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
;
313 if (!s
->cid_checked
&& bs
->backing
) {
314 BlockDriverState
*p_bs
= bs
->backing
->bs
;
316 cur_pcid
= vmdk_read_cid(p_bs
, 0);
317 if (s
->parent_cid
!= cur_pcid
) {
322 s
->cid_checked
= true;
327 /* We have nothing to do for VMDK reopen, stubs just return success */
328 static int vmdk_reopen_prepare(BDRVReopenState
*state
,
329 BlockReopenQueue
*queue
, Error
**errp
)
331 assert(state
!= NULL
);
332 assert(state
->bs
!= NULL
);
336 static int vmdk_parent_open(BlockDriverState
*bs
)
339 char desc
[DESC_SIZE
+ 1];
340 BDRVVmdkState
*s
= bs
->opaque
;
343 desc
[DESC_SIZE
] = '\0';
344 ret
= bdrv_pread(bs
->file
->bs
, s
->desc_offset
, desc
, DESC_SIZE
);
349 p_name
= strstr(desc
, "parentFileNameHint");
350 if (p_name
!= NULL
) {
353 p_name
+= sizeof("parentFileNameHint") + 1;
354 end_name
= strchr(p_name
, '\"');
355 if (end_name
== NULL
) {
358 if ((end_name
- p_name
) > sizeof(bs
->backing_file
) - 1) {
362 pstrcpy(bs
->backing_file
, end_name
- p_name
+ 1, p_name
);
368 /* Create and append extent to the extent array. Return the added VmdkExtent
369 * address. return NULL if allocation failed. */
370 static int vmdk_add_extent(BlockDriverState
*bs
,
371 BdrvChild
*file
, bool flat
, int64_t sectors
,
372 int64_t l1_offset
, int64_t l1_backup_offset
,
374 int l2_size
, uint64_t cluster_sectors
,
375 VmdkExtent
**new_extent
,
379 BDRVVmdkState
*s
= bs
->opaque
;
382 if (cluster_sectors
> 0x200000) {
383 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
384 error_setg(errp
, "Invalid granularity, image may be corrupt");
387 if (l1_size
> 512 * 1024 * 1024) {
388 /* Although with big capacity and small l1_entry_sectors, we can get a
389 * big l1_size, we don't want unbounded value to allocate the table.
390 * Limit it to 512M, which is 16PB for default cluster and L2 table
392 error_setg(errp
, "L1 size too big");
396 nb_sectors
= bdrv_nb_sectors(file
->bs
);
397 if (nb_sectors
< 0) {
401 s
->extents
= g_renew(VmdkExtent
, s
->extents
, s
->num_extents
+ 1);
402 extent
= &s
->extents
[s
->num_extents
];
405 memset(extent
, 0, sizeof(VmdkExtent
));
408 extent
->sectors
= sectors
;
409 extent
->l1_table_offset
= l1_offset
;
410 extent
->l1_backup_table_offset
= l1_backup_offset
;
411 extent
->l1_size
= l1_size
;
412 extent
->l1_entry_sectors
= l2_size
* cluster_sectors
;
413 extent
->l2_size
= l2_size
;
414 extent
->cluster_sectors
= flat
? sectors
: cluster_sectors
;
415 extent
->next_cluster_sector
= ROUND_UP(nb_sectors
, cluster_sectors
);
417 if (s
->num_extents
> 1) {
418 extent
->end_sector
= (*(extent
- 1)).end_sector
+ extent
->sectors
;
420 extent
->end_sector
= extent
->sectors
;
422 bs
->total_sectors
= extent
->end_sector
;
424 *new_extent
= extent
;
429 static int vmdk_init_tables(BlockDriverState
*bs
, VmdkExtent
*extent
,
436 /* read the L1 table */
437 l1_size
= extent
->l1_size
* sizeof(uint32_t);
438 extent
->l1_table
= g_try_malloc(l1_size
);
439 if (l1_size
&& extent
->l1_table
== NULL
) {
443 ret
= bdrv_pread(extent
->file
->bs
,
444 extent
->l1_table_offset
,
448 error_setg_errno(errp
, -ret
,
449 "Could not read l1 table from extent '%s'",
450 extent
->file
->bs
->filename
);
453 for (i
= 0; i
< extent
->l1_size
; i
++) {
454 le32_to_cpus(&extent
->l1_table
[i
]);
457 if (extent
->l1_backup_table_offset
) {
458 extent
->l1_backup_table
= g_try_malloc(l1_size
);
459 if (l1_size
&& extent
->l1_backup_table
== NULL
) {
463 ret
= bdrv_pread(extent
->file
->bs
,
464 extent
->l1_backup_table_offset
,
465 extent
->l1_backup_table
,
468 error_setg_errno(errp
, -ret
,
469 "Could not read l1 backup table from extent '%s'",
470 extent
->file
->bs
->filename
);
473 for (i
= 0; i
< extent
->l1_size
; i
++) {
474 le32_to_cpus(&extent
->l1_backup_table
[i
]);
479 g_new(uint32_t, extent
->l2_size
* L2_CACHE_SIZE
);
482 g_free(extent
->l1_backup_table
);
484 g_free(extent
->l1_table
);
488 static int vmdk_open_vmfs_sparse(BlockDriverState
*bs
,
490 int flags
, Error
**errp
)
497 ret
= bdrv_pread(file
->bs
, sizeof(magic
), &header
, sizeof(header
));
499 error_setg_errno(errp
, -ret
,
500 "Could not read header from file '%s'",
504 ret
= vmdk_add_extent(bs
, file
, false,
505 le32_to_cpu(header
.disk_sectors
),
506 (int64_t)le32_to_cpu(header
.l1dir_offset
) << 9,
508 le32_to_cpu(header
.l1dir_size
),
510 le32_to_cpu(header
.granularity
),
516 ret
= vmdk_init_tables(bs
, extent
, errp
);
518 /* free extent allocated by vmdk_add_extent */
519 vmdk_free_last_extent(bs
);
524 static int vmdk_open_desc_file(BlockDriverState
*bs
, int flags
, char *buf
,
525 QDict
*options
, Error
**errp
);
527 static char *vmdk_read_desc(BlockDriverState
*file
, uint64_t desc_offset
,
534 size
= bdrv_getlength(file
);
536 error_setg_errno(errp
, -size
, "Could not access file");
541 /* Both descriptor file and sparse image must be much larger than 4
542 * bytes, also callers of vmdk_read_desc want to compare the first 4
543 * bytes with VMDK4_MAGIC, let's error out if less is read. */
544 error_setg(errp
, "File is too small, not a valid image");
548 size
= MIN(size
, (1 << 20) - 1); /* avoid unbounded allocation */
549 buf
= g_malloc(size
+ 1);
551 ret
= bdrv_pread(file
, desc_offset
, buf
, size
);
553 error_setg_errno(errp
, -ret
, "Could not read from file");
562 static int vmdk_open_vmdk4(BlockDriverState
*bs
,
564 int flags
, QDict
*options
, Error
**errp
)
568 uint32_t l1_size
, l1_entry_sectors
;
571 BDRVVmdkState
*s
= bs
->opaque
;
572 int64_t l1_backup_offset
= 0;
574 ret
= bdrv_pread(file
->bs
, sizeof(magic
), &header
, sizeof(header
));
576 error_setg_errno(errp
, -ret
,
577 "Could not read header from file '%s'",
581 if (header
.capacity
== 0) {
582 uint64_t desc_offset
= le64_to_cpu(header
.desc_offset
);
584 char *buf
= vmdk_read_desc(file
->bs
, desc_offset
<< 9, errp
);
588 ret
= vmdk_open_desc_file(bs
, flags
, buf
, options
, errp
);
594 if (!s
->create_type
) {
595 s
->create_type
= g_strdup("monolithicSparse");
598 if (le64_to_cpu(header
.gd_offset
) == VMDK4_GD_AT_END
) {
600 * The footer takes precedence over the header, so read it in. The
601 * footer starts at offset -1024 from the end: One sector for the
602 * footer, and another one for the end-of-stream marker.
