xref: /openbmc/qemu/block/vhdx.c (revision db57fef1)
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     return vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
415 }
416 
417 /* opens the specified header block from the VHDX file header section */
418 static void vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s,
419                               Error **errp)
420 {
421     int ret;
422     VHDXHeader *header1;
423     VHDXHeader *header2;
424     bool h1_valid = false;
425     bool h2_valid = false;
426     uint64_t h1_seq = 0;
427     uint64_t h2_seq = 0;
428     uint8_t *buffer;
429 
430     /* header1 & header2 are freed in vhdx_close() */
431     header1 = qemu_blockalign(bs, sizeof(VHDXHeader));
432     header2 = qemu_blockalign(bs, sizeof(VHDXHeader));
433 
434     buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE);
435 
436     s->headers[0] = header1;
437     s->headers[1] = header2;
438 
439     /* We have to read the whole VHDX_HEADER_SIZE instead of
440      * sizeof(VHDXHeader), because the checksum is over the whole
441      * region */
442     ret = bdrv_pread(bs->file, VHDX_HEADER1_OFFSET, buffer,
443                      VHDX_HEADER_SIZE);
444     if (ret < 0) {
445         goto fail;
446     }
447     /* copy over just the relevant portion that we need */
448     memcpy(header1, buffer, sizeof(VHDXHeader));
449 
450     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
451         vhdx_header_le_import(header1);
452         if (header1->signature == VHDX_HEADER_SIGNATURE &&
453             header1->version == 1) {
454             h1_seq = header1->sequence_number;
455             h1_valid = true;
456         }
457     }
458 
459     ret = bdrv_pread(bs->file, VHDX_HEADER2_OFFSET, buffer,
460                      VHDX_HEADER_SIZE);
461     if (ret < 0) {
462         goto fail;
463     }
464     /* copy over just the relevant portion that we need */
465     memcpy(header2, buffer, sizeof(VHDXHeader));
466 
467     if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4)) {
468         vhdx_header_le_import(header2);
469         if (header2->signature == VHDX_HEADER_SIGNATURE &&
470             header2->version == 1) {
471             h2_seq = header2->sequence_number;
472             h2_valid = true;
473         }
474     }
475 
476     /* If there is only 1 valid header (or no valid headers), we
477      * don't care what the sequence numbers are */
478     if (h1_valid && !h2_valid) {
479         s->curr_header = 0;
480     } else if (!h1_valid && h2_valid) {
481         s->curr_header = 1;
482     } else if (!h1_valid && !h2_valid) {
483         goto fail;
484     } else {
485         /* If both headers are valid, then we choose the active one by the
486          * highest sequence number.  If the sequence numbers are equal, that is
487          * invalid */
488         if (h1_seq > h2_seq) {
489             s->curr_header = 0;
490         } else if (h2_seq > h1_seq) {
491             s->curr_header = 1;
492         } else {
493             /* The Microsoft Disk2VHD tool will create 2 identical
494              * headers, with identical sequence numbers.  If the headers are
495              * identical, don't consider the file corrupt */
496             if (!memcmp(header1, header2, sizeof(VHDXHeader))) {
497                 s->curr_header = 0;
498             } else {
499                 goto fail;
500             }
501         }
502     }
503 
504     vhdx_region_register(s, s->headers[s->curr_header]->log_offset,
505                             s->headers[s->curr_header]->log_length);
506     goto exit;
507 
508 fail:
509     error_setg_errno(errp, -ret, "No valid VHDX header found");
510     qemu_vfree(header1);
511     qemu_vfree(header2);
512     s->headers[0] = NULL;
513     s->headers[1] = NULL;
514 exit:
515     qemu_vfree(buffer);
516 }
517 
518 
519 static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s)
520 {
521     int ret = 0;
522     uint8_t *buffer;
523     int offset = 0;
524     VHDXRegionTableEntry rt_entry;
525     uint32_t i;
526     bool bat_rt_found = false;
527     bool metadata_rt_found = false;
528 
529     /* We have to read the whole 64KB block, because the crc32 is over the
530      * whole block */
531     buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE);
532 
533     ret = bdrv_pread(bs->file, VHDX_REGION_TABLE_OFFSET, buffer,
534                      VHDX_HEADER_BLOCK_SIZE);
535     if (ret < 0) {
536         goto fail;
537     }
538     memcpy(&s->rt, buffer, sizeof(s->rt));
539     offset += sizeof(s->rt);
540 
541     if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4)) {
542         ret = -EINVAL;
543         goto fail;
544     }
545 
546     vhdx_region_header_le_import(&s->rt);
547 
548     if (s->rt.signature != VHDX_REGION_SIGNATURE) {
549         ret = -EINVAL;
550         goto fail;
551     }
552 
553 
554     /* Per spec, maximum region table entry count is 2047 */
555     if (s->rt.entry_count > 2047) {
556         ret = -EINVAL;
557         goto fail;
558     }
559 
560     for (i = 0; i < s->rt.entry_count; i++) {
561         memcpy(&rt_entry, buffer + offset, sizeof(rt_entry));
562         offset += sizeof(rt_entry);
563 
564         vhdx_region_entry_le_import(&rt_entry);
565 
566         /* check for region overlap between these entries, and any
567          * other memory regions in the file */
568         ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length);
569         if (ret < 0) {
570             goto fail;
571         }
572 
573         vhdx_region_register(s, rt_entry.file_offset, rt_entry.length);
574 
575         /* see if we recognize the entry */
576         if (guid_eq(rt_entry.guid, bat_guid)) {
577             /* must be unique; if we have already found it this is invalid */
578             if (bat_rt_found) {
579                 ret = -EINVAL;
580                 goto fail;
581             }
582             bat_rt_found = true;
583             s->bat_rt = rt_entry;
584             continue;
585         }
586 
587         if (guid_eq(rt_entry.guid, metadata_guid)) {
588             /* must be unique; if we have already found it this is invalid */
589             if (metadata_rt_found) {
590                 ret = -EINVAL;
591                 goto fail;
592             }
593             metadata_rt_found = true;
594             s->metadata_rt = rt_entry;
595             continue;
596         }
597 
598         if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) {
599             /* cannot read vhdx file - required region table entry that
600              * we do not understand.  per spec, we must fail to open */
601             ret = -ENOTSUP;
602             goto fail;
603         }
604     }
605 
606     if (!bat_rt_found || !metadata_rt_found) {
607         ret = -EINVAL;
608         goto fail;
609     }
610 
611     ret = 0;
612 
613 fail:
614     qemu_vfree(buffer);
615     return ret;
616 }
617 
618 
619 
620 /* Metadata initial parser
621  *
622  * This loads all the metadata entry fields.  This may cause additional
623  * fields to be processed (e.g. parent locator, etc..).
624  *
625  * There are 5 Metadata items that are always required:
626  *      - File Parameters (block size, has a parent)
627  *      - Virtual Disk Size (size, in bytes, of the virtual drive)
628  *      - Page 83 Data (scsi page 83 guid)
629  *      - Logical Sector Size (logical sector size in bytes, either 512 or
630  *                             4096.  We only support 512 currently)
631  *      - Physical Sector Size (512 or 4096)
632  *
633  * Also, if the File Parameters indicate this is a differencing file,
634  * we must also look for the Parent Locator metadata item.
