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 "qemu/units.h" 54 #include "qapi/error.h" 55 #include "qapi/qobject-input-visitor.h" 56 #include "qapi/qapi-visit-block-core.h" 57 #include "block/block_int.h" 58 #include "block/qdict.h" 59 #include "sysemu/block-backend.h" 60 #include "qemu/module.h" 61 #include "qemu/option.h" 62 #include "qemu/bswap.h" 63 #include "migration/blocker.h" 64 #include "qemu/coroutine.h" 65 #include "qemu/cutils.h" 66 #include "qemu/uuid.h" 67 68 /* Code configuration options. */ 69 70 /* Enable debug messages. */ 71 //~ #define CONFIG_VDI_DEBUG 72 73 /* Support write operations on VDI images. */ 74 #define CONFIG_VDI_WRITE 75 76 /* Support non-standard block (cluster) size. This is untested. 77 * Maybe it will be needed for very large images. 78 */ 79 //~ #define CONFIG_VDI_BLOCK_SIZE 80 81 /* Support static (fixed, pre-allocated) images. */ 82 #define CONFIG_VDI_STATIC_IMAGE 83 84 /* Command line option for static images. */ 85 #define BLOCK_OPT_STATIC "static" 86 87 #define SECTOR_SIZE 512 88 #define DEFAULT_CLUSTER_SIZE 1048576 89 /* Note: can't use 1 * MiB, because it's passed to stringify() */ 90 91 #if defined(CONFIG_VDI_DEBUG) 92 #define VDI_DEBUG 1 93 #else 94 #define VDI_DEBUG 0 95 #endif 96 97 #define logout(fmt, ...) \ 98 do { \ 99 if (VDI_DEBUG) { \ 100 fprintf(stderr, "vdi\t%-24s" fmt, __func__, ##__VA_ARGS__); \ 101 } \ 102 } while (0) 103 104 /* Image signature. */ 105 #define VDI_SIGNATURE 0xbeda107f 106 107 /* Image version. */ 108 #define VDI_VERSION_1_1 0x00010001 109 110 /* Image type. */ 111 #define VDI_TYPE_DYNAMIC 1 112 #define VDI_TYPE_STATIC 2 113 114 /* Innotek / SUN images use these strings in header.text: 115 * "<<< innotek VirtualBox Disk Image >>>\n" 116 * "<<< Sun xVM VirtualBox Disk Image >>>\n" 117 * "<<< Sun VirtualBox Disk Image >>>\n" 118 * The value does not matter, so QEMU created images use a different text. 119 */ 120 #define VDI_TEXT "<<< QEMU VM Virtual Disk Image >>>\n" 121 122 /* A never-allocated block; semantically arbitrary content. */ 123 #define VDI_UNALLOCATED 0xffffffffU 124 125 /* A discarded (no longer allocated) block; semantically zero-filled. */ 126 #define VDI_DISCARDED 0xfffffffeU 127 128 #define VDI_IS_ALLOCATED(X) ((X) < VDI_DISCARDED) 129 130 /* The bmap will take up VDI_BLOCKS_IN_IMAGE_MAX * sizeof(uint32_t) bytes; since 131 * the bmap is read and written in a single operation, its size needs to be 132 * limited to INT_MAX; furthermore, when opening an image, the bmap size is 133 * rounded up to be aligned on BDRV_SECTOR_SIZE. 134 * Therefore this should satisfy the following: 135 * VDI_BLOCKS_IN_IMAGE_MAX * sizeof(uint32_t) + BDRV_SECTOR_SIZE == INT_MAX + 1 136 * (INT_MAX + 1 is the first value not representable as an int) 137 * This guarantees that any value below or equal to the constant will, when 138 * multiplied by sizeof(uint32_t) and rounded up to a BDRV_SECTOR_SIZE boundary, 139 * still be below or equal to INT_MAX. */ 140 #define VDI_BLOCKS_IN_IMAGE_MAX \ 141 ((unsigned)((INT_MAX + 1u - BDRV_SECTOR_SIZE) / sizeof(uint32_t))) 142 #define VDI_DISK_SIZE_MAX ((uint64_t)VDI_BLOCKS_IN_IMAGE_MAX * \ 143 (uint64_t)DEFAULT_CLUSTER_SIZE) 144 145 static QemuOptsList vdi_create_opts; 146 147 typedef struct { 148 char text[0x40]; 149 uint32_t signature; 150 uint32_t version; 151 uint32_t header_size; 152 uint32_t image_type; 153 uint32_t image_flags; 154 char description[256]; 155 uint32_t offset_bmap; 156 uint32_t offset_data; 157 uint32_t cylinders; /* disk geometry, unused here */ 158 uint32_t heads; /* disk geometry, unused here */ 159 uint32_t sectors; /* disk geometry, unused here */ 160 uint32_t sector_size; 161 uint32_t unused1; 162 uint64_t disk_size; 163 uint32_t block_size; 164 uint32_t block_extra; /* unused here */ 165 uint32_t blocks_in_image; 166 uint32_t blocks_allocated; 167 QemuUUID uuid_image; 168 QemuUUID uuid_last_snap; 169 QemuUUID uuid_link; 170 