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