1 /* 2 * Block driver for the Virtual Disk Image (VDI) format 3 * 4 * Copyright (c) 2009 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 adjacents 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-common.h" 53 #include "block_int.h" 54 #include "module.h" 55 56 #if defined(CONFIG_UUID) 57 #include <uuid/uuid.h> 58 #else 59 /* TODO: move uuid emulation to some central place in QEMU. */ 60 #include "sysemu.h" /* UUID_FMT */ 61 typedef unsigned char uuid_t[16]; 62 void uuid_generate(uuid_t out); 63 int uuid_is_null(const uuid_t uu); 64 void uuid_unparse(const uuid_t uu, char *out); 65 #endif 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 logout(fmt, ...) \ 94 fprintf(stderr, "vdi\t%-24s" fmt, __func__, ##__VA_ARGS__) 95 #else 96 #define logout(fmt, ...) ((void)0) 97 #endif 98 99 /* Image signature. */ 100 #define VDI_SIGNATURE 0xbeda107f 101 102 /* Image version. */ 103 #define VDI_VERSION_1_1 0x00010001 104 105 /* Image type. */ 106 #define VDI_TYPE_DYNAMIC 1 107 #define VDI_TYPE_STATIC 2 108 109 /* Innotek / SUN images use these strings in header.text: 110 * "<<< innotek VirtualBox Disk Image >>>\n" 111 * "<<< Sun xVM VirtualBox Disk Image >>>\n" 112 * "<<< Sun VirtualBox Disk Image >>>\n" 113 * The value does not matter, so QEMU created images use a different text. 114 */ 115 #define VDI_TEXT "<<< QEMU VM Virtual Disk Image >>>\n" 116 117 /* A never-allocated block; semantically arbitrary content. */ 118 #define VDI_UNALLOCATED 0xffffffffU 119 120 /* A discarded (no longer allocated) block; semantically zero-filled. */ 121 #define VDI_DISCARDED 0xfffffffeU 122 123 #define VDI_IS_ALLOCATED(X) ((X) < VDI_DISCARDED) 124 125 #if !defined(CONFIG_UUID) 126 void uuid_generate(uuid_t out) 127 { 128 memset(out, 0, sizeof(uuid_t)); 129 } 130 131 int uuid_is_null(const uuid_t uu) 132 { 133 uuid_t null_uuid = { 0 }; 134 return memcmp(uu, null_uuid, sizeof(uuid_t)) == 0; 135 } 136 137 void uuid_unparse(const uuid_t uu, char *out) 138 { 139 snprintf(out, 37, UUID_FMT, 140 uu[0], uu[1], uu[2], uu[3], uu[4], uu[5], uu[6], uu[7], 141 uu[8], uu[9], uu[10], uu[11], uu[12], uu[13], uu[14], uu[15]); 142 } 143 #endif 144 145 typedef struct { 146 BlockDriverAIOCB common; 147 int64_t sector_num; 148 QEMUIOVector *qiov; 149 uint8_t *buf; 150 /* Total number of sectors. */ 151 int nb_sectors; 152 /* Number of sectors for current AIO. */ 153 int n_sectors; 154 /* New allocated block map entry. */ 155 uint32_t bmap_first; 156 uint32_t bmap_last; 157 /* Buffer for new allocated block. */ 158 void *block_buffer; 159 void *orig_buf; 160 bool is_write; 161 int header_modified; 162 BlockDriverAIOCB *hd_aiocb; 163 struct iovec hd_iov; 164 QEMUIOVector hd_qiov; 165 QEMUBH *bh; 166 } VdiAIOCB; 167 168 typedef struct { 169 char text[0x40]; 170 uint32_t signature; 171 uint32_t version; 172 uint32_t header_size; 173 uint32_t image_type; 174 uint32_t image_flags; 175 char description[256]; 176 uint32_t offset_bmap; 177 uint32_t offset_data; 178 uint32_t cylinders; /* disk geometry, unused here */ 179 uint32_t heads; /* disk geometry, unused here */ 180 uint32_t sectors; /* disk geometry, unused here */ 181 uint32_t sector_size; 182 uint32_t unused1; 183 uint64_t disk_size; 184 uint32_t block_size; 185 uint32_t block_extra; /* unused here */ 186 uint32_t blocks_in_image; 187 uint32_t blocks_allocated; 188 uuid_t uuid_image; 189 uuid_t uuid_last_snap; 190 uuid_t uuid_link; 191 