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 "qapi/qmp/qerror.h"
29 #include "qemu/error-report.h"
30 #include "qemu/module.h"
31 #include "migration/migration.h"
35 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
36 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
37 #define VMDK4_COMPRESSION_DEFLATE 1
38 #define VMDK4_FLAG_NL_DETECT (1 << 0)
39 #define VMDK4_FLAG_RGD (1 << 1)
40 /* Zeroed-grain enable bit */
41 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
42 #define VMDK4_FLAG_COMPRESS (1 << 16)
43 #define VMDK4_FLAG_MARKER (1 << 17)
44 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
46 #define VMDK_GTE_ZEROED 0x1
48 /* VMDK internal error codes */
50 #define VMDK_ERROR (-1)
51 /* Cluster not allocated */
52 #define VMDK_UNALLOC (-2)
53 #define VMDK_ZEROED (-3)
55 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
60 uint32_t disk_sectors
;
62 uint32_t l1dir_offset
;
64 uint32_t file_sectors
;
67 uint32_t sectors_per_track
;
68 } QEMU_PACKED VMDK3Header
;
77 /* Number of GrainTableEntries per GrainTable */
78 uint32_t num_gtes_per_gt
;
81 uint64_t grain_offset
;
84 uint16_t compressAlgorithm
;
85 } QEMU_PACKED VMDK4Header
;
87 #define L2_CACHE_SIZE 16
89 typedef struct VmdkExtent
{
98 int64_t flat_start_offset
;
99 int64_t l1_table_offset
;
100 int64_t l1_backup_table_offset
;
102 uint32_t *l1_backup_table
;
103 unsigned int l1_size
;
104 uint32_t l1_entry_sectors
;
106 unsigned int l2_size
;
108 uint32_t l2_cache_offsets
[L2_CACHE_SIZE
];
109 uint32_t l2_cache_counts
[L2_CACHE_SIZE
];
111 int64_t cluster_sectors
;
112 int64_t next_cluster_sector
;
116 typedef struct BDRVVmdkState
{
118 uint64_t desc_offset
;
124 /* Extent array with num_extents entries, ascend ordered by address */
126 Error
*migration_blocker
;
130 typedef struct VmdkMetaData
{
131 unsigned int l1_index
;
132 unsigned int l2_index
;
133 unsigned int l2_offset
;
135 uint32_t *l2_cache_entry
;
138 typedef struct VmdkGrainMarker
{
142 } QEMU_PACKED VmdkGrainMarker
;
145 MARKER_END_OF_STREAM
= 0,
146 MARKER_GRAIN_TABLE
= 1,
147 MARKER_GRAIN_DIRECTORY
= 2,
151 static int vmdk_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
158 magic
= be32_to_cpu(*(uint32_t *)buf
);
159 if (magic
== VMDK3_MAGIC
||
160 magic
== VMDK4_MAGIC
) {
163 const char *p
= (const char *)buf
;
164 const char *end
= p
+ buf_size
;
167 /* skip comment line */
168 while (p
< end
&& *p
!= '\n') {
175 while (p
< end
&& *p
== ' ') {
178 /* skip '\r' if windows line endings used. */
179 if (p
< end
&& *p
== '\r') {
182 /* only accept blank lines before 'version=' line */
183 if (p
== end
|| *p
!= '\n') {
189 if (end
- p
>= strlen("version=X\n")) {
190 if (strncmp("version=1\n", p
, strlen("version=1\n")) == 0 ||
191 strncmp("version=2\n", p
, strlen("version=2\n")) == 0) {
195 if (end
- p
>= strlen("version=X\r\n")) {
196 if (strncmp("version=1\r\n", p
, strlen("version=1\r\n")) == 0 ||
197 strncmp("version=2\r\n", p
, strlen("version=2\r\n")) == 0) {
207 #define SECTOR_SIZE 512
208 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
209 #define BUF_SIZE 4096
210 #define HEADER_SIZE 512 /* first sector of 512 bytes */
212 static void vmdk_free_extents(BlockDriverState
*bs
)
215 BDRVVmdkState
*s
= bs
->opaque
;
218 for (i
= 0; i
< s
->num_extents
; i
++) {
222 g_free(e
->l1_backup_table
);
224 if (e
->file
!= bs
->file
) {
225 bdrv_unref_child(bs
, e
->file
);
231 static void vmdk_free_last_extent(BlockDriverState
*bs
)
233 BDRVVmdkState
*s
= bs
->opaque
;
235 if (s
->num_extents
== 0) {
239 s
->extents
= g_renew(VmdkExtent
, s
->extents
, s
->num_extents
);
242 static uint32_t vmdk_read_cid(BlockDriverState
*bs
, int parent
)
244 char desc
[DESC_SIZE
];
245 uint32_t cid
= 0xffffffff;
246 const char *p_name
, *cid_str
;
248 BDRVVmdkState
*s
= bs
->opaque
;
251 ret
= bdrv_pread(bs
->file
->bs
, s
->desc_offset
, desc
, DESC_SIZE
);
257 cid_str
= "parentCID";
258 cid_str_size
= sizeof("parentCID");
261 cid_str_size
= sizeof("CID");
264 desc
[DESC_SIZE
- 1] = '\0';
265 p_name
= strstr(desc
, cid_str
);
266 if (p_name
!= NULL
) {
267 p_name
+= cid_str_size
;
268 sscanf(p_name
, "%" SCNx32
, &cid
);
274 static int vmdk_write_cid(BlockDriverState
*bs
, uint32_t cid
)
276 char desc
[DESC_SIZE
], tmp_desc
[DESC_SIZE
];
277 char *p_name
, *tmp_str
;
278 BDRVVmdkState
*s
= bs
->opaque
;
281 ret
= bdrv_pread(bs
->file
->bs
, s
->desc_offset
, desc
, DESC_SIZE
);
286 desc
[DESC_SIZE
- 1] = '\0';
287 tmp_str
= strstr(desc
, "parentCID");
288 if (tmp_str
== NULL
) {
292 pstrcpy(tmp_desc
, sizeof(tmp_desc
), tmp_str
);
293 p_name
= strstr(desc
, "CID");
294 if (p_name
!= NULL
) {
295 p_name
+= sizeof("CID");
296 snprintf(p_name
, sizeof(desc
) - (p_name
- desc
), "%" PRIx32
"\n", cid
);
297 pstrcat(desc
, sizeof(desc
), tmp_desc
);
300 ret
= bdrv_pwrite_sync(bs
->file
->bs
, s
->desc_offset
, desc
, DESC_SIZE
);
308 static int vmdk_is_cid_valid(BlockDriverState
*bs
)
310 BDRVVmdkState
*s
= bs
->opaque
;
311 BlockDriverState
*p_bs
= bs
->backing_hd
;
314 if (!s
->cid_checked
&& p_bs
) {
315 cur_pcid
= vmdk_read_cid(p_bs
, 0);
316 if (s
->parent_cid
!= cur_pcid
) {
321 s
->cid_checked
= true;
326 /* We have nothing to do for VMDK reopen, stubs just return success */
327 static int vmdk_reopen_prepare(BDRVReopenState
*state
,
328 BlockReopenQueue
*queue
, Error
**errp
)
330 assert(state
!= NULL
);
331 assert(state
->bs
!= NULL
);
335 static int vmdk_parent_open(BlockDriverState
*bs
)
338 char desc
[DESC_SIZE
+ 1];
339 BDRVVmdkState
*s
= bs
->opaque
;
342 desc
[DESC_SIZE
] = '\0';
343 ret
= bdrv_pread(bs
->file
->bs
, s
->desc_offset
, desc
, DESC_SIZE
);
348 p_name
= strstr(desc
, "parentFileNameHint");
349 if (p_name
!= NULL
) {
352 p_name
+= sizeof("parentFileNameHint") + 1;
353 end_name
= strchr(p_name
, '\"');
354 if (end_name
== NULL
) {
357 if ((end_name
- p_name
) > sizeof(bs
->backing_file
) - 1) {
361 pstrcpy(bs
->backing_file
, end_name
- p_name
+ 1, p_name
);
367 /* Create and append extent to the extent array. Return the added VmdkExtent
368 * address. return NULL if allocation failed. */
369 static int vmdk_add_extent(BlockDriverState
*bs
,
370 BdrvChild
*file
, bool flat
, int64_t sectors
,
371 int64_t l1_offset
, int64_t l1_backup_offset
,
373 int l2_size
, uint64_t cluster_sectors
,
374 VmdkExtent
**new_extent
,
378 BDRVVmdkState
*s
= bs
->opaque
;
381 if (cluster_sectors
> 0x200000) {
382 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
383 error_setg(errp
, "Invalid granularity, image may be corrupt");
386 if (l1_size
> 512 * 1024 * 1024) {
387 /* Although with big capacity and small l1_entry_sectors, we can get a
388 * big l1_size, we don't want unbounded value to allocate the table.
