Merge commit '2cc6e0a14210d2e80173ec6b4611e7e3c50c2cce' into upstream-merge
[qemu-kvm.git] / block / vmdk.c
blob8da87acef05148e751e034f7d464c4799efa6f36
1 /*
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
23 * THE SOFTWARE.
26 #include "qemu-common.h"
27 #include "block_int.h"
28 #include "module.h"
30 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
31 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
33 typedef struct {
34 uint32_t version;
35 uint32_t flags;
36 uint32_t disk_sectors;
37 uint32_t granularity;
38 uint32_t l1dir_offset;
39 uint32_t l1dir_size;
40 uint32_t file_sectors;
41 uint32_t cylinders;
42 uint32_t heads;
43 uint32_t sectors_per_track;
44 } VMDK3Header;
46 typedef struct {
47 uint32_t version;
48 uint32_t flags;
49 int64_t capacity;
50 int64_t granularity;
51 int64_t desc_offset;
52 int64_t desc_size;
53 int32_t num_gtes_per_gte;
54 int64_t rgd_offset;
55 int64_t gd_offset;
56 int64_t grain_offset;
57 char filler[1];
58 char check_bytes[4];
59 } __attribute__((packed)) VMDK4Header;
61 #define L2_CACHE_SIZE 16
63 typedef struct VmdkExtent {
64 BlockDriverState *file;
65 bool flat;
66 int64_t sectors;
67 int64_t end_sector;
68 int64_t flat_start_offset;
69 int64_t l1_table_offset;
70 int64_t l1_backup_table_offset;
71 uint32_t *l1_table;
72 uint32_t *l1_backup_table;
73 unsigned int l1_size;
74 uint32_t l1_entry_sectors;
76 unsigned int l2_size;
77 uint32_t *l2_cache;
78 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
79 uint32_t l2_cache_counts[L2_CACHE_SIZE];
81 unsigned int cluster_sectors;
82 } VmdkExtent;
84 typedef struct BDRVVmdkState {
85 int desc_offset;
86 bool cid_updated;
87 uint32_t parent_cid;
88 int num_extents;
89 /* Extent array with num_extents entries, ascend ordered by address */
90 VmdkExtent *extents;
91 } BDRVVmdkState;
93 typedef struct VmdkMetaData {
94 uint32_t offset;
95 unsigned int l1_index;
96 unsigned int l2_index;
97 unsigned int l2_offset;
98 int valid;
99 } VmdkMetaData;
101 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
103 uint32_t magic;
105 if (buf_size < 4) {
106 return 0;
108 magic = be32_to_cpu(*(uint32_t *)buf);
109 if (magic == VMDK3_MAGIC ||
110 magic == VMDK4_MAGIC) {
111 return 100;
112 } else {
113 const char *p = (const char *)buf;
114 const char *end = p + buf_size;
115 while (p < end) {
116 if (*p == '#') {
117 /* skip comment line */
118 while (p < end && *p != '\n') {
119 p++;
121 p++;
122 continue;
124 if (*p == ' ') {
125 while (p < end && *p == ' ') {
126 p++;
128 /* skip '\r' if windows line endings used. */
129 if (p < end && *p == '\r') {
130 p++;
132 /* only accept blank lines before 'version=' line */
133 if (p == end || *p != '\n') {
134 return 0;
136 p++;
137 continue;
139 if (end - p >= strlen("version=X\n")) {
140 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
141 strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
142 return 100;
145 if (end - p >= strlen("version=X\r\n")) {
146 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
147 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
148 return 100;
151 return 0;
153 return 0;
157 #define CHECK_CID 1
159 #define SECTOR_SIZE 512
160 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
161 #define BUF_SIZE 4096
162 #define HEADER_SIZE 512 /* first sector of 512 bytes */
164 static void vmdk_free_extents(BlockDriverState *bs)
166 int i;
167 BDRVVmdkState *s = bs->opaque;
169 for (i = 0; i < s->num_extents; i++) {
170 g_free(s->extents[i].l1_table);
171 g_free(s->extents[i].l2_cache);
172 g_free(s->extents[i].l1_backup_table);
174 g_free(s->extents);
177 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
179 char desc[DESC_SIZE];
180 uint32_t cid;
181 const char *p_name, *cid_str;
182 size_t cid_str_size;
183 BDRVVmdkState *s = bs->opaque;
185 if (bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE) != DESC_SIZE) {
186 return 0;
189 if (parent) {
190 cid_str = "parentCID";
191 cid_str_size = sizeof("parentCID");
192 } else {
193 cid_str = "CID";
194 cid_str_size = sizeof("CID");
197 p_name = strstr(desc, cid_str);
198 if (p_name != NULL) {
199 p_name += cid_str_size;
200 sscanf(p_name, "%x", &cid);
203 return cid;
206 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
208 char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
209 char *p_name, *tmp_str;
