2 * Block driver for the VMDK format
4 * Copyright (c) 2004 Fabrice Bellard
5 * Copyright (c) 2005 Filip Navara
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "block/block_int.h"
29 #include "sysemu/block-backend.h"
30 #include "qapi/qmp/qdict.h"
31 #include "qapi/qmp/qerror.h"
32 #include "qemu/error-report.h"
33 #include "qemu/module.h"
34 #include "qemu/option.h"
35 #include "qemu/bswap.h"
36 #include "qemu/memalign.h"
37 #include "migration/blocker.h"
38 #include "qemu/cutils.h"
41 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
42 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
43 #define VMDK4_COMPRESSION_DEFLATE 1
44 #define VMDK4_FLAG_NL_DETECT (1 << 0)
45 #define VMDK4_FLAG_RGD (1 << 1)
46 /* Zeroed-grain enable bit */
47 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
48 #define VMDK4_FLAG_COMPRESS (1 << 16)
49 #define VMDK4_FLAG_MARKER (1 << 17)
50 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
52 #define VMDK_EXTENT_MAX_SECTORS (1ULL << 32)
54 #define VMDK_GTE_ZEROED 0x1
56 /* VMDK internal error codes */
58 #define VMDK_ERROR (-1)
59 /* Cluster not allocated */
60 #define VMDK_UNALLOC (-2)
61 #define VMDK_ZEROED (-3)
63 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
64 #define BLOCK_OPT_TOOLSVERSION "toolsversion"
69 uint32_t disk_sectors;
71 uint32_t l1dir_offset;
73 uint32_t file_sectors;
76 uint32_t sectors_per_track;
77 } QEMU_PACKED VMDK3Header;
86 /* Number of GrainTableEntries per GrainTable */
87 uint32_t num_gtes_per_gt;
90 uint64_t grain_offset;
93 uint16_t compressAlgorithm;
94 } QEMU_PACKED VMDK4Header;
96 typedef struct VMDKSESparseConstHeader {
101 uint64_t grain_table_size;
107 uint64_t volatile_header_offset;
108 uint64_t volatile_header_size;
109 uint64_t journal_header_offset;
110 uint64_t journal_header_size;
111 uint64_t journal_offset;
112 uint64_t journal_size;
113 uint64_t grain_dir_offset;
114 uint64_t grain_dir_size;
115 uint64_t grain_tables_offset;
116 uint64_t grain_tables_size;
117 uint64_t free_bitmap_offset;
118 uint64_t free_bitmap_size;
119 uint64_t backmap_offset;
120 uint64_t backmap_size;
121 uint64_t grains_offset;
122 uint64_t grains_size;
124 } QEMU_PACKED VMDKSESparseConstHeader;
126 typedef struct VMDKSESparseVolatileHeader {
128 uint64_t free_gt_number;
129 uint64_t next_txn_seq_number;
130 uint64_t replay_journal;
132 } QEMU_PACKED VMDKSESparseVolatileHeader;
134 #define L2_CACHE_SIZE 16
136 typedef struct VmdkExtent {
143 uint64_t sesparse_l2_tables_offset;
144 uint64_t sesparse_clusters_offset;
149 int64_t flat_start_offset;
150 int64_t l1_table_offset;
151 int64_t l1_backup_table_offset;
153 uint32_t *l1_backup_table;
154 unsigned int l1_size;
155 uint32_t l1_entry_sectors;
157 unsigned int l2_size;
159 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
160 uint32_t l2_cache_counts[L2_CACHE_SIZE];
162 int64_t cluster_sectors;
163 int64_t next_cluster_sector;
167 typedef struct BDRVVmdkState {
169 uint64_t desc_offset;
175 /* Extent array with num_extents entries, ascend ordered by address */
177 Error *migration_blocker;
181 typedef struct BDRVVmdkReopenState {
182 bool *extents_using_bs_file;
183 } BDRVVmdkReopenState;
185 typedef struct VmdkMetaData {
186 unsigned int l1_index;
187 unsigned int l2_index;
188 unsigned int l2_offset;
190 uint32_t *l2_cache_entry;
193 typedef struct VmdkGrainMarker {
197 } QEMU_PACKED VmdkGrainMarker;
200 MARKER_END_OF_STREAM = 0,
201 MARKER_GRAIN_TABLE = 1,
202 MARKER_GRAIN_DIRECTORY = 2,
206 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
213 magic = be32_to_cpu(*(uint32_t *)buf);
214 if (magic == VMDK3_MAGIC ||
215 magic == VMDK4_MAGIC) {
218 const char *p = (const char *)buf;
219 const char *end = p + buf_size;
222 /* skip comment line */
223 while (p < end && *p != '\n') {
230 while (p < end && *p == ' ') {
233 /* skip '\r' if windows line endings used. */
234 if (p < end && *p == '\r') {
237 /* only accept blank lines before 'version=' line */
238 if (p == end || *p != '\n') {
244 if (end - p >= strlen("version=X\n")) {
245 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
246 strncmp("version=2\n", p, strlen("version=2\n")) == 0 ||
247 strncmp("version=3\n", p, strlen("version=3\n")) == 0) {
251 if (end - p >= strlen("version=X\r\n")) {
252 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
253 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0 ||
254 strncmp("version=3\r\n", p, strlen("version=3\r\n")) == 0) {
264 #define SECTOR_SIZE 512
265 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
266 #define BUF_SIZE 4096
267 #define HEADER_SIZE 512 /* first sector of 512 bytes */
269 static void vmdk_free_extents(BlockDriverState *bs)
272 BDRVVmdkState *s = bs->opaque;
275 for (i = 0; i < s->num_extents; i++) {
279 g_free(e->l1_backup_table);
281 if (e->file != bs->file) {
282 bdrv_unref_child(bs, e->file);
288 static void vmdk_free_last_extent(BlockDriverState *bs)
290 BDRVVmdkState *s = bs->opaque;
292 if (s->num_extents == 0) {
296 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
299 /* Return -ve errno, or 0 on success and write CID into *pcid. */
300 static int vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
304 const char *p_name, *cid_str;
306 BDRVVmdkState *s = bs->opaque;
309 desc = g_malloc0(DESC_SIZE);
310 ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
316 cid_str = "parentCID";
317 cid_str_size = sizeof("parentCID");
320 cid_str_size = sizeof("CID");
323 desc[DESC_SIZE - 1] = '\0';
324 p_name = strstr(desc, cid_str);
325 if (p_name == NULL) {
329 p_name += cid_str_size;
330 if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
342 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
344 char *desc, *tmp_desc;
345 char *p_name, *tmp_str;
346 BDRVVmdkState *s = bs->opaque;
349 desc = g_malloc0(DESC_SIZE);
350 tmp_desc = g_malloc0(DESC_SIZE);
351 ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
356 desc[DESC_SIZE - 1] = '\0';
357 tmp_str = strstr(desc, "parentCID");
358 if (tmp_str == NULL) {
363 pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
364 p_name = strstr(desc, "CID");
365 if (p_name != NULL) {
366 p_name += sizeof("CID");
367 snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
368 pstrcat(desc, DESC_SIZE, tmp_desc);
371 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
379 static int coroutine_fn vmdk_is_cid_valid(BlockDriverState *bs)
381 BDRVVmdkState *s = bs->opaque;
384 if (!s->cid_checked && bs->backing) {
385 BlockDriverState *p_bs = bs->backing->bs;
387 if (strcmp(p_bs->drv->format_name, "vmdk")) {
388 /* Backing file is not in vmdk format, so it does not have
389 * a CID, which makes the overlay's parent CID invalid */
393 if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
394 /* read failure: report as not valid */
397 if (s->parent_cid != cur_pcid) {
402 s->cid_checked = true;
407 static int vmdk_reopen_prepare(BDRVReopenState *state,
408 BlockReopenQueue *queue, Error **errp)
411 BDRVVmdkReopenState *rs;
414 assert(state != NULL);
415 assert(state->bs != NULL);
416 assert(state->opaque == NULL);
418 s = state->bs->opaque;
420 rs = g_new0(BDRVVmdkReopenState, 1);
424 * Check whether there are any extents stored in bs->file; if bs->file
425 * changes, we will need to update their .file pointers to follow suit
427 rs->extents_using_bs_file = g_new(bool, s->num_extents);
428 for (i = 0; i < s->num_extents; i++) {
429 rs->extents_using_bs_file[i] = s->extents[i].file == state->bs->file;
435 static void vmdk_reopen_clean(BDRVReopenState *state)
437 BDRVVmdkReopenState *rs = state->opaque;
439 g_free(rs->extents_using_bs_file);
441 state->opaque = NULL;
444 static void vmdk_reopen_commit(BDRVReopenState *state)
446 BDRVVmdkState *s = state->bs->opaque;
447 BDRVVmdkReopenState *rs = state->opaque;
450 for (i = 0; i < s->num_extents; i++) {
451 if (rs->extents_using_bs_file[i]) {
452 s->extents[i].file = state->bs->file;
456 vmdk_reopen_clean(state);
459 static void vmdk_reopen_abort(BDRVReopenState *state)
461 vmdk_reopen_clean(state);
464 static int vmdk_parent_open(BlockDriverState *bs)
468 BDRVVmdkState *s = bs->opaque;
471 desc = g_malloc0(DESC_SIZE + 1);
472 ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
477 p_name = strstr(desc, "parentFileNameHint");
478 if (p_name != NULL) {
481 p_name += sizeof("parentFileNameHint") + 1;
482 end_name = strchr(p_name, '\"');
483 if (end_name == NULL) {
487 if ((end_name - p_name) > sizeof(bs->auto_backing_file) - 1) {
492 pstrcpy(bs->auto_backing_file, end_name - p_name + 1, p_name);
493 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
494 bs->auto_backing_file);
495 pstrcpy(bs->backing_format, sizeof(bs->backing_format),
504 /* Create and append extent to the extent array. Return the added VmdkExtent
505 * address. return NULL if allocation failed. */
506 static int vmdk_add_extent(BlockDriverState *bs,
507 BdrvChild *file, bool flat, int64_t sectors,
508 int64_t l1_offset, int64_t l1_backup_offset,
510 int l2_size, uint64_t cluster_sectors,
511 VmdkExtent **new_extent,
515 BDRVVmdkState *s = bs->opaque;
518 if (cluster_sectors > 0x200000) {
519 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
520 error_setg(errp, "Invalid granularity, image may be corrupt");
523 if (l1_size > 32 * 1024 * 1024) {
525 * Although with big capacity and small l1_entry_sectors, we can get a
526 * big l1_size, we don't want unbounded value to allocate the table.
