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-common.h"
19 #include "block/block_int.h"
20 #include "qemu/module.h"
21 #include "qemu/crc32c.h"
22 #include "block/vhdx.h"
23 #include "migration/migration.h"
25 #include <uuid/uuid.h>
28 /* Options for VHDX creation */
30 #define VHDX_BLOCK_OPT_LOG_SIZE "log_size"
31 #define VHDX_BLOCK_OPT_BLOCK_SIZE "block_size"
32 #define VHDX_BLOCK_OPT_ZERO "block_state_zero"
34 typedef enum VHDXImageType
{
35 VHDX_TYPE_DYNAMIC
= 0,
37 VHDX_TYPE_DIFFERENCING
, /* Currently unsupported */
40 /* Several metadata and region table data entries are identified by
41 * guids in a MS-specific GUID format. */
44 /* ------- Known Region Table GUIDs ---------------------- */
45 static const MSGUID bat_guid
= { .data1
= 0x2dc27766,
48 .data4
= { 0x9d, 0x64, 0x11, 0x5e,
49 0x9b, 0xfd, 0x4a, 0x08} };
51 static const MSGUID metadata_guid
= { .data1
= 0x8b7ca206,
54 .data4
= { 0xb8, 0xfe, 0x57, 0x5f,
55 0x05, 0x0f, 0x88, 0x6e} };
59 /* ------- Known Metadata Entry GUIDs ---------------------- */
60 static const MSGUID file_param_guid
= { .data1
= 0xcaa16737,
63 .data4
= { 0xb3, 0xb6, 0x33, 0xf0,
64 0xaa, 0x44, 0xe7, 0x6b} };
66 static const MSGUID virtual_size_guid
= { .data1
= 0x2FA54224,
69 .data4
= { 0xb2, 0x11, 0x5d, 0xbe,
70 0xd8, 0x3b, 0xf4, 0xb8} };
72 static const MSGUID page83_guid
= { .data1
= 0xbeca12ab,
75 .data4
= { 0x93, 0xef, 0xc3, 0x09,
76 0xe0, 0x00, 0xc7, 0x46} };
79 static const MSGUID phys_sector_guid
= { .data1
= 0xcda348c7,
82 .data4
= { 0x9c, 0xc9, 0xe9, 0x88,
83 0x52, 0x51, 0xc5, 0x56} };
85 static const MSGUID parent_locator_guid
= { .data1
= 0xa8d35f2d,
88 .data4
= { 0xab, 0xf7, 0xd3,
92 static const MSGUID logical_sector_guid
= { .data1
= 0x8141bf1d,
95 .data4
= { 0xba, 0x47, 0xf2,
99 /* Each parent type must have a valid GUID; this is for parent images
100 * of type 'VHDX'. If we were to allow e.g. a QCOW2 parent, we would
101 * need to make up our own QCOW2 GUID type */
102 static const MSGUID parent_vhdx_guid
= { .data1
= 0xb04aefb7,
105 .data4
= { 0xb7, 0x89, 0x25, 0xb8,
106 0xe9, 0x44, 0x59, 0x13} };
109 #define META_FILE_PARAMETER_PRESENT 0x01
110 #define META_VIRTUAL_DISK_SIZE_PRESENT 0x02
111 #define META_PAGE_83_PRESENT 0x04
112 #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08
113 #define META_PHYS_SECTOR_SIZE_PRESENT 0x10
114 #define META_PARENT_LOCATOR_PRESENT 0x20
116 #define META_ALL_PRESENT \
117 (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \
118 META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \
119 META_PHYS_SECTOR_SIZE_PRESENT)
122 typedef struct VHDXSectorInfo
{
123 uint32_t bat_idx
; /* BAT entry index */
124 uint32_t sectors_avail
; /* sectors available in payload block */
125 uint32_t bytes_left
; /* bytes left in the block after data to r/w */
126 uint32_t bytes_avail
; /* bytes available in payload block */
127 uint64_t file_offset
; /* absolute offset in bytes, in file */
128 uint64_t block_offset
; /* block offset, in bytes */
131 /* Calculates new checksum.
133 * Zero is substituted during crc calculation for the original crc field
134 * crc_offset: byte offset in buf of the buffer crc
135 * buf: buffer pointer
136 * size: size of buffer (must be > crc_offset+4)
138 * Note: The resulting checksum is in the CPU endianness, not necessarily
139 * in the file format endianness (LE). Any header export to disk should
140 * make sure that vhdx_header_le_export() is used to convert to the
143 uint32_t vhdx_update_checksum(uint8_t *buf
, size_t size
, int crc_offset
)
148 assert(size
> (crc_offset
+ sizeof(crc
)));
150 memset(buf
+ crc_offset
, 0, sizeof(crc
));
151 crc
= crc32c(0xffffffff, buf
, size
);
152 memcpy(buf
+ crc_offset
, &crc
, sizeof(crc
));
157 uint32_t vhdx_checksum_calc(uint32_t crc
, uint8_t *buf
, size_t size
,
164 if (crc_offset
> 0) {
165 memcpy(&crc_orig
, buf
+ crc_offset
, sizeof(crc_orig
));
166 memset(buf
+ crc_offset
, 0, sizeof(crc_orig
));
169 crc_new
= crc32c(crc
, buf
, size
);
170 if (crc_offset
> 0) {
171 memcpy(buf
+ crc_offset
, &crc_orig
, sizeof(crc_orig
));
177 /* Validates the checksum of the buffer, with an in-place CRC.
179 * Zero is substituted during crc calculation for the original crc field,
180 * and the crc field is restored afterwards. But the buffer will be modifed
181 * during the calculation, so this may not be not suitable for multi-threaded
184 * crc_offset: byte offset in buf of the buffer crc
185 * buf: buffer pointer
186 * size: size of buffer (must be > crc_offset+4)
188 * returns true if checksum is valid, false otherwise
190 bool vhdx_checksum_is_valid(uint8_t *buf
, size_t size
, int crc_offset
)
196 assert(size
> (crc_offset
+ 4));
198 memcpy(&crc_orig
, buf
+ crc_offset
, sizeof(crc_orig
));
199 crc_orig
= le32_to_cpu(crc_orig
);
201 crc
= vhdx_checksum_calc(0xffffffff, buf
, size
, crc_offset
);
203 return crc
== crc_orig
;
208 * This generates a UUID that is compliant with the MS GUIDs used
209 * in the VHDX spec (and elsewhere).
211 void vhdx_guid_generate(MSGUID
*guid
)
214 assert(guid
!= NULL
);
217 memcpy(guid
, uuid
, sizeof(MSGUID
));
220 /* Check for region overlaps inside the VHDX image */
221 static int vhdx_region_check(BDRVVHDXState
*s
, uint64_t start
, uint64_t length
)
227 end
= start
+ length
;
228 QLIST_FOREACH(r
, &s
->regions
, entries
) {
229 if (!((start
>= r
->end
) || (end
<= r
->start
))) {
239 /* Register a region for future checks */
240 static void vhdx_region_register(BDRVVHDXState
*s
,
241 uint64_t start
, uint64_t length
)
245 r
= g_malloc0(sizeof(*r
));
248 r
->end
= start
+ length
;
250 QLIST_INSERT_HEAD(&s
->regions
, r
, entries
);
253 /* Free all registered regions */
254 static void vhdx_region_unregister_all(BDRVVHDXState
*s
)
256 VHDXRegionEntry
*r
, *r_next
;
258 QLIST_FOREACH_SAFE(r
, &s
->regions
, entries
, r_next
) {
259 QLIST_REMOVE(r
, entries
);
264 static void vhdx_set_shift_bits(BDRVVHDXState
*s
)
266 s
->logical_sector_size_bits
= 31 - clz32(s
->logical_sector_size
);
267 s
->sectors_per_block_bits
= 31 - clz32(s
->sectors_per_block
);
268 s
->chunk_ratio_bits
= 63 - clz64(s
->chunk_ratio
);
269 s
->block_size_bits
= 31 - clz32(s
->block_size
);
273 * Per the MS VHDX Specification, for every VHDX file:
274 * - The header section is fixed size - 1 MB
275 * - The header section is always the first "object"
276 * - The first 64KB of the header is the File Identifier
277 * - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile")
278 * - The following 512 bytes constitute a UTF-16 string identifiying the
279 * software that created the file, and is optional and diagnostic only.
