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/osdep.h"
  25 #include "qemu-common.h"
  26 #include "block/block_int.h"
  27 #include "sysemu/block-backend.h"
  28 #include "qemu/module.h"
  29 #include <zlib.h>
  30 #include "block/qcow2.h"
  31 #include "qemu/error-report.h"
  32 #include "qapi/qmp/qerror.h"
  33 #include "qapi/qmp/qbool.h"
  34 #include "qapi/util.h"
  35 #include "qapi/qmp/types.h"
  36 #include "qapi-event.h"
  37 #include "trace.h"
  38 #include "qemu/option_int.h"
  39
  40 /*
  41   Differences with QCOW:
  42
  43   - Support for multiple incremental snapshots.
  44   - Memory management by reference counts.
  45   - Clusters which have a reference count of one have the bit
  46     QCOW_OFLAG_COPIED to optimize write performance.
  47   - Size of compressed clusters is stored in sectors to reduce bit usage
  48     in the cluster offsets.
  49   - Support for storing additional data (such as the VM state) in the
  50     snapshots.
  51   - If a backing store is used, the cluster size is not constrained
  52     (could be backported to QCOW).
  53   - L2 tables have always a size of one cluster.
  54 */
  55
  56
  57 typedef struct {
  58     uint32_t magic;
  59     uint32_t len;
  60 } QEMU_PACKED QCowExtension;
  61
  62 #define  QCOW2_EXT_MAGIC_END 0
  63 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
  64 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
  65
  66 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
  67 {
  68     const QCowHeader *cow_header = (const void *)buf;
  69
  70     if (buf_size >= sizeof(QCowHeader) &&
  71         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
  72         be32_to_cpu(cow_header->version) >= 2)
  73         return 100;
  74     else
  75         return 0;
  76 }
  77
  78
  79 /*
  80  * read qcow2 extension and fill bs
  81  * start reading from start_offset
  82  * finish reading upon magic of value 0 or when end_offset reached
  83  * unknown magic is skipped (future extension this version knows nothing about)
  84  * return 0 upon success, non-0 otherwise
  85  */
  86 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
  87                                  uint64_t end_offset, void **p_feature_table,
  88                                  Error **errp)
  89 {
  90     BDRVQcow2State *s = bs->opaque;
  91     QCowExtension ext;
  92     uint64_t offset;
  93     int ret;
  94
  95 #ifdef DEBUG_EXT
  96     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
  97 #endif
  98     offset = start_offset;
  99     while (offset < end_offset) {
 100
 101 #ifdef DEBUG_EXT
 102         /* Sanity check */
 103         if (offset > s->cluster_size)
 104             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
 105
 106         printf("attempting to read extended header in offset %lu\n", offset);
 107 #endif
 108
 109         ret = bdrv_pread(bs->file->bs, offset, &ext, sizeof(ext));
 110         if (ret < 0) {
 111             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
 112                              "pread fail from offset %" PRIu64, offset);
 113             return 1;
 114         }
 115         be32_to_cpus(&ext.magic);
 116         be32_to_cpus(&ext.len);
 117         offset += sizeof(ext);
 118 #ifdef DEBUG_EXT
 119         printf("ext.magic = 0x%x\n", ext.magic);
 120 #endif
 121         if (offset > end_offset || ext.len > end_offset - offset) {
 122             error_setg(errp, "Header extension too large");
 123             return -EINVAL;
 124         }
 125
 126         switch (ext.magic) {
 127         case QCOW2_EXT_MAGIC_END:
 128             return 0;
 129
 130         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
 131             if (ext.len >= sizeof(bs->backing_format)) {
 132                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
 133                            " too large (>=%zu)", ext.len,
 134                            sizeof(bs->backing_format));
 135                 return 2;
 136             }
 137             ret = bdrv_pread(bs->file->bs, offset, bs->backing_format, ext.len);
 138             if (ret < 0) {
 139                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
 140                                  "Could not read format name");
 141                 return 3;
 142             }
 143             bs->backing_format[ext.len] = '\0';
 144             s->image_backing_format = g_strdup(bs->backing_format);
 145 #ifdef DEBUG_EXT
 146             printf("Qcow2: Got format extension %s\n", bs->backing_format);
 147 #endif
 148             break;
 149
 150         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
 151             if (p_feature_table != NULL) {
 152                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
 153                 ret = bdrv_pread(bs->file->bs, offset , feature_table, ext.len);
 154                 if (ret < 0) {
 155                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
 156                                      "Could not read table");
 157                     return ret;
 158                 }
 159
 160                 *p_feature_table = feature_table;
 161             }
 162             break;
 163
 164         default:
 165             /* unknown magic - save it in case we need to rewrite the header */
 166             {
 167                 Qcow2UnknownHeaderExtension *uext;
 168
 169                 uext = g_malloc0(sizeof(*uext)  + ext.len);
 170                 uext->magic = ext.magic;
 171                 uext->len = ext.len;
 172                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
 173
 174                 ret = bdrv_pread(bs->file->bs, offset , uext->data, uext->len);
 175                 if (ret < 0) {
 176                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
 177                                      "Could not read data");
 178                     return ret;
 179                 }
 180             }
 181             break;
 182         }
 183
 184         offset += ((ext.len + 7) & ~7);
 185     }
 186
 187     return 0;
 188 }
 189
 190 static void cleanup_unknown_header_ext(BlockDriverState *bs)
 191 {
 192     BDRVQcow2State *s = bs->opaque;
 193     Qcow2UnknownHeaderExtension *uext, *next;
 194
 195     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
 196         QLIST_REMOVE(uext, next);
 197         g_free(uext);
 198     }
 199 }
 200
 201 static void GCC_FMT_ATTR(3, 4) report_unsupported(BlockDriverState *bs,
 202     Error **errp, const char *fmt, ...)
 203 {
 204     char msg[64];
 205     va_list ap;
 206
 207     va_start(ap, fmt);
 208     vsnprintf(msg, sizeof(msg), fmt, ap);
 209     va_end(ap);
 210
 211     error_setg(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
 212                bdrv_get_device_or_node_name(bs), "qcow2", msg);
 213 }
 214
 215 static void report_unsupported_feature(BlockDriverState *bs,
 216     Error **errp, Qcow2Feature *table, uint64_t mask)
 217 {
 218     char *features = g_strdup("");
 219     char *old;
 220
 221     while (table && table->name[0] != '\0') {
 222         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
 223             if (mask & (1ULL << table->bit)) {
 224                 old = features;
 225                 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
 226                                            table->name);
 227                 g_free(old);
 228                 mask &= ~(1ULL << table->bit);
 229             }
 230         }
 231         table++;
 232     }
 233
 234     if (mask) {
 235         old = features;
 236         features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
 237                                    old, *old ? ", " : "", mask);
 238         g_free(old);
 239     }
 240
 241     report_unsupported(bs, errp, "%s", features);
 242     g_free(features);
 243 }
 244
 245 /*
 246  * Sets the dirty bit and flushes afterwards if necessary.
 247  *
 248  * The incompatible_features bit is only set if the image file header was
 249  * updated successfully.  Therefore it is not required to check the return
 250  * value of this function.
 251  */
 252 int qcow2_mark_dirty(BlockDriverState *bs)
 253 {
 254     BDRVQcow2State *s = bs->opaque;
 255     uint64_t val;
 256     int ret;
 257
 258     assert(s->qcow_version >= 3);
 259
 260     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
 261         return 0; /* already dirty */
 262     }
 263
 264     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
 265     ret = bdrv_pwrite(bs->file->bs, offsetof(QCowHeader, incompatible_features),
 266                       &val, sizeof(val));
 267     if (ret < 0) {
 268         return ret;
 269     }
 270     ret = bdrv_flush(bs->file->bs);
 271     if (ret < 0) {
 272         return ret;
 273     }
 274
 275     /* Only treat image as dirty if the header was updated successfully */
 276     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
 277     return 0;
 278 }
 279
 280 /*
 281  * Clears the dirty bit and flushes before if necessary.  Only call this
 282  * function when there are no pending requests, it does not guard against
 283  * concurrent requests dirtying the image.
 284  */
 285 static int qcow2_mark_clean(BlockDriverState *bs)
 286 {
 287     BDRVQcow2State *s = bs->opaque;
 288
 289     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
 290         int ret;
 291
 292         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
 293
 294         ret = bdrv_flush(bs);
 295         if (ret < 0) {
 296             return ret;
 297         }
 298
 299         return qcow2_update_header(bs);
 300     }
 301     return 0;
 302 }
 303
 304 /*
 305  * Marks the image as corrupt.
 306  */
 307 int qcow2_mark_corrupt(BlockDriverState *bs)
 308 {
 309     BDRVQcow2State *s = bs->opaque;
 310
 311     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
 312     return qcow2_update_header(bs);
 313 }
 314
 315 /*
 316  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
 317  * before if necessary.
 318  */
 319 int qcow2_mark_consistent(BlockDriverState *bs)
 320 {
 321     BDRVQcow2State *s = bs->opaque;
 322
 323     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
 324         int ret = bdrv_flush(bs);
 325         if (ret < 0) {
 326             return ret;
 327         }
 328
 329         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
 330         return qcow2_update_header(bs);
 331     }
 332     return 0;
 333 }
 334
 335 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
 336                        BdrvCheckMode fix)
 337 {
 338     int ret = qcow2_check_refcounts(bs, result, fix);
 339     if (ret < 0) {
 340         return ret;
 341     }
 342
 343     if (fix && result->check_errors == 0 && result->corruptions == 0) {
 344         ret = qcow2_mark_clean(bs);
 345         if (ret < 0) {
 346             return ret;
 347         }
 348         return qcow2_mark_consistent(bs);
 349     }
 350     return ret;
 351 }
 352
 353 static int validate_table_offset(BlockDriverState *bs, uint64_t offset,
 354                                  uint64_t entries, size_t entry_len)
 355 {
 356     BDRVQcow2State *s = bs->opaque;
 357     uint64_t size;
 358
 359     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
 360      * because values will be passed to qemu functions taking int64_t. */
 361     if (entries > INT64_MAX / entry_len) {
 362         return -EINVAL;
 363     }
 364
 365     size = entries * entry_len;
 366
 367     if (INT64_MAX - size < offset) {
 368         return -EINVAL;
 369     }
 370
 371     /* Tables must be cluster aligned */
 372     if (offset & (s->cluster_size - 1)) {
 373         return -EINVAL;
 374     }
 375
 376     return 0;
 377 }
 378
 379 static QemuOptsList qcow2_runtime_opts = {
 380     .name = "qcow2",
 381     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
 382     .desc = {
 383         {
 384             .name = QCOW2_OPT_LAZY_REFCOUNTS,
 385             .type = QEMU_OPT_BOOL,
 386             .help = "Postpone refcount updates",
 387         },
 388         {
 389             .name = QCOW2_OPT_DISCARD_REQUEST,
 390             .type = QEMU_OPT_BOOL,
 391             .help = "Pass guest discard requests to the layer below",
 392         },
 393         {
 394             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
 395             .type = QEMU_OPT_BOOL,
 396             .help = "Generate discard requests when snapshot related space "
 397                     "is freed",
 398         },
 399         {
 400             .name = QCOW2_OPT_DISCARD_OTHER,
 401             .type = QEMU_OPT_BOOL,
 402             .help = "Generate discard requests when other clusters are freed",
 403         },
 404         {
 405             .name = QCOW2_OPT_OVERLAP,
 406             .type = QEMU_OPT_STRING,
 407             .help = "Selects which overlap checks to perform from a range of "
 408                     "templates (none, constant, cached, all)",
 409         },
 410         {
 411             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
 412             .type = QEMU_OPT_STRING,
 413             .help = "Selects which overlap checks to perform from a range of "
 414                     "templates (none, constant, cached, all)",
 415         },
 416         {
 417             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
 418             .type = QEMU_OPT_BOOL,
 419             .help = "Check for unintended writes into the main qcow2 header",
 420         },
 421         {
 422             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
 423             .type = QEMU_OPT_BOOL,
 424             .help = "Check for unintended writes into the active L1 table",
 425         },
 426         {
 427             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
 428             .type = QEMU_OPT_BOOL,
 429             .help = "Check for unintended writes into an active L2 table",
 430         },
 431         {
 432             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
 433             .type = QEMU_OPT_BOOL,
 434             .help = "Check for unintended writes into the refcount table",
 435         },
 436         {
 437             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
 438             .type = QEMU_OPT_BOOL,
 439             .help = "Check for unintended writes into a refcount block",
 440         },
 441         {
 442             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
 443             .type = QEMU_OPT_BOOL,
 444             .help = "Check for unintended writes into the snapshot table",
 445         },
 446         {
 447             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
 448             .type = QEMU_OPT_BOOL,
 449             .help = "Check for unintended writes into an inactive L1 table",
 450         },
 451         {
 452             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
 453             .type = QEMU_OPT_BOOL,
 454             .help = "Check for unintended writes into an inactive L2 table",
 455         },
 456         {
 457             .name = QCOW2_OPT_CACHE_SIZE,
 458             .type = QEMU_OPT_SIZE,
 459             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
 460                     "cache size",
 461         },
 462         {
 463             .name = QCOW2_OPT_L2_CACHE_SIZE,
 464             .type = QEMU_OPT_SIZE,
 465             .help = "Maximum L2 table cache size",
 466         },
 467         {
 468             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
 469             .type = QEMU_OPT_SIZE,
 470             .help = "Maximum refcount block cache size",
 471         },
 472         {
 473             .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
 474             .type = QEMU_OPT_NUMBER,
 475             .help = "Clean unused cache entries after this time (in seconds)",
 476         },
 477         { /* end of list */ }
 478     },
 479 };
 480
