2 * Block driver for the VMDK format
4 * Copyright (c) 2004 Fabrice Bellard
5 * Copyright (c) 2005 Filip Navara
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu-common.h"
27 #include "block/block_int.h"
28 #include "qemu/module.h"
29 #include "migration/migration.h"
32 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
33 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
34 #define VMDK4_COMPRESSION_DEFLATE 1
35 #define VMDK4_FLAG_RGD (1 << 1)
36 #define VMDK4_FLAG_COMPRESS (1 << 16)
37 #define VMDK4_FLAG_MARKER (1 << 17)
38 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
43 uint32_t disk_sectors
;
45 uint32_t l1dir_offset
;
47 uint32_t file_sectors
;
50 uint32_t sectors_per_track
;
60 int32_t num_gtes_per_gte
;
66 uint16_t compressAlgorithm
;
67 } QEMU_PACKED VMDK4Header
;
69 #define L2_CACHE_SIZE 16
71 typedef struct VmdkExtent
{
72 BlockDriverState
*file
;
78 int64_t flat_start_offset
;
79 int64_t l1_table_offset
;
80 int64_t l1_backup_table_offset
;
82 uint32_t *l1_backup_table
;
84 uint32_t l1_entry_sectors
;
88 uint32_t l2_cache_offsets
[L2_CACHE_SIZE
];
89 uint32_t l2_cache_counts
[L2_CACHE_SIZE
];
91 unsigned int cluster_sectors
;
94 typedef struct BDRVVmdkState
{
100 /* Extent array with num_extents entries, ascend ordered by address */
102 Error
*migration_blocker
;
105 typedef struct VmdkMetaData
{
107 unsigned int l1_index
;
108 unsigned int l2_index
;
109 unsigned int l2_offset
;
113 typedef struct VmdkGrainMarker
{
120 MARKER_END_OF_STREAM
= 0,
121 MARKER_GRAIN_TABLE
= 1,
122 MARKER_GRAIN_DIRECTORY
= 2,
126 static int vmdk_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
133 magic
= be32_to_cpu(*(uint32_t *)buf
);
134 if (magic
== VMDK3_MAGIC
||
135 magic
== VMDK4_MAGIC
) {
138 const char *p
= (const char *)buf
;
139 const char *end
= p
+ buf_size
;
142 /* skip comment line */
143 while (p
< end
&& *p
!= '\n') {
150 while (p
< end
&& *p
== ' ') {
153 /* skip '\r' if windows line endings used. */
154 if (p
< end
&& *p
== '\r') {
157 /* only accept blank lines before 'version=' line */
158 if (p
== end
|| *p
!= '\n') {
164 if (end
- p
>= strlen("version=X\n")) {
165 if (strncmp("version=1\n", p
, strlen("version=1\n")) == 0 ||
166 strncmp("version=2\n", p
, strlen("version=2\n")) == 0) {
170 if (end
- p
>= strlen("version=X\r\n")) {
171 if (strncmp("version=1\r\n", p
, strlen("version=1\r\n")) == 0 ||
172 strncmp("version=2\r\n", p
, strlen("version=2\r\n")) == 0) {
184 #define SECTOR_SIZE 512
185 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
186 #define BUF_SIZE 4096
187 #define HEADER_SIZE 512 /* first sector of 512 bytes */
189 static void vmdk_free_extents(BlockDriverState
*bs
)
192 BDRVVmdkState
*s
= bs
->opaque
;
195 for (i
= 0; i
< s
->num_extents
; i
++) {
199 g_free(e
->l1_backup_table
);
200 if (e
->file
!= bs
->file
) {
201 bdrv_delete(e
->file
);
207 static void vmdk_free_last_extent(BlockDriverState
*bs
)
209 BDRVVmdkState
*s
= bs
->opaque
;
211 if (s
->num_extents
== 0) {
215 s
->extents
= g_realloc(s
->extents
, s
->num_extents
* sizeof(VmdkExtent
));
218 static uint32_t vmdk_read_cid(BlockDriverState
*bs
, int parent
)
220 char desc
[DESC_SIZE
];
221 uint32_t cid
= 0xffffffff;
222 const char *p_name
, *cid_str
;
224 BDRVVmdkState
*s
= bs
->opaque
;
227 ret
= bdrv_pread(bs
->file
, s
->desc_offset
, desc
, DESC_SIZE
);
233 cid_str
= "parentCID";
234 cid_str_size
= sizeof("parentCID");
237 cid_str_size
= sizeof("CID");
240 desc
[DESC_SIZE
- 1] = '\0';
241 p_name
= strstr(desc
, cid_str
);
242 if (p_name
!= NULL
) {
243 p_name
+= cid_str_size
;
244 sscanf(p_name
, "%x", &cid
);
250 static int vmdk_write_cid(BlockDriverState
*bs
, uint32_t cid
)
252 char desc
[DESC_SIZE
], tmp_desc
[DESC_SIZE
];
253 char *p_name
, *tmp_str
;
254 BDRVVmdkState
*s
= bs
->opaque
;
257 ret
= bdrv_pread(bs
->file
, s
->desc_offset
, desc
, DESC_SIZE
);
262 desc
[DESC_SIZE
- 1] = '\0';
263 tmp_str
= strstr(desc
, "parentCID");
264 if (tmp_str
== NULL
) {
268 pstrcpy(tmp_desc
, sizeof(tmp_desc
), tmp_str
);
269 p_name
= strstr(desc
, "CID");
270 if (p_name
!= NULL
) {
271 p_name
+= sizeof("CID");
272 snprintf(p_name
, sizeof(desc
) - (p_name
- desc
), "%x\n", cid
);
273 pstrcat(desc
, sizeof(desc
), tmp_desc
);
276 ret
= bdrv_pwrite_sync(bs
->file
, s
->desc_offset
, desc
