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_GTE_ZEROED 0x1
52 /* VMDK internal error codes */
54 #define VMDK_ERROR (-1)
55 /* Cluster not allocated */
56 #define VMDK_UNALLOC (-2)
57 #define VMDK_ZEROED (-3)
59 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
64 uint32_t disk_sectors;
66 uint32_t l1dir_offset;
68 uint32_t file_sectors;
71 uint32_t sectors_per_track;
72 } QEMU_PACKED VMDK3Header;
81 /* Number of GrainTableEntries per GrainTable */
82 uint32_t num_gtes_per_gt;
85 uint64_t grain_offset;
88 uint16_t compressAlgorithm;
89 } QEMU_PACKED VMDK4Header;
91 #define L2_CACHE_SIZE 16
93 typedef struct VmdkExtent {
102 int64_t flat_start_offset;
103 int64_t l1_table_offset;
104 int64_t l1_backup_table_offset;
106 uint32_t *l1_backup_table;
107 unsigned int l1_size;
108 uint32_t l1_entry_sectors;
110 unsigned int l2_size;
112 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
113 uint32_t l2_cache_counts[L2_CACHE_SIZE];
115 int64_t cluster_sectors;
116 int64_t next_cluster_sector;
120 typedef struct BDRVVmdkState {
122 uint64_t desc_offset;
128 /* Extent array with num_extents entries, ascend ordered by address */
130 Error *migration_blocker;
134 typedef struct VmdkMetaData {
135 unsigned int l1_index;
136 unsigned int l2_index;
137 unsigned int l2_offset;
139 uint32_t *l2_cache_entry;
142 typedef struct VmdkGrainMarker {
146 } QEMU_PACKED VmdkGrainMarker;
149 MARKER_END_OF_STREAM = 0,
150 MARKER_GRAIN_TABLE = 1,
151 MARKER_GRAIN_DIRECTORY = 2,
155 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
162 magic = be32_to_cpu(*(uint32_t *)buf);
163 if (magic == VMDK3_MAGIC ||
164 magic == VMDK4_MAGIC) {
167 const char *p = (const char *)buf;
168 const char *end = p + buf_size;
171 /* skip comment line */
172 while (p < end && *p != '\n') {
179 while (p < end && *p == ' ') {
182 /* skip '\r' if windows line endings used. */
183 if (p < end && *p == '\r') {
186 /* only accept blank lines before 'version=' line */
187 if (p == end || *p != '\n') {
193 if (end - p >= strlen("version=X\n")) {
194 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
195 strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
199 if (end - p >= strlen("version=X\r\n")) {
200 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
201 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
211 #define SECTOR_SIZE 512
212 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
213 #define BUF_SIZE 4096
214 #define HEADER_SIZE 512 /* first sector of 512 bytes */
216 static void vmdk_free_extents(BlockDriverState *bs)
219 BDRVVmdkState *s = bs->opaque;
222 for (i = 0; i < s->num_extents; i++) {
226 g_free(e->l1_backup_table);
228 if (e->file != bs->file) {
229 bdrv_unref_child(bs, e->file);
235 static void vmdk_free_last_extent(BlockDriverState *bs)
237 BDRVVmdkState *s = bs->opaque;
239 if (s->num_extents == 0) {
243 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
246 /* Return -ve errno, or 0 on success and write CID into *pcid. */
247 static int vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
251 const char *p_name, *cid_str;
253 BDRVVmdkState *s = bs->opaque;
256 desc = g_malloc0(DESC_SIZE);
257 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
263 cid_str = "parentCID";
264 cid_str_size = sizeof("parentCID");
267 cid_str_size = sizeof("CID");
270 desc[DESC_SIZE - 1] = '\0';
271 p_name = strstr(desc, cid_str);
272 if (p_name == NULL) {
276 p_name += cid_str_size;
277 if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
289 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
291 char *desc, *tmp_desc;
292 char *p_name, *tmp_str;
293 BDRVVmdkState *s = bs->opaque;
296 desc = g_malloc0(DESC_SIZE);
297 tmp_desc = g_malloc0(DESC_SIZE);
298 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
303 desc[DESC_SIZE - 1] = '\0';
304 tmp_str = strstr(desc, "parentCID");
305 if (tmp_str == NULL) {
310 pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
311 p_name = strstr(desc, "CID");
312 if (p_name != NULL) {
313 p_name += sizeof("CID");
314 snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
315 pstrcat(desc, DESC_SIZE, tmp_desc);
318 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
326 static int vmdk_is_cid_valid(BlockDriverState *bs)
328 BDRVVmdkState *s = bs->opaque;
331 if (!s->cid_checked && bs->backing) {
332 BlockDriverState *p_bs = bs->backing->bs;
334 if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
335 /* read failure: report as not valid */
338 if (s->parent_cid != cur_pcid) {
343 s->cid_checked = true;
348 /* We have nothing to do for VMDK reopen, stubs just return success */
349 static int vmdk_reopen_prepare(BDRVReopenState *state,
350 BlockReopenQueue *queue, Error **errp)
352 assert(state != NULL);
353 assert(state->bs != NULL);
357 static int vmdk_parent_open(BlockDriverState *bs)
361 BDRVVmdkState *s = bs->opaque;
364 desc = g_malloc0(DESC_SIZE + 1);
365 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
371 p_name = strstr(desc, "parentFileNameHint");
372 if (p_name != NULL) {
375 p_name += sizeof("parentFileNameHint") + 1;
376 end_name = strchr(p_name, '\"');
377 if (end_name == NULL) {
381 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
386 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
394 /* Create and append extent to the extent array. Return the added VmdkExtent
395 * address. return NULL if allocation failed. */
396 static int vmdk_add_extent(BlockDriverState *bs,
397 BdrvChild *file, bool flat, int64_t sectors,
398 int64_t l1_offset, int64_t l1_backup_offset,
400 int l2_size, uint64_t cluster_sectors,
401 VmdkExtent **new_extent,
405 BDRVVmdkState *s = bs->opaque;
408 if (cluster_sectors > 0x200000) {
409 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
410 error_setg(errp, "Invalid granularity, image may be corrupt");
413 if (l1_size > 512 * 1024 * 1024) {
414 /* Although with big capacity and small l1_entry_sectors, we can get a
415 * big l1_size, we don't want unbounded value to allocate the table.