527 * Limit it to 32M, which is enough to store:
528 * 8TB - for both VMDK3 & VMDK4 with
529 * minimal cluster size: 512B
530 * minimal L2 table size: 512 entries
531 * 8 TB is still more than the maximal value supported for
532 * VMDK3 & VMDK4 which is 2TB.
533 * 64TB - for "ESXi seSparse Extent"
534 * minimal cluster size: 512B (default is 4KB)
535 * L2 table size: 4096 entries (const).
536 * 64TB is more than the maximal value supported for
537 * seSparse VMDKs (which is slightly less than 64TB)
539 error_setg(errp, "L1 size too big");
543 nb_sectors = bdrv_nb_sectors(file->bs);
544 if (nb_sectors < 0) {
548 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
549 extent = &s->extents[s->num_extents];
552 memset(extent, 0, sizeof(VmdkExtent));
555 extent->sectors = sectors;
556 extent->l1_table_offset = l1_offset;
557 extent->l1_backup_table_offset = l1_backup_offset;
558 extent->l1_size = l1_size;
559 extent->l1_entry_sectors = l2_size * cluster_sectors;
560 extent->l2_size = l2_size;
561 extent->cluster_sectors = flat ? sectors : cluster_sectors;
562 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
563 extent->entry_size = sizeof(uint32_t);
565 if (s->num_extents > 1) {
566 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
568 extent->end_sector = extent->sectors;
570 bs->total_sectors = extent->end_sector;
572 *new_extent = extent;
577 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
584 /* read the L1 table */
585 l1_size = extent->l1_size * extent->entry_size;
586 extent->l1_table = g_try_malloc(l1_size);
587 if (l1_size && extent->l1_table == NULL) {
591 ret = bdrv_pread(extent->file, extent->l1_table_offset, l1_size,
592 extent->l1_table, 0);
594 bdrv_refresh_filename(extent->file->bs);
595 error_setg_errno(errp, -ret,
596 "Could not read l1 table from extent '%s'",
597 extent->file->bs->filename);
600 for (i = 0; i < extent->l1_size; i++) {
601 if (extent->entry_size == sizeof(uint64_t)) {
602 le64_to_cpus((uint64_t *)extent->l1_table + i);
604 assert(extent->entry_size == sizeof(uint32_t));
605 le32_to_cpus((uint32_t *)extent->l1_table + i);
609 if (extent->l1_backup_table_offset) {
610 assert(!extent->sesparse);
611 extent->l1_backup_table = g_try_malloc(l1_size);
612 if (l1_size && extent->l1_backup_table == NULL) {
616 ret = bdrv_pread(extent->file, extent->l1_backup_table_offset,
617 l1_size, extent->l1_backup_table, 0);
619 bdrv_refresh_filename(extent->file->bs);
620 error_setg_errno(errp, -ret,
621 "Could not read l1 backup table from extent '%s'",
622 extent->file->bs->filename);
625 for (i = 0; i < extent->l1_size; i++) {
626 le32_to_cpus(&extent->l1_backup_table[i]);
631 g_malloc(extent->entry_size * extent->l2_size * L2_CACHE_SIZE);
634 g_free(extent->l1_backup_table);
636 g_free(extent->l1_table);
640 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
642 int flags, Error **errp)
647 VmdkExtent *extent = NULL;
649 ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0);
651 bdrv_refresh_filename(file->bs);
652 error_setg_errno(errp, -ret,
653 "Could not read header from file '%s'",
657 ret = vmdk_add_extent(bs, file, false,
658 le32_to_cpu(header.disk_sectors),
659 (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
661 le32_to_cpu(header.l1dir_size),
663 le32_to_cpu(header.granularity),
669 ret = vmdk_init_tables(bs, extent, errp);
671 /* free extent allocated by vmdk_add_extent */
672 vmdk_free_last_extent(bs);
677 #define SESPARSE_CONST_HEADER_MAGIC UINT64_C(0x00000000cafebabe)
678 #define SESPARSE_VOLATILE_HEADER_MAGIC UINT64_C(0x00000000cafecafe)
680 /* Strict checks - format not officially documented */
681 static int check_se_sparse_const_header(VMDKSESparseConstHeader *header,
684 header->magic = le64_to_cpu(header->magic);
685 header->version = le64_to_cpu(header->version);
686 header->grain_size = le64_to_cpu(header->grain_size);
687 header->grain_table_size = le64_to_cpu(header->grain_table_size);
688 header->flags = le64_to_cpu(header->flags);
689 header->reserved1 = le64_to_cpu(header->reserved1);
690 header->reserved2 = le64_to_cpu(header->reserved2);
691 header->reserved3 = le64_to_cpu(header->reserved3);
692 header->reserved4 = le64_to_cpu(header->reserved4);
694 header->volatile_header_offset =
695 le64_to_cpu(header->volatile_header_offset);
696 header->volatile_header_size = le64_to_cpu(header->volatile_header_size);
698 header->journal_header_offset = le64_to_cpu(header->journal_header_offset);
699 header->journal_header_size = le64_to_cpu(header->journal_header_size);
701 header->journal_offset = le64_to_cpu(header->journal_offset);
702 header->journal_size = le64_to_cpu(header->journal_size);
704 header->grain_dir_offset = le64_to_cpu(header->grain_dir_offset);
705 header->grain_dir_size = le64_to_cpu(header->grain_dir_size);
707 header->grain_tables_offset = le64_to_cpu(header->grain_tables_offset);
708 header->grain_tables_size = le64_to_cpu(header->grain_tables_size);
710 header->free_bitmap_offset = le64_to_cpu(header->free_bitmap_offset);
711 header->free_bitmap_size = le64_to_cpu(header->free_bitmap_size);
713 header->backmap_offset = le64_to_cpu(header->backmap_offset);
714 header->backmap_size = le64_to_cpu(header->backmap_size);
716 header->grains_offset = le64_to_cpu(header->grains_offset);
717 header->grains_size = le64_to_cpu(header->grains_size);
719 if (header->magic != SESPARSE_CONST_HEADER_MAGIC) {
720 error_setg(errp, "Bad const header magic: 0x%016" PRIx64,
725 if (header->version != 0x0000000200000001) {
726 error_setg(errp, "Unsupported version: 0x%016" PRIx64,
731 if (header->grain_size != 8) {
732 error_setg(errp, "Unsupported grain size: %" PRIu64,
737 if (header->grain_table_size != 64) {
738 error_setg(errp, "Unsupported grain table size: %" PRIu64,
739 header->grain_table_size);
743 if (header->flags != 0) {
744 error_setg(errp, "Unsupported flags: 0x%016" PRIx64,
749 if (header->reserved1 != 0 || header->reserved2 != 0 ||
750 header->reserved3 != 0 || header->reserved4 != 0) {
751 error_setg(errp, "Unsupported reserved bits:"
752 " 0x%016" PRIx64 " 0x%016" PRIx64
753 " 0x%016" PRIx64 " 0x%016" PRIx64,
754 header->reserved1, header->reserved2,
755 header->reserved3, header->reserved4);
759 /* check that padding is 0 */
760 if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
761 error_setg(errp, "Unsupported non-zero const header padding");
768 static int check_se_sparse_volatile_header(VMDKSESparseVolatileHeader *header,
771 header->magic = le64_to_cpu(header->magic);
772 header->free_gt_number = le64_to_cpu(header->free_gt_number);
773 header->next_txn_seq_number = le64_to_cpu(header->next_txn_seq_number);
774 header->replay_journal = le64_to_cpu(header->replay_journal);
776 if (header->magic != SESPARSE_VOLATILE_HEADER_MAGIC) {
777 error_setg(errp, "Bad volatile header magic: 0x%016" PRIx64,
782 if (header->replay_journal) {
783 error_setg(errp, "Image is dirty, Replaying journal not supported");
787 /* check that padding is 0 */
788 if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
789 error_setg(errp, "Unsupported non-zero volatile header padding");
796 static int vmdk_open_se_sparse(BlockDriverState *bs,
798 int flags, Error **errp)
801 VMDKSESparseConstHeader const_header;
802 VMDKSESparseVolatileHeader volatile_header;
803 VmdkExtent *extent = NULL;
805 ret = bdrv_apply_auto_read_only(bs,
806 "No write support for seSparse images available", errp);
811 assert(sizeof(const_header) == SECTOR_SIZE);
813 ret = bdrv_pread(file, 0, sizeof(const_header), &const_header, 0);
815 bdrv_refresh_filename(file->bs);
816 error_setg_errno(errp, -ret,
817 "Could not read const header from file '%s'",
822 /* check const header */
823 ret = check_se_sparse_const_header(&const_header, errp);
828 assert(sizeof(volatile_header) == SECTOR_SIZE);
830 ret = bdrv_pread(file, const_header.volatile_header_offset * SECTOR_SIZE,
831 sizeof(volatile_header), &volatile_header, 0);
833 bdrv_refresh_filename(file->bs);
834 error_setg_errno(errp, -ret,
835 "Could not read volatile header from file '%s'",
840 /* check volatile header */
