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 error_setg_errno(errp, -ret,
483 "Could not read l1 table from extent '%s'",
484 extent->file->bs->filename);
487 for (i = 0; i < extent->l1_size; i++) {
488 le32_to_cpus(&extent->l1_table[i]);
491 if (extent->l1_backup_table_offset) {
492 extent->l1_backup_table = g_try_malloc(l1_size);
493 if (l1_size && extent->l1_backup_table == NULL) {
497 ret = bdrv_pread(extent->file,
498 extent->l1_backup_table_offset,
499 extent->l1_backup_table,
502 error_setg_errno(errp, -ret,
503 "Could not read l1 backup table from extent '%s'",
504 extent->file->bs->filename);
507 for (i = 0; i < extent->l1_size; i++) {
508 le32_to_cpus(&extent->l1_backup_table[i]);
513 g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE);
516 g_free(extent->l1_backup_table);
518 g_free(extent->l1_table);
522 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
524 int flags, Error **errp)
531 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
533 error_setg_errno(errp, -ret,
534 "Could not read header from file '%s'",
538 ret = vmdk_add_extent(bs, file, false,
539 le32_to_cpu(header.disk_sectors),
540 (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
542 le32_to_cpu(header.l1dir_size),
544 le32_to_cpu(header.granularity),
550 ret = vmdk_init_tables(bs, extent, errp);
552 /* free extent allocated by vmdk_add_extent */
553 vmdk_free_last_extent(bs);
558 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
559 QDict *options, Error **errp);
561 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
567 size = bdrv_getlength(file->bs);
569 error_setg_errno(errp, -size, "Could not access file");
574 /* Both descriptor file and sparse image must be much larger than 4
575 * bytes, also callers of vmdk_read_desc want to compare the first 4
576 * bytes with VMDK4_MAGIC, let's error out if less is read. */
577 error_setg(errp, "File is too small, not a valid image");
581 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */
582 buf = g_malloc(size + 1);
584 ret = bdrv_pread(file, desc_offset, buf, size);
586 error_setg_errno(errp, -ret, "Could not read from file");
595 static int vmdk_open_vmdk4(BlockDriverState *bs,
597 int flags, QDict *options, Error **errp)
601 uint32_t l1_size, l1_entry_sectors;
604 BDRVVmdkState *s = bs->opaque;
605 int64_t l1_backup_offset = 0;
608 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
610 error_setg_errno(errp, -ret,
611 "Could not read header from file '%s'",
615 if (header.capacity == 0) {
616 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
618 char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
622 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
628 if (!s->create_type) {
629 s->create_type = g_strdup("monolithicSparse");
632 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
634 * The footer takes precedence over the header, so read it in. The
635 * footer starts at offset -1024 from the end: One sector for the
636 * footer, and another one for the end-of-stream marker.
643 uint8_t pad[512 - 16];
644 } QEMU_PACKED footer_marker;
648 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
654 uint8_t pad[512 - 16];
655 } QEMU_PACKED eos_marker;
656 } QEMU_PACKED footer;
658 ret = bdrv_pread(file,
659 bs->file->bs->total_sectors * 512 - 1536,
660 &footer, sizeof(footer));
662 error_setg_errno(errp, -ret, "Failed to read footer");
666 /* Some sanity checks for the footer */
667 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
668 le32_to_cpu(footer.footer_marker.size) != 0 ||
669 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
670 le64_to_cpu(footer.eos_marker.val) != 0 ||
671 le32_to_cpu(footer.eos_marker.size) != 0 ||
672 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
674 error_setg(errp, "Invalid footer");
678 header = footer.header;
682 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
683 if (le32_to_cpu(header.version) > 3) {
684 error_setg(errp, "Unsupported VMDK version %" PRIu32,
685 le32_to_cpu(header.version));
687 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
689 /* VMware KB 2064959 explains that version 3 added support for
690 * persistent changed block tracking (CBT), and backup software can
691 * read it as version=1 if it doesn't care about the changed area
692 * information. So we are safe to enable read only. */
693 error_setg(errp, "VMDK version 3 must be read only");
697 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
698 error_setg(errp, "L2 table size too big");
702 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
703 * le64_to_cpu(header.granularity);
704 if (l1_entry_sectors == 0) {
705 error_setg(errp, "L1 entry size is invalid");
708 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
710 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
711 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
713 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
714 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
715 (int64_t)(le64_to_cpu(header.grain_offset)
716 * BDRV_SECTOR_SIZE));
720 ret = vmdk_add_extent(bs, file, false,
721 le64_to_cpu(header.capacity),
722 le64_to_cpu(header.gd_offset) << 9,
725 le32_to_cpu(header.num_gtes_per_gt),
726 le64_to_cpu(header.granularity),
733 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
734 if (extent->compressed) {
735 g_free(s->create_type);
736 s->create_type = g_strdup("streamOptimized");
738 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
739 extent->version = le32_to_cpu(header.version);
740 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
