2 * Block driver for the VMDK format
4 * Copyright (c) 2004 Fabrice Bellard
5 * Copyright (c) 2005 Filip Navara
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu-common.h"
27 #include "block/block_int.h"
28 #include "qemu/module.h"
29 #include "migration/migration.h"
33 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
34 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
35 #define VMDK4_COMPRESSION_DEFLATE 1
36 #define VMDK4_FLAG_NL_DETECT (1 << 0)
37 #define VMDK4_FLAG_RGD (1 << 1)
38 /* Zeroed-grain enable bit */
39 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
40 #define VMDK4_FLAG_COMPRESS (1 << 16)
41 #define VMDK4_FLAG_MARKER (1 << 17)
42 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
44 #define VMDK_GTE_ZEROED 0x1
46 /* VMDK internal error codes */
48 #define VMDK_ERROR (-1)
49 /* Cluster not allocated */
50 #define VMDK_UNALLOC (-2)
51 #define VMDK_ZEROED (-3)
53 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
58 uint32_t disk_sectors;
60 uint32_t l1dir_offset;
62 uint32_t file_sectors;
65 uint32_t sectors_per_track;
66 } QEMU_PACKED VMDK3Header;
75 /* Number of GrainTableEntries per GrainTable */
76 uint32_t num_gtes_per_gt;
79 uint64_t grain_offset;
82 uint16_t compressAlgorithm;
83 } QEMU_PACKED VMDK4Header;
85 #define L2_CACHE_SIZE 16
87 typedef struct VmdkExtent {
88 BlockDriverState *file;
96 int64_t flat_start_offset;
97 int64_t l1_table_offset;
98 int64_t l1_backup_table_offset;
100 uint32_t *l1_backup_table;
101 unsigned int l1_size;
102 uint32_t l1_entry_sectors;
104 unsigned int l2_size;
106 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
107 uint32_t l2_cache_counts[L2_CACHE_SIZE];
109 int64_t cluster_sectors;
110 int64_t next_cluster_sector;
114 typedef struct BDRVVmdkState {
116 uint64_t desc_offset;
122 /* Extent array with num_extents entries, ascend ordered by address */
124 Error *migration_blocker;
128 typedef struct VmdkMetaData {
129 unsigned int l1_index;
130 unsigned int l2_index;
131 unsigned int l2_offset;
133 uint32_t *l2_cache_entry;
136 typedef struct VmdkGrainMarker {
140 } QEMU_PACKED VmdkGrainMarker;
143 MARKER_END_OF_STREAM = 0,
144 MARKER_GRAIN_TABLE = 1,
145 MARKER_GRAIN_DIRECTORY = 2,
149 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
156 magic = be32_to_cpu(*(uint32_t *)buf);
157 if (magic == VMDK3_MAGIC ||
158 magic == VMDK4_MAGIC) {
161 const char *p = (const char *)buf;
162 const char *end = p + buf_size;
165 /* skip comment line */
166 while (p < end && *p != '\n') {
173 while (p < end && *p == ' ') {
176 /* skip '\r' if windows line endings used. */
177 if (p < end && *p == '\r') {
180 /* only accept blank lines before 'version=' line */
181 if (p == end || *p != '\n') {
187 if (end - p >= strlen("version=X\n")) {
188 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
189 strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
193 if (end - p >= strlen("version=X\r\n")) {
194 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
195 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
205 #define SECTOR_SIZE 512
206 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
207 #define BUF_SIZE 4096
208 #define HEADER_SIZE 512 /* first sector of 512 bytes */
210 static void vmdk_free_extents(BlockDriverState *bs)
213 BDRVVmdkState *s = bs->opaque;
216 for (i = 0; i < s->num_extents; i++) {
220 g_free(e->l1_backup_table);
222 if (e->file != bs->file) {
229 static void vmdk_free_last_extent(BlockDriverState *bs)
231 BDRVVmdkState *s = bs->opaque;
233 if (s->num_extents == 0) {
237 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
240 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
242 char desc[DESC_SIZE];
243 uint32_t cid = 0xffffffff;
244 const char *p_name, *cid_str;
246 BDRVVmdkState *s = bs->opaque;
249 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
255 cid_str = "parentCID";
256 cid_str_size = sizeof("parentCID");
259 cid_str_size = sizeof("CID");
262 desc[DESC_SIZE - 1] = '\0';
263 p_name = strstr(desc, cid_str);
264 if (p_name != NULL) {
265 p_name += cid_str_size;
266 sscanf(p_name, "%" SCNx32, &cid);
272 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
274 char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
275 char *p_name, *tmp_str;
276 BDRVVmdkState *s = bs->opaque;
279 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
284 desc[DESC_SIZE - 1] = '\0';
285 tmp_str = strstr(desc, "parentCID");
286 if (tmp_str == NULL) {
290 pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
291 p_name = strstr(desc, "CID");
292 if (p_name != NULL) {
293 p_name += sizeof("CID");
294 snprintf(p_name, sizeof(desc) - (p_name - desc), "%" PRIx32 "\n", cid);
295 pstrcat(desc, sizeof(desc), tmp_desc);
298 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
306 static int vmdk_is_cid_valid(BlockDriverState *bs)
308 BDRVVmdkState *s = bs->opaque;
309 BlockDriverState *p_bs = bs->backing_hd;
312 if (!s->cid_checked && p_bs) {
313 cur_pcid = vmdk_read_cid(p_bs, 0);
314 if (s->parent_cid != cur_pcid) {
319 s->cid_checked = true;
324 /* Queue extents, if any, for reopen() */
325 static int vmdk_reopen_prepare(BDRVReopenState *state,
326 BlockReopenQueue *queue, Error **errp)
333 assert(state != NULL);
334 assert(state->bs != NULL);
337 error_setg(errp, "No reopen queue for VMDK extents");
341 s = state->bs->opaque;
345 for (i = 0; i < s->num_extents; i++) {
347 if (e->file != state->bs->file) {
348 bdrv_reopen_queue(queue, e->file, state->flags);
357 static int vmdk_parent_open(BlockDriverState *bs)
360 char desc[DESC_SIZE + 1];
361 BDRVVmdkState *s = bs->opaque;
364 desc[DESC_SIZE] = '\0';
365 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
370 p_name = strstr(desc, "parentFileNameHint");
371 if (p_name != NULL) {
374 p_name += sizeof("parentFileNameHint") + 1;
375 end_name = strchr(p_name, '\"');
376 if (end_name == NULL) {
379 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
383 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
389 /* Create and append extent to the extent array. Return the added VmdkExtent
390 * address. return NULL if allocation failed. */
391 static int vmdk_add_extent(BlockDriverState *bs,
392 BlockDriverState *file, bool flat, int64_t sectors,
393 int64_t l1_offset, int64_t l1_backup_offset,
395 int l2_size, uint64_t cluster_sectors,
396 VmdkExtent **new_extent,
400 BDRVVmdkState *s = bs->opaque;
403 if (cluster_sectors > 0x200000) {
404 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
405 error_setg(errp, "Invalid granularity, image may be corrupt");
408 if (l1_size > 512 * 1024 * 1024) {
409 /* Although with big capacity and small l1_entry_sectors, we can get a
410 * big l1_size, we don't want unbounded value to allocate the table.
