xref: /openbmc/qemu/block/vhdx.c (revision 461a2753)
1 /*
2  * Block driver for Hyper-V VHDX Images
3  *
4  * Copyright (c) 2013 Red Hat, Inc.,
5  *
6  * Authors:
7  *  Jeff Cody <jcody@redhat.com>
8  *
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
12  *
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.
15  *
16  */
17 
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"
24 
25 #include <uuid/uuid.h>
26 #include <glib.h>
27 
28 /* Options for VHDX creation */
29 
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"
33 
34 typedef enum VHDXImageType {
35     VHDX_TYPE_DYNAMIC = 0,
36     VHDX_TYPE_FIXED,
37     VHDX_TYPE_DIFFERENCING,   /* Currently unsupported */
38 } VHDXImageType;
39 
40 /* Several metadata and region table data entries are identified by
41  * guids in  a MS-specific GUID format. */
42 
43 
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} };
50 
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} };
56 
57 
58 
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} };
65 
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} };
71 
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} };
77 
78 
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} };
84 
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} };
91 
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} };
98 
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} };
107 
108 
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
115 
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)
120 
121 
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;
130 
131 /* Calculates new checksum.
132  *
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)
137  *
138  * Note: The buffer should have all multi-byte data in little-endian format,
139  *       and the resulting checksum is in little endian format.
140  */
141 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset)
142 {
143     uint32_t crc;
144 
145     assert(buf != NULL);
146     assert(size > (crc_offset + sizeof(crc)));
147 
148     memset(buf + crc_offset, 0, sizeof(crc));
149     crc =  crc32c(0xffffffff, buf, size);
150     cpu_to_le32s(&crc);
151     memcpy(buf + crc_offset, &crc, sizeof(crc));
152 
153     return crc;
154 }
155 
156 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size,
157                             int crc_offset)
158 {
159     uint32_t crc_new;
160     uint32_t crc_orig;
161     assert(buf != NULL);
162 
163     if (crc_offset > 0) {
164         memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
165         memset(buf + crc_offset, 0, sizeof(crc_orig));
166     }
167 
168     crc_new = crc32c(crc, buf, size);
169     if (crc_offset > 0) {
170         memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig));
171     }
172 
173     return crc_new;
174 }
175 
176 /* Validates the checksum of the buffer, with an in-place CRC.
177  *
178  * Zero is substituted during crc calculation for the original crc field,
179  * and the crc field is restored afterwards.  But the buffer will be modifed
180  * during the calculation, so this may not be not suitable for multi-threaded
181  * use.
182  *
183  * crc_offset: byte offset in buf of the buffer crc
184  * buf: buffer pointer
185  * size: size of buffer (must be > crc_offset+4)
186  *
187  * returns true if checksum is valid, false otherwise
188  */
189 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset)
190 {
191     uint32_t crc_orig;
192     uint32_t crc;
193 
194     assert(buf != NULL);
195     assert(size > (crc_offset + 4));
196 
197     memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
198     crc_orig = le32_to_cpu(crc_orig);
199 
200     crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset);
201 
202     return crc == crc_orig;
203 }
204 
205 
206 /*
207  * This generates a UUID that is compliant with the MS GUIDs used
208  * in the VHDX spec (and elsewhere).
209  */
210 void vhdx_guid_generate(MSGUID *guid)
211 {
212     uuid_t uuid;
213     assert(guid != NULL);
214 
215     uuid_generate(uuid);
216     memcpy(guid, uuid, sizeof(MSGUID));
217 }
218 
219 /* Check for region overlaps inside the VHDX image */
220 static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length)
221 {
222     int ret = 0;
223     uint64_t end;
224     VHDXRegionEntry *r;
225 
226     end = start + length;
227     QLIST_FOREACH(r, &s->regions, entries) {
228         if (!((start >= r->end) || (end <= r->start))) {
229             ret = -EINVAL;
230             goto exit;
231         }
232     }
233 
234 exit:
235     return ret;
236 }
237 
238 /* Register a region for future checks */
239 static void vhdx_region_register(BDRVVHDXState *s,
240                                  uint64_t start, uint64_t length)
241 {
242     VHDXRegionEntry *r;
243 
244     r = g_malloc0(sizeof(*r));
245 
246     r->start = start;
247     r->end = start + length;
248 
249     QLIST_INSERT_HEAD(&s->regions, r, entries);
250 }
251 
252 /* Free all registered regions */
253 static void vhdx_region_unregister_all(BDRVVHDXState *s)
254 {
255     VHDXRegionEntry *r, *r_next;
256 
257     QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) {
258         QLIST_REMOVE(r, entries);
259         g_free(r);
260     }
261 }
262 
263 static void vhdx_set_shift_bits(BDRVVHDXState *s)
264 {
265     s->logical_sector_size_bits = 31 - clz32(s->logical_sector_size);
266     s->sectors_per_block_bits =   31 - clz32(s->sectors_per_block);
267     s->chunk_ratio_bits =         63 - clz64(s->chunk_ratio);
268     s->block_size_bits =          31 - clz32(s->block_size);
269 }
270 
271 /*
272  * Per the MS VHDX Specification, for every VHDX file:
273  *      - The header section is fixed size - 1 MB
274  *      - The header section is always the first "object"
275  *      - The first 64KB of the header is the File Identifier
276  *      - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile")
277  *      - The following 512 bytes constitute a UTF-16 string identifiying the
278  *        software that created the file, and is optional and diagnostic only.
279  *
280  *  Therefore, we probe by looking for the vhdxfile signature "vhdxfile"
281  */
282 static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename)
283 {
284     if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) {
285         return 100;
286     }
287     return 0;
288 }
289 
290 /*
291  * Writes the header to the specified offset.
292  *
293  * This will optionally read in buffer data from disk (otherwise zero-fill),
294  * and then update the header checksum.  Header is converted to proper
295  * endianness before being written to the specified file offset
296  */
297 static int vhdx_write_header(BlockDriverState *bs_file, VHDXHeader *hdr,
298                              uint64_t offset, bool read)
299 {
300     uint8_t *buffer = NULL;
301     int ret;
302     VHDXHeader *header_le;
303 
304     assert(bs_file != NULL);
305     assert(hdr != NULL);
306 
307     /* the header checksum is not over just the packed size of VHDXHeader,
308      * but rather over the entire 'reserved' range for the header, which is
309      * 4KB (VHDX_HEADER_SIZE). */
310 
311     buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE);
312     if (read) {
313         /* if true, we can't assume the extra reserved bytes are 0 */
314         ret = bdrv_pread(bs_file, offset, buffer, VHDX_HEADER_SIZE);
315         if (ret < 0) {
316             goto exit;
317         }
318     } else {
319         memset(buffer, 0, VHDX_HEADER_SIZE);
320     }
321 
322     /* overwrite the actual VHDXHeader portion */
323     header_le = (VHDXHeader *)buffer;
324     memcpy(header_le, hdr, sizeof(VHDXHeader));
325     vhdx_header_le_export(hdr, header_le);
326     vhdx_update_checksum(buffer, VHDX_HEADER_SIZE,
327                          offsetof(VHDXHeader, checksum));
328     ret = bdrv_pwrite_sync(bs_file, offset, header_le, sizeof(VHDXHeader));
329 
330 exit:
331     qemu_vfree(buffer);
332     return ret;
333 }
334 
335 /* Update the VHDX headers
336  *
337  * This follows the VHDX spec procedures for header updates.