741 ret = vmdk_init_tables(bs, extent, errp);
743 /* free extent allocated by vmdk_add_extent */
744 vmdk_free_last_extent(bs);
749 /* find an option value out of descriptor file */
750 static int vmdk_parse_description(const char *desc, const char *opt_name,
751 char *buf, int buf_size)
753 char *opt_pos, *opt_end;
754 const char *end = desc + strlen(desc);
756 opt_pos = strstr(desc, opt_name);
760 /* Skip "=\"" following opt_name */
761 opt_pos += strlen(opt_name) + 2;
762 if (opt_pos >= end) {
766 while (opt_end < end && *opt_end != '"') {
769 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
772 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
776 /* Open an extent file and append to bs array */
777 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
778 char *buf, QDict *options, Error **errp)
782 magic = ldl_be_p(buf);
785 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
788 return vmdk_open_vmdk4(bs, file, flags, options, errp);
791 error_setg(errp, "Image not in VMDK format");
797 static const char *next_line(const char *s)
808 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
809 const char *desc_file_path, QDict *options,
821 BdrvChild *extent_file;
822 BDRVVmdkState *s = bs->opaque;
824 char extent_opt_prefix[32];
825 Error *local_err = NULL;
827 for (p = desc; *p; p = next_line(p)) {
828 /* parse extent line in one of below formats:
830 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
831 * RW [size in sectors] SPARSE "file-name.vmdk"
832 * RW [size in sectors] VMFS "file-name.vmdk"
833 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
836 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
837 access, §ors, type, fname, &flat_offset);
838 if (matches < 4 || strcmp(access, "RW")) {
840 } else if (!strcmp(type, "FLAT")) {
841 if (matches != 5 || flat_offset < 0) {
844 } else if (!strcmp(type, "VMFS")) {
850 } else if (matches != 4) {
855 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
856 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
857 (strcmp(access, "RW"))) {
861 if (!path_is_absolute(fname) && !path_has_protocol(fname) &&
864 error_setg(errp, "Cannot use relative extent paths with VMDK "
865 "descriptor file '%s'", bs->file->bs->filename);
869 extent_path = g_malloc0(PATH_MAX);
870 path_combine(extent_path, PATH_MAX, desc_file_path, fname);
872 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
875 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
876 bs, &child_file, false, &local_err);
879 error_propagate(errp, local_err);
883 /* save to extents array */
884 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
887 ret = vmdk_add_extent(bs, extent_file, true, sectors,
888 0, 0, 0, 0, 0, &extent, errp);
890 bdrv_unref_child(bs, extent_file);
893 extent->flat_start_offset = flat_offset << 9;
894 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
895 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
896 char *buf = vmdk_read_desc(extent_file, 0, errp);
900 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
905 bdrv_unref_child(bs, extent_file);
908 extent = &s->extents[s->num_extents - 1];
910 error_setg(errp, "Unsupported extent type '%s'", type);
911 bdrv_unref_child(bs, extent_file);
914 extent->type = g_strdup(type);
921 if (np[-1] == '\n') {
924 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
928 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
929 QDict *options, Error **errp)
933 BDRVVmdkState *s = bs->opaque;
935 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
936 error_setg(errp, "invalid VMDK image descriptor");
940 if (strcmp(ct, "monolithicFlat") &&
941 strcmp(ct, "vmfs") &&
942 strcmp(ct, "vmfsSparse") &&
943 strcmp(ct, "twoGbMaxExtentSparse") &&
944 strcmp(ct, "twoGbMaxExtentFlat")) {
945 error_setg(errp, "Unsupported image type '%s'", ct);
949 s->create_type = g_strdup(ct);
951 ret = vmdk_parse_extents(buf, bs, bs->file->bs->exact_filename, options,
957 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
962 BDRVVmdkState *s = bs->opaque;
964 Error *local_err = NULL;
966 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
972 buf = vmdk_read_desc(bs->file, 0, errp);
977 magic = ldl_be_p(buf);
981 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
983 s->desc_offset = 0x200;
986 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
993 /* try to open parent images, if exist */
994 ret = vmdk_parent_open(bs);
998 ret = vmdk_read_cid(bs, 0, &s->cid);
1002 ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1006 qemu_co_mutex_init(&s->lock);
1008 /* Disable migration when VMDK images are used */
1009 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1010 "does not support live migration",
1011 bdrv_get_device_or_node_name(bs));
1012 ret = migrate_add_blocker(s->migration_blocker, &local_err);
1014 error_propagate(errp, local_err);
1015 error_free(s->migration_blocker);
1024 g_free(s->create_type);
1025 s->create_type = NULL;
1026 vmdk_free_extents(bs);
1031 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1033 BDRVVmdkState *s = bs->opaque;
1036 for (i = 0; i < s->num_extents; i++) {
1037 if (!s->extents[i].flat) {
1038 bs->bl.pwrite_zeroes_alignment =
1039 MAX(bs->bl.pwrite_zeroes_alignment,
1040 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1048 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1049 * to the cluster at @cluster_sector_num.
1051 * If @skip_start_sector < @skip_end_sector, the relative range
1052 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1053 * it for call to write user data in the request.
