xref: /openbmc/qemu/block/vmdk.c (revision b0e70950)
1 /*
2  * Block driver for the VMDK format
3  *
4  * Copyright (c) 2004 Fabrice Bellard
5  * Copyright (c) 2005 Filip Navara
6  *
7  * Permission is hereby granted, free of charge, to any person obtaining a copy
8  * of this software and associated documentation files (the "Software"), to deal
9  * in the Software without restriction, including without limitation the rights
10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11  * copies of the Software, and to permit persons to whom the Software is
12  * furnished to do so, subject to the following conditions:
13  *
14  * The above copyright notice and this permission notice shall be included in
15  * all copies or substantial portions of the Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23  * THE SOFTWARE.
24  */
25 
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "block/block_int.h"
29 #include "sysemu/block-backend.h"
30 #include "qapi/qmp/qdict.h"
31 #include "qapi/qmp/qerror.h"
32 #include "qemu/error-report.h"
33 #include "qemu/module.h"
34 #include "qemu/option.h"
35 #include "qemu/bswap.h"
36 #include "migration/blocker.h"
37 #include "qemu/cutils.h"
38 #include <zlib.h>
39 
40 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
41 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
42 #define VMDK4_COMPRESSION_DEFLATE 1
43 #define VMDK4_FLAG_NL_DETECT (1 << 0)
44 #define VMDK4_FLAG_RGD (1 << 1)
45 /* Zeroed-grain enable bit */
46 #define VMDK4_FLAG_ZERO_GRAIN   (1 << 2)
47 #define VMDK4_FLAG_COMPRESS (1 << 16)
48 #define VMDK4_FLAG_MARKER (1 << 17)
49 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
50 
51 #define VMDK_EXTENT_MAX_SECTORS (1ULL << 32)
52 
53 #define VMDK_GTE_ZEROED 0x1
54 
55 /* VMDK internal error codes */
56 #define VMDK_OK      0
57 #define VMDK_ERROR   (-1)
58 /* Cluster not allocated */
59 #define VMDK_UNALLOC (-2)
60 #define VMDK_ZEROED  (-3)
61 
62 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
63 
64 typedef struct {
65     uint32_t version;
66     uint32_t flags;
67     uint32_t disk_sectors;
68     uint32_t granularity;
69     uint32_t l1dir_offset;
70     uint32_t l1dir_size;
71     uint32_t file_sectors;
72     uint32_t cylinders;
73     uint32_t heads;
74     uint32_t sectors_per_track;
75 } QEMU_PACKED VMDK3Header;
76 
77 typedef struct {
78     uint32_t version;
79     uint32_t flags;
80     uint64_t capacity;
81     uint64_t granularity;
82     uint64_t desc_offset;
83     uint64_t desc_size;
84     /* Number of GrainTableEntries per GrainTable */
85     uint32_t num_gtes_per_gt;
86     uint64_t rgd_offset;
87     uint64_t gd_offset;
88     uint64_t grain_offset;
89     char filler[1];
90     char check_bytes[4];
91     uint16_t compressAlgorithm;
92 } QEMU_PACKED VMDK4Header;
93 
94 typedef struct VMDKSESparseConstHeader {
95     uint64_t magic;
96     uint64_t version;
97     uint64_t capacity;
98     uint64_t grain_size;
99     uint64_t grain_table_size;
100     uint64_t flags;
101     uint64_t reserved1;
102     uint64_t reserved2;
103     uint64_t reserved3;
104     uint64_t reserved4;
105     uint64_t volatile_header_offset;
106     uint64_t volatile_header_size;
107     uint64_t journal_header_offset;
108     uint64_t journal_header_size;
109     uint64_t journal_offset;
110     uint64_t journal_size;
111     uint64_t grain_dir_offset;
112     uint64_t grain_dir_size;
113     uint64_t grain_tables_offset;
114     uint64_t grain_tables_size;
115     uint64_t free_bitmap_offset;
116     uint64_t free_bitmap_size;
117     uint64_t backmap_offset;
118     uint64_t backmap_size;
119     uint64_t grains_offset;
120     uint64_t grains_size;
121     uint8_t pad[304];
122 } QEMU_PACKED VMDKSESparseConstHeader;
123 
124 typedef struct VMDKSESparseVolatileHeader {
125     uint64_t magic;
126     uint64_t free_gt_number;
127     uint64_t next_txn_seq_number;
128     uint64_t replay_journal;
129     uint8_t pad[480];
130 } QEMU_PACKED VMDKSESparseVolatileHeader;
131 
132 #define L2_CACHE_SIZE 16
133 
134 typedef struct VmdkExtent {
135     BdrvChild *file;
136     bool flat;
137     bool compressed;
138     bool has_marker;
139     bool has_zero_grain;
140     bool sesparse;
141     uint64_t sesparse_l2_tables_offset;
142     uint64_t sesparse_clusters_offset;
143     int32_t entry_size;
144     int version;
145     int64_t sectors;
146     int64_t end_sector;
147     int64_t flat_start_offset;
148     int64_t l1_table_offset;
149     int64_t l1_backup_table_offset;
150     void *l1_table;
151     uint32_t *l1_backup_table;
152     unsigned int l1_size;
153     uint32_t l1_entry_sectors;
154 
155     unsigned int l2_size;
156     void *l2_cache;
157     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
158     uint32_t l2_cache_counts[L2_CACHE_SIZE];
159 
160     int64_t cluster_sectors;
161     int64_t next_cluster_sector;
162     char *type;
163 } VmdkExtent;
164 
165 typedef struct BDRVVmdkState {
166     CoMutex lock;
167     uint64_t desc_offset;
168     bool cid_updated;
169     bool cid_checked;
170     uint32_t cid;
171     uint32_t parent_cid;
172     int num_extents;
173     /* Extent array with num_extents entries, ascend ordered by address */
174     VmdkExtent *extents;
175     Error *migration_blocker;
176     char *create_type;
177 } BDRVVmdkState;
178 
179 typedef struct VmdkMetaData {
180     unsigned int l1_index;
181     unsigned int l2_index;
182     unsigned int l2_offset;
183     int valid;
184     uint32_t *l2_cache_entry;
185 } VmdkMetaData;
186 
187 typedef struct VmdkGrainMarker {
188     uint64_t lba;
189     uint32_t size;
190     uint8_t  data[];
191 } QEMU_PACKED VmdkGrainMarker;
192 
193 enum {
194     MARKER_END_OF_STREAM    = 0,
195     MARKER_GRAIN_TABLE      = 1,
196     MARKER_GRAIN_DIRECTORY  = 2,
197     MARKER_FOOTER           = 3,
198 };
199 
200 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
201 {
202     uint32_t magic;
203 
204     if (buf_size < 4) {
205         return 0;
206     }
207     magic = be32_to_cpu(*(uint32_t *)buf);
208     if (magic == VMDK3_MAGIC ||
209         magic == VMDK4_MAGIC) {
210         return 100;
211     } else {
212         const char *p = (const char *)buf;
213         const char *end = p + buf_size;
214         while (p < end) {
215             if (*p == '#') {
216                 /* skip comment line */
217                 while (p < end && *p != '\n') {
218                     p++;
219                 }
220                 p++;
221                 continue;
222             }
223             if (*p == ' ') {
224                 while (p < end && *p == ' ') {
225                     p++;
226                 }
227                 /* skip '\r' if windows line endings used. */
228                 if (p < end && *p == '\r') {
229                     p++;
230                 }
231                 /* only accept blank lines before 'version=' line */
232                 if (p == end || *p != '\n') {
233                     return 0;
234                 }
235                 p++;
236                 continue;
237             }
238             if (end - p >= strlen("version=X\n")) {
239                 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
240                     strncmp("version=2\n", p, strlen("version=2\n")) == 0 ||
241                     strncmp("version=3\n", p, strlen("version=3\n")) == 0) {
242                     return 100;
243                 }
244             }
245             if (end - p >= strlen("version=X\r\n")) {
246                 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
247                     strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0 ||
248                     strncmp("version=3\r\n", p, strlen("version=3\r\n")) == 0) {
249                     return 100;
250                 }
251             }
252             return 0;
253         }
254         return 0;
255     }
256 }
257 
258 #define SECTOR_SIZE 512
259 #define DESC_SIZE (20 * SECTOR_SIZE)    /* 20 sectors of 512 bytes each */
260 #define BUF_SIZE 4096
261 #define HEADER_SIZE 512                 /* first sector of 512 bytes */
262 
263 static void vmdk_free_extents(BlockDriverState *bs)
264 {
265     int i;
266     BDRVVmdkState *s = bs->opaque;
267     VmdkExtent *e;
268 
269     for (i = 0; i < s->num_extents; i++) {
270         e = &s->extents[i];
271         g_free(e->l1_table);
272         g_free(e->l2_cache);
273         g_free(e->l1_backup_table);
274         g_free(e->type);
275         if (e->file != bs->file) {
276             bdrv_unref_child(bs, e->file);
277         }
278     }
279     g_free(s->extents);
280 }
281 
282 static void vmdk_free_last_extent(BlockDriverState *bs)
283 {
284     BDRVVmdkState *s = bs->opaque;
285 
286     if (s->num_extents == 0) {
287         return;
288     }
289     s->num_extents--;
290     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
291 }
292 
293 /* Return -ve errno, or 0 on success and write CID into *pcid. */
294 static int vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
295 {
296     char *desc;
297     uint32_t cid;
298     const char *p_name, *cid_str;
299     size_t cid_str_size;
300     BDRVVmdkState *s = bs->opaque;
301     int ret;
302 
303     desc = g_malloc0(DESC_SIZE);
304     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
305     if (ret < 0) {
306         goto out;
307     }
308 
309     if (parent) {
310         cid_str = "parentCID";
311         cid_str_size = sizeof("parentCID");
312     } else {
313         cid_str = "CID";
314         cid_str_size = sizeof("CID");
315     }
316 
317     desc[DESC_SIZE - 1] = '\0';
318     p_name = strstr(desc, cid_str);
319     if (p_name == NULL) {
320         ret = -EINVAL;
321         goto out;
322     }
323     p_name += cid_str_size;
324     if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
325         ret = -EINVAL;
326         goto out;
327     }
328     *pcid = cid;
329     ret = 0;
330 
331 out:
332     g_free(desc);
333     return ret;
334 }
335 
336 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
337 {
338     char *desc, *tmp_desc;
339     char *p_name, *tmp_str;
340     BDRVVmdkState *s = bs->opaque;
341     int ret = 0;
342 
343     desc = g_malloc0(DESC_SIZE);
344     tmp_desc = g_malloc0(DESC_SIZE);
345     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
346     if (ret < 0) {
347         goto out;
348     }
349 
350     desc[DESC_SIZE - 1] = '\0';
351     tmp_str = strstr(desc, "parentCID");
352     if (tmp_str == NULL) {
353         ret = -EINVAL;
354         goto out;
355     }
356 
357     pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
358     p_name = strstr(desc, "CID");
359     if (p_name != NULL) {
360         p_name += sizeof("CID");
361         snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
362         pstrcat(desc, DESC_SIZE, tmp_desc);
363     }
364 
365     ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
366 
367 out:
368     g_free(desc);
369     g_free(tmp_desc);
370     return ret;
371 }
372 
373 static int vmdk_is_cid_valid(BlockDriverState *bs)
374 {
375     BDRVVmdkState *s = bs->opaque;
376     uint32_t cur_pcid;
377 
378     if (!s->cid_checked && bs->backing) {
379         BlockDriverState *p_bs = bs->backing->bs;
380 
381         if (strcmp(p_bs->drv->format_name, "vmdk")) {
382             /* Backing file is not in vmdk format, so it does not have
383              * a CID, which makes the overlay's parent CID invalid */
384             return 0;
385         }
386 
387         if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
388             /* read failure: report as not valid */
389             return 0;
390         }
391         if (s->parent_cid != cur_pcid) {
392             /* CID not valid */
393             return 0;
394         }
395     }
396     s->cid_checked = true;
397     /* CID valid */
398     return 1;
399 }
400 
401 /* We have nothing to do for VMDK reopen, stubs just return success */
402 static int vmdk_reopen_prepare(BDRVReopenState *state,
403                                BlockReopenQueue *queue, Error **errp)
404 {
405     assert(state != NULL);
406     assert(state->bs != NULL);
407     return 0;
408 }
409 
410 static int vmdk_parent_open(BlockDriverState *bs)
411 {
412     char *p_name;
413     char *desc;
414     BDRVVmdkState *s = bs->opaque;
415     int ret;
416 
417     desc = g_malloc0(DESC_SIZE + 1);
418     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
419     if (ret < 0) {
420         goto out;
421     }
422     ret = 0;
423 
424     p_name = strstr(desc, "parentFileNameHint");
425     if (p_name != NULL) {
426         char *end_name;
427 
428         p_name += sizeof("parentFileNameHint") + 1;
429         end_name = strchr(p_name, '\"');
430         if (end_name == NULL) {
431             ret = -EINVAL;
432             goto out;
433         }
434         if ((end_name - p_name) > sizeof(bs->auto_backing_file) - 1) {
435             ret = -EINVAL;
436             goto out;
437         }
438 
439         pstrcpy(bs->auto_backing_file, end_name - p_name + 1, p_name);
440         pstrcpy(bs->backing_file, sizeof(bs->backing_file),
441                 bs->auto_backing_file);
442         pstrcpy(bs->backing_format, sizeof(bs->backing_format),
443                 "vmdk");
444     }
445 
446 out:
447     g_free(desc);
448     return ret;
449 }
450 
451 /* Create and append extent to the extent array. Return the added VmdkExtent
452  * address. return NULL if allocation failed. */
453 static int vmdk_add_extent(BlockDriverState *bs,
454                            BdrvChild *file, bool flat, int64_t sectors,
455                            int64_t l1_offset, int64_t l1_backup_offset,
456                            uint32_t l1_size,
457                            int l2_size, uint64_t cluster_sectors,
458                            VmdkExtent **new_extent,
459                            Error **errp)
460 {
461     VmdkExtent *extent;
462     BDRVVmdkState *s = bs->opaque;
463     int64_t nb_sectors;
464 
465     if (cluster_sectors > 0x200000) {
466         /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
467         error_setg(errp, "Invalid granularity, image may be corrupt");
468         return -EFBIG;
469     }
470     if (l1_size > 32 * 1024 * 1024) {
471         /*
472          * Although with big capacity and small l1_entry_sectors, we can get a
473          * big l1_size, we don't want unbounded value to allocate the table.
