pc: use new CPU hotplug interface since 2.7 machine type
[qemu/ar7.git] / block / vhdx.c
blobf5605a2ff4ce688dd36cc6a1361b974610eb1e14
1 /*
2 * Block driver for Hyper-V VHDX Images
4 * Copyright (c) 2013 Red Hat, Inc.,
6 * Authors:
7 * Jeff Cody <jcody@redhat.com>
9 * This is based on the "VHDX Format Specification v1.00", published 8/25/2012
10 * by Microsoft:
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 "qemu-common.h"
21 #include "block/block_int.h"
22 #include "sysemu/block-backend.h"
23 #include "qemu/module.h"
24 #include "qemu/crc32c.h"
25 #include "qemu/bswap.h"
26 #include "block/vhdx.h"
27 #include "migration/migration.h"
29 #include <uuid/uuid.h>
31 /* Options for VHDX creation */
33 #define VHDX_BLOCK_OPT_LOG_SIZE "log_size"
34 #define VHDX_BLOCK_OPT_BLOCK_SIZE "block_size"
35 #define VHDX_BLOCK_OPT_ZERO "block_state_zero"
37 typedef enum VHDXImageType {
38 VHDX_TYPE_DYNAMIC = 0,
39 VHDX_TYPE_FIXED,
40 VHDX_TYPE_DIFFERENCING, /* Currently unsupported */
41 } VHDXImageType;
43 /* Several metadata and region table data entries are identified by
44 * guids in a MS-specific GUID format. */
47 /* ------- Known Region Table GUIDs ---------------------- */
48 static const MSGUID bat_guid = { .data1 = 0x2dc27766,
49 .data2 = 0xf623,
50 .data3 = 0x4200,
51 .data4 = { 0x9d, 0x64, 0x11, 0x5e,
52 0x9b, 0xfd, 0x4a, 0x08} };
54 static const MSGUID metadata_guid = { .data1 = 0x8b7ca206,
55 .data2 = 0x4790,
56 .data3 = 0x4b9a,
57 .data4 = { 0xb8, 0xfe, 0x57, 0x5f,
58 0x05, 0x0f, 0x88, 0x6e} };
62 /* ------- Known Metadata Entry GUIDs ---------------------- */
63 static const MSGUID file_param_guid = { .data1 = 0xcaa16737,
64 .data2 = 0xfa36,
65 .data3 = 0x4d43,
66 .data4 = { 0xb3, 0xb6, 0x33, 0xf0,
67 0xaa, 0x44, 0xe7, 0x6b} };
69 static const MSGUID virtual_size_guid = { .data1 = 0x2FA54224,
70 .data2 = 0xcd1b,
71 .data3 = 0x4876,
72 .data4 = { 0xb2, 0x11, 0x5d, 0xbe,
73 0xd8, 0x3b, 0xf4, 0xb8} };
75 static const MSGUID page83_guid = { .data1 = 0xbeca12ab,
76 .data2 = 0xb2e6,
77 .data3 = 0x4523,
78 .data4 = { 0x93, 0xef, 0xc3, 0x09,
79 0xe0, 0x00, 0xc7, 0x46} };
82 static const MSGUID phys_sector_guid = { .data1 = 0xcda348c7,
83 .data2 = 0x445d,
84 .data3 = 0x4471,
85 .data4 = { 0x9c, 0xc9, 0xe9, 0x88,
86 0x52, 0x51, 0xc5, 0x56} };
88 static const MSGUID parent_locator_guid = { .data1 = 0xa8d35f2d,
89 .data2 = 0xb30b,
90 .data3 = 0x454d,
91 .data4 = { 0xab, 0xf7, 0xd3,
92 0xd8, 0x48, 0x34,
93 0xab, 0x0c} };
95 static const MSGUID logical_sector_guid = { .data1 = 0x8141bf1d,
96 .data2 = 0xa96f,
97 .data3 = 0x4709,
98 .data4 = { 0xba, 0x47, 0xf2,
99 0x33, 0xa8, 0xfa,
100 0xab, 0x5f} };
102 /* Each parent type must have a valid GUID; this is for parent images
103 * of type 'VHDX'. If we were to allow e.g. a QCOW2 parent, we would
104 * need to make up our own QCOW2 GUID type */
105 static const MSGUID parent_vhdx_guid __attribute__((unused))
106 = { .data1 = 0xb04aefb7,
107 .data2 = 0xd19e,
108 .data3 = 0x4a81,
109 .data4 = { 0xb7, 0x89, 0x25, 0xb8,
110 0xe9, 0x44, 0x59, 0x13} };
113 #define META_FILE_PARAMETER_PRESENT 0x01
114 #define META_VIRTUAL_DISK_SIZE_PRESENT 0x02
115 #define META_PAGE_83_PRESENT 0x04
116 #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08
117 #define META_PHYS_SECTOR_SIZE_PRESENT 0x10
118 #define META_PARENT_LOCATOR_PRESENT 0x20
120 #define META_ALL_PRESENT \
121 (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \
122 META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \
123 META_PHYS_SECTOR_SIZE_PRESENT)
126 typedef struct VHDXSectorInfo {
127 uint32_t bat_idx; /* BAT entry index */
128 uint32_t sectors_avail; /* sectors available in payload block */
129 uint32_t bytes_left; /* bytes left in the block after data to r/w */
130 uint32_t bytes_avail; /* bytes available in payload block */
131 uint64_t file_offset; /* absolute offset in bytes, in file */
132 uint64_t block_offset; /* block offset, in bytes */
133 } VHDXSectorInfo;
135 /* Calculates new checksum.
137 * Zero is substituted during crc calculation for the original crc field
138 * crc_offset: byte offset in buf of the buffer crc
139 * buf: buffer pointer
140 * size: size of buffer (must be > crc_offset+4)
142 * Note: The buffer should have all multi-byte data in little-endian format,
143 * and the resulting checksum is in little endian format.
145 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset)
147 uint32_t crc;
149 assert(buf != NULL);
150 assert(size > (crc_offset + sizeof(crc)));
152 memset(buf + crc_offset, 0, sizeof(crc));
153 crc = crc32c(0xffffffff, buf, size);
154 cpu_to_le32s(&crc);
155 memcpy(buf + crc_offset, &crc, sizeof(crc));
157 return crc;
160 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size,
161 int crc_offset)
163 uint32_t crc_new;
164 uint32_t crc_orig;
165 assert(buf != NULL);
167 if (crc_offset > 0) {
168 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
169 memset(buf + crc_offset, 0, sizeof(crc_orig));
172 crc_new = crc32c(crc, buf, size);
173 if (crc_offset > 0) {
174 memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig));
177 return crc_new;
180 /* Validates the checksum of the buffer, with an in-place CRC.
182 * Zero is substituted during crc calculation for the original crc field,
183 * and the crc field is restored afterwards. But the buffer will be modifed
184 * during the calculation, so this may not be not suitable for multi-threaded
185 * use.
187 * crc_offset: byte offset in buf of the buffer crc
188 * buf: buffer pointer
189 * size: size of buffer (must be > crc_offset+4)
191 * returns true if checksum is valid, false otherwise
193 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset)
195 uint32_t crc_orig;
196 uint32_t crc;
198 assert(buf != NULL);
199 assert(size > (crc_offset + 4));
201 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
202 crc_orig = le32_to_cpu(crc_orig);
204 crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset);
206 return crc == crc_orig;
211 * This generates a UUID that is compliant with the MS GUIDs used
212 * in the VHDX spec (and elsewhere).
