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_NL_DETECT (1 << 0)
36 #define VMDK4_FLAG_RGD (1 << 1)
37 /* Zeroed-grain enable bit */
38 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
39 #define VMDK4_FLAG_COMPRESS (1 << 16)
40 #define VMDK4_FLAG_MARKER (1 << 17)
41 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
43 #define VMDK_GTE_ZEROED 0x1
45 /* VMDK internal error codes */
47 #define VMDK_ERROR (-1)
48 /* Cluster not allocated */
49 #define VMDK_UNALLOC (-2)
50 #define VMDK_ZEROED (-3)
52 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
57 uint32_t disk_sectors
;
59 uint32_t l1dir_offset
;
61 uint32_t file_sectors
;
64 uint32_t sectors_per_track
;
65 } QEMU_PACKED VMDK3Header
;
74 /* Number of GrainTableEntries per GrainTable */
75 uint32_t num_gtes_per_gt
;
78 uint64_t grain_offset
;
81 uint16_t compressAlgorithm
;
82 } QEMU_PACKED VMDK4Header
;
84 #define L2_CACHE_SIZE 16
86 typedef struct VmdkExtent
{
87 BlockDriverState
*file
;
95 int64_t flat_start_offset
;
96 int64_t l1_table_offset
;
97 int64_t l1_backup_table_offset
;
99 uint32_t *l1_backup_table
;
100 unsigned int l1_size
;
101 uint32_t l1_entry_sectors
;
103 unsigned int l2_size
;
105 uint32_t l2_cache_offsets
[L2_CACHE_SIZE
];
106 uint32_t l2_cache_counts
[L2_CACHE_SIZE
];
108 int64_t cluster_sectors
;
112 typedef struct BDRVVmdkState
{
114 uint64_t desc_offset
;
120 /* Extent array with num_extents entries, ascend ordered by address */
122 Error
*migration_blocker
;
126 typedef struct VmdkMetaData
{
128 unsigned int l1_index
;
129 unsigned int l2_index
;
130 unsigned int l2_offset
;
132 uint32_t *l2_cache_entry
;
135 typedef struct VmdkGrainMarker
{
139 } QEMU_PACKED VmdkGrainMarker
;
142 MARKER_END_OF_STREAM
= 0,
143 MARKER_GRAIN_TABLE
= 1,
144 MARKER_GRAIN_DIRECTORY
= 2,
148 static int vmdk_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
155 magic
= be32_to_cpu(*(uint32_t *)buf
);
156 if (magic
== VMDK3_MAGIC
||
157 magic
== VMDK4_MAGIC
) {
160 const char *p
= (const char *)buf
;
161 const char *end
= p
+ buf_size
;
164 /* skip comment line */
165 while (p
< end
&& *p
!= '\n') {
172 while (p
< end
&& *p
== ' ') {
175 /* skip '\r' if windows line endings used. */
176 if (p
< end
&& *p
== '\r') {
179 /* only accept blank lines before 'version=' line */
180 if (p
== end
|| *p
!= '\n') {
186 if (end
- p
>= strlen("version=X\n")) {
187 if (strncmp("version=1\n", p
, strlen("version=1\n")) == 0 ||
188 strncmp("version=2\n", p
, strlen("version=2\n")) == 0) {
192 if (end
- p
>= strlen("version=X\r\n")) {
193 if (strncmp("version=1\r\n", p
, strlen("version=1\r\n")) == 0 ||
194 strncmp("version=2\r\n", p
, strlen("version=2\r\n")) == 0) {
204 #define SECTOR_SIZE 512
205 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
206 #define BUF_SIZE 4096
207 #define HEADER_SIZE 512 /* first sector of 512 bytes */
209 static void vmdk_free_extents(BlockDriverState
*bs
)
212 BDRVVmdkState
*s
= bs
->opaque
;
215 for (i
= 0; i
< s
->num_extents
; i
++) {
219 g_free(e
->l1_backup_table
);
221 if (e
->file
!= bs
->file
) {
228 static void vmdk_free_last_extent(BlockDriverState
*bs
)
230 BDRVVmdkState
*s
= bs
->opaque
;
232 if (s
->num_extents
== 0) {
236 s
->extents
= g_realloc(s
->extents
, s
->num_extents
* sizeof(VmdkExtent
));
239 static uint32_t vmdk_read_cid(BlockDriverState
*bs
, int parent
)
241 char desc
[DESC_SIZE
];
242 uint32_t cid
= 0xffffffff;
243 const char *p_name
, *cid_str
;
245 BDRVVmdkState
*s
= bs
->opaque
;
248 ret
= bdrv_pread(bs
->file
, s
->desc_offset
, desc
, DESC_SIZE
);
254 cid_str
= "parentCID";
255 cid_str_size
= sizeof("parentCID");
258 cid_str_size
= sizeof("CID");
261 desc
[DESC_SIZE
- 1] = '\0';
262 p_name
= strstr(desc
, cid_str
);
263 if (p_name
!= NULL
) {
264 p_name
+= cid_str_size
;
265 sscanf(p_name
, "%x", &cid
);
271 static int vmdk_write_cid(BlockDriverState
*bs
, uint32_t cid
)
273 char desc
[DESC_SIZE
], tmp_desc
[DESC_SIZE
];
274 char *p_name
, *tmp_str
;
275 BDRVVmdkState
*s
= bs
->opaque
;
278 ret
= bdrv_pread(bs
->file
, s
->desc_offset
, desc
, DESC_SIZE
);
283 desc
[DESC_SIZE
- 1] = '\0';
284 tmp_str
= strstr(desc
, "parentCID");
285 if (tmp_str
== NULL
) {
289 pstrcpy(tmp_desc
, sizeof(tmp_desc
), tmp_str
);
290 p_name
= strstr(desc
, "CID");
291 if (p_name
!= NULL
) {
292 p_name
+= sizeof("CID");
293 snprintf(p_name
, sizeof(desc
) - (p_name
- desc
), "%x\n", cid
);
294 pstrcat(desc
, sizeof(desc
), tmp_desc
);
297 ret
= bdrv_pwrite_sync(bs
->file
, s
->desc_offset
, desc
, DESC_SIZE
);
305 static int vmdk_is_cid_valid(BlockDriverState
*bs
)
307 BDRVVmdkState
*s
= bs
->opaque
;
308 BlockDriverState
*p_bs
= bs
->backing_hd
;
311 if (!s
->cid_checked
&& p_bs
) {
312 cur_pcid
= vmdk_read_cid(p_bs
, 0);
313 if (s
->parent_cid
!= cur_pcid
) {
318 s
->cid_checked
= true;
323 /* Queue extents, if any, for reopen() */
324 static int vmdk_reopen_prepare(BDRVReopenState
*state
,
325 BlockReopenQueue
*queue
, Error
**errp
)
332 assert(state
!= NULL
);
333 assert(state
->bs
!= NULL
);
336 error_setg(errp
, "No reopen queue for VMDK extents");
340 s
= state
->bs
->opaque
;
344 for (i
= 0; i
< s
->num_extents
; i
++) {
346 if (e
->file
!= state
->bs
->file
) {
347 bdrv_reopen_queue(queue
, e
->file
, state
->flags
);
356 static int vmdk_parent_open(BlockDriverState
*bs
)
359 char desc
[DESC_SIZE
+ 1];
360 BDRVVmdkState
*s
= bs
->opaque
;
363 desc
[DESC_SIZE
] = '\0';
364 ret
= bdrv_pread(bs
->file
, s
->desc_offset
, desc
, DESC_SIZE
);
369 p_name
= strstr(desc
, "parentFileNameHint");
370 if (p_name
!= NULL
) {
373 p_name
+= sizeof("parentFileNameHint") + 1;
374 end_name
= strchr(p_name
, '\"');
375 if (end_name
== NULL
) {
378 if ((end_name
- p_name
) > sizeof(bs
->backing_file
) - 1) {
382 pstrcpy(bs
->backing_file
, end_name
- p_name
+ 1, p_name
);
388 /* Create and append extent to the extent array. Return the added VmdkExtent
389 * address. return NULL if allocation failed. */
390 static int vmdk_add_extent(BlockDriverState
*bs
,
391 BlockDriverState
*file
, bool flat
, int64_t sectors
,
392 int64_t l1_offset
, int64_t l1_backup_offset
,
394 int l2_size
, uint64_t cluster_sectors
,
395 VmdkExtent
**new_extent
,
399 BDRVVmdkState
*s
= bs
->opaque
;
401 if (cluster_sectors
> 0x200000) {
402 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
403 error_setg(errp
, "Invalid granularity, image may be corrupt");
406 if (l1_size
> 512 * 1024 * 1024) {
407 /* Although with big capacity and small l1_entry_sectors, we can get a
408 * big l1_size, we don't want unbounded value to allocate the table.
