1 /* 2 * Block driver for the QCOW version 2 format 3 * 4 * Copyright (c) 2004-2006 Fabrice Bellard 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 #include "qemu/osdep.h" 25 #include "block/block_int.h" 26 #include "sysemu/block-backend.h" 27 #include "qemu/module.h" 28 #include <zlib.h> 29 #include "block/qcow2.h" 30 #include "qemu/error-report.h" 31 #include "qapi/qmp/qerror.h" 32 #include "qapi/qmp/qbool.h" 33 #include "qapi/util.h" 34 #include "qapi/qmp/types.h" 35 #include "qapi-event.h" 36 #include "trace.h" 37 #include "qemu/option_int.h" 38 #include "qemu/cutils.h" 39 #include "qemu/bswap.h" 40 41 /* 42 Differences with QCOW: 43 44 - Support for multiple incremental snapshots. 45 - Memory management by reference counts. 46 - Clusters which have a reference count of one have the bit 47 QCOW_OFLAG_COPIED to optimize write performance. 48 - Size of compressed clusters is stored in sectors to reduce bit usage 49 in the cluster offsets. 50 - Support for storing additional data (such as the VM state) in the 51 snapshots. 52 - If a backing store is used, the cluster size is not constrained 53 (could be backported to QCOW). 54 - L2 tables have always a size of one cluster. 55 */ 56 57 58 typedef struct { 59 uint32_t magic; 60 uint32_t len; 61 } QEMU_PACKED QCowExtension; 62 63 #define QCOW2_EXT_MAGIC_END 0 64 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA 65 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 66 67 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) 68 { 69 const QCowHeader *cow_header = (const void *)buf; 70 71 if (buf_size >= sizeof(QCowHeader) && 72 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 73 be32_to_cpu(cow_header->version) >= 2) 74 return 100; 75 else 76 return 0; 77 } 78 79 80 /* 81 * read qcow2 extension and fill bs 82 * start reading from start_offset 83 * finish reading upon magic of value 0 or when end_offset reached 84 * unknown magic is skipped (future extension this version knows nothing about) 85 * return 0 upon success, non-0 otherwise 86 */ 87 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, 88 uint64_t end_offset, void **p_feature_table, 89 Error **errp) 90 { 91 BDRVQcow2State *s = bs->opaque; 92 QCowExtension ext; 93 uint64_t offset; 94 int ret; 95 96 #ifdef DEBUG_EXT 97 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); 98 #endif 99 offset = start_offset; 100 while (offset < end_offset) { 101 102 #ifdef DEBUG_EXT 103 /* Sanity check */ 104 if (offset > s->cluster_size) 105 printf("qcow2_read_extension: suspicious offset %lu\n", offset); 106 107 printf("attempting to read extended header in offset %lu\n", offset); 108 #endif 109 110 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext)); 111 if (ret < 0) { 112 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: " 113 "pread fail from offset %" PRIu64, offset); 114 return 1; 115 } 116 be32_to_cpus(&ext.magic); 117 be32_to_cpus(&ext.len); 118 offset += sizeof(ext); 119 #ifdef DEBUG_EXT 120 printf("ext.magic = 0x%x\n", ext.magic); 121 #endif 122 if (offset > end_offset || ext.len > end_offset - offset) { 123 error_setg(errp, "Header extension too large"); 124 return -EINVAL; 125 } 126 127 switch (ext.magic) { 128 case QCOW2_EXT_MAGIC_END: 129 return 0; 130 131 case QCOW2_EXT_MAGIC_BACKING_FORMAT: 132 if (ext.len >= sizeof(bs->backing_format)) { 133 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32 134 " too large (>=%zu)", ext.len, 135 sizeof(bs->backing_format)); 136 return 2; 137 } 138 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len); 139 if (ret < 0) { 140 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: " 141 "Could not read format name"); 142 return 3; 143 } 144 bs->backing_format[ext.len] = '\0'; 145 s->image_backing_format = g_strdup(bs->backing_format); 146 #ifdef DEBUG_EXT 147 printf("Qcow2: Got format extension %s\n", bs->backing_format); 148 #endif 149 break; 150 151 case QCOW2_EXT_MAGIC_FEATURE_TABLE: 152 if (p_feature_table != NULL) { 153 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); 154 ret = bdrv_pread(bs->file, offset , feature_table, ext.len); 155 if (ret < 0) { 156 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: " 157 "Could not read table"); 158 return ret; 159 } 160 161 *p_feature_table = feature_table; 162 } 163 break; 164 165 default: 166 /* unknown magic - save it in case we need to rewrite the header */ 167 { 168 Qcow2UnknownHeaderExtension *uext; 169 170 uext = g_malloc0(sizeof(*uext) + ext.len); 171 uext->magic = ext.magic; 172 uext->len = ext.len; 173 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); 174 175 ret = bdrv_pread(bs->file, offset , uext->data, uext->len); 176 if (ret < 0) { 177 error_setg_errno(errp, -ret, "ERROR: unknown extension: " 178 "Could not read data"); 179 return ret; 180 } 181 } 182 break; 183 } 184 185 offset += ((ext.len + 7) & ~7); 186 } 187 188 return 0; 189 } 190 191 static void cleanup_unknown_header_ext(BlockDriverState *bs) 192 { 193 BDRVQcow2State *s = bs->opaque; 194 Qcow2UnknownHeaderExtension *uext, *next; 195 196 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) { 197 QLIST_REMOVE(uext, next); 198 g_free(uext); 199 } 200 } 201 202 static void report_unsupported_feature(Error **errp, Qcow2Feature *table, 203 uint64_t mask) 204 { 205 char *features = g_strdup(""); 206 char *old; 207 208 while (table && table->name[0] != '\0') { 209 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) { 210 if (mask & (1ULL << table->bit)) { 211 old = features; 212 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "", 213 table->name); 214 g_free(old); 215 mask &= ~(1ULL << table->bit); 216 } 217 } 218 table++; 219 } 220 221 if (mask) { 222 old = features; 223 features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64, 224 old, *old ? ", " : "", mask); 225 g_free(old); 226 } 227 228 error_setg(errp, "Unsupported qcow2 feature(s): %s", features); 229 g_free(features); 230 } 231 232 /* 233 * Sets the dirty bit and flushes afterwards if necessary. 234 * 235 * The incompatible_features bit is only set if the image file header was 236 * updated successfully. Therefore it is not required to check the return 237 * value of this function. 238 */ 239 int qcow2_mark_dirty(BlockDriverState *bs) 240 { 241 BDRVQcow2State *s = bs->opaque; 242 uint64_t val; 243 int ret; 244 245 assert(s->qcow_version >= 3); 246 247 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 248 return 0; /* already dirty */ 249 } 250 251 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); 252 ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features), 253 &val, sizeof(val)); 254 if (ret < 0) { 255 return ret; 256 } 257 ret = bdrv_flush(bs->file->bs); 258 if (ret < 0) { 259 return ret; 260 } 261 262 /* Only treat image as dirty if the header was updated successfully */ 263 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; 264 return 0; 265 } 266 267 /* 268 * Clears the dirty bit and flushes before if necessary. Only call this 269 * function when there are no pending requests, it does not guard against 270 * concurrent requests dirtying the image. 271 */ 272 static int qcow2_mark_clean(BlockDriverState *bs) 273 { 274 BDRVQcow2State *s = bs->opaque; 275 276 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 277 int ret; 278 279 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; 280 281 ret = bdrv_flush(bs); 282 if (ret < 0) { 283 return ret; 284 } 285 286 return qcow2_update_header(bs); 287 } 288 return 0; 289 } 290 291 /* 292 * Marks the image as corrupt. 293 */ 294 int qcow2_mark_corrupt(BlockDriverState *bs) 295 { 296 BDRVQcow2State *s = bs->opaque; 297 298 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT; 299 return qcow2_update_header(bs); 300 } 301 302 /* 303 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes 304 * before if necessary. 305 */ 306 int qcow2_mark_consistent(BlockDriverState *bs) 307 { 308 BDRVQcow2State *s = bs->opaque; 309 310 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 311 int ret = bdrv_flush(bs); 312 if (ret < 0) { 313 return ret; 314 } 315 316 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT; 317 return qcow2_update_header(bs); 318 } 319 return 0; 320 } 321 322 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result, 323 BdrvCheckMode fix) 324 { 325 int ret = qcow2_check_refcounts(bs, result, fix); 326 if (ret < 0) { 327 return ret; 328 } 329 330 if (fix && result->check_errors == 0 && result->corruptions == 0) { 331 ret = qcow2_mark_clean(bs); 332 if (ret < 0) { 333 return ret; 334 } 335 return qcow2_mark_consistent(bs); 336 } 337 return ret; 338 } 339 340 static int validate_table_offset(BlockDriverState *bs, uint64_t offset, 341 uint64_t entries, size_t entry_len) 342 { 343 BDRVQcow2State *s = bs->opaque; 344 uint64_t size; 345 346 /* Use signed INT64_MAX as the maximum even for uint64_t header fields, 347 * because values will be passed to qemu functions taking int64_t. */ 348 if (entries > INT64_MAX / entry_len) { 349 return -EINVAL; 350 } 351 352 size = entries * entry_len; 353 354 if (INT64_MAX - size < offset) { 355 return -EINVAL; 356 } 357 358 /* Tables must be cluster aligned */ 359 if (offset & (s->cluster_size - 1)) { 360 return -EINVAL; 361 } 362 363 return 0; 364 } 365 366 static QemuOptsList qcow2_runtime_opts = { 367 .name = "qcow2", 368 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), 369 .desc = { 370 { 371 .name = QCOW2_OPT_LAZY_REFCOUNTS, 372 .type = QEMU_OPT_BOOL, 373 .help = "Postpone refcount updates", 374 }, 375 { 376 .name = QCOW2_OPT_DISCARD_REQUEST, 377 .type = QEMU_OPT_BOOL, 378 .help = "Pass guest discard requests to the layer below", 379 }, 380 { 381 .name = QCOW2_OPT_DISCARD_SNAPSHOT, 382 .type = QEMU_OPT_BOOL, 383 .help = "Generate discard requests when snapshot related space " 384 "is freed", 385 }, 386 { 387 .name = QCOW2_OPT_DISCARD_OTHER, 388 .type = QEMU_OPT_BOOL, 389 .help = "Generate discard requests when other clusters are freed", 390 }, 391 { 392 .name = QCOW2_OPT_OVERLAP, 393 .type = QEMU_OPT_STRING, 394 .help = "Selects which overlap checks to perform from a range of " 395 "templates (none, constant, cached, all)", 396 }, 397 { 398 .name = QCOW2_OPT_OVERLAP_TEMPLATE, 399 .type = QEMU_OPT_STRING, 400 .help = "Selects which overlap checks to perform from a range of " 401 "templates (none, constant, cached, all)", 402 }, 403 { 404 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER, 405 .type = QEMU_OPT_BOOL, 406 .help = "Check for unintended writes into the main qcow2 header", 407 }, 408 { 409 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1, 410 .type = QEMU_OPT_BOOL, 411 .help = "Check for unintended writes into the active L1 table", 412 }, 413 { 414 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2, 415 .type = QEMU_OPT_BOOL, 416 .help = "Check for unintended writes into an active L2 table", 417 }, 418 { 419 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 420 .type = QEMU_OPT_BOOL, 421 .help = "Check for unintended writes into the refcount table", 422 }, 423 { 424 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 425 .type = QEMU_OPT_BOOL, 426 .help = "Check for unintended writes into a refcount block", 427 }, 428 { 429 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 430 .type = QEMU_OPT_BOOL, 431 .help = "Check for unintended writes into the snapshot table", 432 }, 433 { 434 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1, 435 .type = QEMU_OPT_BOOL, 436 .help = "Check for unintended writes into an inactive L1 table", 437 }, 438 { 439 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2, 440 .type = QEMU_OPT_BOOL, 441 .help = "Check for unintended writes into an inactive L2 table", 442 }, 443 { 444 .name = QCOW2_OPT_CACHE_SIZE, 445 .type = QEMU_OPT_SIZE, 446 .help = "Maximum combined metadata (L2 tables and refcount blocks) " 447 "cache size", 448 }, 449 { 450 .name = QCOW2_OPT_L2_CACHE_SIZE, 451 .type = QEMU_OPT_SIZE, 452 .help = "Maximum L2 table cache size", 453 }, 454 { 455 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE, 456 .type = QEMU_OPT_SIZE, 457 .help = "Maximum refcount block cache size", 458 }, 459 { 460 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL, 461 .type = QEMU_OPT_NUMBER, 462 .help = "Clean unused cache entries after this time (in seconds)", 463 }, 464 { /* end of list */ } 465 }, 466 }; 467 468 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { 469 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, 470 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, 471 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, 472 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 473 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 474 