?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'qemu-macppc-20230206' of https://github.com/mcayland/qemu into staging
[qemu.git] / block / vmdk.c
blob5b0eae877ea44530d8c50b960ef69f8d76ac8bc7
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 for (i = 0; i < s->num_extents; i++) {
276 e = &s->extents[i];
277 g_free(e->l1_table);
278 g_free(e->l2_cache);
279 g_free(e->l1_backup_table);
280 g_free(e->type);
281 if (e->file != bs->file) {
282 bdrv_unref_child(bs, e->file);
283 }
284 }
285 g_free(s->extents);
286 }
287
288 static void vmdk_free_last_extent(BlockDriverState *bs)
289 {
290 BDRVVmdkState *s = bs->opaque;
291
292 if (s->num_extents == 0) {
293 return;
294 }
295 s->num_extents--;
296 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
297 }
298
299 /* Return -ve errno, or 0 on success and write CID into *pcid. */
300 static int vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
301 {
302 char *desc;
303 uint32_t cid;
304 const char *p_name, *cid_str;
305 size_t cid_str_size;
306 BDRVVmdkState *s = bs->opaque;
307 int ret;
308
309 desc = g_malloc0(DESC_SIZE);
310 ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
311 if (ret < 0) {
312 goto out;
313 }
314
315 if (parent) {
316 cid_str = "parentCID";
317 cid_str_size = sizeof("parentCID");
318 } else {
319 cid_str = "CID";
320 cid_str_size = sizeof("CID");
321 }
322
323 desc[DESC_SIZE - 1] = '\0';
324 p_name = strstr(desc, cid_str);
325 if (p_name == NULL) {
326 ret = -EINVAL;
327 goto out;
328 }
329 p_name += cid_str_size;
330 if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
331 ret = -EINVAL;
332 goto out;
333 }
334 *pcid = cid;
335 ret = 0;
336
337 out:
338 g_free(desc);
339 return ret;
340 }
341
342 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
343 {
344 char *desc, *tmp_desc;
345 char *p_name, *tmp_str;
346 BDRVVmdkState *s = bs->opaque;
347 int ret = 0;
348
349 desc = g_malloc0(DESC_SIZE);
350 tmp_desc = g_malloc0(DESC_SIZE);
351 ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
352 if (ret < 0) {
353 goto out;
354 }
355
356 desc[DESC_SIZE - 1] = '\0';
357 tmp_str = strstr(desc, "parentCID");
358 if (tmp_str == NULL) {
359 ret = -EINVAL;
360 goto out;
361 }
362
363 pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
364 p_name = strstr(desc, "CID");
365 if (p_name != NULL) {
366 p_name += sizeof("CID");
367 snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
368 pstrcat(desc, DESC_SIZE, tmp_desc);
369 }
370
371 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
372
373 out:
374 g_free(desc);
375 g_free(tmp_desc);
376 return ret;
377 }
378
379 static int vmdk_is_cid_valid(BlockDriverState *bs)
380 {
381 BDRVVmdkState *s = bs->opaque;
382 uint32_t cur_pcid;
383
384 if (!s->cid_checked && bs->backing) {
385 BlockDriverState *p_bs = bs->backing->bs;
386
387 if (strcmp(p_bs->drv->format_name, "vmdk")) {
388 /* Backing file is not in vmdk format, so it does not have
389 * a CID, which makes the overlay's parent CID invalid */
390 return 0;
391 }
392
393 if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
394 /* read failure: report as not valid */
395 return 0;
396 }
397 if (s->parent_cid != cur_pcid) {
398 /* CID not valid */
399 return 0;
400 }
401 }
402 s->cid_checked = true;
403 /* CID valid */
404 return 1;
405 }
406
407 static int vmdk_reopen_prepare(BDRVReopenState *state,
408 BlockReopenQueue *queue, Error **errp)
409 {
410 BDRVVmdkState *s;
411 BDRVVmdkReopenState *rs;
412 int i;
413
414 assert(state != NULL);
415 assert(state->bs != NULL);
416 assert(state->opaque == NULL);
417
418 s = state->bs->opaque;
419
420 rs = g_new0(BDRVVmdkReopenState, 1);
421 state->opaque = rs;
422
423 /*
424 * Check whether there are any extents stored in bs->file; if bs->file
425 * changes, we will need to update their .file pointers to follow suit
426 */
427 rs->extents_using_bs_file = g_new(bool, s->num_extents);
428 for (i = 0; i < s->num_extents; i++) {
429 rs->extents_using_bs_file[i] = s->extents[i].file == state->bs->file;
430 }
431
432 return 0;
433 }
434
435 static void vmdk_reopen_clean(BDRVReopenState *state)
436 {
437 BDRVVmdkReopenState *rs = state->opaque;
438
439 g_free(rs->extents_using_bs_file);
440 g_free(rs);
441 state->opaque = NULL;
442 }
443
444 static void vmdk_reopen_commit(BDRVReopenState *state)
445 {
446 BDRVVmdkState *s = state->bs->opaque;
447 BDRVVmdkReopenState *rs = state->opaque;
448 int i;
449
450 for (i = 0; i < s->num_extents; i++) {
451 if (rs->extents_using_bs_file[i]) {
452 s->extents[i].file = state->bs->file;
453 }
454 }
455
456 vmdk_reopen_clean(state);
457 }
458
459 static void vmdk_reopen_abort(BDRVReopenState *state)
460 {
461 vmdk_reopen_clean(state);
462 }
463
464 static int vmdk_parent_open(BlockDriverState *bs)
465 {
466 char *p_name;
467 char *desc;
468 BDRVVmdkState *s = bs->opaque;
469 int ret;
470
471 desc = g_malloc0(DESC_SIZE + 1);
472 ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
473 if (ret < 0) {
474 goto out;
475 }
476
477 p_name = strstr(desc, "parentFileNameHint");
478 if (p_name != NULL) {
479 char *end_name;
480
481 p_name += sizeof("parentFileNameHint") + 1;
482 end_name = strchr(p_name, '\"');
483 if (end_name == NULL) {
484 ret = -EINVAL;
485 goto out;
486 }
487 if ((end_name - p_name) > sizeof(bs->auto_backing_file) - 1) {
488 ret = -EINVAL;
489 goto out;
490 }
491
492 pstrcpy(bs->auto_backing_file, end_name - p_name + 1, p_name);
493 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
494 bs->auto_backing_file);
495 pstrcpy(bs->backing_format, sizeof(bs->backing_format),
496 "vmdk");
497 }
498
499 out:
500 g_free(desc);
501 return ret;
502 }
503
504 /* Create and append extent to the extent array. Return the added VmdkExtent
505 * address. return NULL if allocation failed. */
506 static int vmdk_add_extent(BlockDriverState *bs,
507 BdrvChild *file, bool flat, int64_t sectors,
508 int64_t l1_offset, int64_t l1_backup_offset,
509 uint32_t l1_size,
510 int l2_size, uint64_t cluster_sectors,
511 VmdkExtent **new_extent,
512 Error **errp)
513 {
514 VmdkExtent *extent;
515 BDRVVmdkState *s = bs->opaque;
516 int64_t nb_sectors;
517
518 if (cluster_sectors > 0x200000) {
519 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
520 error_setg(errp, "Invalid granularity, image may be corrupt");
521 return -EFBIG;
522 }
523 if (l1_size > 32 * 1024 * 1024) {
524 /*
525 * Although with big capacity and small l1_entry_sectors, we can get a
526 * big l1_size, we don't want unbounded value to allocate the table.
527 * Limit it to 32M, which is enough to store:
528 * 8TB - for both VMDK3 & VMDK4 with
529 * minimal cluster size: 512B
530 * minimal L2 table size: 512 entries
531 * 8 TB is still more than the maximal value supported for
532 * VMDK3 & VMDK4 which is 2TB.
533 * 64TB - for "ESXi seSparse Extent"
534 * minimal cluster size: 512B (default is 4KB)
535 * L2 table size: 4096 entries (const).
