qemu/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         if (bdrv_uses_whitelist() &&
 969             s->crypt_method_header == QCOW_CRYPT_AES) {
 970             error_report("qcow2 built-in AES encryption is deprecated");
 971             error_printf("Support for it will be removed in a future release.\n"
 972                          "You can use 'qemu-img convert' to switch to an\n"
 973                          "unencrypted qcow2 image, or a LUKS raw image.\n");
 974         }
 975
 976         bs->encrypted = 1;
 977     }
 978
 979     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
 980     s->l2_size = 1 << s->l2_bits;
 981     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
 982     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
 983     s->refcount_block_size = 1 << s->refcount_block_bits;
 984     bs->total_sectors = header.size / 512;
 985     s->csize_shift = (62 - (s->cluster_bits - 8));
 986     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
 987     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
 988
 989     s->refcount_table_offset = header.refcount_table_offset;
 990     s->refcount_table_size =
 991         header.refcount_table_clusters << (s->cluster_bits - 3);
 992
 993     if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
 994         error_setg(errp, "Reference count table too large");
 995         ret = -EINVAL;
 996         goto fail;
 997     }
 998
 999     ret = validate_table_offset(bs, s->refcount_table_offset,
1000                                 s->refcount_table_size, sizeof(uint64_t));
1001     if (ret < 0) {
1002         error_setg(errp, "Invalid reference count table offset");
1003         goto fail;
1004     }
1005
1006     /* Snapshot table offset/length */
1007     if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
1008         error_setg(errp, "Too many snapshots");
1009         ret = -EINVAL;
1010         goto fail;
1011     }
1012
1013     ret = validate_table_offset(bs, header.snapshots_offset,
1014                                 header.nb_snapshots,
1015                                 sizeof(QCowSnapshotHeader));
1016     if (ret < 0) {
1017         error_setg(errp, "Invalid snapshot table offset");
1018         goto fail;
1019     }
1020
1021     /* read the level 1 table */
1022     if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
1023         error_setg(errp, "Active L1 table too large");
1024         ret = -EFBIG;
1025         goto fail;
1026     }
1027     s->l1_size = header.l1_size;
1028
1029     l1_vm_state_index = size_to_l1(s, header.size);
1030     if (l1_vm_state_index > INT_MAX) {
1031         error_setg(errp, "Image is too big");
1032         ret = -EFBIG;
1033         goto fail;
1034     }
1035     s->l1_vm_state_index = l1_vm_state_index;
1036
1037     /* the L1 table must contain at least enough entries to put
1038        header.size bytes */
1039     if (s->l1_size < s->l1_vm_state_index) {
1040         error_setg(errp, "L1 table is too small");
1041         ret = -EINVAL;
1042         goto fail;
1043     }
1044
1045     ret = validate_table_offset(bs, header.l1_table_offset,
1046                                 header.l1_size, sizeof(uint64_t));
1047     if (ret < 0) {
1048         error_setg(errp, "Invalid L1 table offset");
1049         goto fail;
1050     }
1051     s->l1_table_offset = header.l1_table_offset;
1052
1053
1054     if (s->l1_size > 0) {
1055         s->l1_table = qemu_try_blockalign(bs->file->bs,
1056             align_offset(s->l1_size * sizeof(uint64_t), 512));
1057         if (s->l1_table == NULL) {
1058             error_setg(errp, "Could not allocate L1 table");
1059             ret = -ENOMEM;
1060             goto fail;
1061         }
1062         ret = bdrv_pread(bs->file->bs, s->l1_table_offset, s->l1_table,
1063                          s->l1_size * sizeof(uint64_t));
1064         if (ret < 0) {
1065             error_setg_errno(errp, -ret, "Could not read L1 table");
1066             goto fail;
1067         }
1068         for(i = 0;i < s->l1_size; i++) {
1069             be64_to_cpus(&s->l1_table[i]);
1070         }
1071     }
1072
1073     /* Parse driver-specific options */
1074     ret = qcow2_update_options(bs, options, flags, errp);
1075     if (ret < 0) {
1076         goto fail;
1077     }
1078
1079     s->cluster_cache = g_malloc(s->cluster_size);
1080     /* one more sector for decompressed data alignment */
1081     s->cluster_data = qemu_try_blockalign(bs->file->bs, QCOW_MAX_CRYPT_CLUSTERS
1082                                                     * s->cluster_size + 512);
1083     if (s->cluster_data == NULL) {
1084         error_setg(errp, "Could not allocate temporary cluster buffer");
1085         ret = -ENOMEM;
1086         goto fail;
1087     }
1088
1089     s->cluster_cache_offset = -1;
1090     s->flags = flags;
1091
1092     ret = qcow2_refcount_init(bs);
1093     if (ret != 0) {
1094         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1095         goto fail;
1096     }
1097
1098     QLIST_INIT(&s->cluster_allocs);
1099     QTAILQ_INIT(&s->discards);
1100
1101     /* read qcow2 extensions */
1102     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1103         &local_err)) {
1104         error_propagate(errp, local_err);
1105         ret = -EINVAL;
1106         goto fail;
1107     }
1108
1109     /* read the backing file name */
1110     if (header.backing_file_offset != 0) {
1111         len = header.backing_file_size;
1112         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1113             len >= sizeof(bs->backing_file)) {
1114             error_setg(errp, "Backing file name too long");
1115             ret = -EINVAL;
1116             goto fail;
1117         }
1118         ret = bdrv_pread(bs->file->bs, header.backing_file_offset,
1119                          bs->backing_file, len);
1120         if (ret < 0) {
1121             error_setg_errno(errp, -ret, "Could not read backing file name");
1122             goto fail;
1123         }
1124         bs->backing_file[len] = '\0';
1125         s->image_backing_file = g_strdup(bs->backing_file);
1126     }
1127
1128     /* Internal snapshots */
1129     s->snapshots_offset = header.snapshots_offset;
1130     s->nb_snapshots = header.nb_snapshots;
1131
1132     ret = qcow2_read_snapshots(bs);
1133     if (ret < 0) {
1134         error_setg_errno(errp, -ret, "Could not read snapshots");
1135         goto fail;
1136     }
1137
1138     /* Clear unknown autoclear feature bits */
1139     if (!bs->read_only && !(flags & BDRV_O_INACTIVE) && s->autoclear_features) {
1140         s->autoclear_features = 0;
1141         ret = qcow2_update_header(bs);
1142         if (ret < 0) {
1143             error_setg_errno(errp, -ret, "Could not update qcow2 header");
1144             goto fail;
1145         }
1146     }
1147
1148     /* Initialise locks */
1149     qemu_co_mutex_init(&s->lock);
1150
1151     /* Repair image if dirty */
1152     if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1153         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1154         BdrvCheckResult result = {0};
1155
1156         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1157         if (ret < 0) {
1158             error_setg_errno(errp, -ret, "Could not repair dirty image");
1159             goto fail;
1160         }
1161     }
1162
1163 #ifdef DEBUG_ALLOC
1164     {
1165         BdrvCheckResult result = {0};
1166         qcow2_check_refcounts(bs, &result, 0);
1167     }
1168 #endif
1169     return ret;
1170
1171  fail:
1172     g_free(s->unknown_header_fields);
1173     cleanup_unknown_header_ext(bs);
1174     qcow2_free_snapshots(bs);
1175     qcow2_refcount_close(bs);
1176     qemu_vfree(s->l1_table);
1177     /* else pre-write overlap checks in cache_destroy may crash */
1178     s->l1_table = NULL;
1179     cache_clean_timer_del(bs);
1180     if (s->l2_table_cache) {
1181         qcow2_cache_destroy(bs, s->l2_table_cache);
1182     }
1183     if (s->refcount_block_cache) {
1184         qcow2_cache_destroy(bs, s->refcount_block_cache);
1185     }
1186     g_free(s->cluster_cache);
1187     qemu_vfree(s->cluster_data);
1188     return ret;
1189 }
1190
1191 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1192 {
1193     BDRVQcow2State *s = bs->opaque;
1194
1195     bs->bl.write_zeroes_alignment = s->cluster_sectors;
1196 }
1197
1198 static int qcow2_set_key(BlockDriverState *bs, const char *key)
1199 {
1200     BDRVQcow2State *s = bs->opaque;
1201     uint8_t keybuf[16];
1202     int len, i;
1203     Error *err = NULL;
1204
1205     memset(keybuf, 0, 16);
1206     len = strlen(key);
1207     if (len > 16)
1208         len = 16;
1209     /* XXX: we could compress the chars to 7 bits to increase
1210        entropy */
1211     for(i = 0;i < len;i++) {
1212         keybuf[i] = key[i];
1213     }
1214     assert(bs->encrypted);
1215
1216     qcrypto_cipher_free(s->cipher);
1217     s->cipher = qcrypto_cipher_new(
1218         QCRYPTO_CIPHER_ALG_AES_128,
1219         QCRYPTO_CIPHER_MODE_CBC,
1220         keybuf, G_N_ELEMENTS(keybuf),
1221         &err);
1222
1223     if (!s->cipher) {
1224         /* XXX would be nice if errors in this method could
1225          * be properly propagate to the caller. Would need
1226          * the bdrv_set_key() API signature to be fixed. */
1227         error_free(err);
1228         return -1;
1229     }
1230     return 0;
1231 }
1232
1233 static int qcow2_reopen_prepare(BDRVReopenState *state,
1234                                 BlockReopenQueue *queue, Error **errp)
1235 {
1236     Qcow2ReopenState *r;
1237     int ret;
1238
1239     r = g_new0(Qcow2ReopenState, 1);
1240     state->opaque = r;
1241
1242     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1243                                        state->flags, errp);
1244     if (ret < 0) {
1245         goto fail;
1246     }
1247
1248     /* We need to write out any unwritten data if we reopen read-only. */
1249     if ((state->flags & BDRV_O_RDWR) == 0) {
1250         ret = bdrv_flush(state->bs);
1251         if (ret < 0) {
1252             goto fail;
1253         }
1254
1255         ret = qcow2_mark_clean(state->bs);
1256         if (ret < 0) {
1257             goto fail;
1258         }
1259     }
1260
1261     return 0;
1262
1263 fail:
1264     qcow2_update_options_abort(state->bs, r);
1265     g_free(r);
1266     return ret;
1267 }
1268
1269 static void qcow2_reopen_commit(BDRVReopenState *state)
1270 {
1271     qcow2_update_options_commit(state->bs, state->opaque);
1272     g_free(state->opaque);
1273 }
1274
1275 static void qcow2_reopen_abort(BDRVReopenState *state)
1276 {
1277     qcow2_update_options_abort(state->bs, state->opaque);
1278     g_free(state->opaque);
1279 }
1280
1281 static void qcow2_join_options(QDict *options, QDict *old_options)
1282 {
1283     bool has_new_overlap_template =
1284         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1285         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1286     bool has_new_total_cache_size =
1287         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1288     bool has_all_cache_options;
1289
1290     /* New overlap template overrides all old overlap options */
1291     if (has_new_overlap_template) {
1292         qdict_del(old_options, QCOW2_OPT_OVERLAP);
1293         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1294         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1295         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1296         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1297         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1298         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1299         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1300         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1301         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1302     }
1303
1304     /* New total cache size overrides all old options */
1305     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1306         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1307         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1308     }
1309
1310     qdict_join(options, old_options, false);
1311
1312     /*
1313      * If after merging all cache size options are set, an old total size is
1314      * overwritten. Do keep all options, however, if all three are new. The
1315      * resulting error message is what we want to happen.
