]> git.proxmox.com Git - qemu.git/blob - block/vmdk.c
qmp: fix handling of cmd with Equals in qmp-shell
[qemu.git] / block / vmdk.c
1 /*
2 * Block driver for the VMDK format
3 *
4 * Copyright (c) 2004 Fabrice Bellard
5 * Copyright (c) 2005 Filip Navara
6 *
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:
13 *
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
16 *
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.
24 */
25
26 #include "qemu-common.h"
27 #include "block/block_int.h"
28 #include "qemu/module.h"
29 #include "migration/migration.h"
30 #include <zlib.h>
31
32 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
33 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
34 #define VMDK4_COMPRESSION_DEFLATE 1
35 #define VMDK4_FLAG_NL_DETECT (1 << 0)
36 #define VMDK4_FLAG_RGD (1 << 1)
37 /* Zeroed-grain enable bit */
38 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
39 #define VMDK4_FLAG_COMPRESS (1 << 16)
40 #define VMDK4_FLAG_MARKER (1 << 17)
41 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
42
43 #define VMDK_GTE_ZEROED 0x1
44
45 /* VMDK internal error codes */
46 #define VMDK_OK 0
47 #define VMDK_ERROR (-1)
48 /* Cluster not allocated */
49 #define VMDK_UNALLOC (-2)
50 #define VMDK_ZEROED (-3)
51
52 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
53
54 typedef struct {
55 uint32_t version;
56 uint32_t flags;
57 uint32_t disk_sectors;
58 uint32_t granularity;
59 uint32_t l1dir_offset;
60 uint32_t l1dir_size;
61 uint32_t file_sectors;
62 uint32_t cylinders;
63 uint32_t heads;
64 uint32_t sectors_per_track;
65 } VMDK3Header;
66
67 typedef struct {
68 uint32_t version;
69 uint32_t flags;
70 int64_t capacity;
71 int64_t granularity;
72 int64_t desc_offset;
73 int64_t desc_size;
74 int32_t num_gtes_per_gte;
75 int64_t rgd_offset;
76 int64_t gd_offset;
77 int64_t grain_offset;
78 char filler[1];
79 char check_bytes[4];
80 uint16_t compressAlgorithm;
81 } QEMU_PACKED VMDK4Header;
82
83 #define L2_CACHE_SIZE 16
84
85 typedef struct VmdkExtent {
86 BlockDriverState *file;
87 bool flat;
88 bool compressed;
89 bool has_marker;
90 bool has_zero_grain;
91 int version;
92 int64_t sectors;
93 int64_t end_sector;
94 int64_t flat_start_offset;
95 int64_t l1_table_offset;
96 int64_t l1_backup_table_offset;
97 uint32_t *l1_table;
98 uint32_t *l1_backup_table;
99 unsigned int l1_size;
100 uint32_t l1_entry_sectors;
101
102 unsigned int l2_size;
103 uint32_t *l2_cache;
104 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
105 uint32_t l2_cache_counts[L2_CACHE_SIZE];
106
107 unsigned int cluster_sectors;
108 } VmdkExtent;
109
110 typedef struct BDRVVmdkState {
111 CoMutex lock;
112 int desc_offset;
113 bool cid_updated;
114 uint32_t parent_cid;
115 int num_extents;
116 /* Extent array with num_extents entries, ascend ordered by address */
117 VmdkExtent *extents;
118 Error *migration_blocker;
119 } BDRVVmdkState;
120
121 typedef struct VmdkMetaData {
122 uint32_t offset;
123 unsigned int l1_index;
124 unsigned int l2_index;
125 unsigned int l2_offset;
126 int valid;
127 uint32_t *l2_cache_entry;
128 } VmdkMetaData;
129
130 typedef struct VmdkGrainMarker {
131 uint64_t lba;
132 uint32_t size;
133 uint8_t data[0];
134 } VmdkGrainMarker;
135
136 enum {
137 MARKER_END_OF_STREAM = 0,
138 MARKER_GRAIN_TABLE = 1,
139 MARKER_GRAIN_DIRECTORY = 2,
140 MARKER_FOOTER = 3,
141 };
142
143 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
144 {
145 uint32_t magic;
146
147 if (buf_size < 4) {
148 return 0;
149 }
150 magic = be32_to_cpu(*(uint32_t *)buf);
151 if (magic == VMDK3_MAGIC ||
152 magic == VMDK4_MAGIC) {
153 return 100;
154 } else {
155 const char *p = (const char *)buf;
156 const char *end = p + buf_size;
157 while (p < end) {
158 if (*p == '#') {
159 /* skip comment line */
160 while (p < end && *p != '\n') {
161 p++;
162 }
163 p++;
164 continue;
165 }
166 if (*p == ' ') {
167 while (p < end && *p == ' ') {
168 p++;
169 }
170 /* skip '\r' if windows line endings used. */
171 if (p < end && *p == '\r') {
172 p++;
173 }
174 /* only accept blank lines before 'version=' line */
175 if (p == end || *p != '\n') {
176 return 0;
177 }
178 p++;
179 continue;
180 }
181 if (end - p >= strlen("version=X\n")) {
182 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
183 strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
184 return 100;
185 }
186 }
187 if (end - p >= strlen("version=X\r\n")) {
188 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
189 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
190 return 100;
191 }
192 }
193 return 0;
194 }
195 return 0;
196 }
197 }
198
199 #define CHECK_CID 1
200
201 #define SECTOR_SIZE 512
202 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
203 #define BUF_SIZE 4096
204 #define HEADER_SIZE 512 /* first sector of 512 bytes */
205
206 static void vmdk_free_extents(BlockDriverState *bs)
207 {
208 int i;
209 BDRVVmdkState *s = bs->opaque;
210 VmdkExtent *e;
211
212 for (i = 0; i < s->num_extents; i++) {
213 e = &s->extents[i];
214 g_free(e->l1_table);
215 g_free(e->l2_cache);
216 g_free(e->l1_backup_table);
217 if (e->file != bs->file) {
218 bdrv_delete(e->file);
219 }
220 }
221 g_free(s->extents);
222 }
223
224 static void vmdk_free_last_extent(BlockDriverState *bs)
225 {
226 BDRVVmdkState *s = bs->opaque;
227
228 if (s->num_extents == 0) {
229 return;
230 }
231 s->num_extents--;
232 s->extents = g_realloc(s->extents, s->num_extents * sizeof(VmdkExtent));
233 }
234
235 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
236 {
237 char desc[DESC_SIZE];
238 uint32_t cid = 0xffffffff;
239 const char *p_name, *cid_str;
240 size_t cid_str_size;
241 BDRVVmdkState *s = bs->opaque;
242 int ret;
243
244 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
245 if (ret < 0) {
246 return 0;
247 }
248
249 if (parent) {
250 cid_str = "parentCID";
251 cid_str_size = sizeof("parentCID");
252 } else {
253 cid_str = "CID";
254 cid_str_size = sizeof("CID");
255 }
256
257 desc[DESC_SIZE - 1] = '\0';
258 p_name = strstr(desc, cid_str);
259 if (p_name != NULL) {
260 p_name += cid_str_size;
261 sscanf(p_name, "%x", &cid);
262 }
263
264 return cid;
265 }
266
267 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
268 {
269 char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
