xref: /openbmc/qemu/block/vhdx.c (revision 2a886794)
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 "qemu/error-report.h"
28 #include "vhdx.h"
29 #include "migration/blocker.h"
30 #include "qemu/uuid.h"
31 #include "qapi/qmp/qdict.h"
32 #include "qapi/qobject-input-visitor.h"
33 #include "qapi/qapi-visit-block-core.h"
34 
35 /* Options for VHDX creation */
36 
37 #define VHDX_BLOCK_OPT_LOG_SIZE   "log_size"
38 #define VHDX_BLOCK_OPT_BLOCK_SIZE "block_size"
39 #define VHDX_BLOCK_OPT_ZERO "block_state_zero"
40 
41 typedef enum VHDXImageType {
42     VHDX_TYPE_DYNAMIC = 0,
43     VHDX_TYPE_FIXED,
44     VHDX_TYPE_DIFFERENCING,   /* Currently unsupported */
45 } VHDXImageType;
46 
47 static QemuOptsList vhdx_create_opts;
48 
49 /* Several metadata and region table data entries are identified by
50  * guids in  a MS-specific GUID format. */
51 
52 
53 /* ------- Known Region Table GUIDs ---------------------- */
54 static const MSGUID bat_guid =      { .data1 = 0x2dc27766,
55                                       .data2 = 0xf623,
56                                       .data3 = 0x4200,
57                                       .data4 = { 0x9d, 0x64, 0x11, 0x5e,
58                                                  0x9b, 0xfd, 0x4a, 0x08} };
59 
60 static const MSGUID metadata_guid = { .data1 = 0x8b7ca206,
61                                       .data2 = 0x4790,
62                                       .data3 = 0x4b9a,
63                                       .data4 = { 0xb8, 0xfe, 0x57, 0x5f,
64                                                  0x05, 0x0f, 0x88, 0x6e} };
65 
66 
67 
68 /* ------- Known Metadata Entry GUIDs ---------------------- */
69 static const MSGUID file_param_guid =   { .data1 = 0xcaa16737,
70                                           .data2 = 0xfa36,
71                                           .data3 = 0x4d43,
72                                           .data4 = { 0xb3, 0xb6, 0x33, 0xf0,
73                                                      0xaa, 0x44, 0xe7, 0x6b} };
74 
75 static const MSGUID virtual_size_guid = { .data1 = 0x2FA54224,
76                                           .data2 = 0xcd1b,
77                                           .data3 = 0x4876,
78                                           .data4 = { 0xb2, 0x11, 0x5d, 0xbe,
79                                                      0xd8, 0x3b, 0xf4, 0xb8} };
80 
81 static const MSGUID page83_guid =       { .data1 = 0xbeca12ab,
82                                           .data2 = 0xb2e6,
83                                           .data3 = 0x4523,
84                                           .data4 = { 0x93, 0xef, 0xc3, 0x09,
85                                                      0xe0, 0x00, 0xc7, 0x46} };
86 
87 
88 static const MSGUID phys_sector_guid =  { .data1 = 0xcda348c7,
89                                           .data2 = 0x445d,
90                                           .data3 = 0x4471,
91                                           .data4 = { 0x9c, 0xc9, 0xe9, 0x88,
92                                                      0x52, 0x51, 0xc5, 0x56} };
93 
94 static const MSGUID parent_locator_guid = { .data1 = 0xa8d35f2d,
95                                             .data2 = 0xb30b,
96                                             .data3 = 0x454d,
97                                             .data4 = { 0xab, 0xf7, 0xd3,
98                                                        0xd8, 0x48, 0x34,
99                                                        0xab, 0x0c} };
100 
101 static const MSGUID logical_sector_guid = { .data1 = 0x8141bf1d,
102                                             .data2 = 0xa96f,
103                                             .data3 = 0x4709,
104                                             .data4 = { 0xba, 0x47, 0xf2,
105                                                        0x33, 0xa8, 0xfa,
106                                                        0xab, 0x5f} };
107 
108 /* Each parent type must have a valid GUID; this is for parent images
109  * of type 'VHDX'.  If we were to allow e.g. a QCOW2 parent, we would
110  * need to make up our own QCOW2 GUID type */
111 static const MSGUID parent_vhdx_guid __attribute__((unused))
112                                      = { .data1 = 0xb04aefb7,
113                                          .data2 = 0xd19e,
114                                          .data3 = 0x4a81,
115                                          .data4 = { 0xb7, 0x89, 0x25, 0xb8,
116                                                     0xe9, 0x44, 0x59, 0x13} };
117 
118 
119 #define META_FILE_PARAMETER_PRESENT      0x01
120 #define META_VIRTUAL_DISK_SIZE_PRESENT   0x02
121 #define META_PAGE_83_PRESENT             0x04
122 #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08
123 #define META_PHYS_SECTOR_SIZE_PRESENT    0x10
124 #define META_PARENT_LOCATOR_PRESENT      0x20
125 
126 #define META_ALL_PRESENT    \
127     (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \
128      META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \
129      META_PHYS_SECTOR_SIZE_PRESENT)
130 
131 
132 typedef struct VHDXSectorInfo {
133     uint32_t bat_idx;       /* BAT entry index */
134     uint32_t sectors_avail; /* sectors available in payload block */
135     uint32_t bytes_left;    /* bytes left in the block after data to r/w */
136     uint32_t bytes_avail;   /* bytes available in payload block */
137     uint64_t file_offset;   /* absolute offset in bytes, in file */
138     uint64_t block_offset;  /* block offset, in bytes */
139 } VHDXSectorInfo;
140 
141 /* Calculates new checksum.
142  *
143  * Zero is substituted during crc calculation for the original crc field
144  * crc_offset: byte offset in buf of the buffer crc
145  * buf: buffer pointer
146  * size: size of buffer (must be > crc_offset+4)
147  *
148  * Note: The buffer should have all multi-byte data in little-endian format,
149  *       and the resulting checksum is in little endian format.
150  */
151 uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset)
152 {
153     uint32_t crc;
154 
155     assert(buf != NULL);
156     assert(size > (crc_offset + sizeof(crc)));
157 
158     memset(buf + crc_offset, 0, sizeof(crc));
159     crc =  crc32c(0xffffffff, buf, size);
160     crc = cpu_to_le32(crc);
161     memcpy(buf + crc_offset, &crc, sizeof(crc));
162 
163     return crc;
164 }
165 
166 uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size,
167                             int crc_offset)
168 {
169     uint32_t crc_new;
170     uint32_t crc_orig;
171     assert(buf != NULL);
172 
173     if (crc_offset > 0) {
174         memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
175         memset(buf + crc_offset, 0, sizeof(crc_orig));
176     }
177 
178     crc_new = crc32c(crc, buf, size);
179     if (crc_offset > 0) {
180         memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig));
181     }
182 
183     return crc_new;
184 }
185 
186 /* Validates the checksum of the buffer, with an in-place CRC.
187  *
188  * Zero is substituted during crc calculation for the original crc field,
189  * and the crc field is restored afterwards.  But the buffer will be modified
190  * during the calculation, so this may not be not suitable for multi-threaded
191  * use.
192  *
193  * crc_offset: byte offset in buf of the buffer crc
194  * buf: buffer pointer
195  * size: size of buffer (must be > crc_offset+4)
196  *
197  * returns true if checksum is valid, false otherwise
198  */
199 bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset)
200 {
201     uint32_t crc_orig;
202     uint32_t crc;
203 
204     assert(buf != NULL);
205     assert(size > (crc_offset + 4));
206 
207     memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
208     crc_orig = le32_to_cpu(crc_orig);
209 
210     crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset);
211 
212     return crc == crc_orig;
213 }
214 
215 
216 /*
217  * This generates a UUID that is compliant with the MS GUIDs used
218  * in the VHDX spec (and elsewhere).
219  */
220 void vhdx_guid_generate(MSGUID *guid)
221 {
222     QemuUUID uuid;
223     assert(guid != NULL);
224 
225     qemu_uuid_generate(&uuid);
226     memcpy(guid, &uuid, sizeof(MSGUID));
227 }
228 
229 /* Check for region overlaps inside the VHDX image */
230 static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length)
231 {
232     int ret = 0;
233     uint64_t end;
234     VHDXRegionEntry *r;
235 
236     end = start + length;
237     QLIST_FOREACH(r, &s->regions, entries) {
238         if (!((start >= r->end) || (end <= r->start))) {
239             error_report("VHDX region %" PRIu64 "-%" PRIu64 " overlaps with "
240                          "region %" PRIu64 "-%." PRIu64, start, end, r->start,
241                          r->end);
242             ret = -EINVAL;
243             goto exit;
244         }
245     }
246 
247 exit:
248     return ret;
249 }
250 
251 /* Register a region for future checks */
252 static void vhdx_region_register(BDRVVHDXState *s,
253                                  uint64_t start, uint64_t length)
254 {
255     VHDXRegionEntry *r;
256 
257     r = g_malloc0(sizeof(*r));
258 
259     r->start = start;
260     r->end = start + length;
261 
262     QLIST_INSERT_HEAD(&s->regions, r, entries);
263 }
264 
265 /* Free all registered regions */
266 static void vhdx_region_unregister_all(BDRVVHDXState *s)
267 {
268     VHDXRegionEntry *r, *r_next;
269 
270     QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) {
271         QLIST_REMOVE(r, entries);
272         g_free(r);
273     }
274 }
275 
276 static void vhdx_set_shift_bits(BDRVVHDXState *s)
277 {
278     s->logical_sector_size_bits = ctz32(s->logical_sector_size);
279     s->sectors_per_block_bits =   ctz32(s->sectors_per_block);
280     s->chunk_ratio_bits =         ctz64(s->chunk_ratio);
281     s->block_size_bits =          ctz32(s->block_size);
282 }
283 
284 /*
285  * Per the MS VHDX Specification, for every VHDX file:
286  *      - The header section is fixed size - 1 MB
287  *      - The header section is always the first "object"
288  *      - The first 64KB of the header is the File Identifier
289  *      - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile")
290  *      - The following 512 bytes constitute a UTF-16 string identifiying the
291  *        software that created the file, and is optional and diagnostic only.
