1 /* 2 * Block driver for the Virtual Disk Image (VDI) format 3 * 4 * Copyright (c) 2009, 2012 Stefan Weil 5 * 6 * SPDX-License-Identifier: GPL-2.0-or-later 7 * 8 * This program is free software: you can redistribute it and/or modify 9 * it under the terms of the GNU General Public License as published by 10 * the Free Software Foundation, either version 2 of the License, or 11 * (at your option) any later version. 12 * 13 * This program is distributed in the hope that it will be useful, 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 * GNU General Public License for more details. 17 * 18 * You should have received a copy of the GNU General Public License 19 * along with this program. If not, see <http://www.gnu.org/licenses/>. 20 * 21 * Reference: 22 * http://forums.virtualbox.org/viewtopic.php?t=8046 23 * 24 * This driver supports create / read / write operations on VDI images. 25 * 26 * Todo (see also TODO in code): 27 * 28 * Some features like snapshots are still missing. 29 * 30 * Deallocation of zero-filled blocks and shrinking images are missing, too 31 * (might be added to common block layer). 32 * 33 * Allocation of blocks could be optimized (less writes to block map and 34 * header). 35 * 36 * Read and write of adjacent blocks could be done in one operation 37 * (current code uses one operation per block (1 MiB). 38 * 39 * The code is not thread safe (missing locks for changes in header and 40 * block table, no problem with current QEMU). 41 * 42 * Hints: 43 * 44 * Blocks (VDI documentation) correspond to clusters (QEMU). 45 * QEMU's backing files could be implemented using VDI snapshot files (TODO). 46 * VDI snapshot files may also contain the complete machine state. 47 * Maybe this machine state can be converted to QEMU PC machine snapshot data. 48 * 49 * The driver keeps a block cache (little endian entries) in memory. 50 * For the standard block size (1 MiB), a 1 TiB disk will use 4 MiB RAM, 51 * so this seems to be reasonable. 52 */ 53 54 #include "qemu/osdep.h" 55 #include "qemu/units.h" 56 #include "qapi/error.h" 57 #include "qapi/qobject-input-visitor.h" 58 #include "qapi/qapi-visit-block-core.h" 59 #include "block/block_int.h" 60 #include "block/qdict.h" 61 #include "sysemu/block-backend.h" 62 #include "qemu/module.h" 63 #include "qemu/option.h" 64 #include "qemu/bswap.h" 65 #include "migration/blocker.h" 66 #include "qemu/coroutine.h" 67 #include "qemu/cutils.h" 68 #include "qemu/uuid.h" 69 #include "qemu/memalign.h" 70 71 /* Code configuration options. */ 72 73 /* Enable debug messages. */ 74 //~ #define CONFIG_VDI_DEBUG 75 76 /* Support write operations on VDI images. */ 77 #define CONFIG_VDI_WRITE 78 79 /* Support non-standard block (cluster) size. This is untested. 80 * Maybe it will be needed for very large images. 81 */ 82 //~ #define CONFIG_VDI_BLOCK_SIZE 83 84 /* Support static (fixed, pre-allocated) images. */ 85 #define CONFIG_VDI_STATIC_IMAGE 86 87 /* Command line option for static images. */ 88 #define BLOCK_OPT_STATIC "static" 89 90 #define SECTOR_SIZE 512 91 #define DEFAULT_CLUSTER_SIZE 1048576 92 /* Note: can't use 1 * MiB, because it's passed to stringify() */ 93 94 #if defined(CONFIG_VDI_DEBUG) 95 #define VDI_DEBUG 1 96 #else 97 #define VDI_DEBUG 0 98 #endif 99 100 #define logout(fmt, ...) \ 101 do { \ 102 if (VDI_DEBUG) { \ 103 fprintf(stderr, "vdi\t%-24s" fmt, __func__, ##__VA_ARGS__); \ 104 } \ 105 } while (0) 106 107 /* Image signature. */ 108 #define VDI_SIGNATURE 0xbeda107f 109 110 /* Image version. */ 111 #define VDI_VERSION_1_1 0x00010001 112 113 /* Image type. */ 114 #define VDI_TYPE_DYNAMIC 1 115 #define VDI_TYPE_STATIC 2 116 117 /* Innotek / SUN images use these strings in header.text: 118 * "<<< innotek VirtualBox Disk Image >>>\n" 119 * "<<< Sun xVM VirtualBox Disk Image >>>\n" 120 * "<<< Sun VirtualBox Disk Image >>>\n" 121 * The value does not matter, so QEMU created images use a different text. 122 */ 123 #define VDI_TEXT "<<< QEMU VM Virtual Disk Image >>>\n" 124 125 /* A never-allocated block; semantically arbitrary content. */ 