536 * 64TB is more than the maximal value supported for
537 * seSparse VMDKs (which is slightly less than 64TB)
538 */
539 error_setg(errp, "L1 size too big");
540 return -EFBIG;
541 }
542
543 nb_sectors = bdrv_nb_sectors(file->bs);
544 if (nb_sectors < 0) {
545 return nb_sectors;
546 }
547
548 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
549 extent = &s->extents[s->num_extents];
550 s->num_extents++;
551
552 memset(extent, 0, sizeof(VmdkExtent));
553 extent->file = file;
554 extent->flat = flat;
555 extent->sectors = sectors;
556 extent->l1_table_offset = l1_offset;
557 extent->l1_backup_table_offset = l1_backup_offset;
558 extent->l1_size = l1_size;
559 extent->l1_entry_sectors = l2_size * cluster_sectors;
560 extent->l2_size = l2_size;
561 extent->cluster_sectors = flat ? sectors : cluster_sectors;
562 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
563 extent->entry_size = sizeof(uint32_t);
564
565 if (s->num_extents > 1) {
566 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
567 } else {
568 extent->end_sector = extent->sectors;
569 }
570 bs->total_sectors = extent->end_sector;
571 if (new_extent) {
572 *new_extent = extent;
573 }
574 return 0;
575 }
576
577 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
578 Error **errp)
579 {
580 int ret;
581 size_t l1_size;
582 int i;
583
584 /* read the L1 table */
585 l1_size = extent->l1_size * extent->entry_size;
586 extent->l1_table = g_try_malloc(l1_size);
587 if (l1_size && extent->l1_table == NULL) {
588 return -ENOMEM;
589 }
590
591 ret = bdrv_pread(extent->file, extent->l1_table_offset, l1_size,
592 extent->l1_table, 0);
593 if (ret < 0) {
594 bdrv_refresh_filename(extent->file->bs);
595 error_setg_errno(errp, -ret,
596 "Could not read l1 table from extent '%s'",
597 extent->file->bs->filename);
598 goto fail_l1;
599 }
600 for (i = 0; i < extent->l1_size; i++) {
601 if (extent->entry_size == sizeof(uint64_t)) {
602 le64_to_cpus((uint64_t *)extent->l1_table + i);
603 } else {
604 assert(extent->entry_size == sizeof(uint32_t));
605 le32_to_cpus((uint32_t *)extent->l1_table + i);
606 }
607 }
608
609 if (extent->l1_backup_table_offset) {
610 assert(!extent->sesparse);
611 extent->l1_backup_table = g_try_malloc(l1_size);
612 if (l1_size && extent->l1_backup_table == NULL) {
613 ret = -ENOMEM;
614 goto fail_l1;
615 }
616 ret = bdrv_pread(extent->file, extent->l1_backup_table_offset,
617 l1_size, extent->l1_backup_table, 0);
618 if (ret < 0) {
619 bdrv_refresh_filename(extent->file->bs);
620 error_setg_errno(errp, -ret,
621 "Could not read l1 backup table from extent '%s'",
622 extent->file->bs->filename);
623 goto fail_l1b;
624 }
625 for (i = 0; i < extent->l1_size; i++) {
626 le32_to_cpus(&extent->l1_backup_table[i]);
627 }
628 }
629
630 extent->l2_cache =
631 g_malloc(extent->entry_size * extent->l2_size * L2_CACHE_SIZE);
632 return 0;
633 fail_l1b:
634 g_free(extent->l1_backup_table);
635 fail_l1:
636 g_free(extent->l1_table);
637 return ret;
638 }
639
640 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
641 BdrvChild *file,
642 int flags, Error **errp)
643 {
644 int ret;
645 uint32_t magic;
646 VMDK3Header header;
647 VmdkExtent *extent = NULL;
648
649 ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0);
650 if (ret < 0) {
651 bdrv_refresh_filename(file->bs);
652 error_setg_errno(errp, -ret,
653 "Could not read header from file '%s'",
654 file->bs->filename);
655 return ret;
656 }
657 ret = vmdk_add_extent(bs, file, false,
658 le32_to_cpu(header.disk_sectors),
659 (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
660 0,
661 le32_to_cpu(header.l1dir_size),
662 4096,
663 le32_to_cpu(header.granularity),
664 &extent,
665 errp);
666 if (ret < 0) {
667 return ret;
668 }
669 ret = vmdk_init_tables(bs, extent, errp);
670 if (ret) {
671 /* free extent allocated by vmdk_add_extent */
672 vmdk_free_last_extent(bs);
673 }
674 return ret;
675 }
676
677 #define SESPARSE_CONST_HEADER_MAGIC UINT64_C(0x00000000cafebabe)
678 #define SESPARSE_VOLATILE_HEADER_MAGIC UINT64_C(0x00000000cafecafe)
679
680 /* Strict checks - format not officially documented */
681 static int check_se_sparse_const_header(VMDKSESparseConstHeader *header,
682 Error **errp)
683 {
684 header->magic = le64_to_cpu(header->magic);
685 header->version = le64_to_cpu(header->version);
686 header->grain_size = le64_to_cpu(header->grain_size);
687 header->grain_table_size = le64_to_cpu(header->grain_table_size);
688 header->flags = le64_to_cpu(header->flags);
689 header->reserved1 = le64_to_cpu(header->reserved1);
690 header->reserved2 = le64_to_cpu(header->reserved2);
691 header->reserved3 = le64_to_cpu(header->reserved3);
692 header->reserved4 = le64_to_cpu(header->reserved4);
693
694 header->volatile_header_offset =
695 le64_to_cpu(header->volatile_header_offset);
696 header->volatile_header_size = le64_to_cpu(header->volatile_header_size);
697
698 header->journal_header_offset = le64_to_cpu(header->journal_header_offset);
699 header->journal_header_size = le64_to_cpu(header->journal_header_size);
700
701 header->journal_offset = le64_to_cpu(header->journal_offset);
702 header->journal_size = le64_to_cpu(header->journal_size);
703
704 header->grain_dir_offset = le64_to_cpu(header->grain_dir_offset);
705 header->grain_dir_size = le64_to_cpu(header->grain_dir_size);
706
707 header->grain_tables_offset = le64_to_cpu(header->grain_tables_offset);
708 header->grain_tables_size = le64_to_cpu(header->grain_tables_size);
709
710 header->free_bitmap_offset = le64_to_cpu(header->free_bitmap_offset);
711 header->free_bitmap_size = le64_to_cpu(header->free_bitmap_size);
712
713 header->backmap_offset = le64_to_cpu(header->backmap_offset);
714 header->backmap_size = le64_to_cpu(header->backmap_size);
715
716 header->grains_offset = le64_to_cpu(header->grains_offset);
717 header->grains_size = le64_to_cpu(header->grains_size);
718
719 if (header->magic != SESPARSE_CONST_HEADER_MAGIC) {
720 error_setg(errp, "Bad const header magic: 0x%016" PRIx64,
721 header->magic);
722 return -EINVAL;
723 }
724
725 if (header->version != 0x0000000200000001) {
726 error_setg(errp, "Unsupported version: 0x%016" PRIx64,
727 header->version);
728 return -ENOTSUP;
729 }
730
731 if (header->grain_size != 8) {
732 error_setg(errp, "Unsupported grain size: %" PRIu64,
733 header->grain_size);
734 return -ENOTSUP;
735 }
736
737 if (header->grain_table_size != 64) {
738 error_setg(errp, "Unsupported grain table size: %" PRIu64,
739 header->grain_table_size);
740 return -ENOTSUP;
741 }
742
743 if (header->flags != 0) {
744 error_setg(errp, "Unsupported flags: 0x%016" PRIx64,
745 header->flags);
746 return -ENOTSUP;
747 }
748
749 if (header->reserved1 != 0 || header->reserved2 != 0 ||
750 header->reserved3 != 0 || header->reserved4 != 0) {
751 error_setg(errp, "Unsupported reserved bits:"
752 " 0x%016" PRIx64 " 0x%016" PRIx64
753 " 0x%016" PRIx64 " 0x%016" PRIx64,
754 header->reserved1, header->reserved2,
755 header->reserved3, header->reserved4);
756 return -ENOTSUP;
757 }
758
759 /* check that padding is 0 */
760 if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
761 error_setg(errp, "Unsupported non-zero const header padding");
762 return -ENOTSUP;
763 }
764
765 return 0;
766 }
767
768 static int check_se_sparse_volatile_header(VMDKSESparseVolatileHeader *header,
769 Error **errp)
770 {
771 header->magic = le64_to_cpu(header->magic);
772 header->free_gt_number = le64_to_cpu(header->free_gt_number);
773 header->next_txn_seq_number = le64_to_cpu(header->next_txn_seq_number);
774 header->replay_journal = le64_to_cpu(header->replay_journal);
775
776 if (header->magic != SESPARSE_VOLATILE_HEADER_MAGIC) {
777 error_setg(errp, "Bad volatile header magic: 0x%016" PRIx64,
778 header->magic);
779 return -EINVAL;
780 }
781
782 if (header->replay_journal) {
783 error_setg(errp, "Image is dirty, Replaying journal not supported");
784 return -ENOTSUP;
785 }
786
787 /* check that padding is 0 */
788 if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
789 error_setg(errp, "Unsupported non-zero volatile header padding");
790 return -ENOTSUP;
791 }
792
793 return 0;
794 }
795
796 static int vmdk_open_se_sparse(BlockDriverState *bs,
797 BdrvChild *file,
798 int flags, Error **errp)
799 {
800 int ret;
801 VMDKSESparseConstHeader const_header;
802 VMDKSESparseVolatileHeader volatile_header;
803 VmdkExtent *extent = NULL;
804
805 ret = bdrv_apply_auto_read_only(bs,
806 "No write support for seSparse images available", errp);
807 if (ret < 0) {
808 return ret;
809 }
810
811 assert(sizeof(const_header) == SECTOR_SIZE);
812
813 ret = bdrv_pread(file, 0, sizeof(const_header), &const_header, 0);
814 if (ret < 0) {
815 bdrv_refresh_filename(file->bs);
816 error_setg_errno(errp, -ret,
817 "Could not read const header from file '%s'",
818 file->bs->filename);
819 return ret;
820 }
821
822 /* check const header */
823 ret = check_se_sparse_const_header(&const_header, errp);
824 if (ret < 0) {
825 return ret;
826 }
827
828 assert(sizeof(volatile_header) == SECTOR_SIZE);
829
830 ret = bdrv_pread(file, const_header.volatile_header_offset * SECTOR_SIZE,
831 sizeof(volatile_header), &volatile_header, 0);
832 if (ret < 0) {
833 bdrv_refresh_filename(file->bs);
834 error_setg_errno(errp, -ret,
835 "Could not read volatile header from file '%s'",
836 file->bs->filename);
837 return ret;
838 }
839
840 /* check volatile header */
841 ret = check_se_sparse_volatile_header(&volatile_header, errp);
842 if (ret < 0) {
843 return ret;
844 }
845
846 ret = vmdk_add_extent(bs, file, false,
847 const_header.capacity,
848 const_header.grain_dir_offset * SECTOR_SIZE,
849 0,
850 const_header.grain_dir_size *
851 SECTOR_SIZE / sizeof(uint64_t),
852 const_header.grain_table_size *
853 SECTOR_SIZE / sizeof(uint64_t),
854 const_header.grain_size,
855 &extent,
856 errp);
857 if (ret < 0) {
858 return ret;
859 }
860
861 extent->sesparse = true;
862 extent->sesparse_l2_tables_offset = const_header.grain_tables_offset;
863 extent->sesparse_clusters_offset = const_header.grains_offset;
864 extent->entry_size = sizeof(uint64_t);
865
866 ret = vmdk_init_tables(bs, extent, errp);
867 if (ret) {
868 /* free extent allocated by vmdk_add_extent */
869 vmdk_free_last_extent(bs);
870 }
871
872 return ret;
873 }
874
875 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
876 QDict *options, Error **errp);
877
878 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
879 {
880 int64_t size;
881 char *buf;
882 int ret;
883
884 size = bdrv_getlength(file->bs);
885 if (size < 0) {
886 error_setg_errno(errp, -size, "Could not access file");
887 return NULL;
888 }
889
890 if (size < 4) {
891 /* Both descriptor file and sparse image must be much larger than 4
892 * bytes, also callers of vmdk_read_desc want to compare the first 4
893 * bytes with VMDK4_MAGIC, let's error out if less is read. */
894 error_setg(errp, "File is too small, not a valid image");
895 return NULL;
896 }
897
898 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */
899 buf = g_malloc(size + 1);
900
901 ret = bdrv_pread(file, desc_offset, size, buf, 0);
902 if (ret < 0) {
903 error_setg_errno(errp, -ret, "Could not read from file");
904 g_free(buf);
905 return NULL;
906 }
907 buf[size] = 0;
908
909 return buf;
910 }
911
912 static int vmdk_open_vmdk4(BlockDriverState *bs,
913 BdrvChild *file,
914 int flags, QDict *options, Error **errp)
915 {
916 int ret;
917 uint32_t magic;
918 uint32_t l1_size, l1_entry_sectors;
919 VMDK4Header header;
920 VmdkExtent *extent = NULL;
921 BDRVVmdkState *s = bs->opaque;
922 int64_t l1_backup_offset = 0;
923 bool compressed;
924
925 ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0);
926 if (ret < 0) {
927 bdrv_refresh_filename(file->bs);
928 error_setg_errno(errp, -ret,
929 "Could not read header from file '%s'",
930 file->bs->filename);
931 return -EINVAL;
932 }
933 if (header.capacity == 0) {
934 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
935 if (desc_offset) {
936 char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
937 if (!buf) {
938 return -EINVAL;
939 }
940 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
941 g_free(buf);
942 return ret;
943 }
944 }
945
946 if (!s->create_type) {
947 s->create_type = g_strdup("monolithicSparse");
948 }
949
950 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
951 /*
952 * The footer takes precedence over the header, so read it in. The
953 * footer starts at offset -1024 from the end: One sector for the
954 * footer, and another one for the end-of-stream marker.
