xref: /openbmc/qemu/block/vhdx.c (revision 9edc6313da34699ebd2bae4573ea22339b26450a)
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 "qemu/memalign.h"
29  #include "vhdx.h"
30  #include "migration/blocker.h"
31  #include "qemu/uuid.h"
32  #include "qapi/qmp/qdict.h"
33  #include "qapi/qobject-input-visitor.h"
34  #include "qapi/qapi-visit-block-core.h"
35  
36  /* Options for VHDX creation */
37  
38  #define VHDX_BLOCK_OPT_LOG_SIZE   "log_size"
39  #define VHDX_BLOCK_OPT_BLOCK_SIZE "block_size"
40  #define VHDX_BLOCK_OPT_ZERO "block_state_zero"
41  
42  typedef enum VHDXImageType {
43      VHDX_TYPE_DYNAMIC = 0,
44      VHDX_TYPE_FIXED,
45      VHDX_TYPE_DIFFERENCING,   /* Currently unsupported */
46  } VHDXImageType;
47  
48  static QemuOptsList vhdx_create_opts;
49  
50  /* Several metadata and region table data entries are identified by
51   * guids in  a MS-specific GUID format. */
52  
53  
54  /* ------- Known Region Table GUIDs ---------------------- */
55  static const MSGUID bat_guid =      { .data1 = 0x2dc27766,
56                                        .data2 = 0xf623,
57                                        .data3 = 0x4200,
58                                        .data4 = { 0x9d, 0x64, 0x11, 0x5e,
59                                                   0x9b, 0xfd, 0x4a, 0x08} };
60  
61  static const MSGUID metadata_guid = { .data1 = 0x8b7ca206,
62                                        .data2 = 0x4790,
63                                        .data3 = 0x4b9a,
64                                        .data4 = { 0xb8, 0xfe, 0x57, 0x5f,
65                                                   0x05, 0x0f, 0x88, 0x6e} };
66  
67  
68  
69  /* ------- Known Metadata Entry GUIDs ---------------------- */
70  static const MSGUID file_param_guid =   { .data1 = 0xcaa16737,
71                                            .data2 = 0xfa36,
72                                            .data3 = 0x4d43,
73                                            .data4 = { 0xb3, 0xb6, 0x33, 0xf0,
74                                                       0xaa, 0x44, 0xe7, 0x6b} };
75  
76  static const MSGUID virtual_size_guid = { .data1 = 0x2FA54224,
77                                            .data2 = 0xcd1b,
78                                            .data3 = 0x4876,
79                                            .data4 = { 0xb2, 0x11, 0x5d, 0xbe,
80                                                       0xd8, 0x3b, 0xf4, 0xb8} };
81  
82  static const MSGUID page83_guid =       { .data1 = 0xbeca12ab,
83                                            .data2 = 0xb2e6,
84                                            .data3 = 0x4523,
85                                            .data4 = { 0x93, 0xef, 0xc3, 0x09,
86                                                       0xe0, 0x00, 0xc7, 0x46} };
87  
88  
89  static const MSGUID phys_sector_guid =  { .data1 = 0xcda348c7,
90                                            .data2 = 0x445d,
91                                            .data3 = 0x4471,
92                                            .data4 = { 0x9c, 0xc9, 0xe9, 0x88,
93                                                       0x52, 0x51, 0xc5, 0x56} };
94  
95  static const MSGUID parent_locator_guid = { .data1 = 0xa8d35f2d,
96                                              .data2 = 0xb30b,
97                                              .data3 = 0x454d,
98                                              .data4 = { 0xab, 0xf7, 0xd3,
99                                                         0xd8, 0x48, 0x34,
100                                                         0xab, 0x0c} };
101  
102  static const MSGUID logical_sector_guid = { .data1 = 0x8141bf1d,
103                                              .data2 = 0xa96f,
104                                              .data3 = 0x4709,
105                                              .data4 = { 0xba, 0x47, 0xf2,
106                                                         0x33, 0xa8, 0xfa,
107                                                         0xab, 0x5f} };
108  
109  /* Each parent type must have a valid GUID; this is for parent images
110   * of type 'VHDX'.  If we were to allow e.g. a QCOW2 parent, we would
111   * need to make up our own QCOW2 GUID type */
112  static const MSGUID parent_vhdx_guid __attribute__((unused))
113                                       = { .data1 = 0xb04aefb7,
114                                           .data2 = 0xd19e,
115                                           .data3 = 0x4a81,
116                                           .data4 = { 0xb7, 0x89, 0x25, 0xb8,
117                                                      0xe9, 0x44, 0x59, 0x13} };
118  
119  
120  #define META_FILE_PARAMETER_PRESENT      0x01
121  #define META_VIRTUAL_DISK_SIZE_PRESENT   0x02
122  #define META_PAGE_83_PRESENT             0x04
123  #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08
124  #define META_PHYS_SECTOR_SIZE_PRESENT    0x10
125  #define META_PARENT_LOCATOR_PRESENT      0x20
126  
127  #define META_ALL_PRESENT    \
128      (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \
129       META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \
130       META_PHYS_SECTOR_SIZE_PRESENT)
131  
132  
133  typedef struct VHDXSectorInfo {
134      uint32_t bat_idx;       /* BAT entry index */
135      uint32_t sectors_avail; /* sectors available in payload block */
136      uint32_t bytes_left;    /* bytes left in the block after data to r/w */
137      uint32_t bytes_avail;   /* bytes available in payload block */
138      uint64_t file_offset;   /* absolute offset in bytes, in file */
139      uint64_t block_offset;  /* block offset, in bytes */
140  } VHDXSectorInfo;
141  
142  /* Calculates new checksum.
143   *
144   * Zero is substituted during crc calculation for the original crc field
145   * crc_offset: byte offset in buf of the buffer crc
146   * buf: buffer pointer
147   * size: size of buffer (must be > crc_offset+4)
148   *
149   * Note: The buffer should have all multi-byte data in little-endian format,
150   *       and the resulting checksum is in little endian format.
151   */
152  uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset)
153  {
154      uint32_t crc;
155  
156      assert(buf != NULL);
157      assert(size > (crc_offset + sizeof(crc)));
158  
159      memset(buf + crc_offset, 0, sizeof(crc));
160      crc =  crc32c(0xffffffff, buf, size);
161      crc = cpu_to_le32(crc);
162      memcpy(buf + crc_offset, &crc, sizeof(crc));
163  
164      return crc;
165  }
166  
167  uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size,
168                              int crc_offset)
169  {
170      uint32_t crc_new;
171      uint32_t crc_orig;
172      assert(buf != NULL);
173  
174      if (crc_offset > 0) {
175          memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
176          memset(buf + crc_offset, 0, sizeof(crc_orig));
177      }
178  
179      crc_new = crc32c(crc, buf, size);
180      if (crc_offset > 0) {
181          memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig));
182      }
183  
184      return crc_new;
185  }
186  
187  /* Validates the checksum of the buffer, with an in-place CRC.
188   *
189   * Zero is substituted during crc calculation for the original crc field,
190   * and the crc field is restored afterwards.  But the buffer will be modified
191   * during the calculation, so this may not be not suitable for multi-threaded
192   * use.
