xref: /openbmc/qemu/block/vhdx.c (revision b000325a)
1 /*
2  * Block driver for Hyper-V VHDX Images
3  *
4  * Copyright (c) 2013 Red Hat, Inc.,
5  *
6  * Authors:
7  *  Jeff Cody <jcody@redhat.com>
8  *
9  *  This is based on the "VHDX Format Specification v1.00", published 8/25/2012
10  *  by Microsoft:
11  *      https://www.microsoft.com/en-us/download/details.aspx?id=34750
12  *
13  * This work is licensed under the terms of the GNU LGPL, version 2 or later.
14  * See the COPYING.LIB file in the top-level directory.
15  *
16  */
17 
18 #include "qemu/osdep.h"
19 #include "qapi/error.h"
20 #include "block/block_int.h"
21 #include "block/qdict.h"
22 #include "sysemu/block-backend.h"
23 #include "qemu/module.h"
24 #include "qemu/option.h"
25 #include "qemu/crc32c.h"
26 #include "qemu/bswap.h"
27 #include "vhdx.h"
28 #include "migration/blocker.h"
29 #include "qemu/uuid.h"
30 #include "qapi/qmp/qdict.h"
31 #include "qapi/qobject-input-visitor.h"
32 #include "qapi/qapi-visit-block-core.h"
33 
34 /* Options for VHDX creation */
35 
36 #define VHDX_BLOCK_OPT_LOG_SIZE   "log_size"
37 #define VHDX_BLOCK_OPT_BLOCK_SIZE "block_size"
38 #define VHDX_BLOCK_OPT_ZERO "block_state_zero"
39 
40 typedef enum VHDXImageType {
41     VHDX_TYPE_DYNAMIC = 0,
42     VHDX_TYPE_FIXED,
43     VHDX_TYPE_DIFFERENCING,   /* Currently unsupported */
44 } VHDXImageType;
45 
46 static QemuOptsList vhdx_create_opts;
47 
48 /* Several metadata and region table data entries are identified by
49  * guids in  a MS-specific GUID format. */
50 
51 
52 /* ------- Known Region Table GUIDs ---------------------- */
53 static const MSGUID bat_guid =      { .data1 = 0x2dc27766,
54                                       .data2 = 0xf623,
55                                       .data3 = 0x4200,
56                                       .data4 = { 0x9d, 0x64, 0x11, 0x5e,
57                                                  0x9b, 0xfd, 0x4a, 0x08} };
58 
59 static const MSGUID metadata_guid = { .data1 = 0x8b7ca206,
60                                       .data2 = 0x4790,
61                                       .data3 = 0x4b9a,
62                                       .data4 = { 0xb8, 0xfe, 0x57, 0x5f,
63                                                  0x05, 0x0f, 0x88, 0x6e} };
64 
65 
66 
67 /* ------- Known Metadata Entry GUIDs ---------------------- */
68 static const MSGUID file_param_guid =   { .data1 = 0xcaa16737,
69                                           .data2 = 0xfa36,
70                                           .data3 = 0x4d43,
71                                           .data4 = { 0xb3, 0xb6, 0x33, 0xf0,
72                                                      0xaa, 0x44, 0xe7, 0x6b} };
73 
74 static const MSGUID virtual_size_guid = { .data1 = 0x2FA54224,
75                                           .data2 = 0xcd1b,
76                                           .data3 = 0x4876,
77                                           .data4 = { 0xb2, 0x11, 0x5d, 0xbe,
78                                                      0xd8, 0x3b, 0xf4, 0xb8} };
79 
80 static const MSGUID page83_guid =       { .data1 = 0xbeca12ab,
81                                           .data2 = 0xb2e6,
82                                           .data3 = 0x4523,
83                                           .data4 = { 0x93, 0xef, 0xc3, 0x09,
84                                                      0xe0, 0x00, 0xc7, 0x46} };
85 
86 
87 static const MSGUID phys_sector_guid =  { .data1 = 0xcda348c7,
88                                           .data2 = 0x445d,
89                                           .data3 = 0x4471,
90                                           .data4 = { 0x9c, 0xc9, 0xe9, 0x88,
91                                                      0x52, 0x51, 0xc5, 0x56} };
92 
93 static const MSGUID parent_locator_guid = { .data1 = 0xa8d35f2d,
94                                             .data2 = 0xb30b,
95                                             .data3 = 0x454d,
96                                             .data4 = { 0xab, 0xf7, 0xd3,
97                                                        0xd8, 0x48, 0x34,
98                                                        0xab, 0x0c} };
99 
100 static const MSGUID logical_sector_guid = { .data1 = 0x8141bf1d,
101                                             .data2 = 0xa96f,
102                                             .data3 = 0x4709,
103                                             .data4 = { 0xba, 0x47, 0xf2,
104                                                        0x33, 0xa8, 0xfa,
105                                                        0xab, 0x5f} };
106 
107 /* Each parent type must have a valid GUID; this is for parent images
108  * of type 'VHDX'.  If we were to allow e.g. a QCOW2 parent, we would
109  * need to make up our own QCOW2 GUID type */
110 static const MSGUID parent_vhdx_guid __attribute__((unused))
111                                      = { .data1 = 0xb04aefb7,
112                                          .data2 = 0xd19e,
113                                          .data3 = 0x4a81,
114                                          .data4 = { 0xb7, 0x89, 0x25, 0xb8,
115                                                     0xe9, 0x44, 0x59, 0x13} };
116 
117 
118 #define META_FILE_PARAMETER_PRESENT      0x01
119 #define META_VIRTUAL_DISK_SIZE_PRESENT   0x02
120 #define META_PAGE_83_PRESENT             0x04
121 #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08
122 #define META_PHYS_SECTOR_SIZE_PRESENT    0x10
123 #define META_PARENT_LOCATOR_PRESENT      0x20
124 
125 #define META_ALL_PRESENT    \
126     (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \
127      META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \
128      META_PHYS_SECTOR_SIZE_PRESENT)
129 
130 
131 typedef struct VHDXSectorInfo {
132     uint32_t bat_idx;       /* BAT entry index */
133     uint32_t sectors_avail; /* sectors available in payload block */
134     uint32_t bytes_left;    /* bytes left in the block after data to r/w */
135     uint32_t bytes_avail;   /* bytes available in payload block */
136     uint64_t file_offset;   /* absolute offset in bytes, in file */
137     uint64_t block_offset;  /* block offset, in bytes */
138 } VHDXSectorInfo;
139 
140 /* Calculates new checksum.
141  *
142  * Zero is substituted during crc calculation for the original crc field
143  * crc_offset: byte offset in buf of the buffer crc
144  * buf: buffer pointer
145  * size: size of buffer (must be > crc_offset+4)
146  *
147  * Note: The buffer should have all multi-byte data in little-endian format,
148  *       and the resulting checksum is in little endian format.
149  */
150 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset)
151 {
152     uint32_t crc;
153 
154     assert(buf != NULL);
155     assert(size > (crc_offset + sizeof(crc)));
156 
157     memset(buf + crc_offset, 0, sizeof(crc));
158     crc =  crc32c(0xffffffff, buf, size);
159     cpu_to_le32s(&crc);
160     memcpy(buf + crc_offset, &crc, sizeof(crc));
161 
162     return crc;
163 }
164 
165 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size,
166                             int crc_offset)
167 {
168     uint32_t crc_new;
169     uint32_t crc_orig;
170     assert(buf != NULL);
171 
172     if (crc_offset > 0) {
173         memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
174         memset(buf + crc_offset, 0, sizeof(crc_orig));
175     }
176 
177     crc_new = crc32c(crc, buf, size);
178     if (crc_offset > 0) {
179         memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig));
180     }
181 
182     return crc_new;
183 }
184 
185 /* Validates the checksum of the buffer, with an in-place CRC.
186  *
187  * Zero is substituted during crc calculation for the original crc field,
188  * and the crc field is restored afterwards.  But the buffer will be modifed
189  * during the calculation, so this may not be not suitable for multi-threaded
190  * use.
191  *
192  * crc_offset: byte offset in buf of the buffer crc
193  * buf: buffer pointer
194  * size: size of buffer (must be > crc_offset+4)
195  *
196  * returns true if checksum is valid, false otherwise
197  */
198 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset)
199 {
200     uint32_t crc_orig;
201     uint32_t crc;
202 
203     assert(buf != NULL);
204     assert(size > (crc_offset + 4));
205 
206     memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
207     crc_orig = le32_to_cpu(crc_orig);
208 
209     crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset);
210 
211     return crc == crc_orig;
212 }
213 
214 
215 /*
216  * This generates a UUID that is compliant with the MS GUIDs used
217  * in the VHDX spec (and elsewhere).