409 * Limit it to 512M, which is 16PB for default cluster and L2 table
411 error_setg(errp
, "L1 size too big");
415 s
->extents
= g_realloc(s
->extents
,
416 (s
->num_extents
+ 1) * sizeof(VmdkExtent
));
417 extent
= &s
->extents
[s
->num_extents
];
420 memset(extent
, 0, sizeof(VmdkExtent
));
423 extent
->sectors
= sectors
;
424 extent
->l1_table_offset
= l1_offset
;
425 extent
->l1_backup_table_offset
= l1_backup_offset
;
426 extent
->l1_size
= l1_size
;
427 extent
->l1_entry_sectors
= l2_size
* cluster_sectors
;
428 extent
->l2_size
= l2_size
;
429 extent
->cluster_sectors
= flat
? sectors
: cluster_sectors
;
432 bs
->bl
.write_zeroes_alignment
=
433 MAX(bs
->bl
.write_zeroes_alignment
, cluster_sectors
);
435 if (s
->num_extents
> 1) {
436 extent
->end_sector
= (*(extent
- 1)).end_sector
+ extent
->sectors
;
438 extent
->end_sector
= extent
->sectors
;
440 bs
->total_sectors
= extent
->end_sector
;
442 *new_extent
= extent
;
447 static int vmdk_init_tables(BlockDriverState
*bs
, VmdkExtent
*extent
,
453 /* read the L1 table */
454 l1_size
= extent
->l1_size
* sizeof(uint32_t);
455 extent
->l1_table
= g_malloc(l1_size
);
456 ret
= bdrv_pread(extent
->file
,
457 extent
->l1_table_offset
,
461 error_setg_errno(errp
, -ret
,
462 "Could not read l1 table from extent '%s'",
463 extent
->file
->filename
);
466 for (i
= 0; i
< extent
->l1_size
; i
++) {
467 le32_to_cpus(&extent
->l1_table
[i
]);
470 if (extent
->l1_backup_table_offset
) {
471 extent
->l1_backup_table
= g_malloc(l1_size
);
472 ret
= bdrv_pread(extent
->file
,
473 extent
->l1_backup_table_offset
,
474 extent
->l1_backup_table
,
477 error_setg_errno(errp
, -ret
,
478 "Could not read l1 backup table from extent '%s'",
479 extent
->file
->filename
);
482 for (i
= 0; i
< extent
->l1_size
; i
++) {
483 le32_to_cpus(&extent
->l1_backup_table
[i
]);
488 g_malloc(extent
->l2_size
* L2_CACHE_SIZE
* sizeof(uint32_t));
491 g_free(extent
->l1_backup_table
);
493 g_free(extent
->l1_table
);
497 static int vmdk_open_vmfs_sparse(BlockDriverState
*bs
,
498 BlockDriverState
*file
,
499 int flags
, Error
**errp
)
506 ret
= bdrv_pread(file
, sizeof(magic
), &header
, sizeof(header
));
508 error_setg_errno(errp
, -ret
,
509 "Could not read header from file '%s'",
513 ret
= vmdk_add_extent(bs
, file
, false,
514 le32_to_cpu(header
.disk_sectors
),
515 le32_to_cpu(header
.l1dir_offset
) << 9,
517 le32_to_cpu(header
.l1dir_size
),
519 le32_to_cpu(header
.granularity
),
525 ret
= vmdk_init_tables(bs
, extent
, errp
);
527 /* free extent allocated by vmdk_add_extent */
528 vmdk_free_last_extent(bs
);
533 static int vmdk_open_desc_file(BlockDriverState
*bs
, int flags
,
534 uint64_t desc_offset
, Error
**errp
);
536 static int vmdk_open_vmdk4(BlockDriverState
*bs
,
537 BlockDriverState
*file
,
538 int flags
, Error
**errp
)
542 uint32_t l1_size
, l1_entry_sectors
;
545 BDRVVmdkState
*s
= bs
->opaque
;
546 int64_t l1_backup_offset
= 0;
548 ret
= bdrv_pread(file
, sizeof(magic
), &header
, sizeof(header
));
550 error_setg_errno(errp
, -ret
,
551 "Could not read header from file '%s'",
554 if (header
.capacity
== 0) {
555 uint64_t desc_offset
= le64_to_cpu(header
.desc_offset
);
557 return vmdk_open_desc_file(bs
, flags
, desc_offset
<< 9, errp
);
561 if (!s
->create_type
) {
562 s
->create_type
= g_strdup("monolithicSparse");
565 if (le64_to_cpu(header
.gd_offset
) == VMDK4_GD_AT_END
) {
567 * The footer takes precedence over the header, so read it in. The
568 * footer starts at offset -1024 from the end: One sector for the
569 * footer, and another one for the end-of-stream marker.
