xref: /openbmc/qemu/block/vhdx.c (revision 416296a9)
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/osdep.h"
19 #include "qapi/error.h"
20 #include "qemu-common.h"
21 #include "block/block_int.h"
22 #include "sysemu/block-backend.h"
23 #include "qemu/module.h"
24 #include "qemu/crc32c.h"
25 #include "qemu/bswap.h"
26 #include "block/vhdx.h"
27 #include "migration/migration.h"
28 
29 #include <uuid/uuid.h>
30 
31 /* Options for VHDX creation */
32 
33 #define VHDX_BLOCK_OPT_LOG_SIZE   "log_size"
34 #define VHDX_BLOCK_OPT_BLOCK_SIZE "block_size"
35 #define VHDX_BLOCK_OPT_ZERO "block_state_zero"
36 
37 typedef enum VHDXImageType {
38     VHDX_TYPE_DYNAMIC = 0,
39     VHDX_TYPE_FIXED,
40     VHDX_TYPE_DIFFERENCING,   /* Currently unsupported */
41 } VHDXImageType;
42 
43 /* Several metadata and region table data entries are identified by
44  * guids in  a MS-specific GUID format. */
45 
46 
47 /* ------- Known Region Table GUIDs ---------------------- */
48 static const MSGUID bat_guid =      { .data1 = 0x2dc27766,
49                                       .data2 = 0xf623,
50                                       .data3 = 0x4200,
51                                       .data4 = { 0x9d, 0x64, 0x11, 0x5e,
52                                                  0x9b, 0xfd, 0x4a, 0x08} };
53 
54 static const MSGUID metadata_guid = { .data1 = 0x8b7ca206,
55                                       .data2 = 0x4790,
56                                       .data3 = 0x4b9a,
57                                       .data4 = { 0xb8, 0xfe, 0x57, 0x5f,
58                                                  0x05, 0x0f, 0x88, 0x6e} };
59 
60 
61 
62 /* ------- Known Metadata Entry GUIDs ---------------------- */
63 static const MSGUID file_param_guid =   { .data1 = 0xcaa16737,
64                                           .data2 = 0xfa36,
65                                           .data3 = 0x4d43,
66                                           .data4 = { 0xb3, 0xb6, 0x33, 0xf0,
67                                                      0xaa, 0x44, 0xe7, 0x6b} };
68 
69 static const MSGUID virtual_size_guid = { .data1 = 0x2FA54224,
70                                           .data2 = 0xcd1b,
71                                           .data3 = 0x4876,
72                                           .data4 = { 0xb2, 0x11, 0x5d, 0xbe,
73                                                      0xd8, 0x3b, 0xf4, 0xb8} };
74 
75 static const MSGUID page83_guid =       { .data1 = 0xbeca12ab,
76                                           .data2 = 0xb2e6,
77                                           .data3 = 0x4523,
78                                           .data4 = { 0x93, 0xef, 0xc3, 0x09,
79                                                      0xe0, 0x00, 0xc7, 0x46} };
80 
81 
82 static const MSGUID phys_sector_guid =  { .data1 = 0xcda348c7,
83                                           .data2 = 0x445d,
84                                           .data3 = 0x4471,
85                                           .data4 = { 0x9c, 0xc9, 0xe9, 0x88,
86                                                      0x52, 0x51, 0xc5, 0x56} };
87 
88 static const MSGUID parent_locator_guid = { .data1 = 0xa8d35f2d,
89                                             .data2 = 0xb30b,
90                                             .data3 = 0x454d,
91                                             .data4 = { 0xab, 0xf7, 0xd3,
92                                                        0xd8, 0x48, 0x34,
93                                                        0xab, 0x0c} };
94 
95 static const MSGUID logical_sector_guid = { .data1 = 0x8141bf1d,
96                                             .data2 = 0xa96f,
97                                             .data3 = 0x4709,
98                                             .data4 = { 0xba, 0x47, 0xf2,
99                                                        0x33, 0xa8, 0xfa,
100                                                        0xab, 0x5f} };
101 
102 /* Each parent type must have a valid GUID; this is for parent images
103  * of type 'VHDX'.  If we were to allow e.g. a QCOW2 parent, we would
104  * need to make up our own QCOW2 GUID type */
105 static const MSGUID parent_vhdx_guid __attribute__((unused))
106                                      = { .data1 = 0xb04aefb7,
107                                          .data2 = 0xd19e,
108                                          .data3 = 0x4a81,
109                                          .data4 = { 0xb7, 0x89, 0x25, 0xb8,
110                                                     0xe9, 0x44, 0x59, 0x13} };
111 
112 
113 #define META_FILE_PARAMETER_PRESENT      0x01
114 #define META_VIRTUAL_DISK_SIZE_PRESENT   0x02
115 #define META_PAGE_83_PRESENT             0x04
116 #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08
117 #define META_PHYS_SECTOR_SIZE_PRESENT    0x10
118 #define META_PARENT_LOCATOR_PRESENT      0x20
119 
120 #define META_ALL_PRESENT    \
121     (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \
122      META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \
123      META_PHYS_SECTOR_SIZE_PRESENT)
124 
125 
126 typedef struct VHDXSectorInfo {
127     uint32_t bat_idx;       /* BAT entry index */
128     uint32_t sectors_avail; /* sectors available in payload block */
129     uint32_t bytes_left;    /* bytes left in the block after data to r/w */
130     uint32_t bytes_avail;   /* bytes available in payload block */
131     uint64_t file_offset;   /* absolute offset in bytes, in file */
132     uint64_t block_offset;  /* block offset, in bytes */
133 } VHDXSectorInfo;
134 
135 /* Calculates new checksum.
136  *
137  * Zero is substituted during crc calculation for the original crc field
138  * crc_offset: byte offset in buf of the buffer crc
139  * buf: buffer pointer
140  * size: size of buffer (must be > crc_offset+4)
141  *
142  * Note: The buffer should have all multi-byte data in little-endian format,
143  *       and the resulting checksum is in little endian format.
144  */
145 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset)
146 {
147     uint32_t crc;
148 
149     assert(buf != NULL);
150     assert(size > (crc_offset + sizeof(crc)));
151 
152     memset(buf + crc_offset, 0, sizeof(crc));
153     crc =  crc32c(0xffffffff, buf, size);
154     cpu_to_le32s(&crc);
155     memcpy(buf + crc_offset, &crc, sizeof(crc));
156 
157     return crc;
158 }
159 
160 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size,
161                             int crc_offset)
162 {
163     uint32_t crc_new;
164     uint32_t crc_orig;
165     assert(buf != NULL);
166 
167     if (crc_offset > 0) {
168         memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
169         memset(buf + crc_offset, 0, sizeof(crc_orig));
170     }
171 
172     crc_new = crc32c(crc, buf, size);
173     if (crc_offset > 0) {
174         memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig));
175     }
176 
177     return crc_new;
178 }
179 
180 /* Validates the checksum of the buffer, with an in-place CRC.
181  *
182  * Zero is substituted during crc calculation for the original crc field,
183  * and the crc field is restored afterwards.  But the buffer will be modifed
184  * during the calculation, so this may not be not suitable for multi-threaded
185  * use.
186  *
187  * crc_offset: byte offset in buf of the buffer crc
188  * buf: buffer pointer
189  * size: size of buffer (must be > crc_offset+4)
190  *
191  * returns true if checksum is valid, false otherwise
192  */
193 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset)
194 {
195     uint32_t crc_orig;
196     uint32_t crc;
197 
198     assert(buf != NULL);
199     assert(size > (crc_offset + 4));
200 
201     memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
202     crc_orig = le32_to_cpu(crc_orig);
203 
204     crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset);
205 
206     return crc == crc_orig;
207 }
208 
209 
210 /*
211  * This generates a UUID that is compliant with the MS GUIDs used
212  * in the VHDX spec (and elsewhere).
