1 /* 2 * Block driver for the QCOW version 2 format 3 * 4 * Copyright (c) 2004-2006 Fabrice Bellard 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 #include "qemu/osdep.h" 25 #include "block/block_int.h" 26 #include "sysemu/block-backend.h" 27 #include "qemu/module.h" 28 #include <zlib.h> 29 #include "block/qcow2.h" 30 #include "qemu/error-report.h" 31 #include "qapi/qmp/qerror.h" 32 #include "qapi/qmp/qbool.h" 33 #include "qapi/util.h" 34 #include "qapi/qmp/types.h" 35 #include "qapi-event.h" 36 #include "trace.h" 37 #include "qemu/option_int.h" 38 #include "qemu/cutils.h" 39 #include "qemu/bswap.h" 40 41 /* 42 Differences with QCOW: 43 44 - Support for multiple incremental snapshots. 45 - Memory management by reference counts. 46 - Clusters which have a reference count of one have the bit 47 QCOW_OFLAG_COPIED to optimize write performance. 48 - Size of compressed clusters is stored in sectors to reduce bit usage 49 in the cluster offsets. 50 - Support for storing additional data (such as the VM state) in the 51 snapshots. 52 - If a backing store is used, the cluster size is not constrained 53 (could be backported to QCOW). 54 - L2 tables have always a size of one cluster. 55 */ 56 57 58 typedef struct { 59 uint32_t magic; 60 uint32_t len; 61 } QEMU_PACKED QCowExtension; 62 63 #define QCOW2_EXT_MAGIC_END 0 64 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA 65 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 66 67 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) 68 { 69 const QCowHeader *cow_header = (const void *)buf; 70 71 if (buf_size >= sizeof(QCowHeader) && 72 be32_to_cpu(cow_header->magic) == QCOW_MAGIC && 73 be32_to_cpu(cow_header->version) >= 2) 74 return 100; 75 else 76 return 0; 77 } 78 79 80 /* 81 * read qcow2 extension and fill bs 82 * start reading from start_offset 83 * finish reading upon magic of value 0 or when end_offset reached 84 * unknown magic is skipped (future extension this version knows nothing about) 85 * return 0 upon success, non-0 otherwise 86 */ 87 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, 88 uint64_t end_offset, void **p_feature_table, 89 Error **errp) 90 { 91 BDRVQcow2State *s = bs->opaque; 92 QCowExtension ext; 93 uint64_t offset; 94 int ret; 95 96 #ifdef DEBUG_EXT 97 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); 98 #endif 99 offset = start_offset; 100 while (offset < end_offset) { 101 102 #ifdef DEBUG_EXT 103 /* Sanity check */ 104 if (offset > s->cluster_size) 105 printf("qcow2_read_extension: suspicious offset %lu\n", offset); 106 107 printf("attempting to read extended header in offset %lu\n", offset); 108 #endif 109 110 ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext)); 111 if (ret < 0) { 112 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: " 113 "pread fail from offset %" PRIu64, offset); 114 return 1; 115 } 116 be32_to_cpus(&ext.magic); 117 be32_to_cpus(&ext.len); 118 offset += sizeof(ext); 119 #ifdef DEBUG_EXT 120 printf("ext.magic = 0x%x\n", ext.magic); 121 #endif 122 if (offset > end_offset || ext.len > end_offset - offset) { 123 error_setg(errp, "Header extension too large"); 124 return -EINVAL; 125 } 126 127 switch (ext.magic) { 128 case QCOW2_EXT_MAGIC_END: 129 return 0; 130 131 case QCOW2_EXT_MAGIC_BACKING_FORMAT: 132 if (ext.len >= sizeof(bs->backing_format)) { 133 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32 134 " too large (>=%zu)", ext.len, 135 sizeof(bs->backing_format)); 136 return 2; 137 } 138 ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len); 139 if (ret < 0) { 140 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: " 141 "Could not read format name"); 142 return 3; 143 } 144 bs->backing_format[ext.len] = '\0'; 145 s->image_backing_format = g_strdup(bs->backing_format); 146 #ifdef DEBUG_EXT 147 printf("Qcow2: Got format extension %s\n", bs->backing_format); 148 #endif 149 break; 150 151 case QCOW2_EXT_MAGIC_FEATURE_TABLE: 152 if (p_feature_table != NULL) { 153 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); 154 ret = bdrv_pread(bs->file, offset , feature_table, ext.len); 155 if (ret < 0) { 156 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: " 157 "Could not read table"); 158 return ret; 159 } 160 161 *p_feature_table = feature_table; 162 } 163 break; 164 165 default: 166 /* unknown magic - save it in case we need to rewrite the header */ 167 { 168 Qcow2UnknownHeaderExtension *uext; 169 170 uext = g_malloc0(sizeof(*uext) + ext.len); 171 uext->magic = ext.magic; 172 uext->len = ext.len; 173 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); 174 175 ret = bdrv_pread(bs->file, offset , uext->data, uext->len); 176 if (ret < 0) { 177 error_setg_errno(errp, -ret, "ERROR: unknown extension: " 178 "Could not read data"); 179 return ret; 180 } 181 } 182 break; 183 } 184 185 offset += ((ext.len + 7) & ~7); 186 } 187 188 return 0; 189 } 190 191 static void cleanup_unknown_header_ext(BlockDriverState *bs) 192 { 193 BDRVQcow2State *s = bs->opaque; 194 Qcow2UnknownHeaderExtension *uext, *next; 195 196 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) { 197 QLIST_REMOVE(uext, next); 198 g_free(uext); 199 } 200 } 201 202 static void report_unsupported_feature(Error **errp, Qcow2Feature *table, 203 uint64_t mask) 204 { 205 char *features = g_strdup(""); 206 char *old; 207 208 while (table && table->name[0] != '\0') { 209 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) { 210 if (mask & (1ULL << table->bit)) { 211 old = features; 212 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "", 213 table->name); 214 g_free(old); 215 mask &= ~(1ULL << table->bit); 216 } 217 } 218 table++; 219 } 220 221 if (mask) { 222 old = features; 223 features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64, 224 old, *old ? ", " : "", mask); 225 g_free(old); 226 } 227 228 error_setg(errp, "Unsupported qcow2 feature(s): %s", features); 229 g_free(features); 230 } 231 232 /* 233 * Sets the dirty bit and flushes afterwards if necessary. 234 * 235 * The incompatible_features bit is only set if the image file header was 236 * updated successfully. Therefore it is not required to check the return 237 * value of this function. 238 */ 239 int qcow2_mark_dirty(BlockDriverState *bs) 240 { 241 BDRVQcow2State *s = bs->opaque; 242 uint64_t val; 243 int ret; 244 245 assert(s->qcow_version >= 3); 246 247 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 248 return 0; /* already dirty */ 249 } 250 251 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); 252 ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features), 253 &val, sizeof(val)); 254 if (ret < 0) { 255 return ret; 256 } 257 ret = bdrv_flush(bs->file->bs); 258 if (ret < 0) { 259 return ret; 260 } 261 262 /* Only treat image as dirty if the header was updated successfully */ 263 s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; 264 return 0; 265 } 266 267 /* 268 * Clears the dirty bit and flushes before if necessary. Only call this 269 * function when there are no pending requests, it does not guard against 270 * concurrent requests dirtying the image. 271 */ 272 static int qcow2_mark_clean(BlockDriverState *bs) 273 { 274 BDRVQcow2State *s = bs->opaque; 275 276 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 277 int ret; 278 279 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; 280 281 ret = bdrv_flush(bs); 282 if (ret < 0) { 283 return ret; 284 } 285 286 return qcow2_update_header(bs); 287 } 288 return 0; 289 } 290 291 /* 292 * Marks the image as corrupt. 293 */ 294 int qcow2_mark_corrupt(BlockDriverState *bs) 295 { 296 BDRVQcow2State *s = bs->opaque; 297 298 s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT; 299 return qcow2_update_header(bs); 300 } 301 302 /* 303 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes 304 * before if necessary. 305 */ 306 int qcow2_mark_consistent(BlockDriverState *bs) 307 { 308 BDRVQcow2State *s = bs->opaque; 309 310 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 311 int ret = bdrv_flush(bs); 312 if (ret < 0) { 313 return ret; 314 } 315 316 s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT; 317 return qcow2_update_header(bs); 318 } 319 return 0; 320 } 321 322 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result, 323 BdrvCheckMode fix) 324 { 325 int ret = qcow2_check_refcounts(bs, result, fix); 326 if (ret < 0) { 327 return ret; 328 } 329 330 if (fix && result->check_errors == 0 && result->corruptions == 0) { 331 ret = qcow2_mark_clean(bs); 332 if (ret < 0) { 333 return ret; 334 } 335 return qcow2_mark_consistent(bs); 336 } 337 return ret; 338 } 339 340 static int validate_table_offset(BlockDriverState *bs, uint64_t offset, 341 uint64_t entries, size_t entry_len) 342 { 343 BDRVQcow2State *s = bs->opaque; 344 uint64_t size; 345 346 /* Use signed INT64_MAX as the maximum even for uint64_t header fields, 347 * because values will be passed to qemu functions taking int64_t. */ 348 if (entries > INT64_MAX / entry_len) { 349 return -EINVAL; 350 } 351 352 size = entries * entry_len; 353 354 if (INT64_MAX - size < offset) { 355 return -EINVAL; 356 } 357 358 /* Tables must be cluster aligned */ 359 if (offset & (s->cluster_size - 1)) { 360 return -EINVAL; 361 } 362 363 return 0; 364 } 365 366 static QemuOptsList qcow2_runtime_opts = { 367 .name = "qcow2", 368 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), 369 .desc = { 370 { 371 .name = QCOW2_OPT_LAZY_REFCOUNTS, 372 .type = QEMU_OPT_BOOL, 373 .help = "Postpone refcount updates", 374 }, 375 { 376 .name = QCOW2_OPT_DISCARD_REQUEST, 377 .type = QEMU_OPT_BOOL, 378 .help = "Pass guest discard requests to the layer below", 379 }, 380 { 381 .name = QCOW2_OPT_DISCARD_SNAPSHOT, 382 .type = QEMU_OPT_BOOL, 383 .help = "Generate discard requests when snapshot related space " 384 "is freed", 385 }, 386 { 387 .name = QCOW2_OPT_DISCARD_OTHER, 388 .type = QEMU_OPT_BOOL, 389 .help = "Generate discard requests when other clusters are freed", 390 }, 391 { 392 .name = QCOW2_OPT_OVERLAP, 393 .type = QEMU_OPT_STRING, 394 .help = "Selects which overlap checks to perform from a range of " 395 "templates (none, constant, cached, all)", 396 }, 397 { 398 .name = QCOW2_OPT_OVERLAP_TEMPLATE, 399 .type = QEMU_OPT_STRING, 400 .help = "Selects which overlap checks to perform from a range of " 401 "templates (none, constant, cached, all)", 402 }, 403 { 404 .name = QCOW2_OPT_OVERLAP_MAIN_HEADER, 405 .type = QEMU_OPT_BOOL, 406 .help = "Check for unintended writes into the main qcow2 header", 407 }, 408 { 409 .name = QCOW2_OPT_OVERLAP_ACTIVE_L1, 410 .type = QEMU_OPT_BOOL, 411 .help = "Check for unintended writes into the active L1 table", 412 }, 413 { 414 .name = QCOW2_OPT_OVERLAP_ACTIVE_L2, 415 .type = QEMU_OPT_BOOL, 416 .help = "Check for unintended writes into an active L2 table", 417 }, 418 { 419 .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 420 .type = QEMU_OPT_BOOL, 421 .help = "Check for unintended writes into the refcount table", 422 }, 423 { 424 .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 425 .type = QEMU_OPT_BOOL, 426 .help = "Check for unintended writes into a refcount block", 427 }, 428 { 429 .