?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
migration: simplify blockers
[qemu/kevin.git] / block / vmdk.c
blob1335d39e16efbbf0ee85c3dc7b73a2acbb6dacf7
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 GRAPH_RDLOCK
582 vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent, 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 GRAPH_RDLOCK
645 vmdk_open_vmfs_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
646 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 GRAPH_RDLOCK
801 vmdk_open_se_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
802 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 GRAPH_RDLOCK
917 vmdk_open_vmdk4(BlockDriverState *bs, BdrvChild *file, int flags,
918 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 GRAPH_RDLOCK
1099 vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
1100 char *buf, QDict *options, Error **errp)
1101 {
1102 uint32_t magic;
1103
1104 magic = ldl_be_p(buf);
1105 switch (magic) {
1106 case VMDK3_MAGIC:
1107 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
1108 case VMDK4_MAGIC:
1109 return vmdk_open_vmdk4(bs, file, flags, options, errp);
1110 default:
1111 error_setg(errp, "Image not in VMDK format");
1112 return -EINVAL;
1113 }
1114 }
1115
1116 static const char *next_line(const char *s)
1117 {
1118 while (*s) {
1119 if (*s == '\n') {
1120 return s + 1;
1121 }
1122 s++;
1123 }
1124 return s;
1125 }
1126
1127 static int GRAPH_RDLOCK
1128 vmdk_parse_extents(const char *desc, BlockDriverState *bs, QDict *options,
1129 Error **errp)
1130 {
1131 int ret;
1132 int matches;
1133 char access[11];
1134 char type[11];
1135 char fname[512];
1136 const char *p, *np;
1137 int64_t sectors = 0;
1138 int64_t flat_offset;
1139 char *desc_file_dir = NULL;
1140 char *extent_path;
1141 BdrvChild *extent_file;
1142 BdrvChildRole extent_role;
1143 BDRVVmdkState *s = bs->opaque;
1144 VmdkExtent *extent = NULL;
1145 char extent_opt_prefix[32];
1146 Error *local_err = NULL;
1147
1148 GLOBAL_STATE_CODE();
1149
1150 for (p = desc; *p; p = next_line(p)) {
1151 /* parse extent line in one of below formats:
1152 *
1153 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
1154 * RW [size in sectors] SPARSE "file-name.vmdk"
1155 * RW [size in sectors] VMFS "file-name.vmdk"
1156 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
1157 * RW [size in sectors] SESPARSE "file-name.vmdk"
1158 */
1159 flat_offset = -1;
1160 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
1161 access, &sectors, type, fname, &flat_offset);
1162 if (matches < 4 || strcmp(access, "RW")) {
1163 continue;
1164 } else if (!strcmp(type, "FLAT")) {
1165 if (matches != 5 || flat_offset < 0) {
1166 goto invalid;
1167 }
1168 } else if (!strcmp(type, "VMFS")) {
1169 if (matches == 4) {
1170 flat_offset = 0;
1171 } else {
1172 goto invalid;
1173 }
1174 } else if (matches != 4) {
1175 goto invalid;
1176 }
1177
1178 if (sectors <= 0 ||
1179 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
1180 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") &&
1181 strcmp(type, "SESPARSE")) ||
1182 (strcmp(access, "RW"))) {
1183 continue;
1184 }
1185
1186 if (path_is_absolute(fname)) {
1187 extent_path = g_strdup(fname);
1188 } else {
1189 if (!desc_file_dir) {
1190 desc_file_dir = bdrv_dirname(bs->file->bs, errp);
1191 if (!desc_file_dir) {
1192 bdrv_refresh_filename(bs->file->bs);
1193 error_prepend(errp, "Cannot use relative paths with VMDK "
1194 "descriptor file '%s': ",
1195 bs->file->bs->filename);
1196 ret = -EINVAL;
1197 goto out;
1198 }
1199 }
1200
1201 extent_path = g_strconcat(desc_file_dir, fname, NULL);
1202 }
1203
1204 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
1205 assert(ret < 32);
1206
1207 extent_role = BDRV_CHILD_DATA;
1208 if (strcmp(type, "FLAT") != 0 && strcmp(type, "VMFS") != 0) {
1209 /* non-flat extents have metadata */
1210 extent_role |= BDRV_CHILD_METADATA;
1211 }
1212
1213 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
1214 bs, &child_of_bds, extent_role, false,
1215 &local_err);
1216 g_free(extent_path);
1217 if (!extent_file) {
1218 error_propagate(errp, local_err);
1219 ret = -EINVAL;
1220 goto out;
1221 }
1222
1223 /* save to extents array */
1224 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
1225 /* FLAT extent */
1226
1227 ret = vmdk_add_extent(bs, extent_file, true, sectors,
1228 0, 0, 0, 0, 0, &extent, errp);
1229 if (ret < 0) {
1230 bdrv_graph_rdunlock_main_loop();
1231 bdrv_graph_wrlock(NULL);
1232 bdrv_unref_child(bs, extent_file);
1233 bdrv_graph_wrunlock();
1234 bdrv_graph_rdlock_main_loop();
1235 goto out;
1236 }
1237 extent->flat_start_offset = flat_offset << 9;
1238 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
1239 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
1240 char *buf = vmdk_read_desc(extent_file, 0, errp);
1241 if (!buf) {
1242 ret = -EINVAL;
1243 } else {
1244 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
1245 options, errp);
1246 }
1247 g_free(buf);
1248 if (ret) {
1249 bdrv_graph_rdunlock_main_loop();
1250 bdrv_graph_wrlock(NULL);
1251 bdrv_unref_child(bs, extent_file);
1252 bdrv_graph_wrunlock();
1253 bdrv_graph_rdlock_main_loop();
1254 goto out;
1255 }
1256 extent = &s->extents[s->num_extents - 1];
1257 } else if (!strcmp(type, "SESPARSE")) {
1258 ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
1259 if (ret) {
1260 bdrv_graph_rdunlock_main_loop();
1261 bdrv_graph_wrlock(NULL);
1262 bdrv_unref_child(bs, extent_file);
1263 bdrv_graph_wrunlock();
1264 bdrv_graph_rdlock_main_loop();
1265 goto out;
1266 }
1267 extent = &s->extents[s->num_extents - 1];
1268 } else {
1269 error_setg(errp, "Unsupported extent type '%s'", type);
1270 bdrv_graph_rdunlock_main_loop();
1271 bdrv_graph_wrlock(NULL);
1272 bdrv_unref_child(bs, extent_file);
1273 bdrv_graph_wrunlock();
1274 bdrv_graph_rdlock_main_loop();
1275 ret = -ENOTSUP;
1276 goto out;
1277 }
1278 extent->type = g_strdup(type);
1279 }
1280
1281 ret = 0;
1282 goto out;
1283
1284 invalid:
1285 np = next_line(p);
1286 assert(np != p);
1287 if (np[-1] == '\n') {
1288 np--;
1289 }
1290 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
1291 ret = -EINVAL;
1292
1293 out:
1294 g_free(desc_file_dir);
1295 return ret;
1296 }
1297
1298 static int GRAPH_RDLOCK
1299 vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, QDict *options,
1300 Error **errp)
1301 {
1302 int ret;
1303 char ct[128];
1304 BDRVVmdkState *s = bs->opaque;
1305
1306 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
1307 error_setg(errp, "invalid VMDK image descriptor");
1308 ret = -EINVAL;
1309 goto exit;
1310 }
1311 if (strcmp(ct, "monolithicFlat") &&
1312 strcmp(ct, "vmfs") &&
1313 strcmp(ct, "vmfsSparse") &&
1314 strcmp(ct, "seSparse") &&
1315 strcmp(ct, "twoGbMaxExtentSparse") &&
1316 strcmp(ct, "twoGbMaxExtentFlat")) {
1317 error_setg(errp, "Unsupported image type '%s'", ct);
1318 ret = -ENOTSUP;
1319 goto exit;
1320 }
1321 s->create_type = g_strdup(ct);
1322 s->desc_offset = 0;
1323 ret = vmdk_parse_extents(buf, bs, options, errp);
1324 exit:
1325 return ret;
1326 }
1327
1328 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
1329 Error **errp)
1330 {
1331 char *buf;
1332 int ret;
1333 BDRVVmdkState *s = bs->opaque;
1334 uint32_t magic;
1335
1336 GRAPH_RDLOCK_GUARD_MAINLOOP();
1337
1338 ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
1339 if (ret < 0) {
1340 return ret;
1341 }
1342
1343 buf = vmdk_read_desc(bs->file, 0, errp);
1344 if (!buf) {
1345 return -EINVAL;
1346 }
1347
1348 magic = ldl_be_p(buf);
1349 switch (magic) {
1350 case VMDK3_MAGIC:
1351 case VMDK4_MAGIC:
1352 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
1353 errp);
1354 s->desc_offset = 0x200;
1355 break;
1356 default:
1357 /* No data in the descriptor file */
1358 bs->file->role &= ~BDRV_CHILD_DATA;
1359
1360 /* Must succeed because we have given up permissions if anything */
1361 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1362
1363 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1364 break;
1365 }
1366 if (ret) {
1367 goto fail;
1368 }
1369
1370 /* try to open parent images, if exist */
1371 ret = vmdk_parent_open(bs);
1372 if (ret) {
1373 goto fail;
1374 }
1375 ret = vmdk_read_cid(bs, 0, &s->cid);
1376 if (ret) {
1377 goto fail;
1378 }
1379 ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1380 if (ret) {
1381 goto fail;
1382 }
1383 qemu_co_mutex_init(&s->lock);
1384
1385 /* Disable migration when VMDK images are used */
1386 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1387 "does not support live migration",
1388 bdrv_get_device_or_node_name(bs));
1389 ret = migrate_add_blocker(&s->migration_blocker, errp);
1390 if (ret < 0) {
1391 goto fail;
1392 }
1393
1394 g_free(buf);
1395 return 0;
1396
1397 fail:
1398 g_free(buf);
1399 g_free(s->create_type);
1400 s->create_type = NULL;
1401 vmdk_free_extents(bs);
1402 return ret;
1403 }
1404
1405
1406 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1407 {
1408 BDRVVmdkState *s = bs->opaque;
1409 int i;
1410
1411 for (i = 0; i < s->num_extents; i++) {
1412 if (!s->extents[i].flat) {
1413 bs->bl.pwrite_zeroes_alignment =
1414 MAX(bs->bl.pwrite_zeroes_alignment,
1415 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1416 }
1417 }
1418 }
1419
1420 /**
1421 * get_whole_cluster
1422 *
1423 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1424 * to the cluster at @cluster_sector_num. If @zeroed is true, we're overwriting
1425 * a zeroed cluster in the current layer and must not copy data from the
1426 * backing file.
