1 /* 2 * Block driver for the QCOW version 2 format 3 * 4 * Copyright (c) 2004-2006 Fabrice Bellard 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 #include "qemu/osdep.h" 25 #include "block/block_int.h" 26 #include "sysemu/block-backend.h" 27 #include "qemu/module.h" 28 #include <zlib.h> 29 #include "block/qcow2.h" 30 #include "qemu/error-report.h" 31 #include "qapi/qmp/qerror.h" 32 #include "qapi/qmp/qbool.h" 33 #include "qapi/util.h" 34 #include "qapi/qmp/types.h" 35 #include "qapi-event.h" 36 #include "trace.h" 37 #include "qemu/option_int.h" 38 #include "qemu/cutils.h" 39 #include "qemu/bswap.h" 40 41 /* 42 Differences with QCOW: 43 44 - Support for multiple incremental snapshots. 45 - Memory management by reference counts. 46 - Clusters which have a reference count of one have the bit 47 QCOW_OFLAG_COPIED to optimize write performance. 48 - Size of compressed clusters is stored in sectors to reduce bit usage 49 in the cluster offsets. 50 - Support for storing additional data (such as the VM state) in the 51 snapshots. 52 - If a backing store is used, the cluster size is not constrained 53 (could be backported to QCOW). 54 - L2 tables have always a size of one cluster. 55 */ 56 57 58 typedef struct { 59 uint32_t magic; 60 uint32_t len; 61 } QEMU_PACKED QCowExtension; 62 63 #define QCOW2_EXT_MAGIC_END 0 64 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA 65 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 66 67 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) 68 { 69 const QCowHeader *cow_header = (const void *)buf; 70 71 if (buf_size >= sizeof(QCowHeader) && 72 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 73 be32_to_cpu(cow_header->version) >= 2) 74 return 100; 75 else 76 return 0; 77 } 78 79 80 /* 81 * read qcow2 extension and fill bs 82 * start reading from start_offset 83 * finish reading upon magic of value 0 or when end_offset reached 84 * unknown magic is skipped (future extension this version knows nothing about) 85 * return 0 upon success, non-0 otherwise 86 */ 87 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, 88 uint64_t end_offset, void **p_feature_table, 89 Error **errp) 90 { 91 BDRVQcow2State *s = bs->opaque; 92 QCowExtension ext; 93 uint64_t offset; 94 int ret; 95 96 #ifdef DEBUG_EXT 97 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); 98 #endif 99 offset = start_offset; 100 while (offset < end_offset) { 101 102 #ifdef DEBUG_EXT 103 /* Sanity check */ 104 if (offset > s->cluster_size) 105 printf("qcow2_read_extension: suspicious offset %lu\n", offset); 106 107 printf("attempting to read extended header in offset %lu\n", offset); 108 #endif 109 110 ret = bdrv_pread(bs->file->bs, offset, &ext, sizeof(ext)); 111 if (ret < 0) { 112 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: " 113 "pread fail from offset %" PRIu64, offset); 114 return 1; 115 } 116 be32_to_cpus(&ext.magic); 117 be32_to_cpus(&ext.len); 118 offset += sizeof(ext); 119 #ifdef DEBUG_EXT 120 printf("ext.magic = 0x%x\n", ext.magic); 121 #endif 122 if (offset > end_offset || ext.len > end_offset - offset) { 123 error_setg(errp, "Header extension too large"); 124 return -EINVAL; 125 } 126 127 switch (ext.magic) { 128 case QCOW2_EXT_MAGIC_END: 129 return 0; 130 131 case QCOW2_EXT_MAGIC_BACKING_FORMAT: 132 if (ext.len >= sizeof(bs->backing_format)) { 133 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32 134 " too large (>=%zu)", ext.len, 135 sizeof(bs->backing_format)); 136 return 2; 137 } 138 ret = bdrv_pread(bs->file->bs, offset, bs->backing_format, ext.len); 139 if (ret < 0) { 140 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: " 141 "Could not read format name"); 142 return 3; 143 } 144 bs->backing_format[ext.len] = '\0'; 145 s->image_backing_format = g_strdup(bs->backing_format); 146 #ifdef DEBUG_EXT 147 printf("Qcow2: Got format extension %s\n", bs->backing_format); 148 #endif 149 break; 150 151 case QCOW2_EXT_MAGIC_FEATURE_TABLE: 152 if (p_feature_table != NULL) { 153 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); 154 ret = bdrv_pread(bs->file->bs, offset , feature_table, ext.len); 155 if (ret < 0) { 156 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: " 157 "Could not read table"); 158 return ret; 159 } 160 161 *p_feature_table = feature_table; 162 } 163 break; 164 165 default: 166 /* unknown magic - save it in case we need to rewrite the header */ 167 { 168 Qcow2UnknownHeaderExtension *uext; 169 170 uext = g_malloc0(sizeof(*uext) + ext.len); 171 uext->magic = ext.magic; 172 uext->len = ext.len; 173 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); 174 175 ret = bdrv_pread(bs->file->bs, offset , uext->data, uext->len); 176 if (ret < 0) { 177 error_setg_errno(errp, -ret, "ERROR: unknown extension: " 178 "Could not read data"); 179 return ret; 180 } 181 } 182 break; 183 } 184 185 offset += ((ext.len + 7) & ~7); 186 } 187 188 return 0; 189 } 190 191 static void cleanup_unknown_header_ext(BlockDriverState *bs) 192 { 193 BDRVQcow2State *s = bs->opaque; 194 Qcow2UnknownHeaderExtension *uext, *next; 195 196 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) { 197 QLIST_REMOVE(uext, next); 198 g_free(uext); 199 } 200 } 201 202 static void report_unsupported_feature(Error **errp, Qcow2Feature *table, 203 uint64_t mask) 204 { 205 char *features = g_strdup(""); 206 char *old; 207 208 while (table && table->name[0] != '\0') { 209 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) { 210 if (mask & (1ULL << table->bit)) { 211 old = features; 212 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "", 213 table->name); 214 g_free(old); 215 mask &= ~(1ULL << table->bit); 216 } 217 } 218 table++; 219 } 220 221 if (mask) { 222 old = features; 223 features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64, 224 old, *old ? ", " : "", mask); 225 g_free(old); 226 } 227 228 error_setg(errp, "Unsupported qcow2 feature(s): %s", features); 229 g_free(features); 230 } 231 232 /* 233 * Sets the dirty bit and flushes afterwards if necessary. 234 * 235 * The incompatible_features bit is only set if the image file header was 236 * updated successfully. Therefore it is not required to check the return 237 * value of this function. 238 */ 239 int qcow2_mark_dirty(BlockDriverState *bs) 240 { 241 BDRVQcow2State *s = bs->opaque; 242 uint64_t val; 243 int ret; 244 245 assert(s->qcow_version >= 3); 246 247 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 248 return 0; /* already dirty */ 249 } 250 251 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); 252 ret = bdrv_pwrite(bs->file->bs, offsetof(QCowHeader, incompatible_features), 253 &val, sizeof(val)); 254 if (ret < 0) { 255 return ret; 256 } 257 ret = bdrv_flush(bs->file->bs); 258 if (ret < 0) { 259 return ret; 260 } 261 262 /* Only treat image as dirty if the header was updated successfully */ 263 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; 264 return 0; 265 } 266 267 /* 268 * Clears the dirty bit and flushes before if necessary. Only call this 269 * function when there are no pending requests, it does not guard against 270 * concurrent requests dirtying the image. 271 */ 272 static int qcow2_mark_clean(BlockDriverState *bs) 273 { 274 BDRVQcow2State *s = bs->opaque; 275 276 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 277 int ret; 278 279 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; 280 281 ret = bdrv_flush(bs); 282 if (ret < 0) { 283 return ret; 284 } 285 286 return qcow2_update_header(bs); 287 } 288 return 0; 289 } 290 291 /* 292 * Marks the image as corrupt. 293 */ 294 int qcow2_mark_corrupt(BlockDriverState *bs) 295 { 296 BDRVQcow2State *s = bs->opaque; 297 298 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT; 299 return qcow2_update_header(bs); 300 } 301 302 /* 303 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes 304 * before if necessary. 305 */ 306 int qcow2_mark_consistent(BlockDriverState *bs) 307 { 308 BDRVQcow2State *s = bs->opaque; 309 310 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 311 int ret = bdrv_flush(bs); 312 if (ret < 0) { 313 return ret; 314 } 315 316 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT; 317 return qcow2_update_header(bs); 318 } 319 return 0; 320 } 321 322 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result, 323 BdrvCheckMode fix) 324 { 325 int ret = qcow2_check_refcounts(bs, result, fix); 326 if (ret < 0) { 327 return ret; 328 } 329 330 if (fix && result->check_errors == 0 && result->corruptions == 0) { 331 ret = qcow2_mark_clean(bs); 332 if (ret < 0) { 333 return ret; 334 } 335 return qcow2_mark_consistent(bs); 336 } 337 return ret; 338 } 339 340 static int validate_table_offset(BlockDriverState *bs, uint64_t offset, 341 uint64_t entries, size_t entry_len) 342 { 343 BDRVQcow2State *s = bs->opaque; 344 uint64_t size; 345 346 /* Use signed INT64_MAX as the maximum even for uint64_t header fields, 347 * because values will be passed to qemu functions taking int64_t. */ 348 if (entries > INT64_MAX / entry_len) { 349 return -EINVAL; 350 } 351 352 size = entries * entry_len; 353 354 if (INT64_MAX - size < offset) { 355 return -EINVAL; 356 } 357 358 /* Tables must be cluster aligned */ 359 if (offset & (s->cluster_size - 1)) { 360 return -EINVAL; 361 } 362 363 return 0; 364 } 365 366 static QemuOptsList qcow2_runtime_opts = { 367 .name = "qcow2", 368 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), 369 .desc = { 370 { 371 .name = QCOW2_OPT_LAZY_REFCOUNTS, 372 .type = QEMU_OPT_BOOL, 373 .help = "Postpone refcount updates", 374 }, 375 { 376 .name = QCOW2_OPT_DISCARD_REQUEST, 377 .type = QEMU_OPT_BOOL, 378 .help = "Pass guest discard requests to the layer below", 379 }, 380 { 381 .name = QCOW2_OPT_DISCARD_SNAPSHOT, 382 .type = QEMU_OPT_BOOL, 383 .help = "Generate discard requests when snapshot related space " 384 "is freed", 385 }, 386 { 387 .name = QCOW2_OPT_DISCARD_OTHER, 388 .type = QEMU_OPT_BOOL, 389 .help = "Generate discard requests when other clusters are freed", 390 }, 391 { 392 .name = QCOW2_OPT_OVERLAP, 393 .type = QEMU_OPT_STRING, 394 .help = "Selects which overlap checks to perform from a range of " 395 "templates (none, constant, cached, all)", 396 }, 397 { 398 .name = QCOW2_OPT_OVERLAP_TEMPLATE, 399 .type = QEMU_OPT_STRING, 400 .help = "Selects which overlap checks to perform from a range of " 401 "templates (none, constant, cached, all)", 402 }, 403 { 404 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER, 405 .type = QEMU_OPT_BOOL, 406 .help = "Check for unintended writes into the main qcow2 header", 407 }, 408 { 409 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1, 410 .type = QEMU_OPT_BOOL, 411 .help = "Check for unintended writes into the active L1 table", 412 }, 413 { 414 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2, 415 .type = QEMU_OPT_BOOL, 416 .help = "Check for unintended writes into an active L2 table", 417 }, 418 { 419 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 420 .type = QEMU_OPT_BOOL, 421 .help = "Check for unintended writes into the refcount table", 422 }, 423 { 424 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 425 .type = QEMU_OPT_BOOL, 426 .help = "Check for unintended writes into a refcount block", 427 }, 428 { 429 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 430 .type = QEMU_OPT_BOOL, 431 .help = "Check for unintended writes into the snapshot table", 432 }, 433 { 434 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1, 435 .type = QEMU_OPT_BOOL, 436 .help = "Check for unintended writes into an inactive L1 table", 437 }, 438 { 439 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2, 440 .type = QEMU_OPT_BOOL, 441 .help = "Check for unintended writes into an inactive L2 table", 442 }, 443 { 444 .name = QCOW2_OPT_CACHE_SIZE, 445 .type = QEMU_OPT_SIZE, 446 .help = "Maximum combined metadata (L2 tables and refcount blocks) " 447 "cache size", 448 }, 449 { 450 .name = QCOW2_OPT_L2_CACHE_SIZE, 451 .type = QEMU_OPT_SIZE, 452 .help = "Maximum L2 table cache size", 453 }, 454 { 455 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE, 456 .type = QEMU_OPT_SIZE, 457 .help = "Maximum refcount block cache size", 458 }, 459 { 460 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL, 461 .type = QEMU_OPT_NUMBER, 462 .help = "Clean unused cache entries after this time (in seconds)", 463 }, 464 { /* end of list */ } 465 }, 466 }; 467 468 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { 469 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, 470 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, 471 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, 472 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 473 