, DESC_SIZE
);
284 static int vmdk_is_cid_valid(BlockDriverState
*bs
)
287 BDRVVmdkState
*s
= bs
->opaque
;
288 BlockDriverState
*p_bs
= bs
->backing_hd
;
292 cur_pcid
= vmdk_read_cid(p_bs
, 0);
293 if (s
->parent_cid
!= cur_pcid
) {
303 /* Queue extents, if any, for reopen() */
304 static int vmdk_reopen_prepare(BDRVReopenState
*state
,
305 BlockReopenQueue
*queue
, Error
**errp
)
312 assert(state
!= NULL
);
313 assert(state
->bs
!= NULL
);
316 error_set(errp
, ERROR_CLASS_GENERIC_ERROR
,
317 "No reopen queue for VMDK extents");
321 s
= state
->bs
->opaque
;
325 for (i
= 0; i
< s
->num_extents
; i
++) {
327 if (e
->file
!= state
->bs
->file
) {
328 bdrv_reopen_queue(queue
, e
->file
, state
->flags
);
337 static int vmdk_parent_open(BlockDriverState
*bs
)
340 char desc
[DESC_SIZE
+ 1];
341 BDRVVmdkState
*s
= bs
->opaque
;
344 desc
[DESC_SIZE
] = '\0';
345 ret
= bdrv_pread(bs
->file
, s
->desc_offset
, desc
, DESC_SIZE
);
350 p_name
= strstr(desc
, "parentFileNameHint");
351 if (p_name
!= NULL
) {
354 p_name
+= sizeof("parentFileNameHint") + 1;
355 end_name
= strchr(p_name
, '\"');
356 if (end_name
== NULL
) {
359 if ((end_name
- p_name
) > sizeof(bs
->backing_file
) - 1) {
363 pstrcpy(bs
->backing_file
, end_name
- p_name
+ 1, p_name
);
369 /* Create and append extent to the extent array. Return the added VmdkExtent
370 * address. return NULL if allocation failed. */
371 static VmdkExtent
*vmdk_add_extent(BlockDriverState
*bs
,
372 BlockDriverState
*file
, bool flat
, int64_t sectors
,
373 int64_t l1_offset
, int64_t l1_backup_offset
,
375 int l2_size
, unsigned int cluster_sectors
)
378 BDRVVmdkState
*s
= bs
->opaque
;
380 s
->extents
= g_realloc(s
->extents
,
381 (s
->num_extents
+ 1) * sizeof(VmdkExtent
));
382 extent
= &s
->extents
[s
->num_extents
];
385 memset(extent
, 0, sizeof(VmdkExtent
));
388 extent
->sectors
= sectors
;
389 extent
->l1_table_offset
= l1_offset
;
390 extent
->l1_backup_table_offset
= l1_backup_offset
;
391 extent
->l1_size
= l1_size
;
392 extent
->l1_entry_sectors
= l2_size
* cluster_sectors
;
393 extent
->l2_size
= l2_size
;
394 extent
->cluster_sectors
= cluster_sectors
;
396 if (s
->num_extents
> 1) {
397 extent
->end_sector
= (*(extent
- 1)).end_sector
+ extent
->sectors
;
399 extent
->end_sector
= extent
->sectors
;
401 bs
->total_sectors
= extent
->end_sector
;
405 static int vmdk_init_tables(BlockDriverState
*bs
, VmdkExtent
*extent
)
410 /* read the L1 table */
411 l1_size
= extent
->l1_size
* sizeof(uint32_t);
412 extent
->l1_table
= g_malloc(l1_size
);
413 ret
= bdrv_pread(extent
->file
,
414 extent
->l1_table_offset
,
420 for (i
= 0; i
< extent
->l1_size
; i
++) {
421 le32_to_cpus(&extent
->l1_table
[i
]);
424 if (extent
->l1_backup_table_offset
) {
425 extent
->l1_backup_table
= g_malloc(l1_size
);
426 ret
= bdrv_pread(extent
->file
,
427 extent
->l1_backup_table_offset
,
428 extent
->l1_backup_table
,
433 for (i
= 0; i
< extent
->l1_size
; i
++) {
434 le32_to_cpus(&extent
->l1_backup_table
[i
]);
439 g_malloc(extent
->l2_size
* L2_CACHE_SIZE
* sizeof(uint32_t));
442 g_free(extent
->l1_backup_table
);
444 g_free(extent
->l1_table
);
448 static int vmdk_open_vmdk3(BlockDriverState
*bs
,
449 BlockDriverState
*file
,
457 ret
= bdrv_pread(file
, sizeof(magic
), &header
, sizeof(header
));
461 extent
= vmdk_add_extent(bs
,
463 le32_to_cpu(header
.disk_sectors
),
464 le32_to_cpu(header
.l1dir_offset
) << 9,
466 le32_to_cpu(header
.granularity
));
467 ret
= vmdk_init_tables(bs
, extent
);
469 /* free extent allocated by vmdk_add_extent */
470 vmdk_free_last_extent(bs
);
475 static int vmdk_open_desc_file(BlockDriverState
*bs
, int flags
,
476 int64_t desc_offset
);
478 static int vmdk_open_vmdk4(BlockDriverState
*bs
,
479 BlockDriverState
*file
,
484 uint32_t l1_size
, l1_entry_sectors
;
487 int64_t l1_backup_offset
= 0;
489 ret
= bdrv_pread(file
, sizeof(magic
), &header
, sizeof(header
));
493 if (header
.capacity
== 0 && header
.desc_offset
) {
494 return vmdk_open_desc_file(bs
, flags
, header
.desc_offset
<< 9);
497 if (le64_to_cpu(header
.gd_offset
) == VMDK4_GD_AT_END
) {
499 * The footer takes precedence over the header, so read it in. The
500 * footer starts at offset -1024 from the end: One sector for the
501 * footer, and another one for the end-of-stream marker.
