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/osdep.h"
27 #include "qapi/error.h"
28 #include "block/block_int.h"
29 #include "sysemu/block-backend.h"
30 #include "qapi/qmp/qerror.h"
31 #include "qemu/error-report.h"
32 #include "qemu/module.h"
33 #include "migration/migration.h"
34 #include "qemu/cutils.h"
38 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
39 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
40 #define VMDK4_COMPRESSION_DEFLATE 1
41 #define VMDK4_FLAG_NL_DETECT (1 << 0)
42 #define VMDK4_FLAG_RGD (1 << 1)
43 /* Zeroed-grain enable bit */
44 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
45 #define VMDK4_FLAG_COMPRESS (1 << 16)
46 #define VMDK4_FLAG_MARKER (1 << 17)
47 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
49 #define VMDK_GTE_ZEROED 0x1
51 /* VMDK internal error codes */
53 #define VMDK_ERROR (-1)
54 /* Cluster not allocated */
55 #define VMDK_UNALLOC (-2)
56 #define VMDK_ZEROED (-3)
58 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
63 uint32_t disk_sectors
;
65 uint32_t l1dir_offset
;
67 uint32_t file_sectors
;
70 uint32_t sectors_per_track
;
71 } QEMU_PACKED VMDK3Header
;
80 /* Number of GrainTableEntries per GrainTable */
81 uint32_t num_gtes_per_gt
;
84 uint64_t grain_offset
;
87 uint16_t compressAlgorithm
;
88 } QEMU_PACKED VMDK4Header
;
90 #define L2_CACHE_SIZE 16
92 typedef struct VmdkExtent
{
101 int64_t flat_start_offset
;
102 int64_t l1_table_offset
;
103 int64_t l1_backup_table_offset
;
105 uint32_t *l1_backup_table
;
106 unsigned int l1_size
;
107 uint32_t l1_entry_sectors
;
109 unsigned int l2_size
;
111 uint32_t l2_cache_offsets
[L2_CACHE_SIZE
];
112 uint32_t l2_cache_counts
[L2_CACHE_SIZE
];
114 int64_t cluster_sectors
;
115 int64_t next_cluster_sector
;
119 typedef struct BDRVVmdkState
{
121 uint64_t desc_offset
;
127 /* Extent array with num_extents entries, ascend ordered by address */
129 Error
*migration_blocker
;
133 typedef struct VmdkMetaData
{
134 unsigned int l1_index
;
135 unsigned int l2_index
;
136 unsigned int l2_offset
;
138 uint32_t *l2_cache_entry
;
141 typedef struct VmdkGrainMarker
{
145 } QEMU_PACKED VmdkGrainMarker
;
148 MARKER_END_OF_STREAM
= 0,
149 MARKER_GRAIN_TABLE
= 1,
150 MARKER_GRAIN_DIRECTORY
= 2,
154 static int vmdk_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
161 magic
= be32_to_cpu(*(uint32_t *)buf
);
162 if (magic
== VMDK3_MAGIC
||
163 magic
== VMDK4_MAGIC
) {
166 const char *p
= (const char *)buf
;
167 const char *end
= p
+ buf_size
;
170 /* skip comment line */
171 while (p
< end
&& *p
!= '\n') {
178 while (p
< end
&& *p
== ' ') {
181 /* skip '\r' if windows line endings used. */
182 if (p
< end
&& *p
== '\r') {
185 /* only accept blank lines before 'version=' line */
186 if (p
== end
|| *p
!= '\n') {
192 if (end
- p
>= strlen("version=X\n")) {
193 if (strncmp("version=1\n", p
, strlen("version=1\n")) == 0 ||
194 strncmp("version=2\n", p
, strlen("version=2\n")) == 0) {
198 if (end
- p
>= strlen("version=X\r\n")) {
199 if (strncmp("version=1\r\n", p
, strlen("version=1\r\n")) == 0 ||
200 strncmp("version=2\r\n", p
, strlen("version=2\r\n")) == 0) {
210 #define SECTOR_SIZE 512
211 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
212 #define BUF_SIZE 4096
213 #define HEADER_SIZE 512 /* first sector of 512 bytes */
215 static void vmdk_free_extents(BlockDriverState
*bs
)
218 BDRVVmdkState
*s
= bs
->opaque
;
221 for (i
= 0; i
< s
->num_extents
; i
++) {
225 g_free(e
->l1_backup_table
);
227 if (e
->file
!= bs
->file
) {
228 bdrv_unref_child(bs
, e
->file
);
234 static void vmdk_free_last_extent(BlockDriverState
*bs
)
236 BDRVVmdkState
*s
= bs
->opaque
;
238 if (s
->num_extents
== 0) {
242 s
->extents
= g_renew(VmdkExtent
, s
->extents
, s
->num_extents
);
245 static uint32_t vmdk_read_cid(BlockDriverState
*bs
, int parent
)
248 uint32_t cid
= 0xffffffff;
249 const char *p_name
, *cid_str
;
251 BDRVVmdkState
*s
= bs
->opaque
;
254 desc
= g_malloc0(DESC_SIZE
);
255 ret
= bdrv_pread(bs
->file
->bs
, s
->desc_offset
, desc
, DESC_SIZE
);
262 cid_str
= "parentCID";
263 cid_str_size
= sizeof("parentCID");
266 cid_str_size
= sizeof("CID");
269 desc
[DESC_SIZE
- 1] = '\0';
270 p_name
= strstr(desc
, cid_str
);
271 if (p_name
!= NULL
) {
272 p_name
+= cid_str_size
;
273 sscanf(p_name
, "%" SCNx32
, &cid
);
280 static int vmdk_write_cid(BlockDriverState
*bs
, uint32_t cid
)
282 char *desc
, *tmp_desc
;
283 char *p_name
, *tmp_str
;
284 BDRVVmdkState
*s
= bs
->opaque
;
287 desc
= g_malloc0(DESC_SIZE
);
288 tmp_desc
= g_malloc0(DESC_SIZE
);
289 ret
= bdrv_pread(bs
->file
->bs
, s
->desc_offset
, desc
, DESC_SIZE
);
294 desc
[DESC_SIZE
- 1] = '\0';
295 tmp_str
= strstr(desc
, "parentCID");
296 if (tmp_str
== NULL
) {
301 pstrcpy(tmp_desc
, DESC_SIZE
, tmp_str
);
302 p_name
= strstr(desc
, "CID");
303 if (p_name
!= NULL
) {
304 p_name
+= sizeof("CID");
305 snprintf(p_name
, DESC_SIZE
- (p_name
- desc
), "%" PRIx32
"\n", cid
);
306 pstrcat(desc
, DESC_SIZE
, tmp_desc
);
309 ret
= bdrv_pwrite_sync(bs
->file
->bs
, s
->desc_offset
, desc
, DESC_SIZE
);
317 static int vmdk_is_cid_valid(BlockDriverState
*bs
)
319 BDRVVmdkState
*s
= bs
->opaque
;
322 if (!s
->cid_checked
&& bs
->backing
) {
323 BlockDriverState
*p_bs
= bs
->backing
->bs
;
325 cur_pcid
= vmdk_read_cid(p_bs
, 0);
326 if (s
->parent_cid
!= cur_pcid
) {
331 s
->cid_checked
= true;
336 /* We have nothing to do for VMDK reopen, stubs just return success */
337 static int vmdk_reopen_prepare(BDRVReopenState
*state
,
338 BlockReopenQueue
*queue
, Error
**errp
)
340 assert(state
!= NULL
);
341 assert(state
->bs
!= NULL
);
345 static int vmdk_parent_open(BlockDriverState
*bs
)
349 BDRVVmdkState
*s
= bs
->opaque
;
352 desc
= g_malloc0(DESC_SIZE
+ 1);
353 ret
= bdrv_pread(bs
->file
->bs
, s
->desc_offset
, desc
, DESC_SIZE
);
359 p_name
= strstr(desc
, "parentFileNameHint");
360 if (p_name
!= NULL
) {
363 p_name
+= sizeof("parentFileNameHint") + 1;
364 end_name
= strchr(p_name
, '\"');
365 if (end_name
== NULL
) {
369 if ((end_name
- p_name
) > sizeof(bs
->backing_file
) - 1) {
374 pstrcpy(bs
->backing_file
, end_name
- p_name
+ 1, p_name
);
382 /* Create and append extent to the extent array. Return the added VmdkExtent
383 * address. return NULL if allocation failed. */
384 static int vmdk_add_extent(BlockDriverState
*bs
,
385 BdrvChild
*file
, bool flat
, int64_t sectors
,
386 int64_t l1_offset
, int64_t l1_backup_offset
,
388 int l2_size
, uint64_t cluster_sectors
,
389 VmdkExtent
**new_extent
,
393 BDRVVmdkState
*s
= bs
->opaque
;
396 if (cluster_sectors
> 0x200000) {
397 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
398 error_setg(errp
, "Invalid granularity, image may be corrupt");
401 if (l1_size
> 512 * 1024 * 1024) {
402 /* Although with big capacity and small l1_entry_sectors, we can get a
403 * big l1_size, we don't want unbounded value to allocate the table.