 481 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
 482     [QCOW2_OL_MAIN_HEADER_BITNR]    = QCOW2_OPT_OVERLAP_MAIN_HEADER,
 483     [QCOW2_OL_ACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L1,
 484     [QCOW2_OL_ACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L2,
 485     [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
 486     [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
 487     [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
 488     [QCOW2_OL_INACTIVE_L1_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L1,
 489     [QCOW2_OL_INACTIVE_L2_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L2,
 490 };
 491
 492 static void cache_clean_timer_cb(void *opaque)
 493 {
 494     BlockDriverState *bs = opaque;
 495     BDRVQcow2State *s = bs->opaque;
 496     qcow2_cache_clean_unused(bs, s->l2_table_cache);
 497     qcow2_cache_clean_unused(bs, s->refcount_block_cache);
 498     timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
 499               (int64_t) s->cache_clean_interval * 1000);
 500 }
 501
 502 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
 503 {
 504     BDRVQcow2State *s = bs->opaque;
 505     if (s->cache_clean_interval > 0) {
 506         s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
 507                                              SCALE_MS, cache_clean_timer_cb,
 508                                              bs);
 509         timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
 510                   (int64_t) s->cache_clean_interval * 1000);
 511     }
 512 }
 513
 514 static void cache_clean_timer_del(BlockDriverState *bs)
 515 {
 516     BDRVQcow2State *s = bs->opaque;
 517     if (s->cache_clean_timer) {
 518         timer_del(s->cache_clean_timer);
 519         timer_free(s->cache_clean_timer);
 520         s->cache_clean_timer = NULL;
 521     }
 522 }
 523
 524 static void qcow2_detach_aio_context(BlockDriverState *bs)
 525 {
 526     cache_clean_timer_del(bs);
 527 }
 528
 529 static void qcow2_attach_aio_context(BlockDriverState *bs,
 530                                      AioContext *new_context)
 531 {
 532     cache_clean_timer_init(bs, new_context);
 533 }
 534
 535 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
 536                              uint64_t *l2_cache_size,
 537                              uint64_t *refcount_cache_size, Error **errp)
 538 {
 539     BDRVQcow2State *s = bs->opaque;
 540     uint64_t combined_cache_size;
 541     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
 542
 543     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
 544     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
 545     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
 546
 547     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
 548     *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0);
 549     *refcount_cache_size = qemu_opt_get_size(opts,
 550                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
 551
 552     if (combined_cache_size_set) {
 553         if (l2_cache_size_set && refcount_cache_size_set) {
 554             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
 555                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
 556                        "the same time");
 557             return;
 558         } else if (*l2_cache_size > combined_cache_size) {
 559             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
 560                        QCOW2_OPT_CACHE_SIZE);
 561             return;
 562         } else if (*refcount_cache_size > combined_cache_size) {
 563             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
 564                        QCOW2_OPT_CACHE_SIZE);
 565             return;
 566         }
 567
 568         if (l2_cache_size_set) {
 569             *refcount_cache_size = combined_cache_size - *l2_cache_size;
 570         } else if (refcount_cache_size_set) {
 571             *l2_cache_size = combined_cache_size - *refcount_cache_size;
 572         } else {
 573             *refcount_cache_size = combined_cache_size
 574                                  / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1);
 575             *l2_cache_size = combined_cache_size - *refcount_cache_size;
 576         }
 577     } else {
 578         if (!l2_cache_size_set && !refcount_cache_size_set) {
 579             *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE,
 580                                  (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
 581                                  * s->cluster_size);
 582             *refcount_cache_size = *l2_cache_size
 583                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 584         } else if (!l2_cache_size_set) {
 585             *l2_cache_size = *refcount_cache_size
 586                            * DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 587         } else if (!refcount_cache_size_set) {
 588             *refcount_cache_size = *l2_cache_size
 589                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 590         }
 591     }
 592 }
 593
 594 typedef struct Qcow2ReopenState {
 595     Qcow2Cache *l2_table_cache;
 596     Qcow2Cache *refcount_block_cache;
 597     bool use_lazy_refcounts;
 598     int overlap_check;
 599     bool discard_passthrough[QCOW2_DISCARD_MAX];
 600     uint64_t cache_clean_interval;
 601 } Qcow2ReopenState;
 602
 603 static int qcow2_update_options_prepare(BlockDriverState *bs,
 604                                         Qcow2ReopenState *r,
 605                                         QDict *options, int flags,
 606                                         Error **errp)
 607 {
 608     BDRVQcow2State *s = bs->opaque;
 609     QemuOpts *opts = NULL;
 610     const char *opt_overlap_check, *opt_overlap_check_template;
 611     int overlap_check_template = 0;
 612     uint64_t l2_cache_size, refcount_cache_size;
 613     int i;
 614     Error *local_err = NULL;
 615     int ret;
 616
 617     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
 618     qemu_opts_absorb_qdict(opts, options, &local_err);
 619     if (local_err) {
 620         error_propagate(errp, local_err);
 621         ret = -EINVAL;
 622         goto fail;
 623     }
 624
 625     /* get L2 table/refcount block cache size from command line options */
 626     read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size,
 627                      &local_err);
 628     if (local_err) {
 629         error_propagate(errp, local_err);
 630         ret = -EINVAL;
 631         goto fail;
 632     }
 633
 634     l2_cache_size /= s->cluster_size;
 635     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
 636         l2_cache_size = MIN_L2_CACHE_SIZE;
 637     }
 638     if (l2_cache_size > INT_MAX) {
 639         error_setg(errp, "L2 cache size too big");
 640         ret = -EINVAL;
 641         goto fail;
 642     }
 643
 644     refcount_cache_size /= s->cluster_size;
 645     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
 646         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
 647     }
 648     if (refcount_cache_size > INT_MAX) {
 649         error_setg(errp, "Refcount cache size too big");
 650         ret = -EINVAL;
 651         goto fail;
 652     }
 653
 654     /* alloc new L2 table/refcount block cache, flush old one */
 655     if (s->l2_table_cache) {
 656         ret = qcow2_cache_flush(bs, s->l2_table_cache);
 657         if (ret) {
 658             error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
 659             goto fail;
 660         }
 661     }
 662
 663     if (s->refcount_block_cache) {
 664         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
 665         if (ret) {
 666             error_setg_errno(errp, -ret,
 667                              "Failed to flush the refcount block cache");
 668             goto fail;
 669         }
 670     }
 671
 672     r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size);
 673     r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size);
 674     if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
 675         error_setg(errp, "Could not allocate metadata caches");
 676         ret = -ENOMEM;
 677         goto fail;
 678     }
 679
 680     /* New interval for cache cleanup timer */
 681     r->cache_clean_interval =
 682         qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
 683                             s->cache_clean_interval);
 684     if (r->cache_clean_interval > UINT_MAX) {
 685         error_setg(errp, "Cache clean interval too big");
 686         ret = -EINVAL;
 687         goto fail;
 688     }
 689
 690     /* lazy-refcounts; flush if going from enabled to disabled */
 691     r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
 692         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
 693     if (r->use_lazy_refcounts && s->qcow_version < 3) {
 694         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
 695                    "qemu 1.1 compatibility level");
 696         ret = -EINVAL;
 697         goto fail;
 698     }
 699
 700     if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
 701         ret = qcow2_mark_clean(bs);
 702         if (ret < 0) {
 703             error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
 704             goto fail;
 705         }
 706     }
 707
 708     /* Overlap check options */
 709     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
 710     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
 711     if (opt_overlap_check_template && opt_overlap_check &&
 712         strcmp(opt_overlap_check_template, opt_overlap_check))
 713     {
 714         error_setg(errp, "Conflicting values for qcow2 options '"
 715                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
 716                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
 717         ret = -EINVAL;
 718         goto fail;
 719     }
 720     if (!opt_overlap_check) {
 721         opt_overlap_check = opt_overlap_check_template ?: "cached";
 722     }
 723
 724     if (!strcmp(opt_overlap_check, "none")) {
 725         overlap_check_template = 0;
 726     } else if (!strcmp(opt_overlap_check, "constant")) {
 727         overlap_check_template = QCOW2_OL_CONSTANT;
 728     } else if (!strcmp(opt_overlap_check, "cached")) {
 729         overlap_check_template = QCOW2_OL_CACHED;
 730     } else if (!strcmp(opt_overlap_check, "all")) {
 731         overlap_check_template = QCOW2_OL_ALL;
 732     } else {
 733         error_setg(errp, "Unsupported value '%s' for qcow2 option "
 734                    "'overlap-check'. Allowed are any of the following: "
 735                    "none, constant, cached, all", opt_overlap_check);
 736         ret = -EINVAL;
 737         goto fail;
 738     }
 739
 740     r->overlap_check = 0;
 741     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
 742         /* overlap-check defines a template bitmask, but every flag may be
 743          * overwritten through the associated boolean option */
 744         r->overlap_check |=
 745             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
 746                               overlap_check_template & (1 << i)) << i;
 747     }
 748
 749     r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
 750     r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
 751     r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
 752         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
 753                           flags & BDRV_O_UNMAP);
 754     r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
 755         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
 756     r->discard_passthrough[QCOW2_DISCARD_OTHER] =
 757         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
 758
 759     ret = 0;
 760 fail:
 761     qemu_opts_del(opts);
 762     opts = NULL;
 763     return ret;
 764 }
 765
 766 static void qcow2_update_options_commit(BlockDriverState *bs,
 767                                         Qcow2ReopenState *r)
 768 {
 769     BDRVQcow2State *s = bs->opaque;
 770     int i;
 771
 772     if (s->l2_table_cache) {
 773         qcow2_cache_destroy(bs, s->l2_table_cache);
 774     }
 775     if (s->refcount_block_cache) {
 776         qcow2_cache_destroy(bs, s->refcount_block_cache);
 777     }
 778     s->l2_table_cache = r->l2_table_cache;
 779     s->refcount_block_cache = r->refcount_block_cache;
 780
 781     s->overlap_check = r->overlap_check;
 782     s->use_lazy_refcounts = r->use_lazy_refcounts;
 783
 784     for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
 785         s->discard_passthrough[i] = r->discard_passthrough[i];
 786     }
 787
 788     if (s->cache_clean_interval != r->cache_clean_interval) {
 789         cache_clean_timer_del(bs);
 790         s->cache_clean_interval = r->cache_clean_interval;
 791         cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
 792     }
 793 }
 794
 795 static void qcow2_update_options_abort(BlockDriverState *bs,
 796                                        Qcow2ReopenState *r)
 797 {
 798     if (r->l2_table_cache) {
 799         qcow2_cache_destroy(bs, r->l2_table_cache);
 800     }
 801     if (r->refcount_block_cache) {
 802         qcow2_cache_destroy(bs, r->refcount_block_cache);
 803     }
 804 }
 805
 806 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
 807                                 int flags, Error **errp)
 808 {
 809     Qcow2ReopenState r = {};
 810     int ret;
 811
 812     ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
 813     if (ret >= 0) {
 814         qcow2_update_options_commit(bs, &r);
 815     } else {
 816         qcow2_update_options_abort(bs, &r);
 817     }
 818
 819     return ret;
 820 }
 821
 822 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
 823                       Error **errp)
 824 {
 825     BDRVQcow2State *s = bs->opaque;
 826     unsigned int len, i;
 827     int ret = 0;
 828     QCowHeader header;
 829     Error *local_err = NULL;
 830     uint64_t ext_end;
 831     uint64_t l1_vm_state_index;
 832
 833     ret = bdrv_pread(bs->file->bs, 0, &header, sizeof(header));
 834     if (ret < 0) {
 835         error_setg_errno(errp, -ret, "Could not read qcow2 header");
 836         goto fail;
 837     }
 838     be32_to_cpus(&header.magic);
 839     be32_to_cpus(&header.version);
 840     be64_to_cpus(&header.backing_file_offset);
 841     be32_to_cpus(&header.backing_file_size);
 842     be64_to_cpus(&header.size);
 843     be32_to_cpus(&header.cluster_bits);
 844     be32_to_cpus(&header.crypt_method);
 845     be64_to_cpus(&header.l1_table_offset);
 846     be32_to_cpus(&header.l1_size);
 847     be64_to_cpus(&header.refcount_table_offset);
 848     be32_to_cpus(&header.refcount_table_clusters);
 849     be64_to_cpus(&header.snapshots_offset);
 850     be32_to_cpus(&header.nb_snapshots);
 851
 852     if (header.magic != QCOW_MAGIC) {
 853         error_setg(errp, "Image is not in qcow2 format");
 854         ret = -EINVAL;
 855         goto fail;
 856     }
 857     if (header.version < 2 || header.version > 3) {
 858         report_unsupported(bs, errp, "QCOW version %" PRIu32, header.version);
 859         ret = -ENOTSUP;
 860         goto fail;
 861     }
 862
 863     s->qcow_version = header.version;
 864
 865     /* Initialise cluster size */
 866     if (header.cluster_bits < MIN_CLUSTER_BITS ||
 867         header.cluster_bits > MAX_CLUSTER_BITS) {
 868         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
 869                    header.cluster_bits);
 870         ret = -EINVAL;
 871         goto fail;
 872     }
 873
 874     s->cluster_bits = header.cluster_bits;
 875     s->cluster_size = 1 << s->cluster_bits;
 876     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 877
 878     /* Initialise version 3 header fields */
 879     if (header.version == 2) {
 880         header.incompatible_features    = 0;
