?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
hw/intc/arm_gicv3_its: Factor out "is intid a valid LPI ID?"
[qemu/rayw.git] / block / vmdk.c
blob37c0946066ee9ea1f904234eced1651018ab95d8
1 /*
2 * Block driver for the VMDK format
3 *
4 * Copyright (c) 2004 Fabrice Bellard
5 * Copyright (c) 2005 Filip Navara
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23 * THE SOFTWARE.
24 */
25
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "block/block_int.h"
29 #include "sysemu/block-backend.h"
30 #include "qapi/qmp/qdict.h"
31 #include "qapi/qmp/qerror.h"
32 #include "qemu/error-report.h"
33 #include "qemu/module.h"
34 #include "qemu/option.h"
35 #include "qemu/bswap.h"
36 #include "qemu/memalign.h"
37 #include "migration/blocker.h"
38 #include "qemu/cutils.h"
39 #include <zlib.h>
40
41 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
42 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
43 #define VMDK4_COMPRESSION_DEFLATE 1
44 #define VMDK4_FLAG_NL_DETECT (1 << 0)
45 #define VMDK4_FLAG_RGD (1 << 1)
46 /* Zeroed-grain enable bit */
47 #define VMDK4_FLAG_ZERO_GRAIN (1 << 2)
48 #define VMDK4_FLAG_COMPRESS (1 << 16)
49 #define VMDK4_FLAG_MARKER (1 << 17)
50 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
51
52 #define VMDK_EXTENT_MAX_SECTORS (1ULL << 32)
53
54 #define VMDK_GTE_ZEROED 0x1
55
56 /* VMDK internal error codes */
57 #define VMDK_OK 0
58 #define VMDK_ERROR (-1)
59 /* Cluster not allocated */
60 #define VMDK_UNALLOC (-2)
61 #define VMDK_ZEROED (-3)
62
63 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
64 #define BLOCK_OPT_TOOLSVERSION "toolsversion"
65
66 typedef struct {
67 uint32_t version;
68 uint32_t flags;
69 uint32_t disk_sectors;
70 uint32_t granularity;
71 uint32_t l1dir_offset;
72 uint32_t l1dir_size;
73 uint32_t file_sectors;
74 uint32_t cylinders;
75 uint32_t heads;
76 uint32_t sectors_per_track;
77 } QEMU_PACKED VMDK3Header;
78
79 typedef struct {
80 uint32_t version;
81 uint32_t flags;
82 uint64_t capacity;
83 uint64_t granularity;
84 uint64_t desc_offset;
85 uint64_t desc_size;
86 /* Number of GrainTableEntries per GrainTable */
87 uint32_t num_gtes_per_gt;
88 uint64_t rgd_offset;
89 uint64_t gd_offset;
90 uint64_t grain_offset;
91 char filler[1];
92 char check_bytes[4];
93 uint16_t compressAlgorithm;
94 } QEMU_PACKED VMDK4Header;
95
96 typedef struct VMDKSESparseConstHeader {
97 uint64_t magic;
98 uint64_t version;
99 uint64_t capacity;
100 uint64_t grain_size;
101 uint64_t grain_table_size;
102 uint64_t flags;
103 uint64_t reserved1;
104 uint64_t reserved2;
105 uint64_t reserved3;
106 uint64_t reserved4;
107 uint64_t volatile_header_offset;
108 uint64_t volatile_header_size;
109 uint64_t journal_header_offset;
110 uint64_t journal_header_size;
111 uint64_t journal_offset;
112 uint64_t journal_size;
113 uint64_t grain_dir_offset;
114 uint64_t grain_dir_size;
115 uint64_t grain_tables_offset;
116 uint64_t grain_tables_size;
117 uint64_t free_bitmap_offset;
118 uint64_t free_bitmap_size;
119 uint64_t backmap_offset;
120 uint64_t backmap_size;
121 uint64_t grains_offset;
122 uint64_t grains_size;
123 uint8_t pad[304];
124 } QEMU_PACKED VMDKSESparseConstHeader;
125
126 typedef struct VMDKSESparseVolatileHeader {
127 uint64_t magic;
128 uint64_t free_gt_number;
129 uint64_t next_txn_seq_number;
130 uint64_t replay_journal;
131 uint8_t pad[480];
132 } QEMU_PACKED VMDKSESparseVolatileHeader;
133
134 #define L2_CACHE_SIZE 16
135
136 typedef struct VmdkExtent {
137 BdrvChild *file;
138 bool flat;
139 bool compressed;
140 bool has_marker;
141 bool has_zero_grain;
142 bool sesparse;
143 uint64_t sesparse_l2_tables_offset;
144 uint64_t sesparse_clusters_offset;
145 int32_t entry_size;
146 int version;
147 int64_t sectors;
148 int64_t end_sector;
149 int64_t flat_start_offset;
150 int64_t l1_table_offset;
151 int64_t l1_backup_table_offset;
152 void *l1_table;
153 uint32_t *l1_backup_table;
154 unsigned int l1_size;
155 uint32_t l1_entry_sectors;
156
157 unsigned int l2_size;
158 void *l2_cache;
159 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
160 uint32_t l2_cache_counts[L2_CACHE_SIZE];
161
162 int64_t cluster_sectors;
163 int64_t next_cluster_sector;
164 char *type;
165 } VmdkExtent;
166
167 typedef struct BDRVVmdkState {
168 CoMutex lock;
169 uint64_t desc_offset;
170 bool cid_updated;
171 bool cid_checked;
172 uint32_t cid;
173 uint32_t parent_cid;
174 int num_extents;
175 /* Extent array with num_extents entries, ascend ordered by address */
176 VmdkExtent *extents;
177 Error *migration_blocker;
178 char *create_type;
179 } BDRVVmdkState;
180
181 typedef struct VmdkMetaData {
182 unsigned int l1_index;
183 unsigned int l2_index;
184 unsigned int l2_offset;
185 bool new_allocation;
186 uint32_t *l2_cache_entry;
187 } VmdkMetaData;
188
189 typedef struct VmdkGrainMarker {
190 uint64_t lba;
191 uint32_t size;
192 uint8_t data[];
193 } QEMU_PACKED VmdkGrainMarker;
194
195 enum {
196 MARKER_END_OF_STREAM = 0,
197 MARKER_GRAIN_TABLE = 1,
198 MARKER_GRAIN_DIRECTORY = 2,
199 MARKER_FOOTER = 3,
200 };
201
202 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
203 {
204 uint32_t magic;
205
206 if (buf_size < 4) {
207 return 0;
208 }
209 magic = be32_to_cpu(*(uint32_t *)buf);
210 if (magic == VMDK3_MAGIC ||
211 magic == VMDK4_MAGIC) {
212 return 100;
213 } else {
214 const char *p = (const char *)buf;
215 const char *end = p + buf_size;
216 while (p < end) {
217 if (*p == '#') {
218 /* skip comment line */
219 while (p < end && *p != '\n') {
220 p++;
221 }
222 p++;
223 continue;
224 }
225 if (*p == ' ') {
226 while (p < end && *p == ' ') {
227 p++;
228 }
229 /* skip '\r' if windows line endings used. */
230 if (p < end && *p == '\r') {
231 p++;
232 }
233 /* only accept blank lines before 'version=' line */
234 if (p == end || *p != '\n') {
235 return 0;
236 }
237 p++;
238 continue;
239 }
240 if (end - p >= strlen("version=X\n")) {
241 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
242 strncmp("version=2\n", p, strlen("version=2\n")) == 0 ||
243 strncmp("version=3\n", p, strlen("version=3\n")) == 0) {
244 return 100;
245 }
246 }
247 if (end - p >= strlen("version=X\r\n")) {
248 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
249 strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0 ||
250 strncmp("version=3\r\n", p, strlen("version=3\r\n")) == 0) {
251 return 100;
252 }
253 }
254 return 0;
255 }
256 return 0;
257 }
258 }
259
260 #define SECTOR_SIZE 512
261 #define DESC_SIZE (20 * SECTOR_SIZE) /* 20 sectors of 512 bytes each */
262 #define BUF_SIZE 4096
263 #define HEADER_SIZE 512 /* first sector of 512 bytes */
264
265 static void vmdk_free_extents(BlockDriverState *bs)
266 {
267 int i;
268 BDRVVmdkState *s = bs->opaque;
269 VmdkExtent *e;
270
271 for (i = 0; i < s->num_extents; i++) {
272 e = &s->extents[i];
273 g_free(e->l1_table);
274 g_free(e->l2_cache);
275 g_free(e->l1_backup_table);
276 g_free(e->type);
277 if (e->file != bs->file) {
278 bdrv_unref_child(bs, e->file);
279 }
280 }
281 g_free(s->extents);
282 }
283
284 static void vmdk_free_last_extent(BlockDriverState *bs)
285 {
286 BDRVVmdkState *s = bs->opaque;
287
288 if (s->num_extents == 0) {
289 return;
290 }
291 s->num_extents--;
292 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
293 }
294
295 /* Return -ve errno, or 0 on success and write CID into *pcid. */
296 static int vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
297 {
298 char *desc;
299 uint32_t cid;
300 const char *p_name, *cid_str;
301 size_t cid_str_size;
302 BDRVVmdkState *s = bs->opaque;
303 int ret;
304
305 desc = g_malloc0(DESC_SIZE);
306 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
307 if (ret < 0) {
308 goto out;
309 }
310
311 if (parent) {
312 cid_str = "parentCID";
313 cid_str_size = sizeof("parentCID");
314 } else {
315 cid_str = "CID";
316 cid_str_size = sizeof("CID");
317 }
318
319 desc[DESC_SIZE - 1] = '\0';
320 p_name = strstr(desc, cid_str);
321 if (p_name == NULL) {
322 ret = -EINVAL;
323 goto out;
324 }
325 p_name += cid_str_size;
326 if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
327 ret = -EINVAL;
328 goto out;
329 }
330 *pcid = cid;
331 ret = 0;
332
333 out:
334 g_free(desc);
335 return ret;
336 }
337
338 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
339 {
340 char *desc, *tmp_desc;
341 char *p_name, *tmp_str;
342 BDRVVmdkState *s = bs->opaque;
343 int ret = 0;
344
345 desc = g_malloc0(DESC_SIZE);
346 tmp_desc = g_malloc0(DESC_SIZE);
347 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
348 if (ret < 0) {
349 goto out;
350 }
351
352 desc[DESC_SIZE - 1] = '\0';
353 tmp_str = strstr(desc, "parentCID");
354 if (tmp_str == NULL) {
355 ret = -EINVAL;
356 goto out;
357 }
358
359 pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
360 p_name = strstr(desc, "CID");
361 if (p_name != NULL) {
362 p_name += sizeof("CID");
363 snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
364 pstrcat(desc, DESC_SIZE, tmp_desc);
365 }
366
367 ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
368
369 out:
370 g_free(desc);
371 g_free(tmp_desc);
372 return ret;
373 }
374
375 static int vmdk_is_cid_valid(BlockDriverState *bs)
376 {
377 BDRVVmdkState *s = bs->opaque;
378 uint32_t cur_pcid;
379
380 if (!s->cid_checked && bs->backing) {
381 BlockDriverState *p_bs = bs->backing->bs;
382
383 if (strcmp(p_bs->drv->format_name, "vmdk")) {
384 /* Backing file is not in vmdk format, so it does not have
385 * a CID, which makes the overlay's parent CID invalid */
386 return 0;
387 }
388
389 if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
390 /* read failure: report as not valid */
391 return 0;
392 }
393 if (s->parent_cid != cur_pcid) {
394 /* CID not valid */
395 return 0;
396 }
397 }
398 s->cid_checked = true;
399 /* CID valid */
400 return 1;
401 }
402
403 /* We have nothing to do for VMDK reopen, stubs just return success */
404 static int vmdk_reopen_prepare(BDRVReopenState *state,
405 BlockReopenQueue *queue, Error **errp)
406 {
407 assert(state != NULL);
408 assert(state->bs != NULL);
409 return 0;
410 }
411
412 static int vmdk_parent_open(BlockDriverState *bs)
413 {
414 char *p_name;
415 char *desc;
416 BDRVVmdkState *s = bs->opaque;
417 int ret;
418
419 desc = g_malloc0(DESC_SIZE + 1);
420 ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
421 if (ret < 0) {
422 goto out;
423 }
424 ret = 0;
425
426 p_name = strstr(desc, "parentFileNameHint");
427 if (p_name != NULL) {
428 char *end_name;
429
430 p_name += sizeof("parentFileNameHint") + 1;
431 end_name = strchr(p_name, '\"');
432 if (end_name == NULL) {
433 ret = -EINVAL;
434 goto out;
435 }
436 if ((end_name - p_name) > sizeof(bs->auto_backing_file) - 1) {
437 ret = -EINVAL;
438 goto out;
439 }
440
441 pstrcpy(bs->auto_backing_file, end_name - p_name + 1, p_name);
442 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
443 bs->auto_backing_file);
444 pstrcpy(bs->backing_format, sizeof(bs->backing_format),
445 "vmdk");
446 }
447
448 out:
449 g_free(desc);
450 return ret;
451 }
452
453 /* Create and append extent to the extent array. Return the added VmdkExtent
454 * address. return NULL if allocation failed. */
455 static int vmdk_add_extent(BlockDriverState *bs,
456 BdrvChild *file, bool flat, int64_t sectors,
457 int64_t l1_offset, int64_t l1_backup_offset,
458 uint32_t l1_size,
459 int l2_size, uint64_t cluster_sectors,
460 VmdkExtent **new_extent,
461 Error **errp)
462 {
463 VmdkExtent *extent;
464 BDRVVmdkState *s = bs->opaque;
465 int64_t nb_sectors;
466
467 if (cluster_sectors > 0x200000) {
468 /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
469 error_setg(errp, "Invalid granularity, image may be corrupt");
470 return -EFBIG;
471 }
472 if (l1_size > 32 * 1024 * 1024) {
473 /*
474 * Although with big capacity and small l1_entry_sectors, we can get a
475 * big l1_size, we don't want unbounded value to allocate the table.
