139 {
2140     BDRVQcowState *s = bs->opaque;
2141     z_stream strm;
2142     int ret, out_len;
2143     uint8_t *out_buf;
2144     uint64_t cluster_offset;
2145
2146     if (nb_sectors == 0) {
2147         /* align end of file to a sector boundary to ease reading with
2148            sector based I/Os */
2149         cluster_offset = bdrv_getlength(bs->file);
2150         bdrv_truncate(bs->file, cluster_offset);
2151         return 0;
2152     }
2153
2154     if (nb_sectors != s->cluster_sectors) {
2155         ret = -EINVAL;
2156
2157         /* Zero-pad last write if image size is not cluster aligned */
2158         if (sector_num + nb_sectors == bs->total_sectors &&
2159             nb_sectors < s->cluster_sectors) {
2160             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
2161             memset(pad_buf, 0, s->cluster_size);
2162             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
2163             ret = qcow2_write_compressed(bs, sector_num,
2164                                          pad_buf, s->cluster_sectors);
2165             qemu_vfree(pad_buf);
2166         }
2167         return ret;
2168     }
2169
2170     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
2171
2172     /* best compression, small window, no zlib header */
2173     memset(&strm, 0, sizeof(strm));
2174     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
2175                        Z_DEFLATED, -12,
2176                        9, Z_DEFAULT_STRATEGY);
2177     if (ret != 0) {
2178         ret = -EINVAL;
2179         goto fail;
2180     }
2181
2182     strm.avail_in = s->cluster_size;
2183     strm.next_in = (uint8_t *)buf;
2184     strm.avail_out = s->cluster_size;
2185     strm.next_out = out_buf;
2186
2187     ret = deflate(&strm, Z_FINISH);
2188     if (ret != Z_STREAM_END && ret != Z_OK) {
2189         deflateEnd(&strm);
2190         ret = -EINVAL;
2191         goto fail;
2192     }
2193     out_len = strm.next_out - out_buf;
2194
2195     deflateEnd(&strm);
2196
2197     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
2198         /* could not compress: write normal cluster */
2199         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
2200         if (ret < 0) {
2201             goto fail;
2202         }
2203     } else {
2204         cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
2205             sector_num << 9, out_len);
2206         if (!cluster_offset) {
2207             ret = -EIO;
2208             goto fail;
2209         }
2210         cluster_offset &= s->cluster_offset_mask;
2211
2212         ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
2213         if (ret < 0) {
2214             goto fail;
2215         }
2216
2217         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
2218         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
2219         if (ret < 0) {
2220             goto fail;
2221         }
2222     }
2223
2224     ret = 0;
2225 fail:
2226     g_free(out_buf);
2227     return ret;
2228 }
2229
2230 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
2231 {
2232     BDRVQcowState *s = bs->opaque;
2233     int ret;
2234
2235     qemu_co_mutex_lock(&s->lock);
2236     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2237     if (ret < 0) {
2238         qemu_co_mutex_unlock(&s->lock);
2239         return ret;
2240     }
2241
2242     if (qcow2_need_accurate_refcounts(s)) {
2243         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2244         if (ret < 0) {
2245             qemu_co_mutex_unlock(&s->lock);
2246             return ret;
2247         }
2248     }
2249     qemu_co_mutex_unlock(&s->lock);
2250
2251     return 0;
2252 }
2253
2254 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2255 {
2256     BDRVQcowState *s = bs->opaque;
2257     bdi->unallocated_blocks_are_zero = true;
2258     bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
2259     bdi->cluster_size = s->cluster_size;
2260     bdi->vm_state_offset = qcow2_vm_state_offset(s);
2261     return 0;
2262 }
2263
2264 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
2265 {
2266     BDRVQcowState *s = bs->opaque;
2267     ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
2268
2269     *spec_info = (ImageInfoSpecific){
2270         .kind  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
2271         {
2272             .qcow2 = g_new(ImageInfoSpecificQCow2, 1),
2273         },
2274     };
2275     if (s->qcow_version == 2) {
2276         *spec_info->qcow2 = (ImageInfoSpecificQCow2){
2277             .compat = g_strdup("0.10"),
2278         };
2279     } else if (s->qcow_version == 3) {
2280         *spec_info->qcow2 = (ImageInfoSpecificQCow2){
2281             .compat             = g_strdup("1.1"),
2282             .lazy_refcounts     = s->compatible_features &
2283                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
2284             .has_lazy_refcounts = true,
2285             .corrupt            = s->incompatible_features &
2286                                   QCOW2_INCOMPAT_CORRUPT,
2287             .has_corrupt        = true,
2288         };
2289     }
2290
2291     return spec_info;