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 475 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, 476 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, 477 }; 478 479 static void cache_clean_timer_cb(void *opaque) 480 { 481 BlockDriverState *bs = opaque; 482 BDRVQcow2State *s = bs->opaque; 483 qcow2_cache_clean_unused(bs, s->l2_table_cache); 484 qcow2_cache_clean_unused(bs, s->refcount_block_cache); 485 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 486 (int64_t) s->cache_clean_interval * 1000); 487 } 488 489 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context) 490 { 491 BDRVQcow2State *s = bs->opaque; 492 if (s->cache_clean_interval > 0) { 493 s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL, 494 SCALE_MS, cache_clean_timer_cb, 495 bs); 496 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 497 (int64_t) s->cache_clean_interval * 1000); 498 } 499 } 500 501 static void cache_clean_timer_del(BlockDriverState *bs) 502 { 503 BDRVQcow2State *s = bs->opaque; 504 if (s->cache_clean_timer) { 505 timer_del(s->cache_clean_timer); 506 timer_free(s->cache_clean_timer); 507 s->cache_clean_timer = NULL; 508 } 509 } 510 511 static void qcow2_detach_aio_context(BlockDriverState *bs) 512 { 513 cache_clean_timer_del(bs); 514 } 515 516 static void qcow2_attach_aio_context(BlockDriverState *bs, 517 AioContext *new_context) 518 { 519 cache_clean_timer_init(bs, new_context); 520 } 521 522 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts, 523 uint64_t *l2_cache_size, 524 uint64_t *refcount_cache_size, Error **errp) 525 { 526 BDRVQcow2State *s = bs->opaque; 527 uint64_t combined_cache_size; 528 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set; 529 530 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE); 531 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE); 532 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 533 534 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0); 535 *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0); 536 *refcount_cache_size = qemu_opt_get_size(opts, 537 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0); 538 539 if (combined_cache_size_set) { 540 if (l2_cache_size_set && refcount_cache_size_set) { 541 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE 542 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set " 543 "the same time"); 544 return; 545 } else if (*l2_cache_size > combined_cache_size) { 546 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed " 547 QCOW2_OPT_CACHE_SIZE); 548 return; 549 } else if (*refcount_cache_size > combined_cache_size) { 550 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed " 551 QCOW2_OPT_CACHE_SIZE); 552 return; 553 } 554 555 if (l2_cache_size_set) { 556 *refcount_cache_size = combined_cache_size - *l2_cache_size; 557 } else if (refcount_cache_size_set) { 558 *l2_cache_size = combined_cache_size - *refcount_cache_size; 559 } else { 560 *refcount_cache_size = combined_cache_size 561 / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1); 562 *l2_cache_size = combined_cache_size - *refcount_cache_size; 563 } 564 } else { 565 if (!l2_cache_size_set && !refcount_cache_size_set) { 566 *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE, 567 (uint64_t)DEFAULT_L2_CACHE_CLUSTERS 568 * s->cluster_size); 569 *refcount_cache_size = *l2_cache_size 570 / DEFAULT_L2_REFCOUNT_SIZE_RATIO; 571 } else if (!l2_cache_size_set) { 572 *l2_cache_size = *refcount_cache_size 573 * DEFAULT_L2_REFCOUNT_SIZE_RATIO; 574 } else if (!refcount_cache_size_set) { 575 *refcount_cache_size = *l2_cache_size 576 / DEFAULT_L2_REFCOUNT_SIZE_RATIO; 577 } 578 } 579 } 580 581 typedef struct Qcow2ReopenState { 582 Qcow2Cache *l2_table_cache; 583 Qcow2Cache *refcount_block_cache; 584 bool use_lazy_refcounts; 585 int overlap_check; 586 bool discard_passthrough[QCOW2_DISCARD_MAX]; 587 uint64_t cache_clean_interval; 588 } Qcow2ReopenState; 589 590 static int qcow2_update_options_prepare(BlockDriverState *bs, 591 Qcow2ReopenState *r, 592 QDict *options, int flags, 593 Error **errp) 594 { 595 BDRVQcow2State *s = bs->opaque; 596 QemuOpts *opts = NULL; 597 const char *opt_overlap_check, *opt_overlap_check_template; 598 int overlap_check_template = 0; 599 uint64_t l2_cache_size, refcount_cache_size; 600 int i; 601 Error *local_err = NULL; 602 int ret; 603 604 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); 605 qemu_opts_absorb_qdict(opts, options, &local_err); 606 if (local_err) { 607 error_propagate(errp, local_err); 608 ret = -EINVAL; 609 goto fail; 610 } 611 612 /* get L2 table/refcount block cache size from command line options */ 613 read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size, 614 &local_err); 615 if (local_err) { 616 error_propagate(errp, local_err); 617 ret = -EINVAL; 618 goto fail; 619 } 620 621 l2_cache_size /= s->cluster_size; 622 if (l2_cache_size < MIN_L2_CACHE_SIZE) { 623 l2_cache_size = MIN_L2_CACHE_SIZE; 624 } 625 if (l2_cache_size > INT_MAX) { 626 error_setg(errp, "L2 cache size too big"); 627 ret = -EINVAL; 628 goto fail; 629 } 630 631 refcount_cache_size /= s->cluster_size; 632 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) { 633 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE; 634 } 635 if (refcount_cache_size > INT_MAX) { 636 error_setg(errp, "Refcount cache size too big"); 637 ret = -EINVAL; 638 goto fail; 639 } 640 641 /* alloc new L2 table/refcount block cache, flush old one */ 642 if (s->l2_table_cache) { 643 ret = qcow2_cache_flush(bs, s->l2_table_cache); 644 if (ret) { 645 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache"); 646 goto fail; 647 } 648 } 649 650 if (s->refcount_block_cache) { 651 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 652 if (ret) { 653 error_setg_errno(errp, -ret, 654 "Failed to flush the refcount block cache"); 655 goto fail; 656 } 657 } 658 659 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size); 660 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size); 661 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) { 662 error_setg(errp, "Could not allocate metadata caches"); 663 ret = -ENOMEM; 664 goto fail; 665 } 666 667 /* New interval for cache cleanup timer */ 668 r->cache_clean_interval = 669 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL, 670 s->cache_clean_interval); 671 if (r->cache_clean_interval > UINT_MAX) { 672 error_setg(errp, "Cache clean interval too big"); 673 ret = -EINVAL; 674 goto fail; 675 } 676 677 /* lazy-refcounts; flush if going from enabled to disabled */ 678 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, 679 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); 680 if (r->use_lazy_refcounts && s->qcow_version < 3) { 681 error_setg(errp, "Lazy refcounts require a qcow2 image with at least " 682 "qemu 1.1 compatibility level"); 683 ret = -EINVAL; 684 goto fail; 685 } 686 687 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) { 688 ret = qcow2_mark_clean(bs); 689 if (ret < 0) { 690 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts"); 691 goto fail; 692 } 693 } 694 695 /* Overlap check options */ 696 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP); 697 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE); 698 if (opt_overlap_check_template && opt_overlap_check && 699 strcmp(opt_overlap_check_template, opt_overlap_check)) 700 { 701 error_setg(errp, "Conflicting values for qcow2 options '" 702 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE 703 "' ('%s')", opt_overlap_check, opt_overlap_check_template); 704 ret = -EINVAL; 705 goto fail; 706 } 707 if (!opt_overlap_check) { 708 opt_overlap_check = opt_overlap_check_template ?: "cached"; 709 } 710 711 if (!strcmp(opt_overlap_check, "none")) { 712 overlap_check_template = 0; 713 } else if (!strcmp(opt_overlap_check, "constant")) { 714 overlap_check_template = QCOW2_OL_CONSTANT; 715 } else if (!strcmp(opt_overlap_check, "cached")) { 716 overlap_check_template = QCOW2_OL_CACHED; 717 } else if (!strcmp(opt_overlap_check, "all")) { 718 overlap_check_template = QCOW2_OL_ALL; 719 } else { 720 error_setg(errp, "Unsupported value '%s' for qcow2 option " 721 "'overlap-check'. Allowed are any of the following: " 722 "none, constant, cached, all", opt_overlap_check); 723 ret = -EINVAL; 724 goto fail; 725 } 726 727 r->overlap_check = 0; 728 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { 729 /* overlap-check defines a template bitmask, but every flag may be 730 * overwritten through the associated boolean option */ 731 r->overlap_check |= 732 qemu_opt_get_bool(opts, overlap_bool_option_names[i], 733 overlap_check_template & (1 << i)) << i; 734 } 735 736 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false; 737 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true; 738 r->discard_passthrough[QCOW2_DISCARD_REQUEST] = 739 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, 740 flags & BDRV_O_UNMAP); 741 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = 742 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true); 743 r->discard_passthrough[QCOW2_DISCARD_OTHER] = 744 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false); 745 746 ret = 0; 747 fail: 748 qemu_opts_del(opts); 749 opts = NULL; 750 return ret; 751 } 752 753 static void qcow2_update_options_commit(BlockDriverState *bs, 754 Qcow2ReopenState *r) 755 { 756 BDRVQcow2State *s = bs->opaque; 757 int i; 758 759 if (s->l2_table_cache) { 760 qcow2_cache_destroy(bs, s->l2_table_cache); 761 } 762 if (s->refcount_block_cache) { 763 qcow2_cache_destroy(bs, s->refcount_block_cache); 764 } 765 s->l2_table_cache = r->l2_table_cache; 766 s->refcount_block_cache = r->refcount_block_cache; 767 768 s->overlap_check = r->overlap_check; 769 s->use_lazy_refcounts = r->use_lazy_refcounts; 770 771 for (i = 0; i < QCOW2_DISCARD_MAX; i++) { 772 s->discard_passthrough[i] = r->discard_passthrough[i]; 773 } 774 775 if (s->cache_clean_interval != r->cache_clean_interval) { 776 cache_clean_timer_del(bs); 777 s->cache_clean_interval = r->cache_clean_interval; 778 cache_clean_timer_init(bs, bdrv_get_aio_context(bs)); 779 } 780 } 781 782 static void qcow2_update_options_abort(BlockDriverState *bs, 783 Qcow2ReopenState *r) 784 { 785 if (r->l2_table_cache) { 786 qcow2_cache_destroy(bs, r->l2_table_cache); 787 } 788 if (r->refcount_block_cache) { 789 qcow2_cache_destroy(bs, r->refcount_block_cache); 790 } 791 } 792 793 static int qcow2_update_options(BlockDriverState *bs, QDict *options, 794 int flags, Error **errp) 795 { 796 Qcow2ReopenState r = {}; 797 int ret; 798 799 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp); 800 if (ret >= 0) { 801 qcow2_update_options_commit(bs, &r); 802 } else { 803 qcow2_update_options_abort(bs, &r); 804 } 805 806 return ret; 807 } 808 809 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, 810 Error **errp) 811 { 812 BDRVQcow2State *s = bs->opaque; 813 unsigned int len, i; 814 int ret = 0; 815 QCowHeader header; 816 Error *local_err = NULL; 817 uint64_t ext_end; 818 uint64_t l1_vm_state_index; 819 820 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 821 if (ret < 0) { 822 error_setg_errno(errp, -ret, "Could not read qcow2 header"); 823 goto fail; 824 } 825 be32_to_cpus(&header.magic); 826 be32_to_cpus(&header.version); 827 be64_to_cpus(&header.backing_file_offset); 828 be32_to_cpus(&header.backing_file_size); 829 be64_to_cpus(&header.size); 830 be32_to_cpus(&header.cluster_bits); 831 be32_to_cpus(&header.crypt_method); 832 be64_to_cpus(&header.l1_table_offset); 833 be32_to_cpus(&header.l1_size); 834 be64_to_cpus(&header.refcount_table_offset); 835 be32_to_cpus(&header.refcount_table_clusters); 836 be64_to_cpus(&header.snapshots_offset); 837 be32_to_cpus(&header.nb_snapshots); 838 839 if (header.magic != QCOW_MAGIC) { 840 error_setg(errp, "Image is not in qcow2 format"); 841 ret = -EINVAL; 842 goto fail; 843 } 844 if (header.version < 2 || header.version > 3) { 845 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version); 846 ret = -ENOTSUP; 847 goto fail; 848 } 849 850 s->qcow_version = header.version; 851 852 /* Initialise cluster size */ 853 if (header.cluster_bits < MIN_CLUSTER_BITS || 854 header.cluster_bits > MAX_CLUSTER_BITS) { 855 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, 856 header.cluster_bits); 857 ret = -EINVAL; 858 goto fail; 859 } 860 861 s->cluster_bits = header.cluster_bits; 862 s->cluster_size = 1 << s->cluster_bits; 863 s->cluster_sectors = 1 << (s->cluster_bits - 9); 864 865 /* Initialise version 3 header fields */ 866 if (header.version == 2) { 867 header.incompatible_features = 0; 868 header.compatible_features = 0; 869 header.autoclear_features = 0; 870 header.refcount_order = 4; 871 header.header_length = 72; 872 } else { 873 be64_to_cpus(&header.incompatible_features); 874 be64_to_cpus(&header.compatible_features); 875 be64_to_cpus(&header.autoclear_features); 876 be32_to_cpus(&header.refcount_order); 877 be32_to_cpus(&header.header_length); 878 879 if (header.header_length < 104) { 880 error_setg(errp, "qcow2 header too short"); 881 ret = -EINVAL; 882 goto fail; 883 } 884 } 885 886 if (header.header_length > s->cluster_size) { 887 error_setg(errp, "qcow2 header exceeds cluster size"); 888 ret = -EINVAL; 889 goto fail; 890 } 891 892 if (header.header_length > sizeof(header)) { 893 s->unknown_header_fields_size = header.header_length - sizeof(header); 894 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); 895 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields, 896 s->unknown_header_fields_size); 897 if (ret < 0) { 898 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header " 899 "fields"); 900 goto fail; 901 } 902 } 903 904 if (header.backing_file_offset > s->cluster_size) { 905 error_setg(errp, "Invalid backing file offset"); 906 ret = -EINVAL; 907 goto fail; 908 } 909 910 if (header.backing_file_offset) { 911 ext_end = header.backing_file_offset; 912 } else { 913 ext_end = 1 << header.cluster_bits; 914 } 915 916 /* Handle feature bits */ 917 s->incompatible_features = header.incompatible_features; 918 s->compatible_features = header.compatible_features; 919 s->autoclear_features = header.autoclear_features; 920 921 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) { 922 void *feature_table = NULL; 923 qcow2_read_extensions(bs, header.header_length, ext_end, 924 &feature_table, NULL); 925 report_unsupported_feature(errp, feature_table, 926 s->incompatible_features & 927 ~QCOW2_INCOMPAT_MASK); 928 ret = -ENOTSUP; 929 g_free(feature_table); 930 goto fail; 931 } 932 933 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 934 /* Corrupt images may not be written to unless they are being repaired 935 */ 936 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 937 error_setg(errp, "qcow2: Image is corrupt; cannot be opened " 938 "read/write"); 939 ret = -EACCES; 940 goto fail; 941 } 942 } 943 944 /* Check support for various header values */ 945 if (header.refcount_order > 6) { 946 error_setg(errp, "Reference count entry width too large; may not " 947 "exceed 64 bits"); 948 ret = -EINVAL; 949 goto fail; 950 } 951 s->refcount_order = header.refcount_order; 952 s->refcount_bits = 1 << s->refcount_order; 953 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 954 s->refcount_max += s->refcount_max - 1; 955 956 if (header.crypt_method > QCOW_CRYPT_AES) { 957 error_setg(errp, "Unsupported encryption method: %" PRIu32, 958 header.crypt_method); 959 ret = -EINVAL; 960 goto fail; 961 } 962 if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128, 963 QCRYPTO_CIPHER_MODE_CBC)) { 964 error_setg(errp, "AES cipher not available"); 965 ret = -EINVAL; 966 goto fail; 967 } 968 s->crypt_method_header = header.crypt_method; 969 if (s->crypt_method_header) { 970 if (bdrv_uses_whitelist() && 971 s->crypt_method_header == QCOW_CRYPT_AES) { 972 error_setg(errp, 973 "Use of AES-CBC encrypted qcow2 images is no longer " 974 "supported in system emulators"); 975 error_append_hint(errp, 976 "You can use 'qemu-img convert' to convert your " 977 "image to an alternative supported format, such " 978 "as unencrypted qcow2, or raw with the LUKS " 979 "format instead.\n"); 980 ret = -ENOSYS; 981 goto fail; 982 } 983 984 bs->encrypted = true; 985 } 986 987 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ 988 s->l2_size = 1 << s->l2_bits; 989 /* 2^(s->refcount_order - 3) is the refcount width in bytes */ 990 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3); 991 s->refcount_block_size = 1 << s->refcount_block_bits; 992 bs->total_sectors = header.size / 512; 993 s->csize_shift = (62 - (s->cluster_bits - 8)); 994 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; 995 s->cluster_offset_mask = (1LL << s->csize_shift) - 1; 996 997 s->refcount_table_offset = header.refcount_table_offset; 998 s->refcount_table_size = 999 header.refcount_table_clusters << (s->cluster_bits - 3); 1000 1001 if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) { 1002 error_setg(errp, "Reference count table too large"); 1003 ret = -EINVAL; 1004 goto fail; 1005 } 1006 1007 ret = validate_table_offset(bs, s->refcount_table_offset, 1008 s->refcount_table_size, sizeof(uint64_t)); 1009 if (ret < 0) { 1010 error_setg(errp, "Invalid reference count table offset"); 1011 goto fail; 1012 } 1013 1014 /* Snapshot table offset/length */ 1015 if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) { 1016 error_setg(errp, "Too many snapshots"); 1017 ret = -EINVAL; 1018 goto fail; 1019 } 1020 1021 ret = validate_table_offset(bs, header.snapshots_offset, 1022 header.nb_snapshots, 1023 sizeof(QCowSnapshotHeader)); 1024 if (ret < 0) { 1025 error_setg(errp, "Invalid snapshot table offset"); 1026 goto fail; 1027 } 1028 1029 /* read the level 1 table */ 1030 if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) { 1031 error_setg(errp, "Active L1 table too large"); 1032 ret = -EFBIG; 1033 goto fail; 1034 } 1035 s->l1_size = header.l1_size; 1036 1037 l1_vm_state_index = size_to_l1(s, header.size); 1038 if (l1_vm_state_index > INT_MAX) { 1039 error_setg(errp, "Image is too big"); 1040 ret = -EFBIG; 1041 goto fail; 1042 } 1043 s->l1_vm_state_index = l1_vm_state_index; 1044 1045 /* the L1 table must contain at least enough entries to put 1046 header.size bytes */ 1047 if (s->l1_size < s->l1_vm_state_index) { 1048 error_setg(errp, "L1 table is too small"); 1049 ret = -EINVAL; 1050 goto fail; 1051 } 1052 1053 ret = validate_table_offset(bs, header.l1_table_offset, 1054 header.l1_size, sizeof(uint64_t)); 1055 if (ret < 0) { 1056 error_setg(errp, "Invalid L1 table offset"); 1057 goto fail; 1058 } 1059 s->l1_table_offset = header.l1_table_offset; 1060 1061 1062 if (s->l1_size > 0) { 1063 s->l1_table = qemu_try_blockalign(bs->file->bs, 1064 align_offset(s->l1_size * sizeof(uint64_t), 512)); 1065 if (s->l1_table == NULL) { 1066 error_setg(errp, "Could not allocate L1 table"); 1067 ret = -ENOMEM; 1068 goto fail; 1069 } 1070 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 1071 s->l1_size * sizeof(uint64_t)); 1072 if (ret < 0) { 1073 error_setg_errno(errp, -ret, "Could not read L1 table"); 1074 goto fail; 1075 } 1076 for(i = 0;i < s->l1_size; i++) { 1077 be64_to_cpus(&s->l1_table[i]); 1078 } 1079 } 1080 1081 /* Parse driver-specific options */ 1082 ret = qcow2_update_options(bs, options, flags, errp); 1083 if (ret < 0) { 1084 goto fail; 1085 } 1086 1087 s->cluster_cache = g_malloc(s->cluster_size); 1088 /* one more sector for decompressed data alignment */ 1089 s->cluster_data = qemu_try_blockalign(bs->file->bs, QCOW_MAX_CRYPT_CLUSTERS 1090 * s->cluster_size + 512); 1091 if (s->cluster_data == NULL) { 1092 error_setg(errp, "Could not allocate temporary cluster buffer"); 1093 ret = -ENOMEM; 1094 goto fail; 1095 } 1096 1097 s->cluster_cache_offset = -1; 1098 s->flags = flags; 1099 1100 ret = qcow2_refcount_init(bs); 1101 if (ret != 0) { 1102 error_setg_errno(errp, -ret, "Could not initialize refcount handling"); 1103 goto fail; 1104 } 1105 1106 QLIST_INIT(&s->cluster_allocs); 1107 QTAILQ_INIT(&s->discards); 1108 1109 /* read qcow2 extensions */ 1110 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, 1111 &local_err)) { 1112 error_propagate(errp, local_err); 1113 ret = -EINVAL; 1114 goto fail; 1115 } 1116 1117 /* read the backing file name */ 1118 if (header.backing_file_offset != 0) { 1119 len = header.backing_file_size; 1120 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || 1121 len >= sizeof(bs->backing_file)) { 1122 error_setg(errp, "Backing file name too long"); 1123 ret = -EINVAL; 1124 goto fail; 1125 } 1126 ret = bdrv_pread(bs->file, header.backing_file_offset, 1127 bs->backing_file, len); 1128 if (ret < 0) { 1129 error_setg_errno(errp, -ret, "Could not read backing file name"); 1130 goto fail; 1131 } 1132 bs->backing_file[len] = '\0'; 1133 s->image_backing_file = g_strdup(bs->backing_file); 1134 } 1135 1136 /* Internal snapshots */ 1137 s->snapshots_offset = header.snapshots_offset; 1138 s->nb_snapshots = header.nb_snapshots; 1139 1140 ret = qcow2_read_snapshots(bs); 1141 if (ret < 0) { 1142 error_setg_errno(errp, -ret, "Could not read snapshots"); 1143 goto fail; 1144 } 1145 1146 /* Clear unknown autoclear feature bits */ 1147 if (!bs->read_only && !(flags & BDRV_O_INACTIVE) && s->autoclear_features) { 1148 s->autoclear_features = 0; 1149 ret = qcow2_update_header(bs); 1150 if (ret < 0) { 1151 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 1152 goto fail; 1153 } 1154 } 1155 1156 /* Initialise locks */ 1157 qemu_co_mutex_init(&s->lock); 1158 1159 /* Repair image if dirty */ 1160 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only && 1161 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { 1162 BdrvCheckResult result = {0}; 1163 1164 ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); 1165 if (ret < 0) { 1166 error_setg_errno(errp, -ret, "Could not repair dirty image"); 1167 goto fail; 1168 } 1169 } 1170 1171 #ifdef DEBUG_ALLOC 1172 { 1173 BdrvCheckResult result = {0}; 1174 qcow2_check_refcounts(bs, &result, 0); 1175 } 1176 #endif 1177 return ret; 1178 1179 fail: 1180 g_free(s->unknown_header_fields); 1181 cleanup_unknown_header_ext(bs); 1182 qcow2_free_snapshots(bs); 1183 qcow2_refcount_close(bs); 1184 qemu_vfree(s->l1_table); 1185 /* else pre-write overlap checks in cache_destroy may crash */ 1186 s->l1_table = NULL; 1187 cache_clean_timer_del(bs); 1188 if (s->l2_table_cache) { 1189 qcow2_cache_destroy(bs, s->l2_table_cache); 1190 } 1191 if (s->refcount_block_cache) { 1192 qcow2_cache_destroy(bs, s->refcount_block_cache); 1193 } 1194 g_free(s->cluster_cache); 1195 qemu_vfree(s->cluster_data); 1196 return ret; 1197 } 1198 1199 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) 1200 { 1201 BDRVQcow2State *s = bs->opaque; 1202 1203 if (bs->encrypted) { 1204 /* Encryption works on a sector granularity */ 1205 bs->bl.request_alignment = BDRV_SECTOR_SIZE; 1206 } 1207 bs->bl.pwrite_zeroes_alignment = s->cluster_size; 1208 } 1209 1210 static int qcow2_set_key(BlockDriverState *bs, const char *key) 1211 { 1212 BDRVQcow2State *s = bs->opaque; 1213 uint8_t keybuf[16]; 1214 int len, i; 1215 Error *err = NULL; 1216 1217 memset(keybuf, 0, 16); 1218 len = strlen(key); 1219 if (len > 16) 1220 len = 16; 1221 /* XXX: we could compress the chars to 7 bits to increase 1222 entropy */ 1223 for(i = 0;i < len;i++) { 1224 keybuf[i] = key[i]; 1225 } 1226 assert(bs->encrypted); 1227 1228 qcrypto_cipher_free(s->cipher); 1229 s->cipher = qcrypto_cipher_new( 1230 QCRYPTO_CIPHER_ALG_AES_128, 1231 QCRYPTO_CIPHER_MODE_CBC, 1232 keybuf, G_N_ELEMENTS(keybuf), 1233 &err); 1234 1235 if (!s->cipher) { 1236 /* XXX would be nice if errors in this method could 1237 * be properly propagate to the caller. Would need 1238 * the bdrv_set_key() API signature to be fixed. */ 1239 error_free(err); 1240 return -1; 1241 } 1242 return 0; 1243 } 1244 1245 static int qcow2_reopen_prepare(BDRVReopenState *state, 1246 BlockReopenQueue *queue, Error **errp) 1247 { 1248 Qcow2ReopenState *r; 1249 int ret; 1250 1251 r = g_new0(Qcow2ReopenState, 1); 1252 state->opaque = r; 1253 1254 ret = qcow2_update_options_prepare(state->bs, r, state->options, 1255 state->flags, errp); 1256 if (ret < 0) { 1257 goto fail; 1258 } 1259 1260 /* We need to write out any unwritten data if we reopen read-only. */ 1261 if ((state->flags & BDRV_O_RDWR) == 0) { 1262 ret = bdrv_flush(state->bs); 1263 if (ret < 0) { 1264 goto fail; 1265 } 1266 1267 ret = qcow2_mark_clean(state->bs); 1268 if (ret < 0) { 1269 goto fail; 1270 } 1271 } 1272 1273 return 0; 1274 1275 fail: 1276 qcow2_update_options_abort(state->bs, r); 1277 g_free(r); 1278 return ret; 1279 } 1280 1281 static void qcow2_reopen_commit(BDRVReopenState *state) 1282 { 1283 qcow2_update_options_commit(state->bs, state->opaque); 1284 g_free(state->opaque); 1285 } 1286 1287 static void qcow2_reopen_abort(BDRVReopenState *state) 1288 { 1289 qcow2_update_options_abort(state->bs, state->opaque); 1290 g_free(state->opaque); 1291 } 1292 1293 static void qcow2_join_options(QDict *options, QDict *old_options) 1294 { 1295 bool has_new_overlap_template = 1296 qdict_haskey(options, QCOW2_OPT_OVERLAP) || 1297 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE); 1298 bool has_new_total_cache_size = 1299 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE); 1300 bool has_all_cache_options; 1301 1302 /* New overlap template overrides all old overlap options */ 1303 if (has_new_overlap_template) { 1304 qdict_del(old_options, QCOW2_OPT_OVERLAP); 1305 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE); 1306 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER); 1307 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1); 1308 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2); 1309 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE); 1310 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK); 1311 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE); 1312 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1); 1313 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2); 1314 } 1315 1316 /* New total cache size overrides all old options */ 1317 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) { 1318 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE); 1319 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1320 } 1321 1322 qdict_join(options, old_options, false); 1323 1324 /* 1325 * If after merging all cache size options are set, an old total size is 1326 * overwritten. Do keep all options, however, if all three are new. The 1327 * resulting error message is what we want to happen. 