126 #define VDI_UNALLOCATED 0xffffffffU 127 128 /* A discarded (no longer allocated) block; semantically zero-filled. */ 129 #define VDI_DISCARDED 0xfffffffeU 130 131 #define VDI_IS_ALLOCATED(X) ((X) < VDI_DISCARDED) 132 133 /* The bmap will take up VDI_BLOCKS_IN_IMAGE_MAX * sizeof(uint32_t) bytes; since 134 * the bmap is read and written in a single operation, its size needs to be 135 * limited to INT_MAX; furthermore, when opening an image, the bmap size is 136 * rounded up to be aligned on BDRV_SECTOR_SIZE. 137 * Therefore this should satisfy the following: 138 * VDI_BLOCKS_IN_IMAGE_MAX * sizeof(uint32_t) + BDRV_SECTOR_SIZE == INT_MAX + 1 139 * (INT_MAX + 1 is the first value not representable as an int) 140 * This guarantees that any value below or equal to the constant will, when 141 * multiplied by sizeof(uint32_t) and rounded up to a BDRV_SECTOR_SIZE boundary, 142 * still be below or equal to INT_MAX. */ 143 #define VDI_BLOCKS_IN_IMAGE_MAX \ 144 ((unsigned)((INT_MAX + 1u - BDRV_SECTOR_SIZE) / sizeof(uint32_t))) 145 #define VDI_DISK_SIZE_MAX ((uint64_t)VDI_BLOCKS_IN_IMAGE_MAX * \ 146 (uint64_t)DEFAULT_CLUSTER_SIZE) 147 148 static QemuOptsList vdi_create_opts; 149 150 typedef struct { 151 char text[0x40]; 152 uint32_t signature; 153 uint32_t version; 154 uint32_t header_size; 155 uint32_t image_type; 156 uint32_t image_flags; 157 char description[256]; 158 uint32_t offset_bmap; 159 uint32_t offset_data; 160 uint32_t cylinders; /* disk geometry, unused here */ 161 uint32_t heads; /* disk geometry, unused here */ 162 uint32_t sectors; /* disk geometry, unused here */ 163 uint32_t sector_size; 164 uint32_t unused1; 165 uint64_t disk_size; 166 uint32_t block_size; 167 uint32_t block_extra; /* unused here */ 168 uint32_t blocks_in_image; 169 uint32_t blocks_allocated; 170 QemuUUID uuid_image; 171 QemuUUID uuid_last_snap; 172 QemuUUID uuid_link; 173 QemuUUID uuid_parent; 174 uint64_t unused2[7]; 175 } QEMU_PACKED VdiHeader; 176 177 QEMU_BUILD_BUG_ON(sizeof(VdiHeader) != 512); 178 179 typedef struct { 180 /* The block map entries are little endian (even in memory). */ 181 uint32_t *bmap; 182 /* Size of block (bytes). */ 183 uint32_t block_size; 184 /* First sector of block map. */ 185 uint32_t bmap_sector; 186 /* VDI header (converted to host endianness). */ 187 VdiHeader header; 188 189 CoRwlock bmap_lock; 190 191 Error *migration_blocker; 192 } BDRVVdiState; 193 194 static void vdi_header_to_cpu(VdiHeader *header) 195 { 196 header->signature = le32_to_cpu(header->signature); 197 header->version = le32_to_cpu(header->version); 198 header->header_size = le32_to_cpu(header->header_size); 199 header->image_type = le32_to_cpu(header->image_type); 200 header->image_flags = le32_to_cpu(header->image_flags); 201 header->offset_bmap = le32_to_cpu(header->offset_bmap); 202 header->offset_data = le32_to_cpu(header->offset_data); 203 header->cylinders = le32_to_cpu(header->cylinders); 204 header->heads = le32_to_cpu(header->heads); 205 header->sectors = le32_to_cpu(header->sectors); 206 header->sector_size = le32_to_cpu(header->sector_size); 207 header->disk_size = le64_to_cpu(header->disk_size); 208 header->block_size = le32_to_cpu(header->block_size); 209 header->block_extra = le32_to_cpu(header->block_extra); 210 header->blocks_in_image = le32_to_cpu(header->blocks_in_image); 211 header->blocks_allocated = le32_to_cpu(header->blocks_allocated); 212 header->uuid_image = qemu_uuid_bswap(header->uuid_image); 213 header->uuid_last_snap = qemu_uuid_bswap(header->uuid_last_snap); 214 header->uuid_link = qemu_uuid_bswap(header->uuid_link); 215 header->uuid_parent = qemu_uuid_bswap(header->uuid_parent); 216 } 217 218 static void vdi_header_to_le(VdiHeader *header) 219 { 220 header->signature = cpu_to_le32(header->signature); 221 header->version = cpu_to_le32(header->version); 222 header->header_size = cpu_to_le32(header->header_size); 223 header->image_type = cpu_to_le32(header->image_type); 224 header->image_flags = cpu_to_le32(header->image_flags); 225 header->offset_bmap = cpu_to_le32(header->offset_bmap); 