uuid_t uuid_parent; 192 uint64_t unused2[7]; 193 } VdiHeader; 194 195 typedef struct { 196 /* The block map entries are little endian (even in memory). */ 197 uint32_t *bmap; 198 /* Size of block (bytes). */ 199 uint32_t block_size; 200 /* Size of block (sectors). */ 201 uint32_t block_sectors; 202 /* First sector of block map. */ 203 uint32_t bmap_sector; 204 /* VDI header (converted to host endianness). */ 205 VdiHeader header; 206 } BDRVVdiState; 207 208 /* Change UUID from little endian (IPRT = VirtualBox format) to big endian 209 * format (network byte order, standard, see RFC 4122) and vice versa. 210 */ 211 static void uuid_convert(uuid_t uuid) 212 { 213 bswap32s((uint32_t *)&uuid[0]); 214 bswap16s((uint16_t *)&uuid[4]); 215 bswap16s((uint16_t *)&uuid[6]); 216 } 217 218 static void vdi_header_to_cpu(VdiHeader *header) 219 { 220 le32_to_cpus(&header->signature); 221 le32_to_cpus(&header->version); 222 le32_to_cpus(&header->header_size); 223 le32_to_cpus(&header->image_type); 224 le32_to_cpus(&header->image_flags); 225 le32_to_cpus(&header->offset_bmap); 226 le32_to_cpus(&header->offset_data); 227 le32_to_cpus(&header->cylinders); 228 le32_to_cpus(&header->heads); 229 le32_to_cpus(&header->sectors); 230 le32_to_cpus(&header->sector_size); 231 le64_to_cpus(&header->disk_size); 232 le32_to_cpus(&header->block_size); 233 le32_to_cpus(&header->block_extra); 234 le32_to_cpus(&header->blocks_in_image); 235 le32_to_cpus(&header->blocks_allocated); 236 uuid_convert(header->uuid_image); 237 uuid_convert(header->uuid_last_snap); 238 uuid_convert(header->uuid_link); 239 uuid_convert(header->uuid_parent); 240 } 241 242 static void vdi_header_to_le(VdiHeader *header) 243 { 244 cpu_to_le32s(&header->signature); 245 cpu_to_le32s(&header->version); 246 cpu_to_le32s(&header->header_size); 247 cpu_to_le32s(&header->image_type); 248 cpu_to_le32s(&header->image_flags); 249 cpu_to_le32s(&header->offset_bmap); 250 cpu_to_le32s(&header->offset_data); 251 cpu_to_le32s(&header->cylinders); 252 cpu_to_le32s(&header->heads); 253 cpu_to_le32s(&header->sectors); 254 cpu_to_le32s(&header->sector_size); 255 cpu_to_le64s(&header->disk_size); 256 cpu_to_le32s(&header->block_size); 257 cpu_to_le32s(&header->block_extra); 258 cpu_to_le32s(&header->blocks_in_image); 259 cpu_to_le32s(&header->blocks_allocated); 260 cpu_to_le32s(&header->blocks_allocated); 261 uuid_convert(header->uuid_image); 262 uuid_convert(header->uuid_last_snap); 263 uuid_convert(header->uuid_link); 264 uuid_convert(header->uuid_parent); 265 } 266 267 #if defined(CONFIG_VDI_DEBUG) 268 static void vdi_header_print(VdiHeader *header) 269 { 270 char uuid[37]; 271 logout("text %s", header->text); 272 logout("signature 0x%04x\n", header->signature); 273 logout("header size 0x%04x\n", header->header_size); 274 logout("image type 0x%04x\n", header->image_type); 275 logout("image flags 0x%04x\n", header->image_flags); 276 logout("description %s\n", header->description); 277 logout("offset bmap 0x%04x\n", header->offset_bmap); 278 logout("offset data 0x%04x\n", header->offset_data); 279 logout("cylinders 0x%04x\n", header->cylinders); 280 logout("heads 0x%04x\n", header->heads); 281 logout("sectors 0x%04x\n", header->sectors); 282 logout("sector size 0x%04x\n", header->sector_size); 283 logout("image size 0x%" PRIx64 " B (%" PRIu64 " MiB)\n", 284 header->disk_size, header->disk_size / MiB); 285 logout("block size 