1 /* 2 * Block driver for the Virtual Disk Image (VDI) format 3 * 4 * Copyright (c) 2009, 2012 Stefan Weil 5 * 6 * This program is free software: you can redistribute it and/or modify 7 * it under the terms of the GNU General Public License as published by 8 * the Free Software Foundation, either version 2 of the License, or 9 * (at your option) version 3 or any later version. 10 * 11 * This program is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 14 * GNU General Public License for more details. 15 * 16 * You should have received a copy of the GNU General Public License 17 * along with this program. If not, see <http://www.gnu.org/licenses/>. 18 * 19 * Reference: 20 * http://forums.virtualbox.org/viewtopic.php?t=8046 21 * 22 * This driver supports create / read / write operations on VDI images. 23 * 24 * Todo (see also TODO in code): 25 * 26 * Some features like snapshots are still missing. 27 * 28 * Deallocation of zero-filled blocks and shrinking images are missing, too 29 * (might be added to common block layer). 30 * 31 * Allocation of blocks could be optimized (less writes to block map and 32 * header). 33 * 34 * Read and write of adjacent blocks could be done in one operation 35 * (current code uses one operation per block (1 MiB). 36 * 37 * The code is not thread safe (missing locks for changes in header and 38 * block table, no problem with current QEMU). 39 * 40 * Hints: 41 * 42 * Blocks (VDI documentation) correspond to clusters (QEMU). 43 * QEMU's backing files could be implemented using VDI snapshot files (TODO). 44 * VDI snapshot files may also contain the complete machine state. 45 * Maybe this machine state can be converted to QEMU PC machine snapshot data. 46 * 47 * The driver keeps a block cache (little endian entries) in memory. 48 * For the standard block size (1 MiB), a 1 TiB disk will use 4 MiB RAM, 49 * so this seems to be reasonable. 50 */ 51 52 #include "qemu/osdep.h" 53 #include "qapi/error.h" 54 #include "qapi/qmp/qdict.h" 55 #include "qapi/qobject-input-visitor.h" 56 #include "qapi/qapi-visit-block-core.h" 57 #include "block/block_int.h" 58 #include "sysemu/block-backend.h" 59 #include "qemu/module.h" 60 #include "qemu/option.h" 61 #include "qemu/bswap.h" 62 #include "migration/blocker.h" 63 #include "qemu/coroutine.h" 64 #include "qemu/cutils.h" 65 #include "qemu/uuid.h" 66 67 /* Code configuration options. */ 68 69 /* Enable debug messages. */ 70 //~ #define CONFIG_VDI_DEBUG 71 72 /* Support write operations on VDI images. */ 73 #define CONFIG_VDI_WRITE 74 75 /* Support non-standard block (cluster) size. This is untested. 76 * Maybe it will be needed for very large images. 77 */ 78 //~ #define CONFIG_VDI_BLOCK_SIZE 79 80 /* Support static (fixed, pre-allocated) images. */ 81 #define CONFIG_VDI_STATIC_IMAGE 82 83 /* Command line option for static images. */ 84 #define BLOCK_OPT_STATIC "static" 85 86 #define KiB 1024 87 #define MiB (KiB * KiB) 88 89 #define SECTOR_SIZE 512 90 #define DEFAULT_CLUSTER_SIZE (1 * MiB) 91 92 #if defined(CONFIG_VDI_DEBUG) 93 #define VDI_DEBUG 1 94 #else 95 #define VDI_DEBUG 0 96 #endif 97 98 #define logout(fmt, ...) \ 99 do { \ 100 if (VDI_DEBUG) { \ 101 fprintf(stderr, "vdi\t%-24s" fmt, __func__, ##__VA_ARGS__); \ 102 } \ 103 } while (0) 104 105 /* Image signature. */ 106 #define VDI_SIGNATURE 0xbeda107f 107 108 /* Image version. */ 109 #define VDI_VERSION_1_1 0x00010001 110 111 /* Image type. */ 112 #define VDI_TYPE_DYNAMIC 1 113 #define VDI_TYPE_STATIC 2 114 115 /* Innotek / SUN images use these strings in header.text: 116 * "<<< innotek VirtualBox Disk Image >>>\n" 117 * "<<< Sun xVM VirtualBox Disk Image >>>\n" 118 * "<<< Sun VirtualBox Disk Image >>>\n" 119 * The value does not matter, so QEMU created images use a different text. 120 */ 121 #define VDI_TEXT "<<< QEMU VM Virtual Disk Image >>>\n" 122 123 /* A never-allocated block; semantically arbitrary content. */ 124 #define VDI_UNALLOCATED 0xffffffffU 125 126 /* A