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_NL_DETECT (1 << 0)
36 #define VMDK4_FLAG_RGD (1 << 1)
37 /* Zeroed-grain enable bit */
38 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
39 #define VMDK4_FLAG_COMPRESS (1 << 16)
40 #define VMDK4_FLAG_MARKER (1 << 17)
41 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
43 #define VMDK_GTE_ZEROED 0x1
45 /* VMDK internal error codes */
47 #define VMDK_ERROR (-1)
48 /* Cluster not allocated */
49 #define VMDK_UNALLOC (-2)
50 #define VMDK_ZEROED (-3)
52 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
57 uint32_t disk_sectors;
59 uint32_t l1dir_offset;
61 uint32_t file_sectors;
64 uint32_t sectors_per_track;
65 } QEMU_PACKED VMDK3Header;
74 /* Number of GrainTableEntries per GrainTable */
75 uint32_t num_gtes_per_gt;
78 uint64_t grain_offset;
81 uint16_t compressAlgorithm;
82 } QEMU_PACKED VMDK4Header;
84 #define L2_CACHE_SIZE 16
86 typedef struct VmdkExtent {
87 BlockDriverState *file;
95 int64_t flat_start_offset;
96 int64_t l1_table_offset;
97 int64_t l1_backup_table_offset;
99 uint32_t *l1_backup_table;
100 unsigned int l1_size;
101 uint32_t l1_entry_sectors;
103 unsigned int l2_size;
105 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
106 uint32_t l2_cache_counts[L2_CACHE_SIZE];
108 unsigned int cluster_sectors;
111 typedef struct BDRVVmdkState {
113 uint64_t desc_offset;
117 /* Extent array with num_extents entries, ascend ordered by address */
119 Error *migration_blocker;
122 typedef struct VmdkMetaData {
124 unsigned int l1_index;
125 unsigned int l2_index;
126 unsigned int l2_offset;
128 uint32_t *l2_cache_entry;
131 typedef struct VmdkGrainMarker {
135 } QEMU_PACKED VmdkGrainMarker;
138 MARKER_END_OF_STREAM = 0,
139 MARKER_GRAIN_TABLE = 1,
140 MARKER_GRAIN_DIRECTORY = 2,
144 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
151 magic = be32_to_cpu(*(uint32_t *)buf);
152 if (magic == VMDK3_MAGIC ||
153 magic == VMDK4_MAGIC) {
156 const char *p = (const char *)buf;
157 const char *end = p + buf_size;
160 /* skip comment line */
161 while (p < end && *p != '\n') {
168 while (p < end && *p == ' ') {
171 /* skip '\r' if windows line endings used. */
172 if (p < end && *p == '\r') {
175 /* only accept blank lines before 'version=' line */
176 if (p == end || *p != '\n') {
182 if (end - p >= strlen("version=X\n")) {
183 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
184 strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
188 if (end - p >= strlen("version=X\r\n")) {
189 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
190 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
202 #define SECTOR_SIZE 512
203 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
204 #define BUF_SIZE 4096
205 #define HEADER_SIZE 512 /* first sector of 512 bytes */
207 static void vmdk_free_extents(BlockDriverState *bs)
210 BDRVVmdkState *s = bs->opaque;
213 for (i = 0; i < s->num_extents; i++) {
217 g_free(e->l1_backup_table);
218 if (e->file != bs->file) {
225 static void vmdk_free_last_extent(BlockDriverState *bs)
227 BDRVVmdkState *s = bs->opaque;
229 if (s->num_extents == 0) {
233 s->extents = g_realloc(s->extents, s->num_extents * sizeof(VmdkExtent));
236 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
238 char desc[DESC_SIZE];
239 uint32_t cid = 0xffffffff;
240 const char *p_name, *cid_str;
242 BDRVVmdkState *s = bs->opaque;
245 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
251 cid_str = "parentCID";
252 cid_str_size = sizeof("parentCID");
255 cid_str_size = sizeof("CID");
258 desc[DESC_SIZE - 1] = '\0';
259 p_name = strstr(desc, cid_str);
260 if (p_name != NULL) {
261 p_name += cid_str_size;
262 sscanf(p_name, "%x", &cid);
268 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
270 char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
271 char *p_name, *tmp_str;
272 BDRVVmdkState *s = bs->opaque;
275 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
280 desc[DESC_SIZE - 1] = '\0';
281 tmp_str = strstr(desc, "parentCID");
282 if (tmp_str == NULL) {
286 pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
287 p_name = strstr(desc, "CID");
288 if (p_name != NULL) {
289 p_name += sizeof("CID");
290 snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
291 pstrcat(desc, sizeof(desc), tmp_desc);
294 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
302 static int vmdk_is_cid_valid(BlockDriverState *bs)
305 BDRVVmdkState *s = bs->opaque;
306 BlockDriverState *p_bs = bs->backing_hd;
