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"
33 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
34 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
35 #define VMDK4_COMPRESSION_DEFLATE 1
36 #define VMDK4_FLAG_NL_DETECT (1 << 0)
37 #define VMDK4_FLAG_RGD (1 << 1)
38 /* Zeroed-grain enable bit */
39 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
40 #define VMDK4_FLAG_COMPRESS (1 << 16)
41 #define VMDK4_FLAG_MARKER (1 << 17)
42 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
44 #define VMDK_GTE_ZEROED 0x1
46 /* VMDK internal error codes */
48 #define VMDK_ERROR (-1)
49 /* Cluster not allocated */
50 #define VMDK_UNALLOC (-2)
51 #define VMDK_ZEROED (-3)
53 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
58 uint32_t disk_sectors
;
60 uint32_t l1dir_offset
;
62 uint32_t file_sectors
;
65 uint32_t sectors_per_track
;
66 } QEMU_PACKED VMDK3Header
;
75 /* Number of GrainTableEntries per GrainTable */
76 uint32_t num_gtes_per_gt
;
79 uint64_t grain_offset
;
82 uint16_t compressAlgorithm
;
83 } QEMU_PACKED VMDK4Header
;
85 #define L2_CACHE_SIZE 16
87 typedef struct VmdkExtent
{
88 BlockDriverState
*file
;
96 int64_t flat_start_offset
;
97 int64_t l1_table_offset
;
98 int64_t l1_backup_table_offset
;
100 uint32_t *l1_backup_table
;
101 unsigned int l1_size
;
102 uint32_t l1_entry_sectors
;
104 unsigned int l2_size
;
106 uint32_t l2_cache_offsets
[L2_CACHE_SIZE
];
107 uint32_t l2_cache_counts
[L2_CACHE_SIZE
];
109 int64_t cluster_sectors
;
110 int64_t next_cluster_sector
;
114 typedef struct BDRVVmdkState
{
116 uint64_t desc_offset
;
122 /* Extent array with num_extents entries, ascend ordered by address */
124 Error
*migration_blocker
;
128 typedef struct VmdkMetaData
{
129 unsigned int l1_index
;
130 unsigned int l2_index
;
131 unsigned int l2_offset
;
133 uint32_t *l2_cache_entry
;
136 typedef struct VmdkGrainMarker
{
140 } QEMU_PACKED VmdkGrainMarker
;
143 MARKER_END_OF_STREAM
= 0,
144 MARKER_GRAIN_TABLE
= 1,
145 MARKER_GRAIN_DIRECTORY
= 2,
149 static int vmdk_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
156 magic
= be32_to_cpu(*(uint32_t *)buf
);
157 if (magic
== VMDK3_MAGIC
||
158 magic
== VMDK4_MAGIC
) {
161 const char *p
= (const char *)buf
;
162 const char *end
= p
+ buf_size
;
165 /* skip comment line */
166 while (p
< end
&& *p
!= '\n') {
173 while (p
< end
&& *p
== ' ') {
176 /* skip '\r' if windows line endings used. */
177 if (p
< end
&& *p
== '\r') {
180 /* only accept blank lines before 'version=' line */
181 if (p
== end
|| *p
!= '\n') {
187 if (end
- p
>= strlen("version=X\n")) {
188 if (strncmp("version=1\n", p
, strlen("version=1\n")) == 0 ||
189 strncmp("version=2\n", p
, strlen("version=2\n")) == 0) {
193 if (end
- p
>= strlen("version=X\r\n")) {
194 if (strncmp("version=1\r\n", p
, strlen("version=1\r\n")) == 0 ||
195 strncmp("version=2\r\n", p
, strlen("version=2\r\n")) == 0) {
205 #define SECTOR_SIZE 512
206 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
207 #define BUF_SIZE 4096
208 #define HEADER_SIZE 512 /* first sector of 512 bytes */
210 static void vmdk_free_extents(BlockDriverState
*bs
)
213 BDRVVmdkState
*s
= bs
->opaque
;
216 for (i
= 0; i
< s
->num_extents
; i
++) {
220 g_free(e
->l1_backup_table
);
222 if (e
->file
!= bs
->file
) {
229 static void vmdk_free_last_extent(BlockDriverState
*bs
)
231 BDRVVmdkState
*s
= bs
->opaque
;
233 if (s
->num_extents
== 0) {
237 s
->extents
= g_renew(VmdkExtent
, s
->extents
, s
->num_extents
);
240 static uint32_t vmdk_read_cid(BlockDriverState
*bs
, int parent
)
242 char desc
[DESC_SIZE
];
243 uint32_t cid
= 0xffffffff;
244 const char *p_name
, *cid_str
;
246 BDRVVmdkState
*s
= bs
->opaque
;
249 ret
= bdrv_pread(bs
->file
, s
->desc_offset
, desc
, DESC_SIZE
);
255 cid_str
= "parentCID";
256 cid_str_size
= sizeof("parentCID");
259 cid_str_size
= sizeof("CID");
262 desc
[DESC_SIZE
- 1] = '\0';
263 p_name
= strstr(desc
, cid_str
);
264 if (p_name
!= NULL
) {
265 p_name
+= cid_str_size
;
266 sscanf(p_name
, "%" SCNx32
, &cid
);
272 static int vmdk_write_cid(BlockDriverState
*bs
, uint32_t cid
)
274 char desc
[DESC_SIZE
], tmp_desc
[DESC_SIZE
];
275 char *p_name
, *tmp_str
;
276 BDRVVmdkState
*s
= bs
->opaque
;
279 ret
= bdrv_pread(bs
->file
, s
->desc_offset
, desc
, DESC_SIZE
);
284 desc
[DESC_SIZE
- 1] = '\0';
285 tmp_str
= strstr(desc
, "parentCID");
286 if (tmp_str
== NULL
) {
290 pstrcpy(tmp_desc
, sizeof(tmp_desc
), tmp_str
);
291 p_name
= strstr(desc
, "CID");
292 if (p_name
!= NULL
) {
293 p_name
+= sizeof("CID");
294 snprintf(p_name
, sizeof(desc
) - (p_name
- desc
), "%" PRIx32
"\n", cid
);
295 pstrcat(desc
, sizeof(desc
), tmp_desc
);
298 ret
= bdrv_pwrite_sync(bs
->file
, s
->desc_offset
, desc
, DESC_SIZE
);
306 static int vmdk_is_cid_valid(BlockDriverState
*bs
)
308 BDRVVmdkState
*s
= bs
->opaque
;
309 BlockDriverState
*p_bs
= bs
->backing_hd
;
312 if (!s
->cid_checked
&& p_bs
) {
313 cur_pcid
= vmdk_read_cid(p_bs
, 0);
314 if (s
->parent_cid
!= cur_pcid
) {
319 s
->cid_checked
= true;
324 /* Queue extents, if any, for reopen() */
325 static int vmdk_reopen_prepare(BDRVReopenState
*state
,
326 BlockReopenQueue
*queue
, Error
**errp
)
333 assert(state
!= NULL
);
334 assert(state
->bs
!= NULL
);
337 error_setg(errp
, "No reopen queue for VMDK extents");
341 s
= state
->bs
->opaque
;
345 for (i
= 0; i
< s
->num_extents
; i
++) {
347 if (e
->file
!= state
->bs
->file
) {
348 bdrv_reopen_queue(queue
, e
->file
, state
->flags
);
357 static int vmdk_parent_open(BlockDriverState
*bs
)
360 char desc
[DESC_SIZE
+ 1];
361 BDRVVmdkState
*s
= bs
->opaque
;
364 desc
[DESC_SIZE
] = '\0';
365 ret
= bdrv_pread(bs
->file
, s
->desc_offset
, desc
, DESC_SIZE
);
370 p_name
= strstr(desc
, "parentFileNameHint");
371 if (p_name
!= NULL
) {
374 p_name
+= sizeof("parentFileNameHint") + 1;
375 end_name
= strchr(p_name
, '\"');
376 if (end_name
== NULL
) {
379 if ((end_name
- p_name
) > sizeof(bs
->backing_file
) - 1) {
383 pstrcpy(bs
->backing_file
, end_name
- p_name
+ 1, p_name
);
389 /* Create and append extent to the extent array. Return the added VmdkExtent
390 * address. return NULL if allocation failed. */
391 static int vmdk_add_extent(BlockDriverState
*bs
,
392 BlockDriverState
*file
, bool flat
, int64_t sectors
,
393 int64_t l1_offset
, int64_t l1_backup_offset
,
395 int l2_size
, uint64_t cluster_sectors
,
396 VmdkExtent
**new_extent
,
400 BDRVVmdkState
*s
= bs
->opaque
;
403 if (cluster_sectors
> 0x200000) {
404 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
405 error_setg(errp
, "Invalid granularity, image may be corrupt");
408 if (l1_size
> 512 * 1024 * 1024) {
409 /* Although with big capacity and small l1_entry_sectors, we can get a
410 * big l1_size, we don't want unbounded value to allocate the table.