218  */
219 void vhdx_guid_generate(MSGUID *guid)
220 {
221     QemuUUID uuid;
222     assert(guid != NULL);
223 
224     qemu_uuid_generate(&uuid);
225     memcpy(guid, &uuid, sizeof(MSGUID));
226 }
227 
228 /* Check for region overlaps inside the VHDX image */
229 static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length)
230 {
231     int ret = 0;
232     uint64_t end;
233     VHDXRegionEntry *r;
234 
235     end = start + length;
236     QLIST_FOREACH(r, &s->regions, entries) {
237         if (!((start >= r->end) || (end <= r->start))) {
238             ret = -EINVAL;
239             goto exit;
240         }
241     }
242 
243 exit:
244     return ret;
245 }
246 
247 /* Register a region for future checks */
248 static void vhdx_region_register(BDRVVHDXState *s,
249                                  uint64_t start, uint64_t length)
250 {
251     VHDXRegionEntry *r;
252 
253     r = g_malloc0(sizeof(*r));
254 
255     r->start = start;
256     r->end = start + length;
257 
258     QLIST_INSERT_HEAD(&s->regions, r, entries);
259 }
260 
261 /* Free all registered regions */
262 static void vhdx_region_unregister_all(BDRVVHDXState *s)
263 {
264     VHDXRegionEntry *r, *r_next;
265 
266     QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) {
267         QLIST_REMOVE(r, entries);
268         g_free(r);
269     }
270 }
271 
272 static void vhdx_set_shift_bits(BDRVVHDXState *s)
273 {
274     s->logical_sector_size_bits = ctz32(s->logical_sector_size);
275     s->sectors_per_block_bits =   ctz32(s->sectors_per_block);
276     s->chunk_ratio_bits =         ctz64(s->chunk_ratio);
277     s->block_size_bits =          ctz32(s->block_size);
278 }
279 
280 /*
281  * Per the MS VHDX Specification, for every VHDX file:
282  *      - The header section is fixed size - 1 MB
283  *      - The header section is always the first "object"
284  *      - The first 64KB of the header is the File Identifier
285  *      - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile")
286  *      - The following 512 bytes constitute a UTF-16 string identifiying the
287  *        software that created the file, and is optional and diagnostic only.
288  *
289  *  Therefore, we probe by looking for the vhdxfile signature "vhdxfile"
290  */
291 static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename)
292 {
293     if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) {
294         return 100;
295     }
296     return 0;
297 }
298 
299 /*
300  * Writes the header to the specified offset.
301  *
302  * This will optionally read in buffer data from disk (otherwise zero-fill),
303  * and then update the header checksum.  Header is converted to proper
304  * endianness before being written to the specified file offset
305  */
306 static int vhdx_write_header(BdrvChild *file, VHDXHeader *hdr,
307                              uint64_t offset, bool read)
308 {
309     BlockDriverState *bs_file = file->bs;
310     uint8_t *buffer = NULL;
311     int ret;
312     VHDXHeader *header_le;
313 
314     assert(bs_file != NULL);
315     assert(hdr != NULL);
316 
317     /* the header checksum is not over just the packed size of VHDXHeader,
318      * but rather over the entire 'reserved' range for the header, which is
319      * 4KB (VHDX_HEADER_SIZE). */
320 
321     buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE);
322     if (read) {
323         /* if true, we can't assume the extra reserved bytes are 0 */
324         ret = bdrv_pread(file, offset, buffer, VHDX_HEADER_SIZE);
325         if (ret < 0) {
326             goto exit;
327         }
328     } else {
329         memset(buffer, 0, VHDX_HEADER_SIZE);
330     }
331 
332     /* overwrite the actual VHDXHeader portion */
333     header_le = (VHDXHeader *)buffer;
334     memcpy(header_le, hdr, sizeof(VHDXHeader));
335     vhdx_header_le_export(hdr, header_le);
336     vhdx_update_checksum(buffer, VHDX_HEADER_SIZE,
337                          offsetof(VHDXHeader, checksum));
338     ret = bdrv_pwrite_sync(file, offset, header_le, sizeof(VHDXHeader));
339 
340 exit:
341     qemu_vfree(buffer);
342     return ret;
343 }
344 
345 /* Update the VHDX headers
346  *
347  * This follows the VHDX spec procedures for header updates.
348  *
349  *  - non-current header is updated with largest sequence number
350  */
351 static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s,
352                               bool generate_data_write_guid, MSGUID *log_guid)
353 {
354     int ret = 0;
355     int hdr_idx = 0;
356     uint64_t header_offset = VHDX_HEADER1_OFFSET;
357 
358     VHDXHeader *active_header;
359     VHDXHeader *inactive_header;
360 
361     /* operate on the non-current header */
362     if (s->curr_header == 0) {
363         hdr_idx = 1;
364         header_offset = VHDX_HEADER2_OFFSET;
365     }
366 
367     active_header   = s->headers[s->curr_header];
368     inactive_header = s->headers[hdr_idx];
369 
370     inactive_header->sequence_number = active_header->sequence_number + 1;
371 
372     /* a new file guid must be generated before any file write, including
373      * headers */
374     inactive_header->file_write_guid = s->session_guid;
375 
376     /* a new data guid only needs to be generated before any guest-visible
377      * writes (i.e. something observable via virtual disk read) */
378     if (generate_data_write_guid) {
379         vhdx_guid_generate(&inactive_header->data_write_guid);
380     }
381 
382     /* update the log guid if present */
383     if (log_guid) {
384         inactive_header->log_guid = *log_guid;
385     }
386 
387     ret = vhdx_write_header(bs->file, inactive_header, header_offset, true);
388     if (ret < 0) {
389         goto exit;
390     }
391     s->curr_header = hdr_idx;
392 
393 exit:
394     return ret;
395 }
396 
397 /*
398  * The VHDX spec calls for header updates to be performed twice, so that both
399  * the current and non-current header have valid info
400  */
401 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s,
402                         bool generate_data_write_guid, MSGUID *log_guid)
403 {
404     int ret;
405 
406     ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
407     if (ret < 0) {
408         return ret;
409     }
410     ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
411     return ret;
412 }
413 
414 /* opens the specified header block from the VHDX file header section */
415 static void vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s,
416                               Error **errp)
417 {
418     int ret;
419     VHDXHeader *header1;
420     VHDXHeader *header2;
421     bool h1_valid = false;
422     bool h2_valid = false;
423     uint64_t h1_seq = 0;
424     uint64_t h2_seq = 0;
425     uint8_t *buffer;
426 
427     /* header1 & header2 are freed in vhdx_close() */
428     header1 = qemu_blockalign(bs, sizeof(VHDXHeader));
429     header2 = qemu_blockalign(bs, sizeof(VHDXHeader));
430 
431     buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE);
432 
433     s->headers[0] = header1;
434     s->headers[1] = header2;
435 
436     /* We have to read the whole VHDX_HEADER_SIZE instead of
437      * sizeof(VHDXHeader), because the checksum is over the whole
438      * region */
439     ret = bdrv_pread(bs->file, VHDX_HEADER1_OFFSET, buffer,
440                      VHDX_HEADER_SIZE);
441     if (ret < 0) {
442         goto fail;
443     }
444     /* copy over just the relevant portion that we need */
445     memcpy(header1, buffer, sizeof(VHDXHeader));
446 
447     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
448         vhdx_header_le_import(header1);
449         if (header1->signature == VHDX_HEADER_SIGNATURE &&
450             header1->version == 1) {
451             h1_seq = header1->sequence_number;
452             h1_valid = true;
453         }
454     }
455 
456     ret = bdrv_pread(bs->file, VHDX_HEADER2_OFFSET, buffer,
457                      VHDX_HEADER_SIZE);
458     if (ret < 0) {
459         goto fail;
460     }
461     /* copy over just the relevant portion that we need */
462     memcpy(header2, buffer, sizeof(VHDXHeader));
463 
464     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
465         vhdx_header_le_import(header2);
466         if (header2->signature == VHDX_HEADER_SIGNATURE &&
467             header2->version == 1) {
468             h2_seq = header2->sequence_number;
469             h2_valid = true;
470         }
471     }
472 
473     /* If there is only 1 valid header (or no valid headers), we
474      * don't care what the sequence numbers are */
475     if (h1_valid && !h2_valid) {
476         s->curr_header = 0;
477     } else if (!h1_valid && h2_valid) {
478         s->curr_header = 1;
479     } else if (!h1_valid && !h2_valid) {
480         goto fail;
481     } else {
482         /* If both headers are valid, then we choose the active one by the
483          * highest sequence number.  If the sequence numbers are equal, that is
484          * invalid */
485         if (h1_seq > h2_seq) {
486             s->curr_header = 0;
487         } else if (h2_seq > h1_seq) {
488             s->curr_header = 1;
489         } else {
490             /* The Microsoft Disk2VHD tool will create 2 identical
491              * headers, with identical sequence numbers.  If the headers are
492              * identical, don't consider the file corrupt */
493             if (!memcmp(header1, header2, sizeof(VHDXHeader))) {
494                 s->curr_header = 0;
495             } else {
496                 goto fail;
497             }
498         }
499     }
500 
501     vhdx_region_register(s, s->headers[s->curr_header]->log_offset,
502                             s->headers[s->curr_header]->log_length);
503     goto exit;
504 
505 fail:
506     error_setg_errno(errp, -ret, "No valid VHDX header found");
507     qemu_vfree(header1);
508     qemu_vfree(header2);
509     s->headers[0] = NULL;
510     s->headers[1] = NULL;
511 exit:
512     qemu_vfree(buffer);
513 }
514 
515 
516 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s)
517 {
518     int ret = 0;
519     uint8_t *buffer;
520     int offset = 0;
521     VHDXRegionTableEntry rt_entry;
522     uint32_t i;
523     bool bat_rt_found = false;
524     bool metadata_rt_found = false;
525 
526     /* We have to read the whole 64KB block, because the crc32 is over the
527      * whole block */
528     buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE);
529 
530     ret = bdrv_pread(bs->file, VHDX_REGION_TABLE_OFFSET, buffer,
531                      VHDX_HEADER_BLOCK_SIZE);
532     if (ret < 0) {
533         goto fail;
534     }
535     memcpy(&s->rt, buffer, sizeof(s->rt));
536     offset += sizeof(s->rt);
537 
538     if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4)) {
539         ret = -EINVAL;
540         goto fail;
541     }
542 
543     vhdx_region_header_le_import(&s->rt);
544 
545     if (s->rt.signature != VHDX_REGION_SIGNATURE) {
546         ret = -EINVAL;
547         goto fail;
548     }
549 
550 
551     /* Per spec, maximum region table entry count is 2047 */
552     if (s->rt.entry_count > 2047) {
553         ret = -EINVAL;
554         goto fail;
555     }
556 
557     for (i = 0; i < s->rt.entry_count; i++) {
558         memcpy(&rt_entry, buffer + offset, sizeof(rt_entry));
559         offset += sizeof(rt_entry);
560 
561         vhdx_region_entry_le_import(&rt_entry);
562 
563         /* check for region overlap between these entries, and any
564          * other memory regions in the file */
565         ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length);
566         if (ret < 0) {
567             goto fail;
568         }
569 
570         vhdx_region_register(s, rt_entry.file_offset, rt_entry.length);
571 
572         /* see if we recognize the entry */
573         if (guid_eq(rt_entry.guid, bat_guid)) {
574             /* must be unique; if we have already found it this is invalid */
575             if (bat_rt_found) {
576                 ret = -EINVAL;
577                 goto fail;
578             }
579             bat_rt_found = true;
580             s->bat_rt = rt_entry;
581             continue;
582         }
583 
584         if (guid_eq(rt_entry.guid, metadata_guid)) {
585             /* must be unique; if we have already found it this is invalid */
586             if (metadata_rt_found) {
587                 ret = -EINVAL;
588                 goto fail;
589             }
590             metadata_rt_found = true;
591             s->metadata_rt = rt_entry;
592             continue;
593         }
594 
595         if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) {
596             /* cannot read vhdx file - required region table entry that
597              * we do not understand.  per spec, we must fail to open */
598             ret = -ENOTSUP;
599             goto fail;
600         }
601     }
602 
603     if (!bat_rt_found || !metadata_rt_found) {
604         ret = -EINVAL;
605         goto fail;
606     }
607 
608     ret = 0;
609 
610 fail:
611     qemu_vfree(buffer);
612     return ret;
613 }
614 
615 
616 
617 /* Metadata initial parser
618  *
619  * This loads all the metadata entry fields.  This may cause additional
620  * fields to be processed (e.g. parent locator, etc..).