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 430 .type = QEMU_OPT_BOOL, 431 .help = "Check for unintended writes into the snapshot table", 432 }, 433 { 434 .name = QCOW2_OPT_OVERLAP_INACTIVE_L1, 435 .type = QEMU_OPT_BOOL, 436 .help = "Check for unintended writes into an inactive L1 table", 437 }, 438 { 439 .name = QCOW2_OPT_OVERLAP_INACTIVE_L2, 440 .type = QEMU_OPT_BOOL, 441 .help = "Check for unintended writes into an inactive L2 table", 442 }, 443 { 444 .name = QCOW2_OPT_CACHE_SIZE, 445 .type = QEMU_OPT_SIZE, 446 .help = "Maximum combined metadata (L2 tables and refcount blocks) " 447 "cache size", 448 }, 449 { 450 .name = QCOW2_OPT_L2_CACHE_SIZE, 451 .type = QEMU_OPT_SIZE, 452 .help = "Maximum L2 table cache size", 453 }, 454 { 455 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE, 456 .type = QEMU_OPT_SIZE, 457 .help = "Maximum refcount block cache size", 458 }, 459 { 460 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL, 461 .type = QEMU_OPT_NUMBER, 462 .help = "Clean unused cache entries after this time (in seconds)", 463 }, 464 { /* end of list */ } 465 }, 466 }; 467 468 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { 469 [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, 470 [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, 471 [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, 472 [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, 473 [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, 474 [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, 475 [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, 476 [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, 477 }; 478 479 static void cache_clean_timer_cb(void *opaque) 480 { 481 BlockDriverState *bs = opaque; 482 BDRVQcow2State *s = bs->opaque; 483 qcow2_cache_clean_unused(bs, s->l2_table_cache); 484 qcow2_cache_clean_unused(bs, s->refcount_block_cache); 485 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 486 (int64_t) s->cache_clean_interval * 1000); 487 } 488 489 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context) 490 { 491 BDRVQcow2State *s = bs->opaque; 492 if (s->cache_clean_interval > 0) { 493 s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL, 494 SCALE_MS, cache_clean_timer_cb, 495 bs); 496 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 497 (int64_t) s->cache_clean_interval * 1000); 498 } 499 } 500 501 static void cache_clean_timer_del(BlockDriverState *bs) 502 { 503 BDRVQcow2State *s = bs->opaque; 504 if (s->cache_clean_timer) { 505 timer_del(s->cache_clean_timer); 506 timer_free(s->cache_clean_timer); 507 s->cache_clean_timer = NULL; 508 } 509 } 510 511 static void qcow2_detach_aio_context(BlockDriverState *bs) 512 { 513 cache_clean_timer_del(bs); 514 } 515 516 static void qcow2_attach_aio_context(BlockDriverState *bs, 517 AioContext *new_context) 518 { 519 cache_clean_timer_init(bs, new_context); 520 } 521 522 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts, 523 uint64_t *l2_cache_size, 524 uint64_t *refcount_cache_size, Error **errp) 525 { 526 BDRVQcow2State *s = bs->opaque; 527 uint64_t combined_cache_size; 528 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set; 529 530 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE); 531 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE); 532 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 533 534 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0); 535 *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0); 536 *refcount_cache_size = qemu_opt_get_size(opts, 537 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0); 538 539 if (combined_cache_size_set) { 540 if (l2_cache_size_set && refcount_cache_size_set) { 541 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE 542 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set " 543 "the same time"); 544 return; 545 } else if (*l2_cache_size > combined_cache_size) { 546 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed " 547 QCOW2_OPT_CACHE_SIZE); 548 return; 549 } else if (*refcount_cache_size > combined_cache_size) { 550 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed " 551 QCOW2_OPT_CACHE_SIZE); 552 return; 553 } 554 555 if (l2_cache_size_set) { 556 *refcount_cache_size = combined_cache_size - *l2_cache_size; 557 } else if (refcount_cache_size_set) { 558 *l2_cache_size = combined_cache_size - *refcount_cache_size; 559 } else { 560 *refcount_cache_size = combined_cache_size 561 / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1); 562 *l2_cache_size = combined_cache_size - *refcount_cache_size; 563 } 564 } else { 565 if (!l2_cache_size_set && !refcount_cache_size_set) { 566 *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE, 567 (uint64_t)DEFAULT_L2_CACHE_CLUSTERS 568 * s->cluster_size); 569 *refcount_cache_size = *l2_cache_size 570 / DEFAULT_L2_REFCOUNT_SIZE_RATIO; 571 } else if (!l2_cache_size_set) { 572 *l2_cache_size = *refcount_cache_size 573 * DEFAULT_L2_REFCOUNT_SIZE_RATIO; 574 } else if (!refcount_cache_size_set) { 575 *refcount_cache_size = *l2_cache_size 576 / DEFAULT_L2_REFCOUNT_SIZE_RATIO; 577 } 578 } 579 } 580 581 typedef struct Qcow2ReopenState { 582 Qcow2Cache *l2_table_cache; 583 Qcow2Cache *refcount_block_cache; 584 bool use_lazy_refcounts; 585 int overlap_check; 586 bool discard_passthrough[QCOW2_DISCARD_MAX]; 587 uint64_t cache_clean_interval; 588 } Qcow2ReopenState; 589 590 static int qcow2_update_options_prepare(BlockDriverState *bs, 591 Qcow2ReopenState *r, 592 QDict *options, int flags, 593 Error **errp) 594 { 595 BDRVQcow2State *s = bs->opaque; 596 QemuOpts *opts = NULL; 597 const char *opt_overlap_check, *opt_overlap_check_template; 598 int overlap_check_template = 0; 599 uint64_t l2_cache_size, refcount_cache_size; 600 int i; 601 Error *local_err = NULL; 602 int ret; 603 604 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); 605 qemu_opts_absorb_qdict(opts, options, &local_err); 606 if (local_err) { 607 error_propagate(errp, local_err); 608 ret = -EINVAL; 609 goto fail; 610 } 611 612 /* get L2 table/refcount block cache size from command line options */ 613 read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size, 614 &local_err); 615 if (local_err) { 616 error_propagate(errp, local_err); 617 ret = -EINVAL; 618 goto fail; 619 } 620 621 l2_cache_size /= s->cluster_size; 622 if (l2_cache_size < MIN_L2_CACHE_SIZE) { 623 l2_cache_size = MIN_L2_CACHE_SIZE; 624 } 625 if (l2_cache_size > INT_MAX) { 626 error_setg(errp, "L2 cache size too big"); 627 ret = -EINVAL; 628 goto fail; 629 } 630 631 refcount_cache_size /= s->cluster_size; 632 if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) { 633 refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE; 634 } 635 if (refcount_cache_size > INT_MAX) { 636 error_setg(errp, "Refcount cache size too big"); 637 ret = -EINVAL; 638 goto fail; 639 } 640 641 /* alloc new L2 table/refcount block cache, flush old one */ 642 if (s->l2_table_cache) { 643 ret = qcow2_cache_flush(bs, s->l2_table_cache); 644 if (ret) { 645 error_setg_errno(errp, -ret, "Failed to flush the L2 table cache"); 646 goto fail; 647 } 648 } 649 650 if (s->refcount_block_cache) { 651 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 652 if (ret) { 653 error_setg_errno(errp, -ret, 654 "Failed to flush the refcount block cache"); 655 goto fail; 656 } 657 } 658 659 r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size); 660 r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size); 661 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) { 662 error_setg(errp, "Could not allocate metadata caches"); 663 ret = -ENOMEM; 664 goto fail; 665 } 666 667 /* New interval for cache cleanup timer */ 668 r->cache_clean_interval = 669 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL, 670 s->cache_clean_interval); 671 if (r->cache_clean_interval > UINT_MAX) { 672 error_setg(errp, "Cache clean interval too big"); 673 ret = -EINVAL; 674 goto fail; 675 } 676 677 /* lazy-refcounts; flush if going from enabled to disabled */ 678 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, 679 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); 680 if (r->use_lazy_refcounts && s->qcow_version < 3) { 681 error_setg(errp, "Lazy refcounts require a qcow2 image with at least " 682 "qemu 1.1 compatibility level"); 683 ret = -EINVAL; 684 goto fail; 685 } 686 687 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) { 688 ret = qcow2_mark_clean(bs); 689 if (ret < 0) { 690 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts"); 691 goto fail; 692 } 693 } 694 695 /* Overlap check options */ 696 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP); 697 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE); 698 if (opt_overlap_check_template && opt_overlap_check && 699 strcmp(opt_overlap_check_template, opt_overlap_check)) 700 { 701 error_setg(errp, "Conflicting values for qcow2 options '" 702 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE 703 "' ('%s')", opt_overlap_check, opt_overlap_check_template); 704 ret = -EINVAL; 705 goto fail; 706 } 707 if (!opt_overlap_check) { 708 opt_overlap_check = opt_overlap_check_template ?: "cached"; 709 } 710 711 if (!strcmp(opt_overlap_check, "none")) { 712 overlap_check_template = 0; 713 } else if (!strcmp(opt_overlap_check, "constant")) { 714 overlap_check_template = QCOW2_OL_CONSTANT; 715 } else if (!strcmp(opt_overlap_check, "cached")) { 716 overlap_check_template = QCOW2_OL_CACHED; 717 } else if (!strcmp(opt_overlap_check, "all")) { 718 overlap_check_template = QCOW2_OL_ALL; 719 } else { 720 error_setg(errp, "Unsupported value '%s' for qcow2 option " 721 "'overlap-check'. Allowed are any of the following: " 722 "none, constant, cached, all", opt_overlap_check); 723 ret = -EINVAL; 724 goto fail; 725 } 726 727 r->overlap_check = 0; 728 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { 729 /* overlap-check defines a template bitmask, but every flag may be 730 * overwritten through the associated boolean option */ 731 r->overlap_check |= 732 qemu_opt_get_bool(opts, overlap_bool_option_names[i], 733 overlap_check_template & (1 << i)) << i; 734 } 735 736 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false; 737 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true; 738 r->discard_passthrough[QCOW2_DISCARD_REQUEST] = 739 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, 740 flags & BDRV_O_UNMAP); 741 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = 742 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true); 743 r->discard_passthrough[QCOW2_DISCARD_OTHER] = 744 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false); 745 746 ret = 0; 747 fail: 748 qemu_opts_del(opts); 749 opts = NULL; 750 return ret; 751 } 752 753 static void qcow2_update_options_commit(BlockDriverState *bs, 754 Qcow2ReopenState *r) 755 { 756 BDRVQcow2State *s = bs->opaque; 757 int i; 758 759 if (s->l2_table_cache) { 760 qcow2_cache_destroy(bs, s->l2_table_cache); 761 } 762 if (s->refcount_block_cache) { 763 qcow2_cache_destroy(bs, s->refcount_block_cache); 764 } 765 s->l2_table_cache = r->l2_table_cache; 766 s->refcount_block_cache = r->refcount_block_cache; 767 768 s->overlap_check = r->overlap_check; 769 s->use_lazy_refcounts = r->use_lazy_refcounts; 770 771 for (i = 0; i < QCOW2_DISCARD_MAX; i++) { 772 s->discard_passthrough[i] = r->discard_passthrough[i]; 773 } 774 775 if (s->cache_clean_interval != r->cache_clean_interval) { 776 cache_clean_timer_del(bs); 777 s->cache_clean_interval = r->cache_clean_interval; 778 cache_clean_timer_init(bs, bdrv_get_aio_context(bs)); 779 } 780 } 781 782 static void qcow2_update_options_abort(BlockDriverState *bs, 783 Qcow2ReopenState *r) 784 { 785 if (r->l2_table_cache) { 786 qcow2_cache_destroy(bs, r->l2_table_cache); 787 } 788 if (r->refcount_block_cache) { 789 qcow2_cache_destroy(bs, r->refcount_block_cache); 790 } 791 } 792 793 static int qcow2_update_options(BlockDriverState *bs, QDict *options, 794 int flags, Error **errp) 795 { 796 Qcow2ReopenState r = {}; 797 int ret; 798 799 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp); 800 if (ret >= 0) { 801 qcow2_update_options_commit(bs, &r); 802 } else { 803 qcow2_update_options_abort(bs, &r); 804 } 805 806 return ret; 807 } 808 809 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, 810 Error **errp) 811 { 812 BDRVQcow2State *s = bs->opaque; 813 unsigned int len, i; 814 int ret = 0; 815 QCowHeader header; 816 Error *local_err = NULL; 817 uint64_t ext_end; 818 uint64_t l1_vm_state_index; 819 820 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 821 if (ret < 0) { 822 error_setg_errno(errp, -ret, "Could not read qcow2 header"); 823 goto fail; 824 } 825 be32_to_cpus(&header.magic); 826 be32_to_cpus(&header.version); 827 be64_to_cpus(&header.backing_file_offset); 828 be32_to_cpus(&header.backing_file_size); 829 be64_to_cpus(&header.size); 830 be32_to_cpus(&header.cluster_bits); 831 be32_to_cpus(&header.crypt_method); 832 be64_to_cpus(&header.l1_table_offset); 833 be32_to_cpus(&header.l1_size); 834 be64_to_cpus(&header.refcount_table_offset); 835 be32_to_cpus(&header.refcount_table_clusters); 836 be64_to_cpus(&header.snapshots_offset); 837 be32_to_cpus(&header.nb_snapshots); 838 839 if (header.magic != QCOW_MAGIC) { 840 error_setg(errp, "Image is not in qcow2 format"); 841 ret = -EINVAL; 842 goto fail; 843 } 844 if (header.version < 2 || header.version > 3) { 845 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version); 846 ret = -ENOTSUP; 847 goto fail; 848 } 849 850 s->qcow_version = header.version; 851 852 /* Initialise cluster size */ 853 if (header.cluster_bits < MIN_CLUSTER_BITS || 854 header.cluster_bits > MAX_CLUSTER_BITS) { 855 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, 856 header.cluster_bits); 857 ret = -EINVAL; 858 goto fail; 859 } 860 861 s->cluster_bits = header.cluster_bits; 862 s->cluster_size = 1 << s->cluster_bits; 863 s->cluster_sectors = 1 << (s->cluster_bits - 9); 864 865 /* Initialise version 3 header fields */ 866 if (header.version == 2) { 867 header.incompatible_features = 0; 868 header.compatible_features = 0; 869 header.autoclear_features = 0; 870 header.refcount_order = 4; 871 header.header_length = 72; 872 } else { 873 be64_to_cpus(&header.incompatible_features); 874 