955 */
956 struct {
957 struct {
958 uint64_t val;
959 uint32_t size;
960 uint32_t type;
961 uint8_t pad[512 - 16];
962 } QEMU_PACKED footer_marker;
963
964 uint32_t magic;
965 VMDK4Header header;
966 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
967
968 struct {
969 uint64_t val;
970 uint32_t size;
971 uint32_t type;
972 uint8_t pad[512 - 16];
973 } QEMU_PACKED eos_marker;
974 } QEMU_PACKED footer;
975
976 ret = bdrv_pread(file, bs->file->bs->total_sectors * 512 - 1536,
977 sizeof(footer), &footer, 0);
978 if (ret < 0) {
979 error_setg_errno(errp, -ret, "Failed to read footer");
980 return ret;
981 }
982
983 /* Some sanity checks for the footer */
984 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
985 le32_to_cpu(footer.footer_marker.size) != 0 ||
986 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
987 le64_to_cpu(footer.eos_marker.val) != 0 ||
988 le32_to_cpu(footer.eos_marker.size) != 0 ||
989 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
990 {
991 error_setg(errp, "Invalid footer");
992 return -EINVAL;
993 }
994
995 header = footer.header;
996 }
997
998 compressed =
999 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1000 if (le32_to_cpu(header.version) > 3) {
1001 error_setg(errp, "Unsupported VMDK version %" PRIu32,
1002 le32_to_cpu(header.version));
1003 return -ENOTSUP;
1004 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
1005 !compressed) {
1006 /* VMware KB 2064959 explains that version 3 added support for
1007 * persistent changed block tracking (CBT), and backup software can
1008 * read it as version=1 if it doesn't care about the changed area
1009 * information. So we are safe to enable read only. */
1010 error_setg(errp, "VMDK version 3 must be read only");
1011 return -EINVAL;
1012 }
1013
1014 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
1015 error_setg(errp, "L2 table size too big");
1016 return -EINVAL;
1017 }
1018
1019 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
1020 * le64_to_cpu(header.granularity);
1021 if (l1_entry_sectors == 0) {
1022 error_setg(errp, "L1 entry size is invalid");
1023 return -EINVAL;
1024 }
1025 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
1026 / l1_entry_sectors;
1027 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
1028 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
1029 }
1030 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
1031 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
1032 (int64_t)(le64_to_cpu(header.grain_offset)
1033 * BDRV_SECTOR_SIZE));
1034 return -EINVAL;
1035 }
1036
1037 ret = vmdk_add_extent(bs, file, false,
1038 le64_to_cpu(header.capacity),
1039 le64_to_cpu(header.gd_offset) << 9,
1040 l1_backup_offset,
1041 l1_size,
1042 le32_to_cpu(header.num_gtes_per_gt),
1043 le64_to_cpu(header.granularity),
1044 &extent,
1045 errp);
1046 if (ret < 0) {
1047 return ret;
1048 }
1049 extent->compressed =
1050 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1051 if (extent->compressed) {
1052 g_free(s->create_type);
1053 s->create_type = g_strdup("streamOptimized");
1054 }
1055 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
1056 extent->version = le32_to_cpu(header.version);
1057 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
1058 ret = vmdk_init_tables(bs, extent, errp);
1059 if (ret) {
1060 /* free extent allocated by vmdk_add_extent */
1061 vmdk_free_last_extent(bs);
1062 }
1063 return ret;
1064 }
1065
1066 /* find an option value out of descriptor file */
1067 static int vmdk_parse_description(const char *desc, const char *opt_name,
1068 char *buf, int buf_size)
1069 {
1070 char *opt_pos, *opt_end;
1071 const char *end = desc + strlen(desc);
1072
1073 opt_pos = strstr(desc, opt_name);
1074 if (!opt_pos) {
1075 return VMDK_ERROR;
1076 }
1077 /* Skip "=\"" following opt_name */
1078 opt_pos += strlen(opt_name) + 2;
1079 if (opt_pos >= end) {
1080 return VMDK_ERROR;
1081 }
1082 opt_end = opt_pos;
1083 while (opt_end < end && *opt_end != '"') {
1084 opt_end++;
1085 }
1086 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
1087 return VMDK_ERROR;
1088 }
1089 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
1090 return VMDK_OK;
1091 }
1092
1093 /* Open an extent file and append to bs array */
1094 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
1095 char *buf, QDict *options, Error **errp)
1096 {
1097 uint32_t magic;
1098
1099 magic = ldl_be_p(buf);
1100 switch (magic) {
1101 case VMDK3_MAGIC:
1102 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
1103 case VMDK4_MAGIC:
1104 return vmdk_open_vmdk4(bs, file, flags, options, errp);
1105 default:
1106 error_setg(errp, "Image not in VMDK format");
1107 return -EINVAL;
1108 }
1109 }
1110
1111 static const char *next_line(const char *s)
1112 {
1113 while (*s) {
1114 if (*s == '\n') {
1115 return s + 1;
1116 }
1117 s++;
1118 }
1119 return s;
1120 }
1121
1122 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
1123 QDict *options, Error **errp)
1124 {
1125 int ret;
1126 int matches;
1127 char access[11];
1128 char type[11];
1129 char fname[512];
1130 const char *p, *np;
1131 int64_t sectors = 0;
1132 int64_t flat_offset;
1133 char *desc_file_dir = NULL;
1134 char *extent_path;
1135 BdrvChild *extent_file;
1136 BdrvChildRole extent_role;
1137 BDRVVmdkState *s = bs->opaque;
1138 VmdkExtent *extent = NULL;
1139 char extent_opt_prefix[32];
1140 Error *local_err = NULL;
1141
1142 for (p = desc; *p; p = next_line(p)) {
1143 /* parse extent line in one of below formats:
1144 *
1145 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
1146 * RW [size in sectors] SPARSE "file-name.vmdk"
1147 * RW [size in sectors] VMFS "file-name.vmdk"
1148 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
1149 * RW [size in sectors] SESPARSE "file-name.vmdk"
1150 */
1151 flat_offset = -1;
1152 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
1153 access, &sectors, type, fname, &flat_offset);
1154 if (matches < 4 || strcmp(access, "RW")) {
1155 continue;
1156 } else if (!strcmp(type, "FLAT")) {
1157 if (matches != 5 || flat_offset < 0) {
1158 goto invalid;
1159 }
1160 } else if (!strcmp(type, "VMFS")) {
1161 if (matches == 4) {
1162 flat_offset = 0;
1163 } else {
1164 goto invalid;
1165 }
1166 } else if (matches != 4) {
1167 goto invalid;
1168 }
1169
1170 if (sectors <= 0 ||
1171 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
1172 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") &&
1173 strcmp(type, "SESPARSE")) ||
1174 (strcmp(access, "RW"))) {
1175 continue;
1176 }
1177
1178 if (path_is_absolute(fname)) {
1179 extent_path = g_strdup(fname);
1180 } else {
1181 if (!desc_file_dir) {
1182 desc_file_dir = bdrv_dirname(bs->file->bs, errp);
1183 if (!desc_file_dir) {
1184 bdrv_refresh_filename(bs->file->bs);
1185 error_prepend(errp, "Cannot use relative paths with VMDK "
1186 "descriptor file '%s': ",
1187 bs->file->bs->filename);
1188 ret = -EINVAL;
1189 goto out;
1190 }
1191 }
1192
1193 extent_path = g_strconcat(desc_file_dir, fname, NULL);
1194 }
1195
1196 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
1197 assert(ret < 32);
1198
1199 extent_role = BDRV_CHILD_DATA;
1200 if (strcmp(type, "FLAT") != 0 && strcmp(type, "VMFS") != 0) {
1201 /* non-flat extents have metadata */
1202 extent_role |= BDRV_CHILD_METADATA;
1203 }
1204
1205 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
1206 bs, &child_of_bds, extent_role, false,
1207 &local_err);
1208 g_free(extent_path);
1209 if (local_err) {
1210 error_propagate(errp, local_err);
1211 ret = -EINVAL;
1212 goto out;
1213 }
1214
1215 /* save to extents array */
1216 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
1217 /* FLAT extent */
1218
1219 ret = vmdk_add_extent(bs, extent_file, true, sectors,
1220 0, 0, 0, 0, 0, &extent, errp);
1221 if (ret < 0) {
1222 bdrv_unref_child(bs, extent_file);
1223 goto out;
1224 }
1225 extent->flat_start_offset = flat_offset << 9;
1226 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
1227 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
1228 char *buf = vmdk_read_desc(extent_file, 0, errp);
1229 if (!buf) {
1230 ret = -EINVAL;
1231 } else {
1232 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
1233 options, errp);
1234 }
1235 g_free(buf);
1236 if (ret) {
1237 bdrv_unref_child(bs, extent_file);
1238 goto out;
1239 }
1240 extent = &s->extents[s->num_extents - 1];
1241 } else if (!strcmp(type, "SESPARSE")) {
1242 ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
1243 if (ret) {
1244 bdrv_unref_child(bs, extent_file);
1245 goto out;
1246 }
1247 extent = &s->extents[s->num_extents - 1];
1248 } else {
1249 error_setg(errp, "Unsupported extent type '%s'", type);
1250 bdrv_unref_child(bs, extent_file);
1251 ret = -ENOTSUP;
1252 goto out;
1253 }
1254 extent->type = g_strdup(type);
1255 }
1256
1257 ret = 0;
1258 goto out;
1259
1260 invalid:
1261 np = next_line(p);
1262 assert(np != p);
1263 if (np[-1] == '\n') {
1264 np--;
1265 }
1266 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
1267 ret = -EINVAL;
1268
1269 out:
1270 g_free(desc_file_dir);
1271 return ret;
1272 }
1273
1274 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
1275 QDict *options, Error **errp)