389 * Limit it to 512M, which is 16PB for default cluster and L2 table
391 error_setg(errp
, "L1 size too big");
395 nb_sectors
= bdrv_nb_sectors(file
->bs
);
396 if (nb_sectors
< 0) {
400 s
->extents
= g_renew(VmdkExtent
, s
->extents
, s
->num_extents
+ 1);
401 extent
= &s
->extents
[s
->num_extents
];
404 memset(extent
, 0, sizeof(VmdkExtent
));
407 extent
->sectors
= sectors
;
408 extent
->l1_table_offset
= l1_offset
;
409 extent
->l1_backup_table_offset
= l1_backup_offset
;
410 extent
->l1_size
= l1_size
;
411 extent
->l1_entry_sectors
= l2_size
* cluster_sectors
;
412 extent
->l2_size
= l2_size
;
413 extent
->cluster_sectors
= flat
? sectors
: cluster_sectors
;
414 extent
->next_cluster_sector
= ROUND_UP(nb_sectors
, cluster_sectors
);
416 if (s
->num_extents
> 1) {
417 extent
->end_sector
= (*(extent
- 1)).end_sector
+ extent
->sectors
;
419 extent
->end_sector
= extent
->sectors
;
421 bs
->total_sectors
= extent
->end_sector
;
423 *new_extent
= extent
;
428 static int vmdk_init_tables(BlockDriverState
*bs
, VmdkExtent
*extent
,
435 /* read the L1 table */
436 l1_size
= extent
->l1_size
* sizeof(uint32_t);
437 extent
->l1_table
= g_try_malloc(l1_size
);
438 if (l1_size
&& extent
->l1_table
== NULL
) {
442 ret
= bdrv_pread(extent
->file
->bs
,
443 extent
->l1_table_offset
,
447 error_setg_errno(errp
, -ret
,
448 "Could not read l1 table from extent '%s'",
449 extent
->file
->bs
->filename
);
452 for (i
= 0; i
< extent
->l1_size
; i
++) {
453 le32_to_cpus(&extent
->l1_table
[i
]);
456 if (extent
->l1_backup_table_offset
) {
457 extent
->l1_backup_table
= g_try_malloc(l1_size
);
458 if (l1_size
&& extent
->l1_backup_table
== NULL
) {
462 ret
= bdrv_pread(extent
->file
->bs
,
463 extent
->l1_backup_table_offset
,
464 extent
->l1_backup_table
,
467 error_setg_errno(errp
, -ret
,
468 "Could not read l1 backup table from extent '%s'",
469 extent
->file
->bs
->filename
);
472 for (i
= 0; i
< extent
->l1_size
; i
++) {
473 le32_to_cpus(&extent
->l1_backup_table
[i
]);
478 g_new(uint32_t, extent
->l2_size
* L2_CACHE_SIZE
);
481 g_free(extent
->l1_backup_table
);
483 g_free(extent
->l1_table
);
487 static int vmdk_open_vmfs_sparse(BlockDriverState
*bs
,
489 int flags
, Error
**errp
)
496 ret
= bdrv_pread(file
->bs
, sizeof(magic
), &header
, sizeof(header
));
498 error_setg_errno(errp
, -ret
,
499 "Could not read header from file '%s'",
503 ret
= vmdk_add_extent(bs
, file
, false,
504 le32_to_cpu(header
.disk_sectors
),
505 (int64_t)le32_to_cpu(header
.l1dir_offset
) << 9,
507 le32_to_cpu(header
.l1dir_size
),
509 le32_to_cpu(header
.granularity
),
515 ret
= vmdk_init_tables(bs
, extent
, errp
);
517 /* free extent allocated by vmdk_add_extent */
518 vmdk_free_last_extent(bs
);
523 static int vmdk_open_desc_file(BlockDriverState
*bs
, int flags
, char *buf
,
524 QDict
*options
, Error
**errp
);
526 static char *vmdk_read_desc(BlockDriverState
*file
, uint64_t desc_offset
,
533 size
= bdrv_getlength(file
);
535 error_setg_errno(errp
, -size
, "Could not access file");
540 /* Both descriptor file and sparse image must be much larger than 4
541 * bytes, also callers of vmdk_read_desc want to compare the first 4
542 * bytes with VMDK4_MAGIC, let's error out if less is read. */
543 error_setg(errp
, "File is too small, not a valid image");
547 size
= MIN(size
, (1 << 20) - 1); /* avoid unbounded allocation */
548 buf
= g_malloc(size
+ 1);
550 ret
= bdrv_pread(file
, desc_offset
, buf
, size
);
552 error_setg_errno(errp
, -ret
, "Could not read from file");
561 static int vmdk_open_vmdk4(BlockDriverState
*bs
,
563 int flags
, QDict
*options
, Error
**errp
)
567 uint32_t l1_size
, l1_entry_sectors
;
570 BDRVVmdkState
*s
= bs
->opaque
;
571 int64_t l1_backup_offset
= 0;
573 ret
= bdrv_pread(file
->bs
, sizeof(magic
), &header
, sizeof(header
));
575 error_setg_errno(errp
, -ret
,
576 "Could not read header from file '%s'",
580 if (header
.capacity
== 0) {
581 uint64_t desc_offset
= le64_to_cpu(header
.desc_offset
);
583 char *buf
= vmdk_read_desc(file
->bs
, desc_offset
<< 9, errp
);
587 ret
= vmdk_open_desc_file(bs
, flags
, buf
, options
, errp
);
593 if (!s
->create_type
) {
594 s
->create_type
= g_strdup("monolithicSparse");
597 if (le64_to_cpu(header
.gd_offset
) == VMDK4_GD_AT_END
) {
599 * The footer takes precedence over the header, so read it in. The
600 * footer starts at offset -1024 from the end: One sector for the
601 * footer, and another one for the end-of-stream marker.