310 cur_pcid = vmdk_read_cid(p_bs, 0);
311 if (s->parent_cid != cur_pcid) {
321 /* Queue extents, if any, for reopen() */
322 static int vmdk_reopen_prepare(BDRVReopenState *state,
323 BlockReopenQueue *queue, Error **errp)
330 assert(state != NULL);
331 assert(state->bs != NULL);
334 error_set(errp, ERROR_CLASS_GENERIC_ERROR,
335 "No reopen queue for VMDK extents");
339 s = state->bs->opaque;
343 for (i = 0; i < s->num_extents; i++) {
345 if (e->file != state->bs->file) {
346 bdrv_reopen_queue(queue, e->file, state->flags);
355 static int vmdk_parent_open(BlockDriverState *bs)
358 char desc[DESC_SIZE + 1];
359 BDRVVmdkState *s = bs->opaque;
362 desc[DESC_SIZE] = '\0';
363 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
368 p_name = strstr(desc, "parentFileNameHint");
369 if (p_name != NULL) {
372 p_name += sizeof("parentFileNameHint") + 1;
373 end_name = strchr(p_name, '\"');
374 if (end_name == NULL) {
377 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
381 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
387 /* Create and append extent to the extent array. Return the added VmdkExtent
388 * address. return NULL if allocation failed. */
389 static int vmdk_add_extent(BlockDriverState *bs,
390 BlockDriverState *file, bool flat, int64_t sectors,
391 int64_t l1_offset, int64_t l1_backup_offset,
393 int l2_size, uint64_t cluster_sectors,
394 VmdkExtent **new_extent)
397 BDRVVmdkState *s = bs->opaque;
399 if (cluster_sectors > 0x200000) {
400 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
401 error_report("invalid granularity, image may be corrupt");
404 if (l1_size > 512 * 1024 * 1024) {
405 /* Although with big capacity and small l1_entry_sectors, we can get a
406 * big l1_size, we don't want unbounded value to allocate the table.
407 * Limit it to 512M, which is 16PB for default cluster and L2 table
409 error_report("L1 size too big");
413 s->extents = g_realloc(s->extents,
414 (s->num_extents + 1) * sizeof(VmdkExtent));
415 extent = &s->extents[s->num_extents];
418 memset(extent, 0, sizeof(VmdkExtent));
421 extent->sectors = sectors;
422 extent->l1_table_offset = l1_offset;
423 extent->l1_backup_table_offset = l1_backup_offset;
424 extent->l1_size = l1_size;
425 extent->l1_entry_sectors = l2_size * cluster_sectors;
426 extent->l2_size = l2_size;
427 extent->cluster_sectors = cluster_sectors;
429 if (s->num_extents > 1) {
430 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
432 extent->end_sector = extent->sectors;
434 bs->total_sectors = extent->end_sector;
436 *new_extent = extent;
441 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent)
446 /* read the L1 table */
447 l1_size = extent->l1_size * sizeof(uint32_t);
448 extent->l1_table = g_malloc(l1_size);
449 ret = bdrv_pread(extent->file,
450 extent->l1_table_offset,
456 for (i = 0; i < extent->l1_size; i++) {
457 le32_to_cpus(&extent->l1_table[i]);
460 if (extent->l1_backup_table_offset) {
461 extent->l1_backup_table = g_malloc(l1_size);
462 ret = bdrv_pread(extent->file,
463 extent->l1_backup_table_offset,
464 extent->l1_backup_table,
469 for (i = 0; i < extent->l1_size; i++) {
470 le32_to_cpus(&extent->l1_backup_table[i]);
475 g_malloc(extent->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
478 g_free(extent->l1_backup_table);
480 g_free(extent->l1_table);
484 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
485 BlockDriverState *file,
493 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
497 ret = vmdk_add_extent(bs, file, false,
498 le32_to_cpu(header.disk_sectors),
499 le32_to_cpu(header.l1dir_offset) << 9,
501 le32_to_cpu(header.l1dir_size),
503 le32_to_cpu(header.granularity),
508 ret = vmdk_init_tables(bs, extent);
510 /* free extent allocated by vmdk_add_extent */
511 vmdk_free_last_extent(bs);
516 static int vmdk_open_desc_file(BlockDriverState *bs, int flags,
517 uint64_t desc_offset);
519 static int vmdk_open_vmdk4(BlockDriverState *bs,
520 BlockDriverState *file,
525 uint32_t l1_size, l1_entry_sectors;
528 int64_t l1_backup_offset = 0;
530 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
534 if (header.capacity == 0) {
535 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
537 return vmdk_open_desc_file(bs, flags, desc_offset << 9);
541 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
543 * The footer takes precedence over the header, so read it in. The
544 * footer starts at offset -1024 from the end: One sector for the
545 * footer, and another one for the end-of-stream marker.
