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/qdict.h"
31 #include "qapi/qmp/qerror.h"
32 #include "qemu/error-report.h"
33 #include "qemu/module.h"
34 #include "qemu/option.h"
35 #include "qemu/bswap.h"
36 #include "migration/blocker.h"
37 #include "qemu/cutils.h"
40 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
41 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
42 #define VMDK4_COMPRESSION_DEFLATE 1
43 #define VMDK4_FLAG_NL_DETECT (1 << 0)
44 #define VMDK4_FLAG_RGD (1 << 1)
45 /* Zeroed-grain enable bit */
46 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
47 #define VMDK4_FLAG_COMPRESS (1 << 16)
48 #define VMDK4_FLAG_MARKER (1 << 17)
49 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
51 #define VMDK_EXTENT_MAX_SECTORS (1ULL << 32)
53 #define VMDK_GTE_ZEROED 0x1
55 /* VMDK internal error codes */
57 #define VMDK_ERROR (-1)
58 /* Cluster not allocated */
59 #define VMDK_UNALLOC (-2)
60 #define VMDK_ZEROED (-3)
62 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
67 uint32_t disk_sectors;
69 uint32_t l1dir_offset;
71 uint32_t file_sectors;
74 uint32_t sectors_per_track;
75 } QEMU_PACKED VMDK3Header;
84 /* Number of GrainTableEntries per GrainTable */
85 uint32_t num_gtes_per_gt;
88 uint64_t grain_offset;
91 uint16_t compressAlgorithm;
92 } QEMU_PACKED VMDK4Header;
94 #define L2_CACHE_SIZE 16
96 typedef struct VmdkExtent {
105 int64_t flat_start_offset;
106 int64_t l1_table_offset;
107 int64_t l1_backup_table_offset;
109 uint32_t *l1_backup_table;
110 unsigned int l1_size;
111 uint32_t l1_entry_sectors;
113 unsigned int l2_size;
115 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
116 uint32_t l2_cache_counts[L2_CACHE_SIZE];
118 int64_t cluster_sectors;
119 int64_t next_cluster_sector;
123 typedef struct BDRVVmdkState {
125 uint64_t desc_offset;
131 /* Extent array with num_extents entries, ascend ordered by address */
133 Error *migration_blocker;
137 typedef struct VmdkMetaData {
138 unsigned int l1_index;
139 unsigned int l2_index;
140 unsigned int l2_offset;
142 uint32_t *l2_cache_entry;
145 typedef struct VmdkGrainMarker {
149 } QEMU_PACKED VmdkGrainMarker;
152 MARKER_END_OF_STREAM = 0,
153 MARKER_GRAIN_TABLE = 1,
154 MARKER_GRAIN_DIRECTORY = 2,
158 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
165 magic = be32_to_cpu(*(uint32_t *)buf);
166 if (magic == VMDK3_MAGIC ||
167 magic == VMDK4_MAGIC) {
170 const char *p = (const char *)buf;
171 const char *end = p + buf_size;
174 /* skip comment line */
175 while (p < end && *p != '\n') {
182 while (p < end && *p == ' ') {
185 /* skip '\r' if windows line endings used. */
186 if (p < end && *p == '\r') {
189 /* only accept blank lines before 'version=' line */
190 if (p == end || *p != '\n') {
196 if (end - p >= strlen("version=X\n")) {
197 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
198 strncmp("version=2\n", p, strlen("version=2\n")) == 0 ||
199 strncmp("version=3\n", p, strlen("version=3\n")) == 0) {
203 if (end - p >= strlen("version=X\r\n")) {
204 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
205 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0 ||
206 strncmp("version=3\r\n", p, strlen("version=3\r\n")) == 0) {
216 #define SECTOR_SIZE 512
217 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
218 #define BUF_SIZE 4096
219 #define HEADER_SIZE 512 /* first sector of 512 bytes */
221 static void vmdk_free_extents(BlockDriverState *bs)
224 BDRVVmdkState *s = bs->opaque;
227 for (i = 0; i < s->num_extents; i++) {
231 g_free(e->l1_backup_table);
233 if (e->file != bs->file) {
234 bdrv_unref_child(bs, e->file);
240 static void vmdk_free_last_extent(BlockDriverState *bs)
242 BDRVVmdkState *s = bs->opaque;
244 if (s->num_extents == 0) {
248 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
251 /* Return -ve errno, or 0 on success and write CID into *pcid. */
252 static int vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
256 const char *p_name, *cid_str;
258 BDRVVmdkState *s = bs->opaque;
261 desc = g_malloc0(DESC_SIZE);
262 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
268 cid_str = "parentCID";
269 cid_str_size = sizeof("parentCID");
272 cid_str_size = sizeof("CID");
275 desc[DESC_SIZE - 1] = '\0';
276 p_name = strstr(desc, cid_str);
277 if (p_name == NULL) {
281 p_name += cid_str_size;
282 if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
294 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
296 char *desc, *tmp_desc;
297 char *p_name, *tmp_str;
298 BDRVVmdkState *s = bs->opaque;
301 desc = g_malloc0(DESC_SIZE);
302 tmp_desc = g_malloc0(DESC_SIZE);
303 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
308 desc[DESC_SIZE - 1] = '\0';
309 tmp_str = strstr(desc, "parentCID");
310 if (tmp_str == NULL) {
315 pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
316 p_name = strstr(desc, "CID");
317 if (p_name != NULL) {
318 p_name += sizeof("CID");
319 snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
320 pstrcat(desc, DESC_SIZE, tmp_desc);
323 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
331 static int vmdk_is_cid_valid(BlockDriverState *bs)
333 BDRVVmdkState *s = bs->opaque;
336 if (!s->cid_checked && bs->backing) {
337 BlockDriverState *p_bs = bs->backing->bs;
339 if (strcmp(p_bs->drv->format_name, "vmdk")) {
340 /* Backing file is not in vmdk format, so it does not have
341 * a CID, which makes the overlay's parent CID invalid */
345 if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
346 /* read failure: report as not valid */
349 if (s->parent_cid != cur_pcid) {
354 s->cid_checked = true;
359 /* We have nothing to do for VMDK reopen, stubs just return success */
360 static int vmdk_reopen_prepare(BDRVReopenState *state,
361 BlockReopenQueue *queue, Error **errp)
363 assert(state != NULL);
364 assert(state->bs != NULL);
368 static int vmdk_parent_open(BlockDriverState *bs)
372 BDRVVmdkState *s = bs->opaque;
375 desc = g_malloc0(DESC_SIZE + 1);
376 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
382 p_name = strstr(desc, "parentFileNameHint");
383 if (p_name != NULL) {
386 p_name += sizeof("parentFileNameHint") + 1;
387 end_name = strchr(p_name, '\"');
388 if (end_name == NULL) {
392 if ((end_name - p_name) > sizeof(bs->auto_backing_file) - 1) {
397 pstrcpy(bs->auto_backing_file, end_name - p_name + 1, p_name);
398 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
399 bs->auto_backing_file);
407 /* Create and append extent to the extent array. Return the added VmdkExtent
408 * address. return NULL if allocation failed. */
409 static int vmdk_add_extent(BlockDriverState *bs,
410 BdrvChild *file, bool flat, int64_t sectors,
411 int64_t l1_offset, int64_t l1_backup_offset,
413 int l2_size, uint64_t cluster_sectors,
414 VmdkExtent **new_extent,
418 BDRVVmdkState *s = bs->opaque;
421 if (cluster_sectors > 0x200000) {
422 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
423 error_setg(errp, "Invalid granularity, image may be corrupt");
426 if (l1_size > 512 * 1024 * 1024) {
427 /* Although with big capacity and small l1_entry_sectors, we can get a
428 * big l1_size, we don't want unbounded value to allocate the table.
