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 "block/block_int.h"
  26 #include "sysemu/block-backend.h"
  27 #include "qemu/module.h"
  28 #include <zlib.h>
  29 #include "block/qcow2.h"
  30 #include "qemu/error-report.h"
  31 #include "qapi/qmp/qerror.h"
  32 #include "qapi/qmp/qbool.h"
  33 #include "qapi/util.h"
  34 #include "qapi/qmp/types.h"
  35 #include "qapi-event.h"
  36 #include "trace.h"
  37 #include "qemu/option_int.h"
  38 #include "qemu/cutils.h"
  39 #include "qemu/bswap.h"
  40
  41 /*
  42   Differences with QCOW:
  43
  44   - Support for multiple incremental snapshots.
  45   - Memory management by reference counts.
  46   - Clusters which have a reference count of one have the bit
  47     QCOW_OFLAG_COPIED to optimize write performance.
  48   - Size of compressed clusters is stored in sectors to reduce bit usage
  49     in the cluster offsets.
  50   - Support for storing additional data (such as the VM state) in the
  51     snapshots.
  52   - If a backing store is used, the cluster size is not constrained
  53     (could be backported to QCOW).
  54   - L2 tables have always a size of one cluster.
  55 */
  56
  57
  58 typedef struct {
  59     uint32_t magic;
  60     uint32_t len;
  61 } QEMU_PACKED QCowExtension;
  62
  63 #define  QCOW2_EXT_MAGIC_END 0
  64 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
  65 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
  66
  67 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
  68 {
  69     const QCowHeader *cow_header = (const void *)buf;
  70
  71     if (buf_size >= sizeof(QCowHeader) &&
  72         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
  73         be32_to_cpu(cow_header->version) >= 2)
  74         return 100;
  75     else
  76         return 0;
  77 }
  78
  79
  80 /*
  81  * read qcow2 extension and fill bs
  82  * start reading from start_offset
  83  * finish reading upon magic of value 0 or when end_offset reached
  84  * unknown magic is skipped (future extension this version knows nothing about)
  85  * return 0 upon success, non-0 otherwise
  86  */
  87 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
  88                                  uint64_t end_offset, void **p_feature_table,
  89                                  Error **errp)
  90 {
  91     BDRVQcow2State *s = bs->opaque;
  92     QCowExtension ext;
  93     uint64_t offset;
  94     int ret;
  95
  96 #ifdef DEBUG_EXT
  97     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
  98 #endif
  99     offset = start_offset;
 100     while (offset < end_offset) {
 101
 102 #ifdef DEBUG_EXT
 103         /* Sanity check */
 104         if (offset > s->cluster_size)
 105             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
 106
 107         printf("attempting to read extended header in offset %lu\n", offset);
 108 #endif
 109
 110         ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
 111         if (ret < 0) {
 112             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
 113                              "pread fail from offset %" PRIu64, offset);
 114             return 1;
 115         }
 116         be32_to_cpus(&ext.magic);
 117         be32_to_cpus(&ext.len);
 118         offset += sizeof(ext);
 119 #ifdef DEBUG_EXT
 120         printf("ext.magic = 0x%x\n", ext.magic);
 121 #endif
 122         if (offset > end_offset || ext.len > end_offset - offset) {
 123             error_setg(errp, "Header extension too large");
 124             return -EINVAL;
 125         }
 126
 127         switch (ext.magic) {
 128         case QCOW2_EXT_MAGIC_END:
 129             return 0;
 130
 131         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
 132             if (ext.len >= sizeof(bs->backing_format)) {
 133                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
 134                            " too large (>=%zu)", ext.len,
 135                            sizeof(bs->backing_format));
 136                 return 2;
 137             }
 138             ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len);
 139             if (ret < 0) {
 140                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
 141                                  "Could not read format name");
 142                 return 3;
 143             }
 144             bs->backing_format[ext.len] = '\0';
 145             s->image_backing_format = g_strdup(bs->backing_format);
 146 #ifdef DEBUG_EXT
 147             printf("Qcow2: Got format extension %s\n", bs->backing_format);
 148 #endif
 149             break;
 150
 151         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
 152             if (p_feature_table != NULL) {
 153                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
 154                 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
 155                 if (ret < 0) {
 156                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
 157                                      "Could not read table");
 158                     return ret;
 159                 }
 160
 161                 *p_feature_table = feature_table;
 162             }
 163             break;
 164
 165         default:
 166             /* unknown magic - save it in case we need to rewrite the header */
 167             {
 168                 Qcow2UnknownHeaderExtension *uext;
 169
 170                 uext = g_malloc0(sizeof(*uext)  + ext.len);
 171                 uext->magic = ext.magic;
 172                 uext->len = ext.len;
 173                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
 174
 175                 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
 176                 if (ret < 0) {
 177                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
 178                                      "Could not read data");
 179                     return ret;
 180                 }
 181             }
 182             break;
 183         }
 184
 185         offset += ((ext.len + 7) & ~7);
 186     }
 187
 188     return 0;
 189 }
 190
 191 static void cleanup_unknown_header_ext(BlockDriverState *bs)
 192 {
 193     BDRVQcow2State *s = bs->opaque;
 194     Qcow2UnknownHeaderExtension *uext, *next;
 195
 196     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
 197         QLIST_REMOVE(uext, next);
 198         g_free(uext);
 199     }
 200 }
 201
 202 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
 203                                        uint64_t mask)
 204 {
 205     char *features = g_strdup("");
 206     char *old;
 207
 208     while (table && table->name[0] != '\0') {
 209         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
 210             if (mask & (1ULL << table->bit)) {
 211                 old = features;
 212                 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
 213                                            table->name);
 214                 g_free(old);
 215                 mask &= ~(1ULL << table->bit);
 216             }
 217         }
 218         table++;
 219     }
 220
 221     if (mask) {
 222         old = features;
 223         features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
 224                                    old, *old ? ", " : "", mask);
 225         g_free(old);
 226     }
 227
 228     error_setg(errp, "Unsupported qcow2 feature(s): %s", features);
 229     g_free(features);
 230 }
 231
 232 /*
 233  * Sets the dirty bit and flushes afterwards if necessary.
 234  *
 235  * The incompatible_features bit is only set if the image file header was
 236  * updated successfully.  Therefore it is not required to check the return
 237  * value of this function.
 238  */
 239 int qcow2_mark_dirty(BlockDriverState *bs)
 240 {
 241     BDRVQcow2State *s = bs->opaque;
 242     uint64_t val;
 243     int ret;
 244
 245     assert(s->qcow_version >= 3);
 246
 247     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
 248         return 0; /* already dirty */
 249     }
 250
 251     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
 252     ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
 253                       &val, sizeof(val));
 254     if (ret < 0) {
 255         return ret;
 256     }
 257     ret = bdrv_flush(bs->file->bs);
 258     if (ret < 0) {
 259         return ret;
 260     }
 261
 262     /* Only treat image as dirty if the header was updated successfully */
 263     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
 264     return 0;
 265 }
 266
 267 /*
 268  * Clears the dirty bit and flushes before if necessary.  Only call this
 269  * function when there are no pending requests, it does not guard against
 270  * concurrent requests dirtying the image.
 271  */
 272 static int qcow2_mark_clean(BlockDriverState *bs)
 273 {
 274     BDRVQcow2State *s = bs->opaque;
 275
 276     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
 277         int ret;
 278
 279         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
 280
 281         ret = bdrv_flush(bs);
 282         if (ret < 0) {
 283             return ret;
 284         }
 285
 286         return qcow2_update_header(bs);
 287     }
 288     return 0;
 289 }
 290
 291 /*
 292  * Marks the image as corrupt.
 293  */
 294 int qcow2_mark_corrupt(BlockDriverState *bs)
 295 {
 296     BDRVQcow2State *s = bs->opaque;
 297
 298     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
 299     return qcow2_update_header(bs);
 300 }
 301
 302 /*
 303  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
 304  * before if necessary.
 305  */
 306 int qcow2_mark_consistent(BlockDriverState *bs)
 307 {
 308     BDRVQcow2State *s = bs->opaque;
 309
 310     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
 311         int ret = bdrv_flush(bs);
 312         if (ret < 0) {
 313             return ret;
 314         }
 315
 316         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
 317         return qcow2_update_header(bs);
 318     }
 319     return 0;
 320 }
 321
 322 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
 323                        BdrvCheckMode fix)
 324 {
 325     int ret = qcow2_check_refcounts(bs, result, fix);
 326     if (ret < 0) {
 327         return ret;
 328     }
 329
 330     if (fix && result->check_errors == 0 && result->corruptions == 0) {
 331         ret = qcow2_mark_clean(bs);
 332         if (ret < 0) {
 333             return ret;
 334         }
 335         return qcow2_mark_consistent(bs);
 336     }
 337     return ret;
 338 }
 339
 340 static int validate_table_offset(BlockDriverState *bs, uint64_t offset,
 341                                  uint64_t entries, size_t entry_len)
 342 {
 343     BDRVQcow2State *s = bs->opaque;
 344     uint64_t size;
 345
 346     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
 347      * because values will be passed to qemu functions taking int64_t. */
 348     if (entries > INT64_MAX / entry_len) {
 349         return -EINVAL;
 350     }
 351
 352     size = entries * entry_len;
 353
 354     if (INT64_MAX - size < offset) {
 355         return -EINVAL;
 356     }
 357
 358     /* Tables must be cluster aligned */
 359     if (offset & (s->cluster_size - 1)) {
 360         return -EINVAL;
 361     }
 362
 363     return 0;
 364 }
 365
 366 static QemuOptsList qcow2_runtime_opts = {
 367     .name = "qcow2",
 368     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
 369     .desc = {
 370         {
 371             .name = QCOW2_OPT_LAZY_REFCOUNTS,
 372             .type = QEMU_OPT_BOOL,
 373             .help = "Postpone refcount updates",
 374         },
 375         {
 376             .name = QCOW2_OPT_DISCARD_REQUEST,
 377             .type = QEMU_OPT_BOOL,
 378             .help = "Pass guest discard requests to the layer below",
 379         },
 380         {
 381             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
 382             .type = QEMU_OPT_BOOL,
 383             .help = "Generate discard requests when snapshot related space "
 384                     "is freed",
 385         },
 386         {
 387             .name = QCOW2_OPT_DISCARD_OTHER,
 388             .type = QEMU_OPT_BOOL,
 389             .help = "Generate discard requests when other clusters are freed",
 390         },
 391         {
 392             .name = QCOW2_OPT_OVERLAP,
 393             .type = QEMU_OPT_STRING,
 394             .help = "Selects which overlap checks to perform from a range of "
 395                     "templates (none, constant, cached, all)",
 396         },
 397         {
 398             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
 399             .type = QEMU_OPT_STRING,
 400             .help = "Selects which overlap checks to perform from a range of "
 401                     "templates (none, constant, cached, all)",
 402         },
 403         {
 404             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
 405             .type = QEMU_OPT_BOOL,
 406             .help = "Check for unintended writes into the main qcow2 header",
 407         },
 408         {
 409             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
 410             .type = QEMU_OPT_BOOL,
 411             .help = "Check for unintended writes into the active L1 table",
 412         },
 413         {
 414             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
 415             .type = QEMU_OPT_BOOL,
 416             .help = "Check for unintended writes into an active L2 table",
 417         },
 418         {
 419             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
 420             .type = QEMU_OPT_BOOL,
 421             .help = "Check for unintended writes into the refcount table",
 422         },
 423         {
 424             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
 425             .type = QEMU_OPT_BOOL,
 426             .help = "Check for unintended writes into a refcount block",
 427         },
 428         {
 429             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
 430             .type = QEMU_OPT_BOOL,
 431             .help = "Check for unintended writes into the snapshot table",
 432         },
 433         {
 434             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
 435             .type = QEMU_OPT_BOOL,
 436             .help = "Check for unintended writes into an inactive L1 table",
 437         },
 438         {
 439             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
 440             .type = QEMU_OPT_BOOL,
 441             .help = "Check for unintended writes into an inactive L2 table",
 442         },
 443         {
 444             .name = QCOW2_OPT_CACHE_SIZE,
 445             .type = QEMU_OPT_SIZE,
 446             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
 447                     "cache size",
 448         },
 449         {
 450             .name = QCOW2_OPT_L2_CACHE_SIZE,
 451             .type = QEMU_OPT_SIZE,
 452             .help = "Maximum L2 table cache size",
 453         },
 454         {
 455             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
 456             .type = QEMU_OPT_SIZE,
 457             .help = "Maximum refcount block cache size",
 458         },
 459         {
 460             .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
 461             .type = QEMU_OPT_NUMBER,
 462             .help = "Clean unused cache entries after this time (in seconds)",
 463         },
 464         { /* end of list */ }
 465     },
 466 };
 467
 468 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
 469     [QCOW2_OL_MAIN_HEADER_BITNR]    = QCOW2_OPT_OVERLAP_MAIN_HEADER,
 470     [QCOW2_OL_ACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L1,
 471     [QCOW2_OL_ACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L2,
 472     [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
 473     [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
 474     [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
 475     [QCOW2_OL_INACTIVE_L1_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L1,
 476     [QCOW2_OL_INACTIVE_L2_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L2,
