qdict: Fix memory leak in qdict_do_flatten()
[qemu/cris-port.git] / block / vhdx.c
blob7d1af9663b32ec275f7ebd3e086e911b490c3298
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-common.h"
19 #include "block/block_int.h"
20 #include "qemu/module.h"
21 #include "qemu/crc32c.h"
22 #include "block/vhdx.h"
23 #include "migration/migration.h"
25 #include <uuid/uuid.h>
26 #include <glib.h>
28 /* Options for VHDX creation */
30 #define VHDX_BLOCK_OPT_LOG_SIZE "log_size"
31 #define VHDX_BLOCK_OPT_BLOCK_SIZE "block_size"
32 #define VHDX_BLOCK_OPT_ZERO "block_state_zero"
34 typedef enum VHDXImageType {
35 VHDX_TYPE_DYNAMIC = 0,
36 VHDX_TYPE_FIXED,
37 VHDX_TYPE_DIFFERENCING, /* Currently unsupported */
38 } VHDXImageType;
40 /* Several metadata and region table data entries are identified by
41 * guids in a MS-specific GUID format. */
44 /* ------- Known Region Table GUIDs ---------------------- */
45 static const MSGUID bat_guid = { .data1 = 0x2dc27766,
46 .data2 = 0xf623,
47 .data3 = 0x4200,
48 .data4 = { 0x9d, 0x64, 0x11, 0x5e,
49 0x9b, 0xfd, 0x4a, 0x08} };
51 static const MSGUID metadata_guid = { .data1 = 0x8b7ca206,
52 .data2 = 0x4790,
53 .data3 = 0x4b9a,
54 .data4 = { 0xb8, 0xfe, 0x57, 0x5f,
55 0x05, 0x0f, 0x88, 0x6e} };
59 /* ------- Known Metadata Entry GUIDs ---------------------- */
60 static const MSGUID file_param_guid = { .data1 = 0xcaa16737,
61 .data2 = 0xfa36,
62 .data3 = 0x4d43,
63 .data4 = { 0xb3, 0xb6, 0x33, 0xf0,
64 0xaa, 0x44, 0xe7, 0x6b} };
66 static const MSGUID virtual_size_guid = { .data1 = 0x2FA54224,
67 .data2 = 0xcd1b,
68 .data3 = 0x4876,
69 .data4 = { 0xb2, 0x11, 0x5d, 0xbe,
70 0xd8, 0x3b, 0xf4, 0xb8} };
72 static const MSGUID page83_guid = { .data1 = 0xbeca12ab,
73 .data2 = 0xb2e6,
74 .data3 = 0x4523,
75 .data4 = { 0x93, 0xef, 0xc3, 0x09,
76 0xe0, 0x00, 0xc7, 0x46} };
79 static const MSGUID phys_sector_guid = { .data1 = 0xcda348c7,
80 .data2 = 0x445d,
81 .data3 = 0x4471,
82 .data4 = { 0x9c, 0xc9, 0xe9, 0x88,
83 0x52, 0x51, 0xc5, 0x56} };
85 static const MSGUID parent_locator_guid = { .data1 = 0xa8d35f2d,
86 .data2 = 0xb30b,
87 .data3 = 0x454d,
88 .data4 = { 0xab, 0xf7, 0xd3,
89 0xd8, 0x48, 0x34,
90 0xab, 0x0c} };
92 static const MSGUID logical_sector_guid = { .data1 = 0x8141bf1d,
93 .data2 = 0xa96f,
94 .data3 = 0x4709,
95 .data4 = { 0xba, 0x47, 0xf2,
96 0x33, 0xa8, 0xfa,
97 0xab, 0x5f} };
99 /* Each parent type must have a valid GUID; this is for parent images
100 * of type 'VHDX'. If we were to allow e.g. a QCOW2 parent, we would
101 * need to make up our own QCOW2 GUID type */
102 static const MSGUID parent_vhdx_guid = { .data1 = 0xb04aefb7,
103 .data2 = 0xd19e,
104 .data3 = 0x4a81,
105 .data4 = { 0xb7, 0x89, 0x25, 0xb8,
106 0xe9, 0x44, 0x59, 0x13} };
109 #define META_FILE_PARAMETER_PRESENT 0x01
110 #define META_VIRTUAL_DISK_SIZE_PRESENT 0x02
111 #define META_PAGE_83_PRESENT 0x04
112 #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08
113 #define META_PHYS_SECTOR_SIZE_PRESENT 0x10
114 #define META_PARENT_LOCATOR_PRESENT 0x20
116 #define META_ALL_PRESENT \
117 (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \
118 META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \
119 META_PHYS_SECTOR_SIZE_PRESENT)
122 typedef struct VHDXSectorInfo {
123 uint32_t bat_idx; /* BAT entry index */
124 uint32_t sectors_avail; /* sectors available in payload block */
125 uint32_t bytes_left; /* bytes left in the block after data to r/w */
126 uint32_t bytes_avail; /* bytes available in payload block */
127 uint64_t file_offset; /* absolute offset in bytes, in file */
128 uint64_t block_offset; /* block offset, in bytes */
129 } VHDXSectorInfo;
131 /* Calculates new checksum.
133 * Zero is substituted during crc calculation for the original crc field
134 * crc_offset: byte offset in buf of the buffer crc
135 * buf: buffer pointer
136 * size: size of buffer (must be > crc_offset+4)
138 * Note: The resulting checksum is in the CPU endianness, not necessarily
139 * in the file format endianness (LE). Any header export to disk should
140 * make sure that vhdx_header_le_export() is used to convert to the
141 * correct endianness
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 memcpy(buf + crc_offset, &crc, sizeof(crc));
154 return crc;
157 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size,
158 int crc_offset)
160 uint32_t crc_new;
161 uint32_t crc_orig;
162 assert(buf != NULL);
164 if (crc_offset > 0) {
165 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
166 memset(buf + crc_offset, 0, sizeof(crc_orig));
169 crc_new = crc32c(crc, buf, size);
170 if (crc_offset > 0) {
171 memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig));
174 return crc_new;
177 /* Validates the checksum of the buffer, with an in-place CRC.
179 * Zero is substituted during crc calculation for the original crc field,
180 * and the crc field is restored afterwards. But the buffer will be modifed
181 * during the calculation, so this may not be not suitable for multi-threaded
182 * use.
184 * crc_offset: byte offset in buf of the buffer crc
185 * buf: buffer pointer
186 * size: size of buffer (must be > crc_offset+4)
188 * returns true if checksum is valid, false otherwise
190 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset)
192 uint32_t crc_orig;
193 uint32_t crc;
195 assert(buf != NULL);
196 assert(size > (crc_offset + 4));
198 memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
199 crc_orig = le32_to_cpu(crc_orig);
201 crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset);
203 return crc == crc_orig;
208 * This generates a UUID that is compliant with the MS GUIDs used
209 * in the VHDX spec (and elsewhere).
