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-common.h"
  25 #include "block/block_int.h"
  26 #include "qemu/module.h"
  27 #include <zlib.h>
  28 #include "qemu/aes.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
  39 /*
  40   Differences with QCOW:
  41
  42   - Support for multiple incremental snapshots.
  43   - Memory management by reference counts.
  44   - Clusters which have a reference count of one have the bit
  45     QCOW_OFLAG_COPIED to optimize write performance.
  46   - Size of compressed clusters is stored in sectors to reduce bit usage
  47     in the cluster offsets.
  48   - Support for storing additional data (such as the VM state) in the
  49     snapshots.
  50   - If a backing store is used, the cluster size is not constrained
  51     (could be backported to QCOW).
  52   - L2 tables have always a size of one cluster.
  53 */
  54
  55
  56 typedef struct {
  57     uint32_t magic;
  58     uint32_t len;
  59 } QEMU_PACKED QCowExtension;
  60
  61 #define  QCOW2_EXT_MAGIC_END 0
  62 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
  63 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
  64
  65 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
  66 {
  67     const QCowHeader *cow_header = (const void *)buf;
  68
  69     if (buf_size >= sizeof(QCowHeader) &&
  70         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
  71         be32_to_cpu(cow_header->version) >= 2)
  72         return 100;
  73     else
  74         return 0;
  75 }
  76
  77
  78 /*
  79  * read qcow2 extension and fill bs
  80  * start reading from start_offset
  81  * finish reading upon magic of value 0 or when end_offset reached
  82  * unknown magic is skipped (future extension this version knows nothing about)
  83  * return 0 upon success, non-0 otherwise
  84  */
  85 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
  86                                  uint64_t end_offset, void **p_feature_table,
  87                                  Error **errp)
  88 {
  89     BDRVQcowState *s = bs->opaque;
  90     QCowExtension ext;
  91     uint64_t offset;
  92     int ret;
  93
  94 #ifdef DEBUG_EXT
  95     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
  96 #endif
  97     offset = start_offset;
  98     while (offset < end_offset) {
  99
 100 #ifdef DEBUG_EXT
 101         /* Sanity check */
 102         if (offset > s->cluster_size)
 103             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
 104
 105         printf("attempting to read extended header in offset %lu\n", offset);
 106 #endif
 107
 108         ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
 109         if (ret < 0) {
 110             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
 111                              "pread fail from offset %" PRIu64, offset);
 112             return 1;
 113         }
 114         be32_to_cpus(&ext.magic);
 115         be32_to_cpus(&ext.len);
 116         offset += sizeof(ext);
 117 #ifdef DEBUG_EXT
 118         printf("ext.magic = 0x%x\n", ext.magic);
 119 #endif
 120         if (offset > end_offset || ext.len > end_offset - offset) {
 121             error_setg(errp, "Header extension too large");
 122             return -EINVAL;
 123         }
 124
 125         switch (ext.magic) {
 126         case QCOW2_EXT_MAGIC_END:
 127             return 0;
 128
 129         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
 130             if (ext.len >= sizeof(bs->backing_format)) {
 131                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
 132                            " too large (>=%zu)", ext.len,
 133                            sizeof(bs->backing_format));
 134                 return 2;
 135             }
 136             ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len);
 137             if (ret < 0) {
 138                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
 139                                  "Could not read format name");
 140                 return 3;
 141             }
 142             bs->backing_format[ext.len] = '\0';
 143 #ifdef DEBUG_EXT
 144             printf("Qcow2: Got format extension %s\n", bs->backing_format);
 145 #endif
 146             break;
 147
 148         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
 149             if (p_feature_table != NULL) {
 150                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
 151                 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
 152                 if (ret < 0) {
 153                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
 154                                      "Could not read table");
 155                     return ret;
 156                 }
 157
 158                 *p_feature_table = feature_table;
 159             }
 160             break;
 161
 162         default:
 163             /* unknown magic - save it in case we need to rewrite the header */
 164             {
 165                 Qcow2UnknownHeaderExtension *uext;
 166
 167                 uext = g_malloc0(sizeof(*uext)  + ext.len);
 168                 uext->magic = ext.magic;
 169                 uext->len = ext.len;
 170                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
 171
 172                 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
 173                 if (ret < 0) {
 174                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
 175                                      "Could not read data");
 176                     return ret;
 177                 }
 178             }
 179             break;
 180         }
 181
 182         offset += ((ext.len + 7) & ~7);
 183     }
 184
 185     return 0;
 186 }
 187
 188 static void cleanup_unknown_header_ext(BlockDriverState *bs)
 189 {
 190     BDRVQcowState *s = bs->opaque;
 191     Qcow2UnknownHeaderExtension *uext, *next;
 192
 193     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
 194         QLIST_REMOVE(uext, next);
 195         g_free(uext);
 196     }
 197 }
 198
 199 static void GCC_FMT_ATTR(3, 4) report_unsupported(BlockDriverState *bs,
 200     Error **errp, const char *fmt, ...)
 201 {
 202     char msg[64];
 203     va_list ap;
 204
 205     va_start(ap, fmt);
 206     vsnprintf(msg, sizeof(msg), fmt, ap);
 207     va_end(ap);
 208
 209     error_set(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
 210               bdrv_get_device_name(bs), "qcow2", msg);
 211 }
 212
 213 static void report_unsupported_feature(BlockDriverState *bs,
 214     Error **errp, Qcow2Feature *table, uint64_t mask)
 215 {
 216     char *features = g_strdup("");
 217     char *old;
 218
 219     while (table && table->name[0] != '\0') {
 220         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
 221             if (mask & (1ULL << table->bit)) {
 222                 old = features;
 223                 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
 224                                            table->name);
 225                 g_free(old);
 226                 mask &= ~(1ULL << table->bit);
 227             }
 228         }
 229         table++;
 230     }
 231
 232     if (mask) {
 233         old = features;
 234         features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
 235                                    old, *old ? ", " : "", mask);
 236         g_free(old);
 237     }
 238
 239     report_unsupported(bs, errp, "%s", features);
 240     g_free(features);
 241 }
 242
 243 /*
 244  * Sets the dirty bit and flushes afterwards if necessary.
 245  *
 246  * The incompatible_features bit is only set if the image file header was
 247  * updated successfully.  Therefore it is not required to check the return
 248  * value of this function.
 249  */
 250 int qcow2_mark_dirty(BlockDriverState *bs)
 251 {
 252     BDRVQcowState *s = bs->opaque;
 253     uint64_t val;
 254     int ret;
 255
 256     assert(s->qcow_version >= 3);
 257
 258     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
 259         return 0; /* already dirty */
 260     }
 261
 262     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
 263     ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
 264                       &val, sizeof(val));
 265     if (ret < 0) {
 266         return ret;
 267     }
 268     ret = bdrv_flush(bs->file);
 269     if (ret < 0) {
 270         return ret;
 271     }
 272
 273     /* Only treat image as dirty if the header was updated successfully */
 274     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
 275     return 0;
 276 }
 277
 278 /*
 279  * Clears the dirty bit and flushes before if necessary.  Only call this
 280  * function when there are no pending requests, it does not guard against
 281  * concurrent requests dirtying the image.
 282  */
 283 static int qcow2_mark_clean(BlockDriverState *bs)
 284 {
 285     BDRVQcowState *s = bs->opaque;
 286
 287     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
 288         int ret;
 289
 290         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
 291
 292         ret = bdrv_flush(bs);
 293         if (ret < 0) {
 294             return ret;
 295         }
 296
 297         return qcow2_update_header(bs);
 298     }
 299     return 0;
 300 }
 301
 302 /*
 303  * Marks the image as corrupt.
 304  */
 305 int qcow2_mark_corrupt(BlockDriverState *bs)
 306 {
 307     BDRVQcowState *s = bs->opaque;
 308
 309     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
 310     return qcow2_update_header(bs);
 311 }
 312
 313 /*
 314  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
 315  * before if necessary.
 316  */
 317 int qcow2_mark_consistent(BlockDriverState *bs)
 318 {
 319     BDRVQcowState *s = bs->opaque;
 320
 321     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
 322         int ret = bdrv_flush(bs);
 323         if (ret < 0) {
 324             return ret;
 325         }
 326
 327         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
 328         return qcow2_update_header(bs);
 329     }
 330     return 0;
 331 }
 332
 333 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
 334                        BdrvCheckMode fix)
 335 {
 336     int ret = qcow2_check_refcounts(bs, result, fix);
 337     if (ret < 0) {
 338         return ret;
 339     }
 340
 341     if (fix && result->check_errors == 0 && result->corruptions == 0) {
 342         ret = qcow2_mark_clean(bs);
 343         if (ret < 0) {
 344             return ret;
 345         }
 346         return qcow2_mark_consistent(bs);
 347     }
 348     return ret;
 349 }
 350
 351 static int validate_table_offset(BlockDriverState *bs, uint64_t offset,
 352                                  uint64_t entries, size_t entry_len)
 353 {
 354     BDRVQcowState *s = bs->opaque;
 355     uint64_t size;
 356
 357     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
 358      * because values will be passed to qemu functions taking int64_t. */
 359     if (entries > INT64_MAX / entry_len) {
 360         return -EINVAL;
 361     }
 362
 363     size = entries * entry_len;
 364
 365     if (INT64_MAX - size < offset) {
 366         return -EINVAL;
 367     }
 368
 369     /* Tables must be cluster aligned */
 370     if (offset & (s->cluster_size - 1)) {
 371         return -EINVAL;
 372     }
 373
 374     return 0;
 375 }
 376
 377 static QemuOptsList qcow2_runtime_opts = {
 378     .name = "qcow2",
 379     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
 380     .desc = {
 381         {
 382             .name = QCOW2_OPT_LAZY_REFCOUNTS,
 383             .type = QEMU_OPT_BOOL,
 384             .help = "Postpone refcount updates",
 385         },
 386         {
 387             .name = QCOW2_OPT_DISCARD_REQUEST,
 388             .type = QEMU_OPT_BOOL,
 389             .help = "Pass guest discard requests to the layer below",
 390         },
 391         {
 392             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
 393             .type = QEMU_OPT_BOOL,
 394             .help = "Generate discard requests when snapshot related space "
 395                     "is freed",
 396         },
 397         {
 398             .name = QCOW2_OPT_DISCARD_OTHER,
 399             .type = QEMU_OPT_BOOL,
 400             .help = "Generate discard requests when other clusters are freed",
 401         },
 402         {
 403             .name = QCOW2_OPT_OVERLAP,
 404             .type = QEMU_OPT_STRING,
 405             .help = "Selects which overlap checks to perform from a range of "
 406                     "templates (none, constant, cached, all)",
 407         },
 408         {
 409             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
 410             .type = QEMU_OPT_STRING,
 411             .help = "Selects which overlap checks to perform from a range of "
 412                     "templates (none, constant, cached, all)",
 413         },
 414         {
 415             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
 416             .type = QEMU_OPT_BOOL,
 417             .help = "Check for unintended writes into the main qcow2 header",
 418         },
 419         {
 420             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
 421             .type = QEMU_OPT_BOOL,
 422             .help = "Check for unintended writes into the active L1 table",
 423         },
 424         {
 425             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
 426             .type = QEMU_OPT_BOOL,
 427             .help = "Check for unintended writes into an active L2 table",
 428         },
 429         {
 430             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
 431             .type = QEMU_OPT_BOOL,
 432             .help = "Check for unintended writes into the refcount table",
 433         },
 434         {
 435             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
 436             .type = QEMU_OPT_BOOL,
 437             .help = "Check for unintended writes into a refcount block",
 438         },
 439         {
 440             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
 441             .type = QEMU_OPT_BOOL,
 442             .help = "Check for unintended writes into the snapshot table",
 443         },
 444         {
 445             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
 446             .type = QEMU_OPT_BOOL,
 447             .help = "Check for unintended writes into an inactive L1 table",
 448         },
 449         {
 450             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
 451             .type = QEMU_OPT_BOOL,
 452             .help = "Check for unintended writes into an inactive L2 table",
