xref: /openbmc/qemu/block/vhdx.c (revision 10358b6a)
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 resulting checksum is in the CPU endianness, not necessarily
139  *       in the file format endianness (LE).  Any header export to disk should
140  *       make sure that vhdx_header_le_export() is used to convert to the
141  *       correct endianness
142  */
143 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset)
144 {
145     uint32_t crc;
146 
147     assert(buf != NULL);
148     assert(size > (crc_offset + sizeof(crc)));
149 
150     memset(buf + crc_offset, 0, sizeof(crc));
151     crc =  crc32c(0xffffffff, buf, size);
152     memcpy(buf + crc_offset, &crc, sizeof(crc));
153 
154     return crc;
155 }
156 
157 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size,
158                             int crc_offset)
159 {
160     uint32_t crc_new;
161     uint32_t crc_orig;
162     assert(buf != NULL);
163 
164     if (crc_offset > 0) {
165         memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
166         memset(buf + crc_offset, 0, sizeof(crc_orig));
167     }
168 
169     crc_new = crc32c(crc, buf, size);
170     if (crc_offset > 0) {
171         memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig));
172     }
173 
174     return crc_new;
175 }
176 
177 /* Validates the checksum of the buffer, with an in-place CRC.
178  *
179  * Zero is substituted during crc calculation for the original crc field,
180  * and the crc field is restored afterwards.  But the buffer will be modifed
181  * during the calculation, so this may not be not suitable for multi-threaded
182  * use.
183  *
184  * crc_offset: byte offset in buf of the buffer crc
185  * buf: buffer pointer
186  * size: size of buffer (must be > crc_offset+4)
187  *
188  * returns true if checksum is valid, false otherwise
189  */
190 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset)
191 {
192     uint32_t crc_orig;
193     uint32_t crc;
194 
195     assert(buf != NULL);
196     assert(size > (crc_offset + 4));
197 
198     memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
199     crc_orig = le32_to_cpu(crc_orig);
200 
201     crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset);
202 
203     return crc == crc_orig;
204 }
205 
206 
207 /*
208  * This generates a UUID that is compliant with the MS GUIDs used
209  * in the VHDX spec (and elsewhere).
210  */
211 void vhdx_guid_generate(MSGUID *guid)
212 {
213     uuid_t uuid;
214     assert(guid != NULL);
215 
216     uuid_generate(uuid);
217     memcpy(guid, uuid, sizeof(MSGUID));
218 }
219 
220 /* Check for region overlaps inside the VHDX image */
221 static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length)
222 {
223     int ret = 0;
224     uint64_t end;
225     VHDXRegionEntry *r;
226 
227     end = start + length;
228     QLIST_FOREACH(r, &s->regions, entries) {
229         if (!((start >= r->end) || (end <= r->start))) {
230             ret = -EINVAL;
231             goto exit;
232         }
233     }
234 
235 exit:
236     return ret;
237 }
238 
239 /* Register a region for future checks */
240 static void vhdx_region_register(BDRVVHDXState *s,
241                                  uint64_t start, uint64_t length)
242 {
243     VHDXRegionEntry *r;
244 
245     r = g_malloc0(sizeof(*r));
246 
247     r->start = start;
248     r->end = start + length;
249 
250     QLIST_INSERT_HEAD(&s->regions, r, entries);
251 }
252 
253 /* Free all registered regions */
254 static void vhdx_region_unregister_all(BDRVVHDXState *s)
255 {
256     VHDXRegionEntry *r, *r_next;
257 
258     QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) {
259         QLIST_REMOVE(r, entries);
260         g_free(r);
261     }
262 }
263 
264 static void vhdx_set_shift_bits(BDRVVHDXState *s)
265 {
266     s->logical_sector_size_bits = 31 - clz32(s->logical_sector_size);
267     s->sectors_per_block_bits =   31 - clz32(s->sectors_per_block);
268     s->chunk_ratio_bits =         63 - clz64(s->chunk_ratio);
269     s->block_size_bits =          31 - clz32(s->block_size);
270 }
271 
272 /*
273  * Per the MS VHDX Specification, for every VHDX file:
274  *      - The header section is fixed size - 1 MB
275  *      - The header section is always the first "object"
276  *      - The first 64KB of the header is the File Identifier
277  *      - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile")
278  *      - The following 512 bytes constitute a UTF-16 string identifiying the
279  *        software that created the file, and is optional and diagnostic only.
280  *
281  *  Therefore, we probe by looking for the vhdxfile signature "vhdxfile"
282  */
283 static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename)
284 {
285     if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) {
286         return 100;
287     }
288     return 0;
289 }
290 
291 /*
292  * Writes the header to the specified offset.
293  *
294  * This will optionally read in buffer data from disk (otherwise zero-fill),
295  * and then update the header checksum.  Header is converted to proper
296  * endianness before being written to the specified file offset
297  */
298 static int vhdx_write_header(BlockDriverState *bs_file, VHDXHeader *hdr,
299                              uint64_t offset, bool read)
300 {
301     uint8_t *buffer = NULL;
302     int ret;
303     VHDXHeader header_le;
304 
305     assert(bs_file != NULL);
306     assert(hdr != NULL);
307 
308     /* the header checksum is not over just the packed size of VHDXHeader,
309      * but rather over the entire 'reserved' range for the header, which is
310      * 4KB (VHDX_HEADER_SIZE). */
311 
312     buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE);
313     if (read) {
314         /* if true, we can't assume the extra reserved bytes are 0 */
315         ret = bdrv_pread(bs_file, offset, buffer, VHDX_HEADER_SIZE);
316         if (ret < 0) {
317             goto exit;
318         }
319     } else {
320         memset(buffer, 0, VHDX_HEADER_SIZE);
321     }
322 
323     /* overwrite the actual VHDXHeader portion */
324     memcpy(buffer, hdr, sizeof(VHDXHeader));
325     hdr->checksum = vhdx_update_checksum(buffer, VHDX_HEADER_SIZE,
326                                          offsetof(VHDXHeader, checksum));
327     vhdx_header_le_export(hdr, &header_le);
328     ret = bdrv_pwrite_sync(bs_file, offset, &header_le, sizeof(VHDXHeader));
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     vhdx_header_le_import(header1);
436 
437     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4) &&
438         !memcmp(&header1->signature, "head", 4)             &&
439         header1->version == 1) {
440         h1_seq = header1->sequence_number;
441         h1_valid = true;
442     }
443 
444     ret = bdrv_pread(bs->file, VHDX_HEADER2_OFFSET, buffer, VHDX_HEADER_SIZE);
445     if (ret < 0) {
446         goto fail;
447     }
448     /* copy over just the relevant portion that we need */
449     memcpy(header2, buffer, sizeof(VHDXHeader));
450     vhdx_header_le_import(header2);
451 
452     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4) &&
453         !memcmp(&header2->signature, "head", 4)             &&
454         header2->version == 1) {
455         h2_seq = header2->sequence_number;
456         h2_valid = true;
457     }
458 
459     /* If there is only 1 valid header (or no valid headers), we
460      * don't care what the sequence numbers are */
461     if (h1_valid && !h2_valid) {
462         s->curr_header = 0;
463     } else if (!h1_valid && h2_valid) {
464         s->curr_header = 1;
465     } else if (!h1_valid && !h2_valid) {
466         goto fail;
467     } else {
468         /* If both headers are valid, then we choose the active one by the
