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"
32 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
33 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
34 #define VMDK4_COMPRESSION_DEFLATE 1
35 #define VMDK4_FLAG_RGD (1 << 1)
36 #define VMDK4_FLAG_COMPRESS (1 << 16)
37 #define VMDK4_FLAG_MARKER (1 << 17)
38 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
43 uint32_t disk_sectors;
45 uint32_t l1dir_offset;
47 uint32_t file_sectors;
50 uint32_t sectors_per_track;
60 int32_t num_gtes_per_gte;
66 uint16_t compressAlgorithm;
67 } QEMU_PACKED VMDK4Header;
69 #define L2_CACHE_SIZE 16
71 typedef struct VmdkExtent {
72 BlockDriverState *file;
78 int64_t flat_start_offset;
79 int64_t l1_table_offset;
80 int64_t l1_backup_table_offset;
82 uint32_t *l1_backup_table;
84 uint32_t l1_entry_sectors;
88 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
89 uint32_t l2_cache_counts[L2_CACHE_SIZE];
91 unsigned int cluster_sectors;
94 typedef struct BDRVVmdkState {
100 /* Extent array with num_extents entries, ascend ordered by address */
102 Error *migration_blocker;
105 typedef struct VmdkMetaData {
107 unsigned int l1_index;
108 unsigned int l2_index;
109 unsigned int l2_offset;
113 typedef struct VmdkGrainMarker {
120 MARKER_END_OF_STREAM = 0,
121 MARKER_GRAIN_TABLE = 1,
122 MARKER_GRAIN_DIRECTORY = 2,
126 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
133 magic = be32_to_cpu(*(uint32_t *)buf);
134 if (magic == VMDK3_MAGIC ||
135 magic == VMDK4_MAGIC) {
138 const char *p = (const char *)buf;
139 const char *end = p + buf_size;
142 /* skip comment line */
143 while (p < end && *p != '\n') {
150 while (p < end && *p == ' ') {
153 /* skip '\r' if windows line endings used. */
154 if (p < end && *p == '\r') {
157 /* only accept blank lines before 'version=' line */
158 if (p == end || *p != '\n') {
164 if (end - p >= strlen("version=X\n")) {
165 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
166 strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
170 if (end - p >= strlen("version=X\r\n")) {
171 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
172 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
184 #define SECTOR_SIZE 512
185 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
186 #define BUF_SIZE 4096
187 #define HEADER_SIZE 512 /* first sector of 512 bytes */
189 static void vmdk_free_extents(BlockDriverState *bs)
192 BDRVVmdkState *s = bs->opaque;
195 for (i = 0; i < s->num_extents; i++) {
199 g_free(e->l1_backup_table);
200 if (e->file != bs->file) {
201 bdrv_delete(e->file);
207 static void vmdk_free_last_extent(BlockDriverState *bs)
209 BDRVVmdkState *s = bs->opaque;
211 if (s->num_extents == 0) {
215 s->extents = g_realloc(s->extents, s->num_extents * sizeof(VmdkExtent));
218 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
220 char desc[DESC_SIZE];
221 uint32_t cid = 0xffffffff;
222 const char *p_name, *cid_str;
224 BDRVVmdkState *s = bs->opaque;
227 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
233 cid_str = "parentCID";
234 cid_str_size = sizeof("parentCID");
237 cid_str_size = sizeof("CID");
240 desc[DESC_SIZE - 1] = '\0';
241 p_name = strstr(desc, cid_str);
242 if (p_name != NULL) {
243 p_name += cid_str_size;
244 sscanf(p_name, "%x", &cid);
250 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
252 char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
253 char *p_name, *tmp_str;
254 BDRVVmdkState *s = bs->opaque;
257 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
262 desc[DESC_SIZE - 1] = '\0';
263 tmp_str = strstr(desc, "parentCID");
264 if (tmp_str == NULL) {
268 pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
269 p_name = strstr(desc, "CID");
270 if (p_name != NULL) {
271 p_name += sizeof("CID");
272 snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
273 pstrcat(desc, sizeof(desc), tmp_desc);
276 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
284 static int vmdk_is_cid_valid(BlockDriverState *bs)
