?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
migration: fix multifd_send_pages() next channel
[qemu.git] / block / vmdk.c
blob62da4651263bab7fbce5ad0e298b7e192db3297b
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 "migration/blocker.h"
37 #include "qemu/cutils.h"
38 #include <zlib.h>
39
40 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
41 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
42 #define VMDK4_COMPRESSION_DEFLATE 1
43 #define VMDK4_FLAG_NL_DETECT (1 << 0)
44 #define VMDK4_FLAG_RGD (1 << 1)
45 /* Zeroed-grain enable bit */
46 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
47 #define VMDK4_FLAG_COMPRESS (1 << 16)
48 #define VMDK4_FLAG_MARKER (1 << 17)
49 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
50
51 #define VMDK_EXTENT_MAX_SECTORS (1ULL << 32)
52
53 #define VMDK_GTE_ZEROED 0x1
54
55 /* VMDK internal error codes */
56 #define VMDK_OK 0
57 #define VMDK_ERROR (-1)
58 /* Cluster not allocated */
59 #define VMDK_UNALLOC (-2)
60 #define VMDK_ZEROED (-3)
61
62 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
63
64 typedef struct {
65 uint32_t version;
66 uint32_t flags;
67 uint32_t disk_sectors;
68 uint32_t granularity;
69 uint32_t l1dir_offset;
70 uint32_t l1dir_size;
71 uint32_t file_sectors;
72 uint32_t cylinders;
73 uint32_t heads;
74 uint32_t sectors_per_track;
75 } QEMU_PACKED VMDK3Header;
76
77 typedef struct {
78 uint32_t version;
79 uint32_t flags;
80 uint64_t capacity;
81 uint64_t granularity;
82 uint64_t desc_offset;
83 uint64_t desc_size;
84 /* Number of GrainTableEntries per GrainTable */
85 uint32_t num_gtes_per_gt;
86 uint64_t rgd_offset;
87 uint64_t gd_offset;
88 uint64_t grain_offset;
89 char filler[1];
90 char check_bytes[4];
91 uint16_t compressAlgorithm;
92 } QEMU_PACKED VMDK4Header;
93
94 typedef struct VMDKSESparseConstHeader {
95 uint64_t magic;
96 uint64_t version;
97 uint64_t capacity;
98 uint64_t grain_size;
99 uint64_t grain_table_size;
100 uint64_t flags;
101 uint64_t reserved1;
102 uint64_t reserved2;
103 uint64_t reserved3;
104 uint64_t reserved4;
105 uint64_t volatile_header_offset;
106 uint64_t volatile_header_size;
107 uint64_t journal_header_offset;
108 uint64_t journal_header_size;
109 uint64_t journal_offset;
110 uint64_t journal_size;
111 uint64_t grain_dir_offset;
112 uint64_t grain_dir_size;
113 uint64_t grain_tables_offset;
114 uint64_t grain_tables_size;
115 uint64_t free_bitmap_offset;
116 uint64_t free_bitmap_size;
117 uint64_t backmap_offset;
118 uint64_t backmap_size;
119 uint64_t grains_offset;
120 uint64_t grains_size;
121 uint8_t pad[304];
122 } QEMU_PACKED VMDKSESparseConstHeader;
123
124 typedef struct VMDKSESparseVolatileHeader {
125 uint64_t magic;
126 uint64_t free_gt_number;
127 uint64_t next_txn_seq_number;
128 uint64_t replay_journal;
129 uint8_t pad[480];
130 } QEMU_PACKED VMDKSESparseVolatileHeader;
131
132 #define L2_CACHE_SIZE 16
133
134 typedef struct VmdkExtent {
135 BdrvChild *file;
136 bool flat;
137 bool compressed;
138 bool has_marker;
139 bool has_zero_grain;
140 bool sesparse;
141 uint64_t sesparse_l2_tables_offset;
142 uint64_t sesparse_clusters_offset;
143 int32_t entry_size;
144 int version;
145 int64_t sectors;
146 int64_t end_sector;
147 int64_t flat_start_offset;
148 int64_t l1_table_offset;
149 int64_t l1_backup_table_offset;
150 void *l1_table;
151 uint32_t *l1_backup_table;
152 unsigned int l1_size;
153 uint32_t l1_entry_sectors;
154
155 unsigned int l2_size;
156 void *l2_cache;
157 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
158 uint32_t l2_cache_counts[L2_CACHE_SIZE];
159
160 int64_t cluster_sectors;
161 int64_t next_cluster_sector;
162 char *type;
163 } VmdkExtent;
164
165 typedef struct BDRVVmdkState {
166 CoMutex lock;
167 uint64_t desc_offset;
168 bool cid_updated;
169 bool cid_checked;
170 uint32_t cid;
171 uint32_t parent_cid;
172 int num_extents;
173 /* Extent array with num_extents entries, ascend ordered by address */
174 VmdkExtent *extents;
175 Error *migration_blocker;
176 char *create_type;
177 } BDRVVmdkState;
178
179 typedef struct VmdkMetaData {
180 unsigned int l1_index;
181 unsigned int l2_index;
182 unsigned int l2_offset;
183 bool new_allocation;
184 uint32_t *l2_cache_entry;
185 } VmdkMetaData;
186
187 typedef struct VmdkGrainMarker {
188 uint64_t lba;
189 uint32_t size;
190 uint8_t data[];
191 } QEMU_PACKED VmdkGrainMarker;
192
193 enum {
194 MARKER_END_OF_STREAM = 0,
195 MARKER_GRAIN_TABLE = 1,
196 MARKER_GRAIN_DIRECTORY = 2,
197 MARKER_FOOTER = 3,
198 };
199
200 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
201 {
202 uint32_t magic;
203
204 if (buf_size < 4) {
205 return 0;
206 }
207 magic = be32_to_cpu(*(uint32_t *)buf);
208 if (magic == VMDK3_MAGIC ||
209 magic == VMDK4_MAGIC) {
210 return 100;
211 } else {
212 const char *p = (const char *)buf;
213 const char *end = p + buf_size;
214 while (p < end) {
215 if (*p == '#') {
216 /* skip comment line */
217 while (p < end && *p != '\n') {
218 p++;
219 }
220 p++;
221 continue;
222 }
223 if (*p == ' ') {
224 while (p < end && *p == ' ') {
225 p++;
226 }
227 /* skip '\r' if windows line endings used. */
228 if (p < end && *p == '\r') {
229 p++;
230 }
231 /* only accept blank lines before 'version=' line */
232 if (p == end || *p != '\n') {
233 return 0;
234 }
235 p++;
236 continue;
237 }
238 if (end - p >= strlen("version=X\n")) {
239 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
240 strncmp("version=2\n", p, strlen("version=2\n")) == 0 ||
241 strncmp("version=3\n", p, strlen("version=3\n")) == 0) {
242 return 100;
243 }
244 }
245 if (end - p >= strlen("version=X\r\n")) {
246 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
247 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0 ||
248 strncmp("version=3\r\n", p, strlen("version=3\r\n")) == 0) {
249 return 100;
250 }
251 }
252 return 0;
253 }
254 return 0;
255 }
256 }
257
258 #define SECTOR_SIZE 512
259 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
260 #define BUF_SIZE 4096
261 #define HEADER_SIZE 512 /* first sector of 512 bytes */
262
263 static void vmdk_free_extents(BlockDriverState *bs)
264 {
265 int i;
266 BDRVVmdkState *s = bs->opaque;
267 VmdkExtent *e;
268
269 for (i = 0; i < s->num_extents; i++) {
270 e = &s->extents[i];
271 g_free(e->l1_table);
272 g_free(e->l2_cache);
273 g_free(e->l1_backup_table);
274 g_free(e->type);
275 if (e->file != bs->file) {
276 bdrv_unref_child(bs, e->file);
277 }
278 }
279 g_free(s->extents);
280 }
281
282 static void vmdk_free_last_extent(BlockDriverState *bs)
283 {
284 BDRVVmdkState *s = bs->opaque;
285
286 if (s->num_extents == 0) {
287 return;
288 }
289 s->num_extents--;
290 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
291 }
292
293 /* Return -ve errno, or 0 on success and write CID into *pcid. */
294 static int vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
295 {
296 char *desc;
297 uint32_t cid;
298 const char *p_name, *cid_str;
299 size_t cid_str_size;
300 BDRVVmdkState *s = bs->opaque;
301 int ret;
302
303 desc = g_malloc0(DESC_SIZE);
304 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
305 if (ret < 0) {
306 goto out;
307 }
308
309 if (parent) {
310 cid_str = "parentCID";
311 cid_str_size = sizeof("parentCID");
312 } else {
313 cid_str = "CID";
314 cid_str_size = sizeof("CID");
315 }
316
317 desc[DESC_SIZE - 1] = '\0';
318 p_name = strstr(desc, cid_str);
319 if (p_name == NULL) {
320 ret = -EINVAL;
321 goto out;
322 }
323 p_name += cid_str_size;
324 if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
325 ret = -EINVAL;
326 goto out;
327 }
328 *pcid = cid;
329 ret = 0;
330
331 out:
332 g_free(desc);
333 return ret;
334 }
335
336 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
337 {
338 char *desc, *tmp_desc;
339 char *p_name, *tmp_str;
340 BDRVVmdkState *s = bs->opaque;
341 int ret = 0;
342
343 desc = g_malloc0(DESC_SIZE);
344 tmp_desc = g_malloc0(DESC_SIZE);
345 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
346 if (ret < 0) {
347 goto out;
348 }
349
350 desc[DESC_SIZE - 1] = '\0';
351 tmp_str = strstr(desc, "parentCID");
352 if (tmp_str == NULL) {
353 ret = -EINVAL;
354 goto out;
355 }
356
357 pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
358 p_name = strstr(desc, "CID");
359 if (p_name != NULL) {
360 p_name += sizeof("CID");
361 snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
362 pstrcat(desc, DESC_SIZE, tmp_desc);
363 }
364
365 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
366
367 out:
368 g_free(desc);
369 g_free(tmp_desc);
370 return ret;
371 }
372
373 static int vmdk_is_cid_valid(BlockDriverState *bs)
374 {
375 BDRVVmdkState *s = bs->opaque;
376 uint32_t cur_pcid;
377
378 if (!s->cid_checked && bs->backing) {
379 BlockDriverState *p_bs = bs->backing->bs;
380
381 if (strcmp(p_bs->drv->format_name, "vmdk")) {
382 /* Backing file is not in vmdk format, so it does not have
383 * a CID, which makes the overlay's parent CID invalid */
384 return 0;
385 }
386
387 if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
388 /* read failure: report as not valid */
389 return 0;
390 }
391 if (s->parent_cid != cur_pcid) {
392 /* CID not valid */
393 return 0;
394 }
395 }
396 s->cid_checked = true;
397 /* CID valid */
398 return 1;
399 }
400
401 /* We have nothing to do for VMDK reopen, stubs just return success */
402 static int vmdk_reopen_prepare(BDRVReopenState *state,
403 BlockReopenQueue *queue, Error **errp)
404 {
405 assert(state != NULL);
406 assert(state->bs != NULL);
407 return 0;
408 }
409
410 static int vmdk_parent_open(BlockDriverState *bs)
411 {
412 char *p_name;
413 char *desc;
414 BDRVVmdkState *s = bs->opaque;
415 int ret;
416
417 desc = g_malloc0(DESC_SIZE + 1);
418 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
419 if (ret < 0) {
420 goto out;
421 }
422 ret = 0;
423
424 p_name = strstr(desc, "parentFileNameHint");
425 if (p_name != NULL) {
426 char *end_name;
427
428 p_name += sizeof("parentFileNameHint") + 1;
429 end_name = strchr(p_name, '\"');
430 if (end_name == NULL) {
431 ret = -EINVAL;
432 goto out;
433 }
434 if ((end_name - p_name) > sizeof(bs->auto_backing_file) - 1) {
435 ret = -EINVAL;
436 goto out;
437 }
438
439 pstrcpy(bs->auto_backing_file, end_name - p_name + 1, p_name);
440 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
441 bs->auto_backing_file);
442 pstrcpy(bs->backing_format, sizeof(bs->backing_format),
443 "vmdk");
444 }
445
446 out:
447 g_free(desc);
448 return ret;
449 }
450
451 /* Create and append extent to the extent array. Return the added VmdkExtent
452 * address. return NULL if allocation failed. */
453 static int vmdk_add_extent(BlockDriverState *bs,
454 BdrvChild *file, bool flat, int64_t sectors,
455 int64_t l1_offset, int64_t l1_backup_offset,
456 uint32_t l1_size,
457 int l2_size, uint64_t cluster_sectors,
458 VmdkExtent **new_extent,
459 Error **errp)
460 {
461 VmdkExtent *extent;
462 BDRVVmdkState *s = bs->opaque;
463 int64_t nb_sectors;
464
465 if (cluster_sectors > 0x200000) {
466 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
467 error_setg(errp, "Invalid granularity, image may be corrupt");
468 return -EFBIG;
469 }
470 if (l1_size > 32 * 1024 * 1024) {
471 /*
472 * Although with big capacity and small l1_entry_sectors, we can get a
473 * big l1_size, we don't want unbounded value to allocate the table.
