replay: Set error location properly when parsing options
[qemu/ar7.git] / block / vhdx.c
blob72042e90820e23d308e9090093d92025d79da34e
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 "qemu-common.h"
20 #include "block/block_int.h"
21 #include "qemu/module.h"
22 #include "qemu/crc32c.h"
23 #include "block/vhdx.h"
24 #include "migration/migration.h"
26 #include <uuid/uuid.h>
27 #include <glib.h>
29 /* Options for VHDX creation */
31 #define VHDX_BLOCK_OPT_LOG_SIZE "log_size"
32 #define VHDX_BLOCK_OPT_BLOCK_SIZE "block_size"
33 #define VHDX_BLOCK_OPT_ZERO "block_state_zero"
35 typedef enum VHDXImageType {
36 VHDX_TYPE_DYNAMIC = 0,
37 VHDX_TYPE_FIXED,
38 VHDX_TYPE_DIFFERENCING, /* Currently unsupported */
39 } VHDXImageType;
41 /* Several metadata and region table data entries are identified by
42 * guids in a MS-specific GUID format. */
45 /* ------- Known Region Table GUIDs ---------------------- */
46 static const MSGUID bat_guid = { .data1 = 0x2dc27766,
47 .data2 = 0xf623,
48 .data3 = 0x4200,
49 .data4 = { 0x9d, 0x64, 0x11, 0x5e,
50 0x9b, 0xfd, 0x4a, 0x08} };
52 static const MSGUID metadata_guid = { .data1 = 0x8b7ca206,
53 .data2 = 0x4790,
54 .data3 = 0x4b9a,
55 .data4 = { 0xb8, 0xfe, 0x57, 0x5f,
56 0x05, 0x0f, 0x88, 0x6e} };
60 /* ------- Known Metadata Entry GUIDs ---------------------- */
61 static const MSGUID file_param_guid = { .data1 = 0xcaa16737,
62 .data2 = 0xfa36,
63 .data3 = 0x4d43,
64 .data4 = { 0xb3, 0xb6, 0x33, 0xf0,
65 0xaa, 0x44, 0xe7, 0x6b} };
67 static const MSGUID virtual_size_guid = { .data1 = 0x2FA54224,
68 .data2 = 0xcd1b,
69 .data3 = 0x4876,
70 .data4 = { 0xb2, 0x11, 0x5d, 0xbe,
71 0xd8, 0x3b, 0xf4, 0xb8} };
73 static const MSGUID page83_guid = { .data1 = 0xbeca12ab,
74 .data2 = 0xb2e6,
75 .data3 = 0x4523,
76 .data4 = { 0x93, 0xef, 0xc3, 0x09,
77 0xe0, 0x00, 0xc7, 0x46} };
80 static const MSGUID phys_sector_guid = { .data1 = 0xcda348c7,
81 .data2 = 0x445d,
82 .data3 = 0x4471,
83 .data4 = { 0x9c, 0xc9, 0xe9, 0x88,
84 0x52, 0x51, 0xc5, 0x56} };
86 static const MSGUID parent_locator_guid = { .data1 = 0xa8d35f2d,
87 .data2 = 0xb30b,
88 .data3 = 0x454d,
89 .data4 = { 0xab, 0xf7, 0xd3,
90 0xd8, 0x48, 0x34,
91 0xab, 0x0c} };
93 static const MSGUID logical_sector_guid = { .data1 = 0x8141bf1d,
94 .data2 = 0xa96f,
95 .data3 = 0x4709,
96 .data4 = { 0xba, 0x47, 0xf2,
97 0x33, 0xa8, 0xfa,
98 0xab, 0x5f} };
100 /* Each parent type must have a valid GUID; this is for parent images
101 * of type 'VHDX'. If we were to allow e.g. a QCOW2 parent, we would
102 * need to make up our own QCOW2 GUID type */
103 static const MSGUID parent_vhdx_guid __attribute__((unused))
104 = { .data1 = 0xb04aefb7,
105 .data2 = 0xd19e,
106 .data3 = 0x4a81,
107 .data4 = { 0xb7, 0x89, 0x25, 0xb8,
108 0xe9, 0x44, 0x59, 0x13} };
111 #define META_FILE_PARAMETER_PRESENT 0x01
112 #define META_VIRTUAL_DISK_SIZE_PRESENT 0x02
113 #define META_PAGE_83_PRESENT 0x04
114 #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08
115 #define META_PHYS_SECTOR_SIZE_PRESENT 0x10
116 #define META_PARENT_LOCATOR_PRESENT 0x20
118 #define META_ALL_PRESENT \
119 (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \
120 META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \
121 META_PHYS_SECTOR_SIZE_PRESENT)
124 typedef struct VHDXSectorInfo {
125 uint32_t bat_idx; /* BAT entry index */
126 uint32_t sectors_avail; /* sectors available in payload block */
127 uint32_t bytes_left; /* bytes left in the block after data to r/w */
128 uint32_t bytes_avail; /* bytes available in payload block */
129 uint64_t file_offset; /* absolute offset in bytes, in file */
130 uint64_t block_offset; /* block offset, in bytes */
131 } VHDXSectorInfo;
133 /* Calculates new checksum.
135 * Zero is substituted during crc calculation for the original crc field
136 * crc_offset: byte offset in buf of the buffer crc
137 * buf: buffer pointer
138 * size: size of buffer (must be > crc_offset+4)
140 * Note: The buffer should have all multi-byte data in little-endian format,
141 * and the resulting checksum is in little endian format.
143 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset)
145 uint32_t crc;
147 assert(buf != NULL);
148 assert(size > (crc_offset + sizeof(crc)));
150 memset(buf + crc_offset, 0, sizeof(crc));
151 crc = crc32c(0xffffffff, buf, size);
152 cpu_to_le32s(&crc);
153 memcpy(buf + crc_offset, &crc, sizeof(crc));
155 return crc;
158 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size,
159 int crc_offset)
161 uint32_t crc_new;
162 uint32_t crc_orig;
163 assert(buf != NULL);
165 if (crc_offset > 0) {
166 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
167 memset(buf + crc_offset, 0, sizeof(crc_orig));
170 crc_new = crc32c(crc, buf, size);
171 if (crc_offset > 0) {
172 memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig));
175 return crc_new;
178 /* Validates the checksum of the buffer, with an in-place CRC.
180 * Zero is substituted during crc calculation for the original crc field,
181 * and the crc field is restored afterwards. But the buffer will be modifed
182 * during the calculation, so this may not be not suitable for multi-threaded
183 * use.
185 * crc_offset: byte offset in buf of the buffer crc
186 * buf: buffer pointer
187 * size: size of buffer (must be > crc_offset+4)
189 * returns true if checksum is valid, false otherwise
191 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset)
193 uint32_t crc_orig;
194 uint32_t crc;
196 assert(buf != NULL);
197 assert(size > (crc_offset + 4));
199 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
200 crc_orig = le32_to_cpu(crc_orig);
202 crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset);
204 return crc == crc_orig;
209 * This generates a UUID that is compliant with the MS GUIDs used
210 * in the VHDX spec (and elsewhere).
