xref: /openbmc/qemu/block/vmdk.c (revision 86269823)
1 /*
2  * Block driver for the VMDK format
3  *
4  * Copyright (c) 2004 Fabrice Bellard
5  * Copyright (c) 2005 Filip Navara
6  *
7  * Permission is hereby granted, free of charge, to any person obtaining a copy
8  * of this software and associated documentation files (the "Software"), to deal
9  * in the Software without restriction, including without limitation the rights
10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11  * copies of the Software, and to permit persons to whom the Software is
12  * furnished to do so, subject to the following conditions:
13  *
14  * The above copyright notice and this permission notice shall be included in
15  * all copies or substantial portions of the Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23  * THE SOFTWARE.
24  */
25 
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "block/block_int.h"
29 #include "sysemu/block-backend.h"
30 #include "qapi/qmp/qdict.h"
31 #include "qapi/qmp/qerror.h"
32 #include "qemu/error-report.h"
33 #include "qemu/module.h"
34 #include "qemu/option.h"
35 #include "qemu/bswap.h"
36 #include "qemu/memalign.h"
37 #include "migration/blocker.h"
38 #include "qemu/cutils.h"
39 #include <zlib.h>
40 
41 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
42 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
43 #define VMDK4_COMPRESSION_DEFLATE 1
44 #define VMDK4_FLAG_NL_DETECT (1 << 0)
45 #define VMDK4_FLAG_RGD (1 << 1)
46 /* Zeroed-grain enable bit */
47 #define VMDK4_FLAG_ZERO_GRAIN   (1 << 2)
48 #define VMDK4_FLAG_COMPRESS (1 << 16)
49 #define VMDK4_FLAG_MARKER (1 << 17)
50 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
51 
52 #define VMDK_EXTENT_MAX_SECTORS (1ULL << 32)
53 
54 #define VMDK_GTE_ZEROED 0x1
55 
56 /* VMDK internal error codes */
57 #define VMDK_OK      0
58 #define VMDK_ERROR   (-1)
59 /* Cluster not allocated */
60 #define VMDK_UNALLOC (-2)
61 #define VMDK_ZEROED  (-3)
62 
63 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
64 #define BLOCK_OPT_TOOLSVERSION "toolsversion"
65 
66 typedef struct {
67     uint32_t version;
68     uint32_t flags;
69     uint32_t disk_sectors;
70     uint32_t granularity;
71     uint32_t l1dir_offset;
72     uint32_t l1dir_size;
73     uint32_t file_sectors;
74     uint32_t cylinders;
75     uint32_t heads;
76     uint32_t sectors_per_track;
77 } QEMU_PACKED VMDK3Header;
78 
79 typedef struct {
80     uint32_t version;
81     uint32_t flags;
82     uint64_t capacity;
83     uint64_t granularity;
84     uint64_t desc_offset;
85     uint64_t desc_size;
86     /* Number of GrainTableEntries per GrainTable */
87     uint32_t num_gtes_per_gt;
88     uint64_t rgd_offset;
89     uint64_t gd_offset;
90     uint64_t grain_offset;
91     char filler[1];
92     char check_bytes[4];
93     uint16_t compressAlgorithm;
94 } QEMU_PACKED VMDK4Header;
95 
96 typedef struct VMDKSESparseConstHeader {
97     uint64_t magic;
98     uint64_t version;
99     uint64_t capacity;
100     uint64_t grain_size;
101     uint64_t grain_table_size;
102     uint64_t flags;
103     uint64_t reserved1;
104     uint64_t reserved2;
105     uint64_t reserved3;
106     uint64_t reserved4;
107     uint64_t volatile_header_offset;
108     uint64_t volatile_header_size;
109     uint64_t journal_header_offset;
110     uint64_t journal_header_size;
111     uint64_t journal_offset;
112     uint64_t journal_size;
113     uint64_t grain_dir_offset;
114     uint64_t grain_dir_size;
115     uint64_t grain_tables_offset;
116     uint64_t grain_tables_size;
117     uint64_t free_bitmap_offset;
118     uint64_t free_bitmap_size;
119     uint64_t backmap_offset;
120     uint64_t backmap_size;
121     uint64_t grains_offset;
122     uint64_t grains_size;
123     uint8_t pad[304];
124 } QEMU_PACKED VMDKSESparseConstHeader;
125 
126 typedef struct VMDKSESparseVolatileHeader {
127     uint64_t magic;
128     uint64_t free_gt_number;
129     uint64_t next_txn_seq_number;
130     uint64_t replay_journal;
131     uint8_t pad[480];
132 } QEMU_PACKED VMDKSESparseVolatileHeader;
133 
134 #define L2_CACHE_SIZE 16
135 
136 typedef struct VmdkExtent {
137     BdrvChild *file;
138     bool flat;
139     bool compressed;
140     bool has_marker;
141     bool has_zero_grain;
142     bool sesparse;
143     uint64_t sesparse_l2_tables_offset;
144     uint64_t sesparse_clusters_offset;
145     int32_t entry_size;
146     int version;
147     int64_t sectors;
148     int64_t end_sector;
149     int64_t flat_start_offset;
150     int64_t l1_table_offset;
151     int64_t l1_backup_table_offset;
152     void *l1_table;
153     uint32_t *l1_backup_table;
154     unsigned int l1_size;
155     uint32_t l1_entry_sectors;
156 
157     unsigned int l2_size;
158     void *l2_cache;
159     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
160     uint32_t l2_cache_counts[L2_CACHE_SIZE];
161 
162     int64_t cluster_sectors;
163     int64_t next_cluster_sector;
164     char *type;
165 } VmdkExtent;
166 
167 typedef struct BDRVVmdkState {
168     CoMutex lock;
169     uint64_t desc_offset;
170     bool cid_updated;
171     bool cid_checked;
172     uint32_t cid;
173     uint32_t parent_cid;
174     int num_extents;
175     /* Extent array with num_extents entries, ascend ordered by address */
176     VmdkExtent *extents;
177     Error *migration_blocker;
178     char *create_type;
179 } BDRVVmdkState;
180 
181 typedef struct BDRVVmdkReopenState {
182     bool *extents_using_bs_file;
183 } BDRVVmdkReopenState;
184 
185 typedef struct VmdkMetaData {
186     unsigned int l1_index;
187     unsigned int l2_index;
188     unsigned int l2_offset;
189     bool new_allocation;
190     uint32_t *l2_cache_entry;
191 } VmdkMetaData;
192 
193 typedef struct VmdkGrainMarker {
194     uint64_t lba;
195     uint32_t size;
196     uint8_t  data[];
197 } QEMU_PACKED VmdkGrainMarker;
198 
199 enum {
200     MARKER_END_OF_STREAM    = 0,
201     MARKER_GRAIN_TABLE      = 1,
202     MARKER_GRAIN_DIRECTORY  = 2,
203     MARKER_FOOTER           = 3,
204 };
205 
206 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
207 {
208     uint32_t magic;
209 
210     if (buf_size < 4) {
211         return 0;
212     }
213     magic = be32_to_cpu(*(uint32_t *)buf);
214     if (magic == VMDK3_MAGIC ||
215         magic == VMDK4_MAGIC) {
216         return 100;
217     } else {
218         const char *p = (const char *)buf;
219         const char *end = p + buf_size;
220         while (p < end) {
221             if (*p == '#') {
222                 /* skip comment line */
223                 while (p < end && *p != '\n') {
224                     p++;
225                 }
226                 p++;
227                 continue;
228             }
229             if (*p == ' ') {
230                 while (p < end && *p == ' ') {
231                     p++;
232                 }
233                 /* skip '\r' if windows line endings used. */
234                 if (p < end && *p == '\r') {
235                     p++;
236                 }
237                 /* only accept blank lines before 'version=' line */
238                 if (p == end || *p != '\n') {
239                     return 0;
240                 }
241                 p++;
242                 continue;
243             }
244             if (end - p >= strlen("version=X\n")) {
245                 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
246                     strncmp("version=2\n", p, strlen("version=2\n")) == 0 ||
247                     strncmp("version=3\n", p, strlen("version=3\n")) == 0) {
248                     return 100;
249                 }
250             }
251             if (end - p >= strlen("version=X\r\n")) {
252                 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
253                     strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0 ||
254                     strncmp("version=3\r\n", p, strlen("version=3\r\n")) == 0) {
255                     return 100;
256                 }
257             }
258             return 0;
259         }
260         return 0;
261     }
262 }
263 
264 #define SECTOR_SIZE 512
265 #define DESC_SIZE (20 * SECTOR_SIZE)    /* 20 sectors of 512 bytes each */
266 #define BUF_SIZE 4096
267 #define HEADER_SIZE 512                 /* first sector of 512 bytes */
268 
269 static void vmdk_free_extents(BlockDriverState *bs)
270 {
271     int i;
272     BDRVVmdkState *s = bs->opaque;
273     VmdkExtent *e;
274 
275     for (i = 0; i < s->num_extents; i++) {
276         e = &s->extents[i];
277         g_free(e->l1_table);
278         g_free(e->l2_cache);
279         g_free(e->l1_backup_table);
280         g_free(e->type);
281         if (e->file != bs->file) {
282             bdrv_unref_child(bs, e->file);
283         }
284     }
285     g_free(s->extents);
286 }
287 
288 static void vmdk_free_last_extent(BlockDriverState *bs)
289 {
290     BDRVVmdkState *s = bs->opaque;
291 
292     if (s->num_extents == 0) {
293         return;
294     }
295     s->num_extents--;
296     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
297 }
298 
299 /* Return -ve errno, or 0 on success and write CID into *pcid. */
300 static int vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
301 {
302     char *desc;
303     uint32_t cid;
304     const char *p_name, *cid_str;
305     size_t cid_str_size;
306     BDRVVmdkState *s = bs->opaque;
307     int ret;
308 
309     desc = g_malloc0(DESC_SIZE);
310     ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
311     if (ret < 0) {
312         goto out;
313     }
314 
315     if (parent) {
316         cid_str = "parentCID";
317         cid_str_size = sizeof("parentCID");
318     } else {
319         cid_str = "CID";
320         cid_str_size = sizeof("CID");
321     }
322 
323     desc[DESC_SIZE - 1] = '\0';
324     p_name = strstr(desc, cid_str);
325     if (p_name == NULL) {
326         ret = -EINVAL;
327         goto out;
328     }
329     p_name += cid_str_size;
330     if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
331         ret = -EINVAL;
332         goto out;
333     }
334     *pcid = cid;
335     ret = 0;
336 
337 out:
338     g_free(desc);
339     return ret;
340 }
341 
342 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
343 {
344     char *desc, *tmp_desc;
345     char *p_name, *tmp_str;
346     BDRVVmdkState *s = bs->opaque;
347     int ret = 0;
348 
349     desc = g_malloc0(DESC_SIZE);
350     tmp_desc = g_malloc0(DESC_SIZE);
351     ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
352     if (ret < 0) {
353         goto out;
354     }
355 
356     desc[DESC_SIZE - 1] = '\0';
357     tmp_str = strstr(desc, "parentCID");
358     if (tmp_str == NULL) {
359         ret = -EINVAL;
360         goto out;
361     }
362 
363     pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
364     p_name = strstr(desc, "CID");
365     if (p_name != NULL) {
366         p_name += sizeof("CID");
367         snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
368         pstrcat(desc, DESC_SIZE, tmp_desc);
369     }
370 
371     ret = bdrv_pwrite_sync(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
372 
373 out:
374     g_free(desc);
375     g_free(tmp_desc);
376     return ret;
377 }
378 
379 static int vmdk_is_cid_valid(BlockDriverState *bs)
380 {
381     BDRVVmdkState *s = bs->opaque;
382     uint32_t cur_pcid;
383 
384     if (!s->cid_checked && bs->backing) {
385         BlockDriverState *p_bs = bs->backing->bs;
386 
387         if (strcmp(p_bs->drv->format_name, "vmdk")) {
388             /* Backing file is not in vmdk format, so it does not have
389              * a CID, which makes the overlay's parent CID invalid */
390             return 0;
391         }
392 
393         if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
394             /* read failure: report as not valid */
395             return 0;
396         }
397         if (s->parent_cid != cur_pcid) {
398             /* CID not valid */
399             return 0;
400         }
401     }
402     s->cid_checked = true;
403     /* CID valid */
404     return 1;
405 }
406 
407 static int vmdk_reopen_prepare(BDRVReopenState *state,
408                                BlockReopenQueue *queue, Error **errp)
409 {
410     BDRVVmdkState *s;
411     BDRVVmdkReopenState *rs;
412     int i;
413 
414     assert(state != NULL);
415     assert(state->bs != NULL);
416     assert(state->opaque == NULL);
417 
418     s = state->bs->opaque;
419 
420     rs = g_new0(BDRVVmdkReopenState, 1);
421     state->opaque = rs;
422 
423     /*
424      * Check whether there are any extents stored in bs->file; if bs->file
425      * changes, we will need to update their .file pointers to follow suit
426      */
427     rs->extents_using_bs_file = g_new(bool, s->num_extents);
428     for (i = 0; i < s->num_extents; i++) {
429         rs->extents_using_bs_file[i] = s->extents[i].file == state->bs->file;
430     }
431 
432     return 0;
433 }
434 
435 static void vmdk_reopen_clean(BDRVReopenState *state)
436 {
437     BDRVVmdkReopenState *rs = state->opaque;
438 
439     g_free(rs->extents_using_bs_file);
440     g_free(rs);
441     state->opaque = NULL;
442 }
443 
444 static void vmdk_reopen_commit(BDRVReopenState *state)
445 {
446     BDRVVmdkState *s = state->bs->opaque;
447     BDRVVmdkReopenState *rs = state->opaque;
448     int i;
449 
450     for (i = 0; i < s->num_extents; i++) {
451         if (rs->extents_using_bs_file[i]) {
452             s->extents[i].file = state->bs->file;
453         }
454     }
455 
456     vmdk_reopen_clean(state);
457 }
458 
459 static void vmdk_reopen_abort(BDRVReopenState *state)
460 {
461     vmdk_reopen_clean(state);
462 }
463 
464 static int vmdk_parent_open(BlockDriverState *bs)
465 {
466     char *p_name;
467     char *desc;
468     BDRVVmdkState *s = bs->opaque;
469     int ret;
470 
471     desc = g_malloc0(DESC_SIZE + 1);
472     ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
473     if (ret < 0) {
474         goto out;
475     }
476 
477     p_name = strstr(desc, "parentFileNameHint");
478     if (p_name != NULL) {
479         char *end_name;
480 
481         p_name += sizeof("parentFileNameHint") + 1;
482         end_name = strchr(p_name, '\"');
483         if (end_name == NULL) {
484             ret = -EINVAL;
485             goto out;
486         }
487         if ((end_name - p_name) > sizeof(bs->auto_backing_file) - 1) {
488             ret = -EINVAL;
489             goto out;
490         }
491 
492         pstrcpy(bs->auto_backing_file, end_name - p_name + 1, p_name);
493         pstrcpy(bs->backing_file, sizeof(bs->backing_file),
494                 bs->auto_backing_file);
495         pstrcpy(bs->backing_format, sizeof(bs->backing_format),
496                 "vmdk");
497     }
498 
499 out:
500     g_free(desc);
501     return ret;
502 }
503 
504 /* Create and append extent to the extent array. Return the added VmdkExtent
505  * address. return NULL if allocation failed. */
506 static int vmdk_add_extent(BlockDriverState *bs,
507                            BdrvChild *file, bool flat, int64_t sectors,
508                            int64_t l1_offset, int64_t l1_backup_offset,
509                            uint32_t l1_size,
510                            int l2_size, uint64_t cluster_sectors,
511                            VmdkExtent **new_extent,
512                            Error **errp)
513 {
514     VmdkExtent *extent;
515     BDRVVmdkState *s = bs->opaque;
516     int64_t nb_sectors;
517 
518     if (cluster_sectors > 0x200000) {
519         /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
520         error_setg(errp, "Invalid granularity, image may be corrupt");
521         return -EFBIG;
522     }
523     if (l1_size > 32 * 1024 * 1024) {
524         /*
525          * Although with big capacity and small l1_entry_sectors, we can get a
526          * big l1_size, we don't want unbounded value to allocate the table.