281 * Therefore, we probe by looking for the vhdxfile signature "vhdxfile"
283 static int vhdx_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
285 if (buf_size
>= 8 && !memcmp(buf
, "vhdxfile", 8)) {
292 * Writes the header to the specified offset.
294 * This will optionally read in buffer data from disk (otherwise zero-fill),
295 * and then update the header checksum. Header is converted to proper
296 * endianness before being written to the specified file offset
298 static int vhdx_write_header(BlockDriverState
*bs_file
, VHDXHeader
*hdr
,
299 uint64_t offset
, bool read
)
301 uint8_t *buffer
= NULL
;
303 VHDXHeader header_le
;
305 assert(bs_file
!= NULL
);
308 /* the header checksum is not over just the packed size of VHDXHeader,
309 * but rather over the entire 'reserved' range for the header, which is
310 * 4KB (VHDX_HEADER_SIZE). */
312 buffer
= qemu_blockalign(bs_file
, VHDX_HEADER_SIZE
);
314 /* if true, we can't assume the extra reserved bytes are 0 */
315 ret
= bdrv_pread(bs_file
, offset
, buffer
, VHDX_HEADER_SIZE
);
320 memset(buffer
, 0, VHDX_HEADER_SIZE
);
323 /* overwrite the actual VHDXHeader portion */
324 memcpy(buffer
, hdr
, sizeof(VHDXHeader
));
325 hdr
->checksum
= vhdx_update_checksum(buffer
, VHDX_HEADER_SIZE
,
326 offsetof(VHDXHeader
, checksum
));
327 vhdx_header_le_export(hdr
, &header_le
);
328 ret
= bdrv_pwrite_sync(bs_file
, offset
, &header_le
, sizeof(VHDXHeader
));
335 /* Update the VHDX headers
337 * This follows the VHDX spec procedures for header updates.
339 * - non-current header is updated with largest sequence number
341 static int vhdx_update_header(BlockDriverState
*bs
, BDRVVHDXState
*s
,
342 bool generate_data_write_guid
, MSGUID
*log_guid
)
346 uint64_t header_offset
= VHDX_HEADER1_OFFSET
;
348 VHDXHeader
*active_header
;
349 VHDXHeader
*inactive_header
;
351 /* operate on the non-current header */
352 if (s
->curr_header
== 0) {
354 header_offset
= VHDX_HEADER2_OFFSET
;
357 active_header
= s
->headers
[s
->curr_header
];
358 inactive_header
= s
->headers
[hdr_idx
];
360 inactive_header
->sequence_number
= active_header
->sequence_number
+ 1;
362 /* a new file guid must be generated before any file write, including
364 inactive_header
->file_write_guid
= s
->session_guid
;
366 /* a new data guid only needs to be generated before any guest-visible
367 * writes (i.e. something observable via virtual disk read) */
368 if (generate_data_write_guid
) {
369 vhdx_guid_generate(&inactive_header
->data_write_guid
);
372 /* update the log guid if present */
374 inactive_header
->log_guid
= *log_guid
;
377 ret
= vhdx_write_header(bs
->file
, inactive_header
, header_offset
, true);
381 s
->curr_header
= hdr_idx
;
388 * The VHDX spec calls for header updates to be performed twice, so that both
389 * the current and non-current header have valid info
391 int vhdx_update_headers(BlockDriverState
*bs
, BDRVVHDXState
*s
,
392 bool generate_data_write_guid
, MSGUID
*log_guid
)
396 ret
= vhdx_update_header(bs
, s
, generate_data_write_guid
, log_guid
);
400 ret
= vhdx_update_header(bs
, s
, generate_data_write_guid
, log_guid
);
404 /* opens the specified header block from the VHDX file header section */
405 static void vhdx_parse_header(BlockDriverState
*bs
, BDRVVHDXState
*s
,
411 bool h1_valid
= false;
412 bool h2_valid
= false;
417 /* header1 & header2 are freed in vhdx_close() */
418 header1
= qemu_blockalign(bs
, sizeof(VHDXHeader
));
419 header2
= qemu_blockalign(bs
, sizeof(VHDXHeader
));
421 buffer
= qemu_blockalign(bs
, VHDX_HEADER_SIZE
);
423 s
->headers
[0] = header1
;
424 s
->headers
[1] = header2
;
426 /* We have to read the whole VHDX_HEADER_SIZE instead of
427 * sizeof(VHDXHeader), because the checksum is over the whole
429 ret
= bdrv_pread(bs
->file
, VHDX_HEADER1_OFFSET
, buffer
, VHDX_HEADER_SIZE
);
433 /* copy over just the relevant portion that we need */
434 memcpy(header1
, buffer
, sizeof(VHDXHeader
));
435 vhdx_header_le_import(header1
);
437 if (vhdx_checksum_is_valid(buffer
, VHDX_HEADER_SIZE
, 4) &&
438 !memcmp(&header1
->signature
, "head", 4) &&
439 header1
->version
== 1) {
440 h1_seq
= header1
->sequence_number
;
444 ret
= bdrv_pread(bs
->file
, VHDX_HEADER2_OFFSET
, buffer
, VHDX_HEADER_SIZE
);
448 /* copy over just the relevant portion that we need */
449 memcpy(header2
, buffer
, sizeof(VHDXHeader
));
450 vhdx_header_le_import(header2
);
452 if (vhdx_checksum_is_valid(buffer
, VHDX_HEADER_SIZE
, 4) &&
453 !memcmp(&header2
->signature
, "head", 4) &&
454 header2
->version
== 1) {
455 h2_seq
= header2
->sequence_number
;
459 /* If there is only 1 valid header (or no valid headers), we
460 * don't care what the sequence numbers are */
461 if (h1_valid
&& !h2_valid
) {
463 } else if (!h1_valid
&& h2_valid
) {
465 } else if (!h1_valid
&& !h2_valid
) {
468 /* If both headers are valid, then we choose the active one by the
469 * highest sequence number. If the sequence numbers are equal, that is
471 if (h1_seq
> h2_seq
) {
473 } else if (h2_seq
> h1_seq
) {
480 vhdx_region_register(s
, s
->headers
[s
->curr_header
]->log_offset
,
481 s
->headers
[s
->curr_header
]->log_length
);
485 error_setg_errno(errp
, -ret
, "No valid VHDX header found");
488 s
->headers
[0] = NULL
;
489 s
->headers
[1] = NULL
;
495 static int vhdx_open_region_tables(BlockDriverState
*bs
, BDRVVHDXState
*s
)
500 VHDXRegionTableEntry rt_entry
;
502 bool bat_rt_found
= false;
503 bool metadata_rt_found
= false;
505 /* We have to read the whole 64KB block, because the crc32 is over the
507 buffer
= qemu_blockalign(bs
, VHDX_HEADER_BLOCK_SIZE
);
509 ret
= bdrv_pread(bs
->file
, VHDX_REGION_TABLE_OFFSET
, buffer
,
510 VHDX_HEADER_BLOCK_SIZE
);
514 memcpy(&s
->rt
, buffer
, sizeof(s
->rt
));
515 vhdx_region_header_le_import(&s
->rt
);
516 offset
+= sizeof(s
->rt
);
518 if (!vhdx_checksum_is_valid(buffer
, VHDX_HEADER_BLOCK_SIZE
, 4) ||
519 memcmp(&s
->rt
.signature
, "regi", 4)) {
524 /* Per spec, maximum region table entry count is 2047 */
525 if (s
->rt
.entry_count
> 2047) {
530 for (i
= 0; i
< s
->rt
.entry_count
; i
++) {
531 memcpy(&rt_entry
, buffer
+ offset
, sizeof(rt_entry
));
532 offset
+= sizeof(rt_entry
);
534 vhdx_region_entry_le_import(&rt_entry
);
536 /* check for region overlap between these entries, and any
537 * other memory regions in the file */
538 ret
= vhdx_region_check(s
, rt_entry
.file_offset
, rt_entry
.length
);
543 vhdx_region_register(s
, rt_entry
.file_offset
, rt_entry
.length
);
545 /* see if we recognize the entry */
546 if (guid_eq(rt_entry
.guid
, bat_guid
)) {
547 /* must be unique; if we have already found it this is invalid */
553 s
->bat_rt
= rt_entry
;
557 if (guid_eq(rt_entry
.guid
, metadata_guid
)) {
558 /* must be unique; if we have already found it this is invalid */
559 if (metadata_rt_found
) {
563 metadata_rt_found
= true;
564 s
->metadata_rt
= rt_entry
;
568 if (rt_entry
.data_bits
& VHDX_REGION_ENTRY_REQUIRED
) {
569 /* cannot read vhdx file - required region table entry that
570 * we do not understand. per spec, we must fail to open */
576 if (!bat_rt_found
|| !metadata_rt_found
) {
590 /* Metadata initial parser
592 * This loads all the metadata entry fields. This may cause additional
593 * fields to be processed (e.g. parent locator, etc..).