292  *
293  *  Therefore, we probe by looking for the vhdxfile signature "vhdxfile"
294  */
295 static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename)
296 {
297     if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) {
298         return 100;
299     }
300     return 0;
301 }
302 
303 /*
304  * Writes the header to the specified offset.
305  *
306  * This will optionally read in buffer data from disk (otherwise zero-fill),
307  * and then update the header checksum.  Header is converted to proper
308  * endianness before being written to the specified file offset
309  */
310 static int vhdx_write_header(BdrvChild *file, VHDXHeader *hdr,
311                              uint64_t offset, bool read)
312 {
313     BlockDriverState *bs_file = file->bs;
314     uint8_t *buffer = NULL;
315     int ret;
316     VHDXHeader *header_le;
317 
318     assert(bs_file != NULL);
319     assert(hdr != NULL);
320 
321     /* the header checksum is not over just the packed size of VHDXHeader,
322      * but rather over the entire 'reserved' range for the header, which is
323      * 4KB (VHDX_HEADER_SIZE). */
324 
325     buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE);
326     if (read) {
327         /* if true, we can't assume the extra reserved bytes are 0 */
328         ret = bdrv_pread(file, offset, buffer, VHDX_HEADER_SIZE);
329         if (ret < 0) {
330             goto exit;
331         }
332     } else {
333         memset(buffer, 0, VHDX_HEADER_SIZE);
334     }
335 
336     /* overwrite the actual VHDXHeader portion */
337     header_le = (VHDXHeader *)buffer;
338     memcpy(header_le, hdr, sizeof(VHDXHeader));
339     vhdx_header_le_export(hdr, header_le);
340     vhdx_update_checksum(buffer, VHDX_HEADER_SIZE,
341                          offsetof(VHDXHeader, checksum));
342     ret = bdrv_pwrite_sync(file, offset, header_le, sizeof(VHDXHeader));
343 
344 exit:
345     qemu_vfree(buffer);
346     return ret;
347 }
348 
349 /* Update the VHDX headers
350  *
351  * This follows the VHDX spec procedures for header updates.
352  *
353  *  - non-current header is updated with largest sequence number
354  */
355 static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s,
356                               bool generate_data_write_guid, MSGUID *log_guid)
357 {
358     int ret = 0;
359     int hdr_idx = 0;
360     uint64_t header_offset = VHDX_HEADER1_OFFSET;
361 
362     VHDXHeader *active_header;
363     VHDXHeader *inactive_header;
364 
365     /* operate on the non-current header */
366     if (s->curr_header == 0) {
367         hdr_idx = 1;
368         header_offset = VHDX_HEADER2_OFFSET;
369     }
370 
371     active_header   = s->headers[s->curr_header];
372     inactive_header = s->headers[hdr_idx];
373 
374     inactive_header->sequence_number = active_header->sequence_number + 1;
375 
376     /* a new file guid must be generated before any file write, including
377      * headers */
378     inactive_header->file_write_guid = s->session_guid;
379 
380     /* a new data guid only needs to be generated before any guest-visible
381      * writes (i.e. something observable via virtual disk read) */
382     if (generate_data_write_guid) {
383         vhdx_guid_generate(&inactive_header->data_write_guid);
384     }
385 
386     /* update the log guid if present */
387     if (log_guid) {
388         inactive_header->log_guid = *log_guid;
389     }
390 
391     ret = vhdx_write_header(bs->file, inactive_header, header_offset, true);
392     if (ret < 0) {
393         goto exit;
394     }
395     s->curr_header = hdr_idx;
396 
397 exit:
398     return ret;
399 }
400 
401 /*
402  * The VHDX spec calls for header updates to be performed twice, so that both
403  * the current and non-current header have valid info
404  */
405 int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s,
406                         bool generate_data_write_guid, MSGUID *log_guid)
407 {
408     int ret;
409 
410     ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
411     if (ret < 0) {
412         return ret;
413     }
414     ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
415     return ret;
416 }
417 
418 /* opens the specified header block from the VHDX file header section */
419 static void vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s,
420                               Error **errp)
421 {
422     int ret;
423     VHDXHeader *header1;
424     VHDXHeader *header2;
425     bool h1_valid = false;
426     bool h2_valid = false;
427     uint64_t h1_seq = 0;
428     uint64_t h2_seq = 0;
429     uint8_t *buffer;
430 
431     /* header1 & header2 are freed in vhdx_close() */
432     header1 = qemu_blockalign(bs, sizeof(VHDXHeader));
433     header2 = qemu_blockalign(bs, sizeof(VHDXHeader));
434 
435     buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE);
436 
437     s->headers[0] = header1;
438     s->headers[1] = header2;
439 
440     /* We have to read the whole VHDX_HEADER_SIZE instead of
441      * sizeof(VHDXHeader), because the checksum is over the whole
442      * region */
443     ret = bdrv_pread(bs->file, VHDX_HEADER1_OFFSET, buffer,
444                      VHDX_HEADER_SIZE);
445     if (ret < 0) {
446         goto fail;
447     }
448     /* copy over just the relevant portion that we need */
449     memcpy(header1, buffer, sizeof(VHDXHeader));
450 
451     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
452         vhdx_header_le_import(header1);
453         if (header1->signature == VHDX_HEADER_SIGNATURE &&
454             header1->version == 1) {
455             h1_seq = header1->sequence_number;
456             h1_valid = true;
457         }
458     }
459 
460     ret = bdrv_pread(bs->file, VHDX_HEADER2_OFFSET, buffer,
461                      VHDX_HEADER_SIZE);
462     if (ret < 0) {
463         goto fail;
464     }
465     /* copy over just the relevant portion that we need */
466     memcpy(header2, buffer, sizeof(VHDXHeader));
467 
468     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
469         vhdx_header_le_import(header2);
470         if (header2->signature == VHDX_HEADER_SIGNATURE &&
471             header2->version == 1) {
472             h2_seq = header2->sequence_number;
473             h2_valid = true;
474         }
475     }
476 
477     /* If there is only 1 valid header (or no valid headers), we
478      * don't care what the sequence numbers are */
479     if (h1_valid && !h2_valid) {
480         s->curr_header = 0;
481     } else if (!h1_valid && h2_valid) {
482         s->curr_header = 1;
483     } else if (!h1_valid && !h2_valid) {
484         goto fail;
485     } else {
486         /* If both headers are valid, then we choose the active one by the
487          * highest sequence number.  If the sequence numbers are equal, that is
488          * invalid */
489         if (h1_seq > h2_seq) {
490             s->curr_header = 0;
491         } else if (h2_seq > h1_seq) {
492             s->curr_header = 1;
493         } else {
494             /* The Microsoft Disk2VHD tool will create 2 identical
495              * headers, with identical sequence numbers.  If the headers are
496              * identical, don't consider the file corrupt */
497             if (!memcmp(header1, header2, sizeof(VHDXHeader))) {
498                 s->curr_header = 0;
499             } else {
500                 goto fail;
501             }
502         }
503     }
504 
505     vhdx_region_register(s, s->headers[s->curr_header]->log_offset,
506                             s->headers[s->curr_header]->log_length);
507     goto exit;
508 
509 fail:
510     error_setg_errno(errp, -ret, "No valid VHDX header found");
511     qemu_vfree(header1);
512     qemu_vfree(header2);
513     s->headers[0] = NULL;
514     s->headers[1] = NULL;
515 exit:
516     qemu_vfree(buffer);
517 }
518 
519 
520 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s)
521 {
522     int ret = 0;
523     uint8_t *buffer;
524     int offset = 0;
525     VHDXRegionTableEntry rt_entry;
526     uint32_t i;
527     bool bat_rt_found = false;
528     bool metadata_rt_found = false;
529 
530     /* We have to read the whole 64KB block, because the crc32 is over the
531      * whole block */
532     buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE);
533 
534     ret = bdrv_pread(bs->file, VHDX_REGION_TABLE_OFFSET, buffer,
535                      VHDX_HEADER_BLOCK_SIZE);
536     if (ret < 0) {
537         goto fail;
538     }
539     memcpy(&s->rt, buffer, sizeof(s->rt));
540     offset += sizeof(s->rt);
541 
542     if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4)) {
543         ret = -EINVAL;
544         goto fail;
545     }
546 
547     vhdx_region_header_le_import(&s->rt);
548 
549     if (s->rt.signature != VHDX_REGION_SIGNATURE) {
550         ret = -EINVAL;
551         goto fail;
552     }
553 
554 
555     /* Per spec, maximum region table entry count is 2047 */
556     if (s->rt.entry_count > 2047) {
557         ret = -EINVAL;
558         goto fail;
559     }
560 
561     for (i = 0; i < s->rt.entry_count; i++) {
562         memcpy(&rt_entry, buffer + offset, sizeof(rt_entry));
563         offset += sizeof(rt_entry);
564 
565         vhdx_region_entry_le_import(&rt_entry);
566 
567         /* check for region overlap between these entries, and any
568          * other memory regions in the file */
569         ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length);
570         if (ret < 0) {
571             goto fail;
572         }
573 
574         vhdx_region_register(s, rt_entry.file_offset, rt_entry.length);
575 
576         /* see if we recognize the entry */
577         if (guid_eq(rt_entry.guid, bat_guid)) {
578             /* must be unique; if we have already found it this is invalid */
579             if (bat_rt_found) {
580                 ret = -EINVAL;
581                 goto fail;
582             }
583             bat_rt_found = true;
584             s->bat_rt = rt_entry;
585             continue;
586         }
587 
588         if (guid_eq(rt_entry.guid, metadata_guid)) {
589             /* must be unique; if we have already found it this is invalid */
590             if (metadata_rt_found) {
591                 ret = -EINVAL;
592                 goto fail;
593             }
594             metadata_rt_found = true;
595             s->metadata_rt = rt_entry;
596             continue;
597         }
598 
599         if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) {
600             /* cannot read vhdx file - required region table entry that
601              * we do not understand.  per spec, we must fail to open */
602             ret = -ENOTSUP;
603             goto fail;
604         }
605     }
606 
607     if (!bat_rt_found || !metadata_rt_found) {
608         ret = -EINVAL;
609         goto fail;
610     }
611 
612     ret = 0;
613 
614 fail:
615     qemu_vfree(buffer);
616     return ret;
617 }
618 
619 
620 
621 /* Metadata initial parser
622  *
623  * This loads all the metadata entry fields.  This may cause additional
624  * fields to be processed (e.g. parent locator, etc..).