287 BDRVVmdkState *s = bs->opaque;
288 BlockDriverState *p_bs = bs->backing_hd;
292 cur_pcid = vmdk_read_cid(p_bs, 0);
293 if (s->parent_cid != cur_pcid) {
303 /* Queue extents, if any, for reopen() */
304 static int vmdk_reopen_prepare(BDRVReopenState *state,
305 BlockReopenQueue *queue, Error **errp)
312 assert(state != NULL);
313 assert(state->bs != NULL);
316 error_set(errp, ERROR_CLASS_GENERIC_ERROR,
317 "No reopen queue for VMDK extents");
321 s = state->bs->opaque;
325 for (i = 0; i < s->num_extents; i++) {
327 if (e->file != state->bs->file) {
328 bdrv_reopen_queue(queue, e->file, state->flags);
337 static int vmdk_parent_open(BlockDriverState *bs)
340 char desc[DESC_SIZE + 1];
341 BDRVVmdkState *s = bs->opaque;
344 desc[DESC_SIZE] = '\0';
345 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
350 p_name = strstr(desc, "parentFileNameHint");
351 if (p_name != NULL) {
354 p_name += sizeof("parentFileNameHint") + 1;
355 end_name = strchr(p_name, '\"');
356 if (end_name == NULL) {
359 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
363 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
369 /* Create and append extent to the extent array. Return the added VmdkExtent
370 * address. return NULL if allocation failed. */
371 static VmdkExtent *vmdk_add_extent(BlockDriverState *bs,
372 BlockDriverState *file, bool flat, int64_t sectors,
373 int64_t l1_offset, int64_t l1_backup_offset,
375 int l2_size, unsigned int cluster_sectors)
378 BDRVVmdkState *s = bs->opaque;
380 s->extents = g_realloc(s->extents,
381 (s->num_extents + 1) * sizeof(VmdkExtent));
382 extent = &s->extents[s->num_extents];
385 memset(extent, 0, sizeof(VmdkExtent));
388 extent->sectors = sectors;
389 extent->l1_table_offset = l1_offset;
390 extent->l1_backup_table_offset = l1_backup_offset;
391 extent->l1_size = l1_size;
392 extent->l1_entry_sectors = l2_size * cluster_sectors;
393 extent->l2_size = l2_size;
394 extent->cluster_sectors = cluster_sectors;
396 if (s->num_extents > 1) {
397 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
399 extent->end_sector = extent->sectors;
401 bs->total_sectors = extent->end_sector;
405 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent)
410 /* read the L1 table */
411 l1_size = extent->l1_size * sizeof(uint32_t);
412 extent->l1_table = g_malloc(l1_size);
413 ret = bdrv_pread(extent->file,
414 extent->l1_table_offset,
420 for (i = 0; i < extent->l1_size; i++) {
421 le32_to_cpus(&extent->l1_table[i]);
424 if (extent->l1_backup_table_offset) {
425 extent->l1_backup_table = g_malloc(l1_size);
426 ret = bdrv_pread(extent->file,
427 extent->l1_backup_table_offset,
428 extent->l1_backup_table,
433 for (i = 0; i < extent->l1_size; i++) {
434 le32_to_cpus(&extent->l1_backup_table[i]);
439 g_malloc(extent->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
442 g_free(extent->l1_backup_table);
444 g_free(extent->l1_table);
448 static int vmdk_open_vmdk3(BlockDriverState *bs,
449 BlockDriverState *file,
457 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
461 extent = vmdk_add_extent(bs,
463 le32_to_cpu(header.disk_sectors),
464 le32_to_cpu(header.l1dir_offset) << 9,
466 le32_to_cpu(header.granularity));
467 ret = vmdk_init_tables(bs, extent);
469 /* free extent allocated by vmdk_add_extent */
470 vmdk_free_last_extent(bs);
475 static int vmdk_open_desc_file(BlockDriverState *bs, int flags,
476 int64_t desc_offset);
478 static int vmdk_open_vmdk4(BlockDriverState *bs,
479 BlockDriverState *file,
484 uint32_t l1_size, l1_entry_sectors;
487 int64_t l1_backup_offset = 0;
489 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
493 if (header.capacity == 0 && header.desc_offset) {
494 return vmdk_open_desc_file(bs, flags, header.desc_offset << 9);
497 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
499 * The footer takes precedence over the header, so read it in. The
500 * footer starts at offset -1024 from the end: One sector for the
501 * footer, and another one for the end-of-stream marker.
