2 * Block driver for the VMDK format
4 * Copyright (c) 2004 Fabrice Bellard
5 * Copyright (c) 2005 Filip Navara
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "block/block_int.h"
29 #include "sysemu/block-backend.h"
30 #include "qapi/qmp/qerror.h"
31 #include "qemu/error-report.h"
32 #include "qemu/module.h"
33 #include "qemu/bswap.h"
34 #include "migration/blocker.h"
35 #include "qemu/cutils.h"
38 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
39 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
40 #define VMDK4_COMPRESSION_DEFLATE 1
41 #define VMDK4_FLAG_NL_DETECT (1 << 0)
42 #define VMDK4_FLAG_RGD (1 << 1)
43 /* Zeroed-grain enable bit */
44 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
45 #define VMDK4_FLAG_COMPRESS (1 << 16)
46 #define VMDK4_FLAG_MARKER (1 << 17)
47 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
49 #define VMDK_GTE_ZEROED 0x1
51 /* VMDK internal error codes */
53 #define VMDK_ERROR (-1)
54 /* Cluster not allocated */
55 #define VMDK_UNALLOC (-2)
56 #define VMDK_ZEROED (-3)
58 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
63 uint32_t disk_sectors;
65 uint32_t l1dir_offset;
67 uint32_t file_sectors;
70 uint32_t sectors_per_track;
71 } QEMU_PACKED VMDK3Header;
80 /* Number of GrainTableEntries per GrainTable */
81 uint32_t num_gtes_per_gt;
84 uint64_t grain_offset;
87 uint16_t compressAlgorithm;
88 } QEMU_PACKED VMDK4Header;
90 #define L2_CACHE_SIZE 16
92 typedef struct VmdkExtent {
101 int64_t flat_start_offset;
102 int64_t l1_table_offset;
103 int64_t l1_backup_table_offset;
105 uint32_t *l1_backup_table;
106 unsigned int l1_size;
107 uint32_t l1_entry_sectors;
109 unsigned int l2_size;
111 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
112 uint32_t l2_cache_counts[L2_CACHE_SIZE];
114 int64_t cluster_sectors;
115 int64_t next_cluster_sector;
119 typedef struct BDRVVmdkState {
121 uint64_t desc_offset;
127 /* Extent array with num_extents entries, ascend ordered by address */
129 Error *migration_blocker;
133 typedef struct VmdkMetaData {
134 unsigned int l1_index;
135 unsigned int l2_index;
136 unsigned int l2_offset;
138 uint32_t *l2_cache_entry;
141 typedef struct VmdkGrainMarker {
145 } QEMU_PACKED VmdkGrainMarker;
148 MARKER_END_OF_STREAM = 0,
149 MARKER_GRAIN_TABLE = 1,
150 MARKER_GRAIN_DIRECTORY = 2,
154 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
161 magic = be32_to_cpu(*(uint32_t *)buf);
162 if (magic == VMDK3_MAGIC ||
163 magic == VMDK4_MAGIC) {
166 const char *p = (const char *)buf;
167 const char *end = p + buf_size;
170 /* skip comment line */
171 while (p < end && *p != '\n') {
178 while (p < end && *p == ' ') {
181 /* skip '\r' if windows line endings used. */
182 if (p < end && *p == '\r') {
185 /* only accept blank lines before 'version=' line */
186 if (p == end || *p != '\n') {
192 if (end - p >= strlen("version=X\n")) {
193 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
194 strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
198 if (end - p >= strlen("version=X\r\n")) {
199 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
200 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
210 #define SECTOR_SIZE 512
211 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
212 #define BUF_SIZE 4096
213 #define HEADER_SIZE 512 /* first sector of 512 bytes */
215 static void vmdk_free_extents(BlockDriverState *bs)
218 BDRVVmdkState *s = bs->opaque;
221 for (i = 0; i < s->num_extents; i++) {
225 g_free(e->l1_backup_table);
227 if (e->file != bs->file) {
228 bdrv_unref_child(bs, e->file);
234 static void vmdk_free_last_extent(BlockDriverState *bs)
236 BDRVVmdkState *s = bs->opaque;
238 if (s->num_extents == 0) {
242 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
245 /* Return -ve errno, or 0 on success and write CID into *pcid. */
246 static int vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
250 const char *p_name, *cid_str;
252 BDRVVmdkState *s = bs->opaque;
255 desc = g_malloc0(DESC_SIZE);
256 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
262 cid_str = "parentCID";
263 cid_str_size = sizeof("parentCID");
266 cid_str_size = sizeof("CID");
269 desc[DESC_SIZE - 1] = '\0';
270 p_name = strstr(desc, cid_str);
271 if (p_name == NULL) {
275 p_name += cid_str_size;
276 if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
288 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
290 char *desc, *tmp_desc;
291 char *p_name, *tmp_str;
292 BDRVVmdkState *s = bs->opaque;
295 desc = g_malloc0(DESC_SIZE);
296 tmp_desc = g_malloc0(DESC_SIZE);
297 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
302 desc[DESC_SIZE - 1] = '\0';
303 tmp_str = strstr(desc, "parentCID");
304 if (tmp_str == NULL) {
309 pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
310 p_name = strstr(desc, "CID");
311 if (p_name != NULL) {
312 p_name += sizeof("CID");
313 snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
314 pstrcat(desc, DESC_SIZE, tmp_desc);
317 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
325 static int vmdk_is_cid_valid(BlockDriverState *bs)
327 BDRVVmdkState *s = bs->opaque;
330 if (!s->cid_checked && bs->backing) {
331 BlockDriverState *p_bs = bs->backing->bs;
333 if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
334 /* read failure: report as not valid */
337 if (s->parent_cid != cur_pcid) {
342 s->cid_checked = true;
347 /* We have nothing to do for VMDK reopen, stubs just return success */
348 static int vmdk_reopen_prepare(BDRVReopenState *state,
349 BlockReopenQueue *queue, Error **errp)
351 assert(state != NULL);
352 assert(state->bs != NULL);
356 static int vmdk_parent_open(BlockDriverState *bs)
360 BDRVVmdkState *s = bs->opaque;
363 desc = g_malloc0(DESC_SIZE + 1);
364 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
370 p_name = strstr(desc, "parentFileNameHint");
371 if (p_name != NULL) {
374 p_name += sizeof("parentFileNameHint") + 1;
375 end_name = strchr(p_name, '\"');
376 if (end_name == NULL) {
380 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
385 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
393 /* Create and append extent to the extent array. Return the added VmdkExtent
394 * address. return NULL if allocation failed. */
395 static int vmdk_add_extent(BlockDriverState *bs,
396 BdrvChild *file, bool flat, int64_t sectors,
397 int64_t l1_offset, int64_t l1_backup_offset,
399 int l2_size, uint64_t cluster_sectors,
400 VmdkExtent **new_extent,
404 BDRVVmdkState *s = bs->opaque;
407 if (cluster_sectors > 0x200000) {
408 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
409 error_setg(errp, "Invalid granularity, image may be corrupt");
412 if (l1_size > 512 * 1024 * 1024) {
413 /* Although with big capacity and small l1_entry_sectors, we can get a
414 * big l1_size, we don't want unbounded value to allocate the table.