841 ret = check_se_sparse_volatile_header(&volatile_header, errp);
846 ret = vmdk_add_extent(bs, file, false,
847 const_header.capacity,
848 const_header.grain_dir_offset * SECTOR_SIZE,
850 const_header.grain_dir_size *
851 SECTOR_SIZE / sizeof(uint64_t),
852 const_header.grain_table_size *
853 SECTOR_SIZE / sizeof(uint64_t),
854 const_header.grain_size,
861 extent->sesparse = true;
862 extent->sesparse_l2_tables_offset = const_header.grain_tables_offset;
863 extent->sesparse_clusters_offset = const_header.grains_offset;
864 extent->entry_size = sizeof(uint64_t);
866 ret = vmdk_init_tables(bs, extent, errp);
868 /* free extent allocated by vmdk_add_extent */
869 vmdk_free_last_extent(bs);
875 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
876 QDict *options, Error **errp);
878 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
884 size = bdrv_getlength(file->bs);
886 error_setg_errno(errp, -size, "Could not access file");
891 /* Both descriptor file and sparse image must be much larger than 4
892 * bytes, also callers of vmdk_read_desc want to compare the first 4
893 * bytes with VMDK4_MAGIC, let's error out if less is read. */
894 error_setg(errp, "File is too small, not a valid image");
898 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */
899 buf = g_malloc(size + 1);
901 ret = bdrv_pread(file, desc_offset, size, buf, 0);
903 error_setg_errno(errp, -ret, "Could not read from file");
912 static int vmdk_open_vmdk4(BlockDriverState *bs,
914 int flags, QDict *options, Error **errp)
918 uint32_t l1_size, l1_entry_sectors;
920 VmdkExtent *extent = NULL;
921 BDRVVmdkState *s = bs->opaque;
922 int64_t l1_backup_offset = 0;
925 ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0);
927 bdrv_refresh_filename(file->bs);
928 error_setg_errno(errp, -ret,
929 "Could not read header from file '%s'",
933 if (header.capacity == 0) {
934 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
936 char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
940 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
946 if (!s->create_type) {
947 s->create_type = g_strdup("monolithicSparse");
950 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
952 * The footer takes precedence over the header, so read it in. The
953 * footer starts at offset -1024 from the end: One sector for the
954 * footer, and another one for the end-of-stream marker.
961 uint8_t pad[512 - 16];
962 } QEMU_PACKED footer_marker;
966 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
972 uint8_t pad[512 - 16];
973 } QEMU_PACKED eos_marker;
974 } QEMU_PACKED footer;
976 ret = bdrv_pread(file, bs->file->bs->total_sectors * 512 - 1536,
977 sizeof(footer), &footer, 0);
979 error_setg_errno(errp, -ret, "Failed to read footer");
983 /* Some sanity checks for the footer */
984 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
985 le32_to_cpu(footer.footer_marker.size) != 0 ||
986 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
987 le64_to_cpu(footer.eos_marker.val) != 0 ||
988 le32_to_cpu(footer.eos_marker.size) != 0 ||
989 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
991 error_setg(errp, "Invalid footer");
995 header = footer.header;
999 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1000 if (le32_to_cpu(header.version) > 3) {
1001 error_setg(errp, "Unsupported VMDK version %" PRIu32,
1002 le32_to_cpu(header.version));
1004 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
1006 /* VMware KB 2064959 explains that version 3 added support for
1007 * persistent changed block tracking (CBT), and backup software can
1008 * read it as version=1 if it doesn't care about the changed area
1009 * information. So we are safe to enable read only. */
1010 error_setg(errp, "VMDK version 3 must be read only");
1014 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
1015 error_setg(errp, "L2 table size too big");
1019 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
1020 * le64_to_cpu(header.granularity);
1021 if (l1_entry_sectors == 0) {
1022 error_setg(errp, "L1 entry size is invalid");
1025 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
1027 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
1028 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
1030 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
1031 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
1032 (int64_t)(le64_to_cpu(header.grain_offset)
1033 * BDRV_SECTOR_SIZE));
1037 ret = vmdk_add_extent(bs, file, false,
1038 le64_to_cpu(header.capacity),
1039 le64_to_cpu(header.gd_offset) << 9,
1042 le32_to_cpu(header.num_gtes_per_gt),
1043 le64_to_cpu(header.granularity),
1049 extent->compressed =
1050 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1051 if (extent->compressed) {
1052 g_free(s->create_type);
1053 s->create_type = g_strdup("streamOptimized");
1055 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
1056 extent->version = le32_to_cpu(header.version);
1057 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
1058 ret = vmdk_init_tables(bs, extent, errp);
1060 /* free extent allocated by vmdk_add_extent */
1061 vmdk_free_last_extent(bs);
1066 /* find an option value out of descriptor file */
1067 static int vmdk_parse_description(const char *desc, const char *opt_name,
1068 char *buf, int buf_size)
1070 char *opt_pos, *opt_end;
1071 const char *end = desc + strlen(desc);
1073 opt_pos = strstr(desc, opt_name);
1077 /* Skip "=\"" following opt_name */
1078 opt_pos += strlen(opt_name) + 2;
1079 if (opt_pos >= end) {
1083 while (opt_end < end && *opt_end != '"') {
1086 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
1089 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
1093 /* Open an extent file and append to bs array */
1094 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
1095 char *buf, QDict *options, Error **errp)
1099 magic = ldl_be_p(buf);
1102 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
1104 return vmdk_open_vmdk4(bs, file, flags, options, errp);
1106 error_setg(errp, "Image not in VMDK format");
1111 static const char *next_line(const char *s)
1122 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
1123 QDict *options, Error **errp)
1131 int64_t sectors = 0;
1132 int64_t flat_offset;
1133 char *desc_file_dir = NULL;
1135 BdrvChild *extent_file;
1136 BdrvChildRole extent_role;
1137 BDRVVmdkState *s = bs->opaque;
1138 VmdkExtent *extent = NULL;
1139 char extent_opt_prefix[32];
1140 Error *local_err = NULL;
1142 for (p = desc; *p; p = next_line(p)) {
1143 /* parse extent line in one of below formats:
1145 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
1146 * RW [size in sectors] SPARSE "file-name.vmdk"
1147 * RW [size in sectors] VMFS "file-name.vmdk"
1148 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
1149 * RW [size in sectors] SESPARSE "file-name.vmdk"
1152 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
1153 access, §ors, type, fname, &flat_offset);
1154 if (matches < 4 || strcmp(access, "RW")) {
1156 } else if (!strcmp(type, "FLAT")) {
1157 if (matches != 5 || flat_offset < 0) {
1160 } else if (!strcmp(type, "VMFS")) {
1166 } else if (matches != 4) {
1171 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
1172 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") &&
1173 strcmp(type, "SESPARSE")) ||
1174 (strcmp(access, "RW"))) {
1178 if (path_is_absolute(fname)) {
1179 extent_path = g_strdup(fname);
1181 if (!desc_file_dir) {
1182 desc_file_dir = bdrv_dirname(bs->file->bs, errp);
1183 if (!desc_file_dir) {
1184 bdrv_refresh_filename(bs->file->bs);
1185 error_prepend(errp, "Cannot use relative paths with VMDK "
1186 "descriptor file '%s': ",
1187 bs->file->bs->filename);
1193 extent_path = g_strconcat(desc_file_dir, fname, NULL);
1196 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
1199 extent_role = BDRV_CHILD_DATA;
1200 if (strcmp(type, "FLAT") != 0 && strcmp(type, "VMFS") != 0) {
1201 /* non-flat extents have metadata */
1202 extent_role |= BDRV_CHILD_METADATA;
1205 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
1206 bs, &child_of_bds, extent_role, false,
1208 g_free(extent_path);
1210 error_propagate(errp, local_err);