608 uint8_t pad
[512 - 16];
609 } QEMU_PACKED footer_marker
;
613 uint8_t pad
[512 - 4 - sizeof(VMDK4Header
)];
619 uint8_t pad
[512 - 16];
620 } QEMU_PACKED eos_marker
;
621 } QEMU_PACKED footer
;
623 ret
= bdrv_pread(file
->bs
,
624 bs
->file
->bs
->total_sectors
* 512 - 1536,
625 &footer
, sizeof(footer
));
627 error_setg_errno(errp
, -ret
, "Failed to read footer");
631 /* Some sanity checks for the footer */
632 if (be32_to_cpu(footer
.magic
) != VMDK4_MAGIC
||
633 le32_to_cpu(footer
.footer_marker
.size
) != 0 ||
634 le32_to_cpu(footer
.footer_marker
.type
) != MARKER_FOOTER
||
635 le64_to_cpu(footer
.eos_marker
.val
) != 0 ||
636 le32_to_cpu(footer
.eos_marker
.size
) != 0 ||
637 le32_to_cpu(footer
.eos_marker
.type
) != MARKER_END_OF_STREAM
)
639 error_setg(errp
, "Invalid footer");
643 header
= footer
.header
;
646 if (le32_to_cpu(header
.version
) > 3) {
648 snprintf(buf
, sizeof(buf
), "VMDK version %" PRId32
,
649 le32_to_cpu(header
.version
));
650 error_setg(errp
, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE
,
651 bdrv_get_device_or_node_name(bs
), "vmdk", buf
);
653 } else if (le32_to_cpu(header
.version
) == 3 && (flags
& BDRV_O_RDWR
)) {
654 /* VMware KB 2064959 explains that version 3 added support for
655 * persistent changed block tracking (CBT), and backup software can
656 * read it as version=1 if it doesn't care about the changed area
657 * information. So we are safe to enable read only. */
658 error_setg(errp
, "VMDK version 3 must be read only");
662 if (le32_to_cpu(header
.num_gtes_per_gt
) > 512) {
663 error_setg(errp
, "L2 table size too big");
667 l1_entry_sectors
= le32_to_cpu(header
.num_gtes_per_gt
)
668 * le64_to_cpu(header
.granularity
);
669 if (l1_entry_sectors
== 0) {
670 error_setg(errp
, "L1 entry size is invalid");
673 l1_size
= (le64_to_cpu(header
.capacity
) + l1_entry_sectors
- 1)
675 if (le32_to_cpu(header
.flags
) & VMDK4_FLAG_RGD
) {
676 l1_backup_offset
= le64_to_cpu(header
.rgd_offset
) << 9;
678 if (bdrv_nb_sectors(file
->bs
) < le64_to_cpu(header
.grain_offset
)) {
679 error_setg(errp
, "File truncated, expecting at least %" PRId64
" bytes",
680 (int64_t)(le64_to_cpu(header
.grain_offset
)
681 * BDRV_SECTOR_SIZE
));
685 ret
= vmdk_add_extent(bs
, file
, false,
686 le64_to_cpu(header
.capacity
),
687 le64_to_cpu(header
.gd_offset
) << 9,
690 le32_to_cpu(header
.num_gtes_per_gt
),
691 le64_to_cpu(header
.granularity
),
698 le16_to_cpu(header
.compressAlgorithm
) == VMDK4_COMPRESSION_DEFLATE
;
699 if (extent
->compressed
) {
700 g_free(s
->create_type
);
701 s
->create_type
= g_strdup("streamOptimized");
703 extent
->has_marker
= le32_to_cpu(header
.flags
) & VMDK4_FLAG_MARKER
;
704 extent
->version
= le32_to_cpu(header
.version
);
705 extent
->has_zero_grain
= le32_to_cpu(header
.flags
) & VMDK4_FLAG_ZERO_GRAIN
;
706 ret
= vmdk_init_tables(bs
, extent
, errp
);
708 /* free extent allocated by vmdk_add_extent */
709 vmdk_free_last_extent(bs
);
714 /* find an option value out of descriptor file */
715 static int vmdk_parse_description(const char *desc
, const char *opt_name
,
716 char *buf
, int buf_size
)
718 char *opt_pos
, *opt_end
;
719 const char *end
= desc
+ strlen(desc
);
721 opt_pos
= strstr(desc
, opt_name
);
725 /* Skip "=\"" following opt_name */
726 opt_pos
+= strlen(opt_name
) + 2;
727 if (opt_pos
>= end
) {
731 while (opt_end
< end
&& *opt_end
!= '"') {
734 if (opt_end
== end
|| buf_size
< opt_end
- opt_pos
+ 1) {
737 pstrcpy(buf
, opt_end
- opt_pos
+ 1, opt_pos
);
741 /* Open an extent file and append to bs array */
742 static int vmdk_open_sparse(BlockDriverState
*bs
, BdrvChild
*file
, int flags
,
743 char *buf
, QDict
*options
, Error
**errp
)
747 magic
= ldl_be_p(buf
);
750 return vmdk_open_vmfs_sparse(bs
, file
, flags
, errp
);
753 return vmdk_open_vmdk4(bs
, file
, flags
, options
, errp
);
756 error_setg(errp
, "Image not in VMDK format");
762 static int vmdk_parse_extents(const char *desc
, BlockDriverState
*bs
,
763 const char *desc_file_path
, QDict
*options
,
771 const char *p
= desc
;
775 BdrvChild
*extent_file
;
776 BDRVVmdkState
*s
= bs
->opaque
;
778 char extent_opt_prefix
[32];
779 Error
*local_err
= NULL
;
782 /* parse extent line in one of below formats:
784 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
785 * RW [size in sectors] SPARSE "file-name.vmdk"
786 * RW [size in sectors] VMFS "file-name.vmdk"
787 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
790 matches
= sscanf(p
, "%10s %" SCNd64
" %10s \"%511[^\n\r\"]\" %" SCNd64
,
791 access
, §ors
, type
, fname
, &flat_offset
);
792 if (matches
< 4 || strcmp(access
, "RW")) {
794 } else if (!strcmp(type
, "FLAT")) {
795 if (matches
!= 5 || flat_offset
< 0) {
796 error_setg(errp
, "Invalid extent lines: \n%s", p
);
799 } else if (!strcmp(type
, "VMFS")) {
803 error_setg(errp
, "Invalid extent lines:\n%s", p
);
806 } else if (matches
!= 4) {
807 error_setg(errp
, "Invalid extent lines:\n%s", p
);
812 (strcmp(type
, "FLAT") && strcmp(type
, "SPARSE") &&
813 strcmp(type
, "VMFS") && strcmp(type
, "VMFSSPARSE")) ||
814 (strcmp(access
, "RW"))) {
818 if (!path_is_absolute(fname
) && !path_has_protocol(fname
) &&
821 error_setg(errp
, "Cannot use relative extent paths with VMDK "
822 "descriptor file '%s'", bs
->file
->bs
->filename
);
826 extent_path
= g_malloc0(PATH_MAX
);
827 path_combine(extent_path
, PATH_MAX
, desc_file_path
, fname
);
829 ret
= snprintf(extent_opt_prefix
, 32, "extents.%d", s
->num_extents
);
832 extent_file
= bdrv_open_child(extent_path
, options
, extent_opt_prefix
,
833 bs
, &child_file
, false, &local_err
);
836 error_propagate(errp
, local_err
);
840 /* save to extents array */
841 if (!strcmp(type
, "FLAT") || !strcmp(type
, "VMFS")) {
844 ret
= vmdk_add_extent(bs
, extent_file
, true, sectors
,
845 0, 0, 0, 0, 0, &extent
, errp
);
847 bdrv_unref_child(bs
, extent_file
);
850 extent
->flat_start_offset
= flat_offset
<< 9;
851 } else if (!strcmp(type
, "SPARSE") || !strcmp(type
, "VMFSSPARSE")) {
852 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
853 char *buf
= vmdk_read_desc(extent_file
->bs
, 0, errp
);
857 ret
= vmdk_open_sparse(bs
, extent_file
, bs
->open_flags
, buf
,
862 bdrv_unref_child(bs
, extent_file
);
865 extent
= &s
->extents
[s
->num_extents
- 1];
867 error_setg(errp
, "Unsupported extent type '%s'", type
);
868 bdrv_unref_child(bs
, extent_file
);
871 extent
->type
= g_strdup(type
);
873 /* move to next line */
885 static int vmdk_open_desc_file(BlockDriverState
*bs
, int flags
, char *buf
,
886 QDict
*options
, Error
**errp
)
890 BDRVVmdkState
*s
= bs
->opaque
;
892 if (vmdk_parse_description(buf
, "createType", ct
, sizeof(ct
))) {
893 error_setg(errp
, "invalid VMDK image descriptor");
897 if (strcmp(ct
, "monolithicFlat") &&
898 strcmp(ct
, "vmfs") &&
899 strcmp(ct
, "vmfsSparse") &&
900 strcmp(ct
, "twoGbMaxExtentSparse") &&
901 strcmp(ct
, "twoGbMaxExtentFlat")) {
902 error_setg(errp
, "Unsupported image type '%s'", ct
);
906 s
->create_type
= g_strdup(ct
);
908 ret
= vmdk_parse_extents(buf
, bs
, bs
->file
->bs
->exact_filename
, options
,
914 static int vmdk_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
919 BDRVVmdkState
*s
= bs
->opaque
;
922 buf
= vmdk_read_desc(bs
->file
->bs
, 0, errp
);
927 magic
= ldl_be_p(buf
);
931 ret
= vmdk_open_sparse(bs
, bs
->file
, flags
, buf
, options
,
933 s
->desc_offset
= 0x200;
936 ret
= vmdk_open_desc_file(bs
, flags
, buf
, options
, errp
);
943 /* try to open parent images, if exist */
944 ret
= vmdk_parent_open(bs
);
948 s
->cid
= vmdk_read_cid(bs
, 0);
949 s
->parent_cid
= vmdk_read_cid(bs
, 1);
950 qemu_co_mutex_init(&s
->lock
);
952 /* Disable migration when VMDK images are used */
953 error_setg(&s
->migration_blocker
, "The vmdk format used by node '%s' "
954 "does not support live migration",
955 bdrv_get_device_or_node_name(bs
));
956 migrate_add_blocker(s
->migration_blocker
);
962 g_free(s
->create_type
);
963 s
->create_type
= NULL
;
964 vmdk_free_extents(bs
);
969 static void vmdk_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
971 BDRVVmdkState
*s
= bs
->opaque
;
974 for (i
= 0; i
< s
->num_extents
; i
++) {
975 if (!s
->extents
[i
].flat
) {
976 bs
->bl
.write_zeroes_alignment
=
977 MAX(bs
->bl
.write_zeroes_alignment
,
978 s
->extents
[i
].cluster_sectors
);
986 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
987 * to the cluster at @cluster_sector_num.
