2 * Block driver for Hyper-V VHDX Images
4 * Copyright (c) 2013 Red Hat, Inc.,
7 * Jeff Cody <jcody@redhat.com>
9 * This is based on the "VHDX Format Specification v1.00", published 8/25/2012
11 * https://www.microsoft.com/en-us/download/details.aspx?id=34750
13 * This work is licensed under the terms of the GNU LGPL, version 2 or later.
14 * See the COPYING.LIB file in the top-level directory.
18 #include "qemu/osdep.h"
19 #include "qapi/error.h"
20 #include "block/block_int.h"
21 #include "block/qdict.h"
22 #include "sysemu/block-backend.h"
23 #include "qemu/module.h"
24 #include "qemu/option.h"
25 #include "qemu/crc32c.h"
26 #include "qemu/bswap.h"
27 #include "qemu/error-report.h"
28 #include "qemu/memalign.h"
30 #include "migration/blocker.h"
31 #include "qemu/uuid.h"
32 #include "qapi/qmp/qdict.h"
33 #include "qapi/qobject-input-visitor.h"
34 #include "qapi/qapi-visit-block-core.h"
36 /* Options for VHDX creation */
38 #define VHDX_BLOCK_OPT_LOG_SIZE "log_size"
39 #define VHDX_BLOCK_OPT_BLOCK_SIZE "block_size"
40 #define VHDX_BLOCK_OPT_ZERO "block_state_zero"
42 typedef enum VHDXImageType {
43 VHDX_TYPE_DYNAMIC = 0,
45 VHDX_TYPE_DIFFERENCING, /* Currently unsupported */
48 static QemuOptsList vhdx_create_opts;
50 /* Several metadata and region table data entries are identified by
51 * guids in a MS-specific GUID format. */
54 /* ------- Known Region Table GUIDs ---------------------- */
55 static const MSGUID bat_guid = { .data1 = 0x2dc27766,
58 .data4 = { 0x9d, 0x64, 0x11, 0x5e,
59 0x9b, 0xfd, 0x4a, 0x08} };
61 static const MSGUID metadata_guid = { .data1 = 0x8b7ca206,
64 .data4 = { 0xb8, 0xfe, 0x57, 0x5f,
65 0x05, 0x0f, 0x88, 0x6e} };
69 /* ------- Known Metadata Entry GUIDs ---------------------- */
70 static const MSGUID file_param_guid = { .data1 = 0xcaa16737,
73 .data4 = { 0xb3, 0xb6, 0x33, 0xf0,
74 0xaa, 0x44, 0xe7, 0x6b} };
76 static const MSGUID virtual_size_guid = { .data1 = 0x2FA54224,
79 .data4 = { 0xb2, 0x11, 0x5d, 0xbe,
80 0xd8, 0x3b, 0xf4, 0xb8} };
82 static const MSGUID page83_guid = { .data1 = 0xbeca12ab,
85 .data4 = { 0x93, 0xef, 0xc3, 0x09,
86 0xe0, 0x00, 0xc7, 0x46} };
89 static const MSGUID phys_sector_guid = { .data1 = 0xcda348c7,
92 .data4 = { 0x9c, 0xc9, 0xe9, 0x88,
93 0x52, 0x51, 0xc5, 0x56} };
95 static const MSGUID parent_locator_guid = { .data1 = 0xa8d35f2d,
98 .data4 = { 0xab, 0xf7, 0xd3,
102 static const MSGUID logical_sector_guid = { .data1 = 0x8141bf1d,
105 .data4 = { 0xba, 0x47, 0xf2,
109 /* Each parent type must have a valid GUID; this is for parent images
110 * of type 'VHDX'. If we were to allow e.g. a QCOW2 parent, we would
111 * need to make up our own QCOW2 GUID type */
112 static const MSGUID parent_vhdx_guid __attribute__((unused))
113 = { .data1 = 0xb04aefb7,
116 .data4 = { 0xb7, 0x89, 0x25, 0xb8,
117 0xe9, 0x44, 0x59, 0x13} };
120 #define META_FILE_PARAMETER_PRESENT 0x01
121 #define META_VIRTUAL_DISK_SIZE_PRESENT 0x02
122 #define META_PAGE_83_PRESENT 0x04
123 #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08
124 #define META_PHYS_SECTOR_SIZE_PRESENT 0x10
125 #define META_PARENT_LOCATOR_PRESENT 0x20
127 #define META_ALL_PRESENT \
128 (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \
129 META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \
130 META_PHYS_SECTOR_SIZE_PRESENT)
133 typedef struct VHDXSectorInfo {
134 uint32_t bat_idx; /* BAT entry index */
135 uint32_t sectors_avail; /* sectors available in payload block */
136 uint32_t bytes_left; /* bytes left in the block after data to r/w */
137 uint32_t bytes_avail; /* bytes available in payload block */
138 uint64_t file_offset; /* absolute offset in bytes, in file */
139 uint64_t block_offset; /* block offset, in bytes */
142 /* Calculates new checksum.
144 * Zero is substituted during crc calculation for the original crc field
145 * crc_offset: byte offset in buf of the buffer crc
146 * buf: buffer pointer
147 * size: size of buffer (must be > crc_offset+4)
149 * Note: The buffer should have all multi-byte data in little-endian format,
150 * and the resulting checksum is in little endian format.
152 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset)
157 assert(size > (crc_offset + sizeof(crc)));
159 memset(buf + crc_offset, 0, sizeof(crc));
160 crc = crc32c(0xffffffff, buf, size);
161 crc = cpu_to_le32(crc);
162 memcpy(buf + crc_offset, &crc, sizeof(crc));
167 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size,
174 if (crc_offset > 0) {
175 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
176 memset(buf + crc_offset, 0, sizeof(crc_orig));
179 crc_new = crc32c(crc, buf, size);
180 if (crc_offset > 0) {
181 memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig));
187 /* Validates the checksum of the buffer, with an in-place CRC.
189 * Zero is substituted during crc calculation for the original crc field,
190 * and the crc field is restored afterwards. But the buffer will be modified
191 * during the calculation, so this may not be not suitable for multi-threaded
194 * crc_offset: byte offset in buf of the buffer crc
195 * buf: buffer pointer
196 * size: size of buffer (must be > crc_offset+4)
198 * returns true if checksum is valid, false otherwise
200 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset)
206 assert(size > (crc_offset + 4));
208 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
209 crc_orig = le32_to_cpu(crc_orig);
211 crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset);
213 return crc == crc_orig;
218 * This generates a UUID that is compliant with the MS GUIDs used
219 * in the VHDX spec (and elsewhere).
221 void vhdx_guid_generate(MSGUID *guid)
224 assert(guid != NULL);
226 qemu_uuid_generate(&uuid);
227 memcpy(guid, &uuid, sizeof(MSGUID));
230 /* Check for region overlaps inside the VHDX image */
231 static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length)
237 end = start + length;
238 QLIST_FOREACH(r, &s->regions, entries) {
239 if (!((start >= r->end) || (end <= r->start))) {
240 error_report("VHDX region %" PRIu64 "-%" PRIu64 " overlaps with "
241 "region %" PRIu64 "-%." PRIu64, start, end, r->start,
252 /* Register a region for future checks */
253 static void vhdx_region_register(BDRVVHDXState *s,
254 uint64_t start, uint64_t length)
258 r = g_malloc0(sizeof(*r));
261 r->end = start + length;
263 QLIST_INSERT_HEAD(&s->regions, r, entries);
266 /* Free all registered regions */
267 static void vhdx_region_unregister_all(BDRVVHDXState *s)
269 VHDXRegionEntry *r, *r_next;
271 QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) {
272 QLIST_REMOVE(r, entries);
277 static void vhdx_set_shift_bits(BDRVVHDXState *s)
279 s->logical_sector_size_bits = ctz32(s->logical_sector_size);
280 s->sectors_per_block_bits = ctz32(s->sectors_per_block);
281 s->chunk_ratio_bits = ctz64(s->chunk_ratio);
282 s->block_size_bits = ctz32(s->block_size);
286 * Per the MS VHDX Specification, for every VHDX file:
287 * - The header section is fixed size - 1 MB
288 * - The header section is always the first "object"
289 * - The first 64KB of the header is the File Identifier
290 * - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile")
291 * - The following 512 bytes constitute a UTF-16 string identifiying the
292 * software that created the file, and is optional and diagnostic only.