210 BDRVVmdkState *s = bs->opaque;
212 memset(desc, 0, sizeof(desc));
213 if (bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE) != DESC_SIZE) {
214 return -EIO;
217 tmp_str = strstr(desc, "parentCID");
218 pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
219 p_name = strstr(desc, "CID");
220 if (p_name != NULL) {
221 p_name += sizeof("CID");
222 snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
223 pstrcat(desc, sizeof(desc), tmp_desc);
226 if (bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE) < 0) {
227 return -EIO;
229 return 0;
232 static int vmdk_is_cid_valid(BlockDriverState *bs)
234 #ifdef CHECK_CID
235 BDRVVmdkState *s = bs->opaque;
236 BlockDriverState *p_bs = bs->backing_hd;
237 uint32_t cur_pcid;
239 if (p_bs) {
240 cur_pcid = vmdk_read_cid(p_bs, 0);
241 if (s->parent_cid != cur_pcid) {
242 /* CID not valid */
243 return 0;
246 #endif
247 /* CID valid */
248 return 1;
251 static int vmdk_parent_open(BlockDriverState *bs)
253 char *p_name;
254 char desc[DESC_SIZE + 1];
255 BDRVVmdkState *s = bs->opaque;
257 desc[DESC_SIZE] = '\0';
258 if (bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE) != DESC_SIZE) {
259 return -1;
262 p_name = strstr(desc, "parentFileNameHint");
263 if (p_name != NULL) {
264 char *end_name;
266 p_name += sizeof("parentFileNameHint") + 1;
267 end_name = strchr(p_name, '\"');
268 if (end_name == NULL) {
269 return -1;
271 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
272 return -1;
275 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
278 return 0;
281 /* Create and append extent to the extent array. Return the added VmdkExtent
282 * address. return NULL if allocation failed. */
283 static VmdkExtent *vmdk_add_extent(BlockDriverState *bs,
284 BlockDriverState *file, bool flat, int64_t sectors,
285 int64_t l1_offset, int64_t l1_backup_offset,
286 uint32_t l1_size,
287 int l2_size, unsigned int cluster_sectors)
289 VmdkExtent *extent;
290 BDRVVmdkState *s = bs->opaque;
292 s->extents = g_realloc(s->extents,
293 (s->num_extents + 1) * sizeof(VmdkExtent));
294 extent = &s->extents[s->num_extents];
295 s->num_extents++;
297 memset(extent, 0, sizeof(VmdkExtent));
298 extent->file = file;
299 extent->flat = flat;
300 extent->sectors = sectors;
301 extent->l1_table_offset = l1_offset;
302 extent->l1_backup_table_offset = l1_backup_offset;
303 extent->l1_size = l1_size;
304 extent->l1_entry_sectors = l2_size * cluster_sectors;
305 extent->l2_size = l2_size;
306 extent->cluster_sectors = cluster_sectors;
308 if (s->num_extents > 1) {
309 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
310 } else {
311 extent->end_sector = extent->sectors;
313 bs->total_sectors = extent->end_sector;
314 return extent;
317 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent)
319 int ret;
320 int l1_size, i;
322 /* read the L1 table */
323 l1_size = extent->l1_size * sizeof(uint32_t);
324 extent->l1_table = g_malloc(l1_size);
325 ret = bdrv_pread(extent->file,
326 extent->l1_table_offset,
327 extent->l1_table,
328 l1_size);
329 if (ret < 0) {
330 goto fail_l1;
332 for (i = 0; i < extent->l1_size; i++) {
333 le32_to_cpus(&extent->l1_table[i]);
336 if (extent->l1_backup_table_offset) {
337 extent->l1_backup_table = g_malloc(l1_size);
338 ret = bdrv_pread(extent->file,
339 extent->l1_backup_table_offset,
340 extent->l1_backup_table,
341 l1_size);
342 if (ret < 0) {
343 goto fail_l1b;
345 for (i = 0; i < extent->l1_size; i++) {
346 le32_to_cpus(&extent->l1_backup_table[i]);
350 extent->l2_cache =
351 g_malloc(extent->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
352 return 0;
353 fail_l1b:
354 g_free(extent->l1_backup_table);
355 fail_l1:
356 g_free(extent->l1_table);
357 return ret;
360 static int vmdk_open_vmdk3(BlockDriverState *bs, int flags)
362 int ret;
363 uint32_t magic;
364 VMDK3Header header;
365 BDRVVmdkState *s = bs->opaque;
366 VmdkExtent *extent;
368 s->desc_offset = 0x200;
369 ret = bdrv_pread(bs->file, sizeof(magic), &header, sizeof(header));
370 if (ret < 0) {
371 goto fail;
373 extent = vmdk_add_extent(bs,
374 bs->file, false,
375 le32_to_cpu(header.disk_sectors),
376 le32_to_cpu(header.l1dir_offset) << 9,
377 0, 1 << 6, 1 << 9,
378 le32_to_cpu(header.granularity));
379 ret = vmdk_init_tables(bs, extent);
380 if (ret) {
381 /* vmdk_init_tables cleans up on fail, so only free allocation of