338  *
339  *  - non-current header is updated with largest sequence number
340  */
341 static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s,
342                               bool generate_data_write_guid, MSGUID *log_guid)
343 {
344     int ret = 0;
345     int hdr_idx = 0;
346     uint64_t header_offset = VHDX_HEADER1_OFFSET;
347 
348     VHDXHeader *active_header;
349     VHDXHeader *inactive_header;
350 
351     /* operate on the non-current header */
352     if (s->curr_header == 0) {
353         hdr_idx = 1;
354         header_offset = VHDX_HEADER2_OFFSET;
355     }
356 
357     active_header   = s->headers[s->curr_header];
358     inactive_header = s->headers[hdr_idx];
359 
360     inactive_header->sequence_number = active_header->sequence_number + 1;
361 
362     /* a new file guid must be generated before any file write, including
363      * headers */
364     inactive_header->file_write_guid = s->session_guid;
365 
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);
370     }
371 
372     /* update the log guid if present */
373     if (log_guid) {
374         inactive_header->log_guid = *log_guid;
375     }
376 
377     ret = vhdx_write_header(bs->file, inactive_header, header_offset, true);
378     if (ret < 0) {
379         goto exit;
380     }
381     s->curr_header = hdr_idx;
382 
383 exit:
384     return ret;
385 }
386 
387 /*
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
390  */
391 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s,
392                         bool generate_data_write_guid, MSGUID *log_guid)
393 {
394     int ret;
395 
396     ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
397     if (ret < 0) {
398         return ret;
399     }
400     ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
401     return ret;
402 }
403 
404 /* opens the specified header block from the VHDX file header section */
405 static void vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s,
406                               Error **errp)
407 {
408     int ret;
409     VHDXHeader *header1;
410     VHDXHeader *header2;
411     bool h1_valid = false;
412     bool h2_valid = false;
413     uint64_t h1_seq = 0;
414     uint64_t h2_seq = 0;
415     uint8_t *buffer;
416 
417     /* header1 & header2 are freed in vhdx_close() */
418     header1 = qemu_blockalign(bs, sizeof(VHDXHeader));
419     header2 = qemu_blockalign(bs, sizeof(VHDXHeader));
420 
421     buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE);
422 
423     s->headers[0] = header1;
424     s->headers[1] = header2;
425 
426     /* We have to read the whole VHDX_HEADER_SIZE instead of
427      * sizeof(VHDXHeader), because the checksum is over the whole
428      * region */
429     ret = bdrv_pread(bs->file, VHDX_HEADER1_OFFSET, buffer, VHDX_HEADER_SIZE);
430     if (ret < 0) {
431         goto fail;
432     }
433     /* copy over just the relevant portion that we need */
434     memcpy(header1, buffer, sizeof(VHDXHeader));
435 
436     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
437         vhdx_header_le_import(header1);
438         if (header1->signature == VHDX_HEADER_SIGNATURE &&
439             header1->version == 1) {
440             h1_seq = header1->sequence_number;
441             h1_valid = true;
442         }
443     }
444 
445     ret = bdrv_pread(bs->file, VHDX_HEADER2_OFFSET, buffer, VHDX_HEADER_SIZE);
446     if (ret < 0) {
447         goto fail;
448     }
449     /* copy over just the relevant portion that we need */
450     memcpy(header2, buffer, sizeof(VHDXHeader));
451 
452     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
453         vhdx_header_le_import(header2);
454         if (header2->signature == VHDX_HEADER_SIGNATURE &&
455             header2->version == 1) {
456             h2_seq = header2->sequence_number;
457             h2_valid = true;
458         }
459     }
460 
461     /* If there is only 1 valid header (or no valid headers), we
462      * don't care what the sequence numbers are */
463     if (h1_valid && !h2_valid) {
464         s->curr_header = 0;
465     } else if (!h1_valid && h2_valid) {
466         s->curr_header = 1;
467     } else if (!h1_valid && !h2_valid) {
468         goto fail;
469     } else {
470         /* If both headers are valid, then we choose the active one by the
471          * highest sequence number.  If the sequence numbers are equal, that is
472          * invalid */
473         if (h1_seq > h2_seq) {
474             s->curr_header = 0;
475         } else if (h2_seq > h1_seq) {
476             s->curr_header = 1;
477         } else {
478             /* The Microsoft Disk2VHD tool will create 2 identical
479              * headers, with identical sequence numbers.  If the headers are
480              * identical, don't consider the file corrupt */
481             if (!memcmp(header1, header2, sizeof(VHDXHeader))) {
482                 s->curr_header = 0;
483             } else {
484                 goto fail;
485             }
486         }
487     }
488 
489     vhdx_region_register(s, s->headers[s->curr_header]->log_offset,
490                             s->headers[s->curr_header]->log_length);
491     goto exit;
492 
493 fail:
494     error_setg_errno(errp, -ret, "No valid VHDX header found");
495     qemu_vfree(header1);
496     qemu_vfree(header2);
497     s->headers[0] = NULL;
498     s->headers[1] = NULL;
499 exit:
500     qemu_vfree(buffer);
501 }
502 
503 
504 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s)
505 {
506     int ret = 0;
507     uint8_t *buffer;
508     int offset = 0;
509     VHDXRegionTableEntry rt_entry;
510     uint32_t i;
511     bool bat_rt_found = false;
512     bool metadata_rt_found = false;
513 
514     /* We have to read the whole 64KB block, because the crc32 is over the
515      * whole block */
516     buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE);
517 
518     ret = bdrv_pread(bs->file, VHDX_REGION_TABLE_OFFSET, buffer,
519                      VHDX_HEADER_BLOCK_SIZE);
520     if (ret < 0) {
521         goto fail;
522     }
523     memcpy(&s->rt, buffer, sizeof(s->rt));
524     offset += sizeof(s->rt);
525 
526     if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4)) {
527         ret = -EINVAL;
528         goto fail;
529     }
530 
531     vhdx_region_header_le_import(&s->rt);
532 
533     if (s->rt.signature != VHDX_REGION_SIGNATURE) {
534         ret = -EINVAL;
535         goto fail;
536     }
537 
538 
539     /* Per spec, maximum region table entry count is 2047 */
540     if (s->rt.entry_count > 2047) {
541         ret = -EINVAL;
542         goto fail;
543     }
544 
545     for (i = 0; i < s->rt.entry_count; i++) {
546         memcpy(&rt_entry, buffer + offset, sizeof(rt_entry));
547         offset += sizeof(rt_entry);
548 
549         vhdx_region_entry_le_import(&rt_entry);
550 
551         /* check for region overlap between these entries, and any
552          * other memory regions in the file */
553         ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length);
554         if (ret < 0) {
555             goto fail;
556         }
557 
558         vhdx_region_register(s, rt_entry.file_offset, rt_entry.length);
559 
560         /* see if we recognize the entry */
561         if (guid_eq(rt_entry.guid, bat_guid)) {
562             /* must be unique; if we have already found it this is invalid */
563             if (bat_rt_found) {
564                 ret = -EINVAL;
565                 goto fail;
566             }
567             bat_rt_found = true;
568             s->bat_rt = rt_entry;
569             continue;
570         }
571 
572         if (guid_eq(rt_entry.guid, metadata_guid)) {
573             /* must be unique; if we have already found it this is invalid */
574             if (metadata_rt_found) {
575                 ret = -EINVAL;
576                 goto fail;
577             }
578             metadata_rt_found = true;
579             s->metadata_rt = rt_entry;
580             continue;
581         }
582 
583         if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) {
584             /* cannot read vhdx file - required region table entry that
585              * we do not understand.  per spec, we must fail to open */
586             ret = -ENOTSUP;
587             goto fail;
588         }
589     }
590 
591     if (!bat_rt_found || !metadata_rt_found) {
592         ret = -EINVAL;
593         goto fail;
594     }
595 
596     ret = 0;
597 
598 fail:
599     qemu_vfree(buffer);
600     return ret;
601 }
602 
603 
604 
605 /* Metadata initial parser
606  *
607  * This loads all the metadata entry fields.  This may cause additional
608  * fields to be processed (e.g. parent locator, etc..).
609  *
610  * There are 5 Metadata items that are always required:
611  *      - File Parameters (block size, has a parent)
612  *      - Virtual Disk Size (size, in bytes, of the virtual drive)
613  *      - Page 83 Data (scsi page 83 guid)
614  *      - Logical Sector Size (logical sector size in bytes, either 512 or
615  *                             4096.  We only support 512 currently)
616  *      - Physical Sector Size (512 or 4096)
617  *
618  * Also, if the File Parameters indicate this is a differencing file,
619  * we must also look for the Parent Locator metadata item.