625  *
626  * There are 5 Metadata items that are always required:
627  *      - File Parameters (block size, has a parent)
628  *      - Virtual Disk Size (size, in bytes, of the virtual drive)
629  *      - Page 83 Data (scsi page 83 guid)
630  *      - Logical Sector Size (logical sector size in bytes, either 512 or
631  *                             4096.  We only support 512 currently)
632  *      - Physical Sector Size (512 or 4096)
633  *
634  * Also, if the File Parameters indicate this is a differencing file,
635  * we must also look for the Parent Locator metadata item.
636  */
637 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s)
638 {
639     int ret = 0;
640     uint8_t *buffer;
641     int offset = 0;
642     uint32_t i = 0;
643     VHDXMetadataTableEntry md_entry;
644 
645     buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE);
646 
647     ret = bdrv_pread(bs->file, s->metadata_rt.file_offset, buffer,
648                      VHDX_METADATA_TABLE_MAX_SIZE);
649     if (ret < 0) {
650         goto exit;
651     }
652     memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr));
653     offset += sizeof(s->metadata_hdr);
654 
655     vhdx_metadata_header_le_import(&s->metadata_hdr);
656 
657     if (s->metadata_hdr.signature != VHDX_METADATA_SIGNATURE) {
658         ret = -EINVAL;
659         goto exit;
660     }
661 
662     s->metadata_entries.present = 0;
663 
664     if ((s->metadata_hdr.entry_count * sizeof(md_entry)) >
665         (VHDX_METADATA_TABLE_MAX_SIZE - offset)) {
666         ret = -EINVAL;
667         goto exit;
668     }
669 
670     for (i = 0; i < s->metadata_hdr.entry_count; i++) {
671         memcpy(&md_entry, buffer + offset, sizeof(md_entry));
672         offset += sizeof(md_entry);
673 
674         vhdx_metadata_entry_le_import(&md_entry);
675 
676         if (guid_eq(md_entry.item_id, file_param_guid)) {
677             if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) {
678                 ret = -EINVAL;
679                 goto exit;
680             }
681             s->metadata_entries.file_parameters_entry = md_entry;
682             s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT;
683             continue;
684         }
685 
686         if (guid_eq(md_entry.item_id, virtual_size_guid)) {
687             if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) {
688                 ret = -EINVAL;
689                 goto exit;
690             }
691             s->metadata_entries.virtual_disk_size_entry = md_entry;
692             s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT;
693             continue;
694         }
695 
696         if (guid_eq(md_entry.item_id, page83_guid)) {
697             if (s->metadata_entries.present & META_PAGE_83_PRESENT) {
698                 ret = -EINVAL;
699                 goto exit;
700             }
701             s->metadata_entries.page83_data_entry = md_entry;
702             s->metadata_entries.present |= META_PAGE_83_PRESENT;
703             continue;
704         }
705 
706         if (guid_eq(md_entry.item_id, logical_sector_guid)) {
707             if (s->metadata_entries.present &
708                 META_LOGICAL_SECTOR_SIZE_PRESENT) {
709                 ret = -EINVAL;
710                 goto exit;
711             }
712             s->metadata_entries.logical_sector_size_entry = md_entry;
713             s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT;
714             continue;
715         }
716 
717         if (guid_eq(md_entry.item_id, phys_sector_guid)) {
718             if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) {
719                 ret = -EINVAL;
720                 goto exit;
721             }
722             s->metadata_entries.phys_sector_size_entry = md_entry;
723             s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT;
724             continue;
725         }
726 
727         if (guid_eq(md_entry.item_id, parent_locator_guid)) {
728             if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
729                 ret = -EINVAL;
730                 goto exit;
731             }
732             s->metadata_entries.parent_locator_entry = md_entry;
733             s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT;
734             continue;
735         }
736 
737         if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) {
738             /* cannot read vhdx file - required region table entry that
739              * we do not understand.  per spec, we must fail to open */
740             ret = -ENOTSUP;
741             goto exit;
742         }
743     }
744 
745     if (s->metadata_entries.present != META_ALL_PRESENT) {
746         ret = -ENOTSUP;
747         goto exit;
748     }
749 
750     ret = bdrv_pread(bs->file,
751                      s->metadata_entries.file_parameters_entry.offset
752                                          + s->metadata_rt.file_offset,
753                      &s->params,
754                      sizeof(s->params));
755 
756     if (ret < 0) {
757         goto exit;
758     }
759 
760     s->params.block_size = le32_to_cpu(s->params.block_size);
761     s->params.data_bits = le32_to_cpu(s->params.data_bits);
762 
763 
764     /* We now have the file parameters, so we can tell if this is a
765      * differencing file (i.e.. has_parent), is dynamic or fixed
766      * sized (leave_blocks_allocated), and the block size */
767 
768     /* The parent locator required iff the file parameters has_parent set */
769     if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
770         if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
771             /* TODO: parse  parent locator fields */
772             ret = -ENOTSUP; /* temp, until differencing files are supported */
773             goto exit;
774         } else {
775             /* if has_parent is set, but there is not parent locator present,
776              * then that is an invalid combination */
777             ret = -EINVAL;
778             goto exit;
779         }
780     }
781 
782     /* determine virtual disk size, logical sector size,
783      * and phys sector size */
784 
785     ret = bdrv_pread(bs->file,
786                      s->metadata_entries.virtual_disk_size_entry.offset
787                                            + s->metadata_rt.file_offset,
788                      &s->virtual_disk_size,
789                      sizeof(uint64_t));
790     if (ret < 0) {
791         goto exit;
792     }
793     ret = bdrv_pread(bs->file,
794                      s->metadata_entries.logical_sector_size_entry.offset
795                                              + s->metadata_rt.file_offset,
796                      &s->logical_sector_size,
797                      sizeof(uint32_t));
798     if (ret < 0) {
799         goto exit;
800     }
801     ret = bdrv_pread(bs->file,
802                      s->metadata_entries.phys_sector_size_entry.offset
803                                           + s->metadata_rt.file_offset,
804                      &s->physical_sector_size,
805                      sizeof(uint32_t));
806     if (ret < 0) {
807         goto exit;
808     }
809 
810     s->virtual_disk_size = le64_to_cpu(s->virtual_disk_size);
811     s->logical_sector_size = le32_to_cpu(s->logical_sector_size);
812     s->physical_sector_size = le32_to_cpu(s->physical_sector_size);
813 
814     if (s->params.block_size < VHDX_BLOCK_SIZE_MIN ||
815         s->params.block_size > VHDX_BLOCK_SIZE_MAX) {
816         ret = -EINVAL;
817         goto exit;
818     }
819 
820     /* only 2 supported sector sizes */
821     if (s->logical_sector_size != 512 && s->logical_sector_size != 4096) {
822         ret = -EINVAL;
823         goto exit;
824     }
825 
826     /* Both block_size and sector_size are guaranteed powers of 2, below.
827        Due to range checks above, s->sectors_per_block can never be < 256 */
828     s->sectors_per_block = s->params.block_size / s->logical_sector_size;
829     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
830                      (uint64_t)s->logical_sector_size /
831                      (uint64_t)s->params.block_size;
832 
833     /* These values are ones we will want to use for division / multiplication
834      * later on, and they are all guaranteed (per the spec) to be powers of 2,
835      * so we can take advantage of that for shift operations during
836      * reads/writes */
837     if (s->logical_sector_size & (s->logical_sector_size - 1)) {
838         ret = -EINVAL;
839         goto exit;
840     }
841     if (s->sectors_per_block & (s->sectors_per_block - 1)) {
842         ret = -EINVAL;
843         goto exit;
844     }
845     if (s->chunk_ratio & (s->chunk_ratio - 1)) {
846         ret = -EINVAL;
847         goto exit;
848     }
849     s->block_size = s->params.block_size;
850     if (s->block_size & (s->block_size - 1)) {
851         ret = -EINVAL;
852         goto exit;
853     }
854 
855     vhdx_set_shift_bits(s);
856 
857     ret = 0;
858 
859 exit:
860     qemu_vfree(buffer);
861     return ret;
862 }
863 
864 /*
865  * Calculate the number of BAT entries, including sector
866  * bitmap entries.
867  */
868 static void vhdx_calc_bat_entries(BDRVVHDXState *s)
869 {
870     uint32_t data_blocks_cnt, bitmap_blocks_cnt;
871 
872     data_blocks_cnt = DIV_ROUND_UP(s->virtual_disk_size, s->block_size);
873     bitmap_blocks_cnt = DIV_ROUND_UP(data_blocks_cnt, s->chunk_ratio);
874 
875     if (s->parent_entries) {
876         s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1);
877     } else {
878         s->bat_entries = data_blocks_cnt +
879                          ((data_blocks_cnt - 1) >> s->chunk_ratio_bits);
880     }
881 
882 }
883 
884 static int vhdx_check_bat_entries(BlockDriverState *bs, int *errcnt)
885 {
886     BDRVVHDXState *s = bs->opaque;
887     int64_t image_file_size = bdrv_getlength(bs->file->bs);
888     uint64_t payblocks = s->chunk_ratio;
889     uint64_t i;
890     int ret = 0;
891 
892     if (image_file_size < 0) {
893         error_report("Could not determinate VHDX image file size.");
894         return image_file_size;
895     }
896 
897     for (i = 0; i < s->bat_entries; i++) {
898         if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) ==
899             PAYLOAD_BLOCK_FULLY_PRESENT) {
900             uint64_t offset = s->bat[i] & VHDX_BAT_FILE_OFF_MASK;
901             /*
902              * Allow that the last block exists only partially. The VHDX spec
903              * states that the image file can only grow in blocksize increments,
904              * but QEMU created images with partial last blocks in the past.