1276 {
1277 int ret;
1278 char ct[128];
1279 BDRVVmdkState *s = bs->opaque;
1280
1281 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
1282 error_setg(errp, "invalid VMDK image descriptor");
1283 ret = -EINVAL;
1284 goto exit;
1285 }
1286 if (strcmp(ct, "monolithicFlat") &&
1287 strcmp(ct, "vmfs") &&
1288 strcmp(ct, "vmfsSparse") &&
1289 strcmp(ct, "seSparse") &&
1290 strcmp(ct, "twoGbMaxExtentSparse") &&
1291 strcmp(ct, "twoGbMaxExtentFlat")) {
1292 error_setg(errp, "Unsupported image type '%s'", ct);
1293 ret = -ENOTSUP;
1294 goto exit;
1295 }
1296 s->create_type = g_strdup(ct);
1297 s->desc_offset = 0;
1298 ret = vmdk_parse_extents(buf, bs, options, errp);
1299 exit:
1300 return ret;
1301 }
1302
1303 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
1304 Error **errp)
1305 {
1306 char *buf;
1307 int ret;
1308 BDRVVmdkState *s = bs->opaque;
1309 uint32_t magic;
1310
1311 ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
1312 if (ret < 0) {
1313 return ret;
1314 }
1315
1316 buf = vmdk_read_desc(bs->file, 0, errp);
1317 if (!buf) {
1318 return -EINVAL;
1319 }
1320
1321 magic = ldl_be_p(buf);
1322 switch (magic) {
1323 case VMDK3_MAGIC:
1324 case VMDK4_MAGIC:
1325 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
1326 errp);
1327 s->desc_offset = 0x200;
1328 break;
1329 default:
1330 /* No data in the descriptor file */
1331 bs->file->role &= ~BDRV_CHILD_DATA;
1332
1333 /* Must succeed because we have given up permissions if anything */
1334 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1335
1336 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1337 break;
1338 }
1339 if (ret) {
1340 goto fail;
1341 }
1342
1343 /* try to open parent images, if exist */
1344 ret = vmdk_parent_open(bs);
1345 if (ret) {
1346 goto fail;
1347 }
1348 ret = vmdk_read_cid(bs, 0, &s->cid);
1349 if (ret) {
1350 goto fail;
1351 }
1352 ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1353 if (ret) {
1354 goto fail;
1355 }
1356 qemu_co_mutex_init(&s->lock);
1357
1358 /* Disable migration when VMDK images are used */
1359 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1360 "does not support live migration",
1361 bdrv_get_device_or_node_name(bs));
1362 ret = migrate_add_blocker(s->migration_blocker, errp);
1363 if (ret < 0) {
1364 error_free(s->migration_blocker);
1365 goto fail;
1366 }
1367
1368 g_free(buf);
1369 return 0;
1370
1371 fail:
1372 g_free(buf);
1373 g_free(s->create_type);
1374 s->create_type = NULL;
1375 vmdk_free_extents(bs);
1376 return ret;
1377 }
1378
1379
1380 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1381 {
1382 BDRVVmdkState *s = bs->opaque;
1383 int i;
1384
1385 for (i = 0; i < s->num_extents; i++) {
1386 if (!s->extents[i].flat) {
1387 bs->bl.pwrite_zeroes_alignment =
1388 MAX(bs->bl.pwrite_zeroes_alignment,
1389 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1390 }
1391 }
1392 }
1393
1394 /**
1395 * get_whole_cluster
1396 *
1397 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1398 * to the cluster at @cluster_sector_num. If @zeroed is true, we're overwriting
1399 * a zeroed cluster in the current layer and must not copy data from the
1400 * backing file.
1401 *
1402 * If @skip_start_sector < @skip_end_sector, the relative range
1403 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1404 * it for call to write user data in the request.
1405 */
1406 static int coroutine_fn get_whole_cluster(BlockDriverState *bs,
1407 VmdkExtent *extent,
1408 uint64_t cluster_offset,
1409 uint64_t offset,
1410 uint64_t skip_start_bytes,
1411 uint64_t skip_end_bytes,
1412 bool zeroed)
1413 {
1414 int ret = VMDK_OK;
1415 int64_t cluster_bytes;
1416 uint8_t *whole_grain;
1417 bool copy_from_backing;
1418
1419 /* For COW, align request sector_num to cluster start */
1420 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1421 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1422 whole_grain = qemu_blockalign(bs, cluster_bytes);
1423 copy_from_backing = bs->backing && !zeroed;
1424
1425 if (!copy_from_backing) {
1426 memset(whole_grain, 0, skip_start_bytes);
1427 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1428 }
1429
1430 assert(skip_end_bytes <= cluster_bytes);
1431 /* we will be here if it's first write on non-exist grain(cluster).
1432 * try to read from parent image, if exist */
1433 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1434 ret = VMDK_ERROR;
1435 goto exit;
1436 }
1437
1438 /* Read backing data before skip range */
1439 if (skip_start_bytes > 0) {
1440 if (copy_from_backing) {
1441 /* qcow2 emits this on bs->file instead of bs->backing */
1442 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1443 ret = bdrv_co_pread(bs->backing, offset, skip_start_bytes,
1444 whole_grain, 0);
1445 if (ret < 0) {
1446 ret = VMDK_ERROR;
1447 goto exit;
1448 }
1449 }
1450 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1451 ret = bdrv_co_pwrite(extent->file, cluster_offset, skip_start_bytes,
1452 whole_grain, 0);
1453 if (ret < 0) {
1454 ret = VMDK_ERROR;
1455 goto exit;
1456 }
1457 }
1458 /* Read backing data after skip range */
1459 if (skip_end_bytes < cluster_bytes) {
1460 if (copy_from_backing) {
1461 /* qcow2 emits this on bs->file instead of bs->backing */
1462 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1463 ret = bdrv_co_pread(bs->backing, offset + skip_end_bytes,
1464 cluster_bytes - skip_end_bytes,
1465 whole_grain + skip_end_bytes, 0);
1466 if (ret < 0) {
1467 ret = VMDK_ERROR;
1468 goto exit;
1469 }
1470 }
1471 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1472 ret = bdrv_co_pwrite(extent->file, cluster_offset + skip_end_bytes,
1473 cluster_bytes - skip_end_bytes,
1474 whole_grain + skip_end_bytes, 0);
1475 if (ret < 0) {
1476 ret = VMDK_ERROR;
1477 goto exit;
1478 }
1479 }
1480
1481 ret = VMDK_OK;
1482 exit:
1483 qemu_vfree(whole_grain);
1484 return ret;
1485 }
1486
1487 static int coroutine_fn vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1488 uint32_t offset)
1489 {
1490 offset = cpu_to_le32(offset);
1491 /* update L2 table */
1492 BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1493 if (bdrv_co_pwrite(extent->file,
1494 ((int64_t)m_data->l2_offset * 512)
1495 + (m_data->l2_index * sizeof(offset)),
1496 sizeof(offset), &offset, 0) < 0) {
1497 return VMDK_ERROR;
1498 }
1499 /* update backup L2 table */
1500 if (extent->l1_backup_table_offset != 0) {
1501 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1502 if (bdrv_co_pwrite(extent->file,
1503 ((int64_t)m_data->l2_offset * 512)
1504 + (m_data->l2_index * sizeof(offset)),
1505 sizeof(offset), &offset, 0) < 0) {
1506 return VMDK_ERROR;
1507 }
1508 }
1509 if (bdrv_co_flush(extent->file->bs) < 0) {
1510 return VMDK_ERROR;
1511 }
1512 if (m_data->l2_cache_entry) {
1513 *m_data->l2_cache_entry = offset;
1514 }
1515
1516 return VMDK_OK;
1517 }
1518
1519 /**
1520 * get_cluster_offset
1521 *
1522 * Look up cluster offset in extent file by sector number, and store in
1523 * @cluster_offset.
1524 *
1525 * For flat extents, the start offset as parsed from the description file is
1526 * returned.
1527 *
1528 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1529 * offset for a new cluster and update L2 cache. If there is a backing file,
1530 * COW is done before returning; otherwise, zeroes are written to the allocated
1531 * cluster. Both COW and zero writing skips the sector range
1532 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1533 * has new data to write there.
1534 *
1535 * Returns: VMDK_OK if cluster exists and mapped in the image.
1536 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1537 * VMDK_ERROR if failed.
1538 */
1539 static int coroutine_fn get_cluster_offset(BlockDriverState *bs,
1540 VmdkExtent *extent,
1541 VmdkMetaData *m_data,
1542 uint64_t offset,
1543 bool allocate,
1544 uint64_t *cluster_offset,
1545 uint64_t skip_start_bytes,
1546 uint64_t skip_end_bytes)
1547 {
1548 unsigned int l1_index, l2_offset, l2_index;
1549 int min_index, i, j;
1550 uint32_t min_count;
1551 void *l2_table;
1552 bool zeroed = false;
1553 int64_t ret;
1554 int64_t cluster_sector;
1555 unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
1556
1557 if (m_data) {
1558 m_data->new_allocation = false;
1559 }
1560 if (extent->flat) {
1561 *cluster_offset = extent->flat_start_offset;
1562 return VMDK_OK;
1563 }
1564
1565 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1566 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1567 if (l1_index >= extent->l1_size) {
1568 return VMDK_ERROR;
1569 }
1570 if (extent->sesparse) {
1571 uint64_t l2_offset_u64;
1572
1573 assert(extent->entry_size == sizeof(uint64_t));
1574
1575 l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
1576 if (l2_offset_u64 == 0) {
1577 l2_offset = 0;
1578 } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
1579 /*
1580 * Top most nibble is 0x1 if grain table is allocated.
1581 * strict check - top most 4 bytes must be 0x10000000 since max
1582 * supported size is 64TB for disk - so no more than 64TB / 16MB
1583 * grain directories which is smaller than uint32,
1584 * where 16MB is the only supported default grain table coverage.
