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 return 0; 338 } 339 340 static int vdi_make_empty(BlockDriverState *bs) 341 { 342 /* TODO: missing code. */ 343 logout("\n"); 344 /* The return value for missing code must be 0, see block.c. */ 345 return 0; 346 } 347 348 static int vdi_probe(const uint8_t *buf, int buf_size, const char *filename) 349 { 350 const VdiHeader *header = (const VdiHeader *)buf; 351 int ret = 0; 352 353 logout("\n"); 354 355 if (buf_size < sizeof(*header)) { 356 /* Header too small, no VDI. */ 357 } else if (le32_to_cpu(header->signature) == VDI_SIGNATURE) { 358 ret = 100; 359 } 360 361 if (ret == 0) { 362 logout("no vdi image\n"); 363 } else { 364 logout("%s", header->text); 365 } 366 367 return ret; 368 } 369 370 static int vdi_open(BlockDriverState *bs, QDict *options, int flags, 371 Error **errp) 372 { 373 BDRVVdiState *s = bs->opaque; 374 VdiHeader header; 375 size_t bmap_size; 376 int ret; 377 Error *local_err = NULL; 378 QemuUUID uuid_link, uuid_parent; 379 380 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds, 381 BDRV_CHILD_IMAGE, false, errp); 382 if (!bs->file) { 383 return -EINVAL; 384 } 385 386 logout("\n"); 387 388 ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); 389 if (ret < 0) { 390 goto fail; 391 } 392 393 vdi_header_to_cpu(&header); 394 if (VDI_DEBUG) { 395 vdi_header_print(&header); 396 } 397 398 if (header.disk_size > VDI_DISK_SIZE_MAX) { 399 error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64 400 ", max supported is 0x%" PRIx64 ")", 401 header.disk_size, VDI_DISK_SIZE_MAX); 402 ret = -ENOTSUP; 403 goto fail; 404 } 405 406 uuid_link = header.uuid_link; 407 uuid_parent = header.uuid_parent; 408 409 if (header.disk_size % SECTOR_SIZE != 0) { 410 /* 'VBoxManage convertfromraw' can create images with odd disk sizes. 411 We accept them but round the disk size to the next multiple of 412 SECTOR_SIZE. */ 413 logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size); 414 header.disk_size = ROUND_UP(header.disk_size, SECTOR_SIZE); 415 } 416 417 if (header.signature != VDI_SIGNATURE) { 418 error_setg(errp, "Image not in VDI format (bad signature %08" PRIx32 419 ")", header.signature); 420 ret = -EINVAL; 421 goto fail; 422 } else if (header.version != VDI_VERSION_1_1) { 423 error_setg(errp, "unsupported VDI image (version %" PRIu32 ".%" PRIu32 424 ")", header.version >> 16, header.version & 0xffff); 425 ret = -ENOTSUP; 426 goto fail; 427 } else if (header.offset_bmap % SECTOR_SIZE != 0) { 428 /* We only support block maps which start on a sector boundary. */ 429 error_setg(errp, "unsupported VDI image (unaligned block map offset " 430 "0x%" PRIx32 ")", header.offset_bmap); 431 ret = -ENOTSUP; 432 goto fail; 433 } else if (header.offset_data % SECTOR_SIZE != 0) { 434 /* We only support data blocks which start on a sector boundary. */ 435 error_setg(errp, "unsupported VDI image (unaligned data offset 0x%" 436 PRIx32 ")", header.offset_data); 437 ret = -ENOTSUP; 438 goto fail; 439 } else if (header.sector_size != SECTOR_SIZE) { 440 error_setg(errp, "unsupported VDI image (sector size %" PRIu32 441 " is not %u)", header.sector_size, SECTOR_SIZE); 442 ret = -ENOTSUP; 443 goto fail; 444 } else if (header.block_size != DEFAULT_CLUSTER_SIZE) { 445 error_setg(errp, "unsupported VDI image (block size %" PRIu32 446 " is not %" PRIu32 ")", 447 header.block_size, DEFAULT_CLUSTER_SIZE); 448 ret = -ENOTSUP; 449 goto fail; 450 } else if (header.disk_size > 451 (uint64_t)header.blocks_in_image * header.block_size) { 