411 * Limit it to 512M, which is 16PB for default cluster and L2 table
413 error_setg(errp, "L1 size too big");
417 nb_sectors = bdrv_nb_sectors(file);
418 if (nb_sectors < 0) {
422 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
423 extent = &s->extents[s->num_extents];
426 memset(extent, 0, sizeof(VmdkExtent));
429 extent->sectors = sectors;
430 extent->l1_table_offset = l1_offset;
431 extent->l1_backup_table_offset = l1_backup_offset;
432 extent->l1_size = l1_size;
433 extent->l1_entry_sectors = l2_size * cluster_sectors;
434 extent->l2_size = l2_size;
435 extent->cluster_sectors = flat ? sectors : cluster_sectors;
436 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
438 if (s->num_extents > 1) {
439 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
441 extent->end_sector = extent->sectors;
443 bs->total_sectors = extent->end_sector;
445 *new_extent = extent;
450 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
456 /* read the L1 table */
457 l1_size = extent->l1_size * sizeof(uint32_t);
458 extent->l1_table = g_try_malloc(l1_size);
459 if (l1_size && extent->l1_table == NULL) {
463 ret = bdrv_pread(extent->file,
464 extent->l1_table_offset,
468 error_setg_errno(errp, -ret,
469 "Could not read l1 table from extent '%s'",
470 extent->file->filename);
473 for (i = 0; i < extent->l1_size; i++) {
474 le32_to_cpus(&extent->l1_table[i]);
477 if (extent->l1_backup_table_offset) {
478 extent->l1_backup_table = g_try_malloc(l1_size);
479 if (l1_size && extent->l1_backup_table == NULL) {
483 ret = bdrv_pread(extent->file,
484 extent->l1_backup_table_offset,
485 extent->l1_backup_table,
488 error_setg_errno(errp, -ret,
489 "Could not read l1 backup table from extent '%s'",
490 extent->file->filename);
493 for (i = 0; i < extent->l1_size; i++) {
494 le32_to_cpus(&extent->l1_backup_table[i]);
499 g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE);
502 g_free(extent->l1_backup_table);
504 g_free(extent->l1_table);
508 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
509 BlockDriverState *file,
510 int flags, Error **errp)
517 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
519 error_setg_errno(errp, -ret,
520 "Could not read header from file '%s'",
524 ret = vmdk_add_extent(bs, file, false,
525 le32_to_cpu(header.disk_sectors),
526 le32_to_cpu(header.l1dir_offset) << 9,
528 le32_to_cpu(header.l1dir_size),
530 le32_to_cpu(header.granularity),
536 ret = vmdk_init_tables(bs, extent, errp);
538 /* free extent allocated by vmdk_add_extent */
539 vmdk_free_last_extent(bs);
544 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
547 static char *vmdk_read_desc(BlockDriverState *file, uint64_t desc_offset,
554 size = bdrv_getlength(file);
556 error_setg_errno(errp, -size, "Could not access file");
561 /* Both descriptor file and sparse image must be much larger than 4
562 * bytes, also callers of vmdk_read_desc want to compare the first 4
563 * bytes with VMDK4_MAGIC, let's error out if less is read. */
564 error_setg(errp, "File is too small, not a valid image");
568 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */
569 buf = g_malloc(size + 1);
571 ret = bdrv_pread(file, desc_offset, buf, size);
573 error_setg_errno(errp, -ret, "Could not read from file");
582 static int vmdk_open_vmdk4(BlockDriverState *bs,
583 BlockDriverState *file,
584 int flags, Error **errp)
588 uint32_t l1_size, l1_entry_sectors;
591 BDRVVmdkState *s = bs->opaque;
592 int64_t l1_backup_offset = 0;
594 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
596 error_setg_errno(errp, -ret,
597 "Could not read header from file '%s'",
601 if (header.capacity == 0) {
602 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
604 char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
608 ret = vmdk_open_desc_file(bs, flags, buf, errp);
614 if (!s->create_type) {
615 s->create_type = g_strdup("monolithicSparse");
618 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
620 * The footer takes precedence over the header, so read it in. The
621 * footer starts at offset -1024 from the end: One sector for the
622 * footer, and another one for the end-of-stream marker.
