?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
python/qemu: make 'args' style arguments immutable
[qemu.git] / block / vmdk.c
bloba00dc00eb47a1d478e8c56a644bed04931fe12cb
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 = NULL;
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 = NULL;
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 = NULL;
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 case VMDK4_MAGIC:
1057 return vmdk_open_vmdk4(bs, file, flags, options, errp);
1058 default:
1059 error_setg(errp, "Image not in VMDK format");
1060 return -EINVAL;
1061 }
1062 }
1063
1064 static const char *next_line(const char *s)
1065 {
1066 while (*s) {
1067 if (*s == '\n') {
1068 return s + 1;
1069 }
1070 s++;
1071 }
1072 return s;
1073 }
1074
1075 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
1076 QDict *options, Error **errp)
1077 {
1078 int ret;
1079 int matches;
1080 char access[11];
1081 char type[11];
1082 char fname[512];
1083 const char *p, *np;
1084 int64_t sectors = 0;
1085 int64_t flat_offset;
1086 char *desc_file_dir = NULL;
1087 char *extent_path;
1088 BdrvChild *extent_file;
1089 BdrvChildRole extent_role;
1090 BDRVVmdkState *s = bs->opaque;
1091 VmdkExtent *extent = NULL;
1092 char extent_opt_prefix[32];
1093 Error *local_err = NULL;
1094
1095 for (p = desc; *p; p = next_line(p)) {
1096 /* parse extent line in one of below formats:
1097 *
1098 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
1099 * RW [size in sectors] SPARSE "file-name.vmdk"
1100 * RW [size in sectors] VMFS "file-name.vmdk"
1101 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
1102 * RW [size in sectors] SESPARSE "file-name.vmdk"
1103 */
1104 flat_offset = -1;
1105 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
1106 access, &sectors, type, fname, &flat_offset);
1107 if (matches < 4 || strcmp(access, "RW")) {
1108 continue;
1109 } else if (!strcmp(type, "FLAT")) {
1110 if (matches != 5 || flat_offset < 0) {
1111 goto invalid;
1112 }
1113 } else if (!strcmp(type, "VMFS")) {
1114 if (matches == 4) {
1115 flat_offset = 0;
1116 } else {
1117 goto invalid;
1118 }
1119 } else if (matches != 4) {
1120 goto invalid;
1121 }
1122
1123 if (sectors <= 0 ||
1124 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
1125 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") &&
1126 strcmp(type, "SESPARSE")) ||
1127 (strcmp(access, "RW"))) {
1128 continue;
1129 }
1130
1131 if (path_is_absolute(fname)) {
1132 extent_path = g_strdup(fname);
1133 } else {
1134 if (!desc_file_dir) {
1135 desc_file_dir = bdrv_dirname(bs->file->bs, errp);
1136 if (!desc_file_dir) {
1137 bdrv_refresh_filename(bs->file->bs);
1138 error_prepend(errp, "Cannot use relative paths with VMDK "
1139 "descriptor file '%s': ",
1140 bs->file->bs->filename);
1141 ret = -EINVAL;
1142 goto out;
1143 }
1144 }
1145
1146 extent_path = g_strconcat(desc_file_dir, fname, NULL);
1147 }
1148
1149 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
1150 assert(ret < 32);
1151
1152 extent_role = BDRV_CHILD_DATA;
1153 if (strcmp(type, "FLAT") != 0 && strcmp(type, "VMFS") != 0) {
1154 /* non-flat extents have metadata */
1155 extent_role |= BDRV_CHILD_METADATA;
1156 }
1157
1158 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
1159 bs, &child_of_bds, extent_role, false,
1160 &local_err);
1161 g_free(extent_path);
1162 if (local_err) {
1163 error_propagate(errp, local_err);
1164 ret = -EINVAL;
1165 goto out;
1166 }
1167
1168 /* save to extents array */
1169 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
1170 /* FLAT extent */
1171
1172 ret = vmdk_add_extent(bs, extent_file, true, sectors,
1173 0, 0, 0, 0, 0, &extent, errp);
1174 if (ret < 0) {
1175 bdrv_unref_child(bs, extent_file);
1176 goto out;
1177 }
1178 extent->flat_start_offset = flat_offset << 9;
1179 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
1180 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
1181 char *buf = vmdk_read_desc(extent_file, 0, errp);
1182 if (!buf) {
1183 ret = -EINVAL;
1184 } else {
1185 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
1186 options, errp);
1187 }
1188 g_free(buf);
1189 if (ret) {
1190 bdrv_unref_child(bs, extent_file);
1191 goto out;
1192 }
1193 extent = &s->extents[s->num_extents - 1];
1194 } else if (!strcmp(type, "SESPARSE")) {
1195 ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
1196 if (ret) {
1197 bdrv_unref_child(bs, extent_file);
1198 goto out;
1199 }
1200 extent = &s->extents[s->num_extents - 1];
1201 } else {
1202 error_setg(errp, "Unsupported extent type '%s'", type);
1203 bdrv_unref_child(bs, extent_file);
1204 ret = -ENOTSUP;
1205 goto out;
1206 }
1207 extent->type = g_strdup(type);
1208 }
1209
1210 ret = 0;
1211 goto out;
1212
1213 invalid:
1214 np = next_line(p);
1215 assert(np != p);
1216 if (np[-1] == '\n') {
1217 np--;
1218 }
1219 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
1220 ret = -EINVAL;
1221
1222 out:
1223 g_free(desc_file_dir);
1224 return ret;
1225 }
1226
1227 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
1228 QDict *options, Error **errp)
1229 {
1230 int ret;
1231 char ct[128];
1232 BDRVVmdkState *s = bs->opaque;
1233
1234 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
1235 error_setg(errp, "invalid VMDK image descriptor");
1236 ret = -EINVAL;
1237 goto exit;
1238 }
1239 if (strcmp(ct, "monolithicFlat") &&
1240 strcmp(ct, "vmfs") &&
1241 strcmp(ct, "vmfsSparse") &&
1242 strcmp(ct, "seSparse") &&
1243 strcmp(ct, "twoGbMaxExtentSparse") &&
1244 strcmp(ct, "twoGbMaxExtentFlat")) {
1245 error_setg(errp, "Unsupported image type '%s'", ct);
1246 ret = -ENOTSUP;
1247 goto exit;
1248 }
1249 s->create_type = g_strdup(ct);
1250 s->desc_offset = 0;
1251 ret = vmdk_parse_extents(buf, bs, options, errp);
1252 exit:
1253 return ret;
1254 }
1255
1256 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
1257 Error **errp)
1258 {
1259 char *buf;
1260 int ret;
1261 BDRVVmdkState *s = bs->opaque;
1262 uint32_t magic;
1263
1264 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
1265 BDRV_CHILD_IMAGE, false, errp);
1266 if (!bs->file) {
1267 return -EINVAL;
1268 }
1269
1270 buf = vmdk_read_desc(bs->file, 0, errp);
1271 if (!buf) {
1272 return -EINVAL;
1273 }
1274
1275 magic = ldl_be_p(buf);
1276 switch (magic) {
1277 case VMDK3_MAGIC:
1278 case VMDK4_MAGIC:
1279 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
1280 errp);
1281 s->desc_offset = 0x200;
1282 break;
1283 default:
1284 /* No data in the descriptor file */
1285 bs->file->role &= ~BDRV_CHILD_DATA;
1286
1287 /* Must succeed because we have given up permissions if anything */
1288 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1289
1290 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1291 break;
1292 }
1293 if (ret) {
1294 goto fail;
1295 }
1296
1297 /* try to open parent images, if exist */
1298 ret = vmdk_parent_open(bs);
1299 if (ret) {
1300 goto fail;
1301 }
1302 ret = vmdk_read_cid(bs, 0, &s->cid);
1303 if (ret) {
1304 goto fail;
1305 }
1306 ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1307 if (ret) {
1308 goto fail;
1309 }
1310 qemu_co_mutex_init(&s->lock);
1311
1312 /* Disable migration when VMDK images are used */
1313 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1314 "does not support live migration",
1315 bdrv_get_device_or_node_name(bs));
1316 ret = migrate_add_blocker(s->migration_blocker, errp);
1317 if (ret < 0) {
1318 error_free(s->migration_blocker);
1319 goto fail;
1320 }
1321
1322 g_free(buf);
1323 return 0;
1324
1325 fail:
1326 g_free(buf);
1327 g_free(s->create_type);
1328 s->create_type = NULL;
1329 vmdk_free_extents(bs);
1330 return ret;
1331 }
1332
1333
1334 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1335 {
1336 BDRVVmdkState *s = bs->opaque;
1337 int i;
1338
1339 for (i = 0; i < s->num_extents; i++) {
1340 if (!s->extents[i].flat) {
1341 bs->bl.pwrite_zeroes_alignment =
1342 MAX(bs->bl.pwrite_zeroes_alignment,
1343 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1344 }
1345 }
1346 }
1347
1348 /**
1349 * get_whole_cluster
1350 *
1351 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1352 * to the cluster at @cluster_sector_num. If @zeroed is true, we're overwriting
1353 * a zeroed cluster in the current layer and must not copy data from the
1354 * backing file.
