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/osdep.h"
27 #include "qapi/error.h"
28 #include "block/block_int.h"
29 #include "system/block-backend.h"
30 #include "qobject/qdict.h"
31 #include "qemu/error-report.h"
32 #include "qemu/module.h"
33 #include "qemu/option.h"
34 #include "qemu/bswap.h"
35 #include "qemu/memalign.h"
36 #include "migration/blocker.h"
37 #include "qemu/cutils.h"
38 #include <zlib.h>
39
40 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
41 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
42 #define VMDK4_COMPRESSION_DEFLATE 1
43 #define VMDK4_FLAG_NL_DETECT (1 << 0)
44 #define VMDK4_FLAG_RGD (1 << 1)
45 /* Zeroed-grain enable bit */
46 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
47 #define VMDK4_FLAG_COMPRESS (1 << 16)
48 #define VMDK4_FLAG_MARKER (1 << 17)
49 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
50
51 #define VMDK_EXTENT_MAX_SECTORS (1ULL << 32)
52
53 #define VMDK_GTE_ZEROED 0x1
54
55 /* VMDK internal error codes */
56 #define VMDK_OK 0
57 #define VMDK_ERROR (-1)
58 /* Cluster not allocated */
59 #define VMDK_UNALLOC (-2)
60 #define VMDK_ZEROED (-3)
61
62 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
63 #define BLOCK_OPT_TOOLSVERSION "toolsversion"
64
65 typedef struct {
66 uint32_t version;
67 uint32_t flags;
68 uint32_t disk_sectors;
69 uint32_t granularity;
70 uint32_t l1dir_offset;
71 uint32_t l1dir_size;
72 uint32_t file_sectors;
73 uint32_t cylinders;
74 uint32_t heads;
75 uint32_t sectors_per_track;
76 } QEMU_PACKED VMDK3Header;
77
78 typedef struct {
79 uint32_t version;
80 uint32_t flags;
81 uint64_t capacity;
82 uint64_t granularity;
83 uint64_t desc_offset;
84 uint64_t desc_size;
85 /* Number of GrainTableEntries per GrainTable */
86 uint32_t num_gtes_per_gt;
87 uint64_t rgd_offset;
88 uint64_t gd_offset;
89 uint64_t grain_offset;
90 char filler[1];
91 char check_bytes[4];
92 uint16_t compressAlgorithm;
93 } QEMU_PACKED VMDK4Header;
94
95 typedef struct VMDKSESparseConstHeader {
96 uint64_t magic;
97 uint64_t version;
98 uint64_t capacity;
99 uint64_t grain_size;
100 uint64_t grain_table_size;
101 uint64_t flags;
102 uint64_t reserved1;
103 uint64_t reserved2;
104 uint64_t reserved3;
105 uint64_t reserved4;
106 uint64_t volatile_header_offset;
107 uint64_t volatile_header_size;
108 uint64_t journal_header_offset;
109 uint64_t journal_header_size;
110 uint64_t journal_offset;
111 uint64_t journal_size;
112 uint64_t grain_dir_offset;
113 uint64_t grain_dir_size;
114 uint64_t grain_tables_offset;
115 uint64_t grain_tables_size;
116 uint64_t free_bitmap_offset;
117 uint64_t free_bitmap_size;
118 uint64_t backmap_offset;
119 uint64_t backmap_size;
120 uint64_t grains_offset;
121 uint64_t grains_size;
122 uint8_t pad[304];
123 } QEMU_PACKED VMDKSESparseConstHeader;
124
125 typedef struct VMDKSESparseVolatileHeader {
126 uint64_t magic;
127 uint64_t free_gt_number;
128 uint64_t next_txn_seq_number;
129 uint64_t replay_journal;
130 uint8_t pad[480];
131 } QEMU_PACKED VMDKSESparseVolatileHeader;
132
133 #define L2_CACHE_SIZE 16
134
135 typedef struct VmdkExtent {
136 BdrvChild *file;
137 bool flat;
138 bool compressed;
139 bool has_marker;
140 bool has_zero_grain;
141 bool sesparse;
142 uint64_t sesparse_l2_tables_offset;
143 uint64_t sesparse_clusters_offset;
144 int32_t entry_size;
145 int version;
146 int64_t sectors;
147 int64_t end_sector;
148 int64_t flat_start_offset;
149 int64_t l1_table_offset;
150 int64_t l1_backup_table_offset;
151 void *l1_table;
152 uint32_t *l1_backup_table;
153 unsigned int l1_size;
154 uint32_t l1_entry_sectors;
155
156 unsigned int l2_size;
157 void *l2_cache;
158 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
159 uint32_t l2_cache_counts[L2_CACHE_SIZE];
160
161 int64_t cluster_sectors;
162 int64_t next_cluster_sector;
163 char *type;
164 } VmdkExtent;
165
166 typedef struct BDRVVmdkState {
167 CoMutex lock;
168 uint64_t desc_offset;
169 bool cid_updated;
170 bool cid_checked;
171 uint32_t cid;
172 uint32_t parent_cid;
173 int num_extents;
174 /* Extent array with num_extents entries, ascend ordered by address */
175 VmdkExtent *extents;
176 Error *migration_blocker;
177 char *create_type;
178 } BDRVVmdkState;
179
180 typedef struct BDRVVmdkReopenState {
181 bool *extents_using_bs_file;
182 } BDRVVmdkReopenState;
183
184 typedef struct VmdkMetaData {
185 unsigned int l1_index;
186 unsigned int l2_index;
187 unsigned int l2_offset;
188 bool new_allocation;
189 uint32_t *l2_cache_entry;
190 } VmdkMetaData;
191
192 typedef struct VmdkGrainMarker {
193 uint64_t lba;
194 uint32_t size;
195 uint8_t data[];
196 } QEMU_PACKED VmdkGrainMarker;
197
198 enum {
199 MARKER_END_OF_STREAM = 0,
200 MARKER_GRAIN_TABLE = 1,
201 MARKER_GRAIN_DIRECTORY = 2,
202 MARKER_FOOTER = 3,
203 };
204
vmdk_probe(const uint8_t * buf,int buf_size,const char * filename)205 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
206 {
207 uint32_t magic;
208
209 if (buf_size < 4) {
210 return 0;
211 }
212 magic = be32_to_cpu(*(uint32_t *)buf);
213 if (magic == VMDK3_MAGIC ||
214 magic == VMDK4_MAGIC) {
215 return 100;
216 } else {
217 const char *p = (const char *)buf;
218 const char *end = p + buf_size;
219 while (p < end) {
220 if (*p == '#') {
221 /* skip comment line */
222 while (p < end && *p != '\n') {
223 p++;
224 }
225 p++;
226 continue;
227 }
228 if (*p == ' ') {
229 while (p < end && *p == ' ') {
230 p++;
231 }
232 /* skip '\r' if windows line endings used. */
233 if (p < end && *p == '\r') {
234 p++;
235 }
236 /* only accept blank lines before 'version=' line */
237 if (p == end || *p != '\n') {
238 return 0;
239 }
240 p++;
241 continue;
242 }
243 if (end - p >= strlen("version=X\n")) {
244 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
245 strncmp("version=2\n", p, strlen("version=2\n")) == 0 ||
246 strncmp("version=3\n", p, strlen("version=3\n")) == 0) {
247 return 100;
248 }
249 }
250 if (end - p >= strlen("version=X\r\n")) {
251 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
252 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0 ||
253 strncmp("version=3\r\n", p, strlen("version=3\r\n")) == 0) {
254 return 100;
255 }
256 }
257 return 0;
258 }
259 return 0;
260 }
261 }
262
263 #define SECTOR_SIZE 512
264 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
265 #define BUF_SIZE 4096
266 #define HEADER_SIZE 512 /* first sector of 512 bytes */
267
vmdk_free_extents(BlockDriverState * bs)268 static void vmdk_free_extents(BlockDriverState *bs)
269 {
270 int i;
271 BDRVVmdkState *s = bs->opaque;
272 VmdkExtent *e;
273
274 bdrv_graph_wrlock_drained();
275 for (i = 0; i < s->num_extents; i++) {
276 e = &s->extents[i];
277 g_free(e->l1_table);
278 g_free(e->l2_cache);
279 g_free(e->l1_backup_table);
280 g_free(e->type);
281 if (e->file != bs->file) {
282 bdrv_unref_child(bs, e->file);
283 }
284 }
285 bdrv_graph_wrunlock();
286
287 g_free(s->extents);
288 }
289
vmdk_free_last_extent(BlockDriverState * bs)290 static void vmdk_free_last_extent(BlockDriverState *bs)
291 {
292 BDRVVmdkState *s = bs->opaque;
293
294 if (s->num_extents == 0) {
295 return;
296 }
297 s->num_extents--;
298 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
299 }
300
301 /* Return -ve errno, or 0 on success and write CID into *pcid. */
302 static int GRAPH_RDLOCK
vmdk_read_cid(BlockDriverState * bs,int parent,uint32_t * pcid)303 vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
304 {
305 char *desc;
306 uint32_t cid;
307 const char *p_name, *cid_str;
308 size_t cid_str_size;
309 BDRVVmdkState *s = bs->opaque;
310 int ret;
311
312 desc = g_malloc0(DESC_SIZE);
313 ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
314 if (ret < 0) {
315 goto out;
316 }
317
318 if (parent) {
319 cid_str = "parentCID";
320 cid_str_size = sizeof("parentCID");
321 } else {
322 cid_str = "CID";
323 cid_str_size = sizeof("CID");
324 }
325
326 desc[DESC_SIZE - 1] = '\0';
327 p_name = strstr(desc, cid_str);
328 if (p_name == NULL) {
329 ret = -EINVAL;
330 goto out;
331 }
332 p_name += cid_str_size;
333 if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
334 ret = -EINVAL;
335 goto out;
336 }
337 *pcid = cid;
338 ret = 0;
339
340 out:
341 g_free(desc);
342 return ret;
343 }
344
345 static int coroutine_fn GRAPH_RDLOCK
vmdk_write_cid(BlockDriverState * bs,uint32_t cid)346 vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
347 {
348 char *desc, *tmp_desc;
349 char *p_name, *tmp_str;
350 BDRVVmdkState *s = bs->opaque;
351 int ret = 0;
352
353 size_t desc_buf_size;
354
355 if (s->desc_offset == 0) {
356 desc_buf_size = bdrv_getlength(bs->file->bs);
357 if (desc_buf_size > 16ULL << 20) {
358 error_report("VMDK description file too big");
359 return -EFBIG;
360 }
361 } else {
362 desc_buf_size = DESC_SIZE;
363 }
364
365 desc = g_malloc0(desc_buf_size);
366 tmp_desc = g_malloc0(desc_buf_size);
367 ret = bdrv_co_pread(bs->file, s->desc_offset, desc_buf_size, desc, 0);
368 if (ret < 0) {
369 goto out;
370 }
371
372 desc[desc_buf_size - 1] = '\0';
373 tmp_str = strstr(desc, "parentCID");
374 if (tmp_str == NULL) {
375 ret = -EINVAL;
376 goto out;
377 }
378
379 pstrcpy(tmp_desc, desc_buf_size, tmp_str);
380 p_name = strstr(desc, "CID");
381 if (p_name != NULL) {
382 p_name += sizeof("CID");
383 snprintf(p_name, desc_buf_size - (p_name - desc), "%" PRIx32 "\n", cid);
384 pstrcat(desc, desc_buf_size, tmp_desc);
385 }
386
387 ret = bdrv_co_pwrite_sync(bs->file, s->desc_offset, desc_buf_size, desc, 0);
388
389 out:
390 g_free(desc);
391 g_free(tmp_desc);
392 return ret;
393 }
394
vmdk_is_cid_valid(BlockDriverState * bs)395 static int coroutine_fn GRAPH_RDLOCK vmdk_is_cid_valid(BlockDriverState *bs)
396 {
397 BDRVVmdkState *s = bs->opaque;
398 uint32_t cur_pcid;
399
400 if (!s->cid_checked && bs->backing) {
401 BlockDriverState *p_bs = bs->backing->bs;
402
403 if (strcmp(p_bs->drv->format_name, "vmdk")) {
404 /* Backing file is not in vmdk format, so it does not have
405 * a CID, which makes the overlay's parent CID invalid */
406 return 0;
407 }
408
409 if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
410 /* read failure: report as not valid */
411 return 0;
412 }
413 if (s->parent_cid != cur_pcid) {
414 /* CID not valid */
415 return 0;
416 }
417 }
418 s->cid_checked = true;
419 /* CID valid */
420 return 1;
421 }
422
vmdk_reopen_prepare(BDRVReopenState * state,BlockReopenQueue * queue,Error ** errp)423 static int vmdk_reopen_prepare(BDRVReopenState *state,
424 BlockReopenQueue *queue, Error **errp)
425 {
426 BDRVVmdkState *s;
427 BDRVVmdkReopenState *rs;
428 int i;
429
430 GLOBAL_STATE_CODE();
431 GRAPH_RDLOCK_GUARD_MAINLOOP();
432
433 assert(state != NULL);
434 assert(state->bs != NULL);
435 assert(state->opaque == NULL);
436
437 s = state->bs->opaque;
438
439 rs = g_new0(BDRVVmdkReopenState, 1);
440 state->opaque = rs;
441
442 /*
443 * Check whether there are any extents stored in bs->file; if bs->file
444 * changes, we will need to update their .file pointers to follow suit
445 */
446 rs->extents_using_bs_file = g_new(bool, s->num_extents);
447 for (i = 0; i < s->num_extents; i++) {
448 rs->extents_using_bs_file[i] = s->extents[i].file == state->bs->file;
449 }
450
451 return 0;
452 }
453
vmdk_reopen_clean(BDRVReopenState * state)454 static void vmdk_reopen_clean(BDRVReopenState *state)
455 {
456 BDRVVmdkReopenState *rs = state->opaque;
457
458 g_free(rs->extents_using_bs_file);
459 g_free(rs);
460 state->opaque = NULL;
461 }
462
vmdk_reopen_commit(BDRVReopenState * state)463 static void vmdk_reopen_commit(BDRVReopenState *state)
464 {
465 BDRVVmdkState *s = state->bs->opaque;
466 BDRVVmdkReopenState *rs = state->opaque;
467 int i;
468
469 GLOBAL_STATE_CODE();
470 GRAPH_RDLOCK_GUARD_MAINLOOP();
471
472 for (i = 0; i < s->num_extents; i++) {
473 if (rs->extents_using_bs_file[i]) {
474 s->extents[i].file = state->bs->file;
475 }
476 }
477
478 vmdk_reopen_clean(state);
479 }
480
vmdk_reopen_abort(BDRVReopenState * state)481 static void vmdk_reopen_abort(BDRVReopenState *state)
482 {
483 vmdk_reopen_clean(state);
484 }
485
vmdk_parent_open(BlockDriverState * bs)486 static int GRAPH_RDLOCK vmdk_parent_open(BlockDriverState *bs)
487 {
488 char *p_name;
489 char *desc;
490 BDRVVmdkState *s = bs->opaque;
491 int ret;
492
493 desc = g_malloc0(DESC_SIZE + 1);
494 ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
495 if (ret < 0) {
496 goto out;
497 }
498
499 p_name = strstr(desc, "parentFileNameHint");
500 if (p_name != NULL) {
501 char *end_name;
502
503 p_name += sizeof("parentFileNameHint") + 1;
504 end_name = strchr(p_name, '\"');
505 if (end_name == NULL) {
506 ret = -EINVAL;
507 goto out;
508 }
509 if ((end_name - p_name) > sizeof(bs->auto_backing_file) - 1) {
510 ret = -EINVAL;
511 goto out;
512 }
513
514 pstrcpy(bs->auto_backing_file, end_name - p_name + 1, p_name);
515 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
516 bs->auto_backing_file);
517 pstrcpy(bs->backing_format, sizeof(bs->backing_format),
518 "vmdk");
519 }
520
521 out:
522 g_free(desc);
523 return ret;
524 }
525
526 /* Create and append extent to the extent array. Return the added VmdkExtent
527 * address. return NULL if allocation failed. */
vmdk_add_extent(BlockDriverState * bs,BdrvChild * file,bool flat,int64_t sectors,int64_t l1_offset,int64_t l1_backup_offset,uint32_t l1_size,int l2_size,uint64_t cluster_sectors,VmdkExtent ** new_extent,Error ** errp)528 static int vmdk_add_extent(BlockDriverState *bs,
529 BdrvChild *file, bool flat, int64_t sectors,
530 int64_t l1_offset, int64_t l1_backup_offset,
531 uint32_t l1_size,
532 int l2_size, uint64_t cluster_sectors,
533 VmdkExtent **new_extent,
534 Error **errp)
535 {
536 VmdkExtent *extent;
537 BDRVVmdkState *s = bs->opaque;
538 int64_t nb_sectors;
539
540 if (cluster_sectors > 0x200000) {
541 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
542 error_setg(errp, "Invalid granularity, image may be corrupt");
543 return -EFBIG;
544 }
545 if (l1_size > 32 * 1024 * 1024) {
546 /*
547 * Although with big capacity and small l1_entry_sectors, we can get a
548 * big l1_size, we don't want unbounded value to allocate the table.
