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/option.h"
34 #include "qemu/bswap.h"
35 #include "migration/blocker.h"
36 #include "qemu/cutils.h"
39 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
40 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
41 #define VMDK4_COMPRESSION_DEFLATE 1
42 #define VMDK4_FLAG_NL_DETECT (1 << 0)
43 #define VMDK4_FLAG_RGD (1 << 1)
44 /* Zeroed-grain enable bit */
45 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
46 #define VMDK4_FLAG_COMPRESS (1 << 16)
47 #define VMDK4_FLAG_MARKER (1 << 17)
48 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
50 #define VMDK_EXTENT_MAX_SECTORS (1ULL << 32)
52 #define VMDK_GTE_ZEROED 0x1
54 /* VMDK internal error codes */
56 #define VMDK_ERROR (-1)
57 /* Cluster not allocated */
58 #define VMDK_UNALLOC (-2)
59 #define VMDK_ZEROED (-3)
61 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
66 uint32_t disk_sectors;
68 uint32_t l1dir_offset;
70 uint32_t file_sectors;
73 uint32_t sectors_per_track;
74 } QEMU_PACKED VMDK3Header;
83 /* Number of GrainTableEntries per GrainTable */
84 uint32_t num_gtes_per_gt;
87 uint64_t grain_offset;
90 uint16_t compressAlgorithm;
91 } QEMU_PACKED VMDK4Header;
93 #define L2_CACHE_SIZE 16
95 typedef struct VmdkExtent {
104 int64_t flat_start_offset;
105 int64_t l1_table_offset;
106 int64_t l1_backup_table_offset;
108 uint32_t *l1_backup_table;
109 unsigned int l1_size;
110 uint32_t l1_entry_sectors;
112 unsigned int l2_size;
114 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
115 uint32_t l2_cache_counts[L2_CACHE_SIZE];
117 int64_t cluster_sectors;
118 int64_t next_cluster_sector;
122 typedef struct BDRVVmdkState {
124 uint64_t desc_offset;
130 /* Extent array with num_extents entries, ascend ordered by address */
132 Error *migration_blocker;
136 typedef struct VmdkMetaData {
137 unsigned int l1_index;
138 unsigned int l2_index;
139 unsigned int l2_offset;
141 uint32_t *l2_cache_entry;
144 typedef struct VmdkGrainMarker {
148 } QEMU_PACKED VmdkGrainMarker;
151 MARKER_END_OF_STREAM = 0,
152 MARKER_GRAIN_TABLE = 1,
153 MARKER_GRAIN_DIRECTORY = 2,
157 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
164 magic = be32_to_cpu(*(uint32_t *)buf);
165 if (magic == VMDK3_MAGIC ||
166 magic == VMDK4_MAGIC) {
169 const char *p = (const char *)buf;
170 const char *end = p + buf_size;
173 /* skip comment line */
174 while (p < end && *p != '\n') {
181 while (p < end && *p == ' ') {
184 /* skip '\r' if windows line endings used. */
185 if (p < end && *p == '\r') {
188 /* only accept blank lines before 'version=' line */
189 if (p == end || *p != '\n') {
195 if (end - p >= strlen("version=X\n")) {
196 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
197 strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
201 if (end - p >= strlen("version=X\r\n")) {
202 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
203 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
213 #define SECTOR_SIZE 512
214 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
215 #define BUF_SIZE 4096
216 #define HEADER_SIZE 512 /* first sector of 512 bytes */
218 static void vmdk_free_extents(BlockDriverState *bs)
221 BDRVVmdkState *s = bs->opaque;
224 for (i = 0; i < s->num_extents; i++) {
228 g_free(e->l1_backup_table);
230 if (e->file != bs->file) {
231 bdrv_unref_child(bs, e->file);
237 static void vmdk_free_last_extent(BlockDriverState *bs)
239 BDRVVmdkState *s = bs->opaque;
241 if (s->num_extents == 0) {
245 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
248 /* Return -ve errno, or 0 on success and write CID into *pcid. */
249 static int vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
253 const char *p_name, *cid_str;
255 BDRVVmdkState *s = bs->opaque;
258 desc = g_malloc0(DESC_SIZE);
259 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
265 cid_str = "parentCID";
266 cid_str_size = sizeof("parentCID");
269 cid_str_size = sizeof("CID");
272 desc[DESC_SIZE - 1] = '\0';
273 p_name = strstr(desc, cid_str);
274 if (p_name == NULL) {
278 p_name += cid_str_size;
279 if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
291 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
293 char *desc, *tmp_desc;
294 char *p_name, *tmp_str;
295 BDRVVmdkState *s = bs->opaque;
298 desc = g_malloc0(DESC_SIZE);
299 tmp_desc = g_malloc0(DESC_SIZE);
300 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
305 desc[DESC_SIZE - 1] = '\0';
306 tmp_str = strstr(desc, "parentCID");
307 if (tmp_str == NULL) {
312 pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
313 p_name = strstr(desc, "CID");
314 if (p_name != NULL) {
315 p_name += sizeof("CID");
316 snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
317 pstrcat(desc, DESC_SIZE, tmp_desc);
320 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
328 static int vmdk_is_cid_valid(BlockDriverState *bs)
330 BDRVVmdkState *s = bs->opaque;
333 if (!s->cid_checked && bs->backing) {
334 BlockDriverState *p_bs = bs->backing->bs;
336 if (strcmp(p_bs->drv->format_name, "vmdk")) {
337 /* Backing file is not in vmdk format, so it does not have
338 * a CID, which makes the overlay's parent CID invalid */
342 if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
343 /* read failure: report as not valid */
346 if (s->parent_cid != cur_pcid) {
351 s->cid_checked = true;
356 /* We have nothing to do for VMDK reopen, stubs just return success */
357 static int vmdk_reopen_prepare(BDRVReopenState *state,
358 BlockReopenQueue *queue, Error **errp)
360 assert(state != NULL);
361 assert(state->bs != NULL);
365 static int vmdk_parent_open(BlockDriverState *bs)
369 BDRVVmdkState *s = bs->opaque;
372 desc = g_malloc0(DESC_SIZE + 1);
373 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
379 p_name = strstr(desc, "parentFileNameHint");
380 if (p_name != NULL) {
383 p_name += sizeof("parentFileNameHint") + 1;
384 end_name = strchr(p_name, '\"');
385 if (end_name == NULL) {
389 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
394 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
402 /* Create and append extent to the extent array. Return the added VmdkExtent
403 * address. return NULL if allocation failed. */
404 static int vmdk_add_extent(BlockDriverState *bs,
405 BdrvChild *file, bool flat, int64_t sectors,
406 int64_t l1_offset, int64_t l1_backup_offset,
408 int l2_size, uint64_t cluster_sectors,
409 VmdkExtent **new_extent,
413 BDRVVmdkState *s = bs->opaque;
416 if (cluster_sectors > 0x200000) {
417 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
418 error_setg(errp, "Invalid granularity, image may be corrupt");
421 if (l1_size > 512 * 1024 * 1024) {
422 /* Although with big capacity and small l1_entry_sectors, we can get a
423 * big l1_size, we don't want unbounded value to allocate the table.