382 * vmdk_add_extent here. */
383 goto fail;
385 return 0;
386 fail:
387 vmdk_free_extents(bs);
388 return ret;
391 static int vmdk_open_vmdk4(BlockDriverState *bs, int flags)
393 int ret;
394 uint32_t magic;
395 uint32_t l1_size, l1_entry_sectors;
396 VMDK4Header header;
397 BDRVVmdkState *s = bs->opaque;
398 VmdkExtent *extent;
400 s->desc_offset = 0x200;
401 ret = bdrv_pread(bs->file, sizeof(magic), &header, sizeof(header));
402 if (ret < 0) {
403 goto fail;
405 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gte)
406 * le64_to_cpu(header.granularity);
407 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
408 / l1_entry_sectors;
409 extent = vmdk_add_extent(bs, bs->file, false,
410 le64_to_cpu(header.capacity),
411 le64_to_cpu(header.gd_offset) << 9,
412 le64_to_cpu(header.rgd_offset) << 9,
413 l1_size,
414 le32_to_cpu(header.num_gtes_per_gte),
415 le64_to_cpu(header.granularity));
416 if (extent->l1_entry_sectors <= 0) {
417 ret = -EINVAL;
418 goto fail;
420 /* try to open parent images, if exist */
421 ret = vmdk_parent_open(bs);
422 if (ret) {
423 goto fail;
425 s->parent_cid = vmdk_read_cid(bs, 1);
426 ret = vmdk_init_tables(bs, extent);
427 if (ret) {
428 goto fail;
430 return 0;
431 fail:
432 vmdk_free_extents(bs);
433 return ret;
436 /* find an option value out of descriptor file */
437 static int vmdk_parse_description(const char *desc, const char *opt_name,
438 char *buf, int buf_size)
440 char *opt_pos, *opt_end;
441 const char *end = desc + strlen(desc);
443 opt_pos = strstr(desc, opt_name);
444 if (!opt_pos) {
445 return -1;
447 /* Skip "=\"" following opt_name */
448 opt_pos += strlen(opt_name) + 2;
449 if (opt_pos >= end) {
450 return -1;
452 opt_end = opt_pos;
453 while (opt_end < end && *opt_end != '"') {
454 opt_end++;
456 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
457 return -1;
459 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
460 return 0;
463 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
464 const char *desc_file_path)
466 int ret;
467 char access[11];
468 char type[11];
469 char fname[512];
470 const char *p = desc;
471 int64_t sectors = 0;
472 int64_t flat_offset;
474 while (*p) {
475 /* parse extent line:
476 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
477 * or
478 * RW [size in sectors] SPARSE "file-name.vmdk"
480 flat_offset = -1;
481 ret = sscanf(p, "%10s %" SCNd64 " %10s %511s %" SCNd64,
482 access, &sectors, type, fname, &flat_offset);
483 if (ret < 4 || strcmp(access, "RW")) {
484 goto next_line;
485 } else if (!strcmp(type, "FLAT")) {
486 if (ret != 5 || flat_offset < 0) {
487 return -EINVAL;
489 } else if (ret != 4) {
490 return -EINVAL;
493 /* trim the quotation marks around */
494 if (fname[0] == '"') {
495 memmove(fname, fname + 1, strlen(fname));
496 if (strlen(fname) <= 1 || fname[strlen(fname) - 1] != '"') {
497 return -EINVAL;
499 fname[strlen(fname) - 1] = '\0';
501 if (sectors <= 0 ||
502 (strcmp(type, "FLAT") && strcmp(type, "SPARSE")) ||
503 (strcmp(access, "RW"))) {
504 goto next_line;
507 /* save to extents array */
508 if (!strcmp(type, "FLAT")) {
509 /* FLAT extent */
510 char extent_path[PATH_MAX];
511 BlockDriverState *extent_file;
512 VmdkExtent *extent;
514 path_combine(extent_path, sizeof(extent_path),
515 desc_file_path, fname);
516 ret = bdrv_file_open(&extent_file, extent_path, bs->open_flags);
517 if (ret) {
518 return ret;
520 extent = vmdk_add_extent(bs, extent_file, true, sectors,
521 0, 0, 0, 0, sectors);
522 extent->flat_start_offset = flat_offset;
523 } else {
524 /* SPARSE extent, not supported for now */
525 fprintf(stderr,
526 "VMDK: Not supported extent type \"%s\""".\n", type);
527 return -ENOTSUP;
529 next_line:
530 /* move to next line */
531 while (*p && *p != '\n') {
532 p++;
534 p++;
536 return 0;
539 static int vmdk_open_desc_file(BlockDriverState *bs, int flags)
541 int ret;
542 char buf[2048];
543 char ct[128];
544 BDRVVmdkState *s = bs->opaque;
546 ret = bdrv_pread(bs->file, 0, buf, sizeof(buf));
547 if (ret < 0) {
548 return ret;
550 buf[2047] = '\0';
551 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
552 return -EINVAL;
554 if (strcmp(ct, "monolithicFlat")) {
555 fprintf(stderr,
556 "VMDK: Not supported image type \"%s\""".\n", ct);
557 return -ENOTSUP;
559 s->desc_offset = 0;
560 ret = vmdk_parse_extents(buf, bs, bs->file->filename);