214 void vhdx_guid_generate(MSGUID *guid)
216 uuid_t uuid;
217 assert(guid != NULL);
219 uuid_generate(uuid);
220 memcpy(guid, uuid, sizeof(MSGUID));
223 /* Check for region overlaps inside the VHDX image */
224 static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length)
226 int ret = 0;
227 uint64_t end;
228 VHDXRegionEntry *r;
230 end = start + length;
231 QLIST_FOREACH(r, &s->regions, entries) {
232 if (!((start >= r->end) || (end <= r->start))) {
233 ret = -EINVAL;
234 goto exit;
238 exit:
239 return ret;
242 /* Register a region for future checks */
243 static void vhdx_region_register(BDRVVHDXState *s,
244 uint64_t start, uint64_t length)
246 VHDXRegionEntry *r;
248 r = g_malloc0(sizeof(*r));
250 r->start = start;
251 r->end = start + length;
253 QLIST_INSERT_HEAD(&s->regions, r, entries);
256 /* Free all registered regions */
257 static void vhdx_region_unregister_all(BDRVVHDXState *s)
259 VHDXRegionEntry *r, *r_next;
261 QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) {
262 QLIST_REMOVE(r, entries);
263 g_free(r);
267 static void vhdx_set_shift_bits(BDRVVHDXState *s)
269 s->logical_sector_size_bits = ctz32(s->logical_sector_size);
270 s->sectors_per_block_bits = ctz32(s->sectors_per_block);
271 s->chunk_ratio_bits = ctz64(s->chunk_ratio);
272 s->block_size_bits = ctz32(s->block_size);
276 * Per the MS VHDX Specification, for every VHDX file:
277 * - The header section is fixed size - 1 MB
278 * - The header section is always the first "object"
279 * - The first 64KB of the header is the File Identifier
280 * - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile")
281 * - The following 512 bytes constitute a UTF-16 string identifiying the
282 * software that created the file, and is optional and diagnostic only.
284 * Therefore, we probe by looking for the vhdxfile signature "vhdxfile"
286 static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename)
288 if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) {
289 return 100;
291 return 0;
295 * Writes the header to the specified offset.
297 * This will optionally read in buffer data from disk (otherwise zero-fill),
298 * and then update the header checksum. Header is converted to proper
299 * endianness before being written to the specified file offset
301 static int vhdx_write_header(BlockDriverState *bs_file, VHDXHeader *hdr,
302 uint64_t offset, bool read)
304 uint8_t *buffer = NULL;
305 int ret;
306 VHDXHeader *header_le;
308 assert(bs_file != NULL);
309 assert(hdr != NULL);
311 /* the header checksum is not over just the packed size of VHDXHeader,
312 * but rather over the entire 'reserved' range for the header, which is
313 * 4KB (VHDX_HEADER_SIZE). */
315 buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE);
316 if (read) {
317 /* if true, we can't assume the extra reserved bytes are 0 */
318 ret = bdrv_pread(bs_file, offset, buffer, VHDX_HEADER_SIZE);
319 if (ret < 0) {
320 goto exit;
322 } else {
323 memset(buffer, 0, VHDX_HEADER_SIZE);
326 /* overwrite the actual VHDXHeader portion */
327 header_le = (VHDXHeader *)buffer;
328 memcpy(header_le, hdr, sizeof(VHDXHeader));
329 vhdx_header_le_export(hdr, header_le);
330 vhdx_update_checksum(buffer, VHDX_HEADER_SIZE,
331 offsetof(VHDXHeader, checksum));
332 ret = bdrv_pwrite_sync(bs_file, offset, header_le, sizeof(VHDXHeader));
334 exit:
335 qemu_vfree(buffer);
336 return ret;
339 /* Update the VHDX headers
341 * This follows the VHDX spec procedures for header updates.
343 * - non-current header is updated with largest sequence number
345 static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s,
346 bool generate_data_write_guid, MSGUID *log_guid)
348 int ret = 0;
349 int hdr_idx = 0;
350 uint64_t header_offset = VHDX_HEADER1_OFFSET;
352 VHDXHeader *active_header;
353 VHDXHeader *inactive_header;
355 /* operate on the non-current header */
356 if (s->curr_header == 0) {
357 hdr_idx = 1;
358 header_offset = VHDX_HEADER2_OFFSET;
361 active_header = s->headers[s->curr_header];
362 inactive_header = s->headers[hdr_idx];
364 inactive_header->sequence_number = active_header->sequence_number + 1;
366 /* a new file guid must be generated before any file write, including
367 * headers */
368 inactive_header->file_write_guid = s->session_guid;
370 /* a new data guid only needs to be generated before any guest-visible
371 * writes (i.e. something observable via virtual disk read) */
372 if (generate_data_write_guid) {
373 vhdx_guid_generate(&inactive_header->data_write_guid);
376 /* update the log guid if present */
377 if (log_guid) {
378 inactive_header->log_guid = *log_guid;
381 ret = vhdx_write_header(bs->file->bs, inactive_header, header_offset, true);
382 if (ret < 0) {
383 goto exit;
385 s->curr_header = hdr_idx;
387 exit:
388 return ret;
392 * The VHDX spec calls for header updates to be performed twice, so that both
393 * the current and non-current header have valid info
395 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s,
396 bool generate_data_write_guid, MSGUID *log_guid)
398 int ret;
400 ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
401 if (ret < 0) {
402 return ret;
404 ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
405 return ret;
408 /* opens the specified header block from the VHDX file header section */
409 static void vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s,
410 Error **errp)
412 int ret;
413 VHDXHeader *header1;
414 VHDXHeader *header2;
415 bool h1_valid = false;
416 bool h2_valid = false;
417 uint64_t h1_seq = 0;
418 uint64_t h2_seq = 0;
419 uint8_t *buffer;
421 /* header1 & header2 are freed in vhdx_close() */
422 header1 = qemu_blockalign(bs, sizeof(VHDXHeader));
423 header2 = qemu_blockalign(bs, sizeof(VHDXHeader));
425 buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE);
427 s->headers[0] = header1;
428 s->headers[1] = header2;
430 /* We have to read the whole VHDX_HEADER_SIZE instead of
431 * sizeof(VHDXHeader), because the checksum is over the whole
432 * region */
433 ret = bdrv_pread(bs->file->bs, VHDX_HEADER1_OFFSET, buffer,
434 VHDX_HEADER_SIZE);
435 if (ret < 0) {
436 goto fail;
438 /* copy over just the relevant portion that we need */
439 memcpy(header1, buffer, sizeof(VHDXHeader));
441 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
442 vhdx_header_le_import(header1);
443 if (header1->signature == VHDX_HEADER_SIGNATURE &&
444 header1->version == 1) {
445 h1_seq = header1->sequence_number;
446 h1_valid = true;
450 ret = bdrv_pread(bs->file->bs, VHDX_HEADER2_OFFSET, buffer,
451 VHDX_HEADER_SIZE);
452 if (ret < 0) {
453 goto fail;
455 /* copy over just the relevant portion that we need */
456 memcpy(header2, buffer, sizeof(VHDXHeader));
458 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
459 vhdx_header_le_import(header2);
460 if (header2->signature == VHDX_HEADER_SIGNATURE &&
461 header2->version == 1) {
462 h2_seq = header2->sequence_number;
463 h2_valid = true;
467 /* If there is only 1 valid header (or no valid headers), we
468 * don't care what the sequence numbers are */
469 if (h1_valid && !h2_valid) {
470 s->curr_header = 0;
471 } else if (!h1_valid && h2_valid) {
472 s->curr_header = 1;
473 } else if (!h1_valid && !h2_valid) {
474 goto fail;
475 } else {
476 /* If both headers are valid, then we choose the active one by the
477 * highest sequence number. If the sequence numbers are equal, that is
478 * invalid */
479 if (h1_seq > h2_seq) {
480 s->curr_header = 0;
481 } else if (h2_seq > h1_seq) {
482 s->curr_header = 1;
483 } else {
484 /* The Microsoft Disk2VHD tool will create 2 identical
485 * headers, with identical sequence numbers. If the headers are
486 * identical, don't consider the file corrupt */
487 if (!memcmp(header1, header2, sizeof(VHDXHeader))) {
488 s->curr_header = 0;
489 } else {
490 goto fail;
495 vhdx_region_register(s, s->headers[s->curr_header]->log_offset,
496 s->headers[s->curr_header]->log_length);
497 goto exit;
499 fail:
500 error_setg_errno(errp, -ret, "No valid VHDX header found");
501 qemu_vfree(header1);
502 qemu_vfree(header2);
503 s->headers[0] = NULL;
504 s->headers[1] = NULL;
505 exit:
506 qemu_vfree(buffer);