620  */
621 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s)
622 {
623     int ret = 0;
624     uint8_t *buffer;
625     int offset = 0;
626     uint32_t i = 0;
627     VHDXMetadataTableEntry md_entry;
628 
629     buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE);
630 
631     ret = bdrv_pread(bs->file, s->metadata_rt.file_offset, buffer,
632                      VHDX_METADATA_TABLE_MAX_SIZE);
633     if (ret < 0) {
634         goto exit;
635     }
636     memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr));
637     offset += sizeof(s->metadata_hdr);
638 
639     vhdx_metadata_header_le_import(&s->metadata_hdr);
640 
641     if (s->metadata_hdr.signature != VHDX_METADATA_SIGNATURE) {
642         ret = -EINVAL;
643         goto exit;
644     }
645 
646     s->metadata_entries.present = 0;
647 
648     if ((s->metadata_hdr.entry_count * sizeof(md_entry)) >
649         (VHDX_METADATA_TABLE_MAX_SIZE - offset)) {
650         ret = -EINVAL;
651         goto exit;
652     }
653 
654     for (i = 0; i < s->metadata_hdr.entry_count; i++) {
655         memcpy(&md_entry, buffer + offset, sizeof(md_entry));
656         offset += sizeof(md_entry);
657 
658         vhdx_metadata_entry_le_import(&md_entry);
659 
660         if (guid_eq(md_entry.item_id, file_param_guid)) {
661             if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) {
662                 ret = -EINVAL;
663                 goto exit;
664             }
665             s->metadata_entries.file_parameters_entry = md_entry;
666             s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT;
667             continue;
668         }
669 
670         if (guid_eq(md_entry.item_id, virtual_size_guid)) {
671             if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) {
672                 ret = -EINVAL;
673                 goto exit;
674             }
675             s->metadata_entries.virtual_disk_size_entry = md_entry;
676             s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT;
677             continue;
678         }
679 
680         if (guid_eq(md_entry.item_id, page83_guid)) {
681             if (s->metadata_entries.present & META_PAGE_83_PRESENT) {
682                 ret = -EINVAL;
683                 goto exit;
684             }
685             s->metadata_entries.page83_data_entry = md_entry;
686             s->metadata_entries.present |= META_PAGE_83_PRESENT;
687             continue;
688         }
689 
690         if (guid_eq(md_entry.item_id, logical_sector_guid)) {
691             if (s->metadata_entries.present &
692                 META_LOGICAL_SECTOR_SIZE_PRESENT) {
693                 ret = -EINVAL;
694                 goto exit;
695             }
696             s->metadata_entries.logical_sector_size_entry = md_entry;
697             s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT;
698             continue;
699         }
700 
701         if (guid_eq(md_entry.item_id, phys_sector_guid)) {
702             if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) {
703                 ret = -EINVAL;
704                 goto exit;
705             }
706             s->metadata_entries.phys_sector_size_entry = md_entry;
707             s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT;
708             continue;
709         }
710 
711         if (guid_eq(md_entry.item_id, parent_locator_guid)) {
712             if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
713                 ret = -EINVAL;
714                 goto exit;
715             }
716             s->metadata_entries.parent_locator_entry = md_entry;
717             s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT;
718             continue;
719         }
720 
721         if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) {
722             /* cannot read vhdx file - required region table entry that
723              * we do not understand.  per spec, we must fail to open */
724             ret = -ENOTSUP;
725             goto exit;
726         }
727     }
728 
729     if (s->metadata_entries.present != META_ALL_PRESENT) {
730         ret = -ENOTSUP;
731         goto exit;
732     }
733 
734     ret = bdrv_pread(bs->file,
735                      s->metadata_entries.file_parameters_entry.offset
736                                          + s->metadata_rt.file_offset,
737                      &s->params,
738                      sizeof(s->params));
739 
740     if (ret < 0) {
741         goto exit;
742     }
743 
744     le32_to_cpus(&s->params.block_size);
745     le32_to_cpus(&s->params.data_bits);
746 
747 
748     /* We now have the file parameters, so we can tell if this is a
749      * differencing file (i.e.. has_parent), is dynamic or fixed
750      * sized (leave_blocks_allocated), and the block size */
751 
752     /* The parent locator required iff the file parameters has_parent set */
753     if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
754         if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
755             /* TODO: parse  parent locator fields */
756             ret = -ENOTSUP; /* temp, until differencing files are supported */
757             goto exit;
758         } else {
759             /* if has_parent is set, but there is not parent locator present,
760              * then that is an invalid combination */
761             ret = -EINVAL;
762             goto exit;
763         }
764     }
765 
766     /* determine virtual disk size, logical sector size,
767      * and phys sector size */
768 
769     ret = bdrv_pread(bs->file,
770                      s->metadata_entries.virtual_disk_size_entry.offset
771                                            + s->metadata_rt.file_offset,
772                      &s->virtual_disk_size,
773                      sizeof(uint64_t));
774     if (ret < 0) {
775         goto exit;
776     }
777     ret = bdrv_pread(bs->file,
778                      s->metadata_entries.logical_sector_size_entry.offset
779                                              + s->metadata_rt.file_offset,
780                      &s->logical_sector_size,
781                      sizeof(uint32_t));
782     if (ret < 0) {
783         goto exit;
784     }
785     ret = bdrv_pread(bs->file,
786                      s->metadata_entries.phys_sector_size_entry.offset
787                                           + s->metadata_rt.file_offset,
788                      &s->physical_sector_size,
789                      sizeof(uint32_t));
790     if (ret < 0) {
791         goto exit;
792     }
793 
794     le64_to_cpus(&s->virtual_disk_size);
795     le32_to_cpus(&s->logical_sector_size);
796     le32_to_cpus(&s->physical_sector_size);
797 
798     if (s->params.block_size < VHDX_BLOCK_SIZE_MIN ||
799         s->params.block_size > VHDX_BLOCK_SIZE_MAX) {
800         ret = -EINVAL;
801         goto exit;
802     }
803 
804     /* only 2 supported sector sizes */
805     if (s->logical_sector_size != 512 && s->logical_sector_size != 4096) {
806         ret = -EINVAL;
807         goto exit;
808     }
809 
810     /* Both block_size and sector_size are guaranteed powers of 2, below.
811        Due to range checks above, s->sectors_per_block can never be < 256 */
812     s->sectors_per_block = s->params.block_size / s->logical_sector_size;
813     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
814                      (uint64_t)s->logical_sector_size /
815                      (uint64_t)s->params.block_size;
816 
817     /* These values are ones we will want to use for division / multiplication
818      * later on, and they are all guaranteed (per the spec) to be powers of 2,
819      * so we can take advantage of that for shift operations during
820      * reads/writes */
821     if (s->logical_sector_size & (s->logical_sector_size - 1)) {
822         ret = -EINVAL;
823         goto exit;
824     }
825     if (s->sectors_per_block & (s->sectors_per_block - 1)) {
826         ret = -EINVAL;
827         goto exit;
828     }
829     if (s->chunk_ratio & (s->chunk_ratio - 1)) {
830         ret = -EINVAL;
831         goto exit;
832     }
833     s->block_size = s->params.block_size;
834     if (s->block_size & (s->block_size - 1)) {
835         ret = -EINVAL;
836         goto exit;
837     }
838 
839     vhdx_set_shift_bits(s);
840 
841     ret = 0;
842 
843 exit:
844     qemu_vfree(buffer);
845     return ret;
846 }
847 
848 /*
849  * Calculate the number of BAT entries, including sector
850  * bitmap entries.