193   *
194   * crc_offset: byte offset in buf of the buffer crc
195   * buf: buffer pointer
196   * size: size of buffer (must be > crc_offset+4)
197   *
198   * returns true if checksum is valid, false otherwise
199   */
200  bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset)
201  {
202      uint32_t crc_orig;
203      uint32_t crc;
204  
205      assert(buf != NULL);
206      assert(size > (crc_offset + 4));
207  
208      memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig));
209      crc_orig = le32_to_cpu(crc_orig);
210  
211      crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset);
212  
213      return crc == crc_orig;
214  }
215  
216  
217  /*
218   * This generates a UUID that is compliant with the MS GUIDs used
219   * in the VHDX spec (and elsewhere).
220   */
221  void vhdx_guid_generate(MSGUID *guid)
222  {
223      QemuUUID uuid;
224      assert(guid != NULL);
225  
226      qemu_uuid_generate(&uuid);
227      memcpy(guid, &uuid, sizeof(MSGUID));
228  }
229  
230  /* Check for region overlaps inside the VHDX image */
231  static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length)
232  {
233      int ret = 0;
234      uint64_t end;
235      VHDXRegionEntry *r;
236  
237      end = start + length;
238      QLIST_FOREACH(r, &s->regions, entries) {
239          if (!((start >= r->end) || (end <= r->start))) {
240              error_report("VHDX region %" PRIu64 "-%" PRIu64 " overlaps with "
241                           "region %" PRIu64 "-%." PRIu64, start, end, r->start,
242                           r->end);
243              ret = -EINVAL;
244              goto exit;
245          }
246      }
247  
248  exit:
249      return ret;
250  }
251  
252  /* Register a region for future checks */
253  static void vhdx_region_register(BDRVVHDXState *s,
254                                   uint64_t start, uint64_t length)
255  {
256      VHDXRegionEntry *r;
257  
258      r = g_malloc0(sizeof(*r));
259  
260      r->start = start;
261      r->end = start + length;
262  
263      QLIST_INSERT_HEAD(&s->regions, r, entries);
264  }
265  
266  /* Free all registered regions */
267  static void vhdx_region_unregister_all(BDRVVHDXState *s)
268  {
269      VHDXRegionEntry *r, *r_next;
270  
271      QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) {
272          QLIST_REMOVE(r, entries);
273          g_free(r);
274      }
275  }
276  
277  static void vhdx_set_shift_bits(BDRVVHDXState *s)
278  {
279      s->logical_sector_size_bits = ctz32(s->logical_sector_size);
280      s->sectors_per_block_bits =   ctz32(s->sectors_per_block);
281      s->chunk_ratio_bits =         ctz64(s->chunk_ratio);
282      s->block_size_bits =          ctz32(s->block_size);
283  }
284  
285  /*
286   * Per the MS VHDX Specification, for every VHDX file:
287   *      - The header section is fixed size - 1 MB
288   *      - The header section is always the first "object"
289   *      - The first 64KB of the header is the File Identifier
290   *      - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile")
291   *      - The following 512 bytes constitute a UTF-16 string identifiying the
292   *        software that created the file, and is optional and diagnostic only.
293   *
294   *  Therefore, we probe by looking for the vhdxfile signature "vhdxfile"
295   */
296  static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename)
297  {
298      if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) {
299          return 100;
300      }
301      return 0;
302  }
303  
304  /*
305   * Writes the header to the specified offset.
306   *
307   * This will optionally read in buffer data from disk (otherwise zero-fill),
308   * and then update the header checksum.  Header is converted to proper
309   * endianness before being written to the specified file offset
310   */
311  static int vhdx_write_header(BdrvChild *file, VHDXHeader *hdr,
312                               uint64_t offset, bool read)
313  {
314      BlockDriverState *bs_file = file->bs;
315      uint8_t *buffer = NULL;
316      int ret;
317      VHDXHeader *header_le;
318  
319      assert(bs_file != NULL);
320      assert(hdr != NULL);
321  
322      /* the header checksum is not over just the packed size of VHDXHeader,
323       * but rather over the entire 'reserved' range for the header, which is
324       * 4KB (VHDX_HEADER_SIZE). */
325  
326      buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE);
327      if (read) {
328          /* if true, we can't assume the extra reserved bytes are 0 */
329          ret = bdrv_pread(file, offset, buffer, VHDX_HEADER_SIZE);
330          if (ret < 0) {
331              goto exit;
332          }
333      } else {
334          memset(buffer, 0, VHDX_HEADER_SIZE);
335      }
336  
337      /* overwrite the actual VHDXHeader portion */
338      header_le = (VHDXHeader *)buffer;
339      memcpy(header_le, hdr, sizeof(VHDXHeader));
340      vhdx_header_le_export(hdr, header_le);
341      vhdx_update_checksum(buffer, VHDX_HEADER_SIZE,
342                           offsetof(VHDXHeader, checksum));
343      ret = bdrv_pwrite_sync(file, offset, header_le, sizeof(VHDXHeader));
344  
345  exit:
346      qemu_vfree(buffer);
347      return ret;
348  }
349  
350  /* Update the VHDX headers
351   *
352   * This follows the VHDX spec procedures for header updates.
353   *
354   *  - non-current header is updated with largest sequence number
355   */
356  static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s,
357                                bool generate_data_write_guid, MSGUID *log_guid)
358  {
359      int ret = 0;
360      int hdr_idx = 0;
361      uint64_t header_offset = VHDX_HEADER1_OFFSET;
362  
363      VHDXHeader *active_header;
364      VHDXHeader *inactive_header;
365  
366      /* operate on the non-current header */
367      if (s->curr_header == 0) {
368          hdr_idx = 1;
369          header_offset = VHDX_HEADER2_OFFSET;
370      }
371  
372      active_header   = s->headers[s->curr_header];
373      inactive_header = s->headers[hdr_idx];
374  
375      inactive_header->sequence_number = active_header->sequence_number + 1;
376  
377      /* a new file guid must be generated before any file write, including
378       * headers */
379      inactive_header->file_write_guid = s->session_guid;
380  
381      /* a new data guid only needs to be generated before any guest-visible
382       * writes (i.e. something observable via virtual disk read) */
383      if (generate_data_write_guid) {
384          vhdx_guid_generate(&inactive_header->data_write_guid);
385      }
386  
387      /* update the log guid if present */
388      if (log_guid) {
389          inactive_header->log_guid = *log_guid;
390      }
391  
392      ret = vhdx_write_header(bs->file, inactive_header, header_offset, true);
393      if (ret < 0) {
394          goto exit;
395      }
396      s->curr_header = hdr_idx;
397  
398  exit:
399      return ret;
400  }
401  
402  /*
403   * The VHDX spec calls for header updates to be performed twice, so that both
404   * the current and non-current header have valid info
405   */
406  int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s,
407                          bool generate_data_write_guid, MSGUID *log_guid)
408  {
409      int ret;
410  
411      ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
412      if (ret < 0) {
413          return ret;
414      }
415      return vhdx_update_header(bs, s, generate_data_write_guid, log_guid);
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      /* Currently we only support 512 */
821      if (s->logical_sector_size != 512) {
822          ret = -ENOTSUP;
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_of_bds,
1001                                 BDRV_CHILD_IMAGE, 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, errp);
1094      if (ret < 0) {
1095          error_free(s->migration_blocker);
1096          goto fail;
1097      }
1098  
1099      /* TODO: differencing files */
1100  
1101      return 0;
1102  fail:
1103      vhdx_close(bs);
1104      return ret;
1105  }
1106  
1107  static int vhdx_reopen_prepare(BDRVReopenState *state,
1108                                 BlockReopenQueue *queue, Error **errp)
1109  {
1110      return 0;
1111  }
1112  
1113  
1114  /*
1115   * Perform sector to block offset translations, to get various
1116   * sector and file offsets into the image.  See VHDXSectorInfo
1117   */
1118  static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num,
1119                                   int nb_sectors, VHDXSectorInfo *sinfo)
1120  {
1121      uint32_t block_offset;
1122  
1123      sinfo->bat_idx = sector_num >> s->sectors_per_block_bits;