469          * highest sequence number.  If the sequence numbers are equal, that is
470          * invalid */
471         if (h1_seq > h2_seq) {
472             s->curr_header = 0;
473         } else if (h2_seq > h1_seq) {
474             s->curr_header = 1;
475         } else {
476             /* The Microsoft Disk2VHD tool will create 2 identical
477              * headers, with identical sequence numbers.  If the headers are
478              * identical, don't consider the file corrupt */
479             if (!memcmp(header1, header2, sizeof(VHDXHeader))) {
480                 s->curr_header = 0;
481             } else {
482                 goto fail;
483             }
484         }
485     }
486 
487     vhdx_region_register(s, s->headers[s->curr_header]->log_offset,
488                             s->headers[s->curr_header]->log_length);
489     goto exit;
490 
491 fail:
492     error_setg_errno(errp, -ret, "No valid VHDX header found");
493     qemu_vfree(header1);
494     qemu_vfree(header2);
495     s->headers[0] = NULL;
496     s->headers[1] = NULL;
497 exit:
498     qemu_vfree(buffer);
499 }
500 
501 
502 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s)
503 {
504     int ret = 0;
505     uint8_t *buffer;
506     int offset = 0;
507     VHDXRegionTableEntry rt_entry;
508     uint32_t i;
509     bool bat_rt_found = false;
510     bool metadata_rt_found = false;
511 
512     /* We have to read the whole 64KB block, because the crc32 is over the
513      * whole block */
514     buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE);
515 
516     ret = bdrv_pread(bs->file, VHDX_REGION_TABLE_OFFSET, buffer,
517                      VHDX_HEADER_BLOCK_SIZE);
518     if (ret < 0) {
519         goto fail;
520     }
521     memcpy(&s->rt, buffer, sizeof(s->rt));
522     vhdx_region_header_le_import(&s->rt);
523     offset += sizeof(s->rt);
524 
525     if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4) ||
526         memcmp(&s->rt.signature, "regi", 4)) {
527         ret = -EINVAL;
528         goto fail;
529     }
530 
531     /* Per spec, maximum region table entry count is 2047 */
532     if (s->rt.entry_count > 2047) {
533         ret = -EINVAL;
534         goto fail;
535     }
536 
537     for (i = 0; i < s->rt.entry_count; i++) {
538         memcpy(&rt_entry, buffer + offset, sizeof(rt_entry));
539         offset += sizeof(rt_entry);
540 
541         vhdx_region_entry_le_import(&rt_entry);
542 
543         /* check for region overlap between these entries, and any
544          * other memory regions in the file */
545         ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length);
546         if (ret < 0) {
547             goto fail;
548         }
549 
550         vhdx_region_register(s, rt_entry.file_offset, rt_entry.length);
551 
552         /* see if we recognize the entry */
553         if (guid_eq(rt_entry.guid, bat_guid)) {
554             /* must be unique; if we have already found it this is invalid */
555             if (bat_rt_found) {
556                 ret = -EINVAL;
557                 goto fail;
558             }
559             bat_rt_found = true;
560             s->bat_rt = rt_entry;
561             continue;
562         }
563 
564         if (guid_eq(rt_entry.guid, metadata_guid)) {
565             /* must be unique; if we have already found it this is invalid */
566             if (metadata_rt_found) {
567                 ret = -EINVAL;
568                 goto fail;
569             }
570             metadata_rt_found = true;
571             s->metadata_rt = rt_entry;
572             continue;
573         }
574 
575         if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) {
576             /* cannot read vhdx file - required region table entry that
577              * we do not understand.  per spec, we must fail to open */
578             ret = -ENOTSUP;
579             goto fail;
580         }
581     }
582 
583     if (!bat_rt_found || !metadata_rt_found) {
584         ret = -EINVAL;
585         goto fail;
586     }
587 
588     ret = 0;
589 
590 fail:
591     qemu_vfree(buffer);
592     return ret;
593 }
594 
595 
596 
597 /* Metadata initial parser
598  *
599  * This loads all the metadata entry fields.  This may cause additional
600  * fields to be processed (e.g. parent locator, etc..).
601  *
602  * There are 5 Metadata items that are always required:
603  *      - File Parameters (block size, has a parent)
604  *      - Virtual Disk Size (size, in bytes, of the virtual drive)
605  *      - Page 83 Data (scsi page 83 guid)
606  *      - Logical Sector Size (logical sector size in bytes, either 512 or
607  *                             4096.  We only support 512 currently)
608  *      - Physical Sector Size (512 or 4096)
609  *
610  * Also, if the File Parameters indicate this is a differencing file,
611  * we must also look for the Parent Locator metadata item.
612  */
613 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s)
614 {
615     int ret = 0;
616     uint8_t *buffer;
617     int offset = 0;
618     uint32_t i = 0;
619     VHDXMetadataTableEntry md_entry;
620 
621     buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE);
622 
623     ret = bdrv_pread(bs->file, s->metadata_rt.file_offset, buffer,
624                      VHDX_METADATA_TABLE_MAX_SIZE);
625     if (ret < 0) {
626         goto exit;
627     }
628     memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr));
629     offset += sizeof(s->metadata_hdr);
630 
631     vhdx_metadata_header_le_import(&s->metadata_hdr);
632 
633     if (memcmp(&s->metadata_hdr.signature, "metadata", 8)) {
634         ret = -EINVAL;
635         goto exit;
636     }
637 
638     s->metadata_entries.present = 0;
639 
640     if ((s->metadata_hdr.entry_count * sizeof(md_entry)) >
641         (VHDX_METADATA_TABLE_MAX_SIZE - offset)) {
642         ret = -EINVAL;
643         goto exit;
644     }
645 
646     for (i = 0; i < s->metadata_hdr.entry_count; i++) {
647         memcpy(&md_entry, buffer + offset, sizeof(md_entry));
648         offset += sizeof(md_entry);
649 
650         vhdx_metadata_entry_le_import(&md_entry);
651 
652         if (guid_eq(md_entry.item_id, file_param_guid)) {
653             if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) {
654                 ret = -EINVAL;
655                 goto exit;
656             }
657             s->metadata_entries.file_parameters_entry = md_entry;
658             s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT;
659             continue;
660         }
661 
662         if (guid_eq(md_entry.item_id, virtual_size_guid)) {
663             if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) {
664                 ret = -EINVAL;
665                 goto exit;
666             }
667             s->metadata_entries.virtual_disk_size_entry = md_entry;
668             s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT;
669             continue;
670         }
671 
672         if (guid_eq(md_entry.item_id, page83_guid)) {
673             if (s->metadata_entries.present & META_PAGE_83_PRESENT) {
674                 ret = -EINVAL;
675                 goto exit;
676             }
677             s->metadata_entries.page83_data_entry = md_entry;
678             s->metadata_entries.present |= META_PAGE_83_PRESENT;
679             continue;
680         }
681 
682         if (guid_eq(md_entry.item_id, logical_sector_guid)) {
683             if (s->metadata_entries.present &
684                 META_LOGICAL_SECTOR_SIZE_PRESENT) {
685                 ret = -EINVAL;
686                 goto exit;
687             }
688             s->metadata_entries.logical_sector_size_entry = md_entry;
689             s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT;
