?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
block: Mark bdrv_unref_child() GRAPH_WRLOCK
[qemu/ar7.git] / block / vmdk.c
blob78baa04c0c17033a2ce6d36bff523f7fb4b55686
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 "sysemu/block-backend.h"
30 #include "qapi/qmp/qdict.h"
31 #include "qapi/qmp/qerror.h"
32 #include "qemu/error-report.h"
33 #include "qemu/module.h"
34 #include "qemu/option.h"
35 #include "qemu/bswap.h"
36 #include "qemu/memalign.h"
37 #include "migration/blocker.h"
38 #include "qemu/cutils.h"
39 #include <zlib.h>
40
41 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
42 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
43 #define VMDK4_COMPRESSION_DEFLATE 1
44 #define VMDK4_FLAG_NL_DETECT (1 << 0)
45 #define VMDK4_FLAG_RGD (1 << 1)
46 /* Zeroed-grain enable bit */
47 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
48 #define VMDK4_FLAG_COMPRESS (1 << 16)
49 #define VMDK4_FLAG_MARKER (1 << 17)
50 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
51
52 #define VMDK_EXTENT_MAX_SECTORS (1ULL << 32)
53
54 #define VMDK_GTE_ZEROED 0x1
55
56 /* VMDK internal error codes */
57 #define VMDK_OK 0
58 #define VMDK_ERROR (-1)
59 /* Cluster not allocated */
60 #define VMDK_UNALLOC (-2)
61 #define VMDK_ZEROED (-3)
62
63 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
64 #define BLOCK_OPT_TOOLSVERSION "toolsversion"
65
66 typedef struct {
67 uint32_t version;
68 uint32_t flags;
69 uint32_t disk_sectors;
70 uint32_t granularity;
71 uint32_t l1dir_offset;
72 uint32_t l1dir_size;
73 uint32_t file_sectors;
74 uint32_t cylinders;
75 uint32_t heads;
76 uint32_t sectors_per_track;
77 } QEMU_PACKED VMDK3Header;
78
79 typedef struct {
80 uint32_t version;
81 uint32_t flags;
82 uint64_t capacity;
83 uint64_t granularity;
84 uint64_t desc_offset;
85 uint64_t desc_size;
86 /* Number of GrainTableEntries per GrainTable */
87 uint32_t num_gtes_per_gt;
88 uint64_t rgd_offset;
89 uint64_t gd_offset;
90 uint64_t grain_offset;
91 char filler[1];
92 char check_bytes[4];
93 uint16_t compressAlgorithm;
94 } QEMU_PACKED VMDK4Header;
95
96 typedef struct VMDKSESparseConstHeader {
97 uint64_t magic;
98 uint64_t version;
99 uint64_t capacity;
100 uint64_t grain_size;
101 uint64_t grain_table_size;
102 uint64_t flags;
103 uint64_t reserved1;
104 uint64_t reserved2;
105 uint64_t reserved3;
106 uint64_t reserved4;
107 uint64_t volatile_header_offset;
108 uint64_t volatile_header_size;
109 uint64_t journal_header_offset;
110 uint64_t journal_header_size;
111 uint64_t journal_offset;
112 uint64_t journal_size;
113 uint64_t grain_dir_offset;
114 uint64_t grain_dir_size;
115 uint64_t grain_tables_offset;
116 uint64_t grain_tables_size;
117 uint64_t free_bitmap_offset;
118 uint64_t free_bitmap_size;
119 uint64_t backmap_offset;
120 uint64_t backmap_size;
121 uint64_t grains_offset;
122 uint64_t grains_size;
123 uint8_t pad[304];
124 } QEMU_PACKED VMDKSESparseConstHeader;
125
126 typedef struct VMDKSESparseVolatileHeader {
127 uint64_t magic;
128 uint64_t free_gt_number;
129 uint64_t next_txn_seq_number;
130 uint64_t replay_journal;
131 uint8_t pad[480];
132 } QEMU_PACKED VMDKSESparseVolatileHeader;
133
134 #define L2_CACHE_SIZE 16
135
136 typedef struct VmdkExtent {
137 BdrvChild *file;
138 bool flat;
139 bool compressed;
140 bool has_marker;
141 bool has_zero_grain;
142 bool sesparse;
143 uint64_t sesparse_l2_tables_offset;
144 uint64_t sesparse_clusters_offset;
145 int32_t entry_size;
146 int version;
147 int64_t sectors;
148 int64_t end_sector;
149 int64_t flat_start_offset;
150 int64_t l1_table_offset;
151 int64_t l1_backup_table_offset;
152 void *l1_table;
153 uint32_t *l1_backup_table;
154 unsigned int l1_size;
155 uint32_t l1_entry_sectors;
156
157 unsigned int l2_size;
158 void *l2_cache;
159 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
160 uint32_t l2_cache_counts[L2_CACHE_SIZE];
161
162 int64_t cluster_sectors;
163 int64_t next_cluster_sector;
164 char *type;
165 } VmdkExtent;
166
167 typedef struct BDRVVmdkState {
168 CoMutex lock;
169 uint64_t desc_offset;
170 bool cid_updated;
171 bool cid_checked;
172 uint32_t cid;
173 uint32_t parent_cid;
174 int num_extents;
175 /* Extent array with num_extents entries, ascend ordered by address */
176 VmdkExtent *extents;
177 Error *migration_blocker;
178 char *create_type;
179 } BDRVVmdkState;
180
181 typedef struct BDRVVmdkReopenState {
182 bool *extents_using_bs_file;
183 } BDRVVmdkReopenState;
184
185 typedef struct VmdkMetaData {
186 unsigned int l1_index;
187 unsigned int l2_index;
188 unsigned int l2_offset;
189 bool new_allocation;
190 uint32_t *l2_cache_entry;
191 } VmdkMetaData;
192
193 typedef struct VmdkGrainMarker {
194 uint64_t lba;
195 uint32_t size;
196 uint8_t data[];
197 } QEMU_PACKED VmdkGrainMarker;
198
199 enum {
200 MARKER_END_OF_STREAM = 0,
201 MARKER_GRAIN_TABLE = 1,
202 MARKER_GRAIN_DIRECTORY = 2,
203 MARKER_FOOTER = 3,
204 };
205
206 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
207 {
208 uint32_t magic;
209
210 if (buf_size < 4) {
211 return 0;
212 }
213 magic = be32_to_cpu(*(uint32_t *)buf);
214 if (magic == VMDK3_MAGIC ||
215 magic == VMDK4_MAGIC) {
216 return 100;
217 } else {
218 const char *p = (const char *)buf;
219 const char *end = p + buf_size;
220 while (p < end) {
221 if (*p == '#') {
222 /* skip comment line */
223 while (p < end && *p != '\n') {
224 p++;
225 }
226 p++;
227 continue;
228 }
229 if (*p == ' ') {
230 while (p < end && *p == ' ') {
231 p++;
232 }
233 /* skip '\r' if windows line endings used. */
234 if (p < end && *p == '\r') {
235 p++;
236 }
237 /* only accept blank lines before 'version=' line */
238 if (p == end || *p != '\n') {
239 return 0;
240 }
241 p++;
242 continue;
243 }
244 if (end - p >= strlen("version=X\n")) {
245 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
246 strncmp("version=2\n", p, strlen("version=2\n")) == 0 ||
247 strncmp("version=3\n", p, strlen("version=3\n")) == 0) {
248 return 100;
249 }
250 }
251 if (end - p >= strlen("version=X\r\n")) {
252 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
253 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0 ||
254 strncmp("version=3\r\n", p, strlen("version=3\r\n")) == 0) {
255 return 100;
256 }
257 }
258 return 0;
259 }
260 return 0;
261 }
262 }
263
264 #define SECTOR_SIZE 512
265 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
266 #define BUF_SIZE 4096
267 #define HEADER_SIZE 512 /* first sector of 512 bytes */
268
269 static void vmdk_free_extents(BlockDriverState *bs)
270 {
271 int i;
272 BDRVVmdkState *s = bs->opaque;
273 VmdkExtent *e;
274
275 bdrv_graph_wrlock(NULL);
276 for (i = 0; i < s->num_extents; i++) {
277 e = &s->extents[i];
278 g_free(e->l1_table);
279 g_free(e->l2_cache);
280 g_free(e->l1_backup_table);
281 g_free(e->type);
282 if (e->file != bs->file) {
283 bdrv_unref_child(bs, e->file);
284 }
285 }
286 bdrv_graph_wrunlock();
287
288 g_free(s->extents);
289 }
290
291 static void vmdk_free_last_extent(BlockDriverState *bs)
292 {
293 BDRVVmdkState *s = bs->opaque;
294
295 if (s->num_extents == 0) {
296 return;
297 }
298 s->num_extents--;
299 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
300 }
301
302 /* Return -ve errno, or 0 on success and write CID into *pcid. */
303 static int 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
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 desc = g_malloc0(DESC_SIZE);
354 tmp_desc = g_malloc0(DESC_SIZE);
355 ret = bdrv_co_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
356 if (ret < 0) {
357 goto out;
358 }
359
360 desc[DESC_SIZE - 1] = '\0';
361 tmp_str = strstr(desc, "parentCID");
362 if (tmp_str == NULL) {
363 ret = -EINVAL;
364 goto out;
365 }
366
367 pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
368 p_name = strstr(desc, "CID");
369 if (p_name != NULL) {
370 p_name += sizeof("CID");
371 snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
372 pstrcat(desc, DESC_SIZE, tmp_desc);
373 }
374
375 ret = bdrv_co_pwrite_sync(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
376
377 out:
378 g_free(desc);
379 g_free(tmp_desc);
380 return ret;
381 }
382
383 static int coroutine_fn vmdk_is_cid_valid(BlockDriverState *bs)
384 {
385 BDRVVmdkState *s = bs->opaque;
386 uint32_t cur_pcid;
387
388 if (!s->cid_checked && bs->backing) {
389 BlockDriverState *p_bs = bs->backing->bs;
390
391 if (strcmp(p_bs->drv->format_name, "vmdk")) {
392 /* Backing file is not in vmdk format, so it does not have
393 * a CID, which makes the overlay's parent CID invalid */
394 return 0;
395 }
396
397 if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
398 /* read failure: report as not valid */
399 return 0;
400 }
401 if (s->parent_cid != cur_pcid) {
402 /* CID not valid */
403 return 0;
404 }
405 }
406 s->cid_checked = true;
407 /* CID valid */
408 return 1;
409 }
410
411 static int vmdk_reopen_prepare(BDRVReopenState *state,
412 BlockReopenQueue *queue, Error **errp)
413 {
414 BDRVVmdkState *s;
415 BDRVVmdkReopenState *rs;
416 int i;
417
418 assert(state != NULL);
419 assert(state->bs != NULL);
420 assert(state->opaque == NULL);
421
422 s = state->bs->opaque;
423
424 rs = g_new0(BDRVVmdkReopenState, 1);
425 state->opaque = rs;
426
427 /*
428 * Check whether there are any extents stored in bs->file; if bs->file
429 * changes, we will need to update their .file pointers to follow suit
430 */
431 rs->extents_using_bs_file = g_new(bool, s->num_extents);
432 for (i = 0; i < s->num_extents; i++) {
433 rs->extents_using_bs_file[i] = s->extents[i].file == state->bs->file;
434 }
435
436 return 0;
437 }
438
439 static void vmdk_reopen_clean(BDRVReopenState *state)
440 {
441 BDRVVmdkReopenState *rs = state->opaque;
442
443 g_free(rs->extents_using_bs_file);
444 g_free(rs);
445 state->opaque = NULL;
446 }
447
448 static void vmdk_reopen_commit(BDRVReopenState *state)
449 {
450 BDRVVmdkState *s = state->bs->opaque;
451 BDRVVmdkReopenState *rs = state->opaque;
452 int i;
453
454 for (i = 0; i < s->num_extents; i++) {
455 if (rs->extents_using_bs_file[i]) {
456 s->extents[i].file = state->bs->file;
457 }
458 }
459
460 vmdk_reopen_clean(state);
461 }
462
463 static void vmdk_reopen_abort(BDRVReopenState *state)
464 {
465 vmdk_reopen_clean(state);
466 }
467
468 static int vmdk_parent_open(BlockDriverState *bs)
469 {
470 char *p_name;
471 char *desc;
472 BDRVVmdkState *s = bs->opaque;
473 int ret;
474
475 desc = g_malloc0(DESC_SIZE + 1);
476 ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
477 if (ret < 0) {
478 goto out;
479 }
480
481 p_name = strstr(desc, "parentFileNameHint");
482 if (p_name != NULL) {
483 char *end_name;
484
485 p_name += sizeof("parentFileNameHint") + 1;
486 end_name = strchr(p_name, '\"');
487 if (end_name == NULL) {
488 ret = -EINVAL;
489 goto out;
490 }
491 if ((end_name - p_name) > sizeof(bs->auto_backing_file) - 1) {
492 ret = -EINVAL;
493 goto out;
494 }
495
496 pstrcpy(bs->auto_backing_file, end_name - p_name + 1, p_name);
497 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
498 bs->auto_backing_file);
499 pstrcpy(bs->backing_format, sizeof(bs->backing_format),
500 "vmdk");
501 }
502
503 out:
504 g_free(desc);
505 return ret;
506 }
507
508 /* Create and append extent to the extent array. Return the added VmdkExtent
509 * address. return NULL if allocation failed. */
510 static int vmdk_add_extent(BlockDriverState *bs,
511 BdrvChild *file, bool flat, int64_t sectors,
512 int64_t l1_offset, int64_t l1_backup_offset,
513 uint32_t l1_size,
514 int l2_size, uint64_t cluster_sectors,
515 VmdkExtent **new_extent,
516 Error **errp)
517 {
518 VmdkExtent *extent;
519 BDRVVmdkState *s = bs->opaque;
520 int64_t nb_sectors;
521
522 if (cluster_sectors > 0x200000) {
523 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
524 error_setg(errp, "Invalid granularity, image may be corrupt");
525 return -EFBIG;
526 }
527 if (l1_size > 32 * 1024 * 1024) {
528 /*
529 * Although with big capacity and small l1_entry_sectors, we can get a
530 * big l1_size, we don't want unbounded value to allocate the table.