411 * Limit it to 512M, which is 16PB for default cluster and L2 table
413 error_setg(errp
, "L1 size too big");
417 nb_sectors
= bdrv_nb_sectors(file
);
418 if (nb_sectors
< 0) {
422 s
->extents
= g_renew(VmdkExtent
, s
->extents
, s
->num_extents
+ 1);
423 extent
= &s
->extents
[s
->num_extents
];
426 memset(extent
, 0, sizeof(VmdkExtent
));
429 extent
->sectors
= sectors
;
430 extent
->l1_table_offset
= l1_offset
;
431 extent
->l1_backup_table_offset
= l1_backup_offset
;
432 extent
->l1_size
= l1_size
;
433 extent
->l1_entry_sectors
= l2_size
* cluster_sectors
;
434 extent
->l2_size
= l2_size
;
435 extent
->cluster_sectors
= flat
? sectors
: cluster_sectors
;
436 extent
->next_cluster_sector
= ROUND_UP(nb_sectors
, cluster_sectors
);
438 if (s
->num_extents
> 1) {
439 extent
->end_sector
= (*(extent
- 1)).end_sector
+ extent
->sectors
;
441 extent
->end_sector
= extent
->sectors
;
443 bs
->total_sectors
= extent
->end_sector
;
445 *new_extent
= extent
;
450 static int vmdk_init_tables(BlockDriverState
*bs
, VmdkExtent
*extent
,
456 /* read the L1 table */
457 l1_size
= extent
->l1_size
* sizeof(uint32_t);
458 extent
->l1_table
= g_try_malloc(l1_size
);
459 if (l1_size
&& extent
->l1_table
== NULL
) {
463 ret
= bdrv_pread(extent
->file
,
464 extent
->l1_table_offset
,
468 error_setg_errno(errp
, -ret
,
469 "Could not read l1 table from extent '%s'",
470 extent
->file
->filename
);
473 for (i
= 0; i
< extent
->l1_size
; i
++) {
474 le32_to_cpus(&extent
->l1_table
[i
]);
477 if (extent
->l1_backup_table_offset
) {
478 extent
->l1_backup_table
= g_try_malloc(l1_size
);
479 if (l1_size
&& extent
->l1_backup_table
== NULL
) {
483 ret
= bdrv_pread(extent
->file
,
484 extent
->l1_backup_table_offset
,
485 extent
->l1_backup_table
,
488 error_setg_errno(errp
, -ret
,
489 "Could not read l1 backup table from extent '%s'",
490 extent
->file
->filename
);
493 for (i
= 0; i
< extent
->l1_size
; i
++) {
494 le32_to_cpus(&extent
->l1_backup_table
[i
]);
499 g_new(uint32_t, extent
->l2_size
* L2_CACHE_SIZE
);
502 g_free(extent
->l1_backup_table
);
504 g_free(extent
->l1_table
);
508 static int vmdk_open_vmfs_sparse(BlockDriverState
*bs
,
509 BlockDriverState
*file
,
510 int flags
, Error
**errp
)
517 ret
= bdrv_pread(file
, sizeof(magic
), &header
, sizeof(header
));
519 error_setg_errno(errp
, -ret
,
520 "Could not read header from file '%s'",
524 ret
= vmdk_add_extent(bs
, file
, false,
525 le32_to_cpu(header
.disk_sectors
),
526 le32_to_cpu(header
.l1dir_offset
) << 9,
528 le32_to_cpu(header
.l1dir_size
),
530 le32_to_cpu(header
.granularity
),
536 ret
= vmdk_init_tables(bs
, extent
, errp
);
538 /* free extent allocated by vmdk_add_extent */
539 vmdk_free_last_extent(bs
);
544 static int vmdk_open_desc_file(BlockDriverState
*bs
, int flags
, char *buf
,
547 static char *vmdk_read_desc(BlockDriverState
*file
, uint64_t desc_offset
,
554 size
= bdrv_getlength(file
);
556 error_setg_errno(errp
, -size
, "Could not access file");
560 size
= MIN(size
, (1 << 20) - 1); /* avoid unbounded allocation */
561 buf
= g_malloc(size
+ 1);
563 ret
= bdrv_pread(file
, desc_offset
, buf
, size
);
565 error_setg_errno(errp
, -ret
, "Could not read from file");
574 static int vmdk_open_vmdk4(BlockDriverState
*bs
,
575 BlockDriverState
*file
,
576 int flags
, Error
**errp
)
580 uint32_t l1_size
, l1_entry_sectors
;
583 BDRVVmdkState
*s
= bs
->opaque
;
584 int64_t l1_backup_offset
= 0;
586 ret
= bdrv_pread(file
, sizeof(magic
), &header
, sizeof(header
));
588 error_setg_errno(errp
, -ret
,
589 "Could not read header from file '%s'",
593 if (header
.capacity
== 0) {
594 uint64_t desc_offset
= le64_to_cpu(header
.desc_offset
);
596 char *buf
= vmdk_read_desc(file
, desc_offset
<< 9, errp
);
600 ret
= vmdk_open_desc_file(bs
, flags
, buf
, errp
);
606 if (!s
->create_type
) {
607 s
->create_type
= g_strdup("monolithicSparse");
610 if (le64_to_cpu(header
.gd_offset
) == VMDK4_GD_AT_END
) {
612 * The footer takes precedence over the header, so read it in. The
613 * footer starts at offset -1024 from the end: One sector for the
614 * footer, and another one for the end-of-stream marker.