621  *
622  * There are 5 Metadata items that are always required:
623  *      - File Parameters (block size, has a parent)
624  *      - Virtual Disk Size (size, in bytes, of the virtual drive)
625  *      - Page 83 Data (scsi page 83 guid)
626  *      - Logical Sector Size (logical sector size in bytes, either 512 or
627  *                             4096.  We only support 512 currently)
628  *      - Physical Sector Size (512 or 4096)
629  *
630  * Also, if the File Parameters indicate this is a differencing file,
631  * we must also look for the Parent Locator metadata item.
632  */
633 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s)
634 {
635     int ret = 0;
636     uint8_t *buffer;
637     int offset = 0;
638     uint32_t i = 0;
639     VHDXMetadataTableEntry md_entry;
640 
641     buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE);
642 
643     ret = bdrv_pread(bs->file, s->metadata_rt.file_offset, buffer,
644                      VHDX_METADATA_TABLE_MAX_SIZE);
645     if (ret < 0) {
646         goto exit;
647     }
648     memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr));
649     offset += sizeof(s->metadata_hdr);
650 
651     vhdx_metadata_header_le_import(&s->metadata_hdr);
652 
653     if (s->metadata_hdr.signature != VHDX_METADATA_SIGNATURE) {
654         ret = -EINVAL;
655         goto exit;
656     }
657 
658     s->metadata_entries.present = 0;
659 
660     if ((s->metadata_hdr.entry_count * sizeof(md_entry)) >
661         (VHDX_METADATA_TABLE_MAX_SIZE - offset)) {
662         ret = -EINVAL;
663         goto exit;
664     }
665 
666     for (i = 0; i < s->metadata_hdr.entry_count; i++) {
667         memcpy(&md_entry, buffer + offset, sizeof(md_entry));
668         offset += sizeof(md_entry);
669 
670         vhdx_metadata_entry_le_import(&md_entry);
671 
672         if (guid_eq(md_entry.item_id, file_param_guid)) {
673             if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) {
674                 ret = -EINVAL;
675                 goto exit;
676             }
677             s->metadata_entries.file_parameters_entry = md_entry;
678             s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT;
679             continue;
680         }
681 
682         if (guid_eq(md_entry.item_id, virtual_size_guid)) {
683             if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) {
684                 ret = -EINVAL;
685                 goto exit;
686             }
687             s->metadata_entries.virtual_disk_size_entry = md_entry;
688             s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT;
689             continue;
690         }
691 
692         if (guid_eq(md_entry.item_id, page83_guid)) {
693             if (s->metadata_entries.present & META_PAGE_83_PRESENT) {
694                 ret = -EINVAL;
695                 goto exit;
696             }
697             s->metadata_entries.page83_data_entry = md_entry;
698             s->metadata_entries.present |= META_PAGE_83_PRESENT;
699             continue;
700         }
701 
702         if (guid_eq(md_entry.item_id, logical_sector_guid)) {
703             if (s->metadata_entries.present &
704                 META_LOGICAL_SECTOR_SIZE_PRESENT) {
705                 ret = -EINVAL;
706                 goto exit;
707             }
708             s->metadata_entries.logical_sector_size_entry = md_entry;
709             s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT;
710             continue;
711         }
712 
713         if (guid_eq(md_entry.item_id, phys_sector_guid)) {
714             if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) {
715                 ret = -EINVAL;
716                 goto exit;
717             }
718             s->metadata_entries.phys_sector_size_entry = md_entry;
719             s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT;
720             continue;
721         }
722 
723         if (guid_eq(md_entry.item_id, parent_locator_guid)) {
724             if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
725                 ret = -EINVAL;
726                 goto exit;
727             }
728             s->metadata_entries.parent_locator_entry = md_entry;
729             s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT;
730             continue;
731         }
732 
733         if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) {
734             /* cannot read vhdx file - required region table entry that
735              * we do not understand.  per spec, we must fail to open */
736             ret = -ENOTSUP;
737             goto exit;
738         }
739     }
740 
741     if (s->metadata_entries.present != META_ALL_PRESENT) {
742         ret = -ENOTSUP;
743         goto exit;
744     }
745 
746     ret = bdrv_pread(bs->file,
747                      s->metadata_entries.file_parameters_entry.offset
748                                          + s->metadata_rt.file_offset,
749                      &s->params,
750                      sizeof(s->params));
751 
752     if (ret < 0) {
753         goto exit;
754     }
755 
756     le32_to_cpus(&s->params.block_size);
757     le32_to_cpus(&s->params.data_bits);
758 
759 
760     /* We now have the file parameters, so we can tell if this is a
761      * differencing file (i.e.. has_parent), is dynamic or fixed
762      * sized (leave_blocks_allocated), and the block size */
763 
764     /* The parent locator required iff the file parameters has_parent set */
765     if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
766         if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
767             /* TODO: parse  parent locator fields */
768             ret = -ENOTSUP; /* temp, until differencing files are supported */
769             goto exit;
770         } else {
771             /* if has_parent is set, but there is not parent locator present,
772              * then that is an invalid combination */
773             ret = -EINVAL;
774             goto exit;
775         }
776     }
777 
778     /* determine virtual disk size, logical sector size,
779      * and phys sector size */
780 
781     ret = bdrv_pread(bs->file,
782                      s->metadata_entries.virtual_disk_size_entry.offset
783                                            + s->metadata_rt.file_offset,
784                      &s->virtual_disk_size,
785                      sizeof(uint64_t));
786     if (ret < 0) {
787         goto exit;
788     }
789     ret = bdrv_pread(bs->file,
790                      s->metadata_entries.logical_sector_size_entry.offset
791                                              + s->metadata_rt.file_offset,
792                      &s->logical_sector_size,
793                      sizeof(uint32_t));
794     if (ret < 0) {
795         goto exit;
796     }
797     ret = bdrv_pread(bs->file,
798                      s->metadata_entries.phys_sector_size_entry.offset
799                                           + s->metadata_rt.file_offset,
800                      &s->physical_sector_size,
801                      sizeof(uint32_t));
802     if (ret < 0) {
803         goto exit;
804     }
805 
806     le64_to_cpus(&s->virtual_disk_size);
807     le32_to_cpus(&s->logical_sector_size);
808     le32_to_cpus(&s->physical_sector_size);
809 
810     if (s->params.block_size < VHDX_BLOCK_SIZE_MIN ||
811         s->params.block_size > VHDX_BLOCK_SIZE_MAX) {
812         ret = -EINVAL;
813         goto exit;
814     }
815 
816     /* only 2 supported sector sizes */
817     if (s->logical_sector_size != 512 && s->logical_sector_size != 4096) {
818         ret = -EINVAL;
819         goto exit;
820     }
821 
822     /* Both block_size and sector_size are guaranteed powers of 2, below.
823        Due to range checks above, s->sectors_per_block can never be < 256 */
824     s->sectors_per_block = s->params.block_size / s->logical_sector_size;
825     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
826                      (uint64_t)s->logical_sector_size /
827                      (uint64_t)s->params.block_size;
828 
829     /* These values are ones we will want to use for division / multiplication
830      * later on, and they are all guaranteed (per the spec) to be powers of 2,
831      * so we can take advantage of that for shift operations during
832      * reads/writes */
833     if (s->logical_sector_size & (s->logical_sector_size - 1)) {
834         ret = -EINVAL;
835         goto exit;
836     }
837     if (s->sectors_per_block & (s->sectors_per_block - 1)) {
838         ret = -EINVAL;
839         goto exit;
840     }
841     if (s->chunk_ratio & (s->chunk_ratio - 1)) {
842         ret = -EINVAL;
843         goto exit;
844     }
845     s->block_size = s->params.block_size;
846     if (s->block_size & (s->block_size - 1)) {
847         ret = -EINVAL;
848         goto exit;
849     }
850 
851     vhdx_set_shift_bits(s);
852 
853     ret = 0;
854 
855 exit:
856     qemu_vfree(buffer);
857     return ret;
858 }
859 
860 /*
861  * Calculate the number of BAT entries, including sector
862  * bitmap entries.