 453         },
 454         {
 455             .name = QCOW2_OPT_CACHE_SIZE,
 456             .type = QEMU_OPT_SIZE,
 457             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
 458                     "cache size",
 459         },
 460         {
 461             .name = QCOW2_OPT_L2_CACHE_SIZE,
 462             .type = QEMU_OPT_SIZE,
 463             .help = "Maximum L2 table cache size",
 464         },
 465         {
 466             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
 467             .type = QEMU_OPT_SIZE,
 468             .help = "Maximum refcount block cache size",
 469         },
 470         { /* end of list */ }
 471     },
 472 };
 473
 474 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
 475     [QCOW2_OL_MAIN_HEADER_BITNR]    = QCOW2_OPT_OVERLAP_MAIN_HEADER,
 476     [QCOW2_OL_ACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L1,
 477     [QCOW2_OL_ACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L2,
 478     [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
 479     [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
 480     [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
 481     [QCOW2_OL_INACTIVE_L1_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L1,
 482     [QCOW2_OL_INACTIVE_L2_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L2,
 483 };
 484
 485 static void read_cache_sizes(QemuOpts *opts, uint64_t *l2_cache_size,
 486                              uint64_t *refcount_cache_size, Error **errp)
 487 {
 488     uint64_t combined_cache_size;
 489     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
 490
 491     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
 492     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
 493     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
 494
 495     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
 496     *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0);
 497     *refcount_cache_size = qemu_opt_get_size(opts,
 498                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
 499
 500     if (combined_cache_size_set) {
 501         if (l2_cache_size_set && refcount_cache_size_set) {
 502             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
 503                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
 504                        "the same time");
 505             return;
 506         } else if (*l2_cache_size > combined_cache_size) {
 507             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
 508                        QCOW2_OPT_CACHE_SIZE);
 509             return;
 510         } else if (*refcount_cache_size > combined_cache_size) {
 511             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
 512                        QCOW2_OPT_CACHE_SIZE);
 513             return;
 514         }
 515
 516         if (l2_cache_size_set) {
 517             *refcount_cache_size = combined_cache_size - *l2_cache_size;
 518         } else if (refcount_cache_size_set) {
 519             *l2_cache_size = combined_cache_size - *refcount_cache_size;
 520         } else {
 521             *refcount_cache_size = combined_cache_size
 522                                  / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1);
 523             *l2_cache_size = combined_cache_size - *refcount_cache_size;
 524         }
 525     } else {
 526         if (!l2_cache_size_set && !refcount_cache_size_set) {
 527             *l2_cache_size = DEFAULT_L2_CACHE_BYTE_SIZE;
 528             *refcount_cache_size = *l2_cache_size
 529                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 530         } else if (!l2_cache_size_set) {
 531             *l2_cache_size = *refcount_cache_size
 532                            * DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 533         } else if (!refcount_cache_size_set) {
 534             *refcount_cache_size = *l2_cache_size
 535                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
 536         }
 537     }
 538 }
 539
 540 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
 541                       Error **errp)
 542 {
 543     BDRVQcowState *s = bs->opaque;
 544     unsigned int len, i;
 545     int ret = 0;
 546     QCowHeader header;
 547     QemuOpts *opts = NULL;
 548     Error *local_err = NULL;
 549     uint64_t ext_end;
 550     uint64_t l1_vm_state_index;
 551     const char *opt_overlap_check, *opt_overlap_check_template;
 552     int overlap_check_template = 0;
 553     uint64_t l2_cache_size, refcount_cache_size;
 554
 555     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
 556     if (ret < 0) {
 557         error_setg_errno(errp, -ret, "Could not read qcow2 header");
 558         goto fail;
 559     }
 560     be32_to_cpus(&header.magic);
 561     be32_to_cpus(&header.version);
 562     be64_to_cpus(&header.backing_file_offset);
 563     be32_to_cpus(&header.backing_file_size);
 564     be64_to_cpus(&header.size);
 565     be32_to_cpus(&header.cluster_bits);
 566     be32_to_cpus(&header.crypt_method);
 567     be64_to_cpus(&header.l1_table_offset);
 568     be32_to_cpus(&header.l1_size);
 569     be64_to_cpus(&header.refcount_table_offset);
 570     be32_to_cpus(&header.refcount_table_clusters);
 571     be64_to_cpus(&header.snapshots_offset);
 572     be32_to_cpus(&header.nb_snapshots);
 573
 574     if (header.magic != QCOW_MAGIC) {
 575         error_setg(errp, "Image is not in qcow2 format");
 576         ret = -EINVAL;
 577         goto fail;
 578     }
 579     if (header.version < 2 || header.version > 3) {
 580         report_unsupported(bs, errp, "QCOW version %" PRIu32, header.version);
 581         ret = -ENOTSUP;
 582         goto fail;
 583     }
 584
 585     s->qcow_version = header.version;
 586
 587     /* Initialise cluster size */
 588     if (header.cluster_bits < MIN_CLUSTER_BITS ||
 589         header.cluster_bits > MAX_CLUSTER_BITS) {
 590         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
 591                    header.cluster_bits);
 592         ret = -EINVAL;
 593         goto fail;
 594     }
 595
 596     s->cluster_bits = header.cluster_bits;
 597     s->cluster_size = 1 << s->cluster_bits;
 598     s->cluster_sectors = 1 << (s->cluster_bits - 9);
 599
 600     /* Initialise version 3 header fields */
 601     if (header.version == 2) {
 602         header.incompatible_features    = 0;
 603         header.compatible_features      = 0;
 604         header.autoclear_features       = 0;
 605         header.refcount_order           = 4;
 606         header.header_length            = 72;
 607     } else {
 608         be64_to_cpus(&header.incompatible_features);
 609         be64_to_cpus(&header.compatible_features);
 610         be64_to_cpus(&header.autoclear_features);
 611         be32_to_cpus(&header.refcount_order);
 612         be32_to_cpus(&header.header_length);
 613
 614         if (header.header_length < 104) {
 615             error_setg(errp, "qcow2 header too short");
 616             ret = -EINVAL;
 617             goto fail;
 618         }
 619     }
 620
 621     if (header.header_length > s->cluster_size) {
 622         error_setg(errp, "qcow2 header exceeds cluster size");
 623         ret = -EINVAL;
 624         goto fail;
 625     }
 626
 627     if (header.header_length > sizeof(header)) {
 628         s->unknown_header_fields_size = header.header_length - sizeof(header);
 629         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
 630         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
 631                          s->unknown_header_fields_size);
 632         if (ret < 0) {
 633             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
 634                              "fields");
 635             goto fail;
 636         }
 637     }
 638
 639     if (header.backing_file_offset > s->cluster_size) {
 640         error_setg(errp, "Invalid backing file offset");
 641         ret = -EINVAL;
 642         goto fail;
 643     }
 644
 645     if (header.backing_file_offset) {
 646         ext_end = header.backing_file_offset;
 647     } else {
 648         ext_end = 1 << header.cluster_bits;
 649     }
 650
 651     /* Handle feature bits */
 652     s->incompatible_features    = header.incompatible_features;
 653     s->compatible_features      = header.compatible_features;
 654     s->autoclear_features       = header.autoclear_features;
 655
 656     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
 657         void *feature_table = NULL;
 658         qcow2_read_extensions(bs, header.header_length, ext_end,
 659                               &feature_table, NULL);
 660         report_unsupported_feature(bs, errp, feature_table,
 661                                    s->incompatible_features &
 662                                    ~QCOW2_INCOMPAT_MASK);
 663         ret = -ENOTSUP;
 664         g_free(feature_table);
 665         goto fail;
 666     }
 667
 668     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
 669         /* Corrupt images may not be written to unless they are being repaired
 670          */
 671         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
 672             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
 673                        "read/write");
 674             ret = -EACCES;
 675             goto fail;
 676         }
 677     }
 678
 679     /* Check support for various header values */
 680     if (header.refcount_order != 4) {
 681         report_unsupported(bs, errp, "%d bit reference counts",
 682                            1 << header.refcount_order);
 683         ret = -ENOTSUP;
 684         goto fail;
 685     }
 686     s->refcount_order = header.refcount_order;
 687
 688     if (header.crypt_method > QCOW_CRYPT_AES) {
 689         error_setg(errp, "Unsupported encryption method: %" PRIu32,
 690                    header.crypt_method);
 691         ret = -EINVAL;
 692         goto fail;
 693     }
 694     s->crypt_method_header = header.crypt_method;
 695     if (s->crypt_method_header) {
 696         bs->encrypted = 1;
 697     }
 698
 699     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
 700     s->l2_size = 1 << s->l2_bits;
 701     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
 702     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
 703     s->refcount_block_size = 1 << s->refcount_block_bits;
 704     bs->total_sectors = header.size / 512;
 705     s->csize_shift = (62 - (s->cluster_bits - 8));
 706     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
 707     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
 708
 709     s->refcount_table_offset = header.refcount_table_offset;
 710     s->refcount_table_size =
 711         header.refcount_table_clusters << (s->cluster_bits - 3);
 712
 713     if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
 714         error_setg(errp, "Reference count table too large");
 715         ret = -EINVAL;
 716         goto fail;
 717     }
 718
 719     ret = validate_table_offset(bs, s->refcount_table_offset,
 720                                 s->refcount_table_size, sizeof(uint64_t));
 721     if (ret < 0) {
 722         error_setg(errp, "Invalid reference count table offset");
 723         goto fail;
 724     }
 725
 726     /* Snapshot table offset/length */
 727     if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
 728         error_setg(errp, "Too many snapshots");
 729         ret = -EINVAL;
 730         goto fail;
 731     }
 732
 733     ret = validate_table_offset(bs, header.snapshots_offset,
 734                                 header.nb_snapshots,
 735                                 sizeof(QCowSnapshotHeader));
 736     if (ret < 0) {
 737         error_setg(errp, "Invalid snapshot table offset");
 738         goto fail;
 739     }
 740
 741     /* read the level 1 table */
 742     if (header.l1_size > QCOW_MAX_L1_SIZE) {
 743         error_setg(errp, "Active L1 table too large");
 744         ret = -EFBIG;
 745         goto fail;
 746     }
 747     s->l1_size = header.l1_size;
 748
 749     l1_vm_state_index = size_to_l1(s, header.size);
 750     if (l1_vm_state_index > INT_MAX) {
 751         error_setg(errp, "Image is too big");
 752         ret = -EFBIG;
 753         goto fail;
 754     }
 755     s->l1_vm_state_index = l1_vm_state_index;
 756
 757     /* the L1 table must contain at least enough entries to put
 758        header.size bytes */
 759     if (s->l1_size < s->l1_vm_state_index) {
 760         error_setg(errp, "L1 table is too small");
 761         ret = -EINVAL;
 762         goto fail;
 763     }
 764
 765     ret = validate_table_offset(bs, header.l1_table_offset,
 766                                 header.l1_size, sizeof(uint64_t));
 767     if (ret < 0) {
 768         error_setg(errp, "Invalid L1 table offset");
 769         goto fail;
 770     }
 771     s->l1_table_offset = header.l1_table_offset;
 772
 773
 774     if (s->l1_size > 0) {
 775         s->l1_table = qemu_try_blockalign(bs->file,
 776             align_offset(s->l1_size * sizeof(uint64_t), 512));
 777         if (s->l1_table == NULL) {
 778             error_setg(errp, "Could not allocate L1 table");
 779             ret = -ENOMEM;
 780             goto fail;
 781         }
 782         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
 783                          s->l1_size * sizeof(uint64_t));
 784         if (ret < 0) {
 785             error_setg_errno(errp, -ret, "Could not read L1 table");
 786             goto fail;
 787         }
 788         for(i = 0;i < s->l1_size; i++) {
 789             be64_to_cpus(&s->l1_table[i]);
 790         }
 791     }
 792
 793     /* get L2 table/refcount block cache size from command line options */
 794     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
 795     qemu_opts_absorb_qdict(opts, options, &local_err);
 796     if (local_err) {
 797         error_propagate(errp, local_err);
 798         ret = -EINVAL;
 799         goto fail;
 800     }
 801
 802     read_cache_sizes(opts, &l2_cache_size, &refcount_cache_size, &local_err);
 803     if (local_err) {
 804         error_propagate(errp, local_err);
 805         ret = -EINVAL;
 806         goto fail;
 807     }
 808
 809     l2_cache_size /= s->cluster_size;
 810     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
 811         l2_cache_size = MIN_L2_CACHE_SIZE;
 812     }
 813     if (l2_cache_size > INT_MAX) {
 814         error_setg(errp, "L2 cache size too big");
 815         ret = -EINVAL;
 816         goto fail;
 817     }
 818
 819     refcount_cache_size /= s->cluster_size;