527          * Limit it to 32M, which is enough to store:
528          *     8TB  - for both VMDK3 & VMDK4 with
529          *            minimal cluster size: 512B
530          *            minimal L2 table size: 512 entries
531          *            8 TB is still more than the maximal value supported for
532          *            VMDK3 & VMDK4 which is 2TB.
533          *     64TB - for "ESXi seSparse Extent"
534          *            minimal cluster size: 512B (default is 4KB)
535          *            L2 table size: 4096 entries (const).
536          *            64TB is more than the maximal value supported for
537          *            seSparse VMDKs (which is slightly less than 64TB)
538          */
539         error_setg(errp, "L1 size too big");
540         return -EFBIG;
541     }
542 
543     nb_sectors = bdrv_nb_sectors(file->bs);
544     if (nb_sectors < 0) {
545         return nb_sectors;
546     }
547 
548     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
549     extent = &s->extents[s->num_extents];
550     s->num_extents++;
551 
552     memset(extent, 0, sizeof(VmdkExtent));
553     extent->file = file;
554     extent->flat = flat;
555     extent->sectors = sectors;
556     extent->l1_table_offset = l1_offset;
557     extent->l1_backup_table_offset = l1_backup_offset;
558     extent->l1_size = l1_size;
559     extent->l1_entry_sectors = l2_size * cluster_sectors;
560     extent->l2_size = l2_size;
561     extent->cluster_sectors = flat ? sectors : cluster_sectors;
562     extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
563     extent->entry_size = sizeof(uint32_t);
564 
565     if (s->num_extents > 1) {
566         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
567     } else {
568         extent->end_sector = extent->sectors;
569     }
570     bs->total_sectors = extent->end_sector;
571     if (new_extent) {
572         *new_extent = extent;
573     }
574     return 0;
575 }
576 
577 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
578                             Error **errp)
579 {
580     int ret;
581     size_t l1_size;
582     int i;
583 
584     /* read the L1 table */
585     l1_size = extent->l1_size * extent->entry_size;
586     extent->l1_table = g_try_malloc(l1_size);
587     if (l1_size && extent->l1_table == NULL) {
588         return -ENOMEM;
589     }
590 
591     ret = bdrv_pread(extent->file, extent->l1_table_offset, l1_size,
592                      extent->l1_table, 0);
593     if (ret < 0) {
594         bdrv_refresh_filename(extent->file->bs);
595         error_setg_errno(errp, -ret,
596                          "Could not read l1 table from extent '%s'",
597                          extent->file->bs->filename);
598         goto fail_l1;
599     }
600     for (i = 0; i < extent->l1_size; i++) {
601         if (extent->entry_size == sizeof(uint64_t)) {
602             le64_to_cpus((uint64_t *)extent->l1_table + i);
603         } else {
604             assert(extent->entry_size == sizeof(uint32_t));
605             le32_to_cpus((uint32_t *)extent->l1_table + i);
606         }
607     }
608 
609     if (extent->l1_backup_table_offset) {
610         assert(!extent->sesparse);
611         extent->l1_backup_table = g_try_malloc(l1_size);
612         if (l1_size && extent->l1_backup_table == NULL) {
613             ret = -ENOMEM;
614             goto fail_l1;
615         }
616         ret = bdrv_pread(extent->file, extent->l1_backup_table_offset,
617                          l1_size, extent->l1_backup_table, 0);
618         if (ret < 0) {
619             bdrv_refresh_filename(extent->file->bs);
620             error_setg_errno(errp, -ret,
621                              "Could not read l1 backup table from extent '%s'",
622                              extent->file->bs->filename);
623             goto fail_l1b;
624         }
625         for (i = 0; i < extent->l1_size; i++) {
626             le32_to_cpus(&extent->l1_backup_table[i]);
627         }
628     }
629 
630     extent->l2_cache =
631         g_malloc(extent->entry_size * extent->l2_size * L2_CACHE_SIZE);
632     return 0;
633  fail_l1b:
634     g_free(extent->l1_backup_table);
635  fail_l1:
636     g_free(extent->l1_table);
637     return ret;
638 }
639 
640 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
641                                  BdrvChild *file,
642                                  int flags, Error **errp)
643 {
644     int ret;
645     uint32_t magic;
646     VMDK3Header header;
647     VmdkExtent *extent = NULL;
648 
649     ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0);
650     if (ret < 0) {
651         bdrv_refresh_filename(file->bs);
652         error_setg_errno(errp, -ret,
653                          "Could not read header from file '%s'",
654                          file->bs->filename);
655         return ret;
656     }
657     ret = vmdk_add_extent(bs, file, false,
658                           le32_to_cpu(header.disk_sectors),
659                           (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
660                           0,
661                           le32_to_cpu(header.l1dir_size),
662                           4096,
663                           le32_to_cpu(header.granularity),
664                           &extent,
665                           errp);
666     if (ret < 0) {
667         return ret;
668     }
669     ret = vmdk_init_tables(bs, extent, errp);
670     if (ret) {
671         /* free extent allocated by vmdk_add_extent */
672         vmdk_free_last_extent(bs);
673     }
674     return ret;
675 }
676 
677 #define SESPARSE_CONST_HEADER_MAGIC UINT64_C(0x00000000cafebabe)
678 #define SESPARSE_VOLATILE_HEADER_MAGIC UINT64_C(0x00000000cafecafe)
679 
680 /* Strict checks - format not officially documented */
681 static int check_se_sparse_const_header(VMDKSESparseConstHeader *header,
682                                         Error **errp)
683 {
684     header->magic = le64_to_cpu(header->magic);
685     header->version = le64_to_cpu(header->version);
686     header->grain_size = le64_to_cpu(header->grain_size);
687     header->grain_table_size = le64_to_cpu(header->grain_table_size);
688     header->flags = le64_to_cpu(header->flags);
689     header->reserved1 = le64_to_cpu(header->reserved1);
690     header->reserved2 = le64_to_cpu(header->reserved2);
691     header->reserved3 = le64_to_cpu(header->reserved3);
692     header->reserved4 = le64_to_cpu(header->reserved4);
693 
694     header->volatile_header_offset =
695         le64_to_cpu(header->volatile_header_offset);
696     header->volatile_header_size = le64_to_cpu(header->volatile_header_size);
697 
698     header->journal_header_offset = le64_to_cpu(header->journal_header_offset);
699     header->journal_header_size = le64_to_cpu(header->journal_header_size);
700 
701     header->journal_offset = le64_to_cpu(header->journal_offset);
702     header->journal_size = le64_to_cpu(header->journal_size);
703 
704     header->grain_dir_offset = le64_to_cpu(header->grain_dir_offset);
705     header->grain_dir_size = le64_to_cpu(header->grain_dir_size);
706 
707     header->grain_tables_offset = le64_to_cpu(header->grain_tables_offset);
708     header->grain_tables_size = le64_to_cpu(header->grain_tables_size);
709 
710     header->free_bitmap_offset = le64_to_cpu(header->free_bitmap_offset);
711     header->free_bitmap_size = le64_to_cpu(header->free_bitmap_size);
712 
713     header->backmap_offset = le64_to_cpu(header->backmap_offset);
714     header->backmap_size = le64_to_cpu(header->backmap_size);
715 
716     header->grains_offset = le64_to_cpu(header->grains_offset);
717     header->grains_size = le64_to_cpu(header->grains_size);
718 
719     if (header->magic != SESPARSE_CONST_HEADER_MAGIC) {
720         error_setg(errp, "Bad const header magic: 0x%016" PRIx64,
721                    header->magic);
722         return -EINVAL;
723     }
724 
725     if (header->version != 0x0000000200000001) {
726         error_setg(errp, "Unsupported version: 0x%016" PRIx64,
727                    header->version);
728         return -ENOTSUP;
729     }
730 
731     if (header->grain_size != 8) {
732         error_setg(errp, "Unsupported grain size: %" PRIu64,
733                    header->grain_size);
734         return -ENOTSUP;
735     }
736 
737     if (header->grain_table_size != 64) {
738         error_setg(errp, "Unsupported grain table size: %" PRIu64,
739                    header->grain_table_size);
740         return -ENOTSUP;
741     }
742 
743     if (header->flags != 0) {
744         error_setg(errp, "Unsupported flags: 0x%016" PRIx64,
745                    header->flags);
746         return -ENOTSUP;
747     }
748 
749     if (header->reserved1 != 0 || header->reserved2 != 0 ||
750         header->reserved3 != 0 || header->reserved4 != 0) {
751         error_setg(errp, "Unsupported reserved bits:"
752                    " 0x%016" PRIx64 " 0x%016" PRIx64
753                    " 0x%016" PRIx64 " 0x%016" PRIx64,
754                    header->reserved1, header->reserved2,
755                    header->reserved3, header->reserved4);
756         return -ENOTSUP;
757     }
758 
759     /* check that padding is 0 */
760     if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
761         error_setg(errp, "Unsupported non-zero const header padding");
762         return -ENOTSUP;
763     }
764 
765     return 0;
766 }
767 
768 static int check_se_sparse_volatile_header(VMDKSESparseVolatileHeader *header,
769                                            Error **errp)
770 {
771     header->magic = le64_to_cpu(header->magic);
772     header->free_gt_number = le64_to_cpu(header->free_gt_number);
773     header->next_txn_seq_number = le64_to_cpu(header->next_txn_seq_number);
774     header->replay_journal = le64_to_cpu(header->replay_journal);
775 
776     if (header->magic != SESPARSE_VOLATILE_HEADER_MAGIC) {
777         error_setg(errp, "Bad volatile header magic: 0x%016" PRIx64,
778                    header->magic);
779         return -EINVAL;
780     }
781 
782     if (header->replay_journal) {
783         error_setg(errp, "Image is dirty, Replaying journal not supported");
784         return -ENOTSUP;
785     }
786 
787     /* check that padding is 0 */
788     if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
789         error_setg(errp, "Unsupported non-zero volatile header padding");
790         return -ENOTSUP;
791     }
792 
793     return 0;
794 }
795 
796 static int vmdk_open_se_sparse(BlockDriverState *bs,
797                                BdrvChild *file,
798                                int flags, Error **errp)
799 {
800     int ret;
801     VMDKSESparseConstHeader const_header;
802     VMDKSESparseVolatileHeader volatile_header;
803     VmdkExtent *extent = NULL;
804 
805     ret = bdrv_apply_auto_read_only(bs,
806             "No write support for seSparse images available", errp);
807     if (ret < 0) {
808         return ret;
809     }
810 
811     assert(sizeof(const_header) == SECTOR_SIZE);
812 
813     ret = bdrv_pread(file, 0, sizeof(const_header), &const_header, 0);
814     if (ret < 0) {
815         bdrv_refresh_filename(file->bs);
816         error_setg_errno(errp, -ret,
817                          "Could not read const header from file '%s'",
818                          file->bs->filename);
819         return ret;
820     }
821 
822     /* check const header */
823     ret = check_se_sparse_const_header(&const_header, errp);
824     if (ret < 0) {
825         return ret;
826     }
827 
828     assert(sizeof(volatile_header) == SECTOR_SIZE);
829 
830     ret = bdrv_pread(file, const_header.volatile_header_offset * SECTOR_SIZE,
831                      sizeof(volatile_header), &volatile_header, 0);
832     if (ret < 0) {
833         bdrv_refresh_filename(file->bs);
834         error_setg_errno(errp, -ret,
835                          "Could not read volatile header from file '%s'",
836                          file->bs->filename);
837         return ret;
838     }
839 
840     /* check volatile header */
841     ret = check_se_sparse_volatile_header(&volatile_header, errp);
842     if (ret < 0) {
843         return ret;
844     }
845 
846     ret = vmdk_add_extent(bs, file, false,
847                           const_header.capacity,
848                           const_header.grain_dir_offset * SECTOR_SIZE,
849                           0,
850                           const_header.grain_dir_size *
851                           SECTOR_SIZE / sizeof(uint64_t),
852                           const_header.grain_table_size *
853                           SECTOR_SIZE / sizeof(uint64_t),
854                           const_header.grain_size,
855                           &extent,
856                           errp);
857     if (ret < 0) {
858         return ret;
859     }
860 
861     extent->sesparse = true;
862     extent->sesparse_l2_tables_offset = const_header.grain_tables_offset;
863     extent->sesparse_clusters_offset = const_header.grains_offset;