552 uint8_t pad[512 - 16];
553 } QEMU_PACKED footer_marker;
557 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
563 uint8_t pad[512 - 16];
564 } QEMU_PACKED eos_marker;
565 } QEMU_PACKED footer;
567 ret = bdrv_pread(file,
568 bs->file->total_sectors * 512 - 1536,
569 &footer, sizeof(footer));
574 /* Some sanity checks for the footer */
575 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
576 le32_to_cpu(footer.footer_marker.size) != 0 ||
577 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
578 le64_to_cpu(footer.eos_marker.val) != 0 ||
579 le32_to_cpu(footer.eos_marker.size) != 0 ||
580 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
585 header = footer.header;
588 if (le32_to_cpu(header.version) >= 3) {
590 snprintf(buf, sizeof(buf), "VMDK version %d",
591 le32_to_cpu(header.version));
592 qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
593 bs->device_name, "vmdk", buf);
597 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
598 error_report("L2 table size too big");
602 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
603 * le64_to_cpu(header.granularity);
604 if (l1_entry_sectors == 0) {
607 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
609 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
610 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
612 ret = vmdk_add_extent(bs, file, false,
613 le64_to_cpu(header.capacity),
614 le64_to_cpu(header.gd_offset) << 9,
617 le32_to_cpu(header.num_gtes_per_gt),
618 le64_to_cpu(header.granularity),
624 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
625 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
626 extent->version = le32_to_cpu(header.version);
627 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
628 ret = vmdk_init_tables(bs, extent);
630 /* free extent allocated by vmdk_add_extent */
631 vmdk_free_last_extent(bs);
636 /* find an option value out of descriptor file */
637 static int vmdk_parse_description(const char *desc, const char *opt_name,
638 char *buf, int buf_size)
640 char *opt_pos, *opt_end;
641 const char *end = desc + strlen(desc);
643 opt_pos = strstr(desc, opt_name);
647 /* Skip "=\"" following opt_name */
648 opt_pos += strlen(opt_name) + 2;
649 if (opt_pos >= end) {
653 while (opt_end < end && *opt_end != '"') {
656 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
659 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
663 /* Open an extent file and append to bs array */
664 static int vmdk_open_sparse(BlockDriverState *bs,
665 BlockDriverState *file,
670 if (bdrv_pread(file, 0, &magic, sizeof(magic)) != sizeof(magic)) {
674 magic = be32_to_cpu(magic);
677 return vmdk_open_vmfs_sparse(bs, file, flags);
680 return vmdk_open_vmdk4(bs, file, flags);
688 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
689 const char *desc_file_path)
695 const char *p = desc;
698 char extent_path[PATH_MAX];
699 BlockDriverState *extent_file;
702 /* parse extent line:
703 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
705 * RW [size in sectors] SPARSE "file-name.vmdk"
708 ret = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
709 access, §ors, type, fname, &flat_offset);
710 if (ret < 4 || strcmp(access, "RW")) {
712 } else if (!strcmp(type, "FLAT")) {
713 if (ret != 5 || flat_offset < 0) {
716 } else if (ret != 4) {
721 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
722 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
723 (strcmp(access, "RW"))) {
727 path_combine(extent_path, sizeof(extent_path),
728 desc_file_path, fname);
729 ret = bdrv_file_open(&extent_file, extent_path, NULL, bs->open_flags);
734 /* save to extents array */
735 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
739 ret = vmdk_add_extent(bs, extent_file, true, sectors,
740 0, 0, 0, 0, sectors, &extent);
744 extent->flat_start_offset = flat_offset << 9;
745 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
746 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
747 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags);
749 bdrv_unref(extent_file);
754 "VMDK: Not supported extent type \"%s\""".\n", type);
758 /* move to next line */
759 while (*p && *p != '\n') {
767 static int vmdk_open_desc_file(BlockDriverState *bs, int flags,
768 uint64_t desc_offset)
773 BDRVVmdkState *s = bs->opaque;
776 size = bdrv_getlength(bs->file);
781 size = MIN(size, 1 << 20); /* avoid unbounded allocation */
782 buf = g_malloc0(size + 1);
784 ret = bdrv_pread(bs->file, desc_offset, buf, size);
788 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
792 if (strcmp(ct, "monolithicFlat") &&
793 strcmp(ct, "vmfs") &&
794 strcmp(ct, "vmfsSparse") &&
795 strcmp(ct, "twoGbMaxExtentSparse") &&
796 strcmp(ct, "twoGbMaxExtentFlat")) {
798 "VMDK: Not supported image type \"%s\""".\n", ct);
803 ret = vmdk_parse_extents(buf, bs, bs->file->filename);
809 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags)
812 BDRVVmdkState *s = bs->opaque;
814 if (vmdk_open_sparse(bs, bs->file, flags) == 0) {
815 s->desc_offset = 0x200;
817 ret = vmdk_open_desc_file(bs, flags, 0);
822 /* try to open parent images, if exist */
823 ret = vmdk_parent_open(bs);
827 s->parent_cid = vmdk_read_cid(bs, 1);
828 qemu_co_mutex_init(&s->lock);
830 /* Disable migration when VMDK images are used */
831 error_set(&s->migration_blocker,
832 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
833 "vmdk", bs->device_name, "live migration");
834 migrate_add_blocker(s->migration_blocker);
839 vmdk_free_extents(bs);
843 static int get_whole_cluster(BlockDriverState *bs,
845 uint64_t cluster_offset,
850 uint8_t *whole_grain = NULL;
852 /* we will be here if it's first write on non-exist grain(cluster).