415 * Limit it to 512M, which is 16PB for default cluster and L2 table
417 error_setg(errp, "L1 size too big");
421 nb_sectors = bdrv_nb_sectors(file->bs);
422 if (nb_sectors < 0) {
426 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
427 extent = &s->extents[s->num_extents];
430 memset(extent, 0, sizeof(VmdkExtent));
433 extent->sectors = sectors;
434 extent->l1_table_offset = l1_offset;
435 extent->l1_backup_table_offset = l1_backup_offset;
436 extent->l1_size = l1_size;
437 extent->l1_entry_sectors = l2_size * cluster_sectors;
438 extent->l2_size = l2_size;
439 extent->cluster_sectors = flat ? sectors : cluster_sectors;
440 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
442 if (s->num_extents > 1) {
443 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
445 extent->end_sector = extent->sectors;
447 bs->total_sectors = extent->end_sector;
449 *new_extent = extent;
454 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
461 /* read the L1 table */
462 l1_size = extent->l1_size * sizeof(uint32_t);
463 extent->l1_table = g_try_malloc(l1_size);
464 if (l1_size && extent->l1_table == NULL) {
468 ret = bdrv_pread(extent->file,
469 extent->l1_table_offset,
473 error_setg_errno(errp, -ret,
474 "Could not read l1 table from extent '%s'",
475 extent->file->bs->filename);
478 for (i = 0; i < extent->l1_size; i++) {
479 le32_to_cpus(&extent->l1_table[i]);
482 if (extent->l1_backup_table_offset) {
483 extent->l1_backup_table = g_try_malloc(l1_size);
484 if (l1_size && extent->l1_backup_table == NULL) {
488 ret = bdrv_pread(extent->file,
489 extent->l1_backup_table_offset,
490 extent->l1_backup_table,
493 error_setg_errno(errp, -ret,
494 "Could not read l1 backup table from extent '%s'",
495 extent->file->bs->filename);
498 for (i = 0; i < extent->l1_size; i++) {
499 le32_to_cpus(&extent->l1_backup_table[i]);
504 g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE);
507 g_free(extent->l1_backup_table);
509 g_free(extent->l1_table);
513 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
515 int flags, Error **errp)
522 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
524 error_setg_errno(errp, -ret,
525 "Could not read header from file '%s'",
529 ret = vmdk_add_extent(bs, file, false,
530 le32_to_cpu(header.disk_sectors),
531 (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
533 le32_to_cpu(header.l1dir_size),
535 le32_to_cpu(header.granularity),
541 ret = vmdk_init_tables(bs, extent, errp);
543 /* free extent allocated by vmdk_add_extent */
544 vmdk_free_last_extent(bs);
549 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
550 QDict *options, Error **errp);
552 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
558 size = bdrv_getlength(file->bs);
560 error_setg_errno(errp, -size, "Could not access file");
565 /* Both descriptor file and sparse image must be much larger than 4
566 * bytes, also callers of vmdk_read_desc want to compare the first 4
567 * bytes with VMDK4_MAGIC, let's error out if less is read. */
568 error_setg(errp, "File is too small, not a valid image");
572 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */
573 buf = g_malloc(size + 1);
575 ret = bdrv_pread(file, desc_offset, buf, size);
577 error_setg_errno(errp, -ret, "Could not read from file");
586 static int vmdk_open_vmdk4(BlockDriverState *bs,
588 int flags, QDict *options, Error **errp)
592 uint32_t l1_size, l1_entry_sectors;
595 BDRVVmdkState *s = bs->opaque;
596 int64_t l1_backup_offset = 0;
599 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
601 error_setg_errno(errp, -ret,
602 "Could not read header from file '%s'",
606 if (header.capacity == 0) {
607 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
609 char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
613 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
619 if (!s->create_type) {
620 s->create_type = g_strdup("monolithicSparse");
623 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
625 * The footer takes precedence over the header, so read it in. The
626 * footer starts at offset -1024 from the end: One sector for the
627 * footer, and another one for the end-of-stream marker.