595 * There are 5 Metadata items that are always required:
596 * - File Parameters (block size, has a parent)
597 * - Virtual Disk Size (size, in bytes, of the virtual drive)
598 * - Page 83 Data (scsi page 83 guid)
599 * - Logical Sector Size (logical sector size in bytes, either 512 or
600 * 4096. We only support 512 currently)
601 * - Physical Sector Size (512 or 4096)
603 * Also, if the File Parameters indicate this is a differencing file,
604 * we must also look for the Parent Locator metadata item.
606 static int vhdx_parse_metadata(BlockDriverState
*bs
, BDRVVHDXState
*s
)
612 VHDXMetadataTableEntry md_entry
;
614 buffer
= qemu_blockalign(bs
, VHDX_METADATA_TABLE_MAX_SIZE
);
616 ret
= bdrv_pread(bs
->file
, s
->metadata_rt
.file_offset
, buffer
,
617 VHDX_METADATA_TABLE_MAX_SIZE
);
621 memcpy(&s
->metadata_hdr
, buffer
, sizeof(s
->metadata_hdr
));
622 offset
+= sizeof(s
->metadata_hdr
);
624 vhdx_metadata_header_le_import(&s
->metadata_hdr
);
626 if (memcmp(&s
->metadata_hdr
.signature
, "metadata", 8)) {
631 s
->metadata_entries
.present
= 0;
633 if ((s
->metadata_hdr
.entry_count
* sizeof(md_entry
)) >
634 (VHDX_METADATA_TABLE_MAX_SIZE
- offset
)) {
639 for (i
= 0; i
< s
->metadata_hdr
.entry_count
; i
++) {
640 memcpy(&md_entry
, buffer
+ offset
, sizeof(md_entry
));
641 offset
+= sizeof(md_entry
);
643 vhdx_metadata_entry_le_import(&md_entry
);
645 if (guid_eq(md_entry
.item_id
, file_param_guid
)) {
646 if (s
->metadata_entries
.present
& META_FILE_PARAMETER_PRESENT
) {
650 s
->metadata_entries
.file_parameters_entry
= md_entry
;
651 s
->metadata_entries
.present
|= META_FILE_PARAMETER_PRESENT
;
655 if (guid_eq(md_entry
.item_id
, virtual_size_guid
)) {
656 if (s
->metadata_entries
.present
& META_VIRTUAL_DISK_SIZE_PRESENT
) {
660 s
->metadata_entries
.virtual_disk_size_entry
= md_entry
;
661 s
->metadata_entries
.present
|= META_VIRTUAL_DISK_SIZE_PRESENT
;
665 if (guid_eq(md_entry
.item_id
, page83_guid
)) {
666 if (s
->metadata_entries
.present
& META_PAGE_83_PRESENT
) {
670 s
->metadata_entries
.page83_data_entry
= md_entry
;
671 s
->metadata_entries
.present
|= META_PAGE_83_PRESENT
;
675 if (guid_eq(md_entry
.item_id
, logical_sector_guid
)) {
676 if (s
->metadata_entries
.present
&
677 META_LOGICAL_SECTOR_SIZE_PRESENT
) {
681 s
->metadata_entries
.logical_sector_size_entry
= md_entry
;
682 s
->metadata_entries
.present
|= META_LOGICAL_SECTOR_SIZE_PRESENT
;
686 if (guid_eq(md_entry
.item_id
, phys_sector_guid
)) {
687 if (s
->metadata_entries
.present
& META_PHYS_SECTOR_SIZE_PRESENT
) {
691 s
->metadata_entries
.phys_sector_size_entry
= md_entry
;
692 s
->metadata_entries
.present
|= META_PHYS_SECTOR_SIZE_PRESENT
;
696 if (guid_eq(md_entry
.item_id
, parent_locator_guid
)) {
697 if (s
->metadata_entries
.present
& META_PARENT_LOCATOR_PRESENT
) {
701 s
->metadata_entries
.parent_locator_entry
= md_entry
;
702 s
->metadata_entries
.present
|= META_PARENT_LOCATOR_PRESENT
;
706 if (md_entry
.data_bits
& VHDX_META_FLAGS_IS_REQUIRED
) {
707 /* cannot read vhdx file - required region table entry that
708 * we do not understand. per spec, we must fail to open */
714 if (s
->metadata_entries
.present
!= META_ALL_PRESENT
) {
719 ret
= bdrv_pread(bs
->file
,
720 s
->metadata_entries
.file_parameters_entry
.offset
721 + s
->metadata_rt
.file_offset
,
729 le32_to_cpus(&s
->params
.block_size
);
730 le32_to_cpus(&s
->params
.data_bits
);
733 /* We now have the file parameters, so we can tell if this is a
734 * differencing file (i.e.. has_parent), is dynamic or fixed
735 * sized (leave_blocks_allocated), and the block size */
737 /* The parent locator required iff the file parameters has_parent set */
738 if (s
->params
.data_bits
& VHDX_PARAMS_HAS_PARENT
) {
739 if (s
->metadata_entries
.present
& META_PARENT_LOCATOR_PRESENT
) {
740 /* TODO: parse parent locator fields */
741 ret
= -ENOTSUP
; /* temp, until differencing files are supported */
744 /* if has_parent is set, but there is not parent locator present,
745 * then that is an invalid combination */
751 /* determine virtual disk size, logical sector size,
752 * and phys sector size */
754 ret
= bdrv_pread(bs
->file
,
755 s
->metadata_entries
.virtual_disk_size_entry
.offset
756 + s
->metadata_rt
.file_offset
,
757 &s
->virtual_disk_size
,
762 ret
= bdrv_pread(bs
->file
,
763 s
->metadata_entries
.logical_sector_size_entry
.offset
764 + s
->metadata_rt
.file_offset
,
765 &s
->logical_sector_size
,
770 ret
= bdrv_pread(bs
->file
,
771 s
->metadata_entries
.phys_sector_size_entry
.offset
772 + s
->metadata_rt
.file_offset
,
773 &s
->physical_sector_size
,
779 le64_to_cpus(&s
->virtual_disk_size
);
780 le32_to_cpus(&s
->logical_sector_size
);
781 le32_to_cpus(&s
->physical_sector_size
);
783 if (s
->logical_sector_size
== 0 || s
->params
.block_size
== 0) {
788 /* both block_size and sector_size are guaranteed powers of 2 */
789 s
->sectors_per_block
= s
->params
.block_size
/ s
->logical_sector_size
;
790 s
->chunk_ratio
= (VHDX_MAX_SECTORS_PER_BLOCK
) *
791 (uint64_t)s
->logical_sector_size
/
792 (uint64_t)s
->params
.block_size
;
794 /* These values are ones we will want to use for division / multiplication
795 * later on, and they are all guaranteed (per the spec) to be powers of 2,
796 * so we can take advantage of that for shift operations during
798 if (s
->logical_sector_size
& (s
->logical_sector_size
- 1)) {
802 if (s
->sectors_per_block
& (s
->sectors_per_block
- 1)) {
806 if (s
->chunk_ratio
& (s
->chunk_ratio
- 1)) {
810 s
->block_size
= s
->params
.block_size
;
811 if (s
->block_size
& (s
->block_size
- 1)) {
816 vhdx_set_shift_bits(s
);
826 * Calculate the number of BAT entries, including sector
829 static void vhdx_calc_bat_entries(BDRVVHDXState
*s
)
831 uint32_t data_blocks_cnt
, bitmap_blocks_cnt
;
833 data_blocks_cnt
= s
->virtual_disk_size
>> s
->block_size_bits
;
834 if (s
->virtual_disk_size
- (data_blocks_cnt
<< s
->block_size_bits
)) {
837 bitmap_blocks_cnt
= data_blocks_cnt
>> s
->chunk_ratio_bits
;
838 if (data_blocks_cnt
- (bitmap_blocks_cnt
<< s
->chunk_ratio_bits
)) {
842 if (s
->parent_entries
) {
843 s
->bat_entries
= bitmap_blocks_cnt
* (s
->chunk_ratio
+ 1);
845 s
->bat_entries
= data_blocks_cnt
+
846 ((data_blocks_cnt
- 1) >> s
->chunk_ratio_bits
);
851 static void vhdx_close(BlockDriverState
*bs
)
853 BDRVVHDXState
*s
= bs
->opaque
;
854 qemu_vfree(s
->headers
[0]);
855 s
->headers
[0] = NULL
;
856 qemu_vfree(s
->headers
[1]);
857 s
->headers
