2 * Block driver for the VMDK format
4 * Copyright (c) 2004 Fabrice Bellard
5 * Copyright (c) 2005 Filip Navara
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "block/block_int.h"
29 #include "sysemu/block-backend.h"
30 #include "qapi/qmp/qerror.h"
31 #include "qemu/error-report.h"
32 #include "qemu/module.h"
33 #include "qemu/bswap.h"
34 #include "migration/blocker.h"
35 #include "qemu/cutils.h"
38 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
39 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
40 #define VMDK4_COMPRESSION_DEFLATE 1
41 #define VMDK4_FLAG_NL_DETECT (1 << 0)
42 #define VMDK4_FLAG_RGD (1 << 1)
43 /* Zeroed-grain enable bit */
44 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
45 #define VMDK4_FLAG_COMPRESS (1 << 16)
46 #define VMDK4_FLAG_MARKER (1 << 17)
47 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
49 #define VMDK_GTE_ZEROED 0x1
51 /* VMDK internal error codes */
53 #define VMDK_ERROR (-1)
54 /* Cluster not allocated */
55 #define VMDK_UNALLOC (-2)
56 #define VMDK_ZEROED (-3)
58 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
63 uint32_t disk_sectors;
65 uint32_t l1dir_offset;
67 uint32_t file_sectors;
70 uint32_t sectors_per_track;
71 } QEMU_PACKED VMDK3Header;
80 /* Number of GrainTableEntries per GrainTable */
81 uint32_t num_gtes_per_gt;
84 uint64_t grain_offset;
87 uint16_t compressAlgorithm;
88 } QEMU_PACKED VMDK4Header;
90 #define L2_CACHE_SIZE 16
92 typedef struct VmdkExtent {
101 int64_t flat_start_offset;
102 int64_t l1_table_offset;
103 int64_t l1_backup_table_offset;
105 uint32_t *l1_backup_table;
106 unsigned int l1_size;
107 uint32_t l1_entry_sectors;
109 unsigned int l2_size;
111 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
112 uint32_t l2_cache_counts[L2_CACHE_SIZE];
114 int64_t cluster_sectors;
115 int64_t next_cluster_sector;
119 typedef struct BDRVVmdkState {
121 uint64_t desc_offset;
127 /* Extent array with num_extents entries, ascend ordered by address */
129 Error *migration_blocker;
133 typedef struct VmdkMetaData {
134 unsigned int l1_index;
135 unsigned int l2_index;
136 unsigned int l2_offset;
138 uint32_t *l2_cache_entry;
141 typedef struct VmdkGrainMarker {
145 } QEMU_PACKED VmdkGrainMarker;
148 MARKER_END_OF_STREAM = 0,
149 MARKER_GRAIN_TABLE = 1,
150 MARKER_GRAIN_DIRECTORY = 2,
154 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
161 magic = be32_to_cpu(*(uint32_t *)buf);
162 if (magic == VMDK3_MAGIC ||
163 magic == VMDK4_MAGIC) {
166 const char *p = (const char *)buf;
167 const char *end = p + buf_size;
170 /* skip comment line */
171 while (p < end && *p != '\n') {
178 while (p < end && *p == ' ') {
181 /* skip '\r' if windows line endings used. */
182 if (p < end && *p == '\r') {
185 /* only accept blank lines before 'version=' line */
186 if (p == end || *p != '\n') {
192 if (end - p >= strlen("version=X\n")) {
193 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
194 strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
198 if (end - p >= strlen("version=X\r\n")) {
199 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
200 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
210 #define SECTOR_SIZE 512
211 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
212 #define BUF_SIZE 4096
213 #define HEADER_SIZE 512 /* first sector of 512 bytes */
215 static void vmdk_free_extents(BlockDriverState *bs)
218 BDRVVmdkState *s = bs->opaque;
221 for (i = 0; i < s->num_extents; i++) {
225 g_free(e->l1_backup_table);
227 if (e->file != bs->file) {
228 bdrv_unref_child(bs, e->file);
234 static void vmdk_free_last_extent(BlockDriverState *bs)
236 BDRVVmdkState *s = bs->opaque;
238 if (s->num_extents == 0) {
242 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
245 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
248 uint32_t cid = 0xffffffff;
249 const char *p_name, *cid_str;
251 BDRVVmdkState *s = bs->opaque;
254 desc = g_malloc0(DESC_SIZE);
255 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
262 cid_str = "parentCID";
263 cid_str_size = sizeof("parentCID");
266 cid_str_size = sizeof("CID");
269 desc[DESC_SIZE - 1] = '\0';
270 p_name = strstr(desc, cid_str);
271 if (p_name != NULL) {
272 p_name += cid_str_size;
273 sscanf(p_name, "%" SCNx32, &cid);
280 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
282 char *desc, *tmp_desc;
283 char *p_name, *tmp_str;
284 BDRVVmdkState *s = bs->opaque;
287 desc = g_malloc0(DESC_SIZE);
288 tmp_desc = g_malloc0(DESC_SIZE);
289 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
294 desc[DESC_SIZE - 1] = '\0';
295 tmp_str = strstr(desc, "parentCID");
296 if (tmp_str == NULL) {
301 pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
302 p_name = strstr(desc, "CID");
303 if (p_name != NULL) {
304 p_name += sizeof("CID");
305 snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
306 pstrcat(desc, DESC_SIZE, tmp_desc);
309 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
317 static int vmdk_is_cid_valid(BlockDriverState *bs)
319 BDRVVmdkState *s = bs->opaque;
322 if (!s->cid_checked && bs->backing) {
323 BlockDriverState *p_bs = bs->backing->bs;
325 cur_pcid = vmdk_read_cid(p_bs, 0);
326 if (s->parent_cid != cur_pcid) {
331 s->cid_checked = true;
336 /* We have nothing to do for VMDK reopen, stubs just return success */
337 static int vmdk_reopen_prepare(BDRVReopenState *state,
338 BlockReopenQueue *queue, Error **errp)
340 assert(state != NULL);
341 assert(state->bs != NULL);
345 static int vmdk_parent_open(BlockDriverState *bs)
349 BDRVVmdkState *s = bs->opaque;
352 desc = g_malloc0(DESC_SIZE + 1);
353 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
359 p_name = strstr(desc, "parentFileNameHint");
360 if (p_name != NULL) {
363 p_name += sizeof("parentFileNameHint") + 1;
364 end_name = strchr(p_name, '\"');
365 if (end_name == NULL) {
369 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
374 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
382 /* Create and append extent to the extent array. Return the added VmdkExtent
383 * address. return NULL if allocation failed. */
384 static int vmdk_add_extent(BlockDriverState *bs,
385 BdrvChild *file, bool flat, int64_t sectors,
386 int64_t l1_offset, int64_t l1_backup_offset,
388 int l2_size, uint64_t cluster_sectors,
389 VmdkExtent **new_extent,
393 BDRVVmdkState *s = bs->opaque;
396 if (cluster_sectors > 0x200000) {
397 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
398 error_setg(errp, "Invalid granularity, image may be corrupt");
401 if (l1_size > 512 * 1024 * 1024) {
402 /* Although with big capacity and small l1_entry_sectors, we can get a
403 * big l1_size, we don't want unbounded value to allocate the table.