452 error_setg(errp, "unsupported VDI image (disk size %" PRIu64 ", " 453 "image bitmap has room for %" PRIu64 ")", 454 header.disk_size, 455 (uint64_t)header.blocks_in_image * header.block_size); 456 ret = -ENOTSUP; 457 goto fail; 458 } else if (!qemu_uuid_is_null(&uuid_link)) { 459 error_setg(errp, "unsupported VDI image (non-NULL link UUID)"); 460 ret = -ENOTSUP; 461 goto fail; 462 } else if (!qemu_uuid_is_null(&uuid_parent)) { 463 error_setg(errp, "unsupported VDI image (non-NULL parent UUID)"); 464 ret = -ENOTSUP; 465 goto fail; 466 } else if (header.blocks_in_image > VDI_BLOCKS_IN_IMAGE_MAX) { 467 error_setg(errp, "unsupported VDI image " 468 "(too many blocks %u, max is %u)", 469 header.blocks_in_image, VDI_BLOCKS_IN_IMAGE_MAX); 470 ret = -ENOTSUP; 471 goto fail; 472 } 473 474 bs->total_sectors = header.disk_size / SECTOR_SIZE; 475 476 s->block_size = header.block_size; 477 s->bmap_sector = header.offset_bmap / SECTOR_SIZE; 478 s->header = header; 479 480 bmap_size = header.blocks_in_image * sizeof(uint32_t); 481 bmap_size = DIV_ROUND_UP(bmap_size, SECTOR_SIZE); 482 s->bmap = qemu_try_blockalign(bs->file->bs, bmap_size * SECTOR_SIZE); 483 if (s->bmap == NULL) { 484 ret = -ENOMEM; 485 goto fail; 486 } 487 488 ret = bdrv_pread(bs->file, header.offset_bmap, s->bmap, 489 bmap_size * SECTOR_SIZE); 490 if (ret < 0) { 491 goto fail_free_bmap; 492 } 493 494 /* Disable migration when vdi images are used */ 495 error_setg(&s->migration_blocker, "The vdi format used by node '%s' " 496 "does not support live migration", 497 bdrv_get_device_or_node_name(bs)); 498 ret = migrate_add_blocker(s->migration_blocker, &local_err); 499 if (local_err) { 500 error_propagate(errp, local_err); 501 error_free(s->migration_blocker); 502 goto fail_free_bmap; 503 } 504 505 qemu_co_rwlock_init(&s->bmap_lock); 506 507 return 0; 508 509 fail_free_bmap: 510 qemu_vfree(s->bmap); 511 512 fail: 513 return ret; 514 } 515 516 static int vdi_reopen_prepare(BDRVReopenState *state, 517 BlockReopenQueue *queue, Error **errp) 518 { 519 return 0; 520 } 521 522 static int coroutine_fn vdi_co_block_status(BlockDriverState *bs, 523 bool want_zero, 524 int64_t offset, int64_t bytes, 525 int64_t *pnum, int64_t *map, 526 BlockDriverState **file) 527 { 528 BDRVVdiState *s = (BDRVVdiState *)bs->opaque; 529 size_t bmap_index = offset / s->block_size; 530 size_t index_in_block = offset % s->block_size; 531 uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]); 532 int result; 533 534 logout("%p, %" PRId64 ", %" PRId64 ", %p\n", bs, offset, bytes, pnum); 535 *pnum = MIN(s->block_size - index_in_block, bytes); 536 result = VDI_IS_ALLOCATED(bmap_entry); 537 if (!result) { 538 return BDRV_BLOCK_ZERO; 539 } 540 541 *map = s->header.offset_data + (uint64_t)bmap_entry * s->block_size + 542 index_in_block; 543 *file = bs->file->bs; 544 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | 545 (s->header.image_type == VDI_TYPE_STATIC ? BDRV_BLOCK_RECURSE : 0); 546 } 547 548 static int coroutine_fn 549 vdi_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 550 QEMUIOVector *qiov, int flags) 551 { 552 BDRVVdiState *s = bs->opaque; 553 QEMUIOVector local_qiov; 554 uint32_t bmap_entry; 555 uint32_t block_index; 556 uint32_t offset_in_block; 557 uint32_t n_bytes; 558 uint64_t bytes_done = 0; 559 int ret = 0; 560 561 logout("\n"); 562 563 qemu_iovec_init(&local_qiov, qiov->niov); 564 565 while (ret >= 0 && bytes > 0) { 566 block_index = offset / s->block_size; 567 offset_in_block = offset % s->block_size; 