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 "block/qcow2.h" 29 #include "qemu/error-report.h" 30 #include "qapi/qmp/qerror.h" 31 #include "qapi/qmp/qbool.h" 32 #include "qapi/util.h" 33 #include "qapi/qmp/types.h" 34 #include "qapi-event.h" 35 #include "trace.h" 36 #include "qemu/option_int.h" 37 38 /* 39 Differences with QCOW: 40 41 - Support for multiple incremental snapshots. 42 - Memory management by reference counts. 43 - Clusters which have a reference count of one have the bit 44 QCOW_OFLAG_COPIED to optimize write performance. 45 - Size of compressed clusters is stored in sectors to reduce bit usage 46 in the cluster offsets. 47 - Support for storing additional data (such as the VM state) in the 48 snapshots. 49 - If a backing store is used, the cluster size is not constrained 50 (could be backported to QCOW). 51 - L2 tables have always a size of one cluster. 52 */ 53 54 55 typedef struct { 56 uint32_t magic; 57 uint32_t len; 58 } QEMU_PACKED QCowExtension; 59 60 #define QCOW2_EXT_MAGIC_END 0 61 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA 62 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 63 64 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) 65 { 66 const QCowHeader *cow_header = (const void *)buf; 67 68 if (buf_size >= sizeof(QCowHeader) && 69 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 70 be32_to_cpu(cow_header->version) >= 2) 71 return 100; 72 else 73 return 0; 74 } 75 76 77 /* 78 * read qcow2 extension and fill bs 79 * start reading from start_offset 80 * finish reading upon magic of value 0 or when end_offset reached 81 * unknown magic is skipped (future extension this version knows nothing about) 82 * return 0 upon success, non-0 otherwise 83 */ 84 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, 85 uint64_t end_offset, void **p_feature_table, 86 Error **errp) 87 { 88 BDRVQcow2State *s = bs->opaque; 89 QCowExtension ext; 90 uint64_t offset; 91 int ret; 92 93 #ifdef DEBUG_EXT 94 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); 95 #endif 96 offset = start_offset; 97 while (offset < end_offset) { 98 99 #ifdef DEBUG_EXT 100 /* Sanity check */ 101 if (offset > s->cluster_size) 102 printf("qcow2_read_extension: suspicious offset %lu\n", offset); 103 104 printf("attempting to read extended header in offset %lu\n", offset); 105 #endif 106 107 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext)); 108 if (ret < 0) { 109 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: " 110 "pread fail from offset %" PRIu64, offset); 111 return 1; 112 } 113 be32_to_cpus(&ext.magic); 114 be32_to_cpus(&ext.len); 115 offset += sizeof(ext); 116 #ifdef DEBUG_EXT 117 printf("ext.magic = 0x%x\n", ext.magic); 118 #endif 119 if (offset > end_offset || ext.len > end_offset - offset) { 120 error_setg(errp, "Header extension too large"); 121 return -EINVAL; 122 } 123 124 switch (ext.magic) { 125 case QCOW2_EXT_MAGIC_END: 126 return 0; 127 128 case QCOW2_EXT_MAGIC_BACKING_FORMAT: 129 if (ext.len >= sizeof(bs->backing_format)) { 130 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32 131 " too large (>=%zu)", ext.len, 132 sizeof(bs->backing_format)); 133 return 2; 134 } 135 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len); 136 if (ret < 0) { 137 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: " 138 "Could not read format name"); 139 return 3; 140 } 141 bs->backing_format[ext.len] = '\0'; 142 s->image_backing_format = g_strdup(bs->backing_format); 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 BDRVQcow2State *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_setg(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, 210 bdrv_get_device_or_node_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 BDRVQcow2State *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 BDRVQcow2State *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 BDRVQcow2State *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 BDRVQcow2State *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 BDRVQcow2State *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 { 471 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL, 472 .type = QEMU_OPT_NUMBER, 473 .help = "Clean unused cache entries after this time (in seconds)", 474 }, 475 { /* end of list */ } 476 }, 477 }; 478 479 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { 480 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, 481 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, 482 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, 483 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 484 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 485 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 486 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, 487 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, 488 }; 489 490 static void cache_clean_timer_cb(void *opaque) 491 { 492 BlockDriverState *bs = opaque; 493 BDRVQcow2State *s = bs->opaque; 494 qcow2_cache_clean_unused(bs, s->l2_table_cache); 495 qcow2_cache_clean_unused(bs, s->refcount_block_cache); 496 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 497 (int64_t) s->cache_clean_interval * 1000); 498 } 499 500 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context) 501 { 502 BDRVQcow2State *s = bs->opaque; 503 if (s->cache_clean_interval > 0) { 504 s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL, 505 SCALE_MS, cache_clean_timer_cb, 506 bs); 507 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 508 (int64_t) s->cache_clean_interval * 1000); 509 } 510 } 511 512 static void cache_clean_timer_del(BlockDriverState *bs) 513 { 514 BDRVQcow2State *s = bs->opaque; 515 if (s->cache_clean_timer) { 516 timer_del(s->cache_clean_timer); 517 timer_free(s->cache_clean_timer); 518 s->cache_clean_timer = NULL; 519 } 520 } 521 522 static void qcow2_detach_aio_context(BlockDriverState *bs) 523 { 524 cache_clean_timer_del(bs); 525 } 526 527 static void qcow2_attach_aio_context(BlockDriverState *bs, 528 AioContext *new_context) 529 { 530 cache_clean_timer_init(bs, new_context); 531 } 532 533 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts, 534 uint64_t *l2_cache_size, 535 uint64_t *refcount_cache_size, Error **errp) 536 { 537 BDRVQcow2State *s = bs->opaque; 538 uint64_t combined_cache_size; 539 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set; 540 541 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE); 542 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE); 543 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 544 545 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0); 546 *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0); 547 *refcount_cache_size = qemu_opt_get_size(opts, 548 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0); 549 550 if (combined_cache_size_set) { 551 if (l2_cache_size_set && refcount_cache_size_set) { 552 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE 553 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set " 554 "the same time"); 555 return; 556 } else if (*l2_cache_size > combined_cache_size) { 557 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed " 558 QCOW2_OPT_CACHE_SIZE); 559 return; 560 } else if (*refcount_cache_size > combined_cache_size) { 561 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed " 562 QCOW2_OPT_CACHE_SIZE); 563 return; 564 } 565 566 if (l2_cache_size_set) { 567 *refcount_cache_size = combined_cache_size - *l2_cache_size; 568 } else if (refcount_cache_size_set) { 569 *l2_cache_size = combined_cache_size - *refcount_cache_size; 570 } else { 571 *refcount_cache_size = combined_cache_size 572 / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1); 573 *l2_cache_size = combined_cache_size - *refcount_cache_size; 574 } 575 } else { 576 if (!l2_cache_size_set && !refcount_cache_size_set) { 577 *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE, 578 (uint64_t)DEFAULT_L2_CACHE_CLUSTERS 579 * s->cluster_size); 580 *refcount_cache_size = *l2_cache_size 581 / DEFAULT_L2_REFCOUNT_SIZE_RATIO; 582 } else if (!l2_cache_size_set) { 583 *l2_cache_size = *refcount_cache_size 584 * DEFAULT_L2_REFCOUNT_SIZE_RATIO; 585 } else if (!refcount_cache_size_set) { 586 *refcount_cache_size = *l2_cache_size 587 / DEFAULT_L2_REFCOUNT_SIZE_RATIO; 588 } 589 } 590 } 591 592 typedef struct Qcow2ReopenState { 593 Qcow2Cache *l2_table_cache; 594 Qcow2Cache *refcount_block_cache; 595 bool use_lazy_refcounts; 596 int overlap_check; 597 bool discard_passthrough[QCOW2_DISCARD_MAX]; 598 uint64_t cache_clean_interval; 599 } Qcow2ReopenState; 600 601 static int qcow2_update_options_prepare(BlockDriverState *bs, 602 Qcow2ReopenState *r, 603 QDict *options, int flags, 604 Error **errp) 605 { 606 BDRVQcow2State *s = bs->opaque; 607 QemuOpts *opts = NULL; 608 const char *opt_overlap_check, *opt_overlap_check_template; 609 int overlap_check_template = 0; 610 uint64_t l2_cache_size, refcount_cache_size; 611 int i; 612 Error *local_err = NULL; 613 int ret; 614 615 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); 616 qemu_opts_absorb_qdict(opts, options, &local_err); 617 if (local_err) { 618 error_propagate(errp, local_err); 619 ret = -EINVAL; 620 goto fail; 621 } 622 623 /* get L2 table/refcount block cache size from command line options */ 624 read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size, 625 &local_err); 626 if (local_err) { 627 error_propagate(errp, local_err); 628 ret = -EINVAL; 629 goto fail; 630 } 631 632 l2_cache_size /= s->cluster_size; 633 if (l2_cache_size < MIN_L2_CACHE_SIZE) { 634 l2_cache_size = MIN_L2_CACHE_SIZE; 635 } 636 if (l2_cache_size > INT_MAX) { 637 error_setg(errp, "L2 cache size too big"); 638 ret = -EINVAL; 639 goto fail; 640 } 641 642 refcount_cache_size /= s->cluster_size; 643 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) { 644 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE; 645 } 646 if (refcount_cache_size > INT_MAX) { 647 error_setg(errp, "Refcount cache size too big"); 648 ret = -EINVAL; 649 goto fail; 650 } 651 652 /* alloc new L2 table/refcount block cache, flush old one */ 653 if (s->l2_table_cache) { 654 ret = qcow2_cache_flush(bs, s->l2_table_cache); 655 if (ret) { 656 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache"); 657 goto fail; 658 } 659 } 660 661 if (s->refcount_block_cache) { 662 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 663 if (ret) { 664 error_setg_errno(errp, -ret, 665 "Failed to flush the refcount block cache"); 666 goto fail; 667 } 668 } 669 670 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size); 671 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size); 672 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) { 673 error_setg(errp, "Could not allocate