be64_to_cpus(&header.compatible_features); 875 be64_to_cpus(&header.autoclear_features); 876 be32_to_cpus(&header.refcount_order); 877 be32_to_cpus(&header.header_length); 878 879 if (header.header_length < 104) { 880 error_setg(errp, "qcow2 header too short"); 881 ret = -EINVAL; 882 goto fail; 883 } 884 } 885 886 if (header.header_length > s->cluster_size) { 887 error_setg(errp, "qcow2 header exceeds cluster size"); 888 ret = -EINVAL; 889 goto fail; 890 } 891 892 if (header.header_length > sizeof(header)) { 893 s->unknown_header_fields_size = header.header_length - sizeof(header); 894 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); 895 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields, 896 s->unknown_header_fields_size); 897 if (ret < 0) { 898 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header " 899 "fields"); 900 goto fail; 901 } 902 } 903 904 if (header.backing_file_offset > s->cluster_size) { 905 error_setg(errp, "Invalid backing file offset"); 906 ret = -EINVAL; 907 goto fail; 908 } 909 910 if (header.backing_file_offset) { 911 ext_end = header.backing_file_offset; 912 } else { 913 ext_end = 1 << header.cluster_bits; 914 } 915 916 /* Handle feature bits */ 917 s->incompatible_features = header.incompatible_features; 918 s->compatible_features = header.compatible_features; 919 s->autoclear_features = header.autoclear_features; 920 921 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) { 922 void *feature_table = NULL; 923 qcow2_read_extensions(bs, header.header_length, ext_end, 924 &feature_table, NULL); 925 report_unsupported_feature(errp, feature_table, 926 s->incompatible_features & 927 ~QCOW2_INCOMPAT_MASK); 928 ret = -ENOTSUP; 929 g_free(feature_table); 930 goto fail; 931 } 932 933 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 934 /* Corrupt images may not be written to unless they are being repaired 935 */ 936 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 937 error_setg(errp, "qcow2: Image is corrupt; cannot be opened " 938 "read/write"); 939 ret = -EACCES; 940 goto fail; 941 } 942 } 943 944 /* Check support for various header values */ 945 if (header.refcount_order > 6) { 946 error_setg(errp, "Reference count entry width too large; may not " 947 "exceed 64 bits"); 948 ret = -EINVAL; 949 goto fail; 950 } 951 s->refcount_order = header.refcount_order; 952 s->refcount_bits = 1 << s->refcount_order; 953 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 954 s->refcount_max += s->refcount_max - 1; 955 956 if (header.crypt_method > QCOW_CRYPT_AES) { 957 error_setg(errp, "Unsupported encryption method: %" PRIu32, 958 header.crypt_method); 959 ret = -EINVAL; 960 goto fail; 961 } 962 if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128, 963 QCRYPTO_CIPHER_MODE_CBC)) { 964 error_setg(errp, "AES cipher not available"); 965 ret = -EINVAL; 966 goto fail; 967 } 968 s->crypt_method_header = header.crypt_method; 969 if (s->crypt_method_header) { 970 if (bdrv_uses_whitelist() && 971 s->crypt_method_header == QCOW_CRYPT_AES) { 972 error_setg(errp, 973 "Use of AES-CBC encrypted qcow2 images is no longer " 974 "supported in system emulators"); 975 error_append_hint(errp, 976 "You can use 'qemu-img convert' to convert your " 977 "image to an alternative supported format, such " 978 "as unencrypted qcow2, or raw with the LUKS " 979 "format instead.\n"); 980 ret = -ENOSYS; 981 goto fail; 982 } 983 984 bs->encrypted = true; 985 } 986 987 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ 988 s->l2_size = 1 << s->l2_bits; 989 /* 2^(s->refcount_order - 3) is the refcount width in bytes */ 990 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3); 991 s->refcount_block_size = 1 << s->refcount_block_bits; 992 bs->total_sectors = header.size / 512; 993 s->csize_shift = (62 - (s->cluster_bits - 8)); 994 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; 995 s->cluster_offset_mask = (1LL << s->csize_shift) - 1; 996 997 s->refcount_table_offset = header.refcount_table_offset; 998 s->refcount_table_size = 999 header.refcount_table_clusters << (s->cluster_bits - 3); 1000 1001 if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) { 1002 error_setg(errp, "Reference count table too large"); 1003 ret = -EINVAL; 1004 goto fail; 1005 } 1006 1007 ret = validate_table_offset(bs, s->refcount_table_offset, 1008 s->refcount_table_size, sizeof(uint64_t)); 1009 if (ret < 0) { 1010 error_setg(errp, "Invalid reference count table offset"); 1011 goto fail; 1012 } 1013 1014 /* Snapshot table offset/length */ 1015 if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) { 1016 error_setg(errp, "Too many snapshots"); 1017 ret = -EINVAL; 1018 goto fail; 1019 } 1020 1021 ret = validate_table_offset(bs, header.snapshots_offset, 1022 header.nb_snapshots, 1023 sizeof(QCowSnapshotHeader)); 1024 if (ret < 0) { 1025 error_setg(errp, "Invalid snapshot table offset"); 1026 goto fail; 1027 } 1028 1029 /* read the level 1 table */ 1030 if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) { 1031 error_setg(errp, "Active L1 table too large"); 1032 ret = -EFBIG; 1033 goto fail; 1034 } 1035 s->l1_size = header.l1_size; 1036 1037 l1_vm_state_index = size_to_l1(s, header.size); 1038 if (l1_vm_state_index > INT_MAX) { 1039 error_setg(errp, "Image is too big"); 1040 ret = -EFBIG; 1041 goto fail; 1042 } 1043 s->l1_vm_state_index = l1_vm_state_index; 1044 1045 /* the L1 table must contain at least enough entries to put 1046 header.size bytes */ 1047 if (s->l1_size < s->l1_vm_state_index) { 1048 error_setg(errp, "L1 table is too small"); 1049 ret = -EINVAL; 1050 goto fail; 1051 } 1052 1053 ret = validate_table_offset(bs, header.l1_table_offset, 1054 header.l1_size, sizeof(uint64_t)); 1055 if (ret < 0) { 1056 error_setg(errp, "Invalid L1 table offset"); 1057 goto fail; 1058 } 1059 s->l1_table_offset = header.l1_table_offset; 1060 1061 1062 if (s->l1_size > 0) { 1063 s->l1_table = qemu_try_blockalign(bs->file->bs, 1064 align_offset(s->l1_size * sizeof(uint64_t), 512)); 1065 if (s->l1_table == NULL) { 1066 error_setg(errp, "Could not allocate L1 table"); 1067 ret = -ENOMEM; 1068 goto fail; 1069 } 1070 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 1071 s->l1_size * sizeof(uint64_t)); 1072 if (ret < 0) { 1073 error_setg_errno(errp, -ret, "Could not read L1 table"); 1074 goto fail; 1075 } 1076 for(i = 0;i < s->l1_size; i++) { 1077 be64_to_cpus(&s->l1_table[i]); 1078 } 1079 } 1080 1081 /* Parse driver-specific options */ 1082 ret = qcow2_update_options(bs, options, flags, errp); 1083 if (ret < 0) { 1084 goto fail; 1085 } 1086 1087 s->cluster_cache = g_malloc(s->cluster_size); 1088 /* one more sector for decompressed data alignment */ 1089 s->cluster_data = qemu_try_blockalign(bs->file->bs, QCOW_MAX_CRYPT_CLUSTERS 1090 * s->cluster_size + 512); 1091 if (s->cluster_data == NULL) { 1092 error_setg(errp, "Could not allocate temporary cluster buffer"); 1093 ret = -ENOMEM; 1094 goto fail; 1095 } 1096 1097 s->cluster_cache_offset = -1; 1098 s->flags = flags; 1099 1100 ret = qcow2_refcount_init(bs); 1101 if (ret != 0) { 1102 error_setg_errno(errp, -ret, "Could not initialize refcount handling"); 1103 goto fail; 1104 } 1105 1106 QLIST_INIT(&s->cluster_allocs); 1107 QTAILQ_INIT(&s->discards); 1108 1109 /* read qcow2 extensions */ 1110 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, 1111 &local_err)) { 1112 error_propagate(errp, local_err); 1113 ret = -EINVAL; 1114 goto fail; 1115 } 1116 1117 /* read the backing file name */ 1118 if (header.backing_file_offset != 0) { 1119 len = header.backing_file_size; 1120 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || 1121 len >= sizeof(bs->backing_file)) { 1122 error_setg(errp, "Backing file name too long"); 1123 ret = -EINVAL; 1124 goto fail; 1125 } 1126 ret = bdrv_pread(bs->file, header.backing_file_offset, 1127 bs->backing_file, len); 1128 if (ret < 0) { 1129 error_setg_errno(errp, -ret, "Could not read backing file name"); 1130 goto fail; 1131 } 1132 bs->backing_file[len] = '\0'; 1133 s->image_backing_file = g_strdup(bs->backing_file); 1134 } 1135 1136 /* Internal snapshots */ 1137 s->snapshots_offset = header.snapshots_offset; 1138 s->nb_snapshots = header.nb_snapshots; 1139 1140 ret = qcow2_read_snapshots(bs); 1141 if (ret < 0) { 1142 error_setg_errno(errp, -ret, "Could not read snapshots"); 1143 goto fail; 1144 } 1145 1146 /* Clear unknown autoclear feature bits */ 1147 if (!bs->read_only && !(flags & BDRV_O_INACTIVE) && s->autoclear_features) { 1148 s->autoclear_features = 0; 1149 ret = qcow2_update_header(bs); 1150 if (ret < 0) { 1151 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 1152 goto fail; 1153 } 1154 } 1155 1156 /* Initialise locks */ 1157 qemu_co_mutex_init(&s->lock); 1158 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP; 1159 1160 /* Repair image if dirty */ 1161 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only && 1162 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { 1163 BdrvCheckResult result = {0}; 1164 1165 ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); 1166 if (ret < 0) { 1167 error_setg_errno(errp, -ret, "Could not repair dirty image"); 1168 goto fail; 1169 } 1170 } 1171 1172 #ifdef DEBUG_ALLOC 1173 { 1174 BdrvCheckResult result = {0}; 1175 qcow2_check_refcounts(bs, &result, 0); 1176 } 1177 #endif 1178 return ret; 1179 1180 fail: 1181 g_free(s->unknown_header_fields); 1182 cleanup_unknown_header_ext(bs); 1183 qcow2_free_snapshots(bs); 1184 qcow2_refcount_close(bs); 1185 qemu_vfree(s->l1_table); 1186 /* else pre-write overlap checks in cache_destroy may crash */ 1187 s->l1_table = NULL; 1188 cache_clean_timer_del(bs); 1189 if (s->l2_table_cache) { 1190 qcow2_cache_destroy(bs, s->l2_table_cache); 1191 } 1192 if (s->refcount_block_cache) { 1193 qcow2_cache_destroy(bs, s->refcount_block_cache); 1194 } 1195 g_free(s->cluster_cache); 1196 qemu_vfree(s->cluster_data); 1197 return ret; 1198 } 1199 1200 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) 1201 { 1202 BDRVQcow2State *s = bs->opaque; 1203 1204 if (bs->encrypted) { 1205 /* Encryption works on a sector granularity */ 1206 bs->bl.request_alignment = BDRV_SECTOR_SIZE; 1207 } 1208 bs->bl.pwrite_zeroes_alignment = s->cluster_size; 1209 } 1210 1211 static int qcow2_set_key(BlockDriverState *bs, const char *key) 1212 { 1213 BDRVQcow2State *s = bs->opaque; 1214 uint8_t keybuf[16]; 1215 int len, i; 1216 Error *err = NULL; 1217 1218 memset(keybuf, 0, 16); 1219 len = strlen(key); 1220 if (len > 16) 1221 len = 16; 1222 /* XXX: we could compress the chars to 7 bits to increase 1223 entropy */ 1224 for(i = 0;i < len;i++) { 1225 keybuf[i] = key[i]; 1226 } 1227 assert(bs->encrypted); 1228 1229 qcrypto_cipher_free(s->cipher); 1230 s->cipher = qcrypto_cipher_new( 1231 QCRYPTO_CIPHER_ALG_AES_128, 1232 QCRYPTO_CIPHER_MODE_CBC, 1233 keybuf, G_N_ELEMENTS(keybuf), 1234 &err); 1235 1236 if (!s->cipher) { 1237 /* XXX would be nice if errors in this method could 1238 * be properly propagate to the caller. Would need 1239 * the bdrv_set_key() API signature to be fixed. */ 1240 error_free(err); 1241 return -1; 1242 } 1243 return 0; 1244 } 1245 1246 static int qcow2_reopen_prepare(BDRVReopenState *state, 1247 BlockReopenQueue *queue, Error **errp) 1248 { 1249 Qcow2ReopenState *r; 1250 int ret; 1251 1252 r = g_new0(Qcow2ReopenState, 1); 1253 state->opaque = r; 1254 1255 ret = qcow2_update_options_prepare(state->bs, r, state->options, 1256 state->flags, errp); 1257 if (ret < 0) { 1258 goto fail; 1259 } 1260 1261 /* We need to write out any unwritten data if we reopen read-only. */ 1262 if ((state->flags & BDRV_O_RDWR) == 0) { 1263 ret = bdrv_flush(state->bs); 1264 if (ret < 0) { 1265 goto fail; 1266 } 1267 1268 ret = qcow2_mark_clean(state->bs); 1269 if (ret < 0) { 1270 goto fail; 1271 } 1272 } 1273 1274 return 0; 1275 1276 fail: 1277 qcow2_update_options_abort(state->bs, r); 1278 g_free(r); 1279 return ret; 1280 } 1281 1282 static void qcow2_reopen_commit(BDRVReopenState *state) 1283 { 1284 qcow2_update_options_commit(state->bs, state->opaque); 1285 g_free(state->opaque); 1286 } 1287 1288 static void qcow2_reopen_abort(BDRVReopenState *state) 1289 { 1290 qcow2_update_options_abort(state->bs, state->opaque); 1291 g_free(state->opaque); 1292 } 1293 1294 static void qcow2_join_options(QDict *options, QDict *old_options) 1295 { 1296 bool has_new_overlap_template = 1297 qdict_haskey(options, QCOW2_OPT_OVERLAP) || 1298 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE); 1299 bool has_new_total_cache_size = 1300 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE); 1301 bool has_all_cache_options; 1302 1303 /* New overlap template overrides all old overlap options */ 1304 if (has_new_overlap_template) { 1305 qdict_del(old_options, QCOW2_OPT_OVERLAP); 1306 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE); 1307 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER); 1308 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1); 1309 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2); 1310 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE); 1311 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK); 1312 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE); 1313 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1); 1314 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2); 1315 } 1316 1317 /* New total cache size overrides all old options */ 1318 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) { 1319 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE); 1320 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1321 } 1322 1323 qdict_join(options, old_options, false); 1324 1325 /* 1326 * If after merging all cache size options are set, an old total size is 1327 * overwritten. Do keep all options, however, if all three are new. The 1328 * resulting error message is what we want to happen. 