226 header->offset_data = cpu_to_le32(header->offset_data); 227 header->cylinders = cpu_to_le32(header->cylinders); 228 header->heads = cpu_to_le32(header->heads); 229 header->sectors = cpu_to_le32(header->sectors); 230 header->sector_size = cpu_to_le32(header->sector_size); 231 header->disk_size = cpu_to_le64(header->disk_size); 232 header->block_size = cpu_to_le32(header->block_size); 233 header->block_extra = cpu_to_le32(header->block_extra); 234 header->blocks_in_image = cpu_to_le32(header->blocks_in_image); 235 header->blocks_allocated = cpu_to_le32(header->blocks_allocated); 236 header->uuid_image = qemu_uuid_bswap(header->uuid_image); 237 header->uuid_last_snap = qemu_uuid_bswap(header->uuid_last_snap); 238 header->uuid_link = qemu_uuid_bswap(header->uuid_link); 239 header->uuid_parent = qemu_uuid_bswap(header->uuid_parent); 240 } 241 242 static void vdi_header_print(VdiHeader *header) 243 { 244 char uuidstr[UUID_STR_LEN]; 245 QemuUUID uuid; 246 logout("text %s", header->text); 247 logout("signature 0x%08x\n", header->signature); 248 logout("header size 0x%04x\n", header->header_size); 249 logout("image type 0x%04x\n", header->image_type); 250 logout("image flags 0x%04x\n", header->image_flags); 251 logout("description %s\n", header->description); 252 logout("offset bmap 0x%04x\n", header->offset_bmap); 253 logout("offset data 0x%04x\n", header->offset_data); 254 logout("cylinders 0x%04x\n", header->cylinders); 255 logout("heads 0x%04x\n", header->heads); 256 logout("sectors 0x%04x\n", header->sectors); 257 logout("sector size 0x%04x\n", header->sector_size); 258 logout("image size 0x%" PRIx64 " B (%" PRIu64 " MiB)\n", 259 header->disk_size, header->disk_size / MiB); 260 logout("block size 0x%04x\n", header->block_size); 261 logout("block extra 0x%04x\n", header->block_extra); 262 logout("blocks tot. 0x%04x\n", header->blocks_in_image); 263 logout("blocks all. 0x%04x\n", header->blocks_allocated); 264 uuid = header->uuid_image; 265 qemu_uuid_unparse(&uuid, uuidstr); 266 logout("uuid image %s\n", uuidstr); 267 uuid = header->uuid_last_snap; 268 qemu_uuid_unparse(&uuid, uuidstr); 269 logout("uuid snap %s\n", uuidstr); 270 uuid = header->uuid_link; 271 qemu_uuid_unparse(&uuid, uuidstr); 272 logout("uuid link %s\n", uuidstr); 273 uuid = header->uuid_parent; 274 qemu_uuid_unparse(&uuid, uuidstr); 275 logout("uuid parent %s\n", uuidstr); 276 } 277 278 static int coroutine_fn vdi_co_check(BlockDriverState *bs, BdrvCheckResult *res, 279 BdrvCheckMode fix) 280 { 281 /* TODO: additional checks possible. */ 282 BDRVVdiState *s = (BDRVVdiState *)bs->opaque; 283 uint32_t blocks_allocated = 0; 284 uint32_t block; 285 uint32_t *bmap; 286 logout("\n"); 287 288 if (fix) { 289 return -ENOTSUP; 290 } 291 292 bmap = g_try_new(uint32_t, s->header.blocks_in_image); 293 if (s->header.blocks_in_image && bmap == NULL) { 294 res->check_errors++; 295 return -ENOMEM; 296 } 297 298 memset(bmap, 0xff, s->header.blocks_in_image * sizeof(uint32_t)); 299 300 /* Check block map and value of blocks_allocated. */ 301 for (block = 0; block < s->header.blocks_in_image; block++) { 302 uint32_t bmap_entry = le32_to_cpu(s->bmap[block]); 303 if (VDI_IS_ALLOCATED(bmap_entry)) { 304 if (bmap_entry < s->header.blocks_in_image) { 305 blocks_allocated++; 306 if (!VDI_IS_ALLOCATED(bmap[bmap_entry])) { 307 bmap[bmap_entry] = bmap_entry; 308 } else { 309 fprintf(stderr, "ERROR: block index %" PRIu32 310 " also used by %" PRIu32 "\n", bmap[bmap_entry], bmap_entry); 311 res->corruptions++; 312 } 313 } else { 314 fprintf(stderr, "ERROR: block index %" PRIu32 315 " too large, is %" PRIu32 "\n", block, bmap_entry); 316 res->corruptions++; 317 } 318 } 319 } 320 if (blocks_allocated != s->header.blocks_allocated) { 321 fprintf(stderr, "ERROR: allocated blocks mismatch, is %" PRIu32 322 ", should be %" PRIu32 "\n", 323 blocks_allocated, s->header.blocks_allocated); 324 res->corruptions++; 325 } 326 327 g_free(bmap); 328 329 return 0; 330 } 331 332 static int coroutine_fn 333 vdi_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 334 { 335 /* TODO: vdi_co_get_info would be needed for machine snapshots. 