474          * Limit it to 32M, which is enough to store:
475          *     8TB  - for both VMDK3 & VMDK4 with
476          *            minimal cluster size: 512B
477          *            minimal L2 table size: 512 entries
478          *            8 TB is still more than the maximal value supported for
479          *            VMDK3 & VMDK4 which is 2TB.
480          *     64TB - for "ESXi seSparse Extent"
481          *            minimal cluster size: 512B (default is 4KB)
482          *            L2 table size: 4096 entries (const).
483          *            64TB is more than the maximal value supported for
484          *            seSparse VMDKs (which is slightly less than 64TB)
485          */
486         error_setg(errp, "L1 size too big");
487         return -EFBIG;
488     }
489 
490     nb_sectors = bdrv_nb_sectors(file->bs);
491     if (nb_sectors < 0) {
492         return nb_sectors;
493     }
494 
495     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
496     extent = &s->extents[s->num_extents];
497     s->num_extents++;
498 
499     memset(extent, 0, sizeof(VmdkExtent));
500     extent->file = file;
501     extent->flat = flat;
502     extent->sectors = sectors;
503     extent->l1_table_offset = l1_offset;
504     extent->l1_backup_table_offset = l1_backup_offset;
505     extent->l1_size = l1_size;
506     extent->l1_entry_sectors = l2_size * cluster_sectors;
507     extent->l2_size = l2_size;
508     extent->cluster_sectors = flat ? sectors : cluster_sectors;
509     extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
510     extent->entry_size = sizeof(uint32_t);
511 
512     if (s->num_extents > 1) {
513         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
514     } else {
515         extent->end_sector = extent->sectors;
516     }
517     bs->total_sectors = extent->end_sector;
518     if (new_extent) {
519         *new_extent = extent;
520     }
521     return 0;
522 }
523 
524 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
525                             Error **errp)
526 {
527     int ret;
528     size_t l1_size;
529     int i;
530 
531     /* read the L1 table */
532     l1_size = extent->l1_size * extent->entry_size;
533     extent->l1_table = g_try_malloc(l1_size);
534     if (l1_size && extent->l1_table == NULL) {
535         return -ENOMEM;
536     }
537 
538     ret = bdrv_pread(extent->file,
539                      extent->l1_table_offset,
540                      extent->l1_table,
541                      l1_size);
542     if (ret < 0) {
543         bdrv_refresh_filename(extent->file->bs);
544         error_setg_errno(errp, -ret,
545                          "Could not read l1 table from extent '%s'",
546                          extent->file->bs->filename);
547         goto fail_l1;
548     }
549     for (i = 0; i < extent->l1_size; i++) {
550         if (extent->entry_size == sizeof(uint64_t)) {
551             le64_to_cpus((uint64_t *)extent->l1_table + i);
552         } else {
553             assert(extent->entry_size == sizeof(uint32_t));
554             le32_to_cpus((uint32_t *)extent->l1_table + i);
555         }
556     }
557 
558     if (extent->l1_backup_table_offset) {
559         assert(!extent->sesparse);
560         extent->l1_backup_table = g_try_malloc(l1_size);
561         if (l1_size && extent->l1_backup_table == NULL) {
562             ret = -ENOMEM;
563             goto fail_l1;
564         }
565         ret = bdrv_pread(extent->file,
566                          extent->l1_backup_table_offset,
567                          extent->l1_backup_table,
568                          l1_size);
569         if (ret < 0) {
570             bdrv_refresh_filename(extent->file->bs);
571             error_setg_errno(errp, -ret,
572                              "Could not read l1 backup table from extent '%s'",
573                              extent->file->bs->filename);
574             goto fail_l1b;
575         }
576         for (i = 0; i < extent->l1_size; i++) {
577             le32_to_cpus(&extent->l1_backup_table[i]);
578         }
579     }
580 
581     extent->l2_cache =
582         g_malloc(extent->entry_size * extent->l2_size * L2_CACHE_SIZE);
583     return 0;
584  fail_l1b:
585     g_free(extent->l1_backup_table);
586  fail_l1:
587     g_free(extent->l1_table);
588     return ret;
589 }
590 
591 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
592                                  BdrvChild *file,
593                                  int flags, Error **errp)
594 {
595     int ret;
596     uint32_t magic;
597     VMDK3Header header;
598     VmdkExtent *extent;
599 
600     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
601     if (ret < 0) {
602         bdrv_refresh_filename(file->bs);
603         error_setg_errno(errp, -ret,
604                          "Could not read header from file '%s'",
605                          file->bs->filename);
606         return ret;
607     }
608     ret = vmdk_add_extent(bs, file, false,
609                           le32_to_cpu(header.disk_sectors),
610                           (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
611                           0,
612                           le32_to_cpu(header.l1dir_size),
613                           4096,
614                           le32_to_cpu(header.granularity),
615                           &extent,
616                           errp);
617     if (ret < 0) {
618         return ret;
619     }
620     ret = vmdk_init_tables(bs, extent, errp);
621     if (ret) {
622         /* free extent allocated by vmdk_add_extent */
623         vmdk_free_last_extent(bs);
624     }
625     return ret;
626 }
627 
628 #define SESPARSE_CONST_HEADER_MAGIC UINT64_C(0x00000000cafebabe)
629 #define SESPARSE_VOLATILE_HEADER_MAGIC UINT64_C(0x00000000cafecafe)
630 
631 /* Strict checks - format not officially documented */
632 static int check_se_sparse_const_header(VMDKSESparseConstHeader *header,
633                                         Error **errp)
634 {
635     header->magic = le64_to_cpu(header->magic);
636     header->version = le64_to_cpu(header->version);
637     header->grain_size = le64_to_cpu(header->grain_size);
638     header->grain_table_size = le64_to_cpu(header->grain_table_size);
639     header->flags = le64_to_cpu(header->flags);
640     header->reserved1 = le64_to_cpu(header->reserved1);
641     header->reserved2 = le64_to_cpu(header->reserved2);
642     header->reserved3 = le64_to_cpu(header->reserved3);
643     header->reserved4 = le64_to_cpu(header->reserved4);
644 
645     header->volatile_header_offset =
646         le64_to_cpu(header->volatile_header_offset);
647     header->volatile_header_size = le64_to_cpu(header->volatile_header_size);
648 
649     header->journal_header_offset = le64_to_cpu(header->journal_header_offset);
650     header->journal_header_size = le64_to_cpu(header->journal_header_size);
651 
652     header->journal_offset = le64_to_cpu(header->journal_offset);
653     header->journal_size = le64_to_cpu(header->journal_size);
654 
655     header->grain_dir_offset = le64_to_cpu(header->grain_dir_offset);
656     header->grain_dir_size = le64_to_cpu(header->grain_dir_size);
657 
658     header->grain_tables_offset = le64_to_cpu(header->grain_tables_offset);
659     header->grain_tables_size = le64_to_cpu(header->grain_tables_size);
660 
661     header->free_bitmap_offset = le64_to_cpu(header->free_bitmap_offset);
662     header->free_bitmap_size = le64_to_cpu(header->free_bitmap_size);
663 
664     header->backmap_offset = le64_to_cpu(header->backmap_offset);
665     header->backmap_size = le64_to_cpu(header->backmap_size);
666 
667     header->grains_offset = le64_to_cpu(header->grains_offset);
668     header->grains_size = le64_to_cpu(header->grains_size);
669 
670     if (header->magic != SESPARSE_CONST_HEADER_MAGIC) {
671         error_setg(errp, "Bad const header magic: 0x%016" PRIx64,
672                    header->magic);
673         return -EINVAL;
674     }
675 
676     if (header->version != 0x0000000200000001) {
677         error_setg(errp, "Unsupported version: 0x%016" PRIx64,
678                    header->version);
679         return -ENOTSUP;
680     }
681 
682     if (header->grain_size != 8) {
683         error_setg(errp, "Unsupported grain size: %" PRIu64,
684                    header->grain_size);
685         return -ENOTSUP;
686     }
687 
688     if (header->grain_table_size != 64) {
689         error_setg(errp, "Unsupported grain table size: %" PRIu64,
690                    header->grain_table_size);
691         return -ENOTSUP;
692     }
693 
694     if (header->flags != 0) {
695         error_setg(errp, "Unsupported flags: 0x%016" PRIx64,
696                    header->flags);
697         return -ENOTSUP;
698     }
699 
700     if (header->reserved1 != 0 || header->reserved2 != 0 ||
701         header->reserved3 != 0 || header->reserved4 != 0) {
702         error_setg(errp, "Unsupported reserved bits:"
703                    " 0x%016" PRIx64 " 0x%016" PRIx64
704                    " 0x%016" PRIx64 " 0x%016" PRIx64,
705                    header->reserved1, header->reserved2,
706                    header->reserved3, header->reserved4);
707         return -ENOTSUP;
708     }
709 
710     /* check that padding is 0 */
711     if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
712         error_setg(errp, "Unsupported non-zero const header padding");
713         return -ENOTSUP;
714     }
715 
716     return 0;
717 }
718 
719 static int check_se_sparse_volatile_header(VMDKSESparseVolatileHeader *header,
720                                            Error **errp)
721 {
722     header->magic = le64_to_cpu(header->magic);
723     header->free_gt_number = le64_to_cpu(header->free_gt_number);
724     header->next_txn_seq_number = le64_to_cpu(header->next_txn_seq_number);
725     header->replay_journal = le64_to_cpu(header->replay_journal);
726 
727     if (header->magic != SESPARSE_VOLATILE_HEADER_MAGIC) {
728         error_setg(errp, "Bad volatile header magic: 0x%016" PRIx64,
729                    header->magic);
730         return -EINVAL;
731     }
732 
733     if (header->replay_journal) {
734         error_setg(errp, "Image is dirty, Replaying journal not supported");
735         return -ENOTSUP;
736     }
737 
738     /* check that padding is 0 */
739     if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
740         error_setg(errp, "Unsupported non-zero volatile header padding");
741         return -ENOTSUP;
742     }
743 
744     return 0;
745 }
746 
747 static int vmdk_open_se_sparse(BlockDriverState *bs,
748                                BdrvChild *file,
749                                int flags, Error **errp)
750 {
751     int ret;
752     VMDKSESparseConstHeader const_header;
753     VMDKSESparseVolatileHeader volatile_header;
754     VmdkExtent *extent;
755 
756     ret = bdrv_apply_auto_read_only(bs,
757             "No write support for seSparse images available", errp);
758     if (ret < 0) {
759         return ret;
760     }
761 
762     assert(sizeof(const_header) == SECTOR_SIZE);
763 
764     ret = bdrv_pread(file, 0, &const_header, sizeof(const_header));
765     if (ret < 0) {
766         bdrv_refresh_filename(file->bs);
767         error_setg_errno(errp, -ret,
768                          "Could not read const header from file '%s'",
769                          file->bs->filename);
770         return ret;
771     }
772 
773     /* check const header */
774     ret = check_se_sparse_const_header(&const_header, errp);
775     if (ret < 0) {
776         return ret;
777     }
778 
779     assert(sizeof(volatile_header) == SECTOR_SIZE);
780 
781     ret = bdrv_pread(file,
782                      const_header.volatile_header_offset * SECTOR_SIZE,
783                      &volatile_header, sizeof(volatile_header));
784     if (ret < 0) {
785         bdrv_refresh_filename(file->bs);
786         error_setg_errno(errp, -ret,
787                          "Could not read volatile header from file '%s'",
788                          file->bs->filename);
789         return ret;
790     }
791 
792     /* check volatile header */
793     ret = check_se_sparse_volatile_header(&volatile_header, errp);
794     if (ret < 0) {
795         return ret;
796     }
797 
798     ret = vmdk_add_extent(bs, file, false,
799                           const_header.capacity,
800                           const_header.grain_dir_offset * SECTOR_SIZE,
801                           0,
802                           const_header.grain_dir_size *
803                           SECTOR_SIZE / sizeof(uint64_t),
804                           const_header.grain_table_size *
805                           SECTOR_SIZE / sizeof(uint64_t),
806                           const_header.grain_size,
807                           &extent,
808                           errp);