561 if (ret) {
562 return ret;
565 /* try to open parent images, if exist */
566 if (vmdk_parent_open(bs)) {
567 g_free(s->extents);
568 return -EINVAL;
570 s->parent_cid = vmdk_read_cid(bs, 1);
571 return 0;
574 static int vmdk_open(BlockDriverState *bs, int flags)
576 uint32_t magic;
578 if (bdrv_pread(bs->file, 0, &magic, sizeof(magic)) != sizeof(magic)) {
579 return -EIO;
582 magic = be32_to_cpu(magic);
583 if (magic == VMDK3_MAGIC) {
584 return vmdk_open_vmdk3(bs, flags);
585 } else if (magic == VMDK4_MAGIC) {
586 return vmdk_open_vmdk4(bs, flags);
587 } else {
588 return vmdk_open_desc_file(bs, flags);
592 static int get_whole_cluster(BlockDriverState *bs,
593 VmdkExtent *extent,
594 uint64_t cluster_offset,
595 uint64_t offset,
596 bool allocate)
598 /* 128 sectors * 512 bytes each = grain size 64KB */
599 uint8_t whole_grain[extent->cluster_sectors * 512];
601 /* we will be here if it's first write on non-exist grain(cluster).
602 * try to read from parent image, if exist */
603 if (bs->backing_hd) {
604 int ret;
606 if (!vmdk_is_cid_valid(bs)) {
607 return -1;
610 /* floor offset to cluster */
611 offset -= offset % (extent->cluster_sectors * 512);
612 ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain,
613 extent->cluster_sectors);
614 if (ret < 0) {
615 return -1;
618 /* Write grain only into the active image */
619 ret = bdrv_write(extent->file, cluster_offset, whole_grain,
620 extent->cluster_sectors);
621 if (ret < 0) {
622 return -1;
625 return 0;
628 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data)
630 /* update L2 table */
631 if (bdrv_pwrite_sync(
632 extent->file,
633 ((int64_t)m_data->l2_offset * 512)
634 + (m_data->l2_index * sizeof(m_data->offset)),
635 &(m_data->offset),
636 sizeof(m_data->offset)
637 ) < 0) {
638 return -1;
640 /* update backup L2 table */
641 if (extent->l1_backup_table_offset != 0) {
642 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
643 if (bdrv_pwrite_sync(
644 extent->file,
645 ((int64_t)m_data->l2_offset * 512)
646 + (m_data->l2_index * sizeof(m_data->offset)),
647 &(m_data->offset), sizeof(m_data->offset)
648 ) < 0) {
649 return -1;
653 return 0;
656 static int get_cluster_offset(BlockDriverState *bs,
657 VmdkExtent *extent,
658 VmdkMetaData *m_data,
659 uint64_t offset,
660 int allocate,
661 uint64_t *cluster_offset)
663 unsigned int l1_index, l2_offset, l2_index;
664 int min_index, i, j;
665 uint32_t min_count, *l2_table, tmp = 0;
667 if (m_data) {
668 m_data->valid = 0;
670 if (extent->flat) {
671 *cluster_offset = extent->flat_start_offset;
672 return 0;
675 l1_index = (offset >> 9) / extent->l1_entry_sectors;
676 if (l1_index >= extent->l1_size) {
677 return -1;
679 l2_offset = extent->l1_table[l1_index];
680 if (!l2_offset) {
681 return -1;
683 for (i = 0; i < L2_CACHE_SIZE; i++) {
684 if (l2_offset == extent->l2_cache_offsets[i]) {
685 /* increment the hit count */
686 if (++extent->l2_cache_counts[i] == 0xffffffff) {
687 for (j = 0; j < L2_CACHE_SIZE; j++) {
688 extent->l2_cache_counts[j] >>= 1;
691 l2_table = extent->l2_cache + (i * extent->l2_size);
692 goto found;
695 /* not found: load a new entry in the least used one */
696 min_index = 0;
697 min_count = 0xffffffff;
698 for (i = 0; i < L2_CACHE_SIZE; i++) {
699 if (extent->l2_cache_counts[i] < min_count) {
700 min_count = extent->l2_cache_counts[i];
701 min_index = i;
704 l2_table = extent->l2_cache + (min_index * extent->l2_size);
705 if (bdrv_pread(
706 extent->file,
707 (int64_t)l2_offset * 512,
708 l2_table,
709 extent->l2_size * sizeof(uint32_t)
710 ) != extent->l2_size * sizeof(uint32_t)) {
711 return -1;
714 extent->l2_cache_offsets[min_index] = l2_offset;
715 extent->l2_cache_counts[min_index] = 1;
716 found:
717 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
718 *cluster_offset = le32_to_cpu(l2_table[l2_index]);
720 if (!*cluster_offset) {
721 if (!allocate) {
722 return -1;
725 /* Avoid the L2 tables update for the images that have snapshots. */
726 *cluster_offset = bdrv_getlength(extent->file);
727 bdrv_truncate(
728 extent->file,
729 *cluster_offset + (extent->cluster_sectors << 9)
732 *cluster_offset >>= 9;
733 tmp = cpu_to_le32(*cluster_offset);
734 l2_table[l2_index] = tmp;
736 /* First of all we write grain itself, to avoid race condition
737 * that may to corrupt the image.