864     extent->entry_size = sizeof(uint64_t);
865 
866     ret = vmdk_init_tables(bs, extent, errp);
867     if (ret) {
868         /* free extent allocated by vmdk_add_extent */
869         vmdk_free_last_extent(bs);
870     }
871 
872     return ret;
873 }
874 
875 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
876                                QDict *options, Error **errp);
877 
878 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
879 {
880     int64_t size;
881     char *buf;
882     int ret;
883 
884     size = bdrv_getlength(file->bs);
885     if (size < 0) {
886         error_setg_errno(errp, -size, "Could not access file");
887         return NULL;
888     }
889 
890     if (size < 4) {
891         /* Both descriptor file and sparse image must be much larger than 4
892          * bytes, also callers of vmdk_read_desc want to compare the first 4
893          * bytes with VMDK4_MAGIC, let's error out if less is read. */
894         error_setg(errp, "File is too small, not a valid image");
895         return NULL;
896     }
897 
898     size = MIN(size, (1 << 20) - 1);  /* avoid unbounded allocation */
899     buf = g_malloc(size + 1);
900 
901     ret = bdrv_pread(file, desc_offset, size, buf, 0);
902     if (ret < 0) {
903         error_setg_errno(errp, -ret, "Could not read from file");
904         g_free(buf);
905         return NULL;
906     }
907     buf[size] = 0;
908 
909     return buf;
910 }
911 
912 static int vmdk_open_vmdk4(BlockDriverState *bs,
913                            BdrvChild *file,
914                            int flags, QDict *options, Error **errp)
915 {
916     int ret;
917     uint32_t magic;
918     uint32_t l1_size, l1_entry_sectors;
919     VMDK4Header header;
920     VmdkExtent *extent = NULL;
921     BDRVVmdkState *s = bs->opaque;
922     int64_t l1_backup_offset = 0;
923     bool compressed;
924 
925     ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0);
926     if (ret < 0) {
927         bdrv_refresh_filename(file->bs);
928         error_setg_errno(errp, -ret,
929                          "Could not read header from file '%s'",
930                          file->bs->filename);
931         return -EINVAL;
932     }
933     if (header.capacity == 0) {
934         uint64_t desc_offset = le64_to_cpu(header.desc_offset);
935         if (desc_offset) {
936             char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
937             if (!buf) {
938                 return -EINVAL;
939             }
940             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
941             g_free(buf);
942             return ret;
943         }
944     }
945 
946     if (!s->create_type) {
947         s->create_type = g_strdup("monolithicSparse");
948     }
949 
950     if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
951         /*
952          * The footer takes precedence over the header, so read it in. The
953          * footer starts at offset -1024 from the end: One sector for the
954          * footer, and another one for the end-of-stream marker.
955          */
956         struct {
957             struct {
958                 uint64_t val;
959                 uint32_t size;
960                 uint32_t type;
961                 uint8_t pad[512 - 16];
962             } QEMU_PACKED footer_marker;
963 
964             uint32_t magic;
965             VMDK4Header header;
966             uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
967 
968             struct {
969                 uint64_t val;
970                 uint32_t size;
971                 uint32_t type;
972                 uint8_t pad[512 - 16];
973             } QEMU_PACKED eos_marker;
974         } QEMU_PACKED footer;
975 
976         ret = bdrv_pread(file, bs->file->bs->total_sectors * 512 - 1536,
977                          sizeof(footer), &footer, 0);
978         if (ret < 0) {
979             error_setg_errno(errp, -ret, "Failed to read footer");
980             return ret;
981         }
982 
983         /* Some sanity checks for the footer */
984         if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
985             le32_to_cpu(footer.footer_marker.size) != 0  ||
986             le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
987             le64_to_cpu(footer.eos_marker.val) != 0  ||
988             le32_to_cpu(footer.eos_marker.size) != 0  ||
989             le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
990         {
991             error_setg(errp, "Invalid footer");
992             return -EINVAL;
993         }
994 
995         header = footer.header;
996     }
997 
998     compressed =
999         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1000     if (le32_to_cpu(header.version) > 3) {
1001         error_setg(errp, "Unsupported VMDK version %" PRIu32,
1002                    le32_to_cpu(header.version));
1003         return -ENOTSUP;
1004     } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
1005                !compressed) {
1006         /* VMware KB 2064959 explains that version 3 added support for
1007          * persistent changed block tracking (CBT), and backup software can
1008          * read it as version=1 if it doesn't care about the changed area
1009          * information. So we are safe to enable read only. */
1010         error_setg(errp, "VMDK version 3 must be read only");
1011         return -EINVAL;
1012     }
1013 
1014     if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
1015         error_setg(errp, "L2 table size too big");
1016         return -EINVAL;
1017     }
1018 
1019     l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
1020                         * le64_to_cpu(header.granularity);
1021     if (l1_entry_sectors == 0) {
1022         error_setg(errp, "L1 entry size is invalid");
1023         return -EINVAL;
1024     }
1025     l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
1026                 / l1_entry_sectors;
1027     if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
1028         l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
1029     }
1030     if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
1031         error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
1032                    (int64_t)(le64_to_cpu(header.grain_offset)
1033                              * BDRV_SECTOR_SIZE));
1034         return -EINVAL;
1035     }
1036 
1037     ret = vmdk_add_extent(bs, file, false,
1038                           le64_to_cpu(header.capacity),
1039                           le64_to_cpu(header.gd_offset) << 9,
1040                           l1_backup_offset,
1041                           l1_size,
1042                           le32_to_cpu(header.num_gtes_per_gt),
1043                           le64_to_cpu(header.granularity),
1044                           &extent,
1045                           errp);
1046     if (ret < 0) {
1047         return ret;
1048     }
1049     extent->compressed =
1050         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1051     if (extent->compressed) {
1052         g_free(s->create_type);
1053         s->create_type = g_strdup("streamOptimized");
1054     }
1055     extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
1056     extent->version = le32_to_cpu(header.version);
1057     extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
1058     ret = vmdk_init_tables(bs, extent, errp);
1059     if (ret) {
1060         /* free extent allocated by vmdk_add_extent */
1061         vmdk_free_last_extent(bs);
1062     }
1063     return ret;
1064 }
1065 
1066 /* find an option value out of descriptor file */
1067 static int vmdk_parse_description(const char *desc, const char *opt_name,
1068         char *buf, int buf_size)
1069 {
1070     char *opt_pos, *opt_end;
1071     const char *end = desc + strlen(desc);
1072 
1073     opt_pos = strstr(desc, opt_name);
1074     if (!opt_pos) {
1075         return VMDK_ERROR;
1076     }
1077     /* Skip "=\"" following opt_name */
1078     opt_pos += strlen(opt_name) + 2;
1079     if (opt_pos >= end) {
1080         return VMDK_ERROR;
1081     }
1082     opt_end = opt_pos;
1083     while (opt_end < end && *opt_end != '"') {
1084         opt_end++;
1085     }
1086     if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
1087         return VMDK_ERROR;
1088     }
1089     pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
1090     return VMDK_OK;
1091 }
1092 
1093 /* Open an extent file and append to bs array */
1094 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
1095                             char *buf, QDict *options, Error **errp)
1096 {
1097     uint32_t magic;
1098 
1099     magic = ldl_be_p(buf);
1100     switch (magic) {
1101         case VMDK3_MAGIC:
1102             return vmdk_open_vmfs_sparse(bs, file, flags, errp);
1103         case VMDK4_MAGIC:
1104             return vmdk_open_vmdk4(bs, file, flags, options, errp);
1105         default:
1106             error_setg(errp, "Image not in VMDK format");
1107             return -EINVAL;
1108     }
1109 }
1110 
1111 static const char *next_line(const char *s)
1112 {
1113     while (*s) {
1114         if (*s == '\n') {
1115             return s + 1;
1116         }
1117         s++;
1118     }
1119     return s;
1120 }
1121 
1122 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
1123                               QDict *options, Error **errp)
1124 {
1125     int ret;
1126     int matches;
1127     char access[11];
1128     char type[11];
1129     char fname[512];
1130     const char *p, *np;
1131     int64_t sectors = 0;
1132     int64_t flat_offset;
1133     char *desc_file_dir = NULL;
1134     char *extent_path;
1135     BdrvChild *extent_file;
1136     BdrvChildRole extent_role;
1137     BDRVVmdkState *s = bs->opaque;
1138     VmdkExtent *extent = NULL;
1139     char extent_opt_prefix[32];
1140     Error *local_err = NULL;
1141 
1142     for (p = desc; *p; p = next_line(p)) {
1143         /* parse extent line in one of below formats:
1144          *
1145          * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
1146          * RW [size in sectors] SPARSE "file-name.vmdk"
1147          * RW [size in sectors] VMFS "file-name.vmdk"
1148          * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
1149          * RW [size in sectors] SESPARSE "file-name.vmdk"
1150          */
1151         flat_offset = -1;
1152         matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
1153                          access, &sectors, type, fname, &flat_offset);
1154         if (matches < 4 || strcmp(access, "RW")) {
1155             continue;
1156         } else if (!strcmp(type, "FLAT")) {
1157             if (matches != 5 || flat_offset < 0) {
1158                 goto invalid;
1159             }
1160         } else if (!strcmp(type, "VMFS")) {
1161             if (matches == 4) {
1162                 flat_offset = 0;
1163             } else {
1164                 goto invalid;
1165             }
1166         } else if (matches != 4) {
1167             goto invalid;
1168         }
1169 
1170         if (sectors <= 0 ||
1171             (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
1172              strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") &&
1173              strcmp(type, "SESPARSE")) ||
1174             (strcmp(access, "RW"))) {
1175             continue;
1176         }
1177 
1178         if (path_is_absolute(fname)) {
1179             extent_path = g_strdup(fname);
1180         } else {
1181             if (!desc_file_dir) {
1182                 desc_file_dir = bdrv_dirname(bs->file->bs, errp);
1183                 if (!desc_file_dir) {
1184                     bdrv_refresh_filename(bs->file->bs);
1185                     error_prepend(errp, "Cannot use relative paths with VMDK "
1186                                   "descriptor file '%s': ",
1187                                   bs->file->bs->filename);
1188                     ret = -EINVAL;
1189                     goto out;
1190                 }
1191             }
1192 
1193             extent_path = g_strconcat(desc_file_dir, fname, NULL);
1194         }
1195 
1196         ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
1197         assert(ret < 32);
1198 
1199         extent_role = BDRV_CHILD_DATA;
1200         if (strcmp(type, "FLAT") != 0 && strcmp(type, "VMFS") != 0) {
1201             /* non-flat extents have metadata */
1202             extent_role |= BDRV_CHILD_METADATA;
1203         }
1204 
1205         extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
1206                                       bs, &child_of_bds, extent_role, false,
1207                                       &local_err);
1208         g_free(extent_path);
1209         if (local_err) {
1210             error_propagate(errp, local_err);
1211             ret = -EINVAL;
1212             goto out;
1213         }
1214 
1215         /* save to extents array */
1216         if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
1217             /* FLAT extent */
1218 
