block/qcow2.c

   1 /*
   2  * Block driver for the QCOW version 2 format
   3  *
   4  * Copyright (c) 2004-2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu/osdep.h"
  25 #include "block/block_int.h"
  26 #include "sysemu/block-backend.h"
  27 #include "qemu/module.h"
  28 #include <zlib.h>
  29 #include "block/qcow2.h"
  30 #include "qemu/error-report.h"
  31 #include "qapi/qmp/qerror.h"
  32 #include "qapi/qmp/qbool.h"
  33 #include "qapi/util.h"
  34 #include "qapi/qmp/types.h"
  35 #include "qapi-event.h"
  36 #include "trace.h"
  37 #include "qemu/option_int.h"
  38 #include "qemu/cutils.h"
  39 #include "qemu/bswap.h"
  40
  41 /*
  42   Differences with QCOW:
  43
  44   - Support for multiple incremental snapshots.
  45   - Memory management by reference counts.
  46   - Clusters which have a reference count of one have the bit
  47     QCOW_OFLAG_COPIED to optimize write performance.
  48   - Size of compressed clusters is stored in sectors to reduce bit usage
  49     in the cluster offsets.
  50   - Support for storing additional data (such as the VM state) in the
  51     snapshots.
  52   - If a backing store is used, the cluster size is not constrained
  53     (could be backported to QCOW).
  54   - L2 tables have always a size of one cluster.
  55 */
  56
  57
  58 typedef struct {
  59     uint32_t magic;
  60     uint32_t len;
  61 } QEMU_PACKED QCowExtension;
  62
  63 #define  QCOW2_EXT_MAGIC_END 0
  64 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
  65 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
  66
  67 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
  68 {
  69     const QCowHeader *cow_header = (const void *)buf;
  70
  71     if (buf_size >= sizeof(QCowHeader) &&
  72         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
  73         be32_to_cpu(cow_header->version) >= 2)
  74         return 100;
  75     else
  76         return 0;
  77 }
  78
  79
  80 /*
  81  * read qcow2 extension and fill bs
  82  * start reading from start_offset
  83  * finish reading upon magic of value 0 or when end_offset reached
  84  * unknown magic is skipped (future extension this version knows nothing about)
  85  * return 0 upon success, non-0 otherwise
  86  */
  87 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
  88                                  uint64_t end_offset, void **p_feature_table,
  89                                  Error **errp)
  90 {
  91     BDRVQcow2State *s = bs->opaque;
  92     QCowExtension ext;
  93     uint64_t offset;
  94     int ret;
  95
  96 #ifdef DEBUG_EXT
  97     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
  98 #endif
  99     offset = start_offset;
 100     while (offset < end_offset) {
 101
 102 #ifdef DEBUG_EXT
 103         /* Sanity check */
 104         if (offset > s->cluster_size)
 105             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
 106
 107         printf("attempting to read extended header in offset %lu\n", offset);
 108 #endif
 109
 110         ret = bdrv_pread(bs->file->bs, offset, &ext, sizeof(ext));
 111         if (ret < 0) {
 112             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
 113                              "pread fail from offset %" PRIu64, offset);
 114             return 1;
 115         }
 116         be32_to_cpus(&ext.magic);
 117         be32_to_cpus(&ext.len);
 118         offset += sizeof(ext);
 119 #ifdef DEBUG_EXT
 120         printf("ext.magic = 0x%x\n", ext.magic);
 121 #endif
 122         if (offset > end_offset || ext.len > end_offset - offset) {
 123             error_setg(errp, "Header extension too large");
 124             return -EINVAL;
 125         }
 126
 127         switch (ext.magic) {
 128         case QCOW2_EXT_MAGIC_END:
 129             return 0;
 130
 131         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
 132             if (ext.len >= sizeof(bs->backing_format)) {
 133                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
 134                            " too large (>=%zu)", ext.len,
 135                            sizeof(bs->backing_format));
 136                 return 2;
 137             }
 138             ret = bdrv_pread(bs->file->bs, offset, bs->backing_format, ext.len);
 139             if (ret < 0) {
 140                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
 141                                  "Could not read format name");
 142                 return 3;
 143             }
 144             bs->backing_format[ext.len] = '\0';
 145             s->image_backing_format = g_strdup(bs->backing_format);
 146 #ifdef DEBUG_EXT
 147             printf("Qcow2: Got format extension %s\n", bs->backing_format);
 148 #endif
 149             break;
 150
 151         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
 152             if (p_feature_table != NULL) {
 153                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
 154                 ret = bdrv_pread(bs->file->bs, offset , feature_table, ext.len);
 155                 if (ret < 0) {
 156                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
 157                                      "Could not read table");
 158                     return ret;
 159                 }
 160
 161                 *p_feature_table = feature_table;
 162             }
 163             break;
 164
 165         default:
 166             /* unknown magic - save it in case we need to rewrite the header */
 167             {
 168                 Qcow2UnknownHeaderExtension *uext;
 169
 170                 uext = g_malloc0(sizeof(*uext)  + ext.len);
 171                 uext->magic = ext.magic;
 172                 uext->len = ext.len;
 173                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
 174
 175                 ret = bdrv_pread(bs->file->bs, offset , uext->data, uext->len);
 176                 if (ret < 0) {
 177                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
 178                                      "Could not read data");
 179                     return ret;
 180                 }
 181             }
 182             break;
 183         }
 184
 185         offset += ((ext.len + 7) & ~7);
 186     }
 187
 188     return 0;
 189 }
 190
 191 static void cleanup_unknown_header_ext(BlockDriverState *bs)
 192 {
 193     BDRVQcow2State *s = bs->opaque;
 194     Qcow2UnknownHeaderExtension *uext, *next;
 195
 196     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
 197         QLIST_REMOVE(uext, next);
 198         g_free(uext);
 199     }
 200 }
 201
 202 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
 203                                        uint64_t mask)
 204 {
 205     char *features = g_strdup("");
 206     char *old;
 207
 208     while (table && table->name[0] != '\0') {
 209         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
 210             if (mask & (1ULL << table->bit)) {
 211                 old = features;
 212                 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
 213                                            table->name);
 214                 g_free(old);
 215                 mask &= ~(1ULL << table->bit);
 216             }
 217         }
 218         table++;
 219     }
 220
 221     if (mask) {
 222         old = features;
 223         features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
 224                                    old, *old ? ", " : "", mask);
 225         g_free(old);
 226     }
 227
 228     error_setg(errp, "Unsupported qcow2 feature(s): %s", features);
 229     g_free(features);
 230 }
 231
 232 /*
 233  * Sets the dirty bit and flushes afterwards if necessary.
 234  *
 235  * The incompatible_features bit is only set if the image file header was
 236  * updated successfully.  Therefore it is not required to check the return
 237  * value of this function.
 238  */
 239 int qcow2_mark_dirty(BlockDriverState *bs)
 240 {
 241     BDRVQcow2State *s = bs->opaque;
 242     uint64_t val;
 243     int ret;
 244
 245     assert(s->qcow_version >= 3);
 246
 247     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
 248         return 0; /* already dirty */
 249     }
 250
 251     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
 252     ret = bdrv_pwrite(bs->file->bs, offsetof(QCowHeader, incompatible_features),
 253                       &val, sizeof(val));
 254     if (ret < 0) {
 255         return ret;
 256     }
 257     ret = bdrv_flush(bs->file->bs);
 258     if (ret < 0) {
 259         return ret;
 260     }
 261
 262     /* Only treat image as dirty if the header was updated successfully */
 263     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
 264     return 0;
 265 }
 266
 267 /*
 268  * Clears the dirty bit and flushes before if necessary.  Only call this
 269  * function when there are no pending requests, it does not guard against
 270  * concurrent requests dirtying the image.
 271  */
 272 static int qcow2_mark_clean(BlockDriverState *bs)
 273 {
 274     BDRVQcow2State *s = bs->opaque;
 275
 276     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
 277         int ret;
 278
 279         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
 280
 281         ret = bdrv_flush(bs);
 282         if (ret < 0) {
 283             return ret;
 284         }
 285
 286         return qcow2_update_header(bs);
 287     }
 288     return 0;
 289 }
 290
 291 /*
 292  * Marks the image as corrupt.
 293  */
 294 int qcow2_mark_corrupt(BlockDriverState *bs)
 295 {
 296     BDRVQcow2State *s = bs->opaque;
 297
 298     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
 299     return qcow2_update_header(bs);
 300 }
 301
 302 /*
 303  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
 304  * before if necessary.
 305  */
 306 int qcow2_mark_consistent(BlockDriverState *bs)
 307 {
 308     BDRVQcow2State *s = bs->opaque;
 309
 310     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
 311         int ret = bdrv_flush(bs);
 312         if (ret < 0) {
 313             return ret;
 314         }
 315
 316         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
 317         return qcow2_update_header(bs);
 318     }
 319     return 0;
 320 }
 321
 322 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
 323                        BdrvCheckMode fix)
 324 {
 325     int ret = qcow2_check_refcounts(bs, result, fix);
 326     if (ret < 0) {
 327         return ret;
 328     }
 329
 330     if (fix && result->check_errors == 0 && result->corruptions == 0) {
 331         ret = qcow2_mark_clean(bs);
 332         if (ret < 0) {
 333             return ret;
 334         }
 335         return qcow2_mark_consistent(bs);
 336     }
 337     return ret;
 338 }
 339
 340 static int validate_table_offset(BlockDriverState *bs, uint64_t offset,
 341                                  uint64_t entries, size_t entry_len)
 342 {
 343     BDRVQcow2State *s = bs->opaque;
 344     uint64_t size;
 345
 346     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
 347      * because values will be passed to qemu functions taking int64_t. */
 348     if (entries > INT64_MAX / entry_len) {
 349         return -EINVAL;
 350     }
 351
 352     size = entries * entry_len;
 353
 354     if (INT64_MAX - size < offset) {
 355         return -EINVAL;
 356     }
 357
 358     /* Tables must be cluster aligned */
 359     if (offset & (s->cluster_size - 1)) {
 360         return -EINVAL;
 361     }
 362
 363     return 0;
 364 }
 365
 366 static QemuOptsList qcow2_runtime_opts = {
 367     .name = "qcow2",
 368     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
 369     .desc = {
 370         {
 371             .name = QCOW2_OPT_LAZY_REFCOUNTS,
 372             .type = QEMU_OPT_BOOL,
 373             .help = "Postpone refcount updates",
 374         },
 375         {
 376             .name = QCOW2_OPT_DISCARD_REQUEST,
 377             .type = QEMU_OPT_BOOL,
 378             .help = "Pass guest discard requests to the layer below",
 379         },
 380         {
 381             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
 382             .type = QEMU_OPT_BOOL,
 383             .help = "Generate discard requests when snapshot related space "
 384                     "is freed",
 385         },
 386         {
 387             .name = QCOW2_OPT_DISCARD_OTHER,
 388             .type = QEMU_OPT_BOOL,
 389             .help = "Generate discard requests when other clusters are freed",
 390         },
 391         {
 392             .name = QCOW2_OPT_OVERLAP,
 393             .type = QEMU_OPT_STRING,
 394             .help = "Selects which overlap checks to perform from a range of "
 395                     "templates (none, constant, cached, all)",
 396         },
 397         {
 398             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
 399             .type = QEMU_OPT_STRING,
 400             .help = "Selects which overlap checks to perform from a range of "
 401                     "templates (none, constant, cached, all)",
 402         },
 403         {
 404             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
 405             .type = QEMU_OPT_BOOL,
 406             .help = "Check for unintended writes into the main qcow2 header",
 407         },
 408         {
 409             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
 410             .type = QEMU_OPT_BOOL,
 411             .help = "Check for unintended writes into the active L1 table",
 412         },
 413         {
 414             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
 415             .type = QEMU_OPT_BOOL,
 416             .help = "Check for unintended writes into an active L2 table",
 417         },
 418         {
 419             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
 420             .type = QEMU_OPT_BOOL,
 421             .help = "Check for unintended writes into the refcount table",
 422         },
 423         {
 424             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
 425             .type = QEMU_OPT_BOOL,
 426             .help = "Check for unintended writes into a refcount block",
 427         },
 428         {
 429             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
 430             .type = QEMU_OPT_BOOL,
 431             .help = "Check for unintended writes into the snapshot table",
 432         },
 433         {
 434             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
 435             .type = QEMU_OPT_BOOL,
 436             .help = "Check for unintended writes into an inactive L1 table",
 437         },
 438         {
 439             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
 440             .type = QEMU_OPT_BOOL,
 441             .help = "Check for unintended writes into an inactive L2 table",
 442         },
 443         {
 444             .name = QCOW2_OPT_CACHE_SIZE,
 445             .type = QEMU_OPT_SIZE,
 446             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
 447                     "cache size",
 448         },
 449         {
 450             .name = QCOW2_OPT_L2_CACHE_SIZE,
 451             .type = QEMU_OPT_SIZE,
 452             .help = "Maximum L2 table cache size",
 453         },
 454         {
 455             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
 456             .type = QEMU_OPT_SIZE,
 457             .help = "Maximum refcount block cache size",
 458         },
 459         {
 460             .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
 461             .type = QEMU_OPT_NUMBER,
 462             .help = "Clean unused cache entries after this time (in seconds)",
 463         },
 464         { /* end of list */ }
 465     },
 466 };
 467
 468 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
 469     [QCOW2_OL_MAIN_HEADER_BITNR]    = QCOW2_OPT_OVERLAP_MAIN_HEADER,
 470     [QCOW2_OL_ACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L1,
 471     [QCOW2_OL_ACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L2,
 472     [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