1355 *
1356 * If @skip_start_sector < @skip_end_sector, the relative range
1357 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1358 * it for call to write user data in the request.
1359 */
1360 static int get_whole_cluster(BlockDriverState *bs,
1361 VmdkExtent *extent,
1362 uint64_t cluster_offset,
1363 uint64_t offset,
1364 uint64_t skip_start_bytes,
1365 uint64_t skip_end_bytes,
1366 bool zeroed)
1367 {
1368 int ret = VMDK_OK;
1369 int64_t cluster_bytes;
1370 uint8_t *whole_grain;
1371 bool copy_from_backing;
1372
1373 /* For COW, align request sector_num to cluster start */
1374 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1375 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1376 whole_grain = qemu_blockalign(bs, cluster_bytes);
1377 copy_from_backing = bs->backing && !zeroed;
1378
1379 if (!copy_from_backing) {
1380 memset(whole_grain, 0, skip_start_bytes);
1381 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1382 }
1383
1384 assert(skip_end_bytes <= cluster_bytes);
1385 /* we will be here if it's first write on non-exist grain(cluster).
1386 * try to read from parent image, if exist */
1387 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1388 ret = VMDK_ERROR;
1389 goto exit;
1390 }
1391
1392 /* Read backing data before skip range */
1393 if (skip_start_bytes > 0) {
1394 if (copy_from_backing) {
1395 /* qcow2 emits this on bs->file instead of bs->backing */
1396 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1397 ret = bdrv_pread(bs->backing, offset, whole_grain,
1398 skip_start_bytes);
1399 if (ret < 0) {
1400 ret = VMDK_ERROR;
1401 goto exit;
1402 }
1403 }
1404 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1405 ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1406 skip_start_bytes);
1407 if (ret < 0) {
1408 ret = VMDK_ERROR;
1409 goto exit;
1410 }
1411 }
1412 /* Read backing data after skip range */
1413 if (skip_end_bytes < cluster_bytes) {
1414 if (copy_from_backing) {
1415 /* qcow2 emits this on bs->file instead of bs->backing */
1416 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1417 ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1418 whole_grain + skip_end_bytes,
1419 cluster_bytes - skip_end_bytes);
1420 if (ret < 0) {
1421 ret = VMDK_ERROR;
1422 goto exit;
1423 }
1424 }
1425 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1426 ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1427 whole_grain + skip_end_bytes,
1428 cluster_bytes - skip_end_bytes);
1429 if (ret < 0) {
1430 ret = VMDK_ERROR;
1431 goto exit;
1432 }
1433 }
1434
1435 ret = VMDK_OK;
1436 exit:
1437 qemu_vfree(whole_grain);
1438 return ret;
1439 }
1440
1441 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1442 uint32_t offset)
1443 {
1444 offset = cpu_to_le32(offset);
1445 /* update L2 table */
1446 BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1447 if (bdrv_pwrite(extent->file,
1448 ((int64_t)m_data->l2_offset * 512)
1449 + (m_data->l2_index * sizeof(offset)),
1450 &offset, sizeof(offset)) < 0) {
1451 return VMDK_ERROR;
1452 }
1453 /* update backup L2 table */
1454 if (extent->l1_backup_table_offset != 0) {
1455 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1456 if (bdrv_pwrite(extent->file,
1457 ((int64_t)m_data->l2_offset * 512)
1458 + (m_data->l2_index * sizeof(offset)),
1459 &offset, sizeof(offset)) < 0) {
1460 return VMDK_ERROR;
1461 }
1462 }
1463 if (bdrv_flush(extent->file->bs) < 0) {
1464 return VMDK_ERROR;
1465 }
1466 if (m_data->l2_cache_entry) {
1467 *m_data->l2_cache_entry = offset;
1468 }
1469
1470 return VMDK_OK;
1471 }
1472
1473 /**
1474 * get_cluster_offset
1475 *
1476 * Look up cluster offset in extent file by sector number, and store in
1477 * @cluster_offset.
1478 *
1479 * For flat extents, the start offset as parsed from the description file is
1480 * returned.
1481 *
1482 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1483 * offset for a new cluster and update L2 cache. If there is a backing file,
1484 * COW is done before returning; otherwise, zeroes are written to the allocated
1485 * cluster. Both COW and zero writing skips the sector range
1486 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1487 * has new data to write there.
1488 *
1489 * Returns: VMDK_OK if cluster exists and mapped in the image.
1490 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1491 * VMDK_ERROR if failed.
1492 */
1493 static int get_cluster_offset(BlockDriverState *bs,
1494 VmdkExtent *extent,
1495 VmdkMetaData *m_data,
1496 uint64_t offset,
1497 bool allocate,
1498 uint64_t *cluster_offset,
1499 uint64_t skip_start_bytes,
1500 uint64_t skip_end_bytes)
1501 {
1502 unsigned int l1_index, l2_offset, l2_index;
1503 int min_index, i, j;
1504 uint32_t min_count;
1505 void *l2_table;
1506 bool zeroed = false;
1507 int64_t ret;
1508 int64_t cluster_sector;
1509 unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
1510
1511 if (m_data) {
1512 m_data->new_allocation = false;
1513 }
1514 if (extent->flat) {
1515 *cluster_offset = extent->flat_start_offset;
1516 return VMDK_OK;
1517 }
1518
1519 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1520 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1521 if (l1_index >= extent->l1_size) {
1522 return VMDK_ERROR;
1523 }
1524 if (extent->sesparse) {
1525 uint64_t l2_offset_u64;
1526
1527 assert(extent->entry_size == sizeof(uint64_t));
1528
1529 l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
1530 if (l2_offset_u64 == 0) {
1531 l2_offset = 0;
1532 } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
1533 /*
1534 * Top most nibble is 0x1 if grain table is allocated.
1535 * strict check - top most 4 bytes must be 0x10000000 since max
1536 * supported size is 64TB for disk - so no more than 64TB / 16MB
1537 * grain directories which is smaller than uint32,
1538 * where 16MB is the only supported default grain table coverage.