1329 */ 1330 has_all_cache_options = 1331 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) || 1332 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) || 1333 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1334 1335 if (has_all_cache_options && !has_new_total_cache_size) { 1336 qdict_del(options, QCOW2_OPT_CACHE_SIZE); 1337 } 1338 } 1339 1340 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs, 1341 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file) 1342 { 1343 BDRVQcow2State *s = bs->opaque; 1344 uint64_t cluster_offset; 1345 int index_in_cluster, ret; 1346 unsigned int bytes; 1347 int64_t status = 0; 1348 1349 bytes = MIN(INT_MAX, nb_sectors * BDRV_SECTOR_SIZE); 1350 qemu_co_mutex_lock(&s->lock); 1351 ret = qcow2_get_cluster_offset(bs, sector_num << 9, &bytes, 1352 &cluster_offset); 1353 qemu_co_mutex_unlock(&s->lock); 1354 if (ret < 0) { 1355 return ret; 1356 } 1357 1358 *pnum = bytes >> BDRV_SECTOR_BITS; 1359 1360 if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED && 1361 !s->cipher) { 1362 index_in_cluster = sector_num & (s->cluster_sectors - 1); 1363 cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS); 1364 *file = bs->file->bs; 1365 status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset; 1366 } 1367 if (ret == QCOW2_CLUSTER_ZERO) { 1368 status |= BDRV_BLOCK_ZERO; 1369 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) { 1370 status |= BDRV_BLOCK_DATA; 1371 } 1372 return status; 1373 } 1374 1375 /* handle reading after the end of the backing file */ 1376 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov, 1377 int64_t offset, int bytes) 1378 { 1379 uint64_t bs_size = bs->total_sectors * BDRV_SECTOR_SIZE; 1380 int n1; 1381 1382 if ((offset + bytes) <= bs_size) { 1383 return bytes; 1384 } 1385 1386 if (offset >= bs_size) { 1387 n1 = 0; 1388 } else { 1389 n1 = bs_size - offset; 1390 } 1391 1392 qemu_iovec_memset(qiov, n1, 0, bytes - n1); 1393 1394 return n1; 1395 } 1396 1397 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset, 1398 uint64_t bytes, QEMUIOVector *qiov, 1399 int flags) 1400 { 1401 BDRVQcow2State *s = bs->opaque; 1402 int offset_in_cluster, n1; 1403 int ret; 1404 unsigned int cur_bytes; /* number of bytes in current iteration */ 1405 uint64_t cluster_offset = 0; 1406 uint64_t bytes_done = 0; 1407 QEMUIOVector hd_qiov; 1408 uint8_t *cluster_data = NULL; 1409 1410 qemu_iovec_init(&hd_qiov, qiov->niov); 1411 1412 qemu_co_mutex_lock(&s->lock); 1413 1414 while (bytes != 0) { 1415 1416 /* prepare next request */ 1417 cur_bytes = MIN(bytes, INT_MAX); 1418 if (s->cipher) { 1419 cur_bytes = MIN(cur_bytes, 1420 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1421 } 1422 1423 ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset); 1424 if (ret < 0) { 1425 goto fail; 1426 } 1427 1428 offset_in_cluster = offset_into_cluster(s, offset); 1429 1430 qemu_iovec_reset(&hd_qiov); 1431 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 1432 1433 switch (ret) { 1434 case QCOW2_CLUSTER_UNALLOCATED: 1435 1436 if (bs->backing) { 1437 /* read from the base image */ 1438 n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov, 1439 offset, cur_bytes); 1440 if (n1 > 0) { 1441 QEMUIOVector local_qiov; 1442 1443 qemu_iovec_init(&local_qiov, hd_qiov.niov); 1444 qemu_iovec_concat(&local_qiov, &hd_qiov, 0, n1); 1445 1446 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 1447 qemu_co_mutex_unlock(&s->lock); 1448 ret = bdrv_co_preadv(bs->backing, offset, n1, 1449 &local_qiov, 0); 1450 qemu_co_mutex_lock(&s->lock); 1451 1452 qemu_iovec_destroy(&local_qiov); 1453 1454 if (ret < 0) { 1455 goto fail; 1456 } 1457 } 1458 } else { 1459 /* Note: in this case, no need to wait */ 1460 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 1461 } 1462 break; 1463 1464 case QCOW2_CLUSTER_ZERO: 1465 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 1466 break; 1467 1468 case QCOW2_CLUSTER_COMPRESSED: 1469 /* add AIO support for compressed blocks ? */ 1470 ret = qcow2_decompress_cluster(bs, cluster_offset); 1471 if (ret < 0) { 1472 goto fail; 1473 } 1474 1475 qemu_iovec_from_buf(&hd_qiov, 0, 1476 s->cluster_cache + offset_in_cluster, 1477 cur_bytes); 1478 break; 1479 1480 case QCOW2_CLUSTER_NORMAL: 1481 if ((cluster_offset & 511) != 0) { 1482 ret = -EIO; 1483 goto fail; 1484 } 1485 1486 if (bs->encrypted) { 1487 assert(s->cipher); 1488 1489 /* 1490 * For encrypted images, read everything into a temporary 1491 * contiguous buffer on which the AES functions can work. 1492 */ 1493 if (!cluster_data) { 1494 cluster_data = 1495 qemu_try_blockalign(bs->file->bs, 1496 QCOW_MAX_CRYPT_CLUSTERS 1497 * s->cluster_size); 1498 if (cluster_data == NULL) { 1499 ret = -ENOMEM; 1500 goto fail; 1501 } 1502 } 1503 1504 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1505 qemu_iovec_reset(&hd_qiov); 1506 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 1507 } 1508 1509 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 1510 qemu_co_mutex_unlock(&s->lock); 1511 ret = bdrv_co_preadv(bs->file, 1512 cluster_offset + offset_in_cluster, 1513 cur_bytes, &hd_qiov, 0); 1514 qemu_co_mutex_lock(&s->lock); 1515 if (ret < 0) { 1516 goto fail; 1517 } 1518 if (bs->encrypted) { 1519 assert(s->cipher); 1520 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); 1521 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0); 1522 Error *err = NULL; 1523 if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS, 1524 cluster_data, cluster_data, 1525 cur_bytes >> BDRV_SECTOR_BITS, 1526 false, &err) < 0) { 1527 error_free(err); 1528 ret = -EIO; 1529 goto fail; 1530 } 1531 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes); 1532 } 1533 break; 1534 1535 default: 1536 g_assert_not_reached(); 1537 ret = -EIO; 1538 goto fail; 1539 } 1540 1541 bytes -= cur_bytes; 1542 offset += cur_bytes; 1543 bytes_done += cur_bytes; 1544 } 1545 ret = 0; 1546 1547 fail: 1548 qemu_co_mutex_unlock(&s->lock); 1549 1550 qemu_iovec_destroy(&hd_qiov); 1551 qemu_vfree(cluster_data); 1552 1553 return ret; 1554 } 1555 1556 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset, 1557 uint64_t bytes, QEMUIOVector *qiov, 1558 int flags) 1559 { 1560 BDRVQcow2State *s = bs->opaque; 1561 int offset_in_cluster; 1562 int ret; 1563 unsigned int cur_bytes; /* number of sectors in current iteration */ 1564 uint64_t cluster_offset; 1565 QEMUIOVector hd_qiov; 1566 uint64_t bytes_done = 0; 1567 uint8_t *cluster_data = NULL; 1568 QCowL2Meta *l2meta = NULL; 1569 1570 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes); 1571 1572 qemu_iovec_init(&hd_qiov, qiov->niov); 1573 1574 s->cluster_cache_offset = -1; /* disable compressed cache */ 1575 1576 qemu_co_mutex_lock(&s->lock); 1577 1578 while (bytes != 0) { 1579 1580 l2meta = NULL; 1581 1582 trace_qcow2_writev_start_part(qemu_coroutine_self()); 1583 offset_in_cluster = offset_into_cluster(s, offset); 1584 cur_bytes = MIN(bytes, INT_MAX); 1585 if (bs->encrypted) { 1586 cur_bytes = MIN(cur_bytes, 1587 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size 1588 - offset_in_cluster); 1589 } 1590 1591 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 1592 &cluster_offset, &l2meta); 1593 if (ret < 0) { 1594 goto fail; 1595 } 1596 1597 assert((cluster_offset & 511) == 0); 1598 1599 qemu_iovec_reset(&hd_qiov); 1600 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 1601 1602 if (bs->encrypted) { 1603 Error *err = NULL; 1604 assert(s->cipher); 1605 if (!cluster_data) { 1606 cluster_data = qemu_try_blockalign(bs->file->bs, 1607 QCOW_MAX_CRYPT_CLUSTERS 1608 * s->cluster_size); 1609 if (cluster_data == NULL) { 1610 ret = -ENOMEM; 1611 goto fail; 1612 } 1613 } 1614 1615 assert(hd_qiov.size <= 1616 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1617 qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size); 1618 1619 if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS, 1620 cluster_data, cluster_data, 1621 cur_bytes >>BDRV_SECTOR_BITS, 1622 true, &err) < 0) { 1623 error_free(err); 1624 ret = -EIO; 1625 goto fail; 1626 } 1627 1628 qemu_iovec_reset(&hd_qiov); 1629 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 1630 } 1631 1632 ret = qcow2_pre_write_overlap_check(bs, 0, 1633 cluster_offset + offset_in_cluster, cur_bytes); 1634 if (ret < 0) { 1635 goto fail; 1636 } 1637 1638 qemu_co_mutex_unlock(&s->lock); 1639 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 1640 trace_qcow2_writev_data(qemu_coroutine_self(), 1641 cluster_offset + offset_in_cluster); 1642 ret = bdrv_co_pwritev(bs->file, 1643 cluster_offset + offset_in_cluster, 1644 cur_bytes, &hd_qiov, 0); 1645 qemu_co_mutex_lock(&s->lock); 1646 if (ret < 0) { 1647 goto fail; 1648 } 1649 1650 while (l2meta != NULL) { 1651 QCowL2Meta *next; 1652 1653 ret = qcow2_alloc_cluster_link_l2(bs, l2meta); 1654 if (ret < 0) { 1655 goto fail; 1656 } 1657 1658 /* Take the request off the list of running requests */ 1659 if (l2meta->nb_clusters != 0) { 1660 QLIST_REMOVE(l2meta, next_in_flight); 1661 } 1662 1663 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1664 1665 next = l2meta->next; 1666 g_free(l2meta); 1667 l2meta = next; 1668 } 1669 1670 bytes -= cur_bytes; 1671 offset += cur_bytes; 1672 bytes_done += cur_bytes; 1673 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes); 1674 } 1675 ret = 0; 1676 1677 fail: 1678 qemu_co_mutex_unlock(&s->lock); 1679 1680 while (l2meta != NULL) { 1681 QCowL2Meta *next; 1682 1683 if (l2meta->nb_clusters != 0) { 1684 QLIST_REMOVE(l2meta, next_in_flight); 1685 } 1686 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1687 1688 next = l2meta->next; 1689 g_free(l2meta); 1690 l2meta = next; 1691 } 1692 1693 qemu_iovec_destroy(&hd_qiov); 1694 qemu_vfree(cluster_data); 1695 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 1696 1697 return ret; 1698 } 1699 1700 static int qcow2_inactivate(BlockDriverState *bs) 1701 { 1702 BDRVQcow2State *s = bs->opaque; 1703 int ret, result = 0; 1704 1705 ret = qcow2_cache_flush(bs, s->l2_table_cache); 1706 if (ret) { 1707 result = ret; 1708 error_report("Failed to flush the L2 table cache: %s", 1709 strerror(-ret)); 1710 } 1711 1712 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 1713 if (ret) { 1714 result = ret; 1715 error_report("Failed to flush the refcount block cache: %s", 1716 strerror(-ret)); 1717 } 1718 1719 if (result == 0) { 1720 qcow2_mark_clean(bs); 1721 } 1722 1723 return result; 1724 } 1725 1726 static void qcow2_close(BlockDriverState *bs) 1727 { 1728 BDRVQcow2State *s = bs->opaque; 1729 qemu_vfree(s->l1_table); 1730 /* else pre-write overlap checks in cache_destroy may crash */ 1731 s->l1_table = NULL; 1732 1733 if (!