1124      /* effectively a modulo - this gives us the offset into the block
1125       * (in sector sizes) for our sector number */
1126      block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits);
1127      /* the chunk ratio gives us the interleaving of the sector
1128       * bitmaps, so we need to advance our page block index by the
1129       * sector bitmaps entry number */
1130      sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits;
1131  
1132      /* the number of sectors we can read/write in this cycle */
1133      sinfo->sectors_avail = s->sectors_per_block - block_offset;
1134  
1135      sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits;
1136  
1137      if (sinfo->sectors_avail > nb_sectors) {
1138          sinfo->sectors_avail = nb_sectors;
1139      }
1140  
1141      sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits;
1142  
1143      sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK;
1144  
1145      sinfo->block_offset = block_offset << s->logical_sector_size_bits;
1146  
1147      /* The file offset must be past the header section, so must be > 0 */
1148      if (sinfo->file_offset == 0) {
1149          return;
1150      }
1151  
1152      /* block offset is the offset in vhdx logical sectors, in
1153       * the payload data block. Convert that to a byte offset
1154       * in the block, and add in the payload data block offset
1155       * in the file, in bytes, to get the final read address */
1156  
1157      sinfo->file_offset += sinfo->block_offset;
1158  }
1159  
1160  
1161  static int vhdx_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1162  {
1163      BDRVVHDXState *s = bs->opaque;
1164  
1165      bdi->cluster_size = s->block_size;
1166  
1167      return 0;
1168  }
1169  
1170  
1171  static coroutine_fn int vhdx_co_readv(BlockDriverState *bs, int64_t sector_num,
1172                                        int nb_sectors, QEMUIOVector *qiov)
1173  {
1174      BDRVVHDXState *s = bs->opaque;
1175      int ret = 0;
1176      VHDXSectorInfo sinfo;
1177      uint64_t bytes_done = 0;
1178      QEMUIOVector hd_qiov;
1179  
1180      qemu_iovec_init(&hd_qiov, qiov->niov);
1181  
1182      qemu_co_mutex_lock(&s->lock);
1183  
1184      while (nb_sectors > 0) {
1185          /* We are a differencing file, so we need to inspect the sector bitmap
1186           * to see if we have the data or not */
1187          if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1188              /* not supported yet */
1189              ret = -ENOTSUP;
1190              goto exit;
1191          } else {
1192              vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1193  
1194              qemu_iovec_reset(&hd_qiov);
1195              qemu_iovec_concat(&hd_qiov, qiov,  bytes_done, sinfo.bytes_avail);
1196  
1197              /* check the payload block state */
1198              switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) {
1199              case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1200              case PAYLOAD_BLOCK_UNDEFINED:
1201              case PAYLOAD_BLOCK_UNMAPPED:
1202              case PAYLOAD_BLOCK_UNMAPPED_v095:
1203              case PAYLOAD_BLOCK_ZERO:
1204                  /* return zero */
1205                  qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail);
1206                  break;
1207              case PAYLOAD_BLOCK_FULLY_PRESENT:
1208                  qemu_co_mutex_unlock(&s->lock);
1209                  ret = bdrv_co_preadv(bs->file, sinfo.file_offset,
1210                                       sinfo.sectors_avail * BDRV_SECTOR_SIZE,
1211                                       &hd_qiov, 0);
1212                  qemu_co_mutex_lock(&s->lock);
1213                  if (ret < 0) {
1214                      goto exit;
1215                  }
1216                  break;
1217              case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1218                  /* we don't yet support difference files, fall through
1219                   * to error */
1220              default:
1221                  ret = -EIO;
1222                  goto exit;
1223                  break;
1224              }
1225              nb_sectors -= sinfo.sectors_avail;
1226              sector_num += sinfo.sectors_avail;
1227              bytes_done += sinfo.bytes_avail;
1228          }
1229      }
1230      ret = 0;
1231  exit:
1232      qemu_co_mutex_unlock(&s->lock);
1233      qemu_iovec_destroy(&hd_qiov);
1234      return ret;
1235  }
1236  
1237  /*
1238   * Allocate a new payload block at the end of the file.
1239   *
1240   * Allocation will happen at 1MB alignment inside the file.
1241   *
1242   * If @need_zero is set on entry but not cleared on return, then truncation
1243   * could not guarantee that the new portion reads as zero, and the caller
1244   * will take care of it instead.
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, bool *need_zero)
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      if (*need_zero) {
1267          int ret;
1268  
1269          ret = bdrv_truncate(bs->file, *new_offset + s->block_size, false,
1270                              PREALLOC_MODE_OFF, BDRV_REQ_ZERO_WRITE, NULL);
1271          if (ret != -ENOTSUP) {
1272              *need_zero = false;
1273              return ret;
1274          }
1275      }
1276  
1277      return bdrv_truncate(bs->file, *new_offset + s->block_size, false,
1278                           PREALLOC_MODE_OFF, 0, NULL);
1279  }
1280  
1281  /*
1282   * Update the BAT table entry with the new file offset, and the new entry
1283   * state */
1284  static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s,
1285                                         VHDXSectorInfo *sinfo,
1286                                         uint64_t *bat_entry_le,
1287                                         uint64_t *bat_offset, int state)
1288  {
1289      /* The BAT entry is a uint64, with 44 bits for the file offset in units of
1290       * 1MB, and 3 bits for the block state. */
1291      if ((state == PAYLOAD_BLOCK_ZERO)        ||
1292          (state == PAYLOAD_BLOCK_UNDEFINED)   ||
1293          (state == PAYLOAD_BLOCK_NOT_PRESENT) ||
1294          (state == PAYLOAD_BLOCK_UNMAPPED)) {
1295          s->bat[sinfo->bat_idx]  = 0;  /* For PAYLOAD_BLOCK_ZERO, the
1296                                           FileOffsetMB field is denoted as
1297                                           'reserved' in the v1.0 spec.  If it is
1298                                           non-zero, MS Hyper-V will fail to read
1299                                           the disk image */
1300      } else {
1301          s->bat[sinfo->bat_idx]  = sinfo->file_offset;
1302      }
1303  
1304      s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK;
1305  
1306      *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]);
1307      *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry);
1308  
1309  }
1310  
1311  /* Per the spec, on the first write of guest-visible data to the file the
1312   * data write guid must be updated in the header */
1313  int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s)
1314  {
1315      int ret = 0;
1316      if (s->first_visible_write) {
1317          s->first_visible_write = false;
1318          ret = vhdx_update_headers(bs, s, true, NULL);
1319      }
1320      return ret;
1321  }
1322  
1323  static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num,
1324                                         int nb_sectors, QEMUIOVector *qiov,
1325                                         int flags)
1326  {
1327      int ret = -ENOTSUP;
1328      BDRVVHDXState *s = bs->opaque;
1329      VHDXSectorInfo sinfo;
1330      uint64_t bytes_done = 0;
1331      uint64_t bat_entry = 0;
1332      uint64_t bat_entry_offset = 0;
1333      QEMUIOVector hd_qiov;
1334      struct iovec iov1 = { 0 };
1335      struct iovec iov2 = { 0 };
1336      int sectors_to_write;
1337      int bat_state;
1338      uint64_t bat_prior_offset = 0;
1339      bool bat_update = false;
1340  
1341      assert(!flags);
1342      qemu_iovec_init(&hd_qiov, qiov->niov);
1343  
1344      qemu_co_mutex_lock(&s->lock);
1345  
1346      ret = vhdx_user_visible_write(bs, s);
1347      if (ret < 0) {
1348          goto exit;
1349      }
1350  
1351      while (nb_sectors > 0) {
1352          bool use_zero_buffers = false;
1353          bat_update = false;
1354          if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) {
1355              /* not supported yet */
1356              ret = -ENOTSUP;
1357              goto exit;
1358          } else {
1359              vhdx_block_translate(s, sector_num, nb_sectors, &sinfo);
1360              sectors_to_write = sinfo.sectors_avail;