404 * Limit it to 512M, which is 16PB for default cluster and L2 table
406 error_setg(errp, "L1 size too big");
410 nb_sectors = bdrv_nb_sectors(file->bs);
411 if (nb_sectors < 0) {
415 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
416 extent = &s->extents[s->num_extents];
419 memset(extent, 0, sizeof(VmdkExtent));
422 extent->sectors = sectors;
423 extent->l1_table_offset = l1_offset;
424 extent->l1_backup_table_offset = l1_backup_offset;
425 extent->l1_size = l1_size;
426 extent->l1_entry_sectors = l2_size * cluster_sectors;
427 extent->l2_size = l2_size;
428 extent->cluster_sectors = flat ? sectors : cluster_sectors;
429 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
431 if (s->num_extents > 1) {
432 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
434 extent->end_sector = extent->sectors;
436 bs->total_sectors = extent->end_sector;
438 *new_extent = extent;
443 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
450 /* read the L1 table */
451 l1_size = extent->l1_size * sizeof(uint32_t);
452 extent->l1_table = g_try_malloc(l1_size);
453 if (l1_size && extent->l1_table == NULL) {
457 ret = bdrv_pread(extent->file,
458 extent->l1_table_offset,
462 error_setg_errno(errp, -ret,
463 "Could not read l1 table from extent '%s'",
464 extent->file->bs->filename);
467 for (i = 0; i < extent->l1_size; i++) {
468 le32_to_cpus(&extent->l1_table[i]);
471 if (extent->l1_backup_table_offset) {
472 extent->l1_backup_table = g_try_malloc(l1_size);
473 if (l1_size && extent->l1_backup_table == NULL) {
477 ret = bdrv_pread(extent->file,
478 extent->l1_backup_table_offset,
479 extent->l1_backup_table,
482 error_setg_errno(errp, -ret,
483 "Could not read l1 backup table from extent '%s'",
484 extent->file->bs->filename);
487 for (i = 0; i < extent->l1_size; i++) {
488 le32_to_cpus(&extent->l1_backup_table[i]);
493 g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE);
496 g_free(extent->l1_backup_table);
498 g_free(extent->l1_table);
502 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
504 int flags, Error **errp)
511 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
513 error_setg_errno(errp, -ret,
514 "Could not read header from file '%s'",
518 ret = vmdk_add_extent(bs, file, false,
519 le32_to_cpu(header.disk_sectors),
520 (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
522 le32_to_cpu(header.l1dir_size),
524 le32_to_cpu(header.granularity),
530 ret = vmdk_init_tables(bs, extent, errp);
532 /* free extent allocated by vmdk_add_extent */
533 vmdk_free_last_extent(bs);
538 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
539 QDict *options, Error **errp);
541 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
547 size = bdrv_getlength(file->bs);
549 error_setg_errno(errp, -size, "Could not access file");
554 /* Both descriptor file and sparse image must be much larger than 4
555 * bytes, also callers of vmdk_read_desc want to compare the first 4
556 * bytes with VMDK4_MAGIC, let's error out if less is read. */
557 error_setg(errp, "File is too small, not a valid image");
561 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */
562 buf = g_malloc(size + 1);
564 ret = bdrv_pread(file, desc_offset, buf, size);
566 error_setg_errno(errp, -ret, "Could not read from file");
575 static int vmdk_open_vmdk4(BlockDriverState *bs,
577 int flags, QDict *options, Error **errp)
581 uint32_t l1_size, l1_entry_sectors;
584 BDRVVmdkState *s = bs->opaque;
585 int64_t l1_backup_offset = 0;
588 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
590 error_setg_errno(errp, -ret,
591 "Could not read header from file '%s'",
595 if (header.capacity == 0) {
596 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
598 char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
602 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
608 if (!s->create_type) {
609 s->create_type = g_strdup("monolithicSparse");
612 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
614 * The footer takes precedence over the header, so read it in. The
615 * footer starts at offset -1024 from the end: One sector for the
616 * footer, and another one for the end-of-stream marker.