635  */
636 static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s)
637 {
638     int ret = 0;
639     uint8_t *buffer;
640     int offset = 0;
641     uint32_t i = 0;
642     VHDXMetadataTableEntry md_entry;
643 
644     buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE);
645 
646     ret = bdrv_pread(bs->file, s->metadata_rt.file_offset, buffer,
647                      VHDX_METADATA_TABLE_MAX_SIZE);
648     if (ret < 0) {
649         goto exit;
650     }
651     memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr));
652     offset += sizeof(s->metadata_hdr);
653 
654     vhdx_metadata_header_le_import(&s->metadata_hdr);
655 
656     if (s->metadata_hdr.signature != VHDX_METADATA_SIGNATURE) {
657         ret = -EINVAL;
658         goto exit;
659     }
660 
661     s->metadata_entries.present = 0;
662 
663     if ((s->metadata_hdr.entry_count * sizeof(md_entry)) >
664         (VHDX_METADATA_TABLE_MAX_SIZE - offset)) {
665         ret = -EINVAL;
666         goto exit;
667     }
668 
669     for (i = 0; i < s->metadata_hdr.entry_count; i++) {
670         memcpy(&md_entry, buffer + offset, sizeof(md_entry));
671         offset += sizeof(md_entry);
672 
673         vhdx_metadata_entry_le_import(&md_entry);
674 
675         if (guid_eq(md_entry.item_id, file_param_guid)) {
676             if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) {
677                 ret = -EINVAL;
678                 goto exit;
679             }
680             s->metadata_entries.file_parameters_entry = md_entry;
681             s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT;
682             continue;
683         }
684 
685         if (guid_eq(md_entry.item_id, virtual_size_guid)) {
686             if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) {
687                 ret = -EINVAL;
688                 goto exit;
689             }
690             s->metadata_entries.virtual_disk_size_entry = md_entry;
691             s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT;
692             continue;
693         }
694 
695         if (guid_eq(md_entry.item_id, page83_guid)) {
696             if (s->metadata_entries.present & META_PAGE_83_PRESENT) {
697                 ret = -EINVAL;
698                 goto exit;
699             }
700             s->metadata_entries.page83_data_entry = md_entry;
701             s->metadata_entries.present |= META_PAGE_83_PRESENT;
702             continue;
703         }
704 
705         if (guid_eq(md_entry.item_id, logical_sector_guid)) {
706             if (s->metadata_entries.present &
707                 META_LOGICAL_SECTOR_SIZE_PRESENT) {
708                 ret = -EINVAL;
709                 goto exit;
710             }
711             s->metadata_entries.logical_sector_size_entry = md_entry;
712             s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT;
713             continue;
714         }
715 
716         if (guid_eq(md_entry.item_id, phys_sector_guid)) {
717             if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) {
718                 ret = -EINVAL;
719                 goto exit;
720             }
721             s->metadata_entries.phys_sector_size_entry = md_entry;
722             s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT;
723             continue;
724         }
725 
726         if (guid_eq(md_entry.item_id, parent_locator_guid)) {
727             if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
728                 ret = -EINVAL;
729                 goto exit;
730             }
731             s->metadata_entries.parent_locator_entry = md_entry;
732             s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT;
733             continue;
734         }
735 
736         if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) {
737             /* cannot read vhdx file - required region table entry that
738              * we do not understand.  per spec, we must fail to open */
739             ret = -ENOTSUP;
740             goto exit;
741         }
742     }
743 
744     if (s->metadata_entries.present != META_ALL_PRESENT) {
745         ret = -ENOTSUP;
746         goto exit;
747     }
748 
749     ret = bdrv_pread(bs->file,
750                      s->metadata_entries.file_parameters_entry.offset
751                                          + s->metadata_rt.file_offset,
752                      &s->params,
753                      sizeof(s->params));
754 
755     if (ret < 0) {
756         goto exit;
757     }
758 
759     s->params.block_size = le32_to_cpu(s->params.block_size);
760     s->params.data_bits = le32_to_cpu(s->params.data_bits);
761 
762 
763     /* We now have the file parameters, so we can tell if this is a
764      * differencing file (i.e.. has_parent), is dynamic or fixed
765      * sized (leave_blocks_allocated), and the block size */
766 
767     /* The parent locator required iff the file parameters has_parent set */
768     if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
769         if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) {
770             /* TODO: parse  parent locator fields */
771             ret = -ENOTSUP; /* temp, until differencing files are supported */
772             goto exit;
773         } else {
774             /* if has_parent is set, but there is not parent locator present,
775              * then that is an invalid combination */
776             ret = -EINVAL;
777             goto exit;
778         }
779     }
780 
781     /* determine virtual disk size, logical sector size,
782      * and phys sector size */
783 
784     ret = bdrv_pread(bs->file,
785                      s->metadata_entries.virtual_disk_size_entry.offset
786                                            + s->metadata_rt.file_offset,
787                      &s->virtual_disk_size,
788                      sizeof(uint64_t));
789     if (ret < 0) {
790         goto exit;
791     }
792     ret = bdrv_pread(bs->file,
793                      s->metadata_entries.logical_sector_size_entry.offset
794                                              + s->metadata_rt.file_offset,
795                      &s->logical_sector_size,
796                      sizeof(uint32_t));
797     if (ret < 0) {
798         goto exit;
799     }
800     ret = bdrv_pread(bs->file,
801                      s->metadata_entries.phys_sector_size_entry.offset
802                                           + s->metadata_rt.file_offset,
803                      &s->physical_sector_size,
804                      sizeof(uint32_t));
805     if (ret < 0) {
806         goto exit;
807     }
808 
809     s->virtual_disk_size = le64_to_cpu(s->virtual_disk_size);
810     s->logical_sector_size = le32_to_cpu(s->logical_sector_size);
811     s->physical_sector_size = le32_to_cpu(s->physical_sector_size);
812 
813     if (s->params.block_size < VHDX_BLOCK_SIZE_MIN ||
814         s->params.block_size > VHDX_BLOCK_SIZE_MAX) {
815         ret = -EINVAL;
816         goto exit;
817     }
818 
819     /* only 2 supported sector sizes */
820     if (s->logical_sector_size != 512 && s->logical_sector_size != 4096) {
821         ret = -EINVAL;
822         goto exit;
823     }
824 
825     /* Both block_size and sector_size are guaranteed powers of 2, below.
826        Due to range checks above, s->sectors_per_block can never be < 256 */
827     s->sectors_per_block = s->params.block_size / s->logical_sector_size;
828     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
829                      (uint64_t)s->logical_sector_size /
830                      (uint64_t)s->params.block_size;
831 
832     /* These values are ones we will want to use for division / multiplication
833      * later on, and they are all guaranteed (per the spec) to be powers of 2,
834      * so we can take advantage of that for shift operations during
835      * reads/writes */
836     if (s->logical_sector_size & (s->logical_sector_size - 1)) {
837         ret = -EINVAL;
838         goto exit;
839     }
840     if (s->sectors_per_block & (s->sectors_per_block - 1)) {
841         ret = -EINVAL;
842         goto exit;
843     }
844     if (s->chunk_ratio & (s->chunk_ratio - 1)) {
845         ret = -EINVAL;
846         goto exit;
847     }
848     s->block_size = s->params.block_size;
849     if (s->block_size & (s->block_size - 1)) {
850         ret = -EINVAL;
851         goto exit;
852     }
853 
854     vhdx_set_shift_bits(s);
855 
856     ret = 0;
857 
858 exit:
859     qemu_vfree(buffer);
860     return ret;
861 }
862 
863 /*
864  * Calculate the number of BAT entries, including sector
865  * bitmap entries.
866  */
867 static void vhdx_calc_bat_entries(BDRVVHDXState *s)
868 {
869     uint32_t data_blocks_cnt, bitmap_blocks_cnt;
870 
871     data_blocks_cnt = DIV_ROUND_UP(s->virtual_disk_size, s->block_size);
872     bitmap_blocks_cnt = DIV_ROUND_UP(data_blocks_cnt, s->chunk_ratio);
873 
874     if (s->parent_entries) {
875         s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1);
876     } else {
877         s->bat_entries = data_blocks_cnt +
878                          ((data_blocks_cnt - 1) >> s->chunk_ratio_bits);
879     }
880 
881 }
882 
883 static int vhdx_check_bat_entries(BlockDriverState *bs, int *errcnt)
884 {
885     BDRVVHDXState *s = bs->opaque;
886     int64_t image_file_size = bdrv_getlength(bs->file->bs);
887     uint64_t payblocks = s->chunk_ratio;
888     uint64_t i;
889     int ret = 0;
890 
891     if (image_file_size < 0) {
892         error_report("Could not determinate VHDX image file size.");
893         return image_file_size;
894     }
895 
896     for (i = 0; i < s->bat_entries; i++) {
897         if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) ==
898             PAYLOAD_BLOCK_FULLY_PRESENT) {
899             uint64_t offset = s->bat[i] & VHDX_BAT_FILE_OFF_MASK;
900             /*
901              * Allow that the last block exists only partially. The VHDX spec
902              * states that the image file can only grow in blocksize increments,
903              * but QEMU created images with partial last blocks in the past.
904              */
905             uint32_t block_length = MIN(s->block_size,
906                 bs->total_sectors * BDRV_SECTOR_SIZE - i * s->block_size);
907             /*
908              * Check for BAT entry overflow.