1215 /* save to extents array */
1216 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
1219 ret = vmdk_add_extent(bs, extent_file, true, sectors,
1220 0, 0, 0, 0, 0, &extent, errp);
1222 bdrv_unref_child(bs, extent_file);
1225 extent->flat_start_offset = flat_offset << 9;
1226 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
1227 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
1228 char *buf = vmdk_read_desc(extent_file, 0, errp);
1232 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
1237 bdrv_unref_child(bs, extent_file);
1240 extent = &s->extents[s->num_extents - 1];
1241 } else if (!strcmp(type, "SESPARSE")) {
1242 ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
1244 bdrv_unref_child(bs, extent_file);
1247 extent = &s->extents[s->num_extents - 1];
1249 error_setg(errp, "Unsupported extent type '%s'", type);
1250 bdrv_unref_child(bs, extent_file);
1254 extent->type = g_strdup(type);
1263 if (np[-1] == '\n') {
1266 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
1270 g_free(desc_file_dir);
1274 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
1275 QDict *options, Error **errp)
1279 BDRVVmdkState *s = bs->opaque;
1281 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
1282 error_setg(errp, "invalid VMDK image descriptor");
1286 if (strcmp(ct, "monolithicFlat") &&
1287 strcmp(ct, "vmfs") &&
1288 strcmp(ct, "vmfsSparse") &&
1289 strcmp(ct, "seSparse") &&
1290 strcmp(ct, "twoGbMaxExtentSparse") &&
1291 strcmp(ct, "twoGbMaxExtentFlat")) {
1292 error_setg(errp, "Unsupported image type '%s'", ct);
1296 s->create_type = g_strdup(ct);
1298 ret = vmdk_parse_extents(buf, bs, options, errp);
1303 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
1308 BDRVVmdkState *s = bs->opaque;
1311 ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
1316 buf = vmdk_read_desc(bs->file, 0, errp);
1321 magic = ldl_be_p(buf);
1325 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
1327 s->desc_offset = 0x200;
1330 /* No data in the descriptor file */
1331 bs->file->role &= ~BDRV_CHILD_DATA;
1333 /* Must succeed because we have given up permissions if anything */
1334 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1336 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1343 /* try to open parent images, if exist */
1344 ret = vmdk_parent_open(bs);
1348 ret = vmdk_read_cid(bs, 0, &s->cid);
1352 ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1356 qemu_co_mutex_init(&s->lock);
1358 /* Disable migration when VMDK images are used */
1359 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1360 "does not support live migration",
1361 bdrv_get_device_or_node_name(bs));
1362 ret = migrate_add_blocker(s->migration_blocker, errp);
1364 error_free(s->migration_blocker);
1373 g_free(s->create_type);
1374 s->create_type = NULL;
1375 vmdk_free_extents(bs);
1380 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1382 BDRVVmdkState *s = bs->opaque;
1385 for (i = 0; i < s->num_extents; i++) {
1386 if (!s->extents[i].flat) {
1387 bs->bl.pwrite_zeroes_alignment =
1388 MAX(bs->bl.pwrite_zeroes_alignment,
1389 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1397 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1398 * to the cluster at @cluster_sector_num. If @zeroed is true, we're overwriting
1399 * a zeroed cluster in the current layer and must not copy data from the
1402 * If @skip_start_sector < @skip_end_sector, the relative range
1403 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1404 * it for call to write user data in the request.
1406 static int coroutine_fn GRAPH_RDLOCK
1407 get_whole_cluster(BlockDriverState *bs, VmdkExtent *extent,
1408 uint64_t cluster_offset, uint64_t offset,
1409 uint64_t skip_start_bytes, uint64_t skip_end_bytes,
1413 int64_t cluster_bytes;
1414 uint8_t *whole_grain;
1415 bool copy_from_backing;
1417 /* For COW, align request sector_num to cluster start */
1418 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1419 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1420 whole_grain = qemu_blockalign(bs, cluster_bytes);
1421 copy_from_backing = bs->backing && !zeroed;
1423 if (!copy_from_backing) {
1424 memset(whole_grain, 0, skip_start_bytes);
1425 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1428 assert(skip_end_bytes <= cluster_bytes);
1429 /* we will be here if it's first write on non-exist grain(cluster).
1430 * try to read from parent image, if exist */
1431 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1436 /* Read backing data before skip range */
1437 if (skip_start_bytes > 0) {
1438 if (copy_from_backing) {
1439 /* qcow2 emits this on bs->file instead of bs->backing */
1440 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1441 ret = bdrv_co_pread(bs->backing, offset, skip_start_bytes,
1448 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1449 ret = bdrv_co_pwrite(extent->file, cluster_offset, skip_start_bytes,
1456 /* Read backing data after skip range */
1457 if (skip_end_bytes < cluster_bytes) {
1458 if (copy_from_backing) {
1459 /* qcow2 emits this on bs->file instead of bs->backing */
1460 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1461 ret = bdrv_co_pread(bs->backing, offset + skip_end_bytes,
1462 cluster_bytes - skip_end_bytes,
1463 whole_grain + skip_end_bytes, 0);
1469 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1470 ret = bdrv_co_pwrite(extent->file, cluster_offset + skip_end_bytes,
1471 cluster_bytes - skip_end_bytes,
1472 whole_grain + skip_end_bytes, 0);
1481 qemu_vfree(whole_grain);
1485 static int coroutine_fn GRAPH_RDLOCK
1486 vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data, uint32_t offset)
1488 offset = cpu_to_le32(offset);
1489 /* update L2 table */
1490 BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1491 if (bdrv_co_pwrite(extent->file,
1492 ((int64_t)m_data->l2_offset * 512)
1493 + (m_data->l2_index * sizeof(offset)),
1494 sizeof(offset), &offset, 0) < 0) {
1497 /* update backup L2 table */
1498 if (extent->l1_backup_table_offset != 0) {
1499 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1500 if (bdrv_co_pwrite(extent->file,
1501 ((int64_t)m_data->l2_offset * 512)
1502 + (m_data->l2_index * sizeof(offset)),
1503 sizeof(offset), &offset, 0) < 0) {
1507 if (bdrv_co_flush(extent->file->bs) < 0) {
1510 if (m_data->l2_cache_entry) {
1511 *m_data->l2_cache_entry = offset;
1518 * get_cluster_offset
1520 * Look up cluster offset in extent file by sector number, and store in
1523 * For flat extents, the start offset as parsed from the description file is
1526 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1527 * offset for a new cluster and update L2 cache. If there is a backing file,
1528 * COW is done before returning; otherwise, zeroes are written to the allocated
1529 * cluster. Both COW and zero writing skips the sector range
1530 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1531 * has new data to write there.
1533 * Returns: VMDK_OK if cluster exists and mapped in the image.
1534 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1535 * VMDK_ERROR if failed.
1537 static int coroutine_fn GRAPH_RDLOCK
1538 get_cluster_offset(BlockDriverState *bs, VmdkExtent *extent,
1539 VmdkMetaData *m_data, uint64_t offset, bool allocate,
1540 uint64_t *cluster_offset, uint64_t skip_start_bytes,
1541 uint64_t skip_end_bytes)
1543 unsigned int l1_index, l2_offset, l2_index;
1544 int min_index, i, j;
1547 bool zeroed = false;
1549 int64_t cluster_sector;
1550 unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
1553 m_data->new_allocation = false;
1556 *cluster_offset = extent->flat_start_offset;
1560 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1561 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1562 if (l1_index >= extent->l1_size) {
1565 if (extent->sesparse) {
1566 uint64_t l2_offset_u64;
1568 assert(extent->entry_size == sizeof(uint64_t));
1570 l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
1571 if (l2_offset_u64 == 0) {
1573 } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
1575 * Top most nibble is 0x1 if grain table is allocated.