metadata caches"); 674 ret = -ENOMEM; 675 goto fail; 676 } 677 678 /* New interval for cache cleanup timer */ 679 r->cache_clean_interval = 680 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL, 681 s->cache_clean_interval); 682 if (r->cache_clean_interval > UINT_MAX) { 683 error_setg(errp, "Cache clean interval too big"); 684 ret = -EINVAL; 685 goto fail; 686 } 687 688 /* lazy-refcounts; flush if going from enabled to disabled */ 689 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, 690 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); 691 if (r->use_lazy_refcounts && s->qcow_version < 3) { 692 error_setg(errp, "Lazy refcounts require a qcow2 image with at least " 693 "qemu 1.1 compatibility level"); 694 ret = -EINVAL; 695 goto fail; 696 } 697 698 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) { 699 ret = qcow2_mark_clean(bs); 700 if (ret < 0) { 701 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts"); 702 goto fail; 703 } 704 } 705 706 /* Overlap check options */ 707 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP); 708 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE); 709 if (opt_overlap_check_template && opt_overlap_check && 710 strcmp(opt_overlap_check_template, opt_overlap_check)) 711 { 712 error_setg(errp, "Conflicting values for qcow2 options '" 713 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE 714 "' ('%s')", opt_overlap_check, opt_overlap_check_template); 715 ret = -EINVAL; 716 goto fail; 717 } 718 if (!opt_overlap_check) { 719 opt_overlap_check = opt_overlap_check_template ?: "cached"; 720 } 721 722 if (!strcmp(opt_overlap_check, "none")) { 723 overlap_check_template = 0; 724 } else if (!strcmp(opt_overlap_check, "constant")) { 725 overlap_check_template = QCOW2_OL_CONSTANT; 726 } else if (!strcmp(opt_overlap_check, "cached")) { 727 overlap_check_template = QCOW2_OL_CACHED; 728 } else if (!strcmp(opt_overlap_check, "all")) { 729 overlap_check_template = QCOW2_OL_ALL; 730 } else { 731 error_setg(errp, "Unsupported value '%s' for qcow2 option " 732 "'overlap-check'. Allowed are any of the following: " 733 "none, constant, cached, all", opt_overlap_check); 734 ret = -EINVAL; 735 goto fail; 736 } 737 738 r->overlap_check = 0; 739 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { 740 /* overlap-check defines a template bitmask, but every flag may be 741 * overwritten through the associated boolean option */ 742 r->overlap_check |= 743 qemu_opt_get_bool(opts, overlap_bool_option_names[i], 744 overlap_check_template & (1 << i)) << i; 745 } 746 747 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false; 748 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true; 749 r->discard_passthrough[QCOW2_DISCARD_REQUEST] = 750 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, 751 flags & BDRV_O_UNMAP); 752 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = 753 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true); 754 r->discard_passthrough[QCOW2_DISCARD_OTHER] = 755 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false); 756 757 ret = 0; 758 fail: 759 qemu_opts_del(opts); 760 opts = NULL; 761 return ret; 762 } 763 764 static void qcow2_update_options_commit(BlockDriverState *bs, 765 Qcow2ReopenState *r) 766 { 767 BDRVQcow2State *s = bs->opaque; 768 int i; 769 770 if (s->l2_table_cache) { 771 qcow2_cache_destroy(bs, s->l2_table_cache); 772 } 773 if (s->refcount_block_cache) { 774 qcow2_cache_destroy(bs, s->refcount_block_cache); 775 } 776 s->l2_table_cache = r->l2_table_cache; 777 s->refcount_block_cache = r->refcount_block_cache; 778 779 s->overlap_check = r->overlap_check; 780 s->use_lazy_refcounts = r->use_lazy_refcounts; 781 782 for (i = 0; i < QCOW2_DISCARD_MAX; i++) { 783 s->discard_passthrough[i] = r->discard_passthrough[i]; 784 } 785 786 if (s->cache_clean_interval != r->cache_clean_interval) { 787 cache_clean_timer_del(bs); 788 s->cache_clean_interval = r->cache_clean_interval; 789 cache_clean_timer_init(bs, bdrv_get_aio_context(bs)); 790 } 791 } 792 793 static void qcow2_update_options_abort(BlockDriverState *bs, 794 Qcow2ReopenState *r) 795 { 796 if (r->l2_table_cache) { 797 qcow2_cache_destroy(bs, r->l2_table_cache); 798 } 799 if (r->refcount_block_cache) { 800 qcow2_cache_destroy(bs, r->refcount_block_cache); 801 } 802 } 803 804 static int qcow2_update_options(BlockDriverState *bs, QDict *options, 805 int flags, Error **errp) 806 { 807 Qcow2ReopenState r = {}; 808 int ret; 809 810 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp); 811 if (ret >= 0) { 812 qcow2_update_options_commit(bs, &r); 813 } else { 814 qcow2_update_options_abort(bs, &r); 815 } 816 817 return ret; 818 } 819 820 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, 821 Error **errp) 822 { 823 BDRVQcow2State *s = bs->opaque; 824 unsigned int len, i; 825 int ret = 0; 826 QCowHeader header; 827 Error *local_err = NULL; 828 uint64_t ext_end; 829 uint64_t l1_vm_state_index; 830 831 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 832 if (ret < 0) { 833 error_setg_errno(errp, -ret, "Could not read qcow2 header"); 834 goto fail; 835 } 836 be32_to_cpus(&header.magic); 837 be32_to_cpus(&header.version); 838 be64_to_cpus(&header.backing_file_offset); 839 be32_to_cpus(&header.backing_file_size); 840 be64_to_cpus(&header.size); 841 be32_to_cpus(&header.cluster_bits); 842 be32_to_cpus(&header.crypt_method); 843 be64_to_cpus(&header.l1_table_offset); 844 be32_to_cpus(&header.l1_size); 845 be64_to_cpus(&header.refcount_table_offset); 846 be32_to_cpus(&header.refcount_table_clusters); 847 be64_to_cpus(&header.snapshots_offset); 848 be32_to_cpus(&header.nb_snapshots); 849 850 if (header.magic != QCOW_MAGIC) { 851 error_setg(errp, "Image is not in qcow2 format"); 852 ret = -EINVAL; 853 goto fail; 854 } 855 if (header.version < 2 || header.version > 3) { 856 report_unsupported(bs, errp, "QCOW version %" PRIu32, header.version); 857 ret = -ENOTSUP; 858 goto fail; 859 } 860 861 s->qcow_version = header.version; 862 863 /* Initialise cluster size */ 864 if (header.cluster_bits < MIN_CLUSTER_BITS || 865 header.cluster_bits > MAX_CLUSTER_BITS) { 866 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, 867 header.cluster_bits); 868 ret = -EINVAL; 869 goto fail; 870 } 871 872 s->cluster_bits = header.cluster_bits; 873 s->cluster_size = 1 << s->cluster_bits; 874 s->cluster_sectors = 1 << (s->cluster_bits - 9); 875 876 /* Initialise version 3 header fields */ 877 if (header.version == 2) { 878 header.incompatible_features = 0; 879 header.compatible_features = 0; 880 header.autoclear_features = 0; 881 header.refcount_order = 4; 882 header.header_length = 72; 883 } else { 884 be64_to_cpus(&header.incompatible_features); 885 be64_to_cpus(&header.compatible_features); 886 be64_to_cpus(&header.autoclear_features); 887 be32_to_cpus(&header.refcount_order); 888 be32_to_cpus(&header.header_length); 889 890 if (header.header_length < 104) { 891 error_setg(errp, "qcow2 header too short"); 892 ret = -EINVAL; 893 goto fail; 894 } 895 } 896 897 if (header.header_length > s->cluster_size) { 898 error_setg(errp, "qcow2 header exceeds cluster size"); 899 ret = -EINVAL; 900 goto fail; 901 } 902 903 if (header.header_length > sizeof(header)) { 904 s->unknown_header_fields_size = header.header_length - sizeof(header); 905 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); 906 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields, 907 s->unknown_header_fields_size); 908 if (ret < 0) { 909 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header " 910 "fields"); 911 goto fail; 912 } 913 } 914 915 if (header.backing_file_offset > s->cluster_size) { 916 error_setg(errp, "Invalid backing file offset"); 917 ret = -EINVAL; 918 goto fail; 919 } 920 921 if (header.backing_file_offset) { 922 ext_end = header.backing_file_offset; 923 } else { 924 ext_end = 1 << header.cluster_bits; 925 } 926 927 /* Handle feature bits */ 928 s->incompatible_features = header.incompatible_features; 929 s->compatible_features = header.compatible_features; 930 s->autoclear_features = header.autoclear_features; 931 932 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) { 933 void *feature_table = NULL; 934 qcow2_read_extensions(bs, header.header_length, ext_end, 935 &feature_table, NULL); 936 report_unsupported_feature(bs, errp, feature_table, 937 s->incompatible_features & 938 ~QCOW2_INCOMPAT_MASK); 939 ret = -ENOTSUP; 940 g_free(feature_table); 941 goto fail; 942 } 943 944 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 945 /* Corrupt images may not be written to unless they are being repaired 946 */ 947 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 948 error_setg(errp, "qcow2: Image is corrupt; cannot be opened " 949 "read/write"); 950 ret = -EACCES; 951 goto fail; 952 } 953 } 954 955 /* Check support for various header values */ 956 if (header.refcount_order > 6) { 957 error_setg(errp, "Reference count entry width too large; may not " 958 "exceed 64 bits"); 959 ret = -EINVAL; 960 goto fail; 961 } 962 s->refcount_order = header.refcount_order; 963 s->refcount_bits = 1 << s->refcount_order; 964 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 965 s->refcount_max += s->refcount_max - 1; 966 967 if (header.crypt_method > QCOW_CRYPT_AES) { 968 error_setg(errp, "Unsupported encryption method: %" PRIu32, 969 header.crypt_method); 970 ret = -EINVAL; 971 goto fail; 972 } 973 if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) { 974 error_setg(errp, "AES cipher not available"); 975 ret = -EINVAL; 976 goto fail; 977 } 978 s->crypt_method_header = header.crypt_method; 979 if (s->crypt_method_header) { 980 bs->encrypted = 1; 981 } 982 983 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ 984 s->l2_size = 1 << s->l2_bits; 985 /* 2^(s->refcount_order - 3) is the refcount width in bytes */ 986 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3); 987 s->refcount_block_size = 1 << s->refcount_block_bits; 988 bs->total_sectors = header.size / 512; 989 s->csize_shift = (62 - (s->cluster_bits - 8)); 990 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; 991 s->cluster_offset_mask = (1LL << s->csize_shift) - 1; 992 993 s->refcount_table_offset = header.refcount_table_offset; 994 s->refcount_table_size = 995 header.refcount_table_clusters << (s->cluster_bits - 3); 996 997 if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) { 998 error_setg(errp, "Reference count table too large"); 999 ret = -EINVAL; 1000 goto fail; 1001 } 1002 1003 ret = validate_table_offset(bs, s->refcount_table_offset, 1004 s->refcount_table_size, sizeof(uint64_t)); 1005 if (ret < 0) { 1006 error_setg(errp, "Invalid reference count table offset"); 1007 goto fail; 1008 } 1009 1010 /* Snapshot table offset/length */ 1011 