1219             ret = vmdk_add_extent(bs, extent_file, true, sectors,
1220                             0, 0, 0, 0, 0, &extent, errp);
1221             if (ret < 0) {
1222                 bdrv_unref_child(bs, extent_file);
1223                 goto out;
1224             }
1225             extent->flat_start_offset = flat_offset << 9;
1226         } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
1227             /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
1228             char *buf = vmdk_read_desc(extent_file, 0, errp);
1229             if (!buf) {
1230                 ret = -EINVAL;
1231             } else {
1232                 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
1233                                        options, errp);
1234             }
1235             g_free(buf);
1236             if (ret) {
1237                 bdrv_unref_child(bs, extent_file);
1238                 goto out;
1239             }
1240             extent = &s->extents[s->num_extents - 1];
1241         } else if (!strcmp(type, "SESPARSE")) {
1242             ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
1243             if (ret) {
1244                 bdrv_unref_child(bs, extent_file);
1245                 goto out;
1246             }
1247             extent = &s->extents[s->num_extents - 1];
1248         } else {
1249             error_setg(errp, "Unsupported extent type '%s'", type);
1250             bdrv_unref_child(bs, extent_file);
1251             ret = -ENOTSUP;
1252             goto out;
1253         }
1254         extent->type = g_strdup(type);
1255     }
1256 
1257     ret = 0;
1258     goto out;
1259 
1260 invalid:
1261     np = next_line(p);
1262     assert(np != p);
1263     if (np[-1] == '\n') {
1264         np--;
1265     }
1266     error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
1267     ret = -EINVAL;
1268 
1269 out:
1270     g_free(desc_file_dir);
1271     return ret;
1272 }
1273 
1274 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
1275                                QDict *options, Error **errp)
1276 {
1277     int ret;
1278     char ct[128];
1279     BDRVVmdkState *s = bs->opaque;
1280 
1281     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
1282         error_setg(errp, "invalid VMDK image descriptor");
1283         ret = -EINVAL;
1284         goto exit;
1285     }
1286     if (strcmp(ct, "monolithicFlat") &&
1287         strcmp(ct, "vmfs") &&
1288         strcmp(ct, "vmfsSparse") &&
1289         strcmp(ct, "seSparse") &&
1290         strcmp(ct, "twoGbMaxExtentSparse") &&
1291         strcmp(ct, "twoGbMaxExtentFlat")) {
1292         error_setg(errp, "Unsupported image type '%s'", ct);
1293         ret = -ENOTSUP;
1294         goto exit;
1295     }
1296     s->create_type = g_strdup(ct);
1297     s->desc_offset = 0;
1298     ret = vmdk_parse_extents(buf, bs, options, errp);
1299 exit:
1300     return ret;
1301 }
1302 
1303 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
1304                      Error **errp)
1305 {
1306     char *buf;
1307     int ret;
1308     BDRVVmdkState *s = bs->opaque;
1309     uint32_t magic;
1310 
1311     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
1312                                BDRV_CHILD_IMAGE, false, errp);
1313     if (!bs->file) {
1314         return -EINVAL;
1315     }
1316 
1317     buf = vmdk_read_desc(bs->file, 0, errp);
1318     if (!buf) {
1319         return -EINVAL;
1320     }
1321 
1322     magic = ldl_be_p(buf);
1323     switch (magic) {
1324         case VMDK3_MAGIC:
1325         case VMDK4_MAGIC:
1326             ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
1327                                    errp);
1328             s->desc_offset = 0x200;
1329             break;
1330         default:
1331             /* No data in the descriptor file */
1332             bs->file->role &= ~BDRV_CHILD_DATA;
1333 
1334             /* Must succeed because we have given up permissions if anything */
1335             bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1336 
1337             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1338             break;
1339     }
1340     if (ret) {
1341         goto fail;
1342     }
1343 
1344     /* try to open parent images, if exist */
1345     ret = vmdk_parent_open(bs);
1346     if (ret) {
1347         goto fail;
1348     }
1349     ret = vmdk_read_cid(bs, 0, &s->cid);
1350     if (ret) {
1351         goto fail;
1352     }
1353     ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1354     if (ret) {
1355         goto fail;
1356     }
1357     qemu_co_mutex_init(&s->lock);
1358 
1359     /* Disable migration when VMDK images are used */
1360     error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1361                "does not support live migration",
1362                bdrv_get_device_or_node_name(bs));
1363     ret = migrate_add_blocker(s->migration_blocker, errp);
1364     if (ret < 0) {
1365         error_free(s->migration_blocker);
1366         goto fail;
1367     }
1368 
1369     g_free(buf);
1370     return 0;
1371 
1372 fail:
1373     g_free(buf);
1374     g_free(s->create_type);
1375     s->create_type = NULL;
1376     vmdk_free_extents(bs);
1377     return ret;
1378 }
1379 
1380 
1381 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1382 {
1383     BDRVVmdkState *s = bs->opaque;
1384     int i;
1385 
1386     for (i = 0; i < s->num_extents; i++) {
1387         if (!s->extents[i].flat) {
1388             bs->bl.pwrite_zeroes_alignment =
1389                 MAX(bs->bl.pwrite_zeroes_alignment,
1390                     s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1391         }
1392     }
1393 }
1394 
1395 /**
1396  * get_whole_cluster
1397  *
1398  * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1399  * to the cluster at @cluster_sector_num. If @zeroed is true, we're overwriting
1400  * a zeroed cluster in the current layer and must not copy data from the
1401  * backing file.
1402  *
1403  * If @skip_start_sector < @skip_end_sector, the relative range
1404  * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1405  * it for call to write user data in the request.
1406  */
1407 static int get_whole_cluster(BlockDriverState *bs,
1408                              VmdkExtent *extent,
1409                              uint64_t cluster_offset,
1410                              uint64_t offset,
1411                              uint64_t skip_start_bytes,
1412                              uint64_t skip_end_bytes,
1413                              bool zeroed)
1414 {
1415     int ret = VMDK_OK;
1416     int64_t cluster_bytes;
1417     uint8_t *whole_grain;
1418     bool copy_from_backing;
1419 
1420     /* For COW, align request sector_num to cluster start */
1421     cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1422     offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1423     whole_grain = qemu_blockalign(bs, cluster_bytes);
1424     copy_from_backing = bs->backing && !zeroed;
1425 
1426     if (!copy_from_backing) {
1427         memset(whole_grain, 0, skip_start_bytes);
1428         memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1429     }
1430 
1431     assert(skip_end_bytes <= cluster_bytes);
1432     /* we will be here if it's first write on non-exist grain(cluster).
1433      * try to read from parent image, if exist */
1434     if (bs->backing && !vmdk_is_cid_valid(bs)) {
1435         ret = VMDK_ERROR;
1436         goto exit;
1437     }
1438 
1439     /* Read backing data before skip range */
1440     if (skip_start_bytes > 0) {
1441         if (copy_from_backing) {
1442             /* qcow2 emits this on bs->file instead of bs->backing */
1443             BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1444             ret = bdrv_pread(bs->backing, offset, skip_start_bytes,
1445                              whole_grain, 0);
1446             if (ret < 0) {
1447                 ret = VMDK_ERROR;
1448                 goto exit;
1449             }
1450         }
1451         BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1452         ret = bdrv_pwrite(extent->file, cluster_offset, skip_start_bytes,
1453                           whole_grain, 0);
1454         if (ret < 0) {
1455             ret = VMDK_ERROR;
1456             goto exit;
1457         }
1458     }
1459     /* Read backing data after skip range */
1460     if (skip_end_bytes < cluster_bytes) {
1461         if (copy_from_backing) {
1462             /* qcow2 emits this on bs->file instead of bs->backing */
1463             BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1464             ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1465                              cluster_bytes - skip_end_bytes,
1466                              whole_grain + skip_end_bytes, 0);
1467             if (ret < 0) {
1468                 ret = VMDK_ERROR;
1469                 goto exit;
1470             }
1471         }
1472         BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1473         ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1474                           cluster_bytes - skip_end_bytes,
1475                           whole_grain + skip_end_bytes, 0);
1476         if (ret < 0) {
1477             ret = VMDK_ERROR;
1478             goto exit;
1479         }
1480     }
1481 
1482     ret = VMDK_OK;
1483 exit:
1484     qemu_vfree(whole_grain);
1485     return ret;
1486 }
1487 
1488 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1489                          uint32_t offset)
1490 {
1491     offset = cpu_to_le32(offset);
1492     /* update L2 table */
1493     BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1494     if (bdrv_pwrite(extent->file,
1495                 ((int64_t)m_data->l2_offset * 512)
1496                     + (m_data->l2_index * sizeof(offset)),
1497                 sizeof(offset), &offset, 0) < 0) {
1498         return VMDK_ERROR;
1499     }
1500     /* update backup L2 table */
1501     if (extent->l1_backup_table_offset != 0) {
1502         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1503         if (bdrv_pwrite(extent->file,
1504                     ((int64_t)m_data->l2_offset * 512)
1505                         + (m_data->l2_index * sizeof(offset)),
1506                     sizeof(offset), &offset, 0) < 0) {
1507             return VMDK_ERROR;
1508         }
1509     }
1510     if (bdrv_flush(extent->file->bs) < 0) {
1511         return VMDK_ERROR;
1512     }
1513     if (m_data->l2_cache_entry) {
1514         *m_data->l2_cache_entry = offset;
1515     }
1516 
1517     return VMDK_OK;
1518 }
1519 
1520 /**
1521  * get_cluster_offset
1522  *
1523  * Look up cluster offset in extent file by sector number, and store in
1524  * @cluster_offset.
1525  *
1526  * For flat extents, the start offset as parsed from the description file is
1527  * returned.
1528  *
1529  * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1530  * offset for a new cluster and update L2 cache. If there is a backing file,
1531  * COW is done before returning; otherwise, zeroes are written to the allocated
1532  * cluster. Both COW and zero writing skips the sector range
1533  * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1534  * has new data to write there.
1535  *
1536  * Returns: VMDK_OK if cluster exists and mapped in the image.
1537  *          VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1538  *          VMDK_ERROR if failed.
1539  */
1540 static int get_cluster_offset(BlockDriverState *bs,
1541                               VmdkExtent *extent,
1542                               VmdkMetaData *m_data,
1543                               uint64_t offset,
1544                               bool allocate,
1545                               uint64_t *cluster_offset,
1546                               uint64_t skip_start_bytes,
1547                               uint64_t skip_end_bytes)
1548 {
1549     unsigned int l1_index, l2_offset, l2_index;
1550     int min_index, i, j;
1551     uint32_t min_count;
1552     void *l2_table;
1553     bool zeroed = false;
1554     int64_t ret;
1555     int64_t cluster_sector;
1556     unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
1557 
1558     if (m_data) {
1559         m_data->new_allocation = false;
1560     }
1561     if (extent->flat) {
1562         *cluster_offset = extent->flat_start_offset;
1563         return VMDK_OK;
1564     }
1565 
1566     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1567     l1_index = (offset >> 9) / extent->l1_entry_sectors;
1568     if (l1_index >= extent->l1_size) {
1569         return VMDK_ERROR;
1570     }
1571     if (extent->sesparse) {
1572         uint64_t l2_offset_u64;
1573 
1574         assert(extent->entry_size == sizeof(uint64_t));
1575 
1576         l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
1577         if (l2_offset_u64 == 0) {
1578             l2_offset = 0;
1579         } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
1580             /*
1581              * Top most nibble is 0x1 if grain table is allocated.