576 uint8_t pad
[512 - 16];
577 } QEMU_PACKED footer_marker
;
581 uint8_t pad
[512 - 4 - sizeof(VMDK4Header
)];
587 uint8_t pad
[512 - 16];
588 } QEMU_PACKED eos_marker
;
589 } QEMU_PACKED footer
;
591 ret
= bdrv_pread(file
,
592 bs
->file
->total_sectors
* 512 - 1536,
593 &footer
, sizeof(footer
));
598 /* Some sanity checks for the footer */
599 if (be32_to_cpu(footer
.magic
) != VMDK4_MAGIC
||
600 le32_to_cpu(footer
.footer_marker
.size
) != 0 ||
601 le32_to_cpu(footer
.footer_marker
.type
) != MARKER_FOOTER
||
602 le64_to_cpu(footer
.eos_marker
.val
) != 0 ||
603 le32_to_cpu(footer
.eos_marker
.size
) != 0 ||
604 le32_to_cpu(footer
.eos_marker
.type
) != MARKER_END_OF_STREAM
)
609 header
= footer
.header
;
612 if (le32_to_cpu(header
.version
) > 3) {
614 snprintf(buf
, sizeof(buf
), "VMDK version %d",
615 le32_to_cpu(header
.version
));
616 qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE
,
617 bs
->device_name
, "vmdk", buf
);
619 } else if (le32_to_cpu(header
.version
) == 3 && (flags
& BDRV_O_RDWR
)) {
620 /* VMware KB 2064959 explains that version 3 added support for
621 * persistent changed block tracking (CBT), and backup software can
622 * read it as version=1 if it doesn't care about the changed area
623 * information. So we are safe to enable read only. */
624 error_setg(errp
, "VMDK version 3 must be read only");
628 if (le32_to_cpu(header
.num_gtes_per_gt
) > 512) {
629 error_report("L2 table size too big");
633 l1_entry_sectors
= le32_to_cpu(header
.num_gtes_per_gt
)
634 * le64_to_cpu(header
.granularity
);
635 if (l1_entry_sectors
== 0) {
638 l1_size
= (le64_to_cpu(header
.capacity
) + l1_entry_sectors
- 1)
640 if (le32_to_cpu(header
.flags
) & VMDK4_FLAG_RGD
) {
641 l1_backup_offset
= le64_to_cpu(header
.rgd_offset
) << 9;
643 ret
= vmdk_add_extent(bs
, file
, false,
644 le64_to_cpu(header
.capacity
),
645 le64_to_cpu(header
.gd_offset
) << 9,
648 le32_to_cpu(header
.num_gtes_per_gt
),
649 le64_to_cpu(header
.granularity
),
656 le16_to_cpu(header
.compressAlgorithm
) == VMDK4_COMPRESSION_DEFLATE
;
657 extent
->has_marker
= le32_to_cpu(header
.flags
) & VMDK4_FLAG_MARKER
;
658 extent
->version
= le32_to_cpu(header
.version
);
659 extent
->has_zero_grain
= le32_to_cpu(header
.flags
) & VMDK4_FLAG_ZERO_GRAIN
;
660 ret
= vmdk_init_tables(bs
, extent
, errp
);
662 /* free extent allocated by vmdk_add_extent */
663 vmdk_free_last_extent(bs
);
668 /* find an option value out of descriptor file */
669 static int vmdk_parse_description(const char *desc
, const char *opt_name
,
670 char *buf
, int buf_size
)
672 char *opt_pos
, *opt_end
;
673 const char *end
= desc
+ strlen(desc
);
675 opt_pos
= strstr(desc
, opt_name
);
679 /* Skip "=\"" following opt_name */
680 opt_pos
+= strlen(opt_name
) + 2;
681 if (opt_pos
>= end
) {
685 while (opt_end
< end
&& *opt_end
!= '"') {
688 if (opt_end
== end
|| buf_size
< opt_end
- opt_pos
+ 1) {
691 pstrcpy(buf
, opt_end
- opt_pos
+ 1, opt_pos
);
695 /* Open an extent file and append to bs array */
696 static int vmdk_open_sparse(BlockDriverState
*bs
,
697 BlockDriverState
*file
,
698 int flags
, Error
**errp
)
702 if (bdrv_pread(file
, 0, &magic
, sizeof(magic
)) != sizeof(magic
)) {
706 magic
= be32_to_cpu(magic
);
709 return vmdk_open_vmfs_sparse(bs
, file
, flags
, errp
);
712 return vmdk_open_vmdk4(bs
, file
, flags
, errp
);
720 static int vmdk_parse_extents(const char *desc
, BlockDriverState
*bs
,
721 const char *desc_file_path
, Error
**errp
)
727 const char *p
= desc
;
730 char extent_path
[PATH_MAX
];
731 BlockDriverState
*extent_file
;
732 BDRVVmdkState
*s
= bs
->opaque
;
736 /* parse extent line:
737 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
739 * RW [size in sectors] SPARSE "file-name.vmdk"
742 ret
= sscanf(p
, "%10s %" SCNd64
" %10s \"%511[^\n\r\"]\" %" SCNd64
,
743 access
, §ors
, type
, fname
, &flat_offset
);
744 if (ret
< 4 || strcmp(access
, "RW")) {
746 } else if (!strcmp(type
, "FLAT")) {
747 if (ret
!= 5 || flat_offset
< 0) {
748 error_setg(errp
, "Invalid extent lines: \n%s", p
);
751 } else if (!strcmp(type
, "VMFS")) {
753 } else if (ret
!= 4) {
754 error_setg(errp
, "Invalid extent lines: \n%s", p
);
759 (strcmp(type
, "FLAT") && strcmp(type
, "SPARSE") &&
760 strcmp(type
, "VMFS") && strcmp(type
, "VMFSSPARSE")) ||
761 (strcmp(access
, "RW"))) {
765 path_combine(extent_path
, sizeof(extent_path
),
766 desc_file_path
, fname
);
767 ret
= bdrv_file_open(&extent_file
, extent_path
, NULL
, bs
->open_flags
,
773 /* save to extents array */
774 if (!strcmp(type
, "FLAT") || !strcmp(type
, "VMFS")) {
777 ret
= vmdk_add_extent(bs
, extent_file
, true, sectors
,
778 0, 0, 0, 0, 0, &extent
, errp
);
782 extent
->flat_start_offset
= flat_offset
<< 9;
783 } else if (!strcmp(type
, "SPARSE") || !strcmp(type
, "VMFSSPARSE")) {
784 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
785 ret
= vmdk_open_sparse(bs
, extent_file
, bs
->open_flags
, errp
);
787 bdrv_unref(extent_file
);
790 extent
= &s
->extents
[s
->num_extents
- 1];
792 error_setg(errp
, "Unsupported extent type '%s'", type
);
795 extent
->type
= g_strdup(type
);
797 /* move to next line */
809 static int vmdk_open_desc_file(BlockDriverState
*bs
, int flags
,
810 uint64_t desc_offset
, Error
**errp
)
815 BDRVVmdkState
*s
= bs
->opaque
;
818 size
= bdrv_getlength(bs
->file
);
823 size
= MIN(size
, 1 << 20); /* avoid unbounded allocation */
824 buf
= g_malloc0(size
+ 1);
826 ret
= bdrv_pread(bs
->file
, desc_offset
, buf
, size
);
830 if (vmdk_parse_description(buf
, "createType", ct
, sizeof(ct
))) {
834 if (strcmp(ct
, "monolithicFlat") &&
835 strcmp(ct
, "vmfs") &&
836 strcmp(ct
, "vmfsSparse") &&
837 strcmp(ct
, "twoGbMaxExtentSparse") &&
838 strcmp(ct
, "twoGbMaxExtentFlat")) {
839 error_setg(errp
, "Unsupported image type '%s'", ct
);
843 s
->create_type
= g_strdup(ct
);
845 ret
= vmdk_parse_extents(buf
, bs
, bs
->file
->filename
, errp
);
851 static int vmdk_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
855 BDRVVmdkState
*s
= bs
->opaque
;
857 if (vmdk_open_sparse(bs
, bs
->file
, flags
, errp
) == 0) {
858 s
->desc_offset
= 0x200;
860 ret
= vmdk_open_desc_file(bs
, flags
, 0, errp
);
865 /* try to open parent images, if exist */
866 ret
= vmdk_parent_open(bs
);
870 s
->cid
= vmdk_read_cid(bs
, 0);
871 s
->parent_cid
= vmdk_read_cid(bs
, 1);
872 qemu_co_mutex_init(&s
->lock
);
874 /* Disable migration when VMDK images are used */
875 error_set(&s
->migration_blocker
,
876 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED
,
877 "vmdk", bs
->device_name
, "live migration");
878 migrate_add_blocker(s
->migration_blocker
);
883 g_free(s
->create_type
);
884 s
->create_type
= NULL
;
885 vmdk_free_extents(bs
);
889 static int get_whole_cluster(BlockDriverState
*bs
,
891 uint64_t cluster_offset
,
896 uint8_t *whole_grain
= NULL
;
898 /* we will be here if it's first write on non-exist grain(cluster).