1361  
1362              qemu_iovec_reset(&hd_qiov);
1363              /* check the payload block state */
1364              bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK;
1365              switch (bat_state) {
1366              case PAYLOAD_BLOCK_ZERO:
1367                  /* in this case, we need to preserve zero writes for
1368                   * data that is not part of this write, so we must pad
1369                   * the rest of the buffer to zeroes */
1370                  use_zero_buffers = true;
1371                  /* fall through */
1372              case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */
1373              case PAYLOAD_BLOCK_UNMAPPED:
1374              case PAYLOAD_BLOCK_UNMAPPED_v095:
1375              case PAYLOAD_BLOCK_UNDEFINED:
1376                  bat_prior_offset = sinfo.file_offset;
1377                  ret = vhdx_allocate_block(bs, s, &sinfo.file_offset,
1378                                            &use_zero_buffers);
1379                  if (ret < 0) {
1380                      goto exit;
1381                  }
1382                  /*
1383                   * once we support differencing files, this may also be
1384                   * partially present
1385                   */
1386                  /* update block state to the newly specified state */
1387                  vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1388                                              &bat_entry_offset,
1389                                              PAYLOAD_BLOCK_FULLY_PRESENT);
1390                  bat_update = true;
1391                  /*
1392                   * Since we just allocated a block, file_offset is the
1393                   * beginning of the payload block. It needs to be the
1394                   * write address, which includes the offset into the
1395                   * block, unless the entire block needs to read as
1396                   * zeroes but truncation was not able to provide them,
1397                   * in which case we need to fill in the rest.
1398                   */
1399                  if (!use_zero_buffers) {
1400                      sinfo.file_offset += sinfo.block_offset;
1401                  } else {
1402                      /* zero fill the front, if any */
1403                      if (sinfo.block_offset) {
1404                          iov1.iov_len = sinfo.block_offset;
1405                          iov1.iov_base = qemu_blockalign(bs, iov1.iov_len);
1406                          memset(iov1.iov_base, 0, iov1.iov_len);
1407                          qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0,
1408                                                iov1.iov_len);
1409                          sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS;
1410                      }
1411  
1412                      /* our actual data */
1413                      qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1414                                        sinfo.bytes_avail);
1415  
1416                      /* zero fill the back, if any */
1417                      if ((sinfo.bytes_avail - sinfo.block_offset) <
1418                           s->block_size) {
1419                          iov2.iov_len = s->block_size -
1420                                        (sinfo.bytes_avail + sinfo.block_offset);
1421                          iov2.iov_base = qemu_blockalign(bs, iov2.iov_len);
1422                          memset(iov2.iov_base, 0, iov2.iov_len);
1423                          qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0,
1424                                                iov2.iov_len);
1425                          sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS;
1426                      }
1427                  }
1428  
1429                  /* fall through */
1430              case PAYLOAD_BLOCK_FULLY_PRESENT:
1431                  /* if the file offset address is in the header zone,
1432                   * there is a problem */
1433                  if (sinfo.file_offset < (1 * MiB)) {
1434                      ret = -EFAULT;
1435                      goto error_bat_restore;
1436                  }
1437  
1438                  if (!use_zero_buffers) {
1439                      qemu_iovec_concat(&hd_qiov, qiov,  bytes_done,
1440                                        sinfo.bytes_avail);
1441                  }
1442                  /* block exists, so we can just overwrite it */
1443                  qemu_co_mutex_unlock(&s->lock);
1444                  ret = bdrv_co_pwritev(bs->file, sinfo.file_offset,
1445                                        sectors_to_write * BDRV_SECTOR_SIZE,
1446                                        &hd_qiov, 0);
1447                  qemu_co_mutex_lock(&s->lock);
1448                  if (ret < 0) {
1449                      goto error_bat_restore;
1450                  }
1451                  break;
1452              case PAYLOAD_BLOCK_PARTIALLY_PRESENT:
1453                  /* we don't yet support difference files, fall through
1454                   * to error */
1455              default:
1456                  ret = -EIO;
1457                  goto exit;
1458                  break;
1459              }
1460  
1461              if (bat_update) {
1462                  /* this will update the BAT entry into the log journal, and
1463                   * then flush the log journal out to disk */
1464                  ret =  vhdx_log_write_and_flush(bs, s, &bat_entry,
1465                                                  sizeof(VHDXBatEntry),
1466                                                  bat_entry_offset);
1467                  if (ret < 0) {
1468                      goto exit;
1469                  }
1470              }
1471  
1472              nb_sectors -= sinfo.sectors_avail;
1473              sector_num += sinfo.sectors_avail;
1474              bytes_done += sinfo.bytes_avail;
1475  
1476          }
1477      }
1478  
1479      goto exit;
1480  
1481  error_bat_restore:
1482      if (bat_update) {
1483          /* keep metadata in sync, and restore the bat entry state
1484           * if error. */
1485          sinfo.file_offset = bat_prior_offset;
1486          vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry,
1487                                      &bat_entry_offset, bat_state);
1488      }
1489  exit:
1490      qemu_vfree(iov1.iov_base);
1491      qemu_vfree(iov2.iov_base);
1492      qemu_co_mutex_unlock(&s->lock);
1493      qemu_iovec_destroy(&hd_qiov);
1494      return ret;
1495  }
1496  
1497  
1498  
1499  /*
1500   * Create VHDX Headers
1501   *
1502   * There are 2 headers, and the highest sequence number will represent
1503   * the active header
1504   */
1505  static int vhdx_create_new_headers(BlockBackend *blk, uint64_t image_size,
1506                                     uint32_t log_size)
1507  {
1508      BlockDriverState *bs = blk_bs(blk);
1509      BdrvChild *child;
1510      int ret = 0;
1511      VHDXHeader *hdr = NULL;
1512  
1513      hdr = g_new0(VHDXHeader, 1);
1514  
1515      hdr->signature       = VHDX_HEADER_SIGNATURE;
1516      hdr->sequence_number = g_random_int();
1517      hdr->log_version     = 0;
1518      hdr->version         = 1;
1519      hdr->log_length      = log_size;
1520      hdr->log_offset      = VHDX_HEADER_SECTION_END;
1521      vhdx_guid_generate(&hdr->file_write_guid);
1522      vhdx_guid_generate(&hdr->data_write_guid);
1523  
1524      /* XXX Ugly way to get blk->root, but that's a feature, not a bug. This
1525       * hack makes it obvious that vhdx_write_header() bypasses the BlockBackend
1526       * here, which it really shouldn't be doing. */
1527      child = QLIST_FIRST(&bs->parents);
1528      assert(!QLIST_NEXT(child, next_parent));
1529  
1530      ret = vhdx_write_header(child, hdr, VHDX_HEADER1_OFFSET, false);
1531      if (ret < 0) {
1532          goto exit;
1533      }
1534      hdr->sequence_number++;
1535      ret = vhdx_write_header(child, hdr, VHDX_HEADER2_OFFSET, false);
1536      if (ret < 0) {
1537          goto exit;
1538      }
1539  
1540  exit:
1541      g_free(hdr);
1542      return ret;
1543  }
1544  
1545  #define VHDX_METADATA_ENTRY_BUFFER_SIZE \
1546                                      (sizeof(VHDXFileParameters)               +\
1547                                       sizeof(VHDXVirtualDiskSize)              +\
1548                                       sizeof(VHDXPage83Data)                   +\
1549                                       sizeof(VHDXVirtualDiskLogicalSectorSize) +\
1550                                       sizeof(VHDXVirtualDiskPhysicalSectorSize))
1551  
1552  /*
1553   * Create the Metadata entries.