905              */
906             uint32_t block_length = MIN(s->block_size,
907                 bs->total_sectors * BDRV_SECTOR_SIZE - i * s->block_size);
908             /*
909              * Check for BAT entry overflow.
910              */
911             if (offset > INT64_MAX - s->block_size) {
912                 error_report("VHDX BAT entry %" PRIu64 " offset overflow.", i);
913                 ret = -EINVAL;
914                 if (!errcnt) {
915                     break;
916                 }
917                 (*errcnt)++;
918             }
919             /*
920              * Check if fully allocated BAT entries do not reside after
921              * end of the image file.
922              */
923             if (offset >= image_file_size) {
924                 error_report("VHDX BAT entry %" PRIu64 " start offset %" PRIu64
925                              " points after end of file (%" PRIi64 "). Image"
926                              " has probably been truncated.",
927                              i, offset, image_file_size);
928                 ret = -EINVAL;
929                 if (!errcnt) {
930                     break;
931                 }
932                 (*errcnt)++;
933             } else if (offset + block_length > image_file_size) {
934                 error_report("VHDX BAT entry %" PRIu64 " end offset %" PRIu64
935                              " points after end of file (%" PRIi64 "). Image"
936                              " has probably been truncated.",
937                              i, offset + block_length - 1, image_file_size);
938                 ret = -EINVAL;
939                 if (!errcnt) {
940                     break;
941                 }
942                 (*errcnt)++;
943             }
944 
945             /*
946              * verify populated BAT field file offsets against
947              * region table and log entries
948              */
949             if (payblocks--) {
950                 /* payload bat entries */
951                 int ret2;
952                 ret2 = vhdx_region_check(s, offset, s->block_size);
953                 if (ret2 < 0) {
954                     ret = -EINVAL;
955                     if (!errcnt) {
956                         break;
957                     }
958                     (*errcnt)++;
959                 }
960             } else {
961                 payblocks = s->chunk_ratio;
962                 /*
963                  * Once differencing files are supported, verify sector bitmap
964                  * blocks here
965                  */
966             }
967         }
968     }
969 
970     return ret;
971 }
972 
973 static void vhdx_close(BlockDriverState *bs)
974 {
975     BDRVVHDXState *s = bs->opaque;
976     qemu_vfree(s->headers[0]);
977     s->headers[0] = NULL;
978     qemu_vfree(s->headers[1]);
979     s->headers[1] = NULL;
980     qemu_vfree(s->bat);
981     s->bat = NULL;
982     qemu_vfree(s->parent_entries);
983     s->parent_entries = NULL;
984     migrate_del_blocker(s->migration_blocker);
985     error_free(s->migration_blocker);
986     qemu_vfree(s->log.hdr);
987     s->log.hdr = NULL;
988     vhdx_region_unregister_all(s);
989 }
990 
991 static int vhdx_open(BlockDriverState *bs, QDict *options, int flags,
992                      Error **errp)
993 {
994     BDRVVHDXState *s = bs->opaque;
995     int ret = 0;
996     uint32_t i;
997     uint64_t signature;
998     Error *local_err = NULL;
999 
1000     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
1001                                false, errp);
1002     if (!bs->file) {
1003         return -EINVAL;
1004     }
1005 
1006     s->bat = NULL;
1007     s->first_visible_write = true;
1008 
1009     qemu_co_mutex_init(&s->lock);
1010     QLIST_INIT(&s->regions);
1011 
1012     /* validate the file signature */
1013     ret = bdrv_pread(bs->file, 0, &signature, sizeof(uint64_t));
1014     if (ret < 0) {
1015         goto fail;
1016     }
1017     if (memcmp(&signature, "vhdxfile", 8)) {
1018         ret = -EINVAL;
1019         goto fail;
1020     }
1021 
1022     /* This is used for any header updates, for the file_write_guid.
1023      * The spec dictates that a new value should be used for the first
1024      * header update */
1025     vhdx_guid_generate(&s->session_guid);
1026 
1027     vhdx_parse_header(bs, s, &local_err);
1028     if (local_err != NULL) {
1029         error_propagate(errp, local_err);
1030         ret = -EINVAL;
1031         goto fail;
1032     }
1033 
1034     ret = vhdx_parse_log(bs, s, &s->log_replayed_on_open, errp);
1035     if (ret < 0) {
1036         goto fail;
1037     }
1038 
1039     ret = vhdx_open_region_tables(bs, s);
1040     if (ret < 0) {
1041         goto fail;
1042     }
1043 
1044     ret = vhdx_parse_metadata(bs, s);
1045     if (ret < 0) {
1046         goto fail;
1047     }
1048 
1049     s->block_size = s->params.block_size;
1050 
1051     /* the VHDX spec dictates that virtual_disk_size is always a multiple of
1052      * logical_sector_size */
1053     bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits;
1054 
1055     vhdx_calc_bat_entries(s);
1056 
1057     s->bat_offset = s->bat_rt.file_offset;
1058 
1059     if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) {
1060         /* BAT allocation is not large enough for all entries */
1061         ret = -EINVAL;
1062         goto fail;
1063     }
1064 
1065     /* s->bat is freed in vhdx_close() */
1066     s->bat = qemu_try_blockalign(bs->file->bs, s->bat_rt.length);
1067     if (s->bat == NULL) {
1068         ret = -ENOMEM;
1069         goto fail;
1070     }
1071 
1072     ret = bdrv_pread(bs->file, s->bat_offset, s->bat, s->bat_rt.length);
1073     if (ret < 0) {
1074         goto fail;
1075     }
1076 
1077     /* endian convert populated BAT field entires */
1078     for (i = 0; i < s->bat_entries; i++) {
1079         s->bat[i] = le64_to_cpu(s->bat[i]);
1080     }
1081 
1082     if (!(flags & BDRV_O_CHECK)) {
1083         ret = vhdx_check_bat_entries(bs, NULL);
1084         if (ret < 0) {
1085             goto fail;
1086         }
1087     }
1088 
1089     /* Disable migration when VHDX images are used */
1090     error_setg(&s->migration_blocker, "The vhdx format used by node '%s' "
1091                "does not support live migration",
1092                bdrv_get_device_or_node_name(bs));
1093     ret = migrate_add_blocker(s->migration_blocker, &local_err);
1094     if (local_err) {
1095         error_propagate(errp, local_err);
1096         error_free(s->migration_blocker);
1097         goto fail;
1098     }
1099 
1100     /* TODO: differencing files */
1101 
1102     return 0;
1103 fail:
1104     vhdx_close(bs);
1105     return ret;
1106 }
1107 
1108 static int vhdx_reopen_prepare(BDRVReopenState *state,
1109                                BlockReopenQueue *queue, Error **errp)
1110 {
1111     return 0;
1112 }
1113 
1114 
1115 /*
1116  * Perform sector to block offset translations, to get various
1117  * sector and file offsets into the image.  See VHDXSectorInfo
1118  */
1119 static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num,
1120                                  int nb_sectors, VHDXSectorInfo *sinfo)
1121 {
1122     uint32_t block_offset;
1123 
1124     sinfo->bat_idx = sector_num >> s->sectors_per_block_bits;
1125     /* effectively a modulo - this gives us the offset into the block
1126      * (in sector sizes) for our sector number */
1127     block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits);
1128     /* the chunk ratio gives us the interleaving of the sector
1129      * bitmaps, so we need to advance our page block index by the
1130      * sector bitmaps entry number */
1131     sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits;
1132 
1133     /* the number of sectors we can read/write in this cycle */
1134     sinfo->sectors_avail = s->sectors_per_block - block_offset;
1135 
1136     sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits;
1137 
1138     if (sinfo->sectors_avail > nb_sectors) {
1139         sinfo->sectors_avail = nb_sectors;
1140     }
1141 
1142     sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits;
1143 
1144     sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK;
1145 
1146     sinfo->block_offset = block_offset << s->logical_sector_size_bits;
1147 
1148     /* The file offset must be past the header section, so must be > 0 */
1149     if (sinfo->file_offset == 0) {
1150         return;
1151     }
1152 
1153     /* block offset is the offset in vhdx logical sectors, in
1154      * the payload data block. Convert that to a byte offset
1155      * in the block, and add in the payload data block offset
1156      * in the file, in bytes, to get the final read address */
1157 
1158     sinfo->file_offset += sinfo->block_offset;
1159 }
1160 
1161 
1162 static int vhdx_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1163 {
1164     BDRVVHDXState *s = bs->opaque;
1165 
1166     bdi->cluster_size = s->block_size;
1167 
1168     bdi->unallocated_blocks_are_zero =
1169         (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) == 0;
1170 
1171     return 0;
1172 }
1173 
1174 
1175 static coroutine_fn int vhdx_co_readv(BlockDriverState *bs, int64_t sector_num,
1176                                       int nb_sectors, QEMUIOVector *qiov)
1177 {
1178     BDRVVHDXState *s = bs->opaque;
1179     int ret = 0;
1180     VHDXSectorInfo sinfo;
1181     uint64_t bytes_done = 0;
1182     QEMUIOVector hd_qiov;
1183 
1184     qemu_iovec_init(&hd_qiov, qiov->niov);
1185 
1186     qemu_co_mutex_lock(&s->lock);
1187 
1188     while (nb_sectors > 0) {
1189         /* We are a differencing file, so we need to inspect the sector bitmap
1190          * to see if we have the data or not */
1191         if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1192             /* not supported yet */
1193             ret = -ENOTSUP;
1194             goto exit;
1195         } else {
1196             vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1197 
1198             qemu_iovec_reset(&hd_qiov);
1199             qemu_iovec_concat(&hd_qiov, qiov,  bytes_done, sinfo.bytes_avail);
1200 
1201             /* check the payload block state */
1202             switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) {
1203             case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1204             case PAYLOAD_BLOCK_UNDEFINED:
1205             case PAYLOAD_BLOCK_UNMAPPED:
1206             case PAYLOAD_BLOCK_UNMAPPED_v095:
1207             case PAYLOAD_BLOCK_ZERO:
1208                 /* return zero */
1209                 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail);
1210                 break;
1211             case PAYLOAD_BLOCK_FULLY_PRESENT:
1212                 qemu_co_mutex_unlock(&s->lock);
1213                 ret = bdrv_co_preadv(bs->file, sinfo.file_offset,
1214                                      sinfo.sectors_avail * BDRV_SECTOR_SIZE,
1215                                      &hd_qiov, 0);
1216                 qemu_co_mutex_lock(&s->lock);
1217                 if (ret < 0) {
1218                     goto exit;
1219                 }
1220                 break;
1221             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1222                 /* we don't yet support difference files, fall through
1223                  * to error */
1224             default:
1225                 ret = -EIO;
1226                 goto exit;
1227                 break;
1228             }
1229             nb_sectors -= sinfo.sectors_avail;
1230             sector_num += sinfo.sectors_avail;
1231             bytes_done += sinfo.bytes_avail;
1232         }
1233     }
1234     ret = 0;
1235 exit:
1236     qemu_co_mutex_unlock(&s->lock);
1237     qemu_iovec_destroy(&hd_qiov);
1238     return ret;
1239 }
1240 
1241 /*
1242  * Allocate a new payload block at the end of the file.