294 * Therefore, we probe by looking for the vhdxfile signature "vhdxfile"
296 static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename)
298 if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) {
305 * Writes the header to the specified offset.
307 * This will optionally read in buffer data from disk (otherwise zero-fill),
308 * and then update the header checksum. Header is converted to proper
309 * endianness before being written to the specified file offset
311 static int vhdx_write_header(BdrvChild *file, VHDXHeader *hdr,
312 uint64_t offset, bool read)
314 BlockDriverState *bs_file = file->bs;
315 uint8_t *buffer = NULL;
317 VHDXHeader *header_le;
319 assert(bs_file != NULL);
322 /* the header checksum is not over just the packed size of VHDXHeader,
323 * but rather over the entire 'reserved' range for the header, which is
324 * 4KB (VHDX_HEADER_SIZE). */
326 buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE);
328 /* if true, we can't assume the extra reserved bytes are 0 */
329 ret = bdrv_pread(file, offset, VHDX_HEADER_SIZE, buffer, 0);
334 memset(buffer, 0, VHDX_HEADER_SIZE);
337 /* overwrite the actual VHDXHeader portion */
338 header_le = (VHDXHeader *)buffer;
339 memcpy(header_le, hdr, sizeof(VHDXHeader));
340 vhdx_header_le_export(hdr, header_le);
341 vhdx_update_checksum(buffer, VHDX_HEADER_SIZE,
342 offsetof(VHDXHeader, checksum));
343 ret = bdrv_pwrite_sync(file, offset, sizeof(VHDXHeader), header_le, 0);
350 /* Update the VHDX headers
352 * This follows the VHDX spec procedures for header updates.
354 * - non-current header is updated with largest sequence number
356 static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s,
357 bool generate_data_write_guid, MSGUID *log_guid)
361 uint64_t header_offset = VHDX_HEADER1_OFFSET;
363 VHDXHeader *active_header;
364 VHDXHeader *inactive_header;
366 /* operate on the non-current header */
367 if (s->curr_header == 0) {
369 header_offset = VHDX_HEADER2_OFFSET;
372 active_header = s->headers[s->curr_header];
373 inactive_header = s->headers[hdr_idx];
375 inactive_header->sequence_number = active_header->sequence_number + 1;
377 /* a new file guid must be generated before any file write, including
379 inactive_header->file_write_guid = s->session_guid;
381 /* a new data guid only needs to be generated before any guest-visible
382 * writes (i.e. something observable via virtual disk read) */
383 if (generate_data_write_guid) {
384 vhdx_guid_generate(&inactive_header->data_write_guid);
387 /* update the log guid if present */
389 inactive_header->log_guid = *log_guid;
392 ret = vhdx_write_header(bs->file, inactive_header, header_offset, true);
396 s->curr_header = hdr_idx;
403 * The VHDX spec calls for header updates to be performed twice, so that both
404 * the current and non-current header have valid info
406 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s,
407 bool generate_data_write_guid, MSGUID *log_guid)
411 ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
415 return vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
418 /* opens the specified header block from the VHDX file header section */
419 static void vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s,
425 bool h1_valid = false;
426 bool h2_valid = false;
431 /* header1 & header2 are freed in vhdx_close() */
432 header1 = qemu_blockalign(bs, sizeof(VHDXHeader));
433 header2 = qemu_blockalign(bs, sizeof(VHDXHeader));
435 buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE);
437 s->headers[0] = header1;
438 s->headers[1] = header2;
440 /* We have to read the whole VHDX_HEADER_SIZE instead of
441 * sizeof(VHDXHeader), because the checksum is over the whole
443 ret = bdrv_pread(bs->file, VHDX_HEADER1_OFFSET, VHDX_HEADER_SIZE, buffer,
448 /* copy over just the relevant portion that we need */
449 memcpy(header1, buffer, sizeof(VHDXHeader));
451 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
452 vhdx_header_le_import(header1);
453 if (header1->signature == VHDX_HEADER_SIGNATURE &&
454 header1->version == 1) {
455 h1_seq = header1->sequence_number;
460 ret = bdrv_pread(bs->file, VHDX_HEADER2_OFFSET, VHDX_HEADER_SIZE, buffer,
465 /* copy over just the relevant portion that we need */
466 memcpy(header2, buffer, sizeof(VHDXHeader));
468 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
469 vhdx_header_le_import(header2);
470 if (header2->signature == VHDX_HEADER_SIGNATURE &&
471 header2->version == 1) {
472 h2_seq = header2->sequence_number;
477 /* If there is only 1 valid header (or no valid headers), we
478 * don't care what the sequence numbers are */
479 if (h1_valid && !h2_valid) {
481 } else if (!h1_valid && h2_valid) {
483 } else if (!h1_valid && !h2_valid) {
486 /* If both headers are valid, then we choose the active one by the
487 * highest sequence number. If the sequence numbers are equal, that is
489 if (h1_seq > h2_seq) {
491 } else if (h2_seq > h1_seq) {
494 /* The Microsoft Disk2VHD tool will create 2 identical
495 * headers, with identical sequence numbers. If the headers are
496 * identical, don't consider the file corrupt */
497 if (!memcmp(header1, header2, sizeof(VHDXHeader))) {
505 vhdx_region_register(s, s->headers[s->curr_header]->log_offset,
506 s->headers[s->curr_header]->log_length);
510 error_setg_errno(errp, -ret, "No valid VHDX header found");
513 s->headers[0] = NULL;
514 s->headers[1] = NULL;
520 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s)
525 VHDXRegionTableEntry rt_entry;
527 bool bat_rt_found = false;
528 bool metadata_rt_found = false;
530 /* We have to read the whole 64KB block, because the crc32 is over the
532 buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE);
534 ret = bdrv_pread(bs->file, VHDX_REGION_TABLE_OFFSET,
535 VHDX_HEADER_BLOCK_SIZE, buffer, 0);
539 memcpy(&s->rt, buffer, sizeof(s->rt));
540 offset += sizeof(s->rt);
542 if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4)) {
547 vhdx_region_header_le_import(&s->rt);
549 if (s->rt.signature != VHDX_REGION_SIGNATURE) {
555 /* Per spec, maximum region table entry count is 2047 */
556 if (s->rt.entry_count > 2047) {
561 for (i = 0; i < s->rt.entry_count; i++) {
562 memcpy(&rt_entry, buffer + offset, sizeof(rt_entry));
563 offset += sizeof(rt_entry);
565 vhdx_region_entry_le_import(&rt_entry);
567 /* check for region overlap between these entries, and any
568 * other memory regions in the file */
569 ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length);
574 vhdx_region_register(s, rt_entry.file_offset, rt_entry.length);
576 /* see if we recognize the entry */
577 if (guid_eq(rt_entry.guid, bat_guid)) {
578 /* must be unique; if we have already found it this is invalid */
584 s->bat_rt = rt_entry;
588 if (guid_eq(rt_entry.guid, metadata_guid)) {
589 /* must be unique; if we have already found it this is invalid */
590 if (metadata_rt_found) {
594 metadata_rt_found = true;
595 s->metadata_rt = rt_entry;
599 if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) {
600 /* cannot read vhdx file - required region table entry that
601 * we do not understand. per spec, we must fail to open */
607 if (!bat_rt_found || !metadata_rt_found) {
621 /* Metadata initial parser
623 * This loads all the metadata entry fields. This may cause additional
624 * fields to be processed (e.g. parent locator, etc..).