809     if (ret < 0) {
810         return ret;
811     }
812 
813     extent->sesparse = true;
814     extent->sesparse_l2_tables_offset = const_header.grain_tables_offset;
815     extent->sesparse_clusters_offset = const_header.grains_offset;
816     extent->entry_size = sizeof(uint64_t);
817 
818     ret = vmdk_init_tables(bs, extent, errp);
819     if (ret) {
820         /* free extent allocated by vmdk_add_extent */
821         vmdk_free_last_extent(bs);
822     }
823 
824     return ret;
825 }
826 
827 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
828                                QDict *options, Error **errp);
829 
830 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
831 {
832     int64_t size;
833     char *buf;
834     int ret;
835 
836     size = bdrv_getlength(file->bs);
837     if (size < 0) {
838         error_setg_errno(errp, -size, "Could not access file");
839         return NULL;
840     }
841 
842     if (size < 4) {
843         /* Both descriptor file and sparse image must be much larger than 4
844          * bytes, also callers of vmdk_read_desc want to compare the first 4
845          * bytes with VMDK4_MAGIC, let's error out if less is read. */
846         error_setg(errp, "File is too small, not a valid image");
847         return NULL;
848     }
849 
850     size = MIN(size, (1 << 20) - 1);  /* avoid unbounded allocation */
851     buf = g_malloc(size + 1);
852 
853     ret = bdrv_pread(file, desc_offset, buf, size);
854     if (ret < 0) {
855         error_setg_errno(errp, -ret, "Could not read from file");
856         g_free(buf);
857         return NULL;
858     }
859     buf[ret] = 0;
860 
861     return buf;
862 }
863 
864 static int vmdk_open_vmdk4(BlockDriverState *bs,
865                            BdrvChild *file,
866                            int flags, QDict *options, Error **errp)
867 {
868     int ret;
869     uint32_t magic;
870     uint32_t l1_size, l1_entry_sectors;
871     VMDK4Header header;
872     VmdkExtent *extent;
873     BDRVVmdkState *s = bs->opaque;
874     int64_t l1_backup_offset = 0;
875     bool compressed;
876 
877     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
878     if (ret < 0) {
879         bdrv_refresh_filename(file->bs);
880         error_setg_errno(errp, -ret,
881                          "Could not read header from file '%s'",
882                          file->bs->filename);
883         return -EINVAL;
884     }
885     if (header.capacity == 0) {
886         uint64_t desc_offset = le64_to_cpu(header.desc_offset);
887         if (desc_offset) {
888             char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
889             if (!buf) {
890                 return -EINVAL;
891             }
892             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
893             g_free(buf);
894             return ret;
895         }
896     }
897 
898     if (!s->create_type) {
899         s->create_type = g_strdup("monolithicSparse");
900     }
901 
902     if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
903         /*
904          * The footer takes precedence over the header, so read it in. The
905          * footer starts at offset -1024 from the end: One sector for the
906          * footer, and another one for the end-of-stream marker.
907          */
908         struct {
909             struct {
910                 uint64_t val;
911                 uint32_t size;
912                 uint32_t type;
913                 uint8_t pad[512 - 16];
914             } QEMU_PACKED footer_marker;
915 
916             uint32_t magic;
917             VMDK4Header header;
918             uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
919 
920             struct {
921                 uint64_t val;
922                 uint32_t size;
923                 uint32_t type;
924                 uint8_t pad[512 - 16];
925             } QEMU_PACKED eos_marker;
926         } QEMU_PACKED footer;
927 
928         ret = bdrv_pread(file,
929             bs->file->bs->total_sectors * 512 - 1536,
930             &footer, sizeof(footer));
931         if (ret < 0) {
932             error_setg_errno(errp, -ret, "Failed to read footer");
933             return ret;
934         }
935 
936         /* Some sanity checks for the footer */
937         if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
938             le32_to_cpu(footer.footer_marker.size) != 0  ||
939             le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
940             le64_to_cpu(footer.eos_marker.val) != 0  ||
941             le32_to_cpu(footer.eos_marker.size) != 0  ||
942             le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
943         {
944             error_setg(errp, "Invalid footer");
945             return -EINVAL;
946         }
947 
948         header = footer.header;
949     }
950 
951     compressed =
952         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
953     if (le32_to_cpu(header.version) > 3) {
954         error_setg(errp, "Unsupported VMDK version %" PRIu32,
955                    le32_to_cpu(header.version));
956         return -ENOTSUP;
957     } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
958                !compressed) {
959         /* VMware KB 2064959 explains that version 3 added support for
960          * persistent changed block tracking (CBT), and backup software can
961          * read it as version=1 if it doesn't care about the changed area
962          * information. So we are safe to enable read only. */
963         error_setg(errp, "VMDK version 3 must be read only");
964         return -EINVAL;
965     }
966 
967     if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
968         error_setg(errp, "L2 table size too big");
969         return -EINVAL;
970     }
971 
972     l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
973                         * le64_to_cpu(header.granularity);
974     if (l1_entry_sectors == 0) {
975         error_setg(errp, "L1 entry size is invalid");
976         return -EINVAL;
977     }
978     l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
979                 / l1_entry_sectors;
980     if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
981         l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
982     }
983     if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
984         error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
985                    (int64_t)(le64_to_cpu(header.grain_offset)
986                              * BDRV_SECTOR_SIZE));
987         return -EINVAL;
988     }
989 
990     ret = vmdk_add_extent(bs, file, false,
991                           le64_to_cpu(header.capacity),
992                           le64_to_cpu(header.gd_offset) << 9,
993                           l1_backup_offset,
994                           l1_size,
995                           le32_to_cpu(header.num_gtes_per_gt),
996                           le64_to_cpu(header.granularity),
997                           &extent,
998                           errp);
999     if (ret < 0) {
1000         return ret;
1001     }
1002     extent->compressed =
1003         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1004     if (extent->compressed) {
1005         g_free(s->create_type);
1006         s->create_type = g_strdup("streamOptimized");
1007     }
1008     extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
1009     extent->version = le32_to_cpu(header.version);
1010     extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
1011     ret = vmdk_init_tables(bs, extent, errp);
1012     if (ret) {
1013         /* free extent allocated by vmdk_add_extent */
1014         vmdk_free_last_extent(bs);
1015     }
1016     return ret;
1017 }
1018 
1019 /* find an option value out of descriptor file */
1020 static int vmdk_parse_description(const char *desc, const char *opt_name,
1021         char *buf, int buf_size)
1022 {
1023     char *opt_pos, *opt_end;
1024     const char *end = desc + strlen(desc);
1025 
1026     opt_pos = strstr(desc, opt_name);
1027     if (!opt_pos) {
1028         return VMDK_ERROR;
1029     }
1030     /* Skip "=\"" following opt_name */
1031     opt_pos += strlen(opt_name) + 2;
1032     if (opt_pos >= end) {
1033         return VMDK_ERROR;
1034     }
1035     opt_end = opt_pos;
1036     while (opt_end < end && *opt_end != '"') {
1037         opt_end++;
1038     }
1039     if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
1040         return VMDK_ERROR;
1041     }
1042     pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
1043     return VMDK_OK;
1044 }
1045 
1046 /* Open an extent file and append to bs array */
1047 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
1048                             char *buf, QDict *options, Error **errp)
1049 {
1050     uint32_t magic;
1051 
1052     magic = ldl_be_p(buf);
1053     switch (magic) {
1054         case VMDK3_MAGIC:
1055             return vmdk_open_vmfs_sparse(bs, file, flags, errp);
1056             break;
1057         case VMDK4_MAGIC:
1058             return vmdk_open_vmdk4(bs, file, flags, options, errp);
1059             break;
1060         default:
1061             error_setg(errp, "Image not in VMDK format");
1062             return -EINVAL;
1063             break;
1064     }
1065 }
1066 
1067 static const char *next_line(const char *s)
1068 {
1069     while (*s) {
1070         if (*s == '\n') {
1071             return s + 1;
1072         }
1073         s++;
1074     }
1075     return s;
1076 }
1077 
1078 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
1079                               QDict *options, Error **errp)
1080 {
1081     int ret;
1082     int matches;
1083     char access[11];
1084     char type[11];
1085     char fname[512];
1086     const char *p, *np;
1087     int64_t sectors = 0;
1088     int64_t flat_offset;
1089     char *desc_file_dir = NULL;
1090     char *extent_path;
1091     BdrvChild *extent_file;
1092     BDRVVmdkState *s = bs->opaque;
1093     VmdkExtent *extent;
1094     char extent_opt_prefix[32];
1095     Error *local_err = NULL;
1096 
1097     for (p = desc; *p; p = next_line(p)) {
1098         /* parse extent line in one of below formats:
1099          *
1100          * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
1101          * RW [size in sectors] SPARSE "file-name.vmdk"
1102          * RW [size in sectors] VMFS "file-name.vmdk"
1103          * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
1104          * RW [size in sectors] SESPARSE "file-name.vmdk"
1105          */
1106         flat_offset = -1;
1107         matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
1108                          access, &sectors, type, fname, &flat_offset);
1109         if (matches < 4 || strcmp(access, "RW")) {
1110             continue;
1111         } else if (!strcmp(type, "FLAT")) {
1112             if (matches != 5 || flat_offset < 0) {
1113                 goto invalid;
1114             }
1115         } else if (!strcmp(type, "VMFS")) {
1116             if (matches == 4) {
1117                 flat_offset = 0;
1118             } else {
1119                 goto invalid;
1120             }
1121         } else if (matches != 4) {
1122             goto invalid;
1123         }
1124 
1125         if (sectors <= 0 ||
1126             (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
1127              strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") &&
1128              strcmp(type, "SESPARSE")) ||
1129             (strcmp(access, "RW"))) {
1130             continue;
1131         }
1132 
1133         if (path_is_absolute(fname)) {
1134             extent_path = g_strdup(fname);
1135         } else {
1136             if (!desc_file_dir) {
1137                 desc_file_dir = bdrv_dirname(bs->file->bs, errp);
1138                 if (!desc_file_dir) {
1139                     bdrv_refresh_filename(bs->file->bs);
1140                     error_prepend(errp, "Cannot use relative paths with VMDK "
1141                                   "descriptor file '%s': ",
1142                                   bs->file->bs->filename);
1143                     ret = -EINVAL;
1144                     goto out;
1145                 }
1146             }
1147 
1148             extent_path = g_strconcat(desc_file_dir, fname, NULL);
1149         }
1150 
1151         ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
1152         assert(ret < 32);
1153 
1154         extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
1155                                       bs, &child_file, false, &local_err);
1156         g_free(extent_path);
1157         if (local_err) {
1158             error_propagate(errp, local_err);
1159             ret = -EINVAL;
1160             goto out;
1161         }
1162 
1163         /* save to extents array */