212 void vhdx_guid_generate(MSGUID *guid)
214 uuid_t uuid;
215 assert(guid != NULL);
217 uuid_generate(uuid);
218 memcpy(guid, uuid, sizeof(MSGUID));
221 /* Check for region overlaps inside the VHDX image */
222 static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length)
224 int ret = 0;
225 uint64_t end;
226 VHDXRegionEntry *r;
228 end = start + length;
229 QLIST_FOREACH(r, &s->regions, entries) {
230 if (!((start >= r->end) || (end <= r->start))) {
231 ret = -EINVAL;
232 goto exit;
236 exit:
237 return ret;
240 /* Register a region for future checks */
241 static void vhdx_region_register(BDRVVHDXState *s,
242 uint64_t start, uint64_t length)
244 VHDXRegionEntry *r;
246 r = g_malloc0(sizeof(*r));
248 r->start = start;
249 r->end = start + length;
251 QLIST_INSERT_HEAD(&s->regions, r, entries);
254 /* Free all registered regions */
255 static void vhdx_region_unregister_all(BDRVVHDXState *s)
257 VHDXRegionEntry *r, *r_next;
259 QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) {
260 QLIST_REMOVE(r, entries);
261 g_free(r);
265 static void vhdx_set_shift_bits(BDRVVHDXState *s)
267 s->logical_sector_size_bits = 31 - clz32(s->logical_sector_size);
268 s->sectors_per_block_bits = 31 - clz32(s->sectors_per_block);
269 s->chunk_ratio_bits = 63 - clz64(s->chunk_ratio);
270 s->block_size_bits = 31 - clz32(s->block_size);
274 * Per the MS VHDX Specification, for every VHDX file:
275 * - The header section is fixed size - 1 MB
276 * - The header section is always the first "object"
277 * - The first 64KB of the header is the File Identifier
278 * - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile")
279 * - The following 512 bytes constitute a UTF-16 string identifiying the
280 * software that created the file, and is optional and diagnostic only.
282 * Therefore, we probe by looking for the vhdxfile signature "vhdxfile"
284 static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename)
286 if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) {
287 return 100;
289 return 0;
293 * Writes the header to the specified offset.
295 * This will optionally read in buffer data from disk (otherwise zero-fill),
296 * and then update the header checksum. Header is converted to proper
297 * endianness before being written to the specified file offset
299 static int vhdx_write_header(BlockDriverState *bs_file, VHDXHeader *hdr,
300 uint64_t offset, bool read)
302 uint8_t *buffer = NULL;
303 int ret;
304 VHDXHeader *header_le;
306 assert(bs_file != NULL);
307 assert(hdr != NULL);
309 /* the header checksum is not over just the packed size of VHDXHeader,
310 * but rather over the entire 'reserved' range for the header, which is
311 * 4KB (VHDX_HEADER_SIZE). */
313 buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE);
314 if (read) {
315 /* if true, we can't assume the extra reserved bytes are 0 */
316 ret = bdrv_pread(bs_file, offset, buffer, VHDX_HEADER_SIZE);
317 if (ret < 0) {
318 goto exit;
320 } else {
321 memset(buffer, 0, VHDX_HEADER_SIZE);
324 /* overwrite the actual VHDXHeader portion */
325 header_le = (VHDXHeader *)buffer;
326 memcpy(header_le, hdr, sizeof(VHDXHeader));
327 vhdx_header_le_export(hdr, header_le);
328 vhdx_update_checksum(buffer, VHDX_HEADER_SIZE,
329 offsetof(VHDXHeader, checksum));
330 ret = bdrv_pwrite_sync(bs_file, offset, header_le, sizeof(VHDXHeader));
332 exit:
333 qemu_vfree(buffer);
334 return ret;
337 /* Update the VHDX headers
339 * This follows the VHDX spec procedures for header updates.
341 * - non-current header is updated with largest sequence number
343 static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s,
344 bool generate_data_write_guid, MSGUID *log_guid)
346 int ret = 0;
347 int hdr_idx = 0;
348 uint64_t header_offset = VHDX_HEADER1_OFFSET;
350 VHDXHeader *active_header;
351 VHDXHeader *inactive_header;
353 /* operate on the non-current header */
354 if (s->curr_header == 0) {
355 hdr_idx = 1;
356 header_offset = VHDX_HEADER2_OFFSET;
359 active_header = s->headers[s->curr_header];
360 inactive_header = s->headers[hdr_idx];
362 inactive_header->sequence_number = active_header->sequence_number + 1;
364 /* a new file guid must be generated before any file write, including
365 * headers */
366 inactive_header->file_write_guid = s->session_guid;
368 /* a new data guid only needs to be generated before any guest-visible
369 * writes (i.e. something observable via virtual disk read) */
370 if (generate_data_write_guid) {
371 vhdx_guid_generate(&inactive_header->data_write_guid);
374 /* update the log guid if present */
375 if (log_guid) {
376 inactive_header->log_guid = *log_guid;
379 ret = vhdx_write_header(bs->file->bs, inactive_header, header_offset, true);
380 if (ret < 0) {
381 goto exit;
383 s->curr_header = hdr_idx;
385 exit:
386 return ret;
390 * The VHDX spec calls for header updates to be performed twice, so that both
391 * the current and non-current header have valid info
393 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s,
394 bool generate_data_write_guid, MSGUID *log_guid)
396 int ret;
398 ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
399 if (ret < 0) {
400 return ret;
402 ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
403 return ret;
406 /* opens the specified header block from the VHDX file header section */
407 static void vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s,
408 Error **errp)
410 int ret;
411 VHDXHeader *header1;
412 VHDXHeader *header2;
413 bool h1_valid = false;
414 bool h2_valid = false;
415 uint64_t h1_seq = 0;
416 uint64_t h2_seq = 0;
417 uint8_t *buffer;
419 /* header1 & header2 are freed in vhdx_close() */
420 header1 = qemu_blockalign(bs, sizeof(VHDXHeader));
421 header2 = qemu_blockalign(bs, sizeof(VHDXHeader));
423 buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE);
425 s->headers[0] = header1;
426 s->headers[1] = header2;
428 /* We have to read the whole VHDX_HEADER_SIZE instead of
429 * sizeof(VHDXHeader), because the checksum is over the whole
430 * region */
431 ret = bdrv_pread(bs->file->bs, VHDX_HEADER1_OFFSET, buffer,
432 VHDX_HEADER_SIZE);
433 if (ret < 0) {
434 goto fail;
436 /* copy over just the relevant portion that we need */
437 memcpy(header1, buffer, sizeof(VHDXHeader));
439 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
440 vhdx_header_le_import(header1);
441 if (header1->signature == VHDX_HEADER_SIGNATURE &&
442 header1->version == 1) {
443 h1_seq = header1->sequence_number;
444 h1_valid = true;
448 ret = bdrv_pread(bs->file->bs, VHDX_HEADER2_OFFSET, buffer,
449 VHDX_HEADER_SIZE);
450 if (ret < 0) {
451 goto fail;
453 /* copy over just the relevant portion that we need */
454 memcpy(header2, buffer, sizeof(VHDXHeader));
456 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
457 vhdx_header_le_import(header2);
458 if (header2->signature == VHDX_HEADER_SIGNATURE &&
459 header2->version == 1) {
460 h2_seq = header2->sequence_number;
461 h2_valid = true;
465 /* If there is only 1 valid header (or no valid headers), we
466 * don't care what the sequence numbers are */
467 if (h1_valid && !h2_valid) {
468 s->curr_header = 0;
469 } else if (!h1_valid && h2_valid) {
470 s->curr_header = 1;
471 } else if (!h1_valid && !h2_valid) {
472 goto fail;
473 } else {
474 /* If both headers are valid, then we choose the active one by the
475 * highest sequence number. If the sequence numbers are equal, that is
476 * invalid */
477 if (h1_seq > h2_seq) {
478 s->curr_header = 0;
479 } else if (h2_seq > h1_seq) {
480 s->curr_header = 1;
481 } else {
482 /* The Microsoft Disk2VHD tool will create 2 identical
483 * headers, with identical sequence numbers. If the headers are
484 * identical, don't consider the file corrupt */
485 if (!memcmp(header1, header2, sizeof(VHDXHeader))) {
486 s->curr_header = 0;
487 } else {
488 goto fail;
493 vhdx_region_register(s, s->headers[s->curr_header]->log_offset,
494 s->headers[s->curr_header]->log_length);
495 goto exit;
497 fail:
498 error_setg_errno(errp, -ret, "No valid VHDX header found");
499 qemu_vfree(header1);
500 qemu_vfree(header2);
501 s->headers[0] = NULL;
502 s->headers[1] = NULL;
503 exit:
504 qemu_vfree(buffer);