626 * There are 5 Metadata items that are always required:
627 * - File Parameters (block size, has a parent)
628 * - Virtual Disk Size (size, in bytes, of the virtual drive)
629 * - Page 83 Data (scsi page 83 guid)
630 * - Logical Sector Size (logical sector size in bytes, either 512 or
631 * 4096. We only support 512 currently)
632 * - Physical Sector Size (512 or 4096)
634 * Also, if the File Parameters indicate this is a differencing file,
635 * we must also look for the Parent Locator metadata item.
637 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s)
643 VHDXMetadataTableEntry md_entry;
645 buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE);
647 ret = bdrv_pread(bs->file, s->metadata_rt.file_offset,
648 VHDX_METADATA_TABLE_MAX_SIZE, buffer, 0);
652 memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr));
653 offset += sizeof(s->metadata_hdr);
655 vhdx_metadata_header_le_import(&s->metadata_hdr);
657 if (s->metadata_hdr.signature != VHDX_METADATA_SIGNATURE) {
662 s->metadata_entries.present = 0;
664 if ((s->metadata_hdr.entry_count * sizeof(md_entry)) >
665 (VHDX_METADATA_TABLE_MAX_SIZE - offset)) {
670 for (i = 0; i < s->metadata_hdr.entry_count; i++) {
671 memcpy(&md_entry, buffer + offset, sizeof(md_entry));
672 offset += sizeof(md_entry);
674 vhdx_metadata_entry_le_import(&md_entry);
676 if (guid_eq(md_entry.item_id, file_param_guid)) {
677 if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) {
681 s->metadata_entries.file_parameters_entry = md_entry;
682 s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT;
686 if (guid_eq(md_entry.item_id, virtual_size_guid)) {
687 if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) {
691 s->metadata_entries.virtual_disk_size_entry = md_entry;
692 s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT;
696 if (guid_eq(md_entry.item_id, page83_guid)) {
697 if (s->metadata_entries.present & META_PAGE_83_PRESENT) {
701 s->metadata_entries.page83_data_entry = md_entry;
702 s->metadata_entries.present |= META_PAGE_83_PRESENT;
706 if (guid_eq(md_entry.item_id, logical_sector_guid)) {
707 if (s->metadata_entries.present &
708 META_LOGICAL_SECTOR_SIZE_PRESENT) {
712 s->metadata_entries.logical_sector_size_entry = md_entry;
713 s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT;
717 if (guid_eq(md_entry.item_id, phys_sector_guid)) {
718 if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) {
722 s->metadata_entries.phys_sector_size_entry = md_entry;
723 s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT;
727 if (guid_eq(md_entry.item_id, parent_locator_guid)) {
728 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
732 s->metadata_entries.parent_locator_entry = md_entry;
733 s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT;
737 if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) {
738 /* cannot read vhdx file - required region table entry that
739 * we do not understand. per spec, we must fail to open */
745 if (s->metadata_entries.present != META_ALL_PRESENT) {
750 ret = bdrv_pread(bs->file,
751 s->metadata_entries.file_parameters_entry.offset
752 + s->metadata_rt.file_offset,
761 s->params.block_size = le32_to_cpu(s->params.block_size);
762 s->params.data_bits = le32_to_cpu(s->params.data_bits);
765 /* We now have the file parameters, so we can tell if this is a
766 * differencing file (i.e.. has_parent), is dynamic or fixed
767 * sized (leave_blocks_allocated), and the block size */
769 /* The parent locator required iff the file parameters has_parent set */
770 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
771 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
772 /* TODO: parse parent locator fields */
773 ret = -ENOTSUP; /* temp, until differencing files are supported */
776 /* if has_parent is set, but there is not parent locator present,
777 * then that is an invalid combination */
783 /* determine virtual disk size, logical sector size,
784 * and phys sector size */
786 ret = bdrv_pread(bs->file,
787 s->metadata_entries.virtual_disk_size_entry.offset
788 + s->metadata_rt.file_offset,
790 &s->virtual_disk_size,
795 ret = bdrv_pread(bs->file,
796 s->metadata_entries.logical_sector_size_entry.offset
797 + s->metadata_rt.file_offset,
799 &s->logical_sector_size,
804 ret = bdrv_pread(bs->file,
805 s->metadata_entries.phys_sector_size_entry.offset
806 + s->metadata_rt.file_offset,
808 &s->physical_sector_size,
814 s->virtual_disk_size = le64_to_cpu(s->virtual_disk_size);
815 s->logical_sector_size = le32_to_cpu(s->logical_sector_size);
816 s->physical_sector_size = le32_to_cpu(s->physical_sector_size);
818 if (s->params.block_size < VHDX_BLOCK_SIZE_MIN ||
819 s->params.block_size > VHDX_BLOCK_SIZE_MAX) {
824 /* Currently we only support 512 */
825 if (s->logical_sector_size != 512) {
830 /* Both block_size and sector_size are guaranteed powers of 2, below.
831 Due to range checks above, s->sectors_per_block can never be < 256 */
832 s->sectors_per_block = s->params.block_size / s->logical_sector_size;
833 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
834 (uint64_t)s->logical_sector_size /
835 (uint64_t)s->params.block_size;
837 /* These values are ones we will want to use for division / multiplication
838 * later on, and they are all guaranteed (per the spec) to be powers of 2,
839 * so we can take advantage of that for shift operations during
841 if (s->logical_sector_size & (s->logical_sector_size - 1)) {
845 if (s->sectors_per_block & (s->sectors_per_block - 1)) {
849 if (s->chunk_ratio & (s->chunk_ratio - 1)) {
853 s->block_size = s->params.block_size;
854 if (s->block_size & (s->block_size - 1)) {
859 vhdx_set_shift_bits(s);
869 * Calculate the number of BAT entries, including sector
872 static void vhdx_calc_bat_entries(BDRVVHDXState *s)
874 uint32_t data_blocks_cnt, bitmap_blocks_cnt;
876 data_blocks_cnt = DIV_ROUND_UP(s->virtual_disk_size, s->block_size);
877 bitmap_blocks_cnt = DIV_ROUND_UP(data_blocks_cnt, s->chunk_ratio);
879 if (s->parent_entries) {
880 s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1);
882 s->bat_entries = data_blocks_cnt +
883 ((data_blocks_cnt - 1) >> s->chunk_ratio_bits);
888 static int vhdx_check_bat_entries(BlockDriverState *bs, int *errcnt)
890 BDRVVHDXState *s = bs->opaque;
891 int64_t image_file_size = bdrv_getlength(bs->file->bs);
892 uint64_t payblocks = s->chunk_ratio;
896 if (image_file_size < 0) {
897 error_report("Could not determinate VHDX image file size.");
898 return image_file_size;
901 for (i = 0; i < s->bat_entries; i++) {
902 if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) ==
903 PAYLOAD_BLOCK_FULLY_PRESENT) {
904 uint64_t offset = s->bat[i] & VHDX_BAT_FILE_OFF_MASK;
906 * Allow that the last block exists only partially. The VHDX spec
907 * states that the image file can only grow in blocksize increments,
908 * but QEMU created images with partial last blocks in the past.
910 uint32_t block_length = MIN(s->block_size,
911 bs->total_sectors * BDRV_SECTOR_SIZE - i * s->block_size);
913 * Check for BAT entry overflow.