416 * Limit it to 512M, which is 16PB for default cluster and L2 table
418 error_setg(errp, "L1 size too big");
422 nb_sectors = bdrv_nb_sectors(file->bs);
423 if (nb_sectors < 0) {
427 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
428 extent = &s->extents[s->num_extents];
431 memset(extent, 0, sizeof(VmdkExtent));
434 extent->sectors = sectors;
435 extent->l1_table_offset = l1_offset;
436 extent->l1_backup_table_offset = l1_backup_offset;
437 extent->l1_size = l1_size;
438 extent->l1_entry_sectors = l2_size * cluster_sectors;
439 extent->l2_size = l2_size;
440 extent->cluster_sectors = flat ? sectors : cluster_sectors;
441 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
443 if (s->num_extents > 1) {
444 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
446 extent->end_sector = extent->sectors;
448 bs->total_sectors = extent->end_sector;
450 *new_extent = extent;
455 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
462 /* read the L1 table */
463 l1_size = extent->l1_size * sizeof(uint32_t);
464 extent->l1_table = g_try_malloc(l1_size);
465 if (l1_size && extent->l1_table == NULL) {
469 ret = bdrv_pread(extent->file,
470 extent->l1_table_offset,
474 error_setg_errno(errp, -ret,
475 "Could not read l1 table from extent '%s'",
476 extent->file->bs->filename);
479 for (i = 0; i < extent->l1_size; i++) {
480 le32_to_cpus(&extent->l1_table[i]);
483 if (extent->l1_backup_table_offset) {
484 extent->l1_backup_table = g_try_malloc(l1_size);
485 if (l1_size && extent->l1_backup_table == NULL) {
489 ret = bdrv_pread(extent->file,
490 extent->l1_backup_table_offset,
491 extent->l1_backup_table,
494 error_setg_errno(errp, -ret,
495 "Could not read l1 backup table from extent '%s'",
496 extent->file->bs->filename);
499 for (i = 0; i < extent->l1_size; i++) {
500 le32_to_cpus(&extent->l1_backup_table[i]);
505 g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE);
508 g_free(extent->l1_backup_table);
510 g_free(extent->l1_table);
514 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
516 int flags, Error **errp)
523 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
525 error_setg_errno(errp, -ret,
526 "Could not read header from file '%s'",
530 ret = vmdk_add_extent(bs, file, false,
531 le32_to_cpu(header.disk_sectors),
532 (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
534 le32_to_cpu(header.l1dir_size),
536 le32_to_cpu(header.granularity),
542 ret = vmdk_init_tables(bs, extent, errp);
544 /* free extent allocated by vmdk_add_extent */
545 vmdk_free_last_extent(bs);
550 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
551 QDict *options, Error **errp);
553 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
559 size = bdrv_getlength(file->bs);
561 error_setg_errno(errp, -size, "Could not access file");
566 /* Both descriptor file and sparse image must be much larger than 4
567 * bytes, also callers of vmdk_read_desc want to compare the first 4
568 * bytes with VMDK4_MAGIC, let's error out if less is read. */
569 error_setg(errp, "File is too small, not a valid image");
573 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */
574 buf = g_malloc(size + 1);
576 ret = bdrv_pread(file, desc_offset, buf, size);
578 error_setg_errno(errp, -ret, "Could not read from file");
587 static int vmdk_open_vmdk4(BlockDriverState *bs,
589 int flags, QDict *options, Error **errp)
593 uint32_t l1_size, l1_entry_sectors;
596 BDRVVmdkState *s = bs->opaque;
597 int64_t l1_backup_offset = 0;
600 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
602 error_setg_errno(errp, -ret,
603 "Could not read header from file '%s'",
607 if (header.capacity == 0) {
608 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
610 char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
614 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
620 if (!s->create_type) {
621 s->create_type = g_strdup("monolithicSparse");
624 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
626 * The footer takes precedence over the header, so read it in. The
627 * footer starts at offset -1024 from the end: One sector for the
628 * footer, and another one for the end-of-stream marker.