1055 static int get_whole_cluster(BlockDriverState *bs,
1057 uint64_t cluster_offset,
1059 uint64_t skip_start_bytes,
1060 uint64_t skip_end_bytes)
1063 int64_t cluster_bytes;
1064 uint8_t *whole_grain;
1066 /* For COW, align request sector_num to cluster start */
1067 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1068 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1069 whole_grain = qemu_blockalign(bs, cluster_bytes);
1072 memset(whole_grain, 0, skip_start_bytes);
1073 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1076 assert(skip_end_bytes <= cluster_bytes);
1077 /* we will be here if it's first write on non-exist grain(cluster).
1078 * try to read from parent image, if exist */
1079 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1084 /* Read backing data before skip range */
1085 if (skip_start_bytes > 0) {
1087 /* qcow2 emits this on bs->file instead of bs->backing */
1088 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1089 ret = bdrv_pread(bs->backing, offset, whole_grain,
1096 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1097 ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1104 /* Read backing data after skip range */
1105 if (skip_end_bytes < cluster_bytes) {
1107 /* qcow2 emits this on bs->file instead of bs->backing */
1108 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1109 ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1110 whole_grain + skip_end_bytes,
1111 cluster_bytes - skip_end_bytes);
1117 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1118 ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1119 whole_grain + skip_end_bytes,
1120 cluster_bytes - skip_end_bytes);
1129 qemu_vfree(whole_grain);
1133 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1136 offset = cpu_to_le32(offset);
1137 /* update L2 table */
1138 BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1139 if (bdrv_pwrite_sync(extent->file,
1140 ((int64_t)m_data->l2_offset * 512)
1141 + (m_data->l2_index * sizeof(offset)),
1142 &offset, sizeof(offset)) < 0) {
1145 /* update backup L2 table */
1146 if (extent->l1_backup_table_offset != 0) {
1147 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
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) {
1155 if (m_data->l2_cache_entry) {
1156 *m_data->l2_cache_entry = offset;
1163 * get_cluster_offset
1165 * Look up cluster offset in extent file by sector number, and store in
1168 * For flat extents, the start offset as parsed from the description file is
1171 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1172 * offset for a new cluster and update L2 cache. If there is a backing file,
1173 * COW is done before returning; otherwise, zeroes are written to the allocated
1174 * cluster. Both COW and zero writing skips the sector range
1175 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1176 * has new data to write there.
1178 * Returns: VMDK_OK if cluster exists and mapped in the image.
1179 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1180 * VMDK_ERROR if failed.
1182 static int get_cluster_offset(BlockDriverState *bs,
1184 VmdkMetaData *m_data,
1187 uint64_t *cluster_offset,
1188 uint64_t skip_start_bytes,
1189 uint64_t skip_end_bytes)
1191 unsigned int l1_index, l2_offset, l2_index;
1192 int min_index, i, j;
1193 uint32_t min_count, *l2_table;
1194 bool zeroed = false;
1196 int64_t cluster_sector;
1202 *cluster_offset = extent->flat_start_offset;
1206 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1207 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1208 if (l1_index >= extent->l1_size) {
1211 l2_offset = extent->l1_table[l1_index];
1213 return VMDK_UNALLOC;
1215 for (i = 0; i < L2_CACHE_SIZE; i++) {
1216 if (l2_offset == extent->l2_cache_offsets[i]) {
1217 /* increment the hit count */
1218 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1219 for (j = 0; j < L2_CACHE_SIZE; j++) {
1220 extent->l2_cache_counts[j] >>= 1;
1223 l2_table = extent->l2_cache + (i * extent->l2_size);
1227 /* not found: load a new entry in the least used one */
1229 min_count = 0xffffffff;
1230 for (i = 0; i < L2_CACHE_SIZE; i++) {
1231 if (extent->l2_cache_counts[i] < min_count) {
1232 min_count = extent->l2_cache_counts[i];
1236 l2_table = extent->l2_cache + (min_index * extent->l2_size);
1237 BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1238 if (bdrv_pread(extent->file,
1239 (int64_t)l2_offset * 512,
1241 extent->l2_size * sizeof(uint32_t)
1242 ) != extent->l2_size * sizeof(uint32_t)) {
1246 extent->l2_cache_offsets[min_index] = l2_offset;
1247 extent->l2_cache_counts[min_index] = 1;
1249 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1250 cluster_sector = le32_to_cpu(l2_table[l2_index]);
1252 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1256 if (!cluster_sector || zeroed) {
1258 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1261 if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1265 cluster_sector = extent->next_cluster_sector;
1266 extent->next_cluster_sector += extent->cluster_sectors;
1268 /* First of all we write grain itself, to avoid race condition
1269 * that may to corrupt the image.
1270 * This problem may occur because of insufficient space on host disk
1271 * or inappropriate VM shutdown.