915 if (offset > INT64_MAX - s->block_size) {
916 error_report("VHDX BAT entry %" PRIu64 " offset overflow.", i);
924 * Check if fully allocated BAT entries do not reside after
925 * end of the image file.
927 if (offset >= image_file_size) {
928 error_report("VHDX BAT entry %" PRIu64 " start offset %" PRIu64
929 " points after end of file (%" PRIi64 "). Image"
930 " has probably been truncated.",
931 i, offset, image_file_size);
937 } else if (offset + block_length > image_file_size) {
938 error_report("VHDX BAT entry %" PRIu64 " end offset %" PRIu64
939 " points after end of file (%" PRIi64 "). Image"
940 " has probably been truncated.",
941 i, offset + block_length - 1, image_file_size);
950 * verify populated BAT field file offsets against
951 * region table and log entries
954 /* payload bat entries */
956 ret2 = vhdx_region_check(s, offset, s->block_size);
965 payblocks = s->chunk_ratio;
967 * Once differencing files are supported, verify sector bitmap
977 static void vhdx_close(BlockDriverState *bs)
979 BDRVVHDXState *s = bs->opaque;
980 qemu_vfree(s->headers[0]);
981 s->headers[0] = NULL;
982 qemu_vfree(s->headers[1]);
983 s->headers[1] = NULL;
986 qemu_vfree(s->parent_entries);
987 s->parent_entries = NULL;
988 migrate_del_blocker(s->migration_blocker);
989 error_free(s->migration_blocker);
990 qemu_vfree(s->log.hdr);
992 vhdx_region_unregister_all(s);
995 static int vhdx_open(BlockDriverState *bs, QDict *options, int flags,
998 BDRVVHDXState *s = bs->opaque;
1002 Error *local_err = NULL;
1004 ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
1010 s->first_visible_write = true;
1012 qemu_co_mutex_init(&s->lock);
1013 QLIST_INIT(&s->regions);
1015 /* validate the file signature */
1016 ret = bdrv_pread(bs->file, 0, sizeof(uint64_t), &signature, 0);
1020 if (memcmp(&signature, "vhdxfile", 8)) {
1025 /* This is used for any header updates, for the file_write_guid.
1026 * The spec dictates that a new value should be used for the first
1028 vhdx_guid_generate(&s->session_guid);
1030 vhdx_parse_header(bs, s, &local_err);
1031 if (local_err != NULL) {
1032 error_propagate(errp, local_err);
1037 ret = vhdx_parse_log(bs, s, &s->log_replayed_on_open, errp);
1042 ret = vhdx_open_region_tables(bs, s);
1047 ret = vhdx_parse_metadata(bs, s);
1052 s->block_size = s->params.block_size;
1054 /* the VHDX spec dictates that virtual_disk_size is always a multiple of
1055 * logical_sector_size */
1056 bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits;
1058 vhdx_calc_bat_entries(s);
1060 s->bat_offset = s->bat_rt.file_offset;
1062 if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) {
1063 /* BAT allocation is not large enough for all entries */
1068 /* s->bat is freed in vhdx_close() */
1069 s->bat = qemu_try_blockalign(bs->file->bs, s->bat_rt.length);
1070 if (s->bat == NULL) {
1075 ret = bdrv_pread(bs->file, s->bat_offset, s->bat_rt.length, s->bat, 0);
1080 /* endian convert populated BAT field entires */
1081 for (i = 0; i < s->bat_entries; i++) {
1082 s->bat[i] = le64_to_cpu(s->bat[i]);
1085 if (!(flags & BDRV_O_CHECK)) {
1086 ret = vhdx_check_bat_entries(bs, NULL);
1092 /* Disable migration when VHDX images are used */
1093 error_setg(&s->migration_blocker, "The vhdx format used by node '%s' "
1094 "does not support live migration",
1095 bdrv_get_device_or_node_name(bs));
1096 ret = migrate_add_blocker(s->migration_blocker, errp);
1098 error_free(s->migration_blocker);
1102 /* TODO: differencing files */
1110 static int vhdx_reopen_prepare(BDRVReopenState *state,
1111 BlockReopenQueue *queue, Error **errp)
1118 * Perform sector to block offset translations, to get various
1119 * sector and file offsets into the image. See VHDXSectorInfo
1121 static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num,
1122 int nb_sectors, VHDXSectorInfo *sinfo)
1124 uint32_t block_offset;
1126 sinfo->bat_idx = sector_num >> s->sectors_per_block_bits;
1127 /* effectively a modulo - this gives us the offset into the block
1128 * (in sector sizes) for our sector number */
1129 block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits);
1130 /* the chunk ratio gives us the interleaving of the sector
1131 * bitmaps, so we need to advance our page block index by the
1132 * sector bitmaps entry number */
1133 sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits;
1135 /* the number of sectors we can read/write in this cycle */
1136 sinfo->sectors_avail = s->sectors_per_block - block_offset;
1138 sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits;
1140 if (sinfo->sectors_avail > nb_sectors) {
1141 sinfo->sectors_avail = nb_sectors;
1144 sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits;
1146 sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK;
1148 sinfo->block_offset = block_offset << s->logical_sector_size_bits;
1150 /* The file offset must be past the header section, so must be > 0 */
1151 if (sinfo->file_offset == 0) {
1155 /* block offset is the offset in vhdx logical sectors, in
1156 * the payload data block. Convert that to a byte offset
1157 * in the block, and add in the payload data block offset
1158 * in the file, in bytes, to get the final read address */
1160 sinfo->file_offset += sinfo->block_offset;
1164 static int coroutine_fn
1165 vhdx_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1167 BDRVVHDXState *s = bs->opaque;
1169 bdi->cluster_size = s->block_size;
1175 static int coroutine_fn GRAPH_RDLOCK
1176 vhdx_co_readv(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1179 BDRVVHDXState *s = bs->opaque;
1181 VHDXSectorInfo sinfo;
1182 uint64_t bytes_done = 0;
1183 QEMUIOVector hd_qiov;
1185 qemu_iovec_init(&hd_qiov, qiov->niov);
1187 qemu_co_mutex_lock(&s->lock);
1189 while (nb_sectors > 0) {
1190 /* We are a differencing file, so we need to inspect the sector bitmap
1191 * to see if we have the data or not */
1192 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1193 /* not supported yet */
1197 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1199 qemu_iovec_reset(&hd_qiov);
1200 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, sinfo.bytes_avail);
1202 /* check the payload block state */
1203 switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) {
1204 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1205 case PAYLOAD_BLOCK_UNDEFINED:
1206 case PAYLOAD_BLOCK_UNMAPPED:
1207 case PAYLOAD_BLOCK_UNMAPPED_v095:
1208 case PAYLOAD_BLOCK_ZERO:
1210 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail);
1212 case PAYLOAD_BLOCK_FULLY_PRESENT:
1213 qemu_co_mutex_unlock(&s->lock);
1214 ret = bdrv_co_preadv(bs->file, sinfo.file_offset,
1215 sinfo.sectors_avail * BDRV_SECTOR_SIZE,
1217 qemu_co_mutex_lock(&s->lock);
1222 case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1223 /* we don't yet support difference files, fall through
1230 nb_sectors -= sinfo.sectors_avail;
1231 sector_num += sinfo.sectors_avail;
1232 bytes_done += sinfo.bytes_avail;
1237 qemu_co_mutex_unlock(&s->lock);
1238 qemu_iovec_destroy(&hd_qiov);
1243 * Allocate a new payload block at the end of the file.
1245 * Allocation will happen at 1MB alignment inside the file.
1247 * If @need_zero is set on entry but not cleared on return, then truncation
1248 * could not guarantee that the new portion reads as zero, and the caller
1249 * will take care of it instead.