609 uint8_t pad
[512 - 16];
610 } QEMU_PACKED footer_marker
;
614 uint8_t pad
[512 - 4 - sizeof(VMDK4Header
)];
620 uint8_t pad
[512 - 16];
621 } QEMU_PACKED eos_marker
;
622 } QEMU_PACKED footer
;
624 ret
= bdrv_pread(file
->bs
,
625 bs
->file
->bs
->total_sectors
* 512 - 1536,
626 &footer
, sizeof(footer
));
628 error_setg_errno(errp
, -ret
, "Failed to read footer");
632 /* Some sanity checks for the footer */
633 if (be32_to_cpu(footer
.magic
) != VMDK4_MAGIC
||
634 le32_to_cpu(footer
.footer_marker
.size
) != 0 ||
635 le32_to_cpu(footer
.footer_marker
.type
) != MARKER_FOOTER
||
636 le64_to_cpu(footer
.eos_marker
.val
) != 0 ||
637 le32_to_cpu(footer
.eos_marker
.size
) != 0 ||
638 le32_to_cpu(footer
.eos_marker
.type
) != MARKER_END_OF_STREAM
)
640 error_setg(errp
, "Invalid footer");
644 header
= footer
.header
;
647 if (le32_to_cpu(header
.version
) > 3) {
649 snprintf(buf
, sizeof(buf
), "VMDK version %" PRId32
,
650 le32_to_cpu(header
.version
));
651 error_setg(errp
, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE
,
652 bdrv_get_device_or_node_name(bs
), "vmdk", buf
);
654 } else if (le32_to_cpu(header
.version
) == 3 && (flags
& BDRV_O_RDWR
)) {
655 /* VMware KB 2064959 explains that version 3 added support for
656 * persistent changed block tracking (CBT), and backup software can
657 * read it as version=1 if it doesn't care about the changed area
658 * information. So we are safe to enable read only. */
659 error_setg(errp
, "VMDK version 3 must be read only");
663 if (le32_to_cpu(header
.num_gtes_per_gt
) > 512) {
664 error_setg(errp
, "L2 table size too big");
668 l1_entry_sectors
= le32_to_cpu(header
.num_gtes_per_gt
)
669 * le64_to_cpu(header
.granularity
);
670 if (l1_entry_sectors
== 0) {
671 error_setg(errp
, "L1 entry size is invalid");
674 l1_size
= (le64_to_cpu(header
.capacity
) + l1_entry_sectors
- 1)
676 if (le32_to_cpu(header
.flags
) & VMDK4_FLAG_RGD
) {
677 l1_backup_offset
= le64_to_cpu(header
.rgd_offset
) << 9;
679 if (bdrv_nb_sectors(file
->bs
) < le64_to_cpu(header
.grain_offset
)) {
680 error_setg(errp
, "File truncated, expecting at least %" PRId64
" bytes",
681 (int64_t)(le64_to_cpu(header
.grain_offset
)
682 * BDRV_SECTOR_SIZE
));
686 ret
= vmdk_add_extent(bs
, file
, false,
687 le64_to_cpu(header
.capacity
),
688 le64_to_cpu(header
.gd_offset
) << 9,
691 le32_to_cpu(header
.num_gtes_per_gt
),
692 le64_to_cpu(header
.granularity
),
699 le16_to_cpu(header
.compressAlgorithm
) == VMDK4_COMPRESSION_DEFLATE
;
700 if (extent
->compressed
) {
701 g_free(s
->create_type
);
702 s
->create_type
= g_strdup("streamOptimized");
704 extent
->has_marker
= le32_to_cpu(header
.flags
) & VMDK4_FLAG_MARKER
;
705 extent
->version
= le32_to_cpu(header
.version
);
706 extent
->has_zero_grain
= le32_to_cpu(header
.flags
) & VMDK4_FLAG_ZERO_GRAIN
;
707 ret
= vmdk_init_tables(bs
, extent
, errp
);
709 /* free extent allocated by vmdk_add_extent */
710 vmdk_free_last_extent(bs
);
715 /* find an option value out of descriptor file */
716 static int vmdk_parse_description(const char *desc
, const char *opt_name
,
717 char *buf
, int buf_size
)
719 char *opt_pos
, *opt_end
;
720 const char *end
= desc
+ strlen(desc
);
722 opt_pos
= strstr(desc
, opt_name
);
726 /* Skip "=\"" following opt_name */
727 opt_pos
+= strlen(opt_name
) + 2;
728 if (opt_pos
>= end
) {
732 while (opt_end
< end
&& *opt_end
!= '"') {
735 if (opt_end
== end
|| buf_size
< opt_end
- opt_pos
+ 1) {
738 pstrcpy(buf
, opt_end
- opt_pos
+ 1, opt_pos
);
742 /* Open an extent file and append to bs array */
743 static int vmdk_open_sparse(BlockDriverState
*bs
, BdrvChild
*file
, int flags
,
744 char *buf
, QDict
*options
, Error
**errp
)
748 magic
= ldl_be_p(buf
);
751 return vmdk_open_vmfs_sparse(bs
, file
, flags
, errp
);
754 return vmdk_open_vmdk4(bs
, file
, flags
, options
, errp
);
757 error_setg(errp
, "Image not in VMDK format");
763 static const char *next_line(const char *s
)
774 static int vmdk_parse_extents(const char *desc
, BlockDriverState
*bs
,
775 const char *desc_file_path
, QDict
*options
,
787 BdrvChild
*extent_file
;
788 BDRVVmdkState
*s
= bs
->opaque
;
790 char extent_opt_prefix
[32];
791 Error
*local_err
= NULL
;
793 for (p
= desc
; *p
; p
= next_line(p
)) {
794 /* parse extent line in one of below formats:
796 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
797 * RW [size in sectors] SPARSE "file-name.vmdk"
798 * RW [size in sectors] VMFS "file-name.vmdk"
799 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
802 matches
= sscanf(p
, "%10s %" SCNd64
" %10s \"%511[^\n\r\"]\" %" SCNd64
,
803 access
, §ors
, type
, fname
, &flat_offset
);
804 if (matches
< 4 || strcmp(access
, "RW")) {
806 } else if (!strcmp(type
, "FLAT")) {
807 if (matches
!= 5 || flat_offset
< 0) {
810 } else if (!strcmp(type
, "VMFS")) {
816 } else if (matches
!= 4) {
821 (strcmp(type
, "FLAT") && strcmp(type
, "SPARSE") &&
822 strcmp(type
, "VMFS") && strcmp(type
, "VMFSSPARSE")) ||
823 (strcmp(access
, "RW"))) {
827 if (!path_is_absolute(fname
) && !path_has_protocol(fname
) &&
830 error_setg(errp
, "Cannot use relative extent paths with VMDK "
831 "descriptor file '%s'", bs
->file
->bs
->filename
);
835 extent_path
= g_malloc0(PATH_MAX
);
836 path_combine(extent_path
, PATH_MAX
, desc_file_path
, fname
);
838 ret
= snprintf(extent_opt_prefix
, 32, "extents.%d", s
->num_extents
);
841 extent_file
= bdrv_open_child(extent_path
, options
, extent_opt_prefix
,
842 bs
, &child_file
, false, &local_err
);
845 error_propagate(errp
, local_err
);
849 /* save to extents array */
850 if (!strcmp(type
, "FLAT") || !strcmp(type
, "VMFS")) {
853 ret
= vmdk_add_extent(bs
, extent_file
, true, sectors
,
854 0, 0, 0, 0, 0, &extent
, errp
);
856 bdrv_unref_child(bs
, extent_file
);
859 extent
->flat_start_offset
= flat_offset
<< 9;