if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) { 1012 error_setg(errp, "Too many snapshots"); 1013 ret = -EINVAL; 1014 goto fail; 1015 } 1016 1017 ret = validate_table_offset(bs, header.snapshots_offset, 1018 header.nb_snapshots, 1019 sizeof(QCowSnapshotHeader)); 1020 if (ret < 0) { 1021 error_setg(errp, "Invalid snapshot table offset"); 1022 goto fail; 1023 } 1024 1025 /* read the level 1 table */ 1026 if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) { 1027 error_setg(errp, "Active L1 table too large"); 1028 ret = -EFBIG; 1029 goto fail; 1030 } 1031 s->l1_size = header.l1_size; 1032 1033 l1_vm_state_index = size_to_l1(s, header.size); 1034 if (l1_vm_state_index > INT_MAX) { 1035 error_setg(errp, "Image is too big"); 1036 ret = -EFBIG; 1037 goto fail; 1038 } 1039 s->l1_vm_state_index = l1_vm_state_index; 1040 1041 /* the L1 table must contain at least enough entries to put 1042 header.size bytes */ 1043 if (s->l1_size < s->l1_vm_state_index) { 1044 error_setg(errp, "L1 table is too small"); 1045 ret = -EINVAL; 1046 goto fail; 1047 } 1048 1049 ret = validate_table_offset(bs, header.l1_table_offset, 1050 header.l1_size, sizeof(uint64_t)); 1051 if (ret < 0) { 1052 error_setg(errp, "Invalid L1 table offset"); 1053 goto fail; 1054 } 1055 s->l1_table_offset = header.l1_table_offset; 1056 1057 1058 if (s->l1_size > 0) { 1059 s->l1_table = qemu_try_blockalign(bs->file, 1060 align_offset(s->l1_size * sizeof(uint64_t), 512)); 1061 if (s->l1_table == NULL) { 1062 error_setg(errp, "Could not allocate L1 table"); 1063 ret = -ENOMEM; 1064 goto fail; 1065 } 1066 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 1067 s->l1_size * sizeof(uint64_t)); 1068 if (ret < 0) { 1069 error_setg_errno(errp, -ret, "Could not read L1 table"); 1070 goto fail; 1071 } 1072 for(i = 0;i < s->l1_size; i++) { 1073 be64_to_cpus(&s->l1_table[i]); 1074 } 1075 } 1076 1077 /* Parse driver-specific options */ 1078 ret = qcow2_update_options(bs, options, flags, errp); 1079 if (ret < 0) { 1080 goto fail; 1081 } 1082 1083 s->cluster_cache = g_malloc(s->cluster_size); 1084 /* one more sector for decompressed data alignment */ 1085 s->cluster_data = qemu_try_blockalign(bs->file, QCOW_MAX_CRYPT_CLUSTERS 1086 * s->cluster_size + 512); 1087 if (s->cluster_data == NULL) { 1088 error_setg(errp, "Could not allocate temporary cluster buffer"); 1089 ret = -ENOMEM; 1090 goto fail; 1091 } 1092 1093 s->cluster_cache_offset = -1; 1094 s->flags = flags; 1095 1096 ret = qcow2_refcount_init(bs); 1097 if (ret != 0) { 1098 error_setg_errno(errp, -ret, "Could not initialize refcount handling"); 1099 goto fail; 1100 } 1101 1102 QLIST_INIT(&s->cluster_allocs); 1103 QTAILQ_INIT(&s->discards); 1104 1105 /* read qcow2 extensions */ 1106 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, 1107 &local_err)) { 1108 error_propagate(errp, local_err); 1109 ret = -EINVAL; 1110 goto fail; 1111 } 1112 1113 /* read the backing file name */ 1114 if (header.backing_file_offset != 0) { 1115 len = header.backing_file_size; 1116 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || 1117 len >= sizeof(bs->backing_file)) { 1118 error_setg(errp, "Backing file name too long"); 1119 ret = -EINVAL; 1120 goto fail; 1121 } 1122 ret = bdrv_pread(bs->file, header.backing_file_offset, 1123 bs->backing_file, len); 1124 if (ret < 0) { 1125 error_setg_errno(errp, -ret, "Could not read backing file name"); 1126 goto fail; 1127 } 1128 bs->backing_file[len] = '\0'; 1129 s->image_backing_file = g_strdup(bs->backing_file); 1130 } 1131 1132 /* Internal snapshots */ 1133 s->snapshots_offset = header.snapshots_offset; 1134 s->nb_snapshots = header.nb_snapshots; 1135 1136 ret = qcow2_read_snapshots(bs); 1137 if (ret < 0) { 1138 error_setg_errno(errp, -ret, "Could not read snapshots"); 1139 goto fail; 1140 } 1141 1142 /* Clear unknown autoclear feature bits */ 1143 if (!bs->read_only && !(flags & BDRV_O_INCOMING) && s->autoclear_features) { 1144 s->autoclear_features = 0; 1145 ret = qcow2_update_header(bs); 1146 if (ret < 0) { 1147 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 1148 goto fail; 1149 } 1150 } 1151 1152 /* Initialise locks */ 1153 qemu_co_mutex_init(&s->lock); 1154 1155 /* Repair image if dirty */ 1156 if (!(flags & (BDRV_O_CHECK | BDRV_O_INCOMING)) && !bs->read_only && 1157 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { 1158 BdrvCheckResult result = {0}; 1159 1160 ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); 1161 if (ret < 0) { 1162 error_setg_errno(errp, -ret, "Could not repair dirty image"); 1163 goto fail; 1164 } 1165 } 1166 1167 #ifdef DEBUG_ALLOC 1168 { 1169 BdrvCheckResult result = {0}; 1170 qcow2_check_refcounts(bs, &result, 0); 1171 } 1172 #endif 1173 return ret; 1174 1175 fail: 1176 g_free(s->unknown_header_fields); 1177 cleanup_unknown_header_ext(bs); 1178 qcow2_free_snapshots(bs); 1179 qcow2_refcount_close(bs); 1180 qemu_vfree(s->l1_table); 1181 /* else pre-write overlap checks in cache_destroy may crash */ 1182 s->l1_table = NULL; 1183 cache_clean_timer_del(bs); 1184 if (s->l2_table_cache) { 1185 qcow2_cache_destroy(bs, s->l2_table_cache); 1186 } 1187 if (s->refcount_block_cache) { 1188 qcow2_cache_destroy(bs, s->refcount_block_cache); 1189 } 1190 g_free(s->cluster_cache); 1191 qemu_vfree(s->cluster_data); 1192 return ret; 1193 } 1194 1195 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) 1196 { 1197 BDRVQcow2State *s = bs->opaque; 1198 1199 bs->bl.write_zeroes_alignment = s->cluster_sectors; 1200 } 1201 1202 static int qcow2_set_key(BlockDriverState *bs, const char *key) 1203 { 1204 BDRVQcow2State *s = bs->opaque; 1205 uint8_t keybuf[16]; 1206 int len, i; 1207 Error *err = NULL; 1208 1209 memset(keybuf, 0, 16); 1210 len = strlen(key); 1211 if (len > 16) 1212 len = 16; 1213 /* XXX: we could compress the chars to 7 bits to increase 1214 entropy */ 1215 for(i = 0;i < len;i++) { 1216 keybuf[i] = key[i]; 1217 } 1218 assert(bs->encrypted); 1219 1220 qcrypto_cipher_free(s->cipher); 1221 s->cipher = qcrypto_cipher_new( 1222 QCRYPTO_CIPHER_ALG_AES_128, 1223 QCRYPTO_CIPHER_MODE_CBC, 1224 keybuf, G_N_ELEMENTS(keybuf), 1225 &err); 1226 1227 if (!s->cipher) { 1228 /* XXX would be nice if errors in this method could 1229 * be properly propagate to the caller. Would need 1230 * the bdrv_set_key() API signature to be fixed. */ 1231 error_free(err); 1232 return -1; 1233 } 1234 return 0; 1235 } 1236 1237 static int qcow2_reopen_prepare(BDRVReopenState *state, 1238 BlockReopenQueue *queue, Error **errp) 1239 { 1240 Qcow2ReopenState *r; 1241 int ret; 1242 1243 r = g_new0(Qcow2ReopenState, 1); 1244 state->opaque = r; 1245 1246 ret = qcow2_update_options_prepare(state->bs, r, state->options, 1247 state->flags, errp); 1248 if (ret < 0) { 1249 goto fail; 1250 } 1251 1252 /* We need to write out any unwritten data if we reopen read-only. */ 1253 if ((state->flags & BDRV_O_RDWR) == 0) { 1254 ret = bdrv_flush(state->bs); 1255 if (ret < 0) { 1256 goto fail; 1257 } 1258 1259 ret = qcow2_mark_clean(state->bs); 1260 if (ret < 0) { 1261 goto fail; 1262 } 1263 } 1264 1265 return 0; 1266 1267 fail: 1268 qcow2_update_options_abort(state->bs, r); 1269 g_free(r); 1270 return ret; 1271 } 1272 1273 static void qcow2_reopen_commit(BDRVReopenState *state) 1274 { 1275 qcow2_update_options_commit(state->bs, state->opaque); 1276 g_free(state->opaque); 1277 } 1278 1279 static void qcow2_reopen_abort(BDRVReopenState *state) 1280 { 1281 qcow2_update_options_abort(state->bs, state->opaque); 1282 g_free(state->opaque); 1283 } 1284 1285 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs, 1286 int64_t sector_num, int nb_sectors, int *pnum) 1287 { 1288 BDRVQcow2State *s = bs->opaque; 1289 uint64_t cluster_offset; 1290 int index_in_cluster, ret; 1291 int64_t status = 0; 1292 1293 *pnum = nb_sectors; 1294 qemu_co_mutex_lock(&s->lock); 1295 ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset); 1296 qemu_co_mutex_unlock(&s->lock); 1297 if (ret < 0) { 1298 return ret; 1299 } 1300 1301 if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED && 1302 !s->cipher) { 1303 index_in_cluster = sector_num & (s->cluster_sectors - 1); 1304 cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS); 1305 status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset; 1306 } 1307 if (ret == QCOW2_CLUSTER_ZERO) { 1308 status |= BDRV_BLOCK_ZERO; 1309 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) { 1310 status |= BDRV_BLOCK_DATA; 1311 } 1312 return status; 1313 } 1314 1315 /* handle reading after the end of the backing file */ 1316 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov, 1317 int64_t sector_num, int nb_sectors) 1318 { 1319 int n1; 1320 if ((sector_num + nb_sectors) <= bs->total_sectors) 1321 return nb_sectors; 1322 if (sector_num >= bs->total_sectors) 1323 n1 = 0; 1324 else 1325 n1 = bs->total_sectors - sector_num; 1326 1327 qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1)); 1328 1329 return n1; 1330 } 1331 1332 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num, 1333 int remaining_sectors, QEMUIOVector *qiov) 1334 { 1335 BDRVQcow2State *s = bs->opaque; 1336 int index_in_cluster, n1; 1337 int ret; 1338 int cur_nr_sectors; /* number of sectors in current iteration */ 1339 uint64_t cluster_offset = 0; 1340 uint64_t bytes_done = 0; 1341 QEMUIOVector hd_qiov; 1342 uint8_t *cluster_data = NULL; 1343 1344 qemu_iovec_init(&hd_qiov, qiov->niov); 1345 1346 qemu_co_mutex_lock(&s->lock); 1347 1348 while (remaining_sectors != 0) { 1349 1350 /* prepare next request */ 1351 cur_nr_sectors = remaining_sectors; 1352 if (s->cipher) { 1353 cur_nr_sectors = MIN(cur_nr_sectors, 1354 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); 1355 } 1356 1357 ret = qcow2_get_cluster_offset(bs, sector_num << 9, 1358 &cur_nr_sectors, &cluster_offset); 1359 if (ret < 0) { 1360 goto fail; 1361 } 1362 1363 index_in_cluster = sector_num & (s->cluster_sectors - 1); 1364 1365 qemu_iovec_reset(&hd_qiov); 1366 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, 1367 cur_nr_sectors * 512); 1368 1369 switch (ret) { 1370 case QCOW2_CLUSTER_UNALLOCATED: 1371 1372 if (bs->backing_hd) { 1373 /* read from the base image */ 1374 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov, 1375 sector_num, cur_nr_sectors); 1376 if (n1 > 0) { 1377 QEMUIOVector local_qiov; 1378 1379 qemu_iovec_init(&local_qiov, hd_qiov.niov); 1380 qemu_iovec_concat(&local_qiov, &hd_qiov, 0, 1381 n1 * BDRV_SECTOR_SIZE); 1382 1383 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 1384 qemu_co_mutex_unlock(&s->lock); 1385 ret = bdrv_co_readv(bs->backing_hd, sector_num, 1386 n1, &local_qiov); 1387 qemu_co_mutex_lock(&s->lock); 1388 1389 qemu_iovec_destroy(&local_qiov); 1390 1391 if (ret < 0) { 1392 goto fail; 1393 } 1394 } 1395 } else { 1396 /* Note: in