508 uint8_t pad[512 - 16];
509 } QEMU_PACKED footer_marker;
513 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
519 uint8_t pad[512 - 16];
520 } QEMU_PACKED eos_marker;
521 } QEMU_PACKED footer;
523 ret = bdrv_pread(file,
524 bs->file->total_sectors * 512 - 1536,
525 &footer, sizeof(footer));
530 /* Some sanity checks for the footer */
531 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
532 le32_to_cpu(footer.footer_marker.size) != 0 ||
533 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
534 le64_to_cpu(footer.eos_marker.val) != 0 ||
535 le32_to_cpu(footer.eos_marker.size) != 0 ||
536 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
541 header = footer.header;
544 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gte)
545 * le64_to_cpu(header.granularity);
546 if (l1_entry_sectors == 0) {
549 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
551 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
552 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
554 extent = vmdk_add_extent(bs, file, false,
555 le64_to_cpu(header.capacity),
556 le64_to_cpu(header.gd_offset) << 9,
559 le32_to_cpu(header.num_gtes_per_gte),
560 le64_to_cpu(header.granularity));
562 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
563 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
564 ret = vmdk_init_tables(bs, extent);
566 /* free extent allocated by vmdk_add_extent */
567 vmdk_free_last_extent(bs);
572 /* find an option value out of descriptor file */
573 static int vmdk_parse_description(const char *desc, const char *opt_name,
574 char *buf, int buf_size)
576 char *opt_pos, *opt_end;
577 const char *end = desc + strlen(desc);
579 opt_pos = strstr(desc, opt_name);
583 /* Skip "=\"" following opt_name */
584 opt_pos += strlen(opt_name) + 2;
585 if (opt_pos >= end) {
589 while (opt_end < end && *opt_end != '"') {
592 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
595 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
599 /* Open an extent file and append to bs array */
600 static int vmdk_open_sparse(BlockDriverState *bs,
601 BlockDriverState *file,
606 if (bdrv_pread(file, 0, &magic, sizeof(magic)) != sizeof(magic)) {
610 magic = be32_to_cpu(magic);
613 return vmdk_open_vmdk3(bs, file, flags);
616 return vmdk_open_vmdk4(bs, file, flags);
624 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
625 const char *desc_file_path)
631 const char *p = desc;
634 char extent_path[PATH_MAX];
635 BlockDriverState *extent_file;
638 /* parse extent line:
639 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
641 * RW [size in sectors] SPARSE "file-name.vmdk"
644 ret = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
645 access, §ors, type, fname, &flat_offset);
646 if (ret < 4 || strcmp(access, "RW")) {
648 } else if (!strcmp(type, "FLAT")) {
649 if (ret != 5 || flat_offset < 0) {
652 } else if (ret != 4) {
657 (strcmp(type, "FLAT") && strcmp(type, "SPARSE")) ||
658 (strcmp(access, "RW"))) {
662 path_combine(extent_path, sizeof(extent_path),
663 desc_file_path, fname);
664 ret = bdrv_file_open(&extent_file, extent_path, bs->open_flags);
669 /* save to extents array */
670 if (!strcmp(type, "FLAT")) {
674 extent = vmdk_add_extent(bs, extent_file, true, sectors,
675 0, 0, 0, 0, sectors);
676 extent->flat_start_offset = flat_offset << 9;
677 } else if (!strcmp(type, "SPARSE")) {
679 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags);
681 bdrv_delete(extent_file);
686 "VMDK: Not supported extent type \"%s\""".\n", type);
690 /* move to next line */
691 while (*p && *p != '\n') {
699 static int vmdk_open_desc_file(BlockDriverState *bs, int flags,
705 BDRVVmdkState *s = bs->opaque;
707 ret = bdrv_pread(bs->file, desc_offset, buf, sizeof(buf));
712 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
715 if (strcmp(ct, "monolithicFlat") &&
716 strcmp(ct, "twoGbMaxExtentSparse") &&
717 strcmp(ct, "twoGbMaxExtentFlat")) {
719 "VMDK: Not supported image type \"%s\""".\n", ct);
723 return vmdk_parse_extents(buf, bs, bs->file->filename);
726 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags)
729 BDRVVmdkState *s = bs->opaque;
731 if (vmdk_open_sparse(bs, bs->file, flags) == 0) {
732 s->desc_offset = 0x200;
734 ret = vmdk_open_desc_file(bs, flags, 0);
739 /* try to open parent images, if exist */
740 ret = vmdk_parent_open(bs);
744 s->parent_cid = vmdk_read_cid(bs, 1);
745 qemu_co_mutex_init(&s->lock);
747 /* Disable migration when VMDK images are used */
748 error_set(&s->migration_blocker,
749 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
750 "vmdk", bs->device_name, "live migration");
751 migrate_add_blocker(s->migration_blocker);
756 vmdk_free_extents(bs);
760 static int get_whole_cluster(BlockDriverState *bs,
762 uint64_t cluster_offset,
766 /* 128 sectors * 512 bytes each = grain size 64KB */
767 uint8_t whole_grain[extent->cluster_sectors * 512];
769 /* we will be here if it's first write on non-exist grain(cluster).