213  */
214 void vhdx_guid_generate(MSGUID *guid)
215 {
216     uuid_t uuid;
217     assert(guid != NULL);
218 
219     uuid_generate(uuid);
220     memcpy(guid, uuid, sizeof(MSGUID));
221 }
222 
223 /* Check for region overlaps inside the VHDX image */
224 static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length)
225 {
226     int ret = 0;
227     uint64_t end;
228     VHDXRegionEntry *r;
229 
230     end = start + length;
231     QLIST_FOREACH(r, &s->regions, entries) {
232         if (!((start >= r->end) || (end <= r->start))) {
233             ret = -EINVAL;
234             goto exit;
235         }
236     }
237 
238 exit:
239     return ret;
240 }
241 
242 /* Register a region for future checks */
243 static void vhdx_region_register(BDRVVHDXState *s,
244                                  uint64_t start, uint64_t length)
245 {
246     VHDXRegionEntry *r;
247 
248     r = g_malloc0(sizeof(*r));
249 
250     r->start = start;
251     r->end = start + length;
252 
253     QLIST_INSERT_HEAD(&s->regions, r, entries);
254 }
255 
256 /* Free all registered regions */
257 static void vhdx_region_unregister_all(BDRVVHDXState *s)
258 {
259     VHDXRegionEntry *r, *r_next;
260 
261     QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) {
262         QLIST_REMOVE(r, entries);
263         g_free(r);
264     }
265 }
266 
267 static void vhdx_set_shift_bits(BDRVVHDXState *s)
268 {
269     s->logical_sector_size_bits = ctz32(s->logical_sector_size);
270     s->sectors_per_block_bits =   ctz32(s->sectors_per_block);
271     s->chunk_ratio_bits =         ctz64(s->chunk_ratio);
272     s->block_size_bits =          ctz32(s->block_size);
273 }
274 
275 /*
276  * Per the MS VHDX Specification, for every VHDX file:
277  *      - The header section is fixed size - 1 MB
278  *      - The header section is always the first "object"
279  *      - The first 64KB of the header is the File Identifier
280  *      - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile")
281  *      - The following 512 bytes constitute a UTF-16 string identifiying the
282  *        software that created the file, and is optional and diagnostic only.
283  *
284  *  Therefore, we probe by looking for the vhdxfile signature "vhdxfile"
285  */
286 static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename)
287 {
288     if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) {
289         return 100;
290     }
291     return 0;
292 }
293 
294 /*
295  * Writes the header to the specified offset.
296  *
297  * This will optionally read in buffer data from disk (otherwise zero-fill),
298  * and then update the header checksum.  Header is converted to proper
299  * endianness before being written to the specified file offset
300  */
301 static int vhdx_write_header(BdrvChild *file, VHDXHeader *hdr,
302                              uint64_t offset, bool read)
303 {
304     BlockDriverState *bs_file = file->bs;
305     uint8_t *buffer = NULL;
306     int ret;
307     VHDXHeader *header_le;
308 
309     assert(bs_file != NULL);
310     assert(hdr != NULL);
311 
312     /* the header checksum is not over just the packed size of VHDXHeader,
313      * but rather over the entire 'reserved' range for the header, which is
314      * 4KB (VHDX_HEADER_SIZE). */
315 
316     buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE);
317     if (read) {
318         /* if true, we can't assume the extra reserved bytes are 0 */
319         ret = bdrv_pread(file, offset, buffer, VHDX_HEADER_SIZE);
320         if (ret < 0) {
321             goto exit;
322         }
323     } else {
324         memset(buffer, 0, VHDX_HEADER_SIZE);
325     }
326 
327     /* overwrite the actual VHDXHeader portion */
328     header_le = (VHDXHeader *)buffer;
329     memcpy(header_le, hdr, sizeof(VHDXHeader));
330     vhdx_header_le_export(hdr, header_le);
331     vhdx_update_checksum(buffer, VHDX_HEADER_SIZE,
332                          offsetof(VHDXHeader, checksum));
333     ret = bdrv_pwrite_sync(file, offset, header_le, sizeof(VHDXHeader));
334 
335 exit:
336     qemu_vfree(buffer);
337     return ret;
338 }
339 
340 /* Update the VHDX headers
341  *
342  * This follows the VHDX spec procedures for header updates.
343  *
344  *  - non-current header is updated with largest sequence number
345  */
346 static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s,
347                               bool generate_data_write_guid, MSGUID *log_guid)
348 {
349     int ret = 0;
350     int hdr_idx = 0;
351     uint64_t header_offset = VHDX_HEADER1_OFFSET;
352 
353     VHDXHeader *active_header;
354     VHDXHeader *inactive_header;
355 
356     /* operate on the non-current header */
357     if (s->curr_header == 0) {
358         hdr_idx = 1;
359         header_offset = VHDX_HEADER2_OFFSET;
360     }
361 
362     active_header   = s->headers[s->curr_header];
363     inactive_header = s->headers[hdr_idx];
364 
365     inactive_header->sequence_number = active_header->sequence_number + 1;
366 
367     /* a new file guid must be generated before any file write, including
368      * headers */
369     inactive_header->file_write_guid = s->session_guid;
370 
371     /* a new data guid only needs to be generated before any guest-visible
372      * writes (i.e. something observable via virtual disk read) */
373     if (generate_data_write_guid) {
374         vhdx_guid_generate(&inactive_header->data_write_guid);
375     }
376 
377     /* update the log guid if present */
378     if (log_guid) {
379         inactive_header->log_guid = *log_guid;
380     }
381 
382     ret = vhdx_write_header(bs->file, inactive_header, header_offset, true);
383     if (ret < 0) {
384         goto exit;
385     }
386     s->curr_header = hdr_idx;
387 
388 exit:
389     return ret;
390 }
391 
392 /*
393  * The VHDX spec calls for header updates to be performed twice, so that both
394  * the current and non-current header have valid info
395  */
396 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s,
397                         bool generate_data_write_guid, MSGUID *log_guid)
398 {
399     int ret;
400 
401     ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
402     if (ret < 0) {
403         return ret;
404     }
405     ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
406     return ret;
407 }
408 
409 /* opens the specified header block from the VHDX file header section */
410 static void vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s,
411                               Error **errp)
412 {
413     int ret;
414     VHDXHeader *header1;
415     VHDXHeader *header2;
416     bool h1_valid = false;
417     bool h2_valid = false;
418     uint64_t h1_seq = 0;
419     uint64_t h2_seq = 0;
420     uint8_t *buffer;
421 
422     /* header1 & header2 are freed in vhdx_close() */
423     header1 = qemu_blockalign(bs, sizeof(VHDXHeader));
424     header2 = qemu_blockalign(bs, sizeof(VHDXHeader));
425 
426     buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE);
427 
428     s->headers[0] = header1;
429     s->headers[1] = header2;
430 
431     /* We have to read the whole VHDX_HEADER_SIZE instead of
432      * sizeof(VHDXHeader), because the checksum is over the whole
433      * region */
434     ret = bdrv_pread(bs->file, VHDX_HEADER1_OFFSET, buffer,
435                      VHDX_HEADER_SIZE);
436     if (ret < 0) {
437         goto fail;
438     }
439     /* copy over just the relevant portion that we need */
440     memcpy(header1, buffer, sizeof(VHDXHeader));
441 
442     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
443         vhdx_header_le_import(header1);
444         if (header1->signature == VHDX_HEADER_SIGNATURE &&
445             header1->version == 1) {
446             h1_seq = header1->sequence_number;
447             h1_valid = true;
448         }
449     }
450 
451     ret = bdrv_pread(bs->file, VHDX_HEADER2_OFFSET, buffer,
452                      VHDX_HEADER_SIZE);
453     if (ret < 0) {
454         goto fail;
455     }
456     /* copy over just the relevant portion that we need */
457     memcpy(header2, buffer, sizeof(VHDXHeader));
458 
459     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
460         vhdx_header_le_import(header2);
461         if (header2->signature == VHDX_HEADER_SIGNATURE &&
462             header2->version == 1) {
463             h2_seq = header2->sequence_number;
464             h2_valid = true;
465         }
466     }
467 
468     /* If there is only 1 valid header (or no valid headers), we
469      * don't care what the sequence numbers are */
470     if (h1_valid && !h2_valid) {
471         s->curr_header = 0;
472     } else if (!h1_valid && h2_valid) {
473         s->curr_header = 1;
474     } else if (!h1_valid && !h2_valid) {
475         goto fail;
476     } else {
477         /* If both headers are valid, then we choose the active one by the
478          * highest sequence number.  If the sequence numbers are equal, that is
479          * invalid */
480         if (h1_seq > h2_seq) {
481             s->curr_header = 0;
482         } else if (h2_seq > h1_seq) {
483             s->curr_header = 1;
484         } else {
485             /* The Microsoft Disk2VHD tool will create 2 identical
486              * headers, with identical sequence numbers.  If the headers are
487              * identical, don't consider the file corrupt */
488             if (!memcmp(header1, header2, sizeof(VHDXHeader))) {
489                 s->curr_header = 0;
490             } else {
491                 goto fail;
492             }
493         }
494     }
495 
496     vhdx_region_register(s, s->headers[s->curr_header]->log_offset,
497                             s->headers[s->curr_header]->log_length);
498     goto exit;
499 
500 fail:
501     error_setg_errno(errp, -ret, "No valid VHDX header found");
502     qemu_vfree(header1);
503     qemu_vfree(header2);
504     s->headers[0] = NULL;
505     s->headers[1] = NULL;
506 exit:
507     qemu_vfree(buffer);
508 }
509 
510 
511 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s)
512 {
513     int ret = 0;
514     uint8_t *buffer;
515     int offset = 0;
516     VHDXRegionTableEntry rt_entry;
517     uint32_t i;
518     bool bat_rt_found = false;
519     bool metadata_rt_found = false;
520 
521     /* We have to read the whole 64KB block, because the crc32 is over the
522      * whole block */
523     buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE);
524 
525     ret = bdrv_pread(bs->file, VHDX_REGION_TABLE_OFFSET, buffer,