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 474 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 475 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, 476 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, 477 }; 478 479 static void cache_clean_timer_cb(void *opaque) 480 { 481 BlockDriverState *bs = opaque; 482 BDRVQcow2State *s = bs->opaque; 483 qcow2_cache_clean_unused(bs, s->l2_table_cache); 484 qcow2_cache_clean_unused(bs, s->refcount_block_cache); 485 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 486 (int64_t) s->cache_clean_interval * 1000); 487 } 488 489 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context) 490 { 491 BDRVQcow2State *s = bs->opaque; 492 if (s->cache_clean_interval > 0) { 493 s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL, 494 SCALE_MS, cache_clean_timer_cb, 495 bs); 496 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 497 (int64_t) s->cache_clean_interval * 1000); 498 } 499 } 500 501 static void cache_clean_timer_del(BlockDriverState *bs) 502 { 503 BDRVQcow2State *s = bs->opaque; 504 if (s->cache_clean_timer) { 505 timer_del(s->cache_clean_timer); 506 timer_free(s->cache_clean_timer); 507 s->cache_clean_timer = NULL; 508 } 509 } 510 511 static void qcow2_detach_aio_context(BlockDriverState *bs) 512 { 513 cache_clean_timer_del(bs); 514 } 515 516 static void qcow2_attach_aio_context(BlockDriverState *bs, 517 AioContext *new_context) 518 { 519 cache_clean_timer_init(bs, new_context); 520 } 521 522 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts, 523 uint64_t *l2_cache_size, 524 uint64_t *refcount_cache_size, Error **errp) 525 { 526 BDRVQcow2State *s = bs->opaque; 527 uint64_t combined_cache_size; 528 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set; 529 530 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE); 531 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE); 532 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 533 534 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0); 535 *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0); 536 *refcount_cache_size = qemu_opt_get_size(opts, 537 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0); 538 539 if (combined_cache_size_set) { 540 if (l2_cache_size_set && refcount_cache_size_set) { 541 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE 542 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set " 543 "the same time"); 544 return; 545 } else if (*l2_cache_size > combined_cache_size) { 546 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed " 547 QCOW2_OPT_CACHE_SIZE); 548 return; 549 } else if (*refcount_cache_size > combined_cache_size) { 550 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed " 551 QCOW2_OPT_CACHE_SIZE); 552 return; 553 } 554 555 if (l2_cache_size_set) { 556 *refcount_cache_size = combined_cache_size - *l2_cache_size; 557 } else if (refcount_cache_size_set) { 558 *l2_cache_size = combined_cache_size - *refcount_cache_size; 559 } else { 560 *refcount_cache_size = combined_cache_size 561 / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1); 562 *l2_cache_size = combined_cache_size - *refcount_cache_size; 563 } 564 } else { 565 if (!l2_cache_size_set && !refcount_cache_size_set) { 566 *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE, 567 (uint64_t)DEFAULT_L2_CACHE_CLUSTERS 568 * s->cluster_size); 569 *refcount_cache_size = *l2_cache_size 570 / DEFAULT_L2_REFCOUNT_SIZE_RATIO; 571 } else if (!l2_cache_size_set) { 572 *l2_cache_size = *refcount_cache_size 573 * DEFAULT_L2_REFCOUNT_SIZE_RATIO; 574 } else if (!refcount_cache_size_set) { 575 *refcount_cache_size = *l2_cache_size 576 / DEFAULT_L2_REFCOUNT_SIZE_RATIO; 577 } 578 } 579 } 580 581 typedef struct Qcow2ReopenState { 582 Qcow2Cache *l2_table_cache; 583 Qcow2Cache *refcount_block_cache; 584 bool use_lazy_refcounts; 585 int overlap_check; 586 bool discard_passthrough[QCOW2_DISCARD_MAX]; 587 uint64_t cache_clean_interval; 588 } Qcow2ReopenState; 589 590 static int qcow2_update_options_prepare(BlockDriverState *bs, 591 Qcow2ReopenState *r, 592 QDict *options, int flags, 593 Error **errp) 594 { 595 BDRVQcow2State *s = bs->opaque; 596 QemuOpts *opts = NULL; 597 const char *opt_overlap_check, *opt_overlap_check_template; 598 int overlap_check_template = 0; 599 uint64_t l2_cache_size, refcount_cache_size; 600 int i; 601 Error *local_err = NULL; 602 int ret; 603 604 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); 605 qemu_opts_absorb_qdict(opts, options, &local_err); 606 if (local_err) { 607 error_propagate(errp, local_err); 608 ret = -EINVAL; 609 goto fail; 610 } 611 612 /* get L2 table/refcount block cache size from command line options */ 613 read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size, 614 &local_err); 615 if (local_err) { 616 error_propagate(errp, local_err); 617 ret = -EINVAL; 618 goto fail; 619 } 620 621 l2_cache_size /= s->cluster_size; 622 if (l2_cache_size < MIN_L2_CACHE_SIZE) { 623 l2_cache_size = MIN_L2_CACHE_SIZE; 624 } 625 if (l2_cache_size > INT_MAX) { 626 error_setg(errp, "L2 cache size too big"); 627 ret = -EINVAL; 628 goto fail; 629 } 630 631 refcount_cache_size /= s->cluster_size; 632 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) { 633 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE; 634 } 635 if (refcount_cache_size > INT_MAX) { 636 error_setg(errp, "Refcount cache size too big"); 637 ret = -EINVAL; 638 goto fail; 639 } 640 641 /* alloc new L2 table/refcount block cache, flush old one */ 642 if (s->l2_table_cache) { 643 ret = qcow2_cache_flush(bs, s->l2_table_cache); 644 if (ret) { 645 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache"); 646 goto fail; 647 } 648 } 649 650 if (s->refcount_block_cache) { 651 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 652 if (ret) { 653 error_setg_errno(errp, -ret, 654 "Failed to flush the refcount block cache"); 655 goto fail; 656 } 657 } 658 659 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size); 660 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size); 661 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) { 662 error_setg(errp, "Could not allocate metadata caches"); 663 ret = -ENOMEM; 664 goto fail; 665 } 666 667 /* New interval for cache cleanup timer */ 668 r->cache_clean_interval = 669 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL, 670 s->cache_clean_interval); 671 if (r->cache_clean_interval > UINT_MAX) { 672 error_setg(errp, "Cache clean interval too big"); 673 ret = -EINVAL; 674 goto fail; 675 } 676 677 /* lazy-refcounts; flush if going from enabled to disabled */ 678 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, 679 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); 680 if (r->use_lazy_refcounts && s->qcow_version < 3) { 681 error_setg(errp, "Lazy refcounts require a qcow2 image with at least " 682 "qemu 1.1 compatibility level"); 683 ret = -EINVAL; 684 goto fail; 685 } 686 687 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) { 688 ret = qcow2_mark_clean(bs); 689 if (ret < 0) { 690 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts"); 691 goto fail; 692 } 693 } 694 695 /* Overlap check options */ 696 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP); 697 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE); 698 if (opt_overlap_check_template && opt_overlap_check && 699 strcmp(opt_overlap_check_template, opt_overlap_check)) 700 { 701 error_setg(errp, "Conflicting values for qcow2 options '" 702 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE 703 "' ('%s')", opt_overlap_check, opt_overlap_check_template); 704 ret = -EINVAL; 705 goto fail; 706 } 707 if (!opt_overlap_check) { 708 opt_overlap_check = opt_overlap_check_template ?: "cached"; 709 } 710 711 if (!strcmp(opt_overlap_check, "none")) { 712 overlap_check_template = 0; 713 } else if (!strcmp(opt_overlap_check, "constant")) { 714 overlap_check_template = QCOW2_OL_CONSTANT; 715 } else if (!strcmp(opt_overlap_check, "cached")) { 716 overlap_check_template = QCOW2_OL_CACHED; 717 } else if (!strcmp(opt_overlap_check, "all")) { 718 overlap_check_template = QCOW2_OL_ALL; 719 } else { 720 error_setg(errp, "Unsupported value '%s' for qcow2 option " 721 "'overlap-check'. Allowed are any of the following: " 722 "none, constant, cached, all", opt_overlap_check); 723 ret = -EINVAL; 724 goto fail; 725 } 726 727 r->overlap_check = 0; 728 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { 729 /* overlap-check defines a template bitmask, but every flag may be 730 * overwritten through the associated boolean option */ 731 r->overlap_check |= 732 qemu_opt_get_bool(opts, overlap_bool_option_names[i], 733 overlap_check_template & (1 << i)) << i; 734 } 735 736 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false; 737 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true; 738 r->discard_passthrough[QCOW2_DISCARD_REQUEST] = 739 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, 740 flags & BDRV_O_UNMAP); 741 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = 742 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true); 743 r->discard_passthrough[QCOW2_DISCARD_OTHER] = 744 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false); 745 746 ret = 0; 747 fail: 748 qemu_opts_del(opts); 749 opts = NULL; 750 return ret; 751 } 752 753 static void qcow2_update_options_commit(BlockDriverState *bs, 754 Qcow2ReopenState *r) 755 { 756 BDRVQcow2State *s = bs->opaque; 757 int i; 758 759 if (s->l2_table_cache) { 760 qcow2_cache_destroy(bs, s->l2_table_cache); 761 } 762 if (s->refcount_block_cache) { 763 qcow2_cache_destroy(bs, s->refcount_block_cache); 764 } 765 s->l2_table_cache = r->l2_table_cache; 766 s->refcount_block_cache = r->refcount_block_cache; 767 768 s->overlap_check = r->overlap_check; 769 s->use_lazy_refcounts = r->use_lazy_refcounts; 770 771 for (i = 0; i < QCOW2_DISCARD_MAX; i++) { 772 s->discard_passthrough[i] = r->discard_passthrough[i]; 773 } 774 775 if (s->cache_clean_interval != r->cache_clean_interval) { 776 cache_clean_timer_del(bs); 777 s->cache_clean_interval = r->cache_clean_interval; 778 cache_clean_timer_init(bs, bdrv_get_aio_context(bs)); 779 } 780 } 781 782 static void qcow2_update_options_abort(BlockDriverState *bs, 783 Qcow2ReopenState *r) 784 { 785 if (r->l2_table_cache) { 786 qcow2_cache_destroy(bs, r->l2_table_cache); 787 } 788 if (r->refcount_block_cache) { 789 qcow2_cache_destroy(bs, r->refcount_block_cache); 790 } 791 } 792 793 static int qcow2_update_options(BlockDriverState *bs, QDict *options, 794 int flags, Error **errp) 795 { 796 Qcow2ReopenState r = {}; 797 int ret; 798 799 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp); 800 if (ret >= 0) { 801 qcow2_update_options_commit(bs, &r); 802 } else { 803 qcow2_update_options_abort(bs, &r); 804 } 805 806 return ret; 807 } 808 809 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, 810 Error **errp) 811 { 812 BDRVQcow2State *s = bs->opaque; 813 unsigned int len, i; 814 int ret = 0; 815 QCowHeader header; 816 Error *local_err = NULL; 817 uint64_t ext_end; 818 uint64_t l1_vm_state_index; 819 820 ret = bdrv_pread(bs->file->bs, 0, &header, sizeof(header)); 821 if (ret < 0) { 822 error_setg_errno(errp, -ret, "Could not read qcow2 header"); 823 goto fail; 824 } 825 be32_to_cpus(&header.magic); 826 be32_to_cpus(&header.version); 827 be64_to_cpus(&header.backing_file_offset); 828 be32_to_cpus(&header.backing_file_size); 829 be64_to_cpus(&header.size); 830 be32_to_cpus(&header.cluster_bits); 831 be32_to_cpus(&header.crypt_method); 832 be64_to_cpus(&header.l1_table_offset); 833 be32_to_cpus(&header.l1_size); 834 be64_to_cpus(&header.refcount_table_offset); 835 be32_to_cpus(&header.refcount_table_clusters); 836 be64_to_cpus(&header.snapshots_offset); 837 be32_to_cpus(&header.nb_snapshots); 838 839 if (header.magic != QCOW_MAGIC) { 840 error_setg(errp, "Image is not in qcow2 format"); 841 ret = -EINVAL; 842 goto fail; 843 } 844 if (header.version < 2 || header.version > 3) { 845 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version); 846 ret = -ENOTSUP; 847 goto fail; 848 } 849 850 s->qcow_version = header.version; 851 852 /* Initialise cluster size */ 853 if (header.cluster_bits < MIN_CLUSTER_BITS || 854 header.cluster_bits > MAX_CLUSTER_BITS) { 855 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, 856 header.cluster_bits); 857 ret = -EINVAL; 858 goto fail; 859 } 860 861 s->cluster_bits = header.cluster_bits; 862 s->cluster_size = 1 << s->cluster_bits; 863 s->cluster_sectors = 1 << (s->cluster_bits - 9); 864 865 /* Initialise version 3 header fields */ 866 if (header.version == 2) { 867 header.incompatible_features = 0; 868 header.compatible_features = 0; 869 header.autoclear_features = 0; 870 header.refcount_order = 4; 871 header.header_length = 72; 872 } else { 873 be64_to_cpus(&header.incompatible_features); 874 be64_to_cpus(&header.compatible_features); 875 be64_to_cpus(&header.autoclear_features); 876 be32_to_cpus(&header.refcount_order); 877 be32_to_cpus(&header.header_length); 878 879 if (header.header_length < 104) { 880 error_setg(errp, "qcow2 header too short"); 881 ret = -EINVAL; 882 goto fail; 883 } 884 } 885 886 if (header.header_length > s->cluster_size) { 887 error_setg(errp, "qcow2 header exceeds cluster size"); 888 ret = -EINVAL; 889 goto fail; 890 } 891 892 if (header.header_length > sizeof(header)) { 893 s->unknown_header_fields_size = header.header_length - sizeof(header); 894 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); 895 ret = bdrv_pread(bs->file->bs, sizeof(header), s->unknown_header_fields, 896 s->unknown_header_fields_size); 897 if (ret < 0) { 898 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header " 899 "fields"); 900 goto fail; 901 } 902 } 903 904 if (header.backing_file_offset > s->cluster_size) { 905 error_setg(errp, "Invalid backing file offset"); 906 ret = -EINVAL; 907 goto fail; 908 } 909 910 if (header.backing_file_offset) { 911 ext_end = header.backing_file_offset; 912 } else { 913 ext_end = 1 << header.cluster_bits; 914 } 915 916 /* Handle feature bits */ 917 s->incompatible_features = header.incompatible_features; 918 s->compatible_features = header.compatible_features; 919 s->autoclear_features = header.autoclear_features; 920 921 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) { 922 void *feature_table = NULL; 923 qcow2_read_extensions(bs, header.header_length, ext_end, 924 &feature_table, NULL); 925 report_unsupported_feature(errp, feature_table, 926 s->incompatible_features & 927 ~QCOW2_INCOMPAT_MASK); 928 ret = -ENOTSUP; 929 g_free(feature_table); 930 goto fail; 931 } 932 933 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 934 /* Corrupt images may not be written to unless they are being repaired 935 */ 936 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 937 error_setg(errp, "qcow2: Image is corrupt; cannot be opened " 938 "read/write"); 939 ret = -EACCES; 940 goto fail; 941 } 942 } 943 944 /* Check support for various header values */ 945 if (header.refcount_order > 6) { 946 error_setg(errp, "Reference count entry width too large; may not " 947 "exceed 64 bits"); 948 ret = -EINVAL; 949 goto fail; 950 } 951 s->refcount_order = header.refcount_order; 952 s->refcount_bits = 1 << s->refcount_order; 953 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 954 s->refcount_max += s->refcount_max - 1; 955 956 if (header.crypt_method > QCOW_CRYPT_AES) { 957 error_setg(errp, "Unsupported encryption method: %" PRIu32, 958 header.crypt_method); 959 ret = -EINVAL; 960 goto fail; 961 } 962 if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) { 963 error_setg(errp, "AES cipher not available"); 964 ret = -EINVAL; 965 goto fail; 966 } 967 s->crypt_method_header = header.crypt_method; 968 if (s->crypt_method_header) { 969 if (bdrv_uses_whitelist() && 970 s->crypt_method_header == QCOW_CRYPT_AES) { 971 error_setg(errp, 972 "Use of AES-CBC encrypted qcow2 images is no longer " 973 "supported in system emulators"); 974 error_append_hint(errp, 975 "You can use 'qemu-img convert' to convert your " 976 "image to an alternative supported format, such " 977 "as unencrypted qcow2, or raw with the LUKS " 978 "format instead.\n"); 979 ret = -ENOSYS; 980 goto fail; 981 } 982 983 bs->encrypted = 1; 984 985 /* Encryption works on a sector granularity */ 986 bs->request_alignment = BDRV_SECTOR_SIZE; 987 } 988 989 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ 990 s->l2_size = 1 << s->l2_bits; 991 /* 2^(s->refcount_order - 3) is the refcount width in bytes */ 992 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3); 993 s->refcount_block_size = 1 << s->refcount_block_bits; 994 bs->total_sectors = header.size / 512; 995 s->csize_shift = (62 - (s->cluster_bits - 8)); 996 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; 997 s->cluster_offset_mask = (1LL << s->csize_shift) - 1; 998 999 s->refcount_table_offset = header.refcount_table_offset; 1000 s->refcount_table_size = 1001 header.refcount_table_clusters << (s->cluster_bits - 3); 1002 1003 if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) { 1004 error_setg(errp, "Reference count table too large"); 1005 ret = -EINVAL; 1006 goto fail; 1007 } 1008 1009 ret = validate_table_offset(bs, s->refcount_table_offset, 1010 s->refcount_table_size, sizeof(uint64_t)); 1011 if (ret < 0) { 1012 error_setg(errp, "Invalid reference count table offset"); 1013 goto fail; 1014 } 1015 1016 /* Snapshot table offset/length */ 1017 if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) { 1018 error_setg(errp, "Too many snapshots"); 1019 ret = -EINVAL; 1020 goto fail; 1021 } 1022 1023 ret = validate_table_offset(bs, header.snapshots_offset, 1024 header.nb_snapshots, 1025 sizeof(QCowSnapshotHeader)); 1026 if (ret < 0) { 1027 error_setg(errp, "Invalid snapshot table offset"); 1028 goto fail; 1029 } 1030 1031 /* read the level 1 table */ 1032 if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) { 1033 error_setg(errp, "Active L1 table too large"); 1034 ret = -EFBIG; 1035 goto fail; 1036 } 1037 s->l1_size = header.l1_size; 1038 1039 l1_vm_state_index = size_to_l1(s, header.size); 1040 if (l1_vm_state_index > INT_MAX) { 1041 error_setg(errp, "Image is too big"); 1042 ret = -EFBIG; 1043 goto fail; 1044 } 1045 s->l1_vm_state_index = l1_vm_state_index; 1046 1047 /* the L1 table must contain at least enough entries to put 1048 header.size bytes */ 1049 if (s->l1_size < s->l1_vm_state_index) { 1050 error_setg(errp, "L1 table is too small"); 1051 ret = -EINVAL; 1052 goto fail; 1053 } 1054 1055 ret = validate_table_offset(bs, header.l1_table_offset, 1056 header.l1_size, sizeof(uint64_t)); 1057 if (ret < 0) { 1058 error_setg(errp, "Invalid L1 table offset"); 1059 goto fail; 1060 } 1061 s->l1_table_offset = header.l1_table_offset; 1062 1063 1064 if (s->l1_size > 0) { 1065 s->l1_table = qemu_try_blockalign(bs->file->bs, 1066 align_offset(s->l1_size * sizeof(uint64_t), 512)); 1067 if (s->l1_table == NULL) { 1068 error_setg(errp, "Could not allocate L1 table"); 1069 ret = -ENOMEM; 1070 goto fail; 1071 } 1072 ret = bdrv_pread(bs->file->bs, s->l1_table_offset, s->l1_table, 1073 s->l1_size * sizeof(uint64_t)); 1074 if (ret < 0) { 1075 error_setg_errno(errp, -ret, "Could not read L1 table"); 1076 goto fail; 1077 } 1078 for(i = 0;i < s->l1_size; i++) { 1079 be64_to_cpus(&s->l1_table[i]); 1080 } 1081 } 1082 1083 /* Parse driver-specific options */ 1084 ret = qcow2_update_options(bs, options, flags, errp); 1085 if (ret < 0) { 1086 goto fail; 1087 } 1088 1089 s->cluster_cache = g_malloc(s->cluster_size); 1090 /* one more sector for decompressed data alignment */ 1091 s->cluster_data = qemu_try_blockalign(bs->file->bs, QCOW_MAX_CRYPT_CLUSTERS 1092 * s->cluster_size + 512); 1093 if (s->cluster_data == NULL) { 1094 error_setg(errp, "Could not allocate temporary cluster buffer"); 1095 ret = -ENOMEM; 1096 goto fail; 1097 } 1098 1099 s->cluster_cache_offset = -1; 1100 s->flags = flags; 1101 1102 ret = qcow2_refcount_init(bs); 1103 if (ret != 0) { 1104 error_setg_errno(errp, -ret, "Could not initialize refcount handling"); 1105 goto fail; 1106 } 1107 1108 QLIST_INIT(&s->cluster_allocs); 1109 QTAILQ_INIT(&s->discards); 1110 1111 /* read qcow2 extensions */ 1112 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, 1113 &local_err)) { 1114 error_propagate(errp, local_err); 1115 ret = -EINVAL; 1116 goto fail; 1117 } 1118 1119 /* read the backing file name */ 1120 if (header.backing_file_offset != 0) { 1121 len = header.backing_file_size; 1122 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || 1123 len >= sizeof(bs->backing_file)) { 1124 error_setg(errp, "Backing file name too long"); 1125 ret = -EINVAL; 1126 goto fail; 1127 } 1128 ret = bdrv_pread(bs->file->bs, header.backing_file_offset, 1129 bs->backing_file, len); 1130 if (ret < 0) { 1131 error_setg_errno(errp, -ret, "Could not read backing file name"); 1132 goto fail; 1133 } 1134 bs->backing_file[len] = '\0'; 1135 s->image_backing_file = g_strdup(bs->backing_file); 1136 } 1137 1138 /* Internal snapshots */ 1139 s->snapshots_offset = header.snapshots_offset; 1140 s->nb_snapshots = header.nb_snapshots; 1141 1142 ret = qcow2_read_snapshots(bs); 1143 if (ret < 0) { 1144 error_setg_errno(errp, -ret, "Could not read snapshots"); 1145 goto fail; 1146 } 1147 1148 /* Clear unknown autoclear feature bits */ 1149 if (!bs->read_only && !(flags & BDRV_O_INACTIVE) && s->autoclear_features) { 1150 s->autoclear_features = 0; 1151 ret = qcow2_update_header(bs); 1152 if (ret < 0) { 1153 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 1154 goto fail; 1155 } 1156 } 1157 1158 /* Initialise locks */ 1159 qemu_co_mutex_init(&s->lock); 1160 1161 /* Repair image if dirty */ 1162 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only && 1163 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { 1164 BdrvCheckResult result = {0}; 1165 1166 ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); 1167 if (ret < 0) { 1168 error_setg_errno(errp, -ret, "Could not repair dirty image"); 1169 goto fail; 1170 } 1171 } 1172 1173 #ifdef DEBUG_ALLOC 1174 { 1175 BdrvCheckResult result = {0}; 1176 qcow2_check_refcounts(bs, &result, 0); 1177 } 1178 #endif 1179 return ret; 1180 1181 fail: 1182 g_free(s->unknown_header_fields); 1183 cleanup_unknown_header_ext(bs); 1184 qcow2_free_snapshots(bs); 1185 qcow2_refcount_close(bs); 1186 qemu_vfree(s->l1_table); 1187 /* else pre-write overlap checks in cache_destroy may crash */ 1188 s->l1_table = NULL; 1189 cache_clean_timer_del(bs); 1190 if (s->l2_table_cache) { 1191 qcow2_cache_destroy(bs, s->l2_table_cache); 1192 } 1193 if (s->refcount_block_cache) { 1194 qcow2_cache_destroy(bs, s->refcount_block_cache); 1195 } 1196 g_free(s->cluster_cache); 1197 qemu_vfree(s->cluster_data); 1198 return ret; 1199 } 1200 1201 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) 1202 { 1203 BDRVQcow2State *s = bs->opaque; 1204 1205 bs->bl.pwrite_zeroes_alignment = s->cluster_size; 1206 } 1207 1208 static int qcow2_set_key(BlockDriverState *bs, const char *key) 1209 { 1210 BDRVQcow2State *s = bs->opaque; 1211 uint8_t keybuf[16]; 1212 int len, i; 1213 Error *err = NULL; 1214 1215 memset(keybuf, 0, 16); 1216 len = strlen(key); 1217 if (len > 16) 1218 len = 16; 1219 /* XXX: we could compress the chars to 7 bits to increase 1220 entropy */ 1221 for(i = 0;i < len;i++) { 1222 keybuf[i] = key[i]; 1223 } 1224 assert(bs->encrypted); 1225 1226 qcrypto_cipher_free(s->cipher); 1227 s->cipher = qcrypto_cipher_new( 1228 QCRYPTO_CIPHER_ALG_AES_128, 1229 QCRYPTO_CIPHER_MODE_CBC, 1230 keybuf, G_N_ELEMENTS(keybuf), 1231 &err); 1232 1233 if (!s->cipher) { 1234 /* XXX would be nice if errors in this method could 1235 * be properly propagate to the caller. Would need 1236 * the bdrv_set_key() API signature to be fixed. */ 1237 error_free(err); 1238 return -1; 1239 } 1240 return 0; 1241 } 1242 1243 static int qcow2_reopen_prepare(BDRVReopenState *state, 1244 BlockReopenQueue *queue, Error **errp) 1245 { 1246 Qcow2ReopenState *r; 1247 int ret; 1248 1249 r = g_new0(Qcow2ReopenState, 1); 1250 state->opaque = r; 1251 1252 ret = qcow2_update_options_prepare(state->bs, r, state->options, 1253 state->flags, errp); 1254 if (ret < 0) { 1255 goto fail; 1256 } 1257 1258 /* We need to write out any unwritten data if we reopen read-only. */ 1259 if ((state->flags & BDRV_O_RDWR) == 0) { 1260 ret = bdrv_flush(state->bs); 1261 if (ret < 0) { 1262 goto fail; 1263 } 1264 1265 ret = qcow2_mark_clean(state->bs); 1266 if (ret < 0) { 1267 goto fail; 1268 } 1269 } 1270 1271 return 0; 1272 1273 fail: 1274 qcow2_update_options_abort(state->bs, r); 1275 g_free(r); 1276 return ret; 1277 } 1278 1279 static void qcow2_reopen_commit(BDRVReopenState *state) 1280 { 1281 qcow2_update_options_commit(state->bs, state->opaque); 1282 g_free(state->opaque); 1283 } 1284 1285 static void qcow2_reopen_abort(BDRVReopenState *state) 1286 { 1287 qcow2_update_options_abort(state->bs, state->opaque); 1288 g_free(state->opaque); 1289 } 1290 1291 static void qcow2_join_options(QDict *options, QDict *old_options) 1292 { 1293 bool has_new_overlap_template = 1294 qdict_haskey(options, QCOW2_OPT_OVERLAP) || 1295 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE); 1296 bool has_new_total_cache_size = 1297 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE); 1298 bool has_all_cache_options; 1299 1300 /* New overlap template overrides all old overlap options */ 1301 if (has_new_overlap_template) { 1302 qdict_del(old_options, QCOW2_OPT_OVERLAP); 1303 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE); 1304 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER); 1305 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1); 1306 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2); 1307 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE); 1308 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK); 1309 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE); 1310 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1); 1311 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2); 1312 } 1313 1314 /* New total cache size overrides all old options */ 1315 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) { 1316 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE); 1317 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1318 } 1319 1320 qdict_join(options, old_options, false); 1321 1322 /* 1323 * If after merging all cache size options are set, an old total size is 1324 * overwritten. Do keep all options, however, if all three are new. The 1325 * resulting error message is what we want to happen. 