853 * try to read from parent image, if exist */
854 if (bs->backing_hd) {
856 qemu_blockalign(bs, extent->cluster_sectors << BDRV_SECTOR_BITS);
857 if (!vmdk_is_cid_valid(bs)) {
862 /* floor offset to cluster */
863 offset -= offset % (extent->cluster_sectors * 512);
864 ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain,
865 extent->cluster_sectors);
871 /* Write grain only into the active image */
872 ret = bdrv_write(extent->file, cluster_offset, whole_grain,
873 extent->cluster_sectors);
880 qemu_vfree(whole_grain);
884 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data)
887 QEMU_BUILD_BUG_ON(sizeof(offset) != sizeof(m_data->offset));
888 offset = cpu_to_le32(m_data->offset);
889 /* update L2 table */
890 if (bdrv_pwrite_sync(
892 ((int64_t)m_data->l2_offset * 512)
893 + (m_data->l2_index * sizeof(m_data->offset)),
894 &offset, sizeof(offset)) < 0) {
897 /* update backup L2 table */
898 if (extent->l1_backup_table_offset != 0) {
899 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
900 if (bdrv_pwrite_sync(
902 ((int64_t)m_data->l2_offset * 512)
903 + (m_data->l2_index * sizeof(m_data->offset)),
904 &offset, sizeof(offset)) < 0) {
908 if (m_data->l2_cache_entry) {
909 *m_data->l2_cache_entry = offset;
915 static int get_cluster_offset(BlockDriverState *bs,
917 VmdkMetaData *m_data,
920 uint64_t *cluster_offset)
922 unsigned int l1_index, l2_offset, l2_index;
924 uint32_t min_count, *l2_table;
931 *cluster_offset = extent->flat_start_offset;
935 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
936 l1_index = (offset >> 9) / extent->l1_entry_sectors;
937 if (l1_index >= extent->l1_size) {
940 l2_offset = extent->l1_table[l1_index];
944 for (i = 0; i < L2_CACHE_SIZE; i++) {
945 if (l2_offset == extent->l2_cache_offsets[i]) {
946 /* increment the hit count */
947 if (++extent->l2_cache_counts[i] == 0xffffffff) {
948 for (j = 0; j < L2_CACHE_SIZE; j++) {
949 extent->l2_cache_counts[j] >>= 1;
952 l2_table = extent->l2_cache + (i * extent->l2_size);
956 /* not found: load a new entry in the least used one */
958 min_count = 0xffffffff;
959 for (i = 0; i < L2_CACHE_SIZE; i++) {
960 if (extent->l2_cache_counts[i] < min_count) {
961 min_count = extent->l2_cache_counts[i];
965 l2_table = extent->l2_cache + (min_index * extent->l2_size);
968 (int64_t)l2_offset * 512,
970 extent->l2_size * sizeof(uint32_t)
971 ) != extent->l2_size * sizeof(uint32_t)) {
975 extent->l2_cache_offsets[min_index] = l2_offset;
976 extent->l2_cache_counts[min_index] = 1;
978 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
979 *cluster_offset = le32_to_cpu(l2_table[l2_index]);
983 m_data->l1_index = l1_index;
984 m_data->l2_index = l2_index;
985 m_data->offset = *cluster_offset;
986 m_data->l2_offset = l2_offset;
987 m_data->l2_cache_entry = &l2_table[l2_index];
989 if (extent->has_zero_grain && *cluster_offset == VMDK_GTE_ZEROED) {
993 if (!*cluster_offset || zeroed) {
995 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
998 /* Avoid the L2 tables update for the images that have snapshots. */
999 *cluster_offset = bdrv_getlength(extent->file);
1000 if (!extent->compressed) {
1003 *cluster_offset + (extent->cluster_sectors << 9)
1007 *cluster_offset >>= 9;
1008 l2_table[l2_index] = cpu_to_le32(*cluster_offset);
1010 /* First of all we write grain itself, to avoid race condition
1011 * that may to corrupt the image.
1012 * This problem may occur because of insufficient space on host disk
1013 * or inappropriate VM shutdown.