526                      VHDX_HEADER_BLOCK_SIZE);
527     if (ret < 0) {
528         goto fail;
529     }
530     memcpy(&s->rt, buffer, sizeof(s->rt));
531     offset += sizeof(s->rt);
532 
533     if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4)) {
534         ret = -EINVAL;
535         goto fail;
536     }
537 
538     vhdx_region_header_le_import(&s->rt);
539 
540     if (s->rt.signature != VHDX_REGION_SIGNATURE) {
541         ret = -EINVAL;
542         goto fail;
543     }
544 
545 
546     /* Per spec, maximum region table entry count is 2047 */
547     if (s->rt.entry_count > 2047) {
548         ret = -EINVAL;
549         goto fail;
550     }
551 
552     for (i = 0; i < s->rt.entry_count; i++) {
553         memcpy(&rt_entry, buffer + offset, sizeof(rt_entry));
554         offset += sizeof(rt_entry);
555 
556         vhdx_region_entry_le_import(&rt_entry);
557 
558         /* check for region overlap between these entries, and any
559          * other memory regions in the file */
560         ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length);
561         if (ret < 0) {
562             goto fail;
563         }
564 
565         vhdx_region_register(s, rt_entry.file_offset, rt_entry.length);
566 
567         /* see if we recognize the entry */
568         if (guid_eq(rt_entry.guid, bat_guid)) {
569             /* must be unique; if we have already found it this is invalid */
570             if (bat_rt_found) {
571                 ret = -EINVAL;
572                 goto fail;
573             }
574             bat_rt_found = true;
575             s->bat_rt = rt_entry;
576             continue;
577         }
578 
579         if (guid_eq(rt_entry.guid, metadata_guid)) {
580             /* must be unique; if we have already found it this is invalid */
581             if (metadata_rt_found) {
582                 ret = -EINVAL;
583                 goto fail;
584             }
585             metadata_rt_found = true;
586             s->metadata_rt = rt_entry;
587             continue;
588         }
589 
590         if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) {
591             /* cannot read vhdx file - required region table entry that
592              * we do not understand.  per spec, we must fail to open */
593             ret = -ENOTSUP;
594             goto fail;
595         }
596     }
597 
598     if (!bat_rt_found || !metadata_rt_found) {
599         ret = -EINVAL;
600         goto fail;
601     }
602 
603     ret = 0;
604 
605 fail:
606     qemu_vfree(buffer);
607     return ret;
608 }
609 
610 
611 
612 /* Metadata initial parser
613  *
614  * This loads all the metadata entry fields.  This may cause additional
615  * fields to be processed (e.g. parent locator, etc..).
616  *
617  * There are 5 Metadata items that are always required:
618  *      - File Parameters (block size, has a parent)
619  *      - Virtual Disk Size (size, in bytes, of the virtual drive)
620  *      - Page 83 Data (scsi page 83 guid)
621  *      - Logical Sector Size (logical sector size in bytes, either 512 or
622  *                             4096.  We only support 512 currently)
623  *      - Physical Sector Size (512 or 4096)
624  *
625  * Also, if the File Parameters indicate this is a differencing file,
626  * we must also look for the Parent Locator metadata item.
627  */
628 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s)
629 {
630     int ret = 0;
631     uint8_t *buffer;
632     int offset = 0;
633     uint32_t i = 0;
634     VHDXMetadataTableEntry md_entry;
635 
636     buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE);
637 
638     ret = bdrv_pread(bs->file, s->metadata_rt.file_offset, buffer,
639                      VHDX_METADATA_TABLE_MAX_SIZE);
640     if (ret < 0) {
641         goto exit;
642     }
643     memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr));
644     offset += sizeof(s->metadata_hdr);
645 
646     vhdx_metadata_header_le_import(&s->metadata_hdr);
647 
648     if (s->metadata_hdr.signature != VHDX_METADATA_SIGNATURE) {
649         ret = -EINVAL;
650         goto exit;
651     }
652 
653     s->metadata_entries.present = 0;
654 
655     if ((s->metadata_hdr.entry_count * sizeof(md_entry)) >
656         (VHDX_METADATA_TABLE_MAX_SIZE - offset)) {
657         ret = -EINVAL;
658         goto exit;
659     }
660 
661     for (i = 0; i < s->metadata_hdr.entry_count; i++) {
662         memcpy(&md_entry, buffer + offset, sizeof(md_entry));
663         offset += sizeof(md_entry);
664 
665         vhdx_metadata_entry_le_import(&md_entry);
666 
667         if (guid_eq(md_entry.item_id, file_param_guid)) {
668             if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) {
669                 ret = -EINVAL;
670                 goto exit;
671             }
672             s->metadata_entries.file_parameters_entry = md_entry;
673             s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT;
674             continue;
675         }
676 
677         if (guid_eq(md_entry.item_id, virtual_size_guid)) {
678             if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) {
679                 ret = -EINVAL;
680                 goto exit;
681             }
682             s->metadata_entries.virtual_disk_size_entry = md_entry;
683             s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT;
684             continue;
685         }
686 
687         if (guid_eq(md_entry.item_id, page83_guid)) {
688             if (s->metadata_entries.present & META_PAGE_83_PRESENT) {
689                 ret = -EINVAL;
690                 goto exit;
691             }
692             s->metadata_entries.page83_data_entry = md_entry;
693             s->metadata_entries.present |= META_PAGE_83_PRESENT;
694             continue;
695         }
696 
697         if (guid_eq(md_entry.item_id, logical_sector_guid)) {
698             if (s->metadata_entries.present &
699                 META_LOGICAL_SECTOR_SIZE_PRESENT) {
700                 ret = -EINVAL;
701                 goto exit;
702             }
703             s->metadata_entries.logical_sector_size_entry = md_entry;
704             s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT;
705             continue;
706         }
707 
708         if (guid_eq(md_entry.item_id, phys_sector_guid)) {
709             if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) {
710                 ret = -EINVAL;
711                 goto exit;
712             }
713             s->metadata_entries.phys_sector_size_entry = md_entry;
714             s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT;
715             continue;
716         }
717 
718         if (guid_eq(md_entry.item_id, parent_locator_guid)) {
719             if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
720                 ret = -EINVAL;
721                 goto exit;
722             }
723             s->metadata_entries.parent_locator_entry = md_entry;
724             s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT;
725             continue;
726         }
727 
728         if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) {
729             /* cannot read vhdx file - required region table entry that
730              * we do not understand.  per spec, we must fail to open */
731             ret = -ENOTSUP;
732             goto exit;
733         }
734     }
735 
736     if (s->metadata_entries.present != META_ALL_PRESENT) {
737         ret = -ENOTSUP;
738         goto exit;
739     }
740 
741     ret = bdrv_pread(bs->file,
742                      s->metadata_entries.file_parameters_entry.offset
743                                          + s->metadata_rt.file_offset,
744                      &s->params,
745                      sizeof(s->params));
746 
747     if (ret < 0) {
748         goto exit;
749     }
750 
751     le32_to_cpus(&s->params.block_size);
752     le32_to_cpus(&s->params.data_bits);
753 
754 
755     /* We now have the file parameters, so we can tell if this is a
756      * differencing file (i.e.. has_parent), is dynamic or fixed
757      * sized (leave_blocks_allocated), and the block size */
758 
759     /* The parent locator required iff the file parameters has_parent set */
760     if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
761         if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
762             /* TODO: parse  parent locator fields */
763             ret = -ENOTSUP; /* temp, until differencing files are supported */
764             goto exit;
765         } else {
766             /* if has_parent is set, but there is not parent locator present,
767              * then that is an invalid combination */
768             ret = -EINVAL;
769             goto exit;
770         }
771     }
772 
773     /* determine virtual disk size, logical sector size,
774      * and phys sector size */
775 
776     ret = bdrv_pread(bs->file,
777                      s->metadata_entries.virtual_disk_size_entry.offset
778                                            + s->metadata_rt.file_offset,
779                      &s->virtual_disk_size,
780                      sizeof(uint64_t));
781     if (ret < 0) {
782         goto exit;
783     }
784     ret = bdrv_pread(bs->file,
785                      s->metadata_entries.logical_sector_size_entry.offset
786                                              + s->metadata_rt.file_offset,
787                      &s->logical_sector_size,
788                      sizeof(uint32_t));
789     if (ret < 0) {
790         goto exit;
791     }
792     ret = bdrv_pread(bs->file,
793                      s->metadata_entries.phys_sector_size_entry.offset
794                                           + s->metadata_rt.file_offset,
795                      &s->physical_sector_size,
796                      sizeof(uint32_t));
797     if (ret < 0) {
798         goto exit;
799     }
800 
801     le64_to_cpus(&s->virtual_disk_size);
802     le32_to_cpus(&s->logical_sector_size);
803     le32_to_cpus(&s->physical_sector_size);
804 
805     if (s->params.block_size < VHDX_BLOCK_SIZE_MIN ||
806         s->params.block_size > VHDX_BLOCK_SIZE_MAX) {
807         ret = -EINVAL;
808         goto exit;
809     }
810 
811     /* only 2 supported sector sizes */
812     if (s->logical_sector_size != 512 && s->logical_sector_size != 4096) {
813         ret = -EINVAL;
814         goto exit;
815     }
816 
817     /* Both block_size and sector_size are guaranteed powers of 2, below.