[1] = NULL
;
860 qemu_vfree(s
->parent_entries
);
861 s
->parent_entries
= NULL
;
862 migrate_del_blocker(s
->migration_blocker
);
863 error_free(s
->migration_blocker
);
864 qemu_vfree(s
->log
.hdr
);
866 vhdx_region_unregister_all(s
);
869 static int vhdx_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
872 BDRVVHDXState
*s
= bs
->opaque
;
876 Error
*local_err
= NULL
;
879 s
->first_visible_write
= true;
881 qemu_co_mutex_init(&s
->lock
);
882 QLIST_INIT(&s
->regions
);
884 /* validate the file signature */
885 ret
= bdrv_pread(bs
->file
, 0, &signature
, sizeof(uint64_t));
889 if (memcmp(&signature
, "vhdxfile", 8)) {
894 /* This is used for any header updates, for the file_write_guid.
895 * The spec dictates that a new value should be used for the first
897 vhdx_guid_generate(&s
->session_guid
);
899 vhdx_parse_header(bs
, s
, &local_err
);
900 if (local_err
!= NULL
) {
901 error_propagate(errp
, local_err
);
906 ret
= vhdx_parse_log(bs
, s
, &s
->log_replayed_on_open
, errp
);
911 ret
= vhdx_open_region_tables(bs
, s
);
916 ret
= vhdx_parse_metadata(bs
, s
);
921 s
->block_size
= s
->params
.block_size
;
923 /* the VHDX spec dictates that virtual_disk_size is always a multiple of
924 * logical_sector_size */
925 bs
->total_sectors
= s
->virtual_disk_size
>> s
->logical_sector_size_bits
;
927 vhdx_calc_bat_entries(s
);
929 s
->bat_offset
= s
->bat_rt
.file_offset
;
931 if (s
->bat_entries
> s
->bat_rt
.length
/ sizeof(VHDXBatEntry
)) {
932 /* BAT allocation is not large enough for all entries */
937 /* s->bat is freed in vhdx_close() */
938 s
->bat
= qemu_blockalign(bs
, s
->bat_rt
.length
);
940 ret
= bdrv_pread(bs
->file
, s
->bat_offset
, s
->bat
, s
->bat_rt
.length
);
945 uint64_t payblocks
= s
->chunk_ratio
;
946 /* endian convert, and verify populated BAT field file offsets against
947 * region table and log entries */
948 for (i
= 0; i
< s
->bat_entries
; i
++) {
949 le64_to_cpus(&s
->bat
[i
]);
951 /* payload bat entries */
952 if ((s
->bat
[i
] & VHDX_BAT_STATE_BIT_MASK
) ==
953 PAYLOAD_BLOCK_FULLY_PRESENT
) {
954 ret
= vhdx_region_check(s
, s
->bat
[i
] & VHDX_BAT_FILE_OFF_MASK
,
961 payblocks
= s
->chunk_ratio
;
962 /* Once differencing files are supported, verify sector bitmap
967 if (flags
& BDRV_O_RDWR
) {
968 ret
= vhdx_update_headers(bs
, s
, false, NULL
);
974 /* TODO: differencing files */
976 /* Disable migration when VHDX images are used */
977 error_set(&s
->migration_blocker
,
978 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED
,
979 "vhdx", bs
->device_name
, "live migration");
980 migrate_add_blocker(s
->migration_blocker
);
988 static int vhdx_reopen_prepare(BDRVReopenState
*state
,
989 BlockReopenQueue
*queue
, Error
**errp
)
996 * Perform sector to block offset translations, to get various
997 * sector and file offsets into the image. See VHDXSectorInfo
999 static void vhdx_block_translate(BDRVVHDXState
*s
, int64_t sector_num
,
1000 int nb_sectors
, VHDXSectorInfo
*sinfo
)
1002 uint32_t block_offset
;
1004 sinfo
->bat_idx
= sector_num
>> s
->sectors_per_block_bits
;
1005 /* effectively a modulo - this gives us the offset into the block
1006 * (in sector sizes) for our sector number */
1007 block_offset
= sector_num
- (sinfo
->bat_idx
<< s
->sectors_per_block_bits
);
1008 /* the chunk ratio gives us the interleaving of the sector
1009 * bitmaps, so we need to advance our page block index by the
1010 * sector bitmaps entry number */
1011 sinfo
->bat_idx
+= sinfo
->bat_idx
>> s
->chunk_ratio_bits
;
1013 /* the number of sectors we can read/write in this cycle */
1014 sinfo
->sectors_avail
= s
->sectors_per_block
- block_offset
;
1016 sinfo
->bytes_left
= sinfo
->sectors_avail
<< s
->logical_sector_size_bits
;
1018 if (sinfo
->sectors_avail
> nb_sectors
) {
1019 sinfo
->sectors_avail
= nb_sectors
;
1022 sinfo
->bytes_avail
= sinfo
->sectors_avail
<< s
->logical_sector_size_bits
;
1024 sinfo
->file_offset
= s
->bat
[sinfo
->bat_idx
] & VHDX_BAT_FILE_OFF_MASK
;
1026 sinfo
->block_offset
= block_offset
<< s
->logical_sector_size_bits
;
1028 /* The file offset must be past the header section, so must be > 0 */
1029 if (sinfo
->file_offset
== 0) {
1033 /* block offset is the offset in vhdx logical sectors, in
1034 * the payload data block. Convert that to a byte offset
1035 * in the block, and add in the payload data block offset
1036 * in the file, in bytes, to get the final read address */
1038 sinfo
->file_offset
+= sinfo
->block_offset
;
1042 static int vhdx_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1044 BDRVVHDXState
*s
= bs
->opaque
;
1046 bdi
->cluster_size
= s
->block_size
;
1048 bdi
->unallocated_blocks_are_zero
=
1049 (s
->params
.data_bits
& VHDX_PARAMS_HAS_PARENT
) == 0;
1055 static coroutine_fn
int vhdx_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
1056 int nb_sectors
, QEMUIOVector
*qiov
)
1058 BDRVVHDXState
*s
= bs
->opaque
;
1060 VHDXSectorInfo sinfo
;
1061 uint64_t bytes_done
= 0;
1062 QEMUIOVector hd_qiov
;
1064 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1066 qemu_co_mutex_lock(&s
->lock
);
1068 while (nb_sectors
> 0) {
1069 /* We are a differencing file, so we need to inspect the sector bitmap
1070 * to see if we have the data or not */
1071 if (s
->params
.data_bits
& VHDX_PARAMS_HAS_PARENT
) {
1072 /* not supported yet */
1076 vhdx_block_translate(s
, sector_num
, nb_sectors
, &sinfo
);
1078 qemu_iovec_reset(&hd_qiov
);
1079 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, sinfo
.bytes_avail
);
1081 /* check the payload block state */
1082 switch (s
->bat
[sinfo
.bat_idx
] & VHDX_BAT_STATE_BIT_MASK
) {
1083 case PAYLOAD_BLOCK_NOT_PRESENT
: /* fall through */
1084 case PAYLOAD_BLOCK_UNDEFINED
: /* fall through */
1085 case PAYLOAD_BLOCK_UNMAPPED
: /* fall through */
1086 case PAYLOAD_BLOCK_ZERO
:
1088 qemu_iovec_memset(&hd_qiov
, 0, 0, sinfo
.bytes_avail
);
1090 case PAYLOAD_BLOCK_FULLY_PRESENT
:
1091 qemu_co_mutex_unlock(&s
->lock
);
1092 ret
= bdrv_co_readv(bs
->file
,
1093 sinfo
.file_offset
>> BDRV_SECTOR_BITS
,
1094 sinfo
.sectors_avail
, &hd_qiov
);
1095 qemu_co_mutex_lock(&s
->lock
);
1100 case PAYLOAD_BLOCK_PARTIALLY_PRESENT
:
1101 /* we don't yet support difference files, fall through
1108 nb_sectors
-= sinfo
.sectors_avail
;
1109 sector_num
+= sinfo
.sectors_avail
;
1110 bytes_done
+= sinfo
.bytes_avail
;
1115 qemu_co_mutex_unlock(&s
->lock
);
1116 qemu_iovec_destroy(&hd_qiov
);
1121 * Allocate a new payload block at the end of the file.