623 uint8_t pad[512 - 16];
624 } QEMU_PACKED footer_marker;
628 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
634 uint8_t pad[512 - 16];
635 } QEMU_PACKED eos_marker;
636 } QEMU_PACKED footer;
638 ret = bdrv_pread(file,
639 bs->file->bs->total_sectors * 512 - 1536,
640 &footer, sizeof(footer));
642 error_setg_errno(errp, -ret, "Failed to read footer");
646 /* Some sanity checks for the footer */
647 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
648 le32_to_cpu(footer.footer_marker.size) != 0 ||
649 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
650 le64_to_cpu(footer.eos_marker.val) != 0 ||
651 le32_to_cpu(footer.eos_marker.size) != 0 ||
652 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
654 error_setg(errp, "Invalid footer");
658 header = footer.header;
662 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
663 if (le32_to_cpu(header.version) > 3) {
664 error_setg(errp, "Unsupported VMDK version %" PRIu32,
665 le32_to_cpu(header.version));
667 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
669 /* VMware KB 2064959 explains that version 3 added support for
670 * persistent changed block tracking (CBT), and backup software can
671 * read it as version=1 if it doesn't care about the changed area
672 * information. So we are safe to enable read only. */
673 error_setg(errp, "VMDK version 3 must be read only");
677 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
678 error_setg(errp, "L2 table size too big");
682 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
683 * le64_to_cpu(header.granularity);
684 if (l1_entry_sectors == 0) {
685 error_setg(errp, "L1 entry size is invalid");
688 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
690 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
691 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
693 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
694 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
695 (int64_t)(le64_to_cpu(header.grain_offset)
696 * BDRV_SECTOR_SIZE));
700 ret = vmdk_add_extent(bs, file, false,
701 le64_to_cpu(header.capacity),
702 le64_to_cpu(header.gd_offset) << 9,
705 le32_to_cpu(header.num_gtes_per_gt),
706 le64_to_cpu(header.granularity),
713 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
714 if (extent->compressed) {
715 g_free(s->create_type);
716 s->create_type = g_strdup("streamOptimized");
718 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
719 extent->version = le32_to_cpu(header.version);
720 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
721 ret = vmdk_init_tables(bs, extent, errp);
723 /* free extent allocated by vmdk_add_extent */
724 vmdk_free_last_extent(bs);
729 /* find an option value out of descriptor file */
730 static int vmdk_parse_description(const char *desc, const char *opt_name,
731 char *buf, int buf_size)
733 char *opt_pos, *opt_end;
734 const char *end = desc + strlen(desc);
736 opt_pos = strstr(desc, opt_name);
740 /* Skip "=\"" following opt_name */
741 opt_pos += strlen(opt_name) + 2;
742 if (opt_pos >= end) {
746 while (opt_end < end && *opt_end != '"') {
749 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
752 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
756 /* Open an extent file and append to bs array */
757 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
758 char *buf, QDict *options, Error **errp)
762 magic = ldl_be_p(buf);
765 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
768 return vmdk_open_vmdk4(bs, file, flags, options, errp);
771 error_setg(errp, "Image not in VMDK format");
777 static const char *next_line(const char *s)
788 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
789 const char *desc_file_path, QDict *options,
801 BdrvChild *extent_file;
802 BDRVVmdkState *s = bs->opaque;
804 char extent_opt_prefix[32];
805 Error *local_err = NULL;
807 for (p = desc; *p; p = next_line(p)) {
808 /* parse extent line in one of below formats:
810 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
811 * RW [size in sectors] SPARSE "file-name.vmdk"
812 * RW [size in sectors] VMFS "file-name.vmdk"
813 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
816 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
817 access, §ors, type, fname, &flat_offset);
818 if (matches < 4 || strcmp(access, "RW")) {
820 } else if (!strcmp(type, "FLAT")) {
821 if (matches != 5 || flat_offset < 0) {
824 } else if (!strcmp(type, "VMFS")) {
830 } else if (matches != 4) {
835 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
836 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
837 (strcmp(access, "RW"))) {
841 if (!path_is_absolute(fname) && !path_has_protocol(fname) &&
844 error_setg(errp, "Cannot use relative extent paths with VMDK "
845 "descriptor file '%s'", bs->file->bs->filename);
849 extent_path = g_malloc0(PATH_MAX);
850 path_combine(extent_path, PATH_MAX, desc_file_path, fname);
852 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
855 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
856 bs, &child_file, false, &local_err);
859 error_propagate(errp, local_err);
863 /* save to extents array */
864 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
867 ret = vmdk_add_extent(bs, extent_file, true, sectors,
868 0, 0, 0, 0, 0, &extent, errp);
870 bdrv_unref_child(bs, extent_file);
873 extent->flat_start_offset = flat_offset << 9;
874 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
875 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
876 char *buf = vmdk_read_desc(extent_file, 0, errp);
880 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
885 bdrv_unref_child(bs, extent_file);
888 extent = &s->extents[s->num_extents - 1];
890 error_setg(errp, "Unsupported extent type '%s'", type);
891 bdrv_unref_child(bs, extent_file);
894 extent->type = g_strdup(type);
901 if (np[-1] == '\n') {
904 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
908 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
909 QDict *options, Error **errp)
913 BDRVVmdkState *s = bs->opaque;
915 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
916 error_setg(errp, "invalid VMDK image descriptor");
920 if (strcmp(ct, "monolithicFlat") &&
921 strcmp(ct, "vmfs") &&
922 strcmp(ct, "vmfsSparse") &&
923 strcmp(ct, "twoGbMaxExtentSparse") &&
924 strcmp(ct, "twoGbMaxExtentFlat")) {
925 error_setg(errp, "Unsupported image type '%s'", ct);
929 s->create_type = g_strdup(ct);
931 ret = vmdk_parse_extents(buf, bs, bs->file->bs->exact_filename, options,
937 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
942 BDRVVmdkState *s = bs->opaque;
944 Error *local_err = NULL;
946 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
952 buf = vmdk_read_desc(bs->file, 0, errp);
957 magic = ldl_be_p(buf);
961 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
963 s->desc_offset = 0x200;
966 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
973 /* try to open parent images, if exist */
974 ret = vmdk_parent_open(bs);
978 s->cid = vmdk_read_cid(bs, 0);
979 s->parent_cid = vmdk_read_cid(bs, 1);
980 qemu_co_mutex_init(&s->lock);
982 /* Disable migration when VMDK images are used */
983 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
984 "does not support live migration",
985 bdrv_get_device_or_node_name(bs));
986 ret = migrate_add_blocker(s->migration_blocker, &local_err);
988 error_propagate(errp, local_err);
989 error_free(s->migration_blocker);
998 g_free(s->create_type);
999 s->create_type = NULL;
1000 vmdk_free_extents(bs);
1005 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1007 BDRVVmdkState *s = bs->opaque;
1010 for (i = 0; i < s->num_extents; i++) {
1011 if (!s->extents[i].flat) {
1012 bs->bl.pwrite_zeroes_alignment =
1013 MAX(bs->bl.pwrite_zeroes_alignment,
1014 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1022 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1023 * to the cluster at @cluster_sector_num.