1164         if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
1165             /* FLAT extent */
1166 
1167             ret = vmdk_add_extent(bs, extent_file, true, sectors,
1168                             0, 0, 0, 0, 0, &extent, errp);
1169             if (ret < 0) {
1170                 bdrv_unref_child(bs, extent_file);
1171                 goto out;
1172             }
1173             extent->flat_start_offset = flat_offset << 9;
1174         } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
1175             /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
1176             char *buf = vmdk_read_desc(extent_file, 0, errp);
1177             if (!buf) {
1178                 ret = -EINVAL;
1179             } else {
1180                 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
1181                                        options, errp);
1182             }
1183             g_free(buf);
1184             if (ret) {
1185                 bdrv_unref_child(bs, extent_file);
1186                 goto out;
1187             }
1188             extent = &s->extents[s->num_extents - 1];
1189         } else if (!strcmp(type, "SESPARSE")) {
1190             ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
1191             if (ret) {
1192                 bdrv_unref_child(bs, extent_file);
1193                 goto out;
1194             }
1195             extent = &s->extents[s->num_extents - 1];
1196         } else {
1197             error_setg(errp, "Unsupported extent type '%s'", type);
1198             bdrv_unref_child(bs, extent_file);
1199             ret = -ENOTSUP;
1200             goto out;
1201         }
1202         extent->type = g_strdup(type);
1203     }
1204 
1205     ret = 0;
1206     goto out;
1207 
1208 invalid:
1209     np = next_line(p);
1210     assert(np != p);
1211     if (np[-1] == '\n') {
1212         np--;
1213     }
1214     error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
1215     ret = -EINVAL;
1216 
1217 out:
1218     g_free(desc_file_dir);
1219     return ret;
1220 }
1221 
1222 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
1223                                QDict *options, Error **errp)
1224 {
1225     int ret;
1226     char ct[128];
1227     BDRVVmdkState *s = bs->opaque;
1228 
1229     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
1230         error_setg(errp, "invalid VMDK image descriptor");
1231         ret = -EINVAL;
1232         goto exit;
1233     }
1234     if (strcmp(ct, "monolithicFlat") &&
1235         strcmp(ct, "vmfs") &&
1236         strcmp(ct, "vmfsSparse") &&
1237         strcmp(ct, "seSparse") &&
1238         strcmp(ct, "twoGbMaxExtentSparse") &&
1239         strcmp(ct, "twoGbMaxExtentFlat")) {
1240         error_setg(errp, "Unsupported image type '%s'", ct);
1241         ret = -ENOTSUP;
1242         goto exit;
1243     }
1244     s->create_type = g_strdup(ct);
1245     s->desc_offset = 0;
1246     ret = vmdk_parse_extents(buf, bs, options, errp);
1247 exit:
1248     return ret;
1249 }
1250 
1251 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
1252                      Error **errp)
1253 {
1254     char *buf;
1255     int ret;
1256     BDRVVmdkState *s = bs->opaque;
1257     uint32_t magic;
1258     Error *local_err = NULL;
1259 
1260     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
1261                                false, errp);
1262     if (!bs->file) {
1263         return -EINVAL;
1264     }
1265 
1266     buf = vmdk_read_desc(bs->file, 0, errp);
1267     if (!buf) {
1268         return -EINVAL;
1269     }
1270 
1271     magic = ldl_be_p(buf);
1272     switch (magic) {
1273         case VMDK3_MAGIC:
1274         case VMDK4_MAGIC:
1275             ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
1276                                    errp);
1277             s->desc_offset = 0x200;
1278             break;
1279         default:
1280             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1281             break;
1282     }
1283     if (ret) {
1284         goto fail;
1285     }
1286 
1287     /* try to open parent images, if exist */
1288     ret = vmdk_parent_open(bs);
1289     if (ret) {
1290         goto fail;
1291     }
1292     ret = vmdk_read_cid(bs, 0, &s->cid);
1293     if (ret) {
1294         goto fail;
1295     }
1296     ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1297     if (ret) {
1298         goto fail;
1299     }
1300     qemu_co_mutex_init(&s->lock);
1301 
1302     /* Disable migration when VMDK images are used */
1303     error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1304                "does not support live migration",
1305                bdrv_get_device_or_node_name(bs));
1306     ret = migrate_add_blocker(s->migration_blocker, &local_err);
1307     if (local_err) {
1308         error_propagate(errp, local_err);
1309         error_free(s->migration_blocker);
1310         goto fail;
1311     }
1312 
1313     g_free(buf);
1314     return 0;
1315 
1316 fail:
1317     g_free(buf);
1318     g_free(s->create_type);
1319     s->create_type = NULL;
1320     vmdk_free_extents(bs);
1321     return ret;
1322 }
1323 
1324 
1325 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1326 {
1327     BDRVVmdkState *s = bs->opaque;
1328     int i;
1329 
1330     for (i = 0; i < s->num_extents; i++) {
1331         if (!s->extents[i].flat) {
1332             bs->bl.pwrite_zeroes_alignment =
1333                 MAX(bs->bl.pwrite_zeroes_alignment,
1334                     s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1335         }
1336     }
1337 }
1338 
1339 /**
1340  * get_whole_cluster
1341  *
1342  * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1343  * to the cluster at @cluster_sector_num.
1344  *
1345  * If @skip_start_sector < @skip_end_sector, the relative range
1346  * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1347  * it for call to write user data in the request.
1348  */
1349 static int get_whole_cluster(BlockDriverState *bs,
1350                              VmdkExtent *extent,
1351                              uint64_t cluster_offset,
1352                              uint64_t offset,
1353                              uint64_t skip_start_bytes,
1354                              uint64_t skip_end_bytes)
1355 {
1356     int ret = VMDK_OK;
1357     int64_t cluster_bytes;
1358     uint8_t *whole_grain;
1359 
1360     /* For COW, align request sector_num to cluster start */
1361     cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1362     offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1363     whole_grain = qemu_blockalign(bs, cluster_bytes);
1364 
1365     if (!bs->backing) {
1366         memset(whole_grain, 0, skip_start_bytes);
1367         memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1368     }
1369 
1370     assert(skip_end_bytes <= cluster_bytes);
1371     /* we will be here if it's first write on non-exist grain(cluster).
1372      * try to read from parent image, if exist */
1373     if (bs->backing && !vmdk_is_cid_valid(bs)) {
1374         ret = VMDK_ERROR;
1375         goto exit;
1376     }
1377 
1378     /* Read backing data before skip range */
1379     if (skip_start_bytes > 0) {
1380         if (bs->backing) {
1381             /* qcow2 emits this on bs->file instead of bs->backing */
1382             BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1383             ret = bdrv_pread(bs->backing, offset, whole_grain,
1384                              skip_start_bytes);
1385             if (ret < 0) {
1386                 ret = VMDK_ERROR;
1387                 goto exit;
1388             }
1389         }
1390         BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1391         ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1392                           skip_start_bytes);
1393         if (ret < 0) {
1394             ret = VMDK_ERROR;
1395             goto exit;
1396         }
1397     }
1398     /* Read backing data after skip range */
1399     if (skip_end_bytes < cluster_bytes) {
1400         if (bs->backing) {
1401             /* qcow2 emits this on bs->file instead of bs->backing */
1402             BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1403             ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1404                              whole_grain + skip_end_bytes,
1405                              cluster_bytes - skip_end_bytes);
1406             if (ret < 0) {
1407                 ret = VMDK_ERROR;
1408                 goto exit;
1409             }
1410         }
1411         BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1412         ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1413                           whole_grain + skip_end_bytes,
1414                           cluster_bytes - skip_end_bytes);
1415         if (ret < 0) {
1416             ret = VMDK_ERROR;
1417             goto exit;
1418         }
1419     }
1420 
1421     ret = VMDK_OK;
1422 exit:
1423     qemu_vfree(whole_grain);
1424     return ret;
1425 }
1426 
1427 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1428                          uint32_t offset)
1429 {
1430     offset = cpu_to_le32(offset);
1431     /* update L2 table */
1432     BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1433     if (bdrv_pwrite_sync(extent->file,
1434                 ((int64_t)m_data->l2_offset * 512)
1435                     + (m_data->l2_index * sizeof(offset)),
1436                 &offset, sizeof(offset)) < 0) {
1437         return VMDK_ERROR;
1438     }
1439     /* update backup L2 table */
1440     if (extent->l1_backup_table_offset != 0) {
1441         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1442         if (bdrv_pwrite_sync(extent->file,
1443                     ((int64_t)m_data->l2_offset * 512)
1444                         + (m_data->l2_index * sizeof(offset)),
1445                     &offset, sizeof(offset)) < 0) {
1446             return VMDK_ERROR;
1447         }
1448     }
1449     if (m_data->l2_cache_entry) {
1450         *m_data->l2_cache_entry = offset;
1451     }
1452 
1453     return VMDK_OK;
1454 }
1455 
1456 /**
1457  * get_cluster_offset
1458  *
1459  * Look up cluster offset in extent file by sector number, and store in
1460  * @cluster_offset.
1461  *
1462  * For flat extents, the start offset as parsed from the description file is
1463  * returned.
1464  *
1465  * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1466  * offset for a new cluster and update L2 cache. If there is a backing file,
1467  * COW is done before returning; otherwise, zeroes are written to the allocated
1468  * cluster. Both COW and zero writing skips the sector range
1469  * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1470  * has new data to write there.
1471  *
1472  * Returns: VMDK_OK if cluster exists and mapped in the image.
1473  *          VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1474  *          VMDK_ERROR if failed.
1475  */
1476 static int get_cluster_offset(BlockDriverState *bs,
1477                               VmdkExtent *extent,
1478                               VmdkMetaData *m_data,
1479                               uint64_t offset,
1480                               bool allocate,
1481                               uint64_t *cluster_offset,
1482                               uint64_t skip_start_bytes,
1483                               uint64_t skip_end_bytes)
1484 {
1485     unsigned int l1_index, l2_offset, l2_index;
1486     int min_index, i, j;
1487     uint32_t min_count;
1488     void *l2_table;
1489     bool zeroed = false;
1490     int64_t ret;
1491     int64_t cluster_sector;
1492     unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
1493 
1494     if (m_data) {
1495         m_data->valid = 0;
1496     }
1497     if (extent->flat) {
1498         *cluster_offset = extent->flat_start_offset;
1499         return VMDK_OK;
1500     }
1501 
1502     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1503     l1_index = (offset >> 9) / extent->l1_entry_sectors;
1504     if (l1_index >= extent->l1_size) {
1505         return VMDK_ERROR;
1506     }
1507     if (extent->sesparse) {
1508         uint64_t l2_offset_u64;
1509 
1510         assert(extent->entry_size == sizeof(uint64_t));
1511 
1512         l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
1513         if (l2_offset_u64 == 0) {
1514             l2_offset = 0;
1515         } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
1516             /*
1517              * Top most nibble is 0x1 if grain table is allocated.
1518              * strict check - top most 4 bytes must be 0x10000000 since max
1519              * supported size is 64TB for disk - so no more than 64TB / 16MB
1520              * grain directories which is smaller than uint32,
1521              * where 16MB is the only supported default grain table coverage.