 473     [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
 474     [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
 475     [QCOW2_OL_INACTIVE_L1_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L1,
 476     [QCOW2_OL_INACTIVE_L2_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L2,
 477 };
 478
 479 static void cache_clean_timer_cb(void *opaque)
 480 {
 481     BlockDriverState *bs = opaque;
 482     BDRVQcow2State *s = bs->opaque;
 483     qcow2_cache_clean_unused(bs, s->l2_table_cache);
 484     qcow2_cache_clean_unused(bs, s->refcount_block_cache);
 485     timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
 486               (int64_t) s->cache_clean_interval * 1000);
 487 }
 488
 489 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
 490 {
 491     BDRVQcow2State *s = bs->opaque;
 492     if (s->cache_clean_interval > 0) {
 493         s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
 494                                              SCALE_MS, cache_clean_timer_cb,
 495                                              bs);
 496         timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
 497                   (int64_t) s->cache_clean_interval * 1000);
 498     }
 499 }
 500
 501 static void cache_clean_timer_del(BlockDriverState *bs)
 502 {
 503     BDRVQcow2State *s = bs->opaque;
 504     if (s->cache_clean_timer) {
 505         timer_del(s->cache_clean_timer);
 506         timer_free(s->cache_clean_timer);
 507         s->cache_clean_timer = NULL;
 508     }
 509 }
 510
 511 static void qcow2_detach_aio_context(BlockDriverState *bs)
 512 {
 513     cache_clean_timer_del(bs);
 514 }
 515
 516 static void qcow2_attach_aio_context(BlockDriverState *bs,
 517                                      AioContext *new_context)
 518 {
 519     cache_clean_timer_init(bs, new_context);
 520 }
 521
 522 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
 523                              uint64_t *l2_cache_size,
 524                              uint64_t *refcount_cache_size, Error **errp)
 525 {
 526     BDRVQcow2State *s = bs->opaque;
 527     uint64_t combined_cache_size;
 528     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
 529
 530     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
 531     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
 532     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
 533
 534     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
 535     *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0);
 536     *refcount_cache_size = qemu_opt_get_size(opts,
 537                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
 538
 539     if (combined_cache_size_set) {
 540         if (l2_cache_size_set && refcount_cache_size_set) {
 541             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
 542                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
 543                        "the same time");
 544             return;
 545         } else if (*l2_cache_size > combined_cache_size) {
 546             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
 547                        QCOW2_OPT_CACHE_SIZE);
 548             return;
 549         } else if (*refcount_cache_size > combined_cache_size) {
 550             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
 551                        QCOW2_OPT_CACHE_SIZE);
 552             return;
 553         }
 554
 555         if (l2_cache_size_set) {
 556             *refcount_cache_size = combined_cache_size - *l2_cache_size;
 557         } else if (refcount_cache_size_set) {
 558             *l2_cache_size = combined_cache_size - *refcount_cache_size;
 559         } else {
 560             *refcount_cache_size = combined_cache_size
 561                                  / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1);
 562             *l2_cache_size = combined_cache_size - *refcount_cache_size;
 563         }
 564     } else {
 565         if (!l2_cache_size_set && !refcount_cache_size_set) {
 566             *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE,
 567                                  (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
 568                                  * s->cluster_size);
 569             *refcount_cache_size = *l2_cache_size
 570                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 571         } else if (!l2_cache_size_set) {
 572             *l2_cache_size = *refcount_cache_size
 573                            * DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 574         } else if (!refcount_cache_size_set) {
 575             *refcount_cache_size = *l2_cache_size
 576                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 577         }
 578     }
 579 }
 580
 581 typedef struct Qcow2ReopenState {
 582     Qcow2Cache *l2_table_cache;
 583     Qcow2Cache *refcount_block_cache;
 584     bool use_lazy_refcounts;
 585     int overlap_check;
 586     bool discard_passthrough[QCOW2_DISCARD_MAX];
 587     uint64_t cache_clean_interval;
 588 } Qcow2ReopenState;
 589
 590 static int qcow2_update_options_prepare(BlockDriverState *bs,
 591                                         Qcow2ReopenState *r,
 592                                         QDict *options, int flags,
 593                                         Error **errp)
 594 {
 595     BDRVQcow2State *s = bs->opaque;
 596     QemuOpts *opts = NULL;
 597     const char *opt_overlap_check, *opt_overlap_check_template;
 598     int overlap_check_template = 0;
 599     uint64_t l2_cache_size, refcount_cache_size;
 600     int i;
 601     Error *local_err = NULL;
 602     int ret;
 603
 604     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
 605     qemu_opts_absorb_qdict(opts, options, &local_err);
 606     if (local_err) {
 607         error_propagate(errp, local_err);
 608         ret = -EINVAL;
 609         goto fail;
 610     }
 611
 612     /* get L2 table/refcount block cache size from command line options */
 613     read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size,
 614                      &local_err);
 615     if (local_err) {
 616         error_propagate(errp, local_err);
 617         ret = -EINVAL;
 618         goto fail;
 619     }
 620
 621     l2_cache_size /= s->cluster_size;
 622     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
 623         l2_cache_size = MIN_L2_CACHE_SIZE;
 624     }
 625     if (l2_cache_size > INT_MAX) {
 626         error_setg(errp, "L2 cache size too big");
 627         ret = -EINVAL;
 628         goto fail;
 629     }
 630
 631     refcount_cache_size /= s->cluster_size;
 632     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
 633         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
 634     }
 635     if (refcount_cache_size > INT_MAX) {
 636         error_setg(errp, "Refcount cache size too big");
 637         ret = -EINVAL;
 638         goto fail;
 639     }
 640
 641     /* alloc new L2 table/refcount block cache, flush old one */
 642     if (s->l2_table_cache) {
 643         ret = qcow2_cache_flush(bs, s->l2_table_cache);
 644         if (ret) {
 645             error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
 646             goto fail;
 647         }
 648     }
 649
 650     if (s->refcount_block_cache) {
 651         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
 652         if (ret) {
 653             error_setg_errno(errp, -ret,
 654                              "Failed to flush the refcount block cache");
 655             goto fail;
 656         }
 657     }
 658
 659     r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size);
 660     r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size);
 661     if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
 662         error_setg(errp, "Could not allocate metadata caches");
 663         ret = -ENOMEM;
 664         goto fail;
 665     }
 666
 667     /* New interval for cache cleanup timer */
 668     r->cache_clean_interval =
 669         qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
 670                             s->cache_clean_interval);
 671     if (r->cache_clean_interval > UINT_MAX) {
 672         error_setg(errp, "Cache clean interval too big");
 673         ret = -EINVAL;
 674         goto fail;
 675     }
 676
 677     /* lazy-refcounts; flush if going from enabled to disabled */
 678     r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
 679         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
 680     if (r->use_lazy_refcounts && s->qcow_version < 3) {
 681         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
 682                    "qemu 1.1 compatibility level");
 683         ret = -EINVAL;
 684         goto fail;
 685     }
 686
 687     if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
 688         ret = qcow2_mark_clean(bs);
 689         if (ret < 0) {
 690             error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
 691             goto fail;
 692         }
 693     }
 694
 695     /* Overlap check options */
 696     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
 697     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
 698     if (opt_overlap_check_template && opt_overlap_check &&
 699         strcmp(opt_overlap_check_template, opt_overlap_check))
 700     {
 701         error_setg(errp, "Conflicting values for qcow2 options '"
 702                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
 703                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
 704         ret = -EINVAL;
 705         goto fail;
 706     }
 707     if (!opt_overlap_check) {
 708         opt_overlap_check = opt_overlap_check_template ?: "cached";
 709     }
 710
 711     if (!strcmp(opt_overlap_check, "none")) {
 712         overlap_check_template = 0;
 713     } else if (!strcmp(opt_overlap_check, "constant")) {
 714         overlap_check_template = QCOW2_OL_CONSTANT;
 715     } else if (!strcmp(opt_overlap_check, "cached")) {
 716         overlap_check_template = QCOW2_OL_CACHED;
 717     } else if (!strcmp(opt_overlap_check, "all")) {
 718         overlap_check_template = QCOW2_OL_ALL;
 719     } else {
 720         error_setg(errp, "Unsupported value '%s' for qcow2 option "
 721                    "'overlap-check'. Allowed are any of the following: "
 722                    "none, constant, cached, all", opt_overlap_check);
 723         ret = -EINVAL;
 724         goto fail;
 725     }
 726
 727     r->overlap_check = 0;
 728     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
 729         /* overlap-check defines a template bitmask, but every flag may be
 730          * overwritten through the associated boolean option */
 731         r->overlap_check |=
 732             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
 733                               overlap_check_template & (1 << i)) << i;
 734     }
 735
 736     r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
 737     r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
 738     r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
 739         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
 740                           flags & BDRV_O_UNMAP);
 741     r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
 742         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
 743     r->discard_passthrough[QCOW2_DISCARD_OTHER] =
 744         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
 745
 746     ret = 0;
 747 fail:
 748     qemu_opts_del(opts);
 749     opts = NULL;
 750     return ret;
 751 }
 752
 753 static void qcow2_update_options_commit(BlockDriverState *bs,
 754                                         Qcow2ReopenState *r)
 755 {
 756     BDRVQcow2State *s = bs->opaque;
 757     int i;
 758
 759     if (s->l2_table_cache) {
 760         qcow2_cache_destroy(bs, s->l2_table_cache);
 761     }
 762     if (s->refcount_block_cache) {
 763         qcow2_cache_destroy(bs, s->refcount_block_cache);
 764     }
 765     s->l2_table_cache = r->l2_table_cache;
 766     s->refcount_block_cache = r->refcount_block_cache;
 767
 768     s->overlap_check = r->overlap_check;
 769     s->use_lazy_refcounts = r->use_lazy_refcounts;
 770
 771     for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
 772         s->discard_passthrough[i] = r->discard_passthrough[i];
 773     }
 774
 775     if (s->cache_clean_interval != r->cache_clean_interval) {
 776         cache_clean_timer_del(bs);
 777         s->cache_clean_interval = r->cache_clean_interval;
 778         cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
 779     }
 780 }
 781
 782 static void qcow2_update_options_abort(BlockDriverState *bs,
 783                                        Qcow2ReopenState *r)
 784 {
 785     if (r->l2_table_cache) {
 786         qcow2_cache_destroy(bs, r->l2_table_cache);
 787     }
 788     if (r->refcount_block_cache) {
 789         qcow2_cache_destroy(bs, r->refcount_block_cache);
 790     }
 791 }
 792
 793 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
 794                                 int flags, Error **errp)
 795 {
 796     Qcow2ReopenState r = {};
 797     int ret;
 798
 799     ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
 800     if (ret >= 0) {
 801         qcow2_update_options_commit(bs, &r);
 802     } else {
 803         qcow2_update_options_abort(bs, &r);
 804     }
 805
 806     return ret;
 807 }
 808
 809 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
 810                       Error **errp)
 811 {
 812     BDRVQcow2State *s = bs->opaque;
 813     unsigned int len, i;
 814     int ret = 0;
 815     QCowHeader header;
 816     Error *local_err = NULL;
 817     uint64_t ext_end;
 818     uint64_t l1_vm_state_index;
 819
 820     ret = bdrv_pread(bs->file->bs, 0, &header, sizeof(header));
 821     if (ret < 0) {
 822         error_setg_errno(errp, -ret, "Could not read qcow2 header");
 823         goto fail;
 824     }
 825     be32_to_cpus(&header.magic);
 826     be32_to_cpus(&header.version);
 827     be64_to_cpus(&header.backing_file_offset);
 828     be32_to_cpus(&header.backing_file_size);
 829     be64_to_cpus(&header.size);
 830     be32_to_cpus(&header.cluster_bits);
 831     be32_to_cpus(&header.crypt_method);
 832     be64_to_cpus(&header.l1_table_offset);
 833     be32_to_cpus(&header.l1_size);
 834     be64_to_cpus(&header.refcount_table_offset);
 835     be32_to_cpus(&header.refcount_table_clusters);
 836     be64_to_cpus(&header.snapshots_offset);
 837     be32_to_cpus(&header.nb_snapshots);
 838
 839     if (header.magic != QCOW_MAGIC) {
 840         error_setg(errp, "Image is not in qcow2 format");
 841         ret = -EINVAL;
 842         goto fail;
 843     }
 844     if (header.version < 2 || header.version > 3) {
 845         error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
 846         ret = -ENOTSUP;
 847         goto fail;
 848     }
 849
 850     s->qcow_version = header.version;
 851
 852     /* Initialise cluster size */
 853     if (header.cluster_bits < MIN_CLUSTER_BITS ||
 854         header.cluster_bits > MAX_CLUSTER_BITS) {
 855         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
 856                    header.cluster_bits);
 857         ret = -EINVAL;
 858         goto fail;
 859     }
 860
 861     s->cluster_bits = header.cluster_bits;
 862     s->cluster_size = 1 << s->cluster_bits;
 863     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 864
 865     /* Initialise version 3 header fields */
 866     if (header.version == 2) {
 867         header.incompatible_features    = 0;
 868         header.compatible_features      = 0;
 869         header.autoclear_features       = 0;
 870         header.refcount_order           = 4;
 871         header.header_length            = 72;
 872     } else {
 873         be64_to_cpus(&header.incompatible_features);
 874         be64_to_cpus(&header.compatible_features);
 875         be64_to_cpus(&header.autoclear_features);