1576 * strict check - top most 4 bytes must be 0x10000000 since max
1577 * supported size is 64TB for disk - so no more than 64TB / 16MB
1578 * grain directories which is smaller than uint32,
1579 * where 16MB is the only supported default grain table coverage.
1583 l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
1584 l2_offset_u64 = extent->sesparse_l2_tables_offset +
1585 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
1586 if (l2_offset_u64 > 0x00000000ffffffff) {
1589 l2_offset = (unsigned int)(l2_offset_u64);
1592 assert(extent->entry_size == sizeof(uint32_t));
1593 l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
1596 return VMDK_UNALLOC;
1598 for (i = 0; i < L2_CACHE_SIZE; i++) {
1599 if (l2_offset == extent->l2_cache_offsets[i]) {
1600 /* increment the hit count */
1601 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1602 for (j = 0; j < L2_CACHE_SIZE; j++) {
1603 extent->l2_cache_counts[j] >>= 1;
1606 l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
1610 /* not found: load a new entry in the least used one */
1612 min_count = 0xffffffff;
1613 for (i = 0; i < L2_CACHE_SIZE; i++) {
1614 if (extent->l2_cache_counts[i] < min_count) {
1615 min_count = extent->l2_cache_counts[i];
1619 l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
1620 BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1621 if (bdrv_co_pread(extent->file,
1622 (int64_t)l2_offset * 512,
1629 extent->l2_cache_offsets[min_index] = l2_offset;
1630 extent->l2_cache_counts[min_index] = 1;
1632 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1634 m_data->l1_index = l1_index;
1635 m_data->l2_index = l2_index;
1636 m_data->l2_offset = l2_offset;
1637 m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
1640 if (extent->sesparse) {
1641 cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
1642 switch (cluster_sector & 0xf000000000000000) {
1643 case 0x0000000000000000:
1644 /* unallocated grain */
1645 if (cluster_sector != 0) {
1649 case 0x1000000000000000:
1650 /* scsi-unmapped grain - fallthrough */
1651 case 0x2000000000000000:
1655 case 0x3000000000000000:
1656 /* allocated grain */
1657 cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
1658 ((cluster_sector & 0x0000ffffffffffff) << 12));
1659 cluster_sector = extent->sesparse_clusters_offset +
1660 cluster_sector * extent->cluster_sectors;
1666 cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
1668 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1673 if (!cluster_sector || zeroed) {
1675 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1677 assert(!extent->sesparse);
1679 if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1683 cluster_sector = extent->next_cluster_sector;
1684 extent->next_cluster_sector += extent->cluster_sectors;
1686 /* First of all we write grain itself, to avoid race condition
1687 * that may to corrupt the image.
1688 * This problem may occur because of insufficient space on host disk
1689 * or inappropriate VM shutdown.
1691 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1692 offset, skip_start_bytes, skip_end_bytes,
1698 m_data->new_allocation = true;
1701 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1705 static VmdkExtent *find_extent(BDRVVmdkState *s,
1706 int64_t sector_num, VmdkExtent *start_hint)
1708 VmdkExtent *extent = start_hint;
1711 extent = &s->extents[0];
1713 while (extent < &s->extents[s->num_extents]) {
1714 if (sector_num < extent->end_sector) {
1722 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1725 uint64_t extent_begin_offset, extent_relative_offset;
1726 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1728 extent_begin_offset =
1729 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1730 extent_relative_offset = offset - extent_begin_offset;
1731 return extent_relative_offset % cluster_size;
1734 static int coroutine_fn GRAPH_RDLOCK
1735 vmdk_co_block_status(BlockDriverState *bs, bool want_zero,
1736 int64_t offset, int64_t bytes, int64_t *pnum,
1737 int64_t *map, BlockDriverState **file)
1739 BDRVVmdkState *s = bs->opaque;
1740 int64_t index_in_cluster, n, ret;
1741 uint64_t cluster_offset;
1744 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1748 qemu_co_mutex_lock(&s->lock);
1749 ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1751 qemu_co_mutex_unlock(&s->lock);
1753 index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1762 ret = BDRV_BLOCK_ZERO;
1765 ret = BDRV_BLOCK_DATA;
1766 if (!extent->compressed) {
1767 ret |= BDRV_BLOCK_OFFSET_VALID;
1768 *map = cluster_offset + index_in_cluster;
1770 ret |= BDRV_BLOCK_RECURSE;
1773 *file = extent->file->bs;
1777 n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1778 *pnum = MIN(n, bytes);
1782 static int coroutine_fn GRAPH_RDLOCK
1783 vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1784 int64_t offset_in_cluster, QEMUIOVector *qiov,
1785 uint64_t qiov_offset, uint64_t n_bytes,
1789 VmdkGrainMarker *data = NULL;
1791 QEMUIOVector local_qiov;
1792 int64_t write_offset;
1793 int64_t write_end_sector;
1795 if (extent->compressed) {
1796 void *compressed_data;
1798 /* Only whole clusters */
1799 if (offset_in_cluster ||
1800 n_bytes > (extent->cluster_sectors * SECTOR_SIZE) ||
1801 (n_bytes < (extent->cluster_sectors * SECTOR_SIZE) &&
1802 offset + n_bytes != extent->end_sector * SECTOR_SIZE))
1808 if (!extent->has_marker) {
1812 buf_len = (extent->cluster_sectors << 9) * 2;
1813 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1815 compressed_data = g_malloc(n_bytes);
1816 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1817 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1818 g_free(compressed_data);
1820 if (ret != Z_OK || buf_len == 0) {
1825 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1826 data->size = cpu_to_le32(buf_len);
1828 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1829 qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1831 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1833 qemu_iovec_init(&local_qiov, qiov->niov);
1834 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1836 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1839 write_offset = cluster_offset + offset_in_cluster;
1840 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1843 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1845 if (extent->compressed) {
1846 extent->next_cluster_sector = write_end_sector;
1848 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1858 if (!extent->compressed) {
1859 qemu_iovec_destroy(&local_qiov);
1864 static int coroutine_fn GRAPH_RDLOCK
1865 vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1866 int64_t offset_in_cluster, QEMUIOVector *qiov, int bytes)
1869 int cluster_bytes, buf_bytes;
1870 uint8_t *cluster_buf, *compressed_data;
1871 uint8_t *uncomp_buf;
1873 VmdkGrainMarker *marker;
1877 if (!extent->compressed) {
1878 BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1879 ret = bdrv_co_preadv(extent->file,
1880 cluster_offset + offset_in_cluster, bytes,
1887 cluster_bytes = extent->cluster_sectors * 512;
1888 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1889 buf_bytes = cluster_bytes * 2;
1890 cluster_buf = g_malloc(buf_bytes);
1891 uncomp_buf = g_malloc(cluster_bytes);
1892 BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1893 ret = bdrv_co_pread(extent->file, cluster_offset, buf_bytes, cluster_buf,
1898 compressed_data = cluster_buf;
1899 buf_len = cluster_bytes;
1900 data_len = cluster_bytes;
1901 if (extent->has_marker) {
1902 marker = (VmdkGrainMarker *)cluster_buf;
1903 compressed_data = marker->data;
1904 data_len = le32_to_cpu(marker->size);
1906 if (!data_len || data_len > buf_bytes) {
1910 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1916 if (offset_in_cluster < 0 ||
1917 offset_in_cluster + bytes > buf_len) {
1921 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1926 g_free(cluster_buf);
1930 static int coroutine_fn GRAPH_RDLOCK
1931 vmdk_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
1932 QEMUIOVector *qiov, BdrvRequestFlags flags)
1934 BDRVVmdkState *s = bs->opaque;
1936 uint64_t n_bytes, offset_in_cluster;
1937 VmdkExtent *extent = NULL;
1938 QEMUIOVector local_qiov;
1939 uint64_t cluster_offset;
1940 uint64_t bytes_done = 0;
1942 qemu_iovec_init(&local_qiov, qiov->niov);
1943 qemu_co_mutex_lock(&s->lock);
1946 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1951 ret = get_cluster_offset(bs, extent, NULL,
1952 offset, false, &cluster_offset, 0, 0);
1953 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1955 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1956 - offset_in_cluster);
1958 if (ret != VMDK_OK) {
1959 /* if not allocated, try to read from parent image, if exist */
1960 if (bs->backing && ret != VMDK_ZEROED) {
1961 if (!vmdk_is_cid_valid(bs)) {
1966 qemu_iovec_reset(&local_qiov);
1967 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1969 /* qcow2 emits this on bs->file instead of bs->backing */
1970 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1971 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1977 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1980 qemu_iovec_reset(&local_qiov);
1981 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1983 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1984 &local_qiov, n_bytes);
1991 bytes_done += n_bytes;
1996 qemu_co_mutex_unlock(&s->lock);
1997 qemu_iovec_destroy(&local_qiov);
2004 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
2005 * if possible, otherwise return -ENOTSUP.