1522              */
1523             return VMDK_ERROR;
1524         } else {
1525             l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
1526             l2_offset_u64 = extent->sesparse_l2_tables_offset +
1527                 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
1528             if (l2_offset_u64 > 0x00000000ffffffff) {
1529                 return VMDK_ERROR;
1530             }
1531             l2_offset = (unsigned int)(l2_offset_u64);
1532         }
1533     } else {
1534         assert(extent->entry_size == sizeof(uint32_t));
1535         l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
1536     }
1537     if (!l2_offset) {
1538         return VMDK_UNALLOC;
1539     }
1540     for (i = 0; i < L2_CACHE_SIZE; i++) {
1541         if (l2_offset == extent->l2_cache_offsets[i]) {
1542             /* increment the hit count */
1543             if (++extent->l2_cache_counts[i] == 0xffffffff) {
1544                 for (j = 0; j < L2_CACHE_SIZE; j++) {
1545                     extent->l2_cache_counts[j] >>= 1;
1546                 }
1547             }
1548             l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
1549             goto found;
1550         }
1551     }
1552     /* not found: load a new entry in the least used one */
1553     min_index = 0;
1554     min_count = 0xffffffff;
1555     for (i = 0; i < L2_CACHE_SIZE; i++) {
1556         if (extent->l2_cache_counts[i] < min_count) {
1557             min_count = extent->l2_cache_counts[i];
1558             min_index = i;
1559         }
1560     }
1561     l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
1562     BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1563     if (bdrv_pread(extent->file,
1564                 (int64_t)l2_offset * 512,
1565                 l2_table,
1566                 l2_size_bytes
1567             ) != l2_size_bytes) {
1568         return VMDK_ERROR;
1569     }
1570 
1571     extent->l2_cache_offsets[min_index] = l2_offset;
1572     extent->l2_cache_counts[min_index] = 1;
1573  found:
1574     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1575 
1576     if (extent->sesparse) {
1577         cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
1578         switch (cluster_sector & 0xf000000000000000) {
1579         case 0x0000000000000000:
1580             /* unallocated grain */
1581             if (cluster_sector != 0) {
1582                 return VMDK_ERROR;
1583             }
1584             break;
1585         case 0x1000000000000000:
1586             /* scsi-unmapped grain - fallthrough */
1587         case 0x2000000000000000:
1588             /* zero grain */
1589             zeroed = true;
1590             break;
1591         case 0x3000000000000000:
1592             /* allocated grain */
1593             cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
1594                               ((cluster_sector & 0x0000ffffffffffff) << 12));
1595             cluster_sector = extent->sesparse_clusters_offset +
1596                 cluster_sector * extent->cluster_sectors;
1597             break;
1598         default:
1599             return VMDK_ERROR;
1600         }
1601     } else {
1602         cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
1603 
1604         if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1605             zeroed = true;
1606         }
1607     }
1608 
1609     if (!cluster_sector || zeroed) {
1610         if (!allocate) {
1611             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1612         }
1613         assert(!extent->sesparse);
1614 
1615         if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1616             return VMDK_ERROR;
1617         }
1618 
1619         cluster_sector = extent->next_cluster_sector;
1620         extent->next_cluster_sector += extent->cluster_sectors;
1621 
1622         /* First of all we write grain itself, to avoid race condition
1623          * that may to corrupt the image.
1624          * This problem may occur because of insufficient space on host disk
1625          * or inappropriate VM shutdown.
1626          */
1627         ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1628                                 offset, skip_start_bytes, skip_end_bytes);
1629         if (ret) {
1630             return ret;
1631         }
1632         if (m_data) {
1633             m_data->valid = 1;
1634             m_data->l1_index = l1_index;
1635             m_data->l2_index = l2_index;
1636             m_data->l2_offset = l2_offset;
1637             m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
1638         }
1639     }
1640     *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1641     return VMDK_OK;
1642 }
1643 
1644 static VmdkExtent *find_extent(BDRVVmdkState *s,
1645                                 int64_t sector_num, VmdkExtent *start_hint)
1646 {
1647     VmdkExtent *extent = start_hint;
1648 
1649     if (!extent) {
1650         extent = &s->extents[0];
1651     }
1652     while (extent < &s->extents[s->num_extents]) {
1653         if (sector_num < extent->end_sector) {
1654             return extent;
1655         }
1656         extent++;
1657     }
1658     return NULL;
1659 }
1660 
1661 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1662                                                    int64_t offset)
1663 {
1664     uint64_t extent_begin_offset, extent_relative_offset;
1665     uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1666 
1667     extent_begin_offset =
1668         (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1669     extent_relative_offset = offset - extent_begin_offset;
1670     return extent_relative_offset % cluster_size;
1671 }
1672 
1673 static int coroutine_fn vmdk_co_block_status(BlockDriverState *bs,
1674                                              bool want_zero,
1675                                              int64_t offset, int64_t bytes,
1676                                              int64_t *pnum, int64_t *map,
1677                                              BlockDriverState **file)
1678 {
1679     BDRVVmdkState *s = bs->opaque;
1680     int64_t index_in_cluster, n, ret;
1681     uint64_t cluster_offset;
1682     VmdkExtent *extent;
1683 
1684     extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1685     if (!extent) {
1686         return -EIO;
1687     }
1688     qemu_co_mutex_lock(&s->lock);
1689     ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1690                              0, 0);
1691     qemu_co_mutex_unlock(&s->lock);
1692 
1693     index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1694     switch (ret) {
1695     case VMDK_ERROR:
1696         ret = -EIO;
1697         break;
1698     case VMDK_UNALLOC:
1699         ret = 0;
1700         break;
1701     case VMDK_ZEROED:
1702         ret = BDRV_BLOCK_ZERO;
1703         break;
1704     case VMDK_OK:
1705         ret = BDRV_BLOCK_DATA;
1706         if (!extent->compressed) {
1707             ret |= BDRV_BLOCK_OFFSET_VALID;
1708             *map = cluster_offset + index_in_cluster;
1709             if (extent->flat) {
1710                 ret |= BDRV_BLOCK_RECURSE;
1711             }
1712         }
1713         *file = extent->file->bs;
1714         break;
1715     }
1716 
1717     n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1718     *pnum = MIN(n, bytes);
1719     return ret;
1720 }
1721 
1722 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1723                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1724                             uint64_t qiov_offset, uint64_t n_bytes,
1725                             uint64_t offset)
1726 {
1727     int ret;
1728     VmdkGrainMarker *data = NULL;
1729     uLongf buf_len;
1730     QEMUIOVector local_qiov;
1731     int64_t write_offset;
1732     int64_t write_end_sector;
1733 
1734     if (extent->compressed) {
1735         void *compressed_data;
1736 
1737         /* Only whole clusters */
1738         if (offset_in_cluster ||
1739             n_bytes > (extent->cluster_sectors * SECTOR_SIZE) ||
1740             (n_bytes < (extent->cluster_sectors * SECTOR_SIZE) &&
1741              offset + n_bytes != extent->end_sector * SECTOR_SIZE))
1742         {
1743             ret = -EINVAL;
1744             goto out;
1745         }
1746 
1747         if (!extent->has_marker) {
1748             ret = -EINVAL;
1749             goto out;
1750         }
1751         buf_len = (extent->cluster_sectors << 9) * 2;
1752         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1753 
1754         compressed_data = g_malloc(n_bytes);
1755         qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1756         ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1757         g_free(compressed_data);
1758 
1759         if (ret != Z_OK || buf_len == 0) {
1760             ret = -EINVAL;
1761             goto out;
1762         }
1763 
1764         data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1765         data->size = cpu_to_le32(buf_len);
1766 
1767         n_bytes = buf_len + sizeof(VmdkGrainMarker);
1768         qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1769 
1770         BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1771     } else {
1772         qemu_iovec_init(&local_qiov, qiov->niov);
1773         qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1774 
1775         BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1776     }
1777 
1778     write_offset = cluster_offset + offset_in_cluster;
1779     ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1780                           &local_qiov, 0);
1781 
1782     write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1783 
1784     if (extent->compressed) {
1785         extent->next_cluster_sector = write_end_sector;
1786     } else {
1787         extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1788                                           write_end_sector);
1789     }
1790 
1791     if (ret < 0) {
1792         goto out;
1793     }
1794     ret = 0;
1795  out:
1796     g_free(data);
1797     if (!extent->compressed) {
1798         qemu_iovec_destroy(&local_qiov);
1799     }
1800     return ret;
1801 }
1802 
1803 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1804                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1805                             int bytes)
1806 {
1807     int ret;
1808     int cluster_bytes, buf_bytes;
1809     uint8_t *cluster_buf, *compressed_data;
1810     uint8_t *uncomp_buf;
1811     uint32_t data_len;
1812     VmdkGrainMarker *marker;
1813     uLongf buf_len;
1814 
1815 
1816     if (!extent->compressed) {
1817         BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1818         ret = bdrv_co_preadv(extent->file,
1819                              cluster_offset + offset_in_cluster, bytes,
1820                              qiov, 0);
1821         if (ret < 0) {
1822             return ret;
1823         }
1824         return 0;
1825     }
1826     cluster_bytes = extent->cluster_sectors * 512;
1827     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1828     buf_bytes = cluster_bytes * 2;
1829     cluster_buf = g_malloc(buf_bytes);
1830     uncomp_buf = g_malloc(cluster_bytes);
1831     BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1832     ret = bdrv_pread(extent->file,
1833                 cluster_offset,
1834                 cluster_buf, buf_bytes);
1835     if (ret < 0) {
1836         goto out;
1837     }
1838     compressed_data = cluster_buf;
1839     buf_len = cluster_bytes;
1840     data_len = cluster_bytes;
1841     if (extent->has_marker) {
1842         marker = (VmdkGrainMarker *)cluster_buf;
1843         compressed_data = marker->data;
1844         data_len = le32_to_cpu(marker->size);
1845     }
1846     if (!data_len || data_len > buf_bytes) {
1847         ret = -EINVAL;
1848         goto out;
1849     }
1850     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1851     if (ret != Z_OK) {
1852         ret = -EINVAL;
1853         goto out;
1854 
1855     }
1856     if (offset_in_cluster < 0 ||
1857             offset_in_cluster + bytes > buf_len) {
1858         ret = -EINVAL;
1859         goto out;
1860     }
1861     qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1862     ret = 0;
1863 
1864  out:
1865     g_free(uncomp_buf);
1866     g_free(cluster_buf);
1867     return ret;
1868 }
1869 
1870 static int coroutine_fn
1871 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1872                QEMUIOVector *qiov, int flags)
1873 {
1874     BDRVVmdkState *s = bs->opaque;
1875     int ret;
1876     uint64_t n_bytes, offset_in_cluster;
1877     VmdkExtent *extent = NULL;
1878     QEMUIOVector local_qiov;
1879     uint64_t cluster_offset;
1880     uint64_t bytes_done = 0;
1881 
1882     qemu_iovec_init(&local_qiov, qiov->niov);
1883     qemu_co_mutex_lock(&s->lock);
1884 
1885     while (bytes > 0) {
1886         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1887         if (!extent) {
1888             ret = -EIO;
1889             goto fail;
1890         }
1891         ret = get_cluster_offset(bs, extent, NULL,
1892                                  offset, false, &cluster_offset, 0, 0);
1893         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1894 
1895         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1896                              - offset_in_cluster);
1897 
1898         if (ret != VMDK_OK) {
1899             /* if not allocated, try to read from parent image, if exist */
1900             if (bs->backing && ret != VMDK_ZEROED) {
1901                 if (!vmdk_is_cid_valid(bs)) {
1902                     ret = -EINVAL;
1903                     goto fail;
1904                 }
1905 
1906                 qemu_iovec_reset(&local_qiov);
1907                 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1908 
1909                 /* qcow2 emits this on bs->file instead of bs->backing */
1910                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1911                 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1912                                      &local_qiov, 0);
1913                 if (ret < 0) {
1914                     goto fail;
1915                 }
1916             } else {
1917                 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1918             }
1919         } else {
1920             qemu_iovec_reset(&local_qiov);
1921             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1922 
1923             ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1924                                    &local_qiov, n_bytes);
1925             if (ret) {
1926                 goto fail;
1927             }
1928         }
1929         bytes -= n_bytes;
1930         offset += n_bytes;
1931         bytes_done += n_bytes;
1932     }
1933 
1934     ret = 0;
1935 fail:
1936     qemu_co_mutex_unlock(&s->lock);
1937     qemu_iovec_destroy(&local_qiov);
1938 
1939     return ret;
1940 }
1941 
1942 /**
1943  * vmdk_write:
1944  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
1945  *                if possible, otherwise return -ENOTSUP.