508 uint8_t pad
[512 - 16];
509 } QEMU_PACKED footer_marker
;
513 uint8_t pad
[512 - 4 - sizeof(VMDK4Header
)];
519 uint8_t pad
[512 - 16];
520 } QEMU_PACKED eos_marker
;
521 } QEMU_PACKED footer
;
523 ret
= bdrv_pread(file
,
524 bs
->file
->total_sectors
* 512 - 1536,
525 &footer
, sizeof(footer
));
530 /* Some sanity checks for the footer */
531 if (be32_to_cpu(footer
.magic
) != VMDK4_MAGIC
||
532 le32_to_cpu(footer
.footer_marker
.size
) != 0 ||
533 le32_to_cpu(footer
.footer_marker
.type
) != MARKER_FOOTER
||
534 le64_to_cpu(footer
.eos_marker
.val
) != 0 ||
535 le32_to_cpu(footer
.eos_marker
.size
) != 0 ||
536 le32_to_cpu(footer
.eos_marker
.type
) != MARKER_END_OF_STREAM
)
541 header
= footer
.header
;
544 l1_entry_sectors
= le32_to_cpu(header
.num_gtes_per_gte
)
545 * le64_to_cpu(header
.granularity
);
546 if (l1_entry_sectors
== 0) {
549 l1_size
= (le64_to_cpu(header
.capacity
) + l1_entry_sectors
- 1)
551 if (le32_to_cpu(header
.flags
) & VMDK4_FLAG_RGD
) {
552 l1_backup_offset
= le64_to_cpu(header
.rgd_offset
) << 9;
554 extent
= vmdk_add_extent(bs
, file
, false,
555 le64_to_cpu(header
.capacity
),
556 le64_to_cpu(header
.gd_offset
) << 9,
559 le32_to_cpu(header
.num_gtes_per_gte
),
560 le64_to_cpu(header
.granularity
));
562 le16_to_cpu(header
.compressAlgorithm
) == VMDK4_COMPRESSION_DEFLATE
;
563 extent
->has_marker
= le32_to_cpu(header
.flags
) & VMDK4_FLAG_MARKER
;
564 ret
= vmdk_init_tables(bs
, extent
);
566 /* free extent allocated by vmdk_add_extent */
567 vmdk_free_last_extent(bs
);
572 /* find an option value out of descriptor file */
573 static int vmdk_parse_description(const char *desc
, const char *opt_name
,
574 char *buf
, int buf_size
)
576 char *opt_pos
, *opt_end
;
577 const char *end
= desc
+ strlen(desc
);
579 opt_pos
= strstr(desc
, opt_name
);
583 /* Skip "=\"" following opt_name */
584 opt_pos
+= strlen(opt_name
) + 2;
585 if (opt_pos
>= end
) {
589 while (opt_end
< end
&& *opt_end
!= '"') {
592 if (opt_end
== end
|| buf_size
< opt_end
- opt_pos
+ 1) {
595 pstrcpy(buf
, opt_end
- opt_pos
+ 1, opt_pos
);
599 /* Open an extent file and append to bs array */
600 static int vmdk_open_sparse(BlockDriverState
*bs
,
601 BlockDriverState
*file
,
606 if (bdrv_pread(file
, 0, &magic
, sizeof(magic
)) != sizeof(magic
)) {
610 magic
= be32_to_cpu(magic
);
613 return vmdk_open_vmdk3(bs
, file
, flags
);
616 return vmdk_open_vmdk4(bs
, file
, flags
);
624 static int vmdk_parse_extents(const char *desc
, BlockDriverState
*bs
,
625 const char *desc_file_path
)
631 const char *p
= desc
;
634 char extent_path
[PATH_MAX
];
635 BlockDriverState
*extent_file
;
638 /* parse extent line:
639 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
641 * RW [size in sectors] SPARSE "file-name.vmdk"
644 ret
= sscanf(p
, "%10s %" SCNd64
" %10s %511s %" SCNd64
,
645 access
, §ors
, type
, fname
, &flat_offset
);
646 if (ret
< 4 || strcmp(access
, "RW")) {
648 } else if (!strcmp(type
, "FLAT")) {
649 if (ret
!= 5 || flat_offset
< 0) {
652 } else if (ret
!= 4) {
656 /* trim the quotation marks around */
657 if (fname
[0] == '"') {
658 memmove(fname
, fname
+ 1, strlen(fname
));
659 if (strlen(fname
) <= 1 || fname
[strlen(fname
) - 1] != '"') {
662 fname
[strlen(fname
) - 1] = '\0';
665 (strcmp(type
, "FLAT") && strcmp(type
, "SPARSE")) ||
666 (strcmp(access
, "RW"))) {
670 path_combine(extent_path
, sizeof(extent_path
),
671 desc_file_path
, fname
);
672 ret
= bdrv_file_open(&extent_file
, extent_path
, bs
->open_flags
);
677 /* save to extents array */
678 if (!strcmp(type
, "FLAT")) {
682 extent
= vmdk_add_extent(bs
, extent_file
, true, sectors
,
683 0, 0, 0, 0, sectors
);
684 extent
->flat_start_offset
= flat_offset
<< 9;
685 } else if (!strcmp(type
, "SPARSE")) {
687 ret
= vmdk_open_sparse(bs
, extent_file
, bs
->open_flags
);
689 bdrv_delete(extent_file
);
694 "VMDK: Not supported extent type \"%s\""".\n", type
);
698 /* move to next line */
699 while (*p
&& *p
!= '\n') {
707 static int vmdk_open_desc_file(BlockDriverState
*bs
, int flags
,
713 BDRVVmdkState
*s
= bs
->opaque
;
715 ret
= bdrv_pread(bs
->file
, desc_offset
, buf
, sizeof(buf
));
720 if (vmdk_parse_description(buf
, "createType", ct
, sizeof(ct
))) {
723 if (strcmp(ct
, "monolithicFlat") &&
724 strcmp(ct
, "twoGbMaxExtentSparse") &&
725 strcmp(ct
, "twoGbMaxExtentFlat")) {
727 "VMDK: Not supported image type \"%s\""".\n", ct
);
731 return vmdk_parse_extents(buf
, bs
, bs
->file
->filename
);
734 static int vmdk_open(BlockDriverState
*bs
, int flags
)
737 BDRVVmdkState
*s
= bs
->opaque
;
739 if (vmdk_open_sparse(bs
, bs
->file
, flags
) == 0) {
740 s
->desc_offset
= 0x200;
742 ret
= vmdk_open_desc_file(bs
, flags
, 0);
747 /* try to open parent images, if exist */
748 ret
= vmdk_parent_open(bs
);
752 s
->parent_cid
= vmdk_read_cid(bs
, 1);
753 qemu_co_mutex_init(&s
->lock
);
755 /* Disable migration when VMDK images are used */
756 error_set(&s
->migration_blocker
,
757 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED
,
758 "vmdk", bs
->device_name
, "live migration");
759 migrate_add_blocker(s
->migration_blocker
);
764 vmdk_free_extents(bs
);
768 static int get_whole_cluster(BlockDriverState
*bs
,
770 uint64_t cluster_offset
,
774 /* 128 sectors * 512 bytes each = grain size 64KB */
775 uint8_t whole_grain
[extent
->cluster_sectors
* 512];
777 /* we will be here if it's first write on non-exist grain(cluster).
