xref: /openbmc/qemu/block/vhdx.c (revision 228aa992)
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 __attribute__((unused))
103                                      = { .data1 = 0xb04aefb7,
104                                          .data2 = 0xd19e,
105                                          .data3 = 0x4a81,
106                                          .data4 = { 0xb7, 0x89, 0x25, 0xb8,
107                                                     0xe9, 0x44, 0x59, 0x13} };
108 
109 
110 #define META_FILE_PARAMETER_PRESENT      0x01
111 #define META_VIRTUAL_DISK_SIZE_PRESENT   0x02
112 #define META_PAGE_83_PRESENT             0x04
113 #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08
114 #define META_PHYS_SECTOR_SIZE_PRESENT    0x10
115 #define META_PARENT_LOCATOR_PRESENT      0x20
116 
117 #define META_ALL_PRESENT    \
118     (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \
119      META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \
120      META_PHYS_SECTOR_SIZE_PRESENT)
121 
122 
123 typedef struct VHDXSectorInfo {
124     uint32_t bat_idx;       /* BAT entry index */
125     uint32_t sectors_avail; /* sectors available in payload block */
126     uint32_t bytes_left;    /* bytes left in the block after data to r/w */
127     uint32_t bytes_avail;   /* bytes available in payload block */
128     uint64_t file_offset;   /* absolute offset in bytes, in file */
129     uint64_t block_offset;  /* block offset, in bytes */
130 } VHDXSectorInfo;
131 
132 /* Calculates new checksum.
133  *
134  * Zero is substituted during crc calculation for the original crc field
135  * crc_offset: byte offset in buf of the buffer crc
136  * buf: buffer pointer
137  * size: size of buffer (must be > crc_offset+4)
138  *
139  * Note: The buffer should have all multi-byte data in little-endian format,
140  *       and the resulting checksum is in little endian format.
141  */
142 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset)
143 {
144     uint32_t crc;
145 
146     assert(buf != NULL);
147     assert(size > (crc_offset + sizeof(crc)));
148 
149     memset(buf + crc_offset, 0, sizeof(crc));
150     crc =  crc32c(0xffffffff, buf, size);
151     cpu_to_le32s(&crc);
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     header_le = (VHDXHeader *)buffer;
325     memcpy(header_le, hdr, sizeof(VHDXHeader));
326     vhdx_header_le_export(hdr, header_le);
327     vhdx_update_checksum(buffer, VHDX_HEADER_SIZE,
328                          offsetof(VHDXHeader, checksum));
329     ret = bdrv_pwrite_sync(bs_file, offset, header_le, sizeof(VHDXHeader));
330 
331 exit:
332     qemu_vfree(buffer);
333     return ret;
334 }
335 
336 /* Update the VHDX headers
337  *
338  * This follows the VHDX spec procedures for header updates.
339  *
340  *  - non-current header is updated with largest sequence number
341  */
342 static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s,
343                               bool generate_data_write_guid, MSGUID *log_guid)
344 {
345     int ret = 0;
346     int hdr_idx = 0;
347     uint64_t header_offset = VHDX_HEADER1_OFFSET;
348 
349     VHDXHeader *active_header;
350     VHDXHeader *inactive_header;
351 
352     /* operate on the non-current header */
353     if (s->curr_header == 0) {
354         hdr_idx = 1;
355         header_offset = VHDX_HEADER2_OFFSET;
356     }
357 
358     active_header   = s->headers[s->curr_header];
359     inactive_header = s->headers[hdr_idx];
360 
361     inactive_header->sequence_number = active_header->sequence_number + 1;
362 
363     /* a new file guid must be generated before any file write, including
364      * headers */
365     inactive_header->file_write_guid = s->session_guid;
366 
367     /* a new data guid only needs to be generated before any guest-visible
368      * writes (i.e. something observable via virtual disk read) */
369     if (generate_data_write_guid) {
370         vhdx_guid_generate(&inactive_header->data_write_guid);
371     }
372 
373     /* update the log guid if present */
374     if (log_guid) {
375         inactive_header->log_guid = *log_guid;
376     }
377 
378     ret = vhdx_write_header(bs->file, inactive_header, header_offset, true);
379     if (ret < 0) {
380         goto exit;
381     }
382     s->curr_header = hdr_idx;
383 
384 exit:
385     return ret;
386 }
387 
388 /*
389  * The VHDX spec calls for header updates to be performed twice, so that both
390  * the current and non-current header have valid info
391  */
392 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s,
393                         bool generate_data_write_guid, MSGUID *log_guid)
394 {
395     int ret;
396 
397     ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
398     if (ret < 0) {
399         return ret;
400     }
401     ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
402     return ret;
403 }
404 
405 /* opens the specified header block from the VHDX file header section */
406 static void vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s,
407                               Error **errp)
408 {
409     int ret;
410     VHDXHeader *header1;
411     VHDXHeader *header2;
412     bool h1_valid = false;
413     bool h2_valid = false;
414     uint64_t h1_seq = 0;
415     uint64_t h2_seq = 0;
416     uint8_t *buffer;
417 
418     /* header1 & header2 are freed in vhdx_close() */
419     header1 = qemu_blockalign(bs, sizeof(VHDXHeader));
420     header2 = qemu_blockalign(bs, sizeof(VHDXHeader));
421 
422     buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE);
423 
424     s->headers[0] = header1;
425     s->headers[1] = header2;
426 
427     /* We have to read the whole VHDX_HEADER_SIZE instead of
428      * sizeof(VHDXHeader), because the checksum is over the whole
429      * region */
430     ret = bdrv_pread(bs->file, VHDX_HEADER1_OFFSET, buffer, VHDX_HEADER_SIZE);
431     if (ret < 0) {
432         goto fail;
433     }
434     /* copy over just the relevant portion that we need */
435     memcpy(header1, buffer, sizeof(VHDXHeader));
436 
437     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
438         vhdx_header_le_import(header1);