690             continue;
691         }
692 
693         if (guid_eq(md_entry.item_id, phys_sector_guid)) {
694             if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) {
695                 ret = -EINVAL;
696                 goto exit;
697             }
698             s->metadata_entries.phys_sector_size_entry = md_entry;
699             s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT;
700             continue;
701         }
702 
703         if (guid_eq(md_entry.item_id, parent_locator_guid)) {
704             if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
705                 ret = -EINVAL;
706                 goto exit;
707             }
708             s->metadata_entries.parent_locator_entry = md_entry;
709             s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT;
710             continue;
711         }
712 
713         if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) {
714             /* cannot read vhdx file - required region table entry that
715              * we do not understand.  per spec, we must fail to open */
716             ret = -ENOTSUP;
717             goto exit;
718         }
719     }
720 
721     if (s->metadata_entries.present != META_ALL_PRESENT) {
722         ret = -ENOTSUP;
723         goto exit;
724     }
725 
726     ret = bdrv_pread(bs->file,
727                      s->metadata_entries.file_parameters_entry.offset
728                                          + s->metadata_rt.file_offset,
729                      &s->params,
730                      sizeof(s->params));
731 
732     if (ret < 0) {
733         goto exit;
734     }
735 
736     le32_to_cpus(&s->params.block_size);
737     le32_to_cpus(&s->params.data_bits);
738 
739 
740     /* We now have the file parameters, so we can tell if this is a
741      * differencing file (i.e.. has_parent), is dynamic or fixed
742      * sized (leave_blocks_allocated), and the block size */
743 
744     /* The parent locator required iff the file parameters has_parent set */
745     if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
746         if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
747             /* TODO: parse  parent locator fields */
748             ret = -ENOTSUP; /* temp, until differencing files are supported */
749             goto exit;
750         } else {
751             /* if has_parent is set, but there is not parent locator present,
752              * then that is an invalid combination */
753             ret = -EINVAL;
754             goto exit;
755         }
756     }
757 
758     /* determine virtual disk size, logical sector size,
759      * and phys sector size */
760 
761     ret = bdrv_pread(bs->file,
762                      s->metadata_entries.virtual_disk_size_entry.offset
763                                            + s->metadata_rt.file_offset,
764                      &s->virtual_disk_size,
765                      sizeof(uint64_t));
766     if (ret < 0) {
767         goto exit;
768     }
769     ret = bdrv_pread(bs->file,
770                      s->metadata_entries.logical_sector_size_entry.offset
771                                              + s->metadata_rt.file_offset,
772                      &s->logical_sector_size,
773                      sizeof(uint32_t));
774     if (ret < 0) {
775         goto exit;
776     }
777     ret = bdrv_pread(bs->file,
778                      s->metadata_entries.phys_sector_size_entry.offset
779                                           + s->metadata_rt.file_offset,
780                      &s->physical_sector_size,
781                      sizeof(uint32_t));
782     if (ret < 0) {
783         goto exit;
784     }
785 
786     le64_to_cpus(&s->virtual_disk_size);
787     le32_to_cpus(&s->logical_sector_size);
788     le32_to_cpus(&s->physical_sector_size);
789 
790     if (s->params.block_size < VHDX_BLOCK_SIZE_MIN ||
791         s->params.block_size > VHDX_BLOCK_SIZE_MAX) {
792         ret = -EINVAL;
793         goto exit;
794     }
795 
796     /* only 2 supported sector sizes */
797     if (s->logical_sector_size != 512 && s->logical_sector_size != 4096) {
798         ret = -EINVAL;
799         goto exit;
800     }
801 
802     /* Both block_size and sector_size are guaranteed powers of 2, below.
803        Due to range checks above, s->sectors_per_block can never be < 256 */
804     s->sectors_per_block = s->params.block_size / s->logical_sector_size;
805     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
806                      (uint64_t)s->logical_sector_size /
807                      (uint64_t)s->params.block_size;
808 
809     /* These values are ones we will want to use for division / multiplication
810      * later on, and they are all guaranteed (per the spec) to be powers of 2,
811      * so we can take advantage of that for shift operations during
812      * reads/writes */
813     if (s->logical_sector_size & (s->logical_sector_size - 1)) {
814         ret = -EINVAL;
815         goto exit;
816     }
817     if (s->sectors_per_block & (s->sectors_per_block - 1)) {
818         ret = -EINVAL;
819         goto exit;
820     }
821     if (s->chunk_ratio & (s->chunk_ratio - 1)) {
822         ret = -EINVAL;
823         goto exit;
824     }
825     s->block_size = s->params.block_size;
826     if (s->block_size & (s->block_size - 1)) {
827         ret = -EINVAL;
828         goto exit;
829     }
830 
831     vhdx_set_shift_bits(s);
832 
833     ret = 0;
834 
835 exit:
836     qemu_vfree(buffer);
837     return ret;
838 }
839 
840 /*
841  * Calculate the number of BAT entries, including sector
842  * bitmap entries.
843  */
844 static void vhdx_calc_bat_entries(BDRVVHDXState *s)
845 {
846     uint32_t data_blocks_cnt, bitmap_blocks_cnt;
847 
848     data_blocks_cnt = s->virtual_disk_size >> s->block_size_bits;
849     if (s->virtual_disk_size - (data_blocks_cnt << s->block_size_bits)) {
850         data_blocks_cnt++;
851     }
852     bitmap_blocks_cnt = data_blocks_cnt >> s->chunk_ratio_bits;
853     if (data_blocks_cnt - (bitmap_blocks_cnt << s->chunk_ratio_bits)) {
854         bitmap_blocks_cnt++;
855     }
856 
857     if (s->parent_entries) {
858         s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1);
859     } else {
860         s->bat_entries = data_blocks_cnt +
861                          ((data_blocks_cnt - 1) >> s->chunk_ratio_bits);
862     }
863 
864 }
865 
866 static void vhdx_close(BlockDriverState *bs)
867 {
868     BDRVVHDXState *s = bs->opaque;
869     qemu_vfree(s->headers[0]);
870     s->headers[0] = NULL;
871     qemu_vfree(s->headers[1]);
872     s->headers[1] = NULL;
873     qemu_vfree(s->bat);
874     s->bat = NULL;
875     qemu_vfree(s->parent_entries);
876     s->parent_entries = NULL;
877     migrate_del_blocker(s->migration_blocker);
878     error_free(s->migration_blocker);
879     qemu_vfree(s->log.hdr);
880     s->log.hdr = NULL;
881     vhdx_region_unregister_all(s);
882 }
883 
884 static int vhdx_open(BlockDriverState *bs, QDict *options, int flags,
885                      Error **errp)
886 {
887     BDRVVHDXState *s = bs->opaque;
888     int ret = 0;
889     uint32_t i;
890     uint64_t signature;
891     Error *local_err = NULL;
892 
893     s->bat = NULL;
894     s->first_visible_write = true;
895 
896     qemu_co_mutex_init(&s->lock);
897     QLIST_INIT(&s->regions);
898 
899     /* validate the file signature */
900     ret = bdrv_pread(bs->file, 0, &signature, sizeof(uint64_t));
901     if (ret < 0) {
902         goto fail;
903     }
904     if (memcmp(&signature, "vhdxfile", 8)) {
905         ret = -EINVAL;
906         goto fail;
907     }
908 
909     /* This is used for any header updates, for the file_write_guid.