424 * Limit it to 512M, which is 16PB for default cluster and L2 table
426 error_setg(errp, "L1 size too big");
430 nb_sectors = bdrv_nb_sectors(file->bs);
431 if (nb_sectors < 0) {
435 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
436 extent = &s->extents[s->num_extents];
439 memset(extent, 0, sizeof(VmdkExtent));
442 extent->sectors = sectors;
443 extent->l1_table_offset = l1_offset;
444 extent->l1_backup_table_offset = l1_backup_offset;
445 extent->l1_size = l1_size;
446 extent->l1_entry_sectors = l2_size * cluster_sectors;
447 extent->l2_size = l2_size;
448 extent->cluster_sectors = flat ? sectors : cluster_sectors;
449 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
451 if (s->num_extents > 1) {
452 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
454 extent->end_sector = extent->sectors;
456 bs->total_sectors = extent->end_sector;
458 *new_extent = extent;
463 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
470 /* read the L1 table */
471 l1_size = extent->l1_size * sizeof(uint32_t);
472 extent->l1_table = g_try_malloc(l1_size);
473 if (l1_size && extent->l1_table == NULL) {
477 ret = bdrv_pread(extent->file,
478 extent->l1_table_offset,
482 bdrv_refresh_filename(extent->file->bs);
483 error_setg_errno(errp, -ret,
484 "Could not read l1 table from extent '%s'",
485 extent->file->bs->filename);
488 for (i = 0; i < extent->l1_size; i++) {
489 le32_to_cpus(&extent->l1_table[i]);
492 if (extent->l1_backup_table_offset) {
493 extent->l1_backup_table = g_try_malloc(l1_size);
494 if (l1_size && extent->l1_backup_table == NULL) {
498 ret = bdrv_pread(extent->file,
499 extent->l1_backup_table_offset,
500 extent->l1_backup_table,
503 bdrv_refresh_filename(extent->file->bs);
504 error_setg_errno(errp, -ret,
505 "Could not read l1 backup table from extent '%s'",
506 extent->file->bs->filename);
509 for (i = 0; i < extent->l1_size; i++) {
510 le32_to_cpus(&extent->l1_backup_table[i]);
515 g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE);
518 g_free(extent->l1_backup_table);
520 g_free(extent->l1_table);
524 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
526 int flags, Error **errp)
533 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
535 bdrv_refresh_filename(file->bs);
536 error_setg_errno(errp, -ret,
537 "Could not read header from file '%s'",
541 ret = vmdk_add_extent(bs, file, false,
542 le32_to_cpu(header.disk_sectors),
543 (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
545 le32_to_cpu(header.l1dir_size),
547 le32_to_cpu(header.granularity),
553 ret = vmdk_init_tables(bs, extent, errp);
555 /* free extent allocated by vmdk_add_extent */
556 vmdk_free_last_extent(bs);
561 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
562 QDict *options, Error **errp);
564 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
570 size = bdrv_getlength(file->bs);
572 error_setg_errno(errp, -size, "Could not access file");
577 /* Both descriptor file and sparse image must be much larger than 4
578 * bytes, also callers of vmdk_read_desc want to compare the first 4
579 * bytes with VMDK4_MAGIC, let's error out if less is read. */
580 error_setg(errp, "File is too small, not a valid image");
584 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */
585 buf = g_malloc(size + 1);
587 ret = bdrv_pread(file, desc_offset, buf, size);
589 error_setg_errno(errp, -ret, "Could not read from file");
598 static int vmdk_open_vmdk4(BlockDriverState *bs,
600 int flags, QDict *options, Error **errp)
604 uint32_t l1_size, l1_entry_sectors;
607 BDRVVmdkState *s = bs->opaque;
608 int64_t l1_backup_offset = 0;
611 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
613 bdrv_refresh_filename(file->bs);
614 error_setg_errno(errp, -ret,
615 "Could not read header from file '%s'",
619 if (header.capacity == 0) {
620 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
622 char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
626 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
632 if (!s->create_type) {
633 s->create_type = g_strdup("monolithicSparse");
636 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
638 * The footer takes precedence over the header, so read it in. The
639 * footer starts at offset -1024 from the end: One sector for the
640 * footer, and another one for the end-of-stream marker.