 881         header.compatible_features      = 0;
 882         header.autoclear_features       = 0;
 883         header.refcount_order           = 4;
 884         header.header_length            = 72;
 885     } else {
 886         be64_to_cpus(&header.incompatible_features);
 887         be64_to_cpus(&header.compatible_features);
 888         be64_to_cpus(&header.autoclear_features);
 889         be32_to_cpus(&header.refcount_order);
 890         be32_to_cpus(&header.header_length);
 891
 892         if (header.header_length < 104) {
 893             error_setg(errp, "qcow2 header too short");
 894             ret = -EINVAL;
 895             goto fail;
 896         }
 897     }
 898
 899     if (header.header_length > s->cluster_size) {
 900         error_setg(errp, "qcow2 header exceeds cluster size");
 901         ret = -EINVAL;
 902         goto fail;
 903     }
 904
 905     if (header.header_length > sizeof(header)) {
 906         s->unknown_header_fields_size = header.header_length - sizeof(header);
 907         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
 908         ret = bdrv_pread(bs->file->bs, sizeof(header), s->unknown_header_fields,
 909                          s->unknown_header_fields_size);
 910         if (ret < 0) {
 911             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
 912                              "fields");
 913             goto fail;
 914         }
 915     }
 916
 917     if (header.backing_file_offset > s->cluster_size) {
 918         error_setg(errp, "Invalid backing file offset");
 919         ret = -EINVAL;
 920         goto fail;
 921     }
 922
 923     if (header.backing_file_offset) {
 924         ext_end = header.backing_file_offset;
 925     } else {
 926         ext_end = 1 << header.cluster_bits;
 927     }
 928
 929     /* Handle feature bits */
 930     s->incompatible_features    = header.incompatible_features;
 931     s->compatible_features      = header.compatible_features;
 932     s->autoclear_features       = header.autoclear_features;
 933
 934     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
 935         void *feature_table = NULL;
 936         qcow2_read_extensions(bs, header.header_length, ext_end,
 937                               &feature_table, NULL);
 938         report_unsupported_feature(bs, errp, feature_table,
 939                                    s->incompatible_features &
 940                                    ~QCOW2_INCOMPAT_MASK);
 941         ret = -ENOTSUP;
 942         g_free(feature_table);
 943         goto fail;
 944     }
 945
 946     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
 947         /* Corrupt images may not be written to unless they are being repaired
 948          */
 949         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
 950             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
 951                        "read/write");
 952             ret = -EACCES;
 953             goto fail;
 954         }
 955     }
 956
 957     /* Check support for various header values */
 958     if (header.refcount_order > 6) {
 959         error_setg(errp, "Reference count entry width too large; may not "
 960                    "exceed 64 bits");
 961         ret = -EINVAL;
 962         goto fail;
 963     }
 964     s->refcount_order = header.refcount_order;
 965     s->refcount_bits = 1 << s->refcount_order;
 966     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
 967     s->refcount_max += s->refcount_max - 1;
 968
 969     if (header.crypt_method > QCOW_CRYPT_AES) {
 970         error_setg(errp, "Unsupported encryption method: %" PRIu32,
 971                    header.crypt_method);
 972         ret = -EINVAL;
 973         goto fail;
 974     }
 975     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) {
 976         error_setg(errp, "AES cipher not available");
 977         ret = -EINVAL;
 978         goto fail;
 979     }
 980     s->crypt_method_header = header.crypt_method;
 981     if (s->crypt_method_header) {
 982         bs->encrypted = 1;
 983     }
 984
 985     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
 986     s->l2_size = 1 << s->l2_bits;
 987     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
 988     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
 989     s->refcount_block_size = 1 << s->refcount_block_bits;
 990     bs->total_sectors = header.size / 512;
 991     s->csize_shift = (62 - (s->cluster_bits - 8));
 992     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
 993     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
 994
 995     s->refcount_table_offset = header.refcount_table_offset;
 996     s->refcount_table_size =
 997         header.refcount_table_clusters << (s->cluster_bits - 3);
 998
 999     if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
1000         error_setg(errp, "Reference count table too large");
1001         ret = -EINVAL;
1002         goto fail;
1003     }
1004
1005     ret = validate_table_offset(bs, s->refcount_table_offset,
1006                                 s->refcount_table_size, sizeof(uint64_t));
1007     if (ret < 0) {
1008         error_setg(errp, "Invalid reference count table offset");
1009         goto fail;
1010     }
1011
1012     /* Snapshot table offset/length */
1013     if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
1014         error_setg(errp, "Too many snapshots");
1015         ret = -EINVAL;
1016         goto fail;
1017     }
1018
1019     ret = validate_table_offset(bs, header.snapshots_offset,
1020                                 header.nb_snapshots,
1021                                 sizeof(QCowSnapshotHeader));
1022     if (ret < 0) {
1023         error_setg(errp, "Invalid snapshot table offset");
1024         goto fail;
1025     }
1026
1027     /* read the level 1 table */
1028     if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
1029         error_setg(errp, "Active L1 table too large");
1030         ret = -EFBIG;
1031         goto fail;
1032     }
1033     s->l1_size = header.l1_size;
1034
1035     l1_vm_state_index = size_to_l1(s, header.size);
1036     if (l1_vm_state_index > INT_MAX) {
1037         error_setg(errp, "Image is too big");
1038         ret = -EFBIG;
1039         goto fail;
1040     }
1041     s->l1_vm_state_index = l1_vm_state_index;
1042
1043     /* the L1 table must contain at least enough entries to put
1044        header.size bytes */
1045     if (s->l1_size < s->l1_vm_state_index) {
1046         error_setg(errp, "L1 table is too small");
1047         ret = -EINVAL;
1048         goto fail;
1049     }
1050
1051     ret = validate_table_offset(bs, header.l1_table_offset,
1052                                 header.l1_size, sizeof(uint64_t));
1053     if (ret < 0) {
1054         error_setg(errp, "Invalid L1 table offset");
1055         goto fail;
1056     }
1057     s->l1_table_offset = header.l1_table_offset;
1058
1059
1060     if (s->l1_size > 0) {
1061         s->l1_table = qemu_try_blockalign(bs->file->bs,
1062             align_offset(s->l1_size * sizeof(uint64_t), 512));
1063         if (s->l1_table == NULL) {
1064             error_setg(errp, "Could not allocate L1 table");
1065             ret = -ENOMEM;
1066             goto fail;
1067         }
1068         ret = bdrv_pread(bs->file->bs, s->l1_table_offset, s->l1_table,
1069                          s->l1_size * sizeof(uint64_t));
1070         if (ret < 0) {
1071             error_setg_errno(errp, -ret, "Could not read L1 table");
1072             goto fail;
1073         }
1074         for(i = 0;i < s->l1_size; i++) {
1075             be64_to_cpus(&s->l1_table[i]);
1076         }
1077     }
1078
1079     /* Parse driver-specific options */
1080     ret = qcow2_update_options(bs, options, flags, errp);
1081     if (ret < 0) {
1082         goto fail;
1083     }
1084
1085     s->cluster_cache = g_malloc(s->cluster_size);
1086     /* one more sector for decompressed data alignment */
1087     s->cluster_data = qemu_try_blockalign(bs->file->bs, QCOW_MAX_CRYPT_CLUSTERS
1088                                                     * s->cluster_size + 512);
1089     if (s->cluster_data == NULL) {
1090         error_setg(errp, "Could not allocate temporary cluster buffer");
1091         ret = -ENOMEM;
1092         goto fail;
1093     }
1094
1095     s->cluster_cache_offset = -1;
1096     s->flags = flags;
1097
1098     ret = qcow2_refcount_init(bs);
1099     if (ret != 0) {
1100         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1101         goto fail;
1102     }
1103
1104     QLIST_INIT(&s->cluster_allocs);
1105     QTAILQ_INIT(&s->discards);
1106
1107     /* read qcow2 extensions */
1108     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1109         &local_err)) {
1110         error_propagate(errp, local_err);
1111         ret = -EINVAL;
1112         goto fail;
1113     }
1114
1115     /* read the backing file name */
1116     if (header.backing_file_offset != 0) {
1117         len = header.backing_file_size;
1118         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1119             len >= sizeof(bs->backing_file)) {
1120             error_setg(errp, "Backing file name too long");
1121             ret = -EINVAL;
1122             goto fail;
1123         }
1124         ret = bdrv_pread(bs->file->bs, header.backing_file_offset,
1125                          bs->backing_file, len);
1126         if (ret < 0) {
1127             error_setg_errno(errp, -ret, "Could not read backing file name");
1128             goto fail;
1129         }
1130         bs->backing_file[len] = '\0';
1131         s->image_backing_file = g_strdup(bs->backing_file);
1132     }
1133
1134     /* Internal snapshots */
1135     s->snapshots_offset = header.snapshots_offset;
1136     s->nb_snapshots = header.nb_snapshots;
1137
1138     ret = qcow2_read_snapshots(bs);
1139     if (ret < 0) {
1140         error_setg_errno(errp, -ret, "Could not read snapshots");
1141         goto fail;
1142     }
1143
1144     /* Clear unknown autoclear feature bits */
1145     if (!bs->read_only && !(flags & BDRV_O_INACTIVE) && s->autoclear_features) {
1146         s->autoclear_features = 0;
1147         ret = qcow2_update_header(bs);
1148         if (ret < 0) {
1149             error_setg_errno(errp, -ret, "Could not update qcow2 header");
1150             goto fail;
1151         }
1152     }
1153
1154     /* Initialise locks */
1155     qemu_co_mutex_init(&s->lock);
1156
1157     /* Repair image if dirty */
1158     if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1159         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1160         BdrvCheckResult result = {0};
1161
1162         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1163         if (ret < 0) {
1164             error_setg_errno(errp, -ret, "Could not repair dirty image");
1165             goto fail;
1166         }
1167     }
1168
1169 #ifdef DEBUG_ALLOC
1170     {
1171         BdrvCheckResult result = {0};
1172         qcow2_check_refcounts(bs, &result, 0);
1173     }
1174 #endif
1175     return ret;
1176
1177  fail:
1178     g_free(s->unknown_header_fields);
1179     cleanup_unknown_header_ext(bs);
1180     qcow2_free_snapshots(bs);
1181     qcow2_refcount_close(bs);
1182     qemu_vfree(s->l1_table);
1183     /* else pre-write overlap checks in cache_destroy may crash */
1184     s->l1_table = NULL;
1185     cache_clean_timer_del(bs);
1186     if (s->l2_table_cache) {
1187         qcow2_cache_destroy(bs, s->l2_table_cache);
1188     }
1189     if (s->refcount_block_cache) {
1190         qcow2_cache_destroy(bs, s->refcount_block_cache);
1191     }
1192     g_free(s->cluster_cache);
1193     qemu_vfree(s->cluster_data);
1194     return ret;
1195 }
1196
1197 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1198 {
1199     BDRVQcow2State *s = bs->opaque;
1200
1201     bs->bl.write_zeroes_alignment = s->cluster_sectors;
1202 }
1203
1204 static int qcow2_set_key(BlockDriverState *bs, const char *key)
1205 {
1206     BDRVQcow2State *s = bs->opaque;
1207     uint8_t keybuf[16];
1208     int len, i;
1209     Error *err = NULL;
1210
1211     memset(keybuf, 0, 16);
1212     len = strlen(key);
1213     if (len > 16)
1214         len = 16;
1215     /* XXX: we could compress the chars to 7 bits to increase
1216        entropy */
1217     for(i = 0;i < len;i++) {
1218         keybuf[i] = key[i];
1219     }
1220     assert(bs->encrypted);
1221
1222     qcrypto_cipher_free(s->cipher);
1223     s->cipher = qcrypto_cipher_new(
1224         QCRYPTO_CIPHER_ALG_AES_128,
1225         QCRYPTO_CIPHER_MODE_CBC,
1226         keybuf, G_N_ELEMENTS(keybuf),
1227         &err);
1228
1229     if (!s->cipher) {
1230         /* XXX would be nice if errors in this method could
1231          * be properly propagate to the caller. Would need
1232          * the bdrv_set_key() API signature to be fixed. */
1233         error_free(err);
1234         return -1;
1235     }
1236     return 0;
1237 }
1238
1239 static int qcow2_reopen_prepare(BDRVReopenState *state,
1240                                 BlockReopenQueue *queue, Error **errp)
1241 {
1242     Qcow2ReopenState *r;
1243     int ret;
1244
1245     r = g_new0(Qcow2ReopenState, 1);
1246     state->opaque = r;
1247
1248     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1249                                        state->flags, errp);
1250     if (ret < 0) {
1251         goto fail;
1252     }
1253
1254     /* We need to write out any unwritten data if we reopen read-only. */
1255     if ((state->flags & BDRV_O_RDWR) == 0) {
1256         ret = bdrv_flush(state->bs);
1257         if (ret < 0) {
1258             goto fail;
1259         }
1260
1261         ret = qcow2_mark_clean(state->bs);
1262         if (ret < 0) {
1263             goto fail;
1264         }
1265     }
1266
1267     return 0;
1268
1269 fail:
1270     qcow2_update_options_abort(state->bs, r);
1271     g_free(r);
1272     return ret;
1273 }
1274
1275 static void qcow2_reopen_commit(BDRVReopenState *state)
1276 {
1277     qcow2_update_options_commit(state->bs, state->opaque);
1278     g_free(state->opaque);
1279 }
1280
1281 static void qcow2_reopen_abort(BDRVReopenState *state)
1282 {
1283     qcow2_update_options_abort(state->bs, state->opaque);
1284     g_free(state->opaque);
1285 }
1286
1287 static void qcow2_join_options(QDict *options, QDict *old_options)
1288 {
1289     bool has_new_overlap_template =
1290         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1291         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1292     bool has_new_total_cache_size =
1293         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1294     bool has_all_cache_options;
1295
1296     /* New overlap template overrides all old overlap options */
1297     if (has_new_overlap_template) {
1298         qdict_del(old_options, QCOW2_OPT_OVERLAP);
1299         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1300         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1301         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1302         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1303         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1304         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1305         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1306         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1307         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1308     }
1309
1310     /* New total cache size overrides all old options */
1311     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1312         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1313         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1314     }
1315
1316     qdict_join(options, old_options, false);
1317
1318     /*
1319      * If after merging all cache size options are set, an old total size is
1320      * overwritten. Do keep all options, however, if all three are new. The
1321      * resulting error message is what we want to happen.