1273 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1274 offset, skip_start_bytes, skip_end_bytes);
1280 m_data->l1_index = l1_index;
1281 m_data->l2_index = l2_index;
1282 m_data->l2_offset = l2_offset;
1283 m_data->l2_cache_entry = &l2_table[l2_index];
1286 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1290 static VmdkExtent *find_extent(BDRVVmdkState *s,
1291 int64_t sector_num, VmdkExtent *start_hint)
1293 VmdkExtent *extent = start_hint;
1296 extent = &s->extents[0];
1298 while (extent < &s->extents[s->num_extents]) {
1299 if (sector_num < extent->end_sector) {
1307 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1310 uint64_t extent_begin_offset, extent_relative_offset;
1311 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1313 extent_begin_offset =
1314 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1315 extent_relative_offset = offset - extent_begin_offset;
1316 return extent_relative_offset % cluster_size;
1319 static int coroutine_fn vmdk_co_block_status(BlockDriverState *bs,
1321 int64_t offset, int64_t bytes,
1322 int64_t *pnum, int64_t *map,
1323 BlockDriverState **file)
1325 BDRVVmdkState *s = bs->opaque;
1326 int64_t index_in_cluster, n, ret;
1327 uint64_t cluster_offset;
1330 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1334 qemu_co_mutex_lock(&s->lock);
1335 ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1337 qemu_co_mutex_unlock(&s->lock);
1339 index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1348 ret = BDRV_BLOCK_ZERO;
1351 ret = BDRV_BLOCK_DATA;
1352 if (!extent->compressed) {
1353 ret |= BDRV_BLOCK_OFFSET_VALID;
1354 *map = cluster_offset + index_in_cluster;
1356 *file = extent->file->bs;
1360 n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1361 *pnum = MIN(n, bytes);
1365 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1366 int64_t offset_in_cluster, QEMUIOVector *qiov,
1367 uint64_t qiov_offset, uint64_t n_bytes,
1371 VmdkGrainMarker *data = NULL;
1373 QEMUIOVector local_qiov;
1375 int64_t write_offset;
1376 int64_t write_end_sector;
1378 if (extent->compressed) {
1379 void *compressed_data;
1381 if (!extent->has_marker) {
1385 buf_len = (extent->cluster_sectors << 9) * 2;
1386 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1388 compressed_data = g_malloc(n_bytes);
1389 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1390 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1391 g_free(compressed_data);
1393 if (ret != Z_OK || buf_len == 0) {
1398 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1399 data->size = cpu_to_le32(buf_len);
1401 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1402 iov = (struct iovec) {
1406 qemu_iovec_init_external(&local_qiov, &iov, 1);
1408 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1410 qemu_iovec_init(&local_qiov, qiov->niov);
1411 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1413 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1416 write_offset = cluster_offset + offset_in_cluster;
1417 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1420 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1422 if (extent->compressed) {
1423 extent->next_cluster_sector = write_end_sector;
1425 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1435 if (!extent->compressed) {
1436 qemu_iovec_destroy(&local_qiov);
1441 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1442 int64_t offset_in_cluster, QEMUIOVector *qiov,
1446 int cluster_bytes, buf_bytes;
1447 uint8_t *cluster_buf, *compressed_data;
1448 uint8_t *uncomp_buf;
1450 VmdkGrainMarker *marker;
1454 if (!extent->compressed) {
1455 BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1456 ret = bdrv_co_preadv(extent->file,
1457 cluster_offset + offset_in_cluster, bytes,
1464 cluster_bytes = extent->cluster_sectors * 512;
1465 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1466 buf_bytes = cluster_bytes * 2;
1467 cluster_buf = g_malloc(buf_bytes);
1468 uncomp_buf = g_malloc(cluster_bytes);
1469 BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1470 ret = bdrv_pread(extent->file,
1472 cluster_buf, buf_bytes);
1476 compressed_data = cluster_buf;
1477 buf_len = cluster_bytes;
1478 data_len = cluster_bytes;
1479 if (extent->has_marker) {
1480 marker = (VmdkGrainMarker *)cluster_buf;
1481 compressed_data = marker->data;
1482 data_len = le32_to_cpu(marker->size);
1484 if (!data_len || data_len > buf_bytes) {
1488 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1494 if (offset_in_cluster < 0 ||
1495 offset_in_cluster + bytes > buf_len) {
1499 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1504 g_free(cluster_buf);
1508 static int coroutine_fn
1509 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1510 QEMUIOVector *qiov, int flags)
1512 BDRVVmdkState *s = bs->opaque;
1514 uint64_t n_bytes, offset_in_cluster;
1515 VmdkExtent *extent = NULL;
1516 QEMUIOVector local_qiov;
1517 uint64_t cluster_offset;
1518 uint64_t bytes_done = 0;
1520 qemu_iovec_init(&local_qiov, qiov->niov);
1521 qemu_co_mutex_lock(&s->lock);
1524 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1529 ret = get_cluster_offset(bs, extent, NULL,
1530 offset, false, &cluster_offset, 0, 0);
1531 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1533 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1534 - offset_in_cluster);
1536 if (ret != VMDK_OK) {
1537 /* if not allocated, try to read from parent image, if exist */
1538 if (bs->backing && ret != VMDK_ZEROED) {
1539 if (!vmdk_is_cid_valid(bs)) {
1544 qemu_iovec_reset(&local_qiov);
1545 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1547 /* qcow2 emits this on bs->file instead of bs->backing */
1548 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1549 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1555 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1558 qemu_iovec_reset(&local_qiov);
1559 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1561 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1562 &local_qiov, n_bytes);
1569 bytes_done += n_bytes;
1574 qemu_co_mutex_unlock(&s->lock);
1575 qemu_iovec_destroy(&local_qiov);
1582 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1583 * if possible, otherwise return -ENOTSUP.