910      * The spec dictates that a new value should be used for the first
911      * header update */
912     vhdx_guid_generate(&s->session_guid);
913 
914     vhdx_parse_header(bs, s, &local_err);
915     if (local_err != NULL) {
916         error_propagate(errp, local_err);
917         ret = -EINVAL;
918         goto fail;
919     }
920 
921     ret = vhdx_parse_log(bs, s, &s->log_replayed_on_open, errp);
922     if (ret < 0) {
923         goto fail;
924     }
925 
926     ret = vhdx_open_region_tables(bs, s);
927     if (ret < 0) {
928         goto fail;
929     }
930 
931     ret = vhdx_parse_metadata(bs, s);
932     if (ret < 0) {
933         goto fail;
934     }
935 
936     s->block_size = s->params.block_size;
937 
938     /* the VHDX spec dictates that virtual_disk_size is always a multiple of
939      * logical_sector_size */
940     bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits;
941 
942     vhdx_calc_bat_entries(s);
943 
944     s->bat_offset = s->bat_rt.file_offset;
945 
946     if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) {
947         /* BAT allocation is not large enough for all entries */
948         ret = -EINVAL;
949         goto fail;
950     }
951 
952     /* s->bat is freed in vhdx_close() */
953     s->bat = qemu_blockalign(bs, s->bat_rt.length);
954 
955     ret = bdrv_pread(bs->file, s->bat_offset, s->bat, s->bat_rt.length);
956     if (ret < 0) {
957         goto fail;
958     }
959 
960     uint64_t payblocks = s->chunk_ratio;
961     /* endian convert, and verify populated BAT field file offsets against
962      * region table and log entries */
963     for (i = 0; i < s->bat_entries; i++) {
964         le64_to_cpus(&s->bat[i]);
965         if (payblocks--) {
966             /* payload bat entries */
967             if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) ==
968                     PAYLOAD_BLOCK_FULLY_PRESENT) {
969                 ret = vhdx_region_check(s, s->bat[i] & VHDX_BAT_FILE_OFF_MASK,
970                                         s->block_size);
971                 if (ret < 0) {
972                     goto fail;
973                 }
974             }
975         } else {
976             payblocks = s->chunk_ratio;
977             /* Once differencing files are supported, verify sector bitmap
978              * blocks here */
979         }
980     }
981 
982     if (flags & BDRV_O_RDWR) {
983         ret = vhdx_update_headers(bs, s, false, NULL);
984         if (ret < 0) {
985             goto fail;
986         }
987     }
988 
989     /* TODO: differencing files */
990 
991     /* Disable migration when VHDX images are used */
992     error_set(&s->migration_blocker,
993             QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
994             "vhdx", bs->device_name, "live migration");
995     migrate_add_blocker(s->migration_blocker);
996 
997     return 0;
998 fail:
999     vhdx_close(bs);
1000     return ret;
1001 }
1002 
1003 static int vhdx_reopen_prepare(BDRVReopenState *state,
1004                                BlockReopenQueue *queue, Error **errp)
1005 {
1006     return 0;
1007 }
1008 
1009 
1010 /*
1011  * Perform sector to block offset translations, to get various
1012  * sector and file offsets into the image.  See VHDXSectorInfo
1013  */
1014 static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num,
1015                                  int nb_sectors, VHDXSectorInfo *sinfo)
1016 {
1017     uint32_t block_offset;
1018 
1019     sinfo->bat_idx = sector_num >> s->sectors_per_block_bits;
1020     /* effectively a modulo - this gives us the offset into the block
1021      * (in sector sizes) for our sector number */
1022     block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits);
1023     /* the chunk ratio gives us the interleaving of the sector
1024      * bitmaps, so we need to advance our page block index by the
1025      * sector bitmaps entry number */
1026     sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits;
1027 
1028     /* the number of sectors we can read/write in this cycle */
1029     sinfo->sectors_avail = s->sectors_per_block - block_offset;
1030 
1031     sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits;
1032 
1033     if (sinfo->sectors_avail > nb_sectors) {
1034         sinfo->sectors_avail = nb_sectors;
1035     }
1036 
1037     sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits;
1038 
1039     sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK;
1040 
1041     sinfo->block_offset = block_offset << s->logical_sector_size_bits;
1042 
1043     /* The file offset must be past the header section, so must be > 0 */
1044     if (sinfo->file_offset == 0) {
1045         return;
1046     }
1047 
1048     /* block offset is the offset in vhdx logical sectors, in
1049      * the payload data block. Convert that to a byte offset
1050      * in the block, and add in the payload data block offset
1051      * in the file, in bytes, to get the final read address */
1052 
1053     sinfo->file_offset += sinfo->block_offset;
1054 }
1055 
1056 
1057 static int vhdx_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1058 {
1059     BDRVVHDXState *s = bs->opaque;
1060 
1061     bdi->cluster_size = s->block_size;
1062 
1063     bdi->unallocated_blocks_are_zero =
1064         (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) == 0;
1065 
1066     return 0;
1067 }
1068 
1069 
1070 static coroutine_fn int vhdx_co_readv(BlockDriverState *bs, int64_t sector_num,
1071                                       int nb_sectors, QEMUIOVector *qiov)
1072 {
1073     BDRVVHDXState *s = bs->opaque;
1074     int ret = 0;
1075     VHDXSectorInfo sinfo;
1076     uint64_t bytes_done = 0;
1077     QEMUIOVector hd_qiov;
1078 
1079     qemu_iovec_init(&hd_qiov, qiov->niov);
1080 
1081     qemu_co_mutex_lock(&s->lock);
1082 
1083     while (nb_sectors > 0) {
1084         /* We are a differencing file, so we need to inspect the sector bitmap
1085          * to see if we have the data or not */
1086         if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1087             /* not supported yet */
1088             ret = -ENOTSUP;
1089             goto exit;
1090         } else {
1091             vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1092 
1093             qemu_iovec_reset(&hd_qiov);
1094             qemu_iovec_concat(&hd_qiov, qiov,  bytes_done, sinfo.bytes_avail);
1095 
1096             /* check the payload block state */
1097             switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) {
1098             case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1099             case PAYLOAD_BLOCK_UNDEFINED:   /* fall through */
1100             case PAYLOAD_BLOCK_UNMAPPED:    /* fall through */
1101             case PAYLOAD_BLOCK_ZERO:
1102                 /* return zero */
1103                 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail);
1104                 break;
1105             case PAYLOAD_BLOCK_FULLY_PRESENT:
1106                 qemu_co_mutex_unlock(&s->lock);
1107                 ret = bdrv_co_readv(bs->file,
1108                                     sinfo.file_offset >> BDRV_SECTOR_BITS,
1109                                     sinfo.sectors_avail, &hd_qiov);
1110                 qemu_co_mutex_lock(&s->lock);
1111                 if (ret < 0) {
1112                     goto exit;
1113                 }
1114                 break;
1115             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1116                 /* we don't yet support difference files, fall through
1117                  * to error */
1118             default:
1119                 ret = -EIO;
1120                 goto exit;
1121                 break;
1122             }
1123             nb_sectors -= sinfo.sectors_avail;
1124             sector_num += sinfo.sectors_avail;
1125             bytes_done += sinfo.bytes_avail;
1126         }
1127     }
1128     ret = 0;
1129 exit:
1130     qemu_co_mutex_unlock(&s->lock);
1131     qemu_iovec_destroy(&hd_qiov);
1132     return ret;
1133 }
1134 
1135 /*
1136  * Allocate a new payload block at the end of the file.