1582              * strict check - top most 4 bytes must be 0x10000000 since max
1583              * supported size is 64TB for disk - so no more than 64TB / 16MB
1584              * grain directories which is smaller than uint32,
1585              * where 16MB is the only supported default grain table coverage.
1586              */
1587             return VMDK_ERROR;
1588         } else {
1589             l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
1590             l2_offset_u64 = extent->sesparse_l2_tables_offset +
1591                 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
1592             if (l2_offset_u64 > 0x00000000ffffffff) {
1593                 return VMDK_ERROR;
1594             }
1595             l2_offset = (unsigned int)(l2_offset_u64);
1596         }
1597     } else {
1598         assert(extent->entry_size == sizeof(uint32_t));
1599         l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
1600     }
1601     if (!l2_offset) {
1602         return VMDK_UNALLOC;
1603     }
1604     for (i = 0; i < L2_CACHE_SIZE; i++) {
1605         if (l2_offset == extent->l2_cache_offsets[i]) {
1606             /* increment the hit count */
1607             if (++extent->l2_cache_counts[i] == 0xffffffff) {
1608                 for (j = 0; j < L2_CACHE_SIZE; j++) {
1609                     extent->l2_cache_counts[j] >>= 1;
1610                 }
1611             }
1612             l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
1613             goto found;
1614         }
1615     }
1616     /* not found: load a new entry in the least used one */
1617     min_index = 0;
1618     min_count = 0xffffffff;
1619     for (i = 0; i < L2_CACHE_SIZE; i++) {
1620         if (extent->l2_cache_counts[i] < min_count) {
1621             min_count = extent->l2_cache_counts[i];
1622             min_index = i;
1623         }
1624     }
1625     l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
1626     BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1627     if (bdrv_pread(extent->file,
1628                 (int64_t)l2_offset * 512,
1629                 l2_size_bytes,
1630                 l2_table,
1631                 0
1632             ) < 0) {
1633         return VMDK_ERROR;
1634     }
1635 
1636     extent->l2_cache_offsets[min_index] = l2_offset;
1637     extent->l2_cache_counts[min_index] = 1;
1638  found:
1639     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1640     if (m_data) {
1641         m_data->l1_index = l1_index;
1642         m_data->l2_index = l2_index;
1643         m_data->l2_offset = l2_offset;
1644         m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
1645     }
1646 
1647     if (extent->sesparse) {
1648         cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
1649         switch (cluster_sector & 0xf000000000000000) {
1650         case 0x0000000000000000:
1651             /* unallocated grain */
1652             if (cluster_sector != 0) {
1653                 return VMDK_ERROR;
1654             }
1655             break;
1656         case 0x1000000000000000:
1657             /* scsi-unmapped grain - fallthrough */
1658         case 0x2000000000000000:
1659             /* zero grain */
1660             zeroed = true;
1661             break;
1662         case 0x3000000000000000:
1663             /* allocated grain */
1664             cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
1665                               ((cluster_sector & 0x0000ffffffffffff) << 12));
1666             cluster_sector = extent->sesparse_clusters_offset +
1667                 cluster_sector * extent->cluster_sectors;
1668             break;
1669         default:
1670             return VMDK_ERROR;
1671         }
1672     } else {
1673         cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
1674 
1675         if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1676             zeroed = true;
1677         }
1678     }
1679 
1680     if (!cluster_sector || zeroed) {
1681         if (!allocate) {
1682             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1683         }
1684         assert(!extent->sesparse);
1685 
1686         if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1687             return VMDK_ERROR;
1688         }
1689 
1690         cluster_sector = extent->next_cluster_sector;
1691         extent->next_cluster_sector += extent->cluster_sectors;
1692 
1693         /* First of all we write grain itself, to avoid race condition
1694          * that may to corrupt the image.
1695          * This problem may occur because of insufficient space on host disk
1696          * or inappropriate VM shutdown.
1697          */
1698         ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1699                                 offset, skip_start_bytes, skip_end_bytes,
1700                                 zeroed);
1701         if (ret) {
1702             return ret;
1703         }
1704         if (m_data) {
1705             m_data->new_allocation = true;
1706         }
1707     }
1708     *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1709     return VMDK_OK;
1710 }
1711 
1712 static VmdkExtent *find_extent(BDRVVmdkState *s,
1713                                 int64_t sector_num, VmdkExtent *start_hint)
1714 {
1715     VmdkExtent *extent = start_hint;
1716 
1717     if (!extent) {
1718         extent = &s->extents[0];
1719     }
1720     while (extent < &s->extents[s->num_extents]) {
1721         if (sector_num < extent->end_sector) {
1722             return extent;
1723         }
1724         extent++;
1725     }
1726     return NULL;
1727 }
1728 
1729 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1730                                                    int64_t offset)
1731 {
1732     uint64_t extent_begin_offset, extent_relative_offset;
1733     uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1734 
1735     extent_begin_offset =
1736         (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1737     extent_relative_offset = offset - extent_begin_offset;
1738     return extent_relative_offset % cluster_size;
1739 }
1740 
1741 static int coroutine_fn vmdk_co_block_status(BlockDriverState *bs,
1742                                              bool want_zero,
1743                                              int64_t offset, int64_t bytes,
1744                                              int64_t *pnum, int64_t *map,
1745                                              BlockDriverState **file)
1746 {
1747     BDRVVmdkState *s = bs->opaque;
1748     int64_t index_in_cluster, n, ret;
1749     uint64_t cluster_offset;
1750     VmdkExtent *extent;
1751 
1752     extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1753     if (!extent) {
1754         return -EIO;
1755     }
1756     qemu_co_mutex_lock(&s->lock);
1757     ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1758                              0, 0);
1759     qemu_co_mutex_unlock(&s->lock);
1760 
1761     index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1762     switch (ret) {
1763     case VMDK_ERROR:
1764         ret = -EIO;
1765         break;
1766     case VMDK_UNALLOC:
1767         ret = 0;
1768         break;
1769     case VMDK_ZEROED:
1770         ret = BDRV_BLOCK_ZERO;
1771         break;
1772     case VMDK_OK:
1773         ret = BDRV_BLOCK_DATA;
1774         if (!extent->compressed) {
1775             ret |= BDRV_BLOCK_OFFSET_VALID;
1776             *map = cluster_offset + index_in_cluster;
1777             if (extent->flat) {
1778                 ret |= BDRV_BLOCK_RECURSE;
1779             }
1780         }
1781         *file = extent->file->bs;
1782         break;
1783     }
1784 
1785     n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1786     *pnum = MIN(n, bytes);
1787     return ret;
1788 }
1789 
1790 static int coroutine_fn
1791 vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1792                   int64_t offset_in_cluster, QEMUIOVector *qiov,
1793                   uint64_t qiov_offset, uint64_t n_bytes,
1794                   uint64_t offset)
1795 {
1796     int ret;
1797     VmdkGrainMarker *data = NULL;
1798     uLongf buf_len;
1799     QEMUIOVector local_qiov;
1800     int64_t write_offset;
1801     int64_t write_end_sector;
1802 
1803     if (extent->compressed) {
1804         void *compressed_data;
1805 
1806         /* Only whole clusters */
1807         if (offset_in_cluster ||
1808             n_bytes > (extent->cluster_sectors * SECTOR_SIZE) ||
1809             (n_bytes < (extent->cluster_sectors * SECTOR_SIZE) &&
1810              offset + n_bytes != extent->end_sector * SECTOR_SIZE))
1811         {
1812             ret = -EINVAL;
1813             goto out;
1814         }
1815 
1816         if (!extent->has_marker) {
1817             ret = -EINVAL;
1818             goto out;
1819         }
1820         buf_len = (extent->cluster_sectors << 9) * 2;
1821         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1822 
1823         compressed_data = g_malloc(n_bytes);
1824         qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1825         ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1826         g_free(compressed_data);
1827 
1828         if (ret != Z_OK || buf_len == 0) {
1829             ret = -EINVAL;
1830             goto out;
1831         }
1832 
1833         data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1834         data->size = cpu_to_le32(buf_len);
1835 
1836         n_bytes = buf_len + sizeof(VmdkGrainMarker);
1837         qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1838 
1839         BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1840     } else {
1841         qemu_iovec_init(&local_qiov, qiov->niov);
1842         qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1843 
1844         BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1845     }
1846 
1847     write_offset = cluster_offset + offset_in_cluster;
1848     ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1849                           &local_qiov, 0);
1850 
1851     write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1852 
1853     if (extent->compressed) {
1854         extent->next_cluster_sector = write_end_sector;
1855     } else {
1856         extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1857                                           write_end_sector);
1858     }
1859 
1860     if (ret < 0) {
1861         goto out;
1862     }
1863     ret = 0;
1864  out:
1865     g_free(data);
1866     if (!extent->compressed) {
1867         qemu_iovec_destroy(&local_qiov);
1868     }
1869     return ret;
1870 }
1871 
1872 static int coroutine_fn
1873 vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1874                  int64_t offset_in_cluster, QEMUIOVector *qiov,
1875                  int bytes)
1876 {
1877     int ret;
1878     int cluster_bytes, buf_bytes;
1879     uint8_t *cluster_buf, *compressed_data;
1880     uint8_t *uncomp_buf;
1881     uint32_t data_len;
1882     VmdkGrainMarker *marker;
1883     uLongf buf_len;
1884 
1885 
1886     if (!extent->compressed) {
1887         BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1888         ret = bdrv_co_preadv(extent->file,
1889                              cluster_offset + offset_in_cluster, bytes,
1890                              qiov, 0);
1891         if (ret < 0) {
1892             return ret;
1893         }
1894         return 0;
1895     }
1896     cluster_bytes = extent->cluster_sectors * 512;
1897     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1898     buf_bytes = cluster_bytes * 2;
1899     cluster_buf = g_malloc(buf_bytes);
1900     uncomp_buf = g_malloc(cluster_bytes);
1901     BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1902     ret = bdrv_pread(extent->file, cluster_offset, buf_bytes, cluster_buf, 0);
1903     if (ret < 0) {
1904         goto out;
1905     }
1906     compressed_data = cluster_buf;
1907     buf_len = cluster_bytes;
1908     data_len = cluster_bytes;
1909     if (extent->has_marker) {
1910         marker = (VmdkGrainMarker *)cluster_buf;
1911         compressed_data = marker->data;
1912         data_len = le32_to_cpu(marker->size);
1913     }
1914     if (!data_len || data_len > buf_bytes) {
1915         ret = -EINVAL;
1916         goto out;
1917     }
1918     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1919     if (ret != Z_OK) {
1920         ret = -EINVAL;
1921         goto out;
1922 
1923     }
1924     if (offset_in_cluster < 0 ||
1925             offset_in_cluster + bytes > buf_len) {
1926         ret = -EINVAL;
1927         goto out;
1928     }
1929     qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1930     ret = 0;
1931 
1932  out:
1933     g_free(uncomp_buf);
1934     g_free(cluster_buf);
1935     return ret;
1936 }
1937 
1938 static int coroutine_fn
1939 vmdk_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
1940                QEMUIOVector *qiov, BdrvRequestFlags flags)
1941 {
1942     BDRVVmdkState *s = bs->opaque;
1943     int ret;
1944     uint64_t n_bytes, offset_in_cluster;
1945     VmdkExtent *extent = NULL;
1946     QEMUIOVector local_qiov;
1947     uint64_t cluster_offset;
1948     uint64_t bytes_done = 0;
1949 
1950     qemu_iovec_init(&local_qiov, qiov->niov);
1951     qemu_co_mutex_lock(&s->lock);
1952 
1953     while (bytes > 0) {
1954         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1955         if (!extent) {
1956             ret = -EIO;
1957             goto fail;
1958         }
1959         ret = get_cluster_offset(bs, extent, NULL,
1960                                  offset, false, &cluster_offset, 0, 0);
1961         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1962 
1963         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1964                              - offset_in_cluster);
1965 
1966         if (ret != VMDK_OK) {
1967             /* if not allocated, try to read from parent image, if exist */
1968             if (bs->backing && ret != VMDK_ZEROED) {
1969                 if (!vmdk_is_cid_valid(bs)) {
1970                     ret = -EINVAL;
1971                     goto fail;
1972                 }
1973 
1974                 qemu_iovec_reset(&local_qiov);
1975                 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1976 
1977                 /* qcow2 emits this on bs->file instead of bs->backing */
1978                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1979                 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1980                                      &local_qiov, 0);
1981                 if (ret < 0) {
1982                     goto fail;
1983                 }
1984             } else {
1985                 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1986             }
1987         } else {
1988             qemu_iovec_reset(&local_qiov);
1989             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1990 
1991             ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1992                                    &local_qiov, n_bytes);
1993             if (ret) {
1994                 goto fail;
1995             }
1996         }
1997         bytes -= n_bytes;
1998         offset += n_bytes;
1999         bytes_done += n_bytes;
2000     }
2001 
2002     ret = 0;
2003 fail:
2004     qemu_co_mutex_unlock(&s->lock);
2005     qemu_iovec_destroy(&local_qiov);
2006 
2007     return ret;
2008 }
2009 
2010 /**
2011  * vmdk_write:
2012  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
2013  *                if possible, otherwise return -ENOTSUP.