2292 }
2293
2294 #if 0
2295 static void dump_refcounts(BlockDriverState *bs)
2296 {
2297     BDRVQcowState *s = bs->opaque;
2298     int64_t nb_clusters, k, k1, size;
2299     int refcount;
2300
2301     size = bdrv_getlength(bs->file);
2302     nb_clusters = size_to_clusters(s, size);
2303     for(k = 0; k < nb_clusters;) {
2304         k1 = k;
2305         refcount = get_refcount(bs, k);
2306         k++;
2307         while (k < nb_clusters && get_refcount(bs, k) == refcount)
2308             k++;
2309         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
2310                k - k1);
2311     }
2312 }
2313 #endif
2314
2315 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2316                               int64_t pos)
2317 {
2318     BDRVQcowState *s = bs->opaque;
2319     int64_t total_sectors = bs->total_sectors;
2320     int growable = bs->growable;
2321     bool zero_beyond_eof = bs->zero_beyond_eof;
2322     int ret;
2323
2324     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
2325     bs->growable = 1;
2326     bs->zero_beyond_eof = false;
2327     ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov);
2328     bs->growable = growable;
2329     bs->zero_beyond_eof = zero_beyond_eof;
2330
2331     /* bdrv_co_do_writev will have increased the total_sectors value to include
2332      * the VM state - the VM state is however not an actual part of the block
2333      * device, therefore, we need to restore the old value. */
2334     bs->total_sectors = total_sectors;
2335
2336     return ret;
2337 }
2338
2339 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2340                               int64_t pos, int size)
2341 {
2342     BDRVQcowState *s = bs->opaque;
2343     int growable = bs->growable;
2344     bool zero_beyond_eof = bs->zero_beyond_eof;
2345     int ret;
2346
2347     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
2348     bs->growable = 1;
2349     bs->zero_beyond_eof = false;
2350     ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
2351     bs->growable = growable;
2352     bs->zero_beyond_eof = zero_beyond_eof;
2353
2354     return ret;
2355 }
2356
2357 /*
2358  * Downgrades an image's version. To achieve this, any incompatible features
2359  * have to be removed.
2360  */
2361 static int qcow2_downgrade(BlockDriverState *bs, int target_version)
2362 {
2363     BDRVQcowState *s = bs->opaque;
2364     int current_version = s->qcow_version;
2365     int ret;
2366
2367     if (target_version == current_version) {
2368         return 0;
2369     } else if (target_version > current_version) {
2370         return -EINVAL;
2371     } else if (target_version != 2) {
2372         return -EINVAL;
2373     }
2374
2375     if (s->refcount_order != 4) {
2376         /* we would have to convert the image to a refcount_order == 4 image
2377          * here; however, since qemu (at the time of writing this) does not
2378          * support anything different than 4 anyway, there is no point in doing
2379          * so right now; however, we should error out (if qemu supports this in
2380          * the future and this code has not been adapted) */
2381         error_report("qcow2_downgrade: Image refcount orders other than 4 are "
2382                      "currently not supported.");
2383         return -ENOTSUP;
2384     }
2385
2386     /* clear incompatible features */
2387     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
2388         ret = qcow2_mark_clean(bs);
2389         if (ret < 0) {
2390             return ret;
2391         }
2392     }
2393
2394     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2395      * the first place; if that happens nonetheless, returning -ENOTSUP is the
2396      * best thing to do anyway */
2397
2398     if (s->incompatible_features) {
2399         return -ENOTSUP;
2400     }
2401
2402     /* since we can ignore compatible features, we can set them to 0 as well */
2403     s->compatible_features = 0;
2404     /* if lazy refcounts have been used, they have already been fixed through
2405      * clearing the dirty flag */
2406
2407     /* clearing autoclear features is trivial */
2408     s->autoclear_features = 0;
2409
2410     ret = qcow2_expand_zero_clusters(bs);
2411     if (ret < 0) {
2412         return ret;
2413     }
2414
2415     s->qcow_version = target_version;
2416     ret = qcow2_update_header(bs);
2417     if (ret < 0) {
2418         s->qcow_version = current_version;
2419         return ret;
2420     }
2421     return 0;
2422 }
2423
2424 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts)
2425 {
2426     BDRVQcowState *s = bs->opaque;
2427     int old_version = s->qcow_version, new_version = old_version;
2428     uint64_t new_size = 0;
2429     const char *backing_file = NULL, *backing_format = NULL;
2430     bool lazy_refcounts = s->use_lazy_refcounts;
2431     const char *compat = NULL;
2432     uint64_t cluster_size = s->cluster_size;
2433     bool encrypt;
2434     int ret;
2435     QemuOptDesc *desc = opts->list->desc;
2436
2437     while (desc && desc->name) {
2438         if (!qemu_opt_find(opts, desc->name)) {
2439             /* only change explicitly defined options */