1316      */
1317     has_all_cache_options =
1318         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1319         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1320         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1321
1322     if (has_all_cache_options && !has_new_total_cache_size) {
1323         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1324     }
1325 }
1326
1327 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1328         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1329 {
1330     BDRVQcow2State *s = bs->opaque;
1331     uint64_t cluster_offset;
1332     int index_in_cluster, ret;
1333     int64_t status = 0;
1334
1335     *pnum = nb_sectors;
1336     qemu_co_mutex_lock(&s->lock);
1337     ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
1338     qemu_co_mutex_unlock(&s->lock);
1339     if (ret < 0) {
1340         return ret;
1341     }
1342
1343     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1344         !s->cipher) {
1345         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1346         cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1347         *file = bs->file->bs;
1348         status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1349     }
1350     if (ret == QCOW2_CLUSTER_ZERO) {
1351         status |= BDRV_BLOCK_ZERO;
1352     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1353         status |= BDRV_BLOCK_DATA;
1354     }
1355     return status;
1356 }
1357
1358 /* handle reading after the end of the backing file */
1359 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
1360                   int64_t sector_num, int nb_sectors)
1361 {
1362     int n1;
1363     if ((sector_num + nb_sectors) <= bs->total_sectors)
1364         return nb_sectors;
1365     if (sector_num >= bs->total_sectors)
1366         n1 = 0;
1367     else
1368         n1 = bs->total_sectors - sector_num;
1369
1370     qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1));
1371
1372     return n1;
1373 }
1374
1375 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
1376                           int remaining_sectors, QEMUIOVector *qiov)
1377 {
1378     BDRVQcow2State *s = bs->opaque;
1379     int index_in_cluster, n1;
1380     int ret;
1381     int cur_nr_sectors; /* number of sectors in current iteration */
1382     uint64_t cluster_offset = 0;
1383     uint64_t bytes_done = 0;
1384     QEMUIOVector hd_qiov;
1385     uint8_t *cluster_data = NULL;
1386
1387     qemu_iovec_init(&hd_qiov, qiov->niov);
1388
1389     qemu_co_mutex_lock(&s->lock);
1390
1391     while (remaining_sectors != 0) {
1392
1393         /* prepare next request */
1394         cur_nr_sectors = remaining_sectors;
1395         if (s->cipher) {
1396             cur_nr_sectors = MIN(cur_nr_sectors,
1397                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1398         }
1399
1400         ret = qcow2_get_cluster_offset(bs, sector_num << 9,
1401             &cur_nr_sectors, &cluster_offset);
1402         if (ret < 0) {
1403             goto fail;
1404         }
1405
1406         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1407
1408         qemu_iovec_reset(&hd_qiov);
1409         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1410             cur_nr_sectors * 512);
1411
1412         switch (ret) {
1413         case QCOW2_CLUSTER_UNALLOCATED:
1414
1415             if (bs->backing) {
1416                 /* read from the base image */
1417                 n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov,
1418                     sector_num, cur_nr_sectors);
1419                 if (n1 > 0) {
1420                     QEMUIOVector local_qiov;
1421
1422                     qemu_iovec_init(&local_qiov, hd_qiov.niov);
1423                     qemu_iovec_concat(&local_qiov, &hd_qiov, 0,
1424                                       n1 * BDRV_SECTOR_SIZE);
1425
1426                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1427                     qemu_co_mutex_unlock(&s->lock);
1428                     ret = bdrv_co_readv(bs->backing->bs, sector_num,
1429                                         n1, &local_qiov);
1430                     qemu_co_mutex_lock(&s->lock);
1431
1432                     qemu_iovec_destroy(&local_qiov);
1433
1434                     if (ret < 0) {
1435                         goto fail;
1436                     }
1437                 }
1438             } else {
1439                 /* Note: in this case, no need to wait */
1440                 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1441             }
1442             break;
1443
1444         case QCOW2_CLUSTER_ZERO:
1445             qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1446             break;
1447
1448         case QCOW2_CLUSTER_COMPRESSED:
1449             /* add AIO support for compressed blocks ? */
1450             ret = qcow2_decompress_cluster(bs, cluster_offset);
1451             if (ret < 0) {
1452                 goto fail;
1453             }
1454
1455             qemu_iovec_from_buf(&hd_qiov, 0,
1456                 s->cluster_cache + index_in_cluster * 512,
1457                 512 * cur_nr_sectors);
1458             break;
1459
1460         case QCOW2_CLUSTER_NORMAL:
1461             if ((cluster_offset & 511) != 0) {
1462                 ret = -EIO;
1463                 goto fail;
1464             }
1465
1466             if (bs->encrypted) {
1467                 assert(s->cipher);
1468
1469                 /*
1470                  * For encrypted images, read everything into a temporary
1471                  * contiguous buffer on which the AES functions can work.
1472                  */
1473                 if (!cluster_data) {
1474                     cluster_data =
1475                         qemu_try_blockalign(bs->file->bs,
1476                                             QCOW_MAX_CRYPT_CLUSTERS
1477                                             * s->cluster_size);
1478                     if (cluster_data == NULL) {
1479                         ret = -ENOMEM;
1480                         goto fail;
1481                     }
1482                 }
1483
1484                 assert(cur_nr_sectors <=
1485                     QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1486                 qemu_iovec_reset(&hd_qiov);
1487                 qemu_iovec_add(&hd_qiov, cluster_data,
1488                     512 * cur_nr_sectors);
1489             }
1490
1491             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1492             qemu_co_mutex_unlock(&s->lock);
1493             ret = bdrv_co_readv(bs->file->bs,
1494                                 (cluster_offset >> 9) + index_in_cluster,
1495                                 cur_nr_sectors, &hd_qiov);
1496             qemu_co_mutex_lock(&s->lock);
1497             if (ret < 0) {
1498                 goto fail;
1499             }
1500             if (bs->encrypted) {
1501                 assert(s->cipher);
1502                 Error *err = NULL;
1503                 if (qcow2_encrypt_sectors(s, sector_num,  cluster_data,
1504                                           cluster_data, cur_nr_sectors, false,
1505                                           &err) < 0) {
1506                     error_free(err);
1507                     ret = -EIO;
1508                     goto fail;
1509                 }
1510                 qemu_iovec_from_buf(qiov, bytes_done,
1511                     cluster_data, 512 * cur_nr_sectors);
1512             }
1513             break;
1514
1515         default:
1516             g_assert_not_reached();
1517             ret = -EIO;
1518             goto fail;
1519         }
1520
1521         remaining_sectors -= cur_nr_sectors;
1522         sector_num += cur_nr_sectors;
1523         bytes_done += cur_nr_sectors * 512;
1524     }
1525     ret = 0;
1526
1527 fail:
1528     qemu_co_mutex_unlock(&s->lock);
1529
1530     qemu_iovec_destroy(&hd_qiov);
1531     qemu_vfree(cluster_data);
1532
1533     return ret;
1534 }
1535
1536 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
1537                            int64_t sector_num,
1538                            int remaining_sectors,
1539                            QEMUIOVector *qiov)
1540 {
1541     BDRVQcow2State *s = bs->opaque;
1542     int index_in_cluster;
1543     int ret;
1544     int cur_nr_sectors; /* number of sectors in current iteration */
1545     uint64_t cluster_offset;
1546     QEMUIOVector hd_qiov;
1547     uint64_t bytes_done = 0;
1548     uint8_t *cluster_data = NULL;
1549     QCowL2Meta *l2meta = NULL;
1550