1123 * Allocation will happen at 1MB alignment inside the file
1125 * Returns the file offset start of the new payload block
1127 static int vhdx_allocate_block(BlockDriverState
*bs
, BDRVVHDXState
*s
,
1128 uint64_t *new_offset
)
1130 *new_offset
= bdrv_getlength(bs
->file
);
1132 /* per the spec, the address for a block is in units of 1MB */
1133 *new_offset
= ROUND_UP(*new_offset
, 1024 * 1024);
1135 return bdrv_truncate(bs
->file
, *new_offset
+ s
->block_size
);
1139 * Update the BAT table entry with the new file offset, and the new entry
1141 static void vhdx_update_bat_table_entry(BlockDriverState
*bs
, BDRVVHDXState
*s
,
1142 VHDXSectorInfo
*sinfo
,
1143 uint64_t *bat_entry_le
,
1144 uint64_t *bat_offset
, int state
)
1146 /* The BAT entry is a uint64, with 44 bits for the file offset in units of
1147 * 1MB, and 3 bits for the block state. */
1148 s
->bat
[sinfo
->bat_idx
] = sinfo
->file_offset
;
1150 s
->bat
[sinfo
->bat_idx
] |= state
& VHDX_BAT_STATE_BIT_MASK
;
1152 *bat_entry_le
= cpu_to_le64(s
->bat
[sinfo
->bat_idx
]);
1153 *bat_offset
= s
->bat_offset
+ sinfo
->bat_idx
* sizeof(VHDXBatEntry
);
1157 /* Per the spec, on the first write of guest-visible data to the file the
1158 * data write guid must be updated in the header */
1159 int vhdx_user_visible_write(BlockDriverState
*bs
, BDRVVHDXState
*s
)
1162 if (s
->first_visible_write
) {
1163 s
->first_visible_write
= false;
1164 ret
= vhdx_update_headers(bs
, s
, true, NULL
);
1169 static coroutine_fn
int vhdx_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
1170 int nb_sectors
, QEMUIOVector
*qiov
)
1173 BDRVVHDXState
*s
= bs
->opaque
;
1174 VHDXSectorInfo sinfo
;
1175 uint64_t bytes_done
= 0;
1176 uint64_t bat_entry
= 0;
1177 uint64_t bat_entry_offset
= 0;
1178 QEMUIOVector hd_qiov
;
1179 struct iovec iov1
= { 0 };
1180 struct iovec iov2
= { 0 };
1181 int sectors_to_write
;
1183 uint64_t bat_prior_offset
= 0;
1184 bool bat_update
= false;
1186 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1188 qemu_co_mutex_lock(&s
->lock
);
1190 ret
= vhdx_user_visible_write(bs
, s
);
1195 while (nb_sectors
> 0) {
1196 bool use_zero_buffers
= false;
1198 if (s
->params
.data_bits
& VHDX_PARAMS_HAS_PARENT
) {
1199 /* not supported yet */
1203 vhdx_block_translate(s
, sector_num
, nb_sectors
, &sinfo
);
1204 sectors_to_write
= sinfo
.sectors_avail
;
1206 qemu_iovec_reset(&hd_qiov
);
1207 /* check the payload block state */
1208 bat_state
= s
->bat
[sinfo
.bat_idx
] & VHDX_BAT_STATE_BIT_MASK
;
1209 switch (bat_state
) {
1210 case PAYLOAD_BLOCK_ZERO
:
1211 /* in this case, we need to preserve zero writes for
1212 * data that is not part of this write, so we must pad
1213 * the rest of the buffer to zeroes */
1215 /* if we are on a posix system with ftruncate() that extends
1216 * a file, then it is zero-filled for us. On Win32, the raw
1217 * layer uses SetFilePointer and SetFileEnd, which does not
1218 * zero fill AFAIK */
1220 /* Queue another write of zero buffers if the underlying file
1221 * does not zero-fill on file extension */
1223 if (bdrv_has_zero_init(bs
->file
) == 0) {
1224 use_zero_buffers
= true;
1226 /* zero fill the front, if any */
1227 if (sinfo
.block_offset
) {
1228 iov1
.iov_len
= sinfo
.block_offset
;
1229 iov1
.iov_base
= qemu_blockalign(bs
, iov1
.iov_len
);
1230 memset(iov1
.iov_base
, 0, iov1
.iov_len
);
1231 qemu_iovec_concat_iov(&hd_qiov
, &iov1
, 1, 0,
1232 sinfo
.block_offset
);
1233 sectors_to_write
+= iov1
.iov_len
>> BDRV_SECTOR_BITS
;
1236 /* our actual data */
1237 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
,
1240 /* zero fill the back, if any */
1241 if ((sinfo
.bytes_avail
- sinfo
.block_offset
) <
1243 iov2
.iov_len
= s
->block_size
-
1244 (sinfo
.bytes_avail
+ sinfo
.block_offset
);
1245 iov2
.iov_base
= qemu_blockalign(bs
, iov2
.iov_len
);
1246 memset(iov2
.iov_base
, 0, iov2
.iov_len
);
1247 qemu_iovec_concat_iov(&hd_qiov
, &iov2
, 1, 0,
1248 sinfo
.block_offset
);
1249 sectors_to_write
+= iov2
.iov_len
>> BDRV_SECTOR_BITS
;
1254 case PAYLOAD_BLOCK_NOT_PRESENT
: /* fall through */
1255 case PAYLOAD_BLOCK_UNMAPPED
: /* fall through */
1256 case PAYLOAD_BLOCK_UNDEFINED
: /* fall through */
1257 bat_prior_offset
= sinfo
.file_offset
;
1258 ret
= vhdx_allocate_block(bs
, s
, &sinfo
.file_offset
);
1262 /* once we support differencing files, this may also be
1263 * partially present */
1264 /* update block state to the newly specified state */
1265 vhdx_update_bat_table_entry(bs
, s
, &sinfo
, &bat_entry
,
1267 PAYLOAD_BLOCK_FULLY_PRESENT
);
1269 /* since we just allocated a block, file_offset is the
1270 * beginning of the payload block. It needs to be the
1271 * write address, which includes the offset into the block */
1272 if (!use_zero_buffers
) {
1273 sinfo
.file_offset
+= sinfo
.block_offset
;
1276 case PAYLOAD_BLOCK_FULLY_PRESENT
:
1277 /* if the file offset address is in the header zone,
1278 * there is a problem */
1279 if (sinfo
.file_offset
< (1024 * 1024)) {
1281 goto error_bat_restore
;
1284 if (!use_zero_buffers
) {
1285 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
,
1288 /* block exists, so we can just overwrite it */
1289 qemu_co_mutex_unlock(&s
->lock
);
1290 ret
= bdrv_co_writev(bs
->file
,
1291 sinfo
.file_offset
>> BDRV_SECTOR_BITS
,
1292 sectors_to_write
, &hd_qiov
);
1293 qemu_co_mutex_lock(&s
->lock
);
1295 goto error_bat_restore
;
1298 case PAYLOAD_BLOCK_PARTIALLY_PRESENT
:
1299 /* we don't yet support difference files, fall through
1308 /* this will update the BAT entry into the log journal, and
1309 * then flush the log journal out to disk */
1310 ret
= vhdx_log_write_and_flush(bs
, s
, &bat_entry
,
1311 sizeof(VHDXBatEntry
),
1318 nb_sectors
-= sinfo
.sectors_avail
;
1319 sector_num
+= sinfo
.sectors_avail