0x%04x\n", header->block_size); 286 logout("block extra 0x%04x\n", header->block_extra); 287 logout("blocks tot. 0x%04x\n", header->blocks_in_image); 288 logout("blocks all. 0x%04x\n", header->blocks_allocated); 289 uuid_unparse(header->uuid_image, uuid); 290 logout("uuid image %s\n", uuid); 291 uuid_unparse(header->uuid_last_snap, uuid); 292 logout("uuid snap %s\n", uuid); 293 uuid_unparse(header->uuid_link, uuid); 294 logout("uuid link %s\n", uuid); 295 uuid_unparse(header->uuid_parent, uuid); 296 logout("uuid parent %s\n", uuid); 297 } 298 #endif 299 300 static int vdi_check(BlockDriverState *bs, BdrvCheckResult *res) 301 { 302 /* TODO: additional checks possible. */ 303 BDRVVdiState *s = (BDRVVdiState *)bs->opaque; 304 uint32_t blocks_allocated = 0; 305 uint32_t block; 306 uint32_t *bmap; 307 logout("\n"); 308 309 bmap = g_malloc(s->header.blocks_in_image * sizeof(uint32_t)); 310 memset(bmap, 0xff, s->header.blocks_in_image * sizeof(uint32_t)); 311 312 /* Check block map and value of blocks_allocated. */ 313 for (block = 0; block < s->header.blocks_in_image; block++) { 314 uint32_t bmap_entry = le32_to_cpu(s->bmap[block]); 315 if (VDI_IS_ALLOCATED(bmap_entry)) { 316 if (bmap_entry < s->header.blocks_in_image) { 317 blocks_allocated++; 318 if (!VDI_IS_ALLOCATED(bmap[bmap_entry])) { 319 bmap[bmap_entry] = bmap_entry; 320 } else { 321 fprintf(stderr, "ERROR: block index %" PRIu32 322 " also used by %" PRIu32 "\n", bmap[bmap_entry], bmap_entry); 323 res->corruptions++; 324 } 325 } else { 326 fprintf(stderr, "ERROR: block index %" PRIu32 327 " too large, is %" PRIu32 "\n", block, bmap_entry); 328 res->corruptions++; 329 } 330 } 331 } 332 if (blocks_allocated != s->header.blocks_allocated) { 333 fprintf(stderr, "ERROR: allocated blocks mismatch, is %" PRIu32 334 ", should be %" PRIu32 "\n", 335 blocks_allocated, s->header.blocks_allocated); 336 res->corruptions++; 337 } 338 339 g_free(bmap); 340 341 return 0; 342 } 343 344 static int vdi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 345 { 346 /* TODO: vdi_get_info would be needed for machine snapshots. 347 vm_state_offset is still missing. */ 348 BDRVVdiState *s = (BDRVVdiState *)bs->opaque; 349 logout("\n"); 350 bdi->cluster_size = s->block_size; 351 bdi->vm_state_offset = 0; 352 return 0; 353 } 354 355 static int vdi_make_empty(BlockDriverState *bs) 356 { 357 /* TODO: missing code. */ 358 logout("\n"); 359 /* The return value for missing code must be 0, see block.c. */ 360 return 0; 361 } 362 363 static int vdi_probe(const uint8_t *buf, int buf_size, const char *filename) 364 { 365 const VdiHeader *header = (const VdiHeader *)buf; 366 int result = 0; 367 368 logout("\n"); 369 370 if (buf_size < sizeof(*header)) { 371 /* Header too small, no VDI. */ 372 } else if (le32_to_cpu(header->signature) == VDI_SIGNATURE) { 373 result = 100; 374 } 375 376 if (result == 0) { 377 logout("no vdi image\n"); 378 } else { 379 logout("%s", header->text); 380 } 381 382 return result; 383 } 384 385 static int vdi_open(BlockDriverState *bs, int flags) 386 { 387 BDRVVdiState *s = bs->opaque; 388 VdiHeader header; 389 size_t bmap_size; 390 391 logout("\n"); 392 393 if (bdrv_read(bs->file, 0, (uint8_t *)&header, 1) < 0) { 394 goto fail; 395 } 396 397 vdi_header_to_cpu(&header); 398 #if defined(CONFIG_VDI_DEBUG) 399 vdi_header_print(&header); 400 #endif 401 402 if (header.disk_size % SECTOR_SIZE != 0) { 403 /* 'VBoxManage convertfromraw' can create images with odd disk sizes. 