621 uint8_t pad
[512 - 16];
622 } QEMU_PACKED footer_marker
;
626 uint8_t pad
[512 - 4 - sizeof(VMDK4Header
)];
632 uint8_t pad
[512 - 16];
633 } QEMU_PACKED eos_marker
;
634 } QEMU_PACKED footer
;
636 ret
= bdrv_pread(file
,
637 bs
->file
->total_sectors
* 512 - 1536,
638 &footer
, sizeof(footer
));
643 /* Some sanity checks for the footer */
644 if (be32_to_cpu(footer
.magic
) != VMDK4_MAGIC
||
645 le32_to_cpu(footer
.footer_marker
.size
) != 0 ||
646 le32_to_cpu(footer
.footer_marker
.type
) != MARKER_FOOTER
||
647 le64_to_cpu(footer
.eos_marker
.val
) != 0 ||
648 le32_to_cpu(footer
.eos_marker
.size
) != 0 ||
649 le32_to_cpu(footer
.eos_marker
.type
) != MARKER_END_OF_STREAM
)
654 header
= footer
.header
;
657 if (le32_to_cpu(header
.version
) > 3) {
659 snprintf(buf
, sizeof(buf
), "VMDK version %" PRId32
,
660 le32_to_cpu(header
.version
));
661 error_set(errp
, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE
,
662 bdrv_get_device_name(bs
), "vmdk", buf
);
664 } else if (le32_to_cpu(header
.version
) == 3 && (flags
& BDRV_O_RDWR
)) {
665 /* VMware KB 2064959 explains that version 3 added support for
666 * persistent changed block tracking (CBT), and backup software can
667 * read it as version=1 if it doesn't care about the changed area
668 * information. So we are safe to enable read only. */
669 error_setg(errp
, "VMDK version 3 must be read only");
673 if (le32_to_cpu(header
.num_gtes_per_gt
) > 512) {
674 error_setg(errp
, "L2 table size too big");
678 l1_entry_sectors
= le32_to_cpu(header
.num_gtes_per_gt
)
679 * le64_to_cpu(header
.granularity
);
680 if (l1_entry_sectors
== 0) {
683 l1_size
= (le64_to_cpu(header
.capacity
) + l1_entry_sectors
- 1)
685 if (le32_to_cpu(header
.flags
) & VMDK4_FLAG_RGD
) {
686 l1_backup_offset
= le64_to_cpu(header
.rgd_offset
) << 9;
688 if (bdrv_nb_sectors(file
) < le64_to_cpu(header
.grain_offset
)) {
689 error_setg(errp
, "File truncated, expecting at least %" PRId64
" bytes",
690 (int64_t)(le64_to_cpu(header
.grain_offset
)
691 * BDRV_SECTOR_SIZE
));
695 ret
= vmdk_add_extent(bs
, file
, false,
696 le64_to_cpu(header
.capacity
),
697 le64_to_cpu(header
.gd_offset
) << 9,
700 le32_to_cpu(header
.num_gtes_per_gt
),
701 le64_to_cpu(header
.granularity
),
708 le16_to_cpu(header
.compressAlgorithm
) == VMDK4_COMPRESSION_DEFLATE
;
709 if (extent
->compressed
) {
710 g_free(s
->create_type
);
711 s
->create_type
= g_strdup("streamOptimized");
713 extent
->has_marker
= le32_to_cpu(header
.flags
) & VMDK4_FLAG_MARKER
;
714 extent
->version
= le32_to_cpu(header
.version
);
715 extent
->has_zero_grain
= le32_to_cpu(header
.flags
) & VMDK4_FLAG_ZERO_GRAIN
;
716 ret
= vmdk_init_tables(bs
, extent
, errp
);
718 /* free extent allocated by vmdk_add_extent */
719 vmdk_free_last_extent(bs
);
724 /* find an option value out of descriptor file */
725 static int vmdk_parse_description(const char *desc
, const char *opt_name
,
726 char *buf
, int buf_size
)
728 char *opt_pos
, *opt_end
;
729 const char *end
= desc
+ strlen(desc
);
731 opt_pos
= strstr(desc
, opt_name
);
735 /* Skip "=\"" following opt_name */
736 opt_pos
+= strlen(opt_name
) + 2;
737 if (opt_pos
>= end
) {
741 while (opt_end
< end
&& *opt_end
!= '"') {
744 if (opt_end
== end
|| buf_size
< opt_end
- opt_pos
+ 1) {
747 pstrcpy(buf
, opt_end
- opt_pos
+ 1, opt_pos
);
751 /* Open an extent file and append to bs array */
752 static int vmdk_open_sparse(BlockDriverState
*bs
,
753 BlockDriverState
*file
, int flags
,
754 char *buf
, Error
**errp
)
758 magic
= ldl_be_p(buf
);
761 return vmdk_open_vmfs_sparse(bs
, file
, flags
, errp
);
764 return vmdk_open_vmdk4(bs
, file
, flags
, errp
);
767 error_setg(errp
, "Image not in VMDK format");
773 static int vmdk_parse_extents(const char *desc
, BlockDriverState
*bs
,
774 const char *desc_file_path
, Error
**errp
)
780 const char *p
= desc
;
783 char extent_path
[PATH_MAX
];
784 BlockDriverState
*extent_file
;
785 BDRVVmdkState
*s
= bs
->opaque
;
789 /* parse extent line in one of below formats:
791 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
792 * RW [size in sectors] SPARSE "file-name.vmdk"
793 * RW [size in sectors] VMFS "file-name.vmdk"
794 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
797 ret
= sscanf(p
, "%10s %" SCNd64
" %10s \"%511[^\n\r\"]\" %" SCNd64
,
798 access
, §ors
, type
, fname
, &flat_offset
);
799 if (ret
< 4 || strcmp(access
, "RW")) {
801 } else if (!strcmp(type
, "FLAT")) {
802 if (ret
!= 5 || flat_offset
< 0) {
803 error_setg(errp
, "Invalid extent lines: \n%s", p
);
806 } else if (!strcmp(type
, "VMFS")) {
810 error_setg(errp
, "Invalid extent lines:\n%s", p
);
813 } else if (ret
!= 4) {
814 error_setg(errp
, "Invalid extent lines:\n%s", p
);
819 (strcmp(type
, "FLAT") && strcmp(type
, "SPARSE") &&
820 strcmp(type
, "VMFS") && strcmp(type
, "VMFSSPARSE")) ||
821 (strcmp(access
, "RW"))) {
825 path_combine(extent_path
, sizeof(extent_path
),
826 desc_file_path
, fname
);
828 ret
= bdrv_open(&extent_file
, extent_path
, NULL
, NULL
,
829 bs
->open_flags
| BDRV_O_PROTOCOL
, NULL
, errp
);
834 /* save to extents array */
835 if (!strcmp(type
, "FLAT") || !strcmp(type
, "VMFS")) {
838 ret
= vmdk_add_extent(bs
, extent_file
, true, sectors
,
839 0, 0, 0, 0, 0, &extent
, errp
);
841 bdrv_unref(extent_file
);
844 extent
->flat_start_offset
= flat_offset
<< 9;
845 } else if (!strcmp(type
, "SPARSE") || !strcmp(type
, "VMFSSPARSE")) {
846 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
847 char *buf
= vmdk_read_desc(extent_file
, 0, errp
);
851 ret
= vmdk_open_sparse(bs
, extent_file
, bs
->open_flags
, buf
, errp
);
855 bdrv_unref(extent_file
);
858 extent
= &s
->extents
[s
->num_extents
- 1];
860 error_setg(errp
, "Unsupported extent type '%s'", type
);
861 bdrv_unref(extent_file
);
864 extent
->type
= g_strdup(type
);
866 /* move to next line */
878 static int vmdk_open_desc_file(BlockDriverState
*bs
, int flags
, char *buf
,
883 BDRVVmdkState
*s
= bs
->opaque
;
885 if (vmdk_parse_description(buf
, "createType", ct
, sizeof(ct
))) {
886 error_setg(errp
, "invalid VMDK image descriptor");
890 if (strcmp(ct
, "monolithicFlat") &&
891 strcmp(ct
, "vmfs") &&
892 strcmp(ct
, "vmfsSparse") &&
893 strcmp(ct
, "twoGbMaxExtentSparse") &&
894 strcmp(ct
, "twoGbMaxExtentFlat")) {
895 error_setg(errp
, "Unsupported image type '%s'", ct
);
899 s
->create_type
= g_strdup(ct
);
901 ret
= vmdk_parse_extents(buf
, bs
, bs
->file
->filename
, errp
);
906 static int vmdk_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
911 BDRVVmdkState
*s
= bs
->opaque
;
914 buf
= vmdk_read_desc(bs
->file
, 0, errp
);
919 magic
= ldl_be_p(buf
);
923 ret
= vmdk_open_sparse(bs
, bs
->file
, flags
, buf
, errp
);
924 s
->desc_offset
= 0x200;
927 ret
= vmdk_open_desc_file(bs
, flags
, buf
, errp
);
934 /* try to open parent images, if exist */
935 ret
= vmdk_parent_open(bs
);
939 s
->cid
= vmdk_read_cid(bs
, 0);
940 s
->parent_cid
= vmdk_read_cid(bs
, 1);
941 qemu_co_mutex_init(&s
->lock
);
943 /* Disable migration when VMDK images are used */
944 error_set(&s
->migration_blocker
,
945 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED
,
946 "vmdk", bdrv_get_device_name(bs
), "live migration");
947 migrate_add_blocker(s
->migration_blocker
);
953 g_free(s
->create_type
);
954 s
->create_type
= NULL
;
955 vmdk_free_extents(bs
);
960 static void vmdk_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
962 BDRVVmdkState
*s
= bs
->opaque
;
965 for (i
= 0; i
< s
->num_extents
; i
++) {
966 if (!s
->extents
[i
].flat
) {
967 bs
->bl
.write_zeroes_alignment
=
968 MAX(bs
->bl
.write_zeroes_alignment
,
969 s
->extents
[i
].cluster_sectors
);
977 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
978 * to the cluster at @cluster_sector_num.
980 * If @skip_start_sector < @skip_end_sector, the relative range
981 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
982 * it for call to write user data in the request.