476 * Limit it to 32M, which is enough to store:
477 * 8TB - for both VMDK3 & VMDK4 with
478 * minimal cluster size: 512B
479 * minimal L2 table size: 512 entries
480 * 8 TB is still more than the maximal value supported for
481 * VMDK3 & VMDK4 which is 2TB.
482 * 64TB - for "ESXi seSparse Extent"
483 * minimal cluster size: 512B (default is 4KB)
484 * L2 table size: 4096 entries (const).
485 * 64TB is more than the maximal value supported for
486 * seSparse VMDKs (which is slightly less than 64TB)
487 */
488 error_setg(errp, "L1 size too big");
489 return -EFBIG;
490 }
491
492 nb_sectors = bdrv_nb_sectors(file->bs);
493 if (nb_sectors < 0) {
494 return nb_sectors;
495 }
496
497 s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
498 extent = &s->extents[s->num_extents];
499 s->num_extents++;
500
501 memset(extent, 0, sizeof(VmdkExtent));
502 extent->file = file;
503 extent->flat = flat;
504 extent->sectors = sectors;
505 extent->l1_table_offset = l1_offset;
506 extent->l1_backup_table_offset = l1_backup_offset;
507 extent->l1_size = l1_size;
508 extent->l1_entry_sectors = l2_size * cluster_sectors;
509 extent->l2_size = l2_size;
510 extent->cluster_sectors = flat ? sectors : cluster_sectors;
511 extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
512 extent->entry_size = sizeof(uint32_t);
513
514 if (s->num_extents > 1) {
515 extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
516 } else {
517 extent->end_sector = extent->sectors;
518 }
519 bs->total_sectors = extent->end_sector;
520 if (new_extent) {
521 *new_extent = extent;
522 }
523 return 0;
524 }
525
526 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
527 Error **errp)
528 {
529 int ret;
530 size_t l1_size;
531 int i;
532
533 /* read the L1 table */
534 l1_size = extent->l1_size * extent->entry_size;
535 extent->l1_table = g_try_malloc(l1_size);
536 if (l1_size && extent->l1_table == NULL) {
537 return -ENOMEM;
538 }
539
540 ret = bdrv_pread(extent->file,
541 extent->l1_table_offset,
542 extent->l1_table,
543 l1_size);
544 if (ret < 0) {
545 bdrv_refresh_filename(extent->file->bs);
546 error_setg_errno(errp, -ret,
547 "Could not read l1 table from extent '%s'",
548 extent->file->bs->filename);
549 goto fail_l1;
550 }
551 for (i = 0; i < extent->l1_size; i++) {
552 if (extent->entry_size == sizeof(uint64_t)) {
553 le64_to_cpus((uint64_t *)extent->l1_table + i);
554 } else {
555 assert(extent->entry_size == sizeof(uint32_t));
556 le32_to_cpus((uint32_t *)extent->l1_table + i);
557 }
558 }
559
560 if (extent->l1_backup_table_offset) {
561 assert(!extent->sesparse);
562 extent->l1_backup_table = g_try_malloc(l1_size);
563 if (l1_size && extent->l1_backup_table == NULL) {
564 ret = -ENOMEM;
565 goto fail_l1;
566 }
567 ret = bdrv_pread(extent->file,
568 extent->l1_backup_table_offset,
569 extent->l1_backup_table,
570 l1_size);
571 if (ret < 0) {
572 bdrv_refresh_filename(extent->file->bs);
573 error_setg_errno(errp, -ret,
574 "Could not read l1 backup table from extent '%s'",
575 extent->file->bs->filename);
576 goto fail_l1b;
577 }
578 for (i = 0; i < extent->l1_size; i++) {
579 le32_to_cpus(&extent->l1_backup_table[i]);
580 }
581 }
582
583 extent->l2_cache =
584 g_malloc(extent->entry_size * extent->l2_size * L2_CACHE_SIZE);
585 return 0;
586 fail_l1b:
587 g_free(extent->l1_backup_table);
588 fail_l1:
589 g_free(extent->l1_table);
590 return ret;
591 }
592
593 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
594 BdrvChild *file,
595 int flags, Error **errp)
596 {
597 int ret;
598 uint32_t magic;
599 VMDK3Header header;
600 VmdkExtent *extent = NULL;
601
602 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
603 if (ret < 0) {
604 bdrv_refresh_filename(file->bs);
605 error_setg_errno(errp, -ret,
606 "Could not read header from file '%s'",
607 file->bs->filename);
608 return ret;
609 }
610 ret = vmdk_add_extent(bs, file, false,
611 le32_to_cpu(header.disk_sectors),
612 (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
613 0,
614 le32_to_cpu(header.l1dir_size),
615 4096,
616 le32_to_cpu(header.granularity),
617 &extent,
618 errp);
619 if (ret < 0) {
620 return ret;
621 }
622 ret = vmdk_init_tables(bs, extent, errp);
623 if (ret) {
624 /* free extent allocated by vmdk_add_extent */
625 vmdk_free_last_extent(bs);
626 }
627 return ret;
628 }
629
630 #define SESPARSE_CONST_HEADER_MAGIC UINT64_C(0x00000000cafebabe)
631 #define SESPARSE_VOLATILE_HEADER_MAGIC UINT64_C(0x00000000cafecafe)
632
633 /* Strict checks - format not officially documented */
634 static int check_se_sparse_const_header(VMDKSESparseConstHeader *header,
635 Error **errp)
636 {
637 header->magic = le64_to_cpu(header->magic);
638 header->version = le64_to_cpu(header->version);
639 header->grain_size = le64_to_cpu(header->grain_size);
640 header->grain_table_size = le64_to_cpu(header->grain_table_size);
641 header->flags = le64_to_cpu(header->flags);
642 header->reserved1 = le64_to_cpu(header->reserved1);
643 header->reserved2 = le64_to_cpu(header->reserved2);
644 header->reserved3 = le64_to_cpu(header->reserved3);
645 header->reserved4 = le64_to_cpu(header->reserved4);
646
647 header->volatile_header_offset =
648 le64_to_cpu(header->volatile_header_offset);
649 header->volatile_header_size = le64_to_cpu(header->volatile_header_size);
650
651 header->journal_header_offset = le64_to_cpu(header->journal_header_offset);
652 header->journal_header_size = le64_to_cpu(header->journal_header_size);
653
654 header->journal_offset = le64_to_cpu(header->journal_offset);
655 header->journal_size = le64_to_cpu(header->journal_size);
656
657 header->grain_dir_offset = le64_to_cpu(header->grain_dir_offset);
658 header->grain_dir_size = le64_to_cpu(header->grain_dir_size);
659
660 header->grain_tables_offset = le64_to_cpu(header->grain_tables_offset);
661 header->grain_tables_size = le64_to_cpu(header->grain_tables_size);
662
663 header->free_bitmap_offset = le64_to_cpu(header->free_bitmap_offset);
664 header->free_bitmap_size = le64_to_cpu(header->free_bitmap_size);
665
666 header->backmap_offset = le64_to_cpu(header->backmap_offset);
667 header->backmap_size = le64_to_cpu(header->backmap_size);
668
669 header->grains_offset = le64_to_cpu(header->grains_offset);
670 header->grains_size = le64_to_cpu(header->grains_size);
671
672 if (header->magic != SESPARSE_CONST_HEADER_MAGIC) {
673 error_setg(errp, "Bad const header magic: 0x%016" PRIx64,
674 header->magic);
675 return -EINVAL;
676 }
677
678 if (header->version != 0x0000000200000001) {
679 error_setg(errp, "Unsupported version: 0x%016" PRIx64,
680 header->version);
681 return -ENOTSUP;
682 }
683
684 if (header->grain_size != 8) {
685 error_setg(errp, "Unsupported grain size: %" PRIu64,
686 header->grain_size);
687 return -ENOTSUP;
688 }
689
690 if (header->grain_table_size != 64) {
691 error_setg(errp, "Unsupported grain table size: %" PRIu64,
692 header->grain_table_size);
693 return -ENOTSUP;
694 }
695
696 if (header->flags != 0) {
697 error_setg(errp, "Unsupported flags: 0x%016" PRIx64,
698 header->flags);
699 return -ENOTSUP;
700 }
701
702 if (header->reserved1 != 0 || header->reserved2 != 0 ||
703 header->reserved3 != 0 || header->reserved4 != 0) {
704 error_setg(errp, "Unsupported reserved bits:"
705 " 0x%016" PRIx64 " 0x%016" PRIx64
706 " 0x%016" PRIx64 " 0x%016" PRIx64,
707 header->reserved1, header->reserved2,
708 header->reserved3, header->reserved4);
709 return -ENOTSUP;
710 }
711
712 /* check that padding is 0 */
713 if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
714 error_setg(errp, "Unsupported non-zero const header padding");
715 return -ENOTSUP;
716 }
717
718 return 0;
719 }
720
721 static int check_se_sparse_volatile_header(VMDKSESparseVolatileHeader *header,
722 Error **errp)
723 {
724 header->magic = le64_to_cpu(header->magic);
725 header->free_gt_number = le64_to_cpu(header->free_gt_number);
726 header->next_txn_seq_number = le64_to_cpu(header->next_txn_seq_number);
727 header->replay_journal = le64_to_cpu(header->replay_journal);
728
729 if (header->magic != SESPARSE_VOLATILE_HEADER_MAGIC) {
730 error_setg(errp, "Bad volatile header magic: 0x%016" PRIx64,
731 header->magic);
732 return -EINVAL;
733 }
734
735 if (header->replay_journal) {
736 error_setg(errp, "Image is dirty, Replaying journal not supported");
737 return -ENOTSUP;
738 }
739
740 /* check that padding is 0 */
741 if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
742 error_setg(errp, "Unsupported non-zero volatile header padding");
743 return -ENOTSUP;
744 }
745
746 return 0;
747 }
748
749 static int vmdk_open_se_sparse(BlockDriverState *bs,
750 BdrvChild *file,
751 int flags, Error **errp)
752 {
753 int ret;
754 VMDKSESparseConstHeader const_header;
755 VMDKSESparseVolatileHeader volatile_header;
756 VmdkExtent *extent = NULL;
757
758 ret = bdrv_apply_auto_read_only(bs,
759 "No write support for seSparse images available", errp);
760 if (ret < 0) {
761 return ret;
762 }
763
764 assert(sizeof(const_header) == SECTOR_SIZE);
765
766 ret = bdrv_pread(file, 0, &const_header, sizeof(const_header));
767 if (ret < 0) {
768 bdrv_refresh_filename(file->bs);
769 error_setg_errno(errp, -ret,
770 "Could not read const header from file '%s'",
771 file->bs->filename);
772 return ret;
773 }
774
775 /* check const header */
776 ret = check_se_sparse_const_header(&const_header, errp);
777 if (ret < 0) {
778 return ret;
779 }
780
781 assert(sizeof(volatile_header) == SECTOR_SIZE);
782
783 ret = bdrv_pread(file,
784 const_header.volatile_header_offset * SECTOR_SIZE,
785 &volatile_header, sizeof(volatile_header));
786 if (ret < 0) {
787 bdrv_refresh_filename(file->bs);
788 error_setg_errno(errp, -ret,
789 "Could not read volatile header from file '%s'",
790 file->bs->filename);
791 return ret;
792 }
793
794 /* check volatile header */
795 ret = check_se_sparse_volatile_header(&volatile_header, errp);
796 if (ret < 0) {
797 return ret;
798 }