1946  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1947  *                with each cluster. By dry run we can find if the zero write
1948  *                is possible without modifying image data.
1949  *
1950  * Returns: error code with 0 for success.
1951  */
1952 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1953                        uint64_t bytes, QEMUIOVector *qiov,
1954                        bool zeroed, bool zero_dry_run)
1955 {
1956     BDRVVmdkState *s = bs->opaque;
1957     VmdkExtent *extent = NULL;
1958     int ret;
1959     int64_t offset_in_cluster, n_bytes;
1960     uint64_t cluster_offset;
1961     uint64_t bytes_done = 0;
1962     VmdkMetaData m_data;
1963 
1964     if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1965         error_report("Wrong offset: offset=0x%" PRIx64
1966                      " total_sectors=0x%" PRIx64,
1967                      offset, bs->total_sectors);
1968         return -EIO;
1969     }
1970 
1971     while (bytes > 0) {
1972         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1973         if (!extent) {
1974             return -EIO;
1975         }
1976         if (extent->sesparse) {
1977             return -ENOTSUP;
1978         }
1979         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1980         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1981                              - offset_in_cluster);
1982 
1983         ret = get_cluster_offset(bs, extent, &m_data, offset,
1984                                  !(extent->compressed || zeroed),
1985                                  &cluster_offset, offset_in_cluster,
1986                                  offset_in_cluster + n_bytes);
1987         if (extent->compressed) {
1988             if (ret == VMDK_OK) {
1989                 /* Refuse write to allocated cluster for streamOptimized */
1990                 error_report("Could not write to allocated cluster"
1991                               " for streamOptimized");
1992                 return -EIO;
1993             } else {
1994                 /* allocate */
1995                 ret = get_cluster_offset(bs, extent, &m_data, offset,
1996                                          true, &cluster_offset, 0, 0);
1997             }
1998         }
1999         if (ret == VMDK_ERROR) {
2000             return -EINVAL;
2001         }
2002         if (zeroed) {
2003             /* Do zeroed write, buf is ignored */
2004             if (extent->has_zero_grain &&
2005                     offset_in_cluster == 0 &&
2006                     n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
2007                 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
2008                 if (!zero_dry_run) {
2009                     /* update L2 tables */
2010                     if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
2011                             != VMDK_OK) {
2012                         return -EIO;
2013                     }
2014                 }
2015             } else {
2016                 return -ENOTSUP;
2017             }
2018         } else {
2019             ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
2020                                     qiov, bytes_done, n_bytes, offset);
2021             if (ret) {
2022                 return ret;
2023             }
2024             if (m_data.valid) {
2025                 /* update L2 tables */
2026                 if (vmdk_L2update(extent, &m_data,
2027                                   cluster_offset >> BDRV_SECTOR_BITS)
2028                         != VMDK_OK) {
2029                     return -EIO;
2030                 }
2031             }
2032         }
2033         bytes -= n_bytes;
2034         offset += n_bytes;
2035         bytes_done += n_bytes;
2036 
2037         /* update CID on the first write every time the virtual disk is
2038          * opened */
2039         if (!s->cid_updated) {
2040             ret = vmdk_write_cid(bs, g_random_int());
2041             if (ret < 0) {
2042                 return ret;
2043             }
2044             s->cid_updated = true;
2045         }
2046     }
2047     return 0;
2048 }
2049 
2050 static int coroutine_fn
2051 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2052                 QEMUIOVector *qiov, int flags)
2053 {
2054     int ret;
2055     BDRVVmdkState *s = bs->opaque;
2056     qemu_co_mutex_lock(&s->lock);
2057     ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
2058     qemu_co_mutex_unlock(&s->lock);
2059     return ret;
2060 }
2061 
2062 static int coroutine_fn
2063 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
2064                            uint64_t bytes, QEMUIOVector *qiov)
2065 {
2066     if (bytes == 0) {
2067         /* The caller will write bytes 0 to signal EOF.
2068          * When receive it, we align EOF to a sector boundary. */
2069         BDRVVmdkState *s = bs->opaque;
2070         int i, ret;
2071         int64_t length;
2072 
2073         for (i = 0; i < s->num_extents; i++) {
2074             length = bdrv_getlength(s->extents[i].file->bs);
2075             if (length < 0) {
2076                 return length;
2077             }
2078             length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
2079             ret = bdrv_truncate(s->extents[i].file, length, false,
2080                                 PREALLOC_MODE_OFF, NULL);
2081             if (ret < 0) {
2082                 return ret;
2083             }
2084         }
2085         return 0;
2086     }
2087     return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
2088 }
2089 
2090 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
2091                                               int64_t offset,
2092                                               int bytes,
2093                                               BdrvRequestFlags flags)
2094 {
2095     int ret;
2096     BDRVVmdkState *s = bs->opaque;
2097 
2098     qemu_co_mutex_lock(&s->lock);
2099     /* write zeroes could fail if sectors not aligned to cluster, test it with
2100      * dry_run == true before really updating image */
2101     ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
2102     if (!ret) {
2103         ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
2104     }
2105     qemu_co_mutex_unlock(&s->lock);
2106     return ret;
2107 }
2108 
2109 static int vmdk_init_extent(BlockBackend *blk,
2110                             int64_t filesize, bool flat,
2111                             bool compress, bool zeroed_grain,
2112                             Error **errp)
2113 {
2114     int ret, i;
2115     VMDK4Header header;
2116     uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
2117     uint32_t *gd_buf = NULL;
2118     int gd_buf_size;
2119 
2120     if (flat) {
2121         ret = blk_truncate(blk, filesize, false, PREALLOC_MODE_OFF, errp);
2122         goto exit;
2123     }
2124     magic = cpu_to_be32(VMDK4_MAGIC);
2125     memset(&header, 0, sizeof(header));
2126     if (compress) {
2127         header.version = 3;
2128     } else if (zeroed_grain) {
2129         header.version = 2;
2130     } else {
2131         header.version = 1;
2132     }
2133     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
2134                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
2135                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
2136     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
2137     header.capacity = filesize / BDRV_SECTOR_SIZE;
2138     header.granularity = 128;
2139     header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
2140 
2141     grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
2142     gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
2143                            BDRV_SECTOR_SIZE);
2144     gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
2145     gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
2146 
2147     header.desc_offset = 1;
2148     header.desc_size = 20;
2149     header.rgd_offset = header.desc_offset + header.desc_size;
2150     header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
2151     header.grain_offset =
2152         ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
2153                  header.granularity);
2154     /* swap endianness for all header fields */
2155     header.version = cpu_to_le32(header.version);
2156     header.flags = cpu_to_le32(header.flags);
2157     header.capacity = cpu_to_le64(header.capacity);
2158     header.granularity = cpu_to_le64(header.granularity);
2159     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
2160     header.desc_offset = cpu_to_le64(header.desc_offset);
2161     header.desc_size = cpu_to_le64(header.desc_size);
2162     header.rgd_offset = cpu_to_le64(header.rgd_offset);
2163     header.gd_offset = cpu_to_le64(header.gd_offset);
2164     header.grain_offset = cpu_to_le64(header.grain_offset);
2165     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
2166 
2167     header.check_bytes[0] = 0xa;
2168     header.check_bytes[1] = 0x20;
2169     header.check_bytes[2] = 0xd;
2170     header.check_bytes[3] = 0xa;
2171 
2172     /* write all the data */
2173     ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
2174     if (ret < 0) {
2175         error_setg(errp, QERR_IO_ERROR);
2176         goto exit;
2177     }
2178     ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
2179     if (ret < 0) {
2180         error_setg(errp, QERR_IO_ERROR);
2181         goto exit;
2182     }
2183 
2184     ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false,
2185                        PREALLOC_MODE_OFF, errp);
2186     if (ret < 0) {
2187         goto exit;
2188     }
2189 
2190     /* write grain directory */
2191     gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
2192     gd_buf = g_malloc0(gd_buf_size);
2193     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
2194          i < gt_count; i++, tmp += gt_size) {
2195         gd_buf[i] = cpu_to_le32(tmp);
2196     }
2197     ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
2198                      gd_buf, gd_buf_size, 0);
2199     if (ret < 0) {
2200         error_setg(errp, QERR_IO_ERROR);
2201         goto exit;
2202     }
2203 
2204     /* write backup grain directory */
2205     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
2206          i < gt_count; i++, tmp += gt_size) {
2207         gd_buf[i] = cpu_to_le32(tmp);
2208     }
2209     ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
2210                      gd_buf, gd_buf_size, 0);
2211     if (ret < 0) {
2212         error_setg(errp, QERR_IO_ERROR);
2213     }
2214 
2215     ret = 0;
2216 exit:
2217     g_free(gd_buf);
2218     return ret;
2219 }
2220 
2221 static int vmdk_create_extent(const char *filename, int64_t filesize,
2222                               bool flat, bool compress, bool zeroed_grain,
2223                               BlockBackend **pbb,
2224                               QemuOpts *opts, Error **errp)
2225 {
2226     int ret;
2227     BlockBackend *blk = NULL;
2228     Error *local_err = NULL;
2229 
2230     ret = bdrv_create_file(filename, opts, &local_err);
2231     if (ret < 0) {
2232         error_propagate(errp, local_err);
2233         goto exit;
2234     }
2235 
2236     blk = blk_new_open(filename, NULL, NULL,
2237                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2238                        &local_err);
2239     if (blk == NULL) {
2240         error_propagate(errp, local_err);
2241         ret = -EIO;
2242         goto exit;
2243     }
2244 
2245     blk_set_allow_write_beyond_eof(blk, true);
2246 
2247     ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
2248 exit:
2249     if (blk) {
2250         if (pbb) {
2251             *pbb = blk;
2252         } else {
2253             blk_unref(blk);
2254             blk = NULL;
2255         }
2256     }
2257     return ret;
2258 }
2259 
2260 static int filename_decompose(const char *filename, char *path, char *prefix,
2261                               char *postfix, size_t buf_len, Error **errp)
2262 {
2263     const char *p, *q;
2264 
2265     if (filename == NULL || !strlen(filename)) {
2266         error_setg(errp, "No filename provided");
2267         return VMDK_ERROR;
2268     }
2269     p = strrchr(filename, '/');
2270     if (p == NULL) {
2271         p = strrchr(filename, '\\');
2272     }
2273     if (p == NULL) {
2274         p = strrchr(filename, ':');
2275     }
2276     if (p != NULL) {
2277         p++;
2278         if (p - filename >= buf_len) {
2279             return VMDK_ERROR;
2280         }
2281         pstrcpy(path, p - filename + 1, filename);
2282     } else {
2283         p = filename;
2284         path[0] = '\0';
2285     }
2286     q = strrchr(p, '.');
2287     if (q == NULL) {
2288         pstrcpy(prefix, buf_len, p);
2289         postfix[0] = '\0';
2290     } else {
2291         if (q - p >= buf_len) {
2292             return VMDK_ERROR;
2293         }
2294         pstrcpy(prefix, q - p + 1, p);
2295         pstrcpy(postfix, buf_len, q);
2296     }
2297     return VMDK_OK;
2298 }
2299 
2300 /*
2301  * idx == 0: get or create the descriptor file (also the image file if in a
2302  *           non-split format.