1322      */
1323     has_all_cache_options =
1324         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1325         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1326         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1327
1328     if (has_all_cache_options && !has_new_total_cache_size) {
1329         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1330     }
1331 }
1332
1333 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1334         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1335 {
1336     BDRVQcow2State *s = bs->opaque;
1337     uint64_t cluster_offset;
1338     int index_in_cluster, ret;
1339     int64_t status = 0;
1340
1341     *pnum = nb_sectors;
1342     qemu_co_mutex_lock(&s->lock);
1343     ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
1344     qemu_co_mutex_unlock(&s->lock);
1345     if (ret < 0) {
1346         return ret;
1347     }
1348
1349     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1350         !s->cipher) {
1351         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1352         cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1353         *file = bs->file->bs;
1354         status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1355     }
1356     if (ret == QCOW2_CLUSTER_ZERO) {
1357         status |= BDRV_BLOCK_ZERO;
1358     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1359         status |= BDRV_BLOCK_DATA;
1360     }
1361     return status;
1362 }
1363
1364 /* handle reading after the end of the backing file */
1365 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
1366                   int64_t sector_num, int nb_sectors)
1367 {
1368     int n1;
1369     if ((sector_num + nb_sectors) <= bs->total_sectors)
1370         return nb_sectors;
1371     if (sector_num >= bs->total_sectors)
1372         n1 = 0;
1373     else
1374         n1 = bs->total_sectors - sector_num;
1375
1376     qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1));
1377
1378     return n1;
1379 }
1380
1381 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
1382                           int remaining_sectors, QEMUIOVector *qiov)
1383 {
1384     BDRVQcow2State *s = bs->opaque;
1385     int index_in_cluster, n1;
1386     int ret;
1387     int cur_nr_sectors; /* number of sectors in current iteration */
1388     uint64_t cluster_offset = 0;
1389     uint64_t bytes_done = 0;
1390     QEMUIOVector hd_qiov;
1391     uint8_t *cluster_data = NULL;
1392
1393     qemu_iovec_init(&hd_qiov, qiov->niov);
1394
1395     qemu_co_mutex_lock(&s->lock);
1396
1397     while (remaining_sectors != 0) {
1398
1399         /* prepare next request */
1400         cur_nr_sectors = remaining_sectors;
1401         if (s->cipher) {
1402             cur_nr_sectors = MIN(cur_nr_sectors,
1403                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1404         }
1405
1406         ret = qcow2_get_cluster_offset(bs, sector_num << 9,
1407             &cur_nr_sectors, &cluster_offset);
1408         if (ret < 0) {
1409             goto fail;
1410         }
1411
1412         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1413
1414         qemu_iovec_reset(&hd_qiov);
1415         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1416             cur_nr_sectors * 512);
1417
1418         switch (ret) {
1419         case QCOW2_CLUSTER_UNALLOCATED:
1420
1421             if (bs->backing) {
1422                 /* read from the base image */
1423                 n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov,
1424                     sector_num, cur_nr_sectors);
1425                 if (n1 > 0) {
1426                     QEMUIOVector local_qiov;
1427
1428                     qemu_iovec_init(&local_qiov, hd_qiov.niov);
1429                     qemu_iovec_concat(&local_qiov, &hd_qiov, 0,
1430                                       n1 * BDRV_SECTOR_SIZE);
1431
1432                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1433                     qemu_co_mutex_unlock(&s->lock);
1434                     ret = bdrv_co_readv(bs->backing->bs, sector_num,
1435                                         n1, &local_qiov);
1436                     qemu_co_mutex_lock(&s->lock);
1437
1438                     qemu_iovec_destroy(&local_qiov);
1439
1440                     if (ret < 0) {
1441                         goto fail;
1442                     }
1443                 }
1444             } else {
1445                 /* Note: in this case, no need to wait */
1446                 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1447             }
1448             break;
1449
1450         case QCOW2_CLUSTER_ZERO:
1451             qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1452             break;
1453
1454         case QCOW2_CLUSTER_COMPRESSED:
1455             /* add AIO support for compressed blocks ? */
1456             ret = qcow2_decompress_cluster(bs, cluster_offset);
1457             if (ret < 0) {
1458                 goto fail;
1459             }
1460
1461             qemu_iovec_from_buf(&hd_qiov, 0,
1462                 s->cluster_cache + index_in_cluster * 512,
1463                 512 * cur_nr_sectors);
1464             break;
1465
1466         case QCOW2_CLUSTER_NORMAL:
1467             if ((cluster_offset & 511) != 0) {
1468                 ret = -EIO;
1469                 goto fail;
1470             }
1471
1472             if (bs->encrypted) {
1473                 assert(s->cipher);
1474
1475                 /*
1476                  * For encrypted images, read everything into a temporary
1477                  * contiguous buffer on which the AES functions can work.
1478                  */
1479                 if (!cluster_data) {
1480                     cluster_data =
1481                         qemu_try_blockalign(bs->file->bs,
1482                                             QCOW_MAX_CRYPT_CLUSTERS
1483                                             * s->cluster_size);
1484                     if (cluster_data == NULL) {
1485                         ret = -ENOMEM;
1486                         goto fail;
1487                     }
1488                 }
1489
1490                 assert(cur_nr_sectors <=
1491                     QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1492                 qemu_iovec_reset(&hd_qiov);
1493                 qemu_iovec_add(&hd_qiov, cluster_data,
1494                     512 * cur_nr_sectors);
1495             }
1496
1497             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1498             qemu_co_mutex_unlock(&s->lock);
1499             ret = bdrv_co_readv(bs->file->bs,
1500                                 (cluster_offset >> 9) + index_in_cluster,
1501                                 cur_nr_sectors, &hd_qiov);
1502             qemu_co_mutex_lock(&s->lock);
1503             if (ret < 0) {
1504                 goto fail;
1505             }
1506             if (bs->encrypted) {
1507                 assert(s->cipher);
1508                 Error *err = NULL;
1509                 if (qcow2_encrypt_sectors(s, sector_num,  cluster_data,
1510                                           cluster_data, cur_nr_sectors, false,
1511                                           &err) < 0) {
1512                     error_free(err);
1513                     ret = -EIO;
1514                     goto fail;
1515                 }
1516                 qemu_iovec_from_buf(qiov, bytes_done,
1517                     cluster_data, 512 * cur_nr_sectors);
1518             }
1519             break;
1520
1521         default:
1522             g_assert_not_reached();
1523             ret = -EIO;
1524             goto fail;
1525         }
1526
1527         remaining_sectors -= cur_nr_sectors;
1528         sector_num += cur_nr_sectors;
1529         bytes_done += cur_nr_sectors * 512;
1530     }
1531     ret = 0;
1532
1533 fail:
1534     qemu_co_mutex_unlock(&s->lock);
1535
1536     qemu_iovec_destroy(&hd_qiov);
1537     qemu_vfree(cluster_data);
1538
1539     return ret;
1540 }
1541
1542 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
1543                            int64_t sector_num,
1544                            int remaining_sectors,
1545                            QEMUIOVector *qiov)
1546 {
1547     BDRVQcow2State *s = bs->opaque;
1548     int index_in_cluster;
1549     int ret;
1550     int cur_nr_sectors; /* number of sectors in current iteration */
1551     uint64_t cluster_offset;
1552     QEMUIOVector hd_qiov;
1553     uint64_t bytes_done = 0;
1554     uint8_t *cluster_data = NULL;
1555     QCowL2Meta *l2meta = NULL;
1556
1557     trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
1558                                  remaining_sectors);
1559
1560     qemu_iovec_init(&hd_qiov, qiov->niov);
1561
1562     s->cluster_cache_offset = -1; /* disable compressed cache */
1563
1564     qemu_co_mutex_lock(&s->lock);
1565
1566     while (remaining_sectors != 0) {
1567
1568         l2meta = NULL;
1569
1570         trace_qcow2_writev_start_part(qemu_coroutine_self());
1571         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1572         cur_nr_sectors = remaining_sectors;
1573         if (bs->encrypted &&
1574             cur_nr_sectors >
1575             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster) {
1576             cur_nr_sectors =
1577                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster;
1578         }
1579
1580         ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
1581             &cur_nr_sectors, &cluster_offset, &l2meta);
1582         if (ret < 0) {
1583             goto fail;
1584         }
1585
1586         assert((cluster_offset & 511) == 0);
1587
1588         qemu_iovec_reset(&hd_qiov);
1589         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1590             cur_nr_sectors * 512);
1591
1592         if (bs->encrypted) {
1593             Error *err = NULL;
1594             assert(s->cipher);
1595             if (!cluster_data) {
1596                 cluster_data = qemu_try_blockalign(bs->file->bs,
1597                                                    QCOW_MAX_CRYPT_CLUSTERS
1598                                                    * s->cluster_size);
1599                 if (cluster_data == NULL) {
1600                     ret = -ENOMEM;
1601                     goto fail;
1602                 }
1603             }
1604
1605             assert(hd_qiov.size <=
1606                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1607             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1608
1609             if (qcow2_encrypt_sectors(s, sector_num, cluster_data,
1610                                       cluster_data, cur_nr_sectors,
1611                                       true, &err) < 0) {
1612                 error_free(err);
1613                 ret = -EIO;
1614                 goto fail;
1615             }
1616
1617             qemu_iovec_reset(&hd_qiov);
1618             qemu_iovec_add(&hd_qiov, cluster_data,
1619                 cur_nr_sectors * 512);
1620         }
1621
1622         ret = qcow2_pre_write_overlap_check(bs, 0,
1623                 cluster_offset + index_in_cluster * BDRV_SECTOR_SIZE,
1624                 cur_nr_sectors * BDRV_SECTOR_SIZE);
1625         if (ret < 0) {
1626             goto fail;
1627         }
1628
1629         qemu_co_mutex_unlock(&s->lock);
1630         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1631         trace_qcow2_writev_data(qemu_coroutine_self(),
1632                                 (cluster_offset >> 9) + index_in_cluster);
1633         ret = bdrv_co_writev(bs->file->bs,
1634                              (cluster_offset >> 9) + index_in_cluster,
1635                              cur_nr_sectors, &hd_qiov);
1636         qemu_co_mutex_lock(&s->lock);
1637         if (ret < 0) {
1638             goto fail;
1639         }
1640
1641         while (l2meta != NULL) {
1642             QCowL2Meta *next;
1643
1644             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1645             if (ret < 0) {
1646                 goto fail;
1647             }
1648
1649             /* Take the request off the list of running requests */
1650             if (l2meta->nb_clusters != 0) {
1651                 QLIST_REMOVE(l2meta, next_in_flight);
1652             }
1653
1654             qemu_co_queue_restart_all(&l2meta->dependent_requests);
1655
1656             next = l2meta->next;
1657             g_free(l2meta);
1658             l2meta = next;
1659         }
1660
1661         remaining_sectors -= cur_nr_sectors;
1662         sector_num += cur_nr_sectors;
1663         bytes_done += cur_nr_sectors * 512;
1664         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
1665     }
1666     ret = 0;
1667
1668 fail:
1669     qemu_co_mutex_unlock(&s->lock);
1670
1671     while (l2meta != NULL) {
1672         QCowL2Meta *next;
1673
1674         if (l2meta->nb_clusters != 0) {
1675             QLIST_REMOVE(l2meta, next_in_flight);
1676         }
1677         qemu_co_queue_restart_all(&l2meta->dependent_requests);
1678
1679         next = l2meta->next;
1680         g_free(l2meta);
1681         l2meta = next;
1682     }
1683
1684     qemu_iovec_destroy(&hd_qiov);
1685     qemu_vfree(cluster_data);
1686     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
1687
1688     return ret;
1689 }
1690
1691 static int qcow2_inactivate(BlockDriverState *bs)
1692 {
1693     BDRVQcow2State *s = bs->opaque;
1694     int ret, result = 0;
1695
1696     ret = qcow2_cache_flush(bs, s->l2_table_cache);
1697     if (ret) {
1698         result = ret;
1699         error_report("Failed to flush the L2 table cache: %s",
1700                      strerror(-ret));
1701     }
1702
1703     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1704     if (ret) {
1705         result = ret;
1706         error_report("Failed to flush the refcount block cache: %s",
1707                      strerror(-ret));
1708     }
1709
1710     if (result == 0) {
1711         qcow2_mark_clean(bs);
1712     }
1713
1714     return result;
1715 }
1716
1717 static void qcow2_close(BlockDriverState *bs)
1718 {
1719     BDRVQcow2State *s = bs->opaque;
1720     qemu_vfree(s->l1_table);
1721     /* else pre-write overlap checks in cache_destroy may crash */
1722     s->l1_table = NULL;
1723
1724     if (!(s->flags & BDRV_O_INACTIVE)) {
1725         qcow2_inactivate(bs);
1726     }
1727
1728     cache_clean_timer_del(bs);
1729     qcow2_cache_destroy(bs, s->l2_table_cache);
1730     qcow2_cache_destroy(bs, s->refcount_block_cache);
1731
1732     qcrypto_cipher_free(s->cipher);
1733     s->cipher = NULL;
1734
1735     g_free(s->unknown_header_fields);
1736     cleanup_unknown_header_ext(bs);
1737
1738     g_free(s->image_backing_file);
1739     g_free(s->image_backing_format);
1740
1741     g_free(s->cluster_cache);
1742     qemu_vfree(s->cluster_data);
1743     qcow2_refcount_close(bs);
1744     qcow2_free_snapshots(bs);
1745 }
1746
1747 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
1748 {
1749     BDRVQcow2State *s = bs->opaque;
1750     int flags = s->flags;
1751     QCryptoCipher *cipher = NULL;
1752     QDict *options;
1753     Error *local_err = NULL;
1754     int ret;
1755
1756     /*
1757      * Backing files are read-only which makes all of their metadata immutable,
1758      * that means we don't have to worry about reopening them here.