1025 * If @skip_start_sector < @skip_end_sector, the relative range
1026 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1027 * it for call to write user data in the request.
1029 static int get_whole_cluster(BlockDriverState *bs,
1031 uint64_t cluster_offset,
1033 uint64_t skip_start_bytes,
1034 uint64_t skip_end_bytes)
1037 int64_t cluster_bytes;
1038 uint8_t *whole_grain;
1040 /* For COW, align request sector_num to cluster start */
1041 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1042 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1043 whole_grain = qemu_blockalign(bs, cluster_bytes);
1046 memset(whole_grain, 0, skip_start_bytes);
1047 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1050 assert(skip_end_bytes <= cluster_bytes);
1051 /* we will be here if it's first write on non-exist grain(cluster).
1052 * try to read from parent image, if exist */
1053 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1058 /* Read backing data before skip range */
1059 if (skip_start_bytes > 0) {
1061 ret = bdrv_pread(bs->backing, offset, whole_grain,
1068 ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1075 /* Read backing data after skip range */
1076 if (skip_end_bytes < cluster_bytes) {
1078 ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1079 whole_grain + skip_end_bytes,
1080 cluster_bytes - skip_end_bytes);
1086 ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1087 whole_grain + skip_end_bytes,
1088 cluster_bytes - skip_end_bytes);
1097 qemu_vfree(whole_grain);
1101 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1104 offset = cpu_to_le32(offset);
1105 /* update L2 table */
1106 if (bdrv_pwrite_sync(extent->file,
1107 ((int64_t)m_data->l2_offset * 512)
1108 + (m_data->l2_index * sizeof(offset)),
1109 &offset, sizeof(offset)) < 0) {
1112 /* update backup L2 table */
1113 if (extent->l1_backup_table_offset != 0) {
1114 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1115 if (bdrv_pwrite_sync(extent->file,
1116 ((int64_t)m_data->l2_offset * 512)
1117 + (m_data->l2_index * sizeof(offset)),
1118 &offset, sizeof(offset)) < 0) {
1122 if (m_data->l2_cache_entry) {
1123 *m_data->l2_cache_entry = offset;
1130 * get_cluster_offset
1132 * Look up cluster offset in extent file by sector number, and store in
1135 * For flat extents, the start offset as parsed from the description file is
1138 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1139 * offset for a new cluster and update L2 cache. If there is a backing file,
1140 * COW is done before returning; otherwise, zeroes are written to the allocated
1141 * cluster. Both COW and zero writing skips the sector range
1142 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1143 * has new data to write there.
1145 * Returns: VMDK_OK if cluster exists and mapped in the image.
1146 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1147 * VMDK_ERROR if failed.
1149 static int get_cluster_offset(BlockDriverState *bs,
1151 VmdkMetaData *m_data,
1154 uint64_t *cluster_offset,
1155 uint64_t skip_start_bytes,
1156 uint64_t skip_end_bytes)
1158 unsigned int l1_index, l2_offset, l2_index;
1159 int min_index, i, j;
1160 uint32_t min_count, *l2_table;
1161 bool zeroed = false;
1163 int64_t cluster_sector;
1169 *cluster_offset = extent->flat_start_offset;
1173 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1174 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1175 if (l1_index >= extent->l1_size) {
1178 l2_offset = extent->l1_table[l1_index];
1180 return VMDK_UNALLOC;
1182 for (i = 0; i < L2_CACHE_SIZE; i++) {
1183 if (l2_offset == extent->l2_cache_offsets[i]) {
1184 /* increment the hit count */
1185 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1186 for (j = 0; j < L2_CACHE_SIZE; j++) {
1187 extent->l2_cache_counts[j] >>= 1;
1190 l2_table = extent->l2_cache + (i * extent->l2_size);
1194 /* not found: load a new entry in the least used one */
1196 min_count = 0xffffffff;
1197 for (i = 0; i < L2_CACHE_SIZE; i++) {
1198 if (extent->l2_cache_counts[i] < min_count) {
1199 min_count = extent->l2_cache_counts[i];
1203 l2_table = extent->l2_cache + (min_index * extent->l2_size);
1204 if (bdrv_pread(extent->file,
1205 (int64_t)l2_offset * 512,
1207 extent->l2_size * sizeof(uint32_t)
1208 ) != extent->l2_size * sizeof(uint32_t)) {
1212 extent->l2_cache_offsets[min_index] = l2_offset;
1213 extent->l2_cache_counts[min_index] = 1;
1215 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1216 cluster_sector = le32_to_cpu(l2_table[l2_index]);
1218 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1222 if (!cluster_sector || zeroed) {
1224 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1227 cluster_sector = extent->next_cluster_sector;
1228 extent->next_cluster_sector += extent->cluster_sectors;
1230 /* First of all we write grain itself, to avoid race condition
1231 * that may to corrupt the image.
1232 * This problem may occur because of insufficient space on host disk
1233 * or inappropriate VM shutdown.