1584 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1585 * with each cluster. By dry run we can find if the zero write
1586 * is possible without modifying image data.
1588 * Returns: error code with 0 for success.
1590 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1591 uint64_t bytes, QEMUIOVector *qiov,
1592 bool zeroed, bool zero_dry_run)
1594 BDRVVmdkState *s = bs->opaque;
1595 VmdkExtent *extent = NULL;
1597 int64_t offset_in_cluster, n_bytes;
1598 uint64_t cluster_offset;
1599 uint64_t bytes_done = 0;
1600 VmdkMetaData m_data;
1602 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1603 error_report("Wrong offset: offset=0x%" PRIx64
1604 " total_sectors=0x%" PRIx64,
1605 offset, bs->total_sectors);
1610 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1614 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1615 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1616 - offset_in_cluster);
1618 ret = get_cluster_offset(bs, extent, &m_data, offset,
1619 !(extent->compressed || zeroed),
1620 &cluster_offset, offset_in_cluster,
1621 offset_in_cluster + n_bytes);
1622 if (extent->compressed) {
1623 if (ret == VMDK_OK) {
1624 /* Refuse write to allocated cluster for streamOptimized */
1625 error_report("Could not write to allocated cluster"
1626 " for streamOptimized");
1630 ret = get_cluster_offset(bs, extent, &m_data, offset,
1631 true, &cluster_offset, 0, 0);
1634 if (ret == VMDK_ERROR) {
1638 /* Do zeroed write, buf is ignored */
1639 if (extent->has_zero_grain &&
1640 offset_in_cluster == 0 &&
1641 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
1642 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1643 if (!zero_dry_run) {
1644 /* update L2 tables */
1645 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1654 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
1655 qiov, bytes_done, n_bytes, offset);
1660 /* update L2 tables */
1661 if (vmdk_L2update(extent, &m_data,
1662 cluster_offset >> BDRV_SECTOR_BITS)
1670 bytes_done += n_bytes;
1672 /* update CID on the first write every time the virtual disk is
1674 if (!s->cid_updated) {
1675 ret = vmdk_write_cid(bs, g_random_int());
1679 s->cid_updated = true;
1685 static int coroutine_fn
1686 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1687 QEMUIOVector *qiov, int flags)
1690 BDRVVmdkState *s = bs->opaque;
1691 qemu_co_mutex_lock(&s->lock);
1692 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
1693 qemu_co_mutex_unlock(&s->lock);
1697 static int coroutine_fn
1698 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1699 uint64_t bytes, QEMUIOVector *qiov)
1702 /* The caller will write bytes 0 to signal EOF.
1703 * When receive it, we align EOF to a sector boundary. */
1704 BDRVVmdkState *s = bs->opaque;
1708 for (i = 0; i < s->num_extents; i++) {
1709 length = bdrv_getlength(s->extents[i].file->bs);
1713 length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
1714 ret = bdrv_truncate(s->extents[i].file, length,
1715 PREALLOC_MODE_OFF, NULL);
1722 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
1725 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
1728 BdrvRequestFlags flags)
1731 BDRVVmdkState *s = bs->opaque;
1733 qemu_co_mutex_lock(&s->lock);
1734 /* write zeroes could fail if sectors not aligned to cluster, test it with
1735 * dry_run == true before really updating image */
1736 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
1738 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
1740 qemu_co_mutex_unlock(&s->lock);
1744 static int vmdk_create_extent(const char *filename, int64_t filesize,
1745 bool flat, bool compress, bool zeroed_grain,
1746 QemuOpts *opts, Error **errp)
1749 BlockBackend *blk = NULL;
1751 Error *local_err = NULL;
1752 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1753 uint32_t *gd_buf = NULL;
1756 ret = bdrv_create_file(filename, opts, &local_err);
1758 error_propagate(errp, local_err);
1762 blk = blk_new_open(filename, NULL, NULL,
1763 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
1766 error_propagate(errp, local_err);
1771 blk_set_allow_write_beyond_eof(blk, true);
1774 ret = blk_truncate(blk, filesize, PREALLOC_MODE_OFF, errp);
1777 magic = cpu_to_be32(VMDK4_MAGIC);
1778 memset(&header, 0, sizeof(header));
1781 } else if (zeroed_grain) {
1786 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1787 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1788 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1789 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1790 header.capacity = filesize / BDRV_SECTOR_SIZE;
1791 header.granularity = 128;
1792 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1794 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1795 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1797 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1798 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1800 header.desc_offset = 1;