439         if (header1->signature == VHDX_HEADER_SIGNATURE &&
440             header1->version == 1) {
441             h1_seq = header1->sequence_number;
442             h1_valid = true;
443         }
444     }
445 
446     ret = bdrv_pread(bs->file, VHDX_HEADER2_OFFSET, buffer, VHDX_HEADER_SIZE);
447     if (ret < 0) {
448         goto fail;
449     }
450     /* copy over just the relevant portion that we need */
451     memcpy(header2, buffer, sizeof(VHDXHeader));
452 
453     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
454         vhdx_header_le_import(header2);
455         if (header2->signature == VHDX_HEADER_SIGNATURE &&
456             header2->version == 1) {
457             h2_seq = header2->sequence_number;
458             h2_valid = true;
459         }
460     }
461 
462     /* If there is only 1 valid header (or no valid headers), we
463      * don't care what the sequence numbers are */
464     if (h1_valid && !h2_valid) {
465         s->curr_header = 0;
466     } else if (!h1_valid && h2_valid) {
467         s->curr_header = 1;
468     } else if (!h1_valid && !h2_valid) {
469         goto fail;
470     } else {
471         /* If both headers are valid, then we choose the active one by the
472          * highest sequence number.  If the sequence numbers are equal, that is
473          * invalid */
474         if (h1_seq > h2_seq) {
475             s->curr_header = 0;
476         } else if (h2_seq > h1_seq) {
477             s->curr_header = 1;
478         } else {
479             /* The Microsoft Disk2VHD tool will create 2 identical
480              * headers, with identical sequence numbers.  If the headers are
481              * identical, don't consider the file corrupt */
482             if (!memcmp(header1, header2, sizeof(VHDXHeader))) {
483                 s->curr_header = 0;
484             } else {
485                 goto fail;
486             }
487         }
488     }
489 
490     vhdx_region_register(s, s->headers[s->curr_header]->log_offset,
491                             s->headers[s->curr_header]->log_length);
492     goto exit;
493 
494 fail:
495     error_setg_errno(errp, -ret, "No valid VHDX header found");
496     qemu_vfree(header1);
497     qemu_vfree(header2);
498     s->headers[0] = NULL;
499     s->headers[1] = NULL;
500 exit:
501     qemu_vfree(buffer);
502 }
503 
504 
505 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s)
506 {
507     int ret = 0;
508     uint8_t *buffer;
509     int offset = 0;
510     VHDXRegionTableEntry rt_entry;
511     uint32_t i;
512     bool bat_rt_found = false;
513     bool metadata_rt_found = false;
514 
515     /* We have to read the whole 64KB block, because the crc32 is over the
516      * whole block */
517     buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE);
518 
519     ret = bdrv_pread(bs->file, VHDX_REGION_TABLE_OFFSET, buffer,
520                      VHDX_HEADER_BLOCK_SIZE);
521     if (ret < 0) {
522         goto fail;
523     }
524     memcpy(&s->rt, buffer, sizeof(s->rt));
525     offset += sizeof(s->rt);
526 
527     if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4)) {
528         ret = -EINVAL;
529         goto fail;
530     }
531 
532     vhdx_region_header_le_import(&s->rt);
533 
534     if (s->rt.signature != VHDX_REGION_SIGNATURE) {
535         ret = -EINVAL;
536         goto fail;
537     }
538 
539 
540     /* Per spec, maximum region table entry count is 2047 */
541     if (s->rt.entry_count > 2047) {
542         ret = -EINVAL;
543         goto fail;
544     }
545 
546     for (i = 0; i < s->rt.entry_count; i++) {
547         memcpy(&rt_entry, buffer + offset, sizeof(rt_entry));
548         offset += sizeof(rt_entry);
549 
550         vhdx_region_entry_le_import(&rt_entry);
551 
552         /* check for region overlap between these entries, and any
553          * other memory regions in the file */
554         ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length);
555         if (ret < 0) {
556             goto fail;
557         }
558 
559         vhdx_region_register(s, rt_entry.file_offset, rt_entry.length);
560 
561         /* see if we recognize the entry */
562         if (guid_eq(rt_entry.guid, bat_guid)) {
563             /* must be unique; if we have already found it this is invalid */
564             if (bat_rt_found) {
565                 ret = -EINVAL;
566                 goto fail;
567             }
568             bat_rt_found = true;
569             s->bat_rt = rt_entry;
570             continue;
571         }
572 
573         if (guid_eq(rt_entry.guid, metadata_guid)) {
574             /* must be unique; if we have already found it this is invalid */
575             if (metadata_rt_found) {
576                 ret = -EINVAL;
577                 goto fail;
578             }
579             metadata_rt_found = true;
580             s->metadata_rt = rt_entry;
581             continue;
582         }
583 
584         if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) {
585             /* cannot read vhdx file - required region table entry that
586              * we do not understand.  per spec, we must fail to open */
587             ret = -ENOTSUP;
588             goto fail;
589         }
590     }
591 
592     if (!bat_rt_found || !metadata_rt_found) {
593         ret = -EINVAL;
594         goto fail;
595     }
596 
597     ret = 0;
598 
599 fail:
600     qemu_vfree(buffer);
601     return ret;
602 }
603 
604 
605 
606 /* Metadata initial parser
607  *
608  * This loads all the metadata entry fields.  This may cause additional
609  * fields to be processed (e.g. parent locator, etc..).
610  *
611  * There are 5 Metadata items that are always required:
612  *      - File Parameters (block size, has a parent)
613  *      - Virtual Disk Size (size, in bytes, of the virtual drive)
614  *      - Page 83 Data (scsi page 83 guid)
615  *      - Logical Sector Size (logical sector size in bytes, either 512 or
616  *                             4096.  We only support 512 currently)
617  *      - Physical Sector Size (512 or 4096)
618  *
619  * Also, if the File Parameters indicate this is a differencing file,
620  * we must also look for the Parent Locator metadata item.