1551     trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
1552                                  remaining_sectors);
1553
1554     qemu_iovec_init(&hd_qiov, qiov->niov);
1555
1556     s->cluster_cache_offset = -1; /* disable compressed cache */
1557
1558     qemu_co_mutex_lock(&s->lock);
1559
1560     while (remaining_sectors != 0) {
1561
1562         l2meta = NULL;
1563
1564         trace_qcow2_writev_start_part(qemu_coroutine_self());
1565         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1566         cur_nr_sectors = remaining_sectors;
1567         if (bs->encrypted &&
1568             cur_nr_sectors >
1569             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster) {
1570             cur_nr_sectors =
1571                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster;
1572         }
1573
1574         ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
1575             &cur_nr_sectors, &cluster_offset, &l2meta);
1576         if (ret < 0) {
1577             goto fail;
1578         }
1579
1580         assert((cluster_offset & 511) == 0);
1581
1582         qemu_iovec_reset(&hd_qiov);
1583         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1584             cur_nr_sectors * 512);
1585
1586         if (bs->encrypted) {
1587             Error *err = NULL;
1588             assert(s->cipher);
1589             if (!cluster_data) {
1590                 cluster_data = qemu_try_blockalign(bs->file->bs,
1591                                                    QCOW_MAX_CRYPT_CLUSTERS
1592                                                    * s->cluster_size);
1593                 if (cluster_data == NULL) {
1594                     ret = -ENOMEM;
1595                     goto fail;
1596                 }
1597             }
1598
1599             assert(hd_qiov.size <=
1600                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1601             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1602
1603             if (qcow2_encrypt_sectors(s, sector_num, cluster_data,
1604                                       cluster_data, cur_nr_sectors,
1605                                       true, &err) < 0) {
1606                 error_free(err);
1607                 ret = -EIO;
1608                 goto fail;
1609             }
1610
1611             qemu_iovec_reset(&hd_qiov);
1612             qemu_iovec_add(&hd_qiov, cluster_data,
1613                 cur_nr_sectors * 512);
1614         }
1615
1616         ret = qcow2_pre_write_overlap_check(bs, 0,
1617                 cluster_offset + index_in_cluster * BDRV_SECTOR_SIZE,
1618                 cur_nr_sectors * BDRV_SECTOR_SIZE);
1619         if (ret < 0) {
1620             goto fail;
1621         }
1622
1623         qemu_co_mutex_unlock(&s->lock);
1624         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1625         trace_qcow2_writev_data(qemu_coroutine_self(),
1626                                 (cluster_offset >> 9) + index_in_cluster);
1627         ret = bdrv_co_writev(bs->file->bs,
1628                              (cluster_offset >> 9) + index_in_cluster,
1629                              cur_nr_sectors, &hd_qiov);
1630         qemu_co_mutex_lock(&s->lock);
1631         if (ret < 0) {
1632             goto fail;
1633         }
1634
1635         while (l2meta != NULL) {
1636             QCowL2Meta *next;
1637
1638             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1639             if (ret < 0) {
1640                 goto fail;
1641             }
1642
1643             /* Take the request off the list of running requests */
1644             if (l2meta->nb_clusters != 0) {
1645                 QLIST_REMOVE(l2meta, next_in_flight);
1646             }
1647
1648             qemu_co_queue_restart_all(&l2meta->dependent_requests);
1649
1650             next = l2meta->next;
1651             g_free(l2meta);
1652             l2meta = next;
1653         }
1654
1655         remaining_sectors -= cur_nr_sectors;
1656         sector_num += cur_nr_sectors;
1657         bytes_done += cur_nr_sectors * 512;
1658         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
1659     }
1660     ret = 0;
1661
1662 fail:
1663     qemu_co_mutex_unlock(&s->lock);
1664
1665     while (l2meta != NULL) {
1666         QCowL2Meta *next;
1667
1668         if (l2meta->nb_clusters != 0) {
1669             QLIST_REMOVE(l2meta, next_in_flight);
1670         }
1671         qemu_co_queue_restart_all(&l2meta->dependent_requests);
1672
1673         next = l2meta->next;
1674         g_free(l2meta);
1675         l2meta = next;
1676     }
1677
1678     qemu_iovec_destroy(&hd_qiov);
1679     qemu_vfree(cluster_data);
1680     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
1681
1682     return ret;
1683 }
1684
1685 static int qcow2_inactivate(BlockDriverState *bs)
1686 {
1687     BDRVQcow2State *s = bs->opaque;
1688     int ret, result = 0;
1689
1690     ret = qcow2_cache_flush(bs, s->l2_table_cache);
1691     if (ret) {
1692         result = ret;
1693         error_report("Failed to flush the L2 table cache: %s",
1694                      strerror(-ret));
1695     }
1696
1697     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1698     if (ret) {
1699         result = ret;
1700         error_report("Failed to flush the refcount block cache: %s",
1701                      strerror(-ret));
1702     }
1703
1704     if (result == 0) {
1705         qcow2_mark_clean(bs);
1706     }
1707
1708     return result;
1709 }
1710
1711 static void qcow2_close(BlockDriverState *bs)
1712 {
1713     BDRVQcow2State *s = bs->opaque;
1714     qemu_vfree(s->l1_table);
1715     /* else pre-write overlap checks in cache_destroy may crash */
1716     s->l1_table = NULL;
1717
1718     if (!(s->flags & BDRV_O_INACTIVE)) {
1719         qcow2_inactivate(bs);
1720     }
1721
1722     cache_clean_timer_del(bs);
1723     qcow2_cache_destroy(bs, s->l2_table_cache);
1724     qcow2_cache_destroy(bs, s->refcount_block_cache);
1725
1726     qcrypto_cipher_free(s->cipher);
1727     s->cipher = NULL;
1728
1729     g_free(s->unknown_header_fields);
1730     cleanup_unknown_header_ext(bs);
1731
1732     g_free(s->image_backing_file);
1733     g_free(s->image_backing_format);
1734
1735     g_free(s->cluster_cache);
1736     qemu_vfree(s->cluster_data);
1737     qcow2_refcount_close(bs);
1738     qcow2_free_snapshots(bs);
1739 }
1740
1741 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
1742 {
1743     BDRVQcow2State *s = bs->opaque;
1744     int flags = s->flags;
1745     QCryptoCipher *cipher = NULL;
1746     QDict *options;
1747     Error *local_err = NULL;
1748     int ret;
1749
1750     /*
1751      * Backing files are read-only which makes all of their metadata immutable,
1752      * that means we don't have to worry about reopening them here.
1753      */
1754
1755     cipher = s->cipher;
1756     s->cipher = NULL;
1757
1758     qcow2_close(bs);
1759
1760     bdrv_invalidate_cache(bs->file->bs, &local_err);
1761     if (local_err) {
1762         error_propagate(errp, local_err);
1763         bs->drv = NULL;
1764         return;
1765     }
1766
1767     memset(s, 0, sizeof(BDRVQcow2State));
1768     options = qdict_clone_shallow(bs->options);
1769
1770     flags &= ~BDRV_O_INACTIVE;
1771     ret = qcow2_open(bs, options, flags, &local_err);
1772     QDECREF(options);
1773     if (local_err) {
1774         error_propagate(errp, local_err);
1775         error_prepend(errp, "Could not reopen qcow2 layer: ");
1776         bs->drv = NULL;
1777         return;
1778     } else if (ret < 0) {
1779         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
1780         bs->drv = NULL;
1781         return;
1782     }
1783
1784     s->cipher = cipher;
1785 }
1786
1787 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
1788     size_t len, size_t buflen)
1789 {
1790     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
1791     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
1792
1793     if (buflen < ext_len) {
1794         return -ENOSPC;
1795     }
1796
1797     *ext_backing_fmt = (QCowExtension) {
1798         .magic  = cpu_to_be32(magic),
1799         .len    = cpu_to_be32(len),
1800     };
1801     memcpy(buf + sizeof(QCowExtension), s, len);
1802
1803     return ext_len;
1804 }
1805
1806 /*
1807  * Updates the qcow2 header, including the variable length parts of it, i.e.
1808  * the backing file name and all extensions. qcow2 was not designed to allow
1809  * such changes, so if we run out of space (we can only use the first cluster)
1810  * this function may fail.
1811  *
1812  * Returns 0 on success, -errno in error cases.