 477 };
 478
 479 static void cache_clean_timer_cb(void *opaque)
 480 {
 481     BlockDriverState *bs = opaque;
 482     BDRVQcow2State *s = bs->opaque;
 483     qcow2_cache_clean_unused(bs, s->l2_table_cache);
 484     qcow2_cache_clean_unused(bs, s->refcount_block_cache);
 485     timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
 486               (int64_t) s->cache_clean_interval * 1000);
 487 }
 488
 489 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
 490 {
 491     BDRVQcow2State *s = bs->opaque;
 492     if (s->cache_clean_interval > 0) {
 493         s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
 494                                              SCALE_MS, cache_clean_timer_cb,
 495                                              bs);
 496         timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
 497                   (int64_t) s->cache_clean_interval * 1000);
 498     }
 499 }
 500
 501 static void cache_clean_timer_del(BlockDriverState *bs)
 502 {
 503     BDRVQcow2State *s = bs->opaque;
 504     if (s->cache_clean_timer) {
 505         timer_del(s->cache_clean_timer);
 506         timer_free(s->cache_clean_timer);
 507         s->cache_clean_timer = NULL;
 508     }
 509 }
 510
 511 static void qcow2_detach_aio_context(BlockDriverState *bs)
 512 {
 513     cache_clean_timer_del(bs);
 514 }
 515
 516 static void qcow2_attach_aio_context(BlockDriverState *bs,
 517                                      AioContext *new_context)
 518 {
 519     cache_clean_timer_init(bs, new_context);
 520 }
 521
 522 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
 523                              uint64_t *l2_cache_size,
 524                              uint64_t *refcount_cache_size, Error **errp)
 525 {
 526     BDRVQcow2State *s = bs->opaque;
 527     uint64_t combined_cache_size;
 528     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
 529
 530     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
 531     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
 532     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
 533
 534     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
 535     *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0);
 536     *refcount_cache_size = qemu_opt_get_size(opts,
 537                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
 538
 539     if (combined_cache_size_set) {
 540         if (l2_cache_size_set && refcount_cache_size_set) {
 541             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
 542                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
 543                        "the same time");
 544             return;
 545         } else if (*l2_cache_size > combined_cache_size) {
 546             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
 547                        QCOW2_OPT_CACHE_SIZE);
 548             return;
 549         } else if (*refcount_cache_size > combined_cache_size) {
 550             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
 551                        QCOW2_OPT_CACHE_SIZE);
 552             return;
 553         }
 554
 555         if (l2_cache_size_set) {
 556             *refcount_cache_size = combined_cache_size - *l2_cache_size;
 557         } else if (refcount_cache_size_set) {
 558             *l2_cache_size = combined_cache_size - *refcount_cache_size;
 559         } else {
 560             *refcount_cache_size = combined_cache_size
 561                                  / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1);
 562             *l2_cache_size = combined_cache_size - *refcount_cache_size;
 563         }
 564     } else {
 565         if (!l2_cache_size_set && !refcount_cache_size_set) {
 566             *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE,
 567                                  (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
 568                                  * s->cluster_size);
 569             *refcount_cache_size = *l2_cache_size
 570                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 571         } else if (!l2_cache_size_set) {
 572             *l2_cache_size = *refcount_cache_size
 573                            * DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 574         } else if (!refcount_cache_size_set) {
 575             *refcount_cache_size = *l2_cache_size
 576                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 577         }
 578     }
 579 }
 580
 581 typedef struct Qcow2ReopenState {
 582     Qcow2Cache *l2_table_cache;
 583     Qcow2Cache *refcount_block_cache;
 584     bool use_lazy_refcounts;
 585     int overlap_check;
 586     bool discard_passthrough[QCOW2_DISCARD_MAX];
 587     uint64_t cache_clean_interval;
 588 } Qcow2ReopenState;
 589
 590 static int qcow2_update_options_prepare(BlockDriverState *bs,
 591                                         Qcow2ReopenState *r,
 592                                         QDict *options, int flags,
 593                                         Error **errp)
 594 {
 595     BDRVQcow2State *s = bs->opaque;
 596     QemuOpts *opts = NULL;
 597     const char *opt_overlap_check, *opt_overlap_check_template;
 598     int overlap_check_template = 0;
 599     uint64_t l2_cache_size, refcount_cache_size;
 600     int i;
 601     Error *local_err = NULL;
 602     int ret;
 603
 604     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
 605     qemu_opts_absorb_qdict(opts, options, &local_err);
 606     if (local_err) {
 607         error_propagate(errp, local_err);
 608         ret = -EINVAL;
 609         goto fail;
 610     }
 611
 612     /* get L2 table/refcount block cache size from command line options */
 613     read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size,
 614                      &local_err);
 615     if (local_err) {
 616         error_propagate(errp, local_err);
 617         ret = -EINVAL;
 618         goto fail;
 619     }
 620
 621     l2_cache_size /= s->cluster_size;
 622     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
 623         l2_cache_size = MIN_L2_CACHE_SIZE;
 624     }
 625     if (l2_cache_size > INT_MAX) {
 626         error_setg(errp, "L2 cache size too big");
 627         ret = -EINVAL;
 628         goto fail;
 629     }
 630
 631     refcount_cache_size /= s->cluster_size;
 632     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
 633         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
 634     }
 635     if (refcount_cache_size > INT_MAX) {
 636         error_setg(errp, "Refcount cache size too big");
 637         ret = -EINVAL;
 638         goto fail;
 639     }
 640
 641     /* alloc new L2 table/refcount block cache, flush old one */
 642     if (s->l2_table_cache) {
 643         ret = qcow2_cache_flush(bs, s->l2_table_cache);
 644         if (ret) {
 645             error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
 646             goto fail;
 647         }
 648     }
 649
 650     if (s->refcount_block_cache) {
 651         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
 652         if (ret) {
 653             error_setg_errno(errp, -ret,
 654                              "Failed to flush the refcount block cache");
 655             goto fail;
 656         }
 657     }
 658
 659     r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size);
 660     r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size);
 661     if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
 662         error_setg(errp, "Could not allocate metadata caches");
 663         ret = -ENOMEM;
 664         goto fail;
 665     }
 666
 667     /* New interval for cache cleanup timer */
 668     r->cache_clean_interval =
 669         qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
 670                             s->cache_clean_interval);
 671     if (r->cache_clean_interval > UINT_MAX) {
 672         error_setg(errp, "Cache clean interval too big");
 673         ret = -EINVAL;
 674         goto fail;
 675     }
 676
 677     /* lazy-refcounts; flush if going from enabled to disabled */
 678     r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
 679         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
 680     if (r->use_lazy_refcounts && s->qcow_version < 3) {
 681         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
 682                    "qemu 1.1 compatibility level");
 683         ret = -EINVAL;
 684         goto fail;
 685     }
 686
 687     if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
 688         ret = qcow2_mark_clean(bs);
 689         if (ret < 0) {
 690             error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
 691             goto fail;
 692         }
 693     }
 694
 695     /* Overlap check options */
 696     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
 697     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
 698     if (opt_overlap_check_template && opt_overlap_check &&
 699         strcmp(opt_overlap_check_template, opt_overlap_check))
 700     {
 701         error_setg(errp, "Conflicting values for qcow2 options '"
 702                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
 703                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
 704         ret = -EINVAL;
 705         goto fail;
 706     }
 707     if (!opt_overlap_check) {
 708         opt_overlap_check = opt_overlap_check_template ?: "cached";
 709     }
 710
 711     if (!strcmp(opt_overlap_check, "none")) {
 712         overlap_check_template = 0;
 713     } else if (!strcmp(opt_overlap_check, "constant")) {
 714         overlap_check_template = QCOW2_OL_CONSTANT;
 715     } else if (!strcmp(opt_overlap_check, "cached")) {
 716         overlap_check_template = QCOW2_OL_CACHED;
 717     } else if (!strcmp(opt_overlap_check, "all")) {
 718         overlap_check_template = QCOW2_OL_ALL;
 719     } else {
 720         error_setg(errp, "Unsupported value '%s' for qcow2 option "
 721                    "'overlap-check'. Allowed are any of the following: "
 722                    "none, constant, cached, all", opt_overlap_check);
 723         ret = -EINVAL;
 724         goto fail;
 725     }
 726
 727     r->overlap_check = 0;
 728     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
 729         /* overlap-check defines a template bitmask, but every flag may be
 730          * overwritten through the associated boolean option */
 731         r->overlap_check |=
 732             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
 733                               overlap_check_template & (1 << i)) << i;
 734     }
 735
 736     r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
 737     r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
 738     r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
 739         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
 740                           flags & BDRV_O_UNMAP);
 741     r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
 742         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
 743     r->discard_passthrough[QCOW2_DISCARD_OTHER] =
 744         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
 745
 746     ret = 0;
 747 fail:
 748     qemu_opts_del(opts);
 749     opts = NULL;
 750     return ret;
 751 }
 752
 753 static void qcow2_update_options_commit(BlockDriverState *bs,
 754                                         Qcow2ReopenState *r)
 755 {
 756     BDRVQcow2State *s = bs->opaque;
 757     int i;
 758
 759     if (s->l2_table_cache) {
 760         qcow2_cache_destroy(bs, s->l2_table_cache);
 761     }
 762     if (s->refcount_block_cache) {
 763         qcow2_cache_destroy(bs, s->refcount_block_cache);
 764     }
 765     s->l2_table_cache = r->l2_table_cache;
 766     s->refcount_block_cache = r->refcount_block_cache;
 767
 768     s->overlap_check = r->overlap_check;
 769     s->use_lazy_refcounts = r->use_lazy_refcounts;
 770
 771     for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
 772         s->discard_passthrough[i] = r->discard_passthrough[i];
 773     }
 774
 775     if (s->cache_clean_interval != r->cache_clean_interval) {
 776         cache_clean_timer_del(bs);
 777         s->cache_clean_interval = r->cache_clean_interval;
 778         cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
 779     }
 780 }
 781
 782 static void qcow2_update_options_abort(BlockDriverState *bs,
 783                                        Qcow2ReopenState *r)
 784 {
 785     if (r->l2_table_cache) {
 786         qcow2_cache_destroy(bs, r->l2_table_cache);
 787     }
 788     if (r->refcount_block_cache) {
 789         qcow2_cache_destroy(bs, r->refcount_block_cache);
 790     }
 791 }
 792
 793 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
 794                                 int flags, Error **errp)
 795 {
 796     Qcow2ReopenState r = {};
 797     int ret;
 798
 799     ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
 800     if (ret >= 0) {
 801         qcow2_update_options_commit(bs, &r);
 802     } else {
 803         qcow2_update_options_abort(bs, &r);
 804     }
 805
 806     return ret;
 807 }
 808
 809 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
 810                       Error **errp)
 811 {
 812     BDRVQcow2State *s = bs->opaque;
 813     unsigned int len, i;
 814     int ret = 0;
 815     QCowHeader header;
 816     Error *local_err = NULL;
 817     uint64_t ext_end;
 818     uint64_t l1_vm_state_index;
 819
 820     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
 821     if (ret < 0) {
 822         error_setg_errno(errp, -ret, "Could not read qcow2 header");
 823         goto fail;
 824     }
 825     be32_to_cpus(&header.magic);
 826     be32_to_cpus(&header.version);
 827     be64_to_cpus(&header.backing_file_offset);
 828     be32_to_cpus(&header.backing_file_size);
 829     be64_to_cpus(&header.size);
 830     be32_to_cpus(&header.cluster_bits);
 831     be32_to_cpus(&header.crypt_method);
 832     be64_to_cpus(&header.l1_table_offset);
 833     be32_to_cpus(&header.l1_size);
 834     be64_to_cpus(&header.refcount_table_offset);
 835     be32_to_cpus(&header.refcount_table_clusters);
 836     be64_to_cpus(&header.snapshots_offset);
 837     be32_to_cpus(&header.nb_snapshots);
 838
 839     if (header.magic != QCOW_MAGIC) {
 840         error_setg(errp, "Image is not in qcow2 format");
 841         ret = -EINVAL;
 842         goto fail;
 843     }
 844     if (header.version < 2 || header.version > 3) {
 845         error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
 846         ret = -ENOTSUP;
 847         goto fail;
 848     }
 849
 850     s->qcow_version = header.version;
 851
 852     /* Initialise cluster size */
 853     if (header.cluster_bits < MIN_CLUSTER_BITS ||
 854         header.cluster_bits > MAX_CLUSTER_BITS) {
 855         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
 856                    header.cluster_bits);
 857         ret = -EINVAL;
 858         goto fail;
 859     }
 860
 861     s->cluster_bits = header.cluster_bits;
 862     s->cluster_size = 1 << s->cluster_bits;
 863     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 864
 865     /* Initialise version 3 header fields */
 866     if (header.version == 2) {
 867         header.incompatible_features    = 0;
 868         header.compatible_features      = 0;
 869         header.autoclear_features       = 0;
 870         header.refcount_order           = 4;
 871         header.header_length            = 72;
 872     } else {
 873         be64_to_cpus(&header.incompatible_features);
 874         be64_to_cpus(&header.compatible_features);
 875         be64_to_cpus(&header.autoclear_features);
 876         be32_to_cpus(&header.refcount_order);
 877         be32_to_cpus(&header.header_length);
 878
 879         if (header.header_length < 104) {
 880             error_setg(errp, "qcow2 header too short");
 881             ret = -EINVAL;
 882             goto fail;
 883         }
 884     }
 885
 886     if (header.header_length > s->cluster_size) {
 887         error_setg(errp, "qcow2 header exceeds cluster size");
 888         ret = -EINVAL;
 889         goto fail;