508 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s)
510 int ret = 0;
511 uint8_t *buffer;
512 int offset = 0;
513 VHDXRegionTableEntry rt_entry;
514 uint32_t i;
515 bool bat_rt_found = false;
516 bool metadata_rt_found = false;
518 /* We have to read the whole 64KB block, because the crc32 is over the
519 * whole block */
520 buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE);
522 ret = bdrv_pread(bs->file->bs, VHDX_REGION_TABLE_OFFSET, buffer,
523 VHDX_HEADER_BLOCK_SIZE);
524 if (ret < 0) {
525 goto fail;
527 memcpy(&s->rt, buffer, sizeof(s->rt));
528 offset += sizeof(s->rt);
530 if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4)) {
531 ret = -EINVAL;
532 goto fail;
535 vhdx_region_header_le_import(&s->rt);
537 if (s->rt.signature != VHDX_REGION_SIGNATURE) {
538 ret = -EINVAL;
539 goto fail;
543 /* Per spec, maximum region table entry count is 2047 */
544 if (s->rt.entry_count > 2047) {
545 ret = -EINVAL;
546 goto fail;
549 for (i = 0; i < s->rt.entry_count; i++) {
550 memcpy(&rt_entry, buffer + offset, sizeof(rt_entry));
551 offset += sizeof(rt_entry);
553 vhdx_region_entry_le_import(&rt_entry);
555 /* check for region overlap between these entries, and any
556 * other memory regions in the file */
557 ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length);
558 if (ret < 0) {
559 goto fail;
562 vhdx_region_register(s, rt_entry.file_offset, rt_entry.length);
564 /* see if we recognize the entry */
565 if (guid_eq(rt_entry.guid, bat_guid)) {
566 /* must be unique; if we have already found it this is invalid */
567 if (bat_rt_found) {
568 ret = -EINVAL;
569 goto fail;
571 bat_rt_found = true;
572 s->bat_rt = rt_entry;
573 continue;
576 if (guid_eq(rt_entry.guid, metadata_guid)) {
577 /* must be unique; if we have already found it this is invalid */
578 if (metadata_rt_found) {
579 ret = -EINVAL;
580 goto fail;
582 metadata_rt_found = true;
583 s->metadata_rt = rt_entry;
584 continue;
587 if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) {
588 /* cannot read vhdx file - required region table entry that
589 * we do not understand. per spec, we must fail to open */
590 ret = -ENOTSUP;
591 goto fail;
595 if (!bat_rt_found || !metadata_rt_found) {
596 ret = -EINVAL;
597 goto fail;
600 ret = 0;
602 fail:
603 qemu_vfree(buffer);
604 return ret;
609 /* Metadata initial parser
611 * This loads all the metadata entry fields. This may cause additional
612 * fields to be processed (e.g. parent locator, etc..).
614 * There are 5 Metadata items that are always required:
615 * - File Parameters (block size, has a parent)
616 * - Virtual Disk Size (size, in bytes, of the virtual drive)
617 * - Page 83 Data (scsi page 83 guid)
618 * - Logical Sector Size (logical sector size in bytes, either 512 or
619 * 4096. We only support 512 currently)
620 * - Physical Sector Size (512 or 4096)
622 * Also, if the File Parameters indicate this is a differencing file,
623 * we must also look for the Parent Locator metadata item.
625 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s)
627 int ret = 0;
628 uint8_t *buffer;
629 int offset = 0;
630 uint32_t i = 0;
631 VHDXMetadataTableEntry md_entry;
633 buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE);
635 ret = bdrv_pread(bs->file->bs, s->metadata_rt.file_offset, buffer,
636 VHDX_METADATA_TABLE_MAX_SIZE);
637 if (ret < 0) {
638 goto exit;
640 memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr));
641 offset += sizeof(s->metadata_hdr);
643 vhdx_metadata_header_le_import(&s->metadata_hdr);
645 if (s->metadata_hdr.signature != VHDX_METADATA_SIGNATURE) {
646 ret = -EINVAL;
647 goto exit;
650 s->metadata_entries.present = 0;
652 if ((s->metadata_hdr.entry_count * sizeof(md_entry)) >
653 (VHDX_METADATA_TABLE_MAX_SIZE - offset)) {
654 ret = -EINVAL;
655 goto exit;
658 for (i = 0; i < s->metadata_hdr.entry_count; i++) {
659 memcpy(&md_entry, buffer + offset, sizeof(md_entry));
660 offset += sizeof(md_entry);
662 vhdx_metadata_entry_le_import(&md_entry);
664 if (guid_eq(md_entry.item_id, file_param_guid)) {
665 if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) {
666 ret = -EINVAL;
667 goto exit;
669 s->metadata_entries.file_parameters_entry = md_entry;
670 s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT;
671 continue;
674 if (guid_eq(md_entry.item_id, virtual_size_guid)) {
675 if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) {
676 ret = -EINVAL;
677 goto exit;
679 s->metadata_entries.virtual_disk_size_entry = md_entry;
680 s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT;
681 continue;
684 if (guid_eq(md_entry.item_id, page83_guid)) {
685 if (s->metadata_entries.present & META_PAGE_83_PRESENT) {
686 ret = -EINVAL;
687 goto exit;
689 s->metadata_entries.page83_data_entry = md_entry;
690 s->metadata_entries.present |= META_PAGE_83_PRESENT;
691 continue;
694 if (guid_eq(md_entry.item_id, logical_sector_guid)) {
695 if (s->metadata_entries.present &
696 META_LOGICAL_SECTOR_SIZE_PRESENT) {
697 ret = -EINVAL;
698 goto exit;
700 s->metadata_entries.logical_sector_size_entry = md_entry;
701 s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT;
702 continue;
705 if (guid_eq(md_entry.item_id, phys_sector_guid)) {
706 if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) {
707 ret = -EINVAL;
708 goto exit;
710 s->metadata_entries.phys_sector_size_entry = md_entry;
711 s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT;
712 continue;
715 if (guid_eq(md_entry.item_id, parent_locator_guid)) {
716 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
717 ret = -EINVAL;
718 goto exit;
720 s->metadata_entries.parent_locator_entry = md_entry;
721 s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT;
722 continue;
725 if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) {
726 /* cannot read vhdx file - required region table entry that
727 * we do not understand. per spec, we must fail to open */
728 ret = -ENOTSUP;
729 goto exit;
733 if (s->metadata_entries.present != META_ALL_PRESENT) {
734 ret = -ENOTSUP;
735 goto exit;
738 ret = bdrv_pread(bs->file->bs,
739 s->metadata_entries.file_parameters_entry.offset
740 + s->metadata_rt.file_offset,
741 &s->params,
742 sizeof(s->params));
744 if (ret < 0) {
745 goto exit;
748 le32_to_cpus(&s->params.block_size);
749 le32_to_cpus(&s->params.data_bits);
752 /* We now have the file parameters, so we can tell if this is a
753 * differencing file (i.e.. has_parent), is dynamic or fixed
754 * sized (leave_blocks_allocated), and the block size */
756 /* The parent locator required iff the file parameters has_parent set */
757 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
758 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
759 /* TODO: parse parent locator fields */
760 ret = -ENOTSUP; /* temp, until differencing files are supported */
761 goto exit;
762 } else {
763 /* if has_parent is set, but there is not parent locator present,
764 * then that is an invalid combination */
765 ret = -EINVAL;
766 goto exit;
770 /* determine virtual disk size, logical sector size,
771 * and phys sector size */
773 ret = bdrv_pread(bs->file->bs,
774 s->metadata_entries.virtual_disk_size_entry.offset
775 + s->metadata_rt.file_offset,
776 &s->virtual_disk_size,
777 sizeof(uint64_t));
778 if (ret < 0) {
779 goto exit;
781 ret = bdrv_pread(bs->file->bs,
782 s->metadata_entries.logical_sector_size_entry.offset
783 + s->metadata_rt.file_offset,
784 &s->logical_sector_size,
785 sizeof(uint32_t));
786 if (ret < 0) {
787 goto exit;
789 ret = bdrv_pread(bs->file->bs,
790 s->metadata_entries.phys_sector_size_entry.offset
791 + s->metadata_rt.file_offset,
792 &s->physical_sector_size,
793 sizeof(uint32_t));
794 if (ret < 0) {
795 goto exit;
798 le64_to_cpus(&s->virtual_disk_size);
799 le32_to_cpus(&s->logical_sector_size);
800 le32_to_cpus(&s->physical_sector_size);
802 if (s->params.block_size < VHDX_BLOCK_SIZE_MIN ||
803 s->params.block_size > VHDX_BLOCK_SIZE_MAX) {
804 ret = -EINVAL;
805 goto exit;
808 /* only 2 supported sector sizes */
809 if (s->logical_sector_size != 512 && s->logical_sector_size != 4096) {
810 ret = -EINVAL;
811 goto exit;
814 /* Both block_size and sector_size are guaranteed powers of 2, below.