1251 * Returns the file offset start of the new payload block
1253 static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s,
1254 uint64_t *new_offset, bool *need_zero)
1256 int64_t current_len;
1258 current_len = bdrv_getlength(bs->file->bs);
1259 if (current_len < 0) {
1263 *new_offset = current_len;
1265 /* per the spec, the address for a block is in units of 1MB */
1266 *new_offset = ROUND_UP(*new_offset, 1 * MiB);
1267 if (*new_offset > INT64_MAX) {
1274 ret = bdrv_truncate(bs->file, *new_offset + s->block_size, false,
1275 PREALLOC_MODE_OFF, BDRV_REQ_ZERO_WRITE, NULL);
1276 if (ret != -ENOTSUP) {
1282 return bdrv_truncate(bs->file, *new_offset + s->block_size, false,
1283 PREALLOC_MODE_OFF, 0, NULL);
1287 * Update the BAT table entry with the new file offset, and the new entry
1289 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s,
1290 VHDXSectorInfo *sinfo,
1291 uint64_t *bat_entry_le,
1292 uint64_t *bat_offset, int state)
1294 /* The BAT entry is a uint64, with 44 bits for the file offset in units of
1295 * 1MB, and 3 bits for the block state. */
1296 if ((state == PAYLOAD_BLOCK_ZERO) ||
1297 (state == PAYLOAD_BLOCK_UNDEFINED) ||
1298 (state == PAYLOAD_BLOCK_NOT_PRESENT) ||
1299 (state == PAYLOAD_BLOCK_UNMAPPED)) {
1300 s->bat[sinfo->bat_idx] = 0; /* For PAYLOAD_BLOCK_ZERO, the
1301 FileOffsetMB field is denoted as
1302 'reserved' in the v1.0 spec. If it is
1303 non-zero, MS Hyper-V will fail to read
1306 s->bat[sinfo->bat_idx] = sinfo->file_offset;
1309 s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK;
1311 *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]);
1312 *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry);
1316 /* Per the spec, on the first write of guest-visible data to the file the
1317 * data write guid must be updated in the header */
1318 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s)
1321 if (s->first_visible_write) {
1322 s->first_visible_write = false;
1323 ret = vhdx_update_headers(bs, s, true, NULL);
1328 static int coroutine_fn GRAPH_RDLOCK
1329 vhdx_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1330 QEMUIOVector *qiov, int flags)
1333 BDRVVHDXState *s = bs->opaque;
1334 VHDXSectorInfo sinfo;
1335 uint64_t bytes_done = 0;
1336 uint64_t bat_entry = 0;
1337 uint64_t bat_entry_offset = 0;
1338 QEMUIOVector hd_qiov;
1339 struct iovec iov1 = { 0 };
1340 struct iovec iov2 = { 0 };
1341 int sectors_to_write;
1343 uint64_t bat_prior_offset = 0;
1344 bool bat_update = false;
1346 qemu_iovec_init(&hd_qiov, qiov->niov);
1348 qemu_co_mutex_lock(&s->lock);
1350 ret = vhdx_user_visible_write(bs, s);
1355 while (nb_sectors > 0) {
1356 bool use_zero_buffers = false;
1358 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1359 /* not supported yet */
1363 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1364 sectors_to_write = sinfo.sectors_avail;
1366 qemu_iovec_reset(&hd_qiov);
1367 /* check the payload block state */
1368 bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK;
1369 switch (bat_state) {
1370 case PAYLOAD_BLOCK_ZERO:
1371 /* in this case, we need to preserve zero writes for
1372 * data that is not part of this write, so we must pad
1373 * the rest of the buffer to zeroes */
1374 use_zero_buffers = true;
1376 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1377 case PAYLOAD_BLOCK_UNMAPPED:
1378 case PAYLOAD_BLOCK_UNMAPPED_v095:
1379 case PAYLOAD_BLOCK_UNDEFINED:
1380 bat_prior_offset = sinfo.file_offset;
1381 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset,
1387 * once we support differencing files, this may also be
1390 /* update block state to the newly specified state */
1391 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1393 PAYLOAD_BLOCK_FULLY_PRESENT);
1396 * Since we just allocated a block, file_offset is the
1397 * beginning of the payload block. It needs to be the
1398 * write address, which includes the offset into the
1399 * block, unless the entire block needs to read as
1400 * zeroes but truncation was not able to provide them,
1401 * in which case we need to fill in the rest.
1403 if (!use_zero_buffers) {
1404 sinfo.file_offset += sinfo.block_offset;
1406 /* zero fill the front, if any */
1407 if (sinfo.block_offset) {
1408 iov1.iov_len = sinfo.block_offset;
1409 iov1.iov_base = qemu_blockalign(bs, iov1.iov_len);
1410 memset(iov1.iov_base, 0, iov1.iov_len);
1411 qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0,
1413 sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS;
1416 /* our actual data */
1417 qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1420 /* zero fill the back, if any */
1421 if ((sinfo.bytes_avail - sinfo.block_offset) <
1423 iov2.iov_len = s->block_size -
1424 (sinfo.bytes_avail + sinfo.block_offset);
1425 iov2.iov_base = qemu_blockalign(bs, iov2.iov_len);
1426 memset(iov2.iov_base, 0, iov2.iov_len);
1427 qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0,
1429 sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS;
1434 case PAYLOAD_BLOCK_FULLY_PRESENT:
1435 /* if the file offset address is in the header zone,
1436 * there is a problem */
1437 if (sinfo.file_offset < (1 * MiB)) {
1439 goto error_bat_restore;
1442 if (!use_zero_buffers) {
1443 qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1446 /* block exists, so we can just overwrite it */
1447 qemu_co_mutex_unlock(&s->lock);
1448 ret = bdrv_co_pwritev(bs->file, sinfo.file_offset,
1449 sectors_to_write * BDRV_SECTOR_SIZE,
1451 qemu_co_mutex_lock(&s->lock);
1453 goto error_bat_restore;
1456 case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1457 /* we don't yet support difference files, fall through
1466 /* this will update the BAT entry into the log journal, and
1467 * then flush the log journal out to disk */
1468 ret = vhdx_log_write_and_flush(bs, s, &bat_entry,
1469 sizeof(VHDXBatEntry),
1476 nb_sectors -= sinfo.sectors_avail;
1477 sector_num += sinfo.sectors_avail;
1478 bytes_done += sinfo.bytes_avail;
1487 /* keep metadata in sync, and restore the bat entry state
1489 sinfo.file_offset = bat_prior_offset;
1490 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1491 &bat_entry_offset, bat_state);
1494 qemu_vfree(iov1.iov_base);
1495 qemu_vfree(iov2.iov_base);
1496 qemu_co_mutex_unlock(&s->lock);
1497 qemu_iovec_destroy(&hd_qiov);
1504 * Create VHDX Headers
1506 * There are 2 headers, and the highest sequence number will represent
1509 static int vhdx_create_new_headers(BlockBackend *blk, uint64_t image_size,
1512 BlockDriverState *bs = blk_bs(blk);
1515 VHDXHeader *hdr = NULL;
1517 hdr = g_new0(VHDXHeader, 1);
1519 hdr->signature = VHDX_HEADER_SIGNATURE;
1520 hdr->sequence_number = g_random_int();
1521 hdr->log_version = 0;
1523 hdr->log_length = log_size;
1524 hdr->log_offset = VHDX_HEADER_SECTION_END;
1525 vhdx_guid_generate(&hdr->file_write_guid);
1526 vhdx_guid_generate(&hdr->data_write_guid);
1528 /* XXX Ugly way to get blk->root, but that's a feature, not a bug. This
1529 * hack makes it obvious that vhdx_write_header() bypasses the BlockBackend
1530 * here, which it really shouldn't be doing. */
1531 child = QLIST_FIRST(&bs->parents);
1532 assert(!QLIST_NEXT(child, next_parent));
1534 ret = vhdx_write_header(child, hdr, VHDX_HEADER1_OFFSET, false);
1538 hdr->sequence_number++;
1539 ret = vhdx_write_header(child, hdr, VHDX_HEADER2_OFFSET, false);
1549 #define VHDX_METADATA_ENTRY_BUFFER_SIZE \
1550 (sizeof(VHDXFileParameters) +\
1551 sizeof(VHDXVirtualDiskSize) +\
1552 sizeof(VHDXPage83Data) +\
1553 sizeof(VHDXVirtualDiskLogicalSectorSize) +\
1554 sizeof(VHDXVirtualDiskPhysicalSectorSize))
1557 * Create the Metadata entries.