1539 */
1540 return VMDK_ERROR;
1541 } else {
1542 l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
1543 l2_offset_u64 = extent->sesparse_l2_tables_offset +
1544 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
1545 if (l2_offset_u64 > 0x00000000ffffffff) {
1546 return VMDK_ERROR;
1547 }
1548 l2_offset = (unsigned int)(l2_offset_u64);
1549 }
1550 } else {
1551 assert(extent->entry_size == sizeof(uint32_t));
1552 l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
1553 }
1554 if (!l2_offset) {
1555 return VMDK_UNALLOC;
1556 }
1557 for (i = 0; i < L2_CACHE_SIZE; i++) {
1558 if (l2_offset == extent->l2_cache_offsets[i]) {
1559 /* increment the hit count */
1560 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1561 for (j = 0; j < L2_CACHE_SIZE; j++) {
1562 extent->l2_cache_counts[j] >>= 1;
1563 }
1564 }
1565 l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
1566 goto found;
1567 }
1568 }
1569 /* not found: load a new entry in the least used one */
1570 min_index = 0;
1571 min_count = 0xffffffff;
1572 for (i = 0; i < L2_CACHE_SIZE; i++) {
1573 if (extent->l2_cache_counts[i] < min_count) {
1574 min_count = extent->l2_cache_counts[i];
1575 min_index = i;
1576 }
1577 }
1578 l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
1579 BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1580 if (bdrv_pread(extent->file,
1581 (int64_t)l2_offset * 512,
1582 l2_table,
1583 l2_size_bytes
1584 ) != l2_size_bytes) {
1585 return VMDK_ERROR;
1586 }
1587
1588 extent->l2_cache_offsets[min_index] = l2_offset;
1589 extent->l2_cache_counts[min_index] = 1;
1590 found:
1591 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1592 if (m_data) {
1593 m_data->l1_index = l1_index;
1594 m_data->l2_index = l2_index;
1595 m_data->l2_offset = l2_offset;
1596 m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
1597 }
1598
1599 if (extent->sesparse) {
1600 cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
1601 switch (cluster_sector & 0xf000000000000000) {
1602 case 0x0000000000000000:
1603 /* unallocated grain */
1604 if (cluster_sector != 0) {
1605 return VMDK_ERROR;
1606 }
1607 break;
1608 case 0x1000000000000000:
1609 /* scsi-unmapped grain - fallthrough */
1610 case 0x2000000000000000:
1611 /* zero grain */
1612 zeroed = true;
1613 break;
1614 case 0x3000000000000000:
1615 /* allocated grain */
1616 cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
1617 ((cluster_sector & 0x0000ffffffffffff) << 12));
1618 cluster_sector = extent->sesparse_clusters_offset +
1619 cluster_sector * extent->cluster_sectors;
1620 break;
1621 default:
1622 return VMDK_ERROR;
1623 }
1624 } else {
1625 cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
1626
1627 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1628 zeroed = true;
1629 }
1630 }
1631
1632 if (!cluster_sector || zeroed) {
1633 if (!allocate) {
1634 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1635 }
1636 assert(!extent->sesparse);
1637
1638 if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1639 return VMDK_ERROR;
1640 }
1641
1642 cluster_sector = extent->next_cluster_sector;
1643 extent->next_cluster_sector += extent->cluster_sectors;
1644
1645 /* First of all we write grain itself, to avoid race condition
1646 * that may to corrupt the image.
1647 * This problem may occur because of insufficient space on host disk
1648 * or inappropriate VM shutdown.
1649 */
1650 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1651 offset, skip_start_bytes, skip_end_bytes,
1652 zeroed);
1653 if (ret) {
1654 return ret;
1655 }
1656 if (m_data) {
1657 m_data->new_allocation = true;
1658 }
1659 }
1660 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1661 return VMDK_OK;
1662 }
1663
1664 static VmdkExtent *find_extent(BDRVVmdkState *s,
1665 int64_t sector_num, VmdkExtent *start_hint)
1666 {
1667 VmdkExtent *extent = start_hint;
1668
1669 if (!extent) {
1670 extent = &s->extents[0];
1671 }
1672 while (extent < &s->extents[s->num_extents]) {
1673 if (sector_num < extent->end_sector) {
1674 return extent;
1675 }
1676 extent++;
1677 }
1678 return NULL;
1679 }
1680
1681 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1682 int64_t offset)
1683 {
1684 uint64_t extent_begin_offset, extent_relative_offset;
1685 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1686
1687 extent_begin_offset =
1688 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1689 extent_relative_offset = offset - extent_begin_offset;
1690 return extent_relative_offset % cluster_size;
1691 }
1692
1693 static int coroutine_fn vmdk_co_block_status(BlockDriverState *bs,
1694 bool want_zero,
1695 int64_t offset, int64_t bytes,
1696 int64_t *pnum, int64_t *map,
1697 BlockDriverState **file)
1698 {
1699 BDRVVmdkState *s = bs->opaque;
1700 int64_t index_in_cluster, n, ret;
1701 uint64_t cluster_offset;
1702 VmdkExtent *extent;
1703
1704 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1705 if (!extent) {
1706 return -EIO;
1707 }
1708 qemu_co_mutex_lock(&s->lock);
1709 ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1710 0, 0);
1711 qemu_co_mutex_unlock(&s->lock);
1712
1713 index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1714 switch (ret) {
1715 case VMDK_ERROR:
1716 ret = -EIO;
1717 break;
1718 case VMDK_UNALLOC:
1719 ret = 0;
1720 break;
1721 case VMDK_ZEROED:
1722 ret = BDRV_BLOCK_ZERO;
1723 break;
1724 case VMDK_OK:
1725 ret = BDRV_BLOCK_DATA;
1726 if (!extent->compressed) {
1727 ret |= BDRV_BLOCK_OFFSET_VALID;
1728 *map = cluster_offset + index_in_cluster;
1729 if (extent->flat) {
1730 ret |= BDRV_BLOCK_RECURSE;
1731 }
1732 }
1733 *file = extent->file->bs;
1734 break;
1735 }
1736
1737 n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1738 *pnum = MIN(n, bytes);
1739 return ret;
1740 }
1741
1742 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1743 int64_t offset_in_cluster, QEMUIOVector *qiov,
1744 uint64_t qiov_offset, uint64_t n_bytes,
1745 uint64_t offset)
1746 {
1747 int ret;
1748 VmdkGrainMarker *data = NULL;
1749 uLongf buf_len;
1750 QEMUIOVector local_qiov;
1751 int64_t write_offset;
1752 int64_t write_end_sector;
1753
1754 if (extent->compressed) {
1755 void *compressed_data;
1756
1757 /* Only whole clusters */
1758 if (offset_in_cluster ||
1759 n_bytes > (extent->cluster_sectors * SECTOR_SIZE) ||
1760 (n_bytes < (extent->cluster_sectors * SECTOR_SIZE) &&
1761 offset + n_bytes != extent->end_sector * SECTOR_SIZE))
1762 {
1763 ret = -EINVAL;
1764 goto out;
1765 }
1766
1767 if (!extent->has_marker) {
1768 ret = -EINVAL;
1769 goto out;
1770 }
1771 buf_len = (extent->cluster_sectors << 9) * 2;
1772 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1773
1774 compressed_data = g_malloc(n_bytes);
1775 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1776 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1777 g_free(compressed_data);
1778
1779 if (ret != Z_OK || buf_len == 0) {
1780 ret = -EINVAL;
1781 goto out;
1782 }
1783
1784 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1785 data->size = cpu_to_le32(buf_len);
1786
1787 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1788 qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1789
1790 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1791 } else {
1792 qemu_iovec_init(&local_qiov, qiov->niov);
1793 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1794
1795 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1796 }
1797
1798 write_offset = cluster_offset + offset_in_cluster;
1799 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1800 &local_qiov, 0);
1801
1802 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1803
1804 if (extent->compressed) {
1805 extent->next_cluster_sector = write_end_sector;