1015 if (get_whole_cluster(
1016 bs, extent, *cluster_offset, offset, allocate) == -1) {
1021 m_data->offset = *cluster_offset;
1024 *cluster_offset <<= 9;
1028 static VmdkExtent *find_extent(BDRVVmdkState *s,
1029 int64_t sector_num, VmdkExtent *start_hint)
1031 VmdkExtent *extent = start_hint;
1034 extent = &s->extents[0];
1036 while (extent < &s->extents[s->num_extents]) {
1037 if (sector_num < extent->end_sector) {
1045 static int coroutine_fn vmdk_co_is_allocated(BlockDriverState *bs,
1046 int64_t sector_num, int nb_sectors, int *pnum)
1048 BDRVVmdkState *s = bs->opaque;
1049 int64_t index_in_cluster, n, ret;
1053 extent = find_extent(s, sector_num, NULL);
1057 qemu_co_mutex_lock(&s->lock);
1058 ret = get_cluster_offset(bs, extent, NULL,
1059 sector_num * 512, 0, &offset);
1060 qemu_co_mutex_unlock(&s->lock);
1062 ret = (ret == VMDK_OK || ret == VMDK_ZEROED);
1064 index_in_cluster = sector_num % extent->cluster_sectors;
1065 n = extent->cluster_sectors - index_in_cluster;
1066 if (n > nb_sectors) {
1073 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1074 int64_t offset_in_cluster, const uint8_t *buf,
1075 int nb_sectors, int64_t sector_num)
1078 VmdkGrainMarker *data = NULL;
1080 const uint8_t *write_buf = buf;
1081 int write_len = nb_sectors * 512;
1083 if (extent->compressed) {
1084 if (!extent->has_marker) {
1088 buf_len = (extent->cluster_sectors << 9) * 2;
1089 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1090 if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK ||
1095 data->lba = sector_num;
1096 data->size = buf_len;
1097 write_buf = (uint8_t *)data;
1098 write_len = buf_len + sizeof(VmdkGrainMarker);
1100 ret = bdrv_pwrite(extent->file,
1101 cluster_offset + offset_in_cluster,
1104 if (ret != write_len) {
1105 ret = ret < 0 ? ret : -EIO;
1114 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1115 int64_t offset_in_cluster, uint8_t *buf,
1119 int cluster_bytes, buf_bytes;
1120 uint8_t *cluster_buf, *compressed_data;
1121 uint8_t *uncomp_buf;
1123 VmdkGrainMarker *marker;
1127 if (!extent->compressed) {
1128 ret = bdrv_pread(extent->file,
1129 cluster_offset + offset_in_cluster,
1130 buf, nb_sectors * 512);
1131 if (ret == nb_sectors * 512) {
1137 cluster_bytes = extent->cluster_sectors * 512;
1138 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1139 buf_bytes = cluster_bytes * 2;
1140 cluster_buf = g_malloc(buf_bytes);
1141 uncomp_buf = g_malloc(cluster_bytes);
1142 ret = bdrv_pread(extent->file,
1144 cluster_buf, buf_bytes);
1148 compressed_data = cluster_buf;
1149 buf_len = cluster_bytes;
1150 data_len = cluster_bytes;
1151 if (extent->has_marker) {
1152 marker = (VmdkGrainMarker *)cluster_buf;
1153 compressed_data = marker->data;
1154 data_len = le32_to_cpu(marker->size);
1156 if (!data_len || data_len > buf_bytes) {
1160 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1166 if (offset_in_cluster < 0 ||
1167 offset_in_cluster + nb_sectors * 512 > buf_len) {
1171 memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512);
1176 g_free(cluster_buf);
1180 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
1181 uint8_t *buf, int nb_sectors)
1183 BDRVVmdkState *s = bs->opaque;
1185 uint64_t n, index_in_cluster;
1186 uint64_t extent_begin_sector, extent_relative_sector_num;
1187 VmdkExtent *extent = NULL;
1188 uint64_t cluster_offset;
1190 while (nb_sectors > 0) {
1191 extent = find_extent(s, sector_num, extent);
1195 ret = get_cluster_offset(
1197 sector_num << 9, 0, &cluster_offset);
1198 extent_begin_sector = extent->end_sector - extent->sectors;
1199 extent_relative_sector_num = sector_num - extent_begin_sector;
1200 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1201 n = extent->cluster_sectors - index_in_cluster;
1202 if (n > nb_sectors) {
1205 if (ret != VMDK_OK) {
1206 /* if not allocated, try to read from parent image, if exist */
1207 if (bs->backing_hd && ret != VMDK_ZEROED) {
1208 if (!vmdk_is_cid_valid(bs)) {
1211 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
1216 memset(buf, 0, 512 * n);
1219 ret = vmdk_read_extent(extent,
1220 cluster_offset, index_in_cluster * 512,
1233 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num,
1234 uint8_t *buf, int nb_sectors)
1237 BDRVVmdkState *s = bs->opaque;
1238 qemu_co_mutex_lock(&s->lock);
1239 ret = vmdk_read(bs, sector_num, buf, nb_sectors);
1240 qemu_co_mutex_unlock(&s->lock);
1246 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1247 * if possible, otherwise return -ENOTSUP.
1248 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1249 * with each cluster. By dry run we can find if the zero write
1250 * is possible without modifying image data.
1252 * Returns: error code with 0 for success.