1813  */
1814 int qcow2_update_header(BlockDriverState *bs)
1815 {
1816     BDRVQcow2State *s = bs->opaque;
1817     QCowHeader *header;
1818     char *buf;
1819     size_t buflen = s->cluster_size;
1820     int ret;
1821     uint64_t total_size;
1822     uint32_t refcount_table_clusters;
1823     size_t header_length;
1824     Qcow2UnknownHeaderExtension *uext;
1825
1826     buf = qemu_blockalign(bs, buflen);
1827
1828     /* Header structure */
1829     header = (QCowHeader*) buf;
1830
1831     if (buflen < sizeof(*header)) {
1832         ret = -ENOSPC;
1833         goto fail;
1834     }
1835
1836     header_length = sizeof(*header) + s->unknown_header_fields_size;
1837     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1838     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1839
1840     *header = (QCowHeader) {
1841         /* Version 2 fields */
1842         .magic                  = cpu_to_be32(QCOW_MAGIC),
1843         .version                = cpu_to_be32(s->qcow_version),
1844         .backing_file_offset    = 0,
1845         .backing_file_size      = 0,
1846         .cluster_bits           = cpu_to_be32(s->cluster_bits),
1847         .size                   = cpu_to_be64(total_size),
1848         .crypt_method           = cpu_to_be32(s->crypt_method_header),
1849         .l1_size                = cpu_to_be32(s->l1_size),
1850         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
1851         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
1852         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1853         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
1854         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
1855
1856         /* Version 3 fields */
1857         .incompatible_features  = cpu_to_be64(s->incompatible_features),
1858         .compatible_features    = cpu_to_be64(s->compatible_features),
1859         .autoclear_features     = cpu_to_be64(s->autoclear_features),
1860         .refcount_order         = cpu_to_be32(s->refcount_order),
1861         .header_length          = cpu_to_be32(header_length),
1862     };
1863
1864     /* For older versions, write a shorter header */
1865     switch (s->qcow_version) {
1866     case 2:
1867         ret = offsetof(QCowHeader, incompatible_features);
1868         break;
1869     case 3:
1870         ret = sizeof(*header);
1871         break;
1872     default:
1873         ret = -EINVAL;
1874         goto fail;
1875     }
1876
1877     buf += ret;
1878     buflen -= ret;
1879     memset(buf, 0, buflen);
1880
1881     /* Preserve any unknown field in the header */
1882     if (s->unknown_header_fields_size) {
1883         if (buflen < s->unknown_header_fields_size) {
1884             ret = -ENOSPC;
1885             goto fail;
1886         }
1887
1888         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1889         buf += s->unknown_header_fields_size;
1890         buflen -= s->unknown_header_fields_size;
1891     }
1892
1893     /* Backing file format header extension */
1894     if (s->image_backing_format) {
1895         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1896                              s->image_backing_format,
1897                              strlen(s->image_backing_format),
1898                              buflen);
1899         if (ret < 0) {
1900             goto fail;
1901         }
1902
1903         buf += ret;
1904         buflen -= ret;
1905     }
1906
1907     /* Feature table */
1908     if (s->qcow_version >= 3) {
1909         Qcow2Feature features[] = {
1910             {
1911                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1912                 .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
1913                 .name = "dirty bit",
1914             },
1915             {
1916                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1917                 .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
1918                 .name = "corrupt bit",
1919             },
1920             {
1921                 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1922                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1923                 .name = "lazy refcounts",
1924             },
1925         };
1926
1927         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1928                              features, sizeof(features), buflen);
1929         if (ret < 0) {
1930             goto fail;
1931         }
1932         buf += ret;
1933         buflen -= ret;
1934     }
1935
1936     /* Keep unknown header extensions */
1937     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1938         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1939         if (ret < 0) {
1940             goto fail;
1941         }
1942
1943         buf += ret;
1944         buflen -= ret;
1945     }
1946
1947     /* End of header extensions */
1948     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1949     if (ret < 0) {
1950         goto fail;
1951     }
1952
1953     buf += ret;
1954     buflen -= ret;
1955
1956     /* Backing file name */
1957     if (s->image_backing_file) {
1958         size_t backing_file_len = strlen(s->image_backing_file);
1959
1960         if (buflen < backing_file_len) {
1961             ret = -ENOSPC;
1962             goto fail;
1963         }
1964
1965         /* Using strncpy is ok here, since buf is not NUL-terminated. */
1966         strncpy(buf, s->image_backing_file, buflen);
1967
1968         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1969         header->backing_file_size   = cpu_to_be32(backing_file_len);
1970     }
1971
1972     /* Write the new header */
1973     ret = bdrv_pwrite(bs->file->bs, 0, header, s->cluster_size);
1974     if (ret < 0) {
1975         goto fail;
1976     }
1977
1978     ret = 0;
1979 fail:
1980     qemu_vfree(header);
1981     return ret;
1982 }
1983
1984 static int qcow2_change_backing_file(BlockDriverState *bs,
1985     const char *backing_file, const char *backing_fmt)
1986 {
1987     BDRVQcow2State *s = bs->opaque;
1988
1989     if (backing_file && strlen(backing_file) > 1023) {
1990         return -EINVAL;
1991     }
1992
1993     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1994     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1995
1996     g_free(s->image_backing_file);
1997     g_free(s->image_backing_format);
1998
1999     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
2000     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
2001
2002     return qcow2_update_header(bs);
2003 }
2004
2005 static int preallocate(BlockDriverState *bs)
2006 {
2007     uint64_t nb_sectors;
2008     uint64_t offset;
2009     uint64_t host_offset = 0;
2010     int num;
2011     int ret;
2012     QCowL2Meta *meta;
2013
2014     nb_sectors = bdrv_nb_sectors(bs);
2015     offset = 0;
2016
2017     while (nb_sectors) {
2018         num = MIN(nb_sectors, INT_MAX >> BDRV_SECTOR_BITS);
2019         ret = qcow2_alloc_cluster_offset(bs, offset, &num,
2020                                          &host_offset, &meta);
2021         if (ret < 0) {
2022             return ret;
2023         }
2024
2025         while (meta) {
2026             QCowL2Meta *next = meta->next;
2027
2028             ret = qcow2_alloc_cluster_link_l2(bs, meta);
2029             if (ret < 0) {
2030                 qcow2_free_any_clusters(bs, meta->alloc_offset,
2031                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
2032                 return ret;
2033             }
2034
2035             /* There are no dependent requests, but we need to remove our
2036              * request from the list of in-flight requests */
2037             QLIST_REMOVE(meta, next_in_flight);
2038
2039             g_free(meta);
2040             meta = next;
2041         }
2042
2043         /* TODO Preallocate data if requested */
2044
2045         nb_sectors -= num;
2046         offset += num << BDRV_SECTOR_BITS;
2047     }
2048
2049     /*
2050      * It is expected that the image file is large enough to actually contain
2051      * all of the allocated clusters (otherwise we get failing reads after
2052      * EOF). Extend the image to the last allocated sector.
2053      */
2054     if (host_offset != 0) {
2055         uint8_t buf[BDRV_SECTOR_SIZE];
2056         memset(buf, 0, BDRV_SECTOR_SIZE);
2057         ret = bdrv_write(bs->file->bs,
2058                          (host_offset >> BDRV_SECTOR_BITS) + num - 1,
2059                          buf, 1);
2060         if (ret < 0) {
2061             return ret;
2062         }
2063     }
2064
2065     return 0;
2066 }
2067
2068 static int qcow2_create2(const char *filename, int64_t total_size,
2069                          const char *backing_file, const char *backing_format,
2070                          int flags, size_t cluster_size, PreallocMode prealloc,
2071                          QemuOpts *opts, int version, int refcount_order,
2072                          Error **errp)
2073 {
2074     int cluster_bits;
2075     QDict *options;
2076
2077     /* Calculate cluster_bits */
2078     cluster_bits = ctz32(cluster_size);
2079     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
2080         (1 << cluster_bits) != cluster_size)
2081     {
2082         error_setg(errp, "Cluster size must be a power of two between %d and "
2083                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
2084         return -EINVAL;
2085     }
2086
2087     /*
2088      * Open the image file and write a minimal qcow2 header.
2089      *
2090      * We keep things simple and start with a zero-sized image. We also
2091      * do without refcount blocks or a L1 table for now. We'll fix the
2092      * inconsistency later.
2093      *
2094      * We do need a refcount table because growing the refcount table means
2095      * allocating two new refcount blocks - the seconds of which would be at
2096      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2097      * size for any qcow2 image.
2098      */
2099     BlockBackend *blk;
2100     QCowHeader *header;
2101     uint64_t* refcount_table;
2102     Error *local_err = NULL;
2103     int ret;
2104
2105     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
2106         /* Note: The following calculation does not need to be exact; if it is a
2107          * bit off, either some bytes will be "leaked" (which is fine) or we
2108          * will need to increase the file size by some bytes (which is fine,
2109          * too, as long as the bulk is allocated here). Therefore, using
2110          * floating point arithmetic is fine. */
2111         int64_t meta_size = 0;
2112         uint64_t nreftablee, nrefblocke, nl1e, nl2e;
2113         int64_t aligned_total_size = align_offset(total_size, cluster_size);
2114         int refblock_bits, refblock_size;
2115         /* refcount entry size in bytes */
2116         double rces = (1 << refcount_order) / 8.;
2117
2118         /* see qcow2_open() */
2119         refblock_bits = cluster_bits - (refcount_order - 3);
2120         refblock_size = 1 << refblock_bits;
2121
2122         /* header: 1 cluster */
2123         meta_size += cluster_size;
2124
2125         /* total size of L2 tables */
2126         nl2e = aligned_total_size / cluster_size;
2127         nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
2128         meta_size += nl2e * sizeof(uint64_t);
2129
2130         /* total size of L1 tables */
2131         nl1e = nl2e * sizeof(uint64_t) / cluster_size;
2132         nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
2133         meta_size += nl1e * sizeof(uint64_t);
2134
2135         /* total size of refcount blocks
2136          *
2137          * note: every host cluster is reference-counted, including metadata
2138          * (even refcount blocks are recursively included).