629 uint8_t pad[512 - 16];
630 } QEMU_PACKED footer_marker;
634 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
640 uint8_t pad[512 - 16];
641 } QEMU_PACKED eos_marker;
642 } QEMU_PACKED footer;
644 ret = bdrv_pread(file,
645 bs->file->total_sectors * 512 - 1536,
646 &footer, sizeof(footer));
651 /* Some sanity checks for the footer */
652 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
653 le32_to_cpu(footer.footer_marker.size) != 0 ||
654 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
655 le64_to_cpu(footer.eos_marker.val) != 0 ||
656 le32_to_cpu(footer.eos_marker.size) != 0 ||
657 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
662 header = footer.header;
665 if (le32_to_cpu(header.version) > 3) {
667 snprintf(buf, sizeof(buf), "VMDK version %" PRId32,
668 le32_to_cpu(header.version));
669 error_set(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
670 bdrv_get_device_name(bs), "vmdk", buf);
672 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR)) {
673 /* VMware KB 2064959 explains that version 3 added support for
674 * persistent changed block tracking (CBT), and backup software can
675 * read it as version=1 if it doesn't care about the changed area
676 * information. So we are safe to enable read only. */
677 error_setg(errp, "VMDK version 3 must be read only");
681 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
682 error_setg(errp, "L2 table size too big");
686 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
687 * le64_to_cpu(header.granularity);
688 if (l1_entry_sectors == 0) {
691 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
693 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
694 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
696 if (bdrv_nb_sectors(file) < le64_to_cpu(header.grain_offset)) {
697 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
698 (int64_t)(le64_to_cpu(header.grain_offset)
699 * BDRV_SECTOR_SIZE));
703 ret = vmdk_add_extent(bs, file, false,
704 le64_to_cpu(header.capacity),
705 le64_to_cpu(header.gd_offset) << 9,
708 le32_to_cpu(header.num_gtes_per_gt),
709 le64_to_cpu(header.granularity),
716 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
717 if (extent->compressed) {
718 g_free(s->create_type);
719 s->create_type = g_strdup("streamOptimized");
721 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
722 extent->version = le32_to_cpu(header.version);
723 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
724 ret = vmdk_init_tables(bs, extent, errp);
726 /* free extent allocated by vmdk_add_extent */
727 vmdk_free_last_extent(bs);
732 /* find an option value out of descriptor file */
733 static int vmdk_parse_description(const char *desc, const char *opt_name,
734 char *buf, int buf_size)
736 char *opt_pos, *opt_end;
737 const char *end = desc + strlen(desc);
739 opt_pos = strstr(desc, opt_name);
743 /* Skip "=\"" following opt_name */
744 opt_pos += strlen(opt_name) + 2;
745 if (opt_pos >= end) {
749 while (opt_end < end && *opt_end != '"') {
752 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
755 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
759 /* Open an extent file and append to bs array */
760 static int vmdk_open_sparse(BlockDriverState *bs,
761 BlockDriverState *file, int flags,
762 char *buf, Error **errp)
766 magic = ldl_be_p(buf);
769 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
772 return vmdk_open_vmdk4(bs, file, flags, errp);
775 error_setg(errp, "Image not in VMDK format");
781 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
782 const char *desc_file_path, Error **errp)
788 const char *p = desc;
791 char extent_path[PATH_MAX];
792 BlockDriverState *extent_file;
793 BDRVVmdkState *s = bs->opaque;
797 /* parse extent line in one of below formats:
799 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
800 * RW [size in sectors] SPARSE "file-name.vmdk"
801 * RW [size in sectors] VMFS "file-name.vmdk"
802 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
805 ret = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
806 access, §ors, type, fname, &flat_offset);
807 if (ret < 4 || strcmp(access, "RW")) {
809 } else if (!strcmp(type, "FLAT")) {
810 if (ret != 5 || flat_offset < 0) {
811 error_setg(errp, "Invalid extent lines: \n%s", p);
814 } else if (!strcmp(type, "VMFS")) {
818 error_setg(errp, "Invalid extent lines:\n%s", p);
821 } else if (ret != 4) {
822 error_setg(errp, "Invalid extent lines:\n%s", p);
827 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
828 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
829 (strcmp(access, "RW"))) {
833 path_combine(extent_path, sizeof(extent_path),
834 desc_file_path, fname);
836 ret = bdrv_open(&extent_file, extent_path, NULL, NULL,
837 bs->open_flags | BDRV_O_PROTOCOL, NULL, errp);
842 /* save to extents array */
843 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
846 ret = vmdk_add_extent(bs, extent_file, true, sectors,
847 0, 0, 0, 0, 0, &extent, errp);
849 bdrv_unref(extent_file);
852 extent->flat_start_offset = flat_offset << 9;
853 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
854 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
855 char *buf = vmdk_read_desc(extent_file, 0, errp);
859 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf, errp);
863 bdrv_unref(extent_file);
866 extent = &s->extents[s->num_extents - 1];
868 error_setg(errp, "Unsupported extent type '%s'", type);
869 bdrv_unref(extent_file);
872 extent->type = g_strdup(type);
874 /* move to next line */
886 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
891 BDRVVmdkState *s = bs->opaque;
893 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
894 error_setg(errp, "invalid VMDK image descriptor");
898 if (strcmp(ct, "monolithicFlat") &&
899 strcmp(ct, "vmfs") &&
900 strcmp(ct, "vmfsSparse") &&
901 strcmp(ct, "twoGbMaxExtentSparse") &&
902 strcmp(ct, "twoGbMaxExtentFlat")) {
903 error_setg(errp, "Unsupported image type '%s'", ct);
907 s->create_type = g_strdup(ct);
909 ret = vmdk_parse_extents(buf, bs, bs->file->filename, errp);
914 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
919 BDRVVmdkState *s = bs->opaque;
922 buf = vmdk_read_desc(bs->file, 0, errp);
927 magic = ldl_be_p(buf);
931 ret = vmdk_open_sparse(bs, bs->file, flags, buf, errp);
932 s->desc_offset = 0x200;
935 ret = vmdk_open_desc_file(bs, flags, buf, errp);
942 /* try to open parent images, if exist */
943 ret = vmdk_parent_open(bs);
947 s->cid = vmdk_read_cid(bs, 0);
948 s->parent_cid = vmdk_read_cid(bs, 1);
949 qemu_co_mutex_init(&s->lock);
951 /* Disable migration when VMDK images are used */
952 error_set(&s->migration_blocker,
953 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
954 "vmdk", bdrv_get_device_name(bs), "live migration");
955 migrate_add_blocker(s->migration_blocker);
961 g_free(s->create_type);
962 s->create_type = NULL;
963 vmdk_free_extents(bs);
968 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
970 BDRVVmdkState *s = bs->opaque;
973 for (i = 0; i < s->num_extents; i++) {
974 if (!s->extents[i].flat) {
975 bs->bl.write_zeroes_alignment =
976 MAX(bs->bl.write_zeroes_alignment,
977 s->extents[i].cluster_sectors);
985 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
986 * to the cluster at @cluster_sector_num.
988 * If @skip_start_sector < @skip_end_sector, the relative range
989 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
990 * it for call to write user data in the request.