1427 *
1428 * If @skip_start_sector < @skip_end_sector, the relative range
1429 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1430 * it for call to write user data in the request.
1431 */
1432 static int coroutine_fn GRAPH_RDLOCK
1433 get_whole_cluster(BlockDriverState *bs, VmdkExtent *extent,
1434 uint64_t cluster_offset, uint64_t offset,
1435 uint64_t skip_start_bytes, uint64_t skip_end_bytes,
1436 bool zeroed)
1437 {
1438 int ret = VMDK_OK;
1439 int64_t cluster_bytes;
1440 uint8_t *whole_grain;
1441 bool copy_from_backing;
1442
1443 /* For COW, align request sector_num to cluster start */
1444 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1445 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1446 whole_grain = qemu_blockalign(bs, cluster_bytes);
1447 copy_from_backing = bs->backing && !zeroed;
1448
1449 if (!copy_from_backing) {
1450 memset(whole_grain, 0, skip_start_bytes);
1451 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1452 }
1453
1454 assert(skip_end_bytes <= cluster_bytes);
1455 /* we will be here if it's first write on non-exist grain(cluster).
1456 * try to read from parent image, if exist */
1457 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1458 ret = VMDK_ERROR;
1459 goto exit;
1460 }
1461
1462 /* Read backing data before skip range */
1463 if (skip_start_bytes > 0) {
1464 if (copy_from_backing) {
1465 /* qcow2 emits this on bs->file instead of bs->backing */
1466 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_READ);
1467 ret = bdrv_co_pread(bs->backing, offset, skip_start_bytes,
1468 whole_grain, 0);
1469 if (ret < 0) {
1470 ret = VMDK_ERROR;
1471 goto exit;
1472 }
1473 }
1474 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_WRITE);
1475 ret = bdrv_co_pwrite(extent->file, cluster_offset, skip_start_bytes,
1476 whole_grain, 0);
1477 if (ret < 0) {
1478 ret = VMDK_ERROR;
1479 goto exit;
1480 }
1481 }
1482 /* Read backing data after skip range */
1483 if (skip_end_bytes < cluster_bytes) {
1484 if (copy_from_backing) {
1485 /* qcow2 emits this on bs->file instead of bs->backing */
1486 BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_READ);
1487 ret = bdrv_co_pread(bs->backing, offset + skip_end_bytes,
1488 cluster_bytes - skip_end_bytes,
1489 whole_grain + skip_end_bytes, 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_end_bytes,
1497 cluster_bytes - skip_end_bytes,
1498 whole_grain + skip_end_bytes, 0);
1499 if (ret < 0) {
1500 ret = VMDK_ERROR;
1501 goto exit;
1502 }
1503 }
1504
1505 ret = VMDK_OK;
1506 exit:
1507 qemu_vfree(whole_grain);
1508 return ret;
1509 }
1510
1511 static int coroutine_fn GRAPH_RDLOCK
1512 vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data, uint32_t offset)
1513 {
1514 offset = cpu_to_le32(offset);
1515 /* update L2 table */
1516 BLKDBG_CO_EVENT(extent->file, BLKDBG_L2_UPDATE);
1517 if (bdrv_co_pwrite(extent->file,
1518 ((int64_t)m_data->l2_offset * 512)
1519 + (m_data->l2_index * sizeof(offset)),
1520 sizeof(offset), &offset, 0) < 0) {
1521 return VMDK_ERROR;
1522 }
1523 /* update backup L2 table */
1524 if (extent->l1_backup_table_offset != 0) {
1525 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1526 if (bdrv_co_pwrite(extent->file,
1527 ((int64_t)m_data->l2_offset * 512)
1528 + (m_data->l2_index * sizeof(offset)),
1529 sizeof(offset), &offset, 0) < 0) {
1530 return VMDK_ERROR;
1531 }
1532 }
1533 if (bdrv_co_flush(extent->file->bs) < 0) {
1534 return VMDK_ERROR;
1535 }
1536 if (m_data->l2_cache_entry) {
1537 *m_data->l2_cache_entry = offset;
1538 }
1539
1540 return VMDK_OK;
1541 }
1542
1543 /**
1544 * get_cluster_offset
1545 *
1546 * Look up cluster offset in extent file by sector number, and store in
1547 * @cluster_offset.
1548 *
1549 * For flat extents, the start offset as parsed from the description file is
1550 * returned.
1551 *
1552 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1553 * offset for a new cluster and update L2 cache. If there is a backing file,
1554 * COW is done before returning; otherwise, zeroes are written to the allocated
1555 * cluster. Both COW and zero writing skips the sector range
1556 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1557 * has new data to write there.
1558 *
1559 * Returns: VMDK_OK if cluster exists and mapped in the image.
1560 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1561 * VMDK_ERROR if failed.
1562 */
1563 static int coroutine_fn GRAPH_RDLOCK
1564 get_cluster_offset(BlockDriverState *bs, VmdkExtent *extent,
1565 VmdkMetaData *m_data, uint64_t offset, bool allocate,
1566 uint64_t *cluster_offset, uint64_t skip_start_bytes,
1567 uint64_t skip_end_bytes)
1568 {
1569 unsigned int l1_index, l2_offset, l2_index;
1570 int min_index, i, j;
1571 uint32_t min_count;
1572 void *l2_table;
1573 bool zeroed = false;
1574 int64_t ret;
1575 int64_t cluster_sector;
1576 unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
1577
1578 if (m_data) {
1579 m_data->new_allocation = false;
1580 }
1581 if (extent->flat) {
1582 *cluster_offset = extent->flat_start_offset;
1583 return VMDK_OK;
1584 }
1585
1586 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1587 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1588 if (l1_index >= extent->l1_size) {
1589 return VMDK_ERROR;
1590 }
1591 if (extent->sesparse) {
1592 uint64_t l2_offset_u64;
1593
1594 assert(extent->entry_size == sizeof(uint64_t));
1595
1596 l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
1597 if (l2_offset_u64 == 0) {
1598 l2_offset = 0;
1599 } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
1600 /*
1601 * Top most nibble is 0x1 if grain table is allocated.