2303  * idx >= 1: get the n-th extent if in a split subformat
2304  */
2305 typedef BlockBackend *(*vmdk_create_extent_fn)(int64_t size,
2306                                                int idx,
2307                                                bool flat,
2308                                                bool split,
2309                                                bool compress,
2310                                                bool zeroed_grain,
2311                                                void *opaque,
2312                                                Error **errp);
2313 
2314 static void vmdk_desc_add_extent(GString *desc,
2315                                  const char *extent_line_fmt,
2316                                  int64_t size, const char *filename)
2317 {
2318     char *basename = g_path_get_basename(filename);
2319 
2320     g_string_append_printf(desc, extent_line_fmt,
2321                            DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
2322     g_free(basename);
2323 }
2324 
2325 static int coroutine_fn vmdk_co_do_create(int64_t size,
2326                                           BlockdevVmdkSubformat subformat,
2327                                           BlockdevVmdkAdapterType adapter_type,
2328                                           const char *backing_file,
2329                                           const char *hw_version,
2330                                           bool compat6,
2331                                           bool zeroed_grain,
2332                                           vmdk_create_extent_fn extent_fn,
2333                                           void *opaque,
2334                                           Error **errp)
2335 {
2336     int extent_idx;
2337     BlockBackend *blk = NULL;
2338     BlockBackend *extent_blk;
2339     Error *local_err = NULL;
2340     char *desc = NULL;
2341     int ret = 0;
2342     bool flat, split, compress;
2343     GString *ext_desc_lines;
2344     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
2345     int64_t extent_size;
2346     int64_t created_size = 0;
2347     const char *extent_line_fmt;
2348     char *parent_desc_line = g_malloc0(BUF_SIZE);
2349     uint32_t parent_cid = 0xffffffff;
2350     uint32_t number_heads = 16;
2351     uint32_t desc_offset = 0, desc_len;
2352     const char desc_template[] =
2353         "# Disk DescriptorFile\n"
2354         "version=1\n"
2355         "CID=%" PRIx32 "\n"
2356         "parentCID=%" PRIx32 "\n"
2357         "createType=\"%s\"\n"
2358         "%s"
2359         "\n"
2360         "# Extent description\n"
2361         "%s"
2362         "\n"
2363         "# The Disk Data Base\n"
2364         "#DDB\n"
2365         "\n"
2366         "ddb.virtualHWVersion = \"%s\"\n"
2367         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2368         "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2369         "ddb.geometry.sectors = \"63\"\n"
2370         "ddb.adapterType = \"%s\"\n";
2371 
2372     ext_desc_lines = g_string_new(NULL);
2373 
2374     /* Read out options */
2375     if (compat6) {
2376         if (hw_version) {
2377             error_setg(errp,
2378                        "compat6 cannot be enabled with hwversion set");
2379             ret = -EINVAL;
2380             goto exit;
2381         }
2382         hw_version = "6";
2383     }
2384     if (!hw_version) {
2385         hw_version = "4";
2386     }
2387 
2388     if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2389         /* that's the number of heads with which vmware operates when
2390            creating, exporting, etc. vmdk files with a non-ide adapter type */
2391         number_heads = 255;
2392     }
2393     split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2394             (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2395     flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2396            (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2397     compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2398 
2399     if (flat) {
2400         extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2401     } else {
2402         extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2403     }
2404     if (flat && backing_file) {
2405         error_setg(errp, "Flat image can't have backing file");
2406         ret = -ENOTSUP;
2407         goto exit;
2408     }
2409     if (flat && zeroed_grain) {
2410         error_setg(errp, "Flat image can't enable zeroed grain");
2411         ret = -ENOTSUP;
2412         goto exit;
2413     }
2414 
2415     /* Create extents */
2416     if (split) {
2417         extent_size = split_size;
2418     } else {
2419         extent_size = size;
2420     }
2421     if (!split && !flat) {
2422         created_size = extent_size;
2423     } else {
2424         created_size = 0;
2425     }
2426     /* Get the descriptor file BDS */
2427     blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2428                     opaque, errp);
2429     if (!blk) {
2430         ret = -EIO;
2431         goto exit;
2432     }
2433     if (!split && !flat) {
2434         vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2435                              blk_bs(blk)->filename);
2436     }
2437 
2438     if (backing_file) {
2439         BlockBackend *backing;
2440         char *full_backing =
2441             bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2442                                                          backing_file,
2443                                                          &local_err);
2444         if (local_err) {
2445             error_propagate(errp, local_err);
2446             ret = -ENOENT;
2447             goto exit;
2448         }
2449         assert(full_backing);
2450 
2451         backing = blk_new_open(full_backing, NULL, NULL,
2452                                BDRV_O_NO_BACKING, errp);
2453         g_free(full_backing);
2454         if (backing == NULL) {
2455             ret = -EIO;
2456             goto exit;
2457         }
2458         if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2459             error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2460                        blk_bs(backing)->drv->format_name);
2461             blk_unref(backing);
2462             ret = -EINVAL;
2463             goto exit;
2464         }
2465         ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2466         blk_unref(backing);
2467         if (ret) {
2468             error_setg(errp, "Failed to read parent CID");
2469             goto exit;
2470         }
2471         snprintf(parent_desc_line, BUF_SIZE,
2472                 "parentFileNameHint=\"%s\"", backing_file);
2473     }
2474     extent_idx = 1;
2475     while (created_size < size) {
2476         int64_t cur_size = MIN(size - created_size, extent_size);
2477         extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2478                                zeroed_grain, opaque, errp);
2479         if (!extent_blk) {
2480             ret = -EINVAL;
2481             goto exit;
2482         }
2483         vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2484                              blk_bs(extent_blk)->filename);
2485         created_size += cur_size;
2486         extent_idx++;
2487         blk_unref(extent_blk);
2488     }
2489 
2490     /* Check whether we got excess extents */
2491     extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2492                            opaque, NULL);
2493     if (extent_blk) {
2494         blk_unref(extent_blk);
2495         error_setg(errp, "List of extents contains unused extents");
2496         ret = -EINVAL;
2497         goto exit;
2498     }
2499 
2500     /* generate descriptor file */
2501     desc = g_strdup_printf(desc_template,
2502                            g_random_int(),
2503                            parent_cid,
2504                            BlockdevVmdkSubformat_str(subformat),
2505                            parent_desc_line,
2506                            ext_desc_lines->str,
2507                            hw_version,
2508                            size /
2509                                (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2510                            number_heads,
2511                            BlockdevVmdkAdapterType_str(adapter_type));
2512     desc_len = strlen(desc);
2513     /* the descriptor offset = 0x200 */
2514     if (!split && !flat) {
2515         desc_offset = 0x200;
2516     }
2517 
2518     ret = blk_pwrite(blk, desc_offset, desc, desc_len, 0);
2519     if (ret < 0) {
2520         error_setg_errno(errp, -ret, "Could not write description");
2521         goto exit;
2522     }
2523     /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2524      * for description file */
2525     if (desc_offset == 0) {
2526         ret = blk_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, errp);
2527         if (ret < 0) {
2528             goto exit;
2529         }
2530     }
2531     ret = 0;
2532 exit:
2533     if (blk) {
2534         blk_unref(blk);
2535     }
2536     g_free(desc);
2537     g_free(parent_desc_line);
2538     g_string_free(ext_desc_lines, true);
2539     return ret;
2540 }
2541 
2542 typedef struct {
2543     char *path;
2544     char *prefix;
2545     char *postfix;
2546     QemuOpts *opts;
2547 } VMDKCreateOptsData;
2548 
2549 static BlockBackend *vmdk_co_create_opts_cb(int64_t size, int idx,
2550                                             bool flat, bool split, bool compress,
2551                                             bool zeroed_grain, void *opaque,
2552                                             Error **errp)
2553 {
2554     BlockBackend *blk = NULL;
2555     BlockDriverState *bs = NULL;
2556     VMDKCreateOptsData *data = opaque;
2557     char *ext_filename = NULL;
2558     char *rel_filename = NULL;
2559 
2560     /* We're done, don't create excess extents. */
2561     if (size == -1) {
2562         assert(errp == NULL);
2563         return NULL;
2564     }
2565 
2566     if (idx == 0) {
2567         rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2568     } else if (split) {
2569         rel_filename = g_strdup_printf("%s-%c%03d%s",
2570                                        data->prefix,
2571                                        flat ? 'f' : 's', idx, data->postfix);
2572     } else {
2573         assert(idx == 1);
2574         rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2575     }
2576 
2577     ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2578     g_free(rel_filename);
2579 
2580     if (vmdk_create_extent(ext_filename, size,
2581                            flat, compress, zeroed_grain, &blk, data->opts,
2582                            errp)) {
2583         goto exit;
2584     }
2585     bdrv_unref(bs);
2586 exit:
2587     g_free(ext_filename);
2588     return blk;
2589 }
2590 
2591 static int coroutine_fn vmdk_co_create_opts(BlockDriver *drv,
2592                                             const char *filename,
2593                                             QemuOpts *opts,
2594                                             Error **errp)
2595 {
2596     Error *local_err = NULL;
2597     char *desc = NULL;
2598     int64_t total_size = 0;
2599     char *adapter_type = NULL;
2600     BlockdevVmdkAdapterType adapter_type_enum;
2601     char *backing_file = NULL;
2602     char *hw_version = NULL;
2603     char *fmt = NULL;
2604     BlockdevVmdkSubformat subformat;
2605     int ret = 0;
2606     char *path = g_malloc0(PATH_MAX);
2607     char *prefix = g_malloc0(PATH_MAX);
2608     char *postfix = g_malloc0(PATH_MAX);
2609     char *desc_line = g_malloc0(BUF_SIZE);
2610     char *ext_filename = g_malloc0(PATH_MAX);
2611     char *desc_filename = g_malloc0(PATH_MAX);
2612     char *parent_desc_line = g_malloc0(BUF_SIZE);
2613     bool zeroed_grain;
2614     bool compat6;
2615     VMDKCreateOptsData data;
2616 
2617     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
2618         ret = -EINVAL;
2619         goto exit;
2620     }
2621     /* Read out options */
2622     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2623                           BDRV_SECTOR_SIZE);
2624     adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
2625     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2626     hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
2627     compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false);
2628     if (strcmp(hw_version, "undefined") == 0) {
2629         g_free(hw_version);
2630         hw_version = NULL;
2631     }
2632     fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
2633     zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false);
2634 
2635     if (adapter_type) {
2636         adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup,
2637                                             adapter_type,
2638                                             BLOCKDEV_VMDK_ADAPTER_TYPE_IDE,
2639                                             &local_err);
2640         if (local_err) {
2641             error_propagate(errp, local_err);
2642             ret = -EINVAL;
2643             goto exit;
2644         }
2645     } else {
2646         adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE;
2647     }
2648 
2649     if (!fmt) {
2650         /* Default format to monolithicSparse */
2651         subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE;
2652     } else {
2653         subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup,
2654                                     fmt,
2655                                     BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE,
2656                                     &local_err);
2657         if (local_err) {
2658             error_propagate(errp, local_err);
2659             ret = -EINVAL;
2660             goto exit;
2661         }
2662     }
2663     data = (VMDKCreateOptsData){
2664         .prefix = prefix,
2665         .postfix = postfix,