(s->flags & BDRV_O_INACTIVE)) { 1734 qcow2_inactivate(bs); 1735 } 1736 1737 cache_clean_timer_del(bs); 1738 qcow2_cache_destroy(bs, s->l2_table_cache); 1739 qcow2_cache_destroy(bs, s->refcount_block_cache); 1740 1741 qcrypto_cipher_free(s->cipher); 1742 s->cipher = NULL; 1743 1744 g_free(s->unknown_header_fields); 1745 cleanup_unknown_header_ext(bs); 1746 1747 g_free(s->image_backing_file); 1748 g_free(s->image_backing_format); 1749 1750 g_free(s->cluster_cache); 1751 qemu_vfree(s->cluster_data); 1752 qcow2_refcount_close(bs); 1753 qcow2_free_snapshots(bs); 1754 } 1755 1756 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp) 1757 { 1758 BDRVQcow2State *s = bs->opaque; 1759 int flags = s->flags; 1760 QCryptoCipher *cipher = NULL; 1761 QDict *options; 1762 Error *local_err = NULL; 1763 int ret; 1764 1765 /* 1766 * Backing files are read-only which makes all of their metadata immutable, 1767 * that means we don't have to worry about reopening them here. 1768 */ 1769 1770 cipher = s->cipher; 1771 s->cipher = NULL; 1772 1773 qcow2_close(bs); 1774 1775 memset(s, 0, sizeof(BDRVQcow2State)); 1776 options = qdict_clone_shallow(bs->options); 1777 1778 flags &= ~BDRV_O_INACTIVE; 1779 ret = qcow2_open(bs, options, flags, &local_err); 1780 QDECREF(options); 1781 if (local_err) { 1782 error_propagate(errp, local_err); 1783 error_prepend(errp, "Could not reopen qcow2 layer: "); 1784 bs->drv = NULL; 1785 return; 1786 } else if (ret < 0) { 1787 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer"); 1788 bs->drv = NULL; 1789 return; 1790 } 1791 1792 s->cipher = cipher; 1793 } 1794 1795 static size_t header_ext_add(char *buf, uint32_t magic, const void *s, 1796 size_t len, size_t buflen) 1797 { 1798 QCowExtension *ext_backing_fmt = (QCowExtension*) buf; 1799 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); 1800 1801 if (buflen < ext_len) { 1802 return -ENOSPC; 1803 } 1804 1805 *ext_backing_fmt = (QCowExtension) { 1806 .magic = cpu_to_be32(magic), 1807 .len = cpu_to_be32(len), 1808 }; 1809 1810 if (len) { 1811 memcpy(buf + sizeof(QCowExtension), s, len); 1812 } 1813 1814 return ext_len; 1815 } 1816 1817 /* 1818 * Updates the qcow2 header, including the variable length parts of it, i.e. 1819 * the backing file name and all extensions. qcow2 was not designed to allow 1820 * such changes, so if we run out of space (we can only use the first cluster) 1821 * this function may fail. 1822 * 1823 * Returns 0 on success, -errno in error cases. 1824 */ 1825 int qcow2_update_header(BlockDriverState *bs) 1826 { 1827 BDRVQcow2State *s = bs->opaque; 1828 QCowHeader *header; 1829 char *buf; 1830 size_t buflen = s->cluster_size; 1831 int ret; 1832 uint64_t total_size; 1833 uint32_t refcount_table_clusters; 1834 size_t header_length; 1835 Qcow2UnknownHeaderExtension *uext; 1836 1837 buf = qemu_blockalign(bs, buflen); 1838 1839 /* Header structure */ 1840 header = (QCowHeader*) buf; 1841 1842 if (buflen < sizeof(*header)) { 1843 ret = -ENOSPC; 1844 goto fail; 1845 } 1846 1847 header_length = sizeof(*header) + s->unknown_header_fields_size; 1848 total_size = bs->total_sectors * BDRV_SECTOR_SIZE; 1849 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); 1850 1851 *header = (QCowHeader) { 1852 /* Version 2 fields */ 1853 .magic = cpu_to_be32(QCOW_MAGIC), 1854 .version = cpu_to_be32(s->qcow_version), 1855 .backing_file_offset = 0, 1856 .backing_file_size = 0, 1857 .cluster_bits = cpu_to_be32(s->cluster_bits), 1858 .size = cpu_to_be64(total_size), 1859 .crypt_method = cpu_to_be32(s->crypt_method_header), 1860 .l1_size = cpu_to_be32(s->l1_size), 1861 .l1_table_offset = cpu_to_be64(s->l1_table_offset), 1862 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), 1863 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), 1864 .nb_snapshots = cpu_to_be32(s->nb_snapshots), 1865 .snapshots_offset = cpu_to_be64(s->snapshots_offset), 1866 1867 /* Version 3 fields */ 1868 .incompatible_features = cpu_to_be64(s->incompatible_features), 1869 .compatible_features = cpu_to_be64(s->compatible_features), 1870 .autoclear_features = cpu_to_be64(s->autoclear_features), 1871 .refcount_order = cpu_to_be32(s->refcount_order), 1872 .header_length = cpu_to_be32(header_length), 1873 }; 1874 1875 /* For older versions, write a shorter header */ 1876 switch (s->qcow_version) { 1877 case 2: 1878 ret = offsetof(QCowHeader, incompatible_features); 1879 break; 1880 case 3: 1881 ret = sizeof(*header); 1882 break; 1883 default: 1884 ret = -EINVAL; 1885 goto fail; 1886 } 1887 1888 buf += ret; 1889 buflen -= ret; 1890 memset(buf, 0, buflen); 1891 1892 /* Preserve any unknown field in the header */ 1893 if (s->unknown_header_fields_size) { 1894 if (buflen < s->unknown_header_fields_size) { 1895 ret = -ENOSPC; 1896 goto fail; 1897 } 1898 1899 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); 1900 buf += s->unknown_header_fields_size; 1901 buflen -= s->unknown_header_fields_size; 1902 } 1903 1904 /* Backing file format header extension */ 1905 if (s->image_backing_format) { 1906 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, 1907 s->image_backing_format, 1908 strlen(s->image_backing_format), 1909 buflen); 1910 if (ret < 0) { 1911 goto fail; 1912 } 1913 1914 buf += ret; 1915 buflen -= ret; 1916 } 1917 1918 /* Feature table */ 1919 if (s->qcow_version >= 3) { 1920 Qcow2Feature features[] = { 1921 { 1922 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 1923 .bit = QCOW2_INCOMPAT_DIRTY_BITNR, 1924 .name = "dirty bit", 1925 }, 1926 { 1927 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 1928 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, 1929 .name = "corrupt bit", 1930 }, 1931 { 1932 .type = QCOW2_FEAT_TYPE_COMPATIBLE, 1933 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 1934 .name = "lazy refcounts", 1935 }, 1936 }; 1937 1938 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, 1939 features, sizeof(features), buflen); 1940 if (ret < 0) { 1941 goto fail; 1942 } 1943 buf += ret; 1944 buflen -= ret; 1945 } 1946 1947 /* Keep unknown header extensions */ 1948 QLIST_FOREACH(uext, &s->unknown_header_ext, next) { 1949 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); 1950 if (ret < 0) { 1951 goto fail; 1952 } 1953 1954 buf += ret; 1955 buflen -= ret; 1956 } 1957 1958 /* End of header extensions */ 1959 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); 1960 if (ret < 0) { 1961 goto fail; 1962 } 1963 1964 buf += ret; 1965 buflen -= ret; 1966 1967 /* Backing file name */ 1968 if (s->image_backing_file) { 1969 size_t backing_file_len = strlen(s->image_backing_file); 1970 1971 if (buflen < backing_file_len) { 1972 ret = -ENOSPC; 1973 goto fail; 1974 } 1975 1976 /* Using strncpy is ok here, since buf is not NUL-terminated. */ 1977 strncpy(buf, s->image_backing_file, buflen); 1978 1979 header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); 1980 header->backing_file_size = cpu_to_be32(backing_file_len); 1981 } 1982 1983 /* Write the new header */ 1984 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size); 1985 if (ret < 0) { 1986 goto fail; 1987 } 1988 1989 ret = 0; 1990 fail: 1991 qemu_vfree(header); 1992 return ret; 1993 } 1994 1995 static int qcow2_change_backing_file(BlockDriverState *bs, 1996 const char *backing_file, const char *backing_fmt) 1997 { 1998 BDRVQcow2State *s = bs->opaque; 1999 2000 if (backing_file && strlen(backing_file) > 1023) { 2001 return -EINVAL; 2002 } 2003 2004 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 2005 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 2006 2007 g_free(s->image_backing_file); 2008 g_free(s->image_backing_format); 2009 2010 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL; 2011 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL; 2012 2013 return qcow2_update_header(bs); 2014 } 2015 2016 static int preallocate(BlockDriverState *bs) 2017 { 2018 uint64_t bytes; 2019 uint64_t offset; 2020 uint64_t host_offset = 0; 2021 unsigned int cur_bytes; 2022 int ret; 2023 QCowL2Meta *meta; 2024 2025 bytes = bdrv_getlength(bs); 2026 offset = 0; 2027 2028 while (bytes) { 2029 cur_bytes = MIN(bytes, INT_MAX); 2030 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 2031 &host_offset, &meta); 2032 if (ret < 0) { 2033 return ret; 2034 } 2035 2036 while (meta) { 2037 QCowL2Meta *next = meta->next; 2038 2039 ret = qcow2_alloc_cluster_link_l2(bs, meta); 2040 if (ret < 0) { 2041 qcow2_free_any_clusters(bs, meta->alloc_offset, 2042 meta->nb_clusters, QCOW2_DISCARD_NEVER); 2043 return ret; 2044 } 2045 2046 /* There are no dependent requests, but we need to remove our 2047 * request from the list of in-flight requests */ 2048 QLIST_REMOVE(meta, next_in_flight); 2049 2050 g_free(meta); 2051 meta = next; 2052 } 2053 2054 /* TODO Preallocate data if requested */ 2055 2056 bytes -= cur_bytes; 2057 offset += cur_bytes; 2058 } 2059 2060 /* 2061 * It is expected that the image file is large enough to actually contain 2062 * all of the allocated clusters (otherwise we get failing reads after 2063 * EOF). Extend the image to the last allocated sector. 2064 */ 2065 if (host_offset != 0) { 2066 uint8_t data = 0; 2067 ret = bdrv_pwrite(bs->file, (host_offset + cur_bytes) - 1, 2068 &data, 1); 2069 if (ret < 0) { 2070 return ret; 2071 } 2072 } 2073 2074 return 0; 2075 } 2076 2077 static int qcow2_create2(const char *filename, int64_t total_size, 2078 const char *backing_file, const char *backing_format, 2079 int flags, size_t cluster_size, PreallocMode prealloc, 2080 QemuOpts *opts, int version, int refcount_order, 2081 Error **errp) 2082 { 2083 int cluster_bits; 2084 QDict *options; 2085 2086 /* Calculate cluster_bits */ 2087 cluster_bits = ctz32(cluster_size); 2088 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 2089 (1 << cluster_bits) != cluster_size) 2090 { 2091 error_setg(errp, "Cluster size must be a power of two between %d and " 2092 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 2093 return -EINVAL; 2094 } 2095 2096 /* 2097 * Open the image file and write a minimal qcow2 header. 2098 * 2099 * We keep things simple and start with a zero-sized image. We also 2100 * do without refcount blocks or a L1 table for now. We'll fix the 2101 * inconsistency later. 2102 * 2103 * We do need a refcount table because growing the refcount table means 2104 * allocating two new refcount blocks - the seconds of which would be at 2105 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 2106 * size for any qcow2 image. 2107 */ 2108 BlockBackend *blk; 2109 QCowHeader *header; 2110 uint64_t* refcount_table; 2111 Error *local_err = NULL; 2112 int ret; 2113 2114 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 2115 /* Note: The following calculation does not need to be exact; if it is a 2116 * bit off, either some bytes will be "leaked" (which is fine) or we 2117 * will need to increase the file size by some bytes (which is fine, 2118 * too, as long as the bulk is allocated here). Therefore, using 2119 * floating point arithmetic is fine. */ 2120 int64_t meta_size = 0; 2121 uint64_t nreftablee, nrefblocke, nl1e, nl2e; 2122 int64_t aligned_total_size = align_offset(total_size, cluster_size); 2123 int refblock_bits, refblock_size; 2124 /* refcount entry size in bytes */ 2125 double rces = (1 << refcount_order) / 8.; 2126 2127 /* see qcow2_open() */ 2128 refblock_bits = cluster_bits - (refcount_order - 3); 2129 refblock_size = 1 << refblock_bits; 2130 2131 /* header: 1 cluster */ 2132 meta_size += cluster_size; 2133 2134 /* total size of L2 tables */ 2135 nl2e = aligned_total_size / cluster_size; 2136 nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t)); 2137 meta_size += nl2e * sizeof(uint64_t); 2138 2139 /* total size of L1 tables */ 2140 nl1e = nl2e * sizeof(uint64_t) / cluster_size; 2141 nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t)); 2142 meta_size += nl1e * sizeof(uint64_t); 2143 2144 /* total size of refcount blocks 2145 * 2146 * note: every host cluster is reference-counted, including metadata 2147 * (even refcount blocks are recursively included). 