 890     }
 891
 892     if (header.header_length > sizeof(header)) {
 893         s->unknown_header_fields_size = header.header_length - sizeof(header);
 894         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
 895         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
 896                          s->unknown_header_fields_size);
 897         if (ret < 0) {
 898             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
 899                              "fields");
 900             goto fail;
 901         }
 902     }
 903
 904     if (header.backing_file_offset > s->cluster_size) {
 905         error_setg(errp, "Invalid backing file offset");
 906         ret = -EINVAL;
 907         goto fail;
 908     }
 909
 910     if (header.backing_file_offset) {
 911         ext_end = header.backing_file_offset;
 912     } else {
 913         ext_end = 1 << header.cluster_bits;
 914     }
 915
 916     /* Handle feature bits */
 917     s->incompatible_features    = header.incompatible_features;
 918     s->compatible_features      = header.compatible_features;
 919     s->autoclear_features       = header.autoclear_features;
 920
 921     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
 922         void *feature_table = NULL;
 923         qcow2_read_extensions(bs, header.header_length, ext_end,
 924                               &feature_table, NULL);
 925         report_unsupported_feature(errp, feature_table,
 926                                    s->incompatible_features &
 927                                    ~QCOW2_INCOMPAT_MASK);
 928         ret = -ENOTSUP;
 929         g_free(feature_table);
 930         goto fail;
 931     }
 932
 933     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
 934         /* Corrupt images may not be written to unless they are being repaired
 935          */
 936         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
 937             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
 938                        "read/write");
 939             ret = -EACCES;
 940             goto fail;
 941         }
 942     }
 943
 944     /* Check support for various header values */
 945     if (header.refcount_order > 6) {
 946         error_setg(errp, "Reference count entry width too large; may not "
 947                    "exceed 64 bits");
 948         ret = -EINVAL;
 949         goto fail;
 950     }
 951     s->refcount_order = header.refcount_order;
 952     s->refcount_bits = 1 << s->refcount_order;
 953     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
 954     s->refcount_max += s->refcount_max - 1;
 955
 956     if (header.crypt_method > QCOW_CRYPT_AES) {
 957         error_setg(errp, "Unsupported encryption method: %" PRIu32,
 958                    header.crypt_method);
 959         ret = -EINVAL;
 960         goto fail;
 961     }
 962     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128,
 963                                  QCRYPTO_CIPHER_MODE_CBC)) {
 964         error_setg(errp, "AES cipher not available");
 965         ret = -EINVAL;
 966         goto fail;
 967     }
 968     s->crypt_method_header = header.crypt_method;
 969     if (s->crypt_method_header) {
 970         if (bdrv_uses_whitelist() &&
 971             s->crypt_method_header == QCOW_CRYPT_AES) {
 972             error_setg(errp,
 973                        "Use of AES-CBC encrypted qcow2 images is no longer "
 974                        "supported in system emulators");
 975             error_append_hint(errp,
 976                               "You can use 'qemu-img convert' to convert your "
 977                               "image to an alternative supported format, such "
 978                               "as unencrypted qcow2, or raw with the LUKS "
 979                               "format instead.\n");
 980             ret = -ENOSYS;
 981             goto fail;
 982         }
 983
 984         bs->encrypted = true;
 985     }
 986
 987     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
 988     s->l2_size = 1 << s->l2_bits;
 989     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
 990     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
 991     s->refcount_block_size = 1 << s->refcount_block_bits;
 992     bs->total_sectors = header.size / 512;
 993     s->csize_shift = (62 - (s->cluster_bits - 8));
 994     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
 995     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
 996
 997     s->refcount_table_offset = header.refcount_table_offset;
 998     s->refcount_table_size =
 999         header.refcount_table_clusters << (s->cluster_bits - 3);
1000
1001     if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
1002         error_setg(errp, "Reference count table too large");
1003         ret = -EINVAL;
1004         goto fail;
1005     }
1006
1007     ret = validate_table_offset(bs, s->refcount_table_offset,
1008                                 s->refcount_table_size, sizeof(uint64_t));
1009     if (ret < 0) {
1010         error_setg(errp, "Invalid reference count table offset");
1011         goto fail;
1012     }
1013
1014     /* Snapshot table offset/length */
1015     if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
1016         error_setg(errp, "Too many snapshots");
1017         ret = -EINVAL;
1018         goto fail;
1019     }
1020
1021     ret = validate_table_offset(bs, header.snapshots_offset,
1022                                 header.nb_snapshots,
1023                                 sizeof(QCowSnapshotHeader));
1024     if (ret < 0) {
1025         error_setg(errp, "Invalid snapshot table offset");
1026         goto fail;
1027     }
1028
1029     /* read the level 1 table */
1030     if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
1031         error_setg(errp, "Active L1 table too large");
1032         ret = -EFBIG;
1033         goto fail;
1034     }
1035     s->l1_size = header.l1_size;
1036
1037     l1_vm_state_index = size_to_l1(s, header.size);
1038     if (l1_vm_state_index > INT_MAX) {
1039         error_setg(errp, "Image is too big");
1040         ret = -EFBIG;
1041         goto fail;
1042     }
1043     s->l1_vm_state_index = l1_vm_state_index;
1044
1045     /* the L1 table must contain at least enough entries to put
1046        header.size bytes */
1047     if (s->l1_size < s->l1_vm_state_index) {
1048         error_setg(errp, "L1 table is too small");
1049         ret = -EINVAL;
1050         goto fail;
1051     }
1052
1053     ret = validate_table_offset(bs, header.l1_table_offset,
1054                                 header.l1_size, sizeof(uint64_t));
1055     if (ret < 0) {
1056         error_setg(errp, "Invalid L1 table offset");
1057         goto fail;
1058     }
1059     s->l1_table_offset = header.l1_table_offset;
1060
1061
1062     if (s->l1_size > 0) {
1063         s->l1_table = qemu_try_blockalign(bs->file->bs,
1064             align_offset(s->l1_size * sizeof(uint64_t), 512));
1065         if (s->l1_table == NULL) {
1066             error_setg(errp, "Could not allocate L1 table");
1067             ret = -ENOMEM;
1068             goto fail;
1069         }
1070         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
1071                          s->l1_size * sizeof(uint64_t));
1072         if (ret < 0) {
1073             error_setg_errno(errp, -ret, "Could not read L1 table");
1074             goto fail;
1075         }
1076         for(i = 0;i < s->l1_size; i++) {
1077             be64_to_cpus(&s->l1_table[i]);
1078         }
1079     }
1080
1081     /* Parse driver-specific options */
1082     ret = qcow2_update_options(bs, options, flags, errp);
1083     if (ret < 0) {
1084         goto fail;
1085     }
1086
1087     s->cluster_cache = g_malloc(s->cluster_size);
1088     /* one more sector for decompressed data alignment */
1089     s->cluster_data = qemu_try_blockalign(bs->file->bs, QCOW_MAX_CRYPT_CLUSTERS
1090                                                     * s->cluster_size + 512);
1091     if (s->cluster_data == NULL) {
1092         error_setg(errp, "Could not allocate temporary cluster buffer");
1093         ret = -ENOMEM;
1094         goto fail;
1095     }
1096
1097     s->cluster_cache_offset = -1;
1098     s->flags = flags;
1099
1100     ret = qcow2_refcount_init(bs);
1101     if (ret != 0) {
1102         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1103         goto fail;
1104     }
1105
1106     QLIST_INIT(&s->cluster_allocs);
1107     QTAILQ_INIT(&s->discards);
1108
1109     /* read qcow2 extensions */
1110     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1111         &local_err)) {
1112         error_propagate(errp, local_err);
1113         ret = -EINVAL;
1114         goto fail;
1115     }
1116
1117     /* read the backing file name */
1118     if (header.backing_file_offset != 0) {
1119         len = header.backing_file_size;
1120         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1121             len >= sizeof(bs->backing_file)) {
1122             error_setg(errp, "Backing file name too long");
1123             ret = -EINVAL;
1124             goto fail;
1125         }
1126         ret = bdrv_pread(bs->file, header.backing_file_offset,
1127                          bs->backing_file, len);
1128         if (ret < 0) {
1129             error_setg_errno(errp, -ret, "Could not read backing file name");
1130             goto fail;
1131         }
1132         bs->backing_file[len] = '\0';
1133         s->image_backing_file = g_strdup(bs->backing_file);
1134     }
1135
1136     /* Internal snapshots */
1137     s->snapshots_offset = header.snapshots_offset;
1138     s->nb_snapshots = header.nb_snapshots;
1139
1140     ret = qcow2_read_snapshots(bs);
1141     if (ret < 0) {
1142         error_setg_errno(errp, -ret, "Could not read snapshots");
1143         goto fail;
1144     }
1145
1146     /* Clear unknown autoclear feature bits */
1147     if (!bs->read_only && !(flags & BDRV_O_INACTIVE) && s->autoclear_features) {
1148         s->autoclear_features = 0;
1149         ret = qcow2_update_header(bs);
1150         if (ret < 0) {
1151             error_setg_errno(errp, -ret, "Could not update qcow2 header");
1152             goto fail;
1153         }
1154     }
1155
1156     /* Initialise locks */
1157     qemu_co_mutex_init(&s->lock);
1158     bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
1159
1160     /* Repair image if dirty */
1161     if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1162         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1163         BdrvCheckResult result = {0};
1164
1165         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1166         if (ret < 0) {
1167             error_setg_errno(errp, -ret, "Could not repair dirty image");
1168             goto fail;
1169         }
1170     }
1171
1172 #ifdef DEBUG_ALLOC
1173     {
1174         BdrvCheckResult result = {0};
1175         qcow2_check_refcounts(bs, &result, 0);
1176     }
1177 #endif
1178     return ret;
1179
1180  fail:
1181     g_free(s->unknown_header_fields);
1182     cleanup_unknown_header_ext(bs);
1183     qcow2_free_snapshots(bs);
1184     qcow2_refcount_close(bs);
1185     qemu_vfree(s->l1_table);
1186     /* else pre-write overlap checks in cache_destroy may crash */
1187     s->l1_table = NULL;
1188     cache_clean_timer_del(bs);
1189     if (s->l2_table_cache) {
1190         qcow2_cache_destroy(bs, s->l2_table_cache);
1191     }
1192     if (s->refcount_block_cache) {
1193         qcow2_cache_destroy(bs, s->refcount_block_cache);
1194     }
1195     g_free(s->cluster_cache);
1196     qemu_vfree(s->cluster_data);
1197     return ret;
1198 }
1199
1200 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1201 {
1202     BDRVQcow2State *s = bs->opaque;
1203
1204     if (bs->encrypted) {
1205         /* Encryption works on a sector granularity */
1206         bs->bl.request_alignment = BDRV_SECTOR_SIZE;
1207     }
1208     bs->bl.pwrite_zeroes_alignment = s->cluster_size;
1209     bs->bl.pdiscard_alignment = s->cluster_size;
1210 }
1211
1212 static int qcow2_set_key(BlockDriverState *bs, const char *key)
1213 {
1214     BDRVQcow2State *s = bs->opaque;
1215     uint8_t keybuf[16];
1216     int len, i;
1217     Error *err = NULL;
1218
1219     memset(keybuf, 0, 16);
1220     len = strlen(key);
1221     if (len > 16)
1222         len = 16;
1223     /* XXX: we could compress the chars to 7 bits to increase
1224        entropy */
1225     for(i = 0;i < len;i++) {
1226         keybuf[i] = key[i];
1227     }
1228     assert(bs->encrypted);
1229
1230     qcrypto_cipher_free(s->cipher);
1231     s->cipher = qcrypto_cipher_new(
1232         QCRYPTO_CIPHER_ALG_AES_128,
1233         QCRYPTO_CIPHER_MODE_CBC,
1234         keybuf, G_N_ELEMENTS(keybuf),
1235         &err);
1236
1237     if (!s->cipher) {
1238         /* XXX would be nice if errors in this method could
1239          * be properly propagate to the caller. Would need
1240          * the bdrv_set_key() API signature to be fixed. */
1241         error_free(err);
1242         return -1;
1243     }
1244     return 0;
1245 }
1246
1247 static int qcow2_reopen_prepare(BDRVReopenState *state,
1248                                 BlockReopenQueue *queue, Error **errp)
1249 {
1250     Qcow2ReopenState *r;
1251     int ret;
1252
1253     r = g_new0(Qcow2ReopenState, 1);
1254     state->opaque = r;
1255
1256     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1257                                        state->flags, errp);
1258     if (ret < 0) {
1259         goto fail;
1260     }
1261
1262     /* We need to write out any unwritten data if we reopen read-only. */
1263     if ((state->flags & BDRV_O_RDWR) == 0) {
1264         ret = bdrv_flush(state->bs);
1265         if (ret < 0) {
1266             goto fail;
1267         }
1268
1269         ret = qcow2_mark_clean(state->bs);
1270         if (ret < 0) {
1271             goto fail;
1272         }
1273     }
1274
1275     return 0;
1276
1277 fail:
1278     qcow2_update_options_abort(state->bs, r);
1279     g_free(r);
1280     return ret;
1281 }
1282
1283 static void qcow2_reopen_commit(BDRVReopenState *state)
1284 {
1285     qcow2_update_options_commit(state->bs, state->opaque);
1286     g_free(state->opaque);
1287 }
1288
1289 static void qcow2_reopen_abort(BDRVReopenState *state)
1290 {
1291     qcow2_update_options_abort(state->bs, state->opaque);
1292     g_free(state->opaque);
1293 }
1294
1295 static void qcow2_join_options(QDict *options, QDict *old_options)
1296 {
1297     bool has_new_overlap_template =
1298         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1299         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1300     bool has_new_total_cache_size =
1301         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1302     bool has_all_cache_options;
1303
1304     /* New overlap template overrides all old overlap options */
1305     if (has_new_overlap_template) {
1306         qdict_del(old_options, QCOW2_OPT_OVERLAP);
1307         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1308         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1309         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1310         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1311         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1312         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1313         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1314         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1315         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1316     }
1317
1318     /* New total cache size overrides all old options */
1319     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1320         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1321         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1322     }
1323
1324     qdict_join(options, old_options, false);
1325
1326     /*
1327      * If after merging all cache size options are set, an old total size is
1328      * overwritten. Do keep all options, however, if all three are new. The
1329      * resulting error message is what we want to happen.