404 * Limit it to 512M, which is 16PB for default cluster and L2 table
406 error_setg(errp
, "L1 size too big");
410 nb_sectors
= bdrv_nb_sectors(file
->bs
);
411 if (nb_sectors
< 0) {
415 s
->extents
= g_renew(VmdkExtent
, s
->extents
, s
->num_extents
+ 1);
416 extent
= &s
->extents
[s
->num_extents
];
419 memset(extent
, 0, sizeof(VmdkExtent
));
422 extent
->sectors
= sectors
;
423 extent
->l1_table_offset
= l1_offset
;
424 extent
->l1_backup_table_offset
= l1_backup_offset
;
425 extent
->l1_size
= l1_size
;
426 extent
->l1_entry_sectors
= l2_size
* cluster_sectors
;
427 extent
->l2_size
= l2_size
;
428 extent
->cluster_sectors
= flat
? sectors
: cluster_sectors
;
429 extent
->next_cluster_sector
= ROUND_UP(nb_sectors
, cluster_sectors
);
431 if (s
->num_extents
> 1) {
432 extent
->end_sector
= (*(extent
- 1)).end_sector
+ extent
->sectors
;
434 extent
->end_sector
= extent
->sectors
;
436 bs
->total_sectors
= extent
->end_sector
;
438 *new_extent
= extent
;
443 static int vmdk_init_tables(BlockDriverState
*bs
, VmdkExtent
*extent
,
450 /* read the L1 table */
451 l1_size
= extent
->l1_size
* sizeof(uint32_t);
452 extent
->l1_table
= g_try_malloc(l1_size
);
453 if (l1_size
&& extent
->l1_table
== NULL
) {
457 ret
= bdrv_pread(extent
->file
->bs
,
458 extent
->l1_table_offset
,
462 error_setg_errno(errp
, -ret
,
463 "Could not read l1 table from extent '%s'",
464 extent
->file
->bs
->filename
);
467 for (i
= 0; i
< extent
->l1_size
; i
++) {
468 le32_to_cpus(&extent
->l1_table
[i
]);
471 if (extent
->l1_backup_table_offset
) {
472 extent
->l1_backup_table
= g_try_malloc(l1_size
);
473 if (l1_size
&& extent
->l1_backup_table
== NULL
) {
477 ret
= bdrv_pread(extent
->file
->bs
,
478 extent
->l1_backup_table_offset
,
479 extent
->l1_backup_table
,
482 error_setg_errno(errp
, -ret
,
483 "Could not read l1 backup table from extent '%s'",
484 extent
->file
->bs
->filename
);
487 for (i
= 0; i
< extent
->l1_size
; i
++) {
488 le32_to_cpus(&extent
->l1_backup_table
[i
]);
493 g_new(uint32_t, extent
->l2_size
* L2_CACHE_SIZE
);
496 g_free(extent
->l1_backup_table
);
498 g_free(extent
->l1_table
);
502 static int vmdk_open_vmfs_sparse(BlockDriverState
*bs
,
504 int flags
, Error
**errp
)
511 ret
= bdrv_pread(file
->bs
, sizeof(magic
), &header
, sizeof(header
));
513 error_setg_errno(errp
, -ret
,
514 "Could not read header from file '%s'",
518 ret
= vmdk_add_extent(bs
, file
, false,
519 le32_to_cpu(header
.disk_sectors
),
520 (int64_t)le32_to_cpu(header
.l1dir_offset
) << 9,
522 le32_to_cpu(header
.l1dir_size
),
524 le32_to_cpu(header
.granularity
),
530 ret
= vmdk_init_tables(bs
, extent
, errp
);
532 /* free extent allocated by vmdk_add_extent */
533 vmdk_free_last_extent(bs
);
538 static int vmdk_open_desc_file(BlockDriverState
*bs
, int flags
, char *buf
,
539 QDict
*options
, Error
**errp
);
541 static char *vmdk_read_desc(BlockDriverState
*file
, uint64_t desc_offset
,
548 size
= bdrv_getlength(file
);
550 error_setg_errno(errp
, -size
, "Could not access file");
555 /* Both descriptor file and sparse image must be much larger than 4
556 * bytes, also callers of vmdk_read_desc want to compare the first 4
557 * bytes with VMDK4_MAGIC, let's error out if less is read. */
558 error_setg(errp
, "File is too small, not a valid image");
562 size
= MIN(size
, (1 << 20) - 1); /* avoid unbounded allocation */
563 buf
= g_malloc(size
+ 1);
565 ret
= bdrv_pread(file
, desc_offset
, buf
, size
);
567 error_setg_errno(errp
, -ret
, "Could not read from file");
576 static int vmdk_open_vmdk4(BlockDriverState
*bs
,
578 int flags
, QDict
*options
, Error
**errp
)
582 uint32_t l1_size
, l1_entry_sectors
;
585 BDRVVmdkState
*s
= bs
->opaque
;
586 int64_t l1_backup_offset
= 0;
589 ret
= bdrv_pread(file
->bs
, sizeof(magic
), &header
, sizeof(header
));
591 error_setg_errno(errp
, -ret
,
592 "Could not read header from file '%s'",
596 if (header
.capacity
== 0) {
597 uint64_t desc_offset
= le64_to_cpu(header
.desc_offset
);
599 char *buf
= vmdk_read_desc(file
->bs
, desc_offset
<< 9, errp
);
603 ret
= vmdk_open_desc_file(bs
, flags
, buf
, options
, errp
);
609 if (!s
->create_type
) {
610 s
->create_type
= g_strdup("monolithicSparse");
613 if (le64_to_cpu(header
.gd_offset
) == VMDK4_GD_AT_END
) {
615 * The footer takes precedence over the header, so read it in. The
616 * footer starts at offset -1024 from the end: One sector for the
617 * footer, and another one for the end-of-stream marker.