899 * try to read from parent image, if exist */
900 if (bs
->backing_hd
) {
902 qemu_blockalign(bs
, extent
->cluster_sectors
<< BDRV_SECTOR_BITS
);
903 if (!vmdk_is_cid_valid(bs
)) {
908 /* floor offset to cluster */
909 offset
-= offset
% (extent
->cluster_sectors
* 512);
910 ret
= bdrv_read(bs
->backing_hd
, offset
>> 9, whole_grain
,
911 extent
->cluster_sectors
);
917 /* Write grain only into the active image */
918 ret
= bdrv_write(extent
->file
, cluster_offset
, whole_grain
,
919 extent
->cluster_sectors
);
926 qemu_vfree(whole_grain
);
930 static int vmdk_L2update(VmdkExtent
*extent
, VmdkMetaData
*m_data
)
933 QEMU_BUILD_BUG_ON(sizeof(offset
) != sizeof(m_data
->offset
));
934 offset
= cpu_to_le32(m_data
->offset
);
935 /* update L2 table */
936 if (bdrv_pwrite_sync(
938 ((int64_t)m_data
->l2_offset
* 512)
939 + (m_data
->l2_index
* sizeof(m_data
->offset
)),
940 &offset
, sizeof(offset
)) < 0) {
943 /* update backup L2 table */
944 if (extent
->l1_backup_table_offset
!= 0) {
945 m_data
->l2_offset
= extent
->l1_backup_table
[m_data
->l1_index
];
946 if (bdrv_pwrite_sync(
948 ((int64_t)m_data
->l2_offset
* 512)
949 + (m_data
->l2_index
* sizeof(m_data
->offset
)),
950 &offset
, sizeof(offset
)) < 0) {
954 if (m_data
->l2_cache_entry
) {
955 *m_data
->l2_cache_entry
= offset
;
961 static int get_cluster_offset(BlockDriverState
*bs
,
963 VmdkMetaData
*m_data
,
966 uint64_t *cluster_offset
)
968 unsigned int l1_index
, l2_offset
, l2_index
;
970 uint32_t min_count
, *l2_table
;
977 *cluster_offset
= extent
->flat_start_offset
;
981 offset
-= (extent
->end_sector
- extent
->sectors
) * SECTOR_SIZE
;
982 l1_index
= (offset
>> 9) / extent
->l1_entry_sectors
;
983 if (l1_index
>= extent
->l1_size
) {
986 l2_offset
= extent
->l1_table
[l1_index
];
990 for (i
= 0; i
< L2_CACHE_SIZE
; i
++) {
991 if (l2_offset
== extent
->l2_cache_offsets
[i
]) {
992 /* increment the hit count */
993 if (++extent
->l2_cache_counts
[i
] == 0xffffffff) {
994 for (j
= 0; j
< L2_CACHE_SIZE
; j
++) {
995 extent
->l2_cache_counts
[j
] >>= 1;
998 l2_table
= extent
->l2_cache
+ (i
* extent
->l2_size
);
1002 /* not found: load a new entry in the least used one */
1004 min_count
= 0xffffffff;
1005 for (i
= 0; i
< L2_CACHE_SIZE
; i
++) {
1006 if (extent
->l2_cache_counts
[i
] < min_count
) {
1007 min_count
= extent
->l2_cache_counts
[i
];
1011 l2_table
= extent
->l2_cache
+ (min_index
* extent
->l2_size
);
1014 (int64_t)l2_offset
* 512,
1016 extent
->l2_size
* sizeof(uint32_t)
1017 ) != extent
->l2_size
* sizeof(uint32_t)) {
1021 extent
->l2_cache_offsets
[min_index
] = l2_offset
;
1022 extent
->l2_cache_counts
[min_index
] = 1;
1024 l2_index
= ((offset
>> 9) / extent
->cluster_sectors
) % extent
->l2_size
;
1025 *cluster_offset
= le32_to_cpu(l2_table
[l2_index
]);
1029 m_data
->l1_index
= l1_index
;
1030 m_data
->l2_index
= l2_index
;
1031 m_data
->offset
= *cluster_offset
;
1032 m_data
->l2_offset
= l2_offset
;
1033 m_data
->l2_cache_entry
= &l2_table
[l2_index
];
1035 if (extent
->has_zero_grain
&& *cluster_offset
== VMDK_GTE_ZEROED
) {
1039 if (!*cluster_offset
|| zeroed
) {
1041 return zeroed
? VMDK_ZEROED
: VMDK_UNALLOC
;
1044 /* Avoid the L2 tables update for the images that have snapshots. */
1045 *cluster_offset
= bdrv_getlength(extent
->file
);
1046 if (!extent
->compressed
) {
1049 *cluster_offset
+ (extent
->cluster_sectors
<< 9)
1053 *cluster_offset
>>= 9;
1054 l2_table
[l2_index
] = cpu_to_le32(*cluster_offset
);
1056 /* First of all we write grain itself, to avoid race condition
1057 * that may to corrupt the image.
1058 * This problem may occur because of insufficient space on host disk
1059 * or inappropriate VM shutdown.