510 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s)
512 int ret = 0;
513 uint8_t *buffer;
514 int offset = 0;
515 VHDXRegionTableEntry rt_entry;
516 uint32_t i;
517 bool bat_rt_found = false;
518 bool metadata_rt_found = false;
520 /* We have to read the whole 64KB block, because the crc32 is over the
521 * whole block */
522 buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE);
524 ret = bdrv_pread(bs->file->bs, VHDX_REGION_TABLE_OFFSET, buffer,
525 VHDX_HEADER_BLOCK_SIZE);
526 if (ret < 0) {
527 goto fail;
529 memcpy(&s->rt, buffer, sizeof(s->rt));
530 offset += sizeof(s->rt);
532 if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4)) {
533 ret = -EINVAL;
534 goto fail;
537 vhdx_region_header_le_import(&s->rt);
539 if (s->rt.signature != VHDX_REGION_SIGNATURE) {
540 ret = -EINVAL;
541 goto fail;
545 /* Per spec, maximum region table entry count is 2047 */
546 if (s->rt.entry_count > 2047) {
547 ret = -EINVAL;
548 goto fail;
551 for (i = 0; i < s->rt.entry_count; i++) {
552 memcpy(&rt_entry, buffer + offset, sizeof(rt_entry));
553 offset += sizeof(rt_entry);
555 vhdx_region_entry_le_import(&rt_entry);
557 /* check for region overlap between these entries, and any
558 * other memory regions in the file */
559 ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length);
560 if (ret < 0) {
561 goto fail;
564 vhdx_region_register(s, rt_entry.file_offset, rt_entry.length);
566 /* see if we recognize the entry */
567 if (guid_eq(rt_entry.guid, bat_guid)) {
568 /* must be unique; if we have already found it this is invalid */
569 if (bat_rt_found) {
570 ret = -EINVAL;
571 goto fail;
573 bat_rt_found = true;
574 s->bat_rt = rt_entry;
575 continue;
578 if (guid_eq(rt_entry.guid, metadata_guid)) {
579 /* must be unique; if we have already found it this is invalid */
580 if (metadata_rt_found) {
581 ret = -EINVAL;
582 goto fail;
584 metadata_rt_found = true;
585 s->metadata_rt = rt_entry;
586 continue;
589 if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) {
590 /* cannot read vhdx file - required region table entry that
591 * we do not understand. per spec, we must fail to open */
592 ret = -ENOTSUP;
593 goto fail;
597 if (!bat_rt_found || !metadata_rt_found) {
598 ret = -EINVAL;
599 goto fail;
602 ret = 0;
604 fail:
605 qemu_vfree(buffer);
606 return ret;
611 /* Metadata initial parser
613 * This loads all the metadata entry fields. This may cause additional
614 * fields to be processed (e.g. parent locator, etc..).
616 * There are 5 Metadata items that are always required:
617 * - File Parameters (block size, has a parent)
618 * - Virtual Disk Size (size, in bytes, of the virtual drive)
619 * - Page 83 Data (scsi page 83 guid)
620 * - Logical Sector Size (logical sector size in bytes, either 512 or
621 * 4096. We only support 512 currently)
622 * - Physical Sector Size (512 or 4096)
624 * Also, if the File Parameters indicate this is a differencing file,
625 * we must also look for the Parent Locator metadata item.
627 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s)
629 int ret = 0;
630 uint8_t *buffer;
631 int offset = 0;
632 uint32_t i = 0;
633 VHDXMetadataTableEntry md_entry;
635 buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE);
637 ret = bdrv_pread(bs->file->bs, s->metadata_rt.file_offset, buffer,
638 VHDX_METADATA_TABLE_MAX_SIZE);
639 if (ret < 0) {
640 goto exit;
642 memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr));
643 offset += sizeof(s->metadata_hdr);
645 vhdx_metadata_header_le_import(&s->metadata_hdr);
647 if (s->metadata_hdr.signature != VHDX_METADATA_SIGNATURE) {
648 ret = -EINVAL;
649 goto exit;
652 s->metadata_entries.present = 0;
654 if ((s->metadata_hdr.entry_count * sizeof(md_entry)) >
655 (VHDX_METADATA_TABLE_MAX_SIZE - offset)) {
656 ret = -EINVAL;
657 goto exit;
660 for (i = 0; i < s->metadata_hdr.entry_count; i++) {
661 memcpy(&md_entry, buffer + offset, sizeof(md_entry));
662 offset += sizeof(md_entry);
664 vhdx_metadata_entry_le_import(&md_entry);
666 if (guid_eq(md_entry.item_id, file_param_guid)) {
667 if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) {
668 ret = -EINVAL;
669 goto exit;
671 s->metadata_entries.file_parameters_entry = md_entry;
672 s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT;
673 continue;
676 if (guid_eq(md_entry.item_id, virtual_size_guid)) {
677 if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) {
678 ret = -EINVAL;
679 goto exit;
681 s->metadata_entries.virtual_disk_size_entry = md_entry;
682 s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT;
683 continue;
686 if (guid_eq(md_entry.item_id, page83_guid)) {
687 if (s->metadata_entries.present & META_PAGE_83_PRESENT) {
688 ret = -EINVAL;
689 goto exit;
691 s->metadata_entries.page83_data_entry = md_entry;
692 s->metadata_entries.present |= META_PAGE_83_PRESENT;
693 continue;
696 if (guid_eq(md_entry.item_id, logical_sector_guid)) {
697 if (s->metadata_entries.present &
698 META_LOGICAL_SECTOR_SIZE_PRESENT) {
699 ret = -EINVAL;
700 goto exit;
702 s->metadata_entries.logical_sector_size_entry = md_entry;
703 s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT;
704 continue;
707 if (guid_eq(md_entry.item_id, phys_sector_guid)) {
708 if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) {
709 ret = -EINVAL;
710 goto exit;
712 s->metadata_entries.phys_sector_size_entry = md_entry;
713 s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT;
714 continue;
717 if (guid_eq(md_entry.item_id, parent_locator_guid)) {
718 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
719 ret = -EINVAL;
720 goto exit;
722 s->metadata_entries.parent_locator_entry = md_entry;
723 s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT;
724 continue;
727 if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) {
728 /* cannot read vhdx file - required region table entry that
729 * we do not understand. per spec, we must fail to open */
730 ret = -ENOTSUP;
731 goto exit;
735 if (s->metadata_entries.present != META_ALL_PRESENT) {
736 ret = -ENOTSUP;
737 goto exit;
740 ret = bdrv_pread(bs->file->bs,
741 s->metadata_entries.file_parameters_entry.offset
742 + s->metadata_rt.file_offset,
743 &s->params,
744 sizeof(s->params));
746 if (ret < 0) {
747 goto exit;
750 le32_to_cpus(&s->params.block_size);
751 le32_to_cpus(&s->params.data_bits);
754 /* We now have the file parameters, so we can tell if this is a
755 * differencing file (i.e.. has_parent), is dynamic or fixed
756 * sized (leave_blocks_allocated), and the block size */
758 /* The parent locator required iff the file parameters has_parent set */
759 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
760 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
761 /* TODO: parse parent locator fields */
762 ret = -ENOTSUP; /* temp, until differencing files are supported */
763 goto exit;
764 } else {
765 /* if has_parent is set, but there is not parent locator present,
766 * then that is an invalid combination */
767 ret = -EINVAL;
768 goto exit;
772 /* determine virtual disk size, logical sector size,
773 * and phys sector size */
775 ret = bdrv_pread(bs->file->bs,
776 s->metadata_entries.virtual_disk_size_entry.offset
777 + s->metadata_rt.file_offset,
778 &s->virtual_disk_size,
779 sizeof(uint64_t));
780 if (ret < 0) {
781 goto exit;
783 ret = bdrv_pread(bs->file->bs,
784 s->metadata_entries.logical_sector_size_entry.offset
785 + s->metadata_rt.file_offset,
786 &s->logical_sector_size,
787 sizeof(uint32_t));
788 if (ret < 0) {
789 goto exit;
791 ret = bdrv_pread(bs->file->bs,
792 s->metadata_entries.phys_sector_size_entry.offset
793 + s->metadata_rt.file_offset,
794 &s->physical_sector_size,
795 sizeof(uint32_t));
796 if (ret < 0) {
797 goto exit;
800 le64_to_cpus(&s->virtual_disk_size);
801 le32_to_cpus(&s->logical_sector_size);
802 le32_to_cpus(&s->physical_sector_size);
804 if (s->params.block_size < VHDX_BLOCK_SIZE_MIN ||
805 s->params.block_size > VHDX_BLOCK_SIZE_MAX) {
806 ret = -EINVAL;
807 goto exit;
810 /* only 2 supported sector sizes */
811 if (s->logical_sector_size != 512 && s->logical_sector_size != 4096) {
812 ret = -EINVAL;
813 goto exit;
816 /* Both block_size and sector_size are guaranteed powers of 2, below.