621  */
622 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s)
623 {
624     int ret = 0;
625     uint8_t *buffer;
626     int offset = 0;
627     uint32_t i = 0;
628     VHDXMetadataTableEntry md_entry;
629 
630     buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE);
631 
632     ret = bdrv_pread(bs->file, s->metadata_rt.file_offset, buffer,
633                      VHDX_METADATA_TABLE_MAX_SIZE);
634     if (ret < 0) {
635         goto exit;
636     }
637     memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr));
638     offset += sizeof(s->metadata_hdr);
639 
640     vhdx_metadata_header_le_import(&s->metadata_hdr);
641 
642     if (s->metadata_hdr.signature != VHDX_METADATA_SIGNATURE) {
643         ret = -EINVAL;
644         goto exit;
645     }
646 
647     s->metadata_entries.present = 0;
648 
649     if ((s->metadata_hdr.entry_count * sizeof(md_entry)) >
650         (VHDX_METADATA_TABLE_MAX_SIZE - offset)) {
651         ret = -EINVAL;
652         goto exit;
653     }
654 
655     for (i = 0; i < s->metadata_hdr.entry_count; i++) {
656         memcpy(&md_entry, buffer + offset, sizeof(md_entry));
657         offset += sizeof(md_entry);
658 
659         vhdx_metadata_entry_le_import(&md_entry);
660 
661         if (guid_eq(md_entry.item_id, file_param_guid)) {
662             if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) {
663                 ret = -EINVAL;
664                 goto exit;
665             }
666             s->metadata_entries.file_parameters_entry = md_entry;
667             s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT;
668             continue;
669         }
670 
671         if (guid_eq(md_entry.item_id, virtual_size_guid)) {
672             if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) {
673                 ret = -EINVAL;
674                 goto exit;
675             }
676             s->metadata_entries.virtual_disk_size_entry = md_entry;
677             s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT;
678             continue;
679         }
680 
681         if (guid_eq(md_entry.item_id, page83_guid)) {
682             if (s->metadata_entries.present & META_PAGE_83_PRESENT) {
683                 ret = -EINVAL;
684                 goto exit;
685             }
686             s->metadata_entries.page83_data_entry = md_entry;
687             s->metadata_entries.present |= META_PAGE_83_PRESENT;
688             continue;
689         }
690 
691         if (guid_eq(md_entry.item_id, logical_sector_guid)) {
692             if (s->metadata_entries.present &
693                 META_LOGICAL_SECTOR_SIZE_PRESENT) {
694                 ret = -EINVAL;
695                 goto exit;
696             }
697             s->metadata_entries.logical_sector_size_entry = md_entry;
698             s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT;
699             continue;
700         }
701 
702         if (guid_eq(md_entry.item_id, phys_sector_guid)) {
703             if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) {
704                 ret = -EINVAL;
705                 goto exit;
706             }
707             s->metadata_entries.phys_sector_size_entry = md_entry;
708             s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT;
709             continue;
710         }
711 
712         if (guid_eq(md_entry.item_id, parent_locator_guid)) {
713             if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
714                 ret = -EINVAL;
715                 goto exit;
716             }
717             s->metadata_entries.parent_locator_entry = md_entry;
718             s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT;
719             continue;
720         }
721 
722         if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) {
723             /* cannot read vhdx file - required region table entry that
724              * we do not understand.  per spec, we must fail to open */
725             ret = -ENOTSUP;
726             goto exit;
727         }
728     }
729 
730     if (s->metadata_entries.present != META_ALL_PRESENT) {
731         ret = -ENOTSUP;
732         goto exit;
733     }
734 
735     ret = bdrv_pread(bs->file,
736                      s->metadata_entries.file_parameters_entry.offset
737                                          + s->metadata_rt.file_offset,
738                      &s->params,
739                      sizeof(s->params));
740 
741     if (ret < 0) {
742         goto exit;
743     }
744 
745     le32_to_cpus(&s->params.block_size);
746     le32_to_cpus(&s->params.data_bits);
747 
748 
749     /* We now have the file parameters, so we can tell if this is a
750      * differencing file (i.e.. has_parent), is dynamic or fixed
751      * sized (leave_blocks_allocated), and the block size */
752 
753     /* The parent locator required iff the file parameters has_parent set */
754     if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
755         if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
756             /* TODO: parse  parent locator fields */
757             ret = -ENOTSUP; /* temp, until differencing files are supported */
758             goto exit;
759         } else {
760             /* if has_parent is set, but there is not parent locator present,
761              * then that is an invalid combination */
762             ret = -EINVAL;
763             goto exit;
764         }
765     }
766 
767     /* determine virtual disk size, logical sector size,
768      * and phys sector size */
769 
770     ret = bdrv_pread(bs->file,
771                      s->metadata_entries.virtual_disk_size_entry.offset
772                                            + s->metadata_rt.file_offset,
773                      &s->virtual_disk_size,
774                      sizeof(uint64_t));
775     if (ret < 0) {
776         goto exit;
777     }
778     ret = bdrv_pread(bs->file,
779                      s->metadata_entries.logical_sector_size_entry.offset
780                                              + s->metadata_rt.file_offset,
781                      &s->logical_sector_size,
782                      sizeof(uint32_t));
783     if (ret < 0) {
784         goto exit;
785     }
786     ret = bdrv_pread(bs->file,
787                      s->metadata_entries.phys_sector_size_entry.offset
788                                           + s->metadata_rt.file_offset,
789                      &s->physical_sector_size,
790                      sizeof(uint32_t));
791     if (ret < 0) {
792         goto exit;
793     }
794 
795     le64_to_cpus(&s->virtual_disk_size);
796     le32_to_cpus(&s->logical_sector_size);
797     le32_to_cpus(&s->physical_sector_size);
798 
799     if (s->params.block_size < VHDX_BLOCK_SIZE_MIN ||
800         s->params.block_size > VHDX_BLOCK_SIZE_MAX) {
801         ret = -EINVAL;
802         goto exit;
803     }
804 
805     /* only 2 supported sector sizes */
806     if (s->logical_sector_size != 512 && s->logical_sector_size != 4096) {
807         ret = -EINVAL;
808         goto exit;
809     }
810 
811     /* Both block_size and sector_size are guaranteed powers of 2, below.
812        Due to range checks above, s->sectors_per_block can never be < 256 */
813     s->sectors_per_block = s->params.block_size / s->logical_sector_size;
814     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
815                      (uint64_t)s->logical_sector_size /
816                      (uint64_t)s->params.block_size;
817 
818     /* These values are ones we will want to use for division / multiplication
819      * later on, and they are all guaranteed (per the spec) to be powers of 2,
820      * so we can take advantage of that for shift operations during
821      * reads/writes */
822     if (s->logical_sector_size & (s->logical_sector_size - 1)) {
823         ret = -EINVAL;
824         goto exit;
825     }
826     if (s->sectors_per_block & (s->sectors_per_block - 1)) {
827         ret = -EINVAL;
828         goto exit;
829     }
830     if (s->chunk_ratio & (s->chunk_ratio - 1)) {
831         ret = -EINVAL;
832         goto exit;
833     }
834     s->block_size = s->params.block_size;
835     if (s->block_size & (s->block_size - 1)) {
836         ret = -EINVAL;
837         goto exit;
838     }
839 
840     vhdx_set_shift_bits(s);
841 
842     ret = 0;
843 
844 exit:
845     qemu_vfree(buffer);
846     return ret;
847 }
848 
849 /*
850  * Calculate the number of BAT entries, including sector
851  * bitmap entries.