568 n_bytes = MIN(bytes, s->block_size - offset_in_block); 569 570 logout("will read %u bytes starting at offset %" PRIu64 "\n", 571 n_bytes, offset); 572 573 /* prepare next AIO request */ 574 qemu_co_rwlock_rdlock(&s->bmap_lock); 575 bmap_entry = le32_to_cpu(s->bmap[block_index]); 576 qemu_co_rwlock_unlock(&s->bmap_lock); 577 if (!VDI_IS_ALLOCATED(bmap_entry)) { 578 /* Block not allocated, return zeros, no need to wait. */ 579 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes); 580 ret = 0; 581 } else { 582 uint64_t data_offset = s->header.offset_data + 583 (uint64_t)bmap_entry * s->block_size + 584 offset_in_block; 585 586 qemu_iovec_reset(&local_qiov); 587 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); 588 589 ret = bdrv_co_preadv(bs->file, data_offset, n_bytes, 590 &local_qiov, 0); 591 } 592 logout("%u bytes read\n", n_bytes); 593 594 bytes -= n_bytes; 595 offset += n_bytes; 596 bytes_done += n_bytes; 597 } 598 599 qemu_iovec_destroy(&local_qiov); 600 601 return ret; 602 } 603 604 static int coroutine_fn 605 vdi_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes, 606 QEMUIOVector *qiov, int flags) 607 { 608 BDRVVdiState *s = bs->opaque; 609 QEMUIOVector local_qiov; 610 uint32_t bmap_entry; 611 uint32_t block_index; 612 uint32_t offset_in_block; 613 uint32_t n_bytes; 614 uint64_t data_offset; 615 uint32_t bmap_first = VDI_UNALLOCATED; 616 uint32_t bmap_last = VDI_UNALLOCATED; 617 uint8_t *block = NULL; 618 uint64_t bytes_done = 0; 619 int ret = 0; 620 621 logout("\n"); 622 623 qemu_iovec_init(&local_qiov, qiov->niov); 624 625 while (ret >= 0 && bytes > 0) { 626 block_index = offset / s->block_size; 627 offset_in_block = offset % s->block_size; 628 n_bytes = MIN(bytes, s->block_size - offset_in_block); 629 630 logout("will write %u bytes starting at offset %" PRIu64 "\n", 631 n_bytes, offset); 632 633 /* prepare next AIO request */ 634 qemu_co_rwlock_rdlock(&s->bmap_lock); 635 bmap_entry = le32_to_cpu(s->bmap[block_index]); 636 if (!VDI_IS_ALLOCATED(bmap_entry)) { 637 /* Allocate new block and write to it. */ 638 uint64_t data_offset; 639 qemu_co_rwlock_upgrade(&s->bmap_lock); 640 bmap_entry = le32_to_cpu(s->bmap[block_index]); 641 if (VDI_IS_ALLOCATED(bmap_entry)) { 642 /* A concurrent allocation did the work for us. */ 643 qemu_co_rwlock_downgrade(&s->bmap_lock); 644 goto nonallocating_write; 645 } 646 647 bmap_entry = s->header.blocks_allocated; 648 s->bmap[block_index] = cpu_to_le32(bmap_entry); 649 s->header.blocks_allocated++; 650 data_offset = s->header.offset_data + 651 (uint64_t)bmap_entry * s->block_size; 652 if (block == NULL) { 653 block = g_malloc(s->block_size); 654 bmap_first = block_index; 655 } 656 bmap_last = block_index; 657 /* Copy data to be written to new block and zero unused parts. */ 658 memset(block, 0, offset_in_block); 659 qemu_iovec_to_buf(qiov, bytes_done, block + offset_in_block, 660 n_bytes); 661 memset(block + offset_in_block + n_bytes, 0, 662 s->block_size - n_bytes - offset_in_block); 663 664 /* Write the new block under CoRwLock write-side protection, 665 * so this full-cluster write does not overlap a partial write 666 * of the same cluster, issued from the "else" branch. 