817 Due to range checks above, s->sectors_per_block can never be < 256 */
818 s->sectors_per_block = s->params.block_size / s->logical_sector_size;
819 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
820 (uint64_t)s->logical_sector_size /
821 (uint64_t)s->params.block_size;
823 /* These values are ones we will want to use for division / multiplication
824 * later on, and they are all guaranteed (per the spec) to be powers of 2,
825 * so we can take advantage of that for shift operations during
826 * reads/writes */
827 if (s->logical_sector_size & (s->logical_sector_size - 1)) {
828 ret = -EINVAL;
829 goto exit;
831 if (s->sectors_per_block & (s->sectors_per_block - 1)) {
832 ret = -EINVAL;
833 goto exit;
835 if (s->chunk_ratio & (s->chunk_ratio - 1)) {
836 ret = -EINVAL;
837 goto exit;
839 s->block_size = s->params.block_size;
840 if (s->block_size & (s->block_size - 1)) {
841 ret = -EINVAL;
842 goto exit;
845 vhdx_set_shift_bits(s);
847 ret = 0;
849 exit:
850 qemu_vfree(buffer);
851 return ret;
855 * Calculate the number of BAT entries, including sector
856 * bitmap entries.
858 static void vhdx_calc_bat_entries(BDRVVHDXState *s)
860 uint32_t data_blocks_cnt, bitmap_blocks_cnt;
862 data_blocks_cnt = DIV_ROUND_UP(s->virtual_disk_size, s->block_size);
863 bitmap_blocks_cnt = DIV_ROUND_UP(data_blocks_cnt, s->chunk_ratio);
865 if (s->parent_entries) {
866 s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1);
867 } else {
868 s->bat_entries = data_blocks_cnt +
869 ((data_blocks_cnt - 1) >> s->chunk_ratio_bits);
874 static void vhdx_close(BlockDriverState *bs)
876 BDRVVHDXState *s = bs->opaque;
877 qemu_vfree(s->headers[0]);
878 s->headers[0] = NULL;
879 qemu_vfree(s->headers[1]);
880 s->headers[1] = NULL;
881 qemu_vfree(s->bat);
882 s->bat = NULL;
883 qemu_vfree(s->parent_entries);
884 s->parent_entries = NULL;
885 migrate_del_blocker(s->migration_blocker);
886 error_free(s->migration_blocker);
887 qemu_vfree(s->log.hdr);
888 s->log.hdr = NULL;
889 vhdx_region_unregister_all(s);
892 static int vhdx_open(BlockDriverState *bs, QDict *options, int flags,
893 Error **errp)
895 BDRVVHDXState *s = bs->opaque;
896 int ret = 0;
897 uint32_t i;
898 uint64_t signature;
899 Error *local_err = NULL;
901 s->bat = NULL;
902 s->first_visible_write = true;
904 qemu_co_mutex_init(&s->lock);
905 QLIST_INIT(&s->regions);
907 /* validate the file signature */
908 ret = bdrv_pread(bs->file->bs, 0, &signature, sizeof(uint64_t));
909 if (ret < 0) {
910 goto fail;
912 if (memcmp(&signature, "vhdxfile", 8)) {
913 ret = -EINVAL;
914 goto fail;
917 /* This is used for any header updates, for the file_write_guid.
918 * The spec dictates that a new value should be used for the first
919 * header update */
920 vhdx_guid_generate(&s->session_guid);
922 vhdx_parse_header(bs, s, &local_err);
923 if (local_err != NULL) {
924 error_propagate(errp, local_err);
925 ret = -EINVAL;
926 goto fail;
929 ret = vhdx_parse_log(bs, s, &s->log_replayed_on_open, errp);
930 if (ret < 0) {
931 goto fail;
934 ret = vhdx_open_region_tables(bs, s);
935 if (ret < 0) {
936 goto fail;
939 ret = vhdx_parse_metadata(bs, s);
940 if (ret < 0) {
941 goto fail;
944 s->block_size = s->params.block_size;
946 /* the VHDX spec dictates that virtual_disk_size is always a multiple of
947 * logical_sector_size */
948 bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits;
950 vhdx_calc_bat_entries(s);
952 s->bat_offset = s->bat_rt.file_offset;
954 if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) {
955 /* BAT allocation is not large enough for all entries */
956 ret = -EINVAL;
957 goto fail;
960 /* s->bat is freed in vhdx_close() */
961 s->bat = qemu_try_blockalign(bs->file->bs, s->bat_rt.length);
962 if (s->bat == NULL) {
963 ret = -ENOMEM;
964 goto fail;
967 ret = bdrv_pread(bs->file->bs, s->bat_offset, s->bat, s->bat_rt.length);
968 if (ret < 0) {
969 goto fail;
972 uint64_t payblocks = s->chunk_ratio;
973 /* endian convert, and verify populated BAT field file offsets against
974 * region table and log entries */
975 for (i = 0; i < s->bat_entries; i++) {
976 le64_to_cpus(&s->bat[i]);
977 if (payblocks--) {
978 /* payload bat entries */
979 if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) ==
980 PAYLOAD_BLOCK_FULLY_PRESENT) {
981 ret = vhdx_region_check(s, s->bat[i] & VHDX_BAT_FILE_OFF_MASK,
982 s->block_size);
983 if (ret < 0) {
984 goto fail;
987 } else {
988 payblocks = s->chunk_ratio;
989 /* Once differencing files are supported, verify sector bitmap
990 * blocks here */
994 if (flags & BDRV_O_RDWR) {
995 ret = vhdx_update_headers(bs, s, false, NULL);
996 if (ret < 0) {
997 goto fail;
1001 /* TODO: differencing files */
1003 /* Disable migration when VHDX images are used */
1004 error_setg(&s->migration_blocker, "The vhdx format used by node '%s' "
1005 "does not support live migration",
1006 bdrv_get_device_or_node_name(bs));
1007 migrate_add_blocker(s->migration_blocker);
1009 return 0;
1010 fail:
1011 vhdx_close(bs);
1012 return ret;
1015 static int vhdx_reopen_prepare(BDRVReopenState *state,
1016 BlockReopenQueue *queue, Error **errp)
1018 return 0;
1023 * Perform sector to block offset translations, to get various
1024 * sector and file offsets into the image. See VHDXSectorInfo
1026 static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num,
1027 int nb_sectors, VHDXSectorInfo *sinfo)
1029 uint32_t block_offset;
1031 sinfo->bat_idx = sector_num >> s->sectors_per_block_bits;
1032 /* effectively a modulo - this gives us the offset into the block
1033 * (in sector sizes) for our sector number */
1034 block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits);
1035 /* the chunk ratio gives us the interleaving of the sector
1036 * bitmaps, so we need to advance our page block index by the
1037 * sector bitmaps entry number */
1038 sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits;
1040 /* the number of sectors we can read/write in this cycle */
1041 sinfo->sectors_avail = s->sectors_per_block - block_offset;
1043 sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits;
1045 if (sinfo->sectors_avail > nb_sectors) {
1046 sinfo->sectors_avail = nb_sectors;
1049 sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits;
1051 sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK;
1053 sinfo->block_offset = block_offset << s->logical_sector_size_bits;
1055 /* The file offset must be past the header section, so must be > 0 */
1056 if (sinfo->file_offset == 0) {
1057 return;
1060 /* block offset is the offset in vhdx logical sectors, in
1061 * the payload data block. Convert that to a byte offset
1062 * in the block, and add in the payload data block offset
1063 * in the file, in bytes, to get the final read address */
1065 sinfo->file_offset += sinfo->block_offset;
1069 static int vhdx_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1071 BDRVVHDXState *s = bs->opaque;
1073 bdi->cluster_size = s->block_size;
1075 bdi->unallocated_blocks_are_zero =
1076 (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) == 0;
1078 return 0;
1082 static coroutine_fn int vhdx_co_readv(BlockDriverState *bs, int64_t sector_num,
1083 int nb_sectors, QEMUIOVector *qiov)
1085 BDRVVHDXState *s = bs->opaque;
1086 int ret = 0;
1087 VHDXSectorInfo sinfo;
1088 uint64_t bytes_done = 0;
1089 QEMUIOVector hd_qiov;