211 void vhdx_guid_generate(MSGUID *guid)
213 uuid_t uuid;
214 assert(guid != NULL);
216 uuid_generate(uuid);
217 memcpy(guid, uuid, sizeof(MSGUID));
220 /* Check for region overlaps inside the VHDX image */
221 static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length)
223 int ret = 0;
224 uint64_t end;
225 VHDXRegionEntry *r;
227 end = start + length;
228 QLIST_FOREACH(r, &s->regions, entries) {
229 if (!((start >= r->end) || (end <= r->start))) {
230 ret = -EINVAL;
231 goto exit;
235 exit:
236 return ret;
239 /* Register a region for future checks */
240 static void vhdx_region_register(BDRVVHDXState *s,
241 uint64_t start, uint64_t length)
243 VHDXRegionEntry *r;
245 r = g_malloc0(sizeof(*r));
247 r->start = start;
248 r->end = start + length;
250 QLIST_INSERT_HEAD(&s->regions, r, entries);
253 /* Free all registered regions */
254 static void vhdx_region_unregister_all(BDRVVHDXState *s)
256 VHDXRegionEntry *r, *r_next;
258 QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) {
259 QLIST_REMOVE(r, entries);
260 g_free(r);
264 static void vhdx_set_shift_bits(BDRVVHDXState *s)
266 s->logical_sector_size_bits = 31 - clz32(s->logical_sector_size);
267 s->sectors_per_block_bits = 31 - clz32(s->sectors_per_block);
268 s->chunk_ratio_bits = 63 - clz64(s->chunk_ratio);
269 s->block_size_bits = 31 - clz32(s->block_size);
273 * Per the MS VHDX Specification, for every VHDX file:
274 * - The header section is fixed size - 1 MB
275 * - The header section is always the first "object"
276 * - The first 64KB of the header is the File Identifier
277 * - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile")
278 * - The following 512 bytes constitute a UTF-16 string identifiying the
279 * software that created the file, and is optional and diagnostic only.
281 * Therefore, we probe by looking for the vhdxfile signature "vhdxfile"
283 static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename)
285 if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) {
286 return 100;
288 return 0;
292 * Writes the header to the specified offset.
294 * This will optionally read in buffer data from disk (otherwise zero-fill),
295 * and then update the header checksum. Header is converted to proper
296 * endianness before being written to the specified file offset
298 static int vhdx_write_header(BlockDriverState *bs_file, VHDXHeader *hdr,
299 uint64_t offset, bool read)
301 uint8_t *buffer = NULL;
302 int ret;
303 VHDXHeader header_le;
305 assert(bs_file != NULL);
306 assert(hdr != NULL);
308 /* the header checksum is not over just the packed size of VHDXHeader,
309 * but rather over the entire 'reserved' range for the header, which is
310 * 4KB (VHDX_HEADER_SIZE). */
312 buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE);
313 if (read) {
314 /* if true, we can't assume the extra reserved bytes are 0 */
315 ret = bdrv_pread(bs_file, offset, buffer, VHDX_HEADER_SIZE);
316 if (ret < 0) {
317 goto exit;
319 } else {
320 memset(buffer, 0, VHDX_HEADER_SIZE);
323 /* overwrite the actual VHDXHeader portion */
324 memcpy(buffer, hdr, sizeof(VHDXHeader));
325 hdr->checksum = vhdx_update_checksum(buffer, VHDX_HEADER_SIZE,
326 offsetof(VHDXHeader, checksum));
327 vhdx_header_le_export(hdr, &header_le);
328 ret = bdrv_pwrite_sync(bs_file, offset, &header_le, sizeof(VHDXHeader));
330 exit:
331 qemu_vfree(buffer);
332 return ret;
335 /* Update the VHDX headers
337 * This follows the VHDX spec procedures for header updates.
339 * - non-current header is updated with largest sequence number
341 static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s,
342 bool generate_data_write_guid, MSGUID *log_guid)
344 int ret = 0;
345 int hdr_idx = 0;
346 uint64_t header_offset = VHDX_HEADER1_OFFSET;
348 VHDXHeader *active_header;
349 VHDXHeader *inactive_header;
351 /* operate on the non-current header */
352 if (s->curr_header == 0) {
353 hdr_idx = 1;
354 header_offset = VHDX_HEADER2_OFFSET;
357 active_header = s->headers[s->curr_header];
358 inactive_header = s->headers[hdr_idx];
360 inactive_header->sequence_number = active_header->sequence_number + 1;
362 /* a new file guid must be generated before any file write, including
363 * headers */
364 inactive_header->file_write_guid = s->session_guid;
366 /* a new data guid only needs to be generated before any guest-visible
367 * writes (i.e. something observable via virtual disk read) */
368 if (generate_data_write_guid) {
369 vhdx_guid_generate(&inactive_header->data_write_guid);
372 /* update the log guid if present */
373 if (log_guid) {
374 inactive_header->log_guid = *log_guid;
377 vhdx_write_header(bs->file, inactive_header, header_offset, true);
378 if (ret < 0) {
379 goto exit;
381 s->curr_header = hdr_idx;
383 exit:
384 return ret;
388 * The VHDX spec calls for header updates to be performed twice, so that both
389 * the current and non-current header have valid info
391 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s,
392 bool generate_data_write_guid, MSGUID *log_guid)
394 int ret;
396 ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
397 if (ret < 0) {
398 return ret;
400 ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
401 return ret;
404 /* opens the specified header block from the VHDX file header section */
405 static int vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s)
407 int ret = 0;
408 VHDXHeader *header1;
409 VHDXHeader *header2;
410 bool h1_valid = false;
411 bool h2_valid = false;
412 uint64_t h1_seq = 0;
413 uint64_t h2_seq = 0;
414 uint8_t *buffer;
416 /* header1 & header2 are freed in vhdx_close() */
417 header1 = qemu_blockalign(bs, sizeof(VHDXHeader));
418 header2 = qemu_blockalign(bs, sizeof(VHDXHeader));
420 buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE);
422 s->headers[0] = header1;
423 s->headers[1] = header2;
425 /* We have to read the whole VHDX_HEADER_SIZE instead of
426 * sizeof(VHDXHeader), because the checksum is over the whole
427 * region */
428 ret = bdrv_pread(bs->file, VHDX_HEADER1_OFFSET, buffer, VHDX_HEADER_SIZE);
429 if (ret < 0) {
430 goto fail;
432 /* copy over just the relevant portion that we need */
433 memcpy(header1, buffer, sizeof(VHDXHeader));
434 vhdx_header_le_import(header1);
436 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4) &&
437 !memcmp(&header1->signature, "head", 4) &&
438 header1->version == 1) {
439 h1_seq = header1->sequence_number;
440 h1_valid = true;
443 ret = bdrv_pread(bs->file, VHDX_HEADER2_OFFSET, buffer, VHDX_HEADER_SIZE);
444 if (ret < 0) {
445 goto fail;
447 /* copy over just the relevant portion that we need */
448 memcpy(header2, buffer, sizeof(VHDXHeader));
449 vhdx_header_le_import(header2);
451 if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4) &&
452 !memcmp(&header2->signature, "head", 4) &&
453 header2->version == 1) {
454 h2_seq = header2->sequence_number;
455 h2_valid = true;
458 /* If there is only 1 valid header (or no valid headers), we
459 * don't care what the sequence numbers are */
460 if (h1_valid && !h2_valid) {
461 s->curr_header = 0;
462 } else if (!h1_valid && h2_valid) {
463 s->curr_header = 1;
464 } else if (!h1_valid && !h2_valid) {
465 ret = -EINVAL;
466 goto fail;
467 } else {
468 /* If both headers are valid, then we choose the active one by the
469 * highest sequence number. If the sequence numbers are equal, that is
470 * invalid */
471 if (h1_seq > h2_seq) {