909              */
910             if (offset > INT64_MAX - s->block_size) {
911                 error_report("VHDX BAT entry %" PRIu64 " offset overflow.", i);
912                 ret = -EINVAL;
913                 if (!errcnt) {
914                     break;
915                 }
916                 (*errcnt)++;
917             }
918             /*
919              * Check if fully allocated BAT entries do not reside after
920              * end of the image file.
921              */
922             if (offset >= image_file_size) {
923                 error_report("VHDX BAT entry %" PRIu64 " start offset %" PRIu64
924                              " points after end of file (%" PRIi64 "). Image"
925                              " has probably been truncated.",
926                              i, offset, image_file_size);
927                 ret = -EINVAL;
928                 if (!errcnt) {
929                     break;
930                 }
931                 (*errcnt)++;
932             } else if (offset + block_length > image_file_size) {
933                 error_report("VHDX BAT entry %" PRIu64 " end offset %" PRIu64
934                              " points after end of file (%" PRIi64 "). Image"
935                              " has probably been truncated.",
936                              i, offset + block_length - 1, image_file_size);
937                 ret = -EINVAL;
938                 if (!errcnt) {
939                     break;
940                 }
941                 (*errcnt)++;
942             }
943 
944             /*
945              * verify populated BAT field file offsets against
946              * region table and log entries
947              */
948             if (payblocks--) {
949                 /* payload bat entries */
950                 int ret2;
951                 ret2 = vhdx_region_check(s, offset, s->block_size);
952                 if (ret2 < 0) {
953                     ret = -EINVAL;
954                     if (!errcnt) {
955                         break;
956                     }
957                     (*errcnt)++;
958                 }
959             } else {
960                 payblocks = s->chunk_ratio;
961                 /*
962                  * Once differencing files are supported, verify sector bitmap
963                  * blocks here
964                  */
965             }
966         }
967     }
968 
969     return ret;
970 }
971 
972 static void vhdx_close(BlockDriverState *bs)
973 {
974     BDRVVHDXState *s = bs->opaque;
975     qemu_vfree(s->headers[0]);
976     s->headers[0] = NULL;
977     qemu_vfree(s->headers[1]);
978     s->headers[1] = NULL;
979     qemu_vfree(s->bat);
980     s->bat = NULL;
981     qemu_vfree(s->parent_entries);
982     s->parent_entries = NULL;
983     migrate_del_blocker(s->migration_blocker);
984     error_free(s->migration_blocker);
985     qemu_vfree(s->log.hdr);
986     s->log.hdr = NULL;
987     vhdx_region_unregister_all(s);
988 }
989 
990 static int vhdx_open(BlockDriverState *bs, QDict *options, int flags,
991                      Error **errp)
992 {
993     BDRVVHDXState *s = bs->opaque;
994     int ret = 0;
995     uint32_t i;
996     uint64_t signature;
997     Error *local_err = NULL;
998 
999     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
1000                                BDRV_CHILD_IMAGE, false, errp);
1001     if (!bs->file) {
1002         return -EINVAL;
1003     }
1004 
1005     s->bat = NULL;
1006     s->first_visible_write = true;
1007 
1008     qemu_co_mutex_init(&s->lock);
1009     QLIST_INIT(&s->regions);
1010 
1011     /* validate the file signature */
1012     ret = bdrv_pread(bs->file, 0, &signature, sizeof(uint64_t));
1013     if (ret < 0) {
1014         goto fail;
1015     }
1016     if (memcmp(&signature, "vhdxfile", 8)) {
1017         ret = -EINVAL;
1018         goto fail;
1019     }
1020 
1021     /* This is used for any header updates, for the file_write_guid.
1022      * The spec dictates that a new value should be used for the first
1023      * header update */
1024     vhdx_guid_generate(&s->session_guid);
1025 
1026     vhdx_parse_header(bs, s, &local_err);
1027     if (local_err != NULL) {
1028         error_propagate(errp, local_err);
1029         ret = -EINVAL;
1030         goto fail;
1031     }
1032 
1033     ret = vhdx_parse_log(bs, s, &s->log_replayed_on_open, errp);
1034     if (ret < 0) {
1035         goto fail;
1036     }
1037 
1038     ret = vhdx_open_region_tables(bs, s);
1039     if (ret < 0) {
1040         goto fail;
1041     }
1042 
1043     ret = vhdx_parse_metadata(bs, s);
1044     if (ret < 0) {
1045         goto fail;
1046     }
1047 
1048     s->block_size = s->params.block_size;
1049 
1050     /* the VHDX spec dictates that virtual_disk_size is always a multiple of
1051      * logical_sector_size */
1052     bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits;
1053 
1054     vhdx_calc_bat_entries(s);
1055 
1056     s->bat_offset = s->bat_rt.file_offset;
1057 
1058     if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) {
1059         /* BAT allocation is not large enough for all entries */
1060         ret = -EINVAL;
1061         goto fail;
1062     }
1063 
1064     /* s->bat is freed in vhdx_close() */
1065     s->bat = qemu_try_blockalign(bs->file->bs, s->bat_rt.length);
1066     if (s->bat == NULL) {
1067         ret = -ENOMEM;
1068         goto fail;
1069     }
1070 
1071     ret = bdrv_pread(bs->file, s->bat_offset, s->bat, s->bat_rt.length);
1072     if (ret < 0) {
1073         goto fail;
1074     }
1075 
1076     /* endian convert populated BAT field entires */
1077     for (i = 0; i < s->bat_entries; i++) {
1078         s->bat[i] = le64_to_cpu(s->bat[i]);
1079     }
1080 
1081     if (!(flags & BDRV_O_CHECK)) {
1082         ret = vhdx_check_bat_entries(bs, NULL);
1083         if (ret < 0) {
1084             goto fail;
1085         }
1086     }
1087 
1088     /* Disable migration when VHDX images are used */
1089     error_setg(&s->migration_blocker, "The vhdx format used by node '%s' "
1090                "does not support live migration",
1091                bdrv_get_device_or_node_name(bs));
1092     ret = migrate_add_blocker(s->migration_blocker, errp);
1093     if (ret < 0) {
1094         error_free(s->migration_blocker);
1095         goto fail;
1096     }
1097 
1098     /* TODO: differencing files */
1099 
1100     return 0;
1101 fail:
1102     vhdx_close(bs);
1103     return ret;
1104 }
1105 
1106 static int vhdx_reopen_prepare(BDRVReopenState *state,
1107                                BlockReopenQueue *queue, Error **errp)
1108 {
1109     return 0;
1110 }
1111 
1112 
1113 /*
1114  * Perform sector to block offset translations, to get various
1115  * sector and file offsets into the image.  See VHDXSectorInfo
1116  */
1117 static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num,
1118                                  int nb_sectors, VHDXSectorInfo *sinfo)
1119 {
1120     uint32_t block_offset;
1121 
1122     sinfo->bat_idx = sector_num >> s->sectors_per_block_bits;
1123     /* effectively a modulo - this gives us the offset into the block
1124      * (in sector sizes) for our sector number */
1125     block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits);
1126     /* the chunk ratio gives us the interleaving of the sector
1127      * bitmaps, so we need to advance our page block index by the
1128      * sector bitmaps entry number */
1129     sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits;
1130 
1131     /* the number of sectors we can read/write in this cycle */
1132     sinfo->sectors_avail = s->sectors_per_block - block_offset;
1133 
1134     sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits;
1135 
1136     if (sinfo->sectors_avail > nb_sectors) {
1137         sinfo->sectors_avail = nb_sectors;
1138     }
1139 
1140     sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits;
1141 
1142     sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK;
1143 
1144     sinfo->block_offset = block_offset << s->logical_sector_size_bits;
1145 
1146     /* The file offset must be past the header section, so must be > 0 */
1147     if (sinfo->file_offset == 0) {
1148         return;
1149     }
1150 
1151     /* block offset is the offset in vhdx logical sectors, in
1152      * the payload data block. Convert that to a byte offset
1153      * in the block, and add in the payload data block offset
1154      * in the file, in bytes, to get the final read address */
1155 
1156     sinfo->file_offset += sinfo->block_offset;
1157 }
1158 
1159 
1160 static int vhdx_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1161 {
1162     BDRVVHDXState *s = bs->opaque;
1163 
1164     bdi->cluster_size = s->block_size;
1165 
1166     return 0;
1167 }
1168 
1169 
1170 static coroutine_fn int vhdx_co_readv(BlockDriverState *bs, int64_t sector_num,
1171                                       int nb_sectors, QEMUIOVector *qiov)
1172 {
1173     BDRVVHDXState *s = bs->opaque;
1174     int ret = 0;
1175     VHDXSectorInfo sinfo;
1176     uint64_t bytes_done = 0;
1177     QEMUIOVector hd_qiov;
1178 
1179     qemu_iovec_init(&hd_qiov, qiov->niov);
1180 
1181     qemu_co_mutex_lock(&s->lock);
1182 
1183     while (nb_sectors > 0) {
1184         /* We are a differencing file, so we need to inspect the sector bitmap
1185          * to see if we have the data or not */
1186         if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1187             /* not supported yet */
1188             ret = -ENOTSUP;
1189             goto exit;
1190         } else {
1191             vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1192 
1193             qemu_iovec_reset(&hd_qiov);
1194             qemu_iovec_concat(&hd_qiov, qiov,  bytes_done, sinfo.bytes_avail);
1195 
1196             /* check the payload block state */
1197             switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) {
1198             case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1199             case PAYLOAD_BLOCK_UNDEFINED:
1200             case PAYLOAD_BLOCK_UNMAPPED:
1201             case PAYLOAD_BLOCK_UNMAPPED_v095:
1202             case PAYLOAD_BLOCK_ZERO:
1203                 /* return zero */
1204                 qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail);
1205                 break;
1206             case PAYLOAD_BLOCK_FULLY_PRESENT:
1207                 qemu_co_mutex_unlock(&s->lock);
1208                 ret = bdrv_co_preadv(bs->file, sinfo.file_offset,
1209                                      sinfo.sectors_avail * BDRV_SECTOR_SIZE,
1210                                      &hd_qiov, 0);
1211                 qemu_co_mutex_lock(&s->lock);
1212                 if (ret < 0) {
1213                     goto exit;
1214                 }
1215                 break;
1216             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1217                 /* we don't yet support difference files, fall through
1218                  * to error */
1219             default:
1220                 ret = -EIO;
1221                 goto exit;
1222                 break;
1223             }
1224             nb_sectors -= sinfo.sectors_avail;
1225             sector_num += sinfo.sectors_avail;
1226             bytes_done += sinfo.bytes_avail;
1227         }
1228     }
1229     ret = 0;
1230 exit:
1231     qemu_co_mutex_unlock(&s->lock);
1232     qemu_iovec_destroy(&hd_qiov);
1233     return ret;
1234 }
1235 
1236 /*
1237  * Allocate a new payload block at the end of the file.