815 Due to range checks above, s->sectors_per_block can never be < 256 */
816 s->sectors_per_block = s->params.block_size / s->logical_sector_size;
817 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
818 (uint64_t)s->logical_sector_size /
819 (uint64_t)s->params.block_size;
821 /* These values are ones we will want to use for division / multiplication
822 * later on, and they are all guaranteed (per the spec) to be powers of 2,
823 * so we can take advantage of that for shift operations during
824 * reads/writes */
825 if (s->logical_sector_size & (s->logical_sector_size - 1)) {
826 ret = -EINVAL;
827 goto exit;
829 if (s->sectors_per_block & (s->sectors_per_block - 1)) {
830 ret = -EINVAL;
831 goto exit;
833 if (s->chunk_ratio & (s->chunk_ratio - 1)) {
834 ret = -EINVAL;
835 goto exit;
837 s->block_size = s->params.block_size;
838 if (s->block_size & (s->block_size - 1)) {
839 ret = -EINVAL;
840 goto exit;
843 vhdx_set_shift_bits(s);
845 ret = 0;
847 exit:
848 qemu_vfree(buffer);
849 return ret;
853 * Calculate the number of BAT entries, including sector
854 * bitmap entries.
856 static void vhdx_calc_bat_entries(BDRVVHDXState *s)
858 uint32_t data_blocks_cnt, bitmap_blocks_cnt;
860 data_blocks_cnt = s->virtual_disk_size >> s->block_size_bits;
861 if (s->virtual_disk_size - (data_blocks_cnt << s->block_size_bits)) {
862 data_blocks_cnt++;
864 bitmap_blocks_cnt = data_blocks_cnt >> s->chunk_ratio_bits;
865 if (data_blocks_cnt - (bitmap_blocks_cnt << s->chunk_ratio_bits)) {
866 bitmap_blocks_cnt++;
869 if (s->parent_entries) {
870 s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1);
871 } else {
872 s->bat_entries = data_blocks_cnt +
873 ((data_blocks_cnt - 1) >> s->chunk_ratio_bits);
878 static void vhdx_close(BlockDriverState *bs)
880 BDRVVHDXState *s = bs->opaque;
881 qemu_vfree(s->headers[0]);
882 s->headers[0] = NULL;
883 qemu_vfree(s->headers[1]);
884 s->headers[1] = NULL;
885 qemu_vfree(s->bat);
886 s->bat = NULL;
887 qemu_vfree(s->parent_entries);
888 s->parent_entries = NULL;
889 migrate_del_blocker(s->migration_blocker);
890 error_free(s->migration_blocker);
891 qemu_vfree(s->log.hdr);
892 s->log.hdr = NULL;
893 vhdx_region_unregister_all(s);
896 static int vhdx_open(BlockDriverState *bs, QDict *options, int flags,
897 Error **errp)
899 BDRVVHDXState *s = bs->opaque;
900 int ret = 0;
901 uint32_t i;
902 uint64_t signature;
903 Error *local_err = NULL;
905 s->bat = NULL;
906 s->first_visible_write = true;
908 qemu_co_mutex_init(&s->lock);
909 QLIST_INIT(&s->regions);
911 /* validate the file signature */
912 ret = bdrv_pread(bs->file->bs, 0, &signature, sizeof(uint64_t));
913 if (ret < 0) {
914 goto fail;
916 if (memcmp(&signature, "vhdxfile", 8)) {
917 ret = -EINVAL;
918 goto fail;
921 /* This is used for any header updates, for the file_write_guid.
922 * The spec dictates that a new value should be used for the first
923 * header update */
924 vhdx_guid_generate(&s->session_guid);
926 vhdx_parse_header(bs, s, &local_err);
927 if (local_err != NULL) {
928 error_propagate(errp, local_err);
929 ret = -EINVAL;
930 goto fail;
933 ret = vhdx_parse_log(bs, s, &s->log_replayed_on_open, errp);
934 if (ret < 0) {
935 goto fail;
938 ret = vhdx_open_region_tables(bs, s);
939 if (ret < 0) {
940 goto fail;
943 ret = vhdx_parse_metadata(bs, s);
944 if (ret < 0) {
945 goto fail;
948 s->block_size = s->params.block_size;
950 /* the VHDX spec dictates that virtual_disk_size is always a multiple of
951 * logical_sector_size */
952 bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits;
954 vhdx_calc_bat_entries(s);
956 s->bat_offset = s->bat_rt.file_offset;
958 if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) {
959 /* BAT allocation is not large enough for all entries */
960 ret = -EINVAL;
961 goto fail;
964 /* s->bat is freed in vhdx_close() */
965 s->bat = qemu_try_blockalign(bs->file->bs, s->bat_rt.length);
966 if (s->bat == NULL) {
967 ret = -ENOMEM;
968 goto fail;
971 ret = bdrv_pread(bs->file->bs, s->bat_offset, s->bat, s->bat_rt.length);
972 if (ret < 0) {
973 goto fail;
976 uint64_t payblocks = s->chunk_ratio;
977 /* endian convert, and verify populated BAT field file offsets against
978 * region table and log entries */
979 for (i = 0; i < s->bat_entries; i++) {
980 le64_to_cpus(&s->bat[i]);
981 if (payblocks--) {
982 /* payload bat entries */
983 if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) ==
984 PAYLOAD_BLOCK_FULLY_PRESENT) {
985 ret = vhdx_region_check(s, s->bat[i] & VHDX_BAT_FILE_OFF_MASK,
986 s->block_size);
987 if (ret < 0) {
988 goto fail;
991 } else {
992 payblocks = s->chunk_ratio;
993 /* Once differencing files are supported, verify sector bitmap
994 * blocks here */
998 if (flags & BDRV_O_RDWR) {
999 ret = vhdx_update_headers(bs, s, false, NULL);
1000 if (ret < 0) {
1001 goto fail;
1005 /* TODO: differencing files */
1007 /* Disable migration when VHDX images are used */
1008 error_setg(&s->migration_blocker, "The vhdx format used by node '%s' "
1009 "does not support live migration",
1010 bdrv_get_device_or_node_name(bs));
1011 migrate_add_blocker(s->migration_blocker);
1013 return 0;
1014 fail:
1015 vhdx_close(bs);
1016 return ret;
1019 static int vhdx_reopen_prepare(BDRVReopenState *state,
1020 BlockReopenQueue *queue, Error **errp)
1022 return 0;
1027 * Perform sector to block offset translations, to get various
1028 * sector and file offsets into the image. See VHDXSectorInfo
1030 static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num,
1031 int nb_sectors, VHDXSectorInfo *sinfo)
1033 uint32_t block_offset;
1035 sinfo->bat_idx = sector_num >> s->sectors_per_block_bits;
1036 /* effectively a modulo - this gives us the offset into the block
1037 * (in sector sizes) for our sector number */
1038 block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits);
1039 /* the chunk ratio gives us the interleaving of the sector
1040 * bitmaps, so we need to advance our page block index by the
1041 * sector bitmaps entry number */
1042 sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits;
1044 /* the number of sectors we can read/write in this cycle */
1045 sinfo->sectors_avail = s->sectors_per_block - block_offset;
1047 sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits;
1049 if (sinfo->sectors_avail > nb_sectors) {
1050 sinfo->sectors_avail = nb_sectors;
1053 sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits;
1055 sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK;
1057 sinfo->block_offset = block_offset << s->logical_sector_size_bits;
1059 /* The file offset must be past the header section, so must be > 0 */
1060 if (sinfo->file_offset == 0) {
1061 return;
1064 /* block offset is the offset in vhdx logical sectors, in
1065 * the payload data block. Convert that to a byte offset
1066 * in the block, and add in the payload data block offset
1067 * in the file, in bytes, to get the final read address */
1069 sinfo->file_offset += sinfo->block_offset;
1073 static int vhdx_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1075 BDRVVHDXState *s = bs->opaque;
1077 bdi->cluster_size = s->block_size;
1079 bdi->unallocated_blocks_are_zero =
1080 (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) == 0;
1082 return 0;
1086 static coroutine_fn int vhdx_co_readv(BlockDriverState *bs, int64_t sector_num,
1087 int nb_sectors, QEMUIOVector *qiov)
1089 BDRVVHDXState *s = bs->opaque;