989 * If @skip_start_sector < @skip_end_sector, the relative range
990 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
991 * it for call to write user data in the request.
993 static int get_whole_cluster(BlockDriverState
*bs
,
995 uint64_t cluster_sector_num
,
997 uint64_t skip_start_sector
,
998 uint64_t skip_end_sector
)
1001 int64_t cluster_bytes
;
1002 uint8_t *whole_grain
;
1004 /* For COW, align request sector_num to cluster start */
1005 sector_num
= QEMU_ALIGN_DOWN(sector_num
, extent
->cluster_sectors
);
1006 cluster_bytes
= extent
->cluster_sectors
<< BDRV_SECTOR_BITS
;
1007 whole_grain
= qemu_blockalign(bs
, cluster_bytes
);
1009 if (!bs
->backing_hd
) {
1010 memset(whole_grain
, 0, skip_start_sector
<< BDRV_SECTOR_BITS
);
1011 memset(whole_grain
+ (skip_end_sector
<< BDRV_SECTOR_BITS
), 0,
1012 cluster_bytes
- (skip_end_sector
<< BDRV_SECTOR_BITS
));
1015 assert(skip_end_sector
<= extent
->cluster_sectors
);
1016 /* we will be here if it's first write on non-exist grain(cluster).
1017 * try to read from parent image, if exist */
1018 if (bs
->backing_hd
&& !vmdk_is_cid_valid(bs
)) {
1023 /* Read backing data before skip range */
1024 if (skip_start_sector
> 0) {
1025 if (bs
->backing_hd
) {
1026 ret
= bdrv_read(bs
->backing_hd
, sector_num
,
1027 whole_grain
, skip_start_sector
);
1033 ret
= bdrv_write(extent
->file
->bs
, cluster_sector_num
, whole_grain
,
1040 /* Read backing data after skip range */
1041 if (skip_end_sector
< extent
->cluster_sectors
) {
1042 if (bs
->backing_hd
) {
1043 ret
= bdrv_read(bs
->backing_hd
, sector_num
+ skip_end_sector
,
1044 whole_grain
+ (skip_end_sector
<< BDRV_SECTOR_BITS
),
1045 extent
->cluster_sectors
- skip_end_sector
);
1051 ret
= bdrv_write(extent
->file
->bs
, cluster_sector_num
+ skip_end_sector
,
1052 whole_grain
+ (skip_end_sector
<< BDRV_SECTOR_BITS
),
1053 extent
->cluster_sectors
- skip_end_sector
);
1061 qemu_vfree(whole_grain
);
1065 static int vmdk_L2update(VmdkExtent
*extent
, VmdkMetaData
*m_data
,
1068 offset
= cpu_to_le32(offset
);
1069 /* update L2 table */
1070 if (bdrv_pwrite_sync(
1072 ((int64_t)m_data
->l2_offset
* 512)
1073 + (m_data
->l2_index
* sizeof(offset
)),
1074 &offset
, sizeof(offset
)) < 0) {
1077 /* update backup L2 table */
1078 if (extent
->l1_backup_table_offset
!= 0) {
1079 m_data
->l2_offset
= extent
->l1_backup_table
[m_data
->l1_index
];
1080 if (bdrv_pwrite_sync(
1082 ((int64_t)m_data
->l2_offset
* 512)
1083 + (m_data
->l2_index
* sizeof(offset
)),
1084 &offset
, sizeof(offset
)) < 0) {
1088 if (m_data
->l2_cache_entry
) {
1089 *m_data
->l2_cache_entry
= offset
;
1096 * get_cluster_offset
1098 * Look up cluster offset in extent file by sector number, and store in
1101 * For flat extents, the start offset as parsed from the description file is
1104 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1105 * offset for a new cluster and update L2 cache. If there is a backing file,
1106 * COW is done before returning; otherwise, zeroes are written to the allocated
1107 * cluster. Both COW and zero writing skips the sector range
1108 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1109 * has new data to write there.
1111 * Returns: VMDK_OK if cluster exists and mapped in the image.
1112 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1113 * VMDK_ERROR if failed.
1115 static int get_cluster_offset(BlockDriverState
*bs
,
1117 VmdkMetaData
*m_data
,
1120 uint64_t *cluster_offset
,
1121 uint64_t skip_start_sector
,
1122 uint64_t skip_end_sector
)
1124 unsigned int l1_index
, l2_offset
, l2_index
;
1125 int min_index
, i
, j
;
1126 uint32_t min_count
, *l2_table
;
1127 bool zeroed
= false;
1129 int64_t cluster_sector
;
1135 *cluster_offset
= extent
->flat_start_offset
;
1139 offset
-= (extent
->end_sector
- extent
->sectors
) * SECTOR_SIZE
;
1140 l1_index
= (offset
>> 9) / extent
->l1_entry_sectors
;
1141 if (l1_index
>= extent
->l1_size
) {
1144 l2_offset
= extent
->l1_table
[l1_index
];
1146 return VMDK_UNALLOC
;
1148 for (i
= 0; i
< L2_CACHE_SIZE
; i
++) {
1149 if (l2_offset
== extent
->l2_cache_offsets
[i
]) {
1150 /* increment the hit count */
1151 if (++extent
->l2_cache_counts
[i
] == 0xffffffff) {
1152 for (j
= 0; j
< L2_CACHE_SIZE
; j
++) {
1153 extent
->l2_cache_counts
[j
] >>= 1;
1156 l2_table
= extent
->l2_cache
+ (i
* extent
->l2_size
);
1160 /* not found: load a new entry in the least used one */
1162 min_count
= 0xffffffff;
1163 for (i
= 0; i
< L2_CACHE_SIZE
; i
++) {
1164 if (extent
->l2_cache_counts
[i
] < min_count
) {
1165 min_count
= extent
->l2_cache_counts
[i
];
1169 l2_table
= extent
->l2_cache
+ (min_index
* extent
->l2_size
);
1172 (int64_t)l2_offset
* 512,
1174 extent
->l2_size
* sizeof(uint32_t)
1175 ) != extent
->l2_size
* sizeof(uint32_t)) {
1179 extent
->l2_cache_offsets
[min_index
] = l2_offset
;
1180 extent
->l2_cache_counts
[min_index
] = 1;
1182 l2_index
= ((offset
>> 9) / extent
->cluster_sectors
) % extent
->l2_size
;
1183 cluster_sector
= le32_to_cpu(l2_table
[l2_index
]);
1187 m_data
->l1_index
= l1_index
;
1188 m_data
->l2_index
= l2_index
;
1189 m_data
->l2_offset
= l2_offset
;
1190 m_data
->l2_cache_entry
= &l2_table
[l2_index
];
1192 if (extent
->has_zero_grain
&& cluster_sector
== VMDK_GTE_ZEROED
) {
1196 if (!cluster_sector
|| zeroed
) {
1198 return zeroed
? VMDK_ZEROED
: VMDK_UNALLOC
;
1201 cluster_sector
= extent
->next_cluster_sector
;
1202 extent
->next_cluster_sector
+= extent
->cluster_sectors
;
1204 /* First of all we write grain itself, to avoid race condition
1205 * that may to corrupt the image.
1206 * This problem may occur because of insufficient space on host disk
1207 * or inappropriate VM shutdown.