852  */
853 static void vhdx_calc_bat_entries(BDRVVHDXState *s)
854 {
855     uint32_t data_blocks_cnt, bitmap_blocks_cnt;
856 
857     data_blocks_cnt = s->virtual_disk_size >> s->block_size_bits;
858     if (s->virtual_disk_size - (data_blocks_cnt << s->block_size_bits)) {
859         data_blocks_cnt++;
860     }
861     bitmap_blocks_cnt = data_blocks_cnt >> s->chunk_ratio_bits;
862     if (data_blocks_cnt - (bitmap_blocks_cnt << s->chunk_ratio_bits)) {
863         bitmap_blocks_cnt++;
864     }
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, 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_set(&s->migration_blocker,
1006             QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1007             "vhdx", bdrv_get_device_name(bs), "live migration");
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:   /* fall through */
1113             case PAYLOAD_BLOCK_UNMAPPED:    /* fall through */
1114             case PAYLOAD_BLOCK_ZERO:
1115                 /* return zero */
1116                 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail);
1117                 break;
1118             case PAYLOAD_BLOCK_FULLY_PRESENT:
1119                 qemu_co_mutex_unlock(&s->lock);
1120                 ret = bdrv_co_readv(bs->file,
1121                                     sinfo.file_offset >> BDRV_SECTOR_BITS,
1122                                     sinfo.sectors_avail, &hd_qiov);
1123                 qemu_co_mutex_lock(&s->lock);
1124                 if (ret < 0) {
1125                     goto exit;
1126                 }
1127                 break;
1128             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1129                 /* we don't yet support difference files, fall through
1130                  * to error */
1131             default:
1132                 ret = -EIO;
1133                 goto exit;
1134                 break;
1135             }
1136             nb_sectors -= sinfo.sectors_avail;
1137             sector_num += sinfo.sectors_avail;
1138             bytes_done += sinfo.bytes_avail;
1139         }
1140     }
1141     ret = 0;
1142 exit:
1143     qemu_co_mutex_unlock(&s->lock);
1144     qemu_iovec_destroy(&hd_qiov);
1145     return ret;
1146 }
1147 
1148 /*
1149  * Allocate a new payload block at the end of the file.
1150  *
1151  * Allocation will happen at 1MB alignment inside the file
1152  *
1153  * Returns the file offset start of the new payload block
1154  */
1155 static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s,
1156                                     uint64_t *new_offset)
1157 {
1158     *new_offset = bdrv_getlength(bs->file);
1159 
1160     /* per the spec, the address for a block is in units of 1MB */
1161     *new_offset = ROUND_UP(*new_offset, 1024 * 1024);
1162 
1163     return bdrv_truncate(bs->file, *new_offset + s->block_size);
1164 }
1165 
1166 /*
1167  * Update the BAT table entry with the new file offset, and the new entry
1168  * state */
1169 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s,
1170                                        VHDXSectorInfo *sinfo,
1171                                        uint64_t *bat_entry_le,
1172                                        uint64_t *bat_offset, int state)
1173 {
1174     /* The BAT entry is a uint64, with 44 bits for the file offset in units of
1175      * 1MB, and 3 bits for the block state. */
1176     s->bat[sinfo->bat_idx]  = sinfo->file_offset;
1177 
1178     s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK;
1179 
1180     *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]);
1181     *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry);
1182 
1183 }
1184 
1185 /* Per the spec, on the first write of guest-visible data to the file the
1186  * data write guid must be updated in the header */
1187 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s)
1188 {
1189     int ret = 0;
1190     if (s->first_visible_write) {
1191         s->first_visible_write = false;
1192         ret = vhdx_update_headers(bs, s, true, NULL);
1193     }
1194     return ret;
1195 }
1196 
1197 static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num,
1198                                       int nb_sectors, QEMUIOVector *qiov)
1199 {
1200     int ret = -ENOTSUP;
1201     BDRVVHDXState *s = bs->opaque;
1202     VHDXSectorInfo sinfo;
1203     uint64_t bytes_done = 0;
1204     uint64_t bat_entry = 0;
1205     uint64_t bat_entry_offset = 0;
1206     QEMUIOVector hd_qiov;
1207     struct iovec iov1 = { 0 };
1208     struct iovec iov2 = { 0 };
1209     int sectors_to_write;
1210     int bat_state;
1211     uint64_t bat_prior_offset = 0;
1212     bool bat_update = false;
1213 
1214     qemu_iovec_init(&hd_qiov, qiov->niov);
1215 
1216     qemu_co_mutex_lock(&s->lock);
1217 
1218     ret = vhdx_user_visible_write(bs, s);
1219     if (ret < 0) {
1220         goto exit;
1221     }
1222 
1223     while (nb_sectors > 0) {
1224         bool use_zero_buffers = false;
1225         bat_update = false;
1226         if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1227             /* not supported yet */
1228             ret = -ENOTSUP;
1229             goto exit;
1230         } else {
1231             vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1232             sectors_to_write = sinfo.sectors_avail;
1233 
1234             qemu_iovec_reset(&hd_qiov);
1235             /* check the payload block state */
1236             bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK;
1237             switch (bat_state) {
1238             case PAYLOAD_BLOCK_ZERO:
1239                 /* in this case, we need to preserve zero writes for
1240                  * data that is not part of this write, so we must pad
1241                  * the rest of the buffer to zeroes */
1242 
1243                 /* if we are on a posix system with ftruncate() that extends
1244                  * a file, then it is zero-filled for us.  On Win32, the raw
1245                  * layer uses SetFilePointer and SetFileEnd, which does not
1246                  * zero fill AFAIK */
1247 
1248                 /* Queue another write of zero buffers if the underlying file
1249                  * does not zero-fill on file extension */
1250 
1251                 if (bdrv_has_zero_init(bs->file) == 0) {
1252                     use_zero_buffers = true;
1253 
1254                     /* zero fill the front, if any */
1255                     if (sinfo.block_offset) {
1256                         iov1.iov_len = sinfo.block_offset;
1257                         iov1.iov_base = qemu_blockalign(bs, iov1.iov_len);
1258                         memset(iov1.iov_base, 0, iov1.iov_len);
1259                         qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0,
1260                                               sinfo.block_offset);
1261                         sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS;
1262                     }
1263 
1264                     /* our actual data */
1265                     qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1266                                       sinfo.bytes_avail);