1243  *
1244  * Allocation will happen at 1MB alignment inside the file
1245  *
1246  * Returns the file offset start of the new payload block
1247  */
1248 static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s,
1249                                     uint64_t *new_offset)
1250 {
1251     int64_t current_len;
1252 
1253     current_len = bdrv_getlength(bs->file->bs);
1254     if (current_len < 0) {
1255         return current_len;
1256     }
1257 
1258     *new_offset = current_len;
1259 
1260     /* per the spec, the address for a block is in units of 1MB */
1261     *new_offset = ROUND_UP(*new_offset, 1 * MiB);
1262     if (*new_offset > INT64_MAX) {
1263         return -EINVAL;
1264     }
1265 
1266     return bdrv_truncate(bs->file, *new_offset + s->block_size, false,
1267                          PREALLOC_MODE_OFF, NULL);
1268 }
1269 
1270 /*
1271  * Update the BAT table entry with the new file offset, and the new entry
1272  * state */
1273 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s,
1274                                        VHDXSectorInfo *sinfo,
1275                                        uint64_t *bat_entry_le,
1276                                        uint64_t *bat_offset, int state)
1277 {
1278     /* The BAT entry is a uint64, with 44 bits for the file offset in units of
1279      * 1MB, and 3 bits for the block state. */
1280     if ((state == PAYLOAD_BLOCK_ZERO)        ||
1281         (state == PAYLOAD_BLOCK_UNDEFINED)   ||
1282         (state == PAYLOAD_BLOCK_NOT_PRESENT) ||
1283         (state == PAYLOAD_BLOCK_UNMAPPED)) {
1284         s->bat[sinfo->bat_idx]  = 0;  /* For PAYLOAD_BLOCK_ZERO, the
1285                                          FileOffsetMB field is denoted as
1286                                          'reserved' in the v1.0 spec.  If it is
1287                                          non-zero, MS Hyper-V will fail to read
1288                                          the disk image */
1289     } else {
1290         s->bat[sinfo->bat_idx]  = sinfo->file_offset;
1291     }
1292 
1293     s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK;
1294 
1295     *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]);
1296     *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry);
1297 
1298 }
1299 
1300 /* Per the spec, on the first write of guest-visible data to the file the
1301  * data write guid must be updated in the header */
1302 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s)
1303 {
1304     int ret = 0;
1305     if (s->first_visible_write) {
1306         s->first_visible_write = false;
1307         ret = vhdx_update_headers(bs, s, true, NULL);
1308     }
1309     return ret;
1310 }
1311 
1312 static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num,
1313                                        int nb_sectors, QEMUIOVector *qiov,
1314                                        int flags)
1315 {
1316     int ret = -ENOTSUP;
1317     BDRVVHDXState *s = bs->opaque;
1318     VHDXSectorInfo sinfo;
1319     uint64_t bytes_done = 0;
1320     uint64_t bat_entry = 0;
1321     uint64_t bat_entry_offset = 0;
1322     QEMUIOVector hd_qiov;
1323     struct iovec iov1 = { 0 };
1324     struct iovec iov2 = { 0 };
1325     int sectors_to_write;
1326     int bat_state;
1327     uint64_t bat_prior_offset = 0;
1328     bool bat_update = false;
1329 
1330     assert(!flags);
1331     qemu_iovec_init(&hd_qiov, qiov->niov);
1332 
1333     qemu_co_mutex_lock(&s->lock);
1334 
1335     ret = vhdx_user_visible_write(bs, s);
1336     if (ret < 0) {
1337         goto exit;
1338     }
1339 
1340     while (nb_sectors > 0) {
1341         bool use_zero_buffers = false;
1342         bat_update = false;
1343         if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1344             /* not supported yet */
1345             ret = -ENOTSUP;
1346             goto exit;
1347         } else {
1348             vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1349             sectors_to_write = sinfo.sectors_avail;
1350 
1351             qemu_iovec_reset(&hd_qiov);
1352             /* check the payload block state */
1353             bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK;
1354             switch (bat_state) {
1355             case PAYLOAD_BLOCK_ZERO:
1356                 /* in this case, we need to preserve zero writes for
1357                  * data that is not part of this write, so we must pad
1358                  * the rest of the buffer to zeroes */
1359 
1360                 /* if we are on a posix system with ftruncate() that extends
1361                  * a file, then it is zero-filled for us.  On Win32, the raw
1362                  * layer uses SetFilePointer and SetFileEnd, which does not
1363                  * zero fill AFAIK */
1364 
1365                 /* Queue another write of zero buffers if the underlying file
1366                  * does not zero-fill on file extension */
1367 
1368                 if (bdrv_has_zero_init_truncate(bs->file->bs) == 0) {
1369                     use_zero_buffers = true;
1370 
1371                     /* zero fill the front, if any */
1372                     if (sinfo.block_offset) {
1373                         iov1.iov_len = sinfo.block_offset;
1374                         iov1.iov_base = qemu_blockalign(bs, iov1.iov_len);
1375                         memset(iov1.iov_base, 0, iov1.iov_len);
1376                         qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0,
1377                                               iov1.iov_len);
1378                         sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS;
1379                     }
1380 
1381                     /* our actual data */
1382                     qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1383                                       sinfo.bytes_avail);
1384 
1385                     /* zero fill the back, if any */
1386                     if ((sinfo.bytes_avail - sinfo.block_offset) <
1387                          s->block_size) {
1388                         iov2.iov_len = s->block_size -
1389                                       (sinfo.bytes_avail + sinfo.block_offset);
1390                         iov2.iov_base = qemu_blockalign(bs, iov2.iov_len);
1391                         memset(iov2.iov_base, 0, iov2.iov_len);
1392                         qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0,
1393                                               iov2.iov_len);
1394                         sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS;
1395                     }
1396                 }
1397                 /* fall through */
1398             case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1399             case PAYLOAD_BLOCK_UNMAPPED:
1400             case PAYLOAD_BLOCK_UNMAPPED_v095:
1401             case PAYLOAD_BLOCK_UNDEFINED:
1402                 bat_prior_offset = sinfo.file_offset;
1403                 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset);
1404                 if (ret < 0) {
1405                     goto exit;
1406                 }
1407                 /* once we support differencing files, this may also be
1408                  * partially present */
1409                 /* update block state to the newly specified state */
1410                 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1411                                             &bat_entry_offset,
1412                                             PAYLOAD_BLOCK_FULLY_PRESENT);
1413                 bat_update = true;
1414                 /* since we just allocated a block, file_offset is the
1415                  * beginning of the payload block. It needs to be the
1416                  * write address, which includes the offset into the block */
1417                 if (!use_zero_buffers) {
1418                     sinfo.file_offset += sinfo.block_offset;
1419                 }
1420                 /* fall through */
1421             case PAYLOAD_BLOCK_FULLY_PRESENT:
1422                 /* if the file offset address is in the header zone,
1423                  * there is a problem */
1424                 if (sinfo.file_offset < (1 * MiB)) {
1425                     ret = -EFAULT;
1426                     goto error_bat_restore;
1427                 }
1428 
1429                 if (!use_zero_buffers) {
1430                     qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1431                                       sinfo.bytes_avail);
1432                 }
1433                 /* block exists, so we can just overwrite it */
1434                 qemu_co_mutex_unlock(&s->lock);
1435                 ret = bdrv_co_pwritev(bs->file, sinfo.file_offset,
1436                                       sectors_to_write * BDRV_SECTOR_SIZE,
1437                                       &hd_qiov, 0);
1438                 qemu_co_mutex_lock(&s->lock);
1439                 if (ret < 0) {
1440                     goto error_bat_restore;
1441                 }
1442                 break;
1443             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1444                 /* we don't yet support difference files, fall through
1445                  * to error */
1446             default:
1447                 ret = -EIO;
1448                 goto exit;
1449                 break;
1450             }
1451 
1452             if (bat_update) {
1453                 /* this will update the BAT entry into the log journal, and
1454                  * then flush the log journal out to disk */
1455                 ret =  vhdx_log_write_and_flush(bs, s, &bat_entry,
1456                                                 sizeof(VHDXBatEntry),
1457                                                 bat_entry_offset);
1458                 if (ret < 0) {
1459                     goto exit;
1460                 }
1461             }
1462 
1463             nb_sectors -= sinfo.sectors_avail;
1464             sector_num += sinfo.sectors_avail;
1465             bytes_done += sinfo.bytes_avail;
1466 
1467         }
1468     }
1469 
1470     goto exit;
1471 
1472 error_bat_restore:
1473     if (bat_update) {
1474         /* keep metadata in sync, and restore the bat entry state
1475          * if error. */
1476         sinfo.file_offset = bat_prior_offset;
1477         vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1478                                     &bat_entry_offset, bat_state);
1479     }
1480 exit:
1481     qemu_vfree(iov1.iov_base);
1482     qemu_vfree(iov2.iov_base);
1483     qemu_co_mutex_unlock(&s->lock);
1484     qemu_iovec_destroy(&hd_qiov);
1485     return ret;
1486 }
1487 
1488 
1489 
1490 /*
1491  * Create VHDX Headers
1492  *
1493  * There are 2 headers, and the highest sequence number will represent
1494  * the active header
1495  */
1496 static int vhdx_create_new_headers(BlockBackend *blk, uint64_t image_size,
1497                                    uint32_t log_size)
1498 {
1499     BlockDriverState *bs = blk_bs(blk);
1500     BdrvChild *child;
1501     int ret = 0;
1502     VHDXHeader *hdr = NULL;
1503 
1504     hdr = g_new0(VHDXHeader, 1);
1505 
1506     hdr->signature       = VHDX_HEADER_SIGNATURE;
1507     hdr->sequence_number = g_random_int();
1508     hdr->log_version     = 0;
1509     hdr->version         = 1;
1510     hdr->log_length      = log_size;
1511     hdr->log_offset      = VHDX_HEADER_SECTION_END;
1512     vhdx_guid_generate(&hdr->file_write_guid);
1513     vhdx_guid_generate(&hdr->data_write_guid);
1514 
1515     /* XXX Ugly way to get blk->root, but that's a feature, not a bug. This
1516      * hack makes it obvious that vhdx_write_header() bypasses the BlockBackend
1517      * here, which it really shouldn't be doing. */
1518     child = QLIST_FIRST(&bs->parents);
1519     assert(!QLIST_NEXT(child, next_parent));
1520 
1521     ret = vhdx_write_header(child, hdr, VHDX_HEADER1_OFFSET, false);
1522     if (ret < 0) {
1523         goto exit;
1524     }
1525     hdr->sequence_number++;
1526     ret = vhdx_write_header(child, hdr, VHDX_HEADER2_OFFSET, false);
1527     if (ret < 0) {
1528         goto exit;
1529     }
1530 
1531 exit:
1532     g_free(hdr);
1533     return ret;
1534 }
1535 
1536 #define VHDX_METADATA_ENTRY_BUFFER_SIZE \
1537                                     (sizeof(VHDXFileParameters)               +\
1538                                      sizeof(VHDXVirtualDiskSize)              +\
1539                                      sizeof(VHDXPage83Data)                   +\
1540                                      sizeof(VHDXVirtualDiskLogicalSectorSize) +\
1541                                      sizeof(VHDXVirtualDiskPhysicalSectorSize))
1542 
1543 /*
1544  * Create the Metadata entries.