1602 * strict check - top most 4 bytes must be 0x10000000 since max
1603 * supported size is 64TB for disk - so no more than 64TB / 16MB
1604 * grain directories which is smaller than uint32,
1605 * where 16MB is the only supported default grain table coverage.
1606 */
1607 return VMDK_ERROR;
1608 } else {
1609 l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
1610 l2_offset_u64 = extent->sesparse_l2_tables_offset +
1611 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
1612 if (l2_offset_u64 > 0x00000000ffffffff) {
1613 return VMDK_ERROR;
1614 }
1615 l2_offset = (unsigned int)(l2_offset_u64);
1616 }
1617 } else {
1618 assert(extent->entry_size == sizeof(uint32_t));
1619 l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
1620 }
1621 if (!l2_offset) {
1622 return VMDK_UNALLOC;
1623 }
1624 for (i = 0; i < L2_CACHE_SIZE; i++) {
1625 if (l2_offset == extent->l2_cache_offsets[i]) {
1626 /* increment the hit count */
1627 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1628 for (j = 0; j < L2_CACHE_SIZE; j++) {
1629 extent->l2_cache_counts[j] >>= 1;
1630 }
1631 }
1632 l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
1633 goto found;
1634 }
1635 }
1636 /* not found: load a new entry in the least used one */
1637 min_index = 0;
1638 min_count = 0xffffffff;
1639 for (i = 0; i < L2_CACHE_SIZE; i++) {
1640 if (extent->l2_cache_counts[i] < min_count) {
1641 min_count = extent->l2_cache_counts[i];
1642 min_index = i;
1643 }
1644 }
1645 l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
1646 BLKDBG_CO_EVENT(extent->file, BLKDBG_L2_LOAD);
1647 if (bdrv_co_pread(extent->file,
1648 (int64_t)l2_offset * 512,
1649 l2_size_bytes,
1650 l2_table, 0
1651 ) < 0) {
1652 return VMDK_ERROR;
1653 }
1654
1655 extent->l2_cache_offsets[min_index] = l2_offset;
1656 extent->l2_cache_counts[min_index] = 1;
1657 found:
1658 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1659 if (m_data) {
1660 m_data->l1_index = l1_index;
1661 m_data->l2_index = l2_index;
1662 m_data->l2_offset = l2_offset;
1663 m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
1664 }
1665
1666 if (extent->sesparse) {
1667 cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
1668 switch (cluster_sector & 0xf000000000000000) {
1669 case 0x0000000000000000:
1670 /* unallocated grain */
1671 if (cluster_sector != 0) {
1672 return VMDK_ERROR;
1673 }
1674 break;
1675 case 0x1000000000000000:
1676 /* scsi-unmapped grain - fallthrough */
1677 case 0x2000000000000000:
1678 /* zero grain */
1679 zeroed = true;
1680 break;
1681 case 0x3000000000000000:
1682 /* allocated grain */
1683 cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
1684 ((cluster_sector & 0x0000ffffffffffff) << 12));
1685 cluster_sector = extent->sesparse_clusters_offset +
1686 cluster_sector * extent->cluster_sectors;
1687 break;
1688 default:
1689 return VMDK_ERROR;
1690 }
1691 } else {
1692 cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
1693
1694 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1695 zeroed = true;
1696 }
1697 }
1698
1699 if (!cluster_sector || zeroed) {
1700 if (!allocate) {
1701 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1702 }
1703 assert(!extent->sesparse);
1704
1705 if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1706 return VMDK_ERROR;
1707 }
1708
1709 cluster_sector = extent->next_cluster_sector;
1710 extent->next_cluster_sector += extent->cluster_sectors;
1711
1712 /* First of all we write grain itself, to avoid race condition
1713 * that may to corrupt the image.
1714 * This problem may occur because of insufficient space on host disk
1715 * or inappropriate VM shutdown.
1716 */
1717 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1718 offset, skip_start_bytes, skip_end_bytes,
1719 zeroed);
1720 if (ret) {
1721 return ret;
1722 }
1723 if (m_data) {
1724 m_data->new_allocation = true;
1725 }
1726 }
1727 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1728 return VMDK_OK;
1729 }
1730
1731 static VmdkExtent *find_extent(BDRVVmdkState *s,
1732 int64_t sector_num, VmdkExtent *start_hint)
1733 {
1734 VmdkExtent *extent = start_hint;
1735
1736 if (!extent) {
1737 extent = &s->extents[0];
1738 }
1739 while (extent < &s->extents[s->num_extents]) {
1740 if (sector_num < extent->end_sector) {
1741 return extent;
1742 }
1743 extent++;
1744 }
1745 return NULL;
1746 }
1747
1748 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1749 int64_t offset)
1750 {
1751 uint64_t extent_begin_offset, extent_relative_offset;
1752 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1753
1754 extent_begin_offset =
1755 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1756 extent_relative_offset = offset - extent_begin_offset;
1757 return extent_relative_offset % cluster_size;
1758 }
1759
1760 static int coroutine_fn GRAPH_RDLOCK
1761 vmdk_co_block_status(BlockDriverState *bs, bool want_zero,
1762 int64_t offset, int64_t bytes, int64_t *pnum,
1763 int64_t *map, BlockDriverState **file)
1764 {
1765 BDRVVmdkState *s = bs->opaque;
1766 int64_t index_in_cluster, n, ret;
1767 uint64_t cluster_offset;
1768 VmdkExtent *extent;
1769
1770 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1771 if (!extent) {
1772 return -EIO;
1773 }
1774 qemu_co_mutex_lock(&s->lock);
1775 ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1776 0, 0);
1777 qemu_co_mutex_unlock(&s->lock);
1778
1779 index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1780 switch (ret) {
1781 case VMDK_ERROR:
1782 ret = -EIO;
1783 break;
1784 case VMDK_UNALLOC:
1785 ret = 0;
1786 break;
1787 case VMDK_ZEROED:
1788 ret = BDRV_BLOCK_ZERO;
1789 break;
1790 case VMDK_OK:
1791 ret = BDRV_BLOCK_DATA;
1792 if (!extent->compressed) {
1793 ret |= BDRV_BLOCK_OFFSET_VALID;
1794 *map = cluster_offset + index_in_cluster;
1795 if (extent->flat) {
1796 ret |= BDRV_BLOCK_RECURSE;
1797 }
1798 } else {
1799 ret |= BDRV_BLOCK_COMPRESSED;
1800 }
1801 *file = extent->file->bs;
1802 break;
1803 }
1804
1805 n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1806 *pnum = MIN(n, bytes);
1807 return ret;
1808 }
1809
1810 static int coroutine_fn GRAPH_RDLOCK
1811 vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1812 int64_t offset_in_cluster, QEMUIOVector *qiov,
1813 uint64_t qiov_offset, uint64_t n_bytes,
1814 uint64_t offset)
1815 {
1816 int ret;
1817 VmdkGrainMarker *data = NULL;
1818 uLongf buf_len;
1819 QEMUIOVector local_qiov;
1820 int64_t write_offset;
1821 int64_t write_end_sector;
1822
1823 if (extent->compressed) {
1824 void *compressed_data;
1825
1826 /* Only whole clusters */
1827 if (offset_in_cluster ||
1828 n_bytes > (extent->cluster_sectors * SECTOR_SIZE) ||
1829 (n_bytes < (extent->cluster_sectors * SECTOR_SIZE) &&
1830 offset + n_bytes != extent->end_sector * SECTOR_SIZE))
1831 {
1832 ret = -EINVAL;
1833 goto out;
1834 }
1835
1836 if (!extent->has_marker) {
1837 ret = -EINVAL;
1838 goto out;
1839 }
1840 buf_len = (extent->cluster_sectors << 9) * 2;
1841 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1842
1843 compressed_data = g_malloc(n_bytes);
1844 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1845 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1846 g_free(compressed_data);
1847
1848 if (ret != Z_OK || buf_len == 0) {
1849 ret = -EINVAL;
1850 goto out;
1851 }
1852
1853 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1854 data->size = cpu_to_le32(buf_len);
1855
1856 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1857 qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1858
1859 BLKDBG_CO_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1860 } else {
1861 qemu_iovec_init(&local_qiov, qiov->niov);
1862 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1863
1864 BLKDBG_CO_EVENT(extent->file, BLKDBG_WRITE_AIO);
1865 }
1866
1867 write_offset = cluster_offset + offset_in_cluster;
1868 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1869 &local_qiov, 0);
1870