discarded (no longer allocated) block; semantically zero-filled. */ 127 #define VDI_DISCARDED 0xfffffffeU 128 129 #define VDI_IS_ALLOCATED(X) ((X) < VDI_DISCARDED) 130 131 /* The bmap will take up VDI_BLOCKS_IN_IMAGE_MAX * sizeof(uint32_t) bytes; since 132 * the bmap is read and written in a single operation, its size needs to be 133 * limited to INT_MAX; furthermore, when opening an image, the bmap size is 134 * rounded up to be aligned on BDRV_SECTOR_SIZE. 135 * Therefore this should satisfy the following: 136 * VDI_BLOCKS_IN_IMAGE_MAX * sizeof(uint32_t) + BDRV_SECTOR_SIZE == INT_MAX + 1 137 * (INT_MAX + 1 is the first value not representable as an int) 138 * This guarantees that any value below or equal to the constant will, when 139 * multiplied by sizeof(uint32_t) and rounded up to a BDRV_SECTOR_SIZE boundary, 140 * still be below or equal to INT_MAX. */ 141 #define VDI_BLOCKS_IN_IMAGE_MAX \ 142 ((unsigned)((INT_MAX + 1u - BDRV_SECTOR_SIZE) / sizeof(uint32_t))) 143 #define VDI_DISK_SIZE_MAX ((uint64_t)VDI_BLOCKS_IN_IMAGE_MAX * \ 144 (uint64_t)DEFAULT_CLUSTER_SIZE) 145 146 static QemuOptsList vdi_create_opts; 147 148 typedef struct { 149 char text[0x40]; 150 uint32_t signature; 151 uint32_t version; 152 uint32_t header_size; 153 uint32_t image_type; 154 uint32_t image_flags; 155 char description[256]; 156 uint32_t offset_bmap; 157 uint32_t offset_data; 158 uint32_t cylinders; /* disk geometry, unused here */ 159 uint32_t heads; /* disk geometry, unused here */ 160 uint32_t sectors; /* disk geometry, unused here */ 161 uint32_t sector_size; 162 uint32_t unused1; 163 uint64_t disk_size; 164 uint32_t block_size; 165 uint32_t block_extra; /* unused here */ 166 uint32_t blocks_in_image; 167 uint32_t blocks_allocated; 168 QemuUUID uuid_image; 169 QemuUUID uuid_last_snap; 170 QemuUUID uuid_link; 171 QemuUUID uuid_parent; 172 uint64_t unused2[7]; 173 } QEMU_PACKED VdiHeader; 174 175 typedef struct { 176 /* The block map entries are little endian (even in memory). */ 177 uint32_t *bmap; 178 /* Size of block (bytes). */ 179 uint32_t block_size; 180 /* First sector of block map. */ 181 uint32_t bmap_sector; 182 /* VDI header (converted to host endianness). */ 183 VdiHeader header; 184 185 CoRwlock bmap_lock; 186 187 Error *migration_blocker; 188 } BDRVVdiState; 189 190 static void vdi_header_to_cpu(VdiHeader *header) 191 { 192 le32_to_cpus(&header->signature); 193 le32_to_cpus(&header->version); 194 le32_to_cpus(&header->header_size); 195 le32_to_cpus(&header->image_type); 196 le32_to_cpus(&header->image_flags); 197 le32_to_cpus(&header->offset_bmap); 198 le32_to_cpus(&header->offset_data); 199 le32_to_cpus(&header->cylinders); 200 le32_to_cpus(&header->heads); 201 le32_to_cpus(&header->sectors); 202 le32_to_cpus(&header->sector_size); 203 le64_to_cpus(&header->disk_size); 204 le32_to_cpus(&header->block_size); 205 le32_to_cpus(&header->block_extra); 206 le32_to_cpus(&header->blocks_in_image); 207 le32_to_cpus(&header->blocks_allocated); 208 qemu_uuid_bswap(&header->uuid_image); 209 qemu_uuid_bswap(&header->uuid_last_snap); 210 qemu_uuid_bswap(&header->uuid_link); 211 qemu_uuid_bswap(&header->uuid_parent); 212 } 213 214 static void vdi_header_to_le(VdiHeader *header) 215 { 216 cpu_to_le32s(&header->signature); 217 cpu_to_le32s(&header->version); 218 cpu_to_le32s(&header->header_size); 219 cpu_to_le32s(&header->image_type); 220 cpu_to_le32s(&header->image_flags); 221 cpu_to_le32s(&header->offset_bmap); 222 cpu_to_le32s(&header->offset_data); 223 cpu_to_le32s(&header->cylinders); 224 cpu_to_le32s(&header->heads); 225 cpu_to_le32s(&header->sectors); 226 cpu_to_le32s(&header->sector_size); 227 cpu_to_le64s(&header->disk_size); 228 cpu_to_le32s(&header->block_size); 229 cpu_to_le32s(&header->block_extra); 230 cpu_to_le32s(&header->blocks_in_image); 231 cpu_to_le32s(&header->blocks_allocated); 232 qemu_uuid_bswap(&header->uuid_image); 233 qemu_uuid_bswap(&header->uuid_last_snap); 234 