2139          * Let:
2140          *   a = total_size (this is the guest disk size)
2141          *   m = meta size not including refcount blocks and refcount tables
2142          *   c = cluster size
2143          *   y1 = number of refcount blocks entries
2144          *   y2 = meta size including everything
2145          *   rces = refcount entry size in bytes
2146          * then,
2147          *   y1 = (y2 + a)/c
2148          *   y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m
2149          * we can get y1:
2150          *   y1 = (a + m) / (c - rces - rces * sizeof(u64) / c)
2151          */
2152         nrefblocke = (aligned_total_size + meta_size + cluster_size)
2153                    / (cluster_size - rces - rces * sizeof(uint64_t)
2154                                                  / cluster_size);
2155         meta_size += DIV_ROUND_UP(nrefblocke, refblock_size) * cluster_size;
2156
2157         /* total size of refcount tables */
2158         nreftablee = nrefblocke / refblock_size;
2159         nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t));
2160         meta_size += nreftablee * sizeof(uint64_t);
2161
2162         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
2163                             aligned_total_size + meta_size, &error_abort);
2164         qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc],
2165                      &error_abort);
2166     }
2167
2168     ret = bdrv_create_file(filename, opts, &local_err);
2169     if (ret < 0) {
2170         error_propagate(errp, local_err);
2171         return ret;
2172     }
2173
2174     blk = blk_new_open(filename, NULL, NULL,
2175                        BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
2176     if (blk == NULL) {
2177         error_propagate(errp, local_err);
2178         return -EIO;
2179     }
2180
2181     blk_set_allow_write_beyond_eof(blk, true);
2182
2183     /* Write the header */
2184     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
2185     header = g_malloc0(cluster_size);
2186     *header = (QCowHeader) {
2187         .magic                      = cpu_to_be32(QCOW_MAGIC),
2188         .version                    = cpu_to_be32(version),
2189         .cluster_bits               = cpu_to_be32(cluster_bits),
2190         .size                       = cpu_to_be64(0),
2191         .l1_table_offset            = cpu_to_be64(0),
2192         .l1_size                    = cpu_to_be32(0),
2193         .refcount_table_offset      = cpu_to_be64(cluster_size),
2194         .refcount_table_clusters    = cpu_to_be32(1),
2195         .refcount_order             = cpu_to_be32(refcount_order),
2196         .header_length              = cpu_to_be32(sizeof(*header)),
2197     };
2198
2199     if (flags & BLOCK_FLAG_ENCRYPT) {
2200         header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
2201     } else {
2202         header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
2203     }
2204
2205     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
2206         header->compatible_features |=
2207             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
2208     }
2209
2210     ret = blk_pwrite(blk, 0, header, cluster_size);
2211     g_free(header);
2212     if (ret < 0) {
2213         error_setg_errno(errp, -ret, "Could not write qcow2 header");
2214         goto out;
2215     }
2216
2217     /* Write a refcount table with one refcount block */
2218     refcount_table = g_malloc0(2 * cluster_size);
2219     refcount_table[0] = cpu_to_be64(2 * cluster_size);
2220     ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size);
2221     g_free(refcount_table);
2222
2223     if (ret < 0) {
2224         error_setg_errno(errp, -ret, "Could not write refcount table");
2225         goto out;
2226     }
2227
2228     blk_unref(blk);
2229     blk = NULL;
2230
2231     /*
2232      * And now open the image and make it consistent first (i.e. increase the
2233      * refcount of the cluster that is occupied by the header and the refcount
2234      * table)
2235      */
2236     options = qdict_new();
2237     qdict_put(options, "driver", qstring_from_str("qcow2"));
2238     blk = blk_new_open(filename, NULL, options,
2239                        BDRV_O_RDWR | BDRV_O_NO_FLUSH, &local_err);
2240     if (blk == NULL) {
2241         error_propagate(errp, local_err);
2242         ret = -EIO;
2243         goto out;
2244     }
2245
2246     ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
2247     if (ret < 0) {
2248         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
2249                          "header and refcount table");
2250         goto out;
2251
2252     } else if (ret != 0) {
2253         error_report("Huh, first cluster in empty image is already in use?");
2254         abort();
2255     }
2256
2257     /* Create a full header (including things like feature table) */
2258     ret = qcow2_update_header(blk_bs(blk));
2259     if (ret < 0) {
2260         error_setg_errno(errp, -ret, "Could not update qcow2 header");
2261         goto out;
2262     }
2263
2264     /* Okay, now that we have a valid image, let's give it the right size */
2265     ret = blk_truncate(blk, total_size);
2266     if (ret < 0) {
2267         error_setg_errno(errp, -ret, "Could not resize image");
2268         goto out;
2269     }
2270
2271     /* Want a backing file? There you go.*/
2272     if (backing_file) {
2273         ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format);
2274         if (ret < 0) {
2275             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
2276                              "with format '%s'", backing_file, backing_format);
2277             goto out;
2278         }
2279     }
2280
2281     /* And if we're supposed to preallocate metadata, do that now */
2282     if (prealloc != PREALLOC_MODE_OFF) {
2283         BDRVQcow2State *s = blk_bs(blk)->opaque;
2284         qemu_co_mutex_lock(&s->lock);
2285         ret = preallocate(blk_bs(blk));
2286         qemu_co_mutex_unlock(&s->lock);
2287         if (ret < 0) {
2288             error_setg_errno(errp, -ret, "Could not preallocate metadata");
2289             goto out;
2290         }
2291     }
2292
2293     blk_unref(blk);
2294     blk = NULL;
2295
2296     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
2297     options = qdict_new();
2298     qdict_put(options, "driver", qstring_from_str("qcow2"));
2299     blk = blk_new_open(filename, NULL, options,
2300                        BDRV_O_RDWR | BDRV_O_NO_BACKING, &local_err);
2301     if (blk == NULL) {
2302         error_propagate(errp, local_err);
2303         ret = -EIO;
2304         goto out;
2305     }
2306
2307     ret = 0;
2308 out:
2309     if (blk) {
2310         blk_unref(blk);
2311     }
2312     return ret;
2313 }
2314
2315 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
2316 {
2317     char *backing_file = NULL;
2318     char *backing_fmt = NULL;
2319     char *buf = NULL;
2320     uint64_t size = 0;
2321     int flags = 0;
2322     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
2323     PreallocMode prealloc;
2324     int version = 3;
2325     uint64_t refcount_bits = 16;
2326     int refcount_order;
2327     Error *local_err = NULL;
2328     int ret;
2329
2330     /* Read out options */
2331     size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2332                     BDRV_SECTOR_SIZE);
2333     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2334     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2335     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2336         flags |= BLOCK_FLAG_ENCRYPT;
2337     }
2338     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2339                                          DEFAULT_CLUSTER_SIZE);
2340     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2341     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
2342                                PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
2343                                &local_err);
2344     if (local_err) {
2345         error_propagate(errp, local_err);
2346         ret = -EINVAL;
2347         goto finish;
2348     }
2349     g_free(buf);
2350     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2351     if (!buf) {
2352         /* keep the default */
2353     } else if (!strcmp(buf, "0.10")) {
2354         version = 2;
2355     } else if (!strcmp(buf, "1.1")) {
2356         version = 3;
2357     } else {
2358         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2359         ret = -EINVAL;
2360         goto finish;
2361     }
2362
2363     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2364         flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2365     }
2366
2367     if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2368         error_setg(errp, "Backing file and preallocation cannot be used at "
2369                    "the same time");
2370         ret = -EINVAL;
2371         goto finish;
2372     }
2373
2374     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2375         error_setg(errp, "Lazy refcounts only supported with compatibility "
2376                    "level 1.1 and above (use compat=1.1 or greater)");
2377         ret = -EINVAL;
2378         goto finish;
2379     }
2380
2381     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS,
2382                                             refcount_bits);
2383     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2384         error_setg(errp, "Refcount width must be a power of two and may not "
2385                    "exceed 64 bits");
2386         ret = -EINVAL;
2387         goto finish;
2388     }
2389
2390     if (version < 3 && refcount_bits != 16) {
2391         error_setg(errp, "Different refcount widths than 16 bits require "
2392                    "compatibility level 1.1 or above (use compat=1.1 or "
2393                    "greater)");
2394         ret = -EINVAL;
2395         goto finish;
2396     }
2397
2398     refcount_order = ctz32(refcount_bits);
2399
2400     ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2401                         cluster_size, prealloc, opts, version, refcount_order,
2402                         &local_err);
2403     if (local_err) {
2404         error_propagate(errp, local_err);
2405     }
2406
2407 finish:
2408     g_free(backing_file);
2409     g_free(backing_fmt);
2410     g_free(buf);
2411     return ret;
2412 }
2413
2414 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
2415     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2416 {
2417     int ret;
2418     BDRVQcow2State *s = bs->opaque;
2419
2420     /* Emulate misaligned zero writes */
2421     if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
2422         return -ENOTSUP;
2423     }
2424
2425     /* Whatever is left can use real zero clusters */
2426     qemu_co_mutex_lock(&s->lock);
2427     ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2428         nb_sectors);
2429     qemu_co_mutex_unlock(&s->lock);
2430
2431     return ret;
2432 }
2433
2434 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
2435     int64_t sector_num, int nb_sectors)
2436 {
2437     int ret;
2438     BDRVQcow2State *s = bs->opaque;
2439
2440     qemu_co_mutex_lock(&s->lock);
2441     ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2442         nb_sectors, QCOW2_DISCARD_REQUEST, false);
2443     qemu_co_mutex_unlock(&s->lock);
2444     return ret;
2445 }
2446
2447 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
2448 {
2449     BDRVQcow2State *s = bs->opaque;
2450     int64_t new_l1_size;
2451     int ret;
2452
2453     if (offset & 511) {
2454         error_report("The new size must be a multiple of 512");