851  */
852 static void vhdx_calc_bat_entries(BDRVVHDXState *s)
853 {
854     uint32_t data_blocks_cnt, bitmap_blocks_cnt;
855 
856     data_blocks_cnt = s->virtual_disk_size >> s->block_size_bits;
857     if (s->virtual_disk_size - (data_blocks_cnt << s->block_size_bits)) {
858         data_blocks_cnt++;
859     }
860     bitmap_blocks_cnt = data_blocks_cnt >> s->chunk_ratio_bits;
861     if (data_blocks_cnt - (bitmap_blocks_cnt << s->chunk_ratio_bits)) {
862         bitmap_blocks_cnt++;
863     }
864 
865     if (s->parent_entries) {
866         s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1);
867     } else {
868         s->bat_entries = data_blocks_cnt +
869                          ((data_blocks_cnt - 1) >> s->chunk_ratio_bits);
870     }
871 
872 }
873 
874 static void vhdx_close(BlockDriverState *bs)
875 {
876     BDRVVHDXState *s = bs->opaque;
877     qemu_vfree(s->headers[0]);
878     s->headers[0] = NULL;
879     qemu_vfree(s->headers[1]);
880     s->headers[1] = NULL;
881     qemu_vfree(s->bat);
882     s->bat = NULL;
883     qemu_vfree(s->parent_entries);
884     s->parent_entries = NULL;
885     migrate_del_blocker(s->migration_blocker);
886     error_free(s->migration_blocker);
887     qemu_vfree(s->log.hdr);
888     s->log.hdr = NULL;
889     vhdx_region_unregister_all(s);
890 }
891 
892 static int vhdx_open(BlockDriverState *bs, QDict *options, int flags,
893                      Error **errp)
894 {
895     BDRVVHDXState *s = bs->opaque;
896     int ret = 0;
897     uint32_t i;
898     uint64_t signature;
899     Error *local_err = NULL;
900 
901     s->bat = NULL;
902     s->first_visible_write = true;
903 
904     qemu_co_mutex_init(&s->lock);
905     QLIST_INIT(&s->regions);
906 
907     /* validate the file signature */
908     ret = bdrv_pread(bs->file, 0, &signature, sizeof(uint64_t));
909     if (ret < 0) {
910         goto fail;
911     }
912     if (memcmp(&signature, "vhdxfile", 8)) {
913         ret = -EINVAL;
914         goto fail;
915     }
916 
917     /* This is used for any header updates, for the file_write_guid.
918      * The spec dictates that a new value should be used for the first
919      * header update */
920     vhdx_guid_generate(&s->session_guid);
921 
922     vhdx_parse_header(bs, s, &local_err);
923     if (local_err != NULL) {
924         error_propagate(errp, local_err);
925         ret = -EINVAL;
926         goto fail;
927     }
928 
929     ret = vhdx_parse_log(bs, s, &s->log_replayed_on_open, errp);
930     if (ret < 0) {
931         goto fail;
932     }
933 
934     ret = vhdx_open_region_tables(bs, s);
935     if (ret < 0) {
936         goto fail;
937     }
938 
939     ret = vhdx_parse_metadata(bs, s);
940     if (ret < 0) {
941         goto fail;
942     }
943 
944     s->block_size = s->params.block_size;
945 
946     /* the VHDX spec dictates that virtual_disk_size is always a multiple of
947      * logical_sector_size */
948     bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits;
949 
950     vhdx_calc_bat_entries(s);
951 
952     s->bat_offset = s->bat_rt.file_offset;
953 
954     if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) {
955         /* BAT allocation is not large enough for all entries */
956         ret = -EINVAL;
957         goto fail;
958     }
959 
960     /* s->bat is freed in vhdx_close() */
961     s->bat = qemu_try_blockalign(bs->file, s->bat_rt.length);
962     if (s->bat == NULL) {
963         ret = -ENOMEM;
964         goto fail;
965     }
966 
967     ret = bdrv_pread(bs->file, s->bat_offset, s->bat, s->bat_rt.length);
968     if (ret < 0) {
969         goto fail;
970     }
971 
972     uint64_t payblocks = s->chunk_ratio;
973     /* endian convert, and verify populated BAT field file offsets against
974      * region table and log entries */
975     for (i = 0; i < s->bat_entries; i++) {
976         le64_to_cpus(&s->bat[i]);
977         if (payblocks--) {
978             /* payload bat entries */
979             if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) ==
980                     PAYLOAD_BLOCK_FULLY_PRESENT) {
981                 ret = vhdx_region_check(s, s->bat[i] & VHDX_BAT_FILE_OFF_MASK,
982                                         s->block_size);
983                 if (ret < 0) {
984                     goto fail;
985                 }
986             }
987         } else {
988             payblocks = s->chunk_ratio;
989             /* Once differencing files are supported, verify sector bitmap
990              * blocks here */
991         }
992     }
993 
994     if (flags & BDRV_O_RDWR) {
995         ret = vhdx_update_headers(bs, s, false, NULL);
996         if (ret < 0) {
997             goto fail;
998         }
999     }
1000 
1001     /* TODO: differencing files */
1002 
1003     /* Disable migration when VHDX images are used */
1004     error_set(&s->migration_blocker,
1005             QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1006             "vhdx", bs->device_name, "live migration");
1007     migrate_add_blocker(s->migration_blocker);
1008 
1009     return 0;
1010 fail:
1011     vhdx_close(bs);
1012     return ret;
1013 }
1014 
1015 static int vhdx_reopen_prepare(BDRVReopenState *state,
1016                                BlockReopenQueue *queue, Error **errp)
1017 {
1018     return 0;
1019 }
1020 
1021 
1022 /*
1023  * Perform sector to block offset translations, to get various
1024  * sector and file offsets into the image.  See VHDXSectorInfo
1025  */
1026 static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num,
1027                                  int nb_sectors, VHDXSectorInfo *sinfo)
1028 {
1029     uint32_t block_offset;
1030 
1031     sinfo->bat_idx = sector_num >> s->sectors_per_block_bits;
1032     /* effectively a modulo - this gives us the offset into the block
1033      * (in sector sizes) for our sector number */
1034     block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits);
1035     /* the chunk ratio gives us the interleaving of the sector
1036      * bitmaps, so we need to advance our page block index by the
1037      * sector bitmaps entry number */
1038     sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits;
1039 
1040     /* the number of sectors we can read/write in this cycle */
1041     sinfo->sectors_avail = s->sectors_per_block - block_offset;
1042 
1043     sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits;
1044 
1045     if (sinfo->sectors_avail > nb_sectors) {
1046         sinfo->sectors_avail = nb_sectors;
1047     }
1048 
1049     sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits;
1050 
1051     sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK;
1052 
1053     sinfo->block_offset = block_offset << s->logical_sector_size_bits;
1054 
1055     /* The file offset must be past the header section, so must be > 0 */
1056     if (sinfo->file_offset == 0) {
1057         return;
1058     }
1059 
1060     /* block offset is the offset in vhdx logical sectors, in
1061      * the payload data block. Convert that to a byte offset
1062      * in the block, and add in the payload data block offset
1063      * in the file, in bytes, to get the final read address */
1064 
1065     sinfo->file_offset += sinfo->block_offset;
1066 }
1067 
1068 
1069 static int vhdx_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1070 {
1071     BDRVVHDXState *s = bs->opaque;
1072 
1073     bdi->cluster_size = s->block_size;
1074 
1075     bdi->unallocated_blocks_are_zero =
1076         (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) == 0;
1077 
1078     return 0;
1079 }
1080 
1081 
1082 static coroutine_fn int vhdx_co_readv(BlockDriverState *bs, int64_t sector_num,
1083                                       int nb_sectors, QEMUIOVector *qiov)
1084 {
1085     BDRVVHDXState *s = bs->opaque;
1086     int ret = 0;
1087     VHDXSectorInfo sinfo;
1088     uint64_t bytes_done = 0;
1089     QEMUIOVector hd_qiov;
1090 
1091     qemu_iovec_init(&hd_qiov, qiov->niov);
1092 
1093     qemu_co_mutex_lock(&s->lock);
1094 
1095     while (nb_sectors > 0) {
1096         /* We are a differencing file, so we need to inspect the sector bitmap
1097          * to see if we have the data or not */
1098         if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1099             /* not supported yet */
1100             ret = -ENOTSUP;
1101             goto exit;
1102         } else {
1103             vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1104 
1105             qemu_iovec_reset(&hd_qiov);
1106             qemu_iovec_concat(&hd_qiov, qiov,  bytes_done, sinfo.bytes_avail);
1107 
1108             /* check the payload block state */
1109             switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) {
1110             case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1111             case PAYLOAD_BLOCK_UNDEFINED:   /* fall through */
1112             case PAYLOAD_BLOCK_UNMAPPED:    /* fall through */
1113             case PAYLOAD_BLOCK_ZERO:
1114                 /* return zero */
1115                 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail);
1116                 break;
1117             case PAYLOAD_BLOCK_FULLY_PRESENT:
1118                 qemu_co_mutex_unlock(&s->lock);
1119                 ret = bdrv_co_readv(bs->file,
1120                                     sinfo.file_offset >> BDRV_SECTOR_BITS,
1121                                     sinfo.sectors_avail, &hd_qiov);
1122                 qemu_co_mutex_lock(&s->lock);
1123                 if (ret < 0) {
1124                     goto exit;
1125                 }
1126                 break;
1127             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1128                 /* we don't yet support difference files, fall through
1129                  * to error */
1130             default:
1131                 ret = -EIO;
1132                 goto exit;
1133                 break;
1134             }
1135             nb_sectors -= sinfo.sectors_avail;
1136             sector_num += sinfo.sectors_avail;
1137             bytes_done += sinfo.bytes_avail;
1138         }
1139     }
1140     ret = 0;
1141 exit:
1142     qemu_co_mutex_unlock(&s->lock);
1143     qemu_iovec_destroy(&hd_qiov);
1144     return ret;
1145 }
1146 
1147 /*
1148  * Allocate a new payload block at the end of the file.