QemuUUID uuid_parent; 171 uint64_t unused2[7]; 172 } QEMU_PACKED VdiHeader; 173 174 QEMU_BUILD_BUG_ON(sizeof(VdiHeader) != 512); 175 176 typedef struct { 177 /* The block map entries are little endian (even in memory). */ 178 uint32_t *bmap; 179 /* Size of block (bytes). */ 180 uint32_t block_size; 181 /* First sector of block map. */ 182 uint32_t bmap_sector; 183 /* VDI header (converted to host endianness). */ 184 VdiHeader header; 185 186 CoRwlock bmap_lock; 187 188 Error *migration_blocker; 189 } BDRVVdiState; 190 191 static void vdi_header_to_cpu(VdiHeader *header) 192 { 193 header->signature = le32_to_cpu(header->signature); 194 header->version = le32_to_cpu(header->version); 195 header->header_size = le32_to_cpu(header->header_size); 196 header->image_type = le32_to_cpu(header->image_type); 197 header->image_flags = le32_to_cpu(header->image_flags); 198 header->offset_bmap = le32_to_cpu(header->offset_bmap); 199 header->offset_data = le32_to_cpu(header->offset_data); 200 header->cylinders = le32_to_cpu(header->cylinders); 201 header->heads = le32_to_cpu(header->heads); 202 header->sectors = le32_to_cpu(header->sectors); 203 header->sector_size = le32_to_cpu(header->sector_size); 204 header->disk_size = le64_to_cpu(header->disk_size); 205 header->block_size = le32_to_cpu(header->block_size); 206 header->block_extra = le32_to_cpu(header->block_extra); 207 header->blocks_in_image = le32_to_cpu(header->blocks_in_image); 208 header->blocks_allocated = le32_to_cpu(header->blocks_allocated); 209 header->uuid_image = qemu_uuid_bswap(header->uuid_image); 210 header->uuid_last_snap = qemu_uuid_bswap(header->uuid_last_snap); 211 header->uuid_link = qemu_uuid_bswap(header->uuid_link); 212 header->uuid_parent = qemu_uuid_bswap(header->uuid_parent); 213 } 214 215 static void vdi_header_to_le(VdiHeader *header) 216 { 217 header->signature = cpu_to_le32(header->signature); 218 header->version = cpu_to_le32(header->version); 219 header->header_size = cpu_to_le32(header->header_size); 220 header->image_type = cpu_to_le32(header->image_type); 221 header->image_flags = cpu_to_le32(header->image_flags); 222 header->offset_bmap = cpu_to_le32(header->offset_bmap); 223 header->offset_data = cpu_to_le32(header->offset_data); 224 header->cylinders = cpu_to_le32(header->cylinders); 225 header->heads = cpu_to_le32(header->heads); 226 header->sectors = cpu_to_le32(header->sectors); 227 header->sector_size = cpu_to_le32(header->sector_size); 228 header->disk_size = cpu_to_le64(header->disk_size); 229 header->block_size = cpu_to_le32(header->block_size); 230 header->block_extra = cpu_to_le32(header->block_extra); 231 header->blocks_in_image = cpu_to_le32(header->blocks_in_image); 232 header->blocks_allocated = cpu_to_le32(header->blocks_allocated); 233 header->uuid_image = qemu_uuid_bswap(header->uuid_image); 234 header->uuid_last_snap = qemu_uuid_bswap(header->uuid_last_snap); 235 header->uuid_link = qemu_uuid_bswap(header->uuid_link); 236 header->uuid_parent = qemu_uuid_bswap(header->uuid_parent); 237 } 238 239 static void vdi_header_print(VdiHeader *header) 240 { 241 char uuidstr[37]; 242 QemuUUID uuid; 243 logout("text %s", header->text); 244 logout("signature 0x%08x\n", header->signature); 245 logout("header size 0x%04x\n", header->header_size); 246 logout("image type 0x%04x\n", header->image_type); 247 logout("image flags 0x%04x\n", header->image_flags); 248 logout("description %s\n", header->description); 249 logout("offset bmap 0x%04x\n", header->offset_bmap); 250 logout("offset data 0x%04x\n", header->offset_data); 251 logout("cylinders 0x%04x\n", header->cylinders); 252 logout("heads 0x%04x\n", header->heads); 253 logout("sectors 0x%04x\n", header->sectors); 254 logout("sector size 0x%04x\n", header->sector_size); 255 logout("image size 0x%" PRIx64 " B (%" PRIu64 " MiB)\n", 256 header->disk_size, header->disk_size / MiB); 257 logout("block size 0x%04x\n", header->block_size); 