1559 * For more details on the entries, see section 3.5 (pg 29) in the
1560 * VHDX 1.00 specification.
1562 * We support 5 metadata entries (all required by spec):
1564 * Virtual Disk Size,
1566 * Logical Sector Size,
1567 * Physical Sector Size
1569 * The first 64KB of the Metadata section is reserved for the metadata
1570 * header and entries; beyond that, the metadata items themselves reside.
1572 static int vhdx_create_new_metadata(BlockBackend *blk,
1573 uint64_t image_size,
1574 uint32_t block_size,
1575 uint32_t sector_size,
1576 uint64_t metadata_offset,
1580 uint32_t offset = 0;
1581 void *buffer = NULL;
1583 VHDXMetadataTableHeader *md_table;
1584 VHDXMetadataTableEntry *md_table_entry;
1586 /* Metadata entries */
1587 VHDXFileParameters *mt_file_params;
1588 VHDXVirtualDiskSize *mt_virtual_size;
1589 VHDXPage83Data *mt_page83;
1590 VHDXVirtualDiskLogicalSectorSize *mt_log_sector_size;
1591 VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size;
1593 entry_buffer = g_malloc0(VHDX_METADATA_ENTRY_BUFFER_SIZE);
1595 mt_file_params = entry_buffer;
1596 offset += sizeof(VHDXFileParameters);
1597 mt_virtual_size = entry_buffer + offset;
1598 offset += sizeof(VHDXVirtualDiskSize);
1599 mt_page83 = entry_buffer + offset;
1600 offset += sizeof(VHDXPage83Data);
1601 mt_log_sector_size = entry_buffer + offset;
1602 offset += sizeof(VHDXVirtualDiskLogicalSectorSize);
1603 mt_phys_sector_size = entry_buffer + offset;
1605 mt_file_params->block_size = cpu_to_le32(block_size);
1606 if (type == VHDX_TYPE_FIXED) {
1607 mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED;
1608 mt_file_params->data_bits = cpu_to_le32(mt_file_params->data_bits);
1611 vhdx_guid_generate(&mt_page83->page_83_data);
1612 cpu_to_leguids(&mt_page83->page_83_data);
1613 mt_virtual_size->virtual_disk_size = cpu_to_le64(image_size);
1614 mt_log_sector_size->logical_sector_size = cpu_to_le32(sector_size);
1615 mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size);
1617 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1620 md_table->signature = VHDX_METADATA_SIGNATURE;
1621 md_table->entry_count = 5;
1622 vhdx_metadata_header_le_export(md_table);
1625 /* This will reference beyond the reserved table portion */
1628 md_table_entry = buffer + sizeof(VHDXMetadataTableHeader);
1630 md_table_entry[0].item_id = file_param_guid;
1631 md_table_entry[0].offset = offset;
1632 md_table_entry[0].length = sizeof(VHDXFileParameters);
1633 md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED;
1634 offset += md_table_entry[0].length;
1635 vhdx_metadata_entry_le_export(&md_table_entry[0]);
1637 md_table_entry[1].item_id = virtual_size_guid;
1638 md_table_entry[1].offset = offset;
1639 md_table_entry[1].length = sizeof(VHDXVirtualDiskSize);
1640 md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1641 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1642 offset += md_table_entry[1].length;
1643 vhdx_metadata_entry_le_export(&md_table_entry[1]);
1645 md_table_entry[2].item_id = page83_guid;
1646 md_table_entry[2].offset = offset;
1647 md_table_entry[2].length = sizeof(VHDXPage83Data);
1648 md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1649 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1650 offset += md_table_entry[2].length;
1651 vhdx_metadata_entry_le_export(&md_table_entry[2]);
1653 md_table_entry[3].item_id = logical_sector_guid;
1654 md_table_entry[3].offset = offset;
1655 md_table_entry[3].length = sizeof(VHDXVirtualDiskLogicalSectorSize);
1656 md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1657 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1658 offset += md_table_entry[3].length;
1659 vhdx_metadata_entry_le_export(&md_table_entry[3]);
1661 md_table_entry[4].item_id = phys_sector_guid;
1662 md_table_entry[4].offset = offset;
1663 md_table_entry[4].length = sizeof(VHDXVirtualDiskPhysicalSectorSize);
1664 md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1665 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1666 vhdx_metadata_entry_le_export(&md_table_entry[4]);
1668 ret = blk_pwrite(blk, metadata_offset, VHDX_HEADER_BLOCK_SIZE, buffer, 0);
1673 ret = blk_pwrite(blk, metadata_offset + (64 * KiB),
1674 VHDX_METADATA_ENTRY_BUFFER_SIZE, entry_buffer, 0);
1682 g_free(entry_buffer);
1686 /* This create the actual BAT itself. We currently only support
1687 * 'Dynamic' and 'Fixed' image types.
1689 * Dynamic images: default state of the BAT is all zeroes.
1691 * Fixed images: default state of the BAT is fully populated, with
1692 * file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT.
1694 static int vhdx_create_bat(BlockBackend *blk, BDRVVHDXState *s,
1695 uint64_t image_size, VHDXImageType type,
1696 bool use_zero_blocks, uint64_t file_offset,
1697 uint32_t length, Error **errp)
1700 uint64_t data_file_offset;
1701 uint64_t total_sectors = 0;
1702 uint64_t sector_num = 0;
1705 VHDXSectorInfo sinfo;
1707 assert(s->bat == NULL);
1709 /* this gives a data start after BAT/bitmap entries, and well
1710 * past any metadata entries (with a 4 MB buffer for future
1712 data_file_offset = file_offset + length + 5 * MiB;
1713 total_sectors = image_size >> s->logical_sector_size_bits;
1715 if (type == VHDX_TYPE_DYNAMIC) {
1716 /* All zeroes, so we can just extend the file - the end of the BAT
1717 * is the furthest thing we have written yet */
1718 ret = blk_truncate(blk, data_file_offset, false, PREALLOC_MODE_OFF,
1723 } else if (type == VHDX_TYPE_FIXED) {
1724 ret = blk_truncate(blk, data_file_offset + image_size, false,
1725 PREALLOC_MODE_OFF, 0, errp);
1730 error_setg(errp, "Unsupported image type");
1735 if (type == VHDX_TYPE_FIXED ||
1737 bdrv_has_zero_init(blk_bs(blk)) == 0) {
1738 /* for a fixed file, the default BAT entry is not zero */
1739 s->bat = g_try_malloc0(length);
1740 if (length && s->bat == NULL) {
1741 error_setg(errp, "Failed to allocate memory for the BAT");
1745 block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT :
1746 PAYLOAD_BLOCK_NOT_PRESENT;
1747 block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state;
1748 /* fill the BAT by emulating sector writes of sectors_per_block size */
1749 while (sector_num < total_sectors) {
1750 vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo);
1751 sinfo.file_offset = data_file_offset +
1752 (sector_num << s->logical_sector_size_bits);
1753 sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB);
1754 vhdx_update_bat_table_entry(blk_bs(blk), s, &sinfo, &unused, &unused,
1756 s->bat[sinfo.bat_idx] = cpu_to_le64(s->bat[sinfo.bat_idx]);
1757 sector_num += s->sectors_per_block;
1759 ret = blk_pwrite(blk, file_offset, length, s->bat, 0);
1761 error_setg_errno(errp, -ret, "Failed to write the BAT");
1773 /* Creates the region table header, and region table entries.