863  */
864 static void vhdx_calc_bat_entries(BDRVVHDXState *s)
865 {
866     uint32_t data_blocks_cnt, bitmap_blocks_cnt;
867 
868     data_blocks_cnt = DIV_ROUND_UP(s->virtual_disk_size, s->block_size);
869     bitmap_blocks_cnt = DIV_ROUND_UP(data_blocks_cnt, s->chunk_ratio);
870 
871     if (s->parent_entries) {
872         s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1);
873     } else {
874         s->bat_entries = data_blocks_cnt +
875                          ((data_blocks_cnt - 1) >> s->chunk_ratio_bits);
876     }
877 
878 }
879 
880 static void vhdx_close(BlockDriverState *bs)
881 {
882     BDRVVHDXState *s = bs->opaque;
883     qemu_vfree(s->headers[0]);
884     s->headers[0] = NULL;
885     qemu_vfree(s->headers[1]);
886     s->headers[1] = NULL;
887     qemu_vfree(s->bat);
888     s->bat = NULL;
889     qemu_vfree(s->parent_entries);
890     s->parent_entries = NULL;
891     migrate_del_blocker(s->migration_blocker);
892     error_free(s->migration_blocker);
893     qemu_vfree(s->log.hdr);
894     s->log.hdr = NULL;
895     vhdx_region_unregister_all(s);
896 }
897 
898 static int vhdx_open(BlockDriverState *bs, QDict *options, int flags,
899                      Error **errp)
900 {
901     BDRVVHDXState *s = bs->opaque;
902     int ret = 0;
903     uint32_t i;
904     uint64_t signature;
905     Error *local_err = NULL;
906 
907     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
908                                false, errp);
909     if (!bs->file) {
910         return -EINVAL;
911     }
912 
913     s->bat = NULL;
914     s->first_visible_write = true;
915 
916     qemu_co_mutex_init(&s->lock);
917     QLIST_INIT(&s->regions);
918 
919     /* validate the file signature */
920     ret = bdrv_pread(bs->file, 0, &signature, sizeof(uint64_t));
921     if (ret < 0) {
922         goto fail;
923     }
924     if (memcmp(&signature, "vhdxfile", 8)) {
925         ret = -EINVAL;
926         goto fail;
927     }
928 
929     /* This is used for any header updates, for the file_write_guid.
930      * The spec dictates that a new value should be used for the first
931      * header update */
932     vhdx_guid_generate(&s->session_guid);
933 
934     vhdx_parse_header(bs, s, &local_err);
935     if (local_err != NULL) {
936         error_propagate(errp, local_err);
937         ret = -EINVAL;
938         goto fail;
939     }
940 
941     ret = vhdx_parse_log(bs, s, &s->log_replayed_on_open, errp);
942     if (ret < 0) {
943         goto fail;
944     }
945 
946     ret = vhdx_open_region_tables(bs, s);
947     if (ret < 0) {
948         goto fail;
949     }
950 
951     ret = vhdx_parse_metadata(bs, s);
952     if (ret < 0) {
953         goto fail;
954     }
955 
956     s->block_size = s->params.block_size;
957 
958     /* the VHDX spec dictates that virtual_disk_size is always a multiple of
959      * logical_sector_size */
960     bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits;
961 
962     vhdx_calc_bat_entries(s);
963 
964     s->bat_offset = s->bat_rt.file_offset;
965 
966     if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) {
967         /* BAT allocation is not large enough for all entries */
968         ret = -EINVAL;
969         goto fail;
970     }
971 
972     /* s->bat is freed in vhdx_close() */
973     s->bat = qemu_try_blockalign(bs->file->bs, s->bat_rt.length);
974     if (s->bat == NULL) {
975         ret = -ENOMEM;
976         goto fail;
977     }
978 
979     ret = bdrv_pread(bs->file, s->bat_offset, s->bat, s->bat_rt.length);
980     if (ret < 0) {
981         goto fail;
982     }
983 
984     uint64_t payblocks = s->chunk_ratio;
985     /* endian convert, and verify populated BAT field file offsets against
986      * region table and log entries */
987     for (i = 0; i < s->bat_entries; i++) {
988         le64_to_cpus(&s->bat[i]);
989         if (payblocks--) {
990             /* payload bat entries */
991             if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) ==
992                     PAYLOAD_BLOCK_FULLY_PRESENT) {
993                 ret = vhdx_region_check(s, s->bat[i] & VHDX_BAT_FILE_OFF_MASK,
994                                         s->block_size);
995                 if (ret < 0) {
996                     goto fail;
997                 }
998             }
999         } else {
1000             payblocks = s->chunk_ratio;
1001             /* Once differencing files are supported, verify sector bitmap
1002              * blocks here */
1003         }
1004     }
1005 
1006     /* Disable migration when VHDX images are used */
1007     error_setg(&s->migration_blocker, "The vhdx format used by node '%s' "
1008                "does not support live migration",
1009                bdrv_get_device_or_node_name(bs));
1010     ret = migrate_add_blocker(s->migration_blocker, &local_err);
1011     if (local_err) {
1012         error_propagate(errp, local_err);
1013         error_free(s->migration_blocker);
1014         goto fail;
1015     }
1016 
1017     /* TODO: differencing files */
1018 
1019     return 0;
1020 fail:
1021     vhdx_close(bs);
1022     return ret;
1023 }
1024 
1025 static int vhdx_reopen_prepare(BDRVReopenState *state,
1026                                BlockReopenQueue *queue, Error **errp)
1027 {
1028     return 0;
1029 }
1030 
1031 
1032 /*
1033  * Perform sector to block offset translations, to get various
1034  * sector and file offsets into the image.  See VHDXSectorInfo
1035  */
1036 static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num,
1037                                  int nb_sectors, VHDXSectorInfo *sinfo)
1038 {
1039     uint32_t block_offset;
1040 
1041     sinfo->bat_idx = sector_num >> s->sectors_per_block_bits;
1042     /* effectively a modulo - this gives us the offset into the block
1043      * (in sector sizes) for our sector number */
1044     block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits);
1045     /* the chunk ratio gives us the interleaving of the sector
1046      * bitmaps, so we need to advance our page block index by the
1047      * sector bitmaps entry number */
1048     sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits;
1049 
1050     /* the number of sectors we can read/write in this cycle */
1051     sinfo->sectors_avail = s->sectors_per_block - block_offset;
1052 
1053     sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits;
1054 
1055     if (sinfo->sectors_avail > nb_sectors) {
1056         sinfo->sectors_avail = nb_sectors;
1057     }
1058 
1059     sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits;
1060 
1061     sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK;
1062 
1063     sinfo->block_offset = block_offset << s->logical_sector_size_bits;
1064 
1065     /* The file offset must be past the header section, so must be > 0 */
1066     if (sinfo->file_offset == 0) {
1067         return;
1068     }
1069 
1070     /* block offset is the offset in vhdx logical sectors, in
1071      * the payload data block. Convert that to a byte offset
1072      * in the block, and add in the payload data block offset
1073      * in the file, in bytes, to get the final read address */
1074 
1075     sinfo->file_offset += sinfo->block_offset;
1076 }
1077 
1078 
1079 static int vhdx_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1080 {
1081     BDRVVHDXState *s = bs->opaque;
1082 
1083     bdi->cluster_size = s->block_size;
1084 
1085     bdi->unallocated_blocks_are_zero =
1086         (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) == 0;
1087 
1088     return 0;
1089 }
1090 
1091 
1092 static coroutine_fn int vhdx_co_readv(BlockDriverState *bs, int64_t sector_num,
1093                                       int nb_sectors, QEMUIOVector *qiov)
1094 {
1095     BDRVVHDXState *s = bs->opaque;
1096     int ret = 0;
1097     VHDXSectorInfo sinfo;
1098     uint64_t bytes_done = 0;
1099     QEMUIOVector hd_qiov;
1100 
1101     qemu_iovec_init(&hd_qiov, qiov->niov);
1102 
1103     qemu_co_mutex_lock(&s->lock);
1104 
1105     while (nb_sectors > 0) {
1106         /* We are a differencing file, so we need to inspect the sector bitmap
1107          * to see if we have the data or not */
1108         if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1109             /* not supported yet */
1110             ret = -ENOTSUP;
1111             goto exit;
1112         } else {
1113             vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1114 
1115             qemu_iovec_reset(&hd_qiov);
1116             qemu_iovec_concat(&hd_qiov, qiov,  bytes_done, sinfo.bytes_avail);
1117 
1118             /* check the payload block state */
1119             switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) {
1120             case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1121             case PAYLOAD_BLOCK_UNDEFINED:
1122             case PAYLOAD_BLOCK_UNMAPPED:
1123             case PAYLOAD_BLOCK_UNMAPPED_v095:
1124             case PAYLOAD_BLOCK_ZERO:
1125                 /* return zero */
1126                 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail);
1127                 break;
1128             case PAYLOAD_BLOCK_FULLY_PRESENT:
1129                 qemu_co_mutex_unlock(&s->lock);
1130                 ret = bdrv_co_preadv(bs->file, sinfo.file_offset,
1131                                      sinfo.sectors_avail * BDRV_SECTOR_SIZE,
1132                                      &hd_qiov, 0);
1133                 qemu_co_mutex_lock(&s->lock);
1134                 if (ret < 0) {
1135                     goto exit;
1136                 }
1137                 break;
1138             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1139                 /* we don't yet support difference files, fall through
1140                  * to error */
1141             default:
1142                 ret = -EIO;
1143                 goto exit;
1144                 break;
1145             }
1146             nb_sectors -= sinfo.sectors_avail;
1147             sector_num += sinfo.sectors_avail;
1148             bytes_done += sinfo.bytes_avail;
1149         }
1150     }
1151     ret = 0;
1152 exit:
1153     qemu_co_mutex_unlock(&s->lock);
1154     qemu_iovec_destroy(&hd_qiov);
1155     return ret;
1156 }
1157 
1158 /*
1159  * Allocate a new payload block at the end of the file.