1090 int ret = 0;
1091 VHDXSectorInfo sinfo;
1092 uint64_t bytes_done = 0;
1093 QEMUIOVector hd_qiov;
1095 qemu_iovec_init(&hd_qiov, qiov->niov);
1097 qemu_co_mutex_lock(&s->lock);
1099 while (nb_sectors > 0) {
1100 /* We are a differencing file, so we need to inspect the sector bitmap
1101 * to see if we have the data or not */
1102 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1103 /* not supported yet */
1104 ret = -ENOTSUP;
1105 goto exit;
1106 } else {
1107 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1109 qemu_iovec_reset(&hd_qiov);
1110 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, sinfo.bytes_avail);
1112 /* check the payload block state */
1113 switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) {
1114 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1115 case PAYLOAD_BLOCK_UNDEFINED:
1116 case PAYLOAD_BLOCK_UNMAPPED:
1117 case PAYLOAD_BLOCK_UNMAPPED_v095:
1118 case PAYLOAD_BLOCK_ZERO:
1119 /* return zero */
1120 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail);
1121 break;
1122 case PAYLOAD_BLOCK_FULLY_PRESENT:
1123 qemu_co_mutex_unlock(&s->lock);
1124 ret = bdrv_co_readv(bs->file->bs,
1125 sinfo.file_offset >> BDRV_SECTOR_BITS,
1126 sinfo.sectors_avail, &hd_qiov);
1127 qemu_co_mutex_lock(&s->lock);
1128 if (ret < 0) {
1129 goto exit;
1131 break;
1132 case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1133 /* we don't yet support difference files, fall through
1134 * to error */
1135 default:
1136 ret = -EIO;
1137 goto exit;
1138 break;
1140 nb_sectors -= sinfo.sectors_avail;
1141 sector_num += sinfo.sectors_avail;
1142 bytes_done += sinfo.bytes_avail;
1145 ret = 0;
1146 exit:
1147 qemu_co_mutex_unlock(&s->lock);
1148 qemu_iovec_destroy(&hd_qiov);
1149 return ret;
1153 * Allocate a new payload block at the end of the file.
1155 * Allocation will happen at 1MB alignment inside the file
1157 * Returns the file offset start of the new payload block
1159 static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s,
1160 uint64_t *new_offset)
1162 *new_offset = bdrv_getlength(bs->file->bs);
1164 /* per the spec, the address for a block is in units of 1MB */
1165 *new_offset = ROUND_UP(*new_offset, 1024 * 1024);
1167 return bdrv_truncate(bs->file->bs, *new_offset + s->block_size);
1171 * Update the BAT table entry with the new file offset, and the new entry
1172 * state */
1173 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s,
1174 VHDXSectorInfo *sinfo,
1175 uint64_t *bat_entry_le,
1176 uint64_t *bat_offset, int state)
1178 /* The BAT entry is a uint64, with 44 bits for the file offset in units of
1179 * 1MB, and 3 bits for the block state. */
1180 if ((state == PAYLOAD_BLOCK_ZERO) ||
1181 (state == PAYLOAD_BLOCK_UNDEFINED) ||
1182 (state == PAYLOAD_BLOCK_NOT_PRESENT) ||
1183 (state == PAYLOAD_BLOCK_UNMAPPED)) {
1184 s->bat[sinfo->bat_idx] = 0; /* For PAYLOAD_BLOCK_ZERO, the
1185 FileOffsetMB field is denoted as
1186 'reserved' in the v1.0 spec. If it is
1187 non-zero, MS Hyper-V will fail to read
1188 the disk image */
1189 } else {
1190 s->bat[sinfo->bat_idx] = sinfo->file_offset;
1193 s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK;
1195 *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]);
1196 *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry);
1200 /* Per the spec, on the first write of guest-visible data to the file the
1201 * data write guid must be updated in the header */
1202 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s)
1204 int ret = 0;
1205 if (s->first_visible_write) {
1206 s->first_visible_write = false;
1207 ret = vhdx_update_headers(bs, s, true, NULL);
1209 return ret;
1212 static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num,
1213 int nb_sectors, QEMUIOVector *qiov)
1215 int ret = -ENOTSUP;
1216 BDRVVHDXState *s = bs->opaque;
1217 VHDXSectorInfo sinfo;
1218 uint64_t bytes_done = 0;
1219 uint64_t bat_entry = 0;
1220 uint64_t bat_entry_offset = 0;
1221 QEMUIOVector hd_qiov;
1222 struct iovec iov1 = { 0 };
1223 struct iovec iov2 = { 0 };
1224 int sectors_to_write;
1225 int bat_state;
1226 uint64_t bat_prior_offset = 0;
1227 bool bat_update = false;
1229 qemu_iovec_init(&hd_qiov, qiov->niov);
1231 qemu_co_mutex_lock(&s->lock);
1233 ret = vhdx_user_visible_write(bs, s);
1234 if (ret < 0) {
1235 goto exit;
1238 while (nb_sectors > 0) {
1239 bool use_zero_buffers = false;
1240 bat_update = false;
1241 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1242 /* not supported yet */
1243 ret = -ENOTSUP;
1244 goto exit;
1245 } else {
1246 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1247 sectors_to_write = sinfo.sectors_avail;
1249 qemu_iovec_reset(&hd_qiov);
1250 /* check the payload block state */
1251 bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK;
1252 switch (bat_state) {
1253 case PAYLOAD_BLOCK_ZERO:
1254 /* in this case, we need to preserve zero writes for
1255 * data that is not part of this write, so we must pad
1256 * the rest of the buffer to zeroes */
1258 /* if we are on a posix system with ftruncate() that extends
1259 * a file, then it is zero-filled for us. On Win32, the raw
1260 * layer uses SetFilePointer and SetFileEnd, which does not
1261 * zero fill AFAIK */
1263 /* Queue another write of zero buffers if the underlying file
1264 * does not zero-fill on file extension */
1266 if (bdrv_has_zero_init(bs->file->bs) == 0) {
1267 use_zero_buffers = true;
1269 /* zero fill the front, if any */
1270 if (sinfo.block_offset) {
1271 iov1.iov_len = sinfo.block_offset;
1272 iov1.iov_base = qemu_blockalign(bs, iov1.iov_len);
1273 memset(iov1.iov_base, 0, iov1.iov_len);
1274 qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0,
1275 iov1.iov_len);
1276 sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS;
1279 /* our actual data */
1280 qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1281 sinfo.bytes_avail);
1283 /* zero fill the back, if any */
1284 if ((sinfo.bytes_avail - sinfo.block_offset) <
1285 s->block_size) {
1286 iov2.iov_len = s->block_size -
1287 (sinfo.bytes_avail + sinfo.block_offset);
1288 iov2.iov_base = qemu_blockalign(bs, iov2.iov_len);
1289 memset(iov2.iov_base, 0, iov2.iov_len);
1290 qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0,
1291 iov2.iov_len);
1292 sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS;
1295 /* fall through */
1296 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1297 case PAYLOAD_BLOCK_UNMAPPED:
1298 case PAYLOAD_BLOCK_UNMAPPED_v095:
1299 case PAYLOAD_BLOCK_UNDEFINED:
1300 bat_prior_offset = sinfo.file_offset;
1301 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset);
1302 if (ret < 0) {
1303 goto exit;
1305 /* once we support differencing files, this may also be
1306 * partially present */
1307 /* update block state to the newly specified state */
1308 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1309 &bat_entry_offset,
1310 PAYLOAD_BLOCK_FULLY_PRESENT);
1311 bat_update = true;
1312 /* since we just allocated a block, file_offset is the
1313 * beginning of the payload block. It needs to be the
1314 * write address, which includes the offset into the block */
1315 if (!use_zero_buffers) {
1316 sinfo.file_offset += sinfo.block_offset;
1318 /* fall through */
1319 case PAYLOAD_BLOCK_FULLY_PRESENT:
1320 /* if the file offset address is in the header zone,
1321 * there is a problem */
1322 if (sinfo.file_offset < (1024 * 1024)) {