 876         be32_to_cpus(&header.refcount_order);
 877         be32_to_cpus(&header.header_length);
 878
 879         if (header.header_length < 104) {
 880             error_setg(errp, "qcow2 header too short");
 881             ret = -EINVAL;
 882             goto fail;
 883         }
 884     }
 885
 886     if (header.header_length > s->cluster_size) {
 887         error_setg(errp, "qcow2 header exceeds cluster size");
 888         ret = -EINVAL;
 889         goto fail;
 890     }
 891
 892     if (header.header_length > sizeof(header)) {
 893         s->unknown_header_fields_size = header.header_length - sizeof(header);
 894         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
 895         ret = bdrv_pread(bs->file->bs, sizeof(header), s->unknown_header_fields,
 896                          s->unknown_header_fields_size);
 897         if (ret < 0) {
 898             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
 899                              "fields");
 900             goto fail;
 901         }
 902     }
 903
 904     if (header.backing_file_offset > s->cluster_size) {
 905         error_setg(errp, "Invalid backing file offset");
 906         ret = -EINVAL;
 907         goto fail;
 908     }
 909
 910     if (header.backing_file_offset) {
 911         ext_end = header.backing_file_offset;
 912     } else {
 913         ext_end = 1 << header.cluster_bits;
 914     }
 915
 916     /* Handle feature bits */
 917     s->incompatible_features    = header.incompatible_features;
 918     s->compatible_features      = header.compatible_features;
 919     s->autoclear_features       = header.autoclear_features;
 920
 921     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
 922         void *feature_table = NULL;
 923         qcow2_read_extensions(bs, header.header_length, ext_end,
 924                               &feature_table, NULL);
 925         report_unsupported_feature(errp, feature_table,
 926                                    s->incompatible_features &
 927                                    ~QCOW2_INCOMPAT_MASK);
 928         ret = -ENOTSUP;
 929         g_free(feature_table);
 930         goto fail;
 931     }
 932
 933     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
 934         /* Corrupt images may not be written to unless they are being repaired
 935          */
 936         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
 937             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
 938                        "read/write");
 939             ret = -EACCES;
 940             goto fail;
 941         }
 942     }
 943
 944     /* Check support for various header values */
 945     if (header.refcount_order > 6) {
 946         error_setg(errp, "Reference count entry width too large; may not "
 947                    "exceed 64 bits");
 948         ret = -EINVAL;
 949         goto fail;
 950     }
 951     s->refcount_order = header.refcount_order;
 952     s->refcount_bits = 1 << s->refcount_order;
 953     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
 954     s->refcount_max += s->refcount_max - 1;
 955
 956     if (header.crypt_method > QCOW_CRYPT_AES) {
 957         error_setg(errp, "Unsupported encryption method: %" PRIu32,
 958                    header.crypt_method);
 959         ret = -EINVAL;
 960         goto fail;
 961     }
 962     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) {
 963         error_setg(errp, "AES cipher not available");
 964         ret = -EINVAL;
 965         goto fail;
 966     }
 967     s->crypt_method_header = header.crypt_method;
 968     if (s->crypt_method_header) {
 969         if (bdrv_uses_whitelist() &&
 970             s->crypt_method_header == QCOW_CRYPT_AES) {
 971             error_report("qcow2 built-in AES encryption is deprecated");
 972             error_printf("Support for it will be removed in a future release.\n"
 973                          "You can use 'qemu-img convert' to switch to an\n"
 974                          "unencrypted qcow2 image, or a LUKS raw image.\n");
 975         }
 976
 977         bs->encrypted = 1;
 978     }
 979
 980     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
 981     s->l2_size = 1 << s->l2_bits;
 982     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
 983     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
 984     s->refcount_block_size = 1 << s->refcount_block_bits;
 985     bs->total_sectors = header.size / 512;
 986     s->csize_shift = (62 - (s->cluster_bits - 8));
 987     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
 988     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
 989
 990     s->refcount_table_offset = header.refcount_table_offset;
 991     s->refcount_table_size =
 992         header.refcount_table_clusters << (s->cluster_bits - 3);
 993
 994     if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
 995         error_setg(errp, "Reference count table too large");
 996         ret = -EINVAL;
 997         goto fail;
 998     }
 999
1000     ret = validate_table_offset(bs, s->refcount_table_offset,
1001                                 s->refcount_table_size, sizeof(uint64_t));
1002     if (ret < 0) {
1003         error_setg(errp, "Invalid reference count table offset");
1004         goto fail;
1005     }
1006
1007     /* Snapshot table offset/length */
1008     if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
1009         error_setg(errp, "Too many snapshots");
1010         ret = -EINVAL;
1011         goto fail;
1012     }
1013
1014     ret = validate_table_offset(bs, header.snapshots_offset,
1015                                 header.nb_snapshots,
1016                                 sizeof(QCowSnapshotHeader));
1017     if (ret < 0) {
1018         error_setg(errp, "Invalid snapshot table offset");
1019         goto fail;
1020     }
1021
1022     /* read the level 1 table */
1023     if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
1024         error_setg(errp, "Active L1 table too large");
1025         ret = -EFBIG;
1026         goto fail;
1027     }
1028     s->l1_size = header.l1_size;
1029
1030     l1_vm_state_index = size_to_l1(s, header.size);
1031     if (l1_vm_state_index > INT_MAX) {
1032         error_setg(errp, "Image is too big");
1033         ret = -EFBIG;
1034         goto fail;
1035     }
1036     s->l1_vm_state_index = l1_vm_state_index;
1037
1038     /* the L1 table must contain at least enough entries to put
1039        header.size bytes */
1040     if (s->l1_size < s->l1_vm_state_index) {
1041         error_setg(errp, "L1 table is too small");
1042         ret = -EINVAL;
1043         goto fail;
1044     }
1045
1046     ret = validate_table_offset(bs, header.l1_table_offset,
1047                                 header.l1_size, sizeof(uint64_t));
1048     if (ret < 0) {
1049         error_setg(errp, "Invalid L1 table offset");
1050         goto fail;
1051     }
1052     s->l1_table_offset = header.l1_table_offset;
1053
1054
1055     if (s->l1_size > 0) {
1056         s->l1_table = qemu_try_blockalign(bs->file->bs,
1057             align_offset(s->l1_size * sizeof(uint64_t), 512));
1058         if (s->l1_table == NULL) {
1059             error_setg(errp, "Could not allocate L1 table");
1060             ret = -ENOMEM;
1061             goto fail;
1062         }
1063         ret = bdrv_pread(bs->file->bs, s->l1_table_offset, s->l1_table,
1064                          s->l1_size * sizeof(uint64_t));
1065         if (ret < 0) {
1066             error_setg_errno(errp, -ret, "Could not read L1 table");
1067             goto fail;
1068         }
1069         for(i = 0;i < s->l1_size; i++) {
1070             be64_to_cpus(&s->l1_table[i]);
1071         }
1072     }
1073
1074     /* Parse driver-specific options */
1075     ret = qcow2_update_options(bs, options, flags, errp);
1076     if (ret < 0) {
1077         goto fail;
1078     }
1079
1080     s->cluster_cache = g_malloc(s->cluster_size);
1081     /* one more sector for decompressed data alignment */
1082     s->cluster_data = qemu_try_blockalign(bs->file->bs, QCOW_MAX_CRYPT_CLUSTERS
1083                                                     * s->cluster_size + 512);
1084     if (s->cluster_data == NULL) {
1085         error_setg(errp, "Could not allocate temporary cluster buffer");
1086         ret = -ENOMEM;
1087         goto fail;
1088     }
1089
1090     s->cluster_cache_offset = -1;
1091     s->flags = flags;
1092
1093     ret = qcow2_refcount_init(bs);
1094     if (ret != 0) {
1095         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1096         goto fail;
1097     }
1098
1099     QLIST_INIT(&s->cluster_allocs);
1100     QTAILQ_INIT(&s->discards);
1101
1102     /* read qcow2 extensions */
1103     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1104         &local_err)) {
1105         error_propagate(errp, local_err);
1106         ret = -EINVAL;
1107         goto fail;
1108     }
1109
1110     /* read the backing file name */
1111     if (header.backing_file_offset != 0) {
1112         len = header.backing_file_size;
1113         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1114             len >= sizeof(bs->backing_file)) {
1115             error_setg(errp, "Backing file name too long");
1116             ret = -EINVAL;
1117             goto fail;
1118         }
1119         ret = bdrv_pread(bs->file->bs, header.backing_file_offset,
1120                          bs->backing_file, len);
1121         if (ret < 0) {
1122             error_setg_errno(errp, -ret, "Could not read backing file name");
1123             goto fail;
1124         }
1125         bs->backing_file[len] = '\0';
1126         s->image_backing_file = g_strdup(bs->backing_file);
1127     }
1128
1129     /* Internal snapshots */
1130     s->snapshots_offset = header.snapshots_offset;
1131     s->nb_snapshots = header.nb_snapshots;
1132
1133     ret = qcow2_read_snapshots(bs);
1134     if (ret < 0) {
1135         error_setg_errno(errp, -ret, "Could not read snapshots");
1136         goto fail;
1137     }
1138
1139     /* Clear unknown autoclear feature bits */
1140     if (!bs->read_only && !(flags & BDRV_O_INACTIVE) && s->autoclear_features) {
1141         s->autoclear_features = 0;
1142         ret = qcow2_update_header(bs);
1143         if (ret < 0) {
1144             error_setg_errno(errp, -ret, "Could not update qcow2 header");
1145             goto fail;
1146         }
1147     }
1148
1149     /* Initialise locks */
1150     qemu_co_mutex_init(&s->lock);
1151
1152     /* Repair image if dirty */
1153     if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1154         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1155         BdrvCheckResult result = {0};
1156
1157         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1158         if (ret < 0) {
1159             error_setg_errno(errp, -ret, "Could not repair dirty image");
1160             goto fail;
1161         }
1162     }
1163
1164 #ifdef DEBUG_ALLOC
1165     {
1166         BdrvCheckResult result = {0};
1167         qcow2_check_refcounts(bs, &result, 0);
1168     }
1169 #endif
1170     return ret;
1171
1172  fail:
1173     g_free(s->unknown_header_fields);
1174     cleanup_unknown_header_ext(bs);
1175     qcow2_free_snapshots(bs);
1176     qcow2_refcount_close(bs);
1177     qemu_vfree(s->l1_table);
1178     /* else pre-write overlap checks in cache_destroy may crash */
1179     s->l1_table = NULL;
1180     cache_clean_timer_del(bs);
1181     if (s->l2_table_cache) {
1182         qcow2_cache_destroy(bs, s->l2_table_cache);
1183     }
1184     if (s->refcount_block_cache) {
1185         qcow2_cache_destroy(bs, s->refcount_block_cache);
1186     }
1187     g_free(s->cluster_cache);
1188     qemu_vfree(s->cluster_data);
1189     return ret;
1190 }
1191
1192 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1193 {
1194     BDRVQcow2State *s = bs->opaque;
1195
1196     bs->bl.write_zeroes_alignment = s->cluster_sectors;
1197 }
1198
1199 static int qcow2_set_key(BlockDriverState *bs, const char *key)
1200 {
1201     BDRVQcow2State *s = bs->opaque;
1202     uint8_t keybuf[16];
1203     int len, i;
1204     Error *err = NULL;
1205
1206     memset(keybuf, 0, 16);
1207     len = strlen(key);
1208     if (len > 16)
1209         len = 16;
1210     /* XXX: we could compress the chars to 7 bits to increase
1211        entropy */
1212     for(i = 0;i < len;i++) {
1213         keybuf[i] = key[i];
1214     }
1215     assert(bs->encrypted);
1216
1217     qcrypto_cipher_free(s->cipher);
1218     s->cipher = qcrypto_cipher_new(
1219         QCRYPTO_CIPHER_ALG_AES_128,
1220         QCRYPTO_CIPHER_MODE_CBC,
1221         keybuf, G_N_ELEMENTS(keybuf),
1222         &err);
1223
1224     if (!s->cipher) {
1225         /* XXX would be nice if errors in this method could
1226          * be properly propagate to the caller. Would need
1227          * the bdrv_set_key() API signature to be fixed. */
1228         error_free(err);
1229         return -1;
1230     }
1231     return 0;
1232 }
1233
1234 static int qcow2_reopen_prepare(BDRVReopenState *state,
1235                                 BlockReopenQueue *queue, Error **errp)
1236 {
1237     Qcow2ReopenState *r;
1238     int ret;
1239
1240     r = g_new0(Qcow2ReopenState, 1);
1241     state->opaque = r;
1242
1243     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1244                                        state->flags, errp);
1245     if (ret < 0) {
1246         goto fail;
1247     }
1248
1249     /* We need to write out any unwritten data if we reopen read-only. */
1250     if ((state->flags & BDRV_O_RDWR) == 0) {
1251         ret = bdrv_flush(state->bs);
1252         if (ret < 0) {
1253             goto fail;
1254         }
1255
1256         ret = qcow2_mark_clean(state->bs);
1257         if (ret < 0) {
1258             goto fail;
1259         }
1260     }
1261
1262     return 0;
1263
1264 fail:
1265     qcow2_update_options_abort(state->bs, r);
1266     g_free(r);
1267     return ret;
1268 }
1269
1270 static void qcow2_reopen_commit(BDRVReopenState *state)
1271 {
1272     qcow2_update_options_commit(state->bs, state->opaque);
1273     g_free(state->opaque);
1274 }
1275
1276 static void qcow2_reopen_abort(BDRVReopenState *state)
1277 {
1278     qcow2_update_options_abort(state->bs, state->opaque);
1279     g_free(state->opaque);
1280 }
1281
1282 static void qcow2_join_options(QDict *options, QDict *old_options)
1283 {
1284     bool has_new_overlap_template =
1285         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1286         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1287     bool has_new_total_cache_size =
1288         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1289     bool has_all_cache_options;
1290
1291     /* New overlap template overrides all old overlap options */
1292     if (has_new_overlap_template) {
1293         qdict_del(old_options, QCOW2_OPT_OVERLAP);
1294         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1295         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1296         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1297         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1298         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1299         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1300         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1301         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1302         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1303     }
1304
1305     /* New total cache size overrides all old options */
1306     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1307         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1308         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1309     }
1310
1311     qdict_join(options, old_options, false);
1312
1313     /*
1314      * If after merging all cache size options are set, an old total size is
1315      * overwritten. Do keep all options, however, if all three are new. The
1316      * resulting error message is what we want to happen.
