xref: /openbmc/qemu/block/vmdk.c (revision 9e9b10c6)
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[0];
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                               const char *desc_file_path, QDict *options,
1080                               Error **errp)
1081 {
1082     int ret;
1083     int matches;
1084     char access[11];
1085     char type[11];
1086     char fname[512];
1087     const char *p, *np;
1088     int64_t sectors = 0;
1089     int64_t flat_offset;
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) && !path_has_protocol(fname) &&
1134             !desc_file_path[0])
1135         {
1136             bdrv_refresh_filename(bs->file->bs);
1137             error_setg(errp, "Cannot use relative extent paths with VMDK "
1138                        "descriptor file '%s'", bs->file->bs->filename);
1139             return -EINVAL;
1140         }
1141 
1142         extent_path = path_combine(desc_file_path, fname);
1143 
1144         ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
1145         assert(ret < 32);
1146 
1147         extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
1148                                       bs, &child_file, false, &local_err);
1149         g_free(extent_path);
1150         if (local_err) {
1151             error_propagate(errp, local_err);
1152             return -EINVAL;
1153         }
1154 
1155         /* save to extents array */
1156         if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
1157             /* FLAT extent */
1158 
1159             ret = vmdk_add_extent(bs, extent_file, true, sectors,
1160                             0, 0, 0, 0, 0, &extent, errp);
1161             if (ret < 0) {
1162                 bdrv_unref_child(bs, extent_file);
1163                 return ret;
1164             }
1165             extent->flat_start_offset = flat_offset << 9;
1166         } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
1167             /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
1168             char *buf = vmdk_read_desc(extent_file, 0, errp);
1169             if (!buf) {
1170                 ret = -EINVAL;
1171             } else {
1172                 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
1173                                        options, errp);
1174             }
1175             g_free(buf);
1176             if (ret) {
1177                 bdrv_unref_child(bs, extent_file);
1178                 return ret;
1179             }
1180             extent = &s->extents[s->num_extents - 1];
1181         } else if (!strcmp(type, "SESPARSE")) {
1182             ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
1183             if (ret) {
1184                 bdrv_unref_child(bs, extent_file);
1185                 return ret;
1186             }
1187             extent = &s->extents[s->num_extents - 1];
1188         } else {
1189             error_setg(errp, "Unsupported extent type '%s'", type);
1190             bdrv_unref_child(bs, extent_file);
1191             return -ENOTSUP;
1192         }
1193         extent->type = g_strdup(type);
1194     }
1195     return 0;
1196 
1197 invalid:
1198     np = next_line(p);
1199     assert(np != p);
1200     if (np[-1] == '\n') {
1201         np--;
1202     }
1203     error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
1204     return -EINVAL;
1205 }
1206 
1207 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
1208                                QDict *options, Error **errp)
1209 {
1210     int ret;
1211     char ct[128];
1212     BDRVVmdkState *s = bs->opaque;
1213 
1214     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
1215         error_setg(errp, "invalid VMDK image descriptor");
1216         ret = -EINVAL;
1217         goto exit;
1218     }
1219     if (strcmp(ct, "monolithicFlat") &&
1220         strcmp(ct, "vmfs") &&
1221         strcmp(ct, "vmfsSparse") &&
1222         strcmp(ct, "seSparse") &&
1223         strcmp(ct, "twoGbMaxExtentSparse") &&
1224         strcmp(ct, "twoGbMaxExtentFlat")) {
1225         error_setg(errp, "Unsupported image type '%s'", ct);
1226         ret = -ENOTSUP;
1227         goto exit;