778 * try to read from parent image, if exist */
779 if (bs
->backing_hd
) {
782 if (!vmdk_is_cid_valid(bs
)) {
786 /* floor offset to cluster */
787 offset
-= offset
% (extent
->cluster_sectors
* 512);
788 ret
= bdrv_read(bs
->backing_hd
, offset
>> 9, whole_grain
,
789 extent
->cluster_sectors
);
794 /* Write grain only into the active image */
795 ret
= bdrv_write(extent
->file
, cluster_offset
, whole_grain
,
796 extent
->cluster_sectors
);
804 static int vmdk_L2update(VmdkExtent
*extent
, VmdkMetaData
*m_data
)
806 /* update L2 table */
807 if (bdrv_pwrite_sync(
809 ((int64_t)m_data
->l2_offset
* 512)
810 + (m_data
->l2_index
* sizeof(m_data
->offset
)),
812 sizeof(m_data
->offset
)
816 /* update backup L2 table */
817 if (extent
->l1_backup_table_offset
!= 0) {
818 m_data
->l2_offset
= extent
->l1_backup_table
[m_data
->l1_index
];
819 if (bdrv_pwrite_sync(
821 ((int64_t)m_data
->l2_offset
* 512)
822 + (m_data
->l2_index
* sizeof(m_data
->offset
)),
823 &(m_data
->offset
), sizeof(m_data
->offset
)
832 static int get_cluster_offset(BlockDriverState
*bs
,
834 VmdkMetaData
*m_data
,
837 uint64_t *cluster_offset
)
839 unsigned int l1_index
, l2_offset
, l2_index
;
841 uint32_t min_count
, *l2_table
, tmp
= 0;
847 *cluster_offset
= extent
->flat_start_offset
;
851 offset
-= (extent
->end_sector
- extent
->sectors
) * SECTOR_SIZE
;
852 l1_index
= (offset
>> 9) / extent
->l1_entry_sectors
;
853 if (l1_index
>= extent
->l1_size
) {
856 l2_offset
= extent
->l1_table
[l1_index
];
860 for (i
= 0; i
< L2_CACHE_SIZE
; i
++) {
861 if (l2_offset
== extent
->l2_cache_offsets
[i
]) {
862 /* increment the hit count */
863 if (++extent
->l2_cache_counts
[i
] == 0xffffffff) {
864 for (j
= 0; j
< L2_CACHE_SIZE
; j
++) {
865 extent
->l2_cache_counts
[j
] >>= 1;
868 l2_table
= extent
->l2_cache
+ (i
* extent
->l2_size
);
872 /* not found: load a new entry in the least used one */
874 min_count
= 0xffffffff;
875 for (i
= 0; i
< L2_CACHE_SIZE
; i
++) {
876 if (extent
->l2_cache_counts
[i
] < min_count
) {
877 min_count
= extent
->l2_cache_counts
[i
];
881 l2_table
= extent
->l2_cache
+ (min_index
* extent
->l2_size
);
884 (int64_t)l2_offset
* 512,
886 extent
->l2_size
* sizeof(uint32_t)
887 ) != extent
->l2_size
* sizeof(uint32_t)) {
891 extent
->l2_cache_offsets
[min_index
] = l2_offset
;
892 extent
->l2_cache_counts
[min_index
] = 1;
894 l2_index
= ((offset
>> 9) / extent
->cluster_sectors
) % extent
->l2_size
;
895 *cluster_offset
= le32_to_cpu(l2_table
[l2_index
]);
897 if (!*cluster_offset
) {
902 /* Avoid the L2 tables update for the images that have snapshots. */
903 *cluster_offset
= bdrv_getlength(extent
->file
);
904 if (!extent
->compressed
) {
907 *cluster_offset
+ (extent
->cluster_sectors
<< 9)
911 *cluster_offset
>>= 9;
912 tmp
= cpu_to_le32(*cluster_offset
);
913 l2_table
[l2_index
] = tmp
;
915 /* First of all we write grain itself, to avoid race condition
916 * that may to corrupt the image.
917 * This problem may occur because of insufficient space on host disk
918 * or inappropriate VM shutdown.