1137  *
1138  * Allocation will happen at 1MB alignment inside the file
1139  *
1140  * Returns the file offset start of the new payload block
1141  */
1142 static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s,
1143                                     uint64_t *new_offset)
1144 {
1145     *new_offset = bdrv_getlength(bs->file);
1146 
1147     /* per the spec, the address for a block is in units of 1MB */
1148     *new_offset = ROUND_UP(*new_offset, 1024 * 1024);
1149 
1150     return bdrv_truncate(bs->file, *new_offset + s->block_size);
1151 }
1152 
1153 /*
1154  * Update the BAT table entry with the new file offset, and the new entry
1155  * state */
1156 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s,
1157                                        VHDXSectorInfo *sinfo,
1158                                        uint64_t *bat_entry_le,
1159                                        uint64_t *bat_offset, int state)
1160 {
1161     /* The BAT entry is a uint64, with 44 bits for the file offset in units of
1162      * 1MB, and 3 bits for the block state. */
1163     s->bat[sinfo->bat_idx]  = sinfo->file_offset;
1164 
1165     s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK;
1166 
1167     *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]);
1168     *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry);
1169 
1170 }
1171 
1172 /* Per the spec, on the first write of guest-visible data to the file the
1173  * data write guid must be updated in the header */
1174 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s)
1175 {
1176     int ret = 0;
1177     if (s->first_visible_write) {
1178         s->first_visible_write = false;
1179         ret = vhdx_update_headers(bs, s, true, NULL);
1180     }
1181     return ret;
1182 }
1183 
1184 static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num,
1185                                       int nb_sectors, QEMUIOVector *qiov)
1186 {
1187     int ret = -ENOTSUP;
1188     BDRVVHDXState *s = bs->opaque;
1189     VHDXSectorInfo sinfo;
1190     uint64_t bytes_done = 0;
1191     uint64_t bat_entry = 0;
1192     uint64_t bat_entry_offset = 0;
1193     QEMUIOVector hd_qiov;
1194     struct iovec iov1 = { 0 };
1195     struct iovec iov2 = { 0 };
1196     int sectors_to_write;
1197     int bat_state;
1198     uint64_t bat_prior_offset = 0;
1199     bool bat_update = false;
1200 
1201     qemu_iovec_init(&hd_qiov, qiov->niov);
1202 
1203     qemu_co_mutex_lock(&s->lock);
1204 
1205     ret = vhdx_user_visible_write(bs, s);
1206     if (ret < 0) {
1207         goto exit;
1208     }
1209 
1210     while (nb_sectors > 0) {
1211         bool use_zero_buffers = false;
1212         bat_update = false;
1213         if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1214             /* not supported yet */
1215             ret = -ENOTSUP;
1216             goto exit;
1217         } else {
1218             vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1219             sectors_to_write = sinfo.sectors_avail;
1220 
1221             qemu_iovec_reset(&hd_qiov);
1222             /* check the payload block state */
1223             bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK;
1224             switch (bat_state) {
1225             case PAYLOAD_BLOCK_ZERO:
1226                 /* in this case, we need to preserve zero writes for
1227                  * data that is not part of this write, so we must pad
1228                  * the rest of the buffer to zeroes */
1229 
1230                 /* if we are on a posix system with ftruncate() that extends
1231                  * a file, then it is zero-filled for us.  On Win32, the raw
1232                  * layer uses SetFilePointer and SetFileEnd, which does not
1233                  * zero fill AFAIK */
1234 
1235                 /* Queue another write of zero buffers if the underlying file
1236                  * does not zero-fill on file extension */
1237 
1238                 if (bdrv_has_zero_init(bs->file) == 0) {
1239                     use_zero_buffers = true;
1240 
1241                     /* zero fill the front, if any */
1242                     if (sinfo.block_offset) {
1243                         iov1.iov_len = sinfo.block_offset;
1244                         iov1.iov_base = qemu_blockalign(bs, iov1.iov_len);
1245                         memset(iov1.iov_base, 0, iov1.iov_len);
1246                         qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0,
1247                                               sinfo.block_offset);
1248                         sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS;
1249                     }
1250 
1251                     /* our actual data */
1252                     qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1253                                       sinfo.bytes_avail);
1254 
1255                     /* zero fill the back, if any */
1256                     if ((sinfo.bytes_avail - sinfo.block_offset) <
1257                          s->block_size) {
1258                         iov2.iov_len = s->block_size -
1259                                       (sinfo.bytes_avail + sinfo.block_offset);
1260                         iov2.iov_base = qemu_blockalign(bs, iov2.iov_len);
1261                         memset(iov2.iov_base, 0, iov2.iov_len);
1262                         qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0,
1263                                               sinfo.block_offset);
1264                         sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS;
1265                     }
1266                 }
1267 
1268                 /* fall through */
1269             case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1270             case PAYLOAD_BLOCK_UNMAPPED:    /* fall through */
1271             case PAYLOAD_BLOCK_UNDEFINED:   /* fall through */
1272                 bat_prior_offset = sinfo.file_offset;
1273                 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset);
1274                 if (ret < 0) {
1275                     goto exit;
1276                 }
1277                 /* once we support differencing files, this may also be
1278                  * partially present */
1279                 /* update block state to the newly specified state */
1280                 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1281                                             &bat_entry_offset,
1282                                             PAYLOAD_BLOCK_FULLY_PRESENT);
1283                 bat_update = true;
1284                 /* since we just allocated a block, file_offset is the
1285                  * beginning of the payload block. It needs to be the
1286                  * write address, which includes the offset into the block */
1287                 if (!use_zero_buffers) {
1288                     sinfo.file_offset += sinfo.block_offset;
1289                 }
1290                 /* fall through */
1291             case PAYLOAD_BLOCK_FULLY_PRESENT:
1292                 /* if the file offset address is in the header zone,
1293                  * there is a problem */
1294                 if (sinfo.file_offset < (1024 * 1024)) {
1295                     ret = -EFAULT;
1296                     goto error_bat_restore;
1297                 }
1298 
1299                 if (!use_zero_buffers) {
1300                     qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1301                                       sinfo.bytes_avail);
1302                 }
1303                 /* block exists, so we can just overwrite it */
1304                 qemu_co_mutex_unlock(&s->lock);
1305                 ret = bdrv_co_writev(bs->file,
1306                                     sinfo.file_offset >> BDRV_SECTOR_BITS,
1307                                     sectors_to_write, &hd_qiov);
1308                 qemu_co_mutex_lock(&s->lock);
1309                 if (ret < 0) {
1310                     goto error_bat_restore;
1311                 }
1312                 break;
1313             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1314                 /* we don't yet support difference files, fall through
1315                  * to error */
1316             default:
1317                 ret = -EIO;
1318                 goto exit;
1319                 break;
1320             }
1321 
1322             if (bat_update) {
1323                 /* this will update the BAT entry into the log journal, and
1324                  * then flush the log journal out to disk */
1325                 ret =  vhdx_log_write_and_flush(bs, s, &bat_entry,
1326                                                 sizeof(VHDXBatEntry),
1327                                                 bat_entry_offset);
1328                 if (ret < 0) {
1329                     goto exit;
1330                 }
1331             }
1332 
1333             nb_sectors -= sinfo.sectors_avail;
1334             sector_num += sinfo.sectors_avail;
1335             bytes_done += sinfo.bytes_avail;
1336 
1337         }
1338     }
1339 
1340     goto exit;
1341 
1342 error_bat_restore:
1343     if (bat_update) {
1344         /* keep metadata in sync, and restore the bat entry state
1345          * if error. */
1346         sinfo.file_offset = bat_prior_offset;
1347         vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1348                                     &bat_entry_offset, bat_state);
1349     }
1350 exit:
1351     qemu_vfree(iov1.iov_base);
1352     qemu_vfree(iov2.iov_base);
1353     qemu_co_mutex_unlock(&s->lock);
1354     qemu_iovec_destroy(&hd_qiov);
1355     return ret;
1356 }
1357 
1358 
1359 
1360 /*
1361  * Create VHDX Headers
1362  *
1363  * There are 2 headers, and the highest sequence number will represent
1364  * the active header
1365  */
1366 static int vhdx_create_new_headers(BlockDriverState *bs, uint64_t image_size,
1367                                    uint32_t log_size)
1368 {
1369     int ret = 0;
1370     VHDXHeader *hdr = NULL;
1371 
1372     hdr = g_malloc0(sizeof(VHDXHeader));
1373 
1374     hdr->signature       = VHDX_HEADER_SIGNATURE;
1375     hdr->sequence_number = g_random_int();
1376     hdr->log_version     = 0;
1377     hdr->version         = 1;
1378     hdr->log_length      = log_size;
1379     hdr->log_offset      = VHDX_HEADER_SECTION_END;
1380     vhdx_guid_generate(&hdr->file_write_guid);
1381     vhdx_guid_generate(&hdr->data_write_guid);
1382 
1383     ret = vhdx_write_header(bs, hdr, VHDX_HEADER1_OFFSET, false);
1384     if (ret < 0) {
1385         goto exit;
1386     }
1387     hdr->sequence_number++;
1388     ret = vhdx_write_header(bs, hdr, VHDX_HEADER2_OFFSET, false);
1389     if (ret < 0) {
1390         goto exit;
1391     }
1392 
1393 exit:
1394     g_free(hdr);
1395     return ret;
1396 }
1397 
1398 
1399 /*
1400  * Create the Metadata entries.