624 uint8_t pad
[512 - 16];
625 } QEMU_PACKED footer_marker
;
629 uint8_t pad
[512 - 4 - sizeof(VMDK4Header
)];
635 uint8_t pad
[512 - 16];
636 } QEMU_PACKED eos_marker
;
637 } QEMU_PACKED footer
;
639 ret
= bdrv_pread(file
->bs
,
640 bs
->file
->bs
->total_sectors
* 512 - 1536,
641 &footer
, sizeof(footer
));
643 error_setg_errno(errp
, -ret
, "Failed to read footer");
647 /* Some sanity checks for the footer */
648 if (be32_to_cpu(footer
.magic
) != VMDK4_MAGIC
||
649 le32_to_cpu(footer
.footer_marker
.size
) != 0 ||
650 le32_to_cpu(footer
.footer_marker
.type
) != MARKER_FOOTER
||
651 le64_to_cpu(footer
.eos_marker
.val
) != 0 ||
652 le32_to_cpu(footer
.eos_marker
.size
) != 0 ||
653 le32_to_cpu(footer
.eos_marker
.type
) != MARKER_END_OF_STREAM
)
655 error_setg(errp
, "Invalid footer");
659 header
= footer
.header
;
663 le16_to_cpu(header
.compressAlgorithm
) == VMDK4_COMPRESSION_DEFLATE
;
664 if (le32_to_cpu(header
.version
) > 3) {
665 error_setg(errp
, "Unsupported VMDK version %" PRIu32
,
666 le32_to_cpu(header
.version
));
668 } else if (le32_to_cpu(header
.version
) == 3 && (flags
& BDRV_O_RDWR
) &&
670 /* VMware KB 2064959 explains that version 3 added support for
671 * persistent changed block tracking (CBT), and backup software can
672 * read it as version=1 if it doesn't care about the changed area
673 * information. So we are safe to enable read only. */
674 error_setg(errp
, "VMDK version 3 must be read only");
678 if (le32_to_cpu(header
.num_gtes_per_gt
) > 512) {
679 error_setg(errp
, "L2 table size too big");
683 l1_entry_sectors
= le32_to_cpu(header
.num_gtes_per_gt
)
684 * le64_to_cpu(header
.granularity
);
685 if (l1_entry_sectors
== 0) {
686 error_setg(errp
, "L1 entry size is invalid");
689 l1_size
= (le64_to_cpu(header
.capacity
) + l1_entry_sectors
- 1)
691 if (le32_to_cpu(header
.flags
) & VMDK4_FLAG_RGD
) {
692 l1_backup_offset
= le64_to_cpu(header
.rgd_offset
) << 9;
694 if (bdrv_nb_sectors(file
->bs
) < le64_to_cpu(header
.grain_offset
)) {
695 error_setg(errp
, "File truncated, expecting at least %" PRId64
" bytes",
696 (int64_t)(le64_to_cpu(header
.grain_offset
)
697 * BDRV_SECTOR_SIZE
));
701 ret
= vmdk_add_extent(bs
, file
, false,
702 le64_to_cpu(header
.capacity
),
703 le64_to_cpu(header
.gd_offset
) << 9,
706 le32_to_cpu(header
.num_gtes_per_gt
),
707 le64_to_cpu(header
.granularity
),
714 le16_to_cpu(header
.compressAlgorithm
) == VMDK4_COMPRESSION_DEFLATE
;
715 if (extent
->compressed
) {
716 g_free(s
->create_type
);
717 s
->create_type
= g_strdup("streamOptimized");
719 extent
->has_marker
= le32_to_cpu(header
.flags
) & VMDK4_FLAG_MARKER
;
720 extent
->version
= le32_to_cpu(header
.version
);
721 extent
->has_zero_grain
= le32_to_cpu(header
.flags
) & VMDK4_FLAG_ZERO_GRAIN
;
722 ret
= vmdk_init_tables(bs
, extent
, errp
);
724 /* free extent allocated by vmdk_add_extent */
725 vmdk_free_last_extent(bs
);
730 /* find an option value out of descriptor file */
731 static int vmdk_parse_description(const char *desc
, const char *opt_name
,
732 char *buf
, int buf_size
)
734 char *opt_pos
, *opt_end
;
735 const char *end
= desc
+ strlen(desc
);
737 opt_pos
= strstr(desc
, opt_name
);
741 /* Skip "=\"" following opt_name */
742 opt_pos
+= strlen(opt_name
) + 2;
743 if (opt_pos
>= end
) {
747 while (opt_end
< end
&& *opt_end
!= '"') {
750 if (opt_end
== end
|| buf_size
< opt_end
- opt_pos
+ 1) {
753 pstrcpy(buf
, opt_end
- opt_pos
+ 1, opt_pos
);
757 /* Open an extent file and append to bs array */
758 static int vmdk_open_sparse(BlockDriverState
*bs
, BdrvChild
*file
, int flags
,
759 char *buf
, QDict
*options
, Error
**errp
)
763 magic
= ldl_be_p(buf
);
766 return vmdk_open_vmfs_sparse(bs
, file
, flags
, errp
);
769 return vmdk_open_vmdk4(bs
, file
, flags
, options
, errp
);
772 error_setg(errp
, "Image not in VMDK format");
778 static const char *next_line(const char *s
)
789 static int vmdk_parse_extents(const char *desc
, BlockDriverState
*bs
,
790 const char *desc_file_path
, QDict
*options
,
802 BdrvChild
*extent_file
;
803 BDRVVmdkState
*s
= bs
->opaque
;
805 char extent_opt_prefix
[32];
806 Error
*local_err
= NULL
;
808 for (p
= desc
; *p
; p
= next_line(p
)) {
809 /* parse extent line in one of below formats:
811 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
812 * RW [size in sectors] SPARSE "file-name.vmdk"
813 * RW [size in sectors] VMFS "file-name.vmdk"
814 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
817 matches
= sscanf(p
, "%10s %" SCNd64
" %10s \"%511[^\n\r\"]\" %" SCNd64
,
818 access
, §ors
, type
, fname
, &flat_offset
);
819 if (matches
< 4 || strcmp(access
, "RW")) {
821 } else if (!strcmp(type
, "FLAT")) {
822 if (matches
!= 5 || flat_offset
< 0) {
825 } else if (!strcmp(type
, "VMFS")) {
831 } else if (matches
!= 4) {
836 (strcmp(type
, "FLAT") && strcmp(type
, "SPARSE") &&
837 strcmp(type
, "VMFS") && strcmp(type
, "VMFSSPARSE")) ||
838 (strcmp(access
, "RW"))) {
842 if (!path_is_absolute(fname
) && !path_has_protocol(fname
) &&
845 error_setg(errp
, "Cannot use relative extent paths with VMDK "
846 "descriptor file '%s'", bs
->file
->bs
->filename
);
850 extent_path
= g_malloc0(PATH_MAX
);
851 path_combine(extent_path
, PATH_MAX
, desc_file_path
, fname
);
853 ret
= snprintf(extent_opt_prefix
, 32, "extents.%d", s
->num_extents
);
856 extent_file
= bdrv_open_child(extent_path
, options
, extent_opt_prefix
,
857 bs
, &child_file
, false, &local_err
);
860 error_propagate(errp
, local_err
);
864 /* save to extents array */
865 if (!strcmp(type
, "FLAT") || !strcmp(type
, "VMFS")) {
868 ret
= vmdk_add_extent(bs
, extent_file
, true, sectors
,
869 0, 0, 0, 0, 0, &extent
, errp
);
871 bdrv_unref_child(bs
, extent_file
);
874 extent
->flat_start_offset
= flat_offset
<< 9;
875 } else if (!strcmp(type
, "SPARSE") || !strcmp(type
, "VMFSSPARSE")) {
876 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
877 char *buf
= vmdk_read_desc(extent_file
->bs
, 0, errp
);
881 ret
= vmdk_open_sparse(bs
, extent_file
, bs
->open_flags
, buf
,
886 bdrv_unref_child(bs
, extent_file
);
889 extent
= &s
->extents
[s
->num_extents
- 1];
891 error_setg(errp
, "Unsupported extent type '%s'", type
);
892 bdrv_unref_child(bs
, extent_file
);
895 extent
->type
= g_strdup(type
);
902 if (np
[-1] == '\n') {
905 error_setg(errp
, "Invalid extent line: %.*s", (int)(np
- p
), p
);
909 static int vmdk_open_desc_file(BlockDriverState
*bs
, int flags
, char *buf
,
910 QDict
*options
, Error
**errp
)
914 BDRVVmdkState
*s
= bs
->opaque
;
916 if (vmdk_parse_description(buf
, "createType", ct
, sizeof(ct
))) {
917 error_setg(errp
, "invalid VMDK image descriptor");
921 if (strcmp(ct
, "monolithicFlat") &&
922 strcmp(ct
, "vmfs") &&
923 strcmp(ct
, "vmfsSparse") &&
924 strcmp(ct
, "twoGbMaxExtentSparse") &&
925 strcmp(ct
, "twoGbMaxExtentFlat")) {
926 error_setg(errp
, "Unsupported image type '%s'", ct
);
930 s
->create_type
= g_strdup(ct
);
932 ret
= vmdk_parse_extents(buf
, bs
, bs
->file
->bs
->exact_filename
, options
,
938 static int vmdk_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
943 BDRVVmdkState
*s
= bs
->opaque
;
946 buf
= vmdk_read_desc(bs
->file
->bs
, 0, errp
);
951 magic
= ldl_be_p(buf
);
955 ret
= vmdk_open_sparse(bs
, bs
->file
, flags
, buf
, options
,
957 s
->desc_offset
= 0x200;
960 ret
= vmdk_open_desc_file(bs
, flags
, buf
, options
, errp
);
967 /* try to open parent images, if exist */
968 ret
= vmdk_parent_open(bs
);
972 s
->cid
= vmdk_read_cid(bs
, 0);
973 s
->parent_cid
= vmdk_read_cid(bs
, 1);
974 qemu_co_mutex_init(&s
->lock
);
976 /* Disable migration when VMDK images are used */
977 error_setg(&s
->migration_blocker
, "The vmdk format used by node '%s' "
978 "does not support live migration",
979 bdrv_get_device_or_node_name(bs
));
980 migrate_add_blocker(s
->migration_blocker
);
986 g_free(s
->create_type
);
987 s
->create_type
= NULL
;
988 vmdk_free_extents(bs
);
993 static void vmdk_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
995 BDRVVmdkState
*s
= bs
->opaque
;
998 for (i
= 0; i
< s
->num_extents
; i
++) {
999 if (!s
->extents
[i
].flat
) {
1000 bs
->bl
.write_zeroes_alignment
=
1001 MAX(bs
->bl
.write_zeroes_alignment
,
1002 s
->extents
[i
].cluster_sectors
);
1010 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1011 * to the cluster at @cluster_sector_num.