1238  *
1239  * Allocation will happen at 1MB alignment inside the file.
1240  *
1241  * If @need_zero is set on entry but not cleared on return, then truncation
1242  * could not guarantee that the new portion reads as zero, and the caller
1243  * will take care of it instead.
1244  *
1245  * Returns the file offset start of the new payload block
1246  */
1247 static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s,
1248                                uint64_t *new_offset, bool *need_zero)
1249 {
1250     int64_t current_len;
1251 
1252     current_len = bdrv_getlength(bs->file->bs);
1253     if (current_len < 0) {
1254         return current_len;
1255     }
1256 
1257     *new_offset = current_len;
1258 
1259     /* per the spec, the address for a block is in units of 1MB */
1260     *new_offset = ROUND_UP(*new_offset, 1 * MiB);
1261     if (*new_offset > INT64_MAX) {
1262         return -EINVAL;
1263     }
1264 
1265     if (*need_zero) {
1266         int ret;
1267 
1268         ret = bdrv_truncate(bs->file, *new_offset + s->block_size, false,
1269                             PREALLOC_MODE_OFF, BDRV_REQ_ZERO_WRITE, NULL);
1270         if (ret != -ENOTSUP) {
1271             *need_zero = false;
1272             return ret;
1273         }
1274     }
1275 
1276     return bdrv_truncate(bs->file, *new_offset + s->block_size, false,
1277                          PREALLOC_MODE_OFF, 0, NULL);
1278 }
1279 
1280 /*
1281  * Update the BAT table entry with the new file offset, and the new entry
1282  * state */
1283 static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s,
1284                                        VHDXSectorInfo *sinfo,
1285                                        uint64_t *bat_entry_le,
1286                                        uint64_t *bat_offset, int state)
1287 {
1288     /* The BAT entry is a uint64, with 44 bits for the file offset in units of
1289      * 1MB, and 3 bits for the block state. */
1290     if ((state == PAYLOAD_BLOCK_ZERO)        ||
1291         (state == PAYLOAD_BLOCK_UNDEFINED)   ||
1292         (state == PAYLOAD_BLOCK_NOT_PRESENT) ||
1293         (state == PAYLOAD_BLOCK_UNMAPPED)) {
1294         s->bat[sinfo->bat_idx]  = 0;  /* For PAYLOAD_BLOCK_ZERO, the
1295                                          FileOffsetMB field is denoted as
1296                                          'reserved' in the v1.0 spec.  If it is
1297                                          non-zero, MS Hyper-V will fail to read
1298                                          the disk image */
1299     } else {
1300         s->bat[sinfo->bat_idx]  = sinfo->file_offset;
1301     }
1302 
1303     s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK;
1304 
1305     *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]);
1306     *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry);
1307 
1308 }
1309 
1310 /* Per the spec, on the first write of guest-visible data to the file the
1311  * data write guid must be updated in the header */
1312 int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s)
1313 {
1314     int ret = 0;
1315     if (s->first_visible_write) {
1316         s->first_visible_write = false;
1317         ret = vhdx_update_headers(bs, s, true, NULL);
1318     }
1319     return ret;
1320 }
1321 
1322 static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num,
1323                                        int nb_sectors, QEMUIOVector *qiov,
1324                                        int flags)
1325 {
1326     int ret = -ENOTSUP;
1327     BDRVVHDXState *s = bs->opaque;
1328     VHDXSectorInfo sinfo;
1329     uint64_t bytes_done = 0;
1330     uint64_t bat_entry = 0;
1331     uint64_t bat_entry_offset = 0;
1332     QEMUIOVector hd_qiov;
1333     struct iovec iov1 = { 0 };
1334     struct iovec iov2 = { 0 };
1335     int sectors_to_write;
1336     int bat_state;
1337     uint64_t bat_prior_offset = 0;
1338     bool bat_update = false;
1339 
1340     assert(!flags);
1341     qemu_iovec_init(&hd_qiov, qiov->niov);
1342 
1343     qemu_co_mutex_lock(&s->lock);
1344 
1345     ret = vhdx_user_visible_write(bs, s);
1346     if (ret < 0) {
1347         goto exit;
1348     }
1349 
1350     while (nb_sectors > 0) {
1351         bool use_zero_buffers = false;
1352         bat_update = false;
1353         if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1354             /* not supported yet */
1355             ret = -ENOTSUP;
1356             goto exit;
1357         } else {
1358             vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1359             sectors_to_write = sinfo.sectors_avail;
1360 
1361             qemu_iovec_reset(&hd_qiov);
1362             /* check the payload block state */
1363             bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK;
1364             switch (bat_state) {
1365             case PAYLOAD_BLOCK_ZERO:
1366                 /* in this case, we need to preserve zero writes for
1367                  * data that is not part of this write, so we must pad
1368                  * the rest of the buffer to zeroes */
1369                 use_zero_buffers = true;
1370                 /* fall through */
1371             case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1372             case PAYLOAD_BLOCK_UNMAPPED:
1373             case PAYLOAD_BLOCK_UNMAPPED_v095:
1374             case PAYLOAD_BLOCK_UNDEFINED:
1375                 bat_prior_offset = sinfo.file_offset;
1376                 ret = vhdx_allocate_block(bs, s, &sinfo.file_offset,
1377                                           &use_zero_buffers);
1378                 if (ret < 0) {
1379                     goto exit;
1380                 }
1381                 /*
1382                  * once we support differencing files, this may also be
1383                  * partially present
1384                  */
1385                 /* update block state to the newly specified state */
1386                 vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1387                                             &bat_entry_offset,
1388                                             PAYLOAD_BLOCK_FULLY_PRESENT);
1389                 bat_update = true;
1390                 /*
1391                  * Since we just allocated a block, file_offset is the
1392                  * beginning of the payload block. It needs to be the
1393                  * write address, which includes the offset into the
1394                  * block, unless the entire block needs to read as
1395                  * zeroes but truncation was not able to provide them,
1396                  * in which case we need to fill in the rest.