1061 if (get_whole_cluster(
1062 bs
, extent
, *cluster_offset
, offset
, allocate
) == -1) {
1067 m_data
->offset
= *cluster_offset
;
1070 *cluster_offset
<<= 9;
1074 static VmdkExtent
*find_extent(BDRVVmdkState
*s
,
1075 int64_t sector_num
, VmdkExtent
*start_hint
)
1077 VmdkExtent
*extent
= start_hint
;
1080 extent
= &s
->extents
[0];
1082 while (extent
< &s
->extents
[s
->num_extents
]) {
1083 if (sector_num
< extent
->end_sector
) {
1091 static int64_t coroutine_fn
vmdk_co_get_block_status(BlockDriverState
*bs
,
1092 int64_t sector_num
, int nb_sectors
, int *pnum
)
1094 BDRVVmdkState
*s
= bs
->opaque
;
1095 int64_t index_in_cluster
, n
, ret
;
1099 extent
= find_extent(s
, sector_num
, NULL
);
1103 qemu_co_mutex_lock(&s
->lock
);
1104 ret
= get_cluster_offset(bs
, extent
, NULL
,
1105 sector_num
* 512, 0, &offset
);
1106 qemu_co_mutex_unlock(&s
->lock
);
1116 ret
= BDRV_BLOCK_ZERO
;
1119 ret
= BDRV_BLOCK_DATA
;
1120 if (extent
->file
== bs
->file
) {
1121 ret
|= BDRV_BLOCK_OFFSET_VALID
| offset
;
1127 index_in_cluster
= sector_num
% extent
->cluster_sectors
;
1128 n
= extent
->cluster_sectors
- index_in_cluster
;
1129 if (n
> nb_sectors
) {
1136 static int vmdk_write_extent(VmdkExtent
*extent
, int64_t cluster_offset
,
1137 int64_t offset_in_cluster
, const uint8_t *buf
,
1138 int nb_sectors
, int64_t sector_num
)
1141 VmdkGrainMarker
*data
= NULL
;
1143 const uint8_t *write_buf
= buf
;
1144 int write_len
= nb_sectors
* 512;
1146 if (extent
->compressed
) {
1147 if (!extent
->has_marker
) {
1151 buf_len
= (extent
->cluster_sectors
<< 9) * 2;
1152 data
= g_malloc(buf_len
+ sizeof(VmdkGrainMarker
));
1153 if (compress(data
->data
, &buf_len
, buf
, nb_sectors
<< 9) != Z_OK
||
1158 data
->lba
= sector_num
;
1159 data
->size
= buf_len
;
1160 write_buf
= (uint8_t *)data
;
1161 write_len
= buf_len
+ sizeof(VmdkGrainMarker
);
1163 ret
= bdrv_pwrite(extent
->file
,
1164 cluster_offset
+ offset_in_cluster
,
1167 if (ret
!= write_len
) {
1168 ret
= ret
< 0 ? ret
: -EIO
;
1177 static int vmdk_read_extent(VmdkExtent
*extent
, int64_t cluster_offset
,
1178 int64_t offset_in_cluster
, uint8_t *buf
,
1182 int cluster_bytes
, buf_bytes
;
1183 uint8_t *cluster_buf
, *compressed_data
;
1184 uint8_t *uncomp_buf
;
1186 VmdkGrainMarker
*marker
;
1190 if (!extent
->compressed
) {
1191 ret
= bdrv_pread(extent
->file
,
1192 cluster_offset
+ offset_in_cluster
,
1193 buf
, nb_sectors
* 512);
1194 if (ret
== nb_sectors
* 512) {
1200 cluster_bytes
= extent
->cluster_sectors
* 512;
1201 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1202 buf_bytes
= cluster_bytes
* 2;
1203 cluster_buf
= g_malloc(buf_bytes
);
1204 uncomp_buf
= g_malloc(cluster_bytes
);
1205 ret
= bdrv_pread(extent
->file
,
1207 cluster_buf
, buf_bytes
);
1211 compressed_data
= cluster_buf
;
1212 buf_len
= cluster_bytes
;
1213 data_len
= cluster_bytes
;
1214 if (extent
->has_marker
) {
1215 marker
= (VmdkGrainMarker
*)cluster_buf
;
1216 compressed_data
= marker
->data
;
1217 data_len
= le32_to_cpu(marker
->size
);
1219 if (!data_len
|| data_len
> buf_bytes
) {
1223 ret
= uncompress(uncomp_buf
, &buf_len
, compressed_data
, data_len
);
1229 if (offset_in_cluster
< 0 ||
1230 offset_in_cluster
+ nb_sectors
* 512 > buf_len
) {
1234 memcpy(buf
, uncomp_buf
+ offset_in_cluster
, nb_sectors
* 512);
1239 g_free(cluster_buf
);
1243 static int vmdk_read(BlockDriverState
*bs
, int64_t sector_num
,
1244 uint8_t *buf
, int nb_sectors
)
1246 BDRVVmdkState
*s
= bs
->opaque
;
1248 uint64_t n
, index_in_cluster
;
1249 uint64_t extent_begin_sector
, extent_relative_sector_num
;
1250 VmdkExtent
*extent
= NULL
;
1251 uint64_t cluster_offset
;
1253 while (nb_sectors
> 0) {
1254 extent
= find_extent(s
, sector_num
, extent
);
1258 ret
= get_cluster_offset(
1260 sector_num
<< 9, 0, &cluster_offset
);
1261 extent_begin_sector
= extent
->end_sector
- extent
->sectors
;
1262 extent_relative_sector_num
= sector_num
- extent_begin_sector
;
1263 index_in_cluster
= extent_relative_sector_num
% extent
->cluster_sectors
;
1264 n
= extent
->cluster_sectors
- index_in_cluster
;
1265 if (n
> nb_sectors
) {
1268 if (ret
!= VMDK_OK
) {
1269 /* if not allocated, try to read from parent image, if exist */
1270 if (bs
->backing_hd
&& ret
!= VMDK_ZEROED
) {
1271 if (!vmdk_is_cid_valid(bs
)) {
1274 ret
= bdrv_read(bs
->backing_hd
, sector_num
, buf
, n
);
1279 memset(buf
, 0, 512 * n
);
1282 ret
= vmdk_read_extent(extent
,
1283 cluster_offset
, index_in_cluster
* 512,
1296 static coroutine_fn
int vmdk_co_read(BlockDriverState
*bs
, int64_t sector_num
,
1297 uint8_t *buf
, int nb_sectors
)
1300 BDRVVmdkState
*s
= bs
->opaque
;
1301 qemu_co_mutex_lock(&s
->lock
);
1302 ret
= vmdk_read(bs
, sector_num
, buf
, nb_sectors
);
1303 qemu_co_mutex_unlock(&s
->lock
);
1309 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1310 * if possible, otherwise return -ENOTSUP.
1311 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1312 * with each cluster. By dry run we can find if the zero write
1313 * is possible without modifying image data.
1315 * Returns: error code with 0 for success.