1228     }
1229     s->create_type = g_strdup(ct);
1230     s->desc_offset = 0;
1231     ret = vmdk_parse_extents(buf, bs, bs->file->bs->exact_filename, options,
1232                              errp);
1233 exit:
1234     return ret;
1235 }
1236 
1237 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
1238                      Error **errp)
1239 {
1240     char *buf;
1241     int ret;
1242     BDRVVmdkState *s = bs->opaque;
1243     uint32_t magic;
1244     Error *local_err = NULL;
1245 
1246     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
1247                                false, errp);
1248     if (!bs->file) {
1249         return -EINVAL;
1250     }
1251 
1252     buf = vmdk_read_desc(bs->file, 0, errp);
1253     if (!buf) {
1254         return -EINVAL;
1255     }
1256 
1257     magic = ldl_be_p(buf);
1258     switch (magic) {
1259         case VMDK3_MAGIC:
1260         case VMDK4_MAGIC:
1261             ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
1262                                    errp);
1263             s->desc_offset = 0x200;
1264             break;
1265         default:
1266             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1267             break;
1268     }
1269     if (ret) {
1270         goto fail;
1271     }
1272 
1273     /* try to open parent images, if exist */
1274     ret = vmdk_parent_open(bs);
1275     if (ret) {
1276         goto fail;
1277     }
1278     ret = vmdk_read_cid(bs, 0, &s->cid);
1279     if (ret) {
1280         goto fail;
1281     }
1282     ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1283     if (ret) {
1284         goto fail;
1285     }
1286     qemu_co_mutex_init(&s->lock);
1287 
1288     /* Disable migration when VMDK images are used */
1289     error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1290                "does not support live migration",
1291                bdrv_get_device_or_node_name(bs));
1292     ret = migrate_add_blocker(s->migration_blocker, &local_err);
1293     if (local_err) {
1294         error_propagate(errp, local_err);
1295         error_free(s->migration_blocker);
1296         goto fail;
1297     }
1298 
1299     g_free(buf);
1300     return 0;
1301 
1302 fail:
1303     g_free(buf);
1304     g_free(s->create_type);
1305     s->create_type = NULL;
1306     vmdk_free_extents(bs);
1307     return ret;
1308 }
1309 
1310 
1311 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1312 {
1313     BDRVVmdkState *s = bs->opaque;
1314     int i;
1315 
1316     for (i = 0; i < s->num_extents; i++) {
1317         if (!s->extents[i].flat) {
1318             bs->bl.pwrite_zeroes_alignment =
1319                 MAX(bs->bl.pwrite_zeroes_alignment,
1320                     s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1321         }
1322     }
1323 }
1324 
1325 /**
1326  * get_whole_cluster
1327  *
1328  * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1329  * to the cluster at @cluster_sector_num.
1330  *
1331  * If @skip_start_sector < @skip_end_sector, the relative range
1332  * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1333  * it for call to write user data in the request.
1334  */
1335 static int get_whole_cluster(BlockDriverState *bs,
1336                              VmdkExtent *extent,
1337                              uint64_t cluster_offset,
1338                              uint64_t offset,
1339                              uint64_t skip_start_bytes,
1340                              uint64_t skip_end_bytes)
1341 {
1342     int ret = VMDK_OK;
1343     int64_t cluster_bytes;
1344     uint8_t *whole_grain;
1345 
1346     /* For COW, align request sector_num to cluster start */
1347     cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1348     offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1349     whole_grain = qemu_blockalign(bs, cluster_bytes);
1350 
1351     if (!bs->backing) {
1352         memset(whole_grain, 0, skip_start_bytes);
1353         memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1354     }
1355 
1356     assert(skip_end_bytes <= cluster_bytes);
1357     /* we will be here if it's first write on non-exist grain(cluster).