531 * Limit it to 32M, which is enough to store:
532 * 8TB - for both VMDK3 & VMDK4 with
533 * minimal cluster size: 512B
534 * minimal L2 table size: 512 entries
535 * 8 TB is still more than the maximal value supported for
536 * VMDK3 & VMDK4 which is 2TB.
537 * 64TB - for "ESXi seSparse Extent"
538 * minimal cluster size: 512B (default is 4KB)
539 * L2 table size: 4096 entries (const).
540 * 64TB is more than the maximal value supported for
541 * seSparse VMDKs (which is slightly less than 64TB)
542 */
543 error_setg(errp, "L1 size too big");
544 return -EFBIG;
545 }
546
547 nb_sectors = bdrv_nb_sectors(file->bs);
548 if (nb_sectors < 0) {
549 return nb_sectors;
550 }
551
552 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
553 extent = &s->extents[s->num_extents];
554 s->num_extents++;
555
556 memset(extent, 0, sizeof(VmdkExtent));
557 extent->file = file;
558 extent->flat = flat;
559 extent->sectors = sectors;
560 extent->l1_table_offset = l1_offset;
561 extent->l1_backup_table_offset = l1_backup_offset;
562 extent->l1_size = l1_size;
563 extent->l1_entry_sectors = l2_size * cluster_sectors;
564 extent->l2_size = l2_size;
565 extent->cluster_sectors = flat ? sectors : cluster_sectors;
566 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
567 extent->entry_size = sizeof(uint32_t);
568
569 if (s->num_extents > 1) {
570 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
571 } else {
572 extent->end_sector = extent->sectors;
573 }
574 bs->total_sectors = extent->end_sector;
575 if (new_extent) {
576 *new_extent = extent;
577 }
578 return 0;
579 }
580
581 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
582 Error **errp)
583 {
584 int ret;
585 size_t l1_size;
586 int i;
587
588 /* read the L1 table */
589 l1_size = extent->l1_size * extent->entry_size;
590 extent->l1_table = g_try_malloc(l1_size);
591 if (l1_size && extent->l1_table == NULL) {
592 return -ENOMEM;
593 }
594
595 ret = bdrv_pread(extent->file, extent->l1_table_offset, l1_size,
596 extent->l1_table, 0);
597 if (ret < 0) {
598 bdrv_refresh_filename(extent->file->bs);
599 error_setg_errno(errp, -ret,
600 "Could not read l1 table from extent '%s'",
601 extent->file->bs->filename);
602 goto fail_l1;
603 }
604 for (i = 0; i < extent->l1_size; i++) {
605 if (extent->entry_size == sizeof(uint64_t)) {
606 le64_to_cpus((uint64_t *)extent->l1_table + i);
607 } else {
608 assert(extent->entry_size == sizeof(uint32_t));
609 le32_to_cpus((uint32_t *)extent->l1_table + i);
610 }
611 }
612
613 if (extent->l1_backup_table_offset) {
614 assert(!extent->sesparse);
615 extent->l1_backup_table = g_try_malloc(l1_size);
616 if (l1_size && extent->l1_backup_table == NULL) {
617 ret = -ENOMEM;
618 goto fail_l1;
619 }
620 ret = bdrv_pread(extent->file, extent->l1_backup_table_offset,
621 l1_size, extent->l1_backup_table, 0);
622 if (ret < 0) {
623 bdrv_refresh_filename(extent->file->bs);
624 error_setg_errno(errp, -ret,
625 "Could not read l1 backup table from extent '%s'",
626 extent->file->bs->filename);
627 goto fail_l1b;
628 }
629 for (i = 0; i < extent->l1_size; i++) {
630 le32_to_cpus(&extent->l1_backup_table[i]);
631 }
632 }
633
634 extent->l2_cache =
635 g_malloc(extent->entry_size * extent->l2_size * L2_CACHE_SIZE);
636 return 0;
637 fail_l1b:
638 g_free(extent->l1_backup_table);
639 fail_l1:
640 g_free(extent->l1_table);
641 return ret;
642 }
643
644 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
645 BdrvChild *file,
646 int flags, Error **errp)
647 {
648 int ret;
649 uint32_t magic;
650 VMDK3Header header;
651 VmdkExtent *extent = NULL;
652
653 ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0);
654 if (ret < 0) {
655 bdrv_refresh_filename(file->bs);
656 error_setg_errno(errp, -ret,
657 "Could not read header from file '%s'",
658 file->bs->filename);
659 return ret;
660 }
661 ret = vmdk_add_extent(bs, file, false,
662 le32_to_cpu(header.disk_sectors),
663 (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
664 0,
665 le32_to_cpu(header.l1dir_size),
666 4096,
667 le32_to_cpu(header.granularity),
668 &extent,
669 errp);
670 if (ret < 0) {
671 return ret;
672 }
673 ret = vmdk_init_tables(bs, extent, errp);
674 if (ret) {
675 /* free extent allocated by vmdk_add_extent */
676 vmdk_free_last_extent(bs);
677 }
678 return ret;
679 }
680
681 #define SESPARSE_CONST_HEADER_MAGIC UINT64_C(0x00000000cafebabe)
682 #define SESPARSE_VOLATILE_HEADER_MAGIC UINT64_C(0x00000000cafecafe)
683
684 /* Strict checks - format not officially documented */
685 static int check_se_sparse_const_header(VMDKSESparseConstHeader *header,
686 Error **errp)
687 {
688 header->magic = le64_to_cpu(header->magic);
689 header->version = le64_to_cpu(header->version);
690 header->grain_size = le64_to_cpu(header->grain_size);
691 header->grain_table_size = le64_to_cpu(header->grain_table_size);
692 header->flags = le64_to_cpu(header->flags);
693 header->reserved1 = le64_to_cpu(header->reserved1);
694 header->reserved2 = le64_to_cpu(header->reserved2);
695 header->reserved3 = le64_to_cpu(header->reserved3);
696 header->reserved4 = le64_to_cpu(header->reserved4);
697
698 header->volatile_header_offset =
699 le64_to_cpu(header->volatile_header_offset);
700 header->volatile_header_size = le64_to_cpu(header->volatile_header_size);
701
702 header->journal_header_offset = le64_to_cpu(header->journal_header_offset);
703 header->journal_header_size = le64_to_cpu(header->journal_header_size);
704
705 header->journal_offset = le64_to_cpu(header->journal_offset);
706 header->journal_size = le64_to_cpu(header->journal_size);
707
708 header->grain_dir_offset = le64_to_cpu(header->grain_dir_offset);
709 header->grain_dir_size = le64_to_cpu(header->grain_dir_size);
710
711 header->grain_tables_offset = le64_to_cpu(header->grain_tables_offset);
712 header->grain_tables_size = le64_to_cpu(header->grain_tables_size);
713
714 header->free_bitmap_offset = le64_to_cpu(header->free_bitmap_offset);
715 header->free_bitmap_size = le64_to_cpu(header->free_bitmap_size);
716
717 header->backmap_offset = le64_to_cpu(header->backmap_offset);
718 header->backmap_size = le64_to_cpu(header->backmap_size);
719
720 header->grains_offset = le64_to_cpu(header->grains_offset);
721 header->grains_size = le64_to_cpu(header->grains_size);
722
723 if (header->magic != SESPARSE_CONST_HEADER_MAGIC) {
724 error_setg(errp, "Bad const header magic: 0x%016" PRIx64,
725 header->magic);
726 return -EINVAL;
727 }
728
729 if (header->version != 0x0000000200000001) {
730 error_setg(errp, "Unsupported version: 0x%016" PRIx64,
731 header->version);
732 return -ENOTSUP;
733 }
734
735 if (header->grain_size != 8) {
736 error_setg(errp, "Unsupported grain size: %" PRIu64,
737 header->grain_size);
738 return -ENOTSUP;
739 }
740
741 if (header->grain_table_size != 64) {
742 error_setg(errp, "Unsupported grain table size: %" PRIu64,
743 header->grain_table_size);
744 return -ENOTSUP;
745 }
746
747 if (header->flags != 0) {
748 error_setg(errp, "Unsupported flags: 0x%016" PRIx64,
749 header->flags);
750 return -ENOTSUP;
751 }
752
753 if (header->reserved1 != 0 || header->reserved2 != 0 ||
754 header->reserved3 != 0 || header->reserved4 != 0) {
755 error_setg(errp, "Unsupported reserved bits:"
756 " 0x%016" PRIx64 " 0x%016" PRIx64
757 " 0x%016" PRIx64 " 0x%016" PRIx64,
758 header->reserved1, header->reserved2,
759 header->reserved3, header->reserved4);
760 return -ENOTSUP;
761 }
762
763 /* check that padding is 0 */
764 if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
765 error_setg(errp, "Unsupported non-zero const header padding");
766 return -ENOTSUP;
767 }
768
769 return 0;
770 }
771
772 static int check_se_sparse_volatile_header(VMDKSESparseVolatileHeader *header,
773 Error **errp)
774 {
775 header->magic = le64_to_cpu(header->magic);
776 header->free_gt_number = le64_to_cpu(header->free_gt_number);
777 header->next_txn_seq_number = le64_to_cpu(header->next_txn_seq_number);
778 header->replay_journal = le64_to_cpu(header->replay_journal);
779
780 if (header->magic != SESPARSE_VOLATILE_HEADER_MAGIC) {
781 error_setg(errp, "Bad volatile header magic: 0x%016" PRIx64,
782 header->magic);
783 return -EINVAL;
784 }
785
786 if (header->replay_journal) {
787 error_setg(errp, "Image is dirty, Replaying journal not supported");
788 return -ENOTSUP;
789 }
790
791 /* check that padding is 0 */
792 if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
793 error_setg(errp, "Unsupported non-zero volatile header padding");
794 return -ENOTSUP;
795 }
796
797 return 0;
798 }
799
800 static int vmdk_open_se_sparse(BlockDriverState *bs,
801 BdrvChild *file,
802 int flags, Error **errp)
803 {
804 int ret;
805 VMDKSESparseConstHeader const_header;
806 VMDKSESparseVolatileHeader volatile_header;
807 VmdkExtent *extent = NULL;
808
809 ret = bdrv_apply_auto_read_only(bs,
810 "No write support for seSparse images available", errp);
811 if (ret < 0) {
812 return ret;
813 }
814
815 assert(sizeof(const_header) == SECTOR_SIZE);
816
817 ret = bdrv_pread(file, 0, sizeof(const_header), &const_header, 0);
818 if (ret < 0) {
819 bdrv_refresh_filename(file->bs);
820 error_setg_errno(errp, -ret,
821 "Could not read const header from file '%s'",
822 file->bs->filename);
823 return ret;
824 }
825
826 /* check const header */
827 ret = check_se_sparse_const_header(&const_header, errp);
828 if (ret < 0) {
829 return ret;
830 }
831
832 assert(sizeof(volatile_header) == SECTOR_SIZE);
833
834 ret = bdrv_pread(file, const_header.volatile_header_offset * SECTOR_SIZE,
835 sizeof(volatile_header), &volatile_header, 0);
836 if (ret < 0) {
837 bdrv_refresh_filename(file->bs);
838 error_setg_errno(errp, -ret,
839 "Could not read volatile header from file '%s'",
840 file->bs->filename);
841 return ret;
842 }
843
844 /* check volatile header */
845 ret = check_se_sparse_volatile_header(&volatile_header, errp);
846 if (ret < 0) {
847 return ret;
848 }
849
850 ret = vmdk_add_extent(bs, file, false,
851 const_header.capacity,
852 const_header.grain_dir_offset * SECTOR_SIZE,
853 0,
854 const_header.grain_dir_size *
855 SECTOR_SIZE / sizeof(uint64_t),
856 const_header.grain_table_size *
857 SECTOR_SIZE / sizeof(uint64_t),
858 const_header.grain_size,
859 &extent,
860 errp);
861 if (ret < 0) {
862 return ret;
863 }
864
865 extent->sesparse = true;
866 extent->sesparse_l2_tables_offset = const_header.grain_tables_offset;
867 extent->sesparse_clusters_offset = const_header.grains_offset;
868 extent->entry_size = sizeof(uint64_t);
869
870 ret = vmdk_init_tables(bs, extent, errp);