404 We accept them but round the disk size to the next multiple of 405 SECTOR_SIZE. */ 406 logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size); 407 header.disk_size += SECTOR_SIZE - 1; 408 header.disk_size &= ~(SECTOR_SIZE - 1); 409 } 410 411 if (header.version != VDI_VERSION_1_1) { 412 logout("unsupported version %u.%u\n", 413 header.version >> 16, header.version & 0xffff); 414 goto fail; 415 } else if (header.offset_bmap % SECTOR_SIZE != 0) { 416 /* We only support block maps which start on a sector boundary. */ 417 logout("unsupported block map offset 0x%x B\n", header.offset_bmap); 418 goto fail; 419 } else if (header.offset_data % SECTOR_SIZE != 0) { 420 /* We only support data blocks which start on a sector boundary. */ 421 logout("unsupported data offset 0x%x B\n", header.offset_data); 422 goto fail; 423 } else if (header.sector_size != SECTOR_SIZE) { 424 logout("unsupported sector size %u B\n", header.sector_size); 425 goto fail; 426 } else if (header.block_size != 1 * MiB) { 427 logout("unsupported block size %u B\n", header.block_size); 428 goto fail; 429 } else if (header.disk_size > 430 (uint64_t)header.blocks_in_image * header.block_size) { 431 logout("unsupported disk size %" PRIu64 " B\n", header.disk_size); 432 goto fail; 433 } else if (!uuid_is_null(header.uuid_link)) { 434 logout("link uuid != 0, unsupported\n"); 435 goto fail; 436 } else if (!uuid_is_null(header.uuid_parent)) { 437 logout("parent uuid != 0, unsupported\n"); 438 goto fail; 439 } 440 441 bs->total_sectors = header.disk_size / SECTOR_SIZE; 442 443 s->block_size = header.block_size; 444 s->block_sectors = header.block_size / SECTOR_SIZE; 445 s->bmap_sector = header.offset_bmap / SECTOR_SIZE; 446 s->header = header; 447 448 bmap_size = header.blocks_in_image * sizeof(uint32_t); 449 bmap_size = (bmap_size + SECTOR_SIZE - 1) / SECTOR_SIZE; 450 if (bmap_size > 0) { 451 s->bmap = g_malloc(bmap_size * SECTOR_SIZE); 452 } 453 if (bdrv_read(bs->file, s->bmap_sector, (uint8_t *)s->bmap, bmap_size) < 0) { 454 goto fail_free_bmap; 455 } 456 457 return 0; 458 459 fail_free_bmap: 460 g_free(s->bmap); 461 462 fail: 463 return -1; 464 } 465 466 static int vdi_is_allocated(BlockDriverState *bs, int64_t sector_num, 467 int nb_sectors, int *pnum) 468 { 469 /* TODO: Check for too large sector_num (in bdrv_is_allocated or here). */ 470 BDRVVdiState *s = (BDRVVdiState *)bs->opaque; 471 size_t bmap_index = sector_num / s->block_sectors; 472 size_t sector_in_block = sector_num % s->block_sectors; 473 int n_sectors = s->block_sectors - sector_in_block; 474 uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]); 475 logout("%p, %" PRId64 ", %d, %p\n", bs, sector_num, nb_sectors, pnum); 476 if (n_sectors > nb_sectors) { 477 n_sectors = nb_sectors; 478 } 479 *pnum = n_sectors; 480 return VDI_IS_ALLOCATED(bmap_entry); 481 } 482 483 static void vdi_aio_cancel(BlockDriverAIOCB *blockacb) 484 { 485 /* TODO: This code is untested. How can I get it executed? */ 486 VdiAIOCB *acb = container_of(blockacb, VdiAIOCB, common); 487 logout("\n"); 488 if (acb->hd_aiocb) { 489 bdrv_aio_cancel(acb->hd_aiocb); 490 } 491 qemu_aio_release(acb); 492 } 493 494 static AIOPool vdi_aio_pool = { 495 .aiocb_size = sizeof(VdiAIOCB), 496 .cancel = vdi_aio_cancel, 497 }; 498 499 static VdiAIOCB *vdi_aio_setup(BlockDriverState *bs, int64_t sector_num, 500 QEMUIOVector *qiov, int nb_sectors, 501 BlockDriverCompletionFunc *cb, void *opaque, int is_write) 502 { 503 VdiAIOCB *acb; 504 505 logout("%p, %" PRId64 ", %p, %d, %p, %p, %d\n", 506 bs, sector_num, qiov, nb_sectors, cb, opaque, is_write); 507 508 acb = qemu_aio_get(&vdi_aio_pool, bs, cb, opaque); 509 if (acb) { 510 acb->hd_aiocb = NULL; 511 acb->sector_num = sector_num; 512 acb->qiov = qiov; 513 acb->is_write = is_write; 514 515 if (qiov->niov > 1) { 516 acb->buf = qemu_blockalign(bs, qiov->size); 517 acb->orig_buf = acb->buf; 518 if (is_write) { 519 qemu_iovec_to_buffer(qiov, acb->buf); 520 } 521 } else { 522 acb->buf = (uint8_t *)qiov->iov->iov_base; 523 } 524 acb->nb_sectors = nb_sectors; 525 acb->n_sectors = 0; 526 acb->bmap_first = VDI_UNALLOCATED; 527 acb->bmap_last = VDI_UNALLOCATED; 528 acb->block_buffer = NULL; 529 acb->header_modified = 0; 530 } 531 return acb; 532 } 533 534 static int vdi_schedule_bh(QEMUBHFunc *cb, VdiAIOCB *acb) 535 { 536 logout("\n"); 537 538 if (acb->bh) { 539 return -EIO; 540 } 541 542 acb->bh = qemu_bh_new(cb, acb); 543 if (!acb->bh) { 544 return -EIO; 545 } 546 547 qemu_bh_schedule(acb->bh); 548 549 return 0; 550 } 551 552 static void vdi_aio_read_cb(void *opaque, int ret); 553 static void vdi_aio_write_cb(void *opaque, int ret); 554 555 static void vdi_aio_rw_bh(void *opaque) 556 { 557 VdiAIOCB *acb = opaque; 558 logout("\n"); 559 qemu_bh_delete(acb->bh); 560 acb->bh = NULL; 561 562 if (acb->is_write) { 563 vdi_aio_write_cb(opaque, 0); 564 } else { 565 vdi_aio_read_cb(opaque, 0); 566 } 567 } 568 569 static void vdi_aio_read_cb(void *opaque, int ret) 570 { 571 VdiAIOCB *acb = opaque; 572 BlockDriverState *bs = acb->common.bs; 573 BDRVVdiState *s = bs->opaque; 574 uint32_t bmap_entry; 575 uint32_t block_index; 576 uint32_t sector_in_block; 577 uint32_t n_sectors; 578 579 logout("%u sectors read\n", acb->n_sectors); 580 581 acb->hd_aiocb = NULL; 582 583 if (ret < 0) { 584 goto done; 585 } 586 587 acb->nb_sectors -= acb->n_sectors; 588 589 if (acb->nb_sectors == 0) { 590 /* request completed */ 591 ret = 0; 592 goto done; 593 } 594 595 acb->sector_num += acb->n_sectors; 596 acb->buf += acb->n_sectors * SECTOR_SIZE; 597 598 block_index = acb->sector_num / s->block_sectors; 599 sector_in_block = acb->sector_num % s->block_sectors; 600 n_sectors = s->block_sectors - sector_in_block; 601 if (n_sectors > acb->nb_sectors) { 602 n_sectors = acb->nb_sectors; 603 } 604 605 logout("will read %u sectors starting at sector %" PRIu64 "\n", 606 n_sectors, acb->sector_num); 607 608 /* prepare next AIO request */ 609 acb->n_sectors = n_sectors; 610 bmap_entry = le32_to_cpu(s->bmap[block_index]); 611 if (!VDI_IS_ALLOCATED(bmap_entry)) { 612 /* Block not allocated, return zeros, no need to wait. */ 613 memset(acb->buf, 0, n_sectors * SECTOR_SIZE); 614 ret = vdi_schedule_bh(vdi_aio_rw_bh, acb); 615 if (ret < 0) { 616 goto done; 617 } 618 } else { 619 uint64_t offset = s->header.offset_data / SECTOR_SIZE + 620 (uint64_t)bmap_entry * s->block_sectors + 621 sector_in_block; 622 acb->hd_iov.iov_base = (void *)acb->buf; 623 acb->hd_iov.iov_len = n_sectors * SECTOR_SIZE; 624 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1); 625 acb->hd_aiocb = bdrv_aio_readv(bs->file, offset, &acb->hd_qiov, 626 n_sectors, vdi_aio_read_cb, acb); 627 if (acb->hd_aiocb == NULL) { 628 ret = -EIO; 629 goto done; 630 } 631 } 632 return; 633 done: 634 if (acb->qiov->niov > 