860 } else if (!strcmp(type
, "SPARSE") || !strcmp(type
, "VMFSSPARSE")) {
861 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
862 char *buf
= vmdk_read_desc(extent_file
->bs
, 0, errp
);
866 ret
= vmdk_open_sparse(bs
, extent_file
, bs
->open_flags
, buf
,
871 bdrv_unref_child(bs
, extent_file
);
874 extent
= &s
->extents
[s
->num_extents
- 1];
876 error_setg(errp
, "Unsupported extent type '%s'", type
);
877 bdrv_unref_child(bs
, extent_file
);
880 extent
->type
= g_strdup(type
);
887 if (np
[-1] == '\n') {
890 error_setg(errp
, "Invalid extent line: %.*s", (int)(np
- p
), p
);
894 static int vmdk_open_desc_file(BlockDriverState
*bs
, int flags
, char *buf
,
895 QDict
*options
, Error
**errp
)
899 BDRVVmdkState
*s
= bs
->opaque
;
901 if (vmdk_parse_description(buf
, "createType", ct
, sizeof(ct
))) {
902 error_setg(errp
, "invalid VMDK image descriptor");
906 if (strcmp(ct
, "monolithicFlat") &&
907 strcmp(ct
, "vmfs") &&
908 strcmp(ct
, "vmfsSparse") &&
909 strcmp(ct
, "twoGbMaxExtentSparse") &&
910 strcmp(ct
, "twoGbMaxExtentFlat")) {
911 error_setg(errp
, "Unsupported image type '%s'", ct
);
915 s
->create_type
= g_strdup(ct
);
917 ret
= vmdk_parse_extents(buf
, bs
, bs
->file
->bs
->exact_filename
, options
,
923 static int vmdk_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
928 BDRVVmdkState
*s
= bs
->opaque
;
931 buf
= vmdk_read_desc(bs
->file
->bs
, 0, errp
);
936 magic
= ldl_be_p(buf
);
940 ret
= vmdk_open_sparse(bs
, bs
->file
, flags
, buf
, options
,
942 s
->desc_offset
= 0x200;
945 ret
= vmdk_open_desc_file(bs
, flags
, buf
, options
, errp
);
952 /* try to open parent images, if exist */
953 ret
= vmdk_parent_open(bs
);
957 s
->cid
= vmdk_read_cid(bs
, 0);
958 s
->parent_cid
= vmdk_read_cid(bs
, 1);
959 qemu_co_mutex_init(&s
->lock
);
961 /* Disable migration when VMDK images are used */
962 error_setg(&s
->migration_blocker
, "The vmdk format used by node '%s' "
963 "does not support live migration",
964 bdrv_get_device_or_node_name(bs
));
965 migrate_add_blocker(s
->migration_blocker
);
971 g_free(s
->create_type
);
972 s
->create_type
= NULL
;
973 vmdk_free_extents(bs
);
978 static void vmdk_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
980 BDRVVmdkState
*s
= bs
->opaque
;
983 for (i
= 0; i
< s
->num_extents
; i
++) {
984 if (!s
->extents
[i
].flat
) {
985 bs
->bl
.write_zeroes_alignment
=
986 MAX(bs
->bl
.write_zeroes_alignment
,
987 s
->extents
[i
].cluster_sectors
);
995 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
996 * to the cluster at @cluster_sector_num.
998 * If @skip_start_sector < @skip_end_sector, the relative range
999 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1000 * it for call to write user data in the request.
1002 static int get_whole_cluster(BlockDriverState
*bs
,
1004 uint64_t cluster_sector_num
,
1005 uint64_t sector_num
,
1006 uint64_t skip_start_sector
,
1007 uint64_t skip_end_sector
)
1010 int64_t cluster_bytes
;
1011 uint8_t *whole_grain
;
1013 /* For COW, align request sector_num to cluster start */
1014 sector_num
= QEMU_ALIGN_DOWN(sector_num
, extent
->cluster_sectors
);
1015 cluster_bytes
= extent
->cluster_sectors
<< BDRV_SECTOR_BITS
;
1016 whole_grain
= qemu_blockalign(bs
, cluster_bytes
);
1019 memset(whole_grain
, 0, skip_start_sector
<< BDRV_SECTOR_BITS
);
1020 memset(whole_grain
+ (skip_end_sector
<< BDRV_SECTOR_BITS
), 0,
1021 cluster_bytes
- (skip_end_sector
<< BDRV_SECTOR_BITS
));
1024 assert(skip_end_sector
<= extent
->cluster_sectors
);
1025 /* we will be here if it's first write on non-exist grain(cluster).
1026 * try to read from parent image, if exist */
1027 if (bs
->backing
&& !vmdk_is_cid_valid(bs
)) {
1032 /* Read backing data before skip range */
1033 if (skip_start_sector
> 0) {
1035 ret
= bdrv_read(bs
->backing
->bs
, sector_num
,
1036 whole_grain
, skip_start_sector
);
1042 ret
= bdrv_write(extent
->file
->bs
, cluster_sector_num
, whole_grain
,
1049 /* Read backing data after skip range */
1050 if (skip_end_sector
< extent
->cluster_sectors
) {
1052 ret
= bdrv_read(bs
->backing
->bs
, sector_num
+ skip_end_sector
,
1053 whole_grain
+ (skip_end_sector
<< BDRV_SECTOR_BITS
),
1054 extent
->cluster_sectors
- skip_end_sector
);
1060 ret
= bdrv_write(extent
->file
->bs
, cluster_sector_num
+ skip_end_sector
,
1061 whole_grain
+ (skip_end_sector
<< BDRV_SECTOR_BITS
),
1062 extent
->cluster_sectors
- skip_end_sector
);
1070 qemu_vfree(whole_grain
);
1074 static int vmdk_L2update(VmdkExtent
*extent
, VmdkMetaData
*m_data
,
1077 offset
= cpu_to_le32(offset
);
1078 /* update L2 table */
1079 if (bdrv_pwrite_sync(
1081 ((int64_t)m_data
->l2_offset
* 512)
1082 + (m_data
->l2_index
* sizeof(offset
)),
1083 &offset
, sizeof(offset
)) < 0) {
1086 /* update backup L2 table */
1087 if (extent
->l1_backup_table_offset
!= 0) {
1088 m_data
->l2_offset
= extent
->l1_backup_table
[m_data
->l1_index
];
1089 if (bdrv_pwrite_sync(
1091 ((int64_t)m_data
->l2_offset
* 512)
1092 + (m_data
->l2_index
* sizeof(offset
)),
1093 &offset
, sizeof(offset
)) < 0) {
1097 if (m_data
->l2_cache_entry
) {
1098 *m_data
->l2_cache_entry
= offset
;
1105 * get_cluster_offset
1107 * Look up cluster offset in extent file by sector number, and store in
1110 * For flat extents, the start offset as parsed from the description file is
1113 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1114 * offset for a new cluster and update L2 cache. If there is a backing file,
1115 * COW is done before returning; otherwise, zeroes are written to the allocated
1116 * cluster. Both COW and zero writing skips the sector range
1117 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1118 * has new data to write there.
1120 * Returns: VMDK_OK if cluster exists and mapped in the image.
1121 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1122 * VMDK_ERROR if failed.