799
800 ret = vmdk_add_extent(bs, file, false,
801 const_header.capacity,
802 const_header.grain_dir_offset * SECTOR_SIZE,
803 0,
804 const_header.grain_dir_size *
805 SECTOR_SIZE / sizeof(uint64_t),
806 const_header.grain_table_size *
807 SECTOR_SIZE / sizeof(uint64_t),
808 const_header.grain_size,
809 &extent,
810 errp);
811 if (ret < 0) {
812 return ret;
813 }
814
815 extent->sesparse = true;
816 extent->sesparse_l2_tables_offset = const_header.grain_tables_offset;
817 extent->sesparse_clusters_offset = const_header.grains_offset;
818 extent->entry_size = sizeof(uint64_t);
819
820 ret = vmdk_init_tables(bs, extent, errp);
821 if (ret) {
822 /* free extent allocated by vmdk_add_extent */
823 vmdk_free_last_extent(bs);
824 }
825
826 return ret;
827 }
828
829 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
830 QDict *options, Error **errp);
831
832 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
833 {
834 int64_t size;
835 char *buf;
836 int ret;
837
838 size = bdrv_getlength(file->bs);
839 if (size < 0) {
840 error_setg_errno(errp, -size, "Could not access file");
841 return NULL;
842 }
843
844 if (size < 4) {
845 /* Both descriptor file and sparse image must be much larger than 4
846 * bytes, also callers of vmdk_read_desc want to compare the first 4
847 * bytes with VMDK4_MAGIC, let's error out if less is read. */
848 error_setg(errp, "File is too small, not a valid image");
849 return NULL;
850 }
851
852 size = MIN(size, (1 << 20) - 1); /* avoid unbounded allocation */
853 buf = g_malloc(size + 1);
854
855 ret = bdrv_pread(file, desc_offset, buf, size);
856 if (ret < 0) {
857 error_setg_errno(errp, -ret, "Could not read from file");
858 g_free(buf);
859 return NULL;
860 }
861 buf[ret] = 0;
862
863 return buf;
864 }
865
866 static int vmdk_open_vmdk4(BlockDriverState *bs,
867 BdrvChild *file,
868 int flags, QDict *options, Error **errp)
869 {
870 int ret;
871 uint32_t magic;
872 uint32_t l1_size, l1_entry_sectors;
873 VMDK4Header header;
874 VmdkExtent *extent = NULL;
875 BDRVVmdkState *s = bs->opaque;
876 int64_t l1_backup_offset = 0;
877 bool compressed;
878
879 ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
880 if (ret < 0) {
881 bdrv_refresh_filename(file->bs);
882 error_setg_errno(errp, -ret,
883 "Could not read header from file '%s'",
884 file->bs->filename);
885 return -EINVAL;
886 }
887 if (header.capacity == 0) {
888 uint64_t desc_offset = le64_to_cpu(header.desc_offset);
889 if (desc_offset) {
890 char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
891 if (!buf) {
892 return -EINVAL;
893 }
894 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
895 g_free(buf);
896 return ret;
897 }
898 }
899
900 if (!s->create_type) {
901 s->create_type = g_strdup("monolithicSparse");
902 }
903
904 if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
905 /*
906 * The footer takes precedence over the header, so read it in. The
907 * footer starts at offset -1024 from the end: One sector for the
908 * footer, and another one for the end-of-stream marker.
909 */
910 struct {
911 struct {
912 uint64_t val;
913 uint32_t size;
914 uint32_t type;
915 uint8_t pad[512 - 16];
916 } QEMU_PACKED footer_marker;
917
918 uint32_t magic;
919 VMDK4Header header;
920 uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
921
922 struct {
923 uint64_t val;
924 uint32_t size;
925 uint32_t type;
926 uint8_t pad[512 - 16];
927 } QEMU_PACKED eos_marker;
928 } QEMU_PACKED footer;
929
930 ret = bdrv_pread(file,
931 bs->file->bs->total_sectors * 512 - 1536,
932 &footer, sizeof(footer));
933 if (ret < 0) {
934 error_setg_errno(errp, -ret, "Failed to read footer");
935 return ret;
936 }
937
938 /* Some sanity checks for the footer */
939 if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
940 le32_to_cpu(footer.footer_marker.size) != 0 ||
941 le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
942 le64_to_cpu(footer.eos_marker.val) != 0 ||
943 le32_to_cpu(footer.eos_marker.size) != 0 ||
944 le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
945 {
946 error_setg(errp, "Invalid footer");
947 return -EINVAL;
948 }
949
950 header = footer.header;
951 }
952
953 compressed =
954 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
955 if (le32_to_cpu(header.version) > 3) {
956 error_setg(errp, "Unsupported VMDK version %" PRIu32,
957 le32_to_cpu(header.version));
958 return -ENOTSUP;
959 } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
960 !compressed) {
961 /* VMware KB 2064959 explains that version 3 added support for
962 * persistent changed block tracking (CBT), and backup software can
963 * read it as version=1 if it doesn't care about the changed area
964 * information. So we are safe to enable read only. */
965 error_setg(errp, "VMDK version 3 must be read only");
966 return -EINVAL;
967 }
968
969 if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
970 error_setg(errp, "L2 table size too big");
971 return -EINVAL;
972 }
973
974 l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
975 * le64_to_cpu(header.granularity);
976 if (l1_entry_sectors == 0) {
977 error_setg(errp, "L1 entry size is invalid");
978 return -EINVAL;
979 }
980 l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
981 / l1_entry_sectors;
982 if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
983 l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
984 }
985 if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
986 error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
987 (int64_t)(le64_to_cpu(header.grain_offset)
988 * BDRV_SECTOR_SIZE));
989 return -EINVAL;
990 }
991
992 ret = vmdk_add_extent(bs, file, false,
993 le64_to_cpu(header.capacity),
994 le64_to_cpu(header.gd_offset) << 9,
995 l1_backup_offset,
996 l1_size,
997 le32_to_cpu(header.num_gtes_per_gt),
998 le64_to_cpu(header.granularity),
999 &extent,
1000 errp);
1001 if (ret < 0) {
1002 return ret;
1003 }
1004 extent->compressed =
1005 le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1006 if (extent->compressed) {
1007 g_free(s->create_type);
1008 s->create_type = g_strdup("streamOptimized");
1009 }
1010 extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
1011 extent->version = le32_to_cpu(header.version);
1012 extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
1013 ret = vmdk_init_tables(bs, extent, errp);
1014 if (ret) {
1015 /* free extent allocated by vmdk_add_extent */
1016 vmdk_free_last_extent(bs);
1017 }
1018 return ret;
1019 }
1020
1021 /* find an option value out of descriptor file */
1022 static int vmdk_parse_description(const char *desc, const char *opt_name,
1023 char *buf, int buf_size)
1024 {
1025 char *opt_pos, *opt_end;
1026 const char *end = desc + strlen(desc);
1027
1028 opt_pos = strstr(desc, opt_name);
1029 if (!opt_pos) {
1030 return VMDK_ERROR;
1031 }
1032 /* Skip "=\"" following opt_name */
1033 opt_pos += strlen(opt_name) + 2;
1034 if (opt_pos >= end) {
1035 return VMDK_ERROR;
1036 }
1037 opt_end = opt_pos;
1038 while (opt_end < end && *opt_end != '"') {
1039 opt_end++;
1040 }
1041 if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
1042 return VMDK_ERROR;
1043 }
1044 pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
1045 return VMDK_OK;
1046 }
1047
1048 /* Open an extent file and append to bs array */
1049 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
1050 char *buf, QDict *options, Error **errp)
1051 {
1052 uint32_t magic;
1053
1054 magic = ldl_be_p(buf);
1055 switch (magic) {
1056 case VMDK3_MAGIC:
1057 return vmdk_open_vmfs_sparse(bs, file, flags, errp);
1058 case VMDK4_MAGIC:
1059 return vmdk_open_vmdk4(bs, file, flags, options, errp);
1060 default:
1061 error_setg(errp, "Image not in VMDK format");
1062 return -EINVAL;
1063 }
1064 }
1065
1066 static const char *next_line(const char *s)
1067 {
1068 while (*s) {
1069 if (*s == '\n') {
1070 return s + 1;
1071 }
1072 s++;
1073 }
1074 return s;
1075 }
1076
1077 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
1078 QDict *options, Error **errp)
1079 {
1080 int ret;
1081 int matches;
1082 char access[11];
1083 char type[11];
1084 char fname[512];
1085 const char *p, *np;
1086 int64_t sectors = 0;
1087 int64_t flat_offset;
1088 char *desc_file_dir = NULL;
1089 char *extent_path;
1090 BdrvChild *extent_file;
1091 BdrvChildRole extent_role;
1092 BDRVVmdkState *s = bs->opaque;
1093 VmdkExtent *extent = NULL;
1094 char extent_opt_prefix[32];
1095 Error *local_err = NULL;
1096
1097 for (p = desc; *p; p = next_line(p)) {
1098 /* parse extent line in one of below formats:
1099 *
1100 * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
1101 * RW [size in sectors] SPARSE "file-name.vmdk"
1102 * RW [size in sectors] VMFS "file-name.vmdk"
1103 * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
1104 * RW [size in sectors] SESPARSE "file-name.vmdk"
1105 */
1106 flat_offset = -1;
1107 matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
1108 access, &sectors, type, fname, &flat_offset);
1109 if (matches < 4 || strcmp(access, "RW")) {
1110 continue;
1111 } else if (!strcmp(type, "FLAT")) {
1112 if (matches != 5 || flat_offset < 0) {
1113 goto invalid;
1114 }
1115 } else if (!strcmp(type, "VMFS")) {
1116 if (matches == 4) {
1117 flat_offset = 0;
1118 } else {
1119 goto invalid;
1120 }
1121 } else if (matches != 4) {
1122 goto invalid;
1123 }
1124
1125 if (sectors <= 0 ||
1126 (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
1127 strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") &&
1128 strcmp(type, "SESPARSE")) ||
1129 (strcmp(access, "RW"))) {
1130 continue;
1131 }
1132
1133 if (path_is_absolute(fname)) {
1134 extent_path = g_strdup(fname);
1135 } else {
1136 if (!desc_file_dir) {
1137 desc_file_dir = bdrv_dirname(bs->file->bs, errp);
1138 if (!desc_file_dir) {
1139 bdrv_refresh_filename(bs->file->bs);
1140 error_prepend(errp, "Cannot use relative paths with VMDK "
1141 "descriptor file '%s': ",
1142 bs->file->bs->filename);