1323 ret = -EFAULT;
1324 goto error_bat_restore;
1327 if (!use_zero_buffers) {
1328 qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1329 sinfo.bytes_avail);
1331 /* block exists, so we can just overwrite it */
1332 qemu_co_mutex_unlock(&s->lock);
1333 ret = bdrv_co_writev(bs->file->bs,
1334 sinfo.file_offset >> BDRV_SECTOR_BITS,
1335 sectors_to_write, &hd_qiov);
1336 qemu_co_mutex_lock(&s->lock);
1337 if (ret < 0) {
1338 goto error_bat_restore;
1340 break;
1341 case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1342 /* we don't yet support difference files, fall through
1343 * to error */
1344 default:
1345 ret = -EIO;
1346 goto exit;
1347 break;
1350 if (bat_update) {
1351 /* this will update the BAT entry into the log journal, and
1352 * then flush the log journal out to disk */
1353 ret = vhdx_log_write_and_flush(bs, s, &bat_entry,
1354 sizeof(VHDXBatEntry),
1355 bat_entry_offset);
1356 if (ret < 0) {
1357 goto exit;
1361 nb_sectors -= sinfo.sectors_avail;
1362 sector_num += sinfo.sectors_avail;
1363 bytes_done += sinfo.bytes_avail;
1368 goto exit;
1370 error_bat_restore:
1371 if (bat_update) {
1372 /* keep metadata in sync, and restore the bat entry state
1373 * if error. */
1374 sinfo.file_offset = bat_prior_offset;
1375 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1376 &bat_entry_offset, bat_state);
1378 exit:
1379 qemu_vfree(iov1.iov_base);
1380 qemu_vfree(iov2.iov_base);
1381 qemu_co_mutex_unlock(&s->lock);
1382 qemu_iovec_destroy(&hd_qiov);
1383 return ret;
1389 * Create VHDX Headers
1391 * There are 2 headers, and the highest sequence number will represent
1392 * the active header
1394 static int vhdx_create_new_headers(BlockDriverState *bs, uint64_t image_size,
1395 uint32_t log_size)
1397 int ret = 0;
1398 VHDXHeader *hdr = NULL;
1400 hdr = g_new0(VHDXHeader, 1);
1402 hdr->signature = VHDX_HEADER_SIGNATURE;
1403 hdr->sequence_number = g_random_int();
1404 hdr->log_version = 0;
1405 hdr->version = 1;
1406 hdr->log_length = log_size;
1407 hdr->log_offset = VHDX_HEADER_SECTION_END;
1408 vhdx_guid_generate(&hdr->file_write_guid);
1409 vhdx_guid_generate(&hdr->data_write_guid);
1411 ret = vhdx_write_header(bs, hdr, VHDX_HEADER1_OFFSET, false);
1412 if (ret < 0) {
1413 goto exit;
1415 hdr->sequence_number++;
1416 ret = vhdx_write_header(bs, hdr, VHDX_HEADER2_OFFSET, false);
1417 if (ret < 0) {
1418 goto exit;
1421 exit:
1422 g_free(hdr);
1423 return ret;
1426 #define VHDX_METADATA_ENTRY_BUFFER_SIZE \
1427 (sizeof(VHDXFileParameters) +\
1428 sizeof(VHDXVirtualDiskSize) +\
1429 sizeof(VHDXPage83Data) +\
1430 sizeof(VHDXVirtualDiskLogicalSectorSize) +\
1431 sizeof(VHDXVirtualDiskPhysicalSectorSize))
1434 * Create the Metadata entries.
1436 * For more details on the entries, see section 3.5 (pg 29) in the
1437 * VHDX 1.00 specification.
1439 * We support 5 metadata entries (all required by spec):
1440 * File Parameters,
1441 * Virtual Disk Size,
1442 * Page 83 Data,
1443 * Logical Sector Size,
1444 * Physical Sector Size
1446 * The first 64KB of the Metadata section is reserved for the metadata
1447 * header and entries; beyond that, the metadata items themselves reside.
1449 static int vhdx_create_new_metadata(BlockDriverState *bs,
1450 uint64_t image_size,
1451 uint32_t block_size,
1452 uint32_t sector_size,
1453 uint64_t metadata_offset,
1454 VHDXImageType type)
1456 int ret = 0;
1457 uint32_t offset = 0;
1458 void *buffer = NULL;
1459 void *entry_buffer;
1460 VHDXMetadataTableHeader *md_table;
1461 VHDXMetadataTableEntry *md_table_entry;
1463 /* Metadata entries */
1464 VHDXFileParameters *mt_file_params;
1465 VHDXVirtualDiskSize *mt_virtual_size;
1466 VHDXPage83Data *mt_page83;
1467 VHDXVirtualDiskLogicalSectorSize *mt_log_sector_size;
1468 VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size;
1470 entry_buffer = g_malloc0(VHDX_METADATA_ENTRY_BUFFER_SIZE);
1472 mt_file_params = entry_buffer;
1473 offset += sizeof(VHDXFileParameters);
1474 mt_virtual_size = entry_buffer + offset;
1475 offset += sizeof(VHDXVirtualDiskSize);
1476 mt_page83 = entry_buffer + offset;
1477 offset += sizeof(VHDXPage83Data);
1478 mt_log_sector_size = entry_buffer + offset;
1479 offset += sizeof(VHDXVirtualDiskLogicalSectorSize);
1480 mt_phys_sector_size = entry_buffer + offset;
1482 mt_file_params->block_size = cpu_to_le32(block_size);
1483 if (type == VHDX_TYPE_FIXED) {
1484 mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED;
1485 cpu_to_le32s(&mt_file_params->data_bits);
1488 vhdx_guid_generate(&mt_page83->page_83_data);
1489 cpu_to_leguids(&mt_page83->page_83_data);
1490 mt_virtual_size->virtual_disk_size = cpu_to_le64(image_size);
1491 mt_log_sector_size->logical_sector_size = cpu_to_le32(sector_size);
1492 mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size);
1494 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1495 md_table = buffer;
1497 md_table->signature = VHDX_METADATA_SIGNATURE;
1498 md_table->entry_count = 5;
1499 vhdx_metadata_header_le_export(md_table);
1502 /* This will reference beyond the reserved table portion */
1503 offset = 64 * KiB;
1505 md_table_entry = buffer + sizeof(VHDXMetadataTableHeader);
1507 md_table_entry[0].item_id = file_param_guid;
1508 md_table_entry[0].offset = offset;
1509 md_table_entry[0].length = sizeof(VHDXFileParameters);
1510 md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED;
1511 offset += md_table_entry[0].length;
1512 vhdx_metadata_entry_le_export(&md_table_entry[0]);
1514 md_table_entry[1].item_id = virtual_size_guid;
1515 md_table_entry[1].offset = offset;
1516 md_table_entry[1].length = sizeof(VHDXVirtualDiskSize);
1517 md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1518 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1519 offset += md_table_entry[1].length;
1520 vhdx_metadata_entry_le_export(&md_table_entry[1]);
1522 md_table_entry[2].item_id = page83_guid;
1523 md_table_entry[2].offset = offset;
1524 md_table_entry[2].length = sizeof(VHDXPage83Data);
1525 md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1526 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1527 offset += md_table_entry[2].length;
1528 vhdx_metadata_entry_le_export(&md_table_entry[2]);
1530 md_table_entry[3].item_id = logical_sector_guid;
1531 md_table_entry[3].offset = offset;
1532 md_table_entry[3].length = sizeof(VHDXVirtualDiskLogicalSectorSize);
1533 md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1534 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1535 offset += md_table_entry[3].length;
1536 vhdx_metadata_entry_le_export(&md_table_entry[3]);
1538 md_table_entry[4].item_id = phys_sector_guid;
1539 md_table_entry[4].offset = offset;
1540 md_table_entry[4].length = sizeof(VHDXVirtualDiskPhysicalSectorSize);
1541 md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1542 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1543 vhdx_metadata_entry_le_export(&md_table_entry[4]);
1545 ret = bdrv_pwrite(bs, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE);
1546 if (ret < 0) {
1547 goto exit;
1550 ret = bdrv_pwrite(bs, metadata_offset + (64 * KiB), entry_buffer,
1551 VHDX_METADATA_ENTRY_BUFFER_SIZE);
1552 if (ret < 0) {
1553 goto exit;
1557 exit:
1558 g_free(buffer);
1559 g_free(entry_buffer);
1560 return ret;
1563 /* This create the actual BAT itself. We currently only support
1564 * 'Dynamic' and 'Fixed' image types.