1328 */ 1329 has_all_cache_options = 1330 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) || 1331 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) || 1332 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1333 1334 if (has_all_cache_options && !has_new_total_cache_size) { 1335 qdict_del(options, QCOW2_OPT_CACHE_SIZE); 1336 } 1337 } 1338 1339 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs, 1340 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file) 1341 { 1342 BDRVQcow2State *s = bs->opaque; 1343 uint64_t cluster_offset; 1344 int index_in_cluster, ret; 1345 unsigned int bytes; 1346 int64_t status = 0; 1347 1348 bytes = MIN(INT_MAX, nb_sectors * BDRV_SECTOR_SIZE); 1349 qemu_co_mutex_lock(&s->lock); 1350 ret = qcow2_get_cluster_offset(bs, sector_num << 9, &bytes, 1351 &cluster_offset); 1352 qemu_co_mutex_unlock(&s->lock); 1353 if (ret < 0) { 1354 return ret; 1355 } 1356 1357 *pnum = bytes >> BDRV_SECTOR_BITS; 1358 1359 if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED && 1360 !s->cipher) { 1361 index_in_cluster = sector_num & (s->cluster_sectors - 1); 1362 cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS); 1363 *file = bs->file->bs; 1364 status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset; 1365 } 1366 if (ret == QCOW2_CLUSTER_ZERO) { 1367 status |= BDRV_BLOCK_ZERO; 1368 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) { 1369 status |= BDRV_BLOCK_DATA; 1370 } 1371 return status; 1372 } 1373 1374 /* handle reading after the end of the backing file */ 1375 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov, 1376 int64_t offset, int bytes) 1377 { 1378 uint64_t bs_size = bs->total_sectors * BDRV_SECTOR_SIZE; 1379 int n1; 1380 1381 if ((offset + bytes) <= bs_size) { 1382 return bytes; 1383 } 1384 1385 if (offset >= bs_size) { 1386 n1 = 0; 1387 } else { 1388 n1 = bs_size - offset; 1389 } 1390 1391 qemu_iovec_memset(qiov, n1, 0, bytes - n1); 1392 1393 return n1; 1394 } 1395 1396 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset, 1397 uint64_t bytes, QEMUIOVector *qiov, 1398 int flags) 1399 { 1400 BDRVQcow2State *s = bs->opaque; 1401 int offset_in_cluster, n1; 1402 int ret; 1403 unsigned int cur_bytes; /* number of bytes in current iteration */ 1404 uint64_t cluster_offset = 0; 1405 uint64_t bytes_done = 0; 1406 QEMUIOVector hd_qiov; 1407 uint8_t *cluster_data = NULL; 1408 1409 qemu_iovec_init(&hd_qiov, qiov->niov); 1410 1411 qemu_co_mutex_lock(&s->lock); 1412 1413 while (bytes != 0) { 1414 1415 /* prepare next request */ 1416 cur_bytes = MIN(bytes, INT_MAX); 1417 if (s->cipher) { 1418 cur_bytes = MIN(cur_bytes, 1419 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1420 } 1421 1422 ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset); 1423 if (ret < 0) { 1424 goto fail; 1425 } 1426 1427 offset_in_cluster = offset_into_cluster(s, offset); 1428 1429 qemu_iovec_reset(&hd_qiov); 1430 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 1431 1432 switch (ret) { 1433 case QCOW2_CLUSTER_UNALLOCATED: 1434 1435 if (bs->backing) { 1436 /* read from the base image */ 1437 n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov, 1438 offset, cur_bytes); 1439 if (n1 > 0) { 1440 QEMUIOVector local_qiov; 1441 1442 qemu_iovec_init(&local_qiov, hd_qiov.niov); 1443 qemu_iovec_concat(&local_qiov, &hd_qiov, 0, n1); 1444 1445 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 1446 qemu_co_mutex_unlock(&s->lock); 1447 ret = bdrv_co_preadv(bs->backing, offset, n1, 1448 &local_qiov, 0); 1449 qemu_co_mutex_lock(&s->lock); 1450 1451 qemu_iovec_destroy(&local_qiov); 1452 1453 if (ret < 0) { 1454 goto fail; 1455 } 1456 } 1457 } else { 1458 /* Note: in this case, no need to wait */ 1459 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 1460 } 1461 break; 1462 1463 case QCOW2_CLUSTER_ZERO: 1464 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 1465 break; 1466 1467 case QCOW2_CLUSTER_COMPRESSED: 1468 /* add AIO support for compressed blocks ? */ 1469 ret = qcow2_decompress_cluster(bs, cluster_offset); 1470 if (ret < 0) { 1471 goto fail; 1472 } 1473 1474 qemu_iovec_from_buf(&hd_qiov, 0, 1475 s->cluster_cache + offset_in_cluster, 1476 cur_bytes); 1477 break; 1478 1479 case QCOW2_CLUSTER_NORMAL: 1480 if ((cluster_offset & 511) != 0) { 1481 ret = -EIO; 1482 goto fail; 1483 } 1484 1485 if (bs->encrypted) { 1486 assert(s->cipher); 1487 1488 /* 1489 * For encrypted images, read everything into a temporary 1490 * contiguous buffer on which the AES functions can work. 1491 */ 1492 if (!cluster_data) { 1493 cluster_data = 1494 qemu_try_blockalign(bs->file->bs, 1495 QCOW_MAX_CRYPT_CLUSTERS 1496 * s->cluster_size); 1497 if (cluster_data == NULL) { 1498 ret = -ENOMEM; 1499 goto fail; 1500 } 1501 } 1502 1503 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1504 qemu_iovec_reset(&hd_qiov); 1505 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 1506 } 1507 1508 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 1509 qemu_co_mutex_unlock(&s->lock); 1510 ret = bdrv_co_preadv(bs->file, 1511 cluster_offset + offset_in_cluster, 1512 cur_bytes, &hd_qiov, 0); 1513 qemu_co_mutex_lock(&s->lock); 1514 if (ret < 0) { 1515 goto fail; 1516 } 1517 if (bs->encrypted) { 1518 assert(s->cipher); 1519 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); 1520 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0); 1521 Error *err = NULL; 1522 if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS, 1523 cluster_data, cluster_data, 1524 cur_bytes >> BDRV_SECTOR_BITS, 1525 false, &err) < 0) { 1526 error_free(err); 1527 ret = -EIO; 1528 goto fail; 1529 } 1530 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes); 1531 } 1532 break; 1533 1534 default: 1535 g_assert_not_reached(); 1536 ret = -EIO; 1537 goto fail; 1538 } 1539 1540 bytes -= cur_bytes; 1541 offset += cur_bytes; 1542 bytes_done += cur_bytes; 1543 } 1544 ret = 0; 1545 1546 fail: 1547 qemu_co_mutex_unlock(&s->lock); 1548 1549 qemu_iovec_destroy(&hd_qiov); 1550 qemu_vfree(cluster_data); 1551 1552 return ret; 1553 } 1554 1555 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset, 1556 uint64_t bytes, QEMUIOVector *qiov, 1557 int flags) 1558 { 1559 BDRVQcow2State *s = bs->opaque; 1560 int offset_in_cluster; 1561 int ret; 1562 unsigned int cur_bytes; /* number of sectors in current iteration */ 1563 uint64_t cluster_offset; 1564 QEMUIOVector hd_qiov; 1565 uint64_t bytes_done = 0; 1566 uint8_t *cluster_data = NULL; 1567 QCowL2Meta *l2meta = NULL; 1568 1569 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes); 1570 1571 qemu_iovec_init(&hd_qiov, qiov->niov); 1572 1573 s->cluster_cache_offset = -1; /* disable compressed cache */ 1574 1575 qemu_co_mutex_lock(&s->lock); 1576 1577 while (bytes != 0) { 1578 1579 l2meta = NULL; 1580 1581 trace_qcow2_writev_start_part(qemu_coroutine_self()); 1582 offset_in_cluster = offset_into_cluster(s, offset); 1583 cur_bytes = MIN(bytes, INT_MAX); 1584 if (bs->encrypted) { 1585 cur_bytes = MIN(cur_bytes, 1586 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size 1587 - offset_in_cluster); 1588 } 1589 1590 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 1591 &cluster_offset, &l2meta); 1592 if (ret < 0) { 1593 goto fail; 1594 } 1595 1596 assert((cluster_offset & 511) == 0); 1597 1598 qemu_iovec_reset(&hd_qiov); 1599 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 1600 1601 if (bs->encrypted) { 1602 Error *err = NULL; 1603 assert(s->cipher); 1604 if (!cluster_data) { 1605 cluster_data = qemu_try_blockalign(bs->file->bs, 1606 QCOW_MAX_CRYPT_CLUSTERS 1607 * s->cluster_size); 1608 if (cluster_data == NULL) { 1609 ret = -ENOMEM; 1610 goto fail; 1611 } 1612 } 1613 1614 assert(hd_qiov.size <= 1615 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1616 qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size); 1617 1618 if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS, 1619 cluster_data, cluster_data, 1620 cur_bytes >>BDRV_SECTOR_BITS, 1621 true, &err) < 0) { 1622 error_free(err); 1623 ret = -EIO; 1624 goto fail; 1625 } 1626 1627 qemu_iovec_reset(&hd_qiov); 1628 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 1629 } 1630 1631 ret = qcow2_pre_write_overlap_check(bs, 0, 1632 cluster_offset + offset_in_cluster, cur_bytes); 1633 if (ret < 0) { 1634 goto fail; 1635 } 1636 1637 qemu_co_mutex_unlock(&s->lock); 1638 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 1639 trace_qcow2_writev_data(qemu_coroutine_self(), 1640 cluster_offset + offset_in_cluster); 1641 ret = bdrv_co_pwritev(bs->file, 1642 cluster_offset + offset_in_cluster, 1643 cur_bytes, &hd_qiov, 0); 1644 qemu_co_mutex_lock(&s->lock); 1645 if (ret < 0) { 1646 goto fail; 1647 } 1648 1649 while (l2meta != NULL) { 1650 QCowL2Meta *next; 1651 1652 ret = qcow2_alloc_cluster_link_l2(bs, l2meta); 1653 if (ret < 0) { 1654 goto fail; 1655 } 1656 1657 /* Take the request off the list of running requests */ 1658 if (l2meta->nb_clusters != 0) { 1659 QLIST_REMOVE(l2meta, next_in_flight); 1660 } 1661 1662 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1663 1664 next = l2meta->next; 1665 g_free(l2meta); 1666 l2meta = next; 1667 } 1668 1669 bytes -= cur_bytes; 1670 offset += cur_bytes; 1671 bytes_done += cur_bytes; 1672 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes); 1673 } 1674 ret = 0; 1675 1676 fail: 1677 qemu_co_mutex_unlock(&s->lock); 1678 1679 while (l2meta != NULL) { 1680 QCowL2Meta *next; 1681 1682 if (l2meta->nb_clusters != 0) { 1683 QLIST_REMOVE(l2meta, next_in_flight); 1684 } 1685 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1686 1687 next = l2meta->next; 1688 g_free(l2meta); 1689 l2meta = next; 1690 } 1691 1692 qemu_iovec_destroy(&hd_qiov); 1693 qemu_vfree(cluster_data); 1694 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 1695 1696 return ret; 1697 } 1698 1699 static int qcow2_inactivate(BlockDriverState *bs) 1700 { 1701 BDRVQcow2State *s = bs->opaque; 1702 int ret, result = 0; 1703 1704 ret = qcow2_cache_flush(bs, s->l2_table_cache); 1705 if (ret) { 1706 result = ret; 1707 error_report("Failed to flush the L2 table cache: %s", 1708 strerror(-ret)); 1709 } 1710 1711 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 1712 if (ret) { 1713 result = ret; 1714 error_report("Failed to flush the refcount block cache: %s", 1715 strerror(-ret)); 1716 } 1717 1718 if (result == 0) { 1719 qcow2_mark_clean(bs); 1720 } 1721 1722 return result; 1723 } 1724 1725 static void qcow2_close(BlockDriverState *bs) 1726 { 1727 BDRVQcow2State *s = bs->opaque; 1728 qemu_vfree(s->l1_table); 1729 /* else pre-write overlap checks in cache_destroy may crash */ 1730 s->l1_table = NULL; 1731 1732 if (!