2014  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
2015  *                with each cluster. By dry run we can find if the zero write
2016  *                is possible without modifying image data.
2017  *
2018  * Returns: error code with 0 for success.
2019  */
2020 static int coroutine_fn vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
2021                                      uint64_t bytes, QEMUIOVector *qiov,
2022                                      bool zeroed, bool zero_dry_run)
2023 {
2024     BDRVVmdkState *s = bs->opaque;
2025     VmdkExtent *extent = NULL;
2026     int ret;
2027     int64_t offset_in_cluster, n_bytes;
2028     uint64_t cluster_offset;
2029     uint64_t bytes_done = 0;
2030     VmdkMetaData m_data;
2031 
2032     if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
2033         error_report("Wrong offset: offset=0x%" PRIx64
2034                      " total_sectors=0x%" PRIx64,
2035                      offset, bs->total_sectors);
2036         return -EIO;
2037     }
2038 
2039     while (bytes > 0) {
2040         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
2041         if (!extent) {
2042             return -EIO;
2043         }
2044         if (extent->sesparse) {
2045             return -ENOTSUP;
2046         }
2047         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
2048         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
2049                              - offset_in_cluster);
2050 
2051         ret = get_cluster_offset(bs, extent, &m_data, offset,
2052                                  !(extent->compressed || zeroed),
2053                                  &cluster_offset, offset_in_cluster,
2054                                  offset_in_cluster + n_bytes);
2055         if (extent->compressed) {
2056             if (ret == VMDK_OK) {
2057                 /* Refuse write to allocated cluster for streamOptimized */
2058                 error_report("Could not write to allocated cluster"
2059                               " for streamOptimized");
2060                 return -EIO;
2061             } else if (!zeroed) {
2062                 /* allocate */
2063                 ret = get_cluster_offset(bs, extent, &m_data, offset,
2064                                          true, &cluster_offset, 0, 0);
2065             }
2066         }
2067         if (ret == VMDK_ERROR) {
2068             return -EINVAL;
2069         }
2070         if (zeroed) {
2071             /* Do zeroed write, buf is ignored */
2072             if (extent->has_zero_grain &&
2073                     offset_in_cluster == 0 &&
2074                     n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
2075                 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
2076                 if (!zero_dry_run && ret != VMDK_ZEROED) {
2077                     /* update L2 tables */
2078                     if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
2079                             != VMDK_OK) {
2080                         return -EIO;
2081                     }
2082                 }
2083             } else {
2084                 return -ENOTSUP;
2085             }
2086         } else {
2087             ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
2088                                     qiov, bytes_done, n_bytes, offset);
2089             if (ret) {
2090                 return ret;
2091             }
2092             if (m_data.new_allocation) {
2093                 /* update L2 tables */
2094                 if (vmdk_L2update(extent, &m_data,
2095                                   cluster_offset >> BDRV_SECTOR_BITS)
2096                         != VMDK_OK) {
2097                     return -EIO;
2098                 }
2099             }
2100         }
2101         bytes -= n_bytes;
2102         offset += n_bytes;
2103         bytes_done += n_bytes;
2104 
2105         /* update CID on the first write every time the virtual disk is
2106          * opened */
2107         if (!s->cid_updated) {
2108             ret = vmdk_write_cid(bs, g_random_int());
2109             if (ret < 0) {
2110                 return ret;
2111             }
2112             s->cid_updated = true;
2113         }
2114     }
2115     return 0;
2116 }
2117 
2118 static int coroutine_fn
2119 vmdk_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
2120                 QEMUIOVector *qiov, BdrvRequestFlags flags)
2121 {
2122     int ret;
2123     BDRVVmdkState *s = bs->opaque;
2124     qemu_co_mutex_lock(&s->lock);
2125     ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
2126     qemu_co_mutex_unlock(&s->lock);
2127     return ret;
2128 }
2129 
2130 static int coroutine_fn
2131 vmdk_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes,
2132                            QEMUIOVector *qiov)
2133 {
2134     if (bytes == 0) {
2135         /* The caller will write bytes 0 to signal EOF.
2136          * When receive it, we align EOF to a sector boundary. */
2137         BDRVVmdkState *s = bs->opaque;
2138         int i, ret;
2139         int64_t length;
2140 
2141         for (i = 0; i < s->num_extents; i++) {
2142             length = bdrv_getlength(s->extents[i].file->bs);
2143             if (length < 0) {
2144                 return length;
2145             }
2146             length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
2147             ret = bdrv_truncate(s->extents[i].file, length, false,
2148                                 PREALLOC_MODE_OFF, 0, NULL);
2149             if (ret < 0) {
2150                 return ret;
2151             }
2152         }
2153         return 0;
2154     }
2155     return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
2156 }
2157 
2158 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
2159                                               int64_t offset,
2160                                               int64_t bytes,
2161                                               BdrvRequestFlags flags)
2162 {
2163     int ret;
2164     BDRVVmdkState *s = bs->opaque;
2165 
2166     qemu_co_mutex_lock(&s->lock);
2167     /* write zeroes could fail if sectors not aligned to cluster, test it with
2168      * dry_run == true before really updating image */
2169     ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
2170     if (!ret) {
2171         ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
2172     }
2173     qemu_co_mutex_unlock(&s->lock);
2174     return ret;
2175 }
2176 
2177 static int vmdk_init_extent(BlockBackend *blk,
2178                             int64_t filesize, bool flat,
2179                             bool compress, bool zeroed_grain,
2180                             Error **errp)
2181 {
2182     int ret, i;
2183     VMDK4Header header;
2184     uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
2185     uint32_t *gd_buf = NULL;
2186     int gd_buf_size;
2187 
2188     if (flat) {
2189         ret = blk_truncate(blk, filesize, false, PREALLOC_MODE_OFF, 0, errp);
2190         goto exit;
2191     }
2192     magic = cpu_to_be32(VMDK4_MAGIC);
2193     memset(&header, 0, sizeof(header));
2194     if (compress) {
2195         header.version = 3;
2196     } else if (zeroed_grain) {
2197         header.version = 2;
2198     } else {
2199         header.version = 1;
2200     }
2201     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
2202                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
2203                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
2204     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
2205     header.capacity = filesize / BDRV_SECTOR_SIZE;
2206     header.granularity = 128;
2207     header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
2208 
2209     grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
2210     gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
2211                            BDRV_SECTOR_SIZE);
2212     gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
2213     gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
2214 
2215     header.desc_offset = 1;
2216     header.desc_size = 20;
2217     header.rgd_offset = header.desc_offset + header.desc_size;
2218     header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
2219     header.grain_offset =
2220         ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
2221                  header.granularity);
2222     /* swap endianness for all header fields */
2223     header.version = cpu_to_le32(header.version);
2224     header.flags = cpu_to_le32(header.flags);
2225     header.capacity = cpu_to_le64(header.capacity);
2226     header.granularity = cpu_to_le64(header.granularity);
2227     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
2228     header.desc_offset = cpu_to_le64(header.desc_offset);
2229     header.desc_size = cpu_to_le64(header.desc_size);
2230     header.rgd_offset = cpu_to_le64(header.rgd_offset);
2231     header.gd_offset = cpu_to_le64(header.gd_offset);
2232     header.grain_offset = cpu_to_le64(header.grain_offset);
2233     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
2234 
2235     header.check_bytes[0] = 0xa;
2236     header.check_bytes[1] = 0x20;
2237     header.check_bytes[2] = 0xd;
2238     header.check_bytes[3] = 0xa;
2239 
2240     /* write all the data */
2241     ret = blk_pwrite(blk, 0, sizeof(magic), &magic, 0);
2242     if (ret < 0) {
2243         error_setg(errp, QERR_IO_ERROR);
2244         goto exit;
2245     }
2246     ret = blk_pwrite(blk, sizeof(magic), sizeof(header), &header, 0);
2247     if (ret < 0) {
2248         error_setg(errp, QERR_IO_ERROR);
2249         goto exit;
2250     }
2251 
2252     ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false,
2253                        PREALLOC_MODE_OFF, 0, errp);
2254     if (ret < 0) {
2255         goto exit;
2256     }
2257 
2258     /* write grain directory */
2259     gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
2260     gd_buf = g_malloc0(gd_buf_size);
2261     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
2262          i < gt_count; i++, tmp += gt_size) {
2263         gd_buf[i] = cpu_to_le32(tmp);
2264     }
2265     ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
2266                      gd_buf_size, gd_buf, 0);
2267     if (ret < 0) {
2268         error_setg(errp, QERR_IO_ERROR);
2269         goto exit;
2270     }
2271 
2272     /* write backup grain directory */
2273     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
2274          i < gt_count; i++, tmp += gt_size) {
2275         gd_buf[i] = cpu_to_le32(tmp);
2276     }
2277     ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
2278                      gd_buf_size, gd_buf, 0);
2279     if (ret < 0) {
2280         error_setg(errp, QERR_IO_ERROR);
2281     }
2282 
2283     ret = 0;
2284 exit:
2285     g_free(gd_buf);
2286     return ret;
2287 }
2288 
2289 static int vmdk_create_extent(const char *filename, int64_t filesize,
2290                               bool flat, bool compress, bool zeroed_grain,
2291                               BlockBackend **pbb,
2292                               QemuOpts *opts, Error **errp)
2293 {
2294     int ret;
2295     BlockBackend *blk = NULL;
2296 
2297     ret = bdrv_create_file(filename, opts, errp);
2298     if (ret < 0) {
2299         goto exit;
2300     }
2301 
2302     blk = blk_new_open(filename, NULL, NULL,
2303                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2304                        errp);
2305     if (blk == NULL) {
2306         ret = -EIO;
2307         goto exit;
2308     }
2309 
2310     blk_set_allow_write_beyond_eof(blk, true);
2311 
2312     ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
2313 exit:
2314     if (blk) {
2315         if (pbb) {
2316             *pbb = blk;
2317         } else {
2318             blk_unref(blk);
2319             blk = NULL;
2320         }
2321     }
2322     return ret;
2323 }
2324 
2325 static int filename_decompose(const char *filename, char *path, char *prefix,
2326                               char *postfix, size_t buf_len, Error **errp)
2327 {
2328     const char *p, *q;
2329 
2330     if (filename == NULL || !strlen(filename)) {
2331         error_setg(errp, "No filename provided");
2332         return VMDK_ERROR;
2333     }
2334     p = strrchr(filename, '/');
2335     if (p == NULL) {
2336         p = strrchr(filename, '\\');
2337     }
2338     if (p == NULL) {
2339         p = strrchr(filename, ':');
2340     }
2341     if (p != NULL) {
2342         p++;
2343         if (p - filename >= buf_len) {
2344             return VMDK_ERROR;
2345         }
2346         pstrcpy(path, p - filename + 1, filename);
2347     } else {
2348         p = filename;
2349         path[0] = '\0';
2350     }
2351     q = strrchr(p, '.');
2352     if (q == NULL) {
2353         pstrcpy(prefix, buf_len, p);
2354         postfix[0] = '\0';
2355     } else {
2356         if (q - p >= buf_len) {
2357             return VMDK_ERROR;
2358         }
2359         pstrcpy(prefix, q - p + 1, p);
2360         pstrcpy(postfix, buf_len, q);
2361     }
2362     return VMDK_OK;
2363 }
2364 
2365 /*
2366  * idx == 0: get or create the descriptor file (also the image file if in a
2367  *           non-split format.