1330      */
1331     has_all_cache_options =
1332         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1333         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1334         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1335
1336     if (has_all_cache_options && !has_new_total_cache_size) {
1337         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1338     }
1339 }
1340
1341 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1342         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1343 {
1344     BDRVQcow2State *s = bs->opaque;
1345     uint64_t cluster_offset;
1346     int index_in_cluster, ret;
1347     unsigned int bytes;
1348     int64_t status = 0;
1349
1350     bytes = MIN(INT_MAX, nb_sectors * BDRV_SECTOR_SIZE);
1351     qemu_co_mutex_lock(&s->lock);
1352     ret = qcow2_get_cluster_offset(bs, sector_num << 9, &bytes,
1353                                    &cluster_offset);
1354     qemu_co_mutex_unlock(&s->lock);
1355     if (ret < 0) {
1356         return ret;
1357     }
1358
1359     *pnum = bytes >> BDRV_SECTOR_BITS;
1360
1361     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1362         !s->cipher) {
1363         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1364         cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1365         *file = bs->file->bs;
1366         status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1367     }
1368     if (ret == QCOW2_CLUSTER_ZERO) {
1369         status |= BDRV_BLOCK_ZERO;
1370     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1371         status |= BDRV_BLOCK_DATA;
1372     }
1373     return status;
1374 }
1375
1376 /* handle reading after the end of the backing file */
1377 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
1378                         int64_t offset, int bytes)
1379 {
1380     uint64_t bs_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1381     int n1;
1382
1383     if ((offset + bytes) <= bs_size) {
1384         return bytes;
1385     }
1386
1387     if (offset >= bs_size) {
1388         n1 = 0;
1389     } else {
1390         n1 = bs_size - offset;
1391     }
1392
1393     qemu_iovec_memset(qiov, n1, 0, bytes - n1);
1394
1395     return n1;
1396 }
1397
1398 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset,
1399                                         uint64_t bytes, QEMUIOVector *qiov,
1400                                         int flags)
1401 {
1402     BDRVQcow2State *s = bs->opaque;
1403     int offset_in_cluster, n1;
1404     int ret;
1405     unsigned int cur_bytes; /* number of bytes in current iteration */
1406     uint64_t cluster_offset = 0;
1407     uint64_t bytes_done = 0;
1408     QEMUIOVector hd_qiov;
1409     uint8_t *cluster_data = NULL;
1410
1411     qemu_iovec_init(&hd_qiov, qiov->niov);
1412
1413     qemu_co_mutex_lock(&s->lock);
1414
1415     while (bytes != 0) {
1416
1417         /* prepare next request */
1418         cur_bytes = MIN(bytes, INT_MAX);
1419         if (s->cipher) {
1420             cur_bytes = MIN(cur_bytes,
1421                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1422         }
1423
1424         ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset);
1425         if (ret < 0) {
1426             goto fail;
1427         }
1428
1429         offset_in_cluster = offset_into_cluster(s, offset);
1430
1431         qemu_iovec_reset(&hd_qiov);
1432         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
1433
1434         switch (ret) {
1435         case QCOW2_CLUSTER_UNALLOCATED:
1436
1437             if (bs->backing) {
1438                 /* read from the base image */
1439                 n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov,
1440                                          offset, cur_bytes);
1441                 if (n1 > 0) {
1442                     QEMUIOVector local_qiov;
1443
1444                     qemu_iovec_init(&local_qiov, hd_qiov.niov);
1445                     qemu_iovec_concat(&local_qiov, &hd_qiov, 0, n1);
1446
1447                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1448                     qemu_co_mutex_unlock(&s->lock);
1449                     ret = bdrv_co_preadv(bs->backing, offset, n1,
1450                                          &local_qiov, 0);
1451                     qemu_co_mutex_lock(&s->lock);
1452
1453                     qemu_iovec_destroy(&local_qiov);
1454
1455                     if (ret < 0) {
1456                         goto fail;
1457                     }
1458                 }
1459             } else {
1460                 /* Note: in this case, no need to wait */
1461                 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1462             }
1463             break;
1464
1465         case QCOW2_CLUSTER_ZERO:
1466             qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1467             break;
1468
1469         case QCOW2_CLUSTER_COMPRESSED:
1470             /* add AIO support for compressed blocks ? */
1471             ret = qcow2_decompress_cluster(bs, cluster_offset);
1472             if (ret < 0) {
1473                 goto fail;
1474             }
1475
1476             qemu_iovec_from_buf(&hd_qiov, 0,
1477                                 s->cluster_cache + offset_in_cluster,
1478                                 cur_bytes);
1479             break;
1480
1481         case QCOW2_CLUSTER_NORMAL:
1482             if ((cluster_offset & 511) != 0) {
1483                 ret = -EIO;
1484                 goto fail;
1485             }
1486
1487             if (bs->encrypted) {
1488                 assert(s->cipher);
1489
1490                 /*
1491                  * For encrypted images, read everything into a temporary
1492                  * contiguous buffer on which the AES functions can work.
1493                  */
1494                 if (!cluster_data) {
1495                     cluster_data =
1496                         qemu_try_blockalign(bs->file->bs,
1497                                             QCOW_MAX_CRYPT_CLUSTERS
1498                                             * s->cluster_size);
1499                     if (cluster_data == NULL) {
1500                         ret = -ENOMEM;
1501                         goto fail;
1502                     }
1503                 }
1504
1505                 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1506                 qemu_iovec_reset(&hd_qiov);
1507                 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
1508             }
1509
1510             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1511             qemu_co_mutex_unlock(&s->lock);
1512             ret = bdrv_co_preadv(bs->file,
1513                                  cluster_offset + offset_in_cluster,
1514                                  cur_bytes, &hd_qiov, 0);
1515             qemu_co_mutex_lock(&s->lock);
1516             if (ret < 0) {
1517                 goto fail;
1518             }
1519             if (bs->encrypted) {
1520                 assert(s->cipher);
1521                 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
1522                 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
1523                 Error *err = NULL;
1524                 if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS,
1525                                           cluster_data, cluster_data,
1526                                           cur_bytes >> BDRV_SECTOR_BITS,
1527                                           false, &err) < 0) {
1528                     error_free(err);
1529                     ret = -EIO;
1530                     goto fail;
1531                 }
1532                 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes);
1533             }
1534             break;
1535
1536         default:
1537             g_assert_not_reached();
1538             ret = -EIO;
1539             goto fail;
1540         }
1541
1542         bytes -= cur_bytes;
1543         offset += cur_bytes;
1544         bytes_done += cur_bytes;
1545     }
1546     ret = 0;
1547
1548 fail:
1549     qemu_co_mutex_unlock(&s->lock);
1550
1551     qemu_iovec_destroy(&hd_qiov);
1552     qemu_vfree(cluster_data);
1553
1554     return ret;
1555 }
1556
1557 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset,
1558                                          uint64_t bytes, QEMUIOVector *qiov,
1559                                          int flags)
1560 {
1561     BDRVQcow2State *s = bs->opaque;
1562     int offset_in_cluster;
1563     int ret;
1564     unsigned int cur_bytes; /* number of sectors in current iteration */
1565     uint64_t cluster_offset;
1566     QEMUIOVector hd_qiov;
1567     uint64_t bytes_done = 0;
1568     uint8_t *cluster_data = NULL;
1569     QCowL2Meta *l2meta = NULL;
1570
1571     trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
1572
1573     qemu_iovec_init(&hd_qiov, qiov->niov);
1574
1575     s->cluster_cache_offset = -1; /* disable compressed cache */
1576
1577     qemu_co_mutex_lock(&s->lock);
1578
1579     while (bytes != 0) {
1580
1581         l2meta = NULL;
1582
1583         trace_qcow2_writev_start_part(qemu_coroutine_self());
1584         offset_in_cluster = offset_into_cluster(s, offset);
1585         cur_bytes = MIN(bytes, INT_MAX);
1586         if (bs->encrypted) {
1587             cur_bytes = MIN(cur_bytes,
1588                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
1589                             - offset_in_cluster);
1590         }
1591
1592         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
1593                                          &cluster_offset, &l2meta);
1594         if (ret < 0) {
1595             goto fail;
1596         }
1597
1598         assert((cluster_offset & 511) == 0);
1599
1600         qemu_iovec_reset(&hd_qiov);
1601         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
1602
1603         if (bs->encrypted) {
1604             Error *err = NULL;
1605             assert(s->cipher);
1606             if (!cluster_data) {
1607                 cluster_data = qemu_try_blockalign(bs->file->bs,
1608                                                    QCOW_MAX_CRYPT_CLUSTERS
1609                                                    * s->cluster_size);
1610                 if (cluster_data == NULL) {
1611                     ret = -ENOMEM;
1612                     goto fail;
1613                 }
1614             }
1615
1616             assert(hd_qiov.size <=
1617                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1618             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1619
1620             if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS,
1621                                       cluster_data, cluster_data,
1622                                       cur_bytes >>BDRV_SECTOR_BITS,
1623                                       true, &err) < 0) {
1624                 error_free(err);
1625                 ret = -EIO;
1626                 goto fail;
1627             }
1628
1629             qemu_iovec_reset(&hd_qiov);
1630             qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
1631         }
1632
1633         ret = qcow2_pre_write_overlap_check(bs, 0,
1634                 cluster_offset + offset_in_cluster, cur_bytes);
1635         if (ret < 0) {
1636             goto fail;
1637         }
1638
1639         qemu_co_mutex_unlock(&s->lock);
1640         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1641         trace_qcow2_writev_data(qemu_coroutine_self(),
1642                                 cluster_offset + offset_in_cluster);
1643         ret = bdrv_co_pwritev(bs->file,
1644                               cluster_offset + offset_in_cluster,
1645                               cur_bytes, &hd_qiov, 0);
1646         qemu_co_mutex_lock(&s->lock);
1647         if (ret < 0) {
1648             goto fail;
1649         }
1650
1651         while (l2meta != NULL) {
1652             QCowL2Meta *next;
1653
1654             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1655             if (ret < 0) {
1656                 goto fail;
1657             }
1658
1659             /* Take the request off the list of running requests */
1660             if (l2meta->nb_clusters != 0) {
1661                 QLIST_REMOVE(l2meta, next_in_flight);
1662             }
1663
1664             qemu_co_queue_restart_all(&l2meta->dependent_requests);
1665
1666             next = l2meta->next;
1667             g_free(l2meta);
1668             l2meta = next;
1669         }
1670
1671         bytes -= cur_bytes;
1672         offset += cur_bytes;
1673         bytes_done += cur_bytes;
1674         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
1675     }
1676     ret = 0;
1677
1678 fail:
1679     qemu_co_mutex_unlock(&s->lock);
1680
1681     while (l2meta != NULL) {
1682         QCowL2Meta *next;
1683
1684         if (l2meta->nb_clusters != 0) {
1685             QLIST_REMOVE(l2meta, next_in_flight);
1686         }
1687         qemu_co_queue_restart_all(&l2meta->dependent_requests);
1688
1689         next = l2meta->next;
1690         g_free(l2meta);
1691         l2meta = next;
1692     }
1693
1694     qemu_iovec_destroy(&hd_qiov);
1695     qemu_vfree(cluster_data);
1696     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
1697
1698     return ret;
1699 }
1700
1701 static int qcow2_inactivate(BlockDriverState *bs)
1702 {
1703     BDRVQcow2State *s = bs->opaque;
1704     int ret, result = 0;
1705
1706     ret = qcow2_cache_flush(bs, s->l2_table_cache);
1707     if (ret) {
1708         result = ret;
1709         error_report("Failed to flush the L2 table cache: %s",
1710                      strerror(-ret));
1711     }
1712
1713     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1714     if (ret) {
1715         result = ret;
1716         error_report("Failed to flush the refcount block cache: %s",
1717                      strerror(-ret));
1718     }
1719
1720     if (result == 0) {
1721         qcow2_mark_clean(bs);
1722     }
1723
1724     return result;
1725 }
1726
1727 static void qcow2_close(BlockDriverState *bs)
1728 {
1729     BDRVQcow2State *s = bs->opaque;
1730     qemu_vfree(s->l1_table);
1731     /* else pre-write overlap checks in cache_destroy may crash */
1732     s->l1_table = NULL;
1733
1734     if (!(s->flags & BDRV_O_INACTIVE)) {
1735         qcow2_inactivate(bs);
1736     }
1737
1738     cache_clean_timer_del(bs);
1739     qcow2_cache_destroy(bs, s->l2_table_cache);
1740     qcow2_cache_destroy(bs, s->refcount_block_cache);
1741
1742     qcrypto_cipher_free(s->cipher);
1743     s->cipher = NULL;
1744
1745     g_free(s->unknown_header_fields);
1746     cleanup_unknown_header_ext(bs);
1747
1748     g_free(s->image_backing_file);
1749     g_free(s->image_backing_format);
1750
1751     g_free(s->cluster_cache);
1752     qemu_vfree(s->cluster_data);
1753     qcow2_refcount_close(bs);
1754     qcow2_free_snapshots(bs);
1755 }
1756
1757 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
1758 {
1759     BDRVQcow2State *s = bs->opaque;
1760     int flags = s->flags;
1761     QCryptoCipher *cipher = NULL;
1762     QDict *options;
1763     Error *local_err = NULL;
1764     int ret;
1765
1766     /*
1767      * Backing files are read-only which makes all of their metadata immutable,
1768      * that means we don't have to worry about reopening them here.