472 s->curr_header = 0;
473 } else if (h2_seq > h1_seq) {
474 s->curr_header = 1;
475 } else {
476 ret = -EINVAL;
477 goto fail;
481 vhdx_region_register(s, s->headers[s->curr_header]->log_offset,
482 s->headers[s->curr_header]->log_length);
484 ret = 0;
486 goto exit;
488 fail:
489 qerror_report(ERROR_CLASS_GENERIC_ERROR, "No valid VHDX header found");
490 qemu_vfree(header1);
491 qemu_vfree(header2);
492 s->headers[0] = NULL;
493 s->headers[1] = NULL;
494 exit:
495 qemu_vfree(buffer);
496 return ret;
500 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s)
502 int ret = 0;
503 uint8_t *buffer;
504 int offset = 0;
505 VHDXRegionTableEntry rt_entry;
506 uint32_t i;
507 bool bat_rt_found = false;
508 bool metadata_rt_found = false;
510 /* We have to read the whole 64KB block, because the crc32 is over the
511 * whole block */
512 buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE);
514 ret = bdrv_pread(bs->file, VHDX_REGION_TABLE_OFFSET, buffer,
515 VHDX_HEADER_BLOCK_SIZE);
516 if (ret < 0) {
517 goto fail;
519 memcpy(&s->rt, buffer, sizeof(s->rt));
520 vhdx_region_header_le_import(&s->rt);
521 offset += sizeof(s->rt);
523 if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4) ||
524 memcmp(&s->rt.signature, "regi", 4)) {
525 ret = -EINVAL;
526 goto fail;
529 /* Per spec, maximum region table entry count is 2047 */
530 if (s->rt.entry_count > 2047) {
531 ret = -EINVAL;
532 goto fail;
535 for (i = 0; i < s->rt.entry_count; i++) {
536 memcpy(&rt_entry, buffer + offset, sizeof(rt_entry));
537 offset += sizeof(rt_entry);
539 vhdx_region_entry_le_import(&rt_entry);
541 /* check for region overlap between these entries, and any
542 * other memory regions in the file */
543 ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length);
544 if (ret < 0) {
545 goto fail;
548 vhdx_region_register(s, rt_entry.file_offset, rt_entry.length);
550 /* see if we recognize the entry */
551 if (guid_eq(rt_entry.guid, bat_guid)) {
552 /* must be unique; if we have already found it this is invalid */
553 if (bat_rt_found) {
554 ret = -EINVAL;
555 goto fail;
557 bat_rt_found = true;
558 s->bat_rt = rt_entry;
559 continue;
562 if (guid_eq(rt_entry.guid, metadata_guid)) {
563 /* must be unique; if we have already found it this is invalid */
564 if (metadata_rt_found) {
565 ret = -EINVAL;
566 goto fail;
568 metadata_rt_found = true;
569 s->metadata_rt = rt_entry;
570 continue;
573 if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) {
574 /* cannot read vhdx file - required region table entry that
575 * we do not understand. per spec, we must fail to open */
576 ret = -ENOTSUP;
577 goto fail;
581 if (!bat_rt_found || !metadata_rt_found) {
582 ret = -EINVAL;
583 goto fail;
586 ret = 0;
588 fail:
589 qemu_vfree(buffer);
590 return ret;
595 /* Metadata initial parser
597 * This loads all the metadata entry fields. This may cause additional
598 * fields to be processed (e.g. parent locator, etc..).
600 * There are 5 Metadata items that are always required:
601 * - File Parameters (block size, has a parent)
602 * - Virtual Disk Size (size, in bytes, of the virtual drive)
603 * - Page 83 Data (scsi page 83 guid)
604 * - Logical Sector Size (logical sector size in bytes, either 512 or
605 * 4096. We only support 512 currently)
606 * - Physical Sector Size (512 or 4096)
608 * Also, if the File Parameters indicate this is a differencing file,
609 * we must also look for the Parent Locator metadata item.
611 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s)
613 int ret = 0;
614 uint8_t *buffer;
615 int offset = 0;
616 uint32_t i = 0;
617 VHDXMetadataTableEntry md_entry;
619 buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE);
621 ret = bdrv_pread(bs->file, s->metadata_rt.file_offset, buffer,
622 VHDX_METADATA_TABLE_MAX_SIZE);
623 if (ret < 0) {
624 goto exit;
626 memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr));
627 offset += sizeof(s->metadata_hdr);
629 vhdx_metadata_header_le_import(&s->metadata_hdr);
631 if (memcmp(&s->metadata_hdr.signature, "metadata", 8)) {
632 ret = -EINVAL;
633 goto exit;
636 s->metadata_entries.present = 0;
638 if ((s->metadata_hdr.entry_count * sizeof(md_entry)) >
639 (VHDX_METADATA_TABLE_MAX_SIZE - offset)) {
640 ret = -EINVAL;
641 goto exit;
644 for (i = 0; i < s->metadata_hdr.entry_count; i++) {
645 memcpy(&md_entry, buffer + offset, sizeof(md_entry));
646 offset += sizeof(md_entry);
648 vhdx_metadata_entry_le_import(&md_entry);
650 if (guid_eq(md_entry.item_id, file_param_guid)) {
651 if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) {
652 ret = -EINVAL;
653 goto exit;
655 s->metadata_entries.file_parameters_entry = md_entry;
656 s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT;
657 continue;
660 if (guid_eq(md_entry.item_id, virtual_size_guid)) {
661 if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) {
662 ret = -EINVAL;
663 goto exit;
665 s->metadata_entries.virtual_disk_size_entry = md_entry;
666 s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT;
667 continue;
670 if (guid_eq(md_entry.item_id, page83_guid)) {
671 if (s->metadata_entries.present & META_PAGE_83_PRESENT) {
672 ret = -EINVAL;
673 goto exit;
675 s->metadata_entries.page83_data_entry = md_entry;
676 s->metadata_entries.present |= META_PAGE_83_PRESENT;
677 continue;
680 if (guid_eq(md_entry.item_id, logical_sector_guid)) {
681 if (s->metadata_entries.present &
682 META_LOGICAL_SECTOR_SIZE_PRESENT) {
683 ret = -EINVAL;
684 goto exit;
686 s->metadata_entries.logical_sector_size_entry = md_entry;
687 s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT;
688 continue;
691 if (guid_eq(md_entry.item_id, phys_sector_guid)) {
692 if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) {
693 ret = -EINVAL;
694 goto exit;
696 s->metadata_entries.phys_sector_size_entry = md_entry;
697 s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT;
698 continue;
701 if (guid_eq(md_entry.item_id, parent_locator_guid)) {
702 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
703 ret = -EINVAL;
704 goto exit;
706 s->metadata_entries.parent_locator_entry = md_entry;
707 s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT;
708 continue;
711 if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) {
712 /* cannot read vhdx file - required region table entry that
713 * we do not understand. per spec, we must fail to open */
714 ret = -ENOTSUP;
715 goto exit;
719 if (s->metadata_entries.present != META_ALL_PRESENT) {
720 ret = -ENOTSUP;
721 goto exit;
724 ret = bdrv_pread(bs->file,
725 s->metadata_entries.file_parameters_entry.offset
726 + s->metadata_rt.file_offset,
727 &s->params,
728 sizeof(s->params));
730 if (ret < 0) {
731 goto exit;
734 le32_to_cpus(&s->params.block_size);
735 le32_to_cpus(&s->params.data_bits);
738 /* We now have the file parameters, so we can tell if this is a
739 * differencing file (i.e.. has_parent), is dynamic or fixed
740 * sized (leave_blocks_allocated), and the block size */
742 /* The parent locator required iff the file parameters has_parent set */
743 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
744 if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
745 /* TODO: parse parent locator fields */
746 ret = -ENOTSUP; /* temp, until differencing files are supported */
747 goto exit;
748 } else {
749 /* if has_parent is set, but there is not parent locator present,
750 * then that is an invalid combination */