1358      * try to read from parent image, if exist */
1359     if (bs->backing && !vmdk_is_cid_valid(bs)) {
1360         ret = VMDK_ERROR;
1361         goto exit;
1362     }
1363 
1364     /* Read backing data before skip range */
1365     if (skip_start_bytes > 0) {
1366         if (bs->backing) {
1367             /* qcow2 emits this on bs->file instead of bs->backing */
1368             BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1369             ret = bdrv_pread(bs->backing, offset, whole_grain,
1370                              skip_start_bytes);
1371             if (ret < 0) {
1372                 ret = VMDK_ERROR;
1373                 goto exit;
1374             }
1375         }
1376         BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1377         ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1378                           skip_start_bytes);
1379         if (ret < 0) {
1380             ret = VMDK_ERROR;
1381             goto exit;
1382         }
1383     }
1384     /* Read backing data after skip range */
1385     if (skip_end_bytes < cluster_bytes) {
1386         if (bs->backing) {
1387             /* qcow2 emits this on bs->file instead of bs->backing */
1388             BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1389             ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1390                              whole_grain + skip_end_bytes,
1391                              cluster_bytes - skip_end_bytes);
1392             if (ret < 0) {
1393                 ret = VMDK_ERROR;
1394                 goto exit;
1395             }
1396         }
1397         BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1398         ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1399                           whole_grain + skip_end_bytes,
1400                           cluster_bytes - skip_end_bytes);
1401         if (ret < 0) {
1402             ret = VMDK_ERROR;
1403             goto exit;
1404         }
1405     }
1406 
1407     ret = VMDK_OK;
1408 exit:
1409     qemu_vfree(whole_grain);
1410     return ret;
1411 }
1412 
1413 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1414                          uint32_t offset)
1415 {
1416     offset = cpu_to_le32(offset);
1417     /* update L2 table */
1418     BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1419     if (bdrv_pwrite_sync(extent->file,
1420                 ((int64_t)m_data->l2_offset * 512)
1421                     + (m_data->l2_index * sizeof(offset)),
1422                 &offset, sizeof(offset)) < 0) {
1423         return VMDK_ERROR;
1424     }
1425     /* update backup L2 table */
1426     if (extent->l1_backup_table_offset != 0) {
1427         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1428         if (bdrv_pwrite_sync(extent->file,
1429                     ((int64_t)m_data->l2_offset * 512)
1430                         + (m_data->l2_index * sizeof(offset)),
1431                     &offset, sizeof(offset)) < 0) {
1432             return VMDK_ERROR;
1433         }
1434     }
1435     if (m_data->l2_cache_entry) {
1436         *m_data->l2_cache_entry = offset;
1437     }
1438 
1439     return VMDK_OK;
1440 }
1441 
1442 /**
1443  * get_cluster_offset
1444  *
1445  * Look up cluster offset in extent file by sector number, and store in
1446  * @cluster_offset.
1447  *
1448  * For flat extents, the start offset as parsed from the description file is
1449  * returned.
1450  *
1451  * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1452  * offset for a new cluster and update L2 cache. If there is a backing file,
1453  * COW is done before returning; otherwise, zeroes are written to the allocated
1454  * cluster. Both COW and zero writing skips the sector range
1455  * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1456  * has new data to write there.
1457  *
1458  * Returns: VMDK_OK if cluster exists and mapped in the image.
1459  *          VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1460  *          VMDK_ERROR if failed.
1461  */
1462 static int get_cluster_offset(BlockDriverState *bs,
1463                               VmdkExtent *extent,
1464                               VmdkMetaData *m_data,
1465                               uint64_t offset,
1466                               bool allocate,
1467                               uint64_t *cluster_offset,
1468                               uint64_t skip_start_bytes,
1469                               uint64_t skip_end_bytes)
1470 {
1471     unsigned int l1_index, l2_offset, l2_index;
1472     int min_index, i, j;
1473     uint32_t min_count;
1474     void *l2_table;
1475     bool zeroed = false;
1476     int64_t ret;
1477     int64_t cluster_sector;
1478     unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
1479 
1480     if (m_data) {
1481         m_data->valid = 0;
1482     }
1483     if (extent->flat) {
1484         *cluster_offset = extent->flat_start_offset;
1485         return VMDK_OK;
1486     }
1487 
1488     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1489     l1_index = (offset >> 9) / extent->l1_entry_sectors;
1490     if (l1_index >= extent->l1_size) {
1491         return VMDK_ERROR;
1492     }
1493     if (extent->sesparse) {
1494         uint64_t l2_offset_u64;
1495 
1496         assert(extent->entry_size == sizeof(uint64_t));
1497 
1498         l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
1499         if (l2_offset_u64 == 0) {
1500             l2_offset = 0;
1501         } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
1502             /*
1503              * Top most nibble is 0x1 if grain table is allocated.
1504              * strict check - top most 4 bytes must be 0x10000000 since max
1505              * supported size is 64TB for disk - so no more than 64TB / 16MB
1506              * grain directories which is smaller than uint32,
1507              * where 16MB is the only supported default grain table coverage.