647 uint8_t pad[512 - 16];
648 } QEMU_PACKED footer_marker;
652 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
658 uint8_t pad[512 - 16];
659 } QEMU_PACKED eos_marker;
660 } QEMU_PACKED footer;
662 ret = bdrv_pread(file,
663 bs->file->bs->total_sectors * 512 - 1536,
664 &footer, sizeof(footer));
666 error_setg_errno(errp, -ret, "Failed to read footer");
670 /* Some sanity checks for the footer */
671 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
672 le32_to_cpu(footer.footer_marker.size) != 0 ||
673 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
674 le64_to_cpu(footer.eos_marker.val) != 0 ||
675 le32_to_cpu(footer.eos_marker.size) != 0 ||
676 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
678 error_setg(errp, "Invalid footer");
682 header = footer.header;
686 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
687 if (le32_to_cpu(header.version) > 3) {
688 error_setg(errp, "Unsupported VMDK version %" PRIu32,
689 le32_to_cpu(header.version));
691 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
693 /* VMware KB 2064959 explains that version 3 added support for
694 * persistent changed block tracking (CBT), and backup software can
695 * read it as version=1 if it doesn't care about the changed area
696 * information. So we are safe to enable read only. */
697 error_setg(errp, "VMDK version 3 must be read only");
701 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
702 error_setg(errp, "L2 table size too big");
706 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
707 * le64_to_cpu(header.granularity);
708 if (l1_entry_sectors == 0) {
709 error_setg(errp, "L1 entry size is invalid");
712 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
714 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
715 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
717 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
718 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
719 (int64_t)(le64_to_cpu(header.grain_offset)
720 * BDRV_SECTOR_SIZE));
724 ret = vmdk_add_extent(bs, file, false,
725 le64_to_cpu(header.capacity),
726 le64_to_cpu(header.gd_offset) << 9,
729 le32_to_cpu(header.num_gtes_per_gt),
730 le64_to_cpu(header.granularity),
737 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
738 if (extent->compressed) {
739 g_free(s->create_type);
740 s->create_type = g_strdup("streamOptimized");
742 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
743 extent->version = le32_to_cpu(header.version);
744 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
745 ret = vmdk_init_tables(bs, extent, errp);
747 /* free extent allocated by vmdk_add_extent */
748 vmdk_free_last_extent(bs);
753 /* find an option value out of descriptor file */
754 static int vmdk_parse_description(const char *desc, const char *opt_name,
755 char *buf, int buf_size)
757 char *opt_pos, *opt_end;
758 const char *end = desc + strlen(desc);
760 opt_pos = strstr(desc, opt_name);
764 /* Skip "=\"" following opt_name */
765 opt_pos += strlen(opt_name) + 2;
766 if (opt_pos >= end) {
770 while (opt_end < end && *opt_end != '"') {
773 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
776 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
780 /* Open an extent file and append to bs array */
781 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
782 char *buf, QDict *options, Error **errp)
786 magic = ldl_be_p(buf);
789 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
792 return vmdk_open_vmdk4(bs, file, flags, options, errp);
795 error_setg(errp, "Image not in VMDK format");
801 static const char *next_line(const char *s)
812 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
813 const char *desc_file_path, QDict *options,
825 BdrvChild *extent_file;
826 BDRVVmdkState *s = bs->opaque;
828 char extent_opt_prefix[32];
829 Error *local_err = NULL;
831 for (p = desc; *p; p = next_line(p)) {
832 /* parse extent line in one of below formats:
834 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
835 * RW [size in sectors] SPARSE "file-name.vmdk"
836 * RW [size in sectors] VMFS "file-name.vmdk"
837 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
840 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
841 access, §ors, type, fname, &flat_offset);
842 if (matches < 4 || strcmp(access, "RW")) {
844 } else if (!strcmp(type, "FLAT")) {
845 if (matches != 5 || flat_offset < 0) {
848 } else if (!strcmp(type, "VMFS")) {
854 } else if (matches != 4) {
859 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
860 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
861 (strcmp(access, "RW"))) {
865 if (!path_is_absolute(fname) && !path_has_protocol(fname) &&
868 bdrv_refresh_filename(bs->file->bs);
869 error_setg(errp, "Cannot use relative extent paths with VMDK "
870 "descriptor file '%s'", bs->file->bs->filename);
874 extent_path = g_malloc0(PATH_MAX);
875 path_combine(extent_path, PATH_MAX, desc_file_path, fname);
877 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
880 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
881 bs, &child_file, false, &local_err);
884 error_propagate(errp, local_err);
888 /* save to extents array */
889 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
892 ret = vmdk_add_extent(bs, extent_file, true, sectors,
893 0, 0, 0, 0, 0, &extent, errp);
895 bdrv_unref_child(bs, extent_file);
898 extent->flat_start_offset = flat_offset << 9;
899 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
900 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
901 char *buf = vmdk_read_desc(extent_file, 0, errp);
905 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
910 bdrv_unref_child(bs, extent_file);
913 extent = &s->extents[s->num_extents - 1];
915 error_setg(errp, "Unsupported extent type '%s'", type);
916 bdrv_unref_child(bs, extent_file);
919 extent->type = g_strdup(type);
926 if (np[-1] == '\n') {
929 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
933 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
934 QDict *options, Error **errp)
938 BDRVVmdkState *s = bs->opaque;
940 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
941 error_setg(errp, "invalid VMDK image descriptor");
945 if (strcmp(ct, "monolithicFlat") &&
946 strcmp(ct, "vmfs") &&
947 strcmp(ct, "vmfsSparse") &&
948 strcmp(ct, "twoGbMaxExtentSparse") &&
949 strcmp(ct, "twoGbMaxExtentFlat")) {
950 error_setg(errp, "Unsupported image type '%s'", ct);
954 s->create_type = g_strdup(ct);
956 ret = vmdk_parse_extents(buf, bs, bs->file->bs->exact_filename, options,
962 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
967 BDRVVmdkState *s = bs->opaque;
969 Error *local_err = NULL;
971 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
977 buf = vmdk_read_desc(bs->file, 0, errp);
982 magic = ldl_be_p(buf);
986 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
988 s->desc_offset = 0x200;
991 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
998 /* try to open parent images, if exist */
999 ret = vmdk_parent_open(bs);
1003 ret = vmdk_read_cid(bs, 0, &s->cid);
1007 ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1011 qemu_co_mutex_init(&s->lock);
1013 /* Disable migration when VMDK images are used */
1014 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1015 "does not support live migration",
1016 bdrv_get_device_or_node_name(bs));
1017 ret = migrate_add_blocker(s->migration_blocker, &local_err);
1019 error_propagate(errp, local_err);
1020 error_free(s->migration_blocker);
1029 g_free(s->create_type);
1030 s->create_type = NULL;
1031 vmdk_free_extents(bs);
1036 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1038 BDRVVmdkState *s = bs->opaque;
1041 for (i = 0; i < s->num_extents; i++) {
1042 if (!s->extents[i].flat) {
1043 bs->bl.pwrite_zeroes_alignment =
1044 MAX(bs->bl.pwrite_zeroes_alignment,
1045 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1053 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1054 * to the cluster at @cluster_sector_num.
1056 * If @skip_start_sector < @skip_end_sector, the relative range
1057 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1058 * it for call to write user data in the request.
1060 static int get_whole_cluster(BlockDriverState *bs,
1062 uint64_t cluster_offset,
1064 uint64_t skip_start_bytes,
1065 uint64_t skip_end_bytes)
1068 int64_t cluster_bytes;
1069 uint8_t *whole_grain;
1071 /* For COW, align request sector_num to cluster start */
1072 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1073 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1074 whole_grain = qemu_blockalign(bs, cluster_bytes);
1077 memset(whole_grain, 0, skip_start_bytes);
1078 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1081 assert(skip_end_bytes <= cluster_bytes);
1082 /* we will be here if it's first write on non-exist grain(cluster).