738 * This problem may occur because of insufficient space on host disk
739 * or inappropriate VM shutdown.
741 if (get_whole_cluster(
742 bs, extent, *cluster_offset, offset, allocate) == -1) {
743 return -1;
746 if (m_data) {
747 m_data->offset = tmp;
748 m_data->l1_index = l1_index;
749 m_data->l2_index = l2_index;
750 m_data->l2_offset = l2_offset;
751 m_data->valid = 1;
754 *cluster_offset <<= 9;
755 return 0;
758 static VmdkExtent *find_extent(BDRVVmdkState *s,
759 int64_t sector_num, VmdkExtent *start_hint)
761 VmdkExtent *extent = start_hint;
763 if (!extent) {
764 extent = &s->extents[0];
766 while (extent < &s->extents[s->num_extents]) {
767 if (sector_num < extent->end_sector) {
768 return extent;
770 extent++;
772 return NULL;
775 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
776 int nb_sectors, int *pnum)
778 BDRVVmdkState *s = bs->opaque;
779 int64_t index_in_cluster, n, ret;
780 uint64_t offset;
781 VmdkExtent *extent;
783 extent = find_extent(s, sector_num, NULL);
784 if (!extent) {
785 return 0;
787 ret = get_cluster_offset(bs, extent, NULL,
788 sector_num * 512, 0, &offset);
789 /* get_cluster_offset returning 0 means success */
790 ret = !ret;
792 index_in_cluster = sector_num % extent->cluster_sectors;
793 n = extent->cluster_sectors - index_in_cluster;
794 if (n > nb_sectors) {
795 n = nb_sectors;
797 *pnum = n;
798 return ret;
801 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
802 uint8_t *buf, int nb_sectors)
804 BDRVVmdkState *s = bs->opaque;
805 int ret;
806 uint64_t n, index_in_cluster;
807 VmdkExtent *extent = NULL;
808 uint64_t cluster_offset;
810 while (nb_sectors > 0) {
811 extent = find_extent(s, sector_num, extent);
812 if (!extent) {
813 return -EIO;
815 ret = get_cluster_offset(
816 bs, extent, NULL,
817 sector_num << 9, 0, &cluster_offset);
818 index_in_cluster = sector_num % extent->cluster_sectors;
819 n = extent->cluster_sectors - index_in_cluster;
820 if (n > nb_sectors) {
821 n = nb_sectors;
823 if (ret) {
824 /* if not allocated, try to read from parent image, if exist */
825 if (bs->backing_hd) {
826 if (!vmdk_is_cid_valid(bs)) {
827 return -EINVAL;
829 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
830 if (ret < 0) {
831 return ret;
833 } else {
834 memset(buf, 0, 512 * n);
836 } else {
837 ret = bdrv_pread(extent->file,
838 cluster_offset + index_in_cluster * 512,
839 buf, n * 512);
840 if (ret < 0) {
841 return ret;
844 nb_sectors -= n;
845 sector_num += n;
846 buf += n * 512;
848 return 0;
851 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
852 const uint8_t *buf, int nb_sectors)
854 BDRVVmdkState *s = bs->opaque;
855 VmdkExtent *extent = NULL;
856 int n, ret;
857 int64_t index_in_cluster;
858 uint64_t cluster_offset;
859 VmdkMetaData m_data;
861 if (sector_num > bs->total_sectors) {
862 fprintf(stderr,
863 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
864 " total_sectors=0x%" PRIx64 "\n",
865 sector_num, bs->total_sectors);
866 return -EIO;
869 while (nb_sectors > 0) {
870 extent = find_extent(s, sector_num, extent);
871 if (!extent) {
872 return -EIO;
874 ret = get_cluster_offset(
876 extent,
877 &m_data,
878 sector_num << 9, 1,
879 &cluster_offset);
880 if (ret) {
881 return -EINVAL;
883 index_in_cluster = sector_num % extent->cluster_sectors;
884 n = extent->cluster_sectors - index_in_cluster;
885 if (n > nb_sectors) {
886 n = nb_sectors;
889 ret = bdrv_pwrite(extent->file,
890 cluster_offset + index_in_cluster * 512,
891 buf,
892 n * 512);
893 if (ret < 0) {
894 return ret;
896 if (m_data.valid) {
897 /* update L2 tables */
898 if (vmdk_L2update(extent, &m_data) == -1) {
899 return -EIO;
902 nb_sectors -= n;
903 sector_num += n;
904 buf += n * 512;
906 /* update CID on the first write every time the virtual disk is
907 * opened */
908 if (!s->cid_updated) {
909 vmdk_write_cid(bs, time(NULL));