549 * Limit it to 32M, which is enough to store:
550 * 8TB - for both VMDK3 & VMDK4 with
551 * minimal cluster size: 512B
552 * minimal L2 table size: 512 entries
553 * 8 TB is still more than the maximal value supported for
554 * VMDK3 & VMDK4 which is 2TB.
555 * 64TB - for "ESXi seSparse Extent"
556 * minimal cluster size: 512B (default is 4KB)
557 * L2 table size: 4096 entries (const).
558 * 64TB is more than the maximal value supported for
559 * seSparse VMDKs (which is slightly less than 64TB)
560 */
561 error_setg(errp, "L1 size too big");
562 return -EFBIG;
563 }
564
565 nb_sectors = bdrv_nb_sectors(file->bs);
566 if (nb_sectors < 0) {
567 return nb_sectors;
568 }
569
570 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
571 extent = &s->extents[s->num_extents];
572 s->num_extents++;
573
574 memset(extent, 0, sizeof(VmdkExtent));
575 extent->file = file;
576 extent->flat = flat;
577 extent->sectors = sectors;
578 extent->l1_table_offset = l1_offset;
579 extent->l1_backup_table_offset = l1_backup_offset;
580 extent->l1_size = l1_size;
581 extent->l1_entry_sectors = l2_size * cluster_sectors;
582 extent->l2_size = l2_size;
583 extent->cluster_sectors = flat ? sectors : cluster_sectors;
584 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
585 extent->entry_size = sizeof(uint32_t);
586
587 if (s->num_extents > 1) {
588 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
589 } else {
590 extent->end_sector = extent->sectors;
591 }
592 bs->total_sectors = extent->end_sector;
593 if (new_extent) {
594 *new_extent = extent;
595 }
596 return 0;
597 }
598
599 static int GRAPH_RDLOCK
vmdk_init_tables(BlockDriverState * bs,VmdkExtent * extent,Error ** errp)600 vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent, Error **errp)
601 {
602 int ret;
603 size_t l1_size;
604 int i;
605
606 /* read the L1 table */
607 l1_size = extent->l1_size * extent->entry_size;
608 extent->l1_table = g_try_malloc(l1_size);
609 if (l1_size && extent->l1_table == NULL) {
610 return -ENOMEM;
611 }
612
613 ret = bdrv_pread(extent->file, extent->l1_table_offset, l1_size,
614 extent->l1_table, 0);
615 if (ret < 0) {
616 bdrv_refresh_filename(extent->file->bs);
617 error_setg_errno(errp, -ret,
618 "Could not read l1 table from extent '%s'",
619 extent->file->bs->filename);
620 goto fail_l1;
621 }
622 for (i = 0; i < extent->l1_size; i++) {
623 if (extent->entry_size == sizeof(uint64_t)) {
624 le64_to_cpus((uint64_t *)extent->l1_table + i);
625 } else {
626 assert(extent->entry_size == sizeof(uint32_t));
627 le32_to_cpus((uint32_t *)extent->l1_table + i);
628 }
629 }
630
631 if (extent->l1_backup_table_offset) {
632 assert(!extent->sesparse);
633 extent->l1_backup_table = g_try_malloc(l1_size);
634 if (l1_size && extent->l1_backup_table == NULL) {
635 ret = -ENOMEM;
636 goto fail_l1;
637 }
638 ret = bdrv_pread(extent->file, extent->l1_backup_table_offset,
639 l1_size, extent->l1_backup_table, 0);
640 if (ret < 0) {
641 bdrv_refresh_filename(extent->file->bs);
642 error_setg_errno(errp, -ret,
643 "Could not read l1 backup table from extent '%s'",
644 extent->file->bs->filename);
645 goto fail_l1b;
646 }
647 for (i = 0; i < extent->l1_size; i++) {
648 le32_to_cpus(&extent->l1_backup_table[i]);
649 }
650 }
651
652 extent->l2_cache =
653 g_malloc(extent->entry_size * extent->l2_size * L2_CACHE_SIZE);
654 return 0;
655 fail_l1b:
656 g_free(extent->l1_backup_table);
657 fail_l1:
658 g_free(extent->l1_table);
659 return ret;
660 }
661
662 static int GRAPH_RDLOCK
vmdk_open_vmfs_sparse(BlockDriverState * bs,BdrvChild * file,int flags,Error ** errp)663 vmdk_open_vmfs_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
664 Error **errp)
665 {
666 int ret;
667 uint32_t magic;
668 VMDK3Header header;
669 VmdkExtent *extent = NULL;
670
671 ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0);
672 if (ret < 0) {
673 bdrv_refresh_filename(file->bs);
674 error_setg_errno(errp, -ret,
675 "Could not read header from file '%s'",
676 file->bs->filename);
677 return ret;
678 }
679 ret = vmdk_add_extent(bs, file, false,
680 le32_to_cpu(header.disk_sectors),
681 (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
682 0,
683 le32_to_cpu(header.l1dir_size),
684 4096,
685 le32_to_cpu(header.granularity),
686 &extent,
687 errp);
688 if (ret < 0) {
689 return ret;
690 }
691 ret = vmdk_init_tables(bs, extent, errp);
692 if (ret) {
693 /* free extent allocated by vmdk_add_extent */
694 vmdk_free_last_extent(bs);
695 }
696 return ret;
697 }
698
699 #define SESPARSE_CONST_HEADER_MAGIC UINT64_C(0x00000000cafebabe)
700 #define SESPARSE_VOLATILE_HEADER_MAGIC UINT64_C(0x00000000cafecafe)
701
702 /* Strict checks - format not officially documented */
check_se_sparse_const_header(VMDKSESparseConstHeader * header,Error ** errp)703 static int check_se_sparse_const_header(VMDKSESparseConstHeader *header,
704 Error **errp)
705 {
706 header->magic = le64_to_cpu(header->magic);
707 header->version = le64_to_cpu(header->version);
708 header->grain_size = le64_to_cpu(header->grain_size);
709 header->grain_table_size = le64_to_cpu(header->grain_table_size);
710 header->flags = le64_to_cpu(header->flags);
711 header->reserved1 = le64_to_cpu(header->reserved1);
712 header->reserved2 = le64_to_cpu(header->reserved2);
713 header->reserved3 = le64_to_cpu(header->reserved3);
714 header->reserved4 = le64_to_cpu(header->reserved4);
715
716 header->volatile_header_offset =
717 le64_to_cpu(header->volatile_header_offset);
718 header->volatile_header_size = le64_to_cpu(header->volatile_header_size);
719
720 header->journal_header_offset = le64_to_cpu(header->journal_header_offset);
721 header->journal_header_size = le64_to_cpu(header->journal_header_size);
722
723 header->journal_offset = le64_to_cpu(header->journal_offset);
724 header->journal_size = le64_to_cpu(header->journal_size);
725
726 header->grain_dir_offset = le64_to_cpu(header->grain_dir_offset);
727 header->grain_dir_size = le64_to_cpu(header->grain_dir_size);
728
729 header->grain_tables_offset = le64_to_cpu(header->grain_tables_offset);
730 header->grain_tables_size = le64_to_cpu(header->grain_tables_size);
731
732 header->free_bitmap_offset = le64_to_cpu(header->free_bitmap_offset);
733 header->free_bitmap_size = le64_to_cpu(header->free_bitmap_size);
734
735 header->backmap_offset = le64_to_cpu(header->backmap_offset);
736 header->backmap_size = le64_to_cpu(header->backmap_size);
737
738 header->grains_offset = le64_to_cpu(header->grains_offset);
739 header->grains_size = le64_to_cpu(header->grains_size);
740
741 if (header->magic != SESPARSE_CONST_HEADER_MAGIC) {
742 error_setg(errp, "Bad const header magic: 0x%016" PRIx64,
743 header->magic);
744 return -EINVAL;
745 }
746
747 if (header->version != 0x0000000200000001) {
748 error_setg(errp, "Unsupported version: 0x%016" PRIx64,
749 header->version);
750 return -ENOTSUP;
751 }
752
753 if (header->grain_size != 8) {
754 error_setg(errp, "Unsupported grain size: %" PRIu64,
755 header->grain_size);
756 return -ENOTSUP;
757 }
758
759 if (header->grain_table_size != 64) {
760 error_setg(errp, "Unsupported grain table size: %" PRIu64,
761 header->grain_table_size);
762 return -ENOTSUP;
763 }
764
765 if (header->flags != 0) {
766 error_setg(errp, "Unsupported flags: 0x%016" PRIx64,
767 header->flags);
768 return -ENOTSUP;
769 }
770
771 if (header->reserved1 != 0 || header->reserved2 != 0 ||
772 header->reserved3 != 0 || header->reserved4 != 0) {
773 error_setg(errp, "Unsupported reserved bits:"
774 " 0x%016" PRIx64 " 0x%016" PRIx64
775 " 0x%016" PRIx64 " 0x%016" PRIx64,
776 header->reserved1, header->reserved2,
777 header->reserved3, header->reserved4);
778 return -ENOTSUP;
779 }
780
781 /* check that padding is 0 */
782 if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
783 error_setg(errp, "Unsupported non-zero const header padding");
784 return -ENOTSUP;
785 }
786
787 return 0;
788 }
789
check_se_sparse_volatile_header(VMDKSESparseVolatileHeader * header,Error ** errp)790 static int check_se_sparse_volatile_header(VMDKSESparseVolatileHeader *header,
791 Error **errp)
792 {
793 header->magic = le64_to_cpu(header->magic);
794 header->free_gt_number = le64_to_cpu(header->free_gt_number);
795 header->next_txn_seq_number = le64_to_cpu(header->next_txn_seq_number);
796 header->replay_journal = le64_to_cpu(header->replay_journal);
797
798 if (header->magic != SESPARSE_VOLATILE_HEADER_MAGIC) {
799 error_setg(errp, "Bad volatile header magic: 0x%016" PRIx64,
800 header->magic);
801 return -EINVAL;
802 }
803
804 if (header->replay_journal) {
805 error_setg(errp, "Image is dirty, Replaying journal not supported");
806 return -ENOTSUP;
807 }
808
809 /* check that padding is 0 */
810 if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
811 error_setg(errp, "Unsupported non-zero volatile header padding");
812 return -ENOTSUP;
813 }
814
815 return 0;
816 }
817
818 static int GRAPH_RDLOCK
vmdk_open_se_sparse(BlockDriverState * bs,BdrvChild * file,int flags,Error ** errp)819 vmdk_open_se_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
820 Error **errp)
821 {
822 int ret;
823 VMDKSESparseConstHeader const_header;
824 VMDKSESparseVolatileHeader volatile_header;
825 VmdkExtent *extent = NULL;
826
827 ret = bdrv_apply_auto_read_only(bs,
828 "No write support for seSparse images available", errp);
829 if (ret < 0) {
830 return ret;
831 }
832
833 assert(sizeof(const_header) == SECTOR_SIZE);
834
835 ret = bdrv_pread(file, 0, sizeof(const_header), &const_header, 0);
836 if (ret < 0) {
837 bdrv_refresh_filename(file->bs);
838 error_setg_errno(errp, -ret,
839 "Could not read const header from file '%s'",
840 file->bs->filename);
841 return ret;
842 }
843
844 /* check const header */
845 ret = check_se_sparse_const_header(&const_header, errp);
846 if (ret < 0) {
847 return ret;
848 }
849
850 assert(sizeof(volatile_header) == SECTOR_SIZE);
851
852 ret = bdrv_pread(file, const_header.volatile_header_offset * SECTOR_SIZE,
853 sizeof(volatile_header), &volatile_header, 0);
854 if (ret < 0) {
855 bdrv_refresh_filename(file->bs);
856 error_setg_errno(errp, -ret,
857 "Could not read volatile header from file '%s'",
858 file->bs->filename);
859 return ret;
860 }
861
862 /* check volatile header */
863 ret = check_se_sparse_volatile_header(&volatile_header, errp);
864 if (ret < 0) {
865 return ret;
866 }
867
868 ret = vmdk_add_extent(bs, file, false,
869 const_header.capacity,
870 const_header.grain_dir_offset * SECTOR_SIZE,
871 0,
872 const_header.grain_dir_size *
873 SECTOR_SIZE / sizeof(uint64_t),
874 const_header.grain_table_size *
875 SECTOR_SIZE / sizeof(uint64_t),
876 const_header.grain_size,
877 &extent,
878 errp);
879 if (ret < 0) {
880 return ret;
881 }
882
883 extent->sesparse = true;
884 extent->sesparse_l2_tables_offset = const_header.grain_tables_offset;
885 extent->sesparse_clusters_offset = const_header.grains_offset;
886 extent->entry_size = sizeof(uint64_t);
887
888 ret = vmdk_init_tables(bs, extent, errp);
889 if (ret) {
890 /* free extent allocated by vmdk_add_extent */
891 vmdk_free_last_extent(bs);
892 }
893
894 return ret;
895 }
896