2006 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
2007 * with each cluster. By dry run we can find if the zero write
2008 * is possible without modifying image data.
2010 * Returns: error code with 0 for success.
2012 static int coroutine_fn GRAPH_RDLOCK
2013 vmdk_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2014 QEMUIOVector *qiov, bool zeroed, bool zero_dry_run)
2016 BDRVVmdkState *s = bs->opaque;
2017 VmdkExtent *extent = NULL;
2019 int64_t offset_in_cluster, n_bytes;
2020 uint64_t cluster_offset;
2021 uint64_t bytes_done = 0;
2022 VmdkMetaData m_data;
2024 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
2025 error_report("Wrong offset: offset=0x%" PRIx64
2026 " total_sectors=0x%" PRIx64,
2027 offset, bs->total_sectors);
2032 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
2036 if (extent->sesparse) {
2039 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
2040 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
2041 - offset_in_cluster);
2043 ret = get_cluster_offset(bs, extent, &m_data, offset,
2044 !(extent->compressed || zeroed),
2045 &cluster_offset, offset_in_cluster,
2046 offset_in_cluster + n_bytes);
2047 if (extent->compressed) {
2048 if (ret == VMDK_OK) {
2049 /* Refuse write to allocated cluster for streamOptimized */
2050 error_report("Could not write to allocated cluster"
2051 " for streamOptimized");
2053 } else if (!zeroed) {
2055 ret = get_cluster_offset(bs, extent, &m_data, offset,
2056 true, &cluster_offset, 0, 0);
2059 if (ret == VMDK_ERROR) {
2063 /* Do zeroed write, buf is ignored */
2064 if (extent->has_zero_grain &&
2065 offset_in_cluster == 0 &&
2066 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
2067 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
2068 if (!zero_dry_run && ret != VMDK_ZEROED) {
2069 /* update L2 tables */
2070 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
2079 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
2080 qiov, bytes_done, n_bytes, offset);
2084 if (m_data.new_allocation) {
2085 /* update L2 tables */
2086 if (vmdk_L2update(extent, &m_data,
2087 cluster_offset >> BDRV_SECTOR_BITS)
2095 bytes_done += n_bytes;
2097 /* update CID on the first write every time the virtual disk is
2099 if (!s->cid_updated) {
2100 ret = vmdk_write_cid(bs, g_random_int());
2104 s->cid_updated = true;
2110 static int coroutine_fn GRAPH_RDLOCK
2111 vmdk_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
2112 QEMUIOVector *qiov, BdrvRequestFlags flags)
2115 BDRVVmdkState *s = bs->opaque;
2116 qemu_co_mutex_lock(&s->lock);
2117 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
2118 qemu_co_mutex_unlock(&s->lock);
2122 static int coroutine_fn GRAPH_RDLOCK
2123 vmdk_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes,
2127 /* The caller will write bytes 0 to signal EOF.
2128 * When receive it, we align EOF to a sector boundary. */
2129 BDRVVmdkState *s = bs->opaque;
2133 for (i = 0; i < s->num_extents; i++) {
2134 length = bdrv_getlength(s->extents[i].file->bs);
2138 length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
2139 ret = bdrv_co_truncate(s->extents[i].file, length, false,
2140 PREALLOC_MODE_OFF, 0, NULL);
2147 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
2150 static int coroutine_fn GRAPH_RDLOCK
2151 vmdk_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
2152 BdrvRequestFlags flags)
2155 BDRVVmdkState *s = bs->opaque;
2157 qemu_co_mutex_lock(&s->lock);
2158 /* write zeroes could fail if sectors not aligned to cluster, test it with
2159 * dry_run == true before really updating image */
2160 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
2162 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
2164 qemu_co_mutex_unlock(&s->lock);
2168 static int vmdk_init_extent(BlockBackend *blk,
2169 int64_t filesize, bool flat,
2170 bool compress, bool zeroed_grain,
2175 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
2176 uint32_t *gd_buf = NULL;
2180 ret = blk_truncate(blk, filesize, false, PREALLOC_MODE_OFF, 0, errp);
2183 magic = cpu_to_be32(VMDK4_MAGIC);
2184 memset(&header, 0, sizeof(header));
2187 } else if (zeroed_grain) {
2192 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
2193 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
2194 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
2195 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
2196 header.capacity = filesize / BDRV_SECTOR_SIZE;
2197 header.granularity = 128;
2198 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
2200 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
2201 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
2203 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
2204 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
2206 header.desc_offset = 1;
2207 header.desc_size = 20;
2208 header.rgd_offset = header.desc_offset + header.desc_size;
2209 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
2210 header.grain_offset =
2211 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
2212 header.granularity);
2213 /* swap endianness for all header fields */
2214 header.version = cpu_to_le32(header.version);
2215 header.flags = cpu_to_le32(header.flags);
2216 header.capacity = cpu_to_le64(header.capacity);
2217 header.granularity = cpu_to_le64(header.granularity);
2218 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
2219 header.desc_offset = cpu_to_le64(header.desc_offset);
2220 header.desc_size = cpu_to_le64(header.desc_size);
2221 header.rgd_offset = cpu_to_le64(header.rgd_offset);
2222 header.gd_offset = cpu_to_le64(header.gd_offset);
2223 header.grain_offset = cpu_to_le64(header.grain_offset);
2224 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
2226 header.check_bytes[0] = 0xa;
2227 header.check_bytes[1] = 0x20;
2228 header.check_bytes[2] = 0xd;
2229 header.check_bytes[3] = 0xa;
2231 /* write all the data */
2232 ret = blk_pwrite(blk, 0, sizeof(magic), &magic, 0);
2234 error_setg(errp, QERR_IO_ERROR);
2237 ret = blk_pwrite(blk, sizeof(magic), sizeof(header), &header, 0);
2239 error_setg(errp, QERR_IO_ERROR);
2243 ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false,
2244 PREALLOC_MODE_OFF, 0, errp);
2249 /* write grain directory */
2250 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
2251 gd_buf = g_malloc0(gd_buf_size);
2252 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
2253 i < gt_count; i++, tmp += gt_size) {
2254 gd_buf[i] = cpu_to_le32(tmp);
2256 ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
2257 gd_buf_size, gd_buf, 0);
2259 error_setg(errp, QERR_IO_ERROR);
2263 /* write backup grain directory */
2264 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
2265 i < gt_count; i++, tmp += gt_size) {
2266 gd_buf[i] = cpu_to_le32(tmp);
2268 ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
2269 gd_buf_size, gd_buf, 0);
2271 error_setg(errp, QERR_IO_ERROR);
2280 static int coroutine_fn GRAPH_RDLOCK
2281 vmdk_create_extent(const char *filename, int64_t filesize, bool flat,
2282 bool compress, bool zeroed_grain, BlockBackend **pbb,
2283 QemuOpts *opts, Error **errp)
2286 BlockBackend *blk = NULL;
2288 ret = bdrv_co_create_file(filename, opts, errp);
2293 blk = blk_co_new_open(filename, NULL, NULL,
2294 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2301 blk_set_allow_write_beyond_eof(blk, true);
2303 ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
2316 static int filename_decompose(const char *filename, char *path, char *prefix,
2317 char *postfix, size_t buf_len, Error **errp)
2321 if (filename == NULL || !strlen(filename)) {
2322 error_setg(errp, "No filename provided");
2325 p = strrchr(filename, '/');
2327 p = strrchr(filename, '\\');
2330 p = strrchr(filename, ':');
2334 if (p - filename >= buf_len) {
2337 pstrcpy(path, p - filename + 1, filename);
2342 q = strrchr(p, '.');
2344 pstrcpy(prefix, buf_len, p);
2347 if (q - p >= buf_len) {
2350 pstrcpy(prefix, q - p + 1, p);
2351 pstrcpy(postfix, buf_len, q);
2357 * idx == 0: get or create the descriptor file (also the image file if in a
2359 * idx >= 1: get the n-th extent if in a split subformat
2361 typedef BlockBackend * coroutine_fn /* GRAPH_RDLOCK */