818        Due to range checks above, s->sectors_per_block can never be < 256 */
819     s->sectors_per_block = s->params.block_size / s->logical_sector_size;
820     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
821                      (uint64_t)s->logical_sector_size /
822                      (uint64_t)s->params.block_size;
823 
824     /* These values are ones we will want to use for division / multiplication
825      * later on, and they are all guaranteed (per the spec) to be powers of 2,
826      * so we can take advantage of that for shift operations during
827      * reads/writes */
828     if (s->logical_sector_size & (s->logical_sector_size - 1)) {
829         ret = -EINVAL;
830         goto exit;
831     }
832     if (s->sectors_per_block & (s->sectors_per_block - 1)) {
833         ret = -EINVAL;
834         goto exit;
835     }
836     if (s->chunk_ratio & (s->chunk_ratio - 1)) {
837         ret = -EINVAL;
838         goto exit;
839     }
840     s->block_size = s->params.block_size;
841     if (s->block_size & (s->block_size - 1)) {
842         ret = -EINVAL;
843         goto exit;
844     }
845 
846     vhdx_set_shift_bits(s);
847 
848     ret = 0;
849 
850 exit:
851     qemu_vfree(buffer);
852     return ret;
853 }
854 
855 /*
856  * Calculate the number of BAT entries, including sector
857  * bitmap entries.
858  */
859 static void vhdx_calc_bat_entries(BDRVVHDXState *s)
860 {
861     uint32_t data_blocks_cnt, bitmap_blocks_cnt;
862 
863     data_blocks_cnt = DIV_ROUND_UP(s->virtual_disk_size, s->block_size);
864     bitmap_blocks_cnt = DIV_ROUND_UP(data_blocks_cnt, s->chunk_ratio);
865 
866     if (s->parent_entries) {
867         s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1);
868     } else {
869         s->bat_entries = data_blocks_cnt +
870                          ((data_blocks_cnt - 1) >> s->chunk_ratio_bits);
871     }
872 
873 }
874 
875 static void vhdx_close(BlockDriverState *bs)
876 {
877     BDRVVHDXState *s = bs->opaque;
878     qemu_vfree(s->headers[0]);
879     s->headers[0] = NULL;
880     qemu_vfree(s->headers[1]);
881     s->headers[1] = NULL;
882     qemu_vfree(s->bat);
883     s->bat = NULL;
884     qemu_vfree(s->parent_entries);
885     s->parent_entries = NULL;
886     migrate_del_blocker(s->migration_blocker);
887     error_free(s->migration_blocker);
888     qemu_vfree(s->log.hdr);
889     s->log.hdr = NULL;
890     vhdx_region_unregister_all(s);
891 }
892 
893 static int vhdx_open(BlockDriverState *bs, QDict *options, int flags,
894                      Error **errp)
895 {
896     BDRVVHDXState *s = bs->opaque;
897     int ret = 0;
898     uint32_t i;
899     uint64_t signature;
900     Error *local_err = NULL;
901 
902     s->bat = NULL;
903     s->first_visible_write = true;
904 
905     qemu_co_mutex_init(&s->lock);
906     QLIST_INIT(&s->regions);
907 
908     /* validate the file signature */
909     ret = bdrv_pread(bs->file, 0, &signature, sizeof(uint64_t));
910     if (ret < 0) {
911         goto fail;
912     }
913     if (memcmp(&signature, "vhdxfile", 8)) {
914         ret = -EINVAL;
915         goto fail;
916     }
917 
918     /* This is used for any header updates, for the file_write_guid.
919      * The spec dictates that a new value should be used for the first
920      * header update */
921     vhdx_guid_generate(&s->session_guid);
922 
923     vhdx_parse_header(bs, s, &local_err);
924     if (local_err != NULL) {
925         error_propagate(errp, local_err);
926         ret = -EINVAL;
927         goto fail;
928     }
929 
930     ret = vhdx_parse_log(bs, s, &s->log_replayed_on_open, errp);
931     if (ret < 0) {
932         goto fail;
933     }
934 
935     ret = vhdx_open_region_tables(bs, s);
936     if (ret < 0) {
937         goto fail;
938     }
939 
940     ret = vhdx_parse_metadata(bs, s);
941     if (ret < 0) {
942         goto fail;
943     }
944 
945     s->block_size = s->params.block_size;
946 
947     /* the VHDX spec dictates that virtual_disk_size is always a multiple of
948      * logical_sector_size */
949     bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits;
950 
951     vhdx_calc_bat_entries(s);
952 
953     s->bat_offset = s->bat_rt.file_offset;
954 
955     if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) {
956         /* BAT allocation is not large enough for all entries */
957         ret = -EINVAL;
958         goto fail;
959     }
960 
961     /* s->bat is freed in vhdx_close() */
962     s->bat = qemu_try_blockalign(bs->file->bs, s->bat_rt.length);
963     if (s->bat == NULL) {
964         ret = -ENOMEM;
965         goto fail;
966     }
967 
968     ret = bdrv_pread(bs->file, s->bat_offset, s->bat, s->bat_rt.length);
969     if (ret < 0) {
970         goto fail;
971     }
972 
973     uint64_t payblocks = s->chunk_ratio;
974     /* endian convert, and verify populated BAT field file offsets against
975      * region table and log entries */
976     for (i = 0; i < s->bat_entries; i++) {
977         le64_to_cpus(&s->bat[i]);
978         if (payblocks--) {
979             /* payload bat entries */
980             if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) ==
981                     PAYLOAD_BLOCK_FULLY_PRESENT) {
982                 ret = vhdx_region_check(s, s->bat[i] & VHDX_BAT_FILE_OFF_MASK,
983                                         s->block_size);
984                 if (ret < 0) {
985                     goto fail;
986                 }
987             }
988         } else {
989             payblocks = s->chunk_ratio;
990             /* Once differencing files are supported, verify sector bitmap
991              * blocks here */
992         }
993     }
994 
995     if (flags & BDRV_O_RDWR) {
996         ret = vhdx_update_headers(bs, s, false, NULL);
997         if (ret < 0) {
998             goto fail;
999         }
1000     }
1001 
1002     /* TODO: differencing files */
1003 
1004     /* Disable migration when VHDX images are used */
1005     error_setg(&s->migration_blocker, "The vhdx format used by node '%s' "
1006                "does not support live migration",
1007                bdrv_get_device_or_node_name(bs));
1008     migrate_add_blocker(s->migration_blocker);
1009 
1010     return 0;
1011 fail:
1012     vhdx_close(bs);
1013     return ret;
1014 }
1015 
1016 static int vhdx_reopen_prepare(BDRVReopenState *state,
1017                                BlockReopenQueue *queue, Error **errp)
1018 {
1019     return 0;
1020 }
1021 
1022 
1023 /*
1024  * Perform sector to block offset translations, to get various
1025  * sector and file offsets into the image.  See VHDXSectorInfo
1026  */
1027 static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num,
1028                                  int nb_sectors, VHDXSectorInfo *sinfo)
1029 {
1030     uint32_t block_offset;
1031 
1032     sinfo->bat_idx = sector_num >> s->sectors_per_block_bits;
1033     /* effectively a modulo - this gives us the offset into the block
1034      * (in sector sizes) for our sector number */
1035     block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits);
1036     /* the chunk ratio gives us the interleaving of the sector
1037      * bitmaps, so we need to advance our page block index by the
1038      * sector bitmaps entry number */
1039     sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits;
1040 
1041     /* the number of sectors we can read/write in this cycle */
1042     sinfo->sectors_avail = s->sectors_per_block - block_offset;
1043 
1044     sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits;
1045 
1046     if (sinfo->sectors_avail > nb_sectors) {
1047         sinfo->sectors_avail = nb_sectors;
1048     }
1049 
1050     sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits;
1051 
1052     sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK;
1053 
1054     sinfo->block_offset = block_offset << s->logical_sector_size_bits;
1055 
1056     /* The file offset must be past the header section, so must be > 0 */
1057     if (sinfo->file_offset == 0) {
1058         return;
1059     }
1060 
1061     /* block offset is the offset in vhdx logical sectors, in
1062      * the payload data block. Convert that to a byte offset
1063      * in the block, and add in the payload data block offset
1064      * in the file, in bytes, to get the final read address */
1065 
1066     sinfo->file_offset += sinfo->block_offset;
1067 }
1068 
1069 
1070 static int vhdx_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1071 {
1072     BDRVVHDXState *s = bs->opaque;
1073 
1074     bdi->cluster_size = s->block_size;
1075 
1076     bdi->unallocated_blocks_are_zero =
1077         (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) == 0;
1078 
1079     return 0;
1080 }
1081 
1082 
1083 static coroutine_fn int vhdx_co_readv(BlockDriverState *bs, int64_t sector_num,
1084                                       int nb_sectors, QEMUIOVector *qiov)
1085 {
1086     BDRVVHDXState *s = bs->opaque;
1087     int ret = 0;
1088     VHDXSectorInfo sinfo;
1089     uint64_t bytes_done = 0;
1090     QEMUIOVector hd_qiov;
1091 
1092     qemu_iovec_init(&hd_qiov, qiov->niov);
1093 
1094     qemu_co_mutex_lock(&s->lock);
1095 
1096     while (nb_sectors > 0) {
1097         /* We are a differencing file, so we need to inspect the sector bitmap
1098          * to see if we have the data or not */
1099         if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1100             /* not supported yet */
1101             ret = -ENOTSUP;
1102             goto exit;
1103         } else {
1104             vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1105 
1106             qemu_iovec_reset(&hd_qiov);
1107             qemu_iovec_concat(&hd_qiov, qiov,  bytes_done, sinfo.bytes_avail);
1108 
1109             /* check the payload block state */