1143 ret = -EINVAL;
1144 goto out;
1145 }
1146 }
1147
1148 extent_path = g_strconcat(desc_file_dir, fname, NULL);
1149 }
1150
1151 ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
1152 assert(ret < 32);
1153
1154 extent_role = BDRV_CHILD_DATA;
1155 if (strcmp(type, "FLAT") != 0 && strcmp(type, "VMFS") != 0) {
1156 /* non-flat extents have metadata */
1157 extent_role |= BDRV_CHILD_METADATA;
1158 }
1159
1160 extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
1161 bs, &child_of_bds, extent_role, false,
1162 &local_err);
1163 g_free(extent_path);
1164 if (local_err) {
1165 error_propagate(errp, local_err);
1166 ret = -EINVAL;
1167 goto out;
1168 }
1169
1170 /* save to extents array */
1171 if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
1172 /* FLAT extent */
1173
1174 ret = vmdk_add_extent(bs, extent_file, true, sectors,
1175 0, 0, 0, 0, 0, &extent, errp);
1176 if (ret < 0) {
1177 bdrv_unref_child(bs, extent_file);
1178 goto out;
1179 }
1180 extent->flat_start_offset = flat_offset << 9;
1181 } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
1182 /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
1183 char *buf = vmdk_read_desc(extent_file, 0, errp);
1184 if (!buf) {
1185 ret = -EINVAL;
1186 } else {
1187 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
1188 options, errp);
1189 }
1190 g_free(buf);
1191 if (ret) {
1192 bdrv_unref_child(bs, extent_file);
1193 goto out;
1194 }
1195 extent = &s->extents[s->num_extents - 1];
1196 } else if (!strcmp(type, "SESPARSE")) {
1197 ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
1198 if (ret) {
1199 bdrv_unref_child(bs, extent_file);
1200 goto out;
1201 }
1202 extent = &s->extents[s->num_extents - 1];
1203 } else {
1204 error_setg(errp, "Unsupported extent type '%s'", type);
1205 bdrv_unref_child(bs, extent_file);
1206 ret = -ENOTSUP;
1207 goto out;
1208 }
1209 extent->type = g_strdup(type);
1210 }
1211
1212 ret = 0;
1213 goto out;
1214
1215 invalid:
1216 np = next_line(p);
1217 assert(np != p);
1218 if (np[-1] == '\n') {
1219 np--;
1220 }
1221 error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
1222 ret = -EINVAL;
1223
1224 out:
1225 g_free(desc_file_dir);
1226 return ret;
1227 }
1228
1229 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
1230 QDict *options, Error **errp)
1231 {
1232 int ret;
1233 char ct[128];
1234 BDRVVmdkState *s = bs->opaque;
1235
1236 if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
1237 error_setg(errp, "invalid VMDK image descriptor");
1238 ret = -EINVAL;
1239 goto exit;
1240 }
1241 if (strcmp(ct, "monolithicFlat") &&
1242 strcmp(ct, "vmfs") &&
1243 strcmp(ct, "vmfsSparse") &&
1244 strcmp(ct, "seSparse") &&
1245 strcmp(ct, "twoGbMaxExtentSparse") &&
1246 strcmp(ct, "twoGbMaxExtentFlat")) {
1247 error_setg(errp, "Unsupported image type '%s'", ct);
1248 ret = -ENOTSUP;
1249 goto exit;
1250 }
1251 s->create_type = g_strdup(ct);
1252 s->desc_offset = 0;
1253 ret = vmdk_parse_extents(buf, bs, options, errp);
1254 exit:
1255 return ret;
1256 }
1257
1258 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
1259 Error **errp)
1260 {
1261 char *buf;
1262 int ret;
1263 BDRVVmdkState *s = bs->opaque;
1264 uint32_t magic;
1265
1266 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
1267 BDRV_CHILD_IMAGE, false, errp);
1268 if (!bs->file) {
1269 return -EINVAL;
1270 }
1271
1272 buf = vmdk_read_desc(bs->file, 0, errp);
1273 if (!buf) {
1274 return -EINVAL;
1275 }
1276
1277 magic = ldl_be_p(buf);
1278 switch (magic) {
1279 case VMDK3_MAGIC:
1280 case VMDK4_MAGIC:
1281 ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
1282 errp);
1283 s->desc_offset = 0x200;
1284 break;
1285 default:
1286 /* No data in the descriptor file */
1287 bs->file->role &= ~BDRV_CHILD_DATA;
1288
1289 /* Must succeed because we have given up permissions if anything */
1290 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1291
1292 ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1293 break;
1294 }
1295 if (ret) {
1296 goto fail;
1297 }
1298
1299 /* try to open parent images, if exist */
1300 ret = vmdk_parent_open(bs);
1301 if (ret) {
1302 goto fail;
1303 }
1304 ret = vmdk_read_cid(bs, 0, &s->cid);
1305 if (ret) {
1306 goto fail;
1307 }
1308 ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1309 if (ret) {
1310 goto fail;
1311 }
1312 qemu_co_mutex_init(&s->lock);
1313
1314 /* Disable migration when VMDK images are used */
1315 error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1316 "does not support live migration",
1317 bdrv_get_device_or_node_name(bs));
1318 ret = migrate_add_blocker(s->migration_blocker, errp);
1319 if (ret < 0) {
1320 error_free(s->migration_blocker);
1321 goto fail;
1322 }
1323
1324 g_free(buf);
1325 return 0;
1326
1327 fail:
1328 g_free(buf);
1329 g_free(s->create_type);
1330 s->create_type = NULL;
1331 vmdk_free_extents(bs);
1332 return ret;
1333 }
1334
1335
1336 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1337 {
1338 BDRVVmdkState *s = bs->opaque;
1339 int i;
1340
1341 for (i = 0; i < s->num_extents; i++) {
1342 if (!s->extents[i].flat) {
1343 bs->bl.pwrite_zeroes_alignment =
1344 MAX(bs->bl.pwrite_zeroes_alignment,
1345 s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1346 }
1347 }
1348 }
1349
1350 /**
1351 * get_whole_cluster
1352 *
1353 * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1354 * to the cluster at @cluster_sector_num. If @zeroed is true, we're overwriting
1355 * a zeroed cluster in the current layer and must not copy data from the
1356 * backing file.
1357 *
1358 * If @skip_start_sector < @skip_end_sector, the relative range
1359 * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1360 * it for call to write user data in the request.
1361 */
1362 static int get_whole_cluster(BlockDriverState *bs,
1363 VmdkExtent *extent,
1364 uint64_t cluster_offset,
1365 uint64_t offset,
1366 uint64_t skip_start_bytes,
1367 uint64_t skip_end_bytes,
1368 bool zeroed)
1369 {
1370 int ret = VMDK_OK;
1371 int64_t cluster_bytes;
1372 uint8_t *whole_grain;
1373 bool copy_from_backing;
1374
1375 /* For COW, align request sector_num to cluster start */
1376 cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1377 offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1378 whole_grain = qemu_blockalign(bs, cluster_bytes);
1379 copy_from_backing = bs->backing && !zeroed;
1380
1381 if (!copy_from_backing) {
1382 memset(whole_grain, 0, skip_start_bytes);
1383 memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1384 }
1385
1386 assert(skip_end_bytes <= cluster_bytes);
1387 /* we will be here if it's first write on non-exist grain(cluster).
1388 * try to read from parent image, if exist */
1389 if (bs->backing && !vmdk_is_cid_valid(bs)) {
1390 ret = VMDK_ERROR;
1391 goto exit;
1392 }
1393
1394 /* Read backing data before skip range */
1395 if (skip_start_bytes > 0) {
1396 if (copy_from_backing) {
1397 /* qcow2 emits this on bs->file instead of bs->backing */
1398 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1399 ret = bdrv_pread(bs->backing, offset, whole_grain,
1400 skip_start_bytes);
1401 if (ret < 0) {
1402 ret = VMDK_ERROR;
1403 goto exit;
1404 }
1405 }
1406 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1407 ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1408 skip_start_bytes);
1409 if (ret < 0) {
1410 ret = VMDK_ERROR;
1411 goto exit;
1412 }
1413 }
1414 /* Read backing data after skip range */
1415 if (skip_end_bytes < cluster_bytes) {
1416 if (copy_from_backing) {
1417 /* qcow2 emits this on bs->file instead of bs->backing */
1418 BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1419 ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1420 whole_grain + skip_end_bytes,
1421 cluster_bytes - skip_end_bytes);
1422 if (ret < 0) {
1423 ret = VMDK_ERROR;
1424 goto exit;
1425 }
1426 }
1427 BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1428 ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1429 whole_grain + skip_end_bytes,
1430 cluster_bytes - skip_end_bytes);
1431 if (ret < 0) {
1432 ret = VMDK_ERROR;
1433 goto exit;
1434 }
1435 }
1436
1437 ret = VMDK_OK;
1438 exit:
1439 qemu_vfree(whole_grain);
1440 return ret;
1441 }
1442
1443 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1444 uint32_t offset)
1445 {
1446 offset = cpu_to_le32(offset);
1447 /* update L2 table */
1448 BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1449 if (bdrv_pwrite(extent->file,
1450 ((int64_t)m_data->l2_offset * 512)
1451 + (m_data->l2_index * sizeof(offset)),
1452 &offset, sizeof(offset)) < 0) {
1453 return VMDK_ERROR;
1454 }
1455 /* update backup L2 table */
1456 if (extent->l1_backup_table_offset != 0) {
1457 m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1458 if (bdrv_pwrite(extent->file,
1459 ((int64_t)m_data->l2_offset * 512)
1460 + (m_data->l2_index * sizeof(offset)),
1461 &offset, sizeof(offset)) < 0) {
1462 return VMDK_ERROR;
1463 }
1464 }
1465 if (bdrv_flush(extent->file->bs) < 0) {
1466 return VMDK_ERROR;
1467 }
1468 if (m_data->l2_cache_entry) {
1469 *m_data->l2_cache_entry = offset;
1470 }
1471
1472 return VMDK_OK;
1473 }
1474
1475 /**
1476 * get_cluster_offset
1477 *
1478 * Look up cluster offset in extent file by sector number, and store in
1479 * @cluster_offset.
1480 *
1481 * For flat extents, the start offset as parsed from the description file is
1482 * returned.
1483 *
1484 * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1485 * offset for a new cluster and update L2 cache. If there is a backing file,
1486 * COW is done before returning; otherwise, zeroes are written to the allocated
1487 * cluster. Both COW and zero writing skips the sector range
1488 * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1489 * has new data to write there.
1490 *
1491 * Returns: VMDK_OK if cluster exists and mapped in the image.
1492 * VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1493 * VMDK_ERROR if failed.