1267 
1268                     /* zero fill the back, if any */
1269                     if ((sinfo.bytes_avail - sinfo.block_offset) <
1270                          s->block_size) {
1271                         iov2.iov_len = s->block_size -
1272                                       (sinfo.bytes_avail + sinfo.block_offset);
1273                         iov2.iov_base = qemu_blockalign(bs, iov2.iov_len);
1274                         memset(iov2.iov_base, 0, iov2.iov_len);
1275                         qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0,
1276                                               sinfo.block_offset);
1277                         sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS;
1278                     }
1279                 }
1280 
1281                 /* fall through */
1282             case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1283             case PAYLOAD_BLOCK_UNMAPPED:    /* fall through */
1284             case PAYLOAD_BLOCK_UNDEFINED:   /* fall through */
1285                 bat_prior_offset = sinfo.file_offset;
1286                 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset);
1287                 if (ret < 0) {
1288                     goto exit;
1289                 }
1290                 /* once we support differencing files, this may also be
1291                  * partially present */
1292                 /* update block state to the newly specified state */
1293                 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1294                                             &bat_entry_offset,
1295                                             PAYLOAD_BLOCK_FULLY_PRESENT);
1296                 bat_update = true;
1297                 /* since we just allocated a block, file_offset is the
1298                  * beginning of the payload block. It needs to be the
1299                  * write address, which includes the offset into the block */
1300                 if (!use_zero_buffers) {
1301                     sinfo.file_offset += sinfo.block_offset;
1302                 }
1303                 /* fall through */
1304             case PAYLOAD_BLOCK_FULLY_PRESENT:
1305                 /* if the file offset address is in the header zone,
1306                  * there is a problem */
1307                 if (sinfo.file_offset < (1024 * 1024)) {
1308                     ret = -EFAULT;
1309                     goto error_bat_restore;
1310                 }
1311 
1312                 if (!use_zero_buffers) {
1313                     qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1314                                       sinfo.bytes_avail);
1315                 }
1316                 /* block exists, so we can just overwrite it */
1317                 qemu_co_mutex_unlock(&s->lock);
1318                 ret = bdrv_co_writev(bs->file,
1319                                     sinfo.file_offset >> BDRV_SECTOR_BITS,
1320                                     sectors_to_write, &hd_qiov);
1321                 qemu_co_mutex_lock(&s->lock);
1322                 if (ret < 0) {
1323                     goto error_bat_restore;
1324                 }
1325                 break;
1326             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1327                 /* we don't yet support difference files, fall through
1328                  * to error */
1329             default:
1330                 ret = -EIO;
1331                 goto exit;
1332                 break;
1333             }
1334 
1335             if (bat_update) {
1336                 /* this will update the BAT entry into the log journal, and
1337                  * then flush the log journal out to disk */
1338                 ret =  vhdx_log_write_and_flush(bs, s, &bat_entry,
1339                                                 sizeof(VHDXBatEntry),
1340                                                 bat_entry_offset);
1341                 if (ret < 0) {
1342                     goto exit;
1343                 }
1344             }
1345 
1346             nb_sectors -= sinfo.sectors_avail;
1347             sector_num += sinfo.sectors_avail;
1348             bytes_done += sinfo.bytes_avail;
1349 
1350         }
1351     }
1352 
1353     goto exit;
1354 
1355 error_bat_restore:
1356     if (bat_update) {
1357         /* keep metadata in sync, and restore the bat entry state
1358          * if error. */
1359         sinfo.file_offset = bat_prior_offset;
1360         vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1361                                     &bat_entry_offset, bat_state);
1362     }
1363 exit:
1364     qemu_vfree(iov1.iov_base);
1365     qemu_vfree(iov2.iov_base);
1366     qemu_co_mutex_unlock(&s->lock);
1367     qemu_iovec_destroy(&hd_qiov);
1368     return ret;
1369 }
1370 
1371 
1372 
1373 /*
1374  * Create VHDX Headers
1375  *
1376  * There are 2 headers, and the highest sequence number will represent
1377  * the active header
1378  */
1379 static int vhdx_create_new_headers(BlockDriverState *bs, uint64_t image_size,
1380                                    uint32_t log_size)
1381 {
1382     int ret = 0;
1383     VHDXHeader *hdr = NULL;
1384 
1385     hdr = g_new0(VHDXHeader, 1);
1386 
1387     hdr->signature       = VHDX_HEADER_SIGNATURE;
1388     hdr->sequence_number = g_random_int();
1389     hdr->log_version     = 0;
1390     hdr->version         = 1;
1391     hdr->log_length      = log_size;
1392     hdr->log_offset      = VHDX_HEADER_SECTION_END;
1393     vhdx_guid_generate(&hdr->file_write_guid);
1394     vhdx_guid_generate(&hdr->data_write_guid);
1395 
1396     ret = vhdx_write_header(bs, hdr, VHDX_HEADER1_OFFSET, false);
1397     if (ret < 0) {
1398         goto exit;
1399     }
1400     hdr->sequence_number++;
1401     ret = vhdx_write_header(bs, hdr, VHDX_HEADER2_OFFSET, false);
1402     if (ret < 0) {
1403         goto exit;
1404     }
1405 
1406 exit:
1407     g_free(hdr);
1408     return ret;
1409 }
1410 
1411 #define VHDX_METADATA_ENTRY_BUFFER_SIZE \
1412                                     (sizeof(VHDXFileParameters)               +\
1413                                      sizeof(VHDXVirtualDiskSize)              +\
1414                                      sizeof(VHDXPage83Data)                   +\
1415                                      sizeof(VHDXVirtualDiskLogicalSectorSize) +\
1416                                      sizeof(VHDXVirtualDiskPhysicalSectorSize))
1417 
1418 /*
1419  * Create the Metadata entries.
1420  *
1421  * For more details on the entries, see section 3.5 (pg 29) in the
1422  * VHDX 1.00 specification.
1423  *
1424  * We support 5 metadata entries (all required by spec):
1425  *          File Parameters,
1426  *          Virtual Disk Size,
1427  *          Page 83 Data,
1428  *          Logical Sector Size,
1429  *          Physical Sector Size
1430  *
1431  * The first 64KB of the Metadata section is reserved for the metadata
1432  * header and entries; beyond that, the metadata items themselves reside.