1317      */
1318     has_all_cache_options =
1319         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1320         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1321         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1322
1323     if (has_all_cache_options && !has_new_total_cache_size) {
1324         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1325     }
1326 }
1327
1328 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1329         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1330 {
1331     BDRVQcow2State *s = bs->opaque;
1332     uint64_t cluster_offset;
1333     int index_in_cluster, ret;
1334     int64_t status = 0;
1335
1336     *pnum = nb_sectors;
1337     qemu_co_mutex_lock(&s->lock);
1338     ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
1339     qemu_co_mutex_unlock(&s->lock);
1340     if (ret < 0) {
1341         return ret;
1342     }
1343
1344     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1345         !s->cipher) {
1346         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1347         cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1348         *file = bs->file->bs;
1349         status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1350     }
1351     if (ret == QCOW2_CLUSTER_ZERO) {
1352         status |= BDRV_BLOCK_ZERO;
1353     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1354         status |= BDRV_BLOCK_DATA;
1355     }
1356     return status;
1357 }
1358
1359 /* handle reading after the end of the backing file */
1360 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
1361                   int64_t sector_num, int nb_sectors)
1362 {
1363     int n1;
1364     if ((sector_num + nb_sectors) <= bs->total_sectors)
1365         return nb_sectors;
1366     if (sector_num >= bs->total_sectors)
1367         n1 = 0;
1368     else
1369         n1 = bs->total_sectors - sector_num;
1370
1371     qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1));
1372
1373     return n1;
1374 }
1375
1376 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
1377                           int remaining_sectors, QEMUIOVector *qiov)
1378 {
1379     BDRVQcow2State *s = bs->opaque;
1380     int index_in_cluster, n1;
1381     int ret;
1382     int cur_nr_sectors; /* number of sectors in current iteration */
1383     uint64_t cluster_offset = 0;
1384     uint64_t bytes_done = 0;
1385     QEMUIOVector hd_qiov;
1386     uint8_t *cluster_data = NULL;
1387
1388     qemu_iovec_init(&hd_qiov, qiov->niov);
1389
1390     qemu_co_mutex_lock(&s->lock);
1391
1392     while (remaining_sectors != 0) {
1393
1394         /* prepare next request */
1395         cur_nr_sectors = remaining_sectors;
1396         if (s->cipher) {
1397             cur_nr_sectors = MIN(cur_nr_sectors,
1398                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1399         }
1400
1401         ret = qcow2_get_cluster_offset(bs, sector_num << 9,
1402             &cur_nr_sectors, &cluster_offset);
1403         if (ret < 0) {
1404             goto fail;
1405         }
1406
1407         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1408
1409         qemu_iovec_reset(&hd_qiov);
1410         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1411             cur_nr_sectors * 512);
1412
1413         switch (ret) {
1414         case QCOW2_CLUSTER_UNALLOCATED:
1415
1416             if (bs->backing) {
1417                 /* read from the base image */
1418                 n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov,
1419                     sector_num, cur_nr_sectors);
1420                 if (n1 > 0) {
1421                     QEMUIOVector local_qiov;
1422
1423                     qemu_iovec_init(&local_qiov, hd_qiov.niov);
1424                     qemu_iovec_concat(&local_qiov, &hd_qiov, 0,
1425                                       n1 * BDRV_SECTOR_SIZE);
1426
1427                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1428                     qemu_co_mutex_unlock(&s->lock);
1429                     ret = bdrv_co_readv(bs->backing->bs, sector_num,
1430                                         n1, &local_qiov);
1431                     qemu_co_mutex_lock(&s->lock);
1432
1433                     qemu_iovec_destroy(&local_qiov);
1434
1435                     if (ret < 0) {
1436                         goto fail;
1437                     }
1438                 }
1439             } else {
1440                 /* Note: in this case, no need to wait */
1441                 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1442             }
1443             break;
1444
1445         case QCOW2_CLUSTER_ZERO:
1446             qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1447             break;
1448
1449         case QCOW2_CLUSTER_COMPRESSED:
1450             /* add AIO support for compressed blocks ? */
1451             ret = qcow2_decompress_cluster(bs, cluster_offset);
1452             if (ret < 0) {
1453                 goto fail;
1454             }
1455
1456             qemu_iovec_from_buf(&hd_qiov, 0,
1457                 s->cluster_cache + index_in_cluster * 512,
1458                 512 * cur_nr_sectors);
1459             break;
1460
1461         case QCOW2_CLUSTER_NORMAL:
1462             if ((cluster_offset & 511) != 0) {
1463                 ret = -EIO;
1464                 goto fail;
1465             }
1466
1467             if (bs->encrypted) {
1468                 assert(s->cipher);
1469
1470                 /*
1471                  * For encrypted images, read everything into a temporary
1472                  * contiguous buffer on which the AES functions can work.
1473                  */
1474                 if (!cluster_data) {
1475                     cluster_data =
1476                         qemu_try_blockalign(bs->file->bs,
1477                                             QCOW_MAX_CRYPT_CLUSTERS
1478                                             * s->cluster_size);
1479                     if (cluster_data == NULL) {
1480                         ret = -ENOMEM;
1481                         goto fail;
1482                     }
1483                 }
1484
1485                 assert(cur_nr_sectors <=
1486                     QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1487                 qemu_iovec_reset(&hd_qiov);
1488                 qemu_iovec_add(&hd_qiov, cluster_data,
1489                     512 * cur_nr_sectors);
1490             }
1491
1492             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1493             qemu_co_mutex_unlock(&s->lock);
1494             ret = bdrv_co_readv(bs->file->bs,
1495                                 (cluster_offset >> 9) + index_in_cluster,
1496                                 cur_nr_sectors, &hd_qiov);
1497             qemu_co_mutex_lock(&s->lock);
1498             if (ret < 0) {
1499                 goto fail;
1500             }
1501             if (bs->encrypted) {
1502                 assert(s->cipher);
1503                 Error *err = NULL;
1504                 if (qcow2_encrypt_sectors(s, sector_num,  cluster_data,
1505                                           cluster_data, cur_nr_sectors, false,
1506                                           &err) < 0) {
1507                     error_free(err);
1508                     ret = -EIO;
1509                     goto fail;
1510                 }
1511                 qemu_iovec_from_buf(qiov, bytes_done,
1512                     cluster_data, 512 * cur_nr_sectors);
1513             }
1514             break;
1515
1516         default:
1517             g_assert_not_reached();
1518             ret = -EIO;
1519             goto fail;
1520         }
1521
1522         remaining_sectors -= cur_nr_sectors;
1523         sector_num += cur_nr_sectors;
1524         bytes_done += cur_nr_sectors * 512;
1525     }
1526     ret = 0;
1527
1528 fail:
1529     qemu_co_mutex_unlock(&s->lock);
1530
1531     qemu_iovec_destroy(&hd_qiov);
1532     qemu_vfree(cluster_data);
1533
1534     return ret;
1535 }
1536
1537 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
1538                            int64_t sector_num,
1539                            int remaining_sectors,
1540                            QEMUIOVector *qiov)
1541 {
1542     BDRVQcow2State *s = bs->opaque;
1543     int index_in_cluster;
1544     int ret;
1545     int cur_nr_sectors; /* number of sectors in current iteration */
1546     uint64_t cluster_offset;
1547     QEMUIOVector hd_qiov;
1548     uint64_t bytes_done = 0;
1549     uint8_t *cluster_data = NULL;
1550     QCowL2Meta *l2meta = NULL;
1551
1552     trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
1553                                  remaining_sectors);
1554
1555     qemu_iovec_init(&hd_qiov, qiov->niov);
1556
1557     s->cluster_cache_offset = -1; /* disable compressed cache */
1558
1559     qemu_co_mutex_lock(&s->lock);
1560
1561     while (remaining_sectors != 0) {
1562
1563         l2meta = NULL;
1564
1565         trace_qcow2_writev_start_part(qemu_coroutine_self());
1566         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1567         cur_nr_sectors = remaining_sectors;
1568         if (bs->encrypted &&
1569             cur_nr_sectors >
1570             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster) {
1571             cur_nr_sectors =
1572                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster;
1573         }
1574
1575         ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
1576             &cur_nr_sectors, &cluster_offset, &l2meta);
1577         if (ret < 0) {
1578             goto fail;
1579         }
1580
1581         assert((cluster_offset & 511) == 0);
1582
1583         qemu_iovec_reset(&hd_qiov);
1584         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1585             cur_nr_sectors * 512);
1586
1587         if (bs->encrypted) {
1588             Error *err = NULL;
1589             assert(s->cipher);
1590             if (!cluster_data) {
1591                 cluster_data = qemu_try_blockalign(bs->file->bs,
1592                                                    QCOW_MAX_CRYPT_CLUSTERS
1593                                                    * s->cluster_size);
1594                 if (cluster_data == NULL) {
1595                     ret = -ENOMEM;
1596                     goto fail;
1597                 }
1598             }
1599
1600             assert(hd_qiov.size <=
1601                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1602             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1603
1604             if (qcow2_encrypt_sectors(s, sector_num, cluster_data,
1605                                       cluster_data, cur_nr_sectors,
1606                                       true, &err) < 0) {
1607                 error_free(err);
1608                 ret = -EIO;
1609                 goto fail;
1610             }
1611
1612             qemu_iovec_reset(&hd_qiov);
1613             qemu_iovec_add(&hd_qiov, cluster_data,
1614                 cur_nr_sectors * 512);
1615         }
1616
1617         ret = qcow2_pre_write_overlap_check(bs, 0,
1618                 cluster_offset + index_in_cluster * BDRV_SECTOR_SIZE,
1619                 cur_nr_sectors * BDRV_SECTOR_SIZE);
1620         if (ret < 0) {
1621             goto fail;
1622         }
1623
1624         qemu_co_mutex_unlock(&s->lock);
1625         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1626         trace_qcow2_writev_data(qemu_coroutine_self(),
1627                                 (cluster_offset >> 9) + index_in_cluster);
1628         ret = bdrv_co_writev(bs->file->bs,
1629                              (cluster_offset >> 9) + index_in_cluster,
1630                              cur_nr_sectors, &hd_qiov);
1631         qemu_co_mutex_lock(&s->lock);
1632         if (ret < 0) {
1633             goto fail;
1634         }
1635
1636         while (l2meta != NULL) {
1637             QCowL2Meta *next;
1638
1639             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1640             if (ret < 0) {
1641                 goto fail;
1642             }
1643
1644             /* Take the request off the list of running requests */
1645             if (l2meta->nb_clusters != 0) {
1646                 QLIST_REMOVE(l2meta, next_in_flight);
1647             }
1648
1649             qemu_co_queue_restart_all(&l2meta->dependent_requests);
1650
1651             next = l2meta->next;
1652             g_free(l2meta);
1653             l2meta = next;
1654         }
1655
1656         remaining_sectors -= cur_nr_sectors;
1657         sector_num += cur_nr_sectors;
1658         bytes_done += cur_nr_sectors * 512;
1659         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
1660     }
1661     ret = 0;
1662
1663 fail:
1664     qemu_co_mutex_unlock(&s->lock);
1665
1666     while (l2meta != NULL) {
1667         QCowL2Meta *next;
1668
1669         if (l2meta->nb_clusters != 0) {
1670             QLIST_REMOVE(l2meta, next_in_flight);
1671         }
1672         qemu_co_queue_restart_all(&l2meta->dependent_requests);
1673
1674         next = l2meta->next;
1675         g_free(l2meta);
1676         l2meta = next;
1677     }
1678
1679     qemu_iovec_destroy(&hd_qiov);
1680     qemu_vfree(cluster_data);
1681     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
1682
1683     return ret;
1684 }
1685
1686 static int qcow2_inactivate(BlockDriverState *bs)
1687 {
1688     BDRVQcow2State *s = bs->opaque;
1689     int ret, result = 0;
1690
1691     ret = qcow2_cache_flush(bs, s->l2_table_cache);
1692     if (ret) {
1693         result = ret;
1694         error_report("Failed to flush the L2 table cache: %s",
1695                      strerror(-ret));
1696     }
1697
1698     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1699     if (ret) {
1700         result = ret;
1701         error_report("Failed to flush the refcount block cache: %s",
1702                      strerror(-ret));
1703     }
1704
1705     if (result == 0) {
1706         qcow2_mark_clean(bs);
1707     }
1708
1709     return result;
1710 }
1711
1712 static void qcow2_close(BlockDriverState *bs)
1713 {
1714     BDRVQcow2State *s = bs->opaque;
1715     qemu_vfree(s->l1_table);
1716     /* else pre-write overlap checks in cache_destroy may crash */
1717     s->l1_table = NULL;
1718
1719     if (!(s->flags & BDRV_O_INACTIVE)) {
1720         qcow2_inactivate(bs);
1721     }
1722
1723     cache_clean_timer_del(bs);
1724     qcow2_cache_destroy(bs, s->l2_table_cache);
1725     qcow2_cache_destroy(bs, s->refcount_block_cache);
1726
1727     qcrypto_cipher_free(s->cipher);
1728     s->cipher = NULL;
1729
1730     g_free(s->unknown_header_fields);
1731     cleanup_unknown_header_ext(bs);
1732
1733     g_free(s->image_backing_file);
1734     g_free(s->image_backing_format);
1735
1736     g_free(s->cluster_cache);
1737     qemu_vfree(s->cluster_data);
1738     qcow2_refcount_close(bs);
1739     qcow2_free_snapshots(bs);
1740 }
1741
1742 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
1743 {
1744     BDRVQcow2State *s = bs->opaque;
1745     int flags = s->flags;
1746     QCryptoCipher *cipher = NULL;
1747     QDict *options;
1748     Error *local_err = NULL;
1749     int ret;
1750
1751     /*
1752      * Backing files are read-only which makes all of their metadata immutable,
1753      * that means we don't have to worry about reopening them here.