 820     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
 821         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
 822     }
 823     if (refcount_cache_size > INT_MAX) {
 824         error_setg(errp, "Refcount cache size too big");
 825         ret = -EINVAL;
 826         goto fail;
 827     }
 828
 829     /* alloc L2 table/refcount block cache */
 830     s->l2_table_cache = qcow2_cache_create(bs, l2_cache_size);
 831     s->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size);
 832     if (s->l2_table_cache == NULL || s->refcount_block_cache == NULL) {
 833         error_setg(errp, "Could not allocate metadata caches");
 834         ret = -ENOMEM;
 835         goto fail;
 836     }
 837
 838     s->cluster_cache = g_malloc(s->cluster_size);
 839     /* one more sector for decompressed data alignment */
 840     s->cluster_data = qemu_try_blockalign(bs->file, QCOW_MAX_CRYPT_CLUSTERS
 841                                                     * s->cluster_size + 512);
 842     if (s->cluster_data == NULL) {
 843         error_setg(errp, "Could not allocate temporary cluster buffer");
 844         ret = -ENOMEM;
 845         goto fail;
 846     }
 847
 848     s->cluster_cache_offset = -1;
 849     s->flags = flags;
 850
 851     ret = qcow2_refcount_init(bs);
 852     if (ret != 0) {
 853         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
 854         goto fail;
 855     }
 856
 857     QLIST_INIT(&s->cluster_allocs);
 858     QTAILQ_INIT(&s->discards);
 859
 860     /* read qcow2 extensions */
 861     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
 862         &local_err)) {
 863         error_propagate(errp, local_err);
 864         ret = -EINVAL;
 865         goto fail;
 866     }
 867
 868     /* read the backing file name */
 869     if (header.backing_file_offset != 0) {
 870         len = header.backing_file_size;
 871         if (len > MIN(1023, s->cluster_size - header.backing_file_offset)) {
 872             error_setg(errp, "Backing file name too long");
 873             ret = -EINVAL;
 874             goto fail;
 875         }
 876         ret = bdrv_pread(bs->file, header.backing_file_offset,
 877                          bs->backing_file, len);
 878         if (ret < 0) {
 879             error_setg_errno(errp, -ret, "Could not read backing file name");
 880             goto fail;
 881         }
 882         bs->backing_file[len] = '\0';
 883     }
 884
 885     /* Internal snapshots */
 886     s->snapshots_offset = header.snapshots_offset;
 887     s->nb_snapshots = header.nb_snapshots;
 888
 889     ret = qcow2_read_snapshots(bs);
 890     if (ret < 0) {
 891         error_setg_errno(errp, -ret, "Could not read snapshots");
 892         goto fail;
 893     }
 894
 895     /* Clear unknown autoclear feature bits */
 896     if (!bs->read_only && !(flags & BDRV_O_INCOMING) && s->autoclear_features) {
 897         s->autoclear_features = 0;
 898         ret = qcow2_update_header(bs);
 899         if (ret < 0) {
 900             error_setg_errno(errp, -ret, "Could not update qcow2 header");
 901             goto fail;
 902         }
 903     }
 904
 905     /* Initialise locks */
 906     qemu_co_mutex_init(&s->lock);
 907
 908     /* Repair image if dirty */
 909     if (!(flags & (BDRV_O_CHECK | BDRV_O_INCOMING)) && !bs->read_only &&
 910         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
 911         BdrvCheckResult result = {0};
 912
 913         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
 914         if (ret < 0) {
 915             error_setg_errno(errp, -ret, "Could not repair dirty image");
 916             goto fail;
 917         }
 918     }
 919
 920     /* Enable lazy_refcounts according to image and command line options */
 921     s->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
 922         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
 923
 924     s->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
 925     s->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
 926     s->discard_passthrough[QCOW2_DISCARD_REQUEST] =
 927         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
 928                           flags & BDRV_O_UNMAP);
 929     s->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
 930         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
 931     s->discard_passthrough[QCOW2_DISCARD_OTHER] =
 932         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
 933
 934     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
 935     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
 936     if (opt_overlap_check_template && opt_overlap_check &&
 937         strcmp(opt_overlap_check_template, opt_overlap_check))
 938     {
 939         error_setg(errp, "Conflicting values for qcow2 options '"
 940                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
 941                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
 942         ret = -EINVAL;
 943         goto fail;
 944     }
 945     if (!opt_overlap_check) {
 946         opt_overlap_check = opt_overlap_check_template ?: "cached";
 947     }
 948
 949     if (!strcmp(opt_overlap_check, "none")) {
 950         overlap_check_template = 0;
 951     } else if (!strcmp(opt_overlap_check, "constant")) {
 952         overlap_check_template = QCOW2_OL_CONSTANT;
 953     } else if (!strcmp(opt_overlap_check, "cached")) {
 954         overlap_check_template = QCOW2_OL_CACHED;
 955     } else if (!strcmp(opt_overlap_check, "all")) {
 956         overlap_check_template = QCOW2_OL_ALL;
 957     } else {
 958         error_setg(errp, "Unsupported value '%s' for qcow2 option "
 959                    "'overlap-check'. Allowed are either of the following: "
 960                    "none, constant, cached, all", opt_overlap_check);
 961         ret = -EINVAL;
 962         goto fail;
 963     }
 964
 965     s->overlap_check = 0;
 966     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
 967         /* overlap-check defines a template bitmask, but every flag may be
 968          * overwritten through the associated boolean option */
 969         s->overlap_check |=
 970             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
 971                               overlap_check_template & (1 << i)) << i;
 972     }
 973
 974     qemu_opts_del(opts);
 975     opts = NULL;
 976
 977     if (s->use_lazy_refcounts && s->qcow_version < 3) {
 978         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
 979                    "qemu 1.1 compatibility level");
 980         ret = -EINVAL;
 981         goto fail;
 982     }
 983
 984 #ifdef DEBUG_ALLOC
 985     {
 986         BdrvCheckResult result = {0};
 987         qcow2_check_refcounts(bs, &result, 0);
 988     }
 989 #endif
 990     return ret;
 991
 992  fail:
 993     qemu_opts_del(opts);
 994     g_free(s->unknown_header_fields);
 995     cleanup_unknown_header_ext(bs);
 996     qcow2_free_snapshots(bs);
 997     qcow2_refcount_close(bs);
 998     qemu_vfree(s->l1_table);
 999     /* else pre-write overlap checks in cache_destroy may crash */
1000     s->l1_table = NULL;
1001     if (s->l2_table_cache) {
1002         qcow2_cache_destroy(bs, s->l2_table_cache);
1003     }
1004     if (s->refcount_block_cache) {
1005         qcow2_cache_destroy(bs, s->refcount_block_cache);
1006     }
1007     g_free(s->cluster_cache);
1008     qemu_vfree(s->cluster_data);
1009     return ret;
1010 }
1011
1012 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1013 {
1014     BDRVQcowState *s = bs->opaque;
1015
1016     bs->bl.write_zeroes_alignment = s->cluster_sectors;
1017 }
1018
1019 static int qcow2_set_key(BlockDriverState *bs, const char *key)
1020 {
1021     BDRVQcowState *s = bs->opaque;
1022     uint8_t keybuf[16];
1023     int len, i;
1024
1025     memset(keybuf, 0, 16);
1026     len = strlen(key);
1027     if (len > 16)
1028         len = 16;
1029     /* XXX: we could compress the chars to 7 bits to increase
1030        entropy */
1031     for(i = 0;i < len;i++) {
1032         keybuf[i] = key[i];
1033     }
1034     s->crypt_method = s->crypt_method_header;
1035
1036     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
1037         return -1;
1038     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
1039         return -1;
1040 #if 0
1041     /* test */
1042     {
1043         uint8_t in[16];
1044         uint8_t out[16];
1045         uint8_t tmp[16];
1046         for(i=0;i<16;i++)
1047             in[i] = i;
1048         AES_encrypt(in, tmp, &s->aes_encrypt_key);
1049         AES_decrypt(tmp, out, &s->aes_decrypt_key);
1050         for(i = 0; i < 16; i++)
1051             printf(" %02x", tmp[i]);
1052         printf("\n");
1053         for(i = 0; i < 16; i++)
1054             printf(" %02x", out[i]);
1055         printf("\n");
1056     }
1057 #endif
1058     return 0;
1059 }
1060
1061 /* We have no actual commit/abort logic for qcow2, but we need to write out any
1062  * unwritten data if we reopen read-only. */
1063 static int qcow2_reopen_prepare(BDRVReopenState *state,
1064                                 BlockReopenQueue *queue, Error **errp)
1065 {
1066     int ret;
1067
1068     if ((state->flags & BDRV_O_RDWR) == 0) {
1069         ret = bdrv_flush(state->bs);
1070         if (ret < 0) {
1071             return ret;
1072         }
1073
1074         ret = qcow2_mark_clean(state->bs);
1075         if (ret < 0) {
1076             return ret;
1077         }
1078     }
1079
1080     return 0;
1081 }
1082
1083 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1084         int64_t sector_num, int nb_sectors, int *pnum)
1085 {
1086     BDRVQcowState *s = bs->opaque;
1087     uint64_t cluster_offset;
1088     int index_in_cluster, ret;
1089     int64_t status = 0;
1090
1091     *pnum = nb_sectors;
1092     qemu_co_mutex_lock(&s->lock);
1093     ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
1094     qemu_co_mutex_unlock(&s->lock);
1095     if (ret < 0) {
1096         return ret;
1097     }
1098
1099     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1100         !s->crypt_method) {
1101         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1102         cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1103         status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1104     }
1105     if (ret == QCOW2_CLUSTER_ZERO) {
1106         status |= BDRV_BLOCK_ZERO;
1107     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1108         status |= BDRV_BLOCK_DATA;
1109     }
1110     return status;
1111 }
1112
1113 /* handle reading after the end of the backing file */
1114 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
1115                   int64_t sector_num, int nb_sectors)
1116 {
1117     int n1;
1118     if ((sector_num + nb_sectors) <= bs->total_sectors)
1119         return nb_sectors;
1120     if (sector_num >= bs->total_sectors)
1121         n1 = 0;
1122     else
1123         n1 = bs->total_sectors - sector_num;
1124
1125     qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1));
1126
1127     return n1;
1128 }
1129
1130 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
1131                           int remaining_sectors, QEMUIOVector *qiov)
1132 {
1133     BDRVQcowState *s = bs->opaque;
1134     int index_in_cluster, n1;
1135     int ret;
1136     int cur_nr_sectors; /* number of sectors in current iteration */
1137     uint64_t cluster_offset = 0;
1138     uint64_t bytes_done = 0;
1139     QEMUIOVector hd_qiov;
1140     uint8_t *cluster_data = NULL;
1141
1142     qemu_iovec_init(&hd_qiov, qiov->niov);
1143
1144     qemu_co_mutex_lock(&s->lock);
1145
1146     while (remaining_sectors != 0) {
1147
1148         /* prepare next request */
1149         cur_nr_sectors = remaining_sectors;
1150         if (s->crypt_method) {
1151             cur_nr_sectors = MIN(cur_nr_sectors,
1152                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1153         }
1154
1155         ret = qcow2_get_cluster_offset(bs, sector_num << 9,
1156             &cur_nr_sectors, &cluster_offset);
1157         if (ret < 0) {
1158             goto fail;
1159         }
1160
1161         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1162
1163         qemu_iovec_reset(&hd_qiov);
1164         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1165             cur_nr_sectors * 512);
1166
1167         switch (ret) {
1168         case QCOW2_CLUSTER_UNALLOCATED:
1169
1170             if (bs->backing_hd) {
1171                 /* read from the base image */
1172                 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov,
1173                     sector_num, cur_nr_sectors);
1174                 if (n1 > 0) {
1175                     QEMUIOVector local_qiov;
1176
1177                     qemu_iovec_init(&local_qiov, hd_qiov.niov);
1178                     qemu_iovec_concat(&local_qiov, &hd_qiov, 0,
1179                                       n1 * BDRV_SECTOR_SIZE);
1180
1181                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1182                     qemu_co_mutex_unlock(&s->lock);
1183                     ret = bdrv_co_readv(bs->backing_hd, sector_num,
1184                                         n1, &local_qiov);
1185                     qemu_co_mutex_lock(&s->lock);
1186
1187                     qemu_iovec_destroy(&local_qiov);
1188
1189                     if (ret < 0) {
1190                         goto fail;
1191                     }
1192                 }
1193             } else {
1194                 /* Note: in this case, no need to wait */
1195                 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1196             }
1197             break;
1198
1199         case QCOW2_CLUSTER_ZERO:
1200             qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1201             break;
1202
1203         case QCOW2_CLUSTER_COMPRESSED:
1204             /* add AIO support for compressed blocks ? */
1205             ret = qcow2_decompress_cluster(bs, cluster_offset);
1206             if (ret < 0) {
1207                 goto fail;
1208             }
1209
1210             qemu_iovec_from_buf(&hd_qiov, 0,
1211                 s->cluster_cache + index_in_cluster * 512,
1212                 512 * cur_nr_sectors);
1213             break;
1214
1215         case QCOW2_CLUSTER_NORMAL:
1216             if ((cluster_offset & 511) != 0) {
1217                 ret = -EIO;
1218                 goto fail;
1219             }
1220
1221             if (s->crypt_method) {
1222                 /*
1223                  * For encrypted images, read everything into a temporary
1224                  * contiguous buffer on which the AES functions can work.