474 * Limit it to 32M, which is enough to store:
475 * 8TB - for both VMDK3 & VMDK4 with
476 * minimal cluster size: 512B
477 * minimal L2 table size: 512 entries
478 * 8 TB is still more than the maximal value supported for
479 * VMDK3 & VMDK4 which is 2TB.
480 * 64TB - for "ESXi seSparse Extent"
481 * minimal cluster size: 512B (default is 4KB)
482 * L2 table size: 4096 entries (const).
483 * 64TB is more than the maximal value supported for
484 * seSparse VMDKs (which is slightly less than 64TB)
485 */
486 error_setg(errp, "L1 size too big");
487 return -EFBIG;
488 }
489
490 nb_sectors = bdrv_nb_sectors(file->bs);
491 if (nb_sectors < 0) {
492 return nb_sectors;
493 }
494
495 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
496 extent = &s->extents[s->num_extents];
497 s->num_extents++;
498
499 memset(extent, 0, sizeof(VmdkExtent));
500 extent->file = file;
501 extent->flat = flat;
502 extent->sectors = sectors;
503 extent->l1_table_offset = l1_offset;
504 extent->l1_backup_table_offset = l1_backup_offset;
505 extent->l1_size = l1_size;
506 extent->l1_entry_sectors = l2_size * cluster_sectors;
507 extent->l2_size = l2_size;
508 extent->cluster_sectors = flat ? sectors : cluster_sectors;
509 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
510 extent->entry_size = sizeof(uint32_t);
511
512 if (s->num_extents > 1) {
513 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
514 } else {
515 extent->end_sector = extent->sectors;
516 }
517 bs->total_sectors = extent->end_sector;
518 if (new_extent) {
519 *new_extent = extent;
520 }
521 return 0;
522 }
523
524 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
525 Error **errp)
526 {
527 int ret;
528 size_t l1_size;
529 int i;
530
531 /* read the L1 table */
532 l1_size = extent->l1_size * extent->entry_size;
533 extent->l1_table = g_try_malloc(l1_size);
534 if (l1_size && extent->l1_table == NULL) {
535 return -ENOMEM;
536 }
537
538 ret = bdrv_pread(extent->file,
539 extent->l1_table_offset,
540 extent->l1_table,
541 l1_size);
542 if (ret < 0) {
543 bdrv_refresh_filename(extent->file->bs);
544 error_setg_errno(errp, -ret,
545 "Could not read l1 table from extent '%s'",
546 extent->file->bs->filename);
547 goto fail_l1;
548 }
549 for (i = 0; i < extent->l1_size; i++) {
550 if (extent->entry_size == sizeof(uint64_t)) {
551 le64_to_cpus((uint64_t *)extent->l1_table + i);
552 } else {
553 assert(extent->entry_size == sizeof(uint32_t));
554 le32_to_cpus((uint32_t *)extent->l1_table + i);
555 }
556 }
557
558 if (extent->l1_backup_table_offset) {
559 assert(!extent->sesparse);
560 extent->l1_backup_table = g_try_malloc(l1_size);
561 if (l1_size && extent->l1_backup_table == NULL) {
562 ret = -ENOMEM;
563 goto fail_l1;
564 }
565 ret = bdrv_pread(extent->file,
566 extent->l1_backup_table_offset,
567 extent->l1_backup_table,
568 l1_size);
569 if (ret < 0) {
570 bdrv_refresh_filename(extent->file->bs);
571 error_setg_errno(errp, -ret,
572 "Could not read l1 backup table from extent '%s'",
573 extent->file->bs->filename);
574 goto fail_l1b;
575 }
576 for (i = 0; i < extent->l1_size; i++) {
577 le32_to_cpus(&extent->l1_backup_table[i]);
578 }
579 }
580
581 extent->l2_cache =
582 g_malloc(extent->entry_size * extent->l2_size * L2_CACHE_SIZE);
583 return 0;
584 fail_l1b:
585 g_free(extent->l1_backup_table);
586 fail_l1:
587 g_free(extent->l1_table);
588 return ret;
589 }
590
591 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
592 BdrvChild *file,
593 int flags, Error **errp)
594 {
595 int ret;
596 uint32_t magic;
597 VMDK3Header header;
598 VmdkExtent *extent;
599
600 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
601 if (ret < 0) {
602 bdrv_refresh_filename(file->bs);
603 error_setg_errno(errp, -ret,
604 "Could not read header from file '%s'",
605 file->bs->filename);
606 return ret;
607 }
608 ret = vmdk_add_extent(bs, file, false,
609 le32_to_cpu(header.disk_sectors),
610 (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
611 0,
612 le32_to_cpu(header.l1dir_size),
613 4096,
614 le32_to_cpu(header.granularity),
615 &extent,
616 errp);
617 if (ret < 0) {
618 return ret;
619 }
620 ret = vmdk_init_tables(bs, extent, errp);
621 if (ret) {
622 /* free extent allocated by vmdk_add_extent */
623 vmdk_free_last_extent(bs);
624 }
625 return ret;
626 }
627
628 #define SESPARSE_CONST_HEADER_MAGIC UINT64_C(0x00000000cafebabe)
629 #define SESPARSE_VOLATILE_HEADER_MAGIC UINT64_C(0x00000000cafecafe)
630
631 /* Strict checks - format not officially documented */
632 static int check_se_sparse_const_header(VMDKSESparseConstHeader *header,
633 Error **errp)
634 {
635 header->magic = le64_to_cpu(header->magic);
636 header->version = le64_to_cpu(header->version);
637 header->grain_size = le64_to_cpu(header->grain_size);
638 header->grain_table_size = le64_to_cpu(header->grain_table_size);
639 header->flags = le64_to_cpu(header->flags);
640 header->reserved1 = le64_to_cpu(header->reserved1);
641 header->reserved2 = le64_to_cpu(header->reserved2);
642 header->reserved3 = le64_to_cpu(header->reserved3);
643 header->reserved4 = le64_to_cpu(header->reserved4);
644
645 header->volatile_header_offset =
646 le64_to_cpu(header->volatile_header_offset);
647 header->volatile_header_size = le64_to_cpu(header->volatile_header_size);
648
649 header->journal_header_offset = le64_to_cpu(header->journal_header_offset);
650 header->journal_header_size = le64_to_cpu(header->journal_header_size);
651
652 header->journal_offset = le64_to_cpu(header->journal_offset);
653 header->journal_size = le64_to_cpu(header->journal_size);
654
655 header->grain_dir_offset = le64_to_cpu(header->grain_dir_offset);
656 header->grain_dir_size = le64_to_cpu(header->grain_dir_size);
657
658 header->grain_tables_offset = le64_to_cpu(header->grain_tables_offset);
659 header->grain_tables_size = le64_to_cpu(header->grain_tables_size);
660
661 header->free_bitmap_offset = le64_to_cpu(header->free_bitmap_offset);
662 header->free_bitmap_size = le64_to_cpu(header->free_bitmap_size);
663
664 header->backmap_offset = le64_to_cpu(header->backmap_offset);
665 header->backmap_size = le64_to_cpu(header->backmap_size);
666
667 header->grains_offset = le64_to_cpu(header->grains_offset);
668 header->grains_size = le64_to_cpu(header->grains_size);
669
670 if (header->magic != SESPARSE_CONST_HEADER_MAGIC) {
671 error_setg(errp, "Bad const header magic: 0x%016" PRIx64,
672 header->magic);
673 return -EINVAL;
674 }
675
676 if (header->version != 0x0000000200000001) {
677 error_setg(errp, "Unsupported version: 0x%016" PRIx64,
678 header->version);
679 return -ENOTSUP;
680 }
681
682 if (header->grain_size != 8) {
683 error_setg(errp, "Unsupported grain size: %" PRIu64,
684 header->grain_size);
685 return -ENOTSUP;
686 }
687
688 if (header->grain_table_size != 64) {
689 error_setg(errp, "Unsupported grain table size: %" PRIu64,
690 header->grain_table_size);
691 return -ENOTSUP;
692 }
693
694 if (header->flags != 0) {
695 error_setg(errp, "Unsupported flags: 0x%016" PRIx64,
696 header->flags);
697 return -ENOTSUP;
698 }
699
700 if (header->reserved1 != 0 || header->reserved2 != 0 ||
701 header->reserved3 != 0 || header->reserved4 != 0) {
702 error_setg(errp, "Unsupported reserved bits:"
703 " 0x%016" PRIx64 " 0x%016" PRIx64
704 " 0x%016" PRIx64 " 0x%016" PRIx64,
705 header->reserved1, header->reserved2,
706 header->reserved3, header->reserved4);
707 return -ENOTSUP;
708 }
709
710 /* check that padding is 0 */
711 if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
712 error_setg(errp, "Unsupported non-zero const header padding");
713 return -ENOTSUP;
714 }
715
716 return 0;
717 }
718
719 static int check_se_sparse_volatile_header(VMDKSESparseVolatileHeader *header,
720 Error **errp)
721 {
722 header->magic = le64_to_cpu(header->magic);
723 header->free_gt_number = le64_to_cpu(header->free_gt_number);
724 header->next_txn_seq_number = le64_to_cpu(header->next_txn_seq_number);
725 header->replay_journal = le64_to_cpu(header->replay_journal);
726
727 if (header->magic != SESPARSE_VOLATILE_HEADER_MAGIC) {
728 error_setg(errp, "Bad volatile header magic: 0x%016" PRIx64,
729 header->magic);
730 return -EINVAL;
731 }
732
733 if (header->replay_journal) {
734 error_setg(errp, "Image is dirty, Replaying journal not supported");
735 return -ENOTSUP;
736 }
737
738 /* check that padding is 0 */
739 if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
740 error_setg(errp, "Unsupported non-zero volatile header padding");
741 return -ENOTSUP;
742 }
743
744 return 0;
745 }
746
747 static int vmdk_open_se_sparse(BlockDriverState *bs,
748 BdrvChild *file,
749 int flags, Error **errp)
750 {
751 int ret;
752 VMDKSESparseConstHeader const_header;
753 VMDKSESparseVolatileHeader volatile_header;
754 VmdkExtent *extent;
755
756 ret = bdrv_apply_auto_read_only(bs,
757 "No write support for seSparse images available", errp);
758 if (ret < 0) {
759 return ret;
760 }
761
762 assert(sizeof(const_header) == SECTOR_SIZE);
763
764 ret = bdrv_pread(file, 0, &const_header, sizeof(const_header));
765 if (ret < 0) {
766 bdrv_refresh_filename(file->bs);
767 error_setg_errno(errp, -ret,
768 "Could not read const header from file '%s'",
769 file->bs->filename);
770 return ret;
771 }
772
773 /* check const header */
774 ret = check_se_sparse_const_header(&const_header, errp);
775 if (ret < 0) {
776 return ret;
777 }
778
779 assert(sizeof(volatile_header) == SECTOR_SIZE);
780
781 ret = bdrv_pread(file,
782 const_header.volatile_header_offset * SECTOR_SIZE,
783 &volatile_header, sizeof(volatile_header));
784 if (ret < 0) {
785 bdrv_refresh_filename(file->bs);
786 error_setg_errno(errp, -ret,
787 "Could not read volatile header from file '%s'",
788 file->bs->filename);
789 return ret;
790 }
791
792 /* check volatile header */
793 ret = check_se_sparse_volatile_header(&volatile_header, errp);
794 if (ret < 0) {
795 return ret;
796 }
797
798 ret = vmdk_add_extent(bs, file, false,
799 const_header.capacity,
800 const_header.grain_dir_offset * SECTOR_SIZE,
801 0,
802 const_header.grain_dir_size *
803 SECTOR_SIZE / sizeof(uint64_t),
804 const_header.grain_table_size *
805 SECTOR_SIZE / sizeof(uint64_t),
806 const_header.grain_size,
807 &extent,
808 errp);
809 if (ret < 0) {
810 return ret;
811 }
812
813 extent->sesparse = true;
814 extent->sesparse_l2_tables_offset = const_header.grain_tables_offset;
815 extent->sesparse_clusters_offset = const_header.grains_offset;