871 if (ret) {
872 /* free extent allocated by vmdk_add_extent */
873 vmdk_free_last_extent(bs);
874 }
875
876 return ret;
877 }
878
879 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
880 QDict *options, Error **errp);
881
882 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
883 {
884 int64_t size;
885 char *buf;
886 int ret;
887
888 size = bdrv_getlength(file->bs);
889 if (size < 0) {
890 error_setg_errno(errp, -size, "Could not access file");
891 return NULL;
892 }
893
894 if (size < 4) {
895 /* Both descriptor file and sparse image must be much larger than 4
896 * bytes, also callers of vmdk_read_desc want to compare the first 4
897 * bytes with VMDK4_MAGIC, let's error out if less is read. */
898 error_setg(errp, "File is too small, not a valid image");
899 return NULL;
900 }
901
902 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */
903 buf = g_malloc(size + 1);
904
905 ret = bdrv_pread(file, desc_offset, size, buf, 0);
906 if (ret < 0) {
907 error_setg_errno(errp, -ret, "Could not read from file");
908 g_free(buf);
909 return NULL;
910 }
911 buf[size] = 0;
912
913 return buf;
914 }
915
916 static int vmdk_open_vmdk4(BlockDriverState *bs,
917 BdrvChild *file,
918 int flags, QDict *options, Error **errp)
919 {
920 int ret;
921 uint32_t magic;
922 uint32_t l1_size, l1_entry_sectors;
923 VMDK4Header header;
924 VmdkExtent *extent = NULL;
925 BDRVVmdkState *s = bs->opaque;
926 int64_t l1_backup_offset = 0;
927 bool compressed;
928
929 ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0);
930 if (ret < 0) {
931 bdrv_refresh_filename(file->bs);
932 error_setg_errno(errp, -ret,
933 "Could not read header from file '%s'",
934 file->bs->filename);
935 return -EINVAL;
936 }
937 if (header.capacity == 0) {
938 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
939 if (desc_offset) {
940 char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
941 if (!buf) {
942 return -EINVAL;
943 }
944 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
945 g_free(buf);
946 return ret;
947 }
948 }
949
950 if (!s->create_type) {
951 s->create_type = g_strdup("monolithicSparse");
952 }
953
954 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
955 /*
956 * The footer takes precedence over the header, so read it in. The
957 * footer starts at offset -1024 from the end: One sector for the
958 * footer, and another one for the end-of-stream marker.
959 */
960 struct {
961 struct {
962 uint64_t val;
963 uint32_t size;
964 uint32_t type;
965 uint8_t pad[512 - 16];
966 } QEMU_PACKED footer_marker;
967
968 uint32_t magic;
969 VMDK4Header header;
970 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
971
972 struct {
973 uint64_t val;
974 uint32_t size;
975 uint32_t type;
976 uint8_t pad[512 - 16];
977 } QEMU_PACKED eos_marker;
978 } QEMU_PACKED footer;
979
980 ret = bdrv_pread(file, bs->file->bs->total_sectors * 512 - 1536,
981 sizeof(footer), &footer, 0);
982 if (ret < 0) {
983 error_setg_errno(errp, -ret, "Failed to read footer");
984 return ret;
985 }
986
987 /* Some sanity checks for the footer */
988 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
989 le32_to_cpu(footer.footer_marker.size) != 0 ||
990 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
991 le64_to_cpu(footer.eos_marker.val) != 0 ||
992 le32_to_cpu(footer.eos_marker.size) != 0 ||
993 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
994 {
995 error_setg(errp, "Invalid footer");
996 return -EINVAL;
997 }
998
999 header = footer.header;
1000 }
1001
1002 compressed =
1003 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1004 if (le32_to_cpu(header.version) > 3) {
1005 error_setg(errp, "Unsupported VMDK version %" PRIu32,
1006 le32_to_cpu(header.version));
1007 return -ENOTSUP;
1008 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
1009 !compressed) {
1010 /* VMware KB 2064959 explains that version 3 added support for
1011 * persistent changed block tracking (CBT), and backup software can
1012 * read it as version=1 if it doesn't care about the changed area
1013 * information. So we are safe to enable read only. */
1014 error_setg(errp, "VMDK version 3 must be read only");
1015 return -EINVAL;
1016 }
1017
1018 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
1019 error_setg(errp, "L2 table size too big");
1020 return -EINVAL;
1021 }
1022
1023 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
1024 * le64_to_cpu(header.granularity);
1025 if (l1_entry_sectors == 0) {
1026 error_setg(errp, "L1 entry size is invalid");
1027 return -EINVAL;
1028 }
1029 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
1030 / l1_entry_sectors;
1031 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
1032 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
1033 }
1034 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
1035 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
1036 (int64_t)(le64_to_cpu(header.grain_offset)
1037 * BDRV_SECTOR_SIZE));
1038 return -EINVAL;
1039 }
1040
1041 ret = vmdk_add_extent(bs, file, false,
1042 le64_to_cpu(header.capacity),
1043 le64_to_cpu(header.gd_offset) << 9,
1044 l1_backup_offset,
1045 l1_size,
1046 le32_to_cpu(header.num_gtes_per_gt),
1047 le64_to_cpu(header.granularity),
1048 &extent,
1049 errp);
1050 if (ret < 0) {
1051 return ret;
1052 }
1053 extent->compressed =
1054 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1055 if (extent->compressed) {
1056 g_free(s->create_type);
1057 s->create_type = g_strdup("streamOptimized");
1058 }
1059 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
1060 extent->version = le32_to_cpu(header.version);
1061 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
1062 ret = vmdk_init_tables(bs, extent, errp);
1063 if (ret) {
1064 /* free extent allocated by vmdk_add_extent */
1065 vmdk_free_last_extent(bs);
1066 }
1067 return ret;
1068 }
1069
1070 /* find an option value out of descriptor file */
1071 static int vmdk_parse_description(const char *desc, const char *opt_name,
1072 char *buf, int buf_size)
1073 {
1074 char *opt_pos, *opt_end;
1075 const char *end = desc + strlen(desc);
1076
1077 opt_pos = strstr(desc, opt_name);
1078 if (!opt_pos) {
1079 return VMDK_ERROR;
1080 }
1081 /* Skip "=\"" following opt_name */
1082 opt_pos += strlen(opt_name) + 2;
1083 if (opt_pos >= end) {
1084 return VMDK_ERROR;
1085 }
1086 opt_end = opt_pos;
1087 while (opt_end < end && *opt_end != '"') {
1088 opt_end++;
1089 }
1090 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
1091 return VMDK_ERROR;
1092 }
1093 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
1094 return VMDK_OK;
1095 }
1096
1097 /* Open an extent file and append to bs array */
1098 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
1099 char *buf, QDict *options, Error **errp)
1100 {
1101 uint32_t magic;
1102
1103 magic = ldl_be_p(buf);
1104 switch (magic) {
1105 case VMDK3_MAGIC:
1106 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
1107 case VMDK4_MAGIC:
1108 return vmdk_open_vmdk4(bs, file, flags, options, errp);
1109 default:
1110 error_setg(errp, "Image not in VMDK format");
1111 return -EINVAL;
1112 }
1113 }
1114
1115 static const char *next_line(const char *s)
1116 {
1117 while (*s) {
1118 if (*s == '\n') {
1119 return s + 1;
1120 }
1121 s++;
1122 }
1123 return s;
1124 }
1125
1126 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
1127 QDict *options, Error **errp)
1128 {
1129 int ret;
1130 int matches;
1131 char access[11];
1132 char type[11];
1133 char fname[512];
1134 const char *p, *np;
1135 int64_t sectors = 0;
1136 int64_t flat_offset;
1137 char *desc_file_dir = NULL;
1138 char *extent_path;
1139 BdrvChild *extent_file;
1140 BdrvChildRole extent_role;
1141 BDRVVmdkState *s = bs->opaque;
1142 VmdkExtent *extent = NULL;
1143 char extent_opt_prefix[32];
1144 Error *local_err = NULL;
1145
1146 for (p = desc; *p; p = next_line(p)) {
1147 /* parse extent line in one of below formats:
1148 *
1149 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
1150 * RW [size in sectors] SPARSE "file-name.vmdk"
1151 * RW [size in sectors] VMFS "file-name.vmdk"
1152 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
1153 * RW [size in sectors] SESPARSE "file-name.vmdk"
1154 */
1155 flat_offset = -1;
1156 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
1157 access, &sectors, type, fname, &flat_offset);
1158 if (matches < 4 || strcmp(access, "RW")) {
1159 continue;
1160 } else if (!strcmp(type, "FLAT")) {
1161 if (matches != 5 || flat_offset < 0) {
1162 goto invalid;
1163 }
1164 } else if (!strcmp(type, "VMFS")) {
1165 if (matches == 4) {
1166 flat_offset = 0;
1167 } else {
1168 goto invalid;
1169 }
1170 } else if (matches != 4) {
1171 goto invalid;
1172 }
1173
1174 if (sectors <= 0 ||
1175 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
1176 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") &&
1177 strcmp(type, "SESPARSE")) ||
1178 (strcmp(access, "RW"))) {
1179 continue;
1180 }
1181
1182 if (path_is_absolute(fname)) {
1183 extent_path = g_strdup(fname);
1184 } else {
1185 if (!desc_file_dir) {
1186 desc_file_dir = bdrv_dirname(bs->file->bs, errp);
1187 if (!desc_file_dir) {
1188 bdrv_refresh_filename(bs->file->bs);
1189 error_prepend(errp, "Cannot use relative paths with VMDK "
1190 "descriptor file '%s': ",
1191 bs->file->bs->filename);
1192 ret = -EINVAL;
1193 goto out;
1194 }
1195 }
1196
1197 extent_path = g_strconcat(desc_file_dir, fname, NULL);
1198 }
1199
1200 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
1201 assert(ret < 32);
1202
1203 extent_role = BDRV_CHILD_DATA;
1204 if (strcmp(type, "FLAT") != 0 && strcmp(type, "VMFS") != 0) {
1205 /* non-flat extents have metadata */
1206 extent_role |= BDRV_CHILD_METADATA;
1207 }
1208
1209 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
1210 bs, &child_of_bds, extent_role, false,
1211 &local_err);
1212 g_free(extent_path);
1213 if (!extent_file) {
1214 error_propagate(errp, local_err);
1215 ret = -EINVAL;
1216 goto out;
1217 }
1218
1219 /* save to extents array */
1220 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
1221 /* FLAT extent */
1222
1223 ret = vmdk_add_extent(bs, extent_file, true, sectors,
1224 0, 0, 0, 0, 0, &extent, errp);
1225 if (ret < 0) {
1226 bdrv_graph_wrlock(NULL);
1227 bdrv_unref_child(bs, extent_file);
1228 bdrv_graph_wrunlock();
1229 goto out;
1230 }
1231 extent->flat_start_offset = flat_offset << 9;
1232 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
1233 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
1234 char *buf = vmdk_read_desc(extent_file, 0, errp);