770 * try to read from parent image, if exist */
771 if (bs->backing_hd) {
774 if (!vmdk_is_cid_valid(bs)) {
778 /* floor offset to cluster */
779 offset -= offset % (extent->cluster_sectors * 512);
780 ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain,
781 extent->cluster_sectors);
786 /* Write grain only into the active image */
787 ret = bdrv_write(extent->file, cluster_offset, whole_grain,
788 extent->cluster_sectors);
796 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data)
798 /* update L2 table */
799 if (bdrv_pwrite_sync(
801 ((int64_t)m_data->l2_offset * 512)
802 + (m_data->l2_index * sizeof(m_data->offset)),
804 sizeof(m_data->offset)
808 /* update backup L2 table */
809 if (extent->l1_backup_table_offset != 0) {
810 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
811 if (bdrv_pwrite_sync(
813 ((int64_t)m_data->l2_offset * 512)
814 + (m_data->l2_index * sizeof(m_data->offset)),
815 &(m_data->offset), sizeof(m_data->offset)
824 static int get_cluster_offset(BlockDriverState *bs,
826 VmdkMetaData *m_data,
829 uint64_t *cluster_offset)
831 unsigned int l1_index, l2_offset, l2_index;
833 uint32_t min_count, *l2_table, tmp = 0;
839 *cluster_offset = extent->flat_start_offset;
843 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
844 l1_index = (offset >> 9) / extent->l1_entry_sectors;
845 if (l1_index >= extent->l1_size) {
848 l2_offset = extent->l1_table[l1_index];
852 for (i = 0; i < L2_CACHE_SIZE; i++) {
853 if (l2_offset == extent->l2_cache_offsets[i]) {
854 /* increment the hit count */
855 if (++extent->l2_cache_counts[i] == 0xffffffff) {
856 for (j = 0; j < L2_CACHE_SIZE; j++) {
857 extent->l2_cache_counts[j] >>= 1;
860 l2_table = extent->l2_cache + (i * extent->l2_size);
864 /* not found: load a new entry in the least used one */
866 min_count = 0xffffffff;
867 for (i = 0; i < L2_CACHE_SIZE; i++) {
868 if (extent->l2_cache_counts[i] < min_count) {
869 min_count = extent->l2_cache_counts[i];
873 l2_table = extent->l2_cache + (min_index * extent->l2_size);
876 (int64_t)l2_offset * 512,
878 extent->l2_size * sizeof(uint32_t)
879 ) != extent->l2_size * sizeof(uint32_t)) {
883 extent->l2_cache_offsets[min_index] = l2_offset;
884 extent->l2_cache_counts[min_index] = 1;
886 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
887 *cluster_offset = le32_to_cpu(l2_table[l2_index]);
889 if (!*cluster_offset) {
894 /* Avoid the L2 tables update for the images that have snapshots. */
895 *cluster_offset = bdrv_getlength(extent->file);
896 if (!extent->compressed) {
899 *cluster_offset + (extent->cluster_sectors << 9)
903 *cluster_offset >>= 9;
904 tmp = cpu_to_le32(*cluster_offset);
905 l2_table[l2_index] = tmp;
907 /* First of all we write grain itself, to avoid race condition
908 * that may to corrupt the image.
909 * This problem may occur because of insufficient space on host disk
910 * or inappropriate VM shutdown.