429 * Limit it to 512M, which is 16PB for default cluster and L2 table
431 error_setg(errp, "L1 size too big");
435 nb_sectors = bdrv_nb_sectors(file->bs);
436 if (nb_sectors < 0) {
440 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
441 extent = &s->extents[s->num_extents];
444 memset(extent, 0, sizeof(VmdkExtent));
447 extent->sectors = sectors;
448 extent->l1_table_offset = l1_offset;
449 extent->l1_backup_table_offset = l1_backup_offset;
450 extent->l1_size = l1_size;
451 extent->l1_entry_sectors = l2_size * cluster_sectors;
452 extent->l2_size = l2_size;
453 extent->cluster_sectors = flat ? sectors : cluster_sectors;
454 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
456 if (s->num_extents > 1) {
457 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
459 extent->end_sector = extent->sectors;
461 bs->total_sectors = extent->end_sector;
463 *new_extent = extent;
468 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
475 /* read the L1 table */
476 l1_size = extent->l1_size * sizeof(uint32_t);
477 extent->l1_table = g_try_malloc(l1_size);
478 if (l1_size && extent->l1_table == NULL) {
482 ret = bdrv_pread(extent->file,
483 extent->l1_table_offset,
487 bdrv_refresh_filename(extent->file->bs);
488 error_setg_errno(errp, -ret,
489 "Could not read l1 table from extent '%s'",
490 extent->file->bs->filename);
493 for (i = 0; i < extent->l1_size; i++) {
494 le32_to_cpus(&extent->l1_table[i]);
497 if (extent->l1_backup_table_offset) {
498 extent->l1_backup_table = g_try_malloc(l1_size);
499 if (l1_size && extent->l1_backup_table == NULL) {
503 ret = bdrv_pread(extent->file,
504 extent->l1_backup_table_offset,
505 extent->l1_backup_table,
508 bdrv_refresh_filename(extent->file->bs);
509 error_setg_errno(errp, -ret,
510 "Could not read l1 backup table from extent '%s'",
511 extent->file->bs->filename);
514 for (i = 0; i < extent->l1_size; i++) {
515 le32_to_cpus(&extent->l1_backup_table[i]);
520 g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE);
523 g_free(extent->l1_backup_table);
525 g_free(extent->l1_table);
529 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
531 int flags, Error **errp)
538 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
540 bdrv_refresh_filename(file->bs);
541 error_setg_errno(errp, -ret,
542 "Could not read header from file '%s'",
546 ret = vmdk_add_extent(bs, file, false,
547 le32_to_cpu(header.disk_sectors),
548 (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
550 le32_to_cpu(header.l1dir_size),
552 le32_to_cpu(header.granularity),
558 ret = vmdk_init_tables(bs, extent, errp);
560 /* free extent allocated by vmdk_add_extent */
561 vmdk_free_last_extent(bs);
566 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
567 QDict *options, Error **errp);
569 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
575 size = bdrv_getlength(file->bs);
577 error_setg_errno(errp, -size, "Could not access file");
582 /* Both descriptor file and sparse image must be much larger than 4
583 * bytes, also callers of vmdk_read_desc want to compare the first 4
584 * bytes with VMDK4_MAGIC, let's error out if less is read. */
585 error_setg(errp, "File is too small, not a valid image");
589 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */
590 buf = g_malloc(size + 1);
592 ret = bdrv_pread(file, desc_offset, buf, size);
594 error_setg_errno(errp, -ret, "Could not read from file");
603 static int vmdk_open_vmdk4(BlockDriverState *bs,
605 int flags, QDict *options, Error **errp)
609 uint32_t l1_size, l1_entry_sectors;
612 BDRVVmdkState *s = bs->opaque;
613 int64_t l1_backup_offset = 0;
616 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
618 bdrv_refresh_filename(file->bs);
619 error_setg_errno(errp, -ret,
620 "Could not read header from file '%s'",
624 if (header.capacity == 0) {
625 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
627 char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
631 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
637 if (!s->create_type) {
638 s->create_type = g_strdup("monolithicSparse");
641 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
643 * The footer takes precedence over the header, so read it in. The
644 * footer starts at offset -1024 from the end: One sector for the
645 * footer, and another one for the end-of-stream marker.