(s->flags & BDRV_O_INACTIVE)) { 1733 qcow2_inactivate(bs); 1734 } 1735 1736 cache_clean_timer_del(bs); 1737 qcow2_cache_destroy(bs, s->l2_table_cache); 1738 qcow2_cache_destroy(bs, s->refcount_block_cache); 1739 1740 qcrypto_cipher_free(s->cipher); 1741 s->cipher = NULL; 1742 1743 g_free(s->unknown_header_fields); 1744 cleanup_unknown_header_ext(bs); 1745 1746 g_free(s->image_backing_file); 1747 g_free(s->image_backing_format); 1748 1749 g_free(s->cluster_cache); 1750 qemu_vfree(s->cluster_data); 1751 qcow2_refcount_close(bs); 1752 qcow2_free_snapshots(bs); 1753 } 1754 1755 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp) 1756 { 1757 BDRVQcow2State *s = bs->opaque; 1758 int flags = s->flags; 1759 QCryptoCipher *cipher = NULL; 1760 QDict *options; 1761 Error *local_err = NULL; 1762 int ret; 1763 1764 /* 1765 * Backing files are read-only which makes all of their metadata immutable, 1766 * that means we don't have to worry about reopening them here. 1767 */ 1768 1769 cipher = s->cipher; 1770 s->cipher = NULL; 1771 1772 qcow2_close(bs); 1773 1774 memset(s, 0, sizeof(BDRVQcow2State)); 1775 options = qdict_clone_shallow(bs->options); 1776 1777 flags &= ~BDRV_O_INACTIVE; 1778 ret = qcow2_open(bs, options, flags, &local_err); 1779 QDECREF(options); 1780 if (local_err) { 1781 error_propagate(errp, local_err); 1782 error_prepend(errp, "Could not reopen qcow2 layer: "); 1783 bs->drv = NULL; 1784 return; 1785 } else if (ret < 0) { 1786 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer"); 1787 bs->drv = NULL; 1788 return; 1789 } 1790 1791 s->cipher = cipher; 1792 } 1793 1794 static size_t header_ext_add(char *buf, uint32_t magic, const void *s, 1795 size_t len, size_t buflen) 1796 { 1797 QCowExtension *ext_backing_fmt = (QCowExtension*) buf; 1798 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); 1799 1800 if (buflen < ext_len) { 1801 return -ENOSPC; 1802 } 1803 1804 *ext_backing_fmt = (QCowExtension) { 1805 .magic = cpu_to_be32(magic), 1806 .len = cpu_to_be32(len), 1807 }; 1808 1809 if (len) { 1810 memcpy(buf + sizeof(QCowExtension), s, len); 1811 } 1812 1813 return ext_len; 1814 } 1815 1816 /* 1817 * Updates the qcow2 header, including the variable length parts of it, i.e. 1818 * the backing file name and all extensions. qcow2 was not designed to allow 1819 * such changes, so if we run out of space (we can only use the first cluster) 1820 * this function may fail. 1821 * 1822 * Returns 0 on success, -errno in error cases. 1823 */ 1824 int qcow2_update_header(BlockDriverState *bs) 1825 { 1826 BDRVQcow2State *s = bs->opaque; 1827 QCowHeader *header; 1828 char *buf; 1829 size_t buflen = s->cluster_size; 1830 int ret; 1831 uint64_t total_size; 1832 uint32_t refcount_table_clusters; 1833 size_t header_length; 1834 Qcow2UnknownHeaderExtension *uext; 1835 1836 buf = qemu_blockalign(bs, buflen); 1837 1838 /* Header structure */ 1839 header = (QCowHeader*) buf; 1840 1841 if (buflen < sizeof(*header)) { 1842 ret = -ENOSPC; 1843 goto fail; 1844 } 1845 1846 header_length = sizeof(*header) + s->unknown_header_fields_size; 1847 total_size = bs->total_sectors * BDRV_SECTOR_SIZE; 1848 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); 1849 1850 *header = (QCowHeader) { 1851 /* Version 2 fields */ 1852 .magic = cpu_to_be32(QCOW_MAGIC), 1853 .version = cpu_to_be32(s->qcow_version), 1854 .backing_file_offset = 0, 1855 .backing_file_size = 0, 1856 .cluster_bits = cpu_to_be32(s->cluster_bits), 1857 .size = cpu_to_be64(total_size), 1858 .crypt_method = cpu_to_be32(s->crypt_method_header), 1859 .l1_size = cpu_to_be32(s->l1_size), 1860 .l1_table_offset = cpu_to_be64(s->l1_table_offset), 1861 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), 1862 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), 1863 .nb_snapshots = cpu_to_be32(s->nb_snapshots), 1864 .snapshots_offset = cpu_to_be64(s->snapshots_offset), 1865 1866 /* Version 3 fields */ 1867 .incompatible_features = cpu_to_be64(s->incompatible_features), 1868 .compatible_features = cpu_to_be64(s->compatible_features), 1869 .autoclear_features = cpu_to_be64(s->autoclear_features), 1870 .refcount_order = cpu_to_be32(s->refcount_order), 1871 .header_length = cpu_to_be32(header_length), 1872 }; 1873 1874 /* For older versions, write a shorter header */ 1875 switch (s->qcow_version) { 1876 case 2: 1877 ret = offsetof(QCowHeader, incompatible_features); 1878 break; 1879 case 3: 1880 ret = sizeof(*header); 1881 break; 1882 default: 1883 ret = -EINVAL; 1884 goto fail; 1885 } 1886 1887 buf += ret; 1888 buflen -= ret; 1889 memset(buf, 0, buflen); 1890 1891 /* Preserve any unknown field in the header */ 1892 if (s->unknown_header_fields_size) { 1893 if (buflen < s->unknown_header_fields_size) { 1894 ret = -ENOSPC; 1895 goto fail; 1896 } 1897 1898 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); 1899 buf += s->unknown_header_fields_size; 1900 buflen -= s->unknown_header_fields_size; 1901 } 1902 1903 /* Backing file format header extension */ 1904 if (s->image_backing_format) { 1905 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, 1906 s->image_backing_format, 1907 strlen(s->image_backing_format), 1908 buflen); 1909 if (ret < 0) { 1910 goto fail; 1911 } 1912 1913 buf += ret; 1914 buflen -= ret; 1915 } 1916 1917 /* Feature table */ 1918 if (s->qcow_version >= 3) { 1919 Qcow2Feature features[] = { 1920 { 1921 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 1922 .bit = QCOW2_INCOMPAT_DIRTY_BITNR, 1923 .name = "dirty bit", 1924 }, 1925 { 1926 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 1927 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, 1928 .name = "corrupt bit", 1929 }, 1930 { 1931 .type = QCOW2_FEAT_TYPE_COMPATIBLE, 1932 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 1933 .name = "lazy refcounts", 1934 }, 1935 }; 1936 1937 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, 1938 features, sizeof(features), buflen); 1939 if (ret < 0) { 1940 goto fail; 1941 } 1942 buf += ret; 1943 buflen -= ret; 1944 } 1945 1946 /* Keep unknown header extensions */ 1947 QLIST_FOREACH(uext, &s->unknown_header_ext, next) { 1948 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); 1949 if (ret < 0) { 1950 goto fail; 1951 } 1952 1953 buf += ret; 1954 buflen -= ret; 1955 } 1956 1957 /* End of header extensions */ 1958 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); 1959 if (ret < 0) { 1960 goto fail; 1961 } 1962 1963 buf += ret; 1964 buflen -= ret; 1965 1966 /* Backing file name */ 1967 if (s->image_backing_file) { 1968 size_t backing_file_len = strlen(s->image_backing_file); 1969 1970 if (buflen < backing_file_len) { 1971 ret = -ENOSPC; 1972 goto fail; 1973 } 1974 1975 /* Using strncpy is ok here, since buf is not NUL-terminated. */ 1976 strncpy(buf, s->image_backing_file, buflen); 1977 1978 header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); 1979 header->backing_file_size = cpu_to_be32(backing_file_len); 1980 } 1981 1982 /* Write the new header */ 1983 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size); 1984 if (ret < 0) { 1985 goto fail; 1986 } 1987 1988 ret = 0; 1989 fail: 1990 qemu_vfree(header); 1991 return ret; 1992 } 1993 1994 static int qcow2_change_backing_file(BlockDriverState *bs, 1995 const char *backing_file, const char *backing_fmt) 1996 { 1997 BDRVQcow2State *s = bs->opaque; 1998 1999 if (backing_file && strlen(backing_file) > 1023) { 2000 return -EINVAL; 2001 } 2002 2003 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 2004 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 2005 2006 g_free(s->image_backing_file); 2007 g_free(s->image_backing_format); 2008 2009 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL; 2010 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL; 2011 2012 return qcow2_update_header(bs); 2013 } 2014 2015 static int preallocate(BlockDriverState *bs) 2016 { 2017 uint64_t bytes; 2018 uint64_t offset; 2019 uint64_t host_offset = 0; 2020 unsigned int cur_bytes; 2021 int ret; 2022 QCowL2Meta *meta; 2023 2024 bytes = bdrv_getlength(bs); 2025 offset = 0; 2026 2027 while (bytes) { 2028 cur_bytes = MIN(bytes, INT_MAX); 2029 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 2030 &host_offset, &meta); 2031 if (ret < 0) { 2032 return ret; 2033 } 2034 2035 while (meta) { 2036 QCowL2Meta *next = meta->next; 2037 2038 ret = qcow2_alloc_cluster_link_l2(bs, meta); 2039 if (ret < 0) { 2040 qcow2_free_any_clusters(bs, meta->alloc_offset, 2041 meta->nb_clusters, QCOW2_DISCARD_NEVER); 2042 return ret; 2043 } 2044 2045 /* There are no dependent requests, but we need to remove our 2046 * request from the list of in-flight requests */ 2047 QLIST_REMOVE(meta, next_in_flight); 2048 2049 g_free(meta); 2050 meta = next; 2051 } 2052 2053 /* TODO Preallocate data if requested */ 2054 2055 bytes -= cur_bytes; 2056 offset += cur_bytes; 2057 } 2058 2059 /* 2060 * It is expected that the image file is large enough to actually contain 2061 * all of the allocated clusters (otherwise we get failing reads after 2062 * EOF). Extend the image to the last allocated sector. 2063 */ 2064 if (host_offset != 0) { 2065 uint8_t data = 0; 2066 ret = bdrv_pwrite(bs->file, (host_offset + cur_bytes) - 1, 2067 &data, 1); 2068 if (ret < 0) { 2069 return ret; 2070 } 2071 } 2072 2073 return 0; 2074 } 2075 2076 static int qcow2_create2(const char *filename, int64_t total_size, 2077 const char *backing_file, const char *backing_format, 2078 int flags, size_t cluster_size, PreallocMode prealloc, 2079 QemuOpts *opts, int version, int refcount_order, 2080 Error **errp) 2081 { 2082 int cluster_bits; 2083 QDict *options; 2084 2085 /* Calculate cluster_bits */ 2086 cluster_bits = ctz32(cluster_size); 2087 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 2088 (1 << cluster_bits) != cluster_size) 2089 { 2090 error_setg(errp, "Cluster size must be a power of two between %d and " 2091 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 2092 return -EINVAL; 2093 } 2094 2095 /* 2096 * Open the image file and write a minimal qcow2 header. 2097 * 2098 * We keep things simple and start with a zero-sized image. We also 2099 * do without refcount blocks or a L1 table for now. We'll fix the 2100 * inconsistency later. 2101 * 2102 * We do need a refcount table because growing the refcount table means 2103 * allocating two new refcount blocks - the seconds of which would be at 2104 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 2105 * size for any qcow2 image. 2106 */ 2107 BlockBackend *blk; 2108 QCowHeader *header; 2109 uint64_t* refcount_table; 2110 Error *local_err = NULL; 2111 int ret; 2112 2113 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 2114 /* Note: The following calculation does not need to be exact; if it is a 2115 * bit off, either some bytes will be "leaked" (which is fine) or we 2116 * will need to increase the file size by some bytes (which is fine, 2117 * too, as long as the bulk is allocated here). Therefore, using 2118 * floating point arithmetic is fine. */ 2119 int64_t meta_size = 0; 2120 uint64_t nreftablee, nrefblocke, nl1e, nl2e; 2121 int64_t aligned_total_size = align_offset(total_size, cluster_size); 2122 int refblock_bits, refblock_size; 2123 /* refcount entry size in bytes */ 2124 double rces = (1 << refcount_order) / 8.; 2125 2126 /* see qcow2_open() */ 2127 refblock_bits = cluster_bits - (refcount_order - 3); 2128 refblock_size = 1 << refblock_bits; 2129 2130 /* header: 1 cluster */ 2131 meta_size += cluster_size; 2132 2133 /* total size of L2 tables */ 2134 nl2e = aligned_total_size / cluster_size; 2135 nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t)); 2136 meta_size += nl2e * sizeof(uint64_t); 2137 2138 /* total size of L1 tables */ 2139 nl1e = nl2e * sizeof(uint64_t) / cluster_size; 2140 nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t)); 2141 meta_size += nl1e * sizeof(uint64_t); 2142 2143 /* total size of refcount blocks 2144 * 2145 * note: every host cluster is reference-counted, including metadata 2146 * (even refcount blocks are recursively included). 