1494 */
1495 static int get_cluster_offset(BlockDriverState *bs,
1496 VmdkExtent *extent,
1497 VmdkMetaData *m_data,
1498 uint64_t offset,
1499 bool allocate,
1500 uint64_t *cluster_offset,
1501 uint64_t skip_start_bytes,
1502 uint64_t skip_end_bytes)
1503 {
1504 unsigned int l1_index, l2_offset, l2_index;
1505 int min_index, i, j;
1506 uint32_t min_count;
1507 void *l2_table;
1508 bool zeroed = false;
1509 int64_t ret;
1510 int64_t cluster_sector;
1511 unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
1512
1513 if (m_data) {
1514 m_data->new_allocation = false;
1515 }
1516 if (extent->flat) {
1517 *cluster_offset = extent->flat_start_offset;
1518 return VMDK_OK;
1519 }
1520
1521 offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1522 l1_index = (offset >> 9) / extent->l1_entry_sectors;
1523 if (l1_index >= extent->l1_size) {
1524 return VMDK_ERROR;
1525 }
1526 if (extent->sesparse) {
1527 uint64_t l2_offset_u64;
1528
1529 assert(extent->entry_size == sizeof(uint64_t));
1530
1531 l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
1532 if (l2_offset_u64 == 0) {
1533 l2_offset = 0;
1534 } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
1535 /*
1536 * Top most nibble is 0x1 if grain table is allocated.
1537 * strict check - top most 4 bytes must be 0x10000000 since max
1538 * supported size is 64TB for disk - so no more than 64TB / 16MB
1539 * grain directories which is smaller than uint32,
1540 * where 16MB is the only supported default grain table coverage.
1541 */
1542 return VMDK_ERROR;
1543 } else {
1544 l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
1545 l2_offset_u64 = extent->sesparse_l2_tables_offset +
1546 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
1547 if (l2_offset_u64 > 0x00000000ffffffff) {
1548 return VMDK_ERROR;
1549 }
1550 l2_offset = (unsigned int)(l2_offset_u64);
1551 }
1552 } else {
1553 assert(extent->entry_size == sizeof(uint32_t));
1554 l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
1555 }
1556 if (!l2_offset) {
1557 return VMDK_UNALLOC;
1558 }
1559 for (i = 0; i < L2_CACHE_SIZE; i++) {
1560 if (l2_offset == extent->l2_cache_offsets[i]) {
1561 /* increment the hit count */
1562 if (++extent->l2_cache_counts[i] == 0xffffffff) {
1563 for (j = 0; j < L2_CACHE_SIZE; j++) {
1564 extent->l2_cache_counts[j] >>= 1;
1565 }
1566 }
1567 l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
1568 goto found;
1569 }
1570 }
1571 /* not found: load a new entry in the least used one */
1572 min_index = 0;
1573 min_count = 0xffffffff;
1574 for (i = 0; i < L2_CACHE_SIZE; i++) {
1575 if (extent->l2_cache_counts[i] < min_count) {
1576 min_count = extent->l2_cache_counts[i];
1577 min_index = i;
1578 }
1579 }
1580 l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
1581 BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1582 if (bdrv_pread(extent->file,
1583 (int64_t)l2_offset * 512,
1584 l2_table,
1585 l2_size_bytes
1586 ) != l2_size_bytes) {
1587 return VMDK_ERROR;
1588 }
1589
1590 extent->l2_cache_offsets[min_index] = l2_offset;
1591 extent->l2_cache_counts[min_index] = 1;
1592 found:
1593 l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1594 if (m_data) {
1595 m_data->l1_index = l1_index;
1596 m_data->l2_index = l2_index;
1597 m_data->l2_offset = l2_offset;
1598 m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
1599 }
1600
1601 if (extent->sesparse) {
1602 cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
1603 switch (cluster_sector & 0xf000000000000000) {
1604 case 0x0000000000000000:
1605 /* unallocated grain */
1606 if (cluster_sector != 0) {
1607 return VMDK_ERROR;
1608 }
1609 break;
1610 case 0x1000000000000000:
1611 /* scsi-unmapped grain - fallthrough */
1612 case 0x2000000000000000:
1613 /* zero grain */
1614 zeroed = true;
1615 break;
1616 case 0x3000000000000000:
1617 /* allocated grain */
1618 cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
1619 ((cluster_sector & 0x0000ffffffffffff) << 12));
1620 cluster_sector = extent->sesparse_clusters_offset +
1621 cluster_sector * extent->cluster_sectors;
1622 break;
1623 default:
1624 return VMDK_ERROR;
1625 }
1626 } else {
1627 cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
1628
1629 if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1630 zeroed = true;
1631 }
1632 }
1633
1634 if (!cluster_sector || zeroed) {
1635 if (!allocate) {
1636 return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1637 }
1638 assert(!extent->sesparse);
1639
1640 if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1641 return VMDK_ERROR;
1642 }
1643
1644 cluster_sector = extent->next_cluster_sector;
1645 extent->next_cluster_sector += extent->cluster_sectors;
1646
1647 /* First of all we write grain itself, to avoid race condition
1648 * that may to corrupt the image.
1649 * This problem may occur because of insufficient space on host disk
1650 * or inappropriate VM shutdown.
1651 */
1652 ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1653 offset, skip_start_bytes, skip_end_bytes,
1654 zeroed);
1655 if (ret) {
1656 return ret;
1657 }
1658 if (m_data) {
1659 m_data->new_allocation = true;
1660 }
1661 }
1662 *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1663 return VMDK_OK;
1664 }
1665
1666 static VmdkExtent *find_extent(BDRVVmdkState *s,
1667 int64_t sector_num, VmdkExtent *start_hint)
1668 {
1669 VmdkExtent *extent = start_hint;
1670
1671 if (!extent) {
1672 extent = &s->extents[0];
1673 }
1674 while (extent < &s->extents[s->num_extents]) {
1675 if (sector_num < extent->end_sector) {
1676 return extent;
1677 }
1678 extent++;
1679 }
1680 return NULL;
1681 }
1682
1683 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1684 int64_t offset)
1685 {
1686 uint64_t extent_begin_offset, extent_relative_offset;
1687 uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1688
1689 extent_begin_offset =
1690 (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1691 extent_relative_offset = offset - extent_begin_offset;
1692 return extent_relative_offset % cluster_size;
1693 }
1694
1695 static int coroutine_fn vmdk_co_block_status(BlockDriverState *bs,
1696 bool want_zero,
1697 int64_t offset, int64_t bytes,
1698 int64_t *pnum, int64_t *map,
1699 BlockDriverState **file)
1700 {
1701 BDRVVmdkState *s = bs->opaque;
1702 int64_t index_in_cluster, n, ret;
1703 uint64_t cluster_offset;
1704 VmdkExtent *extent;
1705
1706 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1707 if (!extent) {
1708 return -EIO;
1709 }
1710 qemu_co_mutex_lock(&s->lock);
1711 ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1712 0, 0);
1713 qemu_co_mutex_unlock(&s->lock);
1714
1715 index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1716 switch (ret) {
1717 case VMDK_ERROR:
1718 ret = -EIO;
1719 break;
1720 case VMDK_UNALLOC:
1721 ret = 0;
1722 break;
1723 case VMDK_ZEROED:
1724 ret = BDRV_BLOCK_ZERO;
1725 break;
1726 case VMDK_OK:
1727 ret = BDRV_BLOCK_DATA;
1728 if (!extent->compressed) {
1729 ret |= BDRV_BLOCK_OFFSET_VALID;
1730 *map = cluster_offset + index_in_cluster;
1731 if (extent->flat) {
1732 ret |= BDRV_BLOCK_RECURSE;
1733 }
1734 }
1735 *file = extent->file->bs;
1736 break;
1737 }
1738
1739 n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1740 *pnum = MIN(n, bytes);
1741 return ret;
1742 }
1743
1744 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1745 int64_t offset_in_cluster, QEMUIOVector *qiov,
1746 uint64_t qiov_offset, uint64_t n_bytes,
1747 uint64_t offset)
1748 {
1749 int ret;
1750 VmdkGrainMarker *data = NULL;
1751 uLongf buf_len;
1752 QEMUIOVector local_qiov;
1753 int64_t write_offset;
1754 int64_t write_end_sector;
1755
1756 if (extent->compressed) {
1757 void *compressed_data;
1758
1759 /* Only whole clusters */
1760 if (offset_in_cluster ||
1761 n_bytes > (extent->cluster_sectors * SECTOR_SIZE) ||
1762 (n_bytes < (extent->cluster_sectors * SECTOR_SIZE) &&
1763 offset + n_bytes != extent->end_sector * SECTOR_SIZE))
1764 {
1765 ret = -EINVAL;
1766 goto out;
1767 }
1768
1769 if (!extent->has_marker) {
1770 ret = -EINVAL;
1771 goto out;
1772 }
1773 buf_len = (extent->cluster_sectors << 9) * 2;
1774 data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1775
1776 compressed_data = g_malloc(n_bytes);
1777 qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1778 ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1779 g_free(compressed_data);
1780
1781 if (ret != Z_OK || buf_len == 0) {
1782 ret = -EINVAL;
1783 goto out;
1784 }
1785
1786 data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1787 data->size = cpu_to_le32(buf_len);
1788
1789 n_bytes = buf_len + sizeof(VmdkGrainMarker);
1790 qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1791
1792 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1793 } else {
1794 qemu_iovec_init(&local_qiov, qiov->niov);
1795 qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1796
1797 BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1798 }
1799
1800 write_offset = cluster_offset + offset_in_cluster;
1801 ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1802 &local_qiov, 0);
1803
1804 write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1805
1806 if (extent->compressed) {
1807 extent->next_cluster_sector = write_end_sector;
1808 } else {
1809 extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1810 write_end_sector);
1811 }
1812
1813 if (ret < 0) {
1814 goto out;
1815 }
1816 ret = 0;
1817 out:
1818 g_free(data);
1819 if (!extent->compressed) {
1820 qemu_iovec_destroy(&local_qiov);
1821 }
1822 return ret;
1823 }
1824
1825 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1826 int64_t offset_in_cluster, QEMUIOVector *qiov,
1827 int bytes)
1828 {
1829 int ret;
1830 int cluster_bytes, buf_bytes;
1831 uint8_t *cluster_buf, *compressed_data;
1832 uint8_t *uncomp_buf;
1833 uint32_t data_len;
1834 VmdkGrainMarker *marker;
1835 uLongf buf_len;
1836
1837
1838 if (!extent->compressed) {
1839 BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1840 ret = bdrv_co_preadv(extent->file,
1841 cluster_offset + offset_in_cluster, bytes,
1842 qiov, 0);
1843 if (ret < 0) {
1844 return ret;
1845 }
1846 return 0;
1847 }
1848 cluster_bytes = extent->cluster_sectors * 512;
1849 /* Read two clusters in case GrainMarker + compressed data > one cluster */
1850 buf_bytes = cluster_bytes * 2;
1851 cluster_buf = g_malloc(buf_bytes);
1852 uncomp_buf = g_malloc(cluster_bytes);
1853 BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1854 ret = bdrv_pread(extent->file,
1855 cluster_offset,
1856 cluster_buf, buf_bytes);
1857 if (ret < 0) {
1858 goto out;
1859 }
1860 compressed_data = cluster_buf;
1861 buf_len = cluster_bytes;
1862 data_len = cluster_bytes;
1863 if (extent->has_marker) {
1864 marker = (VmdkGrainMarker *)cluster_buf;
1865 compressed_data = marker->data;
1866 data_len = le32_to_cpu(marker->size);
1867 }
1868 if (!data_len || data_len > buf_bytes) {
1869 ret = -EINVAL;
1870 goto out;
1871 }
1872 ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1873 if (ret != Z_OK) {
1874 ret = -EINVAL;
1875 goto out;
1876
1877 }
1878 if (offset_in_cluster < 0 ||
1879 offset_in_cluster + bytes > buf_len) {
1880 ret = -EINVAL;
1881 goto out;
1882 }
1883 qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1884 ret = 0;
1885
1886 out:
1887 g_free(uncomp_buf);
1888 g_free(cluster_buf);
1889 return ret;
1890 }
1891
1892 static int coroutine_fn
1893 vmdk_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
1894 QEMUIOVector *qiov, BdrvRequestFlags flags)
1895 {
1896 BDRVVmdkState *s = bs->opaque;
1897 int ret;
1898 uint64_t n_bytes, offset_in_cluster;
1899 VmdkExtent *extent = NULL;
1900 QEMUIOVector local_qiov;
1901 uint64_t cluster_offset;
1902 uint64_t bytes_done = 0;
1903
1904 qemu_iovec_init(&local_qiov, qiov->niov);
1905 qemu_co_mutex_lock(&s->lock);
1906
1907 while (bytes > 0) {
1908 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1909 if (!extent) {
1910 ret = -EIO;
1911 goto fail;
1912 }
1913 ret = get_cluster_offset(bs, extent, NULL,
1914 offset, false, &cluster_offset, 0, 0);
1915 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1916
1917 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1918 - offset_in_cluster);
1919
1920 if (ret != VMDK_OK) {
1921 /* if not allocated, try to read from parent image, if exist */
1922 if (bs->backing && ret != VMDK_ZEROED) {
1923 if (!vmdk_is_cid_valid(bs)) {
1924 ret = -EINVAL;
1925 goto fail;
1926 }
1927
1928 qemu_iovec_reset(&local_qiov);
1929 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1930
1931 /* qcow2 emits this on bs->file instead of bs->backing */
1932 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1933 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1934 &local_qiov, 0);
1935 if (ret < 0) {
1936 goto fail;
1937 }
1938 } else {
1939 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1940 }
1941 } else {
1942 qemu_iovec_reset(&local_qiov);
1943 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1944
1945 ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1946 &local_qiov, n_bytes);
1947 if (ret) {
1948 goto fail;
1949 }
1950 }
1951 bytes -= n_bytes;
1952 offset += n_bytes;
1953 bytes_done += n_bytes;
1954 }
1955
1956 ret = 0;
1957 fail:
1958 qemu_co_mutex_unlock(&s->lock);
1959 qemu_iovec_destroy(&local_qiov);
1960
1961 return ret;
1962 }
1963
1964 /**
1965 * vmdk_write:
1966 * @zeroed: buf is ignored (data is zero), use zeroed_grain GTE feature
1967 * if possible, otherwise return -ENOTSUP.
