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_into_cluster(s, offset) != 0) { 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 #ifndef CONFIG_LINUX 672 if (r->cache_clean_interval != 0) { 673 error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL 674 " not supported on this host"); 675 ret = -EINVAL; 676 goto fail; 677 } 678 #endif 679 if (r->cache_clean_interval > UINT_MAX) { 680 error_setg(errp, "Cache clean interval too big"); 681 ret = -EINVAL; 682 goto fail; 683 } 684 685 /* lazy-refcounts; flush if going from enabled to disabled */ 686 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, 687 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); 688 if (r->use_lazy_refcounts && s->qcow_version < 3) { 689 error_setg(errp, "Lazy refcounts require a qcow2 image with at least " 690 "qemu 1.1 compatibility level"); 691 ret = -EINVAL; 692 goto fail; 693 } 694 695 if (s->use_lazy_refcounts && !r->use_lazy_refcounts) { 696 ret = qcow2_mark_clean(bs); 697 if (ret < 0) { 698 error_setg_errno(errp, -ret, "Failed to disable lazy refcounts"); 699 goto fail; 700 } 701 } 702 703 /* Overlap check options */ 704 opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP); 705 opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE); 706 if (opt_overlap_check_template && opt_overlap_check && 707 strcmp(opt_overlap_check_template, opt_overlap_check)) 708 { 709 error_setg(errp, "Conflicting values for qcow2 options '" 710 QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE 711 "' ('%s')", opt_overlap_check, opt_overlap_check_template); 712 ret = -EINVAL; 713 goto fail; 714 } 715 if (!opt_overlap_check) { 716 opt_overlap_check = opt_overlap_check_template ?: "cached"; 717 } 718 719 if (!strcmp(opt_overlap_check, "none")) { 720 overlap_check_template = 0; 721 } else if (!strcmp(opt_overlap_check, "constant")) { 722 overlap_check_template = QCOW2_OL_CONSTANT; 723 } else if (!strcmp(opt_overlap_check, "cached")) { 724 overlap_check_template = QCOW2_OL_CACHED; 725 } else if (!strcmp(opt_overlap_check, "all")) { 726 overlap_check_template = QCOW2_OL_ALL; 727 } else { 728 error_setg(errp, "Unsupported value '%s' for qcow2 option " 729 "'overlap-check'. Allowed are any of the following: " 730 "none, constant, cached, all", opt_overlap_check); 731 ret = -EINVAL; 732 goto fail; 733 } 734 735 r->overlap_check = 0; 736 for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { 737 /* overlap-check defines a template bitmask, but every flag may be 738 * overwritten through the associated boolean option */ 739 r->overlap_check |= 740 qemu_opt_get_bool(opts, overlap_bool_option_names[i], 741 overlap_check_template & (1 << i)) << i; 742 } 743 744 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false; 745 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true; 746 r->discard_passthrough[QCOW2_DISCARD_REQUEST] = 747 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, 748 flags & BDRV_O_UNMAP); 749 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = 750 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true); 751 r->discard_passthrough[QCOW2_DISCARD_OTHER] = 752 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false); 753 754 ret = 0; 755 fail: 756 qemu_opts_del(opts); 757 opts = NULL; 758 return ret; 759 } 760 761 static void qcow2_update_options_commit(BlockDriverState *bs, 762 Qcow2ReopenState *r) 763 { 764 BDRVQcow2State *s = bs->opaque; 765 int i; 766 767 if (s->l2_table_cache) { 768 qcow2_cache_destroy(bs, s->l2_table_cache); 769 } 770 if (s->refcount_block_cache) { 771 qcow2_cache_destroy(bs, s->refcount_block_cache); 772 } 773 s->l2_table_cache = r->l2_table_cache; 774 s->refcount_block_cache = r->refcount_block_cache; 775 776 s->overlap_check = r->overlap_check; 777 s->use_lazy_refcounts = r->use_lazy_refcounts; 778 779 for (i = 0; i < QCOW2_DISCARD_MAX; i++) { 780 s->discard_passthrough[i] = r->discard_passthrough[i]; 781 } 782 783 if (s->cache_clean_interval != r->cache_clean_interval) { 784 cache_clean_timer_del(bs); 785 s->cache_clean_interval = r->cache_clean_interval; 786 cache_clean_timer_init(bs, bdrv_get_aio_context(bs)); 787 } 788 } 789 790 static void qcow2_update_options_abort(BlockDriverState *bs, 791 Qcow2ReopenState *r) 792 { 793 if (r->l2_table_cache) { 794 qcow2_cache_destroy(bs, r->l2_table_cache); 795 } 796 if (r->refcount_block_cache) { 797 qcow2_cache_destroy(bs, r->refcount_block_cache); 798 } 799 } 800 801 static int qcow2_update_options(BlockDriverState *bs, QDict *options, 802 int flags, Error **errp) 803 { 804 Qcow2ReopenState r = {}; 805 int ret; 806 807 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp); 808 if (ret >= 0) { 809 qcow2_update_options_commit(bs, &r); 810 } else { 811 qcow2_update_options_abort(bs, &r); 812 } 813 814 return ret; 815 } 816 817 static int qcow2_do_open(BlockDriverState *bs, QDict *options, int flags, 818 Error **errp) 819 { 820 BDRVQcow2State *s = bs->opaque; 821 unsigned int len, i; 822 int ret = 0; 823 QCowHeader header; 824 Error *local_err = NULL; 825 uint64_t ext_end; 826 uint64_t l1_vm_state_index; 827 828 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 829 if (ret < 0) { 830 error_setg_errno(errp, -ret, "Could not read qcow2 header"); 831 goto fail; 832 } 833 be32_to_cpus(&header.magic); 834 be32_to_cpus(&header.version); 835 be64_to_cpus(&header.backing_file_offset); 836 be32_to_cpus(&header.backing_file_size); 837 be64_to_cpus(&header.size); 838 be32_to_cpus(&header.cluster_bits); 839 be32_to_cpus(&header.crypt_method); 840 be64_to_cpus(&header.l1_table_offset); 841 be32_to_cpus(&header.l1_size); 842 be64_to_cpus(&header.refcount_table_offset); 843 be32_to_cpus(&header.refcount_table_clusters); 844 be64_to_cpus(&header.snapshots_offset); 845 be32_to_cpus(&header.nb_snapshots); 846 847 if (header.magic != QCOW_MAGIC) { 848 error_setg(errp, "Image is not in qcow2 format"); 849 ret = -EINVAL; 850 goto fail; 851 } 852 if (header.version < 2 || header.version > 3) { 853 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version); 854 ret = -ENOTSUP; 855 goto fail; 856 } 857 858 s->qcow_version = header.version; 859 860 /* Initialise cluster size */ 861 if (header.cluster_bits < MIN_CLUSTER_BITS || 862 header.cluster_bits > MAX_CLUSTER_BITS) { 863 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, 864 header.cluster_bits); 865 ret = -EINVAL; 866 goto fail; 867 } 868 869 s->cluster_bits = header.cluster_bits; 870 s->cluster_size = 1 << s->cluster_bits; 871 s->cluster_sectors = 1 << (s->cluster_bits - 9); 872 873 /* Initialise version 3 header fields */ 874 if (header.version == 2) { 875 header.incompatible_features = 0; 876 header.compatible_features = 0; 877 header.autoclear_features = 0; 878 header.refcount_order = 4; 879 header.header_length = 72; 880 } else { 881 be64_to_cpus(&header.incompatible_features); 882 be64_to_cpus(&header.compatible_features); 883 be64_to_cpus(&header.autoclear_features); 884 be32_to_cpus(&header.refcount_order); 885 be32_to_cpus(&header.header_length); 886 887 if (header.header_length < 104) { 888 error_setg(errp, "qcow2 header too short"); 889 ret = -EINVAL; 890 goto fail; 891 } 892 } 893 894 if (header.header_length > s->cluster_size) { 895 error_setg(errp, "qcow2 header exceeds cluster size"); 896 ret = -EINVAL; 897 goto fail; 898 } 899 900 if (header.header_length > sizeof(header)) { 901 s->unknown_header_fields_size = header.header_length - sizeof(header); 902 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); 903 ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields, 904 s->unknown_header_fields_size); 905 if (ret < 0) { 906 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header " 907 "fields"); 908 goto fail; 909 } 910 } 911 912 if (header.backing_file_offset > s->cluster_size) { 913 error_setg(errp, "Invalid backing file offset"); 914 ret = -EINVAL; 915 goto fail; 916 } 917 918 if (header.backing_file_offset) { 919 ext_end = header.backing_file_offset; 920 } else { 921 ext_end = 1 << header.cluster_bits; 922 } 923 924 /* Handle feature bits */ 925 s->incompatible_features = header.incompatible_features; 926 s->compatible_features = header.compatible_features; 927 s->autoclear_features = header.autoclear_features; 928 929 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) { 930 void *feature_table = NULL; 931 qcow2_read_extensions(bs, header.header_length, ext_end, 932 &feature_table, NULL); 933 report_unsupported_feature(errp, feature_table, 934 s->incompatible_features & 935 ~QCOW2_INCOMPAT_MASK); 936 ret = -ENOTSUP; 937 g_free(feature_table); 938 goto fail; 939 } 940 941 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { 942 /* Corrupt images may not be written to unless they are being repaired 943 */ 944 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { 945 error_setg(errp, "qcow2: Image is corrupt; cannot be opened " 946 "read/write"); 947 ret = -EACCES; 948 goto fail; 949 } 950 } 951 952 /* Check support for various header values */ 953 if (header.refcount_order > 6) { 954 error_setg(errp, "Reference count entry width too large; may not " 955 "exceed 64 bits"); 956 ret = -EINVAL; 957 goto fail; 958 } 959 s->refcount_order = header.refcount_order; 960 s->refcount_bits = 1 << s->refcount_order; 961 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); 962 s->refcount_max += s->refcount_max - 1; 963 964 if (header.crypt_method > QCOW_CRYPT_AES) { 965 error_setg(errp, "Unsupported encryption method: %" PRIu32, 966 header.crypt_method); 967 ret = -EINVAL; 968 goto fail; 969 } 970 if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128, 971 QCRYPTO_CIPHER_MODE_CBC)) { 972 error_setg(errp, "AES cipher not available"); 973 ret = -EINVAL; 974 goto fail; 975 } 976 s->crypt_method_header = header.crypt_method; 977 if (s->crypt_method_header) { 978 if (bdrv_uses_whitelist() && 979 s->crypt_method_header == QCOW_CRYPT_AES) { 980 error_setg(errp, 981 "Use of AES-CBC encrypted qcow2 images is no longer " 982 "supported in system emulators"); 983 error_append_hint(errp, 984 "You can use 'qemu-img convert' to convert your " 985 "image to an alternative supported format, such " 986 "as unencrypted qcow2, or raw with the LUKS " 987 "format instead.