910 s->cid_updated = true;
913 return 0;
917 static int vmdk_create_extent(const char *filename, int64_t filesize, bool flat)
919 int ret, i;
920 int fd = 0;
921 VMDK4Header header;
922 uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
924 fd = open(
925 filename,
926 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
927 0644);
928 if (fd < 0) {
929 return -errno;
931 if (flat) {
932 ret = ftruncate(fd, filesize);
933 if (ret < 0) {
934 ret = -errno;
936 goto exit;
938 magic = cpu_to_be32(VMDK4_MAGIC);
939 memset(&header, 0, sizeof(header));
940 header.version = 1;
941 header.flags = 3; /* ?? */
942 header.capacity = filesize / 512;
943 header.granularity = 128;
944 header.num_gtes_per_gte = 512;
946 grains = (filesize / 512 + header.granularity - 1) / header.granularity;
947 gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
948 gt_count =
949 (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
950 gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
952 header.desc_offset = 1;
953 header.desc_size = 20;
954 header.rgd_offset = header.desc_offset + header.desc_size;
955 header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
956 header.grain_offset =
957 ((header.gd_offset + gd_size + (gt_size * gt_count) +
958 header.granularity - 1) / header.granularity) *
959 header.granularity;
960 /* swap endianness for all header fields */
961 header.version = cpu_to_le32(header.version);
962 header.flags = cpu_to_le32(header.flags);
963 header.capacity = cpu_to_le64(header.capacity);
964 header.granularity = cpu_to_le64(header.granularity);
965 header.num_gtes_per_gte = cpu_to_le32(header.num_gtes_per_gte);
966 header.desc_offset = cpu_to_le64(header.desc_offset);
967 header.desc_size = cpu_to_le64(header.desc_size);
968 header.rgd_offset = cpu_to_le64(header.rgd_offset);
969 header.gd_offset = cpu_to_le64(header.gd_offset);
970 header.grain_offset = cpu_to_le64(header.grain_offset);
972 header.check_bytes[0] = 0xa;
973 header.check_bytes[1] = 0x20;
974 header.check_bytes[2] = 0xd;
975 header.check_bytes[3] = 0xa;
977 /* write all the data */
978 ret = qemu_write_full(fd, &magic, sizeof(magic));
979 if (ret != sizeof(magic)) {
980 ret = -errno;
981 goto exit;
983 ret = qemu_write_full(fd, &header, sizeof(header));
984 if (ret != sizeof(header)) {
985 ret = -errno;
986 goto exit;
989 ret = ftruncate(fd, le64_to_cpu(header.grain_offset) << 9);
990 if (ret < 0) {
991 ret = -errno;
992 goto exit;
995 /* write grain directory */
996 lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
997 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_size;
998 i < gt_count; i++, tmp += gt_size) {
999 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1000 if (ret != sizeof(tmp)) {
1001 ret = -errno;
1002 goto exit;
1006 /* write backup grain directory */
1007 lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
1008 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_size;
1009 i < gt_count; i++, tmp += gt_size) {
1010 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1011 if (ret != sizeof(tmp)) {
1012 ret = -errno;
1013 goto exit;
1017 ret = 0;
1018 exit:
1019 close(fd);
1020 return ret;
1023 static int filename_decompose(const char *filename, char *path, char *prefix,
1024 char *postfix, size_t buf_len)
1026 const char *p, *q;
1028 if (filename == NULL || !strlen(filename)) {
1029 fprintf(stderr, "Vmdk: no filename provided.\n");
1030 return -1;
1032 p = strrchr(filename, '/');
1033 if (p == NULL) {
1034 p = strrchr(filename, '\\');
1036 if (p == NULL) {
1037 p = strrchr(filename, ':');
1039 if (p != NULL) {
1040 p++;
1041 if (p - filename >= buf_len) {
1042 return -1;
1044 pstrcpy(path, p - filename + 1, filename);
1045 } else {
1046 p = filename;
1047 path[0] = '\0';
1049 q = strrchr(p, '.');
1050 if (q == NULL) {
1051 pstrcpy(prefix, buf_len, p);
1052 postfix[0] = '\0';