984 static int get_whole_cluster(BlockDriverState
*bs
,
986 uint64_t cluster_sector_num
,
988 uint64_t skip_start_sector
,
989 uint64_t skip_end_sector
)
992 int64_t cluster_bytes
;
993 uint8_t *whole_grain
;
995 /* For COW, align request sector_num to cluster start */
996 sector_num
= QEMU_ALIGN_DOWN(sector_num
, extent
->cluster_sectors
);
997 cluster_bytes
= extent
->cluster_sectors
<< BDRV_SECTOR_BITS
;
998 whole_grain
= qemu_blockalign(bs
, cluster_bytes
);
1000 if (!bs
->backing_hd
) {
1001 memset(whole_grain
, 0, skip_start_sector
<< BDRV_SECTOR_BITS
);
1002 memset(whole_grain
+ (skip_end_sector
<< BDRV_SECTOR_BITS
), 0,
1003 cluster_bytes
- (skip_end_sector
<< BDRV_SECTOR_BITS
));
1006 assert(skip_end_sector
<= extent
->cluster_sectors
);
1007 /* we will be here if it's first write on non-exist grain(cluster).
1008 * try to read from parent image, if exist */
1009 if (bs
->backing_hd
&& !vmdk_is_cid_valid(bs
)) {
1014 /* Read backing data before skip range */
1015 if (skip_start_sector
> 0) {
1016 if (bs
->backing_hd
) {
1017 ret
= bdrv_read(bs
->backing_hd
, sector_num
,
1018 whole_grain
, skip_start_sector
);
1024 ret
= bdrv_write(extent
->file
, cluster_sector_num
, whole_grain
,
1031 /* Read backing data after skip range */
1032 if (skip_end_sector
< extent
->cluster_sectors
) {
1033 if (bs
->backing_hd
) {
1034 ret
= bdrv_read(bs
->backing_hd
, sector_num
+ skip_end_sector
,
1035 whole_grain
+ (skip_end_sector
<< BDRV_SECTOR_BITS
),
1036 extent
->cluster_sectors
- skip_end_sector
);
1042 ret
= bdrv_write(extent
->file
, cluster_sector_num
+ skip_end_sector
,
1043 whole_grain
+ (skip_end_sector
<< BDRV_SECTOR_BITS
),
1044 extent
->cluster_sectors
- skip_end_sector
);
1052 qemu_vfree(whole_grain
);
1056 static int vmdk_L2update(VmdkExtent
*extent
, VmdkMetaData
*m_data
,
1059 offset
= cpu_to_le32(offset
);
1060 /* update L2 table */
1061 if (bdrv_pwrite_sync(
1063 ((int64_t)m_data
->l2_offset
* 512)
1064 + (m_data
->l2_index
* sizeof(offset
)),
1065 &offset
, sizeof(offset
)) < 0) {
1068 /* update backup L2 table */
1069 if (extent
->l1_backup_table_offset
!= 0) {
1070 m_data
->l2_offset
= extent
->l1_backup_table
[m_data
->l1_index
];
1071 if (bdrv_pwrite_sync(
1073 ((int64_t)m_data
->l2_offset
* 512)
1074 + (m_data
->l2_index
* sizeof(offset
)),
1075 &offset
, sizeof(offset
)) < 0) {
1079 if (m_data
->l2_cache_entry
) {
1080 *m_data
->l2_cache_entry
= offset
;
1087 * get_cluster_offset
1089 * Look up cluster offset in extent file by sector number, and store in
1092 * For flat extents, the start offset as parsed from the description file is
1095 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1096 * offset for a new cluster and update L2 cache. If there is a backing file,
1097 * COW is done before returning; otherwise, zeroes are written to the allocated
1098 * cluster. Both COW and zero writing skips the sector range
1099 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1100 * has new data to write there.
1102 * Returns: VMDK_OK if cluster exists and mapped in the image.
1103 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1104 * VMDK_ERROR if failed.
1106 static int get_cluster_offset(BlockDriverState
*bs
,
1108 VmdkMetaData
*m_data
,
1111 uint64_t *cluster_offset
,
1112 uint64_t skip_start_sector
,
1113 uint64_t skip_end_sector
)
1115 unsigned int l1_index
, l2_offset
, l2_index
;
1116 int min_index
, i
, j
;
1117 uint32_t min_count
, *l2_table
;
1118 bool zeroed
= false;
1120 int64_t cluster_sector
;
1126 *cluster_offset
= extent
->flat_start_offset
;
1130 offset
-= (extent
->end_sector
- extent
->sectors
) * SECTOR_SIZE
;
1131 l1_index
= (offset
>> 9) / extent
->l1_entry_sectors
;
1132 if (l1_index
>= extent
->l1_size
) {
1135 l2_offset
= extent
->l1_table
[l1_index
];
1137 return VMDK_UNALLOC
;
1139 for (i
= 0; i
< L2_CACHE_SIZE
; i
++) {
1140 if (l2_offset
== extent
->l2_cache_offsets
[i
]) {
1141 /* increment the hit count */
1142 if (++extent
->l2_cache_counts
[i
] == 0xffffffff) {
1143 for (j
= 0; j
< L2_CACHE_SIZE
; j
++) {
1144 extent
->l2_cache_counts
[j
] >>= 1;
1147 l2_table
= extent
->l2_cache
+ (i
* extent
->l2_size
);
1151 /* not found: load a new entry in the least used one */
1153 min_count
= 0xffffffff;
1154 for (i
= 0; i
< L2_CACHE_SIZE
; i
++) {
1155 if (extent
->l2_cache_counts
[i
] < min_count
) {
1156 min_count
= extent
->l2_cache_counts
[i
];
1160 l2_table
= extent
->l2_cache
+ (min_index
* extent
->l2_size
);
1163 (int64_t)l2_offset
* 512,
1165 extent
->l2_size
* sizeof(uint32_t)
1166 ) != extent
->l2_size
* sizeof(uint32_t)) {
1170 extent
->l2_cache_offsets
[min_index
] = l2_offset
;
1171 extent
->l2_cache_counts
[min_index
] = 1;
1173 l2_index
= ((offset
>> 9) / extent
->cluster_sectors
) % extent
->l2_size
;
1174 cluster_sector
= le32_to_cpu(l2_table
[l2_index
]);
1178 m_data
->l1_index
= l1_index
;
1179 m_data
->l2_index
= l2_index
;
1180 m_data
->l2_offset
= l2_offset
;
1181 m_data
->l2_cache_entry
= &l2_table
[l2_index
];
1183 if (extent
->has_zero_grain
&& cluster_sector
== VMDK_GTE_ZEROED
) {
1187 if (!cluster_sector
|| zeroed
) {
1189 return zeroed
? VMDK_ZEROED
: VMDK_UNALLOC
;
1192 cluster_sector
= extent
->next_cluster_sector
;
1193 extent
->next_cluster_sector
+= extent
->cluster_sectors
;
1195 /* First of all we write grain itself, to avoid race condition
1196 * that may to corrupt the image.
1197 * This problem may occur because of insufficient space on host disk
1198 * or inappropriate VM shutdown.