1317 static int vmdk_write(BlockDriverState
*bs
, int64_t sector_num
,
1318 const uint8_t *buf
, int nb_sectors
,
1319 bool zeroed
, bool zero_dry_run
)
1321 BDRVVmdkState
*s
= bs
->opaque
;
1322 VmdkExtent
*extent
= NULL
;
1324 int64_t index_in_cluster
;
1325 uint64_t extent_begin_sector
, extent_relative_sector_num
;
1326 uint64_t cluster_offset
;
1327 VmdkMetaData m_data
;
1329 if (sector_num
> bs
->total_sectors
) {
1330 error_report("Wrong offset: sector_num=0x%" PRIx64
1331 " total_sectors=0x%" PRIx64
"\n",
1332 sector_num
, bs
->total_sectors
);
1336 while (nb_sectors
> 0) {
1337 extent
= find_extent(s
, sector_num
, extent
);
1341 ret
= get_cluster_offset(
1345 sector_num
<< 9, !extent
->compressed
,
1347 if (extent
->compressed
) {
1348 if (ret
== VMDK_OK
) {
1349 /* Refuse write to allocated cluster for streamOptimized */
1350 error_report("Could not write to allocated cluster"
1351 " for streamOptimized");
1355 ret
= get_cluster_offset(
1363 if (ret
== VMDK_ERROR
) {
1366 extent_begin_sector
= extent
->end_sector
- extent
->sectors
;
1367 extent_relative_sector_num
= sector_num
- extent_begin_sector
;
1368 index_in_cluster
= extent_relative_sector_num
% extent
->cluster_sectors
;
1369 n
= extent
->cluster_sectors
- index_in_cluster
;
1370 if (n
> nb_sectors
) {
1374 /* Do zeroed write, buf is ignored */
1375 if (extent
->has_zero_grain
&&
1376 index_in_cluster
== 0 &&
1377 n
>= extent
->cluster_sectors
) {
1378 n
= extent
->cluster_sectors
;
1379 if (!zero_dry_run
) {
1380 m_data
.offset
= VMDK_GTE_ZEROED
;
1381 /* update L2 tables */
1382 if (vmdk_L2update(extent
, &m_data
) != VMDK_OK
) {
1390 ret
= vmdk_write_extent(extent
,
1391 cluster_offset
, index_in_cluster
* 512,
1392 buf
, n
, sector_num
);
1397 /* update L2 tables */
1398 if (vmdk_L2update(extent
, &m_data
) != VMDK_OK
) {
1407 /* update CID on the first write every time the virtual disk is
1409 if (!s
->cid_updated
) {
1410 ret
= vmdk_write_cid(bs
, time(NULL
));
1414 s
->cid_updated
= true;
1420 static coroutine_fn
int vmdk_co_write(BlockDriverState
*bs
, int64_t sector_num
,
1421 const uint8_t *buf
, int nb_sectors
)
1424 BDRVVmdkState
*s
= bs
->opaque
;
1425 qemu_co_mutex_lock(&s
->lock
);
1426 ret
= vmdk_write(bs
, sector_num
, buf
, nb_sectors
, false, false);
1427 qemu_co_mutex_unlock(&s
->lock
);
1431 static int coroutine_fn
vmdk_co_write_zeroes(BlockDriverState
*bs
,
1434 BdrvRequestFlags flags
)
1437 BDRVVmdkState
*s
= bs
->opaque
;
1438 qemu_co_mutex_lock(&s
->lock
);
1439 /* write zeroes could fail if sectors not aligned to cluster, test it with
1440 * dry_run == true before really updating image */
1441 ret
= vmdk_write(bs
, sector_num
, NULL
, nb_sectors
, true, true);
1443 ret
= vmdk_write(bs
, sector_num
, NULL
, nb_sectors
, true, false);
1445 qemu_co_mutex_unlock(&s
->lock
);
1449 static int vmdk_create_extent(const char *filename
, int64_t filesize
,
1450 bool flat
, bool compress
, bool zeroed_grain
)
1455 uint32_t tmp
, magic
, grains
, gd_size
, gt_size
, gt_count
;
1457 fd
= qemu_open(filename
,
1458 O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
| O_LARGEFILE
,
1464 ret
= ftruncate(fd
, filesize
);
1470 magic
= cpu_to_be32(VMDK4_MAGIC
);
1471 memset(&header
, 0, sizeof(header
));
1472 header
.version
= zeroed_grain
? 2 : 1;
1473 header
.flags
= VMDK4_FLAG_RGD
| VMDK4_FLAG_NL_DETECT
1474 | (compress
? VMDK4_FLAG_COMPRESS
| VMDK4_FLAG_MARKER
: 0)
1475 | (zeroed_grain
? VMDK4_FLAG_ZERO_GRAIN
: 0);
1476 header
.compressAlgorithm
= compress
? VMDK4_COMPRESSION_DEFLATE
: 0;
1477 header
.capacity
= filesize
/ 512;
1478 header
.granularity
= 128;
1479 header
.num_gtes_per_gt
= 512;
1481 grains
= (filesize
/ 512 + header
.granularity
- 1) / header
.granularity
;
1482 gt_size
= ((header
.num_gtes_per_gt
* sizeof(uint32_t)) + 511) >> 9;
1484 (grains
+ header
.num_gtes_per_gt
- 1) / header
.num_gtes_per_gt
;
1485 gd_size
= (gt_count
* sizeof(uint32_t) + 511) >> 9;
1487 header
.desc_offset
= 1;
1488 header
.desc_size
= 20;
1489 header
.rgd_offset
= header
.desc_offset
+ header
.desc_size
;
1490 header
.gd_offset
= header
.rgd_offset
+ gd_size
+ (gt_size
* gt_count
);
1491 header
.grain_offset
=
1492 ((header
.gd_offset
+ gd_size
+ (gt_size
* gt_count
) +
1493 header
.granularity
- 1) / header
.granularity
) *
1495 /* swap endianness for all header fields */
1496 header
.version
= cpu_to_le32(header
.version
);
1497 header
.flags
= cpu_to_le32(header
.flags
);
1498 header
.capacity
= cpu_to_le64(header
.capacity
);
1499 header
.granularity
= cpu_to_le64(header
.granularity
);
1500 header
.num_gtes_per_gt
= cpu_to_le32(header
.num_gtes_per_gt
);
1501 header
.desc_offset
= cpu_to_le64(header
.desc_offset
);
1502 header
.desc_size
= cpu_to_le64(header
.desc_size
);
1503 header
.rgd_offset
= cpu_to_le64(header
.rgd_offset
);
1504 header
.gd_offset
= cpu_to_le64(header
.gd_offset
);
1505 header
.grain_offset
= cpu_to_le64(header
.grain_offset
);
1506 header
.compressAlgorithm
= cpu_to_le16(header
.compressAlgorithm
);
1508 header
.check_bytes
[0] = 0xa;
1509 header
.check_bytes
[1] = 0x20;
1510 header
.check_bytes
[2] = 0xd;
1511 header
.check_bytes
[3] = 0xa;
1513 /* write all the data */