1225                  */
1226                 if (!cluster_data) {
1227                     cluster_data =
1228                         qemu_try_blockalign(bs->file, QCOW_MAX_CRYPT_CLUSTERS
1229                                                       * s->cluster_size);
1230                     if (cluster_data == NULL) {
1231                         ret = -ENOMEM;
1232                         goto fail;
1233                     }
1234                 }
1235
1236                 assert(cur_nr_sectors <=
1237                     QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1238                 qemu_iovec_reset(&hd_qiov);
1239                 qemu_iovec_add(&hd_qiov, cluster_data,
1240                     512 * cur_nr_sectors);
1241             }
1242
1243             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1244             qemu_co_mutex_unlock(&s->lock);
1245             ret = bdrv_co_readv(bs->file,
1246                                 (cluster_offset >> 9) + index_in_cluster,
1247                                 cur_nr_sectors, &hd_qiov);
1248             qemu_co_mutex_lock(&s->lock);
1249             if (ret < 0) {
1250                 goto fail;
1251             }
1252             if (s->crypt_method) {
1253                 qcow2_encrypt_sectors(s, sector_num,  cluster_data,
1254                     cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
1255                 qemu_iovec_from_buf(qiov, bytes_done,
1256                     cluster_data, 512 * cur_nr_sectors);
1257             }
1258             break;
1259
1260         default:
1261             g_assert_not_reached();
1262             ret = -EIO;
1263             goto fail;
1264         }
1265
1266         remaining_sectors -= cur_nr_sectors;
1267         sector_num += cur_nr_sectors;
1268         bytes_done += cur_nr_sectors * 512;
1269     }
1270     ret = 0;
1271
1272 fail:
1273     qemu_co_mutex_unlock(&s->lock);
1274
1275     qemu_iovec_destroy(&hd_qiov);
1276     qemu_vfree(cluster_data);
1277
1278     return ret;
1279 }
1280
1281 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
1282                            int64_t sector_num,
1283                            int remaining_sectors,
1284                            QEMUIOVector *qiov)
1285 {
1286     BDRVQcowState *s = bs->opaque;
1287     int index_in_cluster;
1288     int ret;
1289     int cur_nr_sectors; /* number of sectors in current iteration */
1290     uint64_t cluster_offset;
1291     QEMUIOVector hd_qiov;
1292     uint64_t bytes_done = 0;
1293     uint8_t *cluster_data = NULL;
1294     QCowL2Meta *l2meta = NULL;
1295
1296     trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
1297                                  remaining_sectors);
1298
1299     qemu_iovec_init(&hd_qiov, qiov->niov);
1300
1301     s->cluster_cache_offset = -1; /* disable compressed cache */
1302
1303     qemu_co_mutex_lock(&s->lock);
1304
1305     while (remaining_sectors != 0) {
1306
1307         l2meta = NULL;
1308
1309         trace_qcow2_writev_start_part(qemu_coroutine_self());
1310         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1311         cur_nr_sectors = remaining_sectors;
1312         if (s->crypt_method &&
1313             cur_nr_sectors >
1314             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster) {
1315             cur_nr_sectors =
1316                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster;
1317         }
1318
1319         ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
1320             &cur_nr_sectors, &cluster_offset, &l2meta);
1321         if (ret < 0) {
1322             goto fail;
1323         }
1324
1325         assert((cluster_offset & 511) == 0);
1326
1327         qemu_iovec_reset(&hd_qiov);
1328         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1329             cur_nr_sectors * 512);
1330
1331         if (s->crypt_method) {
1332             if (!cluster_data) {
1333                 cluster_data = qemu_try_blockalign(bs->file,
1334                                                    QCOW_MAX_CRYPT_CLUSTERS
1335                                                    * s->cluster_size);
1336                 if (cluster_data == NULL) {
1337                     ret = -ENOMEM;
1338                     goto fail;
1339                 }
1340             }
1341
1342             assert(hd_qiov.size <=
1343                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1344             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1345
1346             qcow2_encrypt_sectors(s, sector_num, cluster_data,
1347                 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
1348
1349             qemu_iovec_reset(&hd_qiov);
1350             qemu_iovec_add(&hd_qiov, cluster_data,
1351                 cur_nr_sectors * 512);
1352         }
1353
1354         ret = qcow2_pre_write_overlap_check(bs, 0,
1355                 cluster_offset + index_in_cluster * BDRV_SECTOR_SIZE,
1356                 cur_nr_sectors * BDRV_SECTOR_SIZE);
1357         if (ret < 0) {
1358             goto fail;
1359         }
1360
1361         qemu_co_mutex_unlock(&s->lock);
1362         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1363         trace_qcow2_writev_data(qemu_coroutine_self(),
1364                                 (cluster_offset >> 9) + index_in_cluster);
1365         ret = bdrv_co_writev(bs->file,
1366                              (cluster_offset >> 9) + index_in_cluster,
1367                              cur_nr_sectors, &hd_qiov);
1368         qemu_co_mutex_lock(&s->lock);
1369         if (ret < 0) {
1370             goto fail;
1371         }
1372
1373         while (l2meta != NULL) {
1374             QCowL2Meta *next;
1375
1376             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1377             if (ret < 0) {
1378                 goto fail;
1379             }
1380
1381             /* Take the request off the list of running requests */
1382             if (l2meta->nb_clusters != 0) {
1383                 QLIST_REMOVE(l2meta, next_in_flight);
1384             }
1385
1386             qemu_co_queue_restart_all(&l2meta->dependent_requests);
1387
1388             next = l2meta->next;
1389             g_free(l2meta);
1390             l2meta = next;
1391         }
1392
1393         remaining_sectors -= cur_nr_sectors;
1394         sector_num += cur_nr_sectors;
1395         bytes_done += cur_nr_sectors * 512;
1396         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
1397     }
1398     ret = 0;
1399
1400 fail:
1401     qemu_co_mutex_unlock(&s->lock);
1402
1403     while (l2meta != NULL) {
1404         QCowL2Meta *next;
1405
1406         if (l2meta->nb_clusters != 0) {
1407             QLIST_REMOVE(l2meta, next_in_flight);
1408         }
1409         qemu_co_queue_restart_all(&l2meta->dependent_requests);
1410
1411         next = l2meta->next;
1412         g_free(l2meta);
1413         l2meta = next;
1414     }
1415
1416     qemu_iovec_destroy(&hd_qiov);
1417     qemu_vfree(cluster_data);
1418     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
1419
1420     return ret;
1421 }
1422
1423 static void qcow2_close(BlockDriverState *bs)
1424 {
1425     BDRVQcowState *s = bs->opaque;
1426     qemu_vfree(s->l1_table);
1427     /* else pre-write overlap checks in cache_destroy may crash */
1428     s->l1_table = NULL;
1429
1430     if (!(bs->open_flags & BDRV_O_INCOMING)) {
1431         int ret1, ret2;
1432
1433         ret1 = qcow2_cache_flush(bs, s->l2_table_cache);
1434         ret2 = qcow2_cache_flush(bs, s->refcount_block_cache);
1435
1436         if (ret1) {
1437             error_report("Failed to flush the L2 table cache: %s",
1438                          strerror(-ret1));
1439         }
1440         if (ret2) {
1441             error_report("Failed to flush the refcount block cache: %s",
1442                          strerror(-ret2));
1443         }
1444
1445         if (!ret1 && !ret2) {
1446             qcow2_mark_clean(bs);
1447         }
1448     }
1449
1450     qcow2_cache_destroy(bs, s->l2_table_cache);
1451     qcow2_cache_destroy(bs, s->refcount_block_cache);
1452
1453     g_free(s->unknown_header_fields);
1454     cleanup_unknown_header_ext(bs);
1455
1456     g_free(s->cluster_cache);
1457     qemu_vfree(s->cluster_data);
1458     qcow2_refcount_close(bs);
1459     qcow2_free_snapshots(bs);
1460 }
1461
1462 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
1463 {
1464     BDRVQcowState *s = bs->opaque;
1465     int flags = s->flags;
1466     AES_KEY aes_encrypt_key;
1467     AES_KEY aes_decrypt_key;
1468     uint32_t crypt_method = 0;
1469     QDict *options;
1470     Error *local_err = NULL;
1471     int ret;
1472
1473     /*
1474      * Backing files are read-only which makes all of their metadata immutable,
1475      * that means we don't have to worry about reopening them here.