1200 ret
= get_whole_cluster(bs
, extent
,
1202 offset
>> BDRV_SECTOR_BITS
,
1203 skip_start_sector
, skip_end_sector
);
1208 *cluster_offset
= cluster_sector
<< BDRV_SECTOR_BITS
;
1212 static VmdkExtent
*find_extent(BDRVVmdkState
*s
,
1213 int64_t sector_num
, VmdkExtent
*start_hint
)
1215 VmdkExtent
*extent
= start_hint
;
1218 extent
= &s
->extents
[0];
1220 while (extent
< &s
->extents
[s
->num_extents
]) {
1221 if (sector_num
< extent
->end_sector
) {
1229 static int64_t coroutine_fn
vmdk_co_get_block_status(BlockDriverState
*bs
,
1230 int64_t sector_num
, int nb_sectors
, int *pnum
)
1232 BDRVVmdkState
*s
= bs
->opaque
;
1233 int64_t index_in_cluster
, n
, ret
;
1237 extent
= find_extent(s
, sector_num
, NULL
);
1241 qemu_co_mutex_lock(&s
->lock
);
1242 ret
= get_cluster_offset(bs
, extent
, NULL
,
1243 sector_num
* 512, false, &offset
,
1245 qemu_co_mutex_unlock(&s
->lock
);
1255 ret
= BDRV_BLOCK_ZERO
;
1258 ret
= BDRV_BLOCK_DATA
;
1259 if (extent
->file
== bs
->file
&& !extent
->compressed
) {
1260 ret
|= BDRV_BLOCK_OFFSET_VALID
| offset
;
1266 index_in_cluster
= sector_num
% extent
->cluster_sectors
;
1267 n
= extent
->cluster_sectors
- index_in_cluster
;
1268 if (n
> nb_sectors
) {
1275 static int vmdk_write_extent(VmdkExtent
*extent
, int64_t cluster_offset
,
1276 int64_t offset_in_cluster
, const uint8_t *buf
,
1277 int nb_sectors
, int64_t sector_num
)
1280 VmdkGrainMarker
*data
= NULL
;
1282 const uint8_t *write_buf
= buf
;
1283 int write_len
= nb_sectors
* 512;
1285 if (extent
->compressed
) {
1286 if (!extent
->has_marker
) {
1290 buf_len
= (extent
->cluster_sectors
<< 9) * 2;
1291 data
= g_malloc(buf_len
+ sizeof(VmdkGrainMarker
));
1292 if (compress(data
->data
, &buf_len
, buf
, nb_sectors
<< 9) != Z_OK
||
1297 data
->lba
= sector_num
;
1298 data
->size
= buf_len
;
1299 write_buf
= (uint8_t *)data
;
1300 write_len
= buf_len
+ sizeof(VmdkGrainMarker
);
1302 ret
= bdrv_pwrite(extent
->file
,
1303 cluster_offset
+ offset_in_cluster
,
1306 if (ret
!= write_len
) {
1307 ret
= ret
< 0 ? ret
: -EIO
;
1316 static int vmdk_read_extent(VmdkExtent
*extent
, int64_t cluster_offset
,
1317 int64_t offset_in_cluster
, uint8_t *buf
,
1321 int cluster_bytes
, buf_bytes
;
1322 uint8_t *cluster_buf
, *compressed_data
;
1323 uint8_t *uncomp_buf
;
1325 VmdkGrainMarker
*marker
;
1329 if (!extent
->compressed
) {
1330 ret
= bdrv_pread(extent
->file
,
1331 cluster_offset
+ offset_in_cluster
,
1332 buf
, nb_sectors
* 512);
1333 if (ret
== nb_sectors
* 512) {
1339 cluster_bytes
= extent
->cluster_sectors
* 512;
1340 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1341 buf_bytes
= cluster_bytes
* 2;
1342 cluster_buf
= g_malloc(buf_bytes
);
1343 uncomp_buf
= g_malloc(cluster_bytes
);
1344 ret
= bdrv_pread(extent
->file
,
1346 cluster_buf
, buf_bytes
);
1350 compressed_data
= cluster_buf
;
1351 buf_len
= cluster_bytes
;
1352 data_len
= cluster_bytes
;
1353 if (extent
->has_marker
) {
1354 marker
= (VmdkGrainMarker
*)cluster_buf
;
1355 compressed_data
= marker
->data
;
1356 data_len
= le32_to_cpu(marker
->size
);
1358 if (!data_len
|| data_len
> buf_bytes
) {
1362 ret
= uncompress(uncomp_buf
, &buf_len
, compressed_data
, data_len
);
1368 if (offset_in_cluster
< 0 ||
1369 offset_in_cluster
+ nb_sectors
* 512 > buf_len
) {
1373 memcpy(buf
, uncomp_buf
+ offset_in_cluster
, nb_sectors
* 512);
1378 g_free(cluster_buf
);
1382 static int vmdk_read(BlockDriverState
*bs
, int64_t sector_num
,
1383 uint8_t *buf
, int nb_sectors
)
1385 BDRVVmdkState
*s
= bs
->opaque
;
1387 uint64_t n
, index_in_cluster
;
1388 uint64_t extent_begin_sector
, extent_relative_sector_num
;
1389 VmdkExtent
*extent
= NULL
;
1390 uint64_t cluster_offset
;
1392 while (nb_sectors
> 0) {
1393 extent
= find_extent(s
, sector_num
, extent
);
1397 ret
= get_cluster_offset(bs
, extent
, NULL
,
1398 sector_num
<< 9, false, &cluster_offset
,
1400 extent_begin_sector
= extent
->end_sector
- extent
->sectors
;
1401 extent_relative_sector_num
= sector_num
- extent_begin_sector
;
1402 index_in_cluster
= extent_relative_sector_num
% extent
->cluster_sectors
;
1403 n
= extent
->cluster_sectors
- index_in_cluster
;
1404 if (n
> nb_sectors
) {
1407 if (ret
!= VMDK_OK
) {
1408 /* if not allocated, try to read from parent image, if exist */
1409 if (bs
->backing_hd
&& ret
!= VMDK_ZEROED
) {
1410 if (!vmdk_is_cid_valid(bs
)) {
1413 ret
= bdrv_read(bs
->backing_hd
, sector_num
, buf
, n
);
1418 memset(buf
, 0, 512 * n
);
1421 ret
= vmdk_read_extent(extent
,
1422 cluster_offset
, index_in_cluster
* 512,
1435 static coroutine_fn
int vmdk_co_read(BlockDriverState
*bs
, int64_t sector_num
,
1436 uint8_t *buf
, int nb_sectors
)
1439 BDRVVmdkState
*s
= bs
->opaque
;
1440 qemu_co_mutex_lock(&s
->lock
);
1441 ret
= vmdk_read(bs
, sector_num
, buf
, nb_sectors
);
1442 qemu_co_mutex_unlock(&s
->lock
);
1448 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1449 * if possible, otherwise return -ENOTSUP.
1450 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1451 * with each cluster. By dry run we can find if the zero write
1452 * is possible without modifying image data.
1454 * Returns: error code with 0 for success.
1456 static int vmdk_write(BlockDriverState
*bs
, int64_t sector_num
,
1457 const uint8_t *buf
, int nb_sectors
,
1458 bool zeroed
, bool zero_dry_run
)
1460 BDRVVmdkState
*s
= bs
->opaque
;
1461 VmdkExtent
*extent
= NULL
;
1463 int64_t index_in_cluster
, n
;
1464 uint64_t extent_begin_sector
, extent_relative_sector_num
;
1465 uint64_t cluster_offset
;
1466 VmdkMetaData m_data
;
1468 if (sector_num
> bs
->total_sectors
) {
1469 error_report("Wrong offset: sector_num=0x%" PRIx64
1470 " total_sectors=0x%" PRIx64
"\n",
1471 sector_num
, bs
->total_sectors
);
1475 while (nb_sectors
> 0) {
1476 extent
= find_extent(s
, sector_num
, extent
);
1480 extent_begin_sector
= extent
->end_sector
- extent
->sectors
;
1481 extent_relative_sector_num
= sector_num
- extent_begin_sector
;
1482 index_in_cluster
= extent_relative_sector_num
% extent
->cluster_sectors
;
1483 n
= extent
->cluster_sectors
- index_in_cluster
;
1484 if (n
> nb_sectors
) {
1487 ret
= get_cluster_offset(bs
, extent
, &m_data
, sector_num
<< 9,
1488 !(extent
->compressed