751 ret = -EINVAL;
752 goto exit;
756 /* determine virtual disk size, logical sector size,
757 * and phys sector size */
759 ret = bdrv_pread(bs->file,
760 s->metadata_entries.virtual_disk_size_entry.offset
761 + s->metadata_rt.file_offset,
762 &s->virtual_disk_size,
763 sizeof(uint64_t));
764 if (ret < 0) {
765 goto exit;
767 ret = bdrv_pread(bs->file,
768 s->metadata_entries.logical_sector_size_entry.offset
769 + s->metadata_rt.file_offset,
770 &s->logical_sector_size,
771 sizeof(uint32_t));
772 if (ret < 0) {
773 goto exit;
775 ret = bdrv_pread(bs->file,
776 s->metadata_entries.phys_sector_size_entry.offset
777 + s->metadata_rt.file_offset,
778 &s->physical_sector_size,
779 sizeof(uint32_t));
780 if (ret < 0) {
781 goto exit;
784 le64_to_cpus(&s->virtual_disk_size);
785 le32_to_cpus(&s->logical_sector_size);
786 le32_to_cpus(&s->physical_sector_size);
788 if (s->logical_sector_size == 0 || s->params.block_size == 0) {
789 ret = -EINVAL;
790 goto exit;
793 /* both block_size and sector_size are guaranteed powers of 2 */
794 s->sectors_per_block = s->params.block_size / s->logical_sector_size;
795 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
796 (uint64_t)s->logical_sector_size /
797 (uint64_t)s->params.block_size;
799 /* These values are ones we will want to use for division / multiplication
800 * later on, and they are all guaranteed (per the spec) to be powers of 2,
801 * so we can take advantage of that for shift operations during
802 * reads/writes */
803 if (s->logical_sector_size & (s->logical_sector_size - 1)) {
804 ret = -EINVAL;
805 goto exit;
807 if (s->sectors_per_block & (s->sectors_per_block - 1)) {
808 ret = -EINVAL;
809 goto exit;
811 if (s->chunk_ratio & (s->chunk_ratio - 1)) {
812 ret = -EINVAL;
813 goto exit;
815 s->block_size = s->params.block_size;
816 if (s->block_size & (s->block_size - 1)) {
817 ret = -EINVAL;
818 goto exit;
821 vhdx_set_shift_bits(s);
823 ret = 0;
825 exit:
826 qemu_vfree(buffer);
827 return ret;
831 * Calculate the number of BAT entries, including sector
832 * bitmap entries.
834 static void vhdx_calc_bat_entries(BDRVVHDXState *s)
836 uint32_t data_blocks_cnt, bitmap_blocks_cnt;
838 data_blocks_cnt = s->virtual_disk_size >> s->block_size_bits;
839 if (s->virtual_disk_size - (data_blocks_cnt << s->block_size_bits)) {
840 data_blocks_cnt++;
842 bitmap_blocks_cnt = data_blocks_cnt >> s->chunk_ratio_bits;
843 if (data_blocks_cnt - (bitmap_blocks_cnt << s->chunk_ratio_bits)) {
844 bitmap_blocks_cnt++;
847 if (s->parent_entries) {
848 s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1);
849 } else {
850 s->bat_entries = data_blocks_cnt +
851 ((data_blocks_cnt - 1) >> s->chunk_ratio_bits);
856 static void vhdx_close(BlockDriverState *bs)
858 BDRVVHDXState *s = bs->opaque;
859 qemu_vfree(s->headers[0]);
860 s->headers[0] = NULL;
861 qemu_vfree(s->headers[1]);
862 s->headers[1] = NULL;
863 qemu_vfree(s->bat);
864 s->bat = NULL;
865 qemu_vfree(s->parent_entries);
866 s->parent_entries = NULL;
867 migrate_del_blocker(s->migration_blocker);
868 error_free(s->migration_blocker);
869 qemu_vfree(s->log.hdr);
870 s->log.hdr = NULL;
871 vhdx_region_unregister_all(s);
874 static int vhdx_open(BlockDriverState *bs, QDict *options, int flags,
875 Error **errp)
877 BDRVVHDXState *s = bs->opaque;
878 int ret = 0;
879 uint32_t i;
880 uint64_t signature;
881 bool log_flushed = false;
884 s->bat = NULL;
885 s->first_visible_write = true;
887 qemu_co_mutex_init(&s->lock);
888 QLIST_INIT(&s->regions);
890 /* validate the file signature */
891 ret = bdrv_pread(bs->file, 0, &signature, sizeof(uint64_t));
892 if (ret < 0) {
893 goto fail;
895 if (memcmp(&signature, "vhdxfile", 8)) {
896 ret = -EINVAL;
897 goto fail;
900 /* This is used for any header updates, for the file_write_guid.
901 * The spec dictates that a new value should be used for the first
902 * header update */
903 vhdx_guid_generate(&s->session_guid);
905 ret = vhdx_parse_header(bs, s);
906 if (ret < 0) {
907 goto fail;
910 ret = vhdx_parse_log(bs, s, &log_flushed);
911 if (ret < 0) {
912 goto fail;
915 ret = vhdx_open_region_tables(bs, s);
916 if (ret < 0) {
917 goto fail;
920 ret = vhdx_parse_metadata(bs, s);
921 if (ret < 0) {
922 goto fail;
925 s->block_size = s->params.block_size;
927 /* the VHDX spec dictates that virtual_disk_size is always a multiple of
928 * logical_sector_size */
929 bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits;
931 vhdx_calc_bat_entries(s);
933 s->bat_offset = s->bat_rt.file_offset;
935 if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) {
936 /* BAT allocation is not large enough for all entries */
937 ret = -EINVAL;
938 goto fail;
941 /* s->bat is freed in vhdx_close() */
942 s->bat = qemu_blockalign(bs, s->bat_rt.length);
944 ret = bdrv_pread(bs->file, s->bat_offset, s->bat, s->bat_rt.length);
945 if (ret < 0) {
946 goto fail;
949 uint64_t payblocks = s->chunk_ratio;
950 /* endian convert, and verify populated BAT field file offsets against
951 * region table and log entries */
952 for (i = 0; i < s->bat_entries; i++) {
953 le64_to_cpus(&s->bat[i]);
954 if (payblocks--) {
955 /* payload bat entries */
956 if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) ==
957 PAYLOAD_BLOCK_FULLY_PRESENT) {
958 ret = vhdx_region_check(s, s->bat[i] & VHDX_BAT_FILE_OFF_MASK,
959 s->block_size);
960 if (ret < 0) {
961 goto fail;
964 } else {
965 payblocks = s->chunk_ratio;
966 /* Once differencing files are supported, verify sector bitmap
967 * blocks here */
971 if (flags & BDRV_O_RDWR) {
972 ret = vhdx_update_headers(bs, s, false, NULL);
973 if (ret < 0) {
974 goto fail;
978 /* TODO: differencing files */
980 /* Disable migration when VHDX images are used */
981 error_set(&s->migration_blocker,
982 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
983 "vhdx", bs->device_name, "live migration");
984 migrate_add_blocker(s->migration_blocker);
986 return 0;
987 fail:
988 vhdx_close(bs);
989 return ret;
992 static int vhdx_reopen_prepare(BDRVReopenState *state,
993 BlockReopenQueue *queue, Error **errp)
995 return 0;
1000 * Perform sector to block offset translations, to get various
1001 * sector and file offsets into the image. See VHDXSectorInfo
1003 static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num,
1004 int nb_sectors, VHDXSectorInfo *sinfo)
1006 uint32_t block_offset;
1008 sinfo->bat_idx = sector_num >> s->sectors_per_block_bits;
1009 /* effectively a modulo - this gives us the offset into the block
1010 * (in sector sizes) for our sector number */
1011 block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits);
1012 /* the chunk ratio gives us the interleaving of the sector
1013 * bitmaps, so we need to advance our page block index by the
1014 * sector bitmaps entry number */
1015 sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits;
1017 /* the number of sectors we can read/write in this cycle */
1018 sinfo->sectors_avail = s->sectors_per_block - block_offset;
1020 sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits;
1022 if (sinfo->sectors_avail > nb_sectors) {
1023 sinfo->sectors_avail = nb_sectors;
1026 sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits;
1028 sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK;
1030 sinfo->block_offset = block_offset << s->logical_sector_size_bits;
1032 /* The file offset must be past the header section, so must be > 0 */