2455         return -EINVAL;
2456     }
2457
2458     /* cannot proceed if image has snapshots */
2459     if (s->nb_snapshots) {
2460         error_report("Can't resize an image which has snapshots");
2461         return -ENOTSUP;
2462     }
2463
2464     /* shrinking is currently not supported */
2465     if (offset < bs->total_sectors * 512) {
2466         error_report("qcow2 doesn't support shrinking images yet");
2467         return -ENOTSUP;
2468     }
2469
2470     new_l1_size = size_to_l1(s, offset);
2471     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
2472     if (ret < 0) {
2473         return ret;
2474     }
2475
2476     /* write updated header.size */
2477     offset = cpu_to_be64(offset);
2478     ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, size),
2479                            &offset, sizeof(uint64_t));
2480     if (ret < 0) {
2481         return ret;
2482     }
2483
2484     s->l1_vm_state_index = new_l1_size;
2485     return 0;
2486 }
2487
2488 /* XXX: put compressed sectors first, then all the cluster aligned
2489    tables to avoid losing bytes in alignment */
2490 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
2491                                   const uint8_t *buf, int nb_sectors)
2492 {
2493     BDRVQcow2State *s = bs->opaque;
2494     z_stream strm;
2495     int ret, out_len;
2496     uint8_t *out_buf;
2497     uint64_t cluster_offset;
2498
2499     if (nb_sectors == 0) {
2500         /* align end of file to a sector boundary to ease reading with
2501            sector based I/Os */
2502         cluster_offset = bdrv_getlength(bs->file->bs);
2503         return bdrv_truncate(bs->file->bs, cluster_offset);
2504     }
2505
2506     if (nb_sectors != s->cluster_sectors) {
2507         ret = -EINVAL;
2508
2509         /* Zero-pad last write if image size is not cluster aligned */
2510         if (sector_num + nb_sectors == bs->total_sectors &&
2511             nb_sectors < s->cluster_sectors) {
2512             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
2513             memset(pad_buf, 0, s->cluster_size);
2514             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
2515             ret = qcow2_write_compressed(bs, sector_num,
2516                                          pad_buf, s->cluster_sectors);
2517             qemu_vfree(pad_buf);
2518         }
2519         return ret;
2520     }
2521
2522     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
2523
2524     /* best compression, small window, no zlib header */
2525     memset(&strm, 0, sizeof(strm));
2526     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
2527                        Z_DEFLATED, -12,
2528                        9, Z_DEFAULT_STRATEGY);
2529     if (ret != 0) {
2530         ret = -EINVAL;
2531         goto fail;
2532     }
2533
2534     strm.avail_in = s->cluster_size;
2535     strm.next_in = (uint8_t *)buf;
2536     strm.avail_out = s->cluster_size;
2537     strm.next_out = out_buf;
2538
2539     ret = deflate(&strm, Z_FINISH);
2540     if (ret != Z_STREAM_END && ret != Z_OK) {
2541         deflateEnd(&strm);
2542         ret = -EINVAL;
2543         goto fail;
2544     }
2545     out_len = strm.next_out - out_buf;
2546
2547     deflateEnd(&strm);
2548
2549     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
2550         /* could not compress: write normal cluster */
2551         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
2552         if (ret < 0) {
2553             goto fail;
2554         }
2555     } else {
2556         cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
2557             sector_num << 9, out_len);
2558         if (!cluster_offset) {
2559             ret = -EIO;
2560             goto fail;
2561         }
2562         cluster_offset &= s->cluster_offset_mask;
2563
2564         ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
2565         if (ret < 0) {
2566             goto fail;
2567         }
2568
2569         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
2570         ret = bdrv_pwrite(bs->file->bs, cluster_offset, out_buf, out_len);
2571         if (ret < 0) {
2572             goto fail;
2573         }
2574     }
2575
2576     ret = 0;
2577 fail:
2578     g_free(out_buf);
2579     return ret;
2580 }
2581
2582 static int make_completely_empty(BlockDriverState *bs)
2583 {
2584     BDRVQcow2State *s = bs->opaque;
2585     int ret, l1_clusters;
2586     int64_t offset;
2587     uint64_t *new_reftable = NULL;
2588     uint64_t rt_entry, l1_size2;
2589     struct {
2590         uint64_t l1_offset;
2591         uint64_t reftable_offset;
2592         uint32_t reftable_clusters;
2593     } QEMU_PACKED l1_ofs_rt_ofs_cls;
2594
2595     ret = qcow2_cache_empty(bs, s->l2_table_cache);
2596     if (ret < 0) {
2597         goto fail;
2598     }
2599
2600     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
2601     if (ret < 0) {
2602         goto fail;
2603     }
2604
2605     /* Refcounts will be broken utterly */
2606     ret = qcow2_mark_dirty(bs);
2607     if (ret < 0) {
2608         goto fail;
2609     }
2610
2611     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2612
2613     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2614     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
2615
2616     /* After this call, neither the in-memory nor the on-disk refcount
2617      * information accurately describe the actual references */
2618
2619     ret = bdrv_write_zeroes(bs->file->bs, s->l1_table_offset / BDRV_SECTOR_SIZE,
2620                             l1_clusters * s->cluster_sectors, 0);
2621     if (ret < 0) {
2622         goto fail_broken_refcounts;
2623     }
2624     memset(s->l1_table, 0, l1_size2);
2625
2626     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
2627
2628     /* Overwrite enough clusters at the beginning of the sectors to place
2629      * the refcount table, a refcount block and the L1 table in; this may
2630      * overwrite parts of the existing refcount and L1 table, which is not
2631      * an issue because the dirty flag is set, complete data loss is in fact
2632      * desired and partial data loss is consequently fine as well */
2633     ret = bdrv_write_zeroes(bs->file->bs, s->cluster_size / BDRV_SECTOR_SIZE,
2634                             (2 + l1_clusters) * s->cluster_size /
2635                             BDRV_SECTOR_SIZE, 0);
2636     /* This call (even if it failed overall) may have overwritten on-disk
2637      * refcount structures; in that case, the in-memory refcount information
2638      * will probably differ from the on-disk information which makes the BDS
2639      * unusable */
2640     if (ret < 0) {
2641         goto fail_broken_refcounts;
2642     }
2643
2644     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2645     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
2646
2647     /* "Create" an empty reftable (one cluster) directly after the image
2648      * header and an empty L1 table three clusters after the image header;
2649      * the cluster between those two will be used as the first refblock */
2650     cpu_to_be64w(&l1_ofs_rt_ofs_cls.l1_offset, 3 * s->cluster_size);
2651     cpu_to_be64w(&l1_ofs_rt_ofs_cls.reftable_offset, s->cluster_size);
2652     cpu_to_be32w(&l1_ofs_rt_ofs_cls.reftable_clusters, 1);
2653     ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, l1_table_offset),
2654                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
2655     if (ret < 0) {
2656         goto fail_broken_refcounts;
2657     }
2658
2659     s->l1_table_offset = 3 * s->cluster_size;
2660
2661     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
2662     if (!new_reftable) {
2663         ret = -ENOMEM;
2664         goto fail_broken_refcounts;
2665     }
2666
2667     s->refcount_table_offset = s->cluster_size;
2668     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
2669
2670     g_free(s->refcount_table);
2671     s->refcount_table = new_reftable;
2672     new_reftable = NULL;
2673
2674     /* Now the in-memory refcount information again corresponds to the on-disk
2675      * information (reftable is empty and no refblocks (the refblock cache is
2676      * empty)); however, this means some clusters (e.g. the image header) are
2677      * referenced, but not refcounted, but the normal qcow2 code assumes that
2678      * the in-memory information is always correct */
2679
2680     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
2681
2682     /* Enter the first refblock into the reftable */
2683     rt_entry = cpu_to_be64(2 * s->cluster_size);
2684     ret = bdrv_pwrite_sync(bs->file->bs, s->cluster_size,
2685                            &rt_entry, sizeof(rt_entry));
2686     if (ret < 0) {
2687         goto fail_broken_refcounts;
2688     }
2689     s->refcount_table[0] = 2 * s->cluster_size;
2690
2691     s->free_cluster_index = 0;
2692     assert(3 + l1_clusters <= s->refcount_block_size);
2693     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
2694     if (offset < 0) {
2695         ret = offset;
2696         goto fail_broken_refcounts;
2697     } else if (offset > 0) {
2698         error_report("First cluster in emptied image is in use");
2699         abort();
2700     }
2701
2702     /* Now finally the in-memory information corresponds to the on-disk
2703      * structures and is correct */
2704     ret = qcow2_mark_clean(bs);
2705     if (ret < 0) {
2706         goto fail;
2707     }
2708
2709     ret = bdrv_truncate(bs->file->bs, (3 + l1_clusters) * s->cluster_size);
2710     if (ret < 0) {
2711         goto fail;
2712     }
2713
2714     return 0;
2715
2716 fail_broken_refcounts:
2717     /* The BDS is unusable at this point. If we wanted to make it usable, we
2718      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
2719      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
2720      * again. However, because the functions which could have caused this error
2721      * path to be taken are used by those functions as well, it's very likely
2722      * that that sequence will fail as well. Therefore, just eject the BDS. */
2723     bs->drv = NULL;
2724
2725 fail:
2726     g_free(new_reftable);
2727     return ret;
2728 }
2729
2730 static int qcow2_make_empty(BlockDriverState *bs)
2731 {
2732     BDRVQcow2State *s = bs->opaque;
2733     uint64_t start_sector;
2734     int sector_step = INT_MAX / BDRV_SECTOR_SIZE;
2735     int l1_clusters, ret = 0;
2736
2737     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2738
2739     if (s->qcow_version >= 3 && !s->snapshots &&
2740         3 + l1_clusters <= s->refcount_block_size) {
2741         /* The following function only works for qcow2 v3 images (it requires
2742          * the dirty flag) and only as long as there are no snapshots (because
2743          * it completely empties the image). Furthermore, the L1 table and three
2744          * additional clusters (image header, refcount table, one refcount
2745          * block) have to fit inside one refcount block. */
2746         return make_completely_empty(bs);
2747     }
2748
2749     /* This fallback code simply discards every active cluster; this is slow,
2750      * but works in all cases */
2751     for (start_sector = 0; start_sector < bs->total_sectors;
2752          start_sector += sector_step)
2753     {
2754         /* As this function is generally used after committing an external
2755          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
2756          * default action for this kind of discard is to pass the discard,
2757          * which will ideally result in an actually smaller image file, as
2758          * is probably desired. */