897 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
898 QDict *options, Error **errp);
899
vmdk_read_desc(BdrvChild * file,uint64_t desc_offset,Error ** errp)900 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
901 {
902 int64_t size;
903 char *buf;
904 int ret;
905
906 size = bdrv_getlength(file->bs);
907 if (size < 0) {
908 error_setg_errno(errp, -size, "Could not access file");
909 return NULL;
910 }
911
912 if (size < 4) {
913 /* Both descriptor file and sparse image must be much larger than 4
914 * bytes, also callers of vmdk_read_desc want to compare the first 4
915 * bytes with VMDK4_MAGIC, let's error out if less is read. */
916 error_setg(errp, "File is too small, not a valid image");
917 return NULL;
918 }
919
920 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */
921 buf = g_malloc(size + 1);
922
923 ret = bdrv_pread(file, desc_offset, size, buf, 0);
924 if (ret < 0) {
925 error_setg_errno(errp, -ret, "Could not read from file");
926 g_free(buf);
927 return NULL;
928 }
929 buf[size] = 0;
930
931 return buf;
932 }
933
934 static int GRAPH_RDLOCK
vmdk_open_vmdk4(BlockDriverState * bs,BdrvChild * file,int flags,QDict * options,Error ** errp)935 vmdk_open_vmdk4(BlockDriverState *bs, BdrvChild *file, int flags,
936 QDict *options, Error **errp)
937 {
938 int ret;
939 uint32_t magic;
940 uint32_t l1_size, l1_entry_sectors;
941 VMDK4Header header;
942 VmdkExtent *extent = NULL;
943 BDRVVmdkState *s = bs->opaque;
944 int64_t l1_backup_offset = 0;
945 bool compressed;
946
947 ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0);
948 if (ret < 0) {
949 bdrv_refresh_filename(file->bs);
950 error_setg_errno(errp, -ret,
951 "Could not read header from file '%s'",
952 file->bs->filename);
953 return -EINVAL;
954 }
955 if (header.capacity == 0) {
956 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
957 if (desc_offset) {
958 char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
959 if (!buf) {
960 return -EINVAL;
961 }
962 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
963 g_free(buf);
964 return ret;
965 }
966 }
967
968 if (!s->create_type) {
969 s->create_type = g_strdup("monolithicSparse");
970 }
971
972 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
973 /*
974 * The footer takes precedence over the header, so read it in. The
975 * footer starts at offset -1024 from the end: One sector for the
976 * footer, and another one for the end-of-stream marker.
977 */
978 struct {
979 struct {
980 uint64_t val;
981 uint32_t size;
982 uint32_t type;
983 uint8_t pad[512 - 16];
984 } QEMU_PACKED footer_marker;
985
986 uint32_t magic;
987 VMDK4Header header;
988 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
989
990 struct {
991 uint64_t val;
992 uint32_t size;
993 uint32_t type;
994 uint8_t pad[512 - 16];
995 } QEMU_PACKED eos_marker;
996 } QEMU_PACKED footer;
997
998 ret = bdrv_pread(file, bs->file->bs->total_sectors * 512 - 1536,
999 sizeof(footer), &footer, 0);
1000 if (ret < 0) {
1001 error_setg_errno(errp, -ret, "Failed to read footer");
1002 return ret;
1003 }
1004
1005 /* Some sanity checks for the footer */
1006 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
1007 le32_to_cpu(footer.footer_marker.size) != 0 ||
1008 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
1009 le64_to_cpu(footer.eos_marker.val) != 0 ||
1010 le32_to_cpu(footer.eos_marker.size) != 0 ||
1011 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
1012 {
1013 error_setg(errp, "Invalid footer");
1014 return -EINVAL;
1015 }
1016
1017 header = footer.header;
1018 }
1019
1020 compressed =
1021 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1022 if (le32_to_cpu(header.version) > 3) {
1023 error_setg(errp, "Unsupported VMDK version %" PRIu32,
1024 le32_to_cpu(header.version));
1025 return -ENOTSUP;
1026 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
1027 !compressed) {
1028 /* VMware KB 2064959 explains that version 3 added support for
1029 * persistent changed block tracking (CBT), and backup software can
1030 * read it as version=1 if it doesn't care about the changed area
1031 * information. So we are safe to enable read only. */
1032 error_setg(errp, "VMDK version 3 must be read only");
1033 return -EINVAL;
1034 }
1035
1036 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
1037 error_setg(errp, "L2 table size too big");
1038 return -EINVAL;
1039 }
1040
1041 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
1042 * le64_to_cpu(header.granularity);
1043 if (l1_entry_sectors == 0) {
1044 error_setg(errp, "L1 entry size is invalid");
1045 return -EINVAL;
1046 }
1047 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
1048 / l1_entry_sectors;
1049 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
1050 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
1051 }
1052 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
1053 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
1054 (int64_t)(le64_to_cpu(header.grain_offset)
1055 * BDRV_SECTOR_SIZE));
1056 return -EINVAL;
1057 }
1058
1059 ret = vmdk_add_extent(bs, file, false,
1060 le64_to_cpu(header.capacity),
1061 le64_to_cpu(header.gd_offset) << 9,
1062 l1_backup_offset,
1063 l1_size,
1064 le32_to_cpu(header.num_gtes_per_gt),
1065 le64_to_cpu(header.granularity),
1066 &extent,
1067 errp);
1068 if (ret < 0) {
1069 return ret;
1070 }
1071 extent->compressed =
1072 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1073 if (extent->compressed) {
1074 g_free(s->create_type);
1075 s->create_type = g_strdup("streamOptimized");
1076 }
1077 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
1078 extent->version = le32_to_cpu(header.version);
1079 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
1080 ret = vmdk_init_tables(bs, extent, errp);
1081 if (ret) {
1082 /* free extent allocated by vmdk_add_extent */
1083 vmdk_free_last_extent(bs);
1084 }
1085 return ret;
1086 }
1087
1088 /* find an option value out of descriptor file */
vmdk_parse_description(const char * desc,const char * opt_name,char * buf,int buf_size)1089 static int vmdk_parse_description(const char *desc, const char *opt_name,
1090 char *buf, int buf_size)
1091 {
1092 char *opt_pos, *opt_end;
1093 const char *end = desc + strlen(desc);
1094
1095 opt_pos = strstr(desc, opt_name);
1096 if (!opt_pos) {
1097 return VMDK_ERROR;
1098 }
1099 /* Skip "=\"" following opt_name */
1100 opt_pos += strlen(opt_name) + 2;
1101 if (opt_pos >= end) {
1102 return VMDK_ERROR;
1103 }
1104 opt_end = opt_pos;
1105 while (opt_end < end && *opt_end != '"') {
1106 opt_end++;
1107 }
1108 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
1109 return VMDK_ERROR;
1110 }
1111 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
1112 return VMDK_OK;
1113 }
1114
1115 /* Open an extent file and append to bs array */
1116 static int GRAPH_RDLOCK
vmdk_open_sparse(BlockDriverState * bs,BdrvChild * file,int flags,char * buf,QDict * options,Error ** errp)1117 vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
1118 char *buf, QDict *options, Error **errp)
1119 {
1120 uint32_t magic;
1121
1122 magic = ldl_be_p(buf);
1123 switch (magic) {
1124 case VMDK3_MAGIC:
1125 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
1126 case VMDK4_MAGIC:
1127 return vmdk_open_vmdk4(bs, file, flags, options, errp);
1128 default:
1129 error_setg(errp, "Image not in VMDK format");
1130 return -EINVAL;
1131 }
1132 }
1133
next_line(const char * s)1134 static const char *next_line(const char *s)
1135 {
1136 while (*s) {
1137 if (*s == '\n') {
1138 return s + 1;
1139 }
1140 s++;
1141 }
1142 return s;
1143 }
1144
1145 static int GRAPH_RDLOCK
vmdk_parse_extents(const char * desc,BlockDriverState * bs,QDict * options,Error ** errp)1146 vmdk_parse_extents(const char *desc, BlockDriverState *bs, QDict *options,
1147 Error **errp)
1148 {
1149 ERRP_GUARD();
1150 int ret;
1151 int matches;
1152 char access[11];
1153 char type[11];
1154 char fname[512];
1155 const char *p, *np;
1156 int64_t sectors = 0;
1157 int64_t flat_offset;
1158 char *desc_file_dir = NULL;
1159 char *extent_path;
1160 BdrvChild *extent_file;
1161 BdrvChildRole extent_role;
1162 BDRVVmdkState *s = bs->opaque;
1163 VmdkExtent *extent = NULL;
1164 char extent_opt_prefix[32];
1165 Error *local_err = NULL;
1166
1167 GLOBAL_STATE_CODE();
1168
1169 for (p = desc; *p; p = next_line(p)) {
1170 /* parse extent line in one of below formats:
1171 *
1172 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
1173 * RW [size in sectors] SPARSE "file-name.vmdk"
1174 * RW [size in sectors] VMFS "file-name.vmdk"
1175 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
1176 * RW [size in sectors] SESPARSE "file-name.vmdk"
1177 */
1178 flat_offset = -1;
1179 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
1180 access, §ors, type, fname, &flat_offset);
1181 if (matches < 4 || strcmp(access, "RW")) {
1182 continue;
1183 } else if (!strcmp(type, "FLAT")) {
1184 if (matches != 5 || flat_offset < 0) {
1185 goto invalid;
1186 }
1187 } else if (!strcmp(type, "VMFS")) {
1188 if (matches == 4) {
1189 flat_offset = 0;
1190 } else {
1191 goto invalid;
1192 }
1193 } else if (matches != 4) {
1194 goto invalid;
1195 }
1196
1197 if (sectors <= 0 ||
1198 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
1199 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") &&
1200 strcmp(type, "SESPARSE")) ||
1201 (strcmp(access, "RW"))) {
1202 continue;
1203 }
1204
1205 if (path_is_absolute(fname)) {
1206 extent_path = g_strdup(fname);
1207 } else {
1208 if (!desc_file_dir) {
1209 desc_file_dir = bdrv_dirname(bs->file->bs, errp);
1210 if (!desc_file_dir) {
1211 bdrv_refresh_filename(bs->file->bs);
1212 error_prepend(errp, "Cannot use relative paths with VMDK "
1213 "descriptor file '%s': ",
1214 bs->file->bs->filename);
1215 ret = -EINVAL;
1216 goto out;
1217 }
1218 }
1219
1220 extent_path = g_strconcat(desc_file_dir, fname, NULL);
1221 }
1222
1223 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
1224 assert(ret < 32);
1225
1226 extent_role = BDRV_CHILD_DATA;
1227 if (strcmp(type, "FLAT") != 0 && strcmp(type, "VMFS") != 0) {
1228 /* non-flat extents have metadata */
1229 extent_role |= BDRV_CHILD_METADATA;
1230 }
1231
1232 bdrv_graph_rdunlock_main_loop();
1233 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
1234 bs, &child_of_bds, extent_role, false,
1235 &local_err);
1236 bdrv_graph_rdlock_main_loop();
1237 g_free(extent_path);
1238 if (!extent_file) {
1239 error_propagate(errp, local_err);
1240 ret = -EINVAL;
1241 goto out;
1242 }
1243
1244 /* save to extents array */
1245 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
1246 /* FLAT extent */
1247
1248 ret = vmdk_add_extent(bs, extent_file, true, sectors,
1249 0, 0, 0, 0, 0, &extent, errp);
1250 if (ret < 0) {
1251 bdrv_graph_rdunlock_main_loop();
1252 bdrv_graph_wrlock_drained();
1253 bdrv_unref_child(bs, extent_file);
1254 bdrv_graph_wrunlock();
1255 bdrv_graph_rdlock_main_loop();
1256 goto out;
1257 }
1258 extent->flat_start_offset = flat_offset << 9;
1259 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
1260 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
1261 char *buf = vmdk_read_desc(extent_file, 0, errp);