635 uint8_t pad[512 - 16];
636 } QEMU_PACKED footer_marker;
640 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
646 uint8_t pad[512 - 16];
647 } QEMU_PACKED eos_marker;
648 } QEMU_PACKED footer;
650 ret = bdrv_pread(file,
651 bs->file->bs->total_sectors * 512 - 1536,
652 &footer, sizeof(footer));
654 error_setg_errno(errp, -ret, "Failed to read footer");
658 /* Some sanity checks for the footer */
659 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
660 le32_to_cpu(footer.footer_marker.size) != 0 ||
661 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
662 le64_to_cpu(footer.eos_marker.val) != 0 ||
663 le32_to_cpu(footer.eos_marker.size) != 0 ||
664 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
666 error_setg(errp, "Invalid footer");
670 header = footer.header;
674 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
675 if (le32_to_cpu(header.version) > 3) {
676 error_setg(errp, "Unsupported VMDK version %" PRIu32,
677 le32_to_cpu(header.version));
679 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
681 /* VMware KB 2064959 explains that version 3 added support for
682 * persistent changed block tracking (CBT), and backup software can
683 * read it as version=1 if it doesn't care about the changed area
684 * information. So we are safe to enable read only. */
685 error_setg(errp, "VMDK version 3 must be read only");
689 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
690 error_setg(errp, "L2 table size too big");
694 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
695 * le64_to_cpu(header.granularity);
696 if (l1_entry_sectors == 0) {
697 error_setg(errp, "L1 entry size is invalid");
700 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
702 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
703 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
705 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
706 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
707 (int64_t)(le64_to_cpu(header.grain_offset)
708 * BDRV_SECTOR_SIZE));
712 ret = vmdk_add_extent(bs, file, false,
713 le64_to_cpu(header.capacity),
714 le64_to_cpu(header.gd_offset) << 9,
717 le32_to_cpu(header.num_gtes_per_gt),
718 le64_to_cpu(header.granularity),
725 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
726 if (extent->compressed) {
727 g_free(s->create_type);
728 s->create_type = g_strdup("streamOptimized");
730 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
731 extent->version = le32_to_cpu(header.version);
732 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
733 ret = vmdk_init_tables(bs, extent, errp);
735 /* free extent allocated by vmdk_add_extent */
736 vmdk_free_last_extent(bs);
741 /* find an option value out of descriptor file */
742 static int vmdk_parse_description(const char *desc, const char *opt_name,
743 char *buf, int buf_size)
745 char *opt_pos, *opt_end;
746 const char *end = desc + strlen(desc);
748 opt_pos = strstr(desc, opt_name);
752 /* Skip "=\"" following opt_name */
753 opt_pos += strlen(opt_name) + 2;
754 if (opt_pos >= end) {
758 while (opt_end < end && *opt_end != '"') {
761 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
764 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
768 /* Open an extent file and append to bs array */
769 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
770 char *buf, QDict *options, Error **errp)
774 magic = ldl_be_p(buf);
777 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
780 return vmdk_open_vmdk4(bs, file, flags, options, errp);
783 error_setg(errp, "Image not in VMDK format");
789 static const char *next_line(const char *s)
800 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
801 const char *desc_file_path, QDict *options,
813 BdrvChild *extent_file;
814 BDRVVmdkState *s = bs->opaque;
816 char extent_opt_prefix[32];
817 Error *local_err = NULL;
819 for (p = desc; *p; p = next_line(p)) {
820 /* parse extent line in one of below formats:
822 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
823 * RW [size in sectors] SPARSE "file-name.vmdk"
824 * RW [size in sectors] VMFS "file-name.vmdk"
825 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
828 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
829 access, §ors, type, fname, &flat_offset);
830 if (matches < 4 || strcmp(access, "RW")) {
832 } else if (!strcmp(type, "FLAT")) {
833 if (matches != 5 || flat_offset < 0) {
836 } else if (!strcmp(type, "VMFS")) {
842 } else if (matches != 4) {
847 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
848 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
849 (strcmp(access, "RW"))) {
853 if (!path_is_absolute(fname) && !path_has_protocol(fname) &&
856 error_setg(errp, "Cannot use relative extent paths with VMDK "
857 "descriptor file '%s'", bs->file->bs->filename);
861 extent_path = g_malloc0(PATH_MAX);
862 path_combine(extent_path, PATH_MAX, desc_file_path, fname);
864 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
867 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
868 bs, &child_file, false, &local_err);
871 error_propagate(errp, local_err);
875 /* save to extents array */
876 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
879 ret = vmdk_add_extent(bs, extent_file, true, sectors,
880 0, 0, 0, 0, 0, &extent, errp);
882 bdrv_unref_child(bs, extent_file);
885 extent->flat_start_offset = flat_offset << 9;
886 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
887 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
888 char *buf = vmdk_read_desc(extent_file, 0, errp);
892 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
897 bdrv_unref_child(bs, extent_file);
900 extent = &s->extents[s->num_extents - 1];
902 error_setg(errp, "Unsupported extent type '%s'", type);
903 bdrv_unref_child(bs, extent_file);
906 extent->type = g_strdup(type);
913 if (np[-1] == '\n') {
916 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
920 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
921 QDict *options, Error **errp)
925 BDRVVmdkState *s = bs->opaque;
927 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
928 error_setg(errp, "invalid VMDK image descriptor");
932 if (strcmp(ct, "monolithicFlat") &&
933 strcmp(ct, "vmfs") &&
934 strcmp(ct, "vmfsSparse") &&
935 strcmp(ct, "twoGbMaxExtentSparse") &&
936 strcmp(ct, "twoGbMaxExtentFlat")) {
937 error_setg(errp, "Unsupported image type '%s'", ct);
941 s->create_type = g_strdup(ct);
943 ret = vmdk_parse_extents(buf, bs, bs->file->bs->exact_filename, options,
949 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
954 BDRVVmdkState *s = bs->opaque;
956 Error *local_err = NULL;
958 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
964 buf = vmdk_read_desc(bs->file, 0, errp);
969 magic = ldl_be_p(buf);
973 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
975 s->desc_offset = 0x200;
978 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
985 /* try to open parent images, if exist */
986 ret = vmdk_parent_open(bs);
990 ret = vmdk_read_cid(bs, 0, &s->cid);
994 ret = vmdk_read_cid(bs, 1, &s->parent_cid);
998 qemu_co_mutex_init(&s->lock);
1000 /* Disable migration when VMDK images are used */
1001 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1002 "does not support live migration",
1003 bdrv_get_device_or_node_name(bs));
1004 ret = migrate_add_blocker(s->migration_blocker, &local_err);
1006 error_propagate(errp, local_err);
1007 error_free(s->migration_blocker);
1016 g_free(s->create_type);
1017 s->create_type = NULL;
1018 vmdk_free_extents(bs);
1023 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1025 BDRVVmdkState *s = bs->opaque;
1028 for (i = 0; i < s->num_extents; i++) {
1029 if (!s->extents[i].flat) {
1030 bs->bl.pwrite_zeroes_alignment =
1031 MAX(bs->bl.pwrite_zeroes_alignment,
1032 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1040 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1041 * to the cluster at @cluster_sector_num.
1043 * If @skip_start_sector < @skip_end_sector, the relative range
1044 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1045 * it for call to write user data in the request.
1047 static int get_whole_cluster(BlockDriverState *bs,
1049 uint64_t cluster_offset,
1051 uint64_t skip_start_bytes,
1052 uint64_t skip_end_bytes)
1055 int64_t cluster_bytes;
1056 uint8_t *whole_grain;
1058 /* For COW, align request sector_num to cluster start */
1059 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1060 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1061 whole_grain = qemu_blockalign(bs, cluster_bytes);
1064 memset(whole_grain, 0, skip_start_bytes);
1065 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1068 assert(skip_end_bytes <= cluster_bytes);
1069 /* we will be here if it's first write on non-exist grain(cluster).