1871 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1872
1873 if (extent->compressed) {
1874 extent->next_cluster_sector = write_end_sector;
1875 } else {
1876 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1877 write_end_sector);
1878 }
1879
1880 if (ret < 0) {
1881 goto out;
1882 }
1883 ret = 0;
1884 out:
1885 g_free(data);
1886 if (!extent->compressed) {
1887 qemu_iovec_destroy(&local_qiov);
1888 }
1889 return ret;
1890 }
1891
1892 static int coroutine_fn GRAPH_RDLOCK
1893 vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1894 int64_t offset_in_cluster, QEMUIOVector *qiov, int bytes)
1895 {
1896 int ret;
1897 int cluster_bytes, buf_bytes;
1898 uint8_t *cluster_buf, *compressed_data;
1899 uint8_t *uncomp_buf;
1900 uint32_t data_len;
1901 VmdkGrainMarker *marker;
1902 uLongf buf_len;
1903
1904
1905 if (!extent->compressed) {
1906 BLKDBG_CO_EVENT(extent->file, BLKDBG_READ_AIO);
1907 ret = bdrv_co_preadv(extent->file,
1908 cluster_offset + offset_in_cluster, bytes,
1909 qiov, 0);
1910 if (ret < 0) {
1911 return ret;
1912 }
1913 return 0;
1914 }
1915 cluster_bytes = extent->cluster_sectors * 512;
1916 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1917 buf_bytes = cluster_bytes * 2;
1918 cluster_buf = g_malloc(buf_bytes);
1919 uncomp_buf = g_malloc(cluster_bytes);
1920 BLKDBG_CO_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1921 ret = bdrv_co_pread(extent->file, cluster_offset, buf_bytes, cluster_buf,
1922 0);
1923 if (ret < 0) {
1924 goto out;
1925 }
1926 compressed_data = cluster_buf;
1927 buf_len = cluster_bytes;
1928 data_len = cluster_bytes;
1929 if (extent->has_marker) {
1930 marker = (VmdkGrainMarker *)cluster_buf;
1931 compressed_data = marker->data;
1932 data_len = le32_to_cpu(marker->size);
1933 }
1934 if (!data_len || data_len > buf_bytes) {
1935 ret = -EINVAL;
1936 goto out;
1937 }
1938 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1939 if (ret != Z_OK) {
1940 ret = -EINVAL;
1941 goto out;
1942
1943 }
1944 if (offset_in_cluster < 0 ||
1945 offset_in_cluster + bytes > buf_len) {
1946 ret = -EINVAL;
1947 goto out;
1948 }
1949 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1950 ret = 0;
1951
1952 out:
1953 g_free(uncomp_buf);
1954 g_free(cluster_buf);
1955 return ret;
1956 }
1957
1958 static int coroutine_fn GRAPH_RDLOCK
1959 vmdk_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
1960 QEMUIOVector *qiov, BdrvRequestFlags flags)
1961 {
1962 BDRVVmdkState *s = bs->opaque;
1963 int ret;
1964 uint64_t n_bytes, offset_in_cluster;
1965 VmdkExtent *extent = NULL;
1966 QEMUIOVector local_qiov;
1967 uint64_t cluster_offset;
1968 uint64_t bytes_done = 0;
1969
1970 qemu_iovec_init(&local_qiov, qiov->niov);
1971 qemu_co_mutex_lock(&s->lock);
1972
1973 while (bytes > 0) {
1974 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1975 if (!extent) {
1976 ret = -EIO;
1977 goto fail;
1978 }
1979 ret = get_cluster_offset(bs, extent, NULL,
1980 offset, false, &cluster_offset, 0, 0);
1981 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1982
1983 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1984 - offset_in_cluster);
1985
1986 if (ret != VMDK_OK) {
1987 /* if not allocated, try to read from parent image, if exist */
1988 if (bs->backing && ret != VMDK_ZEROED) {
1989 if (!vmdk_is_cid_valid(bs)) {
1990 ret = -EINVAL;
1991 goto fail;
1992 }
1993
1994 qemu_iovec_reset(&local_qiov);
1995 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1996
1997 /* qcow2 emits this on bs->file instead of bs->backing */
1998 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1999 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
2000 &local_qiov, 0);
2001 if (ret < 0) {
2002 goto fail;
2003 }
2004 } else {
2005 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
2006 }
2007 } else {
2008 qemu_iovec_reset(&local_qiov);
2009 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
2010
2011 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
2012 &local_qiov, n_bytes);
2013 if (ret) {
2014 goto fail;
2015 }
2016 }
2017 bytes -= n_bytes;
2018 offset += n_bytes;
2019 bytes_done += n_bytes;
2020 }
2021
2022 ret = 0;
2023 fail:
2024 qemu_co_mutex_unlock(&s->lock);
2025 qemu_iovec_destroy(&local_qiov);
2026
2027 return ret;
2028 }
2029
2030 /**
2031 * vmdk_write:
2032 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
2033 * if possible, otherwise return -ENOTSUP.
2034 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
2035 * with each cluster. By dry run we can find if the zero write
2036 * is possible without modifying image data.
2037 *
2038 * Returns: error code with 0 for success.
2039 */
2040 static int coroutine_fn GRAPH_RDLOCK
2041 vmdk_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2042 QEMUIOVector *qiov, bool zeroed, bool zero_dry_run)
2043 {
2044 BDRVVmdkState *s = bs->opaque;
2045 VmdkExtent *extent = NULL;
2046 int ret;
2047 int64_t offset_in_cluster, n_bytes;
2048 uint64_t cluster_offset;
2049 uint64_t bytes_done = 0;
2050 VmdkMetaData m_data;
2051
2052 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
2053 error_report("Wrong offset: offset=0x%" PRIx64
2054 " total_sectors=0x%" PRIx64,
2055 offset, bs->total_sectors);
2056 return -EIO;
2057 }
2058
2059 while (bytes > 0) {
2060 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
2061 if (!extent) {
2062 return -EIO;
2063 }
2064 if (extent->sesparse) {
2065 return -ENOTSUP;
2066 }
2067 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
2068 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
2069 - offset_in_cluster);
2070
2071 ret = get_cluster_offset(bs, extent, &m_data, offset,
2072 !(extent->compressed || zeroed),
2073 &cluster_offset, offset_in_cluster,
2074 offset_in_cluster + n_bytes);
2075 if (extent->compressed) {
2076 if (ret == VMDK_OK) {
2077 /* Refuse write to allocated cluster for streamOptimized */
2078 error_report("Could not write to allocated cluster"
2079 " for streamOptimized");
2080 return -EIO;
2081 } else if (!zeroed) {
2082 /* allocate */
2083 ret = get_cluster_offset(bs, extent, &m_data, offset,
2084 true, &cluster_offset, 0, 0);
2085 }
2086 }
2087 if (ret == VMDK_ERROR) {
2088 return -EINVAL;
2089 }
2090 if (zeroed) {
2091 /* Do zeroed write, buf is ignored */
2092 if (extent->has_zero_grain &&
2093 offset_in_cluster == 0 &&
2094 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
2095 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
2096 if (!zero_dry_run && ret != VMDK_ZEROED) {
2097 /* update L2 tables */
2098 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
2099 != VMDK_OK) {
2100 return -EIO;
2101 }
2102 }
2103 } else {
2104 return -ENOTSUP;
2105 }
2106 } else {
2107 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
2108 qiov, bytes_done, n_bytes, offset);
2109 if (ret) {
2110 return ret;
2111 }
2112 if (m_data.new_allocation) {
2113 /* update L2 tables */
2114 if (vmdk_L2update(extent, &m_data,
2115 cluster_offset >> BDRV_SECTOR_BITS)
2116 != VMDK_OK) {
2117 return -EIO;
2118 }
2119 }
2120 }
2121 bytes -= n_bytes;
2122 offset += n_bytes;
2123 bytes_done += n_bytes;
2124
2125 /* update CID on the first write every time the virtual disk is
2126 * opened */
2127 if (!s->cid_updated) {
2128 ret = vmdk_write_cid(bs, g_random_int());
2129 if (ret < 0) {
2130 return ret;
2131 }
2132 s->cid_updated = true;
2133 }
2134 }
2135 return 0;
2136 }
2137
2138 static int coroutine_fn GRAPH_RDLOCK
2139 vmdk_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
2140 QEMUIOVector *qiov, BdrvRequestFlags flags)
2141 {
2142 int ret;
2143 BDRVVmdkState *s = bs->opaque;
2144 qemu_co_mutex_lock(&s->lock);
2145 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
2146 qemu_co_mutex_unlock(&s->lock);
2147 return ret;
2148 }
2149
2150 static int coroutine_fn GRAPH_RDLOCK
2151 vmdk_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes,
2152 QEMUIOVector *qiov)
2153 {
2154 if (bytes == 0) {
2155 /* The caller will write bytes 0 to signal EOF.