1476      */
1477
1478     if (s->crypt_method) {
1479         crypt_method = s->crypt_method;
1480         memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key));
1481         memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key));
1482     }
1483
1484     qcow2_close(bs);
1485
1486     bdrv_invalidate_cache(bs->file, &local_err);
1487     if (local_err) {
1488         error_propagate(errp, local_err);
1489         return;
1490     }
1491
1492     memset(s, 0, sizeof(BDRVQcowState));
1493     options = qdict_clone_shallow(bs->options);
1494
1495     ret = qcow2_open(bs, options, flags, &local_err);
1496     QDECREF(options);
1497     if (local_err) {
1498         error_setg(errp, "Could not reopen qcow2 layer: %s",
1499                    error_get_pretty(local_err));
1500         error_free(local_err);
1501         return;
1502     } else if (ret < 0) {
1503         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
1504         return;
1505     }
1506
1507     if (crypt_method) {
1508         s->crypt_method = crypt_method;
1509         memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key));
1510         memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key));
1511     }
1512 }
1513
1514 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
1515     size_t len, size_t buflen)
1516 {
1517     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
1518     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
1519
1520     if (buflen < ext_len) {
1521         return -ENOSPC;
1522     }
1523
1524     *ext_backing_fmt = (QCowExtension) {
1525         .magic  = cpu_to_be32(magic),
1526         .len    = cpu_to_be32(len),
1527     };
1528     memcpy(buf + sizeof(QCowExtension), s, len);
1529
1530     return ext_len;
1531 }
1532
1533 /*
1534  * Updates the qcow2 header, including the variable length parts of it, i.e.
1535  * the backing file name and all extensions. qcow2 was not designed to allow
1536  * such changes, so if we run out of space (we can only use the first cluster)
1537  * this function may fail.
1538  *
1539  * Returns 0 on success, -errno in error cases.
1540  */
1541 int qcow2_update_header(BlockDriverState *bs)
1542 {
1543     BDRVQcowState *s = bs->opaque;
1544     QCowHeader *header;
1545     char *buf;
1546     size_t buflen = s->cluster_size;
1547     int ret;
1548     uint64_t total_size;
1549     uint32_t refcount_table_clusters;
1550     size_t header_length;
1551     Qcow2UnknownHeaderExtension *uext;
1552
1553     buf = qemu_blockalign(bs, buflen);
1554
1555     /* Header structure */
1556     header = (QCowHeader*) buf;
1557
1558     if (buflen < sizeof(*header)) {
1559         ret = -ENOSPC;
1560         goto fail;
1561     }
1562
1563     header_length = sizeof(*header) + s->unknown_header_fields_size;
1564     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1565     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1566
1567     *header = (QCowHeader) {
1568         /* Version 2 fields */
1569         .magic                  = cpu_to_be32(QCOW_MAGIC),
1570         .version                = cpu_to_be32(s->qcow_version),
1571         .backing_file_offset    = 0,
1572         .backing_file_size      = 0,
1573         .cluster_bits           = cpu_to_be32(s->cluster_bits),
1574         .size                   = cpu_to_be64(total_size),
1575         .crypt_method           = cpu_to_be32(s->crypt_method_header),
1576         .l1_size                = cpu_to_be32(s->l1_size),
1577         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
1578         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
1579         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1580         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
1581         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
1582
1583         /* Version 3 fields */
1584         .incompatible_features  = cpu_to_be64(s->incompatible_features),
1585         .compatible_features    = cpu_to_be64(s->compatible_features),
1586         .autoclear_features     = cpu_to_be64(s->autoclear_features),
1587         .refcount_order         = cpu_to_be32(s->refcount_order),
1588         .header_length          = cpu_to_be32(header_length),
1589     };
1590
1591     /* For older versions, write a shorter header */
1592     switch (s->qcow_version) {
1593     case 2:
1594         ret = offsetof(QCowHeader, incompatible_features);
1595         break;
1596     case 3:
1597         ret = sizeof(*header);
1598         break;
1599     default:
1600         ret = -EINVAL;
1601         goto fail;
1602     }
1603
1604     buf += ret;
1605     buflen -= ret;
1606     memset(buf, 0, buflen);
1607
1608     /* Preserve any unknown field in the header */
1609     if (s->unknown_header_fields_size) {
1610         if (buflen < s->unknown_header_fields_size) {
1611             ret = -ENOSPC;
1612             goto fail;
1613         }
1614
1615         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1616         buf += s->unknown_header_fields_size;
1617         buflen -= s->unknown_header_fields_size;
1618     }
1619
1620     /* Backing file format header extension */
1621     if (*bs->backing_format) {
1622         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1623                              bs->backing_format, strlen(bs->backing_format),
1624                              buflen);
1625         if (ret < 0) {
1626             goto fail;
1627         }
1628
1629         buf += ret;
1630         buflen -= ret;
1631     }
1632
1633     /* Feature table */
1634     Qcow2Feature features[] = {
1635         {
1636             .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1637             .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
1638             .name = "dirty bit",
1639         },
1640         {
1641             .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1642             .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
1643             .name = "corrupt bit",
1644         },
1645         {
1646             .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1647             .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1648             .name = "lazy refcounts",
1649         },
1650     };
1651
1652     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1653                          features, sizeof(features), buflen);
1654     if (ret < 0) {
1655         goto fail;
1656     }
1657     buf += ret;
1658     buflen -= ret;
1659
1660     /* Keep unknown header extensions */
1661     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1662         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1663         if (ret < 0) {
1664             goto fail;
1665         }
1666
1667         buf += ret;
1668         buflen -= ret;
1669     }
1670
1671     /* End of header extensions */
1672     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1673     if (ret < 0) {
1674         goto fail;
1675     }
1676
1677     buf += ret;
1678     buflen -= ret;
1679
1680     /* Backing file name */
1681     if (*bs->backing_file) {
1682         size_t backing_file_len = strlen(bs->backing_file);
1683
1684         if (buflen < backing_file_len) {
1685             ret = -ENOSPC;
1686             goto fail;
1687         }
1688
1689         /* Using strncpy is ok here, since buf is not NUL-terminated. */
1690         strncpy(buf, bs->backing_file, buflen);
1691
1692         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1693         header->backing_file_size   = cpu_to_be32(backing_file_len);
1694     }
1695
1696     /* Write the new header */
1697     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
1698     if (ret < 0) {
1699         goto fail;
1700     }
1701
1702     ret = 0;
1703 fail:
1704     qemu_vfree(header);
1705     return ret;
1706 }
1707
1708 static int qcow2_change_backing_file(BlockDriverState *bs,
1709     const char *backing_file, const char *backing_fmt)
1710 {
1711     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1712     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1713
1714     return qcow2_update_header(bs);
1715 }
1716
1717 static int preallocate(BlockDriverState *bs)
1718 {
1719     uint64_t nb_sectors;
1720     uint64_t offset;
1721     uint64_t host_offset = 0;
1722     int num;
1723     int ret;
1724     QCowL2Meta *meta;
1725
1726     nb_sectors = bdrv_nb_sectors(bs);
1727     offset = 0;
1728
1729     while (nb_sectors) {
1730         num = MIN(nb_sectors, INT_MAX >> BDRV_SECTOR_BITS);
1731         ret = qcow2_alloc_cluster_offset(bs, offset, &num,
1732                                          &host_offset, &meta);
1733         if (ret < 0) {
1734             return ret;
1735         }
1736
1737         while (meta) {
1738             QCowL2Meta *next = meta->next;
1739
1740             ret = qcow2_alloc_cluster_link_l2(bs, meta);
1741             if (ret < 0) {
1742                 qcow2_free_any_clusters(bs, meta->alloc_offset,
1743                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
1744                 return ret;
1745             }
1746
1747             /* There are no dependent requests, but we need to remove our
1748              * request from the list of in-flight requests */
1749             QLIST_REMOVE(meta, next_in_flight);
1750
1751             g_free(meta);
1752             meta = next;
1753         }
1754
1755         /* TODO Preallocate data if requested */
1756
1757         nb_sectors -= num;
1758         offset += num << BDRV_SECTOR_BITS;
1759     }
1760
1761     /*
1762      * It is expected that the image file is large enough to actually contain
1763      * all of the allocated clusters (otherwise we get failing reads after
1764      * EOF). Extend the image to the last allocated sector.
1765      */
1766     if (host_offset != 0) {
1767         uint8_t buf[BDRV_SECTOR_SIZE];
1768         memset(buf, 0, BDRV_SECTOR_SIZE);
1769         ret = bdrv_write(bs->file, (host_offset >> BDRV_SECTOR_BITS) + num - 1,
1770                          buf, 1);
1771         if (ret < 0) {
1772             return ret;
1773         }
1774     }
1775
1776     return 0;
1777 }
1778
1779 static int qcow2_create2(const char *filename, int64_t total_size,
1780                          const char *backing_file, const char *backing_format,
1781                          int flags, size_t cluster_size, PreallocMode prealloc,
1782                          QemuOpts *opts, int version,
1783                          Error **errp)
1784 {
1785     /* Calculate cluster_bits */
1786     int cluster_bits;
1787     cluster_bits = ffs(cluster_size) - 1;
1788     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
1789         (1 << cluster_bits) != cluster_size)
1790     {
1791         error_setg(errp, "Cluster size must be a power of two between %d and "
1792                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
1793         return -EINVAL;
1794     }
1795
1796     /*
1797      * Open the image file and write a minimal qcow2 header.
1798      *
1799      * We keep things simple and start with a zero-sized image. We also
1800      * do without refcount blocks or a L1 table for now. We'll fix the
1801      * inconsistency later.
1802      *
1803      * We do need a refcount table because growing the refcount table means
1804      * allocating two new refcount blocks - the seconds of which would be at
1805      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
1806      * size for any qcow2 image.
1807      */
1808     BlockDriverState* bs;
1809     QCowHeader *header;
1810     uint64_t* refcount_table;
1811     Error *local_err = NULL;
1812     int ret;
1813
1814     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
1815         int64_t meta_size = 0;
1816         uint64_t nreftablee, nrefblocke, nl1e, nl2e;
1817         int64_t aligned_total_size = align_offset(total_size, cluster_size);
1818
1819         /* header: 1 cluster */
1820         meta_size += cluster_size;
1821
1822         /* total size of L2 tables */
1823         nl2e = aligned_total_size / cluster_size;
1824         nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
1825         meta_size += nl2e * sizeof(uint64_t);
1826
1827         /* total size of L1 tables */
1828         nl1e = nl2e * sizeof(uint64_t) / cluster_size;
1829         nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
1830         meta_size += nl1e * sizeof(uint64_t);
1831
1832         /* total size of refcount blocks
1833          *
1834          * note: every host cluster is reference-counted, including metadata
1835          * (even refcount blocks are recursively included).