336 vm_state_offset is still missing. */ 337 BDRVVdiState *s = (BDRVVdiState *)bs->opaque; 338 logout("\n"); 339 bdi->cluster_size = s->block_size; 340 bdi->vm_state_offset = 0; 341 return 0; 342 } 343 344 static int vdi_make_empty(BlockDriverState *bs) 345 { 346 /* TODO: missing code. */ 347 logout("\n"); 348 /* The return value for missing code must be 0, see block.c. */ 349 return 0; 350 } 351 352 static int vdi_probe(const uint8_t *buf, int buf_size, const char *filename) 353 { 354 const VdiHeader *header = (const VdiHeader *)buf; 355 int ret = 0; 356 357 logout("\n"); 358 359 if (buf_size < sizeof(*header)) { 360 /* Header too small, no VDI. */ 361 } else if (le32_to_cpu(header->signature) == VDI_SIGNATURE) { 362 ret = 100; 363 } 364 365 if (ret == 0) { 366 logout("no vdi image\n"); 367 } else { 368 logout("%s", header->text); 369 } 370 371 return ret; 372 } 373 374 static int vdi_open(BlockDriverState *bs, QDict *options, int flags, 375 Error **errp) 376 { 377 BDRVVdiState *s = bs->opaque; 378 VdiHeader header; 379 size_t bmap_size; 380 int ret; 381 QemuUUID uuid_link, uuid_parent; 382 383 ret = bdrv_open_file_child(NULL, options, "file", bs, errp); 384 if (ret < 0) { 385 return ret; 386 } 387 388 GRAPH_RDLOCK_GUARD_MAINLOOP(); 389 390 logout("\n"); 391 392 ret = bdrv_pread(bs->file, 0, sizeof(header), &header, 0); 393 if (ret < 0) { 394 goto fail; 395 } 396 397 vdi_header_to_cpu(&header); 398 if (VDI_DEBUG) { 399 vdi_header_print(&header); 400 } 401 402 if (header.disk_size > VDI_DISK_SIZE_MAX) { 403 error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64 404 ", max supported is 0x%" PRIx64 ")", 405 header.disk_size, VDI_DISK_SIZE_MAX); 406 ret = -ENOTSUP; 407 goto fail; 408 } 409 410 uuid_link = header.uuid_link; 411 uuid_parent = header.uuid_parent; 412 413 if (header.disk_size % SECTOR_SIZE != 0) { 414 /* 'VBoxManage convertfromraw' can create images with odd disk sizes. 415 We accept them but round the disk size to the next multiple of 416 SECTOR_SIZE. */ 417 logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size); 418 header.disk_size = ROUND_UP(header.disk_size, SECTOR_SIZE); 419 } 420 421 if (header.signature != VDI_SIGNATURE) { 422 error_setg(errp, "Image not in VDI format (bad signature %08" PRIx32 423 ")", header.signature); 424 ret = -EINVAL; 425 goto fail; 426 } else if (header.version != VDI_VERSION_1_1) { 427 error_setg(errp, "unsupported VDI image (version %" PRIu32 ".%" PRIu32 428 ")", header.version >> 16, header.version & 0xffff); 429 ret = -ENOTSUP; 430 goto fail; 431 } else if (header.offset_bmap % SECTOR_SIZE != 0) { 432 /* We only support block maps which start on a sector boundary. */ 433 error_setg(errp, "unsupported VDI image (unaligned block map offset " 434 "0x%" PRIx32 ")", header.offset_bmap); 435 ret = -ENOTSUP; 436 goto fail; 437 } else if (header.offset_data % SECTOR_SIZE != 0) { 438 /* We only support data blocks which start on a sector boundary. */ 439 error_setg(errp, "unsupported VDI image (unaligned data offset 0x%" 440 PRIx32 ")", header.offset_data); 441 ret = -ENOTSUP; 442 goto fail; 443 } else if (header.sector_size != SECTOR_SIZE) { 444 error_setg(errp, "unsupported VDI image (sector size %" PRIu32 445 " is not %u)", header.sector_size, SECTOR_SIZE); 446 ret = -ENOTSUP; 447 goto fail; 448 } else if (header.block_size != DEFAULT_CLUSTER_SIZE) { 449 error_setg(errp, "unsupported VDI image (block size %" PRIu32 450 " is not %" PRIu32 ")", 451 header.block_size, DEFAULT_CLUSTER_SIZE); 452 ret = -ENOTSUP; 453 goto fail; 454 } else if (header.disk_size > 455 (uint64_t)header.blocks_in_image * header.block_size) { 456 error_setg(errp, "unsupported VDI image (disk size %" PRIu64 ", " 457 "image bitmap has room for %" PRIu64 ")", 458 header.disk_size, 459 (uint64_t)header.blocks_in_image * header.block_size); 460 ret = -ENOTSUP; 461 goto fail; 462 } else if (!qemu_uuid_is_null(&uuid_link)) { 463 error_setg(errp, "unsupported VDI image (non-NULL