1754      */
1755
1756     cipher = s->cipher;
1757     s->cipher = NULL;
1758
1759     qcow2_close(bs);
1760
1761     memset(s, 0, sizeof(BDRVQcow2State));
1762     options = qdict_clone_shallow(bs->options);
1763
1764     flags &= ~BDRV_O_INACTIVE;
1765     ret = qcow2_open(bs, options, flags, &local_err);
1766     QDECREF(options);
1767     if (local_err) {
1768         error_propagate(errp, local_err);
1769         error_prepend(errp, "Could not reopen qcow2 layer: ");
1770         bs->drv = NULL;
1771         return;
1772     } else if (ret < 0) {
1773         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
1774         bs->drv = NULL;
1775         return;
1776     }
1777
1778     s->cipher = cipher;
1779 }
1780
1781 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
1782     size_t len, size_t buflen)
1783 {
1784     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
1785     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
1786
1787     if (buflen < ext_len) {
1788         return -ENOSPC;
1789     }
1790
1791     *ext_backing_fmt = (QCowExtension) {
1792         .magic  = cpu_to_be32(magic),
1793         .len    = cpu_to_be32(len),
1794     };
1795     memcpy(buf + sizeof(QCowExtension), s, len);
1796
1797     return ext_len;
1798 }
1799
1800 /*
1801  * Updates the qcow2 header, including the variable length parts of it, i.e.
1802  * the backing file name and all extensions. qcow2 was not designed to allow
1803  * such changes, so if we run out of space (we can only use the first cluster)
1804  * this function may fail.
1805  *
1806  * Returns 0 on success, -errno in error cases.
1807  */
1808 int qcow2_update_header(BlockDriverState *bs)
1809 {
1810     BDRVQcow2State *s = bs->opaque;
1811     QCowHeader *header;
1812     char *buf;
1813     size_t buflen = s->cluster_size;
1814     int ret;
1815     uint64_t total_size;
1816     uint32_t refcount_table_clusters;
1817     size_t header_length;
1818     Qcow2UnknownHeaderExtension *uext;
1819
1820     buf = qemu_blockalign(bs, buflen);
1821
1822     /* Header structure */
1823     header = (QCowHeader*) buf;
1824
1825     if (buflen < sizeof(*header)) {
1826         ret = -ENOSPC;
1827         goto fail;
1828     }
1829
1830     header_length = sizeof(*header) + s->unknown_header_fields_size;
1831     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1832     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1833
1834     *header = (QCowHeader) {
1835         /* Version 2 fields */
1836         .magic                  = cpu_to_be32(QCOW_MAGIC),
1837         .version                = cpu_to_be32(s->qcow_version),
1838         .backing_file_offset    = 0,
1839         .backing_file_size      = 0,
1840         .cluster_bits           = cpu_to_be32(s->cluster_bits),
1841         .size                   = cpu_to_be64(total_size),
1842         .crypt_method           = cpu_to_be32(s->crypt_method_header),
1843         .l1_size                = cpu_to_be32(s->l1_size),
1844         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
1845         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
1846         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1847         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
1848         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
1849
1850         /* Version 3 fields */
1851         .incompatible_features  = cpu_to_be64(s->incompatible_features),
1852         .compatible_features    = cpu_to_be64(s->compatible_features),
1853         .autoclear_features     = cpu_to_be64(s->autoclear_features),
1854         .refcount_order         = cpu_to_be32(s->refcount_order),
1855         .header_length          = cpu_to_be32(header_length),
1856     };
1857
1858     /* For older versions, write a shorter header */
1859     switch (s->qcow_version) {
1860     case 2:
1861         ret = offsetof(QCowHeader, incompatible_features);
1862         break;
1863     case 3:
1864         ret = sizeof(*header);
1865         break;
1866     default:
1867         ret = -EINVAL;
1868         goto fail;
1869     }
1870
1871     buf += ret;
1872     buflen -= ret;
1873     memset(buf, 0, buflen);
1874
1875     /* Preserve any unknown field in the header */
1876     if (s->unknown_header_fields_size) {
1877         if (buflen < s->unknown_header_fields_size) {
1878             ret = -ENOSPC;
1879             goto fail;
1880         }
1881
1882         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1883         buf += s->unknown_header_fields_size;
1884         buflen -= s->unknown_header_fields_size;
1885     }
1886
1887     /* Backing file format header extension */
1888     if (s->image_backing_format) {
1889         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1890                              s->image_backing_format,
1891                              strlen(s->image_backing_format),
1892                              buflen);
1893         if (ret < 0) {
1894             goto fail;
1895         }
1896
1897         buf += ret;
1898         buflen -= ret;
1899     }
1900
1901     /* Feature table */
1902     if (s->qcow_version >= 3) {
1903         Qcow2Feature features[] = {
1904             {
1905                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1906                 .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
1907                 .name = "dirty bit",
1908             },
1909             {
1910                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1911                 .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
1912                 .name = "corrupt bit",
1913             },
1914             {
1915                 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1916                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1917                 .name = "lazy refcounts",
1918             },
1919         };
1920
1921         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1922                              features, sizeof(features), buflen);
1923         if (ret < 0) {
1924             goto fail;
1925         }
1926         buf += ret;
1927         buflen -= ret;
1928     }
1929
1930     /* Keep unknown header extensions */
1931     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1932         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1933         if (ret < 0) {
1934             goto fail;
1935         }
1936
1937         buf += ret;
1938         buflen -= ret;
1939     }
1940
1941     /* End of header extensions */
1942     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1943     if (ret < 0) {
1944         goto fail;
1945     }
1946
1947     buf += ret;
1948     buflen -= ret;
1949
1950     /* Backing file name */
1951     if (s->image_backing_file) {
1952         size_t backing_file_len = strlen(s->image_backing_file);
1953
1954         if (buflen < backing_file_len) {
1955             ret = -ENOSPC;
1956             goto fail;
1957         }
1958
1959         /* Using strncpy is ok here, since buf is not NUL-terminated. */
1960         strncpy(buf, s->image_backing_file, buflen);
1961
1962         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1963         header->backing_file_size   = cpu_to_be32(backing_file_len);
1964     }
1965
1966     /* Write the new header */
1967     ret = bdrv_pwrite(bs->file->bs, 0, header, s->cluster_size);
1968     if (ret < 0) {
1969         goto fail;
1970     }
1971
1972     ret = 0;
1973 fail:
1974     qemu_vfree(header);
1975     return ret;
1976 }
1977
1978 static int qcow2_change_backing_file(BlockDriverState *bs,
1979     const char *backing_file, const char *backing_fmt)
1980 {
1981     BDRVQcow2State *s = bs->opaque;
1982
1983     if (backing_file && strlen(backing_file) > 1023) {
1984         return -EINVAL;
1985     }
1986
1987     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1988     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1989
1990     g_free(s->image_backing_file);
1991     g_free(s->image_backing_format);
1992
1993     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
1994     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
1995
1996     return qcow2_update_header(bs);
1997 }
1998
1999 static int preallocate(BlockDriverState *bs)
2000 {
2001     uint64_t nb_sectors;
2002     uint64_t offset;
2003     uint64_t host_offset = 0;
2004     int num;
2005     int ret;
2006     QCowL2Meta *meta;
2007
2008     nb_sectors = bdrv_nb_sectors(bs);
2009     offset = 0;
2010
2011     while (nb_sectors) {
2012         num = MIN(nb_sectors, INT_MAX >> BDRV_SECTOR_BITS);
2013         ret = qcow2_alloc_cluster_offset(bs, offset, &num,
2014                                          &host_offset, &meta);
2015         if (ret < 0) {
2016             return ret;
2017         }
2018
2019         while (meta) {
2020             QCowL2Meta *next = meta->next;
2021
2022             ret = qcow2_alloc_cluster_link_l2(bs, meta);
2023             if (ret < 0) {
2024                 qcow2_free_any_clusters(bs, meta->alloc_offset,
2025                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
2026                 return ret;
2027             }
2028
2029             /* There are no dependent requests, but we need to remove our
2030              * request from the list of in-flight requests */
2031             QLIST_REMOVE(meta, next_in_flight);
2032
2033             g_free(meta);
2034             meta = next;
2035         }
2036
2037         /* TODO Preallocate data if requested */
2038
2039         nb_sectors -= num;
2040         offset += num << BDRV_SECTOR_BITS;
2041     }
2042
2043     /*
2044      * It is expected that the image file is large enough to actually contain
2045      * all of the allocated clusters (otherwise we get failing reads after
2046      * EOF). Extend the image to the last allocated sector.
2047      */
2048     if (host_offset != 0) {
2049         uint8_t buf[BDRV_SECTOR_SIZE];
2050         memset(buf, 0, BDRV_SECTOR_SIZE);
2051         ret = bdrv_write(bs->file->bs,
2052                          (host_offset >> BDRV_SECTOR_BITS) + num - 1,
2053                          buf, 1);
2054         if (ret < 0) {
2055             return ret;
2056         }
2057     }
2058
2059     return 0;
2060 }
2061
2062 static int qcow2_create2(const char *filename, int64_t total_size,
2063                          const char *backing_file, const char *backing_format,
2064                          int flags, size_t cluster_size, PreallocMode prealloc,
2065                          QemuOpts *opts, int version, int refcount_order,
2066                          Error **errp)
2067 {
2068     int cluster_bits;
2069     QDict *options;
2070
2071     /* Calculate cluster_bits */
2072     cluster_bits = ctz32(cluster_size);
2073     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
2074         (1 << cluster_bits) != cluster_size)
2075     {
2076         error_setg(errp, "Cluster size must be a power of two between %d and "
2077                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
2078         return -EINVAL;
2079     }
2080
2081     /*
2082      * Open the image file and write a minimal qcow2 header.
2083      *
2084      * We keep things simple and start with a zero-sized image. We also
2085      * do without refcount blocks or a L1 table for now. We'll fix the
2086      * inconsistency later.
2087      *
2088      * We do need a refcount table because growing the refcount table means
2089      * allocating two new refcount blocks - the seconds of which would be at
2090      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2091      * size for any qcow2 image.
2092      */
2093     BlockBackend *blk;
2094     QCowHeader *header;
2095     uint64_t* refcount_table;
2096     Error *local_err = NULL;
2097     int ret;
2098
2099     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
2100         /* Note: The following calculation does not need to be exact; if it is a
2101          * bit off, either some bytes will be "leaked" (which is fine) or we
2102          * will need to increase the file size by some bytes (which is fine,
2103          * too, as long as the bulk is allocated here). Therefore, using
2104          * floating point arithmetic is fine. */
2105         int64_t meta_size = 0;
2106         uint64_t nreftablee, nrefblocke, nl1e, nl2e;
2107         int64_t aligned_total_size = align_offset(total_size, cluster_size);
2108         int refblock_bits, refblock_size;
2109         /* refcount entry size in bytes */
2110         double rces = (1 << refcount_order) / 8.;
2111
2112         /* see qcow2_open() */
2113         refblock_bits = cluster_bits - (refcount_order - 3);
2114         refblock_size = 1 << refblock_bits;
2115
2116         /* header: 1 cluster */
2117         meta_size += cluster_size;
2118
2119         /* total size of L2 tables */
2120         nl2e = aligned_total_size / cluster_size;
2121         nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
2122         meta_size += nl2e * sizeof(uint64_t);
2123
2124         /* total size of L1 tables */
2125         nl1e = nl2e * sizeof(uint64_t) / cluster_size;
2126         nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
2127         meta_size += nl1e * sizeof(uint64_t);
2128
2129         /* total size of refcount blocks
2130          *
2131          * note: every host cluster is reference-counted, including metadata
2132          * (even refcount blocks are recursively included).