1836          * Let:
1837          *   a = total_size (this is the guest disk size)
1838          *   m = meta size not including refcount blocks and refcount tables
1839          *   c = cluster size
1840          *   y1 = number of refcount blocks entries
1841          *   y2 = meta size including everything
1842          * then,
1843          *   y1 = (y2 + a)/c
1844          *   y2 = y1 * sizeof(u16) + y1 * sizeof(u16) * sizeof(u64) / c + m
1845          * we can get y1:
1846          *   y1 = (a + m) / (c - sizeof(u16) - sizeof(u16) * sizeof(u64) / c)
1847          */
1848         nrefblocke = (aligned_total_size + meta_size + cluster_size) /
1849             (cluster_size - sizeof(uint16_t) -
1850              1.0 * sizeof(uint16_t) * sizeof(uint64_t) / cluster_size);
1851         nrefblocke = align_offset(nrefblocke, cluster_size / sizeof(uint16_t));
1852         meta_size += nrefblocke * sizeof(uint16_t);
1853
1854         /* total size of refcount tables */
1855         nreftablee = nrefblocke * sizeof(uint16_t) / cluster_size;
1856         nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t));
1857         meta_size += nreftablee * sizeof(uint64_t);
1858
1859         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
1860                             aligned_total_size + meta_size);
1861         qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc]);
1862     }
1863
1864     ret = bdrv_create_file(filename, opts, &local_err);
1865     if (ret < 0) {
1866         error_propagate(errp, local_err);
1867         return ret;
1868     }
1869
1870     bs = NULL;
1871     ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1872                     NULL, &local_err);
1873     if (ret < 0) {
1874         error_propagate(errp, local_err);
1875         return ret;
1876     }
1877
1878     /* Write the header */
1879     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
1880     header = g_malloc0(cluster_size);
1881     *header = (QCowHeader) {
1882         .magic                      = cpu_to_be32(QCOW_MAGIC),
1883         .version                    = cpu_to_be32(version),
1884         .cluster_bits               = cpu_to_be32(cluster_bits),
1885         .size                       = cpu_to_be64(0),
1886         .l1_table_offset            = cpu_to_be64(0),
1887         .l1_size                    = cpu_to_be32(0),
1888         .refcount_table_offset      = cpu_to_be64(cluster_size),
1889         .refcount_table_clusters    = cpu_to_be32(1),
1890         .refcount_order             = cpu_to_be32(4),
1891         .header_length              = cpu_to_be32(sizeof(*header)),
1892     };
1893
1894     if (flags & BLOCK_FLAG_ENCRYPT) {
1895         header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1896     } else {
1897         header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1898     }
1899
1900     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
1901         header->compatible_features |=
1902             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
1903     }
1904
1905     ret = bdrv_pwrite(bs, 0, header, cluster_size);
1906     g_free(header);
1907     if (ret < 0) {
1908         error_setg_errno(errp, -ret, "Could not write qcow2 header");
1909         goto out;
1910     }
1911
1912     /* Write a refcount table with one refcount block */
1913     refcount_table = g_malloc0(2 * cluster_size);
1914     refcount_table[0] = cpu_to_be64(2 * cluster_size);
1915     ret = bdrv_pwrite(bs, cluster_size, refcount_table, 2 * cluster_size);
1916     g_free(refcount_table);
1917
1918     if (ret < 0) {
1919         error_setg_errno(errp, -ret, "Could not write refcount table");
1920         goto out;
1921     }
1922
1923     bdrv_unref(bs);
1924     bs = NULL;
1925
1926     /*
1927      * And now open the image and make it consistent first (i.e. increase the
1928      * refcount of the cluster that is occupied by the header and the refcount
1929      * table)
1930      */
1931     ret = bdrv_open(&bs, filename, NULL, NULL,
1932                     BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH,
1933                     &bdrv_qcow2, &local_err);
1934     if (ret < 0) {
1935         error_propagate(errp, local_err);
1936         goto out;
1937     }
1938
1939     ret = qcow2_alloc_clusters(bs, 3 * cluster_size);
1940     if (ret < 0) {
1941         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
1942                          "header and refcount table");
1943         goto out;
1944
1945     } else if (ret != 0) {
1946         error_report("Huh, first cluster in empty image is already in use?");
1947         abort();
1948     }
1949
1950     /* Okay, now that we have a valid image, let's give it the right size */
1951     ret = bdrv_truncate(bs, total_size);
1952     if (ret < 0) {
1953         error_setg_errno(errp, -ret, "Could not resize image");
1954         goto out;
1955     }
1956
1957     /* Want a backing file? There you go.*/
1958     if (backing_file) {
1959         ret = bdrv_change_backing_file(bs, backing_file, backing_format);
1960         if (ret < 0) {
1961             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
1962                              "with format '%s'", backing_file, backing_format);
1963             goto out;
1964         }
1965     }
1966
1967     /* And if we're supposed to preallocate metadata, do that now */
1968     if (prealloc != PREALLOC_MODE_OFF) {
1969         BDRVQcowState *s = bs->opaque;
1970         qemu_co_mutex_lock(&s->lock);
1971         ret = preallocate(bs);
1972         qemu_co_mutex_unlock(&s->lock);
1973         if (ret < 0) {
1974             error_setg_errno(errp, -ret, "Could not preallocate metadata");
1975             goto out;
1976         }
1977     }
1978
1979     bdrv_unref(bs);
1980     bs = NULL;
1981
1982     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
1983     ret = bdrv_open(&bs, filename, NULL, NULL,
1984                     BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_BACKING,
1985                     &bdrv_qcow2, &local_err);
1986     if (local_err) {
1987         error_propagate(errp, local_err);
1988         goto out;
1989     }
1990
1991     ret = 0;
1992 out:
1993     if (bs) {
1994         bdrv_unref(bs);
1995     }
1996     return ret;
1997 }
1998
1999 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
2000 {
2001     char *backing_file = NULL;
2002     char *backing_fmt = NULL;
2003     char *buf = NULL;
2004     uint64_t size = 0;
2005     int flags = 0;
2006     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
2007     PreallocMode prealloc;
2008     int version = 3;
2009     Error *local_err = NULL;
2010     int ret;
2011
2012     /* Read out options */
2013     size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2014                     BDRV_SECTOR_SIZE);
2015     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2016     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2017     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2018         flags |= BLOCK_FLAG_ENCRYPT;
2019     }
2020     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2021                                          DEFAULT_CLUSTER_SIZE);
2022     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2023     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
2024                                PREALLOC_MODE_MAX, PREALLOC_MODE_OFF,
2025                                &local_err);
2026     if (local_err) {
2027         error_propagate(errp, local_err);
2028         ret = -EINVAL;
2029         goto finish;
2030     }
2031     g_free(buf);
2032     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2033     if (!buf) {
2034         /* keep the default */
2035     } else if (!strcmp(buf, "0.10")) {
2036         version = 2;
2037     } else if (!strcmp(buf, "1.1")) {
2038         version = 3;
2039     } else {
2040         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2041         ret = -EINVAL;
2042         goto finish;
2043     }
2044
2045     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2046         flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2047     }
2048
2049     if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2050         error_setg(errp, "Backing file and preallocation cannot be used at "
2051                    "the same time");
2052         ret = -EINVAL;
2053         goto finish;
2054     }
2055
2056     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2057         error_setg(errp, "Lazy refcounts only supported with compatibility "
2058                    "level 1.1 and above (use compat=1.1 or greater)");
2059         ret = -EINVAL;
2060         goto finish;
2061     }
2062
2063     ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2064                         cluster_size, prealloc, opts, version, &local_err);
2065     if (local_err) {
2066         error_propagate(errp, local_err);
2067     }
2068
2069 finish:
2070     g_free(backing_file);
2071     g_free(backing_fmt);
2072     g_free(buf);
2073     return ret;
2074 }
2075
2076 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
2077     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2078 {
2079     int ret;
2080     BDRVQcowState *s = bs->opaque;
2081
2082     /* Emulate misaligned zero writes */
2083     if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
2084         return -ENOTSUP;
2085     }
2086
2087     /* Whatever is left can use real zero clusters */
2088     qemu_co_mutex_lock(&s->lock);
2089     ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2090         nb_sectors);
2091     qemu_co_mutex_unlock(&s->lock);
2092
2093     return ret;
2094 }
2095
2096 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
2097     int64_t sector_num, int nb_sectors)
2098 {
2099     int ret;
2100     BDRVQcowState *s = bs->opaque;
2101
2102     qemu_co_mutex_lock(&s->lock);
2103     ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2104         nb_sectors, QCOW2_DISCARD_REQUEST, false);
2105     qemu_co_mutex_unlock(&s->lock);
2106     return ret;
2107 }
2108
2109 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
2110 {
2111     BDRVQcowState *s = bs->opaque;
2112     int64_t new_l1_size;
2113     int ret;
2114
2115     if (offset & 511) {
2116         error_report("The new size must be a multiple of 512");
2117         return -EINVAL;
2118     }
2119
2120     /* cannot proceed if image has snapshots */
2121     if (s->nb_snapshots) {
2122         error_report("Can't resize an image which has snapshots");
2123         return -ENOTSUP;
2124     }
2125
2126     /* shrinking is currently not supported */
2127     if (offset < bs->total_sectors * 512) {
2128         error_report("qcow2 doesn't support shrinking images yet");
2129         return -ENOTSUP;
2130     }
2131
2132     new_l1_size = size_to_l1(s, offset);
2133     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
2134     if (ret < 0) {
2135         return ret;
2136     }
2137
2138     /* write updated header.size */
2139     offset = cpu_to_be64(offset);
2140     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
2141                            &offset, sizeof(uint64_t));
2142     if (ret < 0) {
2143         return ret;
2144     }
2145
2146     s->l1_vm_state_index = new_l1_size;
2147     return 0;
2148 }
2149
2150 /* XXX: put compressed sectors first, then all the cluster aligned
2151    tables to avoid losing bytes in alignment */
2152 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
2153                                   const uint8_t *buf, int nb_sectors)
2154 {
2155     BDRVQcowState *s = bs->opaque;
2156     z_stream strm;
2157     int ret, out_len;
2158     uint8_t *out_buf;
2159     uint64_t cluster_offset;
2160
2161     if (nb_sectors == 0) {
2162         /* align end of file to a sector boundary to ease reading with
2163            sector based I/Os */
2164         cluster_offset = bdrv_getlength(bs->file);
2165         return bdrv_truncate(bs->file, cluster_offset);
2166     }
2167
2168     if (nb_sectors != s->cluster_sectors) {
2169         ret = -EINVAL;
2170
2171         /* Zero-pad last write if image size is not cluster aligned */
2172         if (sector_num + nb_sectors == bs->total_sectors &&
2173             nb_sectors < s->cluster_sectors) {
2174             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
2175             memset(pad_buf, 0, s->cluster_size);
2176             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
2177             ret = qcow2_write_compressed(bs, sector_num,
2178                                          pad_buf, s->cluster_sectors);
2179             qemu_vfree(pad_buf);
2180         }
2181         return ret;
2182     }
2183
2184     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
2185
2186     /* best compression, small window, no zlib header */
2187     memset(&strm, 0, sizeof(strm));
2188     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
2189                        Z_DEFLATED, -12,
2190                        9, Z_DEFAULT_STRATEGY);
2191     if (ret != 0) {
2192         ret = -EINVAL;
2193         goto fail;
2194     }
2195
2196     strm.avail_in = s->cluster_size;
2197     strm.next_in = (uint8_t *)buf;
2198     strm.avail_out = s->cluster_size;
2199     strm.next_out = out_buf;
2200
2201     ret = deflate(&strm, Z_FINISH);