1160  *
1161  * Allocation will happen at 1MB alignment inside the file
1162  *
1163  * Returns the file offset start of the new payload block
1164  */
1165 static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s,
1166                                     uint64_t *new_offset)
1167 {
1168     int64_t current_len;
1169 
1170     current_len = bdrv_getlength(bs->file->bs);
1171     if (current_len < 0) {
1172         return current_len;
1173     }
1174 
1175     *new_offset = current_len;
1176 
1177     /* per the spec, the address for a block is in units of 1MB */
1178     *new_offset = ROUND_UP(*new_offset, 1024 * 1024);
1179     if (*new_offset > INT64_MAX) {
1180         return -EINVAL;
1181     }
1182 
1183     return bdrv_truncate(bs->file, *new_offset + s->block_size,
1184                          PREALLOC_MODE_OFF, NULL);
1185 }
1186 
1187 /*
1188  * Update the BAT table entry with the new file offset, and the new entry
1189  * state */
1190 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s,
1191                                        VHDXSectorInfo *sinfo,
1192                                        uint64_t *bat_entry_le,
1193                                        uint64_t *bat_offset, int state)
1194 {
1195     /* The BAT entry is a uint64, with 44 bits for the file offset in units of
1196      * 1MB, and 3 bits for the block state. */
1197     if ((state == PAYLOAD_BLOCK_ZERO)        ||
1198         (state == PAYLOAD_BLOCK_UNDEFINED)   ||
1199         (state == PAYLOAD_BLOCK_NOT_PRESENT) ||
1200         (state == PAYLOAD_BLOCK_UNMAPPED)) {
1201         s->bat[sinfo->bat_idx]  = 0;  /* For PAYLOAD_BLOCK_ZERO, the
1202                                          FileOffsetMB field is denoted as
1203                                          'reserved' in the v1.0 spec.  If it is
1204                                          non-zero, MS Hyper-V will fail to read
1205                                          the disk image */
1206     } else {
1207         s->bat[sinfo->bat_idx]  = sinfo->file_offset;
1208     }
1209 
1210     s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK;
1211 
1212     *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]);
1213     *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry);
1214 
1215 }
1216 
1217 /* Per the spec, on the first write of guest-visible data to the file the
1218  * data write guid must be updated in the header */
1219 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s)
1220 {
1221     int ret = 0;
1222     if (s->first_visible_write) {
1223         s->first_visible_write = false;
1224         ret = vhdx_update_headers(bs, s, true, NULL);
1225     }
1226     return ret;
1227 }
1228 
1229 static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num,
1230                                        int nb_sectors, QEMUIOVector *qiov,
1231                                        int flags)
1232 {
1233     int ret = -ENOTSUP;
1234     BDRVVHDXState *s = bs->opaque;
1235     VHDXSectorInfo sinfo;
1236     uint64_t bytes_done = 0;
1237     uint64_t bat_entry = 0;
1238     uint64_t bat_entry_offset = 0;
1239     QEMUIOVector hd_qiov;
1240     struct iovec iov1 = { 0 };
1241     struct iovec iov2 = { 0 };
1242     int sectors_to_write;
1243     int bat_state;
1244     uint64_t bat_prior_offset = 0;
1245     bool bat_update = false;
1246 
1247     assert(!flags);
1248     qemu_iovec_init(&hd_qiov, qiov->niov);
1249 
1250     qemu_co_mutex_lock(&s->lock);
1251 
1252     ret = vhdx_user_visible_write(bs, s);
1253     if (ret < 0) {
1254         goto exit;
1255     }
1256 
1257     while (nb_sectors > 0) {
1258         bool use_zero_buffers = false;
1259         bat_update = false;
1260         if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1261             /* not supported yet */
1262             ret = -ENOTSUP;
1263             goto exit;
1264         } else {
1265             vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1266             sectors_to_write = sinfo.sectors_avail;
1267 
1268             qemu_iovec_reset(&hd_qiov);
1269             /* check the payload block state */
1270             bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK;
1271             switch (bat_state) {
1272             case PAYLOAD_BLOCK_ZERO:
1273                 /* in this case, we need to preserve zero writes for
1274                  * data that is not part of this write, so we must pad
1275                  * the rest of the buffer to zeroes */
1276 
1277                 /* if we are on a posix system with ftruncate() that extends
1278                  * a file, then it is zero-filled for us.  On Win32, the raw
1279                  * layer uses SetFilePointer and SetFileEnd, which does not
1280                  * zero fill AFAIK */
1281 
1282                 /* Queue another write of zero buffers if the underlying file
1283                  * does not zero-fill on file extension */
1284 
1285                 if (bdrv_has_zero_init(bs->file->bs) == 0) {
1286                     use_zero_buffers = true;
1287 
1288                     /* zero fill the front, if any */
1289                     if (sinfo.block_offset) {
1290                         iov1.iov_len = sinfo.block_offset;
1291                         iov1.iov_base = qemu_blockalign(bs, iov1.iov_len);
1292                         memset(iov1.iov_base, 0, iov1.iov_len);
1293                         qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0,
1294                                               iov1.iov_len);
1295                         sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS;
1296                     }
1297 
1298                     /* our actual data */
1299                     qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1300                                       sinfo.bytes_avail);
1301 
1302                     /* zero fill the back, if any */
1303                     if ((sinfo.bytes_avail - sinfo.block_offset) <
1304                          s->block_size) {
1305                         iov2.iov_len = s->block_size -
1306                                       (sinfo.bytes_avail + sinfo.block_offset);
1307                         iov2.iov_base = qemu_blockalign(bs, iov2.iov_len);
1308                         memset(iov2.iov_base, 0, iov2.iov_len);
1309                         qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0,
1310                                               iov2.iov_len);
1311                         sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS;
1312                     }
1313                 }
1314                 /* fall through */
1315             case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1316             case PAYLOAD_BLOCK_UNMAPPED:
1317             case PAYLOAD_BLOCK_UNMAPPED_v095:
1318             case PAYLOAD_BLOCK_UNDEFINED:
1319                 bat_prior_offset = sinfo.file_offset;
1320                 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset);
1321                 if (ret < 0) {
1322                     goto exit;
1323                 }
1324                 /* once we support differencing files, this may also be
1325                  * partially present */
1326                 /* update block state to the newly specified state */
1327                 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1328                                             &bat_entry_offset,
1329                                             PAYLOAD_BLOCK_FULLY_PRESENT);
1330                 bat_update = true;
1331                 /* since we just allocated a block, file_offset is the
1332                  * beginning of the payload block. It needs to be the
1333                  * write address, which includes the offset into the block */
1334                 if (!use_zero_buffers) {
1335                     sinfo.file_offset += sinfo.block_offset;
1336                 }
1337                 /* fall through */
1338             case PAYLOAD_BLOCK_FULLY_PRESENT:
1339                 /* if the file offset address is in the header zone,
1340                  * there is a problem */
1341                 if (sinfo.file_offset < (1024 * 1024)) {
1342                     ret = -EFAULT;
1343                     goto error_bat_restore;
1344                 }
1345 
1346                 if (!use_zero_buffers) {
1347                     qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1348                                       sinfo.bytes_avail);
1349                 }
1350                 /* block exists, so we can just overwrite it */
1351                 qemu_co_mutex_unlock(&s->lock);
1352                 ret = bdrv_co_pwritev(bs->file, sinfo.file_offset,
1353                                       sectors_to_write * BDRV_SECTOR_SIZE,
1354                                       &hd_qiov, 0);
1355                 qemu_co_mutex_lock(&s->lock);
1356                 if (ret < 0) {
1357                     goto error_bat_restore;
1358                 }
1359                 break;
1360             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1361                 /* we don't yet support difference files, fall through
1362                  * to error */
1363             default:
1364                 ret = -EIO;
1365                 goto exit;
1366                 break;
1367             }
1368 
1369             if (bat_update) {
1370                 /* this will update the BAT entry into the log journal, and
1371                  * then flush the log journal out to disk */
1372                 ret =  vhdx_log_write_and_flush(bs, s, &bat_entry,
1373                                                 sizeof(VHDXBatEntry),
1374                                                 bat_entry_offset);
1375                 if (ret < 0) {
1376                     goto exit;
1377                 }
1378             }
1379 
1380             nb_sectors -= sinfo.sectors_avail;
1381             sector_num += sinfo.sectors_avail;
1382             bytes_done += sinfo.bytes_avail;
1383 
1384         }
1385     }
1386 
1387     goto exit;
1388 
1389 error_bat_restore:
1390     if (bat_update) {
1391         /* keep metadata in sync, and restore the bat entry state
1392          * if error. */
1393         sinfo.file_offset = bat_prior_offset;
1394         vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1395                                     &bat_entry_offset, bat_state);
1396     }
1397 exit:
1398     qemu_vfree(iov1.iov_base);
1399     qemu_vfree(iov2.iov_base);
1400     qemu_co_mutex_unlock(&s->lock);
1401     qemu_iovec_destroy(&hd_qiov);
1402     return ret;
1403 }
1404 
1405 
1406 
1407 /*
1408  * Create VHDX Headers
1409  *
1410  * There are 2 headers, and the highest sequence number will represent
1411  * the active header
1412  */
1413 static int vhdx_create_new_headers(BlockBackend *blk, uint64_t image_size,
1414                                    uint32_t log_size)
1415 {
1416     BlockDriverState *bs = blk_bs(blk);
1417     BdrvChild *child;
1418     int ret = 0;
1419     VHDXHeader *hdr = NULL;
1420 
1421     hdr = g_new0(VHDXHeader, 1);
1422 
1423     hdr->signature       = VHDX_HEADER_SIGNATURE;
1424     hdr->sequence_number = g_random_int();
1425     hdr->log_version     = 0;
1426     hdr->version         = 1;
1427     hdr->log_length      = log_size;
1428     hdr->log_offset      = VHDX_HEADER_SECTION_END;
1429     vhdx_guid_generate(&hdr->file_write_guid);
1430     vhdx_guid_generate(&hdr->data_write_guid);
1431 
1432     /* XXX Ugly way to get blk->root, but that's a feature, not a bug. This
1433      * hack makes it obvious that vhdx_write_header() bypasses the BlockBackend
1434      * here, which it really shouldn't be doing. */
1435     child = QLIST_FIRST(&bs->parents);
1436     assert(!QLIST_NEXT(child, next_parent));
1437 
1438     ret = vhdx_write_header(child, hdr, VHDX_HEADER1_OFFSET, false);
1439     if (ret < 0) {
1440         goto exit;
1441     }
1442     hdr->sequence_number++;
1443     ret = vhdx_write_header(child, hdr, VHDX_HEADER2_OFFSET, false);
1444     if (ret < 0) {
1445         goto exit;
1446     }
1447 
1448 exit:
1449     g_free(hdr);
1450     return ret;
1451 }
1452 
1453 #define VHDX_METADATA_ENTRY_BUFFER_SIZE \
1454                                     (sizeof(VHDXFileParameters)               +\
1455                                      sizeof(VHDXVirtualDiskSize)              +\
1456                                      sizeof(VHDXPage83Data)                   +\
1457                                      sizeof(VHDXVirtualDiskLogicalSectorSize) +\
1458                                      sizeof(VHDXVirtualDiskPhysicalSectorSize))
1459 
1460 /*
1461  * Create the Metadata entries.