1759      */
1760
1761     cipher = s->cipher;
1762     s->cipher = NULL;
1763
1764     qcow2_close(bs);
1765
1766     bdrv_invalidate_cache(bs->file->bs, &local_err);
1767     if (local_err) {
1768         error_propagate(errp, local_err);
1769         bs->drv = NULL;
1770         return;
1771     }
1772
1773     memset(s, 0, sizeof(BDRVQcow2State));
1774     options = qdict_clone_shallow(bs->options);
1775
1776     flags &= ~BDRV_O_INACTIVE;
1777     ret = qcow2_open(bs, options, flags, &local_err);
1778     QDECREF(options);
1779     if (local_err) {
1780         error_propagate(errp, local_err);
1781         error_prepend(errp, "Could not reopen qcow2 layer: ");
1782         bs->drv = NULL;
1783         return;
1784     } else if (ret < 0) {
1785         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
1786         bs->drv = NULL;
1787         return;
1788     }
1789
1790     s->cipher = cipher;
1791 }
1792
1793 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
1794     size_t len, size_t buflen)
1795 {
1796     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
1797     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
1798
1799     if (buflen < ext_len) {
1800         return -ENOSPC;
1801     }
1802
1803     *ext_backing_fmt = (QCowExtension) {
1804         .magic  = cpu_to_be32(magic),
1805         .len    = cpu_to_be32(len),
1806     };
1807     memcpy(buf + sizeof(QCowExtension), s, len);
1808
1809     return ext_len;
1810 }
1811
1812 /*
1813  * Updates the qcow2 header, including the variable length parts of it, i.e.
1814  * the backing file name and all extensions. qcow2 was not designed to allow
1815  * such changes, so if we run out of space (we can only use the first cluster)
1816  * this function may fail.
1817  *
1818  * Returns 0 on success, -errno in error cases.
1819  */
1820 int qcow2_update_header(BlockDriverState *bs)
1821 {
1822     BDRVQcow2State *s = bs->opaque;
1823     QCowHeader *header;
1824     char *buf;
1825     size_t buflen = s->cluster_size;
1826     int ret;
1827     uint64_t total_size;
1828     uint32_t refcount_table_clusters;
1829     size_t header_length;
1830     Qcow2UnknownHeaderExtension *uext;
1831
1832     buf = qemu_blockalign(bs, buflen);
1833
1834     /* Header structure */
1835     header = (QCowHeader*) buf;
1836
1837     if (buflen < sizeof(*header)) {
1838         ret = -ENOSPC;
1839         goto fail;
1840     }
1841
1842     header_length = sizeof(*header) + s->unknown_header_fields_size;
1843     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1844     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1845
1846     *header = (QCowHeader) {
1847         /* Version 2 fields */
1848         .magic                  = cpu_to_be32(QCOW_MAGIC),
1849         .version                = cpu_to_be32(s->qcow_version),
1850         .backing_file_offset    = 0,
1851         .backing_file_size      = 0,
1852         .cluster_bits           = cpu_to_be32(s->cluster_bits),
1853         .size                   = cpu_to_be64(total_size),
1854         .crypt_method           = cpu_to_be32(s->crypt_method_header),
1855         .l1_size                = cpu_to_be32(s->l1_size),
1856         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
1857         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
1858         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1859         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
1860         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
1861
1862         /* Version 3 fields */
1863         .incompatible_features  = cpu_to_be64(s->incompatible_features),
1864         .compatible_features    = cpu_to_be64(s->compatible_features),
1865         .autoclear_features     = cpu_to_be64(s->autoclear_features),
1866         .refcount_order         = cpu_to_be32(s->refcount_order),
1867         .header_length          = cpu_to_be32(header_length),
1868     };
1869
1870     /* For older versions, write a shorter header */
1871     switch (s->qcow_version) {
1872     case 2:
1873         ret = offsetof(QCowHeader, incompatible_features);
1874         break;
1875     case 3:
1876         ret = sizeof(*header);
1877         break;
1878     default:
1879         ret = -EINVAL;
1880         goto fail;
1881     }
1882
1883     buf += ret;
1884     buflen -= ret;
1885     memset(buf, 0, buflen);
1886
1887     /* Preserve any unknown field in the header */
1888     if (s->unknown_header_fields_size) {
1889         if (buflen < s->unknown_header_fields_size) {
1890             ret = -ENOSPC;
1891             goto fail;
1892         }
1893
1894         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1895         buf += s->unknown_header_fields_size;
1896         buflen -= s->unknown_header_fields_size;
1897     }
1898
1899     /* Backing file format header extension */
1900     if (s->image_backing_format) {
1901         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1902                              s->image_backing_format,
1903                              strlen(s->image_backing_format),
1904                              buflen);
1905         if (ret < 0) {
1906             goto fail;
1907         }
1908
1909         buf += ret;
1910         buflen -= ret;
1911     }
1912
1913     /* Feature table */
1914     if (s->qcow_version >= 3) {
1915         Qcow2Feature features[] = {
1916             {
1917                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1918                 .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
1919                 .name = "dirty bit",
1920             },
1921             {
1922                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1923                 .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
1924                 .name = "corrupt bit",
1925             },
1926             {
1927                 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1928                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1929                 .name = "lazy refcounts",
1930             },
1931         };
1932
1933         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1934                              features, sizeof(features), buflen);
1935         if (ret < 0) {
1936             goto fail;
1937         }
1938         buf += ret;
1939         buflen -= ret;
1940     }
1941
1942     /* Keep unknown header extensions */
1943     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1944         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1945         if (ret < 0) {
1946             goto fail;
1947         }
1948
1949         buf += ret;
1950         buflen -= ret;
1951     }
1952
1953     /* End of header extensions */
1954     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1955     if (ret < 0) {
1956         goto fail;
1957     }
1958
1959     buf += ret;
1960     buflen -= ret;
1961
1962     /* Backing file name */
1963     if (s->image_backing_file) {
1964         size_t backing_file_len = strlen(s->image_backing_file);
1965
1966         if (buflen < backing_file_len) {
1967             ret = -ENOSPC;
1968             goto fail;
1969         }
1970
1971         /* Using strncpy is ok here, since buf is not NUL-terminated. */
1972         strncpy(buf, s->image_backing_file, buflen);
1973
1974         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1975         header->backing_file_size   = cpu_to_be32(backing_file_len);
1976     }
1977
1978     /* Write the new header */
1979     ret = bdrv_pwrite(bs->file->bs, 0, header, s->cluster_size);
1980     if (ret < 0) {
1981         goto fail;
1982     }
1983
1984     ret = 0;
1985 fail:
1986     qemu_vfree(header);
1987     return ret;
1988 }
1989
1990 static int qcow2_change_backing_file(BlockDriverState *bs,
1991     const char *backing_file, const char *backing_fmt)
1992 {
1993     BDRVQcow2State *s = bs->opaque;
1994
1995     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1996     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1997
1998     g_free(s->image_backing_file);
1999     g_free(s->image_backing_format);
2000
2001     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
2002     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
2003
2004     return qcow2_update_header(bs);
2005 }
2006
2007 static int preallocate(BlockDriverState *bs)
2008 {
2009     uint64_t nb_sectors;
2010     uint64_t offset;
2011     uint64_t host_offset = 0;
2012     int num;
2013     int ret;
2014     QCowL2Meta *meta;
2015
2016     nb_sectors = bdrv_nb_sectors(bs);
2017     offset = 0;
2018
2019     while (nb_sectors) {
2020         num = MIN(nb_sectors, INT_MAX >> BDRV_SECTOR_BITS);
2021         ret = qcow2_alloc_cluster_offset(bs, offset, &num,
2022                                          &host_offset, &meta);
2023         if (ret < 0) {
2024             return ret;
2025         }
2026
2027         while (meta) {
2028             QCowL2Meta *next = meta->next;
2029
2030             ret = qcow2_alloc_cluster_link_l2(bs, meta);
2031             if (ret < 0) {
2032                 qcow2_free_any_clusters(bs, meta->alloc_offset,
2033                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
2034                 return ret;
2035             }
2036
2037             /* There are no dependent requests, but we need to remove our
2038              * request from the list of in-flight requests */
2039             QLIST_REMOVE(meta, next_in_flight);
2040
2041             g_free(meta);
2042             meta = next;
2043         }
2044
2045         /* TODO Preallocate data if requested */
2046
2047         nb_sectors -= num;
2048         offset += num << BDRV_SECTOR_BITS;
2049     }
2050
2051     /*
2052      * It is expected that the image file is large enough to actually contain
2053      * all of the allocated clusters (otherwise we get failing reads after
2054      * EOF). Extend the image to the last allocated sector.
2055      */
2056     if (host_offset != 0) {
2057         uint8_t buf[BDRV_SECTOR_SIZE];
2058         memset(buf, 0, BDRV_SECTOR_SIZE);
2059         ret = bdrv_write(bs->file->bs,
2060                          (host_offset >> BDRV_SECTOR_BITS) + num - 1,
2061                          buf, 1);
2062         if (ret < 0) {
2063             return ret;
2064         }
2065     }
2066
2067     return 0;
2068 }
2069
2070 static int qcow2_create2(const char *filename, int64_t total_size,
2071                          const char *backing_file, const char *backing_format,
2072                          int flags, size_t cluster_size, PreallocMode prealloc,
2073                          QemuOpts *opts, int version, int refcount_order,
2074                          Error **errp)
2075 {
2076     int cluster_bits;
2077     QDict *options;
2078
2079     /* Calculate cluster_bits */
2080     cluster_bits = ctz32(cluster_size);
2081     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
2082         (1 << cluster_bits) != cluster_size)
2083     {
2084         error_setg(errp, "Cluster size must be a power of two between %d and "
2085                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
2086         return -EINVAL;
2087     }
2088
2089     /*
2090      * Open the image file and write a minimal qcow2 header.
2091      *
2092      * We keep things simple and start with a zero-sized image. We also
2093      * do without refcount blocks or a L1 table for now. We'll fix the
2094      * inconsistency later.
2095      *
2096      * We do need a refcount table because growing the refcount table means
2097      * allocating two new refcount blocks - the seconds of which would be at
2098      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2099      * size for any qcow2 image.
2100      */
2101     BlockBackend *blk;
2102     QCowHeader *header;
2103     uint64_t* refcount_table;
2104     Error *local_err = NULL;
2105     int ret;
2106
2107     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
2108         /* Note: The following calculation does not need to be exact; if it is a
2109          * bit off, either some bytes will be "leaked" (which is fine) or we
2110          * will need to increase the file size by some bytes (which is fine,
2111          * too, as long as the bulk is allocated here). Therefore, using
2112          * floating point arithmetic is fine. */
2113         int64_t meta_size = 0;
2114         uint64_t nreftablee, nrefblocke, nl1e, nl2e;
2115         int64_t aligned_total_size = align_offset(total_size, cluster_size);
2116         int refblock_bits, refblock_size;
2117         /* refcount entry size in bytes */
2118         double rces = (1 << refcount_order) / 8.;
2119
2120         /* see qcow2_open() */
2121         refblock_bits = cluster_bits - (refcount_order - 3);
2122         refblock_size = 1 << refblock_bits;
2123
2124         /* header: 1 cluster */
2125         meta_size += cluster_size;
2126
2127         /* total size of L2 tables */
2128         nl2e = aligned_total_size / cluster_size;
2129         nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
2130         meta_size += nl2e * sizeof(uint64_t);
2131
2132         /* total size of L1 tables */
2133         nl1e = nl2e * sizeof(uint64_t) / cluster_size;
2134         nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
2135         meta_size += nl1e * sizeof(uint64_t);
2136
2137         /* total size of refcount blocks
2138          *
2139          * note: every host cluster is reference-counted, including metadata
2140          * (even refcount blocks are recursively included).