1806 } else {
1807 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1808 write_end_sector);
1809 }
1810
1811 if (ret < 0) {
1812 goto out;
1813 }
1814 ret = 0;
1815 out:
1816 g_free(data);
1817 if (!extent->compressed) {
1818 qemu_iovec_destroy(&local_qiov);
1819 }
1820 return ret;
1821 }
1822
1823 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1824 int64_t offset_in_cluster, QEMUIOVector *qiov,
1825 int bytes)
1826 {
1827 int ret;
1828 int cluster_bytes, buf_bytes;
1829 uint8_t *cluster_buf, *compressed_data;
1830 uint8_t *uncomp_buf;
1831 uint32_t data_len;
1832 VmdkGrainMarker *marker;
1833 uLongf buf_len;
1834
1835
1836 if (!extent->compressed) {
1837 BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1838 ret = bdrv_co_preadv(extent->file,
1839 cluster_offset + offset_in_cluster, bytes,
1840 qiov, 0);
1841 if (ret < 0) {
1842 return ret;
1843 }
1844 return 0;
1845 }
1846 cluster_bytes = extent->cluster_sectors * 512;
1847 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1848 buf_bytes = cluster_bytes * 2;
1849 cluster_buf = g_malloc(buf_bytes);
1850 uncomp_buf = g_malloc(cluster_bytes);
1851 BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1852 ret = bdrv_pread(extent->file,
1853 cluster_offset,
1854 cluster_buf, buf_bytes);
1855 if (ret < 0) {
1856 goto out;
1857 }
1858 compressed_data = cluster_buf;
1859 buf_len = cluster_bytes;
1860 data_len = cluster_bytes;
1861 if (extent->has_marker) {
1862 marker = (VmdkGrainMarker *)cluster_buf;
1863 compressed_data = marker->data;
1864 data_len = le32_to_cpu(marker->size);
1865 }
1866 if (!data_len || data_len > buf_bytes) {
1867 ret = -EINVAL;
1868 goto out;
1869 }
1870 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1871 if (ret != Z_OK) {
1872 ret = -EINVAL;
1873 goto out;
1874
1875 }
1876 if (offset_in_cluster < 0 ||
1877 offset_in_cluster + bytes > buf_len) {
1878 ret = -EINVAL;
1879 goto out;
1880 }
1881 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1882 ret = 0;
1883
1884 out:
1885 g_free(uncomp_buf);
1886 g_free(cluster_buf);
1887 return ret;
1888 }
1889
1890 static int coroutine_fn
1891 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1892 QEMUIOVector *qiov, int flags)
1893 {
1894 BDRVVmdkState *s = bs->opaque;
1895 int ret;
1896 uint64_t n_bytes, offset_in_cluster;
1897 VmdkExtent *extent = NULL;
1898 QEMUIOVector local_qiov;
1899 uint64_t cluster_offset;
1900 uint64_t bytes_done = 0;
1901
1902 qemu_iovec_init(&local_qiov, qiov->niov);
1903 qemu_co_mutex_lock(&s->lock);
1904
1905 while (bytes > 0) {
1906 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1907 if (!extent) {
1908 ret = -EIO;
1909 goto fail;
1910 }
1911 ret = get_cluster_offset(bs, extent, NULL,
1912 offset, false, &cluster_offset, 0, 0);
1913 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1914
1915 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1916 - offset_in_cluster);
1917
1918 if (ret != VMDK_OK) {
1919 /* if not allocated, try to read from parent image, if exist */
1920 if (bs->backing && ret != VMDK_ZEROED) {
1921 if (!vmdk_is_cid_valid(bs)) {
1922 ret = -EINVAL;
1923 goto fail;
1924 }
1925
1926 qemu_iovec_reset(&local_qiov);
1927 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1928
1929 /* qcow2 emits this on bs->file instead of bs->backing */
1930 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1931 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1932 &local_qiov, 0);
1933 if (ret < 0) {
1934 goto fail;
1935 }
1936 } else {
1937 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1938 }
1939 } else {
1940 qemu_iovec_reset(&local_qiov);
1941 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1942
1943 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1944 &local_qiov, n_bytes);
1945 if (ret) {
1946 goto fail;
1947 }
1948 }
1949 bytes -= n_bytes;
1950 offset += n_bytes;
1951 bytes_done += n_bytes;
1952 }
1953
1954 ret = 0;
1955 fail:
1956 qemu_co_mutex_unlock(&s->lock);
1957 qemu_iovec_destroy(&local_qiov);
1958
1959 return ret;
1960 }
1961
1962 /**
1963 * vmdk_write:
1964 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1965 * if possible, otherwise return -ENOTSUP.
1966 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1967 * with each cluster. By dry run we can find if the zero write
1968 * is possible without modifying image data.
1969 *
1970 * Returns: error code with 0 for success.
1971 */
1972 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1973 uint64_t bytes, QEMUIOVector *qiov,
1974 bool zeroed, bool zero_dry_run)
1975 {
1976 BDRVVmdkState *s = bs->opaque;
1977 VmdkExtent *extent = NULL;
1978 int ret;
1979 int64_t offset_in_cluster, n_bytes;
1980 uint64_t cluster_offset;
1981 uint64_t bytes_done = 0;
1982 VmdkMetaData m_data;
1983
1984 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1985 error_report("Wrong offset: offset=0x%" PRIx64
1986 " total_sectors=0x%" PRIx64,
1987 offset, bs->total_sectors);
1988 return -EIO;
1989 }
1990
1991 while (bytes > 0) {
1992 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1993 if (!extent) {
1994 return -EIO;
1995 }
1996 if (extent->sesparse) {
1997 return -ENOTSUP;
1998 }
1999 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
2000 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
2001 - offset_in_cluster);
2002
2003 ret = get_cluster_offset(bs, extent, &m_data, offset,
2004 !(extent->compressed || zeroed),
2005 &cluster_offset, offset_in_cluster,
2006 offset_in_cluster + n_bytes);
2007 if (extent->compressed) {
2008 if (ret == VMDK_OK) {
2009 /* Refuse write to allocated cluster for streamOptimized */
2010 error_report("Could not write to allocated cluster"
2011 " for streamOptimized");
2012 return -EIO;
2013 } else if (!zeroed) {
2014 /* allocate */
2015 ret = get_cluster_offset(bs, extent, &m_data, offset,
2016 true, &cluster_offset, 0, 0);
2017 }
2018 }
2019 if (ret == VMDK_ERROR) {
2020 return -EINVAL;
2021 }
2022 if (zeroed) {
2023 /* Do zeroed write, buf is ignored */
2024 if (extent->has_zero_grain &&
2025 offset_in_cluster == 0 &&
2026 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
2027 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
2028 if (!zero_dry_run && ret != VMDK_ZEROED) {
2029 /* update L2 tables */
2030 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
2031 != VMDK_OK) {
2032 return -EIO;
2033 }
2034 }
2035 } else {
2036 return -ENOTSUP;
2037 }
2038 } else {
2039 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
2040 qiov, bytes_done, n_bytes, offset);
2041 if (ret) {
2042 return ret;
2043 }
2044 if (m_data.new_allocation) {
2045 /* update L2 tables */
2046 if (vmdk_L2update(extent, &m_data,
2047 cluster_offset >> BDRV_SECTOR_BITS)
2048 != VMDK_OK) {
2049 return -EIO;
2050 }
2051 }
2052 }
2053 bytes -= n_bytes;
2054 offset += n_bytes;
2055 bytes_done += n_bytes;
2056
2057 /* update CID on the first write every time the virtual disk is
2058 * opened */
2059 if (!s->cid_updated) {
2060 ret = vmdk_write_cid(bs, g_random_int());
2061 if (ret < 0) {
2062 return ret;
2063 }
2064 s->cid_updated = true;
2065 }
2066 }
2067 return 0;
2068 }
2069
2070 static int coroutine_fn
2071 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2072 QEMUIOVector *qiov, int flags)
2073 {
2074 int ret;
2075 BDRVVmdkState *s = bs->opaque;
2076 qemu_co_mutex_lock(&s->lock);
2077 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
2078 qemu_co_mutex_unlock(&s->lock);
2079 return ret;
2080 }
2081
2082 static int coroutine_fn
2083 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
2084 uint64_t bytes, QEMUIOVector *qiov)
2085 {
2086 if (bytes == 0) {
2087 /* The caller will write bytes 0 to signal EOF.