634 uint8_t pad[512 - 16];
635 } QEMU_PACKED footer_marker;
639 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
645 uint8_t pad[512 - 16];
646 } QEMU_PACKED eos_marker;
647 } QEMU_PACKED footer;
649 ret = bdrv_pread(file,
650 bs->file->bs->total_sectors * 512 - 1536,
651 &footer, sizeof(footer));
653 error_setg_errno(errp, -ret, "Failed to read footer");
657 /* Some sanity checks for the footer */
658 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
659 le32_to_cpu(footer.footer_marker.size) != 0 ||
660 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
661 le64_to_cpu(footer.eos_marker.val) != 0 ||
662 le32_to_cpu(footer.eos_marker.size) != 0 ||
663 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
665 error_setg(errp, "Invalid footer");
669 header = footer.header;
673 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
674 if (le32_to_cpu(header.version) > 3) {
675 error_setg(errp, "Unsupported VMDK version %" PRIu32,
676 le32_to_cpu(header.version));
678 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
680 /* VMware KB 2064959 explains that version 3 added support for
681 * persistent changed block tracking (CBT), and backup software can
682 * read it as version=1 if it doesn't care about the changed area
683 * information. So we are safe to enable read only. */
684 error_setg(errp, "VMDK version 3 must be read only");
688 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
689 error_setg(errp, "L2 table size too big");
693 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
694 * le64_to_cpu(header.granularity);
695 if (l1_entry_sectors == 0) {
696 error_setg(errp, "L1 entry size is invalid");
699 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
701 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
702 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
704 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
705 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
706 (int64_t)(le64_to_cpu(header.grain_offset)
707 * BDRV_SECTOR_SIZE));
711 ret = vmdk_add_extent(bs, file, false,
712 le64_to_cpu(header.capacity),
713 le64_to_cpu(header.gd_offset) << 9,
716 le32_to_cpu(header.num_gtes_per_gt),
717 le64_to_cpu(header.granularity),
724 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
725 if (extent->compressed) {
726 g_free(s->create_type);
727 s->create_type = g_strdup("streamOptimized");
729 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
730 extent->version = le32_to_cpu(header.version);
731 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
732 ret = vmdk_init_tables(bs, extent, errp);
734 /* free extent allocated by vmdk_add_extent */
735 vmdk_free_last_extent(bs);
740 /* find an option value out of descriptor file */
741 static int vmdk_parse_description(const char *desc, const char *opt_name,
742 char *buf, int buf_size)
744 char *opt_pos, *opt_end;
745 const char *end = desc + strlen(desc);
747 opt_pos = strstr(desc, opt_name);
751 /* Skip "=\"" following opt_name */
752 opt_pos += strlen(opt_name) + 2;
753 if (opt_pos >= end) {
757 while (opt_end < end && *opt_end != '"') {
760 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
763 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
767 /* Open an extent file and append to bs array */
768 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
769 char *buf, QDict *options, Error **errp)
773 magic = ldl_be_p(buf);
776 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
779 return vmdk_open_vmdk4(bs, file, flags, options, errp);
782 error_setg(errp, "Image not in VMDK format");
788 static const char *next_line(const char *s)
799 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
800 const char *desc_file_path, QDict *options,
812 BdrvChild *extent_file;
813 BDRVVmdkState *s = bs->opaque;
815 char extent_opt_prefix[32];
816 Error *local_err = NULL;
818 for (p = desc; *p; p = next_line(p)) {
819 /* parse extent line in one of below formats:
821 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
822 * RW [size in sectors] SPARSE "file-name.vmdk"
823 * RW [size in sectors] VMFS "file-name.vmdk"
824 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
827 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
828 access, §ors, type, fname, &flat_offset);
829 if (matches < 4 || strcmp(access, "RW")) {
831 } else if (!strcmp(type, "FLAT")) {
832 if (matches != 5 || flat_offset < 0) {
835 } else if (!strcmp(type, "VMFS")) {
841 } else if (matches != 4) {
846 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
847 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
848 (strcmp(access, "RW"))) {
852 if (!path_is_absolute(fname) && !path_has_protocol(fname) &&
855 error_setg(errp, "Cannot use relative extent paths with VMDK "
856 "descriptor file '%s'", bs->file->bs->filename);
860 extent_path = g_malloc0(PATH_MAX);
861 path_combine(extent_path, PATH_MAX, desc_file_path, fname);
863 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
866 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
867 bs, &child_file, false, &local_err);
870 error_propagate(errp, local_err);
874 /* save to extents array */
875 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
878 ret = vmdk_add_extent(bs, extent_file, true, sectors,
879 0, 0, 0, 0, 0, &extent, errp);
881 bdrv_unref_child(bs, extent_file);
884 extent->flat_start_offset = flat_offset << 9;
885 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
886 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
887 char *buf = vmdk_read_desc(extent_file, 0, errp);
891 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
896 bdrv_unref_child(bs, extent_file);
899 extent = &s->extents[s->num_extents - 1];
901 error_setg(errp, "Unsupported extent type '%s'", type);
902 bdrv_unref_child(bs, extent_file);
905 extent->type = g_strdup(type);
912 if (np[-1] == '\n') {
915 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
919 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
920 QDict *options, Error **errp)
924 BDRVVmdkState *s = bs->opaque;
926 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
927 error_setg(errp, "invalid VMDK image descriptor");
931 if (strcmp(ct, "monolithicFlat") &&
932 strcmp(ct, "vmfs") &&
933 strcmp(ct, "vmfsSparse") &&
934 strcmp(ct, "twoGbMaxExtentSparse") &&
935 strcmp(ct, "twoGbMaxExtentFlat")) {
936 error_setg(errp, "Unsupported image type '%s'", ct);
940 s->create_type = g_strdup(ct);
942 ret = vmdk_parse_extents(buf, bs, bs->file->bs->exact_filename, options,
948 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
953 BDRVVmdkState *s = bs->opaque;
955 Error *local_err = NULL;
957 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
963 buf = vmdk_read_desc(bs->file, 0, errp);
968 magic = ldl_be_p(buf);
972 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
974 s->desc_offset = 0x200;
977 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
984 /* try to open parent images, if exist */
985 ret = vmdk_parent_open(bs);
989 ret = vmdk_read_cid(bs, 0, &s->cid);
993 ret = vmdk_read_cid(bs, 1, &s->parent_cid);
997 qemu_co_mutex_init(&s->lock);
999 /* Disable migration when VMDK images are used */
1000 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1001 "does not support live migration",
1002 bdrv_get_device_or_node_name(bs));
1003 ret = migrate_add_blocker(s->migration_blocker, &local_err);
1005 error_propagate(errp, local_err);
1006 error_free(s->migration_blocker);
1015 g_free(s->create_type);
1016 s->create_type = NULL;
1017 vmdk_free_extents(bs);
1022 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1024 BDRVVmdkState *s = bs->opaque;
1027 for (i = 0; i < s->num_extents; i++) {
1028 if (!s->extents[i].flat) {
1029 bs->bl.pwrite_zeroes_alignment =
1030 MAX(bs->bl.pwrite_zeroes_alignment,
1031 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1039 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1040 * to the cluster at @cluster_sector_num.