1566 * Dynamic images: default state of the BAT is all zeroes.
1568 * Fixed images: default state of the BAT is fully populated, with
1569 * file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT.
1571 static int vhdx_create_bat(BlockDriverState *bs, BDRVVHDXState *s,
1572 uint64_t image_size, VHDXImageType type,
1573 bool use_zero_blocks, uint64_t file_offset,
1574 uint32_t length)
1576 int ret = 0;
1577 uint64_t data_file_offset;
1578 uint64_t total_sectors = 0;
1579 uint64_t sector_num = 0;
1580 uint64_t unused;
1581 int block_state;
1582 VHDXSectorInfo sinfo;
1584 assert(s->bat == NULL);
1586 /* this gives a data start after BAT/bitmap entries, and well
1587 * past any metadata entries (with a 4 MB buffer for future
1588 * expansion */
1589 data_file_offset = file_offset + length + 5 * MiB;
1590 total_sectors = image_size >> s->logical_sector_size_bits;
1592 if (type == VHDX_TYPE_DYNAMIC) {
1593 /* All zeroes, so we can just extend the file - the end of the BAT
1594 * is the furthest thing we have written yet */
1595 ret = bdrv_truncate(bs, data_file_offset);
1596 if (ret < 0) {
1597 goto exit;
1599 } else if (type == VHDX_TYPE_FIXED) {
1600 ret = bdrv_truncate(bs, data_file_offset + image_size);
1601 if (ret < 0) {
1602 goto exit;
1604 } else {
1605 ret = -ENOTSUP;
1606 goto exit;
1609 if (type == VHDX_TYPE_FIXED ||
1610 use_zero_blocks ||
1611 bdrv_has_zero_init(bs) == 0) {
1612 /* for a fixed file, the default BAT entry is not zero */
1613 s->bat = g_try_malloc0(length);
1614 if (length && s->bat == NULL) {
1615 ret = -ENOMEM;
1616 goto exit;
1618 block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT :
1619 PAYLOAD_BLOCK_NOT_PRESENT;
1620 block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state;
1621 /* fill the BAT by emulating sector writes of sectors_per_block size */
1622 while (sector_num < total_sectors) {
1623 vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo);
1624 sinfo.file_offset = data_file_offset +
1625 (sector_num << s->logical_sector_size_bits);
1626 sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB);
1627 vhdx_update_bat_table_entry(bs, s, &sinfo, &unused, &unused,
1628 block_state);
1629 cpu_to_le64s(&s->bat[sinfo.bat_idx]);
1630 sector_num += s->sectors_per_block;
1632 ret = bdrv_pwrite(bs, file_offset, s->bat, length);
1633 if (ret < 0) {
1634 goto exit;
1640 exit:
1641 g_free(s->bat);
1642 return ret;
1645 /* Creates the region table header, and region table entries.
1646 * There are 2 supported region table entries: BAT, and Metadata/
1648 * As the calculations for the BAT region table are also needed
1649 * to create the BAT itself, we will also cause the BAT to be
1650 * created.
1652 static int vhdx_create_new_region_table(BlockDriverState *bs,
1653 uint64_t image_size,
1654 uint32_t block_size,
1655 uint32_t sector_size,
1656 uint32_t log_size,
1657 bool use_zero_blocks,
1658 VHDXImageType type,
1659 uint64_t *metadata_offset)
1661 int ret = 0;
1662 uint32_t offset = 0;
1663 void *buffer = NULL;
1664 uint64_t bat_file_offset;
1665 uint32_t bat_length;
1666 BDRVVHDXState *s = NULL;
1667 VHDXRegionTableHeader *region_table;
1668 VHDXRegionTableEntry *rt_bat;
1669 VHDXRegionTableEntry *rt_metadata;
1671 assert(metadata_offset != NULL);
1673 /* Populate enough of the BDRVVHDXState to be able to use the
1674 * pre-existing BAT calculation, translation, and update functions */
1675 s = g_new0(BDRVVHDXState, 1);
1677 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
1678 (uint64_t) sector_size / (uint64_t) block_size;
1680 s->sectors_per_block = block_size / sector_size;
1681 s->virtual_disk_size = image_size;
1682 s->block_size = block_size;
1683 s->logical_sector_size = sector_size;
1685 vhdx_set_shift_bits(s);
1687 vhdx_calc_bat_entries(s);
1689 /* At this point the VHDX state is populated enough for creation */
1691 /* a single buffer is used so we can calculate the checksum over the
1692 * entire 64KB block */
1693 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1694 region_table = buffer;
1695 offset += sizeof(VHDXRegionTableHeader);
1696 rt_bat = buffer + offset;
1697 offset += sizeof(VHDXRegionTableEntry);
1698 rt_metadata = buffer + offset;
1700 region_table->signature = VHDX_REGION_SIGNATURE;
1701 region_table->entry_count = 2; /* BAT and Metadata */
1703 rt_bat->guid = bat_guid;
1704 rt_bat->length = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB);
1705 rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB);
1706 s->bat_offset = rt_bat->file_offset;
1708 rt_metadata->guid = metadata_guid;
1709 rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length,
1710 MiB);
1711 rt_metadata->length = 1 * MiB; /* min size, and more than enough */
1712 *metadata_offset = rt_metadata->file_offset;
1714 bat_file_offset = rt_bat->file_offset;
1715 bat_length = rt_bat->length;
1717 vhdx_region_header_le_export(region_table);
1718 vhdx_region_entry_le_export(rt_bat);
1719 vhdx_region_entry_le_export(rt_metadata);
1721 vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE,
1722 offsetof(VHDXRegionTableHeader, checksum));
1725 /* The region table gives us the data we need to create the BAT,
1726 * so do that now */
1727 ret = vhdx_create_bat(bs, s, image_size, type, use_zero_blocks,
1728 bat_file_offset, bat_length);
1729 if (ret < 0) {
1730 goto exit;
1733 /* Now write out the region headers to disk */
1734 ret = bdrv_pwrite(bs, VHDX_REGION_TABLE_OFFSET, buffer,
1735 VHDX_HEADER_BLOCK_SIZE);
1736 if (ret < 0) {
1737 goto exit;
1740 ret = bdrv_pwrite(bs, VHDX_REGION_TABLE2_OFFSET, buffer,
1741 VHDX_HEADER_BLOCK_SIZE);
1742 if (ret < 0) {
1743 goto exit;
1746 exit:
1747 g_free(s);
1748 g_free(buffer);
1749 return ret;
1752 /* We need to create the following elements:
1754 * .-----------------------------------------------------------------.