1774 * There are 2 supported region table entries: BAT, and Metadata/
1776 * As the calculations for the BAT region table are also needed
1777 * to create the BAT itself, we will also cause the BAT to be
1780 static int vhdx_create_new_region_table(BlockBackend *blk,
1781 uint64_t image_size,
1782 uint32_t block_size,
1783 uint32_t sector_size,
1785 bool use_zero_blocks,
1787 uint64_t *metadata_offset,
1791 uint32_t offset = 0;
1792 void *buffer = NULL;
1793 uint64_t bat_file_offset;
1794 uint32_t bat_length;
1795 BDRVVHDXState *s = NULL;
1796 VHDXRegionTableHeader *region_table;
1797 VHDXRegionTableEntry *rt_bat;
1798 VHDXRegionTableEntry *rt_metadata;
1800 assert(metadata_offset != NULL);
1802 /* Populate enough of the BDRVVHDXState to be able to use the
1803 * pre-existing BAT calculation, translation, and update functions */
1804 s = g_new0(BDRVVHDXState, 1);
1806 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
1807 (uint64_t) sector_size / (uint64_t) block_size;
1809 s->sectors_per_block = block_size / sector_size;
1810 s->virtual_disk_size = image_size;
1811 s->block_size = block_size;
1812 s->logical_sector_size = sector_size;
1814 vhdx_set_shift_bits(s);
1816 vhdx_calc_bat_entries(s);
1818 /* At this point the VHDX state is populated enough for creation */
1820 /* a single buffer is used so we can calculate the checksum over the
1821 * entire 64KB block */
1822 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1823 region_table = buffer;
1824 offset += sizeof(VHDXRegionTableHeader);
1825 rt_bat = buffer + offset;
1826 offset += sizeof(VHDXRegionTableEntry);
1827 rt_metadata = buffer + offset;
1829 region_table->signature = VHDX_REGION_SIGNATURE;
1830 region_table->entry_count = 2; /* BAT and Metadata */
1832 rt_bat->guid = bat_guid;
1833 rt_bat->length = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB);
1834 rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB);
1835 s->bat_offset = rt_bat->file_offset;
1837 rt_metadata->guid = metadata_guid;
1838 rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length,
1840 rt_metadata->length = 1 * MiB; /* min size, and more than enough */
1841 *metadata_offset = rt_metadata->file_offset;
1843 bat_file_offset = rt_bat->file_offset;
1844 bat_length = rt_bat->length;
1846 vhdx_region_header_le_export(region_table);
1847 vhdx_region_entry_le_export(rt_bat);
1848 vhdx_region_entry_le_export(rt_metadata);
1850 vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE,
1851 offsetof(VHDXRegionTableHeader, checksum));
1854 /* The region table gives us the data we need to create the BAT,
1856 ret = vhdx_create_bat(blk, s, image_size, type, use_zero_blocks,
1857 bat_file_offset, bat_length, errp);
1862 /* Now write out the region headers to disk */
1863 ret = blk_pwrite(blk, VHDX_REGION_TABLE_OFFSET, VHDX_HEADER_BLOCK_SIZE,
1866 error_setg_errno(errp, -ret, "Failed to write first region table");
1870 ret = blk_pwrite(blk, VHDX_REGION_TABLE2_OFFSET, VHDX_HEADER_BLOCK_SIZE,
1873 error_setg_errno(errp, -ret, "Failed to write second region table");
1883 /* We need to create the following elements:
1885 * .-----------------------------------------------------------------.
1886 * | (A) | (B) | (C) | (D) | (E) |
1887 * | File ID | Header1 | Header 2 | Region Tbl 1 | Region Tbl 2 |
1889 * .-----------------------------------------------------------------.
1890 * 0 64KB 128KB 192KB 256KB 320KB
1893 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1894 * | (F) | (G) | (H) | |
1895 * | Journal Log | BAT / Bitmap | Metadata | .... data ...... |
1897 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1900 static int coroutine_fn vhdx_co_create(BlockdevCreateOptions *opts,
1903 BlockdevCreateOptionsVhdx *vhdx_opts;
1904 BlockBackend *blk = NULL;
1905 BlockDriverState *bs = NULL;
1908 uint64_t image_size;
1910 uint32_t block_size;
1912 uint64_t metadata_offset;
1913 bool use_zero_blocks = false;
1915 gunichar2 *creator = NULL;
1916 glong creator_items;
1917 VHDXImageType image_type;
1919 assert(opts->driver == BLOCKDEV_DRIVER_VHDX);
1920 vhdx_opts = &opts->u.vhdx;
1922 /* Validate options and set default values */
1923 image_size = vhdx_opts->size;
1924 if (image_size > VHDX_MAX_IMAGE_SIZE) {
1925 error_setg(errp, "Image size too large; max of 64TB");
1929 if (!vhdx_opts->has_log_size) {
1930 log_size = DEFAULT_LOG_SIZE;
1932 if (vhdx_opts->log_size > UINT32_MAX) {
1933 error_setg(errp, "Log size must be smaller than 4 GB");
1936 log_size = vhdx_opts->log_size;
1938 if (log_size < MiB || (log_size % MiB) != 0) {
1939 error_setg(errp, "Log size must be a multiple of 1 MB");
1943 if (!vhdx_opts->has_block_state_zero) {
1944 use_zero_blocks = true;
1946 use_zero_blocks = vhdx_opts->block_state_zero;
1949 if (!vhdx_opts->has_subformat) {
1950 vhdx_opts->subformat = BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC;
1953 switch (vhdx_opts->subformat) {
1954 case BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC:
1955 image_type = VHDX_TYPE_DYNAMIC;
1957 case BLOCKDEV_VHDX_SUBFORMAT_FIXED:
1958 image_type = VHDX_TYPE_FIXED;
1961 g_assert_not_reached();
1964 /* These are pretty arbitrary, and mainly designed to keep the BAT
1965 * size reasonable to load into RAM */
1966 if (vhdx_opts->has_block_size) {
1967 block_size = vhdx_opts->block_size;
1969 if (image_size > 32 * TiB) {
1970 block_size = 64 * MiB;
1971 } else if (image_size > (uint64_t) 100 * GiB) {
1972 block_size = 32 * MiB;
1973 } else if (image_size > 1 * GiB) {
1974 block_size = 16 * MiB;
1976 block_size = 8 * MiB;
1980 if (block_size < MiB || (block_size % MiB) != 0) {
1981 error_setg(errp, "Block size must be a multiple of 1 MB");
1984 if (!is_power_of_2(block_size)) {
1985 error_setg(errp, "Block size must be a power of two");
1988 if (block_size > VHDX_BLOCK_SIZE_MAX) {
1989 error_setg(errp, "Block size must not exceed %" PRId64,
1990 VHDX_BLOCK_SIZE_MAX);
1994 /* Create BlockBackend to write to the image */
1995 bs = bdrv_co_open_blockdev_ref(vhdx_opts->file, errp);
2000 blk = blk_co_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
2004 goto delete_and_exit;
2006 blk_set_allow_write_beyond_eof(blk, true);
2010 /* The creator field is optional, but may be useful for