1091 qemu_iovec_init(&hd_qiov, qiov->niov);
1093 qemu_co_mutex_lock(&s->lock);
1095 while (nb_sectors > 0) {
1096 /* We are a differencing file, so we need to inspect the sector bitmap
1097 * to see if we have the data or not */
1098 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1099 /* not supported yet */
1100 ret = -ENOTSUP;
1101 goto exit;
1102 } else {
1103 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1105 qemu_iovec_reset(&hd_qiov);
1106 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, sinfo.bytes_avail);
1108 /* check the payload block state */
1109 switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) {
1110 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1111 case PAYLOAD_BLOCK_UNDEFINED:
1112 case PAYLOAD_BLOCK_UNMAPPED:
1113 case PAYLOAD_BLOCK_UNMAPPED_v095:
1114 case PAYLOAD_BLOCK_ZERO:
1115 /* return zero */
1116 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail);
1117 break;
1118 case PAYLOAD_BLOCK_FULLY_PRESENT:
1119 qemu_co_mutex_unlock(&s->lock);
1120 ret = bdrv_co_readv(bs->file->bs,
1121 sinfo.file_offset >> BDRV_SECTOR_BITS,
1122 sinfo.sectors_avail, &hd_qiov);
1123 qemu_co_mutex_lock(&s->lock);
1124 if (ret < 0) {
1125 goto exit;
1127 break;
1128 case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1129 /* we don't yet support difference files, fall through
1130 * to error */
1131 default:
1132 ret = -EIO;
1133 goto exit;
1134 break;
1136 nb_sectors -= sinfo.sectors_avail;
1137 sector_num += sinfo.sectors_avail;
1138 bytes_done += sinfo.bytes_avail;
1141 ret = 0;
1142 exit:
1143 qemu_co_mutex_unlock(&s->lock);
1144 qemu_iovec_destroy(&hd_qiov);
1145 return ret;
1149 * Allocate a new payload block at the end of the file.
1151 * Allocation will happen at 1MB alignment inside the file
1153 * Returns the file offset start of the new payload block
1155 static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s,
1156 uint64_t *new_offset)
1158 *new_offset = bdrv_getlength(bs->file->bs);
1160 /* per the spec, the address for a block is in units of 1MB */
1161 *new_offset = ROUND_UP(*new_offset, 1024 * 1024);
1163 return bdrv_truncate(bs->file->bs, *new_offset + s->block_size);
1167 * Update the BAT table entry with the new file offset, and the new entry
1168 * state */
1169 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s,
1170 VHDXSectorInfo *sinfo,
1171 uint64_t *bat_entry_le,
1172 uint64_t *bat_offset, int state)
1174 /* The BAT entry is a uint64, with 44 bits for the file offset in units of
1175 * 1MB, and 3 bits for the block state. */
1176 if ((state == PAYLOAD_BLOCK_ZERO) ||
1177 (state == PAYLOAD_BLOCK_UNDEFINED) ||
1178 (state == PAYLOAD_BLOCK_NOT_PRESENT) ||
1179 (state == PAYLOAD_BLOCK_UNMAPPED)) {
1180 s->bat[sinfo->bat_idx] = 0; /* For PAYLOAD_BLOCK_ZERO, the
1181 FileOffsetMB field is denoted as
1182 'reserved' in the v1.0 spec. If it is
1183 non-zero, MS Hyper-V will fail to read
1184 the disk image */
1185 } else {
1186 s->bat[sinfo->bat_idx] = sinfo->file_offset;
1189 s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK;
1191 *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]);
1192 *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry);
1196 /* Per the spec, on the first write of guest-visible data to the file the
1197 * data write guid must be updated in the header */
1198 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s)
1200 int ret = 0;
1201 if (s->first_visible_write) {
1202 s->first_visible_write = false;
1203 ret = vhdx_update_headers(bs, s, true, NULL);
1205 return ret;
1208 static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num,
1209 int nb_sectors, QEMUIOVector *qiov)
1211 int ret = -ENOTSUP;
1212 BDRVVHDXState *s = bs->opaque;
1213 VHDXSectorInfo sinfo;
1214 uint64_t bytes_done = 0;
1215 uint64_t bat_entry = 0;
1216 uint64_t bat_entry_offset = 0;
1217 QEMUIOVector hd_qiov;
1218 struct iovec iov1 = { 0 };
1219 struct iovec iov2 = { 0 };
1220 int sectors_to_write;
1221 int bat_state;
1222 uint64_t bat_prior_offset = 0;
1223 bool bat_update = false;
1225 qemu_iovec_init(&hd_qiov, qiov->niov);
1227 qemu_co_mutex_lock(&s->lock);
1229 ret = vhdx_user_visible_write(bs, s);
1230 if (ret < 0) {
1231 goto exit;
1234 while (nb_sectors > 0) {
1235 bool use_zero_buffers = false;
1236 bat_update = false;
1237 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1238 /* not supported yet */
1239 ret = -ENOTSUP;
1240 goto exit;
1241 } else {
1242 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1243 sectors_to_write = sinfo.sectors_avail;
1245 qemu_iovec_reset(&hd_qiov);
1246 /* check the payload block state */
1247 bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK;
1248 switch (bat_state) {
1249 case PAYLOAD_BLOCK_ZERO:
1250 /* in this case, we need to preserve zero writes for
1251 * data that is not part of this write, so we must pad
1252 * the rest of the buffer to zeroes */
1254 /* if we are on a posix system with ftruncate() that extends
1255 * a file, then it is zero-filled for us. On Win32, the raw
1256 * layer uses SetFilePointer and SetFileEnd, which does not
1257 * zero fill AFAIK */
1259 /* Queue another write of zero buffers if the underlying file
1260 * does not zero-fill on file extension */
1262 if (bdrv_has_zero_init(bs->file->bs) == 0) {
1263 use_zero_buffers = true;
1265 /* zero fill the front, if any */
1266 if (sinfo.block_offset) {
1267 iov1.iov_len = sinfo.block_offset;
1268 iov1.iov_base = qemu_blockalign(bs, iov1.iov_len);
1269 memset(iov1.iov_base, 0, iov1.iov_len);
1270 qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0,
1271 iov1.iov_len);
1272 sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS;
1275 /* our actual data */
1276 qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1277 sinfo.bytes_avail);
1279 /* zero fill the back, if any */
1280 if ((sinfo.bytes_avail - sinfo.block_offset) <
1281 s->block_size) {
1282 iov2.iov_len = s->block_size -
1283 (sinfo.bytes_avail + sinfo.block_offset);
1284 iov2.iov_base = qemu_blockalign(bs, iov2.iov_len);
1285 memset(iov2.iov_base, 0, iov2.iov_len);
1286 qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0,
1287 iov2.iov_len);
1288 sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS;
1291 /* fall through */
1292 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1293 case PAYLOAD_BLOCK_UNMAPPED:
1294 case PAYLOAD_BLOCK_UNMAPPED_v095:
1295 case PAYLOAD_BLOCK_UNDEFINED:
1296 bat_prior_offset = sinfo.file_offset;
1297 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset);
1298 if (ret < 0) {
1299 goto exit;
1301 /* once we support differencing files, this may also be
1302 * partially present */
1303 /* update block state to the newly specified state */
1304 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1305 &bat_entry_offset,
1306 PAYLOAD_BLOCK_FULLY_PRESENT);
1307 bat_update = true;
1308 /* since we just allocated a block, file_offset is the
1309 * beginning of the payload block. It needs to be the
1310 * write address, which includes the offset into the block */
1311 if (!use_zero_buffers) {
1312 sinfo.file_offset += sinfo.block_offset;
1314 /* fall through */
1315 case PAYLOAD_BLOCK_FULLY_PRESENT:
1316 /* if the file offset address is in the header zone,
1317 * there is a problem */
1318 if (sinfo.file_offset < (1024 * 1024)) {
1319 ret = -EFAULT;
1320 goto error_bat_restore;