1235 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1236 offset, skip_start_bytes, skip_end_bytes);
1242 m_data->l1_index = l1_index;
1243 m_data->l2_index = l2_index;
1244 m_data->l2_offset = l2_offset;
1245 m_data->l2_cache_entry = &l2_table[l2_index];
1248 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1252 static VmdkExtent *find_extent(BDRVVmdkState *s,
1253 int64_t sector_num, VmdkExtent *start_hint)
1255 VmdkExtent *extent = start_hint;
1258 extent = &s->extents[0];
1260 while (extent < &s->extents[s->num_extents]) {
1261 if (sector_num < extent->end_sector) {
1269 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1272 uint64_t extent_begin_offset, extent_relative_offset;
1273 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1275 extent_begin_offset =
1276 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1277 extent_relative_offset = offset - extent_begin_offset;
1278 return extent_relative_offset % cluster_size;
1281 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent *extent,
1285 offset = vmdk_find_offset_in_cluster(extent, sector_num * BDRV_SECTOR_SIZE);
1286 return offset / BDRV_SECTOR_SIZE;
1289 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs,
1290 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1292 BDRVVmdkState *s = bs->opaque;
1293 int64_t index_in_cluster, n, ret;
1297 extent = find_extent(s, sector_num, NULL);
1301 qemu_co_mutex_lock(&s->lock);
1302 ret = get_cluster_offset(bs, extent, NULL,
1303 sector_num * 512, false, &offset,
1305 qemu_co_mutex_unlock(&s->lock);
1307 index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num);
1316 ret = BDRV_BLOCK_ZERO;
1319 ret = BDRV_BLOCK_DATA;
1320 if (!extent->compressed) {
1321 ret |= BDRV_BLOCK_OFFSET_VALID;
1322 ret |= (offset + (index_in_cluster << BDRV_SECTOR_BITS))
1323 & BDRV_BLOCK_OFFSET_MASK;
1325 *file = extent->file->bs;
1329 n = extent->cluster_sectors - index_in_cluster;
1330 if (n > nb_sectors) {
1337 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1338 int64_t offset_in_cluster, QEMUIOVector *qiov,
1339 uint64_t qiov_offset, uint64_t n_bytes,
1343 VmdkGrainMarker *data = NULL;
1345 QEMUIOVector local_qiov;
1347 int64_t write_offset;
1348 int64_t write_end_sector;
1350 if (extent->compressed) {
1351 void *compressed_data;
1353 if (!extent->has_marker) {
1357 buf_len = (extent->cluster_sectors << 9) * 2;
1358 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1360 compressed_data = g_malloc(n_bytes);
1361 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1362 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1363 g_free(compressed_data);
1365 if (ret != Z_OK || buf_len == 0) {
1370 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1371 data->size = cpu_to_le32(buf_len);
1373 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1374 iov = (struct iovec) {
1378 qemu_iovec_init_external(&local_qiov, &iov, 1);
1380 qemu_iovec_init(&local_qiov, qiov->niov);
1381 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1384 write_offset = cluster_offset + offset_in_cluster,
1385 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1388 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1390 if (extent->compressed) {
1391 extent->next_cluster_sector = write_end_sector;
1393 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1403 if (!extent->compressed) {
1404 qemu_iovec_destroy(&local_qiov);
1409 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1410 int64_t offset_in_cluster, QEMUIOVector *qiov,
1414 int cluster_bytes, buf_bytes;
1415 uint8_t *cluster_buf, *compressed_data;
1416 uint8_t *uncomp_buf;
1418 VmdkGrainMarker *marker;
1422 if (!extent->compressed) {
1423 ret = bdrv_co_preadv(extent->file,
1424 cluster_offset + offset_in_cluster, bytes,
1431 cluster_bytes = extent->cluster_sectors * 512;
1432 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1433 buf_bytes = cluster_bytes * 2;
1434 cluster_buf = g_malloc(buf_bytes);
1435 uncomp_buf = g_malloc(cluster_bytes);
1436 ret = bdrv_pread(extent->file,
1438 cluster_buf, buf_bytes);
1442 compressed_data = cluster_buf;
1443 buf_len = cluster_bytes;
1444 data_len = cluster_bytes;
1445 if (extent->has_marker) {
1446 marker = (VmdkGrainMarker *)cluster_buf;
1447 compressed_data = marker->data;
1448 data_len = le32_to_cpu(marker->size);
1450 if (!data_len || data_len > buf_bytes) {
1454 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1460 if (offset_in_cluster < 0 ||
1461 offset_in_cluster + bytes > buf_len) {
1465 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1470 g_free(cluster_buf);
1474 static int coroutine_fn
1475 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1476 QEMUIOVector *qiov, int flags)
1478 BDRVVmdkState *s = bs->opaque;
1480 uint64_t n_bytes, offset_in_cluster;
1481 VmdkExtent *extent = NULL;
1482 QEMUIOVector local_qiov;
1483 uint64_t cluster_offset;
1484 uint64_t bytes_done = 0;
1486 qemu_iovec_init(&local_qiov, qiov->niov);
1487 qemu_co_mutex_lock(&s->lock);
1490 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1495 ret = get_cluster_offset(bs, extent, NULL,
1496 offset, false, &cluster_offset, 0, 0);
1497 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1499 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1500 - offset_in_cluster);
1502 if (ret != VMDK_OK) {
1503 /* if not allocated, try to read from parent image, if exist */
1504 if (bs->backing && ret != VMDK_ZEROED) {
1505 if (!vmdk_is_cid_valid(bs)) {
1510 qemu_iovec_reset(&local_qiov);
1511 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1513 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1519 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1522 qemu_iovec_reset(&local_qiov);
1523 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1525 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1526 &local_qiov, n_bytes);
1533 bytes_done += n_bytes;
1538 qemu_co_mutex_unlock(&s->lock);
1539 qemu_iovec_destroy(&local_qiov);
1546 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1547 * if possible, otherwise return -ENOTSUP.
1548 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1549 * with each cluster. By dry run we can find if the zero write
1550 * is possible without modifying image data.
1552 * Returns: error code with 0 for success.