qemu_uuid_bswap(&header->uuid_link); 235 qemu_uuid_bswap(&header->uuid_parent); 236 } 237 238 static void vdi_header_print(VdiHeader *header) 239 { 240 char uuid[37]; 241 logout("text %s", header->text); 242 logout("signature 0x%08x\n", header->signature); 243 logout("header size 0x%04x\n", header->header_size); 244 logout("image type 0x%04x\n", header->image_type); 245 logout("image flags 0x%04x\n", header->image_flags); 246 logout("description %s\n", header->description); 247 logout("offset bmap 0x%04x\n", header->offset_bmap); 248 logout("offset data 0x%04x\n", header->offset_data); 249 logout("cylinders 0x%04x\n", header->cylinders); 250 logout("heads 0x%04x\n", header->heads); 251 logout("sectors 0x%04x\n", header->sectors); 252 logout("sector size 0x%04x\n", header->sector_size); 253 logout("image size 0x%" PRIx64 " B (%" PRIu64 " MiB)\n", 254 header->disk_size, header->disk_size / MiB); 255 logout("block size 0x%04x\n", header->block_size); 256 logout("block extra 0x%04x\n", header->block_extra); 257 logout("blocks tot. 0x%04x\n", header->blocks_in_image); 258 logout("blocks all. 0x%04x\n", header->blocks_allocated); 259 qemu_uuid_unparse(&header->uuid_image, uuid); 260 logout("uuid image %s\n", uuid); 261 qemu_uuid_unparse(&header->uuid_last_snap, uuid); 262 logout("uuid snap %s\n", uuid); 263 qemu_uuid_unparse(&header->uuid_link, uuid); 264 logout("uuid link %s\n", uuid); 265 qemu_uuid_unparse(&header->uuid_parent, uuid); 266 logout("uuid parent %s\n", uuid); 267 } 268 269 static int coroutine_fn vdi_co_check(BlockDriverState *bs, BdrvCheckResult *res, 270 BdrvCheckMode fix) 271 { 272 /* TODO: additional checks possible. */ 273 BDRVVdiState *s = (BDRVVdiState *)bs->opaque; 274 uint32_t blocks_allocated = 0; 275 uint32_t block; 276 uint32_t *bmap; 277 logout("\n"); 278 279 if (fix) { 280 return -ENOTSUP; 281 } 282 283 bmap = g_try_new(uint32_t, s->header.blocks_in_image); 284 if (s->header.blocks_in_image && bmap == NULL) { 285 res->check_errors++; 286 return -ENOMEM; 287 } 288 289 memset(bmap, 0xff, s->header.blocks_in_image * sizeof(uint32_t)); 290 291 /* Check block map and value of blocks_allocated. */ 292 for (block = 0; block < s->header.blocks_in_image; block++) { 293 uint32_t bmap_entry = le32_to_cpu(s->bmap[block]); 294 if (VDI_IS_ALLOCATED(bmap_entry)) { 295 if (bmap_entry < s->header.blocks_in_image) { 296 blocks_allocated++; 297 if (!VDI_IS_ALLOCATED(bmap[bmap_entry])) { 298 bmap[bmap_entry] = bmap_entry; 299 } else { 300 fprintf(stderr, "ERROR: block index %" PRIu32 301 " also used by %" PRIu32 "\n", bmap[bmap_entry], bmap_entry); 302 res->corruptions++; 303 } 304 } else { 305 fprintf(stderr, "ERROR: block index %" PRIu32 306 " too large, is %" PRIu32 "\n", block, bmap_entry); 307 res->corruptions++; 308 } 309 } 310 } 311 if (blocks_allocated != s->header.blocks_allocated) { 312 fprintf(stderr, "ERROR: allocated blocks mismatch, is %" PRIu32 313 ", should be %" PRIu32 "\n", 314 blocks_allocated, s->header.blocks_allocated); 315 res->corruptions++; 316 } 317 318 g_free(bmap); 319 320 return 0; 321 } 322 323 static int vdi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 324 { 325 /* TODO: vdi_get_info would be needed for machine snapshots. 326 vm_state_offset is still missing. */ 327 BDRVVdiState *s = (BDRVVdiState *)bs->opaque; 328 logout("\n"); 329 bdi->cluster_size = s->block_size; 330 bdi->vm_state_offset = 0; 331 bdi->unallocated_blocks_are_zero = true; 332 return 0; 333 } 334 335 static int vdi_make_empty(BlockDriverState *bs) 336 { 337 /* TODO: missing code. */ 338 logout("\n"); 339 /* The return value for missing code must be 0, see block.c. */ 340 return 0; 341 } 342 343 static int vdi_probe(const uint8_t *buf, int buf_size, const char *filename) 344 { 345 const VdiHeader *header = (const VdiHeader *)buf; 346 int ret = 0; 347 348 logout("\n"); 349 350 if (buf_size < sizeof(*header)) { 351 /* Header too small, no VDI. */ 352 } else if (le32_to_cpu(header->signature) == VDI_SIGNATURE) { 