\n"); 988 ret = -ENOSYS; 989 goto fail; 990 } 991 992 bs->encrypted = true; 993 } 994 995 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ 996 s->l2_size = 1 << s->l2_bits; 997 /* 2^(s->refcount_order - 3) is the refcount width in bytes */ 998 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3); 999 s->refcount_block_size = 1 << s->refcount_block_bits; 1000 bs->total_sectors = header.size / 512; 1001 s->csize_shift = (62 - (s->cluster_bits - 8)); 1002 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; 1003 s->cluster_offset_mask = (1LL << s->csize_shift) - 1; 1004 1005 s->refcount_table_offset = header.refcount_table_offset; 1006 s->refcount_table_size = 1007 header.refcount_table_clusters << (s->cluster_bits - 3); 1008 1009 if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) { 1010 error_setg(errp, "Reference count table too large"); 1011 ret = -EINVAL; 1012 goto fail; 1013 } 1014 1015 ret = validate_table_offset(bs, s->refcount_table_offset, 1016 s->refcount_table_size, sizeof(uint64_t)); 1017 if (ret < 0) { 1018 error_setg(errp, "Invalid reference count table offset"); 1019 goto fail; 1020 } 1021 1022 /* Snapshot table offset/length */ 1023 if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) { 1024 error_setg(errp, "Too many snapshots"); 1025 ret = -EINVAL; 1026 goto fail; 1027 } 1028 1029 ret = validate_table_offset(bs, header.snapshots_offset, 1030 header.nb_snapshots, 1031 sizeof(QCowSnapshotHeader)); 1032 if (ret < 0) { 1033 error_setg(errp, "Invalid snapshot table offset"); 1034 goto fail; 1035 } 1036 1037 /* read the level 1 table */ 1038 if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) { 1039 error_setg(errp, "Active L1 table too large"); 1040 ret = -EFBIG; 1041 goto fail; 1042 } 1043 s->l1_size = header.l1_size; 1044 1045 l1_vm_state_index = size_to_l1(s, header.size); 1046 if (l1_vm_state_index > INT_MAX) { 1047 error_setg(errp, "Image is too big"); 1048 ret = -EFBIG; 1049 goto fail; 1050 } 1051 s->l1_vm_state_index = l1_vm_state_index; 1052 1053 /* the L1 table must contain at least enough entries to put 1054 header.size bytes */ 1055 if (s->l1_size < s->l1_vm_state_index) { 1056 error_setg(errp, "L1 table is too small"); 1057 ret = -EINVAL; 1058 goto fail; 1059 } 1060 1061 ret = validate_table_offset(bs, header.l1_table_offset, 1062 header.l1_size, sizeof(uint64_t)); 1063 if (ret < 0) { 1064 error_setg(errp, "Invalid L1 table offset"); 1065 goto fail; 1066 } 1067 s->l1_table_offset = header.l1_table_offset; 1068 1069 1070 if (s->l1_size > 0) { 1071 s->l1_table = qemu_try_blockalign(bs->file->bs, 1072 align_offset(s->l1_size * sizeof(uint64_t), 512)); 1073 if (s->l1_table == NULL) { 1074 error_setg(errp, "Could not allocate L1 table"); 1075 ret = -ENOMEM; 1076 goto fail; 1077 } 1078 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, 1079 s->l1_size * sizeof(uint64_t)); 1080 if (ret < 0) { 1081 error_setg_errno(errp, -ret, "Could not read L1 table"); 1082 goto fail; 1083 } 1084 for(i = 0;i < s->l1_size; i++) { 1085 be64_to_cpus(&s->l1_table[i]); 1086 } 1087 } 1088 1089 /* Parse driver-specific options */ 1090 ret = qcow2_update_options(bs, options, flags, errp); 1091 if (ret < 0) { 1092 goto fail; 1093 } 1094 1095 s->cluster_cache = g_malloc(s->cluster_size); 1096 /* one more sector for decompressed data alignment */ 1097 s->cluster_data = qemu_try_blockalign(bs->file->bs, QCOW_MAX_CRYPT_CLUSTERS 1098 * s->cluster_size + 512); 1099 if (s->cluster_data == NULL) { 1100 error_setg(errp, "Could not allocate temporary cluster buffer"); 1101 ret = -ENOMEM; 1102 goto fail; 1103 } 1104 1105 s->cluster_cache_offset = -1; 1106 s->flags = flags; 1107 1108 ret = qcow2_refcount_init(bs); 1109 if (ret != 0) { 1110 error_setg_errno(errp, -ret, "Could not initialize refcount handling"); 1111 goto fail; 1112 } 1113 1114 QLIST_INIT(&s->cluster_allocs); 1115 QTAILQ_INIT(&s->discards); 1116 1117 /* read qcow2 extensions */ 1118 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, 1119 &local_err)) { 1120 error_propagate(errp, local_err); 1121 ret = -EINVAL; 1122 goto fail; 1123 } 1124 1125 /* read the backing file name */ 1126 if (header.backing_file_offset != 0) { 1127 len = header.backing_file_size; 1128 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || 1129 len >= sizeof(bs->backing_file)) { 1130 error_setg(errp, "Backing file name too long"); 1131 ret = -EINVAL; 1132 goto fail; 1133 } 1134 ret = bdrv_pread(bs->file, header.backing_file_offset, 1135 bs->backing_file, len); 1136 if (ret < 0) { 1137 error_setg_errno(errp, -ret, "Could not read backing file name"); 1138 goto fail; 1139 } 1140 bs->backing_file[len] = '\0'; 1141 s->image_backing_file = g_strdup(bs->backing_file); 1142 } 1143 1144 /* Internal snapshots */ 1145 s->snapshots_offset = header.snapshots_offset; 1146 s->nb_snapshots = header.nb_snapshots; 1147 1148 ret = qcow2_read_snapshots(bs); 1149 if (ret < 0) { 1150 error_setg_errno(errp, -ret, "Could not read snapshots"); 1151 goto fail; 1152 } 1153 1154 /* Clear unknown autoclear feature bits */ 1155 if (!bs->read_only && !(flags & BDRV_O_INACTIVE) && s->autoclear_features) { 1156 s->autoclear_features = 0; 1157 ret = qcow2_update_header(bs); 1158 if (ret < 0) { 1159 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 1160 goto fail; 1161 } 1162 } 1163 1164 /* Initialise locks */ 1165 qemu_co_mutex_init(&s->lock); 1166 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP; 1167 1168 /* Repair image if dirty */ 1169 if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only && 1170 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { 1171 BdrvCheckResult result = {0}; 1172 1173 ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); 1174 if (ret < 0) { 1175 error_setg_errno(errp, -ret, "Could not repair dirty image"); 1176 goto fail; 1177 } 1178 } 1179 1180 #ifdef DEBUG_ALLOC 1181 { 1182 BdrvCheckResult result = {0}; 1183 qcow2_check_refcounts(bs, &result, 0); 1184 } 1185 #endif 1186 return ret; 1187 1188 fail: 1189 g_free(s->unknown_header_fields); 1190 cleanup_unknown_header_ext(bs); 1191 qcow2_free_snapshots(bs); 1192 qcow2_refcount_close(bs); 1193 qemu_vfree(s->l1_table); 1194 /* else pre-write overlap checks in cache_destroy may crash */ 1195 s->l1_table = NULL; 1196 cache_clean_timer_del(bs); 1197 if (s->l2_table_cache) { 1198 qcow2_cache_destroy(bs, s->l2_table_cache); 1199 } 1200 if (s->refcount_block_cache) { 1201 qcow2_cache_destroy(bs, s->refcount_block_cache); 1202 } 1203 g_free(s->cluster_cache); 1204 qemu_vfree(s->cluster_data); 1205 return ret; 1206 } 1207 1208 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, 1209 Error **errp) 1210 { 1211 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, 1212 false, errp); 1213 if (!bs->file) { 1214 return -EINVAL; 1215 } 1216 1217 return qcow2_do_open(bs, options, flags, errp); 1218 } 1219 1220 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) 1221 { 1222 BDRVQcow2State *s = bs->opaque; 1223 1224 if (bs->encrypted) { 1225 /* Encryption works on a sector granularity */ 1226 bs->bl.request_alignment = BDRV_SECTOR_SIZE; 1227 } 1228 bs->bl.pwrite_zeroes_alignment = s->cluster_size; 1229 bs->bl.pdiscard_alignment = s->cluster_size; 1230 } 1231 1232 static int qcow2_set_key(BlockDriverState *bs, const char *key) 1233 { 1234 BDRVQcow2State *s = bs->opaque; 1235 uint8_t keybuf[16]; 1236 int len, i; 1237 Error *err = NULL; 1238 1239 memset(keybuf, 0, 16); 1240 len = strlen(key); 1241 if (len > 16) 1242 len = 16; 1243 /* XXX: we could compress the chars to 7 bits to increase 1244 entropy */ 1245 for(i = 0;i < len;i++) { 1246 keybuf[i] = key[i]; 1247 } 1248 assert(bs->encrypted); 1249 1250 qcrypto_cipher_free(s->cipher); 1251 s->cipher = qcrypto_cipher_new( 1252 QCRYPTO_CIPHER_ALG_AES_128, 1253 QCRYPTO_CIPHER_MODE_CBC, 1254 keybuf, G_N_ELEMENTS(keybuf), 1255 &err); 1256 1257 if (!s->cipher) { 1258 /* XXX would be nice if errors in this method could 1259 * be properly propagate to the caller. Would need 1260 * the bdrv_set_key() API signature to be fixed. */ 1261 error_free(err); 1262 return -1; 1263 } 1264 return 0; 1265 } 1266 1267 static int qcow2_reopen_prepare(BDRVReopenState *state, 1268 BlockReopenQueue *queue, Error **errp) 1269 { 1270 Qcow2ReopenState *r; 1271 int ret; 1272 1273 r = g_new0(Qcow2ReopenState, 1); 1274 state->opaque = r; 1275 1276 ret = qcow2_update_options_prepare(state->bs, r, state->options, 1277 state->flags, errp); 1278 if (ret < 0) { 1279 goto fail; 1280 } 1281 1282 /* We need to write out any unwritten data if we reopen read-only. */ 1283 if ((state->flags & BDRV_O_RDWR) == 0) { 1284 ret = bdrv_flush(state->bs); 1285 if (ret < 0) { 1286 goto fail; 1287 } 1288 1289 ret = qcow2_mark_clean(state->bs); 1290 if (ret < 0) { 1291 goto fail; 1292 } 1293 } 1294 1295 return 0; 1296 1297 fail: 1298 qcow2_update_options_abort(state->bs, r); 1299 g_free(r); 1300 return ret; 1301 } 1302 1303 static void qcow2_reopen_commit(BDRVReopenState *state) 1304 { 1305 qcow2_update_options_commit(state->bs, state->opaque); 1306 g_free(state->opaque); 1307 } 1308 1309 static void qcow2_reopen_abort(BDRVReopenState *state) 1310 { 1311 qcow2_update_options_abort(state->bs, state->opaque); 1312 g_free(state->opaque); 1313 } 1314 1315 static void qcow2_join_options(QDict *options, QDict *old_options) 1316 { 1317 bool has_new_overlap_template = 1318 qdict_haskey(options, QCOW2_OPT_OVERLAP) || 1319 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE); 1320 bool has_new_total_cache_size = 1321 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE); 1322 bool has_all_cache_options; 1323 1324 /* New overlap template overrides all old overlap options */ 1325 if (has_new_overlap_template) { 1326 qdict_del(old_options, QCOW2_OPT_OVERLAP); 1327 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE); 1328 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER); 1329 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1); 1330 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2); 1331 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE); 1332 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK); 1333 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE); 1334 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1); 1335 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2); 1336 } 1337 1338 /* New total cache size overrides all old options */ 1339 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) { 1340 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE); 1341 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1342 } 1343 1344 qdict_join(options, old_options, false); 1345 1346 /* 1347 * If after merging all cache size options are set, an old total size is 1348 * overwritten. Do keep all options, however, if all three are new. The 1349 * resulting error message is what we want to happen. 