1070 * try to read from parent image, if exist */
1071 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1076 /* Read backing data before skip range */
1077 if (skip_start_bytes > 0) {
1079 /* qcow2 emits this on bs->file instead of bs->backing */
1080 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1081 ret = bdrv_pread(bs->backing, offset, whole_grain,
1088 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1089 ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1096 /* Read backing data after skip range */
1097 if (skip_end_bytes < cluster_bytes) {
1099 /* qcow2 emits this on bs->file instead of bs->backing */
1100 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1101 ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1102 whole_grain + skip_end_bytes,
1103 cluster_bytes - skip_end_bytes);
1109 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1110 ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1111 whole_grain + skip_end_bytes,
1112 cluster_bytes - skip_end_bytes);
1121 qemu_vfree(whole_grain);
1125 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1128 offset = cpu_to_le32(offset);
1129 /* update L2 table */
1130 BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1131 if (bdrv_pwrite_sync(extent->file,
1132 ((int64_t)m_data->l2_offset * 512)
1133 + (m_data->l2_index * sizeof(offset)),
1134 &offset, sizeof(offset)) < 0) {
1137 /* update backup L2 table */
1138 if (extent->l1_backup_table_offset != 0) {
1139 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1140 if (bdrv_pwrite_sync(extent->file,
1141 ((int64_t)m_data->l2_offset * 512)
1142 + (m_data->l2_index * sizeof(offset)),
1143 &offset, sizeof(offset)) < 0) {
1147 if (m_data->l2_cache_entry) {
1148 *m_data->l2_cache_entry = offset;
1155 * get_cluster_offset
1157 * Look up cluster offset in extent file by sector number, and store in
1160 * For flat extents, the start offset as parsed from the description file is
1163 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1164 * offset for a new cluster and update L2 cache. If there is a backing file,
1165 * COW is done before returning; otherwise, zeroes are written to the allocated
1166 * cluster. Both COW and zero writing skips the sector range
1167 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1168 * has new data to write there.
1170 * Returns: VMDK_OK if cluster exists and mapped in the image.
1171 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1172 * VMDK_ERROR if failed.
1174 static int get_cluster_offset(BlockDriverState *bs,
1176 VmdkMetaData *m_data,
1179 uint64_t *cluster_offset,
1180 uint64_t skip_start_bytes,
1181 uint64_t skip_end_bytes)
1183 unsigned int l1_index, l2_offset, l2_index;
1184 int min_index, i, j;
1185 uint32_t min_count, *l2_table;
1186 bool zeroed = false;
1188 int64_t cluster_sector;
1194 *cluster_offset = extent->flat_start_offset;
1198 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1199 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1200 if (l1_index >= extent->l1_size) {
1203 l2_offset = extent->l1_table[l1_index];
1205 return VMDK_UNALLOC;
1207 for (i = 0; i < L2_CACHE_SIZE; i++) {
1208 if (l2_offset == extent->l2_cache_offsets[i]) {
1209 /* increment the hit count */
1210 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1211 for (j = 0; j < L2_CACHE_SIZE; j++) {
1212 extent->l2_cache_counts[j] >>= 1;
1215 l2_table = extent->l2_cache + (i * extent->l2_size);
1219 /* not found: load a new entry in the least used one */
1221 min_count = 0xffffffff;
1222 for (i = 0; i < L2_CACHE_SIZE; i++) {
1223 if (extent->l2_cache_counts[i] < min_count) {
1224 min_count = extent->l2_cache_counts[i];
1228 l2_table = extent->l2_cache + (min_index * extent->l2_size);
1229 BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1230 if (bdrv_pread(extent->file,
1231 (int64_t)l2_offset * 512,
1233 extent->l2_size * sizeof(uint32_t)
1234 ) != extent->l2_size * sizeof(uint32_t)) {
1238 extent->l2_cache_offsets[min_index] = l2_offset;
1239 extent->l2_cache_counts[min_index] = 1;
1241 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1242 cluster_sector = le32_to_cpu(l2_table[l2_index]);
1244 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1248 if (!cluster_sector || zeroed) {
1250 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1253 cluster_sector = extent->next_cluster_sector;
1254 extent->next_cluster_sector += extent->cluster_sectors;
1256 /* First of all we write grain itself, to avoid race condition
1257 * that may to corrupt the image.
1258 * This problem may occur because of insufficient space on host disk
1259 * or inappropriate VM shutdown.