816 extent->entry_size = sizeof(uint64_t);
817
818 ret = vmdk_init_tables(bs, extent, errp);
819 if (ret) {
820 /* free extent allocated by vmdk_add_extent */
821 vmdk_free_last_extent(bs);
822 }
823
824 return ret;
825 }
826
827 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
828 QDict *options, Error **errp);
829
830 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
831 {
832 int64_t size;
833 char *buf;
834 int ret;
835
836 size = bdrv_getlength(file->bs);
837 if (size < 0) {
838 error_setg_errno(errp, -size, "Could not access file");
839 return NULL;
840 }
841
842 if (size < 4) {
843 /* Both descriptor file and sparse image must be much larger than 4
844 * bytes, also callers of vmdk_read_desc want to compare the first 4
845 * bytes with VMDK4_MAGIC, let's error out if less is read. */
846 error_setg(errp, "File is too small, not a valid image");
847 return NULL;
848 }
849
850 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */
851 buf = g_malloc(size + 1);
852
853 ret = bdrv_pread(file, desc_offset, buf, size);
854 if (ret < 0) {
855 error_setg_errno(errp, -ret, "Could not read from file");
856 g_free(buf);
857 return NULL;
858 }
859 buf[ret] = 0;
860
861 return buf;
862 }
863
864 static int vmdk_open_vmdk4(BlockDriverState *bs,
865 BdrvChild *file,
866 int flags, QDict *options, Error **errp)
867 {
868 int ret;
869 uint32_t magic;
870 uint32_t l1_size, l1_entry_sectors;
871 VMDK4Header header;
872 VmdkExtent *extent;
873 BDRVVmdkState *s = bs->opaque;
874 int64_t l1_backup_offset = 0;
875 bool compressed;
876
877 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
878 if (ret < 0) {
879 bdrv_refresh_filename(file->bs);
880 error_setg_errno(errp, -ret,
881 "Could not read header from file '%s'",
882 file->bs->filename);
883 return -EINVAL;
884 }
885 if (header.capacity == 0) {
886 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
887 if (desc_offset) {
888 char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
889 if (!buf) {
890 return -EINVAL;
891 }
892 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
893 g_free(buf);
894 return ret;
895 }
896 }
897
898 if (!s->create_type) {
899 s->create_type = g_strdup("monolithicSparse");
900 }
901
902 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
903 /*
904 * The footer takes precedence over the header, so read it in. The
905 * footer starts at offset -1024 from the end: One sector for the
906 * footer, and another one for the end-of-stream marker.
907 */
908 struct {
909 struct {
910 uint64_t val;
911 uint32_t size;
912 uint32_t type;
913 uint8_t pad[512 - 16];
914 } QEMU_PACKED footer_marker;
915
916 uint32_t magic;
917 VMDK4Header header;
918 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
919
920 struct {
921 uint64_t val;
922 uint32_t size;
923 uint32_t type;
924 uint8_t pad[512 - 16];
925 } QEMU_PACKED eos_marker;
926 } QEMU_PACKED footer;
927
928 ret = bdrv_pread(file,
929 bs->file->bs->total_sectors * 512 - 1536,
930 &footer, sizeof(footer));
931 if (ret < 0) {
932 error_setg_errno(errp, -ret, "Failed to read footer");
933 return ret;
934 }
935
936 /* Some sanity checks for the footer */
937 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
938 le32_to_cpu(footer.footer_marker.size) != 0 ||
939 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
940 le64_to_cpu(footer.eos_marker.val) != 0 ||
941 le32_to_cpu(footer.eos_marker.size) != 0 ||
942 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
943 {
944 error_setg(errp, "Invalid footer");
945 return -EINVAL;
946 }
947
948 header = footer.header;
949 }
950
951 compressed =
952 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
953 if (le32_to_cpu(header.version) > 3) {
954 error_setg(errp, "Unsupported VMDK version %" PRIu32,
955 le32_to_cpu(header.version));
956 return -ENOTSUP;
957 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
958 !compressed) {
959 /* VMware KB 2064959 explains that version 3 added support for
960 * persistent changed block tracking (CBT), and backup software can
961 * read it as version=1 if it doesn't care about the changed area
962 * information. So we are safe to enable read only. */
963 error_setg(errp, "VMDK version 3 must be read only");
964 return -EINVAL;
965 }
966
967 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
968 error_setg(errp, "L2 table size too big");
969 return -EINVAL;
970 }
971
972 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
973 * le64_to_cpu(header.granularity);
974 if (l1_entry_sectors == 0) {
975 error_setg(errp, "L1 entry size is invalid");
976 return -EINVAL;
977 }
978 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
979 / l1_entry_sectors;
980 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
981 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
982 }
983 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
984 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
985 (int64_t)(le64_to_cpu(header.grain_offset)
986 * BDRV_SECTOR_SIZE));
987 return -EINVAL;
988 }
989
990 ret = vmdk_add_extent(bs, file, false,
991 le64_to_cpu(header.capacity),
992 le64_to_cpu(header.gd_offset) << 9,
993 l1_backup_offset,
994 l1_size,
995 le32_to_cpu(header.num_gtes_per_gt),
996 le64_to_cpu(header.granularity),
997 &extent,
998 errp);
999 if (ret < 0) {
1000 return ret;
1001 }
1002 extent->compressed =
1003 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1004 if (extent->compressed) {
1005 g_free(s->create_type);
1006 s->create_type = g_strdup("streamOptimized");
1007 }
1008 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
1009 extent->version = le32_to_cpu(header.version);
1010 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
1011 ret = vmdk_init_tables(bs, extent, errp);
1012 if (ret) {
1013 /* free extent allocated by vmdk_add_extent */
1014 vmdk_free_last_extent(bs);
1015 }
1016 return ret;
1017 }
1018
1019 /* find an option value out of descriptor file */
1020 static int vmdk_parse_description(const char *desc, const char *opt_name,
1021 char *buf, int buf_size)
1022 {
1023 char *opt_pos, *opt_end;
1024 const char *end = desc + strlen(desc);
1025
1026 opt_pos = strstr(desc, opt_name);
1027 if (!opt_pos) {
1028 return VMDK_ERROR;
1029 }
1030 /* Skip "=\"" following opt_name */
1031 opt_pos += strlen(opt_name) + 2;
1032 if (opt_pos >= end) {
1033 return VMDK_ERROR;
1034 }
1035 opt_end = opt_pos;
1036 while (opt_end < end && *opt_end != '"') {
1037 opt_end++;
1038 }
1039 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
1040 return VMDK_ERROR;
1041 }
1042 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
1043 return VMDK_OK;
1044 }
1045
1046 /* Open an extent file and append to bs array */
1047 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
1048 char *buf, QDict *options, Error **errp)
1049 {
1050 uint32_t magic;
1051
1052 magic = ldl_be_p(buf);
1053 switch (magic) {
1054 case VMDK3_MAGIC:
1055 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
1056 break;
1057 case VMDK4_MAGIC:
1058 return vmdk_open_vmdk4(bs, file, flags, options, errp);
1059 break;
1060 default:
1061 error_setg(errp, "Image not in VMDK format");
1062 return -EINVAL;
1063 break;
1064 }
1065 }
1066
1067 static const char *next_line(const char *s)
1068 {
1069 while (*s) {
1070 if (*s == '\n') {
1071 return s + 1;
1072 }
1073 s++;
1074 }
1075 return s;
1076 }
1077
1078 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
1079 QDict *options, Error **errp)
1080 {
1081 int ret;
1082 int matches;
1083 char access[11];
1084 char type[11];
1085 char fname[512];
1086 const char *p, *np;
1087 int64_t sectors = 0;
1088 int64_t flat_offset;
1089 char *desc_file_dir = NULL;
1090 char *extent_path;
1091 BdrvChild *extent_file;
1092 BdrvChildRole extent_role;
1093 BDRVVmdkState *s = bs->opaque;
1094 VmdkExtent *extent;
1095 char extent_opt_prefix[32];
1096 Error *local_err = NULL;
1097
1098 for (p = desc; *p; p = next_line(p)) {
1099 /* parse extent line in one of below formats:
1100 *
1101 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
1102 * RW [size in sectors] SPARSE "file-name.vmdk"
1103 * RW [size in sectors] VMFS "file-name.vmdk"
1104 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
1105 * RW [size in sectors] SESPARSE "file-name.vmdk"
1106 */
1107 flat_offset = -1;
1108 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
1109 access, &sectors, type, fname, &flat_offset);
1110 if (matches < 4 || strcmp(access, "RW")) {
1111 continue;
1112 } else if (!strcmp(type, "FLAT")) {
1113 if (matches != 5 || flat_offset < 0) {
1114 goto invalid;
1115 }
1116 } else if (!strcmp(type, "VMFS")) {
1117 if (matches == 4) {
1118 flat_offset = 0;
1119 } else {
1120 goto invalid;
1121 }
1122 } else if (matches != 4) {
1123 goto invalid;
1124 }
1125
1126 if (sectors <= 0 ||
1127 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
1128 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") &&
1129 strcmp(type, "SESPARSE")) ||
1130 (strcmp(access, "RW"))) {
1131 continue;
1132 }
1133
1134 if (path_is_absolute(fname)) {
1135 extent_path = g_strdup(fname);
1136 } else {
1137 if (!desc_file_dir) {
1138 desc_file_dir = bdrv_dirname(bs->file->bs, errp);
1139 if (!desc_file_dir) {
1140 bdrv_refresh_filename(bs->file->bs);
1141 error_prepend(errp, "Cannot use relative paths with VMDK "
1142 "descriptor file '%s': ",
1143 bs->file->bs->filename);
1144 ret = -EINVAL;
1145 goto out;
1146 }
1147 }
1148
1149 extent_path = g_strconcat(desc_file_dir, fname, NULL);
1150 }
1151
1152 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
1153 assert(ret < 32);
1154
1155 extent_role = BDRV_CHILD_DATA;
1156 if (strcmp(type, "FLAT") != 0 && strcmp(type, "VMFS") != 0) {
1157 /* non-flat extents have metadata */
1158 extent_role |= BDRV_CHILD_METADATA;
1159 }
1160
1161 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
1162 bs, &child_of_bds, extent_role, false,
1163 &local_err);
1164 g_free(extent_path);
1165 if (local_err) {
1166 error_propagate(errp, local_err);
1167 ret = -EINVAL;
1168 goto out;
1169 }
1170
1171 /* save to extents array */
1172 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
1173 /* FLAT extent */
1174
1175 ret = vmdk_add_extent(bs, extent_file, true, sectors,
1176 0, 0, 0, 0, 0, &extent, errp);
1177 if (ret < 0) {
1178 bdrv_unref_child(bs, extent_file);
1179 goto out;
1180 }
1181 extent->flat_start_offset = flat_offset << 9;