1769      */
1770
1771     cipher = s->cipher;
1772     s->cipher = NULL;
1773
1774     qcow2_close(bs);
1775
1776     memset(s, 0, sizeof(BDRVQcow2State));
1777     options = qdict_clone_shallow(bs->options);
1778
1779     flags &= ~BDRV_O_INACTIVE;
1780     ret = qcow2_open(bs, options, flags, &local_err);
1781     QDECREF(options);
1782     if (local_err) {
1783         error_propagate(errp, local_err);
1784         error_prepend(errp, "Could not reopen qcow2 layer: ");
1785         bs->drv = NULL;
1786         return;
1787     } else if (ret < 0) {
1788         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
1789         bs->drv = NULL;
1790         return;
1791     }
1792
1793     s->cipher = cipher;
1794 }
1795
1796 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
1797     size_t len, size_t buflen)
1798 {
1799     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
1800     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
1801
1802     if (buflen < ext_len) {
1803         return -ENOSPC;
1804     }
1805
1806     *ext_backing_fmt = (QCowExtension) {
1807         .magic  = cpu_to_be32(magic),
1808         .len    = cpu_to_be32(len),
1809     };
1810
1811     if (len) {
1812         memcpy(buf + sizeof(QCowExtension), s, len);
1813     }
1814
1815     return ext_len;
1816 }
1817
1818 /*
1819  * Updates the qcow2 header, including the variable length parts of it, i.e.
1820  * the backing file name and all extensions. qcow2 was not designed to allow
1821  * such changes, so if we run out of space (we can only use the first cluster)
1822  * this function may fail.
1823  *
1824  * Returns 0 on success, -errno in error cases.
1825  */
1826 int qcow2_update_header(BlockDriverState *bs)
1827 {
1828     BDRVQcow2State *s = bs->opaque;
1829     QCowHeader *header;
1830     char *buf;
1831     size_t buflen = s->cluster_size;
1832     int ret;
1833     uint64_t total_size;
1834     uint32_t refcount_table_clusters;
1835     size_t header_length;
1836     Qcow2UnknownHeaderExtension *uext;
1837
1838     buf = qemu_blockalign(bs, buflen);
1839
1840     /* Header structure */
1841     header = (QCowHeader*) buf;
1842
1843     if (buflen < sizeof(*header)) {
1844         ret = -ENOSPC;
1845         goto fail;
1846     }
1847
1848     header_length = sizeof(*header) + s->unknown_header_fields_size;
1849     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1850     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1851
1852     *header = (QCowHeader) {
1853         /* Version 2 fields */
1854         .magic                  = cpu_to_be32(QCOW_MAGIC),
1855         .version                = cpu_to_be32(s->qcow_version),
1856         .backing_file_offset    = 0,
1857         .backing_file_size      = 0,
1858         .cluster_bits           = cpu_to_be32(s->cluster_bits),
1859         .size                   = cpu_to_be64(total_size),
1860         .crypt_method           = cpu_to_be32(s->crypt_method_header),
1861         .l1_size                = cpu_to_be32(s->l1_size),
1862         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
1863         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
1864         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1865         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
1866         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
1867
1868         /* Version 3 fields */
1869         .incompatible_features  = cpu_to_be64(s->incompatible_features),
1870         .compatible_features    = cpu_to_be64(s->compatible_features),
1871         .autoclear_features     = cpu_to_be64(s->autoclear_features),
1872         .refcount_order         = cpu_to_be32(s->refcount_order),
1873         .header_length          = cpu_to_be32(header_length),
1874     };
1875
1876     /* For older versions, write a shorter header */
1877     switch (s->qcow_version) {
1878     case 2:
1879         ret = offsetof(QCowHeader, incompatible_features);
1880         break;
1881     case 3:
1882         ret = sizeof(*header);
1883         break;
1884     default:
1885         ret = -EINVAL;
1886         goto fail;
1887     }
1888
1889     buf += ret;
1890     buflen -= ret;
1891     memset(buf, 0, buflen);
1892
1893     /* Preserve any unknown field in the header */
1894     if (s->unknown_header_fields_size) {
1895         if (buflen < s->unknown_header_fields_size) {
1896             ret = -ENOSPC;
1897             goto fail;
1898         }
1899
1900         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1901         buf += s->unknown_header_fields_size;
1902         buflen -= s->unknown_header_fields_size;
1903     }
1904
1905     /* Backing file format header extension */
1906     if (s->image_backing_format) {
1907         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1908                              s->image_backing_format,
1909                              strlen(s->image_backing_format),
1910                              buflen);
1911         if (ret < 0) {
1912             goto fail;
1913         }
1914
1915         buf += ret;
1916         buflen -= ret;
1917     }
1918
1919     /* Feature table */
1920     if (s->qcow_version >= 3) {
1921         Qcow2Feature features[] = {
1922             {
1923                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1924                 .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
1925                 .name = "dirty bit",
1926             },
1927             {
1928                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1929                 .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
1930                 .name = "corrupt bit",
1931             },
1932             {
1933                 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1934                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1935                 .name = "lazy refcounts",
1936             },
1937         };
1938
1939         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1940                              features, sizeof(features), buflen);
1941         if (ret < 0) {
1942             goto fail;
1943         }
1944         buf += ret;
1945         buflen -= ret;
1946     }
1947
1948     /* Keep unknown header extensions */
1949     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1950         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1951         if (ret < 0) {
1952             goto fail;
1953         }
1954
1955         buf += ret;
1956         buflen -= ret;
1957     }
1958
1959     /* End of header extensions */
1960     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1961     if (ret < 0) {
1962         goto fail;
1963     }
1964
1965     buf += ret;
1966     buflen -= ret;
1967
1968     /* Backing file name */
1969     if (s->image_backing_file) {
1970         size_t backing_file_len = strlen(s->image_backing_file);
1971
1972         if (buflen < backing_file_len) {
1973             ret = -ENOSPC;
1974             goto fail;
1975         }
1976
1977         /* Using strncpy is ok here, since buf is not NUL-terminated. */
1978         strncpy(buf, s->image_backing_file, buflen);
1979
1980         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1981         header->backing_file_size   = cpu_to_be32(backing_file_len);
1982     }
1983
1984     /* Write the new header */
1985     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
1986     if (ret < 0) {
1987         goto fail;
1988     }
1989
1990     ret = 0;
1991 fail:
1992     qemu_vfree(header);
1993     return ret;
1994 }
1995
1996 static int qcow2_change_backing_file(BlockDriverState *bs,
1997     const char *backing_file, const char *backing_fmt)
1998 {
1999     BDRVQcow2State *s = bs->opaque;
2000
2001     if (backing_file && strlen(backing_file) > 1023) {
2002         return -EINVAL;
2003     }
2004
2005     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2006     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2007
2008     g_free(s->image_backing_file);
2009     g_free(s->image_backing_format);
2010
2011     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
2012     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
2013
2014     return qcow2_update_header(bs);
2015 }
2016
2017 static int preallocate(BlockDriverState *bs)
2018 {
2019     uint64_t bytes;
2020     uint64_t offset;
2021     uint64_t host_offset = 0;
2022     unsigned int cur_bytes;
2023     int ret;
2024     QCowL2Meta *meta;
2025
2026     bytes = bdrv_getlength(bs);
2027     offset = 0;
2028
2029     while (bytes) {
2030         cur_bytes = MIN(bytes, INT_MAX);
2031         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2032                                          &host_offset, &meta);
2033         if (ret < 0) {
2034             return ret;
2035         }
2036
2037         while (meta) {
2038             QCowL2Meta *next = meta->next;
2039
2040             ret = qcow2_alloc_cluster_link_l2(bs, meta);
2041             if (ret < 0) {
2042                 qcow2_free_any_clusters(bs, meta->alloc_offset,
2043                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
2044                 return ret;
2045             }
2046
2047             /* There are no dependent requests, but we need to remove our
2048              * request from the list of in-flight requests */
2049             QLIST_REMOVE(meta, next_in_flight);
2050
2051             g_free(meta);
2052             meta = next;
2053         }
2054
2055         /* TODO Preallocate data if requested */
2056
2057         bytes -= cur_bytes;
2058         offset += cur_bytes;
2059     }
2060
2061     /*
2062      * It is expected that the image file is large enough to actually contain
2063      * all of the allocated clusters (otherwise we get failing reads after
2064      * EOF). Extend the image to the last allocated sector.
2065      */
2066     if (host_offset != 0) {
2067         uint8_t data = 0;
2068         ret = bdrv_pwrite(bs->file, (host_offset + cur_bytes) - 1,
2069                           &data, 1);
2070         if (ret < 0) {
2071             return ret;
2072         }
2073     }
2074
2075     return 0;
2076 }
2077
2078 static int qcow2_create2(const char *filename, int64_t total_size,
2079                          const char *backing_file, const char *backing_format,
2080                          int flags, size_t cluster_size, PreallocMode prealloc,
2081                          QemuOpts *opts, int version, int refcount_order,
2082                          Error **errp)
2083 {
2084     int cluster_bits;
2085     QDict *options;
2086
2087     /* Calculate cluster_bits */
2088     cluster_bits = ctz32(cluster_size);
2089     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
2090         (1 << cluster_bits) != cluster_size)
2091     {
2092         error_setg(errp, "Cluster size must be a power of two between %d and "
2093                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
2094         return -EINVAL;
2095     }
2096
2097     /*
2098      * Open the image file and write a minimal qcow2 header.
2099      *
2100      * We keep things simple and start with a zero-sized image. We also
2101      * do without refcount blocks or a L1 table for now. We'll fix the
2102      * inconsistency later.
2103      *
2104      * We do need a refcount table because growing the refcount table means
2105      * allocating two new refcount blocks - the seconds of which would be at
2106      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2107      * size for any qcow2 image.
2108      */
2109     BlockBackend *blk;
2110     QCowHeader *header;
2111     uint64_t* refcount_table;
2112     Error *local_err = NULL;
2113     int ret;
2114
2115     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
2116         /* Note: The following calculation does not need to be exact; if it is a
2117          * bit off, either some bytes will be "leaked" (which is fine) or we
2118          * will need to increase the file size by some bytes (which is fine,
2119          * too, as long as the bulk is allocated here). Therefore, using
2120          * floating point arithmetic is fine. */
2121         int64_t meta_size = 0;
2122         uint64_t nreftablee, nrefblocke, nl1e, nl2e;
2123         int64_t aligned_total_size = align_offset(total_size, cluster_size);
2124         int refblock_bits, refblock_size;
2125         /* refcount entry size in bytes */
2126         double rces = (1 << refcount_order) / 8.;
2127
2128         /* see qcow2_open() */
2129         refblock_bits = cluster_bits - (refcount_order - 3);
2130         refblock_size = 1 << refblock_bits;
2131
2132         /* header: 1 cluster */
2133         meta_size += cluster_size;
2134
2135         /* total size of L2 tables */
2136         nl2e = aligned_total_size / cluster_size;
2137         nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
2138         meta_size += nl2e * sizeof(uint64_t);
2139
2140         /* total size of L1 tables */
2141         nl1e = nl2e * sizeof(uint64_t) / cluster_size;
2142         nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
2143         meta_size += nl1e * sizeof(uint64_t);
2144
2145         /* total size of refcount blocks
2146          *
2147          * note: every host cluster is reference-counted, including metadata
2148          * (even refcount blocks are recursively included).