2368  * idx >= 1: get the n-th extent if in a split subformat
2369  */
2370 typedef BlockBackend *(*vmdk_create_extent_fn)(int64_t size,
2371                                                int idx,
2372                                                bool flat,
2373                                                bool split,
2374                                                bool compress,
2375                                                bool zeroed_grain,
2376                                                void *opaque,
2377                                                Error **errp);
2378 
2379 static void vmdk_desc_add_extent(GString *desc,
2380                                  const char *extent_line_fmt,
2381                                  int64_t size, const char *filename)
2382 {
2383     char *basename = g_path_get_basename(filename);
2384 
2385     g_string_append_printf(desc, extent_line_fmt,
2386                            DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
2387     g_free(basename);
2388 }
2389 
2390 static int coroutine_fn vmdk_co_do_create(int64_t size,
2391                                           BlockdevVmdkSubformat subformat,
2392                                           BlockdevVmdkAdapterType adapter_type,
2393                                           const char *backing_file,
2394                                           const char *hw_version,
2395                                           const char *toolsversion,
2396                                           bool compat6,
2397                                           bool zeroed_grain,
2398                                           vmdk_create_extent_fn extent_fn,
2399                                           void *opaque,
2400                                           Error **errp)
2401 {
2402     int extent_idx;
2403     BlockBackend *blk = NULL;
2404     BlockBackend *extent_blk;
2405     Error *local_err = NULL;
2406     char *desc = NULL;
2407     int ret = 0;
2408     bool flat, split, compress;
2409     GString *ext_desc_lines;
2410     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
2411     int64_t extent_size;
2412     int64_t created_size = 0;
2413     const char *extent_line_fmt;
2414     char *parent_desc_line = g_malloc0(BUF_SIZE);
2415     uint32_t parent_cid = 0xffffffff;
2416     uint32_t number_heads = 16;
2417     uint32_t desc_offset = 0, desc_len;
2418     const char desc_template[] =
2419         "# Disk DescriptorFile\n"
2420         "version=1\n"
2421         "CID=%" PRIx32 "\n"
2422         "parentCID=%" PRIx32 "\n"
2423         "createType=\"%s\"\n"
2424         "%s"
2425         "\n"
2426         "# Extent description\n"
2427         "%s"
2428         "\n"
2429         "# The Disk Data Base\n"
2430         "#DDB\n"
2431         "\n"
2432         "ddb.virtualHWVersion = \"%s\"\n"
2433         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2434         "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2435         "ddb.geometry.sectors = \"63\"\n"
2436         "ddb.adapterType = \"%s\"\n"
2437         "ddb.toolsVersion = \"%s\"\n";
2438 
2439     ext_desc_lines = g_string_new(NULL);
2440 
2441     /* Read out options */
2442     if (compat6) {
2443         if (hw_version) {
2444             error_setg(errp,
2445                        "compat6 cannot be enabled with hwversion set");
2446             ret = -EINVAL;
2447             goto exit;
2448         }
2449         hw_version = "6";
2450     }
2451     if (!hw_version) {
2452         hw_version = "4";
2453     }
2454     if (!toolsversion) {
2455         toolsversion = "2147483647";
2456     }
2457 
2458     if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2459         /* that's the number of heads with which vmware operates when
2460            creating, exporting, etc. vmdk files with a non-ide adapter type */
2461         number_heads = 255;
2462     }
2463     split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2464             (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2465     flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2466            (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2467     compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2468 
2469     if (flat) {
2470         extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2471     } else {
2472         extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2473     }
2474     if (flat && backing_file) {
2475         error_setg(errp, "Flat image can't have backing file");
2476         ret = -ENOTSUP;
2477         goto exit;
2478     }
2479     if (flat && zeroed_grain) {
2480         error_setg(errp, "Flat image can't enable zeroed grain");
2481         ret = -ENOTSUP;
2482         goto exit;
2483     }
2484 
2485     /* Create extents */
2486     if (split) {
2487         extent_size = split_size;
2488     } else {
2489         extent_size = size;
2490     }
2491     if (!split && !flat) {
2492         created_size = extent_size;
2493     } else {
2494         created_size = 0;
2495     }
2496     /* Get the descriptor file BDS */
2497     blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2498                     opaque, errp);
2499     if (!blk) {
2500         ret = -EIO;
2501         goto exit;
2502     }
2503     if (!split && !flat) {
2504         vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2505                              blk_bs(blk)->filename);
2506     }
2507 
2508     if (backing_file) {
2509         BlockBackend *backing;
2510         char *full_backing =
2511             bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2512                                                          backing_file,
2513                                                          &local_err);
2514         if (local_err) {
2515             error_propagate(errp, local_err);
2516             ret = -ENOENT;
2517             goto exit;
2518         }
2519         assert(full_backing);
2520 
2521         backing = blk_new_open(full_backing, NULL, NULL,
2522                                BDRV_O_NO_BACKING, errp);
2523         g_free(full_backing);
2524         if (backing == NULL) {
2525             ret = -EIO;
2526             goto exit;
2527         }
2528         if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2529             error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2530                        blk_bs(backing)->drv->format_name);
2531             blk_unref(backing);
2532             ret = -EINVAL;
2533             goto exit;
2534         }
2535         ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2536         blk_unref(backing);
2537         if (ret) {
2538             error_setg(errp, "Failed to read parent CID");
2539             goto exit;
2540         }
2541         snprintf(parent_desc_line, BUF_SIZE,
2542                 "parentFileNameHint=\"%s\"", backing_file);
2543     }
2544     extent_idx = 1;
2545     while (created_size < size) {
2546         int64_t cur_size = MIN(size - created_size, extent_size);
2547         extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2548                                zeroed_grain, opaque, errp);
2549         if (!extent_blk) {
2550             ret = -EINVAL;
2551             goto exit;
2552         }
2553         vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2554                              blk_bs(extent_blk)->filename);
2555         created_size += cur_size;
2556         extent_idx++;
2557         blk_unref(extent_blk);
2558     }
2559 
2560     /* Check whether we got excess extents */
2561     extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2562                            opaque, NULL);
2563     if (extent_blk) {
2564         blk_unref(extent_blk);
2565         error_setg(errp, "List of extents contains unused extents");
2566         ret = -EINVAL;
2567         goto exit;
2568     }
2569 
2570     /* generate descriptor file */
2571     desc = g_strdup_printf(desc_template,
2572                            g_random_int(),
2573                            parent_cid,
2574                            BlockdevVmdkSubformat_str(subformat),
2575                            parent_desc_line,
2576                            ext_desc_lines->str,
2577                            hw_version,
2578                            size /
2579                                (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2580                            number_heads,
2581                            BlockdevVmdkAdapterType_str(adapter_type),
2582                            toolsversion);
2583     desc_len = strlen(desc);
2584     /* the descriptor offset = 0x200 */
2585     if (!split && !flat) {
2586         desc_offset = 0x200;
2587     }
2588 
2589     ret = blk_pwrite(blk, desc_offset, desc_len, desc, 0);
2590     if (ret < 0) {
2591         error_setg_errno(errp, -ret, "Could not write description");
2592         goto exit;
2593     }
2594     /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2595      * for description file */
2596     if (desc_offset == 0) {
2597         ret = blk_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, 0, errp);
2598         if (ret < 0) {
2599             goto exit;
2600         }
2601     }
2602     ret = 0;
2603 exit:
2604     if (blk) {
2605         blk_unref(blk);
2606     }
2607     g_free(desc);
2608     g_free(parent_desc_line);
2609     g_string_free(ext_desc_lines, true);
2610     return ret;
2611 }
2612 
2613 typedef struct {
2614     char *path;
2615     char *prefix;
2616     char *postfix;
2617     QemuOpts *opts;
2618 } VMDKCreateOptsData;
2619 
2620 static BlockBackend *vmdk_co_create_opts_cb(int64_t size, int idx,
2621                                             bool flat, bool split, bool compress,
2622                                             bool zeroed_grain, void *opaque,
2623                                             Error **errp)
2624 {
2625     BlockBackend *blk = NULL;
2626     BlockDriverState *bs = NULL;
2627     VMDKCreateOptsData *data = opaque;
2628     char *ext_filename = NULL;
2629     char *rel_filename = NULL;
2630 
2631     /* We're done, don't create excess extents. */
2632     if (size == -1) {
2633         assert(errp == NULL);
2634         return NULL;
2635     }
2636 
2637     if (idx == 0) {
2638         rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2639     } else if (split) {
2640         rel_filename = g_strdup_printf("%s-%c%03d%s",
2641                                        data->prefix,
2642                                        flat ? 'f' : 's', idx, data->postfix);
2643     } else {
2644         assert(idx == 1);
2645         rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2646     }
2647 
2648     ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2649     g_free(rel_filename);
2650 
2651     if (vmdk_create_extent(ext_filename, size,
2652                            flat, compress, zeroed_grain, &blk, data->opts,
2653                            errp)) {
2654         goto exit;
2655     }
2656     bdrv_unref(bs);
2657 exit:
2658     g_free(ext_filename);
2659     return blk;
2660 }
2661 
2662 static int coroutine_fn vmdk_co_create_opts(BlockDriver *drv,
2663                                             const char *filename,
2664                                             QemuOpts *opts,
2665                                             Error **errp)
2666 {
2667     Error *local_err = NULL;
2668     char *desc = NULL;
2669     int64_t total_size = 0;
2670     char *adapter_type = NULL;
2671     BlockdevVmdkAdapterType adapter_type_enum;
2672     char *backing_file = NULL;
2673     char *hw_version = NULL;
2674     char *toolsversion = NULL;
2675     char *fmt = NULL;
2676     BlockdevVmdkSubformat subformat;
2677     int ret = 0;
2678     char *path = g_malloc0(PATH_MAX);
2679     char *prefix = g_malloc0(PATH_MAX);
2680     char *postfix = g_malloc0(PATH_MAX);
2681     char *desc_line = g_malloc0(BUF_SIZE);
2682     char *ext_filename = g_malloc0(PATH_MAX);
2683     char *desc_filename = g_malloc0(PATH_MAX);
2684     char *parent_desc_line = g_malloc0(BUF_SIZE);
2685     bool zeroed_grain;
2686     bool compat6;
2687     VMDKCreateOptsData data;
2688     char *backing_fmt = NULL;
2689 
2690     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2691     if (backing_fmt && strcmp(backing_fmt, "vmdk") != 0) {
2692         error_setg(errp, "backing_file must be a vmdk image");
2693         ret = -EINVAL;
2694         goto exit;
2695     }
2696 
2697     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
2698         ret = -EINVAL;
2699         goto exit;
2700     }
2701     /* Read out options */
2702     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2703                           BDRV_SECTOR_SIZE);
2704     adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
2705     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2706     hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
2707     toolsversion = qemu_opt_get_del(opts, BLOCK_OPT_TOOLSVERSION);
2708     compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false);
2709     if (strcmp(hw_version, "undefined") == 0) {
2710         g_free(hw_version);
2711         hw_version = NULL;
2712     }
2713     fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
2714     zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false);
2715 
2716     if (adapter_type) {
2717         adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup,
2718                                             adapter_type,
2719                                             BLOCKDEV_VMDK_ADAPTER_TYPE_IDE,
2720                                             &local_err);
2721         if (local_err) {
2722             error_propagate(errp, local_err);
2723             ret = -EINVAL;
2724             goto exit;
2725         }
2726     } else {
2727         adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE;
2728     }
2729 
2730     if (!fmt) {
2731         /* Default format to monolithicSparse */
2732         subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE;
2733     } else {
2734         subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup,
2735                                     fmt,
2736                                     BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE,