2141          * Let:
2142          *   a = total_size (this is the guest disk size)
2143          *   m = meta size not including refcount blocks and refcount tables
2144          *   c = cluster size
2145          *   y1 = number of refcount blocks entries
2146          *   y2 = meta size including everything
2147          *   rces = refcount entry size in bytes
2148          * then,
2149          *   y1 = (y2 + a)/c
2150          *   y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m
2151          * we can get y1:
2152          *   y1 = (a + m) / (c - rces - rces * sizeof(u64) / c)
2153          */
2154         nrefblocke = (aligned_total_size + meta_size + cluster_size)
2155                    / (cluster_size - rces - rces * sizeof(uint64_t)
2156                                                  / cluster_size);
2157         meta_size += DIV_ROUND_UP(nrefblocke, refblock_size) * cluster_size;
2158
2159         /* total size of refcount tables */
2160         nreftablee = nrefblocke / refblock_size;
2161         nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t));
2162         meta_size += nreftablee * sizeof(uint64_t);
2163
2164         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
2165                             aligned_total_size + meta_size, &error_abort);
2166         qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc],
2167                      &error_abort);
2168     }
2169
2170     ret = bdrv_create_file(filename, opts, &local_err);
2171     if (ret < 0) {
2172         error_propagate(errp, local_err);
2173         return ret;
2174     }
2175
2176     blk = blk_new_open("image", filename, NULL, NULL,
2177                        BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_PROTOCOL,
2178                        &local_err);
2179     if (blk == NULL) {
2180         error_propagate(errp, local_err);
2181         return -EIO;
2182     }
2183
2184     blk_set_allow_write_beyond_eof(blk, true);
2185
2186     /* Write the header */
2187     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
2188     header = g_malloc0(cluster_size);
2189     *header = (QCowHeader) {
2190         .magic                      = cpu_to_be32(QCOW_MAGIC),
2191         .version                    = cpu_to_be32(version),
2192         .cluster_bits               = cpu_to_be32(cluster_bits),
2193         .size                       = cpu_to_be64(0),
2194         .l1_table_offset            = cpu_to_be64(0),
2195         .l1_size                    = cpu_to_be32(0),
2196         .refcount_table_offset      = cpu_to_be64(cluster_size),
2197         .refcount_table_clusters    = cpu_to_be32(1),
2198         .refcount_order             = cpu_to_be32(refcount_order),
2199         .header_length              = cpu_to_be32(sizeof(*header)),
2200     };
2201
2202     if (flags & BLOCK_FLAG_ENCRYPT) {
2203         header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
2204     } else {
2205         header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
2206     }
2207
2208     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
2209         header->compatible_features |=
2210             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
2211     }
2212
2213     ret = blk_pwrite(blk, 0, header, cluster_size);
2214     g_free(header);
2215     if (ret < 0) {
2216         error_setg_errno(errp, -ret, "Could not write qcow2 header");
2217         goto out;
2218     }
2219
2220     /* Write a refcount table with one refcount block */
2221     refcount_table = g_malloc0(2 * cluster_size);
2222     refcount_table[0] = cpu_to_be64(2 * cluster_size);
2223     ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size);
2224     g_free(refcount_table);
2225
2226     if (ret < 0) {
2227         error_setg_errno(errp, -ret, "Could not write refcount table");
2228         goto out;
2229     }
2230
2231     blk_unref(blk);
2232     blk = NULL;
2233
2234     /*
2235      * And now open the image and make it consistent first (i.e. increase the
2236      * refcount of the cluster that is occupied by the header and the refcount
2237      * table)
2238      */
2239     options = qdict_new();
2240     qdict_put(options, "driver", qstring_from_str("qcow2"));
2241     blk = blk_new_open("image-qcow2", filename, NULL, options,
2242                        BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH,
2243                        &local_err);
2244     if (blk == NULL) {
2245         error_propagate(errp, local_err);
2246         ret = -EIO;
2247         goto out;
2248     }
2249
2250     ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
2251     if (ret < 0) {
2252         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
2253                          "header and refcount table");
2254         goto out;
2255
2256     } else if (ret != 0) {
2257         error_report("Huh, first cluster in empty image is already in use?");
2258         abort();
2259     }
2260
2261     /* Create a full header (including things like feature table) */
2262     ret = qcow2_update_header(blk_bs(blk));
2263     if (ret < 0) {
2264         error_setg_errno(errp, -ret, "Could not update qcow2 header");
2265         goto out;
2266     }
2267
2268     /* Okay, now that we have a valid image, let's give it the right size */
2269     ret = blk_truncate(blk, total_size);
2270     if (ret < 0) {
2271         error_setg_errno(errp, -ret, "Could not resize image");
2272         goto out;
2273     }
2274
2275     /* Want a backing file? There you go.*/
2276     if (backing_file) {
2277         ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format);
2278         if (ret < 0) {
2279             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
2280                              "with format '%s'", backing_file, backing_format);
2281             goto out;
2282         }
2283     }
2284
2285     /* And if we're supposed to preallocate metadata, do that now */
2286     if (prealloc != PREALLOC_MODE_OFF) {
2287         BDRVQcow2State *s = blk_bs(blk)->opaque;
2288         qemu_co_mutex_lock(&s->lock);
2289         ret = preallocate(blk_bs(blk));
2290         qemu_co_mutex_unlock(&s->lock);
2291         if (ret < 0) {
2292             error_setg_errno(errp, -ret, "Could not preallocate metadata");
2293             goto out;
2294         }
2295     }
2296
2297     blk_unref(blk);
2298     blk = NULL;
2299
2300     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
2301     options = qdict_new();
2302     qdict_put(options, "driver", qstring_from_str("qcow2"));
2303     blk = blk_new_open("image-flush", filename, NULL, options,
2304                        BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_BACKING,
2305                        &local_err);
2306     if (blk == NULL) {
2307         error_propagate(errp, local_err);
2308         ret = -EIO;
2309         goto out;
2310     }
2311
2312     ret = 0;
2313 out:
2314     if (blk) {
2315         blk_unref(blk);
2316     }
2317     return ret;
2318 }
2319
2320 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
2321 {
2322     char *backing_file = NULL;
2323     char *backing_fmt = NULL;
2324     char *buf = NULL;
2325     uint64_t size = 0;
2326     int flags = 0;
2327     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
2328     PreallocMode prealloc;
2329     int version = 3;
2330     uint64_t refcount_bits = 16;
2331     int refcount_order;
2332     Error *local_err = NULL;
2333     int ret;
2334
2335     /* Read out options */
2336     size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2337                     BDRV_SECTOR_SIZE);
2338     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2339     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2340     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2341         flags |= BLOCK_FLAG_ENCRYPT;
2342     }
2343     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2344                                          DEFAULT_CLUSTER_SIZE);
2345     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2346     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
2347                                PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
2348                                &local_err);
2349     if (local_err) {
2350         error_propagate(errp, local_err);
2351         ret = -EINVAL;
2352         goto finish;
2353     }
2354     g_free(buf);
2355     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2356     if (!buf) {
2357         /* keep the default */
2358     } else if (!strcmp(buf, "0.10")) {
2359         version = 2;
2360     } else if (!strcmp(buf, "1.1")) {
2361         version = 3;
2362     } else {
2363         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2364         ret = -EINVAL;
2365         goto finish;
2366     }
2367
2368     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2369         flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2370     }
2371
2372     if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2373         error_setg(errp, "Backing file and preallocation cannot be used at "
2374                    "the same time");
2375         ret = -EINVAL;
2376         goto finish;
2377     }
2378
2379     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2380         error_setg(errp, "Lazy refcounts only supported with compatibility "
2381                    "level 1.1 and above (use compat=1.1 or greater)");
2382         ret = -EINVAL;
2383         goto finish;
2384     }
2385
2386     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS,
2387                                             refcount_bits);
2388     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2389         error_setg(errp, "Refcount width must be a power of two and may not "
2390                    "exceed 64 bits");
2391         ret = -EINVAL;
2392         goto finish;
2393     }
2394
2395     if (version < 3 && refcount_bits != 16) {
2396         error_setg(errp, "Different refcount widths than 16 bits require "
2397                    "compatibility level 1.1 or above (use compat=1.1 or "
2398                    "greater)");
2399         ret = -EINVAL;
2400         goto finish;
2401     }
2402
2403     refcount_order = ctz32(refcount_bits);
2404
2405     ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2406                         cluster_size, prealloc, opts, version, refcount_order,
2407                         &local_err);
2408     if (local_err) {
2409         error_propagate(errp, local_err);
2410     }
2411
2412 finish:
2413     g_free(backing_file);
2414     g_free(backing_fmt);
2415     g_free(buf);
2416     return ret;
2417 }
2418
2419 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
2420     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2421 {
2422     int ret;
2423     BDRVQcow2State *s = bs->opaque;
2424
2425     /* Emulate misaligned zero writes */
2426     if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
2427         return -ENOTSUP;
2428     }
2429
2430     /* Whatever is left can use real zero clusters */
2431     qemu_co_mutex_lock(&s->lock);
2432     ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2433         nb_sectors);
2434     qemu_co_mutex_unlock(&s->lock);
2435
2436     return ret;
2437 }
2438
2439 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
2440     int64_t sector_num, int nb_sectors)
2441 {
2442     int ret;
2443     BDRVQcow2State *s = bs->opaque;
2444
2445     qemu_co_mutex_lock(&s->lock);
2446     ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2447         nb_sectors, QCOW2_DISCARD_REQUEST, false);
2448     qemu_co_mutex_unlock(&s->lock);
2449     return ret;
2450 }
2451
2452 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
2453 {
2454     BDRVQcow2State *s = bs->opaque;
2455     int64_t new_l1_size;
2456     int ret;
2457
2458     if (offset & 511) {
2459         error_report("The new size must be a multiple of 512");
2460         return -EINVAL;
2461     }
2462
2463     /* cannot proceed if image has snapshots */
2464     if (s->nb_snapshots) {
2465         error_report("Can't resize an image which has snapshots");
2466         return -ENOTSUP;
2467     }
2468
2469     /* shrinking is currently not supported */
2470     if (offset < bs->total_sectors * 512) {
2471         error_report("qcow2 doesn't support shrinking images yet");
2472         return -ENOTSUP;
2473     }
2474
2475     new_l1_size = size_to_l1(s, offset);
2476     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
2477     if (ret < 0) {
2478         return ret;
2479     }
2480
2481     /* write updated header.size */
2482     offset = cpu_to_be64(offset);
2483     ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, size),
2484                            &offset, sizeof(uint64_t));
2485     if (ret < 0) {
2486         return ret;
2487     }
2488
2489     s->l1_vm_state_index = new_l1_size;
2490     return 0;
2491 }
2492
2493 /* XXX: put compressed sectors first, then all the cluster aligned
2494    tables to avoid losing bytes in alignment */
2495 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
2496                                   const uint8_t *buf, int nb_sectors)
2497 {
2498     BDRVQcow2State *s = bs->opaque;
2499     z_stream strm;
2500     int ret, out_len;
2501     uint8_t *out_buf;
2502     uint64_t cluster_offset;
2503
2504     if (nb_sectors == 0) {
2505         /* align end of file to a sector boundary to ease reading with
2506            sector based I/Os */
2507         cluster_offset = bdrv_getlength(bs->file->bs);
2508         return bdrv_truncate(bs->file->bs, cluster_offset);
2509     }
2510
2511     if (nb_sectors != s->cluster_sectors) {
2512         ret = -EINVAL;
2513
2514         /* Zero-pad last write if image size is not cluster aligned */
2515         if (sector_num + nb_sectors == bs->total_sectors &&
2516             nb_sectors < s->cluster_sectors) {
2517             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
2518             memset(pad_buf, 0, s->cluster_size);
2519             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
2520             ret = qcow2_write_compressed(bs, sector_num,
2521                                          pad_buf, s->cluster_sectors);