this case, no need to wait */ 1397 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors); 1398 } 1399 break; 1400 1401 case QCOW2_CLUSTER_ZERO: 1402 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors); 1403 break; 1404 1405 case QCOW2_CLUSTER_COMPRESSED: 1406 /* add AIO support for compressed blocks ? */ 1407 ret = qcow2_decompress_cluster(bs, cluster_offset); 1408 if (ret < 0) { 1409 goto fail; 1410 } 1411 1412 qemu_iovec_from_buf(&hd_qiov, 0, 1413 s->cluster_cache + index_in_cluster * 512, 1414 512 * cur_nr_sectors); 1415 break; 1416 1417 case QCOW2_CLUSTER_NORMAL: 1418 if ((cluster_offset & 511) != 0) { 1419 ret = -EIO; 1420 goto fail; 1421 } 1422 1423 if (bs->encrypted) { 1424 assert(s->cipher); 1425 1426 /* 1427 * For encrypted images, read everything into a temporary 1428 * contiguous buffer on which the AES functions can work. 1429 */ 1430 if (!cluster_data) { 1431 cluster_data = 1432 qemu_try_blockalign(bs->file, QCOW_MAX_CRYPT_CLUSTERS 1433 * s->cluster_size); 1434 if (cluster_data == NULL) { 1435 ret = -ENOMEM; 1436 goto fail; 1437 } 1438 } 1439 1440 assert(cur_nr_sectors <= 1441 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); 1442 qemu_iovec_reset(&hd_qiov); 1443 qemu_iovec_add(&hd_qiov, cluster_data, 1444 512 * cur_nr_sectors); 1445 } 1446 1447 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 1448 qemu_co_mutex_unlock(&s->lock); 1449 ret = bdrv_co_readv(bs->file, 1450 (cluster_offset >> 9) + index_in_cluster, 1451 cur_nr_sectors, &hd_qiov); 1452 qemu_co_mutex_lock(&s->lock); 1453 if (ret < 0) { 1454 goto fail; 1455 } 1456 if (bs->encrypted) { 1457 assert(s->cipher); 1458 Error *err = NULL; 1459 if (qcow2_encrypt_sectors(s, sector_num, cluster_data, 1460 cluster_data, cur_nr_sectors, false, 1461 &err) < 0) { 1462 error_free(err); 1463 ret = -EIO; 1464 goto fail; 1465 } 1466 qemu_iovec_from_buf(qiov, bytes_done, 1467 cluster_data, 512 * cur_nr_sectors); 1468 } 1469 break; 1470 1471 default: 1472 g_assert_not_reached(); 1473 ret = -EIO; 1474 goto fail; 1475 } 1476 1477 remaining_sectors -= cur_nr_sectors; 1478 sector_num += cur_nr_sectors; 1479 bytes_done += cur_nr_sectors * 512; 1480 } 1481 ret = 0; 1482 1483 fail: 1484 qemu_co_mutex_unlock(&s->lock); 1485 1486 qemu_iovec_destroy(&hd_qiov); 1487 qemu_vfree(cluster_data); 1488 1489 return ret; 1490 } 1491 1492 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs, 1493 int64_t sector_num, 1494 int remaining_sectors, 1495 QEMUIOVector *qiov) 1496 { 1497 BDRVQcow2State *s = bs->opaque; 1498 int index_in_cluster; 1499 int ret; 1500 int cur_nr_sectors; /* number of sectors in current iteration */ 1501 uint64_t cluster_offset; 1502 QEMUIOVector hd_qiov; 1503 uint64_t bytes_done = 0; 1504 uint8_t *cluster_data = NULL; 1505 QCowL2Meta *l2meta = NULL; 1506 1507 trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num, 1508 remaining_sectors); 1509 1510 qemu_iovec_init(&hd_qiov, qiov->niov); 1511 1512 s->cluster_cache_offset = -1; /* disable compressed cache */ 1513 1514 qemu_co_mutex_lock(&s->lock); 1515 1516 while (remaining_sectors != 0) { 1517 1518 l2meta = NULL; 1519 1520 trace_qcow2_writev_start_part(qemu_coroutine_self()); 1521 index_in_cluster = sector_num & (s->cluster_sectors - 1); 1522 cur_nr_sectors = remaining_sectors; 1523 if (bs->encrypted && 1524 cur_nr_sectors > 1525 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster) { 1526 cur_nr_sectors = 1527 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster; 1528 } 1529 1530 ret = qcow2_alloc_cluster_offset(bs, sector_num << 9, 1531 &cur_nr_sectors, &cluster_offset, &l2meta); 1532 if (ret < 0) { 1533 goto fail; 1534 } 1535 1536 assert((cluster_offset & 511) == 0); 1537 1538 qemu_iovec_reset(&hd_qiov); 1539 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, 1540 cur_nr_sectors * 512); 1541 1542 if (bs->encrypted) { 1543 Error *err = NULL; 1544 assert(s->cipher); 1545 if (!cluster_data) { 1546 cluster_data = qemu_try_blockalign(bs->file, 1547 QCOW_MAX_CRYPT_CLUSTERS 1548 * s->cluster_size); 1549 if (cluster_data == NULL) { 1550 ret = -ENOMEM; 1551 goto fail; 1552 } 1553 } 1554 1555 assert(hd_qiov.size <= 1556 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1557 qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size); 1558 1559 if (qcow2_encrypt_sectors(s, sector_num, cluster_data, 1560 cluster_data, cur_nr_sectors, 1561 true, &err) < 0) { 1562 error_free(err); 1563 ret = -EIO; 1564 goto fail; 1565 } 1566 1567 qemu_iovec_reset(&hd_qiov); 1568 qemu_iovec_add(&hd_qiov, cluster_data, 1569 cur_nr_sectors * 512); 1570 } 1571 1572 ret = qcow2_pre_write_overlap_check(bs, 0, 1573 cluster_offset + index_in_cluster * BDRV_SECTOR_SIZE, 1574 cur_nr_sectors * BDRV_SECTOR_SIZE); 1575 if (ret < 0) { 1576 goto fail; 1577 } 1578 1579 qemu_co_mutex_unlock(&s->lock); 1580 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 1581 trace_qcow2_writev_data(qemu_coroutine_self(), 1582 (cluster_offset >> 9) + index_in_cluster); 1583 ret = bdrv_co_writev(bs->file, 1584 (cluster_offset >> 9) + index_in_cluster, 1585 cur_nr_sectors, &hd_qiov); 1586 qemu_co_mutex_lock(&s->lock); 1587 if (ret < 0) { 1588 goto fail; 1589 } 1590 1591 while (l2meta != NULL) { 1592 QCowL2Meta *next; 1593 1594 ret = qcow2_alloc_cluster_link_l2(bs, l2meta); 1595 if (ret < 0) { 1596 goto fail; 1597 } 1598 1599 /* Take the request off the list of running requests */ 1600 if (l2meta->nb_clusters != 0) { 1601 QLIST_REMOVE(l2meta, next_in_flight); 1602 } 1603 1604 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1605 1606 next = l2meta->next; 1607 g_free(l2meta); 1608 l2meta = next; 1609 } 1610 1611 remaining_sectors -= cur_nr_sectors; 1612 sector_num += cur_nr_sectors; 1613 bytes_done += cur_nr_sectors * 512; 1614 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors); 1615 } 1616 ret = 0; 1617 1618 fail: 1619 qemu_co_mutex_unlock(&s->lock); 1620 1621 while (l2meta != NULL) { 1622 QCowL2Meta *next; 1623 1624 if (l2meta->nb_clusters != 0) { 1625 QLIST_REMOVE(l2meta, next_in_flight); 1626 } 1627 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1628 1629 next = l2meta->next; 1630 g_free(l2meta); 1631 l2meta = next; 1632 } 1633 1634 qemu_iovec_destroy(&hd_qiov); 1635 qemu_vfree(cluster_data); 1636 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 1637 1638 return ret; 1639 } 1640 1641 static void qcow2_close(BlockDriverState *bs) 1642 { 1643 BDRVQcow2State *s = bs->opaque; 1644 qemu_vfree(s->l1_table); 1645 /* else pre-write overlap checks in cache_destroy may crash */ 1646 s->l1_table = NULL; 1647 1648 if (!(bs->open_flags & BDRV_O_INCOMING)) { 1649 int ret1, ret2; 1650 1651 ret1 = qcow2_cache_flush(bs, s->l2_table_cache); 1652 ret2 = qcow2_cache_flush(bs, s->refcount_block_cache); 1653 1654 if (ret1) { 1655 error_report("Failed to flush the L2 table cache: %s", 1656 strerror(-ret1)); 1657 } 1658 if (ret2) { 1659 error_report("Failed to flush the refcount block cache: %s", 1660 strerror(-ret2)); 1661 } 1662 1663 if (!ret1 && !ret2) { 1664 qcow2_mark_clean(bs); 1665 } 1666 } 1667 1668 cache_clean_timer_del(bs); 1669 qcow2_cache_destroy(bs, s->l2_table_cache); 1670 qcow2_cache_destroy(bs, s->refcount_block_cache); 1671 1672 qcrypto_cipher_free(s->cipher); 1673 s->cipher = NULL; 1674 1675 g_free(s->unknown_header_fields); 1676 cleanup_unknown_header_ext(bs); 1677 1678 g_free(s->image_backing_file); 1679 g_free(s->image_backing_format); 1680 1681 g_free(s->cluster_cache); 1682 qemu_vfree(s->cluster_data); 1683 qcow2_refcount_close(bs); 1684 qcow2_free_snapshots(bs); 1685 } 1686 1687 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp) 1688 { 1689 BDRVQcow2State *s = bs->opaque; 1690 int flags = s->flags; 1691 QCryptoCipher *cipher = NULL; 1692 QDict *options; 1693 Error *local_err = NULL; 1694 int ret; 1695 1696 /* 1697 * Backing files are read-only which makes all of their metadata immutable, 1698 * that means we don't have to worry about reopening them here. 1699 */ 1700 1701 cipher = s->cipher; 1702 s->cipher = NULL; 1703 1704 qcow2_close(bs); 1705 1706 bdrv_invalidate_cache(bs->file, &local_err); 1707 if (local_err) { 1708 error_propagate(errp, local_err); 1709 return; 1710 } 1711 1712 memset(s, 0, sizeof(BDRVQcow2State)); 1713 options = qdict_clone_shallow(bs->options); 1714 1715 ret = qcow2_open(bs, options, flags, &local_err); 1716 QDECREF(options); 1717 if (local_err) { 1718 error_setg(errp, "Could not reopen qcow2 layer: %s", 1719 error_get_pretty(local_err)); 1720 error_free(local_err); 1721 return; 1722 } else if (ret < 0) { 1723 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer"); 1724 return; 1725 } 1726 1727 s->cipher = cipher; 1728 } 1729 1730 static size_t header_ext_add(char *buf, uint32_t magic, const void *s, 1731 size_t len, size_t buflen) 1732 { 1733 QCowExtension *ext_backing_fmt = (QCowExtension*) buf; 1734 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); 1735 1736 if (buflen < ext_len) { 1737 return -ENOSPC; 1738 } 1739 1740 *ext_backing_fmt = (QCowExtension) { 1741 .magic = cpu_to_be32(magic), 1742 .len = cpu_to_be32(len), 1743 }; 1744 memcpy(buf + sizeof(QCowExtension), s, len); 1745 1746 return ext_len; 1747 } 1748 1749 /* 1750 * Updates the qcow2 header, including the variable length parts of it, i.e. 1751 * the backing file name and all extensions. qcow2 was not designed to allow 1752 * such changes, so if we run out of space (we can only use the first cluster) 1753 * this function may fail. 1754 * 1755 * Returns 0 on success, -errno in error cases. 