2088 * When receive it, we align EOF to a sector boundary. */
2089 BDRVVmdkState *s = bs->opaque;
2090 int i, ret;
2091 int64_t length;
2092
2093 for (i = 0; i < s->num_extents; i++) {
2094 length = bdrv_getlength(s->extents[i].file->bs);
2095 if (length < 0) {
2096 return length;
2097 }
2098 length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
2099 ret = bdrv_truncate(s->extents[i].file, length, false,
2100 PREALLOC_MODE_OFF, 0, NULL);
2101 if (ret < 0) {
2102 return ret;
2103 }
2104 }
2105 return 0;
2106 }
2107 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
2108 }
2109
2110 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
2111 int64_t offset,
2112 int bytes,
2113 BdrvRequestFlags flags)
2114 {
2115 int ret;
2116 BDRVVmdkState *s = bs->opaque;
2117
2118 qemu_co_mutex_lock(&s->lock);
2119 /* write zeroes could fail if sectors not aligned to cluster, test it with
2120 * dry_run == true before really updating image */
2121 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
2122 if (!ret) {
2123 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
2124 }
2125 qemu_co_mutex_unlock(&s->lock);
2126 return ret;
2127 }
2128
2129 static int vmdk_init_extent(BlockBackend *blk,
2130 int64_t filesize, bool flat,
2131 bool compress, bool zeroed_grain,
2132 Error **errp)
2133 {
2134 int ret, i;
2135 VMDK4Header header;
2136 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
2137 uint32_t *gd_buf = NULL;
2138 int gd_buf_size;
2139
2140 if (flat) {
2141 ret = blk_truncate(blk, filesize, false, PREALLOC_MODE_OFF, 0, errp);
2142 goto exit;
2143 }
2144 magic = cpu_to_be32(VMDK4_MAGIC);
2145 memset(&header, 0, sizeof(header));
2146 if (compress) {
2147 header.version = 3;
2148 } else if (zeroed_grain) {
2149 header.version = 2;
2150 } else {
2151 header.version = 1;
2152 }
2153 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
2154 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
2155 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
2156 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
2157 header.capacity = filesize / BDRV_SECTOR_SIZE;
2158 header.granularity = 128;
2159 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
2160
2161 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
2162 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
2163 BDRV_SECTOR_SIZE);
2164 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
2165 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
2166
2167 header.desc_offset = 1;
2168 header.desc_size = 20;
2169 header.rgd_offset = header.desc_offset + header.desc_size;
2170 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
2171 header.grain_offset =
2172 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
2173 header.granularity);
2174 /* swap endianness for all header fields */
2175 header.version = cpu_to_le32(header.version);
2176 header.flags = cpu_to_le32(header.flags);
2177 header.capacity = cpu_to_le64(header.capacity);
2178 header.granularity = cpu_to_le64(header.granularity);
2179 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
2180 header.desc_offset = cpu_to_le64(header.desc_offset);
2181 header.desc_size = cpu_to_le64(header.desc_size);
2182 header.rgd_offset = cpu_to_le64(header.rgd_offset);
2183 header.gd_offset = cpu_to_le64(header.gd_offset);
2184 header.grain_offset = cpu_to_le64(header.grain_offset);
2185 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
2186
2187 header.check_bytes[0] = 0xa;
2188 header.check_bytes[1] = 0x20;
2189 header.check_bytes[2] = 0xd;
2190 header.check_bytes[3] = 0xa;
2191
2192 /* write all the data */
2193 ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
2194 if (ret < 0) {
2195 error_setg(errp, QERR_IO_ERROR);
2196 goto exit;
2197 }
2198 ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
2199 if (ret < 0) {
2200 error_setg(errp, QERR_IO_ERROR);
2201 goto exit;
2202 }
2203
2204 ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false,
2205 PREALLOC_MODE_OFF, 0, errp);
2206 if (ret < 0) {
2207 goto exit;
2208 }
2209
2210 /* write grain directory */
2211 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
2212 gd_buf = g_malloc0(gd_buf_size);
2213 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
2214 i < gt_count; i++, tmp += gt_size) {
2215 gd_buf[i] = cpu_to_le32(tmp);
2216 }
2217 ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
2218 gd_buf, gd_buf_size, 0);
2219 if (ret < 0) {
2220 error_setg(errp, QERR_IO_ERROR);
2221 goto exit;
2222 }
2223
2224 /* write backup grain directory */
2225 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
2226 i < gt_count; i++, tmp += gt_size) {
2227 gd_buf[i] = cpu_to_le32(tmp);
2228 }
2229 ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
2230 gd_buf, gd_buf_size, 0);
2231 if (ret < 0) {
2232 error_setg(errp, QERR_IO_ERROR);
2233 }
2234
2235 ret = 0;
2236 exit:
2237 g_free(gd_buf);
2238 return ret;
2239 }
2240
2241 static int vmdk_create_extent(const char *filename, int64_t filesize,
2242 bool flat, bool compress, bool zeroed_grain,
2243 BlockBackend **pbb,
2244 QemuOpts *opts, Error **errp)
2245 {
2246 int ret;
2247 BlockBackend *blk = NULL;
2248
2249 ret = bdrv_create_file(filename, opts, errp);
2250 if (ret < 0) {
2251 goto exit;
2252 }
2253
2254 blk = blk_new_open(filename, NULL, NULL,
2255 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2256 errp);
2257 if (blk == NULL) {
2258 ret = -EIO;
2259 goto exit;
2260 }
2261
2262 blk_set_allow_write_beyond_eof(blk, true);
2263
2264 ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
2265 exit:
2266 if (blk) {
2267 if (pbb) {
2268 *pbb = blk;
2269 } else {
2270 blk_unref(blk);
2271 blk = NULL;
2272 }
2273 }
2274 return ret;
2275 }
2276
2277 static int filename_decompose(const char *filename, char *path, char *prefix,
2278 char *postfix, size_t buf_len, Error **errp)
2279 {
2280 const char *p, *q;
2281
2282 if (filename == NULL || !strlen(filename)) {
2283 error_setg(errp, "No filename provided");
2284 return VMDK_ERROR;
2285 }
2286 p = strrchr(filename, '/');
2287 if (p == NULL) {
2288 p = strrchr(filename, '\\');
2289 }
2290 if (p == NULL) {
2291 p = strrchr(filename, ':');
2292 }
2293 if (p != NULL) {
2294 p++;
2295 if (p - filename >= buf_len) {
2296 return VMDK_ERROR;
2297 }
2298 pstrcpy(path, p - filename + 1, filename);
2299 } else {
2300 p = filename;
2301 path[0] = '\0';
2302 }
2303 q = strrchr(p, '.');
2304 if (q == NULL) {
2305 pstrcpy(prefix, buf_len, p);
2306 postfix[0] = '\0';
2307 } else {
2308 if (q - p >= buf_len) {
2309 return VMDK_ERROR;
2310 }
2311 pstrcpy(prefix, q - p + 1, p);
2312 pstrcpy(postfix, buf_len, q);
2313 }
2314 return VMDK_OK;
2315 }
2316
2317 /*
2318 * idx == 0: get or create the descriptor file (also the image file if in a
2319 * non-split format.