2759         ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE,
2760                                      MIN(sector_step,
2761                                          bs->total_sectors - start_sector),
2762                                      QCOW2_DISCARD_SNAPSHOT, true);
2763         if (ret < 0) {
2764             break;
2765         }
2766     }
2767
2768     return ret;
2769 }
2770
2771 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
2772 {
2773     BDRVQcow2State *s = bs->opaque;
2774     int ret;
2775
2776     qemu_co_mutex_lock(&s->lock);
2777     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2778     if (ret < 0) {
2779         qemu_co_mutex_unlock(&s->lock);
2780         return ret;
2781     }
2782
2783     if (qcow2_need_accurate_refcounts(s)) {
2784         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2785         if (ret < 0) {
2786             qemu_co_mutex_unlock(&s->lock);
2787             return ret;
2788         }
2789     }
2790     qemu_co_mutex_unlock(&s->lock);
2791
2792     return 0;
2793 }
2794
2795 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2796 {
2797     BDRVQcow2State *s = bs->opaque;
2798     bdi->unallocated_blocks_are_zero = true;
2799     bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
2800     bdi->cluster_size = s->cluster_size;
2801     bdi->vm_state_offset = qcow2_vm_state_offset(s);
2802     return 0;
2803 }
2804
2805 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
2806 {
2807     BDRVQcow2State *s = bs->opaque;
2808     ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
2809
2810     *spec_info = (ImageInfoSpecific){
2811         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
2812         .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1),
2813     };
2814     if (s->qcow_version == 2) {
2815         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
2816             .compat             = g_strdup("0.10"),
2817             .refcount_bits      = s->refcount_bits,
2818         };
2819     } else if (s->qcow_version == 3) {
2820         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
2821             .compat             = g_strdup("1.1"),
2822             .lazy_refcounts     = s->compatible_features &
2823                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
2824             .has_lazy_refcounts = true,
2825             .corrupt            = s->incompatible_features &
2826                                   QCOW2_INCOMPAT_CORRUPT,
2827             .has_corrupt        = true,
2828             .refcount_bits      = s->refcount_bits,
2829         };
2830     } else {
2831         /* if this assertion fails, this probably means a new version was
2832          * added without having it covered here */
2833         assert(false);
2834     }
2835
2836     return spec_info;
2837 }
2838
2839 #if 0
2840 static void dump_refcounts(BlockDriverState *bs)
2841 {
2842     BDRVQcow2State *s = bs->opaque;
2843     int64_t nb_clusters, k, k1, size;
2844     int refcount;
2845
2846     size = bdrv_getlength(bs->file->bs);
2847     nb_clusters = size_to_clusters(s, size);
2848     for(k = 0; k < nb_clusters;) {
2849         k1 = k;
2850         refcount = get_refcount(bs, k);
2851         k++;
2852         while (k < nb_clusters && get_refcount(bs, k) == refcount)
2853             k++;
2854         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
2855                k - k1);
2856     }
2857 }
2858 #endif
2859
2860 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2861                               int64_t pos)
2862 {
2863     BDRVQcow2State *s = bs->opaque;
2864     int64_t total_sectors = bs->total_sectors;
2865     bool zero_beyond_eof = bs->zero_beyond_eof;
2866     int ret;
2867
2868     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
2869     bs->zero_beyond_eof = false;
2870     ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov);
2871     bs->zero_beyond_eof = zero_beyond_eof;
2872
2873     /* bdrv_co_do_writev will have increased the total_sectors value to include
2874      * the VM state - the VM state is however not an actual part of the block
2875      * device, therefore, we need to restore the old value. */
2876     bs->total_sectors = total_sectors;
2877
2878     return ret;
2879 }
2880
2881 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2882                               int64_t pos, int size)
2883 {
2884     BDRVQcow2State *s = bs->opaque;
2885     bool zero_beyond_eof = bs->zero_beyond_eof;
2886     int ret;
2887
2888     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
2889     bs->zero_beyond_eof = false;
2890     ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
2891     bs->zero_beyond_eof = zero_beyond_eof;
2892
2893     return ret;
2894 }
2895
2896 /*
2897  * Downgrades an image's version. To achieve this, any incompatible features
2898  * have to be removed.
2899  */
2900 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
2901                            BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
2902 {
2903     BDRVQcow2State *s = bs->opaque;
2904     int current_version = s->qcow_version;
2905     int ret;
2906
2907     if (target_version == current_version) {
2908         return 0;
2909     } else if (target_version > current_version) {
2910         return -EINVAL;
2911     } else if (target_version != 2) {
2912         return -EINVAL;
2913     }
2914
2915     if (s->refcount_order != 4) {
2916         error_report("compat=0.10 requires refcount_bits=16");
2917         return -ENOTSUP;
2918     }
2919
2920     /* clear incompatible features */
2921     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
2922         ret = qcow2_mark_clean(bs);
2923         if (ret < 0) {
2924             return ret;
2925         }
2926     }
2927
2928     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2929      * the first place; if that happens nonetheless, returning -ENOTSUP is the
2930      * best thing to do anyway */
2931
2932     if (s->incompatible_features) {
2933         return -ENOTSUP;
2934     }
2935
2936     /* since we can ignore compatible features, we can set them to 0 as well */
2937     s->compatible_features = 0;
2938     /* if lazy refcounts have been used, they have already been fixed through
2939      * clearing the dirty flag */
2940
2941     /* clearing autoclear features is trivial */
2942     s->autoclear_features = 0;
2943
2944     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
2945     if (ret < 0) {
2946         return ret;
2947     }
2948
2949     s->qcow_version = target_version;
2950     ret = qcow2_update_header(bs);
2951     if (ret < 0) {
2952         s->qcow_version = current_version;
2953         return ret;
2954     }
2955     return 0;
2956 }
2957
2958 typedef enum Qcow2AmendOperation {
2959     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
2960      * statically initialized to so that the helper CB can discern the first
2961      * invocation from an operation change */
2962     QCOW2_NO_OPERATION = 0,
2963
2964     QCOW2_CHANGING_REFCOUNT_ORDER,
2965     QCOW2_DOWNGRADING,
2966 } Qcow2AmendOperation;
2967
2968 typedef struct Qcow2AmendHelperCBInfo {
2969     /* The code coordinating the amend operations should only modify
2970      * these four fields; the rest will be managed by the CB */
2971     BlockDriverAmendStatusCB *original_status_cb;
2972     void *original_cb_opaque;
2973
2974     Qcow2AmendOperation current_operation;
2975
2976     /* Total number of operations to perform (only set once) */
2977     int total_operations;
2978
2979     /* The following fields are managed by the CB */
2980
2981     /* Number of operations completed */
2982     int operations_completed;
2983
2984     /* Cumulative offset of all completed operations */
2985     int64_t offset_completed;
2986
2987     Qcow2AmendOperation last_operation;
2988     int64_t last_work_size;
2989 } Qcow2AmendHelperCBInfo;
2990
2991 static void qcow2_amend_helper_cb(BlockDriverState *bs,
2992                                   int64_t operation_offset,
2993                                   int64_t operation_work_size, void *opaque)
2994 {
2995     Qcow2AmendHelperCBInfo *info = opaque;
2996     int64_t current_work_size;
2997     int64_t projected_work_size;
2998
2999     if (info->current_operation != info->last_operation) {
3000         if (info->last_operation != QCOW2_NO_OPERATION) {
3001             info->offset_completed += info->last_work_size;
3002             info->operations_completed++;
3003         }
3004
3005         info->last_operation = info->current_operation;
3006     }
3007
3008     assert(info->total_operations > 0);
3009     assert(info->operations_completed < info->total_operations);
3010
3011     info->last_work_size = operation_work_size;
3012
3013     current_work_size = info->offset_completed + operation_work_size;
3014
3015     /* current_work_size is the total work size for (operations_completed + 1)
3016      * operations (which includes this one), so multiply it by the number of
3017      * operations not covered and divide it by the number of operations
3018      * covered to get a projection for the operations not covered */
3019     projected_work_size = current_work_size * (info->total_operations -
3020                                                info->operations_completed - 1)
3021                                             / (info->operations_completed + 1);
3022
3023     info->original_status_cb(bs, info->offset_completed + operation_offset,
3024                              current_work_size + projected_work_size,
3025                              info->original_cb_opaque);
3026 }
3027
3028 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
3029                                BlockDriverAmendStatusCB *status_cb,
3030                                void *cb_opaque)
3031 {
3032     BDRVQcow2State *s = bs->opaque;
3033     int old_version = s->qcow_version, new_version = old_version;
3034     uint64_t new_size = 0;
3035     const char *backing_file = NULL, *backing_format = NULL;
3036     bool lazy_refcounts = s->use_lazy_refcounts;
3037     const char *compat = NULL;
3038     uint64_t cluster_size = s->cluster_size;
3039     bool encrypt;
3040     int refcount_bits = s->refcount_bits;
3041     int ret;
3042     QemuOptDesc *desc = opts->list->desc;
3043     Qcow2AmendHelperCBInfo helper_cb_info;
3044
3045     while (desc && desc->name) {
3046         if (!qemu_opt_find(opts, desc->name)) {
3047             /* only change explicitly defined options */
3048             desc++;
3049             continue;
3050         }
3051
3052         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
3053             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
3054             if (!compat) {
3055                 /* preserve default */
3056             } else if (!strcmp(compat, "0.10")) {
3057                 new_version = 2;
3058             } else if (!strcmp(compat, "1.1")) {
3059                 new_version = 3;
3060             } else {
3061                 error_report("Unknown compatibility level %s", compat);
3062                 return -EINVAL;
3063             }
3064         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
3065             error_report("Cannot change preallocation mode");
3066             return -ENOTSUP;
3067         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
3068             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
3069         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
3070             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
3071         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
3072             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