2666         .path = path,
2667         .opts = opts,
2668     };
2669     ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum,
2670                             backing_file, hw_version, compat6, zeroed_grain,
2671                             vmdk_co_create_opts_cb, &data, errp);
2672 
2673 exit:
2674     g_free(adapter_type);
2675     g_free(backing_file);
2676     g_free(hw_version);
2677     g_free(fmt);
2678     g_free(desc);
2679     g_free(path);
2680     g_free(prefix);
2681     g_free(postfix);
2682     g_free(desc_line);
2683     g_free(ext_filename);
2684     g_free(desc_filename);
2685     g_free(parent_desc_line);
2686     return ret;
2687 }
2688 
2689 static BlockBackend *vmdk_co_create_cb(int64_t size, int idx,
2690                                        bool flat, bool split, bool compress,
2691                                        bool zeroed_grain, void *opaque,
2692                                        Error **errp)
2693 {
2694     int ret;
2695     BlockDriverState *bs;
2696     BlockBackend *blk;
2697     BlockdevCreateOptionsVmdk *opts = opaque;
2698 
2699     if (idx == 0) {
2700         bs = bdrv_open_blockdev_ref(opts->file, errp);
2701     } else {
2702         int i;
2703         BlockdevRefList *list = opts->extents;
2704         for (i = 1; i < idx; i++) {
2705             if (!list || !list->next) {
2706                 error_setg(errp, "Extent [%d] not specified", i);
2707                 return NULL;
2708             }
2709             list = list->next;
2710         }
2711         if (!list) {
2712             error_setg(errp, "Extent [%d] not specified", idx - 1);
2713             return NULL;
2714         }
2715         bs = bdrv_open_blockdev_ref(list->value, errp);
2716     }
2717     if (!bs) {
2718         return NULL;
2719     }
2720     blk = blk_new(bdrv_get_aio_context(bs),
2721                   BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | BLK_PERM_RESIZE,
2722                   BLK_PERM_ALL);
2723     if (blk_insert_bs(blk, bs, errp)) {
2724         bdrv_unref(bs);
2725         return NULL;
2726     }
2727     blk_set_allow_write_beyond_eof(blk, true);
2728     bdrv_unref(bs);
2729 
2730     if (size != -1) {
2731         ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2732         if (ret) {
2733             blk_unref(blk);
2734             blk = NULL;
2735         }
2736     }
2737     return blk;
2738 }
2739 
2740 static int coroutine_fn vmdk_co_create(BlockdevCreateOptions *create_options,
2741                                        Error **errp)
2742 {
2743     int ret;
2744     BlockdevCreateOptionsVmdk *opts;
2745 
2746     opts = &create_options->u.vmdk;
2747 
2748     /* Validate options */
2749     if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2750         error_setg(errp, "Image size must be a multiple of 512 bytes");
2751         ret = -EINVAL;
2752         goto out;
2753     }
2754 
2755     ret = vmdk_co_do_create(opts->size,
2756                             opts->subformat,
2757                             opts->adapter_type,
2758                             opts->backing_file,
2759                             opts->hwversion,
2760                             false,
2761                             opts->zeroed_grain,
2762                             vmdk_co_create_cb,
2763                             opts, errp);
2764     return ret;
2765 
2766 out:
2767     return ret;
2768 }
2769 
2770 static void vmdk_close(BlockDriverState *bs)
2771 {
2772     BDRVVmdkState *s = bs->opaque;
2773 
2774     vmdk_free_extents(bs);
2775     g_free(s->create_type);
2776 
2777     migrate_del_blocker(s->migration_blocker);
2778     error_free(s->migration_blocker);
2779 }
2780 
2781 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2782 {
2783     BDRVVmdkState *s = bs->opaque;
2784     int i, err;
2785     int ret = 0;
2786 
2787     for (i = 0; i < s->num_extents; i++) {
2788         err = bdrv_co_flush(s->extents[i].file->bs);
2789         if (err < 0) {
2790             ret = err;
2791         }
2792     }
2793     return ret;
2794 }
2795 
2796 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2797 {
2798     int i;
2799     int64_t ret = 0;
2800     int64_t r;
2801     BDRVVmdkState *s = bs->opaque;
2802 
2803     ret = bdrv_get_allocated_file_size(bs->file->bs);
2804     if (ret < 0) {
2805         return ret;
2806     }
2807     for (i = 0; i < s->num_extents; i++) {
2808         if (s->extents[i].file == bs->file) {
2809             continue;
2810         }
2811         r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2812         if (r < 0) {
2813             return r;
2814         }
2815         ret += r;
2816     }
2817     return ret;
2818 }
2819 
2820 static int vmdk_has_zero_init(BlockDriverState *bs)
2821 {
2822     int i;
2823     BDRVVmdkState *s = bs->opaque;
2824 
2825     /* If has a flat extent and its underlying storage doesn't have zero init,
2826      * return 0. */
2827     for (i = 0; i < s->num_extents; i++) {
2828         if (s->extents[i].flat) {
2829             if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2830                 return 0;
2831             }
2832         }
2833     }
2834     return 1;
2835 }
2836 
2837 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2838 {
2839     ImageInfo *info = g_new0(ImageInfo, 1);
2840 
2841     bdrv_refresh_filename(extent->file->bs);
2842     *info = (ImageInfo){
2843         .filename         = g_strdup(extent->file->bs->filename),
2844         .format           = g_strdup(extent->type),
2845         .virtual_size     = extent->sectors * BDRV_SECTOR_SIZE,
2846         .compressed       = extent->compressed,
2847         .has_compressed   = extent->compressed,
2848         .cluster_size     = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2849         .has_cluster_size = !extent->flat,
2850     };
2851 
2852     return info;
2853 }
2854 
2855 static int coroutine_fn vmdk_co_check(BlockDriverState *bs,
2856                                       BdrvCheckResult *result,
2857                                       BdrvCheckMode fix)
2858 {
2859     BDRVVmdkState *s = bs->opaque;
2860     VmdkExtent *extent = NULL;
2861     int64_t sector_num = 0;
2862     int64_t total_sectors = bdrv_nb_sectors(bs);
2863     int ret;
2864     uint64_t cluster_offset;
2865 
2866     if (fix) {
2867         return -ENOTSUP;
2868     }
2869 
2870     for (;;) {
2871         if (sector_num >= total_sectors) {
2872             return 0;
2873         }
2874         extent = find_extent(s, sector_num, extent);
2875         if (!extent) {
2876             fprintf(stderr,
2877                     "ERROR: could not find extent for sector %" PRId64 "\n",
2878                     sector_num);
2879             ret = -EINVAL;
2880             break;
2881         }
2882         ret = get_cluster_offset(bs, extent, NULL,
2883                                  sector_num << BDRV_SECTOR_BITS,
2884                                  false, &cluster_offset, 0, 0);
2885         if (ret == VMDK_ERROR) {
2886             fprintf(stderr,
2887                     "ERROR: could not get cluster_offset for sector %"
2888                     PRId64 "\n", sector_num);
2889             break;
2890         }
2891         if (ret == VMDK_OK) {
2892             int64_t extent_len = bdrv_getlength(extent->file->bs);
2893             if (extent_len < 0) {
2894                 fprintf(stderr,
2895                         "ERROR: could not get extent file length for sector %"
2896                         PRId64 "\n", sector_num);
2897                 ret = extent_len;
2898                 break;
2899             }
2900             if (cluster_offset >= extent_len) {
2901                 fprintf(stderr,
2902                         "ERROR: cluster offset for sector %"
2903                         PRId64 " points after EOF\n", sector_num);
2904                 ret = -EINVAL;
2905                 break;
2906             }
2907         }
2908         sector_num += extent->cluster_sectors;
2909     }
2910 
2911     result->corruptions++;
2912     return ret;
2913 }
2914 
2915 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs,
2916                                                  Error **errp)
2917 {
2918     int i;
2919     BDRVVmdkState *s = bs->opaque;
2920     ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2921     ImageInfoList **next;
2922 
2923     *spec_info = (ImageInfoSpecific){
2924         .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2925         .u = {
2926             .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2927         },
2928     };
2929 
2930     *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2931         .create_type = g_strdup(s->create_type),
2932         .cid = s->cid,
2933         .parent_cid = s->parent_cid,
2934     };
2935 
2936     next = &spec_info->u.vmdk.data->extents;
2937     for (i = 0; i < s->num_extents; i++) {
2938         *next = g_new0(ImageInfoList, 1);
2939         (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2940         (*next)->next = NULL;
2941         next = &(*next)->next;
2942     }
2943 
2944     return spec_info;
2945 }
2946 
2947 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2948 {
2949     return a->flat == b->flat &&
2950            a->compressed == b->compressed &&
2951            (a->flat || a->cluster_sectors == b->cluster_sectors);
2952 }
2953 
2954 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2955 {
2956     int i;
2957     BDRVVmdkState *s = bs->opaque;
2958     assert(s->num_extents);
2959 
2960     /* See if we have multiple extents but they have different cases */
2961     for (i = 1; i < s->num_extents; i++) {
2962         if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2963             return -ENOTSUP;
2964         }
2965     }
2966     bdi->needs_compressed_writes = s->extents[0].compressed;
2967     if (!s->extents[0].flat) {
2968         bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2969     }
2970     return 0;
2971 }
2972 
2973 static void vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
2974                                       bool backing_overridden)
2975 {
2976     /* No children but file and backing can be explicitly specified (TODO) */
2977     qdict_put(target, "file",
2978               qobject_ref(bs->file->bs->full_open_options));
2979 
2980     if (backing_overridden) {
2981         if (bs->backing) {
2982             qdict_put(target, "backing",
2983                       qobject_ref(bs->backing->bs->full_open_options));
2984         } else {
2985             qdict_put_null(target, "backing");
2986         }
2987     }
2988 }
2989 
2990 static QemuOptsList vmdk_create_opts = {
2991     .name = "vmdk-create-opts",
2992     .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2993     .desc = {
2994         {
2995             .name = BLOCK_OPT_SIZE,
2996             .type = QEMU_OPT_SIZE,
2997             .help = "Virtual disk size"
2998         },
2999         {
3000             .name = BLOCK_OPT_ADAPTER_TYPE,
3001             .type = QEMU_OPT_STRING,
3002             .help = "Virtual adapter type, can be one of "
3003                     "ide (default), lsilogic, buslogic or legacyESX"
3004         },
3005         {
3006             .name = BLOCK_OPT_BACKING_FILE,
3007             .type = QEMU_OPT_STRING,
3008             .help = "File name of a base image"
3009         },
3010         {
3011             .name = BLOCK_OPT_COMPAT6,
3012             .type = QEMU_OPT_BOOL,
3013             .help = "VMDK version 6 image",
3014             .def_value_str = "off"
3015         },
3016         {
3017             .name = BLOCK_OPT_HWVERSION,
3018             .type = QEMU_OPT_STRING,
3019             .help = "VMDK hardware version",
3020             .def_value_str = "undefined"
3021         },
3022         {
3023             .name = BLOCK_OPT_SUBFMT,
3024             .type = QEMU_OPT_STRING,
3025             .help =
3026                 "VMDK flat extent format, can be one of "
3027                 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
3028         },
3029         {
3030             .name = BLOCK_OPT_ZEROED_GRAIN,
3031             .type = QEMU_OPT_BOOL,
3032             .help = "Enable efficient zero writes "
3033                     "using the zeroed-grain GTE feature"
3034         },
3035         { /* end of list */ }
3036     }
3037 };
3038 
3039 static BlockDriver bdrv_vmdk = {
3040     .format_name                  = "vmdk",
3041     .instance_size                = sizeof(BDRVVmdkState),
3042     .bdrv_probe                   = vmdk_probe,
3043     .bdrv_open                    = vmdk_open,
3044     .bdrv_co_check                = vmdk_co_check,
3045     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
3046     .bdrv_child_perm              = bdrv_format_default_perms,
3047     .bdrv_co_preadv               = vmdk_co_preadv,
3048     .bdrv_co_pwritev              = vmdk_co_pwritev,
3049     .bdrv_co_pwritev_compressed   = vmdk_co_pwritev_compressed,
3050     .bdrv_co_pwrite_zeroes        = vmdk_co_pwrite_zeroes,
3051     .bdrv_close                   = vmdk_close,
3052     .bdrv_co_create_opts          = vmdk_co_create_opts,
3053     .bdrv_co_create               = vmdk_co_create,
3054     .bdrv_co_flush_to_disk        = vmdk_co_flush,
3055     .bdrv_co_block_status         = vmdk_co_block_status,
3056     .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
3057     .bdrv_has_zero_init           = vmdk_has_zero_init,
3058     .bdrv_get_specific_info       = vmdk_get_specific_info,
3059     .bdrv_refresh_limits          = vmdk_refresh_limits,
3060     .bdrv_get_info                = vmdk_get_info,
3061     .bdrv_gather_child_options    = vmdk_gather_child_options,
3062 
3063     .supports_backing             = true,
3064     .create_opts                  = &vmdk_create_opts,
3065 };
3066 
3067 static void bdrv_vmdk_init(void)
3068 {
3069     bdrv_register(&bdrv_vmdk);
3070 }
3071 
3072 block_init(bdrv_vmdk_init);
3073