258 logout("block extra 0x%04x\n", header->block_extra); 259 logout("blocks tot. 0x%04x\n", header->blocks_in_image); 260 logout("blocks all. 0x%04x\n", header->blocks_allocated); 261 uuid = header->uuid_image; 262 qemu_uuid_unparse(&uuid, uuidstr); 263 logout("uuid image %s\n", uuidstr); 264 uuid = header->uuid_last_snap; 265 qemu_uuid_unparse(&uuid, uuidstr); 266 logout("uuid snap %s\n", uuidstr); 267 uuid = header->uuid_link; 268 qemu_uuid_unparse(&uuid, uuidstr); 269 logout("uuid link %s\n", uuidstr); 270 uuid = header->uuid_parent; 271 qemu_uuid_unparse(&uuid, uuidstr); 272 logout("uuid parent %s\n", uuidstr); 273 } 274 275 static int coroutine_fn vdi_co_check(BlockDriverState *bs, BdrvCheckResult *res, 276 BdrvCheckMode fix) 277 { 278 /* TODO: additional checks possible. */ 279 BDRVVdiState *s = (BDRVVdiState *)bs->opaque; 280 uint32_t blocks_allocated = 0; 281 uint32_t block; 282 uint32_t *bmap; 283 logout("\n"); 284 285 if (fix) { 286 return -ENOTSUP; 287 } 288 289 bmap = g_try_new(uint32_t, s->header.blocks_in_image); 290 if (s->header.blocks_in_image && bmap == NULL) { 291 res->check_errors++; 292 return -ENOMEM; 293 } 294 295 memset(bmap, 0xff, s->header.blocks_in_image * sizeof(uint32_t)); 296 297 /* Check block map and value of blocks_allocated. */ 298 for (block = 0; block < s->header.blocks_in_image; block++) { 299 uint32_t bmap_entry = le32_to_cpu(s->bmap[block]); 300 if (VDI_IS_ALLOCATED(bmap_entry)) { 301 if (bmap_entry < s->header.blocks_in_image) { 302 blocks_allocated++; 303 if (!VDI_IS_ALLOCATED(bmap[bmap_entry])) { 304 bmap[bmap_entry] = bmap_entry; 305 } else { 306 fprintf(stderr, "ERROR: block index %" PRIu32 307 " also used by %" PRIu32 "\n", bmap[bmap_entry], bmap_entry); 308 res->corruptions++; 309 } 310 } else { 311 fprintf(stderr, "ERROR: block index %" PRIu32 312 " too large, is %" PRIu32 "\n", block, bmap_entry); 313 res->corruptions++; 314 } 315 } 316 } 317 if (blocks_allocated != s->header.blocks_allocated) { 318 fprintf(stderr, "ERROR: allocated blocks mismatch, is %" PRIu32 319 ", should be %" PRIu32 "\n", 320 blocks_allocated, s->header.blocks_allocated); 321 res->corruptions++; 322 } 323 324 g_free(bmap); 325 326 return 0; 327 } 328 329 static int vdi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 330 { 331 /* TODO: vdi_get_info would be needed for machine snapshots. 332 vm_state_offset is still missing. */ 333 BDRVVdiState *s = (BDRVVdiState *)bs->opaque; 334 logout("\n"); 335 bdi->cluster_size = s->block_size; 336 bdi->vm_state_offset = 0; 337 bdi->unallocated_blocks_are_zero = true; 338 return 0; 339 } 340 341 static int vdi_make_empty(BlockDriverState *bs) 342 { 343 /* TODO: missing code. */ 344 logout("\n"); 345 /* The return value for missing code must be 0, see block.c. */ 346 return 0; 347 } 348 349 static int vdi_probe(const uint8_t *buf, int buf_size, const char *filename) 350 { 351 const VdiHeader *header = (const VdiHeader *)buf; 352 int ret = 0; 353 354 logout("\n"); 355 356 if (buf_size < sizeof(*header)) { 357 /* Header too small, no VDI. */ 358 } else if (le32_to_cpu(header->signature) == VDI_SIGNATURE) { 359 ret = 100; 360 } 361 362 if (ret == 0) { 363 logout("no vdi image\n"); 364 } else { 365 logout("%s", header->text); 366 } 367 368 return ret; 369 } 370 371 static int vdi_open(BlockDriverState *bs, QDict *options, int flags, 372 Error **errp) 373 { 374 BDRVVdiState *s = bs->opaque; 375 VdiHeader header; 376 size_t bmap_size; 377 int ret; 378 Error *local_err = NULL; 379 QemuUUID uuid_link, uuid_parent; 380 381 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, 382 false, errp); 383 if (!bs->file) { 384 return -EINVAL; 385 } 386 387 logout("\n"); 388 389 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 390 if (ret < 0) { 391 goto fail; 392 } 393 394 vdi_header_to_cpu(&header); 395 if (VDI_DEBUG) { 396 vdi_header_print(&header); 397 } 398 399 if (header.disk_size > VDI_DISK_SIZE_MAX) { 400 error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64 401 ", max supported is 0x%" PRIx64 ")", 402 header.disk_size, VDI_DISK_SIZE_MAX); 403 ret = -ENOTSUP; 404 goto fail; 405 } 406 407 uuid_link = header.uuid_link; 408 uuid_parent = header.uuid_parent; 409 410 if (header.disk_size % SECTOR_SIZE != 0) { 411 /* 'VBoxManage convertfromraw' can create images with odd disk sizes. 