920 if (get_whole_cluster(
921 bs
, extent
, *cluster_offset
, offset
, allocate
) == -1) {
926 m_data
->offset
= tmp
;
927 m_data
->l1_index
= l1_index
;
928 m_data
->l2_index
= l2_index
;
929 m_data
->l2_offset
= l2_offset
;
933 *cluster_offset
<<= 9;
937 static VmdkExtent
*find_extent(BDRVVmdkState
*s
,
938 int64_t sector_num
, VmdkExtent
*start_hint
)
940 VmdkExtent
*extent
= start_hint
;
943 extent
= &s
->extents
[0];
945 while (extent
< &s
->extents
[s
->num_extents
]) {
946 if (sector_num
< extent
->end_sector
) {
954 static int coroutine_fn
vmdk_co_is_allocated(BlockDriverState
*bs
,
955 int64_t sector_num
, int nb_sectors
, int *pnum
)
957 BDRVVmdkState
*s
= bs
->opaque
;
958 int64_t index_in_cluster
, n
, ret
;
962 extent
= find_extent(s
, sector_num
, NULL
);
966 qemu_co_mutex_lock(&s
->lock
);
967 ret
= get_cluster_offset(bs
, extent
, NULL
,
968 sector_num
* 512, 0, &offset
);
969 qemu_co_mutex_unlock(&s
->lock
);
970 /* get_cluster_offset returning 0 means success */
973 index_in_cluster
= sector_num
% extent
->cluster_sectors
;
974 n
= extent
->cluster_sectors
- index_in_cluster
;
975 if (n
> nb_sectors
) {
982 static int vmdk_write_extent(VmdkExtent
*extent
, int64_t cluster_offset
,
983 int64_t offset_in_cluster
, const uint8_t *buf
,
984 int nb_sectors
, int64_t sector_num
)
987 VmdkGrainMarker
*data
= NULL
;
989 const uint8_t *write_buf
= buf
;
990 int write_len
= nb_sectors
* 512;
992 if (extent
->compressed
) {
993 if (!extent
->has_marker
) {
997 buf_len
= (extent
->cluster_sectors
<< 9) * 2;
998 data
= g_malloc(buf_len
+ sizeof(VmdkGrainMarker
));
999 if (compress(data
->data
, &buf_len
, buf
, nb_sectors
<< 9) != Z_OK
||
1004 data
->lba
= sector_num
;
1005 data
->size
= buf_len
;
1006 write_buf
= (uint8_t *)data
;
1007 write_len
= buf_len
+ sizeof(VmdkGrainMarker
);
1009 ret
= bdrv_pwrite(extent
->file
,
1010 cluster_offset
+ offset_in_cluster
,
1013 if (ret
!= write_len
) {
1014 ret
= ret
< 0 ? ret
: -EIO
;
1023 static int vmdk_read_extent(VmdkExtent
*extent
, int64_t cluster_offset
,
1024 int64_t offset_in_cluster
, uint8_t *buf
,
1028 int cluster_bytes
, buf_bytes
;
1029 uint8_t *cluster_buf
, *compressed_data
;
1030 uint8_t *uncomp_buf
;
1032 VmdkGrainMarker
*marker
;
1036 if (!extent
->compressed
) {
1037 ret
= bdrv_pread(extent
->file
,
1038 cluster_offset
+ offset_in_cluster
,
1039 buf
, nb_sectors
* 512);
1040 if (ret
== nb_sectors
* 512) {
1046 cluster_bytes
= extent
->cluster_sectors
* 512;
1047 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1048 buf_bytes
= cluster_bytes
* 2;
1049 cluster_buf
= g_malloc(buf_bytes
);
1050 uncomp_buf
= g_malloc(cluster_bytes
);
1051 ret
= bdrv_pread(extent
->file
,
1053 cluster_buf
, buf_bytes
);
1057 compressed_data
= cluster_buf
;
1058 buf_len
= cluster_bytes
;
1059 data_len
= cluster_bytes
;
1060 if (extent
->has_marker
) {
1061 marker
= (VmdkGrainMarker
*)cluster_buf
;
1062 compressed_data
= marker
->data
;
1063 data_len
= le32_to_cpu(marker
->size
);
1065 if (!data_len
|| data_len
> buf_bytes
) {
1069 ret
= uncompress(uncomp_buf
, &buf_len
, compressed_data
, data_len
);
1075 if (offset_in_cluster
< 0 ||
1076 offset_in_cluster
+ nb_sectors
* 512 > buf_len
) {
1080 memcpy(buf
, uncomp_buf
+ offset_in_cluster
, nb_sectors
* 512);
1085 g_free(cluster_buf
);
1089 static int vmdk_read(BlockDriverState
*bs
, int64_t sector_num
,
1090 uint8_t *buf
, int nb_sectors
)
1092 BDRVVmdkState
*s
= bs
->opaque
;
1094 uint64_t n
, index_in_cluster
;
1095 uint64_t extent_begin_sector
, extent_relative_sector_num
;
1096 VmdkExtent
*extent
= NULL
;
1097 uint64_t cluster_offset
;
1099 while (nb_sectors
> 0) {
1100 extent
= find_extent(s
, sector_num
, extent
);
1104 ret
= get_cluster_offset(
1106 sector_num
<< 9, 0, &cluster_offset
);
1107 extent_begin_sector
= extent
->end_sector
- extent
->sectors
;
1108 extent_relative_sector_num
= sector_num
- extent_begin_sector
;
1109 index_in_cluster
= extent_relative_sector_num
% extent
->cluster_sectors
;
1110 n
= extent
->cluster_sectors
- index_in_cluster
;
1111 if (n
> nb_sectors
) {
1115 /* if not allocated, try to read from parent image, if exist */
1116 if (bs
->backing_hd