2147 * Let: 2148 * a = total_size (this is the guest disk size) 2149 * m = meta size not including refcount blocks and refcount tables 2150 * c = cluster size 2151 * y1 = number of refcount blocks entries 2152 * y2 = meta size including everything 2153 * rces = refcount entry size in bytes 2154 * then, 2155 * y1 = (y2 + a)/c 2156 * y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m 2157 * we can get y1: 2158 * y1 = (a + m) / (c - rces - rces * sizeof(u64) / c) 2159 */ 2160 nrefblocke = (aligned_total_size + meta_size + cluster_size) 2161 / (cluster_size - rces - rces * sizeof(uint64_t) 2162 / cluster_size); 2163 meta_size += DIV_ROUND_UP(nrefblocke, refblock_size) * cluster_size; 2164 2165 /* total size of refcount tables */ 2166 nreftablee = nrefblocke / refblock_size; 2167 nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t)); 2168 meta_size += nreftablee * sizeof(uint64_t); 2169 2170 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, 2171 aligned_total_size + meta_size, &error_abort); 2172 qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc], 2173 &error_abort); 2174 } 2175 2176 ret = bdrv_create_file(filename, opts, &local_err); 2177 if (ret < 0) { 2178 error_propagate(errp, local_err); 2179 return ret; 2180 } 2181 2182 blk = blk_new_open(filename, NULL, NULL, 2183 BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err); 2184 if (blk == NULL) { 2185 error_propagate(errp, local_err); 2186 return -EIO; 2187 } 2188 2189 blk_set_allow_write_beyond_eof(blk, true); 2190 2191 /* Write the header */ 2192 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); 2193 header = g_malloc0(cluster_size); 2194 *header = (QCowHeader) { 2195 .magic = cpu_to_be32(QCOW_MAGIC), 2196 .version = cpu_to_be32(version), 2197 .cluster_bits = cpu_to_be32(cluster_bits), 2198 .size = cpu_to_be64(0), 2199 .l1_table_offset = cpu_to_be64(0), 2200 .l1_size = cpu_to_be32(0), 2201 .refcount_table_offset = cpu_to_be64(cluster_size), 2202 .refcount_table_clusters = cpu_to_be32(1), 2203 .refcount_order = cpu_to_be32(refcount_order), 2204 .header_length = cpu_to_be32(sizeof(*header)), 2205 }; 2206 2207 if (flags & BLOCK_FLAG_ENCRYPT) { 2208 header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES); 2209 } else { 2210 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 2211 } 2212 2213 if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) { 2214 header->compatible_features |= 2215 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); 2216 } 2217 2218 ret = blk_pwrite(blk, 0, header, cluster_size, 0); 2219 g_free(header); 2220 if (ret < 0) { 2221 error_setg_errno(errp, -ret, "Could not write qcow2 header"); 2222 goto out; 2223 } 2224 2225 /* Write a refcount table with one refcount block */ 2226 refcount_table = g_malloc0(2 * cluster_size); 2227 refcount_table[0] = cpu_to_be64(2 * cluster_size); 2228 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0); 2229 g_free(refcount_table); 2230 2231 if (ret < 0) { 2232 error_setg_errno(errp, -ret, "Could not write refcount table"); 2233 goto out; 2234 } 2235 2236 blk_unref(blk); 2237 blk = NULL; 2238 2239 /* 2240 * And now open the image and make it consistent first (i.e. increase the 2241 * refcount of the cluster that is occupied by the header and the refcount 2242 * table) 2243 */ 2244 options = qdict_new(); 2245 qdict_put(options, "driver", qstring_from_str("qcow2")); 2246 blk = blk_new_open(filename, NULL, options, 2247 BDRV_O_RDWR | BDRV_O_NO_FLUSH, &local_err); 2248 if (blk == NULL) { 2249 error_propagate(errp, local_err); 2250 ret = -EIO; 2251 goto out; 2252 } 2253 2254 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size); 2255 if (ret < 0) { 2256 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " 2257 "header and refcount table"); 2258 goto out; 2259 2260 } else if (ret != 0) { 2261 error_report("Huh, first cluster in empty image is already in use?"); 2262 abort(); 2263 } 2264 2265 /* Create a full header (including things like feature table) */ 2266 ret = qcow2_update_header(blk_bs(blk)); 2267 if (ret < 0) { 2268 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 2269 goto out; 2270 } 2271 2272 /* Okay, now that we have a valid image, let's give it the right size */ 2273 ret = blk_truncate(blk, total_size); 2274 if (ret < 0) { 2275 error_setg_errno(errp, -ret, "Could not resize image"); 2276 goto out; 2277 } 2278 2279 /* Want a backing file? There you go.*/ 2280 if (backing_file) { 2281 ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format); 2282 if (ret < 0) { 2283 error_setg_errno(errp, -ret, "Could not assign backing file '%s' " 2284 "with format '%s'", backing_file, backing_format); 2285 goto out; 2286 } 2287 } 2288 2289 /* And if we're supposed to preallocate metadata, do that now */ 2290 if (prealloc != PREALLOC_MODE_OFF) { 2291 BDRVQcow2State *s = blk_bs(blk)->opaque; 2292 qemu_co_mutex_lock(&s->lock); 2293 ret = preallocate(blk_bs(blk)); 2294 qemu_co_mutex_unlock(&s->lock); 2295 if (ret < 0) { 2296 error_setg_errno(errp, -ret, "Could not preallocate metadata"); 2297 goto out; 2298 } 2299 } 2300 2301 blk_unref(blk); 2302 blk = NULL; 2303 2304 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */ 2305 options = qdict_new(); 2306 qdict_put(options, "driver", qstring_from_str("qcow2")); 2307 blk = blk_new_open(filename, NULL, options, 2308 BDRV_O_RDWR | BDRV_O_NO_BACKING, &local_err); 2309 if (blk == NULL) { 2310 error_propagate(errp, local_err); 2311 ret = -EIO; 2312 goto out; 2313 } 2314 2315 ret = 0; 2316 out: 2317 if (blk) { 2318 blk_unref(blk); 2319 } 2320 return ret; 2321 } 2322 2323 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp) 2324 { 2325 char *backing_file = NULL; 2326 char *backing_fmt = NULL; 2327 char *buf = NULL; 2328 uint64_t size = 0; 2329 int flags = 0; 2330 size_t cluster_size = DEFAULT_CLUSTER_SIZE; 2331 PreallocMode prealloc; 2332 int version = 3; 2333 uint64_t refcount_bits = 16; 2334 int refcount_order; 2335 Error *local_err = NULL; 2336 int ret; 2337 2338 /* Read out options */ 2339 size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2340 BDRV_SECTOR_SIZE); 2341 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 2342 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT); 2343 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) { 2344 flags |= BLOCK_FLAG_ENCRYPT; 2345 } 2346 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 2347 DEFAULT_CLUSTER_SIZE); 2348 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2349 prealloc = qapi_enum_parse(PreallocMode_lookup, buf, 2350 PREALLOC_MODE__MAX, PREALLOC_MODE_OFF, 2351 &local_err); 2352 if (local_err) { 2353 error_propagate(errp, local_err); 2354 ret = -EINVAL; 2355 goto finish; 2356 } 2357 g_free(buf); 2358 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); 2359 if (!buf) { 2360 /* keep the default */ 2361 } else if (!strcmp(buf, "0.10")) { 2362 version = 2; 2363 } else if (!strcmp(buf, "1.1")) { 2364 version = 3; 2365 } else { 2366 error_setg(errp, "Invalid compatibility level: '%s'", buf); 2367 ret = -EINVAL; 2368 goto finish; 2369 } 2370 2371 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) { 2372 flags |= BLOCK_FLAG_LAZY_REFCOUNTS; 2373 } 2374 2375 if (backing_file && prealloc != PREALLOC_MODE_OFF) { 2376 error_setg(errp, "Backing file and preallocation cannot be used at " 2377 "the same time"); 2378 ret = -EINVAL; 2379 goto finish; 2380 } 2381 2382 if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) { 2383 error_setg(errp, "Lazy refcounts only supported with compatibility " 2384 "level 1.1 and above (use compat=1.1 or greater)"); 2385 ret = -EINVAL; 2386 goto finish; 2387 } 2388 2389 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 2390 refcount_bits); 2391 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) { 2392 error_setg(errp, "Refcount width must be a power of two and may not " 2393 "exceed 64 bits"); 2394 ret = -EINVAL; 2395 goto finish; 2396 } 2397 2398 if (version < 3 && refcount_bits != 16) { 2399 error_setg(errp, "Different refcount widths than 16 bits require " 2400 "compatibility level 1.1 or above (use compat=1.1 or " 2401 "greater)"); 2402 ret = -EINVAL; 2403 goto finish; 2404 } 2405 2406 refcount_order = ctz32(refcount_bits); 2407 2408 ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags, 2409 cluster_size, prealloc, opts, version, refcount_order, 2410 &local_err); 2411 error_propagate(errp, local_err); 2412 2413 finish: 2414 g_free(backing_file); 2415 g_free(backing_fmt); 2416 g_free(buf); 2417 return ret; 2418 } 2419 2420 2421 static bool is_zero_sectors(BlockDriverState *bs, int64_t start, 2422 uint32_t count) 2423 { 2424 int nr; 2425 BlockDriverState *file; 2426 int64_t res; 2427 2428 if (!count) { 2429 return true; 2430 } 2431 res = bdrv_get_block_status_above(bs, NULL, start, count, 2432 &nr, &file); 2433 return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == count; 2434 } 2435 2436 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs, 2437 int64_t offset, int count, BdrvRequestFlags flags) 2438 { 2439 int ret; 2440 BDRVQcow2State *s = bs->opaque; 2441 2442 uint32_t head = offset % s->cluster_size; 2443 uint32_t tail = (offset + count) % s->cluster_size; 2444 2445 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, count); 2446 2447 if (head || tail) { 2448 int64_t cl_start = (offset - head) >> BDRV_SECTOR_BITS; 2449 uint64_t off; 2450 unsigned int nr; 2451 2452 assert(head + count <= s->cluster_size); 2453 2454 /* check whether remainder of cluster already reads as zero */ 2455 if (!(is_zero_sectors(bs, cl_start, 2456 DIV_ROUND_UP(head, BDRV_SECTOR_SIZE)) && 2457 is_zero_sectors(bs, (offset + count) >> BDRV_SECTOR_BITS, 2458 DIV_ROUND_UP(-tail & (s->cluster_size - 1), 2459 BDRV_SECTOR_SIZE)))) { 2460 return -ENOTSUP; 2461 } 2462 2463 qemu_co_mutex_lock(&s->lock); 2464 /* We can have new write after previous check */ 2465 offset = cl_start << BDRV_SECTOR_BITS; 2466 count = s->cluster_size; 2467 nr = s->cluster_size; 2468 ret = qcow2_get_cluster_offset(bs, offset, &nr, &off); 2469 if (ret != QCOW2_CLUSTER_UNALLOCATED && ret != QCOW2_CLUSTER_ZERO) { 2470 qemu_co_mutex_unlock(&s->lock); 2471 return -ENOTSUP; 2472 } 2473 } else { 2474 qemu_co_mutex_lock(&s->lock); 2475 } 2476 2477 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, count); 2478 2479 /* Whatever is left can use real zero clusters */ 2480 ret = qcow2_zero_clusters(bs, offset, count >> BDRV_SECTOR_BITS); 2481 qemu_co_mutex_unlock(&s->lock); 2482 2483 return ret; 2484 } 2485 2486 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs, 2487 int64_t offset, int count) 2488 { 2489 int ret; 2490 BDRVQcow2State *s = bs->opaque; 2491 2492 qemu_co_mutex_lock(&s->lock); 2493 ret = qcow2_discard_clusters(bs, offset, count >> BDRV_SECTOR_BITS, 2494 QCOW2_DISCARD_REQUEST, false); 2495 qemu_co_mutex_unlock(&s->lock); 2496 return ret; 2497 } 2498 2499 static int qcow2_truncate(BlockDriverState *bs, int64_t offset) 2500 { 2501 BDRVQcow2State *s = bs->opaque; 2502 int64_t new_l1_size; 2503 int ret; 2504 2505 if (offset & 511) { 2506 error_report("The new size must be a multiple of 512"); 2507 return -EINVAL; 2508 } 2509 2510 /* cannot proceed if image has snapshots */ 2511 if (s->nb_snapshots) { 2512 error_report("Can't resize an image which has snapshots"); 2513 return -ENOTSUP; 2514 } 2515 2516 /* shrinking is currently not supported */ 2517 if (offset < bs->total_sectors * 512) { 2518 error_report("qcow2 doesn't support shrinking images yet"); 2519 return -ENOTSUP; 2520 } 2521 2522 new_l1_size = size_to_l1(s, offset); 2523 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 2524 if (ret < 0) { 2525 return ret; 2526 } 2527 2528 /* write updated header.size */ 2529 offset = cpu_to_be64(offset); 2530 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), 2531 &offset, sizeof(uint64_t)); 2532 if (ret < 0) { 2533 return ret; 2534 } 2535 2536 s->l1_vm_state_index = new_l1_size; 2537 return 0; 2538 } 2539 2540 /* XXX: put compressed sectors first, then all the cluster aligned 2541 tables to avoid losing bytes in alignment */ 2542 static coroutine_fn int 2543 qcow2_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset, 2544 uint64_t bytes, QEMUIOVector *qiov) 2545 { 2546 BDRVQcow2State *s = bs->opaque; 2547 QEMUIOVector hd_qiov; 2548 struct iovec iov; 2549 z_stream strm; 2550 int ret, out_len; 2551 uint8_t *buf, *out_buf; 2552 uint64_t cluster_offset; 2553 2554 if (bytes == 0) { 2555 /* align end of file to a sector boundary to ease reading with 2556 sector based I/Os */ 2557 cluster_offset = bdrv_getlength(bs->file->bs); 2558 return bdrv_truncate(bs->file->bs, cluster_offset); 2559 } 2560 2561 buf = qemu_blockalign(bs, s->cluster_size); 2562 if (bytes != s->cluster_size) { 2563 if (bytes > s->cluster_size || 2564 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS) 2565 { 2566 qemu_vfree(buf); 2567 return -EINVAL; 2568 } 2569 /* Zero-pad last write if image size is not cluster aligned */ 2570 memset(buf + bytes, 0, s->cluster_size - bytes); 2571 } 2572 qemu_iovec_to_buf(qiov, 0, buf, bytes); 2573 2574 out_buf = g_malloc(s->cluster_size); 2575 2576 /* best compression, small window, no zlib header */ 2577 memset(&strm, 0, sizeof(strm)); 2578 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, 2579 Z_DEFLATED, -12, 2580 9, Z_DEFAULT_STRATEGY); 2581 if (ret != 0) { 2582 ret = -EINVAL; 2583 goto fail; 2584 } 2585 2586 strm.avail_in = s->cluster_size; 2587 strm.next_in = (uint8_t *)buf; 2588 strm.avail_out = s->cluster_size; 2589 strm.next_out = out_buf; 2590 2591 ret = deflate(&strm, Z_FINISH); 2592 if (ret != Z_STREAM_END && ret != Z_OK) { 2593 deflateEnd(&strm); 2594 ret = -EINVAL; 2595 goto fail; 2596 } 2597 out_len = strm.next_out - out_buf; 2598 2599 deflateEnd(&strm); 2600 2601 if (ret != Z_STREAM_END || out_len >= s->cluster_size) { 2602 /* could not compress: write normal cluster */ 2603 ret = qcow2_co_pwritev(bs, offset, bytes, qiov, 0); 2604 if (ret < 0) { 2605 goto fail; 2606 } 2607 goto success; 2608 } 2609 2610 qemu_co_mutex_lock(&s->lock); 2611 cluster_offset = 2612 qcow2_alloc_compressed_cluster_offset(bs, offset, out_len); 2613 if (!cluster_offset) { 2614 qemu_co_mutex_unlock(&s->lock); 2615 ret = -EIO; 2616 goto fail; 2617 } 2618 cluster_offset &= s->cluster_offset_mask; 2619 2620 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len); 