1149  *
1150  * Allocation will happen at 1MB alignment inside the file
1151  *
1152  * Returns the file offset start of the new payload block
1153  */
1154 static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s,
1155                                     uint64_t *new_offset)
1156 {
1157     *new_offset = bdrv_getlength(bs->file);
1158 
1159     /* per the spec, the address for a block is in units of 1MB */
1160     *new_offset = ROUND_UP(*new_offset, 1024 * 1024);
1161 
1162     return bdrv_truncate(bs->file, *new_offset + s->block_size);
1163 }
1164 
1165 /*
1166  * Update the BAT table entry with the new file offset, and the new entry
1167  * state */
1168 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s,
1169                                        VHDXSectorInfo *sinfo,
1170                                        uint64_t *bat_entry_le,
1171                                        uint64_t *bat_offset, int state)
1172 {
1173     /* The BAT entry is a uint64, with 44 bits for the file offset in units of
1174      * 1MB, and 3 bits for the block state. */
1175     s->bat[sinfo->bat_idx]  = sinfo->file_offset;
1176 
1177     s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK;
1178 
1179     *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]);
1180     *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry);
1181 
1182 }
1183 
1184 /* Per the spec, on the first write of guest-visible data to the file the
1185  * data write guid must be updated in the header */
1186 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s)
1187 {
1188     int ret = 0;
1189     if (s->first_visible_write) {
1190         s->first_visible_write = false;
1191         ret = vhdx_update_headers(bs, s, true, NULL);
1192     }
1193     return ret;
1194 }
1195 
1196 static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num,
1197                                       int nb_sectors, QEMUIOVector *qiov)
1198 {
1199     int ret = -ENOTSUP;
1200     BDRVVHDXState *s = bs->opaque;
1201     VHDXSectorInfo sinfo;
1202     uint64_t bytes_done = 0;
1203     uint64_t bat_entry = 0;
1204     uint64_t bat_entry_offset = 0;
1205     QEMUIOVector hd_qiov;
1206     struct iovec iov1 = { 0 };
1207     struct iovec iov2 = { 0 };
1208     int sectors_to_write;
1209     int bat_state;
1210     uint64_t bat_prior_offset = 0;
1211     bool bat_update = false;
1212 
1213     qemu_iovec_init(&hd_qiov, qiov->niov);
1214 
1215     qemu_co_mutex_lock(&s->lock);
1216 
1217     ret = vhdx_user_visible_write(bs, s);
1218     if (ret < 0) {
1219         goto exit;
1220     }
1221 
1222     while (nb_sectors > 0) {
1223         bool use_zero_buffers = false;
1224         bat_update = false;
1225         if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1226             /* not supported yet */
1227             ret = -ENOTSUP;
1228             goto exit;
1229         } else {
1230             vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1231             sectors_to_write = sinfo.sectors_avail;
1232 
1233             qemu_iovec_reset(&hd_qiov);
1234             /* check the payload block state */
1235             bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK;
1236             switch (bat_state) {
1237             case PAYLOAD_BLOCK_ZERO:
1238                 /* in this case, we need to preserve zero writes for
1239                  * data that is not part of this write, so we must pad
1240                  * the rest of the buffer to zeroes */
1241 
1242                 /* if we are on a posix system with ftruncate() that extends
1243                  * a file, then it is zero-filled for us.  On Win32, the raw
1244                  * layer uses SetFilePointer and SetFileEnd, which does not
1245                  * zero fill AFAIK */
1246 
1247                 /* Queue another write of zero buffers if the underlying file
1248                  * does not zero-fill on file extension */
1249 
1250                 if (bdrv_has_zero_init(bs->file) == 0) {
1251                     use_zero_buffers = true;
1252 
1253                     /* zero fill the front, if any */
1254                     if (sinfo.block_offset) {
1255                         iov1.iov_len = sinfo.block_offset;
1256                         iov1.iov_base = qemu_blockalign(bs, iov1.iov_len);
1257                         memset(iov1.iov_base, 0, iov1.iov_len);
1258                         qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0,
1259                                               sinfo.block_offset);
1260                         sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS;
1261                     }
1262 
1263                     /* our actual data */
1264                     qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1265                                       sinfo.bytes_avail);
1266 
1267                     /* zero fill the back, if any */
1268                     if ((sinfo.bytes_avail - sinfo.block_offset) <
1269                          s->block_size) {
1270                         iov2.iov_len = s->block_size -
1271                                       (sinfo.bytes_avail + sinfo.block_offset);
1272                         iov2.iov_base = qemu_blockalign(bs, iov2.iov_len);
1273                         memset(iov2.iov_base, 0, iov2.iov_len);
1274                         qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0,
1275                                               sinfo.block_offset);
1276                         sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS;
1277                     }
1278                 }
1279 
1280                 /* fall through */
1281             case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1282             case PAYLOAD_BLOCK_UNMAPPED:    /* fall through */
1283             case PAYLOAD_BLOCK_UNDEFINED:   /* fall through */
1284                 bat_prior_offset = sinfo.file_offset;
1285                 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset);
1286                 if (ret < 0) {
1287                     goto exit;
1288                 }
1289                 /* once we support differencing files, this may also be
1290                  * partially present */
1291                 /* update block state to the newly specified state */
1292                 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1293                                             &bat_entry_offset,
1294                                             PAYLOAD_BLOCK_FULLY_PRESENT);
1295                 bat_update = true;
1296                 /* since we just allocated a block, file_offset is the
1297                  * beginning of the payload block. It needs to be the
1298                  * write address, which includes the offset into the block */
1299                 if (!use_zero_buffers) {
1300                     sinfo.file_offset += sinfo.block_offset;
1301                 }
1302                 /* fall through */
1303             case PAYLOAD_BLOCK_FULLY_PRESENT:
1304                 /* if the file offset address is in the header zone,
1305                  * there is a problem */
1306                 if (sinfo.file_offset < (1024 * 1024)) {
1307                     ret = -EFAULT;
1308                     goto error_bat_restore;
1309                 }
1310 
1311                 if (!use_zero_buffers) {
1312                     qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1313                                       sinfo.bytes_avail);
1314                 }
1315                 /* block exists, so we can just overwrite it */
1316                 qemu_co_mutex_unlock(&s->lock);
1317                 ret = bdrv_co_writev(bs->file,
1318                                     sinfo.file_offset >> BDRV_SECTOR_BITS,
1319                                     sectors_to_write, &hd_qiov);
1320                 qemu_co_mutex_lock(&s->lock);
1321                 if (ret < 0) {
1322                     goto error_bat_restore;
1323                 }
1324                 break;
1325             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1326                 /* we don't yet support difference files, fall through
1327                  * to error */
1328             default:
1329                 ret = -EIO;
1330                 goto exit;
1331                 break;