1209 ret
= get_whole_cluster(bs
, extent
,
1211 offset
>> BDRV_SECTOR_BITS
,
1212 skip_start_sector
, skip_end_sector
);
1217 *cluster_offset
= cluster_sector
<< BDRV_SECTOR_BITS
;
1221 static VmdkExtent
*find_extent(BDRVVmdkState
*s
,
1222 int64_t sector_num
, VmdkExtent
*start_hint
)
1224 VmdkExtent
*extent
= start_hint
;
1227 extent
= &s
->extents
[0];
1229 while (extent
< &s
->extents
[s
->num_extents
]) {
1230 if (sector_num
< extent
->end_sector
) {
1238 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent
*extent
,
1241 uint64_t index_in_cluster
, extent_begin_sector
, extent_relative_sector_num
;
1243 extent_begin_sector
= extent
->end_sector
- extent
->sectors
;
1244 extent_relative_sector_num
= sector_num
- extent_begin_sector
;
1245 index_in_cluster
= extent_relative_sector_num
% extent
->cluster_sectors
;
1246 return index_in_cluster
;
1249 static int64_t coroutine_fn
vmdk_co_get_block_status(BlockDriverState
*bs
,
1250 int64_t sector_num
, int nb_sectors
, int *pnum
)
1252 BDRVVmdkState
*s
= bs
->opaque
;
1253 int64_t index_in_cluster
, n
, ret
;
1257 extent
= find_extent(s
, sector_num
, NULL
);
1261 qemu_co_mutex_lock(&s
->lock
);
1262 ret
= get_cluster_offset(bs
, extent
, NULL
,
1263 sector_num
* 512, false, &offset
,
1265 qemu_co_mutex_unlock(&s
->lock
);
1275 ret
= BDRV_BLOCK_ZERO
;
1278 ret
= BDRV_BLOCK_DATA
;
1279 if (extent
->file
== bs
->file
&& !extent
->compressed
) {
1280 ret
|= BDRV_BLOCK_OFFSET_VALID
| offset
;
1286 index_in_cluster
= vmdk_find_index_in_cluster(extent
, sector_num
);
1287 n
= extent
->cluster_sectors
- index_in_cluster
;
1288 if (n
> nb_sectors
) {
1295 static int vmdk_write_extent(VmdkExtent
*extent
, int64_t cluster_offset
,
1296 int64_t offset_in_cluster
, const uint8_t *buf
,
1297 int nb_sectors
, int64_t sector_num
)
1300 VmdkGrainMarker
*data
= NULL
;
1302 const uint8_t *write_buf
= buf
;
1303 int write_len
= nb_sectors
* 512;
1304 int64_t write_offset
;
1305 int64_t write_end_sector
;
1307 if (extent
->compressed
) {
1308 if (!extent
->has_marker
) {
1312 buf_len
= (extent
->cluster_sectors
<< 9) * 2;
1313 data
= g_malloc(buf_len
+ sizeof(VmdkGrainMarker
));
1314 if (compress(data
->data
, &buf_len
, buf
, nb_sectors
<< 9) != Z_OK
||
1319 data
->lba
= sector_num
;
1320 data
->size
= buf_len
;
1321 write_buf
= (uint8_t *)data
;
1322 write_len
= buf_len
+ sizeof(VmdkGrainMarker
);
1324 write_offset
= cluster_offset
+ offset_in_cluster
,
1325 ret
= bdrv_pwrite(extent
->file
->bs
, write_offset
, write_buf
, write_len
);
1327 write_end_sector
= DIV_ROUND_UP(write_offset
+ write_len
, BDRV_SECTOR_SIZE
);
1329 if (extent
->compressed
) {
1330 extent
->next_cluster_sector
= write_end_sector
;
1332 extent
->next_cluster_sector
= MAX(extent
->next_cluster_sector
,
1336 if (ret
!= write_len
) {
1337 ret
= ret
< 0 ? ret
: -EIO
;
1346 static int vmdk_read_extent(VmdkExtent
*extent
, int64_t cluster_offset
,
1347 int64_t offset_in_cluster
, uint8_t *buf
,
1351 int cluster_bytes
, buf_bytes
;
1352 uint8_t *cluster_buf
, *compressed_data
;
1353 uint8_t *uncomp_buf
;
1355 VmdkGrainMarker
*marker
;
1359 if (!extent
->compressed
) {
1360 ret
= bdrv_pread(extent
->file
->bs
,
1361 cluster_offset
+ offset_in_cluster
,
1362 buf
, nb_sectors
* 512);
1363 if (ret
== nb_sectors
* 512) {
1369 cluster_bytes
= extent
->cluster_sectors
* 512;
1370 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1371 buf_bytes
= cluster_bytes
* 2;
1372 cluster_buf
= g_malloc(buf_bytes
);
1373 uncomp_buf
= g_malloc(cluster_bytes
);
1374 ret
= bdrv_pread(extent
->file
->bs
,
1376 cluster_buf
, buf_bytes
);
1380 compressed_data
= cluster_buf
;
1381 buf_len
= cluster_bytes
;
1382 data_len
= cluster_bytes
;
1383 if (extent
->has_marker
) {
1384 marker
= (VmdkGrainMarker
*)cluster_buf
;
1385 compressed_data
= marker
->data
;
1386 data_len
= le32_to_cpu(marker
->size
);
1388 if (!data_len
|| data_len
> buf_bytes
) {
1392 ret
= uncompress(uncomp_buf
, &buf_len
, compressed_data
, data_len
);
1398 if (offset_in_cluster
< 0 ||
1399 offset_in_cluster
+ nb_sectors
* 512 > buf_len
) {
1403 memcpy(buf
, uncomp_buf
+ offset_in_cluster
, nb_sectors
* 512);
1408 g_free(cluster_buf
);
1412 static int vmdk_read(BlockDriverState
*bs
, int64_t sector_num
,
1413 uint8_t *buf
, int nb_sectors
)
1415 BDRVVmdkState
*s
= bs
->opaque
;
1417 uint64_t n
, index_in_cluster
;
1418 VmdkExtent
*extent
= NULL
;
1419 uint64_t cluster_offset
;
1421 while (nb_sectors
> 0) {
1422 extent
= find_extent(s
, sector_num
, extent
);
1426 ret
= get_cluster_offset(bs
, extent
, NULL
,
1427 sector_num
<< 9, false, &cluster_offset
,
1429 index_in_cluster
= vmdk_find_index_in_cluster(extent
, sector_num
);
1430 n
= extent
->cluster_sectors
- index_in_cluster
;
1431 if (n
> nb_sectors
) {
1434 if (ret
!= VMDK_OK
) {
1435 /* if not allocated, try to read from parent image, if exist */
1436 if (bs
->backing_hd
&& ret
!= VMDK_ZEROED
) {
1437 if (!vmdk_is_cid_valid(bs
)) {
1440 ret
= bdrv_read(bs
->backing_hd
, sector_num
, buf
, n
);
1445 memset(buf
, 0, 512 * n
);
1448 ret
= vmdk_read_extent(extent
,
1449 cluster_offset
, index_in_cluster
* 512,
1462 static coroutine_fn
int vmdk_co_read(BlockDriverState
*bs
, int64_t sector_num
,
1463 uint8_t *buf
, int nb_sectors
)
1466 BDRVVmdkState
*s
= bs
->opaque
;
1467 qemu_co_mutex_lock(&s
->lock
);
1468 ret
= vmdk_read(bs
, sector_num
, buf
, nb_sectors
);
1469 qemu_co_mutex_unlock(&s
->lock
);
1475 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1476 * if possible, otherwise return -ENOTSUP.
1477 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1478 * with each cluster. By dry run we can find if the zero write
1479 * is possible without modifying image data.
1481 * Returns: error code with 0 for success.