1326 */ 1327 has_all_cache_options = 1328 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) || 1329 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) || 1330 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1331 1332 if (has_all_cache_options && !has_new_total_cache_size) { 1333 qdict_del(options, QCOW2_OPT_CACHE_SIZE); 1334 } 1335 } 1336 1337 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs, 1338 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file) 1339 { 1340 BDRVQcow2State *s = bs->opaque; 1341 uint64_t cluster_offset; 1342 int index_in_cluster, ret; 1343 unsigned int bytes; 1344 int64_t status = 0; 1345 1346 bytes = MIN(INT_MAX, nb_sectors * BDRV_SECTOR_SIZE); 1347 qemu_co_mutex_lock(&s->lock); 1348 ret = qcow2_get_cluster_offset(bs, sector_num << 9, &bytes, 1349 &cluster_offset); 1350 qemu_co_mutex_unlock(&s->lock); 1351 if (ret < 0) { 1352 return ret; 1353 } 1354 1355 *pnum = bytes >> BDRV_SECTOR_BITS; 1356 1357 if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED && 1358 !s->cipher) { 1359 index_in_cluster = sector_num & (s->cluster_sectors - 1); 1360 cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS); 1361 *file = bs->file->bs; 1362 status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset; 1363 } 1364 if (ret == QCOW2_CLUSTER_ZERO) { 1365 status |= BDRV_BLOCK_ZERO; 1366 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) { 1367 status |= BDRV_BLOCK_DATA; 1368 } 1369 return status; 1370 } 1371 1372 /* handle reading after the end of the backing file */ 1373 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov, 1374 int64_t offset, int bytes) 1375 { 1376 uint64_t bs_size = bs->total_sectors * BDRV_SECTOR_SIZE; 1377 int n1; 1378 1379 if ((offset + bytes) <= bs_size) { 1380 return bytes; 1381 } 1382 1383 if (offset >= bs_size) { 1384 n1 = 0; 1385 } else { 1386 n1 = bs_size - offset; 1387 } 1388 1389 qemu_iovec_memset(qiov, n1, 0, bytes - n1); 1390 1391 return n1; 1392 } 1393 1394 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset, 1395 uint64_t bytes, QEMUIOVector *qiov, 1396 int flags) 1397 { 1398 BDRVQcow2State *s = bs->opaque; 1399 int offset_in_cluster, n1; 1400 int ret; 1401 unsigned int cur_bytes; /* number of bytes in current iteration */ 1402 uint64_t cluster_offset = 0; 1403 uint64_t bytes_done = 0; 1404 QEMUIOVector hd_qiov; 1405 uint8_t *cluster_data = NULL; 1406 1407 qemu_iovec_init(&hd_qiov, qiov->niov); 1408 1409 qemu_co_mutex_lock(&s->lock); 1410 1411 while (bytes != 0) { 1412 1413 /* prepare next request */ 1414 cur_bytes = MIN(bytes, INT_MAX); 1415 if (s->cipher) { 1416 cur_bytes = MIN(cur_bytes, 1417 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1418 } 1419 1420 ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset); 1421 if (ret < 0) { 1422 goto fail; 1423 } 1424 1425 offset_in_cluster = offset_into_cluster(s, offset); 1426 1427 qemu_iovec_reset(&hd_qiov); 1428 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 1429 1430 switch (ret) { 1431 case QCOW2_CLUSTER_UNALLOCATED: 1432 1433 if (bs->backing) { 1434 /* read from the base image */ 1435 n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov, 1436 offset, cur_bytes); 1437 if (n1 > 0) { 1438 QEMUIOVector local_qiov; 1439 1440 qemu_iovec_init(&local_qiov, hd_qiov.niov); 1441 qemu_iovec_concat(&local_qiov, &hd_qiov, 0, n1); 1442 1443 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 1444 qemu_co_mutex_unlock(&s->lock); 1445 ret = bdrv_co_preadv(bs->backing->bs, offset, n1, 1446 &local_qiov, 0); 1447 qemu_co_mutex_lock(&s->lock); 1448 1449 qemu_iovec_destroy(&local_qiov); 1450 1451 if (ret < 0) { 1452 goto fail; 1453 } 1454 } 1455 } else { 1456 /* Note: in this case, no need to wait */ 1457 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 1458 } 1459 break; 1460 1461 case QCOW2_CLUSTER_ZERO: 1462 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 1463 break; 1464 1465 case QCOW2_CLUSTER_COMPRESSED: 1466 /* add AIO support for compressed blocks ? */ 1467 ret = qcow2_decompress_cluster(bs, cluster_offset); 1468 if (ret < 0) { 1469 goto fail; 1470 } 1471 1472 qemu_iovec_from_buf(&hd_qiov, 0, 1473 s->cluster_cache + offset_in_cluster, 1474 cur_bytes); 1475 break; 1476 1477 case QCOW2_CLUSTER_NORMAL: 1478 if ((cluster_offset & 511) != 0) { 1479 ret = -EIO; 1480 goto fail; 1481 } 1482 1483 if (bs->encrypted) { 1484 assert(s->cipher); 1485 1486 /* 1487 * For encrypted images, read everything into a temporary 1488 * contiguous buffer on which the AES functions can work. 1489 */ 1490 if (!cluster_data) { 1491 cluster_data = 1492 qemu_try_blockalign(bs->file->bs, 1493 QCOW_MAX_CRYPT_CLUSTERS 1494 * s->cluster_size); 1495 if (cluster_data == NULL) { 1496 ret = -ENOMEM; 1497 goto fail; 1498 } 1499 } 1500 1501 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1502 qemu_iovec_reset(&hd_qiov); 1503 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 1504 } 1505 1506 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 1507 qemu_co_mutex_unlock(&s->lock); 1508 ret = bdrv_co_preadv(bs->file->bs, 1509 cluster_offset + offset_in_cluster, 1510 cur_bytes, &hd_qiov, 0); 1511 qemu_co_mutex_lock(&s->lock); 1512 if (ret < 0) { 1513 goto fail; 1514 } 1515 if (bs->encrypted) { 1516 assert(s->cipher); 1517 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); 1518 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0); 1519 Error *err = NULL; 1520 if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS, 1521 cluster_data, cluster_data, 1522 cur_bytes >> BDRV_SECTOR_BITS, 1523 false, &err) < 0) { 1524 error_free(err); 1525 ret = -EIO; 1526 goto fail; 1527 } 1528 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes); 1529 } 1530 break; 1531 1532 default: 1533 g_assert_not_reached(); 1534 ret = -EIO; 1535 goto fail; 1536 } 1537 1538 bytes -= cur_bytes; 1539 offset += cur_bytes; 1540 bytes_done += cur_bytes; 1541 } 1542 ret = 0; 1543 1544 fail: 1545 qemu_co_mutex_unlock(&s->lock); 1546 1547 qemu_iovec_destroy(&hd_qiov); 1548 qemu_vfree(cluster_data); 1549 1550 return ret; 1551 } 1552 1553 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset, 1554 uint64_t bytes, QEMUIOVector *qiov, 1555 int flags) 1556 { 1557 BDRVQcow2State *s = bs->opaque; 1558 int offset_in_cluster; 1559 int ret; 1560 unsigned int cur_bytes; /* number of sectors in current iteration */ 1561 uint64_t cluster_offset; 1562 QEMUIOVector hd_qiov; 1563 uint64_t bytes_done = 0; 1564 uint8_t *cluster_data = NULL; 1565 QCowL2Meta *l2meta = NULL; 1566 1567 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes); 1568 1569 qemu_iovec_init(&hd_qiov, qiov->niov); 1570 1571 s->cluster_cache_offset = -1; /* disable compressed cache */ 1572 1573 qemu_co_mutex_lock(&s->lock); 1574 1575 while (bytes != 0) { 1576 1577 l2meta = NULL; 1578 1579 trace_qcow2_writev_start_part(qemu_coroutine_self()); 1580 offset_in_cluster = offset_into_cluster(s, offset); 1581 cur_bytes = MIN(bytes, INT_MAX); 1582 if (bs->encrypted) { 1583 cur_bytes = MIN(cur_bytes, 1584 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size 1585 - offset_in_cluster); 1586 } 1587 1588 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 1589 &cluster_offset, &l2meta); 1590 if (ret < 0) { 1591 goto fail; 1592 } 1593 1594 assert((cluster_offset & 511) == 0); 1595 1596 qemu_iovec_reset(&hd_qiov); 1597 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 1598 1599 if (bs->encrypted) { 1600 Error *err = NULL; 1601 assert(s->cipher); 1602 if (!cluster_data) { 1603 cluster_data = qemu_try_blockalign(bs->file->bs, 1604 QCOW_MAX_CRYPT_CLUSTERS 1605 * s->cluster_size); 1606 if (cluster_data == NULL) { 1607 ret = -ENOMEM; 1608 goto fail; 1609 } 1610 } 1611 1612 assert(hd_qiov.size <= 1613 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1614 qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size); 1615 1616 if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS, 1617 cluster_data, cluster_data, 1618 cur_bytes >>BDRV_SECTOR_BITS, 1619 true, &err) < 0) { 1620 error_free(err); 1621 ret = -EIO; 1622 goto fail; 1623 } 1624 1625 qemu_iovec_reset(&hd_qiov); 1626 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 1627 } 1628 1629 ret = qcow2_pre_write_overlap_check(bs, 0, 1630 cluster_offset + offset_in_cluster, cur_bytes); 1631 if (ret < 0) { 1632 goto fail; 1633 } 1634 1635 qemu_co_mutex_unlock(&s->lock); 1636 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 1637 trace_qcow2_writev_data(qemu_coroutine_self(), 1638 cluster_offset + offset_in_cluster); 1639 ret = bdrv_co_pwritev(bs->file->bs, 1640 cluster_offset + offset_in_cluster, 1641 cur_bytes, &hd_qiov, 0); 1642 qemu_co_mutex_lock(&s->lock); 1643 if (ret < 0) { 1644 goto fail; 1645 } 1646 1647 while (l2meta != NULL) { 1648 QCowL2Meta *next; 1649 1650 ret = qcow2_alloc_cluster_link_l2(bs, l2meta); 1651 if (ret < 0) { 1652 goto fail; 1653 } 1654 1655 /* Take the request off the list of running requests */ 1656 if (l2meta->nb_clusters != 0) { 1657 QLIST_REMOVE(l2meta, next_in_flight); 1658 } 1659 1660 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1661 1662 next = l2meta->next; 1663 g_free(l2meta); 1664 l2meta = next; 1665 } 1666 1667 bytes -= cur_bytes; 1668 offset += cur_bytes; 1669 bytes_done += cur_bytes; 1670 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes); 1671 } 1672 ret = 0; 1673 1674 fail: 1675 qemu_co_mutex_unlock(&s->lock); 1676 1677 while (l2meta != NULL) { 1678 QCowL2Meta *next; 1679 1680 if (l2meta->nb_clusters != 0) { 1681 QLIST_REMOVE(l2meta, next_in_flight); 1682 } 1683 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1684 1685 next = l2meta->next; 1686 g_free(l2meta); 1687 l2meta = next; 1688 } 1689 1690 qemu_iovec_destroy(&hd_qiov); 1691 qemu_vfree(cluster_data); 1692 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 1693 1694 return ret; 1695 } 1696 1697 static int qcow2_inactivate(BlockDriverState *bs) 1698 { 1699 BDRVQcow2State *s = bs->opaque; 1700 int ret, result = 0; 1701 1702 ret = qcow2_cache_flush(bs, s->l2_table_cache); 1703 if (ret) { 1704 result = ret; 1705 error_report("Failed to flush the L2 table cache: %s", 1706 strerror(-ret)); 1707 } 1708 1709 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 1710 if (ret) { 1711 result = ret; 1712 error_report("Failed to flush the refcount block cache: %s", 1713 strerror(-ret)); 1714 } 1715 1716 if (result == 0) { 1717 qcow2_mark_clean(bs); 1718 } 1719 1720 return result; 1721 } 1722 1723 static void qcow2_close(BlockDriverState *bs) 1724 { 1725 BDRVQcow2State *s = bs->opaque; 1726 qemu_vfree(s->l1_table); 1727 /* else pre-write overlap checks in cache_destroy may crash */ 1728 s->l1_table = NULL; 1729 1730 if (!