992 static int get_whole_cluster(BlockDriverState *bs,
994 uint64_t cluster_sector_num,
996 uint64_t skip_start_sector,
997 uint64_t skip_end_sector)
1000 int64_t cluster_bytes;
1001 uint8_t *whole_grain;
1003 /* For COW, align request sector_num to cluster start */
1004 sector_num = QEMU_ALIGN_DOWN(sector_num, extent->cluster_sectors);
1005 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1006 whole_grain = qemu_blockalign(bs, cluster_bytes);
1008 if (!bs->backing_hd) {
1009 memset(whole_grain, 0, skip_start_sector << BDRV_SECTOR_BITS);
1010 memset(whole_grain + (skip_end_sector << BDRV_SECTOR_BITS), 0,
1011 cluster_bytes - (skip_end_sector << BDRV_SECTOR_BITS));
1014 assert(skip_end_sector <= extent->cluster_sectors);
1015 /* we will be here if it's first write on non-exist grain(cluster).
1016 * try to read from parent image, if exist */
1017 if (bs->backing_hd && !vmdk_is_cid_valid(bs)) {
1022 /* Read backing data before skip range */
1023 if (skip_start_sector > 0) {
1024 if (bs->backing_hd) {
1025 ret = bdrv_read(bs->backing_hd, sector_num,
1026 whole_grain, skip_start_sector);
1032 ret = bdrv_write(extent->file, cluster_sector_num, whole_grain,
1039 /* Read backing data after skip range */
1040 if (skip_end_sector < extent->cluster_sectors) {
1041 if (bs->backing_hd) {
1042 ret = bdrv_read(bs->backing_hd, sector_num + skip_end_sector,
1043 whole_grain + (skip_end_sector << BDRV_SECTOR_BITS),
1044 extent->cluster_sectors - skip_end_sector);
1050 ret = bdrv_write(extent->file, cluster_sector_num + skip_end_sector,
1051 whole_grain + (skip_end_sector << BDRV_SECTOR_BITS),
1052 extent->cluster_sectors - skip_end_sector);
1060 qemu_vfree(whole_grain);
1064 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1067 offset = cpu_to_le32(offset);
1068 /* update L2 table */
1069 if (bdrv_pwrite_sync(
1071 ((int64_t)m_data->l2_offset * 512)
1072 + (m_data->l2_index * sizeof(offset)),
1073 &offset, sizeof(offset)) < 0) {
1076 /* update backup L2 table */
1077 if (extent->l1_backup_table_offset != 0) {
1078 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
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) {
1087 if (m_data->l2_cache_entry) {
1088 *m_data->l2_cache_entry = offset;
1095 * get_cluster_offset
1097 * Look up cluster offset in extent file by sector number, and store in
1100 * For flat extents, the start offset as parsed from the description file is
1103 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1104 * offset for a new cluster and update L2 cache. If there is a backing file,
1105 * COW is done before returning; otherwise, zeroes are written to the allocated
1106 * cluster. Both COW and zero writing skips the sector range
1107 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1108 * has new data to write there.
1110 * Returns: VMDK_OK if cluster exists and mapped in the image.
1111 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1112 * VMDK_ERROR if failed.
1114 static int get_cluster_offset(BlockDriverState *bs,
1116 VmdkMetaData *m_data,
1119 uint64_t *cluster_offset,
1120 uint64_t skip_start_sector,
1121 uint64_t skip_end_sector)
1123 unsigned int l1_index, l2_offset, l2_index;
1124 int min_index, i, j;
1125 uint32_t min_count, *l2_table;
1126 bool zeroed = false;
1128 int64_t cluster_sector;
1134 *cluster_offset = extent->flat_start_offset;
1138 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1139 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1140 if (l1_index >= extent->l1_size) {
1143 l2_offset = extent->l1_table[l1_index];
1145 return VMDK_UNALLOC;
1147 for (i = 0; i < L2_CACHE_SIZE; i++) {
1148 if (l2_offset == extent->l2_cache_offsets[i]) {
1149 /* increment the hit count */
1150 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1151 for (j = 0; j < L2_CACHE_SIZE; j++) {
1152 extent->l2_cache_counts[j] >>= 1;
1155 l2_table = extent->l2_cache + (i * extent->l2_size);
1159 /* not found: load a new entry in the least used one */
1161 min_count = 0xffffffff;
1162 for (i = 0; i < L2_CACHE_SIZE; i++) {
1163 if (extent->l2_cache_counts[i] < min_count) {
1164 min_count = extent->l2_cache_counts[i];
1168 l2_table = extent->l2_cache + (min_index * extent->l2_size);
1171 (int64_t)l2_offset * 512,
1173 extent->l2_size * sizeof(uint32_t)
1174 ) != extent->l2_size * sizeof(uint32_t)) {
1178 extent->l2_cache_offsets[min_index] = l2_offset;
1179 extent->l2_cache_counts[min_index] = 1;
1181 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1182 cluster_sector = le32_to_cpu(l2_table[l2_index]);
1186 m_data->l1_index = l1_index;
1187 m_data->l2_index = l2_index;
1188 m_data->l2_offset = l2_offset;
1189 m_data->l2_cache_entry = &l2_table[l2_index];
1191 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1195 if (!cluster_sector || zeroed) {
1197 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1200 cluster_sector = extent->next_cluster_sector;
1201 extent->next_cluster_sector += extent->cluster_sectors;
1203 /* First of all we write grain itself, to avoid race condition
1204 * that may to corrupt the image.
1205 * This problem may occur because of insufficient space on host disk
1206 * or inappropriate VM shutdown.