1554 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1555 uint64_t bytes, QEMUIOVector *qiov,
1556 bool zeroed, bool zero_dry_run)
1558 BDRVVmdkState *s = bs->opaque;
1559 VmdkExtent *extent = NULL;
1561 int64_t offset_in_cluster, n_bytes;
1562 uint64_t cluster_offset;
1563 uint64_t bytes_done = 0;
1564 VmdkMetaData m_data;
1566 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1567 error_report("Wrong offset: offset=0x%" PRIx64
1568 " total_sectors=0x%" PRIx64,
1569 offset, bs->total_sectors);
1574 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1578 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1579 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1580 - offset_in_cluster);
1582 ret = get_cluster_offset(bs, extent, &m_data, offset,
1583 !(extent->compressed || zeroed),
1584 &cluster_offset, offset_in_cluster,
1585 offset_in_cluster + n_bytes);
1586 if (extent->compressed) {
1587 if (ret == VMDK_OK) {
1588 /* Refuse write to allocated cluster for streamOptimized */
1589 error_report("Could not write to allocated cluster"
1590 " for streamOptimized");
1594 ret = get_cluster_offset(bs, extent, &m_data, offset,
1595 true, &cluster_offset, 0, 0);
1598 if (ret == VMDK_ERROR) {
1602 /* Do zeroed write, buf is ignored */
1603 if (extent->has_zero_grain &&
1604 offset_in_cluster == 0 &&
1605 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
1606 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1607 if (!zero_dry_run) {
1608 /* update L2 tables */
1609 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1618 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
1619 qiov, bytes_done, n_bytes, offset);
1624 /* update L2 tables */
1625 if (vmdk_L2update(extent, &m_data,
1626 cluster_offset >> BDRV_SECTOR_BITS)
1634 bytes_done += n_bytes;
1636 /* update CID on the first write every time the virtual disk is
1638 if (!s->cid_updated) {
1639 ret = vmdk_write_cid(bs, g_random_int());
1643 s->cid_updated = true;
1649 static int coroutine_fn
1650 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1651 QEMUIOVector *qiov, int flags)
1654 BDRVVmdkState *s = bs->opaque;
1655 qemu_co_mutex_lock(&s->lock);
1656 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
1657 qemu_co_mutex_unlock(&s->lock);
1661 static int coroutine_fn
1662 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1663 uint64_t bytes, QEMUIOVector *qiov)
1665 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
1668 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
1671 BdrvRequestFlags flags)
1674 BDRVVmdkState *s = bs->opaque;
1676 qemu_co_mutex_lock(&s->lock);
1677 /* write zeroes could fail if sectors not aligned to cluster, test it with
1678 * dry_run == true before really updating image */
1679 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
1681 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
1683 qemu_co_mutex_unlock(&s->lock);
1687 static int vmdk_create_extent(const char *filename, int64_t filesize,
1688 bool flat, bool compress, bool zeroed_grain,
1689 QemuOpts *opts, Error **errp)
1692 BlockBackend *blk = NULL;
1694 Error *local_err = NULL;
1695 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1696 uint32_t *gd_buf = NULL;
1699 ret = bdrv_create_file(filename, opts, &local_err);
1701 error_propagate(errp, local_err);
1705 blk = blk_new_open(filename, NULL, NULL,
1706 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
1709 error_propagate(errp, local_err);
1714 blk_set_allow_write_beyond_eof(blk, true);
1717 ret = blk_truncate(blk, filesize, PREALLOC_MODE_OFF, errp);
1720 magic = cpu_to_be32(VMDK4_MAGIC);
1721 memset(&header, 0, sizeof(header));
1724 } else if (zeroed_grain) {
1729 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1730 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1731 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1732 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1733 header.capacity = filesize / BDRV_SECTOR_SIZE;
1734 header.granularity = 128;
1735 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1737 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1738 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1740 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1741 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1743 header.desc_offset = 1;
1744 header.desc_size = 20;
1745 header.rgd_offset = header.desc_offset + header.desc_size;
1746 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1747 header.grain_offset =
1748 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1749 header.granularity);
1750 /* swap endianness for all header fields */
1751 header.version = cpu_to_le32(header.version);
1752 header.flags = cpu_to_le32(header.flags);
1753 header.capacity = cpu_to_le64(header.capacity);
1754 header.granularity = cpu_to_le64(header.granularity);
1755 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1756 header.desc_offset = cpu_to_le64(header.desc_offset);
1757 header.desc_size = cpu_to_le64(header.desc_size);
1758 header.rgd_offset = cpu_to_le64(header.rgd_offset);
1759 header.gd_offset = cpu_to_le64(header.gd_offset);
1760 header.grain_offset = cpu_to_le64(header.grain_offset);
1761 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1763 header.check_bytes[0] = 0xa;
1764 header.check_bytes[1] = 0x20;
1765 header.check_bytes[2] = 0xd;
1766 header.check_bytes[3] = 0xa;
1768 /* write all the data */
1769 ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
1771 error_setg(errp, QERR_IO_ERROR);
1774 ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
1776 error_setg(errp, QERR_IO_ERROR);
1780 ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9,
1781 PREALLOC_MODE_OFF, errp);
1786 /* write grain directory */
1787 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1788 gd_buf = g_malloc0(gd_buf_size);
1789 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1790 i < gt_count; i++, tmp += gt_size) {
1791 gd_buf[i] = cpu_to_le32(tmp);
1793 ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1794 gd_buf, gd_buf_size, 0);
1796 error_setg(errp, QERR_IO_ERROR);
1800 /* write backup grain directory */
1801 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1802 i < gt_count; i++, tmp += gt_size) {
1803 gd_buf[i] = cpu_to_le32(tmp);