(s->flags & BDRV_O_INACTIVE)) { 1731 qcow2_inactivate(bs); 1732 } 1733 1734 cache_clean_timer_del(bs); 1735 qcow2_cache_destroy(bs, s->l2_table_cache); 1736 qcow2_cache_destroy(bs, s->refcount_block_cache); 1737 1738 qcrypto_cipher_free(s->cipher); 1739 s->cipher = NULL; 1740 1741 g_free(s->unknown_header_fields); 1742 cleanup_unknown_header_ext(bs); 1743 1744 g_free(s->image_backing_file); 1745 g_free(s->image_backing_format); 1746 1747 g_free(s->cluster_cache); 1748 qemu_vfree(s->cluster_data); 1749 qcow2_refcount_close(bs); 1750 qcow2_free_snapshots(bs); 1751 } 1752 1753 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp) 1754 { 1755 BDRVQcow2State *s = bs->opaque; 1756 int flags = s->flags; 1757 QCryptoCipher *cipher = NULL; 1758 QDict *options; 1759 Error *local_err = NULL; 1760 int ret; 1761 1762 /* 1763 * Backing files are read-only which makes all of their metadata immutable, 1764 * that means we don't have to worry about reopening them here. 1765 */ 1766 1767 cipher = s->cipher; 1768 s->cipher = NULL; 1769 1770 qcow2_close(bs); 1771 1772 memset(s, 0, sizeof(BDRVQcow2State)); 1773 options = qdict_clone_shallow(bs->options); 1774 1775 flags &= ~BDRV_O_INACTIVE; 1776 ret = qcow2_open(bs, options, flags, &local_err); 1777 QDECREF(options); 1778 if (local_err) { 1779 error_propagate(errp, local_err); 1780 error_prepend(errp, "Could not reopen qcow2 layer: "); 1781 bs->drv = NULL; 1782 return; 1783 } else if (ret < 0) { 1784 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer"); 1785 bs->drv = NULL; 1786 return; 1787 } 1788 1789 s->cipher = cipher; 1790 } 1791 1792 static size_t header_ext_add(char *buf, uint32_t magic, const void *s, 1793 size_t len, size_t buflen) 1794 { 1795 QCowExtension *ext_backing_fmt = (QCowExtension*) buf; 1796 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); 1797 1798 if (buflen < ext_len) { 1799 return -ENOSPC; 1800 } 1801 1802 *ext_backing_fmt = (QCowExtension) { 1803 .magic = cpu_to_be32(magic), 1804 .len = cpu_to_be32(len), 1805 }; 1806 memcpy(buf + sizeof(QCowExtension), s, len); 1807 1808 return ext_len; 1809 } 1810 1811 /* 1812 * Updates the qcow2 header, including the variable length parts of it, i.e. 1813 * the backing file name and all extensions. qcow2 was not designed to allow 1814 * such changes, so if we run out of space (we can only use the first cluster) 1815 * this function may fail. 1816 * 1817 * Returns 0 on success, -errno in error cases. 1818 */ 1819 int qcow2_update_header(BlockDriverState *bs) 1820 { 1821 BDRVQcow2State *s = bs->opaque; 1822 QCowHeader *header; 1823 char *buf; 1824 size_t buflen = s->cluster_size; 1825 int ret; 1826 uint64_t total_size; 1827 uint32_t refcount_table_clusters; 1828 size_t header_length; 1829 Qcow2UnknownHeaderExtension *uext; 1830 1831 buf = qemu_blockalign(bs, buflen); 1832 1833 /* Header structure */ 1834 header = (QCowHeader*) buf; 1835 1836 if (buflen < sizeof(*header)) { 1837 ret = -ENOSPC; 1838 goto fail; 1839 } 1840 1841 header_length = sizeof(*header) + s->unknown_header_fields_size; 1842 total_size = bs->total_sectors * BDRV_SECTOR_SIZE; 1843 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); 1844 1845 *header = (QCowHeader) { 1846 /* Version 2 fields */ 1847 .magic = cpu_to_be32(QCOW_MAGIC), 1848 .version = cpu_to_be32(s->qcow_version), 1849 .backing_file_offset = 0, 1850 .backing_file_size = 0, 1851 .cluster_bits = cpu_to_be32(s->cluster_bits), 1852 .size = cpu_to_be64(total_size), 1853 .crypt_method = cpu_to_be32(s->crypt_method_header), 1854 .l1_size = cpu_to_be32(s->l1_size), 1855 .l1_table_offset = cpu_to_be64(s->l1_table_offset), 1856 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), 1857 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), 1858 .nb_snapshots = cpu_to_be32(s->nb_snapshots), 1859 .snapshots_offset = cpu_to_be64(s->snapshots_offset), 1860 1861 /* Version 3 fields */ 1862 .incompatible_features = cpu_to_be64(s->incompatible_features), 1863 .compatible_features = cpu_to_be64(s->compatible_features), 1864 .autoclear_features = cpu_to_be64(s->autoclear_features), 1865 .refcount_order = cpu_to_be32(s->refcount_order), 1866 .header_length = cpu_to_be32(header_length), 1867 }; 1868 1869 /* For older versions, write a shorter header */ 1870 switch (s->qcow_version) { 1871 case 2: 1872 ret = offsetof(QCowHeader, incompatible_features); 1873 break; 1874 case 3: 1875 ret = sizeof(*header); 1876 break; 1877 default: 1878 ret = -EINVAL; 1879 goto fail; 1880 } 1881 1882 buf += ret; 1883 buflen -= ret; 1884 memset(buf, 0, buflen); 1885 1886 /* Preserve any unknown field in the header */ 1887 if (s->unknown_header_fields_size) { 1888 if (buflen < s->unknown_header_fields_size) { 1889 ret = -ENOSPC; 1890 goto fail; 1891 } 1892 1893 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); 1894 buf += s->unknown_header_fields_size; 1895 buflen -= s->unknown_header_fields_size; 1896 } 1897 1898 /* Backing file format header extension */ 1899 if (s->image_backing_format) { 1900 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, 1901 s->image_backing_format, 1902 strlen(s->image_backing_format), 1903 buflen); 1904 if (ret < 0) { 1905 goto fail; 1906 } 1907 1908 buf += ret; 1909 buflen -= ret; 1910 } 1911 1912 /* Feature table */ 1913 if (s->qcow_version >= 3) { 1914 Qcow2Feature features[] = { 1915 { 1916 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 1917 .bit = QCOW2_INCOMPAT_DIRTY_BITNR, 1918 .name = "dirty bit", 1919 }, 1920 { 1921 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 1922 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, 1923 .name = "corrupt bit", 1924 }, 1925 { 1926 .type = QCOW2_FEAT_TYPE_COMPATIBLE, 1927 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 1928 .name = "lazy refcounts", 1929 }, 1930 }; 1931 1932 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, 1933 features, sizeof(features), buflen); 1934 if (ret < 0) { 1935 goto fail; 1936 } 1937 buf += ret; 1938 buflen -= ret; 1939 } 1940 1941 /* Keep unknown header extensions */ 1942 QLIST_FOREACH(uext, &s->unknown_header_ext, next) { 1943 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); 1944 if (ret < 0) { 1945 goto fail; 1946 } 1947 1948 buf += ret; 1949 buflen -= ret; 1950 } 1951 1952 /* End of header extensions */ 1953 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); 1954 if (ret < 0) { 1955 goto fail; 1956 } 1957 1958 buf += ret; 1959 buflen -= ret; 1960 1961 /* Backing file name */ 1962 if (s->image_backing_file) { 1963 size_t backing_file_len = strlen(s->image_backing_file); 1964 1965 if (buflen < backing_file_len) { 1966 ret = -ENOSPC; 1967 goto fail; 1968 } 1969 1970 /* Using strncpy is ok here, since buf is not NUL-terminated. */ 1971 strncpy(buf, s->image_backing_file, buflen); 1972 1973 header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); 1974 header->backing_file_size = cpu_to_be32(backing_file_len); 1975 } 1976 1977 /* Write the new header */ 1978 ret = bdrv_pwrite(bs->file->bs, 0, header, s->cluster_size); 1979 if (ret < 0) { 1980 goto fail; 1981 } 1982 1983 ret = 0; 1984 fail: 1985 qemu_vfree(header); 1986 return ret; 1987 } 1988 1989 static int qcow2_change_backing_file(BlockDriverState *bs, 1990 const char *backing_file, const char *backing_fmt) 1991 { 1992 BDRVQcow2State *s = bs->opaque; 1993 1994 if (backing_file && strlen(backing_file) > 1023) { 1995 return -EINVAL; 1996 } 1997 1998 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 1999 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 2000 2001 g_free(s->image_backing_file); 2002 g_free(s->image_backing_format); 2003 2004 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL; 2005 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL; 2006 2007 return qcow2_update_header(bs); 2008 } 2009 2010 static int preallocate(BlockDriverState *bs) 2011 { 2012 uint64_t bytes; 2013 uint64_t offset; 2014 uint64_t host_offset = 0; 2015 unsigned int cur_bytes; 2016 int ret; 2017 QCowL2Meta *meta; 2018 2019 bytes = bdrv_getlength(bs); 2020 offset = 0; 2021 2022 while (bytes) { 2023 cur_bytes = MIN(bytes, INT_MAX); 2024 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 2025 &host_offset, &meta); 2026 if (ret < 0) { 2027 return ret; 2028 } 2029 2030 while (meta) { 2031 QCowL2Meta *next = meta->next; 2032 2033 ret = qcow2_alloc_cluster_link_l2(bs, meta); 2034 if (ret < 0) { 2035 qcow2_free_any_clusters(bs, meta->alloc_offset, 2036 meta->nb_clusters, QCOW2_DISCARD_NEVER); 2037 return ret; 2038 } 2039 2040 /* There are no dependent requests, but we need to remove our 2041 * request from the list of in-flight requests */ 2042 QLIST_REMOVE(meta, next_in_flight); 2043 2044 g_free(meta); 2045 meta = next; 2046 } 2047 2048 /* TODO Preallocate data if requested */ 2049 2050 bytes -= cur_bytes; 2051 offset += cur_bytes; 2052 } 2053 2054 /* 2055 * It is expected that the image file is large enough to actually contain 2056 * all of the allocated clusters (otherwise we get failing reads after 2057 * EOF). Extend the image to the last allocated sector. 2058 */ 2059 if (host_offset != 0) { 2060 uint8_t data = 0; 2061 ret = bdrv_pwrite(bs->file->bs, (host_offset + cur_bytes) - 1, 2062 &data, 1); 2063 if (ret < 0) { 2064 return ret; 2065 } 2066 } 2067 2068 return 0; 2069 } 2070 2071 static int qcow2_create2(const char *filename, int64_t total_size, 2072 const char *backing_file, const char *backing_format, 2073 int flags, size_t cluster_size, PreallocMode prealloc, 2074 QemuOpts *opts, int version, int refcount_order, 2075 Error **errp) 2076 { 2077 int cluster_bits; 2078 QDict *options; 2079 2080 /* Calculate cluster_bits */ 2081 cluster_bits = ctz32(cluster_size); 2082 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 2083 (1 << cluster_bits) != cluster_size) 2084 { 2085 error_setg(errp, "Cluster size must be a power of two between %d and " 2086 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 2087 return -EINVAL; 2088 } 2089 2090 /* 2091 * Open the image file and write a minimal qcow2 header. 2092 * 2093 * We keep things simple and start with a zero-sized image. We also 2094 * do without refcount blocks or a L1 table for now. We'll fix the 2095 * inconsistency later. 2096 * 2097 * We do need a refcount table because growing the refcount table means 2098 * allocating two new refcount blocks - the seconds of which would be at 2099 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 2100 * size for any qcow2 image. 2101 */ 2102 BlockBackend *blk; 2103 QCowHeader *header; 2104 uint64_t* refcount_table; 2105 Error *local_err = NULL; 2106 int ret; 2107 2108 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 2109 /* Note: The following calculation does not need to be exact; if it is a 2110 * bit off, either some bytes will be "leaked" (which is fine) or we 2111 * will need to increase the file size by some bytes (which is fine, 2112 * too, as long as the bulk is allocated here). Therefore, using 2113 * floating point arithmetic is fine. */ 2114 int64_t meta_size = 0; 2115 uint64_t nreftablee, nrefblocke, nl1e, nl2e; 2116 int64_t aligned_total_size = align_offset(total_size, cluster_size); 2117 int refblock_bits, refblock_size; 2118 /* refcount entry size in bytes */ 2119 double rces = (1 << refcount_order) / 8.; 2120 2121 /* see qcow2_open() */ 2122 refblock_bits = cluster_bits - (refcount_order - 3); 2123 refblock_size = 1 << refblock_bits; 2124 2125 /* header: 1 cluster */ 2126 meta_size += cluster_size; 2127 2128 /* total size of L2 tables */ 2129 nl2e = aligned_total_size / cluster_size; 2130 nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t)); 2131 meta_size += nl2e * sizeof(uint64_t); 2132 2133 /* total size of L1 tables */ 2134 nl1e = nl2e * sizeof(uint64_t) / cluster_size; 2135 nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t)); 2136 meta_size += nl1e * sizeof(uint64_t); 2137 2138 /* total size of refcount blocks 2139 * 2140 * note: every host cluster is reference-counted, including metadata 2141 * (even refcount blocks are recursively included). 