270 char *p_name, *tmp_str;
271 BDRVVmdkState *s = bs->opaque;
272 int ret;
273
274 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
275 if (ret < 0) {
276 return ret;
277 }
278
279 desc[DESC_SIZE - 1] = '\0';
280 tmp_str = strstr(desc, "parentCID");
281 if (tmp_str == NULL) {
282 return -EINVAL;
283 }
284
285 pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
286 p_name = strstr(desc, "CID");
287 if (p_name != NULL) {
288 p_name += sizeof("CID");
289 snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
290 pstrcat(desc, sizeof(desc), tmp_desc);
291 }
292
293 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
294 if (ret < 0) {
295 return ret;
296 }
297
298 return 0;
299 }
300
301 static int vmdk_is_cid_valid(BlockDriverState *bs)
302 {
303 #ifdef CHECK_CID
304 BDRVVmdkState *s = bs->opaque;
305 BlockDriverState *p_bs = bs->backing_hd;
306 uint32_t cur_pcid;
307
308 if (p_bs) {
309 cur_pcid = vmdk_read_cid(p_bs, 0);
310 if (s->parent_cid != cur_pcid) {
311 /* CID not valid */
312 return 0;
313 }
314 }
315 #endif
316 /* CID valid */
317 return 1;
318 }
319
320 /* Queue extents, if any, for reopen() */
321 static int vmdk_reopen_prepare(BDRVReopenState *state,
322 BlockReopenQueue *queue, Error **errp)
323 {
324 BDRVVmdkState *s;
325 int ret = -1;
326 int i;
327 VmdkExtent *e;
328
329 assert(state != NULL);
330 assert(state->bs != NULL);
331
332 if (queue == NULL) {
333 error_set(errp, ERROR_CLASS_GENERIC_ERROR,
334 "No reopen queue for VMDK extents");
335 goto exit;
336 }
337
338 s = state->bs->opaque;
339
340 assert(s != NULL);
341
342 for (i = 0; i < s->num_extents; i++) {
343 e = &s->extents[i];
344 if (e->file != state->bs->file) {
345 bdrv_reopen_queue(queue, e->file, state->flags);
346 }
347 }
348 ret = 0;
349
350 exit:
351 return ret;
352 }
353
354 static int vmdk_parent_open(BlockDriverState *bs)
355 {
356 char *p_name;
357 char desc[DESC_SIZE + 1];
358 BDRVVmdkState *s = bs->opaque;
359 int ret;
360
361 desc[DESC_SIZE] = '\0';
362 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
363 if (ret < 0) {
364 return ret;
365 }
366
367 p_name = strstr(desc, "parentFileNameHint");
368 if (p_name != NULL) {
369 char *end_name;
370
371 p_name += sizeof("parentFileNameHint") + 1;
372 end_name = strchr(p_name, '\"');
373 if (end_name == NULL) {
374 return -EINVAL;
375 }
376 if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
377 return -EINVAL;
378 }
379
380 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
381 }
382
383 return 0;
384 }
385
386 /* Create and append extent to the extent array. Return the added VmdkExtent
387 * address. return NULL if allocation failed. */
388 static VmdkExtent *vmdk_add_extent(BlockDriverState *bs,
389 BlockDriverState *file, bool flat, int64_t sectors,
390 int64_t l1_offset, int64_t l1_backup_offset,
391 uint32_t l1_size,
392 int l2_size, unsigned int cluster_sectors)
393 {
394 VmdkExtent *extent;
395 BDRVVmdkState *s = bs->opaque;
396
397 s->extents = g_realloc(s->extents,
398 (s->num_extents + 1) * sizeof(VmdkExtent));
399 extent = &s->extents[s->num_extents];
400 s->num_extents++;
401
402 memset(extent, 0, sizeof(VmdkExtent));
403 extent->file = file;
404 extent->flat = flat;
405 extent->sectors = sectors;
406 extent->l1_table_offset = l1_offset;
407 extent->l1_backup_table_offset = l1_backup_offset;
408 extent->l1_size = l1_size;
409 extent->l1_entry_sectors = l2_size * cluster_sectors;
410 extent->l2_size = l2_size;
411 extent->cluster_sectors = cluster_sectors;
412
413 if (s->num_extents > 1) {
414 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
415 } else {
416 extent->end_sector = extent->sectors;
417 }
418 bs->total_sectors = extent->end_sector;
419 return extent;
420 }
421
422 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent)
423 {
424 int ret;
425 int l1_size, i;
426
427 /* read the L1 table */
428 l1_size = extent->l1_size * sizeof(uint32_t);
429 extent->l1_table = g_malloc(l1_size);
430 ret = bdrv_pread(extent->file,
431 extent->l1_table_offset,
432 extent->l1_table,
433 l1_size);
434 if (ret < 0) {
435 goto fail_l1;
436 }
437 for (i = 0; i < extent->l1_size; i++) {
438 le32_to_cpus(&extent->l1_table[i]);
439 }
440
441 if (extent->l1_backup_table_offset) {
442 extent->l1_backup_table = g_malloc(l1_size);
443 ret = bdrv_pread(extent->file,
444 extent->l1_backup_table_offset,
445 extent->l1_backup_table,
446 l1_size);
447 if (ret < 0) {
448 goto fail_l1b;
449 }
450 for (i = 0; i < extent->l1_size; i++) {
451 le32_to_cpus(&extent->l1_backup_table[i]);
452 }
453 }
454
455 extent->l2_cache =
456 g_malloc(extent->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
457 return 0;
458 fail_l1b:
459 g_free(extent->l1_backup_table);
460 fail_l1:
461 g_free(extent->l1_table);
462 return ret;
463 }
464
465 static int vmdk_open_vmdk3(BlockDriverState *bs,
466 BlockDriverState *file,
467 int flags)
468 {
469 int ret;
470 uint32_t magic;
471 VMDK3Header header;
472 VmdkExtent *extent;
473
474 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
475 if (ret < 0) {
476 return ret;
477 }
478 extent = vmdk_add_extent(bs,
479 bs->file, false,
480 le32_to_cpu(header.disk_sectors),
481 le32_to_cpu(header.l1dir_offset) << 9,
482 0, 1 << 6, 1 << 9,
483 le32_to_cpu(header.granularity));
484 ret = vmdk_init_tables(bs, extent);
485 if (ret) {
486 /* free extent allocated by vmdk_add_extent */
487 vmdk_free_last_extent(bs);
488 }
489 return ret;
490 }
491
492 static int vmdk_open_desc_file(BlockDriverState *bs, int flags,
493 int64_t desc_offset);
494
495 static int vmdk_open_vmdk4(BlockDriverState *bs,
496 BlockDriverState *file,
497 int flags)
498 {
499 int ret;
500 uint32_t magic;
501 uint32_t l1_size, l1_entry_sectors;
502 VMDK4Header header;
503 VmdkExtent *extent;
504 int64_t l1_backup_offset = 0;
505
506 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
507 if (ret < 0) {
508 return ret;
509 }
510 if (header.capacity == 0 && header.desc_offset) {
511 return vmdk_open_desc_file(bs, flags, header.desc_offset << 9);
512 }
513
514 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
515 /*
516 * The footer takes precedence over the header, so read it in. The
517 * footer starts at offset -1024 from the end: One sector for the
518 * footer, and another one for the end-of-stream marker.