1805 ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1806 gd_buf, gd_buf_size, 0);
1808 error_setg(errp, QERR_IO_ERROR);
1821 static int filename_decompose(const char *filename, char *path, char *prefix,
1822 char *postfix, size_t buf_len, Error **errp)
1826 if (filename == NULL || !strlen(filename)) {
1827 error_setg(errp, "No filename provided");
1830 p = strrchr(filename, '/');
1832 p = strrchr(filename, '\\');
1835 p = strrchr(filename, ':');
1839 if (p - filename >= buf_len) {
1842 pstrcpy(path, p - filename + 1, filename);
1847 q = strrchr(p, '.');
1849 pstrcpy(prefix, buf_len, p);
1852 if (q - p >= buf_len) {
1855 pstrcpy(prefix, q - p + 1, p);
1856 pstrcpy(postfix, buf_len, q);
1861 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp)
1864 BlockBackend *new_blk = NULL;
1865 Error *local_err = NULL;
1867 int64_t total_size = 0, filesize;
1868 char *adapter_type = NULL;
1869 char *backing_file = NULL;
1870 char *hw_version = NULL;
1873 bool flat, split, compress;
1874 GString *ext_desc_lines;
1875 char *path = g_malloc0(PATH_MAX);
1876 char *prefix = g_malloc0(PATH_MAX);
1877 char *postfix = g_malloc0(PATH_MAX);
1878 char *desc_line = g_malloc0(BUF_SIZE);
1879 char *ext_filename = g_malloc0(PATH_MAX);
1880 char *desc_filename = g_malloc0(PATH_MAX);
1881 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
1882 const char *desc_extent_line;
1883 char *parent_desc_line = g_malloc0(BUF_SIZE);
1884 uint32_t parent_cid = 0xffffffff;
1885 uint32_t number_heads = 16;
1886 bool zeroed_grain = false;
1887 uint32_t desc_offset = 0, desc_len;
1888 const char desc_template[] =
1889 "# Disk DescriptorFile\n"
1892 "parentCID=%" PRIx32 "\n"
1893 "createType=\"%s\"\n"
1896 "# Extent description\n"
1899 "# The Disk Data Base\n"
1902 "ddb.virtualHWVersion = \"%s\"\n"
1903 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1904 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
1905 "ddb.geometry.sectors = \"63\"\n"
1906 "ddb.adapterType = \"%s\"\n";
1908 ext_desc_lines = g_string_new(NULL);
1910 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
1914 /* Read out options */
1915 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1917 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
1918 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1919 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
1920 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) {
1921 if (strcmp(hw_version, "undefined")) {
1923 "compat6 cannot be enabled with hwversion set");
1928 hw_version = g_strdup("6");
1930 if (strcmp(hw_version, "undefined") == 0) {
1932 hw_version = g_strdup("4");
1934 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1935 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) {
1936 zeroed_grain = true;
1939 if (!adapter_type) {
1940 adapter_type = g_strdup("ide");
1941 } else if (strcmp(adapter_type, "ide") &&
1942 strcmp(adapter_type, "buslogic") &&
1943 strcmp(adapter_type, "lsilogic") &&
1944 strcmp(adapter_type, "legacyESX")) {
1945 error_setg(errp, "Unknown adapter type: '%s'", adapter_type);
1949 if (strcmp(adapter_type, "ide") != 0) {
1950 /* that's the number of heads with which vmware operates when
1951 creating, exporting, etc. vmdk files with a non-ide adapter type */
1955 /* Default format to monolithicSparse */
1956 fmt = g_strdup("monolithicSparse");
1957 } else if (strcmp(fmt, "monolithicFlat") &&
1958 strcmp(fmt, "monolithicSparse") &&
1959 strcmp(fmt, "twoGbMaxExtentSparse") &&
1960 strcmp(fmt, "twoGbMaxExtentFlat") &&
1961 strcmp(fmt, "streamOptimized")) {
1962 error_setg(errp, "Unknown subformat: '%s'", fmt);
1966 split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1967 strcmp(fmt, "twoGbMaxExtentSparse"));
1968 flat = !(strcmp(fmt, "monolithicFlat") &&
1969 strcmp(fmt, "twoGbMaxExtentFlat"));
1970 compress = !strcmp(fmt, "streamOptimized");
1972 desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n";
1974 desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n";
1976 if (flat && backing_file) {
1977 error_setg(errp, "Flat image can't have backing file");
1981 if (flat && zeroed_grain) {
1982 error_setg(errp, "Flat image can't enable zeroed grain");
1988 char *full_backing = g_new0(char, PATH_MAX);
1989 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
1990 full_backing, PATH_MAX,
1993 g_free(full_backing);
1994 error_propagate(errp, local_err);
1999 blk = blk_new_open(full_backing, NULL, NULL,
2000 BDRV_O_NO_BACKING, errp);
2001 g_free(full_backing);
2006 if (strcmp(blk_bs(blk)->drv->format_name, "vmdk")) {
2011 parent_cid = vmdk_read_cid(blk_bs(blk), 0);
2013 snprintf(parent_desc_line, BUF_SIZE,
2014 "parentFileNameHint=\"%s\"", backing_file);
2017 /* Create extents */
2018 filesize = total_size;
2019 while (filesize > 0) {
2020 int64_t size = filesize;
2022 if (split && size > split_size) {
2026 snprintf(desc_filename, PATH_MAX, "%s-%c%03d%s",
2027 prefix, flat ? 'f' : 's', ++idx, postfix);
2029 snprintf(desc_filename, PATH_MAX, "%s-flat%s", prefix, postfix);
2031 snprintf(desc_filename, PATH_MAX, "%s%s", prefix, postfix);
2033 snprintf(ext_filename, PATH_MAX, "%s%s", path, desc_filename);
2035 if (vmdk_create_extent(ext_filename, size,
2036 flat, compress, zeroed_grain, opts, errp)) {
2042 /* Format description line */
2043 snprintf(desc_line, BUF_SIZE,
2044 desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename);
2045 g_string_append(ext_desc_lines, desc_line);
2047 /* generate descriptor file */
2048 desc = g_strdup_printf(desc_template,
2053 ext_desc_lines->str,
2056 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2059 desc_len = strlen(desc);
2060 /* the descriptor offset = 0x200 */
2061 if (!split && !flat) {
2062 desc_offset = 0x200;
2064 ret = bdrv_create_file(filename, opts, &local_err);
2066 error_propagate(errp, local_err);
2071 new_blk = blk_new_open(filename, NULL, NULL,
2072 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2074 if (new_blk == NULL) {
2075 error_propagate(errp, local_err);
2080 blk_set_allow_write_beyond_eof(new_blk, true);
2082 ret = blk_pwrite(new_blk, desc_offset, desc, desc_len, 0);
2084 error_setg_errno(errp, -ret, "Could not write description");
2087 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2088 * for description file */