1350 */ 1351 has_all_cache_options = 1352 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) || 1353 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) || 1354 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); 1355 1356 if (has_all_cache_options && !has_new_total_cache_size) { 1357 qdict_del(options, QCOW2_OPT_CACHE_SIZE); 1358 } 1359 } 1360 1361 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs, 1362 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file) 1363 { 1364 BDRVQcow2State *s = bs->opaque; 1365 uint64_t cluster_offset; 1366 int index_in_cluster, ret; 1367 unsigned int bytes; 1368 int64_t status = 0; 1369 1370 bytes = MIN(INT_MAX, nb_sectors * BDRV_SECTOR_SIZE); 1371 qemu_co_mutex_lock(&s->lock); 1372 ret = qcow2_get_cluster_offset(bs, sector_num << 9, &bytes, 1373 &cluster_offset); 1374 qemu_co_mutex_unlock(&s->lock); 1375 if (ret < 0) { 1376 return ret; 1377 } 1378 1379 *pnum = bytes >> BDRV_SECTOR_BITS; 1380 1381 if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED && 1382 !s->cipher) { 1383 index_in_cluster = sector_num & (s->cluster_sectors - 1); 1384 cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS); 1385 *file = bs->file->bs; 1386 status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset; 1387 } 1388 if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) { 1389 status |= BDRV_BLOCK_ZERO; 1390 } else if (ret != QCOW2_CLUSTER_UNALLOCATED) { 1391 status |= BDRV_BLOCK_DATA; 1392 } 1393 return status; 1394 } 1395 1396 /* handle reading after the end of the backing file */ 1397 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov, 1398 int64_t offset, int bytes) 1399 { 1400 uint64_t bs_size = bs->total_sectors * BDRV_SECTOR_SIZE; 1401 int n1; 1402 1403 if ((offset + bytes) <= bs_size) { 1404 return bytes; 1405 } 1406 1407 if (offset >= bs_size) { 1408 n1 = 0; 1409 } else { 1410 n1 = bs_size - offset; 1411 } 1412 1413 qemu_iovec_memset(qiov, n1, 0, bytes - n1); 1414 1415 return n1; 1416 } 1417 1418 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset, 1419 uint64_t bytes, QEMUIOVector *qiov, 1420 int flags) 1421 { 1422 BDRVQcow2State *s = bs->opaque; 1423 int offset_in_cluster, n1; 1424 int ret; 1425 unsigned int cur_bytes; /* number of bytes in current iteration */ 1426 uint64_t cluster_offset = 0; 1427 uint64_t bytes_done = 0; 1428 QEMUIOVector hd_qiov; 1429 uint8_t *cluster_data = NULL; 1430 1431 qemu_iovec_init(&hd_qiov, qiov->niov); 1432 1433 qemu_co_mutex_lock(&s->lock); 1434 1435 while (bytes != 0) { 1436 1437 /* prepare next request */ 1438 cur_bytes = MIN(bytes, INT_MAX); 1439 if (s->cipher) { 1440 cur_bytes = MIN(cur_bytes, 1441 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1442 } 1443 1444 ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset); 1445 if (ret < 0) { 1446 goto fail; 1447 } 1448 1449 offset_in_cluster = offset_into_cluster(s, offset); 1450 1451 qemu_iovec_reset(&hd_qiov); 1452 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 1453 1454 switch (ret) { 1455 case QCOW2_CLUSTER_UNALLOCATED: 1456 1457 if (bs->backing) { 1458 /* read from the base image */ 1459 n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov, 1460 offset, cur_bytes); 1461 if (n1 > 0) { 1462 QEMUIOVector local_qiov; 1463 1464 qemu_iovec_init(&local_qiov, hd_qiov.niov); 1465 qemu_iovec_concat(&local_qiov, &hd_qiov, 0, n1); 1466 1467 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); 1468 qemu_co_mutex_unlock(&s->lock); 1469 ret = bdrv_co_preadv(bs->backing, offset, n1, 1470 &local_qiov, 0); 1471 qemu_co_mutex_lock(&s->lock); 1472 1473 qemu_iovec_destroy(&local_qiov); 1474 1475 if (ret < 0) { 1476 goto fail; 1477 } 1478 } 1479 } else { 1480 /* Note: in this case, no need to wait */ 1481 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 1482 } 1483 break; 1484 1485 case QCOW2_CLUSTER_ZERO_PLAIN: 1486 case QCOW2_CLUSTER_ZERO_ALLOC: 1487 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes); 1488 break; 1489 1490 case QCOW2_CLUSTER_COMPRESSED: 1491 /* add AIO support for compressed blocks ? */ 1492 ret = qcow2_decompress_cluster(bs, cluster_offset); 1493 if (ret < 0) { 1494 goto fail; 1495 } 1496 1497 qemu_iovec_from_buf(&hd_qiov, 0, 1498 s->cluster_cache + offset_in_cluster, 1499 cur_bytes); 1500 break; 1501 1502 case QCOW2_CLUSTER_NORMAL: 1503 if ((cluster_offset & 511) != 0) { 1504 ret = -EIO; 1505 goto fail; 1506 } 1507 1508 if (bs->encrypted) { 1509 assert(s->cipher); 1510 1511 /* 1512 * For encrypted images, read everything into a temporary 1513 * contiguous buffer on which the AES functions can work. 1514 */ 1515 if (!cluster_data) { 1516 cluster_data = 1517 qemu_try_blockalign(bs->file->bs, 1518 QCOW_MAX_CRYPT_CLUSTERS 1519 * s->cluster_size); 1520 if (cluster_data == NULL) { 1521 ret = -ENOMEM; 1522 goto fail; 1523 } 1524 } 1525 1526 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1527 qemu_iovec_reset(&hd_qiov); 1528 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 1529 } 1530 1531 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); 1532 qemu_co_mutex_unlock(&s->lock); 1533 ret = bdrv_co_preadv(bs->file, 1534 cluster_offset + offset_in_cluster, 1535 cur_bytes, &hd_qiov, 0); 1536 qemu_co_mutex_lock(&s->lock); 1537 if (ret < 0) { 1538 goto fail; 1539 } 1540 if (bs->encrypted) { 1541 assert(s->cipher); 1542 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); 1543 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0); 1544 Error *err = NULL; 1545 if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS, 1546 cluster_data, cluster_data, 1547 cur_bytes >> BDRV_SECTOR_BITS, 1548 false, &err) < 0) { 1549 error_free(err); 1550 ret = -EIO; 1551 goto fail; 1552 } 1553 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes); 1554 } 1555 break; 1556 1557 default: 1558 g_assert_not_reached(); 1559 ret = -EIO; 1560 goto fail; 1561 } 1562 1563 bytes -= cur_bytes; 1564 offset += cur_bytes; 1565 bytes_done += cur_bytes; 1566 } 1567 ret = 0; 1568 1569 fail: 1570 qemu_co_mutex_unlock(&s->lock); 1571 1572 qemu_iovec_destroy(&hd_qiov); 1573 qemu_vfree(cluster_data); 1574 1575 return ret; 1576 } 1577 1578 /* Check if it's possible to merge a write request with the writing of 1579 * the data from the COW regions */ 1580 static bool merge_cow(uint64_t offset, unsigned bytes, 1581 QEMUIOVector *hd_qiov, QCowL2Meta *l2meta) 1582 { 1583 QCowL2Meta *m; 1584 1585 for (m = l2meta; m != NULL; m = m->next) { 1586 /* If both COW regions are empty then there's nothing to merge */ 1587 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) { 1588 continue; 1589 } 1590 1591 /* The data (middle) region must be immediately after the 1592 * start region */ 1593 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) { 1594 continue; 1595 } 1596 1597 /* The end region must be immediately after the data (middle) 1598 * region */ 1599 if (m->offset + m->cow_end.offset != offset + bytes) { 1600 continue; 1601 } 1602 1603 /* Make sure that adding both COW regions to the QEMUIOVector 1604 * does not exceed IOV_MAX */ 1605 if (hd_qiov->niov > IOV_MAX - 2) { 1606 continue; 1607 } 1608 1609 m->data_qiov = hd_qiov; 1610 return true; 1611 } 1612 1613 return false; 1614 } 1615 1616 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset, 1617 uint64_t bytes, QEMUIOVector *qiov, 1618 int flags) 1619 { 1620 BDRVQcow2State *s = bs->opaque; 1621 int offset_in_cluster; 1622 int ret; 1623 unsigned int cur_bytes; /* number of sectors in current iteration */ 1624 uint64_t cluster_offset; 1625 QEMUIOVector hd_qiov; 1626 uint64_t bytes_done = 0; 1627 uint8_t *cluster_data = NULL; 1628 QCowL2Meta *l2meta = NULL; 1629 1630 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes); 1631 1632 qemu_iovec_init(&hd_qiov, qiov->niov); 1633 1634 s->cluster_cache_offset = -1; /* disable compressed cache */ 1635 1636 qemu_co_mutex_lock(&s->lock); 1637 1638 while (bytes != 0) { 1639 1640 l2meta = NULL; 1641 1642 trace_qcow2_writev_start_part(qemu_coroutine_self()); 1643 offset_in_cluster = offset_into_cluster(s, offset); 1644 cur_bytes = MIN(bytes, INT_MAX); 1645 if (bs->encrypted) { 1646 cur_bytes = MIN(cur_bytes, 1647 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size 1648 - offset_in_cluster); 1649 } 1650 1651 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 1652 &cluster_offset, &l2meta); 1653 if (ret < 0) { 1654 goto fail; 1655 } 1656 1657 assert((cluster_offset & 511) == 0); 1658 1659 qemu_iovec_reset(&hd_qiov); 1660 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes); 1661 1662 if (bs->encrypted) { 1663 Error *err = NULL; 1664 assert(s->cipher); 1665 if (!cluster_data) { 1666 cluster_data = qemu_try_blockalign(bs->file->bs, 1667 QCOW_MAX_CRYPT_CLUSTERS 1668 * s->cluster_size); 1669 if (cluster_data == NULL) { 1670 ret = -ENOMEM; 1671 goto fail; 1672 } 1673 } 1674 1675 assert(hd_qiov.size <= 1676 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); 1677 qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size); 1678 1679 if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS, 1680 cluster_data, cluster_data, 1681 cur_bytes >>BDRV_SECTOR_BITS, 1682 true, &err) < 0) { 1683 error_free(err); 1684 ret = -EIO; 1685 goto fail; 1686 } 1687 1688 qemu_iovec_reset(&hd_qiov); 1689 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes); 1690 } 1691 1692 ret = qcow2_pre_write_overlap_check(bs, 0, 1693 cluster_offset + offset_in_cluster, cur_bytes); 1694 if (ret < 0) { 1695 goto fail; 1696 } 1697 1698 /* If we need to do COW, check if it's possible to merge the 1699 * writing of the guest data together with that of the COW regions. 1700 * If it's not possible (or not necessary) then write the 1701 * guest data now. */ 1702 if (!merge_cow(offset, cur_bytes, &hd_qiov, l2meta)) { 1703 qemu_co_mutex_unlock(&s->lock); 1704 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); 1705 trace_qcow2_writev_data(qemu_coroutine_self(), 1706 cluster_offset + offset_in_cluster); 1707 ret = bdrv_co_pwritev(bs->file, 1708 cluster_offset + offset_in_cluster, 1709 cur_bytes, &hd_qiov, 0); 1710 qemu_co_mutex_lock(&s->lock); 1711 if (ret < 0) { 1712 goto fail; 1713 } 1714 } 1715 1716 while (l2meta != NULL) { 1717 QCowL2Meta *next; 1718 1719 ret = qcow2_alloc_cluster_link_l2(bs, l2meta); 1720 if (ret < 0) { 1721 goto fail; 1722 } 1723 1724 /* Take the request off the list of running requests */ 1725 if (l2meta->nb_clusters != 0) { 1726 QLIST_REMOVE(l2meta, next_in_flight); 1727 } 1728 1729 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1730 1731 next = l2meta->next; 1732 g_free(l2meta); 1733 l2meta = next; 1734 } 1735 1736 bytes -= cur_bytes; 1737 offset += cur_bytes; 1738 bytes_done += cur_bytes; 1739 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes); 1740 } 1741 ret = 0; 1742 1743 fail: 1744 qemu_co_mutex_unlock(&s->lock); 1745 1746 while (l2meta != NULL) { 1747 QCowL2Meta *next; 1748 1749 if (l2meta->nb_clusters != 0) { 1750 QLIST_REMOVE(l2meta, next_in_flight); 1751 } 1752 qemu_co_queue_restart_all(&l2meta->dependent_requests); 1753 1754 next = l2meta->next; 1755 g_free(l2meta); 1756 l2meta = next; 1757 } 1758 1759 qemu_iovec_destroy(&hd_qiov); 1760 qemu_vfree(cluster_data); 1761 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); 1762 1763 return ret; 1764 } 1765 1766 static int qcow2_inactivate(BlockDriverState *bs) 1767 { 1768 BDRVQcow2State *s = bs->opaque; 1769 int ret, result = 0; 1770 1771 ret = qcow2_cache_flush(bs, s->l2_table_cache); 1772 if (ret) { 1773 result = ret; 1774 error_report("Failed to flush the L2 table cache: %s", 1775 strerror(-ret)); 1776 } 1777 1778 ret = qcow2_cache_flush(bs, s->refcount_block_cache); 1779 if (ret) { 1780 result = ret; 1781 error_report("Failed to flush the refcount block cache: %s", 1782 strerror(-ret)); 1783 } 1784 1785 if (result == 0) { 1786 qcow2_mark_clean(bs); 1787 } 1788 1789 return result; 1790 } 1791 1792 static void qcow2_close(BlockDriverState *bs) 1793 { 1794 BDRVQcow2State *s = bs->opaque; 1795 qemu_vfree(s->l1_table); 1796 /* else pre-write overlap checks in cache_destroy may crash */ 1797 s->l1_table = NULL; 1798 1799 if (!