2362 (*vmdk_create_extent_fn)(int64_t size, int idx, bool flat, bool split,
2363 bool compress, bool zeroed_grain, void *opaque,
2366 static void vmdk_desc_add_extent(GString *desc,
2367 const char *extent_line_fmt,
2368 int64_t size, const char *filename)
2370 char *basename = g_path_get_basename(filename);
2372 g_string_append_printf(desc, extent_line_fmt,
2373 DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
2377 static int coroutine_fn GRAPH_RDLOCK
2378 vmdk_co_do_create(int64_t size,
2379 BlockdevVmdkSubformat subformat,
2380 BlockdevVmdkAdapterType adapter_type,
2381 const char *backing_file,
2382 const char *hw_version,
2383 const char *toolsversion,
2386 vmdk_create_extent_fn extent_fn,
2391 BlockBackend *blk = NULL;
2392 BlockBackend *extent_blk;
2393 Error *local_err = NULL;
2396 bool flat, split, compress;
2397 GString *ext_desc_lines;
2398 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
2399 int64_t extent_size;
2400 int64_t created_size = 0;
2401 const char *extent_line_fmt;
2402 char *parent_desc_line = g_malloc0(BUF_SIZE);
2403 uint32_t parent_cid = 0xffffffff;
2404 uint32_t number_heads = 16;
2405 uint32_t desc_offset = 0, desc_len;
2406 const char desc_template[] =
2407 "# Disk DescriptorFile\n"
2410 "parentCID=%" PRIx32 "\n"
2411 "createType=\"%s\"\n"
2414 "# Extent description\n"
2417 "# The Disk Data Base\n"
2420 "ddb.virtualHWVersion = \"%s\"\n"
2421 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2422 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2423 "ddb.geometry.sectors = \"63\"\n"
2424 "ddb.adapterType = \"%s\"\n"
2425 "ddb.toolsVersion = \"%s\"\n";
2427 ext_desc_lines = g_string_new(NULL);
2429 /* Read out options */
2433 "compat6 cannot be enabled with hwversion set");
2442 if (!toolsversion) {
2443 toolsversion = "2147483647";
2446 if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2447 /* that's the number of heads with which vmware operates when
2448 creating, exporting, etc. vmdk files with a non-ide adapter type */
2451 split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2452 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2453 flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2454 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2455 compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2458 extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2460 extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2462 if (flat && backing_file) {
2463 error_setg(errp, "Flat image can't have backing file");
2467 if (flat && zeroed_grain) {
2468 error_setg(errp, "Flat image can't enable zeroed grain");
2473 /* Create extents */
2475 extent_size = split_size;
2479 if (!split && !flat) {
2480 created_size = extent_size;
2484 /* Get the descriptor file BDS */
2485 blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2491 if (!split && !flat) {
2492 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2493 blk_bs(blk)->filename);
2497 BlockBackend *backing;
2498 char *full_backing =
2499 bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2503 error_propagate(errp, local_err);
2507 assert(full_backing);
2509 backing = blk_co_new_open(full_backing, NULL, NULL,
2510 BDRV_O_NO_BACKING, errp);
2511 g_free(full_backing);
2512 if (backing == NULL) {
2516 if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2517 error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2518 blk_bs(backing)->drv->format_name);
2519 blk_co_unref(backing);
2523 ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2524 blk_co_unref(backing);
2526 error_setg(errp, "Failed to read parent CID");
2529 snprintf(parent_desc_line, BUF_SIZE,
2530 "parentFileNameHint=\"%s\"", backing_file);
2533 while (created_size < size) {
2534 int64_t cur_size = MIN(size - created_size, extent_size);
2535 extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2536 zeroed_grain, opaque, errp);
2541 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2542 blk_bs(extent_blk)->filename);
2543 created_size += cur_size;
2545 blk_co_unref(extent_blk);
2548 /* Check whether we got excess extents */
2549 extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2552 blk_co_unref(extent_blk);
2553 error_setg(errp, "List of extents contains unused extents");
2558 /* generate descriptor file */
2559 desc = g_strdup_printf(desc_template,
2562 BlockdevVmdkSubformat_str(subformat),
2564 ext_desc_lines->str,
2567 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2569 BlockdevVmdkAdapterType_str(adapter_type),
2571 desc_len = strlen(desc);
2572 /* the descriptor offset = 0x200 */
2573 if (!split && !flat) {
2574 desc_offset = 0x200;
2577 ret = blk_co_pwrite(blk, desc_offset, desc_len, desc, 0);
2579 error_setg_errno(errp, -ret, "Could not write description");
2582 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2583 * for description file */
2584 if (desc_offset == 0) {
2585 ret = blk_co_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, 0, errp);
2596 g_free(parent_desc_line);
2597 g_string_free(ext_desc_lines, true);
2606 } VMDKCreateOptsData;
2608 static BlockBackend * coroutine_fn GRAPH_RDLOCK
2609 vmdk_co_create_opts_cb(int64_t size, int idx, bool flat, bool split,
2610 bool compress, bool zeroed_grain, void *opaque,
2613 BlockBackend *blk = NULL;
2614 BlockDriverState *bs = NULL;
2615 VMDKCreateOptsData *data = opaque;
2616 char *ext_filename = NULL;
2617 char *rel_filename = NULL;
2619 /* We're done, don't create excess extents. */
2621 assert(errp == NULL);
2626 rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2628 rel_filename = g_strdup_printf("%s-%c%03d%s",
2630 flat ? 'f' : 's', idx, data->postfix);
2633 rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2636 ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2637 g_free(rel_filename);
2639 if (vmdk_create_extent(ext_filename, size,
2640 flat, compress, zeroed_grain, &blk, data->opts,
2646 g_free(ext_filename);
2650 static int coroutine_fn GRAPH_RDLOCK
2651 vmdk_co_create_opts(BlockDriver *drv, const char *filename,
2652 QemuOpts *opts, Error **errp)
2654 Error *local_err = NULL;
2656 int64_t total_size = 0;
2657 char *adapter_type = NULL;
2658 BlockdevVmdkAdapterType adapter_type_enum;
2659 char *backing_file = NULL;
2660 char *hw_version = NULL;
2661 char *toolsversion = NULL;
2663 BlockdevVmdkSubformat subformat;
2665 char *path = g_malloc0(PATH_MAX);
2666 char *prefix = g_malloc0(PATH_MAX);
2667 char *postfix = g_malloc0(PATH_MAX);
2668 char *desc_line = g_malloc0(BUF_SIZE);
2669 char *ext_filename = g_malloc0(PATH_MAX);
2670 char *desc_filename = g_malloc0(PATH_MAX);
2671 char *parent_desc_line = g_malloc0(BUF_SIZE);
2674 VMDKCreateOptsData data;
2675 char *backing_fmt = NULL;
2677 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2678 if (backing_fmt && strcmp(backing_fmt, "vmdk") != 0) {
2679 error_setg(errp, "backing_file must be a vmdk image");
2684 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
2688 /* Read out options */
2689 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2691 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
2692 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2693 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
2694 toolsversion = qemu_opt_get_del(opts, BLOCK_OPT_TOOLSVERSION);
2695 compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false);
2696 if (strcmp(hw_version, "undefined") == 0) {
2700 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
2701 zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false);
2704 adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup,
2706 BLOCKDEV_VMDK_ADAPTER_TYPE_IDE,
2709 error_propagate(errp, local_err);
2714 adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE;
2718 /* Default format to monolithicSparse */
2719 subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE;
2721 subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup,
2723 BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE,
2726 error_propagate(errp, local_err);
2731 data = (VMDKCreateOptsData){
2737 ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum,
2738 backing_file, hw_version, toolsversion, compat6,
2739 zeroed_grain, vmdk_co_create_opts_cb, &data, errp);