2440             desc++;
2441             continue;
2442         }
2443
2444         if (!strcmp(desc->name, "compat")) {
2445             compat = qemu_opt_get(opts, "compat");
2446             if (!compat) {
2447                 /* preserve default */
2448             } else if (!strcmp(compat, "0.10")) {
2449                 new_version = 2;
2450             } else if (!strcmp(compat, "1.1")) {
2451                 new_version = 3;
2452             } else {
2453                 fprintf(stderr, "Unknown compatibility level %s.\n", compat);
2454                 return -EINVAL;
2455             }
2456         } else if (!strcmp(desc->name, "preallocation")) {
2457             fprintf(stderr, "Cannot change preallocation mode.\n");
2458             return -ENOTSUP;
2459         } else if (!strcmp(desc->name, "size")) {
2460             new_size = qemu_opt_get_size(opts, "size", 0);
2461         } else if (!strcmp(desc->name, "backing_file")) {
2462             backing_file = qemu_opt_get(opts, "backing_file");
2463         } else if (!strcmp(desc->name, "backing_fmt")) {
2464             backing_format = qemu_opt_get(opts, "backing_fmt");
2465         } else if (!strcmp(desc->name, "encryption")) {
2466             encrypt = qemu_opt_get_bool(opts, "encryption", s->crypt_method);
2467             if (encrypt != !!s->crypt_method) {
2468                 fprintf(stderr, "Changing the encryption flag is not "
2469                         "supported.\n");
2470                 return -ENOTSUP;
2471             }
2472         } else if (!strcmp(desc->name, "cluster_size")) {
2473             cluster_size = qemu_opt_get_size(opts, "cluster_size",
2474                                              cluster_size);
2475             if (cluster_size != s->cluster_size) {
2476                 fprintf(stderr, "Changing the cluster size is not "
2477                         "supported.\n");
2478                 return -ENOTSUP;
2479             }
2480         } else if (!strcmp(desc->name, "lazy_refcounts")) {
2481             lazy_refcounts = qemu_opt_get_bool(opts, "lazy_refcounts",
2482                                                lazy_refcounts);
2483         } else {
2484             /* if this assertion fails, this probably means a new option was
2485              * added without having it covered here */
2486             assert(false);
2487         }
2488
2489         desc++;
2490     }
2491
2492     if (new_version != old_version) {
2493         if (new_version > old_version) {
2494             /* Upgrade */
2495             s->qcow_version = new_version;
2496             ret = qcow2_update_header(bs);
2497             if (ret < 0) {
2498                 s->qcow_version = old_version;
2499                 return ret;
2500             }
2501         } else {
2502             ret = qcow2_downgrade(bs, new_version);
2503             if (ret < 0) {
2504                 return ret;
2505             }
2506         }
2507     }
2508
2509     if (backing_file || backing_format) {
2510         ret = qcow2_change_backing_file(bs, backing_file ?: bs->backing_file,
2511                                         backing_format ?: bs->backing_format);
2512         if (ret < 0) {
2513             return ret;
2514         }
2515     }
2516
2517     if (s->use_lazy_refcounts != lazy_refcounts) {
2518         if (lazy_refcounts) {
2519             if (s->qcow_version < 3) {
2520                 fprintf(stderr, "Lazy refcounts only supported with compatibility "
2521                         "level 1.1 and above (use compat=1.1 or greater)\n");
2522                 return -EINVAL;
2523             }
2524             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
2525             ret = qcow2_update_header(bs);
2526             if (ret < 0) {
2527                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
2528                 return ret;
2529             }
2530             s->use_lazy_refcounts = true;
2531         } else {
2532             /* make image clean first */
2533             ret = qcow2_mark_clean(bs);
2534             if (ret < 0) {
2535                 return ret;
2536             }
2537             /* now disallow lazy refcounts */
2538             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
2539             ret = qcow2_update_header(bs);
2540             if (ret < 0) {
2541                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
2542                 return ret;
2543             }
2544             s->use_lazy_refcounts = false;
2545         }
2546     }
2547
2548     if (new_size) {
2549         ret = bdrv_truncate(bs, new_size);
2550         if (ret < 0) {
2551             return ret;
2552         }
2553     }
2554
2555     return 0;
2556 }
2557
2558 /*
2559  * If offset or size are negative, respectively, they will not be included in
2560  * the BLOCK_IMAGE_CORRUPTED event emitted.
2561  * fatal will be ignored for read-only BDS; corruptions found there will always
2562  * be considered non-fatal.
2563  */
2564 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
2565                              int64_t size, const char *message_format, ...)