1235 if (!buf) {
1236 ret = -EINVAL;
1237 } else {
1238 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
1239 options, errp);
1240 }
1241 g_free(buf);
1242 if (ret) {
1243 bdrv_graph_wrlock(NULL);
1244 bdrv_unref_child(bs, extent_file);
1245 bdrv_graph_wrunlock();
1246 goto out;
1247 }
1248 extent = &s->extents[s->num_extents - 1];
1249 } else if (!strcmp(type, "SESPARSE")) {
1250 ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
1251 if (ret) {
1252 bdrv_graph_wrlock(NULL);
1253 bdrv_unref_child(bs, extent_file);
1254 bdrv_graph_wrunlock();
1255 goto out;
1256 }
1257 extent = &s->extents[s->num_extents - 1];
1258 } else {
1259 error_setg(errp, "Unsupported extent type '%s'", type);
1260 bdrv_graph_wrlock(NULL);
1261 bdrv_unref_child(bs, extent_file);
1262 bdrv_graph_wrunlock();
1263 ret = -ENOTSUP;
1264 goto out;
1265 }
1266 extent->type = g_strdup(type);
1267 }
1268
1269 ret = 0;
1270 goto out;
1271
1272 invalid:
1273 np = next_line(p);
1274 assert(np != p);
1275 if (np[-1] == '\n') {
1276 np--;
1277 }
1278 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
1279 ret = -EINVAL;
1280
1281 out:
1282 g_free(desc_file_dir);
1283 return ret;
1284 }
1285
1286 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
1287 QDict *options, Error **errp)
1288 {
1289 int ret;
1290 char ct[128];
1291 BDRVVmdkState *s = bs->opaque;
1292
1293 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
1294 error_setg(errp, "invalid VMDK image descriptor");
1295 ret = -EINVAL;
1296 goto exit;
1297 }
1298 if (strcmp(ct, "monolithicFlat") &&
1299 strcmp(ct, "vmfs") &&
1300 strcmp(ct, "vmfsSparse") &&
1301 strcmp(ct, "seSparse") &&
1302 strcmp(ct, "twoGbMaxExtentSparse") &&
1303 strcmp(ct, "twoGbMaxExtentFlat")) {
1304 error_setg(errp, "Unsupported image type '%s'", ct);
1305 ret = -ENOTSUP;
1306 goto exit;
1307 }
1308 s->create_type = g_strdup(ct);
1309 s->desc_offset = 0;
1310 ret = vmdk_parse_extents(buf, bs, options, errp);
1311 exit:
1312 return ret;
1313 }
1314
1315 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
1316 Error **errp)
1317 {
1318 char *buf;
1319 int ret;
1320 BDRVVmdkState *s = bs->opaque;
1321 uint32_t magic;
1322
1323 GRAPH_RDLOCK_GUARD_MAINLOOP();
1324
1325 ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
1326 if (ret < 0) {
1327 return ret;
1328 }
1329
1330 buf = vmdk_read_desc(bs->file, 0, errp);
1331 if (!buf) {
1332 return -EINVAL;
1333 }
1334
1335 magic = ldl_be_p(buf);
1336 switch (magic) {
1337 case VMDK3_MAGIC:
1338 case VMDK4_MAGIC:
1339 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
1340 errp);
1341 s->desc_offset = 0x200;
1342 break;
1343 default:
1344 /* No data in the descriptor file */
1345 bs->file->role &= ~BDRV_CHILD_DATA;
1346
1347 /* Must succeed because we have given up permissions if anything */
1348 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1349
1350 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1351 break;
1352 }
1353 if (ret) {
1354 goto fail;
1355 }
1356
1357 /* try to open parent images, if exist */
1358 ret = vmdk_parent_open(bs);
1359 if (ret) {
1360 goto fail;
1361 }
1362 ret = vmdk_read_cid(bs, 0, &s->cid);
1363 if (ret) {
1364 goto fail;
1365 }
1366 ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1367 if (ret) {
1368 goto fail;
1369 }
1370 qemu_co_mutex_init(&s->lock);
1371
1372 /* Disable migration when VMDK images are used */
1373 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1374 "does not support live migration",
1375 bdrv_get_device_or_node_name(bs));
1376 ret = migrate_add_blocker(s->migration_blocker, errp);
1377 if (ret < 0) {
1378 error_free(s->migration_blocker);
1379 goto fail;
1380 }
1381
1382 g_free(buf);
1383 return 0;
1384
1385 fail:
1386 g_free(buf);
1387 g_free(s->create_type);
1388 s->create_type = NULL;
1389 vmdk_free_extents(bs);
1390 return ret;
1391 }
1392
1393
1394 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1395 {
1396 BDRVVmdkState *s = bs->opaque;
1397 int i;
1398
1399 for (i = 0; i < s->num_extents; i++) {
1400 if (!s->extents[i].flat) {
1401 bs->bl.pwrite_zeroes_alignment =
1402 MAX(bs->bl.pwrite_zeroes_alignment,
1403 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1404 }
1405 }
1406 }
1407
1408 /**
1409 * get_whole_cluster
1410 *
1411 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1412 * to the cluster at @cluster_sector_num. If @zeroed is true, we're overwriting
1413 * a zeroed cluster in the current layer and must not copy data from the
1414 * backing file.
1415 *
1416 * If @skip_start_sector < @skip_end_sector, the relative range
1417 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1418 * it for call to write user data in the request.
1419 */
1420 static int coroutine_fn GRAPH_RDLOCK
1421 get_whole_cluster(BlockDriverState *bs, VmdkExtent *extent,
1422 uint64_t cluster_offset, uint64_t offset,
1423 uint64_t skip_start_bytes, uint64_t skip_end_bytes,
1424 bool zeroed)
1425 {
1426 int ret = VMDK_OK;
1427 int64_t cluster_bytes;
1428 uint8_t *whole_grain;
1429 bool copy_from_backing;
1430
1431 /* For COW, align request sector_num to cluster start */
1432 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1433 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1434 whole_grain = qemu_blockalign(bs, cluster_bytes);
1435 copy_from_backing = bs->backing && !zeroed;
1436
1437 if (!copy_from_backing) {
1438 memset(whole_grain, 0, skip_start_bytes);
1439 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1440 }
1441
1442 assert(skip_end_bytes <= cluster_bytes);
1443 /* we will be here if it's first write on non-exist grain(cluster).
1444 * try to read from parent image, if exist */
1445 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1446 ret = VMDK_ERROR;
1447 goto exit;
1448 }
1449
1450 /* Read backing data before skip range */
1451 if (skip_start_bytes > 0) {
1452 if (copy_from_backing) {
1453 /* qcow2 emits this on bs->file instead of bs->backing */
1454 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_READ);
1455 ret = bdrv_co_pread(bs->backing, offset, skip_start_bytes,
1456 whole_grain, 0);
1457 if (ret < 0) {
1458 ret = VMDK_ERROR;
1459 goto exit;
1460 }
1461 }
1462 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_WRITE);
1463 ret = bdrv_co_pwrite(extent->file, cluster_offset, skip_start_bytes,
1464 whole_grain, 0);
1465 if (ret < 0) {
1466 ret = VMDK_ERROR;
1467 goto exit;
1468 }
1469 }
1470 /* Read backing data after skip range */
1471 if (skip_end_bytes < cluster_bytes) {
1472 if (copy_from_backing) {
1473 /* qcow2 emits this on bs->file instead of bs->backing */
1474 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_READ);
1475 ret = bdrv_co_pread(bs->backing, offset + skip_end_bytes,
1476 cluster_bytes - skip_end_bytes,
1477 whole_grain + skip_end_bytes, 0);
1478 if (ret < 0) {
1479 ret = VMDK_ERROR;
1480 goto exit;
1481 }
1482 }
1483 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_WRITE);
1484 ret = bdrv_co_pwrite(extent->file, cluster_offset + skip_end_bytes,
1485 cluster_bytes - skip_end_bytes,
1486 whole_grain + skip_end_bytes, 0);
1487 if (ret < 0) {
1488 ret = VMDK_ERROR;
1489 goto exit;
1490 }
1491 }
1492
1493 ret = VMDK_OK;
1494 exit:
1495 qemu_vfree(whole_grain);
1496 return ret;
1497 }
1498
1499 static int coroutine_fn GRAPH_RDLOCK
1500 vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data, uint32_t offset)
1501 {
1502 offset = cpu_to_le32(offset);
1503 /* update L2 table */
1504 BLKDBG_CO_EVENT(extent->file, BLKDBG_L2_UPDATE);
1505 if (bdrv_co_pwrite(extent->file,
1506 ((int64_t)m_data->l2_offset * 512)
1507 + (m_data->l2_index * sizeof(offset)),
1508 sizeof(offset), &offset, 0) < 0) {
1509 return VMDK_ERROR;
1510 }
1511 /* update backup L2 table */
1512 if (extent->l1_backup_table_offset != 0) {
1513 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1514 if (bdrv_co_pwrite(extent->file,
1515 ((int64_t)m_data->l2_offset * 512)
1516 + (m_data->l2_index * sizeof(offset)),
1517 sizeof(offset), &offset, 0) < 0) {
1518 return VMDK_ERROR;
1519 }
1520 }
1521 if (bdrv_co_flush(extent->file->bs) < 0) {
1522 return VMDK_ERROR;
1523 }
1524 if (m_data->l2_cache_entry) {
1525 *m_data->l2_cache_entry = offset;
1526 }
1527
1528 return VMDK_OK;
1529 }
1530
1531 /**
1532 * get_cluster_offset
1533 *
1534 * Look up cluster offset in extent file by sector number, and store in
1535 * @cluster_offset.
1536 *
1537 * For flat extents, the start offset as parsed from the description file is
1538 * returned.
1539 *
1540 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1541 * offset for a new cluster and update L2 cache. If there is a backing file,
1542 * COW is done before returning; otherwise, zeroes are written to the allocated
1543 * cluster. Both COW and zero writing skips the sector range
1544 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1545 * has new data to write there.
1546 *
1547 * Returns: VMDK_OK if cluster exists and mapped in the image.
1548 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1549 * VMDK_ERROR if failed.
1550 */
1551 static int coroutine_fn GRAPH_RDLOCK
1552 get_cluster_offset(BlockDriverState *bs, VmdkExtent *extent,
1553 VmdkMetaData *m_data, uint64_t offset, bool allocate,
1554 uint64_t *cluster_offset, uint64_t skip_start_bytes,
1555 uint64_t skip_end_bytes)
1556 {
1557 unsigned int l1_index, l2_offset, l2_index;
1558 int min_index, i, j;
1559 uint32_t min_count;
1560 void *l2_table;
1561 bool zeroed = false;
1562 int64_t ret;
1563 int64_t cluster_sector;
1564 unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
1565
1566 if (m_data) {
1567 m_data->new_allocation = false;
1568 }
1569 if (extent->flat) {
1570 *cluster_offset = extent->flat_start_offset;
1571 return VMDK_OK;
1572 }
1573
1574 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1575 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1576 if (l1_index >= extent->l1_size) {
1577 return VMDK_ERROR;
1578 }
1579 if (extent->sesparse) {
1580 uint64_t l2_offset_u64;
1581
1582 assert(extent->entry_size == sizeof(uint64_t));
1583
1584 l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
1585 if (l2_offset_u64 == 0) {
1586 l2_offset = 0;
1587 } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
1588 /*
1589 * Top most nibble is 0x1 if grain table is allocated.
1590 * strict check - top most 4 bytes must be 0x10000000 since max
1591 * supported size is 64TB for disk - so no more than 64TB / 16MB
1592 * grain directories which is smaller than uint32,
1593 * where 16MB is the only supported default grain table coverage.