2149          * Let:
2150          *   a = total_size (this is the guest disk size)
2151          *   m = meta size not including refcount blocks and refcount tables
2152          *   c = cluster size
2153          *   y1 = number of refcount blocks entries
2154          *   y2 = meta size including everything
2155          *   rces = refcount entry size in bytes
2156          * then,
2157          *   y1 = (y2 + a)/c
2158          *   y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m
2159          * we can get y1:
2160          *   y1 = (a + m) / (c - rces - rces * sizeof(u64) / c)
2161          */
2162         nrefblocke = (aligned_total_size + meta_size + cluster_size)
2163                    / (cluster_size - rces - rces * sizeof(uint64_t)
2164                                                  / cluster_size);
2165         meta_size += DIV_ROUND_UP(nrefblocke, refblock_size) * cluster_size;
2166
2167         /* total size of refcount tables */
2168         nreftablee = nrefblocke / refblock_size;
2169         nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t));
2170         meta_size += nreftablee * sizeof(uint64_t);
2171
2172         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
2173                             aligned_total_size + meta_size, &error_abort);
2174         qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc],
2175                      &error_abort);
2176     }
2177
2178     ret = bdrv_create_file(filename, opts, &local_err);
2179     if (ret < 0) {
2180         error_propagate(errp, local_err);
2181         return ret;
2182     }
2183
2184     blk = blk_new_open(filename, NULL, NULL,
2185                        BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
2186     if (blk == NULL) {
2187         error_propagate(errp, local_err);
2188         return -EIO;
2189     }
2190
2191     blk_set_allow_write_beyond_eof(blk, true);
2192
2193     /* Write the header */
2194     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
2195     header = g_malloc0(cluster_size);
2196     *header = (QCowHeader) {
2197         .magic                      = cpu_to_be32(QCOW_MAGIC),
2198         .version                    = cpu_to_be32(version),
2199         .cluster_bits               = cpu_to_be32(cluster_bits),
2200         .size                       = cpu_to_be64(0),
2201         .l1_table_offset            = cpu_to_be64(0),
2202         .l1_size                    = cpu_to_be32(0),
2203         .refcount_table_offset      = cpu_to_be64(cluster_size),
2204         .refcount_table_clusters    = cpu_to_be32(1),
2205         .refcount_order             = cpu_to_be32(refcount_order),
2206         .header_length              = cpu_to_be32(sizeof(*header)),
2207     };
2208
2209     if (flags & BLOCK_FLAG_ENCRYPT) {
2210         header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
2211     } else {
2212         header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
2213     }
2214
2215     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
2216         header->compatible_features |=
2217             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
2218     }
2219
2220     ret = blk_pwrite(blk, 0, header, cluster_size, 0);
2221     g_free(header);
2222     if (ret < 0) {
2223         error_setg_errno(errp, -ret, "Could not write qcow2 header");
2224         goto out;
2225     }
2226
2227     /* Write a refcount table with one refcount block */
2228     refcount_table = g_malloc0(2 * cluster_size);
2229     refcount_table[0] = cpu_to_be64(2 * cluster_size);
2230     ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
2231     g_free(refcount_table);
2232
2233     if (ret < 0) {
2234         error_setg_errno(errp, -ret, "Could not write refcount table");
2235         goto out;
2236     }
2237
2238     blk_unref(blk);
2239     blk = NULL;
2240
2241     /*
2242      * And now open the image and make it consistent first (i.e. increase the
2243      * refcount of the cluster that is occupied by the header and the refcount
2244      * table)
2245      */
2246     options = qdict_new();
2247     qdict_put(options, "driver", qstring_from_str("qcow2"));
2248     blk = blk_new_open(filename, NULL, options,
2249                        BDRV_O_RDWR | BDRV_O_NO_FLUSH, &local_err);
2250     if (blk == NULL) {
2251         error_propagate(errp, local_err);
2252         ret = -EIO;
2253         goto out;
2254     }
2255
2256     ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
2257     if (ret < 0) {
2258         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
2259                          "header and refcount table");
2260         goto out;
2261
2262     } else if (ret != 0) {
2263         error_report("Huh, first cluster in empty image is already in use?");
2264         abort();
2265     }
2266
2267     /* Create a full header (including things like feature table) */
2268     ret = qcow2_update_header(blk_bs(blk));
2269     if (ret < 0) {
2270         error_setg_errno(errp, -ret, "Could not update qcow2 header");
2271         goto out;
2272     }
2273
2274     /* Okay, now that we have a valid image, let's give it the right size */
2275     ret = blk_truncate(blk, total_size);
2276     if (ret < 0) {
2277         error_setg_errno(errp, -ret, "Could not resize image");
2278         goto out;
2279     }
2280
2281     /* Want a backing file? There you go.*/
2282     if (backing_file) {
2283         ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format);
2284         if (ret < 0) {
2285             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
2286                              "with format '%s'", backing_file, backing_format);
2287             goto out;
2288         }
2289     }
2290
2291     /* And if we're supposed to preallocate metadata, do that now */
2292     if (prealloc != PREALLOC_MODE_OFF) {
2293         BDRVQcow2State *s = blk_bs(blk)->opaque;
2294         qemu_co_mutex_lock(&s->lock);
2295         ret = preallocate(blk_bs(blk));
2296         qemu_co_mutex_unlock(&s->lock);
2297         if (ret < 0) {
2298             error_setg_errno(errp, -ret, "Could not preallocate metadata");
2299             goto out;
2300         }
2301     }
2302
2303     blk_unref(blk);
2304     blk = NULL;
2305
2306     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
2307     options = qdict_new();
2308     qdict_put(options, "driver", qstring_from_str("qcow2"));
2309     blk = blk_new_open(filename, NULL, options,
2310                        BDRV_O_RDWR | BDRV_O_NO_BACKING, &local_err);
2311     if (blk == NULL) {
2312         error_propagate(errp, local_err);
2313         ret = -EIO;
2314         goto out;
2315     }
2316
2317     ret = 0;
2318 out:
2319     if (blk) {
2320         blk_unref(blk);
2321     }
2322     return ret;
2323 }
2324
2325 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
2326 {
2327     char *backing_file = NULL;
2328     char *backing_fmt = NULL;
2329     char *buf = NULL;
2330     uint64_t size = 0;
2331     int flags = 0;
2332     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
2333     PreallocMode prealloc;
2334     int version = 3;
2335     uint64_t refcount_bits = 16;
2336     int refcount_order;
2337     Error *local_err = NULL;
2338     int ret;
2339
2340     /* Read out options */
2341     size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2342                     BDRV_SECTOR_SIZE);
2343     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2344     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2345     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2346         flags |= BLOCK_FLAG_ENCRYPT;
2347     }
2348     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2349                                          DEFAULT_CLUSTER_SIZE);
2350     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2351     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
2352                                PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
2353                                &local_err);
2354     if (local_err) {
2355         error_propagate(errp, local_err);
2356         ret = -EINVAL;
2357         goto finish;
2358     }
2359     g_free(buf);
2360     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2361     if (!buf) {
2362         /* keep the default */
2363     } else if (!strcmp(buf, "0.10")) {
2364         version = 2;
2365     } else if (!strcmp(buf, "1.1")) {
2366         version = 3;
2367     } else {
2368         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2369         ret = -EINVAL;
2370         goto finish;
2371     }
2372
2373     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2374         flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2375     }
2376
2377     if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2378         error_setg(errp, "Backing file and preallocation cannot be used at "
2379                    "the same time");
2380         ret = -EINVAL;
2381         goto finish;
2382     }
2383
2384     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2385         error_setg(errp, "Lazy refcounts only supported with compatibility "
2386                    "level 1.1 and above (use compat=1.1 or greater)");
2387         ret = -EINVAL;
2388         goto finish;
2389     }
2390
2391     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS,
2392                                             refcount_bits);
2393     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2394         error_setg(errp, "Refcount width must be a power of two and may not "
2395                    "exceed 64 bits");
2396         ret = -EINVAL;
2397         goto finish;
2398     }
2399
2400     if (version < 3 && refcount_bits != 16) {
2401         error_setg(errp, "Different refcount widths than 16 bits require "
2402                    "compatibility level 1.1 or above (use compat=1.1 or "
2403                    "greater)");
2404         ret = -EINVAL;
2405         goto finish;
2406     }
2407
2408     refcount_order = ctz32(refcount_bits);
2409
2410     ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2411                         cluster_size, prealloc, opts, version, refcount_order,
2412                         &local_err);
2413     error_propagate(errp, local_err);
2414
2415 finish:
2416     g_free(backing_file);
2417     g_free(backing_fmt);
2418     g_free(buf);
2419     return ret;
2420 }
2421
2422
2423 static bool is_zero_sectors(BlockDriverState *bs, int64_t start,
2424                             uint32_t count)
2425 {
2426     int nr;
2427     BlockDriverState *file;
2428     int64_t res;
2429
2430     if (!count) {
2431         return true;
2432     }
2433     res = bdrv_get_block_status_above(bs, NULL, start, count,
2434                                       &nr, &file);
2435     return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == count;
2436 }
2437
2438 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
2439     int64_t offset, int count, BdrvRequestFlags flags)
2440 {
2441     int ret;
2442     BDRVQcow2State *s = bs->opaque;
2443
2444     uint32_t head = offset % s->cluster_size;
2445     uint32_t tail = (offset + count) % s->cluster_size;
2446
2447     trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, count);
2448
2449     if (head || tail) {
2450         int64_t cl_start = (offset - head) >> BDRV_SECTOR_BITS;
2451         uint64_t off;
2452         unsigned int nr;
2453
2454         assert(head + count <= s->cluster_size);
2455
2456         /* check whether remainder of cluster already reads as zero */
2457         if (!(is_zero_sectors(bs, cl_start,
2458                               DIV_ROUND_UP(head, BDRV_SECTOR_SIZE)) &&
2459               is_zero_sectors(bs, (offset + count) >> BDRV_SECTOR_BITS,
2460                               DIV_ROUND_UP(-tail & (s->cluster_size - 1),
2461                                            BDRV_SECTOR_SIZE)))) {
2462             return -ENOTSUP;
2463         }
2464
2465         qemu_co_mutex_lock(&s->lock);
2466         /* We can have new write after previous check */
2467         offset = cl_start << BDRV_SECTOR_BITS;
2468         count = s->cluster_size;
2469         nr = s->cluster_size;
2470         ret = qcow2_get_cluster_offset(bs, offset, &nr, &off);
2471         if (ret != QCOW2_CLUSTER_UNALLOCATED && ret != QCOW2_CLUSTER_ZERO) {
2472             qemu_co_mutex_unlock(&s->lock);
2473             return -ENOTSUP;
2474         }
2475     } else {
2476         qemu_co_mutex_lock(&s->lock);
2477     }
2478
2479     trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, count);
2480
2481     /* Whatever is left can use real zero clusters */
2482     ret = qcow2_zero_clusters(bs, offset, count >> BDRV_SECTOR_BITS, flags);
2483     qemu_co_mutex_unlock(&s->lock);
2484
2485     return ret;
2486 }
2487
2488 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
2489                                           int64_t offset, int count)
2490 {
2491     int ret;
2492     BDRVQcow2State *s = bs->opaque;
2493
2494     if (!QEMU_IS_ALIGNED(offset | count, s->cluster_size)) {
2495         assert(count < s->cluster_size);
2496         return -ENOTSUP;
2497     }
2498
2499     qemu_co_mutex_lock(&s->lock);
2500     ret = qcow2_discard_clusters(bs, offset, count >> BDRV_SECTOR_BITS,
2501                                  QCOW2_DISCARD_REQUEST, false);
2502     qemu_co_mutex_unlock(&s->lock);
2503     return ret;
2504 }
2505
2506 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
2507 {
2508     BDRVQcow2State *s = bs->opaque;
2509     int64_t new_l1_size;
2510     int ret;
2511
2512     if (offset & 511) {
2513         error_report("The new size must be a multiple of 512");
2514         return -EINVAL;
2515     }
2516
2517     /* cannot proceed if image has snapshots */
2518     if (s->nb_snapshots) {
2519         error_report("Can't resize an image which has snapshots");
2520         return -ENOTSUP;
2521     }
2522
2523     /* shrinking is currently not supported */
2524     if (offset < bs->total_sectors * 512) {
2525         error_report("qcow2 doesn't support shrinking images yet");
2526         return -ENOTSUP;
2527     }
2528
2529     new_l1_size = size_to_l1(s, offset);
2530     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
2531     if (ret < 0) {
2532         return ret;
2533     }
2534
2535     /* write updated header.size */
2536     offset = cpu_to_be64(offset);
2537     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
2538                            &offset, sizeof(uint64_t));
2539     if (ret < 0) {
2540         return ret;
2541     }
2542
2543     s->l1_vm_state_index = new_l1_size;
2544     return 0;
2545 }
2546
2547 /* XXX: put compressed sectors first, then all the cluster aligned
2548    tables to avoid losing bytes in alignment */
2549 static coroutine_fn int