912 if (get_whole_cluster(
913 bs, extent, *cluster_offset, offset, allocate) == -1) {
918 m_data->offset = tmp;
919 m_data->l1_index = l1_index;
920 m_data->l2_index = l2_index;
921 m_data->l2_offset = l2_offset;
925 *cluster_offset <<= 9;
929 static VmdkExtent *find_extent(BDRVVmdkState *s,
930 int64_t sector_num, VmdkExtent *start_hint)
932 VmdkExtent *extent = start_hint;
935 extent = &s->extents[0];
937 while (extent < &s->extents[s->num_extents]) {
938 if (sector_num < extent->end_sector) {
946 static int coroutine_fn vmdk_co_is_allocated(BlockDriverState *bs,
947 int64_t sector_num, int nb_sectors, int *pnum)
949 BDRVVmdkState *s = bs->opaque;
950 int64_t index_in_cluster, n, ret;
954 extent = find_extent(s, sector_num, NULL);
958 qemu_co_mutex_lock(&s->lock);
959 ret = get_cluster_offset(bs, extent, NULL,
960 sector_num * 512, 0, &offset);
961 qemu_co_mutex_unlock(&s->lock);
962 /* get_cluster_offset returning 0 means success */
965 index_in_cluster = sector_num % extent->cluster_sectors;
966 n = extent->cluster_sectors - index_in_cluster;
967 if (n > nb_sectors) {
974 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
975 int64_t offset_in_cluster, const uint8_t *buf,
976 int nb_sectors, int64_t sector_num)
979 VmdkGrainMarker *data = NULL;
981 const uint8_t *write_buf = buf;
982 int write_len = nb_sectors * 512;
984 if (extent->compressed) {
985 if (!extent->has_marker) {
989 buf_len = (extent->cluster_sectors << 9) * 2;
990 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
991 if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK ||
996 data->lba = sector_num;
997 data->size = buf_len;
998 write_buf = (uint8_t *)data;
999 write_len = buf_len + sizeof(VmdkGrainMarker);
1001 ret = bdrv_pwrite(extent->file,
1002 cluster_offset + offset_in_cluster,
1005 if (ret != write_len) {
1006 ret = ret < 0 ? ret : -EIO;
1015 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1016 int64_t offset_in_cluster, uint8_t *buf,
1020 int cluster_bytes, buf_bytes;
1021 uint8_t *cluster_buf, *compressed_data;
1022 uint8_t *uncomp_buf;
1024 VmdkGrainMarker *marker;
1028 if (!extent->compressed) {
1029 ret = bdrv_pread(extent->file,
1030 cluster_offset + offset_in_cluster,
1031 buf, nb_sectors * 512);
1032 if (ret == nb_sectors * 512) {
1038 cluster_bytes = extent->cluster_sectors * 512;
1039 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1040 buf_bytes = cluster_bytes * 2;
1041 cluster_buf = g_malloc(buf_bytes);
1042 uncomp_buf = g_malloc(cluster_bytes);
1043 ret = bdrv_pread(extent->file,
1045 cluster_buf, buf_bytes);
1049 compressed_data = cluster_buf;
1050 buf_len = cluster_bytes;
1051 data_len = cluster_bytes;
1052 if (extent->has_marker) {
1053 marker = (VmdkGrainMarker *)cluster_buf;
1054 compressed_data = marker->data;
1055 data_len = le32_to_cpu(marker->size);
1057 if (!data_len || data_len > buf_bytes) {
1061 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1067 if (offset_in_cluster < 0 ||
1068 offset_in_cluster + nb_sectors * 512 > buf_len) {
1072 memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512);
1077 g_free(cluster_buf);
1081 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
1082 uint8_t *buf, int nb_sectors)
1084 BDRVVmdkState *s = bs->opaque;
1086 uint64_t n, index_in_cluster;
1087 uint64_t extent_begin_sector, extent_relative_sector_num;
1088 VmdkExtent *extent = NULL;
1089 uint64_t cluster_offset;
1091 while (nb_sectors > 0) {
1092 extent = find_extent(s, sector_num, extent);
1096 ret = get_cluster_offset(
1098 sector_num << 9, 0, &cluster_offset);
1099 extent_begin_sector = extent->end_sector - extent->sectors;
1100 extent_relative_sector_num = sector_num - extent_begin_sector;
1101 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1102 n = extent->cluster_sectors - index_in_cluster;
1103 if (n > nb_sectors) {
1107 /* if not allocated, try to read from parent image, if exist */
1108 if (bs->backing_hd) {
1109 if (!vmdk_is_cid_valid(bs)) {
1112 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
1117 memset(buf, 0, 512 * n);
1120 ret = vmdk_read_extent(extent,
1121 cluster_offset, index_in_cluster * 512,