1262 if (!buf) {
1263 ret = -EINVAL;
1264 } else {
1265 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
1266 options, errp);
1267 }
1268 g_free(buf);
1269 if (ret) {
1270 bdrv_graph_rdunlock_main_loop();
1271 bdrv_graph_wrlock_drained();
1272 bdrv_unref_child(bs, extent_file);
1273 bdrv_graph_wrunlock();
1274 bdrv_graph_rdlock_main_loop();
1275 goto out;
1276 }
1277 extent = &s->extents[s->num_extents - 1];
1278 } else if (!strcmp(type, "SESPARSE")) {
1279 ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
1280 if (ret) {
1281 bdrv_graph_rdunlock_main_loop();
1282 bdrv_graph_wrlock_drained();
1283 bdrv_unref_child(bs, extent_file);
1284 bdrv_graph_wrunlock();
1285 bdrv_graph_rdlock_main_loop();
1286 goto out;
1287 }
1288 extent = &s->extents[s->num_extents - 1];
1289 } else {
1290 error_setg(errp, "Unsupported extent type '%s'", type);
1291 bdrv_graph_rdunlock_main_loop();
1292 bdrv_graph_wrlock_drained();
1293 bdrv_unref_child(bs, extent_file);
1294 bdrv_graph_wrunlock();
1295 bdrv_graph_rdlock_main_loop();
1296 ret = -ENOTSUP;
1297 goto out;
1298 }
1299 extent->type = g_strdup(type);
1300 }
1301
1302 ret = 0;
1303 goto out;
1304
1305 invalid:
1306 np = next_line(p);
1307 assert(np != p);
1308 if (np[-1] == '\n') {
1309 np--;
1310 }
1311 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
1312 ret = -EINVAL;
1313
1314 out:
1315 g_free(desc_file_dir);
1316 return ret;
1317 }
1318
1319 static int GRAPH_RDLOCK
vmdk_open_desc_file(BlockDriverState * bs,int flags,char * buf,QDict * options,Error ** errp)1320 vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, QDict *options,
1321 Error **errp)
1322 {
1323 int ret;
1324 char ct[128];
1325 BDRVVmdkState *s = bs->opaque;
1326
1327 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
1328 error_setg(errp, "invalid VMDK image descriptor");
1329 ret = -EINVAL;
1330 goto exit;
1331 }
1332 if (strcmp(ct, "monolithicFlat") &&
1333 strcmp(ct, "vmfs") &&
1334 strcmp(ct, "vmfsSparse") &&
1335 strcmp(ct, "seSparse") &&
1336 strcmp(ct, "twoGbMaxExtentSparse") &&
1337 strcmp(ct, "twoGbMaxExtentFlat")) {
1338 error_setg(errp, "Unsupported image type '%s'", ct);
1339 ret = -ENOTSUP;
1340 goto exit;
1341 }
1342 s->create_type = g_strdup(ct);
1343 s->desc_offset = 0;
1344 ret = vmdk_parse_extents(buf, bs, options, errp);
1345 exit:
1346 return ret;
1347 }
1348
vmdk_open(BlockDriverState * bs,QDict * options,int flags,Error ** errp)1349 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
1350 Error **errp)
1351 {
1352 char *buf;
1353 int ret;
1354 BDRVVmdkState *s = bs->opaque;
1355 uint32_t magic;
1356
1357 ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
1358 if (ret < 0) {
1359 return ret;
1360 }
1361
1362 GRAPH_RDLOCK_GUARD_MAINLOOP();
1363
1364 buf = vmdk_read_desc(bs->file, 0, errp);
1365 if (!buf) {
1366 return -EINVAL;
1367 }
1368
1369 magic = ldl_be_p(buf);
1370 switch (magic) {
1371 case VMDK3_MAGIC:
1372 case VMDK4_MAGIC:
1373 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
1374 errp);
1375 s->desc_offset = 0x200;
1376 break;
1377 default:
1378 /* No data in the descriptor file */
1379 bs->file->role &= ~BDRV_CHILD_DATA;
1380
1381 /* Must succeed because we have given up permissions if anything */
1382 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1383
1384 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1385 break;
1386 }
1387 if (ret) {
1388 goto fail;
1389 }
1390
1391 /* try to open parent images, if exist */
1392 ret = vmdk_parent_open(bs);
1393 if (ret) {
1394 goto fail;
1395 }
1396 ret = vmdk_read_cid(bs, 0, &s->cid);
1397 if (ret) {
1398 goto fail;
1399 }
1400 ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1401 if (ret) {
1402 goto fail;
1403 }
1404 qemu_co_mutex_init(&s->lock);
1405
1406 /* Disable migration when VMDK images are used */
1407 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1408 "does not support live migration",
1409 bdrv_get_device_or_node_name(bs));
1410 ret = migrate_add_blocker_normal(&s->migration_blocker, errp);
1411 if (ret < 0) {
1412 goto fail;
1413 }
1414
1415 g_free(buf);
1416 return 0;
1417
1418 fail:
1419 g_free(buf);
1420 g_free(s->create_type);
1421 s->create_type = NULL;
1422 vmdk_free_extents(bs);
1423 return ret;
1424 }
1425
1426
vmdk_refresh_limits(BlockDriverState * bs,Error ** errp)1427 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1428 {
1429 BDRVVmdkState *s = bs->opaque;
1430 int i;
1431
1432 for (i = 0; i < s->num_extents; i++) {
1433 if (!s->extents[i].flat) {
1434 bs->bl.pwrite_zeroes_alignment =
1435 MAX(bs->bl.pwrite_zeroes_alignment,
1436 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1437 }
1438 }
1439 }
1440
1441 /**
1442 * get_whole_cluster
1443 *
1444 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1445 * to the cluster at @cluster_sector_num. If @zeroed is true, we're overwriting
1446 * a zeroed cluster in the current layer and must not copy data from the
1447 * backing file.
1448 *
1449 * If @skip_start_sector < @skip_end_sector, the relative range
1450 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1451 * it for call to write user data in the request.
1452 */
1453 static int coroutine_fn GRAPH_RDLOCK
get_whole_cluster(BlockDriverState * bs,VmdkExtent * extent,uint64_t cluster_offset,uint64_t offset,uint64_t skip_start_bytes,uint64_t skip_end_bytes,bool zeroed)1454 get_whole_cluster(BlockDriverState *bs, VmdkExtent *extent,
1455 uint64_t cluster_offset, uint64_t offset,
1456 uint64_t skip_start_bytes, uint64_t skip_end_bytes,
1457 bool zeroed)
1458 {
1459 int ret = VMDK_OK;
1460 int64_t cluster_bytes;
1461 uint8_t *whole_grain;
1462 bool copy_from_backing;
1463
1464 /* For COW, align request sector_num to cluster start */
1465 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1466 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1467 whole_grain = qemu_blockalign(bs, cluster_bytes);
1468 copy_from_backing = bs->backing && !zeroed;
1469
1470 if (!copy_from_backing) {
1471 memset(whole_grain, 0, skip_start_bytes);
1472 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1473 }
1474
1475 assert(skip_end_bytes <= cluster_bytes);
1476 /* we will be here if it's first write on non-exist grain(cluster).
1477 * try to read from parent image, if exist */
1478 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1479 ret = VMDK_ERROR;
1480 goto exit;
1481 }
1482
1483 /* Read backing data before skip range */
1484 if (skip_start_bytes > 0) {
1485 if (copy_from_backing) {
1486 /* qcow2 emits this on bs->file instead of bs->backing */
1487 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_READ);
1488 ret = bdrv_co_pread(bs->backing, offset, skip_start_bytes,
1489 whole_grain, 0);
1490 if (ret < 0) {
1491 ret = VMDK_ERROR;
1492 goto exit;
1493 }
1494 }
1495 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_WRITE);
1496 ret = bdrv_co_pwrite(extent->file, cluster_offset, skip_start_bytes,
1497 whole_grain, 0);
1498 if (ret < 0) {
1499 ret = VMDK_ERROR;
1500 goto exit;
1501 }
1502 }
1503 /* Read backing data after skip range */
1504 if (skip_end_bytes < cluster_bytes) {
1505 if (copy_from_backing) {
1506 /* qcow2 emits this on bs->file instead of bs->backing */
1507 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_READ);
1508 ret = bdrv_co_pread(bs->backing, offset + skip_end_bytes,
1509 cluster_bytes - skip_end_bytes,
1510 whole_grain + skip_end_bytes, 0);
1511 if (ret < 0) {
1512 ret = VMDK_ERROR;
1513 goto exit;
1514 }
1515 }
1516 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_WRITE);
1517 ret = bdrv_co_pwrite(extent->file, cluster_offset + skip_end_bytes,
1518 cluster_bytes - skip_end_bytes,
1519 whole_grain + skip_end_bytes, 0);
1520 if (ret < 0) {
1521 ret = VMDK_ERROR;
1522 goto exit;
1523 }
1524 }
1525
1526 ret = VMDK_OK;
1527 exit:
1528 qemu_vfree(whole_grain);
1529 return ret;
1530 }
1531
1532 static int coroutine_fn GRAPH_RDLOCK
vmdk_L2update(VmdkExtent * extent,VmdkMetaData * m_data,uint32_t offset)1533 vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data, uint32_t offset)
1534 {
1535 offset = cpu_to_le32(offset);
1536 /* update L2 table */
1537 BLKDBG_CO_EVENT(extent->file, BLKDBG_L2_UPDATE);
1538 if (bdrv_co_pwrite(extent->file,
1539 ((int64_t)m_data->l2_offset * 512)
1540 + (m_data->l2_index * sizeof(offset)),
1541 sizeof(offset), &offset, 0) < 0) {
1542 return VMDK_ERROR;
1543 }
1544 /* update backup L2 table */
1545 if (extent->l1_backup_table_offset != 0) {
1546 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1547 if (bdrv_co_pwrite(extent->file,
1548 ((int64_t)m_data->l2_offset * 512)
1549 + (m_data->l2_index * sizeof(offset)),
1550 sizeof(offset), &offset, 0) < 0) {
1551 return VMDK_ERROR;
1552 }
1553 }
1554 if (bdrv_co_flush(extent->file->bs) < 0) {
1555 return VMDK_ERROR;
1556 }
1557 if (m_data->l2_cache_entry) {
1558 *m_data->l2_cache_entry = offset;
1559 }
1560
1561 return VMDK_OK;
1562 }
1563
1564 /**
1565 * get_cluster_offset
1566 *
1567 * Look up cluster offset in extent file by sector number, and store in
1568 * @cluster_offset.
1569 *
1570 * For flat extents, the start offset as parsed from the description file is
1571 * returned.
1572 *
1573 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1574 * offset for a new cluster and update L2 cache. If there is a backing file,
1575 * COW is done before returning; otherwise, zeroes are written to the allocated
1576 * cluster. Both COW and zero writing skips the sector range
1577 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1578 * has new data to write there.
1579 *
1580 * Returns: VMDK_OK if cluster exists and mapped in the image.
1581 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1582 * VMDK_ERROR if failed.
1583 */
1584 static int coroutine_fn GRAPH_RDLOCK
get_cluster_offset(BlockDriverState * bs,VmdkExtent * extent,VmdkMetaData * m_data,uint64_t offset,bool allocate,uint64_t * cluster_offset,uint64_t skip_start_bytes,uint64_t skip_end_bytes)1585 get_cluster_offset(BlockDriverState *bs, VmdkExtent *extent,
1586 VmdkMetaData *m_data, uint64_t offset, bool allocate,
1587 uint64_t *cluster_offset, uint64_t skip_start_bytes,
1588 uint64_t skip_end_bytes)
1589 {
1590 unsigned int l1_index, l2_offset, l2_index;
1591 int min_index, i, j;
1592 uint32_t min_count;
1593 void *l2_table;
1594 bool zeroed = false;
1595 int64_t ret;
1596 int64_t cluster_sector;
1597 unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
1598
1599 if (m_data) {
1600 m_data->new_allocation = false;
1601 }
1602 if (extent->flat) {
1603 *cluster_offset = extent->flat_start_offset;
1604 return VMDK_OK;
1605 }
1606
1607 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1608 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1609 if (l1_index >= extent->l1_size) {
1610 return VMDK_ERROR;
1611 }
1612 if (extent->sesparse) {
1613 uint64_t l2_offset_u64;
1614
1615 assert(extent->entry_size == sizeof(uint64_t));
1616
1617 l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
1618 if (l2_offset_u64 == 0) {
1619 l2_offset = 0;
1620 } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
1621 /*
1622 * Top most nibble is 0x1 if grain table is allocated.