1083 * try to read from parent image, if exist */
1084 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1089 /* Read backing data before skip range */
1090 if (skip_start_bytes > 0) {
1092 /* qcow2 emits this on bs->file instead of bs->backing */
1093 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1094 ret = bdrv_pread(bs->backing, offset, whole_grain,
1101 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1102 ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1109 /* Read backing data after skip range */
1110 if (skip_end_bytes < cluster_bytes) {
1112 /* qcow2 emits this on bs->file instead of bs->backing */
1113 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1114 ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1115 whole_grain + skip_end_bytes,
1116 cluster_bytes - skip_end_bytes);
1122 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1123 ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1124 whole_grain + skip_end_bytes,
1125 cluster_bytes - skip_end_bytes);
1134 qemu_vfree(whole_grain);
1138 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1141 offset = cpu_to_le32(offset);
1142 /* update L2 table */
1143 BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1144 if (bdrv_pwrite_sync(extent->file,
1145 ((int64_t)m_data->l2_offset * 512)
1146 + (m_data->l2_index * sizeof(offset)),
1147 &offset, sizeof(offset)) < 0) {
1150 /* update backup L2 table */
1151 if (extent->l1_backup_table_offset != 0) {
1152 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1153 if (bdrv_pwrite_sync(extent->file,
1154 ((int64_t)m_data->l2_offset * 512)
1155 + (m_data->l2_index * sizeof(offset)),
1156 &offset, sizeof(offset)) < 0) {
1160 if (m_data->l2_cache_entry) {
1161 *m_data->l2_cache_entry = offset;
1168 * get_cluster_offset
1170 * Look up cluster offset in extent file by sector number, and store in
1173 * For flat extents, the start offset as parsed from the description file is
1176 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1177 * offset for a new cluster and update L2 cache. If there is a backing file,
1178 * COW is done before returning; otherwise, zeroes are written to the allocated
1179 * cluster. Both COW and zero writing skips the sector range
1180 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1181 * has new data to write there.
1183 * Returns: VMDK_OK if cluster exists and mapped in the image.
1184 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1185 * VMDK_ERROR if failed.
1187 static int get_cluster_offset(BlockDriverState *bs,
1189 VmdkMetaData *m_data,
1192 uint64_t *cluster_offset,
1193 uint64_t skip_start_bytes,
1194 uint64_t skip_end_bytes)
1196 unsigned int l1_index, l2_offset, l2_index;
1197 int min_index, i, j;
1198 uint32_t min_count, *l2_table;
1199 bool zeroed = false;
1201 int64_t cluster_sector;
1207 *cluster_offset = extent->flat_start_offset;
1211 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1212 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1213 if (l1_index >= extent->l1_size) {
1216 l2_offset = extent->l1_table[l1_index];
1218 return VMDK_UNALLOC;
1220 for (i = 0; i < L2_CACHE_SIZE; i++) {
1221 if (l2_offset == extent->l2_cache_offsets[i]) {
1222 /* increment the hit count */
1223 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1224 for (j = 0; j < L2_CACHE_SIZE; j++) {
1225 extent->l2_cache_counts[j] >>= 1;
1228 l2_table = extent->l2_cache + (i * extent->l2_size);
1232 /* not found: load a new entry in the least used one */
1234 min_count = 0xffffffff;
1235 for (i = 0; i < L2_CACHE_SIZE; i++) {
1236 if (extent->l2_cache_counts[i] < min_count) {
1237 min_count = extent->l2_cache_counts[i];
1241 l2_table = extent->l2_cache + (min_index * extent->l2_size);
1242 BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1243 if (bdrv_pread(extent->file,
1244 (int64_t)l2_offset * 512,
1246 extent->l2_size * sizeof(uint32_t)
1247 ) != extent->l2_size * sizeof(uint32_t)) {
1251 extent->l2_cache_offsets[min_index] = l2_offset;
1252 extent->l2_cache_counts[min_index] = 1;
1254 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1255 cluster_sector = le32_to_cpu(l2_table[l2_index]);
1257 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1261 if (!cluster_sector || zeroed) {
1263 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1266 if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1270 cluster_sector = extent->next_cluster_sector;
1271 extent->next_cluster_sector += extent->cluster_sectors;
1273 /* First of all we write grain itself, to avoid race condition
1274 * that may to corrupt the image.
1275 * This problem may occur because of insufficient space on host disk
1276 * or inappropriate VM shutdown.
1278 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1279 offset, skip_start_bytes, skip_end_bytes);
1285 m_data->l1_index = l1_index;
1286 m_data->l2_index = l2_index;
1287 m_data->l2_offset = l2_offset;
1288 m_data->l2_cache_entry = &l2_table[l2_index];
1291 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1295 static VmdkExtent *find_extent(BDRVVmdkState *s,
1296 int64_t sector_num, VmdkExtent *start_hint)
1298 VmdkExtent *extent = start_hint;
1301 extent = &s->extents[0];
1303 while (extent < &s->extents[s->num_extents]) {
1304 if (sector_num < extent->end_sector) {
1312 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1315 uint64_t extent_begin_offset, extent_relative_offset;
1316 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1318 extent_begin_offset =
1319 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1320 extent_relative_offset = offset - extent_begin_offset;
1321 return extent_relative_offset % cluster_size;
1324 static int coroutine_fn vmdk_co_block_status(BlockDriverState *bs,
1326 int64_t offset, int64_t bytes,
1327 int64_t *pnum, int64_t *map,
1328 BlockDriverState **file)
1330 BDRVVmdkState *s = bs->opaque;
1331 int64_t index_in_cluster, n, ret;
1332 uint64_t cluster_offset;
1335 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1339 qemu_co_mutex_lock(&s->lock);
1340 ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1342 qemu_co_mutex_unlock(&s->lock);
1344 index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1353 ret = BDRV_BLOCK_ZERO;
1356 ret = BDRV_BLOCK_DATA;
1357 if (!extent->compressed) {
1358 ret |= BDRV_BLOCK_OFFSET_VALID;
1359 *map = cluster_offset + index_in_cluster;
1361 *file = extent->file->bs;
1365 n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1366 *pnum = MIN(n, bytes);
1370 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1371 int64_t offset_in_cluster, QEMUIOVector *qiov,
1372 uint64_t qiov_offset, uint64_t n_bytes,
1376 VmdkGrainMarker *data = NULL;
1378 QEMUIOVector local_qiov;
1380 int64_t write_offset;
1381 int64_t write_end_sector;
1383 if (extent->compressed) {