652 uint8_t pad[512 - 16];
653 } QEMU_PACKED footer_marker;
657 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
663 uint8_t pad[512 - 16];
664 } QEMU_PACKED eos_marker;
665 } QEMU_PACKED footer;
667 ret = bdrv_pread(file,
668 bs->file->bs->total_sectors * 512 - 1536,
669 &footer, sizeof(footer));
671 error_setg_errno(errp, -ret, "Failed to read footer");
675 /* Some sanity checks for the footer */
676 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
677 le32_to_cpu(footer.footer_marker.size) != 0 ||
678 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
679 le64_to_cpu(footer.eos_marker.val) != 0 ||
680 le32_to_cpu(footer.eos_marker.size) != 0 ||
681 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
683 error_setg(errp, "Invalid footer");
687 header = footer.header;
691 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
692 if (le32_to_cpu(header.version) > 3) {
693 error_setg(errp, "Unsupported VMDK version %" PRIu32,
694 le32_to_cpu(header.version));
696 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
698 /* VMware KB 2064959 explains that version 3 added support for
699 * persistent changed block tracking (CBT), and backup software can
700 * read it as version=1 if it doesn't care about the changed area
701 * information. So we are safe to enable read only. */
702 error_setg(errp, "VMDK version 3 must be read only");
706 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
707 error_setg(errp, "L2 table size too big");
711 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
712 * le64_to_cpu(header.granularity);
713 if (l1_entry_sectors == 0) {
714 error_setg(errp, "L1 entry size is invalid");
717 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
719 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
720 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
722 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
723 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
724 (int64_t)(le64_to_cpu(header.grain_offset)
725 * BDRV_SECTOR_SIZE));
729 ret = vmdk_add_extent(bs, file, false,
730 le64_to_cpu(header.capacity),
731 le64_to_cpu(header.gd_offset) << 9,
734 le32_to_cpu(header.num_gtes_per_gt),
735 le64_to_cpu(header.granularity),
742 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
743 if (extent->compressed) {
744 g_free(s->create_type);
745 s->create_type = g_strdup("streamOptimized");
747 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
748 extent->version = le32_to_cpu(header.version);
749 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
750 ret = vmdk_init_tables(bs, extent, errp);
752 /* free extent allocated by vmdk_add_extent */
753 vmdk_free_last_extent(bs);
758 /* find an option value out of descriptor file */
759 static int vmdk_parse_description(const char *desc, const char *opt_name,
760 char *buf, int buf_size)
762 char *opt_pos, *opt_end;
763 const char *end = desc + strlen(desc);
765 opt_pos = strstr(desc, opt_name);
769 /* Skip "=\"" following opt_name */
770 opt_pos += strlen(opt_name) + 2;
771 if (opt_pos >= end) {
775 while (opt_end < end && *opt_end != '"') {
778 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
781 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
785 /* Open an extent file and append to bs array */
786 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
787 char *buf, QDict *options, Error **errp)
791 magic = ldl_be_p(buf);
794 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
797 return vmdk_open_vmdk4(bs, file, flags, options, errp);
800 error_setg(errp, "Image not in VMDK format");
806 static const char *next_line(const char *s)
817 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
818 const char *desc_file_path, QDict *options,
830 BdrvChild *extent_file;
831 BDRVVmdkState *s = bs->opaque;
833 char extent_opt_prefix[32];
834 Error *local_err = NULL;
836 for (p = desc; *p; p = next_line(p)) {
837 /* parse extent line in one of below formats:
839 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
840 * RW [size in sectors] SPARSE "file-name.vmdk"
841 * RW [size in sectors] VMFS "file-name.vmdk"
842 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
845 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
846 access, §ors, type, fname, &flat_offset);
847 if (matches < 4 || strcmp(access, "RW")) {
849 } else if (!strcmp(type, "FLAT")) {
850 if (matches != 5 || flat_offset < 0) {
853 } else if (!strcmp(type, "VMFS")) {
859 } else if (matches != 4) {
864 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
865 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
866 (strcmp(access, "RW"))) {
870 if (!path_is_absolute(fname) && !path_has_protocol(fname) &&
873 bdrv_refresh_filename(bs->file->bs);
874 error_setg(errp, "Cannot use relative extent paths with VMDK "
875 "descriptor file '%s'", bs->file->bs->filename);
879 extent_path = path_combine(desc_file_path, fname);
881 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
884 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
885 bs, &child_file, false, &local_err);
888 error_propagate(errp, local_err);
892 /* save to extents array */
893 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
896 ret = vmdk_add_extent(bs, extent_file, true, sectors,
897 0, 0, 0, 0, 0, &extent, errp);
899 bdrv_unref_child(bs, extent_file);
902 extent->flat_start_offset = flat_offset << 9;
903 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
904 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
905 char *buf = vmdk_read_desc(extent_file, 0, errp);
909 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
914 bdrv_unref_child(bs, extent_file);
917 extent = &s->extents[s->num_extents - 1];
919 error_setg(errp, "Unsupported extent type '%s'", type);
920 bdrv_unref_child(bs, extent_file);
923 extent->type = g_strdup(type);
930 if (np[-1] == '\n') {
933 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
937 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
938 QDict *options, Error **errp)
942 BDRVVmdkState *s = bs->opaque;
944 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
945 error_setg(errp, "invalid VMDK image descriptor");
949 if (strcmp(ct, "monolithicFlat") &&
950 strcmp(ct, "vmfs") &&
951 strcmp(ct, "vmfsSparse") &&
952 strcmp(ct, "twoGbMaxExtentSparse") &&
953 strcmp(ct, "twoGbMaxExtentFlat")) {
954 error_setg(errp, "Unsupported image type '%s'", ct);
958 s->create_type = g_strdup(ct);
960 ret = vmdk_parse_extents(buf, bs, bs->file->bs->exact_filename, options,
966 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
971 BDRVVmdkState *s = bs->opaque;
973 Error *local_err = NULL;
975 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
981 buf = vmdk_read_desc(bs->file, 0, errp);
986 magic = ldl_be_p(buf);
990 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
992 s->desc_offset = 0x200;
995 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1002 /* try to open parent images, if exist */
1003 ret = vmdk_parent_open(bs);
1007 ret = vmdk_read_cid(bs, 0, &s->cid);
1011 ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1015 qemu_co_mutex_init(&s->lock);
1017 /* Disable migration when VMDK images are used */
1018 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1019 "does not support live migration",
1020 bdrv_get_device_or_node_name(bs));
1021 ret = migrate_add_blocker(s->migration_blocker, &local_err);
1023 error_propagate(errp, local_err);
1024 error_free(s->migration_blocker);
1033 g_free(s->create_type);
1034 s->create_type = NULL;
1035 vmdk_free_extents(bs);
1040 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1042 BDRVVmdkState *s = bs->opaque;
1045 for (i = 0; i < s->num_extents; i++) {
1046 if (!s->extents[i].flat) {
1047 bs->bl.pwrite_zeroes_alignment =
1048 MAX(bs->bl.pwrite_zeroes_alignment,
1049 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1057 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1058 * to the cluster at @cluster_sector_num.
1060 * If @skip_start_sector < @skip_end_sector, the relative range
1061 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1062 * it for call to write user data in the request.
1064 static int get_whole_cluster(BlockDriverState *bs,
1066 uint64_t cluster_offset,
1068 uint64_t skip_start_bytes,
1069 uint64_t skip_end_bytes)
1072 int64_t cluster_bytes;
1073 uint8_t *whole_grain;
1075 /* For COW, align request sector_num to cluster start */
1076 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1077 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1078 whole_grain = qemu_blockalign(bs, cluster_bytes);
1081 memset(whole_grain, 0, skip_start_bytes);
1082 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1085 assert(skip_end_bytes <= cluster_bytes);
1086 /* we will be here if it's first write on non-exist grain(cluster).