1208 ret = get_whole_cluster(bs, extent,
1210 offset >> BDRV_SECTOR_BITS,
1211 skip_start_sector, skip_end_sector);
1216 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1220 static VmdkExtent *find_extent(BDRVVmdkState *s,
1221 int64_t sector_num, VmdkExtent *start_hint)
1223 VmdkExtent *extent = start_hint;
1226 extent = &s->extents[0];
1228 while (extent < &s->extents[s->num_extents]) {
1229 if (sector_num < extent->end_sector) {
1237 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs,
1238 int64_t sector_num, int nb_sectors, int *pnum)
1240 BDRVVmdkState *s = bs->opaque;
1241 int64_t index_in_cluster, n, ret;
1245 extent = find_extent(s, sector_num, NULL);
1249 qemu_co_mutex_lock(&s->lock);
1250 ret = get_cluster_offset(bs, extent, NULL,
1251 sector_num * 512, false, &offset,
1253 qemu_co_mutex_unlock(&s->lock);
1263 ret = BDRV_BLOCK_ZERO;
1266 ret = BDRV_BLOCK_DATA;
1267 if (extent->file == bs->file && !extent->compressed) {
1268 ret |= BDRV_BLOCK_OFFSET_VALID | offset;
1274 index_in_cluster = sector_num % extent->cluster_sectors;
1275 n = extent->cluster_sectors - index_in_cluster;
1276 if (n > nb_sectors) {
1283 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1284 int64_t offset_in_cluster, const uint8_t *buf,
1285 int nb_sectors, int64_t sector_num)
1288 VmdkGrainMarker *data = NULL;
1290 const uint8_t *write_buf = buf;
1291 int write_len = nb_sectors * 512;
1293 if (extent->compressed) {
1294 if (!extent->has_marker) {
1298 buf_len = (extent->cluster_sectors << 9) * 2;
1299 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1300 if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK ||
1305 data->lba = sector_num;
1306 data->size = buf_len;
1307 write_buf = (uint8_t *)data;
1308 write_len = buf_len + sizeof(VmdkGrainMarker);
1310 ret = bdrv_pwrite(extent->file,
1311 cluster_offset + offset_in_cluster,
1314 if (ret != write_len) {
1315 ret = ret < 0 ? ret : -EIO;
1324 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1325 int64_t offset_in_cluster, uint8_t *buf,
1329 int cluster_bytes, buf_bytes;
1330 uint8_t *cluster_buf, *compressed_data;
1331 uint8_t *uncomp_buf;
1333 VmdkGrainMarker *marker;
1337 if (!extent->compressed) {
1338 ret = bdrv_pread(extent->file,
1339 cluster_offset + offset_in_cluster,
1340 buf, nb_sectors * 512);
1341 if (ret == nb_sectors * 512) {
1347 cluster_bytes = extent->cluster_sectors * 512;
1348 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1349 buf_bytes = cluster_bytes * 2;
1350 cluster_buf = g_malloc(buf_bytes);
1351 uncomp_buf = g_malloc(cluster_bytes);
1352 ret = bdrv_pread(extent->file,
1354 cluster_buf, buf_bytes);
1358 compressed_data = cluster_buf;
1359 buf_len = cluster_bytes;
1360 data_len = cluster_bytes;
1361 if (extent->has_marker) {
1362 marker = (VmdkGrainMarker *)cluster_buf;
1363 compressed_data = marker->data;
1364 data_len = le32_to_cpu(marker->size);
1366 if (!data_len || data_len > buf_bytes) {
1370 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1376 if (offset_in_cluster < 0 ||
1377 offset_in_cluster + nb_sectors * 512 > buf_len) {
1381 memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512);
1386 g_free(cluster_buf);
1390 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
1391 uint8_t *buf, int nb_sectors)
1393 BDRVVmdkState *s = bs->opaque;
1395 uint64_t n, index_in_cluster;
1396 uint64_t extent_begin_sector, extent_relative_sector_num;
1397 VmdkExtent *extent = NULL;
1398 uint64_t cluster_offset;
1400 while (nb_sectors > 0) {
1401 extent = find_extent(s, sector_num, extent);
1405 ret = get_cluster_offset(bs, extent, NULL,
1406 sector_num << 9, false, &cluster_offset,
1408 extent_begin_sector = extent->end_sector - extent->sectors;
1409 extent_relative_sector_num = sector_num - extent_begin_sector;
1410 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1411 n = extent->cluster_sectors - index_in_cluster;
1412 if (n > nb_sectors) {
1415 if (ret != VMDK_OK) {
1416 /* if not allocated, try to read from parent image, if exist */
1417 if (bs->backing_hd && ret != VMDK_ZEROED) {
1418 if (!vmdk_is_cid_valid(bs)) {
1421 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
1426 memset(buf, 0, 512 * n);
1429 ret = vmdk_read_extent(extent,
1430 cluster_offset, index_in_cluster * 512,
1443 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num,
1444 uint8_t *buf, int nb_sectors)
1447 BDRVVmdkState *s = bs->opaque;
1448 qemu_co_mutex_lock(&s->lock);
1449 ret = vmdk_read(bs, sector_num, buf, nb_sectors);
1450 qemu_co_mutex_unlock(&s->lock);
1456 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1457 * if possible, otherwise return -ENOTSUP.
1458 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1459 * with each cluster. By dry run we can find if the zero write
1460 * is possible without modifying image data.
1462 * Returns: error code with 0 for success.