(s->flags & BDRV_O_INACTIVE)) { 1800 qcow2_inactivate(bs); 1801 } 1802 1803 cache_clean_timer_del(bs); 1804 qcow2_cache_destroy(bs, s->l2_table_cache); 1805 qcow2_cache_destroy(bs, s->refcount_block_cache); 1806 1807 qcrypto_cipher_free(s->cipher); 1808 s->cipher = NULL; 1809 1810 g_free(s->unknown_header_fields); 1811 cleanup_unknown_header_ext(bs); 1812 1813 g_free(s->image_backing_file); 1814 g_free(s->image_backing_format); 1815 1816 g_free(s->cluster_cache); 1817 qemu_vfree(s->cluster_data); 1818 qcow2_refcount_close(bs); 1819 qcow2_free_snapshots(bs); 1820 } 1821 1822 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp) 1823 { 1824 BDRVQcow2State *s = bs->opaque; 1825 int flags = s->flags; 1826 QCryptoCipher *cipher = NULL; 1827 QDict *options; 1828 Error *local_err = NULL; 1829 int ret; 1830 1831 /* 1832 * Backing files are read-only which makes all of their metadata immutable, 1833 * that means we don't have to worry about reopening them here. 1834 */ 1835 1836 cipher = s->cipher; 1837 s->cipher = NULL; 1838 1839 qcow2_close(bs); 1840 1841 memset(s, 0, sizeof(BDRVQcow2State)); 1842 options = qdict_clone_shallow(bs->options); 1843 1844 flags &= ~BDRV_O_INACTIVE; 1845 ret = qcow2_do_open(bs, options, flags, &local_err); 1846 QDECREF(options); 1847 if (local_err) { 1848 error_propagate(errp, local_err); 1849 error_prepend(errp, "Could not reopen qcow2 layer: "); 1850 bs->drv = NULL; 1851 return; 1852 } else if (ret < 0) { 1853 error_setg_errno(errp, -ret, "Could not reopen qcow2 layer"); 1854 bs->drv = NULL; 1855 return; 1856 } 1857 1858 s->cipher = cipher; 1859 } 1860 1861 static size_t header_ext_add(char *buf, uint32_t magic, const void *s, 1862 size_t len, size_t buflen) 1863 { 1864 QCowExtension *ext_backing_fmt = (QCowExtension*) buf; 1865 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); 1866 1867 if (buflen < ext_len) { 1868 return -ENOSPC; 1869 } 1870 1871 *ext_backing_fmt = (QCowExtension) { 1872 .magic = cpu_to_be32(magic), 1873 .len = cpu_to_be32(len), 1874 }; 1875 1876 if (len) { 1877 memcpy(buf + sizeof(QCowExtension), s, len); 1878 } 1879 1880 return ext_len; 1881 } 1882 1883 /* 1884 * Updates the qcow2 header, including the variable length parts of it, i.e. 1885 * the backing file name and all extensions. qcow2 was not designed to allow 1886 * such changes, so if we run out of space (we can only use the first cluster) 1887 * this function may fail. 1888 * 1889 * Returns 0 on success, -errno in error cases. 1890 */ 1891 int qcow2_update_header(BlockDriverState *bs) 1892 { 1893 BDRVQcow2State *s = bs->opaque; 1894 QCowHeader *header; 1895 char *buf; 1896 size_t buflen = s->cluster_size; 1897 int ret; 1898 uint64_t total_size; 1899 uint32_t refcount_table_clusters; 1900 size_t header_length; 1901 Qcow2UnknownHeaderExtension *uext; 1902 1903 buf = qemu_blockalign(bs, buflen); 1904 1905 /* Header structure */ 1906 header = (QCowHeader*) buf; 1907 1908 if (buflen < sizeof(*header)) { 1909 ret = -ENOSPC; 1910 goto fail; 1911 } 1912 1913 header_length = sizeof(*header) + s->unknown_header_fields_size; 1914 total_size = bs->total_sectors * BDRV_SECTOR_SIZE; 1915 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); 1916 1917 *header = (QCowHeader) { 1918 /* Version 2 fields */ 1919 .magic = cpu_to_be32(QCOW_MAGIC), 1920 .version = cpu_to_be32(s->qcow_version), 1921 .backing_file_offset = 0, 1922 .backing_file_size = 0, 1923 .cluster_bits = cpu_to_be32(s->cluster_bits), 1924 .size = cpu_to_be64(total_size), 1925 .crypt_method = cpu_to_be32(s->crypt_method_header), 1926 .l1_size = cpu_to_be32(s->l1_size), 1927 .l1_table_offset = cpu_to_be64(s->l1_table_offset), 1928 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), 1929 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), 1930 .nb_snapshots = cpu_to_be32(s->nb_snapshots), 1931 .snapshots_offset = cpu_to_be64(s->snapshots_offset), 1932 1933 /* Version 3 fields */ 1934 .incompatible_features = cpu_to_be64(s->incompatible_features), 1935 .compatible_features = cpu_to_be64(s->compatible_features), 1936 .autoclear_features = cpu_to_be64(s->autoclear_features), 1937 .refcount_order = cpu_to_be32(s->refcount_order), 1938 .header_length = cpu_to_be32(header_length), 1939 }; 1940 1941 /* For older versions, write a shorter header */ 1942 switch (s->qcow_version) { 1943 case 2: 1944 ret = offsetof(QCowHeader, incompatible_features); 1945 break; 1946 case 3: 1947 ret = sizeof(*header); 1948 break; 1949 default: 1950 ret = -EINVAL; 1951 goto fail; 1952 } 1953 1954 buf += ret; 1955 buflen -= ret; 1956 memset(buf, 0, buflen); 1957 1958 /* Preserve any unknown field in the header */ 1959 if (s->unknown_header_fields_size) { 1960 if (buflen < s->unknown_header_fields_size) { 1961 ret = -ENOSPC; 1962 goto fail; 1963 } 1964 1965 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); 1966 buf += s->unknown_header_fields_size; 1967 buflen -= s->unknown_header_fields_size; 1968 } 1969 1970 /* Backing file format header extension */ 1971 if (s->image_backing_format) { 1972 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, 1973 s->image_backing_format, 1974 strlen(s->image_backing_format), 1975 buflen); 1976 if (ret < 0) { 1977 goto fail; 1978 } 1979 1980 buf += ret; 1981 buflen -= ret; 1982 } 1983 1984 /* Feature table */ 1985 if (s->qcow_version >= 3) { 1986 Qcow2Feature features[] = { 1987 { 1988 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 1989 .bit = QCOW2_INCOMPAT_DIRTY_BITNR, 1990 .name = "dirty bit", 1991 }, 1992 { 1993 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, 1994 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, 1995 .name = "corrupt bit", 1996 }, 1997 { 1998 .type = QCOW2_FEAT_TYPE_COMPATIBLE, 1999 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, 2000 .name = "lazy refcounts", 2001 }, 2002 }; 2003 2004 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, 2005 features, sizeof(features), buflen); 2006 if (ret < 0) { 2007 goto fail; 2008 } 2009 buf += ret; 2010 buflen -= ret; 2011 } 2012 2013 /* Keep unknown header extensions */ 2014 QLIST_FOREACH(uext, &s->unknown_header_ext, next) { 2015 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); 2016 if (ret < 0) { 2017 goto fail; 2018 } 2019 2020 buf += ret; 2021 buflen -= ret; 2022 } 2023 2024 /* End of header extensions */ 2025 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); 2026 if (ret < 0) { 2027 goto fail; 2028 } 2029 2030 buf += ret; 2031 buflen -= ret; 2032 2033 /* Backing file name */ 2034 if (s->image_backing_file) { 2035 size_t backing_file_len = strlen(s->image_backing_file); 2036 2037 if (buflen < backing_file_len) { 2038 ret = -ENOSPC; 2039 goto fail; 2040 } 2041 2042 /* Using strncpy is ok here, since buf is not NUL-terminated. */ 2043 strncpy(buf, s->image_backing_file, buflen); 2044 2045 header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); 2046 header->backing_file_size = cpu_to_be32(backing_file_len); 2047 } 2048 2049 /* Write the new header */ 2050 ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size); 2051 if (ret < 0) { 2052 goto fail; 2053 } 2054 2055 ret = 0; 2056 fail: 2057 qemu_vfree(header); 2058 return ret; 2059 } 2060 2061 static int qcow2_change_backing_file(BlockDriverState *bs, 2062 const char *backing_file, const char *backing_fmt) 2063 { 2064 BDRVQcow2State *s = bs->opaque; 2065 2066 if (backing_file && strlen(backing_file) > 1023) { 2067 return -EINVAL; 2068 } 2069 2070 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); 2071 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); 2072 2073 g_free(s->image_backing_file); 2074 g_free(s->image_backing_format); 2075 2076 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL; 2077 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL; 2078 2079 return qcow2_update_header(bs); 2080 } 2081 2082 static int preallocate(BlockDriverState *bs) 2083 { 2084 uint64_t bytes; 2085 uint64_t offset; 2086 uint64_t host_offset = 0; 2087 unsigned int cur_bytes; 2088 int ret; 2089 QCowL2Meta *meta; 2090 2091 bytes = bdrv_getlength(bs); 2092 offset = 0; 2093 2094 while (bytes) { 2095 cur_bytes = MIN(bytes, INT_MAX); 2096 ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes, 2097 &host_offset, &meta); 2098 if (ret < 0) { 2099 return ret; 2100 } 2101 2102 while (meta) { 2103 QCowL2Meta *next = meta->next; 2104 2105 ret = qcow2_alloc_cluster_link_l2(bs, meta); 2106 if (ret < 0) { 2107 qcow2_free_any_clusters(bs, meta->alloc_offset, 2108 meta->nb_clusters, QCOW2_DISCARD_NEVER); 2109 return ret; 2110 } 2111 2112 /* There are no dependent requests, but we need to remove our 2113 * request from the list of in-flight requests */ 2114 QLIST_REMOVE(meta, next_in_flight); 2115 2116 g_free(meta); 2117 meta = next; 2118 } 2119 2120 /* TODO Preallocate data if requested */ 2121 2122 bytes -= cur_bytes; 2123 offset += cur_bytes; 2124 } 2125 2126 /* 2127 * It is expected that the image file is large enough to actually contain 2128 * all of the allocated clusters (otherwise we get failing reads after 2129 * EOF). Extend the image to the last allocated sector. 2130 */ 2131 if (host_offset != 0) { 2132 uint8_t data = 0; 2133 ret = bdrv_pwrite(bs->file, (host_offset + cur_bytes) - 1, 2134 &data, 1); 2135 if (ret < 0) { 2136 return ret; 2137 } 2138 } 2139 2140 return 0; 2141 } 2142 2143 static int qcow2_create2(const char *filename, int64_t total_size, 2144 const char *backing_file, const char *backing_format, 2145 int flags, size_t cluster_size, PreallocMode prealloc, 2146 QemuOpts *opts, int version, int refcount_order, 2147 Error **errp) 2148 { 2149 int cluster_bits; 2150 QDict *options; 2151 2152 /* Calculate cluster_bits */ 2153 cluster_bits = ctz32(cluster_size); 2154 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || 2155 (1 << cluster_bits) != cluster_size) 2156 { 2157 error_setg(errp, "Cluster size must be a power of two between %d and " 2158 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); 2159 return -EINVAL; 2160 } 2161 2162 /* 2163 * Open the image file and write a minimal qcow2 header. 