1087 * try to read from parent image, if exist */
1088 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1093 /* Read backing data before skip range */
1094 if (skip_start_bytes > 0) {
1096 /* qcow2 emits this on bs->file instead of bs->backing */
1097 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1098 ret = bdrv_pread(bs->backing, offset, whole_grain,
1105 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1106 ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1113 /* Read backing data after skip range */
1114 if (skip_end_bytes < cluster_bytes) {
1116 /* qcow2 emits this on bs->file instead of bs->backing */
1117 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1118 ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1119 whole_grain + skip_end_bytes,
1120 cluster_bytes - skip_end_bytes);
1126 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1127 ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1128 whole_grain + skip_end_bytes,
1129 cluster_bytes - skip_end_bytes);
1138 qemu_vfree(whole_grain);
1142 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1145 offset = cpu_to_le32(offset);
1146 /* update L2 table */
1147 BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1148 if (bdrv_pwrite_sync(extent->file,
1149 ((int64_t)m_data->l2_offset * 512)
1150 + (m_data->l2_index * sizeof(offset)),
1151 &offset, sizeof(offset)) < 0) {
1154 /* update backup L2 table */
1155 if (extent->l1_backup_table_offset != 0) {
1156 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1157 if (bdrv_pwrite_sync(extent->file,
1158 ((int64_t)m_data->l2_offset * 512)
1159 + (m_data->l2_index * sizeof(offset)),
1160 &offset, sizeof(offset)) < 0) {
1164 if (m_data->l2_cache_entry) {
1165 *m_data->l2_cache_entry = offset;
1172 * get_cluster_offset
1174 * Look up cluster offset in extent file by sector number, and store in
1177 * For flat extents, the start offset as parsed from the description file is
1180 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1181 * offset for a new cluster and update L2 cache. If there is a backing file,
1182 * COW is done before returning; otherwise, zeroes are written to the allocated
1183 * cluster. Both COW and zero writing skips the sector range
1184 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1185 * has new data to write there.
1187 * Returns: VMDK_OK if cluster exists and mapped in the image.
1188 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1189 * VMDK_ERROR if failed.
1191 static int get_cluster_offset(BlockDriverState *bs,
1193 VmdkMetaData *m_data,
1196 uint64_t *cluster_offset,
1197 uint64_t skip_start_bytes,
1198 uint64_t skip_end_bytes)
1200 unsigned int l1_index, l2_offset, l2_index;
1201 int min_index, i, j;
1202 uint32_t min_count, *l2_table;
1203 bool zeroed = false;
1205 int64_t cluster_sector;
1211 *cluster_offset = extent->flat_start_offset;
1215 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1216 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1217 if (l1_index >= extent->l1_size) {
1220 l2_offset = extent->l1_table[l1_index];
1222 return VMDK_UNALLOC;
1224 for (i = 0; i < L2_CACHE_SIZE; i++) {
1225 if (l2_offset == extent->l2_cache_offsets[i]) {
1226 /* increment the hit count */
1227 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1228 for (j = 0; j < L2_CACHE_SIZE; j++) {
1229 extent->l2_cache_counts[j] >>= 1;
1232 l2_table = extent->l2_cache + (i * extent->l2_size);
1236 /* not found: load a new entry in the least used one */
1238 min_count = 0xffffffff;
1239 for (i = 0; i < L2_CACHE_SIZE; i++) {
1240 if (extent->l2_cache_counts[i] < min_count) {
1241 min_count = extent->l2_cache_counts[i];
1245 l2_table = extent->l2_cache + (min_index * extent->l2_size);
1246 BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1247 if (bdrv_pread(extent->file,
1248 (int64_t)l2_offset * 512,
1250 extent->l2_size * sizeof(uint32_t)
1251 ) != extent->l2_size * sizeof(uint32_t)) {
1255 extent->l2_cache_offsets[min_index] = l2_offset;
1256 extent->l2_cache_counts[min_index] = 1;
1258 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1259 cluster_sector = le32_to_cpu(l2_table[l2_index]);
1261 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1265 if (!cluster_sector || zeroed) {
1267 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1270 if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1274 cluster_sector = extent->next_cluster_sector;
1275 extent->next_cluster_sector += extent->cluster_sectors;
1277 /* First of all we write grain itself, to avoid race condition
1278 * that may to corrupt the image.
1279 * This problem may occur because of insufficient space on host disk
1280 * or inappropriate VM shutdown.
1282 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1283 offset, skip_start_bytes, skip_end_bytes);
1289 m_data->l1_index = l1_index;
1290 m_data->l2_index = l2_index;
1291 m_data->l2_offset = l2_offset;
1292 m_data->l2_cache_entry = &l2_table[l2_index];
1295 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1299 static VmdkExtent *find_extent(BDRVVmdkState *s,
1300 int64_t sector_num, VmdkExtent *start_hint)
1302 VmdkExtent *extent = start_hint;
1305 extent = &s->extents[0];
1307 while (extent < &s->extents[s->num_extents]) {
1308 if (sector_num < extent->end_sector) {
1316 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1319 uint64_t extent_begin_offset, extent_relative_offset;
1320 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1322 extent_begin_offset =
1323 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1324 extent_relative_offset = offset - extent_begin_offset;
1325 return extent_relative_offset % cluster_size;
1328 static int coroutine_fn vmdk_co_block_status(BlockDriverState *bs,
1330 int64_t offset, int64_t bytes,
1331 int64_t *pnum, int64_t *map,
1332 BlockDriverState **file)
1334 BDRVVmdkState *s = bs->opaque;
1335 int64_t index_in_cluster, n, ret;
1336 uint64_t cluster_offset;
1339 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1343 qemu_co_mutex_lock(&s->lock);
1344 ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1346 qemu_co_mutex_unlock(&s->lock);
1348 index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1357 ret = BDRV_BLOCK_ZERO;
1360 ret = BDRV_BLOCK_DATA;
1361 if (!extent->compressed) {
1362 ret |= BDRV_BLOCK_OFFSET_VALID;
1363 *map = cluster_offset + index_in_cluster;
1365 *file = extent->file->bs;
1369 n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1370 *pnum = MIN(n, bytes);
1374 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1375 int64_t offset_in_cluster, QEMUIOVector *qiov,
1376 uint64_t qiov_offset, uint64_t n_bytes,
1380 VmdkGrainMarker *data = NULL;
1382 QEMUIOVector local_qiov;
1383 int64_t write_offset;
1384 int64_t write_end_sector;
1386 if (extent->compressed) {