1182 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
1183 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
1184 char *buf = vmdk_read_desc(extent_file, 0, errp);
1185 if (!buf) {
1186 ret = -EINVAL;
1187 } else {
1188 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
1189 options, errp);
1190 }
1191 g_free(buf);
1192 if (ret) {
1193 bdrv_unref_child(bs, extent_file);
1194 goto out;
1195 }
1196 extent = &s->extents[s->num_extents - 1];
1197 } else if (!strcmp(type, "SESPARSE")) {
1198 ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
1199 if (ret) {
1200 bdrv_unref_child(bs, extent_file);
1201 goto out;
1202 }
1203 extent = &s->extents[s->num_extents - 1];
1204 } else {
1205 error_setg(errp, "Unsupported extent type '%s'", type);
1206 bdrv_unref_child(bs, extent_file);
1207 ret = -ENOTSUP;
1208 goto out;
1209 }
1210 extent->type = g_strdup(type);
1211 }
1212
1213 ret = 0;
1214 goto out;
1215
1216 invalid:
1217 np = next_line(p);
1218 assert(np != p);
1219 if (np[-1] == '\n') {
1220 np--;
1221 }
1222 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
1223 ret = -EINVAL;
1224
1225 out:
1226 g_free(desc_file_dir);
1227 return ret;
1228 }
1229
1230 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
1231 QDict *options, Error **errp)
1232 {
1233 int ret;
1234 char ct[128];
1235 BDRVVmdkState *s = bs->opaque;
1236
1237 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
1238 error_setg(errp, "invalid VMDK image descriptor");
1239 ret = -EINVAL;
1240 goto exit;
1241 }
1242 if (strcmp(ct, "monolithicFlat") &&
1243 strcmp(ct, "vmfs") &&
1244 strcmp(ct, "vmfsSparse") &&
1245 strcmp(ct, "seSparse") &&
1246 strcmp(ct, "twoGbMaxExtentSparse") &&
1247 strcmp(ct, "twoGbMaxExtentFlat")) {
1248 error_setg(errp, "Unsupported image type '%s'", ct);
1249 ret = -ENOTSUP;
1250 goto exit;
1251 }
1252 s->create_type = g_strdup(ct);
1253 s->desc_offset = 0;
1254 ret = vmdk_parse_extents(buf, bs, options, errp);
1255 exit:
1256 return ret;
1257 }
1258
1259 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
1260 Error **errp)
1261 {
1262 char *buf;
1263 int ret;
1264 BDRVVmdkState *s = bs->opaque;
1265 uint32_t magic;
1266 Error *local_err = NULL;
1267
1268 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
1269 BDRV_CHILD_IMAGE, false, errp);
1270 if (!bs->file) {
1271 return -EINVAL;
1272 }
1273
1274 buf = vmdk_read_desc(bs->file, 0, errp);
1275 if (!buf) {
1276 return -EINVAL;
1277 }
1278
1279 magic = ldl_be_p(buf);
1280 switch (magic) {
1281 case VMDK3_MAGIC:
1282 case VMDK4_MAGIC:
1283 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
1284 errp);
1285 s->desc_offset = 0x200;
1286 break;
1287 default:
1288 /* No data in the descriptor file */
1289 bs->file->role &= ~BDRV_CHILD_DATA;
1290
1291 /* Must succeed because we have given up permissions if anything */
1292 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1293
1294 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1295 break;
1296 }
1297 if (ret) {
1298 goto fail;
1299 }
1300
1301 /* try to open parent images, if exist */
1302 ret = vmdk_parent_open(bs);
1303 if (ret) {
1304 goto fail;
1305 }
1306 ret = vmdk_read_cid(bs, 0, &s->cid);
1307 if (ret) {
1308 goto fail;
1309 }
1310 ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1311 if (ret) {
1312 goto fail;
1313 }
1314 qemu_co_mutex_init(&s->lock);
1315
1316 /* Disable migration when VMDK images are used */
1317 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1318 "does not support live migration",
1319 bdrv_get_device_or_node_name(bs));
1320 ret = migrate_add_blocker(s->migration_blocker, &local_err);
1321 if (local_err) {
1322 error_propagate(errp, local_err);
1323 error_free(s->migration_blocker);
1324 goto fail;
1325 }
1326
1327 g_free(buf);
1328 return 0;
1329
1330 fail:
1331 g_free(buf);
1332 g_free(s->create_type);
1333 s->create_type = NULL;
1334 vmdk_free_extents(bs);
1335 return ret;
1336 }
1337
1338
1339 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1340 {
1341 BDRVVmdkState *s = bs->opaque;
1342 int i;
1343
1344 for (i = 0; i < s->num_extents; i++) {
1345 if (!s->extents[i].flat) {
1346 bs->bl.pwrite_zeroes_alignment =
1347 MAX(bs->bl.pwrite_zeroes_alignment,
1348 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1349 }
1350 }
1351 }
1352
1353 /**
1354 * get_whole_cluster
1355 *
1356 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1357 * to the cluster at @cluster_sector_num. If @zeroed is true, we're overwriting
1358 * a zeroed cluster in the current layer and must not copy data from the
1359 * backing file.
1360 *
1361 * If @skip_start_sector < @skip_end_sector, the relative range
1362 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1363 * it for call to write user data in the request.
1364 */
1365 static int get_whole_cluster(BlockDriverState *bs,
1366 VmdkExtent *extent,
1367 uint64_t cluster_offset,
1368 uint64_t offset,
1369 uint64_t skip_start_bytes,
1370 uint64_t skip_end_bytes,
1371 bool zeroed)
1372 {
1373 int ret = VMDK_OK;
1374 int64_t cluster_bytes;
1375 uint8_t *whole_grain;
1376 bool copy_from_backing;
1377
1378 /* For COW, align request sector_num to cluster start */
1379 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1380 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1381 whole_grain = qemu_blockalign(bs, cluster_bytes);
1382 copy_from_backing = bs->backing && !zeroed;
1383
1384 if (!copy_from_backing) {
1385 memset(whole_grain, 0, skip_start_bytes);
1386 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1387 }
1388
1389 assert(skip_end_bytes <= cluster_bytes);
1390 /* we will be here if it's first write on non-exist grain(cluster).
1391 * try to read from parent image, if exist */
1392 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1393 ret = VMDK_ERROR;
1394 goto exit;
1395 }
1396
1397 /* Read backing data before skip range */
1398 if (skip_start_bytes > 0) {
1399 if (copy_from_backing) {
1400 /* qcow2 emits this on bs->file instead of bs->backing */
1401 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1402 ret = bdrv_pread(bs->backing, offset, whole_grain,
1403 skip_start_bytes);
1404 if (ret < 0) {
1405 ret = VMDK_ERROR;
1406 goto exit;
1407 }
1408 }
1409 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1410 ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1411 skip_start_bytes);
1412 if (ret < 0) {
1413 ret = VMDK_ERROR;
1414 goto exit;
1415 }
1416 }
1417 /* Read backing data after skip range */
1418 if (skip_end_bytes < cluster_bytes) {
1419 if (copy_from_backing) {
1420 /* qcow2 emits this on bs->file instead of bs->backing */
1421 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1422 ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1423 whole_grain + skip_end_bytes,
1424 cluster_bytes - skip_end_bytes);
1425 if (ret < 0) {
1426 ret = VMDK_ERROR;
1427 goto exit;
1428 }
1429 }
1430 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1431 ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1432 whole_grain + skip_end_bytes,
1433 cluster_bytes - skip_end_bytes);
1434 if (ret < 0) {
1435 ret = VMDK_ERROR;
1436 goto exit;
1437 }
1438 }
1439
1440 ret = VMDK_OK;
1441 exit:
1442 qemu_vfree(whole_grain);
1443 return ret;
1444 }
1445
1446 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1447 uint32_t offset)
1448 {
1449 offset = cpu_to_le32(offset);
1450 /* update L2 table */
1451 BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1452 if (bdrv_pwrite(extent->file,
1453 ((int64_t)m_data->l2_offset * 512)
1454 + (m_data->l2_index * sizeof(offset)),
1455 &offset, sizeof(offset)) < 0) {
1456 return VMDK_ERROR;
1457 }
1458 /* update backup L2 table */
1459 if (extent->l1_backup_table_offset != 0) {
1460 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1461 if (bdrv_pwrite(extent->file,
1462 ((int64_t)m_data->l2_offset * 512)
1463 + (m_data->l2_index * sizeof(offset)),
1464 &offset, sizeof(offset)) < 0) {
1465 return VMDK_ERROR;
1466 }
1467 }
1468 if (bdrv_flush(extent->file->bs) < 0) {
1469 return VMDK_ERROR;
1470 }
1471 if (m_data->l2_cache_entry) {
1472 *m_data->l2_cache_entry = offset;
1473 }
1474
1475 return VMDK_OK;
1476 }
1477
1478 /**
1479 * get_cluster_offset
1480 *
1481 * Look up cluster offset in extent file by sector number, and store in
1482 * @cluster_offset.
1483 *
1484 * For flat extents, the start offset as parsed from the description file is
1485 * returned.
1486 *
1487 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1488 * offset for a new cluster and update L2 cache. If there is a backing file,
1489 * COW is done before returning; otherwise, zeroes are written to the allocated
1490 * cluster. Both COW and zero writing skips the sector range
1491 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1492 * has new data to write there.
1493 *
1494 * Returns: VMDK_OK if cluster exists and mapped in the image.
1495 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1496 * VMDK_ERROR if failed.
1497 */
1498 static int get_cluster_offset(BlockDriverState *bs,
1499 VmdkExtent *extent,
1500 VmdkMetaData *m_data,
1501 uint64_t offset,
1502 bool allocate,
1503 uint64_t *cluster_offset,
1504 uint64_t skip_start_bytes,
1505 uint64_t skip_end_bytes)
1506 {
1507 unsigned int l1_index, l2_offset, l2_index;
1508 int min_index, i, j;
1509 uint32_t min_count;
1510 void *l2_table;
1511 bool zeroed = false;
1512 int64_t ret;
1513 int64_t cluster_sector;
1514 unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
1515
1516 if (m_data) {
1517 m_data->new_allocation = false;
1518 }
1519 if (extent->flat) {
1520 *cluster_offset = extent->flat_start_offset;
1521 return VMDK_OK;
1522 }
1523
1524 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1525 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1526 if (l1_index >= extent->l1_size) {
1527 return VMDK_ERROR;
1528 }
1529 if (extent->sesparse) {
1530 uint64_t l2_offset_u64;
1531
1532 assert(extent->entry_size == sizeof(uint64_t));
1533
1534 l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
1535 if (l2_offset_u64 == 0) {
1536 l2_offset = 0;
1537 } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
1538 /*
1539 * Top most nibble is 0x1 if grain table is allocated.
1540 * strict check - top most 4 bytes must be 0x10000000 since max
1541 * supported size is 64TB for disk - so no more than 64TB / 16MB
1542 * grain directories which is smaller than uint32,
1543 * where 16MB is the only supported default grain table coverage.