) {
1117 if (!vmdk_is_cid_valid(bs
)) {
1120 ret
= bdrv_read(bs
->backing_hd
, sector_num
, buf
, n
);
1125 memset(buf
, 0, 512 * n
);
1128 ret
= vmdk_read_extent(extent
,
1129 cluster_offset
, index_in_cluster
* 512,
1142 static coroutine_fn
int vmdk_co_read(BlockDriverState
*bs
, int64_t sector_num
,
1143 uint8_t *buf
, int nb_sectors
)
1146 BDRVVmdkState
*s
= bs
->opaque
;
1147 qemu_co_mutex_lock(&s
->lock
);
1148 ret
= vmdk_read(bs
, sector_num
, buf
, nb_sectors
);
1149 qemu_co_mutex_unlock(&s
->lock
);
1153 static int vmdk_write(BlockDriverState
*bs
, int64_t sector_num
,
1154 const uint8_t *buf
, int nb_sectors
)
1156 BDRVVmdkState
*s
= bs
->opaque
;
1157 VmdkExtent
*extent
= NULL
;
1159 int64_t index_in_cluster
;
1160 uint64_t extent_begin_sector
, extent_relative_sector_num
;
1161 uint64_t cluster_offset
;
1162 VmdkMetaData m_data
;
1164 if (sector_num
> bs
->total_sectors
) {
1166 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
1167 " total_sectors=0x%" PRIx64
"\n",
1168 sector_num
, bs
->total_sectors
);
1172 while (nb_sectors
> 0) {
1173 extent
= find_extent(s
, sector_num
, extent
);
1177 ret
= get_cluster_offset(
1181 sector_num
<< 9, !extent
->compressed
,
1183 if (extent
->compressed
) {
1185 /* Refuse write to allocated cluster for streamOptimized */
1187 "VMDK: can't write to allocated cluster"
1188 " for streamOptimized\n");
1192 ret
= get_cluster_offset(
1203 extent_begin_sector
= extent
->end_sector
- extent
->sectors
;
1204 extent_relative_sector_num
= sector_num
- extent_begin_sector
;
1205 index_in_cluster
= extent_relative_sector_num
% extent
->cluster_sectors
;
1206 n
= extent
->cluster_sectors
- index_in_cluster
;
1207 if (n
> nb_sectors
) {
1211 ret
= vmdk_write_extent(extent
,
1212 cluster_offset
, index_in_cluster
* 512,
1213 buf
, n
, sector_num
);
1218 /* update L2 tables */
1219 if (vmdk_L2update(extent
, &m_data
) == -1) {
1227 /* update CID on the first write every time the virtual disk is
1229 if (!s
->cid_updated
) {
1230 ret
= vmdk_write_cid(bs
, time(NULL
));
1234 s
->cid_updated
= true;
1240 static coroutine_fn
int vmdk_co_write(BlockDriverState
*bs
, int64_t sector_num
,
1241 const uint8_t *buf
, int nb_sectors
)
1244 BDRVVmdkState
*s
= bs
->opaque
;
1245 qemu_co_mutex_lock(&s
->lock
);
1246 ret
= vmdk_write(bs
, sector_num
, buf
, nb_sectors
);
1247 qemu_co_mutex_unlock(&s
->lock
);
1252 static int vmdk_create_extent(const char *filename
, int64_t filesize
,
1253 bool flat
, bool compress
)
1258 uint32_t tmp
, magic
, grains
, gd_size
, gt_size
, gt_count
;
1260 fd
= qemu_open(filename
,
1261 O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
| O_LARGEFILE
,
1267 ret
= ftruncate(fd
, filesize
);
1273 magic
= cpu_to_be32(VMDK4_MAGIC
);
1274 memset(&header
, 0, sizeof(header
));
1277 3 | (compress
? VMDK4_FLAG_COMPRESS
| VMDK4_FLAG_MARKER
: 0);
1278 header
.compressAlgorithm
= compress
? VMDK4_COMPRESSION_DEFLATE
: 0;
1279 header
.capacity
= filesize
/ 512;
1280 header
.granularity
= 128;
1281 header
.num_gtes_per_gte
= 512;
1283 grains
= (filesize
/ 512 + header
.granularity
- 1) / header
.granularity
;
1284 gt_size
= ((header
.num_gtes_per_gte
* sizeof(uint32_t)) + 511) >> 9;
1286 (grains
+ header
.num_gtes_per_gte
- 1) / header
.num_gtes_per_gte
;
1287 gd_size
= (gt_count
* sizeof(uint32_t) + 511) >> 9;
1289 header
.desc_offset
= 1;
1290 header
.desc_size
= 20;
1291 header
.rgd_offset
= header
.desc_offset
+ header
.desc_size
;
1292 header
.gd_offset
= header
.rgd_offset
+ gd_size
+ (gt_size
* gt_count
);
1293 header
.grain_offset
=
1294 ((header
.gd_offset
+ gd_size
+ (gt_size
* gt_count
) +
1295 header
.granularity
- 1) / header
.granularity
) *
1297 /* swap endianness for all header fields */
1298 header
.version
= cpu_to_le32(header
.version
);
1299 header
.flags
= cpu_to_le32(header
.flags
);
1300 header
.capacity
= cpu_to_le64(header
.capacity
);
1301 header
.granularity
= cpu_to_le64(header
.granularity
);
1302 header
.num_gtes_per_gte
= cpu_to_le32(header
.num_gtes_per_gte
);
1303 header
.desc_offset
= cpu_to_le64(header
.desc_offset
);
1304 header
.desc_size
= cpu_to_le64(header
.desc_size
);
1305 header