1) { 635 qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size); 636 qemu_vfree(acb->orig_buf); 637 } 638 acb->common.cb(acb->common.opaque, ret); 639 qemu_aio_release(acb); 640 } 641 642 static BlockDriverAIOCB *vdi_aio_readv(BlockDriverState *bs, 643 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, 644 BlockDriverCompletionFunc *cb, void *opaque) 645 { 646 VdiAIOCB *acb; 647 int ret; 648 649 logout("\n"); 650 acb = vdi_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0); 651 if (!acb) { 652 return NULL; 653 } 654 655 ret = vdi_schedule_bh(vdi_aio_rw_bh, acb); 656 if (ret < 0) { 657 if (acb->qiov->niov > 1) { 658 qemu_vfree(acb->orig_buf); 659 } 660 qemu_aio_release(acb); 661 return NULL; 662 } 663 664 return &acb->common; 665 } 666 667 static void vdi_aio_write_cb(void *opaque, int ret) 668 { 669 VdiAIOCB *acb = opaque; 670 BlockDriverState *bs = acb->common.bs; 671 BDRVVdiState *s = bs->opaque; 672 uint32_t bmap_entry; 673 uint32_t block_index; 674 uint32_t sector_in_block; 675 uint32_t n_sectors; 676 677 acb->hd_aiocb = NULL; 678 679 if (ret < 0) { 680 goto done; 681 } 682 683 acb->nb_sectors -= acb->n_sectors; 684 acb->sector_num += acb->n_sectors; 685 acb->buf += acb->n_sectors * SECTOR_SIZE; 686 687 if (acb->nb_sectors == 0) { 688 logout("finished data write\n"); 689 acb->n_sectors = 0; 690 if (acb->header_modified) { 691 VdiHeader *header = acb->block_buffer; 692 logout("now writing modified header\n"); 693 assert(VDI_IS_ALLOCATED(acb->bmap_first)); 694 *header = s->header; 695 vdi_header_to_le(header); 696 acb->header_modified = 0; 697 acb->hd_iov.iov_base = acb->block_buffer; 698 acb->hd_iov.iov_len = SECTOR_SIZE; 699 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1); 700 acb->hd_aiocb = bdrv_aio_writev(bs->file, 0, &acb->hd_qiov, 1, 701 vdi_aio_write_cb, acb); 702 if (acb->hd_aiocb == NULL) { 703 ret = -EIO; 704 goto done; 705 } 706 return; 707 } else if (VDI_IS_ALLOCATED(acb->bmap_first)) { 708 /* One or more new blocks were allocated. */ 709 uint64_t offset; 710 uint32_t bmap_first; 711 uint32_t bmap_last; 712 g_free(acb->block_buffer); 713 acb->block_buffer = NULL; 714 bmap_first = acb->bmap_first; 715 bmap_last = acb->bmap_last; 716 logout("now writing modified block map entry %u...%u\n", 717 bmap_first, bmap_last); 718 /* Write modified sectors from block map. */ 719 bmap_first /= (SECTOR_SIZE / sizeof(uint32_t)); 720 bmap_last /= (SECTOR_SIZE / sizeof(uint32_t)); 721 n_sectors = bmap_last - bmap_first + 1; 722 offset = s->bmap_sector + bmap_first; 723 acb->bmap_first = VDI_UNALLOCATED; 724 acb->hd_iov.iov_base = (void *)((uint8_t *)&s->bmap[0] + 725 bmap_first * SECTOR_SIZE); 726 acb->hd_iov.iov_len = n_sectors * SECTOR_SIZE; 727 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1); 728 logout("will write %u block map sectors starting from entry %u\n", 729 n_sectors, bmap_first); 730 acb->hd_aiocb = bdrv_aio_writev(bs->file, offset, &acb->hd_qiov, 731 n_sectors, vdi_aio_write_cb, acb); 732 if (acb->hd_aiocb == NULL) { 733 ret = -EIO; 734 goto done; 735 } 736 return; 737 } 738 ret = 0; 739 goto done; 740 } 741 742 logout("%u sectors written\n", acb->n_sectors); 743 744 block_index = acb->sector_num / s->block_sectors; 745 sector_in_block = acb->sector_num % s->block_sectors; 746 n_sectors = s->block_sectors - sector_in_block; 747 if (n_sectors > acb->nb_sectors) { 748 n_sectors = acb->nb_sectors; 749 } 750 751 