1397                  */
1398                 if (!use_zero_buffers) {
1399                     sinfo.file_offset += sinfo.block_offset;
1400                 } else {
1401                     /* zero fill the front, if any */
1402                     if (sinfo.block_offset) {
1403                         iov1.iov_len = sinfo.block_offset;
1404                         iov1.iov_base = qemu_blockalign(bs, iov1.iov_len);
1405                         memset(iov1.iov_base, 0, iov1.iov_len);
1406                         qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0,
1407                                               iov1.iov_len);
1408                         sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS;
1409                     }
1410 
1411                     /* our actual data */
1412                     qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1413                                       sinfo.bytes_avail);
1414 
1415                     /* zero fill the back, if any */
1416                     if ((sinfo.bytes_avail - sinfo.block_offset) <
1417                          s->block_size) {
1418                         iov2.iov_len = s->block_size -
1419                                       (sinfo.bytes_avail + sinfo.block_offset);
1420                         iov2.iov_base = qemu_blockalign(bs, iov2.iov_len);
1421                         memset(iov2.iov_base, 0, iov2.iov_len);
1422                         qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0,
1423                                               iov2.iov_len);
1424                         sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS;
1425                     }
1426                 }
1427 
1428                 /* fall through */
1429             case PAYLOAD_BLOCK_FULLY_PRESENT:
1430                 /* if the file offset address is in the header zone,
1431                  * there is a problem */
1432                 if (sinfo.file_offset < (1 * MiB)) {
1433                     ret = -EFAULT;
1434                     goto error_bat_restore;
1435                 }
1436 
1437                 if (!use_zero_buffers) {
1438                     qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1439                                       sinfo.bytes_avail);
1440                 }
1441                 /* block exists, so we can just overwrite it */
1442                 qemu_co_mutex_unlock(&s->lock);
1443                 ret = bdrv_co_pwritev(bs->file, sinfo.file_offset,
1444                                       sectors_to_write * BDRV_SECTOR_SIZE,
1445                                       &hd_qiov, 0);
1446                 qemu_co_mutex_lock(&s->lock);
1447                 if (ret < 0) {
1448                     goto error_bat_restore;
1449                 }
1450                 break;
1451             case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1452                 /* we don't yet support difference files, fall through
1453                  * to error */
1454             default:
1455                 ret = -EIO;
1456                 goto exit;
1457                 break;
1458             }
1459 
1460             if (bat_update) {
1461                 /* this will update the BAT entry into the log journal, and
1462                  * then flush the log journal out to disk */
1463                 ret =  vhdx_log_write_and_flush(bs, s, &bat_entry,
1464                                                 sizeof(VHDXBatEntry),
1465                                                 bat_entry_offset);
1466                 if (ret < 0) {
1467                     goto exit;
1468                 }
1469             }
1470 
1471             nb_sectors -= sinfo.sectors_avail;
1472             sector_num += sinfo.sectors_avail;
1473             bytes_done += sinfo.bytes_avail;
1474 
1475         }
1476     }
1477 
1478     goto exit;
1479 
1480 error_bat_restore:
1481     if (bat_update) {
1482         /* keep metadata in sync, and restore the bat entry state
1483          * if error. */
1484         sinfo.file_offset = bat_prior_offset;
1485         vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1486                                     &bat_entry_offset, bat_state);
1487     }
1488 exit:
1489     qemu_vfree(iov1.iov_base);
1490     qemu_vfree(iov2.iov_base);
1491     qemu_co_mutex_unlock(&s->lock);
1492     qemu_iovec_destroy(&hd_qiov);
1493     return ret;
1494 }
1495 
1496 
1497 
1498 /*
1499  * Create VHDX Headers
1500  *
1501  * There are 2 headers, and the highest sequence number will represent
1502  * the active header
1503  */
1504 static int vhdx_create_new_headers(BlockBackend *blk, uint64_t image_size,
1505                                    uint32_t log_size)
1506 {
1507     BlockDriverState *bs = blk_bs(blk);
1508     BdrvChild *child;
1509     int ret = 0;
1510     VHDXHeader *hdr = NULL;
1511 
1512     hdr = g_new0(VHDXHeader, 1);
1513 
1514     hdr->signature       = VHDX_HEADER_SIGNATURE;
1515     hdr->sequence_number = g_random_int();
1516     hdr->log_version     = 0;
1517     hdr->version         = 1;
1518     hdr->log_length      = log_size;
1519     hdr->log_offset      = VHDX_HEADER_SECTION_END;
1520     vhdx_guid_generate(&hdr->file_write_guid);
1521     vhdx_guid_generate(&hdr->data_write_guid);
1522 
1523     /* XXX Ugly way to get blk->root, but that's a feature, not a bug. This
1524      * hack makes it obvious that vhdx_write_header() bypasses the BlockBackend
1525      * here, which it really shouldn't be doing. */
1526     child = QLIST_FIRST(&bs->parents);
1527     assert(!QLIST_NEXT(child, next_parent));
1528 
1529     ret = vhdx_write_header(child, hdr, VHDX_HEADER1_OFFSET, false);
1530     if (ret < 0) {
1531         goto exit;
1532     }
1533     hdr->sequence_number++;
1534     ret = vhdx_write_header(child, hdr, VHDX_HEADER2_OFFSET, false);
1535     if (ret < 0) {
1536         goto exit;
1537     }
1538 
1539 exit:
1540     g_free(hdr);
1541     return ret;
1542 }
1543 
1544 #define VHDX_METADATA_ENTRY_BUFFER_SIZE \
1545                                     (sizeof(VHDXFileParameters)               +\
1546                                      sizeof(VHDXVirtualDiskSize)              +\
1547                                      sizeof(VHDXPage83Data)                   +\
1548                                      sizeof(VHDXVirtualDiskLogicalSectorSize) +\
1549                                      sizeof(VHDXVirtualDiskPhysicalSectorSize))
1550 
1551 /*
1552  * Create the Metadata entries.
1553  *
1554  * For more details on the entries, see section 3.5 (pg 29) in the
1555  * VHDX 1.00 specification.
1556  *
1557  * We support 5 metadata entries (all required by spec):
1558  *          File Parameters,
1559  *          Virtual Disk Size,
1560  *          Page 83 Data,
1561  *          Logical Sector Size,
1562  *          Physical Sector Size
1563  *
1564  * The first 64KB of the Metadata section is reserved for the metadata
1565  * header and entries; beyond that, the metadata items themselves reside.