1332             }
1333 
1334             if (bat_update) {
1335                 /* this will update the BAT entry into the log journal, and
1336                  * then flush the log journal out to disk */
1337                 ret =  vhdx_log_write_and_flush(bs, s, &bat_entry,
1338                                                 sizeof(VHDXBatEntry),
1339                                                 bat_entry_offset);
1340                 if (ret < 0) {
1341                     goto exit;
1342                 }
1343             }
1344 
1345             nb_sectors -= sinfo.sectors_avail;
1346             sector_num += sinfo.sectors_avail;
1347             bytes_done += sinfo.bytes_avail;
1348 
1349         }
1350     }
1351 
1352     goto exit;
1353 
1354 error_bat_restore:
1355     if (bat_update) {
1356         /* keep metadata in sync, and restore the bat entry state
1357          * if error. */
1358         sinfo.file_offset = bat_prior_offset;
1359         vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1360                                     &bat_entry_offset, bat_state);
1361     }
1362 exit:
1363     qemu_vfree(iov1.iov_base);
1364     qemu_vfree(iov2.iov_base);
1365     qemu_co_mutex_unlock(&s->lock);
1366     qemu_iovec_destroy(&hd_qiov);
1367     return ret;
1368 }
1369 
1370 
1371 
1372 /*
1373  * Create VHDX Headers
1374  *
1375  * There are 2 headers, and the highest sequence number will represent
1376  * the active header
1377  */
1378 static int vhdx_create_new_headers(BlockDriverState *bs, uint64_t image_size,
1379                                    uint32_t log_size)
1380 {
1381     int ret = 0;
1382     VHDXHeader *hdr = NULL;
1383 
1384     hdr = g_new0(VHDXHeader, 1);
1385 
1386     hdr->signature       = VHDX_HEADER_SIGNATURE;
1387     hdr->sequence_number = g_random_int();
1388     hdr->log_version     = 0;
1389     hdr->version         = 1;
1390     hdr->log_length      = log_size;
1391     hdr->log_offset      = VHDX_HEADER_SECTION_END;
1392     vhdx_guid_generate(&hdr->file_write_guid);
1393     vhdx_guid_generate(&hdr->data_write_guid);
1394 
1395     ret = vhdx_write_header(bs, hdr, VHDX_HEADER1_OFFSET, false);
1396     if (ret < 0) {
1397         goto exit;
1398     }
1399     hdr->sequence_number++;
1400     ret = vhdx_write_header(bs, hdr, VHDX_HEADER2_OFFSET, false);
1401     if (ret < 0) {
1402         goto exit;
1403     }
1404 
1405 exit:
1406     g_free(hdr);
1407     return ret;
1408 }
1409 
1410 
1411 /*
1412  * Create the Metadata entries.
1413  *
1414  * For more details on the entries, see section 3.5 (pg 29) in the
1415  * VHDX 1.00 specification.
1416  *
1417  * We support 5 metadata entries (all required by spec):
1418  *          File Parameters,
1419  *          Virtual Disk Size,
1420  *          Page 83 Data,
1421  *          Logical Sector Size,
1422  *          Physical Sector Size
1423  *
1424  * The first 64KB of the Metadata section is reserved for the metadata
1425  * header and entries; beyond that, the metadata items themselves reside.
1426  */
1427 static int vhdx_create_new_metadata(BlockDriverState *bs,
1428                                     uint64_t image_size,
1429                                     uint32_t block_size,
1430                                     uint32_t sector_size,
1431                                     uint64_t metadata_offset,
1432                                     VHDXImageType type)
1433 {
1434     int ret = 0;
1435     uint32_t offset = 0;
1436     void *buffer = NULL;
1437     void *entry_buffer;
1438     VHDXMetadataTableHeader *md_table;;
1439     VHDXMetadataTableEntry  *md_table_entry;
1440 
1441     /* Metadata entries */
1442     VHDXFileParameters     *mt_file_params;
1443     VHDXVirtualDiskSize    *mt_virtual_size;
1444     VHDXPage83Data         *mt_page83;
1445     VHDXVirtualDiskLogicalSectorSize  *mt_log_sector_size;
1446     VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size;
1447 
1448     entry_buffer = g_malloc0(sizeof(VHDXFileParameters)               +
1449                              sizeof(VHDXVirtualDiskSize)              +
1450                              sizeof(VHDXPage83Data)                   +
1451                              sizeof(VHDXVirtualDiskLogicalSectorSize) +
1452                              sizeof(VHDXVirtualDiskPhysicalSectorSize));
1453 
1454     mt_file_params = entry_buffer;
1455     offset += sizeof(VHDXFileParameters);
1456     mt_virtual_size = entry_buffer + offset;
1457     offset += sizeof(VHDXVirtualDiskSize);
1458     mt_page83 = entry_buffer + offset;
1459     offset += sizeof(VHDXPage83Data);
1460     mt_log_sector_size = entry_buffer + offset;
1461     offset += sizeof(VHDXVirtualDiskLogicalSectorSize);
1462     mt_phys_sector_size = entry_buffer + offset;
1463 
1464     mt_file_params->block_size = cpu_to_le32(block_size);
1465     if (type == VHDX_TYPE_FIXED) {
1466         mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED;
1467         cpu_to_le32s(&mt_file_params->data_bits);
1468     }
1469 
1470     vhdx_guid_generate(&mt_page83->page_83_data);
1471     cpu_to_leguids(&mt_page83->page_83_data);
1472     mt_virtual_size->virtual_disk_size        = cpu_to_le64(image_size);
1473     mt_log_sector_size->logical_sector_size   = cpu_to_le32(sector_size);
1474     mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size);
1475 
1476     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1477     md_table = buffer;
1478 
1479     md_table->signature   = VHDX_METADATA_SIGNATURE;
1480     md_table->entry_count = 5;
1481     vhdx_metadata_header_le_export(md_table);
1482 
1483 
1484     /* This will reference beyond the reserved table portion */
1485     offset = 64 * KiB;
1486 
1487     md_table_entry = buffer + sizeof(VHDXMetadataTableHeader);
1488 
1489     md_table_entry[0].item_id = file_param_guid;
1490     md_table_entry[0].offset  = offset;
1491     md_table_entry[0].length  = sizeof(VHDXFileParameters);
1492     md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED;
1493     offset += md_table_entry[0].length;
1494     vhdx_metadata_entry_le_export(&md_table_entry[0]);
1495 
1496     md_table_entry[1].item_id = virtual_size_guid;
1497     md_table_entry[1].offset  = offset;
1498     md_table_entry[1].length  = sizeof(VHDXVirtualDiskSize);
1499     md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1500                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1501     offset += md_table_entry[1].length;
1502     vhdx_metadata_entry_le_export(&md_table_entry[1]);
1503 
1504     md_table_entry[2].item_id = page83_guid;
1505     md_table_entry[2].offset  = offset;
1506     md_table_entry[2].length  = sizeof(VHDXPage83Data);
1507     md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1508                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1509     offset += md_table_entry[2].length;
1510     vhdx_metadata_entry_le_export(&md_table_entry[2]);
1511 
1512     md_table_entry[3].item_id = logical_sector_guid;
1513     md_table_entry[3].offset  = offset;
1514     md_table_entry[3].length  = sizeof(VHDXVirtualDiskLogicalSectorSize);
1515     md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1516                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1517     offset += md_table_entry[3].length;
1518     vhdx_metadata_entry_le_export(&md_table_entry[3]);
1519 
1520     md_table_entry[4].item_id = phys_sector_guid;
1521     md_table_entry[4].offset  = offset;
1522     md_table_entry[4].length  = sizeof(VHDXVirtualDiskPhysicalSectorSize);
1523     md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1524                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1525     vhdx_metadata_entry_le_export(&md_table_entry[4]);
1526 
1527     ret = bdrv_pwrite(bs, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE);
1528     if (ret < 0) {
1529         goto exit;
1530     }
1531 
1532     ret = bdrv_pwrite(bs, metadata_offset + (64 * KiB), entry_buffer,
1533                       VHDX_HEADER_BLOCK_SIZE);
1534     if (ret < 0) {
1535         goto exit;
1536     }
1537 
1538 
1539 exit:
1540     g_free(buffer);
1541     g_free(entry_buffer);
1542     return ret;
1543 }
1544 
1545 /* This create the actual BAT itself.  We currently only support
1546  * 'Dynamic' and 'Fixed' image types.