1623 * strict check - top most 4 bytes must be 0x10000000 since max
1624 * supported size is 64TB for disk - so no more than 64TB / 16MB
1625 * grain directories which is smaller than uint32,
1626 * where 16MB is the only supported default grain table coverage.
1627 */
1628 return VMDK_ERROR;
1629 } else {
1630 l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
1631 l2_offset_u64 = extent->sesparse_l2_tables_offset +
1632 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
1633 if (l2_offset_u64 > 0x00000000ffffffff) {
1634 return VMDK_ERROR;
1635 }
1636 l2_offset = (unsigned int)(l2_offset_u64);
1637 }
1638 } else {
1639 assert(extent->entry_size == sizeof(uint32_t));
1640 l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
1641 }
1642 if (!l2_offset) {
1643 return VMDK_UNALLOC;
1644 }
1645 for (i = 0; i < L2_CACHE_SIZE; i++) {
1646 if (l2_offset == extent->l2_cache_offsets[i]) {
1647 /* increment the hit count */
1648 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1649 for (j = 0; j < L2_CACHE_SIZE; j++) {
1650 extent->l2_cache_counts[j] >>= 1;
1651 }
1652 }
1653 l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
1654 goto found;
1655 }
1656 }
1657 /* not found: load a new entry in the least used one */
1658 min_index = 0;
1659 min_count = 0xffffffff;
1660 for (i = 0; i < L2_CACHE_SIZE; i++) {
1661 if (extent->l2_cache_counts[i] < min_count) {
1662 min_count = extent->l2_cache_counts[i];
1663 min_index = i;
1664 }
1665 }
1666 l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
1667 BLKDBG_CO_EVENT(extent->file, BLKDBG_L2_LOAD);
1668 if (bdrv_co_pread(extent->file,
1669 (int64_t)l2_offset * 512,
1670 l2_size_bytes,
1671 l2_table, 0
1672 ) < 0) {
1673 return VMDK_ERROR;
1674 }
1675
1676 extent->l2_cache_offsets[min_index] = l2_offset;
1677 extent->l2_cache_counts[min_index] = 1;
1678 found:
1679 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1680 if (m_data) {
1681 m_data->l1_index = l1_index;
1682 m_data->l2_index = l2_index;
1683 m_data->l2_offset = l2_offset;
1684 m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
1685 }
1686
1687 if (extent->sesparse) {
1688 cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
1689 switch (cluster_sector & 0xf000000000000000) {
1690 case 0x0000000000000000:
1691 /* unallocated grain */
1692 if (cluster_sector != 0) {
1693 return VMDK_ERROR;
1694 }
1695 break;
1696 case 0x1000000000000000:
1697 /* scsi-unmapped grain - fallthrough */
1698 case 0x2000000000000000:
1699 /* zero grain */
1700 zeroed = true;
1701 break;
1702 case 0x3000000000000000:
1703 /* allocated grain */
1704 cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
1705 ((cluster_sector & 0x0000ffffffffffff) << 12));
1706 cluster_sector = extent->sesparse_clusters_offset +
1707 cluster_sector * extent->cluster_sectors;
1708 break;
1709 default:
1710 return VMDK_ERROR;
1711 }
1712 } else {
1713 cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
1714
1715 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1716 zeroed = true;
1717 }
1718 }
1719
1720 if (!cluster_sector || zeroed) {
1721 if (!allocate) {
1722 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1723 }
1724 assert(!extent->sesparse);
1725
1726 if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1727 return VMDK_ERROR;
1728 }
1729
1730 cluster_sector = extent->next_cluster_sector;
1731 extent->next_cluster_sector += extent->cluster_sectors;
1732
1733 /* First of all we write grain itself, to avoid race condition
1734 * that may to corrupt the image.
1735 * This problem may occur because of insufficient space on host disk
1736 * or inappropriate VM shutdown.
1737 */
1738 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1739 offset, skip_start_bytes, skip_end_bytes,
1740 zeroed);
1741 if (ret) {
1742 return ret;
1743 }
1744 if (m_data) {
1745 m_data->new_allocation = true;
1746 }
1747 }
1748 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1749 return VMDK_OK;
1750 }
1751
find_extent(BDRVVmdkState * s,int64_t sector_num,VmdkExtent * start_hint)1752 static VmdkExtent *find_extent(BDRVVmdkState *s,
1753 int64_t sector_num, VmdkExtent *start_hint)
1754 {
1755 VmdkExtent *extent = start_hint;
1756
1757 if (!extent) {
1758 extent = &s->extents[0];
1759 }
1760 while (extent < &s->extents[s->num_extents]) {
1761 if (sector_num < extent->end_sector) {
1762 return extent;
1763 }
1764 extent++;
1765 }
1766 return NULL;
1767 }
1768
vmdk_find_offset_in_cluster(VmdkExtent * extent,int64_t offset)1769 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1770 int64_t offset)
1771 {
1772 uint64_t extent_begin_offset, extent_relative_offset;
1773 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1774
1775 extent_begin_offset =
1776 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1777 extent_relative_offset = offset - extent_begin_offset;
1778 return extent_relative_offset % cluster_size;
1779 }
1780
1781 static int coroutine_fn GRAPH_RDLOCK
vmdk_co_block_status(BlockDriverState * bs,unsigned int mode,int64_t offset,int64_t bytes,int64_t * pnum,int64_t * map,BlockDriverState ** file)1782 vmdk_co_block_status(BlockDriverState *bs, unsigned int mode,
1783 int64_t offset, int64_t bytes, int64_t *pnum,
1784 int64_t *map, BlockDriverState **file)
1785 {
1786 BDRVVmdkState *s = bs->opaque;
1787 int64_t index_in_cluster, n, ret;
1788 uint64_t cluster_offset;
1789 VmdkExtent *extent;
1790
1791 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1792 if (!extent) {
1793 return -EIO;
1794 }
1795 qemu_co_mutex_lock(&s->lock);
1796 ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1797 0, 0);
1798 qemu_co_mutex_unlock(&s->lock);
1799
1800 index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1801 switch (ret) {
1802 case VMDK_ERROR:
1803 ret = -EIO;
1804 break;
1805 case VMDK_UNALLOC:
1806 ret = 0;
1807 break;
1808 case VMDK_ZEROED:
1809 ret = BDRV_BLOCK_ZERO;
1810 break;
1811 case VMDK_OK:
1812 ret = BDRV_BLOCK_DATA;
1813 if (!extent->compressed) {
1814 ret |= BDRV_BLOCK_OFFSET_VALID;
1815 *map = cluster_offset + index_in_cluster;
1816 if (extent->flat) {
1817 ret |= BDRV_BLOCK_RECURSE;
1818 }
1819 } else {
1820 ret |= BDRV_BLOCK_COMPRESSED;
1821 }
1822 *file = extent->file->bs;
1823 break;
1824 }
1825
1826 n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1827 *pnum = MIN(n, bytes);
1828 return ret;
1829 }
1830
1831 static int coroutine_fn GRAPH_RDLOCK
vmdk_write_extent(VmdkExtent * extent,int64_t cluster_offset,int64_t offset_in_cluster,QEMUIOVector * qiov,uint64_t qiov_offset,uint64_t n_bytes,uint64_t offset)1832 vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1833 int64_t offset_in_cluster, QEMUIOVector *qiov,
1834 uint64_t qiov_offset, uint64_t n_bytes,
1835 uint64_t offset)
1836 {
1837 int ret;
1838 VmdkGrainMarker *data = NULL;
1839 uLongf buf_len;
1840 QEMUIOVector local_qiov;
1841 int64_t write_offset;
1842 int64_t write_end_sector;
1843
1844 if (extent->compressed) {
1845 void *compressed_data;
1846
1847 /* Only whole clusters */
1848 if (offset_in_cluster ||
1849 n_bytes > (extent->cluster_sectors * SECTOR_SIZE) ||
1850 (n_bytes < (extent->cluster_sectors * SECTOR_SIZE) &&
1851 offset + n_bytes != extent->end_sector * SECTOR_SIZE))
1852 {
1853 ret = -EINVAL;
1854 goto out;
1855 }
1856
1857 if (!extent->has_marker) {
1858 ret = -EINVAL;
1859 goto out;
1860 }
1861 buf_len = (extent->cluster_sectors << 9) * 2;
1862 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1863
1864 compressed_data = g_malloc(n_bytes);
1865 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1866 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1867 g_free(compressed_data);
1868
1869 if (ret != Z_OK || buf_len == 0) {
1870 ret = -EINVAL;
1871 goto out;
1872 }
1873
1874 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1875 data->size = cpu_to_le32(buf_len);
1876
1877 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1878 qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1879
1880 BLKDBG_CO_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1881 } else {
1882 qemu_iovec_init(&local_qiov, qiov->niov);
1883 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1884
1885 BLKDBG_CO_EVENT(extent->file, BLKDBG_WRITE_AIO);
1886 }
1887
1888 write_offset = cluster_offset + offset_in_cluster;
1889 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1890 &local_qiov, 0);
1891
1892 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1893
1894 if (extent->compressed) {
1895 extent->next_cluster_sector = write_end_sector;
1896 } else {
1897 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1898 write_end_sector);
1899 }
1900
1901 if (ret < 0) {
1902 goto out;
1903 }
1904 ret = 0;
1905 out:
1906 g_free(data);
1907 if (!extent->compressed) {
1908 qemu_iovec_destroy(&local_qiov);
1909 }
1910 return ret;
1911 }
1912
1913 static int coroutine_fn GRAPH_RDLOCK
vmdk_read_extent(VmdkExtent * extent,int64_t cluster_offset,int64_t offset_in_cluster,QEMUIOVector * qiov,int bytes)1914 vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1915 int64_t offset_in_cluster, QEMUIOVector *qiov, int bytes)
1916 {
1917 int ret;
1918 int cluster_bytes, buf_bytes;
1919 uint8_t *cluster_buf, *compressed_data;
1920 uint8_t *uncomp_buf;
1921 uint32_t data_len;
1922 VmdkGrainMarker *marker;
1923 uLongf buf_len;
1924
1925
1926 if (!extent->compressed) {
1927 BLKDBG_CO_EVENT(extent->file, BLKDBG_READ_AIO);
1928 ret = bdrv_co_preadv(extent->file,
1929 cluster_offset + offset_in_cluster, bytes,
1930 qiov, 0);
1931 if (ret < 0) {
1932 return ret;
1933 }
1934 return 0;
1935 }
1936 cluster_bytes = extent->cluster_sectors * 512;
1937 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1938 buf_bytes = cluster_bytes * 2;
1939 cluster_buf = g_malloc(buf_bytes);
1940 uncomp_buf = g_malloc(cluster_bytes);
1941 BLKDBG_CO_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1942 ret = bdrv_co_pread(extent->file, cluster_offset, buf_bytes, cluster_buf,
1943 0);
1944 if (ret < 0) {
1945 goto out;
1946 }
1947 compressed_data = cluster_buf;
1948 buf_len = cluster_bytes;
1949 data_len = cluster_bytes;
1950 if (extent->has_marker) {
1951 marker = (VmdkGrainMarker *)cluster_buf;
1952 compressed_data = marker->data;
1953 data_len = le32_to_cpu(marker->size);
1954 }
1955 if (!data_len || data_len > buf_bytes) {
1956 ret = -EINVAL;
1957 goto out;
1958 }
1959 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1960 if (ret != Z_OK) {
1961 ret = -EINVAL;
1962 goto out;
1963
1964 }
1965 if (offset_in_cluster < 0 ||
1966 offset_in_cluster + bytes > buf_len) {
1967 ret = -EINVAL;
1968 goto out;
1969 }
1970 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1971 ret = 0;
1972
1973 out:
1974 g_free(uncomp_buf);
1975 g_free(cluster_buf);
1976 return ret;
1977 }
1978
1979 static int coroutine_fn GRAPH_RDLOCK
vmdk_co_preadv(BlockDriverState * bs,int64_t offset,int64_t bytes,QEMUIOVector * qiov,BdrvRequestFlags flags)1980 vmdk_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
1981 QEMUIOVector *qiov, BdrvRequestFlags flags)
1982 {
1983 BDRVVmdkState *s = bs->opaque;
1984 int ret;
1985 uint64_t n_bytes, offset_in_cluster;
1986 VmdkExtent *extent = NULL;
1987 QEMUIOVector local_qiov;
1988 uint64_t cluster_offset;
1989 uint64_t bytes_done = 0;
1990
1991 qemu_iovec_init(&local_qiov, qiov->niov);
1992 qemu_co_mutex_lock(&s->lock);
1993
1994 while (bytes > 0) {
1995 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1996 if (!extent) {
1997 ret = -EIO;
1998 goto fail;
1999 }
2000 ret = get_cluster_offset(bs, extent, NULL,
2001 offset, false, &cluster_offset, 0, 0);
2002 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
2003
2004 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
2005 - offset_in_cluster);
2006
2007 if (ret != VMDK_OK) {
2008 /* if not allocated, try to read from parent image, if exist */
2009 if (bs->backing && ret != VMDK_ZEROED) {
2010 if (!vmdk_is_cid_valid(bs)) {
2011 ret = -EINVAL;
2012 goto fail;
2013 }
2014
2015 qemu_iovec_reset(&local_qiov);
2016 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
2017
2018 /* qcow2 emits this on bs->file instead of bs->backing */
2019 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
2020 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
2021 &local_qiov, 0);
2022 if (ret < 0) {
2023 goto fail;
2024 }
2025 } else {
2026 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
2027 }
2028 } else {
2029 qemu_iovec_reset(&local_qiov);
2030 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
2031
2032 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
2033 &local_qiov, n_bytes);
2034 if (ret) {
2035 goto fail;
2036 }
2037 }
2038 bytes -= n_bytes;
2039 offset += n_bytes;
2040 bytes_done += n_bytes;
2041 }
2042
2043 ret = 0;
2044 fail:
2045 qemu_co_mutex_unlock(&s->lock);
2046 qemu_iovec_destroy(&local_qiov);
2047
2048 return ret;
2049 }
2050
2051 /**
2052 * vmdk_write:
2053 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
2054 * if possible, otherwise return -ENOTSUP.