519 */
520 struct {
521 struct {
522 uint64_t val;
523 uint32_t size;
524 uint32_t type;
525 uint8_t pad[512 - 16];
526 } QEMU_PACKED footer_marker;
527
528 uint32_t magic;
529 VMDK4Header header;
530 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
531
532 struct {
533 uint64_t val;
534 uint32_t size;
535 uint32_t type;
536 uint8_t pad[512 - 16];
537 } QEMU_PACKED eos_marker;
538 } QEMU_PACKED footer;
539
540 ret = bdrv_pread(file,
541 bs->file->total_sectors * 512 - 1536,
542 &footer, sizeof(footer));
543 if (ret < 0) {
544 return ret;
545 }
546
547 /* Some sanity checks for the footer */
548 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
549 le32_to_cpu(footer.footer_marker.size) != 0 ||
550 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
551 le64_to_cpu(footer.eos_marker.val) != 0 ||
552 le32_to_cpu(footer.eos_marker.size) != 0 ||
553 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
554 {
555 return -EINVAL;
556 }
557
558 header = footer.header;
559 }
560
561 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gte)
562 * le64_to_cpu(header.granularity);
563 if (l1_entry_sectors == 0) {
564 return -EINVAL;
565 }
566 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
567 / l1_entry_sectors;
568 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
569 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
570 }
571 extent = vmdk_add_extent(bs, file, false,
572 le64_to_cpu(header.capacity),
573 le64_to_cpu(header.gd_offset) << 9,
574 l1_backup_offset,
575 l1_size,
576 le32_to_cpu(header.num_gtes_per_gte),
577 le64_to_cpu(header.granularity));
578 extent->compressed =
579 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
580 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
581 extent->version = le32_to_cpu(header.version);
582 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
583 ret = vmdk_init_tables(bs, extent);
584 if (ret) {
585 /* free extent allocated by vmdk_add_extent */
586 vmdk_free_last_extent(bs);
587 }
588 return ret;
589 }
590
591 /* find an option value out of descriptor file */
592 static int vmdk_parse_description(const char *desc, const char *opt_name,
593 char *buf, int buf_size)
594 {
595 char *opt_pos, *opt_end;
596 const char *end = desc + strlen(desc);
597
598 opt_pos = strstr(desc, opt_name);
599 if (!opt_pos) {
600 return VMDK_ERROR;
601 }
602 /* Skip "=\"" following opt_name */
603 opt_pos += strlen(opt_name) + 2;
604 if (opt_pos >= end) {
605 return VMDK_ERROR;
606 }
607 opt_end = opt_pos;
608 while (opt_end < end && *opt_end != '"') {
609 opt_end++;
610 }
611 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
612 return VMDK_ERROR;
613 }
614 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
615 return VMDK_OK;
616 }
617
618 /* Open an extent file and append to bs array */
619 static int vmdk_open_sparse(BlockDriverState *bs,
620 BlockDriverState *file,
621 int flags)
622 {
623 uint32_t magic;
624
625 if (bdrv_pread(file, 0, &magic, sizeof(magic)) != sizeof(magic)) {
626 return -EIO;
627 }
628
629 magic = be32_to_cpu(magic);
630 switch (magic) {
631 case VMDK3_MAGIC:
632 return vmdk_open_vmdk3(bs, file, flags);
633 break;
634 case VMDK4_MAGIC:
635 return vmdk_open_vmdk4(bs, file, flags);
636 break;
637 default:
638 return -EMEDIUMTYPE;
639 break;
640 }
641 }
642
643 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
644 const char *desc_file_path)
645 {
646 int ret;
647 char access[11];
648 char type[11];
649 char fname[512];
650 const char *p = desc;
651 int64_t sectors = 0;
652 int64_t flat_offset;
653 char extent_path[PATH_MAX];
654 BlockDriverState *extent_file;
655
656 while (*p) {
657 /* parse extent line:
658 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
659 * or
660 * RW [size in sectors] SPARSE "file-name.vmdk"
661 */
662 flat_offset = -1;
663 ret = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
664 access, &sectors, type, fname, &flat_offset);
665 if (ret < 4 || strcmp(access, "RW")) {
666 goto next_line;
667 } else if (!strcmp(type, "FLAT")) {
668 if (ret != 5 || flat_offset < 0) {
669 return -EINVAL;
670 }
671 } else if (ret != 4) {
672 return -EINVAL;
673 }
674
675 if (sectors <= 0 ||
676 (strcmp(type, "FLAT") && strcmp(type, "SPARSE")) ||
677 (strcmp(access, "RW"))) {
678 goto next_line;
679 }
680
681 path_combine(extent_path, sizeof(extent_path),
682 desc_file_path, fname);
683 ret = bdrv_file_open(&extent_file, extent_path, NULL, bs->open_flags);
684 if (ret) {
685 return ret;
686 }
687
688 /* save to extents array */
689 if (!strcmp(type, "FLAT")) {
690 /* FLAT extent */
691 VmdkExtent *extent;
692
693 extent = vmdk_add_extent(bs, extent_file, true, sectors,
694 0, 0, 0, 0, sectors);
695 extent->flat_start_offset = flat_offset << 9;
696 } else if (!strcmp(type, "SPARSE")) {
697 /* SPARSE extent */
698 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags);
699 if (ret) {
700 bdrv_delete(extent_file);
701 return ret;
702 }
703 } else {
704 fprintf(stderr,
705 "VMDK: Not supported extent type \"%s\""".\n", type);
706 return -ENOTSUP;
707 }
708 next_line:
709 /* move to next line */
710 while (*p && *p != '\n') {
711 p++;
712 }
713 p++;
714 }
715 return 0;
716 }
717
718 static int vmdk_open_desc_file(BlockDriverState *bs, int flags,
719 int64_t desc_offset)
720 {
721 int ret;
722 char buf[2048];
723 char ct[128];
724 BDRVVmdkState *s = bs->opaque;
725
726 ret = bdrv_pread(bs->file, desc_offset, buf, sizeof(buf));
727 if (ret < 0) {
728 return ret;
729 }
730 buf[2047] = '\0';
731 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
732 return -EMEDIUMTYPE;
733 }
734 if (strcmp(ct, "monolithicFlat") &&
735 strcmp(ct, "twoGbMaxExtentSparse") &&
736 strcmp(ct, "twoGbMaxExtentFlat")) {
737 fprintf(stderr,
738 "VMDK: Not supported image type \"%s\""".\n", ct);
739 return -ENOTSUP;
740 }
741 s->desc_offset = 0;
742 return vmdk_parse_extents(buf, bs, bs->file->filename);
743 }
744
745 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags)
746 {
747 int ret;
748 BDRVVmdkState *s = bs->opaque;
749
750 if (vmdk_open_sparse(bs, bs->file, flags) == 0) {
751 s->desc_offset = 0x200;
752 } else {
753 ret = vmdk_open_desc_file(bs, flags, 0);
754 if (ret) {
755 goto fail;
756 }
757 }
758 /* try to open parent images, if exist */
759 ret = vmdk_parent_open(bs);
760 if (ret) {
761 goto fail;
762 }
763 s->parent_cid = vmdk_read_cid(bs, 1);
764 qemu_co_mutex_init(&s->lock);
765
766 /* Disable migration when VMDK images are used */
767 error_set(&s->migration_blocker,
768 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
769 "vmdk", bs->device_name, "live migration");
770 migrate_add_blocker(s->migration_blocker);
771
772 return 0;
773
774 fail:
775 vmdk_free_extents(bs);
776 return ret;
777 }
778
779 static int get_whole_cluster(BlockDriverState *bs,
780 VmdkExtent *extent,
781 uint64_t cluster_offset,
782 uint64_t offset,
783 bool allocate)
784 {
785 /* 128 sectors * 512 bytes each = grain size 64KB */
786 uint8_t whole_grain[extent->cluster_sectors * 512];
787
788 /* we will be here if it's first write on non-exist grain(cluster).