1401  *
1402  * For more details on the entries, see section 3.5 (pg 29) in the
1403  * VHDX 1.00 specification.
1404  *
1405  * We support 5 metadata entries (all required by spec):
1406  *          File Parameters,
1407  *          Virtual Disk Size,
1408  *          Page 83 Data,
1409  *          Logical Sector Size,
1410  *          Physical Sector Size
1411  *
1412  * The first 64KB of the Metadata section is reserved for the metadata
1413  * header and entries; beyond that, the metadata items themselves reside.
1414  */
1415 static int vhdx_create_new_metadata(BlockDriverState *bs,
1416                                     uint64_t image_size,
1417                                     uint32_t block_size,
1418                                     uint32_t sector_size,
1419                                     uint64_t metadata_offset,
1420                                     VHDXImageType type)
1421 {
1422     int ret = 0;
1423     uint32_t offset = 0;
1424     void *buffer = NULL;
1425     void *entry_buffer;
1426     VHDXMetadataTableHeader *md_table;;
1427     VHDXMetadataTableEntry  *md_table_entry;
1428 
1429     /* Metadata entries */
1430     VHDXFileParameters     *mt_file_params;
1431     VHDXVirtualDiskSize    *mt_virtual_size;
1432     VHDXPage83Data         *mt_page83;
1433     VHDXVirtualDiskLogicalSectorSize  *mt_log_sector_size;
1434     VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size;
1435 
1436     entry_buffer = g_malloc0(sizeof(VHDXFileParameters)               +
1437                              sizeof(VHDXVirtualDiskSize)              +
1438                              sizeof(VHDXPage83Data)                   +
1439                              sizeof(VHDXVirtualDiskLogicalSectorSize) +
1440                              sizeof(VHDXVirtualDiskPhysicalSectorSize));
1441 
1442     mt_file_params = entry_buffer;
1443     offset += sizeof(VHDXFileParameters);
1444     mt_virtual_size = entry_buffer + offset;
1445     offset += sizeof(VHDXVirtualDiskSize);
1446     mt_page83 = entry_buffer + offset;
1447     offset += sizeof(VHDXPage83Data);
1448     mt_log_sector_size = entry_buffer + offset;
1449     offset += sizeof(VHDXVirtualDiskLogicalSectorSize);
1450     mt_phys_sector_size = entry_buffer + offset;
1451 
1452     mt_file_params->block_size = cpu_to_le32(block_size);
1453     if (type == VHDX_TYPE_FIXED) {
1454         mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED;
1455         cpu_to_le32s(&mt_file_params->data_bits);
1456     }
1457 
1458     vhdx_guid_generate(&mt_page83->page_83_data);
1459     cpu_to_leguids(&mt_page83->page_83_data);
1460     mt_virtual_size->virtual_disk_size        = cpu_to_le64(image_size);
1461     mt_log_sector_size->logical_sector_size   = cpu_to_le32(sector_size);
1462     mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size);
1463 
1464     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1465     md_table = buffer;
1466 
1467     md_table->signature   = VHDX_METADATA_SIGNATURE;
1468     md_table->entry_count = 5;
1469     vhdx_metadata_header_le_export(md_table);
1470 
1471 
1472     /* This will reference beyond the reserved table portion */
1473     offset = 64 * KiB;
1474 
1475     md_table_entry = buffer + sizeof(VHDXMetadataTableHeader);
1476 
1477     md_table_entry[0].item_id = file_param_guid;
1478     md_table_entry[0].offset  = offset;
1479     md_table_entry[0].length  = sizeof(VHDXFileParameters);
1480     md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED;
1481     offset += md_table_entry[0].length;
1482     vhdx_metadata_entry_le_export(&md_table_entry[0]);
1483 
1484     md_table_entry[1].item_id = virtual_size_guid;
1485     md_table_entry[1].offset  = offset;
1486     md_table_entry[1].length  = sizeof(VHDXVirtualDiskSize);
1487     md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1488                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1489     offset += md_table_entry[1].length;
1490     vhdx_metadata_entry_le_export(&md_table_entry[1]);
1491 
1492     md_table_entry[2].item_id = page83_guid;
1493     md_table_entry[2].offset  = offset;
1494     md_table_entry[2].length  = sizeof(VHDXPage83Data);
1495     md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1496                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1497     offset += md_table_entry[2].length;
1498     vhdx_metadata_entry_le_export(&md_table_entry[2]);
1499 
1500     md_table_entry[3].item_id = logical_sector_guid;
1501     md_table_entry[3].offset  = offset;
1502     md_table_entry[3].length  = sizeof(VHDXVirtualDiskLogicalSectorSize);
1503     md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1504                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1505     offset += md_table_entry[3].length;
1506     vhdx_metadata_entry_le_export(&md_table_entry[3]);
1507 
1508     md_table_entry[4].item_id = phys_sector_guid;
1509     md_table_entry[4].offset  = offset;
1510     md_table_entry[4].length  = sizeof(VHDXVirtualDiskPhysicalSectorSize);
1511     md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1512                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1513     vhdx_metadata_entry_le_export(&md_table_entry[4]);
1514 
1515     ret = bdrv_pwrite(bs, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE);
1516     if (ret < 0) {
1517         goto exit;
1518     }
1519 
1520     ret = bdrv_pwrite(bs, metadata_offset + (64 * KiB), entry_buffer,
1521                       VHDX_HEADER_BLOCK_SIZE);
1522     if (ret < 0) {
1523         goto exit;
1524     }
1525 
1526 
1527 exit:
1528     g_free(buffer);
1529     g_free(entry_buffer);
1530     return ret;
1531 }
1532 
1533 /* This create the actual BAT itself.  We currently only support
1534  * 'Dynamic' and 'Fixed' image types.
1535  *
1536  *  Dynamic images: default state of the BAT is all zeroes.
1537  *
1538  *  Fixed images: default state of the BAT is fully populated, with
1539  *                file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT.