1110             switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) {
1111             case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1112             case PAYLOAD_BLOCK_UNDEFINED:
1113             case PAYLOAD_BLOCK_UNMAPPED:
1114             case PAYLOAD_BLOCK_UNMAPPED_v095:
1115             case PAYLOAD_BLOCK_ZERO:
1116                 /* return zero */
1117                 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail);
1118                 break;
1119             case PAYLOAD_BLOCK_FULLY_PRESENT:
1120                 qemu_co_mutex_unlock(&s->lock);
1121                 ret = bdrv_co_readv(bs->file,
1122                                     sinfo.file_offset >> BDRV_SECTOR_BITS,
1123                                     sinfo.sectors_avail, &hd_qiov);
1124                 qemu_co_mutex_lock(&s->lock);
1125                 if (ret < 0) {
1126                     goto exit;
1127                 }
1128                 break;
1129             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1130                 /* we don't yet support difference files, fall through
1131                  * to error */
1132             default:
1133                 ret = -EIO;
1134                 goto exit;
1135                 break;
1136             }
1137             nb_sectors -= sinfo.sectors_avail;
1138             sector_num += sinfo.sectors_avail;
1139             bytes_done += sinfo.bytes_avail;
1140         }
1141     }
1142     ret = 0;
1143 exit:
1144     qemu_co_mutex_unlock(&s->lock);
1145     qemu_iovec_destroy(&hd_qiov);
1146     return ret;
1147 }
1148 
1149 /*
1150  * Allocate a new payload block at the end of the file.
1151  *
1152  * Allocation will happen at 1MB alignment inside the file
1153  *
1154  * Returns the file offset start of the new payload block
1155  */
1156 static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s,
1157                                     uint64_t *new_offset)
1158 {
1159     *new_offset = bdrv_getlength(bs->file->bs);
1160 
1161     /* per the spec, the address for a block is in units of 1MB */
1162     *new_offset = ROUND_UP(*new_offset, 1024 * 1024);
1163 
1164     return bdrv_truncate(bs->file->bs, *new_offset + s->block_size);
1165 }
1166 
1167 /*
1168  * Update the BAT table entry with the new file offset, and the new entry
1169  * state */
1170 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s,
1171                                        VHDXSectorInfo *sinfo,
1172                                        uint64_t *bat_entry_le,
1173                                        uint64_t *bat_offset, int state)
1174 {
1175     /* The BAT entry is a uint64, with 44 bits for the file offset in units of
1176      * 1MB, and 3 bits for the block state. */
1177     if ((state == PAYLOAD_BLOCK_ZERO)        ||
1178         (state == PAYLOAD_BLOCK_UNDEFINED)   ||
1179         (state == PAYLOAD_BLOCK_NOT_PRESENT) ||
1180         (state == PAYLOAD_BLOCK_UNMAPPED)) {
1181         s->bat[sinfo->bat_idx]  = 0;  /* For PAYLOAD_BLOCK_ZERO, the
1182                                          FileOffsetMB field is denoted as
1183                                          'reserved' in the v1.0 spec.  If it is
1184                                          non-zero, MS Hyper-V will fail to read
1185                                          the disk image */
1186     } else {
1187         s->bat[sinfo->bat_idx]  = sinfo->file_offset;
1188     }
1189 
1190     s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK;
1191 
1192     *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]);
1193     *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry);
1194 
1195 }
1196 
1197 /* Per the spec, on the first write of guest-visible data to the file the
1198  * data write guid must be updated in the header */
1199 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s)
1200 {
1201     int ret = 0;
1202     if (s->first_visible_write) {
1203         s->first_visible_write = false;
1204         ret = vhdx_update_headers(bs, s, true, NULL);
1205     }
1206     return ret;
1207 }
1208 
1209 static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num,
1210                                       int nb_sectors, QEMUIOVector *qiov)
1211 {
1212     int ret = -ENOTSUP;
1213     BDRVVHDXState *s = bs->opaque;
1214     VHDXSectorInfo sinfo;
1215     uint64_t bytes_done = 0;
1216     uint64_t bat_entry = 0;
1217     uint64_t bat_entry_offset = 0;
1218     QEMUIOVector hd_qiov;
1219     struct iovec iov1 = { 0 };
1220     struct iovec iov2 = { 0 };
1221     int sectors_to_write;
1222     int bat_state;
1223     uint64_t bat_prior_offset = 0;
1224     bool bat_update = false;
1225 
1226     qemu_iovec_init(&hd_qiov, qiov->niov);
1227 
1228     qemu_co_mutex_lock(&s->lock);
1229 
1230     ret = vhdx_user_visible_write(bs, s);
1231     if (ret < 0) {
1232         goto exit;
1233     }
1234 
1235     while (nb_sectors > 0) {
1236         bool use_zero_buffers = false;
1237         bat_update = false;
1238         if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1239             /* not supported yet */
1240             ret = -ENOTSUP;
1241             goto exit;
1242         } else {
1243             vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1244             sectors_to_write = sinfo.sectors_avail;
1245 
1246             qemu_iovec_reset(&hd_qiov);
1247             /* check the payload block state */
1248             bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK;
1249             switch (bat_state) {
1250             case PAYLOAD_BLOCK_ZERO:
1251                 /* in this case, we need to preserve zero writes for
1252                  * data that is not part of this write, so we must pad
1253                  * the rest of the buffer to zeroes */
1254 
1255                 /* if we are on a posix system with ftruncate() that extends
1256                  * a file, then it is zero-filled for us.  On Win32, the raw
1257                  * layer uses SetFilePointer and SetFileEnd, which does not
1258                  * zero fill AFAIK */
1259 
1260                 /* Queue another write of zero buffers if the underlying file
1261                  * does not zero-fill on file extension */
1262 
1263                 if (bdrv_has_zero_init(bs->file->bs) == 0) {
1264                     use_zero_buffers = true;
1265 
1266                     /* zero fill the front, if any */
1267                     if (sinfo.block_offset) {
1268                         iov1.iov_len = sinfo.block_offset;
1269                         iov1.iov_base = qemu_blockalign(bs, iov1.iov_len);
1270                         memset(iov1.iov_base, 0, iov1.iov_len);
1271                         qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0,
1272                                               iov1.iov_len);
1273                         sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS;
1274                     }
1275 
1276                     /* our actual data */
1277                     qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1278                                       sinfo.bytes_avail);
1279 
1280                     /* zero fill the back, if any */
1281                     if ((sinfo.bytes_avail - sinfo.block_offset) <
1282                          s->block_size) {
1283                         iov2.iov_len = s->block_size -
1284                                       (sinfo.bytes_avail + sinfo.block_offset);
1285                         iov2.iov_base = qemu_blockalign(bs, iov2.iov_len);
1286                         memset(iov2.iov_base, 0, iov2.iov_len);
1287                         qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0,
1288                                               iov2.iov_len);
1289                         sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS;
1290                     }
1291                 }
1292                 /* fall through */
1293             case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1294             case PAYLOAD_BLOCK_UNMAPPED:
1295             case PAYLOAD_BLOCK_UNMAPPED_v095:
1296             case PAYLOAD_BLOCK_UNDEFINED:
1297                 bat_prior_offset = sinfo.file_offset;
1298                 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset);
1299                 if (ret < 0) {
1300                     goto exit;
1301                 }
1302                 /* once we support differencing files, this may also be
1303                  * partially present */
1304                 /* update block state to the newly specified state */
1305                 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1306                                             &bat_entry_offset,
1307                                             PAYLOAD_BLOCK_FULLY_PRESENT);
1308                 bat_update = true;
1309                 /* since we just allocated a block, file_offset is the
1310                  * beginning of the payload block. It needs to be the
1311                  * write address, which includes the offset into the block */
1312                 if (!use_zero_buffers) {
1313                     sinfo.file_offset += sinfo.block_offset;
1314                 }
1315                 /* fall through */
1316             case PAYLOAD_BLOCK_FULLY_PRESENT:
1317                 /* if the file offset address is in the header zone,
1318                  * there is a problem */
1319                 if (sinfo.file_offset < (1024 * 1024)) {
1320                     ret = -EFAULT;
1321                     goto error_bat_restore;
1322                 }
1323 
1324                 if (!use_zero_buffers) {
1325                     qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1326                                       sinfo.bytes_avail);