789 * try to read from parent image, if exist */
790 if (bs->backing_hd) {
791 int ret;
792
793 if (!vmdk_is_cid_valid(bs)) {
794 return VMDK_ERROR;
795 }
796
797 /* floor offset to cluster */
798 offset -= offset % (extent->cluster_sectors * 512);
799 ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain,
800 extent->cluster_sectors);
801 if (ret < 0) {
802 return VMDK_ERROR;
803 }
804
805 /* Write grain only into the active image */
806 ret = bdrv_write(extent->file, cluster_offset, whole_grain,
807 extent->cluster_sectors);
808 if (ret < 0) {
809 return VMDK_ERROR;
810 }
811 }
812 return VMDK_OK;
813 }
814
815 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data)
816 {
817 uint32_t offset;
818 QEMU_BUILD_BUG_ON(sizeof(offset) != sizeof(m_data->offset));
819 offset = cpu_to_le32(m_data->offset);
820 /* update L2 table */
821 if (bdrv_pwrite_sync(
822 extent->file,
823 ((int64_t)m_data->l2_offset * 512)
824 + (m_data->l2_index * sizeof(m_data->offset)),
825 &offset, sizeof(offset)) < 0) {
826 return VMDK_ERROR;
827 }
828 /* update backup L2 table */
829 if (extent->l1_backup_table_offset != 0) {
830 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
831 if (bdrv_pwrite_sync(
832 extent->file,
833 ((int64_t)m_data->l2_offset * 512)
834 + (m_data->l2_index * sizeof(m_data->offset)),
835 &offset, sizeof(offset)) < 0) {
836 return VMDK_ERROR;
837 }
838 }
839 if (m_data->l2_cache_entry) {
840 *m_data->l2_cache_entry = offset;
841 }
842
843 return VMDK_OK;
844 }
845
846 static int get_cluster_offset(BlockDriverState *bs,
847 VmdkExtent *extent,
848 VmdkMetaData *m_data,
849 uint64_t offset,
850 int allocate,
851 uint64_t *cluster_offset)
852 {
853 unsigned int l1_index, l2_offset, l2_index;
854 int min_index, i, j;
855 uint32_t min_count, *l2_table;
856 bool zeroed = false;
857
858 if (m_data) {
859 m_data->valid = 0;
860 }
861 if (extent->flat) {
862 *cluster_offset = extent->flat_start_offset;
863 return VMDK_OK;
864 }
865
866 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
867 l1_index = (offset >> 9) / extent->l1_entry_sectors;
868 if (l1_index >= extent->l1_size) {
869 return VMDK_ERROR;
870 }
871 l2_offset = extent->l1_table[l1_index];
872 if (!l2_offset) {
873 return VMDK_UNALLOC;
874 }
875 for (i = 0; i < L2_CACHE_SIZE; i++) {
876 if (l2_offset == extent->l2_cache_offsets[i]) {
877 /* increment the hit count */
878 if (++extent->l2_cache_counts[i] == 0xffffffff) {
879 for (j = 0; j < L2_CACHE_SIZE; j++) {
880 extent->l2_cache_counts[j] >>= 1;
881 }
882 }
883 l2_table = extent->l2_cache + (i * extent->l2_size);
884 goto found;
885 }
886 }
887 /* not found: load a new entry in the least used one */
888 min_index = 0;
889 min_count = 0xffffffff;
890 for (i = 0; i < L2_CACHE_SIZE; i++) {
891 if (extent->l2_cache_counts[i] < min_count) {
892 min_count = extent->l2_cache_counts[i];
893 min_index = i;
894 }
895 }
896 l2_table = extent->l2_cache + (min_index * extent->l2_size);
897 if (bdrv_pread(
898 extent->file,
899 (int64_t)l2_offset * 512,
900 l2_table,
901 extent->l2_size * sizeof(uint32_t)
902 ) != extent->l2_size * sizeof(uint32_t)) {
903 return VMDK_ERROR;
904 }
905
906 extent->l2_cache_offsets[min_index] = l2_offset;
907 extent->l2_cache_counts[min_index] = 1;
908 found:
909 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
910 *cluster_offset = le32_to_cpu(l2_table[l2_index]);
911
912 if (m_data) {
913 m_data->valid = 1;
914 m_data->l1_index = l1_index;
915 m_data->l2_index = l2_index;
916 m_data->offset = *cluster_offset;
917 m_data->l2_offset = l2_offset;
918 m_data->l2_cache_entry = &l2_table[l2_index];
919 }
920 if (extent->has_zero_grain && *cluster_offset == VMDK_GTE_ZEROED) {
921 zeroed = true;
922 }
923
924 if (!*cluster_offset || zeroed) {
925 if (!allocate) {
926 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
927 }
928
929 /* Avoid the L2 tables update for the images that have snapshots. */
930 *cluster_offset = bdrv_getlength(extent->file);
931 if (!extent->compressed) {
932 bdrv_truncate(
933 extent->file,
934 *cluster_offset + (extent->cluster_sectors << 9)
935 );
936 }
937
938 *cluster_offset >>= 9;
939 l2_table[l2_index] = cpu_to_le32(*cluster_offset);
940
941 /* First of all we write grain itself, to avoid race condition
942 * that may to corrupt the image.
943 * This problem may occur because of insufficient space on host disk
944 * or inappropriate VM shutdown.