2320 * idx >= 1: get the n-th extent if in a split subformat
2321 */
2322 typedef BlockBackend *(*vmdk_create_extent_fn)(int64_t size,
2323 int idx,
2324 bool flat,
2325 bool split,
2326 bool compress,
2327 bool zeroed_grain,
2328 void *opaque,
2329 Error **errp);
2330
2331 static void vmdk_desc_add_extent(GString *desc,
2332 const char *extent_line_fmt,
2333 int64_t size, const char *filename)
2334 {
2335 char *basename = g_path_get_basename(filename);
2336
2337 g_string_append_printf(desc, extent_line_fmt,
2338 DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
2339 g_free(basename);
2340 }
2341
2342 static int coroutine_fn vmdk_co_do_create(int64_t size,
2343 BlockdevVmdkSubformat subformat,
2344 BlockdevVmdkAdapterType adapter_type,
2345 const char *backing_file,
2346 const char *hw_version,
2347 bool compat6,
2348 bool zeroed_grain,
2349 vmdk_create_extent_fn extent_fn,
2350 void *opaque,
2351 Error **errp)
2352 {
2353 int extent_idx;
2354 BlockBackend *blk = NULL;
2355 BlockBackend *extent_blk;
2356 Error *local_err = NULL;
2357 char *desc = NULL;
2358 int ret = 0;
2359 bool flat, split, compress;
2360 GString *ext_desc_lines;
2361 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
2362 int64_t extent_size;
2363 int64_t created_size = 0;
2364 const char *extent_line_fmt;
2365 char *parent_desc_line = g_malloc0(BUF_SIZE);
2366 uint32_t parent_cid = 0xffffffff;
2367 uint32_t number_heads = 16;
2368 uint32_t desc_offset = 0, desc_len;
2369 const char desc_template[] =
2370 "# Disk DescriptorFile\n"
2371 "version=1\n"
2372 "CID=%" PRIx32 "\n"
2373 "parentCID=%" PRIx32 "\n"
2374 "createType=\"%s\"\n"
2375 "%s"
2376 "\n"
2377 "# Extent description\n"
2378 "%s"
2379 "\n"
2380 "# The Disk Data Base\n"
2381 "#DDB\n"
2382 "\n"
2383 "ddb.virtualHWVersion = \"%s\"\n"
2384 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2385 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2386 "ddb.geometry.sectors = \"63\"\n"
2387 "ddb.adapterType = \"%s\"\n";
2388
2389 ext_desc_lines = g_string_new(NULL);
2390
2391 /* Read out options */
2392 if (compat6) {
2393 if (hw_version) {
2394 error_setg(errp,
2395 "compat6 cannot be enabled with hwversion set");
2396 ret = -EINVAL;
2397 goto exit;
2398 }
2399 hw_version = "6";
2400 }
2401 if (!hw_version) {
2402 hw_version = "4";
2403 }
2404
2405 if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2406 /* that's the number of heads with which vmware operates when
2407 creating, exporting, etc. vmdk files with a non-ide adapter type */
2408 number_heads = 255;
2409 }
2410 split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2411 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2412 flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2413 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2414 compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2415
2416 if (flat) {
2417 extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2418 } else {
2419 extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2420 }
2421 if (flat && backing_file) {
2422 error_setg(errp, "Flat image can't have backing file");
2423 ret = -ENOTSUP;
2424 goto exit;
2425 }
2426 if (flat && zeroed_grain) {
2427 error_setg(errp, "Flat image can't enable zeroed grain");
2428 ret = -ENOTSUP;
2429 goto exit;
2430 }
2431
2432 /* Create extents */
2433 if (split) {
2434 extent_size = split_size;
2435 } else {
2436 extent_size = size;
2437 }
2438 if (!split && !flat) {
2439 created_size = extent_size;
2440 } else {
2441 created_size = 0;
2442 }
2443 /* Get the descriptor file BDS */
2444 blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2445 opaque, errp);
2446 if (!blk) {
2447 ret = -EIO;
2448 goto exit;
2449 }
2450 if (!split && !flat) {
2451 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2452 blk_bs(blk)->filename);
2453 }
2454
2455 if (backing_file) {
2456 BlockBackend *backing;
2457 char *full_backing =
2458 bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2459 backing_file,
2460 &local_err);
2461 if (local_err) {
2462 error_propagate(errp, local_err);
2463 ret = -ENOENT;
2464 goto exit;
2465 }
2466 assert(full_backing);
2467
2468 backing = blk_new_open(full_backing, NULL, NULL,
2469 BDRV_O_NO_BACKING, errp);
2470 g_free(full_backing);
2471 if (backing == NULL) {
2472 ret = -EIO;
2473 goto exit;
2474 }
2475 if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2476 error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2477 blk_bs(backing)->drv->format_name);
2478 blk_unref(backing);
2479 ret = -EINVAL;
2480 goto exit;
2481 }
2482 ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2483 blk_unref(backing);
2484 if (ret) {
2485 error_setg(errp, "Failed to read parent CID");
2486 goto exit;
2487 }
2488 snprintf(parent_desc_line, BUF_SIZE,
2489 "parentFileNameHint=\"%s\"", backing_file);
2490 }
2491 extent_idx = 1;
2492 while (created_size < size) {
2493 int64_t cur_size = MIN(size - created_size, extent_size);
2494 extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2495 zeroed_grain, opaque, errp);
2496 if (!extent_blk) {
2497 ret = -EINVAL;
2498 goto exit;
2499 }
2500 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2501 blk_bs(extent_blk)->filename);
2502 created_size += cur_size;
2503 extent_idx++;
2504 blk_unref(extent_blk);
2505 }
2506
2507 /* Check whether we got excess extents */
2508 extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2509 opaque, NULL);
2510 if (extent_blk) {
2511 blk_unref(extent_blk);
2512 error_setg(errp, "List of extents contains unused extents");
2513 ret = -EINVAL;
2514 goto exit;
2515 }
2516
2517 /* generate descriptor file */
2518 desc = g_strdup_printf(desc_template,
2519 g_random_int(),
2520 parent_cid,
2521 BlockdevVmdkSubformat_str(subformat),
2522 parent_desc_line,
2523 ext_desc_lines->str,
2524 hw_version,
2525 size /
2526 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2527 number_heads,
2528 BlockdevVmdkAdapterType_str(adapter_type));
2529 desc_len = strlen(desc);
2530 /* the descriptor offset = 0x200 */
2531 if (!split && !flat) {
2532 desc_offset = 0x200;
2533 }
2534
2535 ret = blk_pwrite(blk, desc_offset, desc, desc_len, 0);
2536 if (ret < 0) {
2537 error_setg_errno(errp, -ret, "Could not write description");
2538 goto exit;
2539 }
2540 /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2541 * for description file */
2542 if (desc_offset == 0) {
2543 ret = blk_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, 0, errp);