1433  */
1434 static int vhdx_create_new_metadata(BlockDriverState *bs,
1435                                     uint64_t image_size,
1436                                     uint32_t block_size,
1437                                     uint32_t sector_size,
1438                                     uint64_t metadata_offset,
1439                                     VHDXImageType type)
1440 {
1441     int ret = 0;
1442     uint32_t offset = 0;
1443     void *buffer = NULL;
1444     void *entry_buffer;
1445     VHDXMetadataTableHeader *md_table;;
1446     VHDXMetadataTableEntry  *md_table_entry;
1447 
1448     /* Metadata entries */
1449     VHDXFileParameters     *mt_file_params;
1450     VHDXVirtualDiskSize    *mt_virtual_size;
1451     VHDXPage83Data         *mt_page83;
1452     VHDXVirtualDiskLogicalSectorSize  *mt_log_sector_size;
1453     VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size;
1454 
1455     entry_buffer = g_malloc0(VHDX_METADATA_ENTRY_BUFFER_SIZE);
1456 
1457     mt_file_params = entry_buffer;
1458     offset += sizeof(VHDXFileParameters);
1459     mt_virtual_size = entry_buffer + offset;
1460     offset += sizeof(VHDXVirtualDiskSize);
1461     mt_page83 = entry_buffer + offset;
1462     offset += sizeof(VHDXPage83Data);
1463     mt_log_sector_size = entry_buffer + offset;
1464     offset += sizeof(VHDXVirtualDiskLogicalSectorSize);
1465     mt_phys_sector_size = entry_buffer + offset;
1466 
1467     mt_file_params->block_size = cpu_to_le32(block_size);
1468     if (type == VHDX_TYPE_FIXED) {
1469         mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED;
1470         cpu_to_le32s(&mt_file_params->data_bits);
1471     }
1472 
1473     vhdx_guid_generate(&mt_page83->page_83_data);
1474     cpu_to_leguids(&mt_page83->page_83_data);
1475     mt_virtual_size->virtual_disk_size        = cpu_to_le64(image_size);
1476     mt_log_sector_size->logical_sector_size   = cpu_to_le32(sector_size);
1477     mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size);
1478 
1479     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1480     md_table = buffer;
1481 
1482     md_table->signature   = VHDX_METADATA_SIGNATURE;
1483     md_table->entry_count = 5;
1484     vhdx_metadata_header_le_export(md_table);
1485 
1486 
1487     /* This will reference beyond the reserved table portion */
1488     offset = 64 * KiB;
1489 
1490     md_table_entry = buffer + sizeof(VHDXMetadataTableHeader);
1491 
1492     md_table_entry[0].item_id = file_param_guid;
1493     md_table_entry[0].offset  = offset;
1494     md_table_entry[0].length  = sizeof(VHDXFileParameters);
1495     md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED;
1496     offset += md_table_entry[0].length;
1497     vhdx_metadata_entry_le_export(&md_table_entry[0]);
1498 
1499     md_table_entry[1].item_id = virtual_size_guid;
1500     md_table_entry[1].offset  = offset;
1501     md_table_entry[1].length  = sizeof(VHDXVirtualDiskSize);
1502     md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1503                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1504     offset += md_table_entry[1].length;
1505     vhdx_metadata_entry_le_export(&md_table_entry[1]);
1506 
1507     md_table_entry[2].item_id = page83_guid;
1508     md_table_entry[2].offset  = offset;
1509     md_table_entry[2].length  = sizeof(VHDXPage83Data);
1510     md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1511                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1512     offset += md_table_entry[2].length;
1513     vhdx_metadata_entry_le_export(&md_table_entry[2]);
1514 
1515     md_table_entry[3].item_id = logical_sector_guid;
1516     md_table_entry[3].offset  = offset;
1517     md_table_entry[3].length  = sizeof(VHDXVirtualDiskLogicalSectorSize);
1518     md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1519                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1520     offset += md_table_entry[3].length;
1521     vhdx_metadata_entry_le_export(&md_table_entry[3]);
1522 
1523     md_table_entry[4].item_id = phys_sector_guid;
1524     md_table_entry[4].offset  = offset;
1525     md_table_entry[4].length  = sizeof(VHDXVirtualDiskPhysicalSectorSize);
1526     md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1527                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1528     vhdx_metadata_entry_le_export(&md_table_entry[4]);
1529 
1530     ret = bdrv_pwrite(bs, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE);
1531     if (ret < 0) {
1532         goto exit;
1533     }
1534 
1535     ret = bdrv_pwrite(bs, metadata_offset + (64 * KiB), entry_buffer,
1536                       VHDX_METADATA_ENTRY_BUFFER_SIZE);
1537     if (ret < 0) {
1538         goto exit;
1539     }
1540 
1541 
1542 exit:
1543     g_free(buffer);
1544     g_free(entry_buffer);
1545     return ret;
1546 }
1547 
1548 /* This create the actual BAT itself.  We currently only support
1549  * 'Dynamic' and 'Fixed' image types.
1550  *
1551  *  Dynamic images: default state of the BAT is all zeroes.
1552  *
1553  *  Fixed images: default state of the BAT is fully populated, with
1554  *                file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT.
1555  */
1556 static int vhdx_create_bat(BlockDriverState *bs, BDRVVHDXState *s,
1557                            uint64_t image_size, VHDXImageType type,
1558                            bool use_zero_blocks, uint64_t file_offset,
1559                            uint32_t length)
1560 {
1561     int ret = 0;
1562     uint64_t data_file_offset;
1563     uint64_t total_sectors = 0;
1564     uint64_t sector_num = 0;
1565     uint64_t unused;
1566     int block_state;
1567     VHDXSectorInfo sinfo;
1568 
1569     assert(s->bat == NULL);
1570 
1571     /* this gives a data start after BAT/bitmap entries, and well
1572      * past any metadata entries (with a 4 MB buffer for future
1573      * expansion */
1574     data_file_offset = file_offset + length + 5 * MiB;
1575     total_sectors = image_size >> s->logical_sector_size_bits;
1576 
1577     if (type == VHDX_TYPE_DYNAMIC) {
1578         /* All zeroes, so we can just extend the file - the end of the BAT
1579          * is the furthest thing we have written yet */
1580         ret = bdrv_truncate(bs, data_file_offset);
1581         if (ret < 0) {
1582             goto exit;
1583         }
1584     } else if (type == VHDX_TYPE_FIXED) {
1585         ret = bdrv_truncate(bs, data_file_offset + image_size);
1586         if (ret < 0) {
1587             goto exit;
1588         }
1589     } else {
1590         ret = -ENOTSUP;
1591         goto exit;
1592     }
1593 
1594     if (type == VHDX_TYPE_FIXED ||
1595                 use_zero_blocks ||
1596                 bdrv_has_zero_init(bs) == 0) {
1597         /* for a fixed file, the default BAT entry is not zero */
1598         s->bat = g_try_malloc0(length);
1599         if (length && s->bat == NULL) {
1600             ret = -ENOMEM;
1601             goto exit;
1602         }
1603         block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT :
1604                                                 PAYLOAD_BLOCK_NOT_PRESENT;
1605         block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state;
1606         /* fill the BAT by emulating sector writes of sectors_per_block size */
1607         while (sector_num < total_sectors) {
1608             vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo);
1609             sinfo.file_offset = data_file_offset +
1610                                 (sector_num << s->logical_sector_size_bits);
1611             sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB);
1612             vhdx_update_bat_table_entry(bs, s, &sinfo, &unused, &unused,
1613                                         block_state);
1614             cpu_to_le64s(&s->bat[sinfo.bat_idx]);
1615             sector_num += s->sectors_per_block;
1616         }
1617         ret = bdrv_pwrite(bs, file_offset, s->bat, length);
1618         if (ret < 0) {
1619             goto exit;
1620         }
1621     }
1622 
1623 
1624 
1625 exit:
1626     g_free(s->bat);
1627     return ret;
1628 }
1629 
1630 /* Creates the region table header, and region table entries.