945 */
946 if (get_whole_cluster(
947 bs, extent, *cluster_offset, offset, allocate) == -1) {
948 return VMDK_ERROR;
949 }
950
951 if (m_data) {
952 m_data->offset = *cluster_offset;
953 }
954 }
955 *cluster_offset <<= 9;
956 return VMDK_OK;
957 }
958
959 static VmdkExtent *find_extent(BDRVVmdkState *s,
960 int64_t sector_num, VmdkExtent *start_hint)
961 {
962 VmdkExtent *extent = start_hint;
963
964 if (!extent) {
965 extent = &s->extents[0];
966 }
967 while (extent < &s->extents[s->num_extents]) {
968 if (sector_num < extent->end_sector) {
969 return extent;
970 }
971 extent++;
972 }
973 return NULL;
974 }
975
976 static int coroutine_fn vmdk_co_is_allocated(BlockDriverState *bs,
977 int64_t sector_num, int nb_sectors, int *pnum)
978 {
979 BDRVVmdkState *s = bs->opaque;
980 int64_t index_in_cluster, n, ret;
981 uint64_t offset;
982 VmdkExtent *extent;
983
984 extent = find_extent(s, sector_num, NULL);
985 if (!extent) {
986 return 0;
987 }
988 qemu_co_mutex_lock(&s->lock);
989 ret = get_cluster_offset(bs, extent, NULL,
990 sector_num * 512, 0, &offset);
991 qemu_co_mutex_unlock(&s->lock);
992
993 ret = (ret == VMDK_OK || ret == VMDK_ZEROED);
994
995 index_in_cluster = sector_num % extent->cluster_sectors;
996 n = extent->cluster_sectors - index_in_cluster;
997 if (n > nb_sectors) {
998 n = nb_sectors;
999 }
1000 *pnum = n;
1001 return ret;
1002 }
1003
1004 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1005 int64_t offset_in_cluster, const uint8_t *buf,
1006 int nb_sectors, int64_t sector_num)
1007 {
1008 int ret;
1009 VmdkGrainMarker *data = NULL;
1010 uLongf buf_len;
1011 const uint8_t *write_buf = buf;
1012 int write_len = nb_sectors * 512;
1013
1014 if (extent->compressed) {
1015 if (!extent->has_marker) {
1016 ret = -EINVAL;
1017 goto out;
1018 }
1019 buf_len = (extent->cluster_sectors << 9) * 2;
1020 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1021 if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK ||
1022 buf_len == 0) {
1023 ret = -EINVAL;
1024 goto out;
1025 }
1026 data->lba = sector_num;
1027 data->size = buf_len;
1028 write_buf = (uint8_t *)data;
1029 write_len = buf_len + sizeof(VmdkGrainMarker);
1030 }
1031 ret = bdrv_pwrite(extent->file,
1032 cluster_offset + offset_in_cluster,
1033 write_buf,
1034 write_len);
1035 if (ret != write_len) {
1036 ret = ret < 0 ? ret : -EIO;
1037 goto out;
1038 }
1039 ret = 0;
1040 out:
1041 g_free(data);
1042 return ret;
1043 }
1044
1045 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1046 int64_t offset_in_cluster, uint8_t *buf,
1047 int nb_sectors)
1048 {
1049 int ret;
1050 int cluster_bytes, buf_bytes;
1051 uint8_t *cluster_buf, *compressed_data;
1052 uint8_t *uncomp_buf;
1053 uint32_t data_len;
1054 VmdkGrainMarker *marker;
1055 uLongf buf_len;
1056
1057
1058 if (!extent->compressed) {
1059 ret = bdrv_pread(extent->file,
1060 cluster_offset + offset_in_cluster,
1061 buf, nb_sectors * 512);
1062 if (ret == nb_sectors * 512) {
1063 return 0;
1064 } else {
1065 return -EIO;
1066 }
1067 }
1068 cluster_bytes = extent->cluster_sectors * 512;
1069 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1070 buf_bytes = cluster_bytes * 2;
1071 cluster_buf = g_malloc(buf_bytes);
1072 uncomp_buf = g_malloc(cluster_bytes);
1073 ret = bdrv_pread(extent->file,
1074 cluster_offset,
1075 cluster_buf, buf_bytes);
1076 if (ret < 0) {
1077 goto out;
1078 }
1079 compressed_data = cluster_buf;
1080 buf_len = cluster_bytes;
1081 data_len = cluster_bytes;
1082 if (extent->has_marker) {
1083 marker = (VmdkGrainMarker *)cluster_buf;
1084 compressed_data = marker->data;
1085 data_len = le32_to_cpu(marker->size);
1086 }
1087 if (!data_len || data_len > buf_bytes) {
1088 ret = -EINVAL;
1089 goto out;
1090 }
1091 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1092 if (ret != Z_OK) {
1093 ret = -EINVAL;
1094 goto out;
1095
1096 }
1097 if (offset_in_cluster < 0 ||
1098 offset_in_cluster + nb_sectors * 512 > buf_len) {
1099 ret = -EINVAL;
1100 goto out;
1101 }
1102 memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512);
1103 ret = 0;
1104
1105 out:
1106 g_free(uncomp_buf);
1107 g_free(cluster_buf);
1108 return ret;
1109 }
1110
1111 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
1112 uint8_t *buf, int nb_sectors)
1113 {
1114 BDRVVmdkState *s = bs->opaque;
1115 int ret;
1116 uint64_t n, index_in_cluster;
1117 uint64_t extent_begin_sector, extent_relative_sector_num;
1118 VmdkExtent *extent = NULL;
1119 uint64_t cluster_offset;
1120
1121 while (nb_sectors > 0) {
1122 extent = find_extent(s, sector_num, extent);
1123 if (!extent) {
1124 return -EIO;
1125 }
1126 ret = get_cluster_offset(
1127 bs, extent, NULL,
1128 sector_num << 9, 0, &cluster_offset);
1129 extent_begin_sector = extent->end_sector - extent->sectors;
1130 extent_relative_sector_num = sector_num - extent_begin_sector;
1131 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1132 n = extent->cluster_sectors - index_in_cluster;
1133 if (n > nb_sectors) {
1134 n = nb_sectors;
1135 }
1136 if (ret != VMDK_OK) {
1137 /* if not allocated, try to read from parent image, if exist */
1138 if (bs->backing_hd && ret != VMDK_ZEROED) {
1139 if (!vmdk_is_cid_valid(bs)) {
1140 return -EINVAL;
1141 }
1142 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
1143 if (ret < 0) {
1144 return ret;
1145 }
1146 } else {
1147 memset(buf, 0, 512 * n);
1148 }
1149 } else {
1150 ret = vmdk_read_extent(extent,
1151 cluster_offset, index_in_cluster * 512,
1152 buf, n);
1153 if (ret) {
1154 return ret;
1155 }
1156 }
1157 nb_sectors -= n;
1158 sector_num += n;
1159 buf += n * 512;
1160 }
1161 return 0;
1162 }
1163
1164 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num,
1165 uint8_t *buf, int nb_sectors)
1166 {
1167 int ret;
1168 BDRVVmdkState *s = bs->opaque;
1169 qemu_co_mutex_lock(&s->lock);
1170 ret = vmdk_read(bs, sector_num, buf, nb_sectors);
1171 qemu_co_mutex_unlock(&s->lock);
1172 return ret;
1173 }
1174
1175 /**
1176 * vmdk_write:
1177 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1178 * if possible, otherwise return -ENOTSUP.
1179 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just
1180 *
1181 * Returns: error code with 0 for success.