2156 * When receive it, we align EOF to a sector boundary. */
2157 BDRVVmdkState *s = bs->opaque;
2158 int i, ret;
2159 int64_t length;
2160
2161 for (i = 0; i < s->num_extents; i++) {
2162 length = bdrv_co_getlength(s->extents[i].file->bs);
2163 if (length < 0) {
2164 return length;
2165 }
2166 length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
2167 ret = bdrv_co_truncate(s->extents[i].file, length, false,
2168 PREALLOC_MODE_OFF, 0, NULL);
2169 if (ret < 0) {
2170 return ret;
2171 }
2172 }
2173 return 0;
2174 }
2175 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
2176 }
2177
2178 static int coroutine_fn GRAPH_RDLOCK
2179 vmdk_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
2180 BdrvRequestFlags flags)
2181 {
2182 int ret;
2183 BDRVVmdkState *s = bs->opaque;
2184
2185 qemu_co_mutex_lock(&s->lock);
2186 /* write zeroes could fail if sectors not aligned to cluster, test it with
2187 * dry_run == true before really updating image */
2188 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
2189 if (!ret) {
2190 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
2191 }
2192 qemu_co_mutex_unlock(&s->lock);
2193 return ret;
2194 }
2195
2196 static int coroutine_fn GRAPH_UNLOCKED
2197 vmdk_init_extent(BlockBackend *blk, int64_t filesize, bool flat, bool compress,
2198 bool zeroed_grain, Error **errp)
2199 {
2200 int ret, i;
2201 VMDK4Header header;
2202 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
2203 uint32_t *gd_buf = NULL;
2204 int gd_buf_size;
2205
2206 if (flat) {
2207 ret = blk_co_truncate(blk, filesize, false, PREALLOC_MODE_OFF, 0, errp);
2208 goto exit;
2209 }
2210 magic = cpu_to_be32(VMDK4_MAGIC);
2211 memset(&header, 0, sizeof(header));
2212 if (compress) {
2213 header.version = 3;
2214 } else if (zeroed_grain) {
2215 header.version = 2;
2216 } else {
2217 header.version = 1;
2218 }
2219 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
2220 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
2221 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
2222 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
2223 header.capacity = filesize / BDRV_SECTOR_SIZE;
2224 header.granularity = 128;
2225 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
2226
2227 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
2228 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
2229 BDRV_SECTOR_SIZE);
2230 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
2231 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
2232
2233 header.desc_offset = 1;
2234 header.desc_size = 20;
2235 header.rgd_offset = header.desc_offset + header.desc_size;
2236 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
2237 header.grain_offset =
2238 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
2239 header.granularity);
2240 /* swap endianness for all header fields */
2241 header.version = cpu_to_le32(header.version);
2242 header.flags = cpu_to_le32(header.flags);
2243 header.capacity = cpu_to_le64(header.capacity);
2244 header.granularity = cpu_to_le64(header.granularity);
2245 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
2246 header.desc_offset = cpu_to_le64(header.desc_offset);
2247 header.desc_size = cpu_to_le64(header.desc_size);
2248 header.rgd_offset = cpu_to_le64(header.rgd_offset);
2249 header.gd_offset = cpu_to_le64(header.gd_offset);
2250 header.grain_offset = cpu_to_le64(header.grain_offset);
2251 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
2252
2253 header.check_bytes[0] = 0xa;
2254 header.check_bytes[1] = 0x20;
2255 header.check_bytes[2] = 0xd;
2256 header.check_bytes[3] = 0xa;
2257
2258 /* write all the data */
2259 ret = blk_co_pwrite(blk, 0, sizeof(magic), &magic, 0);
2260 if (ret < 0) {
2261 error_setg(errp, QERR_IO_ERROR);
2262 goto exit;
2263 }
2264 ret = blk_co_pwrite(blk, sizeof(magic), sizeof(header), &header, 0);
2265 if (ret < 0) {
2266 error_setg(errp, QERR_IO_ERROR);
2267 goto exit;
2268 }
2269
2270 ret = blk_co_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false,
2271 PREALLOC_MODE_OFF, 0, errp);
2272 if (ret < 0) {
2273 goto exit;
2274 }
2275
2276 /* write grain directory */
2277 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
2278 gd_buf = g_malloc0(gd_buf_size);
2279 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
2280 i < gt_count; i++, tmp += gt_size) {
2281 gd_buf[i] = cpu_to_le32(tmp);
2282 }
2283 ret = blk_co_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
2284 gd_buf_size, gd_buf, 0);
2285 if (ret < 0) {
2286 error_setg(errp, QERR_IO_ERROR);
2287 goto exit;
2288 }
2289
2290 /* write backup grain directory */
2291 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
2292 i < gt_count; i++, tmp += gt_size) {
2293 gd_buf[i] = cpu_to_le32(tmp);
2294 }
2295 ret = blk_co_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
2296 gd_buf_size, gd_buf, 0);
2297 if (ret < 0) {
2298 error_setg(errp, QERR_IO_ERROR);
2299 }
2300
2301 ret = 0;
2302 exit:
2303 g_free(gd_buf);
2304 return ret;
2305 }
2306
2307 static int coroutine_fn GRAPH_UNLOCKED
2308 vmdk_create_extent(const char *filename, int64_t filesize, bool flat,
2309 bool compress, bool zeroed_grain, BlockBackend **pbb,
2310 QemuOpts *opts, Error **errp)
2311 {
2312 int ret;
2313 BlockBackend *blk = NULL;
2314
2315 ret = bdrv_co_create_file(filename, opts, errp);
2316 if (ret < 0) {
2317 goto exit;
2318 }
2319
2320 blk = blk_co_new_open(filename, NULL, NULL,
2321 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2322 errp);
2323 if (blk == NULL) {
2324 ret = -EIO;
2325 goto exit;
2326 }
2327
2328 blk_set_allow_write_beyond_eof(blk, true);
2329
2330 ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
2331 exit:
2332 if (blk) {
2333 if (pbb) {
2334 *pbb = blk;
2335 } else {
2336 blk_co_unref(blk);
2337 blk = NULL;
2338 }
2339 }
2340 return ret;
2341 }
2342
2343 static int filename_decompose(const char *filename, char *path, char *prefix,
2344 char *postfix, size_t buf_len, Error **errp)
2345 {
2346 const char *p, *q;
2347
2348 if (filename == NULL || !strlen(filename)) {
2349 error_setg(errp, "No filename provided");
2350 return VMDK_ERROR;
2351 }
2352 p = strrchr(filename, '/');
2353 if (p == NULL) {
2354 p = strrchr(filename, '\\');
2355 }
2356 if (p == NULL) {
2357 p = strrchr(filename, ':');
2358 }
2359 if (p != NULL) {
2360 p++;
2361 if (p - filename >= buf_len) {
2362 return VMDK_ERROR;
2363 }
2364 pstrcpy(path, p - filename + 1, filename);
2365 } else {
2366 p = filename;
2367 path[0] = '\0';
2368 }
2369 q = strrchr(p, '.');
2370 if (q == NULL) {
2371 pstrcpy(prefix, buf_len, p);
2372 postfix[0] = '\0';
2373 } else {
2374 if (q - p >= buf_len) {
2375 return VMDK_ERROR;
2376 }
2377 pstrcpy(prefix, q - p + 1, p);
2378 pstrcpy(postfix, buf_len, q);
2379 }
2380 return VMDK_OK;
2381 }
2382
2383 /*
2384 * idx == 0: get or create the descriptor file (also the image file if in a
2385 * non-split format.