1545  *
1546  * For more details on the entries, see section 3.5 (pg 29) in the
1547  * VHDX 1.00 specification.
1548  *
1549  * We support 5 metadata entries (all required by spec):
1550  *          File Parameters,
1551  *          Virtual Disk Size,
1552  *          Page 83 Data,
1553  *          Logical Sector Size,
1554  *          Physical Sector Size
1555  *
1556  * The first 64KB of the Metadata section is reserved for the metadata
1557  * header and entries; beyond that, the metadata items themselves reside.
1558  */
1559 static int vhdx_create_new_metadata(BlockBackend *blk,
1560                                     uint64_t image_size,
1561                                     uint32_t block_size,
1562                                     uint32_t sector_size,
1563                                     uint64_t metadata_offset,
1564                                     VHDXImageType type)
1565 {
1566     int ret = 0;
1567     uint32_t offset = 0;
1568     void *buffer = NULL;
1569     void *entry_buffer;
1570     VHDXMetadataTableHeader *md_table;
1571     VHDXMetadataTableEntry  *md_table_entry;
1572 
1573     /* Metadata entries */
1574     VHDXFileParameters     *mt_file_params;
1575     VHDXVirtualDiskSize    *mt_virtual_size;
1576     VHDXPage83Data         *mt_page83;
1577     VHDXVirtualDiskLogicalSectorSize  *mt_log_sector_size;
1578     VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size;
1579 
1580     entry_buffer = g_malloc0(VHDX_METADATA_ENTRY_BUFFER_SIZE);
1581 
1582     mt_file_params = entry_buffer;
1583     offset += sizeof(VHDXFileParameters);
1584     mt_virtual_size = entry_buffer + offset;
1585     offset += sizeof(VHDXVirtualDiskSize);
1586     mt_page83 = entry_buffer + offset;
1587     offset += sizeof(VHDXPage83Data);
1588     mt_log_sector_size = entry_buffer + offset;
1589     offset += sizeof(VHDXVirtualDiskLogicalSectorSize);
1590     mt_phys_sector_size = entry_buffer + offset;
1591 
1592     mt_file_params->block_size = cpu_to_le32(block_size);
1593     if (type == VHDX_TYPE_FIXED) {
1594         mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED;
1595         mt_file_params->data_bits = cpu_to_le32(mt_file_params->data_bits);
1596     }
1597 
1598     vhdx_guid_generate(&mt_page83->page_83_data);
1599     cpu_to_leguids(&mt_page83->page_83_data);
1600     mt_virtual_size->virtual_disk_size        = cpu_to_le64(image_size);
1601     mt_log_sector_size->logical_sector_size   = cpu_to_le32(sector_size);
1602     mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size);
1603 
1604     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1605     md_table = buffer;
1606 
1607     md_table->signature   = VHDX_METADATA_SIGNATURE;
1608     md_table->entry_count = 5;
1609     vhdx_metadata_header_le_export(md_table);
1610 
1611 
1612     /* This will reference beyond the reserved table portion */
1613     offset = 64 * KiB;
1614 
1615     md_table_entry = buffer + sizeof(VHDXMetadataTableHeader);
1616 
1617     md_table_entry[0].item_id = file_param_guid;
1618     md_table_entry[0].offset  = offset;
1619     md_table_entry[0].length  = sizeof(VHDXFileParameters);
1620     md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED;
1621     offset += md_table_entry[0].length;
1622     vhdx_metadata_entry_le_export(&md_table_entry[0]);
1623 
1624     md_table_entry[1].item_id = virtual_size_guid;
1625     md_table_entry[1].offset  = offset;
1626     md_table_entry[1].length  = sizeof(VHDXVirtualDiskSize);
1627     md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1628                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1629     offset += md_table_entry[1].length;
1630     vhdx_metadata_entry_le_export(&md_table_entry[1]);
1631 
1632     md_table_entry[2].item_id = page83_guid;
1633     md_table_entry[2].offset  = offset;
1634     md_table_entry[2].length  = sizeof(VHDXPage83Data);
1635     md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1636                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1637     offset += md_table_entry[2].length;
1638     vhdx_metadata_entry_le_export(&md_table_entry[2]);
1639 
1640     md_table_entry[3].item_id = logical_sector_guid;
1641     md_table_entry[3].offset  = offset;
1642     md_table_entry[3].length  = sizeof(VHDXVirtualDiskLogicalSectorSize);
1643     md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1644                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1645     offset += md_table_entry[3].length;
1646     vhdx_metadata_entry_le_export(&md_table_entry[3]);
1647 
1648     md_table_entry[4].item_id = phys_sector_guid;
1649     md_table_entry[4].offset  = offset;
1650     md_table_entry[4].length  = sizeof(VHDXVirtualDiskPhysicalSectorSize);
1651     md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1652                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1653     vhdx_metadata_entry_le_export(&md_table_entry[4]);
1654 
1655     ret = blk_pwrite(blk, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE, 0);
1656     if (ret < 0) {
1657         goto exit;
1658     }
1659 
1660     ret = blk_pwrite(blk, metadata_offset + (64 * KiB), entry_buffer,
1661                      VHDX_METADATA_ENTRY_BUFFER_SIZE, 0);
1662     if (ret < 0) {
1663         goto exit;
1664     }
1665 
1666 
1667 exit:
1668     g_free(buffer);
1669     g_free(entry_buffer);
1670     return ret;
1671 }
1672 
1673 /* This create the actual BAT itself.  We currently only support
1674  * 'Dynamic' and 'Fixed' image types.
1675  *
1676  *  Dynamic images: default state of the BAT is all zeroes.
1677  *
1678  *  Fixed images: default state of the BAT is fully populated, with
1679  *                file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT.
1680  */
1681 static int vhdx_create_bat(BlockBackend *blk, BDRVVHDXState *s,
1682                            uint64_t image_size, VHDXImageType type,
1683                            bool use_zero_blocks, uint64_t file_offset,
1684                            uint32_t length, Error **errp)
1685 {
1686     int ret = 0;
1687     uint64_t data_file_offset;
1688     uint64_t total_sectors = 0;
1689     uint64_t sector_num = 0;
1690     uint64_t unused;
1691     int block_state;
1692     VHDXSectorInfo sinfo;
1693 
1694     assert(s->bat == NULL);
1695 
1696     /* this gives a data start after BAT/bitmap entries, and well
1697      * past any metadata entries (with a 4 MB buffer for future
1698      * expansion */
1699     data_file_offset = file_offset + length + 5 * MiB;
1700     total_sectors = image_size >> s->logical_sector_size_bits;
1701 
1702     if (type == VHDX_TYPE_DYNAMIC) {
1703         /* All zeroes, so we can just extend the file - the end of the BAT
1704          * is the furthest thing we have written yet */
1705         ret = blk_truncate(blk, data_file_offset, false, PREALLOC_MODE_OFF,
1706                            errp);
1707         if (ret < 0) {
1708             goto exit;
1709         }
1710     } else if (type == VHDX_TYPE_FIXED) {
1711         ret = blk_truncate(blk, data_file_offset + image_size, false,
1712                            PREALLOC_MODE_OFF, errp);
1713         if (ret < 0) {
1714             goto exit;
1715         }
1716     } else {
1717         error_setg(errp, "Unsupported image type");
1718         ret = -ENOTSUP;
1719         goto exit;
1720     }
1721 
1722     if (type == VHDX_TYPE_FIXED ||
1723                 use_zero_blocks ||
1724                 bdrv_has_zero_init(blk_bs(blk)) == 0) {
1725         /* for a fixed file, the default BAT entry is not zero */
1726         s->bat = g_try_malloc0(length);
1727         if (length && s->bat == NULL) {
1728             error_setg(errp, "Failed to allocate memory for the BAT");
1729             ret = -ENOMEM;
1730             goto exit;
1731         }
1732         block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT :
1733                                                 PAYLOAD_BLOCK_NOT_PRESENT;
1734         block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state;
1735         /* fill the BAT by emulating sector writes of sectors_per_block size */
1736         while (sector_num < total_sectors) {
1737             vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo);
1738             sinfo.file_offset = data_file_offset +
1739                                 (sector_num << s->logical_sector_size_bits);
1740             sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB);
1741             vhdx_update_bat_table_entry(blk_bs(blk), s, &sinfo, &unused, &unused,
1742                                         block_state);
1743             s->bat[sinfo.bat_idx] = cpu_to_le64(s->bat[sinfo.bat_idx]);
1744             sector_num += s->sectors_per_block;
1745         }
1746         ret = blk_pwrite(blk, file_offset, s->bat, length, 0);
1747         if (ret < 0) {
1748             error_setg_errno(errp, -ret, "Failed to write the BAT");
1749             goto exit;
1750         }
1751     }
1752 
1753 
1754 
1755 exit:
1756     g_free(s->bat);
1757     return ret;
1758 }
1759 
1760 /* Creates the region table header, and region table entries.