1182 */
1183 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
1184 const uint8_t *buf, int nb_sectors,
1185 bool zeroed, bool zero_dry_run)
1186 {
1187 BDRVVmdkState *s = bs->opaque;
1188 VmdkExtent *extent = NULL;
1189 int n, ret;
1190 int64_t index_in_cluster;
1191 uint64_t extent_begin_sector, extent_relative_sector_num;
1192 uint64_t cluster_offset;
1193 VmdkMetaData m_data;
1194
1195 if (sector_num > bs->total_sectors) {
1196 fprintf(stderr,
1197 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
1198 " total_sectors=0x%" PRIx64 "\n",
1199 sector_num, bs->total_sectors);
1200 return -EIO;
1201 }
1202
1203 while (nb_sectors > 0) {
1204 extent = find_extent(s, sector_num, extent);
1205 if (!extent) {
1206 return -EIO;
1207 }
1208 ret = get_cluster_offset(
1209 bs,
1210 extent,
1211 &m_data,
1212 sector_num << 9, !extent->compressed,
1213 &cluster_offset);
1214 if (extent->compressed) {
1215 if (ret == VMDK_OK) {
1216 /* Refuse write to allocated cluster for streamOptimized */
1217 fprintf(stderr,
1218 "VMDK: can't write to allocated cluster"
1219 " for streamOptimized\n");
1220 return -EIO;
1221 } else {
1222 /* allocate */
1223 ret = get_cluster_offset(
1224 bs,
1225 extent,
1226 &m_data,
1227 sector_num << 9, 1,
1228 &cluster_offset);
1229 }
1230 }
1231 if (ret == VMDK_ERROR) {
1232 return -EINVAL;
1233 }
1234 extent_begin_sector = extent->end_sector - extent->sectors;
1235 extent_relative_sector_num = sector_num - extent_begin_sector;
1236 index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1237 n = extent->cluster_sectors - index_in_cluster;
1238 if (n > nb_sectors) {
1239 n = nb_sectors;
1240 }
1241 if (zeroed) {
1242 /* Do zeroed write, buf is ignored */
1243 if (extent->has_zero_grain &&
1244 index_in_cluster == 0 &&
1245 n >= extent->cluster_sectors) {
1246 n = extent->cluster_sectors;
1247 if (!zero_dry_run) {
1248 m_data.offset = VMDK_GTE_ZEROED;
1249 /* update L2 tables */
1250 if (vmdk_L2update(extent, &m_data) != VMDK_OK) {
1251 return -EIO;
1252 }
1253 }
1254 } else {
1255 return -ENOTSUP;
1256 }
1257 } else {
1258 ret = vmdk_write_extent(extent,
1259 cluster_offset, index_in_cluster * 512,
1260 buf, n, sector_num);
1261 if (ret) {
1262 return ret;
1263 }
1264 if (m_data.valid) {
1265 /* update L2 tables */
1266 if (vmdk_L2update(extent, &m_data) != VMDK_OK) {
1267 return -EIO;
1268 }
1269 }
1270 }
1271 nb_sectors -= n;
1272 sector_num += n;
1273 buf += n * 512;
1274
1275 /* update CID on the first write every time the virtual disk is
1276 * opened */
1277 if (!s->cid_updated) {
1278 ret = vmdk_write_cid(bs, time(NULL));
1279 if (ret < 0) {
1280 return ret;
1281 }
1282 s->cid_updated = true;
1283 }
1284 }
1285 return 0;
1286 }
1287
1288 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num,
1289 const uint8_t *buf, int nb_sectors)
1290 {
1291 int ret;
1292 BDRVVmdkState *s = bs->opaque;
1293 qemu_co_mutex_lock(&s->lock);
1294 ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false);
1295 qemu_co_mutex_unlock(&s->lock);
1296 return ret;
1297 }
1298
1299 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs,
1300 int64_t sector_num,
1301 int nb_sectors)
1302 {
1303 int ret;
1304 BDRVVmdkState *s = bs->opaque;
1305 qemu_co_mutex_lock(&s->lock);
1306 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true);
1307 if (!ret) {
1308 ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false);
1309 }
1310 qemu_co_mutex_unlock(&s->lock);
1311 return ret;
1312 }
1313
1314
1315 static int vmdk_create_extent(const char *filename, int64_t filesize,
1316 bool flat, bool compress, bool zeroed_grain)
1317 {
1318 int ret, i;
1319 int fd = 0;
1320 VMDK4Header header;
1321 uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
1322
1323 fd = qemu_open(filename,
1324 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
1325 0644);
1326 if (fd < 0) {
1327 return -errno;
1328 }
1329 if (flat) {
1330 ret = ftruncate(fd, filesize);
1331 if (ret < 0) {
1332 ret = -errno;
1333 }
1334 goto exit;
1335 }
1336 magic = cpu_to_be32(VMDK4_MAGIC);
1337 memset(&header, 0, sizeof(header));
1338 header.version = zeroed_grain ? 2 : 1;
1339 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1340 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1341 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1342 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1343 header.capacity = filesize / 512;
1344 header.granularity = 128;
1345 header.num_gtes_per_gte = 512;
1346
1347 grains = (filesize / 512 + header.granularity - 1) / header.granularity;
1348 gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
1349 gt_count =
1350 (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
1351 gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
1352
1353 header.desc_offset = 1;
1354 header.desc_size = 20;
1355 header.rgd_offset = header.desc_offset + header.desc_size;
1356 header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
1357 header.grain_offset =
1358 ((header.gd_offset + gd_size + (gt_size * gt_count) +
1359 header.granularity - 1) / header.granularity) *
1360 header.granularity;
1361 /* swap endianness for all header fields */
1362 header.version = cpu_to_le32(header.version);
1363 header.flags = cpu_to_le32(header.flags);
1364 header.capacity = cpu_to_le64(header.capacity);
1365 header.granularity = cpu_to_le64(header.granularity);
1366 header.num_gtes_per_gte = cpu_to_le32(header.num_gtes_per_gte);
1367 header.desc_offset = cpu_to_le64(header.desc_offset);
1368 header.desc_size = cpu_to_le64(header.desc_size);
1369 header.rgd_offset = cpu_to_le64(header.rgd_offset);
1370 header.gd_offset = cpu_to_le64(header.gd_offset);
1371 header.grain_offset = cpu_to_le64(header.grain_offset);
1372 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1373
1374 header.check_bytes[0] = 0xa;
1375 header.check_bytes[1] = 0x20;
1376 header.check_bytes[2] = 0xd;