1544 */
1545 return VMDK_ERROR;
1546 } else {
1547 l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
1548 l2_offset_u64 = extent->sesparse_l2_tables_offset +
1549 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
1550 if (l2_offset_u64 > 0x00000000ffffffff) {
1551 return VMDK_ERROR;
1552 }
1553 l2_offset = (unsigned int)(l2_offset_u64);
1554 }
1555 } else {
1556 assert(extent->entry_size == sizeof(uint32_t));
1557 l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
1558 }
1559 if (!l2_offset) {
1560 return VMDK_UNALLOC;
1561 }
1562 for (i = 0; i < L2_CACHE_SIZE; i++) {
1563 if (l2_offset == extent->l2_cache_offsets[i]) {
1564 /* increment the hit count */
1565 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1566 for (j = 0; j < L2_CACHE_SIZE; j++) {
1567 extent->l2_cache_counts[j] >>= 1;
1568 }
1569 }
1570 l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
1571 goto found;
1572 }
1573 }
1574 /* not found: load a new entry in the least used one */
1575 min_index = 0;
1576 min_count = 0xffffffff;
1577 for (i = 0; i < L2_CACHE_SIZE; i++) {
1578 if (extent->l2_cache_counts[i] < min_count) {
1579 min_count = extent->l2_cache_counts[i];
1580 min_index = i;
1581 }
1582 }
1583 l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
1584 BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1585 if (bdrv_pread(extent->file,
1586 (int64_t)l2_offset * 512,
1587 l2_table,
1588 l2_size_bytes
1589 ) != l2_size_bytes) {
1590 return VMDK_ERROR;
1591 }
1592
1593 extent->l2_cache_offsets[min_index] = l2_offset;
1594 extent->l2_cache_counts[min_index] = 1;
1595 found:
1596 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1597 if (m_data) {
1598 m_data->l1_index = l1_index;
1599 m_data->l2_index = l2_index;
1600 m_data->l2_offset = l2_offset;
1601 m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
1602 }
1603
1604 if (extent->sesparse) {
1605 cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
1606 switch (cluster_sector & 0xf000000000000000) {
1607 case 0x0000000000000000:
1608 /* unallocated grain */
1609 if (cluster_sector != 0) {
1610 return VMDK_ERROR;
1611 }
1612 break;
1613 case 0x1000000000000000:
1614 /* scsi-unmapped grain - fallthrough */
1615 case 0x2000000000000000:
1616 /* zero grain */
1617 zeroed = true;
1618 break;
1619 case 0x3000000000000000:
1620 /* allocated grain */
1621 cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
1622 ((cluster_sector & 0x0000ffffffffffff) << 12));
1623 cluster_sector = extent->sesparse_clusters_offset +
1624 cluster_sector * extent->cluster_sectors;
1625 break;
1626 default:
1627 return VMDK_ERROR;
1628 }
1629 } else {
1630 cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
1631
1632 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1633 zeroed = true;
1634 }
1635 }
1636
1637 if (!cluster_sector || zeroed) {
1638 if (!allocate) {
1639 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1640 }
1641 assert(!extent->sesparse);
1642
1643 if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1644 return VMDK_ERROR;
1645 }
1646
1647 cluster_sector = extent->next_cluster_sector;
1648 extent->next_cluster_sector += extent->cluster_sectors;
1649
1650 /* First of all we write grain itself, to avoid race condition
1651 * that may to corrupt the image.
1652 * This problem may occur because of insufficient space on host disk
1653 * or inappropriate VM shutdown.
1654 */
1655 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1656 offset, skip_start_bytes, skip_end_bytes,
1657 zeroed);
1658 if (ret) {
1659 return ret;
1660 }
1661 if (m_data) {
1662 m_data->new_allocation = true;
1663 }
1664 }
1665 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1666 return VMDK_OK;
1667 }
1668
1669 static VmdkExtent *find_extent(BDRVVmdkState *s,
1670 int64_t sector_num, VmdkExtent *start_hint)
1671 {
1672 VmdkExtent *extent = start_hint;
1673
1674 if (!extent) {
1675 extent = &s->extents[0];
1676 }
1677 while (extent < &s->extents[s->num_extents]) {
1678 if (sector_num < extent->end_sector) {
1679 return extent;
1680 }
1681 extent++;
1682 }
1683 return NULL;
1684 }
1685
1686 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1687 int64_t offset)
1688 {
1689 uint64_t extent_begin_offset, extent_relative_offset;
1690 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1691
1692 extent_begin_offset =
1693 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1694 extent_relative_offset = offset - extent_begin_offset;
1695 return extent_relative_offset % cluster_size;
1696 }
1697
1698 static int coroutine_fn vmdk_co_block_status(BlockDriverState *bs,
1699 bool want_zero,
1700 int64_t offset, int64_t bytes,
1701 int64_t *pnum, int64_t *map,
1702 BlockDriverState **file)
1703 {
1704 BDRVVmdkState *s = bs->opaque;
1705 int64_t index_in_cluster, n, ret;
1706 uint64_t cluster_offset;
1707 VmdkExtent *extent;
1708
1709 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1710 if (!extent) {
1711 return -EIO;
1712 }
1713 qemu_co_mutex_lock(&s->lock);
1714 ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1715 0, 0);
1716 qemu_co_mutex_unlock(&s->lock);
1717
1718 index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1719 switch (ret) {
1720 case VMDK_ERROR:
1721 ret = -EIO;
1722 break;
1723 case VMDK_UNALLOC:
1724 ret = 0;
1725 break;
1726 case VMDK_ZEROED:
1727 ret = BDRV_BLOCK_ZERO;
1728 break;
1729 case VMDK_OK:
1730 ret = BDRV_BLOCK_DATA;
1731 if (!extent->compressed) {
1732 ret |= BDRV_BLOCK_OFFSET_VALID;
1733 *map = cluster_offset + index_in_cluster;
1734 if (extent->flat) {
1735 ret |= BDRV_BLOCK_RECURSE;
1736 }
1737 }
1738 *file = extent->file->bs;
1739 break;
1740 }
1741
1742 n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1743 *pnum = MIN(n, bytes);
1744 return ret;
1745 }
1746
1747 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1748 int64_t offset_in_cluster, QEMUIOVector *qiov,
1749 uint64_t qiov_offset, uint64_t n_bytes,
1750 uint64_t offset)
1751 {
1752 int ret;
1753 VmdkGrainMarker *data = NULL;
1754 uLongf buf_len;
1755 QEMUIOVector local_qiov;
1756 int64_t write_offset;
1757 int64_t write_end_sector;
1758
1759 if (extent->compressed) {
1760 void *compressed_data;
1761
1762 /* Only whole clusters */
1763 if (offset_in_cluster ||
1764 n_bytes > (extent->cluster_sectors * SECTOR_SIZE) ||
1765 (n_bytes < (extent->cluster_sectors * SECTOR_SIZE) &&
1766 offset + n_bytes != extent->end_sector * SECTOR_SIZE))
1767 {
1768 ret = -EINVAL;
1769 goto out;
1770 }
1771
1772 if (!extent->has_marker) {
1773 ret = -EINVAL;
1774 goto out;
1775 }
1776 buf_len = (extent->cluster_sectors << 9) * 2;
1777 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1778
1779 compressed_data = g_malloc(n_bytes);
1780 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1781 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1782 g_free(compressed_data);
1783
1784 if (ret != Z_OK || buf_len == 0) {
1785 ret = -EINVAL;
1786 goto out;
1787 }
1788
1789 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1790 data->size = cpu_to_le32(buf_len);
1791
1792 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1793 qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1794
1795 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1796 } else {
1797 qemu_iovec_init(&local_qiov, qiov->niov);
1798 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1799
1800 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1801 }
1802
1803 write_offset = cluster_offset + offset_in_cluster;
1804 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1805 &local_qiov, 0);
1806
1807 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1808
1809 if (extent->compressed) {
1810 extent->next_cluster_sector = write_end_sector;
1811 } else {
1812 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1813 write_end_sector);
1814 }
1815
1816 if (ret < 0) {
1817 goto out;
1818 }
1819 ret = 0;
1820 out:
1821 g_free(data);
1822 if (!extent->compressed) {
1823 qemu_iovec_destroy(&local_qiov);
1824 }
1825 return ret;
1826 }
1827
1828 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1829 int64_t offset_in_cluster, QEMUIOVector *qiov,
1830 int bytes)
1831 {
1832 int ret;
1833 int cluster_bytes, buf_bytes;
1834 uint8_t *cluster_buf, *compressed_data;
1835 uint8_t *uncomp_buf;
1836 uint32_t data_len;
1837 VmdkGrainMarker *marker;
1838 uLongf buf_len;
1839
1840
1841 if (!extent->compressed) {
1842 BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1843 ret = bdrv_co_preadv(extent->file,
1844 cluster_offset + offset_in_cluster, bytes,
1845 qiov, 0);
1846 if (ret < 0) {
1847 return ret;
1848 }
1849 return 0;
1850 }
1851 cluster_bytes = extent->cluster_sectors * 512;
1852 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1853 buf_bytes = cluster_bytes * 2;
1854 cluster_buf = g_malloc(buf_bytes);
1855 uncomp_buf = g_malloc(cluster_bytes);
1856 BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1857 ret = bdrv_pread(extent->file,
1858 cluster_offset,
1859 cluster_buf, buf_bytes);
1860 if (ret < 0) {
1861 goto out;
1862 }
1863 compressed_data = cluster_buf;
1864 buf_len = cluster_bytes;
1865 data_len = cluster_bytes;
1866 if (extent->has_marker) {
1867 marker = (VmdkGrainMarker *)cluster_buf;
1868 compressed_data = marker->data;
1869 data_len = le32_to_cpu(marker->size);
1870 }
1871 if (!data_len || data_len > buf_bytes) {
1872 ret = -EINVAL;
1873 goto out;
1874 }
1875 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1876 if (ret != Z_OK) {
1877 ret = -EINVAL;
1878 goto out;
1879
1880 }
1881 if (offset_in_cluster < 0 ||
1882 offset_in_cluster + bytes > buf_len) {
1883 ret = -EINVAL;
1884 goto out;
1885 }
1886 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1887 ret = 0;
1888
1889 out:
1890 g_free(uncomp_buf);
1891 g_free(cluster_buf);
1892 return ret;
1893 }
1894
1895 static int coroutine_fn
1896 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1897 QEMUIOVector *qiov, int flags)
1898 {
1899 BDRVVmdkState *s = bs->opaque;
1900 int ret;
1901 uint64_t n_bytes, offset_in_cluster;
1902 VmdkExtent *extent = NULL;
1903 QEMUIOVector local_qiov;
1904 uint64_t cluster_offset;
1905 uint64_t bytes_done = 0;
1906
1907 qemu_iovec_init(&local_qiov, qiov->niov);
1908 qemu_co_mutex_lock(&s->lock);
1909
1910 while (bytes > 0) {
1911 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1912 if (!extent) {
1913 ret = -EIO;
1914 goto fail;
1915 }
1916 ret = get_cluster_offset(bs, extent, NULL,
1917 offset, false, &cluster_offset, 0, 0);
1918 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1919
1920 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1921 - offset_in_cluster);
1922
1923 if (ret != VMDK_OK) {
1924 /* if not allocated, try to read from parent image, if exist */
1925 if (bs->backing && ret != VMDK_ZEROED) {
1926 if (!vmdk_is_cid_valid(bs)) {
1927 ret = -EINVAL;
1928 goto fail;
1929 }
1930
1931 qemu_iovec_reset(&local_qiov);
1932 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1933
1934 /* qcow2 emits this on bs->file instead of bs->backing */
1935 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1936 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1937 &local_qiov, 0);
1938 if (ret < 0) {
1939 goto fail;
1940 }
1941 } else {
1942 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1943 }
1944 } else {
1945 qemu_iovec_reset(&local_qiov);
1946 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1947
1948 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1949 &local_qiov, n_bytes);
1950 if (ret) {
1951 goto fail;
1952 }
1953 }
1954 bytes -= n_bytes;
1955 offset += n_bytes;
1956 bytes_done += n_bytes;
1957 }
1958
1959 ret = 0;
1960 fail:
1961 qemu_co_mutex_unlock(&s->lock);
1962 qemu_iovec_destroy(&local_qiov);
1963
1964 return ret;
1965 }
1966
1967 /**
1968 * vmdk_write:
1969 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1970 * if possible, otherwise return -ENOTSUP.