2133          * Let:
2134          *   a = total_size (this is the guest disk size)
2135          *   m = meta size not including refcount blocks and refcount tables
2136          *   c = cluster size
2137          *   y1 = number of refcount blocks entries
2138          *   y2 = meta size including everything
2139          *   rces = refcount entry size in bytes
2140          * then,
2141          *   y1 = (y2 + a)/c
2142          *   y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m
2143          * we can get y1:
2144          *   y1 = (a + m) / (c - rces - rces * sizeof(u64) / c)
2145          */
2146         nrefblocke = (aligned_total_size + meta_size + cluster_size)
2147                    / (cluster_size - rces - rces * sizeof(uint64_t)
2148                                                  / cluster_size);
2149         meta_size += DIV_ROUND_UP(nrefblocke, refblock_size) * cluster_size;
2150
2151         /* total size of refcount tables */
2152         nreftablee = nrefblocke / refblock_size;
2153         nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t));
2154         meta_size += nreftablee * sizeof(uint64_t);
2155
2156         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
2157                             aligned_total_size + meta_size, &error_abort);
2158         qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc],
2159                      &error_abort);
2160     }
2161
2162     ret = bdrv_create_file(filename, opts, &local_err);
2163     if (ret < 0) {
2164         error_propagate(errp, local_err);
2165         return ret;
2166     }
2167
2168     blk = blk_new_open(filename, NULL, NULL,
2169                        BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
2170     if (blk == NULL) {
2171         error_propagate(errp, local_err);
2172         return -EIO;
2173     }
2174
2175     blk_set_allow_write_beyond_eof(blk, true);
2176
2177     /* Write the header */
2178     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
2179     header = g_malloc0(cluster_size);
2180     *header = (QCowHeader) {
2181         .magic                      = cpu_to_be32(QCOW_MAGIC),
2182         .version                    = cpu_to_be32(version),
2183         .cluster_bits               = cpu_to_be32(cluster_bits),
2184         .size                       = cpu_to_be64(0),
2185         .l1_table_offset            = cpu_to_be64(0),
2186         .l1_size                    = cpu_to_be32(0),
2187         .refcount_table_offset      = cpu_to_be64(cluster_size),
2188         .refcount_table_clusters    = cpu_to_be32(1),
2189         .refcount_order             = cpu_to_be32(refcount_order),
2190         .header_length              = cpu_to_be32(sizeof(*header)),
2191     };
2192
2193     if (flags & BLOCK_FLAG_ENCRYPT) {
2194         header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
2195     } else {
2196         header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
2197     }
2198
2199     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
2200         header->compatible_features |=
2201             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
2202     }
2203
2204     ret = blk_pwrite(blk, 0, header, cluster_size, 0);
2205     g_free(header);
2206     if (ret < 0) {
2207         error_setg_errno(errp, -ret, "Could not write qcow2 header");
2208         goto out;
2209     }
2210
2211     /* Write a refcount table with one refcount block */
2212     refcount_table = g_malloc0(2 * cluster_size);
2213     refcount_table[0] = cpu_to_be64(2 * cluster_size);
2214     ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
2215     g_free(refcount_table);
2216
2217     if (ret < 0) {
2218         error_setg_errno(errp, -ret, "Could not write refcount table");
2219         goto out;
2220     }
2221
2222     blk_unref(blk);
2223     blk = NULL;
2224
2225     /*
2226      * And now open the image and make it consistent first (i.e. increase the
2227      * refcount of the cluster that is occupied by the header and the refcount
2228      * table)
2229      */
2230     options = qdict_new();
2231     qdict_put(options, "driver", qstring_from_str("qcow2"));
2232     blk = blk_new_open(filename, NULL, options,
2233                        BDRV_O_RDWR | BDRV_O_NO_FLUSH, &local_err);
2234     if (blk == NULL) {
2235         error_propagate(errp, local_err);
2236         ret = -EIO;
2237         goto out;
2238     }
2239
2240     ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
2241     if (ret < 0) {
2242         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
2243                          "header and refcount table");
2244         goto out;
2245
2246     } else if (ret != 0) {
2247         error_report("Huh, first cluster in empty image is already in use?");
2248         abort();
2249     }
2250
2251     /* Create a full header (including things like feature table) */
2252     ret = qcow2_update_header(blk_bs(blk));
2253     if (ret < 0) {
2254         error_setg_errno(errp, -ret, "Could not update qcow2 header");
2255         goto out;
2256     }
2257
2258     /* Okay, now that we have a valid image, let's give it the right size */
2259     ret = blk_truncate(blk, total_size);
2260     if (ret < 0) {
2261         error_setg_errno(errp, -ret, "Could not resize image");
2262         goto out;
2263     }
2264
2265     /* Want a backing file? There you go.*/
2266     if (backing_file) {
2267         ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format);
2268         if (ret < 0) {
2269             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
2270                              "with format '%s'", backing_file, backing_format);
2271             goto out;
2272         }
2273     }
2274
2275     /* And if we're supposed to preallocate metadata, do that now */
2276     if (prealloc != PREALLOC_MODE_OFF) {
2277         BDRVQcow2State *s = blk_bs(blk)->opaque;
2278         qemu_co_mutex_lock(&s->lock);
2279         ret = preallocate(blk_bs(blk));
2280         qemu_co_mutex_unlock(&s->lock);
2281         if (ret < 0) {
2282             error_setg_errno(errp, -ret, "Could not preallocate metadata");
2283             goto out;
2284         }
2285     }
2286
2287     blk_unref(blk);
2288     blk = NULL;
2289
2290     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
2291     options = qdict_new();
2292     qdict_put(options, "driver", qstring_from_str("qcow2"));
2293     blk = blk_new_open(filename, NULL, options,
2294                        BDRV_O_RDWR | BDRV_O_NO_BACKING, &local_err);
2295     if (blk == NULL) {
2296         error_propagate(errp, local_err);
2297         ret = -EIO;
2298         goto out;
2299     }
2300
2301     ret = 0;
2302 out:
2303     if (blk) {
2304         blk_unref(blk);
2305     }
2306     return ret;
2307 }
2308
2309 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
2310 {
2311     char *backing_file = NULL;
2312     char *backing_fmt = NULL;
2313     char *buf = NULL;
2314     uint64_t size = 0;
2315     int flags = 0;
2316     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
2317     PreallocMode prealloc;
2318     int version = 3;
2319     uint64_t refcount_bits = 16;
2320     int refcount_order;
2321     Error *local_err = NULL;
2322     int ret;
2323
2324     /* Read out options */
2325     size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2326                     BDRV_SECTOR_SIZE);
2327     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2328     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2329     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2330         flags |= BLOCK_FLAG_ENCRYPT;
2331     }
2332     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2333                                          DEFAULT_CLUSTER_SIZE);
2334     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2335     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
2336                                PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
2337                                &local_err);
2338     if (local_err) {
2339         error_propagate(errp, local_err);
2340         ret = -EINVAL;
2341         goto finish;
2342     }
2343     g_free(buf);
2344     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2345     if (!buf) {
2346         /* keep the default */
2347     } else if (!strcmp(buf, "0.10")) {
2348         version = 2;
2349     } else if (!strcmp(buf, "1.1")) {
2350         version = 3;
2351     } else {
2352         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2353         ret = -EINVAL;
2354         goto finish;
2355     }
2356
2357     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2358         flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2359     }
2360
2361     if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2362         error_setg(errp, "Backing file and preallocation cannot be used at "
2363                    "the same time");
2364         ret = -EINVAL;
2365         goto finish;
2366     }
2367
2368     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2369         error_setg(errp, "Lazy refcounts only supported with compatibility "
2370                    "level 1.1 and above (use compat=1.1 or greater)");
2371         ret = -EINVAL;
2372         goto finish;
2373     }
2374
2375     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS,
2376                                             refcount_bits);
2377     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2378         error_setg(errp, "Refcount width must be a power of two and may not "
2379                    "exceed 64 bits");
2380         ret = -EINVAL;
2381         goto finish;
2382     }
2383
2384     if (version < 3 && refcount_bits != 16) {
2385         error_setg(errp, "Different refcount widths than 16 bits require "
2386                    "compatibility level 1.1 or above (use compat=1.1 or "
2387                    "greater)");
2388         ret = -EINVAL;
2389         goto finish;
2390     }
2391
2392     refcount_order = ctz32(refcount_bits);
2393
2394     ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2395                         cluster_size, prealloc, opts, version, refcount_order,
2396                         &local_err);
2397     if (local_err) {
2398         error_propagate(errp, local_err);
2399     }
2400
2401 finish:
2402     g_free(backing_file);
2403     g_free(backing_fmt);
2404     g_free(buf);
2405     return ret;
2406 }
2407
2408
2409 static bool is_zero_cluster(BlockDriverState *bs, int64_t start)
2410 {
2411     BDRVQcow2State *s = bs->opaque;
2412     int nr;
2413     BlockDriverState *file;
2414     int64_t res = bdrv_get_block_status_above(bs, NULL, start,
2415                                               s->cluster_sectors, &nr, &file);
2416     return res >= 0 && ((res & BDRV_BLOCK_ZERO) || !(res & BDRV_BLOCK_DATA));
2417 }
2418
2419 static bool is_zero_cluster_top_locked(BlockDriverState *bs, int64_t start)
2420 {
2421     BDRVQcow2State *s = bs->opaque;
2422     int nr = s->cluster_sectors;
2423     uint64_t off;
2424     int ret;
2425
2426     ret = qcow2_get_cluster_offset(bs, start << BDRV_SECTOR_BITS, &nr, &off);
2427     return ret == QCOW2_CLUSTER_UNALLOCATED || ret == QCOW2_CLUSTER_ZERO;
2428 }
2429
2430 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
2431     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2432 {
2433     int ret;
2434     BDRVQcow2State *s = bs->opaque;
2435
2436     int head = sector_num % s->cluster_sectors;
2437     int tail = (sector_num + nb_sectors) % s->cluster_sectors;
2438
2439     if (head != 0 || tail != 0) {
2440         int64_t cl_end = -1;
2441
2442         sector_num -= head;
2443         nb_sectors += head;
2444
2445         if (tail != 0) {
2446             nb_sectors += s->cluster_sectors - tail;
2447         }
2448
2449         if (!is_zero_cluster(bs, sector_num)) {
2450             return -ENOTSUP;
2451         }
2452
2453         if (nb_sectors > s->cluster_sectors) {
2454             /* Technically the request can cover 2 clusters, f.e. 4k write
2455                at s->cluster_sectors - 2k offset. One of these cluster can
2456                be zeroed, one unallocated */
2457             cl_end = sector_num + nb_sectors - s->cluster_sectors;
2458             if (!is_zero_cluster(bs, cl_end)) {
2459                 return -ENOTSUP;
2460             }
2461         }
2462
2463         qemu_co_mutex_lock(&s->lock);
2464         /* We can have new write after previous check */
2465         if (!is_zero_cluster_top_locked(bs, sector_num) ||
2466                 (cl_end > 0 && !is_zero_cluster_top_locked(bs, cl_end))) {
2467             qemu_co_mutex_unlock(&s->lock);
2468             return -ENOTSUP;
2469         }
2470     } else {
2471         qemu_co_mutex_lock(&s->lock);
2472     }
2473
2474     /* Whatever is left can use real zero clusters */
2475     ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS, nb_sectors);
2476     qemu_co_mutex_unlock(&s->lock);
2477
2478     return ret;
2479 }
2480
2481 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
2482     int64_t sector_num, int nb_sectors)
2483 {
2484     int ret;
2485     BDRVQcow2State *s = bs->opaque;
2486
2487     qemu_co_mutex_lock(&s->lock);
2488     ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2489         nb_sectors, QCOW2_DISCARD_REQUEST, false);
2490     qemu_co_mutex_unlock(&s->lock);
2491     return ret;
2492 }
2493
2494 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
2495 {
2496     BDRVQcow2State *s = bs->opaque;
2497     int64_t new_l1_size;
2498     int ret;
2499
2500     if (offset & 511) {
2501         error_report("The new size must be a multiple of 512");
2502         return -EINVAL;
2503     }
2504
2505     /* cannot proceed if image has snapshots */
2506     if (s->nb_snapshots) {
2507         error_report("Can't resize an image which has snapshots");
2508         return -ENOTSUP;
2509     }
2510
2511     /* shrinking is currently not supported */
2512     if (offset < bs->total_sectors * 512) {
2513         error_report("qcow2 doesn't support shrinking images yet");
2514         return -ENOTSUP;
2515     }
2516
2517     new_l1_size = size_to_l1(s, offset);
2518     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
2519     if (ret < 0) {
2520         return ret;
2521     }
2522
2523     /* write updated header.size */
2524     offset = cpu_to_be64(offset);
2525     ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, size),
2526                            &offset, sizeof(uint64_t));
2527     if (ret < 0) {
2528         return ret;
2529     }
2530
2531     s->l1_vm_state_index = new_l1_size;
2532     return 0;
2533 }
2534
2535 /* XXX: put compressed sectors first, then all the cluster aligned
2536    tables to avoid losing bytes in alignment */
2537 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
2538                                   const uint8_t *buf, int nb_sectors)
2539 {
2540     BDRVQcow2State *s = bs->opaque;
2541     z_stream strm;
2542     int ret, out_len;
2543     uint8_t *out_buf;
2544     uint64_t cluster_offset;
2545
2546     if (nb_sectors == 0) {