2142 * Let: 2143 * a = total_size (this is the guest disk size) 2144 * m = meta size not including refcount blocks and refcount tables 2145 * c = cluster size 2146 * y1 = number of refcount blocks entries 2147 * y2 = meta size including everything 2148 * rces = refcount entry size in bytes 2149 * then, 2150 * y1 = (y2 + a)/c 2151 * y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m 2152 * we can get y1: 2153 * y1 = (a + m) / (c - rces - rces * sizeof(u64) / c) 2154 */ 2155 nrefblocke = (aligned_total_size + meta_size + cluster_size) 2156 / (cluster_size - rces - rces * sizeof(uint64_t) 2157 / cluster_size); 2158 meta_size += DIV_ROUND_UP(nrefblocke, refblock_size) * cluster_size; 2159 2160 /* total size of refcount tables */ 2161 nreftablee = nrefblocke / refblock_size; 2162 nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t)); 2163 meta_size += nreftablee * sizeof(uint64_t); 2164 2165 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, 2166 aligned_total_size + meta_size, &error_abort); 2167 qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc], 2168 &error_abort); 2169 } 2170 2171 ret = bdrv_create_file(filename, opts, &local_err); 2172 if (ret < 0) { 2173 error_propagate(errp, local_err); 2174 return ret; 2175 } 2176 2177 blk = blk_new_open(filename, NULL, NULL, 2178 BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err); 2179 if (blk == NULL) { 2180 error_propagate(errp, local_err); 2181 return -EIO; 2182 } 2183 2184 blk_set_allow_write_beyond_eof(blk, true); 2185 2186 /* Write the header */ 2187 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); 2188 header = g_malloc0(cluster_size); 2189 *header = (QCowHeader) { 2190 .magic = cpu_to_be32(QCOW_MAGIC), 2191 .version = cpu_to_be32(version), 2192 .cluster_bits = cpu_to_be32(cluster_bits), 2193 .size = cpu_to_be64(0), 2194 .l1_table_offset = cpu_to_be64(0), 2195 .l1_size = cpu_to_be32(0), 2196 .refcount_table_offset = cpu_to_be64(cluster_size), 2197 .refcount_table_clusters = cpu_to_be32(1), 2198 .refcount_order = cpu_to_be32(refcount_order), 2199 .header_length = cpu_to_be32(sizeof(*header)), 2200 }; 2201 2202 if (flags & BLOCK_FLAG_ENCRYPT) { 2203 header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES); 2204 } else { 2205 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 2206 } 2207 2208 if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) { 2209 header->compatible_features |= 2210 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); 2211 } 2212 2213 ret = blk_pwrite(blk, 0, header, cluster_size, 0); 2214 g_free(header); 2215 if (ret < 0) { 2216 error_setg_errno(errp, -ret, "Could not write qcow2 header"); 2217 goto out; 2218 } 2219 2220 /* Write a refcount table with one refcount block */ 2221 refcount_table = g_malloc0(2 * cluster_size); 2222 refcount_table[0] = cpu_to_be64(2 * cluster_size); 2223 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0); 2224 g_free(refcount_table); 2225 2226 if (ret < 0) { 2227 error_setg_errno(errp, -ret, "Could not write refcount table"); 2228 goto out; 2229 } 2230 2231 blk_unref(blk); 2232 blk = NULL; 2233 2234 /* 2235 * And now open the image and make it consistent first (i.e. increase the 2236 * refcount of the cluster that is occupied by the header and the refcount 2237 * table) 2238 */ 2239 options = qdict_new(); 2240 qdict_put(options, "driver", qstring_from_str("qcow2")); 2241 blk = blk_new_open(filename, NULL, options, 2242 BDRV_O_RDWR | BDRV_O_NO_FLUSH, &local_err); 2243 if (blk == NULL) { 2244 error_propagate(errp, local_err); 2245 ret = -EIO; 2246 goto out; 2247 } 2248 2249 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size); 2250 if (ret < 0) { 2251 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " 2252 "header and refcount table"); 2253 goto out; 2254 2255 } else if (ret != 0) { 2256 error_report("Huh, first cluster in empty image is already in use?"); 2257 abort(); 2258 } 2259 2260 /* Create a full header (including things like feature table) */ 2261 ret = qcow2_update_header(blk_bs(blk)); 2262 if (ret < 0) { 2263 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 2264 goto out; 2265 } 2266 2267 /* Okay, now that we have a valid image, let's give it the right size */ 2268 ret = blk_truncate(blk, total_size); 2269 if (ret < 0) { 2270 error_setg_errno(errp, -ret, "Could not resize image"); 2271 goto out; 2272 } 2273 2274 /* Want a backing file? There you go.*/ 2275 if (backing_file) { 2276 ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format); 2277 if (ret < 0) { 2278 error_setg_errno(errp, -ret, "Could not assign backing file '%s' " 2279 "with format '%s'", backing_file, backing_format); 2280 goto out; 2281 } 2282 } 2283 2284 /* And if we're supposed to preallocate metadata, do that now */ 2285 if (prealloc != PREALLOC_MODE_OFF) { 2286 BDRVQcow2State *s = blk_bs(blk)->opaque; 2287 qemu_co_mutex_lock(&s->lock); 2288 ret = preallocate(blk_bs(blk)); 2289 qemu_co_mutex_unlock(&s->lock); 2290 if (ret < 0) { 2291 error_setg_errno(errp, -ret, "Could not preallocate metadata"); 2292 goto out; 2293 } 2294 } 2295 2296 blk_unref(blk); 2297 blk = NULL; 2298 2299 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */ 2300 options = qdict_new(); 2301 qdict_put(options, "driver", qstring_from_str("qcow2")); 2302 blk = blk_new_open(filename, NULL, options, 2303 BDRV_O_RDWR | BDRV_O_NO_BACKING, &local_err); 2304 if (blk == NULL) { 2305 error_propagate(errp, local_err); 2306 ret = -EIO; 2307 goto out; 2308 } 2309 2310 ret = 0; 2311 out: 2312 if (blk) { 2313 blk_unref(blk); 2314 } 2315 return ret; 2316 } 2317 2318 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp) 2319 { 2320 char *backing_file = NULL; 2321 char *backing_fmt = NULL; 2322 char *buf = NULL; 2323 uint64_t size = 0; 2324 int flags = 0; 2325 size_t cluster_size = DEFAULT_CLUSTER_SIZE; 2326 PreallocMode prealloc; 2327 int version = 3; 2328 uint64_t refcount_bits = 16; 2329 int refcount_order; 2330 Error *local_err = NULL; 2331 int ret; 2332 2333 /* Read out options */ 2334 size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2335 BDRV_SECTOR_SIZE); 2336 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 2337 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT); 2338 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) { 2339 flags |= BLOCK_FLAG_ENCRYPT; 2340 } 2341 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 2342 DEFAULT_CLUSTER_SIZE); 2343 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2344 prealloc = qapi_enum_parse(PreallocMode_lookup, buf, 2345 PREALLOC_MODE__MAX, PREALLOC_MODE_OFF, 2346 &local_err); 2347 if (local_err) { 2348 error_propagate(errp, local_err); 2349 ret = -EINVAL; 2350 goto finish; 2351 } 2352 g_free(buf); 2353 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); 2354 if (!buf) { 2355 /* keep the default */ 2356 } else if (!strcmp(buf, "0.10")) { 2357 version = 2; 2358 } else if (!strcmp(buf, "1.1")) { 2359 version = 3; 2360 } else { 2361 error_setg(errp, "Invalid compatibility level: '%s'", buf); 2362 ret = -EINVAL; 2363 goto finish; 2364 } 2365 2366 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) { 2367 flags |= BLOCK_FLAG_LAZY_REFCOUNTS; 2368 } 2369 2370 if (backing_file && prealloc != PREALLOC_MODE_OFF) { 2371 error_setg(errp, "Backing file and preallocation cannot be used at " 2372 "the same time"); 2373 ret = -EINVAL; 2374 goto finish; 2375 } 2376 2377 if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) { 2378 error_setg(errp, "Lazy refcounts only supported with compatibility " 2379 "level 1.1 and above (use compat=1.1 or greater)"); 2380 ret = -EINVAL; 2381 goto finish; 2382 } 2383 2384 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 2385 refcount_bits); 2386 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) { 2387 error_setg(errp, "Refcount width must be a power of two and may not " 2388 "exceed 64 bits"); 2389 ret = -EINVAL; 2390 goto finish; 2391 } 2392 2393 if (version < 3 && refcount_bits != 16) { 2394 error_setg(errp, "Different refcount widths than 16 bits require " 2395 "compatibility level 1.1 or above (use compat=1.1 or " 2396 "greater)"); 2397 ret = -EINVAL; 2398 goto finish; 2399 } 2400 2401 refcount_order = ctz32(refcount_bits); 2402 2403 ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags, 2404 cluster_size, prealloc, opts, version, refcount_order, 2405 &local_err); 2406 if (local_err) { 2407 error_propagate(errp, local_err); 2408 } 2409 2410 finish: 2411 g_free(backing_file); 2412 g_free(backing_fmt); 2413 g_free(buf); 2414 return ret; 2415 } 2416 2417 2418 static bool is_zero_sectors(BlockDriverState *bs, int64_t start, 2419 uint32_t count) 2420 { 2421 int nr; 2422 BlockDriverState *file; 2423 int64_t res; 2424 2425 if (!count) { 2426 return true; 2427 } 2428 res = bdrv_get_block_status_above(bs, NULL, start, count, 2429 &nr, &file); 2430 return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == count; 2431 } 2432 2433 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs, 2434 int64_t offset, int count, BdrvRequestFlags flags) 2435 { 2436 int ret; 2437 BDRVQcow2State *s = bs->opaque; 2438 2439 uint32_t head = offset % s->cluster_size; 2440 uint32_t tail = (offset + count) % s->cluster_size; 2441 2442 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, count); 2443 2444 if (head || tail) { 2445 int64_t cl_start = (offset - head) >> BDRV_SECTOR_BITS; 2446 uint64_t off; 2447 unsigned int nr; 2448 2449 assert(head + count <= s->cluster_size); 2450 2451 /* check whether remainder of cluster already reads as zero */ 2452 if (!(is_zero_sectors(bs, cl_start, 2453 DIV_ROUND_UP(head, BDRV_SECTOR_SIZE)) && 2454 is_zero_sectors(bs, (offset + count) >> BDRV_SECTOR_BITS, 2455 DIV_ROUND_UP(-tail & (s->cluster_size - 1), 2456 BDRV_SECTOR_SIZE)))) { 2457 return -ENOTSUP; 2458 } 2459 2460 qemu_co_mutex_lock(&s->lock); 2461 /* We can have new write after previous check */ 2462 offset = cl_start << BDRV_SECTOR_BITS; 2463 count = s->cluster_size; 2464 nr = s->cluster_size; 2465 ret = qcow2_get_cluster_offset(bs, offset, &nr, &off); 2466 if (ret != QCOW2_CLUSTER_UNALLOCATED && ret != QCOW2_CLUSTER_ZERO) { 2467 qemu_co_mutex_unlock(&s->lock); 2468 return -ENOTSUP; 2469 } 2470 } else { 2471 qemu_co_mutex_lock(&s->lock); 2472 } 2473 2474 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, count); 2475 2476 /* Whatever is left can use real zero clusters */ 2477 ret = qcow2_zero_clusters(bs, offset, count >> BDRV_SECTOR_BITS); 2478 qemu_co_mutex_unlock(&s->lock); 2479 2480 return ret; 2481 } 2482 2483 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs, 2484 int64_t sector_num, int nb_sectors) 2485 { 2486 int ret; 2487 BDRVQcow2State *s = bs->opaque; 2488 2489 qemu_co_mutex_lock(&s->lock); 2490 ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS, 2491 nb_sectors, QCOW2_DISCARD_REQUEST, false); 2492 qemu_co_mutex_unlock(&s->lock); 2493 return ret; 2494 } 2495 2496 static int qcow2_truncate(BlockDriverState *bs, int64_t offset) 2497 { 2498 BDRVQcow2State *s = bs->opaque; 2499 int64_t new_l1_size; 2500 int ret; 2501 2502 if (offset & 511) { 2503 error_report("The new size must be a multiple of 512"); 2504 return -EINVAL; 2505 } 2506 2507 /* cannot proceed if image has snapshots */ 2508 if (s->nb_snapshots) { 2509 error_report("Can't resize an image which has snapshots"); 2510 return -ENOTSUP; 2511 } 2512 2513 /* shrinking is currently not supported */ 2514 if (offset < bs->total_sectors * 512) { 2515 error_report("qcow2 doesn't support shrinking images yet"); 2516 return -ENOTSUP; 2517 } 2518 2519 new_l1_size = size_to_l1(s, offset); 2520 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 2521 if (ret < 0) { 2522 return ret; 2523 } 2524 2525 /* write updated header.size */ 2526 offset = cpu_to_be64(offset); 2527 ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, size), 2528 &offset, sizeof(uint64_t)); 2529 if (ret < 0) { 2530 return ret; 2531 } 2532 2533 s->l1_vm_state_index = new_l1_size; 2534 return 0; 2535 } 2536 2537 /* XXX: put compressed sectors first, then all the cluster aligned 2538 tables to avoid losing bytes in alignment */ 2539 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num, 2540 const uint8_t *buf, int nb_sectors) 2541 { 2542 BDRVQcow2State *s = bs->opaque; 2543 z_stream strm; 2544 int ret, out_len; 2545 uint8_t *out_buf; 2546 uint64_t cluster_offset; 2547 2548 if (nb_sectors == 0) { 2549 /* align end of file to a sector boundary to ease reading with 2550 sector based I/Os */ 2551 cluster_offset = bdrv_getlength(bs->file->bs); 2552 return bdrv_truncate(bs->file->bs, cluster_offset); 2553 } 2554 2555 if (nb_sectors != s->cluster_sectors) { 2556 ret = -EINVAL; 2557 2558 /* Zero-pad last write if image size is not cluster aligned */ 2559 if (sector_num + nb_sectors == bs->total_sectors && 2560 nb_sectors < s->cluster_sectors) { 2561 uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size); 2562 memset(pad_buf, 0, s->cluster_size); 2563 memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE); 2564 ret = qcow2_write_compressed(bs, sector_num, 2565 pad_buf, s->cluster_sectors); 2566 qemu_vfree(pad_buf); 2567 } 2568 return ret; 2569 } 2570 2571 out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128); 2572 2573 /* best compression, small window, no zlib header */ 2574 memset(&strm, 0, sizeof(strm)); 2575 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, 2576 Z_DEFLATED, -12, 2577 9, Z_DEFAULT_STRATEGY); 2578 if (ret != 0) { 2579 ret = -EINVAL; 2580 goto fail; 2581 } 2582 2583 strm.avail_in = s->cluster_size; 2584 strm.next_in = (uint8_t *)buf; 2585 strm.avail_out = s->cluster_size; 2586 strm.next_out = out_buf; 2587 2588 ret = deflate(&strm, Z_FINISH); 2589 if (ret != Z_STREAM_END && ret != Z_OK) { 2590 deflateEnd(&strm); 2591 ret = -EINVAL; 2592 goto fail; 2593 } 2594 out_len = strm.next_out - out_buf; 2595 2596 deflateEnd(&strm); 2597 2598 if (ret != Z_STREAM_END || out_len >= s->cluster_size) { 2599 /* could not compress: write normal cluster */ 2600 ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors); 2601 if (ret < 0) { 2602 goto fail; 2603 } 2604 } else { 2605 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs, 2606 sector_num << 9, out_len); 2607 if (!cluster_offset) { 2608 ret = -EIO; 2609 goto fail; 2610 } 2611 cluster_offset &= s->cluster_offset_mask; 2612 2613 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len); 2614 if (ret < 0) { 2615 goto fail; 2616 } 2617 2618 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); 2619 ret = bdrv_pwrite(bs->file->bs, cluster_offset, out_buf, out_len); 2620 if (ret < 0) { 2621 goto fail; 2622 } 2623 } 2624 2625 ret = 0; 2626 fail: 2627 g_free(out_buf); 2628 return ret; 2629 } 2630 2631 static int make_completely_empty(BlockDriverState *bs) 2632 { 2633 BDRVQcow2State *s = bs->opaque; 2634 int ret, l1_clusters; 2635 int64_t offset; 2636 uint64_t *new_reftable = NULL; 2637 uint64_t rt_entry, l1_size2; 2638 struct { 2639 uint64_t l1_offset; 2640 uint64_t reftable_offset; 2641 uint32_t reftable_clusters; 2642 } QEMU_PACKED l1_ofs_rt_ofs_cls; 2643 2644 ret = qcow2_cache_empty(bs, s->l2_table_cache); 2645 if (ret < 0) { 2646 goto fail; 2647 } 2648 2649 ret = qcow2_cache_empty(bs, s->refcount_block_cache); 2650 if (ret < 0) { 2651 goto fail; 2652 } 2653 2654 /* Refcounts will be broken utterly */ 2655 ret = qcow2_mark_dirty(bs); 2656 if (ret < 0) { 2657 goto fail; 2658 } 2659 2660 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 2661 2662 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 2663 l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t); 2664 2665 /* After this call, neither the in-memory nor the on-disk refcount 2666 * information accurately describe the actual references */ 2667 2668 ret = bdrv_pwrite_zeroes(bs->file->bs, s->l1_table_offset, 2669 l1_clusters * s->cluster_size, 0); 2670 if (ret < 0) { 2671 goto fail_broken_refcounts; 2672 } 2673 memset(s->l1_table, 0, l1_size2); 2674 2675 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); 2676 2677 /* Overwrite enough clusters at the beginning of the sectors to place 2678 * the refcount table, a refcount block and the L1 table in; this may 2679 * overwrite parts of the existing refcount and L1 table, which is not 2680 * an issue because the dirty flag is set, complete data loss is in fact 2681 * desired and partial data loss is consequently fine as well */ 2682 ret = bdrv_pwrite_zeroes(bs->file->bs, s->cluster_size, 2683 (2 + l1_clusters) * s->cluster_size, 0); 2684 /* This call (even if it failed overall) may have overwritten on-disk 2685 * refcount structures; in that case, the in-memory refcount information 2686 * will probably differ from the on-disk information which makes the BDS 2687 * unusable */ 2688 if (ret < 0) { 2689 goto fail_broken_refcounts; 2690 } 2691 2692 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 2693 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); 2694 2695 /* "Create" an empty reftable (one cluster) directly after the image 2696 * header and an empty L1 table three clusters after the image header; 2697 * the cluster between those two will be used as the first refblock */ 2698 cpu_to_be64w(&l1_ofs_rt_ofs_cls.l1_offset, 3 * s->cluster_size); 2699 cpu_to_be64w(&l1_ofs_rt_ofs_cls.reftable_offset, s->cluster_size); 2700 cpu_to_be32w(&l1_ofs_rt_ofs_cls.reftable_clusters, 1); 2701 ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, l1_table_offset), 2702 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); 2703 if (ret < 0) { 2704 goto fail_broken_refcounts; 2705 } 2706 2707 s->l1_table_offset = 3 * s->cluster_size; 2708 2709 new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t)); 2710 if (!new_reftable) { 2711 ret = -ENOMEM; 2712 goto fail_broken_refcounts; 2713 } 2714 2715 s->refcount_table_offset = s->cluster_size; 2716 s->refcount_table_size = s->cluster_size / sizeof(uint64_t); 2717 2718 g_free(s->refcount_table); 2719 s->refcount_table = new_reftable; 2720 new_reftable = NULL; 2721 2722 /* Now the in-memory refcount information again corresponds to the on-disk 2723 * information (reftable is empty and no refblocks (the refblock cache is 2724 * empty)); however, this means some clusters (e.g. the image header) are 2725 * referenced, but not refcounted, but the normal qcow2 code assumes that 2726 * the in-memory information is always correct */ 2727 2728 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 2729 2730 /* Enter the first refblock into the reftable */ 2731 rt_entry = cpu_to_be64(2 * s->cluster_size); 2732 ret = bdrv_pwrite_sync(bs->file->bs, s->cluster_size, 2733 &rt_entry, sizeof(rt_entry)); 2734 if (ret < 0) { 2735 goto fail_broken_refcounts; 2736 } 2737 s->refcount_table[0] = 2 * s->cluster_size; 2738 2739 s->free_cluster_index = 0; 2740 assert(3 + l1_clusters <= s->refcount_block_size); 2741 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); 2742 if (offset < 0) { 2743 ret = offset; 2744 goto fail_broken_refcounts; 2745 } else if (offset > 0) { 2746 error_report("First cluster in emptied image is in use"); 2747 abort(); 2748 } 2749 2750 /* Now finally the in-memory information corresponds to the on-disk 2751 * structures and is correct */ 2752 ret = qcow2_mark_clean(bs); 2753 if (ret < 0) { 2754 goto fail; 2755 } 2756 2757 ret = bdrv_truncate(bs->file->bs, (3 + l1_clusters) * s->cluster_size); 2758 if (ret < 0) { 2759 goto fail; 2760 } 2761 2762 return 0; 2763 2764 fail_broken_refcounts: 2765 /* The BDS is unusable at this point. If we wanted to make it usable, we 2766 * would have to call qcow2_refcount_close(), qcow2_refcount_init(), 2767 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() 2768 * again. However, because the functions which could have caused this error 2769 * path to be taken are used by those functions as well, it's very likely 2770 * that that sequence will fail as well. Therefore, just eject the BDS. */ 2771 bs->drv = NULL; 2772 2773 fail: 2774 g_free(new_reftable); 2775 return ret; 2776 } 2777 2778 static int qcow2_make_empty(BlockDriverState *bs) 2779 { 2780 BDRVQcow2State *s = bs->opaque; 2781 uint64_t start_sector; 2782 int sector_step = INT_MAX / BDRV_SECTOR_SIZE; 2783 int l1_clusters, ret = 0; 2784 2785 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 2786 2787 if (s->qcow_version >= 3 && !s->snapshots && 2788 3 + l1_clusters <= s->refcount_block_size) { 2789 /* The following function only works for qcow2 v3 images (it requires 2790 * the dirty flag) and only as long as there are no snapshots (because 2791 * it completely empties the image). Furthermore, the L1 table and three 2792 * additional clusters (image header, refcount table, one refcount 2793 * block) have to fit inside one refcount block. */ 2794 return make_completely_empty(bs); 2795 } 2796 2797 /* This fallback code simply discards every active cluster; this is slow, 2798 * but works in all cases */ 2799 for (start_sector = 0; start_sector < bs->total_sectors; 2800 start_sector += sector_step) 2801 { 2802 /* As this function is generally used after committing an external 2803 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the 2804 * default action for this kind of discard is to pass the discard, 2805 * which will ideally result in an actually smaller image file, as 2806 * is probably desired. */ 2807 ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE, 2808 MIN(sector_step, 2809 bs->total_sectors - start_sector), 2810 QCOW2_DISCARD_SNAPSHOT, true); 2811 if (ret < 0) { 2812 break; 2813 } 2814 } 2815 2816 return ret; 2817 } 2818 2819 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) 2820 { 2821 BDRVQcow2State *s = bs->opaque; 2822 int ret; 2823 2824 qemu_co_mutex_lock(&s->lock); 2825 ret = qcow2_cache_write(bs, s->l2_table_cache); 2826 if (ret < 0) { 2827 qemu_co_mutex_unlock(&s->lock); 2828 return ret; 2829 } 2830 2831 if (qcow2_need_accurate_refcounts(s)) { 2832 ret = qcow2_cache_write(bs, s->refcount_block_cache); 2833 if (ret < 0) { 2834 qemu_co_mutex_unlock(&s->lock); 2835 return ret; 2836 } 2837 } 2838 qemu_co_mutex_unlock(&s->lock); 2839 2840 return 0; 2841 } 2842 2843 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2844 { 2845 BDRVQcow2State *s = bs->opaque; 2846 bdi->unallocated_blocks_are_zero = true; 2847 bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3); 2848 bdi->cluster_size = s->cluster_size; 2849 bdi->vm_state_offset = qcow2_vm_state_offset(s); 2850 return 0; 2851 } 2852 2853 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs) 2854 { 2855 BDRVQcow2State *s = bs->opaque; 2856 ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1); 2857 2858 *spec_info = (ImageInfoSpecific){ 2859 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2, 2860 .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1), 2861 }; 2862 if (s->qcow_version == 2) { 2863 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 2864 .compat = g_strdup("0.10"), 2865 .refcount_bits = s->refcount_bits, 2866 }; 2867 } else if (s->qcow_version == 3) { 2868 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 2869 .compat = g_strdup("1.1"), 2870 .lazy_refcounts = s->compatible_features & 2871 QCOW2_COMPAT_LAZY_REFCOUNTS, 2872 .has_lazy_refcounts = true, 2873 .corrupt = s->incompatible_features & 2874 QCOW2_INCOMPAT_CORRUPT, 2875 .has_corrupt = true, 2876 .refcount_bits = s->refcount_bits, 2877 }; 2878 } else { 2879 /* if this assertion fails, this probably means a new version was 2880 * added without having it covered here */ 2881 assert(false); 2882 } 2883 2884 return spec_info; 2885 } 2886 2887 #if 0 2888 static void dump_refcounts(BlockDriverState *bs) 2889 { 2890 BDRVQcow2State *s = bs->opaque; 2891 int64_t nb_clusters, k, k1, size; 2892 int refcount; 2893 2894 size = bdrv_getlength(bs->file->bs); 2895 nb_clusters = size_to_clusters(s, size); 2896 for(k = 0; k < nb_clusters;) { 2897 k1 = k; 2898 refcount = get_refcount(bs, k); 2899 k++; 2900 while (k < nb_clusters && get_refcount(bs, k) == refcount) 2901 k++; 2902 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount, 2903 k - k1); 2904 } 2905 } 2906 #endif 2907 2908 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 2909 int64_t pos) 2910 { 2911 BDRVQcow2State *s = bs->opaque; 2912 2913 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 2914 return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos, 2915 qiov->size, qiov, 0); 2916 } 2917 2918 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 2919 int64_t pos) 2920 { 2921 BDRVQcow2State *s = bs->opaque; 2922 2923 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 2924 return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos, 2925 qiov->size, qiov, 0); 2926 } 2927 2928 /* 2929 * Downgrades an image's version. To achieve this, any incompatible features 2930 * have to be removed. 2931 */ 2932 static int qcow2_downgrade(BlockDriverState *bs, int target_version, 2933 BlockDriverAmendStatusCB *status_cb, void *cb_opaque) 2934 { 2935 BDRVQcow2State *s = bs->opaque; 2936 int current_version = s->qcow_version; 2937 int ret; 2938 2939 if (target_version == current_version) { 2940 return 0; 2941 } else if (target_version > current_version) { 2942 return -EINVAL; 2943 } else if (target_version != 2) { 2944 return -EINVAL; 2945 } 2946 2947 if (s->refcount_order != 4) { 2948 error_report("compat=0.10 requires refcount_bits=16"); 2949 return -ENOTSUP; 2950 } 2951 2952 /* clear incompatible features */ 2953 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 2954 ret = qcow2_mark_clean(bs); 2955 if (ret < 0) { 2956 return ret; 2957 } 2958 } 2959 2960 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in 2961 * the first place; if that happens nonetheless, returning -ENOTSUP is the 2962 * best thing to do anyway */ 2963 2964 if (s->incompatible_features) { 2965 return -ENOTSUP; 2966 } 2967 2968 /* since we can ignore compatible features, we can set them to 0 as well */ 2969 s->compatible_features = 0; 2970 /* if lazy refcounts have been used, they have already been fixed through 2971 * clearing the dirty flag */ 2972 2973 /* clearing autoclear features is trivial */ 2974 s->autoclear_features = 0; 2975 2976 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque); 2977 if (ret < 0) { 2978 return ret; 2979 } 2980 2981 s->qcow_version = target_version; 2982 ret = qcow2_update_header(bs); 2983 if (ret < 0) { 2984 s->qcow_version = current_version; 2985 return ret; 2986 } 2987 return 0; 2988 } 2989 2990 typedef enum Qcow2AmendOperation { 2991 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be 2992 * statically initialized to so that the helper CB can discern the first 2993 * invocation from an operation change */ 2994 QCOW2_NO_OPERATION = 0, 2995 2996 QCOW2_CHANGING_REFCOUNT_ORDER, 2997 QCOW2_DOWNGRADING, 2998 } Qcow2AmendOperation; 2999 3000 