1631  * There are 2 supported region table entries: BAT, and Metadata/
1632  *
1633  * As the calculations for the BAT region table are also needed
1634  * to create the BAT itself, we will also cause the BAT to be
1635  * created.
1636  */
1637 static int vhdx_create_new_region_table(BlockDriverState *bs,
1638                                         uint64_t image_size,
1639                                         uint32_t block_size,
1640                                         uint32_t sector_size,
1641                                         uint32_t log_size,
1642                                         bool use_zero_blocks,
1643                                         VHDXImageType type,
1644                                         uint64_t *metadata_offset)
1645 {
1646     int ret = 0;
1647     uint32_t offset = 0;
1648     void *buffer = NULL;
1649     uint64_t bat_file_offset;
1650     uint32_t bat_length;
1651     BDRVVHDXState *s = NULL;
1652     VHDXRegionTableHeader *region_table;
1653     VHDXRegionTableEntry *rt_bat;
1654     VHDXRegionTableEntry *rt_metadata;
1655 
1656     assert(metadata_offset != NULL);
1657 
1658     /* Populate enough of the BDRVVHDXState to be able to use the
1659      * pre-existing BAT calculation, translation, and update functions */
1660     s = g_new0(BDRVVHDXState, 1);
1661 
1662     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
1663                      (uint64_t) sector_size / (uint64_t) block_size;
1664 
1665     s->sectors_per_block = block_size / sector_size;
1666     s->virtual_disk_size = image_size;
1667     s->block_size = block_size;
1668     s->logical_sector_size = sector_size;
1669 
1670     vhdx_set_shift_bits(s);
1671 
1672     vhdx_calc_bat_entries(s);
1673 
1674     /* At this point the VHDX state is populated enough for creation */
1675 
1676     /* a single buffer is used so we can calculate the checksum over the
1677      * entire 64KB block */
1678     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1679     region_table = buffer;
1680     offset += sizeof(VHDXRegionTableHeader);
1681     rt_bat = buffer + offset;
1682     offset += sizeof(VHDXRegionTableEntry);
1683     rt_metadata  = buffer + offset;
1684 
1685     region_table->signature = VHDX_REGION_SIGNATURE;
1686     region_table->entry_count = 2;   /* BAT and Metadata */
1687 
1688     rt_bat->guid        = bat_guid;
1689     rt_bat->length      = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB);
1690     rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB);
1691     s->bat_offset = rt_bat->file_offset;
1692 
1693     rt_metadata->guid        = metadata_guid;
1694     rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length,
1695                                         MiB);
1696     rt_metadata->length      = 1 * MiB; /* min size, and more than enough */
1697     *metadata_offset = rt_metadata->file_offset;
1698 
1699     bat_file_offset = rt_bat->file_offset;
1700     bat_length = rt_bat->length;
1701 
1702     vhdx_region_header_le_export(region_table);
1703     vhdx_region_entry_le_export(rt_bat);
1704     vhdx_region_entry_le_export(rt_metadata);
1705 
1706     vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE,
1707                          offsetof(VHDXRegionTableHeader, checksum));
1708 
1709 
1710     /* The region table gives us the data we need to create the BAT,
1711      * so do that now */
1712     ret = vhdx_create_bat(bs, s, image_size, type, use_zero_blocks,
1713                           bat_file_offset, bat_length);
1714     if (ret < 0) {
1715         goto exit;
1716     }
1717 
1718     /* Now write out the region headers to disk */
1719     ret = bdrv_pwrite(bs, VHDX_REGION_TABLE_OFFSET, buffer,
1720                       VHDX_HEADER_BLOCK_SIZE);
1721     if (ret < 0) {
1722         goto exit;
1723     }
1724 
1725     ret = bdrv_pwrite(bs, VHDX_REGION_TABLE2_OFFSET, buffer,
1726                       VHDX_HEADER_BLOCK_SIZE);
1727     if (ret < 0) {
1728         goto exit;
1729     }
1730 
1731 exit:
1732     g_free(s);
1733     g_free(buffer);
1734     return ret;
1735 }
1736 
1737 /* We need to create the following elements:
1738  *
1739  *    .-----------------------------------------------------------------.
1740  *    |   (A)    |   (B)    |    (C)    |     (D)       |     (E)       |
1741  *    |  File ID |  Header1 |  Header 2 |  Region Tbl 1 |  Region Tbl 2 |
1742  *    |          |          |           |               |               |
1743  *    .-----------------------------------------------------------------.