1462  *
1463  * For more details on the entries, see section 3.5 (pg 29) in the
1464  * VHDX 1.00 specification.
1465  *
1466  * We support 5 metadata entries (all required by spec):
1467  *          File Parameters,
1468  *          Virtual Disk Size,
1469  *          Page 83 Data,
1470  *          Logical Sector Size,
1471  *          Physical Sector Size
1472  *
1473  * The first 64KB of the Metadata section is reserved for the metadata
1474  * header and entries; beyond that, the metadata items themselves reside.
1475  */
1476 static int vhdx_create_new_metadata(BlockBackend *blk,
1477                                     uint64_t image_size,
1478                                     uint32_t block_size,
1479                                     uint32_t sector_size,
1480                                     uint64_t metadata_offset,
1481                                     VHDXImageType type)
1482 {
1483     int ret = 0;
1484     uint32_t offset = 0;
1485     void *buffer = NULL;
1486     void *entry_buffer;
1487     VHDXMetadataTableHeader *md_table;
1488     VHDXMetadataTableEntry  *md_table_entry;
1489 
1490     /* Metadata entries */
1491     VHDXFileParameters     *mt_file_params;
1492     VHDXVirtualDiskSize    *mt_virtual_size;
1493     VHDXPage83Data         *mt_page83;
1494     VHDXVirtualDiskLogicalSectorSize  *mt_log_sector_size;
1495     VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size;
1496 
1497     entry_buffer = g_malloc0(VHDX_METADATA_ENTRY_BUFFER_SIZE);
1498 
1499     mt_file_params = entry_buffer;
1500     offset += sizeof(VHDXFileParameters);
1501     mt_virtual_size = entry_buffer + offset;
1502     offset += sizeof(VHDXVirtualDiskSize);
1503     mt_page83 = entry_buffer + offset;
1504     offset += sizeof(VHDXPage83Data);
1505     mt_log_sector_size = entry_buffer + offset;
1506     offset += sizeof(VHDXVirtualDiskLogicalSectorSize);
1507     mt_phys_sector_size = entry_buffer + offset;
1508 
1509     mt_file_params->block_size = cpu_to_le32(block_size);
1510     if (type == VHDX_TYPE_FIXED) {
1511         mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED;
1512         cpu_to_le32s(&mt_file_params->data_bits);
1513     }
1514 
1515     vhdx_guid_generate(&mt_page83->page_83_data);
1516     cpu_to_leguids(&mt_page83->page_83_data);
1517     mt_virtual_size->virtual_disk_size        = cpu_to_le64(image_size);
1518     mt_log_sector_size->logical_sector_size   = cpu_to_le32(sector_size);
1519     mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size);
1520 
1521     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1522     md_table = buffer;
1523 
1524     md_table->signature   = VHDX_METADATA_SIGNATURE;
1525     md_table->entry_count = 5;
1526     vhdx_metadata_header_le_export(md_table);
1527 
1528 
1529     /* This will reference beyond the reserved table portion */
1530     offset = 64 * KiB;
1531 
1532     md_table_entry = buffer + sizeof(VHDXMetadataTableHeader);
1533 
1534     md_table_entry[0].item_id = file_param_guid;
1535     md_table_entry[0].offset  = offset;
1536     md_table_entry[0].length  = sizeof(VHDXFileParameters);
1537     md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED;
1538     offset += md_table_entry[0].length;
1539     vhdx_metadata_entry_le_export(&md_table_entry[0]);
1540 
1541     md_table_entry[1].item_id = virtual_size_guid;
1542     md_table_entry[1].offset  = offset;
1543     md_table_entry[1].length  = sizeof(VHDXVirtualDiskSize);
1544     md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1545                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1546     offset += md_table_entry[1].length;
1547     vhdx_metadata_entry_le_export(&md_table_entry[1]);
1548 
1549     md_table_entry[2].item_id = page83_guid;
1550     md_table_entry[2].offset  = offset;
1551     md_table_entry[2].length  = sizeof(VHDXPage83Data);
1552     md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1553                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1554     offset += md_table_entry[2].length;
1555     vhdx_metadata_entry_le_export(&md_table_entry[2]);
1556 
1557     md_table_entry[3].item_id = logical_sector_guid;
1558     md_table_entry[3].offset  = offset;
1559     md_table_entry[3].length  = sizeof(VHDXVirtualDiskLogicalSectorSize);
1560     md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1561                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1562     offset += md_table_entry[3].length;
1563     vhdx_metadata_entry_le_export(&md_table_entry[3]);
1564 
1565     md_table_entry[4].item_id = phys_sector_guid;
1566     md_table_entry[4].offset  = offset;
1567     md_table_entry[4].length  = sizeof(VHDXVirtualDiskPhysicalSectorSize);
1568     md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1569                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1570     vhdx_metadata_entry_le_export(&md_table_entry[4]);
1571 
1572     ret = blk_pwrite(blk, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE, 0);
1573     if (ret < 0) {
1574         goto exit;
1575     }
1576 
1577     ret = blk_pwrite(blk, metadata_offset + (64 * KiB), entry_buffer,
1578                      VHDX_METADATA_ENTRY_BUFFER_SIZE, 0);
1579     if (ret < 0) {
1580         goto exit;
1581     }
1582 
1583 
1584 exit:
1585     g_free(buffer);
1586     g_free(entry_buffer);
1587     return ret;
1588 }
1589 
1590 /* This create the actual BAT itself.  We currently only support
1591  * 'Dynamic' and 'Fixed' image types.
1592  *
1593  *  Dynamic images: default state of the BAT is all zeroes.
1594  *
1595  *  Fixed images: default state of the BAT is fully populated, with
1596  *                file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT.
1597  */
1598 static int vhdx_create_bat(BlockBackend *blk, BDRVVHDXState *s,
1599                            uint64_t image_size, VHDXImageType type,
1600                            bool use_zero_blocks, uint64_t file_offset,
1601                            uint32_t length, Error **errp)
1602 {
1603     int ret = 0;
1604     uint64_t data_file_offset;
1605     uint64_t total_sectors = 0;
1606     uint64_t sector_num = 0;
1607     uint64_t unused;
1608     int block_state;
1609     VHDXSectorInfo sinfo;
1610 
1611     assert(s->bat == NULL);
1612 
1613     /* this gives a data start after BAT/bitmap entries, and well
1614      * past any metadata entries (with a 4 MB buffer for future
1615      * expansion */
1616     data_file_offset = file_offset + length + 5 * MiB;
1617     total_sectors = image_size >> s->logical_sector_size_bits;
1618 
1619     if (type == VHDX_TYPE_DYNAMIC) {
1620         /* All zeroes, so we can just extend the file - the end of the BAT
1621          * is the furthest thing we have written yet */
1622         ret = blk_truncate(blk, data_file_offset, PREALLOC_MODE_OFF, errp);
1623         if (ret < 0) {
1624             goto exit;
1625         }
1626     } else if (type == VHDX_TYPE_FIXED) {
1627         ret = blk_truncate(blk, data_file_offset + image_size,
1628                            PREALLOC_MODE_OFF, errp);
1629         if (ret < 0) {
1630             goto exit;
1631         }
1632     } else {
1633         error_setg(errp, "Unsupported image type");
1634         ret = -ENOTSUP;
1635         goto exit;
1636     }
1637 
1638     if (type == VHDX_TYPE_FIXED ||
1639                 use_zero_blocks ||
1640                 bdrv_has_zero_init(blk_bs(blk)) == 0) {
1641         /* for a fixed file, the default BAT entry is not zero */
1642         s->bat = g_try_malloc0(length);
1643         if (length && s->bat == NULL) {
1644             error_setg(errp, "Failed to allocate memory for the BAT");
1645             ret = -ENOMEM;
1646             goto exit;
1647         }
1648         block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT :
1649                                                 PAYLOAD_BLOCK_NOT_PRESENT;
1650         block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state;
1651         /* fill the BAT by emulating sector writes of sectors_per_block size */
1652         while (sector_num < total_sectors) {
1653             vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo);
1654             sinfo.file_offset = data_file_offset +
1655                                 (sector_num << s->logical_sector_size_bits);
1656             sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB);
1657             vhdx_update_bat_table_entry(blk_bs(blk), s, &sinfo, &unused, &unused,
1658                                         block_state);
1659             cpu_to_le64s(&s->bat[sinfo.bat_idx]);
1660             sector_num += s->sectors_per_block;
1661         }
1662         ret = blk_pwrite(blk, file_offset, s->bat, length, 0);
1663         if (ret < 0) {
1664             error_setg_errno(errp, -ret, "Failed to write the BAT");
1665             goto exit;
1666         }
1667     }
1668 
1669 
1670 
1671 exit:
1672     g_free(s->bat);
1673     return ret;
1674 }
1675 
1676 /* Creates the region table header, and region table entries.