353 ret = 100; 354 } 355 356 if (ret == 0) { 357 logout("no vdi image\n"); 358 } else { 359 logout("%s", header->text); 360 } 361 362 return ret; 363 } 364 365 static int vdi_open(BlockDriverState *bs, QDict *options, int flags, 366 Error **errp) 367 { 368 BDRVVdiState *s = bs->opaque; 369 VdiHeader header; 370 size_t bmap_size; 371 int ret; 372 Error *local_err = NULL; 373 374 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, 375 false, errp); 376 if (!bs->file) { 377 return -EINVAL; 378 } 379 380 logout("\n"); 381 382 ret = bdrv_read(bs->file, 0, (uint8_t *)&header, 1); 383 if (ret < 0) { 384 goto fail; 385 } 386 387 vdi_header_to_cpu(&header); 388 if (VDI_DEBUG) { 389 vdi_header_print(&header); 390 } 391 392 if (header.disk_size > VDI_DISK_SIZE_MAX) { 393 error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64 394 ", max supported is 0x%" PRIx64 ")", 395 header.disk_size, VDI_DISK_SIZE_MAX); 396 ret = -ENOTSUP; 397 goto fail; 398 } 399 400 if (header.disk_size % SECTOR_SIZE != 0) { 401 /* 'VBoxManage convertfromraw' can create images with odd disk sizes. 402 We accept them but round the disk size to the next multiple of 403 SECTOR_SIZE. */ 404 logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size); 405 header.disk_size = ROUND_UP(header.disk_size, SECTOR_SIZE); 406 } 407 408 if (header.signature != VDI_SIGNATURE) { 409 error_setg(errp, "Image not in VDI format (bad signature %08" PRIx32 410 ")", header.signature); 411 ret = -EINVAL; 412 goto fail; 413 } else if (header.version != VDI_VERSION_1_1) { 414 error_setg(errp, "unsupported VDI image (version %" PRIu32 ".%" PRIu32 415 ")", header.version >> 16, header.version & 0xffff); 416 ret = -ENOTSUP; 417 goto fail; 418 } else if (header.offset_bmap % SECTOR_SIZE != 0) { 419 /* We only support block maps which start on a sector boundary. */ 420 error_setg(errp, "unsupported VDI image (unaligned block map offset " 421 "0x%" PRIx32 ")", header.offset_bmap); 422 ret = -ENOTSUP; 423 goto fail; 424 } else if (header.offset_data % SECTOR_SIZE != 0) { 425 /* We only support data blocks which start on a sector boundary. */ 426 error_setg(errp, "unsupported VDI image (unaligned data offset 0x%" 427 PRIx32 ")", header.offset_data); 428 ret = -ENOTSUP; 429 goto fail; 430 } else if (header.sector_size != SECTOR_SIZE) { 431 error_setg(errp, "unsupported VDI image (sector size %" PRIu32 432 " is not %u)", header.sector_size, SECTOR_SIZE); 433 ret = -ENOTSUP; 434 goto fail; 435 } else if (header.block_size != DEFAULT_CLUSTER_SIZE) { 436 error_setg(errp, "unsupported VDI image (block size %" PRIu32 437 " is not %u)", header.block_size, DEFAULT_CLUSTER_SIZE); 438 ret = -ENOTSUP; 439 goto fail; 440 } else if (header.disk_size > 441 (uint64_t)header.blocks_in_image * header.block_size) { 442 error_setg(errp, "unsupported VDI image (disk size %" PRIu64 ", " 443 "image bitmap has room for %" PRIu64 ")", 444 header.disk_size, 445 (uint64_t)header.blocks_in_image * header.block_size); 446 ret = -ENOTSUP; 447 goto fail; 448 } else if (!qemu_uuid_is_null(&header.uuid_link)) { 449 error_setg(errp, "unsupported VDI image (non-NULL link UUID)"); 450 ret = -ENOTSUP; 451 goto fail; 452 } else if (!qemu_uuid_is_null(&header.uuid_parent)) { 453 error_setg(errp, "unsupported VDI image (non-NULL parent UUID)"); 454 ret = -ENOTSUP; 455 goto fail; 456 } else if (header.blocks_in_image > VDI_BLOCKS_IN_IMAGE_MAX) { 457 error_setg(errp, "unsupported VDI image " 458 "(too many blocks %u, max is %u)", 459 header.blocks_in_image, VDI_BLOCKS_IN_IMAGE_MAX); 460 ret = -ENOTSUP; 461 goto fail; 462 } 463 464 bs->total_sectors = header.disk_size / SECTOR_SIZE; 465 466 s->block_size = header.block_size; 467 s->bmap_sector = header.offset_bmap / SECTOR_SIZE; 468 s->header = header; 469 470 bmap_size = header.blocks_in_image * sizeof(uint32_t); 471 bmap_size = DIV_ROUND_UP(bmap_size, SECTOR_SIZE); 472 s->bmap = qemu_try_blockalign(bs->file->bs, bmap_size * SECTOR_SIZE); 473 if (s->bmap == NULL) { 474 ret = -ENOMEM; 475 goto fail; 476 } 477 478 ret = bdrv_read(bs->file, s->bmap_sector, (uint8_t *)s->bmap, 479 bmap_size); 480 if (ret < 0) { 481 goto fail_free_bmap; 482 } 483 484 /* Disable migration when vdi images are used */ 485 error_setg(&s->migration_blocker, "The vdi format used by node '%s' " 486 "does not support live migration", 487 bdrv_get_device_or_node_name(bs)); 488 ret = migrate_add_blocker(s->migration_blocker, &local_err); 489 if (local_err) { 490 error_propagate(errp, local_err); 491 error_free(s->migration_blocker); 492 goto fail_free_bmap; 493 } 494 495 qemu_co_rwlock_init(&s->bmap_lock); 496 497 return 0; 498 499 fail_free_bmap: 500 qemu_vfree(s->bmap); 501 502 fail: 503 return ret; 504 } 505 506 static int vdi_reopen_prepare(BDRVReopenState *state, 507 BlockReopenQueue *queue, Error **errp) 508 { 509 return 0; 510 } 511 512 static int coroutine_fn vdi_co_block_status(BlockDriverState *bs, 513 bool want_zero, 514 int64_t offset, int64_t bytes, 515 int64_t *pnum, int64_t *map, 516 BlockDriverState **file) 517 { 518 BDRVVdiState *s = (BDRVVdiState *)bs->opaque; 519 size_t bmap_index = offset / s->block_size; 520 size_t index_in_block = offset % s->block_size; 521 uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]); 522 int result; 523 524 logout("%p, %" PRId64 ", %" PRId64 ", %p\n", bs, offset, bytes, pnum); 525 *pnum = MIN(s->block_size - index_in_block, bytes); 526 result = VDI_IS_ALLOCATED(bmap_entry); 527 if (!result) { 528 return 0; 529 } 530 531 *map = s->header.offset_data + (uint64_t)bmap_entry * s->block_size + 532 index_in_block; 533 *file = bs->file->bs; 534 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 535 } 536 537 static int coroutine_fn 538 vdi_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 539 QEMUIOVector *qiov, int flags) 540 { 541 BDRVVdiState *s = bs->opaque; 542 QEMUIOVector local_qiov; 543 uint32_t bmap_entry; 544 uint32_t block_index; 545 uint32_t offset_in_block; 546 uint32_t n_bytes; 547 uint64_t bytes_done = 0; 548 int ret = 0; 549 550 logout("\n"); 551 552 qemu_iovec_init(&local_qiov, qiov->niov); 553 554 while (ret >= 0 && bytes > 0) { 555 block_index = offset / s->block_size; 556 offset_in_block = offset % s->block_size; 557 n_bytes = MIN(bytes, s->block_size - offset_in_block); 558 559 logout("will read %u bytes starting at offset %" PRIu64 "\n", 560 n_bytes, offset); 561 562 /* prepare next AIO request */ 563 qemu_co_rwlock_rdlock(&s->bmap_lock); 564 bmap_entry = le32_to_cpu(s->bmap[block_index]); 565 qemu_co_rwlock_unlock(&s->bmap_lock); 566 if (!VDI_IS_ALLOCATED(bmap_entry)) { 567 /* Block not allocated, return zeros, no need to wait. */ 568 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes); 569 ret = 0; 570 } else { 571 uint64_t data_offset = s->header.offset_data + 572 (uint64_t)bmap_entry * s->block_size + 573 offset_in_block; 574 575 qemu_iovec_reset(&local_qiov); 576 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); 577 578 ret = bdrv_co_preadv(bs->file, data_offset, n_bytes, 579 &local_qiov, 0); 580 } 581 logout("%u bytes read\n", n_bytes); 582 583 bytes -= n_bytes; 584 offset += n_bytes; 585 bytes_done += n_bytes; 586 } 587 588 qemu_iovec_destroy(&local_qiov); 589 590 return ret; 591 } 592 593 static int coroutine_fn 594 vdi_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 595 QEMUIOVector *qiov, int flags) 596 { 597 BDRVVdiState *s = bs->opaque; 598 QEMUIOVector local_qiov; 599 uint32_t bmap_entry; 600 uint32_t block_index; 601 uint32_t offset_in_block; 602 uint32_t n_bytes; 603 uint64_t data_offset; 604 uint32_t bmap_first = VDI_UNALLOCATED; 605 uint32_t bmap_last = VDI_UNALLOCATED; 606 uint8_t *block = NULL; 607 uint64_t bytes_done = 0; 608 int ret = 0; 609 610 logout("\n"); 611 612 qemu_iovec_init(&local_qiov, qiov->niov); 613 614 while (ret >= 0 && bytes > 0) { 615 block_index = offset / s->block_size; 616 offset_in_block = offset % s->block_size; 617 n_bytes = MIN(bytes, s->block_size - offset_in_block); 618 619 logout("will write %u bytes starting at offset %" PRIu64 "\n", 620 n_bytes, offset); 621 622 /* prepare next AIO request */ 623 qemu_co_rwlock_rdlock(&s->bmap_lock); 624 bmap_entry = le32_to_cpu(s->bmap[block_index]); 625 if (!VDI_IS_ALLOCATED(bmap_entry)) { 626 /* Allocate new block and write to it. */ 627 uint64_t data_offset; 628 qemu_co_rwlock_upgrade(&s->bmap_lock); 629 bmap_entry = le32_to_cpu(s->bmap[block_index]); 630 if (VDI_IS_ALLOCATED(bmap_entry)) { 631 /* A concurrent allocation did the work for us. */ 632 qemu_co_rwlock_downgrade(&s->bmap_lock); 633 goto nonallocating_write; 634 } 635 636 bmap_entry = s->header.blocks_allocated; 637 s->bmap[block_index] = cpu_to_le32(bmap_entry); 638 s->header.blocks_allocated++; 639 data_offset = s->header.offset_data + 640 (uint64_t)bmap_entry * s->block_size; 641 if (block == NULL) { 642 block = g_malloc(s->block_size); 643 bmap_first = block_index; 644 } 645 bmap_last = block_index; 646 /* Copy data to be written to new block and zero unused parts. */ 647 memset(block, 0, offset_in_block); 648 qemu_iovec_to_buf(qiov, bytes_done, block + offset_in_block, 649 n_bytes); 650 memset(block + offset_in_block + n_bytes, 0, 651 s->block_size - n_bytes - offset_in_block); 652 653 /* Write the new block under CoRwLock write-side protection, 654 * so this full-cluster write does not overlap a partial write 655 * of the same cluster, issued from the "else" branch. 656 */ 657 ret = bdrv_pwrite(bs->file, data_offset, block, s->block_size); 658 qemu_co_rwlock_unlock(&s->bmap_lock); 659 } else { 660 nonallocating_write: 661 data_offset = s->header.offset_data + 662 (uint64_t)bmap_entry * s->block_size + 663 offset_in_block; 664 qemu_co_rwlock_unlock(&s->bmap_lock); 665 666 qemu_iovec_reset(&local_qiov); 667 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); 668 669 ret = bdrv_co_pwritev(bs->file, data_offset, n_bytes, 670 &local_qiov, 0); 671 } 672 673 bytes -= n_bytes; 674 offset += n_bytes; 675 bytes_done += n_bytes; 676 677 logout("%u bytes written\n", n_bytes); 678 } 679 680 qemu_iovec_destroy(&local_qiov); 681 682 logout("finished data write\n"); 683 if (ret < 0) { 684 return ret; 685 } 686 687 if (block) { 688 /* One or more new blocks were allocated. */ 689 VdiHeader *header = (VdiHeader *) block; 690 uint8_t *base; 691 uint64_t offset; 692 uint32_t n_sectors; 693 694 logout("now writing modified header\n"); 695 assert(VDI_IS_ALLOCATED(bmap_first)); 696 *header = s->header; 697 vdi_header_to_le(header); 698 ret = bdrv_write(bs->file, 0, block, 1); 699 g_free(block); 700 block = NULL; 701 702 if (ret < 0) { 703 return ret; 704 } 705 706 logout("now writing modified block map entry %u...%u\n", 707 bmap_first, bmap_last); 708 /* Write modified sectors from block map. */ 709 bmap_first /= (SECTOR_SIZE / sizeof(uint32_t)); 710 bmap_last /= (SECTOR_SIZE / sizeof(uint32_t)); 711 n_sectors = bmap_last - bmap_first + 1; 712 offset = s->bmap_sector + bmap_first; 713 base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE; 714 logout("will write %u block map sectors starting from entry %u\n", 715 n_sectors, bmap_first); 716 ret = bdrv_write(bs->file, offset, base, n_sectors); 717 } 718 719 return ret; 720 } 721 722 static int coroutine_fn vdi_co_do_create(BlockdevCreateOptions *create_options, 723 size_t block_size, Error **errp) 724 { 725 BlockdevCreateOptionsVdi *vdi_opts; 726 int ret = 0; 727 uint64_t bytes = 0; 728 uint32_t blocks; 729 uint32_t image_type; 730 VdiHeader header; 731 size_t i; 732 size_t bmap_size; 733 int64_t offset = 0; 734 BlockDriverState *bs_file = NULL; 735 BlockBackend *blk = NULL; 736 uint32_t *bmap = NULL; 737 738 assert(create_options->driver == BLOCKDEV_DRIVER_VDI); 739 vdi_opts = &create_options->u.vdi; 740 741 logout("\n"); 742 743 /* Validate options and set default values */ 744 bytes = vdi_opts->size; 745 746 if (!vdi_opts->has_preallocation) { 