2386 * idx >= 1: get the n-th extent if in a split subformat
2387 */
2388 typedef BlockBackend * coroutine_fn GRAPH_UNLOCKED_PTR
2389 (*vmdk_create_extent_fn)(int64_t size, int idx, bool flat, bool split,
2390 bool compress, bool zeroed_grain, void *opaque,
2391 Error **errp);
2392
2393 static void vmdk_desc_add_extent(GString *desc,
2394 const char *extent_line_fmt,
2395 int64_t size, const char *filename)
2396 {
2397 char *basename = g_path_get_basename(filename);
2398
2399 g_string_append_printf(desc, extent_line_fmt,
2400 DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
2401 g_free(basename);
2402 }
2403
2404 static int coroutine_fn GRAPH_UNLOCKED
2405 vmdk_co_do_create(int64_t size,
2406 BlockdevVmdkSubformat subformat,
2407 BlockdevVmdkAdapterType adapter_type,
2408 const char *backing_file,
2409 const char *hw_version,
2410 const char *toolsversion,
2411 bool compat6,
2412 bool zeroed_grain,
2413 vmdk_create_extent_fn extent_fn,
2414 void *opaque,
2415 Error **errp)
2416 {
2417 int extent_idx;
2418 BlockBackend *blk = NULL;
2419 BlockBackend *extent_blk;
2420 Error *local_err = NULL;
2421 char *desc = NULL;
2422 int ret = 0;
2423 bool flat, split, compress;
2424 GString *ext_desc_lines;
2425 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
2426 int64_t extent_size;
2427 int64_t created_size = 0;
2428 const char *extent_line_fmt;
2429 char *parent_desc_line = g_malloc0(BUF_SIZE);
2430 uint32_t parent_cid = 0xffffffff;
2431 uint32_t number_heads = 16;
2432 uint32_t desc_offset = 0, desc_len;
2433 const char desc_template[] =
2434 "# Disk DescriptorFile\n"
2435 "version=1\n"
2436 "CID=%" PRIx32 "\n"
2437 "parentCID=%" PRIx32 "\n"
2438 "createType=\"%s\"\n"
2439 "%s"
2440 "\n"
2441 "# Extent description\n"
2442 "%s"
2443 "\n"
2444 "# The Disk Data Base\n"
2445 "#DDB\n"
2446 "\n"
2447 "ddb.virtualHWVersion = \"%s\"\n"
2448 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2449 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2450 "ddb.geometry.sectors = \"63\"\n"
2451 "ddb.adapterType = \"%s\"\n"
2452 "ddb.toolsVersion = \"%s\"\n";
2453
2454 ext_desc_lines = g_string_new(NULL);
2455
2456 /* Read out options */
2457 if (compat6) {
2458 if (hw_version) {
2459 error_setg(errp,
2460 "compat6 cannot be enabled with hwversion set");
2461 ret = -EINVAL;
2462 goto exit;
2463 }
2464 hw_version = "6";
2465 }
2466 if (!hw_version) {
2467 hw_version = "4";
2468 }
2469 if (!toolsversion) {
2470 toolsversion = "2147483647";
2471 }
2472
2473 if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2474 /* that's the number of heads with which vmware operates when
2475 creating, exporting, etc. vmdk files with a non-ide adapter type */
2476 number_heads = 255;
2477 }
2478 split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2479 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2480 flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2481 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2482 compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2483
2484 if (flat) {
2485 extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2486 } else {
2487 extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2488 }
2489 if (flat && backing_file) {
2490 error_setg(errp, "Flat image can't have backing file");
2491 ret = -ENOTSUP;
2492 goto exit;
2493 }
2494 if (flat && zeroed_grain) {
2495 error_setg(errp, "Flat image can't enable zeroed grain");
2496 ret = -ENOTSUP;
2497 goto exit;
2498 }
2499
2500 /* Create extents */
2501 if (split) {
2502 extent_size = split_size;
2503 } else {
2504 extent_size = size;
2505 }
2506 if (!split && !flat) {
2507 created_size = extent_size;
2508 } else {
2509 created_size = 0;
2510 }
2511 /* Get the descriptor file BDS */
2512 blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2513 opaque, errp);
2514 if (!blk) {
2515 ret = -EIO;
2516 goto exit;
2517 }
2518 if (!split && !flat) {
2519 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2520 blk_bs(blk)->filename);
2521 }
2522
2523 if (backing_file) {
2524 BlockBackend *backing;
2525 char *full_backing =
2526 bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2527 backing_file,
2528 &local_err);
2529 if (local_err) {
2530 error_propagate(errp, local_err);
2531 ret = -ENOENT;
2532 goto exit;
2533 }
2534 assert(full_backing);
2535
2536 backing = blk_co_new_open(full_backing, NULL, NULL,
2537 BDRV_O_NO_BACKING, errp);
2538 g_free(full_backing);
2539 if (backing == NULL) {
2540 ret = -EIO;
2541 goto exit;
2542 }
2543 if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2544 error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2545 blk_bs(backing)->drv->format_name);
2546 blk_co_unref(backing);
2547 ret = -EINVAL;
2548 goto exit;
2549 }
2550 ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2551 blk_co_unref(backing);
2552 if (ret) {
2553 error_setg(errp, "Failed to read parent CID");
2554 goto exit;
2555 }
2556 snprintf(parent_desc_line, BUF_SIZE,
2557 "parentFileNameHint=\"%s\"", backing_file);
2558 }
2559 extent_idx = 1;
2560 while (created_size < size) {
2561 int64_t cur_size = MIN(size - created_size, extent_size);
2562 extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2563 zeroed_grain, opaque, errp);
2564 if (!extent_blk) {
2565 ret = -EINVAL;
2566 goto exit;
2567 }
2568 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2569 blk_bs(extent_blk)->filename);
2570 created_size += cur_size;
2571 extent_idx++;
2572 blk_co_unref(extent_blk);
2573 }
2574
2575 /* Check whether we got excess extents */
2576 extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2577 opaque, NULL);
2578 if (extent_blk) {
2579 blk_co_unref(extent_blk);
2580 error_setg(errp, "List of extents contains unused extents");
2581 ret = -EINVAL;
2582 goto exit;
2583 }
2584
2585 /* generate descriptor file */
2586 desc = g_strdup_printf(desc_template,
2587 g_random_int(),
2588 parent_cid,
2589 BlockdevVmdkSubformat_str(subformat),
2590 parent_desc_line,
2591 ext_desc_lines->str,
2592 hw_version,
2593 size /
2594 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2595 number_heads,
2596 BlockdevVmdkAdapterType_str(adapter_type),
2597 toolsversion);
2598 desc_len = strlen(desc);
2599 /* the descriptor offset = 0x200 */
2600 if (!split && !flat) {
2601 desc_offset = 0x200;
2602 }
2603
2604 ret = blk_co_pwrite(blk, desc_offset, desc_len, desc, 0);
2605 if (ret < 0) {
2606 error_setg_errno(errp, -ret, "Could not write description");
2607 goto exit;
2608 }
2609 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2610 * for description file */
2611 if (desc_offset == 0) {
2612 ret = blk_co_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, 0, errp);
2613 if (ret < 0) {
2614 goto exit;
2615 }
2616 }
2617 ret = 0;
2618 exit:
2619 if (blk) {
2620 blk_co_unref(blk);
2621 }
2622 g_free(desc);
2623 g_free(parent_desc_line);
2624 g_string_free(ext_desc_lines, true);
2625 return ret;
2626 }
2627
2628 typedef struct {
2629 char *path;
2630 char *prefix;
2631 char *postfix;
2632 QemuOpts *opts;
2633 } VMDKCreateOptsData;
2634
2635 static BlockBackend * coroutine_fn GRAPH_UNLOCKED
2636 vmdk_co_create_opts_cb(int64_t size, int idx, bool flat, bool split,
2637 bool compress, bool zeroed_grain, void *opaque,
2638 Error **errp)
2639 {
2640 BlockBackend *blk = NULL;