1323 if (!use_zero_buffers) {
1324 qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1325 sinfo.bytes_avail);
1327 /* block exists, so we can just overwrite it */
1328 qemu_co_mutex_unlock(&s->lock);
1329 ret = bdrv_co_writev(bs->file->bs,
1330 sinfo.file_offset >> BDRV_SECTOR_BITS,
1331 sectors_to_write, &hd_qiov);
1332 qemu_co_mutex_lock(&s->lock);
1333 if (ret < 0) {
1334 goto error_bat_restore;
1336 break;
1337 case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1338 /* we don't yet support difference files, fall through
1339 * to error */
1340 default:
1341 ret = -EIO;
1342 goto exit;
1343 break;
1346 if (bat_update) {
1347 /* this will update the BAT entry into the log journal, and
1348 * then flush the log journal out to disk */
1349 ret = vhdx_log_write_and_flush(bs, s, &bat_entry,
1350 sizeof(VHDXBatEntry),
1351 bat_entry_offset);
1352 if (ret < 0) {
1353 goto exit;
1357 nb_sectors -= sinfo.sectors_avail;
1358 sector_num += sinfo.sectors_avail;
1359 bytes_done += sinfo.bytes_avail;
1364 goto exit;
1366 error_bat_restore:
1367 if (bat_update) {
1368 /* keep metadata in sync, and restore the bat entry state
1369 * if error. */
1370 sinfo.file_offset = bat_prior_offset;
1371 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1372 &bat_entry_offset, bat_state);
1374 exit:
1375 qemu_vfree(iov1.iov_base);
1376 qemu_vfree(iov2.iov_base);
1377 qemu_co_mutex_unlock(&s->lock);
1378 qemu_iovec_destroy(&hd_qiov);
1379 return ret;
1385 * Create VHDX Headers
1387 * There are 2 headers, and the highest sequence number will represent
1388 * the active header
1390 static int vhdx_create_new_headers(BlockDriverState *bs, uint64_t image_size,
1391 uint32_t log_size)
1393 int ret = 0;
1394 VHDXHeader *hdr = NULL;
1396 hdr = g_new0(VHDXHeader, 1);
1398 hdr->signature = VHDX_HEADER_SIGNATURE;
1399 hdr->sequence_number = g_random_int();
1400 hdr->log_version = 0;
1401 hdr->version = 1;
1402 hdr->log_length = log_size;
1403 hdr->log_offset = VHDX_HEADER_SECTION_END;
1404 vhdx_guid_generate(&hdr->file_write_guid);
1405 vhdx_guid_generate(&hdr->data_write_guid);
1407 ret = vhdx_write_header(bs, hdr, VHDX_HEADER1_OFFSET, false);
1408 if (ret < 0) {
1409 goto exit;
1411 hdr->sequence_number++;
1412 ret = vhdx_write_header(bs, hdr, VHDX_HEADER2_OFFSET, false);
1413 if (ret < 0) {
1414 goto exit;
1417 exit:
1418 g_free(hdr);
1419 return ret;
1422 #define VHDX_METADATA_ENTRY_BUFFER_SIZE \
1423 (sizeof(VHDXFileParameters) +\
1424 sizeof(VHDXVirtualDiskSize) +\
1425 sizeof(VHDXPage83Data) +\
1426 sizeof(VHDXVirtualDiskLogicalSectorSize) +\
1427 sizeof(VHDXVirtualDiskPhysicalSectorSize))
1430 * Create the Metadata entries.
1432 * For more details on the entries, see section 3.5 (pg 29) in the
1433 * VHDX 1.00 specification.
1435 * We support 5 metadata entries (all required by spec):
1436 * File Parameters,
1437 * Virtual Disk Size,
1438 * Page 83 Data,
1439 * Logical Sector Size,
1440 * Physical Sector Size
1442 * The first 64KB of the Metadata section is reserved for the metadata
1443 * header and entries; beyond that, the metadata items themselves reside.
1445 static int vhdx_create_new_metadata(BlockDriverState *bs,
1446 uint64_t image_size,
1447 uint32_t block_size,
1448 uint32_t sector_size,
1449 uint64_t metadata_offset,
1450 VHDXImageType type)
1452 int ret = 0;
1453 uint32_t offset = 0;
1454 void *buffer = NULL;
1455 void *entry_buffer;
1456 VHDXMetadataTableHeader *md_table;
1457 VHDXMetadataTableEntry *md_table_entry;
1459 /* Metadata entries */
1460 VHDXFileParameters *mt_file_params;
1461 VHDXVirtualDiskSize *mt_virtual_size;
1462 VHDXPage83Data *mt_page83;
1463 VHDXVirtualDiskLogicalSectorSize *mt_log_sector_size;
1464 VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size;
1466 entry_buffer = g_malloc0(VHDX_METADATA_ENTRY_BUFFER_SIZE);
1468 mt_file_params = entry_buffer;
1469 offset += sizeof(VHDXFileParameters);
1470 mt_virtual_size = entry_buffer + offset;
1471 offset += sizeof(VHDXVirtualDiskSize);
1472 mt_page83 = entry_buffer + offset;
1473 offset += sizeof(VHDXPage83Data);
1474 mt_log_sector_size = entry_buffer + offset;
1475 offset += sizeof(VHDXVirtualDiskLogicalSectorSize);
1476 mt_phys_sector_size = entry_buffer + offset;
1478 mt_file_params->block_size = cpu_to_le32(block_size);
1479 if (type == VHDX_TYPE_FIXED) {
1480 mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED;
1481 cpu_to_le32s(&mt_file_params->data_bits);
1484 vhdx_guid_generate(&mt_page83->page_83_data);
1485 cpu_to_leguids(&mt_page83->page_83_data);
1486 mt_virtual_size->virtual_disk_size = cpu_to_le64(image_size);
1487 mt_log_sector_size->logical_sector_size = cpu_to_le32(sector_size);
1488 mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size);
1490 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1491 md_table = buffer;
1493 md_table->signature = VHDX_METADATA_SIGNATURE;
1494 md_table->entry_count = 5;
1495 vhdx_metadata_header_le_export(md_table);
1498 /* This will reference beyond the reserved table portion */
1499 offset = 64 * KiB;
1501 md_table_entry = buffer + sizeof(VHDXMetadataTableHeader);
1503 md_table_entry[0].item_id = file_param_guid;
1504 md_table_entry[0].offset = offset;
1505 md_table_entry[0].length = sizeof(VHDXFileParameters);
1506 md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED;
1507 offset += md_table_entry[0].length;
1508 vhdx_metadata_entry_le_export(&md_table_entry[0]);
1510 md_table_entry[1].item_id = virtual_size_guid;
1511 md_table_entry[1].offset = offset;
1512 md_table_entry[1].length = sizeof(VHDXVirtualDiskSize);
1513 md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1514 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1515 offset += md_table_entry[1].length;
1516 vhdx_metadata_entry_le_export(&md_table_entry[1]);
1518 md_table_entry[2].item_id = page83_guid;
1519 md_table_entry[2].offset = offset;
1520 md_table_entry[2].length = sizeof(VHDXPage83Data);
1521 md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1522 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1523 offset += md_table_entry[2].length;
1524 vhdx_metadata_entry_le_export(&md_table_entry[2]);
1526 md_table_entry[3].item_id = logical_sector_guid;
1527 md_table_entry[3].offset = offset;
1528 md_table_entry[3].length = sizeof(VHDXVirtualDiskLogicalSectorSize);
1529 md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1530 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1531 offset += md_table_entry[3].length;
1532 vhdx_metadata_entry_le_export(&md_table_entry[3]);
1534 md_table_entry[4].item_id = phys_sector_guid;
1535 md_table_entry[4].offset = offset;
1536 md_table_entry[4].length = sizeof(VHDXVirtualDiskPhysicalSectorSize);
1537 md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1538 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1539 vhdx_metadata_entry_le_export(&md_table_entry[4]);
1541 ret = bdrv_pwrite(bs, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE);
1542 if (ret < 0) {
1543 goto exit;
1546 ret = bdrv_pwrite(bs, metadata_offset + (64 * KiB), entry_buffer,
1547 VHDX_METADATA_ENTRY_BUFFER_SIZE);
1548 if (ret < 0) {
1549 goto exit;
1553 exit:
1554 g_free(buffer);
1555 g_free(entry_buffer);
1556 return ret;
1559 /* This create the actual BAT itself. We currently only support
1560 * 'Dynamic' and 'Fixed' image types.