1566  */
1567 static int vhdx_create_new_metadata(BlockBackend *blk,
1568                                     uint64_t image_size,
1569                                     uint32_t block_size,
1570                                     uint32_t sector_size,
1571                                     uint64_t metadata_offset,
1572                                     VHDXImageType type)
1573 {
1574     int ret = 0;
1575     uint32_t offset = 0;
1576     void *buffer = NULL;
1577     void *entry_buffer;
1578     VHDXMetadataTableHeader *md_table;
1579     VHDXMetadataTableEntry  *md_table_entry;
1580 
1581     /* Metadata entries */
1582     VHDXFileParameters     *mt_file_params;
1583     VHDXVirtualDiskSize    *mt_virtual_size;
1584     VHDXPage83Data         *mt_page83;
1585     VHDXVirtualDiskLogicalSectorSize  *mt_log_sector_size;
1586     VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size;
1587 
1588     entry_buffer = g_malloc0(VHDX_METADATA_ENTRY_BUFFER_SIZE);
1589 
1590     mt_file_params = entry_buffer;
1591     offset += sizeof(VHDXFileParameters);
1592     mt_virtual_size = entry_buffer + offset;
1593     offset += sizeof(VHDXVirtualDiskSize);
1594     mt_page83 = entry_buffer + offset;
1595     offset += sizeof(VHDXPage83Data);
1596     mt_log_sector_size = entry_buffer + offset;
1597     offset += sizeof(VHDXVirtualDiskLogicalSectorSize);
1598     mt_phys_sector_size = entry_buffer + offset;
1599 
1600     mt_file_params->block_size = cpu_to_le32(block_size);
1601     if (type == VHDX_TYPE_FIXED) {
1602         mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED;
1603         mt_file_params->data_bits = cpu_to_le32(mt_file_params->data_bits);
1604     }
1605 
1606     vhdx_guid_generate(&mt_page83->page_83_data);
1607     cpu_to_leguids(&mt_page83->page_83_data);
1608     mt_virtual_size->virtual_disk_size        = cpu_to_le64(image_size);
1609     mt_log_sector_size->logical_sector_size   = cpu_to_le32(sector_size);
1610     mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size);
1611 
1612     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1613     md_table = buffer;
1614 
1615     md_table->signature   = VHDX_METADATA_SIGNATURE;
1616     md_table->entry_count = 5;
1617     vhdx_metadata_header_le_export(md_table);
1618 
1619 
1620     /* This will reference beyond the reserved table portion */
1621     offset = 64 * KiB;
1622 
1623     md_table_entry = buffer + sizeof(VHDXMetadataTableHeader);
1624 
1625     md_table_entry[0].item_id = file_param_guid;
1626     md_table_entry[0].offset  = offset;
1627     md_table_entry[0].length  = sizeof(VHDXFileParameters);
1628     md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED;
1629     offset += md_table_entry[0].length;
1630     vhdx_metadata_entry_le_export(&md_table_entry[0]);
1631 
1632     md_table_entry[1].item_id = virtual_size_guid;
1633     md_table_entry[1].offset  = offset;
1634     md_table_entry[1].length  = sizeof(VHDXVirtualDiskSize);
1635     md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1636                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1637     offset += md_table_entry[1].length;
1638     vhdx_metadata_entry_le_export(&md_table_entry[1]);
1639 
1640     md_table_entry[2].item_id = page83_guid;
1641     md_table_entry[2].offset  = offset;
1642     md_table_entry[2].length  = sizeof(VHDXPage83Data);
1643     md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1644                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1645     offset += md_table_entry[2].length;
1646     vhdx_metadata_entry_le_export(&md_table_entry[2]);
1647 
1648     md_table_entry[3].item_id = logical_sector_guid;
1649     md_table_entry[3].offset  = offset;
1650     md_table_entry[3].length  = sizeof(VHDXVirtualDiskLogicalSectorSize);
1651     md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1652                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1653     offset += md_table_entry[3].length;
1654     vhdx_metadata_entry_le_export(&md_table_entry[3]);
1655 
1656     md_table_entry[4].item_id = phys_sector_guid;
1657     md_table_entry[4].offset  = offset;
1658     md_table_entry[4].length  = sizeof(VHDXVirtualDiskPhysicalSectorSize);
1659     md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1660                                    VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1661     vhdx_metadata_entry_le_export(&md_table_entry[4]);
1662 
1663     ret = blk_pwrite(blk, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE, 0);
1664     if (ret < 0) {
1665         goto exit;
1666     }
1667 
1668     ret = blk_pwrite(blk, metadata_offset + (64 * KiB), entry_buffer,
1669                      VHDX_METADATA_ENTRY_BUFFER_SIZE, 0);
1670     if (ret < 0) {
1671         goto exit;
1672     }
1673 
1674 
1675 exit:
1676     g_free(buffer);
1677     g_free(entry_buffer);
1678     return ret;
1679 }
1680 
1681 /* This create the actual BAT itself.  We currently only support
1682  * 'Dynamic' and 'Fixed' image types.
1683  *
1684  *  Dynamic images: default state of the BAT is all zeroes.
1685  *
1686  *  Fixed images: default state of the BAT is fully populated, with
1687  *                file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT.
1688  */
1689 static int vhdx_create_bat(BlockBackend *blk, BDRVVHDXState *s,
1690                            uint64_t image_size, VHDXImageType type,
1691                            bool use_zero_blocks, uint64_t file_offset,
1692                            uint32_t length, Error **errp)
1693 {
1694     int ret = 0;
1695     uint64_t data_file_offset;
1696     uint64_t total_sectors = 0;
1697     uint64_t sector_num = 0;
1698     uint64_t unused;
1699     int block_state;
1700     VHDXSectorInfo sinfo;
1701 
1702     assert(s->bat == NULL);
1703 
1704     /* this gives a data start after BAT/bitmap entries, and well
1705      * past any metadata entries (with a 4 MB buffer for future
1706      * expansion */
1707     data_file_offset = file_offset + length + 5 * MiB;
1708     total_sectors = image_size >> s->logical_sector_size_bits;
1709 
1710     if (type == VHDX_TYPE_DYNAMIC) {
1711         /* All zeroes, so we can just extend the file - the end of the BAT
1712          * is the furthest thing we have written yet */
1713         ret = blk_truncate(blk, data_file_offset, false, PREALLOC_MODE_OFF,
1714                            0, errp);
1715         if (ret < 0) {
1716             goto exit;
1717         }
1718     } else if (type == VHDX_TYPE_FIXED) {
1719         ret = blk_truncate(blk, data_file_offset + image_size, false,
1720                            PREALLOC_MODE_OFF, 0, errp);
1721         if (ret < 0) {
1722             goto exit;
1723         }
1724     } else {
1725         error_setg(errp, "Unsupported image type");
1726         ret = -ENOTSUP;
1727         goto exit;
1728     }
1729 
1730     if (type == VHDX_TYPE_FIXED ||
1731                 use_zero_blocks ||
1732                 bdrv_has_zero_init(blk_bs(blk)) == 0) {
1733         /* for a fixed file, the default BAT entry is not zero */
1734         s->bat = g_try_malloc0(length);
1735         if (length && s->bat == NULL) {
1736             error_setg(errp, "Failed to allocate memory for the BAT");
1737             ret = -ENOMEM;
1738             goto exit;
1739         }
1740         block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT :
1741                                                 PAYLOAD_BLOCK_NOT_PRESENT;
1742         block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state;
1743         /* fill the BAT by emulating sector writes of sectors_per_block size */
1744         while (sector_num < total_sectors) {
1745             vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo);
1746             sinfo.file_offset = data_file_offset +
1747                                 (sector_num << s->logical_sector_size_bits);
1748             sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB);
1749             vhdx_update_bat_table_entry(blk_bs(blk), s, &sinfo, &unused, &unused,
1750                                         block_state);
1751             s->bat[sinfo.bat_idx] = cpu_to_le64(s->bat[sinfo.bat_idx]);
1752             sector_num += s->sectors_per_block;
1753         }
1754         ret = blk_pwrite(blk, file_offset, s->bat, length, 0);
1755         if (ret < 0) {
1756             error_setg_errno(errp, -ret, "Failed to write the BAT");
1757             goto exit;
1758         }
1759     }
1760 
1761 
1762 
1763 exit:
1764     g_free(s->bat);
1765     return ret;
1766 }
1767 
1768 /* Creates the region table header, and region table entries.
1769  * There are 2 supported region table entries: BAT, and Metadata/
1770  *
1771  * As the calculations for the BAT region table are also needed
1772  * to create the BAT itself, we will also cause the BAT to be
1773  * created.