1042 * If @skip_start_sector < @skip_end_sector, the relative range
1043 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1044 * it for call to write user data in the request.
1046 static int get_whole_cluster(BlockDriverState *bs,
1048 uint64_t cluster_offset,
1050 uint64_t skip_start_bytes,
1051 uint64_t skip_end_bytes)
1054 int64_t cluster_bytes;
1055 uint8_t *whole_grain;
1057 /* For COW, align request sector_num to cluster start */
1058 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1059 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1060 whole_grain = qemu_blockalign(bs, cluster_bytes);
1063 memset(whole_grain, 0, skip_start_bytes);
1064 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1067 assert(skip_end_bytes <= cluster_bytes);
1068 /* we will be here if it's first write on non-exist grain(cluster).
1069 * try to read from parent image, if exist */
1070 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1075 /* Read backing data before skip range */
1076 if (skip_start_bytes > 0) {
1078 ret = bdrv_pread(bs->backing, offset, whole_grain,
1085 ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1092 /* Read backing data after skip range */
1093 if (skip_end_bytes < cluster_bytes) {
1095 ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1096 whole_grain + skip_end_bytes,
1097 cluster_bytes - skip_end_bytes);
1103 ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1104 whole_grain + skip_end_bytes,
1105 cluster_bytes - skip_end_bytes);
1114 qemu_vfree(whole_grain);
1118 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1121 offset = cpu_to_le32(offset);
1122 /* update L2 table */
1123 if (bdrv_pwrite_sync(extent->file,
1124 ((int64_t)m_data->l2_offset * 512)
1125 + (m_data->l2_index * sizeof(offset)),
1126 &offset, sizeof(offset)) < 0) {
1129 /* update backup L2 table */
1130 if (extent->l1_backup_table_offset != 0) {
1131 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1132 if (bdrv_pwrite_sync(extent->file,
1133 ((int64_t)m_data->l2_offset * 512)
1134 + (m_data->l2_index * sizeof(offset)),
1135 &offset, sizeof(offset)) < 0) {
1139 if (m_data->l2_cache_entry) {
1140 *m_data->l2_cache_entry = offset;
1147 * get_cluster_offset
1149 * Look up cluster offset in extent file by sector number, and store in
1152 * For flat extents, the start offset as parsed from the description file is
1155 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1156 * offset for a new cluster and update L2 cache. If there is a backing file,
1157 * COW is done before returning; otherwise, zeroes are written to the allocated
1158 * cluster. Both COW and zero writing skips the sector range
1159 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1160 * has new data to write there.
1162 * Returns: VMDK_OK if cluster exists and mapped in the image.
1163 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1164 * VMDK_ERROR if failed.
1166 static int get_cluster_offset(BlockDriverState *bs,
1168 VmdkMetaData *m_data,
1171 uint64_t *cluster_offset,
1172 uint64_t skip_start_bytes,
1173 uint64_t skip_end_bytes)
1175 unsigned int l1_index, l2_offset, l2_index;
1176 int min_index, i, j;
1177 uint32_t min_count, *l2_table;
1178 bool zeroed = false;
1180 int64_t cluster_sector;
1186 *cluster_offset = extent->flat_start_offset;
1190 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1191 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1192 if (l1_index >= extent->l1_size) {
1195 l2_offset = extent->l1_table[l1_index];
1197 return VMDK_UNALLOC;
1199 for (i = 0; i < L2_CACHE_SIZE; i++) {
1200 if (l2_offset == extent->l2_cache_offsets[i]) {
1201 /* increment the hit count */
1202 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1203 for (j = 0; j < L2_CACHE_SIZE; j++) {
1204 extent->l2_cache_counts[j] >>= 1;
1207 l2_table = extent->l2_cache + (i * extent->l2_size);
1211 /* not found: load a new entry in the least used one */
1213 min_count = 0xffffffff;
1214 for (i = 0; i < L2_CACHE_SIZE; i++) {
1215 if (extent->l2_cache_counts[i] < min_count) {
1216 min_count = extent->l2_cache_counts[i];
1220 l2_table = extent->l2_cache + (min_index * extent->l2_size);
1221 if (bdrv_pread(extent->file,
1222 (int64_t)l2_offset * 512,
1224 extent->l2_size * sizeof(uint32_t)
1225 ) != extent->l2_size * sizeof(uint32_t)) {
1229 extent->l2_cache_offsets[min_index] = l2_offset;
1230 extent->l2_cache_counts[min_index] = 1;
1232 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1233 cluster_sector = le32_to_cpu(l2_table[l2_index]);
1235 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1239 if (!cluster_sector || zeroed) {
1241 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1244 cluster_sector = extent->next_cluster_sector;
1245 extent->next_cluster_sector += extent->cluster_sectors;
1247 /* First of all we write grain itself, to avoid race condition
1248 * that may to corrupt the image.
1249 * This problem may occur because of insufficient space on host disk
1250 * or inappropriate VM shutdown.