1013 * If @skip_start_sector < @skip_end_sector, the relative range
1014 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1015 * it for call to write user data in the request.
1017 static int get_whole_cluster(BlockDriverState
*bs
,
1019 uint64_t cluster_sector_num
,
1020 uint64_t sector_num
,
1021 uint64_t skip_start_sector
,
1022 uint64_t skip_end_sector
)
1025 int64_t cluster_bytes
;
1026 uint8_t *whole_grain
;
1028 /* For COW, align request sector_num to cluster start */
1029 sector_num
= QEMU_ALIGN_DOWN(sector_num
, extent
->cluster_sectors
);
1030 cluster_bytes
= extent
->cluster_sectors
<< BDRV_SECTOR_BITS
;
1031 whole_grain
= qemu_blockalign(bs
, cluster_bytes
);
1034 memset(whole_grain
, 0, skip_start_sector
<< BDRV_SECTOR_BITS
);
1035 memset(whole_grain
+ (skip_end_sector
<< BDRV_SECTOR_BITS
), 0,
1036 cluster_bytes
- (skip_end_sector
<< BDRV_SECTOR_BITS
));
1039 assert(skip_end_sector
<= extent
->cluster_sectors
);
1040 /* we will be here if it's first write on non-exist grain(cluster).
1041 * try to read from parent image, if exist */
1042 if (bs
->backing
&& !vmdk_is_cid_valid(bs
)) {
1047 /* Read backing data before skip range */
1048 if (skip_start_sector
> 0) {
1050 ret
= bdrv_read(bs
->backing
->bs
, sector_num
,
1051 whole_grain
, skip_start_sector
);
1057 ret
= bdrv_write(extent
->file
->bs
, cluster_sector_num
, whole_grain
,
1064 /* Read backing data after skip range */
1065 if (skip_end_sector
< extent
->cluster_sectors
) {
1067 ret
= bdrv_read(bs
->backing
->bs
, sector_num
+ skip_end_sector
,
1068 whole_grain
+ (skip_end_sector
<< BDRV_SECTOR_BITS
),
1069 extent
->cluster_sectors
- skip_end_sector
);
1075 ret
= bdrv_write(extent
->file
->bs
, cluster_sector_num
+ skip_end_sector
,
1076 whole_grain
+ (skip_end_sector
<< BDRV_SECTOR_BITS
),
1077 extent
->cluster_sectors
- skip_end_sector
);
1085 qemu_vfree(whole_grain
);
1089 static int vmdk_L2update(VmdkExtent
*extent
, VmdkMetaData
*m_data
,
1092 offset
= cpu_to_le32(offset
);
1093 /* update L2 table */
1094 if (bdrv_pwrite_sync(
1096 ((int64_t)m_data
->l2_offset
* 512)
1097 + (m_data
->l2_index
* sizeof(offset
)),
1098 &offset
, sizeof(offset
)) < 0) {
1101 /* update backup L2 table */
1102 if (extent
->l1_backup_table_offset
!= 0) {
1103 m_data
->l2_offset
= extent
->l1_backup_table
[m_data
->l1_index
];
1104 if (bdrv_pwrite_sync(
1106 ((int64_t)m_data
->l2_offset
* 512)
1107 + (m_data
->l2_index
* sizeof(offset
)),
1108 &offset
, sizeof(offset
)) < 0) {
1112 if (m_data
->l2_cache_entry
) {
1113 *m_data
->l2_cache_entry
= offset
;
1120 * get_cluster_offset
1122 * Look up cluster offset in extent file by sector number, and store in
1125 * For flat extents, the start offset as parsed from the description file is
1128 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1129 * offset for a new cluster and update L2 cache. If there is a backing file,
1130 * COW is done before returning; otherwise, zeroes are written to the allocated
1131 * cluster. Both COW and zero writing skips the sector range
1132 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1133 * has new data to write there.
1135 * Returns: VMDK_OK if cluster exists and mapped in the image.
1136 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1137 * VMDK_ERROR if failed.
1139 static int get_cluster_offset(BlockDriverState
*bs
,
1141 VmdkMetaData
*m_data
,
1144 uint64_t *cluster_offset
,
1145 uint64_t skip_start_sector
,
1146 uint64_t skip_end_sector
)
1148 unsigned int l1_index
, l2_offset
, l2_index
;
1149 int min_index
, i
, j
;
1150 uint32_t min_count
, *l2_table
;
1151 bool zeroed
= false;
1153 int64_t cluster_sector
;
1159 *cluster_offset
= extent
->flat_start_offset
;
1163 offset
-= (extent
->end_sector
- extent
->sectors
) * SECTOR_SIZE
;
1164 l1_index
= (offset
>> 9) / extent
->l1_entry_sectors
;
1165 if (l1_index
>= extent
->l1_size
) {
1168 l2_offset
= extent
->l1_table
[l1_index
];
1170 return VMDK_UNALLOC
;
1172 for (i
= 0; i
< L2_CACHE_SIZE
; i
++) {
1173 if (l2_offset
== extent
->l2_cache_offsets
[i
]) {
1174 /* increment the hit count */
1175 if (++extent
->l2_cache_counts
[i
] == 0xffffffff) {
1176 for (j
= 0; j
< L2_CACHE_SIZE
; j
++) {
1177 extent
->l2_cache_counts
[j
] >>= 1;
1180 l2_table
= extent
->l2_cache
+ (i
* extent
->l2_size
);
1184 /* not found: load a new entry in the least used one */
1186 min_count
= 0xffffffff;
1187 for (i
= 0; i
< L2_CACHE_SIZE
; i
++) {
1188 if (extent
->l2_cache_counts
[i
] < min_count
) {
1189 min_count
= extent
->l2_cache_counts
[i
];
1193 l2_table
= extent
->l2_cache
+ (min_index
* extent
->l2_size
);
1196 (int64_t)l2_offset
* 512,
1198 extent
->l2_size
* sizeof(uint32_t)
1199 ) != extent
->l2_size
* sizeof(uint32_t)) {
1203 extent
->l2_cache_offsets
[min_index
] = l2_offset
;
1204 extent
->l2_cache_counts
[min_index
] = 1;
1206 l2_index
= ((offset
>> 9) / extent
->cluster_sectors
) % extent
->l2_size
;
1207 cluster_sector
= le32_to_cpu(l2_table
[l2_index
]);
1211 m_data
->l1_index
= l1_index
;
1212 m_data
->l2_index
= l2_index
;
1213 m_data
->l2_offset
= l2_offset
;
1214 m_data
->l2_cache_entry
= &l2_table
[l2_index
];
1216 if (extent
->has_zero_grain
&& cluster_sector
== VMDK_GTE_ZEROED
) {
1220 if (!cluster_sector
|| zeroed
) {
1222 return zeroed
? VMDK_ZEROED
: VMDK_UNALLOC
;
1225 cluster_sector
= extent
->next_cluster_sector
;
1226 extent
->next_cluster_sector
+= extent
->cluster_sectors
;
1228 /* First of all we write grain itself, to avoid race condition
1229 * that may to corrupt the image.
1230 * This problem may occur because of insufficient space on host disk
1231 * or inappropriate VM shutdown.