1971 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1972 * with each cluster. By dry run we can find if the zero write
1973 * is possible without modifying image data.
1974 *
1975 * Returns: error code with 0 for success.
1976 */
1977 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1978 uint64_t bytes, QEMUIOVector *qiov,
1979 bool zeroed, bool zero_dry_run)
1980 {
1981 BDRVVmdkState *s = bs->opaque;
1982 VmdkExtent *extent = NULL;
1983 int ret;
1984 int64_t offset_in_cluster, n_bytes;
1985 uint64_t cluster_offset;
1986 uint64_t bytes_done = 0;
1987 VmdkMetaData m_data;
1988
1989 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1990 error_report("Wrong offset: offset=0x%" PRIx64
1991 " total_sectors=0x%" PRIx64,
1992 offset, bs->total_sectors);
1993 return -EIO;
1994 }
1995
1996 while (bytes > 0) {
1997 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1998 if (!extent) {
1999 return -EIO;
2000 }
2001 if (extent->sesparse) {
2002 return -ENOTSUP;
2003 }
2004 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
2005 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
2006 - offset_in_cluster);
2007
2008 ret = get_cluster_offset(bs, extent, &m_data, offset,
2009 !(extent->compressed || zeroed),
2010 &cluster_offset, offset_in_cluster,
2011 offset_in_cluster + n_bytes);
2012 if (extent->compressed) {
2013 if (ret == VMDK_OK) {
2014 /* Refuse write to allocated cluster for streamOptimized */
2015 error_report("Could not write to allocated cluster"
2016 " for streamOptimized");
2017 return -EIO;
2018 } else if (!zeroed) {
2019 /* allocate */
2020 ret = get_cluster_offset(bs, extent, &m_data, offset,
2021 true, &cluster_offset, 0, 0);
2022 }
2023 }
2024 if (ret == VMDK_ERROR) {
2025 return -EINVAL;
2026 }
2027 if (zeroed) {
2028 /* Do zeroed write, buf is ignored */
2029 if (extent->has_zero_grain &&
2030 offset_in_cluster == 0 &&
2031 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
2032 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
2033 if (!zero_dry_run && ret != VMDK_ZEROED) {
2034 /* update L2 tables */
2035 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
2036 != VMDK_OK) {
2037 return -EIO;
2038 }
2039 }
2040 } else {
2041 return -ENOTSUP;
2042 }
2043 } else {
2044 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
2045 qiov, bytes_done, n_bytes, offset);
2046 if (ret) {
2047 return ret;
2048 }
2049 if (m_data.new_allocation) {
2050 /* update L2 tables */
2051 if (vmdk_L2update(extent, &m_data,
2052 cluster_offset >> BDRV_SECTOR_BITS)
2053 != VMDK_OK) {
2054 return -EIO;
2055 }
2056 }
2057 }
2058 bytes -= n_bytes;
2059 offset += n_bytes;
2060 bytes_done += n_bytes;
2061
2062 /* update CID on the first write every time the virtual disk is
2063 * opened */
2064 if (!s->cid_updated) {
2065 ret = vmdk_write_cid(bs, g_random_int());
2066 if (ret < 0) {
2067 return ret;
2068 }
2069 s->cid_updated = true;
2070 }
2071 }
2072 return 0;
2073 }
2074
2075 static int coroutine_fn
2076 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2077 QEMUIOVector *qiov, int flags)
2078 {
2079 int ret;
2080 BDRVVmdkState *s = bs->opaque;
2081 qemu_co_mutex_lock(&s->lock);
2082 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
2083 qemu_co_mutex_unlock(&s->lock);
2084 return ret;
2085 }
2086
2087 static int coroutine_fn
2088 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
2089 uint64_t bytes, QEMUIOVector *qiov)
2090 {
2091 if (bytes == 0) {
2092 /* The caller will write bytes 0 to signal EOF.
2093 * When receive it, we align EOF to a sector boundary. */
2094 BDRVVmdkState *s = bs->opaque;
2095 int i, ret;
2096 int64_t length;
2097
2098 for (i = 0; i < s->num_extents; i++) {
2099 length = bdrv_getlength(s->extents[i].file->bs);
2100 if (length < 0) {
2101 return length;
2102 }
2103 length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
2104 ret = bdrv_truncate(s->extents[i].file, length, false,
2105 PREALLOC_MODE_OFF, 0, NULL);
2106 if (ret < 0) {
2107 return ret;
2108 }
2109 }
2110 return 0;
2111 }
2112 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
2113 }
2114
2115 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
2116 int64_t offset,
2117 int bytes,
2118 BdrvRequestFlags flags)
2119 {
2120 int ret;
2121 BDRVVmdkState *s = bs->opaque;
2122
2123 qemu_co_mutex_lock(&s->lock);
2124 /* write zeroes could fail if sectors not aligned to cluster, test it with
2125 * dry_run == true before really updating image */
2126 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
2127 if (!ret) {
2128 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
2129 }
2130 qemu_co_mutex_unlock(&s->lock);
2131 return ret;
2132 }
2133
2134 static int vmdk_init_extent(BlockBackend *blk,
2135 int64_t filesize, bool flat,
2136 bool compress, bool zeroed_grain,
2137 Error **errp)
2138 {
2139 int ret, i;
2140 VMDK4Header header;
2141 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
2142 uint32_t *gd_buf = NULL;
2143 int gd_buf_size;
2144
2145 if (flat) {
2146 ret = blk_truncate(blk, filesize, false, PREALLOC_MODE_OFF, 0, errp);
2147 goto exit;
2148 }
2149 magic = cpu_to_be32(VMDK4_MAGIC);
2150 memset(&header, 0, sizeof(header));
2151 if (compress) {
2152 header.version = 3;
2153 } else if (zeroed_grain) {
2154 header.version = 2;
2155 } else {
2156 header.version = 1;
2157 }
2158 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
2159 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
2160 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
2161 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
2162 header.capacity = filesize / BDRV_SECTOR_SIZE;
2163 header.granularity = 128;
2164 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
2165
2166 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
2167 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
2168 BDRV_SECTOR_SIZE);
2169 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
2170 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
2171
2172 header.desc_offset = 1;
2173 header.desc_size = 20;
2174 header.rgd_offset = header.desc_offset + header.desc_size;
2175 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
2176 header.grain_offset =
2177 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
2178 header.granularity);
2179 /* swap endianness for all header fields */
2180 header.version = cpu_to_le32(header.version);
2181 header.flags = cpu_to_le32(header.flags);
2182 header.capacity = cpu_to_le64(header.capacity);
2183 header.granularity = cpu_to_le64(header.granularity);
2184 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
2185 header.desc_offset = cpu_to_le64(header.desc_offset);
2186 header.desc_size = cpu_to_le64(header.desc_size);
2187 header.rgd_offset = cpu_to_le64(header.rgd_offset);
2188 header.gd_offset = cpu_to_le64(header.gd_offset);
2189 header.grain_offset = cpu_to_le64(header.grain_offset);
2190 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
2191
2192 header.check_bytes[0] = 0xa;
2193 header.check_bytes[1] = 0x20;
2194 header.check_bytes[2] = 0xd;
2195 header.check_bytes[3] = 0xa;
2196
2197 /* write all the data */
2198 ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
2199 if (ret < 0) {
2200 error_setg(errp, QERR_IO_ERROR);
2201 goto exit;
2202 }
2203 ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
2204 if (ret < 0) {
2205 error_setg(errp, QERR_IO_ERROR);
2206 goto exit;
2207 }
2208
2209 ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false,
2210 PREALLOC_MODE_OFF, 0, errp);
2211 if (ret < 0) {
2212 goto exit;
2213 }
2214
2215 /* write grain directory */
2216 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
2217 gd_buf = g_malloc0(gd_buf_size);
2218 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
2219 i < gt_count; i++, tmp += gt_size) {
2220 gd_buf[i] = cpu_to_le32(tmp);
2221 }
2222 ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
2223 gd_buf, gd_buf_size, 0);
2224 if (ret < 0) {
2225 error_setg(errp, QERR_IO_ERROR);
2226 goto exit;
2227 }
2228
2229 /* write backup grain directory */
2230 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
2231 i < gt_count; i++, tmp += gt_size) {
2232 gd_buf[i] = cpu_to_le32(tmp);
2233 }
2234 ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
2235 gd_buf, gd_buf_size, 0);
2236 if (ret < 0) {
2237 error_setg(errp, QERR_IO_ERROR);
2238 }
2239
2240 ret = 0;
2241 exit:
2242 g_free(gd_buf);
2243 return ret;
2244 }
2245
2246 static int vmdk_create_extent(const char *filename, int64_t filesize,
2247 bool flat, bool compress, bool zeroed_grain,
2248 BlockBackend **pbb,
2249 QemuOpts *opts, Error **errp)
2250 {
2251 int ret;
2252 BlockBackend *blk = NULL;
2253 Error *local_err = NULL;
2254
2255 ret = bdrv_create_file(filename, opts, &local_err);
2256 if (ret < 0) {
2257 error_propagate(errp, local_err);
2258 goto exit;
2259 }
2260
2261 blk = blk_new_open(filename, NULL, NULL,
2262 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2263 &local_err);
2264 if (blk == NULL) {
2265 error_propagate(errp, local_err);
2266 ret = -EIO;
2267 goto exit;
2268 }
2269
2270 blk_set_allow_write_beyond_eof(blk, true);
2271
2272 ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
2273 exit:
2274 if (blk) {
2275 if (pbb) {
2276 *pbb = blk;
2277 } else {
2278 blk_unref(blk);
2279 blk = NULL;
2280 }
2281 }
2282 return ret;
2283 }
2284
2285 static int filename_decompose(const char *filename, char *path, char *prefix,
2286 char *postfix, size_t buf_len, Error **errp)
2287 {
2288 const char *p, *q;
2289
2290 if (filename == NULL || !strlen(filename)) {
2291 error_setg(errp, "No filename provided");
2292 return VMDK_ERROR;
2293 }
2294 p = strrchr(filename, '/');
2295 if (p == NULL) {
2296 p = strrchr(filename, '\\');
2297 }
2298 if (p == NULL) {
2299 p = strrchr(filename, ':');
2300 }
2301 if (p != NULL) {
2302 p++;
2303 if (p - filename >= buf_len) {
2304 return VMDK_ERROR;
2305 }
2306 pstrcpy(path, p - filename + 1, filename);
2307 } else {
2308 p = filename;
2309 path[0] = '\0';
2310 }
2311 q = strrchr(p, '.');
2312 if (q == NULL) {
2313 pstrcpy(prefix, buf_len, p);
2314 postfix[0] = '\0';
2315 } else {
2316 if (q - p >= buf_len) {
2317 return VMDK_ERROR;
2318 }
2319 pstrcpy(prefix, q - p + 1, p);
2320 pstrcpy(postfix, buf_len, q);
2321 }
2322 return VMDK_OK;
2323 }
2324
2325 /*
2326 * idx == 0: get or create the descriptor file (also the image file if in a
2327 * non-split format.