1756 */ 1757 int qcow2_update_header(BlockDriverState *bs) 1758 { 1759 BDRVQcow2State *s = bs->opaque; 1760 QCowHeader *header; 1761 char *buf; 1762 size_t buflen = s->cluster_size; 1763 int ret; 1764 uint64_t total_size; 1765 uint32_t refcount_table_clusters; 1766 size_t header_length; 1767 Qcow2UnknownHeaderExtension *uext; 1768 1769 buf = qemu_blockalign(bs, buflen); 1770 1771 /* Header structure */ 1772 header = (QCowHeader*) buf; 1773 1774 if (buflen < sizeof(*header)) { 1775 ret = -ENOSPC; 1776 goto fail; 1777 } 1778 1779 header_length = sizeof(*header) + s->unknown_header_fields_size; 1780 total_size = bs->total_sectors * BDRV_SECTOR_SIZE; 1781 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); 1782 1783 *header = (QCowHeader) { 1784 /* Version 2 fields */ 1785 .magic = cpu_to_be32(QCOW_MAGIC), 1786 .version = cpu_to_be32(s->qcow_version), 1787 .backing_file_offset = 0, 1788 .backing_file_size = 0, 1789 .cluster_bits = cpu_to_be32(s->cluster_bits), 1790 .size = cpu_to_be64(total_size), 1791 .crypt_method = cpu_to_be32(s->crypt_method_header), 1792 .l1_size = cpu_to_be32(s->l1_size), 1793 .l1_table_offset = cpu_to_be64(s->l1_table_offset), 1794 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), 1795 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), 1796 .nb_snapshots = cpu_to_be32(s->nb_snapshots), 1797 .snapshots_offset = cpu_to_be64(s->snapshots_offset), 1798 1799 /* Version 3 fields */ 1800 .incompatible_features = cpu_to_be64(s->incompatible_features), 1801 .compatible_features = cpu_to_be64(s->compatible_features), 1802 .autoclear_features = cpu_to_be64(s->autoclear_features), 1803 .refcount_order = cpu_to_be32(s->refcount_order), 1804 .header_length = cpu_to_be32(header_length), 1805 }; 1806 1807 /* For older versions, write a shorter header */ 1808 switch (s->qcow_version) { 1809 case 2: 1810 ret = offsetof(QCowHeader, incompatible_features); 1811 break; 1812 case 3: 1813 ret = sizeof(*header); 1814 break; 1815 default: 1816 ret = -EINVAL; 1817 goto fail; 1818 } 1819 1820 buf += ret; 1821 buflen -= ret; 1822 memset(buf, 0, buflen); 1823 1824 /* Preserve any unknown field in the header */ 1825 if (s->unknown_header_fields_size) { 1826 if (buflen < s->unknown_header_fields_size) { 1827 ret = -ENOSPC; 1828 goto fail; 1829 } 1830 1831 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); 1832 buf += s->unknown_header_fields_size; 1833 buflen -= s->unknown_header_fields_size; 1834 } 1835 1836 /* Backing file format header extension */ 1837 if (s->image_backing_format) { 1838 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, 1839 s->image_backing_format, 1840 strlen(s->image_backing_format), 1841 buflen); 1842 if (ret < 0) { 1843 goto fail; 1844 } 1845 1846 buf += ret; 1847 buflen -= ret; 1848 } 1849 1850 /* Feature table */ 1851 Qcow2Feature features[] = { 1852 { 1853 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 1854 .bit = QCOW2_INCOMPAT_DIRTY_BITNR, 1855 .name = "dirty bit", 1856 }, 1857 { 1858 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 1859 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, 1860 .name = "corrupt bit", 1861 }, 1862 { 1863 .type = QCOW2_FEAT_TYPE_COMPATIBLE, 1864 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 1865 .name = "lazy refcounts", 1866 }, 1867 }; 1868 1869 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, 1870 features, sizeof(features), buflen); 1871 if (ret < 0) { 1872 goto fail; 1873 } 1874 buf += ret; 1875 buflen -= ret; 1876 1877 /* Keep unknown header extensions */ 1878 QLIST_FOREACH(uext, &s->unknown_header_ext, next) { 1879 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); 1880 if (ret < 0) { 1881 goto fail; 1882 } 1883 1884 buf += ret; 1885 buflen -= ret; 1886 } 1887 1888 /* End of header extensions */ 1889 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); 1890 if (ret < 0) { 1891 goto fail; 1892 } 1893 1894 buf += ret; 1895 buflen -= ret; 1896 1897 /* Backing file name */ 1898 if (s->image_backing_file) { 1899 size_t backing_file_len = strlen(s->image_backing_file); 1900 1901 if (buflen < backing_file_len) { 1902 ret = -ENOSPC; 1903 goto fail; 1904 } 1905 1906 /* Using strncpy is ok here, since buf is not NUL-terminated. */ 1907 strncpy(buf, s->image_backing_file, buflen); 1908 1909 header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); 1910 header->backing_file_size = cpu_to_be32(backing_file_len); 1911 } 1912 1913 /* Write the new header */ 1914 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size); 1915 if (ret < 0) { 1916 goto fail; 1917 } 1918 1919 ret = 0; 1920 fail: 1921 qemu_vfree(header); 1922 return ret; 1923 } 1924 1925 static int qcow2_change_backing_file(BlockDriverState *bs, 1926 const char *backing_file, const char *backing_fmt) 1927 { 1928 BDRVQcow2State *s = bs->opaque; 1929 1930 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 1931 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 1932 1933 g_free(s->image_backing_file); 1934 g_free(s->image_backing_format); 1935 1936 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL; 1937 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL; 1938 1939 return qcow2_update_header(bs); 1940 } 1941 1942 static int preallocate(BlockDriverState *bs) 1943 { 1944 uint64_t nb_sectors; 1945 uint64_t offset; 1946 uint64_t host_offset = 0; 1947 int num; 1948 int ret; 1949 QCowL2Meta *meta; 1950 1951 nb_sectors = bdrv_nb_sectors(bs); 1952 offset = 0; 1953 1954 while (nb_sectors) { 1955 num = MIN(nb_sectors, INT_MAX >> BDRV_SECTOR_BITS); 1956 ret = qcow2_alloc_cluster_offset(bs, offset, &num, 1957 &host_offset, &meta); 1958 if (ret < 0) { 1959 return ret; 1960 } 1961 1962 while (meta) { 1963 QCowL2Meta *next = meta->next; 1964 1965 ret = qcow2_alloc_cluster_link_l2(bs, meta); 1966 if (ret < 0) { 1967 qcow2_free_any_clusters(bs, meta->alloc_offset, 1968 meta->nb_clusters, QCOW2_DISCARD_NEVER); 1969 return ret; 1970 } 1971 1972 /* There are no dependent requests, but we need to remove our 1973 * request from the list of in-flight requests */ 1974 QLIST_REMOVE(meta, next_in_flight); 1975 1976 g_free(meta); 1977 meta = next; 1978 } 1979 1980 /* TODO Preallocate data if requested */ 1981 1982 nb_sectors -= num; 1983 offset += num << BDRV_SECTOR_BITS; 1984 } 1985 1986 /* 1987 * It is expected that the image file is large enough to actually contain 1988 * all of the allocated clusters (otherwise we get failing reads after 1989 * EOF). Extend the image to the last allocated sector. 1990 */ 1991 if (host_offset != 0) { 1992 uint8_t buf[BDRV_SECTOR_SIZE]; 1993 memset(buf, 0, BDRV_SECTOR_SIZE); 1994 ret = bdrv_write(bs->file, (host_offset >> BDRV_SECTOR_BITS) + num - 1, 1995 buf, 1); 1996 if (ret < 0) { 1997 return ret; 1998 } 1999 } 2000 2001 return 0; 2002 } 2003 2004 static int qcow2_create2(const char *filename, int64_t total_size, 2005 const char *backing_file, const char *backing_format, 2006 int flags, size_t cluster_size, PreallocMode prealloc, 2007 QemuOpts *opts, int version, int refcount_order, 2008 Error **errp) 2009 { 2010 int cluster_bits; 2011 QDict *options; 2012 2013 /* Calculate cluster_bits */ 2014 cluster_bits = ctz32(cluster_size); 2015 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 2016 (1 << cluster_bits) != cluster_size) 2017 { 2018 error_setg(errp, "Cluster size must be a power of two between %d and " 2019 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 2020 return -EINVAL; 2021 } 2022 2023 /* 2024 * Open the image file and write a minimal qcow2 header. 2025 * 2026 * We keep things simple and start with a zero-sized image. We also 2027 * do without refcount blocks or a L1 table for now. We'll fix the 2028 * inconsistency later. 2029 * 2030 * We do need a refcount table because growing the refcount table means 2031 * allocating two new refcount blocks - the seconds of which would be at 2032 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 2033 * size for any qcow2 image. 2034 */ 2035 BlockDriverState* bs; 2036 QCowHeader *header; 2037 uint64_t* refcount_table; 2038 Error *local_err = NULL; 2039 int ret; 2040 2041 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 2042 /* Note: The following calculation does not need to be exact; if it is a 2043 * bit off, either some bytes will be "leaked" (which is fine) or we 2044 * will need to increase the file size by some bytes (which is fine, 2045 * too, as long as the bulk is allocated here). Therefore, using 2046 * floating point arithmetic is fine. */ 2047 int64_t meta_size = 0; 2048 uint64_t nreftablee, nrefblocke, nl1e, nl2e; 2049 int64_t aligned_total_size = align_offset(total_size, cluster_size); 2050 int refblock_bits, refblock_size; 2051 /* refcount entry size in bytes */ 2052 double rces = (1 << refcount_order) / 8.; 2053 2054 /* see qcow2_open() */ 2055 refblock_bits = cluster_bits - (refcount_order - 3); 2056 refblock_size = 1 << refblock_bits; 2057 2058 /* header: 1 cluster */ 2059 meta_size += cluster_size; 2060 2061 /* total size of L2 tables */ 2062 nl2e = aligned_total_size / cluster_size; 2063 nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t)); 2064 meta_size += nl2e * sizeof(uint64_t); 2065 2066 /* total size of L1 tables */ 2067 nl1e = nl2e * sizeof(uint64_t) / cluster_size; 2068 nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t)); 2069 meta_size += nl1e * sizeof(uint64_t); 2070 2071 /* total size of refcount blocks 2072 * 2073 * note: every host cluster is reference-counted, including metadata 2074 * (even refcount blocks are recursively included). 2075 * Let: 2076 * a = total_size (this is the guest disk size) 2077 * m = meta size not including refcount blocks and refcount tables 2078 * c = cluster size 2079 * y1 = number of refcount blocks entries 2080 * y2 = meta size including everything 2081 * rces = refcount entry size in bytes 2082 * then, 2083 * y1 = (y2 + a)/c 2084 * y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m 2085 * we can get y1: 2086 * y1 = (a + m) / (c - rces - rces * sizeof(u64) / c) 2087 */ 2088 nrefblocke = (aligned_total_size + meta_size + cluster_size) 2089 / (cluster_size - rces - rces * sizeof(uint64_t) 2090 / cluster_size); 2091 meta_size += DIV_ROUND_UP(nrefblocke, refblock_size) * cluster_size; 2092 2093 /* total size of refcount tables */ 2094 nreftablee = nrefblocke / refblock_size; 2095 nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t)); 2096 meta_size += nreftablee * sizeof(uint64_t); 2097 2098 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, 2099 aligned_total_size + meta_size, &error_abort); 2100 qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc], 2101 &error_abort); 2102 } 2103 2104 ret = bdrv_create_file(filename, opts, &local_err); 2105 if (ret < 0) { 2106 error_propagate(errp, local_err); 2107 return ret; 2108 } 2109 2110 bs = NULL; 2111 ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL, 2112 &local_err); 2113 if (ret < 0) { 2114 error_propagate(errp, local_err); 2115 return ret; 2116 } 2117 2118 /* Write the header */ 2119 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); 2120 header = g_malloc0(cluster_size); 2121 *header = (QCowHeader) { 2122 .magic = cpu_to_be32(QCOW_MAGIC), 2123 .version = cpu_to_be32(version), 2124 .cluster_bits = cpu_to_be32(cluster_bits), 2125 .size = cpu_to_be64(0), 2126 .l1_table_offset = cpu_to_be64(0), 2127 .l1_size = cpu_to_be32(0), 2128 .refcount_table_offset = cpu_to_be64(cluster_size), 2129 .refcount_table_clusters = cpu_to_be32(1), 2130 .refcount_order = cpu_to_be32(refcount_order), 2131 .header_length = cpu_to_be32(sizeof(*header)), 2132 }; 2133 2134 if (flags & BLOCK_FLAG_ENCRYPT) { 2135 header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES); 2136 } else { 2137 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 2138 } 2139 2140 if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) { 2141 header->compatible_features |= 2142 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); 2143 } 2144 2145 ret = bdrv_pwrite(bs, 0, header, cluster_size); 2146 