1033 if (sinfo->file_offset == 0) {
1034 return;
1037 /* block offset is the offset in vhdx logical sectors, in
1038 * the payload data block. Convert that to a byte offset
1039 * in the block, and add in the payload data block offset
1040 * in the file, in bytes, to get the final read address */
1042 sinfo->file_offset += sinfo->block_offset;
1047 static coroutine_fn int vhdx_co_readv(BlockDriverState *bs, int64_t sector_num,
1048 int nb_sectors, QEMUIOVector *qiov)
1050 BDRVVHDXState *s = bs->opaque;
1051 int ret = 0;
1052 VHDXSectorInfo sinfo;
1053 uint64_t bytes_done = 0;
1054 QEMUIOVector hd_qiov;
1056 qemu_iovec_init(&hd_qiov, qiov->niov);
1058 qemu_co_mutex_lock(&s->lock);
1060 while (nb_sectors > 0) {
1061 /* We are a differencing file, so we need to inspect the sector bitmap
1062 * to see if we have the data or not */
1063 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1064 /* not supported yet */
1065 ret = -ENOTSUP;
1066 goto exit;
1067 } else {
1068 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1070 qemu_iovec_reset(&hd_qiov);
1071 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, sinfo.bytes_avail);
1073 /* check the payload block state */
1074 switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) {
1075 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1076 case PAYLOAD_BLOCK_UNDEFINED: /* fall through */
1077 case PAYLOAD_BLOCK_UNMAPPED: /* fall through */
1078 case PAYLOAD_BLOCK_ZERO:
1079 /* return zero */
1080 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail);
1081 break;
1082 case PAYLOAD_BLOCK_FULLY_PRESENT:
1083 qemu_co_mutex_unlock(&s->lock);
1084 ret = bdrv_co_readv(bs->file,
1085 sinfo.file_offset >> BDRV_SECTOR_BITS,
1086 sinfo.sectors_avail, &hd_qiov);
1087 qemu_co_mutex_lock(&s->lock);
1088 if (ret < 0) {
1089 goto exit;
1091 break;
1092 case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1093 /* we don't yet support difference files, fall through
1094 * to error */
1095 default:
1096 ret = -EIO;
1097 goto exit;
1098 break;
1100 nb_sectors -= sinfo.sectors_avail;
1101 sector_num += sinfo.sectors_avail;
1102 bytes_done += sinfo.bytes_avail;
1105 ret = 0;
1106 exit:
1107 qemu_co_mutex_unlock(&s->lock);
1108 qemu_iovec_destroy(&hd_qiov);
1109 return ret;
1113 * Allocate a new payload block at the end of the file.
1115 * Allocation will happen at 1MB alignment inside the file
1117 * Returns the file offset start of the new payload block
1119 static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s,
1120 uint64_t *new_offset)
1122 *new_offset = bdrv_getlength(bs->file);
1124 /* per the spec, the address for a block is in units of 1MB */
1125 *new_offset = ROUND_UP(*new_offset, 1024 * 1024);
1127 return bdrv_truncate(bs->file, *new_offset + s->block_size);
1131 * Update the BAT table entry with the new file offset, and the new entry
1132 * state */
1133 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s,
1134 VHDXSectorInfo *sinfo,
1135 uint64_t *bat_entry_le,
1136 uint64_t *bat_offset, int state)
1138 /* The BAT entry is a uint64, with 44 bits for the file offset in units of
1139 * 1MB, and 3 bits for the block state. */
1140 s->bat[sinfo->bat_idx] = sinfo->file_offset;
1142 s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK;
1144 *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]);
1145 *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry);
1149 /* Per the spec, on the first write of guest-visible data to the file the
1150 * data write guid must be updated in the header */
1151 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s)
1153 int ret = 0;
1154 if (s->first_visible_write) {
1155 s->first_visible_write = false;
1156 ret = vhdx_update_headers(bs, s, true, NULL);
1158 return ret;
1161 static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num,
1162 int nb_sectors, QEMUIOVector *qiov)
1164 int ret = -ENOTSUP;
1165 BDRVVHDXState *s = bs->opaque;
1166 VHDXSectorInfo sinfo;
1167 uint64_t bytes_done = 0;
1168 uint64_t bat_entry = 0;
1169 uint64_t bat_entry_offset = 0;
1170 QEMUIOVector hd_qiov;
1171 struct iovec iov1 = { 0 };
1172 struct iovec iov2 = { 0 };
1173 int sectors_to_write;
1174 int bat_state;
1175 uint64_t bat_prior_offset = 0;
1176 bool bat_update = false;
1178 qemu_iovec_init(&hd_qiov, qiov->niov);
1180 qemu_co_mutex_lock(&s->lock);
1182 ret = vhdx_user_visible_write(bs, s);
1183 if (ret < 0) {
1184 goto exit;
1187 while (nb_sectors > 0) {
1188 bool use_zero_buffers = false;
1189 bat_update = false;
1190 if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1191 /* not supported yet */
1192 ret = -ENOTSUP;
1193 goto exit;
1194 } else {
1195 vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1196 sectors_to_write = sinfo.sectors_avail;
1198 qemu_iovec_reset(&hd_qiov);
1199 /* check the payload block state */
1200 bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK;
1201 switch (bat_state) {
1202 case PAYLOAD_BLOCK_ZERO:
1203 /* in this case, we need to preserve zero writes for
1204 * data that is not part of this write, so we must pad
1205 * the rest of the buffer to zeroes */
1207 /* if we are on a posix system with ftruncate() that extends
1208 * a file, then it is zero-filled for us. On Win32, the raw
1209 * layer uses SetFilePointer and SetFileEnd, which does not
1210 * zero fill AFAIK */
1212 /* Queue another write of zero buffers if the underlying file
1213 * does not zero-fill on file extension */
1215 if (bdrv_has_zero_init(bs->file) == 0) {
1216 use_zero_buffers = true;
1218 /* zero fill the front, if any */
1219 if (sinfo.block_offset) {
1220 iov1.iov_len = sinfo.block_offset;
1221 iov1.iov_base = qemu_blockalign(bs, iov1.iov_len);
1222 memset(iov1.iov_base, 0, iov1.iov_len);
1223 qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0,
1224 sinfo.block_offset);
1225 sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS;
1228 /* our actual data */
1229 qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1230 sinfo.bytes_avail);
1232 /* zero fill the back, if any */
1233 if ((sinfo.bytes_avail - sinfo.block_offset) <
1234 s->block_size) {
1235 iov2.iov_len = s->block_size -
1236 (sinfo.bytes_avail + sinfo.block_offset);
1237 iov2.iov_base = qemu_blockalign(bs, iov2.iov_len);
1238 memset(iov2.iov_base, 0, iov2.iov_len);
1239 qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0,
1240 sinfo.block_offset);
1241 sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS;
1245 /* fall through */
1246 case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1247 case PAYLOAD_BLOCK_UNMAPPED: /* fall through */
1248 case PAYLOAD_BLOCK_UNDEFINED: /* fall through */
1249 bat_prior_offset = sinfo.file_offset;
1250 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset);
1251 if (ret < 0) {
1252 goto exit;
1254 /* once we support differencing files, this may also be
1255 * partially present */
1256 /* update block state to the newly specified state */
1257 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1258 &bat_entry_offset,
1259 PAYLOAD_BLOCK_FULLY_PRESENT);
1260 bat_update = true;
1261 /* since we just allocated a block, file_offset is the
1262 * beginning of the payload block. It needs to be the
1263 * write address, which includes the offset into the block */
1264 if (!use_zero_buffers) {
1265 sinfo.file_offset += sinfo.block_offset;
1267 /* fall through */
1268 case PAYLOAD_BLOCK_FULLY_PRESENT:
1269 /* if the file offset address is in the header zone,
1270 * there is a problem */
1271 if (sinfo.file_offset < (1024 * 1024)) {
1272 ret = -EFAULT;
1273 goto error_bat_restore;
1276 if (!use_zero_buffers) {
1277 qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1278 sinfo.bytes_avail);
1280 /* block exists, so we can just overwrite it */
1281 qemu_co_mutex_unlock(&s->lock);
1282 ret = bdrv_co_writev(bs->file,
1283 sinfo.file_offset >> BDRV_SECTOR_BITS,
1284 sectors_to_write, &hd_qiov);
1285 qemu_co_mutex_lock(&s->lock);
1286 if (ret < 0) {
1287 goto error_bat_restore;
1289 break;
1290 case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1291 /* we don't yet support difference files, fall through
1292 * to error */
1293 default:
1294 ret = -EIO;
1295 goto exit;
1296 break;
1299 if (bat_update) {
1300 /* this will update the BAT entry into the log journal, and
1301 * then flush the log journal out to disk */
1302 ret = vhdx_log_write_and_flush(bs, s, &bat_entry,
1303 sizeof(VHDXBatEntry),
1304 bat_entry_offset);
1305 if (ret < 0) {
1306 goto exit;
1310 nb_sectors -= sinfo.sectors_avail;
1311 sector_num += sinfo.sectors_avail;
1312 bytes_done += sinfo.bytes_avail;
1317 goto exit;
1319 error_bat_restore:
1320 if (bat_update) {
1321 /* keep metadata in sync, and restore the bat entry state
1322 * if error. */
1323 sinfo.file_offset = bat_prior_offset;
1324 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1325 &bat_entry_offset, bat_state);
1327 exit:
1328 qemu_vfree(iov1.iov_base);
1329 qemu_vfree(iov2.iov_base);
1330 qemu_co_mutex_unlock(&s->lock);
1331 qemu_iovec_destroy(&hd_qiov);
1332 return ret;
1338 * Create VHDX Headers
1340 * There are 2 headers, and the highest sequence number will represent
1341 * the active header
1343 static int vhdx_create_new_headers(BlockDriverState *bs, uint64_t image_size,
1344 uint32_t log_size)
1346 int ret = 0;
1347 VHDXHeader *hdr = NULL;
1349 hdr = g_malloc0(sizeof(VHDXHeader));
1351 hdr->signature = VHDX_HEADER_SIGNATURE;
1352 hdr->sequence_number = g_random_int();
1353 hdr->log_version = 0;
1354 hdr->version = 1;
1355 hdr->log_length = log_size;
1356 hdr->log_offset = VHDX_HEADER_SECTION_END;
1357 vhdx_guid_generate(&hdr->file_write_guid);
1358 vhdx_guid_generate(&hdr->data_write_guid);
1360 ret = vhdx_write_header(bs, hdr, VHDX_HEADER1_OFFSET, false);
1361 if (ret < 0) {
1362 goto exit;
1364 hdr->sequence_number++;
1365 ret = vhdx_write_header(bs, hdr, VHDX_HEADER2_OFFSET, false);
1366 if (ret < 0) {
1367 goto exit;
1370 exit:
1371 g_free(hdr);
1372 return ret;
1377 * Create the Metadata entries.
1379 * For more details on the entries, see section 3.5 (pg 29) in the
1380 * VHDX 1.00 specification.
1382 * We support 5 metadata entries (all required by spec):
1383 * File Parameters,
1384 * Virtual Disk Size,
1385 * Page 83 Data,
1386 * Logical Sector Size,
1387 * Physical Sector Size
1389 * The first 64KB of the Metadata section is reserved for the metadata
1390 * header and entries; beyond that, the metadata items themselves reside.
1392 static int vhdx_create_new_metadata(BlockDriverState *bs,
1393 uint64_t image_size,
1394 uint32_t block_size,
1395 uint32_t sector_size,
1396 uint64_t metadata_offset,
1397 VHDXImageType type)
1399 int ret = 0;
1400 uint32_t offset = 0;
1401 void *buffer = NULL;
1402 void *entry_buffer;
1403 VHDXMetadataTableHeader *md_table;;
1404 VHDXMetadataTableEntry *md_table_entry;
1406 /* Metadata entries */
1407 VHDXFileParameters *mt_file_params;
1408 VHDXVirtualDiskSize *mt_virtual_size;
1409 VHDXPage83Data *mt_page83;
1410 VHDXVirtualDiskLogicalSectorSize *mt_log_sector_size;
1411 VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size;
1413 entry_buffer = g_malloc0(sizeof(VHDXFileParameters) +
1414 sizeof(VHDXVirtualDiskSize) +
1415 sizeof(VHDXPage83Data) +
1416 sizeof(VHDXVirtualDiskLogicalSectorSize) +
1417 sizeof(VHDXVirtualDiskPhysicalSectorSize));
1419 mt_file_params = entry_buffer;
1420 offset += sizeof(VHDXFileParameters);
1421 mt_virtual_size = entry_buffer + offset;
1422 offset += sizeof(VHDXVirtualDiskSize);
1423 mt_page83 = entry_buffer + offset;
1424 offset += sizeof(VHDXPage83Data);
1425 mt_log_sector_size = entry_buffer + offset;
1426 offset += sizeof(VHDXVirtualDiskLogicalSectorSize);
1427 mt_phys_sector_size = entry_buffer + offset;
1429 mt_file_params->block_size = cpu_to_le32(block_size);
1430 if (type == VHDX_TYPE_FIXED) {
1431 mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED;
1432 cpu_to_le32s(&mt_file_params->data_bits);
1435 vhdx_guid_generate(&mt_page83->page_83_data);
1436 cpu_to_leguids(&mt_page83->page_83_data);
1437 mt_virtual_size->virtual_disk_size = cpu_to_le64(image_size);
1438 mt_log_sector_size->logical_sector_size = cpu_to_le32(sector_size);
1439 mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size);
1441 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1442 md_table = buffer;
1444 md_table->signature = VHDX_METADATA_SIGNATURE;
1445 md_table->entry_count = 5;
1446 vhdx_metadata_header_le_export(md_table);
1449 /* This will reference beyond the reserved table portion */
1450 offset = 64 * KiB;
1452 md_table_entry = buffer + sizeof(VHDXMetadataTableHeader);
1454 md_table_entry[0].item_id = file_param_guid;
1455 md_table_entry[0].offset = offset;
1456 md_table_entry[0].length = sizeof(VHDXFileParameters);
1457 md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED;
1458 offset += md_table_entry[0].length;
1459 vhdx_metadata_entry_le_export(&md_table_entry[0]);
1461 md_table_entry[1].item_id = virtual_size_guid;
1462 md_table_entry[1].offset = offset;
1463 md_table_entry[1].length = sizeof(VHDXVirtualDiskSize);
1464 md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1465 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1466 offset += md_table_entry[1].length;
1467 vhdx_metadata_entry_le_export(&md_table_entry[1]);
1469 md_table_entry[2].item_id = page83_guid;
1470 md_table_entry[2].offset = offset;
1471 md_table_entry[2].length = sizeof(VHDXPage83Data);
1472 md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1473 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1474 offset += md_table_entry[2].length;
1475 vhdx_metadata_entry_le_export(&md_table_entry[2]);
1477 md_table_entry[3].item_id = logical_sector_guid;
1478 md_table_entry[3].offset = offset;
1479 md_table_entry[3].length = sizeof(VHDXVirtualDiskLogicalSectorSize);
1480 md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1481 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1482 offset += md_table_entry[3].length;
1483 vhdx_metadata_entry_le_export(&md_table_entry[3]);
1485 md_table_entry[4].item_id = phys_sector_guid;
1486 md_table_entry[4].offset = offset;
1487 md_table_entry[4].length = sizeof(VHDXVirtualDiskPhysicalSectorSize);
1488 md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1489 VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1490 vhdx_metadata_entry_le_export(&md_table_entry[4]);
1492 ret = bdrv_pwrite(bs, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE);
1493 if (ret < 0) {
1494 goto exit;
1497 ret = bdrv_pwrite(bs, metadata_offset + (64 * KiB), entry_buffer,
1498 VHDX_HEADER_BLOCK_SIZE);
1499 if (ret < 0) {
1500 goto exit;
1504 exit:
1505 g_free(buffer);
1506 g_free(entry_buffer);
1507 return ret;
1510 /* This create the actual BAT itself. We currently only support
1511 * 'Dynamic' and 'Fixed' image types.