1508              */
1509             return VMDK_ERROR;
1510         } else {
1511             l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
1512             l2_offset_u64 = extent->sesparse_l2_tables_offset +
1513                 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
1514             if (l2_offset_u64 > 0x00000000ffffffff) {
1515                 return VMDK_ERROR;
1516             }
1517             l2_offset = (unsigned int)(l2_offset_u64);
1518         }
1519     } else {
1520         assert(extent->entry_size == sizeof(uint32_t));
1521         l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
1522     }
1523     if (!l2_offset) {
1524         return VMDK_UNALLOC;
1525     }
1526     for (i = 0; i < L2_CACHE_SIZE; i++) {
1527         if (l2_offset == extent->l2_cache_offsets[i]) {
1528             /* increment the hit count */
1529             if (++extent->l2_cache_counts[i] == 0xffffffff) {
1530                 for (j = 0; j < L2_CACHE_SIZE; j++) {
1531                     extent->l2_cache_counts[j] >>= 1;
1532                 }
1533             }
1534             l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
1535             goto found;
1536         }
1537     }
1538     /* not found: load a new entry in the least used one */
1539     min_index = 0;
1540     min_count = 0xffffffff;
1541     for (i = 0; i < L2_CACHE_SIZE; i++) {
1542         if (extent->l2_cache_counts[i] < min_count) {
1543             min_count = extent->l2_cache_counts[i];
1544             min_index = i;
1545         }
1546     }
1547     l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
1548     BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1549     if (bdrv_pread(extent->file,
1550                 (int64_t)l2_offset * 512,
1551                 l2_table,
1552                 l2_size_bytes
1553             ) != l2_size_bytes) {
1554         return VMDK_ERROR;
1555     }
1556 
1557     extent->l2_cache_offsets[min_index] = l2_offset;
1558     extent->l2_cache_counts[min_index] = 1;
1559  found:
1560     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1561 
1562     if (extent->sesparse) {
1563         cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
1564         switch (cluster_sector & 0xf000000000000000) {
1565         case 0x0000000000000000:
1566             /* unallocated grain */
1567             if (cluster_sector != 0) {
1568                 return VMDK_ERROR;
1569             }
1570             break;
1571         case 0x1000000000000000:
1572             /* scsi-unmapped grain - fallthrough */
1573         case 0x2000000000000000:
1574             /* zero grain */
1575             zeroed = true;
1576             break;
1577         case 0x3000000000000000:
1578             /* allocated grain */
1579             cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
1580                               ((cluster_sector & 0x0000ffffffffffff) << 12));
1581             cluster_sector = extent->sesparse_clusters_offset +
1582                 cluster_sector * extent->cluster_sectors;
1583             break;
1584         default:
1585             return VMDK_ERROR;
1586         }
1587     } else {
1588         cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
1589 
1590         if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1591             zeroed = true;
1592         }
1593     }
1594 
1595     if (!cluster_sector || zeroed) {
1596         if (!allocate) {
1597             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1598         }
1599         assert(!extent->sesparse);
1600 
1601         if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1602             return VMDK_ERROR;
1603         }
1604 
1605         cluster_sector = extent->next_cluster_sector;
1606         extent->next_cluster_sector += extent->cluster_sectors;
1607 
1608         /* First of all we write grain itself, to avoid race condition
1609          * that may to corrupt the image.
1610          * This problem may occur because of insufficient space on host disk
1611          * or inappropriate VM shutdown.