1384 void *compressed_data;
1386 if (!extent->has_marker) {
1390 buf_len = (extent->cluster_sectors << 9) * 2;
1391 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1393 compressed_data = g_malloc(n_bytes);
1394 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1395 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1396 g_free(compressed_data);
1398 if (ret != Z_OK || buf_len == 0) {
1403 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1404 data->size = cpu_to_le32(buf_len);
1406 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1407 iov = (struct iovec) {
1411 qemu_iovec_init_external(&local_qiov, &iov, 1);
1413 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1415 qemu_iovec_init(&local_qiov, qiov->niov);
1416 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1418 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1421 write_offset = cluster_offset + offset_in_cluster;
1422 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1425 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1427 if (extent->compressed) {
1428 extent->next_cluster_sector = write_end_sector;
1430 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1440 if (!extent->compressed) {
1441 qemu_iovec_destroy(&local_qiov);
1446 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1447 int64_t offset_in_cluster, QEMUIOVector *qiov,
1451 int cluster_bytes, buf_bytes;
1452 uint8_t *cluster_buf, *compressed_data;
1453 uint8_t *uncomp_buf;
1455 VmdkGrainMarker *marker;
1459 if (!extent->compressed) {
1460 BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1461 ret = bdrv_co_preadv(extent->file,
1462 cluster_offset + offset_in_cluster, bytes,
1469 cluster_bytes = extent->cluster_sectors * 512;
1470 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1471 buf_bytes = cluster_bytes * 2;
1472 cluster_buf = g_malloc(buf_bytes);
1473 uncomp_buf = g_malloc(cluster_bytes);
1474 BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1475 ret = bdrv_pread(extent->file,
1477 cluster_buf, buf_bytes);
1481 compressed_data = cluster_buf;
1482 buf_len = cluster_bytes;
1483 data_len = cluster_bytes;
1484 if (extent->has_marker) {
1485 marker = (VmdkGrainMarker *)cluster_buf;
1486 compressed_data = marker->data;
1487 data_len = le32_to_cpu(marker->size);
1489 if (!data_len || data_len > buf_bytes) {
1493 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1499 if (offset_in_cluster < 0 ||
1500 offset_in_cluster + bytes > buf_len) {
1504 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1509 g_free(cluster_buf);
1513 static int coroutine_fn
1514 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1515 QEMUIOVector *qiov, int flags)
1517 BDRVVmdkState *s = bs->opaque;
1519 uint64_t n_bytes, offset_in_cluster;
1520 VmdkExtent *extent = NULL;
1521 QEMUIOVector local_qiov;
1522 uint64_t cluster_offset;
1523 uint64_t bytes_done = 0;
1525 qemu_iovec_init(&local_qiov, qiov->niov);
1526 qemu_co_mutex_lock(&s->lock);
1529 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1534 ret = get_cluster_offset(bs, extent, NULL,
1535 offset, false, &cluster_offset, 0, 0);
1536 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1538 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1539 - offset_in_cluster);
1541 if (ret != VMDK_OK) {
1542 /* if not allocated, try to read from parent image, if exist */
1543 if (bs->backing && ret != VMDK_ZEROED) {
1544 if (!vmdk_is_cid_valid(bs)) {
1549 qemu_iovec_reset(&local_qiov);
1550 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1552 /* qcow2 emits this on bs->file instead of bs->backing */
1553 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1554 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1560 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1563 qemu_iovec_reset(&local_qiov);
1564 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1566 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1567 &local_qiov, n_bytes);
1574 bytes_done += n_bytes;
1579 qemu_co_mutex_unlock(&s->lock);
1580 qemu_iovec_destroy(&local_qiov);
1587 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1588 * if possible, otherwise return -ENOTSUP.
1589 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1590 * with each cluster. By dry run we can find if the zero write
1591 * is possible without modifying image data.
1593 * Returns: error code with 0 for success.
1595 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1596 uint64_t bytes, QEMUIOVector *qiov,
1597 bool zeroed, bool zero_dry_run)
1599 BDRVVmdkState *s = bs->opaque;
1600 VmdkExtent *extent = NULL;
1602 int64_t offset_in_cluster, n_bytes;
1603 uint64_t cluster_offset;
1604 uint64_t bytes_done = 0;
1605 VmdkMetaData m_data;
1607 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1608 error_report("Wrong offset: offset=0x%" PRIx64
1609 " total_sectors=0x%" PRIx64,
1610 offset, bs->total_sectors);
1615 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1619 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1620 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1621 - offset_in_cluster);
1623 ret = get_cluster_offset(bs, extent, &m_data, offset,
1624 !(extent->compressed || zeroed),
1625 &cluster_offset, offset_in_cluster,
1626 offset_in_cluster + n_bytes);
1627 if (extent->compressed) {
1628 if (ret == VMDK_OK) {
1629 /* Refuse write to allocated cluster for streamOptimized */
1630 error_report("Could not write to allocated cluster"
1631 " for streamOptimized");
1635 ret = get_cluster_offset(bs, extent, &m_data, offset,
1636 true, &cluster_offset, 0, 0);
1639 if (ret == VMDK_ERROR) {
1643 /* Do zeroed write, buf is ignored */
1644 if (extent->has_zero_grain &&
1645 offset_in_cluster == 0 &&
1646 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
1647 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1648 if (!zero_dry_run) {
1649 /* update L2 tables */
1650 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1659 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
1660 qiov, bytes_done, n_bytes, offset);
1665 /* update L2 tables */
1666 if (vmdk_L2update(extent, &m_data,
1667 cluster_offset >> BDRV_SECTOR_BITS)
1675 bytes_done += n_bytes;
1677 /* update CID on the first write every time the virtual disk is
1679 if (!s->cid_updated) {
1680 ret = vmdk_write_cid(bs, g_random_int());
1684 s->cid_updated = true;
1690 static int coroutine_fn
1691 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1692 QEMUIOVector *qiov, int flags)
1695 BDRVVmdkState *s = bs->opaque;
1696 qemu_co_mutex_lock(&s->lock);
1697 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
1698 qemu_co_mutex_unlock(&s->lock);
1702 static int coroutine_fn
1703 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1704 uint64_t bytes, QEMUIOVector *qiov)
1707 /* The caller will write bytes 0 to signal EOF.