1124 static int get_cluster_offset(BlockDriverState
*bs
,
1126 VmdkMetaData
*m_data
,
1129 uint64_t *cluster_offset
,
1130 uint64_t skip_start_sector
,
1131 uint64_t skip_end_sector
)
1133 unsigned int l1_index
, l2_offset
, l2_index
;
1134 int min_index
, i
, j
;
1135 uint32_t min_count
, *l2_table
;
1136 bool zeroed
= false;
1138 int64_t cluster_sector
;
1144 *cluster_offset
= extent
->flat_start_offset
;
1148 offset
-= (extent
->end_sector
- extent
->sectors
) * SECTOR_SIZE
;
1149 l1_index
= (offset
>> 9) / extent
->l1_entry_sectors
;
1150 if (l1_index
>= extent
->l1_size
) {
1153 l2_offset
= extent
->l1_table
[l1_index
];
1155 return VMDK_UNALLOC
;
1157 for (i
= 0; i
< L2_CACHE_SIZE
; i
++) {
1158 if (l2_offset
== extent
->l2_cache_offsets
[i
]) {
1159 /* increment the hit count */
1160 if (++extent
->l2_cache_counts
[i
] == 0xffffffff) {
1161 for (j
= 0; j
< L2_CACHE_SIZE
; j
++) {
1162 extent
->l2_cache_counts
[j
] >>= 1;
1165 l2_table
= extent
->l2_cache
+ (i
* extent
->l2_size
);
1169 /* not found: load a new entry in the least used one */
1171 min_count
= 0xffffffff;
1172 for (i
= 0; i
< L2_CACHE_SIZE
; i
++) {
1173 if (extent
->l2_cache_counts
[i
] < min_count
) {
1174 min_count
= extent
->l2_cache_counts
[i
];
1178 l2_table
= extent
->l2_cache
+ (min_index
* extent
->l2_size
);
1181 (int64_t)l2_offset
* 512,
1183 extent
->l2_size
* sizeof(uint32_t)
1184 ) != extent
->l2_size
* sizeof(uint32_t)) {
1188 extent
->l2_cache_offsets
[min_index
] = l2_offset
;
1189 extent
->l2_cache_counts
[min_index
] = 1;
1191 l2_index
= ((offset
>> 9) / extent
->cluster_sectors
) % extent
->l2_size
;
1192 cluster_sector
= le32_to_cpu(l2_table
[l2_index
]);
1196 m_data
->l1_index
= l1_index
;
1197 m_data
->l2_index
= l2_index
;
1198 m_data
->l2_offset
= l2_offset
;
1199 m_data
->l2_cache_entry
= &l2_table
[l2_index
];
1201 if (extent
->has_zero_grain
&& cluster_sector
== VMDK_GTE_ZEROED
) {
1205 if (!cluster_sector
|| zeroed
) {
1207 return zeroed
? VMDK_ZEROED
: VMDK_UNALLOC
;
1210 cluster_sector
= extent
->next_cluster_sector
;
1211 extent
->next_cluster_sector
+= extent
->cluster_sectors
;
1213 /* First of all we write grain itself, to avoid race condition
1214 * that may to corrupt the image.
1215 * This problem may occur because of insufficient space on host disk
1216 * or inappropriate VM shutdown.
1218 ret
= get_whole_cluster(bs
, extent
,
1220 offset
>> BDRV_SECTOR_BITS
,
1221 skip_start_sector
, skip_end_sector
);
1226 *cluster_offset
= cluster_sector
<< BDRV_SECTOR_BITS
;
1230 static VmdkExtent
*find_extent(BDRVVmdkState
*s
,
1231 int64_t sector_num
, VmdkExtent
*start_hint
)
1233 VmdkExtent
*extent
= start_hint
;
1236 extent
= &s
->extents
[0];
1238 while (extent
< &s
->extents
[s
->num_extents
]) {
1239 if (sector_num
< extent
->end_sector
) {
1247 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent
*extent
,
1250 uint64_t index_in_cluster
, extent_begin_sector
, extent_relative_sector_num
;
1252 extent_begin_sector
= extent
->end_sector
- extent
->sectors
;
1253 extent_relative_sector_num
= sector_num
- extent_begin_sector
;
1254 index_in_cluster
= extent_relative_sector_num
% extent
->cluster_sectors
;
1255 return index_in_cluster
;
1258 static int64_t coroutine_fn
vmdk_co_get_block_status(BlockDriverState
*bs
,
1259 int64_t sector_num
, int nb_sectors
, int *pnum
)
1261 BDRVVmdkState
*s
= bs
->opaque
;
1262 int64_t index_in_cluster
, n
, ret
;
1266 extent
= find_extent(s
, sector_num
, NULL
);
1270 qemu_co_mutex_lock(&s
->lock
);
1271 ret
= get_cluster_offset(bs
, extent
, NULL
,
1272 sector_num
* 512, false, &offset
,
1274 qemu_co_mutex_unlock(&s
->lock
);
1284 ret
= BDRV_BLOCK_ZERO
;
1287 ret
= BDRV_BLOCK_DATA
;
1288 if (extent
->file
== bs
->file
&& !extent
->compressed
) {
1289 ret
|= BDRV_BLOCK_OFFSET_VALID
| offset
;
1295 index_in_cluster
= vmdk_find_index_in_cluster(extent
, sector_num
);
1296 n
= extent
->cluster_sectors
- index_in_cluster
;
1297 if (n
> nb_sectors
) {
1304 static int vmdk_write_extent(VmdkExtent
*extent
, int64_t cluster_offset
,
1305 int64_t offset_in_cluster
, const uint8_t *buf
,
1306 int nb_sectors
, int64_t sector_num
)
1309 VmdkGrainMarker
*data
= NULL
;
1311 const uint8_t *write_buf
= buf
;
1312 int write_len
= nb_sectors
* 512;
1313 int64_t write_offset
;
1314 int64_t write_end_sector
;
1316 if (extent
->compressed
) {
1317 if (!extent
->has_marker
) {
1321 buf_len
= (extent
->cluster_sectors
<< 9) * 2;
1322 data
= g_malloc(buf_len
+ sizeof(VmdkGrainMarker
));
1323 if (compress(data
->data
, &buf_len
, buf
, nb_sectors
<< 9) != Z_OK
||
1328 data
->lba
= sector_num
;
1329 data
->size
= buf_len
;
1330 write_buf
= (uint8_t *)data
;
1331 write_len
= buf_len
+ sizeof(VmdkGrainMarker
);
1333 write_offset
= cluster_offset
+ offset_in_cluster
,
1334 ret
= bdrv_pwrite(extent
->file
->bs
, write_offset
, write_buf
, write_len
);
1336 write_end_sector
= DIV_ROUND_UP(write_offset
+ write_len
, BDRV_SECTOR_SIZE
);
1338 if (extent
->compressed
) {
1339 extent
->next_cluster_sector
= write_end_sector
;
1341 extent
->next_cluster_sector
= MAX(extent
->next_cluster_sector
,
1345 if (ret
!= write_len
) {
1346 ret
= ret
< 0 ? ret
: -EIO
;
1355 static int vmdk_read_extent(VmdkExtent
*extent
, int64_t cluster_offset
,
1356 int64_t offset_in_cluster
, uint8_t *buf
,
1360 int cluster_bytes
, buf_bytes
;
1361 uint8_t *cluster_buf
, *compressed_data
;
1362 uint8_t *uncomp_buf
;
1364 VmdkGrainMarker
*marker
;
1368 if (!extent
->compressed
) {
1369 ret
= bdrv_pread(extent
->file
->bs
,
1370 cluster_offset
+ offset_in_cluster
,
1371 buf
, nb_sectors