1547  *
1548  *  Dynamic images: default state of the BAT is all zeroes.
1549  *
1550  *  Fixed images: default state of the BAT is fully populated, with
1551  *                file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT.
1552  */
1553 static int vhdx_create_bat(BlockDriverState *bs, BDRVVHDXState *s,
1554                            uint64_t image_size, VHDXImageType type,
1555                            bool use_zero_blocks, uint64_t file_offset,
1556                            uint32_t length)
1557 {
1558     int ret = 0;
1559     uint64_t data_file_offset;
1560     uint64_t total_sectors = 0;
1561     uint64_t sector_num = 0;
1562     uint64_t unused;
1563     int block_state;
1564     VHDXSectorInfo sinfo;
1565 
1566     assert(s->bat == NULL);
1567 
1568     /* this gives a data start after BAT/bitmap entries, and well
1569      * past any metadata entries (with a 4 MB buffer for future
1570      * expansion */
1571     data_file_offset = file_offset + length + 5 * MiB;
1572     total_sectors = image_size >> s->logical_sector_size_bits;
1573 
1574     if (type == VHDX_TYPE_DYNAMIC) {
1575         /* All zeroes, so we can just extend the file - the end of the BAT
1576          * is the furthest thing we have written yet */
1577         ret = bdrv_truncate(bs, data_file_offset);
1578         if (ret < 0) {
1579             goto exit;
1580         }
1581     } else if (type == VHDX_TYPE_FIXED) {
1582         ret = bdrv_truncate(bs, data_file_offset + image_size);
1583         if (ret < 0) {
1584             goto exit;
1585         }
1586     } else {
1587         ret = -ENOTSUP;
1588         goto exit;
1589     }
1590 
1591     if (type == VHDX_TYPE_FIXED ||
1592                 use_zero_blocks ||
1593                 bdrv_has_zero_init(bs) == 0) {
1594         /* for a fixed file, the default BAT entry is not zero */
1595         s->bat = g_try_malloc0(length);
1596         if (length && s->bat != NULL) {
1597             ret = -ENOMEM;
1598             goto exit;
1599         }
1600         block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT :
1601                                                 PAYLOAD_BLOCK_NOT_PRESENT;
1602         block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state;
1603         /* fill the BAT by emulating sector writes of sectors_per_block size */
1604         while (sector_num < total_sectors) {
1605             vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo);
1606             sinfo.file_offset = data_file_offset +
1607                                 (sector_num << s->logical_sector_size_bits);
1608             sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB);
1609             vhdx_update_bat_table_entry(bs, s, &sinfo, &unused, &unused,
1610                                         block_state);
1611             cpu_to_le64s(&s->bat[sinfo.bat_idx]);
1612             sector_num += s->sectors_per_block;
1613         }
1614         ret = bdrv_pwrite(bs, file_offset, s->bat, length);
1615         if (ret < 0) {
1616             goto exit;
1617         }
1618     }
1619 
1620 
1621 
1622 exit:
1623     g_free(s->bat);
1624     return ret;
1625 }
1626 
1627 /* Creates the region table header, and region table entries.
1628  * There are 2 supported region table entries: BAT, and Metadata/
1629  *
1630  * As the calculations for the BAT region table are also needed
1631  * to create the BAT itself, we will also cause the BAT to be
1632  * created.
1633  */
1634 static int vhdx_create_new_region_table(BlockDriverState *bs,
1635                                         uint64_t image_size,
1636                                         uint32_t block_size,
1637                                         uint32_t sector_size,
1638                                         uint32_t log_size,
1639                                         bool use_zero_blocks,
1640                                         VHDXImageType type,
1641                                         uint64_t *metadata_offset)
1642 {
1643     int ret = 0;
1644     uint32_t offset = 0;
1645     void *buffer = NULL;
1646     uint64_t bat_file_offset;
1647     uint32_t bat_length;
1648     BDRVVHDXState *s = NULL;
1649     VHDXRegionTableHeader *region_table;
1650     VHDXRegionTableEntry *rt_bat;
1651     VHDXRegionTableEntry *rt_metadata;
1652 
1653     assert(metadata_offset != NULL);
1654 
1655     /* Populate enough of the BDRVVHDXState to be able to use the
1656      * pre-existing BAT calculation, translation, and update functions */
1657     s = g_new0(BDRVVHDXState, 1);
1658 
1659     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
1660                      (uint64_t) sector_size / (uint64_t) block_size;
1661 
1662     s->sectors_per_block = block_size / sector_size;
1663     s->virtual_disk_size = image_size;
1664     s->block_size = block_size;
1665     s->logical_sector_size = sector_size;
1666 
1667     vhdx_set_shift_bits(s);
1668 
1669     vhdx_calc_bat_entries(s);
1670 
1671     /* At this point the VHDX state is populated enough for creation */
1672 
1673     /* a single buffer is used so we can calculate the checksum over the
1674      * entire 64KB block */
1675     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1676     region_table = buffer;
1677     offset += sizeof(VHDXRegionTableHeader);
1678     rt_bat = buffer + offset;
1679     offset += sizeof(VHDXRegionTableEntry);
1680     rt_metadata  = buffer + offset;
1681 
1682     region_table->signature = VHDX_REGION_SIGNATURE;
1683     region_table->entry_count = 2;   /* BAT and Metadata */
1684 
1685     rt_bat->guid        = bat_guid;
1686     rt_bat->length      = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB);
1687     rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB);
1688     s->bat_offset = rt_bat->file_offset;
1689 
1690     rt_metadata->guid        = metadata_guid;
1691     rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length,
1692                                         MiB);
1693     rt_metadata->length      = 1 * MiB; /* min size, and more than enough */
1694     *metadata_offset = rt_metadata->file_offset;
1695 
1696     bat_file_offset = rt_bat->file_offset;
1697     bat_length = rt_bat->length;
1698 
1699     vhdx_region_header_le_export(region_table);
1700     vhdx_region_entry_le_export(rt_bat);
1701     vhdx_region_entry_le_export(rt_metadata);
1702 
1703     vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE,
1704                          offsetof(VHDXRegionTableHeader, checksum));
1705 
1706 
1707     /* The region table gives us the data we need to create the BAT,
1708      * so do that now */
1709     ret = vhdx_create_bat(bs, s, image_size, type, use_zero_blocks,
1710                           bat_file_offset, bat_length);
1711     if (ret < 0) {
1712         goto exit;
1713     }
1714 
1715     /* Now write out the region headers to disk */
1716     ret = bdrv_pwrite(bs, VHDX_REGION_TABLE_OFFSET, buffer,
1717                       VHDX_HEADER_BLOCK_SIZE);
1718     if (ret < 0) {
1719         goto exit;
1720     }
1721 
1722     ret = bdrv_pwrite(bs, VHDX_REGION_TABLE2_OFFSET, buffer,
1723                       VHDX_HEADER_BLOCK_SIZE);
1724     if (ret < 0) {
1725         goto exit;
1726     }
1727 
1728 
1729 exit:
1730     g_free(s);
1731     g_free(buffer);
1732     return ret;
1733 }
1734 
1735 /* We need to create the following elements:
1736  *
1737  *    .-----------------------------------------------------------------.
1738  *    |   (A)    |   (B)    |    (C)    |     (D)       |     (E)       |
1739  *    |  File ID |  Header1 |  Header 2 |  Region Tbl 1 |  Region Tbl 2 |
1740  *    |          |          |           |               |               |
1741  *    .-----------------------------------------------------------------.