1252 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1253 offset, skip_start_bytes, skip_end_bytes);
1259 m_data->l1_index = l1_index;
1260 m_data->l2_index = l2_index;
1261 m_data->l2_offset = l2_offset;
1262 m_data->l2_cache_entry = &l2_table[l2_index];
1265 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1269 static VmdkExtent *find_extent(BDRVVmdkState *s,
1270 int64_t sector_num, VmdkExtent *start_hint)
1272 VmdkExtent *extent = start_hint;
1275 extent = &s->extents[0];
1277 while (extent < &s->extents[s->num_extents]) {
1278 if (sector_num < extent->end_sector) {
1286 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1289 uint64_t extent_begin_offset, extent_relative_offset;
1290 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1292 extent_begin_offset =
1293 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1294 extent_relative_offset = offset - extent_begin_offset;
1295 return extent_relative_offset % cluster_size;
1298 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent *extent,
1302 offset = vmdk_find_offset_in_cluster(extent, sector_num * BDRV_SECTOR_SIZE);
1303 return offset / BDRV_SECTOR_SIZE;
1306 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs,
1307 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1309 BDRVVmdkState *s = bs->opaque;
1310 int64_t index_in_cluster, n, ret;
1314 extent = find_extent(s, sector_num, NULL);
1318 qemu_co_mutex_lock(&s->lock);
1319 ret = get_cluster_offset(bs, extent, NULL,
1320 sector_num * 512, false, &offset,
1322 qemu_co_mutex_unlock(&s->lock);
1324 index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num);
1333 ret = BDRV_BLOCK_ZERO;
1336 ret = BDRV_BLOCK_DATA;
1337 if (!extent->compressed) {
1338 ret |= BDRV_BLOCK_OFFSET_VALID;
1339 ret |= (offset + (index_in_cluster << BDRV_SECTOR_BITS))
1340 & BDRV_BLOCK_OFFSET_MASK;
1342 *file = extent->file->bs;
1346 n = extent->cluster_sectors - index_in_cluster;
1347 if (n > nb_sectors) {
1354 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1355 int64_t offset_in_cluster, QEMUIOVector *qiov,
1356 uint64_t qiov_offset, uint64_t n_bytes,
1360 VmdkGrainMarker *data = NULL;
1362 QEMUIOVector local_qiov;
1364 int64_t write_offset;
1365 int64_t write_end_sector;
1367 if (extent->compressed) {
1368 void *compressed_data;
1370 if (!extent->has_marker) {
1374 buf_len = (extent->cluster_sectors << 9) * 2;
1375 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1377 compressed_data = g_malloc(n_bytes);
1378 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1379 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1380 g_free(compressed_data);
1382 if (ret != Z_OK || buf_len == 0) {
1387 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1388 data->size = cpu_to_le32(buf_len);
1390 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1391 iov = (struct iovec) {
1395 qemu_iovec_init_external(&local_qiov, &iov, 1);
1397 qemu_iovec_init(&local_qiov, qiov->niov);
1398 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1401 write_offset = cluster_offset + offset_in_cluster,
1402 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1405 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1407 if (extent->compressed) {
1408 extent->next_cluster_sector = write_end_sector;
1410 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1420 if (!extent->compressed) {
1421 qemu_iovec_destroy(&local_qiov);
1426 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1427 int64_t offset_in_cluster, QEMUIOVector *qiov,
1431 int cluster_bytes, buf_bytes;
1432 uint8_t *cluster_buf, *compressed_data;
1433 uint8_t *uncomp_buf;
1435 VmdkGrainMarker *marker;
1439 if (!extent->compressed) {
1440 ret = bdrv_co_preadv(extent->file,
1441 cluster_offset + offset_in_cluster, bytes,
1448 cluster_bytes = extent->cluster_sectors * 512;
1449 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1450 buf_bytes = cluster_bytes * 2;
1451 cluster_buf = g_malloc(buf_bytes);
1452 uncomp_buf = g_malloc(cluster_bytes);
1453 ret = bdrv_pread(extent->file,
1455 cluster_buf, buf_bytes);
1459 compressed_data = cluster_buf;
1460 buf_len = cluster_bytes;
1461 data_len = cluster_bytes;
1462 if (extent->has_marker) {
1463 marker = (VmdkGrainMarker *)cluster_buf;
1464 compressed_data = marker->data;
1465 data_len = le32_to_cpu(marker->size);
1467 if (!data_len || data_len > buf_bytes) {
1471 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1477 if (offset_in_cluster < 0 ||
1478 offset_in_cluster + bytes > buf_len) {
1482 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1487 g_free(cluster_buf);
1491 static int coroutine_fn
1492 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1493 QEMUIOVector *qiov, int flags)
1495 BDRVVmdkState *s = bs->opaque;
1497 uint64_t n_bytes, offset_in_cluster;
1498 VmdkExtent *extent = NULL;
1499 QEMUIOVector local_qiov;
1500 uint64_t cluster_offset;
1501 uint64_t bytes_done = 0;
1503 qemu_iovec_init(&local_qiov, qiov->niov);
1504 qemu_co_mutex_lock(&s->lock);
1507 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1512 ret = get_cluster_offset(bs, extent, NULL,
1513 offset, false, &cluster_offset, 0, 0);
1514 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1516 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1517 - offset_in_cluster);
1519 if (ret != VMDK_OK) {
1520 /* if not allocated, try to read from parent image, if exist */
1521 if (bs->backing && ret != VMDK_ZEROED) {
1522 if (!vmdk_is_cid_valid(bs)) {
1527 qemu_iovec_reset(&local_qiov);
1528 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1530 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1536 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1539 qemu_iovec_reset(&local_qiov);
1540 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1542 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1543 &local_qiov, n_bytes);
1550 bytes_done += n_bytes;
1555 qemu_co_mutex_unlock(&s->lock);
1556 qemu_iovec_destroy(&local_qiov);
1563 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1564 * if possible, otherwise return -ENOTSUP.
1565 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1566 * with each cluster. By dry run we can find if the zero write
1567 * is possible without modifying image data.
1569 * Returns: error code with 0 for success.