3073         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
3074             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
3075                                         !!s->cipher);
3076
3077             if (encrypt != !!s->cipher) {
3078                 error_report("Changing the encryption flag is not supported");
3079                 return -ENOTSUP;
3080             }
3081         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
3082             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
3083                                              cluster_size);
3084             if (cluster_size != s->cluster_size) {
3085                 error_report("Changing the cluster size is not supported");
3086                 return -ENOTSUP;
3087             }
3088         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
3089             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
3090                                                lazy_refcounts);
3091         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
3092             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
3093                                                 refcount_bits);
3094
3095             if (refcount_bits <= 0 || refcount_bits > 64 ||
3096                 !is_power_of_2(refcount_bits))
3097             {
3098                 error_report("Refcount width must be a power of two and may "
3099                              "not exceed 64 bits");
3100                 return -EINVAL;
3101             }
3102         } else {
3103             /* if this point is reached, this probably means a new option was
3104              * added without having it covered here */
3105             abort();
3106         }
3107
3108         desc++;
3109     }
3110
3111     helper_cb_info = (Qcow2AmendHelperCBInfo){
3112         .original_status_cb = status_cb,
3113         .original_cb_opaque = cb_opaque,
3114         .total_operations = (new_version < old_version)
3115                           + (s->refcount_bits != refcount_bits)
3116     };
3117
3118     /* Upgrade first (some features may require compat=1.1) */
3119     if (new_version > old_version) {
3120         s->qcow_version = new_version;
3121         ret = qcow2_update_header(bs);
3122         if (ret < 0) {
3123             s->qcow_version = old_version;
3124             return ret;
3125         }
3126     }
3127
3128     if (s->refcount_bits != refcount_bits) {
3129         int refcount_order = ctz32(refcount_bits);
3130         Error *local_error = NULL;
3131
3132         if (new_version < 3 && refcount_bits != 16) {
3133             error_report("Different refcount widths than 16 bits require "
3134                          "compatibility level 1.1 or above (use compat=1.1 or "
3135                          "greater)");
3136             return -EINVAL;
3137         }
3138
3139         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
3140         ret = qcow2_change_refcount_order(bs, refcount_order,
3141                                           &qcow2_amend_helper_cb,
3142                                           &helper_cb_info, &local_error);
3143         if (ret < 0) {
3144             error_report_err(local_error);
3145             return ret;
3146         }
3147     }
3148
3149     if (backing_file || backing_format) {
3150         ret = qcow2_change_backing_file(bs,
3151                     backing_file ?: s->image_backing_file,
3152                     backing_format ?: s->image_backing_format);
3153         if (ret < 0) {
3154             return ret;
3155         }
3156     }
3157
3158     if (s->use_lazy_refcounts != lazy_refcounts) {
3159         if (lazy_refcounts) {
3160             if (new_version < 3) {
3161                 error_report("Lazy refcounts only supported with compatibility "
3162                              "level 1.1 and above (use compat=1.1 or greater)");
3163                 return -EINVAL;
3164             }
3165             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3166             ret = qcow2_update_header(bs);
3167             if (ret < 0) {
3168                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3169                 return ret;
3170             }
3171             s->use_lazy_refcounts = true;
3172         } else {
3173             /* make image clean first */
3174             ret = qcow2_mark_clean(bs);
3175             if (ret < 0) {
3176                 return ret;
3177             }
3178             /* now disallow lazy refcounts */
3179             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3180             ret = qcow2_update_header(bs);
3181             if (ret < 0) {
3182                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3183                 return ret;
3184             }
3185             s->use_lazy_refcounts = false;
3186         }
3187     }
3188
3189     if (new_size) {
3190         ret = bdrv_truncate(bs, new_size);
3191         if (ret < 0) {
3192             return ret;
3193         }
3194     }
3195
3196     /* Downgrade last (so unsupported features can be removed before) */
3197     if (new_version < old_version) {
3198         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
3199         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
3200                               &helper_cb_info);
3201         if (ret < 0) {
3202             return ret;
3203         }
3204     }
3205
3206     return 0;
3207 }
3208
3209 /*
3210  * If offset or size are negative, respectively, they will not be included in
3211  * the BLOCK_IMAGE_CORRUPTED event emitted.
3212  * fatal will be ignored for read-only BDS; corruptions found there will always
3213  * be considered non-fatal.
3214  */
3215 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
3216                              int64_t size, const char *message_format, ...)
3217 {
3218     BDRVQcow2State *s = bs->opaque;
3219     const char *node_name;
3220     char *message;
3221     va_list ap;
3222
3223     fatal = fatal && !bs->read_only;
3224
3225     if (s->signaled_corruption &&
3226         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
3227     {
3228         return;
3229     }
3230
3231     va_start(ap, message_format);
3232     message = g_strdup_vprintf(message_format, ap);
3233     va_end(ap);
3234
3235     if (fatal) {
3236         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
3237                 "corruption events will be suppressed\n", message);
3238     } else {
3239         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
3240                 "corruption events will be suppressed\n", message);
3241     }
3242
3243     node_name = bdrv_get_node_name(bs);
3244     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
3245                                           *node_name != '\0', node_name,
3246                                           message, offset >= 0, offset,
3247                                           size >= 0, size,
3248                                           fatal, &error_abort);
3249     g_free(message);
3250
3251     if (fatal) {
3252         qcow2_mark_corrupt(bs);
3253         bs->drv = NULL; /* make BDS unusable */
3254     }
3255
3256     s->signaled_corruption = true;
3257 }
3258
3259 static QemuOptsList qcow2_create_opts = {
3260     .name = "qcow2-create-opts",
3261     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
3262     .desc = {
3263         {
3264             .name = BLOCK_OPT_SIZE,
3265             .type = QEMU_OPT_SIZE,
3266             .help = "Virtual disk size"
3267         },
3268         {
3269             .name = BLOCK_OPT_COMPAT_LEVEL,
3270             .type = QEMU_OPT_STRING,
3271             .help = "Compatibility level (0.10 or 1.1)"
3272         },
3273         {
3274             .name = BLOCK_OPT_BACKING_FILE,
3275             .type = QEMU_OPT_STRING,
3276             .help = "File name of a base image"
3277         },
3278         {
3279             .name = BLOCK_OPT_BACKING_FMT,
3280             .type = QEMU_OPT_STRING,
3281             .help = "Image format of the base image"
3282         },
3283         {
3284             .name = BLOCK_OPT_ENCRYPT,
3285             .type = QEMU_OPT_BOOL,
3286             .help = "Encrypt the image",
3287             .def_value_str = "off"
3288         },
3289         {
3290             .name = BLOCK_OPT_CLUSTER_SIZE,
3291             .type = QEMU_OPT_SIZE,
3292             .help = "qcow2 cluster size",
3293             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
3294         },
3295         {
3296             .name = BLOCK_OPT_PREALLOC,
3297             .type = QEMU_OPT_STRING,
3298             .help = "Preallocation mode (allowed values: off, metadata, "
3299                     "falloc, full)"
3300         },
3301         {
3302             .name = BLOCK_OPT_LAZY_REFCOUNTS,
3303             .type = QEMU_OPT_BOOL,
3304             .help = "Postpone refcount updates",
3305             .def_value_str = "off"
3306         },
3307         {
3308             .name = BLOCK_OPT_REFCOUNT_BITS,
3309             .type = QEMU_OPT_NUMBER,
3310             .help = "Width of a reference count entry in bits",
3311             .def_value_str = "16"
3312         },
3313         { /* end of list */ }
3314     }
3315 };
3316
3317 BlockDriver bdrv_qcow2 = {
3318     .format_name        = "qcow2",
3319     .instance_size      = sizeof(BDRVQcow2State),
3320     .bdrv_probe         = qcow2_probe,
3321     .bdrv_open          = qcow2_open,
3322     .bdrv_close         = qcow2_close,
3323     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
3324     .bdrv_reopen_commit   = qcow2_reopen_commit,
3325     .bdrv_reopen_abort    = qcow2_reopen_abort,
3326     .bdrv_join_options    = qcow2_join_options,
3327     .bdrv_create        = qcow2_create,
3328     .bdrv_has_zero_init = bdrv_has_zero_init_1,
3329     .bdrv_co_get_block_status = qcow2_co_get_block_status,
3330     .bdrv_set_key       = qcow2_set_key,
3331
3332     .bdrv_co_readv          = qcow2_co_readv,
3333     .bdrv_co_writev         = qcow2_co_writev,
3334     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
3335
3336     .bdrv_co_write_zeroes   = qcow2_co_write_zeroes,
3337     .bdrv_co_discard        = qcow2_co_discard,
3338     .bdrv_truncate          = qcow2_truncate,
3339     .bdrv_write_compressed  = qcow2_write_compressed,
3340     .bdrv_make_empty        = qcow2_make_empty,
3341
3342     .bdrv_snapshot_create   = qcow2_snapshot_create,
3343     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
3344     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
3345     .bdrv_snapshot_list     = qcow2_snapshot_list,
3346     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
3347     .bdrv_get_info          = qcow2_get_info,
3348     .bdrv_get_specific_info = qcow2_get_specific_info,
3349
3350     .bdrv_save_vmstate    = qcow2_save_vmstate,
3351     .bdrv_load_vmstate    = qcow2_load_vmstate,
3352
3353     .supports_backing           = true,
3354     .bdrv_change_backing_file   = qcow2_change_backing_file,
3355
3356     .bdrv_refresh_limits        = qcow2_refresh_limits,
3357     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
3358     .bdrv_inactivate            = qcow2_inactivate,
3359
3360     .create_opts         = &qcow2_create_opts,
3361     .bdrv_check          = qcow2_check,
3362     .bdrv_amend_options  = qcow2_amend_options,
3363
3364     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
3365     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
3366 };
3367
3368 static void bdrv_qcow2_init(void)
3369 {
3370     bdrv_register(&bdrv_qcow2);
3371 }
3372
3373 block_init(bdrv_qcow2_init);