link UUID)"); 464 ret = -ENOTSUP; 465 goto fail; 466 } else if (!qemu_uuid_is_null(&uuid_parent)) { 467 error_setg(errp, "unsupported VDI image (non-NULL parent UUID)"); 468 ret = -ENOTSUP; 469 goto fail; 470 } else if (header.blocks_in_image > VDI_BLOCKS_IN_IMAGE_MAX) { 471 error_setg(errp, "unsupported VDI image " 472 "(too many blocks %u, max is %u)", 473 header.blocks_in_image, VDI_BLOCKS_IN_IMAGE_MAX); 474 ret = -ENOTSUP; 475 goto fail; 476 } 477 478 bs->total_sectors = header.disk_size / SECTOR_SIZE; 479 480 s->block_size = header.block_size; 481 s->bmap_sector = header.offset_bmap / SECTOR_SIZE; 482 s->header = header; 483 484 bmap_size = header.blocks_in_image * sizeof(uint32_t); 485 bmap_size = DIV_ROUND_UP(bmap_size, SECTOR_SIZE); 486 s->bmap = qemu_try_blockalign(bs->file->bs, bmap_size * SECTOR_SIZE); 487 if (s->bmap == NULL) { 488 ret = -ENOMEM; 489 goto fail; 490 } 491 492 ret = bdrv_pread(bs->file, header.offset_bmap, bmap_size * SECTOR_SIZE, 493 s->bmap, 0); 494 if (ret < 0) { 495 goto fail_free_bmap; 496 } 497 498 /* Disable migration when vdi images are used */ 499 error_setg(&s->migration_blocker, "The vdi format used by node '%s' " 500 "does not support live migration", 501 bdrv_get_device_or_node_name(bs)); 502 503 ret = migrate_add_blocker_normal(&s->migration_blocker, errp); 504 if (ret < 0) { 505 goto fail_free_bmap; 506 } 507 508 qemu_co_rwlock_init(&s->bmap_lock); 509 510 return 0; 511 512 fail_free_bmap: 513 qemu_vfree(s->bmap); 514 515 fail: 516 return ret; 517 } 518 519 static int vdi_reopen_prepare(BDRVReopenState *state, 520 BlockReopenQueue *queue, Error **errp) 521 { 522 return 0; 523 } 524 525 static int coroutine_fn GRAPH_RDLOCK 526 vdi_co_block_status(BlockDriverState *bs, bool want_zero, int64_t offset, 527 int64_t bytes, int64_t *pnum, int64_t *map, 528 BlockDriverState **file) 529 { 530 BDRVVdiState *s = (BDRVVdiState *)bs->opaque; 531 size_t bmap_index = offset / s->block_size; 532 size_t index_in_block = offset % s->block_size; 533 uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]); 534 int result; 535 536 logout("%p, %" PRId64 ", %" PRId64 ", %p\n", bs, offset, bytes, pnum); 537 *pnum = MIN(s->block_size - index_in_block, bytes); 538 result = VDI_IS_ALLOCATED(bmap_entry); 539 if (!result) { 540 return BDRV_BLOCK_ZERO; 541 } 542 543 *map = s->header.offset_data + (uint64_t)bmap_entry * s->block_size + 544 index_in_block; 545 *file = bs->file->bs; 546 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | 547 (s->header.image_type == VDI_TYPE_STATIC ? BDRV_BLOCK_RECURSE : 0); 548 } 549 550 static int coroutine_fn GRAPH_RDLOCK 551 vdi_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes, 552 QEMUIOVector *qiov, BdrvRequestFlags flags) 553 { 554 BDRVVdiState *s = bs->opaque; 555 QEMUIOVector local_qiov; 556 uint32_t bmap_entry; 557 uint32_t block_index; 558 uint32_t offset_in_block; 559 uint32_t n_bytes; 560 uint64_t bytes_done = 0; 561 int ret = 0; 562 563 logout("\n"); 564 565 qemu_iovec_init(&local_qiov, qiov->niov); 566 567 while (ret >= 0 && bytes > 0) { 568 block_index = offset / s->block_size; 569 offset_in_block = offset % s->block_size; 570 n_bytes = MIN(bytes, s->block_size - offset_in_block); 571 572 logout("will read %u bytes starting at offset %" PRIu64 "\n", 573 n_bytes, offset); 574 575 /* prepare next AIO request */ 576 qemu_co_rwlock_rdlock(&s->bmap_lock); 577 bmap_entry = le32_to_cpu(s->bmap[block_index]); 578 qemu_co_rwlock_unlock(&s->bmap_lock); 579 if (!VDI_IS_ALLOCATED(bmap_entry)) { 580 /* Block not allocated, return zeros, no need to wait. */ 581 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes); 582 ret = 0; 583 } else { 584 uint64_t data_offset = s->header.offset_data + 585 (uint64_t)bmap_entry * s->block_size + 586 offset_in_block; 587 588 qemu_iovec_reset(&local_qiov); 589 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); 590 591 ret = bdrv_co_preadv(bs->file, data_offset, n_bytes, 592 &local_qiov, 0); 593 } 594 logout("%u bytes read\n", n_bytes); 595 596 bytes -= n_bytes; 597 offset += n_bytes; 598 