1483 static int vmdk_write(BlockDriverState
*bs
, int64_t sector_num
,
1484 const uint8_t *buf
, int nb_sectors
,
1485 bool zeroed
, bool zero_dry_run
)
1487 BDRVVmdkState
*s
= bs
->opaque
;
1488 VmdkExtent
*extent
= NULL
;
1490 int64_t index_in_cluster
, n
;
1491 uint64_t cluster_offset
;
1492 VmdkMetaData m_data
;
1494 if (sector_num
> bs
->total_sectors
) {
1495 error_report("Wrong offset: sector_num=0x%" PRIx64
1496 " total_sectors=0x%" PRIx64
"\n",
1497 sector_num
, bs
->total_sectors
);
1501 while (nb_sectors
> 0) {
1502 extent
= find_extent(s
, sector_num
, extent
);
1506 index_in_cluster
= vmdk_find_index_in_cluster(extent
, sector_num
);
1507 n
= extent
->cluster_sectors
- index_in_cluster
;
1508 if (n
> nb_sectors
) {
1511 ret
= get_cluster_offset(bs
, extent
, &m_data
, sector_num
<< 9,
1512 !(extent
->compressed
|| zeroed
),
1514 index_in_cluster
, index_in_cluster
+ n
);
1515 if (extent
->compressed
) {
1516 if (ret
== VMDK_OK
) {
1517 /* Refuse write to allocated cluster for streamOptimized */
1518 error_report("Could not write to allocated cluster"
1519 " for streamOptimized");
1523 ret
= get_cluster_offset(bs
, extent
, &m_data
, sector_num
<< 9,
1524 true, &cluster_offset
, 0, 0);
1527 if (ret
== VMDK_ERROR
) {
1531 /* Do zeroed write, buf is ignored */
1532 if (extent
->has_zero_grain
&&
1533 index_in_cluster
== 0 &&
1534 n
>= extent
->cluster_sectors
) {
1535 n
= extent
->cluster_sectors
;
1536 if (!zero_dry_run
) {
1537 /* update L2 tables */
1538 if (vmdk_L2update(extent
, &m_data
, VMDK_GTE_ZEROED
)
1547 ret
= vmdk_write_extent(extent
,
1548 cluster_offset
, index_in_cluster
* 512,
1549 buf
, n
, sector_num
);
1554 /* update L2 tables */
1555 if (vmdk_L2update(extent
, &m_data
,
1556 cluster_offset
>> BDRV_SECTOR_BITS
)
1566 /* update CID on the first write every time the virtual disk is
1568 if (!s
->cid_updated
) {
1569 ret
= vmdk_write_cid(bs
, g_random_int());
1573 s
->cid_updated
= true;
1579 static coroutine_fn
int vmdk_co_write(BlockDriverState
*bs
, int64_t sector_num
,
1580 const uint8_t *buf
, int nb_sectors
)
1583 BDRVVmdkState
*s
= bs
->opaque
;
1584 qemu_co_mutex_lock(&s
->lock
);
1585 ret
= vmdk_write(bs
, sector_num
, buf
, nb_sectors
, false, false);
1586 qemu_co_mutex_unlock(&s
->lock
);
1590 static int vmdk_write_compressed(BlockDriverState
*bs
,
1595 BDRVVmdkState
*s
= bs
->opaque
;
1596 if (s
->num_extents
== 1 && s
->extents
[0].compressed
) {
1597 return vmdk_write(bs
, sector_num
, buf
, nb_sectors
, false, false);
1603 static int coroutine_fn
vmdk_co_write_zeroes(BlockDriverState
*bs
,
1606 BdrvRequestFlags flags
)
1609 BDRVVmdkState
*s
= bs
->opaque
;
1610 qemu_co_mutex_lock(&s
->lock
);
1611 /* write zeroes could fail if sectors not aligned to cluster, test it with
1612 * dry_run == true before really updating image */
1613 ret
= vmdk_write(bs
, sector_num
, NULL
, nb_sectors
, true, true);
1615 ret
= vmdk_write(bs
, sector_num
, NULL
, nb_sectors
, true, false);
1617 qemu_co_mutex_unlock(&s
->lock
);
1621 static int vmdk_create_extent(const char *filename
, int64_t filesize
,
1622 bool flat
, bool compress
, bool zeroed_grain
,
1623 QemuOpts
*opts
, Error
**errp
)
1626 BlockDriverState
*bs
= NULL
;
1628 Error
*local_err
= NULL
;
1629 uint32_t tmp
, magic
, grains
, gd_sectors
, gt_size
, gt_count
;
1630 uint32_t *gd_buf
= NULL
;
1633 ret
= bdrv_create_file(filename
, opts
, &local_err
);
1635 error_propagate(errp
, local_err
);
1640 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
, BDRV_O_RDWR
| BDRV_O_PROTOCOL
,
1643 error_propagate(errp
, local_err
);
1648 ret
= bdrv_truncate(bs
, filesize
);
1650 error_setg_errno(errp
, -ret
, "Could not truncate file");
1654 magic
= cpu_to_be32(VMDK4_MAGIC
);
1655 memset(&header
, 0, sizeof(header
));
1656 header
.version
= zeroed_grain
? 2 : 1;
1657 header
.flags
= VMDK4_FLAG_RGD
| VMDK4_FLAG_NL_DETECT
1658 | (compress
? VMDK4_FLAG_COMPRESS
| VMDK4_FLAG_MARKER
: 0)
1659 | (zeroed_grain
? VMDK4_FLAG_ZERO_GRAIN
: 0);
1660 header
.compressAlgorithm
= compress
? VMDK4_COMPRESSION_DEFLATE
: 0;
1661 header
.capacity
= filesize
/ BDRV_SECTOR_SIZE
;
1662 header
.granularity
= 128;
1663 header
.num_gtes_per_gt
= BDRV_SECTOR_SIZE
;
1665 grains
= DIV_ROUND_UP(filesize
/ BDRV_SECTOR_SIZE
, header
.granularity
);
1666 gt_size
= DIV_ROUND_UP(header
.num_gtes_per_gt
* sizeof(uint32_t),
1668 gt_count
= DIV_ROUND_UP(grains
, header
.num_gtes_per_gt
);
1669 gd_sectors
= DIV_ROUND_UP(gt_count
* sizeof(uint32_t), BDRV_SECTOR_SIZE
);
1671 header
.desc_offset
= 1;
1672 header
.desc_size
= 20;
1673 header
.rgd_offset
= header
.desc_offset
+ header
.desc_size
;
1674 header
.gd_offset
= header
.rgd_offset
+ gd_sectors
+ (gt_size
* gt_count
);
1675 header
.grain_offset
=
1676 ROUND_UP(header
.gd_offset
+ gd_sectors
+ (gt_size
* gt_count
),
1677 header
.granularity
);
1678 /* swap endianness for all header fields */
1679 header
.version
= cpu_to_le32(header
.version
);
1680 header
.flags
= cpu_to_le32(header
.flags
);
1681 header
.capacity
= cpu_to_le64(header
.capacity
);
1682 header
.granularity
= cpu_to_le64(header
.granularity
);
1683 header
.num_gtes_per_gt
= cpu_to_le32(header
.num_gtes_per_gt
);
1684 header
.desc_offset
= cpu_to_le64(header
.desc_offset
);
1685 header
.desc_size
= cpu_to_le64(header
.desc_size
);
1686 header
.rgd_offset
= cpu_to_le64(header
.rgd_offset
);
1687 header
.gd_offset
= cpu_to_le64(header
.gd_offset
);
1688 header
.grain_offset
= cpu_to_le64(header
.grain_offset
);
1689 header
.compressAlgorithm
= cpu_to_le16(header
.compressAlgorithm
);
1691 header
.check_bytes
[0] = 0xa;
1692 header
.check_bytes
[1] = 0x20;
1693 header
.check_bytes
[2] = 0xd;
1694 header
.check_bytes
[3] = 0xa;
1696 /* write all the data */
1697 ret
= bdrv_pwrite(bs
, 0, &magic
, sizeof(magic
));
1699 error_setg(errp
, QERR_IO_ERROR
);
1702 ret
= bdrv_pwrite(bs
, sizeof(magic
), &header
, sizeof(header
));
1704 error_setg(errp
, QERR_IO_ERROR
);
1708 ret
= bdrv_truncate(bs
, le64_to_cpu(header
.grain_offset
) << 9);
1710 error_setg_errno(errp
, -ret
, "Could not truncate file");
1714 /* write grain directory */
1715 gd_buf_size
= gd_sectors
* BDRV_SECTOR_SIZE
;
1716 gd_buf
= g_malloc0(gd_buf_size
);
1717 for (i
= 0, tmp
= le64_to_cpu(header
.rgd_offset
) + gd_sectors
;
1718 i
< gt_count
; i
++, tmp
+= gt_size
) {
1719 gd_buf
[i
] = cpu_to_le32(tmp
);
1721 ret
= bdrv_pwrite(bs
, le64_to_cpu(header
.rgd_offset
) * BDRV_SECTOR_SIZE
,
1722 gd_buf
, gd_buf_size
);
1724 error_setg(errp
, QERR_IO_ERROR
);
1728 /* write backup grain directory */
1729 for (i
= 0, tmp
= le64_to_cpu(header
.gd_offset
) + gd_sectors
;
1730 i