1571 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1572 uint64_t bytes, QEMUIOVector *qiov,
1573 bool zeroed, bool zero_dry_run)
1575 BDRVVmdkState *s = bs->opaque;
1576 VmdkExtent *extent = NULL;
1578 int64_t offset_in_cluster, n_bytes;
1579 uint64_t cluster_offset;
1580 uint64_t bytes_done = 0;
1581 VmdkMetaData m_data;
1583 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1584 error_report("Wrong offset: offset=0x%" PRIx64
1585 " total_sectors=0x%" PRIx64,
1586 offset, bs->total_sectors);
1591 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1595 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1596 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1597 - offset_in_cluster);
1599 ret = get_cluster_offset(bs, extent, &m_data, offset,
1600 !(extent->compressed || zeroed),
1601 &cluster_offset, offset_in_cluster,
1602 offset_in_cluster + n_bytes);
1603 if (extent->compressed) {
1604 if (ret == VMDK_OK) {
1605 /* Refuse write to allocated cluster for streamOptimized */
1606 error_report("Could not write to allocated cluster"
1607 " for streamOptimized");
1611 ret = get_cluster_offset(bs, extent, &m_data, offset,
1612 true, &cluster_offset, 0, 0);
1615 if (ret == VMDK_ERROR) {
1619 /* Do zeroed write, buf is ignored */
1620 if (extent->has_zero_grain &&
1621 offset_in_cluster == 0 &&
1622 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
1623 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1624 if (!zero_dry_run) {
1625 /* update L2 tables */
1626 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1635 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
1636 qiov, bytes_done, n_bytes, offset);
1641 /* update L2 tables */
1642 if (vmdk_L2update(extent, &m_data,
1643 cluster_offset >> BDRV_SECTOR_BITS)
1651 bytes_done += n_bytes;
1653 /* update CID on the first write every time the virtual disk is
1655 if (!s->cid_updated) {
1656 ret = vmdk_write_cid(bs, g_random_int());
1660 s->cid_updated = true;
1666 static int coroutine_fn
1667 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1668 QEMUIOVector *qiov, int flags)
1671 BDRVVmdkState *s = bs->opaque;
1672 qemu_co_mutex_lock(&s->lock);
1673 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
1674 qemu_co_mutex_unlock(&s->lock);
1678 static int coroutine_fn
1679 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1680 uint64_t bytes, QEMUIOVector *qiov)
1682 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
1685 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
1688 BdrvRequestFlags flags)
1691 BDRVVmdkState *s = bs->opaque;
1693 qemu_co_mutex_lock(&s->lock);
1694 /* write zeroes could fail if sectors not aligned to cluster, test it with
1695 * dry_run == true before really updating image */
1696 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
1698 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
1700 qemu_co_mutex_unlock(&s->lock);
1704 static int vmdk_create_extent(const char *filename, int64_t filesize,
1705 bool flat, bool compress, bool zeroed_grain,
1706 QemuOpts *opts, Error **errp)
1709 BlockBackend *blk = NULL;
1711 Error *local_err = NULL;
1712 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1713 uint32_t *gd_buf = NULL;
1716 ret = bdrv_create_file(filename, opts, &local_err);
1718 error_propagate(errp, local_err);
1722 blk = blk_new_open(filename, NULL, NULL,
1723 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
1726 error_propagate(errp, local_err);
1731 blk_set_allow_write_beyond_eof(blk, true);
1734 ret = blk_truncate(blk, filesize, PREALLOC_MODE_OFF, errp);
1737 magic = cpu_to_be32(VMDK4_MAGIC);
1738 memset(&header, 0, sizeof(header));
1741 } else if (zeroed_grain) {
1746 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1747 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1748 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1749 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1750 header.capacity = filesize / BDRV_SECTOR_SIZE;
1751 header.granularity = 128;
1752 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1754 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1755 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1757 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1758 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1760 header.desc_offset = 1;
1761 header.desc_size = 20;
1762 header.rgd_offset = header.desc_offset + header.desc_size;
1763 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1764 header.grain_offset =
1765 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1766 header.granularity);
1767 /* swap endianness for all header fields */
1768 header.version = cpu_to_le32(header.version);
1769 header.flags = cpu_to_le32(header.flags);
1770 header.capacity = cpu_to_le64(header.capacity);
1771 header.granularity = cpu_to_le64(header.granularity);
1772 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1773 header.desc_offset = cpu_to_le64(header.desc_offset);
1774 header.desc_size = cpu_to_le64(header.desc_size);
1775 header.rgd_offset = cpu_to_le64(header.rgd_offset);
1776 header.gd_offset = cpu_to_le64(header.gd_offset);
1777 header.grain_offset = cpu_to_le64(header.grain_offset);
1778 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1780 header.check_bytes[0] = 0xa;
1781 header.check_bytes[1] = 0x20;
1782 header.check_bytes[2] = 0xd;
1783 header.check_bytes[3] = 0xa;
1785 /* write all the data */
1786 ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
1788 error_setg(errp, QERR_IO_ERROR);
1791 ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
1793 error_setg(errp, QERR_IO_ERROR);
1797 ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9,
1798 PREALLOC_MODE_OFF, errp);
1803 /* write grain directory */
1804 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1805 gd_buf = g_malloc0(gd_buf_size);
1806 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1807 i < gt_count; i++, tmp += gt_size) {
1808 gd_buf[i] = cpu_to_le32(tmp);
1810 ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1811 gd_buf, gd_buf_size, 0);
1813 error_setg(errp, QERR_IO_ERROR);
1817 /* write backup grain directory */
1818 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1819 i < gt_count; i++, tmp += gt_size) {
1820 gd_buf[i] = cpu_to_le32(tmp);
1822 ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1823 gd_buf, gd_buf_size, 0);