1612          */
1613         ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1614                                 offset, skip_start_bytes, skip_end_bytes);
1615         if (ret) {
1616             return ret;
1617         }
1618         if (m_data) {
1619             m_data->valid = 1;
1620             m_data->l1_index = l1_index;
1621             m_data->l2_index = l2_index;
1622             m_data->l2_offset = l2_offset;
1623             m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
1624         }
1625     }
1626     *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1627     return VMDK_OK;
1628 }
1629 
1630 static VmdkExtent *find_extent(BDRVVmdkState *s,
1631                                 int64_t sector_num, VmdkExtent *start_hint)
1632 {
1633     VmdkExtent *extent = start_hint;
1634 
1635     if (!extent) {
1636         extent = &s->extents[0];
1637     }
1638     while (extent < &s->extents[s->num_extents]) {
1639         if (sector_num < extent->end_sector) {
1640             return extent;
1641         }
1642         extent++;
1643     }
1644     return NULL;
1645 }
1646 
1647 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1648                                                    int64_t offset)
1649 {
1650     uint64_t extent_begin_offset, extent_relative_offset;
1651     uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1652 
1653     extent_begin_offset =
1654         (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1655     extent_relative_offset = offset - extent_begin_offset;
1656     return extent_relative_offset % cluster_size;
1657 }
1658 
1659 static int coroutine_fn vmdk_co_block_status(BlockDriverState *bs,
1660                                              bool want_zero,
1661                                              int64_t offset, int64_t bytes,
1662                                              int64_t *pnum, int64_t *map,
1663                                              BlockDriverState **file)
1664 {
1665     BDRVVmdkState *s = bs->opaque;
1666     int64_t index_in_cluster, n, ret;
1667     uint64_t cluster_offset;
1668     VmdkExtent *extent;
1669 
1670     extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1671     if (!extent) {
1672         return -EIO;
1673     }
1674     qemu_co_mutex_lock(&s->lock);
1675     ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1676                              0, 0);
1677     qemu_co_mutex_unlock(&s->lock);
1678 
1679     index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1680     switch (ret) {
1681     case VMDK_ERROR:
1682         ret = -EIO;
1683         break;
1684     case VMDK_UNALLOC:
1685         ret = 0;
1686         break;
1687     case VMDK_ZEROED:
1688         ret = BDRV_BLOCK_ZERO;
1689         break;
1690     case VMDK_OK:
1691         ret = BDRV_BLOCK_DATA;
1692         if (!extent->compressed) {
1693             ret |= BDRV_BLOCK_OFFSET_VALID;
1694             *map = cluster_offset + index_in_cluster;
1695             if (extent->flat) {
1696                 ret |= BDRV_BLOCK_RECURSE;
1697             }
1698         }
1699         *file = extent->file->bs;
1700         break;
1701     }
1702 
1703     n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1704     *pnum = MIN(n, bytes);
1705     return ret;
1706 }
1707 
1708 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1709                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1710                             uint64_t qiov_offset, uint64_t n_bytes,
1711                             uint64_t offset)
1712 {
1713     int ret;
1714     VmdkGrainMarker *data = NULL;
1715     uLongf buf_len;
1716     QEMUIOVector local_qiov;
1717     int64_t write_offset;
1718     int64_t write_end_sector;
1719 
1720     if (extent->compressed) {
1721         void *compressed_data;
1722 
1723         if (!extent->has_marker) {
1724             ret = -EINVAL;
1725             goto out;
1726         }
1727         buf_len = (extent->cluster_sectors << 9) * 2;
1728         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1729 
1730         compressed_data = g_malloc(n_bytes);
1731         qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1732         ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1733         g_free(compressed_data);
1734 
1735         if (ret != Z_OK || buf_len == 0) {
1736             ret = -EINVAL;
1737             goto out;
1738         }
1739 
1740         data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1741         data->size = cpu_to_le32(buf_len);
1742 
1743         n_bytes = buf_len + sizeof(VmdkGrainMarker);
1744         qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1745 
1746         BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1747     } else {
1748         qemu_iovec_init(&local_qiov, qiov->niov);
1749         qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1750 
1751         BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1752     }
1753 
1754     write_offset = cluster_offset + offset_in_cluster;
1755     ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1756                           &local_qiov, 0);
1757 
1758     write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1759 
1760     if (extent->compressed) {
1761         extent->next_cluster_sector = write_end_sector;
1762     } else {
1763         extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1764                                           write_end_sector);