1540  */
1541 static int vhdx_create_bat(BlockDriverState *bs, BDRVVHDXState *s,
1542                            uint64_t image_size, VHDXImageType type,
1543                            bool use_zero_blocks, VHDXRegionTableEntry *rt_bat)
1544 {
1545     int ret = 0;
1546     uint64_t data_file_offset;
1547     uint64_t total_sectors = 0;
1548     uint64_t sector_num = 0;
1549     uint64_t unused;
1550     int block_state;
1551     VHDXSectorInfo sinfo;
1552 
1553     assert(s->bat == NULL);
1554 
1555     /* this gives a data start after BAT/bitmap entries, and well
1556      * past any metadata entries (with a 4 MB buffer for future
1557      * expansion */
1558     data_file_offset = rt_bat->file_offset + rt_bat->length + 5 * MiB;
1559     total_sectors = image_size >> s->logical_sector_size_bits;
1560 
1561     if (type == VHDX_TYPE_DYNAMIC) {
1562         /* All zeroes, so we can just extend the file - the end of the BAT
1563          * is the furthest thing we have written yet */
1564         ret = bdrv_truncate(bs, data_file_offset);
1565         if (ret < 0) {
1566             goto exit;
1567         }
1568     } else if (type == VHDX_TYPE_FIXED) {
1569         ret = bdrv_truncate(bs, data_file_offset + image_size);
1570         if (ret < 0) {
1571             goto exit;
1572         }
1573     } else {
1574         ret = -ENOTSUP;
1575         goto exit;
1576     }
1577 
1578     if (type == VHDX_TYPE_FIXED ||
1579                 use_zero_blocks ||
1580                 bdrv_has_zero_init(bs) == 0) {
1581         /* for a fixed file, the default BAT entry is not zero */
1582         s->bat = g_malloc0(rt_bat->length);
1583         block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT :
1584                                                 PAYLOAD_BLOCK_NOT_PRESENT;
1585         block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state;
1586         /* fill the BAT by emulating sector writes of sectors_per_block size */
1587         while (sector_num < total_sectors) {
1588             vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo);
1589             sinfo.file_offset = data_file_offset +
1590                                 (sector_num << s->logical_sector_size_bits);
1591             sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB);
1592             vhdx_update_bat_table_entry(bs, s, &sinfo, &unused, &unused,
1593                                         block_state);
1594             cpu_to_le64s(&s->bat[sinfo.bat_idx]);
1595             sector_num += s->sectors_per_block;
1596         }
1597         ret = bdrv_pwrite(bs, rt_bat->file_offset, s->bat, rt_bat->length);
1598         if (ret < 0) {
1599             goto exit;
1600         }
1601     }
1602 
1603 
1604 
1605 exit:
1606     g_free(s->bat);
1607     return ret;
1608 }
1609 
1610 /* Creates the region table header, and region table entries.
1611  * There are 2 supported region table entries: BAT, and Metadata/
1612  *
1613  * As the calculations for the BAT region table are also needed
1614  * to create the BAT itself, we will also cause the BAT to be
1615  * created.
1616  */
1617 static int vhdx_create_new_region_table(BlockDriverState *bs,
1618                                         uint64_t image_size,
1619                                         uint32_t block_size,
1620                                         uint32_t sector_size,
1621                                         uint32_t log_size,
1622                                         bool use_zero_blocks,
1623                                         VHDXImageType type,
1624                                         uint64_t *metadata_offset)
1625 {
1626     int ret = 0;
1627     uint32_t offset = 0;
1628     void *buffer = NULL;
1629     BDRVVHDXState *s = NULL;
1630     VHDXRegionTableHeader *region_table;
1631     VHDXRegionTableEntry *rt_bat;
1632     VHDXRegionTableEntry *rt_metadata;
1633 
1634     assert(metadata_offset != NULL);
1635 
1636     /* Populate enough of the BDRVVHDXState to be able to use the
1637      * pre-existing BAT calculation, translation, and update functions */
1638     s = g_malloc0(sizeof(BDRVVHDXState));
1639 
1640     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
1641                      (uint64_t) sector_size / (uint64_t) block_size;
1642 
1643     s->sectors_per_block = block_size / sector_size;
1644     s->virtual_disk_size = image_size;
1645     s->block_size = block_size;
1646     s->logical_sector_size = sector_size;
1647 
1648     vhdx_set_shift_bits(s);
1649 
1650     vhdx_calc_bat_entries(s);
1651 
1652     /* At this point the VHDX state is populated enough for creation */
1653 
1654     /* a single buffer is used so we can calculate the checksum over the
1655      * entire 64KB block */
1656     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1657     region_table = buffer;
1658     offset += sizeof(VHDXRegionTableHeader);
1659     rt_bat = buffer + offset;
1660     offset += sizeof(VHDXRegionTableEntry);
1661     rt_metadata  = buffer + offset;
1662 
1663     region_table->signature = VHDX_REGION_SIGNATURE;
1664     region_table->entry_count = 2;   /* BAT and Metadata */
1665 
1666     rt_bat->guid        = bat_guid;
1667     rt_bat->length      = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB);
1668     rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB);
1669     s->bat_offset = rt_bat->file_offset;
1670 
1671     rt_metadata->guid        = metadata_guid;
1672     rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length,
1673                                         MiB);
1674     rt_metadata->length      = 1 * MiB; /* min size, and more than enough */
1675     *metadata_offset = rt_metadata->file_offset;
1676 
1677     vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE,
1678                          offsetof(VHDXRegionTableHeader, checksum));
1679 
1680 
1681     /* The region table gives us the data we need to create the BAT,
1682      * so do that now */
1683     ret = vhdx_create_bat(bs, s, image_size, type, use_zero_blocks, rt_bat);
1684 
1685     /* Now write out the region headers to disk */
1686     vhdx_region_header_le_export(region_table);
1687     vhdx_region_entry_le_export(rt_bat);
1688     vhdx_region_entry_le_export(rt_metadata);
1689 
1690     ret = bdrv_pwrite(bs, VHDX_REGION_TABLE_OFFSET, buffer,
1691                       VHDX_HEADER_BLOCK_SIZE);
1692     if (ret < 0) {
1693         goto exit;
1694     }
1695 
1696     ret = bdrv_pwrite(bs, VHDX_REGION_TABLE2_OFFSET, buffer,
1697                       VHDX_HEADER_BLOCK_SIZE);
1698     if (ret < 0) {
1699         goto exit;
1700     }
1701 
1702 
1703 exit:
1704     g_free(s);
1705     g_free(buffer);
1706     return ret;
1707 }
1708 
1709 /* We need to create the following elements:
1710  *
1711  *    .-----------------------------------------------------------------.
1712  *    |   (A)    |   (B)    |    (C)    |     (D)       |     (E)       |
1713  *    |  File ID |  Header1 |  Header 2 |  Region Tbl 1 |  Region Tbl 2 |
1714  *    |          |          |           |               |               |
1715  *    .-----------------------------------------------------------------.