2148 * Let: 2149 * a = total_size (this is the guest disk size) 2150 * m = meta size not including refcount blocks and refcount tables 2151 * c = cluster size 2152 * y1 = number of refcount blocks entries 2153 * y2 = meta size including everything 2154 * rces = refcount entry size in bytes 2155 * then, 2156 * y1 = (y2 + a)/c 2157 * y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m 2158 * we can get y1: 2159 * y1 = (a + m) / (c - rces - rces * sizeof(u64) / c) 2160 */ 2161 nrefblocke = (aligned_total_size + meta_size + cluster_size) 2162 / (cluster_size - rces - rces * sizeof(uint64_t) 2163 / cluster_size); 2164 meta_size += DIV_ROUND_UP(nrefblocke, refblock_size) * cluster_size; 2165 2166 /* total size of refcount tables */ 2167 nreftablee = nrefblocke / refblock_size; 2168 nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t)); 2169 meta_size += nreftablee * sizeof(uint64_t); 2170 2171 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, 2172 aligned_total_size + meta_size, &error_abort); 2173 qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc], 2174 &error_abort); 2175 } 2176 2177 ret = bdrv_create_file(filename, opts, &local_err); 2178 if (ret < 0) { 2179 error_propagate(errp, local_err); 2180 return ret; 2181 } 2182 2183 blk = blk_new_open(filename, NULL, NULL, 2184 BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err); 2185 if (blk == NULL) { 2186 error_propagate(errp, local_err); 2187 return -EIO; 2188 } 2189 2190 blk_set_allow_write_beyond_eof(blk, true); 2191 2192 /* Write the header */ 2193 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); 2194 header = g_malloc0(cluster_size); 2195 *header = (QCowHeader) { 2196 .magic = cpu_to_be32(QCOW_MAGIC), 2197 .version = cpu_to_be32(version), 2198 .cluster_bits = cpu_to_be32(cluster_bits), 2199 .size = cpu_to_be64(0), 2200 .l1_table_offset = cpu_to_be64(0), 2201 .l1_size = cpu_to_be32(0), 2202 .refcount_table_offset = cpu_to_be64(cluster_size), 2203 .refcount_table_clusters = cpu_to_be32(1), 2204 .refcount_order = cpu_to_be32(refcount_order), 2205 .header_length = cpu_to_be32(sizeof(*header)), 2206 }; 2207 2208 if (flags & BLOCK_FLAG_ENCRYPT) { 2209 header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES); 2210 } else { 2211 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 2212 } 2213 2214 if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) { 2215 header->compatible_features |= 2216 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); 2217 } 2218 2219 ret = blk_pwrite(blk, 0, header, cluster_size, 0); 2220 g_free(header); 2221 if (ret < 0) { 2222 error_setg_errno(errp, -ret, "Could not write qcow2 header"); 2223 goto out; 2224 } 2225 2226 /* Write a refcount table with one refcount block */ 2227 refcount_table = g_malloc0(2 * cluster_size); 2228 refcount_table[0] = cpu_to_be64(2 * cluster_size); 2229 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0); 2230 g_free(refcount_table); 2231 2232 if (ret < 0) { 2233 error_setg_errno(errp, -ret, "Could not write refcount table"); 2234 goto out; 2235 } 2236 2237 blk_unref(blk); 2238 blk = NULL; 2239 2240 /* 2241 * And now open the image and make it consistent first (i.e. increase the 2242 * refcount of the cluster that is occupied by the header and the refcount 2243 * table) 2244 */ 2245 options = qdict_new(); 2246 qdict_put(options, "driver", qstring_from_str("qcow2")); 2247 blk = blk_new_open(filename, NULL, options, 2248 BDRV_O_RDWR | BDRV_O_NO_FLUSH, &local_err); 2249 if (blk == NULL) { 2250 error_propagate(errp, local_err); 2251 ret = -EIO; 2252 goto out; 2253 } 2254 2255 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size); 2256 if (ret < 0) { 2257 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " 2258 "header and refcount table"); 2259 goto out; 2260 2261 } else if (ret != 0) { 2262 error_report("Huh, first cluster in empty image is already in use?"); 2263 abort(); 2264 } 2265 2266 /* Create a full header (including things like feature table) */ 2267 ret = qcow2_update_header(blk_bs(blk)); 2268 if (ret < 0) { 2269 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 2270 goto out; 2271 } 2272 2273 /* Okay, now that we have a valid image, let's give it the right size */ 2274 ret = blk_truncate(blk, total_size); 2275 if (ret < 0) { 2276 error_setg_errno(errp, -ret, "Could not resize image"); 2277 goto out; 2278 } 2279 2280 /* Want a backing file? There you go.*/ 2281 if (backing_file) { 2282 ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format); 2283 if (ret < 0) { 2284 error_setg_errno(errp, -ret, "Could not assign backing file '%s' " 2285 "with format '%s'", backing_file, backing_format); 2286 goto out; 2287 } 2288 } 2289 2290 /* And if we're supposed to preallocate metadata, do that now */ 2291 if (prealloc != PREALLOC_MODE_OFF) { 2292 BDRVQcow2State *s = blk_bs(blk)->opaque; 2293 qemu_co_mutex_lock(&s->lock); 2294 ret = preallocate(blk_bs(blk)); 2295 qemu_co_mutex_unlock(&s->lock); 2296 if (ret < 0) { 2297 error_setg_errno(errp, -ret, "Could not preallocate metadata"); 2298 goto out; 2299 } 2300 } 2301 2302 blk_unref(blk); 2303 blk = NULL; 2304 2305 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */ 2306 options = qdict_new(); 2307 qdict_put(options, "driver", qstring_from_str("qcow2")); 2308 blk = blk_new_open(filename, NULL, options, 2309 BDRV_O_RDWR | BDRV_O_NO_BACKING, &local_err); 2310 if (blk == NULL) { 2311 error_propagate(errp, local_err); 2312 ret = -EIO; 2313 goto out; 2314 } 2315 2316 ret = 0; 2317 out: 2318 if (blk) { 2319 blk_unref(blk); 2320 } 2321 return ret; 2322 } 2323 2324 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp) 2325 { 2326 char *backing_file = NULL; 2327 char *backing_fmt = NULL; 2328 char *buf = NULL; 2329 uint64_t size = 0; 2330 int flags = 0; 2331 size_t cluster_size = DEFAULT_CLUSTER_SIZE; 2332 PreallocMode prealloc; 2333 int version = 3; 2334 uint64_t refcount_bits = 16; 2335 int refcount_order; 2336 Error *local_err = NULL; 2337 int ret; 2338 2339 /* Read out options */ 2340 size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2341 BDRV_SECTOR_SIZE); 2342 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 2343 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT); 2344 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) { 2345 flags |= BLOCK_FLAG_ENCRYPT; 2346 } 2347 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 2348 DEFAULT_CLUSTER_SIZE); 2349 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2350 prealloc = qapi_enum_parse(PreallocMode_lookup, buf, 2351 PREALLOC_MODE__MAX, PREALLOC_MODE_OFF, 2352 &local_err); 2353 if (local_err) { 2354 error_propagate(errp, local_err); 2355 ret = -EINVAL; 2356 goto finish; 2357 } 2358 g_free(buf); 2359 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); 2360 if (!buf) { 2361 /* keep the default */ 2362 } else if (!strcmp(buf, "0.10")) { 2363 version = 2; 2364 } else if (!strcmp(buf, "1.1")) { 2365 version = 3; 2366 } else { 2367 error_setg(errp, "Invalid compatibility level: '%s'", buf); 2368 ret = -EINVAL; 2369 goto finish; 2370 } 2371 2372 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) { 2373 flags |= BLOCK_FLAG_LAZY_REFCOUNTS; 2374 } 2375 2376 if (backing_file && prealloc != PREALLOC_MODE_OFF) { 2377 error_setg(errp, "Backing file and preallocation cannot be used at " 2378 "the same time"); 2379 ret = -EINVAL; 2380 goto finish; 2381 } 2382 2383 if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) { 2384 error_setg(errp, "Lazy refcounts only supported with compatibility " 2385 "level 1.1 and above (use compat=1.1 or greater)"); 2386 ret = -EINVAL; 2387 goto finish; 2388 } 2389 2390 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 2391 refcount_bits); 2392 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) { 2393 error_setg(errp, "Refcount width must be a power of two and may not " 2394 "exceed 64 bits"); 2395 ret = -EINVAL; 2396 goto finish; 2397 } 2398 2399 if (version < 3 && refcount_bits != 16) { 2400 error_setg(errp, "Different refcount widths than 16 bits require " 2401 "compatibility level 1.1 or above (use compat=1.1 or " 2402 "greater)"); 2403 ret = -EINVAL; 2404 goto finish; 2405 } 2406 2407 refcount_order = ctz32(refcount_bits); 2408 2409 ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags, 2410 cluster_size, prealloc, opts, version, refcount_order, 2411 &local_err); 2412 error_propagate(errp, local_err); 2413 2414 finish: 2415 g_free(backing_file); 2416 g_free(backing_fmt); 2417 g_free(buf); 2418 return ret; 2419 } 2420 2421 2422 static bool is_zero_sectors(BlockDriverState *bs, int64_t start, 2423 uint32_t count) 2424 { 2425 int nr; 2426 BlockDriverState *file; 2427 int64_t res; 2428 2429 if (!count) { 2430 return true; 2431 } 2432 res = bdrv_get_block_status_above(bs, NULL, start, count, 2433 &nr, &file); 2434 return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == count; 2435 } 2436 2437 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs, 2438 int64_t offset, int count, BdrvRequestFlags flags) 2439 { 2440 int ret; 2441 BDRVQcow2State *s = bs->opaque; 2442 2443 uint32_t head = offset % s->cluster_size; 2444 uint32_t tail = (offset + count) % s->cluster_size; 2445 2446 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, count); 2447 2448 if (head || tail) { 2449 int64_t cl_start = (offset - head) >> BDRV_SECTOR_BITS; 2450 uint64_t off; 2451 unsigned int nr; 2452 2453 assert(head + count <= s->cluster_size); 2454 2455 /* check whether remainder of cluster already reads as zero */ 2456 if (!(is_zero_sectors(bs, cl_start, 2457 DIV_ROUND_UP(head, BDRV_SECTOR_SIZE)) && 2458 is_zero_sectors(bs, (offset + count) >> BDRV_SECTOR_BITS, 2459 DIV_ROUND_UP(-tail & (s->cluster_size - 1), 2460 BDRV_SECTOR_SIZE)))) { 2461 return -ENOTSUP; 2462 } 2463 2464 qemu_co_mutex_lock(&s->lock); 2465 /* We can have new write after previous check */ 2466 offset = cl_start << BDRV_SECTOR_BITS; 2467 count = s->cluster_size; 2468 nr = s->cluster_size; 2469 ret = qcow2_get_cluster_offset(bs, offset, &nr, &off); 2470 if (ret != QCOW2_CLUSTER_UNALLOCATED && ret != QCOW2_CLUSTER_ZERO) { 2471 qemu_co_mutex_unlock(&s->lock); 2472 return -ENOTSUP; 2473 } 2474 } else { 2475 qemu_co_mutex_lock(&s->lock); 2476 } 2477 2478 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, count); 2479 2480 /* Whatever is left can use real zero clusters */ 2481 ret = qcow2_zero_clusters(bs, offset, count >> BDRV_SECTOR_BITS, flags); 2482 qemu_co_mutex_unlock(&s->lock); 2483 2484 return ret; 2485 } 2486 2487 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs, 2488 int64_t offset, int count) 2489 { 2490 int ret; 2491 BDRVQcow2State *s = bs->opaque; 2492 2493 qemu_co_mutex_lock(&s->lock); 2494 ret = qcow2_discard_clusters(bs, offset, count >> BDRV_SECTOR_BITS, 2495 QCOW2_DISCARD_REQUEST, false); 2496 qemu_co_mutex_unlock(&s->lock); 2497 return ret; 2498 } 2499 2500 static int qcow2_truncate(BlockDriverState *bs, int64_t offset) 2501 { 2502 BDRVQcow2State *s = bs->opaque; 2503 int64_t new_l1_size; 2504 int ret; 2505 2506 if (offset & 511) { 2507 error_report("The new size must be a multiple of 512"); 2508 return -EINVAL; 2509 } 2510 2511 /* cannot proceed if image has snapshots */ 2512 if (s->nb_snapshots) { 2513 error_report("Can't resize an image which has snapshots"); 2514 return -ENOTSUP; 2515 } 2516 2517 /* shrinking is currently not supported */ 2518 if (offset < bs->total_sectors * 512) { 2519 error_report("qcow2 doesn't support shrinking images yet"); 2520 return -ENOTSUP; 2521 } 2522 2523 new_l1_size = size_to_l1(s, offset); 2524 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 2525 if (ret < 0) { 2526 return ret; 2527 } 2528 2529 /* write updated header.size */ 2530 offset = cpu_to_be64(offset); 2531 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), 2532 &offset, sizeof(uint64_t)); 2533 if (ret < 0) { 2534 return ret; 2535 } 2536 2537 s->l1_vm_state_index = new_l1_size; 2538 return 0; 2539 } 2540 2541 /* XXX: put compressed sectors first, then all the cluster aligned 2542 tables to avoid losing bytes in alignment */ 2543 static coroutine_fn int 2544 qcow2_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset, 2545 uint64_t bytes, QEMUIOVector *qiov) 2546 { 2547 BDRVQcow2State *s = bs->opaque; 2548 QEMUIOVector hd_qiov; 2549 struct iovec iov; 2550 z_stream strm; 2551 int ret, out_len; 2552 uint8_t *buf, *out_buf; 2553 uint64_t cluster_offset; 2554 2555 if (bytes == 0) { 2556 /* align end of file to a sector boundary to ease reading with 2557 sector based I/Os */ 2558 cluster_offset = bdrv_getlength(bs->file->bs); 2559 return bdrv_truncate(bs->file->bs, cluster_offset); 2560 } 2561 2562 buf = qemu_blockalign(bs, s->cluster_size); 2563 if (bytes != s->cluster_size) { 2564 if (bytes > s->cluster_size || 2565 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS) 2566 { 2567 qemu_vfree(buf); 2568 return -EINVAL; 2569 } 2570 /* Zero-pad last write if image size is not cluster aligned */ 2571 memset(buf + bytes, 0, s->cluster_size - bytes); 2572 } 2573 qemu_iovec_to_buf(qiov, 0, buf, bytes); 2574 2575 out_buf = g_malloc(s->cluster_size); 2576 2577 /* best compression, small window, no zlib header */ 2578 memset(&strm, 0, sizeof(strm)); 2579 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, 2580 Z_DEFLATED, -12, 2581 9, Z_DEFAULT_STRATEGY); 2582 if (ret != 0) { 2583 ret = -EINVAL; 2584 goto fail; 2585 } 2586 2587 strm.avail_in = s->cluster_size; 2588 strm.next_in = (uint8_t *)buf; 2589 strm.avail_out = s->cluster_size; 2590 strm.next_out = out_buf; 2591 2592 ret = deflate(&strm, Z_FINISH); 2593 if (ret != Z_STREAM_END && ret != Z_OK) { 2594 deflateEnd(&strm); 2595 ret = -EINVAL; 2596 goto fail; 2597 } 2598 out_len = strm.next_out - out_buf; 2599 2600 deflateEnd(&strm); 2601 2602 if (ret != Z_STREAM_END || out_len >= s->cluster_size) { 2603 /* could not compress: write normal cluster */ 2604 ret = qcow2_co_pwritev(bs, offset, bytes, qiov, 0); 2605 if (ret < 0) { 2606 goto fail; 2607 } 2608 goto success; 2609 } 2610 2611 qemu_co_mutex_lock(&s->lock); 2612 cluster_offset = 2613 qcow2_alloc_compressed_cluster_offset(bs, offset, out_len); 2614 if (!cluster_offset) { 2615 qemu_co_mutex_unlock(&s->lock); 2616 ret = -EIO; 2617 goto fail; 2618 } 2619 cluster_offset &= s->cluster_offset_mask; 2620 2621 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len); 