667 */ 668 ret = bdrv_pwrite(bs->file, data_offset, block, s->block_size); 669 qemu_co_rwlock_unlock(&s->bmap_lock); 670 } else { 671 nonallocating_write: 672 data_offset = s->header.offset_data + 673 (uint64_t)bmap_entry * s->block_size + 674 offset_in_block; 675 qemu_co_rwlock_unlock(&s->bmap_lock); 676 677 qemu_iovec_reset(&local_qiov); 678 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); 679 680 ret = bdrv_co_pwritev(bs->file, data_offset, n_bytes, 681 &local_qiov, 0); 682 } 683 684 bytes -= n_bytes; 685 offset += n_bytes; 686 bytes_done += n_bytes; 687 688 logout("%u bytes written\n", n_bytes); 689 } 690 691 qemu_iovec_destroy(&local_qiov); 692 693 logout("finished data write\n"); 694 if (ret < 0) { 695 return ret; 696 } 697 698 if (block) { 699 /* One or more new blocks were allocated. */ 700 VdiHeader *header = (VdiHeader *) block; 701 uint8_t *base; 702 uint64_t offset; 703 uint32_t n_sectors; 704 705 logout("now writing modified header\n"); 706 assert(VDI_IS_ALLOCATED(bmap_first)); 707 *header = s->header; 708 vdi_header_to_le(header); 709 ret = bdrv_pwrite(bs->file, 0, block, sizeof(VdiHeader)); 710 g_free(block); 711 block = NULL; 712 713 if (ret < 0) { 714 return ret; 715 } 716 717 logout("now writing modified block map entry %u...%u\n", 718 bmap_first, bmap_last); 719 /* Write modified sectors from block map. */ 720 bmap_first /= (SECTOR_SIZE / sizeof(uint32_t)); 721 bmap_last /= (SECTOR_SIZE / sizeof(uint32_t)); 722 n_sectors = bmap_last - bmap_first + 1; 723 offset = s->bmap_sector + bmap_first; 724 base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE; 725 logout("will write %u block map sectors starting from entry %u\n", 726 n_sectors, bmap_first); 727 ret = bdrv_pwrite(bs->file, offset * SECTOR_SIZE, base, 728 n_sectors * SECTOR_SIZE); 729 } 730 731 return ret < 0 ? ret : 0; 732 } 733 734 static int coroutine_fn vdi_co_do_create(BlockdevCreateOptions *create_options, 735 size_t block_size, Error **errp) 736 { 737 BlockdevCreateOptionsVdi *vdi_opts; 738 int ret = 0; 739 uint64_t bytes = 0; 740 uint32_t blocks; 741 uint32_t image_type; 742 VdiHeader header; 743 size_t i; 744 size_t bmap_size; 745 int64_t offset = 0; 746 BlockDriverState *bs_file = NULL; 747 BlockBackend *blk = NULL; 748 uint32_t *bmap = NULL; 749 QemuUUID uuid; 750 751 assert(create_options->driver == BLOCKDEV_DRIVER_VDI); 752 vdi_opts = &create_options->u.vdi; 753 754 logout("\n"); 755 756 /* Validate options and set default values */ 757 bytes = vdi_opts->size; 758 759 if (!vdi_opts->has_preallocation) { 760 vdi_opts->preallocation = PREALLOC_MODE_OFF; 761 } 762 switch (vdi_opts->preallocation) { 763 case PREALLOC_MODE_OFF: 764 image_type = VDI_TYPE_DYNAMIC; 765 break; 766 case PREALLOC_MODE_METADATA: 767 image_type = VDI_TYPE_STATIC; 768 break; 769 default: 770 error_setg(errp, "Preallocation mode not supported for vdi"); 771 return -EINVAL; 772 } 773 774 #ifndef CONFIG_VDI_STATIC_IMAGE 775 if (image_type == VDI_TYPE_STATIC) { 776 ret = -ENOTSUP; 777 error_setg(errp, "Statically allocated images cannot be created in " 778 "this build"); 779 goto exit; 780 } 781 #endif 782 #ifndef CONFIG_VDI_BLOCK_SIZE 783 if (block_size != DEFAULT_CLUSTER_SIZE) { 784 ret = -ENOTSUP; 785 error_setg(errp, 786 "A non-default cluster size is not supported in this build"); 787 goto exit; 788 } 789 #endif 790 791 if (bytes > VDI_DISK_SIZE_MAX) { 792 ret = -ENOTSUP; 793 error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64 794 ", max supported is 0x%" PRIx64 ")", 795 bytes, VDI_DISK_SIZE_MAX); 796 goto exit; 797 } 798 799 /* Create BlockBackend to write to the image */ 800 bs_file = bdrv_open_blockdev_ref(vdi_opts->file, errp); 801 if (!bs_file) { 802 ret = -EIO; 803 goto exit; 804 } 805 806 blk = blk_new_with_bs(bs_file, BLK_PERM_WRITE | BLK_PERM_RESIZE, 807 BLK_PERM_ALL, errp); 808 if (!blk) { 