< gt_count
; i
++, tmp
+= gt_size
) {
1731 gd_buf
[i
] = cpu_to_le32(tmp
);
1733 ret
= bdrv_pwrite(bs
, le64_to_cpu(header
.gd_offset
) * BDRV_SECTOR_SIZE
,
1734 gd_buf
, gd_buf_size
);
1736 error_setg(errp
, QERR_IO_ERROR
);
1749 static int filename_decompose(const char *filename
, char *path
, char *prefix
,
1750 char *postfix
, size_t buf_len
, Error
**errp
)
1754 if (filename
== NULL
|| !strlen(filename
)) {
1755 error_setg(errp
, "No filename provided");
1758 p
= strrchr(filename
, '/');
1760 p
= strrchr(filename
, '\\');
1763 p
= strrchr(filename
, ':');
1767 if (p
- filename
>= buf_len
) {
1770 pstrcpy(path
, p
- filename
+ 1, filename
);
1775 q
= strrchr(p
, '.');
1777 pstrcpy(prefix
, buf_len
, p
);
1780 if (q
- p
>= buf_len
) {
1783 pstrcpy(prefix
, q
- p
+ 1, p
);
1784 pstrcpy(postfix
, buf_len
, q
);
1789 static int vmdk_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
1792 BlockDriverState
*new_bs
= NULL
;
1793 Error
*local_err
= NULL
;
1795 int64_t total_size
= 0, filesize
;
1796 char *adapter_type
= NULL
;
1797 char *backing_file
= NULL
;
1801 bool flat
, split
, compress
;
1802 GString
*ext_desc_lines
;
1803 char *path
= g_malloc0(PATH_MAX
);
1804 char *prefix
= g_malloc0(PATH_MAX
);
1805 char *postfix
= g_malloc0(PATH_MAX
);
1806 char *desc_line
= g_malloc0(BUF_SIZE
);
1807 char *ext_filename
= g_malloc0(PATH_MAX
);
1808 char *desc_filename
= g_malloc0(PATH_MAX
);
1809 const int64_t split_size
= 0x80000000; /* VMDK has constant split size */
1810 const char *desc_extent_line
;
1811 char *parent_desc_line
= g_malloc0(BUF_SIZE
);
1812 uint32_t parent_cid
= 0xffffffff;
1813 uint32_t number_heads
= 16;
1814 bool zeroed_grain
= false;
1815 uint32_t desc_offset
= 0, desc_len
;
1816 const char desc_template
[] =
1817 "# Disk DescriptorFile\n"
1820 "parentCID=%" PRIx32
"\n"
1821 "createType=\"%s\"\n"
1824 "# Extent description\n"
1827 "# The Disk Data Base\n"
1830 "ddb.virtualHWVersion = \"%d\"\n"
1831 "ddb.geometry.cylinders = \"%" PRId64
"\"\n"
1832 "ddb.geometry.heads = \"%" PRIu32
"\"\n"
1833 "ddb.geometry.sectors = \"63\"\n"
1834 "ddb.adapterType = \"%s\"\n";
1836 ext_desc_lines
= g_string_new(NULL
);
1838 if (filename_decompose(filename
, path
, prefix
, postfix
, PATH_MAX
, errp
)) {
1842 /* Read out options */
1843 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
1845 adapter_type
= qemu_opt_get_del(opts
, BLOCK_OPT_ADAPTER_TYPE
);
1846 backing_file
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
1847 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_COMPAT6
, false)) {
1848 flags
|= BLOCK_FLAG_COMPAT6
;
1850 fmt
= qemu_opt_get_del(opts
, BLOCK_OPT_SUBFMT
);
1851 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_ZEROED_GRAIN
, false)) {
1852 zeroed_grain
= true;
1855 if (!adapter_type
) {
1856 adapter_type
= g_strdup("ide");
1857 } else if (strcmp(adapter_type
, "ide") &&
1858 strcmp(adapter_type
, "buslogic") &&
1859 strcmp(adapter_type
, "lsilogic") &&
1860 strcmp(adapter_type
, "legacyESX")) {
1861 error_setg(errp
, "Unknown adapter type: '%s'", adapter_type
);
1865 if (strcmp(adapter_type
, "ide") != 0) {
1866 /* that's the number of heads with which vmware operates when
1867 creating, exporting, etc. vmdk files with a non-ide adapter type */
1871 /* Default format to monolithicSparse */
1872 fmt
= g_strdup("monolithicSparse");
1873 } else if (strcmp(fmt
, "monolithicFlat") &&
1874 strcmp(fmt
, "monolithicSparse") &&
1875 strcmp(fmt
, "twoGbMaxExtentSparse") &&
1876 strcmp(fmt
, "twoGbMaxExtentFlat") &&
1877 strcmp(fmt
, "streamOptimized")) {
1878 error_setg(errp
, "Unknown subformat: '%s'", fmt
);
1882 split
= !(strcmp(fmt
, "twoGbMaxExtentFlat") &&
1883 strcmp(fmt
, "twoGbMaxExtentSparse"));
1884 flat
= !(strcmp(fmt
, "monolithicFlat") &&
1885 strcmp(fmt
, "twoGbMaxExtentFlat"));
1886 compress
= !strcmp(fmt
, "streamOptimized");
1888 desc_extent_line
= "RW %" PRId64
" FLAT \"%s\" 0\n";
1890 desc_extent_line
= "RW %" PRId64
" SPARSE \"%s\"\n";
1892 if (flat
&& backing_file
) {
1893 error_setg(errp
, "Flat image can't have backing file");
1897 if (flat
&& zeroed_grain
) {
1898 error_setg(errp
, "Flat image can't enable zeroed grain");
1903 BlockDriverState
*bs
= NULL
;
1904 char *full_backing
= g_new0(char, PATH_MAX
);
1905 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
1906 full_backing
, PATH_MAX
,
1909 g_free(full_backing
);
1910 error_propagate(errp
, local_err
);
1914 ret
= bdrv_open(&bs
, full_backing
, NULL
, NULL
, BDRV_O_NO_BACKING
, errp
);
1915 g_free(full_backing
);
1919 if (strcmp(bs
->drv
->format_name
, "vmdk")) {
1924 parent_cid
= vmdk_read_cid(bs
, 0);
1926 snprintf(parent_desc_line
, BUF_SIZE
,
1927 "parentFileNameHint=\"%s\"", backing_file
);
1930 /* Create extents */
1931 filesize
= total_size
;
1932 while (filesize
> 0) {
1933 int64_t size
= filesize
;
1935 if (split
&& size
> split_size
) {
1939 snprintf(desc_filename
, PATH_MAX
, "%s-%c%03d%s",
1940 prefix
, flat
? 'f' : 's', ++idx
, postfix
);
1942 snprintf(desc_filename
, PATH_MAX
, "%s-flat%s", prefix
, postfix
);
1944 snprintf(desc_filename
, PATH_MAX
, "%s%s", prefix
, postfix
);
1946 snprintf(ext_filename
, PATH_MAX
, "%s%s", path
, desc_filename
);
1948 if (vmdk_create_extent(ext_filename
, size
,
1949 flat
, compress
, zeroed_grain
, opts
, errp
)) {
1955 /* Format description line */
1956 snprintf(desc_line
, BUF_SIZE
,
1957 desc_extent_line
, size
/ BDRV_SECTOR_SIZE
, desc_filename
);
1958 g_string_append(ext_desc_lines
, desc_line
);
1960 /* generate descriptor file */
1961 desc
= g_strdup_printf(desc_template
,
1966 ext_desc_lines
->str
,
1967 (flags
& BLOCK_FLAG_COMPAT6
? 6 : 4),
1969 (int64_t)(63 * number_heads
* BDRV_SECTOR_SIZE
),
1972 desc_len
= strlen(desc
);
1973 /* the descriptor offset = 0x200 */
1974 if (!split
&& !flat
) {
1975 desc_offset
= 0x200;
1977 ret
= bdrv_create_file(filename
, opts
, &local_err
);
1979 error_propagate(errp
, local_err
);
1983 assert(new_bs
== NULL
);
1984 ret
= bdrv_open(&new_bs
, filename
, NULL
, NULL
,
1985 BDRV_O_RDWR
| BDRV_O_PROTOCOL
, &local_err
);
1987 error_propagate(errp
, local_err
);
1990 ret
= bdrv_pwrite(new_bs
, desc_offset
, desc
, desc_len
);
1992 error_setg_errno(errp
, -ret
, "Could not write description");
1995 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
1996 * for description file */
1997 if (desc_offset
== 0) {
1998 ret
= bdrv_truncate(new_bs
, desc_len
);
2000 error_setg_errno(errp
, -ret
, "Could not truncate file");
2007 g_free(adapter_type
);
2008 g_free(backing_file
);
2015 g_free(ext_filename
);
2016 g_free(desc_filename
);
2017 g_free(parent_desc_line
);