|| zeroed
),
1490 index_in_cluster
, index_in_cluster
+ n
);
1491 if (extent
->compressed
) {
1492 if (ret
== VMDK_OK
) {
1493 /* Refuse write to allocated cluster for streamOptimized */
1494 error_report("Could not write to allocated cluster"
1495 " for streamOptimized");
1499 ret
= get_cluster_offset(bs
, extent
, &m_data
, sector_num
<< 9,
1500 true, &cluster_offset
, 0, 0);
1503 if (ret
== VMDK_ERROR
) {
1507 /* Do zeroed write, buf is ignored */
1508 if (extent
->has_zero_grain
&&
1509 index_in_cluster
== 0 &&
1510 n
>= extent
->cluster_sectors
) {
1511 n
= extent
->cluster_sectors
;
1512 if (!zero_dry_run
) {
1513 /* update L2 tables */
1514 if (vmdk_L2update(extent
, &m_data
, VMDK_GTE_ZEROED
)
1523 ret
= vmdk_write_extent(extent
,
1524 cluster_offset
, index_in_cluster
* 512,
1525 buf
, n
, sector_num
);
1530 /* update L2 tables */
1531 if (vmdk_L2update(extent
, &m_data
,
1532 cluster_offset
>> BDRV_SECTOR_BITS
)
1542 /* update CID on the first write every time the virtual disk is
1544 if (!s
->cid_updated
) {
1545 ret
= vmdk_write_cid(bs
, g_random_int());
1549 s
->cid_updated
= true;
1555 static coroutine_fn
int vmdk_co_write(BlockDriverState
*bs
, int64_t sector_num
,
1556 const uint8_t *buf
, int nb_sectors
)
1559 BDRVVmdkState
*s
= bs
->opaque
;
1560 qemu_co_mutex_lock(&s
->lock
);
1561 ret
= vmdk_write(bs
, sector_num
, buf
, nb_sectors
, false, false);
1562 qemu_co_mutex_unlock(&s
->lock
);
1566 static int vmdk_write_compressed(BlockDriverState
*bs
,
1571 BDRVVmdkState
*s
= bs
->opaque
;
1572 if (s
->num_extents
== 1 && s
->extents
[0].compressed
) {
1573 return vmdk_write(bs
, sector_num
, buf
, nb_sectors
, false, false);
1579 static int coroutine_fn
vmdk_co_write_zeroes(BlockDriverState
*bs
,
1582 BdrvRequestFlags flags
)
1585 BDRVVmdkState
*s
= bs
->opaque
;
1586 qemu_co_mutex_lock(&s
->lock
);
1587 /* write zeroes could fail if sectors not aligned to cluster, test it with
1588 * dry_run == true before really updating image */
1589 ret
= vmdk_write(bs
, sector_num
, NULL
, nb_sectors
, true, true);
1591 ret
= vmdk_write(bs
, sector_num
, NULL
, nb_sectors
, true, false);
1593 qemu_co_mutex_unlock(&s
->lock
);
1597 static int vmdk_create_extent(const char *filename
, int64_t filesize
,
1598 bool flat
, bool compress
, bool zeroed_grain
,
1599 QemuOpts
*opts
, Error
**errp
)
1602 BlockDriverState
*bs
= NULL
;
1604 Error
*local_err
= NULL
;
1605 uint32_t tmp
, magic
, grains
, gd_sectors
, gt_size
, gt_count
;
1606 uint32_t *gd_buf
= NULL
;
1609 ret
= bdrv_create_file(filename
, opts
, &local_err
);
1611 error_propagate(errp
, local_err
);
1616 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
, BDRV_O_RDWR
| BDRV_O_PROTOCOL
,
1619 error_propagate(errp
, local_err
);
1624 ret
= bdrv_truncate(bs
, filesize
);
1626 error_setg_errno(errp
, -ret
, "Could not truncate file");
1630 magic
= cpu_to_be32(VMDK4_MAGIC
);
1631 memset(&header
, 0, sizeof(header
));
1632 header
.version
= zeroed_grain
? 2 : 1;
1633 header
.flags
= VMDK4_FLAG_RGD
| VMDK4_FLAG_NL_DETECT
1634 | (compress
? VMDK4_FLAG_COMPRESS
| VMDK4_FLAG_MARKER
: 0)
1635 | (zeroed_grain
? VMDK4_FLAG_ZERO_GRAIN
: 0);
1636 header
.compressAlgorithm
= compress
? VMDK4_COMPRESSION_DEFLATE
: 0;
1637 header
.capacity
= filesize
/ BDRV_SECTOR_SIZE
;
1638 header
.granularity
= 128;
1639 header
.num_gtes_per_gt
= BDRV_SECTOR_SIZE
;
1641 grains
= DIV_ROUND_UP(filesize
/ BDRV_SECTOR_SIZE
, header
.granularity
);
1642 gt_size
= DIV_ROUND_UP(header
.num_gtes_per_gt
* sizeof(uint32_t),
1644 gt_count
= DIV_ROUND_UP(grains
, header
.num_gtes_per_gt
);
1645 gd_sectors
= DIV_ROUND_UP(gt_count
* sizeof(uint32_t), BDRV_SECTOR_SIZE
);
1647 header
.desc_offset
= 1;
1648 header
.desc_size
= 20;
1649 header
.rgd_offset
= header
.desc_offset
+ header
.desc_size
;
1650 header
.gd_offset
= header
.rgd_offset
+ gd_sectors
+ (gt_size
* gt_count
);
1651 header
.grain_offset
=
1652 ROUND_UP(header
.gd_offset
+ gd_sectors
+ (gt_size
* gt_count
),
1653 header
.granularity
);
1654 /* swap endianness for all header fields */
1655 header
.version
= cpu_to_le32(header
.version
);
1656 header
.flags
= cpu_to_le32(header
.flags
);
1657 header
.capacity
= cpu_to_le64(header
.capacity
);
1658 header
.granularity
= cpu_to_le64(header
.granularity
);
1659 header
.num_gtes_per_gt
= cpu_to_le32(header
.num_gtes_per_gt
);
1660 header
.desc_offset
= cpu_to_le64(header
.desc_offset
);
1661 header
.desc_size
= cpu_to_le64(header
.desc_size
);
1662 header
.rgd_offset
= cpu_to_le64(header
.rgd_offset
);
1663 header
.gd_offset
= cpu_to_le64(header
.gd_offset
);
1664 header
.grain_offset
= cpu_to_le64(header
.grain_offset
);
1665 header
.compressAlgorithm
= cpu_to_le16(header
.compressAlgorithm
);
1667 header
.check_bytes
[0] = 0xa;
1668 header
.check_bytes
[1] = 0x20;
1669 header
.check_bytes
[2] = 0xd;
1670 header
.check_bytes
[3] = 0xa;
1672 /* write all the data */
1673 ret
= bdrv_pwrite(bs
, 0, &magic
, sizeof(magic
));
1675 error_set(errp
, QERR_IO_ERROR
);
1678 ret
= bdrv_pwrite(bs
, sizeof(magic
), &header
, sizeof(header
));
1680 error_set(errp
, QERR_IO_ERROR
);
1684 ret
= bdrv_truncate(bs
, le64_to_cpu(header
.grain_offset
) << 9);
1686 error_setg_errno(errp
, -ret
, "Could not truncate file");
1690 /* write grain directory */
1691 gd_buf_size
= gd_sectors
* BDRV_SECTOR_SIZE
;
1692 gd_buf
= g_malloc0(gd_buf_size
);
1693 for (i
= 0, tmp
= le64_to_cpu(header
.rgd_offset
) + gd_sectors
;
1694 i
< gt_count
; i
++, tmp
+= gt_size
) {
1695 gd_buf
[i
] = cpu_to_le32(tmp
);
1697 ret
= bdrv_pwrite(bs
, le64_to_cpu(header
.rgd_offset
) * BDRV_SECTOR_SIZE
,
1698 gd_buf
, gd_buf_size
);
1700 error_set(errp
, QERR_IO_ERROR
);
1704 /* write backup grain directory */
1705 for (i
= 0, tmp
= le64_to_cpu(header
.gd_offset
) + gd_sectors
;
1706 i
< gt_count
; i
++, tmp
+= gt_size
) {
1707 gd_buf
[i
] = cpu_to_le32(tmp
);
1709 ret
= bdrv_pwrite(bs
, le64_to_cpu(header
.gd_offset
) * BDRV_SECTOR_SIZE
,
1710 gd_buf
, gd_buf_size
);
1712 error_set(errp
, QERR_IO_ERROR
);
1725 static int filename_decompose(const char *filename
, char *path
, char *prefix
,
1726 char *postfix
, size_t buf_len
, Error
**errp
)
1730 if (filename
== NULL
|| !strlen(filename
)) {