1765     }
1766 
1767     if (ret < 0) {
1768         goto out;
1769     }
1770     ret = 0;
1771  out:
1772     g_free(data);
1773     if (!extent->compressed) {
1774         qemu_iovec_destroy(&local_qiov);
1775     }
1776     return ret;
1777 }
1778 
1779 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1780                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1781                             int bytes)
1782 {
1783     int ret;
1784     int cluster_bytes, buf_bytes;
1785     uint8_t *cluster_buf, *compressed_data;
1786     uint8_t *uncomp_buf;
1787     uint32_t data_len;
1788     VmdkGrainMarker *marker;
1789     uLongf buf_len;
1790 
1791 
1792     if (!extent->compressed) {
1793         BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1794         ret = bdrv_co_preadv(extent->file,
1795                              cluster_offset + offset_in_cluster, bytes,
1796                              qiov, 0);
1797         if (ret < 0) {
1798             return ret;
1799         }
1800         return 0;
1801     }
1802     cluster_bytes = extent->cluster_sectors * 512;
1803     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1804     buf_bytes = cluster_bytes * 2;
1805     cluster_buf = g_malloc(buf_bytes);
1806     uncomp_buf = g_malloc(cluster_bytes);
1807     BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1808     ret = bdrv_pread(extent->file,
1809                 cluster_offset,
1810                 cluster_buf, buf_bytes);
1811     if (ret < 0) {
1812         goto out;
1813     }
1814     compressed_data = cluster_buf;
1815     buf_len = cluster_bytes;
1816     data_len = cluster_bytes;
1817     if (extent->has_marker) {
1818         marker = (VmdkGrainMarker *)cluster_buf;
1819         compressed_data = marker->data;
1820         data_len = le32_to_cpu(marker->size);
1821     }
1822     if (!data_len || data_len > buf_bytes) {
1823         ret = -EINVAL;
1824         goto out;
1825     }
1826     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1827     if (ret != Z_OK) {
1828         ret = -EINVAL;
1829         goto out;
1830 
1831     }
1832     if (offset_in_cluster < 0 ||
1833             offset_in_cluster + bytes > buf_len) {
1834         ret = -EINVAL;
1835         goto out;
1836     }
1837     qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1838     ret = 0;
1839 
1840  out:
1841     g_free(uncomp_buf);
1842     g_free(cluster_buf);
1843     return ret;
1844 }
1845 
1846 static int coroutine_fn
1847 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1848                QEMUIOVector *qiov, int flags)
1849 {
1850     BDRVVmdkState *s = bs->opaque;
1851     int ret;
1852     uint64_t n_bytes, offset_in_cluster;
1853     VmdkExtent *extent = NULL;
1854     QEMUIOVector local_qiov;
1855     uint64_t cluster_offset;
1856     uint64_t bytes_done = 0;
1857 
1858     qemu_iovec_init(&local_qiov, qiov->niov);
1859     qemu_co_mutex_lock(&s->lock);
1860 
1861     while (bytes > 0) {
1862         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1863         if (!extent) {
1864             ret = -EIO;
1865             goto fail;
1866         }
1867         ret = get_cluster_offset(bs, extent, NULL,
1868                                  offset, false, &cluster_offset, 0, 0);
1869         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1870 
1871         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1872                              - offset_in_cluster);
1873 
1874         if (ret != VMDK_OK) {
1875             /* if not allocated, try to read from parent image, if exist */
1876             if (bs->backing && ret != VMDK_ZEROED) {
1877                 if (!vmdk_is_cid_valid(bs)) {
1878                     ret = -EINVAL;
1879                     goto fail;
1880                 }
1881 
1882                 qemu_iovec_reset(&local_qiov);
1883                 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1884 
1885                 /* qcow2 emits this on bs->file instead of bs->backing */
1886                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1887                 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1888                                      &local_qiov, 0);
1889                 if (ret < 0) {
1890                     goto fail;
1891                 }
1892             } else {
1893                 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1894             }
1895         } else {
1896             qemu_iovec_reset(&local_qiov);
1897             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1898 
1899             ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1900                                    &local_qiov, n_bytes);
1901             if (ret) {
1902                 goto fail;
1903             }
1904         }
1905         bytes -= n_bytes;
1906         offset += n_bytes;
1907         bytes_done += n_bytes;
1908     }
1909 
1910     ret = 0;
1911 fail:
1912     qemu_co_mutex_unlock(&s->lock);
1913     qemu_iovec_destroy(&local_qiov);
1914 
1915     return ret;
1916 }
1917 
1918 /**
1919  * vmdk_write:
1920  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
1921  *                if possible, otherwise return -ENOTSUP.
1922  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1923  *                with each cluster. By dry run we can find if the zero write
1924  *                is possible without modifying image data.
1925  *
1926  * Returns: error code with 0 for success.