1774  */
1775 static int vhdx_create_new_region_table(BlockBackend *blk,
1776                                         uint64_t image_size,
1777                                         uint32_t block_size,
1778                                         uint32_t sector_size,
1779                                         uint32_t log_size,
1780                                         bool use_zero_blocks,
1781                                         VHDXImageType type,
1782                                         uint64_t *metadata_offset,
1783                                         Error **errp)
1784 {
1785     int ret = 0;
1786     uint32_t offset = 0;
1787     void *buffer = NULL;
1788     uint64_t bat_file_offset;
1789     uint32_t bat_length;
1790     BDRVVHDXState *s = NULL;
1791     VHDXRegionTableHeader *region_table;
1792     VHDXRegionTableEntry *rt_bat;
1793     VHDXRegionTableEntry *rt_metadata;
1794 
1795     assert(metadata_offset != NULL);
1796 
1797     /* Populate enough of the BDRVVHDXState to be able to use the
1798      * pre-existing BAT calculation, translation, and update functions */
1799     s = g_new0(BDRVVHDXState, 1);
1800 
1801     s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
1802                      (uint64_t) sector_size / (uint64_t) block_size;
1803 
1804     s->sectors_per_block = block_size / sector_size;
1805     s->virtual_disk_size = image_size;
1806     s->block_size = block_size;
1807     s->logical_sector_size = sector_size;
1808 
1809     vhdx_set_shift_bits(s);
1810 
1811     vhdx_calc_bat_entries(s);
1812 
1813     /* At this point the VHDX state is populated enough for creation */
1814 
1815     /* a single buffer is used so we can calculate the checksum over the
1816      * entire 64KB block */
1817     buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1818     region_table = buffer;
1819     offset += sizeof(VHDXRegionTableHeader);
1820     rt_bat = buffer + offset;
1821     offset += sizeof(VHDXRegionTableEntry);
1822     rt_metadata  = buffer + offset;
1823 
1824     region_table->signature = VHDX_REGION_SIGNATURE;
1825     region_table->entry_count = 2;   /* BAT and Metadata */
1826 
1827     rt_bat->guid        = bat_guid;
1828     rt_bat->length      = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB);
1829     rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB);
1830     s->bat_offset = rt_bat->file_offset;
1831 
1832     rt_metadata->guid        = metadata_guid;
1833     rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length,
1834                                         MiB);
1835     rt_metadata->length      = 1 * MiB; /* min size, and more than enough */
1836     *metadata_offset = rt_metadata->file_offset;
1837 
1838     bat_file_offset = rt_bat->file_offset;
1839     bat_length = rt_bat->length;
1840 
1841     vhdx_region_header_le_export(region_table);
1842     vhdx_region_entry_le_export(rt_bat);
1843     vhdx_region_entry_le_export(rt_metadata);
1844 
1845     vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE,
1846                          offsetof(VHDXRegionTableHeader, checksum));
1847 
1848 
1849     /* The region table gives us the data we need to create the BAT,
1850      * so do that now */
1851     ret = vhdx_create_bat(blk, s, image_size, type, use_zero_blocks,
1852                           bat_file_offset, bat_length, errp);
1853     if (ret < 0) {
1854         goto exit;
1855     }
1856 
1857     /* Now write out the region headers to disk */
1858     ret = blk_pwrite(blk, VHDX_REGION_TABLE_OFFSET, buffer,
1859                      VHDX_HEADER_BLOCK_SIZE, 0);
1860     if (ret < 0) {
1861         error_setg_errno(errp, -ret, "Failed to write first region table");
1862         goto exit;
1863     }
1864 
1865     ret = blk_pwrite(blk, VHDX_REGION_TABLE2_OFFSET, buffer,
1866                      VHDX_HEADER_BLOCK_SIZE, 0);
1867     if (ret < 0) {
1868         error_setg_errno(errp, -ret, "Failed to write second region table");
1869         goto exit;
1870     }
1871 
1872 exit:
1873     g_free(s);
1874     g_free(buffer);
1875     return ret;
1876 }
1877 
1878 /* We need to create the following elements:
1879  *
1880  *    .-----------------------------------------------------------------.
1881  *    |   (A)    |   (B)    |    (C)    |     (D)       |     (E)       |
1882  *    |  File ID |  Header1 |  Header 2 |  Region Tbl 1 |  Region Tbl 2 |
1883  *    |          |          |           |               |               |
1884  *    .-----------------------------------------------------------------.
1885  *    0         64KB      128KB       192KB           256KB           320KB
1886  *
1887  *
1888  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1889  *    |     (F)     |     (G)       |    (H)    |                        |
1890  *    | Journal Log |  BAT / Bitmap |  Metadata |  .... data ......      |
1891  *    |             |               |           |                        |
1892  *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1893  *   1MB
1894  */
1895 static int coroutine_fn vhdx_co_create(BlockdevCreateOptions *opts,
1896                                        Error **errp)
1897 {
1898     BlockdevCreateOptionsVhdx *vhdx_opts;
1899     BlockBackend *blk = NULL;
1900     BlockDriverState *bs = NULL;
1901 
1902     int ret = 0;
1903     uint64_t image_size;
1904     uint32_t log_size;
1905     uint32_t block_size;
1906     uint64_t signature;
1907     uint64_t metadata_offset;
1908     bool use_zero_blocks = false;
1909 
1910     gunichar2 *creator = NULL;
1911     glong creator_items;
1912     VHDXImageType image_type;
1913 
1914     assert(opts->driver == BLOCKDEV_DRIVER_VHDX);
1915     vhdx_opts = &opts->u.vhdx;
1916 
1917     /* Validate options and set default values */
1918     image_size = vhdx_opts->size;
1919     if (image_size > VHDX_MAX_IMAGE_SIZE) {
1920         error_setg(errp, "Image size too large; max of 64TB");
1921         return -EINVAL;
1922     }
1923 
1924     if (!vhdx_opts->has_log_size) {
1925         log_size = DEFAULT_LOG_SIZE;
1926     } else {
1927         if (vhdx_opts->log_size > UINT32_MAX) {
1928             error_setg(errp, "Log size must be smaller than 4 GB");
1929             return -EINVAL;
1930         }
1931         log_size = vhdx_opts->log_size;
1932     }
1933     if (log_size < MiB || (log_size % MiB) != 0) {
1934         error_setg(errp, "Log size must be a multiple of 1 MB");
1935         return -EINVAL;
1936     }
1937 
1938     if (!vhdx_opts->has_block_state_zero) {
1939         use_zero_blocks = true;
1940     } else {
1941         use_zero_blocks = vhdx_opts->block_state_zero;
1942     }
1943 
1944     if (!vhdx_opts->has_subformat) {
1945         vhdx_opts->subformat = BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC;
1946     }
1947 
1948     switch (vhdx_opts->subformat) {
1949     case BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC:
1950         image_type = VHDX_TYPE_DYNAMIC;
1951         break;
1952     case BLOCKDEV_VHDX_SUBFORMAT_FIXED:
1953         image_type = VHDX_TYPE_FIXED;
1954         break;
1955     default:
1956         g_assert_not_reached();
1957     }
1958 
1959     /* These are pretty arbitrary, and mainly designed to keep the BAT
1960      * size reasonable to load into RAM */
1961     if (vhdx_opts->has_block_size) {
1962         block_size = vhdx_opts->block_size;
1963     } else {
1964         if (image_size > 32 * TiB) {
1965             block_size = 64 * MiB;
1966         } else if (image_size > (uint64_t) 100 * GiB) {
1967             block_size = 32 * MiB;
1968         } else if (image_size > 1 * GiB) {
1969             block_size = 16 * MiB;
1970         } else {
1971             block_size = 8 * MiB;
1972         }
1973     }
1974 
1975     if (block_size < MiB || (block_size % MiB) != 0) {
1976         error_setg(errp, "Block size must be a multiple of 1 MB");
1977         return -EINVAL;
1978     }
1979     if (!is_power_of_2(block_size)) {
1980         error_setg(errp, "Block size must be a power of two");
1981         return -EINVAL;
1982     }
1983     if (block_size > VHDX_BLOCK_SIZE_MAX) {
1984         error_setg(errp, "Block size must not exceed %" PRId64,
1985                    VHDX_BLOCK_SIZE_MAX);
1986         return -EINVAL;
1987     }
1988 
1989     /* Create BlockBackend to write to the image */
1990     bs = bdrv_open_blockdev_ref(vhdx_opts->file, errp);
1991     if (bs == NULL) {
1992         return -EIO;
1993     }
1994 
1995     blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
1996                           errp);
1997     if (!blk) {
1998         ret = -EPERM;
1999         goto delete_and_exit;
2000     }
2001     blk_set_allow_write_beyond_eof(blk, true);
2002 
2003     /* Create (A) */
2004 
2005     /* The creator field is optional, but may be useful for
2006      * debugging / diagnostics */
2007     creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL,
2008                               &creator_items, NULL);
2009     signature = cpu_to_le64(VHDX_FILE_SIGNATURE);