2622 qemu_co_mutex_unlock(&s->lock); 2623 if (ret < 0) { 2624 goto fail; 2625 } 2626 2627 iov = (struct iovec) { 2628 .iov_base = out_buf, 2629 .iov_len = out_len, 2630 }; 2631 qemu_iovec_init_external(&hd_qiov, &iov, 1); 2632 2633 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); 2634 ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0); 2635 if (ret < 0) { 2636 goto fail; 2637 } 2638 success: 2639 ret = 0; 2640 fail: 2641 qemu_vfree(buf); 2642 g_free(out_buf); 2643 return ret; 2644 } 2645 2646 static int make_completely_empty(BlockDriverState *bs) 2647 { 2648 BDRVQcow2State *s = bs->opaque; 2649 int ret, l1_clusters; 2650 int64_t offset; 2651 uint64_t *new_reftable = NULL; 2652 uint64_t rt_entry, l1_size2; 2653 struct { 2654 uint64_t l1_offset; 2655 uint64_t reftable_offset; 2656 uint32_t reftable_clusters; 2657 } QEMU_PACKED l1_ofs_rt_ofs_cls; 2658 2659 ret = qcow2_cache_empty(bs, s->l2_table_cache); 2660 if (ret < 0) { 2661 goto fail; 2662 } 2663 2664 ret = qcow2_cache_empty(bs, s->refcount_block_cache); 2665 if (ret < 0) { 2666 goto fail; 2667 } 2668 2669 /* Refcounts will be broken utterly */ 2670 ret = qcow2_mark_dirty(bs); 2671 if (ret < 0) { 2672 goto fail; 2673 } 2674 2675 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 2676 2677 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 2678 l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t); 2679 2680 /* After this call, neither the in-memory nor the on-disk refcount 2681 * information accurately describe the actual references */ 2682 2683 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset, 2684 l1_clusters * s->cluster_size, 0); 2685 if (ret < 0) { 2686 goto fail_broken_refcounts; 2687 } 2688 memset(s->l1_table, 0, l1_size2); 2689 2690 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); 2691 2692 /* Overwrite enough clusters at the beginning of the sectors to place 2693 * the refcount table, a refcount block and the L1 table in; this may 2694 * overwrite parts of the existing refcount and L1 table, which is not 2695 * an issue because the dirty flag is set, complete data loss is in fact 2696 * desired and partial data loss is consequently fine as well */ 2697 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size, 2698 (2 + l1_clusters) * s->cluster_size, 0); 2699 /* This call (even if it failed overall) may have overwritten on-disk 2700 * refcount structures; in that case, the in-memory refcount information 2701 * will probably differ from the on-disk information which makes the BDS 2702 * unusable */ 2703 if (ret < 0) { 2704 goto fail_broken_refcounts; 2705 } 2706 2707 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 2708 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); 2709 2710 /* "Create" an empty reftable (one cluster) directly after the image 2711 * header and an empty L1 table three clusters after the image header; 2712 * the cluster between those two will be used as the first refblock */ 2713 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size); 2714 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size); 2715 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1); 2716 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), 2717 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); 2718 if (ret < 0) { 2719 goto fail_broken_refcounts; 2720 } 2721 2722 s->l1_table_offset = 3 * s->cluster_size; 2723 2724 new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t)); 2725 if (!new_reftable) { 2726 ret = -ENOMEM; 2727 goto fail_broken_refcounts; 2728 } 2729 2730 s->refcount_table_offset = s->cluster_size; 2731 s->refcount_table_size = s->cluster_size / sizeof(uint64_t); 2732 2733 g_free(s->refcount_table); 2734 s->refcount_table = new_reftable; 2735 new_reftable = NULL; 2736 2737 /* Now the in-memory refcount information again corresponds to the on-disk 2738 * information (reftable is empty and no refblocks (the refblock cache is 2739 * empty)); however, this means some clusters (e.g. the image header) are 2740 * referenced, but not refcounted, but the normal qcow2 code assumes that 2741 * the in-memory information is always correct */ 2742 2743 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 2744 2745 /* Enter the first refblock into the reftable */ 2746 rt_entry = cpu_to_be64(2 * s->cluster_size); 2747 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, 2748 &rt_entry, sizeof(rt_entry)); 2749 if (ret < 0) { 2750 goto fail_broken_refcounts; 2751 } 2752 s->refcount_table[0] = 2 * s->cluster_size; 2753 2754 s->free_cluster_index = 0; 2755 assert(3 + l1_clusters <= s->refcount_block_size); 2756 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); 2757 if (offset < 0) { 2758 ret = offset; 2759 goto fail_broken_refcounts; 2760 } else if (offset > 0) { 2761 error_report("First cluster in emptied image is in use"); 2762 abort(); 2763 } 2764 2765 /* Now finally the in-memory information corresponds to the on-disk 2766 * structures and is correct */ 2767 ret = qcow2_mark_clean(bs); 2768 if (ret < 0) { 2769 goto fail; 2770 } 2771 2772 ret = bdrv_truncate(bs->file->bs, (3 + l1_clusters) * s->cluster_size); 2773 if (ret < 0) { 2774 goto fail; 2775 } 2776 2777 return 0; 2778 2779 fail_broken_refcounts: 2780 /* The BDS is unusable at this point. If we wanted to make it usable, we 2781 * would have to call qcow2_refcount_close(), qcow2_refcount_init(), 2782 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() 2783 * again. However, because the functions which could have caused this error 2784 * path to be taken are used by those functions as well, it's very likely 2785 * that that sequence will fail as well. Therefore, just eject the BDS. */ 2786 bs->drv = NULL; 2787 2788 fail: 2789 g_free(new_reftable); 2790 return ret; 2791 } 2792 2793 static int qcow2_make_empty(BlockDriverState *bs) 2794 { 2795 BDRVQcow2State *s = bs->opaque; 2796 uint64_t start_sector; 2797 int sector_step = INT_MAX / BDRV_SECTOR_SIZE; 2798 int l1_clusters, ret = 0; 2799 2800 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 2801 2802 if (s->qcow_version >= 3 && !s->snapshots && 2803 3 + l1_clusters <= s->refcount_block_size) { 2804 /* The following function only works for qcow2 v3 images (it requires 2805 * the dirty flag) and only as long as there are no snapshots (because 2806 * it completely empties the image). Furthermore, the L1 table and three 2807 * additional clusters (image header, refcount table, one refcount 2808 * block) have to fit inside one refcount block. */ 2809 return make_completely_empty(bs); 2810 } 2811 2812 /* This fallback code simply discards every active cluster; this is slow, 2813 * but works in all cases */ 2814 for (start_sector = 0; start_sector < bs->total_sectors; 2815 start_sector += sector_step) 2816 { 2817 /* As this function is generally used after committing an external 2818 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the 2819 * default action for this kind of discard is to pass the discard, 2820 * which will ideally result in an actually smaller image file, as 2821 * is probably desired. */ 2822 ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE, 2823 MIN(sector_step, 2824 bs->total_sectors - start_sector), 2825 QCOW2_DISCARD_SNAPSHOT, true); 2826 if (ret < 0) { 2827 break; 2828 } 2829 } 2830 2831 return ret; 2832 } 2833 2834 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) 2835 { 2836 BDRVQcow2State *s = bs->opaque; 2837 int ret; 2838 2839 qemu_co_mutex_lock(&s->lock); 2840 ret = qcow2_cache_write(bs, s->l2_table_cache); 2841 if (ret < 0) { 2842 qemu_co_mutex_unlock(&s->lock); 2843 return ret; 2844 } 2845 2846 if (qcow2_need_accurate_refcounts(s)) { 2847 ret = qcow2_cache_write(bs, s->refcount_block_cache); 2848 if (ret < 0) { 2849 qemu_co_mutex_unlock(&s->lock); 2850 return ret; 2851 } 2852 } 2853 qemu_co_mutex_unlock(&s->lock); 2854 2855 return 0; 2856 } 2857 2858 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2859 { 2860 BDRVQcow2State *s = bs->opaque; 2861 bdi->unallocated_blocks_are_zero = true; 2862 bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3); 2863 bdi->cluster_size = s->cluster_size; 2864 bdi->vm_state_offset = qcow2_vm_state_offset(s); 2865 return 0; 2866 } 2867 2868 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs) 2869 { 2870 BDRVQcow2State *s = bs->opaque; 2871 ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1); 2872 2873 *spec_info = (ImageInfoSpecific){ 2874 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2, 2875 .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1), 2876 }; 2877 if (s->qcow_version == 2) { 2878 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 2879 .compat = g_strdup("0.10"), 2880 .refcount_bits = s->refcount_bits, 2881 }; 2882 } else if (s->qcow_version == 3) { 2883 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 2884 .compat = g_strdup("1.1"), 2885 .lazy_refcounts = s->compatible_features & 2886 QCOW2_COMPAT_LAZY_REFCOUNTS, 2887 .has_lazy_refcounts = true, 2888 .corrupt = s->incompatible_features & 2889 QCOW2_INCOMPAT_CORRUPT, 2890 .has_corrupt = true, 2891 .refcount_bits = s->refcount_bits, 2892 }; 2893 } else { 2894 /* if this assertion fails, this probably means a new version was 2895 * added without having it covered here */ 2896 assert(false); 2897 } 2898 2899 return spec_info; 2900 } 2901 2902 #if 0 2903 static void dump_refcounts(BlockDriverState *bs) 2904 { 2905 BDRVQcow2State *s = bs->opaque; 2906 int64_t nb_clusters, k, k1, size; 2907 int refcount; 2908 2909 size = bdrv_getlength(bs->file->bs); 2910 nb_clusters = size_to_clusters(s, size); 2911 for(k = 0; k < nb_clusters;) { 2912 k1 = k; 2913 refcount = get_refcount(bs, k); 2914 k++; 2915 while (k < nb_clusters && get_refcount(bs, k) == refcount) 2916 k++; 2917 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount, 2918 k - k1); 2919 } 2920 } 2921 #endif 2922 2923 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 2924 int64_t pos) 2925 { 2926 BDRVQcow2State *s = bs->opaque; 2927 2928 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 2929 return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos, 2930 qiov->size, qiov, 0); 2931 } 2932 2933 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 2934 int64_t pos) 2935 { 2936 BDRVQcow2State *s = bs->opaque; 2937 2938 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 2939 return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos, 2940 qiov->size, qiov, 0); 2941 } 2942 2943 /* 2944 * Downgrades an image's version. To achieve this, any incompatible features 2945 * have to be removed. 2946 */ 2947 static int qcow2_downgrade(BlockDriverState *bs, int target_version, 2948 BlockDriverAmendStatusCB *status_cb, void *cb_opaque) 2949 { 2950 BDRVQcow2State *s = bs->opaque; 2951 int current_version = s->qcow_version; 2952 int ret; 2953 2954 if (target_version == current_version) { 2955 return 0; 2956 } else if (target_version > current_version) { 2957 return -EINVAL; 2958 } else if (target_version != 2) { 2959 return -EINVAL; 2960 } 2961 2962 if (s->refcount_order != 4) { 2963 error_report("compat=0.10 requires refcount_bits=16"); 2964 return -ENOTSUP; 2965 } 2966 2967 /* clear incompatible features */ 2968 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 2969 ret = qcow2_mark_clean(bs); 2970 if (ret < 0) { 2971 return ret; 2972 } 2973 } 2974 2975 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in 2976 * the first place; if that happens nonetheless, returning -ENOTSUP is the 2977 * best thing to do anyway */ 2978 2979 if (s->incompatible_features) { 2980 return -ENOTSUP; 2981 } 2982 2983 /* since we can ignore compatible features, we can set them to 0 as well */ 2984 s->compatible_features = 0; 2985 /* if lazy refcounts have been used, they have already been fixed through 2986 * clearing the dirty flag */ 2987 2988 /* clearing autoclear features is trivial */ 2989 s->autoclear_features = 0; 2990 2991 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque); 2992 if (ret < 0) { 2993 return ret; 2994 } 2995 2996 s->qcow_version = target_version; 2997 ret = qcow2_update_header(bs); 2998 if (ret < 0) { 2999 s->qcow_version = current_version; 3000 return ret; 3001 } 3002 return 0; 3003 } 3004 3005 typedef enum Qcow2AmendOperation { 3006 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be 3007 * statically initialized to so that the helper CB can discern the first 3008 * invocation from an operation change */ 3009 QCOW2_NO_OPERATION = 0, 3010 3011 QCOW2_CHANGING_REFCOUNT_ORDER, 3012 QCOW2_DOWNGRADING, 3013 } Qcow2AmendOperation; 3014 3015 typedef struct Qcow2AmendHelperCBInfo { 3016 /* The code coordinating the amend operations should only modify 3017 * these four fields; the