2550 qcow2_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
2551                             uint64_t bytes, QEMUIOVector *qiov)
2552 {
2553     BDRVQcow2State *s = bs->opaque;
2554     QEMUIOVector hd_qiov;
2555     struct iovec iov;
2556     z_stream strm;
2557     int ret, out_len;
2558     uint8_t *buf, *out_buf;
2559     uint64_t cluster_offset;
2560
2561     if (bytes == 0) {
2562         /* align end of file to a sector boundary to ease reading with
2563            sector based I/Os */
2564         cluster_offset = bdrv_getlength(bs->file->bs);
2565         return bdrv_truncate(bs->file->bs, cluster_offset);
2566     }
2567
2568     buf = qemu_blockalign(bs, s->cluster_size);
2569     if (bytes != s->cluster_size) {
2570         if (bytes > s->cluster_size ||
2571             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
2572         {
2573             qemu_vfree(buf);
2574             return -EINVAL;
2575         }
2576         /* Zero-pad last write if image size is not cluster aligned */
2577         memset(buf + bytes, 0, s->cluster_size - bytes);
2578     }
2579     qemu_iovec_to_buf(qiov, 0, buf, bytes);
2580
2581     out_buf = g_malloc(s->cluster_size);
2582
2583     /* best compression, small window, no zlib header */
2584     memset(&strm, 0, sizeof(strm));
2585     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
2586                        Z_DEFLATED, -12,
2587                        9, Z_DEFAULT_STRATEGY);
2588     if (ret != 0) {
2589         ret = -EINVAL;
2590         goto fail;
2591     }
2592
2593     strm.avail_in = s->cluster_size;
2594     strm.next_in = (uint8_t *)buf;
2595     strm.avail_out = s->cluster_size;
2596     strm.next_out = out_buf;
2597
2598     ret = deflate(&strm, Z_FINISH);
2599     if (ret != Z_STREAM_END && ret != Z_OK) {
2600         deflateEnd(&strm);
2601         ret = -EINVAL;
2602         goto fail;
2603     }
2604     out_len = strm.next_out - out_buf;
2605
2606     deflateEnd(&strm);
2607
2608     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
2609         /* could not compress: write normal cluster */
2610         ret = qcow2_co_pwritev(bs, offset, bytes, qiov, 0);
2611         if (ret < 0) {
2612             goto fail;
2613         }
2614         goto success;
2615     }
2616
2617     qemu_co_mutex_lock(&s->lock);
2618     cluster_offset =
2619         qcow2_alloc_compressed_cluster_offset(bs, offset, out_len);
2620     if (!cluster_offset) {
2621         qemu_co_mutex_unlock(&s->lock);
2622         ret = -EIO;
2623         goto fail;
2624     }
2625     cluster_offset &= s->cluster_offset_mask;
2626
2627     ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
2628     qemu_co_mutex_unlock(&s->lock);
2629     if (ret < 0) {
2630         goto fail;
2631     }
2632
2633     iov = (struct iovec) {
2634         .iov_base   = out_buf,
2635         .iov_len    = out_len,
2636     };
2637     qemu_iovec_init_external(&hd_qiov, &iov, 1);
2638
2639     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
2640     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
2641     if (ret < 0) {
2642         goto fail;
2643     }
2644 success:
2645     ret = 0;
2646 fail:
2647     qemu_vfree(buf);
2648     g_free(out_buf);
2649     return ret;
2650 }
2651
2652 static int make_completely_empty(BlockDriverState *bs)
2653 {
2654     BDRVQcow2State *s = bs->opaque;
2655     int ret, l1_clusters;
2656     int64_t offset;
2657     uint64_t *new_reftable = NULL;
2658     uint64_t rt_entry, l1_size2;
2659     struct {
2660         uint64_t l1_offset;
2661         uint64_t reftable_offset;
2662         uint32_t reftable_clusters;
2663     } QEMU_PACKED l1_ofs_rt_ofs_cls;
2664
2665     ret = qcow2_cache_empty(bs, s->l2_table_cache);
2666     if (ret < 0) {
2667         goto fail;
2668     }
2669
2670     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
2671     if (ret < 0) {
2672         goto fail;
2673     }
2674
2675     /* Refcounts will be broken utterly */
2676     ret = qcow2_mark_dirty(bs);
2677     if (ret < 0) {
2678         goto fail;
2679     }
2680
2681     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2682
2683     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2684     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
2685
2686     /* After this call, neither the in-memory nor the on-disk refcount
2687      * information accurately describe the actual references */
2688
2689     ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
2690                              l1_clusters * s->cluster_size, 0);
2691     if (ret < 0) {
2692         goto fail_broken_refcounts;
2693     }
2694     memset(s->l1_table, 0, l1_size2);
2695
2696     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
2697
2698     /* Overwrite enough clusters at the beginning of the sectors to place
2699      * the refcount table, a refcount block and the L1 table in; this may
2700      * overwrite parts of the existing refcount and L1 table, which is not
2701      * an issue because the dirty flag is set, complete data loss is in fact
2702      * desired and partial data loss is consequently fine as well */
2703     ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
2704                              (2 + l1_clusters) * s->cluster_size, 0);
2705     /* This call (even if it failed overall) may have overwritten on-disk
2706      * refcount structures; in that case, the in-memory refcount information
2707      * will probably differ from the on-disk information which makes the BDS
2708      * unusable */
2709     if (ret < 0) {
2710         goto fail_broken_refcounts;
2711     }
2712
2713     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2714     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
2715
2716     /* "Create" an empty reftable (one cluster) directly after the image
2717      * header and an empty L1 table three clusters after the image header;
2718      * the cluster between those two will be used as the first refblock */
2719     l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
2720     l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
2721     l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
2722     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
2723                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
2724     if (ret < 0) {
2725         goto fail_broken_refcounts;
2726     }
2727
2728     s->l1_table_offset = 3 * s->cluster_size;
2729
2730     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
2731     if (!new_reftable) {
2732         ret = -ENOMEM;
2733         goto fail_broken_refcounts;
2734     }
2735
2736     s->refcount_table_offset = s->cluster_size;
2737     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
2738
2739     g_free(s->refcount_table);
2740     s->refcount_table = new_reftable;
2741     new_reftable = NULL;
2742
2743     /* Now the in-memory refcount information again corresponds to the on-disk
2744      * information (reftable is empty and no refblocks (the refblock cache is
2745      * empty)); however, this means some clusters (e.g. the image header) are
2746      * referenced, but not refcounted, but the normal qcow2 code assumes that
2747      * the in-memory information is always correct */
2748
2749     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
2750
2751     /* Enter the first refblock into the reftable */
2752     rt_entry = cpu_to_be64(2 * s->cluster_size);
2753     ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
2754                            &rt_entry, sizeof(rt_entry));
2755     if (ret < 0) {
2756         goto fail_broken_refcounts;
2757     }
2758     s->refcount_table[0] = 2 * s->cluster_size;
2759
2760     s->free_cluster_index = 0;
2761     assert(3 + l1_clusters <= s->refcount_block_size);
2762     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
2763     if (offset < 0) {
2764         ret = offset;
2765         goto fail_broken_refcounts;
2766     } else if (offset > 0) {
2767         error_report("First cluster in emptied image is in use");
2768         abort();
2769     }
2770
2771     /* Now finally the in-memory information corresponds to the on-disk
2772      * structures and is correct */
2773     ret = qcow2_mark_clean(bs);
2774     if (ret < 0) {
2775         goto fail;
2776     }
2777
2778     ret = bdrv_truncate(bs->file->bs, (3 + l1_clusters) * s->cluster_size);
2779     if (ret < 0) {
2780         goto fail;
2781     }
2782
2783     return 0;
2784
2785 fail_broken_refcounts:
2786     /* The BDS is unusable at this point. If we wanted to make it usable, we
2787      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
2788      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
2789      * again. However, because the functions which could have caused this error
2790      * path to be taken are used by those functions as well, it's very likely
2791      * that that sequence will fail as well. Therefore, just eject the BDS. */
2792     bs->drv = NULL;
2793
2794 fail:
2795     g_free(new_reftable);
2796     return ret;
2797 }
2798
2799 static int qcow2_make_empty(BlockDriverState *bs)
2800 {
2801     BDRVQcow2State *s = bs->opaque;
2802     uint64_t start_sector;
2803     int sector_step = INT_MAX / BDRV_SECTOR_SIZE;
2804     int l1_clusters, ret = 0;
2805
2806     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2807
2808     if (s->qcow_version >= 3 && !s->snapshots &&
2809         3 + l1_clusters <= s->refcount_block_size) {
2810         /* The following function only works for qcow2 v3 images (it requires
2811          * the dirty flag) and only as long as there are no snapshots (because
2812          * it completely empties the image). Furthermore, the L1 table and three
2813          * additional clusters (image header, refcount table, one refcount
2814          * block) have to fit inside one refcount block. */
2815         return make_completely_empty(bs);
2816     }
2817
2818     /* This fallback code simply discards every active cluster; this is slow,
2819      * but works in all cases */
2820     for (start_sector = 0; start_sector < bs->total_sectors;
2821          start_sector += sector_step)
2822     {
2823         /* As this function is generally used after committing an external
2824          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
2825          * default action for this kind of discard is to pass the discard,
2826          * which will ideally result in an actually smaller image file, as
2827          * is probably desired. */
2828         ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE,
2829                                      MIN(sector_step,
2830                                          bs->total_sectors - start_sector),
2831                                      QCOW2_DISCARD_SNAPSHOT, true);
2832         if (ret < 0) {
2833             break;
2834         }
2835     }
2836
2837     return ret;
2838 }
2839
2840 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
2841 {
2842     BDRVQcow2State *s = bs->opaque;
2843     int ret;
2844
2845     qemu_co_mutex_lock(&s->lock);
2846     ret = qcow2_cache_write(bs, s->l2_table_cache);
2847     if (ret < 0) {
2848         qemu_co_mutex_unlock(&s->lock);
2849         return ret;
2850     }
2851
2852     if (qcow2_need_accurate_refcounts(s)) {
2853         ret = qcow2_cache_write(bs, s->refcount_block_cache);
2854         if (ret < 0) {
2855             qemu_co_mutex_unlock(&s->lock);
2856             return ret;
2857         }
2858     }
2859     qemu_co_mutex_unlock(&s->lock);
2860
2861     return 0;
2862 }
2863
2864 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2865 {
2866     BDRVQcow2State *s = bs->opaque;
2867     bdi->unallocated_blocks_are_zero = true;
2868     bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
2869     bdi->cluster_size = s->cluster_size;
2870     bdi->vm_state_offset = qcow2_vm_state_offset(s);
2871     return 0;
2872 }
2873
2874 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
2875 {
2876     BDRVQcow2State *s = bs->opaque;
2877     ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
2878
2879     *spec_info = (ImageInfoSpecific){
2880         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
2881         .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1),
2882     };
2883     if (s->qcow_version == 2) {
2884         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
2885             .compat             = g_strdup("0.10"),
2886             .refcount_bits      = s->refcount_bits,
2887         };
2888     } else if (s->qcow_version == 3) {
2889         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
2890             .compat             = g_strdup("1.1"),
2891             .lazy_refcounts     = s->compatible_features &
2892                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
2893             .has_lazy_refcounts = true,
2894             .corrupt            = s->incompatible_features &
2895                                   QCOW2_INCOMPAT_CORRUPT,
2896             .has_corrupt        = true,
2897             .refcount_bits      = s->refcount_bits,
2898         };
2899     } else {
2900         /* if this assertion fails, this probably means a new version was
2901          * added without having it covered here */
2902         assert(false);
2903     }
2904
2905     return spec_info;
2906 }
2907
2908 #if 0
2909 static void dump_refcounts(BlockDriverState *bs)
2910 {
2911     BDRVQcow2State *s = bs->opaque;
2912     int64_t nb_clusters, k, k1, size;
2913     int refcount;
2914
2915     size = bdrv_getlength(bs->file->bs);
2916     nb_clusters = size_to_clusters(s, size);
2917     for(k = 0; k < nb_clusters;) {
2918         k1 = k;
2919         refcount = get_refcount(bs, k);
2920         k++;
2921         while (k < nb_clusters && get_refcount(bs, k) == refcount)
2922             k++;
2923         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
2924                k - k1);
2925     }
2926 }
2927 #endif
2928
2929 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2930                               int64_t pos)
2931 {
2932     BDRVQcow2State *s = bs->opaque;
2933
2934     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
2935     return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos,
2936                                     qiov->size, qiov, 0);
2937 }
2938
2939 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2940                               int64_t pos)
2941 {
2942     BDRVQcow2State *s = bs->opaque;
2943
2944     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
2945     return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos,
2946                                    qiov->size, qiov, 0);
2947 }
2948
2949 /*
2950  * Downgrades an image's version. To achieve this, any incompatible features
2951  * have to be removed.