1327                 }
1328                 /* block exists, so we can just overwrite it */
1329                 qemu_co_mutex_unlock(&s->lock);
1330                 ret = bdrv_co_writev(bs->file,
1331                                     sinfo.file_offset >> BDRV_SECTOR_BITS,
1332                                     sectors_to_write, &hd_qiov);
1333                 qemu_co_mutex_lock(&s->lock);
1334                 if (ret < 0) {
1335                     goto error_bat_restore;
1336                 }
1337                 break;
1338             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1339                 /* we don't yet support difference files, fall through
1340                  * to error */
1341             default:
1342                 ret = -EIO;
1343                 goto exit;
1344                 break;
1345             }
1346 
1347             if (bat_update) {
1348                 /* this will update the BAT entry into the log journal, and
1349                  * then flush the log journal out to disk */
1350                 ret =  vhdx_log_write_and_flush(bs, s, &bat_entry,
1351                                                 sizeof(VHDXBatEntry),
1352                                                 bat_entry_offset);
1353                 if (ret < 0) {
1354                     goto exit;
1355                 }
1356             }
1357 
1358             nb_sectors -= sinfo.sectors_avail;
1359             sector_num += sinfo.sectors_avail;
1360             bytes_done += sinfo.bytes_avail;
1361 
1362         }
1363     }
1364 
1365     goto exit;
1366 
1367 error_bat_restore:
1368     if (bat_update) {
1369         /* keep metadata in sync, and restore the bat entry state
1370          * if error. */
1371         sinfo.file_offset = bat_prior_offset;
1372         vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1373                                     &bat_entry_offset, bat_state);
1374     }
1375 exit:
1376     qemu_vfree(iov1.iov_base);
1377     qemu_vfree(iov2.iov_base);
1378     qemu_co_mutex_unlock(&s->lock);
1379     qemu_iovec_destroy(&hd_qiov);
1380     return ret;
1381 }
1382 
1383 
1384 
1385 /*
1386  * Create VHDX Headers
1387  *
1388  * There are 2 headers, and the highest sequence number will represent
1389  * the active header
1390  */
1391 static int vhdx_create_new_headers(BlockBackend *blk, uint64_t image_size,
1392                                    uint32_t log_size)
1393 {
1394     BlockDriverState *bs = blk_bs(blk);
1395     BdrvChild *child;
1396     int ret = 0;
1397     VHDXHeader *hdr = NULL;
1398 
1399     hdr = g_new0(VHDXHeader, 1);
1400 
1401     hdr->signature       = VHDX_HEADER_SIGNATURE;
1402     hdr->sequence_number = g_random_int();
1403     hdr->log_version     = 0;
1404     hdr->version         = 1;
1405     hdr->log_length      = log_size;
1406     hdr->log_offset      = VHDX_HEADER_SECTION_END;
1407     vhdx_guid_generate(&hdr->file_write_guid);
1408     vhdx_guid_generate(&hdr->data_write_guid);
1409 
1410     /* XXX Ugly way to get blk->root, but that's a feature, not a bug. This
1411      * hack makes it obvious that vhdx_write_header() bypasses the BlockBackend
1412      * here, which it really shouldn't be doing. */
1413     child = QLIST_FIRST(&bs->parents);
1414     assert(!QLIST_NEXT(child, next_parent));
1415 
1416     ret = vhdx_write_header(child, hdr, VHDX_HEADER1_OFFSET, false);
1417     if (ret < 0) {
1418         goto exit;
1419     }
1420     hdr->sequence_number++;
1421     ret = vhdx_write_header(child, hdr, VHDX_HEADER2_OFFSET, false);
1422     if (ret < 0) {
1423         goto exit;
1424     }
1425 
1426 exit:
1427     g_free(hdr);
1428     return ret;
1429 }
1430 
1431 #define VHDX_METADATA_ENTRY_BUFFER_SIZE \
1432                                     (sizeof(VHDXFileParameters)               +\
1433                                      sizeof(VHDXVirtualDiskSize)              +\
1434                                      sizeof(VHDXPage83Data)                   +\
1435                                      sizeof(VHDXVirtualDiskLogicalSectorSize) +\
1436                                      sizeof(VHDXVirtualDiskPhysicalSectorSize))
1437 
1438 /*
1439  * Create the Metadata entries.
1440  *
1441  * For more details on the entries, see section 3.5 (pg 29) in the
1442  * VHDX 1.00 specification.
1443  *
1444  * We support 5 metadata entries (all required by spec):
1445  *          File Parameters,
1446  *          Virtual Disk Size,
1447  *          Page 83 Data,
1448  *          Logical Sector Size,
1449  *          Physical Sector Size
1450  *
1451  * The first 64KB of the Metadata section is reserved for the metadata
1452  * header and entries; beyond that, the metadata items themselves reside.
1453  */
1454 static int vhdx_create_new_metadata(BlockBackend *blk,
1455                                     uint64_t image_size,
1456                                     uint32_t block_size,
1457                                     uint32_t sector_size,
1458                                     uint64_t metadata_offset,
1459                                     VHDXImageType type)
1460 {
1461     int ret = 0;
1462     uint32_t offset = 0;
1463     void *buffer = NULL;
1464     void *entry_buffer;
1465     VHDXMetadataTableHeader *md_table;
1466     VHDXMetadataTableEntry  *md_table_entry;
1467 
1468     /* Metadata entries */
1469     VHDXFileParameters     *mt_file_params;
1470     VHDXVirtualDiskSize    *mt_virtual_size;
1471     VHDXPage83Data         *mt_page83;
1472     VHDXVirtualDiskLogicalSectorSize  *mt_log_sector_size;
1473     VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size;
1474 
1475     entry_buffer = g_malloc0(VHDX_METADATA_ENTRY_BUFFER_SIZE);
1476 
1477     mt_file_params = entry_buffer;
1478     offset += sizeof(VHDXFileParameters);
1479     mt_virtual_size = entry_buffer + offset;
1480     offset += sizeof(VHDXVirtualDiskSize);
1481     mt_page83 = entry_buffer + offset;
1482     offset += sizeof(VHDXPage83Data);
1483     mt_log_sector_size = entry_buffer + offset;
1484     offset += sizeof(VHDXVirtualDiskLogicalSectorSize);
1485     mt_phys_sector_size = entry_buffer + offset;
1486 
1487     mt_file_params->block_size = cpu_to_le32(block_size);
1488     if (type == VHDX_TYPE_FIXED) {
1489         mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED;
1490         cpu_to_le32s(&mt_file_params->data_bits);
1491     }
1492 
1493     vhdx_guid_generate(&mt_page83->page_83_data);
1494     cpu_to_leguids(&mt_page83->page_83_data);
1495     mt_virtual_size->virtual_disk_size        = cpu_to_le64(image_size);
1496     mt_log_sector_size->logical_sector_size   = cpu_to_le32(sector_size);
1497     mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size);
1498 
1499     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1500     md_table = buffer;
1501 
1502     md_table->signature   = VHDX_METADATA_SIGNATURE;
1503     md_table->entry_count = 5;
1504     vhdx_metadata_header_le_export(md_table);
1505 
1506 
1507     /* This will reference beyond the reserved table portion */
1508     offset = 64 * KiB;
1509 
1510     md_table_entry = buffer + sizeof(VHDXMetadataTableHeader);
1511 
1512     md_table_entry[0].item_id = file_param_guid;
1513     md_table_entry[0].offset  = offset;
1514     md_table_entry[0].length  = sizeof(VHDXFileParameters);
1515     md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED;
1516     offset += md_table_entry[0].length;
1517     vhdx_metadata_entry_le_export(&md_table_entry[0]);
1518 
1519     md_table_entry[1].item_id = virtual_size_guid;
1520     md_table_entry[1].offset  = offset;
1521     md_table_entry[1].length  = sizeof(VHDXVirtualDiskSize);
1522     md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1523                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1524     offset += md_table_entry[1].length;
1525     vhdx_metadata_entry_le_export(&md_table_entry[1]);
1526 
1527     md_table_entry[2].item_id = page83_guid;
1528     md_table_entry[2].offset  = offset;
1529     md_table_entry[2].length  = sizeof(VHDXPage83Data);
1530     md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1531                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1532     offset += md_table_entry[2].length;
1533     vhdx_metadata_entry_le_export(&md_table_entry[2]);
1534 
1535     md_table_entry[3].item_id = logical_sector_guid;
1536     md_table_entry[3].offset  = offset;
1537     md_table_entry[3].length  = sizeof(VHDXVirtualDiskLogicalSectorSize);
1538     md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1539                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1540     offset += md_table_entry[3].length;
1541     vhdx_metadata_entry_le_export(&md_table_entry[3]);
1542 
1543     md_table_entry[4].item_id = phys_sector_guid;
1544     md_table_entry[4].offset  = offset;
1545     md_table_entry[4].length  = sizeof(VHDXVirtualDiskPhysicalSectorSize);
1546     md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1547                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1548     vhdx_metadata_entry_le_export(&md_table_entry[4]);
1549 
1550     ret = blk_pwrite(blk, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE, 0);
1551     if (ret < 0) {
1552         goto exit;
1553     }
1554 
1555     ret = blk_pwrite(blk, metadata_offset + (64 * KiB), entry_buffer,
1556                      VHDX_METADATA_ENTRY_BUFFER_SIZE, 0);
1557     if (ret < 0) {
1558         goto exit;
1559     }
1560 
1561 
1562 exit:
1563     g_free(buffer);
1564     g_free(entry_buffer);
1565     return ret;
1566 }
1567 
1568 /* This create the actual BAT itself.  We currently only support
1569  * 'Dynamic' and 'Fixed' image types.