1233 ret
= get_whole_cluster(bs
, extent
,
1235 offset
>> BDRV_SECTOR_BITS
,
1236 skip_start_sector
, skip_end_sector
);
1241 *cluster_offset
= cluster_sector
<< BDRV_SECTOR_BITS
;
1245 static VmdkExtent
*find_extent(BDRVVmdkState
*s
,
1246 int64_t sector_num
, VmdkExtent
*start_hint
)
1248 VmdkExtent
*extent
= start_hint
;
1251 extent
= &s
->extents
[0];
1253 while (extent
< &s
->extents
[s
->num_extents
]) {
1254 if (sector_num
< extent
->end_sector
) {
1262 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent
*extent
,
1265 uint64_t index_in_cluster
, extent_begin_sector
, extent_relative_sector_num
;
1267 extent_begin_sector
= extent
->end_sector
- extent
->sectors
;
1268 extent_relative_sector_num
= sector_num
- extent_begin_sector
;
1269 index_in_cluster
= extent_relative_sector_num
% extent
->cluster_sectors
;
1270 return index_in_cluster
;
1273 static int64_t coroutine_fn
vmdk_co_get_block_status(BlockDriverState
*bs
,
1274 int64_t sector_num
, int nb_sectors
, int *pnum
, BlockDriverState
**file
)
1276 BDRVVmdkState
*s
= bs
->opaque
;
1277 int64_t index_in_cluster
, n
, ret
;
1281 extent
= find_extent(s
, sector_num
, NULL
);
1285 qemu_co_mutex_lock(&s
->lock
);
1286 ret
= get_cluster_offset(bs
, extent
, NULL
,
1287 sector_num
* 512, false, &offset
,
1289 qemu_co_mutex_unlock(&s
->lock
);
1291 index_in_cluster
= vmdk_find_index_in_cluster(extent
, sector_num
);
1300 ret
= BDRV_BLOCK_ZERO
;
1303 ret
= BDRV_BLOCK_DATA
;
1304 if (!extent
->compressed
) {
1305 ret
|= BDRV_BLOCK_OFFSET_VALID
;
1306 ret
|= (offset
+ (index_in_cluster
<< BDRV_SECTOR_BITS
))
1307 & BDRV_BLOCK_OFFSET_MASK
;
1309 *file
= extent
->file
->bs
;
1313 n
= extent
->cluster_sectors
- index_in_cluster
;
1314 if (n
> nb_sectors
) {
1321 static int vmdk_write_extent(VmdkExtent
*extent
, int64_t cluster_offset
,
1322 int64_t offset_in_cluster
, const uint8_t *buf
,
1323 int nb_sectors
, int64_t sector_num
)
1326 VmdkGrainMarker
*data
= NULL
;
1328 const uint8_t *write_buf
= buf
;
1329 int write_len
= nb_sectors
* 512;
1330 int64_t write_offset
;
1331 int64_t write_end_sector
;
1333 if (extent
->compressed
) {
1334 if (!extent
->has_marker
) {
1338 buf_len
= (extent
->cluster_sectors
<< 9) * 2;
1339 data
= g_malloc(buf_len
+ sizeof(VmdkGrainMarker
));
1340 if (compress(data
->data
, &buf_len
, buf
, nb_sectors
<< 9) != Z_OK
||
1345 data
->lba
= sector_num
;
1346 data
->size
= buf_len
;
1347 write_buf
= (uint8_t *)data
;
1348 write_len
= buf_len
+ sizeof(VmdkGrainMarker
);
1350 write_offset
= cluster_offset
+ offset_in_cluster
,
1351 ret
= bdrv_pwrite(extent
->file
->bs
, write_offset
, write_buf
, write_len
);
1353 write_end_sector
= DIV_ROUND_UP(write_offset
+ write_len
, BDRV_SECTOR_SIZE
);
1355 if (extent
->compressed
) {
1356 extent
->next_cluster_sector
= write_end_sector
;
1358 extent
->next_cluster_sector
= MAX(extent
->next_cluster_sector
,
1362 if (ret
!= write_len
) {
1363 ret
= ret
< 0 ? ret
: -EIO
;
1372 static int vmdk_read_extent(VmdkExtent
*extent
, int64_t cluster_offset
,
1373 int64_t offset_in_cluster
, uint8_t *buf
,
1377 int cluster_bytes
, buf_bytes
;
1378 uint8_t *cluster_buf
, *compressed_data
;
1379 uint8_t *uncomp_buf
;
1381 VmdkGrainMarker
*marker
;
1385 if (!extent
->compressed
) {
1386 ret
= bdrv_pread(extent
->file
->bs
,
1387 cluster_offset
+ offset_in_cluster
,
1388 buf
, nb_sectors
* 512);
1389 if (ret
== nb_sectors
* 512) {
1395 cluster_bytes
= extent
->cluster_sectors
* 512;
1396 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1397 buf_bytes
= cluster_bytes
* 2;
1398 cluster_buf
= g_malloc(buf_bytes
);
1399 uncomp_buf
= g_malloc(cluster_bytes
);
1400 ret
= bdrv_pread(extent
->file
->bs
,
1402 cluster_buf
, buf_bytes
);
1406 compressed_data
= cluster_buf
;
1407 buf_len
= cluster_bytes
;
1408 data_len
= cluster_bytes
;
1409 if (extent
->has_marker
) {
1410 marker
= (VmdkGrainMarker
*)cluster_buf
;
1411 compressed_data
= marker
->data
;
1412 data_len
= le32_to_cpu(marker
->size
);
1414 if (!data_len
|| data_len
> buf_bytes
) {
1418 ret
= uncompress(uncomp_buf
, &buf_len
, compressed_data
, data_len
);
1424 if (offset_in_cluster
< 0 ||
1425 offset_in_cluster
+ nb_sectors
* 512 > buf_len
) {
1429 memcpy(buf
, uncomp_buf
+ offset_in_cluster
, nb_sectors
* 512);
1434 g_free(cluster_buf
);
1438 static int vmdk_read(BlockDriverState
*bs
, int64_t sector_num
,
1439 uint8_t *buf
, int nb_sectors
)
1441 BDRVVmdkState
*s
= bs
->opaque
;
1443 uint64_t n
, index_in_cluster
;
1444 VmdkExtent
*extent
= NULL
;
1445 uint64_t cluster_offset
;
1447 while (nb_sectors
> 0) {
1448 extent
= find_extent(s
, sector_num
, extent
);
1452 ret
= get_cluster_offset(bs
, extent
, NULL
,
1453 sector_num
<< 9, false, &cluster_offset
,
1455 index_in_cluster
= vmdk_find_index_in_cluster(extent
, sector_num
);
1456 n
= extent
->cluster_sectors
- index_in_cluster
;
1457 if (n
> nb_sectors
) {
1460 if (ret
!= VMDK_OK
) {
1461 /* if not allocated, try to read from parent image, if exist */
1462 if (bs
->backing
&& ret
!= VMDK_ZEROED
) {
1463 if (!vmdk_is_cid_valid(bs
)) {
1466 ret
= bdrv_read(bs
->backing
->bs
, sector_num
, buf
, n
);
1471 memset(buf
, 0, 512 * n
);
1474 ret
= vmdk_read_extent(extent
,
1475 cluster_offset
, index_in_cluster
* 512,
1488 static coroutine_fn
int vmdk_co_read(BlockDriverState
*bs
, int64_t sector_num
,
1489 uint8_t *buf
, int nb_sectors
)
1492 BDRVVmdkState
*s
= bs
->opaque
;
1493 qemu_co_mutex_lock(&s
->lock
);
1494 ret
= vmdk_read(bs
, sector_num
, buf
, nb_sectors
);
1495 qemu_co_mutex_unlock(&s
->lock
);
1501 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1502 * if possible, otherwise return -ENOTSUP.
1503 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1504 * with each cluster. By dry run we can find if the zero write
1505 * is possible without modifying image data.
1507 * Returns: error code with 0 for success.