1464 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
1465 const uint8_t *buf, int nb_sectors,
1466 bool zeroed, bool zero_dry_run)
1468 BDRVVmdkState *s = bs->opaque;
1469 VmdkExtent *extent = NULL;
1471 int64_t index_in_cluster, n;
1472 uint64_t extent_begin_sector, extent_relative_sector_num;
1473 uint64_t cluster_offset;
1474 VmdkMetaData m_data;
1476 if (sector_num > bs->total_sectors) {
1477 error_report("Wrong offset: sector_num=0x%" PRIx64
1478 " total_sectors=0x%" PRIx64 "\n",
1479 sector_num, bs->total_sectors);
1483 while (nb_sectors > 0) {
1484 extent = find_extent(s, sector_num, extent);
1488 extent_begin_sector = extent->end_sector - extent->sectors;
1489 extent_relative_sector_num = sector_num - extent_begin_sector;
1490 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1491 n = extent->cluster_sectors - index_in_cluster;
1492 if (n > nb_sectors) {
1495 ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9,
1496 !(extent->compressed || zeroed),
1498 index_in_cluster, index_in_cluster + n);
1499 if (extent->compressed) {
1500 if (ret == VMDK_OK) {
1501 /* Refuse write to allocated cluster for streamOptimized */
1502 error_report("Could not write to allocated cluster"
1503 " for streamOptimized");
1507 ret = get_cluster_offset(bs, extent, &m_data, sector_num << 9,
1508 true, &cluster_offset, 0, 0);
1511 if (ret == VMDK_ERROR) {
1515 /* Do zeroed write, buf is ignored */
1516 if (extent->has_zero_grain &&
1517 index_in_cluster == 0 &&
1518 n >= extent->cluster_sectors) {
1519 n = extent->cluster_sectors;
1520 if (!zero_dry_run) {
1521 /* update L2 tables */
1522 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1531 ret = vmdk_write_extent(extent,
1532 cluster_offset, index_in_cluster * 512,
1533 buf, n, sector_num);
1538 /* update L2 tables */
1539 if (vmdk_L2update(extent, &m_data,
1540 cluster_offset >> BDRV_SECTOR_BITS)
1550 /* update CID on the first write every time the virtual disk is
1552 if (!s->cid_updated) {
1553 ret = vmdk_write_cid(bs, g_random_int());
1557 s->cid_updated = true;
1563 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num,
1564 const uint8_t *buf, int nb_sectors)
1567 BDRVVmdkState *s = bs->opaque;
1568 qemu_co_mutex_lock(&s->lock);
1569 ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false);
1570 qemu_co_mutex_unlock(&s->lock);
1574 static int vmdk_write_compressed(BlockDriverState *bs,
1579 BDRVVmdkState *s = bs->opaque;
1580 if (s->num_extents == 1 && s->extents[0].compressed) {
1581 return vmdk_write(bs, sector_num, buf, nb_sectors, false, false);
1587 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs,
1590 BdrvRequestFlags flags)
1593 BDRVVmdkState *s = bs->opaque;
1594 qemu_co_mutex_lock(&s->lock);
1595 /* write zeroes could fail if sectors not aligned to cluster, test it with
1596 * dry_run == true before really updating image */
1597 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true);
1599 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false);
1601 qemu_co_mutex_unlock(&s->lock);
1605 static int vmdk_create_extent(const char *filename, int64_t filesize,
1606 bool flat, bool compress, bool zeroed_grain,
1607 QemuOpts *opts, Error **errp)
1610 BlockDriverState *bs = NULL;
1612 Error *local_err = NULL;
1613 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1614 uint32_t *gd_buf = NULL;
1617 ret = bdrv_create_file(filename, opts, &local_err);
1619 error_propagate(errp, local_err);
1624 ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1627 error_propagate(errp, local_err);
1632 ret = bdrv_truncate(bs, filesize);
1634 error_setg_errno(errp, -ret, "Could not truncate file");
1638 magic = cpu_to_be32(VMDK4_MAGIC);
1639 memset(&header, 0, sizeof(header));
1640 header.version = zeroed_grain ? 2 : 1;
1641 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1642 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1643 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1644 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1645 header.capacity = filesize / BDRV_SECTOR_SIZE;
1646 header.granularity = 128;
1647 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1649 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1650 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1652 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1653 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1655 header.desc_offset = 1;
1656 header.desc_size = 20;
1657 header.rgd_offset = header.desc_offset + header.desc_size;
1658 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1659 header.grain_offset =
1660 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1661 header.granularity);
1662 /* swap endianness for all header fields */
1663 header.version = cpu_to_le32(header.version);
1664 header.flags = cpu_to_le32(header.flags);
1665 header.capacity = cpu_to_le64(header.capacity);
1666 header.granularity = cpu_to_le64(header.granularity);
1667 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1668 header.desc_offset = cpu_to_le64(header.desc_offset);
1669 header.desc_size = cpu_to_le64(header.desc_size);
1670 header.rgd_offset = cpu_to_le64(header.rgd_offset);
1671 header.gd_offset = cpu_to_le64(header.gd_offset);
1672 header.grain_offset = cpu_to_le64(header.grain_offset);
1673 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1675 header.check_bytes[0] = 0xa;
1676 header.check_bytes[1] = 0x20;
1677 header.check_bytes[2] = 0xd;
1678 header.check_bytes[3] = 0xa;
1680 /* write all the data */
1681 ret = bdrv_pwrite(bs, 0, &magic, sizeof(magic));
1683 error_set(errp, QERR_IO_ERROR);
1686 ret = bdrv_pwrite(bs, sizeof(magic), &header, sizeof(header));
1688 error_set(errp, QERR_IO_ERROR);
1692 ret = bdrv_truncate(bs, le64_to_cpu(header.grain_offset) << 9);
1694 error_setg_errno(errp, -ret, "Could not truncate file");
1698 /* write grain directory */
1699 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1700 gd_buf = g_malloc0(gd_buf_size);
1701 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1702 i < gt_count; i++, tmp += gt_size) {
1703 gd_buf[i] = cpu_to_le32(tmp);
1705 ret = bdrv_pwrite(bs, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1706 gd_buf, gd_buf_size);