2742 g_free(backing_fmt);
2743 g_free(adapter_type);
2744 g_free(backing_file);
2746 g_free(toolsversion);
2753 g_free(ext_filename);
2754 g_free(desc_filename);
2755 g_free(parent_desc_line);
2759 static BlockBackend * coroutine_fn vmdk_co_create_cb(int64_t size, int idx,
2760 bool flat, bool split,
2763 void *opaque, Error **errp)
2766 BlockDriverState *bs;
2768 BlockdevCreateOptionsVmdk *opts = opaque;
2771 bs = bdrv_co_open_blockdev_ref(opts->file, errp);
2774 BlockdevRefList *list = opts->extents;
2775 for (i = 1; i < idx; i++) {
2776 if (!list || !list->next) {
2777 error_setg(errp, "Extent [%d] not specified", i);
2783 error_setg(errp, "Extent [%d] not specified", idx - 1);
2786 bs = bdrv_co_open_blockdev_ref(list->value, errp);
2791 blk = blk_co_new_with_bs(bs,
2792 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
2799 blk_set_allow_write_beyond_eof(blk, true);
2803 ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2812 static int coroutine_fn GRAPH_RDLOCK
2813 vmdk_co_create(BlockdevCreateOptions *create_options, Error **errp)
2815 BlockdevCreateOptionsVmdk *opts;
2817 opts = &create_options->u.vmdk;
2819 /* Validate options */
2820 if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2821 error_setg(errp, "Image size must be a multiple of 512 bytes");
2825 return vmdk_co_do_create(opts->size,
2837 static void vmdk_close(BlockDriverState *bs)
2839 BDRVVmdkState *s = bs->opaque;
2841 vmdk_free_extents(bs);
2842 g_free(s->create_type);
2844 migrate_del_blocker(s->migration_blocker);
2845 error_free(s->migration_blocker);
2848 static int64_t coroutine_fn GRAPH_RDLOCK
2849 vmdk_co_get_allocated_file_size(BlockDriverState *bs)
2854 BDRVVmdkState *s = bs->opaque;
2856 ret = bdrv_co_get_allocated_file_size(bs->file->bs);
2860 for (i = 0; i < s->num_extents; i++) {
2861 if (s->extents[i].file == bs->file) {
2864 r = bdrv_co_get_allocated_file_size(s->extents[i].file->bs);
2873 static int vmdk_has_zero_init(BlockDriverState *bs)
2876 BDRVVmdkState *s = bs->opaque;
2878 /* If has a flat extent and its underlying storage doesn't have zero init,
2880 for (i = 0; i < s->num_extents; i++) {
2881 if (s->extents[i].flat) {
2882 if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2890 static VmdkExtentInfo *vmdk_get_extent_info(VmdkExtent *extent)
2892 VmdkExtentInfo *info = g_new0(VmdkExtentInfo, 1);
2894 bdrv_refresh_filename(extent->file->bs);
2895 *info = (VmdkExtentInfo){
2896 .filename = g_strdup(extent->file->bs->filename),
2897 .format = g_strdup(extent->type),
2898 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2899 .compressed = extent->compressed,
2900 .has_compressed = extent->compressed,
2901 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2902 .has_cluster_size = !extent->flat,
2908 static int coroutine_fn GRAPH_RDLOCK
2909 vmdk_co_check(BlockDriverState *bs, BdrvCheckResult *result, BdrvCheckMode fix)
2911 BDRVVmdkState *s = bs->opaque;
2912 VmdkExtent *extent = NULL;
2913 int64_t sector_num = 0;
2914 int64_t total_sectors = bdrv_nb_sectors(bs);
2916 uint64_t cluster_offset;
2923 if (sector_num >= total_sectors) {
2926 extent = find_extent(s, sector_num, extent);
2929 "ERROR: could not find extent for sector %" PRId64 "\n",
2934 ret = get_cluster_offset(bs, extent, NULL,
2935 sector_num << BDRV_SECTOR_BITS,
2936 false, &cluster_offset, 0, 0);
2937 if (ret == VMDK_ERROR) {
2939 "ERROR: could not get cluster_offset for sector %"
2940 PRId64 "\n", sector_num);
2943 if (ret == VMDK_OK) {
2944 int64_t extent_len = bdrv_getlength(extent->file->bs);
2945 if (extent_len < 0) {
2947 "ERROR: could not get extent file length for sector %"
2948 PRId64 "\n", sector_num);
2952 if (cluster_offset >= extent_len) {
2954 "ERROR: cluster offset for sector %"
2955 PRId64 " points after EOF\n", sector_num);
2960 sector_num += extent->cluster_sectors;
2963 result->corruptions++;
2967 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs,
2971 BDRVVmdkState *s = bs->opaque;
2972 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2973 VmdkExtentInfoList **tail;
2975 *spec_info = (ImageInfoSpecific){
2976 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2978 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2982 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2983 .create_type = g_strdup(s->create_type),
2985 .parent_cid = s->parent_cid,
2988 tail = &spec_info->u.vmdk.data->extents;
2989 for (i = 0; i < s->num_extents; i++) {
2990 QAPI_LIST_APPEND(tail, vmdk_get_extent_info(&s->extents[i]));
2996 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2998 return a->flat == b->flat &&
2999 a->compressed == b->compressed &&
3000 (a->flat || a->cluster_sectors == b->cluster_sectors);
3003 static int coroutine_fn
3004 vmdk_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3007 BDRVVmdkState *s = bs->opaque;
3008 assert(s->num_extents);
3010 /* See if we have multiple extents but they have different cases */
3011 for (i = 1; i < s->num_extents; i++) {
3012 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
3016 bdi->needs_compressed_writes = s->extents[0].compressed;
3017 if (!s->extents[0].flat) {
3018 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
3023 static void vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
3024 bool backing_overridden)
3026 /* No children but file and backing can be explicitly specified (TODO) */
3027 qdict_put(target, "file",
3028 qobject_ref(bs->file->bs->full_open_options));
3030 if (backing_overridden) {
3032 qdict_put(target, "backing",
3033 qobject_ref(bs->backing->bs->full_open_options));
3035 qdict_put_null(target, "backing");
3040 static QemuOptsList vmdk_create_opts = {
3041 .name = "vmdk-create-opts",
3042 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
3045 .name = BLOCK_OPT_SIZE,
3046 .type = QEMU_OPT_SIZE,
3047 .help = "Virtual disk size"
3050 .name = BLOCK_OPT_ADAPTER_TYPE,
3051 .type = QEMU_OPT_STRING,
3052 .help = "Virtual adapter type, can be one of "
3053 "ide (default), lsilogic, buslogic or legacyESX"
3056 .name = BLOCK_OPT_BACKING_FILE,
3057 .type = QEMU_OPT_STRING,
3058 .help = "File name of a base image"
3061 .name = BLOCK_OPT_BACKING_FMT,
3062 .type = QEMU_OPT_STRING,
3063 .help = "Must be 'vmdk' if present",
3066 .name = BLOCK_OPT_COMPAT6,
3067 .type = QEMU_OPT_BOOL,
3068 .help = "VMDK version 6 image",
3069 .def_value_str = "off"
3072 .name = BLOCK_OPT_HWVERSION,
3073 .type = QEMU_OPT_STRING,
3074 .help = "VMDK hardware version",
3075 .def_value_str = "undefined"
3078 .name = BLOCK_OPT_TOOLSVERSION,
3079 .type = QEMU_OPT_STRING,
3080 .help = "VMware guest tools version",
3083 .name = BLOCK_OPT_SUBFMT,
3084 .type = QEMU_OPT_STRING,
3086 "VMDK flat extent format, can be one of "
3087 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
3090 .name = BLOCK_OPT_ZEROED_GRAIN,
3091 .type = QEMU_OPT_BOOL,
3092 .help = "Enable efficient zero writes "
3093 "using the zeroed-grain GTE feature"
3095 { /* end of list */ }
3099 static BlockDriver bdrv_vmdk = {
3100 .format_name = "vmdk",
3101 .instance_size = sizeof(BDRVVmdkState),
3102 .bdrv_probe = vmdk_probe,
3103 .bdrv_open = vmdk_open,
3104 .bdrv_co_check = vmdk_co_check,
3105 .bdrv_reopen_prepare = vmdk_reopen_prepare,
3106 .bdrv_reopen_commit = vmdk_reopen_commit,
3107 .bdrv_reopen_abort = vmdk_reopen_abort,
3108 .bdrv_child_perm = bdrv_default_perms,
3109 .bdrv_co_preadv = vmdk_co_preadv,
3110 .bdrv_co_pwritev = vmdk_co_pwritev,
3111 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
3112 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
3113 .bdrv_close = vmdk_close,
3114 .bdrv_co_create_opts = vmdk_co_create_opts,
3115 .bdrv_co_create = vmdk_co_create,
3116 .bdrv_co_block_status = vmdk_co_block_status,
3117 .bdrv_co_get_allocated_file_size = vmdk_co_get_allocated_file_size,
3118 .bdrv_has_zero_init = vmdk_has_zero_init,
3119 .bdrv_get_specific_info = vmdk_get_specific_info,
3120 .bdrv_refresh_limits = vmdk_refresh_limits,
3121 .bdrv_co_get_info = vmdk_co_get_info,
3122 .bdrv_gather_child_options = vmdk_gather_child_options,
3125 .supports_backing = true,
3126 .create_opts = &vmdk_create_opts,
3129 static void bdrv_vmdk_init(void)
3131 bdrv_register(&bdrv_vmdk);
3134 block_init(bdrv_vmdk_init);