1801 header.desc_size = 20;
1802 header.rgd_offset = header.desc_offset + header.desc_size;
1803 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1804 header.grain_offset =
1805 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1806 header.granularity);
1807 /* swap endianness for all header fields */
1808 header.version = cpu_to_le32(header.version);
1809 header.flags = cpu_to_le32(header.flags);
1810 header.capacity = cpu_to_le64(header.capacity);
1811 header.granularity = cpu_to_le64(header.granularity);
1812 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1813 header.desc_offset = cpu_to_le64(header.desc_offset);
1814 header.desc_size = cpu_to_le64(header.desc_size);
1815 header.rgd_offset = cpu_to_le64(header.rgd_offset);
1816 header.gd_offset = cpu_to_le64(header.gd_offset);
1817 header.grain_offset = cpu_to_le64(header.grain_offset);
1818 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1820 header.check_bytes[0] = 0xa;
1821 header.check_bytes[1] = 0x20;
1822 header.check_bytes[2] = 0xd;
1823 header.check_bytes[3] = 0xa;
1825 /* write all the data */
1826 ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
1828 error_setg(errp, QERR_IO_ERROR);
1831 ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
1833 error_setg(errp, QERR_IO_ERROR);
1837 ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9,
1838 PREALLOC_MODE_OFF, errp);
1843 /* write grain directory */
1844 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1845 gd_buf = g_malloc0(gd_buf_size);
1846 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1847 i < gt_count; i++, tmp += gt_size) {
1848 gd_buf[i] = cpu_to_le32(tmp);
1850 ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1851 gd_buf, gd_buf_size, 0);
1853 error_setg(errp, QERR_IO_ERROR);
1857 /* write backup grain directory */
1858 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1859 i < gt_count; i++, tmp += gt_size) {
1860 gd_buf[i] = cpu_to_le32(tmp);
1862 ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1863 gd_buf, gd_buf_size, 0);
1865 error_setg(errp, QERR_IO_ERROR);
1878 static int filename_decompose(const char *filename, char *path, char *prefix,
1879 char *postfix, size_t buf_len, Error **errp)
1883 if (filename == NULL || !strlen(filename)) {
1884 error_setg(errp, "No filename provided");
1887 p = strrchr(filename, '/');
1889 p = strrchr(filename, '\\');
1892 p = strrchr(filename, ':');
1896 if (p - filename >= buf_len) {
1899 pstrcpy(path, p - filename + 1, filename);
1904 q = strrchr(p, '.');
1906 pstrcpy(prefix, buf_len, p);
1909 if (q - p >= buf_len) {
1912 pstrcpy(prefix, q - p + 1, p);
1913 pstrcpy(postfix, buf_len, q);
1918 static int coroutine_fn vmdk_co_create_opts(const char *filename, QemuOpts *opts,
1922 BlockBackend *new_blk = NULL;
1923 Error *local_err = NULL;
1925 int64_t total_size = 0, filesize;
1926 char *adapter_type = NULL;
1927 char *backing_file = NULL;
1928 char *hw_version = NULL;
1931 bool flat, split, compress;
1932 GString *ext_desc_lines;
1933 char *path = g_malloc0(PATH_MAX);
1934 char *prefix = g_malloc0(PATH_MAX);
1935 char *postfix = g_malloc0(PATH_MAX);
1936 char *desc_line = g_malloc0(BUF_SIZE);
1937 char *ext_filename = g_malloc0(PATH_MAX);
1938 char *desc_filename = g_malloc0(PATH_MAX);
1939 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
1940 const char *desc_extent_line;
1941 char *parent_desc_line = g_malloc0(BUF_SIZE);
1942 uint32_t parent_cid = 0xffffffff;
1943 uint32_t number_heads = 16;
1944 bool zeroed_grain = false;
1945 uint32_t desc_offset = 0, desc_len;
1946 const char desc_template[] =
1947 "# Disk DescriptorFile\n"
1950 "parentCID=%" PRIx32 "\n"
1951 "createType=\"%s\"\n"
1954 "# Extent description\n"
1957 "# The Disk Data Base\n"
1960 "ddb.virtualHWVersion = \"%s\"\n"
1961 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1962 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
1963 "ddb.geometry.sectors = \"63\"\n"
1964 "ddb.adapterType = \"%s\"\n";
1966 ext_desc_lines = g_string_new(NULL);
1968 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
1972 /* Read out options */
1973 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1975 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
1976 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1977 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
1978 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) {
1979 if (strcmp(hw_version, "undefined")) {
1981 "compat6 cannot be enabled with hwversion set");
1986 hw_version = g_strdup("6");
1988 if (strcmp(hw_version, "undefined") == 0) {
1990 hw_version = g_strdup("4");
1992 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1993 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) {
1994 zeroed_grain = true;
1997 if (!adapter_type) {
1998 adapter_type = g_strdup("ide");
1999 } else if (strcmp(adapter_type, "ide") &&