1761  * There are 2 supported region table entries: BAT, and Metadata/
1762  *
1763  * As the calculations for the BAT region table are also needed
1764  * to create the BAT itself, we will also cause the BAT to be
1765  * created.
1766  */
1767 static int vhdx_create_new_region_table(BlockBackend *blk,
1768                                         uint64_t image_size,
1769                                         uint32_t block_size,
1770                                         uint32_t sector_size,
1771                                         uint32_t log_size,
1772                                         bool use_zero_blocks,
1773                                         VHDXImageType type,
1774                                         uint64_t *metadata_offset,
1775                                         Error **errp)
1776 {
1777     int ret = 0;
1778     uint32_t offset = 0;
1779     void *buffer = NULL;
1780     uint64_t bat_file_offset;
1781     uint32_t bat_length;
1782     BDRVVHDXState *s = NULL;
1783     VHDXRegionTableHeader *region_table;
1784     VHDXRegionTableEntry *rt_bat;
1785     VHDXRegionTableEntry *rt_metadata;
1786 
1787     assert(metadata_offset != NULL);
1788 
1789     /* Populate enough of the BDRVVHDXState to be able to use the
1790      * pre-existing BAT calculation, translation, and update functions */
1791     s = g_new0(BDRVVHDXState, 1);
1792 
1793     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
1794                      (uint64_t) sector_size / (uint64_t) block_size;
1795 
1796     s->sectors_per_block = block_size / sector_size;
1797     s->virtual_disk_size = image_size;
1798     s->block_size = block_size;
1799     s->logical_sector_size = sector_size;
1800 
1801     vhdx_set_shift_bits(s);
1802 
1803     vhdx_calc_bat_entries(s);
1804 
1805     /* At this point the VHDX state is populated enough for creation */
1806 
1807     /* a single buffer is used so we can calculate the checksum over the
1808      * entire 64KB block */
1809     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1810     region_table = buffer;
1811     offset += sizeof(VHDXRegionTableHeader);
1812     rt_bat = buffer + offset;
1813     offset += sizeof(VHDXRegionTableEntry);
1814     rt_metadata  = buffer + offset;
1815 
1816     region_table->signature = VHDX_REGION_SIGNATURE;
1817     region_table->entry_count = 2;   /* BAT and Metadata */
1818 
1819     rt_bat->guid        = bat_guid;
1820     rt_bat->length      = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB);
1821     rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB);
1822     s->bat_offset = rt_bat->file_offset;
1823 
1824     rt_metadata->guid        = metadata_guid;
1825     rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length,
1826                                         MiB);
1827     rt_metadata->length      = 1 * MiB; /* min size, and more than enough */
1828     *metadata_offset = rt_metadata->file_offset;
1829 
1830     bat_file_offset = rt_bat->file_offset;
1831     bat_length = rt_bat->length;
1832 
1833     vhdx_region_header_le_export(region_table);
1834     vhdx_region_entry_le_export(rt_bat);
1835     vhdx_region_entry_le_export(rt_metadata);
1836 
1837     vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE,
1838                          offsetof(VHDXRegionTableHeader, checksum));
1839 
1840 
1841     /* The region table gives us the data we need to create the BAT,
1842      * so do that now */
1843     ret = vhdx_create_bat(blk, s, image_size, type, use_zero_blocks,
1844                           bat_file_offset, bat_length, errp);
1845     if (ret < 0) {
1846         goto exit;
1847     }
1848 
1849     /* Now write out the region headers to disk */
1850     ret = blk_pwrite(blk, VHDX_REGION_TABLE_OFFSET, buffer,
1851                      VHDX_HEADER_BLOCK_SIZE, 0);
1852     if (ret < 0) {
1853         error_setg_errno(errp, -ret, "Failed to write first region table");
1854         goto exit;
1855     }
1856 
1857     ret = blk_pwrite(blk, VHDX_REGION_TABLE2_OFFSET, buffer,
1858                      VHDX_HEADER_BLOCK_SIZE, 0);
1859     if (ret < 0) {
1860         error_setg_errno(errp, -ret, "Failed to write second region table");
1861         goto exit;
1862     }
1863 
1864 exit:
1865     g_free(s);
1866     g_free(buffer);
1867     return ret;
1868 }
1869 
1870 /* We need to create the following elements:
1871  *
1872  *    .-----------------------------------------------------------------.
1873  *    |   (A)    |   (B)    |    (C)    |     (D)       |     (E)       |
1874  *    |  File ID |  Header1 |  Header 2 |  Region Tbl 1 |  Region Tbl 2 |
1875  *    |          |          |           |               |               |
1876  *    .-----------------------------------------------------------------.
1877  *    0         64KB      128KB       192KB           256KB           320KB
1878  *
1879  *
1880  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1881  *    |     (F)     |     (G)       |    (H)    |                        |
1882  *    | Journal Log |  BAT / Bitmap |  Metadata |  .... data ......      |
1883  *    |             |               |           |                        |
1884  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1885  *   1MB
1886  */
1887 static int coroutine_fn vhdx_co_create(BlockdevCreateOptions *opts,
1888                                        Error **errp)
1889 {
1890     BlockdevCreateOptionsVhdx *vhdx_opts;
1891     BlockBackend *blk = NULL;
1892     BlockDriverState *bs = NULL;
1893 
1894     int ret = 0;
1895     uint64_t image_size;
1896     uint32_t log_size;
1897     uint32_t block_size;
1898     uint64_t signature;
1899     uint64_t metadata_offset;
1900     bool use_zero_blocks = false;
1901 
1902     gunichar2 *creator = NULL;
1903     glong creator_items;
1904     VHDXImageType image_type;
1905 
1906     assert(opts->driver == BLOCKDEV_DRIVER_VHDX);
1907     vhdx_opts = &opts->u.vhdx;
1908 
1909     /* Validate options and set default values */
1910     image_size = vhdx_opts->size;
1911     if (image_size > VHDX_MAX_IMAGE_SIZE) {
1912         error_setg(errp, "Image size too large; max of 64TB");
1913         return -EINVAL;
1914     }
1915 
1916     if (!vhdx_opts->has_log_size) {
1917         log_size = DEFAULT_LOG_SIZE;
1918     } else {
1919         if (vhdx_opts->log_size > UINT32_MAX) {
1920             error_setg(errp, "Log size must be smaller than 4 GB");
1921             return -EINVAL;
1922         }
1923         log_size = vhdx_opts->log_size;
1924     }
1925     if (log_size < MiB || (log_size % MiB) != 0) {
1926         error_setg(errp, "Log size must be a multiple of 1 MB");
1927         return -EINVAL;
1928     }
1929 
1930     if (!vhdx_opts->has_block_state_zero) {
1931         use_zero_blocks = true;
1932     } else {
1933         use_zero_blocks = vhdx_opts->block_state_zero;
1934     }
1935 
1936     if (!vhdx_opts->has_subformat) {
1937         vhdx_opts->subformat = BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC;
1938     }
1939 
1940     switch (vhdx_opts->subformat) {
1941     case BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC:
1942         image_type = VHDX_TYPE_DYNAMIC;
1943         break;
1944     case BLOCKDEV_VHDX_SUBFORMAT_FIXED:
1945         image_type = VHDX_TYPE_FIXED;
1946         break;
1947     default:
1948         g_assert_not_reached();
1949     }
1950 
1951     /* These are pretty arbitrary, and mainly designed to keep the BAT
1952      * size reasonable to load into RAM */
1953     if (vhdx_opts->has_block_size) {
1954         block_size = vhdx_opts->block_size;
1955     } else {
1956         if (image_size > 32 * TiB) {
1957             block_size = 64 * MiB;
1958         } else if (image_size > (uint64_t) 100 * GiB) {
1959             block_size = 32 * MiB;
1960         } else if (image_size > 1 * GiB) {
1961             block_size = 16 * MiB;
1962         } else {
1963             block_size = 8 * MiB;
1964         }
1965     }
1966 
1967     if (block_size < MiB || (block_size % MiB) != 0) {
1968         error_setg(errp, "Block size must be a multiple of 1 MB");
1969         return -EINVAL;
1970     }
1971     if (!is_power_of_2(block_size)) {
1972         error_setg(errp, "Block size must be a power of two");
1973         return -EINVAL;
1974     }
1975     if (block_size > VHDX_BLOCK_SIZE_MAX) {
1976         error_setg(errp, "Block size must not exceed %" PRId64,
1977                    VHDX_BLOCK_SIZE_MAX);
1978         return -EINVAL;
1979     }
1980 
1981     /* Create BlockBackend to write to the image */
1982     bs = bdrv_open_blockdev_ref(vhdx_opts->file, errp);
1983     if (bs == NULL) {
1984         return -EIO;
1985     }
1986 
1987     blk = blk_new(bdrv_get_aio_context(bs),
1988                   BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
1989     ret = blk_insert_bs(blk, bs, errp);
1990     if (ret < 0) {
1991         goto delete_and_exit;
1992     }
1993     blk_set_allow_write_beyond_eof(blk, true);
1994 
1995     /* Create (A) */
1996 
1997     /* The creator field is optional, but may be useful for
1998      * debugging / diagnostics */
1999     creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL,