1744  *    0         64KB      128KB       192KB           256KB           320KB
1745  *
1746  *
1747  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1748  *    |     (F)     |     (G)       |    (H)    |                        |
1749  *    | Journal Log |  BAT / Bitmap |  Metadata |  .... data ......      |
1750  *    |             |               |           |                        |
1751  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1752  *   1MB
1753  */
1754 static int vhdx_create(const char *filename, QemuOpts *opts, Error **errp)
1755 {
1756     int ret = 0;
1757     uint64_t image_size = (uint64_t) 2 * GiB;
1758     uint32_t log_size   = 1 * MiB;
1759     uint32_t block_size = 0;
1760     uint64_t signature;
1761     uint64_t metadata_offset;
1762     bool use_zero_blocks = false;
1763 
1764     gunichar2 *creator = NULL;
1765     glong creator_items;
1766     BlockDriverState *bs;
1767     char *type = NULL;
1768     VHDXImageType image_type;
1769     Error *local_err = NULL;
1770 
1771     image_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1772                           BDRV_SECTOR_SIZE);
1773     log_size = qemu_opt_get_size_del(opts, VHDX_BLOCK_OPT_LOG_SIZE, 0);
1774     block_size = qemu_opt_get_size_del(opts, VHDX_BLOCK_OPT_BLOCK_SIZE, 0);
1775     type = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1776     use_zero_blocks = qemu_opt_get_bool_del(opts, VHDX_BLOCK_OPT_ZERO, false);
1777 
1778     if (image_size > VHDX_MAX_IMAGE_SIZE) {
1779         error_setg_errno(errp, EINVAL, "Image size too large; max of 64TB");
1780         ret = -EINVAL;
1781         goto exit;
1782     }
1783 
1784     if (type == NULL) {
1785         type = g_strdup("dynamic");
1786     }
1787 
1788     if (!strcmp(type, "dynamic")) {
1789         image_type = VHDX_TYPE_DYNAMIC;
1790     } else if (!strcmp(type, "fixed")) {
1791         image_type = VHDX_TYPE_FIXED;
1792     } else if (!strcmp(type, "differencing")) {
1793         error_setg_errno(errp, ENOTSUP,
1794                          "Differencing files not yet supported");
1795         ret = -ENOTSUP;
1796         goto exit;
1797     } else {
1798         ret = -EINVAL;
1799         goto exit;
1800     }
1801 
1802     /* These are pretty arbitrary, and mainly designed to keep the BAT
1803      * size reasonable to load into RAM */
1804     if (block_size == 0) {
1805         if (image_size > 32 * TiB) {
1806             block_size = 64 * MiB;
1807         } else if (image_size > (uint64_t) 100 * GiB) {
1808             block_size = 32 * MiB;
1809         } else if (image_size > 1 * GiB) {
1810             block_size = 16 * MiB;
1811         } else {
1812             block_size = 8 * MiB;
1813         }
1814     }
1815 
1816 
1817     /* make the log size close to what was specified, but must be
1818      * min 1MB, and multiple of 1MB */
1819     log_size = ROUND_UP(log_size, MiB);
1820 
1821     block_size = ROUND_UP(block_size, MiB);
1822     block_size = block_size > VHDX_BLOCK_SIZE_MAX ? VHDX_BLOCK_SIZE_MAX :
1823                                                     block_size;
1824 
1825     ret = bdrv_create_file(filename, opts, &local_err);
1826     if (ret < 0) {
1827         error_propagate(errp, local_err);
1828         goto exit;
1829     }
1830 
1831     bs = NULL;
1832     ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1833                     NULL, &local_err);
1834     if (ret < 0) {
1835         error_propagate(errp, local_err);
1836         goto exit;
1837     }
1838 
1839     /* Create (A) */
1840 
1841     /* The creator field is optional, but may be useful for
1842      * debugging / diagnostics */
1843     creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL,
1844                               &creator_items, NULL);
1845     signature = cpu_to_le64(VHDX_FILE_SIGNATURE);
1846     ret = bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature));
1847     if (ret < 0) {
1848         goto delete_and_exit;
1849     }
1850     if (creator) {
1851         ret = bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET + sizeof(signature),
1852                           creator, creator_items * sizeof(gunichar2));
1853         if (ret < 0) {
1854             goto delete_and_exit;
1855         }
1856     }
1857 
1858 
1859     /* Creates (B),(C) */
1860     ret = vhdx_create_new_headers(bs, image_size, log_size);
1861     if (ret < 0) {
1862         goto delete_and_exit;
1863     }
1864 
1865     /* Creates (D),(E),(G) explicitly. (F) created as by-product */
1866     ret = vhdx_create_new_region_table(bs, image_size, block_size, 512,
1867                                        log_size, use_zero_blocks, image_type,
1868                                        &metadata_offset);
1869     if (ret < 0) {
1870         goto delete_and_exit;
1871     }
1872 
1873     /* Creates (H) */
1874     ret = vhdx_create_new_metadata(bs, image_size, block_size, 512,
1875                                    metadata_offset, image_type);
1876     if (ret < 0) {
1877         goto delete_and_exit;
1878     }
1879 
1880 
1881 delete_and_exit:
1882     bdrv_unref(bs);
1883 exit:
1884     g_free(type);
1885     g_free(creator);
1886     return ret;
1887 }
1888 
1889 /* If opened r/w, the VHDX driver will automatically replay the log,
1890  * if one is present, inside the vhdx_open() call.
1891  *
1892  * If qemu-img check -r all is called, the image is automatically opened
1893  * r/w and any log has already been replayed, so there is nothing (currently)
1894  * for us to do here
1895  */
1896 static int vhdx_check(BlockDriverState *bs, BdrvCheckResult *result,
1897                        BdrvCheckMode fix)
1898 {
1899     BDRVVHDXState *s = bs->opaque;
1900 
1901     if (s->log_replayed_on_open) {
1902         result->corruptions_fixed++;
1903     }
1904     return 0;
1905 }
1906 
1907 static QemuOptsList vhdx_create_opts = {
1908     .name = "vhdx-create-opts",
1909     .head = QTAILQ_HEAD_INITIALIZER(vhdx_create_opts.head),
1910     .desc = {
1911         {
1912            .name = BLOCK_OPT_SIZE,
1913            .type = QEMU_OPT_SIZE,
1914            .help = "Virtual disk size; max of 64TB."
1915        },
1916        {
1917            .name = VHDX_BLOCK_OPT_LOG_SIZE,
1918            .type = QEMU_OPT_SIZE,
1919            .def_value_str = stringify(DEFAULT_LOG_SIZE),
1920            .help = "Log size; min 1MB."
1921        },
1922        {
1923            .name = VHDX_BLOCK_OPT_BLOCK_SIZE,
1924            .type = QEMU_OPT_SIZE,
1925            .def_value_str = stringify(0),
1926            .help = "Block Size; min 1MB, max 256MB. " \
1927                    "0 means auto-calculate based on image size."
1928        },
1929        {
1930            .name = BLOCK_OPT_SUBFMT,
1931            .type = QEMU_OPT_STRING,
1932            .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "\
1933                    "Default is 'dynamic'."
1934        },
1935        {
1936            .name = VHDX_BLOCK_OPT_ZERO,
1937            .type = QEMU_OPT_BOOL,
1938            .help = "Force use of payload blocks of type 'ZERO'.  Non-standard."
1939        },
1940        { NULL }
1941     }
1942 };
1943 
1944 static BlockDriver bdrv_vhdx = {
1945     .format_name            = "vhdx",
1946     .instance_size          = sizeof(BDRVVHDXState),
1947     .bdrv_probe             = vhdx_probe,
1948     .bdrv_open              = vhdx_open,
1949     .bdrv_close             = vhdx_close,
1950     .bdrv_reopen_prepare    = vhdx_reopen_prepare,
1951     .bdrv_co_readv          = vhdx_co_readv,
1952     .bdrv_co_writev         = vhdx_co_writev,
1953     .bdrv_create            = vhdx_create,
1954     .bdrv_get_info          = vhdx_get_info,
1955     .bdrv_check             = vhdx_check,
1956 
1957     .create_opts            = &vhdx_create_opts,
1958 };
1959 
1960 static void bdrv_vhdx_init(void)
1961 {
1962     bdrv_register(&bdrv_vhdx);
1963 }
1964 
1965 block_init(bdrv_vhdx_init);
1966