.rgd_offset
= cpu_to_le64(header
.rgd_offset
);
1306 header
.gd_offset
= cpu_to_le64(header
.gd_offset
);
1307 header
.grain_offset
= cpu_to_le64(header
.grain_offset
);
1308 header
.compressAlgorithm
= cpu_to_le16(header
.compressAlgorithm
);
1310 header
.check_bytes
[0] = 0xa;
1311 header
.check_bytes
[1] = 0x20;
1312 header
.check_bytes
[2] = 0xd;
1313 header
.check_bytes
[3] = 0xa;
1315 /* write all the data */
1316 ret
= qemu_write_full(fd
, &magic
, sizeof(magic
));
1317 if (ret
!= sizeof(magic
)) {
1321 ret
= qemu_write_full(fd
, &header
, sizeof(header
));
1322 if (ret
!= sizeof(header
)) {
1327 ret
= ftruncate(fd
, le64_to_cpu(header
.grain_offset
) << 9);
1333 /* write grain directory */
1334 lseek(fd
, le64_to_cpu(header
.rgd_offset
) << 9, SEEK_SET
);
1335 for (i
= 0, tmp
= le64_to_cpu(header
.rgd_offset
) + gd_size
;
1336 i
< gt_count
; i
++, tmp
+= gt_size
) {
1337 ret
= qemu_write_full(fd
, &tmp
, sizeof(tmp
));
1338 if (ret
!= sizeof(tmp
)) {
1344 /* write backup grain directory */
1345 lseek(fd
, le64_to_cpu(header
.gd_offset
) << 9, SEEK_SET
);
1346 for (i
= 0, tmp
= le64_to_cpu(header
.gd_offset
) + gd_size
;
1347 i
< gt_count
; i
++, tmp
+= gt_size
) {
1348 ret
= qemu_write_full(fd
, &tmp
, sizeof(tmp
));
1349 if (ret
!= sizeof(tmp
)) {
1361 static int filename_decompose(const char *filename
, char *path
, char *prefix
,
1362 char *postfix
, size_t buf_len
)
1366 if (filename
== NULL
|| !strlen(filename
)) {
1367 fprintf(stderr
, "Vmdk: no filename provided.\n");
1370 p
= strrchr(filename
, '/');
1372 p
= strrchr(filename
, '\\');
1375 p
= strrchr(filename
, ':');
1379 if (p
- filename
>= buf_len
) {
1382 pstrcpy(path
, p
- filename
+ 1, filename
);
1387 q
= strrchr(p
, '.');
1389 pstrcpy(prefix
, buf_len
, p
);
1392 if (q
- p
>= buf_len
) {
1395 pstrcpy(prefix
, q
- p
+ 1, p
);
1396 pstrcpy(postfix
, buf_len
, q
);
1401 static int relative_path(char *dest
, int dest_size
,
1402 const char *base
, const char *target
)
1408 const char *sep
= "\\";
1410 const char *sep
= "/";
1413 if (!(dest
&& base
&& target
)) {
1416 if (path_is_absolute(target
)) {
1417 pstrcpy(dest
, dest_size
, target
);
1420 while (base
[i
] == target
[i
]) {
1433 pstrcat(dest
, dest_size
, "..");
1434 pstrcat(dest
, dest_size
, sep
);
1436 pstrcat(dest
, dest_size
, q
);
1440 static int vmdk_create(const char *filename
, QEMUOptionParameter
*options
)
1443 char desc
[BUF_SIZE
];
1444 int64_t total_size
= 0, filesize
;
1445 const char *backing_file
= NULL
;
1446 const char *fmt
= NULL
;
1449 bool flat
, split
, compress
;
1450 char ext_desc_lines
[BUF_SIZE
] = "";
1451 char path
[PATH_MAX
], prefix
[PATH_MAX
], postfix
[PATH_MAX
];
1452 const int64_t split_size
= 0x80000000; /* VMDK has constant split size */
1453 const char *desc_extent_line
;
1454 char parent_desc_line
[BUF_SIZE
] = "";
1455 uint32_t parent_cid
= 0xffffffff;
1456 const char desc_template
[] =
1457 "# Disk DescriptorFile\n"
1461 "createType=\"%s\"\n"
1464 "# Extent description\n"
1467 "# The Disk Data Base\n"
1470 "ddb.virtualHWVersion = \"%d\"\n"
1471 "ddb.geometry.cylinders = \"%" PRId64
"\"\n"
1472 "ddb.geometry.heads = \"16\"\n"
1473 "ddb.geometry.sectors = \"63\"\n"
1474 "ddb.adapterType = \"ide\"\n";
1476 if (filename_decompose(filename
, path
, prefix
, postfix
, PATH_MAX
)) {
1479 /* Read out options */
1480 while (options
&& options
->name
) {
1481 if (!strcmp(options
->name
, BLOCK_OPT_SIZE
)) {
1482 total_size
= options
->value
.n
;
1483 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FILE
)) {
1484 backing_file
= options
->value
.s
;
1485 } else if (!strcmp(options
->name
, BLOCK_OPT_COMPAT6
)) {
1486 flags
|= options
->value
.n
? BLOCK_FLAG_COMPAT6
: 0;
1487 } else if (!strcmp(options
->name
, BLOCK_OPT_SUBFMT
)) {
1488 fmt
= options
->value
.s
;
1493 /* Default format to monolithicSparse */
1494 fmt
= "monolithicSparse";
1495 } else if (strcmp(fmt
, "monolithicFlat") &&
1496 strcmp(fmt
, "monolithicSparse") &&
1497 strcmp(fmt
, "twoGbMaxExtentSparse") &&
1498 strcmp(fmt
, "twoGbMaxExtentFlat") &&
1499 strcmp(fmt
, "streamOptimized")) {
1500 fprintf(stderr
, "VMDK: Unknown subformat: %s\n", fmt
);