412 We accept them but round the disk size to the next multiple of 413 SECTOR_SIZE. */ 414 logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size); 415 header.disk_size = ROUND_UP(header.disk_size, SECTOR_SIZE); 416 } 417 418 if (header.signature != VDI_SIGNATURE) { 419 error_setg(errp, "Image not in VDI format (bad signature %08" PRIx32 420 ")", header.signature); 421 ret = -EINVAL; 422 goto fail; 423 } else if (header.version != VDI_VERSION_1_1) { 424 error_setg(errp, "unsupported VDI image (version %" PRIu32 ".%" PRIu32 425 ")", header.version >> 16, header.version & 0xffff); 426 ret = -ENOTSUP; 427 goto fail; 428 } else if (header.offset_bmap % SECTOR_SIZE != 0) { 429 /* We only support block maps which start on a sector boundary. */ 430 error_setg(errp, "unsupported VDI image (unaligned block map offset " 431 "0x%" PRIx32 ")", header.offset_bmap); 432 ret = -ENOTSUP; 433 goto fail; 434 } else if (header.offset_data % SECTOR_SIZE != 0) { 435 /* We only support data blocks which start on a sector boundary. */ 436 error_setg(errp, "unsupported VDI image (unaligned data offset 0x%" 437 PRIx32 ")", header.offset_data); 438 ret = -ENOTSUP; 439 goto fail; 440 } else if (header.sector_size != SECTOR_SIZE) { 441 error_setg(errp, "unsupported VDI image (sector size %" PRIu32 442 " is not %u)", header.sector_size, SECTOR_SIZE); 443 ret = -ENOTSUP; 444 goto fail; 445 } else if (header.block_size != DEFAULT_CLUSTER_SIZE) { 446 error_setg(errp, "unsupported VDI image (block size %" PRIu32 447 " is not %" PRIu32 ")", 448 header.block_size, DEFAULT_CLUSTER_SIZE); 449 ret = -ENOTSUP; 450 goto fail; 451 } else if (header.disk_size > 452 (uint64_t)header.blocks_in_image * header.block_size) { 453 error_setg(errp, "unsupported VDI image (disk size %" PRIu64 ", " 454 "image bitmap has room for %" PRIu64 ")", 455 header.disk_size, 456 (uint64_t)header.blocks_in_image * header.block_size); 457 ret = -ENOTSUP; 458 goto fail; 459 } else if (!qemu_uuid_is_null(&uuid_link)) { 460 error_setg(errp, "unsupported VDI image (non-NULL link UUID)"); 461 ret = -ENOTSUP; 462 goto fail; 463 } else if (!qemu_uuid_is_null(&uuid_parent)) { 464 error_setg(errp, "unsupported VDI image (non-NULL parent UUID)"); 465 ret = -ENOTSUP; 466 goto fail; 467 } else if (header.blocks_in_image > VDI_BLOCKS_IN_IMAGE_MAX) { 468 error_setg(errp, "unsupported VDI image " 469 "(too many blocks %u, max is %u)", 470 header.blocks_in_image, VDI_BLOCKS_IN_IMAGE_MAX); 471 ret = -ENOTSUP; 472 goto fail; 473 } 474 475 bs->total_sectors = header.disk_size / SECTOR_SIZE; 476 477 s->block_size = header.block_size; 478 s->bmap_sector = header.offset_bmap / SECTOR_SIZE; 479 s->header = header; 480 481 bmap_size = header.blocks_in_image * sizeof(uint32_t); 482 bmap_size = DIV_ROUND_UP(bmap_size, SECTOR_SIZE); 483 s->bmap = qemu_try_blockalign(bs->file->bs, bmap_size * SECTOR_SIZE); 484 if (s->bmap == NULL) { 485 ret = -ENOMEM; 486 goto fail; 487 } 488 489 ret = bdrv_pread(bs->file, header.offset_bmap, s->bmap, 490 bmap_size * SECTOR_SIZE); 491 if (ret < 0) { 492 goto fail_free_bmap; 493 } 494 495 /* Disable migration when vdi images are used */ 496 error_setg(&s->migration_blocker, "The vdi format used by node '%s' " 497 "does not support live migration", 498 bdrv_get_device_or_node_name(bs)); 499 ret = migrate_add_blocker(s->migration_blocker, &local_err); 500 if (local_err) { 501 error_propagate(errp, local_err); 502 error_free(s->migration_blocker); 503 goto fail_free_bmap; 504 } 505 506 qemu_co_rwlock_init(&s->bmap_lock); 507 508 return 