1261 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1262 offset, skip_start_bytes, skip_end_bytes);
1268 m_data->l1_index = l1_index;
1269 m_data->l2_index = l2_index;
1270 m_data->l2_offset = l2_offset;
1271 m_data->l2_cache_entry = &l2_table[l2_index];
1274 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1278 static VmdkExtent *find_extent(BDRVVmdkState *s,
1279 int64_t sector_num, VmdkExtent *start_hint)
1281 VmdkExtent *extent = start_hint;
1284 extent = &s->extents[0];
1286 while (extent < &s->extents[s->num_extents]) {
1287 if (sector_num < extent->end_sector) {
1295 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1298 uint64_t extent_begin_offset, extent_relative_offset;
1299 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1301 extent_begin_offset =
1302 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1303 extent_relative_offset = offset - extent_begin_offset;
1304 return extent_relative_offset % cluster_size;
1307 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent *extent,
1311 offset = vmdk_find_offset_in_cluster(extent, sector_num * BDRV_SECTOR_SIZE);
1312 return offset / BDRV_SECTOR_SIZE;
1315 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs,
1316 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1318 BDRVVmdkState *s = bs->opaque;
1319 int64_t index_in_cluster, n, ret;
1323 extent = find_extent(s, sector_num, NULL);
1327 qemu_co_mutex_lock(&s->lock);
1328 ret = get_cluster_offset(bs, extent, NULL,
1329 sector_num * 512, false, &offset,
1331 qemu_co_mutex_unlock(&s->lock);
1333 index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num);
1342 ret = BDRV_BLOCK_ZERO;
1345 ret = BDRV_BLOCK_DATA;
1346 if (!extent->compressed) {
1347 ret |= BDRV_BLOCK_OFFSET_VALID;
1348 ret |= (offset + (index_in_cluster << BDRV_SECTOR_BITS))
1349 & BDRV_BLOCK_OFFSET_MASK;
1351 *file = extent->file->bs;
1355 n = extent->cluster_sectors - index_in_cluster;
1356 if (n > nb_sectors) {
1363 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1364 int64_t offset_in_cluster, QEMUIOVector *qiov,
1365 uint64_t qiov_offset, uint64_t n_bytes,
1369 VmdkGrainMarker *data = NULL;
1371 QEMUIOVector local_qiov;
1373 int64_t write_offset;
1374 int64_t write_end_sector;
1376 if (extent->compressed) {
1377 void *compressed_data;
1379 if (!extent->has_marker) {
1383 buf_len = (extent->cluster_sectors << 9) * 2;
1384 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1386 compressed_data = g_malloc(n_bytes);
1387 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1388 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1389 g_free(compressed_data);
1391 if (ret != Z_OK || buf_len == 0) {
1396 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1397 data->size = cpu_to_le32(buf_len);
1399 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1400 iov = (struct iovec) {
1404 qemu_iovec_init_external(&local_qiov, &iov, 1);
1406 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1408 qemu_iovec_init(&local_qiov, qiov->niov);
1409 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1411 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1414 write_offset = cluster_offset + offset_in_cluster;
1415 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1418 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1420 if (extent->compressed) {
1421 extent->next_cluster_sector = write_end_sector;
1423 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1433 if (!extent->compressed) {
1434 qemu_iovec_destroy(&local_qiov);
1439 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1440 int64_t offset_in_cluster, QEMUIOVector *qiov,
1444 int cluster_bytes, buf_bytes;
1445 uint8_t *cluster_buf, *compressed_data;
1446 uint8_t *uncomp_buf;
1448 VmdkGrainMarker *marker;
1452 if (!extent->compressed) {
1453 BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1454 ret = bdrv_co_preadv(extent->file,
1455 cluster_offset + offset_in_cluster, bytes,
1462 cluster_bytes = extent->cluster_sectors * 512;
1463 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1464 buf_bytes = cluster_bytes * 2;
1465 cluster_buf = g_malloc(buf_bytes);
1466 uncomp_buf = g_malloc(cluster_bytes);
1467 BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1468 ret = bdrv_pread(extent->file,
1470 cluster_buf, buf_bytes);
1474 compressed_data = cluster_buf;
1475 buf_len = cluster_bytes;
1476 data_len = cluster_bytes;
1477 if (extent->has_marker) {
1478 marker = (VmdkGrainMarker *)cluster_buf;
1479 compressed_data = marker->data;
1480 data_len = le32_to_cpu(marker->size);
1482 if (!data_len || data_len > buf_bytes) {
1486 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1492 if (offset_in_cluster < 0 ||
1493 offset_in_cluster + bytes > buf_len) {
1497 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1502 g_free(cluster_buf);
1506 static int coroutine_fn
1507 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1508 QEMUIOVector *qiov, int flags)
1510 BDRVVmdkState *s = bs->opaque;
1512 uint64_t n_bytes, offset_in_cluster;
1513 VmdkExtent *extent = NULL;
1514 QEMUIOVector local_qiov;
1515 uint64_t cluster_offset;
1516 uint64_t bytes_done = 0;
1518 qemu_iovec_init(&local_qiov, qiov->niov);
1519 qemu_co_mutex_lock(&s->lock);
1522 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1527 ret = get_cluster_offset(bs, extent, NULL,
1528 offset, false, &cluster_offset, 0, 0);
1529 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1531 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1532 - offset_in_cluster);
1534 if (ret != VMDK_OK) {
1535 /* if not allocated, try to read from parent image, if exist */
1536 if (bs->backing && ret != VMDK_ZEROED) {
1537 if (!vmdk_is_cid_valid(bs)) {
1542 qemu_iovec_reset(&local_qiov);
1543 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1545 /* qcow2 emits this on bs->file instead of bs->backing */
1546 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1547 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1553 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1556 qemu_iovec_reset(&local_qiov);
1557 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1559 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1560 &local_qiov, n_bytes);
1567 bytes_done += n_bytes;
1572 qemu_co_mutex_unlock(&s->lock);
1573 qemu_iovec_destroy(&local_qiov);
1580 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1581 * if possible, otherwise return -ENOTSUP.
1582 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1583 * with each cluster. By dry run we can find if the zero write
1584 * is possible without modifying image data.
1586 * Returns: error code with 0 for success.