1585 */
1586 return VMDK_ERROR;
1587 } else {
1588 l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
1589 l2_offset_u64 = extent->sesparse_l2_tables_offset +
1590 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
1591 if (l2_offset_u64 > 0x00000000ffffffff) {
1592 return VMDK_ERROR;
1593 }
1594 l2_offset = (unsigned int)(l2_offset_u64);
1595 }
1596 } else {
1597 assert(extent->entry_size == sizeof(uint32_t));
1598 l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
1599 }
1600 if (!l2_offset) {
1601 return VMDK_UNALLOC;
1602 }
1603 for (i = 0; i < L2_CACHE_SIZE; i++) {
1604 if (l2_offset == extent->l2_cache_offsets[i]) {
1605 /* increment the hit count */
1606 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1607 for (j = 0; j < L2_CACHE_SIZE; j++) {
1608 extent->l2_cache_counts[j] >>= 1;
1609 }
1610 }
1611 l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
1612 goto found;
1613 }
1614 }
1615 /* not found: load a new entry in the least used one */
1616 min_index = 0;
1617 min_count = 0xffffffff;
1618 for (i = 0; i < L2_CACHE_SIZE; i++) {
1619 if (extent->l2_cache_counts[i] < min_count) {
1620 min_count = extent->l2_cache_counts[i];
1621 min_index = i;
1622 }
1623 }
1624 l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
1625 BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1626 if (bdrv_co_pread(extent->file,
1627 (int64_t)l2_offset * 512,
1628 l2_size_bytes,
1629 l2_table, 0
1630 ) < 0) {
1631 return VMDK_ERROR;
1632 }
1633
1634 extent->l2_cache_offsets[min_index] = l2_offset;
1635 extent->l2_cache_counts[min_index] = 1;
1636 found:
1637 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1638 if (m_data) {
1639 m_data->l1_index = l1_index;
1640 m_data->l2_index = l2_index;
1641 m_data->l2_offset = l2_offset;
1642 m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
1643 }
1644
1645 if (extent->sesparse) {
1646 cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
1647 switch (cluster_sector & 0xf000000000000000) {
1648 case 0x0000000000000000:
1649 /* unallocated grain */
1650 if (cluster_sector != 0) {
1651 return VMDK_ERROR;
1652 }
1653 break;
1654 case 0x1000000000000000:
1655 /* scsi-unmapped grain - fallthrough */
1656 case 0x2000000000000000:
1657 /* zero grain */
1658 zeroed = true;
1659 break;
1660 case 0x3000000000000000:
1661 /* allocated grain */
1662 cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
1663 ((cluster_sector & 0x0000ffffffffffff) << 12));
1664 cluster_sector = extent->sesparse_clusters_offset +
1665 cluster_sector * extent->cluster_sectors;
1666 break;
1667 default:
1668 return VMDK_ERROR;
1669 }
1670 } else {
1671 cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
1672
1673 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1674 zeroed = true;
1675 }
1676 }
1677
1678 if (!cluster_sector || zeroed) {
1679 if (!allocate) {
1680 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1681 }
1682 assert(!extent->sesparse);
1683
1684 if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1685 return VMDK_ERROR;
1686 }
1687
1688 cluster_sector = extent->next_cluster_sector;
1689 extent->next_cluster_sector += extent->cluster_sectors;
1690
1691 /* First of all we write grain itself, to avoid race condition
1692 * that may to corrupt the image.
1693 * This problem may occur because of insufficient space on host disk
1694 * or inappropriate VM shutdown.
1695 */
1696 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1697 offset, skip_start_bytes, skip_end_bytes,
1698 zeroed);
1699 if (ret) {
1700 return ret;
1701 }
1702 if (m_data) {
1703 m_data->new_allocation = true;
1704 }
1705 }
1706 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1707 return VMDK_OK;
1708 }
1709
1710 static VmdkExtent *find_extent(BDRVVmdkState *s,
1711 int64_t sector_num, VmdkExtent *start_hint)
1712 {
1713 VmdkExtent *extent = start_hint;
1714
1715 if (!extent) {
1716 extent = &s->extents[0];
1717 }
1718 while (extent < &s->extents[s->num_extents]) {
1719 if (sector_num < extent->end_sector) {
1720 return extent;
1721 }
1722 extent++;
1723 }
1724 return NULL;
1725 }
1726
1727 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1728 int64_t offset)
1729 {
1730 uint64_t extent_begin_offset, extent_relative_offset;
1731 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1732
1733 extent_begin_offset =
1734 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1735 extent_relative_offset = offset - extent_begin_offset;
1736 return extent_relative_offset % cluster_size;
1737 }
1738
1739 static int coroutine_fn vmdk_co_block_status(BlockDriverState *bs,
1740 bool want_zero,
1741 int64_t offset, int64_t bytes,
1742 int64_t *pnum, int64_t *map,
1743 BlockDriverState **file)
1744 {
1745 BDRVVmdkState *s = bs->opaque;
1746 int64_t index_in_cluster, n, ret;
1747 uint64_t cluster_offset;
1748 VmdkExtent *extent;
1749
1750 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1751 if (!extent) {
1752 return -EIO;
1753 }
1754 qemu_co_mutex_lock(&s->lock);
1755 ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1756 0, 0);
1757 qemu_co_mutex_unlock(&s->lock);
1758
1759 index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1760 switch (ret) {
1761 case VMDK_ERROR:
1762 ret = -EIO;
1763 break;
1764 case VMDK_UNALLOC:
1765 ret = 0;
1766 break;
1767 case VMDK_ZEROED:
1768 ret = BDRV_BLOCK_ZERO;
1769 break;
1770 case VMDK_OK:
1771 ret = BDRV_BLOCK_DATA;
1772 if (!extent->compressed) {
1773 ret |= BDRV_BLOCK_OFFSET_VALID;
1774 *map = cluster_offset + index_in_cluster;
1775 if (extent->flat) {
1776 ret |= BDRV_BLOCK_RECURSE;
1777 }
1778 }
1779 *file = extent->file->bs;
1780 break;
1781 }
1782
1783 n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1784 *pnum = MIN(n, bytes);
1785 return ret;
1786 }
1787
1788 static int coroutine_fn
1789 vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1790 int64_t offset_in_cluster, QEMUIOVector *qiov,
1791 uint64_t qiov_offset, uint64_t n_bytes,
1792 uint64_t offset)
1793 {
1794 int ret;
1795 VmdkGrainMarker *data = NULL;
1796 uLongf buf_len;
1797 QEMUIOVector local_qiov;
1798 int64_t write_offset;
1799 int64_t write_end_sector;
1800
1801 if (extent->compressed) {
1802 void *compressed_data;
1803
1804 /* Only whole clusters */
1805 if (offset_in_cluster ||
1806 n_bytes > (extent->cluster_sectors * SECTOR_SIZE) ||
1807 (n_bytes < (extent->cluster_sectors * SECTOR_SIZE) &&
1808 offset + n_bytes != extent->end_sector * SECTOR_SIZE))
1809 {
1810 ret = -EINVAL;
1811 goto out;
1812 }
1813
1814 if (!extent->has_marker) {
1815 ret = -EINVAL;
1816 goto out;
1817 }
1818 buf_len = (extent->cluster_sectors << 9) * 2;
1819 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1820
1821 compressed_data = g_malloc(n_bytes);
1822 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1823 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1824 g_free(compressed_data);
1825
1826 if (ret != Z_OK || buf_len == 0) {
1827 ret = -EINVAL;
1828 goto out;
1829 }
1830
1831 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1832 data->size = cpu_to_le32(buf_len);
1833
1834 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1835 qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1836
1837 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1838 } else {
1839 qemu_iovec_init(&local_qiov, qiov->niov);
1840 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1841
1842 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1843 }
1844
1845 write_offset = cluster_offset + offset_in_cluster;
1846 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1847 &local_qiov, 0);
1848
1849 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1850
1851 if (extent->compressed) {
1852 extent->next_cluster_sector = write_end_sector;
1853 } else {
1854 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1855 write_end_sector);
1856 }
1857
1858 if (ret < 0) {
1859 goto out;
1860 }
1861 ret = 0;
1862 out:
1863 g_free(data);
1864 if (!extent->compressed) {
1865 qemu_iovec_destroy(&local_qiov);
1866 }
1867 return ret;
1868 }
1869
1870 static int coroutine_fn
1871 vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1872 int64_t offset_in_cluster, QEMUIOVector *qiov,
1873 int bytes)
1874 {
1875 int ret;
1876 int cluster_bytes, buf_bytes;
1877 uint8_t *cluster_buf, *compressed_data;
1878 uint8_t *uncomp_buf;
1879 uint32_t data_len;
1880 VmdkGrainMarker *marker;
1881 uLongf buf_len;
1882
1883
1884 if (!extent->compressed) {
1885 BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1886 ret = bdrv_co_preadv(extent->file,
1887 cluster_offset + offset_in_cluster, bytes,
1888 qiov, 0);
1889 if (ret < 0) {
1890 return ret;
1891 }
1892 return 0;
1893 }
1894 cluster_bytes = extent->cluster_sectors * 512;
1895 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1896 buf_bytes = cluster_bytes * 2;
1897 cluster_buf = g_malloc(buf_bytes);
1898 uncomp_buf = g_malloc(cluster_bytes);
1899 BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1900 ret = bdrv_co_pread(extent->file, cluster_offset, buf_bytes, cluster_buf,
1901 0);
1902 if (ret < 0) {
1903 goto out;
1904 }
1905 compressed_data = cluster_buf;
1906 buf_len = cluster_bytes;
1907 data_len = cluster_bytes;
1908 if (extent->has_marker) {
1909 marker = (VmdkGrainMarker *)cluster_buf;
1910 compressed_data = marker->data;
1911 data_len = le32_to_cpu(marker->size);
1912 }
1913 if (!data_len || data_len > buf_bytes) {
1914 ret = -EINVAL;
1915 goto out;
1916 }
1917 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1918 if (ret != Z_OK) {
1919 ret = -EINVAL;
1920 goto out;
1921
1922 }
1923 if (offset_in_cluster < 0 ||
1924 offset_in_cluster + bytes > buf_len) {
1925 ret = -EINVAL;
1926 goto out;
1927 }
1928 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1929 ret = 0;
1930
1931 out:
1932 g_free(uncomp_buf);
1933 g_free(cluster_buf);
1934 return ret;
1935 }
1936
1937 static int coroutine_fn
1938 vmdk_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
1939 QEMUIOVector *qiov, BdrvRequestFlags flags)
1940 {
1941 BDRVVmdkState *s = bs->opaque;
1942 int ret;
1943 uint64_t n_bytes, offset_in_cluster;
1944 VmdkExtent *extent = NULL;
1945 QEMUIOVector local_qiov;
1946 uint64_t cluster_offset;
1947 uint64_t bytes_done = 0;
1948
1949 qemu_iovec_init(&local_qiov, qiov->niov);
1950 qemu_co_mutex_lock(&s->lock);
1951
1952 while (bytes > 0) {
1953 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1954 if (!extent) {
1955 ret = -EIO;
1956 goto fail;
1957 }
1958 ret = get_cluster_offset(bs, extent, NULL,
1959 offset, false, &cluster_offset, 0, 0);
1960 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1961
1962 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1963 - offset_in_cluster);
1964
1965 if (ret != VMDK_OK) {
1966 /* if not allocated, try to read from parent image, if exist */
1967 if (bs->backing && ret != VMDK_ZEROED) {
1968 if (!vmdk_is_cid_valid(bs)) {
1969 ret = -EINVAL;
1970 goto fail;
1971 }
1972
1973 qemu_iovec_reset(&local_qiov);
1974 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1975
1976 /* qcow2 emits this on bs->file instead of bs->backing */
1977 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1978 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1979 &local_qiov, 0);
1980 if (ret < 0) {
1981 goto fail;
1982 }
1983 } else {
1984 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1985 }
1986 } else {
1987 qemu_iovec_reset(&local_qiov);
1988 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1989
1990 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1991 &local_qiov, n_bytes);
1992 if (ret) {
1993 goto fail;
1994 }
1995 }
1996 bytes -= n_bytes;
1997 offset += n_bytes;
1998 bytes_done += n_bytes;
1999 }
2000
2001 ret = 0;
2002 fail:
2003 qemu_co_mutex_unlock(&s->lock);
2004 qemu_iovec_destroy(&local_qiov);
2005
2006 return ret;
2007 }
2008
2009 /**
2010 * vmdk_write:
2011 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
2012 * if possible, otherwise return -ENOTSUP.