1554   *
1555   * For more details on the entries, see section 3.5 (pg 29) in the
1556   * VHDX 1.00 specification.
1557   *
1558   * We support 5 metadata entries (all required by spec):
1559   *          File Parameters,
1560   *          Virtual Disk Size,
1561   *          Page 83 Data,
1562   *          Logical Sector Size,
1563   *          Physical Sector Size
1564   *
1565   * The first 64KB of the Metadata section is reserved for the metadata
1566   * header and entries; beyond that, the metadata items themselves reside.
1567   */
1568  static int vhdx_create_new_metadata(BlockBackend *blk,
1569                                      uint64_t image_size,
1570                                      uint32_t block_size,
1571                                      uint32_t sector_size,
1572                                      uint64_t metadata_offset,
1573                                      VHDXImageType type)
1574  {
1575      int ret = 0;
1576      uint32_t offset = 0;
1577      void *buffer = NULL;
1578      void *entry_buffer;
1579      VHDXMetadataTableHeader *md_table;
1580      VHDXMetadataTableEntry  *md_table_entry;
1581  
1582      /* Metadata entries */
1583      VHDXFileParameters     *mt_file_params;
1584      VHDXVirtualDiskSize    *mt_virtual_size;
1585      VHDXPage83Data         *mt_page83;
1586      VHDXVirtualDiskLogicalSectorSize  *mt_log_sector_size;
1587      VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size;
1588  
1589      entry_buffer = g_malloc0(VHDX_METADATA_ENTRY_BUFFER_SIZE);
1590  
1591      mt_file_params = entry_buffer;
1592      offset += sizeof(VHDXFileParameters);
1593      mt_virtual_size = entry_buffer + offset;
1594      offset += sizeof(VHDXVirtualDiskSize);
1595      mt_page83 = entry_buffer + offset;
1596      offset += sizeof(VHDXPage83Data);
1597      mt_log_sector_size = entry_buffer + offset;
1598      offset += sizeof(VHDXVirtualDiskLogicalSectorSize);
1599      mt_phys_sector_size = entry_buffer + offset;
1600  
1601      mt_file_params->block_size = cpu_to_le32(block_size);
1602      if (type == VHDX_TYPE_FIXED) {
1603          mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED;
1604          mt_file_params->data_bits = cpu_to_le32(mt_file_params->data_bits);
1605      }
1606  
1607      vhdx_guid_generate(&mt_page83->page_83_data);
1608      cpu_to_leguids(&mt_page83->page_83_data);
1609      mt_virtual_size->virtual_disk_size        = cpu_to_le64(image_size);
1610      mt_log_sector_size->logical_sector_size   = cpu_to_le32(sector_size);
1611      mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size);
1612  
1613      buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1614      md_table = buffer;
1615  
1616      md_table->signature   = VHDX_METADATA_SIGNATURE;
1617      md_table->entry_count = 5;
1618      vhdx_metadata_header_le_export(md_table);
1619  
1620  
1621      /* This will reference beyond the reserved table portion */
1622      offset = 64 * KiB;
1623  
1624      md_table_entry = buffer + sizeof(VHDXMetadataTableHeader);
1625  
1626      md_table_entry[0].item_id = file_param_guid;
1627      md_table_entry[0].offset  = offset;
1628      md_table_entry[0].length  = sizeof(VHDXFileParameters);
1629      md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED;
1630      offset += md_table_entry[0].length;
1631      vhdx_metadata_entry_le_export(&md_table_entry[0]);
1632  
1633      md_table_entry[1].item_id = virtual_size_guid;
1634      md_table_entry[1].offset  = offset;
1635      md_table_entry[1].length  = sizeof(VHDXVirtualDiskSize);
1636      md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1637                                     VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1638      offset += md_table_entry[1].length;
1639      vhdx_metadata_entry_le_export(&md_table_entry[1]);
1640  
1641      md_table_entry[2].item_id = page83_guid;
1642      md_table_entry[2].offset  = offset;
1643      md_table_entry[2].length  = sizeof(VHDXPage83Data);
1644      md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1645                                     VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1646      offset += md_table_entry[2].length;
1647      vhdx_metadata_entry_le_export(&md_table_entry[2]);
1648  
1649      md_table_entry[3].item_id = logical_sector_guid;
1650      md_table_entry[3].offset  = offset;
1651      md_table_entry[3].length  = sizeof(VHDXVirtualDiskLogicalSectorSize);
1652      md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1653                                     VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1654      offset += md_table_entry[3].length;
1655      vhdx_metadata_entry_le_export(&md_table_entry[3]);
1656  
1657      md_table_entry[4].item_id = phys_sector_guid;
1658      md_table_entry[4].offset  = offset;
1659      md_table_entry[4].length  = sizeof(VHDXVirtualDiskPhysicalSectorSize);
1660      md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED |
1661                                     VHDX_META_FLAGS_IS_VIRTUAL_DISK;
1662      vhdx_metadata_entry_le_export(&md_table_entry[4]);
1663  
1664      ret = blk_pwrite(blk, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE, 0);
1665      if (ret < 0) {
1666          goto exit;
1667      }
1668  
1669      ret = blk_pwrite(blk, metadata_offset + (64 * KiB), entry_buffer,
1670                       VHDX_METADATA_ENTRY_BUFFER_SIZE, 0);
1671      if (ret < 0) {
1672          goto exit;
1673      }
1674  
1675  
1676  exit:
1677      g_free(buffer);
1678      g_free(entry_buffer);
1679      return ret;
1680  }
1681  
1682  /* This create the actual BAT itself.  We currently only support
1683   * 'Dynamic' and 'Fixed' image types.
1684   *
1685   *  Dynamic images: default state of the BAT is all zeroes.
1686   *
1687   *  Fixed images: default state of the BAT is fully populated, with
1688   *                file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT.