1588 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1589 uint64_t bytes, QEMUIOVector *qiov,
1590 bool zeroed, bool zero_dry_run)
1592 BDRVVmdkState *s = bs->opaque;
1593 VmdkExtent *extent = NULL;
1595 int64_t offset_in_cluster, n_bytes;
1596 uint64_t cluster_offset;
1597 uint64_t bytes_done = 0;
1598 VmdkMetaData m_data;
1600 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1601 error_report("Wrong offset: offset=0x%" PRIx64
1602 " total_sectors=0x%" PRIx64,
1603 offset, bs->total_sectors);
1608 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1612 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1613 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1614 - offset_in_cluster);
1616 ret = get_cluster_offset(bs, extent, &m_data, offset,
1617 !(extent->compressed || zeroed),
1618 &cluster_offset, offset_in_cluster,
1619 offset_in_cluster + n_bytes);
1620 if (extent->compressed) {
1621 if (ret == VMDK_OK) {
1622 /* Refuse write to allocated cluster for streamOptimized */
1623 error_report("Could not write to allocated cluster"
1624 " for streamOptimized");
1628 ret = get_cluster_offset(bs, extent, &m_data, offset,
1629 true, &cluster_offset, 0, 0);
1632 if (ret == VMDK_ERROR) {
1636 /* Do zeroed write, buf is ignored */
1637 if (extent->has_zero_grain &&
1638 offset_in_cluster == 0 &&
1639 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
1640 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1641 if (!zero_dry_run) {
1642 /* update L2 tables */
1643 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1652 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
1653 qiov, bytes_done, n_bytes, offset);
1658 /* update L2 tables */
1659 if (vmdk_L2update(extent, &m_data,
1660 cluster_offset >> BDRV_SECTOR_BITS)
1668 bytes_done += n_bytes;
1670 /* update CID on the first write every time the virtual disk is
1672 if (!s->cid_updated) {
1673 ret = vmdk_write_cid(bs, g_random_int());
1677 s->cid_updated = true;
1683 static int coroutine_fn
1684 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1685 QEMUIOVector *qiov, int flags)
1688 BDRVVmdkState *s = bs->opaque;
1689 qemu_co_mutex_lock(&s->lock);
1690 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
1691 qemu_co_mutex_unlock(&s->lock);
1695 static int coroutine_fn
1696 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1697 uint64_t bytes, QEMUIOVector *qiov)
1699 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
1702 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
1705 BdrvRequestFlags flags)
1708 BDRVVmdkState *s = bs->opaque;
1710 qemu_co_mutex_lock(&s->lock);
1711 /* write zeroes could fail if sectors not aligned to cluster, test it with
1712 * dry_run == true before really updating image */
1713 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
1715 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
1717 qemu_co_mutex_unlock(&s->lock);
1721 static int vmdk_create_extent(const char *filename, int64_t filesize,
1722 bool flat, bool compress, bool zeroed_grain,
1723 QemuOpts *opts, Error **errp)
1726 BlockBackend *blk = NULL;
1728 Error *local_err = NULL;
1729 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1730 uint32_t *gd_buf = NULL;
1733 ret = bdrv_create_file(filename, opts, &local_err);
1735 error_propagate(errp, local_err);
1739 blk = blk_new_open(filename, NULL, NULL,
1740 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
1743 error_propagate(errp, local_err);
1748 blk_set_allow_write_beyond_eof(blk, true);
1751 ret = blk_truncate(blk, filesize, PREALLOC_MODE_OFF, errp);
1754 magic = cpu_to_be32(VMDK4_MAGIC);
1755 memset(&header, 0, sizeof(header));
1758 } else if (zeroed_grain) {
1763 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1764 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1765 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1766 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1767 header.capacity = filesize / BDRV_SECTOR_SIZE;
1768 header.granularity = 128;
1769 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1771 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1772 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1774 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1775 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1777 header.desc_offset = 1;
1778 header.desc_size = 20;
1779 header.rgd_offset = header.desc_offset + header.desc_size;
1780 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1781 header.grain_offset =
1782 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1783 header.granularity);
1784 /* swap endianness for all header fields */
1785 header.version = cpu_to_le32(header.version);
1786 header.flags = cpu_to_le32(header.flags);
1787 header.capacity = cpu_to_le64(header.capacity);
1788 header.granularity = cpu_to_le64(header.granularity);
1789 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1790 header.desc_offset = cpu_to_le64(header.desc_offset);
1791 header.desc_size = cpu_to_le64(header.desc_size);
1792 header.rgd_offset = cpu_to_le64(header.rgd_offset);
1793 header.gd_offset = cpu_to_le64(header.gd_offset);
1794 header.grain_offset = cpu_to_le64(header.grain_offset);
1795 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1797 header.check_bytes[0] = 0xa;
1798 header.check_bytes[1] = 0x20;
1799 header.check_bytes[2] = 0xd;
1800 header.check_bytes[3] = 0xa;
1802 /* write all the data */
1803 ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
1805 error_setg(errp, QERR_IO_ERROR);
1808 ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
1810 error_setg(errp, QERR_IO_ERROR);
1814 ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9,
1815 PREALLOC_MODE_OFF, errp);
1820 /* write grain directory */
1821 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1822 gd_buf = g_malloc0(gd_buf_size);
1823 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1824 i < gt_count; i++, tmp += gt_size) {
1825 gd_buf[i] = cpu_to_le32(tmp);
1827 ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1828 gd_buf, gd_buf_size, 0);
1830 error_setg(errp, QERR_IO_ERROR);
1834 /* write backup grain directory */
1835 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1836 i < gt_count; i++, tmp += gt_size) {
1837 gd_buf[i] = cpu_to_le32(tmp);
1839 ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1840 gd_buf, gd_buf_size, 0);