1134 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num,
1135 uint8_t *buf, int nb_sectors)
1138 BDRVVmdkState *s = bs->opaque;
1139 qemu_co_mutex_lock(&s->lock);
1140 ret = vmdk_read(bs, sector_num, buf, nb_sectors);
1141 qemu_co_mutex_unlock(&s->lock);
1145 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
1146 const uint8_t *buf, int nb_sectors)
1148 BDRVVmdkState *s = bs->opaque;
1149 VmdkExtent *extent = NULL;
1151 int64_t index_in_cluster;
1152 uint64_t extent_begin_sector, extent_relative_sector_num;
1153 uint64_t cluster_offset;
1154 VmdkMetaData m_data;
1156 if (sector_num > bs->total_sectors) {
1158 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
1159 " total_sectors=0x%" PRIx64 "\n",
1160 sector_num, bs->total_sectors);
1164 while (nb_sectors > 0) {
1165 extent = find_extent(s, sector_num, extent);
1169 ret = get_cluster_offset(
1173 sector_num << 9, !extent->compressed,
1175 if (extent->compressed) {
1177 /* Refuse write to allocated cluster for streamOptimized */
1179 "VMDK: can't write to allocated cluster"
1180 " for streamOptimized\n");
1184 ret = get_cluster_offset(
1195 extent_begin_sector = extent->end_sector - extent->sectors;
1196 extent_relative_sector_num = sector_num - extent_begin_sector;
1197 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1198 n = extent->cluster_sectors - index_in_cluster;
1199 if (n > nb_sectors) {
1203 ret = vmdk_write_extent(extent,
1204 cluster_offset, index_in_cluster * 512,
1205 buf, n, sector_num);
1210 /* update L2 tables */
1211 if (vmdk_L2update(extent, &m_data) == -1) {
1219 /* update CID on the first write every time the virtual disk is
1221 if (!s->cid_updated) {
1222 ret = vmdk_write_cid(bs, time(NULL));
1226 s->cid_updated = true;
1232 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num,
1233 const uint8_t *buf, int nb_sectors)
1236 BDRVVmdkState *s = bs->opaque;
1237 qemu_co_mutex_lock(&s->lock);
1238 ret = vmdk_write(bs, sector_num, buf, nb_sectors);
1239 qemu_co_mutex_unlock(&s->lock);
1244 static int vmdk_create_extent(const char *filename, int64_t filesize,
1245 bool flat, bool compress)
1250 uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
1252 fd = qemu_open(filename,
1253 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
1259 ret = ftruncate(fd, filesize);
1265 magic = cpu_to_be32(VMDK4_MAGIC);
1266 memset(&header, 0, sizeof(header));
1269 3 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0);
1270 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1271 header.capacity = filesize / 512;
1272 header.granularity = 128;
1273 header.num_gtes_per_gte = 512;
1275 grains = (filesize / 512 + header.granularity - 1) / header.granularity;
1276 gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
1278 (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
1279 gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
1281 header.desc_offset = 1;
1282 header.desc_size = 20;
1283 header.rgd_offset = header.desc_offset + header.desc_size;
1284 header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
1285 header.grain_offset =
1286 ((header.gd_offset + gd_size + (gt_size * gt_count) +
1287 header.granularity - 1) / header.granularity) *
1289 /* swap endianness for all header fields */
1290 header.version = cpu_to_le32(header.version);
1291 header.flags = cpu_to_le32(header.flags);
1292 header.capacity = cpu_to_le64(header.capacity);
1293 header.granularity = cpu_to_le64(header.granularity);
1294 header.num_gtes_per_gte = cpu_to_le32(header.num_gtes_per_gte);
1295 header.desc_offset = cpu_to_le64(header.desc_offset);
1296 header.desc_size = cpu_to_le64(header.desc_size);
1297 header.rgd_offset = cpu_to_le64(header.rgd_offset);
1298 header.gd_offset = cpu_to_le64(header.gd_offset);
1299 header.grain_offset = cpu_to_le64(header.grain_offset);
1300 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1302 header.check_bytes[0] = 0xa;
1303 header.check_bytes[1] = 0x20;
1304 header.check_bytes[2] = 0xd;
1305 header.check_bytes[3] = 0xa;
1307 /* write all the data */
1308 ret = qemu_write_full(fd, &magic, sizeof(magic));
1309 if (ret != sizeof(magic)) {
1313 ret = qemu_write_full(fd, &header, sizeof(header));