2164 * 2165 * We keep things simple and start with a zero-sized image. We also 2166 * do without refcount blocks or a L1 table for now. We'll fix the 2167 * inconsistency later. 2168 * 2169 * We do need a refcount table because growing the refcount table means 2170 * allocating two new refcount blocks - the seconds of which would be at 2171 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file 2172 * size for any qcow2 image. 2173 */ 2174 BlockBackend *blk; 2175 QCowHeader *header; 2176 uint64_t* refcount_table; 2177 Error *local_err = NULL; 2178 int ret; 2179 2180 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { 2181 /* Note: The following calculation does not need to be exact; if it is a 2182 * bit off, either some bytes will be "leaked" (which is fine) or we 2183 * will need to increase the file size by some bytes (which is fine, 2184 * too, as long as the bulk is allocated here). Therefore, using 2185 * floating point arithmetic is fine. */ 2186 int64_t meta_size = 0; 2187 uint64_t nreftablee, nrefblocke, nl1e, nl2e, refblock_count; 2188 int64_t aligned_total_size = align_offset(total_size, cluster_size); 2189 int refblock_bits, refblock_size; 2190 /* refcount entry size in bytes */ 2191 double rces = (1 << refcount_order) / 8.; 2192 2193 /* see qcow2_open() */ 2194 refblock_bits = cluster_bits - (refcount_order - 3); 2195 refblock_size = 1 << refblock_bits; 2196 2197 /* header: 1 cluster */ 2198 meta_size += cluster_size; 2199 2200 /* total size of L2 tables */ 2201 nl2e = aligned_total_size / cluster_size; 2202 nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t)); 2203 meta_size += nl2e * sizeof(uint64_t); 2204 2205 /* total size of L1 tables */ 2206 nl1e = nl2e * sizeof(uint64_t) / cluster_size; 2207 nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t)); 2208 meta_size += nl1e * sizeof(uint64_t); 2209 2210 /* total size of refcount blocks 2211 * 2212 * note: every host cluster is reference-counted, including metadata 2213 * (even refcount blocks are recursively included). 2214 * Let: 2215 * a = total_size (this is the guest disk size) 2216 * m = meta size not including refcount blocks and refcount tables 2217 * c = cluster size 2218 * y1 = number of refcount blocks entries 2219 * y2 = meta size including everything 2220 * rces = refcount entry size in bytes 2221 * then, 2222 * y1 = (y2 + a)/c 2223 * y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m 2224 * we can get y1: 2225 * y1 = (a + m) / (c - rces - rces * sizeof(u64) / c) 2226 */ 2227 nrefblocke = (aligned_total_size + meta_size + cluster_size) 2228 / (cluster_size - rces - rces * sizeof(uint64_t) 2229 / cluster_size); 2230 refblock_count = DIV_ROUND_UP(nrefblocke, refblock_size); 2231 meta_size += refblock_count * cluster_size; 2232 2233 /* total size of refcount tables */ 2234 nreftablee = align_offset(refblock_count, 2235 cluster_size / sizeof(uint64_t)); 2236 meta_size += nreftablee * sizeof(uint64_t); 2237 2238 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, 2239 aligned_total_size + meta_size, &error_abort); 2240 qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc], 2241 &error_abort); 2242 } 2243 2244 ret = bdrv_create_file(filename, opts, &local_err); 2245 if (ret < 0) { 2246 error_propagate(errp, local_err); 2247 return ret; 2248 } 2249 2250 blk = blk_new_open(filename, NULL, NULL, 2251 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, 2252 &local_err); 2253 if (blk == NULL) { 2254 error_propagate(errp, local_err); 2255 return -EIO; 2256 } 2257 2258 blk_set_allow_write_beyond_eof(blk, true); 2259 2260 /* Write the header */ 2261 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); 2262 header = g_malloc0(cluster_size); 2263 *header = (QCowHeader) { 2264 .magic = cpu_to_be32(QCOW_MAGIC), 2265 .version = cpu_to_be32(version), 2266 .cluster_bits = cpu_to_be32(cluster_bits), 2267 .size = cpu_to_be64(0), 2268 .l1_table_offset = cpu_to_be64(0), 2269 .l1_size = cpu_to_be32(0), 2270 .refcount_table_offset = cpu_to_be64(cluster_size), 2271 .refcount_table_clusters = cpu_to_be32(1), 2272 .refcount_order = cpu_to_be32(refcount_order), 2273 .header_length = cpu_to_be32(sizeof(*header)), 2274 }; 2275 2276 if (flags & BLOCK_FLAG_ENCRYPT) { 2277 header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES); 2278 } else { 2279 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); 2280 } 2281 2282 if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) { 2283 header->compatible_features |= 2284 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); 2285 } 2286 2287 ret = blk_pwrite(blk, 0, header, cluster_size, 0); 2288 g_free(header); 2289 if (ret < 0) { 2290 error_setg_errno(errp, -ret, "Could not write qcow2 header"); 2291 goto out; 2292 } 2293 2294 /* Write a refcount table with one refcount block */ 2295 refcount_table = g_malloc0(2 * cluster_size); 2296 refcount_table[0] = cpu_to_be64(2 * cluster_size); 2297 ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0); 2298 g_free(refcount_table); 2299 2300 if (ret < 0) { 2301 error_setg_errno(errp, -ret, "Could not write refcount table"); 2302 goto out; 2303 } 2304 2305 blk_unref(blk); 2306 blk = NULL; 2307 2308 /* 2309 * And now open the image and make it consistent first (i.e. increase the 2310 * refcount of the cluster that is occupied by the header and the refcount 2311 * table) 2312 */ 2313 options = qdict_new(); 2314 qdict_put_str(options, "driver", "qcow2"); 2315 blk = blk_new_open(filename, NULL, options, 2316 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH, 2317 &local_err); 2318 if (blk == NULL) { 2319 error_propagate(errp, local_err); 2320 ret = -EIO; 2321 goto out; 2322 } 2323 2324 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size); 2325 if (ret < 0) { 2326 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " 2327 "header and refcount table"); 2328 goto out; 2329 2330 } else if (ret != 0) { 2331 error_report("Huh, first cluster in empty image is already in use?"); 2332 abort(); 2333 } 2334 2335 /* Create a full header (including things like feature table) */ 2336 ret = qcow2_update_header(blk_bs(blk)); 2337 if (ret < 0) { 2338 error_setg_errno(errp, -ret, "Could not update qcow2 header"); 2339 goto out; 2340 } 2341 2342 /* Okay, now that we have a valid image, let's give it the right size */ 2343 ret = blk_truncate(blk, total_size, errp); 2344 if (ret < 0) { 2345 error_prepend(errp, "Could not resize image: "); 2346 goto out; 2347 } 2348 2349 /* Want a backing file? There you go.*/ 2350 if (backing_file) { 2351 ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format); 2352 if (ret < 0) { 2353 error_setg_errno(errp, -ret, "Could not assign backing file '%s' " 2354 "with format '%s'", backing_file, backing_format); 2355 goto out; 2356 } 2357 } 2358 2359 /* And if we're supposed to preallocate metadata, do that now */ 2360 if (prealloc != PREALLOC_MODE_OFF) { 2361 BDRVQcow2State *s = blk_bs(blk)->opaque; 2362 qemu_co_mutex_lock(&s->lock); 2363 ret = preallocate(blk_bs(blk)); 2364 qemu_co_mutex_unlock(&s->lock); 2365 if (ret < 0) { 2366 error_setg_errno(errp, -ret, "Could not preallocate metadata"); 2367 goto out; 2368 } 2369 } 2370 2371 blk_unref(blk); 2372 blk = NULL; 2373 2374 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */ 2375 options = qdict_new(); 2376 qdict_put_str(options, "driver", "qcow2"); 2377 blk = blk_new_open(filename, NULL, options, 2378 BDRV_O_RDWR | BDRV_O_NO_BACKING, &local_err); 2379 if (blk == NULL) { 2380 error_propagate(errp, local_err); 2381 ret = -EIO; 2382 goto out; 2383 } 2384 2385 ret = 0; 2386 out: 2387 if (blk) { 2388 blk_unref(blk); 2389 } 2390 return ret; 2391 } 2392 2393 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp) 2394 { 2395 char *backing_file = NULL; 2396 char *backing_fmt = NULL; 2397 char *buf = NULL; 2398 uint64_t size = 0; 2399 int flags = 0; 2400 size_t cluster_size = DEFAULT_CLUSTER_SIZE; 2401 PreallocMode prealloc; 2402 int version = 3; 2403 uint64_t refcount_bits = 16; 2404 int refcount_order; 2405 Error *local_err = NULL; 2406 int ret; 2407 2408 /* Read out options */ 2409 size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2410 BDRV_SECTOR_SIZE); 2411 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); 2412 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT); 2413 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) { 2414 flags |= BLOCK_FLAG_ENCRYPT; 2415 } 2416 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 2417 DEFAULT_CLUSTER_SIZE); 2418 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2419 prealloc = qapi_enum_parse(PreallocMode_lookup, buf, 2420 PREALLOC_MODE__MAX, PREALLOC_MODE_OFF, 2421 &local_err); 2422 if (local_err) { 2423 error_propagate(errp, local_err); 2424 ret = -EINVAL; 2425 goto finish; 2426 } 2427 g_free(buf); 2428 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); 2429 if (!buf) { 2430 /* keep the default */ 2431 } else if (!strcmp(buf, "0.10")) { 2432 version = 2; 2433 } else if (!strcmp(buf, "1.1")) { 2434 version = 3; 2435 } else { 2436 error_setg(errp, "Invalid compatibility level: '%s'", buf); 2437 ret = -EINVAL; 2438 goto finish; 2439 } 2440 2441 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) { 2442 flags |= BLOCK_FLAG_LAZY_REFCOUNTS; 2443 } 2444 2445 if (backing_file && prealloc != PREALLOC_MODE_OFF) { 2446 error_setg(errp, "Backing file and preallocation cannot be used at " 2447 "the same time"); 2448 ret = -EINVAL; 2449 goto finish; 2450 } 2451 2452 if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) { 2453 error_setg(errp, "Lazy refcounts only supported with compatibility " 2454 "level 1.1 and above (use compat=1.1 or greater)"); 2455 ret = -EINVAL; 2456 goto finish; 2457 } 2458 2459 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 2460 refcount_bits); 2461 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) { 2462 error_setg(errp, "Refcount width must be a power of two and may not " 2463 "exceed 64 bits"); 2464 ret = -EINVAL; 2465 goto finish; 2466 } 2467 2468 if (version < 3 && refcount_bits != 16) { 2469 error_setg(errp, "Different refcount widths than 16 bits require " 2470 "compatibility level 1.1 or above (use compat=1.1 or " 2471 "greater)"); 2472 ret = -EINVAL; 2473 goto finish; 2474 } 2475 2476 refcount_order = ctz32(refcount_bits); 2477 2478 ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags, 2479 cluster_size, prealloc, opts, version, refcount_order, 2480 &local_err); 2481 error_propagate(errp, local_err); 2482 2483 finish: 2484 g_free(backing_file); 2485 g_free(backing_fmt); 2486 g_free(buf); 2487 return ret; 2488 } 2489 2490 2491 static bool is_zero_sectors(BlockDriverState *bs, int64_t start, 2492 uint32_t count) 2493 { 2494 int nr; 2495 BlockDriverState *file; 2496 int64_t res; 2497 2498 if (start + count > bs->total_sectors) { 2499 count = bs->total_sectors - start; 2500 } 2501 2502 if (!count) { 2503 return true; 2504 } 2505 res = bdrv_get_block_status_above(bs, NULL, start, count, 2506 &nr, &file); 2507 return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == count; 2508 } 2509 2510 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs, 2511 int64_t offset, int bytes, BdrvRequestFlags flags) 2512 { 2513 int ret; 2514 BDRVQcow2State *s = bs->opaque; 2515 2516 uint32_t head = offset % s->cluster_size; 2517 uint32_t tail = (offset + bytes) % s->cluster_size; 2518 2519 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes); 2520 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) { 2521 tail = 0; 2522 } 2523 2524 if (head || tail) { 2525 int64_t cl_start = (offset - head) >> BDRV_SECTOR_BITS; 2526 uint64_t off; 2527 unsigned int nr; 2528 2529 assert(head + bytes <= s->cluster_size); 2530 2531 /* check whether remainder of cluster already reads as zero */ 2532 if (!