747 vdi_opts->preallocation = PREALLOC_MODE_OFF; 748 } 749 switch (vdi_opts->preallocation) { 750 case PREALLOC_MODE_OFF: 751 image_type = VDI_TYPE_DYNAMIC; 752 break; 753 case PREALLOC_MODE_METADATA: 754 image_type = VDI_TYPE_STATIC; 755 break; 756 default: 757 error_setg(errp, "Preallocation mode not supported for vdi"); 758 return -EINVAL; 759 } 760 761 #ifndef CONFIG_VDI_STATIC_IMAGE 762 if (image_type == VDI_TYPE_STATIC) { 763 ret = -ENOTSUP; 764 error_setg(errp, "Statically allocated images cannot be created in " 765 "this build"); 766 goto exit; 767 } 768 #endif 769 #ifndef CONFIG_VDI_BLOCK_SIZE 770 if (block_size != DEFAULT_CLUSTER_SIZE) { 771 ret = -ENOTSUP; 772 error_setg(errp, 773 "A non-default cluster size is not supported in this build"); 774 goto exit; 775 } 776 #endif 777 778 if (bytes > VDI_DISK_SIZE_MAX) { 779 ret = -ENOTSUP; 780 error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64 781 ", max supported is 0x%" PRIx64 ")", 782 bytes, VDI_DISK_SIZE_MAX); 783 goto exit; 784 } 785 786 /* Create BlockBackend to write to the image */ 787 bs_file = bdrv_open_blockdev_ref(vdi_opts->file, errp); 788 if (!bs_file) { 789 ret = -EIO; 790 goto exit; 791 } 792 793 blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL); 794 ret = blk_insert_bs(blk, bs_file, errp); 795 if (ret < 0) { 796 goto exit; 797 } 798 799 blk_set_allow_write_beyond_eof(blk, true); 800 801 /* We need enough blocks to store the given disk size, 802 so always round up. */ 803 blocks = DIV_ROUND_UP(bytes, block_size); 804 805 bmap_size = blocks * sizeof(uint32_t); 806 bmap_size = ROUND_UP(bmap_size, SECTOR_SIZE); 807 808 memset(&header, 0, sizeof(header)); 809 pstrcpy(header.text, sizeof(header.text), VDI_TEXT); 810 header.signature = VDI_SIGNATURE; 811 header.version = VDI_VERSION_1_1; 812 header.header_size = 0x180; 813 header.image_type = image_type; 814 header.offset_bmap = 0x200; 815 header.offset_data = 0x200 + bmap_size; 816 header.sector_size = SECTOR_SIZE; 817 header.disk_size = bytes; 818 header.block_size = block_size; 819 header.blocks_in_image = blocks; 820 if (image_type == VDI_TYPE_STATIC) { 821 header.blocks_allocated = blocks; 822 } 823 qemu_uuid_generate(&header.uuid_image); 824 qemu_uuid_generate(&header.uuid_last_snap); 825 /* There is no need to set header.uuid_link or header.uuid_parent here. */ 826 if (VDI_DEBUG) { 827 vdi_header_print(&header); 828 } 829 vdi_header_to_le(&header); 830 ret = blk_pwrite(blk, offset, &header, sizeof(header), 0); 831 if (ret < 0) { 832 error_setg(errp, "Error writing header"); 833 goto exit; 834 } 835 offset += sizeof(header); 836 837 if (bmap_size > 0) { 838 bmap = g_try_malloc0(bmap_size); 839 if (bmap == NULL) { 840 ret = -ENOMEM; 841 error_setg(errp, "Could not allocate bmap"); 842 goto exit; 843 } 844 for (i = 0; i < blocks; i++) { 845 if (image_type == VDI_TYPE_STATIC) { 846 bmap[i] = i; 847 } else { 848 bmap[i] = VDI_UNALLOCATED; 849 } 850 } 851 ret = blk_pwrite(blk, offset, bmap, bmap_size, 0); 852 if (ret < 0) { 853 error_setg(errp, "Error writing bmap"); 854 goto exit; 855 } 856 offset += bmap_size; 857 } 858 859 if (image_type == VDI_TYPE_STATIC) { 860 ret = blk_truncate(blk, offset + blocks * block_size, 861 PREALLOC_MODE_OFF, errp); 862 if (ret < 0) { 863 error_prepend(errp, "Failed to statically allocate file"); 864 goto exit; 865 } 866 } 867 868 exit: 869 blk_unref(blk); 870 bdrv_unref(bs_file); 871 g_free(bmap); 872 return ret; 873 } 874 875 static int coroutine_fn vdi_co_create(BlockdevCreateOptions *create_options, 876 Error **errp) 877 { 878 return vdi_co_do_create(create_options, DEFAULT_CLUSTER_SIZE, errp); 879 } 880 881 static int coroutine_fn vdi_co_create_opts(const char *filename, QemuOpts *opts, 882 Error **errp) 883 { 884 QDict *qdict = NULL; 885 BlockdevCreateOptions *create_options = NULL; 886 BlockDriverState *bs_file = NULL; 887 uint64_t block_size = DEFAULT_CLUSTER_SIZE; 888 bool is_static = false; 889 Visitor *v; 890 Error *local_err = NULL; 891 int ret; 892 893 /* Parse options and convert legacy syntax. 