1314 if (ret != sizeof(header)) {
1319 ret = ftruncate(fd, le64_to_cpu(header.grain_offset) << 9);
1325 /* write grain directory */
1326 lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
1327 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_size;
1328 i < gt_count; i++, tmp += gt_size) {
1329 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1330 if (ret != sizeof(tmp)) {
1336 /* write backup grain directory */
1337 lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
1338 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_size;
1339 i < gt_count; i++, tmp += gt_size) {
1340 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1341 if (ret != sizeof(tmp)) {
1353 static int filename_decompose(const char *filename, char *path, char *prefix,
1354 char *postfix, size_t buf_len)
1358 if (filename == NULL || !strlen(filename)) {
1359 fprintf(stderr, "Vmdk: no filename provided.\n");
1362 p = strrchr(filename, '/');
1364 p = strrchr(filename, '\\');
1367 p = strrchr(filename, ':');
1371 if (p - filename >= buf_len) {
1374 pstrcpy(path, p - filename + 1, filename);
1379 q = strrchr(p, '.');
1381 pstrcpy(prefix, buf_len, p);
1384 if (q - p >= buf_len) {
1387 pstrcpy(prefix, q - p + 1, p);
1388 pstrcpy(postfix, buf_len, q);
1393 static int relative_path(char *dest, int dest_size,
1394 const char *base, const char *target)
1400 const char *sep = "\\";
1402 const char *sep = "/";
1405 if (!(dest && base && target)) {
1408 if (path_is_absolute(target)) {
1409 pstrcpy(dest, dest_size, target);
1412 while (base[i] == target[i]) {
1425 pstrcat(dest, dest_size, "..");
1426 pstrcat(dest, dest_size, sep);
1428 pstrcat(dest, dest_size, q);
1432 static int vmdk_create(const char *filename, QEMUOptionParameter *options)
1435 char desc[BUF_SIZE];
1436 int64_t total_size = 0, filesize;
1437 const char *adapter_type = NULL;
1438 const char *backing_file = NULL;
1439 const char *fmt = NULL;
1442 bool flat, split, compress;
1443 char ext_desc_lines[BUF_SIZE] = "";
1444 char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX];
1445 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
1446 const char *desc_extent_line;
1447 char parent_desc_line[BUF_SIZE] = "";
1448 uint32_t parent_cid = 0xffffffff;
1449 uint32_t number_heads = 16;
1450 const char desc_template[] =
1451 "# Disk DescriptorFile\n"
1455 "createType=\"%s\"\n"
1458 "# Extent description\n"
1461 "# The Disk Data Base\n"
1464 "ddb.virtualHWVersion = \"%d\"\n"
1465 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1466 "ddb.geometry.heads = \"%d\"\n"
1467 "ddb.geometry.sectors = \"63\"\n"
1468 "ddb.adapterType = \"%s\"\n";
1470 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX)) {
1473 /* Read out options */
1474 while (options && options->name) {
1475 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1476 total_size = options->value.n;
1477 } else if (!strcmp(options->name, BLOCK_OPT_ADAPTER_TYPE)) {
1478 adapter_type = options->value.s;
1479 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1480 backing_file = options->value.s;
1481 } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) {
1482 flags |= options->value.n ? BLOCK_FLAG_COMPAT6 : 0;
1483 } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) {
1484 fmt = options->value.s;
1488 if (!adapter_type) {
1489 adapter_type = "ide";
1490 } else if (strcmp(adapter_type, "ide") &&
1491 strcmp(adapter_type, "buslogic") &&
1492 strcmp(adapter_type, "lsilogic") &&
1493 strcmp(adapter_type, "legacyESX")) {
1494 fprintf(stderr, "VMDK: Unknown adapter type: '%s'.\n", adapter_type);
1497 if (strcmp(adapter_type, "ide") != 0) {
1498 /* that's the number of heads with which vmware operates when
1499 creating, exporting, etc. vmdk files with a non-ide adapter type */
1503 /* Default format to monolithicSparse */
1504 fmt = "monolithicSparse";
1505 } else if (strcmp(fmt, "monolithicFlat") &&
1506 strcmp(fmt, "monolithicSparse") &&
1507 strcmp(fmt, "twoGbMaxExtentSparse") &&
1508 strcmp(fmt, "twoGbMaxExtentFlat") &&
1509 strcmp(fmt, "streamOptimized")) {
1510 fprintf(stderr, "VMDK: Unknown subformat: %s\n", fmt);