1968 * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1969 * with each cluster. By dry run we can find if the zero write
1970 * is possible without modifying image data.
1971 *
1972 * Returns: error code with 0 for success.
1973 */
1974 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1975 uint64_t bytes, QEMUIOVector *qiov,
1976 bool zeroed, bool zero_dry_run)
1977 {
1978 BDRVVmdkState *s = bs->opaque;
1979 VmdkExtent *extent = NULL;
1980 int ret;
1981 int64_t offset_in_cluster, n_bytes;
1982 uint64_t cluster_offset;
1983 uint64_t bytes_done = 0;
1984 VmdkMetaData m_data;
1985
1986 if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1987 error_report("Wrong offset: offset=0x%" PRIx64
1988 " total_sectors=0x%" PRIx64,
1989 offset, bs->total_sectors);
1990 return -EIO;
1991 }
1992
1993 while (bytes > 0) {
1994 extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1995 if (!extent) {
1996 return -EIO;
1997 }
1998 if (extent->sesparse) {
1999 return -ENOTSUP;
2000 }
2001 offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
2002 n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
2003 - offset_in_cluster);
2004
2005 ret = get_cluster_offset(bs, extent, &m_data, offset,
2006 !(extent->compressed || zeroed),
2007 &cluster_offset, offset_in_cluster,
2008 offset_in_cluster + n_bytes);
2009 if (extent->compressed) {
2010 if (ret == VMDK_OK) {
2011 /* Refuse write to allocated cluster for streamOptimized */
2012 error_report("Could not write to allocated cluster"
2013 " for streamOptimized");
2014 return -EIO;
2015 } else if (!zeroed) {
2016 /* allocate */
2017 ret = get_cluster_offset(bs, extent, &m_data, offset,
2018 true, &cluster_offset, 0, 0);
2019 }
2020 }
2021 if (ret == VMDK_ERROR) {
2022 return -EINVAL;
2023 }
2024 if (zeroed) {
2025 /* Do zeroed write, buf is ignored */
2026 if (extent->has_zero_grain &&
2027 offset_in_cluster == 0 &&
2028 n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
2029 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
2030 if (!zero_dry_run && ret != VMDK_ZEROED) {
2031 /* update L2 tables */
2032 if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
2033 != VMDK_OK) {
2034 return -EIO;
2035 }
2036 }
2037 } else {
2038 return -ENOTSUP;
2039 }
2040 } else {
2041 ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
2042 qiov, bytes_done, n_bytes, offset);
2043 if (ret) {
2044 return ret;
2045 }
2046 if (m_data.new_allocation) {
2047 /* update L2 tables */
2048 if (vmdk_L2update(extent, &m_data,
2049 cluster_offset >> BDRV_SECTOR_BITS)
2050 != VMDK_OK) {
2051 return -EIO;
2052 }
2053 }
2054 }
2055 bytes -= n_bytes;
2056 offset += n_bytes;
2057 bytes_done += n_bytes;
2058
2059 /* update CID on the first write every time the virtual disk is
2060 * opened */
2061 if (!s->cid_updated) {
2062 ret = vmdk_write_cid(bs, g_random_int());
2063 if (ret < 0) {
2064 return ret;
2065 }
2066 s->cid_updated = true;
2067 }
2068 }
2069 return 0;
2070 }
2071
2072 static int coroutine_fn
2073 vmdk_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
2074 QEMUIOVector *qiov, BdrvRequestFlags flags)
2075 {
2076 int ret;
2077 BDRVVmdkState *s = bs->opaque;
2078 qemu_co_mutex_lock(&s->lock);
2079 ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
2080 qemu_co_mutex_unlock(&s->lock);
2081 return ret;
2082 }
2083
2084 static int coroutine_fn
2085 vmdk_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes,
2086 QEMUIOVector *qiov)
2087 {
2088 if (bytes == 0) {
2089 /* The caller will write bytes 0 to signal EOF.
2090 * When receive it, we align EOF to a sector boundary. */
2091 BDRVVmdkState *s = bs->opaque;
2092 int i, ret;
2093 int64_t length;
2094
2095 for (i = 0; i < s->num_extents; i++) {
2096 length = bdrv_getlength(s->extents[i].file->bs);
2097 if (length < 0) {
2098 return length;
2099 }
2100 length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
2101 ret = bdrv_truncate(s->extents[i].file, length, false,
2102 PREALLOC_MODE_OFF, 0, NULL);
2103 if (ret < 0) {
2104 return ret;
2105 }
2106 }
2107 return 0;
2108 }
2109 return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
2110 }
2111
2112 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
2113 int64_t offset,
2114 int64_t bytes,
2115 BdrvRequestFlags flags)
2116 {
2117 int ret;
2118 BDRVVmdkState *s = bs->opaque;
2119
2120 qemu_co_mutex_lock(&s->lock);
2121 /* write zeroes could fail if sectors not aligned to cluster, test it with
2122 * dry_run == true before really updating image */
2123 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
2124 if (!ret) {
2125 ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
2126 }
2127 qemu_co_mutex_unlock(&s->lock);
2128 return ret;
2129 }
2130
2131 static int vmdk_init_extent(BlockBackend *blk,
2132 int64_t filesize, bool flat,
2133 bool compress, bool zeroed_grain,
2134 Error **errp)
2135 {
2136 int ret, i;
2137 VMDK4Header header;
2138 uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
2139 uint32_t *gd_buf = NULL;
2140 int gd_buf_size;
2141
2142 if (flat) {
2143 ret = blk_truncate(blk, filesize, false, PREALLOC_MODE_OFF, 0, errp);
2144 goto exit;
2145 }
2146 magic = cpu_to_be32(VMDK4_MAGIC);
2147 memset(&header, 0, sizeof(header));
2148 if (compress) {
2149 header.version = 3;
2150 } else if (zeroed_grain) {
2151 header.version = 2;
2152 } else {
2153 header.version = 1;
2154 }
2155 header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
2156 | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
2157 | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
2158 header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
2159 header.capacity = filesize / BDRV_SECTOR_SIZE;
2160 header.granularity = 128;
2161 header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
2162
2163 grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
2164 gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
2165 BDRV_SECTOR_SIZE);
2166 gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
2167 gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
2168
2169 header.desc_offset = 1;
2170 header.desc_size = 20;
2171 header.rgd_offset = header.desc_offset + header.desc_size;
2172 header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
2173 header.grain_offset =
2174 ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
2175 header.granularity);
2176 /* swap endianness for all header fields */
2177 header.version = cpu_to_le32(header.version);
2178 header.flags = cpu_to_le32(header.flags);
2179 header.capacity = cpu_to_le64(header.capacity);
2180 header.granularity = cpu_to_le64(header.granularity);
2181 header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
2182 header.desc_offset = cpu_to_le64(header.desc_offset);
2183 header.desc_size = cpu_to_le64(header.desc_size);
2184 header.rgd_offset = cpu_to_le64(header.rgd_offset);
2185 header.gd_offset = cpu_to_le64(header.gd_offset);
2186 header.grain_offset = cpu_to_le64(header.grain_offset);
2187 header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
2188
2189 header.check_bytes[0] = 0xa;
2190 header.check_bytes[1] = 0x20;
2191 header.check_bytes[2] = 0xd;
2192 header.check_bytes[3] = 0xa;
2193
2194 /* write all the data */
2195 ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
2196 if (ret < 0) {
2197 error_setg(errp, QERR_IO_ERROR);
2198 goto exit;
2199 }
2200 ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
2201 if (ret < 0) {
2202 error_setg(errp, QERR_IO_ERROR);
2203 goto exit;
2204 }
2205
2206 ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false,
2207 PREALLOC_MODE_OFF, 0, errp);
2208 if (ret < 0) {
2209 goto exit;
2210 }
2211
2212 /* write grain directory */
2213 gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
2214 gd_buf = g_malloc0(gd_buf_size);
2215 for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
2216 i < gt_count; i++, tmp += gt_size) {
2217 gd_buf[i] = cpu_to_le32(tmp);
2218 }
2219 ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
2220 gd_buf, gd_buf_size, 0);
2221 if (ret < 0) {
2222 error_setg(errp, QERR_IO_ERROR);
2223 goto exit;
2224 }
2225
2226 /* write backup grain directory */
2227 for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
2228 i < gt_count; i++, tmp += gt_size) {
2229 gd_buf[i] = cpu_to_le32(tmp);
2230 }
2231 ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
2232 gd_buf, gd_buf_size, 0);
2233 if (ret < 0) {
2234 error_setg(errp, QERR_IO_ERROR);
2235 }
2236
2237 ret = 0;
2238 exit:
2239 g_free(gd_buf);
2240 return ret;
2241 }
2242
2243 static int vmdk_create_extent(const char *filename, int64_t filesize,
2244 bool flat, bool compress, bool zeroed_grain,
2245 BlockBackend **pbb,
2246 QemuOpts *opts, Error **errp)
2247 {
2248 int ret;
2249 BlockBackend *blk = NULL;
2250
2251 ret = bdrv_create_file(filename, opts, errp);
2252 if (ret < 0) {
2253 goto exit;
2254 }
2255
2256 blk = blk_new_open(filename, NULL, NULL,
2257 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2258 errp);
2259 if (blk == NULL) {
2260 ret = -EIO;
2261 goto exit;
2262 }
2263
2264 blk_set_allow_write_beyond_eof(blk, true);
2265
2266 ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
2267 exit:
2268 if (blk) {
2269 if (pbb) {
2270 *pbb = blk;
2271 } else {
2272 blk_unref(blk);
2273 blk = NULL;
2274 }
2275 }
2276 return ret;
2277 }
2278
2279 static int filename_decompose(const char *filename, char *path, char *prefix,
2280 char *postfix, size_t buf_len, Error **errp)
2281 {
2282 const char *p, *q;
2283
2284 if (filename == NULL || !strlen(filename)) {
2285 error_setg(errp, "No filename provided");
2286 return VMDK_ERROR;
2287 }
2288 p = strrchr(filename, '/');
2289 if (p == NULL) {
2290 p = strrchr(filename, '\\');
2291 }
2292 if (p == NULL) {
2293 p = strrchr(filename, ':');
2294 }
2295 if (p != NULL) {
2296 p++;
2297 if (p - filename >= buf_len) {
2298 return VMDK_ERROR;
2299 }
2300 pstrcpy(path, p - filename + 1, filename);
2301 } else {
2302 p = filename;
2303 path[0] = '\0';
2304 }
2305 q = strrchr(p, '.');
2306 if (q == NULL) {
2307 pstrcpy(prefix, buf_len, p);
2308 postfix[0] = '\0';
2309 } else {
2310 if (q - p >= buf_len) {
2311 return VMDK_ERROR;
2312 }
2313 pstrcpy(prefix, q - p + 1, p);
2314 pstrcpy(postfix, buf_len, q);
2315 }
2316 return VMDK_OK;
2317 }
2318
2319 /*
2320 * idx == 0: get or create the descriptor file (also the image file if in a
2321 * non-split format.