1503 split
= !(strcmp(fmt
, "twoGbMaxExtentFlat") &&
1504 strcmp(fmt
, "twoGbMaxExtentSparse"));
1505 flat
= !(strcmp(fmt
, "monolithicFlat") &&
1506 strcmp(fmt
, "twoGbMaxExtentFlat"));
1507 compress
= !strcmp(fmt
, "streamOptimized");
1509 desc_extent_line
= "RW %lld FLAT \"%s\" 0\n";
1511 desc_extent_line
= "RW %lld SPARSE \"%s\"\n";
1513 if (flat
&& backing_file
) {
1514 /* not supporting backing file for flat image */
1518 char parent_filename
[PATH_MAX
];
1519 BlockDriverState
*bs
= bdrv_new("");
1520 ret
= bdrv_open(bs
, backing_file
, 0, NULL
);
1525 if (strcmp(bs
->drv
->format_name
, "vmdk")) {
1529 parent_cid
= vmdk_read_cid(bs
, 0);
1531 relative_path(parent_filename
, sizeof(parent_filename
),
1532 filename
, backing_file
);
1533 snprintf(parent_desc_line
, sizeof(parent_desc_line
),
1534 "parentFileNameHint=\"%s\"", parent_filename
);
1537 /* Create extents */
1538 filesize
= total_size
;
1539 while (filesize
> 0) {
1540 char desc_line
[BUF_SIZE
];
1541 char ext_filename
[PATH_MAX
];
1542 char desc_filename
[PATH_MAX
];
1543 int64_t size
= filesize
;
1545 if (split
&& size
> split_size
) {
1549 snprintf(desc_filename
, sizeof(desc_filename
), "%s-%c%03d%s",
1550 prefix
, flat
? 'f' : 's', ++idx
, postfix
);
1552 snprintf(desc_filename
, sizeof(desc_filename
), "%s-flat%s",
1555 snprintf(desc_filename
, sizeof(desc_filename
), "%s%s",
1558 snprintf(ext_filename
, sizeof(ext_filename
), "%s%s",
1559 path
, desc_filename
);
1561 if (vmdk_create_extent(ext_filename
, size
, flat
, compress
)) {
1566 /* Format description line */
1567 snprintf(desc_line
, sizeof(desc_line
),
1568 desc_extent_line
, size
/ 512, desc_filename
);
1569 pstrcat(ext_desc_lines
, sizeof(ext_desc_lines
), desc_line
);
1571 /* generate descriptor file */
1572 snprintf(desc
, sizeof(desc
), desc_template
,
1573 (unsigned int)time(NULL
),
1578 (flags
& BLOCK_FLAG_COMPAT6
? 6 : 4),
1579 total_size
/ (int64_t)(63 * 16 * 512));
1580 if (split
|| flat
) {
1581 fd
= qemu_open(filename
,
1582 O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
| O_LARGEFILE
,
1585 fd
= qemu_open(filename
,
1586 O_WRONLY
| O_BINARY
| O_LARGEFILE
,
1592 /* the descriptor offset = 0x200 */
1593 if (!split
&& !flat
&& 0x200 != lseek(fd
, 0x200, SEEK_SET
)) {
1597 ret
= qemu_write_full(fd
, desc
, strlen(desc
));
1598 if (ret
!= strlen(desc
)) {
1608 static void vmdk_close(BlockDriverState
*bs
)
1610 BDRVVmdkState
*s
= bs
->opaque
;
1612 vmdk_free_extents(bs
);
1614 migrate_del_blocker(s
->migration_blocker
);
1615 error_free(s
->migration_blocker
);
1618 static coroutine_fn
int vmdk_co_flush(BlockDriverState
*bs
)
1620 BDRVVmdkState
*s
= bs
->opaque
;
1624 for (i
= 0; i
< s
->num_extents
; i
++) {
1625 err
= bdrv_co_flush(s
->extents
[i
].file
);
1633 static int64_t vmdk_get_allocated_file_size(BlockDriverState
*bs
)
1638 BDRVVmdkState
*s
= bs
->opaque
;
1640 ret
= bdrv_get_allocated_file_size(bs
->file
);
1644 for (i
= 0; i
< s
->num_extents
; i
++) {
1645 if (s
->extents
[i
].file
== bs
->file
) {
1648 r
= bdrv_get_allocated_file_size(s
->extents
[i
].file
);
1657 static QEMUOptionParameter vmdk_create_options
[] = {
1659 .name
= BLOCK_OPT_SIZE
,
1661 .help
= "Virtual disk size"
1664 .name
= BLOCK_OPT_BACKING_FILE
,
1666 .help
= "File name of a base image"
1669 .name
= BLOCK_OPT_COMPAT6
,
1671 .help
= "VMDK version 6 image"
1674 .name
= BLOCK_OPT_SUBFMT
,
1677 "VMDK flat extent format, can be one of "
1678 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
1683 static BlockDriver bdrv_vmdk
= {
1684 .format_name
= "vmdk",
1685 .instance_size
= sizeof(BDRVVmdkState
),
1686 .bdrv_probe
= vmdk_probe
,
1687 .bdrv_open
= vmdk_open
,
1688 .bdrv_reopen_prepare
= vmdk_reopen_prepare
,
1689 .bdrv_read
= vmdk_co_read
,
1690 .bdrv_write
= vmdk_co_write
,
1691 .bdrv_close
= vmdk_close
,
1692 .bdrv_create
= vmdk_create
,
1693 .bdrv_co_flush_to_disk
= vmdk_co_flush
,
1694 .bdrv_co_is_allocated
= vmdk_co_is_allocated
,
1695 .bdrv_get_allocated_file_size
= vmdk_get_allocated_file_size
,
1697 .create_options
= vmdk_create_options
,
1700 static void bdrv_vmdk_init(void)
1702 bdrv_register(&bdrv_vmdk
);
1705 block_init(bdrv_vmdk_init
);