2013 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
2014 * with each cluster. By dry run we can find if the zero write
2015 * is possible without modifying image data.
2016 *
2017 * Returns: error code with 0 for success.
2018 */
2019 static int coroutine_fn vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
2020 uint64_t bytes, QEMUIOVector *qiov,
2021 bool zeroed, bool zero_dry_run)
2022 {
2023 BDRVVmdkState *s = bs->opaque;
2024 VmdkExtent *extent = NULL;
2025 int ret;
2026 int64_t offset_in_cluster, n_bytes;
2027 uint64_t cluster_offset;
2028 uint64_t bytes_done = 0;
2029 VmdkMetaData m_data;
2030
2031 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
2032 error_report("Wrong offset: offset=0x%" PRIx64
2033 " total_sectors=0x%" PRIx64,
2034 offset, bs->total_sectors);
2035 return -EIO;
2036 }
2037
2038 while (bytes > 0) {
2039 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
2040 if (!extent) {
2041 return -EIO;
2042 }
2043 if (extent->sesparse) {
2044 return -ENOTSUP;
2045 }
2046 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
2047 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
2048 - offset_in_cluster);
2049
2050 ret = get_cluster_offset(bs, extent, &m_data, offset,
2051 !(extent->compressed || zeroed),
2052 &cluster_offset, offset_in_cluster,
2053 offset_in_cluster + n_bytes);
2054 if (extent->compressed) {
2055 if (ret == VMDK_OK) {
2056 /* Refuse write to allocated cluster for streamOptimized */
2057 error_report("Could not write to allocated cluster"
2058 " for streamOptimized");
2059 return -EIO;
2060 } else if (!zeroed) {
2061 /* allocate */
2062 ret = get_cluster_offset(bs, extent, &m_data, offset,
2063 true, &cluster_offset, 0, 0);
2064 }
2065 }
2066 if (ret == VMDK_ERROR) {
2067 return -EINVAL;
2068 }
2069 if (zeroed) {
2070 /* Do zeroed write, buf is ignored */
2071 if (extent->has_zero_grain &&
2072 offset_in_cluster == 0 &&
2073 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
2074 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
2075 if (!zero_dry_run && ret != VMDK_ZEROED) {
2076 /* update L2 tables */
2077 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
2078 != VMDK_OK) {
2079 return -EIO;
2080 }
2081 }
2082 } else {
2083 return -ENOTSUP;
2084 }
2085 } else {
2086 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
2087 qiov, bytes_done, n_bytes, offset);
2088 if (ret) {
2089 return ret;
2090 }
2091 if (m_data.new_allocation) {
2092 /* update L2 tables */
2093 if (vmdk_L2update(extent, &m_data,
2094 cluster_offset >> BDRV_SECTOR_BITS)
2095 != VMDK_OK) {
2096 return -EIO;
2097 }
2098 }
2099 }
2100 bytes -= n_bytes;
2101 offset += n_bytes;
2102 bytes_done += n_bytes;
2103
2104 /* update CID on the first write every time the virtual disk is
2105 * opened */
2106 if (!s->cid_updated) {
2107 ret = vmdk_write_cid(bs, g_random_int());
2108 if (ret < 0) {
2109 return ret;
2110 }
2111 s->cid_updated = true;
2112 }
2113 }
2114 return 0;
2115 }
2116
2117 static int coroutine_fn
2118 vmdk_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
2119 QEMUIOVector *qiov, BdrvRequestFlags flags)
2120 {
2121 int ret;
2122 BDRVVmdkState *s = bs->opaque;
2123 qemu_co_mutex_lock(&s->lock);
2124 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
2125 qemu_co_mutex_unlock(&s->lock);
2126 return ret;
2127 }
2128
2129 static int coroutine_fn
2130 vmdk_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes,
2131 QEMUIOVector *qiov)
2132 {
2133 if (bytes == 0) {
2134 /* The caller will write bytes 0 to signal EOF.
2135 * When receive it, we align EOF to a sector boundary. */
2136 BDRVVmdkState *s = bs->opaque;
2137 int i, ret;
2138 int64_t length;
2139
2140 for (i = 0; i < s->num_extents; i++) {
2141 length = bdrv_getlength(s->extents[i].file->bs);
2142 if (length < 0) {
2143 return length;
2144 }
2145 length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
2146 ret = bdrv_co_truncate(s->extents[i].file, length, false,
2147 PREALLOC_MODE_OFF, 0, NULL);
2148 if (ret < 0) {
2149 return ret;
2150 }
2151 }
2152 return 0;
2153 }
2154 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
2155 }
2156
2157 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
2158 int64_t offset,
2159 int64_t bytes,
2160 BdrvRequestFlags flags)
2161 {
2162 int ret;
2163 BDRVVmdkState *s = bs->opaque;
2164
2165 qemu_co_mutex_lock(&s->lock);
2166 /* write zeroes could fail if sectors not aligned to cluster, test it with
2167 * dry_run == true before really updating image */
2168 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
2169 if (!ret) {
2170 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
2171 }
2172 qemu_co_mutex_unlock(&s->lock);
2173 return ret;
2174 }
2175
2176 static int vmdk_init_extent(BlockBackend *blk,
2177 int64_t filesize, bool flat,
2178 bool compress, bool zeroed_grain,
2179 Error **errp)
2180 {
2181 int ret, i;
2182 VMDK4Header header;
2183 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
2184 uint32_t *gd_buf = NULL;
2185 int gd_buf_size;
2186
2187 if (flat) {
2188 ret = blk_truncate(blk, filesize, false, PREALLOC_MODE_OFF, 0, errp);
2189 goto exit;
2190 }
2191 magic = cpu_to_be32(VMDK4_MAGIC);
2192 memset(&header, 0, sizeof(header));
2193 if (compress) {
2194 header.version = 3;
2195 } else if (zeroed_grain) {
2196 header.version = 2;
2197 } else {
2198 header.version = 1;
2199 }
2200 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
2201 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
2202 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
2203 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
2204 header.capacity = filesize / BDRV_SECTOR_SIZE;
2205 header.granularity = 128;
2206 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
2207
2208 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
2209 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
2210 BDRV_SECTOR_SIZE);
2211 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
2212 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
2213
2214 header.desc_offset = 1;
2215 header.desc_size = 20;
2216 header.rgd_offset = header.desc_offset + header.desc_size;
2217 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
2218 header.grain_offset =
2219 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
2220 header.granularity);
2221 /* swap endianness for all header fields */
2222 header.version = cpu_to_le32(header.version);
2223 header.flags = cpu_to_le32(header.flags);
2224 header.capacity = cpu_to_le64(header.capacity);
2225 header.granularity = cpu_to_le64(header.granularity);
2226 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
2227 header.desc_offset = cpu_to_le64(header.desc_offset);
2228 header.desc_size = cpu_to_le64(header.desc_size);
2229 header.rgd_offset = cpu_to_le64(header.rgd_offset);
2230 header.gd_offset = cpu_to_le64(header.gd_offset);
2231 header.grain_offset = cpu_to_le64(header.grain_offset);
2232 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
2233
2234 header.check_bytes[0] = 0xa;
2235 header.check_bytes[1] = 0x20;
2236 header.check_bytes[2] = 0xd;
2237 header.check_bytes[3] = 0xa;
2238
2239 /* write all the data */
2240 ret = blk_pwrite(blk, 0, sizeof(magic), &magic, 0);
2241 if (ret < 0) {
2242 error_setg(errp, QERR_IO_ERROR);
2243 goto exit;
2244 }
2245 ret = blk_pwrite(blk, sizeof(magic), sizeof(header), &header, 0);
2246 if (ret < 0) {
2247 error_setg(errp, QERR_IO_ERROR);
2248 goto exit;
2249 }
2250
2251 ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false,
2252 PREALLOC_MODE_OFF, 0, errp);
2253 if (ret < 0) {
2254 goto exit;
2255 }
2256
2257 /* write grain directory */
2258 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
2259 gd_buf = g_malloc0(gd_buf_size);
2260 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
2261 i < gt_count; i++, tmp += gt_size) {
2262 gd_buf[i] = cpu_to_le32(tmp);
2263 }
2264 ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
2265 gd_buf_size, gd_buf, 0);
2266 if (ret < 0) {
2267 error_setg(errp, QERR_IO_ERROR);
2268 goto exit;
2269 }
2270
2271 /* write backup grain directory */
2272 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
2273 i < gt_count; i++, tmp += gt_size) {
2274 gd_buf[i] = cpu_to_le32(tmp);
2275 }
2276 ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
2277 gd_buf_size, gd_buf, 0);
2278 if (ret < 0) {
2279 error_setg(errp, QERR_IO_ERROR);
2280 }
2281
2282 ret = 0;
2283 exit:
2284 g_free(gd_buf);
2285 return ret;
2286 }
2287
2288 static int coroutine_fn vmdk_create_extent(const char *filename,
2289 int64_t filesize, bool flat,
2290 bool compress, bool zeroed_grain,
2291 BlockBackend **pbb,
2292 QemuOpts *opts, Error **errp)
2293 {
2294 int ret;
2295 BlockBackend *blk = NULL;
2296
2297 ret = bdrv_co_create_file(filename, opts, errp);
2298 if (ret < 0) {
2299 goto exit;
2300 }
2301
2302 blk = blk_new_open(filename, NULL, NULL,
2303 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2304 errp);
2305 if (blk == NULL) {
2306 ret = -EIO;
2307 goto exit;
2308 }
2309
2310 blk_set_allow_write_beyond_eof(blk, true);
2311
2312 ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
2313 exit:
2314 if (blk) {
2315 if (pbb) {
2316 *pbb = blk;
2317 } else {
2318 blk_unref(blk);
2319 blk = NULL;
2320 }
2321 }
2322 return ret;
2323 }
2324
2325 static int filename_decompose(const char *filename, char *path, char *prefix,
2326 char *postfix, size_t buf_len, Error **errp)
2327 {
2328 const char *p, *q;
2329
2330 if (filename == NULL || !strlen(filename)) {
2331 error_setg(errp, "No filename provided");
2332 return VMDK_ERROR;
2333 }
2334 p = strrchr(filename, '/');
2335 if (p == NULL) {
2336 p = strrchr(filename, '\\');
2337 }
2338 if (p == NULL) {
2339 p = strrchr(filename, ':');
2340 }
2341 if (p != NULL) {
2342 p++;
2343 if (p - filename >= buf_len) {
2344 return VMDK_ERROR;
2345 }
2346 pstrcpy(path, p - filename + 1, filename);
2347 } else {
2348 p = filename;
2349 path[0] = '\0';
2350 }
2351 q = strrchr(p, '.');
2352 if (q == NULL) {
2353 pstrcpy(prefix, buf_len, p);
2354 postfix[0] = '\0';
2355 } else {
2356 if (q - p >= buf_len) {
2357 return VMDK_ERROR;
2358 }
2359 pstrcpy(prefix, q - p + 1, p);
2360 pstrcpy(postfix, buf_len, q);
2361 }
2362 return VMDK_OK;
2363 }
2364
2365 /*
2366 * idx == 0: get or create the descriptor file (also the image file if in a
2367 * non-split format.