1716  *    0         64KB      128KB       192KB           256KB           320KB
1717  *
1718  *
1719  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1720  *    |     (F)     |     (G)       |    (H)    |                        |
1721  *    | Journal Log |  BAT / Bitmap |  Metadata |  .... data ......      |
1722  *    |             |               |           |                        |
1723  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1724  *   1MB
1725  */
1726 static int vhdx_create(const char *filename, QemuOpts *opts, Error **errp)
1727 {
1728     int ret = 0;
1729     uint64_t image_size = (uint64_t) 2 * GiB;
1730     uint32_t log_size   = 1 * MiB;
1731     uint32_t block_size = 0;
1732     uint64_t signature;
1733     uint64_t metadata_offset;
1734     bool use_zero_blocks = false;
1735 
1736     gunichar2 *creator = NULL;
1737     glong creator_items;
1738     BlockDriverState *bs;
1739     char *type = NULL;
1740     VHDXImageType image_type;
1741     Error *local_err = NULL;
1742 
1743     image_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
1744     log_size = qemu_opt_get_size_del(opts, VHDX_BLOCK_OPT_LOG_SIZE, 0);
1745     block_size = qemu_opt_get_size_del(opts, VHDX_BLOCK_OPT_BLOCK_SIZE, 0);
1746     type = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1747     use_zero_blocks = qemu_opt_get_bool_del(opts, VHDX_BLOCK_OPT_ZERO, false);
1748 
1749     if (image_size > VHDX_MAX_IMAGE_SIZE) {
1750         error_setg_errno(errp, EINVAL, "Image size too large; max of 64TB");
1751         ret = -EINVAL;
1752         goto exit;
1753     }
1754 
1755     if (type == NULL) {
1756         type = g_strdup("dynamic");
1757     }
1758 
1759     if (!strcmp(type, "dynamic")) {
1760         image_type = VHDX_TYPE_DYNAMIC;
1761     } else if (!strcmp(type, "fixed")) {
1762         image_type = VHDX_TYPE_FIXED;
1763     } else if (!strcmp(type, "differencing")) {
1764         error_setg_errno(errp, ENOTSUP,
1765                          "Differencing files not yet supported");
1766         ret = -ENOTSUP;
1767         goto exit;
1768     } else {
1769         ret = -EINVAL;
1770         goto exit;
1771     }
1772 
1773     /* These are pretty arbitrary, and mainly designed to keep the BAT
1774      * size reasonable to load into RAM */
1775     if (block_size == 0) {
1776         if (image_size > 32 * TiB) {
1777             block_size = 64 * MiB;
1778         } else if (image_size > (uint64_t) 100 * GiB) {
1779             block_size = 32 * MiB;
1780         } else if (image_size > 1 * GiB) {
1781             block_size = 16 * MiB;
1782         } else {
1783             block_size = 8 * MiB;
1784         }
1785     }
1786 
1787 
1788     /* make the log size close to what was specified, but must be
1789      * min 1MB, and multiple of 1MB */
1790     log_size = ROUND_UP(log_size, MiB);
1791 
1792     block_size = ROUND_UP(block_size, MiB);
1793     block_size = block_size > VHDX_BLOCK_SIZE_MAX ? VHDX_BLOCK_SIZE_MAX :
1794                                                     block_size;
1795 
1796     ret = bdrv_create_file(filename, opts, &local_err);
1797     if (ret < 0) {
1798         error_propagate(errp, local_err);
1799         goto exit;
1800     }
1801 
1802     bs = NULL;
1803     ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1804                     NULL, &local_err);
1805     if (ret < 0) {
1806         error_propagate(errp, local_err);
1807         goto exit;
1808     }
1809 
1810     /* Create (A) */
1811 
1812     /* The creator field is optional, but may be useful for
1813      * debugging / diagnostics */
1814     creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL,
1815                               &creator_items, NULL);
1816     signature = cpu_to_le64(VHDX_FILE_SIGNATURE);
1817     ret = bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature));
1818     if (ret < 0) {
1819         goto delete_and_exit;
1820     }
1821     if (creator) {
1822         ret = bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET + sizeof(signature),
1823                           creator, creator_items * sizeof(gunichar2));
1824         if (ret < 0) {
1825             goto delete_and_exit;
1826         }
1827     }
1828 
1829 
1830     /* Creates (B),(C) */
1831     ret = vhdx_create_new_headers(bs, image_size, log_size);
1832     if (ret < 0) {
1833         goto delete_and_exit;
1834     }
1835 
1836     /* Creates (D),(E),(G) explicitly. (F) created as by-product */
1837     ret = vhdx_create_new_region_table(bs, image_size, block_size, 512,
1838                                        log_size, use_zero_blocks, image_type,
1839                                        &metadata_offset);
1840     if (ret < 0) {
1841         goto delete_and_exit;
1842     }
1843 
1844     /* Creates (H) */
1845     ret = vhdx_create_new_metadata(bs, image_size, block_size, 512,
1846                                    metadata_offset, image_type);
1847     if (ret < 0) {
1848         goto delete_and_exit;
1849     }
1850 
1851 
1852 
1853 delete_and_exit:
1854     bdrv_unref(bs);
1855 exit:
1856     g_free(type);
1857     g_free(creator);
1858     return ret;
1859 }
1860 
1861 /* If opened r/w, the VHDX driver will automatically replay the log,
1862  * if one is present, inside the vhdx_open() call.
1863  *
1864  * If qemu-img check -r all is called, the image is automatically opened
1865  * r/w and any log has already been replayed, so there is nothing (currently)
1866  * for us to do here
1867  */
1868 static int vhdx_check(BlockDriverState *bs, BdrvCheckResult *result,
1869                        BdrvCheckMode fix)
1870 {
1871     BDRVVHDXState *s = bs->opaque;
1872 
1873     if (s->log_replayed_on_open) {
1874         result->corruptions_fixed++;
1875     }
1876     return 0;
1877 }
1878 
1879 static QemuOptsList vhdx_create_opts = {
1880     .name = "vhdx-create-opts",
1881     .head = QTAILQ_HEAD_INITIALIZER(vhdx_create_opts.head),
1882     .desc = {
1883         {
1884            .name = BLOCK_OPT_SIZE,
1885            .type = QEMU_OPT_SIZE,
1886            .help = "Virtual disk size; max of 64TB."
1887        },
1888        {
1889            .name = VHDX_BLOCK_OPT_LOG_SIZE,
1890            .type = QEMU_OPT_SIZE,
1891            .def_value_str = stringify(DEFAULT_LOG_SIZE),
1892            .help = "Log size; min 1MB."
1893        },
1894        {
1895            .name = VHDX_BLOCK_OPT_BLOCK_SIZE,
1896            .type = QEMU_OPT_SIZE,
1897            .def_value_str = stringify(0),
1898            .help = "Block Size; min 1MB, max 256MB. " \
1899                    "0 means auto-calculate based on image size."
1900        },
1901        {
1902            .name = BLOCK_OPT_SUBFMT,
1903            .type = QEMU_OPT_STRING,
1904            .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "\
1905                    "Default is 'dynamic'."
1906        },
1907        {
1908            .name = VHDX_BLOCK_OPT_ZERO,
1909            .type = QEMU_OPT_BOOL,
1910            .help = "Force use of payload blocks of type 'ZERO'.  Non-standard."
1911        },
1912        { NULL }
1913     }
1914 };
1915 
1916 static BlockDriver bdrv_vhdx = {
1917     .format_name            = "vhdx",
1918     .instance_size          = sizeof(BDRVVHDXState),
1919     .bdrv_probe             = vhdx_probe,
1920     .bdrv_open              = vhdx_open,
1921     .bdrv_close             = vhdx_close,
1922     .bdrv_reopen_prepare    = vhdx_reopen_prepare,
1923     .bdrv_co_readv          = vhdx_co_readv,
1924     .bdrv_co_writev         = vhdx_co_writev,
1925     .bdrv_create            = vhdx_create,
1926     .bdrv_get_info          = vhdx_get_info,
1927     .bdrv_check             = vhdx_check,
1928 
1929     .create_opts            = &vhdx_create_opts,
1930 };
1931 
1932 static void bdrv_vhdx_init(void)
1933 {
1934     bdrv_register(&bdrv_vhdx);
1935 }
1936 
1937 block_init(bdrv_vhdx_init);
1938