894 * 895 * Since CONFIG_VDI_BLOCK_SIZE is disabled by default, 896 * cluster-size is not part of the QAPI schema; therefore we have 897 * to parse it before creating the QAPI object. */ 898 #if defined(CONFIG_VDI_BLOCK_SIZE) 899 block_size = qemu_opt_get_size_del(opts, 900 BLOCK_OPT_CLUSTER_SIZE, 901 DEFAULT_CLUSTER_SIZE); 902 if (block_size < BDRV_SECTOR_SIZE || block_size > UINT32_MAX || 903 !is_power_of_2(block_size)) 904 { 905 error_setg(errp, "Invalid cluster size"); 906 ret = -EINVAL; 907 goto done; 908 } 909 #endif 910 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_STATIC, false)) { 911 is_static = true; 912 } 913 914 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vdi_create_opts, true); 915 916 /* Create and open the file (protocol layer) */ 917 ret = bdrv_create_file(filename, opts, errp); 918 if (ret < 0) { 919 goto done; 920 } 921 922 bs_file = bdrv_open(filename, NULL, NULL, 923 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 924 if (!bs_file) { 925 ret = -EIO; 926 goto done; 927 } 928 929 qdict_put_str(qdict, "driver", "vdi"); 930 qdict_put_str(qdict, "file", bs_file->node_name); 931 if (is_static) { 932 qdict_put_str(qdict, "preallocation", "metadata"); 933 } 934 935 /* Get the QAPI object */ 936 v = qobject_input_visitor_new_keyval(QOBJECT(qdict)); 937 visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err); 938 visit_free(v); 939 940 if (local_err) { 941 error_propagate(errp, local_err); 942 ret = -EINVAL; 943 goto done; 944 } 945 946 /* Silently round up size */ 947 assert(create_options->driver == BLOCKDEV_DRIVER_VDI); 948 create_options->u.vdi.size = ROUND_UP(create_options->u.vdi.size, 949 BDRV_SECTOR_SIZE); 950 951 /* Create the vdi image (format layer) */ 952 ret = vdi_co_do_create(create_options, block_size, errp); 953 done: 954 qobject_unref(qdict); 955 qapi_free_BlockdevCreateOptions(create_options); 956 bdrv_unref(bs_file); 957 return ret; 958 } 959 960 static void vdi_close(BlockDriverState *bs) 961 { 962 BDRVVdiState *s = bs->opaque; 963 964 qemu_vfree(s->bmap); 965 966 migrate_del_blocker(s->migration_blocker); 967 error_free(s->migration_blocker); 968 } 969 970 static QemuOptsList vdi_create_opts = { 971 .name = "vdi-create-opts", 972 .head = QTAILQ_HEAD_INITIALIZER(vdi_create_opts.head), 973 .desc = { 974 { 975 .name = BLOCK_OPT_SIZE, 976 .type = QEMU_OPT_SIZE, 977 .help = "Virtual disk size" 978 }, 979 #if defined(CONFIG_VDI_BLOCK_SIZE) 980 { 981 .name = BLOCK_OPT_CLUSTER_SIZE, 982 .type = QEMU_OPT_SIZE, 983 .help = "VDI cluster (block) size", 984 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) 985 }, 986 #endif 987 #if defined(CONFIG_VDI_STATIC_IMAGE) 988 { 989 .name = BLOCK_OPT_STATIC, 990 .type = QEMU_OPT_BOOL, 991 .help = "VDI static (pre-allocated) image", 992 .def_value_str = "off" 993 }, 994 #endif 995 /* TODO: An additional option to set UUID values might be useful. */ 996 { /* end of list */ } 997 } 998 }; 999 1000 static BlockDriver bdrv_vdi = { 1001 .format_name = "vdi", 1002 .instance_size = sizeof(BDRVVdiState), 1003 .bdrv_probe = vdi_probe, 1004 .bdrv_open = vdi_open, 1005 .bdrv_close = vdi_close, 1006 .bdrv_reopen_prepare = vdi_reopen_prepare, 1007 .bdrv_child_perm = bdrv_format_default_perms, 1008 .bdrv_co_create = vdi_co_create, 1009 .bdrv_co_create_opts = vdi_co_create_opts, 1010 .bdrv_has_zero_init = bdrv_has_zero_init_1, 1011 .bdrv_co_block_status = vdi_co_block_status, 1012 .bdrv_make_empty = vdi_make_empty, 1013 1014 .bdrv_co_preadv = vdi_co_preadv, 1015 #if defined(CONFIG_VDI_WRITE) 1016 .bdrv_co_pwritev = vdi_co_pwritev, 1017 #endif 1018 1019 .bdrv_get_info = vdi_get_info, 1020 1021 .create_opts = &vdi_create_opts, 1022 .bdrv_co_check = vdi_co_check, 1023 }; 1024 1025 static void bdrv_vdi_init(void) 1026 { 1027 logout("\n"); 1028 bdrv_register(&bdrv_vdi); 1029 } 1030 1031 block_init(bdrv_vdi_init); 1032