2089 if (desc_offset == 0) {
2090 ret = blk_truncate(new_blk, desc_len, PREALLOC_MODE_OFF, errp);
2096 g_free(adapter_type);
2097 g_free(backing_file);
2105 g_free(ext_filename);
2106 g_free(desc_filename);
2107 g_free(parent_desc_line);
2108 g_string_free(ext_desc_lines, true);
2112 static void vmdk_close(BlockDriverState *bs)
2114 BDRVVmdkState *s = bs->opaque;
2116 vmdk_free_extents(bs);
2117 g_free(s->create_type);
2119 migrate_del_blocker(s->migration_blocker);
2120 error_free(s->migration_blocker);
2123 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2125 BDRVVmdkState *s = bs->opaque;
2129 for (i = 0; i < s->num_extents; i++) {
2130 err = bdrv_co_flush(s->extents[i].file->bs);
2138 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2143 BDRVVmdkState *s = bs->opaque;
2145 ret = bdrv_get_allocated_file_size(bs->file->bs);
2149 for (i = 0; i < s->num_extents; i++) {
2150 if (s->extents[i].file == bs->file) {
2153 r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2162 static int vmdk_has_zero_init(BlockDriverState *bs)
2165 BDRVVmdkState *s = bs->opaque;
2167 /* If has a flat extent and its underlying storage doesn't have zero init,
2169 for (i = 0; i < s->num_extents; i++) {
2170 if (s->extents[i].flat) {
2171 if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2179 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2181 ImageInfo *info = g_new0(ImageInfo, 1);
2183 *info = (ImageInfo){
2184 .filename = g_strdup(extent->file->bs->filename),
2185 .format = g_strdup(extent->type),
2186 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2187 .compressed = extent->compressed,
2188 .has_compressed = extent->compressed,
2189 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2190 .has_cluster_size = !extent->flat,
2196 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result,
2199 BDRVVmdkState *s = bs->opaque;
2200 VmdkExtent *extent = NULL;
2201 int64_t sector_num = 0;
2202 int64_t total_sectors = bdrv_nb_sectors(bs);
2204 uint64_t cluster_offset;
2211 if (sector_num >= total_sectors) {
2214 extent = find_extent(s, sector_num, extent);
2217 "ERROR: could not find extent for sector %" PRId64 "\n",
2221 ret = get_cluster_offset(bs, extent, NULL,
2222 sector_num << BDRV_SECTOR_BITS,
2223 false, &cluster_offset, 0, 0);
2224 if (ret == VMDK_ERROR) {
2226 "ERROR: could not get cluster_offset for sector %"
2227 PRId64 "\n", sector_num);
2230 if (ret == VMDK_OK &&
2231 cluster_offset >= bdrv_getlength(extent->file->bs))
2234 "ERROR: cluster offset for sector %"
2235 PRId64 " points after EOF\n", sector_num);
2238 sector_num += extent->cluster_sectors;
2241 result->corruptions++;
2245 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs)
2248 BDRVVmdkState *s = bs->opaque;
2249 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2250 ImageInfoList **next;
2252 *spec_info = (ImageInfoSpecific){
2253 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2255 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2259 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2260 .create_type = g_strdup(s->create_type),
2262 .parent_cid = s->parent_cid,
2265 next = &spec_info->u.vmdk.data->extents;
2266 for (i = 0; i < s->num_extents; i++) {
2267 *next = g_new0(ImageInfoList, 1);
2268 (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2269 (*next)->next = NULL;
2270 next = &(*next)->next;
2276 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2278 return a->flat == b->flat &&
2279 a->compressed == b->compressed &&
2280 (a->flat || a->cluster_sectors == b->cluster_sectors);
2283 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2286 BDRVVmdkState *s = bs->opaque;
2287 assert(s->num_extents);
2289 /* See if we have multiple extents but they have different cases */
2290 for (i = 1; i < s->num_extents; i++) {
2291 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2295 bdi->needs_compressed_writes = s->extents[0].compressed;
2296 if (!s->extents[0].flat) {
2297 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2302 static QemuOptsList vmdk_create_opts = {
2303 .name = "vmdk-create-opts",
2304 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2307 .name = BLOCK_OPT_SIZE,
2308 .type = QEMU_OPT_SIZE,
2309 .help = "Virtual disk size"
2312 .name = BLOCK_OPT_ADAPTER_TYPE,
2313 .type = QEMU_OPT_STRING,
2314 .help = "Virtual adapter type, can be one of "
2315 "ide (default), lsilogic, buslogic or legacyESX"
2318 .name = BLOCK_OPT_BACKING_FILE,
2319 .type = QEMU_OPT_STRING,
2320 .help = "File name of a base image"
2323 .name = BLOCK_OPT_COMPAT6,
2324 .type = QEMU_OPT_BOOL,
2325 .help = "VMDK version 6 image",
2326 .def_value_str = "off"
2329 .name = BLOCK_OPT_HWVERSION,
2330 .type = QEMU_OPT_STRING,
2331 .help = "VMDK hardware version",
2332 .def_value_str = "undefined"
2335 .name = BLOCK_OPT_SUBFMT,
2336 .type = QEMU_OPT_STRING,
2338 "VMDK flat extent format, can be one of "
2339 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2342 .name = BLOCK_OPT_ZEROED_GRAIN,
2343 .type = QEMU_OPT_BOOL,
2344 .help = "Enable efficient zero writes "
2345 "using the zeroed-grain GTE feature"
2347 { /* end of list */ }
2351 static BlockDriver bdrv_vmdk = {
2352 .format_name = "vmdk",
2353 .instance_size = sizeof(BDRVVmdkState),
2354 .bdrv_probe = vmdk_probe,
2355 .bdrv_open = vmdk_open,
2356 .bdrv_check = vmdk_check,
2357 .bdrv_reopen_prepare = vmdk_reopen_prepare,
2358 .bdrv_child_perm = bdrv_format_default_perms,
2359 .bdrv_co_preadv = vmdk_co_preadv,
2360 .bdrv_co_pwritev = vmdk_co_pwritev,
2361 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
2362 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
2363 .bdrv_close = vmdk_close,
2364 .bdrv_create = vmdk_create,
2365 .bdrv_co_flush_to_disk = vmdk_co_flush,
2366 .bdrv_co_get_block_status = vmdk_co_get_block_status,
2367 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2368 .bdrv_has_zero_init = vmdk_has_zero_init,
2369 .bdrv_get_specific_info = vmdk_get_specific_info,
2370 .bdrv_refresh_limits = vmdk_refresh_limits,
2371 .bdrv_get_info = vmdk_get_info,
2373 .supports_backing = true,
2374 .create_opts = &vmdk_create_opts,
2377 static void bdrv_vmdk_init(void)
2379 bdrv_register(&bdrv_vmdk);
2382 block_init(bdrv_vmdk_init);