809 ret = -EPERM; 810 goto exit; 811 } 812 813 blk_set_allow_write_beyond_eof(blk, true); 814 815 /* We need enough blocks to store the given disk size, 816 so always round up. */ 817 blocks = DIV_ROUND_UP(bytes, block_size); 818 819 bmap_size = blocks * sizeof(uint32_t); 820 bmap_size = ROUND_UP(bmap_size, SECTOR_SIZE); 821 822 memset(&header, 0, sizeof(header)); 823 pstrcpy(header.text, sizeof(header.text), VDI_TEXT); 824 header.signature = VDI_SIGNATURE; 825 header.version = VDI_VERSION_1_1; 826 header.header_size = 0x180; 827 header.image_type = image_type; 828 header.offset_bmap = 0x200; 829 header.offset_data = 0x200 + bmap_size; 830 header.sector_size = SECTOR_SIZE; 831 header.disk_size = bytes; 832 header.block_size = block_size; 833 header.blocks_in_image = blocks; 834 if (image_type == VDI_TYPE_STATIC) { 835 header.blocks_allocated = blocks; 836 } 837 qemu_uuid_generate(&uuid); 838 header.uuid_image = uuid; 839 qemu_uuid_generate(&uuid); 840 header.uuid_last_snap = uuid; 841 /* There is no need to set header.uuid_link or header.uuid_parent here. */ 842 if (VDI_DEBUG) { 843 vdi_header_print(&header); 844 } 845 vdi_header_to_le(&header); 846 ret = blk_pwrite(blk, offset, &header, sizeof(header), 0); 847 if (ret < 0) { 848 error_setg(errp, "Error writing header"); 849 goto exit; 850 } 851 offset += sizeof(header); 852 853 if (bmap_size > 0) { 854 bmap = g_try_malloc0(bmap_size); 855 if (bmap == NULL) { 856 ret = -ENOMEM; 857 error_setg(errp, "Could not allocate bmap"); 858 goto exit; 859 } 860 for (i = 0; i < blocks; i++) { 861 if (image_type == VDI_TYPE_STATIC) { 862 bmap[i] = i; 863 } else { 864 bmap[i] = VDI_UNALLOCATED; 865 } 866 } 867 ret = blk_pwrite(blk, offset, bmap, bmap_size, 0); 868 if (ret < 0) { 869 error_setg(errp, "Error writing bmap"); 870 goto exit; 871 } 872 offset += bmap_size; 873 } 874 875 if (image_type == VDI_TYPE_STATIC) { 876 ret = blk_truncate(blk, offset + blocks * block_size, false, 877 PREALLOC_MODE_OFF, 0, errp); 878 if (ret < 0) { 879 error_prepend(errp, "Failed to statically allocate file"); 880 goto exit; 881 } 882 } 883 884 ret = 0; 885 exit: 886 blk_unref(blk); 887 bdrv_unref(bs_file); 888 g_free(bmap); 889 return ret; 890 } 891 892 static int coroutine_fn vdi_co_create(BlockdevCreateOptions *create_options, 893 Error **errp) 894 { 895 return vdi_co_do_create(create_options, DEFAULT_CLUSTER_SIZE, errp); 896 } 897 898 static int coroutine_fn vdi_co_create_opts(BlockDriver *drv, 899 const char *filename, 900 QemuOpts *opts, 901 Error **errp) 902 { 903 QDict *qdict = NULL; 904 BlockdevCreateOptions *create_options = NULL; 905 BlockDriverState *bs_file = NULL; 906 uint64_t block_size = DEFAULT_CLUSTER_SIZE; 907 bool is_static = false; 908 Visitor *v; 909 Error *local_err = NULL; 910 int ret; 911 912 /* Parse options and convert legacy syntax. 913 * 914 * Since CONFIG_VDI_BLOCK_SIZE is disabled by default, 915 * cluster-size is not part of the QAPI schema; therefore we have 916 * to parse it before creating the QAPI object. */ 917 #if defined(CONFIG_VDI_BLOCK_SIZE) 918 block_size = qemu_opt_get_size_del(opts, 919 BLOCK_OPT_CLUSTER_SIZE, 920 DEFAULT_CLUSTER_SIZE); 921 if (block_size < BDRV_SECTOR_SIZE || block_size > UINT32_MAX || 922 !is_power_of_2(block_size)) 923 { 924 error_setg(errp, "Invalid cluster size"); 925 ret = -EINVAL; 926 goto done; 927 } 928 #endif 929 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_STATIC, false)) { 930 is_static = true; 931 } 932 933 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vdi_create_opts, true); 