1509 static int vmdk_write(BlockDriverState
*bs
, int64_t sector_num
,
1510 const uint8_t *buf
, int nb_sectors
,
1511 bool zeroed
, bool zero_dry_run
)
1513 BDRVVmdkState
*s
= bs
->opaque
;
1514 VmdkExtent
*extent
= NULL
;
1516 int64_t index_in_cluster
, n
;
1517 uint64_t cluster_offset
;
1518 VmdkMetaData m_data
;
1520 if (sector_num
> bs
->total_sectors
) {
1521 error_report("Wrong offset: sector_num=0x%" PRIx64
1522 " total_sectors=0x%" PRIx64
,
1523 sector_num
, bs
->total_sectors
);
1527 while (nb_sectors
> 0) {
1528 extent
= find_extent(s
, sector_num
, extent
);
1532 index_in_cluster
= vmdk_find_index_in_cluster(extent
, sector_num
);
1533 n
= extent
->cluster_sectors
- index_in_cluster
;
1534 if (n
> nb_sectors
) {
1537 ret
= get_cluster_offset(bs
, extent
, &m_data
, sector_num
<< 9,
1538 !(extent
->compressed
|| zeroed
),
1540 index_in_cluster
, index_in_cluster
+ n
);
1541 if (extent
->compressed
) {
1542 if (ret
== VMDK_OK
) {
1543 /* Refuse write to allocated cluster for streamOptimized */
1544 error_report("Could not write to allocated cluster"
1545 " for streamOptimized");
1549 ret
= get_cluster_offset(bs
, extent
, &m_data
, sector_num
<< 9,
1550 true, &cluster_offset
, 0, 0);
1553 if (ret
== VMDK_ERROR
) {
1557 /* Do zeroed write, buf is ignored */
1558 if (extent
->has_zero_grain
&&
1559 index_in_cluster
== 0 &&
1560 n
>= extent
->cluster_sectors
) {
1561 n
= extent
->cluster_sectors
;
1562 if (!zero_dry_run
) {
1563 /* update L2 tables */
1564 if (vmdk_L2update(extent
, &m_data
, VMDK_GTE_ZEROED
)
1573 ret
= vmdk_write_extent(extent
,
1574 cluster_offset
, index_in_cluster
* 512,
1575 buf
, n
, sector_num
);
1580 /* update L2 tables */
1581 if (vmdk_L2update(extent
, &m_data
,
1582 cluster_offset
>> BDRV_SECTOR_BITS
)
1592 /* update CID on the first write every time the virtual disk is
1594 if (!s
->cid_updated
) {
1595 ret
= vmdk_write_cid(bs
, g_random_int());
1599 s
->cid_updated
= true;
1605 static coroutine_fn
int vmdk_co_write(BlockDriverState
*bs
, int64_t sector_num
,
1606 const uint8_t *buf
, int nb_sectors
)
1609 BDRVVmdkState
*s
= bs
->opaque
;
1610 qemu_co_mutex_lock(&s
->lock
);
1611 ret
= vmdk_write(bs
, sector_num
, buf
, nb_sectors
, false, false);
1612 qemu_co_mutex_unlock(&s
->lock
);
1616 static int vmdk_write_compressed(BlockDriverState
*bs
,
1621 BDRVVmdkState
*s
= bs
->opaque
;
1622 if (s
->num_extents
== 1 && s
->extents
[0].compressed
) {
1623 return vmdk_write(bs
, sector_num
, buf
, nb_sectors
, false, false);
1629 static int coroutine_fn
vmdk_co_write_zeroes(BlockDriverState
*bs
,
1632 BdrvRequestFlags flags
)
1635 BDRVVmdkState
*s
= bs
->opaque
;
1636 qemu_co_mutex_lock(&s
->lock
);
1637 /* write zeroes could fail if sectors not aligned to cluster, test it with
1638 * dry_run == true before really updating image */
1639 ret
= vmdk_write(bs
, sector_num
, NULL
, nb_sectors
, true, true);
1641 ret
= vmdk_write(bs
, sector_num
, NULL
, nb_sectors
, true, false);
1643 qemu_co_mutex_unlock(&s
->lock
);
1647 static int vmdk_create_extent(const char *filename
, int64_t filesize
,
1648 bool flat
, bool compress
, bool zeroed_grain
,
1649 QemuOpts
*opts
, Error
**errp
)
1652 BlockBackend
*blk
= NULL
;
1654 Error
*local_err
= NULL
;
1655 uint32_t tmp
, magic
, grains
, gd_sectors
, gt_size
, gt_count
;
1656 uint32_t *gd_buf
= NULL
;
1659 ret
= bdrv_create_file(filename
, opts
, &local_err
);
1661 error_propagate(errp
, local_err
);
1665 blk
= blk_new_open(filename
, NULL
, NULL
,
1666 BDRV_O_RDWR
| BDRV_O_CACHE_WB
| BDRV_O_PROTOCOL
,
1669 error_propagate(errp
, local_err
);
1674 blk_set_allow_write_beyond_eof(blk
, true);
1677 ret
= blk_truncate(blk
, filesize
);
1679 error_setg_errno(errp
, -ret
, "Could not truncate file");
1683 magic
= cpu_to_be32(VMDK4_MAGIC
);
1684 memset(&header
, 0, sizeof(header
));
1687 } else if (zeroed_grain
) {
1692 header
.flags
= VMDK4_FLAG_RGD
| VMDK4_FLAG_NL_DETECT
1693 | (compress
? VMDK4_FLAG_COMPRESS
| VMDK4_FLAG_MARKER
: 0)
1694 | (zeroed_grain
? VMDK4_FLAG_ZERO_GRAIN
: 0);
1695 header
.compressAlgorithm
= compress
? VMDK4_COMPRESSION_DEFLATE
: 0;
1696 header
.capacity
= filesize
/ BDRV_SECTOR_SIZE
;
1697 header
.granularity
= 128;
1698 header
.num_gtes_per_gt
= BDRV_SECTOR_SIZE
;
1700 grains
= DIV_ROUND_UP(filesize
/ BDRV_SECTOR_SIZE
, header
.granularity
);
1701 gt_size
= DIV_ROUND_UP(header
.num_gtes_per_gt
* sizeof(uint32_t),
1703 gt_count
= DIV_ROUND_UP(grains
, header
.num_gtes_per_gt
);
1704 gd_sectors
= DIV_ROUND_UP(gt_count
* sizeof(uint32_t), BDRV_SECTOR_SIZE
);
1706 header
.desc_offset
= 1;
1707 header
.desc_size
= 20;
1708 header
.rgd_offset
= header
.desc_offset
+ header
.desc_size
;
1709 header
.gd_offset
= header
.rgd_offset
+ gd_sectors
+ (gt_size
* gt_count
);
1710 header
.grain_offset
=
1711 ROUND_UP(header
.gd_offset
+ gd_sectors
+ (gt_size
* gt_count
),
1712 header
.granularity
);
1713 /* swap endianness for all header fields */
1714 header
.version
= cpu_to_le32(header
.version
);
1715 header
.flags
= cpu_to_le32(header
.flags
);
1716 header
.capacity
= cpu_to_le64(header
.capacity
);
1717 header
.granularity
= cpu_to_le64(header
.granularity
);
1718 header
.num_gtes_per_gt
= cpu_to_le32(header
.num_gtes_per_gt
);
1719 header
.desc_offset
= cpu_to_le64(header
.desc_offset
);
1720 header
.desc_size
= cpu_to_le64(header
.desc_size
);
1721 header
.rgd_offset
= cpu_to_le64(header
.rgd_offset
);
1722 header
.gd_offset
= cpu_to_le64(header
.gd_offset
);
1723 header
.grain_offset
= cpu_to_le64(header
.grain_offset
);
1724 header
.compressAlgorithm
= cpu_to_le16(header
.compressAlgorithm
);
1726 header
.check_bytes
[0] = 0xa;
1727 header
.check_bytes
[1] = 0x20;
1728 header
.check_bytes
[2] = 0xd;
1729 header
.check_bytes
[3] = 0xa;
1731 /* write all the data */
1732 ret
= blk_pwrite(blk
, 0, &magic
, sizeof(magic
));
1734 error_setg(errp
, QERR_IO_ERROR
);
1737 ret
= blk_pwrite(blk
, sizeof(magic
), &header
, sizeof(header
));
1739 error_setg(errp
, QERR_IO_ERROR
);
1743 ret
= blk_truncate(blk
, le64_to_cpu(header
.grain_offset
) << 9);
1745 error_setg_errno(errp
, -ret
, "Could not truncate file");
1749 /* write grain directory */
1750 gd_buf_size
= gd_sectors
* BDRV_SECTOR_SIZE
;
1751 gd_buf
= g_malloc0(gd_buf_size
);
1752 for (i
= 0, tmp
= le64_to_cpu(header
.rgd_offset
) + gd_sectors
;
1753 i
< gt_count
; i
++, tmp
+= gt_size
) {
1754 gd_buf
[i
] = cpu_to_le32(tmp
);
1756 ret
= blk_pwrite(blk
, le64_to_cpu(header
.rgd_offset
) * BDRV_SECTOR_SIZE
,
1757 gd_buf
, gd_buf_size
);
1759 error_setg(errp
, QERR_IO_ERROR
);
1763 /* write backup grain directory */
1764 for (i
= 0, tmp
= le64_to_cpu(header
.gd_offset
) + gd_sectors
;
1765 i