1927  */
1928 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1929                        uint64_t bytes, QEMUIOVector *qiov,
1930                        bool zeroed, bool zero_dry_run)
1931 {
1932     BDRVVmdkState *s = bs->opaque;
1933     VmdkExtent *extent = NULL;
1934     int ret;
1935     int64_t offset_in_cluster, n_bytes;
1936     uint64_t cluster_offset;
1937     uint64_t bytes_done = 0;
1938     VmdkMetaData m_data;
1939 
1940     if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1941         error_report("Wrong offset: offset=0x%" PRIx64
1942                      " total_sectors=0x%" PRIx64,
1943                      offset, bs->total_sectors);
1944         return -EIO;
1945     }
1946 
1947     while (bytes > 0) {
1948         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1949         if (!extent) {
1950             return -EIO;
1951         }
1952         if (extent->sesparse) {
1953             return -ENOTSUP;
1954         }
1955         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1956         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1957                              - offset_in_cluster);
1958 
1959         ret = get_cluster_offset(bs, extent, &m_data, offset,
1960                                  !(extent->compressed || zeroed),
1961                                  &cluster_offset, offset_in_cluster,
1962                                  offset_in_cluster + n_bytes);
1963         if (extent->compressed) {
1964             if (ret == VMDK_OK) {
1965                 /* Refuse write to allocated cluster for streamOptimized */
1966                 error_report("Could not write to allocated cluster"
1967                               " for streamOptimized");
1968                 return -EIO;
1969             } else {
1970                 /* allocate */
1971                 ret = get_cluster_offset(bs, extent, &m_data, offset,
1972                                          true, &cluster_offset, 0, 0);
1973             }
1974         }
1975         if (ret == VMDK_ERROR) {
1976             return -EINVAL;
1977         }
1978         if (zeroed) {
1979             /* Do zeroed write, buf is ignored */
1980             if (extent->has_zero_grain &&
1981                     offset_in_cluster == 0 &&
1982                     n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
1983                 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1984                 if (!zero_dry_run) {
1985                     /* update L2 tables */
1986                     if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1987                             != VMDK_OK) {
1988                         return -EIO;
1989                     }
1990                 }
1991             } else {
1992                 return -ENOTSUP;
1993             }
1994         } else {
1995             ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
1996                                     qiov, bytes_done, n_bytes, offset);
1997             if (ret) {
1998                 return ret;
1999             }
2000             if (m_data.valid) {
2001                 /* update L2 tables */
2002                 if (vmdk_L2update(extent, &m_data,
2003                                   cluster_offset >> BDRV_SECTOR_BITS)
2004                         != VMDK_OK) {
2005                     return -EIO;
2006                 }
2007             }
2008         }
2009         bytes -= n_bytes;
2010         offset += n_bytes;
2011         bytes_done += n_bytes;
2012 
2013         /* update CID on the first write every time the virtual disk is
2014          * opened */
2015         if (!s->cid_updated) {
2016             ret = vmdk_write_cid(bs, g_random_int());
2017             if (ret < 0) {
2018                 return ret;
2019             }
2020             s->cid_updated = true;
2021         }
2022     }
2023     return 0;
2024 }
2025 
2026 static int coroutine_fn
2027 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2028                 QEMUIOVector *qiov, int flags)
2029 {
2030     int ret;
2031     BDRVVmdkState *s = bs->opaque;
2032     qemu_co_mutex_lock(&s->lock);
2033     ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
2034     qemu_co_mutex_unlock(&s->lock);
2035     return ret;
2036 }
2037 
2038 static int coroutine_fn
2039 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
2040                            uint64_t bytes, QEMUIOVector *qiov)
2041 {
2042     if (bytes == 0) {
2043         /* The caller will write bytes 0 to signal EOF.