1689   */
1690  static int vhdx_create_bat(BlockBackend *blk, BDRVVHDXState *s,
1691                             uint64_t image_size, VHDXImageType type,
1692                             bool use_zero_blocks, uint64_t file_offset,
1693                             uint32_t length, Error **errp)
1694  {
1695      int ret = 0;
1696      uint64_t data_file_offset;
1697      uint64_t total_sectors = 0;
1698      uint64_t sector_num = 0;
1699      uint64_t unused;
1700      int block_state;
1701      VHDXSectorInfo sinfo;
1702  
1703      assert(s->bat == NULL);
1704  
1705      /* this gives a data start after BAT/bitmap entries, and well
1706       * past any metadata entries (with a 4 MB buffer for future
1707       * expansion */
1708      data_file_offset = file_offset + length + 5 * MiB;
1709      total_sectors = image_size >> s->logical_sector_size_bits;
1710  
1711      if (type == VHDX_TYPE_DYNAMIC) {
1712          /* All zeroes, so we can just extend the file - the end of the BAT
1713           * is the furthest thing we have written yet */
1714          ret = blk_truncate(blk, data_file_offset, false, PREALLOC_MODE_OFF,
1715                             0, errp);
1716          if (ret < 0) {
1717              goto exit;
1718          }
1719      } else if (type == VHDX_TYPE_FIXED) {
1720          ret = blk_truncate(blk, data_file_offset + image_size, false,
1721                             PREALLOC_MODE_OFF, 0, errp);
1722          if (ret < 0) {
1723              goto exit;
1724          }
1725      } else {
1726          error_setg(errp, "Unsupported image type");
1727          ret = -ENOTSUP;
1728          goto exit;
1729      }
1730  
1731      if (type == VHDX_TYPE_FIXED ||
1732                  use_zero_blocks ||
1733                  bdrv_has_zero_init(blk_bs(blk)) == 0) {
1734          /* for a fixed file, the default BAT entry is not zero */
1735          s->bat = g_try_malloc0(length);
1736          if (length && s->bat == NULL) {
1737              error_setg(errp, "Failed to allocate memory for the BAT");
1738              ret = -ENOMEM;
1739              goto exit;
1740          }
1741          block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT :
1742                                                  PAYLOAD_BLOCK_NOT_PRESENT;
1743          block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state;
1744          /* fill the BAT by emulating sector writes of sectors_per_block size */
1745          while (sector_num < total_sectors) {
1746              vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo);
1747              sinfo.file_offset = data_file_offset +
1748                                  (sector_num << s->logical_sector_size_bits);
1749              sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB);
1750              vhdx_update_bat_table_entry(blk_bs(blk), s, &sinfo, &unused, &unused,
1751                                          block_state);
1752              s->bat[sinfo.bat_idx] = cpu_to_le64(s->bat[sinfo.bat_idx]);
1753              sector_num += s->sectors_per_block;
1754          }
1755          ret = blk_pwrite(blk, file_offset, s->bat, length, 0);
1756          if (ret < 0) {
1757              error_setg_errno(errp, -ret, "Failed to write the BAT");
1758              goto exit;
1759          }
1760      }
1761  
1762  
1763  
1764  exit:
1765      g_free(s->bat);
1766      return ret;
1767  }
1768  
1769  /* Creates the region table header, and region table entries.
1770   * There are 2 supported region table entries: BAT, and Metadata/
1771   *
1772   * As the calculations for the BAT region table are also needed
1773   * to create the BAT itself, we will also cause the BAT to be
1774   * created.
1775   */
1776  static int vhdx_create_new_region_table(BlockBackend *blk,
1777                                          uint64_t image_size,
1778                                          uint32_t block_size,
1779                                          uint32_t sector_size,
1780                                          uint32_t log_size,
1781                                          bool use_zero_blocks,
1782                                          VHDXImageType type,
1783                                          uint64_t *metadata_offset,
1784                                          Error **errp)
1785  {
1786      int ret = 0;
1787      uint32_t offset = 0;
1788      void *buffer = NULL;
1789      uint64_t bat_file_offset;
1790      uint32_t bat_length;
1791      BDRVVHDXState *s = NULL;
1792      VHDXRegionTableHeader *region_table;
1793      VHDXRegionTableEntry *rt_bat;
1794      VHDXRegionTableEntry *rt_metadata;
1795  
1796      assert(metadata_offset != NULL);
1797  
1798      /* Populate enough of the BDRVVHDXState to be able to use the
1799       * pre-existing BAT calculation, translation, and update functions */
1800      s = g_new0(BDRVVHDXState, 1);
1801  
1802      s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) *
1803                       (uint64_t) sector_size / (uint64_t) block_size;
1804  
1805      s->sectors_per_block = block_size / sector_size;
1806      s->virtual_disk_size = image_size;
1807      s->block_size = block_size;
1808      s->logical_sector_size = sector_size;
1809  
1810      vhdx_set_shift_bits(s);
1811  
1812      vhdx_calc_bat_entries(s);
1813  
1814      /* At this point the VHDX state is populated enough for creation */
1815  
1816      /* a single buffer is used so we can calculate the checksum over the
1817       * entire 64KB block */
1818      buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE);
1819      region_table = buffer;
1820      offset += sizeof(VHDXRegionTableHeader);
1821      rt_bat = buffer + offset;
1822      offset += sizeof(VHDXRegionTableEntry);
1823      rt_metadata  = buffer + offset;
1824  
1825      region_table->signature = VHDX_REGION_SIGNATURE;
1826      region_table->entry_count = 2;   /* BAT and Metadata */
1827  
1828      rt_bat->guid        = bat_guid;
1829      rt_bat->length      = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB);
1830      rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB);
1831      s->bat_offset = rt_bat->file_offset;
1832  
1833      rt_metadata->guid        = metadata_guid;
1834      rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length,
1835                                          MiB);
1836      rt_metadata->length      = 1 * MiB; /* min size, and more than enough */
1837      *metadata_offset = rt_metadata->file_offset;
1838  
1839      bat_file_offset = rt_bat->file_offset;
1840      bat_length = rt_bat->length;
1841  
1842      vhdx_region_header_le_export(region_table);
1843      vhdx_region_entry_le_export(rt_bat);
1844      vhdx_region_entry_le_export(rt_metadata);
1845  
1846      vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE,
1847                           offsetof(VHDXRegionTableHeader, checksum));
1848  
1849  
1850      /* The region table gives us the data we need to create the BAT,
1851       * so do that now */
1852      ret = vhdx_create_bat(blk, s, image_size, type, use_zero_blocks,
1853                            bat_file_offset, bat_length, errp);
1854      if (ret < 0) {
1855          goto exit;
1856      }
1857  
1858      /* Now write out the region headers to disk */
1859      ret = blk_pwrite(blk, VHDX_REGION_TABLE_OFFSET, buffer,
1860                       VHDX_HEADER_BLOCK_SIZE, 0);
1861      if (ret < 0) {
1862          error_setg_errno(errp, -ret, "Failed to write first region table");
1863          goto exit;
1864      }
1865  
1866      ret = blk_pwrite(blk, VHDX_REGION_TABLE2_OFFSET, buffer,
1867                       VHDX_HEADER_BLOCK_SIZE, 0);
1868      if (ret < 0) {
1869          error_setg_errno(errp, -ret, "Failed to write second region table");
1870          goto exit;
1871      }
1872  
1873  exit:
1874      g_free(s);
1875      g_free(buffer);
1876      return ret;
1877  }
1878  
1879  /* We need to create the following elements:
1880   *
1881   *    .-----------------------------------------------------------------.