logout("will write %u sectors starting at sector %" PRIu64 "\n", 752 n_sectors, acb->sector_num); 753 754 /* prepare next AIO request */ 755 acb->n_sectors = n_sectors; 756 bmap_entry = le32_to_cpu(s->bmap[block_index]); 757 if (!VDI_IS_ALLOCATED(bmap_entry)) { 758 /* Allocate new block and write to it. */ 759 uint64_t offset; 760 uint8_t *block; 761 bmap_entry = s->header.blocks_allocated; 762 s->bmap[block_index] = cpu_to_le32(bmap_entry); 763 s->header.blocks_allocated++; 764 offset = s->header.offset_data / SECTOR_SIZE + 765 (uint64_t)bmap_entry * s->block_sectors; 766 block = acb->block_buffer; 767 if (block == NULL) { 768 block = g_malloc0(s->block_size); 769 acb->block_buffer = block; 770 acb->bmap_first = block_index; 771 assert(!acb->header_modified); 772 acb->header_modified = 1; 773 } 774 acb->bmap_last = block_index; 775 memcpy(block + sector_in_block * SECTOR_SIZE, 776 acb->buf, n_sectors * SECTOR_SIZE); 777 acb->hd_iov.iov_base = (void *)block; 778 acb->hd_iov.iov_len = s->block_size; 779 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1); 780 acb->hd_aiocb = bdrv_aio_writev(bs->file, offset, 781 &acb->hd_qiov, s->block_sectors, 782 vdi_aio_write_cb, acb); 783 if (acb->hd_aiocb == NULL) { 784 ret = -EIO; 785 goto done; 786 } 787 } else { 788 uint64_t offset = s->header.offset_data / SECTOR_SIZE + 789 (uint64_t)bmap_entry * s->block_sectors + 790 sector_in_block; 791 acb->hd_iov.iov_base = (void *)acb->buf; 792 acb->hd_iov.iov_len = n_sectors * SECTOR_SIZE; 793 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1); 794 acb->hd_aiocb = bdrv_aio_writev(bs->file, offset, &acb->hd_qiov, 795 n_sectors, vdi_aio_write_cb, acb); 796 if (acb->hd_aiocb == NULL) { 797 ret = -EIO; 798 goto done; 799 } 800 } 801 802 return; 803 804 done: 805 if (acb->qiov->niov > 1) { 806 qemu_vfree(acb->orig_buf); 807 } 808 acb->common.cb(acb->common.opaque, ret); 809 qemu_aio_release(acb); 810 } 811 812 static BlockDriverAIOCB *vdi_aio_writev(BlockDriverState *bs, 813 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, 814 BlockDriverCompletionFunc *cb, void *opaque) 815 { 816 VdiAIOCB *acb; 817 int ret; 818 819 logout("\n"); 820 acb = vdi_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1); 821 if (!acb) { 822 return NULL; 823 } 824 825 ret = vdi_schedule_bh(vdi_aio_rw_bh, acb); 826 if (ret < 0) { 827 if (acb->qiov->niov > 1) { 828 qemu_vfree(acb->orig_buf); 829 } 830 qemu_aio_release(acb); 831 return NULL; 832 } 833 834 return &acb->common; 835 } 836 837 static int vdi_create(const char *filename, QEMUOptionParameter *options) 838 { 839 int fd; 840 int result = 0; 841 uint64_t bytes = 0; 842 uint32_t blocks; 843 size_t block_size = DEFAULT_CLUSTER_SIZE; 844 uint32_t image_type = VDI_TYPE_DYNAMIC; 845 VdiHeader header; 846 size_t i; 847 size_t bmap_size; 848 uint32_t *bmap; 849 850 logout("\n"); 851 852 /* Read out options. */ 853 while (options && options->name) { 854 if (!strcmp(options->name, BLOCK_OPT_SIZE)) { 855 bytes = options->value.n; 856 #if defined(CONFIG_VDI_BLOCK_SIZE) 857 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) { 858 if (options->value.n) { 859 /* TODO: Additional checks (SECTOR_SIZE * 2^n, ...). */ 860 block_size = options->value.n; 861 } 862 #endif 863 #if defined(CONFIG_VDI_STATIC_IMAGE) 864 } else if (!strcmp(options->name, BLOCK_OPT_STATIC)) { 865 if (options->value.n) { 866 image_type = VDI_TYPE_STATIC; 867 } 868 #endif 