2011 * debugging / diagnostics */
2012 creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL,
2013 &creator_items, NULL);
2014 signature = cpu_to_le64(VHDX_FILE_SIGNATURE);
2015 ret = blk_co_pwrite(blk, VHDX_FILE_ID_OFFSET, sizeof(signature), &signature,
2018 error_setg_errno(errp, -ret, "Failed to write file signature");
2019 goto delete_and_exit;
2022 ret = blk_co_pwrite(blk, VHDX_FILE_ID_OFFSET + sizeof(signature),
2023 creator_items * sizeof(gunichar2), creator, 0);
2025 error_setg_errno(errp, -ret, "Failed to write creator field");
2026 goto delete_and_exit;
2031 /* Creates (B),(C) */
2032 ret = vhdx_create_new_headers(blk, image_size, log_size);
2034 error_setg_errno(errp, -ret, "Failed to write image headers");
2035 goto delete_and_exit;
2038 /* Creates (D),(E),(G) explicitly. (F) created as by-product */
2039 ret = vhdx_create_new_region_table(blk, image_size, block_size, 512,
2040 log_size, use_zero_blocks, image_type,
2041 &metadata_offset, errp);
2043 goto delete_and_exit;
2047 ret = vhdx_create_new_metadata(blk, image_size, block_size, 512,
2048 metadata_offset, image_type);
2050 error_setg_errno(errp, -ret, "Failed to initialize metadata");
2051 goto delete_and_exit;
2062 static int coroutine_fn vhdx_co_create_opts(BlockDriver *drv,
2063 const char *filename,
2067 BlockdevCreateOptions *create_options = NULL;
2070 BlockDriverState *bs = NULL;
2073 static const QDictRenames opt_renames[] = {
2074 { VHDX_BLOCK_OPT_LOG_SIZE, "log-size" },
2075 { VHDX_BLOCK_OPT_BLOCK_SIZE, "block-size" },
2076 { VHDX_BLOCK_OPT_ZERO, "block-state-zero" },
2080 /* Parse options and convert legacy syntax */
2081 qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vhdx_create_opts, true);
2083 if (!qdict_rename_keys(qdict, opt_renames, errp)) {
2088 /* Create and open the file (protocol layer) */
2089 ret = bdrv_co_create_file(filename, opts, errp);
2094 bs = bdrv_co_open(filename, NULL, NULL,
2095 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
2101 /* Now get the QAPI type BlockdevCreateOptions */
2102 qdict_put_str(qdict, "driver", "vhdx");
2103 qdict_put_str(qdict, "file", bs->node_name);
2105 v = qobject_input_visitor_new_flat_confused(qdict, errp);
2111 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
2113 if (!create_options) {
2118 /* Silently round up sizes:
2119 * The image size is rounded to 512 bytes. Make the block and log size
2120 * close to what was specified, but must be at least 1MB, and a multiple of
2121 * 1 MB. Also respect VHDX_BLOCK_SIZE_MAX for block sizes. block_size = 0
2122 * means auto, which is represented by a missing key in QAPI. */
2123 assert(create_options->driver == BLOCKDEV_DRIVER_VHDX);
2124 create_options->u.vhdx.size =
2125 ROUND_UP(create_options->u.vhdx.size, BDRV_SECTOR_SIZE);
2127 if (create_options->u.vhdx.has_log_size) {
2128 create_options->u.vhdx.log_size =
2129 ROUND_UP(create_options->u.vhdx.log_size, MiB);
2131 if (create_options->u.vhdx.has_block_size) {
2132 create_options->u.vhdx.block_size =
2133 ROUND_UP(create_options->u.vhdx.block_size, MiB);
2135 if (create_options->u.vhdx.block_size == 0) {
2136 create_options->u.vhdx.has_block_size = false;
2138 if (create_options->u.vhdx.block_size > VHDX_BLOCK_SIZE_MAX) {
2139 create_options->u.vhdx.block_size = VHDX_BLOCK_SIZE_MAX;
2143 /* Create the vhdx image (format layer) */
2144 ret = vhdx_co_create(create_options, errp);
2147 qobject_unref(qdict);
2149 qapi_free_BlockdevCreateOptions(create_options);
2153 /* If opened r/w, the VHDX driver will automatically replay the log,
2154 * if one is present, inside the vhdx_open() call.
2156 * If qemu-img check -r all is called, the image is automatically opened
2157 * r/w and any log has already been replayed, so there is nothing (currently)
2160 static int coroutine_fn vhdx_co_check(BlockDriverState *bs,
2161 BdrvCheckResult *result,
2164 BDRVVHDXState *s = bs->opaque;
2166 if (s->log_replayed_on_open) {
2167 result->corruptions_fixed++;
2170 vhdx_check_bat_entries(bs, &result->corruptions);
2175 static int vhdx_has_zero_init(BlockDriverState *bs)
2177 BDRVVHDXState *s = bs->opaque;
2181 * Check the subformat: Fixed images have all BAT entries present,
2182 * dynamic images have none (right after creation). It is
2183 * therefore enough to check the first BAT entry.
2185 if (!s->bat_entries) {
2189 state = s->bat[0] & VHDX_BAT_STATE_BIT_MASK;
2190 if (state == PAYLOAD_BLOCK_FULLY_PRESENT) {
2191 /* Fixed subformat */
2192 return bdrv_has_zero_init(bs->file->bs);
2195 /* Dynamic subformat */
2199 static QemuOptsList vhdx_create_opts = {
2200 .name = "vhdx-create-opts",
2201 .head = QTAILQ_HEAD_INITIALIZER(vhdx_create_opts.head),
2204 .name = BLOCK_OPT_SIZE,
2205 .type = QEMU_OPT_SIZE,
2206 .help = "Virtual disk size; max of 64TB."
2209 .name = VHDX_BLOCK_OPT_LOG_SIZE,
2210 .type = QEMU_OPT_SIZE,
2211 .def_value_str = stringify(DEFAULT_LOG_SIZE),
2212 .help = "Log size; min 1MB."
2215 .name = VHDX_BLOCK_OPT_BLOCK_SIZE,
2216 .type = QEMU_OPT_SIZE,
2217 .def_value_str = stringify(0),
2218 .help = "Block Size; min 1MB, max 256MB. "
2219 "0 means auto-calculate based on image size."
2222 .name = BLOCK_OPT_SUBFMT,
2223 .type = QEMU_OPT_STRING,
2224 .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "
2225 "Default is 'dynamic'."
2228 .name = VHDX_BLOCK_OPT_ZERO,
2229 .type = QEMU_OPT_BOOL,
2230 .help = "Force use of payload blocks of type 'ZERO'. "
2231 "Non-standard, but default. Do not set to 'off' when "
2232 "using 'qemu-img convert' with subformat=dynamic."
2238 static BlockDriver bdrv_vhdx = {
2239 .format_name = "vhdx",
2240 .instance_size = sizeof(BDRVVHDXState),
2241 .bdrv_probe = vhdx_probe,
2242 .bdrv_open = vhdx_open,
2243 .bdrv_close = vhdx_close,
2244 .bdrv_reopen_prepare = vhdx_reopen_prepare,
2245 .bdrv_child_perm = bdrv_default_perms,
2246 .bdrv_co_readv = vhdx_co_readv,
2247 .bdrv_co_writev = vhdx_co_writev,
2248 .bdrv_co_create = vhdx_co_create,
2249 .bdrv_co_create_opts = vhdx_co_create_opts,
2250 .bdrv_co_get_info = vhdx_co_get_info,
2251 .bdrv_co_check = vhdx_co_check,
2252 .bdrv_has_zero_init = vhdx_has_zero_init,
2255 .create_opts = &vhdx_create_opts,
2258 static void bdrv_vhdx_init(void)
2260 bdrv_register(&bdrv_vhdx);
2263 block_init(bdrv_vhdx_init);