2566 {
2567     BDRVQcowState *s = bs->opaque;
2568     char *message;
2569     va_list ap;
2570
2571     fatal = fatal && !bs->read_only;
2572
2573     if (s->signaled_corruption &&
2574         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
2575     {
2576         return;
2577     }
2578
2579     va_start(ap, message_format);
2580     message = g_strdup_vprintf(message_format, ap);
2581     va_end(ap);
2582
2583     if (fatal) {
2584         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
2585                 "corruption events will be suppressed\n", message);
2586     } else {
2587         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
2588                 "corruption events will be suppressed\n", message);
2589     }
2590
2591     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), message,
2592                                           offset >= 0, offset, size >= 0, size,
2593                                           fatal, &error_abort);
2594     g_free(message);
2595
2596     if (fatal) {
2597         qcow2_mark_corrupt(bs);
2598         bs->drv = NULL; /* make BDS unusable */
2599     }
2600
2601     s->signaled_corruption = true;
2602 }
2603
2604 static QemuOptsList qcow2_create_opts = {
2605     .name = "qcow2-create-opts",
2606     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
2607     .desc = {
2608         {
2609             .name = BLOCK_OPT_SIZE,
2610             .type = QEMU_OPT_SIZE,
2611             .help = "Virtual disk size"
2612         },
2613         {
2614             .name = BLOCK_OPT_COMPAT_LEVEL,
2615             .type = QEMU_OPT_STRING,
2616             .help = "Compatibility level (0.10 or 1.1)"
2617         },
2618         {
2619             .name = BLOCK_OPT_BACKING_FILE,
2620             .type = QEMU_OPT_STRING,
2621             .help = "File name of a base image"
2622         },
2623         {
2624             .name = BLOCK_OPT_BACKING_FMT,
2625             .type = QEMU_OPT_STRING,
2626             .help = "Image format of the base image"
2627         },
2628         {
2629             .name = BLOCK_OPT_ENCRYPT,
2630             .type = QEMU_OPT_BOOL,
2631             .help = "Encrypt the image",
2632             .def_value_str = "off"
2633         },
2634         {
2635             .name = BLOCK_OPT_CLUSTER_SIZE,
2636             .type = QEMU_OPT_SIZE,
2637             .help = "qcow2 cluster size",
2638             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
2639         },
2640         {
2641             .name = BLOCK_OPT_PREALLOC,
2642             .type = QEMU_OPT_STRING,
2643             .help = "Preallocation mode (allowed values: off, metadata, "
2644                     "falloc, full)"
2645         },
2646         {
2647             .name = BLOCK_OPT_LAZY_REFCOUNTS,
2648             .type = QEMU_OPT_BOOL,
2649             .help = "Postpone refcount updates",
2650             .def_value_str = "off"
2651         },
2652         { /* end of list */ }
2653     }
2654 };
2655
2656 static BlockDriver bdrv_qcow2 = {
2657     .format_name        = "qcow2",
2658     .instance_size      = sizeof(BDRVQcowState),
2659     .bdrv_probe         = qcow2_probe,
2660     .bdrv_open          = qcow2_open,
2661     .bdrv_close         = qcow2_close,
2662     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
2663     .bdrv_create        = qcow2_create,
2664     .bdrv_has_zero_init = bdrv_has_zero_init_1,
2665     .bdrv_co_get_block_status = qcow2_co_get_block_status,
2666     .bdrv_set_key       = qcow2_set_key,
2667
2668     .bdrv_co_readv          = qcow2_co_readv,
2669     .bdrv_co_writev         = qcow2_co_writev,
2670     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
2671
2672     .bdrv_co_write_zeroes   = qcow2_co_write_zeroes,
2673     .bdrv_co_discard        = qcow2_co_discard,
2674     .bdrv_truncate          = qcow2_truncate,
2675     .bdrv_write_compressed  = qcow2_write_compressed,
2676
2677     .bdrv_snapshot_create   = qcow2_snapshot_create,
2678     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
2679     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
2680     .bdrv_snapshot_list     = qcow2_snapshot_list,
2681     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
2682     .bdrv_get_info          = qcow2_get_info,
2683     .bdrv_get_specific_info = qcow2_get_specific_info,
2684
2685     .bdrv_save_vmstate    = qcow2_save_vmstate,
2686     .bdrv_load_vmstate    = qcow2_load_vmstate,
2687
2688     .supports_backing           = true,
2689     .bdrv_change_backing_file   = qcow2_change_backing_file,
2690
2691     .bdrv_refresh_limits        = qcow2_refresh_limits,
2692     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
2693
2694     .create_opts         = &qcow2_create_opts,
2695     .bdrv_check          = qcow2_check,
2696     .bdrv_amend_options  = qcow2_amend_options,
2697 };
2698
2699 static void bdrv_qcow2_init(void)
2700 {
2701     bdrv_register(&bdrv_qcow2);
2702 }
2703
2704 block_init(bdrv_qcow2_init);