1882   *    |   (A)    |   (B)    |    (C)    |     (D)       |     (E)       |
1883   *    |  File ID |  Header1 |  Header 2 |  Region Tbl 1 |  Region Tbl 2 |
1884   *    |          |          |           |               |               |
1885   *    .-----------------------------------------------------------------.
1886   *    0         64KB      128KB       192KB           256KB           320KB
1887   *
1888   *
1889   *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1890   *    |     (F)     |     (G)       |    (H)    |                        |
1891   *    | Journal Log |  BAT / Bitmap |  Metadata |  .... data ......      |
1892   *    |             |               |           |                        |
1893   *    .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------.
1894   *   1MB
1895   */
1896  static int coroutine_fn vhdx_co_create(BlockdevCreateOptions *opts,
1897                                         Error **errp)
1898  {
1899      BlockdevCreateOptionsVhdx *vhdx_opts;
1900      BlockBackend *blk = NULL;
1901      BlockDriverState *bs = NULL;
1902  
1903      int ret = 0;
1904      uint64_t image_size;
1905      uint32_t log_size;
1906      uint32_t block_size;
1907      uint64_t signature;
1908      uint64_t metadata_offset;
1909      bool use_zero_blocks = false;
1910  
1911      gunichar2 *creator = NULL;
1912      glong creator_items;
1913      VHDXImageType image_type;
1914  
1915      assert(opts->driver == BLOCKDEV_DRIVER_VHDX);
1916      vhdx_opts = &opts->u.vhdx;
1917  
1918      /* Validate options and set default values */
1919      image_size = vhdx_opts->size;
1920      if (image_size > VHDX_MAX_IMAGE_SIZE) {
1921          error_setg(errp, "Image size too large; max of 64TB");
1922          return -EINVAL;
1923      }
1924  
1925      if (!vhdx_opts->has_log_size) {
1926          log_size = DEFAULT_LOG_SIZE;
1927      } else {
1928          if (vhdx_opts->log_size > UINT32_MAX) {
1929              error_setg(errp, "Log size must be smaller than 4 GB");
1930              return -EINVAL;
1931          }
1932          log_size = vhdx_opts->log_size;
1933      }
1934      if (log_size < MiB || (log_size % MiB) != 0) {
1935          error_setg(errp, "Log size must be a multiple of 1 MB");
1936          return -EINVAL;
1937      }
1938  
1939      if (!vhdx_opts->has_block_state_zero) {
1940          use_zero_blocks = true;
1941      } else {
1942          use_zero_blocks = vhdx_opts->block_state_zero;
1943      }
1944  
1945      if (!vhdx_opts->has_subformat) {
1946          vhdx_opts->subformat = BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC;
1947      }
1948  
1949      switch (vhdx_opts->subformat) {
1950      case BLOCKDEV_VHDX_SUBFORMAT_DYNAMIC:
1951          image_type = VHDX_TYPE_DYNAMIC;
1952          break;
1953      case BLOCKDEV_VHDX_SUBFORMAT_FIXED:
1954          image_type = VHDX_TYPE_FIXED;
1955          break;
1956      default:
1957          g_assert_not_reached();
1958      }
1959  
1960      /* These are pretty arbitrary, and mainly designed to keep the BAT
1961       * size reasonable to load into RAM */
1962      if (vhdx_opts->has_block_size) {
1963          block_size = vhdx_opts->block_size;
1964      } else {
1965          if (image_size > 32 * TiB) {
1966              block_size = 64 * MiB;
1967          } else if (image_size > (uint64_t) 100 * GiB) {
1968              block_size = 32 * MiB;
1969          } else if (image_size > 1 * GiB) {
1970              block_size = 16 * MiB;
1971          } else {
1972              block_size = 8 * MiB;
1973          }
1974      }
1975  
1976      if (block_size < MiB || (block_size % MiB) != 0) {
1977          error_setg(errp, "Block size must be a multiple of 1 MB");
1978          return -EINVAL;
1979      }
1980      if (!is_power_of_2(block_size)) {
1981          error_setg(errp, "Block size must be a power of two");
1982          return -EINVAL;
1983      }
1984      if (block_size > VHDX_BLOCK_SIZE_MAX) {
1985          error_setg(errp, "Block size must not exceed %" PRId64,
1986                     VHDX_BLOCK_SIZE_MAX);
1987          return -EINVAL;
1988      }
1989  
1990      /* Create BlockBackend to write to the image */
1991      bs = bdrv_open_blockdev_ref(vhdx_opts->file, errp);
1992      if (bs == NULL) {
1993          return -EIO;
1994      }
1995  
1996      blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
1997                            errp);
1998      if (!blk) {
1999          ret = -EPERM;
2000          goto delete_and_exit;
2001      }
2002      blk_set_allow_write_beyond_eof(blk, true);
2003  
2004      /* Create (A) */
2005  
2006      /* The creator field is optional, but may be useful for
2007       * debugging / diagnostics */
2008      creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL,
2009                                &creator_items, NULL);
2010      signature = cpu_to_le64(VHDX_FILE_SIGNATURE);
2011      ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature),
2012                       0);
2013      if (ret < 0) {
2014          error_setg_errno(errp, -ret, "Failed to write file signature");
2015          goto delete_and_exit;
2016      }
2017      if (creator) {
2018          ret = blk_pwrite(blk, VHDX_FILE_ID_OFFSET + sizeof(signature),
2019                           creator, creator_items * sizeof(gunichar2), 0);
2020          if (ret < 0) {
2021              error_setg_errno(errp, -ret, "Failed to write creator field");
2022              goto delete_and_exit;
2023          }
2024      }
2025  
2026  
2027      /* Creates (B),(C) */
2028      ret = vhdx_create_new_headers(blk, image_size, log_size);
2029      if (ret < 0) {
2030          error_setg_errno(errp, -ret, "Failed to write image headers");
2031          goto delete_and_exit;
2032      }
2033  
2034      /* Creates (D),(E),(G) explicitly. (F) created as by-product */
2035      ret = vhdx_create_new_region_table(blk, image_size, block_size, 512,
2036                                         log_size, use_zero_blocks, image_type,
2037                                         &metadata_offset, errp);
2038      if (ret < 0) {
2039          goto delete_and_exit;
2040      }
2041  
2042      /* Creates (H) */
2043      ret = vhdx_create_new_metadata(blk, image_size, block_size, 512,
2044                                     metadata_offset, image_type);
2045      if (ret < 0) {
2046          error_setg_errno(errp, -ret, "Failed to initialize metadata");
2047          goto delete_and_exit;
2048      }
2049  
2050      ret = 0;
2051  delete_and_exit:
2052      blk_unref(blk);
2053      bdrv_unref(bs);
2054      g_free(creator);
2055      return ret;
2056  }
2057  
2058  static int coroutine_fn vhdx_co_create_opts(BlockDriver *drv,
2059                                              const char *filename,
2060                                              QemuOpts *opts,
2061                                              Error **errp)
2062  {
2063      BlockdevCreateOptions *create_options = NULL;
2064      QDict *qdict;
2065      Visitor *v;
2066      BlockDriverState *bs = NULL;