2000 strcmp(adapter_type, "buslogic") &&
2001 strcmp(adapter_type, "lsilogic") &&
2002 strcmp(adapter_type, "legacyESX")) {
2003 error_setg(errp, "Unknown adapter type: '%s'", adapter_type);
2007 if (strcmp(adapter_type, "ide") != 0) {
2008 /* that's the number of heads with which vmware operates when
2009 creating, exporting, etc. vmdk files with a non-ide adapter type */
2013 /* Default format to monolithicSparse */
2014 fmt = g_strdup("monolithicSparse");
2015 } else if (strcmp(fmt, "monolithicFlat") &&
2016 strcmp(fmt, "monolithicSparse") &&
2017 strcmp(fmt, "twoGbMaxExtentSparse") &&
2018 strcmp(fmt, "twoGbMaxExtentFlat") &&
2019 strcmp(fmt, "streamOptimized")) {
2020 error_setg(errp, "Unknown subformat: '%s'", fmt);
2024 split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
2025 strcmp(fmt, "twoGbMaxExtentSparse"));
2026 flat = !(strcmp(fmt, "monolithicFlat") &&
2027 strcmp(fmt, "twoGbMaxExtentFlat"));
2028 compress = !strcmp(fmt, "streamOptimized");
2030 desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n";
2032 desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n";
2034 if (flat && backing_file) {
2035 error_setg(errp, "Flat image can't have backing file");
2039 if (flat && zeroed_grain) {
2040 error_setg(errp, "Flat image can't enable zeroed grain");
2046 char *full_backing = g_new0(char, PATH_MAX);
2047 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
2048 full_backing, PATH_MAX,
2051 g_free(full_backing);
2052 error_propagate(errp, local_err);
2057 blk = blk_new_open(full_backing, NULL, NULL,
2058 BDRV_O_NO_BACKING, errp);
2059 g_free(full_backing);
2064 if (strcmp(blk_bs(blk)->drv->format_name, "vmdk")) {
2069 ret = vmdk_read_cid(blk_bs(blk), 0, &parent_cid);
2074 snprintf(parent_desc_line, BUF_SIZE,
2075 "parentFileNameHint=\"%s\"", backing_file);
2078 /* Create extents */
2079 filesize = total_size;
2080 while (filesize > 0) {
2081 int64_t size = filesize;
2083 if (split && size > split_size) {
2087 snprintf(desc_filename, PATH_MAX, "%s-%c%03d%s",
2088 prefix, flat ? 'f' : 's', ++idx, postfix);
2090 snprintf(desc_filename, PATH_MAX, "%s-flat%s", prefix, postfix);
2092 snprintf(desc_filename, PATH_MAX, "%s%s", prefix, postfix);
2094 snprintf(ext_filename, PATH_MAX, "%s%s", path, desc_filename);
2096 if (vmdk_create_extent(ext_filename, size,
2097 flat, compress, zeroed_grain, opts, errp)) {
2103 /* Format description line */
2104 snprintf(desc_line, BUF_SIZE,
2105 desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename);
2106 g_string_append(ext_desc_lines, desc_line);
2108 /* generate descriptor file */
2109 desc = g_strdup_printf(desc_template,
2114 ext_desc_lines->str,
2117 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2120 desc_len = strlen(desc);
2121 /* the descriptor offset = 0x200 */
2122 if (!split && !flat) {
2123 desc_offset = 0x200;
2125 ret = bdrv_create_file(filename, opts, &local_err);
2127 error_propagate(errp, local_err);
2132 new_blk = blk_new_open(filename, NULL, NULL,
2133 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2135 if (new_blk == NULL) {
2136 error_propagate(errp, local_err);
2141 blk_set_allow_write_beyond_eof(new_blk, true);
2143 ret = blk_pwrite(new_blk, desc_offset, desc, desc_len, 0);
2145 error_setg_errno(errp, -ret, "Could not write description");
2148 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2149 * for description file */
2150 if (desc_offset == 0) {
2151 ret = blk_truncate(new_blk, desc_len, PREALLOC_MODE_OFF, errp);
2157 g_free(adapter_type);
2158 g_free(backing_file);
2166 g_free(ext_filename);
2167 g_free(desc_filename);
2168 g_free(parent_desc_line);
2169 g_string_free(ext_desc_lines, true);
2173 static void vmdk_close(BlockDriverState *bs)
2175 BDRVVmdkState *s = bs->opaque;
2177 vmdk_free_extents(bs);
2178 g_free(s->create_type);
2180 migrate_del_blocker(s->migration_blocker);
2181 error_free(s->migration_blocker);
2184 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2186 BDRVVmdkState *s = bs->opaque;
2190 for (i = 0; i < s->num_extents; i++) {
2191 err = bdrv_co_flush(s->extents[i].file->bs);
2199 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2204 BDRVVmdkState *s = bs->opaque;
2206 ret = bdrv_get_allocated_file_size(bs->file->bs);
2210 for (i = 0; i < s->num_extents; i++) {
2211 if (s->extents[i].file == bs->file) {
2214 r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2223 static int vmdk_has_zero_init(BlockDriverState *bs)
2226 BDRVVmdkState *s = bs->opaque;
2228 /* If has a flat extent and its underlying storage doesn't have zero init,
2230 for (i = 0; i < s->num_extents; i++) {
2231 if (s->extents[i].flat) {
2232 if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2240 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2242 ImageInfo *info = g_new0(ImageInfo, 1);
2244 *info = (ImageInfo){
2245 .filename = g_strdup(extent->file->bs->filename),
2246 .format = g_strdup(extent->type),