* 512);
1372 if (ret
== nb_sectors
* 512) {
1378 cluster_bytes
= extent
->cluster_sectors
* 512;
1379 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1380 buf_bytes
= cluster_bytes
* 2;
1381 cluster_buf
= g_malloc(buf_bytes
);
1382 uncomp_buf
= g_malloc(cluster_bytes
);
1383 ret
= bdrv_pread(extent
->file
->bs
,
1385 cluster_buf
, buf_bytes
);
1389 compressed_data
= cluster_buf
;
1390 buf_len
= cluster_bytes
;
1391 data_len
= cluster_bytes
;
1392 if (extent
->has_marker
) {
1393 marker
= (VmdkGrainMarker
*)cluster_buf
;
1394 compressed_data
= marker
->data
;
1395 data_len
= le32_to_cpu(marker
->size
);
1397 if (!data_len
|| data_len
> buf_bytes
) {
1401 ret
= uncompress(uncomp_buf
, &buf_len
, compressed_data
, data_len
);
1407 if (offset_in_cluster
< 0 ||
1408 offset_in_cluster
+ nb_sectors
* 512 > buf_len
) {
1412 memcpy(buf
, uncomp_buf
+ offset_in_cluster
, nb_sectors
* 512);
1417 g_free(cluster_buf
);
1421 static int vmdk_read(BlockDriverState
*bs
, int64_t sector_num
,
1422 uint8_t *buf
, int nb_sectors
)
1424 BDRVVmdkState
*s
= bs
->opaque
;
1426 uint64_t n
, index_in_cluster
;
1427 VmdkExtent
*extent
= NULL
;
1428 uint64_t cluster_offset
;
1430 while (nb_sectors
> 0) {
1431 extent
= find_extent(s
, sector_num
, extent
);
1435 ret
= get_cluster_offset(bs
, extent
, NULL
,
1436 sector_num
<< 9, false, &cluster_offset
,
1438 index_in_cluster
= vmdk_find_index_in_cluster(extent
, sector_num
);
1439 n
= extent
->cluster_sectors
- index_in_cluster
;
1440 if (n
> nb_sectors
) {
1443 if (ret
!= VMDK_OK
) {
1444 /* if not allocated, try to read from parent image, if exist */
1445 if (bs
->backing
&& ret
!= VMDK_ZEROED
) {
1446 if (!vmdk_is_cid_valid(bs
)) {
1449 ret
= bdrv_read(bs
->backing
->bs
, sector_num
, buf
, n
);
1454 memset(buf
, 0, 512 * n
);
1457 ret
= vmdk_read_extent(extent
,
1458 cluster_offset
, index_in_cluster
* 512,
1471 static coroutine_fn
int vmdk_co_read(BlockDriverState
*bs
, int64_t sector_num
,
1472 uint8_t *buf
, int nb_sectors
)
1475 BDRVVmdkState
*s
= bs
->opaque
;
1476 qemu_co_mutex_lock(&s
->lock
);
1477 ret
= vmdk_read(bs
, sector_num
, buf
, nb_sectors
);
1478 qemu_co_mutex_unlock(&s
->lock
);
1484 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1485 * if possible, otherwise return -ENOTSUP.
1486 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1487 * with each cluster. By dry run we can find if the zero write
1488 * is possible without modifying image data.
1490 * Returns: error code with 0 for success.
1492 static int vmdk_write(BlockDriverState
*bs
, int64_t sector_num
,
1493 const uint8_t *buf
, int nb_sectors
,
1494 bool zeroed
, bool zero_dry_run
)
1496 BDRVVmdkState
*s
= bs
->opaque
;
1497 VmdkExtent
*extent
= NULL
;
1499 int64_t index_in_cluster
, n
;
1500 uint64_t cluster_offset
;
1501 VmdkMetaData m_data
;
1503 if (sector_num
> bs
->total_sectors
) {
1504 error_report("Wrong offset: sector_num=0x%" PRIx64
1505 " total_sectors=0x%" PRIx64
,
1506 sector_num
, bs
->total_sectors
);
1510 while (nb_sectors
> 0) {
1511 extent
= find_extent(s
, sector_num
, extent
);
1515 index_in_cluster
= vmdk_find_index_in_cluster(extent
, sector_num
);
1516 n
= extent
->cluster_sectors
- index_in_cluster
;
1517 if (n
> nb_sectors
) {
1520 ret
= get_cluster_offset(bs
, extent
, &m_data
, sector_num
<< 9,
1521 !(extent
->compressed
|| zeroed
),
1523 index_in_cluster
, index_in_cluster
+ n
);
1524 if (extent
->compressed
) {
1525 if (ret
== VMDK_OK
) {
1526 /* Refuse write to allocated cluster for streamOptimized */
1527 error_report("Could not write to allocated cluster"
1528 " for streamOptimized");
1532 ret
= get_cluster_offset(bs
, extent
, &m_data
, sector_num
<< 9,
1533 true, &cluster_offset
, 0, 0);
1536 if (ret
== VMDK_ERROR
) {
1540 /* Do zeroed write, buf is ignored */
1541 if (extent
->has_zero_grain
&&
1542 index_in_cluster
== 0 &&
1543 n
>= extent
->cluster_sectors
) {
1544 n
= extent
->cluster_sectors
;
1545 if (!zero_dry_run
) {
1546 /* update L2 tables */
1547 if (vmdk_L2update(extent
, &m_data
, VMDK_GTE_ZEROED
)
1556 ret
= vmdk_write_extent(extent
,
1557 cluster_offset
, index_in_cluster
* 512,
1558 buf
, n
, sector_num
);
1563 /* update L2 tables */
1564 if (vmdk_L2update(extent
, &m_data
,
1565 cluster_offset
>> BDRV_SECTOR_BITS
)
1575 /* update CID on the first write every time the virtual disk is
1577 if (!s
->cid_updated
) {
1578 ret
= vmdk_write_cid(bs
, g_random_int());
1582 s
->cid_updated
= true;
1588 static coroutine_fn
int vmdk_co_write(BlockDriverState
*bs
, int64_t sector_num
,
1589 const uint8_t *buf
, int nb_sectors
)
1592 BDRVVmdkState
*s
= bs
->opaque
;
1593 qemu_co_mutex_lock(&s
->lock
);
1594 ret
= vmdk_write(bs
, sector_num
, buf
, nb_sectors
, false, false);
1595 qemu_co_mutex_unlock(&s
->lock
);
1599 static int vmdk_write_compressed(BlockDriverState
*bs
,
1604 BDRVVmdkState
*s
= bs
->opaque
;
1605 if (s
->num_extents
== 1 && s
->extents
[0].compressed
) {
1606 return vmdk_write(bs
, sector_num
, buf
, nb_sectors
, false, false);
1612 static int coroutine_fn
vmdk_co_write_zeroes(BlockDriverState
*bs
,
1615 BdrvRequestFlags flags
)
1618 BDRVVmdkState
*s
= bs
->opaque
;
1619 qemu_co_mutex_lock(&s
->lock
);
1620 /* write zeroes could fail if sectors not aligned to cluster, test it with
1621 * dry_run == true before really updating image */
1622 ret
= vmdk_write(bs
, sector_num
, NULL
, nb_sectors
, true, true);
1624 ret
= vmdk_write(bs
, sector_num
, NULL
, nb_sectors
, true, false);
1626 qemu_co_mutex_unlock(&s
->lock
);
1630 static int vmdk_create_extent(const char *filename
, int64_t filesize
,