1825 error_setg(errp, QERR_IO_ERROR);
1838 static int filename_decompose(const char *filename, char *path, char *prefix,
1839 char *postfix, size_t buf_len, Error **errp)
1843 if (filename == NULL || !strlen(filename)) {
1844 error_setg(errp, "No filename provided");
1847 p = strrchr(filename, '/');
1849 p = strrchr(filename, '\\');
1852 p = strrchr(filename, ':');
1856 if (p - filename >= buf_len) {
1859 pstrcpy(path, p - filename + 1, filename);
1864 q = strrchr(p, '.');
1866 pstrcpy(prefix, buf_len, p);
1869 if (q - p >= buf_len) {
1872 pstrcpy(prefix, q - p + 1, p);
1873 pstrcpy(postfix, buf_len, q);
1878 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp)
1881 BlockBackend *new_blk = NULL;
1882 Error *local_err = NULL;
1884 int64_t total_size = 0, filesize;
1885 char *adapter_type = NULL;
1886 char *backing_file = NULL;
1887 char *hw_version = NULL;
1890 bool flat, split, compress;
1891 GString *ext_desc_lines;
1892 char *path = g_malloc0(PATH_MAX);
1893 char *prefix = g_malloc0(PATH_MAX);
1894 char *postfix = g_malloc0(PATH_MAX);
1895 char *desc_line = g_malloc0(BUF_SIZE);
1896 char *ext_filename = g_malloc0(PATH_MAX);
1897 char *desc_filename = g_malloc0(PATH_MAX);
1898 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
1899 const char *desc_extent_line;
1900 char *parent_desc_line = g_malloc0(BUF_SIZE);
1901 uint32_t parent_cid = 0xffffffff;
1902 uint32_t number_heads = 16;
1903 bool zeroed_grain = false;
1904 uint32_t desc_offset = 0, desc_len;
1905 const char desc_template[] =
1906 "# Disk DescriptorFile\n"
1909 "parentCID=%" PRIx32 "\n"
1910 "createType=\"%s\"\n"
1913 "# Extent description\n"
1916 "# The Disk Data Base\n"
1919 "ddb.virtualHWVersion = \"%s\"\n"
1920 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1921 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
1922 "ddb.geometry.sectors = \"63\"\n"
1923 "ddb.adapterType = \"%s\"\n";
1925 ext_desc_lines = g_string_new(NULL);
1927 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
1931 /* Read out options */
1932 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1934 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
1935 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1936 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
1937 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) {
1938 if (strcmp(hw_version, "undefined")) {
1940 "compat6 cannot be enabled with hwversion set");
1945 hw_version = g_strdup("6");
1947 if (strcmp(hw_version, "undefined") == 0) {
1949 hw_version = g_strdup("4");
1951 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1952 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) {
1953 zeroed_grain = true;
1956 if (!adapter_type) {
1957 adapter_type = g_strdup("ide");
1958 } else if (strcmp(adapter_type, "ide") &&
1959 strcmp(adapter_type, "buslogic") &&
1960 strcmp(adapter_type, "lsilogic") &&
1961 strcmp(adapter_type, "legacyESX")) {
1962 error_setg(errp, "Unknown adapter type: '%s'", adapter_type);
1966 if (strcmp(adapter_type, "ide") != 0) {
1967 /* that's the number of heads with which vmware operates when
1968 creating, exporting, etc. vmdk files with a non-ide adapter type */
1972 /* Default format to monolithicSparse */
1973 fmt = g_strdup("monolithicSparse");
1974 } else if (strcmp(fmt, "monolithicFlat") &&
1975 strcmp(fmt, "monolithicSparse") &&
1976 strcmp(fmt, "twoGbMaxExtentSparse") &&
1977 strcmp(fmt, "twoGbMaxExtentFlat") &&
1978 strcmp(fmt, "streamOptimized")) {
1979 error_setg(errp, "Unknown subformat: '%s'", fmt);
1983 split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1984 strcmp(fmt, "twoGbMaxExtentSparse"));
1985 flat = !(strcmp(fmt, "monolithicFlat") &&
1986 strcmp(fmt, "twoGbMaxExtentFlat"));
1987 compress = !strcmp(fmt, "streamOptimized");
1989 desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n";
1991 desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n";
1993 if (flat && backing_file) {
1994 error_setg(errp, "Flat image can't have backing file");
1998 if (flat && zeroed_grain) {
1999 error_setg(errp, "Flat image can't enable zeroed grain");
2005 char *full_backing = g_new0(char, PATH_MAX);
2006 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
2007 full_backing, PATH_MAX,
2010 g_free(full_backing);
2011 error_propagate(errp, local_err);
2016 blk = blk_new_open(full_backing, NULL, NULL,
2017 BDRV_O_NO_BACKING, errp);
2018 g_free(full_backing);
2023 if (strcmp(blk_bs(blk)->drv->format_name, "vmdk")) {
2028 ret = vmdk_read_cid(blk_bs(blk), 0, &parent_cid);
2033 snprintf(parent_desc_line, BUF_SIZE,
2034 "parentFileNameHint=\"%s\"", backing_file);
2037 /* Create extents */
2038 filesize = total_size;
2039 while (filesize > 0) {
2040 int64_t size = filesize;
2042 if (split && size > split_size) {
2046 snprintf(desc_filename, PATH_MAX, "%s-%c%03d%s",
2047 prefix, flat ? 'f' : 's', ++idx, postfix);
2049 snprintf(desc_filename, PATH_MAX, "%s-flat%s", prefix, postfix);
2051 snprintf(desc_filename, PATH_MAX, "%s%s", prefix, postfix);
2053 snprintf(ext_filename, PATH_MAX, "%s%s", path, desc_filename);
2055 if (vmdk_create_extent(ext_filename, size,
2056 flat, compress, zeroed_grain, opts, errp)) {
2062 /* Format description line */
2063 snprintf(desc_line, BUF_SIZE,
2064 desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename);
2065 g_string_append(ext_desc_lines, desc_line);
2067 /* generate descriptor file */
2068 desc = g_strdup_printf(desc_template,
2073 ext_desc_lines->str,
2076 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2079 desc_len = strlen(desc);
2080 /* the descriptor offset = 0x200 */
2081 if (!split && !flat) {
2082 desc_offset = 0x200;
2084 ret = bdrv_create_file(filename, opts, &local_err);
2086 error_propagate(errp, local_err);
2091 new_blk = blk_new_open(filename, NULL, NULL,
2092 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2094 if (new_blk == NULL) {
2095 error_propagate(errp, local_err);
2100 blk_set_allow_write_beyond_eof(new_blk, true);
2102 ret = blk_pwrite(new_blk, desc_offset, desc, desc_len, 0);
2104 error_setg_errno(errp, -ret, "Could not write description");
2107 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2108 * for description file */
2109 if (desc_offset == 0) {
2110 ret = blk_truncate(new_blk, desc_len, PREALLOC_MODE_OFF, errp);