1842 error_setg(errp, QERR_IO_ERROR);
1855 static int filename_decompose(const char *filename, char *path, char *prefix,
1856 char *postfix, size_t buf_len, Error **errp)
1860 if (filename == NULL || !strlen(filename)) {
1861 error_setg(errp, "No filename provided");
1864 p = strrchr(filename, '/');
1866 p = strrchr(filename, '\\');
1869 p = strrchr(filename, ':');
1873 if (p - filename >= buf_len) {
1876 pstrcpy(path, p - filename + 1, filename);
1881 q = strrchr(p, '.');
1883 pstrcpy(prefix, buf_len, p);
1886 if (q - p >= buf_len) {
1889 pstrcpy(prefix, q - p + 1, p);
1890 pstrcpy(postfix, buf_len, q);
1895 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp)
1898 BlockBackend *new_blk = NULL;
1899 Error *local_err = NULL;
1901 int64_t total_size = 0, filesize;
1902 char *adapter_type = NULL;
1903 char *backing_file = NULL;
1904 char *hw_version = NULL;
1907 bool flat, split, compress;
1908 GString *ext_desc_lines;
1909 char *path = g_malloc0(PATH_MAX);
1910 char *prefix = g_malloc0(PATH_MAX);
1911 char *postfix = g_malloc0(PATH_MAX);
1912 char *desc_line = g_malloc0(BUF_SIZE);
1913 char *ext_filename = g_malloc0(PATH_MAX);
1914 char *desc_filename = g_malloc0(PATH_MAX);
1915 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
1916 const char *desc_extent_line;
1917 char *parent_desc_line = g_malloc0(BUF_SIZE);
1918 uint32_t parent_cid = 0xffffffff;
1919 uint32_t number_heads = 16;
1920 bool zeroed_grain = false;
1921 uint32_t desc_offset = 0, desc_len;
1922 const char desc_template[] =
1923 "# Disk DescriptorFile\n"
1926 "parentCID=%" PRIx32 "\n"
1927 "createType=\"%s\"\n"
1930 "# Extent description\n"
1933 "# The Disk Data Base\n"
1936 "ddb.virtualHWVersion = \"%s\"\n"
1937 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1938 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
1939 "ddb.geometry.sectors = \"63\"\n"
1940 "ddb.adapterType = \"%s\"\n";
1942 ext_desc_lines = g_string_new(NULL);
1944 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
1948 /* Read out options */
1949 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1951 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
1952 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1953 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
1954 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) {
1955 if (strcmp(hw_version, "undefined")) {
1957 "compat6 cannot be enabled with hwversion set");
1962 hw_version = g_strdup("6");
1964 if (strcmp(hw_version, "undefined") == 0) {
1966 hw_version = g_strdup("4");
1968 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1969 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) {
1970 zeroed_grain = true;
1973 if (!adapter_type) {
1974 adapter_type = g_strdup("ide");
1975 } else if (strcmp(adapter_type, "ide") &&
1976 strcmp(adapter_type, "buslogic") &&
1977 strcmp(adapter_type, "lsilogic") &&
1978 strcmp(adapter_type, "legacyESX")) {
1979 error_setg(errp, "Unknown adapter type: '%s'", adapter_type);
1983 if (strcmp(adapter_type, "ide") != 0) {
1984 /* that's the number of heads with which vmware operates when
1985 creating, exporting, etc. vmdk files with a non-ide adapter type */
1989 /* Default format to monolithicSparse */
1990 fmt = g_strdup("monolithicSparse");
1991 } else if (strcmp(fmt, "monolithicFlat") &&
1992 strcmp(fmt, "monolithicSparse") &&
1993 strcmp(fmt, "twoGbMaxExtentSparse") &&
1994 strcmp(fmt, "twoGbMaxExtentFlat") &&
1995 strcmp(fmt, "streamOptimized")) {
1996 error_setg(errp, "Unknown subformat: '%s'", fmt);
2000 split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
2001 strcmp(fmt, "twoGbMaxExtentSparse"));
2002 flat = !(strcmp(fmt, "monolithicFlat") &&
2003 strcmp(fmt, "twoGbMaxExtentFlat"));
2004 compress = !strcmp(fmt, "streamOptimized");
2006 desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n";
2008 desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n";
2010 if (flat && backing_file) {
2011 error_setg(errp, "Flat image can't have backing file");
2015 if (flat && zeroed_grain) {
2016 error_setg(errp, "Flat image can't enable zeroed grain");
2022 char *full_backing = g_new0(char, PATH_MAX);
2023 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
2024 full_backing, PATH_MAX,
2027 g_free(full_backing);
2028 error_propagate(errp, local_err);
2033 blk = blk_new_open(full_backing, NULL, NULL,
2034 BDRV_O_NO_BACKING, errp);
2035 g_free(full_backing);
2040 if (strcmp(blk_bs(blk)->drv->format_name, "vmdk")) {
2045 ret = vmdk_read_cid(blk_bs(blk), 0, &parent_cid);
2050 snprintf(parent_desc_line, BUF_SIZE,
2051 "parentFileNameHint=\"%s\"", backing_file);
2054 /* Create extents */
2055 filesize = total_size;
2056 while (filesize > 0) {
2057 int64_t size = filesize;
2059 if (split && size > split_size) {
2063 snprintf(desc_filename, PATH_MAX, "%s-%c%03d%s",
2064 prefix, flat ? 'f' : 's', ++idx, postfix);
2066 snprintf(desc_filename, PATH_MAX, "%s-flat%s", prefix, postfix);
2068 snprintf(desc_filename, PATH_MAX, "%s%s", prefix, postfix);
2070 snprintf(ext_filename, PATH_MAX, "%s%s", path, desc_filename);
2072 if (vmdk_create_extent(ext_filename, size,
2073 flat, compress, zeroed_grain, opts, errp)) {
2079 /* Format description line */
2080 snprintf(desc_line, BUF_SIZE,
2081 desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename);
2082 g_string_append(ext_desc_lines, desc_line);
2084 /* generate descriptor file */
2085 desc = g_strdup_printf(desc_template,
2090 ext_desc_lines->str,
2093 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2096 desc_len = strlen(desc);
2097 /* the descriptor offset = 0x200 */
2098 if (!split && !flat) {
2099 desc_offset = 0x200;
2101 ret = bdrv_create_file(filename, opts, &local_err);
2103 error_propagate(errp, local_err);
2108 new_blk = blk_new_open(filename, NULL, NULL,
2109 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2111 if (new_blk == NULL) {
2112 error_propagate(errp, local_err);
2117 blk_set_allow_write_beyond_eof(new_blk, true);
2119 ret = blk_pwrite(new_blk, desc_offset, desc, desc_len, 0);
2121 error_setg_errno(errp, -ret, "Could not write description");
2124 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2125 * for description file */
2126 if (desc_offset == 0) {
2127 ret = blk_truncate(new_blk, desc_len, PREALLOC_MODE_OFF, errp);