1594 */
1595 return VMDK_ERROR;
1596 } else {
1597 l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
1598 l2_offset_u64 = extent->sesparse_l2_tables_offset +
1599 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
1600 if (l2_offset_u64 > 0x00000000ffffffff) {
1601 return VMDK_ERROR;
1602 }
1603 l2_offset = (unsigned int)(l2_offset_u64);
1604 }
1605 } else {
1606 assert(extent->entry_size == sizeof(uint32_t));
1607 l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
1608 }
1609 if (!l2_offset) {
1610 return VMDK_UNALLOC;
1611 }
1612 for (i = 0; i < L2_CACHE_SIZE; i++) {
1613 if (l2_offset == extent->l2_cache_offsets[i]) {
1614 /* increment the hit count */
1615 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1616 for (j = 0; j < L2_CACHE_SIZE; j++) {
1617 extent->l2_cache_counts[j] >>= 1;
1618 }
1619 }
1620 l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
1621 goto found;
1622 }
1623 }
1624 /* not found: load a new entry in the least used one */
1625 min_index = 0;
1626 min_count = 0xffffffff;
1627 for (i = 0; i < L2_CACHE_SIZE; i++) {
1628 if (extent->l2_cache_counts[i] < min_count) {
1629 min_count = extent->l2_cache_counts[i];
1630 min_index = i;
1631 }
1632 }
1633 l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
1634 BLKDBG_CO_EVENT(extent->file, BLKDBG_L2_LOAD);
1635 if (bdrv_co_pread(extent->file,
1636 (int64_t)l2_offset * 512,
1637 l2_size_bytes,
1638 l2_table, 0
1639 ) < 0) {
1640 return VMDK_ERROR;
1641 }
1642
1643 extent->l2_cache_offsets[min_index] = l2_offset;
1644 extent->l2_cache_counts[min_index] = 1;
1645 found:
1646 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1647 if (m_data) {
1648 m_data->l1_index = l1_index;
1649 m_data->l2_index = l2_index;
1650 m_data->l2_offset = l2_offset;
1651 m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
1652 }
1653
1654 if (extent->sesparse) {
1655 cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
1656 switch (cluster_sector & 0xf000000000000000) {
1657 case 0x0000000000000000:
1658 /* unallocated grain */
1659 if (cluster_sector != 0) {
1660 return VMDK_ERROR;
1661 }
1662 break;
1663 case 0x1000000000000000:
1664 /* scsi-unmapped grain - fallthrough */
1665 case 0x2000000000000000:
1666 /* zero grain */
1667 zeroed = true;
1668 break;
1669 case 0x3000000000000000:
1670 /* allocated grain */
1671 cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
1672 ((cluster_sector & 0x0000ffffffffffff) << 12));
1673 cluster_sector = extent->sesparse_clusters_offset +
1674 cluster_sector * extent->cluster_sectors;
1675 break;
1676 default:
1677 return VMDK_ERROR;
1678 }
1679 } else {
1680 cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
1681
1682 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1683 zeroed = true;
1684 }
1685 }
1686
1687 if (!cluster_sector || zeroed) {
1688 if (!allocate) {
1689 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1690 }
1691 assert(!extent->sesparse);
1692
1693 if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1694 return VMDK_ERROR;
1695 }
1696
1697 cluster_sector = extent->next_cluster_sector;
1698 extent->next_cluster_sector += extent->cluster_sectors;
1699
1700 /* First of all we write grain itself, to avoid race condition
1701 * that may to corrupt the image.
1702 * This problem may occur because of insufficient space on host disk
1703 * or inappropriate VM shutdown.
1704 */
1705 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1706 offset, skip_start_bytes, skip_end_bytes,
1707 zeroed);
1708 if (ret) {
1709 return ret;
1710 }
1711 if (m_data) {
1712 m_data->new_allocation = true;
1713 }
1714 }
1715 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1716 return VMDK_OK;
1717 }
1718
1719 static VmdkExtent *find_extent(BDRVVmdkState *s,
1720 int64_t sector_num, VmdkExtent *start_hint)
1721 {
1722 VmdkExtent *extent = start_hint;
1723
1724 if (!extent) {
1725 extent = &s->extents[0];
1726 }
1727 while (extent < &s->extents[s->num_extents]) {
1728 if (sector_num < extent->end_sector) {
1729 return extent;
1730 }
1731 extent++;
1732 }
1733 return NULL;
1734 }
1735
1736 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1737 int64_t offset)
1738 {
1739 uint64_t extent_begin_offset, extent_relative_offset;
1740 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1741
1742 extent_begin_offset =
1743 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1744 extent_relative_offset = offset - extent_begin_offset;
1745 return extent_relative_offset % cluster_size;
1746 }
1747
1748 static int coroutine_fn GRAPH_RDLOCK
1749 vmdk_co_block_status(BlockDriverState *bs, bool want_zero,
1750 int64_t offset, int64_t bytes, int64_t *pnum,
1751 int64_t *map, BlockDriverState **file)
1752 {
1753 BDRVVmdkState *s = bs->opaque;
1754 int64_t index_in_cluster, n, ret;
1755 uint64_t cluster_offset;
1756 VmdkExtent *extent;
1757
1758 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1759 if (!extent) {
1760 return -EIO;
1761 }
1762 qemu_co_mutex_lock(&s->lock);
1763 ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1764 0, 0);
1765 qemu_co_mutex_unlock(&s->lock);
1766
1767 index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1768 switch (ret) {
1769 case VMDK_ERROR:
1770 ret = -EIO;
1771 break;
1772 case VMDK_UNALLOC:
1773 ret = 0;
1774 break;
1775 case VMDK_ZEROED:
1776 ret = BDRV_BLOCK_ZERO;
1777 break;
1778 case VMDK_OK:
1779 ret = BDRV_BLOCK_DATA;
1780 if (!extent->compressed) {
1781 ret |= BDRV_BLOCK_OFFSET_VALID;
1782 *map = cluster_offset + index_in_cluster;
1783 if (extent->flat) {
1784 ret |= BDRV_BLOCK_RECURSE;
1785 }
1786 }
1787 *file = extent->file->bs;
1788 break;
1789 }
1790
1791 n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1792 *pnum = MIN(n, bytes);
1793 return ret;
1794 }
1795
1796 static int coroutine_fn GRAPH_RDLOCK
1797 vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1798 int64_t offset_in_cluster, QEMUIOVector *qiov,
1799 uint64_t qiov_offset, uint64_t n_bytes,
1800 uint64_t offset)
1801 {
1802 int ret;
1803 VmdkGrainMarker *data = NULL;
1804 uLongf buf_len;
1805 QEMUIOVector local_qiov;
1806 int64_t write_offset;
1807 int64_t write_end_sector;
1808
1809 if (extent->compressed) {
1810 void *compressed_data;
1811
1812 /* Only whole clusters */
1813 if (offset_in_cluster ||
1814 n_bytes > (extent->cluster_sectors * SECTOR_SIZE) ||
1815 (n_bytes < (extent->cluster_sectors * SECTOR_SIZE) &&
1816 offset + n_bytes != extent->end_sector * SECTOR_SIZE))
1817 {
1818 ret = -EINVAL;
1819 goto out;
1820 }
1821
1822 if (!extent->has_marker) {
1823 ret = -EINVAL;
1824 goto out;
1825 }
1826 buf_len = (extent->cluster_sectors << 9) * 2;
1827 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1828
1829 compressed_data = g_malloc(n_bytes);
1830 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1831 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1832 g_free(compressed_data);
1833
1834 if (ret != Z_OK || buf_len == 0) {
1835 ret = -EINVAL;
1836 goto out;
1837 }
1838
1839 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1840 data->size = cpu_to_le32(buf_len);
1841
1842 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1843 qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1844
1845 BLKDBG_CO_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1846 } else {
1847 qemu_iovec_init(&local_qiov, qiov->niov);
1848 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1849
1850 BLKDBG_CO_EVENT(extent->file, BLKDBG_WRITE_AIO);
1851 }
1852
1853 write_offset = cluster_offset + offset_in_cluster;
1854 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1855 &local_qiov, 0);
1856
1857 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1858
1859 if (extent->compressed) {
1860 extent->next_cluster_sector = write_end_sector;
1861 } else {
1862 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1863 write_end_sector);
1864 }
1865
1866 if (ret < 0) {
1867 goto out;
1868 }
1869 ret = 0;
1870 out:
1871 g_free(data);
1872 if (!extent->compressed) {
1873 qemu_iovec_destroy(&local_qiov);
1874 }
1875 return ret;
1876 }
1877
1878 static int coroutine_fn GRAPH_RDLOCK
1879 vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1880 int64_t offset_in_cluster, QEMUIOVector *qiov, int bytes)
1881 {
1882 int ret;
1883 int cluster_bytes, buf_bytes;
1884 uint8_t *cluster_buf, *compressed_data;
1885 uint8_t *uncomp_buf;
1886 uint32_t data_len;
1887 VmdkGrainMarker *marker;
1888 uLongf buf_len;
1889
1890
1891 if (!extent->compressed) {
1892 BLKDBG_CO_EVENT(extent->file, BLKDBG_READ_AIO);
1893 ret = bdrv_co_preadv(extent->file,
1894 cluster_offset + offset_in_cluster, bytes,
1895 qiov, 0);
1896 if (ret < 0) {
1897 return ret;
1898 }
1899 return 0;
1900 }
1901 cluster_bytes = extent->cluster_sectors * 512;
1902 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1903 buf_bytes = cluster_bytes * 2;
1904 cluster_buf = g_malloc(buf_bytes);
1905 uncomp_buf = g_malloc(cluster_bytes);
1906 BLKDBG_CO_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1907 ret = bdrv_co_pread(extent->file, cluster_offset, buf_bytes, cluster_buf,
1908 0);
1909 if (ret < 0) {
1910 goto out;
1911 }
1912 compressed_data = cluster_buf;
1913 buf_len = cluster_bytes;
1914 data_len = cluster_bytes;
1915 if (extent->has_marker) {
1916 marker = (VmdkGrainMarker *)cluster_buf;
1917 compressed_data = marker->data;
1918 data_len = le32_to_cpu(marker->size);
1919 }
1920 if (!data_len || data_len > buf_bytes) {
1921 ret = -EINVAL;
1922 goto out;
1923 }
1924 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1925 if (ret != Z_OK) {
1926 ret = -EINVAL;
1927 goto out;
1928
1929 }
1930 if (offset_in_cluster < 0 ||
1931 offset_in_cluster + bytes > buf_len) {
1932 ret = -EINVAL;
1933 goto out;
1934 }
1935 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1936 ret = 0;
1937
1938 out:
1939 g_free(uncomp_buf);
1940 g_free(cluster_buf);
1941 return ret;
1942 }
1943
1944 static int coroutine_fn GRAPH_RDLOCK
1945 vmdk_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
1946 QEMUIOVector *qiov, BdrvRequestFlags flags)
1947 {
1948 BDRVVmdkState *s = bs->opaque;
1949 int ret;
1950 uint64_t n_bytes, offset_in_cluster;
1951 VmdkExtent *extent = NULL;
1952 QEMUIOVector local_qiov;
1953 uint64_t cluster_offset;
1954 uint64_t bytes_done = 0;
1955
1956 qemu_iovec_init(&local_qiov, qiov->niov);
1957 qemu_co_mutex_lock(&s->lock);
1958
1959 while (bytes > 0) {
1960 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1961 if (!extent) {
1962 ret = -EIO;
1963 goto fail;
1964 }
1965 ret = get_cluster_offset(bs, extent, NULL,
1966 offset, false, &cluster_offset, 0, 0);
1967 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1968
1969 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1970 - offset_in_cluster);
1971
1972 if (ret != VMDK_OK) {
1973 /* if not allocated, try to read from parent image, if exist */
1974 if (bs->backing && ret != VMDK_ZEROED) {
1975 if (!vmdk_is_cid_valid(bs)) {
1976 ret = -EINVAL;
1977 goto fail;
1978 }
1979
1980 qemu_iovec_reset(&local_qiov);
1981 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1982
1983 /* qcow2 emits this on bs->file instead of bs->backing */
1984 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1985 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1986 &local_qiov, 0);
1987 if (ret < 0) {
1988 goto fail;
1989 }
1990 } else {
1991 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1992 }
1993 } else {
1994 qemu_iovec_reset(&local_qiov);
1995 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1996
1997 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1998 &local_qiov, n_bytes);
1999 if (ret) {
2000 goto fail;
2001 }
2002 }
2003 bytes -= n_bytes;
2004 offset += n_bytes;
2005 bytes_done += n_bytes;
2006 }
2007
2008 ret = 0;
2009 fail:
2010 qemu_co_mutex_unlock(&s->lock);
2011 qemu_iovec_destroy(&local_qiov);
2012
2013 return ret;
2014 }
2015
2016 /**
2017 * vmdk_write:
2018 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
2019 * if possible, otherwise return -ENOTSUP.