;
1320 bytes_done
+= sinfo
.bytes_avail
;
1329 /* keep metadata in sync, and restore the bat entry state
1331 sinfo
.file_offset
= bat_prior_offset
;
1332 vhdx_update_bat_table_entry(bs
, s
, &sinfo
, &bat_entry
,
1333 &bat_entry_offset
, bat_state
);
1336 qemu_vfree(iov1
.iov_base
);
1337 qemu_vfree(iov2
.iov_base
);
1338 qemu_co_mutex_unlock(&s
->lock
);
1339 qemu_iovec_destroy(&hd_qiov
);
1346 * Create VHDX Headers
1348 * There are 2 headers, and the highest sequence number will represent
1351 static int vhdx_create_new_headers(BlockDriverState
*bs
, uint64_t image_size
,
1355 VHDXHeader
*hdr
= NULL
;
1357 hdr
= g_malloc0(sizeof(VHDXHeader
));
1359 hdr
->signature
= VHDX_HEADER_SIGNATURE
;
1360 hdr
->sequence_number
= g_random_int();
1361 hdr
->log_version
= 0;
1363 hdr
->log_length
= log_size
;
1364 hdr
->log_offset
= VHDX_HEADER_SECTION_END
;
1365 vhdx_guid_generate(&hdr
->file_write_guid
);
1366 vhdx_guid_generate(&hdr
->data_write_guid
);
1368 ret
= vhdx_write_header(bs
, hdr
, VHDX_HEADER1_OFFSET
, false);
1372 hdr
->sequence_number
++;
1373 ret
= vhdx_write_header(bs
, hdr
, VHDX_HEADER2_OFFSET
, false);
1385 * Create the Metadata entries.
1387 * For more details on the entries, see section 3.5 (pg 29) in the
1388 * VHDX 1.00 specification.
1390 * We support 5 metadata entries (all required by spec):
1392 * Virtual Disk Size,
1394 * Logical Sector Size,
1395 * Physical Sector Size
1397 * The first 64KB of the Metadata section is reserved for the metadata
1398 * header and entries; beyond that, the metadata items themselves reside.
1400 static int vhdx_create_new_metadata(BlockDriverState
*bs
,
1401 uint64_t image_size
,
1402 uint32_t block_size
,
1403 uint32_t sector_size
,
1404 uint64_t metadata_offset
,
1408 uint32_t offset
= 0;
1409 void *buffer
= NULL
;
1411 VHDXMetadataTableHeader
*md_table
;;
1412 VHDXMetadataTableEntry
*md_table_entry
;
1414 /* Metadata entries */
1415 VHDXFileParameters
*mt_file_params
;
1416 VHDXVirtualDiskSize
*mt_virtual_size
;
1417 VHDXPage83Data
*mt_page83
;
1418 VHDXVirtualDiskLogicalSectorSize
*mt_log_sector_size
;
1419 VHDXVirtualDiskPhysicalSectorSize
*mt_phys_sector_size
;
1421 entry_buffer
= g_malloc0(sizeof(VHDXFileParameters
) +
1422 sizeof(VHDXVirtualDiskSize
) +
1423 sizeof(VHDXPage83Data
) +
1424 sizeof(VHDXVirtualDiskLogicalSectorSize
) +
1425 sizeof(VHDXVirtualDiskPhysicalSectorSize
));
1427 mt_file_params
= entry_buffer
;
1428 offset
+= sizeof(VHDXFileParameters
);
1429 mt_virtual_size
= entry_buffer
+ offset
;
1430 offset
+= sizeof(VHDXVirtualDiskSize
);
1431 mt_page83
= entry_buffer
+ offset
;
1432 offset
+= sizeof(VHDXPage83Data
);
1433 mt_log_sector_size
= entry_buffer
+ offset
;
1434 offset
+= sizeof(VHDXVirtualDiskLogicalSectorSize
);
1435 mt_phys_sector_size
= entry_buffer
+ offset
;
1437 mt_file_params
->block_size
= cpu_to_le32(block_size
);
1438 if (type
== VHDX_TYPE_FIXED
) {
1439 mt_file_params
->data_bits
|= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED
;
1440 cpu_to_le32s(&mt_file_params
->data_bits
);
1443 vhdx_guid_generate(&mt_page83
->page_83_data
);
1444 cpu_to_leguids(&mt_page83
->page_83_data
);
1445 mt_virtual_size
->virtual_disk_size
= cpu_to_le64(image_size
);
1446 mt_log_sector_size
->logical_sector_size
= cpu_to_le32(sector_size
);
1447 mt_phys_sector_size
->physical_sector_size
= cpu_to_le32(sector_size
);
1449 buffer
= g_malloc0(VHDX_HEADER_BLOCK_SIZE
);
1452 md_table
->signature
= VHDX_METADATA_SIGNATURE
;
1453 md_table
->entry_count
= 5;
1454 vhdx_metadata_header_le_export(md_table
);
1457 /* This will reference beyond the reserved table portion */
1460 md_table_entry
= buffer
+ sizeof(VHDXMetadataTableHeader
);
1462 md_table_entry
[0].item_id
= file_param_guid
;
1463 md_table_entry
[0].offset
= offset
;
1464 md_table_entry
[0].length
= sizeof(VHDXFileParameters
);
1465 md_table_entry
[0].data_bits
|= VHDX_META_FLAGS_IS_REQUIRED
;
1466 offset
+= md_table_entry
[0].length
;
1467 vhdx_metadata_entry_le_export(&md_table_entry
[0]);
1469 md_table_entry
[1].item_id
= virtual_size_guid
;
1470 md_table_entry
[1].offset
= offset
;
1471 md_table_entry
[1].length
= sizeof(VHDXVirtualDiskSize
);
1472 md_table_entry
[1].data_bits
|= VHDX_META_FLAGS_IS_REQUIRED
|
1473 VHDX_META_FLAGS_IS_VIRTUAL_DISK
;
1474 offset
+= md_table_entry
[1].length
;
1475 vhdx_metadata_entry_le_export(&md_table_entry
[1]);
1477 md_table_entry
[2].item_id
= page83_guid
;
1478 md_table_entry
[2].offset
= offset
;
1479 md_table_entry
[2].length
= sizeof(VHDXPage83Data
);
1480 md_table_entry
[2].data_bits
|= VHDX_META_FLAGS_IS_REQUIRED
|
1481 VHDX_META_FLAGS_IS_VIRTUAL_DISK
;
1482 offset
+= md_table_entry
[2].length
;
1483 vhdx_metadata_entry_le_export(&md_table_entry
[2]);
1485 md_table_entry
[3].item_id
= logical_sector_guid
;
1486 md_table_entry
[3].offset
= offset
;
1487 md_table_entry
[3].length
= sizeof(VHDXVirtualDiskLogicalSectorSize
);
1488 md_table_entry
[3].data_bits
|= VHDX_META_FLAGS_IS_REQUIRED
|
1489 VHDX_META_FLAGS_IS_VIRTUAL_DISK
;
1490 offset
+= md_table_entry
[3].length
;
1491 vhdx_metadata_entry_le_export(&md_table_entry
[3]);
1493 md_table_entry
[4].item_id
= phys_sector_guid
;
1494 md_table_entry
[4].offset
= offset
;
1495 md_table_entry
[4].length
= sizeof(VHDXVirtualDiskPhysicalSectorSize
);
1496 md_table_entry
[4].data_bits
|= VHDX_META_FLAGS_IS_REQUIRED
|
1497 VHDX_META_FLAGS_IS_VIRTUAL_DISK
;
1498 vhdx_metadata_entry_le_export(&md_table_entry
[4]);
1500 ret
= bdrv_pwrite(bs
, metadata_offset
, buffer
, VHDX_HEADER_BLOCK_SIZE
);
1505 ret
= bdrv_pwrite(bs
, metadata_offset
+ (64 * KiB
), entry_buffer
,
1506 VHDX_HEADER_BLOCK_SIZE
);
1514 g_free(entry_buffer
);
1518 /* This create the actual BAT itself. We currently only support
1519 * 'Dynamic' and 'Fixed' image types.