0; 509 510 fail_free_bmap: 511 qemu_vfree(s->bmap); 512 513 fail: 514 return ret; 515 } 516 517 static int vdi_reopen_prepare(BDRVReopenState *state, 518 BlockReopenQueue *queue, Error **errp) 519 { 520 return 0; 521 } 522 523 static int coroutine_fn vdi_co_block_status(BlockDriverState *bs, 524 bool want_zero, 525 int64_t offset, int64_t bytes, 526 int64_t *pnum, int64_t *map, 527 BlockDriverState **file) 528 { 529 BDRVVdiState *s = (BDRVVdiState *)bs->opaque; 530 size_t bmap_index = offset / s->block_size; 531 size_t index_in_block = offset % s->block_size; 532 uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]); 533 int result; 534 535 logout("%p, %" PRId64 ", %" PRId64 ", %p\n", bs, offset, bytes, pnum); 536 *pnum = MIN(s->block_size - index_in_block, bytes); 537 result = VDI_IS_ALLOCATED(bmap_entry); 538 if (!result) { 539 return 0; 540 } 541 542 *map = s->header.offset_data + (uint64_t)bmap_entry * s->block_size + 543 index_in_block; 544 *file = bs->file->bs; 545 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | 546 (s->header.image_type == VDI_TYPE_STATIC ? BDRV_BLOCK_RECURSE : 0); 547 } 548 549 static int coroutine_fn 550 vdi_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 551 QEMUIOVector *qiov, int flags) 552 { 553 BDRVVdiState *s = bs->opaque; 554 QEMUIOVector local_qiov; 555 uint32_t bmap_entry; 556 uint32_t block_index; 557 uint32_t offset_in_block; 558 uint32_t n_bytes; 559 uint64_t bytes_done = 0; 560 int ret = 0; 561 562 logout("\n"); 563 564 qemu_iovec_init(&local_qiov, qiov->niov); 565 566 while (ret >= 0 && bytes > 0) { 567 block_index = offset / s->block_size; 568 offset_in_block = offset % s->block_size; 569 n_bytes = MIN(bytes, s->block_size - offset_in_block); 570 571 logout("will read %u bytes starting at offset %" PRIu64 "\n", 572 n_bytes, offset); 573 574 /* prepare next AIO request */ 575 qemu_co_rwlock_rdlock(&s->bmap_lock); 576 bmap_entry = le32_to_cpu(s->bmap[block_index]); 577 qemu_co_rwlock_unlock(&s->bmap_lock); 578 if (!VDI_IS_ALLOCATED(bmap_entry)) { 579 /* Block not allocated, return zeros, no need to wait. */ 580 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes); 581 ret = 0; 582 } else { 583 uint64_t data_offset = s->header.offset_data + 584 (uint64_t)bmap_entry * s->block_size + 585 offset_in_block; 586 587 qemu_iovec_reset(&local_qiov); 588 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); 589 590 ret = bdrv_co_preadv(bs->file, data_offset, n_bytes, 591 &local_qiov, 0); 592 } 593 logout("%u bytes read\n", n_bytes); 594 595 bytes -= n_bytes; 596 offset += n_bytes; 597 bytes_done += n_bytes; 598 } 599 600 qemu_iovec_destroy(&local_qiov); 601 602 return ret; 603 } 604 605 static int coroutine_fn 606 vdi_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 607 QEMUIOVector *qiov, int flags) 608 { 609 BDRVVdiState *s = bs->opaque; 610 QEMUIOVector local_qiov; 611 uint32_t bmap_entry; 612 uint32_t block_index; 613 uint32_t offset_in_block; 614 uint32_t n_bytes; 615 uint64_t data_offset; 616 uint32_t bmap_first = VDI_UNALLOCATED; 617 uint32_t bmap_last = VDI_UNALLOCATED; 618 uint8_t *block = NULL; 619 uint64_t bytes_done = 0; 620 int ret = 0; 621 622 logout("\n"); 623 624 qemu_iovec_init(&local_qiov, qiov->niov); 625 626 while (ret >= 0 && bytes > 0) { 627 block_index = offset / s->block_size; 628 offset_in_block = offset % s->block_size; 629 n_bytes = MIN(bytes, s->block_size - offset_in_block); 630 631 logout("will write %u bytes starting at offset %" PRIu64 "\n", 632 n_bytes, offset); 633 634 /* prepare next AIO request */ 635 qemu_co_rwlock_rdlock(&s->bmap_lock); 636 bmap_entry = le32_to_cpu(s->bmap[block_index]); 637 if (!VDI_IS_ALLOCATED(bmap_entry)) { 638 /* Allocate new block and write to it. */ 639 uint64_t data_offset; 640 qemu_co_rwlock_upgrade(&s->bmap_lock); 641 bmap_entry = le32_to_cpu(s->bmap[block_index]); 642 if (VDI_IS_ALLOCATED(bmap_entry)) { 643 /* A concurrent allocation did the