rest will be managed by the CB */ 3018 BlockDriverAmendStatusCB *original_status_cb; 3019 void *original_cb_opaque; 3020 3021 Qcow2AmendOperation current_operation; 3022 3023 /* Total number of operations to perform (only set once) */ 3024 int total_operations; 3025 3026 /* The following fields are managed by the CB */ 3027 3028 /* Number of operations completed */ 3029 int operations_completed; 3030 3031 /* Cumulative offset of all completed operations */ 3032 int64_t offset_completed; 3033 3034 Qcow2AmendOperation last_operation; 3035 int64_t last_work_size; 3036 } Qcow2AmendHelperCBInfo; 3037 3038 static void qcow2_amend_helper_cb(BlockDriverState *bs, 3039 int64_t operation_offset, 3040 int64_t operation_work_size, void *opaque) 3041 { 3042 Qcow2AmendHelperCBInfo *info = opaque; 3043 int64_t current_work_size; 3044 int64_t projected_work_size; 3045 3046 if (info->current_operation != info->last_operation) { 3047 if (info->last_operation != QCOW2_NO_OPERATION) { 3048 info->offset_completed += info->last_work_size; 3049 info->operations_completed++; 3050 } 3051 3052 info->last_operation = info->current_operation; 3053 } 3054 3055 assert(info->total_operations > 0); 3056 assert(info->operations_completed < info->total_operations); 3057 3058 info->last_work_size = operation_work_size; 3059 3060 current_work_size = info->offset_completed + operation_work_size; 3061 3062 /* current_work_size is the total work size for (operations_completed + 1) 3063 * operations (which includes this one), so multiply it by the number of 3064 * operations not covered and divide it by the number of operations 3065 * covered to get a projection for the operations not covered */ 3066 projected_work_size = current_work_size * (info->total_operations - 3067 info->operations_completed - 1) 3068 / (info->operations_completed + 1); 3069 3070 info->original_status_cb(bs, info->offset_completed + operation_offset, 3071 current_work_size + projected_work_size, 3072 info->original_cb_opaque); 3073 } 3074 3075 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, 3076 BlockDriverAmendStatusCB *status_cb, 3077 void *cb_opaque) 3078 { 3079 BDRVQcow2State *s = bs->opaque; 3080 int old_version = s->qcow_version, new_version = old_version; 3081 uint64_t new_size = 0; 3082 const char *backing_file = NULL, *backing_format = NULL; 3083 bool lazy_refcounts = s->use_lazy_refcounts; 3084 const char *compat = NULL; 3085 uint64_t cluster_size = s->cluster_size; 3086 bool encrypt; 3087 int refcount_bits = s->refcount_bits; 3088 int ret; 3089 QemuOptDesc *desc = opts->list->desc; 3090 Qcow2AmendHelperCBInfo helper_cb_info; 3091 3092 while (desc && desc->name) { 3093 if (!qemu_opt_find(opts, desc->name)) { 3094 /* only change explicitly defined options */ 3095 desc++; 3096 continue; 3097 } 3098 3099 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) { 3100 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL); 3101 if (!compat) { 3102 /* preserve default */ 3103 } else if (!strcmp(compat, "0.10")) { 3104 new_version = 2; 3105 } else if (!strcmp(compat, "1.1")) { 3106 new_version = 3; 3107 } else { 3108 error_report("Unknown compatibility level %s", compat); 3109 return -EINVAL; 3110 } 3111 } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) { 3112 error_report("Cannot change preallocation mode"); 3113 return -ENOTSUP; 3114 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) { 3115 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); 3116 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) { 3117 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); 3118 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) { 3119 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); 3120 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) { 3121 encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, 3122 !!s->cipher); 3123 3124 if (encrypt != !!s->cipher) { 3125 error_report("Changing the encryption flag is not supported"); 3126 return -ENOTSUP; 3127 } 3128 } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) { 3129 cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 3130 cluster_size); 3131 if (cluster_size != s->cluster_size) { 3132 error_report("Changing the cluster size is not supported"); 3133 return -ENOTSUP; 3134 } 3135 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) { 3136 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS, 3137 lazy_refcounts); 3138 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) { 3139 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS, 3140 refcount_bits); 3141 3142 if (refcount_bits <= 0 || refcount_bits > 64 || 3143 !is_power_of_2(refcount_bits)) 3144 { 3145 error_report("Refcount width must be a power of two and may " 3146 "not exceed 64 bits"); 3147 return -EINVAL; 3148 } 3149 } else { 3150 /* if this point is reached, this probably means a new option was 3151 * added without having it covered here */ 3152 abort(); 3153 } 3154 3155 desc++; 3156 } 3157 3158 helper_cb_info = (Qcow2AmendHelperCBInfo){ 3159 .original_status_cb = status_cb, 3160 .original_cb_opaque = cb_opaque, 3161 .total_operations = (new_version < old_version) 3162 + (s->refcount_bits != refcount_bits) 3163 }; 3164 3165 /* Upgrade first (some features may require compat=1.1) */ 3166 if (new_version > old_version) { 3167 s->qcow_version = new_version; 3168 ret = qcow2_update_header(bs); 3169 if (ret < 0) { 3170 s->qcow_version = old_version; 3171 return ret; 3172 } 3173 } 3174 3175 if (s->refcount_bits != refcount_bits) { 3176 int refcount_order = ctz32(refcount_bits); 3177 Error *local_error = NULL; 3178 3179 if (new_version < 3 && refcount_bits != 16) { 3180 error_report("Different refcount widths than 16 bits require " 3181 "compatibility level 1.1 or above (use compat=1.1 or " 3182 "greater)"); 3183 return -EINVAL; 3184 } 3185 3186 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER; 3187 ret = qcow2_change_refcount_order(bs, refcount_order, 3188 &qcow2_amend_helper_cb, 3189 &helper_cb_info, &local_error); 3190 if (ret < 0) { 3191 error_report_err(local_error); 3192 return ret; 3193 } 3194 } 3195 3196 if (backing_file || backing_format) { 3197 ret = qcow2_change_backing_file(bs, 3198 backing_file ?: s->image_backing_file, 3199 backing_format ?: s->image_backing_format); 3200 if (ret < 0) { 3201 return ret; 3202 } 3203 } 3204 3205 if (s->use_lazy_refcounts != lazy_refcounts) { 3206 if (lazy_refcounts) { 3207 if (new_version < 3) { 3208 error_report("Lazy refcounts only supported with compatibility " 3209 "level 1.1 and above (use compat=1.1 or greater)"); 3210 return -EINVAL; 3211 } 3212 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 3213 ret = qcow2_update_header(bs); 3214 if (ret < 0) { 3215 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 3216 return ret; 3217 } 3218 s->use_lazy_refcounts = true; 3219 } else { 3220 /* make image clean first */ 3221 ret = qcow2_mark_clean(bs); 3222 if (ret < 0) { 3223 return ret; 3224 } 3225 /* now disallow lazy refcounts */ 3226 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 3227 ret = qcow2_update_header(bs); 3228 if (ret < 0) { 3229 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 3230 return ret; 3231 } 3232 s->use_lazy_refcounts = false; 3233 } 3234 } 3235 3236 if (new_size) { 3237 ret = bdrv_truncate(bs, new_size); 3238 if (ret < 0) { 3239 return ret; 3240 } 3241 } 3242 3243 /* Downgrade last (so unsupported features can be removed before) */ 3244 if (new_version < old_version) { 3245 helper_cb_info.current_operation = QCOW2_DOWNGRADING; 3246 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb, 3247 &helper_cb_info); 3248 if (ret < 0) { 3249 return ret; 3250 } 3251 } 3252 3253 return 0; 3254 } 3255 3256 /* 3257 * If offset or size are negative, respectively, they will not be included in 3258 * the BLOCK_IMAGE_CORRUPTED event emitted. 3259 * fatal will be ignored for read-only BDS; corruptions found there will always 3260 * be considered non-fatal. 3261 */ 3262 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 3263 int64_t size, const char *message_format, ...) 3264 { 3265 BDRVQcow2State *s = bs->opaque; 3266 const char *node_name; 3267 char *message; 3268 va_list ap; 3269 3270 fatal = fatal && !bs->read_only; 3271 3272 if (s->signaled_corruption && 3273 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) 3274 { 3275 return; 3276 } 3277 3278 va_start(ap, message_format); 3279 message = g_strdup_vprintf(message_format, ap); 3280 va_end(ap); 3281 3282 if (fatal) { 3283 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " 3284 "corruption events will be suppressed\n", message); 3285 } else { 3286 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " 3287 "corruption events will be suppressed\n", message); 3288 } 3289 3290 node_name = bdrv_get_node_name(bs); 3291 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), 3292 *node_name != '\0', node_name, 3293 message, offset >= 0, offset, 3294 size >= 0, size, 3295 fatal, &error_abort); 3296 g_free(message); 3297 3298 if (fatal) { 3299 qcow2_mark_corrupt(bs); 3300 bs->drv = NULL; /* make BDS unusable */ 3301 } 3302 3303 s->signaled_corruption = true; 3304 } 3305 3306 static QemuOptsList qcow2_create_opts = { 3307 .name = "qcow2-create-opts", 3308 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), 3309 .desc = { 3310 { 3311 .name = BLOCK_OPT_SIZE, 3312 .type = QEMU_OPT_SIZE, 3313 .help = "Virtual disk size" 3314 }, 3315 { 3316 .name = BLOCK_OPT_COMPAT_LEVEL, 3317 .type = QEMU_OPT_STRING, 3318 .help = "Compatibility level (0.10 or 1.1)" 3319 }, 3320 { 3321 .name = BLOCK_OPT_BACKING_FILE, 3322 .type = QEMU_OPT_STRING, 3323 .help = "File name of a base image" 3324 }, 3325 { 3326 .name = BLOCK_OPT_BACKING_FMT, 3327 .type = QEMU_OPT_STRING, 3328 .help = "Image format of the base image" 3329 }, 3330 { 3331 .name = BLOCK_OPT_ENCRYPT, 3332 .type = QEMU_OPT_BOOL, 3333 .help = "Encrypt the image", 3334 .def_value_str = "off" 3335 }, 3336 { 3337 .name = BLOCK_OPT_CLUSTER_SIZE, 3338 .type = QEMU_OPT_SIZE, 3339 .help = "qcow2 cluster size", 3340 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) 3341 }, 3342 { 3343 .name = BLOCK_OPT_PREALLOC, 3344 .type = QEMU_OPT_STRING, 3345 .help = "Preallocation mode (allowed values: off, metadata, " 3346 "falloc, full)" 3347 }, 3348 { 3349 .name = BLOCK_OPT_LAZY_REFCOUNTS, 3350 .type = QEMU_OPT_BOOL, 3351 .help = "Postpone refcount updates", 3352 .def_value_str = "off" 3353 }, 3354 { 3355 .name = BLOCK_OPT_REFCOUNT_BITS, 3356 .type = QEMU_OPT_NUMBER, 3357 .help = "Width of a reference count entry in bits", 3358 .def_value_str = "16" 3359 }, 3360 { /* end of list */ } 3361 } 3362 }; 3363 3364 BlockDriver bdrv_qcow2 = { 3365 .format_name = "qcow2", 3366 .instance_size = sizeof(BDRVQcow2State), 3367 .bdrv_probe = qcow2_probe, 3368 .bdrv_open = qcow2_open, 3369 .bdrv_close = qcow2_close, 3370 .bdrv_reopen_prepare = qcow2_reopen_prepare, 3371 .bdrv_reopen_commit = qcow2_reopen_commit, 3372 .bdrv_reopen_abort = qcow2_reopen_abort, 3373 .bdrv_join_options = qcow2_join_options, 3374 .bdrv_create = qcow2_create, 3375 .bdrv_has_zero_init = bdrv_has_zero_init_1, 3376 .bdrv_co_get_block_status = qcow2_co_get_block_status, 3377 .bdrv_set_key = qcow2_set_key, 3378 3379 .bdrv_co_preadv = qcow2_co_preadv, 3380 .bdrv_co_pwritev = qcow2_co_pwritev, 3381 .bdrv_co_flush_to_os = qcow2_co_flush_to_os, 3382 3383 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes, 3384 .bdrv_co_pdiscard = qcow2_co_pdiscard, 3385 .bdrv_truncate = qcow2_truncate, 3386 .bdrv_co_pwritev_compressed = qcow2_co_pwritev_compressed, 3387 .bdrv_make_empty = qcow2_make_empty, 3388 3389 .bdrv_snapshot_create = qcow2_snapshot_create, 3390 .bdrv_snapshot_goto = qcow2_snapshot_goto, 3391 .bdrv_snapshot_delete = qcow2_snapshot_delete, 3392 .bdrv_snapshot_list = qcow2_snapshot_list, 3393 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 3394 .bdrv_get_info = qcow2_get_info, 3395 .bdrv_get_specific_info = qcow2_get_specific_info, 3396 3397 .bdrv_save_vmstate = qcow2_save_vmstate, 3398 .bdrv_load_vmstate = qcow2_load_vmstate, 3399 3400 .supports_backing = true, 3401 .bdrv_change_backing_file = qcow2_change_backing_file, 3402 3403 .bdrv_refresh_limits = qcow2_refresh_limits, 3404 .bdrv_invalidate_cache = qcow2_invalidate_cache, 3405 .bdrv_inactivate = qcow2_inactivate, 3406 3407 .create_opts = &qcow2_create_opts, 3408 .bdrv_check = qcow2_check, 3409 .bdrv_amend_options = qcow2_amend_options, 3410 3411 .bdrv_detach_aio_context = qcow2_detach_aio_context, 3412 .bdrv_attach_aio_context = qcow2_attach_aio_context, 3413 }; 3414 3415 static void bdrv_qcow2_init(void) 3416 { 3417 bdrv_register(&bdrv_qcow2); 3418 } 3419 3420 block_init(bdrv_qcow2_init); 3421