2000                               &creator_items, NULL);
2001     signature = cpu_to_le64(VHDX_FILE_SIGNATURE);
2002     ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature),
2003                      0);
2004     if (ret < 0) {
2005         error_setg_errno(errp, -ret, "Failed to write file signature");
2006         goto delete_and_exit;
2007     }
2008     if (creator) {
2009         ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET + sizeof(signature),
2010                          creator, creator_items * sizeof(gunichar2), 0);
2011         if (ret < 0) {
2012             error_setg_errno(errp, -ret, "Failed to write creator field");
2013             goto delete_and_exit;
2014         }
2015     }
2016 
2017 
2018     /* Creates (B),(C) */
2019     ret = vhdx_create_new_headers(blk, image_size, log_size);
2020     if (ret < 0) {
2021         error_setg_errno(errp, -ret, "Failed to write image headers");
2022         goto delete_and_exit;
2023     }
2024 
2025     /* Creates (D),(E),(G) explicitly. (F) created as by-product */
2026     ret = vhdx_create_new_region_table(blk, image_size, block_size, 512,
2027                                        log_size, use_zero_blocks, image_type,
2028                                        &metadata_offset, errp);
2029     if (ret < 0) {
2030         goto delete_and_exit;
2031     }
2032 
2033     /* Creates (H) */
2034     ret = vhdx_create_new_metadata(blk, image_size, block_size, 512,
2035                                    metadata_offset, image_type);
2036     if (ret < 0) {
2037         error_setg_errno(errp, -ret, "Failed to initialize metadata");
2038         goto delete_and_exit;
2039     }
2040 
2041     ret = 0;
2042 delete_and_exit:
2043     blk_unref(blk);
2044     bdrv_unref(bs);
2045     g_free(creator);
2046     return ret;
2047 }
2048 
2049 static int coroutine_fn vhdx_co_create_opts(const char *filename,
2050                                             QemuOpts *opts,
2051                                             Error **errp)
2052 {
2053     BlockdevCreateOptions *create_options = NULL;
2054     QDict *qdict;
2055     Visitor *v;
2056     BlockDriverState *bs = NULL;
2057     Error *local_err = NULL;
2058     int ret;
2059 
2060     static const QDictRenames opt_renames[] = {
2061         { VHDX_BLOCK_OPT_LOG_SIZE,      "log-size" },
2062         { VHDX_BLOCK_OPT_BLOCK_SIZE,    "block-size" },
2063         { VHDX_BLOCK_OPT_ZERO,          "block-state-zero" },
2064         { NULL, NULL },
2065     };
2066 
2067     /* Parse options and convert legacy syntax */
2068     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vhdx_create_opts, true);
2069 
2070     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
2071         ret = -EINVAL;
2072         goto fail;
2073     }
2074 
2075     /* Create and open the file (protocol layer) */
2076     ret = bdrv_create_file(filename, opts, &local_err);
2077     if (ret < 0) {
2078         error_propagate(errp, local_err);
2079         goto fail;
2080     }
2081 
2082     bs = bdrv_open(filename, NULL, NULL,
2083                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
2084     if (bs == NULL) {
2085         ret = -EIO;
2086         goto fail;
2087     }
2088 
2089     /* Now get the QAPI type BlockdevCreateOptions */
2090     qdict_put_str(qdict, "driver", "vhdx");
2091     qdict_put_str(qdict, "file", bs->node_name);
2092 
2093     v = qobject_input_visitor_new_flat_confused(qdict, errp);
2094     if (!v) {
2095         ret = -EINVAL;
2096         goto fail;
2097     }
2098 
2099     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
2100     visit_free(v);
2101 
2102     if (local_err) {
2103         error_propagate(errp, local_err);
2104         ret = -EINVAL;
2105         goto fail;
2106     }
2107 
2108     /* Silently round up sizes:
2109      * The image size is rounded to 512 bytes. Make the block and log size
2110      * close to what was specified, but must be at least 1MB, and a multiple of
2111      * 1 MB. Also respect VHDX_BLOCK_SIZE_MAX for block sizes. block_size = 0
2112      * means auto, which is represented by a missing key in QAPI. */
2113     assert(create_options->driver == BLOCKDEV_DRIVER_VHDX);
2114     create_options->u.vhdx.size =
2115         ROUND_UP(create_options->u.vhdx.size, BDRV_SECTOR_SIZE);
2116 
2117     if (create_options->u.vhdx.has_log_size) {
2118         create_options->u.vhdx.log_size =
2119             ROUND_UP(create_options->u.vhdx.log_size, MiB);
2120     }
2121     if (create_options->u.vhdx.has_block_size) {
2122         create_options->u.vhdx.block_size =
2123             ROUND_UP(create_options->u.vhdx.block_size, MiB);
2124 
2125         if (create_options->u.vhdx.block_size == 0) {
2126             create_options->u.vhdx.has_block_size = false;
2127         }
2128         if (create_options->u.vhdx.block_size > VHDX_BLOCK_SIZE_MAX) {
2129             create_options->u.vhdx.block_size = VHDX_BLOCK_SIZE_MAX;
2130         }
2131     }
2132 
2133     /* Create the vhdx image (format layer) */
2134     ret = vhdx_co_create(create_options, errp);
2135 
2136 fail:
2137     qobject_unref(qdict);
2138     bdrv_unref(bs);
2139     qapi_free_BlockdevCreateOptions(create_options);
2140     return ret;
2141 }
2142 
2143 /* If opened r/w, the VHDX driver will automatically replay the log,
2144  * if one is present, inside the vhdx_open() call.
2145  *
2146  * If qemu-img check -r all is called, the image is automatically opened
2147  * r/w and any log has already been replayed, so there is nothing (currently)
2148  * for us to do here
2149  */
2150 static int coroutine_fn vhdx_co_check(BlockDriverState *bs,
2151                                       BdrvCheckResult *result,
2152                                       BdrvCheckMode fix)
2153 {
2154     BDRVVHDXState *s = bs->opaque;
2155 
2156     if (s->log_replayed_on_open) {
2157         result->corruptions_fixed++;
2158     }
2159 
2160     vhdx_check_bat_entries(bs, &result->corruptions);
2161 
2162     return 0;
2163 }
2164 
2165 static int vhdx_has_zero_init(BlockDriverState *bs)
2166 {
2167     BDRVVHDXState *s = bs->opaque;
2168     int state;
2169 
2170     /*
2171      * Check the subformat: Fixed images have all BAT entries present,
2172      * dynamic images have none (right after creation).  It is
2173      * therefore enough to check the first BAT entry.
2174      */
2175     if (!s->bat_entries) {
2176         return 1;
2177     }
2178 
2179     state = s->bat[0] & VHDX_BAT_STATE_BIT_MASK;
2180     if (state == PAYLOAD_BLOCK_FULLY_PRESENT) {
2181         /* Fixed subformat */
2182         return bdrv_has_zero_init(bs->file->bs);
2183     }
2184 
2185     /* Dynamic subformat */
2186     return 1;
2187 }
2188 
2189 static QemuOptsList vhdx_create_opts = {
2190     .name = "vhdx-create-opts",
2191     .head = QTAILQ_HEAD_INITIALIZER(vhdx_create_opts.head),
2192     .desc = {
2193         {
2194            .name = BLOCK_OPT_SIZE,
2195            .type = QEMU_OPT_SIZE,
2196            .help = "Virtual disk size; max of 64TB."
2197        },
2198        {
2199            .name = VHDX_BLOCK_OPT_LOG_SIZE,
2200            .type = QEMU_OPT_SIZE,
2201            .def_value_str = stringify(DEFAULT_LOG_SIZE),
2202            .help = "Log size; min 1MB."
2203        },
2204        {
2205            .name = VHDX_BLOCK_OPT_BLOCK_SIZE,
2206            .type = QEMU_OPT_SIZE,
2207            .def_value_str = stringify(0),
2208            .help = "Block Size; min 1MB, max 256MB. " \
2209                    "0 means auto-calculate based on image size."
2210        },
2211        {
2212            .name = BLOCK_OPT_SUBFMT,
2213            .type = QEMU_OPT_STRING,
2214            .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "\
2215                    "Default is 'dynamic'."
2216        },
2217        {
2218            .name = VHDX_BLOCK_OPT_ZERO,
2219            .type = QEMU_OPT_BOOL,
2220            .help = "Force use of payload blocks of type 'ZERO'. "\
2221                    "Non-standard, but default.  Do not set to 'off' when "\
2222                    "using 'qemu-img convert' with subformat=dynamic."
2223        },
2224        { NULL }
2225     }
2226 };
2227 
2228 static BlockDriver bdrv_vhdx = {
2229     .format_name            = "vhdx",
2230     .instance_size          = sizeof(BDRVVHDXState),
2231     .bdrv_probe             = vhdx_probe,
2232     .bdrv_open              = vhdx_open,
2233     .bdrv_close             = vhdx_close,
2234     .bdrv_reopen_prepare    = vhdx_reopen_prepare,
2235     .bdrv_child_perm        = bdrv_format_default_perms,
2236     .bdrv_co_readv          = vhdx_co_readv,
2237     .bdrv_co_writev         = vhdx_co_writev,
2238     .bdrv_co_create         = vhdx_co_create,
2239     .bdrv_co_create_opts    = vhdx_co_create_opts,
2240     .bdrv_get_info          = vhdx_get_info,
2241     .bdrv_co_check          = vhdx_co_check,
2242     .bdrv_has_zero_init     = vhdx_has_zero_init,
2243 
2244     .create_opts            = &vhdx_create_opts,
2245 };
2246 
2247 static void bdrv_vhdx_init(void)
2248 {
2249     bdrv_register(&bdrv_vhdx);
2250 }
2251 
2252 block_init(bdrv_vhdx_init);
2253