work for us. */ 644 qemu_co_rwlock_downgrade(&s->bmap_lock); 645 goto nonallocating_write; 646 } 647 648 bmap_entry = s->header.blocks_allocated; 649 s->bmap[block_index] = cpu_to_le32(bmap_entry); 650 s->header.blocks_allocated++; 651 data_offset = s->header.offset_data + 652 (uint64_t)bmap_entry * s->block_size; 653 if (block == NULL) { 654 block = g_malloc(s->block_size); 655 bmap_first = block_index; 656 } 657 bmap_last = block_index; 658 /* Copy data to be written to new block and zero unused parts. */ 659 memset(block, 0, offset_in_block); 660 qemu_iovec_to_buf(qiov, bytes_done, block + offset_in_block, 661 n_bytes); 662 memset(block + offset_in_block + n_bytes, 0, 663 s->block_size - n_bytes - offset_in_block); 664 665 /* Write the new block under CoRwLock write-side protection, 666 * so this full-cluster write does not overlap a partial write 667 * of the same cluster, issued from the "else" branch. 668 */ 669 ret = bdrv_pwrite(bs->file, data_offset, block, s->block_size); 670 qemu_co_rwlock_unlock(&s->bmap_lock); 671 } else { 672 nonallocating_write: 673 data_offset = s->header.offset_data + 674 (uint64_t)bmap_entry * s->block_size + 675 offset_in_block; 676 qemu_co_rwlock_unlock(&s->bmap_lock); 677 678 qemu_iovec_reset(&local_qiov); 679 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); 680 681 ret = bdrv_co_pwritev(bs->file, data_offset, n_bytes, 682 &local_qiov, 0); 683 } 684 685 bytes -= n_bytes; 686 offset += n_bytes; 687 bytes_done += n_bytes; 688 689 logout("%u bytes written\n", n_bytes); 690 } 691 692 qemu_iovec_destroy(&local_qiov); 693 694 logout("finished data write\n"); 695 if (ret < 0) { 696 return ret; 697 } 698 699 if (block) { 700 /* One or more new blocks were allocated. */ 701 VdiHeader *header = (VdiHeader *) block; 702 uint8_t *base; 703 uint64_t offset; 704 uint32_t n_sectors; 705 706 logout("now writing modified header\n"); 707 assert(VDI_IS_ALLOCATED(bmap_first)); 708 *header = s->header; 709 vdi_header_to_le(header); 710 ret = bdrv_pwrite(bs->file, 0, block, sizeof(VdiHeader)); 711 g_free(block); 712 block = NULL; 713 714 if (ret < 0) { 715 return ret; 716 } 717 718 logout("now writing modified block map entry %u...%u\n", 719 bmap_first, bmap_last); 720 /* Write modified sectors from block map. */ 721 bmap_first /= (SECTOR_SIZE / sizeof(uint32_t)); 722 bmap_last /= (SECTOR_SIZE / sizeof(uint32_t)); 723 n_sectors = bmap_last - bmap_first + 1; 724 offset = s->bmap_sector + bmap_first; 725 base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE; 726 logout("will write %u block map sectors starting from entry %u\n", 727 n_sectors, bmap_first); 728 ret = bdrv_pwrite(bs->file, offset * SECTOR_SIZE, base, 729 n_sectors * SECTOR_SIZE); 730 } 731 732 return ret < 0 ? ret : 0; 733 } 734 735 static int coroutine_fn vdi_co_do_create(BlockdevCreateOptions *create_options, 736 size_t block_size, Error **errp) 737 { 738 BlockdevCreateOptionsVdi *vdi_opts; 739 int ret = 0; 740 uint64_t bytes = 0; 741 uint32_t blocks; 742 uint32_t image_type; 743 VdiHeader header; 744 size_t i; 745 size_t bmap_size; 746 int64_t offset = 0; 747 BlockDriverState *bs_file = NULL; 748 BlockBackend *blk = NULL; 749 uint32_t *bmap = NULL; 750 QemuUUID uuid; 751 752 assert(create_options->driver == BLOCKDEV_DRIVER_VDI); 753 vdi_opts = &create_options->u.vdi; 754 755 logout("\n"); 756 757 /* Validate options and set default values */ 758 bytes = vdi_opts->size; 759 760 if (!vdi_opts->has_preallocation) { 761 vdi_opts->preallocation = PREALLOC_MODE_OFF; 762 } 763 switch (vdi_opts->preallocation) { 764 case PREALLOC_MODE_OFF: 765 image_type = VDI_TYPE_DYNAMIC; 766 break; 767 case PREALLOC_MODE_METADATA: 768 image_type = VDI_TYPE_STATIC; 769 break; 770 default: 771 error_setg(errp, "Preallocation mode not supported for vdi"); 772 return -EINVAL; 773 } 774 775 #ifndef CONFIG_VDI_STATIC_IMAGE 776 if (image_type == VDI_TYPE_STATIC) { 777 ret = -ENOTSUP; 778 