2010     ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature),
2011                      0);
2012     if (ret < 0) {
2013         error_setg_errno(errp, -ret, "Failed to write file signature");
2014         goto delete_and_exit;
2015     }
2016     if (creator) {
2017         ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET + sizeof(signature),
2018                          creator, creator_items * sizeof(gunichar2), 0);
2019         if (ret < 0) {
2020             error_setg_errno(errp, -ret, "Failed to write creator field");
2021             goto delete_and_exit;
2022         }
2023     }
2024 
2025 
2026     /* Creates (B),(C) */
2027     ret = vhdx_create_new_headers(blk, image_size, log_size);
2028     if (ret < 0) {
2029         error_setg_errno(errp, -ret, "Failed to write image headers");
2030         goto delete_and_exit;
2031     }
2032 
2033     /* Creates (D),(E),(G) explicitly. (F) created as by-product */
2034     ret = vhdx_create_new_region_table(blk, image_size, block_size, 512,
2035                                        log_size, use_zero_blocks, image_type,
2036                                        &metadata_offset, errp);
2037     if (ret < 0) {
2038         goto delete_and_exit;
2039     }
2040 
2041     /* Creates (H) */
2042     ret = vhdx_create_new_metadata(blk, image_size, block_size, 512,
2043                                    metadata_offset, image_type);
2044     if (ret < 0) {
2045         error_setg_errno(errp, -ret, "Failed to initialize metadata");
2046         goto delete_and_exit;
2047     }
2048 
2049     ret = 0;
2050 delete_and_exit:
2051     blk_unref(blk);
2052     bdrv_unref(bs);
2053     g_free(creator);
2054     return ret;
2055 }
2056 
2057 static int coroutine_fn vhdx_co_create_opts(BlockDriver *drv,
2058                                             const char *filename,
2059                                             QemuOpts *opts,
2060                                             Error **errp)
2061 {
2062     BlockdevCreateOptions *create_options = NULL;
2063     QDict *qdict;
2064     Visitor *v;
2065     BlockDriverState *bs = NULL;
2066     int ret;
2067 
2068     static const QDictRenames opt_renames[] = {
2069         { VHDX_BLOCK_OPT_LOG_SIZE,      "log-size" },
2070         { VHDX_BLOCK_OPT_BLOCK_SIZE,    "block-size" },
2071         { VHDX_BLOCK_OPT_ZERO,          "block-state-zero" },
2072         { NULL, NULL },
2073     };
2074 
2075     /* Parse options and convert legacy syntax */
2076     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vhdx_create_opts, true);
2077 
2078     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
2079         ret = -EINVAL;
2080         goto fail;
2081     }
2082 
2083     /* Create and open the file (protocol layer) */
2084     ret = bdrv_create_file(filename, opts, errp);
2085     if (ret < 0) {
2086         goto fail;
2087     }
2088 
2089     bs = bdrv_open(filename, NULL, NULL,
2090                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
2091     if (bs == NULL) {
2092         ret = -EIO;
2093         goto fail;
2094     }
2095 
2096     /* Now get the QAPI type BlockdevCreateOptions */
2097     qdict_put_str(qdict, "driver", "vhdx");
2098     qdict_put_str(qdict, "file", bs->node_name);
2099 
2100     v = qobject_input_visitor_new_flat_confused(qdict, errp);
2101     if (!v) {
2102         ret = -EINVAL;
2103         goto fail;
2104     }
2105 
2106     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
2107     visit_free(v);
2108     if (!create_options) {
2109         ret = -EINVAL;
2110         goto fail;
2111     }
2112 
2113     /* Silently round up sizes:
2114      * The image size is rounded to 512 bytes. Make the block and log size
2115      * close to what was specified, but must be at least 1MB, and a multiple of
2116      * 1 MB. Also respect VHDX_BLOCK_SIZE_MAX for block sizes. block_size = 0
2117      * means auto, which is represented by a missing key in QAPI. */
2118     assert(create_options->driver == BLOCKDEV_DRIVER_VHDX);
2119     create_options->u.vhdx.size =
2120         ROUND_UP(create_options->u.vhdx.size, BDRV_SECTOR_SIZE);
2121 
2122     if (create_options->u.vhdx.has_log_size) {
2123         create_options->u.vhdx.log_size =
2124             ROUND_UP(create_options->u.vhdx.log_size, MiB);
2125     }
2126     if (create_options->u.vhdx.has_block_size) {
2127         create_options->u.vhdx.block_size =
2128             ROUND_UP(create_options->u.vhdx.block_size, MiB);
2129 
2130         if (create_options->u.vhdx.block_size == 0) {
2131             create_options->u.vhdx.has_block_size = false;
2132         }
2133         if (create_options->u.vhdx.block_size > VHDX_BLOCK_SIZE_MAX) {
2134             create_options->u.vhdx.block_size = VHDX_BLOCK_SIZE_MAX;
2135         }
2136     }
2137 
2138     /* Create the vhdx image (format layer) */
2139     ret = vhdx_co_create(create_options, errp);
2140 
2141 fail:
2142     qobject_unref(qdict);
2143     bdrv_unref(bs);
2144     qapi_free_BlockdevCreateOptions(create_options);
2145     return ret;
2146 }
2147 
2148 /* If opened r/w, the VHDX driver will automatically replay the log,
2149  * if one is present, inside the vhdx_open() call.
2150  *
2151  * If qemu-img check -r all is called, the image is automatically opened
2152  * r/w and any log has already been replayed, so there is nothing (currently)
2153  * for us to do here
2154  */
2155 static int coroutine_fn vhdx_co_check(BlockDriverState *bs,
2156                                       BdrvCheckResult *result,
2157                                       BdrvCheckMode fix)
2158 {
2159     BDRVVHDXState *s = bs->opaque;
2160 
2161     if (s->log_replayed_on_open) {
2162         result->corruptions_fixed++;
2163     }
2164 
2165     vhdx_check_bat_entries(bs, &result->corruptions);
2166 
2167     return 0;
2168 }
2169 
2170 static int vhdx_has_zero_init(BlockDriverState *bs)
2171 {
2172     BDRVVHDXState *s = bs->opaque;
2173     int state;
2174 
2175     /*
2176      * Check the subformat: Fixed images have all BAT entries present,
2177      * dynamic images have none (right after creation).  It is
2178      * therefore enough to check the first BAT entry.
2179      */
2180     if (!s->bat_entries) {
2181         return 1;
2182     }
2183 
2184     state = s->bat[0] & VHDX_BAT_STATE_BIT_MASK;
2185     if (state == PAYLOAD_BLOCK_FULLY_PRESENT) {
2186         /* Fixed subformat */
2187         return bdrv_has_zero_init(bs->file->bs);
2188     }
2189 
2190     /* Dynamic subformat */
2191     return 1;
2192 }
2193 
2194 static QemuOptsList vhdx_create_opts = {
2195     .name = "vhdx-create-opts",
2196     .head = QTAILQ_HEAD_INITIALIZER(vhdx_create_opts.head),
2197     .desc = {
2198         {
2199            .name = BLOCK_OPT_SIZE,
2200            .type = QEMU_OPT_SIZE,
2201            .help = "Virtual disk size; max of 64TB."
2202        },
2203        {
2204            .name = VHDX_BLOCK_OPT_LOG_SIZE,
2205            .type = QEMU_OPT_SIZE,
2206            .def_value_str = stringify(DEFAULT_LOG_SIZE),
2207            .help = "Log size; min 1MB."
2208        },
2209        {
2210            .name = VHDX_BLOCK_OPT_BLOCK_SIZE,
2211            .type = QEMU_OPT_SIZE,
2212            .def_value_str = stringify(0),
2213            .help = "Block Size; min 1MB, max 256MB. "
2214                    "0 means auto-calculate based on image size."
2215        },
2216        {
2217            .name = BLOCK_OPT_SUBFMT,
2218            .type = QEMU_OPT_STRING,
2219            .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "
2220                    "Default is 'dynamic'."
2221        },
2222        {
2223            .name = VHDX_BLOCK_OPT_ZERO,
2224            .type = QEMU_OPT_BOOL,
2225            .help = "Force use of payload blocks of type 'ZERO'. "
2226                    "Non-standard, but default.  Do not set to 'off' when "
2227                    "using 'qemu-img convert' with subformat=dynamic."
2228        },
2229        { NULL }
2230     }
2231 };
2232 
2233 static BlockDriver bdrv_vhdx = {
2234     .format_name            = "vhdx",
2235     .instance_size          = sizeof(BDRVVHDXState),
2236     .bdrv_probe             = vhdx_probe,
2237     .bdrv_open              = vhdx_open,
2238     .bdrv_close             = vhdx_close,
2239     .bdrv_reopen_prepare    = vhdx_reopen_prepare,
2240     .bdrv_child_perm        = bdrv_default_perms,
2241     .bdrv_co_readv          = vhdx_co_readv,
2242     .bdrv_co_writev         = vhdx_co_writev,
2243     .bdrv_co_create         = vhdx_co_create,
2244     .bdrv_co_create_opts    = vhdx_co_create_opts,
2245     .bdrv_get_info          = vhdx_get_info,
2246     .bdrv_co_check          = vhdx_co_check,
2247     .bdrv_has_zero_init     = vhdx_has_zero_init,
2248 
2249     .is_format              = true,
2250     .create_opts            = &vhdx_create_opts,
2251 };
2252 
2253 static void bdrv_vhdx_init(void)
2254 {
2255     bdrv_register(&bdrv_vhdx);
2256 }
2257 
2258 block_init(bdrv_vhdx_init);
2259