2368 * idx >= 1: get the n-th extent if in a split subformat
2369 */
2370 typedef BlockBackend * coroutine_fn (*vmdk_create_extent_fn)(int64_t size,
2371 int idx,
2372 bool flat,
2373 bool split,
2374 bool compress,
2375 bool zeroed_grain,
2376 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 vmdk_co_do_create(int64_t size,
2391 BlockdevVmdkSubformat subformat,
2392 BlockdevVmdkAdapterType adapter_type,
2393 const char *backing_file,
2394 const char *hw_version,
2395 const char *toolsversion,
2396 bool compat6,
2397 bool zeroed_grain,
2398 vmdk_create_extent_fn extent_fn,
2399 void *opaque,
2400 Error **errp)
2401 {
2402 int extent_idx;
2403 BlockBackend *blk = NULL;
2404 BlockBackend *extent_blk;
2405 Error *local_err = NULL;
2406 char *desc = NULL;
2407 int ret = 0;
2408 bool flat, split, compress;
2409 GString *ext_desc_lines;
2410 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
2411 int64_t extent_size;
2412 int64_t created_size = 0;
2413 const char *extent_line_fmt;
2414 char *parent_desc_line = g_malloc0(BUF_SIZE);
2415 uint32_t parent_cid = 0xffffffff;
2416 uint32_t number_heads = 16;
2417 uint32_t desc_offset = 0, desc_len;
2418 const char desc_template[] =
2419 "# Disk DescriptorFile\n"
2420 "version=1\n"
2421 "CID=%" PRIx32 "\n"
2422 "parentCID=%" PRIx32 "\n"
2423 "createType=\"%s\"\n"
2424 "%s"
2425 "\n"
2426 "# Extent description\n"
2427 "%s"
2428 "\n"
2429 "# The Disk Data Base\n"
2430 "#DDB\n"
2431 "\n"
2432 "ddb.virtualHWVersion = \"%s\"\n"
2433 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2434 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2435 "ddb.geometry.sectors = \"63\"\n"
2436 "ddb.adapterType = \"%s\"\n"
2437 "ddb.toolsVersion = \"%s\"\n";
2438
2439 ext_desc_lines = g_string_new(NULL);
2440
2441 /* Read out options */
2442 if (compat6) {
2443 if (hw_version) {
2444 error_setg(errp,
2445 "compat6 cannot be enabled with hwversion set");
2446 ret = -EINVAL;
2447 goto exit;
2448 }
2449 hw_version = "6";
2450 }
2451 if (!hw_version) {
2452 hw_version = "4";
2453 }
2454 if (!toolsversion) {
2455 toolsversion = "2147483647";
2456 }
2457
2458 if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2459 /* that's the number of heads with which vmware operates when
2460 creating, exporting, etc. vmdk files with a non-ide adapter type */
2461 number_heads = 255;
2462 }
2463 split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2464 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2465 flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2466 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2467 compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2468
2469 if (flat) {
2470 extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2471 } else {
2472 extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2473 }
2474 if (flat && backing_file) {
2475 error_setg(errp, "Flat image can't have backing file");
2476 ret = -ENOTSUP;
2477 goto exit;
2478 }
2479 if (flat && zeroed_grain) {
2480 error_setg(errp, "Flat image can't enable zeroed grain");
2481 ret = -ENOTSUP;
2482 goto exit;
2483 }
2484
2485 /* Create extents */
2486 if (split) {
2487 extent_size = split_size;
2488 } else {
2489 extent_size = size;
2490 }
2491 if (!split && !flat) {
2492 created_size = extent_size;
2493 } else {
2494 created_size = 0;
2495 }
2496 /* Get the descriptor file BDS */
2497 blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2498 opaque, errp);
2499 if (!blk) {
2500 ret = -EIO;
2501 goto exit;
2502 }
2503 if (!split && !flat) {
2504 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2505 blk_bs(blk)->filename);
2506 }
2507
2508 if (backing_file) {
2509 BlockBackend *backing;
2510 char *full_backing =
2511 bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2512 backing_file,
2513 &local_err);
2514 if (local_err) {
2515 error_propagate(errp, local_err);
2516 ret = -ENOENT;
2517 goto exit;
2518 }
2519 assert(full_backing);
2520
2521 backing = blk_new_open(full_backing, NULL, NULL,
2522 BDRV_O_NO_BACKING, errp);
2523 g_free(full_backing);
2524 if (backing == NULL) {
2525 ret = -EIO;
2526 goto exit;
2527 }
2528 if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2529 error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2530 blk_bs(backing)->drv->format_name);
2531 blk_unref(backing);
2532 ret = -EINVAL;
2533 goto exit;
2534 }
2535 ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2536 blk_unref(backing);
2537 if (ret) {
2538 error_setg(errp, "Failed to read parent CID");
2539 goto exit;
2540 }
2541 snprintf(parent_desc_line, BUF_SIZE,
2542 "parentFileNameHint=\"%s\"", backing_file);
2543 }
2544 extent_idx = 1;
2545 while (created_size < size) {
2546 int64_t cur_size = MIN(size - created_size, extent_size);
2547 extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2548 zeroed_grain, opaque, errp);
2549 if (!extent_blk) {
2550 ret = -EINVAL;
2551 goto exit;
2552 }
2553 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2554 blk_bs(extent_blk)->filename);
2555 created_size += cur_size;
2556 extent_idx++;
2557 blk_unref(extent_blk);
2558 }
2559
2560 /* Check whether we got excess extents */
2561 extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2562 opaque, NULL);
2563 if (extent_blk) {
2564 blk_unref(extent_blk);
2565 error_setg(errp, "List of extents contains unused extents");
2566 ret = -EINVAL;
2567 goto exit;
2568 }
2569
2570 /* generate descriptor file */
2571 desc = g_strdup_printf(desc_template,
2572 g_random_int(),
2573 parent_cid,
2574 BlockdevVmdkSubformat_str(subformat),
2575 parent_desc_line,
2576 ext_desc_lines->str,
2577 hw_version,
2578 size /
2579 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2580 number_heads,
2581 BlockdevVmdkAdapterType_str(adapter_type),
2582 toolsversion);
2583 desc_len = strlen(desc);
2584 /* the descriptor offset = 0x200 */
2585 if (!split && !flat) {
2586 desc_offset = 0x200;
2587 }
2588
2589 ret = blk_co_pwrite(blk, desc_offset, desc_len, desc, 0);
2590 if (ret < 0) {
2591 error_setg_errno(errp, -ret, "Could not write description");
2592 goto exit;
2593 }
2594 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2595 * for description file */
2596 if (desc_offset == 0) {
2597 ret = blk_co_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, 0, errp);
2598 if (ret < 0) {
2599 goto exit;
2600 }
2601 }
2602 ret = 0;
2603 exit:
2604 if (blk) {
2605 blk_unref(blk);
2606 }
2607 g_free(desc);
2608 g_free(parent_desc_line);
2609 g_string_free(ext_desc_lines, true);
2610 return ret;
2611 }
2612
2613 typedef struct {
2614 char *path;
2615 char *prefix;
2616 char *postfix;
2617 QemuOpts *opts;
2618 } VMDKCreateOptsData;
2619
2620 static BlockBackend * coroutine_fn vmdk_co_create_opts_cb(int64_t size, int idx,
2621 bool flat, bool split, bool compress,
2622 bool zeroed_grain, void *opaque,
2623 Error **errp)
2624 {
2625 BlockBackend *blk = NULL;
2626 BlockDriverState *bs = NULL;
2627 VMDKCreateOptsData *data = opaque;
2628 char *ext_filename = NULL;
2629 char *rel_filename = NULL;
2630
2631 /* We're done, don't create excess extents. */
2632 if (size == -1) {
2633 assert(errp == NULL);
2634 return NULL;
2635 }
2636
2637 if (idx == 0) {
2638 rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2639 } else if (split) {
2640 rel_filename = g_strdup_printf("%s-%c%03d%s",
2641 data->prefix,
2642 flat ? 'f' : 's', idx, data->postfix);
2643 } else {
2644 assert(idx == 1);
2645 rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2646 }
2647
2648 ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2649 g_free(rel_filename);
2650
2651 if (vmdk_create_extent(ext_filename, size,
2652 flat, compress, zeroed_grain, &blk, data->opts,
2653 errp)) {
2654 goto exit;
2655 }
2656 bdrv_unref(bs);
2657 exit:
2658 g_free(ext_filename);
2659 return blk;
2660 }
2661
2662 static int coroutine_fn vmdk_co_create_opts(BlockDriver *drv,
2663 const char *filename,
2664 QemuOpts *opts,