1254 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
1255 const uint8_t *buf, int nb_sectors,
1256 bool zeroed, bool zero_dry_run)
1258 BDRVVmdkState *s = bs->opaque;
1259 VmdkExtent *extent = NULL;
1261 int64_t index_in_cluster;
1262 uint64_t extent_begin_sector, extent_relative_sector_num;
1263 uint64_t cluster_offset;
1264 VmdkMetaData m_data;
1266 if (sector_num > bs->total_sectors) {
1268 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
1269 " total_sectors=0x%" PRIx64 "\n",
1270 sector_num, bs->total_sectors);
1274 while (nb_sectors > 0) {
1275 extent = find_extent(s, sector_num, extent);
1279 ret = get_cluster_offset(
1283 sector_num << 9, !extent->compressed,
1285 if (extent->compressed) {
1286 if (ret == VMDK_OK) {
1287 /* Refuse write to allocated cluster for streamOptimized */
1289 "VMDK: can't write to allocated cluster"
1290 " for streamOptimized\n");
1294 ret = get_cluster_offset(
1302 if (ret == VMDK_ERROR) {
1305 extent_begin_sector = extent->end_sector - extent->sectors;
1306 extent_relative_sector_num = sector_num - extent_begin_sector;
1307 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1308 n = extent->cluster_sectors - index_in_cluster;
1309 if (n > nb_sectors) {
1313 /* Do zeroed write, buf is ignored */
1314 if (extent->has_zero_grain &&
1315 index_in_cluster == 0 &&
1316 n >= extent->cluster_sectors) {
1317 n = extent->cluster_sectors;
1318 if (!zero_dry_run) {
1319 m_data.offset = VMDK_GTE_ZEROED;
1320 /* update L2 tables */
1321 if (vmdk_L2update(extent, &m_data) != VMDK_OK) {
1329 ret = vmdk_write_extent(extent,
1330 cluster_offset, index_in_cluster * 512,
1331 buf, n, sector_num);
1336 /* update L2 tables */
1337 if (vmdk_L2update(extent, &m_data) != VMDK_OK) {
1346 /* update CID on the first write every time the virtual disk is
1348 if (!s->cid_updated) {
1349 ret = vmdk_write_cid(bs, time(NULL));
1353 s->cid_updated = true;
1359 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num,
1360 const uint8_t *buf, int nb_sectors)
1363 BDRVVmdkState *s = bs->opaque;
1364 qemu_co_mutex_lock(&s->lock);
1365 ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false);
1366 qemu_co_mutex_unlock(&s->lock);
1370 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs,
1375 BDRVVmdkState *s = bs->opaque;
1376 qemu_co_mutex_lock(&s->lock);
1377 /* write zeroes could fail if sectors not aligned to cluster, test it with
1378 * dry_run == true before really updating image */
1379 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true);
1381 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false);
1383 qemu_co_mutex_unlock(&s->lock);
1387 static int vmdk_create_extent(const char *filename, int64_t filesize,
1388 bool flat, bool compress, bool zeroed_grain)
1393 uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
1395 fd = qemu_open(filename,
1396 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
1402 ret = ftruncate(fd, filesize);
1408 magic = cpu_to_be32(VMDK4_MAGIC);
1409 memset(&header, 0, sizeof(header));
1410 header.version = zeroed_grain ? 2 : 1;
1411 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1412 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1413 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1414 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1415 header.capacity = filesize / 512;
1416 header.granularity = 128;
1417 header.num_gtes_per_gt = 512;
1419 grains = (filesize / 512 + header.granularity - 1) / header.granularity;
1420 gt_size = ((header.num_gtes_per_gt * sizeof(uint32_t)) + 511) >> 9;
1422 (grains + header.num_gtes_per_gt - 1) / header.num_gtes_per_gt;
1423 gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
1425 header.desc_offset = 1;
1426 header.desc_size = 20;
1427 header.rgd_offset = header.desc_offset + header.desc_size;
1428 header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
1429 header.grain_offset =
1430 ((header.gd_offset + gd_size + (gt_size * gt_count) +
1431 header.granularity - 1) / header.granularity) *
1433 /* swap endianness for all header fields */
1434 header.version = cpu_to_le32(header.version);
1435 header.flags = cpu_to_le32(header.flags);
1436 header.capacity = cpu_to_le64(header.capacity);
1437 header.granularity = cpu_to_le64(header.granularity);
1438 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1439 header.desc_offset = cpu_to_le64(header.desc_offset);
1440 header.desc_size = cpu_to_le64(header.desc_size);
1441 header.rgd_offset = cpu_to_le64(header.rgd_offset);