2067      int ret;
2068  
2069      static const QDictRenames opt_renames[] = {
2070          { VHDX_BLOCK_OPT_LOG_SIZE,      "log-size" },
2071          { VHDX_BLOCK_OPT_BLOCK_SIZE,    "block-size" },
2072          { VHDX_BLOCK_OPT_ZERO,          "block-state-zero" },
2073          { NULL, NULL },
2074      };
2075  
2076      /* Parse options and convert legacy syntax */
2077      qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vhdx_create_opts, true);
2078  
2079      if (!qdict_rename_keys(qdict, opt_renames, errp)) {
2080          ret = -EINVAL;
2081          goto fail;
2082      }
2083  
2084      /* Create and open the file (protocol layer) */
2085      ret = bdrv_create_file(filename, opts, errp);
2086      if (ret < 0) {
2087          goto fail;
2088      }
2089  
2090      bs = bdrv_open(filename, NULL, NULL,
2091                     BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
2092      if (bs == NULL) {
2093          ret = -EIO;
2094          goto fail;
2095      }
2096  
2097      /* Now get the QAPI type BlockdevCreateOptions */
2098      qdict_put_str(qdict, "driver", "vhdx");
2099      qdict_put_str(qdict, "file", bs->node_name);
2100  
2101      v = qobject_input_visitor_new_flat_confused(qdict, errp);
2102      if (!v) {
2103          ret = -EINVAL;
2104          goto fail;
2105      }
2106  
2107      visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
2108      visit_free(v);
2109      if (!create_options) {
2110          ret = -EINVAL;
2111          goto fail;
2112      }
2113  
2114      /* Silently round up sizes:
2115       * The image size is rounded to 512 bytes. Make the block and log size
2116       * close to what was specified, but must be at least 1MB, and a multiple of
2117       * 1 MB. Also respect VHDX_BLOCK_SIZE_MAX for block sizes. block_size = 0
2118       * means auto, which is represented by a missing key in QAPI. */
2119      assert(create_options->driver == BLOCKDEV_DRIVER_VHDX);
2120      create_options->u.vhdx.size =
2121          ROUND_UP(create_options->u.vhdx.size, BDRV_SECTOR_SIZE);
2122  
2123      if (create_options->u.vhdx.has_log_size) {
2124          create_options->u.vhdx.log_size =
2125              ROUND_UP(create_options->u.vhdx.log_size, MiB);
2126      }
2127      if (create_options->u.vhdx.has_block_size) {
2128          create_options->u.vhdx.block_size =
2129              ROUND_UP(create_options->u.vhdx.block_size, MiB);
2130  
2131          if (create_options->u.vhdx.block_size == 0) {
2132              create_options->u.vhdx.has_block_size = false;
2133          }
2134          if (create_options->u.vhdx.block_size > VHDX_BLOCK_SIZE_MAX) {
2135              create_options->u.vhdx.block_size = VHDX_BLOCK_SIZE_MAX;
2136          }
2137      }
2138  
2139      /* Create the vhdx image (format layer) */
2140      ret = vhdx_co_create(create_options, errp);
2141  
2142  fail:
2143      qobject_unref(qdict);
2144      bdrv_unref(bs);
2145      qapi_free_BlockdevCreateOptions(create_options);
2146      return ret;
2147  }
2148  
2149  /* If opened r/w, the VHDX driver will automatically replay the log,
2150   * if one is present, inside the vhdx_open() call.
2151   *
2152   * If qemu-img check -r all is called, the image is automatically opened
2153   * r/w and any log has already been replayed, so there is nothing (currently)
2154   * for us to do here
2155   */
2156  static int coroutine_fn vhdx_co_check(BlockDriverState *bs,
2157                                        BdrvCheckResult *result,
2158                                        BdrvCheckMode fix)
2159  {
2160      BDRVVHDXState *s = bs->opaque;
2161  
2162      if (s->log_replayed_on_open) {
2163          result->corruptions_fixed++;
2164      }
2165  
2166      vhdx_check_bat_entries(bs, &result->corruptions);
2167  
2168      return 0;
2169  }
2170  
2171  static int vhdx_has_zero_init(BlockDriverState *bs)
2172  {
2173      BDRVVHDXState *s = bs->opaque;
2174      int state;
2175  
2176      /*
2177       * Check the subformat: Fixed images have all BAT entries present,
2178       * dynamic images have none (right after creation).  It is
2179       * therefore enough to check the first BAT entry.
2180       */
2181      if (!s->bat_entries) {
2182          return 1;
2183      }
2184  
2185      state = s->bat[0] & VHDX_BAT_STATE_BIT_MASK;
2186      if (state == PAYLOAD_BLOCK_FULLY_PRESENT) {
2187          /* Fixed subformat */
2188          return bdrv_has_zero_init(bs->file->bs);
2189      }
2190  
2191      /* Dynamic subformat */
2192      return 1;
2193  }
2194  
2195  static QemuOptsList vhdx_create_opts = {
2196      .name = "vhdx-create-opts",
2197      .head = QTAILQ_HEAD_INITIALIZER(vhdx_create_opts.head),
2198      .desc = {
2199          {
2200             .name = BLOCK_OPT_SIZE,
2201             .type = QEMU_OPT_SIZE,
2202             .help = "Virtual disk size; max of 64TB."
2203         },
2204         {
2205             .name = VHDX_BLOCK_OPT_LOG_SIZE,
2206             .type = QEMU_OPT_SIZE,
2207             .def_value_str = stringify(DEFAULT_LOG_SIZE),
2208             .help = "Log size; min 1MB."
2209         },
2210         {
2211             .name = VHDX_BLOCK_OPT_BLOCK_SIZE,
2212             .type = QEMU_OPT_SIZE,
2213             .def_value_str = stringify(0),
2214             .help = "Block Size; min 1MB, max 256MB. "
2215                     "0 means auto-calculate based on image size."
2216         },
2217         {
2218             .name = BLOCK_OPT_SUBFMT,
2219             .type = QEMU_OPT_STRING,
2220             .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "
2221                     "Default is 'dynamic'."
2222         },
2223         {
2224             .name = VHDX_BLOCK_OPT_ZERO,
2225             .type = QEMU_OPT_BOOL,
2226             .help = "Force use of payload blocks of type 'ZERO'. "
2227                     "Non-standard, but default.  Do not set to 'off' when "
2228                     "using 'qemu-img convert' with subformat=dynamic."
2229         },
2230         { NULL }
2231      }
2232  };
2233  
2234  static BlockDriver bdrv_vhdx = {
2235      .format_name            = "vhdx",
2236      .instance_size          = sizeof(BDRVVHDXState),
2237      .bdrv_probe             = vhdx_probe,
2238      .bdrv_open              = vhdx_open,
2239      .bdrv_close             = vhdx_close,
2240      .bdrv_reopen_prepare    = vhdx_reopen_prepare,
2241      .bdrv_child_perm        = bdrv_default_perms,
2242      .bdrv_co_readv          = vhdx_co_readv,
2243      .bdrv_co_writev         = vhdx_co_writev,
2244      .bdrv_co_create         = vhdx_co_create,
2245      .bdrv_co_create_opts    = vhdx_co_create_opts,
2246      .bdrv_get_info          = vhdx_get_info,
2247      .bdrv_co_check          = vhdx_co_check,
2248      .bdrv_has_zero_init     = vhdx_has_zero_init,
2249  
2250      .is_format              = true,
2251      .create_opts            = &vhdx_create_opts,
2252  };
2253  
2254  static void bdrv_vhdx_init(void)
2255  {
2256      bdrv_register(&bdrv_vhdx);
2257  }
2258  
2259  block_init(bdrv_vhdx_init);
2260