1377 header.check_bytes[3] = 0xa;
1378
1379 /* write all the data */
1380 ret = qemu_write_full(fd, &magic, sizeof(magic));
1381 if (ret != sizeof(magic)) {
1382 ret = -errno;
1383 goto exit;
1384 }
1385 ret = qemu_write_full(fd, &header, sizeof(header));
1386 if (ret != sizeof(header)) {
1387 ret = -errno;
1388 goto exit;
1389 }
1390
1391 ret = ftruncate(fd, le64_to_cpu(header.grain_offset) << 9);
1392 if (ret < 0) {
1393 ret = -errno;
1394 goto exit;
1395 }
1396
1397 /* write grain directory */
1398 lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
1399 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_size;
1400 i < gt_count; i++, tmp += gt_size) {
1401 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1402 if (ret != sizeof(tmp)) {
1403 ret = -errno;
1404 goto exit;
1405 }
1406 }
1407
1408 /* write backup grain directory */
1409 lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
1410 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_size;
1411 i < gt_count; i++, tmp += gt_size) {
1412 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1413 if (ret != sizeof(tmp)) {
1414 ret = -errno;
1415 goto exit;
1416 }
1417 }
1418
1419 ret = 0;
1420 exit:
1421 qemu_close(fd);
1422 return ret;
1423 }
1424
1425 static int filename_decompose(const char *filename, char *path, char *prefix,
1426 char *postfix, size_t buf_len)
1427 {
1428 const char *p, *q;
1429
1430 if (filename == NULL || !strlen(filename)) {
1431 fprintf(stderr, "Vmdk: no filename provided.\n");
1432 return VMDK_ERROR;
1433 }
1434 p = strrchr(filename, '/');
1435 if (p == NULL) {
1436 p = strrchr(filename, '\\');
1437 }
1438 if (p == NULL) {
1439 p = strrchr(filename, ':');
1440 }
1441 if (p != NULL) {
1442 p++;
1443 if (p - filename >= buf_len) {
1444 return VMDK_ERROR;
1445 }
1446 pstrcpy(path, p - filename + 1, filename);
1447 } else {
1448 p = filename;
1449 path[0] = '\0';
1450 }
1451 q = strrchr(p, '.');
1452 if (q == NULL) {
1453 pstrcpy(prefix, buf_len, p);
1454 postfix[0] = '\0';
1455 } else {
1456 if (q - p >= buf_len) {
1457 return VMDK_ERROR;
1458 }
1459 pstrcpy(prefix, q - p + 1, p);
1460 pstrcpy(postfix, buf_len, q);
1461 }
1462 return VMDK_OK;
1463 }
1464
1465 static int relative_path(char *dest, int dest_size,
1466 const char *base, const char *target)
1467 {
1468 int i = 0;
1469 int n = 0;
1470 const char *p, *q;
1471 #ifdef _WIN32
1472 const char *sep = "\\";
1473 #else
1474 const char *sep = "/";
1475 #endif
1476
1477 if (!(dest && base && target)) {
1478 return VMDK_ERROR;
1479 }
1480 if (path_is_absolute(target)) {
1481 pstrcpy(dest, dest_size, target);
1482 return VMDK_OK;
1483 }
1484 while (base[i] == target[i]) {
1485 i++;
1486 }
1487 p = &base[i];
1488 q = &target[i];
1489 while (*p) {
1490 if (*p == *sep) {
1491 n++;
1492 }
1493 p++;
1494 }
1495 dest[0] = '\0';
1496 for (; n; n--) {
1497 pstrcat(dest, dest_size, "..");
1498 pstrcat(dest, dest_size, sep);
1499 }
1500 pstrcat(dest, dest_size, q);
1501 return VMDK_OK;
1502 }
1503
1504 static int vmdk_create(const char *filename, QEMUOptionParameter *options)
1505 {
1506 int fd, idx = 0;
1507 char desc[BUF_SIZE];
1508 int64_t total_size = 0, filesize;
1509 const char *adapter_type = NULL;
1510 const char *backing_file = NULL;
1511 const char *fmt = NULL;
1512 int flags = 0;
1513 int ret = 0;
1514 bool flat, split, compress;
1515 char ext_desc_lines[BUF_SIZE] = "";
1516 char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX];
1517 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
1518 const char *desc_extent_line;
1519 char parent_desc_line[BUF_SIZE] = "";
1520 uint32_t parent_cid = 0xffffffff;
1521 uint32_t number_heads = 16;
1522 bool zeroed_grain = false;
1523 const char desc_template[] =
1524 "# Disk DescriptorFile\n"
1525 "version=1\n"
1526 "CID=%x\n"
1527 "parentCID=%x\n"
1528 "createType=\"%s\"\n"
1529 "%s"
1530 "\n"
1531 "# Extent description\n"
1532 "%s"
1533 "\n"
1534 "# The Disk Data Base\n"
1535 "#DDB\n"
1536 "\n"
1537 "ddb.virtualHWVersion = \"%d\"\n"
1538 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1539 "ddb.geometry.heads = \"%d\"\n"
1540 "ddb.geometry.sectors = \"63\"\n"
1541 "ddb.adapterType = \"%s\"\n";
1542
1543 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX)) {
1544 return -EINVAL;
1545 }
1546 /* Read out options */
1547 while (options && options->name) {
1548 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1549 total_size = options->value.n;
1550 } else if (!strcmp(options->name, BLOCK_OPT_ADAPTER_TYPE)) {
1551 adapter_type = options->value.s;
1552 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1553 backing_file = options->value.s;
1554 } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) {
1555 flags |= options->value.n ? BLOCK_FLAG_COMPAT6 : 0;
1556 } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) {
1557 fmt = options->value.s;
1558 } else if (!strcmp(options->name, BLOCK_OPT_ZEROED_GRAIN)) {
1559 zeroed_grain |= options->value.n;
1560 }
1561 options++;
1562 }
1563 if (!adapter_type) {
1564 adapter_type = "ide";
1565 } else if (strcmp(adapter_type, "ide") &&
1566 strcmp(adapter_type, "buslogic") &&
1567 strcmp(adapter_type, "lsilogic") &&
1568 strcmp(adapter_type, "legacyESX")) {
1569 fprintf(stderr, "VMDK: Unknown adapter type: '%s'.\n", adapter_type);
1570 return -EINVAL;
1571 }
1572 if (strcmp(adapter_type, "ide") != 0) {
1573 /* that's the number of heads with which vmware operates when
1574 creating, exporting, etc. vmdk files with a non-ide adapter type */
1575 number_heads = 255;
1576 }
1577 if (!fmt) {
1578 /* Default format to monolithicSparse */
1579 fmt = "monolithicSparse";
1580 } else if (strcmp(fmt, "monolithicFlat") &&
1581 strcmp(fmt, "monolithicSparse") &&
1582 strcmp(fmt, "twoGbMaxExtentSparse") &&
1583 strcmp(fmt, "twoGbMaxExtentFlat") &&
1584 strcmp(fmt, "streamOptimized")) {
1585 fprintf(stderr, "VMDK: Unknown subformat: %s\n", fmt);