1631 bool flat
, bool compress
, bool zeroed_grain
,
1632 QemuOpts
*opts
, Error
**errp
)
1635 BlockDriverState
*bs
= NULL
;
1637 Error
*local_err
= NULL
;
1638 uint32_t tmp
, magic
, grains
, gd_sectors
, gt_size
, gt_count
;
1639 uint32_t *gd_buf
= NULL
;
1642 ret
= bdrv_create_file(filename
, opts
, &local_err
);
1644 error_propagate(errp
, local_err
);
1649 ret
= bdrv_open(&bs
, filename
, NULL
, NULL
, BDRV_O_RDWR
| BDRV_O_PROTOCOL
,
1652 error_propagate(errp
, local_err
);
1657 ret
= bdrv_truncate(bs
, filesize
);
1659 error_setg_errno(errp
, -ret
, "Could not truncate file");
1663 magic
= cpu_to_be32(VMDK4_MAGIC
);
1664 memset(&header
, 0, sizeof(header
));
1665 header
.version
= zeroed_grain
? 2 : 1;
1666 header
.flags
= VMDK4_FLAG_RGD
| VMDK4_FLAG_NL_DETECT
1667 | (compress
? VMDK4_FLAG_COMPRESS
| VMDK4_FLAG_MARKER
: 0)
1668 | (zeroed_grain
? VMDK4_FLAG_ZERO_GRAIN
: 0);
1669 header
.compressAlgorithm
= compress
? VMDK4_COMPRESSION_DEFLATE
: 0;
1670 header
.capacity
= filesize
/ BDRV_SECTOR_SIZE
;
1671 header
.granularity
= 128;
1672 header
.num_gtes_per_gt
= BDRV_SECTOR_SIZE
;
1674 grains
= DIV_ROUND_UP(filesize
/ BDRV_SECTOR_SIZE
, header
.granularity
);
1675 gt_size
= DIV_ROUND_UP(header
.num_gtes_per_gt
* sizeof(uint32_t),
1677 gt_count
= DIV_ROUND_UP(grains
, header
.num_gtes_per_gt
);
1678 gd_sectors
= DIV_ROUND_UP(gt_count
* sizeof(uint32_t), BDRV_SECTOR_SIZE
);
1680 header
.desc_offset
= 1;
1681 header
.desc_size
= 20;
1682 header
.rgd_offset
= header
.desc_offset
+ header
.desc_size
;
1683 header
.gd_offset
= header
.rgd_offset
+ gd_sectors
+ (gt_size
* gt_count
);
1684 header
.grain_offset
=
1685 ROUND_UP(header
.gd_offset
+ gd_sectors
+ (gt_size
* gt_count
),
1686 header
.granularity
);
1687 /* swap endianness for all header fields */
1688 header
.version
= cpu_to_le32(header
.version
);
1689 header
.flags
= cpu_to_le32(header
.flags
);
1690 header
.capacity
= cpu_to_le64(header
.capacity
);
1691 header
.granularity
= cpu_to_le64(header
.granularity
);
1692 header
.num_gtes_per_gt
= cpu_to_le32(header
.num_gtes_per_gt
);
1693 header
.desc_offset
= cpu_to_le64(header
.desc_offset
);
1694 header
.desc_size
= cpu_to_le64(header
.desc_size
);
1695 header
.rgd_offset
= cpu_to_le64(header
.rgd_offset
);
1696 header
.gd_offset
= cpu_to_le64(header
.gd_offset
);
1697 header
.grain_offset
= cpu_to_le64(header
.grain_offset
);
1698 header
.compressAlgorithm
= cpu_to_le16(header
.compressAlgorithm
);
1700 header
.check_bytes
[0] = 0xa;
1701 header
.check_bytes
[1] = 0x20;
1702 header
.check_bytes
[2] = 0xd;
1703 header
.check_bytes
[3] = 0xa;
1705 /* write all the data */
1706 ret
= bdrv_pwrite(bs
, 0, &magic
, sizeof(magic
));
1708 error_setg(errp
, QERR_IO_ERROR
);
1711 ret
= bdrv_pwrite(bs
, sizeof(magic
), &header
, sizeof(header
));
1713 error_setg(errp
, QERR_IO_ERROR
);
1717 ret
= bdrv_truncate(bs
, le64_to_cpu(header
.grain_offset
) << 9);
1719 error_setg_errno(errp
, -ret
, "Could not truncate file");
1723 /* write grain directory */
1724 gd_buf_size
= gd_sectors
* BDRV_SECTOR_SIZE
;
1725 gd_buf
= g_malloc0(gd_buf_size
);
1726 for (i
= 0, tmp
= le64_to_cpu(header
.rgd_offset
) + gd_sectors
;
1727 i
< gt_count
; i
++, tmp
+= gt_size
) {
1728 gd_buf
[i
] = cpu_to_le32(tmp
);
1730 ret
= bdrv_pwrite(bs
, le64_to_cpu(header
.rgd_offset
) * BDRV_SECTOR_SIZE
,
1731 gd_buf
, gd_buf_size
);
1733 error_setg(errp
, QERR_IO_ERROR
);
1737 /* write backup grain directory */
1738 for (i
= 0, tmp
= le64_to_cpu(header
.gd_offset
) + gd_sectors
;
1739 i
< gt_count
; i
++, tmp
+= gt_size
) {
1740 gd_buf
[i
] = cpu_to_le32(tmp
);
1742 ret
= bdrv_pwrite(bs
, le64_to_cpu(header
.gd_offset
) * BDRV_SECTOR_SIZE
,
1743 gd_buf
, gd_buf_size
);
1745 error_setg(errp
, QERR_IO_ERROR
);
1758 static int filename_decompose(const char *filename
, char *path
, char *prefix
,
1759 char *postfix
, size_t buf_len
, Error
**errp
)
1763 if (filename
== NULL
|| !strlen(filename
)) {
1764 error_setg(errp
, "No filename provided");
1767 p
= strrchr(filename
, '/');
1769 p
= strrchr(filename
, '\\');
1772 p
= strrchr(filename
, ':');
1776 if (p
- filename
>= buf_len
) {
1779 pstrcpy(path
, p
- filename
+ 1, filename
);
1784 q
= strrchr(p
, '.');
1786 pstrcpy(prefix
, buf_len
, p
);
1789 if (q
- p
>= buf_len
) {
1792 pstrcpy(prefix
, q
- p
+ 1, p
);
1793 pstrcpy(postfix
, buf_len
, q
);
1798 static int vmdk_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
1801 BlockDriverState
*new_bs
= NULL
;
1802 Error
*local_err
= NULL
;
1804 int64_t total_size
= 0, filesize
;
1805 char *adapter_type
= NULL
;
1806 char *backing_file
= NULL
;
1810 bool flat
, split
, compress
;
1811 GString
*ext_desc_lines
;
1812 char *path
= g_malloc0(PATH_MAX
);
1813 char *prefix
= g_malloc0(PATH_MAX
);
1814 char *postfix
= g_malloc0(PATH_MAX
);
1815 char *desc_line
= g_malloc0(BUF_SIZE
);
1816 char *ext_filename
= g_malloc0(PATH_MAX
);
1817 char *desc_filename
= g_malloc0(PATH_MAX
);
1818 const int64_t split_size
= 0x80000000; /* VMDK has constant split size */
1819 const char *desc_extent_line
;
1820 char *parent_desc_line
= g_malloc0(BUF_SIZE
);
1821 uint32_t parent_cid
= 0xffffffff;
1822 uint32_t number_heads
= 16;
1823 bool zeroed_grain
= false;
1824 uint32_t desc_offset
= 0, desc_len
;
1825 const char desc_template
[] =
1826 "# Disk DescriptorFile\n"
1829 "parentCID=%" PRIx32
"\n"
1830 "createType=\"%s\"\n"
1833 "# Extent description\n"
1836 "# The Disk Data Base\n"
1839 "ddb.virtualHWVersion = \"%d\"\n"
1840 "ddb.geometry.cylinders = \"%" PRId64
"\"\n"
1841 "ddb.geometry.heads = \"%" PRIu32
"\"\n"
1842 "ddb.geometry.sectors = \"63\"\n"