2020 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
2021 * with each cluster. By dry run we can find if the zero write
2022 * is possible without modifying image data.
2023 *
2024 * Returns: error code with 0 for success.
2025 */
2026 static int coroutine_fn GRAPH_RDLOCK
2027 vmdk_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2028 QEMUIOVector *qiov, bool zeroed, bool zero_dry_run)
2029 {
2030 BDRVVmdkState *s = bs->opaque;
2031 VmdkExtent *extent = NULL;
2032 int ret;
2033 int64_t offset_in_cluster, n_bytes;
2034 uint64_t cluster_offset;
2035 uint64_t bytes_done = 0;
2036 VmdkMetaData m_data;
2037
2038 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
2039 error_report("Wrong offset: offset=0x%" PRIx64
2040 " total_sectors=0x%" PRIx64,
2041 offset, bs->total_sectors);
2042 return -EIO;
2043 }
2044
2045 while (bytes > 0) {
2046 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
2047 if (!extent) {
2048 return -EIO;
2049 }
2050 if (extent->sesparse) {
2051 return -ENOTSUP;
2052 }
2053 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
2054 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
2055 - offset_in_cluster);
2056
2057 ret = get_cluster_offset(bs, extent, &m_data, offset,
2058 !(extent->compressed || zeroed),
2059 &cluster_offset, offset_in_cluster,
2060 offset_in_cluster + n_bytes);
2061 if (extent->compressed) {
2062 if (ret == VMDK_OK) {
2063 /* Refuse write to allocated cluster for streamOptimized */
2064 error_report("Could not write to allocated cluster"
2065 " for streamOptimized");
2066 return -EIO;
2067 } else if (!zeroed) {
2068 /* allocate */
2069 ret = get_cluster_offset(bs, extent, &m_data, offset,
2070 true, &cluster_offset, 0, 0);
2071 }
2072 }
2073 if (ret == VMDK_ERROR) {
2074 return -EINVAL;
2075 }
2076 if (zeroed) {
2077 /* Do zeroed write, buf is ignored */
2078 if (extent->has_zero_grain &&
2079 offset_in_cluster == 0 &&
2080 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
2081 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
2082 if (!zero_dry_run && ret != VMDK_ZEROED) {
2083 /* update L2 tables */
2084 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
2085 != VMDK_OK) {
2086 return -EIO;
2087 }
2088 }
2089 } else {
2090 return -ENOTSUP;
2091 }
2092 } else {
2093 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
2094 qiov, bytes_done, n_bytes, offset);
2095 if (ret) {
2096 return ret;
2097 }
2098 if (m_data.new_allocation) {
2099 /* update L2 tables */
2100 if (vmdk_L2update(extent, &m_data,
2101 cluster_offset >> BDRV_SECTOR_BITS)
2102 != VMDK_OK) {
2103 return -EIO;
2104 }
2105 }
2106 }
2107 bytes -= n_bytes;
2108 offset += n_bytes;
2109 bytes_done += n_bytes;
2110
2111 /* update CID on the first write every time the virtual disk is
2112 * opened */
2113 if (!s->cid_updated) {
2114 ret = vmdk_write_cid(bs, g_random_int());
2115 if (ret < 0) {
2116 return ret;
2117 }
2118 s->cid_updated = true;
2119 }
2120 }
2121 return 0;
2122 }
2123
2124 static int coroutine_fn GRAPH_RDLOCK
2125 vmdk_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
2126 QEMUIOVector *qiov, BdrvRequestFlags flags)
2127 {
2128 int ret;
2129 BDRVVmdkState *s = bs->opaque;
2130 qemu_co_mutex_lock(&s->lock);
2131 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
2132 qemu_co_mutex_unlock(&s->lock);
2133 return ret;
2134 }
2135
2136 static int coroutine_fn GRAPH_RDLOCK
2137 vmdk_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes,
2138 QEMUIOVector *qiov)
2139 {
2140 if (bytes == 0) {
2141 /* The caller will write bytes 0 to signal EOF.
2142 * When receive it, we align EOF to a sector boundary. */
2143 BDRVVmdkState *s = bs->opaque;
2144 int i, ret;
2145 int64_t length;
2146
2147 for (i = 0; i < s->num_extents; i++) {
2148 length = bdrv_co_getlength(s->extents[i].file->bs);
2149 if (length < 0) {
2150 return length;
2151 }
2152 length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
2153 ret = bdrv_co_truncate(s->extents[i].file, length, false,
2154 PREALLOC_MODE_OFF, 0, NULL);
2155 if (ret < 0) {
2156 return ret;
2157 }
2158 }
2159 return 0;
2160 }
2161 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
2162 }
2163
2164 static int coroutine_fn GRAPH_RDLOCK
2165 vmdk_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
2166 BdrvRequestFlags flags)
2167 {
2168 int ret;
2169 BDRVVmdkState *s = bs->opaque;
2170
2171 qemu_co_mutex_lock(&s->lock);
2172 /* write zeroes could fail if sectors not aligned to cluster, test it with
2173 * dry_run == true before really updating image */
2174 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
2175 if (!ret) {
2176 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
2177 }
2178 qemu_co_mutex_unlock(&s->lock);
2179 return ret;
2180 }
2181
2182 static int coroutine_fn GRAPH_UNLOCKED
2183 vmdk_init_extent(BlockBackend *blk, int64_t filesize, bool flat, bool compress,
2184 bool zeroed_grain, Error **errp)
2185 {
2186 int ret, i;
2187 VMDK4Header header;
2188 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
2189 uint32_t *gd_buf = NULL;
2190 int gd_buf_size;
2191
2192 if (flat) {
2193 ret = blk_co_truncate(blk, filesize, false, PREALLOC_MODE_OFF, 0, errp);
2194 goto exit;
2195 }
2196 magic = cpu_to_be32(VMDK4_MAGIC);
2197 memset(&header, 0, sizeof(header));
2198 if (compress) {
2199 header.version = 3;
2200 } else if (zeroed_grain) {
2201 header.version = 2;
2202 } else {
2203 header.version = 1;
2204 }
2205 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
2206 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
2207 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
2208 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
2209 header.capacity = filesize / BDRV_SECTOR_SIZE;
2210 header.granularity = 128;
2211 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
2212
2213 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
2214 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
2215 BDRV_SECTOR_SIZE);
2216 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
2217 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
2218
2219 header.desc_offset = 1;
2220 header.desc_size = 20;
2221 header.rgd_offset = header.desc_offset + header.desc_size;
2222 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
2223 header.grain_offset =
2224 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
2225 header.granularity);
2226 /* swap endianness for all header fields */
2227 header.version = cpu_to_le32(header.version);
2228 header.flags = cpu_to_le32(header.flags);
2229 header.capacity = cpu_to_le64(header.capacity);
2230 header.granularity = cpu_to_le64(header.granularity);
2231 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
2232 header.desc_offset = cpu_to_le64(header.desc_offset);
2233 header.desc_size = cpu_to_le64(header.desc_size);
2234 header.rgd_offset = cpu_to_le64(header.rgd_offset);
2235 header.gd_offset = cpu_to_le64(header.gd_offset);
2236 header.grain_offset = cpu_to_le64(header.grain_offset);
2237 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
2238
2239 header.check_bytes[0] = 0xa;
2240 header.check_bytes[1] = 0x20;
2241 header.check_bytes[2] = 0xd;
2242 header.check_bytes[3] = 0xa;
2243
2244 /* write all the data */
2245 ret = blk_co_pwrite(blk, 0, sizeof(magic), &magic, 0);
2246 if (ret < 0) {
2247 error_setg(errp, QERR_IO_ERROR);
2248 goto exit;
2249 }
2250 ret = blk_co_pwrite(blk, sizeof(magic), sizeof(header), &header, 0);
2251 if (ret < 0) {
2252 error_setg(errp, QERR_IO_ERROR);
2253 goto exit;
2254 }
2255
2256 ret = blk_co_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false,
2257 PREALLOC_MODE_OFF, 0, errp);
2258 if (ret < 0) {
2259 goto exit;
2260 }
2261
2262 /* write grain directory */
2263 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
2264 gd_buf = g_malloc0(gd_buf_size);
2265 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
2266 i < gt_count; i++, tmp += gt_size) {
2267 gd_buf[i] = cpu_to_le32(tmp);
2268 }
2269 ret = blk_co_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
2270 gd_buf_size, gd_buf, 0);
2271 if (ret < 0) {
2272 error_setg(errp, QERR_IO_ERROR);
2273 goto exit;
2274 }
2275
2276 /* write backup grain directory */
2277 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
2278 i < gt_count; i++, tmp += gt_size) {
2279 gd_buf[i] = cpu_to_le32(tmp);
2280 }
2281 ret = blk_co_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
2282 gd_buf_size, gd_buf, 0);
2283 if (ret < 0) {
2284 error_setg(errp, QERR_IO_ERROR);
2285 }
2286
2287 ret = 0;
2288 exit:
2289 g_free(gd_buf);
2290 return ret;
2291 }
2292
2293 static int coroutine_fn GRAPH_UNLOCKED
2294 vmdk_create_extent(const char *filename, int64_t filesize, bool flat,
2295 bool compress, bool zeroed_grain, BlockBackend **pbb,
2296 QemuOpts *opts, Error **errp)
2297 {
2298 int ret;
2299 BlockBackend *blk = NULL;
2300
2301 ret = bdrv_co_create_file(filename, opts, errp);
2302 if (ret < 0) {
2303 goto exit;
2304 }
2305
2306 blk = blk_co_new_open(filename, NULL, NULL,
2307 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2308 errp);
2309 if (blk == NULL) {
2310 ret = -EIO;
2311 goto exit;
2312 }
2313
2314 blk_set_allow_write_beyond_eof(blk, true);
2315
2316 ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
2317 exit:
2318 if (blk) {
2319 if (pbb) {
2320 *pbb = blk;
2321 } else {
2322 blk_co_unref(blk);
2323 blk = NULL;
2324 }
2325 }
2326 return ret;
2327 }
2328
2329 static int filename_decompose(const char *filename, char *path, char *prefix,
2330 char *postfix, size_t buf_len, Error **errp)
2331 {
2332 const char *p, *q;
2333
2334 if (filename == NULL || !strlen(filename)) {
2335 error_setg(errp, "No filename provided");
2336 return VMDK_ERROR;
2337 }
2338 p = strrchr(filename, '/');
2339 if (p == NULL) {
2340 p = strrchr(filename, '\\');
2341 }
2342 if (p == NULL) {
2343 p = strrchr(filename, ':');
2344 }
2345 if (p != NULL) {
2346 p++;
2347 if (p - filename >= buf_len) {
2348 return VMDK_ERROR;
2349 }
2350 pstrcpy(path, p - filename + 1, filename);
2351 } else {
2352 p = filename;
2353 path[0] = '\0';
2354 }
2355 q = strrchr(p, '.');
2356 if (q == NULL) {
2357 pstrcpy(prefix, buf_len, p);
2358 postfix[0] = '\0';
2359 } else {
2360 if (q - p >= buf_len) {
2361 return VMDK_ERROR;
2362 }
2363 pstrcpy(prefix, q - p + 1, p);
2364 pstrcpy(postfix, buf_len, q);
2365 }
2366 return VMDK_OK;
2367 }
2368
2369 /*
2370 * idx == 0: get or create the descriptor file (also the image file if in a
2371 * non-split format.