1677  * There are 2 supported region table entries: BAT, and Metadata/
1678  *
1679  * As the calculations for the BAT region table are also needed
1680  * to create the BAT itself, we will also cause the BAT to be
1681  * created.
1682  */
1683 static int vhdx_create_new_region_table(BlockBackend *blk,
1684                                         uint64_t image_size,
1685                                         uint32_t block_size,
1686                                         uint32_t sector_size,
1687                                         uint32_t log_size,
1688                                         bool use_zero_blocks,
1689                                         VHDXImageType type,
1690                                         uint64_t *metadata_offset,
1691                                         Error **errp)
1692 {
1693     int ret = 0;
1694     uint32_t offset = 0;
1695     void *buffer = NULL;
1696     uint64_t bat_file_offset;
1697     uint32_t bat_length;
1698     BDRVVHDXState *s = NULL;
1699     VHDXRegionTableHeader *region_table;
1700     VHDXRegionTableEntry *rt_bat;
1701     VHDXRegionTableEntry *rt_metadata;
1702 
1703     assert(metadata_offset != NULL);
1704 
1705     /* Populate enough of the BDRVVHDXState to be able to use the
1706      * pre-existing BAT calculation, translation, and update functions */
1707     s = g_new0(BDRVVHDXState, 1);
1708 
1709     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
1710                      (uint64_t) sector_size / (uint64_t) block_size;
1711 
1712     s->sectors_per_block = block_size / sector_size;
1713     s->virtual_disk_size = image_size;
1714     s->block_size = block_size;
1715     s->logical_sector_size = sector_size;
1716 
1717     vhdx_set_shift_bits(s);
1718 
1719     vhdx_calc_bat_entries(s);
1720 
1721     /* At this point the VHDX state is populated enough for creation */
1722 
1723     /* a single buffer is used so we can calculate the checksum over the
1724      * entire 64KB block */
1725     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1726     region_table = buffer;
1727     offset += sizeof(VHDXRegionTableHeader);
1728     rt_bat = buffer + offset;
1729     offset += sizeof(VHDXRegionTableEntry);
1730     rt_metadata  = buffer + offset;
1731 
1732     region_table->signature = VHDX_REGION_SIGNATURE;
1733     region_table->entry_count = 2;   /* BAT and Metadata */
1734 
1735     rt_bat->guid        = bat_guid;
1736     rt_bat->length      = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB);
1737     rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB);
1738     s->bat_offset = rt_bat->file_offset;
1739 
1740     rt_metadata->guid        = metadata_guid;
1741     rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length,
1742                                         MiB);
1743     rt_metadata->length      = 1 * MiB; /* min size, and more than enough */
1744     *metadata_offset = rt_metadata->file_offset;
1745 
1746     bat_file_offset = rt_bat->file_offset;
1747     bat_length = rt_bat->length;
1748 
1749     vhdx_region_header_le_export(region_table);
1750     vhdx_region_entry_le_export(rt_bat);
1751     vhdx_region_entry_le_export(rt_metadata);
1752 
1753     vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE,
1754                          offsetof(VHDXRegionTableHeader, checksum));
1755 
1756 
1757     /* The region table gives us the data we need to create the BAT,
1758      * so do that now */
1759     ret = vhdx_create_bat(blk, s, image_size, type, use_zero_blocks,
1760                           bat_file_offset, bat_length, errp);
1761     if (ret < 0) {
1762         goto exit;
1763     }
1764 
1765     /* Now write out the region headers to disk */
1766     ret = blk_pwrite(blk, VHDX_REGION_TABLE_OFFSET, buffer,
1767                      VHDX_HEADER_BLOCK_SIZE, 0);
1768     if (ret < 0) {
1769         error_setg_errno(errp, -ret, "Failed to write first region table");
1770         goto exit;
1771     }
1772 
1773     ret = blk_pwrite(blk, VHDX_REGION_TABLE2_OFFSET, buffer,
1774                      VHDX_HEADER_BLOCK_SIZE, 0);
1775     if (ret < 0) {
1776         error_setg_errno(errp, -ret, "Failed to write second region table");
1777         goto exit;
1778     }
1779 
1780 exit:
1781     g_free(s);
1782     g_free(buffer);
1783     return ret;
1784 }
1785 
1786 /* We need to create the following elements:
1787  *
1788  *    .-----------------------------------------------------------------.
1789  *    |   (A)    |   (B)    |    (C)    |     (D)       |     (E)       |
1790  *    |  File ID |  Header1 |  Header 2 |  Region Tbl 1 |  Region Tbl 2 |
1791  *    |          |          |           |               |               |
1792  *    .-----------------------------------------------------------------.
1793  *    0         64KB      128KB       192KB           256KB           320KB
1794  *
1795  *
1796  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1797  *    |     (F)     |     (G)       |    (H)    |                        |
1798  *    | Journal Log |  BAT / Bitmap |  Metadata |  .... data ......      |
1799  *    |             |               |           |                        |
1800  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1801  *   1MB
1802  */
1803 static int coroutine_fn vhdx_co_create(BlockdevCreateOptions *opts,
1804                                        Error **errp)
1805 {
1806     BlockdevCreateOptionsVhdx *vhdx_opts;
1807     BlockBackend *blk = NULL;
1808     BlockDriverState *bs = NULL;
1809 
1810     int ret = 0;
1811     uint64_t image_size;
1812     uint32_t log_size;
1813     uint32_t block_size;
1814     uint64_t signature;
1815     uint64_t metadata_offset;
1816     bool use_zero_blocks = false;
1817 
1818     gunichar2 *creator = NULL;
1819     glong creator_items;
1820     VHDXImageType image_type;
1821 
1822     assert(opts->driver == BLOCKDEV_DRIVER_VHDX);
1823     vhdx_opts = &opts->u.vhdx;
1824 
1825     /* Validate options and set default values */
1826     image_size = vhdx_opts->size;
1827     if (image_size > VHDX_MAX_IMAGE_SIZE) {
1828         error_setg(errp, "Image size too large; max of 64TB");
1829         return -EINVAL;
1830     }
1831 
1832     if (!vhdx_opts->has_log_size) {
1833         log_size = DEFAULT_LOG_SIZE;
1834     } else {
1835         if (vhdx_opts->log_size > UINT32_MAX) {
1836             error_setg(errp, "Log size must be smaller than 4 GB");
1837             return -EINVAL;
1838         }
1839         log_size = vhdx_opts->log_size;
1840     }
1841     if (log_size < MiB || (log_size % MiB) != 0) {
1842         error_setg(errp, "Log size must be a multiple of 1 MB");
1843         return -EINVAL;
1844     }
1845 
1846     if (!vhdx_opts->has_block_state_zero) {
1847         use_zero_blocks = true;
1848     } else {
1849         use_zero_blocks = vhdx_opts->block_state_zero;
1850     }
1851 
1852     if (!vhdx_opts->has_subformat) {
1853         vhdx_opts->subformat = BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC;
1854     }
1855 
1856     switch (vhdx_opts->subformat) {
1857     case BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC:
1858         image_type = VHDX_TYPE_DYNAMIC;
1859         break;
1860     case BLOCKDEV_VHDX_SUBFORMAT_FIXED:
1861         image_type = VHDX_TYPE_FIXED;
1862         break;
1863     default:
1864         g_assert_not_reached();
1865     }
1866 
1867     /* These are pretty arbitrary, and mainly designed to keep the BAT
1868      * size reasonable to load into RAM */
1869     if (vhdx_opts->has_block_size) {
1870         block_size = vhdx_opts->block_size;
1871     } else {
1872         if (image_size > 32 * TiB) {
1873             block_size = 64 * MiB;
1874         } else if (image_size > (uint64_t) 100 * GiB) {
1875             block_size = 32 * MiB;
1876         } else if (image_size > 1 * GiB) {
1877             block_size = 16 * MiB;
1878         } else {
1879             block_size = 8 * MiB;
1880         }
1881     }
1882 
1883     if (block_size < MiB || (block_size % MiB) != 0) {
1884         error_setg(errp, "Block size must be a multiple of 1 MB");
1885         return -EINVAL;
1886     }
1887     if (!is_power_of_2(block_size)) {
1888         error_setg(errp, "Block size must be a power of two");
1889         return -EINVAL;
1890     }
1891     if (block_size > VHDX_BLOCK_SIZE_MAX) {
1892         error_setg(errp, "Block size must not exceed %d", VHDX_BLOCK_SIZE_MAX);
1893         return -EINVAL;
1894     }
1895 
1896     /* Create BlockBackend to write to the image */
1897     bs = bdrv_open_blockdev_ref(vhdx_opts->file, errp);
1898     if (bs == NULL) {
1899         return -EIO;
1900     }
1901 
1902     blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