2621 qemu_co_mutex_unlock(&s->lock); 2622 if (ret < 0) { 2623 goto fail; 2624 } 2625 2626 iov = (struct iovec) { 2627 .iov_base = out_buf, 2628 .iov_len = out_len, 2629 }; 2630 qemu_iovec_init_external(&hd_qiov, &iov, 1); 2631 2632 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); 2633 ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0); 2634 if (ret < 0) { 2635 goto fail; 2636 } 2637 success: 2638 ret = 0; 2639 fail: 2640 qemu_vfree(buf); 2641 g_free(out_buf); 2642 return ret; 2643 } 2644 2645 static int make_completely_empty(BlockDriverState *bs) 2646 { 2647 BDRVQcow2State *s = bs->opaque; 2648 int ret, l1_clusters; 2649 int64_t offset; 2650 uint64_t *new_reftable = NULL; 2651 uint64_t rt_entry, l1_size2; 2652 struct { 2653 uint64_t l1_offset; 2654 uint64_t reftable_offset; 2655 uint32_t reftable_clusters; 2656 } QEMU_PACKED l1_ofs_rt_ofs_cls; 2657 2658 ret = qcow2_cache_empty(bs, s->l2_table_cache); 2659 if (ret < 0) { 2660 goto fail; 2661 } 2662 2663 ret = qcow2_cache_empty(bs, s->refcount_block_cache); 2664 if (ret < 0) { 2665 goto fail; 2666 } 2667 2668 /* Refcounts will be broken utterly */ 2669 ret = qcow2_mark_dirty(bs); 2670 if (ret < 0) { 2671 goto fail; 2672 } 2673 2674 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 2675 2676 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 2677 l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t); 2678 2679 /* After this call, neither the in-memory nor the on-disk refcount 2680 * information accurately describe the actual references */ 2681 2682 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset, 2683 l1_clusters * s->cluster_size, 0); 2684 if (ret < 0) { 2685 goto fail_broken_refcounts; 2686 } 2687 memset(s->l1_table, 0, l1_size2); 2688 2689 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); 2690 2691 /* Overwrite enough clusters at the beginning of the sectors to place 2692 * the refcount table, a refcount block and the L1 table in; this may 2693 * overwrite parts of the existing refcount and L1 table, which is not 2694 * an issue because the dirty flag is set, complete data loss is in fact 2695 * desired and partial data loss is consequently fine as well */ 2696 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size, 2697 (2 + l1_clusters) * s->cluster_size, 0); 2698 /* This call (even if it failed overall) may have overwritten on-disk 2699 * refcount structures; in that case, the in-memory refcount information 2700 * will probably differ from the on-disk information which makes the BDS 2701 * unusable */ 2702 if (ret < 0) { 2703 goto fail_broken_refcounts; 2704 } 2705 2706 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 2707 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); 2708 2709 /* "Create" an empty reftable (one cluster) directly after the image 2710 * header and an empty L1 table three clusters after the image header; 2711 * the cluster between those two will be used as the first refblock */ 2712 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size); 2713 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size); 2714 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1); 2715 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), 2716 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); 2717 if (ret < 0) { 2718 goto fail_broken_refcounts; 2719 } 2720 2721 s->l1_table_offset = 3 * s->cluster_size; 2722 2723 new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t)); 2724 if (!new_reftable) { 2725 ret = -ENOMEM; 2726 goto fail_broken_refcounts; 2727 } 2728 2729 s->refcount_table_offset = s->cluster_size; 2730 s->refcount_table_size = s->cluster_size / sizeof(uint64_t); 2731 2732 g_free(s->refcount_table); 2733 s->refcount_table = new_reftable; 2734 new_reftable = NULL; 2735 2736 /* Now the in-memory refcount information again corresponds to the on-disk 2737 * information (reftable is empty and no refblocks (the refblock cache is 2738 * empty)); however, this means some clusters (e.g. the image header) are 2739 * referenced, but not refcounted, but the normal qcow2 code assumes that 2740 * the in-memory information is always correct */ 2741 2742 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 2743 2744 /* Enter the first refblock into the reftable */ 2745 rt_entry = cpu_to_be64(2 * s->cluster_size); 2746 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, 2747 &rt_entry, sizeof(rt_entry)); 2748 if (ret < 0) { 2749 goto fail_broken_refcounts; 2750 } 2751 s->refcount_table[0] = 2 * s->cluster_size; 2752 2753 s->free_cluster_index = 0; 2754 assert(3 + l1_clusters <= s->refcount_block_size); 2755 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); 2756 if (offset < 0) { 2757 ret = offset; 2758 goto fail_broken_refcounts; 2759 } else if (offset > 0) { 2760 error_report("First cluster in emptied image is in use"); 2761 abort(); 2762 } 2763 2764 /* Now finally the in-memory information corresponds to the on-disk 2765 * structures and is correct */ 2766 ret = qcow2_mark_clean(bs); 2767 if (ret < 0) { 2768 goto fail; 2769 } 2770 2771 ret = bdrv_truncate(bs->file->bs, (3 + l1_clusters) * s->cluster_size); 2772 if (ret < 0) { 2773 goto fail; 2774 } 2775 2776 return 0; 2777 2778 fail_broken_refcounts: 2779 /* The BDS is unusable at this point. If we wanted to make it usable, we 2780 * would have to call qcow2_refcount_close(), qcow2_refcount_init(), 2781 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() 2782 * again. However, because the functions which could have caused this error 2783 * path to be taken are used by those functions as well, it's very likely 2784 * that that sequence will fail as well. Therefore, just eject the BDS. */ 2785 bs->drv = NULL; 2786 2787 fail: 2788 g_free(new_reftable); 2789 return ret; 2790 } 2791 2792 static int qcow2_make_empty(BlockDriverState *bs) 2793 { 2794 BDRVQcow2State *s = bs->opaque; 2795 uint64_t start_sector; 2796 int sector_step = INT_MAX / BDRV_SECTOR_SIZE; 2797 int l1_clusters, ret = 0; 2798 2799 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 2800 2801 if (s->qcow_version >= 3 && !s->snapshots && 2802 3 + l1_clusters <= s->refcount_block_size) { 2803 /* The following function only works for qcow2 v3 images (it requires 2804 * the dirty flag) and only as long as there are no snapshots (because 2805 * it completely empties the image). Furthermore, the L1 table and three 2806 * additional clusters (image header, refcount table, one refcount 2807 * block) have to fit inside one refcount block. */ 2808 return make_completely_empty(bs); 2809 } 2810 2811 /* This fallback code simply discards every active cluster; this is slow, 2812 * but works in all cases */ 2813 for (start_sector = 0; start_sector < bs->total_sectors; 2814 start_sector += sector_step) 2815 { 2816 /* As this function is generally used after committing an external 2817 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the 2818 * default action for this kind of discard is to pass the discard, 2819 * which will ideally result in an actually smaller image file, as 2820 * is probably desired. */ 2821 ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE, 2822 MIN(sector_step, 2823 bs->total_sectors - start_sector), 2824 QCOW2_DISCARD_SNAPSHOT, true); 2825 if (ret < 0) { 2826 break; 2827 } 2828 } 2829 2830 return ret; 2831 } 2832 2833 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) 2834 { 2835 BDRVQcow2State *s = bs->opaque; 2836 int ret; 2837 2838 qemu_co_mutex_lock(&s->lock); 2839 ret = qcow2_cache_write(bs, s->l2_table_cache); 2840 if (ret < 0) { 2841 qemu_co_mutex_unlock(&s->lock); 2842 return ret; 2843 } 2844 2845 if (qcow2_need_accurate_refcounts(s)) { 2846 ret = qcow2_cache_write(bs, s->refcount_block_cache); 2847 if (ret < 0) { 2848 qemu_co_mutex_unlock(&s->lock); 2849 return ret; 2850 } 2851 } 2852 qemu_co_mutex_unlock(&s->lock); 2853 2854 return 0; 2855 } 2856 2857 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2858 { 2859 BDRVQcow2State *s = bs->opaque; 2860 bdi->unallocated_blocks_are_zero = true; 2861 bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3); 2862 bdi->cluster_size = s->cluster_size; 2863 bdi->vm_state_offset = qcow2_vm_state_offset(s); 2864 return 0; 2865 } 2866 2867 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs) 2868 { 2869 BDRVQcow2State *s = bs->opaque; 2870 ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1); 2871 2872 *spec_info = (ImageInfoSpecific){ 2873 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2, 2874 .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1), 2875 }; 2876 if (s->qcow_version == 2) { 2877 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 2878 .compat = g_strdup("0.10"), 2879 .refcount_bits = s->refcount_bits, 2880 }; 2881 } else if (s->qcow_version == 3) { 2882 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 2883 .compat = g_strdup("1.1"), 2884 .lazy_refcounts = s->compatible_features & 2885 QCOW2_COMPAT_LAZY_REFCOUNTS, 2886 .has_lazy_refcounts = true, 2887 .corrupt = s->incompatible_features & 2888 QCOW2_INCOMPAT_CORRUPT, 2889 .has_corrupt = true, 2890 .refcount_bits = s->refcount_bits, 2891 }; 2892 } else { 2893 /* if this assertion fails, this probably means a new version was 2894 * added without having it covered here */ 2895 assert(false); 2896 } 2897 2898 return spec_info; 2899 } 2900 2901 #if 0 2902 static void dump_refcounts(BlockDriverState *bs) 2903 { 2904 BDRVQcow2State *s = bs->opaque; 2905 int64_t nb_clusters, k, k1, size; 2906 int refcount; 2907 2908 size = bdrv_getlength(bs->file->bs); 2909 nb_clusters = size_to_clusters(s, size); 2910 for(k = 0; k < nb_clusters;) { 2911 k1 = k; 2912 refcount = get_refcount(bs, k); 2913 k++; 2914 while (k < nb_clusters && get_refcount(bs, k) == refcount) 2915 k++; 2916 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount, 2917 k - k1); 2918 } 2919 } 2920 #endif 2921 2922 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 2923 int64_t pos) 2924 { 2925 BDRVQcow2State *s = bs->opaque; 2926 2927 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 2928 return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos, 2929 qiov->size, qiov, 0); 2930 } 2931 2932 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 2933 int64_t pos) 2934 { 2935 BDRVQcow2State *s = bs->opaque; 2936 2937 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 2938 return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos, 2939 qiov->size, qiov, 0); 2940 } 2941 2942 /* 2943 * Downgrades an image's version. To achieve this, any incompatible features 2944 * have to be removed. 2945 */ 2946 static int qcow2_downgrade(BlockDriverState *bs, int target_version, 2947 BlockDriverAmendStatusCB *status_cb, void *cb_opaque) 2948 { 2949 BDRVQcow2State *s = bs->opaque; 2950 int current_version = s->qcow_version; 2951 int ret; 2952 2953 if (target_version == current_version) { 2954 return 0; 2955 } else if (target_version > current_version) { 2956 return -EINVAL; 2957 } else if (target_version != 2) { 2958 return -EINVAL; 2959 } 2960 2961 if (s->refcount_order != 4) { 2962 error_report("compat=0.10 requires refcount_bits=16"); 2963 return -ENOTSUP; 2964 } 2965 2966 /* clear incompatible features */ 2967 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 2968 ret = qcow2_mark_clean(bs); 2969 if (ret < 0) { 2970 return ret; 2971 } 2972 } 2973 2974 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in 2975 * the first place; if that happens nonetheless, returning -ENOTSUP is the 2976 * best thing to do anyway */ 2977 2978 if (s->incompatible_features) { 2979 return -ENOTSUP; 2980 } 2981 2982 /* since we can ignore compatible features, we can set them to 0 as well */ 2983 s->compatible_features = 0; 2984 /* if lazy refcounts have been used, they have already been fixed through 2985 * clearing the dirty flag */ 2986 2987 /* clearing autoclear features is trivial */ 2988 s->autoclear_features = 0; 2989 2990 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque); 2991 if (ret < 0) { 2992 return ret; 2993 } 2994 2995 s->qcow_version = target_version; 2996 ret = qcow2_update_header(bs); 2997 if (ret < 0) { 2998 s->qcow_version = current_version; 2999 return ret; 3000 } 3001 return 0; 3002 } 3003 3004 typedef enum Qcow2AmendOperation { 3005 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be 3006 * statically initialized to so that the helper CB can discern the first 3007 * invocation from an operation change */ 3008 QCOW2_NO_OPERATION = 0, 3009 3010 QCOW2_CHANGING_REFCOUNT_ORDER, 3011 QCOW2_DOWNGRADING, 3012 } Qcow2AmendOperation; 3013 3014 typedef struct Qcow2AmendHelperCBInfo { 3015 /* The code coordinating the amend operations should only modify 3016 * these four fields; the