869 } 870 options++; 871 } 872 873 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 874 0644); 875 if (fd < 0) { 876 return -errno; 877 } 878 879 /* We need enough blocks to store the given disk size, 880 so always round up. */ 881 blocks = (bytes + block_size - 1) / block_size; 882 883 bmap_size = blocks * sizeof(uint32_t); 884 bmap_size = ((bmap_size + SECTOR_SIZE - 1) & ~(SECTOR_SIZE -1)); 885 886 memset(&header, 0, sizeof(header)); 887 pstrcpy(header.text, sizeof(header.text), VDI_TEXT); 888 header.signature = VDI_SIGNATURE; 889 header.version = VDI_VERSION_1_1; 890 header.header_size = 0x180; 891 header.image_type = image_type; 892 header.offset_bmap = 0x200; 893 header.offset_data = 0x200 + bmap_size; 894 header.sector_size = SECTOR_SIZE; 895 header.disk_size = bytes; 896 header.block_size = block_size; 897 header.blocks_in_image = blocks; 898 if (image_type == VDI_TYPE_STATIC) { 899 header.blocks_allocated = blocks; 900 } 901 uuid_generate(header.uuid_image); 902 uuid_generate(header.uuid_last_snap); 903 /* There is no need to set header.uuid_link or header.uuid_parent here. */ 904 #if defined(CONFIG_VDI_DEBUG) 905 vdi_header_print(&header); 906 #endif 907 vdi_header_to_le(&header); 908 if (write(fd, &header, sizeof(header)) < 0) { 909 result = -errno; 910 } 911 912 bmap = NULL; 913 if (bmap_size > 0) { 914 bmap = (uint32_t *)g_malloc0(bmap_size); 915 } 916 for (i = 0; i < blocks; i++) { 917 if (image_type == VDI_TYPE_STATIC) { 918 bmap[i] = i; 919 } else { 920 bmap[i] = VDI_UNALLOCATED; 921 } 922 } 923 if (write(fd, bmap, bmap_size) < 0) { 924 result = -errno; 925 } 926 g_free(bmap); 927 if (image_type == VDI_TYPE_STATIC) { 928 if (ftruncate(fd, sizeof(header) + bmap_size + blocks * block_size)) { 929 result = -errno; 930 } 931 } 932 933 if (close(fd) < 0) { 934 result = -errno; 935 } 936 937 return result; 938 } 939 940 static void vdi_close(BlockDriverState *bs) 941 { 942 } 943 944 static coroutine_fn int vdi_co_flush(BlockDriverState *bs) 945 { 946 logout("\n"); 947 return bdrv_co_flush(bs->file); 948 } 949 950 951 static QEMUOptionParameter vdi_create_options[] = { 952 { 953 .name = BLOCK_OPT_SIZE, 954 .type = OPT_SIZE, 955 .help = "Virtual disk size" 956 }, 957 #if defined(CONFIG_VDI_BLOCK_SIZE) 958 { 959 .name = BLOCK_OPT_CLUSTER_SIZE, 960 .type = OPT_SIZE, 961 .help = "VDI cluster (block) size", 962 .value = { .n = DEFAULT_CLUSTER_SIZE }, 963 }, 964 #endif 965 #if defined(CONFIG_VDI_STATIC_IMAGE) 966 { 967 .name = BLOCK_OPT_STATIC, 968 .type = OPT_FLAG, 969 .help = "VDI static (pre-allocated) image" 970 }, 971 #endif 972 /* TODO: An additional option to set UUID values might be useful. */ 973 { NULL } 974 }; 975 976 static BlockDriver bdrv_vdi = { 977 .format_name = "vdi", 978 .instance_size = sizeof(BDRVVdiState), 979 .bdrv_probe = vdi_probe, 980 .bdrv_open = vdi_open, 981 .bdrv_close = vdi_close, 982 .bdrv_create = vdi_create, 983 .bdrv_co_flush_to_disk = vdi_co_flush, 984 .bdrv_is_allocated = vdi_is_allocated, 985 .bdrv_make_empty = vdi_make_empty, 986 987 .bdrv_aio_readv = vdi_aio_readv, 988 #if defined(CONFIG_VDI_WRITE) 989 .bdrv_aio_writev = vdi_aio_writev, 990 #endif 991 992 .bdrv_get_info = vdi_get_info, 993 994 .create_options = vdi_create_options, 995 .bdrv_check = vdi_check, 996 }; 997 998 static void bdrv_vdi_init(void) 999 { 1000 logout("\n"); 1001 bdrv_register(&bdrv_vdi); 1002 } 1003 1004 block_init(bdrv_vdi_init); 1005