error_setg(errp, "Statically allocated images cannot be created in " 779 "this build"); 780 goto exit; 781 } 782 #endif 783 #ifndef CONFIG_VDI_BLOCK_SIZE 784 if (block_size != DEFAULT_CLUSTER_SIZE) { 785 ret = -ENOTSUP; 786 error_setg(errp, 787 "A non-default cluster size is not supported in this build"); 788 goto exit; 789 } 790 #endif 791 792 if (bytes > VDI_DISK_SIZE_MAX) { 793 ret = -ENOTSUP; 794 error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64 795 ", max supported is 0x%" PRIx64 ")", 796 bytes, VDI_DISK_SIZE_MAX); 797 goto exit; 798 } 799 800 /* Create BlockBackend to write to the image */ 801 bs_file = bdrv_open_blockdev_ref(vdi_opts->file, errp); 802 if (!bs_file) { 803 ret = -EIO; 804 goto exit; 805 } 806 807 blk = blk_new(bdrv_get_aio_context(bs_file), 808 BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL); 809 ret = blk_insert_bs(blk, bs_file, errp); 810 if (ret < 0) { 811 goto exit; 812 } 813 814 blk_set_allow_write_beyond_eof(blk, true); 815 816 /* We need enough blocks to store the given disk size, 817 so always round up. */ 818 blocks = DIV_ROUND_UP(bytes, block_size); 819 820 bmap_size = blocks * sizeof(uint32_t); 821 bmap_size = ROUND_UP(bmap_size, SECTOR_SIZE); 822 823 memset(&header, 0, sizeof(header)); 824 pstrcpy(header.text, sizeof(header.text), VDI_TEXT); 825 header.signature = VDI_SIGNATURE; 826 header.version = VDI_VERSION_1_1; 827 header.header_size = 0x180; 828 header.image_type = image_type; 829 header.offset_bmap = 0x200; 830 header.offset_data = 0x200 + bmap_size; 831 header.sector_size = SECTOR_SIZE; 832 header.disk_size = bytes; 833 header.block_size = block_size; 834 header.blocks_in_image = blocks; 835 if (image_type == VDI_TYPE_STATIC) { 836 header.blocks_allocated = blocks; 837 } 838 qemu_uuid_generate(&uuid); 839 header.uuid_image = uuid; 840 qemu_uuid_generate(&uuid); 841 header.uuid_last_snap = uuid; 842 /* There is no need to set header.uuid_link or header.uuid_parent here. */ 843 if (VDI_DEBUG) { 844 vdi_header_print(&header); 845 } 846 vdi_header_to_le(&header); 847 ret = blk_pwrite(blk, offset, &header, sizeof(header), 0); 848 if (ret < 0) { 849 error_setg(errp, "Error writing header"); 850 goto exit; 851 } 852 offset += sizeof(header); 853 854 if (bmap_size > 0) { 855 bmap = g_try_malloc0(bmap_size); 856 if (bmap == NULL) { 857 ret = -ENOMEM; 858 error_setg(errp, "Could not allocate bmap"); 859 goto exit; 860 } 861 for (i = 0; i < blocks; i++) { 862 if (image_type == VDI_TYPE_STATIC) { 863 bmap[i] = i; 864 } else { 865 bmap[i] = VDI_UNALLOCATED; 866 } 867 } 868 ret = blk_pwrite(blk, offset, bmap, bmap_size, 0); 869 if (ret < 0) { 870 error_setg(errp, "Error writing bmap"); 871 goto exit; 872 } 873 offset += bmap_size; 874 } 875 876 if (image_type == VDI_TYPE_STATIC) { 877 ret = blk_truncate(blk, offset + blocks * block_size, false, 878 PREALLOC_MODE_OFF, errp); 879 if (ret < 0) { 880 error_prepend(errp, "Failed to statically allocate file"); 881 goto exit; 882 } 883 } 884 885 ret = 0; 886 exit: 887 blk_unref(blk); 888 bdrv_unref(bs_file); 889 g_free(bmap); 890 return ret; 891 } 892 893 static int coroutine_fn vdi_co_create(BlockdevCreateOptions *create_options, 894 Error **errp) 895 { 896 return vdi_co_do_create(create_options, DEFAULT_CLUSTER_SIZE, errp); 897 } 898 899 static int coroutine_fn vdi_co_create_opts(const char *filename, QemuOpts *opts, 900 Error **errp) 901 { 902 QDict *qdict = NULL; 903 BlockdevCreateOptions *create_options = NULL; 904 BlockDriverState *bs_file = NULL; 905 uint64_t block_size = DEFAULT_CLUSTER_SIZE; 906 bool is_static = false; 907 Visitor *v; 908 Error *local_err = NULL; 909 int ret; 910 911 /* Parse options and convert legacy syntax. 