2665 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 vmdk_co_create_cb(int64_t size, int idx,
2773 bool flat, bool split,
2774 bool compress,
2775 bool zeroed_grain,
2776 void *opaque, Error **errp)
2777 {
2778 int ret;
2779 BlockDriverState *bs;
2780 BlockBackend *blk;
2781 BlockdevCreateOptionsVmdk *opts = opaque;
2782
2783 if (idx == 0) {
2784 bs = bdrv_open_blockdev_ref(opts->file, errp);
2785 } else {
2786 int i;
2787 BlockdevRefList *list = opts->extents;
2788 for (i = 1; i < idx; i++) {
2789 if (!list || !list->next) {
2790 error_setg(errp, "Extent [%d] not specified", i);
2791 return NULL;
2792 }
2793 list = list->next;
2794 }
2795 if (!list) {
2796 error_setg(errp, "Extent [%d] not specified", idx - 1);
2797 return NULL;
2798 }
2799 bs = bdrv_open_blockdev_ref(list->value, errp);
2800 }
2801 if (!bs) {
2802 return NULL;
2803 }
2804 blk = blk_new_with_bs(bs,
2805 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | BLK_PERM_RESIZE,
2806 BLK_PERM_ALL, errp);
2807 if (!blk) {
2808 return NULL;
2809 }
2810 blk_set_allow_write_beyond_eof(blk, true);
2811 bdrv_unref(bs);
2812
2813 if (size != -1) {
2814 ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2815 if (ret) {
2816 blk_unref(blk);
2817 blk = NULL;
2818 }
2819 }
2820 return blk;
2821 }
2822
2823 static int coroutine_fn vmdk_co_create(BlockdevCreateOptions *create_options,
2824 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
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 vmdk_co_check(BlockDriverState *bs,
2920 BdrvCheckResult *result,
2921 BdrvCheckMode fix)
2922 {
2923 BDRVVmdkState *s = bs->opaque;
2924 VmdkExtent *extent = NULL;
2925 int64_t sector_num = 0;
2926 int64_t total_sectors = bdrv_nb_sectors(bs);
2927 int ret;
2928 uint64_t cluster_offset;
2929
2930 if (fix) {
2931 return -ENOTSUP;
2932 }
2933
2934 for (;;) {
2935 if (sector_num >= total_sectors) {
2936 return 0;
2937 }
2938 extent = find_extent(s, sector_num, extent);
2939 if (!extent) {
2940 fprintf(stderr,
2941 "ERROR: could not find extent for sector %" PRId64 "\n",
2942 sector_num);
2943 ret = -EINVAL;
2944 break;
2945 }
2946 ret = get_cluster_offset(bs, extent, NULL,
2947 sector_num << BDRV_SECTOR_BITS,
2948 false, &cluster_offset, 0, 0);
2949 if (ret == VMDK_ERROR) {
2950 fprintf(stderr,
2951 "ERROR: could not get cluster_offset for sector %"
2952 PRId64 "\n", sector_num);
2953 break;
2954 }
2955 if (ret == VMDK_OK) {
2956 int64_t extent_len = bdrv_getlength(extent->file->bs);
2957 if (extent_len < 0) {
2958 fprintf(stderr,
2959 "ERROR: could not get extent file length for sector %"
2960 PRId64 "\n", sector_num);
2961 ret = extent_len;
2962 break;
2963 }
2964 if (cluster_offset >= extent_len) {
2965 fprintf(stderr,
2966 "ERROR: cluster offset for sector %"
2967 PRId64 " points after EOF\n", sector_num);
2968 ret = -EINVAL;
2969 break;
2970 }
2971 }
2972 sector_num += extent->cluster_sectors;
2973 }
2974
2975 result->corruptions++;
2976 return ret;
2977 }
2978
2979 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs,
2980 Error **errp)
2981 {
2982 int i;
2983 BDRVVmdkState *s = bs->opaque;
2984 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2985 VmdkExtentInfoList **tail;
2986
2987 *spec_info = (ImageInfoSpecific){
2988 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2989 .u = {
2990 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2991 },
2992 };
2993
2994 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2995 .create_type = g_strdup(s->create_type),
2996 .cid = s->cid,
2997 .parent_cid = s->parent_cid,
2998 };
2999
3000 tail = &spec_info->u.vmdk.data->extents;
3001 for (i = 0; i < s->num_extents; i++) {
3002 QAPI_LIST_APPEND(tail, vmdk_get_extent_info(&s->extents[i]));
3003 }
3004
3005 return spec_info;
3006 }
3007
3008 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
3009 {
3010 return a->flat == b->flat &&
3011 a->compressed == b->compressed &&
3012 (a->flat || a->cluster_sectors == b->cluster_sectors);
3013 }
3014
3015 static int coroutine_fn
3016 vmdk_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3017 {
3018 int i;
3019 BDRVVmdkState *s = bs->opaque;
3020 assert(s->num_extents);
3021
3022 /* See if we have multiple extents but they have different cases */
3023 for (i = 1; i < s->num_extents; i++) {
3024 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
3025 return -ENOTSUP;
3026 }
3027 }
3028 bdi->needs_compressed_writes = s->extents[0].compressed;
3029 if (!s->extents[0].flat) {
3030 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
3031 }
3032 return 0;
3033 }
3034
3035 static void vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
3036 bool backing_overridden)
3037 {
3038 /* No children but file and backing can be explicitly specified (TODO) */
3039 qdict_put(target, "file",
3040 qobject_ref(bs->file->bs->full_open_options));
3041
3042 if (backing_overridden) {
3043 if (bs->backing) {
3044 qdict_put(target, "backing",
3045 qobject_ref(bs->backing->bs->full_open_options));
3046 } else {
3047 qdict_put_null(target, "backing");
3048 }
3049 }
3050 }
3051
3052 static QemuOptsList vmdk_create_opts = {
3053 .name = "vmdk-create-opts",
3054 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
3055 .desc = {
3056 {
3057 .name = BLOCK_OPT_SIZE,
3058 .type = QEMU_OPT_SIZE,
3059 .help = "Virtual disk size"
3060 },
3061 {
3062 .name = BLOCK_OPT_ADAPTER_TYPE,
3063 .type = QEMU_OPT_STRING,
3064 .help = "Virtual adapter type, can be one of "
3065 "ide (default), lsilogic, buslogic or legacyESX"
3066 },
3067 {
3068 .name = BLOCK_OPT_BACKING_FILE,
3069 .type = QEMU_OPT_STRING,
3070 .help = "File name of a base image"
3071 },
3072 {
3073 .name = BLOCK_OPT_BACKING_FMT,
3074 .type = QEMU_OPT_STRING,
3075 .help = "Must be 'vmdk' if present",
3076 },
3077 {
3078 .name = BLOCK_OPT_COMPAT6,
3079 .type = QEMU_OPT_BOOL,
3080 .help = "VMDK version 6 image",
3081 .def_value_str = "off"
3082 },
3083 {
3084 .name = BLOCK_OPT_HWVERSION,
3085 .type = QEMU_OPT_STRING,
3086 .help = "VMDK hardware version",
3087 .def_value_str = "undefined"
3088 },
3089 {
3090 .name = BLOCK_OPT_TOOLSVERSION,
3091 .type = QEMU_OPT_STRING,
3092 .help = "VMware guest tools version",
3093 },
3094 {
3095 .name = BLOCK_OPT_SUBFMT,
3096 .type = QEMU_OPT_STRING,
3097 .help =
3098 "VMDK flat extent format, can be one of "
3099 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
3100 },
3101 {
3102 .name = BLOCK_OPT_ZEROED_GRAIN,
3103 .type = QEMU_OPT_BOOL,
3104 .help = "Enable efficient zero writes "
3105 "using the zeroed-grain GTE feature"
3106 },
3107 { /* end of list */ }
3108 }
3109 };
3110
3111 static BlockDriver bdrv_vmdk = {
3112 .format_name = "vmdk",
3113 .instance_size = sizeof(BDRVVmdkState),
3114 .bdrv_probe = vmdk_probe,
3115 .bdrv_open = vmdk_open,
3116 .bdrv_co_check = vmdk_co_check,
3117 .bdrv_reopen_prepare = vmdk_reopen_prepare,
3118 .bdrv_reopen_commit = vmdk_reopen_commit,
3119 .bdrv_reopen_abort = vmdk_reopen_abort,
3120 .bdrv_child_perm = bdrv_default_perms,
3121 .bdrv_co_preadv = vmdk_co_preadv,
3122 .bdrv_co_pwritev = vmdk_co_pwritev,
3123 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
3124 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
3125 .bdrv_close = vmdk_close,
3126 .bdrv_co_create_opts = vmdk_co_create_opts,
3127 .bdrv_co_create = vmdk_co_create,
3128 .bdrv_co_block_status = vmdk_co_block_status,
3129 .bdrv_co_get_allocated_file_size = vmdk_co_get_allocated_file_size,
3130 .bdrv_has_zero_init = vmdk_has_zero_init,
3131 .bdrv_get_specific_info = vmdk_get_specific_info,
3132 .bdrv_refresh_limits = vmdk_refresh_limits,
3133 .bdrv_co_get_info = vmdk_co_get_info,
3134 .bdrv_gather_child_options = vmdk_gather_child_options,
3135
3136 .is_format = true,
3137 .supports_backing = true,
3138 .create_opts = &vmdk_create_opts,
3139 };
3140
3141 static void bdrv_vmdk_init(void)
3142 {
3143 bdrv_register(&bdrv_vmdk);
3144 }
3145
3146 block_init(bdrv_vmdk_init);
QEmu