typedef struct Qcow2AmendHelperCBInfo { 3001 /* The code coordinating the amend operations should only modify 3002 * these four fields; the rest will be managed by the CB */ 3003 BlockDriverAmendStatusCB *original_status_cb; 3004 void *original_cb_opaque; 3005 3006 Qcow2AmendOperation current_operation; 3007 3008 /* Total number of operations to perform (only set once) */ 3009 int total_operations; 3010 3011 /* The following fields are managed by the CB */ 3012 3013 /* Number of operations completed */ 3014 int operations_completed; 3015 3016 /* Cumulative offset of all completed operations */ 3017 int64_t offset_completed; 3018 3019 Qcow2AmendOperation last_operation; 3020 int64_t last_work_size; 3021 } Qcow2AmendHelperCBInfo; 3022 3023 static void qcow2_amend_helper_cb(BlockDriverState *bs, 3024 int64_t operation_offset, 3025 int64_t operation_work_size, void *opaque) 3026 { 3027 Qcow2AmendHelperCBInfo *info = opaque; 3028 int64_t current_work_size; 3029 int64_t projected_work_size; 3030 3031 if (info->current_operation != info->last_operation) { 3032 if (info->last_operation != QCOW2_NO_OPERATION) { 3033 info->offset_completed += info->last_work_size; 3034 info->operations_completed++; 3035 } 3036 3037 info->last_operation = info->current_operation; 3038 } 3039 3040 assert(info->total_operations > 0); 3041 assert(info->operations_completed < info->total_operations); 3042 3043 info->last_work_size = operation_work_size; 3044 3045 current_work_size = info->offset_completed + operation_work_size; 3046 3047 /* current_work_size is the total work size for (operations_completed + 1) 3048 * operations (which includes this one), so multiply it by the number of 3049 * operations not covered and divide it by the number of operations 3050 * covered to get a projection for the operations not covered */ 3051 projected_work_size = current_work_size * (info->total_operations - 3052 info->operations_completed - 1) 3053 / (info->operations_completed + 1); 3054 3055 info->original_status_cb(bs, info->offset_completed + operation_offset, 3056 current_work_size + projected_work_size, 3057 info->original_cb_opaque); 3058 } 3059 3060 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, 3061 BlockDriverAmendStatusCB *status_cb, 3062 void *cb_opaque) 3063 { 3064 BDRVQcow2State *s = bs->opaque; 3065 int old_version = s->qcow_version, new_version = old_version; 3066 uint64_t new_size = 0; 3067 const char *backing_file = NULL, *backing_format = NULL; 3068 bool lazy_refcounts = s->use_lazy_refcounts; 3069 const char *compat = NULL; 3070 uint64_t cluster_size = s->cluster_size; 3071 bool encrypt; 3072 int refcount_bits = s->refcount_bits; 3073 int ret; 3074 QemuOptDesc *desc = opts->list->desc; 3075 Qcow2AmendHelperCBInfo helper_cb_info; 3076 3077 while (desc && desc->name) { 3078 if (!qemu_opt_find(opts, desc->name)) { 3079 /* only change explicitly defined options */ 3080 desc++; 3081 continue; 3082 } 3083 3084 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) { 3085 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL); 3086 if (!compat) { 3087 /* preserve default */ 3088 } else if (!strcmp(compat, "0.10")) { 3089 new_version = 2; 3090 } else if (!strcmp(compat, "1.1")) { 3091 new_version = 3; 3092 } else { 3093 error_report("Unknown compatibility level %s", compat); 3094 return -EINVAL; 3095 } 3096 } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) { 3097 error_report("Cannot change preallocation mode"); 3098 return -ENOTSUP; 3099 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) { 3100 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); 3101 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) { 3102 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); 3103 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) { 3104 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); 3105 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) { 3106 encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, 3107 !!s->cipher); 3108 3109 if (encrypt != !!s->cipher) { 3110 error_report("Changing the encryption flag is not supported"); 3111 return -ENOTSUP; 3112 } 3113 } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) { 3114 cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 3115 cluster_size); 3116 if (cluster_size != s->cluster_size) { 3117 error_report("Changing the cluster size is not supported"); 3118 return -ENOTSUP; 3119 } 3120 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) { 3121 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS, 3122 lazy_refcounts); 3123 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) { 3124 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS, 3125 refcount_bits); 3126 3127 if (refcount_bits <= 0 || refcount_bits > 64 || 3128 !is_power_of_2(refcount_bits)) 3129 { 3130 error_report("Refcount width must be a power of two and may " 3131 "not exceed 64 bits"); 3132 return -EINVAL; 3133 } 3134 } else { 3135 /* if this point is reached, this probably means a new option was 3136 * added without having it covered here */ 3137 abort(); 3138 } 3139 3140 desc++; 3141 } 3142 3143 helper_cb_info = (Qcow2AmendHelperCBInfo){ 3144 .original_status_cb = status_cb, 3145 .original_cb_opaque = cb_opaque, 3146 .total_operations = (new_version < old_version) 3147 + (s->refcount_bits != refcount_bits) 3148 }; 3149 3150 /* Upgrade first (some features may require compat=1.1) */ 3151 if (new_version > old_version) { 3152 s->qcow_version = new_version; 3153 ret = qcow2_update_header(bs); 3154 if (ret < 0) { 3155 s->qcow_version = old_version; 3156 return ret; 3157 } 3158 } 3159 3160 if (s->refcount_bits != refcount_bits) { 3161 int refcount_order = ctz32(refcount_bits); 3162 Error *local_error = NULL; 3163 3164 if (new_version < 3 && refcount_bits != 16) { 3165 error_report("Different refcount widths than 16 bits require " 3166 "compatibility level 1.1 or above (use compat=1.1 or " 3167 "greater)"); 3168 return -EINVAL; 3169 } 3170 3171 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER; 3172 ret = qcow2_change_refcount_order(bs, refcount_order, 3173 &qcow2_amend_helper_cb, 3174 &helper_cb_info, &local_error); 3175 if (ret < 0) { 3176 error_report_err(local_error); 3177 return ret; 3178 } 3179 } 3180 3181 if (backing_file || backing_format) { 3182 ret = qcow2_change_backing_file(bs, 3183 backing_file ?: s->image_backing_file, 3184 backing_format ?: s->image_backing_format); 3185 if (ret < 0) { 3186 return ret; 3187 } 3188 } 3189 3190 if (s->use_lazy_refcounts != lazy_refcounts) { 3191 if (lazy_refcounts) { 3192 if (new_version < 3) { 3193 error_report("Lazy refcounts only supported with compatibility " 3194 "level 1.1 and above (use compat=1.1 or greater)"); 3195 return -EINVAL; 3196 } 3197 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 3198 ret = qcow2_update_header(bs); 3199 if (ret < 0) { 3200 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 3201 return ret; 3202 } 3203 s->use_lazy_refcounts = true; 3204 } else { 3205 /* make image clean first */ 3206 ret = qcow2_mark_clean(bs); 3207 if (ret < 0) { 3208 return ret; 3209 } 3210 /* now disallow lazy refcounts */ 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 = false; 3218 } 3219 } 3220 3221 if (new_size) { 3222 ret = bdrv_truncate(bs, new_size); 3223 if (ret < 0) { 3224 return ret; 3225 } 3226 } 3227 3228 /* Downgrade last (so unsupported features can be removed before) */ 3229 if (new_version < old_version) { 3230 helper_cb_info.current_operation = QCOW2_DOWNGRADING; 3231 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb, 3232 &helper_cb_info); 3233 if (ret < 0) { 3234 return ret; 3235 } 3236 } 3237 3238 return 0; 3239 } 3240 3241 /* 3242 * If offset or size are negative, respectively, they will not be included in 3243 * the BLOCK_IMAGE_CORRUPTED event emitted. 3244 * fatal will be ignored for read-only BDS; corruptions found there will always 3245 * be considered non-fatal. 3246 */ 3247 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 3248 int64_t size, const char *message_format, ...) 3249 { 3250 BDRVQcow2State *s = bs->opaque; 3251 const char *node_name; 3252 char *message; 3253 va_list ap; 3254 3255 fatal = fatal && !bs->read_only; 3256 3257 if (s->signaled_corruption && 3258 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) 3259 { 3260 return; 3261 } 3262 3263 va_start(ap, message_format); 3264 message = g_strdup_vprintf(message_format, ap); 3265 va_end(ap); 3266 3267 if (fatal) { 3268 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " 3269 "corruption events will be suppressed\n", message); 3270 } else { 3271 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " 3272 "corruption events will be suppressed\n", message); 3273 } 3274 3275 node_name = bdrv_get_node_name(bs); 3276 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), 3277 *node_name != '\0', node_name, 3278 message, offset >= 0, offset, 3279 size >= 0, size, 3280 fatal, &error_abort); 3281 g_free(message); 3282 3283 if (fatal) { 3284 qcow2_mark_corrupt(bs); 3285 bs->drv = NULL; /* make BDS unusable */ 3286 } 3287 3288 s->signaled_corruption = true; 3289 } 3290 3291 static QemuOptsList qcow2_create_opts = { 3292 .name = "qcow2-create-opts", 3293 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), 3294 .desc = { 3295 { 3296 .name = BLOCK_OPT_SIZE, 3297 .type = QEMU_OPT_SIZE, 3298 .help = "Virtual disk size" 3299 }, 3300 { 3301 .name = BLOCK_OPT_COMPAT_LEVEL, 3302 .type = QEMU_OPT_STRING, 3303 .help = "Compatibility level (0.10 or 1.1)" 3304 }, 3305 { 3306 .name = BLOCK_OPT_BACKING_FILE, 3307 .type = QEMU_OPT_STRING, 3308 .help = "File name of a base image" 3309 }, 3310 { 3311 .name = BLOCK_OPT_BACKING_FMT, 3312 .type = QEMU_OPT_STRING, 3313 .help = "Image format of the base image" 3314 }, 3315 { 3316 .name = BLOCK_OPT_ENCRYPT, 3317 .type = QEMU_OPT_BOOL, 3318 .help = "Encrypt the image", 3319 .def_value_str = "off" 3320 }, 3321 { 3322 .name = BLOCK_OPT_CLUSTER_SIZE, 3323 .type = QEMU_OPT_SIZE, 3324 .help = "qcow2 cluster size", 3325 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) 3326 }, 3327 { 3328 .name = BLOCK_OPT_PREALLOC, 3329 .type = QEMU_OPT_STRING, 3330 .help = "Preallocation mode (allowed values: off, metadata, " 3331 "falloc, full)" 3332 }, 3333 { 3334 .name = BLOCK_OPT_LAZY_REFCOUNTS, 3335 .type = QEMU_OPT_BOOL, 3336 .help = "Postpone refcount updates", 3337 .def_value_str = "off" 3338 }, 3339 { 3340 .name = BLOCK_OPT_REFCOUNT_BITS, 3341 .type = QEMU_OPT_NUMBER, 3342 .help = "Width of a reference count entry in bits", 3343 .def_value_str = "16" 3344 }, 3345 { /* end of list */ } 3346 } 3347 }; 3348 3349 BlockDriver bdrv_qcow2 = { 3350 .format_name = "qcow2", 3351 .instance_size = sizeof(BDRVQcow2State), 3352 .bdrv_probe = qcow2_probe, 3353 .bdrv_open = qcow2_open, 3354 .bdrv_close = qcow2_close, 3355 .bdrv_reopen_prepare = qcow2_reopen_prepare, 3356 .bdrv_reopen_commit = qcow2_reopen_commit, 3357 .bdrv_reopen_abort = qcow2_reopen_abort, 3358 .bdrv_join_options = qcow2_join_options, 3359 .bdrv_create = qcow2_create, 3360 .bdrv_has_zero_init = bdrv_has_zero_init_1, 3361 .bdrv_co_get_block_status = qcow2_co_get_block_status, 3362 .bdrv_set_key = qcow2_set_key, 3363 3364 .bdrv_co_preadv = qcow2_co_preadv, 3365 .bdrv_co_pwritev = qcow2_co_pwritev, 3366 .bdrv_co_flush_to_os = qcow2_co_flush_to_os, 3367 3368 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes, 3369 .bdrv_co_discard = qcow2_co_discard, 3370 .bdrv_truncate = qcow2_truncate, 3371 .bdrv_write_compressed = qcow2_write_compressed, 3372 .bdrv_make_empty = qcow2_make_empty, 3373 3374 .bdrv_snapshot_create = qcow2_snapshot_create, 3375 .bdrv_snapshot_goto = qcow2_snapshot_goto, 3376 .bdrv_snapshot_delete = qcow2_snapshot_delete, 3377 .bdrv_snapshot_list = qcow2_snapshot_list, 3378 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 3379 .bdrv_get_info = qcow2_get_info, 3380 .bdrv_get_specific_info = qcow2_get_specific_info, 3381 3382 .bdrv_save_vmstate = qcow2_save_vmstate, 3383 .bdrv_load_vmstate = qcow2_load_vmstate, 3384 3385 .supports_backing = true, 3386 .bdrv_change_backing_file = qcow2_change_backing_file, 3387 3388 .bdrv_refresh_limits = qcow2_refresh_limits, 3389 .bdrv_invalidate_cache = qcow2_invalidate_cache, 3390 .bdrv_inactivate = qcow2_inactivate, 3391 3392 .create_opts = &qcow2_create_opts, 3393 .bdrv_check = qcow2_check, 3394 .bdrv_amend_options = qcow2_amend_options, 3395 3396 .bdrv_detach_aio_context = qcow2_detach_aio_context, 3397 .bdrv_attach_aio_context = qcow2_attach_aio_context, 3398 }; 3399 3400 static void bdrv_qcow2_init(void) 3401 { 3402 bdrv_register(&bdrv_qcow2); 3403 } 3404 3405 block_init(bdrv_qcow2_init); 3406