1708 error_set(errp, QERR_IO_ERROR);
1712 /* write backup grain directory */
1713 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1714 i < gt_count; i++, tmp += gt_size) {
1715 gd_buf[i] = cpu_to_le32(tmp);
1717 ret = bdrv_pwrite(bs, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1718 gd_buf, gd_buf_size);
1720 error_set(errp, QERR_IO_ERROR);
1733 static int filename_decompose(const char *filename, char *path, char *prefix,
1734 char *postfix, size_t buf_len, Error **errp)
1738 if (filename == NULL || !strlen(filename)) {
1739 error_setg(errp, "No filename provided");
1742 p = strrchr(filename, '/');
1744 p = strrchr(filename, '\\');
1747 p = strrchr(filename, ':');
1751 if (p - filename >= buf_len) {
1754 pstrcpy(path, p - filename + 1, filename);
1759 q = strrchr(p, '.');
1761 pstrcpy(prefix, buf_len, p);
1764 if (q - p >= buf_len) {
1767 pstrcpy(prefix, q - p + 1, p);
1768 pstrcpy(postfix, buf_len, q);
1773 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp)
1776 BlockDriverState *new_bs = NULL;
1777 Error *local_err = NULL;
1779 int64_t total_size = 0, filesize;
1780 char *adapter_type = NULL;
1781 char *backing_file = NULL;
1785 bool flat, split, compress;
1786 GString *ext_desc_lines;
1787 char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX];
1788 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
1789 const char *desc_extent_line;
1790 char parent_desc_line[BUF_SIZE] = "";
1791 uint32_t parent_cid = 0xffffffff;
1792 uint32_t number_heads = 16;
1793 bool zeroed_grain = false;
1794 uint32_t desc_offset = 0, desc_len;
1795 const char desc_template[] =
1796 "# Disk DescriptorFile\n"
1799 "parentCID=%" PRIx32 "\n"
1800 "createType=\"%s\"\n"
1803 "# Extent description\n"
1806 "# The Disk Data Base\n"
1809 "ddb.virtualHWVersion = \"%d\"\n"
1810 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1811 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
1812 "ddb.geometry.sectors = \"63\"\n"
1813 "ddb.adapterType = \"%s\"\n";
1815 ext_desc_lines = g_string_new(NULL);
1817 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
1821 /* Read out options */
1822 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1824 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
1825 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1826 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) {
1827 flags |= BLOCK_FLAG_COMPAT6;
1829 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1830 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) {
1831 zeroed_grain = true;
1834 if (!adapter_type) {
1835 adapter_type = g_strdup("ide");
1836 } else if (strcmp(adapter_type, "ide") &&
1837 strcmp(adapter_type, "buslogic") &&
1838 strcmp(adapter_type, "lsilogic") &&
1839 strcmp(adapter_type, "legacyESX")) {
1840 error_setg(errp, "Unknown adapter type: '%s'", adapter_type);
1844 if (strcmp(adapter_type, "ide") != 0) {
1845 /* that's the number of heads with which vmware operates when
1846 creating, exporting, etc. vmdk files with a non-ide adapter type */
1850 /* Default format to monolithicSparse */
1851 fmt = g_strdup("monolithicSparse");
1852 } else if (strcmp(fmt, "monolithicFlat") &&
1853 strcmp(fmt, "monolithicSparse") &&
1854 strcmp(fmt, "twoGbMaxExtentSparse") &&
1855 strcmp(fmt, "twoGbMaxExtentFlat") &&
1856 strcmp(fmt, "streamOptimized")) {
1857 error_setg(errp, "Unknown subformat: '%s'", fmt);
1861 split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1862 strcmp(fmt, "twoGbMaxExtentSparse"));
1863 flat = !(strcmp(fmt, "monolithicFlat") &&
1864 strcmp(fmt, "twoGbMaxExtentFlat"));
1865 compress = !strcmp(fmt, "streamOptimized");
1867 desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n";
1869 desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n";
1871 if (flat && backing_file) {
1872 error_setg(errp, "Flat image can't have backing file");
1876 if (flat && zeroed_grain) {
1877 error_setg(errp, "Flat image can't enable zeroed grain");
1882 BlockDriverState *bs = NULL;
1883 ret = bdrv_open(&bs, backing_file, NULL, NULL, BDRV_O_NO_BACKING, NULL,
1888 if (strcmp(bs->drv->format_name, "vmdk")) {
1893 parent_cid = vmdk_read_cid(bs, 0);
1895 snprintf(parent_desc_line, sizeof(parent_desc_line),
1896 "parentFileNameHint=\"%s\"", backing_file);
1899 /* Create extents */
1900 filesize = total_size;
1901 while (filesize > 0) {
1902 char desc_line[BUF_SIZE];
1903 char ext_filename[PATH_MAX];
1904 char desc_filename[PATH_MAX];
1905 int64_t size = filesize;
1907 if (split && size > split_size) {
1911 snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s",
1912 prefix, flat ? 'f' : 's', ++idx, postfix);
1914 snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s",
1917 snprintf(desc_filename, sizeof(desc_filename), "%s%s",
1920 snprintf(ext_filename, sizeof(ext_filename), "%s%s",
1921 path, desc_filename);
1923 if (vmdk_create_extent(ext_filename, size,
1924 flat, compress, zeroed_grain, opts, errp)) {
1930 /* Format description line */
1931 snprintf(desc_line, sizeof(desc_line),
1932 desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename);
1933 g_string_append(ext_desc_lines, desc_line);
1935 /* generate descriptor file */
1936 desc = g_strdup_printf(desc_template,
1941 ext_desc_lines->str,
1942 (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
1944 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
1947 desc_len = strlen(desc);
1948 /* the descriptor offset = 0x200 */
1949 if (!split && !flat) {
1950 desc_offset = 0x200;
1952 ret = bdrv_create_file(filename, opts, &local_err);
1954 error_propagate(errp, local_err);
1958 assert(new_bs == NULL);
1959 ret = bdrv_open(&new_bs, filename, NULL, NULL,
1960 BDRV_O_RDWR | BDRV_O_PROTOCOL, NULL, &local_err);
1962 error_propagate(errp, local_err);
1965 ret = bdrv_pwrite(new_bs, desc_offset, desc, desc_len);
1967 error_setg_errno(errp, -ret, "Could not write description");
1970 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
1971 * for description file */
1972 if (desc_offset == 0) {
1973 ret = bdrv_truncate(new_bs, desc_len);
1975 error_setg_errno(errp, -ret, "Could not truncate file");
1982 g_free(adapter_type);
1983 g_free(backing_file);
1986 g_string_free(ext_desc_lines, true);