1053 } else {
1054 if (q - p >= buf_len) {
1055 return -1;
1057 pstrcpy(prefix, q - p + 1, p);
1058 pstrcpy(postfix, buf_len, q);
1060 return 0;
1063 static int relative_path(char *dest, int dest_size,
1064 const char *base, const char *target)
1066 int i = 0;
1067 int n = 0;
1068 const char *p, *q;
1069 #ifdef _WIN32
1070 const char *sep = "\\";
1071 #else
1072 const char *sep = "/";
1073 #endif
1075 if (!(dest && base && target)) {
1076 return -1;
1078 if (path_is_absolute(target)) {
1079 dest[dest_size - 1] = '\0';
1080 strncpy(dest, target, dest_size - 1);
1081 return 0;
1083 while (base[i] == target[i]) {
1084 i++;
1086 p = &base[i];
1087 q = &target[i];
1088 while (*p) {
1089 if (*p == *sep) {
1090 n++;
1092 p++;
1094 dest[0] = '\0';
1095 for (; n; n--) {
1096 pstrcat(dest, dest_size, "..");
1097 pstrcat(dest, dest_size, sep);
1099 pstrcat(dest, dest_size, q);
1100 return 0;
1103 static int vmdk_create(const char *filename, QEMUOptionParameter *options)
1105 int fd, idx = 0;
1106 char desc[BUF_SIZE];
1107 int64_t total_size = 0, filesize;
1108 const char *backing_file = NULL;
1109 const char *fmt = NULL;
1110 int flags = 0;
1111 int ret = 0;
1112 bool flat, split;
1113 char ext_desc_lines[BUF_SIZE] = "";
1114 char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX];
1115 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
1116 const char *desc_extent_line;
1117 char parent_desc_line[BUF_SIZE] = "";
1118 uint32_t parent_cid = 0xffffffff;
1119 const char desc_template[] =
1120 "# Disk DescriptorFile\n"
1121 "version=1\n"
1122 "CID=%x\n"
1123 "parentCID=%x\n"
1124 "createType=\"%s\"\n"
1125 "%s"
1126 "\n"
1127 "# Extent description\n"
1128 "%s"
1129 "\n"
1130 "# The Disk Data Base\n"
1131 "#DDB\n"
1132 "\n"
1133 "ddb.virtualHWVersion = \"%d\"\n"
1134 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1135 "ddb.geometry.heads = \"16\"\n"
1136 "ddb.geometry.sectors = \"63\"\n"
1137 "ddb.adapterType = \"ide\"\n";
1139 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX)) {
1140 return -EINVAL;
1142 /* Read out options */
1143 while (options && options->name) {
1144 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1145 total_size = options->value.n;
1146 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1147 backing_file = options->value.s;
1148 } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) {
1149 flags |= options->value.n ? BLOCK_FLAG_COMPAT6 : 0;
1150 } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) {
1151 fmt = options->value.s;
1153 options++;
1155 if (!fmt) {
1156 /* Default format to monolithicSparse */
1157 fmt = "monolithicSparse";
1158 } else if (strcmp(fmt, "monolithicFlat") &&
1159 strcmp(fmt, "monolithicSparse") &&
1160 strcmp(fmt, "twoGbMaxExtentSparse") &&
1161 strcmp(fmt, "twoGbMaxExtentFlat")) {
1162 fprintf(stderr, "VMDK: Unknown subformat: %s\n", fmt);
1163 return -EINVAL;
1165 split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1166 strcmp(fmt, "twoGbMaxExtentSparse"));
1167 flat = !(strcmp(fmt, "monolithicFlat") &&
1168 strcmp(fmt, "twoGbMaxExtentFlat"));
1169 if (flat) {
1170 desc_extent_line = "RW %lld FLAT \"%s\" 0\n";
1171 } else {
1172 desc_extent_line = "RW %lld SPARSE \"%s\"\n";
1174 if (flat && backing_file) {
1175 /* not supporting backing file for flat image */
1176 return -ENOTSUP;
1178 if (backing_file) {
1179 char parent_filename[PATH_MAX];
1180 BlockDriverState *bs = bdrv_new("");
1181 ret = bdrv_open(bs, backing_file, 0, NULL);
1182 if (ret != 0) {
1183 bdrv_delete(bs);
1184 return ret;
1186 if (strcmp(bs->drv->format_name, "vmdk")) {
1187 bdrv_delete(bs);
1188 return -EINVAL;
1190 filesize = bdrv_getlength(bs);
1191 parent_cid = vmdk_read_cid(bs, 0);
1192 bdrv_delete(bs);
1193 relative_path(parent_filename, sizeof(parent_filename),
1194 filename, backing_file);
1195 snprintf(parent_desc_line, sizeof(parent_desc_line),