1742  *    0         64KB      128KB       192KB           256KB           320KB
1743  *
1744  *
1745  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1746  *    |     (F)     |     (G)       |    (H)    |                        |
1747  *    | Journal Log |  BAT / Bitmap |  Metadata |  .... data ......      |
1748  *    |             |               |           |                        |
1749  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1750  *   1MB
1751  */
1752 static int vhdx_create(const char *filename, QemuOpts *opts, Error **errp)
1753 {
1754     int ret = 0;
1755     uint64_t image_size = (uint64_t) 2 * GiB;
1756     uint32_t log_size   = 1 * MiB;
1757     uint32_t block_size = 0;
1758     uint64_t signature;
1759     uint64_t metadata_offset;
1760     bool use_zero_blocks = false;
1761 
1762     gunichar2 *creator = NULL;
1763     glong creator_items;
1764     BlockDriverState *bs;
1765     char *type = NULL;
1766     VHDXImageType image_type;
1767     Error *local_err = NULL;
1768 
1769     image_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
1770     log_size = qemu_opt_get_size_del(opts, VHDX_BLOCK_OPT_LOG_SIZE, 0);
1771     block_size = qemu_opt_get_size_del(opts, VHDX_BLOCK_OPT_BLOCK_SIZE, 0);
1772     type = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1773     use_zero_blocks = qemu_opt_get_bool_del(opts, VHDX_BLOCK_OPT_ZERO, false);
1774 
1775     if (image_size > VHDX_MAX_IMAGE_SIZE) {
1776         error_setg_errno(errp, EINVAL, "Image size too large; max of 64TB");
1777         ret = -EINVAL;
1778         goto exit;
1779     }
1780 
1781     if (type == NULL) {
1782         type = g_strdup("dynamic");
1783     }
1784 
1785     if (!strcmp(type, "dynamic")) {
1786         image_type = VHDX_TYPE_DYNAMIC;
1787     } else if (!strcmp(type, "fixed")) {
1788         image_type = VHDX_TYPE_FIXED;
1789     } else if (!strcmp(type, "differencing")) {
1790         error_setg_errno(errp, ENOTSUP,
1791                          "Differencing files not yet supported");
1792         ret = -ENOTSUP;
1793         goto exit;
1794     } else {
1795         ret = -EINVAL;
1796         goto exit;
1797     }
1798 
1799     /* These are pretty arbitrary, and mainly designed to keep the BAT
1800      * size reasonable to load into RAM */
1801     if (block_size == 0) {
1802         if (image_size > 32 * TiB) {
1803             block_size = 64 * MiB;
1804         } else if (image_size > (uint64_t) 100 * GiB) {
1805             block_size = 32 * MiB;
1806         } else if (image_size > 1 * GiB) {
1807             block_size = 16 * MiB;
1808         } else {
1809             block_size = 8 * MiB;
1810         }
1811     }
1812 
1813 
1814     /* make the log size close to what was specified, but must be
1815      * min 1MB, and multiple of 1MB */
1816     log_size = ROUND_UP(log_size, MiB);
1817 
1818     block_size = ROUND_UP(block_size, MiB);
1819     block_size = block_size > VHDX_BLOCK_SIZE_MAX ? VHDX_BLOCK_SIZE_MAX :
1820                                                     block_size;
1821 
1822     ret = bdrv_create_file(filename, opts, &local_err);
1823     if (ret < 0) {
1824         error_propagate(errp, local_err);
1825         goto exit;
1826     }
1827 
1828     bs = NULL;
1829     ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1830                     NULL, &local_err);
1831     if (ret < 0) {
1832         error_propagate(errp, local_err);
1833         goto exit;
1834     }
1835 
1836     /* Create (A) */
1837 
1838     /* The creator field is optional, but may be useful for
1839      * debugging / diagnostics */
1840     creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL,
1841                               &creator_items, NULL);
1842     signature = cpu_to_le64(VHDX_FILE_SIGNATURE);
1843     ret = bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature));
1844     if (ret < 0) {
1845         goto delete_and_exit;
1846     }
1847     if (creator) {
1848         ret = bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET + sizeof(signature),
1849                           creator, creator_items * sizeof(gunichar2));
1850         if (ret < 0) {
1851             goto delete_and_exit;
1852         }
1853     }
1854 
1855 
1856     /* Creates (B),(C) */
1857     ret = vhdx_create_new_headers(bs, image_size, log_size);
1858     if (ret < 0) {
1859         goto delete_and_exit;
1860     }
1861 
1862     /* Creates (D),(E),(G) explicitly. (F) created as by-product */
1863     ret = vhdx_create_new_region_table(bs, image_size, block_size, 512,
1864                                        log_size, use_zero_blocks, image_type,
1865                                        &metadata_offset);
1866     if (ret < 0) {
1867         goto delete_and_exit;
1868     }
1869 
1870     /* Creates (H) */
1871     ret = vhdx_create_new_metadata(bs, image_size, block_size, 512,
1872                                    metadata_offset, image_type);
1873     if (ret < 0) {
1874         goto delete_and_exit;
1875     }
1876 
1877 
1878 
1879 delete_and_exit:
1880     bdrv_unref(bs);
1881 exit:
1882     g_free(type);
1883     g_free(creator);
1884     return ret;
1885 }
1886 
1887 /* If opened r/w, the VHDX driver will automatically replay the log,
1888  * if one is present, inside the vhdx_open() call.
1889  *
1890  * If qemu-img check -r all is called, the image is automatically opened
1891  * r/w and any log has already been replayed, so there is nothing (currently)
1892  * for us to do here
1893  */
1894 static int vhdx_check(BlockDriverState *bs, BdrvCheckResult *result,
1895                        BdrvCheckMode fix)
1896 {
1897     BDRVVHDXState *s = bs->opaque;
1898 
1899     if (s->log_replayed_on_open) {
1900         result->corruptions_fixed++;
1901     }
1902     return 0;
1903 }
1904 
1905 static QemuOptsList vhdx_create_opts = {
1906     .name = "vhdx-create-opts",
1907     .head = QTAILQ_HEAD_INITIALIZER(vhdx_create_opts.head),
1908     .desc = {
1909         {
1910            .name = BLOCK_OPT_SIZE,
1911            .type = QEMU_OPT_SIZE,
1912            .help = "Virtual disk size; max of 64TB."
1913        },
1914        {
1915            .name = VHDX_BLOCK_OPT_LOG_SIZE,
1916            .type = QEMU_OPT_SIZE,
1917            .def_value_str = stringify(DEFAULT_LOG_SIZE),
1918            .help = "Log size; min 1MB."
1919        },
1920        {
1921            .name = VHDX_BLOCK_OPT_BLOCK_SIZE,
1922            .type = QEMU_OPT_SIZE,
1923            .def_value_str = stringify(0),
1924            .help = "Block Size; min 1MB, max 256MB. " \
1925                    "0 means auto-calculate based on image size."
1926        },
1927        {
1928            .name = BLOCK_OPT_SUBFMT,
1929            .type = QEMU_OPT_STRING,
1930            .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "\
1931                    "Default is 'dynamic'."
1932        },
1933        {
1934            .name = VHDX_BLOCK_OPT_ZERO,
1935            .type = QEMU_OPT_BOOL,
1936            .help = "Force use of payload blocks of type 'ZERO'.  Non-standard."
1937        },
1938        { NULL }
1939     }
1940 };
1941 
1942 static BlockDriver bdrv_vhdx = {
1943     .format_name            = "vhdx",
1944     .instance_size          = sizeof(BDRVVHDXState),
1945     .bdrv_probe             = vhdx_probe,
1946     .bdrv_open              = vhdx_open,
1947     .bdrv_close             = vhdx_close,
1948     .bdrv_reopen_prepare    = vhdx_reopen_prepare,
1949     .bdrv_co_readv          = vhdx_co_readv,
1950     .bdrv_co_writev         = vhdx_co_writev,
1951     .bdrv_create            = vhdx_create,
1952     .bdrv_get_info          = vhdx_get_info,
1953     .bdrv_check             = vhdx_check,
1954 
1955     .create_opts            = &vhdx_create_opts,
1956 };
1957 
1958 static void bdrv_vhdx_init(void)
1959 {
1960     bdrv_register(&bdrv_vhdx);
1961 }
1962 
1963 block_init(bdrv_vhdx_init);
1964