1521 * Dynamic images: default state of the BAT is all zeroes.
1523 * Fixed images: default state of the BAT is fully populated, with
1524 * file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT.
1526 static int vhdx_create_bat(BlockDriverState
*bs
, BDRVVHDXState
*s
,
1527 uint64_t image_size
, VHDXImageType type
,
1528 bool use_zero_blocks
, VHDXRegionTableEntry
*rt_bat
)
1531 uint64_t data_file_offset
;
1532 uint64_t total_sectors
= 0;
1533 uint64_t sector_num
= 0;
1536 VHDXSectorInfo sinfo
;
1538 assert(s
->bat
== NULL
);
1540 /* this gives a data start after BAT/bitmap entries, and well
1541 * past any metadata entries (with a 4 MB buffer for future
1543 data_file_offset
= rt_bat
->file_offset
+ rt_bat
->length
+ 5 * MiB
;
1544 total_sectors
= image_size
>> s
->logical_sector_size_bits
;
1546 if (type
== VHDX_TYPE_DYNAMIC
) {
1547 /* All zeroes, so we can just extend the file - the end of the BAT
1548 * is the furthest thing we have written yet */
1549 ret
= bdrv_truncate(bs
, data_file_offset
);
1553 } else if (type
== VHDX_TYPE_FIXED
) {
1554 ret
= bdrv_truncate(bs
, data_file_offset
+ image_size
);
1563 if (type
== VHDX_TYPE_FIXED
||
1565 bdrv_has_zero_init(bs
) == 0) {
1566 /* for a fixed file, the default BAT entry is not zero */
1567 s
->bat
= g_malloc0(rt_bat
->length
);
1568 block_state
= type
== VHDX_TYPE_FIXED
? PAYLOAD_BLOCK_FULLY_PRESENT
:
1569 PAYLOAD_BLOCK_NOT_PRESENT
;
1570 block_state
= use_zero_blocks
? PAYLOAD_BLOCK_ZERO
: block_state
;
1571 /* fill the BAT by emulating sector writes of sectors_per_block size */
1572 while (sector_num
< total_sectors
) {
1573 vhdx_block_translate(s
, sector_num
, s
->sectors_per_block
, &sinfo
);
1574 sinfo
.file_offset
= data_file_offset
+
1575 (sector_num
<< s
->logical_sector_size_bits
);
1576 sinfo
.file_offset
= ROUND_UP(sinfo
.file_offset
, MiB
);
1577 vhdx_update_bat_table_entry(bs
, s
, &sinfo
, &unused
, &unused
,
1579 cpu_to_le64s(&s
->bat
[sinfo
.bat_idx
]);
1580 sector_num
+= s
->sectors_per_block
;
1582 ret
= bdrv_pwrite(bs
, rt_bat
->file_offset
, s
->bat
, rt_bat
->length
);
1595 /* Creates the region table header, and region table entries.
1596 * There are 2 supported region table entries: BAT, and Metadata/
1598 * As the calculations for the BAT region table are also needed
1599 * to create the BAT itself, we will also cause the BAT to be
1602 static int vhdx_create_new_region_table(BlockDriverState
*bs
,
1603 uint64_t image_size
,
1604 uint32_t block_size
,
1605 uint32_t sector_size
,
1607 bool use_zero_blocks
,
1609 uint64_t *metadata_offset
)
1612 uint32_t offset
= 0;
1613 void *buffer
= NULL
;
1614 BDRVVHDXState
*s
= NULL
;
1615 VHDXRegionTableHeader
*region_table
;
1616 VHDXRegionTableEntry
*rt_bat
;
1617 VHDXRegionTableEntry
*rt_metadata
;
1619 assert(metadata_offset
!= NULL
);
1621 /* Populate enough of the BDRVVHDXState to be able to use the
1622 * pre-existing BAT calculation, translation, and update functions */
1623 s
= g_malloc0(sizeof(BDRVVHDXState
));
1625 s
->chunk_ratio
= (VHDX_MAX_SECTORS_PER_BLOCK
) *
1626 (uint64_t) sector_size
/ (uint64_t) block_size
;
1628 s
->sectors_per_block
= block_size
/ sector_size
;
1629 s
->virtual_disk_size
= image_size
;
1630 s
->block_size
= block_size
;
1631 s
->logical_sector_size
= sector_size
;
1633 vhdx_set_shift_bits(s
);
1635 vhdx_calc_bat_entries(s
);
1637 /* At this point the VHDX state is populated enough for creation */
1639 /* a single buffer is used so we can calculate the checksum over the
1640 * entire 64KB block */
1641 buffer
= g_malloc0(VHDX_HEADER_BLOCK_SIZE
);
1642 region_table
= buffer
;
1643 offset
+= sizeof(VHDXRegionTableHeader
);
1644 rt_bat
= buffer
+ offset
;
1645 offset
+= sizeof(VHDXRegionTableEntry
);
1646 rt_metadata
= buffer
+ offset
;
1648 region_table
->signature
= VHDX_REGION_SIGNATURE
;
1649 region_table
->entry_count
= 2; /* BAT and Metadata */
1651 rt_bat
->guid
= bat_guid
;
1652 rt_bat
->length
= ROUND_UP(s
->bat_entries
* sizeof(VHDXBatEntry
), MiB
);
1653 rt_bat
->file_offset
= ROUND_UP(VHDX_HEADER_SECTION_END
+ log_size
, MiB
);
1654 s
->bat_offset
= rt_bat
->file_offset
;
1656 rt_metadata
->guid
= metadata_guid
;
1657 rt_metadata
->file_offset
= ROUND_UP(rt_bat
->file_offset
+ rt_bat
->length
,
1659 rt_metadata
->length
= 1 * MiB
; /* min size, and more than enough */
1660 *metadata_offset
= rt_metadata
->file_offset
;
1662 vhdx_update_checksum(buffer
, VHDX_HEADER_BLOCK_SIZE
,
1663 offsetof(VHDXRegionTableHeader
, checksum
));
1666 /* The region table gives us the data we need to create the BAT,
1668 ret
= vhdx_create_bat(bs
, s
, image_size
, type
, use_zero_blocks
, rt_bat
);
1670 /* Now write out the region headers to disk */
1671 vhdx_region_header_le_export(region_table
);
1672 vhdx_region_entry_le_export(rt_bat
);
1673 vhdx_region_entry_le_export(rt_metadata
);
1675 ret
= bdrv_pwrite(bs
, VHDX_REGION_TABLE_OFFSET
, buffer
,
1676 VHDX_HEADER_BLOCK_SIZE
);
1681 ret
= bdrv_pwrite(bs
, VHDX_REGION_TABLE2_OFFSET
, buffer
,
1682 VHDX_HEADER_BLOCK_SIZE
);
1694 /* We need to create the following elements:
1696 * .-----------------------------------------------------------------.