1387 void *compressed_data;
1389 if (!extent->has_marker) {
1393 buf_len = (extent->cluster_sectors << 9) * 2;
1394 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1396 compressed_data = g_malloc(n_bytes);
1397 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1398 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1399 g_free(compressed_data);
1401 if (ret != Z_OK || buf_len == 0) {
1406 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1407 data->size = cpu_to_le32(buf_len);
1409 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1410 qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1412 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1414 qemu_iovec_init(&local_qiov, qiov->niov);
1415 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1417 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1420 write_offset = cluster_offset + offset_in_cluster;
1421 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1424 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1426 if (extent->compressed) {
1427 extent->next_cluster_sector = write_end_sector;
1429 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1439 if (!extent->compressed) {
1440 qemu_iovec_destroy(&local_qiov);
1445 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1446 int64_t offset_in_cluster, QEMUIOVector *qiov,
1450 int cluster_bytes, buf_bytes;
1451 uint8_t *cluster_buf, *compressed_data;
1452 uint8_t *uncomp_buf;
1454 VmdkGrainMarker *marker;
1458 if (!extent->compressed) {
1459 BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1460 ret = bdrv_co_preadv(extent->file,
1461 cluster_offset + offset_in_cluster, bytes,
1468 cluster_bytes = extent->cluster_sectors * 512;
1469 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1470 buf_bytes = cluster_bytes * 2;
1471 cluster_buf = g_malloc(buf_bytes);
1472 uncomp_buf = g_malloc(cluster_bytes);
1473 BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1474 ret = bdrv_pread(extent->file,
1476 cluster_buf, buf_bytes);
1480 compressed_data = cluster_buf;
1481 buf_len = cluster_bytes;
1482 data_len = cluster_bytes;
1483 if (extent->has_marker) {
1484 marker = (VmdkGrainMarker *)cluster_buf;
1485 compressed_data = marker->data;
1486 data_len = le32_to_cpu(marker->size);
1488 if (!data_len || data_len > buf_bytes) {
1492 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1498 if (offset_in_cluster < 0 ||
1499 offset_in_cluster + bytes > buf_len) {
1503 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1508 g_free(cluster_buf);
1512 static int coroutine_fn
1513 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1514 QEMUIOVector *qiov, int flags)
1516 BDRVVmdkState *s = bs->opaque;
1518 uint64_t n_bytes, offset_in_cluster;
1519 VmdkExtent *extent = NULL;
1520 QEMUIOVector local_qiov;
1521 uint64_t cluster_offset;
1522 uint64_t bytes_done = 0;
1524 qemu_iovec_init(&local_qiov, qiov->niov);
1525 qemu_co_mutex_lock(&s->lock);
1528 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1533 ret = get_cluster_offset(bs, extent, NULL,
1534 offset, false, &cluster_offset, 0, 0);
1535 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1537 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1538 - offset_in_cluster);
1540 if (ret != VMDK_OK) {
1541 /* if not allocated, try to read from parent image, if exist */
1542 if (bs->backing && ret != VMDK_ZEROED) {
1543 if (!vmdk_is_cid_valid(bs)) {
1548 qemu_iovec_reset(&local_qiov);
1549 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1551 /* qcow2 emits this on bs->file instead of bs->backing */
1552 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1553 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1559 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1562 qemu_iovec_reset(&local_qiov);
1563 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1565 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1566 &local_qiov, n_bytes);
1573 bytes_done += n_bytes;
1578 qemu_co_mutex_unlock(&s->lock);
1579 qemu_iovec_destroy(&local_qiov);
1586 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1587 * if possible, otherwise return -ENOTSUP.
1588 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1589 * with each cluster. By dry run we can find if the zero write
1590 * is possible without modifying image data.
1592 * Returns: error code with 0 for success.
1594 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1595 uint64_t bytes, QEMUIOVector *qiov,
1596 bool zeroed, bool zero_dry_run)
1598 BDRVVmdkState *s = bs->opaque;
1599 VmdkExtent *extent = NULL;
1601 int64_t offset_in_cluster, n_bytes;
1602 uint64_t cluster_offset;
1603 uint64_t bytes_done = 0;
1604 VmdkMetaData m_data;
1606 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1607 error_report("Wrong offset: offset=0x%" PRIx64
1608 " total_sectors=0x%" PRIx64,
1609 offset, bs->total_sectors);
1614 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1618 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1619 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1620 - offset_in_cluster);
1622 ret = get_cluster_offset(bs, extent, &m_data, offset,
1623 !(extent->compressed || zeroed),
1624 &cluster_offset, offset_in_cluster,
1625 offset_in_cluster + n_bytes);
1626 if (extent->compressed) {
1627 if (ret == VMDK_OK) {
1628 /* Refuse write to allocated cluster for streamOptimized */
1629 error_report("Could not write to allocated cluster"
1630 " for streamOptimized");
1634 ret = get_cluster_offset(bs, extent, &m_data, offset,
1635 true, &cluster_offset, 0, 0);
1638 if (ret == VMDK_ERROR) {
1642 /* Do zeroed write, buf is ignored */
1643 if (extent->has_zero_grain &&
1644 offset_in_cluster == 0 &&
1645 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
1646 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1647 if (!zero_dry_run) {
1648 /* update L2 tables */
1649 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1658 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
1659 qiov, bytes_done, n_bytes, offset);
1664 /* update L2 tables */
1665 if (vmdk_L2update(extent, &m_data,
1666 cluster_offset >> BDRV_SECTOR_BITS)
1674 bytes_done += n_bytes;
1676 /* update CID on the first write every time the virtual disk is
1678 if (!s->cid_updated) {
1679 ret = vmdk_write_cid(bs, g_random_int());
1683 s->cid_updated = true;
1689 static int coroutine_fn
1690 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1691 QEMUIOVector *qiov, int flags)
1694 BDRVVmdkState *s = bs->opaque;
1695 qemu_co_mutex_lock(&s->lock);
1696 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
1697 qemu_co_mutex_unlock(&s->lock);
1701 static int coroutine_fn
1702 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1703 uint64_t bytes, QEMUIOVector *qiov)
1706 /* The caller will write bytes 0 to signal EOF.