2202     if (ret != Z_STREAM_END && ret != Z_OK) {
2203         deflateEnd(&strm);
2204         ret = -EINVAL;
2205         goto fail;
2206     }
2207     out_len = strm.next_out - out_buf;
2208
2209     deflateEnd(&strm);
2210
2211     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
2212         /* could not compress: write normal cluster */
2213         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
2214         if (ret < 0) {
2215             goto fail;
2216         }
2217     } else {
2218         cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
2219             sector_num << 9, out_len);
2220         if (!cluster_offset) {
2221             ret = -EIO;
2222             goto fail;
2223         }
2224         cluster_offset &= s->cluster_offset_mask;
2225
2226         ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
2227         if (ret < 0) {
2228             goto fail;
2229         }
2230
2231         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
2232         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
2233         if (ret < 0) {
2234             goto fail;
2235         }
2236     }
2237
2238     ret = 0;
2239 fail:
2240     g_free(out_buf);
2241     return ret;
2242 }
2243
2244 static int make_completely_empty(BlockDriverState *bs)
2245 {
2246     BDRVQcowState *s = bs->opaque;
2247     int ret, l1_clusters;
2248     int64_t offset;
2249     uint64_t *new_reftable = NULL;
2250     uint64_t rt_entry, l1_size2;
2251     struct {
2252         uint64_t l1_offset;
2253         uint64_t reftable_offset;
2254         uint32_t reftable_clusters;
2255     } QEMU_PACKED l1_ofs_rt_ofs_cls;
2256
2257     ret = qcow2_cache_empty(bs, s->l2_table_cache);
2258     if (ret < 0) {
2259         goto fail;
2260     }
2261
2262     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
2263     if (ret < 0) {
2264         goto fail;
2265     }
2266
2267     /* Refcounts will be broken utterly */
2268     ret = qcow2_mark_dirty(bs);
2269     if (ret < 0) {
2270         goto fail;
2271     }
2272
2273     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2274
2275     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2276     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
2277
2278     /* After this call, neither the in-memory nor the on-disk refcount
2279      * information accurately describe the actual references */
2280
2281     ret = bdrv_write_zeroes(bs->file, s->l1_table_offset / BDRV_SECTOR_SIZE,
2282                             l1_clusters * s->cluster_sectors, 0);
2283     if (ret < 0) {
2284         goto fail_broken_refcounts;
2285     }
2286     memset(s->l1_table, 0, l1_size2);
2287
2288     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
2289
2290     /* Overwrite enough clusters at the beginning of the sectors to place
2291      * the refcount table, a refcount block and the L1 table in; this may
2292      * overwrite parts of the existing refcount and L1 table, which is not
2293      * an issue because the dirty flag is set, complete data loss is in fact
2294      * desired and partial data loss is consequently fine as well */
2295     ret = bdrv_write_zeroes(bs->file, s->cluster_size / BDRV_SECTOR_SIZE,
2296                             (2 + l1_clusters) * s->cluster_size /
2297                             BDRV_SECTOR_SIZE, 0);
2298     /* This call (even if it failed overall) may have overwritten on-disk
2299      * refcount structures; in that case, the in-memory refcount information
2300      * will probably differ from the on-disk information which makes the BDS
2301      * unusable */
2302     if (ret < 0) {
2303         goto fail_broken_refcounts;
2304     }
2305
2306     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2307     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
2308
2309     /* "Create" an empty reftable (one cluster) directly after the image
2310      * header and an empty L1 table three clusters after the image header;
2311      * the cluster between those two will be used as the first refblock */
2312     cpu_to_be64w(&l1_ofs_rt_ofs_cls.l1_offset, 3 * s->cluster_size);
2313     cpu_to_be64w(&l1_ofs_rt_ofs_cls.reftable_offset, s->cluster_size);
2314     cpu_to_be32w(&l1_ofs_rt_ofs_cls.reftable_clusters, 1);
2315     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
2316                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
2317     if (ret < 0) {
2318         goto fail_broken_refcounts;
2319     }
2320
2321     s->l1_table_offset = 3 * s->cluster_size;
2322
2323     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
2324     if (!new_reftable) {
2325         ret = -ENOMEM;
2326         goto fail_broken_refcounts;
2327     }
2328
2329     s->refcount_table_offset = s->cluster_size;
2330     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
2331
2332     g_free(s->refcount_table);
2333     s->refcount_table = new_reftable;
2334     new_reftable = NULL;
2335
2336     /* Now the in-memory refcount information again corresponds to the on-disk
2337      * information (reftable is empty and no refblocks (the refblock cache is
2338      * empty)); however, this means some clusters (e.g. the image header) are
2339      * referenced, but not refcounted, but the normal qcow2 code assumes that
2340      * the in-memory information is always correct */
2341
2342     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
2343
2344     /* Enter the first refblock into the reftable */
2345     rt_entry = cpu_to_be64(2 * s->cluster_size);
2346     ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
2347                            &rt_entry, sizeof(rt_entry));
2348     if (ret < 0) {
2349         goto fail_broken_refcounts;
2350     }
2351     s->refcount_table[0] = 2 * s->cluster_size;
2352
2353     s->free_cluster_index = 0;
2354     assert(3 + l1_clusters <= s->refcount_block_size);
2355     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
2356     if (offset < 0) {
2357         ret = offset;
2358         goto fail_broken_refcounts;
2359     } else if (offset > 0) {
2360         error_report("First cluster in emptied image is in use");
2361         abort();
2362     }
2363
2364     /* Now finally the in-memory information corresponds to the on-disk
2365      * structures and is correct */
2366     ret = qcow2_mark_clean(bs);
2367     if (ret < 0) {
2368         goto fail;
2369     }
2370
2371     ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size);
2372     if (ret < 0) {
2373         goto fail;
2374     }
2375
2376     return 0;
2377
2378 fail_broken_refcounts:
2379     /* The BDS is unusable at this point. If we wanted to make it usable, we
2380      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
2381      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
2382      * again. However, because the functions which could have caused this error
2383      * path to be taken are used by those functions as well, it's very likely
2384      * that that sequence will fail as well. Therefore, just eject the BDS. */
2385     bs->drv = NULL;
2386
2387 fail:
2388     g_free(new_reftable);
2389     return ret;
2390 }
2391
2392 static int qcow2_make_empty(BlockDriverState *bs)
2393 {
2394     BDRVQcowState *s = bs->opaque;
2395     uint64_t start_sector;
2396     int sector_step = INT_MAX / BDRV_SECTOR_SIZE;
2397     int l1_clusters, ret = 0;
2398
2399     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2400
2401     if (s->qcow_version >= 3 && !s->snapshots &&
2402         3 + l1_clusters <= s->refcount_block_size) {
2403         /* The following function only works for qcow2 v3 images (it requires
2404          * the dirty flag) and only as long as there are no snapshots (because
2405          * it completely empties the image). Furthermore, the L1 table and three
2406          * additional clusters (image header, refcount table, one refcount
2407          * block) have to fit inside one refcount block. */
2408         return make_completely_empty(bs);
2409     }
2410
2411     /* This fallback code simply discards every active cluster; this is slow,
2412      * but works in all cases */
2413     for (start_sector = 0; start_sector < bs->total_sectors;
2414          start_sector += sector_step)
2415     {
2416         /* As this function is generally used after committing an external
2417          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
2418          * default action for this kind of discard is to pass the discard,
2419          * which will ideally result in an actually smaller image file, as
2420          * is probably desired. */
2421         ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE,
2422                                      MIN(sector_step,
2423                                          bs->total_sectors - start_sector),
2424                                      QCOW2_DISCARD_SNAPSHOT, true);
2425         if (ret < 0) {
2426             break;
2427         }
2428     }
2429
2430     return ret;
2431 }
2432
2433 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
2434 {
2435     BDRVQcowState *s = bs->opaque;
2436     int ret;
2437
2438     qemu_co_mutex_lock(&s->lock);
2439     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2440     if (ret < 0) {
2441         qemu_co_mutex_unlock(&s->lock);
2442         return ret;
2443     }
2444
2445     if (qcow2_need_accurate_refcounts(s)) {
2446         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2447         if (ret < 0) {
2448             qemu_co_mutex_unlock(&s->lock);
2449             return ret;
2450         }
2451     }
2452     qemu_co_mutex_unlock(&s->lock);
2453
2454     return 0;
2455 }
2456
2457 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2458 {
2459     BDRVQcowState *s = bs->opaque;
2460     bdi->unallocated_blocks_are_zero = true;
2461     bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
2462     bdi->cluster_size = s->cluster_size;
2463     bdi->vm_state_offset = qcow2_vm_state_offset(s);
2464     return 0;
2465 }
2466
2467 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
2468 {
2469     BDRVQcowState *s = bs->opaque;
2470     ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
2471
2472     *spec_info = (ImageInfoSpecific){
2473         .kind  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
2474         {
2475             .qcow2 = g_new(ImageInfoSpecificQCow2, 1),
2476         },
2477     };
2478     if (s->qcow_version == 2) {
2479         *spec_info->qcow2 = (ImageInfoSpecificQCow2){
2480             .compat = g_strdup("0.10"),
2481         };
2482     } else if (s->qcow_version == 3) {
2483         *spec_info->qcow2 = (ImageInfoSpecificQCow2){
2484             .compat             = g_strdup("1.1"),
2485             .lazy_refcounts     = s->compatible_features &
2486                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
2487             .has_lazy_refcounts = true,
2488             .corrupt            = s->incompatible_features &
2489                                   QCOW2_INCOMPAT_CORRUPT,
2490             .has_corrupt        = true,
2491         };
2492     }
2493
2494     return spec_info;
2495 }
2496
2497 #if 0
2498 static void dump_refcounts(BlockDriverState *bs)
2499 {
2500     BDRVQcowState *s = bs->opaque;
2501     int64_t nb_clusters, k, k1, size;
2502     int refcount;
2503
2504     size = bdrv_getlength(bs->file);
2505     nb_clusters = size_to_clusters(s, size);
2506     for(k = 0; k < nb_clusters;) {
2507         k1 = k;
2508         refcount = get_refcount(bs, k);
2509         k++;
2510         while (k < nb_clusters && get_refcount(bs, k) == refcount)
2511             k++;
2512         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
2513                k - k1);
2514     }
2515 }
2516 #endif
2517
2518 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2519                               int64_t pos)
2520 {
2521     BDRVQcowState *s = bs->opaque;
2522     int64_t total_sectors = bs->total_sectors;
2523     int growable = bs->growable;
2524     bool zero_beyond_eof = bs->zero_beyond_eof;
2525     int ret;
2526
2527     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
2528     bs->growable = 1;
2529     bs->zero_beyond_eof = false;
2530     ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov);
2531     bs->growable = growable;
2532     bs->zero_beyond_eof = zero_beyond_eof;
2533
2534     /* bdrv_co_do_writev will have increased the total_sectors value to include
2535      * the VM state - the VM state is however not an actual part of the block
2536      * device, therefore, we need to restore the old value. */
2537     bs->total_sectors = total_sectors;
2538
2539     return ret;
2540 }
2541
2542 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2543                               int64_t pos, int size)
2544 {
2545     BDRVQcowState *s = bs->opaque;
2546     int growable = bs->growable;
2547     bool zero_beyond_eof = bs->zero_beyond_eof;
2548     int ret;
2549
2550     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
2551     bs->growable = 1;
2552     bs->zero_beyond_eof = false;
2553     ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
2554     bs->growable = growable;
2555     bs->zero_beyond_eof = zero_beyond_eof;
2556
2557     return ret;
2558 }
2559
2560 /*
2561  * Downgrades an image's version. To achieve this, any incompatible features
2562  * have to be removed.