1990 static void vmdk_close(BlockDriverState *bs)
1992 BDRVVmdkState *s = bs->opaque;
1994 vmdk_free_extents(bs);
1995 g_free(s->create_type);
1997 migrate_del_blocker(s->migration_blocker);
1998 error_free(s->migration_blocker);
2001 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2003 BDRVVmdkState *s = bs->opaque;
2007 for (i = 0; i < s->num_extents; i++) {
2008 err = bdrv_co_flush(s->extents[i].file);
2016 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2021 BDRVVmdkState *s = bs->opaque;
2023 ret = bdrv_get_allocated_file_size(bs->file);
2027 for (i = 0; i < s->num_extents; i++) {
2028 if (s->extents[i].file == bs->file) {
2031 r = bdrv_get_allocated_file_size(s->extents[i].file);
2040 static int vmdk_has_zero_init(BlockDriverState *bs)
2043 BDRVVmdkState *s = bs->opaque;
2045 /* If has a flat extent and its underlying storage doesn't have zero init,
2047 for (i = 0; i < s->num_extents; i++) {
2048 if (s->extents[i].flat) {
2049 if (!bdrv_has_zero_init(s->extents[i].file)) {
2057 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2059 ImageInfo *info = g_new0(ImageInfo, 1);
2061 *info = (ImageInfo){
2062 .filename = g_strdup(extent->file->filename),
2063 .format = g_strdup(extent->type),
2064 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2065 .compressed = extent->compressed,
2066 .has_compressed = extent->compressed,
2067 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2068 .has_cluster_size = !extent->flat,
2074 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result,
2077 BDRVVmdkState *s = bs->opaque;
2078 VmdkExtent *extent = NULL;
2079 int64_t sector_num = 0;
2080 int64_t total_sectors = bdrv_nb_sectors(bs);
2082 uint64_t cluster_offset;
2089 if (sector_num >= total_sectors) {
2092 extent = find_extent(s, sector_num, extent);
2095 "ERROR: could not find extent for sector %" PRId64 "\n",
2099 ret = get_cluster_offset(bs, extent, NULL,
2100 sector_num << BDRV_SECTOR_BITS,
2101 false, &cluster_offset, 0, 0);
2102 if (ret == VMDK_ERROR) {
2104 "ERROR: could not get cluster_offset for sector %"
2105 PRId64 "\n", sector_num);
2108 if (ret == VMDK_OK && cluster_offset >= bdrv_getlength(extent->file)) {
2110 "ERROR: cluster offset for sector %"
2111 PRId64 " points after EOF\n", sector_num);
2114 sector_num += extent->cluster_sectors;
2117 result->corruptions++;
2121 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs)
2124 BDRVVmdkState *s = bs->opaque;
2125 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2126 ImageInfoList **next;
2128 *spec_info = (ImageInfoSpecific){
2129 .kind = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2131 .vmdk = g_new0(ImageInfoSpecificVmdk, 1),
2135 *spec_info->vmdk = (ImageInfoSpecificVmdk) {
2136 .create_type = g_strdup(s->create_type),
2138 .parent_cid = s->parent_cid,
2141 next = &spec_info->vmdk->extents;
2142 for (i = 0; i < s->num_extents; i++) {
2143 *next = g_new0(ImageInfoList, 1);
2144 (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2145 (*next)->next = NULL;
2146 next = &(*next)->next;
2152 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2154 return a->flat == b->flat &&
2155 a->compressed == b->compressed &&
2156 (a->flat || a->cluster_sectors == b->cluster_sectors);
2159 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2162 BDRVVmdkState *s = bs->opaque;
2163 assert(s->num_extents);
2165 /* See if we have multiple extents but they have different cases */
2166 for (i = 1; i < s->num_extents; i++) {
2167 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2171 bdi->needs_compressed_writes = s->extents[0].compressed;
2172 if (!s->extents[0].flat) {
2173 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2178 static void vmdk_detach_aio_context(BlockDriverState *bs)
2180 BDRVVmdkState *s = bs->opaque;
2183 for (i = 0; i < s->num_extents; i++) {
2184 bdrv_detach_aio_context(s->extents[i].file);
2188 static void vmdk_attach_aio_context(BlockDriverState *bs,
2189 AioContext *new_context)
2191 BDRVVmdkState *s = bs->opaque;
2194 for (i = 0; i < s->num_extents; i++) {
2195 bdrv_attach_aio_context(s->extents[i].file, new_context);
2199 static QemuOptsList vmdk_create_opts = {
2200 .name = "vmdk-create-opts",
2201 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2204 .name = BLOCK_OPT_SIZE,
2205 .type = QEMU_OPT_SIZE,
2206 .help = "Virtual disk size"
2209 .name = BLOCK_OPT_ADAPTER_TYPE,
2210 .type = QEMU_OPT_STRING,
2211 .help = "Virtual adapter type, can be one of "
2212 "ide (default), lsilogic, buslogic or legacyESX"
2215 .name = BLOCK_OPT_BACKING_FILE,
2216 .type = QEMU_OPT_STRING,
2217 .help = "File name of a base image"
2220 .name = BLOCK_OPT_COMPAT6,
2221 .type = QEMU_OPT_BOOL,
2222 .help = "VMDK version 6 image",
2223 .def_value_str = "off"
2226 .name = BLOCK_OPT_SUBFMT,
2227 .type = QEMU_OPT_STRING,
2229 "VMDK flat extent format, can be one of "
2230 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2233 .name = BLOCK_OPT_ZEROED_GRAIN,
2234 .type = QEMU_OPT_BOOL,
2235 .help = "Enable efficient zero writes "
2236 "using the zeroed-grain GTE feature"
2238 { /* end of list */ }
2242 static BlockDriver bdrv_vmdk = {
2243 .format_name = "vmdk",
2244 .instance_size = sizeof(BDRVVmdkState),
2245 .bdrv_probe = vmdk_probe,
2246 .bdrv_open = vmdk_open,
2247 .bdrv_check = vmdk_check,
2248 .bdrv_reopen_prepare = vmdk_reopen_prepare,
2249 .bdrv_read = vmdk_co_read,
2250 .bdrv_write = vmdk_co_write,
2251 .bdrv_write_compressed = vmdk_write_compressed,
2252 .bdrv_co_write_zeroes = vmdk_co_write_zeroes,
2253 .bdrv_close = vmdk_close,
2254 .bdrv_create = vmdk_create,
2255 .bdrv_co_flush_to_disk = vmdk_co_flush,
2256 .bdrv_co_get_block_status = vmdk_co_get_block_status,
2257 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2258 .bdrv_has_zero_init = vmdk_has_zero_init,
2259 .bdrv_get_specific_info = vmdk_get_specific_info,
2260 .bdrv_refresh_limits = vmdk_refresh_limits,
2261 .bdrv_get_info = vmdk_get_info,
2262 .bdrv_detach_aio_context = vmdk_detach_aio_context,
2263 .bdrv_attach_aio_context = vmdk_attach_aio_context,
2265 .supports_backing = true,
2266 .create_opts = &vmdk_create_opts,
2269 static void bdrv_vmdk_init(void)
2271 bdrv_register(&bdrv_vmdk);
2274 block_init(bdrv_vmdk_init);