2133 g_free(adapter_type);
2134 g_free(backing_file);
2142 g_free(ext_filename);
2143 g_free(desc_filename);
2144 g_free(parent_desc_line);
2145 g_string_free(ext_desc_lines, true);
2149 static void vmdk_close(BlockDriverState *bs)
2151 BDRVVmdkState *s = bs->opaque;
2153 vmdk_free_extents(bs);
2154 g_free(s->create_type);
2156 migrate_del_blocker(s->migration_blocker);
2157 error_free(s->migration_blocker);
2160 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2162 BDRVVmdkState *s = bs->opaque;
2166 for (i = 0; i < s->num_extents; i++) {
2167 err = bdrv_co_flush(s->extents[i].file->bs);
2175 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2180 BDRVVmdkState *s = bs->opaque;
2182 ret = bdrv_get_allocated_file_size(bs->file->bs);
2186 for (i = 0; i < s->num_extents; i++) {
2187 if (s->extents[i].file == bs->file) {
2190 r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2199 static int vmdk_has_zero_init(BlockDriverState *bs)
2202 BDRVVmdkState *s = bs->opaque;
2204 /* If has a flat extent and its underlying storage doesn't have zero init,
2206 for (i = 0; i < s->num_extents; i++) {
2207 if (s->extents[i].flat) {
2208 if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2216 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2218 ImageInfo *info = g_new0(ImageInfo, 1);
2220 *info = (ImageInfo){
2221 .filename = g_strdup(extent->file->bs->filename),
2222 .format = g_strdup(extent->type),
2223 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2224 .compressed = extent->compressed,
2225 .has_compressed = extent->compressed,
2226 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2227 .has_cluster_size = !extent->flat,
2233 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result,
2236 BDRVVmdkState *s = bs->opaque;
2237 VmdkExtent *extent = NULL;
2238 int64_t sector_num = 0;
2239 int64_t total_sectors = bdrv_nb_sectors(bs);
2241 uint64_t cluster_offset;
2248 if (sector_num >= total_sectors) {
2251 extent = find_extent(s, sector_num, extent);
2254 "ERROR: could not find extent for sector %" PRId64 "\n",
2259 ret = get_cluster_offset(bs, extent, NULL,
2260 sector_num << BDRV_SECTOR_BITS,
2261 false, &cluster_offset, 0, 0);
2262 if (ret == VMDK_ERROR) {
2264 "ERROR: could not get cluster_offset for sector %"
2265 PRId64 "\n", sector_num);
2268 if (ret == VMDK_OK) {
2269 int64_t extent_len = bdrv_getlength(extent->file->bs);
2270 if (extent_len < 0) {
2272 "ERROR: could not get extent file length for sector %"
2273 PRId64 "\n", sector_num);
2277 if (cluster_offset >= extent_len) {
2279 "ERROR: cluster offset for sector %"
2280 PRId64 " points after EOF\n", sector_num);
2285 sector_num += extent->cluster_sectors;
2288 result->corruptions++;
2292 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs)
2295 BDRVVmdkState *s = bs->opaque;
2296 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2297 ImageInfoList **next;
2299 *spec_info = (ImageInfoSpecific){
2300 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2302 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2306 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2307 .create_type = g_strdup(s->create_type),
2309 .parent_cid = s->parent_cid,
2312 next = &spec_info->u.vmdk.data->extents;
2313 for (i = 0; i < s->num_extents; i++) {
2314 *next = g_new0(ImageInfoList, 1);
2315 (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2316 (*next)->next = NULL;
2317 next = &(*next)->next;
2323 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2325 return a->flat == b->flat &&
2326 a->compressed == b->compressed &&
2327 (a->flat || a->cluster_sectors == b->cluster_sectors);
2330 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2333 BDRVVmdkState *s = bs->opaque;
2334 assert(s->num_extents);
2336 /* See if we have multiple extents but they have different cases */
2337 for (i = 1; i < s->num_extents; i++) {
2338 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2342 bdi->needs_compressed_writes = s->extents[0].compressed;
2343 if (!s->extents[0].flat) {
2344 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2349 static QemuOptsList vmdk_create_opts = {
2350 .name = "vmdk-create-opts",
2351 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2354 .name = BLOCK_OPT_SIZE,
2355 .type = QEMU_OPT_SIZE,
2356 .help = "Virtual disk size"
2359 .name = BLOCK_OPT_ADAPTER_TYPE,
2360 .type = QEMU_OPT_STRING,
2361 .help = "Virtual adapter type, can be one of "
2362 "ide (default), lsilogic, buslogic or legacyESX"
2365 .name = BLOCK_OPT_BACKING_FILE,
2366 .type = QEMU_OPT_STRING,
2367 .help = "File name of a base image"
2370 .name = BLOCK_OPT_COMPAT6,
2371 .type = QEMU_OPT_BOOL,
2372 .help = "VMDK version 6 image",
2373 .def_value_str = "off"
2376 .name = BLOCK_OPT_HWVERSION,
2377 .type = QEMU_OPT_STRING,
2378 .help = "VMDK hardware version",
2379 .def_value_str = "undefined"
2382 .name = BLOCK_OPT_SUBFMT,
2383 .type = QEMU_OPT_STRING,
2385 "VMDK flat extent format, can be one of "
2386 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2389 .name = BLOCK_OPT_ZEROED_GRAIN,
2390 .type = QEMU_OPT_BOOL,
2391 .help = "Enable efficient zero writes "
2392 "using the zeroed-grain GTE feature"
2394 { /* end of list */ }
2398 static BlockDriver bdrv_vmdk = {
2399 .format_name = "vmdk",
2400 .instance_size = sizeof(BDRVVmdkState),
2401 .bdrv_probe = vmdk_probe,
2402 .bdrv_open = vmdk_open,
2403 .bdrv_check = vmdk_check,
2404 .bdrv_reopen_prepare = vmdk_reopen_prepare,
2405 .bdrv_child_perm = bdrv_format_default_perms,
2406 .bdrv_co_preadv = vmdk_co_preadv,
2407 .bdrv_co_pwritev = vmdk_co_pwritev,
2408 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
2409 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
2410 .bdrv_close = vmdk_close,
2411 .bdrv_create = vmdk_create,
2412 .bdrv_co_flush_to_disk = vmdk_co_flush,
2413 .bdrv_co_get_block_status = vmdk_co_get_block_status,
2414 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2415 .bdrv_has_zero_init = vmdk_has_zero_init,
2416 .bdrv_get_specific_info = vmdk_get_specific_info,
2417 .bdrv_refresh_limits = vmdk_refresh_limits,
2418 .bdrv_get_info = vmdk_get_info,
2420 .supports_backing = true,
2421 .create_opts = &vmdk_create_opts,
2424 static void bdrv_vmdk_init(void)
2426 bdrv_register(&bdrv_vmdk);
2429 block_init(bdrv_vmdk_init);