2322 * idx >= 1: get the n-th extent if in a split subformat
2323 */
2324 typedef BlockBackend *(*vmdk_create_extent_fn)(int64_t size,
2325 int idx,
2326 bool flat,
2327 bool split,
2328 bool compress,
2329 bool zeroed_grain,
2330 void *opaque,
2331 Error **errp);
2332
2333 static void vmdk_desc_add_extent(GString *desc,
2334 const char *extent_line_fmt,
2335 int64_t size, const char *filename)
2336 {
2337 char *basename = g_path_get_basename(filename);
2338
2339 g_string_append_printf(desc, extent_line_fmt,
2340 DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
2341 g_free(basename);
2342 }
2343
2344 static int coroutine_fn vmdk_co_do_create(int64_t size,
2345 BlockdevVmdkSubformat subformat,
2346 BlockdevVmdkAdapterType adapter_type,
2347 const char *backing_file,
2348 const char *hw_version,
2349 const char *toolsversion,
2350 bool compat6,
2351 bool zeroed_grain,
2352 vmdk_create_extent_fn extent_fn,
2353 void *opaque,
2354 Error **errp)
2355 {
2356 int extent_idx;
2357 BlockBackend *blk = NULL;
2358 BlockBackend *extent_blk;
2359 Error *local_err = NULL;
2360 char *desc = NULL;
2361 int ret = 0;
2362 bool flat, split, compress;
2363 GString *ext_desc_lines;
2364 const int64_t split_size = 0x80000000; /* VMDK has constant split size */
2365 int64_t extent_size;
2366 int64_t created_size = 0;
2367 const char *extent_line_fmt;
2368 char *parent_desc_line = g_malloc0(BUF_SIZE);
2369 uint32_t parent_cid = 0xffffffff;
2370 uint32_t number_heads = 16;
2371 uint32_t desc_offset = 0, desc_len;
2372 const char desc_template[] =
2373 "# Disk DescriptorFile\n"
2374 "version=1\n"
2375 "CID=%" PRIx32 "\n"
2376 "parentCID=%" PRIx32 "\n"
2377 "createType=\"%s\"\n"
2378 "%s"
2379 "\n"
2380 "# Extent description\n"
2381 "%s"
2382 "\n"
2383 "# The Disk Data Base\n"
2384 "#DDB\n"
2385 "\n"
2386 "ddb.virtualHWVersion = \"%s\"\n"
2387 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2388 "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2389 "ddb.geometry.sectors = \"63\"\n"
2390 "ddb.adapterType = \"%s\"\n"
2391 "ddb.toolsVersion = \"%s\"\n";
2392
2393 ext_desc_lines = g_string_new(NULL);
2394
2395 /* Read out options */
2396 if (compat6) {
2397 if (hw_version) {
2398 error_setg(errp,
2399 "compat6 cannot be enabled with hwversion set");
2400 ret = -EINVAL;
2401 goto exit;
2402 }
2403 hw_version = "6";
2404 }
2405 if (!hw_version) {
2406 hw_version = "4";
2407 }
2408 if (!toolsversion) {
2409 toolsversion = "2147483647";
2410 }
2411
2412 if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2413 /* that's the number of heads with which vmware operates when
2414 creating, exporting, etc. vmdk files with a non-ide adapter type */
2415 number_heads = 255;
2416 }
2417 split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2418 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2419 flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2420 (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2421 compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2422
2423 if (flat) {
2424 extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2425 } else {
2426 extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2427 }
2428 if (flat && backing_file) {
2429 error_setg(errp, "Flat image can't have backing file");
2430 ret = -ENOTSUP;
2431 goto exit;
2432 }
2433 if (flat && zeroed_grain) {
2434 error_setg(errp, "Flat image can't enable zeroed grain");
2435 ret = -ENOTSUP;
2436 goto exit;
2437 }
2438
2439 /* Create extents */
2440 if (split) {
2441 extent_size = split_size;
2442 } else {
2443 extent_size = size;
2444 }
2445 if (!split && !flat) {
2446 created_size = extent_size;
2447 } else {
2448 created_size = 0;
2449 }
2450 /* Get the descriptor file BDS */
2451 blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2452 opaque, errp);
2453 if (!blk) {
2454 ret = -EIO;
2455 goto exit;
2456 }
2457 if (!split && !flat) {
2458 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2459 blk_bs(blk)->filename);
2460 }
2461
2462 if (backing_file) {
2463 BlockBackend *backing;
2464 char *full_backing =
2465 bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2466 backing_file,
2467 &local_err);
2468 if (local_err) {
2469 error_propagate(errp, local_err);
2470 ret = -ENOENT;
2471 goto exit;
2472 }
2473 assert(full_backing);
2474
2475 backing = blk_new_open(full_backing, NULL, NULL,
2476 BDRV_O_NO_BACKING, errp);
2477 g_free(full_backing);
2478 if (backing == NULL) {
2479 ret = -EIO;
2480 goto exit;
2481 }
2482 if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2483 error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2484 blk_bs(backing)->drv->format_name);
2485 blk_unref(backing);
2486 ret = -EINVAL;
2487 goto exit;
2488 }
2489 ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2490 blk_unref(backing);
2491 if (ret) {
2492 error_setg(errp, "Failed to read parent CID");
2493 goto exit;
2494 }
2495 snprintf(parent_desc_line, BUF_SIZE,
2496 "parentFileNameHint=\"%s\"", backing_file);
2497 }
2498 extent_idx = 1;
2499 while (created_size < size) {
2500 int64_t cur_size = MIN(size - created_size, extent_size);
2501 extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2502 zeroed_grain, opaque, errp);
2503 if (!extent_blk) {
2504 ret = -EINVAL;
2505 goto exit;
2506 }
2507 vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2508 blk_bs(extent_blk)->filename);
2509 created_size += cur_size;
2510 extent_idx++;
2511 blk_unref(extent_blk);
2512 }
2513
2514 /* Check whether we got excess extents */
2515 extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2516 opaque, NULL);
2517 if (extent_blk) {
2518 blk_unref(extent_blk);
2519 error_setg(errp, "List of extents contains unused extents");
2520 ret = -EINVAL;
2521 goto exit;
2522 }
2523
2524 /* generate descriptor file */
2525 desc = g_strdup_printf(desc_template,
2526 g_random_int(),
2527 parent_cid,
2528 BlockdevVmdkSubformat_str(subformat),
2529 parent_desc_line,
2530 ext_desc_lines->str,
2531 hw_version,
2532 size /
2533 (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2534 number_heads,
2535 BlockdevVmdkAdapterType_str(adapter_type),
2536 toolsversion);
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 *toolsversion = NULL;
2629 char *fmt = NULL;
2630 BlockdevVmdkSubformat subformat;
2631 int ret = 0;
2632 char *path = g_malloc0(PATH_MAX);
2633 char *prefix = g_malloc0(PATH_MAX);
2634 char *postfix = g_malloc0(PATH_MAX);
2635 char *desc_line = g_malloc0(BUF_SIZE);
2636 char *ext_filename = g_malloc0(PATH_MAX);
2637 char *desc_filename = g_malloc0(PATH_MAX);
2638 char *parent_desc_line = g_malloc0(BUF_SIZE);
2639 bool zeroed_grain;
2640 bool compat6;
2641 VMDKCreateOptsData data;
2642 char *backing_fmt = NULL;
2643
2644 backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2645 if (backing_fmt && strcmp(backing_fmt, "vmdk") != 0) {
2646 error_setg(errp, "backing_file must be a vmdk image");
2647 ret = -EINVAL;
2648 goto exit;
2649 }
2650
2651 if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
2652 ret = -EINVAL;
2653 goto exit;
2654 }
2655 /* Read out options */
2656 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2657 BDRV_SECTOR_SIZE);
2658 adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
2659 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2660 hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
2661 toolsversion = qemu_opt_get_del(opts, BLOCK_OPT_TOOLSVERSION);
2662 compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false);
2663 if (strcmp(hw_version, "undefined") == 0) {
2664 g_free(hw_version);
2665 hw_version = NULL;
2666 }
2667 fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
2668 zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false);
2669
2670 if (adapter_type) {
2671 adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup,
2672 adapter_type,
2673 BLOCKDEV_VMDK_ADAPTER_TYPE_IDE,
2674 &local_err);
2675 if (local_err) {
2676 error_propagate(errp, local_err);
2677 ret = -EINVAL;
2678 goto exit;
2679 }
2680 } else {
2681 adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE;
2682 }
2683
2684 if (!fmt) {
2685 /* Default format to monolithicSparse */
2686 subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE;
2687 } else {
2688 subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup,
2689 fmt,
2690 BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE,
2691 &local_err);
2692 if (local_err) {
2693 error_propagate(errp, local_err);
2694 ret = -EINVAL;
2695 goto exit;
2696 }
2697 }
2698 data = (VMDKCreateOptsData){
2699 .prefix = prefix,
2700 .postfix = postfix,
2701 .path = path,
2702 .opts = opts,
2703 };
2704 ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum,
2705 backing_file, hw_version, toolsversion, compat6,
2706 zeroed_grain, vmdk_co_create_opts_cb, &data, errp);
2707
2708 exit:
2709 g_free(backing_fmt);
2710 g_free(adapter_type);
2711 g_free(backing_file);
2712 g_free(hw_version);
2713 g_free(toolsversion);
2714 g_free(fmt);
2715 g_free(desc);
2716 g_free(path);
2717 g_free(prefix);
2718 g_free(postfix);
2719 g_free(desc_line);
2720 g_free(ext_filename);
2721 g_free(desc_filename);
2722 g_free(parent_desc_line);
2723 return ret;
2724 }
2725
2726 static BlockBackend *vmdk_co_create_cb(int64_t size, int idx,
2727 bool flat, bool split, bool compress,
2728 bool zeroed_grain, void *opaque,
2729 Error **errp)
2730 {
2731 int ret;
2732 BlockDriverState *bs;
2733 BlockBackend *blk;
2734 BlockdevCreateOptionsVmdk *opts = opaque;
2735
2736 if (idx == 0) {