1707 * When receive it, we align EOF to a sector boundary. */
1708 BDRVVmdkState *s = bs->opaque;
1712 for (i = 0; i < s->num_extents; i++) {
1713 length = bdrv_getlength(s->extents[i].file->bs);
1717 length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
1718 ret = bdrv_truncate(s->extents[i].file, length,
1719 PREALLOC_MODE_OFF, NULL);
1726 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
1729 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
1732 BdrvRequestFlags flags)
1735 BDRVVmdkState *s = bs->opaque;
1737 qemu_co_mutex_lock(&s->lock);
1738 /* write zeroes could fail if sectors not aligned to cluster, test it with
1739 * dry_run == true before really updating image */
1740 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
1742 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
1744 qemu_co_mutex_unlock(&s->lock);
1748 static int vmdk_init_extent(BlockBackend *blk,
1749 int64_t filesize, bool flat,
1750 bool compress, bool zeroed_grain,
1755 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1756 uint32_t *gd_buf = NULL;
1760 ret = blk_truncate(blk, filesize, PREALLOC_MODE_OFF, errp);
1763 magic = cpu_to_be32(VMDK4_MAGIC);
1764 memset(&header, 0, sizeof(header));
1767 } else if (zeroed_grain) {
1772 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1773 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1774 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1775 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1776 header.capacity = filesize / BDRV_SECTOR_SIZE;
1777 header.granularity = 128;
1778 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1780 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1781 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1783 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1784 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1786 header.desc_offset = 1;
1787 header.desc_size = 20;
1788 header.rgd_offset = header.desc_offset + header.desc_size;
1789 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1790 header.grain_offset =
1791 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1792 header.granularity);
1793 /* swap endianness for all header fields */
1794 header.version = cpu_to_le32(header.version);
1795 header.flags = cpu_to_le32(header.flags);
1796 header.capacity = cpu_to_le64(header.capacity);
1797 header.granularity = cpu_to_le64(header.granularity);
1798 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1799 header.desc_offset = cpu_to_le64(header.desc_offset);
1800 header.desc_size = cpu_to_le64(header.desc_size);
1801 header.rgd_offset = cpu_to_le64(header.rgd_offset);
1802 header.gd_offset = cpu_to_le64(header.gd_offset);
1803 header.grain_offset = cpu_to_le64(header.grain_offset);
1804 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1806 header.check_bytes[0] = 0xa;
1807 header.check_bytes[1] = 0x20;
1808 header.check_bytes[2] = 0xd;
1809 header.check_bytes[3] = 0xa;
1811 /* write all the data */
1812 ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
1814 error_setg(errp, QERR_IO_ERROR);
1817 ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
1819 error_setg(errp, QERR_IO_ERROR);
1823 ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9,
1824 PREALLOC_MODE_OFF, errp);
1829 /* write grain directory */
1830 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1831 gd_buf = g_malloc0(gd_buf_size);
1832 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1833 i < gt_count; i++, tmp += gt_size) {
1834 gd_buf[i] = cpu_to_le32(tmp);
1836 ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1837 gd_buf, gd_buf_size, 0);
1839 error_setg(errp, QERR_IO_ERROR);
1843 /* write backup grain directory */
1844 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1845 i < gt_count; i++, tmp += gt_size) {
1846 gd_buf[i] = cpu_to_le32(tmp);
1848 ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1849 gd_buf, gd_buf_size, 0);
1851 error_setg(errp, QERR_IO_ERROR);
1860 static int vmdk_create_extent(const char *filename, int64_t filesize,
1861 bool flat, bool compress, bool zeroed_grain,
1863 QemuOpts *opts, Error **errp)
1866 BlockBackend *blk = NULL;
1867 Error *local_err = NULL;
1869 ret = bdrv_create_file(filename, opts, &local_err);
1871 error_propagate(errp, local_err);
1875 blk = blk_new_open(filename, NULL, NULL,
1876 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
1879 error_propagate(errp, local_err);
1884 blk_set_allow_write_beyond_eof(blk, true);
1886 ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
1899 static int filename_decompose(const char *filename, char *path, char *prefix,
1900 char *postfix, size_t buf_len, Error **errp)
1904 if (filename == NULL || !strlen(filename)) {
1905 error_setg(errp, "No filename provided");
1908 p = strrchr(filename, '/');
1910 p = strrchr(filename, '\\');
1913 p = strrchr(filename, ':');
1917 if (p - filename >= buf_len) {
1920 pstrcpy(path, p - filename + 1, filename);
1925 q = strrchr(p, '.');
1927 pstrcpy(prefix, buf_len, p);
1930 if (q - p >= buf_len) {
1933 pstrcpy(prefix, q - p + 1, p);
1934 pstrcpy(postfix, buf_len, q);
1940 * idx == 0: get or create the descriptor file (also the image file if in a
1942 * idx >= 1: get the n-th extent if in a split subformat
1944 typedef BlockBackend *(*vmdk_create_extent_fn)(int64_t size,
1953 static void vmdk_desc_add_extent(GString *desc,
1954 const char *extent_line_fmt,
1955 int64_t size, const char *filename)
1957 char *basename = g_path_get_basename(filename);
1959 g_string_append_printf(desc, extent_line_fmt,
1960 DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
1964 static int coroutine_fn vmdk_co_do_create(int64_t size,
1965 BlockdevVmdkSubformat subformat,
1966 BlockdevVmdkAdapterType adapter_type,
1967 const char *backing_file,
1968 const char *hw_version,
1971 vmdk_create_extent_fn extent_fn,
1976 BlockBackend *blk = NULL;
1977 BlockBackend *extent_blk;
1978 Error *local_err = NULL;
1981 bool flat, split, compress;
1982 GString *ext_desc_lines;
1983 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
1984 int64_t extent_size;
1985 int64_t created_size = 0;
1986 const char *extent_line_fmt;
1987 char *parent_desc_line = g_malloc0(BUF_SIZE);
1988 uint32_t parent_cid = 0xffffffff;
1989 uint32_t number_heads = 16;
1990 uint32_t desc_offset = 0, desc_len;
1991 const char desc_template[] =
1992 "# Disk DescriptorFile\n"
1995 "parentCID=%" PRIx32 "\n"
1996 "createType=\"%s\"\n"
1999 "# Extent description\n"
2002 "# The Disk Data Base\n"
2005 "ddb.virtualHWVersion = \"%s\"\n"
2006 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2007 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2008 "ddb.geometry.sectors = \"63\"\n"
2009 "ddb.adapterType = \"%s\"\n";
2011 ext_desc_lines = g_string_new(NULL);
2013 /* Read out options */
2017 "compat6 cannot be enabled with hwversion set");
2027 if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2028 /* that's the number of heads with which vmware operates when