2641 BlockDriverState *bs = NULL;
2642 VMDKCreateOptsData *data = opaque;
2643 char *ext_filename = NULL;
2644 char *rel_filename = NULL;
2645
2646 /* We're done, don't create excess extents. */
2647 if (size == -1) {
2648 assert(errp == NULL);
2649 return NULL;
2650 }
2651
2652 if (idx == 0) {
2653 rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2654 } else if (split) {
2655 rel_filename = g_strdup_printf("%s-%c%03d%s",
2656 data->prefix,
2657 flat ? 'f' : 's', idx, data->postfix);
2658 } else {
2659 assert(idx == 1);
2660 rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2661 }
2662
2663 ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2664 g_free(rel_filename);
2665
2666 if (vmdk_create_extent(ext_filename, size,
2667 flat, compress, zeroed_grain, &blk, data->opts,
2668 errp)) {
2669 goto exit;
2670 }
2671 bdrv_co_unref(bs);
2672 exit:
2673 g_free(ext_filename);
2674 return blk;
2675 }
2676
2677 static int coroutine_fn GRAPH_UNLOCKED
2678 vmdk_co_create_opts(BlockDriver *drv, const char *filename,
2679 QemuOpts *opts, Error **errp)
2680 {
2681 Error *local_err = NULL;
2682 char *desc = NULL;
2683 int64_t total_size = 0;
2684 char *adapter_type = NULL;
2685 BlockdevVmdkAdapterType adapter_type_enum;
2686 char *backing_file = NULL;
2687 char *hw_version = NULL;
2688 char *toolsversion = NULL;
2689 char *fmt = NULL;
2690 BlockdevVmdkSubformat subformat;
2691 int ret = 0;
2692 char *path = g_malloc0(PATH_MAX);
2693 char *prefix = g_malloc0(PATH_MAX);
2694 char *postfix = g_malloc0(PATH_MAX);
2695 char *desc_line = g_malloc0(BUF_SIZE);
2696 char *ext_filename = g_malloc0(PATH_MAX);
2697 char *desc_filename = g_malloc0(PATH_MAX);
2698 char *parent_desc_line = g_malloc0(BUF_SIZE);
2699 bool zeroed_grain;
2700 bool compat6;
2701 VMDKCreateOptsData data;
2702 char *backing_fmt = NULL;
2703
2704 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2705 if (backing_fmt && strcmp(backing_fmt, "vmdk") != 0) {
2706 error_setg(errp, "backing_file must be a vmdk image");
2707 ret = -EINVAL;
2708 goto exit;
2709 }
2710
2711 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
2712 ret = -EINVAL;
2713 goto exit;
2714 }
2715 /* Read out options */
2716 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2717 BDRV_SECTOR_SIZE);
2718 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
2719 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2720 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
2721 toolsversion = qemu_opt_get_del(opts, BLOCK_OPT_TOOLSVERSION);
2722 compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false);
2723 if (strcmp(hw_version, "undefined") == 0) {
2724 g_free(hw_version);
2725 hw_version = NULL;
2726 }
2727 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
2728 zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false);
2729
2730 if (adapter_type) {
2731 adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup,
2732 adapter_type,
2733 BLOCKDEV_VMDK_ADAPTER_TYPE_IDE,
2734 &local_err);
2735 if (local_err) {
2736 error_propagate(errp, local_err);
2737 ret = -EINVAL;
2738 goto exit;
2739 }
2740 } else {
2741 adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE;
2742 }
2743
2744 if (!fmt) {
2745 /* Default format to monolithicSparse */
2746 subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE;
2747 } else {
2748 subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup,
2749 fmt,
2750 BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE,
2751 &local_err);
2752 if (local_err) {
2753 error_propagate(errp, local_err);
2754 ret = -EINVAL;
2755 goto exit;
2756 }
2757 }
2758 data = (VMDKCreateOptsData){
2759 .prefix = prefix,
2760 .postfix = postfix,
2761 .path = path,
2762 .opts = opts,
2763 };
2764 ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum,
2765 backing_file, hw_version, toolsversion, compat6,
2766 zeroed_grain, vmdk_co_create_opts_cb, &data, errp);
2767
2768 exit:
2769 g_free(backing_fmt);
2770 g_free(adapter_type);
2771 g_free(backing_file);
2772 g_free(hw_version);
2773 g_free(toolsversion);
2774 g_free(fmt);
2775 g_free(desc);
2776 g_free(path);
2777 g_free(prefix);
2778 g_free(postfix);
2779 g_free(desc_line);
2780 g_free(ext_filename);
2781 g_free(desc_filename);
2782 g_free(parent_desc_line);
2783 return ret;
2784 }
2785
2786 static BlockBackend * coroutine_fn GRAPH_UNLOCKED
2787 vmdk_co_create_cb(int64_t size, int idx, bool flat, bool split, bool compress,
2788 bool zeroed_grain, void *opaque, Error **errp)
2789 {
2790 int ret;
2791 BlockDriverState *bs;
2792 BlockBackend *blk;
2793 BlockdevCreateOptionsVmdk *opts = opaque;
2794
2795 if (idx == 0) {
2796 bs = bdrv_co_open_blockdev_ref(opts->file, errp);
2797 } else {
2798 int i;
2799 BlockdevRefList *list = opts->extents;
2800 for (i = 1; i < idx; i++) {
2801 if (!list || !list->next) {
2802 error_setg(errp, "Extent [%d] not specified", i);
2803 return NULL;
2804 }
2805 list = list->next;
2806 }
2807 if (!list) {
2808 error_setg(errp, "Extent [%d] not specified", idx - 1);
2809 return NULL;
2810 }
2811 bs = bdrv_co_open_blockdev_ref(list->value, errp);
2812 }
2813 if (!bs) {
2814 return NULL;
2815 }
2816 blk = blk_co_new_with_bs(bs,
2817 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
2818 BLK_PERM_RESIZE,
2819 BLK_PERM_ALL,
2820 errp);
2821 if (!blk) {
2822 return NULL;
2823 }
2824 blk_set_allow_write_beyond_eof(blk, true);
2825 bdrv_co_unref(bs);
2826
2827 if (size != -1) {
2828 ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2829 if (ret) {
2830 blk_co_unref(blk);
2831 blk = NULL;
2832 }
2833 }
2834 return blk;
2835 }
2836
2837 static int coroutine_fn GRAPH_UNLOCKED
2838 vmdk_co_create(BlockdevCreateOptions *create_options, Error **errp)
2839 {
2840 BlockdevCreateOptionsVmdk *opts;
2841
2842 opts = &create_options->u.vmdk;
2843
2844 /* Validate options */
2845 if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2846 error_setg(errp, "Image size must be a multiple of 512 bytes");
2847 return -EINVAL;
2848 }
2849
2850 return vmdk_co_do_create(opts->size,
2851 opts->subformat,
2852 opts->adapter_type,
2853 opts->backing_file,
2854 opts->hwversion,
2855 opts->toolsversion,
2856 false,
2857 opts->zeroed_grain,
2858 vmdk_co_create_cb,
2859 opts, errp);
2860 }
2861
2862 static void vmdk_close(BlockDriverState *bs)
2863 {
2864 BDRVVmdkState *s = bs->opaque;
2865
2866 vmdk_free_extents(bs);
2867 g_free(s->create_type);
2868
2869 migrate_del_blocker(&s->migration_blocker);
2870 }
2871
2872 static int64_t coroutine_fn GRAPH_RDLOCK
2873 vmdk_co_get_allocated_file_size(BlockDriverState *bs)
2874 {
2875 int i;
2876 int64_t ret = 0;
2877 int64_t r;
2878 BDRVVmdkState *s = bs->opaque;
2879
2880 ret = bdrv_co_get_allocated_file_size(bs->file->bs);
2881 if (ret < 0) {
2882 return ret;
2883 }
2884 for (i = 0; i < s->num_extents; i++) {
2885 if (s->extents[i].file == bs->file) {
2886 continue;
2887 }
2888 r = bdrv_co_get_allocated_file_size(s->extents[i].file->bs);
2889 if (r < 0) {
2890 return r;
2891 }
2892 ret += r;
2893 }
2894 return ret;
2895 }
2896
2897 static int vmdk_has_zero_init(BlockDriverState *bs)
2898 {
2899 int i;
2900 BDRVVmdkState *s = bs->opaque;
2901
2902 /* If has a flat extent and its underlying storage doesn't have zero init,
2903 * return 0. */
2904 for (i = 0; i < s->num_extents; i++) {