1708 * When receive it, we align EOF to a sector boundary. */
1709 BDRVVmdkState *s = bs->opaque;
1713 for (i = 0; i < s->num_extents; i++) {
1714 length = bdrv_getlength(s->extents[i].file->bs);
1718 length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
1719 ret = bdrv_truncate(s->extents[i].file, length,
1720 PREALLOC_MODE_OFF, NULL);
1727 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
1730 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
1733 BdrvRequestFlags flags)
1736 BDRVVmdkState *s = bs->opaque;
1738 qemu_co_mutex_lock(&s->lock);
1739 /* write zeroes could fail if sectors not aligned to cluster, test it with
1740 * dry_run == true before really updating image */
1741 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
1743 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
1745 qemu_co_mutex_unlock(&s->lock);
1749 static int vmdk_init_extent(BlockBackend *blk,
1750 int64_t filesize, bool flat,
1751 bool compress, bool zeroed_grain,
1756 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1757 uint32_t *gd_buf = NULL;
1761 ret = blk_truncate(blk, filesize, PREALLOC_MODE_OFF, errp);
1764 magic = cpu_to_be32(VMDK4_MAGIC);
1765 memset(&header, 0, sizeof(header));
1768 } else if (zeroed_grain) {
1773 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1774 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1775 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1776 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1777 header.capacity = filesize / BDRV_SECTOR_SIZE;
1778 header.granularity = 128;
1779 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1781 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1782 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1784 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1785 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1787 header.desc_offset = 1;
1788 header.desc_size = 20;
1789 header.rgd_offset = header.desc_offset + header.desc_size;
1790 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1791 header.grain_offset =
1792 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1793 header.granularity);
1794 /* swap endianness for all header fields */
1795 header.version = cpu_to_le32(header.version);
1796 header.flags = cpu_to_le32(header.flags);
1797 header.capacity = cpu_to_le64(header.capacity);
1798 header.granularity = cpu_to_le64(header.granularity);
1799 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1800 header.desc_offset = cpu_to_le64(header.desc_offset);
1801 header.desc_size = cpu_to_le64(header.desc_size);
1802 header.rgd_offset = cpu_to_le64(header.rgd_offset);
1803 header.gd_offset = cpu_to_le64(header.gd_offset);
1804 header.grain_offset = cpu_to_le64(header.grain_offset);
1805 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1807 header.check_bytes[0] = 0xa;
1808 header.check_bytes[1] = 0x20;
1809 header.check_bytes[2] = 0xd;
1810 header.check_bytes[3] = 0xa;
1812 /* write all the data */
1813 ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
1815 error_setg(errp, QERR_IO_ERROR);
1818 ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
1820 error_setg(errp, QERR_IO_ERROR);
1824 ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9,
1825 PREALLOC_MODE_OFF, errp);
1830 /* write grain directory */
1831 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1832 gd_buf = g_malloc0(gd_buf_size);
1833 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1834 i < gt_count; i++, tmp += gt_size) {
1835 gd_buf[i] = cpu_to_le32(tmp);
1837 ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1838 gd_buf, gd_buf_size, 0);
1840 error_setg(errp, QERR_IO_ERROR);
1844 /* write backup grain directory */
1845 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1846 i < gt_count; i++, tmp += gt_size) {
1847 gd_buf[i] = cpu_to_le32(tmp);
1849 ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1850 gd_buf, gd_buf_size, 0);
1852 error_setg(errp, QERR_IO_ERROR);
1861 static int vmdk_create_extent(const char *filename, int64_t filesize,
1862 bool flat, bool compress, bool zeroed_grain,
1864 QemuOpts *opts, Error **errp)
1867 BlockBackend *blk = NULL;
1868 Error *local_err = NULL;
1870 ret = bdrv_create_file(filename, opts, &local_err);
1872 error_propagate(errp, local_err);
1876 blk = blk_new_open(filename, NULL, NULL,
1877 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
1880 error_propagate(errp, local_err);
1885 blk_set_allow_write_beyond_eof(blk, true);
1887 ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
1900 static int filename_decompose(const char *filename, char *path, char *prefix,
1901 char *postfix, size_t buf_len, Error **errp)
1905 if (filename == NULL || !strlen(filename)) {
1906 error_setg(errp, "No filename provided");
1909 p = strrchr(filename, '/');
1911 p = strrchr(filename, '\\');
1914 p = strrchr(filename, ':');
1918 if (p - filename >= buf_len) {
1921 pstrcpy(path, p - filename + 1, filename);
1926 q = strrchr(p, '.');
1928 pstrcpy(prefix, buf_len, p);
1931 if (q - p >= buf_len) {
1934 pstrcpy(prefix, q - p + 1, p);
1935 pstrcpy(postfix, buf_len, q);
1941 * idx == 0: get or create the descriptor file (also the image file if in a
1943 * idx >= 1: get the n-th extent if in a split subformat
1945 typedef BlockBackend *(*vmdk_create_extent_fn)(int64_t size,
1954 static void vmdk_desc_add_extent(GString *desc,
1955 const char *extent_line_fmt,
1956 int64_t size, const char *filename)
1958 char *basename = g_path_get_basename(filename);
1960 g_string_append_printf(desc, extent_line_fmt,
1961 DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
1965 static int coroutine_fn vmdk_co_do_create(int64_t size,
1966 BlockdevVmdkSubformat subformat,
1967 BlockdevVmdkAdapterType adapter_type,
1968 const char *backing_file,
1969 const char *hw_version,
1972 vmdk_create_extent_fn extent_fn,
1977 BlockBackend *blk = NULL;
1978 BlockBackend *extent_blk;
1979 Error *local_err = NULL;
1982 bool flat, split, compress;
1983 GString *ext_desc_lines;
1984 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
1985 int64_t extent_size;
1986 int64_t created_size = 0;
1987 const char *extent_line_fmt;
1988 char *parent_desc_line = g_malloc0(BUF_SIZE);
1989 uint32_t parent_cid = 0xffffffff;
1990 uint32_t number_heads = 16;
1991 uint32_t desc_offset = 0, desc_len;
1992 const char desc_template[] =
1993 "# Disk DescriptorFile\n"
1996 "parentCID=%" PRIx32 "\n"
1997 "createType=\"%s\"\n"
2000 "# Extent description\n"
2003 "# The Disk Data Base\n"
2006 "ddb.virtualHWVersion = \"%s\"\n"
2007 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2008 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2009 "ddb.geometry.sectors = \"63\"\n"
2010 "ddb.adapterType = \"%s\"\n";
2012 ext_desc_lines = g_string_new(NULL);
2014 /* Read out options */