1570  *
1571  *  Dynamic images: default state of the BAT is all zeroes.
1572  *
1573  *  Fixed images: default state of the BAT is fully populated, with
1574  *                file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT.
1575  */
1576 static int vhdx_create_bat(BlockBackend *blk, BDRVVHDXState *s,
1577                            uint64_t image_size, VHDXImageType type,
1578                            bool use_zero_blocks, uint64_t file_offset,
1579                            uint32_t length)
1580 {
1581     int ret = 0;
1582     uint64_t data_file_offset;
1583     uint64_t total_sectors = 0;
1584     uint64_t sector_num = 0;
1585     uint64_t unused;
1586     int block_state;
1587     VHDXSectorInfo sinfo;
1588 
1589     assert(s->bat == NULL);
1590 
1591     /* this gives a data start after BAT/bitmap entries, and well
1592      * past any metadata entries (with a 4 MB buffer for future
1593      * expansion */
1594     data_file_offset = file_offset + length + 5 * MiB;
1595     total_sectors = image_size >> s->logical_sector_size_bits;
1596 
1597     if (type == VHDX_TYPE_DYNAMIC) {
1598         /* All zeroes, so we can just extend the file - the end of the BAT
1599          * is the furthest thing we have written yet */
1600         ret = blk_truncate(blk, data_file_offset);
1601         if (ret < 0) {
1602             goto exit;
1603         }
1604     } else if (type == VHDX_TYPE_FIXED) {
1605         ret = blk_truncate(blk, data_file_offset + image_size);
1606         if (ret < 0) {
1607             goto exit;
1608         }
1609     } else {
1610         ret = -ENOTSUP;
1611         goto exit;
1612     }
1613 
1614     if (type == VHDX_TYPE_FIXED ||
1615                 use_zero_blocks ||
1616                 bdrv_has_zero_init(blk_bs(blk)) == 0) {
1617         /* for a fixed file, the default BAT entry is not zero */
1618         s->bat = g_try_malloc0(length);
1619         if (length && s->bat == NULL) {
1620             ret = -ENOMEM;
1621             goto exit;
1622         }
1623         block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT :
1624                                                 PAYLOAD_BLOCK_NOT_PRESENT;
1625         block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state;
1626         /* fill the BAT by emulating sector writes of sectors_per_block size */
1627         while (sector_num < total_sectors) {
1628             vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo);
1629             sinfo.file_offset = data_file_offset +
1630                                 (sector_num << s->logical_sector_size_bits);
1631             sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB);
1632             vhdx_update_bat_table_entry(blk_bs(blk), s, &sinfo, &unused, &unused,
1633                                         block_state);
1634             cpu_to_le64s(&s->bat[sinfo.bat_idx]);
1635             sector_num += s->sectors_per_block;
1636         }
1637         ret = blk_pwrite(blk, file_offset, s->bat, length, 0);
1638         if (ret < 0) {
1639             goto exit;
1640         }
1641     }
1642 
1643 
1644 
1645 exit:
1646     g_free(s->bat);
1647     return ret;
1648 }
1649 
1650 /* Creates the region table header, and region table entries.
1651  * There are 2 supported region table entries: BAT, and Metadata/
1652  *
1653  * As the calculations for the BAT region table are also needed
1654  * to create the BAT itself, we will also cause the BAT to be
1655  * created.
1656  */
1657 static int vhdx_create_new_region_table(BlockBackend *blk,
1658                                         uint64_t image_size,
1659                                         uint32_t block_size,
1660                                         uint32_t sector_size,
1661                                         uint32_t log_size,
1662                                         bool use_zero_blocks,
1663                                         VHDXImageType type,
1664                                         uint64_t *metadata_offset)
1665 {
1666     int ret = 0;
1667     uint32_t offset = 0;
1668     void *buffer = NULL;
1669     uint64_t bat_file_offset;
1670     uint32_t bat_length;
1671     BDRVVHDXState *s = NULL;
1672     VHDXRegionTableHeader *region_table;
1673     VHDXRegionTableEntry *rt_bat;
1674     VHDXRegionTableEntry *rt_metadata;
1675 
1676     assert(metadata_offset != NULL);
1677 
1678     /* Populate enough of the BDRVVHDXState to be able to use the
1679      * pre-existing BAT calculation, translation, and update functions */
1680     s = g_new0(BDRVVHDXState, 1);
1681 
1682     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
1683                      (uint64_t) sector_size / (uint64_t) block_size;
1684 
1685     s->sectors_per_block = block_size / sector_size;
1686     s->virtual_disk_size = image_size;
1687     s->block_size = block_size;
1688     s->logical_sector_size = sector_size;
1689 
1690     vhdx_set_shift_bits(s);
1691 
1692     vhdx_calc_bat_entries(s);
1693 
1694     /* At this point the VHDX state is populated enough for creation */
1695 
1696     /* a single buffer is used so we can calculate the checksum over the
1697      * entire 64KB block */
1698     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1699     region_table = buffer;
1700     offset += sizeof(VHDXRegionTableHeader);
1701     rt_bat = buffer + offset;
1702     offset += sizeof(VHDXRegionTableEntry);
1703     rt_metadata  = buffer + offset;
1704 
1705     region_table->signature = VHDX_REGION_SIGNATURE;
1706     region_table->entry_count = 2;   /* BAT and Metadata */
1707 
1708     rt_bat->guid        = bat_guid;
1709     rt_bat->length      = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB);
1710     rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB);
1711     s->bat_offset = rt_bat->file_offset;
1712 
1713     rt_metadata->guid        = metadata_guid;
1714     rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length,
1715                                         MiB);
1716     rt_metadata->length      = 1 * MiB; /* min size, and more than enough */
1717     *metadata_offset = rt_metadata->file_offset;
1718 
1719     bat_file_offset = rt_bat->file_offset;
1720     bat_length = rt_bat->length;
1721 
1722     vhdx_region_header_le_export(region_table);
1723     vhdx_region_entry_le_export(rt_bat);
1724     vhdx_region_entry_le_export(rt_metadata);
1725 
1726     vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE,
1727                          offsetof(VHDXRegionTableHeader, checksum));
1728 
1729 
1730     /* The region table gives us the data we need to create the BAT,
1731      * so do that now */
1732     ret = vhdx_create_bat(blk, s, image_size, type, use_zero_blocks,
1733                           bat_file_offset, bat_length);
1734     if (ret < 0) {
1735         goto exit;
1736     }
1737 
1738     /* Now write out the region headers to disk */
1739     ret = blk_pwrite(blk, VHDX_REGION_TABLE_OFFSET, buffer,
1740                      VHDX_HEADER_BLOCK_SIZE, 0);
1741     if (ret < 0) {
1742         goto exit;
1743     }
1744 
1745     ret = blk_pwrite(blk, VHDX_REGION_TABLE2_OFFSET, buffer,
1746                      VHDX_HEADER_BLOCK_SIZE, 0);
1747     if (ret < 0) {
1748         goto exit;
1749     }
1750 
1751 exit:
1752     g_free(s);
1753     g_free(buffer);
1754     return ret;
1755 }
1756 
1757 /* We need to create the following elements:
1758  *
1759  *    .-----------------------------------------------------------------.