2522             qemu_vfree(pad_buf);
2523         }
2524         return ret;
2525     }
2526
2527     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
2528
2529     /* best compression, small window, no zlib header */
2530     memset(&strm, 0, sizeof(strm));
2531     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
2532                        Z_DEFLATED, -12,
2533                        9, Z_DEFAULT_STRATEGY);
2534     if (ret != 0) {
2535         ret = -EINVAL;
2536         goto fail;
2537     }
2538
2539     strm.avail_in = s->cluster_size;
2540     strm.next_in = (uint8_t *)buf;
2541     strm.avail_out = s->cluster_size;
2542     strm.next_out = out_buf;
2543
2544     ret = deflate(&strm, Z_FINISH);
2545     if (ret != Z_STREAM_END && ret != Z_OK) {
2546         deflateEnd(&strm);
2547         ret = -EINVAL;
2548         goto fail;
2549     }
2550     out_len = strm.next_out - out_buf;
2551
2552     deflateEnd(&strm);
2553
2554     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
2555         /* could not compress: write normal cluster */
2556         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
2557         if (ret < 0) {
2558             goto fail;
2559         }
2560     } else {
2561         cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
2562             sector_num << 9, out_len);
2563         if (!cluster_offset) {
2564             ret = -EIO;
2565             goto fail;
2566         }
2567         cluster_offset &= s->cluster_offset_mask;
2568
2569         ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
2570         if (ret < 0) {
2571             goto fail;
2572         }
2573
2574         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
2575         ret = bdrv_pwrite(bs->file->bs, cluster_offset, out_buf, out_len);
2576         if (ret < 0) {
2577             goto fail;
2578         }
2579     }
2580
2581     ret = 0;
2582 fail:
2583     g_free(out_buf);
2584     return ret;
2585 }
2586
2587 static int make_completely_empty(BlockDriverState *bs)
2588 {
2589     BDRVQcow2State *s = bs->opaque;
2590     int ret, l1_clusters;
2591     int64_t offset;
2592     uint64_t *new_reftable = NULL;
2593     uint64_t rt_entry, l1_size2;
2594     struct {
2595         uint64_t l1_offset;
2596         uint64_t reftable_offset;
2597         uint32_t reftable_clusters;
2598     } QEMU_PACKED l1_ofs_rt_ofs_cls;
2599
2600     ret = qcow2_cache_empty(bs, s->l2_table_cache);
2601     if (ret < 0) {
2602         goto fail;
2603     }
2604
2605     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
2606     if (ret < 0) {
2607         goto fail;
2608     }
2609
2610     /* Refcounts will be broken utterly */
2611     ret = qcow2_mark_dirty(bs);
2612     if (ret < 0) {
2613         goto fail;
2614     }
2615
2616     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2617
2618     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2619     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
2620
2621     /* After this call, neither the in-memory nor the on-disk refcount
2622      * information accurately describe the actual references */
2623
2624     ret = bdrv_write_zeroes(bs->file->bs, s->l1_table_offset / BDRV_SECTOR_SIZE,
2625                             l1_clusters * s->cluster_sectors, 0);
2626     if (ret < 0) {
2627         goto fail_broken_refcounts;
2628     }
2629     memset(s->l1_table, 0, l1_size2);
2630
2631     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
2632
2633     /* Overwrite enough clusters at the beginning of the sectors to place
2634      * the refcount table, a refcount block and the L1 table in; this may
2635      * overwrite parts of the existing refcount and L1 table, which is not
2636      * an issue because the dirty flag is set, complete data loss is in fact
2637      * desired and partial data loss is consequently fine as well */
2638     ret = bdrv_write_zeroes(bs->file->bs, s->cluster_size / BDRV_SECTOR_SIZE,
2639                             (2 + l1_clusters) * s->cluster_size /
2640                             BDRV_SECTOR_SIZE, 0);
2641     /* This call (even if it failed overall) may have overwritten on-disk
2642      * refcount structures; in that case, the in-memory refcount information
2643      * will probably differ from the on-disk information which makes the BDS
2644      * unusable */
2645     if (ret < 0) {
2646         goto fail_broken_refcounts;
2647     }
2648
2649     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2650     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
2651
2652     /* "Create" an empty reftable (one cluster) directly after the image
2653      * header and an empty L1 table three clusters after the image header;
2654      * the cluster between those two will be used as the first refblock */
2655     cpu_to_be64w(&l1_ofs_rt_ofs_cls.l1_offset, 3 * s->cluster_size);
2656     cpu_to_be64w(&l1_ofs_rt_ofs_cls.reftable_offset, s->cluster_size);
2657     cpu_to_be32w(&l1_ofs_rt_ofs_cls.reftable_clusters, 1);
2658     ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, l1_table_offset),
2659                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
2660     if (ret < 0) {
2661         goto fail_broken_refcounts;
2662     }
2663
2664     s->l1_table_offset = 3 * s->cluster_size;
2665
2666     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
2667     if (!new_reftable) {
2668         ret = -ENOMEM;
2669         goto fail_broken_refcounts;
2670     }
2671
2672     s->refcount_table_offset = s->cluster_size;
2673     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
2674
2675     g_free(s->refcount_table);
2676     s->refcount_table = new_reftable;
2677     new_reftable = NULL;
2678
2679     /* Now the in-memory refcount information again corresponds to the on-disk
2680      * information (reftable is empty and no refblocks (the refblock cache is
2681      * empty)); however, this means some clusters (e.g. the image header) are
2682      * referenced, but not refcounted, but the normal qcow2 code assumes that
2683      * the in-memory information is always correct */
2684
2685     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
2686
2687     /* Enter the first refblock into the reftable */
2688     rt_entry = cpu_to_be64(2 * s->cluster_size);
2689     ret = bdrv_pwrite_sync(bs->file->bs, s->cluster_size,
2690                            &rt_entry, sizeof(rt_entry));
2691     if (ret < 0) {
2692         goto fail_broken_refcounts;
2693     }
2694     s->refcount_table[0] = 2 * s->cluster_size;
2695
2696     s->free_cluster_index = 0;
2697     assert(3 + l1_clusters <= s->refcount_block_size);
2698     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
2699     if (offset < 0) {
2700         ret = offset;
2701         goto fail_broken_refcounts;
2702     } else if (offset > 0) {
2703         error_report("First cluster in emptied image is in use");
2704         abort();
2705     }
2706
2707     /* Now finally the in-memory information corresponds to the on-disk
2708      * structures and is correct */
2709     ret = qcow2_mark_clean(bs);
2710     if (ret < 0) {
2711         goto fail;
2712     }
2713
2714     ret = bdrv_truncate(bs->file->bs, (3 + l1_clusters) * s->cluster_size);
2715     if (ret < 0) {
2716         goto fail;
2717     }
2718
2719     return 0;
2720
2721 fail_broken_refcounts:
2722     /* The BDS is unusable at this point. If we wanted to make it usable, we
2723      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
2724      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
2725      * again. However, because the functions which could have caused this error
2726      * path to be taken are used by those functions as well, it's very likely
2727      * that that sequence will fail as well. Therefore, just eject the BDS. */
2728     bs->drv = NULL;
2729
2730 fail:
2731     g_free(new_reftable);
2732     return ret;
2733 }
2734
2735 static int qcow2_make_empty(BlockDriverState *bs)
2736 {
2737     BDRVQcow2State *s = bs->opaque;
2738     uint64_t start_sector;
2739     int sector_step = INT_MAX / BDRV_SECTOR_SIZE;
2740     int l1_clusters, ret = 0;
2741
2742     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2743
2744     if (s->qcow_version >= 3 && !s->snapshots &&
2745         3 + l1_clusters <= s->refcount_block_size) {
2746         /* The following function only works for qcow2 v3 images (it requires
2747          * the dirty flag) and only as long as there are no snapshots (because
2748          * it completely empties the image). Furthermore, the L1 table and three
2749          * additional clusters (image header, refcount table, one refcount
2750          * block) have to fit inside one refcount block. */
2751         return make_completely_empty(bs);
2752     }
2753
2754     /* This fallback code simply discards every active cluster; this is slow,
2755      * but works in all cases */
2756     for (start_sector = 0; start_sector < bs->total_sectors;
2757          start_sector += sector_step)
2758     {
2759         /* As this function is generally used after committing an external
2760          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
2761          * default action for this kind of discard is to pass the discard,
2762          * which will ideally result in an actually smaller image file, as
2763          * is probably desired. */
2764         ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE,
2765                                      MIN(sector_step,
2766                                          bs->total_sectors - start_sector),
2767                                      QCOW2_DISCARD_SNAPSHOT, true);
2768         if (ret < 0) {
2769             break;
2770         }
2771     }
2772
2773     return ret;
2774 }
2775
2776 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
2777 {
2778     BDRVQcow2State *s = bs->opaque;
2779     int ret;
2780
2781     qemu_co_mutex_lock(&s->lock);
2782     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2783     if (ret < 0) {
2784         qemu_co_mutex_unlock(&s->lock);
2785         return ret;
2786     }
2787
2788     if (qcow2_need_accurate_refcounts(s)) {
2789         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2790         if (ret < 0) {
2791             qemu_co_mutex_unlock(&s->lock);
2792             return ret;
2793         }
2794     }
2795     qemu_co_mutex_unlock(&s->lock);
2796
2797     return 0;
2798 }
2799
2800 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2801 {
2802     BDRVQcow2State *s = bs->opaque;
2803     bdi->unallocated_blocks_are_zero = true;
2804     bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
2805     bdi->cluster_size = s->cluster_size;
2806     bdi->vm_state_offset = qcow2_vm_state_offset(s);
2807     return 0;
2808 }
2809
2810 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
2811 {
2812     BDRVQcow2State *s = bs->opaque;
2813     ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
2814
2815     *spec_info = (ImageInfoSpecific){
2816         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
2817         .u.qcow2 = g_new(ImageInfoSpecificQCow2, 1),
2818     };
2819     if (s->qcow_version == 2) {
2820         *spec_info->u.qcow2 = (ImageInfoSpecificQCow2){
2821             .compat             = g_strdup("0.10"),
2822             .refcount_bits      = s->refcount_bits,
2823         };
2824     } else if (s->qcow_version == 3) {
2825         *spec_info->u.qcow2 = (ImageInfoSpecificQCow2){
2826             .compat             = g_strdup("1.1"),
2827             .lazy_refcounts     = s->compatible_features &
2828                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
2829             .has_lazy_refcounts = true,
2830             .corrupt            = s->incompatible_features &
2831                                   QCOW2_INCOMPAT_CORRUPT,
2832             .has_corrupt        = true,
2833             .refcount_bits      = s->refcount_bits,
2834         };
2835     } else {
2836         /* if this assertion fails, this probably means a new version was
2837          * added without having it covered here */
2838         assert(false);
2839     }
2840
2841     return spec_info;
2842 }
2843
2844 #if 0
2845 static void dump_refcounts(BlockDriverState *bs)
2846 {
2847     BDRVQcow2State *s = bs->opaque;
2848     int64_t nb_clusters, k, k1, size;
2849     int refcount;
2850
2851     size = bdrv_getlength(bs->file->bs);
2852     nb_clusters = size_to_clusters(s, size);
2853     for(k = 0; k < nb_clusters;) {
2854         k1 = k;
2855         refcount = get_refcount(bs, k);
2856         k++;
2857         while (k < nb_clusters && get_refcount(bs, k) == refcount)
2858             k++;
2859         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
2860                k - k1);
2861     }
2862 }
2863 #endif
2864
2865 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2866                               int64_t pos)
2867 {
2868     BDRVQcow2State *s = bs->opaque;
2869     int64_t total_sectors = bs->total_sectors;
2870     bool zero_beyond_eof = bs->zero_beyond_eof;
2871     int ret;
2872
2873     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
2874     bs->zero_beyond_eof = false;
2875     ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov);
2876     bs->zero_beyond_eof = zero_beyond_eof;
2877
2878     /* bdrv_co_do_writev will have increased the total_sectors value to include
2879      * the VM state - the VM state is however not an actual part of the block
2880      * device, therefore, we need to restore the old value. */
2881     bs->total_sectors = total_sectors;
2882
2883     return ret;
2884 }
2885
2886 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2887                               int64_t pos, int size)
2888 {
2889     BDRVQcow2State *s = bs->opaque;
2890     bool zero_beyond_eof = bs->zero_beyond_eof;
2891     int ret;
2892
2893     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
2894     bs->zero_beyond_eof = false;
2895     ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
2896     bs->zero_beyond_eof = zero_beyond_eof;
2897
2898     return ret;
2899 }
2900
2901 /*
2902  * Downgrades an image's version. To achieve this, any incompatible features
2903  * have to be removed.