1562 * Dynamic images: default state of the BAT is all zeroes.
1564 * Fixed images: default state of the BAT is fully populated, with
1565 * file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT.
1567 static int vhdx_create_bat(BlockDriverState *bs, BDRVVHDXState *s,
1568 uint64_t image_size, VHDXImageType type,
1569 bool use_zero_blocks, uint64_t file_offset,
1570 uint32_t length)
1572 int ret = 0;
1573 uint64_t data_file_offset;
1574 uint64_t total_sectors = 0;
1575 uint64_t sector_num = 0;
1576 uint64_t unused;
1577 int block_state;
1578 VHDXSectorInfo sinfo;
1580 assert(s->bat == NULL);
1582 /* this gives a data start after BAT/bitmap entries, and well
1583 * past any metadata entries (with a 4 MB buffer for future
1584 * expansion */
1585 data_file_offset = file_offset + length + 5 * MiB;
1586 total_sectors = image_size >> s->logical_sector_size_bits;
1588 if (type == VHDX_TYPE_DYNAMIC) {
1589 /* All zeroes, so we can just extend the file - the end of the BAT
1590 * is the furthest thing we have written yet */
1591 ret = bdrv_truncate(bs, data_file_offset);
1592 if (ret < 0) {
1593 goto exit;
1595 } else if (type == VHDX_TYPE_FIXED) {
1596 ret = bdrv_truncate(bs, data_file_offset + image_size);
1597 if (ret < 0) {
1598 goto exit;
1600 } else {
1601 ret = -ENOTSUP;
1602 goto exit;
1605 if (type == VHDX_TYPE_FIXED ||
1606 use_zero_blocks ||
1607 bdrv_has_zero_init(bs) == 0) {
1608 /* for a fixed file, the default BAT entry is not zero */
1609 s->bat = g_try_malloc0(length);
1610 if (length && s->bat == NULL) {
1611 ret = -ENOMEM;
1612 goto exit;
1614 block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT :
1615 PAYLOAD_BLOCK_NOT_PRESENT;
1616 block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state;
1617 /* fill the BAT by emulating sector writes of sectors_per_block size */
1618 while (sector_num < total_sectors) {
1619 vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo);
1620 sinfo.file_offset = data_file_offset +
1621 (sector_num << s->logical_sector_size_bits);
1622 sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB);
1623 vhdx_update_bat_table_entry(bs, s, &sinfo, &unused, &unused,
1624 block_state);
1625 cpu_to_le64s(&s->bat[sinfo.bat_idx]);
1626 sector_num += s->sectors_per_block;
1628 ret = bdrv_pwrite(bs, file_offset, s->bat, length);
1629 if (ret < 0) {
1630 goto exit;
1636 exit:
1637 g_free(s->bat);
1638 return ret;
1641 /* Creates the region table header, and region table entries.
1642 * There are 2 supported region table entries: BAT, and Metadata/
1644 * As the calculations for the BAT region table are also needed
1645 * to create the BAT itself, we will also cause the BAT to be
1646 * created.
1648 static int vhdx_create_new_region_table(BlockDriverState *bs,
1649 uint64_t image_size,
1650 uint32_t block_size,
1651 uint32_t sector_size,
1652 uint32_t log_size,
1653 bool use_zero_blocks,
1654 VHDXImageType type,
1655 uint64_t *metadata_offset)
1657 int ret = 0;
1658 uint32_t offset = 0;
1659 void *buffer = NULL;
1660 uint64_t bat_file_offset;
1661 uint32_t bat_length;
1662 BDRVVHDXState *s = NULL;
1663 VHDXRegionTableHeader *region_table;
1664 VHDXRegionTableEntry *rt_bat;
1665 VHDXRegionTableEntry *rt_metadata;
1667 assert(metadata_offset != NULL);
1669 /* Populate enough of the BDRVVHDXState to be able to use the
1670 * pre-existing BAT calculation, translation, and update functions */
1671 s = g_new0(BDRVVHDXState, 1);
1673 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
1674 (uint64_t) sector_size / (uint64_t) block_size;
1676 s->sectors_per_block = block_size / sector_size;
1677 s->virtual_disk_size = image_size;
1678 s->block_size = block_size;
1679 s->logical_sector_size = sector_size;
1681 vhdx_set_shift_bits(s);
1683 vhdx_calc_bat_entries(s);
1685 /* At this point the VHDX state is populated enough for creation */
1687 /* a single buffer is used so we can calculate the checksum over the
1688 * entire 64KB block */
1689 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1690 region_table = buffer;
1691 offset += sizeof(VHDXRegionTableHeader);
1692 rt_bat = buffer + offset;
1693 offset += sizeof(VHDXRegionTableEntry);
1694 rt_metadata = buffer + offset;
1696 region_table->signature = VHDX_REGION_SIGNATURE;
1697 region_table->entry_count = 2; /* BAT and Metadata */
1699 rt_bat->guid = bat_guid;
1700 rt_bat->length = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB);
1701 rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB);
1702 s->bat_offset = rt_bat->file_offset;
1704 rt_metadata->guid = metadata_guid;
1705 rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length,
1706 MiB);
1707 rt_metadata->length = 1 * MiB; /* min size, and more than enough */
1708 *metadata_offset = rt_metadata->file_offset;
1710 bat_file_offset = rt_bat->file_offset;
1711 bat_length = rt_bat->length;
1713 vhdx_region_header_le_export(region_table);
1714 vhdx_region_entry_le_export(rt_bat);
1715 vhdx_region_entry_le_export(rt_metadata);
1717 vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE,
1718 offsetof(VHDXRegionTableHeader, checksum));
1721 /* The region table gives us the data we need to create the BAT,
1722 * so do that now */
1723 ret = vhdx_create_bat(bs, s, image_size, type, use_zero_blocks,
1724 bat_file_offset, bat_length);
1725 if (ret < 0) {
1726 goto exit;
1729 /* Now write out the region headers to disk */
1730 ret = bdrv_pwrite(bs, VHDX_REGION_TABLE_OFFSET, buffer,
1731 VHDX_HEADER_BLOCK_SIZE);
1732 if (ret < 0) {
1733 goto exit;
1736 ret = bdrv_pwrite(bs, VHDX_REGION_TABLE2_OFFSET, buffer,
1737 VHDX_HEADER_BLOCK_SIZE);
1738 if (ret < 0) {
1739 goto exit;
1742 exit:
1743 g_free(s);
1744 g_free(buffer);
1745 return ret;
1748 /* We need to create the following elements:
1750 * .-----------------------------------------------------------------.