934 935 /* Create and open the file (protocol layer) */ 936 ret = bdrv_create_file(filename, opts, errp); 937 if (ret < 0) { 938 goto done; 939 } 940 941 bs_file = bdrv_open(filename, NULL, NULL, 942 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 943 if (!bs_file) { 944 ret = -EIO; 945 goto done; 946 } 947 948 qdict_put_str(qdict, "driver", "vdi"); 949 qdict_put_str(qdict, "file", bs_file->node_name); 950 if (is_static) { 951 qdict_put_str(qdict, "preallocation", "metadata"); 952 } 953 954 /* Get the QAPI object */ 955 v = qobject_input_visitor_new_flat_confused(qdict, errp); 956 if (!v) { 957 ret = -EINVAL; 958 goto done; 959 } 960 visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err); 961 visit_free(v); 962 963 if (local_err) { 964 error_propagate(errp, local_err); 965 ret = -EINVAL; 966 goto done; 967 } 968 969 /* Silently round up size */ 970 assert(create_options->driver == BLOCKDEV_DRIVER_VDI); 971 create_options->u.vdi.size = ROUND_UP(create_options->u.vdi.size, 972 BDRV_SECTOR_SIZE); 973 974 /* Create the vdi image (format layer) */ 975 ret = vdi_co_do_create(create_options, block_size, errp); 976 done: 977 qobject_unref(qdict); 978 qapi_free_BlockdevCreateOptions(create_options); 979 bdrv_unref(bs_file); 980 return ret; 981 } 982 983 static void vdi_close(BlockDriverState *bs) 984 { 985 BDRVVdiState *s = bs->opaque; 986 987 qemu_vfree(s->bmap); 988 989 migrate_del_blocker(s->migration_blocker); 990 error_free(s->migration_blocker); 991 } 992 993 static int vdi_has_zero_init(BlockDriverState *bs) 994 { 995 BDRVVdiState *s = bs->opaque; 996 997 if (s->header.image_type == VDI_TYPE_STATIC) { 998 return bdrv_has_zero_init(bs->file->bs); 999 } else { 1000 return 1; 1001 } 1002 } 1003 1004 static QemuOptsList vdi_create_opts = { 1005 .name = "vdi-create-opts", 1006 .head = QTAILQ_HEAD_INITIALIZER(vdi_create_opts.head), 1007 .desc = { 1008 { 1009 .name = BLOCK_OPT_SIZE, 1010 .type = QEMU_OPT_SIZE, 1011 .help = "Virtual disk size" 1012 }, 1013 #if defined(CONFIG_VDI_BLOCK_SIZE) 1014 { 1015 .name = BLOCK_OPT_CLUSTER_SIZE, 1016 .type = QEMU_OPT_SIZE, 1017 .help = "VDI cluster (block) size", 1018 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) 1019 }, 1020 #endif 1021 #if defined(CONFIG_VDI_STATIC_IMAGE) 1022 { 1023 .name = BLOCK_OPT_STATIC, 1024 .type = QEMU_OPT_BOOL, 1025 .help = "VDI static (pre-allocated) image", 1026 .def_value_str = "off" 1027 }, 1028 #endif 1029 /* TODO: An additional option to set UUID values might be useful. */ 1030 { /* end of list */ } 1031 } 1032 }; 1033 1034 static BlockDriver bdrv_vdi = { 1035 .format_name = "vdi", 1036 .instance_size = sizeof(BDRVVdiState), 1037 .bdrv_probe = vdi_probe, 1038 .bdrv_open = vdi_open, 1039 .bdrv_close = vdi_close, 1040 .bdrv_reopen_prepare = vdi_reopen_prepare, 1041 .bdrv_child_perm = bdrv_default_perms, 1042 .bdrv_co_create = vdi_co_create, 1043 .bdrv_co_create_opts = vdi_co_create_opts, 1044 .bdrv_has_zero_init = vdi_has_zero_init, 1045 .bdrv_co_block_status = vdi_co_block_status, 1046 .bdrv_make_empty = vdi_make_empty, 1047 1048 .bdrv_co_preadv = vdi_co_preadv, 1049 #if defined(CONFIG_VDI_WRITE) 1050 .bdrv_co_pwritev = vdi_co_pwritev, 1051 #endif 1052 1053 .bdrv_get_info = vdi_get_info, 1054 1055 .is_format = true, 1056 .create_opts = &vdi_create_opts, 1057 .bdrv_co_check = vdi_co_check, 1058 }; 1059 1060 static void bdrv_vdi_init(void) 1061 { 1062 logout("\n"); 1063 bdrv_register(&bdrv_vdi); 1064 } 1065 1066 block_init(bdrv_vdi_init); 1067