2563  */
2564 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
2565                            BlockDriverAmendStatusCB *status_cb)
2566 {
2567     BDRVQcowState *s = bs->opaque;
2568     int current_version = s->qcow_version;
2569     int ret;
2570
2571     if (target_version == current_version) {
2572         return 0;
2573     } else if (target_version > current_version) {
2574         return -EINVAL;
2575     } else if (target_version != 2) {
2576         return -EINVAL;
2577     }
2578
2579     if (s->refcount_order != 4) {
2580         /* we would have to convert the image to a refcount_order == 4 image
2581          * here; however, since qemu (at the time of writing this) does not
2582          * support anything different than 4 anyway, there is no point in doing
2583          * so right now; however, we should error out (if qemu supports this in
2584          * the future and this code has not been adapted) */
2585         error_report("qcow2_downgrade: Image refcount orders other than 4 are "
2586                      "currently not supported.");
2587         return -ENOTSUP;
2588     }
2589
2590     /* clear incompatible features */
2591     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
2592         ret = qcow2_mark_clean(bs);
2593         if (ret < 0) {
2594             return ret;
2595         }
2596     }
2597
2598     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2599      * the first place; if that happens nonetheless, returning -ENOTSUP is the
2600      * best thing to do anyway */
2601
2602     if (s->incompatible_features) {
2603         return -ENOTSUP;
2604     }
2605
2606     /* since we can ignore compatible features, we can set them to 0 as well */
2607     s->compatible_features = 0;
2608     /* if lazy refcounts have been used, they have already been fixed through
2609      * clearing the dirty flag */
2610
2611     /* clearing autoclear features is trivial */
2612     s->autoclear_features = 0;
2613
2614     ret = qcow2_expand_zero_clusters(bs, status_cb);
2615     if (ret < 0) {
2616         return ret;
2617     }
2618
2619     s->qcow_version = target_version;
2620     ret = qcow2_update_header(bs);
2621     if (ret < 0) {
2622         s->qcow_version = current_version;
2623         return ret;
2624     }
2625     return 0;
2626 }
2627
2628 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
2629                                BlockDriverAmendStatusCB *status_cb)
2630 {
2631     BDRVQcowState *s = bs->opaque;
2632     int old_version = s->qcow_version, new_version = old_version;
2633     uint64_t new_size = 0;
2634     const char *backing_file = NULL, *backing_format = NULL;
2635     bool lazy_refcounts = s->use_lazy_refcounts;
2636     const char *compat = NULL;
2637     uint64_t cluster_size = s->cluster_size;
2638     bool encrypt;
2639     int ret;
2640     QemuOptDesc *desc = opts->list->desc;
2641
2642     while (desc && desc->name) {
2643         if (!qemu_opt_find(opts, desc->name)) {
2644             /* only change explicitly defined options */
2645             desc++;
2646             continue;
2647         }
2648
2649         if (!strcmp(desc->name, "compat")) {
2650             compat = qemu_opt_get(opts, "compat");
2651             if (!compat) {
2652                 /* preserve default */
2653             } else if (!strcmp(compat, "0.10")) {
2654                 new_version = 2;
2655             } else if (!strcmp(compat, "1.1")) {
2656                 new_version = 3;
2657             } else {
2658                 fprintf(stderr, "Unknown compatibility level %s.\n", compat);
2659                 return -EINVAL;
2660             }
2661         } else if (!strcmp(desc->name, "preallocation")) {
2662             fprintf(stderr, "Cannot change preallocation mode.\n");
2663             return -ENOTSUP;
2664         } else if (!strcmp(desc->name, "size")) {
2665             new_size = qemu_opt_get_size(opts, "size", 0);
2666         } else if (!strcmp(desc->name, "backing_file")) {
2667             backing_file = qemu_opt_get(opts, "backing_file");
2668         } else if (!strcmp(desc->name, "backing_fmt")) {
2669             backing_format = qemu_opt_get(opts, "backing_fmt");
2670         } else if (!strcmp(desc->name, "encryption")) {
2671             encrypt = qemu_opt_get_bool(opts, "encryption", s->crypt_method);
2672             if (encrypt != !!s->crypt_method) {
2673                 fprintf(stderr, "Changing the encryption flag is not "
2674                         "supported.\n");
2675                 return -ENOTSUP;
2676             }
2677         } else if (!strcmp(desc->name, "cluster_size")) {
2678             cluster_size = qemu_opt_get_size(opts, "cluster_size",
2679                                              cluster_size);
2680             if (cluster_size != s->cluster_size) {
2681                 fprintf(stderr, "Changing the cluster size is not "
2682                         "supported.\n");
2683                 return -ENOTSUP;
2684             }
2685         } else if (!strcmp(desc->name, "lazy_refcounts")) {
2686             lazy_refcounts = qemu_opt_get_bool(opts, "lazy_refcounts",
2687                                                lazy_refcounts);
2688         } else {
2689             /* if this assertion fails, this probably means a new option was
2690              * added without having it covered here */
2691             assert(false);
2692         }
2693
2694         desc++;
2695     }
2696
2697     if (new_version != old_version) {
2698         if (new_version > old_version) {
2699             /* Upgrade */
2700             s->qcow_version = new_version;
2701             ret = qcow2_update_header(bs);
2702             if (ret < 0) {
2703                 s->qcow_version = old_version;
2704                 return ret;
2705             }
2706         } else {
2707             ret = qcow2_downgrade(bs, new_version, status_cb);
2708             if (ret < 0) {
2709                 return ret;
2710             }
2711         }
2712     }
2713
2714     if (backing_file || backing_format) {
2715         ret = qcow2_change_backing_file(bs, backing_file ?: bs->backing_file,
2716                                         backing_format ?: bs->backing_format);
2717         if (ret < 0) {
2718             return ret;
2719         }
2720     }
2721
2722     if (s->use_lazy_refcounts != lazy_refcounts) {
2723         if (lazy_refcounts) {
2724             if (s->qcow_version < 3) {
2725                 fprintf(stderr, "Lazy refcounts only supported with compatibility "
2726                         "level 1.1 and above (use compat=1.1 or greater)\n");
2727                 return -EINVAL;
2728             }
2729             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
2730             ret = qcow2_update_header(bs);
2731             if (ret < 0) {
2732                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
2733                 return ret;
2734             }
2735             s->use_lazy_refcounts = true;
2736         } else {
2737             /* make image clean first */
2738             ret = qcow2_mark_clean(bs);
2739             if (ret < 0) {
2740                 return ret;
2741             }
2742             /* now disallow lazy refcounts */
2743             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
2744             ret = qcow2_update_header(bs);
2745             if (ret < 0) {
2746                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
2747                 return ret;
2748             }
2749             s->use_lazy_refcounts = false;
2750         }
2751     }
2752
2753     if (new_size) {
2754         ret = bdrv_truncate(bs, new_size);
2755         if (ret < 0) {
2756             return ret;
2757         }
2758     }
2759
2760     return 0;
2761 }
2762
2763 /*
2764  * If offset or size are negative, respectively, they will not be included in
2765  * the BLOCK_IMAGE_CORRUPTED event emitted.
2766  * fatal will be ignored for read-only BDS; corruptions found there will always
2767  * be considered non-fatal.
2768  */
2769 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
2770                              int64_t size, const char *message_format, ...)
2771 {
2772     BDRVQcowState *s = bs->opaque;
2773     char *message;
2774     va_list ap;
2775
2776     fatal = fatal && !bs->read_only;
2777
2778     if (s->signaled_corruption &&
2779         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
2780     {
2781         return;
2782     }
2783
2784     va_start(ap, message_format);
2785     message = g_strdup_vprintf(message_format, ap);
2786     va_end(ap);
2787
2788     if (fatal) {
2789         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
2790                 "corruption events will be suppressed\n", message);
2791     } else {
2792         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
2793                 "corruption events will be suppressed\n", message);
2794     }
2795
2796     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), message,
2797                                           offset >= 0, offset, size >= 0, size,
2798                                           fatal, &error_abort);
2799     g_free(message);
2800
2801     if (fatal) {
2802         qcow2_mark_corrupt(bs);
2803         bs->drv = NULL; /* make BDS unusable */
2804     }
2805
2806     s->signaled_corruption = true;
2807 }
2808
2809 static QemuOptsList qcow2_create_opts = {
2810     .name = "qcow2-create-opts",
2811     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
2812     .desc = {
2813         {
2814             .name = BLOCK_OPT_SIZE,
2815             .type = QEMU_OPT_SIZE,
2816             .help = "Virtual disk size"
2817         },
2818         {
2819             .name = BLOCK_OPT_COMPAT_LEVEL,
2820             .type = QEMU_OPT_STRING,
2821             .help = "Compatibility level (0.10 or 1.1)"
2822         },
2823         {
2824             .name = BLOCK_OPT_BACKING_FILE,
2825             .type = QEMU_OPT_STRING,
2826             .help = "File name of a base image"
2827         },
2828         {
2829             .name = BLOCK_OPT_BACKING_FMT,
2830             .type = QEMU_OPT_STRING,
2831             .help = "Image format of the base image"
2832         },
2833         {
2834             .name = BLOCK_OPT_ENCRYPT,
2835             .type = QEMU_OPT_BOOL,
2836             .help = "Encrypt the image",
2837             .def_value_str = "off"
2838         },
2839         {
2840             .name = BLOCK_OPT_CLUSTER_SIZE,
2841             .type = QEMU_OPT_SIZE,
2842             .help = "qcow2 cluster size",
2843             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
2844         },
2845         {
2846             .name = BLOCK_OPT_PREALLOC,
2847             .type = QEMU_OPT_STRING,
2848             .help = "Preallocation mode (allowed values: off, metadata, "
2849                     "falloc, full)"
2850         },
2851         {
2852             .name = BLOCK_OPT_LAZY_REFCOUNTS,
2853             .type = QEMU_OPT_BOOL,
2854             .help = "Postpone refcount updates",
2855             .def_value_str = "off"
2856         },
2857         { /* end of list */ }
2858     }
2859 };
2860
2861 BlockDriver bdrv_qcow2 = {
2862     .format_name        = "qcow2",
2863     .instance_size      = sizeof(BDRVQcowState),
2864     .bdrv_probe         = qcow2_probe,
2865     .bdrv_open          = qcow2_open,
2866     .bdrv_close         = qcow2_close,
2867     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
2868     .bdrv_create        = qcow2_create,
2869     .bdrv_has_zero_init = bdrv_has_zero_init_1,
2870     .bdrv_co_get_block_status = qcow2_co_get_block_status,
2871     .bdrv_set_key       = qcow2_set_key,
2872
2873     .bdrv_co_readv          = qcow2_co_readv,
2874     .bdrv_co_writev         = qcow2_co_writev,
2875     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
2876
2877     .bdrv_co_write_zeroes   = qcow2_co_write_zeroes,
2878     .bdrv_co_discard        = qcow2_co_discard,
2879     .bdrv_truncate          = qcow2_truncate,
2880     .bdrv_write_compressed  = qcow2_write_compressed,
2881     .bdrv_make_empty        = qcow2_make_empty,
2882
2883     .bdrv_snapshot_create   = qcow2_snapshot_create,
2884     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
2885     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
2886     .bdrv_snapshot_list     = qcow2_snapshot_list,
2887     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
2888     .bdrv_get_info          = qcow2_get_info,
2889     .bdrv_get_specific_info = qcow2_get_specific_info,
2890
2891     .bdrv_save_vmstate    = qcow2_save_vmstate,
2892     .bdrv_load_vmstate    = qcow2_load_vmstate,
2893
2894     .supports_backing           = true,
2895     .bdrv_change_backing_file   = qcow2_change_backing_file,
2896
2897     .bdrv_refresh_limits        = qcow2_refresh_limits,
2898     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
2899
2900     .create_opts         = &qcow2_create_opts,
2901     .bdrv_check          = qcow2_check,
2902     .bdrv_amend_options  = qcow2_amend_options,
2903 };
2904
2905 static void bdrv_qcow2_init(void)
2906 {
2907     bdrv_register(&bdrv_qcow2);
2908 }
2909
2910 block_init(bdrv_qcow2_init);