1514 ret
= qemu_write_full(fd
, &magic
, sizeof(magic
));
1515 if (ret
!= sizeof(magic
)) {
1519 ret
= qemu_write_full(fd
, &header
, sizeof(header
));
1520 if (ret
!= sizeof(header
)) {
1525 ret
= ftruncate(fd
, le64_to_cpu(header
.grain_offset
) << 9);
1531 /* write grain directory */
1532 lseek(fd
, le64_to_cpu(header
.rgd_offset
) << 9, SEEK_SET
);
1533 for (i
= 0, tmp
= le64_to_cpu(header
.rgd_offset
) + gd_size
;
1534 i
< gt_count
; i
++, tmp
+= gt_size
) {
1535 ret
= qemu_write_full(fd
, &tmp
, sizeof(tmp
));
1536 if (ret
!= sizeof(tmp
)) {
1542 /* write backup grain directory */
1543 lseek(fd
, le64_to_cpu(header
.gd_offset
) << 9, SEEK_SET
);
1544 for (i
= 0, tmp
= le64_to_cpu(header
.gd_offset
) + gd_size
;
1545 i
< gt_count
; i
++, tmp
+= gt_size
) {
1546 ret
= qemu_write_full(fd
, &tmp
, sizeof(tmp
));
1547 if (ret
!= sizeof(tmp
)) {
1559 static int filename_decompose(const char *filename
, char *path
, char *prefix
,
1560 char *postfix
, size_t buf_len
, Error
**errp
)
1564 if (filename
== NULL
|| !strlen(filename
)) {
1565 error_setg(errp
, "No filename provided");
1568 p
= strrchr(filename
, '/');
1570 p
= strrchr(filename
, '\\');
1573 p
= strrchr(filename
, ':');
1577 if (p
- filename
>= buf_len
) {
1580 pstrcpy(path
, p
- filename
+ 1, filename
);
1585 q
= strrchr(p
, '.');
1587 pstrcpy(prefix
, buf_len
, p
);
1590 if (q
- p
>= buf_len
) {
1593 pstrcpy(prefix
, q
- p
+ 1, p
);
1594 pstrcpy(postfix
, buf_len
, q
);
1599 static int vmdk_create(const char *filename
, QEMUOptionParameter
*options
,
1604 int64_t total_size
= 0, filesize
;
1605 const char *adapter_type
= NULL
;
1606 const char *backing_file
= NULL
;
1607 const char *fmt
= NULL
;
1610 bool flat
, split
, compress
;
1611 GString
*ext_desc_lines
;
1612 char path
[PATH_MAX
], prefix
[PATH_MAX
], postfix
[PATH_MAX
];
1613 const int64_t split_size
= 0x80000000; /* VMDK has constant split size */
1614 const char *desc_extent_line
;
1615 char parent_desc_line
[BUF_SIZE
] = "";
1616 uint32_t parent_cid
= 0xffffffff;
1617 uint32_t number_heads
= 16;
1618 bool zeroed_grain
= false;
1619 const char desc_template
[] =
1620 "# Disk DescriptorFile\n"
1624 "createType=\"%s\"\n"
1627 "# Extent description\n"
1630 "# The Disk Data Base\n"
1633 "ddb.virtualHWVersion = \"%d\"\n"
1634 "ddb.geometry.cylinders = \"%" PRId64
"\"\n"
1635 "ddb.geometry.heads = \"%d\"\n"
1636 "ddb.geometry.sectors = \"63\"\n"
1637 "ddb.adapterType = \"%s\"\n";
1639 ext_desc_lines
= g_string_new(NULL
);
1641 if (filename_decompose(filename
, path
, prefix
, postfix
, PATH_MAX
, errp
)) {
1645 /* Read out options */
1646 while (options
&& options
->name
) {
1647 if (!strcmp(options
->name
, BLOCK_OPT_SIZE
)) {
1648 total_size
= options
->value
.n
;
1649 } else if (!strcmp(options
->name
, BLOCK_OPT_ADAPTER_TYPE
)) {
1650 adapter_type
= options
->value
.s
;
1651 } else if (!strcmp(options
->name
, BLOCK_OPT_BACKING_FILE
)) {
1652 backing_file
= options
->value
.s
;
1653 } else if (!strcmp(options
->name
, BLOCK_OPT_COMPAT6
)) {
1654 flags
|= options
->value
.n
? BLOCK_FLAG_COMPAT6
: 0;
1655 } else if (!strcmp(options
->name
, BLOCK_OPT_SUBFMT
)) {
1656 fmt
= options
->value
.s
;
1657 } else if (!strcmp(options
->name
, BLOCK_OPT_ZEROED_GRAIN
)) {
1658 zeroed_grain
|= options
->value
.n
;
1662 if (!adapter_type
) {
1663 adapter_type
= "ide";
1664 } else if (strcmp(adapter_type
, "ide") &&
1665 strcmp(adapter_type
, "buslogic") &&
1666 strcmp(adapter_type
, "lsilogic") &&
1667 strcmp(adapter_type
, "legacyESX")) {
1668 error_setg(errp
, "Unknown adapter type: '%s'", adapter_type
);
1672 if (strcmp(adapter_type
, "ide") != 0) {
1673 /* that's the number of heads with which vmware operates when
1674 creating, exporting, etc. vmdk files with a non-ide adapter type */
1678 /* Default format to monolithicSparse */
1679 fmt
= "monolithicSparse";
1680 } else if (strcmp(fmt
, "monolithicFlat") &&
1681 strcmp(fmt
, "monolithicSparse") &&
1682 strcmp(fmt
, "twoGbMaxExtentSparse") &&
1683 strcmp(fmt
, "twoGbMaxExtentFlat") &&
1684 strcmp(fmt
, "streamOptimized")) {
1685 error_setg(errp
, "Unknown subformat: '%s'", fmt
);
1689 split
= !(strcmp(fmt
, "twoGbMaxExtentFlat") &&
1690 strcmp(fmt
, "twoGbMaxExtentSparse"));
1691 flat
= !(strcmp(fmt
, "monolithicFlat") &&
1692 strcmp(fmt
, "twoGbMaxExtentFlat"));
1693 compress
= !strcmp(fmt
, "streamOptimized");
1695 desc_extent_line
= "RW %lld FLAT \"%s\" 0\n";
1697 desc_extent_line
= "RW %lld SPARSE \"%s\"\n";
1699 if (flat
&& backing_file
) {
1700 error_setg(errp
, "Flat image can't have backing file");
1704 if (flat
&& zeroed_grain
) {
1705 error_setg(errp
, "Flat image can't enable zeroed grain");
1710 BlockDriverState
*bs
= bdrv_new("");
1711 ret
= bdrv_open(bs
, backing_file
, NULL
, BDRV_O_NO_BACKING
, NULL
, errp
);
1716 if (strcmp(bs
->drv
->format_name
, "vmdk")) {
1721 parent_cid
= vmdk_read_cid(bs
, 0);
1723 snprintf(parent_desc_line
, sizeof(parent_desc_line
),
1724 "parentFileNameHint=\"%s\"", backing_file
);
1727 /* Create extents */
1728 filesize
= total_size
;
1729 while (filesize
> 0) {
1730 char desc_line
[BUF_SIZE
];
1731 char ext_filename
[PATH_MAX
];
1732 char desc_filename
[PATH_MAX
];