912 * 913 * Since CONFIG_VDI_BLOCK_SIZE is disabled by default, 914 * cluster-size is not part of the QAPI schema; therefore we have 915 * to parse it before creating the QAPI object. */ 916 #if defined(CONFIG_VDI_BLOCK_SIZE) 917 block_size = qemu_opt_get_size_del(opts, 918 BLOCK_OPT_CLUSTER_SIZE, 919 DEFAULT_CLUSTER_SIZE); 920 if (block_size < BDRV_SECTOR_SIZE || block_size > UINT32_MAX || 921 !is_power_of_2(block_size)) 922 { 923 error_setg(errp, "Invalid cluster size"); 924 ret = -EINVAL; 925 goto done; 926 } 927 #endif 928 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_STATIC, false)) { 929 is_static = true; 930 } 931 932 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vdi_create_opts, true); 933 934 /* Create and open the file (protocol layer) */ 935 ret = bdrv_create_file(filename, opts, errp); 936 if (ret < 0) { 937 goto done; 938 } 939 940 bs_file = bdrv_open(filename, NULL, NULL, 941 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 942 if (!bs_file) { 943 ret = -EIO; 944 goto done; 945 } 946 947 qdict_put_str(qdict, "driver", "vdi"); 948 qdict_put_str(qdict, "file", bs_file->node_name); 949 if (is_static) { 950 qdict_put_str(qdict, "preallocation", "metadata"); 951 } 952 953 /* Get the QAPI object */ 954 v = qobject_input_visitor_new_flat_confused(qdict, errp); 955 if (!v) { 956 ret = -EINVAL; 957 goto done; 958 } 959 visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err); 960 visit_free(v); 961 962 if (local_err) { 963 error_propagate(errp, local_err); 964 ret = -EINVAL; 965 goto done; 966 } 967 968 /* Silently round up size */ 969 assert(create_options->driver == BLOCKDEV_DRIVER_VDI); 970 create_options->u.vdi.size = ROUND_UP(create_options->u.vdi.size, 971 BDRV_SECTOR_SIZE); 972 973 /* Create the vdi image (format layer) */ 974 ret = vdi_co_do_create(create_options, block_size, errp); 975 done: 976 qobject_unref(qdict); 977 qapi_free_BlockdevCreateOptions(create_options); 978 bdrv_unref(bs_file); 979 return ret; 980 } 981 982 static void vdi_close(BlockDriverState *bs) 983 { 984 BDRVVdiState *s = bs->opaque; 985 986 qemu_vfree(s->bmap); 987 988 migrate_del_blocker(s->migration_blocker); 989 error_free(s->migration_blocker); 990 } 991 992 static int vdi_has_zero_init(BlockDriverState *bs) 993 { 994 BDRVVdiState *s = bs->opaque; 995 996 if (s->header.image_type == VDI_TYPE_STATIC) { 997 return bdrv_has_zero_init(bs->file->bs); 998 } else { 999 return 1; 1000 } 1001 } 1002 1003 static QemuOptsList vdi_create_opts = { 1004 .name = "vdi-create-opts", 1005 .head = QTAILQ_HEAD_INITIALIZER(vdi_create_opts.head), 1006 .desc = { 1007 { 1008 .name = BLOCK_OPT_SIZE, 1009 .type = QEMU_OPT_SIZE, 1010 .help = "Virtual disk size" 1011 }, 1012 #if defined(CONFIG_VDI_BLOCK_SIZE) 1013 { 1014 .name = BLOCK_OPT_CLUSTER_SIZE, 1015 .type = QEMU_OPT_SIZE, 1016 .help = "VDI cluster (block) size", 1017 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) 1018 }, 1019 #endif 1020 #if defined(CONFIG_VDI_STATIC_IMAGE) 1021 { 1022 .name = BLOCK_OPT_STATIC, 1023 .type = QEMU_OPT_BOOL, 1024 .help = "VDI static (pre-allocated) image", 1025 .def_value_str = "off" 1026 }, 1027 #endif 1028 /* TODO: An additional option to set UUID values might be useful. */ 1029 { /* end of list */ } 1030 } 1031 }; 1032 1033 static BlockDriver bdrv_vdi = { 1034 .format_name = "vdi", 1035 .instance_size = sizeof(BDRVVdiState), 1036 .bdrv_probe = vdi_probe, 1037 .bdrv_open = vdi_open, 1038 .bdrv_close = vdi_close, 1039 .bdrv_reopen_prepare = vdi_reopen_prepare, 1040 .bdrv_child_perm = bdrv_format_default_perms, 1041 .bdrv_co_create = vdi_co_create, 1042 .bdrv_co_create_opts = vdi_co_create_opts, 1043 .bdrv_has_zero_init = vdi_has_zero_init, 1044 .bdrv_co_block_status = vdi_co_block_status, 1045 .bdrv_make_empty = vdi_make_empty, 1046 1047 .bdrv_co_preadv = vdi_co_preadv, 1048 #if defined(CONFIG_VDI_WRITE) 1049 .bdrv_co_pwritev = vdi_co_pwritev, 1050 #endif 1051 1052 .bdrv_get_info = vdi_get_info, 1053 1054 .create_opts = &vdi_create_opts, 1055 .bdrv_co_check = vdi_co_check, 1056 }; 1057 1058 static void bdrv_vdi_init(void) 1059 { 1060 logout("\n"); 1061 bdrv_register(&bdrv_vdi); 1062 } 1063 1064 block_init(bdrv_vdi_init); 1065