2116 g_free(adapter_type);
2117 g_free(backing_file);
2125 g_free(ext_filename);
2126 g_free(desc_filename);
2127 g_free(parent_desc_line);
2128 g_string_free(ext_desc_lines, true);
2132 static void vmdk_close(BlockDriverState *bs)
2134 BDRVVmdkState *s = bs->opaque;
2136 vmdk_free_extents(bs);
2137 g_free(s->create_type);
2139 migrate_del_blocker(s->migration_blocker);
2140 error_free(s->migration_blocker);
2143 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2145 BDRVVmdkState *s = bs->opaque;
2149 for (i = 0; i < s->num_extents; i++) {
2150 err = bdrv_co_flush(s->extents[i].file->bs);
2158 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2163 BDRVVmdkState *s = bs->opaque;
2165 ret = bdrv_get_allocated_file_size(bs->file->bs);
2169 for (i = 0; i < s->num_extents; i++) {
2170 if (s->extents[i].file == bs->file) {
2173 r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2182 static int vmdk_has_zero_init(BlockDriverState *bs)
2185 BDRVVmdkState *s = bs->opaque;
2187 /* If has a flat extent and its underlying storage doesn't have zero init,
2189 for (i = 0; i < s->num_extents; i++) {
2190 if (s->extents[i].flat) {
2191 if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2199 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2201 ImageInfo *info = g_new0(ImageInfo, 1);
2203 *info = (ImageInfo){
2204 .filename = g_strdup(extent->file->bs->filename),
2205 .format = g_strdup(extent->type),
2206 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2207 .compressed = extent->compressed,
2208 .has_compressed = extent->compressed,
2209 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2210 .has_cluster_size = !extent->flat,
2216 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result,
2219 BDRVVmdkState *s = bs->opaque;
2220 VmdkExtent *extent = NULL;
2221 int64_t sector_num = 0;
2222 int64_t total_sectors = bdrv_nb_sectors(bs);
2224 uint64_t cluster_offset;
2231 if (sector_num >= total_sectors) {
2234 extent = find_extent(s, sector_num, extent);
2237 "ERROR: could not find extent for sector %" PRId64 "\n",
2242 ret = get_cluster_offset(bs, extent, NULL,
2243 sector_num << BDRV_SECTOR_BITS,
2244 false, &cluster_offset, 0, 0);
2245 if (ret == VMDK_ERROR) {
2247 "ERROR: could not get cluster_offset for sector %"
2248 PRId64 "\n", sector_num);
2251 if (ret == VMDK_OK) {
2252 int64_t extent_len = bdrv_getlength(extent->file->bs);
2253 if (extent_len < 0) {
2255 "ERROR: could not get extent file length for sector %"
2256 PRId64 "\n", sector_num);
2260 if (cluster_offset >= extent_len) {
2262 "ERROR: cluster offset for sector %"
2263 PRId64 " points after EOF\n", sector_num);
2268 sector_num += extent->cluster_sectors;
2271 result->corruptions++;
2275 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs)
2278 BDRVVmdkState *s = bs->opaque;
2279 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2280 ImageInfoList **next;
2282 *spec_info = (ImageInfoSpecific){
2283 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2285 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2289 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2290 .create_type = g_strdup(s->create_type),
2292 .parent_cid = s->parent_cid,
2295 next = &spec_info->u.vmdk.data->extents;
2296 for (i = 0; i < s->num_extents; i++) {
2297 *next = g_new0(ImageInfoList, 1);
2298 (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2299 (*next)->next = NULL;
2300 next = &(*next)->next;
2306 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2308 return a->flat == b->flat &&
2309 a->compressed == b->compressed &&
2310 (a->flat || a->cluster_sectors == b->cluster_sectors);
2313 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2316 BDRVVmdkState *s = bs->opaque;
2317 assert(s->num_extents);
2319 /* See if we have multiple extents but they have different cases */
2320 for (i = 1; i < s->num_extents; i++) {
2321 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2325 bdi->needs_compressed_writes = s->extents[0].compressed;
2326 if (!s->extents[0].flat) {
2327 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2332 static QemuOptsList vmdk_create_opts = {
2333 .name = "vmdk-create-opts",
2334 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2337 .name = BLOCK_OPT_SIZE,
2338 .type = QEMU_OPT_SIZE,
2339 .help = "Virtual disk size"
2342 .name = BLOCK_OPT_ADAPTER_TYPE,
2343 .type = QEMU_OPT_STRING,
2344 .help = "Virtual adapter type, can be one of "
2345 "ide (default), lsilogic, buslogic or legacyESX"
2348 .name = BLOCK_OPT_BACKING_FILE,
2349 .type = QEMU_OPT_STRING,
2350 .help = "File name of a base image"
2353 .name = BLOCK_OPT_COMPAT6,
2354 .type = QEMU_OPT_BOOL,
2355 .help = "VMDK version 6 image",
2356 .def_value_str = "off"
2359 .name = BLOCK_OPT_HWVERSION,
2360 .type = QEMU_OPT_STRING,
2361 .help = "VMDK hardware version",
2362 .def_value_str = "undefined"
2365 .name = BLOCK_OPT_SUBFMT,
2366 .type = QEMU_OPT_STRING,
2368 "VMDK flat extent format, can be one of "
2369 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2372 .name = BLOCK_OPT_ZEROED_GRAIN,
2373 .type = QEMU_OPT_BOOL,
2374 .help = "Enable efficient zero writes "
2375 "using the zeroed-grain GTE feature"
2377 { /* end of list */ }
2381 static BlockDriver bdrv_vmdk = {
2382 .format_name = "vmdk",
2383 .instance_size = sizeof(BDRVVmdkState),
2384 .bdrv_probe = vmdk_probe,
2385 .bdrv_open = vmdk_open,
2386 .bdrv_check = vmdk_check,
2387 .bdrv_reopen_prepare = vmdk_reopen_prepare,
2388 .bdrv_child_perm = bdrv_format_default_perms,
2389 .bdrv_co_preadv = vmdk_co_preadv,
2390 .bdrv_co_pwritev = vmdk_co_pwritev,
2391 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
2392 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
2393 .bdrv_close = vmdk_close,
2394 .bdrv_create = vmdk_create,
2395 .bdrv_co_flush_to_disk = vmdk_co_flush,
2396 .bdrv_co_get_block_status = vmdk_co_get_block_status,
2397 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2398 .bdrv_has_zero_init = vmdk_has_zero_init,
2399 .bdrv_get_specific_info = vmdk_get_specific_info,
2400 .bdrv_refresh_limits = vmdk_refresh_limits,
2401 .bdrv_get_info = vmdk_get_info,
2403 .supports_backing = true,
2404 .create_opts = &vmdk_create_opts,
2407 static void bdrv_vmdk_init(void)
2409 bdrv_register(&bdrv_vmdk);
2412 block_init(bdrv_vmdk_init);