2247 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2248 .compressed = extent->compressed,
2249 .has_compressed = extent->compressed,
2250 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2251 .has_cluster_size = !extent->flat,
2257 static int coroutine_fn vmdk_co_check(BlockDriverState *bs,
2258 BdrvCheckResult *result,
2261 BDRVVmdkState *s = bs->opaque;
2262 VmdkExtent *extent = NULL;
2263 int64_t sector_num = 0;
2264 int64_t total_sectors = bdrv_nb_sectors(bs);
2266 uint64_t cluster_offset;
2273 if (sector_num >= total_sectors) {
2276 extent = find_extent(s, sector_num, extent);
2279 "ERROR: could not find extent for sector %" PRId64 "\n",
2284 ret = get_cluster_offset(bs, extent, NULL,
2285 sector_num << BDRV_SECTOR_BITS,
2286 false, &cluster_offset, 0, 0);
2287 if (ret == VMDK_ERROR) {
2289 "ERROR: could not get cluster_offset for sector %"
2290 PRId64 "\n", sector_num);
2293 if (ret == VMDK_OK) {
2294 int64_t extent_len = bdrv_getlength(extent->file->bs);
2295 if (extent_len < 0) {
2297 "ERROR: could not get extent file length for sector %"
2298 PRId64 "\n", sector_num);
2302 if (cluster_offset >= extent_len) {
2304 "ERROR: cluster offset for sector %"
2305 PRId64 " points after EOF\n", sector_num);
2310 sector_num += extent->cluster_sectors;
2313 result->corruptions++;
2317 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs)
2320 BDRVVmdkState *s = bs->opaque;
2321 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2322 ImageInfoList **next;
2324 *spec_info = (ImageInfoSpecific){
2325 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2327 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2331 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2332 .create_type = g_strdup(s->create_type),
2334 .parent_cid = s->parent_cid,
2337 next = &spec_info->u.vmdk.data->extents;
2338 for (i = 0; i < s->num_extents; i++) {
2339 *next = g_new0(ImageInfoList, 1);
2340 (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2341 (*next)->next = NULL;
2342 next = &(*next)->next;
2348 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2350 return a->flat == b->flat &&
2351 a->compressed == b->compressed &&
2352 (a->flat || a->cluster_sectors == b->cluster_sectors);
2355 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2358 BDRVVmdkState *s = bs->opaque;
2359 assert(s->num_extents);
2361 /* See if we have multiple extents but they have different cases */
2362 for (i = 1; i < s->num_extents; i++) {
2363 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2367 bdi->needs_compressed_writes = s->extents[0].compressed;
2368 if (!s->extents[0].flat) {
2369 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2374 static QemuOptsList vmdk_create_opts = {
2375 .name = "vmdk-create-opts",
2376 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2379 .name = BLOCK_OPT_SIZE,
2380 .type = QEMU_OPT_SIZE,
2381 .help = "Virtual disk size"
2384 .name = BLOCK_OPT_ADAPTER_TYPE,
2385 .type = QEMU_OPT_STRING,
2386 .help = "Virtual adapter type, can be one of "
2387 "ide (default), lsilogic, buslogic or legacyESX"
2390 .name = BLOCK_OPT_BACKING_FILE,
2391 .type = QEMU_OPT_STRING,
2392 .help = "File name of a base image"
2395 .name = BLOCK_OPT_COMPAT6,
2396 .type = QEMU_OPT_BOOL,
2397 .help = "VMDK version 6 image",
2398 .def_value_str = "off"
2401 .name = BLOCK_OPT_HWVERSION,
2402 .type = QEMU_OPT_STRING,
2403 .help = "VMDK hardware version",
2404 .def_value_str = "undefined"
2407 .name = BLOCK_OPT_SUBFMT,
2408 .type = QEMU_OPT_STRING,
2410 "VMDK flat extent format, can be one of "
2411 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2414 .name = BLOCK_OPT_ZEROED_GRAIN,
2415 .type = QEMU_OPT_BOOL,
2416 .help = "Enable efficient zero writes "
2417 "using the zeroed-grain GTE feature"
2419 { /* end of list */ }
2423 static BlockDriver bdrv_vmdk = {
2424 .format_name = "vmdk",
2425 .instance_size = sizeof(BDRVVmdkState),
2426 .bdrv_probe = vmdk_probe,
2427 .bdrv_open = vmdk_open,
2428 .bdrv_co_check = vmdk_co_check,
2429 .bdrv_reopen_prepare = vmdk_reopen_prepare,
2430 .bdrv_child_perm = bdrv_format_default_perms,
2431 .bdrv_co_preadv = vmdk_co_preadv,
2432 .bdrv_co_pwritev = vmdk_co_pwritev,
2433 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
2434 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
2435 .bdrv_close = vmdk_close,
2436 .bdrv_co_create_opts = vmdk_co_create_opts,
2437 .bdrv_co_flush_to_disk = vmdk_co_flush,
2438 .bdrv_co_block_status = vmdk_co_block_status,
2439 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2440 .bdrv_has_zero_init = vmdk_has_zero_init,
2441 .bdrv_get_specific_info = vmdk_get_specific_info,
2442 .bdrv_refresh_limits = vmdk_refresh_limits,
2443 .bdrv_get_info = vmdk_get_info,
2445 .supports_backing = true,
2446 .create_opts = &vmdk_create_opts,
2449 static void bdrv_vmdk_init(void)
2451 bdrv_register(&bdrv_vmdk);
2454 block_init(bdrv_vmdk_init);