1196 "parentFileNameHint=\"%s\"", parent_filename);
1199 /* Create extents */
1200 filesize = total_size;
1201 while (filesize > 0) {
1202 char desc_line[BUF_SIZE];
1203 char ext_filename[PATH_MAX];
1204 char desc_filename[PATH_MAX];
1205 int64_t size = filesize;
1207 if (split && size > split_size) {
1208 size = split_size;
1210 if (split) {
1211 snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s",
1212 prefix, flat ? 'f' : 's', ++idx, postfix);
1213 } else if (flat) {
1214 snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s",
1215 prefix, postfix);
1216 } else {
1217 snprintf(desc_filename, sizeof(desc_filename), "%s%s",
1218 prefix, postfix);
1220 snprintf(ext_filename, sizeof(ext_filename), "%s%s",
1221 path, desc_filename);
1223 if (vmdk_create_extent(ext_filename, size, flat)) {
1224 return -EINVAL;
1226 filesize -= size;
1228 /* Format description line */
1229 snprintf(desc_line, sizeof(desc_line),
1230 desc_extent_line, size / 512, desc_filename);
1231 pstrcat(ext_desc_lines, sizeof(ext_desc_lines), desc_line);
1233 /* generate descriptor file */
1234 snprintf(desc, sizeof(desc), desc_template,
1235 (unsigned int)time(NULL),
1236 parent_cid,
1237 fmt,
1238 parent_desc_line,
1239 ext_desc_lines,
1240 (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
1241 total_size / (int64_t)(63 * 16 * 512));
1242 if (split || flat) {
1243 fd = open(
1244 filename,
1245 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
1246 0644);
1247 } else {
1248 fd = open(
1249 filename,
1250 O_WRONLY | O_BINARY | O_LARGEFILE,
1251 0644);
1253 if (fd < 0) {
1254 return -errno;
1256 /* the descriptor offset = 0x200 */
1257 if (!split && !flat && 0x200 != lseek(fd, 0x200, SEEK_SET)) {
1258 ret = -errno;
1259 goto exit;
1261 ret = qemu_write_full(fd, desc, strlen(desc));
1262 if (ret != strlen(desc)) {
1263 ret = -errno;
1264 goto exit;
1266 ret = 0;
1267 exit:
1268 close(fd);
1269 return ret;
1272 static void vmdk_close(BlockDriverState *bs)
1274 vmdk_free_extents(bs);
1277 static int vmdk_flush(BlockDriverState *bs)
1279 int i, ret, err;
1280 BDRVVmdkState *s = bs->opaque;
1282 ret = bdrv_flush(bs->file);
1283 for (i = 0; i < s->num_extents; i++) {
1284 err = bdrv_flush(s->extents[i].file);
1285 if (err < 0) {
1286 ret = err;
1289 return ret;
1292 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
1294 int i;
1295 int64_t ret = 0;
1296 int64_t r;
1297 BDRVVmdkState *s = bs->opaque;
1299 ret = bdrv_get_allocated_file_size(bs->file);
1300 if (ret < 0) {
1301 return ret;
1303 for (i = 0; i < s->num_extents; i++) {
1304 if (s->extents[i].file == bs->file) {
1305 continue;
1307 r = bdrv_get_allocated_file_size(s->extents[i].file);
1308 if (r < 0) {
1309 return r;
1311 ret += r;
1313 return ret;
1316 static QEMUOptionParameter vmdk_create_options[] = {
1318 .name = BLOCK_OPT_SIZE,
1319 .type = OPT_SIZE,
1320 .help = "Virtual disk size"
1323 .name = BLOCK_OPT_BACKING_FILE,
1324 .type = OPT_STRING,
1325 .help = "File name of a base image"
1328 .name = BLOCK_OPT_COMPAT6,
1329 .type = OPT_FLAG,
1330 .help = "VMDK version 6 image"
1333 .name = BLOCK_OPT_SUBFMT,
1334 .type = OPT_STRING,
1335 .help =
1336 "VMDK flat extent format, can be one of "
1337 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat} "
1339 { NULL }
1342 static BlockDriver bdrv_vmdk = {
1343 .format_name = "vmdk",
1344 .instance_size = sizeof(BDRVVmdkState),
1345 .bdrv_probe = vmdk_probe,
1346 .bdrv_open = vmdk_open,
1347 .bdrv_read = vmdk_read,
1348 .bdrv_write = vmdk_write,
1349 .bdrv_close = vmdk_close,
1350 .bdrv_create = vmdk_create,
1351 .bdrv_flush = vmdk_flush,
1352 .bdrv_is_allocated = vmdk_is_allocated,
1353 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
1355 .create_options = vmdk_create_options,
1358 static void bdrv_vmdk_init(void)
1360 bdrv_register(&bdrv_vmdk);
1363 block_init(bdrv_vmdk_init);