1731 error_setg(errp
, "No filename provided");
1734 p
= strrchr(filename
, '/');
1736 p
= strrchr(filename
, '\\');
1739 p
= strrchr(filename
, ':');
1743 if (p
- filename
>= buf_len
) {
1746 pstrcpy(path
, p
- filename
+ 1, filename
);
1751 q
= strrchr(p
, '.');
1753 pstrcpy(prefix
, buf_len
, p
);
1756 if (q
- p
>= buf_len
) {
1759 pstrcpy(prefix
, q
- p
+ 1, p
);
1760 pstrcpy(postfix
, buf_len
, q
);
1765 static int vmdk_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
1768 BlockDriverState
*new_bs
= NULL
;
1769 Error
*local_err
= NULL
;
1771 int64_t total_size
= 0, filesize
;
1772 char *adapter_type
= NULL
;
1773 char *backing_file
= NULL
;
1777 bool flat
, split
, compress
;
1778 GString
*ext_desc_lines
;
1779 char path
[PATH_MAX
], prefix
[PATH_MAX
], postfix
[PATH_MAX
];
1780 const int64_t split_size
= 0x80000000; /* VMDK has constant split size */
1781 const char *desc_extent_line
;
1782 char parent_desc_line
[BUF_SIZE
] = "";
1783 uint32_t parent_cid
= 0xffffffff;
1784 uint32_t number_heads
= 16;
1785 bool zeroed_grain
= false;
1786 uint32_t desc_offset
= 0, desc_len
;
1787 const char desc_template
[] =
1788 "# Disk DescriptorFile\n"
1791 "parentCID=%" PRIx32
"\n"
1792 "createType=\"%s\"\n"
1795 "# Extent description\n"
1798 "# The Disk Data Base\n"
1801 "ddb.virtualHWVersion = \"%d\"\n"
1802 "ddb.geometry.cylinders = \"%" PRId64
"\"\n"
1803 "ddb.geometry.heads = \"%" PRIu32
"\"\n"
1804 "ddb.geometry.sectors = \"63\"\n"
1805 "ddb.adapterType = \"%s\"\n";
1807 ext_desc_lines
= g_string_new(NULL
);
1809 if (filename_decompose(filename
, path
, prefix
, postfix
, PATH_MAX
, errp
)) {
1813 /* Read out options */
1814 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
1816 adapter_type
= qemu_opt_get_del(opts
, BLOCK_OPT_ADAPTER_TYPE
);
1817 backing_file
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
1818 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_COMPAT6
, false)) {
1819 flags
|= BLOCK_FLAG_COMPAT6
;
1821 fmt
= qemu_opt_get_del(opts
, BLOCK_OPT_SUBFMT
);
1822 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_ZEROED_GRAIN
, false)) {
1823 zeroed_grain
= true;
1826 if (!adapter_type
) {
1827 adapter_type
= g_strdup("ide");
1828 } else if (strcmp(adapter_type
, "ide") &&
1829 strcmp(adapter_type
, "buslogic") &&
1830 strcmp(adapter_type
, "lsilogic") &&
1831 strcmp(adapter_type
, "legacyESX")) {
1832 error_setg(errp
, "Unknown adapter type: '%s'", adapter_type
);
1836 if (strcmp(adapter_type
, "ide") != 0) {
1837 /* that's the number of heads with which vmware operates when
1838 creating, exporting, etc. vmdk files with a non-ide adapter type */
1842 /* Default format to monolithicSparse */
1843 fmt
= g_strdup("monolithicSparse");
1844 } else if (strcmp(fmt
, "monolithicFlat") &&
1845 strcmp(fmt
, "monolithicSparse") &&
1846 strcmp(fmt
, "twoGbMaxExtentSparse") &&
1847 strcmp(fmt
, "twoGbMaxExtentFlat") &&
1848 strcmp(fmt
, "streamOptimized")) {
1849 error_setg(errp
, "Unknown subformat: '%s'", fmt
);
1853 split
= !(strcmp(fmt
, "twoGbMaxExtentFlat") &&
1854 strcmp(fmt
, "twoGbMaxExtentSparse"));
1855 flat
= !(strcmp(fmt
, "monolithicFlat") &&
1856 strcmp(fmt
, "twoGbMaxExtentFlat"));
1857 compress
= !strcmp(fmt
, "streamOptimized");
1859 desc_extent_line
= "RW %" PRId64
" FLAT \"%s\" 0\n";
1861 desc_extent_line
= "RW %" PRId64
" SPARSE \"%s\"\n";
1863 if (flat
&& backing_file
) {
1864 error_setg(errp
, "Flat image can't have backing file");
1868 if (flat
&& zeroed_grain
) {
1869 error_setg(errp
, "Flat image can't enable zeroed grain");
1874 BlockDriverState
*bs
= NULL
;
1875 ret
= bdrv_open(&bs
, backing_file
, NULL
, NULL
, BDRV_O_NO_BACKING
, NULL
,
1880 if (strcmp(bs
->drv
->format_name
, "vmdk")) {
1885 parent_cid
= vmdk_read_cid(bs
, 0);
1887 snprintf(parent_desc_line
, sizeof(parent_desc_line
),
1888 "parentFileNameHint=\"%s\"", backing_file
);
1891 /* Create extents */
1892 filesize
= total_size
;
1893 while (filesize
> 0) {
1894 char desc_line
[BUF_SIZE
];
1895 char ext_filename
[PATH_MAX
];
1896 char desc_filename
[PATH_MAX
];
1897 int64_t size
= filesize
;
1899 if (split
&& size
> split_size
) {
1903 snprintf(desc_filename
, sizeof(desc_filename
), "%s-%c%03d%s",
1904 prefix
, flat
? 'f' : 's', ++idx
, postfix
);
1906 snprintf(desc_filename
, sizeof(desc_filename
), "%s-flat%s",
1909 snprintf(desc_filename
, sizeof(desc_filename
), "%s%s",
1912 snprintf(ext_filename
, sizeof(ext_filename
), "%s%s",
1913 path
, desc_filename
);
1915 if (vmdk_create_extent(ext_filename
, size
,
1916 flat
, compress
, zeroed_grain
, opts
, errp
)) {
1922 /* Format description line */
1923 snprintf(desc_line
, sizeof(desc_line
),
1924 desc_extent_line
, size
/ BDRV_SECTOR_SIZE
, desc_filename
);
1925 g_string_append(ext_desc_lines
, desc_line
);
1927 /* generate descriptor file */
1928 desc
= g_strdup_printf(desc_template
,
1933 ext_desc_lines
->str
,
1934 (flags
& BLOCK_FLAG_COMPAT6
? 6 : 4),
1936 (int64_t)(63 * number_heads
* BDRV_SECTOR_SIZE
),
1939 desc_len
= strlen(desc
);
1940 /* the descriptor offset = 0x200 */
1941 if (!split
&& !flat
) {
1942 desc_offset
= 0x200;
1944 ret
= bdrv_create_file(filename
, opts
, &local_err
);
1946 error_propagate(errp
, local_err
);
1950 assert(new_bs
== NULL
);
1951 ret
= bdrv_open(&new_bs
, filename
, NULL
, NULL
,
1952 BDRV_O_RDWR
| BDRV_O_PROTOCOL
, NULL
, &local_err
);
1954 error_propagate(errp
, local_err
);
1957 ret
= bdrv_pwrite(new_bs
, desc_offset
, desc
, desc_len
);
1959 error_setg_errno(errp
, -ret
, "Could not write description");
1962 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
1963 * for description file */
1964 if (desc_offset
== 0) {
1965 ret
= bdrv_truncate(new_bs
, desc_len
);
1967 error_setg_errno(errp
, -ret
, "Could not truncate file");
1974 g_free(adapter_type
);
1975 g_free(backing_file
);
1978 g_string_free(ext_desc_lines
, true);
1982 static void vmdk_close(BlockDriverState
*bs
)
1984 BDRVVmdkState
*s
= bs
->opaque
;
1986 vmdk_free_extents(bs
);
1987 g_free(s
->create_type
);
1989 migrate_del_blocker(s
->migration_blocker
);
1990 error_free(s
->migration_blocker
);
1993 static coroutine_fn
int vmdk_co_flush(BlockDriverState
*bs
)
1995 BDRVVmdkState
*s
= bs
->opaque
;