1760  *    |   (A)    |   (B)    |    (C)    |     (D)       |     (E)       |
1761  *    |  File ID |  Header1 |  Header 2 |  Region Tbl 1 |  Region Tbl 2 |
1762  *    |          |          |           |               |               |
1763  *    .-----------------------------------------------------------------.
1764  *    0         64KB      128KB       192KB           256KB           320KB
1765  *
1766  *
1767  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1768  *    |     (F)     |     (G)       |    (H)    |                        |
1769  *    | Journal Log |  BAT / Bitmap |  Metadata |  .... data ......      |
1770  *    |             |               |           |                        |
1771  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1772  *   1MB
1773  */
1774 static int vhdx_create(const char *filename, QemuOpts *opts, Error **errp)
1775 {
1776     int ret = 0;
1777     uint64_t image_size = (uint64_t) 2 * GiB;
1778     uint32_t log_size   = 1 * MiB;
1779     uint32_t block_size = 0;
1780     uint64_t signature;
1781     uint64_t metadata_offset;
1782     bool use_zero_blocks = false;
1783 
1784     gunichar2 *creator = NULL;
1785     glong creator_items;
1786     BlockBackend *blk;
1787     char *type = NULL;
1788     VHDXImageType image_type;
1789     Error *local_err = NULL;
1790 
1791     image_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1792                           BDRV_SECTOR_SIZE);
1793     log_size = qemu_opt_get_size_del(opts, VHDX_BLOCK_OPT_LOG_SIZE, 0);
1794     block_size = qemu_opt_get_size_del(opts, VHDX_BLOCK_OPT_BLOCK_SIZE, 0);
1795     type = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1796     use_zero_blocks = qemu_opt_get_bool_del(opts, VHDX_BLOCK_OPT_ZERO, true);
1797 
1798     if (image_size > VHDX_MAX_IMAGE_SIZE) {
1799         error_setg_errno(errp, EINVAL, "Image size too large; max of 64TB");
1800         ret = -EINVAL;
1801         goto exit;
1802     }
1803 
1804     if (type == NULL) {
1805         type = g_strdup("dynamic");
1806     }
1807 
1808     if (!strcmp(type, "dynamic")) {
1809         image_type = VHDX_TYPE_DYNAMIC;
1810     } else if (!strcmp(type, "fixed")) {
1811         image_type = VHDX_TYPE_FIXED;
1812     } else if (!strcmp(type, "differencing")) {
1813         error_setg_errno(errp, ENOTSUP,
1814                          "Differencing files not yet supported");
1815         ret = -ENOTSUP;
1816         goto exit;
1817     } else {
1818         ret = -EINVAL;
1819         goto exit;
1820     }
1821 
1822     /* These are pretty arbitrary, and mainly designed to keep the BAT
1823      * size reasonable to load into RAM */
1824     if (block_size == 0) {
1825         if (image_size > 32 * TiB) {
1826             block_size = 64 * MiB;
1827         } else if (image_size > (uint64_t) 100 * GiB) {
1828             block_size = 32 * MiB;
1829         } else if (image_size > 1 * GiB) {
1830             block_size = 16 * MiB;
1831         } else {
1832             block_size = 8 * MiB;
1833         }
1834     }
1835 
1836 
1837     /* make the log size close to what was specified, but must be
1838      * min 1MB, and multiple of 1MB */
1839     log_size = ROUND_UP(log_size, MiB);
1840 
1841     block_size = ROUND_UP(block_size, MiB);
1842     block_size = block_size > VHDX_BLOCK_SIZE_MAX ? VHDX_BLOCK_SIZE_MAX :
1843                                                     block_size;
1844 
1845     ret = bdrv_create_file(filename, opts, &local_err);
1846     if (ret < 0) {
1847         error_propagate(errp, local_err);
1848         goto exit;
1849     }
1850 
1851     blk = blk_new_open(filename, NULL, NULL,
1852                        BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
1853     if (blk == NULL) {
1854         error_propagate(errp, local_err);
1855         ret = -EIO;
1856         goto exit;
1857     }
1858 
1859     blk_set_allow_write_beyond_eof(blk, true);
1860 
1861     /* Create (A) */
1862 
1863     /* The creator field is optional, but may be useful for
1864      * debugging / diagnostics */
1865     creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL,
1866                               &creator_items, NULL);
1867     signature = cpu_to_le64(VHDX_FILE_SIGNATURE);
1868     ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature),
1869                      0);
1870     if (ret < 0) {
1871         goto delete_and_exit;
1872     }
1873     if (creator) {
1874         ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET + sizeof(signature),
1875                          creator, creator_items * sizeof(gunichar2), 0);
1876         if (ret < 0) {
1877             goto delete_and_exit;
1878         }
1879     }
1880 
1881 
1882     /* Creates (B),(C) */
1883     ret = vhdx_create_new_headers(blk, image_size, log_size);
1884     if (ret < 0) {
1885         goto delete_and_exit;
1886     }
1887 
1888     /* Creates (D),(E),(G) explicitly. (F) created as by-product */
1889     ret = vhdx_create_new_region_table(blk, image_size, block_size, 512,
1890                                        log_size, use_zero_blocks, image_type,
1891                                        &metadata_offset);
1892     if (ret < 0) {
1893         goto delete_and_exit;
1894     }
1895 
1896     /* Creates (H) */
1897     ret = vhdx_create_new_metadata(blk, image_size, block_size, 512,
1898                                    metadata_offset, image_type);
1899     if (ret < 0) {
1900         goto delete_and_exit;
1901     }
1902 
1903 
1904 delete_and_exit:
1905     blk_unref(blk);
1906 exit:
1907     g_free(type);
1908     g_free(creator);
1909     return ret;
1910 }
1911 
1912 /* If opened r/w, the VHDX driver will automatically replay the log,
1913  * if one is present, inside the vhdx_open() call.
1914  *
1915  * If qemu-img check -r all is called, the image is automatically opened
1916  * r/w and any log has already been replayed, so there is nothing (currently)
1917  * for us to do here
1918  */
1919 static int vhdx_check(BlockDriverState *bs, BdrvCheckResult *result,
1920                        BdrvCheckMode fix)
1921 {
1922     BDRVVHDXState *s = bs->opaque;
1923 
1924     if (s->log_replayed_on_open) {
1925         result->corruptions_fixed++;
1926     }
1927     return 0;
1928 }
1929 
1930 static QemuOptsList vhdx_create_opts = {
1931     .name = "vhdx-create-opts",
1932     .head = QTAILQ_HEAD_INITIALIZER(vhdx_create_opts.head),
1933     .desc = {
1934         {
1935            .name = BLOCK_OPT_SIZE,
1936            .type = QEMU_OPT_SIZE,
1937            .help = "Virtual disk size; max of 64TB."
1938        },
1939        {
1940            .name = VHDX_BLOCK_OPT_LOG_SIZE,
1941            .type = QEMU_OPT_SIZE,
1942            .def_value_str = stringify(DEFAULT_LOG_SIZE),
1943            .help = "Log size; min 1MB."
1944        },
1945        {
1946            .name = VHDX_BLOCK_OPT_BLOCK_SIZE,
1947            .type = QEMU_OPT_SIZE,
1948            .def_value_str = stringify(0),
1949            .help = "Block Size; min 1MB, max 256MB. " \
1950                    "0 means auto-calculate based on image size."
1951        },
1952        {
1953            .name = BLOCK_OPT_SUBFMT,
1954            .type = QEMU_OPT_STRING,
1955            .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "\
1956                    "Default is 'dynamic'."
1957        },
1958        {
1959            .name = VHDX_BLOCK_OPT_ZERO,
1960            .type = QEMU_OPT_BOOL,
1961            .help = "Force use of payload blocks of type 'ZERO'. "\
1962                    "Non-standard, but default.  Do not set to 'off' when "\
1963                    "using 'qemu-img convert' with subformat=dynamic."
1964        },
1965        { NULL }
1966     }
1967 };
1968 
1969 static BlockDriver bdrv_vhdx = {
1970     .format_name            = "vhdx",
1971     .instance_size          = sizeof(BDRVVHDXState),
1972     .bdrv_probe             = vhdx_probe,
1973     .bdrv_open              = vhdx_open,
1974     .bdrv_close             = vhdx_close,
1975     .bdrv_reopen_prepare    = vhdx_reopen_prepare,
1976     .bdrv_co_readv          = vhdx_co_readv,
1977     .bdrv_co_writev         = vhdx_co_writev,
1978     .bdrv_create            = vhdx_create,
1979     .bdrv_get_info          = vhdx_get_info,
1980     .bdrv_check             = vhdx_check,
1981     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1982 
1983     .create_opts            = &vhdx_create_opts,
1984 };
1985 
1986 static void bdrv_vhdx_init(void)
1987 {
1988     bdrv_register(&bdrv_vhdx);
1989 }
1990 
1991 block_init(bdrv_vhdx_init);
1992