g_free(header); 2147 if (ret < 0) { 2148 error_setg_errno(errp, -ret, "Could not write qcow2 header"); 2149 goto out; 2150 } 2151 2152 /* Write a refcount table with one refcount block */ 2153 refcount_table = g_malloc0(2 * cluster_size); 2154 refcount_table[0] = cpu_to_be64(2 * cluster_size); 2155 ret = bdrv_pwrite(bs, cluster_size, refcount_table, 2 * cluster_size); 2156 g_free(refcount_table); 2157 2158 if (ret < 0) { 2159 error_setg_errno(errp, -ret, "Could not write refcount table"); 2160 goto out; 2161 } 2162 2163 bdrv_unref(bs); 2164 bs = NULL; 2165 2166 /* 2167 * And now open the image and make it consistent first (i.e. increase the 2168 * refcount of the cluster that is occupied by the header and the refcount 2169 * table) 2170 */ 2171 options = qdict_new(); 2172 qdict_put(options, "driver", qstring_from_str("qcow2")); 2173 ret = bdrv_open(&bs, filename, NULL, options, 2174 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, 2175 &local_err); 2176 if (ret < 0) { 2177 error_propagate(errp, local_err); 2178 goto out; 2179 } 2180 2181 ret = qcow2_alloc_clusters(bs, 3 * cluster_size); 2182 if (ret < 0) { 2183 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " 2184 "header and refcount table"); 2185 goto out; 2186 2187 } else if (ret != 0) { 2188 error_report("Huh, first cluster in empty image is already in use?"); 2189 abort(); 2190 } 2191 2192 /* Okay, now that we have a valid image, let's give it the right size */ 2193 ret = bdrv_truncate(bs, total_size); 2194 if (ret < 0) { 2195 error_setg_errno(errp, -ret, "Could not resize image"); 2196 goto out; 2197 } 2198 2199 /* Want a backing file? There you go.*/ 2200 if (backing_file) { 2201 ret = bdrv_change_backing_file(bs, backing_file, backing_format); 2202 if (ret < 0) { 2203 error_setg_errno(errp, -ret, "Could not assign backing file '%s' " 2204 "with format '%s'", backing_file, backing_format); 2205 goto out; 2206 } 2207 } 2208 2209 /* And if we're supposed to preallocate metadata, do that now */ 2210 if (prealloc != PREALLOC_MODE_OFF) { 2211 BDRVQcow2State *s = bs->opaque; 2212 qemu_co_mutex_lock(&s->lock); 2213 ret = preallocate(bs); 2214 qemu_co_mutex_unlock(&s->lock); 2215 if (ret < 0) { 2216 error_setg_errno(errp, -ret, "Could not preallocate metadata"); 2217 goto out; 2218 } 2219 } 2220 2221 bdrv_unref(bs); 2222 bs = NULL; 2223 2224 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */ 2225 options = qdict_new(); 2226 qdict_put(options, "driver", qstring_from_str("qcow2")); 2227 ret = bdrv_open(&bs, filename, NULL, options, 2228 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_BACKING, 2229 &local_err); 2230 if (local_err) { 2231 error_propagate(errp, local_err); 2232 goto out; 2233 } 2234 2235 ret = 0; 2236 out: 2237 if (bs) { 2238 bdrv_unref(bs); 2239 } 2240 return ret; 2241 } 2242 2243 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp) 2244 { 2245 char *backing_file = NULL; 2246 char *backing_fmt = NULL; 2247 char *buf = NULL; 2248 uint64_t size = 0; 2249 int flags = 0; 2250 size_t cluster_size = DEFAULT_CLUSTER_SIZE; 2251 PreallocMode prealloc; 2252 int version = 3; 2253 uint64_t refcount_bits = 16; 2254 int refcount_order; 2255 Error *local_err = NULL; 2256 int ret; 2257 2258 /* Read out options */ 2259 size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2260 BDRV_SECTOR_SIZE); 2261 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 2262 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT); 2263 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) { 2264 flags |= BLOCK_FLAG_ENCRYPT; 2265 } 2266 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 2267 DEFAULT_CLUSTER_SIZE); 2268 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2269 prealloc = qapi_enum_parse(PreallocMode_lookup, buf, 2270 PREALLOC_MODE_MAX, PREALLOC_MODE_OFF, 2271 &local_err); 2272 if (local_err) { 2273 error_propagate(errp, local_err); 2274 ret = -EINVAL; 2275 goto finish; 2276 } 2277 g_free(buf); 2278 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); 2279 if (!buf) { 2280 /* keep the default */ 2281 } else if (!strcmp(buf, "0.10")) { 2282 version = 2; 2283 } else if (!strcmp(buf, "1.1")) { 2284 version = 3; 2285 } else { 2286 error_setg(errp, "Invalid compatibility level: '%s'", buf); 2287 ret = -EINVAL; 2288 goto finish; 2289 } 2290 2291 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) { 2292 flags |= BLOCK_FLAG_LAZY_REFCOUNTS; 2293 } 2294 2295 if (backing_file && prealloc != PREALLOC_MODE_OFF) { 2296 error_setg(errp, "Backing file and preallocation cannot be used at " 2297 "the same time"); 2298 ret = -EINVAL; 2299 goto finish; 2300 } 2301 2302 if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) { 2303 error_setg(errp, "Lazy refcounts only supported with compatibility " 2304 "level 1.1 and above (use compat=1.1 or greater)"); 2305 ret = -EINVAL; 2306 goto finish; 2307 } 2308 2309 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 2310 refcount_bits); 2311 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) { 2312 error_setg(errp, "Refcount width must be a power of two and may not " 2313 "exceed 64 bits"); 2314 ret = -EINVAL; 2315 goto finish; 2316 } 2317 2318 if (version < 3 && refcount_bits != 16) { 2319 error_setg(errp, "Different refcount widths than 16 bits require " 2320 "compatibility level 1.1 or above (use compat=1.1 or " 2321 "greater)"); 2322 ret = -EINVAL; 2323 goto finish; 2324 } 2325 2326 refcount_order = ctz32(refcount_bits); 2327 2328 ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags, 2329 cluster_size, prealloc, opts, version, refcount_order, 2330 &local_err); 2331 if (local_err) { 2332 error_propagate(errp, local_err); 2333 } 2334 2335 finish: 2336 g_free(backing_file); 2337 g_free(backing_fmt); 2338 g_free(buf); 2339 return ret; 2340 } 2341 2342 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs, 2343 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags) 2344 { 2345 int ret; 2346 BDRVQcow2State *s = bs->opaque; 2347 2348 /* Emulate misaligned zero writes */ 2349 if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) { 2350 return -ENOTSUP; 2351 } 2352 2353 /* Whatever is left can use real zero clusters */ 2354 qemu_co_mutex_lock(&s->lock); 2355 ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS, 2356 nb_sectors); 2357 qemu_co_mutex_unlock(&s->lock); 2358 2359 return ret; 2360 } 2361 2362 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs, 2363 int64_t sector_num, int nb_sectors) 2364 { 2365 int ret; 2366 BDRVQcow2State *s = bs->opaque; 2367 2368 qemu_co_mutex_lock(&s->lock); 2369 ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS, 2370 nb_sectors, QCOW2_DISCARD_REQUEST, false); 2371 qemu_co_mutex_unlock(&s->lock); 2372 return ret; 2373 } 2374 2375 static int qcow2_truncate(BlockDriverState *bs, int64_t offset) 2376 { 2377 BDRVQcow2State *s = bs->opaque; 2378 int64_t new_l1_size; 2379 int ret; 2380 2381 if (offset & 511) { 2382 error_report("The new size must be a multiple of 512"); 2383 return -EINVAL; 2384 } 2385 2386 /* cannot proceed if image has snapshots */ 2387 if (s->nb_snapshots) { 2388 error_report("Can't resize an image which has snapshots"); 2389 return -ENOTSUP; 2390 } 2391 2392 /* shrinking is currently not supported */ 2393 if (offset < bs->total_sectors * 512) { 2394 error_report("qcow2 doesn't support shrinking images yet"); 2395 return -ENOTSUP; 2396 } 2397 2398 new_l1_size = size_to_l1(s, offset); 2399 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 2400 if (ret < 0) { 2401 return ret; 2402 } 2403 2404 /* write updated header.size */ 2405 offset = cpu_to_be64(offset); 2406 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), 2407 &offset, sizeof(uint64_t)); 2408 if (ret < 0) { 2409 return ret; 2410 } 2411 2412 s->l1_vm_state_index = new_l1_size; 2413 return 0; 2414 } 2415 2416 /* XXX: put compressed sectors first, then all the cluster aligned 2417 tables to avoid losing bytes in alignment */ 2418 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num, 2419 const uint8_t *buf, int nb_sectors) 2420 { 2421 BDRVQcow2State *s = bs->opaque; 2422 z_stream strm; 2423 int ret, out_len; 2424 uint8_t *out_buf; 2425 uint64_t cluster_offset; 2426 2427 if (nb_sectors == 0) { 2428 /* align end of file to a sector boundary to ease reading with 2429 sector based I/Os */ 2430 cluster_offset = bdrv_getlength(bs->file); 2431 return bdrv_truncate(bs->file, cluster_offset); 2432 } 2433 2434 if (nb_sectors != s->cluster_sectors) { 2435 ret = -EINVAL; 2436 2437 /* Zero-pad last write if image size is not cluster aligned */ 2438 if (sector_num + nb_sectors == bs->total_sectors && 2439 nb_sectors < s->cluster_sectors) { 2440 uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size); 2441 memset(pad_buf, 0, s->cluster_size); 2442 memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE); 2443 ret = qcow2_write_compressed(bs, sector_num, 2444 pad_buf, s->cluster_sectors); 2445 qemu_vfree(pad_buf); 2446 } 2447 return ret; 2448 } 2449 2450 out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128); 2451 2452 /* best compression, small window, no zlib header */ 2453 memset(&strm, 0, sizeof(strm)); 2454 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, 2455 Z_DEFLATED, -12, 2456 9, Z_DEFAULT_STRATEGY); 2457 if (ret != 0) { 2458 ret = -EINVAL; 2459 goto fail; 2460 } 2461 2462 strm.avail_in = s->cluster_size; 2463 strm.next_in = (uint8_t *)buf; 2464 strm.avail_out = s->cluster_size; 2465 strm.next_out = out_buf; 2466 2467 ret = deflate(&strm, Z_FINISH); 2468 if (ret != Z_STREAM_END && ret != Z_OK) { 2469 deflateEnd(&strm); 2470 ret = -EINVAL; 2471 goto fail; 2472 } 2473 out_len = strm.next_out - out_buf; 2474 2475 deflateEnd(&strm); 2476 2477 if (ret != Z_STREAM_END || out_len >= s->cluster_size) { 2478 /* could not compress: write normal cluster */ 2479 ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors); 2480 if (ret < 0) { 2481 goto fail; 2482 } 2483 } else { 2484 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs, 2485 sector_num << 9, out_len); 2486 if (!cluster_offset) { 2487 ret = -EIO; 2488 goto fail; 2489 } 2490 cluster_offset &= s->cluster_offset_mask; 2491 2492 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len); 2493 if (ret < 0) { 2494 goto fail; 2495 } 2496 2497 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); 2498 ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len); 2499 if (ret < 0) { 2500 goto fail; 2501 } 2502 } 2503 2504 ret = 0; 2505 fail: 2506 g_free(out_buf); 2507 return ret; 2508 } 2509 2510 static int make_completely_empty(BlockDriverState *bs) 2511 { 2512 BDRVQcow2State *s = bs->opaque; 2513 int ret, l1_clusters; 2514 int64_t offset; 2515 uint64_t *new_reftable = NULL; 2516 uint64_t rt_entry, l1_size2; 2517 struct { 2518 uint64_t