2952  */
2953 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
2954                            BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
2955 {
2956     BDRVQcow2State *s = bs->opaque;
2957     int current_version = s->qcow_version;
2958     int ret;
2959
2960     if (target_version == current_version) {
2961         return 0;
2962     } else if (target_version > current_version) {
2963         return -EINVAL;
2964     } else if (target_version != 2) {
2965         return -EINVAL;
2966     }
2967
2968     if (s->refcount_order != 4) {
2969         error_report("compat=0.10 requires refcount_bits=16");
2970         return -ENOTSUP;
2971     }
2972
2973     /* clear incompatible features */
2974     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
2975         ret = qcow2_mark_clean(bs);
2976         if (ret < 0) {
2977             return ret;
2978         }
2979     }
2980
2981     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2982      * the first place; if that happens nonetheless, returning -ENOTSUP is the
2983      * best thing to do anyway */
2984
2985     if (s->incompatible_features) {
2986         return -ENOTSUP;
2987     }
2988
2989     /* since we can ignore compatible features, we can set them to 0 as well */
2990     s->compatible_features = 0;
2991     /* if lazy refcounts have been used, they have already been fixed through
2992      * clearing the dirty flag */
2993
2994     /* clearing autoclear features is trivial */
2995     s->autoclear_features = 0;
2996
2997     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
2998     if (ret < 0) {
2999         return ret;
3000     }
3001
3002     s->qcow_version = target_version;
3003     ret = qcow2_update_header(bs);
3004     if (ret < 0) {
3005         s->qcow_version = current_version;
3006         return ret;
3007     }
3008     return 0;
3009 }
3010
3011 typedef enum Qcow2AmendOperation {
3012     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
3013      * statically initialized to so that the helper CB can discern the first
3014      * invocation from an operation change */
3015     QCOW2_NO_OPERATION = 0,
3016
3017     QCOW2_CHANGING_REFCOUNT_ORDER,
3018     QCOW2_DOWNGRADING,
3019 } Qcow2AmendOperation;
3020
3021 typedef struct Qcow2AmendHelperCBInfo {
3022     /* The code coordinating the amend operations should only modify
3023      * these four fields; the rest will be managed by the CB */
3024     BlockDriverAmendStatusCB *original_status_cb;
3025     void *original_cb_opaque;
3026
3027     Qcow2AmendOperation current_operation;
3028
3029     /* Total number of operations to perform (only set once) */
3030     int total_operations;
3031
3032     /* The following fields are managed by the CB */
3033
3034     /* Number of operations completed */
3035     int operations_completed;
3036
3037     /* Cumulative offset of all completed operations */
3038     int64_t offset_completed;
3039
3040     Qcow2AmendOperation last_operation;
3041     int64_t last_work_size;
3042 } Qcow2AmendHelperCBInfo;
3043
3044 static void qcow2_amend_helper_cb(BlockDriverState *bs,
3045                                   int64_t operation_offset,
3046                                   int64_t operation_work_size, void *opaque)
3047 {
3048     Qcow2AmendHelperCBInfo *info = opaque;
3049     int64_t current_work_size;
3050     int64_t projected_work_size;
3051
3052     if (info->current_operation != info->last_operation) {
3053         if (info->last_operation != QCOW2_NO_OPERATION) {
3054             info->offset_completed += info->last_work_size;
3055             info->operations_completed++;
3056         }
3057
3058         info->last_operation = info->current_operation;
3059     }
3060
3061     assert(info->total_operations > 0);
3062     assert(info->operations_completed < info->total_operations);
3063
3064     info->last_work_size = operation_work_size;
3065
3066     current_work_size = info->offset_completed + operation_work_size;
3067
3068     /* current_work_size is the total work size for (operations_completed + 1)
3069      * operations (which includes this one), so multiply it by the number of
3070      * operations not covered and divide it by the number of operations
3071      * covered to get a projection for the operations not covered */
3072     projected_work_size = current_work_size * (info->total_operations -
3073                                                info->operations_completed - 1)
3074                                             / (info->operations_completed + 1);
3075
3076     info->original_status_cb(bs, info->offset_completed + operation_offset,
3077                              current_work_size + projected_work_size,
3078                              info->original_cb_opaque);
3079 }
3080
3081 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
3082                                BlockDriverAmendStatusCB *status_cb,
3083                                void *cb_opaque)
3084 {
3085     BDRVQcow2State *s = bs->opaque;
3086     int old_version = s->qcow_version, new_version = old_version;
3087     uint64_t new_size = 0;
3088     const char *backing_file = NULL, *backing_format = NULL;
3089     bool lazy_refcounts = s->use_lazy_refcounts;
3090     const char *compat = NULL;
3091     uint64_t cluster_size = s->cluster_size;
3092     bool encrypt;
3093     int refcount_bits = s->refcount_bits;
3094     int ret;
3095     QemuOptDesc *desc = opts->list->desc;
3096     Qcow2AmendHelperCBInfo helper_cb_info;
3097
3098     while (desc && desc->name) {
3099         if (!qemu_opt_find(opts, desc->name)) {
3100             /* only change explicitly defined options */
3101             desc++;
3102             continue;
3103         }
3104
3105         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
3106             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
3107             if (!compat) {
3108                 /* preserve default */
3109             } else if (!strcmp(compat, "0.10")) {
3110                 new_version = 2;
3111             } else if (!strcmp(compat, "1.1")) {
3112                 new_version = 3;
3113             } else {
3114                 error_report("Unknown compatibility level %s", compat);
3115                 return -EINVAL;
3116             }
3117         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
3118             error_report("Cannot change preallocation mode");
3119             return -ENOTSUP;
3120         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
3121             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
3122         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
3123             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
3124         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
3125             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
3126         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
3127             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
3128                                         !!s->cipher);
3129
3130             if (encrypt != !!s->cipher) {
3131                 error_report("Changing the encryption flag is not supported");
3132                 return -ENOTSUP;
3133             }
3134         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
3135             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
3136                                              cluster_size);
3137             if (cluster_size != s->cluster_size) {
3138                 error_report("Changing the cluster size is not supported");
3139                 return -ENOTSUP;
3140             }
3141         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
3142             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
3143                                                lazy_refcounts);
3144         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
3145             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
3146                                                 refcount_bits);
3147
3148             if (refcount_bits <= 0 || refcount_bits > 64 ||
3149                 !is_power_of_2(refcount_bits))
3150             {
3151                 error_report("Refcount width must be a power of two and may "
3152                              "not exceed 64 bits");
3153                 return -EINVAL;
3154             }
3155         } else {
3156             /* if this point is reached, this probably means a new option was
3157              * added without having it covered here */
3158             abort();
3159         }
3160
3161         desc++;
3162     }
3163
3164     helper_cb_info = (Qcow2AmendHelperCBInfo){
3165         .original_status_cb = status_cb,
3166         .original_cb_opaque = cb_opaque,
3167         .total_operations = (new_version < old_version)
3168                           + (s->refcount_bits != refcount_bits)
3169     };
3170
3171     /* Upgrade first (some features may require compat=1.1) */
3172     if (new_version > old_version) {
3173         s->qcow_version = new_version;
3174         ret = qcow2_update_header(bs);
3175         if (ret < 0) {
3176             s->qcow_version = old_version;
3177             return ret;
3178         }
3179     }
3180
3181     if (s->refcount_bits != refcount_bits) {
3182         int refcount_order = ctz32(refcount_bits);
3183         Error *local_error = NULL;
3184
3185         if (new_version < 3 && refcount_bits != 16) {
3186             error_report("Different refcount widths than 16 bits require "
3187                          "compatibility level 1.1 or above (use compat=1.1 or "
3188                          "greater)");
3189             return -EINVAL;
3190         }
3191
3192         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
3193         ret = qcow2_change_refcount_order(bs, refcount_order,
3194                                           &qcow2_amend_helper_cb,
3195                                           &helper_cb_info, &local_error);
3196         if (ret < 0) {
3197             error_report_err(local_error);
3198             return ret;
3199         }
3200     }
3201
3202     if (backing_file || backing_format) {
3203         ret = qcow2_change_backing_file(bs,
3204                     backing_file ?: s->image_backing_file,
3205                     backing_format ?: s->image_backing_format);
3206         if (ret < 0) {
3207             return ret;
3208         }
3209     }
3210
3211     if (s->use_lazy_refcounts != lazy_refcounts) {
3212         if (lazy_refcounts) {
3213             if (new_version < 3) {
3214                 error_report("Lazy refcounts only supported with compatibility "
3215                              "level 1.1 and above (use compat=1.1 or greater)");
3216                 return -EINVAL;
3217             }
3218             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3219             ret = qcow2_update_header(bs);
3220             if (ret < 0) {
3221                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3222                 return ret;
3223             }
3224             s->use_lazy_refcounts = true;
3225         } else {
3226             /* make image clean first */
3227             ret = qcow2_mark_clean(bs);
3228             if (ret < 0) {
3229                 return ret;
3230             }
3231             /* now disallow lazy refcounts */
3232             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3233             ret = qcow2_update_header(bs);
3234             if (ret < 0) {
3235                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3236                 return ret;
3237             }
3238             s->use_lazy_refcounts = false;
3239         }
3240     }
3241
3242     if (new_size) {
3243         ret = bdrv_truncate(bs, new_size);
3244         if (ret < 0) {
3245             return ret;
3246         }
3247     }
3248
3249     /* Downgrade last (so unsupported features can be removed before) */
3250     if (new_version < old_version) {
3251         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
3252         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
3253                               &helper_cb_info);
3254         if (ret < 0) {
3255             return ret;
3256         }
3257     }
3258
3259     return 0;
3260 }
3261
3262 /*
3263  * If offset or size are negative, respectively, they will not be included in
3264  * the BLOCK_IMAGE_CORRUPTED event emitted.
3265  * fatal will be ignored for read-only BDS; corruptions found there will always
3266  * be considered non-fatal.
3267  */
3268 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
3269                              int64_t size, const char *message_format, ...)
3270 {
3271     BDRVQcow2State *s = bs->opaque;
3272     const char *node_name;
3273     char *message;
3274     va_list ap;
3275
3276     fatal = fatal && !bs->read_only;
3277
3278     if (s->signaled_corruption &&
3279         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
3280     {
3281         return;
3282     }
3283
3284     va_start(ap, message_format);
3285     message = g_strdup_vprintf(message_format, ap);
3286     va_end(ap);
3287
3288     if (fatal) {
3289         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
3290                 "corruption events will be suppressed\n", message);
3291     } else {
3292         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
3293                 "corruption events will be suppressed\n", message);
3294     }
3295
3296     node_name = bdrv_get_node_name(bs);
3297     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
3298                                           *node_name != '\0', node_name,
3299                                           message, offset >= 0, offset,
3300                                           size >= 0, size,
3301                                           fatal, &error_abort);
3302     g_free(message);
3303
3304     if (fatal) {
3305         qcow2_mark_corrupt(bs);
3306         bs->drv = NULL; /* make BDS unusable */
3307     }
3308
3309     s->signaled_corruption = true;
3310 }
3311
3312 static QemuOptsList qcow2_create_opts = {
3313     .name = "qcow2-create-opts",
3314     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
3315     .desc = {
3316         {
3317             .name = BLOCK_OPT_SIZE,
3318             .type = QEMU_OPT_SIZE,
3319             .help = "Virtual disk size"
3320         },
3321         {
3322             .name = BLOCK_OPT_COMPAT_LEVEL,
3323             .type = QEMU_OPT_STRING,
3324             .help = "Compatibility level (0.10 or 1.1)"
3325         },
3326         {
3327             .name = BLOCK_OPT_BACKING_FILE,
3328             .type = QEMU_OPT_STRING,
3329             .help = "File name of a base image"
3330         },
3331         {
3332             .name = BLOCK_OPT_BACKING_FMT,
3333             .type = QEMU_OPT_STRING,
3334             .help = "Image format of the base image"
3335         },
3336         {
3337             .name = BLOCK_OPT_ENCRYPT,
3338             .type = QEMU_OPT_BOOL,
3339             .help = "Encrypt the image",
3340             .def_value_str = "off"
3341         },
3342         {
3343             .name = BLOCK_OPT_CLUSTER_SIZE,
3344             .type = QEMU_OPT_SIZE,
3345             .help = "qcow2 cluster size",
3346             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
3347         },
3348         {
3349             .name = BLOCK_OPT_PREALLOC,
3350             .type = QEMU_OPT_STRING,
3351             .help = "Preallocation mode (allowed values: off, metadata, "
3352                     "falloc, full)"
3353         },
3354         {
3355             .name = BLOCK_OPT_LAZY_REFCOUNTS,
3356             .type = QEMU_OPT_BOOL,
3357             .help = "Postpone refcount updates",
3358             .def_value_str = "off"
3359         },
3360         {
3361             .name = BLOCK_OPT_REFCOUNT_BITS,
3362             .type = QEMU_OPT_NUMBER,
3363             .help = "Width of a reference count entry in bits",
3364             .def_value_str = "16"
3365         },
3366         { /* end of list */ }
3367     }
3368 };
3369
3370 BlockDriver bdrv_qcow2 = {
3371     .format_name        = "qcow2",
3372     .instance_size      = sizeof(BDRVQcow2State),
3373     .bdrv_probe         = qcow2_probe,
3374     .bdrv_open          = qcow2_open,
3375     .bdrv_close         = qcow2_close,
3376     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
3377     .bdrv_reopen_commit   = qcow2_reopen_commit,
3378     .bdrv_reopen_abort    = qcow2_reopen_abort,
3379     .bdrv_join_options    = qcow2_join_options,
3380     .bdrv_create        = qcow2_create,
3381     .bdrv_has_zero_init = bdrv_has_zero_init_1,
3382     .bdrv_co_get_block_status = qcow2_co_get_block_status,
3383     .bdrv_set_key       = qcow2_set_key,
3384
3385     .bdrv_co_preadv         = qcow2_co_preadv,
3386     .bdrv_co_pwritev        = qcow2_co_pwritev,
3387     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
3388
3389     .bdrv_co_pwrite_zeroes  = qcow2_co_pwrite_zeroes,
3390     .bdrv_co_pdiscard       = qcow2_co_pdiscard,
3391     .bdrv_truncate          = qcow2_truncate,
3392     .bdrv_co_pwritev_compressed = qcow2_co_pwritev_compressed,
3393     .bdrv_make_empty        = qcow2_make_empty,
3394
3395     .bdrv_snapshot_create   = qcow2_snapshot_create,
3396     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
3397     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
3398     .bdrv_snapshot_list     = qcow2_snapshot_list,
3399     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
3400     .bdrv_get_info          = qcow2_get_info,
3401     .bdrv_get_specific_info = qcow2_get_specific_info,
3402
3403     .bdrv_save_vmstate    = qcow2_save_vmstate,
3404     .bdrv_load_vmstate    = qcow2_load_vmstate,
3405
3406     .supports_backing           = true,
3407     .bdrv_change_backing_file   = qcow2_change_backing_file,
3408
3409     .bdrv_refresh_limits        = qcow2_refresh_limits,
3410     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
3411     .bdrv_inactivate            = qcow2_inactivate,
3412
3413     .create_opts         = &qcow2_create_opts,
3414     .bdrv_check          = qcow2_check,
3415     .bdrv_amend_options  = qcow2_amend_options,
3416
3417     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
3418     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
3419 };
3420
3421 static void bdrv_qcow2_init(void)
3422 {
3423     bdrv_register(&bdrv_qcow2);
3424 }
3425
3426 block_init(bdrv_qcow2_init);