2328 * idx >= 1: get the n-th extent if in a split subformat
2329 */
2330 typedef BlockBackend *(*vmdk_create_extent_fn)(int64_t size,
2331 int idx,
2332 bool flat,
2333 bool split,
2334 bool compress,
2335 bool zeroed_grain,
2336 void *opaque,
2337 Error **errp);
2338
2339 static void vmdk_desc_add_extent(GString *desc,
2340 const char *extent_line_fmt,
2341 int64_t size, const char *filename)
2342 {
2343 char *basename = g_path_get_basename(filename);
2344
2345 g_string_append_printf(desc, extent_line_fmt,
2346 DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
2347 g_free(basename);
2348 }
2349
2350 static int coroutine_fn vmdk_co_do_create(int64_t size,
2351 BlockdevVmdkSubformat subformat,
2352 BlockdevVmdkAdapterType adapter_type,
2353 const char *backing_file,
2354 const char *hw_version,
2355 bool compat6,
2356 bool zeroed_grain,
2357 vmdk_create_extent_fn extent_fn,
2358 void *opaque,
2359 Error **errp)
2360 {
2361 int extent_idx;
2362 BlockBackend *blk = NULL;
2363 BlockBackend *extent_blk;
2364 Error *local_err = NULL;
2365 char *desc = NULL;
2366 int ret = 0;
2367 bool flat, split, compress;
2368 GString *ext_desc_lines;
2369 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
2370 int64_t extent_size;
2371 int64_t created_size = 0;
2372 const char *extent_line_fmt;
2373 char *parent_desc_line = g_malloc0(BUF_SIZE);
2374 uint32_t parent_cid = 0xffffffff;
2375 uint32_t number_heads = 16;
2376 uint32_t desc_offset = 0, desc_len;
2377 const char desc_template[] =
2378 "# Disk DescriptorFile\n"
2379 "version=1\n"
2380 "CID=%" PRIx32 "\n"
2381 "parentCID=%" PRIx32 "\n"
2382 "createType=\"%s\"\n"
2383 "%s"
2384 "\n"
2385 "# Extent description\n"
2386 "%s"
2387 "\n"
2388 "# The Disk Data Base\n"
2389 "#DDB\n"
2390 "\n"
2391 "ddb.virtualHWVersion = \"%s\"\n"
2392 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2393 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2394 "ddb.geometry.sectors = \"63\"\n"
2395 "ddb.adapterType = \"%s\"\n";
2396
2397 ext_desc_lines = g_string_new(NULL);
2398
2399 /* Read out options */
2400 if (compat6) {
2401 if (hw_version) {
2402 error_setg(errp,
2403 "compat6 cannot be enabled with hwversion set");
2404 ret = -EINVAL;
2405 goto exit;
2406 }
2407 hw_version = "6";
2408 }
2409 if (!hw_version) {
2410 hw_version = "4";
2411 }
2412
2413 if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2414 /* that's the number of heads with which vmware operates when
2415 creating, exporting, etc. vmdk files with a non-ide adapter type */
2416 number_heads = 255;
2417 }
2418 split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2419 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2420 flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2421 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2422 compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2423
2424 if (flat) {
2425 extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2426 } else {
2427 extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2428 }
2429 if (flat && backing_file) {
2430 error_setg(errp, "Flat image can't have backing file");
2431 ret = -ENOTSUP;
2432 goto exit;
2433 }
2434 if (flat && zeroed_grain) {
2435 error_setg(errp, "Flat image can't enable zeroed grain");
2436 ret = -ENOTSUP;
2437 goto exit;
2438 }
2439
2440 /* Create extents */
2441 if (split) {
2442 extent_size = split_size;
2443 } else {
2444 extent_size = size;
2445 }
2446 if (!split && !flat) {
2447 created_size = extent_size;
2448 } else {
2449 created_size = 0;
2450 }
2451 /* Get the descriptor file BDS */
2452 blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2453 opaque, errp);
2454 if (!blk) {
2455 ret = -EIO;
2456 goto exit;
2457 }
2458 if (!split && !flat) {
2459 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2460 blk_bs(blk)->filename);
2461 }
2462
2463 if (backing_file) {
2464 BlockBackend *backing;
2465 char *full_backing =
2466 bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2467 backing_file,
2468 &local_err);
2469 if (local_err) {
2470 error_propagate(errp, local_err);
2471 ret = -ENOENT;
2472 goto exit;
2473 }
2474 assert(full_backing);
2475
2476 backing = blk_new_open(full_backing, NULL, NULL,
2477 BDRV_O_NO_BACKING, errp);
2478 g_free(full_backing);
2479 if (backing == NULL) {
2480 ret = -EIO;
2481 goto exit;
2482 }
2483 if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2484 error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2485 blk_bs(backing)->drv->format_name);
2486 blk_unref(backing);
2487 ret = -EINVAL;
2488 goto exit;
2489 }
2490 ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2491 blk_unref(backing);
2492 if (ret) {
2493 error_setg(errp, "Failed to read parent CID");
2494 goto exit;
2495 }
2496 snprintf(parent_desc_line, BUF_SIZE,
2497 "parentFileNameHint=\"%s\"", backing_file);
2498 }
2499 extent_idx = 1;
2500 while (created_size < size) {
2501 int64_t cur_size = MIN(size - created_size, extent_size);
2502 extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2503 zeroed_grain, opaque, errp);
2504 if (!extent_blk) {
2505 ret = -EINVAL;
2506 goto exit;
2507 }
2508 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2509 blk_bs(extent_blk)->filename);
2510 created_size += cur_size;
2511 extent_idx++;
2512 blk_unref(extent_blk);
2513 }
2514
2515 /* Check whether we got excess extents */
2516 extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2517 opaque, NULL);
2518 if (extent_blk) {
2519 blk_unref(extent_blk);
2520 error_setg(errp, "List of extents contains unused extents");
2521 ret = -EINVAL;
2522 goto exit;
2523 }
2524
2525 /* generate descriptor file */
2526 desc = g_strdup_printf(desc_template,
2527 g_random_int(),
2528 parent_cid,
2529 BlockdevVmdkSubformat_str(subformat),
2530 parent_desc_line,
2531 ext_desc_lines->str,
2532 hw_version,
2533 size /
2534 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2535 number_heads,
2536 BlockdevVmdkAdapterType_str(adapter_type));
2537 desc_len = strlen(desc);
2538 /* the descriptor offset = 0x200 */
2539 if (!split && !flat) {
2540 desc_offset = 0x200;
2541 }
2542
2543 ret = blk_pwrite(blk, desc_offset, desc, desc_len, 0);
2544 if (ret < 0) {
2545 error_setg_errno(errp, -ret, "Could not write description");
2546 goto exit;
2547 }
2548 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2549 * for description file */
2550 if (desc_offset == 0) {
2551 ret = blk_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, 0, errp);
2552 if (ret < 0) {
2553 goto exit;
2554 }
2555 }
2556 ret = 0;
2557 exit:
2558 if (blk) {
2559 blk_unref(blk);
2560 }
2561 g_free(desc);
2562 g_free(parent_desc_line);
2563 g_string_free(ext_desc_lines, true);
2564 return ret;
2565 }
2566
2567 typedef struct {
2568 char *path;
2569 char *prefix;
2570 char *postfix;
2571 QemuOpts *opts;
2572 } VMDKCreateOptsData;
2573
2574 static BlockBackend *vmdk_co_create_opts_cb(int64_t size, int idx,
2575 bool flat, bool split, bool compress,
2576 bool zeroed_grain, void *opaque,
2577 Error **errp)
2578 {
2579 BlockBackend *blk = NULL;
2580 BlockDriverState *bs = NULL;
2581 VMDKCreateOptsData *data = opaque;
2582 char *ext_filename = NULL;
2583 char *rel_filename = NULL;
2584
2585 /* We're done, don't create excess extents. */
2586 if (size == -1) {
2587 assert(errp == NULL);
2588 return NULL;
2589 }
2590
2591 if (idx == 0) {
2592 rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2593 } else if (split) {
2594 rel_filename = g_strdup_printf("%s-%c%03d%s",
2595 data->prefix,
2596 flat ? 'f' : 's', idx, data->postfix);
2597 } else {
2598 assert(idx == 1);
2599 rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2600 }
2601
2602 ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2603 g_free(rel_filename);
2604
2605 if (vmdk_create_extent(ext_filename, size,
2606 flat, compress, zeroed_grain, &blk, data->opts,
2607 errp)) {
2608 goto exit;
2609 }
2610 bdrv_unref(bs);
2611 exit:
2612 g_free(ext_filename);
2613 return blk;
2614 }
2615
2616 static int coroutine_fn vmdk_co_create_opts(BlockDriver *drv,
2617 const char *filename,
2618 QemuOpts *opts,
2619 Error **errp)
2620 {
2621 Error *local_err = NULL;
2622 char *desc = NULL;
2623 int64_t total_size = 0;
2624 char *adapter_type = NULL;
2625 BlockdevVmdkAdapterType adapter_type_enum;
2626 char *backing_file = NULL;
2627 char *hw_version = NULL;
2628 char *fmt = NULL;
2629 BlockdevVmdkSubformat subformat;
2630 int ret = 0;
2631 char *path = g_malloc0(PATH_MAX);
2632 char *prefix = g_malloc0(PATH_MAX);
2633 char *postfix = g_malloc0(PATH_MAX);
2634 char *desc_line = g_malloc0(BUF_SIZE);
2635 char *ext_filename = g_malloc0(PATH_MAX);
2636 char *desc_filename = g_malloc0(PATH_MAX);
2637 char *parent_desc_line = g_malloc0(BUF_SIZE);
2638 bool zeroed_grain;
2639 bool compat6;
2640 VMDKCreateOptsData data;
2641
2642 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
2643 ret = -EINVAL;
2644 goto exit;
2645 }
2646 /* Read out options */
2647 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2648 BDRV_SECTOR_SIZE);
2649 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
2650 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2651 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
2652 compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false);
2653 if (strcmp(hw_version, "undefined") == 0) {
2654 g_free(hw_version);
2655 hw_version = NULL;
2656 }
2657 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
2658 zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false);
2659
2660 if (adapter_type) {
2661 adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup,
2662 adapter_type,
2663 BLOCKDEV_VMDK_ADAPTER_TYPE_IDE,
2664 &local_err);
2665 if (local_err) {
2666 error_propagate(errp, local_err);
2667 ret = -EINVAL;
2668 goto exit;
2669 }
2670 } else {
2671 adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE;
2672 }
2673
2674 if (!fmt) {
2675 /* Default format to monolithicSparse */
2676 subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE;
2677 } else {
2678 subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup,
2679 fmt,
2680 BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE,
2681 &local_err);
2682 if (local_err) {
2683 error_propagate(errp, local_err);
2684 ret = -EINVAL;
2685 goto exit;
2686 }
2687 }
2688 data = (VMDKCreateOptsData){
2689 .prefix = prefix,
2690 .postfix = postfix,
2691 .path = path,
2692 .opts = opts,
2693 };
2694 ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum,
2695 backing_file, hw_version, compat6, zeroed_grain,
2696 vmdk_co_create_opts_cb, &data, errp);