2904  */
2905 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
2906                            BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
2907 {
2908     BDRVQcow2State *s = bs->opaque;
2909     int current_version = s->qcow_version;
2910     int ret;
2911
2912     if (target_version == current_version) {
2913         return 0;
2914     } else if (target_version > current_version) {
2915         return -EINVAL;
2916     } else if (target_version != 2) {
2917         return -EINVAL;
2918     }
2919
2920     if (s->refcount_order != 4) {
2921         error_report("compat=0.10 requires refcount_bits=16");
2922         return -ENOTSUP;
2923     }
2924
2925     /* clear incompatible features */
2926     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
2927         ret = qcow2_mark_clean(bs);
2928         if (ret < 0) {
2929             return ret;
2930         }
2931     }
2932
2933     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2934      * the first place; if that happens nonetheless, returning -ENOTSUP is the
2935      * best thing to do anyway */
2936
2937     if (s->incompatible_features) {
2938         return -ENOTSUP;
2939     }
2940
2941     /* since we can ignore compatible features, we can set them to 0 as well */
2942     s->compatible_features = 0;
2943     /* if lazy refcounts have been used, they have already been fixed through
2944      * clearing the dirty flag */
2945
2946     /* clearing autoclear features is trivial */
2947     s->autoclear_features = 0;
2948
2949     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
2950     if (ret < 0) {
2951         return ret;
2952     }
2953
2954     s->qcow_version = target_version;
2955     ret = qcow2_update_header(bs);
2956     if (ret < 0) {
2957         s->qcow_version = current_version;
2958         return ret;
2959     }
2960     return 0;
2961 }
2962
2963 typedef enum Qcow2AmendOperation {
2964     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
2965      * statically initialized to so that the helper CB can discern the first
2966      * invocation from an operation change */
2967     QCOW2_NO_OPERATION = 0,
2968
2969     QCOW2_CHANGING_REFCOUNT_ORDER,
2970     QCOW2_DOWNGRADING,
2971 } Qcow2AmendOperation;
2972
2973 typedef struct Qcow2AmendHelperCBInfo {
2974     /* The code coordinating the amend operations should only modify
2975      * these four fields; the rest will be managed by the CB */
2976     BlockDriverAmendStatusCB *original_status_cb;
2977     void *original_cb_opaque;
2978
2979     Qcow2AmendOperation current_operation;
2980
2981     /* Total number of operations to perform (only set once) */
2982     int total_operations;
2983
2984     /* The following fields are managed by the CB */
2985
2986     /* Number of operations completed */
2987     int operations_completed;
2988
2989     /* Cumulative offset of all completed operations */
2990     int64_t offset_completed;
2991
2992     Qcow2AmendOperation last_operation;
2993     int64_t last_work_size;
2994 } Qcow2AmendHelperCBInfo;
2995
2996 static void qcow2_amend_helper_cb(BlockDriverState *bs,
2997                                   int64_t operation_offset,
2998                                   int64_t operation_work_size, void *opaque)
2999 {
3000     Qcow2AmendHelperCBInfo *info = opaque;
3001     int64_t current_work_size;
3002     int64_t projected_work_size;
3003
3004     if (info->current_operation != info->last_operation) {
3005         if (info->last_operation != QCOW2_NO_OPERATION) {
3006             info->offset_completed += info->last_work_size;
3007             info->operations_completed++;
3008         }
3009
3010         info->last_operation = info->current_operation;
3011     }
3012
3013     assert(info->total_operations > 0);
3014     assert(info->operations_completed < info->total_operations);
3015
3016     info->last_work_size = operation_work_size;
3017
3018     current_work_size = info->offset_completed + operation_work_size;
3019
3020     /* current_work_size is the total work size for (operations_completed + 1)
3021      * operations (which includes this one), so multiply it by the number of
3022      * operations not covered and divide it by the number of operations
3023      * covered to get a projection for the operations not covered */
3024     projected_work_size = current_work_size * (info->total_operations -
3025                                                info->operations_completed - 1)
3026                                             / (info->operations_completed + 1);
3027
3028     info->original_status_cb(bs, info->offset_completed + operation_offset,
3029                              current_work_size + projected_work_size,
3030                              info->original_cb_opaque);
3031 }
3032
3033 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
3034                                BlockDriverAmendStatusCB *status_cb,
3035                                void *cb_opaque)
3036 {
3037     BDRVQcow2State *s = bs->opaque;
3038     int old_version = s->qcow_version, new_version = old_version;
3039     uint64_t new_size = 0;
3040     const char *backing_file = NULL, *backing_format = NULL;
3041     bool lazy_refcounts = s->use_lazy_refcounts;
3042     const char *compat = NULL;
3043     uint64_t cluster_size = s->cluster_size;
3044     bool encrypt;
3045     int refcount_bits = s->refcount_bits;
3046     int ret;
3047     QemuOptDesc *desc = opts->list->desc;
3048     Qcow2AmendHelperCBInfo helper_cb_info;
3049
3050     while (desc && desc->name) {
3051         if (!qemu_opt_find(opts, desc->name)) {
3052             /* only change explicitly defined options */
3053             desc++;
3054             continue;
3055         }
3056
3057         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
3058             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
3059             if (!compat) {
3060                 /* preserve default */
3061             } else if (!strcmp(compat, "0.10")) {
3062                 new_version = 2;
3063             } else if (!strcmp(compat, "1.1")) {
3064                 new_version = 3;
3065             } else {
3066                 error_report("Unknown compatibility level %s", compat);
3067                 return -EINVAL;
3068             }
3069         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
3070             error_report("Cannot change preallocation mode");
3071             return -ENOTSUP;
3072         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
3073             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
3074         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
3075             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
3076         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
3077             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
3078         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
3079             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
3080                                         !!s->cipher);
3081
3082             if (encrypt != !!s->cipher) {
3083                 error_report("Changing the encryption flag is not supported");
3084                 return -ENOTSUP;
3085             }
3086         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
3087             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
3088                                              cluster_size);
3089             if (cluster_size != s->cluster_size) {
3090                 error_report("Changing the cluster size is not supported");
3091                 return -ENOTSUP;
3092             }
3093         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
3094             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
3095                                                lazy_refcounts);
3096         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
3097             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
3098                                                 refcount_bits);
3099
3100             if (refcount_bits <= 0 || refcount_bits > 64 ||
3101                 !is_power_of_2(refcount_bits))
3102             {
3103                 error_report("Refcount width must be a power of two and may "
3104                              "not exceed 64 bits");
3105                 return -EINVAL;
3106             }
3107         } else {
3108             /* if this point is reached, this probably means a new option was
3109              * added without having it covered here */
3110             abort();
3111         }
3112
3113         desc++;
3114     }
3115
3116     helper_cb_info = (Qcow2AmendHelperCBInfo){
3117         .original_status_cb = status_cb,
3118         .original_cb_opaque = cb_opaque,
3119         .total_operations = (new_version < old_version)
3120                           + (s->refcount_bits != refcount_bits)
3121     };
3122
3123     /* Upgrade first (some features may require compat=1.1) */
3124     if (new_version > old_version) {
3125         s->qcow_version = new_version;
3126         ret = qcow2_update_header(bs);
3127         if (ret < 0) {
3128             s->qcow_version = old_version;
3129             return ret;
3130         }
3131     }
3132
3133     if (s->refcount_bits != refcount_bits) {
3134         int refcount_order = ctz32(refcount_bits);
3135         Error *local_error = NULL;
3136
3137         if (new_version < 3 && refcount_bits != 16) {
3138             error_report("Different refcount widths than 16 bits require "
3139                          "compatibility level 1.1 or above (use compat=1.1 or "
3140                          "greater)");
3141             return -EINVAL;
3142         }
3143
3144         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
3145         ret = qcow2_change_refcount_order(bs, refcount_order,
3146                                           &qcow2_amend_helper_cb,
3147                                           &helper_cb_info, &local_error);
3148         if (ret < 0) {
3149             error_report_err(local_error);
3150             return ret;
3151         }
3152     }
3153
3154     if (backing_file || backing_format) {
3155         ret = qcow2_change_backing_file(bs,
3156                     backing_file ?: s->image_backing_file,
3157                     backing_format ?: s->image_backing_format);
3158         if (ret < 0) {
3159             return ret;
3160         }
3161     }
3162
3163     if (s->use_lazy_refcounts != lazy_refcounts) {
3164         if (lazy_refcounts) {
3165             if (new_version < 3) {
3166                 error_report("Lazy refcounts only supported with compatibility "
3167                              "level 1.1 and above (use compat=1.1 or greater)");
3168                 return -EINVAL;
3169             }
3170             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3171             ret = qcow2_update_header(bs);
3172             if (ret < 0) {
3173                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3174                 return ret;
3175             }
3176             s->use_lazy_refcounts = true;
3177         } else {
3178             /* make image clean first */
3179             ret = qcow2_mark_clean(bs);
3180             if (ret < 0) {
3181                 return ret;
3182             }
3183             /* now disallow lazy refcounts */
3184             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3185             ret = qcow2_update_header(bs);
3186             if (ret < 0) {
3187                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3188                 return ret;
3189             }
3190             s->use_lazy_refcounts = false;
3191         }
3192     }
3193
3194     if (new_size) {
3195         ret = bdrv_truncate(bs, new_size);
3196         if (ret < 0) {
3197             return ret;
3198         }
3199     }
3200
3201     /* Downgrade last (so unsupported features can be removed before) */
3202     if (new_version < old_version) {
3203         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
3204         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
3205                               &helper_cb_info);
3206         if (ret < 0) {
3207             return ret;
3208         }
3209     }
3210
3211     return 0;
3212 }
3213
3214 /*
3215  * If offset or size are negative, respectively, they will not be included in
3216  * the BLOCK_IMAGE_CORRUPTED event emitted.
3217  * fatal will be ignored for read-only BDS; corruptions found there will always
3218  * be considered non-fatal.
3219  */
3220 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
3221                              int64_t size, const char *message_format, ...)
3222 {
3223     BDRVQcow2State *s = bs->opaque;
3224     const char *node_name;
3225     char *message;
3226     va_list ap;
3227
3228     fatal = fatal && !bs->read_only;
3229
3230     if (s->signaled_corruption &&
3231         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
3232     {
3233         return;
3234     }
3235
3236     va_start(ap, message_format);
3237     message = g_strdup_vprintf(message_format, ap);
3238     va_end(ap);
3239
3240     if (fatal) {
3241         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
3242                 "corruption events will be suppressed\n", message);
3243     } else {
3244         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
3245                 "corruption events will be suppressed\n", message);
3246     }
3247
3248     node_name = bdrv_get_node_name(bs);
3249     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
3250                                           *node_name != '\0', node_name,
3251                                           message, offset >= 0, offset,
3252                                           size >= 0, size,
3253                                           fatal, &error_abort);
3254     g_free(message);
3255
3256     if (fatal) {
3257         qcow2_mark_corrupt(bs);
3258         bs->drv = NULL; /* make BDS unusable */
3259     }
3260
3261     s->signaled_corruption = true;
3262 }
3263
3264 static QemuOptsList qcow2_create_opts = {
3265     .name = "qcow2-create-opts",
3266     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
3267     .desc = {
3268         {
3269             .name = BLOCK_OPT_SIZE,
3270             .type = QEMU_OPT_SIZE,
3271             .help = "Virtual disk size"
3272         },
3273         {
3274             .name = BLOCK_OPT_COMPAT_LEVEL,
3275             .type = QEMU_OPT_STRING,
3276             .help = "Compatibility level (0.10 or 1.1)"
3277         },
3278         {
3279             .name = BLOCK_OPT_BACKING_FILE,
3280             .type = QEMU_OPT_STRING,
3281             .help = "File name of a base image"
3282         },
3283         {
3284             .name = BLOCK_OPT_BACKING_FMT,
3285             .type = QEMU_OPT_STRING,
3286             .help = "Image format of the base image"
3287         },
3288         {
3289             .name = BLOCK_OPT_ENCRYPT,
3290             .type = QEMU_OPT_BOOL,
3291             .help = "Encrypt the image",
3292             .def_value_str = "off"
3293         },
3294         {
3295             .name = BLOCK_OPT_CLUSTER_SIZE,
3296             .type = QEMU_OPT_SIZE,
3297             .help = "qcow2 cluster size",
3298             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
3299         },
3300         {
3301             .name = BLOCK_OPT_PREALLOC,
3302             .type = QEMU_OPT_STRING,
3303             .help = "Preallocation mode (allowed values: off, metadata, "
3304                     "falloc, full)"
3305         },
3306         {
3307             .name = BLOCK_OPT_LAZY_REFCOUNTS,
3308             .type = QEMU_OPT_BOOL,
3309             .help = "Postpone refcount updates",
3310             .def_value_str = "off"
3311         },
3312         {
3313             .name = BLOCK_OPT_REFCOUNT_BITS,
3314             .type = QEMU_OPT_NUMBER,
3315             .help = "Width of a reference count entry in bits",
3316             .def_value_str = "16"
3317         },
3318         { /* end of list */ }
3319     }
3320 };
3321
3322 BlockDriver bdrv_qcow2 = {
3323     .format_name        = "qcow2",
3324     .instance_size      = sizeof(BDRVQcow2State),
3325     .bdrv_probe         = qcow2_probe,
3326     .bdrv_open          = qcow2_open,
3327     .bdrv_close         = qcow2_close,
3328     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
3329     .bdrv_reopen_commit   = qcow2_reopen_commit,
3330     .bdrv_reopen_abort    = qcow2_reopen_abort,
3331     .bdrv_join_options    = qcow2_join_options,
3332     .bdrv_create        = qcow2_create,
3333     .bdrv_has_zero_init = bdrv_has_zero_init_1,
3334     .bdrv_co_get_block_status = qcow2_co_get_block_status,
3335     .bdrv_set_key       = qcow2_set_key,
3336
3337     .bdrv_co_readv          = qcow2_co_readv,
3338     .bdrv_co_writev         = qcow2_co_writev,
3339     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
3340
3341     .bdrv_co_write_zeroes   = qcow2_co_write_zeroes,
3342     .bdrv_co_discard        = qcow2_co_discard,
3343     .bdrv_truncate          = qcow2_truncate,
3344     .bdrv_write_compressed  = qcow2_write_compressed,
3345     .bdrv_make_empty        = qcow2_make_empty,
3346
3347     .bdrv_snapshot_create   = qcow2_snapshot_create,
3348     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
3349     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
3350     .bdrv_snapshot_list     = qcow2_snapshot_list,
3351     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
3352     .bdrv_get_info          = qcow2_get_info,
3353     .bdrv_get_specific_info = qcow2_get_specific_info,
3354
3355     .bdrv_save_vmstate    = qcow2_save_vmstate,
3356     .bdrv_load_vmstate    = qcow2_load_vmstate,
3357
3358     .supports_backing           = true,
3359     .bdrv_change_backing_file   = qcow2_change_backing_file,
3360
3361     .bdrv_refresh_limits        = qcow2_refresh_limits,
3362     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
3363     .bdrv_inactivate            = qcow2_inactivate,
3364
3365     .create_opts         = &qcow2_create_opts,
3366     .bdrv_check          = qcow2_check,
3367     .bdrv_amend_options  = qcow2_amend_options,
3368
3369     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
3370     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
3371 };
3372
3373 static void bdrv_qcow2_init(void)
3374 {
3375     bdrv_register(&bdrv_qcow2);
3376 }
3377
3378 block_init(bdrv_qcow2_init);