1751 * | (A) | (B) | (C) | (D) | (E) |
1752 * | File ID | Header1 | Header 2 | Region Tbl 1 | Region Tbl 2 |
1753 * | | | | | |
1754 * .-----------------------------------------------------------------.
1755 * 0 64KB 128KB 192KB 256KB 320KB
1758 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1759 * | (F) | (G) | (H) | |
1760 * | Journal Log | BAT / Bitmap | Metadata | .... data ...... |
1761 * | | | | |
1762 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1763 * 1MB
1765 static int vhdx_create(const char *filename, QemuOpts *opts, Error **errp)
1767 int ret = 0;
1768 uint64_t image_size = (uint64_t) 2 * GiB;
1769 uint32_t log_size = 1 * MiB;
1770 uint32_t block_size = 0;
1771 uint64_t signature;
1772 uint64_t metadata_offset;
1773 bool use_zero_blocks = false;
1775 gunichar2 *creator = NULL;
1776 glong creator_items;
1777 BlockBackend *blk;
1778 char *type = NULL;
1779 VHDXImageType image_type;
1780 Error *local_err = NULL;
1782 image_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1783 BDRV_SECTOR_SIZE);
1784 log_size = qemu_opt_get_size_del(opts, VHDX_BLOCK_OPT_LOG_SIZE, 0);
1785 block_size = qemu_opt_get_size_del(opts, VHDX_BLOCK_OPT_BLOCK_SIZE, 0);
1786 type = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1787 use_zero_blocks = qemu_opt_get_bool_del(opts, VHDX_BLOCK_OPT_ZERO, true);
1789 if (image_size > VHDX_MAX_IMAGE_SIZE) {
1790 error_setg_errno(errp, EINVAL, "Image size too large; max of 64TB");
1791 ret = -EINVAL;
1792 goto exit;
1795 if (type == NULL) {
1796 type = g_strdup("dynamic");
1799 if (!strcmp(type, "dynamic")) {
1800 image_type = VHDX_TYPE_DYNAMIC;
1801 } else if (!strcmp(type, "fixed")) {
1802 image_type = VHDX_TYPE_FIXED;
1803 } else if (!strcmp(type, "differencing")) {
1804 error_setg_errno(errp, ENOTSUP,
1805 "Differencing files not yet supported");
1806 ret = -ENOTSUP;
1807 goto exit;
1808 } else {
1809 ret = -EINVAL;
1810 goto exit;
1813 /* These are pretty arbitrary, and mainly designed to keep the BAT
1814 * size reasonable to load into RAM */
1815 if (block_size == 0) {
1816 if (image_size > 32 * TiB) {
1817 block_size = 64 * MiB;
1818 } else if (image_size > (uint64_t) 100 * GiB) {
1819 block_size = 32 * MiB;
1820 } else if (image_size > 1 * GiB) {
1821 block_size = 16 * MiB;
1822 } else {
1823 block_size = 8 * MiB;
1828 /* make the log size close to what was specified, but must be
1829 * min 1MB, and multiple of 1MB */
1830 log_size = ROUND_UP(log_size, MiB);
1832 block_size = ROUND_UP(block_size, MiB);
1833 block_size = block_size > VHDX_BLOCK_SIZE_MAX ? VHDX_BLOCK_SIZE_MAX :
1834 block_size;
1836 ret = bdrv_create_file(filename, opts, &local_err);
1837 if (ret < 0) {
1838 error_propagate(errp, local_err);
1839 goto exit;
1842 blk = blk_new_open(filename, NULL, NULL,
1843 BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
1844 if (blk == NULL) {
1845 error_propagate(errp, local_err);
1846 ret = -EIO;
1847 goto exit;
1850 blk_set_allow_write_beyond_eof(blk, true);
1852 /* Create (A) */
1854 /* The creator field is optional, but may be useful for
1855 * debugging / diagnostics */
1856 creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL,
1857 &creator_items, NULL);
1858 signature = cpu_to_le64(VHDX_FILE_SIGNATURE);
1859 ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature),
1861 if (ret < 0) {
1862 goto delete_and_exit;
1864 if (creator) {
1865 ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET + sizeof(signature),
1866 creator, creator_items * sizeof(gunichar2), 0);
1867 if (ret < 0) {
1868 goto delete_and_exit;
1873 /* Creates (B),(C) */
1874 ret = vhdx_create_new_headers(blk_bs(blk), image_size, log_size);
1875 if (ret < 0) {
1876 goto delete_and_exit;
1879 /* Creates (D),(E),(G) explicitly. (F) created as by-product */
1880 ret = vhdx_create_new_region_table(blk_bs(blk), image_size, block_size, 512,
1881 log_size, use_zero_blocks, image_type,
1882 &metadata_offset);
1883 if (ret < 0) {
1884 goto delete_and_exit;
1887 /* Creates (H) */
1888 ret = vhdx_create_new_metadata(blk_bs(blk), image_size, block_size, 512,
1889 metadata_offset, image_type);
1890 if (ret < 0) {
1891 goto delete_and_exit;
1895 delete_and_exit:
1896 blk_unref(blk);
1897 exit:
1898 g_free(type);
1899 g_free(creator);
1900 return ret;
1903 /* If opened r/w, the VHDX driver will automatically replay the log,
1904 * if one is present, inside the vhdx_open() call.
1906 * If qemu-img check -r all is called, the image is automatically opened
1907 * r/w and any log has already been replayed, so there is nothing (currently)
1908 * for us to do here
1910 static int vhdx_check(BlockDriverState *bs, BdrvCheckResult *result,
1911 BdrvCheckMode fix)
1913 BDRVVHDXState *s = bs->opaque;
1915 if (s->log_replayed_on_open) {
1916 result->corruptions_fixed++;
1918 return 0;
1921 static QemuOptsList vhdx_create_opts = {
1922 .name = "vhdx-create-opts",
1923 .head = QTAILQ_HEAD_INITIALIZER(vhdx_create_opts.head),
1924 .desc = {
1926 .name = BLOCK_OPT_SIZE,
1927 .type = QEMU_OPT_SIZE,
1928 .help = "Virtual disk size; max of 64TB."
1931 .name = VHDX_BLOCK_OPT_LOG_SIZE,
1932 .type = QEMU_OPT_SIZE,
1933 .def_value_str = stringify(DEFAULT_LOG_SIZE),
1934 .help = "Log size; min 1MB."
1937 .name = VHDX_BLOCK_OPT_BLOCK_SIZE,
1938 .type = QEMU_OPT_SIZE,
1939 .def_value_str = stringify(0),
1940 .help = "Block Size; min 1MB, max 256MB. " \
1941 "0 means auto-calculate based on image size."
1944 .name = BLOCK_OPT_SUBFMT,
1945 .type = QEMU_OPT_STRING,
1946 .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "\
1947 "Default is 'dynamic'."
1950 .name = VHDX_BLOCK_OPT_ZERO,
1951 .type = QEMU_OPT_BOOL,
1952 .help = "Force use of payload blocks of type 'ZERO'. "\
1953 "Non-standard, but default. Do not set to 'off' when "\
1954 "using 'qemu-img convert' with subformat=dynamic."
1956 { NULL }
1960 static BlockDriver bdrv_vhdx = {
1961 .format_name = "vhdx",
1962 .instance_size = sizeof(BDRVVHDXState),
1963 .bdrv_probe = vhdx_probe,
1964 .bdrv_open = vhdx_open,
1965 .bdrv_close = vhdx_close,
1966 .bdrv_reopen_prepare = vhdx_reopen_prepare,
1967 .bdrv_co_readv = vhdx_co_readv,
1968 .bdrv_co_writev = vhdx_co_writev,
1969 .bdrv_create = vhdx_create,
1970 .bdrv_get_info = vhdx_get_info,
1971 .bdrv_check = vhdx_check,
1972 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1974 .create_opts = &vhdx_create_opts,
1977 static void bdrv_vhdx_init(void)
1979 bdrv_register(&bdrv_vhdx);
1982 block_init(bdrv_vhdx_init);