1586 return -EINVAL;
1587 }
1588 split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1589 strcmp(fmt, "twoGbMaxExtentSparse"));
1590 flat = !(strcmp(fmt, "monolithicFlat") &&
1591 strcmp(fmt, "twoGbMaxExtentFlat"));
1592 compress = !strcmp(fmt, "streamOptimized");
1593 if (flat) {
1594 desc_extent_line = "RW %lld FLAT \"%s\" 0\n";
1595 } else {
1596 desc_extent_line = "RW %lld SPARSE \"%s\"\n";
1597 }
1598 if (flat && backing_file) {
1599 /* not supporting backing file for flat image */
1600 return -ENOTSUP;
1601 }
1602 if (backing_file) {
1603 char parent_filename[PATH_MAX];
1604 BlockDriverState *bs = bdrv_new("");
1605 ret = bdrv_open(bs, backing_file, NULL, 0, NULL);
1606 if (ret != 0) {
1607 bdrv_delete(bs);
1608 return ret;
1609 }
1610 if (strcmp(bs->drv->format_name, "vmdk")) {
1611 bdrv_delete(bs);
1612 return -EINVAL;
1613 }
1614 parent_cid = vmdk_read_cid(bs, 0);
1615 bdrv_delete(bs);
1616 relative_path(parent_filename, sizeof(parent_filename),
1617 filename, backing_file);
1618 snprintf(parent_desc_line, sizeof(parent_desc_line),
1619 "parentFileNameHint=\"%s\"", parent_filename);
1620 }
1621
1622 /* Create extents */
1623 filesize = total_size;
1624 while (filesize > 0) {
1625 char desc_line[BUF_SIZE];
1626 char ext_filename[PATH_MAX];
1627 char desc_filename[PATH_MAX];
1628 int64_t size = filesize;
1629
1630 if (split && size > split_size) {
1631 size = split_size;
1632 }
1633 if (split) {
1634 snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s",
1635 prefix, flat ? 'f' : 's', ++idx, postfix);
1636 } else if (flat) {
1637 snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s",
1638 prefix, postfix);
1639 } else {
1640 snprintf(desc_filename, sizeof(desc_filename), "%s%s",
1641 prefix, postfix);
1642 }
1643 snprintf(ext_filename, sizeof(ext_filename), "%s%s",
1644 path, desc_filename);
1645
1646 if (vmdk_create_extent(ext_filename, size,
1647 flat, compress, zeroed_grain)) {
1648 return -EINVAL;
1649 }
1650 filesize -= size;
1651
1652 /* Format description line */
1653 snprintf(desc_line, sizeof(desc_line),
1654 desc_extent_line, size / 512, desc_filename);
1655 pstrcat(ext_desc_lines, sizeof(ext_desc_lines), desc_line);
1656 }
1657 /* generate descriptor file */
1658 snprintf(desc, sizeof(desc), desc_template,
1659 (unsigned int)time(NULL),
1660 parent_cid,
1661 fmt,
1662 parent_desc_line,
1663 ext_desc_lines,
1664 (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
1665 total_size / (int64_t)(63 * number_heads * 512), number_heads,
1666 adapter_type);
1667 if (split || flat) {
1668 fd = qemu_open(filename,
1669 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
1670 0644);
1671 } else {
1672 fd = qemu_open(filename,
1673 O_WRONLY | O_BINARY | O_LARGEFILE,
1674 0644);
1675 }
1676 if (fd < 0) {
1677 return -errno;
1678 }
1679 /* the descriptor offset = 0x200 */
1680 if (!split && !flat && 0x200 != lseek(fd, 0x200, SEEK_SET)) {
1681 ret = -errno;
1682 goto exit;
1683 }
1684 ret = qemu_write_full(fd, desc, strlen(desc));
1685 if (ret != strlen(desc)) {
1686 ret = -errno;
1687 goto exit;
1688 }
1689 ret = 0;
1690 exit:
1691 qemu_close(fd);
1692 return ret;
1693 }
1694
1695 static void vmdk_close(BlockDriverState *bs)
1696 {
1697 BDRVVmdkState *s = bs->opaque;
1698
1699 vmdk_free_extents(bs);
1700
1701 migrate_del_blocker(s->migration_blocker);
1702 error_free(s->migration_blocker);
1703 }
1704
1705 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
1706 {
1707 BDRVVmdkState *s = bs->opaque;
1708 int i, err;
1709 int ret = 0;
1710
1711 for (i = 0; i < s->num_extents; i++) {
1712 err = bdrv_co_flush(s->extents[i].file);
1713 if (err < 0) {
1714 ret = err;
1715 }
1716 }
1717 return ret;
1718 }
1719
1720 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
1721 {
1722 int i;
1723 int64_t ret = 0;
1724 int64_t r;
1725 BDRVVmdkState *s = bs->opaque;
1726
1727 ret = bdrv_get_allocated_file_size(bs->file);
1728 if (ret < 0) {
1729 return ret;
1730 }
1731 for (i = 0; i < s->num_extents; i++) {
1732 if (s->extents[i].file == bs->file) {
1733 continue;
1734 }
1735 r = bdrv_get_allocated_file_size(s->extents[i].file);
1736 if (r < 0) {
1737 return r;
1738 }
1739 ret += r;
1740 }
1741 return ret;
1742 }
1743
1744 static QEMUOptionParameter vmdk_create_options[] = {
1745 {
1746 .name = BLOCK_OPT_SIZE,
1747 .type = OPT_SIZE,
1748 .help = "Virtual disk size"
1749 },
1750 {
1751 .name = BLOCK_OPT_ADAPTER_TYPE,
1752 .type = OPT_STRING,
1753 .help = "Virtual adapter type, can be one of "
1754 "ide (default), lsilogic, buslogic or legacyESX"
1755 },
1756 {
1757 .name = BLOCK_OPT_BACKING_FILE,
1758 .type = OPT_STRING,
1759 .help = "File name of a base image"
1760 },
1761 {
1762 .name = BLOCK_OPT_COMPAT6,
1763 .type = OPT_FLAG,
1764 .help = "VMDK version 6 image"
1765 },
1766 {
1767 .name = BLOCK_OPT_SUBFMT,
1768 .type = OPT_STRING,
1769 .help =
1770 "VMDK flat extent format, can be one of "
1771 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
1772 },
1773 {
1774 .name = BLOCK_OPT_ZEROED_GRAIN,
1775 .type = OPT_FLAG,
1776 .help = "Enable efficient zero writes using the zeroed-grain GTE feature"
1777 },
1778 { NULL }
1779 };
1780
1781 static BlockDriver bdrv_vmdk = {
1782 .format_name = "vmdk",
1783 .instance_size = sizeof(BDRVVmdkState),
1784 .bdrv_probe = vmdk_probe,
1785 .bdrv_open = vmdk_open,
1786 .bdrv_reopen_prepare = vmdk_reopen_prepare,
1787 .bdrv_read = vmdk_co_read,
1788 .bdrv_write = vmdk_co_write,
1789 .bdrv_co_write_zeroes = vmdk_co_write_zeroes,
1790 .bdrv_close = vmdk_close,
1791 .bdrv_create = vmdk_create,
1792 .bdrv_co_flush_to_disk = vmdk_co_flush,
1793 .bdrv_co_is_allocated = vmdk_co_is_allocated,
1794 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
1795
1796 .create_options = vmdk_create_options,
1797 };
1798
1799 static void bdrv_vmdk_init(void)
1800 {
1801 bdrv_register(&bdrv_vmdk);
1802 }
1803
1804 block_init(bdrv_vmdk_init);