2547         /* align end of file to a sector boundary to ease reading with
2548            sector based I/Os */
2549         cluster_offset = bdrv_getlength(bs->file->bs);
2550         return bdrv_truncate(bs->file->bs, cluster_offset);
2551     }
2552
2553     if (nb_sectors != s->cluster_sectors) {
2554         ret = -EINVAL;
2555
2556         /* Zero-pad last write if image size is not cluster aligned */
2557         if (sector_num + nb_sectors == bs->total_sectors &&
2558             nb_sectors < s->cluster_sectors) {
2559             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
2560             memset(pad_buf, 0, s->cluster_size);
2561             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
2562             ret = qcow2_write_compressed(bs, sector_num,
2563                                          pad_buf, s->cluster_sectors);
2564             qemu_vfree(pad_buf);
2565         }
2566         return ret;
2567     }
2568
2569     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
2570
2571     /* best compression, small window, no zlib header */
2572     memset(&strm, 0, sizeof(strm));
2573     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
2574                        Z_DEFLATED, -12,
2575                        9, Z_DEFAULT_STRATEGY);
2576     if (ret != 0) {
2577         ret = -EINVAL;
2578         goto fail;
2579     }
2580
2581     strm.avail_in = s->cluster_size;
2582     strm.next_in = (uint8_t *)buf;
2583     strm.avail_out = s->cluster_size;
2584     strm.next_out = out_buf;
2585
2586     ret = deflate(&strm, Z_FINISH);
2587     if (ret != Z_STREAM_END && ret != Z_OK) {
2588         deflateEnd(&strm);
2589         ret = -EINVAL;
2590         goto fail;
2591     }
2592     out_len = strm.next_out - out_buf;
2593
2594     deflateEnd(&strm);
2595
2596     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
2597         /* could not compress: write normal cluster */
2598         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
2599         if (ret < 0) {
2600             goto fail;
2601         }
2602     } else {
2603         cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
2604             sector_num << 9, out_len);
2605         if (!cluster_offset) {
2606             ret = -EIO;
2607             goto fail;
2608         }
2609         cluster_offset &= s->cluster_offset_mask;
2610
2611         ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
2612         if (ret < 0) {
2613             goto fail;
2614         }
2615
2616         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
2617         ret = bdrv_pwrite(bs->file->bs, cluster_offset, out_buf, out_len);
2618         if (ret < 0) {
2619             goto fail;
2620         }
2621     }
2622
2623     ret = 0;
2624 fail:
2625     g_free(out_buf);
2626     return ret;
2627 }
2628
2629 static int make_completely_empty(BlockDriverState *bs)
2630 {
2631     BDRVQcow2State *s = bs->opaque;
2632     int ret, l1_clusters;
2633     int64_t offset;
2634     uint64_t *new_reftable = NULL;
2635     uint64_t rt_entry, l1_size2;
2636     struct {
2637         uint64_t l1_offset;
2638         uint64_t reftable_offset;
2639         uint32_t reftable_clusters;
2640     } QEMU_PACKED l1_ofs_rt_ofs_cls;
2641
2642     ret = qcow2_cache_empty(bs, s->l2_table_cache);
2643     if (ret < 0) {
2644         goto fail;
2645     }
2646
2647     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
2648     if (ret < 0) {
2649         goto fail;
2650     }
2651
2652     /* Refcounts will be broken utterly */
2653     ret = qcow2_mark_dirty(bs);
2654     if (ret < 0) {
2655         goto fail;
2656     }
2657
2658     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2659
2660     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2661     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
2662
2663     /* After this call, neither the in-memory nor the on-disk refcount
2664      * information accurately describe the actual references */
2665
2666     ret = bdrv_write_zeroes(bs->file->bs, s->l1_table_offset / BDRV_SECTOR_SIZE,
2667                             l1_clusters * s->cluster_sectors, 0);
2668     if (ret < 0) {
2669         goto fail_broken_refcounts;
2670     }
2671     memset(s->l1_table, 0, l1_size2);
2672
2673     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
2674
2675     /* Overwrite enough clusters at the beginning of the sectors to place
2676      * the refcount table, a refcount block and the L1 table in; this may
2677      * overwrite parts of the existing refcount and L1 table, which is not
2678      * an issue because the dirty flag is set, complete data loss is in fact
2679      * desired and partial data loss is consequently fine as well */
2680     ret = bdrv_write_zeroes(bs->file->bs, s->cluster_size / BDRV_SECTOR_SIZE,
2681                             (2 + l1_clusters) * s->cluster_size /
2682                             BDRV_SECTOR_SIZE, 0);
2683     /* This call (even if it failed overall) may have overwritten on-disk
2684      * refcount structures; in that case, the in-memory refcount information
2685      * will probably differ from the on-disk information which makes the BDS
2686      * unusable */
2687     if (ret < 0) {
2688         goto fail_broken_refcounts;
2689     }
2690
2691     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2692     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
2693
2694     /* "Create" an empty reftable (one cluster) directly after the image
2695      * header and an empty L1 table three clusters after the image header;
2696      * the cluster between those two will be used as the first refblock */
2697     cpu_to_be64w(&l1_ofs_rt_ofs_cls.l1_offset, 3 * s->cluster_size);
2698     cpu_to_be64w(&l1_ofs_rt_ofs_cls.reftable_offset, s->cluster_size);
2699     cpu_to_be32w(&l1_ofs_rt_ofs_cls.reftable_clusters, 1);
2700     ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, l1_table_offset),
2701                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
2702     if (ret < 0) {
2703         goto fail_broken_refcounts;
2704     }
2705
2706     s->l1_table_offset = 3 * s->cluster_size;
2707
2708     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
2709     if (!new_reftable) {
2710         ret = -ENOMEM;
2711         goto fail_broken_refcounts;
2712     }
2713
2714     s->refcount_table_offset = s->cluster_size;
2715     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
2716
2717     g_free(s->refcount_table);
2718     s->refcount_table = new_reftable;
2719     new_reftable = NULL;
2720
2721     /* Now the in-memory refcount information again corresponds to the on-disk
2722      * information (reftable is empty and no refblocks (the refblock cache is
2723      * empty)); however, this means some clusters (e.g. the image header) are
2724      * referenced, but not refcounted, but the normal qcow2 code assumes that
2725      * the in-memory information is always correct */
2726
2727     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
2728
2729     /* Enter the first refblock into the reftable */
2730     rt_entry = cpu_to_be64(2 * s->cluster_size);
2731     ret = bdrv_pwrite_sync(bs->file->bs, s->cluster_size,
2732                            &rt_entry, sizeof(rt_entry));
2733     if (ret < 0) {
2734         goto fail_broken_refcounts;
2735     }
2736     s->refcount_table[0] = 2 * s->cluster_size;
2737
2738     s->free_cluster_index = 0;
2739     assert(3 + l1_clusters <= s->refcount_block_size);
2740     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
2741     if (offset < 0) {
2742         ret = offset;
2743         goto fail_broken_refcounts;
2744     } else if (offset > 0) {
2745         error_report("First cluster in emptied image is in use");
2746         abort();
2747     }
2748
2749     /* Now finally the in-memory information corresponds to the on-disk
2750      * structures and is correct */
2751     ret = qcow2_mark_clean(bs);
2752     if (ret < 0) {
2753         goto fail;
2754     }
2755
2756     ret = bdrv_truncate(bs->file->bs, (3 + l1_clusters) * s->cluster_size);
2757     if (ret < 0) {
2758         goto fail;
2759     }
2760
2761     return 0;
2762
2763 fail_broken_refcounts:
2764     /* The BDS is unusable at this point. If we wanted to make it usable, we
2765      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
2766      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
2767      * again. However, because the functions which could have caused this error
2768      * path to be taken are used by those functions as well, it's very likely
2769      * that that sequence will fail as well. Therefore, just eject the BDS. */
2770     bs->drv = NULL;
2771
2772 fail:
2773     g_free(new_reftable);
2774     return ret;
2775 }
2776
2777 static int qcow2_make_empty(BlockDriverState *bs)
2778 {
2779     BDRVQcow2State *s = bs->opaque;
2780     uint64_t start_sector;
2781     int sector_step = INT_MAX / BDRV_SECTOR_SIZE;
2782     int l1_clusters, ret = 0;
2783
2784     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2785
2786     if (s->qcow_version >= 3 && !s->snapshots &&
2787         3 + l1_clusters <= s->refcount_block_size) {
2788         /* The following function only works for qcow2 v3 images (it requires
2789          * the dirty flag) and only as long as there are no snapshots (because
2790          * it completely empties the image). Furthermore, the L1 table and three
2791          * additional clusters (image header, refcount table, one refcount
2792          * block) have to fit inside one refcount block. */
2793         return make_completely_empty(bs);
2794     }
2795
2796     /* This fallback code simply discards every active cluster; this is slow,
2797      * but works in all cases */
2798     for (start_sector = 0; start_sector < bs->total_sectors;
2799          start_sector += sector_step)
2800     {
2801         /* As this function is generally used after committing an external
2802          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
2803          * default action for this kind of discard is to pass the discard,
2804          * which will ideally result in an actually smaller image file, as
2805          * is probably desired. */
2806         ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE,
2807                                      MIN(sector_step,
2808                                          bs->total_sectors - start_sector),
2809                                      QCOW2_DISCARD_SNAPSHOT, true);
2810         if (ret < 0) {
2811             break;
2812         }
2813     }
2814
2815     return ret;
2816 }
2817
2818 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
2819 {
2820     BDRVQcow2State *s = bs->opaque;
2821     int ret;
2822
2823     qemu_co_mutex_lock(&s->lock);
2824     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2825     if (ret < 0) {
2826         qemu_co_mutex_unlock(&s->lock);
2827         return ret;
2828     }
2829
2830     if (qcow2_need_accurate_refcounts(s)) {
2831         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2832         if (ret < 0) {
2833             qemu_co_mutex_unlock(&s->lock);
2834             return ret;
2835         }
2836     }
2837     qemu_co_mutex_unlock(&s->lock);
2838
2839     return 0;
2840 }
2841
2842 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2843 {
2844     BDRVQcow2State *s = bs->opaque;
2845     bdi->unallocated_blocks_are_zero = true;
2846     bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
2847     bdi->cluster_size = s->cluster_size;
2848     bdi->vm_state_offset = qcow2_vm_state_offset(s);
2849     return 0;
2850 }
2851
2852 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
2853 {
2854     BDRVQcow2State *s = bs->opaque;
2855     ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
2856
2857     *spec_info = (ImageInfoSpecific){
2858         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
2859         .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1),
2860     };
2861     if (s->qcow_version == 2) {
2862         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
2863             .compat             = g_strdup("0.10"),
2864             .refcount_bits      = s->refcount_bits,
2865         };
2866     } else if (s->qcow_version == 3) {
2867         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
2868             .compat             = g_strdup("1.1"),
2869             .lazy_refcounts     = s->compatible_features &
2870                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
2871             .has_lazy_refcounts = true,
2872             .corrupt            = s->incompatible_features &
2873                                   QCOW2_INCOMPAT_CORRUPT,
2874             .has_corrupt        = true,
2875             .refcount_bits      = s->refcount_bits,
2876         };
2877     } else {
2878         /* if this assertion fails, this probably means a new version was
2879          * added without having it covered here */
2880         assert(false);
2881     }
2882
2883     return spec_info;
2884 }
2885
2886 #if 0
2887 static void dump_refcounts(BlockDriverState *bs)
2888 {
2889     BDRVQcow2State *s = bs->opaque;
2890     int64_t nb_clusters, k, k1, size;
2891     int refcount;
2892
2893     size = bdrv_getlength(bs->file->bs);
2894     nb_clusters = size_to_clusters(s, size);
2895     for(k = 0; k < nb_clusters;) {
2896         k1 = k;
2897         refcount = get_refcount(bs, k);
2898         k++;
2899         while (k < nb_clusters && get_refcount(bs, k) == refcount)
2900             k++;
2901         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
2902                k - k1);
2903     }
2904 }
2905 #endif
2906
2907 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2908                               int64_t pos)
2909 {
2910     BDRVQcow2State *s = bs->opaque;
2911     int64_t total_sectors = bs->total_sectors;
2912     bool zero_beyond_eof = bs->zero_beyond_eof;
2913     int ret;
2914
2915     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
2916     bs->zero_beyond_eof = false;
2917     ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov);
2918     bs->zero_beyond_eof = zero_beyond_eof;
2919
2920     /* bdrv_co_do_writev will have increased the total_sectors value to include
2921      * the VM state - the VM state is however not an actual part of the block
2922      * device, therefore, we need to restore the old value. */
2923     bs->total_sectors = total_sectors;
2924
2925     return ret;
2926 }
2927
2928 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2929                               int64_t pos, int size)
2930 {
2931     BDRVQcow2State *s = bs->opaque;
2932     bool zero_beyond_eof = bs->zero_beyond_eof;
2933     int ret;
2934
2935     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
2936     bs->zero_beyond_eof = false;
2937     ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
2938     bs->zero_beyond_eof = zero_beyond_eof;
2939
2940     return ret;
2941 }
2942
2943 /*
2944  * Downgrades an image's version. To achieve this, any incompatible features
2945  * have to be removed.