l1_offset; 2519 uint64_t reftable_offset; 2520 uint32_t reftable_clusters; 2521 } QEMU_PACKED l1_ofs_rt_ofs_cls; 2522 2523 ret = qcow2_cache_empty(bs, s->l2_table_cache); 2524 if (ret < 0) { 2525 goto fail; 2526 } 2527 2528 ret = qcow2_cache_empty(bs, s->refcount_block_cache); 2529 if (ret < 0) { 2530 goto fail; 2531 } 2532 2533 /* Refcounts will be broken utterly */ 2534 ret = qcow2_mark_dirty(bs); 2535 if (ret < 0) { 2536 goto fail; 2537 } 2538 2539 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 2540 2541 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 2542 l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t); 2543 2544 /* After this call, neither the in-memory nor the on-disk refcount 2545 * information accurately describe the actual references */ 2546 2547 ret = bdrv_write_zeroes(bs->file, s->l1_table_offset / BDRV_SECTOR_SIZE, 2548 l1_clusters * s->cluster_sectors, 0); 2549 if (ret < 0) { 2550 goto fail_broken_refcounts; 2551 } 2552 memset(s->l1_table, 0, l1_size2); 2553 2554 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); 2555 2556 /* Overwrite enough clusters at the beginning of the sectors to place 2557 * the refcount table, a refcount block and the L1 table in; this may 2558 * overwrite parts of the existing refcount and L1 table, which is not 2559 * an issue because the dirty flag is set, complete data loss is in fact 2560 * desired and partial data loss is consequently fine as well */ 2561 ret = bdrv_write_zeroes(bs->file, s->cluster_size / BDRV_SECTOR_SIZE, 2562 (2 + l1_clusters) * s->cluster_size / 2563 BDRV_SECTOR_SIZE, 0); 2564 /* This call (even if it failed overall) may have overwritten on-disk 2565 * refcount structures; in that case, the in-memory refcount information 2566 * will probably differ from the on-disk information which makes the BDS 2567 * unusable */ 2568 if (ret < 0) { 2569 goto fail_broken_refcounts; 2570 } 2571 2572 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 2573 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); 2574 2575 /* "Create" an empty reftable (one cluster) directly after the image 2576 * header and an empty L1 table three clusters after the image header; 2577 * the cluster between those two will be used as the first refblock */ 2578 cpu_to_be64w(&l1_ofs_rt_ofs_cls.l1_offset, 3 * s->cluster_size); 2579 cpu_to_be64w(&l1_ofs_rt_ofs_cls.reftable_offset, s->cluster_size); 2580 cpu_to_be32w(&l1_ofs_rt_ofs_cls.reftable_clusters, 1); 2581 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), 2582 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); 2583 if (ret < 0) { 2584 goto fail_broken_refcounts; 2585 } 2586 2587 s->l1_table_offset = 3 * s->cluster_size; 2588 2589 new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t)); 2590 if (!new_reftable) { 2591 ret = -ENOMEM; 2592 goto fail_broken_refcounts; 2593 } 2594 2595 s->refcount_table_offset = s->cluster_size; 2596 s->refcount_table_size = s->cluster_size / sizeof(uint64_t); 2597 2598 g_free(s->refcount_table); 2599 s->refcount_table = new_reftable; 2600 new_reftable = NULL; 2601 2602 /* Now the in-memory refcount information again corresponds to the on-disk 2603 * information (reftable is empty and no refblocks (the refblock cache is 2604 * empty)); however, this means some clusters (e.g. the image header) are 2605 * referenced, but not refcounted, but the normal qcow2 code assumes that 2606 * the in-memory information is always correct */ 2607 2608 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 2609 2610 /* Enter the first refblock into the reftable */ 2611 rt_entry = cpu_to_be64(2 * s->cluster_size); 2612 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, 2613 &rt_entry, sizeof(rt_entry)); 2614 if (ret < 0) { 2615 goto fail_broken_refcounts; 2616 } 2617 s->refcount_table[0] = 2 * s->cluster_size; 2618 2619 s->free_cluster_index = 0; 2620 assert(3 + l1_clusters <= s->refcount_block_size); 2621 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); 2622 if (offset < 0) { 2623 ret = offset; 2624 goto fail_broken_refcounts; 2625 } else if (offset > 0) { 2626 error_report("First cluster in emptied image is in use"); 2627 abort(); 2628 } 2629 2630 /* Now finally the in-memory information corresponds to the on-disk 2631 * structures and is correct */ 2632 ret = qcow2_mark_clean(bs); 2633 if (ret < 0) { 2634 goto fail; 2635 } 2636 2637 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size); 2638 if (ret < 0) { 2639 goto fail; 2640 } 2641 2642 return 0; 2643 2644 fail_broken_refcounts: 2645 /* The BDS is unusable at this point. If we wanted to make it usable, we 2646 * would have to call qcow2_refcount_close(), qcow2_refcount_init(), 2647 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() 2648 * again. However, because the functions which could have caused this error 2649 * path to be taken are used by those functions as well, it's very likely 2650 * that that sequence will fail as well. Therefore, just eject the BDS. */ 2651 bs->drv = NULL; 2652 2653 fail: 2654 g_free(new_reftable); 2655 return ret; 2656 } 2657 2658 static int qcow2_make_empty(BlockDriverState *bs) 2659 { 2660 BDRVQcow2State *s = bs->opaque; 2661 uint64_t start_sector; 2662 int sector_step = INT_MAX / BDRV_SECTOR_SIZE; 2663 int l1_clusters, ret = 0; 2664 2665 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 2666 2667 if (s->qcow_version >= 3 && !s->snapshots && 2668 3 + l1_clusters <= s->refcount_block_size) { 2669 /* The following function only works for qcow2 v3 images (it requires 2670 * the dirty flag) and only as long as there are no snapshots (because 2671 * it completely empties the image). Furthermore, the L1 table and three 2672 * additional clusters (image header, refcount table, one refcount 2673 * block) have to fit inside one refcount block. */ 2674 return make_completely_empty(bs); 2675 } 2676 2677 /* This fallback code simply discards every active cluster; this is slow, 2678 * but works in all cases */ 2679 for (start_sector = 0; start_sector < bs->total_sectors; 2680 start_sector += sector_step) 2681 { 2682 /* As this function is generally used after committing an external 2683 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the 2684 * default action for this kind of discard is to pass the discard, 2685 * which will ideally result in an actually smaller image file, as 2686 * is probably desired. */ 2687 ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE, 2688 MIN(sector_step, 2689 bs->total_sectors - start_sector), 2690 QCOW2_DISCARD_SNAPSHOT, true); 2691 if (ret < 0) { 2692 break; 2693 } 2694 } 2695 2696 return ret; 2697 } 2698 2699 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) 2700 { 2701 BDRVQcow2State *s = bs->opaque; 2702 int ret; 2703 2704 qemu_co_mutex_lock(&s->lock); 2705 ret = qcow2_cache_flush(bs, s->l2_table_cache); 2706 if (ret < 0) { 2707 qemu_co_mutex_unlock(&s->lock); 2708 return ret; 2709 } 2710 2711 if (qcow2_need_accurate_refcounts(s)) { 2712 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 2713 if (ret < 0) { 2714 qemu_co_mutex_unlock(&s->lock); 2715 return ret; 2716 } 2717 } 2718 qemu_co_mutex_unlock(&s->lock); 2719 2720 return 0; 2721 } 2722 2723 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2724 { 2725 BDRVQcow2State *s = bs->opaque; 2726 bdi->unallocated_blocks_are_zero = true; 2727 bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3); 2728 bdi->cluster_size = s->cluster_size; 2729 bdi->vm_state_offset = qcow2_vm_state_offset(s); 2730 return 0; 2731 } 2732 2733 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs) 2734 { 2735 BDRVQcow2State *s = bs->opaque; 2736 ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1); 2737 2738 *spec_info = (ImageInfoSpecific){ 2739 .kind = IMAGE_INFO_SPECIFIC_KIND_QCOW2, 2740 { 2741 .qcow2 = g_new(ImageInfoSpecificQCow2, 1), 2742 }, 2743 }; 2744 if (s->qcow_version == 2) { 2745 *spec_info->qcow2 = (ImageInfoSpecificQCow2){ 2746 .compat = g_strdup("0.10"), 2747 .refcount_bits = s->refcount_bits, 2748 }; 2749 } else if (s->qcow_version == 3) { 2750 *spec_info->qcow2 = (ImageInfoSpecificQCow2){ 2751 .compat = g_strdup("1.1"), 2752 .lazy_refcounts = s->compatible_features & 2753 QCOW2_COMPAT_LAZY_REFCOUNTS, 2754 .has_lazy_refcounts = true, 2755 .corrupt = s->incompatible_features & 2756 QCOW2_INCOMPAT_CORRUPT, 2757 .has_corrupt = true, 2758 .refcount_bits = s->refcount_bits, 2759 }; 2760 } 2761 2762 return spec_info; 2763 } 2764 2765 #if 0 2766 static void dump_refcounts(BlockDriverState *bs) 2767 { 2768 BDRVQcow2State *s = bs->opaque; 2769 int64_t nb_clusters, k, k1, size; 2770 int refcount; 2771 2772 size = bdrv_getlength(bs->file); 2773 nb_clusters = size_to_clusters(s, size); 2774 for(k = 0; k < nb_clusters;) { 2775 k1 = k; 2776 refcount = get_refcount(bs, k); 2777 k++; 2778 while (k < nb_clusters && get_refcount(bs, k) == refcount) 2779 k++; 2780 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount, 2781 k - k1); 2782 } 2783 } 2784 #endif 2785 2786 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 2787 int64_t pos) 2788 { 2789 BDRVQcow2State *s = bs->opaque; 2790 int64_t total_sectors = bs->total_sectors; 2791 bool zero_beyond_eof = bs->zero_beyond_eof; 2792 int ret; 2793 2794 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 2795 bs->zero_beyond_eof = false; 2796 ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov); 2797 bs->zero_beyond_eof = zero_beyond_eof; 2798 2799 /* bdrv_co_do_writev will have increased the total_sectors value to include 2800 * the VM state - the VM state is however not an actual part of the block 2801 * device, therefore, we need to restore the old value. */ 2802 bs->total_sectors = total_sectors; 2803 2804 return ret; 2805 } 2806 2807 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf, 2808 int64_t pos, int size) 2809 { 2810 BDRVQcow2State *s = bs->opaque; 2811 bool zero_beyond_eof = bs->zero_beyond_eof; 2812 int ret; 2813 2814 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 2815 bs->zero_beyond_eof = false; 2816 ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size); 2817 bs->zero_beyond_eof = zero_beyond_eof; 2818 2819 return ret; 2820 } 2821 2822 /* 2823 * Downgrades an image's version. To achieve this, any incompatible features 2824 * have to be removed. 