1755 * | (A) | (B) | (C) | (D) | (E) |
1756 * | File ID | Header1 | Header 2 | Region Tbl 1 | Region Tbl 2 |
1757 * | | | | | |
1758 * .-----------------------------------------------------------------.
1759 * 0 64KB 128KB 192KB 256KB 320KB
1762 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1763 * | (F) | (G) | (H) | |
1764 * | Journal Log | BAT / Bitmap | Metadata | .... data ...... |
1765 * | | | | |
1766 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1767 * 1MB
1769 static int vhdx_create(const char *filename, QemuOpts *opts, Error **errp)
1771 int ret = 0;
1772 uint64_t image_size = (uint64_t) 2 * GiB;
1773 uint32_t log_size = 1 * MiB;
1774 uint32_t block_size = 0;
1775 uint64_t signature;
1776 uint64_t metadata_offset;
1777 bool use_zero_blocks = false;
1779 gunichar2 *creator = NULL;
1780 glong creator_items;
1781 BlockDriverState *bs;
1782 char *type = NULL;
1783 VHDXImageType image_type;
1784 Error *local_err = NULL;
1786 image_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1787 BDRV_SECTOR_SIZE);
1788 log_size = qemu_opt_get_size_del(opts, VHDX_BLOCK_OPT_LOG_SIZE, 0);
1789 block_size = qemu_opt_get_size_del(opts, VHDX_BLOCK_OPT_BLOCK_SIZE, 0);
1790 type = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1791 use_zero_blocks = qemu_opt_get_bool_del(opts, VHDX_BLOCK_OPT_ZERO, true);
1793 if (image_size > VHDX_MAX_IMAGE_SIZE) {
1794 error_setg_errno(errp, EINVAL, "Image size too large; max of 64TB");
1795 ret = -EINVAL;
1796 goto exit;
1799 if (type == NULL) {
1800 type = g_strdup("dynamic");
1803 if (!strcmp(type, "dynamic")) {
1804 image_type = VHDX_TYPE_DYNAMIC;
1805 } else if (!strcmp(type, "fixed")) {
1806 image_type = VHDX_TYPE_FIXED;
1807 } else if (!strcmp(type, "differencing")) {
1808 error_setg_errno(errp, ENOTSUP,
1809 "Differencing files not yet supported");
1810 ret = -ENOTSUP;
1811 goto exit;
1812 } else {
1813 ret = -EINVAL;
1814 goto exit;
1817 /* These are pretty arbitrary, and mainly designed to keep the BAT
1818 * size reasonable to load into RAM */
1819 if (block_size == 0) {
1820 if (image_size > 32 * TiB) {
1821 block_size = 64 * MiB;
1822 } else if (image_size > (uint64_t) 100 * GiB) {
1823 block_size = 32 * MiB;
1824 } else if (image_size > 1 * GiB) {
1825 block_size = 16 * MiB;
1826 } else {
1827 block_size = 8 * MiB;
1832 /* make the log size close to what was specified, but must be
1833 * min 1MB, and multiple of 1MB */
1834 log_size = ROUND_UP(log_size, MiB);
1836 block_size = ROUND_UP(block_size, MiB);
1837 block_size = block_size > VHDX_BLOCK_SIZE_MAX ? VHDX_BLOCK_SIZE_MAX :
1838 block_size;
1840 ret = bdrv_create_file(filename, opts, &local_err);
1841 if (ret < 0) {
1842 error_propagate(errp, local_err);
1843 goto exit;
1846 bs = NULL;
1847 ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1848 &local_err);
1849 if (ret < 0) {
1850 error_propagate(errp, local_err);
1851 goto exit;
1854 /* Create (A) */
1856 /* The creator field is optional, but may be useful for
1857 * debugging / diagnostics */
1858 creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL,
1859 &creator_items, NULL);
1860 signature = cpu_to_le64(VHDX_FILE_SIGNATURE);
1861 ret = bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature));
1862 if (ret < 0) {
1863 goto delete_and_exit;
1865 if (creator) {
1866 ret = bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET + sizeof(signature),
1867 creator, creator_items * sizeof(gunichar2));
1868 if (ret < 0) {
1869 goto delete_and_exit;
1874 /* Creates (B),(C) */
1875 ret = vhdx_create_new_headers(bs, image_size, log_size);
1876 if (ret < 0) {
1877 goto delete_and_exit;
1880 /* Creates (D),(E),(G) explicitly. (F) created as by-product */
1881 ret = vhdx_create_new_region_table(bs, image_size, block_size, 512,
1882 log_size, use_zero_blocks, image_type,
1883 &metadata_offset);
1884 if (ret < 0) {
1885 goto delete_and_exit;
1888 /* Creates (H) */
1889 ret = vhdx_create_new_metadata(bs, image_size, block_size, 512,
1890 metadata_offset, image_type);
1891 if (ret < 0) {
1892 goto delete_and_exit;
1896 delete_and_exit:
1897 bdrv_unref(bs);
1898 exit:
1899 g_free(type);
1900 g_free(creator);
1901 return ret;
1904 /* If opened r/w, the VHDX driver will automatically replay the log,
1905 * if one is present, inside the vhdx_open() call.
1907 * If qemu-img check -r all is called, the image is automatically opened
1908 * r/w and any log has already been replayed, so there is nothing (currently)
1909 * for us to do here
1911 static int vhdx_check(BlockDriverState *bs, BdrvCheckResult *result,
1912 BdrvCheckMode fix)
1914 BDRVVHDXState *s = bs->opaque;
1916 if (s->log_replayed_on_open) {
1917 result->corruptions_fixed++;
1919 return 0;
1922 static QemuOptsList vhdx_create_opts = {
1923 .name = "vhdx-create-opts",
1924 .head = QTAILQ_HEAD_INITIALIZER(vhdx_create_opts.head),
1925 .desc = {
1927 .name = BLOCK_OPT_SIZE,
1928 .type = QEMU_OPT_SIZE,
1929 .help = "Virtual disk size; max of 64TB."
1932 .name = VHDX_BLOCK_OPT_LOG_SIZE,
1933 .type = QEMU_OPT_SIZE,
1934 .def_value_str = stringify(DEFAULT_LOG_SIZE),
1935 .help = "Log size; min 1MB."
1938 .name = VHDX_BLOCK_OPT_BLOCK_SIZE,
1939 .type = QEMU_OPT_SIZE,
1940 .def_value_str = stringify(0),
1941 .help = "Block Size; min 1MB, max 256MB. " \
1942 "0 means auto-calculate based on image size."
1945 .name = BLOCK_OPT_SUBFMT,
1946 .type = QEMU_OPT_STRING,
1947 .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "\
1948 "Default is 'dynamic'."
1951 .name = VHDX_BLOCK_OPT_ZERO,
1952 .type = QEMU_OPT_BOOL,
1953 .help = "Force use of payload blocks of type 'ZERO'. "\
1954 "Non-standard, but default. Do not set to 'off' when "\
1955 "using 'qemu-img convert' with subformat=dynamic."
1957 { NULL }
1961 static BlockDriver bdrv_vhdx = {
1962 .format_name = "vhdx",
1963 .instance_size = sizeof(BDRVVHDXState),
1964 .bdrv_probe = vhdx_probe,
1965 .bdrv_open = vhdx_open,
1966 .bdrv_close = vhdx_close,
1967 .bdrv_reopen_prepare = vhdx_reopen_prepare,
1968 .bdrv_co_readv = vhdx_co_readv,
1969 .bdrv_co_writev = vhdx_co_writev,
1970 .bdrv_create = vhdx_create,
1971 .bdrv_get_info = vhdx_get_info,
1972 .bdrv_check = vhdx_check,
1973 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1975 .create_opts = &vhdx_create_opts,
1978 static void bdrv_vhdx_init(void)
1980 bdrv_register(&bdrv_vhdx);
1983 block_init(bdrv_vhdx_init);