1697 * | (A) | (B) | (C) | (D) | (E) |
1698 * | File ID | Header1 | Header 2 | Region Tbl 1 | Region Tbl 2 |
1700 * .-----------------------------------------------------------------.
1701 * 0 64KB 128KB 192KB 256KB 320KB
1704 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1705 * | (F) | (G) | (H) | |
1706 * | Journal Log | BAT / Bitmap | Metadata | .... data ...... |
1708 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1711 static int vhdx_create(const char *filename
, QEMUOptionParameter
*options
,
1715 uint64_t image_size
= (uint64_t) 2 * GiB
;
1716 uint32_t log_size
= 1 * MiB
;
1717 uint32_t block_size
= 0;
1719 uint64_t metadata_offset
;
1720 bool use_zero_blocks
= false;
1722 gunichar2
*creator
= NULL
;
1723 glong creator_items
;
1724 BlockDriverState
*bs
;
1725 const char *type
= NULL
;
1726 VHDXImageType image_type
;
1727 Error
*local_err
= NULL
;
1729 while (options
&& options
->name
) {
1730 if (!strcmp(options
->name
, BLOCK_OPT_SIZE
)) {
1731 image_size
= options
->value
.n
;
1732 } else if (!strcmp(options
->name
, VHDX_BLOCK_OPT_LOG_SIZE
)) {
1733 log_size
= options
->value
.n
;
1734 } else if (!strcmp(options
->name
, VHDX_BLOCK_OPT_BLOCK_SIZE
)) {
1735 block_size
= options
->value
.n
;
1736 } else if (!strcmp(options
->name
, BLOCK_OPT_SUBFMT
)) {
1737 type
= options
->value
.s
;
1738 } else if (!strcmp(options
->name
, VHDX_BLOCK_OPT_ZERO
)) {
1739 use_zero_blocks
= options
->value
.n
!= 0;
1744 if (image_size
> VHDX_MAX_IMAGE_SIZE
) {
1745 error_setg_errno(errp
, EINVAL
, "Image size too large; max of 64TB");
1754 if (!strcmp(type
, "dynamic")) {
1755 image_type
= VHDX_TYPE_DYNAMIC
;
1756 } else if (!strcmp(type
, "fixed")) {
1757 image_type
= VHDX_TYPE_FIXED
;
1758 } else if (!strcmp(type
, "differencing")) {
1759 error_setg_errno(errp
, ENOTSUP
,
1760 "Differencing files not yet supported");
1768 /* These are pretty arbitrary, and mainly designed to keep the BAT
1769 * size reasonable to load into RAM */
1770 if (block_size
== 0) {
1771 if (image_size
> 32 * TiB
) {
1772 block_size
= 64 * MiB
;
1773 } else if (image_size
> (uint64_t) 100 * GiB
) {
1774 block_size
= 32 * MiB
;
1775 } else if (image_size
> 1 * GiB
) {
1776 block_size
= 16 * MiB
;
1778 block_size
= 8 * MiB
;
1783 /* make the log size close to what was specified, but must be
1784 * min 1MB, and multiple of 1MB */
1785 log_size
= ROUND_UP(log_size
, MiB
);
1787 block_size
= ROUND_UP(block_size
, MiB
);
1788 block_size
= block_size
> VHDX_BLOCK_SIZE_MAX
? VHDX_BLOCK_SIZE_MAX
:
1791 ret
= bdrv_create_file(filename
, options
, &local_err
);
1793 error_propagate(errp
, local_err
);
1798 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
, BDRV_O_RDWR
| BDRV_O_PROTOCOL
,
1801 error_propagate(errp
, local_err
);
1807 /* The creator field is optional, but may be useful for
1808 * debugging / diagnostics */
1809 creator
= g_utf8_to_utf16("QEMU v" QEMU_VERSION
, -1, NULL
,
1810 &creator_items
, NULL
);
1811 signature
= cpu_to_le64(VHDX_FILE_SIGNATURE
);
1812 ret
= bdrv_pwrite(bs
, VHDX_FILE_ID_OFFSET
, &signature
, sizeof(signature
));
1814 goto delete_and_exit
;
1817 ret
= bdrv_pwrite(bs
, VHDX_FILE_ID_OFFSET
+ sizeof(signature
),
1818 creator
, creator_items
* sizeof(gunichar2
));
1820 goto delete_and_exit
;
1825 /* Creates (B),(C) */
1826 ret
= vhdx_create_new_headers(bs
, image_size
, log_size
);
1828 goto delete_and_exit
;
1831 /* Creates (D),(E),(G) explicitly. (F) created as by-product */
1832 ret
= vhdx_create_new_region_table(bs
, image_size
, block_size
, 512,
1833 log_size
, use_zero_blocks
, image_type
,
1836 goto delete_and_exit
;
1840 ret
= vhdx_create_new_metadata(bs
, image_size
, block_size
, 512,
1841 metadata_offset
, image_type
);
1843 goto delete_and_exit
;
1855 /* If opened r/w, the VHDX driver will automatically replay the log,
1856 * if one is present, inside the vhdx_open() call.
1858 * If qemu-img check -r all is called, the image is automatically opened
1859 * r/w and any log has already been replayed, so there is nothing (currently)
1862 static int vhdx_check(BlockDriverState
*bs
, BdrvCheckResult
*result
,
1865 BDRVVHDXState
*s
= bs
->opaque
;
1867 if (s
->log_replayed_on_open
) {
1868 result
->corruptions_fixed
++;
1873 static QEMUOptionParameter vhdx_create_options
[] = {
1875 .name
= BLOCK_OPT_SIZE
,
1877 .help
= "Virtual disk size; max of 64TB."
1880 .name
= VHDX_BLOCK_OPT_LOG_SIZE
,
1883 .help
= "Log size; min 1MB."
1886 .name
= VHDX_BLOCK_OPT_BLOCK_SIZE
,
1889 .help
= "Block Size; min 1MB, max 256MB. " \
1890 "0 means auto-calculate based on image size."
1893 .name
= BLOCK_OPT_SUBFMT
,
1895 .help
= "VHDX format type, can be either 'dynamic' or 'fixed'. "\
1896 "Default is 'dynamic'."
1899 .name
= VHDX_BLOCK_OPT_ZERO
,
1901 .help
= "Force use of payload blocks of type 'ZERO'. Non-standard."
1906 static BlockDriver bdrv_vhdx
= {
1907 .format_name
= "vhdx",
1908 .instance_size
= sizeof(BDRVVHDXState
),
1909 .bdrv_probe
= vhdx_probe
,
1910 .bdrv_open
= vhdx_open
,
1911 .bdrv_close
= vhdx_close
,
1912 .bdrv_reopen_prepare
= vhdx_reopen_prepare
,
1913 .bdrv_co_readv
= vhdx_co_readv
,
1914 .bdrv_co_writev
= vhdx_co_writev
,
1915 .bdrv_create
= vhdx_create
,
1916 .bdrv_get_info
= vhdx_get_info
,
1917 .bdrv_check
= vhdx_check
,
1919 .create_options
= vhdx_create_options
,
1922 static void bdrv_vhdx_init(void)
1924 bdrv_register(&bdrv_vhdx
);
1927 block_init(bdrv_vhdx_init
);