2737 bs = bdrv_open_blockdev_ref(opts->file, errp);
2738 } else {
2739 int i;
2740 BlockdevRefList *list = opts->extents;
2741 for (i = 1; i < idx; i++) {
2742 if (!list || !list->next) {
2743 error_setg(errp, "Extent [%d] not specified", i);
2744 return NULL;
2745 }
2746 list = list->next;
2747 }
2748 if (!list) {
2749 error_setg(errp, "Extent [%d] not specified", idx - 1);
2750 return NULL;
2751 }
2752 bs = bdrv_open_blockdev_ref(list->value, errp);
2753 }
2754 if (!bs) {
2755 return NULL;
2756 }
2757 blk = blk_new_with_bs(bs,
2758 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | BLK_PERM_RESIZE,
2759 BLK_PERM_ALL, errp);
2760 if (!blk) {
2761 return NULL;
2762 }
2763 blk_set_allow_write_beyond_eof(blk, true);
2764 bdrv_unref(bs);
2765
2766 if (size != -1) {
2767 ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2768 if (ret) {
2769 blk_unref(blk);
2770 blk = NULL;
2771 }
2772 }
2773 return blk;
2774 }
2775
2776 static int coroutine_fn vmdk_co_create(BlockdevCreateOptions *create_options,
2777 Error **errp)
2778 {
2779 int ret;
2780 BlockdevCreateOptionsVmdk *opts;
2781
2782 opts = &create_options->u.vmdk;
2783
2784 /* Validate options */
2785 if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2786 error_setg(errp, "Image size must be a multiple of 512 bytes");
2787 ret = -EINVAL;
2788 goto out;
2789 }
2790
2791 ret = vmdk_co_do_create(opts->size,
2792 opts->subformat,
2793 opts->adapter_type,
2794 opts->backing_file,
2795 opts->hwversion,
2796 opts->toolsversion,
2797 false,
2798 opts->zeroed_grain,
2799 vmdk_co_create_cb,
2800 opts, errp);
2801 return ret;
2802
2803 out:
2804 return ret;
2805 }
2806
2807 static void vmdk_close(BlockDriverState *bs)
2808 {
2809 BDRVVmdkState *s = bs->opaque;
2810
2811 vmdk_free_extents(bs);
2812 g_free(s->create_type);
2813
2814 migrate_del_blocker(s->migration_blocker);
2815 error_free(s->migration_blocker);
2816 }
2817
2818 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2819 {
2820 int i;
2821 int64_t ret = 0;
2822 int64_t r;
2823 BDRVVmdkState *s = bs->opaque;
2824
2825 ret = bdrv_get_allocated_file_size(bs->file->bs);
2826 if (ret < 0) {
2827 return ret;
2828 }
2829 for (i = 0; i < s->num_extents; i++) {
2830 if (s->extents[i].file == bs->file) {
2831 continue;
2832 }
2833 r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2834 if (r < 0) {
2835 return r;
2836 }
2837 ret += r;
2838 }
2839 return ret;
2840 }
2841
2842 static int vmdk_has_zero_init(BlockDriverState *bs)
2843 {
2844 int i;
2845 BDRVVmdkState *s = bs->opaque;
2846
2847 /* If has a flat extent and its underlying storage doesn't have zero init,
2848 * return 0. */
2849 for (i = 0; i < s->num_extents; i++) {
2850 if (s->extents[i].flat) {
2851 if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2852 return 0;
2853 }
2854 }
2855 }
2856 return 1;
2857 }
2858
2859 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2860 {
2861 ImageInfo *info = g_new0(ImageInfo, 1);
2862
2863 bdrv_refresh_filename(extent->file->bs);
2864 *info = (ImageInfo){
2865 .filename = g_strdup(extent->file->bs->filename),
2866 .format = g_strdup(extent->type),
2867 .virtual_size = extent->sectors * BDRV_SECTOR_SIZE,
2868 .compressed = extent->compressed,
2869 .has_compressed = extent->compressed,
2870 .cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2871 .has_cluster_size = !extent->flat,
2872 };
2873
2874 return info;
2875 }
2876
2877 static int coroutine_fn vmdk_co_check(BlockDriverState *bs,
2878 BdrvCheckResult *result,
2879 BdrvCheckMode fix)
2880 {
2881 BDRVVmdkState *s = bs->opaque;
2882 VmdkExtent *extent = NULL;
2883 int64_t sector_num = 0;
2884 int64_t total_sectors = bdrv_nb_sectors(bs);
2885 int ret;
2886 uint64_t cluster_offset;
2887
2888 if (fix) {
2889 return -ENOTSUP;
2890 }
2891
2892 for (;;) {
2893 if (sector_num >= total_sectors) {
2894 return 0;
2895 }
2896 extent = find_extent(s, sector_num, extent);
2897 if (!extent) {
2898 fprintf(stderr,
2899 "ERROR: could not find extent for sector %" PRId64 "\n",
2900 sector_num);
2901 ret = -EINVAL;
2902 break;
2903 }
2904 ret = get_cluster_offset(bs, extent, NULL,
2905 sector_num << BDRV_SECTOR_BITS,
2906 false, &cluster_offset, 0, 0);
2907 if (ret == VMDK_ERROR) {
2908 fprintf(stderr,
2909 "ERROR: could not get cluster_offset for sector %"
2910 PRId64 "\n", sector_num);
2911 break;
2912 }
2913 if (ret == VMDK_OK) {
2914 int64_t extent_len = bdrv_getlength(extent->file->bs);
2915 if (extent_len < 0) {
2916 fprintf(stderr,
2917 "ERROR: could not get extent file length for sector %"
2918 PRId64 "\n", sector_num);
2919 ret = extent_len;
2920 break;
2921 }
2922 if (cluster_offset >= extent_len) {
2923 fprintf(stderr,
2924 "ERROR: cluster offset for sector %"
2925 PRId64 " points after EOF\n", sector_num);
2926 ret = -EINVAL;
2927 break;
2928 }
2929 }
2930 sector_num += extent->cluster_sectors;
2931 }
2932
2933 result->corruptions++;
2934 return ret;
2935 }
2936
2937 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs,
2938 Error **errp)
2939 {
2940 int i;
2941 BDRVVmdkState *s = bs->opaque;
2942 ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2943 ImageInfoList **tail;
2944
2945 *spec_info = (ImageInfoSpecific){
2946 .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2947 .u = {
2948 .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2949 },
2950 };
2951
2952 *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2953 .create_type = g_strdup(s->create_type),
2954 .cid = s->cid,
2955 .parent_cid = s->parent_cid,
2956 };
2957
2958 tail = &spec_info->u.vmdk.data->extents;
2959 for (i = 0; i < s->num_extents; i++) {
2960 QAPI_LIST_APPEND(tail, vmdk_get_extent_info(&s->extents[i]));
2961 }
2962
2963 return spec_info;
2964 }
2965
2966 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2967 {
2968 return a->flat == b->flat &&
2969 a->compressed == b->compressed &&
2970 (a->flat || a->cluster_sectors == b->cluster_sectors);
2971 }
2972
2973 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2974 {
2975 int i;
2976 BDRVVmdkState *s = bs->opaque;
2977 assert(s->num_extents);
2978
2979 /* See if we have multiple extents but they have different cases */
2980 for (i = 1; i < s->num_extents; i++) {
2981 if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2982 return -ENOTSUP;
2983 }
2984 }
2985 bdi->needs_compressed_writes = s->extents[0].compressed;
2986 if (!s->extents[0].flat) {
2987 bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2988 }
2989 return 0;
2990 }
2991
2992 static void vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
2993 bool backing_overridden)
2994 {
2995 /* No children but file and backing can be explicitly specified (TODO) */
2996 qdict_put(target, "file",
2997 qobject_ref(bs->file->bs->full_open_options));
2998
2999 if (backing_overridden) {
3000 if (bs->backing) {
3001 qdict_put(target, "backing",
3002 qobject_ref(bs->backing->bs->full_open_options));
3003 } else {
3004 qdict_put_null(target, "backing");
3005 }
3006 }
3007 }
3008
3009 static QemuOptsList vmdk_create_opts = {
3010 .name = "vmdk-create-opts",
3011 .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
3012 .desc = {
3013 {
3014 .name = BLOCK_OPT_SIZE,
3015 .type = QEMU_OPT_SIZE,
3016 .help = "Virtual disk size"
3017 },
3018 {
3019 .name = BLOCK_OPT_ADAPTER_TYPE,
3020 .type = QEMU_OPT_STRING,
3021 .help = "Virtual adapter type, can be one of "
3022 "ide (default), lsilogic, buslogic or legacyESX"
3023 },
3024 {
3025 .name = BLOCK_OPT_BACKING_FILE,
3026 .type = QEMU_OPT_STRING,
3027 .help = "File name of a base image"
3028 },
3029 {
3030 .name = BLOCK_OPT_BACKING_FMT,
3031 .type = QEMU_OPT_STRING,
3032 .help = "Must be 'vmdk' if present",
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_TOOLSVERSION,
3048 .type = QEMU_OPT_STRING,
3049 .help = "VMware guest tools version",
3050 },
3051 {
3052 .name = BLOCK_OPT_SUBFMT,
3053 .type = QEMU_OPT_STRING,
3054 .help =
3055 "VMDK flat extent format, can be one of "
3056 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
3057 },
3058 {
3059 .name = BLOCK_OPT_ZEROED_GRAIN,
3060 .type = QEMU_OPT_BOOL,
3061 .help = "Enable efficient zero writes "
3062 "using the zeroed-grain GTE feature"
3063 },
3064 { /* end of list */ }
3065 }
3066 };
3067
3068 static BlockDriver bdrv_vmdk = {
3069 .format_name = "vmdk",
3070 .instance_size = sizeof(BDRVVmdkState),
3071 .bdrv_probe = vmdk_probe,
3072 .bdrv_open = vmdk_open,
3073 .bdrv_co_check = vmdk_co_check,
3074 .bdrv_reopen_prepare = vmdk_reopen_prepare,
3075 .bdrv_child_perm = bdrv_default_perms,
3076 .bdrv_co_preadv = vmdk_co_preadv,
3077 .bdrv_co_pwritev = vmdk_co_pwritev,
3078 .bdrv_co_pwritev_compressed = vmdk_co_pwritev_compressed,
3079 .bdrv_co_pwrite_zeroes = vmdk_co_pwrite_zeroes,
3080 .bdrv_close = vmdk_close,
3081 .bdrv_co_create_opts = vmdk_co_create_opts,
3082 .bdrv_co_create = vmdk_co_create,
3083 .bdrv_co_block_status = vmdk_co_block_status,
3084 .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
3085 .bdrv_has_zero_init = vmdk_has_zero_init,
3086 .bdrv_get_specific_info = vmdk_get_specific_info,
3087 .bdrv_refresh_limits = vmdk_refresh_limits,
3088 .bdrv_get_info = vmdk_get_info,
3089 .bdrv_gather_child_options = vmdk_gather_child_options,
3090
3091 .is_format = true,
3092 .supports_backing = true,
3093 .create_opts = &vmdk_create_opts,
3094 };
3095
3096 static void bdrv_vmdk_init(void)
3097 {
3098 bdrv_register(&bdrv_vmdk);
3099 }
3100
3101 block_init(bdrv_vmdk_init);
Ray Wang's QEMU Repository