bytes_done += n_bytes; 599 } 600 601 qemu_iovec_destroy(&local_qiov); 602 603 return ret; 604 } 605 606 static int coroutine_fn GRAPH_RDLOCK 607 vdi_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes, 608 QEMUIOVector *qiov, BdrvRequestFlags flags) 609 { 610 BDRVVdiState *s = bs->opaque; 611 QEMUIOVector local_qiov; 612 uint32_t bmap_entry; 613 uint32_t block_index; 614 uint32_t offset_in_block; 615 uint32_t n_bytes; 616 uint64_t data_offset; 617 uint32_t bmap_first = VDI_UNALLOCATED; 618 uint32_t bmap_last = VDI_UNALLOCATED; 619 uint8_t *block = NULL; 620 uint64_t bytes_done = 0; 621 int ret = 0; 622 623 logout("\n"); 624 625 qemu_iovec_init(&local_qiov, qiov->niov); 626 627 while (ret >= 0 && bytes > 0) { 628 block_index = offset / s->block_size; 629 offset_in_block = offset % s->block_size; 630 n_bytes = MIN(bytes, s->block_size - offset_in_block); 631 632 logout("will write %u bytes starting at offset %" PRIu64 "\n", 633 n_bytes, offset); 634 635 /* prepare next AIO request */ 636 qemu_co_rwlock_rdlock(&s->bmap_lock); 637 bmap_entry = le32_to_cpu(s->bmap[block_index]); 638 if (!VDI_IS_ALLOCATED(bmap_entry)) { 639 /* Allocate new block and write to it. */ 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_co_pwrite(bs->file, data_offset, s->block_size, block, 670 0); 671 qemu_co_rwlock_unlock(&s->bmap_lock); 672 } else { 673 nonallocating_write: 674 data_offset = s->header.offset_data + 675 (uint64_t)bmap_entry * s->block_size + 676 offset_in_block; 677 qemu_co_rwlock_unlock(&s->bmap_lock); 678 679 qemu_iovec_reset(&local_qiov); 680 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes); 681 682 ret = bdrv_co_pwritev(bs->file, data_offset, n_bytes, 683 &local_qiov, 0); 684 } 685 686 bytes -= n_bytes; 687 offset += n_bytes; 688 bytes_done += n_bytes; 689 690 logout("%u bytes written\n", n_bytes); 691 } 692 693 qemu_iovec_destroy(&local_qiov); 694 695 logout("finished data write\n"); 696 if (ret < 0) { 697 g_free(block); 698 return ret; 699 } 700 701 if (block) { 702 /* One or more new blocks were allocated. */ 703 VdiHeader *header; 704 uint8_t *base; 705 uint64_t bmap_offset; 706 uint32_t n_sectors; 707 708 g_free(block); 709 header = g_malloc(sizeof(*header)); 710 711 logout("now writing modified header\n"); 712 assert(VDI_IS_ALLOCATED(bmap_first)); 713 *header = s->header; 714 vdi_header_to_le(header); 715 ret = bdrv_co_pwrite(bs->file, 0, sizeof(*header), header, 0); 716 g_free(header); 717 718 if (ret < 0) { 719 return ret; 720 } 721 722 logout("now writing modified block map entry %u...%u\n", 723 bmap_first, bmap_last); 724 /* Write modified sectors from block map. */ 725 bmap_first /= (SECTOR_SIZE / sizeof(uint32_t)); 726 bmap_last /= (SECTOR_SIZE / sizeof(uint32_t)); 727 n_sectors = bmap_last - bmap_first + 1; 728 bmap_offset = s->bmap_sector + bmap_first; 729 base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE; 730 logout("will write %u block map sectors starting from entry %u\n", 731 n_sectors, bmap_first); 732 ret = bdrv_co_pwrite(bs->file, bmap_offset * SECTOR_SIZE, 733 n_sectors * SECTOR_SIZE, base, 0); 734 } 735 736 return ret; 737 } 738 739 static int coroutine_fn GRAPH_UNLOCKED 740 vdi_co_do_create(BlockdevCreateOptions *create_options, size_t block_size, 741 Error **errp) 742 { 743 ERRP_GUARD(); 744 BlockdevCreateOptionsVdi *vdi_opts; 745 int ret = 0; 746 uint64_t bytes = 0; 747 uint32_t blocks; 748 uint32_t image_type; 749 VdiHeader header; 750 size_t i; 751 size_t bmap_size; 752 int64_t offset = 0; 753 BlockDriverState *bs_file = NULL; 754 BlockBackend *blk = NULL; 755 uint32_t *bmap = NULL; 756 QemuUUID uuid; 757 758 assert(create_options->driver == BLOCKDEV_DRIVER_VDI); 759 vdi_opts = &create_options->u.vdi; 760 761 logout("\n"); 762 763 /* Validate options and set default values */ 764 bytes = vdi_opts->size; 765 766 if (!vdi_opts->has_preallocation) { 767 vdi_opts->preallocation = PREALLOC_MODE_OFF; 768 } 769 switch (vdi_opts->preallocation) { 770 case PREALLOC_MODE_OFF: 771 image_type = VDI_TYPE_DYNAMIC; 772 break; 773 case PREALLOC_MODE_METADATA: 