2544 if (ret < 0) {
2545 goto exit;
2546 }
2547 }
2548 ret = 0;
2549 exit:
2550 if (blk) {
2551 blk_unref(blk);
2552 }
2553 g_free(desc);
2554 g_free(parent_desc_line);
2555 g_string_free(ext_desc_lines, true);
2556 return ret;
2557 }
2558
2559 typedef struct {
2560 char *path;
2561 char *prefix;
2562 char *postfix;
2563 QemuOpts *opts;
2564 } VMDKCreateOptsData;
2565
2566 static BlockBackend *vmdk_co_create_opts_cb(int64_t size, int idx,
2567 bool flat, bool split, bool compress,
2568 bool zeroed_grain, void *opaque,
2569 Error **errp)
2570 {
2571 BlockBackend *blk = NULL;
2572 BlockDriverState *bs = NULL;
2573 VMDKCreateOptsData *data = opaque;
2574 char *ext_filename = NULL;
2575 char *rel_filename = NULL;
2576
2577 /* We're done, don't create excess extents. */
2578 if (size == -1) {
2579 assert(errp == NULL);
2580 return NULL;
2581 }
2582
2583 if (idx == 0) {
2584 rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2585 } else if (split) {
2586 rel_filename = g_strdup_printf("%s-%c%03d%s",
2587 data->prefix,
2588 flat ? 'f' : 's', idx, data->postfix);
2589 } else {
2590 assert(idx == 1);
2591 rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2592 }
2593
2594 ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2595 g_free(rel_filename);
2596
2597 if (vmdk_create_extent(ext_filename, size,
2598 flat, compress, zeroed_grain, &blk, data->opts,
2599 errp)) {
2600 goto exit;
2601 }
2602 bdrv_unref(bs);
2603 exit:
2604 g_free(ext_filename);
2605 return blk;
2606 }
2607
2608 static int coroutine_fn vmdk_co_create_opts(BlockDriver *drv,
2609 const char *filename,
2610 QemuOpts *opts,
2611 Error **errp)
2612 {
2613 Error *local_err = NULL;
2614 char *desc = NULL;
2615 int64_t total_size = 0;
2616 char *adapter_type = NULL;
2617 BlockdevVmdkAdapterType adapter_type_enum;
2618 char *backing_file = NULL;
2619 char *hw_version = NULL;
2620 char *fmt = NULL;
2621 BlockdevVmdkSubformat subformat;
2622 int ret = 0;
2623 char *path = g_malloc0(PATH_MAX);
2624 char *prefix = g_malloc0(PATH_MAX);
2625 char *postfix = g_malloc0(PATH_MAX);
2626 char *desc_line = g_malloc0(BUF_SIZE);
2627 char *ext_filename = g_malloc0(PATH_MAX);
2628 char *desc_filename = g_malloc0(PATH_MAX);
2629 char *parent_desc_line = g_malloc0(BUF_SIZE);
2630 bool zeroed_grain;
2631 bool compat6;
2632 VMDKCreateOptsData data;
2633 char *backing_fmt = NULL;
2634
2635 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2636 if (backing_fmt && strcmp(backing_fmt, "vmdk") != 0) {
2637 error_setg(errp, "backing_file must be a vmdk image");
2638 ret = -EINVAL;
2639 goto exit;
2640 }
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(backing_fmt);
2700 g_free(adapter_type);
2701 g_free(backing_file);
2702 g_free(hw_version);
2703 g_free(fmt);
2704 g_free(desc);
2705 g_free(path);
2706 g_free(prefix);
2707 g_free(postfix);
2708 g_free(desc_line);
2709 g_free(ext_filename);
2710 g_free(desc_filename);
2711 g_free(parent_desc_line);
2712 return ret;
2713 }
2714
2715 static BlockBackend *vmdk_co_create_cb(int64_t size, int idx,
2716 bool flat, bool split, bool compress,
2717 bool zeroed_grain, void *opaque,
2718 Error **errp)
2719 {
2720 int ret;
2721 BlockDriverState *bs;
2722 BlockBackend *blk;
2723 BlockdevCreateOptionsVmdk *opts = opaque;
2724
2725 if (idx == 0) {
2726 bs = bdrv_open_blockdev_ref(opts->file, errp);
2727 } else {
2728 int i;
2729 BlockdevRefList *list = opts->extents;
2730 for (i = 1; i < idx; i++) {
2731 if (!list || !list->next) {
2732 error_setg(errp, "Extent [%d] not specified", i);
2733 return NULL;
2734 }
2735 list = list->next;
2736 }
2737 if (!list) {
2738 error_setg(errp, "Extent [%d] not specified", idx - 1);
2739 return NULL;
2740 }
2741 bs = bdrv_open_blockdev_ref(list->value, errp);
2742 }
2743 if (!bs) {
2744 return NULL;
2745 }
2746 blk = blk_new_with_bs(bs,
2747 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | BLK_PERM_RESIZE,
2748 BLK_PERM_ALL, errp);
2749 if (!blk) {
2750 return NULL;
2751 }
2752 blk_set_allow_write_beyond_eof(blk, true);
2753 bdrv_unref(bs);
2754
2755 if (size != -1) {
2756 ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2757 if (ret) {
2758 blk_unref(blk);
2759 blk = NULL;
2760 }
2761 }
2762 return blk;
2763 }
2764
2765 static int coroutine_fn vmdk_co_create(BlockdevCreateOptions *create_options,
2766 Error **errp)
2767 {
2768 int ret;
2769 BlockdevCreateOptionsVmdk *opts;
2770
2771 opts = &create_options->u.vmdk;
2772
2773 /* Validate options */
2774 if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2775 error_setg(errp, "Image size must be a multiple of 512 bytes");
2776 ret = -EINVAL;
2777 goto out;
2778 }
2779
2780 ret = vmdk_co_do_create(opts->size,
2781 opts->subformat,
2782 opts->adapter_type,
2783 opts->backing_file,
2784 opts->hwversion,
2785 false,
2786 opts->zeroed_grain,
2787 vmdk_co_create_cb,
2788 opts, errp);
2789 return ret;
2790
2791 out:
2792 return ret;
2793 }
2794
2795 static void vmdk_close(BlockDriverState *bs)
2796 {
2797 BDRVVmdkState *s = bs->opaque;
2798
2799 vmdk_free_extents(bs);
2800 g_free(s->create_type);
2801
2802 migrate_del_blocker(s->migration_blocker);
2803 error_free(s->migration_blocker);
2804 }
2805
2806 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2807 {
2808 int i;
2809 int64_t ret = 0;
2810 int64_t r;
2811 BDRVVmdkState *s = bs->opaque;
2812
2813 ret = bdrv_get_allocated_file_size(bs->file->bs);
2814 if (ret < 0) {
2815 return ret;
2816 }
2817 for (i = 0; i < s->num_extents; i++) {
2818 if (s->extents[i].file == bs->file) {
2819 continue;
2820 }
2821 r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2822 if (r < 0) {
2823 return r;
2824 }
2825 ret += r;
2826 }
2827 return ret;
2828 }
2829
2830 static int vmdk_has_zero_init(BlockDriverState *bs)
2831 {
2832 int i;
2833 BDRVVmdkState *s = bs->opaque;
2834
2835 /* If has a flat extent and its underlying storage doesn't have zero init,
2836 * return 0. */
2837 for (i = 0; i < s->num_extents; i++) {
2838 if (s->extents[i].flat) {
2839 if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2840 return 0;
2841 }
2842 }
2843 }
2844 return 1;
2845 }
2846
2847 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2848 {
2849 ImageInfo *info = g_new0(ImageInfo, 1);
2850
2851 bdrv_refresh_filename(extent->file->bs);
2852 *info = (ImageInfo){
2853 .filename = g_strdup(extent->file->bs->filename),
2854 .format = g_strdup(extent->type),
2855 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2856 .compressed = extent->compressed,