1513 split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1514 strcmp(fmt, "twoGbMaxExtentSparse"));
1515 flat = !(strcmp(fmt, "monolithicFlat") &&
1516 strcmp(fmt, "twoGbMaxExtentFlat"));
1517 compress = !strcmp(fmt, "streamOptimized");
1519 desc_extent_line = "RW %lld FLAT \"%s\" 0\n";
1521 desc_extent_line = "RW %lld SPARSE \"%s\"\n";
1523 if (flat && backing_file) {
1524 /* not supporting backing file for flat image */
1528 char parent_filename[PATH_MAX];
1529 BlockDriverState *bs = bdrv_new("");
1530 ret = bdrv_open(bs, backing_file, NULL, 0, NULL);
1535 if (strcmp(bs->drv->format_name, "vmdk")) {
1539 parent_cid = vmdk_read_cid(bs, 0);
1541 relative_path(parent_filename, sizeof(parent_filename),
1542 filename, backing_file);
1543 snprintf(parent_desc_line, sizeof(parent_desc_line),
1544 "parentFileNameHint=\"%s\"", parent_filename);
1547 /* Create extents */
1548 filesize = total_size;
1549 while (filesize > 0) {
1550 char desc_line[BUF_SIZE];
1551 char ext_filename[PATH_MAX];
1552 char desc_filename[PATH_MAX];
1553 int64_t size = filesize;
1555 if (split && size > split_size) {
1559 snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s",
1560 prefix, flat ? 'f' : 's', ++idx, postfix);
1562 snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s",
1565 snprintf(desc_filename, sizeof(desc_filename), "%s%s",
1568 snprintf(ext_filename, sizeof(ext_filename), "%s%s",
1569 path, desc_filename);
1571 if (vmdk_create_extent(ext_filename, size, flat, compress)) {
1576 /* Format description line */
1577 snprintf(desc_line, sizeof(desc_line),
1578 desc_extent_line, size / 512, desc_filename);
1579 pstrcat(ext_desc_lines, sizeof(ext_desc_lines), desc_line);
1581 /* generate descriptor file */
1582 snprintf(desc, sizeof(desc), desc_template,
1583 (unsigned int)time(NULL),
1588 (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
1589 total_size / (int64_t)(63 * number_heads * 512), number_heads,
1591 if (split || flat) {
1592 fd = qemu_open(filename,
1593 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
1596 fd = qemu_open(filename,
1597 O_WRONLY | O_BINARY | O_LARGEFILE,
1603 /* the descriptor offset = 0x200 */
1604 if (!split && !flat && 0x200 != lseek(fd, 0x200, SEEK_SET)) {
1608 ret = qemu_write_full(fd, desc, strlen(desc));
1609 if (ret != strlen(desc)) {
1619 static void vmdk_close(BlockDriverState *bs)
1621 BDRVVmdkState *s = bs->opaque;
1623 vmdk_free_extents(bs);
1625 migrate_del_blocker(s->migration_blocker);
1626 error_free(s->migration_blocker);
1629 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
1631 BDRVVmdkState *s = bs->opaque;
1635 for (i = 0; i < s->num_extents; i++) {
1636 err = bdrv_co_flush(s->extents[i].file);
1644 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
1649 BDRVVmdkState *s = bs->opaque;
1651 ret = bdrv_get_allocated_file_size(bs->file);
1655 for (i = 0; i < s->num_extents; i++) {
1656 if (s->extents[i].file == bs->file) {
1659 r = bdrv_get_allocated_file_size(s->extents[i].file);
1668 static QEMUOptionParameter vmdk_create_options[] = {
1670 .name = BLOCK_OPT_SIZE,
1672 .help = "Virtual disk size"
1675 .name = BLOCK_OPT_ADAPTER_TYPE,
1677 .help = "Virtual adapter type, can be one of "
1678 "ide (default), lsilogic, buslogic or legacyESX"
1681 .name = BLOCK_OPT_BACKING_FILE,
1683 .help = "File name of a base image"
1686 .name = BLOCK_OPT_COMPAT6,
1688 .help = "VMDK version 6 image"
1691 .name = BLOCK_OPT_SUBFMT,
1694 "VMDK flat extent format, can be one of "
1695 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
1700 static BlockDriver bdrv_vmdk = {
1701 .format_name = "vmdk",
1702 .instance_size = sizeof(BDRVVmdkState),
1703 .bdrv_probe = vmdk_probe,
1704 .bdrv_open = vmdk_open,
1705 .bdrv_reopen_prepare = vmdk_reopen_prepare,
1706 .bdrv_read = vmdk_co_read,
1707 .bdrv_write = vmdk_co_write,
1708 .bdrv_close = vmdk_close,
1709 .bdrv_create = vmdk_create,
1710 .bdrv_co_flush_to_disk = vmdk_co_flush,
1711 .bdrv_co_is_allocated = vmdk_co_is_allocated,
1712 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
1714 .create_options = vmdk_create_options,
1717 static void bdrv_vmdk_init(void)
1719 bdrv_register(&bdrv_vmdk);
1722 block_init(bdrv_vmdk_init);