774 image_type = VDI_TYPE_STATIC; 775 break; 776 default: 777 error_setg(errp, "Preallocation mode not supported for vdi"); 778 return -EINVAL; 779 } 780 781 #ifndef CONFIG_VDI_STATIC_IMAGE 782 if (image_type == VDI_TYPE_STATIC) { 783 ret = -ENOTSUP; 784 error_setg(errp, "Statically allocated images cannot be created in " 785 "this build"); 786 goto exit; 787 } 788 #endif 789 #ifndef CONFIG_VDI_BLOCK_SIZE 790 if (block_size != DEFAULT_CLUSTER_SIZE) { 791 ret = -ENOTSUP; 792 error_setg(errp, 793 "A non-default cluster size is not supported in this build"); 794 goto exit; 795 } 796 #endif 797 798 if (bytes > VDI_DISK_SIZE_MAX) { 799 ret = -ENOTSUP; 800 error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64 801 ", max supported is 0x%" PRIx64 ")", 802 bytes, VDI_DISK_SIZE_MAX); 803 goto exit; 804 } 805 806 /* Create BlockBackend to write to the image */ 807 bs_file = bdrv_co_open_blockdev_ref(vdi_opts->file, errp); 808 if (!bs_file) { 809 ret = -EIO; 810 goto exit; 811 } 812 813 blk = blk_co_new_with_bs(bs_file, BLK_PERM_WRITE | BLK_PERM_RESIZE, 814 BLK_PERM_ALL, errp); 815 if (!blk) { 816 ret = -EPERM; 817 goto exit; 818 } 819 820 blk_set_allow_write_beyond_eof(blk, true); 821 822 /* We need enough blocks to store the given disk size, 823 so always round up. */ 824 blocks = DIV_ROUND_UP(bytes, block_size); 825 826 bmap_size = blocks * sizeof(uint32_t); 827 bmap_size = ROUND_UP(bmap_size, SECTOR_SIZE); 828 829 memset(&header, 0, sizeof(header)); 830 pstrcpy(header.text, sizeof(header.text), VDI_TEXT); 831 header.signature = VDI_SIGNATURE; 832 header.version = VDI_VERSION_1_1; 833 header.header_size = 0x180; 834 header.image_type = image_type; 835 header.offset_bmap = 0x200; 836 header.offset_data = 0x200 + bmap_size; 837 header.sector_size = SECTOR_SIZE; 838 header.disk_size = bytes; 839 header.block_size = block_size; 840 header.blocks_in_image = blocks; 841 if (image_type == VDI_TYPE_STATIC) { 842 header.blocks_allocated = blocks; 843 } 844 qemu_uuid_generate(&uuid); 845 header.uuid_image = uuid; 846 qemu_uuid_generate(&uuid); 847 header.uuid_last_snap = uuid; 848 /* There is no need to set header.uuid_link or header.uuid_parent here. */ 849 if (VDI_DEBUG) { 850 vdi_header_print(&header); 851 } 852 vdi_header_to_le(&header); 853 ret = blk_co_pwrite(blk, offset, sizeof(header), &header, 0); 854 if (ret < 0) { 855 error_setg(errp, "Error writing header"); 856 goto exit; 857 } 858 offset += sizeof(header); 859 860 if (bmap_size > 0) { 861 bmap = g_try_malloc0(bmap_size); 862 if (bmap == NULL) { 863 ret = -ENOMEM; 864 error_setg(errp, "Could not allocate bmap"); 865 goto exit; 866 } 867 for (i = 0; i < blocks; i++) { 868 if (image_type == VDI_TYPE_STATIC) { 869 bmap[i] = i; 870 } else { 871 bmap[i] = VDI_UNALLOCATED; 872 } 873 } 874 ret = blk_co_pwrite(blk, offset, bmap_size, bmap, 0); 875 if (ret < 0) { 876 error_setg(errp, "Error writing bmap"); 877 goto exit; 878 } 879 offset += bmap_size; 880 } 881 882 if (image_type == VDI_TYPE_STATIC) { 883 ret = blk_co_truncate(blk, offset + blocks * block_size, false, 884 PREALLOC_MODE_OFF, 0, errp); 885 if (ret < 0) { 886 error_prepend(errp, "Failed to statically allocate file"); 887 goto exit; 888 } 889 } 890 891 ret = 0; 892 exit: 893 blk_co_unref(blk); 894 bdrv_co_unref(bs_file); 895 g_free(bmap); 896 return ret; 897 } 898 899 static int coroutine_fn GRAPH_UNLOCKED 900 vdi_co_create(BlockdevCreateOptions *create_options, Error **errp) 901 { 902 return vdi_co_do_create(create_options, DEFAULT_CLUSTER_SIZE, errp); 903 } 904 905 static int coroutine_fn GRAPH_UNLOCKED 906 vdi_co_create_opts(BlockDriver *drv, const char *filename, 907 QemuOpts *opts, Error **errp) 908 { 909 QDict *qdict = NULL; 910 BlockdevCreateOptions *create_options = NULL; 911 BlockDriverState *bs_file = NULL; 912 uint64_t block_size = DEFAULT_CLUSTER_SIZE; 913 bool is_static = false; 914 Visitor *v; 915 int ret; 916 917 /* Parse options and convert legacy syntax. 