1442 header.gd_offset = cpu_to_le64(header.gd_offset);
1443 header.grain_offset = cpu_to_le64(header.grain_offset);
1444 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1446 header.check_bytes[0] = 0xa;
1447 header.check_bytes[1] = 0x20;
1448 header.check_bytes[2] = 0xd;
1449 header.check_bytes[3] = 0xa;
1451 /* write all the data */
1452 ret = qemu_write_full(fd, &magic, sizeof(magic));
1453 if (ret != sizeof(magic)) {
1457 ret = qemu_write_full(fd, &header, sizeof(header));
1458 if (ret != sizeof(header)) {
1463 ret = ftruncate(fd, le64_to_cpu(header.grain_offset) << 9);
1469 /* write grain directory */
1470 lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
1471 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_size;
1472 i < gt_count; i++, tmp += gt_size) {
1473 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1474 if (ret != sizeof(tmp)) {
1480 /* write backup grain directory */
1481 lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
1482 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_size;
1483 i < gt_count; i++, tmp += gt_size) {
1484 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1485 if (ret != sizeof(tmp)) {
1497 static int filename_decompose(const char *filename, char *path, char *prefix,
1498 char *postfix, size_t buf_len)
1502 if (filename == NULL || !strlen(filename)) {
1503 fprintf(stderr, "Vmdk: no filename provided.\n");
1506 p = strrchr(filename, '/');
1508 p = strrchr(filename, '\\');
1511 p = strrchr(filename, ':');
1515 if (p - filename >= buf_len) {
1518 pstrcpy(path, p - filename + 1, filename);
1523 q = strrchr(p, '.');
1525 pstrcpy(prefix, buf_len, p);
1528 if (q - p >= buf_len) {
1531 pstrcpy(prefix, q - p + 1, p);
1532 pstrcpy(postfix, buf_len, q);
1537 static int vmdk_create(const char *filename, QEMUOptionParameter *options)
1540 char desc[BUF_SIZE];
1541 int64_t total_size = 0, filesize;
1542 const char *adapter_type = NULL;
1543 const char *backing_file = NULL;
1544 const char *fmt = NULL;
1547 bool flat, split, compress;
1548 char ext_desc_lines[BUF_SIZE] = "";
1549 char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX];
1550 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
1551 const char *desc_extent_line;
1552 char parent_desc_line[BUF_SIZE] = "";
1553 uint32_t parent_cid = 0xffffffff;
1554 uint32_t number_heads = 16;
1555 bool zeroed_grain = false;
1556 const char desc_template[] =
1557 "# Disk DescriptorFile\n"
1561 "createType=\"%s\"\n"
1564 "# Extent description\n"
1567 "# The Disk Data Base\n"
1570 "ddb.virtualHWVersion = \"%d\"\n"
1571 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1572 "ddb.geometry.heads = \"%d\"\n"
1573 "ddb.geometry.sectors = \"63\"\n"
1574 "ddb.adapterType = \"%s\"\n";
1576 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX)) {
1579 /* Read out options */
1580 while (options && options->name) {
1581 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1582 total_size = options->value.n;
1583 } else if (!strcmp(options->name, BLOCK_OPT_ADAPTER_TYPE)) {
1584 adapter_type = options->value.s;
1585 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1586 backing_file = options->value.s;
1587 } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) {
1588 flags |= options->value.n ? BLOCK_FLAG_COMPAT6 : 0;
1589 } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) {
1590 fmt = options->value.s;
1591 } else if (!strcmp(options->name, BLOCK_OPT_ZEROED_GRAIN)) {
1592 zeroed_grain |= options->value.n;
1596 if (!adapter_type) {
1597 adapter_type = "ide";
1598 } else if (strcmp(adapter_type, "ide") &&
1599 strcmp(adapter_type, "buslogic") &&
1600 strcmp(adapter_type, "lsilogic") &&
1601 strcmp(adapter_type, "legacyESX")) {
1602 fprintf(stderr, "VMDK: Unknown adapter type: '%s'.\n", adapter_type);
1605 if (strcmp(adapter_type, "ide") != 0) {
1606 /* that's the number of heads with which vmware operates when
1607 creating, exporting, etc. vmdk files with a non-ide adapter type */
1611 /* Default format to monolithicSparse */
1612 fmt = "monolithicSparse";
1613 } else if (strcmp(fmt, "monolithicFlat") &&
1614 strcmp(fmt, "monolithicSparse") &&
1615 strcmp(fmt, "twoGbMaxExtentSparse") &&
1616 strcmp(fmt, "twoGbMaxExtentFlat") &&
1617 strcmp(fmt, "streamOptimized")) {
1618 fprintf(stderr, "VMDK: Unknown subformat: %s\n", fmt);
1621 split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1622 strcmp(fmt, "twoGbMaxExtentSparse"));
1623 flat = !(strcmp(fmt, "monolithicFlat") &&