2697
2698 exit:
2699 g_free(adapter_type);
2700 g_free(backing_file);
2701 g_free(hw_version);
2702 g_free(fmt);
2703 g_free(desc);
2704 g_free(path);
2705 g_free(prefix);
2706 g_free(postfix);
2707 g_free(desc_line);
2708 g_free(ext_filename);
2709 g_free(desc_filename);
2710 g_free(parent_desc_line);
2711 return ret;
2712 }
2713
2714 static BlockBackend *vmdk_co_create_cb(int64_t size, int idx,
2715 bool flat, bool split, bool compress,
2716 bool zeroed_grain, void *opaque,
2717 Error **errp)
2718 {
2719 int ret;
2720 BlockDriverState *bs;
2721 BlockBackend *blk;
2722 BlockdevCreateOptionsVmdk *opts = opaque;
2723
2724 if (idx == 0) {
2725 bs = bdrv_open_blockdev_ref(opts->file, errp);
2726 } else {
2727 int i;
2728 BlockdevRefList *list = opts->extents;
2729 for (i = 1; i < idx; i++) {
2730 if (!list || !list->next) {
2731 error_setg(errp, "Extent [%d] not specified", i);
2732 return NULL;
2733 }
2734 list = list->next;
2735 }
2736 if (!list) {
2737 error_setg(errp, "Extent [%d] not specified", idx - 1);
2738 return NULL;
2739 }
2740 bs = bdrv_open_blockdev_ref(list->value, errp);
2741 }
2742 if (!bs) {
2743 return NULL;
2744 }
2745 blk = blk_new_with_bs(bs,
2746 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | BLK_PERM_RESIZE,
2747 BLK_PERM_ALL, errp);
2748 if (!blk) {
2749 return NULL;
2750 }
2751 blk_set_allow_write_beyond_eof(blk, true);
2752 bdrv_unref(bs);
2753
2754 if (size != -1) {
2755 ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2756 if (ret) {
2757 blk_unref(blk);
2758 blk = NULL;
2759 }
2760 }
2761 return blk;
2762 }
2763
2764 static int coroutine_fn vmdk_co_create(BlockdevCreateOptions *create_options,
2765 Error **errp)
2766 {
2767 int ret;
2768 BlockdevCreateOptionsVmdk *opts;
2769
2770 opts = &create_options->u.vmdk;
2771
2772 /* Validate options */
2773 if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2774 error_setg(errp, "Image size must be a multiple of 512 bytes");
2775 ret = -EINVAL;
2776 goto out;
2777 }
2778
2779 ret = vmdk_co_do_create(opts->size,
2780 opts->subformat,
2781 opts->adapter_type,
2782 opts->backing_file,
2783 opts->hwversion,
2784 false,
2785 opts->zeroed_grain,
2786 vmdk_co_create_cb,
2787 opts, errp);
2788 return ret;
2789
2790 out:
2791 return ret;
2792 }
2793
2794 static void vmdk_close(BlockDriverState *bs)
2795 {
2796 BDRVVmdkState *s = bs->opaque;
2797
2798 vmdk_free_extents(bs);
2799 g_free(s->create_type);
2800
2801 migrate_del_blocker(s->migration_blocker);
2802 error_free(s->migration_blocker);
2803 }
2804
2805 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2806 {
2807 BDRVVmdkState *s = bs->opaque;
2808 int i, err;
2809 int ret = 0;
2810
2811 for (i = 0; i < s->num_extents; i++) {
2812 err = bdrv_co_flush(s->extents[i].file->bs);
2813 if (err < 0) {
2814 ret = err;
2815 }
2816 }
2817 return ret;
2818 }
2819
2820 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2821 {
2822 int i;
2823 int64_t ret = 0;
2824 int64_t r;
2825 BDRVVmdkState *s = bs->opaque;
2826
2827 ret = bdrv_get_allocated_file_size(bs->file->bs);
2828 if (ret < 0) {
2829 return ret;
2830 }
2831 for (i = 0; i < s->num_extents; i++) {
2832 if (s->extents[i].file == bs->file) {
2833 continue;
2834 }
2835 r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2836 if (r < 0) {
2837 return r;
2838 }
2839 ret += r;
2840 }
2841 return ret;
2842 }
2843
2844 static int vmdk_has_zero_init(BlockDriverState *bs)
2845 {
2846 int i;
2847 BDRVVmdkState *s = bs->opaque;
2848
2849 /* If has a flat extent and its underlying storage doesn't have zero init,
2850 * return 0. */
2851 for (i = 0; i < s->num_extents; i++) {
2852 if (s->extents[i].flat) {
2853 if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2854 return 0;
2855 }
2856 }
2857 }
2858 return 1;
2859 }
2860
2861 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2862 {
2863 ImageInfo *info = g_new0(ImageInfo, 1);
2864
2865 bdrv_refresh_filename(extent->file->bs);
2866 *info = (ImageInfo){
2867 .filename = g_strdup(extent->file->bs->filename),
2868 .format = g_strdup(extent->type),
2869 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2870 .compressed = extent->compressed,
2871 .has_compressed = extent->compressed,
2872 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2873 .has_cluster_size = !extent->flat,
2874 };
2875
2876 return info;
2877 }
2878
2879 static int coroutine_fn vmdk_co_check(BlockDriverState *bs,
2880 BdrvCheckResult *result,
2881 BdrvCheckMode fix)
2882 {
2883 BDRVVmdkState *s = bs->opaque;
2884 VmdkExtent *extent = NULL;
2885 int64_t sector_num = 0;
2886 int64_t total_sectors = bdrv_nb_sectors(bs);
2887 int ret;
2888 uint64_t cluster_offset;
2889
2890 if (fix) {
2891 return -ENOTSUP;
2892 }
2893
2894 for (;;) {
2895 if (sector_num >= total_sectors) {
2896 return 0;
2897 }
2898 extent = find_extent(s, sector_num, extent);
2899 if (!extent) {
2900 fprintf(stderr,
2901 "ERROR: could not find extent for sector %" PRId64 "\n",
2902 sector_num);
2903 ret = -EINVAL;
2904 break;
2905 }
2906 ret = get_cluster_offset(bs, extent, NULL,
2907 sector_num << BDRV_SECTOR_BITS,
2908 false, &cluster_offset, 0, 0);
2909 if (ret == VMDK_ERROR) {
2910 fprintf(stderr,
2911 "ERROR: could not get cluster_offset for sector %"
2912 PRId64 "\n", sector_num);
2913 break;
2914 }
2915 if (ret == VMDK_OK) {
2916 int64_t extent_len = bdrv_getlength(extent->file->bs);
2917 if (extent_len < 0) {
2918 fprintf(stderr,
2919 "ERROR: could not get extent file length for sector %"
2920 PRId64 "\n", sector_num);
2921 ret = extent_len;
2922 break;
2923 }
2924 if (cluster_offset >= extent_len) {
2925 fprintf(stderr,
2926 "ERROR: cluster offset for sector %"
2927 PRId64 " points after EOF\n", sector_num);
2928 ret = -EINVAL;
2929 break;
2930 }
2931 }
2932 sector_num += extent->cluster_sectors;
2933 }
2934
2935 result->corruptions++;
2936 return ret;
2937 }
2938
2939 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs,
2940 Error **errp)
2941 {
2942 int i;
2943 BDRVVmdkState *s = bs->opaque;
2944 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2945 ImageInfoList **next;
2946
2947 *spec_info = (ImageInfoSpecific){
2948 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2949 .u = {
2950 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2951 },
2952 };
2953
2954 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2955 .create_type = g_strdup(s->create_type),
2956 .cid = s->cid,
2957 .parent_cid = s->parent_cid,
2958 };
2959
2960 next = &spec_info->u.vmdk.data->extents;
2961 for (i = 0; i < s->num_extents; i++) {
2962 *next = g_new0(ImageInfoList, 1);
2963 (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2964 (*next)->next = NULL;
2965 next = &(*next)->next;
2966 }
2967
2968 return spec_info;
2969 }
2970
2971 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2972 {
2973 return a->flat == b->flat &&
2974 a->compressed == b->compressed &&
2975 (a->flat || a->cluster_sectors == b->cluster_sectors);
2976 }
2977
2978 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2979 {
2980 int i;
2981 BDRVVmdkState *s = bs->opaque;
2982 assert(s->num_extents);
2983
2984 /* See if we have multiple extents but they have different cases */
2985 for (i = 1; i < s->num_extents; i++) {
2986 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2987 return -ENOTSUP;
2988 }
2989 }
2990 bdi->needs_compressed_writes = s->extents[0].compressed;
2991 if (!s->extents[0].flat) {
2992 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2993 }
2994 return 0;
2995 }
2996
2997 static void vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
2998 bool backing_overridden)
2999 {
3000 /* No children but file and backing can be explicitly specified (TODO) */
3001 qdict_put(target, "file",
3002 qobject_ref(bs->file->bs->full_open_options));
3003
3004 if (backing_overridden) {
3005 if (bs->backing) {
3006 qdict_put(target, "backing",
3007 qobject_ref(bs->backing->bs->full_open_options));
3008 } else {
3009 qdict_put_null(target, "backing");
3010 }
3011 }
3012 }
3013
3014 static QemuOptsList vmdk_create_opts = {
3015 .name = "vmdk-create-opts",
3016 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
3017 .desc = {
3018 {
3019 .name = BLOCK_OPT_SIZE,
3020 .type = QEMU_OPT_SIZE,
3021 .help = "Virtual disk size"
3022 },
3023 {
3024 .name = BLOCK_OPT_ADAPTER_TYPE,
3025 .type = QEMU_OPT_STRING,
3026 .help = "Virtual adapter type, can be one of "
3027 "ide (default), lsilogic, buslogic or legacyESX"
3028 },
3029 {
3030 .name = BLOCK_OPT_BACKING_FILE,
3031 .type = QEMU_OPT_STRING,
3032 .help = "File name of a base image"
3033 },
3034 {
3035 .name = BLOCK_OPT_COMPAT6,
3036 .type = QEMU_OPT_BOOL,
3037 .help = "VMDK version 6 image",
3038 .def_value_str = "off"
3039 },
3040 {
3041 .name = BLOCK_OPT_HWVERSION,
3042 .type = QEMU_OPT_STRING,
3043 .help = "VMDK hardware version",
3044 .def_value_str = "undefined"
3045 },
3046 {
3047 .name = BLOCK_OPT_SUBFMT,
3048 .type = QEMU_OPT_STRING,
3049 .help =
3050 "VMDK flat extent format, can be one of "
3051 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
3052 },
3053 {
3054 .name = BLOCK_OPT_ZEROED_GRAIN,
3055 .type = QEMU_OPT_BOOL,
3056 .help = "Enable efficient zero writes "
3057 "using the zeroed-grain GTE feature"
3058 },
3059 { /* end of list */ }
3060 }
3061 };
3062
3063 static BlockDriver bdrv_vmdk = {
3064 .format_name = "vmdk",
3065 .instance_size = sizeof(BDRVVmdkState),
3066 .bdrv_probe = vmdk_probe,
3067 .bdrv_open = vmdk_open,
3068 .bdrv_co_check = vmdk_co_check,
3069 .bdrv_reopen_prepare = vmdk_reopen_prepare,
3070 .bdrv_child_perm = bdrv_default_perms,
3071 .bdrv_co_preadv = vmdk_co_preadv,
3072 .bdrv_co_pwritev = vmdk_co_pwritev,
3073 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
3074 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
3075 .bdrv_close = vmdk_close,
3076 .bdrv_co_create_opts = vmdk_co_create_opts,
3077 .bdrv_co_create = vmdk_co_create,
3078 .bdrv_co_flush_to_disk = vmdk_co_flush,
3079 .bdrv_co_block_status = vmdk_co_block_status,
3080 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
3081 .bdrv_has_zero_init = vmdk_has_zero_init,
3082 .bdrv_get_specific_info = vmdk_get_specific_info,
3083 .bdrv_refresh_limits = vmdk_refresh_limits,
3084 .bdrv_get_info = vmdk_get_info,
3085 .bdrv_gather_child_options = vmdk_gather_child_options,
3086
3087 .is_format = true,
3088 .supports_backing = true,
3089 .create_opts = &vmdk_create_opts,
3090 };
3091
3092 static void bdrv_vmdk_init(void)
3093 {
3094 bdrv_register(&bdrv_vmdk);
3095 }
3096
3097 block_init(bdrv_vmdk_init);
QEmu