(is_zero_sectors(bs, cl_start, 2533 DIV_ROUND_UP(head, BDRV_SECTOR_SIZE)) && 2534 is_zero_sectors(bs, (offset + bytes) >> BDRV_SECTOR_BITS, 2535 DIV_ROUND_UP(-tail & (s->cluster_size - 1), 2536 BDRV_SECTOR_SIZE)))) { 2537 return -ENOTSUP; 2538 } 2539 2540 qemu_co_mutex_lock(&s->lock); 2541 /* We can have new write after previous check */ 2542 offset = cl_start << BDRV_SECTOR_BITS; 2543 bytes = s->cluster_size; 2544 nr = s->cluster_size; 2545 ret = qcow2_get_cluster_offset(bs, offset, &nr, &off); 2546 if (ret != QCOW2_CLUSTER_UNALLOCATED && 2547 ret != QCOW2_CLUSTER_ZERO_PLAIN && 2548 ret != QCOW2_CLUSTER_ZERO_ALLOC) { 2549 qemu_co_mutex_unlock(&s->lock); 2550 return -ENOTSUP; 2551 } 2552 } else { 2553 qemu_co_mutex_lock(&s->lock); 2554 } 2555 2556 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes); 2557 2558 /* Whatever is left can use real zero clusters */ 2559 ret = qcow2_cluster_zeroize(bs, offset, bytes, flags); 2560 qemu_co_mutex_unlock(&s->lock); 2561 2562 return ret; 2563 } 2564 2565 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs, 2566 int64_t offset, int bytes) 2567 { 2568 int ret; 2569 BDRVQcow2State *s = bs->opaque; 2570 2571 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) { 2572 assert(bytes < s->cluster_size); 2573 /* Ignore partial clusters, except for the special case of the 2574 * complete partial cluster at the end of an unaligned file */ 2575 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) || 2576 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) { 2577 return -ENOTSUP; 2578 } 2579 } 2580 2581 qemu_co_mutex_lock(&s->lock); 2582 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST, 2583 false); 2584 qemu_co_mutex_unlock(&s->lock); 2585 return ret; 2586 } 2587 2588 static int qcow2_truncate(BlockDriverState *bs, int64_t offset, Error **errp) 2589 { 2590 BDRVQcow2State *s = bs->opaque; 2591 int64_t new_l1_size; 2592 int ret; 2593 2594 if (offset & 511) { 2595 error_setg(errp, "The new size must be a multiple of 512"); 2596 return -EINVAL; 2597 } 2598 2599 /* cannot proceed if image has snapshots */ 2600 if (s->nb_snapshots) { 2601 error_setg(errp, "Can't resize an image which has snapshots"); 2602 return -ENOTSUP; 2603 } 2604 2605 /* shrinking is currently not supported */ 2606 if (offset < bs->total_sectors * 512) { 2607 error_setg(errp, "qcow2 doesn't support shrinking images yet"); 2608 return -ENOTSUP; 2609 } 2610 2611 new_l1_size = size_to_l1(s, offset); 2612 ret = qcow2_grow_l1_table(bs, new_l1_size, true); 2613 if (ret < 0) { 2614 error_setg_errno(errp, -ret, "Failed to grow the L1 table"); 2615 return ret; 2616 } 2617 2618 /* write updated header.size */ 2619 offset = cpu_to_be64(offset); 2620 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), 2621 &offset, sizeof(uint64_t)); 2622 if (ret < 0) { 2623 error_setg_errno(errp, -ret, "Failed to update the image size"); 2624 return ret; 2625 } 2626 2627 s->l1_vm_state_index = new_l1_size; 2628 return 0; 2629 } 2630 2631 /* XXX: put compressed sectors first, then all the cluster aligned 2632 tables to avoid losing bytes in alignment */ 2633 static coroutine_fn int 2634 qcow2_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset, 2635 uint64_t bytes, QEMUIOVector *qiov) 2636 { 2637 BDRVQcow2State *s = bs->opaque; 2638 QEMUIOVector hd_qiov; 2639 struct iovec iov; 2640 z_stream strm; 2641 int ret, out_len; 2642 uint8_t *buf, *out_buf; 2643 uint64_t cluster_offset; 2644 2645 if (bytes == 0) { 2646 /* align end of file to a sector boundary to ease reading with 2647 sector based I/Os */ 2648 cluster_offset = bdrv_getlength(bs->file->bs); 2649 return bdrv_truncate(bs->file, cluster_offset, NULL); 2650 } 2651 2652 buf = qemu_blockalign(bs, s->cluster_size); 2653 if (bytes != s->cluster_size) { 2654 if (bytes > s->cluster_size || 2655 offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS) 2656 { 2657 qemu_vfree(buf); 2658 return -EINVAL; 2659 } 2660 /* Zero-pad last write if image size is not cluster aligned */ 2661 memset(buf + bytes, 0, s->cluster_size - bytes); 2662 } 2663 qemu_iovec_to_buf(qiov, 0, buf, bytes); 2664 2665 out_buf = g_malloc(s->cluster_size); 2666 2667 /* best compression, small window, no zlib header */ 2668 memset(&strm, 0, sizeof(strm)); 2669 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, 2670 Z_DEFLATED, -12, 2671 9, Z_DEFAULT_STRATEGY); 2672 if (ret != 0) { 2673 ret = -EINVAL; 2674 goto fail; 2675 } 2676 2677 strm.avail_in = s->cluster_size; 2678 strm.next_in = (uint8_t *)buf; 2679 strm.avail_out = s->cluster_size; 2680 strm.next_out = out_buf; 2681 2682 ret = deflate(&strm, Z_FINISH); 2683 if (ret != Z_STREAM_END && ret != Z_OK) { 2684 deflateEnd(&strm); 2685 ret = -EINVAL; 2686 goto fail; 2687 } 2688 out_len = strm.next_out - out_buf; 2689 2690 deflateEnd(&strm); 2691 2692 if (ret != Z_STREAM_END || out_len >= s->cluster_size) { 2693 /* could not compress: write normal cluster */ 2694 ret = qcow2_co_pwritev(bs, offset, bytes, qiov, 0); 2695 if (ret < 0) { 2696 goto fail; 2697 } 2698 goto success; 2699 } 2700 2701 qemu_co_mutex_lock(&s->lock); 2702 cluster_offset = 2703 qcow2_alloc_compressed_cluster_offset(bs, offset, out_len); 2704 if (!cluster_offset) { 2705 qemu_co_mutex_unlock(&s->lock); 2706 ret = -EIO; 2707 goto fail; 2708 } 2709 cluster_offset &= s->cluster_offset_mask; 2710 2711 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len); 2712 qemu_co_mutex_unlock(&s->lock); 2713 if (ret < 0) { 2714 goto fail; 2715 } 2716 2717 iov = (struct iovec) { 2718 .iov_base = out_buf, 2719 .iov_len = out_len, 2720 }; 2721 qemu_iovec_init_external(&hd_qiov, &iov, 1); 2722 2723 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); 2724 ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0); 2725 if (ret < 0) { 2726 goto fail; 2727 } 2728 success: 2729 ret = 0; 2730 fail: 2731 qemu_vfree(buf); 2732 g_free(out_buf); 2733 return ret; 2734 } 2735 2736 static int make_completely_empty(BlockDriverState *bs) 2737 { 2738 BDRVQcow2State *s = bs->opaque; 2739 Error *local_err = NULL; 2740 int ret, l1_clusters; 2741 int64_t offset; 2742 uint64_t *new_reftable = NULL; 2743 uint64_t rt_entry, l1_size2; 2744 struct { 2745 uint64_t l1_offset; 2746 uint64_t reftable_offset; 2747 uint32_t reftable_clusters; 2748 } QEMU_PACKED l1_ofs_rt_ofs_cls; 2749 2750 ret = qcow2_cache_empty(bs, s->l2_table_cache); 2751 if (ret < 0) { 2752 goto fail; 2753 } 2754 2755 ret = qcow2_cache_empty(bs, s->refcount_block_cache); 2756 if (ret < 0) { 2757 goto fail; 2758 } 2759 2760 /* Refcounts will be broken utterly */ 2761 ret = qcow2_mark_dirty(bs); 2762 if (ret < 0) { 2763 goto fail; 2764 } 2765 2766 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 2767 2768 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 2769 l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t); 2770 2771 /* After this call, neither the in-memory nor the on-disk refcount 2772 * information accurately describe the actual references */ 2773 2774 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset, 2775 l1_clusters * s->cluster_size, 0); 2776 if (ret < 0) { 2777 goto fail_broken_refcounts; 2778 } 2779 memset(s->l1_table, 0, l1_size2); 2780 2781 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); 2782 2783 /* Overwrite enough clusters at the beginning of the sectors to place 2784 * the refcount table, a refcount block and the L1 table in; this may 2785 * overwrite parts of the existing refcount and L1 table, which is not 2786 * an issue because the dirty flag is set, complete data loss is in fact 2787 * desired and partial data loss is consequently fine as well */ 2788 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size, 2789 (2 + l1_clusters) * s->cluster_size, 0); 2790 /* This call (even if it failed overall) may have overwritten on-disk 2791 * refcount structures; in that case, the in-memory refcount information 2792 * will probably differ from the on-disk information which makes the BDS 2793 * unusable */ 2794 if (ret < 0) { 2795 goto fail_broken_refcounts; 2796 } 2797 2798 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); 2799 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); 2800 2801 /* "Create" an empty reftable (one cluster) directly after the image 2802 * header and an empty L1 table three clusters after the image header; 2803 * the cluster between those two will be used as the first refblock */ 2804 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size); 2805 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size); 2806 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1); 2807 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), 2808 &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); 2809 if (ret < 0) { 2810 goto fail_broken_refcounts; 2811 } 2812 2813 s->l1_table_offset = 3 * s->cluster_size; 2814 2815 new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t)); 2816 if (!new_reftable) { 2817 ret = -ENOMEM; 2818 goto fail_broken_refcounts; 2819 } 2820 2821 s->refcount_table_offset = s->cluster_size; 2822 s->refcount_table_size = s->cluster_size / sizeof(uint64_t); 2823 s->max_refcount_table_index = 0; 2824 2825 g_free(s->refcount_table); 2826 s->refcount_table = new_reftable; 2827 new_reftable = NULL; 2828 2829 /* Now the in-memory refcount information again corresponds to the on-disk 2830 * information (reftable is empty and no refblocks (the refblock cache is 2831 * empty)); however, this means some clusters (e.g. the image header) are 2832 * referenced, but not refcounted, but the normal qcow2 code assumes that 2833 * the in-memory information is always correct */ 2834 2835 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); 2836 2837 /* Enter the first refblock