1624 strcmp(fmt, "twoGbMaxExtentFlat"));
1625 compress = !strcmp(fmt, "streamOptimized");
1627 desc_extent_line = "RW %lld FLAT \"%s\" 0\n";
1629 desc_extent_line = "RW %lld SPARSE \"%s\"\n";
1631 if (flat && backing_file) {
1632 /* not supporting backing file for flat image */
1636 BlockDriverState *bs = bdrv_new("");
1637 ret = bdrv_open(bs, backing_file, NULL, 0, NULL);
1642 if (strcmp(bs->drv->format_name, "vmdk")) {
1646 parent_cid = vmdk_read_cid(bs, 0);
1648 snprintf(parent_desc_line, sizeof(parent_desc_line),
1649 "parentFileNameHint=\"%s\"", backing_file);
1652 /* Create extents */
1653 filesize = total_size;
1654 while (filesize > 0) {
1655 char desc_line[BUF_SIZE];
1656 char ext_filename[PATH_MAX];
1657 char desc_filename[PATH_MAX];
1658 int64_t size = filesize;
1660 if (split && size > split_size) {
1664 snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s",
1665 prefix, flat ? 'f' : 's', ++idx, postfix);
1667 snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s",
1670 snprintf(desc_filename, sizeof(desc_filename), "%s%s",
1673 snprintf(ext_filename, sizeof(ext_filename), "%s%s",
1674 path, desc_filename);
1676 if (vmdk_create_extent(ext_filename, size,
1677 flat, compress, zeroed_grain)) {
1682 /* Format description line */
1683 snprintf(desc_line, sizeof(desc_line),
1684 desc_extent_line, size / 512, desc_filename);
1685 pstrcat(ext_desc_lines, sizeof(ext_desc_lines), desc_line);
1687 /* generate descriptor file */
1688 snprintf(desc, sizeof(desc), desc_template,
1689 (unsigned int)time(NULL),
1694 (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
1695 total_size / (int64_t)(63 * number_heads * 512), number_heads,
1697 if (split || flat) {
1698 fd = qemu_open(filename,
1699 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
1702 fd = qemu_open(filename,
1703 O_WRONLY | O_BINARY | O_LARGEFILE,
1709 /* the descriptor offset = 0x200 */
1710 if (!split && !flat && 0x200 != lseek(fd, 0x200, SEEK_SET)) {
1714 ret = qemu_write_full(fd, desc, strlen(desc));
1715 if (ret != strlen(desc)) {
1725 static void vmdk_close(BlockDriverState *bs)
1727 BDRVVmdkState *s = bs->opaque;
1729 vmdk_free_extents(bs);
1731 migrate_del_blocker(s->migration_blocker);
1732 error_free(s->migration_blocker);
1735 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
1737 BDRVVmdkState *s = bs->opaque;
1741 for (i = 0; i < s->num_extents; i++) {
1742 err = bdrv_co_flush(s->extents[i].file);
1750 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
1755 BDRVVmdkState *s = bs->opaque;
1757 ret = bdrv_get_allocated_file_size(bs->file);
1761 for (i = 0; i < s->num_extents; i++) {
1762 if (s->extents[i].file == bs->file) {
1765 r = bdrv_get_allocated_file_size(s->extents[i].file);
1774 static int vmdk_has_zero_init(BlockDriverState *bs)
1777 BDRVVmdkState *s = bs->opaque;
1779 /* If has a flat extent and its underlying storage doesn't have zero init,
1781 for (i = 0; i < s->num_extents; i++) {
1782 if (s->extents[i].flat) {
1783 if (!bdrv_has_zero_init(s->extents[i].file)) {
1791 static QEMUOptionParameter vmdk_create_options[] = {
1793 .name = BLOCK_OPT_SIZE,
1795 .help = "Virtual disk size"
1798 .name = BLOCK_OPT_ADAPTER_TYPE,
1800 .help = "Virtual adapter type, can be one of "
1801 "ide (default), lsilogic, buslogic or legacyESX"
1804 .name = BLOCK_OPT_BACKING_FILE,
1806 .help = "File name of a base image"
1809 .name = BLOCK_OPT_COMPAT6,
1811 .help = "VMDK version 6 image"
1814 .name = BLOCK_OPT_SUBFMT,
1817 "VMDK flat extent format, can be one of "
1818 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
1821 .name = BLOCK_OPT_ZEROED_GRAIN,
1823 .help = "Enable efficient zero writes using the zeroed-grain GTE feature"
1828 static BlockDriver bdrv_vmdk = {
1829 .format_name = "vmdk",
1830 .instance_size = sizeof(BDRVVmdkState),
1831 .bdrv_probe = vmdk_probe,
1832 .bdrv_open = vmdk_open,
1833 .bdrv_reopen_prepare = vmdk_reopen_prepare,
1834 .bdrv_read = vmdk_co_read,
1835 .bdrv_write = vmdk_co_write,
1836 .bdrv_co_write_zeroes = vmdk_co_write_zeroes,
1837 .bdrv_close = vmdk_close,
1838 .bdrv_create = vmdk_create,
1839 .bdrv_co_flush_to_disk = vmdk_co_flush,
1840 .bdrv_co_is_allocated = vmdk_co_is_allocated,
1841 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
1842 .bdrv_has_zero_init = vmdk_has_zero_init,
1844 .create_options = vmdk_create_options,
1847 static void bdrv_vmdk_init(void)
1849 bdrv_register(&bdrv_vmdk);
1852 block_init(bdrv_vmdk_init);