2029 creating, exporting, etc. vmdk files with a non-ide adapter type */
2032 split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2033 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2034 flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2035 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2036 compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2039 extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2041 extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2043 if (flat && backing_file) {
2044 error_setg(errp, "Flat image can't have backing file");
2048 if (flat && zeroed_grain) {
2049 error_setg(errp, "Flat image can't enable zeroed grain");
2054 /* Create extents */
2056 extent_size = split_size;
2060 if (!split && !flat) {
2061 created_size = extent_size;
2065 /* Get the descriptor file BDS */
2066 blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2072 if (!split && !flat) {
2073 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2074 blk_bs(blk)->filename);
2078 BlockBackend *backing;
2079 char *full_backing =
2080 bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2084 error_propagate(errp, local_err);
2088 assert(full_backing);
2090 backing = blk_new_open(full_backing, NULL, NULL,
2091 BDRV_O_NO_BACKING, errp);
2092 g_free(full_backing);
2093 if (backing == NULL) {
2097 if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2098 error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2099 blk_bs(backing)->drv->format_name);
2104 ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2107 error_setg(errp, "Failed to read parent CID");
2110 snprintf(parent_desc_line, BUF_SIZE,
2111 "parentFileNameHint=\"%s\"", backing_file);
2114 while (created_size < size) {
2115 int64_t cur_size = MIN(size - created_size, extent_size);
2116 extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2117 zeroed_grain, opaque, errp);
2122 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2123 blk_bs(extent_blk)->filename);
2124 created_size += cur_size;
2126 blk_unref(extent_blk);
2129 /* Check whether we got excess extents */
2130 extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2133 blk_unref(extent_blk);
2134 error_setg(errp, "List of extents contains unused extents");
2139 /* generate descriptor file */
2140 desc = g_strdup_printf(desc_template,
2143 BlockdevVmdkSubformat_str(subformat),
2145 ext_desc_lines->str,
2148 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2150 BlockdevVmdkAdapterType_str(adapter_type));
2151 desc_len = strlen(desc);
2152 /* the descriptor offset = 0x200 */
2153 if (!split && !flat) {
2154 desc_offset = 0x200;
2157 ret = blk_pwrite(blk, desc_offset, desc, desc_len, 0);
2159 error_setg_errno(errp, -ret, "Could not write description");
2162 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2163 * for description file */
2164 if (desc_offset == 0) {
2165 ret = blk_truncate(blk, desc_len, PREALLOC_MODE_OFF, errp);
2176 g_free(parent_desc_line);
2177 g_string_free(ext_desc_lines, true);
2186 } VMDKCreateOptsData;
2188 static BlockBackend *vmdk_co_create_opts_cb(int64_t size, int idx,
2189 bool flat, bool split, bool compress,
2190 bool zeroed_grain, void *opaque,
2193 BlockBackend *blk = NULL;
2194 BlockDriverState *bs = NULL;
2195 VMDKCreateOptsData *data = opaque;
2196 char *ext_filename = NULL;
2197 char *rel_filename = NULL;
2199 /* We're done, don't create excess extents. */
2201 assert(errp == NULL);
2206 rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2208 rel_filename = g_strdup_printf("%s-%c%03d%s",
2210 flat ? 'f' : 's', idx, data->postfix);
2213 rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2216 ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2217 g_free(rel_filename);
2219 if (vmdk_create_extent(ext_filename, size,
2220 flat, compress, zeroed_grain, &blk, data->opts,
2226 g_free(ext_filename);
2230 static int coroutine_fn vmdk_co_create_opts(const char *filename, QemuOpts *opts,
2233 Error *local_err = NULL;
2235 int64_t total_size = 0;
2236 char *adapter_type = NULL;
2237 BlockdevVmdkAdapterType adapter_type_enum;
2238 char *backing_file = NULL;
2239 char *hw_version = NULL;
2241 BlockdevVmdkSubformat subformat;
2243 char *path = g_malloc0(PATH_MAX);
2244 char *prefix = g_malloc0(PATH_MAX);
2245 char *postfix = g_malloc0(PATH_MAX);
2246 char *desc_line = g_malloc0(BUF_SIZE);
2247 char *ext_filename = g_malloc0(PATH_MAX);
2248 char *desc_filename = g_malloc0(PATH_MAX);
2249 char *parent_desc_line = g_malloc0(BUF_SIZE);
2252 VMDKCreateOptsData data;
2254 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
2258 /* Read out options */
2259 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2261 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
2262 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2263 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
2264 compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false);
2265 if (strcmp(hw_version, "undefined") == 0) {
2269 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
2270 zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false);
2273 adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup,
2275 BLOCKDEV_VMDK_ADAPTER_TYPE_IDE,
2278 error_propagate(errp, local_err);
2283 adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE;
2287 /* Default format to monolithicSparse */
2288 subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE;
2290 subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup,
2292 BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE,
2295 error_propagate(errp, local_err);
2300 data = (VMDKCreateOptsData){
2306 ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum,
2307 backing_file, hw_version, compat6, zeroed_grain,
2308 vmdk_co_create_opts_cb, &data, errp);
2311 g_free(adapter_type);
2312 g_free(backing_file);
2320 g_free(ext_filename);
2321 g_free(desc_filename);
2322 g_free(parent_desc_line);
2326 static BlockBackend *vmdk_co_create_cb(int64_t size, int idx,
2327 bool flat, bool split, bool compress,
2328 bool zeroed_grain, void *opaque,
2332 BlockDriverState *bs;
2334 BlockdevCreateOptionsVmdk *opts = opaque;
2337 bs = bdrv_open_blockdev_ref(opts->file, errp);
2340 BlockdevRefList *list = opts->extents;
2341 for (i = 1; i < idx; i++) {
2342 if (!list || !list->next) {
2343 error_setg(errp, "Extent [%d] not specified", i);
2349 error_setg(errp, "Extent [%d] not specified", idx - 1);
2352 bs = bdrv_open_blockdev_ref(list->value, errp);
2357 blk = blk_new(BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | BLK_PERM_RESIZE,
2359 if (blk_insert_bs(blk, bs, errp)) {
2363 blk_set_allow_write_beyond_eof(blk, true);
2367 ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2376 static int coroutine_fn vmdk_co_create(BlockdevCreateOptions *create_options,
2380 BlockdevCreateOptionsVmdk *opts;
2382 opts = &create_options->u.vmdk;