< gt_count
; i
++, tmp
+= gt_size
) {
1766 gd_buf
[i
] = cpu_to_le32(tmp
);
1768 ret
= blk_pwrite(blk
, le64_to_cpu(header
.gd_offset
) * BDRV_SECTOR_SIZE
,
1769 gd_buf
, gd_buf_size
);
1771 error_setg(errp
, QERR_IO_ERROR
);
1784 static int filename_decompose(const char *filename
, char *path
, char *prefix
,
1785 char *postfix
, size_t buf_len
, Error
**errp
)
1789 if (filename
== NULL
|| !strlen(filename
)) {
1790 error_setg(errp
, "No filename provided");
1793 p
= strrchr(filename
, '/');
1795 p
= strrchr(filename
, '\\');
1798 p
= strrchr(filename
, ':');
1802 if (p
- filename
>= buf_len
) {
1805 pstrcpy(path
, p
- filename
+ 1, filename
);
1810 q
= strrchr(p
, '.');
1812 pstrcpy(prefix
, buf_len
, p
);
1815 if (q
- p
>= buf_len
) {
1818 pstrcpy(prefix
, q
- p
+ 1, p
);
1819 pstrcpy(postfix
, buf_len
, q
);
1824 static int vmdk_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
1827 BlockBackend
*new_blk
= NULL
;
1828 Error
*local_err
= NULL
;
1830 int64_t total_size
= 0, filesize
;
1831 char *adapter_type
= NULL
;
1832 char *backing_file
= NULL
;
1836 bool flat
, split
, compress
;
1837 GString
*ext_desc_lines
;
1838 char *path
= g_malloc0(PATH_MAX
);
1839 char *prefix
= g_malloc0(PATH_MAX
);
1840 char *postfix
= g_malloc0(PATH_MAX
);
1841 char *desc_line
= g_malloc0(BUF_SIZE
);
1842 char *ext_filename
= g_malloc0(PATH_MAX
);
1843 char *desc_filename
= g_malloc0(PATH_MAX
);
1844 const int64_t split_size
= 0x80000000; /* VMDK has constant split size */
1845 const char *desc_extent_line
;
1846 char *parent_desc_line
= g_malloc0(BUF_SIZE
);
1847 uint32_t parent_cid
= 0xffffffff;
1848 uint32_t number_heads
= 16;
1849 bool zeroed_grain
= false;
1850 uint32_t desc_offset
= 0, desc_len
;
1851 const char desc_template
[] =
1852 "# Disk DescriptorFile\n"
1855 "parentCID=%" PRIx32
"\n"
1856 "createType=\"%s\"\n"
1859 "# Extent description\n"
1862 "# The Disk Data Base\n"
1865 "ddb.virtualHWVersion = \"%d\"\n"
1866 "ddb.geometry.cylinders = \"%" PRId64
"\"\n"
1867 "ddb.geometry.heads = \"%" PRIu32
"\"\n"
1868 "ddb.geometry.sectors = \"63\"\n"
1869 "ddb.adapterType = \"%s\"\n";
1871 ext_desc_lines
= g_string_new(NULL
);
1873 if (filename_decompose(filename
, path
, prefix
, postfix
, PATH_MAX
, errp
)) {
1877 /* Read out options */
1878 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
1880 adapter_type
= qemu_opt_get_del(opts
, BLOCK_OPT_ADAPTER_TYPE
);
1881 backing_file
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
1882 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_COMPAT6
, false)) {
1883 flags
|= BLOCK_FLAG_COMPAT6
;
1885 fmt
= qemu_opt_get_del(opts
, BLOCK_OPT_SUBFMT
);
1886 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_ZEROED_GRAIN
, false)) {
1887 zeroed_grain
= true;
1890 if (!adapter_type
) {
1891 adapter_type
= g_strdup("ide");
1892 } else if (strcmp(adapter_type
, "ide") &&
1893 strcmp(adapter_type
, "buslogic") &&
1894 strcmp(adapter_type
, "lsilogic") &&
1895 strcmp(adapter_type
, "legacyESX")) {
1896 error_setg(errp
, "Unknown adapter type: '%s'", adapter_type
);
1900 if (strcmp(adapter_type
, "ide") != 0) {
1901 /* that's the number of heads with which vmware operates when
1902 creating, exporting, etc. vmdk files with a non-ide adapter type */
1906 /* Default format to monolithicSparse */
1907 fmt
= g_strdup("monolithicSparse");
1908 } else if (strcmp(fmt
, "monolithicFlat") &&
1909 strcmp(fmt
, "monolithicSparse") &&
1910 strcmp(fmt
, "twoGbMaxExtentSparse") &&
1911 strcmp(fmt
, "twoGbMaxExtentFlat") &&
1912 strcmp(fmt
, "streamOptimized")) {
1913 error_setg(errp
, "Unknown subformat: '%s'", fmt
);
1917 split
= !(strcmp(fmt
, "twoGbMaxExtentFlat") &&
1918 strcmp(fmt
, "twoGbMaxExtentSparse"));
1919 flat
= !(strcmp(fmt
, "monolithicFlat") &&
1920 strcmp(fmt
, "twoGbMaxExtentFlat"));
1921 compress
= !strcmp(fmt
, "streamOptimized");
1923 desc_extent_line
= "RW %" PRId64
" FLAT \"%s\" 0\n";
1925 desc_extent_line
= "RW %" PRId64
" SPARSE \"%s\"\n";
1927 if (flat
&& backing_file
) {
1928 error_setg(errp
, "Flat image can't have backing file");
1932 if (flat
&& zeroed_grain
) {
1933 error_setg(errp
, "Flat image can't enable zeroed grain");
1939 char *full_backing
= g_new0(char, PATH_MAX
);
1940 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
1941 full_backing
, PATH_MAX
,
1944 g_free(full_backing
);
1945 error_propagate(errp
, local_err
);
1950 blk
= blk_new_open(full_backing
, NULL
, NULL
,
1951 BDRV_O_NO_BACKING
| BDRV_O_CACHE_WB
, errp
);
1952 g_free(full_backing
);
1957 if (strcmp(blk_bs(blk
)->drv
->format_name
, "vmdk")) {
1962 parent_cid
= vmdk_read_cid(blk_bs(blk
), 0);
1964 snprintf(parent_desc_line
, BUF_SIZE
,
1965 "parentFileNameHint=\"%s\"", backing_file
);
1968 /* Create extents */
1969 filesize
= total_size
;
1970 while (filesize
> 0) {
1971 int64_t size
= filesize
;
1973 if (split
&& size
> split_size
) {
1977 snprintf(desc_filename
, PATH_MAX
, "%s-%c%03d%s",
1978 prefix
, flat
? 'f' : 's', ++idx
, postfix
);
1980 snprintf(desc_filename
, PATH_MAX
, "%s-flat%s", prefix
, postfix
);
1982 snprintf(desc_filename
, PATH_MAX
, "%s%s", prefix
, postfix
);
1984 snprintf(ext_filename
, PATH_MAX
, "%s%s", path
, desc_filename
);
1986 if (vmdk_create_extent(ext_filename
, size
,
1987 flat
, compress
, zeroed_grain
, opts
, errp
)) {
1993 /* Format description line */
1994 snprintf(desc_line
, BUF_SIZE
,
1995 desc_extent_line
, size
/ BDRV_SECTOR_SIZE
, desc_filename
);
1996 g_string_append(ext_desc_lines
, desc_line
);
1998 /* generate descriptor file */
1999 desc
= g_strdup_printf(desc_template
,
2004 ext_desc_lines
->str
,
2005 (flags
& BLOCK_FLAG_COMPAT6
? 6 : 4),
2007 (int64_t)(63 * number_heads
* BDRV_SECTOR_SIZE
),
2010 desc_len
= strlen(desc
);
2011 /* the descriptor offset = 0x200 */
2012 if (!split
&& !flat
) {
2013 desc_offset
= 0x200;
2015 ret
= bdrv_create_file(filename
, opts
, &local_err
);
2017 error_propagate(errp
, local_err
);
2022 new_blk
= blk_new_open(filename
, NULL
, NULL
,
2023 BDRV_O_RDWR
| BDRV_O_CACHE_WB
| BDRV_O_PROTOCOL
,
2025 if (new_blk
== NULL
) {
2026 error_propagate(errp
, local_err
);
2031 blk_set_allow_write_beyond_eof(new_blk
, true);
2033 ret
= blk_pwrite(new_blk
, desc_offset
, desc
, desc_len
);
2035 error_setg_errno(errp
, -ret
, "Could not write description");
2038 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2039 * for description file */
2040 if (desc_offset
== 0) {
2041 ret
= blk_truncate(new_blk
, desc_len
);
2043 error_setg_errno(errp
, -ret
, "Could not truncate file");
2050 g_free(adapter_type
);
2051 g_free(backing_file
);
2058 g_free(ext_filename
);
2059 g_free(desc_filename
);
2060 g_free(parent_desc_line
);