2055 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
2056 * with each cluster. By dry run we can find if the zero write
2057 * is possible without modifying image data.
2058 *
2059 * Returns: error code with 0 for success.
2060 */
2061 static int coroutine_fn GRAPH_RDLOCK
vmdk_pwritev(BlockDriverState * bs,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov,bool zeroed,bool zero_dry_run)2062 vmdk_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2063 QEMUIOVector *qiov, bool zeroed, bool zero_dry_run)
2064 {
2065 BDRVVmdkState *s = bs->opaque;
2066 VmdkExtent *extent = NULL;
2067 int ret;
2068 int64_t offset_in_cluster, n_bytes;
2069 uint64_t cluster_offset;
2070 uint64_t bytes_done = 0;
2071 VmdkMetaData m_data;
2072
2073 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
2074 error_report("Wrong offset: offset=0x%" PRIx64
2075 " total_sectors=0x%" PRIx64,
2076 offset, bs->total_sectors);
2077 return -EIO;
2078 }
2079
2080 while (bytes > 0) {
2081 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
2082 if (!extent) {
2083 return -EIO;
2084 }
2085 if (extent->sesparse) {
2086 return -ENOTSUP;
2087 }
2088 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
2089 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
2090 - offset_in_cluster);
2091
2092 ret = get_cluster_offset(bs, extent, &m_data, offset,
2093 !(extent->compressed || zeroed),
2094 &cluster_offset, offset_in_cluster,
2095 offset_in_cluster + n_bytes);
2096 if (extent->compressed) {
2097 if (ret == VMDK_OK) {
2098 /* Refuse write to allocated cluster for streamOptimized */
2099 error_report("Could not write to allocated cluster"
2100 " for streamOptimized");
2101 return -EIO;
2102 } else if (!zeroed) {
2103 /* allocate */
2104 ret = get_cluster_offset(bs, extent, &m_data, offset,
2105 true, &cluster_offset, 0, 0);
2106 }
2107 }
2108 if (ret == VMDK_ERROR) {
2109 return -EINVAL;
2110 }
2111 if (zeroed) {
2112 /* Do zeroed write, buf is ignored */
2113 if (extent->has_zero_grain &&
2114 offset_in_cluster == 0 &&
2115 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
2116 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
2117 if (!zero_dry_run && ret != VMDK_ZEROED) {
2118 /* update L2 tables */
2119 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
2120 != VMDK_OK) {
2121 return -EIO;
2122 }
2123 }
2124 } else {
2125 return -ENOTSUP;
2126 }
2127 } else {
2128 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
2129 qiov, bytes_done, n_bytes, offset);
2130 if (ret) {
2131 return ret;
2132 }
2133 if (m_data.new_allocation) {
2134 /* update L2 tables */
2135 if (vmdk_L2update(extent, &m_data,
2136 cluster_offset >> BDRV_SECTOR_BITS)
2137 != VMDK_OK) {
2138 return -EIO;
2139 }
2140 }
2141 }
2142 bytes -= n_bytes;
2143 offset += n_bytes;
2144 bytes_done += n_bytes;
2145
2146 /* update CID on the first write every time the virtual disk is
2147 * opened */
2148 if (!s->cid_updated) {
2149 ret = vmdk_write_cid(bs, g_random_int());
2150 if (ret < 0) {
2151 return ret;
2152 }
2153 s->cid_updated = true;
2154 }
2155 }
2156 return 0;
2157 }
2158
2159 static int coroutine_fn GRAPH_RDLOCK
vmdk_co_pwritev(BlockDriverState * bs,int64_t offset,int64_t bytes,QEMUIOVector * qiov,BdrvRequestFlags flags)2160 vmdk_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
2161 QEMUIOVector *qiov, BdrvRequestFlags flags)
2162 {
2163 int ret;
2164 BDRVVmdkState *s = bs->opaque;
2165 qemu_co_mutex_lock(&s->lock);
2166 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
2167 qemu_co_mutex_unlock(&s->lock);
2168 return ret;
2169 }
2170
2171 static int coroutine_fn GRAPH_RDLOCK
vmdk_co_pwritev_compressed(BlockDriverState * bs,int64_t offset,int64_t bytes,QEMUIOVector * qiov)2172 vmdk_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes,
2173 QEMUIOVector *qiov)
2174 {
2175 if (bytes == 0) {
2176 /* The caller will write bytes 0 to signal EOF.
2177 * When receive it, we align EOF to a sector boundary. */
2178 BDRVVmdkState *s = bs->opaque;
2179 int i, ret;
2180 int64_t length;
2181
2182 for (i = 0; i < s->num_extents; i++) {
2183 length = bdrv_co_getlength(s->extents[i].file->bs);
2184 if (length < 0) {
2185 return length;
2186 }
2187 length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
2188 ret = bdrv_co_truncate(s->extents[i].file, length, false,
2189 PREALLOC_MODE_OFF, 0, NULL);
2190 if (ret < 0) {
2191 return ret;
2192 }
2193 }
2194 return 0;
2195 }
2196 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
2197 }
2198
2199 static int coroutine_fn GRAPH_RDLOCK
vmdk_co_pwrite_zeroes(BlockDriverState * bs,int64_t offset,int64_t bytes,BdrvRequestFlags flags)2200 vmdk_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
2201 BdrvRequestFlags flags)
2202 {
2203 int ret;
2204 BDRVVmdkState *s = bs->opaque;
2205
2206 qemu_co_mutex_lock(&s->lock);
2207 /* write zeroes could fail if sectors not aligned to cluster, test it with
2208 * dry_run == true before really updating image */
2209 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
2210 if (!ret) {
2211 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
2212 }
2213 qemu_co_mutex_unlock(&s->lock);
2214 return ret;
2215 }
2216
2217 static int coroutine_fn GRAPH_UNLOCKED
vmdk_init_extent(BlockBackend * blk,int64_t filesize,bool flat,bool compress,bool zeroed_grain,Error ** errp)2218 vmdk_init_extent(BlockBackend *blk, int64_t filesize, bool flat, bool compress,
2219 bool zeroed_grain, Error **errp)
2220 {
2221 int ret, i;
2222 VMDK4Header header;
2223 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
2224 uint32_t *gd_buf = NULL;
2225 int gd_buf_size;
2226
2227 if (flat) {
2228 ret = blk_co_truncate(blk, filesize, false, PREALLOC_MODE_OFF, 0, errp);
2229 goto exit;
2230 }
2231 magic = cpu_to_be32(VMDK4_MAGIC);
2232 memset(&header, 0, sizeof(header));
2233 if (compress) {
2234 header.version = 3;
2235 } else if (zeroed_grain) {
2236 header.version = 2;
2237 } else {
2238 header.version = 1;
2239 }
2240 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
2241 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
2242 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
2243 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
2244 header.capacity = filesize / BDRV_SECTOR_SIZE;
2245 header.granularity = 128;
2246 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
2247
2248 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
2249 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
2250 BDRV_SECTOR_SIZE);
2251 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
2252 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
2253
2254 header.desc_offset = 1;
2255 header.desc_size = 20;
2256 header.rgd_offset = header.desc_offset + header.desc_size;
2257 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
2258 header.grain_offset =
2259 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
2260 header.granularity);
2261 /* swap endianness for all header fields */
2262 header.version = cpu_to_le32(header.version);
2263 header.flags = cpu_to_le32(header.flags);
2264 header.capacity = cpu_to_le64(header.capacity);
2265 header.granularity = cpu_to_le64(header.granularity);
2266 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
2267 header.desc_offset = cpu_to_le64(header.desc_offset);
2268 header.desc_size = cpu_to_le64(header.desc_size);
2269 header.rgd_offset = cpu_to_le64(header.rgd_offset);
2270 header.gd_offset = cpu_to_le64(header.gd_offset);
2271 header.grain_offset = cpu_to_le64(header.grain_offset);
2272 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
2273
2274 header.check_bytes[0] = 0xa;
2275 header.check_bytes[1] = 0x20;
2276 header.check_bytes[2] = 0xd;
2277 header.check_bytes[3] = 0xa;
2278
2279 /* write all the data */
2280 ret = blk_co_pwrite(blk, 0, sizeof(magic), &magic, 0);
2281 if (ret < 0) {
2282 error_setg_errno(errp, -ret, "failed to write VMDK magic");
2283 goto exit;
2284 }
2285 ret = blk_co_pwrite(blk, sizeof(magic), sizeof(header), &header, 0);
2286 if (ret < 0) {
2287 error_setg_errno(errp, -ret, "failed to write VMDK header");
2288 goto exit;
2289 }
2290
2291 ret = blk_co_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false,
2292 PREALLOC_MODE_OFF, 0, errp);
2293 if (ret < 0) {
2294 goto exit;
2295 }
2296
2297 /* write grain directory */
2298 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
2299 gd_buf = g_malloc0(gd_buf_size);
2300 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
2301 i < gt_count; i++, tmp += gt_size) {
2302 gd_buf[i] = cpu_to_le32(tmp);
2303 }
2304 ret = blk_co_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
2305 gd_buf_size, gd_buf, 0);
2306 if (ret < 0) {
2307 error_setg_errno(errp, -ret, "failed to write VMDK grain directory");
2308 goto exit;
2309 }
2310
2311 /* write backup grain directory */
2312 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
2313 i < gt_count; i++, tmp += gt_size) {
2314 gd_buf[i] = cpu_to_le32(tmp);
2315 }
2316 ret = blk_co_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
2317 gd_buf_size, gd_buf, 0);
2318 if (ret < 0) {
2319 error_setg_errno(errp, -ret,
2320 "failed to write VMDK backup grain directory");
2321 }
2322
2323 ret = 0;
2324 exit:
2325 g_free(gd_buf);
2326 return ret;
2327 }
2328
2329 static int coroutine_fn GRAPH_UNLOCKED
vmdk_create_extent(const char * filename,int64_t filesize,bool flat,bool compress,bool zeroed_grain,BlockBackend ** pbb,QemuOpts * opts,Error ** errp)2330 vmdk_create_extent(const char *filename, int64_t filesize, bool flat,
2331 bool compress, bool zeroed_grain, BlockBackend **pbb,
2332 QemuOpts *opts, Error **errp)
2333 {
2334 int ret;
2335 BlockBackend *blk = NULL;
2336
2337 ret = bdrv_co_create_file(filename, opts, errp);
2338 if (ret < 0) {
2339 goto exit;
2340 }
2341
2342 blk = blk_co_new_open(filename, NULL, NULL,
2343 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2344 errp);
2345 if (blk == NULL) {
2346 ret = -EIO;
2347 goto exit;
2348 }
2349
2350 blk_set_allow_write_beyond_eof(blk, true);
2351
2352 ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
2353 exit:
2354 if (blk) {
2355 if (pbb) {
2356 *pbb = blk;
2357 } else {
2358 blk_co_unref(blk);
2359 blk = NULL;
2360 }
2361 }
2362 return ret;
2363 }
2364
filename_decompose(const char * filename,char * path,char * prefix,char * postfix,size_t buf_len,Error ** errp)2365 static int filename_decompose(const char *filename, char *path, char *prefix,
2366 char *postfix, size_t buf_len, Error **errp)
2367 {
2368 const char *p, *q;
2369
2370 if (filename == NULL || !strlen(filename)) {
2371 error_setg(errp, "No filename provided");
2372 return VMDK_ERROR;
2373 }
2374 p = strrchr(filename, '/');
2375 if (p == NULL) {
2376 p = strrchr(filename, '\\');
2377 }
2378 if (p == NULL) {
2379 p = strrchr(filename, ':');
2380 }
2381 if (p != NULL) {
2382 p++;
2383 if (p - filename >= buf_len) {
2384 return VMDK_ERROR;
2385 }
2386 pstrcpy(path, p - filename + 1, filename);
2387 } else {
2388 p = filename;
2389 path[0] = '\0';
2390 }
2391 q = strrchr(p, '.');
2392 if (q == NULL) {
2393 pstrcpy(prefix, buf_len, p);
2394 postfix[0] = '\0';
2395 } else {
2396 if (q - p >= buf_len) {
2397 return VMDK_ERROR;
2398 }
2399 pstrcpy(prefix, q - p + 1, p);
2400 pstrcpy(postfix, buf_len, q);
2401 }
2402 return VMDK_OK;
2403 }
2404
2405 /*
2406 * idx == 0: get or create the descriptor file (also the image file if in a
2407 * non-split format.