2372 * idx >= 1: get the n-th extent if in a split subformat
2373 */
2374 typedef BlockBackend * coroutine_fn GRAPH_UNLOCKED_PTR
2375 (*vmdk_create_extent_fn)(int64_t size, int idx, bool flat, bool split,
2376 bool compress, bool zeroed_grain, void *opaque,
2377 Error **errp);
2378
2379 static void vmdk_desc_add_extent(GString *desc,
2380 const char *extent_line_fmt,
2381 int64_t size, const char *filename)
2382 {
2383 char *basename = g_path_get_basename(filename);
2384
2385 g_string_append_printf(desc, extent_line_fmt,
2386 DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
2387 g_free(basename);
2388 }
2389
2390 static int coroutine_fn GRAPH_UNLOCKED
2391 vmdk_co_do_create(int64_t size,
2392 BlockdevVmdkSubformat subformat,
2393 BlockdevVmdkAdapterType adapter_type,
2394 const char *backing_file,
2395 const char *hw_version,
2396 const char *toolsversion,
2397 bool compat6,
2398 bool zeroed_grain,
2399 vmdk_create_extent_fn extent_fn,
2400 void *opaque,
2401 Error **errp)
2402 {
2403 int extent_idx;
2404 BlockBackend *blk = NULL;
2405 BlockBackend *extent_blk;
2406 Error *local_err = NULL;
2407 char *desc = NULL;
2408 int ret = 0;
2409 bool flat, split, compress;
2410 GString *ext_desc_lines;
2411 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
2412 int64_t extent_size;
2413 int64_t created_size = 0;
2414 const char *extent_line_fmt;
2415 char *parent_desc_line = g_malloc0(BUF_SIZE);
2416 uint32_t parent_cid = 0xffffffff;
2417 uint32_t number_heads = 16;
2418 uint32_t desc_offset = 0, desc_len;
2419 const char desc_template[] =
2420 "# Disk DescriptorFile\n"
2421 "version=1\n"
2422 "CID=%" PRIx32 "\n"
2423 "parentCID=%" PRIx32 "\n"
2424 "createType=\"%s\"\n"
2425 "%s"
2426 "\n"
2427 "# Extent description\n"
2428 "%s"
2429 "\n"
2430 "# The Disk Data Base\n"
2431 "#DDB\n"
2432 "\n"
2433 "ddb.virtualHWVersion = \"%s\"\n"
2434 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2435 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2436 "ddb.geometry.sectors = \"63\"\n"
2437 "ddb.adapterType = \"%s\"\n"
2438 "ddb.toolsVersion = \"%s\"\n";
2439
2440 ext_desc_lines = g_string_new(NULL);
2441
2442 /* Read out options */
2443 if (compat6) {
2444 if (hw_version) {
2445 error_setg(errp,
2446 "compat6 cannot be enabled with hwversion set");
2447 ret = -EINVAL;
2448 goto exit;
2449 }
2450 hw_version = "6";
2451 }
2452 if (!hw_version) {
2453 hw_version = "4";
2454 }
2455 if (!toolsversion) {
2456 toolsversion = "2147483647";
2457 }
2458
2459 if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2460 /* that's the number of heads with which vmware operates when
2461 creating, exporting, etc. vmdk files with a non-ide adapter type */
2462 number_heads = 255;
2463 }
2464 split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2465 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2466 flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2467 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2468 compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2469
2470 if (flat) {
2471 extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2472 } else {
2473 extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2474 }
2475 if (flat && backing_file) {
2476 error_setg(errp, "Flat image can't have backing file");
2477 ret = -ENOTSUP;
2478 goto exit;
2479 }
2480 if (flat && zeroed_grain) {
2481 error_setg(errp, "Flat image can't enable zeroed grain");
2482 ret = -ENOTSUP;
2483 goto exit;
2484 }
2485
2486 /* Create extents */
2487 if (split) {
2488 extent_size = split_size;
2489 } else {
2490 extent_size = size;
2491 }
2492 if (!split && !flat) {
2493 created_size = extent_size;
2494 } else {
2495 created_size = 0;
2496 }
2497 /* Get the descriptor file BDS */
2498 blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2499 opaque, errp);
2500 if (!blk) {
2501 ret = -EIO;
2502 goto exit;
2503 }
2504 if (!split && !flat) {
2505 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2506 blk_bs(blk)->filename);
2507 }
2508
2509 if (backing_file) {
2510 BlockBackend *backing;
2511 char *full_backing =
2512 bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2513 backing_file,
2514 &local_err);
2515 if (local_err) {
2516 error_propagate(errp, local_err);
2517 ret = -ENOENT;
2518 goto exit;
2519 }
2520 assert(full_backing);
2521
2522 backing = blk_co_new_open(full_backing, NULL, NULL,
2523 BDRV_O_NO_BACKING, errp);
2524 g_free(full_backing);
2525 if (backing == NULL) {
2526 ret = -EIO;
2527 goto exit;
2528 }
2529 if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2530 error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2531 blk_bs(backing)->drv->format_name);
2532 blk_co_unref(backing);
2533 ret = -EINVAL;
2534 goto exit;
2535 }
2536 ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2537 blk_co_unref(backing);
2538 if (ret) {
2539 error_setg(errp, "Failed to read parent CID");
2540 goto exit;
2541 }
2542 snprintf(parent_desc_line, BUF_SIZE,
2543 "parentFileNameHint=\"%s\"", backing_file);
2544 }
2545 extent_idx = 1;
2546 while (created_size < size) {
2547 int64_t cur_size = MIN(size - created_size, extent_size);
2548 extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2549 zeroed_grain, opaque, errp);
2550 if (!extent_blk) {
2551 ret = -EINVAL;
2552 goto exit;
2553 }
2554 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2555 blk_bs(extent_blk)->filename);
2556 created_size += cur_size;
2557 extent_idx++;
2558 blk_co_unref(extent_blk);
2559 }
2560
2561 /* Check whether we got excess extents */
2562 extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2563 opaque, NULL);
2564 if (extent_blk) {
2565 blk_co_unref(extent_blk);
2566 error_setg(errp, "List of extents contains unused extents");
2567 ret = -EINVAL;
2568 goto exit;
2569 }
2570
2571 /* generate descriptor file */
2572 desc = g_strdup_printf(desc_template,
2573 g_random_int(),
2574 parent_cid,
2575 BlockdevVmdkSubformat_str(subformat),
2576 parent_desc_line,
2577 ext_desc_lines->str,
2578 hw_version,
2579 size /
2580 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2581 number_heads,
2582 BlockdevVmdkAdapterType_str(adapter_type),
2583 toolsversion);
2584 desc_len = strlen(desc);
2585 /* the descriptor offset = 0x200 */
2586 if (!split && !flat) {
2587 desc_offset = 0x200;
2588 }
2589
2590 ret = blk_co_pwrite(blk, desc_offset, desc_len, desc, 0);
2591 if (ret < 0) {
2592 error_setg_errno(errp, -ret, "Could not write description");
2593 goto exit;
2594 }
2595 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2596 * for description file */
2597 if (desc_offset == 0) {
2598 ret = blk_co_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, 0, errp);
2599 if (ret < 0) {
2600 goto exit;
2601 }
2602 }
2603 ret = 0;
2604 exit:
2605 if (blk) {
2606 blk_co_unref(blk);
2607 }
2608 g_free(desc);
2609 g_free(parent_desc_line);
2610 g_string_free(ext_desc_lines, true);
2611 return ret;
2612 }
2613
2614 typedef struct {
2615 char *path;
2616 char *prefix;
2617 char *postfix;
2618 QemuOpts *opts;
2619 } VMDKCreateOptsData;
2620
2621 static BlockBackend * coroutine_fn GRAPH_UNLOCKED
2622 vmdk_co_create_opts_cb(int64_t size, int idx, bool flat, bool split,
2623 bool compress, bool zeroed_grain, void *opaque,
2624 Error **errp)
2625 {
2626 BlockBackend *blk = NULL;
2627 BlockDriverState *bs = NULL;
2628 VMDKCreateOptsData *data = opaque;
2629 char *ext_filename = NULL;
2630 char *rel_filename = NULL;
2631
2632 /* We're done, don't create excess extents. */
2633 if (size == -1) {
2634 assert(errp == NULL);
2635 return NULL;
2636 }
2637
2638 if (idx == 0) {
2639 rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2640 } else if (split) {
2641 rel_filename = g_strdup_printf("%s-%c%03d%s",
2642 data->prefix,
2643 flat ? 'f' : 's', idx, data->postfix);
2644 } else {
2645 assert(idx == 1);
2646 rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2647 }
2648
2649 ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2650 g_free(rel_filename);
2651
2652 if (vmdk_create_extent(ext_filename, size,
2653 flat, compress, zeroed_grain, &blk, data->opts,
2654 errp)) {
2655 goto exit;
2656 }
2657 bdrv_co_unref(bs);
2658 exit:
2659 g_free(ext_filename);
2660 return blk;
2661 }
2662
2663 static int coroutine_fn GRAPH_UNLOCKED
2664 vmdk_co_create_opts(BlockDriver *drv, const char *filename,
2665 QemuOpts *opts, Error **errp)
2666 {
2667 Error *local_err = NULL;
2668 char *desc = NULL;
2669 int64_t total_size = 0;
2670 char *adapter_type = NULL;
2671 BlockdevVmdkAdapterType adapter_type_enum;
2672 char *backing_file = NULL;
2673 char *hw_version = NULL;
2674 char *toolsversion = NULL;
2675 char *fmt = NULL;
2676 BlockdevVmdkSubformat subformat;
2677 int ret = 0;
2678 char *path = g_malloc0(PATH_MAX);
2679 char *prefix = g_malloc0(PATH_MAX);
2680 char *postfix = g_malloc0(PATH_MAX);
2681 char *desc_line = g_malloc0(BUF_SIZE);
2682 char *ext_filename = g_malloc0(PATH_MAX);
2683 char *desc_filename = g_malloc0(PATH_MAX);
2684 char *parent_desc_line = g_malloc0(BUF_SIZE);
2685 bool zeroed_grain;
2686 bool compat6;
2687 VMDKCreateOptsData data;
2688 char *backing_fmt = NULL;
2689
2690 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2691 if (backing_fmt && strcmp(backing_fmt, "vmdk") != 0) {
2692 error_setg(errp, "backing_file must be a vmdk image");
2693 ret = -EINVAL;
2694 goto exit;
2695 }
2696
2697 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
2698 ret = -EINVAL;
2699 goto exit;
2700 }
2701 /* Read out options */
2702 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2703 BDRV_SECTOR_SIZE);
2704 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
2705 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2706 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
2707 toolsversion = qemu_opt_get_del(opts, BLOCK_OPT_TOOLSVERSION);
2708 compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false);
2709 if (strcmp(hw_version, "undefined") == 0) {
2710 g_free(hw_version);
2711 hw_version = NULL;
2712 }
2713 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
2714 zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false);
2715
2716 if (adapter_type) {
2717 adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup,
2718 adapter_type,
2719 BLOCKDEV_VMDK_ADAPTER_TYPE_IDE,
2720 &local_err);
2721 if (local_err) {
2722 error_propagate(errp, local_err);
2723 ret = -EINVAL;
2724 goto exit;
2725 }
2726 } else {
2727 adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE;
2728 }
2729
2730 if (!fmt) {
2731 /* Default format to monolithicSparse */
2732 subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE;
2733 } else {
2734 subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup,
2735 fmt,
2736 BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE,
2737 &local_err);
2738 if (local_err) {
2739 error_propagate(errp, local_err);
2740 ret = -EINVAL;
2741 goto exit;
2742 }
2743 }
2744 data = (VMDKCreateOptsData){