into the reftable */ 2838 rt_entry = cpu_to_be64(2 * s->cluster_size); 2839 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, 2840 &rt_entry, sizeof(rt_entry)); 2841 if (ret < 0) { 2842 goto fail_broken_refcounts; 2843 } 2844 s->refcount_table[0] = 2 * s->cluster_size; 2845 2846 s->free_cluster_index = 0; 2847 assert(3 + l1_clusters <= s->refcount_block_size); 2848 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); 2849 if (offset < 0) { 2850 ret = offset; 2851 goto fail_broken_refcounts; 2852 } else if (offset > 0) { 2853 error_report("First cluster in emptied image is in use"); 2854 abort(); 2855 } 2856 2857 /* Now finally the in-memory information corresponds to the on-disk 2858 * structures and is correct */ 2859 ret = qcow2_mark_clean(bs); 2860 if (ret < 0) { 2861 goto fail; 2862 } 2863 2864 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, 2865 &local_err); 2866 if (ret < 0) { 2867 error_report_err(local_err); 2868 goto fail; 2869 } 2870 2871 return 0; 2872 2873 fail_broken_refcounts: 2874 /* The BDS is unusable at this point. If we wanted to make it usable, we 2875 * would have to call qcow2_refcount_close(), qcow2_refcount_init(), 2876 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() 2877 * again. However, because the functions which could have caused this error 2878 * path to be taken are used by those functions as well, it's very likely 2879 * that that sequence will fail as well. Therefore, just eject the BDS. */ 2880 bs->drv = NULL; 2881 2882 fail: 2883 g_free(new_reftable); 2884 return ret; 2885 } 2886 2887 static int qcow2_make_empty(BlockDriverState *bs) 2888 { 2889 BDRVQcow2State *s = bs->opaque; 2890 uint64_t offset, end_offset; 2891 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size); 2892 int l1_clusters, ret = 0; 2893 2894 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t)); 2895 2896 if (s->qcow_version >= 3 && !s->snapshots && 2897 3 + l1_clusters <= s->refcount_block_size) { 2898 /* The following function only works for qcow2 v3 images (it requires 2899 * the dirty flag) and only as long as there are no snapshots (because 2900 * it completely empties the image). Furthermore, the L1 table and three 2901 * additional clusters (image header, refcount table, one refcount 2902 * block) have to fit inside one refcount block. */ 2903 return make_completely_empty(bs); 2904 } 2905 2906 /* This fallback code simply discards every active cluster; this is slow, 2907 * but works in all cases */ 2908 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE; 2909 for (offset = 0; offset < end_offset; offset += step) { 2910 /* As this function is generally used after committing an external 2911 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the 2912 * default action for this kind of discard is to pass the discard, 2913 * which will ideally result in an actually smaller image file, as 2914 * is probably desired. */ 2915 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset), 2916 QCOW2_DISCARD_SNAPSHOT, true); 2917 if (ret < 0) { 2918 break; 2919 } 2920 } 2921 2922 return ret; 2923 } 2924 2925 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) 2926 { 2927 BDRVQcow2State *s = bs->opaque; 2928 int ret; 2929 2930 qemu_co_mutex_lock(&s->lock); 2931 ret = qcow2_cache_write(bs, s->l2_table_cache); 2932 if (ret < 0) { 2933 qemu_co_mutex_unlock(&s->lock); 2934 return ret; 2935 } 2936 2937 if (qcow2_need_accurate_refcounts(s)) { 2938 ret = qcow2_cache_write(bs, s->refcount_block_cache); 2939 if (ret < 0) { 2940 qemu_co_mutex_unlock(&s->lock); 2941 return ret; 2942 } 2943 } 2944 qemu_co_mutex_unlock(&s->lock); 2945 2946 return 0; 2947 } 2948 2949 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2950 { 2951 BDRVQcow2State *s = bs->opaque; 2952 bdi->unallocated_blocks_are_zero = true; 2953 bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3); 2954 bdi->cluster_size = s->cluster_size; 2955 bdi->vm_state_offset = qcow2_vm_state_offset(s); 2956 return 0; 2957 } 2958 2959 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs) 2960 { 2961 BDRVQcow2State *s = bs->opaque; 2962 ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1); 2963 2964 *spec_info = (ImageInfoSpecific){ 2965 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2, 2966 .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1), 2967 }; 2968 if (s->qcow_version == 2) { 2969 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 2970 .compat = g_strdup("0.10"), 2971 .refcount_bits = s->refcount_bits, 2972 }; 2973 } else if (s->qcow_version == 3) { 2974 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ 2975 .compat = g_strdup("1.1"), 2976 .lazy_refcounts = s->compatible_features & 2977 QCOW2_COMPAT_LAZY_REFCOUNTS, 2978 .has_lazy_refcounts = true, 2979 .corrupt = s->incompatible_features & 2980 QCOW2_INCOMPAT_CORRUPT, 2981 .has_corrupt = true, 2982 .refcount_bits = s->refcount_bits, 2983 }; 2984 } else { 2985 /* if this assertion fails, this probably means a new version was 2986 * added without having it covered here */ 2987 assert(false); 2988 } 2989 2990 return spec_info; 2991 } 2992 2993 #if 0 2994 static void dump_refcounts(BlockDriverState *bs) 2995 { 2996 BDRVQcow2State *s = bs->opaque; 2997 int64_t nb_clusters, k, k1, size; 2998 int refcount; 2999 3000 size = bdrv_getlength(bs->file->bs); 3001 nb_clusters = size_to_clusters(s, size); 3002 for(k = 0; k < nb_clusters;) { 3003 k1 = k; 3004 refcount = get_refcount(bs, k); 3005 k++; 3006 while (k < nb_clusters && get_refcount(bs, k) == refcount) 3007 k++; 3008 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount, 3009 k - k1); 3010 } 3011 } 3012 #endif 3013 3014 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 3015 int64_t pos) 3016 { 3017 BDRVQcow2State *s = bs->opaque; 3018 3019 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); 3020 return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos, 3021 qiov->size, qiov, 0); 3022 } 3023 3024 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, 3025 int64_t pos) 3026 { 3027 BDRVQcow2State *s = bs->opaque; 3028 3029 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); 3030 return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos, 3031 qiov->size, qiov, 0); 3032 } 3033 3034 /* 3035 * Downgrades an image's version. To achieve this, any incompatible features 3036 * have to be removed. 3037 */ 3038 static int qcow2_downgrade(BlockDriverState *bs, int target_version, 3039 BlockDriverAmendStatusCB *status_cb, void *cb_opaque) 3040 { 3041 BDRVQcow2State *s = bs->opaque; 3042 int current_version = s->qcow_version; 3043 int ret; 3044 3045 if (target_version == current_version) { 3046 return 0; 3047 } else if (target_version > current_version) { 3048 return -EINVAL; 3049 } else if (target_version != 2) { 3050 return -EINVAL; 3051 } 3052 3053 if (s->refcount_order != 4) { 3054 error_report("compat=0.10 requires refcount_bits=16"); 3055 return -ENOTSUP; 3056 } 3057 3058 /* clear incompatible features */ 3059 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { 3060 ret = qcow2_mark_clean(bs); 3061 if (ret < 0) { 3062 return ret; 3063 } 3064 } 3065 3066 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in 3067 * the first place; if that happens nonetheless, returning -ENOTSUP is the 3068 * best thing to do anyway */ 3069 3070 if (s->incompatible_features) { 3071 return -ENOTSUP; 3072 } 3073 3074 /* since we can ignore compatible features, we can set them to 0 as well */ 3075 s->compatible_features = 0; 3076 /* if lazy refcounts have been used, they have already been fixed through 3077 * clearing the dirty flag */ 3078 3079 /* clearing autoclear features is trivial */ 3080 s->autoclear_features = 0; 3081 3082 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque); 3083 if (ret < 0) { 3084 return ret; 3085 } 3086 3087 s->qcow_version = target_version; 3088 ret = qcow2_update_header(bs); 3089 if (ret < 0) { 3090 s->qcow_version = current_version; 3091 return ret; 3092 } 3093 return 0; 3094 } 3095 3096 typedef enum Qcow2AmendOperation { 3097 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be 3098 * statically initialized to so that the helper CB can discern the first 3099 * invocation from an operation change */ 3100 QCOW2_NO_OPERATION = 0, 3101 3102 QCOW2_CHANGING_REFCOUNT_ORDER, 3103 QCOW2_DOWNGRADING, 3104 } Qcow2AmendOperation; 3105 3106 typedef struct Qcow2AmendHelperCBInfo { 3107 /* The code coordinating the amend operations should only modify 3108 * these four fields; the rest will be managed by the CB */ 3109 BlockDriverAmendStatusCB *original_status_cb; 3110 void *original_cb_opaque; 3111 3112 Qcow2AmendOperation current_operation; 3113 3114 /* Total number of operations to perform (only set once) */ 3115 int total_operations; 3116 3117 /* The following fields are managed by the CB */ 3118 3119 /* Number of operations completed */ 3120 int operations_completed; 3121 3122 /* Cumulative offset of all completed operations */ 3123 int64_t offset_completed; 3124 3125 Qcow2AmendOperation last_operation; 3126 int64_t last_work_size; 3127 } Qcow2AmendHelperCBInfo; 3128 3129 static void qcow2_amend_helper_cb(BlockDriverState *bs, 3130 int64_t operation_offset, 3131 int64_t operation_work_size, void *opaque) 3132 { 3133 Qcow2AmendHelperCBInfo *info = opaque; 3134 int64_t current_work_size; 3135 int64_t projected_work_size; 3136 3137 if (info->current_operation != info->last_operation) { 3138 if (info->last_operation != QCOW2_NO_OPERATION) { 3139 info->offset_completed += info->last_work_size; 3140 info->operations_completed++; 3141 } 3142 3143 info->last_operation = info->current_operation; 3144 } 3145 3146 assert(info->total_operations > 0); 3147 assert(info->operations_completed < info->total_operations); 3148 3149 info->last_work_size = operation_work_size; 3150 3151 current_work_size = info->offset_completed + operation_work_size; 3152 3153 /* current_work_size is the total work size for (operations_completed + 1) 3154 * operations (which includes this one), so multiply it by the number of 3155 * operations not covered and divide it by the number of operations 3156 * covered to get a projection for the operations not covered */ 3157 projected_work_size = current_work_size * (info->total_operations - 3158 info->operations_completed - 1) 3159 / (info->operations_completed + 1); 3160 3161 info->original_status_cb(bs, info->offset_completed + operation_offset, 3162 current_work_size + projected_work_size, 3163 info->original_cb_opaque); 3164 } 3165 3166 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, 3167 BlockDriverAmendStatusCB *status_cb, 3168 void *cb_opaque) 3169 { 3170 BDRVQcow2State *s = bs->opaque; 3171 int old_version = s->qcow_version, new_version = old_version; 3172 uint64_t new_size = 0; 3173 const