1513 * Dynamic images: default state of the BAT is all zeroes.
1515 * Fixed images: default state of the BAT is fully populated, with
1516 * file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT.
1518 static int vhdx_create_bat(BlockDriverState *bs, BDRVVHDXState *s,
1519 uint64_t image_size, VHDXImageType type,
1520 bool use_zero_blocks, VHDXRegionTableEntry *rt_bat)
1522 int ret = 0;
1523 uint64_t data_file_offset;
1524 uint64_t total_sectors = 0;
1525 uint64_t sector_num = 0;
1526 uint64_t unused;
1527 int block_state;
1528 VHDXSectorInfo sinfo;
1530 assert(s->bat == NULL);
1532 /* this gives a data start after BAT/bitmap entries, and well
1533 * past any metadata entries (with a 4 MB buffer for future
1534 * expansion */
1535 data_file_offset = rt_bat->file_offset + rt_bat->length + 5 * MiB;
1536 total_sectors = image_size >> s->logical_sector_size_bits;
1538 if (type == VHDX_TYPE_DYNAMIC) {
1539 /* All zeroes, so we can just extend the file - the end of the BAT
1540 * is the furthest thing we have written yet */
1541 ret = bdrv_truncate(bs, data_file_offset);
1542 if (ret < 0) {
1543 goto exit;
1545 } else if (type == VHDX_TYPE_FIXED) {
1546 ret = bdrv_truncate(bs, data_file_offset + image_size);
1547 if (ret < 0) {
1548 goto exit;
1550 } else {
1551 ret = -ENOTSUP;
1552 goto exit;
1555 if (type == VHDX_TYPE_FIXED ||
1556 use_zero_blocks ||
1557 bdrv_has_zero_init(bs) == 0) {
1558 /* for a fixed file, the default BAT entry is not zero */
1559 s->bat = g_malloc0(rt_bat->length);
1560 block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT :
1561 PAYLOAD_BLOCK_NOT_PRESENT;
1562 block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state;
1563 /* fill the BAT by emulating sector writes of sectors_per_block size */
1564 while (sector_num < total_sectors) {
1565 vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo);
1566 sinfo.file_offset = data_file_offset +
1567 (sector_num << s->logical_sector_size_bits);
1568 sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB);
1569 vhdx_update_bat_table_entry(bs, s, &sinfo, &unused, &unused,
1570 block_state);
1571 cpu_to_le64s(&s->bat[sinfo.bat_idx]);
1572 sector_num += s->sectors_per_block;
1574 ret = bdrv_pwrite(bs, rt_bat->file_offset, s->bat, rt_bat->length);
1575 if (ret < 0) {
1576 goto exit;
1582 exit:
1583 g_free(s->bat);
1584 return ret;
1587 /* Creates the region table header, and region table entries.
1588 * There are 2 supported region table entries: BAT, and Metadata/
1590 * As the calculations for the BAT region table are also needed
1591 * to create the BAT itself, we will also cause the BAT to be
1592 * created.
1594 static int vhdx_create_new_region_table(BlockDriverState *bs,
1595 uint64_t image_size,
1596 uint32_t block_size,
1597 uint32_t sector_size,
1598 uint32_t log_size,
1599 bool use_zero_blocks,
1600 VHDXImageType type,
1601 uint64_t *metadata_offset)
1603 int ret = 0;
1604 uint32_t offset = 0;
1605 void *buffer = NULL;
1606 BDRVVHDXState *s = NULL;
1607 VHDXRegionTableHeader *region_table;
1608 VHDXRegionTableEntry *rt_bat;
1609 VHDXRegionTableEntry *rt_metadata;
1611 assert(metadata_offset != NULL);
1613 /* Populate enough of the BDRVVHDXState to be able to use the
1614 * pre-existing BAT calculation, translation, and update functions */
1615 s = g_malloc0(sizeof(BDRVVHDXState));
1617 s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
1618 (uint64_t) sector_size / (uint64_t) block_size;
1620 s->sectors_per_block = block_size / sector_size;
1621 s->virtual_disk_size = image_size;
1622 s->block_size = block_size;
1623 s->logical_sector_size = sector_size;
1625 vhdx_set_shift_bits(s);
1627 vhdx_calc_bat_entries(s);
1629 /* At this point the VHDX state is populated enough for creation */
1631 /* a single buffer is used so we can calculate the checksum over the
1632 * entire 64KB block */
1633 buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1634 region_table = buffer;
1635 offset += sizeof(VHDXRegionTableHeader);
1636 rt_bat = buffer + offset;
1637 offset += sizeof(VHDXRegionTableEntry);
1638 rt_metadata = buffer + offset;
1640 region_table->signature = VHDX_REGION_SIGNATURE;
1641 region_table->entry_count = 2; /* BAT and Metadata */
1643 rt_bat->guid = bat_guid;
1644 rt_bat->length = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB);
1645 rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB);
1646 s->bat_offset = rt_bat->file_offset;
1648 rt_metadata->guid = metadata_guid;
1649 rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length,
1650 MiB);
1651 rt_metadata->length = 1 * MiB; /* min size, and more than enough */
1652 *metadata_offset = rt_metadata->file_offset;
1654 vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE,
1655 offsetof(VHDXRegionTableHeader, checksum));
1658 /* The region table gives us the data we need to create the BAT,
1659 * so do that now */
1660 ret = vhdx_create_bat(bs, s, image_size, type, use_zero_blocks, rt_bat);
1662 /* Now write out the region headers to disk */
1663 vhdx_region_header_le_export(region_table);
1664 vhdx_region_entry_le_export(rt_bat);
1665 vhdx_region_entry_le_export(rt_metadata);
1667 ret = bdrv_pwrite(bs, VHDX_REGION_TABLE_OFFSET, buffer,
1668 VHDX_HEADER_BLOCK_SIZE);
1669 if (ret < 0) {
1670 goto exit;
1673 ret = bdrv_pwrite(bs, VHDX_REGION_TABLE2_OFFSET, buffer,
1674 VHDX_HEADER_BLOCK_SIZE);
1675 if (ret < 0) {
1676 goto exit;
1680 exit:
1681 g_free(s);
1682 g_free(buffer);
1683 return ret;
1686 /* We need to create the following elements:
1688 * .-----------------------------------------------------------------.