2018 "compat6 cannot be enabled with hwversion set");
2028 if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2029 /* that's the number of heads with which vmware operates when
2030 creating, exporting, etc. vmdk files with a non-ide adapter type */
2033 split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2034 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2035 flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2036 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2037 compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2040 extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2042 extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2044 if (flat && backing_file) {
2045 error_setg(errp, "Flat image can't have backing file");
2049 if (flat && zeroed_grain) {
2050 error_setg(errp, "Flat image can't enable zeroed grain");
2055 /* Create extents */
2057 extent_size = split_size;
2061 if (!split && !flat) {
2062 created_size = extent_size;
2066 /* Get the descriptor file BDS */
2067 blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2073 if (!split && !flat) {
2074 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2075 blk_bs(blk)->filename);
2079 BlockBackend *backing;
2080 char *full_backing = g_new0(char, PATH_MAX);
2081 bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename, backing_file,
2082 full_backing, PATH_MAX,
2085 g_free(full_backing);
2086 error_propagate(errp, local_err);
2091 backing = blk_new_open(full_backing, NULL, NULL,
2092 BDRV_O_NO_BACKING, errp);
2093 g_free(full_backing);
2094 if (backing == NULL) {
2098 if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2099 error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2100 blk_bs(backing)->drv->format_name);
2105 ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2108 error_setg(errp, "Failed to read parent CID");
2111 snprintf(parent_desc_line, BUF_SIZE,
2112 "parentFileNameHint=\"%s\"", backing_file);
2115 while (created_size < size) {
2116 int64_t cur_size = MIN(size - created_size, extent_size);
2117 extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2118 zeroed_grain, opaque, errp);
2123 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2124 blk_bs(extent_blk)->filename);
2125 created_size += cur_size;
2127 blk_unref(extent_blk);
2130 /* Check whether we got excess extents */
2131 extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2134 blk_unref(extent_blk);
2135 error_setg(errp, "List of extents contains unused extents");
2140 /* generate descriptor file */
2141 desc = g_strdup_printf(desc_template,
2144 BlockdevVmdkSubformat_str(subformat),
2146 ext_desc_lines->str,
2149 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2151 BlockdevVmdkAdapterType_str(adapter_type));
2152 desc_len = strlen(desc);
2153 /* the descriptor offset = 0x200 */
2154 if (!split && !flat) {
2155 desc_offset = 0x200;
2158 ret = blk_pwrite(blk, desc_offset, desc, desc_len, 0);
2160 error_setg_errno(errp, -ret, "Could not write description");
2163 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2164 * for description file */
2165 if (desc_offset == 0) {
2166 ret = blk_truncate(blk, desc_len, PREALLOC_MODE_OFF, errp);
2177 g_free(parent_desc_line);
2178 g_string_free(ext_desc_lines, true);
2187 } VMDKCreateOptsData;
2189 static BlockBackend *vmdk_co_create_opts_cb(int64_t size, int idx,
2190 bool flat, bool split, bool compress,
2191 bool zeroed_grain, void *opaque,
2194 BlockBackend *blk = NULL;
2195 BlockDriverState *bs = NULL;
2196 VMDKCreateOptsData *data = opaque;
2197 char *ext_filename = NULL;
2198 char *rel_filename = NULL;
2200 /* We're done, don't create excess extents. */
2202 assert(errp == NULL);
2207 rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2209 rel_filename = g_strdup_printf("%s-%c%03d%s",
2211 flat ? 'f' : 's', idx, data->postfix);
2214 rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2217 ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2218 g_free(rel_filename);
2220 if (vmdk_create_extent(ext_filename, size,
2221 flat, compress, zeroed_grain, &blk, data->opts,
2227 g_free(ext_filename);
2231 static int coroutine_fn vmdk_co_create_opts(const char *filename, QemuOpts *opts,
2234 Error *local_err = NULL;
2236 int64_t total_size = 0;
2237 char *adapter_type = NULL;
2238 BlockdevVmdkAdapterType adapter_type_enum;
2239 char *backing_file = NULL;
2240 char *hw_version = NULL;
2242 BlockdevVmdkSubformat subformat;
2244 char *path = g_malloc0(PATH_MAX);
2245 char *prefix = g_malloc0(PATH_MAX);
2246 char *postfix = g_malloc0(PATH_MAX);
2247 char *desc_line = g_malloc0(BUF_SIZE);
2248 char *ext_filename = g_malloc0(PATH_MAX);
2249 char *desc_filename = g_malloc0(PATH_MAX);
2250 char *parent_desc_line = g_malloc0(BUF_SIZE);
2253 VMDKCreateOptsData data;
2255 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
2259 /* Read out options */
2260 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2262 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
2263 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2264 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
2265 compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false);
2266 if (strcmp(hw_version, "undefined") == 0) {
2268 hw_version = g_strdup("4");
2270 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
2271 zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false);
2274 adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup,
2276 BLOCKDEV_VMDK_ADAPTER_TYPE_IDE,
2279 error_propagate(errp, local_err);
2284 adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE;
2288 /* Default format to monolithicSparse */
2289 subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE;
2291 subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup,
2293 BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE,
2296 error_propagate(errp, local_err);
2301 data = (VMDKCreateOptsData){
2307 ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum,
2308 backing_file, hw_version, compat6, zeroed_grain,
2309 vmdk_co_create_opts_cb, &data, errp);
2312 g_free(adapter_type);
2313 g_free(backing_file);
2321 g_free(ext_filename);
2322 g_free(desc_filename);
2323 g_free(parent_desc_line);
2327 static BlockBackend *vmdk_co_create_cb(int64_t size, int idx,
2328 bool flat, bool split, bool compress,
2329 bool zeroed_grain, void *opaque,
2333 BlockDriverState *bs;
2335 BlockdevCreateOptionsVmdk *opts = opaque;
2338 bs = bdrv_open_blockdev_ref(opts->file, errp);
2341 BlockdevRefList *list = opts->extents;
2342 for (i = 1; i < idx; i++) {
2343 if (!list || !list->next) {
2344 error_setg(errp, "Extent [%d] not specified", i);
2350 error_setg(errp, "Extent [%d] not specified", idx - 1);
2353 bs = bdrv_open_blockdev_ref(list->value, errp);