2737                                     &local_err);
2738         if (local_err) {
2739             error_propagate(errp, local_err);
2740             ret = -EINVAL;
2741             goto exit;
2742         }
2743     }
2744     data = (VMDKCreateOptsData){
2745         .prefix = prefix,
2746         .postfix = postfix,
2747         .path = path,
2748         .opts = opts,
2749     };
2750     ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum,
2751                             backing_file, hw_version, toolsversion, compat6,
2752                             zeroed_grain, vmdk_co_create_opts_cb, &data, errp);
2753 
2754 exit:
2755     g_free(backing_fmt);
2756     g_free(adapter_type);
2757     g_free(backing_file);
2758     g_free(hw_version);
2759     g_free(toolsversion);
2760     g_free(fmt);
2761     g_free(desc);
2762     g_free(path);
2763     g_free(prefix);
2764     g_free(postfix);
2765     g_free(desc_line);
2766     g_free(ext_filename);
2767     g_free(desc_filename);
2768     g_free(parent_desc_line);
2769     return ret;
2770 }
2771 
2772 static BlockBackend *vmdk_co_create_cb(int64_t size, int idx,
2773                                        bool flat, bool split, bool compress,
2774                                        bool zeroed_grain, void *opaque,
2775                                        Error **errp)
2776 {
2777     int ret;
2778     BlockDriverState *bs;
2779     BlockBackend *blk;
2780     BlockdevCreateOptionsVmdk *opts = opaque;
2781 
2782     if (idx == 0) {
2783         bs = bdrv_open_blockdev_ref(opts->file, errp);
2784     } else {
2785         int i;
2786         BlockdevRefList *list = opts->extents;
2787         for (i = 1; i < idx; i++) {
2788             if (!list || !list->next) {
2789                 error_setg(errp, "Extent [%d] not specified", i);
2790                 return NULL;
2791             }
2792             list = list->next;
2793         }
2794         if (!list) {
2795             error_setg(errp, "Extent [%d] not specified", idx - 1);
2796             return NULL;
2797         }
2798         bs = bdrv_open_blockdev_ref(list->value, errp);
2799     }
2800     if (!bs) {
2801         return NULL;
2802     }
2803     blk = blk_new_with_bs(bs,
2804                           BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | BLK_PERM_RESIZE,
2805                           BLK_PERM_ALL, errp);
2806     if (!blk) {
2807         return NULL;
2808     }
2809     blk_set_allow_write_beyond_eof(blk, true);
2810     bdrv_unref(bs);
2811 
2812     if (size != -1) {
2813         ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2814         if (ret) {
2815             blk_unref(blk);
2816             blk = NULL;
2817         }
2818     }
2819     return blk;
2820 }
2821 
2822 static int coroutine_fn vmdk_co_create(BlockdevCreateOptions *create_options,
2823                                        Error **errp)
2824 {
2825     int ret;
2826     BlockdevCreateOptionsVmdk *opts;
2827 
2828     opts = &create_options->u.vmdk;
2829 
2830     /* Validate options */
2831     if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2832         error_setg(errp, "Image size must be a multiple of 512 bytes");
2833         ret = -EINVAL;
2834         goto out;
2835     }
2836 
2837     ret = vmdk_co_do_create(opts->size,
2838                             opts->subformat,
2839                             opts->adapter_type,
2840                             opts->backing_file,
2841                             opts->hwversion,
2842                             opts->toolsversion,
2843                             false,
2844                             opts->zeroed_grain,
2845                             vmdk_co_create_cb,
2846                             opts, errp);
2847     return ret;
2848 
2849 out:
2850     return ret;
2851 }
2852 
2853 static void vmdk_close(BlockDriverState *bs)
2854 {
2855     BDRVVmdkState *s = bs->opaque;
2856 
2857     vmdk_free_extents(bs);
2858     g_free(s->create_type);
2859 
2860     migrate_del_blocker(s->migration_blocker);
2861     error_free(s->migration_blocker);
2862 }
2863 
2864 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2865 {
2866     int i;
2867     int64_t ret = 0;
2868     int64_t r;
2869     BDRVVmdkState *s = bs->opaque;
2870 
2871     ret = bdrv_get_allocated_file_size(bs->file->bs);
2872     if (ret < 0) {
2873         return ret;
2874     }
2875     for (i = 0; i < s->num_extents; i++) {
2876         if (s->extents[i].file == bs->file) {
2877             continue;
2878         }
2879         r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2880         if (r < 0) {
2881             return r;
2882         }
2883         ret += r;
2884     }
2885     return ret;
2886 }
2887 
2888 static int vmdk_has_zero_init(BlockDriverState *bs)
2889 {
2890     int i;
2891     BDRVVmdkState *s = bs->opaque;
2892 
2893     /* If has a flat extent and its underlying storage doesn't have zero init,
2894      * return 0. */
2895     for (i = 0; i < s->num_extents; i++) {
2896         if (s->extents[i].flat) {
2897             if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2898                 return 0;
2899             }
2900         }
2901     }
2902     return 1;
2903 }
2904 
2905 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2906 {
2907     ImageInfo *info = g_new0(ImageInfo, 1);
2908 
2909     bdrv_refresh_filename(extent->file->bs);
2910     *info = (ImageInfo){
2911         .filename         = g_strdup(extent->file->bs->filename),
2912         .format           = g_strdup(extent->type),
2913         .virtual_size     = extent->sectors * BDRV_SECTOR_SIZE,
2914         .compressed       = extent->compressed,
2915         .has_compressed   = extent->compressed,
2916         .cluster_size     = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2917         .has_cluster_size = !extent->flat,
2918     };
2919 
2920     return info;
2921 }
2922 
2923 static int coroutine_fn vmdk_co_check(BlockDriverState *bs,
2924                                       BdrvCheckResult *result,
2925                                       BdrvCheckMode fix)
2926 {
2927     BDRVVmdkState *s = bs->opaque;
2928     VmdkExtent *extent = NULL;
2929     int64_t sector_num = 0;
2930     int64_t total_sectors = bdrv_nb_sectors(bs);
2931     int ret;
2932     uint64_t cluster_offset;
2933 
2934     if (fix) {
2935         return -ENOTSUP;
2936     }
2937 
2938     for (;;) {
2939         if (sector_num >= total_sectors) {
2940             return 0;
2941         }
2942         extent = find_extent(s, sector_num, extent);
2943         if (!extent) {
2944             fprintf(stderr,
2945                     "ERROR: could not find extent for sector %" PRId64 "\n",
2946                     sector_num);
2947             ret = -EINVAL;
2948             break;
2949         }
2950         ret = get_cluster_offset(bs, extent, NULL,
2951                                  sector_num << BDRV_SECTOR_BITS,
2952                                  false, &cluster_offset, 0, 0);
2953         if (ret == VMDK_ERROR) {
2954             fprintf(stderr,
2955                     "ERROR: could not get cluster_offset for sector %"
2956                     PRId64 "\n", sector_num);
2957             break;
2958         }
2959         if (ret == VMDK_OK) {
2960             int64_t extent_len = bdrv_getlength(extent->file->bs);
2961             if (extent_len < 0) {
2962                 fprintf(stderr,
2963                         "ERROR: could not get extent file length for sector %"
2964                         PRId64 "\n", sector_num);
2965                 ret = extent_len;
2966                 break;
2967             }
2968             if (cluster_offset >= extent_len) {
2969                 fprintf(stderr,
2970                         "ERROR: cluster offset for sector %"
2971                         PRId64 " points after EOF\n", sector_num);
2972                 ret = -EINVAL;
2973                 break;
2974             }
2975         }
2976         sector_num += extent->cluster_sectors;
2977     }
2978 
2979     result->corruptions++;
2980     return ret;
2981 }
2982 
2983 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs,
2984                                                  Error **errp)
2985 {
2986     int i;
2987     BDRVVmdkState *s = bs->opaque;
2988     ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2989     ImageInfoList **tail;
2990 
2991     *spec_info = (ImageInfoSpecific){
2992         .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2993         .u = {
2994             .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2995         },
2996     };
2997 
2998     *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2999         .create_type = g_strdup(s->create_type),
3000         .cid = s->cid,
3001         .parent_cid = s->parent_cid,
3002     };
3003 
3004     tail = &spec_info->u.vmdk.data->extents;
3005     for (i = 0; i < s->num_extents; i++) {
3006         QAPI_LIST_APPEND(tail, vmdk_get_extent_info(&s->extents[i]));
3007     }
3008 
3009     return spec_info;
3010 }
3011 
3012 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
3013 {
3014     return a->flat == b->flat &&
3015            a->compressed == b->compressed &&
3016            (a->flat || a->cluster_sectors == b->cluster_sectors);
3017 }
3018 
3019 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3020 {
3021     int i;
3022     BDRVVmdkState *s = bs->opaque;
3023     assert(s->num_extents);
3024 
3025     /* See if we have multiple extents but they have different cases */
3026     for (i = 1; i < s->num_extents; i++) {
3027         if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
3028             return -ENOTSUP;
3029         }
3030     }
3031     bdi->needs_compressed_writes = s->extents[0].compressed;
3032     if (!s->extents[0].flat) {
3033         bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
3034     }
3035     return 0;
3036 }
3037 
3038 static void vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
3039                                       bool backing_overridden)
3040 {
3041     /* No children but file and backing can be explicitly specified (TODO) */
3042     qdict_put(target, "file",
3043               qobject_ref(bs->file->bs->full_open_options));
3044 
3045     if (backing_overridden) {
3046         if (bs->backing) {
3047             qdict_put(target, "backing",
3048                       qobject_ref(bs->backing->bs->full_open_options));
3049         } else {
3050             qdict_put_null(target, "backing");
3051         }
3052     }
3053 }
3054 
3055 static QemuOptsList vmdk_create_opts = {
3056     .name = "vmdk-create-opts",
3057     .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
3058     .desc = {
3059         {
3060             .name = BLOCK_OPT_SIZE,
3061             .type = QEMU_OPT_SIZE,
3062             .help = "Virtual disk size"
3063         },
3064         {
3065             .name = BLOCK_OPT_ADAPTER_TYPE,
3066             .type = QEMU_OPT_STRING,
3067             .help = "Virtual adapter type, can be one of "
3068                     "ide (default), lsilogic, buslogic or legacyESX"
3069         },
3070         {
3071             .name = BLOCK_OPT_BACKING_FILE,
3072             .type = QEMU_OPT_STRING,
3073             .help = "File name of a base image"
3074         },
3075         {
3076             .name = BLOCK_OPT_BACKING_FMT,
3077             .type = QEMU_OPT_STRING,
3078             .help = "Must be 'vmdk' if present",
3079         },
3080         {
3081             .name = BLOCK_OPT_COMPAT6,
3082             .type = QEMU_OPT_BOOL,
3083             .help = "VMDK version 6 image",
3084             .def_value_str = "off"
3085         },
3086         {
3087             .name = BLOCK_OPT_HWVERSION,
3088             .type = QEMU_OPT_STRING,
3089             .help = "VMDK hardware version",
3090             .def_value_str = "undefined"
3091         },
3092         {
3093             .name = BLOCK_OPT_TOOLSVERSION,
3094             .type = QEMU_OPT_STRING,
3095             .help = "VMware guest tools version",
3096         },
3097         {
3098             .name = BLOCK_OPT_SUBFMT,
3099             .type = QEMU_OPT_STRING,
3100             .help =
3101                 "VMDK flat extent format, can be one of "
3102                 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
3103         },
3104         {
3105             .name = BLOCK_OPT_ZEROED_GRAIN,
3106             .type = QEMU_OPT_BOOL,
3107             .help = "Enable efficient zero writes "
3108                     "using the zeroed-grain GTE feature"
3109         },
3110         { /* end of list */ }
3111     }
3112 };
3113 
3114 static BlockDriver bdrv_vmdk = {
3115     .format_name                  = "vmdk",
3116     .instance_size                = sizeof(BDRVVmdkState),
3117     .bdrv_probe                   = vmdk_probe,
3118     .bdrv_open                    = vmdk_open,
3119     .bdrv_co_check                = vmdk_co_check,
3120     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
3121     .bdrv_reopen_commit           = vmdk_reopen_commit,
3122     .bdrv_reopen_abort            = vmdk_reopen_abort,
3123     .bdrv_child_perm              = bdrv_default_perms,
3124     .bdrv_co_preadv               = vmdk_co_preadv,
3125     .bdrv_co_pwritev              = vmdk_co_pwritev,
3126     .bdrv_co_pwritev_compressed   = vmdk_co_pwritev_compressed,
3127     .bdrv_co_pwrite_zeroes        = vmdk_co_pwrite_zeroes,
3128     .bdrv_close                   = vmdk_close,
3129     .bdrv_co_create_opts          = vmdk_co_create_opts,
3130     .bdrv_co_create               = vmdk_co_create,
3131     .bdrv_co_block_status         = vmdk_co_block_status,
3132     .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
3133     .bdrv_has_zero_init           = vmdk_has_zero_init,
3134     .bdrv_get_specific_info       = vmdk_get_specific_info,
3135     .bdrv_refresh_limits          = vmdk_refresh_limits,
3136     .bdrv_get_info                = vmdk_get_info,
3137     .bdrv_gather_child_options    = vmdk_gather_child_options,
3138 
3139     .is_format                    = true,
3140     .supports_backing             = true,
3141     .create_opts                  = &vmdk_create_opts,
3142 };
3143 
3144 static void bdrv_vmdk_init(void)
3145 {
3146     bdrv_register(&bdrv_vmdk);
3147 }
3148 
3149 block_init(bdrv_vmdk_init);
3150