1843 "ddb.adapterType = \"%s\"\n";
1845 ext_desc_lines
= g_string_new(NULL
);
1847 if (filename_decompose(filename
, path
, prefix
, postfix
, PATH_MAX
, errp
)) {
1851 /* Read out options */
1852 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
1854 adapter_type
= qemu_opt_get_del(opts
, BLOCK_OPT_ADAPTER_TYPE
);
1855 backing_file
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
1856 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_COMPAT6
, false)) {
1857 flags
|= BLOCK_FLAG_COMPAT6
;
1859 fmt
= qemu_opt_get_del(opts
, BLOCK_OPT_SUBFMT
);
1860 if (qemu_opt_get_bool_del(opts
, BLOCK_OPT_ZEROED_GRAIN
, false)) {
1861 zeroed_grain
= true;
1864 if (!adapter_type
) {
1865 adapter_type
= g_strdup("ide");
1866 } else if (strcmp(adapter_type
, "ide") &&
1867 strcmp(adapter_type
, "buslogic") &&
1868 strcmp(adapter_type
, "lsilogic") &&
1869 strcmp(adapter_type
, "legacyESX")) {
1870 error_setg(errp
, "Unknown adapter type: '%s'", adapter_type
);
1874 if (strcmp(adapter_type
, "ide") != 0) {
1875 /* that's the number of heads with which vmware operates when
1876 creating, exporting, etc. vmdk files with a non-ide adapter type */
1880 /* Default format to monolithicSparse */
1881 fmt
= g_strdup("monolithicSparse");
1882 } else if (strcmp(fmt
, "monolithicFlat") &&
1883 strcmp(fmt
, "monolithicSparse") &&
1884 strcmp(fmt
, "twoGbMaxExtentSparse") &&
1885 strcmp(fmt
, "twoGbMaxExtentFlat") &&
1886 strcmp(fmt
, "streamOptimized")) {
1887 error_setg(errp
, "Unknown subformat: '%s'", fmt
);
1891 split
= !(strcmp(fmt
, "twoGbMaxExtentFlat") &&
1892 strcmp(fmt
, "twoGbMaxExtentSparse"));
1893 flat
= !(strcmp(fmt
, "monolithicFlat") &&
1894 strcmp(fmt
, "twoGbMaxExtentFlat"));
1895 compress
= !strcmp(fmt
, "streamOptimized");
1897 desc_extent_line
= "RW %" PRId64
" FLAT \"%s\" 0\n";
1899 desc_extent_line
= "RW %" PRId64
" SPARSE \"%s\"\n";
1901 if (flat
&& backing_file
) {
1902 error_setg(errp
, "Flat image can't have backing file");
1906 if (flat
&& zeroed_grain
) {
1907 error_setg(errp
, "Flat image can't enable zeroed grain");
1912 BlockDriverState
*bs
= NULL
;
1913 char *full_backing
= g_new0(char, PATH_MAX
);
1914 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
1915 full_backing
, PATH_MAX
,
1918 g_free(full_backing
);
1919 error_propagate(errp
, local_err
);
1923 ret
= bdrv_open(&bs
, full_backing
, NULL
, NULL
, BDRV_O_NO_BACKING
, errp
);
1924 g_free(full_backing
);
1928 if (strcmp(bs
->drv
->format_name
, "vmdk")) {
1933 parent_cid
= vmdk_read_cid(bs
, 0);
1935 snprintf(parent_desc_line
, BUF_SIZE
,
1936 "parentFileNameHint=\"%s\"", backing_file
);
1939 /* Create extents */
1940 filesize
= total_size
;
1941 while (filesize
> 0) {
1942 int64_t size
= filesize
;
1944 if (split
&& size
> split_size
) {
1948 snprintf(desc_filename
, PATH_MAX
, "%s-%c%03d%s",
1949 prefix
, flat
? 'f' : 's', ++idx
, postfix
);
1951 snprintf(desc_filename
, PATH_MAX
, "%s-flat%s", prefix
, postfix
);
1953 snprintf(desc_filename
, PATH_MAX
, "%s%s", prefix
, postfix
);
1955 snprintf(ext_filename
, PATH_MAX
, "%s%s", path
, desc_filename
);
1957 if (vmdk_create_extent(ext_filename
, size
,
1958 flat
, compress
, zeroed_grain
, opts
, errp
)) {
1964 /* Format description line */
1965 snprintf(desc_line
, BUF_SIZE
,
1966 desc_extent_line
, size
/ BDRV_SECTOR_SIZE
, desc_filename
);
1967 g_string_append(ext_desc_lines
, desc_line
);
1969 /* generate descriptor file */
1970 desc
= g_strdup_printf(desc_template
,
1975 ext_desc_lines
->str
,
1976 (flags
& BLOCK_FLAG_COMPAT6
? 6 : 4),
1978 (int64_t)(63 * number_heads
* BDRV_SECTOR_SIZE
),
1981 desc_len
= strlen(desc
);
1982 /* the descriptor offset = 0x200 */
1983 if (!split
&& !flat
) {
1984 desc_offset
= 0x200;
1986 ret
= bdrv_create_file(filename
, opts
, &local_err
);
1988 error_propagate(errp
, local_err
);
1992 assert(new_bs
== NULL
);
1993 ret
= bdrv_open(&new_bs
, filename
, NULL
, NULL
,
1994 BDRV_O_RDWR
| BDRV_O_PROTOCOL
, &local_err
);
1996 error_propagate(errp
, local_err
);
1999 ret
= bdrv_pwrite(new_bs
, desc_offset
, desc
, desc_len
);
2001 error_setg_errno(errp
, -ret
, "Could not write description");
2004 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2005 * for description file */
2006 if (desc_offset
== 0) {
2007 ret
= bdrv_truncate(new_bs
, desc_len
);
2009 error_setg_errno(errp
, -ret
, "Could not truncate file");
2016 g_free(adapter_type
);
2017 g_free(backing_file
);
2024 g_free(ext_filename
);
2025 g_free(desc_filename
);
2026 g_free(parent_desc_line
);
2027 g_string_free(ext_desc_lines
, true);
2031 static void vmdk_close(BlockDriverState
*bs
)
2033 BDRVVmdkState
*s
= bs
->opaque
;
2035 vmdk_free_extents(bs
);
2036 g_free(s
->create_type
);
2038 migrate_del_blocker(s
->migration_blocker
);
2039 error_free(s
->migration_blocker
);
2042 static coroutine_fn
int vmdk_co_flush(BlockDriverState
*bs
)
2044 BDRVVmdkState
*s
= bs
->opaque
;
2048 for (i
= 0; i
< s
->num_extents
; i
++) {
2049 err
= bdrv_co_flush(s
->extents
[i
].file
->bs
);
2057 static int64_t vmdk_get_allocated_file_size(BlockDriverState
*bs
)
2062 BDRVVmdkState
*s
= bs
->opaque
;
2064 ret
= bdrv_get_allocated_file_size(bs
->file
->bs
);
2068 for (i
= 0; i
< s
->num_extents
; i
++) {
2069 if (s
->extents
[i
].file
== bs
->file
) {
2072 r
= bdrv_get_allocated_file_size(s
->extents
[i
].file
->bs
);
2081 static int vmdk_has_zero_init(BlockDriverState
*bs
)
2084 BDRVVmdkState
*s
= bs
->opaque
;
2086 /* If has a flat extent and its underlying storage doesn't have zero init,
2088 for (i
= 0; i
< s
->num_extents
; i
++) {
2089 if (s