2408 * idx >= 1: get the n-th extent if in a split subformat
2409 */
2410 typedef BlockBackend * coroutine_fn GRAPH_UNLOCKED_PTR
2411 (*vmdk_create_extent_fn)(int64_t size, int idx, bool flat, bool split,
2412 bool compress, bool zeroed_grain, void *opaque,
2413 Error **errp);
2414
vmdk_desc_add_extent(GString * desc,const char * extent_line_fmt,int64_t size,const char * filename)2415 static void vmdk_desc_add_extent(GString *desc,
2416 const char *extent_line_fmt,
2417 int64_t size, const char *filename)
2418 {
2419 char *basename = g_path_get_basename(filename);
2420
2421 g_string_append_printf(desc, extent_line_fmt,
2422 DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
2423 g_free(basename);
2424 }
2425
2426 static int coroutine_fn GRAPH_UNLOCKED
vmdk_co_do_create(int64_t size,BlockdevVmdkSubformat subformat,BlockdevVmdkAdapterType adapter_type,const char * backing_file,const char * hw_version,const char * toolsversion,bool compat6,bool zeroed_grain,vmdk_create_extent_fn extent_fn,void * opaque,Error ** errp)2427 vmdk_co_do_create(int64_t size,
2428 BlockdevVmdkSubformat subformat,
2429 BlockdevVmdkAdapterType adapter_type,
2430 const char *backing_file,
2431 const char *hw_version,
2432 const char *toolsversion,
2433 bool compat6,
2434 bool zeroed_grain,
2435 vmdk_create_extent_fn extent_fn,
2436 void *opaque,
2437 Error **errp)
2438 {
2439 int extent_idx;
2440 BlockBackend *blk = NULL;
2441 BlockBackend *extent_blk;
2442 Error *local_err = NULL;
2443 char *desc = NULL;
2444 int ret = 0;
2445 bool flat, split, compress;
2446 GString *ext_desc_lines;
2447 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
2448 int64_t extent_size;
2449 int64_t created_size = 0;
2450 const char *extent_line_fmt;
2451 char *parent_desc_line = g_malloc0(BUF_SIZE);
2452 uint32_t parent_cid = 0xffffffff;
2453 uint32_t number_heads = 16;
2454 uint32_t desc_offset = 0, desc_len;
2455 const char desc_template[] =
2456 "# Disk DescriptorFile\n"
2457 "version=1\n"
2458 "CID=%" PRIx32 "\n"
2459 "parentCID=%" PRIx32 "\n"
2460 "createType=\"%s\"\n"
2461 "%s"
2462 "\n"
2463 "# Extent description\n"
2464 "%s"
2465 "\n"
2466 "# The Disk Data Base\n"
2467 "#DDB\n"
2468 "\n"
2469 "ddb.virtualHWVersion = \"%s\"\n"
2470 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2471 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2472 "ddb.geometry.sectors = \"63\"\n"
2473 "ddb.adapterType = \"%s\"\n"
2474 "ddb.toolsVersion = \"%s\"\n";
2475
2476 ext_desc_lines = g_string_new(NULL);
2477
2478 /* Read out options */
2479 if (compat6) {
2480 if (hw_version) {
2481 error_setg(errp,
2482 "compat6 cannot be enabled with hwversion set");
2483 ret = -EINVAL;
2484 goto exit;
2485 }
2486 hw_version = "6";
2487 }
2488 if (!hw_version) {
2489 hw_version = "4";
2490 }
2491 if (!toolsversion) {
2492 toolsversion = "2147483647";
2493 }
2494
2495 if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2496 /* that's the number of heads with which vmware operates when
2497 creating, exporting, etc. vmdk files with a non-ide adapter type */
2498 number_heads = 255;
2499 }
2500 split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2501 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2502 flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2503 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2504 compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2505
2506 if (flat) {
2507 extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2508 } else {
2509 extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2510 }
2511 if (flat && backing_file) {
2512 error_setg(errp, "Flat image can't have backing file");
2513 ret = -ENOTSUP;
2514 goto exit;
2515 }
2516 if (flat && zeroed_grain) {
2517 error_setg(errp, "Flat image can't enable zeroed grain");
2518 ret = -ENOTSUP;
2519 goto exit;
2520 }
2521
2522 /* Create extents */
2523 if (split) {
2524 extent_size = split_size;
2525 } else {
2526 extent_size = size;
2527 }
2528 if (!split && !flat) {
2529 created_size = extent_size;
2530 } else {
2531 created_size = 0;
2532 }
2533 /* Get the descriptor file BDS */
2534 blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2535 opaque, errp);
2536 if (!blk) {
2537 ret = -EIO;
2538 goto exit;
2539 }
2540 if (!split && !flat) {
2541 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2542 blk_bs(blk)->filename);
2543 }
2544
2545 if (backing_file) {
2546 BlockBackend *backing;
2547 char *full_backing =
2548 bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2549 backing_file,
2550 &local_err);
2551 if (local_err) {
2552 error_propagate(errp, local_err);
2553 ret = -ENOENT;
2554 goto exit;
2555 }
2556 assert(full_backing);
2557
2558 backing = blk_co_new_open(full_backing, NULL, NULL,
2559 BDRV_O_NO_BACKING, errp);
2560 g_free(full_backing);
2561 if (backing == NULL) {
2562 ret = -EIO;
2563 goto exit;
2564 }
2565 if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2566 error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2567 blk_bs(backing)->drv->format_name);
2568 blk_co_unref(backing);
2569 ret = -EINVAL;
2570 goto exit;
2571 }
2572
2573 bdrv_graph_co_rdlock();
2574 ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2575 bdrv_graph_co_rdunlock();
2576 blk_co_unref(backing);
2577 if (ret) {
2578 error_setg(errp, "Failed to read parent CID");
2579 goto exit;
2580 }
2581 snprintf(parent_desc_line, BUF_SIZE,
2582 "parentFileNameHint=\"%s\"", backing_file);
2583 }
2584 extent_idx = 1;
2585 while (created_size < size) {
2586 int64_t cur_size = MIN(size - created_size, extent_size);
2587 extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2588 zeroed_grain, opaque, errp);
2589 if (!extent_blk) {
2590 ret = -EINVAL;
2591 goto exit;
2592 }
2593 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2594 blk_bs(extent_blk)->filename);
2595 created_size += cur_size;
2596 extent_idx++;
2597 blk_co_unref(extent_blk);
2598 }
2599
2600 /* Check whether we got excess extents */
2601 extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2602 opaque, NULL);
2603 if (extent_blk) {
2604 blk_co_unref(extent_blk);
2605 error_setg(errp, "List of extents contains unused extents");
2606 ret = -EINVAL;
2607 goto exit;
2608 }
2609
2610 /* generate descriptor file */
2611 desc = g_strdup_printf(desc_template,
2612 g_random_int(),
2613 parent_cid,
2614 BlockdevVmdkSubformat_str(subformat),
2615 parent_desc_line,
2616 ext_desc_lines->str,
2617 hw_version,
2618 size /
2619 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2620 number_heads,
2621 BlockdevVmdkAdapterType_str(adapter_type),
2622 toolsversion);
2623 desc_len = strlen(desc);
2624 /* the descriptor offset = 0x200 */
2625 if (!split && !flat) {
2626 desc_offset = 0x200;
2627 }
2628
2629 ret = blk_co_pwrite(blk, desc_offset, desc_len, desc, 0);
2630 if (ret < 0) {
2631 error_setg_errno(errp, -ret, "Could not write description");
2632 goto exit;
2633 }
2634 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2635 * for description file */
2636 if (desc_offset == 0) {
2637 ret = blk_co_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, 0, errp);
2638 if (ret < 0) {
2639 goto exit;
2640 }
2641 }
2642 ret = 0;
2643 exit:
2644 if (blk) {
2645 blk_co_unref(blk);
2646 }
2647 g_free(desc);
2648 g_free(parent_desc_line);
2649 g_string_free(ext_desc_lines, true);
2650 return ret;
2651 }
2652
2653 typedef struct {
2654 char *path;
2655 char *prefix;
2656 char *postfix;
2657 QemuOpts *opts;
2658 } VMDKCreateOptsData;
2659
2660 static BlockBackend * coroutine_fn GRAPH_UNLOCKED
vmdk_co_create_opts_cb(int64_t size,int idx,bool flat,bool split,bool compress,bool zeroed_grain,void * opaque,Error ** errp)2661 vmdk_co_create_opts_cb(int64_t size, int idx, bool flat, bool split,
2662 bool compress, bool zeroed_grain, void *opaque,
2663 Error **errp)
2664 {
2665 BlockBackend *blk = NULL;
2666 BlockDriverState *bs = NULL;
2667 VMDKCreateOptsData *data = opaque;
2668 char *ext_filename = NULL;
2669 char *rel_filename = NULL;
2670
2671 /* We're done, don't create excess extents. */
2672 if (size == -1) {
2673 assert(errp == NULL);
2674 return NULL;
2675 }
2676
2677 if (idx == 0) {
2678 rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2679 } else if (split) {
2680 rel_filename = g_strdup_printf("%s-%c%03d%s",
2681 data->prefix,
2682 flat ? 'f' : 's', idx, data->postfix);
2683 } else {
2684 assert(idx == 1);
2685 rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2686 }
2687
2688 ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2689 g_free(rel_filename);
2690
2691 if (vmdk_create_extent(ext_filename, size,
2692 flat, compress, zeroed_grain, &blk, data->opts,
2693 errp)) {
2694 goto exit;
2695 }
2696 bdrv_co_unref(bs);
2697 exit:
2698 g_free(ext_filename);
2699 return blk;
2700 }
2701
2702 static int coroutine_fn GRAPH_UNLOCKED
vmdk_co_create_opts(BlockDriver * drv,const char * filename,QemuOpts * opts,Error ** errp)2703 vmdk_co_create_opts(BlockDriver *drv, const char *filename,
2704 QemuOpts *opts, Error **errp)
2705 {
2706 Error *local_err = NULL;
2707 char *desc = NULL;
2708 int64_t total_size = 0;
2709 char *adapter_type = NULL;
2710 BlockdevVmdkAdapterType adapter_type_enum;
2711 char *backing_file = NULL;
2712 char *hw_version = NULL;
2713 char *toolsversion = NULL;
2714 char *fmt = NULL;
2715 BlockdevVmdkSubformat subformat;
2716 int ret = 0;
2717 char *path = g_malloc0(PATH_MAX);
2718 char *prefix = g_malloc0(PATH_MAX);
2719 char *postfix = g_malloc0(PATH_MAX);
2720 char *desc_line = g_malloc0(BUF_SIZE);
2721 char *ext_filename = g_malloc0(PATH_MAX);
2722 char *desc_filename = g_malloc0(PATH_MAX);
2723 char *parent_desc_line = g_malloc0(BUF_SIZE);
2724 bool zeroed_grain;
2725 bool compat6;
2726 VMDKCreateOptsData data;
2727 char *backing_fmt = NULL;
2728
2729 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2730 if (backing_fmt && strcmp(backing_fmt, "vmdk") != 0) {
2731 error_setg(errp, "backing_file must be a vmdk image");
2732 ret = -EINVAL;
2733 goto exit;
2734 }
2735
2736 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
2737 ret = -EINVAL;
2738 goto exit;
2739 }
2740 /* Read out options */
2741 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2742 BDRV_SECTOR_SIZE);
2743 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
2744 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2745 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
2746 toolsversion = qemu_opt_get_del(opts, BLOCK_OPT_TOOLSVERSION);
2747 compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false);
2748 if (strcmp(hw_version, "undefined") == 0) {
2749 g_free(hw_version);
2750 hw_version = NULL;
2751 }
2752 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
2753 zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false);
2754
2755 if (adapter_type) {
2756 adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup,
2757 adapter_type,
2758 BLOCKDEV_VMDK_ADAPTER_TYPE_IDE,
2759 &local_err);
2760 if (local_err) {
2761 error_propagate(errp, local_err);
2762 ret = -EINVAL;
2763 goto exit;
2764 }
2765 } else {
2766 adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE;
2767 }
2768
2769 if (!fmt) {
2770 /* Default format to monolithicSparse */
2771 subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE;
2772 } else {
2773 subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup,
2774 fmt,
2775 BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE,
2776 &local_err);
2777 if (local_err) {
2778 error_propagate(errp, local_err);
2779 ret = -EINVAL;
2780 goto exit;
2781 }
2782 }
2783 data = (VMDKCreateOptsData){
2784 .prefix = prefix,
2785 .postfix = postfix,
2786 .path = path,