2384 /* Validate options */
2385 if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2386 error_setg(errp, "Image size must be a multiple of 512 bytes");
2391 ret = vmdk_co_do_create(opts->size,
2406 static void vmdk_close(BlockDriverState *bs)
2408 BDRVVmdkState *s = bs->opaque;
2410 vmdk_free_extents(bs);
2411 g_free(s->create_type);
2413 migrate_del_blocker(s->migration_blocker);
2414 error_free(s->migration_blocker);
2417 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2419 BDRVVmdkState *s = bs->opaque;
2423 for (i = 0; i < s->num_extents; i++) {
2424 err = bdrv_co_flush(s->extents[i].file->bs);
2432 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2437 BDRVVmdkState *s = bs->opaque;
2439 ret = bdrv_get_allocated_file_size(bs->file->bs);
2443 for (i = 0; i < s->num_extents; i++) {
2444 if (s->extents[i].file == bs->file) {
2447 r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2456 static int vmdk_has_zero_init(BlockDriverState *bs)
2459 BDRVVmdkState *s = bs->opaque;
2461 /* If has a flat extent and its underlying storage doesn't have zero init,
2463 for (i = 0; i < s->num_extents; i++) {
2464 if (s->extents[i].flat) {
2465 if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2473 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2475 ImageInfo *info = g_new0(ImageInfo, 1);
2477 bdrv_refresh_filename(extent->file->bs);
2478 *info = (ImageInfo){
2479 .filename = g_strdup(extent->file->bs->filename),
2480 .format = g_strdup(extent->type),
2481 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2482 .compressed = extent->compressed,
2483 .has_compressed = extent->compressed,
2484 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2485 .has_cluster_size = !extent->flat,
2491 static int coroutine_fn vmdk_co_check(BlockDriverState *bs,
2492 BdrvCheckResult *result,
2495 BDRVVmdkState *s = bs->opaque;
2496 VmdkExtent *extent = NULL;
2497 int64_t sector_num = 0;
2498 int64_t total_sectors = bdrv_nb_sectors(bs);
2500 uint64_t cluster_offset;
2507 if (sector_num >= total_sectors) {
2510 extent = find_extent(s, sector_num, extent);
2513 "ERROR: could not find extent for sector %" PRId64 "\n",
2518 ret = get_cluster_offset(bs, extent, NULL,
2519 sector_num << BDRV_SECTOR_BITS,
2520 false, &cluster_offset, 0, 0);
2521 if (ret == VMDK_ERROR) {
2523 "ERROR: could not get cluster_offset for sector %"
2524 PRId64 "\n", sector_num);
2527 if (ret == VMDK_OK) {
2528 int64_t extent_len = bdrv_getlength(extent->file->bs);
2529 if (extent_len < 0) {
2531 "ERROR: could not get extent file length for sector %"
2532 PRId64 "\n", sector_num);
2536 if (cluster_offset >= extent_len) {
2538 "ERROR: cluster offset for sector %"
2539 PRId64 " points after EOF\n", sector_num);
2544 sector_num += extent->cluster_sectors;
2547 result->corruptions++;
2551 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs,
2555 BDRVVmdkState *s = bs->opaque;
2556 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2557 ImageInfoList **next;
2559 *spec_info = (ImageInfoSpecific){
2560 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2562 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2566 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2567 .create_type = g_strdup(s->create_type),
2569 .parent_cid = s->parent_cid,
2572 next = &spec_info->u.vmdk.data->extents;
2573 for (i = 0; i < s->num_extents; i++) {
2574 *next = g_new0(ImageInfoList, 1);
2575 (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2576 (*next)->next = NULL;
2577 next = &(*next)->next;
2583 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2585 return a->flat == b->flat &&
2586 a->compressed == b->compressed &&
2587 (a->flat || a->cluster_sectors == b->cluster_sectors);
2590 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2593 BDRVVmdkState *s = bs->opaque;
2594 assert(s->num_extents);
2596 /* See if we have multiple extents but they have different cases */
2597 for (i = 1; i < s->num_extents; i++) {
2598 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2602 bdi->needs_compressed_writes = s->extents[0].compressed;
2603 if (!s->extents[0].flat) {
2604 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2609 static void vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
2610 bool backing_overridden)
2612 /* No children but file and backing can be explicitly specified (TODO) */
2613 qdict_put(target, "file",
2614 qobject_ref(bs->file->bs->full_open_options));
2616 if (backing_overridden) {
2618 qdict_put(target, "backing",
2619 qobject_ref(bs->backing->bs->full_open_options));
2621 qdict_put_null(target, "backing");
2626 static QemuOptsList vmdk_create_opts = {
2627 .name = "vmdk-create-opts",
2628 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2631 .name = BLOCK_OPT_SIZE,
2632 .type = QEMU_OPT_SIZE,
2633 .help = "Virtual disk size"
2636 .name = BLOCK_OPT_ADAPTER_TYPE,
2637 .type = QEMU_OPT_STRING,
2638 .help = "Virtual adapter type, can be one of "
2639 "ide (default), lsilogic, buslogic or legacyESX"
2642 .name = BLOCK_OPT_BACKING_FILE,
2643 .type = QEMU_OPT_STRING,
2644 .help = "File name of a base image"
2647 .name = BLOCK_OPT_COMPAT6,
2648 .type = QEMU_OPT_BOOL,
2649 .help = "VMDK version 6 image",
2650 .def_value_str = "off"
2653 .name = BLOCK_OPT_HWVERSION,
2654 .type = QEMU_OPT_STRING,
2655 .help = "VMDK hardware version",
2656 .def_value_str = "undefined"
2659 .name = BLOCK_OPT_SUBFMT,
2660 .type = QEMU_OPT_STRING,
2662 "VMDK flat extent format, can be one of "
2663 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2666 .name = BLOCK_OPT_ZEROED_GRAIN,
2667 .type = QEMU_OPT_BOOL,
2668 .help = "Enable efficient zero writes "
2669 "using the zeroed-grain GTE feature"
2671 { /* end of list */ }
2675 static BlockDriver bdrv_vmdk = {
2676 .format_name = "vmdk",
2677 .instance_size = sizeof(BDRVVmdkState),
2678 .bdrv_probe = vmdk_probe,
2679 .bdrv_open = vmdk_open,
2680 .bdrv_co_check = vmdk_co_check,
2681 .bdrv_reopen_prepare = vmdk_reopen_prepare,
2682 .bdrv_child_perm = bdrv_format_default_perms,
2683 .bdrv_co_preadv = vmdk_co_preadv,
2684 .bdrv_co_pwritev = vmdk_co_pwritev,
2685 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
2686 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
2687 .bdrv_close = vmdk_close,
2688 .bdrv_co_create_opts = vmdk_co_create_opts,
2689 .bdrv_co_create = vmdk_co_create,
2690 .bdrv_co_flush_to_disk = vmdk_co_flush,
2691 .bdrv_co_block_status = vmdk_co_block_status,
2692 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2693 .bdrv_has_zero_init = vmdk_has_zero_init,
2694 .bdrv_get_specific_info = vmdk_get_specific_info,
2695 .bdrv_refresh_limits = vmdk_refresh_limits,
2696 .bdrv_get_info = vmdk_get_info,
2697 .bdrv_gather_child_options = vmdk_gather_child_options,
2699 .supports_backing = true,
2700 .create_opts = &vmdk_create_opts,
2703 static void bdrv_vmdk_init(void)
2705 bdrv_register(&bdrv_vmdk);
2708 block_init(bdrv_vmdk_init);