2358 blk = blk_new(BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | BLK_PERM_RESIZE,
2360 if (blk_insert_bs(blk, bs, errp)) {
2364 blk_set_allow_write_beyond_eof(blk, true);
2368 ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2377 static int coroutine_fn vmdk_co_create(BlockdevCreateOptions *create_options,
2381 BlockdevCreateOptionsVmdk *opts;
2383 opts = &create_options->u.vmdk;
2385 /* Validate options */
2386 if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2387 error_setg(errp, "Image size must be a multiple of 512 bytes");
2392 ret = vmdk_co_do_create(opts->size,
2407 static void vmdk_close(BlockDriverState *bs)
2409 BDRVVmdkState *s = bs->opaque;
2411 vmdk_free_extents(bs);
2412 g_free(s->create_type);
2414 migrate_del_blocker(s->migration_blocker);
2415 error_free(s->migration_blocker);
2418 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2420 BDRVVmdkState *s = bs->opaque;
2424 for (i = 0; i < s->num_extents; i++) {
2425 err = bdrv_co_flush(s->extents[i].file->bs);
2433 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2438 BDRVVmdkState *s = bs->opaque;
2440 ret = bdrv_get_allocated_file_size(bs->file->bs);
2444 for (i = 0; i < s->num_extents; i++) {
2445 if (s->extents[i].file == bs->file) {
2448 r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2457 static int vmdk_has_zero_init(BlockDriverState *bs)
2460 BDRVVmdkState *s = bs->opaque;
2462 /* If has a flat extent and its underlying storage doesn't have zero init,
2464 for (i = 0; i < s->num_extents; i++) {
2465 if (s->extents[i].flat) {
2466 if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2474 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2476 ImageInfo *info = g_new0(ImageInfo, 1);
2478 bdrv_refresh_filename(extent->file->bs);
2479 *info = (ImageInfo){
2480 .filename = g_strdup(extent->file->bs->filename),
2481 .format = g_strdup(extent->type),
2482 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2483 .compressed = extent->compressed,
2484 .has_compressed = extent->compressed,
2485 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2486 .has_cluster_size = !extent->flat,
2492 static int coroutine_fn vmdk_co_check(BlockDriverState *bs,
2493 BdrvCheckResult *result,
2496 BDRVVmdkState *s = bs->opaque;
2497 VmdkExtent *extent = NULL;
2498 int64_t sector_num = 0;
2499 int64_t total_sectors = bdrv_nb_sectors(bs);
2501 uint64_t cluster_offset;
2508 if (sector_num >= total_sectors) {
2511 extent = find_extent(s, sector_num, extent);
2514 "ERROR: could not find extent for sector %" PRId64 "\n",
2519 ret = get_cluster_offset(bs, extent, NULL,
2520 sector_num << BDRV_SECTOR_BITS,
2521 false, &cluster_offset, 0, 0);
2522 if (ret == VMDK_ERROR) {
2524 "ERROR: could not get cluster_offset for sector %"
2525 PRId64 "\n", sector_num);
2528 if (ret == VMDK_OK) {
2529 int64_t extent_len = bdrv_getlength(extent->file->bs);
2530 if (extent_len < 0) {
2532 "ERROR: could not get extent file length for sector %"
2533 PRId64 "\n", sector_num);
2537 if (cluster_offset >= extent_len) {
2539 "ERROR: cluster offset for sector %"
2540 PRId64 " points after EOF\n", sector_num);
2545 sector_num += extent->cluster_sectors;
2548 result->corruptions++;
2552 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs,
2556 BDRVVmdkState *s = bs->opaque;
2557 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2558 ImageInfoList **next;
2560 *spec_info = (ImageInfoSpecific){
2561 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2563 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2567 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2568 .create_type = g_strdup(s->create_type),
2570 .parent_cid = s->parent_cid,
2573 next = &spec_info->u.vmdk.data->extents;
2574 for (i = 0; i < s->num_extents; i++) {
2575 *next = g_new0(ImageInfoList, 1);
2576 (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2577 (*next)->next = NULL;
2578 next = &(*next)->next;
2584 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2586 return a->flat == b->flat &&
2587 a->compressed == b->compressed &&
2588 (a->flat || a->cluster_sectors == b->cluster_sectors);
2591 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2594 BDRVVmdkState *s = bs->opaque;
2595 assert(s->num_extents);
2597 /* See if we have multiple extents but they have different cases */
2598 for (i = 1; i < s->num_extents; i++) {
2599 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2603 bdi->needs_compressed_writes = s->extents[0].compressed;
2604 if (!s->extents[0].flat) {
2605 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2610 static QemuOptsList vmdk_create_opts = {
2611 .name = "vmdk-create-opts",
2612 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2615 .name = BLOCK_OPT_SIZE,
2616 .type = QEMU_OPT_SIZE,
2617 .help = "Virtual disk size"
2620 .name = BLOCK_OPT_ADAPTER_TYPE,
2621 .type = QEMU_OPT_STRING,
2622 .help = "Virtual adapter type, can be one of "
2623 "ide (default), lsilogic, buslogic or legacyESX"
2626 .name = BLOCK_OPT_BACKING_FILE,
2627 .type = QEMU_OPT_STRING,
2628 .help = "File name of a base image"
2631 .name = BLOCK_OPT_COMPAT6,
2632 .type = QEMU_OPT_BOOL,
2633 .help = "VMDK version 6 image",
2634 .def_value_str = "off"
2637 .name = BLOCK_OPT_HWVERSION,
2638 .type = QEMU_OPT_STRING,
2639 .help = "VMDK hardware version",
2640 .def_value_str = "undefined"
2643 .name = BLOCK_OPT_SUBFMT,
2644 .type = QEMU_OPT_STRING,
2646 "VMDK flat extent format, can be one of "
2647 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2650 .name = BLOCK_OPT_ZEROED_GRAIN,
2651 .type = QEMU_OPT_BOOL,
2652 .help = "Enable efficient zero writes "
2653 "using the zeroed-grain GTE feature"
2655 { /* end of list */ }
2659 static BlockDriver bdrv_vmdk = {
2660 .format_name = "vmdk",
2661 .instance_size = sizeof(BDRVVmdkState),
2662 .bdrv_probe = vmdk_probe,
2663 .bdrv_open = vmdk_open,
2664 .bdrv_co_check = vmdk_co_check,
2665 .bdrv_reopen_prepare = vmdk_reopen_prepare,
2666 .bdrv_child_perm = bdrv_format_default_perms,
2667 .bdrv_co_preadv = vmdk_co_preadv,
2668 .bdrv_co_pwritev = vmdk_co_pwritev,
2669 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
2670 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
2671 .bdrv_close = vmdk_close,
2672 .bdrv_co_create_opts = vmdk_co_create_opts,
2673 .bdrv_co_create = vmdk_co_create,
2674 .bdrv_co_flush_to_disk = vmdk_co_flush,
2675 .bdrv_co_block_status = vmdk_co_block_status,
2676 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2677 .bdrv_has_zero_init = vmdk_has_zero_init,
2678 .bdrv_get_specific_info = vmdk_get_specific_info,
2679 .bdrv_refresh_limits = vmdk_refresh_limits,
2680 .bdrv_get_info = vmdk_get_info,
2682 .supports_backing = true,
2683 .create_opts = &vmdk_create_opts,
2686 static void bdrv_vmdk_init(void)
2688 bdrv_register(&bdrv_vmdk);
2691 block_init(bdrv_vmdk_init);