918 * 919 * Since CONFIG_VDI_BLOCK_SIZE is disabled by default, 920 * cluster-size is not part of the QAPI schema; therefore we have 921 * to parse it before creating the QAPI object. */ 922 #if defined(CONFIG_VDI_BLOCK_SIZE) 923 block_size = qemu_opt_get_size_del(opts, 924 BLOCK_OPT_CLUSTER_SIZE, 925 DEFAULT_CLUSTER_SIZE); 926 if (block_size < BDRV_SECTOR_SIZE || block_size > UINT32_MAX || 927 !is_power_of_2(block_size)) 928 { 929 error_setg(errp, "Invalid cluster size"); 930 ret = -EINVAL; 931 goto done; 932 } 933 #endif 934 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_STATIC, false)) { 935 is_static = true; 936 } 937 938 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vdi_create_opts, true); 939 940 /* Create and open the file (protocol layer) */ 941 ret = bdrv_co_create_file(filename, opts, errp); 942 if (ret < 0) { 943 goto done; 944 } 945 946 bs_file = bdrv_co_open(filename, NULL, NULL, 947 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); 948 if (!bs_file) { 949 ret = -EIO; 950 goto done; 951 } 952 953 qdict_put_str(qdict, "driver", "vdi"); 954 qdict_put_str(qdict, "file", bs_file->node_name); 955 if (is_static) { 956 qdict_put_str(qdict, "preallocation", "metadata"); 957 } 958 959 /* Get the QAPI object */ 960 v = qobject_input_visitor_new_flat_confused(qdict, errp); 961 if (!v) { 962 ret = -EINVAL; 963 goto done; 964 } 965 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp); 966 visit_free(v); 967 if (!create_options) { 968 ret = -EINVAL; 969 goto done; 970 } 971 972 /* Silently round up size */ 973 assert(create_options->driver == BLOCKDEV_DRIVER_VDI); 974 create_options->u.vdi.size = ROUND_UP(create_options->u.vdi.size, 975 BDRV_SECTOR_SIZE); 976 977 /* Create the vdi image (format layer) */ 978 ret = vdi_co_do_create(create_options, block_size, errp); 979 done: 980 qobject_unref(qdict); 981 qapi_free_BlockdevCreateOptions(create_options); 982 bdrv_co_unref(bs_file); 983 return ret; 984 } 985 986 static void vdi_close(BlockDriverState *bs) 987 { 988 BDRVVdiState *s = bs->opaque; 989 990 qemu_vfree(s->bmap); 991 992 migrate_del_blocker(&s->migration_blocker); 993 } 994 995 static int GRAPH_RDLOCK vdi_has_zero_init(BlockDriverState *bs) 996 { 997 BDRVVdiState *s = bs->opaque; 998 999 if (s->header.image_type == VDI_TYPE_STATIC) { 1000 return bdrv_has_zero_init(bs->file->bs); 1001 } else { 1002 return 1; 1003 } 1004 } 1005 1006 static QemuOptsList vdi_create_opts = { 1007 .name = "vdi-create-opts", 1008 .head = QTAILQ_HEAD_INITIALIZER(vdi_create_opts.head), 1009 .desc = { 1010 { 1011 .name = BLOCK_OPT_SIZE, 1012 .type = QEMU_OPT_SIZE, 1013 .help = "Virtual disk size" 1014 }, 1015 #if defined(CONFIG_VDI_BLOCK_SIZE) 1016 { 1017 .name = BLOCK_OPT_CLUSTER_SIZE, 1018 .type = QEMU_OPT_SIZE, 1019 .help = "VDI cluster (block) size", 1020 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) 1021 }, 1022 #endif 1023 #if defined(CONFIG_VDI_STATIC_IMAGE) 1024 { 1025 .name = BLOCK_OPT_STATIC, 1026 .type = QEMU_OPT_BOOL, 1027 .help = "VDI static (pre-allocated) image", 1028 .def_value_str = "off" 1029 }, 1030 #endif 1031 /* TODO: An additional option to set UUID values might be useful. */ 1032 { /* end of list */ } 1033 } 1034 }; 1035 1036 static BlockDriver bdrv_vdi = { 1037 .format_name = "vdi", 1038 .instance_size = sizeof(BDRVVdiState), 1039 .bdrv_probe = vdi_probe, 1040 .bdrv_open = vdi_open, 1041 .bdrv_close = vdi_close, 1042 .bdrv_reopen_prepare = vdi_reopen_prepare, 1043 .bdrv_child_perm = bdrv_default_perms, 1044 .bdrv_co_create = vdi_co_create, 1045 .bdrv_co_create_opts = vdi_co_create_opts, 1046 .bdrv_has_zero_init = vdi_has_zero_init, 1047 .bdrv_co_block_status = vdi_co_block_status, 1048 .bdrv_make_empty = vdi_make_empty, 1049 1050 .bdrv_co_preadv = vdi_co_preadv, 1051 #if defined(CONFIG_VDI_WRITE) 1052 .bdrv_co_pwritev = vdi_co_pwritev, 1053 #endif 1054 1055 .bdrv_co_get_info = vdi_co_get_info, 1056 1057 .is_format = true, 1058 .create_opts = &vdi_create_opts, 1059 .bdrv_co_check = vdi_co_check, 1060 }; 1061 1062 static void bdrv_vdi_init(void) 1063 { 1064 logout("\n"); 1065 bdrv_register(&bdrv_vdi); 1066 } 1067 1068 block_init(bdrv_vdi_init); 1069