1689 * | (A) | (B) | (C) | (D) | (E) |
1690 * | File ID | Header1 | Header 2 | Region Tbl 1 | Region Tbl 2 |
1691 * | | | | | |
1692 * .-----------------------------------------------------------------.
1693 * 0 64KB 128KB 192KB 256KB 320KB
1696 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1697 * | (F) | (G) | (H) | |
1698 * | Journal Log | BAT / Bitmap | Metadata | .... data ...... |
1699 * | | | | |
1700 * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1701 * 1MB
1703 static int vhdx_create(const char *filename, QEMUOptionParameter *options,
1704 Error **errp)
1706 int ret = 0;
1707 uint64_t image_size = (uint64_t) 2 * GiB;
1708 uint32_t log_size = 1 * MiB;
1709 uint32_t block_size = 0;
1710 uint64_t signature;
1711 uint64_t metadata_offset;
1712 bool use_zero_blocks = false;
1714 gunichar2 *creator = NULL;
1715 glong creator_items;
1716 BlockDriverState *bs;
1717 const char *type = NULL;
1718 VHDXImageType image_type;
1719 Error *local_err = NULL;
1721 while (options && options->name) {
1722 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1723 image_size = options->value.n;
1724 } else if (!strcmp(options->name, VHDX_BLOCK_OPT_LOG_SIZE)) {
1725 log_size = options->value.n;
1726 } else if (!strcmp(options->name, VHDX_BLOCK_OPT_BLOCK_SIZE)) {
1727 block_size = options->value.n;
1728 } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) {
1729 type = options->value.s;
1730 } else if (!strcmp(options->name, VHDX_BLOCK_OPT_ZERO)) {
1731 use_zero_blocks = options->value.n != 0;
1733 options++;
1736 if (image_size > VHDX_MAX_IMAGE_SIZE) {
1737 error_setg_errno(errp, EINVAL, "Image size too large; max of 64TB");
1738 ret = -EINVAL;
1739 goto exit;
1742 if (type == NULL) {
1743 type = "dynamic";
1746 if (!strcmp(type, "dynamic")) {
1747 image_type = VHDX_TYPE_DYNAMIC;
1748 } else if (!strcmp(type, "fixed")) {
1749 image_type = VHDX_TYPE_FIXED;
1750 } else if (!strcmp(type, "differencing")) {
1751 error_setg_errno(errp, ENOTSUP,
1752 "Differencing files not yet supported");
1753 ret = -ENOTSUP;
1754 goto exit;
1755 } else {
1756 ret = -EINVAL;
1757 goto exit;
1760 /* These are pretty arbitrary, and mainly designed to keep the BAT
1761 * size reasonable to load into RAM */
1762 if (block_size == 0) {
1763 if (image_size > 32 * TiB) {
1764 block_size = 64 * MiB;
1765 } else if (image_size > (uint64_t) 100 * GiB) {
1766 block_size = 32 * MiB;
1767 } else if (image_size > 1 * GiB) {
1768 block_size = 16 * MiB;
1769 } else {
1770 block_size = 8 * MiB;
1775 /* make the log size close to what was specified, but must be
1776 * min 1MB, and multiple of 1MB */
1777 log_size = ROUND_UP(log_size, MiB);
1779 block_size = ROUND_UP(block_size, MiB);
1780 block_size = block_size > VHDX_BLOCK_SIZE_MAX ? VHDX_BLOCK_SIZE_MAX :
1781 block_size;
1783 ret = bdrv_create_file(filename, options, &local_err);
1784 if (ret < 0) {
1785 error_propagate(errp, local_err);
1786 goto exit;
1789 ret = bdrv_file_open(&bs, filename, NULL, BDRV_O_RDWR, &local_err);
1790 if (ret < 0) {
1791 error_propagate(errp, local_err);
1792 goto exit;
1795 /* Create (A) */
1797 /* The creator field is optional, but may be useful for
1798 * debugging / diagnostics */
1799 creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL,
1800 &creator_items, NULL);
1801 signature = cpu_to_le64(VHDX_FILE_SIGNATURE);
1802 bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature));
1803 if (ret < 0) {
1804 goto delete_and_exit;
1806 if (creator) {
1807 bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET + sizeof(signature), creator,
1808 creator_items * sizeof(gunichar2));
1809 if (ret < 0) {
1810 goto delete_and_exit;
1815 /* Creates (B),(C) */
1816 ret = vhdx_create_new_headers(bs, image_size, log_size);
1817 if (ret < 0) {
1818 goto delete_and_exit;
1821 /* Creates (D),(E),(G) explicitly. (F) created as by-product */
1822 ret = vhdx_create_new_region_table(bs, image_size, block_size, 512,
1823 log_size, use_zero_blocks, image_type,
1824 &metadata_offset);
1825 if (ret < 0) {
1826 goto delete_and_exit;
1829 /* Creates (H) */
1830 ret = vhdx_create_new_metadata(bs, image_size, block_size, 512,
1831 metadata_offset, image_type);
1832 if (ret < 0) {
1833 goto delete_and_exit;
1838 delete_and_exit:
1839 bdrv_unref(bs);
1840 exit:
1841 g_free(creator);
1842 return ret;
1845 static QEMUOptionParameter vhdx_create_options[] = {
1847 .name = BLOCK_OPT_SIZE,
1848 .type = OPT_SIZE,
1849 .help = "Virtual disk size; max of 64TB."
1852 .name = VHDX_BLOCK_OPT_LOG_SIZE,
1853 .type = OPT_SIZE,
1854 .value.n = 1 * MiB,
1855 .help = "Log size; min 1MB."
1858 .name = VHDX_BLOCK_OPT_BLOCK_SIZE,
1859 .type = OPT_SIZE,
1860 .value.n = 0,
1861 .help = "Block Size; min 1MB, max 256MB. " \
1862 "0 means auto-calculate based on image size."
1865 .name = BLOCK_OPT_SUBFMT,
1866 .type = OPT_STRING,
1867 .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "\
1868 "Default is 'dynamic'."
1871 .name = VHDX_BLOCK_OPT_ZERO,
1872 .type = OPT_FLAG,
1873 .help = "Force use of payload blocks of type 'ZERO'. Non-standard."
1875 { NULL }
1878 static BlockDriver bdrv_vhdx = {
1879 .format_name = "vhdx",
1880 .instance_size = sizeof(BDRVVHDXState),
1881 .bdrv_probe = vhdx_probe,
1882 .bdrv_open = vhdx_open,
1883 .bdrv_close = vhdx_close,
1884 .bdrv_reopen_prepare = vhdx_reopen_prepare,
1885 .bdrv_co_readv = vhdx_co_readv,
1886 .bdrv_co_writev = vhdx_co_writev,
1887 .bdrv_create = vhdx_create,
1889 .create_options = vhdx_create_options,
1892 static void bdrv_vhdx_init(void)
1894 bdrv_register(&bdrv_vhdx);
1897 block_init(bdrv_vhdx_init);