2905 if (s->extents[i].flat) {
2906 if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2907 return 0;
2908 }
2909 }
2910 }
2911 return 1;
2912 }
2913
2914 static VmdkExtentInfo * GRAPH_RDLOCK vmdk_get_extent_info(VmdkExtent *extent)
2915 {
2916 VmdkExtentInfo *info = g_new0(VmdkExtentInfo, 1);
2917
2918 bdrv_refresh_filename(extent->file->bs);
2919 *info = (VmdkExtentInfo){
2920 .filename = g_strdup(extent->file->bs->filename),
2921 .format = g_strdup(extent->type),
2922 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2923 .compressed = extent->compressed,
2924 .has_compressed = extent->compressed,
2925 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2926 .has_cluster_size = !extent->flat,
2927 };
2928
2929 return info;
2930 }
2931
2932 static int coroutine_fn GRAPH_RDLOCK
2933 vmdk_co_check(BlockDriverState *bs, BdrvCheckResult *result, BdrvCheckMode fix)
2934 {
2935 BDRVVmdkState *s = bs->opaque;
2936 VmdkExtent *extent = NULL;
2937 int64_t sector_num = 0;
2938 int64_t total_sectors = bdrv_co_nb_sectors(bs);
2939 int ret;
2940 uint64_t cluster_offset;
2941
2942 if (fix) {
2943 return -ENOTSUP;
2944 }
2945
2946 for (;;) {
2947 if (sector_num >= total_sectors) {
2948 return 0;
2949 }
2950 extent = find_extent(s, sector_num, extent);
2951 if (!extent) {
2952 fprintf(stderr,
2953 "ERROR: could not find extent for sector %" PRId64 "\n",
2954 sector_num);
2955 ret = -EINVAL;
2956 break;
2957 }
2958 ret = get_cluster_offset(bs, extent, NULL,
2959 sector_num << BDRV_SECTOR_BITS,
2960 false, &cluster_offset, 0, 0);
2961 if (ret == VMDK_ERROR) {
2962 fprintf(stderr,
2963 "ERROR: could not get cluster_offset for sector %"
2964 PRId64 "\n", sector_num);
2965 break;
2966 }
2967 if (ret == VMDK_OK) {
2968 int64_t extent_len = bdrv_co_getlength(extent->file->bs);
2969 if (extent_len < 0) {
2970 fprintf(stderr,
2971 "ERROR: could not get extent file length for sector %"
2972 PRId64 "\n", sector_num);
2973 ret = extent_len;
2974 break;
2975 }
2976 if (cluster_offset >= extent_len) {
2977 fprintf(stderr,
2978 "ERROR: cluster offset for sector %"
2979 PRId64 " points after EOF\n", sector_num);
2980 ret = -EINVAL;
2981 break;
2982 }
2983 }
2984 sector_num += extent->cluster_sectors;
2985 }
2986
2987 result->corruptions++;
2988 return ret;
2989 }
2990
2991 static ImageInfoSpecific * GRAPH_RDLOCK
2992 vmdk_get_specific_info(BlockDriverState *bs, Error **errp)
2993 {
2994 int i;
2995 BDRVVmdkState *s = bs->opaque;
2996 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2997 VmdkExtentInfoList **tail;
2998
2999 *spec_info = (ImageInfoSpecific){
3000 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
3001 .u = {
3002 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
3003 },
3004 };
3005
3006 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
3007 .create_type = g_strdup(s->create_type),
3008 .cid = s->cid,
3009 .parent_cid = s->parent_cid,
3010 };
3011
3012 tail = &spec_info->u.vmdk.data->extents;
3013 for (i = 0; i < s->num_extents; i++) {
3014 QAPI_LIST_APPEND(tail, vmdk_get_extent_info(&s->extents[i]));
3015 }
3016
3017 return spec_info;
3018 }
3019
3020 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
3021 {
3022 return a->flat == b->flat &&
3023 a->compressed == b->compressed &&
3024 (a->flat || a->cluster_sectors == b->cluster_sectors);
3025 }
3026
3027 static int coroutine_fn
3028 vmdk_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3029 {
3030 int i;
3031 BDRVVmdkState *s = bs->opaque;
3032 assert(s->num_extents);
3033
3034 /* See if we have multiple extents but they have different cases */
3035 for (i = 1; i < s->num_extents; i++) {
3036 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
3037 return -ENOTSUP;
3038 }
3039 }
3040 bdi->needs_compressed_writes = s->extents[0].compressed;
3041 if (!s->extents[0].flat) {
3042 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
3043 }
3044 return 0;
3045 }
3046
3047 static void vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
3048 bool backing_overridden)
3049 {
3050 /* No children but file and backing can be explicitly specified (TODO) */
3051 qdict_put(target, "file",
3052 qobject_ref(bs->file->bs->full_open_options));
3053
3054 if (backing_overridden) {
3055 if (bs->backing) {
3056 qdict_put(target, "backing",
3057 qobject_ref(bs->backing->bs->full_open_options));
3058 } else {
3059 qdict_put_null(target, "backing");
3060 }
3061 }
3062 }
3063
3064 static QemuOptsList vmdk_create_opts = {
3065 .name = "vmdk-create-opts",
3066 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
3067 .desc = {
3068 {
3069 .name = BLOCK_OPT_SIZE,
3070 .type = QEMU_OPT_SIZE,
3071 .help = "Virtual disk size"
3072 },
3073 {
3074 .name = BLOCK_OPT_ADAPTER_TYPE,
3075 .type = QEMU_OPT_STRING,
3076 .help = "Virtual adapter type, can be one of "
3077 "ide (default), lsilogic, buslogic or legacyESX"
3078 },
3079 {
3080 .name = BLOCK_OPT_BACKING_FILE,
3081 .type = QEMU_OPT_STRING,
3082 .help = "File name of a base image"
3083 },
3084 {
3085 .name = BLOCK_OPT_BACKING_FMT,
3086 .type = QEMU_OPT_STRING,
3087 .help = "Must be 'vmdk' if present",
3088 },
3089 {
3090 .name = BLOCK_OPT_COMPAT6,
3091 .type = QEMU_OPT_BOOL,
3092 .help = "VMDK version 6 image",
3093 .def_value_str = "off"
3094 },
3095 {
3096 .name = BLOCK_OPT_HWVERSION,
3097 .type = QEMU_OPT_STRING,
3098 .help = "VMDK hardware version",
3099 .def_value_str = "undefined"
3100 },
3101 {
3102 .name = BLOCK_OPT_TOOLSVERSION,
3103 .type = QEMU_OPT_STRING,
3104 .help = "VMware guest tools version",
3105 },
3106 {
3107 .name = BLOCK_OPT_SUBFMT,
3108 .type = QEMU_OPT_STRING,
3109 .help =
3110 "VMDK flat extent format, can be one of "
3111 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
3112 },
3113 {
3114 .name = BLOCK_OPT_ZEROED_GRAIN,
3115 .type = QEMU_OPT_BOOL,
3116 .help = "Enable efficient zero writes "
3117 "using the zeroed-grain GTE feature"
3118 },
3119 { /* end of list */ }
3120 }
3121 };
3122
3123 static BlockDriver bdrv_vmdk = {
3124 .format_name = "vmdk",
3125 .instance_size = sizeof(BDRVVmdkState),
3126 .bdrv_probe = vmdk_probe,
3127 .bdrv_open = vmdk_open,
3128 .bdrv_co_check = vmdk_co_check,
3129 .bdrv_reopen_prepare = vmdk_reopen_prepare,
3130 .bdrv_reopen_commit = vmdk_reopen_commit,
3131 .bdrv_reopen_abort = vmdk_reopen_abort,
3132 .bdrv_child_perm = bdrv_default_perms,
3133 .bdrv_co_preadv = vmdk_co_preadv,
3134 .bdrv_co_pwritev = vmdk_co_pwritev,
3135 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
3136 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
3137 .bdrv_close = vmdk_close,
3138 .bdrv_co_create_opts = vmdk_co_create_opts,
3139 .bdrv_co_create = vmdk_co_create,
3140 .bdrv_co_block_status = vmdk_co_block_status,
3141 .bdrv_co_get_allocated_file_size = vmdk_co_get_allocated_file_size,
3142 .bdrv_has_zero_init = vmdk_has_zero_init,
3143 .bdrv_get_specific_info = vmdk_get_specific_info,
3144 .bdrv_refresh_limits = vmdk_refresh_limits,
3145 .bdrv_co_get_info = vmdk_co_get_info,
3146 .bdrv_gather_child_options = vmdk_gather_child_options,
3147
3148 .is_format = true,
3149 .supports_backing = true,
3150 .create_opts = &vmdk_create_opts,
3151 };
3152
3153 static void bdrv_vmdk_init(void)
3154 {
3155 bdrv_register(&bdrv_vmdk);
3156 }
3157
3158 block_init(bdrv_vmdk_init);
Random patches