2825 */ 2826 static int qcow2_downgrade(BlockDriverState *bs, int target_version, 2827 BlockDriverAmendStatusCB *status_cb) 2828 { 2829 BDRVQcow2State *s = bs->opaque; 2830 int current_version = s->qcow_version; 2831 int ret; 2832 2833 if (target_version == current_version) { 2834 return 0; 2835 } else if (target_version > current_version) { 2836 return -EINVAL; 2837 } else if (target_version != 2) { 2838 return -EINVAL; 2839 } 2840 2841 if (s->refcount_order != 4) { 2842 /* we would have to convert the image to a refcount_order == 4 image 2843 * here; however, since qemu (at the time of writing this) does not 2844 * support anything different than 4 anyway, there is no point in doing 2845 * so right now; however, we should error out (if qemu supports this in 2846 * the future and this code has not been adapted) */ 2847 error_report("qcow2_downgrade: Image refcount orders other than 4 are " 2848 "currently not supported."); 2849 return -ENOTSUP; 2850 } 2851 2852 /* clear incompatible features */ 2853 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 2854 ret = qcow2_mark_clean(bs); 2855 if (ret < 0) { 2856 return ret; 2857 } 2858 } 2859 2860 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in 2861 * the first place; if that happens nonetheless, returning -ENOTSUP is the 2862 * best thing to do anyway */ 2863 2864 if (s->incompatible_features) { 2865 return -ENOTSUP; 2866 } 2867 2868 /* since we can ignore compatible features, we can set them to 0 as well */ 2869 s->compatible_features = 0; 2870 /* if lazy refcounts have been used, they have already been fixed through 2871 * clearing the dirty flag */ 2872 2873 /* clearing autoclear features is trivial */ 2874 s->autoclear_features = 0; 2875 2876 ret = qcow2_expand_zero_clusters(bs, status_cb); 2877 if (ret < 0) { 2878 return ret; 2879 } 2880 2881 s->qcow_version = target_version; 2882 ret = qcow2_update_header(bs); 2883 if (ret < 0) { 2884 s->qcow_version = current_version; 2885 return ret; 2886 } 2887 return 0; 2888 } 2889 2890 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, 2891 BlockDriverAmendStatusCB *status_cb) 2892 { 2893 BDRVQcow2State *s = bs->opaque; 2894 int old_version = s->qcow_version, new_version = old_version; 2895 uint64_t new_size = 0; 2896 const char *backing_file = NULL, *backing_format = NULL; 2897 bool lazy_refcounts = s->use_lazy_refcounts; 2898 const char *compat = NULL; 2899 uint64_t cluster_size = s->cluster_size; 2900 bool encrypt; 2901 int ret; 2902 QemuOptDesc *desc = opts->list->desc; 2903 2904 while (desc && desc->name) { 2905 if (!qemu_opt_find(opts, desc->name)) { 2906 /* only change explicitly defined options */ 2907 desc++; 2908 continue; 2909 } 2910 2911 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) { 2912 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL); 2913 if (!compat) { 2914 /* preserve default */ 2915 } else if (!strcmp(compat, "0.10")) { 2916 new_version = 2; 2917 } else if (!strcmp(compat, "1.1")) { 2918 new_version = 3; 2919 } else { 2920 fprintf(stderr, "Unknown compatibility level %s.\n", compat); 2921 return -EINVAL; 2922 } 2923 } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) { 2924 fprintf(stderr, "Cannot change preallocation mode.\n"); 2925 return -ENOTSUP; 2926 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) { 2927 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); 2928 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) { 2929 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); 2930 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) { 2931 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); 2932 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) { 2933 encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, 2934 !!s->cipher); 2935 2936 if (encrypt != !!s->cipher) { 2937 fprintf(stderr, "Changing the encryption flag is not " 2938 "supported.\n"); 2939 return -ENOTSUP; 2940 } 2941 } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) { 2942 cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 2943 cluster_size); 2944 if (cluster_size != s->cluster_size) { 2945 fprintf(stderr, "Changing the cluster size is not " 2946 "supported.\n"); 2947 return -ENOTSUP; 2948 } 2949 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) { 2950 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS, 2951 lazy_refcounts); 2952 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) { 2953 error_report("Cannot change refcount entry width"); 2954 return -ENOTSUP; 2955 } else { 2956 /* if this assertion fails, this probably means a new option was 2957 * added without having it covered here */ 2958 assert(false); 2959 } 2960 2961 desc++; 2962 } 2963 2964 if (new_version != old_version) { 2965 if (new_version > old_version) { 2966 /* Upgrade */ 2967 s->qcow_version = new_version; 2968 ret = qcow2_update_header(bs); 2969 if (ret < 0) { 2970 s->qcow_version = old_version; 2971 return ret; 2972 } 2973 } else { 2974 ret = qcow2_downgrade(bs, new_version, status_cb); 2975 if (ret < 0) { 2976 return ret; 2977 } 2978 } 2979 } 2980 2981 if (backing_file || backing_format) { 2982 ret = qcow2_change_backing_file(bs, 2983 backing_file ?: s->image_backing_file, 2984 backing_format ?: s->image_backing_format); 2985 if (ret < 0) { 2986 return ret; 2987 } 2988 } 2989 2990 if (s->use_lazy_refcounts != lazy_refcounts) { 2991 if (lazy_refcounts) { 2992 if (s->qcow_version < 3) { 2993 fprintf(stderr, "Lazy refcounts only supported with compatibility " 2994 "level 1.1 and above (use compat=1.1 or greater)\n"); 2995 return -EINVAL; 2996 } 2997 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 2998 ret = qcow2_update_header(bs); 2999 if (ret < 0) { 3000 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 3001 return ret; 3002 } 3003 s->use_lazy_refcounts = true; 3004 } else { 3005 /* make image clean first */ 3006 ret = qcow2_mark_clean(bs); 3007 if (ret < 0) { 3008 return ret; 3009 } 3010 /* now disallow lazy refcounts */ 3011 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 3012 ret = qcow2_update_header(bs); 3013 if (ret < 0) { 3014 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 3015 return ret; 3016 } 3017 s->use_lazy_refcounts = false; 3018 } 3019 } 3020 3021 if (new_size) { 3022 ret = bdrv_truncate(bs, new_size); 3023 if (ret < 0) { 3024 return ret; 3025 } 3026 } 3027 3028 return 0; 3029 } 3030 3031 /* 3032 * If offset or size are negative, respectively, they will not be included in 3033 * the BLOCK_IMAGE_CORRUPTED event emitted. 3034 * fatal will be ignored for read-only BDS; corruptions found there will always 3035 * be considered non-fatal. 3036 */ 3037 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 3038 int64_t size, const char *message_format, ...) 3039 { 3040 BDRVQcow2State *s = bs->opaque; 3041 const char *node_name; 3042 char *message; 3043 va_list ap; 3044 3045 fatal = fatal && !bs->read_only; 3046 3047 if (s->signaled_corruption && 3048 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) 3049 { 3050 return; 3051 } 3052 3053 va_start(ap, message_format); 3054 message = g_strdup_vprintf(message_format, ap); 3055 va_end(ap); 3056 3057 if (fatal) { 3058 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " 3059 "corruption events will be suppressed\n", message); 3060 } else { 3061 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " 3062 "corruption events will be suppressed\n", message); 3063 } 3064 3065 node_name = bdrv_get_node_name(bs); 3066 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), 3067 *node_name != '\0', node_name, 3068 message, offset >= 0, offset, 3069 size >= 0, size, 3070 fatal, &error_abort); 3071 g_free(message); 3072 3073 if (fatal) { 3074 qcow2_mark_corrupt(bs); 3075 bs->drv = NULL; /* make BDS unusable */ 3076 } 3077 3078 s->signaled_corruption = true; 3079 } 3080 3081 static QemuOptsList qcow2_create_opts = { 3082 .name = "qcow2-create-opts", 3083 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), 3084 .desc = { 3085 { 3086 .name = BLOCK_OPT_SIZE, 3087 .type = QEMU_OPT_SIZE, 3088 .help = "Virtual disk size" 3089 }, 3090 { 3091 .name = BLOCK_OPT_COMPAT_LEVEL, 3092 .type = QEMU_OPT_STRING, 3093 .help = "Compatibility level (0.10 or 1.1)" 3094 }, 3095 { 3096 .name = BLOCK_OPT_BACKING_FILE, 3097 .type = QEMU_OPT_STRING, 3098 .help = "File name of a base image" 3099 }, 3100 { 3101 .name = BLOCK_OPT_BACKING_FMT, 3102 .type = QEMU_OPT_STRING, 3103 .help = "Image format of the base image" 3104 }, 3105 { 3106 .name = BLOCK_OPT_ENCRYPT, 3107 .type = QEMU_OPT_BOOL, 3108 .help = "Encrypt the image", 3109 .def_value_str = "off" 3110 }, 3111 { 3112 .name = BLOCK_OPT_CLUSTER_SIZE, 3113 .type = QEMU_OPT_SIZE, 3114 .help = "qcow2 cluster size", 3115 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) 3116 }, 3117 { 3118 .name = BLOCK_OPT_PREALLOC, 3119 .type = QEMU_OPT_STRING, 3120 .help = "Preallocation mode (allowed values: off, metadata, " 3121 "falloc, full)" 3122 }, 3123 { 3124 .name = BLOCK_OPT_LAZY_REFCOUNTS, 3125 .type = QEMU_OPT_BOOL, 3126 .help = "Postpone refcount updates", 3127 .def_value_str = "off" 3128 }, 3129 { 3130 .name = BLOCK_OPT_REFCOUNT_BITS, 3131 .type = QEMU_OPT_NUMBER, 3132 .help = "Width of a reference count entry in bits", 3133 .def_value_str = "16" 3134 }, 3135 { /* end of list */ } 3136 } 3137 }; 3138 3139 BlockDriver bdrv_qcow2 = { 3140 .format_name = "qcow2", 3141 .instance_size = sizeof(BDRVQcow2State), 3142 .bdrv_probe = qcow2_probe, 3143 .bdrv_open = qcow2_open, 3144 .bdrv_close = qcow2_close, 3145 .bdrv_reopen_prepare = qcow2_reopen_prepare, 3146 .bdrv_reopen_commit = qcow2_reopen_commit, 3147 .bdrv_reopen_abort = qcow2_reopen_abort, 3148 .bdrv_create = qcow2_create, 3149 .bdrv_has_zero_init = bdrv_has_zero_init_1, 3150 .bdrv_co_get_block_status = qcow2_co_get_block_status, 3151 .bdrv_set_key = qcow2_set_key, 3152 3153 .bdrv_co_readv = qcow2_co_readv, 3154 .bdrv_co_writev = qcow2_co_writev, 3155 .bdrv_co_flush_to_os = qcow2_co_flush_to_os, 3156 3157 .bdrv_co_write_zeroes = qcow2_co_write_zeroes, 3158 .bdrv_co_discard = qcow2_co_discard, 3159 .bdrv_truncate = qcow2_truncate, 3160 .bdrv_write_compressed = qcow2_write_compressed, 3161 .bdrv_make_empty = qcow2_make_empty, 3162 3163 .bdrv_snapshot_create = qcow2_snapshot_create, 3164 .bdrv_snapshot_goto = qcow2_snapshot_goto, 3165 .bdrv_snapshot_delete = qcow2_snapshot_delete, 3166 .bdrv_snapshot_list = qcow2_snapshot_list, 3167 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 3168 .bdrv_get_info = qcow2_get_info, 3169 .bdrv_get_specific_info = qcow2_get_specific_info, 3170 3171 .bdrv_save_vmstate = qcow2_save_vmstate, 3172 .bdrv_load_vmstate = qcow2_load_vmstate, 3173 3174 .supports_backing = true, 3175 .bdrv_change_backing_file = qcow2_change_backing_file, 3176 3177 .bdrv_refresh_limits = qcow2_refresh_limits, 3178 .bdrv_invalidate_cache = qcow2_invalidate_cache, 3179 3180 .create_opts = &qcow2_create_opts, 3181 .bdrv_check = qcow2_check, 3182 .bdrv_amend_options = qcow2_amend_options, 3183 3184 .bdrv_detach_aio_context = qcow2_detach_aio_context, 3185 .bdrv_attach_aio_context = qcow2_attach_aio_context, 3186 }; 3187 3188 static void bdrv_qcow2_init(void) 3189 { 3190 bdrv_register(&bdrv_qcow2); 3191 } 3192 3193 block_init(bdrv_qcow2_init); 3194