1733 int64_t size
= filesize
;
1735 if (split
&& size
> split_size
) {
1739 snprintf(desc_filename
, sizeof(desc_filename
), "%s-%c%03d%s",
1740 prefix
, flat
? 'f' : 's', ++idx
, postfix
);
1742 snprintf(desc_filename
, sizeof(desc_filename
), "%s-flat%s",
1745 snprintf(desc_filename
, sizeof(desc_filename
), "%s%s",
1748 snprintf(ext_filename
, sizeof(ext_filename
), "%s%s",
1749 path
, desc_filename
);
1751 if (vmdk_create_extent(ext_filename
, size
,
1752 flat
, compress
, zeroed_grain
)) {
1758 /* Format description line */
1759 snprintf(desc_line
, sizeof(desc_line
),
1760 desc_extent_line
, size
/ 512, desc_filename
);
1761 g_string_append(ext_desc_lines
, desc_line
);
1763 /* generate descriptor file */
1764 desc
= g_strdup_printf(desc_template
,
1765 (unsigned int)time(NULL
),
1769 ext_desc_lines
->str
,
1770 (flags
& BLOCK_FLAG_COMPAT6
? 6 : 4),
1771 total_size
/ (int64_t)(63 * number_heads
* 512),
1774 if (split
|| flat
) {
1775 fd
= qemu_open(filename
,
1776 O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
| O_LARGEFILE
,
1779 fd
= qemu_open(filename
,
1780 O_WRONLY
| O_BINARY
| O_LARGEFILE
,
1787 /* the descriptor offset = 0x200 */
1788 if (!split
&& !flat
&& 0x200 != lseek(fd
, 0x200, SEEK_SET
)) {
1792 ret
= qemu_write_full(fd
, desc
, strlen(desc
));
1793 if (ret
!= strlen(desc
)) {
1802 g_string_free(ext_desc_lines
, true);
1806 static void vmdk_close(BlockDriverState
*bs
)
1808 BDRVVmdkState
*s
= bs
->opaque
;
1810 vmdk_free_extents(bs
);
1811 g_free(s
->create_type
);
1813 migrate_del_blocker(s
->migration_blocker
);
1814 error_free(s
->migration_blocker
);
1817 static coroutine_fn
int vmdk_co_flush(BlockDriverState
*bs
)
1819 BDRVVmdkState
*s
= bs
->opaque
;
1823 for (i
= 0; i
< s
->num_extents
; i
++) {
1824 err
= bdrv_co_flush(s
->extents
[i
].file
);
1832 static int64_t vmdk_get_allocated_file_size(BlockDriverState
*bs
)
1837 BDRVVmdkState
*s
= bs
->opaque
;
1839 ret
= bdrv_get_allocated_file_size(bs
->file
);
1843 for (i
= 0; i
< s
->num_extents
; i
++) {
1844 if (s
->extents
[i
].file
== bs
->file
) {
1847 r
= bdrv_get_allocated_file_size(s
->extents
[i
].file
);
1856 static int vmdk_has_zero_init(BlockDriverState
*bs
)
1859 BDRVVmdkState
*s
= bs
->opaque
;
1861 /* If has a flat extent and its underlying storage doesn't have zero init,
1863 for (i
= 0; i
< s
->num_extents
; i
++) {
1864 if (s
->extents
[i
].flat
) {
1865 if (!bdrv_has_zero_init(s
->extents
[i
].file
)) {
1873 static ImageInfo
*vmdk_get_extent_info(VmdkExtent
*extent
)
1875 ImageInfo
*info
= g_new0(ImageInfo
, 1);
1877 *info
= (ImageInfo
){
1878 .filename
= g_strdup(extent
->file
->filename
),
1879 .format
= g_strdup(extent
->type
),
1880 .virtual_size
= extent
->sectors
* BDRV_SECTOR_SIZE
,
1881 .compressed
= extent
->compressed
,
1882 .has_compressed
= extent
->compressed
,
1883 .cluster_size
= extent
->cluster_sectors
* BDRV_SECTOR_SIZE
,
1884 .has_cluster_size
= !extent
->flat
,
1890 static ImageInfoSpecific
*vmdk_get_specific_info(BlockDriverState
*bs
)
1893 BDRVVmdkState
*s
= bs
->opaque
;
1894 ImageInfoSpecific
*spec_info
= g_new0(ImageInfoSpecific
, 1);
1895 ImageInfoList
**next
;
1897 *spec_info
= (ImageInfoSpecific
){
1898 .kind
= IMAGE_INFO_SPECIFIC_KIND_VMDK
,
1900 .vmdk
= g_new0(ImageInfoSpecificVmdk
, 1),
1904 *spec_info
->vmdk
= (ImageInfoSpecificVmdk
) {
1905 .create_type
= g_strdup(s
->create_type
),
1907 .parent_cid
= s
->parent_cid
,
1910 next
= &spec_info
->vmdk
->extents
;
1911 for (i
= 0; i
< s
->num_extents
; i
++) {
1912 *next
= g_new0(ImageInfoList
, 1);
1913 (*next
)->value
= vmdk_get_extent_info(&s
->extents
[i
]);
1914 (*next
)->next
= NULL
;
1915 next
= &(*next
)->next
;
1921 static QEMUOptionParameter vmdk_create_options
[] = {
1923 .name
= BLOCK_OPT_SIZE
,
1925 .help
= "Virtual disk size"
1928 .name
= BLOCK_OPT_ADAPTER_TYPE
,
1930 .help
= "Virtual adapter type, can be one of "
1931 "ide (default), lsilogic, buslogic or legacyESX"
1934 .name
= BLOCK_OPT_BACKING_FILE
,
1936 .help
= "File name of a base image"
1939 .name
= BLOCK_OPT_COMPAT6
,
1941 .help
= "VMDK version 6 image"
1944 .name
= BLOCK_OPT_SUBFMT
,
1947 "VMDK flat extent format, can be one of "
1948 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
1951 .name
= BLOCK_OPT_ZEROED_GRAIN
,
1953 .help
= "Enable efficient zero writes using the zeroed-grain GTE feature"
1958 static BlockDriver bdrv_vmdk
= {
1959 .format_name
= "vmdk",
1960 .instance_size
= sizeof(BDRVVmdkState
),
1961 .bdrv_probe
= vmdk_probe
,
1962 .bdrv_open
= vmdk_open
,
1963 .bdrv_reopen_prepare
= vmdk_reopen_prepare
,
1964 .bdrv_read
= vmdk_co_read
,
1965 .bdrv_write
= vmdk_co_write
,
1966 .bdrv_co_write_zeroes
= vmdk_co_write_zeroes
,
1967 .bdrv_close
= vmdk_close
,
1968 .bdrv_create
= vmdk_create
,
1969 .bdrv_co_flush_to_disk
= vmdk_co_flush
,
1970 .bdrv_co_get_block_status
= vmdk_co_get_block_status
,
1971 .bdrv_get_allocated_file_size
= vmdk_get_allocated_file_size
,
1972 .bdrv_has_zero_init
= vmdk_has_zero_init
,
1973 .bdrv_get_specific_info
= vmdk_get_specific_info
,
1975 .create_options
= vmdk_create_options
,
1978 static void bdrv_vmdk_init(void)
1980 bdrv_register(&bdrv_vmdk
);
1983 block_init(bdrv_vmdk_init
);