2018 g_string_free(ext_desc_lines
, true);
2022 static void vmdk_close(BlockDriverState
*bs
)
2024 BDRVVmdkState
*s
= bs
->opaque
;
2026 vmdk_free_extents(bs
);
2027 g_free(s
->create_type
);
2029 migrate_del_blocker(s
->migration_blocker
);
2030 error_free(s
->migration_blocker
);
2033 static coroutine_fn
int vmdk_co_flush(BlockDriverState
*bs
)
2035 BDRVVmdkState
*s
= bs
->opaque
;
2039 for (i
= 0; i
< s
->num_extents
; i
++) {
2040 err
= bdrv_co_flush(s
->extents
[i
].file
->bs
);
2048 static int64_t vmdk_get_allocated_file_size(BlockDriverState
*bs
)
2053 BDRVVmdkState
*s
= bs
->opaque
;
2055 ret
= bdrv_get_allocated_file_size(bs
->file
->bs
);
2059 for (i
= 0; i
< s
->num_extents
; i
++) {
2060 if (s
->extents
[i
].file
== bs
->file
) {
2063 r
= bdrv_get_allocated_file_size(s
->extents
[i
].file
->bs
);
2072 static int vmdk_has_zero_init(BlockDriverState
*bs
)
2075 BDRVVmdkState
*s
= bs
->opaque
;
2077 /* If has a flat extent and its underlying storage doesn't have zero init,
2079 for (i
= 0; i
< s
->num_extents
; i
++) {
2080 if (s
->extents
[i
].flat
) {
2081 if (!bdrv_has_zero_init(s
->extents
[i
].file
->bs
)) {
2089 static ImageInfo
*vmdk_get_extent_info(VmdkExtent
*extent
)
2091 ImageInfo
*info
= g_new0(ImageInfo
, 1);
2093 *info
= (ImageInfo
){
2094 .filename
= g_strdup(extent
->file
->bs
->filename
),
2095 .format
= g_strdup(extent
->type
),
2096 .virtual_size
= extent
->sectors
* BDRV_SECTOR_SIZE
,
2097 .compressed
= extent
->compressed
,
2098 .has_compressed
= extent
->compressed
,
2099 .cluster_size
= extent
->cluster_sectors
* BDRV_SECTOR_SIZE
,
2100 .has_cluster_size
= !extent
->flat
,
2106 static int vmdk_check(BlockDriverState
*bs
, BdrvCheckResult
*result
,
2109 BDRVVmdkState
*s
= bs
->opaque
;
2110 VmdkExtent
*extent
= NULL
;
2111 int64_t sector_num
= 0;
2112 int64_t total_sectors
= bdrv_nb_sectors(bs
);
2114 uint64_t cluster_offset
;
2121 if (sector_num
>= total_sectors
) {
2124 extent
= find_extent(s
, sector_num
, extent
);
2127 "ERROR: could not find extent for sector %" PRId64
"\n",
2131 ret
= get_cluster_offset(bs
, extent
, NULL
,
2132 sector_num
<< BDRV_SECTOR_BITS
,
2133 false, &cluster_offset
, 0, 0);
2134 if (ret
== VMDK_ERROR
) {
2136 "ERROR: could not get cluster_offset for sector %"
2137 PRId64
"\n", sector_num
);
2140 if (ret
== VMDK_OK
&&
2141 cluster_offset
>= bdrv_getlength(extent
->file
->bs
))
2144 "ERROR: cluster offset for sector %"
2145 PRId64
" points after EOF\n", sector_num
);
2148 sector_num
+= extent
->cluster_sectors
;
2151 result
->corruptions
++;
2155 static ImageInfoSpecific
*vmdk_get_specific_info(BlockDriverState
*bs
)
2158 BDRVVmdkState
*s
= bs
->opaque
;
2159 ImageInfoSpecific
*spec_info
= g_new0(ImageInfoSpecific
, 1);
2160 ImageInfoList
**next
;
2162 *spec_info
= (ImageInfoSpecific
){
2163 .kind
= IMAGE_INFO_SPECIFIC_KIND_VMDK
,
2165 .vmdk
= g_new0(ImageInfoSpecificVmdk
, 1),
2169 *spec_info
->vmdk
= (ImageInfoSpecificVmdk
) {
2170 .create_type
= g_strdup(s
->create_type
),
2172 .parent_cid
= s
->parent_cid
,
2175 next
= &spec_info
->vmdk
->extents
;
2176 for (i
= 0; i
< s
->num_extents
; i
++) {
2177 *next
= g_new0(ImageInfoList
, 1);
2178 (*next
)->value
= vmdk_get_extent_info(&s
->extents
[i
]);
2179 (*next
)->next
= NULL
;
2180 next
= &(*next
)->next
;
2186 static bool vmdk_extents_type_eq(const VmdkExtent
*a
, const VmdkExtent
*b
)
2188 return a
->flat
== b
->flat
&&
2189 a
->compressed
== b
->compressed
&&
2190 (a
->flat
|| a
->cluster_sectors
== b
->cluster_sectors
);
2193 static int vmdk_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2196 BDRVVmdkState
*s
= bs
->opaque
;
2197 assert(s
->num_extents
);
2199 /* See if we have multiple extents but they have different cases */
2200 for (i
= 1; i
< s
->num_extents
; i
++) {
2201 if (!vmdk_extents_type_eq(&s
->extents
[0], &s
->extents
[i
])) {
2205 bdi
->needs_compressed_writes
= s
->extents
[0].compressed
;
2206 if (!s
->extents
[0].flat
) {
2207 bdi
->cluster_size
= s
->extents
[0].cluster_sectors
<< BDRV_SECTOR_BITS
;
2212 static void vmdk_detach_aio_context(BlockDriverState
*bs
)
2214 BDRVVmdkState
*s
= bs
->opaque
;
2217 for (i
= 0; i
< s
->num_extents
; i
++) {
2218 bdrv_detach_aio_context(s
->extents
[i
].file
->bs
);
2222 static void vmdk_attach_aio_context(BlockDriverState
*bs
,
2223 AioContext
*new_context
)
2225 BDRVVmdkState
*s
= bs
->opaque
;
2228 for (i
= 0; i
< s
->num_extents
; i
++) {
2229 bdrv_attach_aio_context(s
->extents
[i
].file
->bs
, new_context
);
2233 static QemuOptsList vmdk_create_opts
= {
2234 .name
= "vmdk-create-opts",
2235 .head
= QTAILQ_HEAD_INITIALIZER(vmdk_create_opts
.head
),
2238 .name
= BLOCK_OPT_SIZE
,
2239 .type
= QEMU_OPT_SIZE
,
2240 .help
= "Virtual disk size"
2243 .name
= BLOCK_OPT_ADAPTER_TYPE
,
2244 .type
= QEMU_OPT_STRING
,
2245 .help
= "Virtual adapter type, can be one of "
2246 "ide (default), lsilogic, buslogic or legacyESX"
2249 .name
= BLOCK_OPT_BACKING_FILE
,
2250 .type
= QEMU_OPT_STRING
,
2251 .help
= "File name of a base image"
2254 .name
= BLOCK_OPT_COMPAT6
,
2255 .type
= QEMU_OPT_BOOL
,
2256 .help
= "VMDK version 6 image",
2257 .def_value_str
= "off"
2260 .name
= BLOCK_OPT_SUBFMT
,
2261 .type
= QEMU_OPT_STRING
,
2263 "VMDK flat extent format, can be one of "
2264 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2267 .name
= BLOCK_OPT_ZEROED_GRAIN
,
2268 .type
= QEMU_OPT_BOOL
,
2269 .help
= "Enable efficient zero writes "
2270 "using the zeroed-grain GTE feature"
2272 { /* end of list */ }
2276 static BlockDriver bdrv_vmdk
= {
2277 .format_name
= "vmdk",
2278 .instance_size
= sizeof(BDRVVmdkState
),
2279 .bdrv_probe
= vmdk_probe
,
2280 .bdrv_open
= vmdk_open
,
2281 .bdrv_check
= vmdk_check
,
2282 .bdrv_reopen_prepare
= vmdk_reopen_prepare
,
2283 .bdrv_read
= vmdk_co_read
,
2284 .bdrv_write
= vmdk_co_write
,
2285 .bdrv_write_compressed
= vmdk_write_compressed
,
2286 .bdrv_co_write_zeroes
= vmdk_co_write_zeroes
,
2287 .bdrv_close
= vmdk_close
,
2288 .bdrv_create
= vmdk_create
,
2289 .bdrv_co_flush_to_disk
= vmdk_co_flush
,
2290 .bdrv_co_get_block_status
= vmdk_co_get_block_status
,
2291 .bdrv_get_allocated_file_size
= vmdk_get_allocated_file_size
,
2292 .bdrv_has_zero_init
= vmdk_has_zero_init
,
2293 .bdrv_get_specific_info
= vmdk_get_specific_info
,
2294 .bdrv_refresh_limits
= vmdk_refresh_limits
,
2295 .bdrv_get_info
= vmdk_get_info
,
2296 .bdrv_detach_aio_context
= vmdk_detach_aio_context
,
2297 .bdrv_attach_aio_context
= vmdk_attach_aio_context
,
2299 .supports_backing
= true,
2300 .create_opts
= &vmdk_create_opts
,
2303 static void bdrv_vmdk_init(void)
2305 bdrv_register(&bdrv_vmdk
);
2308 block_init(bdrv_vmdk_init
);