2946  */
2947 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
2948                            BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
2949 {
2950     BDRVQcow2State *s = bs->opaque;
2951     int current_version = s->qcow_version;
2952     int ret;
2953
2954     if (target_version == current_version) {
2955         return 0;
2956     } else if (target_version > current_version) {
2957         return -EINVAL;
2958     } else if (target_version != 2) {
2959         return -EINVAL;
2960     }
2961
2962     if (s->refcount_order != 4) {
2963         error_report("compat=0.10 requires refcount_bits=16");
2964         return -ENOTSUP;
2965     }
2966
2967     /* clear incompatible features */
2968     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
2969         ret = qcow2_mark_clean(bs);
2970         if (ret < 0) {
2971             return ret;
2972         }
2973     }
2974
2975     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2976      * the first place; if that happens nonetheless, returning -ENOTSUP is the
2977      * best thing to do anyway */
2978
2979     if (s->incompatible_features) {
2980         return -ENOTSUP;
2981     }
2982
2983     /* since we can ignore compatible features, we can set them to 0 as well */
2984     s->compatible_features = 0;
2985     /* if lazy refcounts have been used, they have already been fixed through
2986      * clearing the dirty flag */
2987
2988     /* clearing autoclear features is trivial */
2989     s->autoclear_features = 0;
2990
2991     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
2992     if (ret < 0) {
2993         return ret;
2994     }
2995
2996     s->qcow_version = target_version;
2997     ret = qcow2_update_header(bs);
2998     if (ret < 0) {
2999         s->qcow_version = current_version;
3000         return ret;
3001     }
3002     return 0;
3003 }
3004
3005 typedef enum Qcow2AmendOperation {
3006     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
3007      * statically initialized to so that the helper CB can discern the first
3008      * invocation from an operation change */
3009     QCOW2_NO_OPERATION = 0,
3010
3011     QCOW2_CHANGING_REFCOUNT_ORDER,
3012     QCOW2_DOWNGRADING,
3013 } Qcow2AmendOperation;
3014
3015 typedef struct Qcow2AmendHelperCBInfo {
3016     /* The code coordinating the amend operations should only modify
3017      * these four fields; the rest will be managed by the CB */
3018     BlockDriverAmendStatusCB *original_status_cb;
3019     void *original_cb_opaque;
3020
3021     Qcow2AmendOperation current_operation;
3022
3023     /* Total number of operations to perform (only set once) */
3024     int total_operations;
3025
3026     /* The following fields are managed by the CB */
3027
3028     /* Number of operations completed */
3029     int operations_completed;
3030
3031     /* Cumulative offset of all completed operations */
3032     int64_t offset_completed;
3033
3034     Qcow2AmendOperation last_operation;
3035     int64_t last_work_size;
3036 } Qcow2AmendHelperCBInfo;
3037
3038 static void qcow2_amend_helper_cb(BlockDriverState *bs,
3039                                   int64_t operation_offset,
3040                                   int64_t operation_work_size, void *opaque)
3041 {
3042     Qcow2AmendHelperCBInfo *info = opaque;
3043     int64_t current_work_size;
3044     int64_t projected_work_size;
3045
3046     if (info->current_operation != info->last_operation) {
3047         if (info->last_operation != QCOW2_NO_OPERATION) {
3048             info->offset_completed += info->last_work_size;
3049             info->operations_completed++;
3050         }
3051
3052         info->last_operation = info->current_operation;
3053     }
3054
3055     assert(info->total_operations > 0);
3056     assert(info->operations_completed < info->total_operations);
3057
3058     info->last_work_size = operation_work_size;
3059
3060     current_work_size = info->offset_completed + operation_work_size;
3061
3062     /* current_work_size is the total work size for (operations_completed + 1)
3063      * operations (which includes this one), so multiply it by the number of
3064      * operations not covered and divide it by the number of operations
3065      * covered to get a projection for the operations not covered */
3066     projected_work_size = current_work_size * (info->total_operations -
3067                                                info->operations_completed - 1)
3068                                             / (info->operations_completed + 1);
3069
3070     info->original_status_cb(bs, info->offset_completed + operation_offset,
3071                              current_work_size + projected_work_size,
3072                              info->original_cb_opaque);
3073 }
3074
3075 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
3076                                BlockDriverAmendStatusCB *status_cb,
3077                                void *cb_opaque)
3078 {
3079     BDRVQcow2State *s = bs->opaque;
3080     int old_version = s->qcow_version, new_version = old_version;
3081     uint64_t new_size = 0;
3082     const char *backing_file = NULL, *backing_format = NULL;
3083     bool lazy_refcounts = s->use_lazy_refcounts;
3084     const char *compat = NULL;
3085     uint64_t cluster_size = s->cluster_size;
3086     bool encrypt;
3087     int refcount_bits = s->refcount_bits;
3088     int ret;
3089     QemuOptDesc *desc = opts->list->desc;
3090     Qcow2AmendHelperCBInfo helper_cb_info;
3091
3092     while (desc && desc->name) {
3093         if (!qemu_opt_find(opts, desc->name)) {
3094             /* only change explicitly defined options */
3095             desc++;
3096             continue;
3097         }
3098
3099         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
3100             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
3101             if (!compat) {
3102                 /* preserve default */
3103             } else if (!strcmp(compat, "0.10")) {
3104                 new_version = 2;
3105             } else if (!strcmp(compat, "1.1")) {
3106                 new_version = 3;
3107             } else {
3108                 error_report("Unknown compatibility level %s", compat);
3109                 return -EINVAL;
3110             }
3111         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
3112             error_report("Cannot change preallocation mode");
3113             return -ENOTSUP;
3114         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
3115             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
3116         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
3117             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
3118         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
3119             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
3120         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
3121             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
3122                                         !!s->cipher);
3123
3124             if (encrypt != !!s->cipher) {
3125                 error_report("Changing the encryption flag is not supported");
3126                 return -ENOTSUP;
3127             }
3128         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
3129             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
3130                                              cluster_size);
3131             if (cluster_size != s->cluster_size) {
3132                 error_report("Changing the cluster size is not supported");
3133                 return -ENOTSUP;
3134             }
3135         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
3136             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
3137                                                lazy_refcounts);
3138         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
3139             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
3140                                                 refcount_bits);
3141
3142             if (refcount_bits <= 0 || refcount_bits > 64 ||
3143                 !is_power_of_2(refcount_bits))
3144             {
3145                 error_report("Refcount width must be a power of two and may "
3146                              "not exceed 64 bits");
3147                 return -EINVAL;
3148             }
3149         } else {
3150             /* if this point is reached, this probably means a new option was
3151              * added without having it covered here */
3152             abort();
3153         }
3154
3155         desc++;
3156     }
3157
3158     helper_cb_info = (Qcow2AmendHelperCBInfo){
3159         .original_status_cb = status_cb,
3160         .original_cb_opaque = cb_opaque,
3161         .total_operations = (new_version < old_version)
3162                           + (s->refcount_bits != refcount_bits)
3163     };
3164
3165     /* Upgrade first (some features may require compat=1.1) */
3166     if (new_version > old_version) {
3167         s->qcow_version = new_version;
3168         ret = qcow2_update_header(bs);
3169         if (ret < 0) {
3170             s->qcow_version = old_version;
3171             return ret;
3172         }
3173     }
3174
3175     if (s->refcount_bits != refcount_bits) {
3176         int refcount_order = ctz32(refcount_bits);
3177         Error *local_error = NULL;
3178
3179         if (new_version < 3 && refcount_bits != 16) {
3180             error_report("Different refcount widths than 16 bits require "
3181                          "compatibility level 1.1 or above (use compat=1.1 or "
3182                          "greater)");
3183             return -EINVAL;
3184         }
3185
3186         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
3187         ret = qcow2_change_refcount_order(bs, refcount_order,
3188                                           &qcow2_amend_helper_cb,
3189                                           &helper_cb_info, &local_error);
3190         if (ret < 0) {
3191             error_report_err(local_error);
3192             return ret;
3193         }
3194     }
3195
3196     if (backing_file || backing_format) {
3197         ret = qcow2_change_backing_file(bs,
3198                     backing_file ?: s->image_backing_file,
3199                     backing_format ?: s->image_backing_format);
3200         if (ret < 0) {
3201             return ret;
3202         }
3203     }
3204
3205     if (s->use_lazy_refcounts != lazy_refcounts) {
3206         if (lazy_refcounts) {
3207             if (new_version < 3) {
3208                 error_report("Lazy refcounts only supported with compatibility "
3209                              "level 1.1 and above (use compat=1.1 or greater)");
3210                 return -EINVAL;
3211             }
3212             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3213             ret = qcow2_update_header(bs);
3214             if (ret < 0) {
3215                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3216                 return ret;
3217             }
3218             s->use_lazy_refcounts = true;
3219         } else {
3220             /* make image clean first */
3221             ret = qcow2_mark_clean(bs);
3222             if (ret < 0) {
3223                 return ret;
3224             }
3225             /* now disallow lazy refcounts */
3226             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3227             ret = qcow2_update_header(bs);
3228             if (ret < 0) {
3229                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3230                 return ret;
3231             }
3232             s->use_lazy_refcounts = false;
3233         }
3234     }
3235
3236     if (new_size) {
3237         ret = bdrv_truncate(bs, new_size);
3238         if (ret < 0) {
3239             return ret;
3240         }
3241     }
3242
3243     /* Downgrade last (so unsupported features can be removed before) */
3244     if (new_version < old_version) {
3245         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
3246         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
3247                               &helper_cb_info);
3248         if (ret < 0) {
3249             return ret;
3250         }
3251     }
3252
3253     return 0;
3254 }
3255
3256 /*
3257  * If offset or size are negative, respectively, they will not be included in
3258  * the BLOCK_IMAGE_CORRUPTED event emitted.
3259  * fatal will be ignored for read-only BDS; corruptions found there will always
3260  * be considered non-fatal.
3261  */
3262 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
3263                              int64_t size, const char *message_format, ...)
3264 {
3265     BDRVQcow2State *s = bs->opaque;
3266     const char *node_name;
3267     char *message;
3268     va_list ap;
3269
3270     fatal = fatal && !bs->read_only;
3271
3272     if (s->signaled_corruption &&
3273         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
3274     {
3275         return;
3276     }
3277
3278     va_start(ap, message_format);
3279     message = g_strdup_vprintf(message_format, ap);
3280     va_end(ap);
3281
3282     if (fatal) {
3283         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
3284                 "corruption events will be suppressed\n", message);
3285     } else {
3286         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
3287                 "corruption events will be suppressed\n", message);
3288     }
3289
3290     node_name = bdrv_get_node_name(bs);
3291     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
3292                                           *node_name != '\0', node_name,
3293                                           message, offset >= 0, offset,
3294                                           size >= 0, size,
3295                                           fatal, &error_abort);
3296     g_free(message);
3297
3298     if (fatal) {
3299         qcow2_mark_corrupt(bs);
3300         bs->drv = NULL; /* make BDS unusable */
3301     }
3302
3303     s->signaled_corruption = true;
3304 }
3305
3306 static QemuOptsList qcow2_create_opts = {
3307     .name = "qcow2-create-opts",
3308     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
3309     .desc = {
3310         {
3311             .name = BLOCK_OPT_SIZE,
3312             .type = QEMU_OPT_SIZE,
3313             .help = "Virtual disk size"
3314         },
3315         {
3316             .name = BLOCK_OPT_COMPAT_LEVEL,
3317             .type = QEMU_OPT_STRING,
3318             .help = "Compatibility level (0.10 or 1.1)"
3319         },
3320         {
3321             .name = BLOCK_OPT_BACKING_FILE,
3322             .type = QEMU_OPT_STRING,
3323             .help = "File name of a base image"
3324         },
3325         {
3326             .name = BLOCK_OPT_BACKING_FMT,
3327             .type = QEMU_OPT_STRING,
3328             .help = "Image format of the base image"
3329         },
3330         {
3331             .name = BLOCK_OPT_ENCRYPT,
3332             .type = QEMU_OPT_BOOL,
3333             .help = "Encrypt the image",
3334             .def_value_str = "off"
3335         },
3336         {
3337             .name = BLOCK_OPT_CLUSTER_SIZE,
3338             .type = QEMU_OPT_SIZE,
3339             .help = "qcow2 cluster size",
3340             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
3341         },
3342         {
3343             .name = BLOCK_OPT_PREALLOC,
3344             .type = QEMU_OPT_STRING,
3345             .help = "Preallocation mode (allowed values: off, metadata, "
3346                     "falloc, full)"
3347         },
3348         {
3349             .name = BLOCK_OPT_LAZY_REFCOUNTS,
3350             .type = QEMU_OPT_BOOL,
3351             .help = "Postpone refcount updates",
3352             .def_value_str = "off"
3353         },
3354         {
3355             .name = BLOCK_OPT_REFCOUNT_BITS,
3356             .type = QEMU_OPT_NUMBER,
3357             .help = "Width of a reference count entry in bits",
3358             .def_value_str = "16"
3359         },
3360         { /* end of list */ }
3361     }
3362 };
3363
3364 BlockDriver bdrv_qcow2 = {
3365     .format_name        = "qcow2",
3366     .instance_size      = sizeof(BDRVQcow2State),
3367     .bdrv_probe         = qcow2_probe,
3368     .bdrv_open          = qcow2_open,
3369     .bdrv_close         = qcow2_close,
3370     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
3371     .bdrv_reopen_commit   = qcow2_reopen_commit,
3372     .bdrv_reopen_abort    = qcow2_reopen_abort,
3373     .bdrv_join_options    = qcow2_join_options,
3374     .bdrv_create        = qcow2_create,
3375     .bdrv_has_zero_init = bdrv_has_zero_init_1,
3376     .bdrv_co_get_block_status = qcow2_co_get_block_status,
3377     .bdrv_set_key       = qcow2_set_key,
3378
3379     .bdrv_co_readv          = qcow2_co_readv,
3380     .bdrv_co_writev         = qcow2_co_writev,
3381     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
3382
3383     .bdrv_co_write_zeroes   = qcow2_co_write_zeroes,
3384     .bdrv_co_discard        = qcow2_co_discard,
3385     .bdrv_truncate          = qcow2_truncate,
3386     .bdrv_write_compressed  = qcow2_write_compressed,
3387     .bdrv_make_empty        = qcow2_make_empty,
3388
3389     .bdrv_snapshot_create   = qcow2_snapshot_create,
3390     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
3391     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
3392     .bdrv_snapshot_list     = qcow2_snapshot_list,
3393     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
3394     .bdrv_get_info          = qcow2_get_info,
3395     .bdrv_get_specific_info = qcow2_get_specific_info,
3396
3397     .bdrv_save_vmstate    = qcow2_save_vmstate,
3398     .bdrv_load_vmstate    = qcow2_load_vmstate,
3399
3400     .supports_backing           = true,
3401     .bdrv_change_backing_file   = qcow2_change_backing_file,
3402
3403     .bdrv_refresh_limits        = qcow2_refresh_limits,
3404     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
3405     .bdrv_inactivate            = qcow2_inactivate,
3406
3407     .create_opts         = &qcow2_create_opts,
3408     .bdrv_check          = qcow2_check,
3409     .bdrv_amend_options  = qcow2_amend_options,
3410
3411     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
3412     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
3413 };
3414
3415 static void bdrv_qcow2_init(void)
3416 {
3417     bdrv_register(&bdrv_qcow2);
3418 }
3419
3420 block_init(bdrv_qcow2_init);