2857 .has_compressed = extent->compressed,
2858 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2859 .has_cluster_size = !extent->flat,
2860 };
2861
2862 return info;
2863 }
2864
2865 static int coroutine_fn vmdk_co_check(BlockDriverState *bs,
2866 BdrvCheckResult *result,
2867 BdrvCheckMode fix)
2868 {
2869 BDRVVmdkState *s = bs->opaque;
2870 VmdkExtent *extent = NULL;
2871 int64_t sector_num = 0;
2872 int64_t total_sectors = bdrv_nb_sectors(bs);
2873 int ret;
2874 uint64_t cluster_offset;
2875
2876 if (fix) {
2877 return -ENOTSUP;
2878 }
2879
2880 for (;;) {
2881 if (sector_num >= total_sectors) {
2882 return 0;
2883 }
2884 extent = find_extent(s, sector_num, extent);
2885 if (!extent) {
2886 fprintf(stderr,
2887 "ERROR: could not find extent for sector %" PRId64 "\n",
2888 sector_num);
2889 ret = -EINVAL;
2890 break;
2891 }
2892 ret = get_cluster_offset(bs, extent, NULL,
2893 sector_num << BDRV_SECTOR_BITS,
2894 false, &cluster_offset, 0, 0);
2895 if (ret == VMDK_ERROR) {
2896 fprintf(stderr,
2897 "ERROR: could not get cluster_offset for sector %"
2898 PRId64 "\n", sector_num);
2899 break;
2900 }
2901 if (ret == VMDK_OK) {
2902 int64_t extent_len = bdrv_getlength(extent->file->bs);
2903 if (extent_len < 0) {
2904 fprintf(stderr,
2905 "ERROR: could not get extent file length for sector %"
2906 PRId64 "\n", sector_num);
2907 ret = extent_len;
2908 break;
2909 }
2910 if (cluster_offset >= extent_len) {
2911 fprintf(stderr,
2912 "ERROR: cluster offset for sector %"
2913 PRId64 " points after EOF\n", sector_num);
2914 ret = -EINVAL;
2915 break;
2916 }
2917 }
2918 sector_num += extent->cluster_sectors;
2919 }
2920
2921 result->corruptions++;
2922 return ret;
2923 }
2924
2925 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs,
2926 Error **errp)
2927 {
2928 int i;
2929 BDRVVmdkState *s = bs->opaque;
2930 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2931 ImageInfoList **next;
2932
2933 *spec_info = (ImageInfoSpecific){
2934 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2935 .u = {
2936 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2937 },
2938 };
2939
2940 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2941 .create_type = g_strdup(s->create_type),
2942 .cid = s->cid,
2943 .parent_cid = s->parent_cid,
2944 };
2945
2946 next = &spec_info->u.vmdk.data->extents;
2947 for (i = 0; i < s->num_extents; i++) {
2948 *next = g_new0(ImageInfoList, 1);
2949 (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2950 (*next)->next = NULL;
2951 next = &(*next)->next;
2952 }
2953
2954 return spec_info;
2955 }
2956
2957 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2958 {
2959 return a->flat == b->flat &&
2960 a->compressed == b->compressed &&
2961 (a->flat || a->cluster_sectors == b->cluster_sectors);
2962 }
2963
2964 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2965 {
2966 int i;
2967 BDRVVmdkState *s = bs->opaque;
2968 assert(s->num_extents);
2969
2970 /* See if we have multiple extents but they have different cases */
2971 for (i = 1; i < s->num_extents; i++) {
2972 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2973 return -ENOTSUP;
2974 }
2975 }
2976 bdi->needs_compressed_writes = s->extents[0].compressed;
2977 if (!s->extents[0].flat) {
2978 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2979 }
2980 return 0;
2981 }
2982
2983 static void vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
2984 bool backing_overridden)
2985 {
2986 /* No children but file and backing can be explicitly specified (TODO) */
2987 qdict_put(target, "file",
2988 qobject_ref(bs->file->bs->full_open_options));
2989
2990 if (backing_overridden) {
2991 if (bs->backing) {
2992 qdict_put(target, "backing",
2993 qobject_ref(bs->backing->bs->full_open_options));
2994 } else {
2995 qdict_put_null(target, "backing");
2996 }
2997 }
2998 }
2999
3000 static QemuOptsList vmdk_create_opts = {
3001 .name = "vmdk-create-opts",
3002 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
3003 .desc = {
3004 {
3005 .name = BLOCK_OPT_SIZE,
3006 .type = QEMU_OPT_SIZE,
3007 .help = "Virtual disk size"
3008 },
3009 {
3010 .name = BLOCK_OPT_ADAPTER_TYPE,
3011 .type = QEMU_OPT_STRING,
3012 .help = "Virtual adapter type, can be one of "
3013 "ide (default), lsilogic, buslogic or legacyESX"
3014 },
3015 {
3016 .name = BLOCK_OPT_BACKING_FILE,
3017 .type = QEMU_OPT_STRING,
3018 .help = "File name of a base image"
3019 },
3020 {
3021 .name = BLOCK_OPT_BACKING_FMT,
3022 .type = QEMU_OPT_STRING,
3023 .help = "Must be 'vmdk' if present",
3024 },
3025 {
3026 .name = BLOCK_OPT_COMPAT6,
3027 .type = QEMU_OPT_BOOL,
3028 .help = "VMDK version 6 image",
3029 .def_value_str = "off"
3030 },
3031 {
3032 .name = BLOCK_OPT_HWVERSION,
3033 .type = QEMU_OPT_STRING,
3034 .help = "VMDK hardware version",
3035 .def_value_str = "undefined"
3036 },
3037 {
3038 .name = BLOCK_OPT_SUBFMT,
3039 .type = QEMU_OPT_STRING,
3040 .help =
3041 "VMDK flat extent format, can be one of "
3042 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
3043 },
3044 {
3045 .name = BLOCK_OPT_ZEROED_GRAIN,
3046 .type = QEMU_OPT_BOOL,
3047 .help = "Enable efficient zero writes "
3048 "using the zeroed-grain GTE feature"
3049 },
3050 { /* end of list */ }
3051 }
3052 };
3053
3054 static BlockDriver bdrv_vmdk = {
3055 .format_name = "vmdk",
3056 .instance_size = sizeof(BDRVVmdkState),
3057 .bdrv_probe = vmdk_probe,
3058 .bdrv_open = vmdk_open,
3059 .bdrv_co_check = vmdk_co_check,
3060 .bdrv_reopen_prepare = vmdk_reopen_prepare,
3061 .bdrv_child_perm = bdrv_default_perms,
3062 .bdrv_co_preadv = vmdk_co_preadv,
3063 .bdrv_co_pwritev = vmdk_co_pwritev,
3064 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
3065 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
3066 .bdrv_close = vmdk_close,
3067 .bdrv_co_create_opts = vmdk_co_create_opts,
3068 .bdrv_co_create = vmdk_co_create,
3069 .bdrv_co_block_status = vmdk_co_block_status,
3070 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
3071 .bdrv_has_zero_init = vmdk_has_zero_init,
3072 .bdrv_get_specific_info = vmdk_get_specific_info,
3073 .bdrv_refresh_limits = vmdk_refresh_limits,
3074 .bdrv_get_info = vmdk_get_info,
3075 .bdrv_gather_child_options = vmdk_gather_child_options,
3076
3077 .is_format = true,
3078 .supports_backing = true,
3079 .create_opts = &vmdk_create_opts,
3080 };
3081
3082 static void bdrv_vmdk_init(void)
3083 {
3084 bdrv_register(&bdrv_vmdk);
3085 }
3086
3087 block_init(bdrv_vmdk_init);
QEmu