1999 for (i
= 0; i
< s
->num_extents
; i
++) {
2000 err
= bdrv_co_flush(s
->extents
[i
].file
);
2008 static int64_t vmdk_get_allocated_file_size(BlockDriverState
*bs
)
2013 BDRVVmdkState
*s
= bs
->opaque
;
2015 ret
= bdrv_get_allocated_file_size(bs
->file
);
2019 for (i
= 0; i
< s
->num_extents
; i
++) {
2020 if (s
->extents
[i
].file
== bs
->file
) {
2023 r
= bdrv_get_allocated_file_size(s
->extents
[i
].file
);
2032 static int vmdk_has_zero_init(BlockDriverState
*bs
)
2035 BDRVVmdkState
*s
= bs
->opaque
;
2037 /* If has a flat extent and its underlying storage doesn't have zero init,
2039 for (i
= 0; i
< s
->num_extents
; i
++) {
2040 if (s
->extents
[i
].flat
) {
2041 if (!bdrv_has_zero_init(s
->extents
[i
].file
)) {
2049 static ImageInfo
*vmdk_get_extent_info(VmdkExtent
*extent
)
2051 ImageInfo
*info
= g_new0(ImageInfo
, 1);
2053 *info
= (ImageInfo
){
2054 .filename
= g_strdup(extent
->file
->filename
),
2055 .format
= g_strdup(extent
->type
),
2056 .virtual_size
= extent
->sectors
* BDRV_SECTOR_SIZE
,
2057 .compressed
= extent
->compressed
,
2058 .has_compressed
= extent
->compressed
,
2059 .cluster_size
= extent
->cluster_sectors
* BDRV_SECTOR_SIZE
,
2060 .has_cluster_size
= !extent
->flat
,
2066 static int vmdk_check(BlockDriverState
*bs
, BdrvCheckResult
*result
,
2069 BDRVVmdkState
*s
= bs
->opaque
;
2070 VmdkExtent
*extent
= NULL
;
2071 int64_t sector_num
= 0;
2072 int64_t total_sectors
= bdrv_nb_sectors(bs
);
2074 uint64_t cluster_offset
;
2081 if (sector_num
>= total_sectors
) {
2084 extent
= find_extent(s
, sector_num
, extent
);
2087 "ERROR: could not find extent for sector %" PRId64
"\n",
2091 ret
= get_cluster_offset(bs
, extent
, NULL
,
2092 sector_num
<< BDRV_SECTOR_BITS
,
2093 false, &cluster_offset
, 0, 0);
2094 if (ret
== VMDK_ERROR
) {
2096 "ERROR: could not get cluster_offset for sector %"
2097 PRId64
"\n", sector_num
);
2100 if (ret
== VMDK_OK
&& cluster_offset
>= bdrv_getlength(extent
->file
)) {
2102 "ERROR: cluster offset for sector %"
2103 PRId64
" points after EOF\n", sector_num
);
2106 sector_num
+= extent
->cluster_sectors
;
2109 result
->corruptions
++;
2113 static ImageInfoSpecific
*vmdk_get_specific_info(BlockDriverState
*bs
)
2116 BDRVVmdkState
*s
= bs
->opaque
;
2117 ImageInfoSpecific
*spec_info
= g_new0(ImageInfoSpecific
, 1);
2118 ImageInfoList
**next
;
2120 *spec_info
= (ImageInfoSpecific
){
2121 .kind
= IMAGE_INFO_SPECIFIC_KIND_VMDK
,
2123 .vmdk
= g_new0(ImageInfoSpecificVmdk
, 1),
2127 *spec_info
->vmdk
= (ImageInfoSpecificVmdk
) {
2128 .create_type
= g_strdup(s
->create_type
),
2130 .parent_cid
= s
->parent_cid
,
2133 next
= &spec_info
->vmdk
->extents
;
2134 for (i
= 0; i
< s
->num_extents
; i
++) {
2135 *next
= g_new0(ImageInfoList
, 1);
2136 (*next
)->value
= vmdk_get_extent_info(&s
->extents
[i
]);
2137 (*next
)->next
= NULL
;
2138 next
= &(*next
)->next
;
2144 static bool vmdk_extents_type_eq(const VmdkExtent
*a
, const VmdkExtent
*b
)
2146 return a
->flat
== b
->flat
&&
2147 a
->compressed
== b
->compressed
&&
2148 (a
->flat
|| a
->cluster_sectors
== b
->cluster_sectors
);
2151 static int vmdk_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2154 BDRVVmdkState
*s
= bs
->opaque
;
2155 assert(s
->num_extents
);
2157 /* See if we have multiple extents but they have different cases */
2158 for (i
= 1; i
< s
->num_extents
; i
++) {
2159 if (!vmdk_extents_type_eq(&s
->extents
[0], &s
->extents
[i
])) {
2163 bdi
->needs_compressed_writes
= s
->extents
[0].compressed
;
2164 if (!s
->extents
[0].flat
) {
2165 bdi
->cluster_size
= s
->extents
[0].cluster_sectors
<< BDRV_SECTOR_BITS
;
2170 static void vmdk_detach_aio_context(BlockDriverState
*bs
)
2172 BDRVVmdkState
*s
= bs
->opaque
;
2175 for (i
= 0; i
< s
->num_extents
; i
++) {
2176 bdrv_detach_aio_context(s
->extents
[i
].file
);
2180 static void vmdk_attach_aio_context(BlockDriverState
*bs
,
2181 AioContext
*new_context
)
2183 BDRVVmdkState
*s
= bs
->opaque
;
2186 for (i
= 0; i
< s
->num_extents
; i
++) {
2187 bdrv_attach_aio_context(s
->extents
[i
].file
, new_context
);
2191 static QemuOptsList vmdk_create_opts
= {
2192 .name
= "vmdk-create-opts",
2193 .head
= QTAILQ_HEAD_INITIALIZER(vmdk_create_opts
.head
),
2196 .name
= BLOCK_OPT_SIZE
,
2197 .type
= QEMU_OPT_SIZE
,
2198 .help
= "Virtual disk size"
2201 .name
= BLOCK_OPT_ADAPTER_TYPE
,
2202 .type
= QEMU_OPT_STRING
,
2203 .help
= "Virtual adapter type, can be one of "
2204 "ide (default), lsilogic, buslogic or legacyESX"
2207 .name
= BLOCK_OPT_BACKING_FILE
,
2208 .type
= QEMU_OPT_STRING
,
2209 .help
= "File name of a base image"
2212 .name
= BLOCK_OPT_COMPAT6
,
2213 .type
= QEMU_OPT_BOOL
,
2214 .help
= "VMDK version 6 image",
2215 .def_value_str
= "off"
2218 .name
= BLOCK_OPT_SUBFMT
,
2219 .type
= QEMU_OPT_STRING
,
2221 "VMDK flat extent format, can be one of "
2222 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2225 .name
= BLOCK_OPT_ZEROED_GRAIN
,
2226 .type
= QEMU_OPT_BOOL
,
2227 .help
= "Enable efficient zero writes "
2228 "using the zeroed-grain GTE feature"
2230 { /* end of list */ }
2234 static BlockDriver bdrv_vmdk
= {
2235 .format_name
= "vmdk",
2236 .instance_size
= sizeof(BDRVVmdkState
),
2237 .bdrv_probe
= vmdk_probe
,
2238 .bdrv_open
= vmdk_open
,
2239 .bdrv_check
= vmdk_check
,
2240 .bdrv_reopen_prepare
= vmdk_reopen_prepare
,
2241 .bdrv_read
= vmdk_co_read
,
2242 .bdrv_write
= vmdk_co_write
,
2243 .bdrv_write_compressed
= vmdk_write_compressed
,
2244 .bdrv_co_write_zeroes
= vmdk_co_write_zeroes
,
2245 .bdrv_close
= vmdk_close
,
2246 .bdrv_create
= vmdk_create
,
2247 .bdrv_co_flush_to_disk
= vmdk_co_flush
,
2248 .bdrv_co_get_block_status
= vmdk_co_get_block_status
,
2249 .bdrv_get_allocated_file_size
= vmdk_get_allocated_file_size
,
2250 .bdrv_has_zero_init
= vmdk_has_zero_init
,
2251 .bdrv_get_specific_info
= vmdk_get_specific_info
,
2252 .bdrv_refresh_limits
= vmdk_refresh_limits
,
2253 .bdrv_get_info
= vmdk_get_info
,
2254 .bdrv_detach_aio_context
= vmdk_detach_aio_context
,
2255 .bdrv_attach_aio_context
= vmdk_attach_aio_context
,
2257 .supports_backing
= true,
2258 .create_opts
= &vmdk_create_opts
,
2261 static void bdrv_vmdk_init(void)
2263 bdrv_register(&bdrv_vmdk
);
2266 block_init(bdrv_vmdk_init
);