1903     ret = blk_insert_bs(blk, bs, errp);
1904     if (ret < 0) {
1905         goto delete_and_exit;
1906     }
1907     blk_set_allow_write_beyond_eof(blk, true);
1908 
1909     /* Create (A) */
1910 
1911     /* The creator field is optional, but may be useful for
1912      * debugging / diagnostics */
1913     creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL,
1914                               &creator_items, NULL);
1915     signature = cpu_to_le64(VHDX_FILE_SIGNATURE);
1916     ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature),
1917                      0);
1918     if (ret < 0) {
1919         error_setg_errno(errp, -ret, "Failed to write file signature");
1920         goto delete_and_exit;
1921     }
1922     if (creator) {
1923         ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET + sizeof(signature),
1924                          creator, creator_items * sizeof(gunichar2), 0);
1925         if (ret < 0) {
1926             error_setg_errno(errp, -ret, "Failed to write creator field");
1927             goto delete_and_exit;
1928         }
1929     }
1930 
1931 
1932     /* Creates (B),(C) */
1933     ret = vhdx_create_new_headers(blk, image_size, log_size);
1934     if (ret < 0) {
1935         error_setg_errno(errp, -ret, "Failed to write image headers");
1936         goto delete_and_exit;
1937     }
1938 
1939     /* Creates (D),(E),(G) explicitly. (F) created as by-product */
1940     ret = vhdx_create_new_region_table(blk, image_size, block_size, 512,
1941                                        log_size, use_zero_blocks, image_type,
1942                                        &metadata_offset, errp);
1943     if (ret < 0) {
1944         goto delete_and_exit;
1945     }
1946 
1947     /* Creates (H) */
1948     ret = vhdx_create_new_metadata(blk, image_size, block_size, 512,
1949                                    metadata_offset, image_type);
1950     if (ret < 0) {
1951         error_setg_errno(errp, -ret, "Failed to initialize metadata");
1952         goto delete_and_exit;
1953     }
1954 
1955     ret = 0;
1956 delete_and_exit:
1957     blk_unref(blk);
1958     bdrv_unref(bs);
1959     g_free(creator);
1960     return ret;
1961 }
1962 
1963 static int coroutine_fn vhdx_co_create_opts(const char *filename,
1964                                             QemuOpts *opts,
1965                                             Error **errp)
1966 {
1967     BlockdevCreateOptions *create_options = NULL;
1968     QDict *qdict;
1969     Visitor *v;
1970     BlockDriverState *bs = NULL;
1971     Error *local_err = NULL;
1972     int ret;
1973 
1974     static const QDictRenames opt_renames[] = {
1975         { VHDX_BLOCK_OPT_LOG_SIZE,      "log-size" },
1976         { VHDX_BLOCK_OPT_BLOCK_SIZE,    "block-size" },
1977         { VHDX_BLOCK_OPT_ZERO,          "block-state-zero" },
1978         { NULL, NULL },
1979     };
1980 
1981     /* Parse options and convert legacy syntax */
1982     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vhdx_create_opts, true);
1983 
1984     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
1985         ret = -EINVAL;
1986         goto fail;
1987     }
1988 
1989     /* Create and open the file (protocol layer) */
1990     ret = bdrv_create_file(filename, opts, &local_err);
1991     if (ret < 0) {
1992         error_propagate(errp, local_err);
1993         goto fail;
1994     }
1995 
1996     bs = bdrv_open(filename, NULL, NULL,
1997                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
1998     if (bs == NULL) {
1999         ret = -EIO;
2000         goto fail;
2001     }
2002 
2003     /* Now get the QAPI type BlockdevCreateOptions */
2004     qdict_put_str(qdict, "driver", "vhdx");
2005     qdict_put_str(qdict, "file", bs->node_name);
2006 
2007     v = qobject_input_visitor_new_flat_confused(qdict, errp);
2008     if (!v) {
2009         ret = -EINVAL;
2010         goto fail;
2011     }
2012 
2013     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
2014     visit_free(v);
2015 
2016     if (local_err) {
2017         error_propagate(errp, local_err);
2018         ret = -EINVAL;
2019         goto fail;
2020     }
2021 
2022     /* Silently round up sizes:
2023      * The image size is rounded to 512 bytes. Make the block and log size
2024      * close to what was specified, but must be at least 1MB, and a multiple of
2025      * 1 MB. Also respect VHDX_BLOCK_SIZE_MAX for block sizes. block_size = 0
2026      * means auto, which is represented by a missing key in QAPI. */
2027     assert(create_options->driver == BLOCKDEV_DRIVER_VHDX);
2028     create_options->u.vhdx.size =
2029         ROUND_UP(create_options->u.vhdx.size, BDRV_SECTOR_SIZE);
2030 
2031     if (create_options->u.vhdx.has_log_size) {
2032         create_options->u.vhdx.log_size =
2033             ROUND_UP(create_options->u.vhdx.log_size, MiB);
2034     }
2035     if (create_options->u.vhdx.has_block_size) {
2036         create_options->u.vhdx.block_size =
2037             ROUND_UP(create_options->u.vhdx.block_size, MiB);
2038 
2039         if (create_options->u.vhdx.block_size == 0) {
2040             create_options->u.vhdx.has_block_size = false;
2041         }
2042         if (create_options->u.vhdx.block_size > VHDX_BLOCK_SIZE_MAX) {
2043             create_options->u.vhdx.block_size = VHDX_BLOCK_SIZE_MAX;
2044         }
2045     }
2046 
2047     /* Create the vhdx image (format layer) */
2048     ret = vhdx_co_create(create_options, errp);
2049 
2050 fail:
2051     qobject_unref(qdict);
2052     bdrv_unref(bs);
2053     qapi_free_BlockdevCreateOptions(create_options);
2054     return ret;
2055 }
2056 
2057 /* If opened r/w, the VHDX driver will automatically replay the log,
2058  * if one is present, inside the vhdx_open() call.
2059  *
2060  * If qemu-img check -r all is called, the image is automatically opened
2061  * r/w and any log has already been replayed, so there is nothing (currently)
2062  * for us to do here
2063  */
2064 static int coroutine_fn vhdx_co_check(BlockDriverState *bs,
2065                                       BdrvCheckResult *result,
2066                                       BdrvCheckMode fix)
2067 {
2068     BDRVVHDXState *s = bs->opaque;
2069 
2070     if (s->log_replayed_on_open) {
2071         result->corruptions_fixed++;
2072     }
2073     return 0;
2074 }
2075 
2076 static QemuOptsList vhdx_create_opts = {
2077     .name = "vhdx-create-opts",
2078     .head = QTAILQ_HEAD_INITIALIZER(vhdx_create_opts.head),
2079     .desc = {
2080         {
2081            .name = BLOCK_OPT_SIZE,
2082            .type = QEMU_OPT_SIZE,
2083            .help = "Virtual disk size; max of 64TB."
2084        },
2085        {
2086            .name = VHDX_BLOCK_OPT_LOG_SIZE,
2087            .type = QEMU_OPT_SIZE,
2088            .def_value_str = stringify(DEFAULT_LOG_SIZE),
2089            .help = "Log size; min 1MB."
2090        },
2091        {
2092            .name = VHDX_BLOCK_OPT_BLOCK_SIZE,
2093            .type = QEMU_OPT_SIZE,
2094            .def_value_str = stringify(0),
2095            .help = "Block Size; min 1MB, max 256MB. " \
2096                    "0 means auto-calculate based on image size."
2097        },
2098        {
2099            .name = BLOCK_OPT_SUBFMT,
2100            .type = QEMU_OPT_STRING,
2101            .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "\
2102                    "Default is 'dynamic'."
2103        },
2104        {
2105            .name = VHDX_BLOCK_OPT_ZERO,
2106            .type = QEMU_OPT_BOOL,
2107            .help = "Force use of payload blocks of type 'ZERO'. "\
2108                    "Non-standard, but default.  Do not set to 'off' when "\
2109                    "using 'qemu-img convert' with subformat=dynamic."
2110        },
2111        { NULL }
2112     }
2113 };
2114 
2115 static BlockDriver bdrv_vhdx = {
2116     .format_name            = "vhdx",
2117     .instance_size          = sizeof(BDRVVHDXState),
2118     .bdrv_probe             = vhdx_probe,
2119     .bdrv_open              = vhdx_open,
2120     .bdrv_close             = vhdx_close,
2121     .bdrv_reopen_prepare    = vhdx_reopen_prepare,
2122     .bdrv_child_perm        = bdrv_format_default_perms,
2123     .bdrv_co_readv          = vhdx_co_readv,
2124     .bdrv_co_writev         = vhdx_co_writev,
2125     .bdrv_co_create         = vhdx_co_create,
2126     .bdrv_co_create_opts    = vhdx_co_create_opts,
2127     .bdrv_get_info          = vhdx_get_info,
2128     .bdrv_co_check          = vhdx_co_check,
2129     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
2130 
2131     .create_opts            = &vhdx_create_opts,
2132 };
2133 
2134 static void bdrv_vhdx_init(void)
2135 {
2136     bdrv_register(&bdrv_vhdx);
2137 }
2138 
2139 block_init(bdrv_vhdx_init);
2140