2061 g_string_free(ext_desc_lines
, true);
2065 static void vmdk_close(BlockDriverState
*bs
)
2067 BDRVVmdkState
*s
= bs
->opaque
;
2069 vmdk_free_extents(bs
);
2070 g_free(s
->create_type
);
2072 migrate_del_blocker(s
->migration_blocker
);
2073 error_free(s
->migration_blocker
);
2076 static coroutine_fn
int vmdk_co_flush(BlockDriverState
*bs
)
2078 BDRVVmdkState
*s
= bs
->opaque
;
2082 for (i
= 0; i
< s
->num_extents
; i
++) {
2083 err
= bdrv_co_flush(s
->extents
[i
].file
->bs
);
2091 static int64_t vmdk_get_allocated_file_size(BlockDriverState
*bs
)
2096 BDRVVmdkState
*s
= bs
->opaque
;
2098 ret
= bdrv_get_allocated_file_size(bs
->file
->bs
);
2102 for (i
= 0; i
< s
->num_extents
; i
++) {
2103 if (s
->extents
[i
].file
== bs
->file
) {
2106 r
= bdrv_get_allocated_file_size(s
->extents
[i
].file
->bs
);
2115 static int vmdk_has_zero_init(BlockDriverState
*bs
)
2118 BDRVVmdkState
*s
= bs
->opaque
;
2120 /* If has a flat extent and its underlying storage doesn't have zero init,
2122 for (i
= 0; i
< s
->num_extents
; i
++) {
2123 if (s
->extents
[i
].flat
) {
2124 if (!bdrv_has_zero_init(s
->extents
[i
].file
->bs
)) {
2132 static ImageInfo
*vmdk_get_extent_info(VmdkExtent
*extent
)
2134 ImageInfo
*info
= g_new0(ImageInfo
, 1);
2136 *info
= (ImageInfo
){
2137 .filename
= g_strdup(extent
->file
->bs
->filename
),
2138 .format
= g_strdup(extent
->type
),
2139 .virtual_size
= extent
->sectors
* BDRV_SECTOR_SIZE
,
2140 .compressed
= extent
->compressed
,
2141 .has_compressed
= extent
->compressed
,
2142 .cluster_size
= extent
->cluster_sectors
* BDRV_SECTOR_SIZE
,
2143 .has_cluster_size
= !extent
->flat
,
2149 static int vmdk_check(BlockDriverState
*bs
, BdrvCheckResult
*result
,
2152 BDRVVmdkState
*s
= bs
->opaque
;
2153 VmdkExtent
*extent
= NULL
;
2154 int64_t sector_num
= 0;
2155 int64_t total_sectors
= bdrv_nb_sectors(bs
);
2157 uint64_t cluster_offset
;
2164 if (sector_num
>= total_sectors
) {
2167 extent
= find_extent(s
, sector_num
, extent
);
2170 "ERROR: could not find extent for sector %" PRId64
"\n",
2174 ret
= get_cluster_offset(bs
, extent
, NULL
,
2175 sector_num
<< BDRV_SECTOR_BITS
,
2176 false, &cluster_offset
, 0, 0);
2177 if (ret
== VMDK_ERROR
) {
2179 "ERROR: could not get cluster_offset for sector %"
2180 PRId64
"\n", sector_num
);
2183 if (ret
== VMDK_OK
&&
2184 cluster_offset
>= bdrv_getlength(extent
->file
->bs
))
2187 "ERROR: cluster offset for sector %"
2188 PRId64
" points after EOF\n", sector_num
);
2191 sector_num
+= extent
->cluster_sectors
;
2194 result
->corruptions
++;
2198 static ImageInfoSpecific
*vmdk_get_specific_info(BlockDriverState
*bs
)
2201 BDRVVmdkState
*s
= bs
->opaque
;
2202 ImageInfoSpecific
*spec_info
= g_new0(ImageInfoSpecific
, 1);
2203 ImageInfoList
**next
;
2205 *spec_info
= (ImageInfoSpecific
){
2206 .type
= IMAGE_INFO_SPECIFIC_KIND_VMDK
,
2208 .vmdk
.data
= g_new0(ImageInfoSpecificVmdk
, 1),
2212 *spec_info
->u
.vmdk
.data
= (ImageInfoSpecificVmdk
) {
2213 .create_type
= g_strdup(s
->create_type
),
2215 .parent_cid
= s
->parent_cid
,
2218 next
= &spec_info
->u
.vmdk
.data
->extents
;
2219 for (i
= 0; i
< s
->num_extents
; i
++) {
2220 *next
= g_new0(ImageInfoList
, 1);
2221 (*next
)->value
= vmdk_get_extent_info(&s
->extents
[i
]);
2222 (*next
)->next
= NULL
;
2223 next
= &(*next
)->next
;
2229 static bool vmdk_extents_type_eq(const VmdkExtent
*a
, const VmdkExtent
*b
)
2231 return a
->flat
== b
->flat
&&
2232 a
->compressed
== b
->compressed
&&
2233 (a
->flat
|| a
->cluster_sectors
== b
->cluster_sectors
);
2236 static int vmdk_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2239 BDRVVmdkState
*s
= bs
->opaque
;
2240 assert(s
->num_extents
);
2242 /* See if we have multiple extents but they have different cases */
2243 for (i
= 1; i
< s
->num_extents
; i
++) {
2244 if (!vmdk_extents_type_eq(&s
->extents
[0], &s
->extents
[i
])) {
2248 bdi
->needs_compressed_writes
= s
->extents
[0].compressed
;
2249 if (!s
->extents
[0].flat
) {
2250 bdi
->cluster_size
= s
->extents
[0].cluster_sectors
<< BDRV_SECTOR_BITS
;
2255 static void vmdk_detach_aio_context(BlockDriverState
*bs
)
2257 BDRVVmdkState
*s
= bs
->opaque
;
2260 for (i
= 0; i
< s
->num_extents
; i
++) {
2261 bdrv_detach_aio_context(s
->extents
[i
].file
->bs
);
2265 static void vmdk_attach_aio_context(BlockDriverState
*bs
,
2266 AioContext
*new_context
)
2268 BDRVVmdkState
*s
= bs
->opaque
;
2271 for (i
= 0; i
< s
->num_extents
; i
++) {
2272 bdrv_attach_aio_context(s
->extents
[i
].file
->bs
, new_context
);
2276 static QemuOptsList vmdk_create_opts
= {
2277 .name
= "vmdk-create-opts",
2278 .head
= QTAILQ_HEAD_INITIALIZER(vmdk_create_opts
.head
),
2281 .name
= BLOCK_OPT_SIZE
,
2282 .type
= QEMU_OPT_SIZE
,
2283 .help
= "Virtual disk size"
2286 .name
= BLOCK_OPT_ADAPTER_TYPE
,
2287 .type
= QEMU_OPT_STRING
,
2288 .help
= "Virtual adapter type, can be one of "
2289 "ide (default), lsilogic, buslogic or legacyESX"
2292 .name
= BLOCK_OPT_BACKING_FILE
,
2293 .type
= QEMU_OPT_STRING
,
2294 .help
= "File name of a base image"
2297 .name
= BLOCK_OPT_COMPAT6
,
2298 .type
= QEMU_OPT_BOOL
,
2299 .help
= "VMDK version 6 image",
2300 .def_value_str
= "off"
2303 .name
= BLOCK_OPT_SUBFMT
,
2304 .type
= QEMU_OPT_STRING
,
2306 "VMDK flat extent format, can be one of "
2307 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2310 .name
= BLOCK_OPT_ZEROED_GRAIN
,
2311 .type
= QEMU_OPT_BOOL
,
2312 .help
= "Enable efficient zero writes "
2313 "using the zeroed-grain GTE feature"
2315 { /* end of list */ }
2319 static BlockDriver bdrv_vmdk
= {
2320 .format_name
= "vmdk",
2321 .instance_size
= sizeof(BDRVVmdkState
),
2322 .bdrv_probe
= vmdk_probe
,
2323 .bdrv_open
= vmdk_open
,
2324 .bdrv_check
= vmdk_check
,
2325 .bdrv_reopen_prepare
= vmdk_reopen_prepare
,
2326 .bdrv_read
= vmdk_co_read
,
2327 .bdrv_write
= vmdk_co_write
,
2328 .bdrv_write_compressed
= vmdk_write_compressed
,
2329 .bdrv_co_write_zeroes
= vmdk_co_write_zeroes
,
2330 .bdrv_close
= vmdk_close
,
2331 .bdrv_create
= vmdk_create
,
2332 .bdrv_co_flush_to_disk
= vmdk_co_flush
,
2333 .bdrv_co_get_block_status
= vmdk_co_get_block_status
,
2334 .bdrv_get_allocated_file_size
= vmdk_get_allocated_file_size
,
2335 .bdrv_has_zero_init
= vmdk_has_zero_init
,
2336 .bdrv_get_specific_info
= vmdk_get_specific_info
,
2337 .bdrv_refresh_limits
= vmdk_refresh_limits
,
2338 .bdrv_get_info
= vmdk_get_info
,
2339 .bdrv_detach_aio_context
= vmdk_detach_aio_context
,
2340 .bdrv_attach_aio_context
= vmdk_attach_aio_context
,
2342 .supports_backing
= true,
2343 .create_opts
= &vmdk_create_opts
,
2346 static void bdrv_vmdk_init(void)
2348 bdrv_register(&bdrv_vmdk
);
2351 block_init(bdrv_vmdk_init
);