rest will be managed by the CB */ 3017 BlockDriverAmendStatusCB *original_status_cb; 3018 void *original_cb_opaque; 3019 3020 Qcow2AmendOperation current_operation; 3021 3022 /* Total number of operations to perform (only set once) */ 3023 int total_operations; 3024 3025 /* The following fields are managed by the CB */ 3026 3027 /* Number of operations completed */ 3028 int operations_completed; 3029 3030 /* Cumulative offset of all completed operations */ 3031 int64_t offset_completed; 3032 3033 Qcow2AmendOperation last_operation; 3034 int64_t last_work_size; 3035 } Qcow2AmendHelperCBInfo; 3036 3037 static void qcow2_amend_helper_cb(BlockDriverState *bs, 3038 int64_t operation_offset, 3039 int64_t operation_work_size, void *opaque) 3040 { 3041 Qcow2AmendHelperCBInfo *info = opaque; 3042 int64_t current_work_size; 3043 int64_t projected_work_size; 3044 3045 if (info->current_operation != info->last_operation) { 3046 if (info->last_operation != QCOW2_NO_OPERATION) { 3047 info->offset_completed += info->last_work_size; 3048 info->operations_completed++; 3049 } 3050 3051 info->last_operation = info->current_operation; 3052 } 3053 3054 assert(info->total_operations > 0); 3055 assert(info->operations_completed < info->total_operations); 3056 3057 info->last_work_size = operation_work_size; 3058 3059 current_work_size = info->offset_completed + operation_work_size; 3060 3061 /* current_work_size is the total work size for (operations_completed + 1) 3062 * operations (which includes this one), so multiply it by the number of 3063 * operations not covered and divide it by the number of operations 3064 * covered to get a projection for the operations not covered */ 3065 projected_work_size = current_work_size * (info->total_operations - 3066 info->operations_completed - 1) 3067 / (info->operations_completed + 1); 3068 3069 info->original_status_cb(bs, info->offset_completed + operation_offset, 3070 current_work_size + projected_work_size, 3071 info->original_cb_opaque); 3072 } 3073 3074 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, 3075 BlockDriverAmendStatusCB *status_cb, 3076 void *cb_opaque) 3077 { 3078 BDRVQcow2State *s = bs->opaque; 3079 int old_version = s->qcow_version, new_version = old_version; 3080 uint64_t new_size = 0; 3081 const char *backing_file = NULL, *backing_format = NULL; 3082 bool lazy_refcounts = s->use_lazy_refcounts; 3083 const char *compat = NULL; 3084 uint64_t cluster_size = s->cluster_size; 3085 bool encrypt; 3086 int refcount_bits = s->refcount_bits; 3087 int ret; 3088 QemuOptDesc *desc = opts->list->desc; 3089 Qcow2AmendHelperCBInfo helper_cb_info; 3090 3091 while (desc && desc->name) { 3092 if (!qemu_opt_find(opts, desc->name)) { 3093 /* only change explicitly defined options */ 3094 desc++; 3095 continue; 3096 } 3097 3098 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) { 3099 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL); 3100 if (!compat) { 3101 /* preserve default */ 3102 } else if (!strcmp(compat, "0.10")) { 3103 new_version = 2; 3104 } else if (!strcmp(compat, "1.1")) { 3105 new_version = 3; 3106 } else { 3107 error_report("Unknown compatibility level %s", compat); 3108 return -EINVAL; 3109 } 3110 } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) { 3111 error_report("Cannot change preallocation mode"); 3112 return -ENOTSUP; 3113 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) { 3114 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); 3115 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) { 3116 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); 3117 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) { 3118 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); 3119 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) { 3120 encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, 3121 !!s->cipher); 3122 3123 if (encrypt != !!s->cipher) { 3124 error_report("Changing the encryption flag is not supported"); 3125 return -ENOTSUP; 3126 } 3127 } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) { 3128 cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 3129 cluster_size); 3130 if (cluster_size != s->cluster_size) { 3131 error_report("Changing the cluster size is not supported"); 3132 return -ENOTSUP; 3133 } 3134 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) { 3135 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS, 3136 lazy_refcounts); 3137 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) { 3138 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS, 3139 refcount_bits); 3140 3141 if (refcount_bits <= 0 || refcount_bits > 64 || 3142 !is_power_of_2(refcount_bits)) 3143 { 3144 error_report("Refcount width must be a power of two and may " 3145 "not exceed 64 bits"); 3146 return -EINVAL; 3147 } 3148 } else { 3149 /* if this point is reached, this probably means a new option was 3150 * added without having it covered here */ 3151 abort(); 3152 } 3153 3154 desc++; 3155 } 3156 3157 helper_cb_info = (Qcow2AmendHelperCBInfo){ 3158 .original_status_cb = status_cb, 3159 .original_cb_opaque = cb_opaque, 3160 .total_operations = (new_version < old_version) 3161 + (s->refcount_bits != refcount_bits) 3162 }; 3163 3164 /* Upgrade first (some features may require compat=1.1) */ 3165 if (new_version > old_version) { 3166 s->qcow_version = new_version; 3167 ret = qcow2_update_header(bs); 3168 if (ret < 0) { 3169 s->qcow_version = old_version; 3170 return ret; 3171 } 3172 } 3173 3174 if (s->refcount_bits != refcount_bits) { 3175 int refcount_order = ctz32(refcount_bits); 3176 Error *local_error = NULL; 3177 3178 if (new_version < 3 && refcount_bits != 16) { 3179 error_report("Different refcount widths than 16 bits require " 3180 "compatibility level 1.1 or above (use compat=1.1 or " 3181 "greater)"); 3182 return -EINVAL; 3183 } 3184 3185 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER; 3186 ret = qcow2_change_refcount_order(bs, refcount_order, 3187 &qcow2_amend_helper_cb, 3188 &helper_cb_info, &local_error); 3189 if (ret < 0) { 3190 error_report_err(local_error); 3191 return ret; 3192 } 3193 } 3194 3195 if (backing_file || backing_format) { 3196 ret = qcow2_change_backing_file(bs, 3197 backing_file ?: s->image_backing_file, 3198 backing_format ?: s->image_backing_format); 3199 if (ret < 0) { 3200 return ret; 3201 } 3202 } 3203 3204 if (s->use_lazy_refcounts != lazy_refcounts) { 3205 if (lazy_refcounts) { 3206 if (new_version < 3) { 3207 error_report("Lazy refcounts only supported with compatibility " 3208 "level 1.1 and above (use compat=1.1 or greater)"); 3209 return -EINVAL; 3210 } 3211 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 3212 ret = qcow2_update_header(bs); 3213 if (ret < 0) { 3214 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 3215 return ret; 3216 } 3217 s->use_lazy_refcounts = true; 3218 } else { 3219 /* make image clean first */ 3220 ret = qcow2_mark_clean(bs); 3221 if (ret < 0) { 3222 return ret; 3223 } 3224 /* now disallow lazy refcounts */ 3225 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 3226 ret = qcow2_update_header(bs); 3227 if (ret < 0) { 3228 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 3229 return ret; 3230 } 3231 s->use_lazy_refcounts = false; 3232 } 3233 } 3234 3235 if (new_size) { 3236 ret = bdrv_truncate(bs, new_size); 3237 if (ret < 0) { 3238 return ret; 3239 } 3240 } 3241 3242 /* Downgrade last (so unsupported features can be removed before) */ 3243 if (new_version < old_version) { 3244 helper_cb_info.current_operation = QCOW2_DOWNGRADING; 3245 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb, 3246 &helper_cb_info); 3247 if (ret < 0) { 3248 return ret; 3249 } 3250 } 3251 3252 return 0; 3253 } 3254 3255 /* 3256 * If offset or size are negative, respectively, they will not be included in 3257 * the BLOCK_IMAGE_CORRUPTED event emitted. 3258 * fatal will be ignored for read-only BDS; corruptions found there will always 3259 * be considered non-fatal. 3260 */ 3261 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 3262 int64_t size, const char *message_format, ...) 3263 { 3264 BDRVQcow2State *s = bs->opaque; 3265 const char *node_name; 3266 char *message; 3267 va_list ap; 3268 3269 fatal = fatal && !bs->read_only; 3270 3271 if (s->signaled_corruption && 3272 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) 3273 { 3274 return; 3275 } 3276 3277 va_start(ap, message_format); 3278 message = g_strdup_vprintf(message_format, ap); 3279 va_end(ap); 3280 3281 if (fatal) { 3282 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " 3283 "corruption events will be suppressed\n", message); 3284 } else { 3285 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " 3286 "corruption events will be suppressed\n", message); 3287 } 3288 3289 node_name = bdrv_get_node_name(bs); 3290 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), 3291 *node_name != '\0', node_name, 3292 message, offset >= 0, offset, 3293 size >= 0, size, 3294 fatal, &error_abort); 3295 g_free(message); 3296 3297 if (fatal) { 3298 qcow2_mark_corrupt(bs); 3299 bs->drv = NULL; /* make BDS unusable */ 3300 } 3301 3302 s->signaled_corruption = true; 3303 } 3304 3305 static QemuOptsList qcow2_create_opts = { 3306 .name = "qcow2-create-opts", 3307 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), 3308 .desc = { 3309 { 3310 .name = BLOCK_OPT_SIZE, 3311 .type = QEMU_OPT_SIZE, 3312 .help = "Virtual disk size" 3313 }, 3314 { 3315 .name = BLOCK_OPT_COMPAT_LEVEL, 3316 .type = QEMU_OPT_STRING, 3317 .help = "Compatibility level (0.10 or 1.1)" 3318 }, 3319 { 3320 .name = BLOCK_OPT_BACKING_FILE, 3321 .type = QEMU_OPT_STRING, 3322 .help = "File name of a base image" 3323 }, 3324 { 3325 .name = BLOCK_OPT_BACKING_FMT, 3326 .type = QEMU_OPT_STRING, 3327 .help = "Image format of the base image" 3328 }, 3329 { 3330 .name = BLOCK_OPT_ENCRYPT, 3331 .type = QEMU_OPT_BOOL, 3332 .help = "Encrypt the image", 3333 .def_value_str = "off" 3334 }, 3335 { 3336 .name = BLOCK_OPT_CLUSTER_SIZE, 3337 .type = QEMU_OPT_SIZE, 3338 .help = "qcow2 cluster size", 3339 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) 3340 }, 3341 { 3342 .name = BLOCK_OPT_PREALLOC, 3343 .type = QEMU_OPT_STRING, 3344 .help = "Preallocation mode (allowed values: off, metadata, " 3345 "falloc, full)" 3346 }, 3347 { 3348 .name = BLOCK_OPT_LAZY_REFCOUNTS, 3349 .type = QEMU_OPT_BOOL, 3350 .help = "Postpone refcount updates", 3351 .def_value_str = "off" 3352 }, 3353 { 3354 .name = BLOCK_OPT_REFCOUNT_BITS, 3355 .type = QEMU_OPT_NUMBER, 3356 .help = "Width of a reference count entry in bits", 3357 .def_value_str = "16" 3358 }, 3359 { /* end of list */ } 3360 } 3361 }; 3362 3363 BlockDriver bdrv_qcow2 = { 3364 .format_name = "qcow2", 3365 .instance_size = sizeof(BDRVQcow2State), 3366 .bdrv_probe = qcow2_probe, 3367 .bdrv_open = qcow2_open, 3368 .bdrv_close = qcow2_close, 3369 .bdrv_reopen_prepare = qcow2_reopen_prepare, 3370 .bdrv_reopen_commit = qcow2_reopen_commit, 3371 .bdrv_reopen_abort = qcow2_reopen_abort, 3372 .bdrv_join_options = qcow2_join_options, 3373 .bdrv_create = qcow2_create, 3374 .bdrv_has_zero_init = bdrv_has_zero_init_1, 3375 .bdrv_co_get_block_status = qcow2_co_get_block_status, 3376 .bdrv_set_key = qcow2_set_key, 3377 3378 .bdrv_co_preadv = qcow2_co_preadv, 3379 .bdrv_co_pwritev = qcow2_co_pwritev, 3380 .bdrv_co_flush_to_os = qcow2_co_flush_to_os, 3381 3382 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes, 3383 .bdrv_co_pdiscard = qcow2_co_pdiscard, 3384 .bdrv_truncate = qcow2_truncate, 3385 .bdrv_co_pwritev_compressed = qcow2_co_pwritev_compressed, 3386 .bdrv_make_empty = qcow2_make_empty, 3387 3388 .bdrv_snapshot_create = qcow2_snapshot_create, 3389 .bdrv_snapshot_goto = qcow2_snapshot_goto, 3390 .bdrv_snapshot_delete = qcow2_snapshot_delete, 3391 .bdrv_snapshot_list = qcow2_snapshot_list, 3392 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 3393 .bdrv_get_info = qcow2_get_info, 3394 .bdrv_get_specific_info = qcow2_get_specific_info, 3395 3396 .bdrv_save_vmstate = qcow2_save_vmstate, 3397 .bdrv_load_vmstate = qcow2_load_vmstate, 3398 3399 .supports_backing = true, 3400 .bdrv_change_backing_file = qcow2_change_backing_file, 3401 3402 .bdrv_refresh_limits = qcow2_refresh_limits, 3403 .bdrv_invalidate_cache = qcow2_invalidate_cache, 3404 .bdrv_inactivate = qcow2_inactivate, 3405 3406 .create_opts = &qcow2_create_opts, 3407 .bdrv_check = qcow2_check, 3408 .bdrv_amend_options = qcow2_amend_options, 3409 3410 .bdrv_detach_aio_context = qcow2_detach_aio_context, 3411 .bdrv_attach_aio_context = qcow2_attach_aio_context, 3412 }; 3413 3414 static void bdrv_qcow2_init(void) 3415 { 3416 bdrv_register(&bdrv_qcow2); 3417 } 3418 3419 block_init(bdrv_qcow2_init); 3420