2787 .opts = opts,
2788 };
2789 ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum,
2790 backing_file, hw_version, toolsversion, compat6,
2791 zeroed_grain, vmdk_co_create_opts_cb, &data, errp);
2792
2793 exit:
2794 g_free(backing_fmt);
2795 g_free(adapter_type);
2796 g_free(backing_file);
2797 g_free(hw_version);
2798 g_free(toolsversion);
2799 g_free(fmt);
2800 g_free(desc);
2801 g_free(path);
2802 g_free(prefix);
2803 g_free(postfix);
2804 g_free(desc_line);
2805 g_free(ext_filename);
2806 g_free(desc_filename);
2807 g_free(parent_desc_line);
2808 return ret;
2809 }
2810
2811 static BlockBackend * coroutine_fn GRAPH_UNLOCKED
vmdk_co_create_cb(int64_t size,int idx,bool flat,bool split,bool compress,bool zeroed_grain,void * opaque,Error ** errp)2812 vmdk_co_create_cb(int64_t size, int idx, bool flat, bool split, bool compress,
2813 bool zeroed_grain, void *opaque, Error **errp)
2814 {
2815 int ret;
2816 BlockDriverState *bs;
2817 BlockBackend *blk;
2818 BlockdevCreateOptionsVmdk *opts = opaque;
2819
2820 if (idx == 0) {
2821 bs = bdrv_co_open_blockdev_ref(opts->file, errp);
2822 } else {
2823 int i;
2824 BlockdevRefList *list = opts->extents;
2825 for (i = 1; i < idx; i++) {
2826 if (!list || !list->next) {
2827 error_setg(errp, "Extent [%d] not specified", i);
2828 return NULL;
2829 }
2830 list = list->next;
2831 }
2832 if (!list) {
2833 error_setg(errp, "Extent [%d] not specified", idx - 1);
2834 return NULL;
2835 }
2836 bs = bdrv_co_open_blockdev_ref(list->value, errp);
2837 }
2838 if (!bs) {
2839 return NULL;
2840 }
2841 blk = blk_co_new_with_bs(bs,
2842 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
2843 BLK_PERM_RESIZE,
2844 BLK_PERM_ALL,
2845 errp);
2846 if (!blk) {
2847 return NULL;
2848 }
2849 blk_set_allow_write_beyond_eof(blk, true);
2850 bdrv_co_unref(bs);
2851
2852 if (size != -1) {
2853 ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2854 if (ret) {
2855 blk_co_unref(blk);
2856 blk = NULL;
2857 }
2858 }
2859 return blk;
2860 }
2861
2862 static int coroutine_fn GRAPH_UNLOCKED
vmdk_co_create(BlockdevCreateOptions * create_options,Error ** errp)2863 vmdk_co_create(BlockdevCreateOptions *create_options, Error **errp)
2864 {
2865 BlockdevCreateOptionsVmdk *opts;
2866
2867 opts = &create_options->u.vmdk;
2868
2869 /* Validate options */
2870 if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2871 error_setg(errp, "Image size must be a multiple of 512 bytes");
2872 return -EINVAL;
2873 }
2874
2875 return vmdk_co_do_create(opts->size,
2876 opts->subformat,
2877 opts->adapter_type,
2878 opts->backing_file,
2879 opts->hwversion,
2880 opts->toolsversion,
2881 false,
2882 opts->zeroed_grain,
2883 vmdk_co_create_cb,
2884 opts, errp);
2885 }
2886
vmdk_close(BlockDriverState * bs)2887 static void vmdk_close(BlockDriverState *bs)
2888 {
2889 BDRVVmdkState *s = bs->opaque;
2890
2891 vmdk_free_extents(bs);
2892 g_free(s->create_type);
2893
2894 migrate_del_blocker(&s->migration_blocker);
2895 }
2896
2897 static int64_t coroutine_fn GRAPH_RDLOCK
vmdk_co_get_allocated_file_size(BlockDriverState * bs)2898 vmdk_co_get_allocated_file_size(BlockDriverState *bs)
2899 {
2900 int i;
2901 int64_t ret = 0;
2902 int64_t r;
2903 BDRVVmdkState *s = bs->opaque;
2904
2905 ret = bdrv_co_get_allocated_file_size(bs->file->bs);
2906 if (ret < 0) {
2907 return ret;
2908 }
2909 for (i = 0; i < s->num_extents; i++) {
2910 if (s->extents[i].file == bs->file) {
2911 continue;
2912 }
2913 r = bdrv_co_get_allocated_file_size(s->extents[i].file->bs);
2914 if (r < 0) {
2915 return r;
2916 }
2917 ret += r;
2918 }
2919 return ret;
2920 }
2921
vmdk_has_zero_init(BlockDriverState * bs)2922 static int GRAPH_RDLOCK vmdk_has_zero_init(BlockDriverState *bs)
2923 {
2924 int i;
2925 BDRVVmdkState *s = bs->opaque;
2926
2927 /* If has a flat extent and its underlying storage doesn't have zero init,
2928 * return 0. */
2929 for (i = 0; i < s->num_extents; i++) {
2930 if (s->extents[i].flat) {
2931 if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2932 return 0;
2933 }
2934 }
2935 }
2936 return 1;
2937 }
2938
vmdk_get_extent_info(VmdkExtent * extent)2939 static VmdkExtentInfo * GRAPH_RDLOCK vmdk_get_extent_info(VmdkExtent *extent)
2940 {
2941 VmdkExtentInfo *info = g_new0(VmdkExtentInfo, 1);
2942
2943 bdrv_refresh_filename(extent->file->bs);
2944 *info = (VmdkExtentInfo){
2945 .filename = g_strdup(extent->file->bs->filename),
2946 .format = g_strdup(extent->type),
2947 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2948 .compressed = extent->compressed,
2949 .has_compressed = extent->compressed,
2950 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2951 .has_cluster_size = !extent->flat,
2952 };
2953
2954 return info;
2955 }
2956
2957 static int coroutine_fn GRAPH_RDLOCK
vmdk_co_check(BlockDriverState * bs,BdrvCheckResult * result,BdrvCheckMode fix)2958 vmdk_co_check(BlockDriverState *bs, BdrvCheckResult *result, BdrvCheckMode fix)
2959 {
2960 BDRVVmdkState *s = bs->opaque;
2961 VmdkExtent *extent = NULL;
2962 int64_t sector_num = 0;
2963 int64_t total_sectors = bdrv_co_nb_sectors(bs);
2964 int ret;
2965 uint64_t cluster_offset;
2966
2967 if (fix) {
2968 return -ENOTSUP;
2969 }
2970
2971 for (;;) {
2972 if (sector_num >= total_sectors) {
2973 return 0;
2974 }
2975 extent = find_extent(s, sector_num, extent);
2976 if (!extent) {
2977 fprintf(stderr,
2978 "ERROR: could not find extent for sector %" PRId64 "\n",
2979 sector_num);
2980 ret = -EINVAL;
2981 break;
2982 }
2983 ret = get_cluster_offset(bs, extent, NULL,
2984 sector_num << BDRV_SECTOR_BITS,
2985 false, &cluster_offset, 0, 0);
2986 if (ret == VMDK_ERROR) {
2987 fprintf(stderr,
2988 "ERROR: could not get cluster_offset for sector %"
2989 PRId64 "\n", sector_num);
2990 break;
2991 }
2992 if (ret == VMDK_OK) {
2993 int64_t extent_len = bdrv_co_getlength(extent->file->bs);
2994 if (extent_len < 0) {
2995 fprintf(stderr,
2996 "ERROR: could not get extent file length for sector %"
2997 PRId64 "\n", sector_num);
2998 ret = extent_len;
2999 break;
3000 }
3001 if (cluster_offset >= extent_len) {
3002 fprintf(stderr,
3003 "ERROR: cluster offset for sector %"
3004 PRId64 " points after EOF\n", sector_num);
3005 ret = -EINVAL;
3006 break;
3007 }
3008 }
3009 sector_num += extent->cluster_sectors;
3010 }
3011
3012 result->corruptions++;
3013 return ret;
3014 }
3015
3016 static ImageInfoSpecific * GRAPH_RDLOCK
vmdk_get_specific_info(BlockDriverState * bs,Error ** errp)3017 vmdk_get_specific_info(BlockDriverState *bs, Error **errp)
3018 {
3019 int i;
3020 BDRVVmdkState *s = bs->opaque;
3021 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
3022 VmdkExtentInfoList **tail;
3023
3024 *spec_info = (ImageInfoSpecific){
3025 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
3026 .u = {
3027 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
3028 },
3029 };
3030
3031 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
3032 .create_type = g_strdup(s->create_type),
3033 .cid = s->cid,
3034 .parent_cid = s->parent_cid,
3035 };
3036
3037 tail = &spec_info->u.vmdk.data->extents;
3038 for (i = 0; i < s->num_extents; i++) {
3039 QAPI_LIST_APPEND(tail, vmdk_get_extent_info(&s->extents[i]));
3040 }
3041
3042 return spec_info;
3043 }
3044
vmdk_extents_type_eq(const VmdkExtent * a,const VmdkExtent * b)3045 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
3046 {
3047 return a->flat == b->flat &&
3048 a->compressed == b->compressed &&
3049 (a->flat || a->cluster_sectors == b->cluster_sectors);
3050 }
3051
3052 static int coroutine_fn
vmdk_co_get_info(BlockDriverState * bs,BlockDriverInfo * bdi)3053 vmdk_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3054 {
3055 int i;
3056 BDRVVmdkState *s = bs->opaque;
3057 assert(s->num_extents);
3058
3059 /* See if we have multiple extents but they have different cases */
3060 for (i = 1; i < s->num_extents; i++) {
3061 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
3062 return -ENOTSUP;
3063 }
3064 }
3065 bdi->needs_compressed_writes = s->extents[0].compressed;
3066 if (!s->extents[0].flat) {
3067 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
3068 }
3069 return 0;
3070 }
3071
3072 static void GRAPH_RDLOCK
vmdk_gather_child_options(BlockDriverState * bs,QDict * target,bool backing_overridden)3073 vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
3074 bool backing_overridden)
3075 {
3076 /* No children but file and backing can be explicitly specified (TODO) */
3077 qdict_put(target, "file",
3078 qobject_ref(bs->file->bs->full_open_options));
3079
3080 if (backing_overridden) {
3081 if (bs->backing) {
3082 qdict_put(target, "backing",
3083 qobject_ref(bs->backing->bs->full_open_options));
3084 } else {
3085 qdict_put_null(target, "backing");
3086 }
3087 }
3088 }
3089
3090 static QemuOptsList vmdk_create_opts = {
3091 .name = "vmdk-create-opts",
3092 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
3093 .desc = {
3094 {
3095 .name = BLOCK_OPT_SIZE,
3096 .type = QEMU_OPT_SIZE,
3097 .help = "Virtual disk size"
3098 },
3099 {
3100 .name = BLOCK_OPT_ADAPTER_TYPE,
3101 .type = QEMU_OPT_STRING,
3102 .help = "Virtual adapter type, can be one of "
3103 "ide (default), lsilogic, buslogic or legacyESX"
3104 },
3105 {
3106 .name = BLOCK_OPT_BACKING_FILE,
3107 .type = QEMU_OPT_STRING,
3108 .help = "File name of a base image"
3109 },
3110 {
3111 .name = BLOCK_OPT_BACKING_FMT,
3112 .type = QEMU_OPT_STRING,
3113 .help = "Must be 'vmdk' if present",
3114 },
3115 {
3116 .name = BLOCK_OPT_COMPAT6,
3117 .type = QEMU_OPT_BOOL,
3118 .help = "VMDK version 6 image",
3119 .def_value_str = "off"
3120 },
3121 {
3122 .name = BLOCK_OPT_HWVERSION,
3123 .type = QEMU_OPT_STRING,
3124 .help = "VMDK hardware version",
3125 .def_value_str = "undefined"
3126 },
3127 {
3128 .name = BLOCK_OPT_TOOLSVERSION,
3129 .type = QEMU_OPT_STRING,
3130 .help = "VMware guest tools version",
3131 },
3132 {
3133 .name = BLOCK_OPT_SUBFMT,
3134 .type = QEMU_OPT_STRING,
3135 .help =
3136 "VMDK flat extent format, can be one of "
3137 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
3138 },
3139 {
3140 .name = BLOCK_OPT_ZEROED_GRAIN,
3141 .type = QEMU_OPT_BOOL,
3142 .help = "Enable efficient zero writes "
3143 "using the zeroed-grain GTE feature"
3144 },
3145 { /* end of list */ }
3146 }
3147 };
3148
3149 static BlockDriver bdrv_vmdk = {
3150 .format_name = "vmdk",
3151 .instance_size = sizeof(BDRVVmdkState),
3152 .bdrv_probe = vmdk_probe,
3153 .bdrv_open = vmdk_open,
3154 .bdrv_co_check = vmdk_co_check,
3155 .bdrv_reopen_prepare = vmdk_reopen_prepare,
3156 .bdrv_reopen_commit = vmdk_reopen_commit,
3157 .bdrv_reopen_abort = vmdk_reopen_abort,
3158 .bdrv_child_perm = bdrv_default_perms,
3159 .bdrv_co_preadv = vmdk_co_preadv,
3160 .bdrv_co_pwritev = vmdk_co_pwritev,
3161 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
3162 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
3163 .bdrv_close = vmdk_close,
3164 .bdrv_co_create_opts = vmdk_co_create_opts,
3165 .bdrv_co_create = vmdk_co_create,
3166 .bdrv_co_block_status = vmdk_co_block_status,
3167 .bdrv_co_get_allocated_file_size = vmdk_co_get_allocated_file_size,
3168 .bdrv_has_zero_init = vmdk_has_zero_init,
3169 .bdrv_get_specific_info = vmdk_get_specific_info,
3170 .bdrv_refresh_limits = vmdk_refresh_limits,
3171 .bdrv_co_get_info = vmdk_co_get_info,
3172 .bdrv_gather_child_options = vmdk_gather_child_options,
3173
3174 .is_format = true,
3175 .supports_backing = true,
3176 .create_opts = &vmdk_create_opts,
3177 };
3178
bdrv_vmdk_init(void)3179 static void bdrv_vmdk_init(void)
3180 {
3181 bdrv_register(&bdrv_vmdk);
3182 }
3183
3184 block_init(bdrv_vmdk_init);
3185