->extents
[i
].flat
) {
2090 if (!bdrv_has_zero_init(s
->extents
[i
].file
->bs
)) {
2098 static ImageInfo
*vmdk_get_extent_info(VmdkExtent
*extent
)
2100 ImageInfo
*info
= g_new0(ImageInfo
, 1);
2102 *info
= (ImageInfo
){
2103 .filename
= g_strdup(extent
->file
->bs
->filename
),
2104 .format
= g_strdup(extent
->type
),
2105 .virtual_size
= extent
->sectors
* BDRV_SECTOR_SIZE
,
2106 .compressed
= extent
->compressed
,
2107 .has_compressed
= extent
->compressed
,
2108 .cluster_size
= extent
->cluster_sectors
* BDRV_SECTOR_SIZE
,
2109 .has_cluster_size
= !extent
->flat
,
2115 static int vmdk_check(BlockDriverState
*bs
, BdrvCheckResult
*result
,
2118 BDRVVmdkState
*s
= bs
->opaque
;
2119 VmdkExtent
*extent
= NULL
;
2120 int64_t sector_num
= 0;
2121 int64_t total_sectors
= bdrv_nb_sectors(bs
);
2123 uint64_t cluster_offset
;
2130 if (sector_num
>= total_sectors
) {
2133 extent
= find_extent(s
, sector_num
, extent
);
2136 "ERROR: could not find extent for sector %" PRId64
"\n",
2140 ret
= get_cluster_offset(bs
, extent
, NULL
,
2141 sector_num
<< BDRV_SECTOR_BITS
,
2142 false, &cluster_offset
, 0, 0);
2143 if (ret
== VMDK_ERROR
) {
2145 "ERROR: could not get cluster_offset for sector %"
2146 PRId64
"\n", sector_num
);
2149 if (ret
== VMDK_OK
&&
2150 cluster_offset
>= bdrv_getlength(extent
->file
->bs
))
2153 "ERROR: cluster offset for sector %"
2154 PRId64
" points after EOF\n", sector_num
);
2157 sector_num
+= extent
->cluster_sectors
;
2160 result
->corruptions
++;
2164 static ImageInfoSpecific
*vmdk_get_specific_info(BlockDriverState
*bs
)
2167 BDRVVmdkState
*s
= bs
->opaque
;
2168 ImageInfoSpecific
*spec_info
= g_new0(ImageInfoSpecific
, 1);
2169 ImageInfoList
**next
;
2171 *spec_info
= (ImageInfoSpecific
){
2172 .type
= IMAGE_INFO_SPECIFIC_KIND_VMDK
,
2174 .vmdk
= g_new0(ImageInfoSpecificVmdk
, 1),
2178 *spec_info
->u
.vmdk
= (ImageInfoSpecificVmdk
) {
2179 .create_type
= g_strdup(s
->create_type
),
2181 .parent_cid
= s
->parent_cid
,
2184 next
= &spec_info
->u
.vmdk
->extents
;
2185 for (i
= 0; i
< s
->num_extents
; i
++) {
2186 *next
= g_new0(ImageInfoList
, 1);
2187 (*next
)->value
= vmdk_get_extent_info(&s
->extents
[i
]);
2188 (*next
)->next
= NULL
;
2189 next
= &(*next
)->next
;
2195 static bool vmdk_extents_type_eq(const VmdkExtent
*a
, const VmdkExtent
*b
)
2197 return a
->flat
== b
->flat
&&
2198 a
->compressed
== b
->compressed
&&
2199 (a
->flat
|| a
->cluster_sectors
== b
->cluster_sectors
);
2202 static int vmdk_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2205 BDRVVmdkState
*s
= bs
->opaque
;
2206 assert(s
->num_extents
);
2208 /* See if we have multiple extents but they have different cases */
2209 for (i
= 1; i
< s
->num_extents
; i
++) {
2210 if (!vmdk_extents_type_eq(&s
->extents
[0], &s
->extents
[i
])) {
2214 bdi
->needs_compressed_writes
= s
->extents
[0].compressed
;
2215 if (!s
->extents
[0].flat
) {
2216 bdi
->cluster_size
= s
->extents
[0].cluster_sectors
<< BDRV_SECTOR_BITS
;
2221 static void vmdk_detach_aio_context(BlockDriverState
*bs
)
2223 BDRVVmdkState
*s
= bs
->opaque
;
2226 for (i
= 0; i
< s
->num_extents
; i
++) {
2227 bdrv_detach_aio_context(s
->extents
[i
].file
->bs
);
2231 static void vmdk_attach_aio_context(BlockDriverState
*bs
,
2232 AioContext
*new_context
)
2234 BDRVVmdkState
*s
= bs
->opaque
;
2237 for (i
= 0; i
< s
->num_extents
; i
++) {
2238 bdrv_attach_aio_context(s
->extents
[i
].file
->bs
, new_context
);
2242 static QemuOptsList vmdk_create_opts
= {
2243 .name
= "vmdk-create-opts",
2244 .head
= QTAILQ_HEAD_INITIALIZER(vmdk_create_opts
.head
),
2247 .name
= BLOCK_OPT_SIZE
,
2248 .type
= QEMU_OPT_SIZE
,
2249 .help
= "Virtual disk size"
2252 .name
= BLOCK_OPT_ADAPTER_TYPE
,
2253 .type
= QEMU_OPT_STRING
,
2254 .help
= "Virtual adapter type, can be one of "
2255 "ide (default), lsilogic, buslogic or legacyESX"
2258 .name
= BLOCK_OPT_BACKING_FILE
,
2259 .type
= QEMU_OPT_STRING
,
2260 .help
= "File name of a base image"
2263 .name
= BLOCK_OPT_COMPAT6
,
2264 .type
= QEMU_OPT_BOOL
,
2265 .help
= "VMDK version 6 image",
2266 .def_value_str
= "off"
2269 .name
= BLOCK_OPT_SUBFMT
,
2270 .type
= QEMU_OPT_STRING
,
2272 "VMDK flat extent format, can be one of "
2273 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2276 .name
= BLOCK_OPT_ZEROED_GRAIN
,
2277 .type
= QEMU_OPT_BOOL
,
2278 .help
= "Enable efficient zero writes "
2279 "using the zeroed-grain GTE feature"
2281 { /* end of list */ }
2285 static BlockDriver bdrv_vmdk
= {
2286 .format_name
= "vmdk",
2287 .instance_size
= sizeof(BDRVVmdkState
),
2288 .bdrv_probe
= vmdk_probe
,
2289 .bdrv_open
= vmdk_open
,
2290 .bdrv_check
= vmdk_check
,
2291 .bdrv_reopen_prepare
= vmdk_reopen_prepare
,
2292 .bdrv_read
= vmdk_co_read
,
2293 .bdrv_write
= vmdk_co_write
,
2294 .bdrv_write_compressed
= vmdk_write_compressed
,
2295 .bdrv_co_write_zeroes
= vmdk_co_write_zeroes
,
2296 .bdrv_close
= vmdk_close
,
2297 .bdrv_create
= vmdk_create
,
2298 .bdrv_co_flush_to_disk
= vmdk_co_flush
,
2299 .bdrv_co_get_block_status
= vmdk_co_get_block_status
,
2300 .bdrv_get_allocated_file_size
= vmdk_get_allocated_file_size
,
2301 .bdrv_has_zero_init
= vmdk_has_zero_init
,
2302 .bdrv_get_specific_info
= vmdk_get_specific_info
,
2303 .bdrv_refresh_limits
= vmdk_refresh_limits
,
2304 .bdrv_get_info
= vmdk_get_info
,
2305 .bdrv_detach_aio_context
= vmdk_detach_aio_context
,
2306 .bdrv_attach_aio_context
= vmdk_attach_aio_context
,
2308 .supports_backing
= true,
2309 .create_opts
= &vmdk_create_opts
,
2312 static void bdrv_vmdk_init(void)
2314 bdrv_register(&bdrv_vmdk
);
2317 block_init(bdrv_vmdk_init
);