char *backing_file = NULL, *backing_format = NULL; 3174 bool lazy_refcounts = s->use_lazy_refcounts; 3175 const char *compat = NULL; 3176 uint64_t cluster_size = s->cluster_size; 3177 bool encrypt; 3178 int refcount_bits = s->refcount_bits; 3179 Error *local_err = NULL; 3180 int ret; 3181 QemuOptDesc *desc = opts->list->desc; 3182 Qcow2AmendHelperCBInfo helper_cb_info; 3183 3184 while (desc && desc->name) { 3185 if (!qemu_opt_find(opts, desc->name)) { 3186 /* only change explicitly defined options */ 3187 desc++; 3188 continue; 3189 } 3190 3191 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) { 3192 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL); 3193 if (!compat) { 3194 /* preserve default */ 3195 } else if (!strcmp(compat, "0.10")) { 3196 new_version = 2; 3197 } else if (!strcmp(compat, "1.1")) { 3198 new_version = 3; 3199 } else { 3200 error_report("Unknown compatibility level %s", compat); 3201 return -EINVAL; 3202 } 3203 } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) { 3204 error_report("Cannot change preallocation mode"); 3205 return -ENOTSUP; 3206 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) { 3207 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); 3208 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) { 3209 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); 3210 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) { 3211 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); 3212 } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) { 3213 encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, 3214 !!s->cipher); 3215 3216 if (encrypt != !!s->cipher) { 3217 error_report("Changing the encryption flag is not supported"); 3218 return -ENOTSUP; 3219 } 3220 } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) { 3221 cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 3222 cluster_size); 3223 if (cluster_size != s->cluster_size) { 3224 error_report("Changing the cluster size is not supported"); 3225 return -ENOTSUP; 3226 } 3227 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) { 3228 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS, 3229 lazy_refcounts); 3230 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) { 3231 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS, 3232 refcount_bits); 3233 3234 if (refcount_bits <= 0 || refcount_bits > 64 || 3235 !is_power_of_2(refcount_bits)) 3236 { 3237 error_report("Refcount width must be a power of two and may " 3238 "not exceed 64 bits"); 3239 return -EINVAL; 3240 } 3241 } else { 3242 /* if this point is reached, this probably means a new option was 3243 * added without having it covered here */ 3244 abort(); 3245 } 3246 3247 desc++; 3248 } 3249 3250 helper_cb_info = (Qcow2AmendHelperCBInfo){ 3251 .original_status_cb = status_cb, 3252 .original_cb_opaque = cb_opaque, 3253 .total_operations = (new_version < old_version) 3254 + (s->refcount_bits != refcount_bits) 3255 }; 3256 3257 /* Upgrade first (some features may require compat=1.1) */ 3258 if (new_version > old_version) { 3259 s->qcow_version = new_version; 3260 ret = qcow2_update_header(bs); 3261 if (ret < 0) { 3262 s->qcow_version = old_version; 3263 return ret; 3264 } 3265 } 3266 3267 if (s->refcount_bits != refcount_bits) { 3268 int refcount_order = ctz32(refcount_bits); 3269 3270 if (new_version < 3 && refcount_bits != 16) { 3271 error_report("Different refcount widths than 16 bits require " 3272 "compatibility level 1.1 or above (use compat=1.1 or " 3273 "greater)"); 3274 return -EINVAL; 3275 } 3276 3277 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER; 3278 ret = qcow2_change_refcount_order(bs, refcount_order, 3279 &qcow2_amend_helper_cb, 3280 &helper_cb_info, &local_err); 3281 if (ret < 0) { 3282 error_report_err(local_err); 3283 return ret; 3284 } 3285 } 3286 3287 if (backing_file || backing_format) { 3288 ret = qcow2_change_backing_file(bs, 3289 backing_file ?: s->image_backing_file, 3290 backing_format ?: s->image_backing_format); 3291 if (ret < 0) { 3292 return ret; 3293 } 3294 } 3295 3296 if (s->use_lazy_refcounts != lazy_refcounts) { 3297 if (lazy_refcounts) { 3298 if (new_version < 3) { 3299 error_report("Lazy refcounts only supported with compatibility " 3300 "level 1.1 and above (use compat=1.1 or greater)"); 3301 return -EINVAL; 3302 } 3303 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 3304 ret = qcow2_update_header(bs); 3305 if (ret < 0) { 3306 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 3307 return ret; 3308 } 3309 s->use_lazy_refcounts = true; 3310 } else { 3311 /* make image clean first */ 3312 ret = qcow2_mark_clean(bs); 3313 if (ret < 0) { 3314 return ret; 3315 } 3316 /* now disallow lazy refcounts */ 3317 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; 3318 ret = qcow2_update_header(bs); 3319 if (ret < 0) { 3320 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; 3321 return ret; 3322 } 3323 s->use_lazy_refcounts = false; 3324 } 3325 } 3326 3327 if (new_size) { 3328 BlockBackend *blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL); 3329 ret = blk_insert_bs(blk, bs, &local_err); 3330 if (ret < 0) { 3331 error_report_err(local_err); 3332 blk_unref(blk); 3333 return ret; 3334 } 3335 3336 ret = blk_truncate(blk, new_size, &local_err); 3337 blk_unref(blk); 3338 if (ret < 0) { 3339 error_report_err(local_err); 3340 return ret; 3341 } 3342 } 3343 3344 /* Downgrade last (so unsupported features can be removed before) */ 3345 if (new_version < old_version) { 3346 helper_cb_info.current_operation = QCOW2_DOWNGRADING; 3347 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb, 3348 &helper_cb_info); 3349 if (ret < 0) { 3350 return ret; 3351 } 3352 } 3353 3354 return 0; 3355 } 3356 3357 /* 3358 * If offset or size are negative, respectively, they will not be included in 3359 * the BLOCK_IMAGE_CORRUPTED event emitted. 3360 * fatal will be ignored for read-only BDS; corruptions found there will always 3361 * be considered non-fatal. 3362 */ 3363 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, 3364 int64_t size, const char *message_format, ...) 3365 { 3366 BDRVQcow2State *s = bs->opaque; 3367 const char *node_name; 3368 char *message; 3369 va_list ap; 3370 3371 fatal = fatal && !bs->read_only; 3372 3373 if (s->signaled_corruption && 3374 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) 3375 { 3376 return; 3377 } 3378 3379 va_start(ap, message_format); 3380 message = g_strdup_vprintf(message_format, ap); 3381 va_end(ap); 3382 3383 if (fatal) { 3384 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " 3385 "corruption events will be suppressed\n", message); 3386 } else { 3387 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " 3388 "corruption events will be suppressed\n", message); 3389 } 3390 3391 node_name = bdrv_get_node_name(bs); 3392 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), 3393 *node_name != '\0', node_name, 3394 message, offset >= 0, offset, 3395 size >= 0, size, 3396 fatal, &error_abort); 3397 g_free(message); 3398 3399 if (fatal) { 3400 qcow2_mark_corrupt(bs); 3401 bs->drv = NULL; /* make BDS unusable */ 3402 } 3403 3404 s->signaled_corruption = true; 3405 } 3406 3407 static QemuOptsList qcow2_create_opts = { 3408 .name = "qcow2-create-opts", 3409 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), 3410 .desc = { 3411 { 3412 .name = BLOCK_OPT_SIZE, 3413 .type = QEMU_OPT_SIZE, 3414 .help = "Virtual disk size" 3415 }, 3416 { 3417 .name = BLOCK_OPT_COMPAT_LEVEL, 3418 .type = QEMU_OPT_STRING, 3419 .help = "Compatibility level (0.10 or 1.1)" 3420 }, 3421 { 3422 .name = BLOCK_OPT_BACKING_FILE, 3423 .type = QEMU_OPT_STRING, 3424 .help = "File name of a base image" 3425 }, 3426 { 3427 .name = BLOCK_OPT_BACKING_FMT, 3428 .type = QEMU_OPT_STRING, 3429 .help = "Image format of the base image" 3430 }, 3431 { 3432 .name = BLOCK_OPT_ENCRYPT, 3433 .type = QEMU_OPT_BOOL, 3434 .help = "Encrypt the image", 3435 .def_value_str = "off" 3436 }, 3437 { 3438 .name = BLOCK_OPT_CLUSTER_SIZE, 3439 .type = QEMU_OPT_SIZE, 3440 .help = "qcow2 cluster size", 3441 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) 3442 }, 3443 { 3444 .name = BLOCK_OPT_PREALLOC, 3445 .type = QEMU_OPT_STRING, 3446 .help = "Preallocation mode (allowed values: off, metadata, " 3447 "falloc, full)" 3448 }, 3449 { 3450 .name = BLOCK_OPT_LAZY_REFCOUNTS, 3451 .type = QEMU_OPT_BOOL, 3452 .help = "Postpone refcount updates", 3453 .def_value_str = "off" 3454 }, 3455 { 3456 .name = BLOCK_OPT_REFCOUNT_BITS, 3457 .type = QEMU_OPT_NUMBER, 3458 .help = "Width of a reference count entry in bits", 3459 .def_value_str = "16" 3460 }, 3461 { /* end of list */ } 3462 } 3463 }; 3464 3465 BlockDriver bdrv_qcow2 = { 3466 .format_name = "qcow2", 3467 .instance_size = sizeof(BDRVQcow2State), 3468 .bdrv_probe = qcow2_probe, 3469 .bdrv_open = qcow2_open, 3470 .bdrv_close = qcow2_close, 3471 .bdrv_reopen_prepare = qcow2_reopen_prepare, 3472 .bdrv_reopen_commit = qcow2_reopen_commit, 3473 .bdrv_reopen_abort = qcow2_reopen_abort, 3474 .bdrv_join_options = qcow2_join_options, 3475 .bdrv_child_perm = bdrv_format_default_perms, 3476 .bdrv_create = qcow2_create, 3477 .bdrv_has_zero_init = bdrv_has_zero_init_1, 3478 .bdrv_co_get_block_status = qcow2_co_get_block_status, 3479 .bdrv_set_key = qcow2_set_key, 3480 3481 .bdrv_co_preadv = qcow2_co_preadv, 3482 .bdrv_co_pwritev = qcow2_co_pwritev, 3483 .bdrv_co_flush_to_os = qcow2_co_flush_to_os, 3484 3485 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes, 3486 .bdrv_co_pdiscard = qcow2_co_pdiscard, 3487 .bdrv_truncate = qcow2_truncate, 3488 .bdrv_co_pwritev_compressed = qcow2_co_pwritev_compressed, 3489 .bdrv_make_empty = qcow2_make_empty, 3490 3491 .bdrv_snapshot_create = qcow2_snapshot_create, 3492 .bdrv_snapshot_goto = qcow2_snapshot_goto, 3493 .bdrv_snapshot_delete = qcow2_snapshot_delete, 3494 .bdrv_snapshot_list = qcow2_snapshot_list, 3495 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, 3496 .bdrv_get_info = qcow2_get_info, 3497 .bdrv_get_specific_info = qcow2_get_specific_info, 3498 3499 .bdrv_save_vmstate = qcow2_save_vmstate, 3500 .bdrv_load_vmstate = qcow2_load_vmstate, 3501 3502 .supports_backing = true, 3503 .bdrv_change_backing_file = qcow2_change_backing_file, 3504 3505 .bdrv_refresh_limits = qcow2_refresh_limits, 3506 .bdrv_invalidate_cache = qcow2_invalidate_cache, 3507 .bdrv_inactivate = qcow2_inactivate, 3508 3509 .create_opts = &qcow2_create_opts, 3510 .bdrv_check = qcow2_check, 3511 .bdrv_amend_options = qcow2_amend_options, 3512 3513 .bdrv_detach_aio_context = qcow2_detach_aio_context, 3514 .bdrv_attach_aio_context = qcow2_attach_aio_context, 3515 }; 3516 3517 static void bdrv_qcow2_init(void) 3518 { 3519 bdrv_register(&bdrv_qcow2); 3520 } 3521 3522 block_init(bdrv_qcow2_init); 3523