2745 .prefix = prefix,
2746 .postfix = postfix,
2747 .path = path,
2748 .opts = opts,
2749 };
2750 ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum,
2751 backing_file, hw_version, toolsversion, compat6,
2752 zeroed_grain, vmdk_co_create_opts_cb, &data, errp);
2753
2754 exit:
2755 g_free(backing_fmt);
2756 g_free(adapter_type);
2757 g_free(backing_file);
2758 g_free(hw_version);
2759 g_free(toolsversion);
2760 g_free(fmt);
2761 g_free(desc);
2762 g_free(path);
2763 g_free(prefix);
2764 g_free(postfix);
2765 g_free(desc_line);
2766 g_free(ext_filename);
2767 g_free(desc_filename);
2768 g_free(parent_desc_line);
2769 return ret;
2770 }
2771
2772 static BlockBackend * coroutine_fn GRAPH_UNLOCKED
2773 vmdk_co_create_cb(int64_t size, int idx, bool flat, bool split, bool compress,
2774 bool zeroed_grain, void *opaque, Error **errp)
2775 {
2776 int ret;
2777 BlockDriverState *bs;
2778 BlockBackend *blk;
2779 BlockdevCreateOptionsVmdk *opts = opaque;
2780
2781 if (idx == 0) {
2782 bs = bdrv_co_open_blockdev_ref(opts->file, errp);
2783 } else {
2784 int i;
2785 BlockdevRefList *list = opts->extents;
2786 for (i = 1; i < idx; i++) {
2787 if (!list || !list->next) {
2788 error_setg(errp, "Extent [%d] not specified", i);
2789 return NULL;
2790 }
2791 list = list->next;
2792 }
2793 if (!list) {
2794 error_setg(errp, "Extent [%d] not specified", idx - 1);
2795 return NULL;
2796 }
2797 bs = bdrv_co_open_blockdev_ref(list->value, errp);
2798 }
2799 if (!bs) {
2800 return NULL;
2801 }
2802 blk = blk_co_new_with_bs(bs,
2803 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
2804 BLK_PERM_RESIZE,
2805 BLK_PERM_ALL,
2806 errp);
2807 if (!blk) {
2808 return NULL;
2809 }
2810 blk_set_allow_write_beyond_eof(blk, true);
2811 bdrv_co_unref(bs);
2812
2813 if (size != -1) {
2814 ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2815 if (ret) {
2816 blk_co_unref(blk);
2817 blk = NULL;
2818 }
2819 }
2820 return blk;
2821 }
2822
2823 static int coroutine_fn GRAPH_UNLOCKED
2824 vmdk_co_create(BlockdevCreateOptions *create_options, Error **errp)
2825 {
2826 BlockdevCreateOptionsVmdk *opts;
2827
2828 opts = &create_options->u.vmdk;
2829
2830 /* Validate options */
2831 if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2832 error_setg(errp, "Image size must be a multiple of 512 bytes");
2833 return -EINVAL;
2834 }
2835
2836 return vmdk_co_do_create(opts->size,
2837 opts->subformat,
2838 opts->adapter_type,
2839 opts->backing_file,
2840 opts->hwversion,
2841 opts->toolsversion,
2842 false,
2843 opts->zeroed_grain,
2844 vmdk_co_create_cb,
2845 opts, errp);
2846 }
2847
2848 static void vmdk_close(BlockDriverState *bs)
2849 {
2850 BDRVVmdkState *s = bs->opaque;
2851
2852 vmdk_free_extents(bs);
2853 g_free(s->create_type);
2854
2855 migrate_del_blocker(s->migration_blocker);
2856 error_free(s->migration_blocker);
2857 }
2858
2859 static int64_t coroutine_fn GRAPH_RDLOCK
2860 vmdk_co_get_allocated_file_size(BlockDriverState *bs)
2861 {
2862 int i;
2863 int64_t ret = 0;
2864 int64_t r;
2865 BDRVVmdkState *s = bs->opaque;
2866
2867 ret = bdrv_co_get_allocated_file_size(bs->file->bs);
2868 if (ret < 0) {
2869 return ret;
2870 }
2871 for (i = 0; i < s->num_extents; i++) {
2872 if (s->extents[i].file == bs->file) {
2873 continue;
2874 }
2875 r = bdrv_co_get_allocated_file_size(s->extents[i].file->bs);
2876 if (r < 0) {
2877 return r;
2878 }
2879 ret += r;
2880 }
2881 return ret;
2882 }
2883
2884 static int vmdk_has_zero_init(BlockDriverState *bs)
2885 {
2886 int i;
2887 BDRVVmdkState *s = bs->opaque;
2888
2889 /* If has a flat extent and its underlying storage doesn't have zero init,
2890 * return 0. */
2891 for (i = 0; i < s->num_extents; i++) {
2892 if (s->extents[i].flat) {
2893 if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2894 return 0;
2895 }
2896 }
2897 }
2898 return 1;
2899 }
2900
2901 static VmdkExtentInfo *vmdk_get_extent_info(VmdkExtent *extent)
2902 {
2903 VmdkExtentInfo *info = g_new0(VmdkExtentInfo, 1);
2904
2905 bdrv_refresh_filename(extent->file->bs);
2906 *info = (VmdkExtentInfo){
2907 .filename = g_strdup(extent->file->bs->filename),
2908 .format = g_strdup(extent->type),
2909 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2910 .compressed = extent->compressed,
2911 .has_compressed = extent->compressed,
2912 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2913 .has_cluster_size = !extent->flat,
2914 };
2915
2916 return info;
2917 }
2918
2919 static int coroutine_fn GRAPH_RDLOCK
2920 vmdk_co_check(BlockDriverState *bs, BdrvCheckResult *result, BdrvCheckMode fix)
2921 {
2922 BDRVVmdkState *s = bs->opaque;
2923 VmdkExtent *extent = NULL;
2924 int64_t sector_num = 0;
2925 int64_t total_sectors = bdrv_co_nb_sectors(bs);
2926 int ret;
2927 uint64_t cluster_offset;
2928
2929 if (fix) {
2930 return -ENOTSUP;
2931 }
2932
2933 for (;;) {
2934 if (sector_num >= total_sectors) {
2935 return 0;
2936 }
2937 extent = find_extent(s, sector_num, extent);
2938 if (!extent) {
2939 fprintf(stderr,
2940 "ERROR: could not find extent for sector %" PRId64 "\n",
2941 sector_num);
2942 ret = -EINVAL;
2943 break;
2944 }
2945 ret = get_cluster_offset(bs, extent, NULL,
2946 sector_num << BDRV_SECTOR_BITS,
2947 false, &cluster_offset, 0, 0);
2948 if (ret == VMDK_ERROR) {
2949 fprintf(stderr,
2950 "ERROR: could not get cluster_offset for sector %"
2951 PRId64 "\n", sector_num);
2952 break;
2953 }
2954 if (ret == VMDK_OK) {
2955 int64_t extent_len = bdrv_co_getlength(extent->file->bs);
2956 if (extent_len < 0) {
2957 fprintf(stderr,
2958 "ERROR: could not get extent file length for sector %"
2959 PRId64 "\n", sector_num);
2960 ret = extent_len;
2961 break;
2962 }
2963 if (cluster_offset >= extent_len) {
2964 fprintf(stderr,
2965 "ERROR: cluster offset for sector %"
2966 PRId64 " points after EOF\n", sector_num);
2967 ret = -EINVAL;
2968 break;
2969 }
2970 }
2971 sector_num += extent->cluster_sectors;
2972 }
2973
2974 result->corruptions++;
2975 return ret;
2976 }
2977
2978 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs,
2979 Error **errp)
2980 {
2981 int i;
2982 BDRVVmdkState *s = bs->opaque;
2983 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2984 VmdkExtentInfoList **tail;
2985
2986 *spec_info = (ImageInfoSpecific){
2987 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2988 .u = {
2989 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2990 },
2991 };
2992
2993 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2994 .create_type = g_strdup(s->create_type),
2995 .cid = s->cid,
2996 .parent_cid = s->parent_cid,
2997 };
2998
2999 tail = &spec_info->u.vmdk.data->extents;
3000 for (i = 0; i < s->num_extents; i++) {
3001 QAPI_LIST_APPEND(tail, vmdk_get_extent_info(&s->extents[i]));
3002 }
3003
3004 return spec_info;
3005 }
3006
3007 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
3008 {
3009 return a->flat == b->flat &&
3010 a->compressed == b->compressed &&
3011 (a->flat || a->cluster_sectors == b->cluster_sectors);
3012 }
3013
3014 static int coroutine_fn
3015 vmdk_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3016 {
3017 int i;
3018 BDRVVmdkState *s = bs->opaque;
3019 assert(s->num_extents);
3020
3021 /* See if we have multiple extents but they have different cases */
3022 for (i = 1; i < s->num_extents; i++) {
3023 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
3024 return -ENOTSUP;
3025 }
3026 }
3027 bdi->needs_compressed_writes = s->extents[0].compressed;
3028 if (!s->extents[0].flat) {
3029 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
3030 }
3031 return 0;
3032 }
3033
3034 static void vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
3035 bool backing_overridden)
3036 {
3037 /* No children but file and backing can be explicitly specified (TODO) */
3038 qdict_put(target, "file",
3039 qobject_ref(bs->file->bs->full_open_options));
3040
3041 if (backing_overridden) {
3042 if (bs->backing) {
3043 qdict_put(target, "backing",
3044 qobject_ref(bs->backing->bs->full_open_options));
3045 } else {
3046 qdict_put_null(target, "backing");
3047 }
3048 }
3049 }
3050
3051 static QemuOptsList vmdk_create_opts = {
3052 .name = "vmdk-create-opts",
3053 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
3054 .desc = {
3055 {
3056 .name = BLOCK_OPT_SIZE,
3057 .type = QEMU_OPT_SIZE,
3058 .help = "Virtual disk size"
3059 },
3060 {
3061 .name = BLOCK_OPT_ADAPTER_TYPE,
3062 .type = QEMU_OPT_STRING,
3063 .help = "Virtual adapter type, can be one of "
3064 "ide (default), lsilogic, buslogic or legacyESX"
3065 },
3066 {
3067 .name = BLOCK_OPT_BACKING_FILE,
3068 .type = QEMU_OPT_STRING,
3069 .help = "File name of a base image"
3070 },
3071 {
3072 .name = BLOCK_OPT_BACKING_FMT,
3073 .type = QEMU_OPT_STRING,
3074 .help = "Must be 'vmdk' if present",
3075 },
3076 {
3077 .name = BLOCK_OPT_COMPAT6,
3078 .type = QEMU_OPT_BOOL,
3079 .help = "VMDK version 6 image",
3080 .def_value_str = "off"
3081 },
3082 {
3083 .name = BLOCK_OPT_HWVERSION,
3084 .type = QEMU_OPT_STRING,
3085 .help = "VMDK hardware version",
3086 .def_value_str = "undefined"
3087 },
3088 {
3089 .name = BLOCK_OPT_TOOLSVERSION,
3090 .type = QEMU_OPT_STRING,
3091 .help = "VMware guest tools version",
3092 },
3093 {
3094 .name = BLOCK_OPT_SUBFMT,
3095 .type = QEMU_OPT_STRING,
3096 .help =
3097 "VMDK flat extent format, can be one of "
3098 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
3099 },
3100 {
3101 .name = BLOCK_OPT_ZEROED_GRAIN,
3102 .type = QEMU_OPT_BOOL,
3103 .help = "Enable efficient zero writes "
3104 "using the zeroed-grain GTE feature"
3105 },
3106 { /* end of list */ }
3107 }
3108 };
3109
3110 static BlockDriver bdrv_vmdk = {
3111 .format_name = "vmdk",
3112 .instance_size = sizeof(BDRVVmdkState),
3113 .bdrv_probe = vmdk_probe,
3114 .bdrv_open = vmdk_open,
3115 .bdrv_co_check = vmdk_co_check,
3116 .bdrv_reopen_prepare = vmdk_reopen_prepare,
3117 .bdrv_reopen_commit = vmdk_reopen_commit,
3118 .bdrv_reopen_abort = vmdk_reopen_abort,
3119 .bdrv_child_perm = bdrv_default_perms,
3120 .bdrv_co_preadv = vmdk_co_preadv,
3121 .bdrv_co_pwritev = vmdk_co_pwritev,
3122 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
3123 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
3124 .bdrv_close = vmdk_close,
3125 .bdrv_co_create_opts = vmdk_co_create_opts,
3126 .bdrv_co_create = vmdk_co_create,
3127 .bdrv_co_block_status = vmdk_co_block_status,
3128 .bdrv_co_get_allocated_file_size = vmdk_co_get_allocated_file_size,
3129 .bdrv_has_zero_init = vmdk_has_zero_init,
3130 .bdrv_get_specific_info = vmdk_get_specific_info,
3131 .bdrv_refresh_limits = vmdk_refresh_limits,
3132 .bdrv_co_get_info = vmdk_co_get_info,
3133 .bdrv_gather_child_options = vmdk_gather_child_options,
3134
3135 .is_format = true,
3136 .supports_backing = true,
3137 .create_opts = &vmdk_create_opts,
3138 };
3139
3140 static void bdrv_vmdk_init(void)
3141 {
3142 bdrv_register(&bdrv_vmdk);
3143 }
3144
3145 block_init(bdrv_vmdk_init);
AR7 emulation for QEMU