2044          * When receive it, we align EOF to a sector boundary. */
2045         BDRVVmdkState *s = bs->opaque;
2046         int i, ret;
2047         int64_t length;
2048 
2049         for (i = 0; i < s->num_extents; i++) {
2050             length = bdrv_getlength(s->extents[i].file->bs);
2051             if (length < 0) {
2052                 return length;
2053             }
2054             length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
2055             ret = bdrv_truncate(s->extents[i].file, length,
2056                                 PREALLOC_MODE_OFF, NULL);
2057             if (ret < 0) {
2058                 return ret;
2059             }
2060         }
2061         return 0;
2062     }
2063     return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
2064 }
2065 
2066 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
2067                                               int64_t offset,
2068                                               int bytes,
2069                                               BdrvRequestFlags flags)
2070 {
2071     int ret;
2072     BDRVVmdkState *s = bs->opaque;
2073 
2074     qemu_co_mutex_lock(&s->lock);
2075     /* write zeroes could fail if sectors not aligned to cluster, test it with
2076      * dry_run == true before really updating image */
2077     ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
2078     if (!ret) {
2079         ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
2080     }
2081     qemu_co_mutex_unlock(&s->lock);
2082     return ret;
2083 }
2084 
2085 static int vmdk_init_extent(BlockBackend *blk,
2086                             int64_t filesize, bool flat,
2087                             bool compress, bool zeroed_grain,
2088                             Error **errp)
2089 {
2090     int ret, i;
2091     VMDK4Header header;
2092     uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
2093     uint32_t *gd_buf = NULL;
2094     int gd_buf_size;
2095 
2096     if (flat) {
2097         ret = blk_truncate(blk, filesize, PREALLOC_MODE_OFF, errp);
2098         goto exit;
2099     }
2100     magic = cpu_to_be32(VMDK4_MAGIC);
2101     memset(&header, 0, sizeof(header));
2102     if (compress) {
2103         header.version = 3;
2104     } else if (zeroed_grain) {
2105         header.version = 2;
2106     } else {
2107         header.version = 1;
2108     }
2109     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
2110                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
2111                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
2112     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
2113     header.capacity = filesize / BDRV_SECTOR_SIZE;
2114     header.granularity = 128;
2115     header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
2116 
2117     grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
2118     gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
2119                            BDRV_SECTOR_SIZE);
2120     gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
2121     gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
2122 
2123     header.desc_offset = 1;
2124     header.desc_size = 20;
2125     header.rgd_offset = header.desc_offset + header.desc_size;
2126     header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
2127     header.grain_offset =
2128         ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
2129                  header.granularity);
2130     /* swap endianness for all header fields */
2131     header.version = cpu_to_le32(header.version);
2132     header.flags = cpu_to_le32(header.flags);
2133     header.capacity = cpu_to_le64(header.capacity);
2134     header.granularity = cpu_to_le64(header.granularity);
2135     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
2136     header.desc_offset = cpu_to_le64(header.desc_offset);
2137     header.desc_size = cpu_to_le64(header.desc_size);
2138     header.rgd_offset = cpu_to_le64(header.rgd_offset);
2139     header.gd_offset = cpu_to_le64(header.gd_offset);
2140     header.grain_offset = cpu_to_le64(header.grain_offset);
2141     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
2142 
2143     header.check_bytes[0] = 0xa;
2144     header.check_bytes[1] = 0x20;
2145     header.check_bytes[2] = 0xd;
2146     header.check_bytes[3] = 0xa;
2147 
2148     /* write all the data */
2149     ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
2150     if (ret < 0) {
2151         error_setg(errp, QERR_IO_ERROR);
2152         goto exit;
2153     }
2154     ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
2155     if (ret < 0) {
2156         error_setg(errp, QERR_IO_ERROR);
2157         goto exit;
2158     }
2159 
2160     ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9,
2161                        PREALLOC_MODE_OFF, errp);
2162     if (ret < 0) {
2163         goto exit;
2164     }
2165 
2166     /* write grain directory */
2167     gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
2168     gd_buf = g_malloc0(gd_buf_size);
2169     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
2170          i < gt_count; i++, tmp += gt_size) {
2171         gd_buf[i] = cpu_to_le32(tmp);
2172     }
2173     ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
2174                      gd_buf, gd_buf_size, 0);
2175     if (ret < 0) {
2176         error_setg(errp, QERR_IO_ERROR);
2177         goto exit;
2178     }
2179 
2180     /* write backup grain directory */
2181     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
2182          i < gt_count; i++, tmp += gt_size) {
2183         gd_buf[i] = cpu_to_le32(tmp);
2184     }
2185     ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
2186                      gd_buf, gd_buf_size, 0);
2187     if (ret < 0) {
2188         error_setg(errp, QERR_IO_ERROR);
2189     }
2190 
2191     ret = 0;
2192 exit:
2193     g_free(gd_buf);
2194     return ret;
2195 }
2196 
2197 static int vmdk_create_extent(const char *filename, int64_t filesize,
2198                               bool flat, bool compress, bool zeroed_grain,
2199                               BlockBackend **pbb,
2200                               QemuOpts *opts, Error **errp)
2201 {
2202     int ret;
2203     BlockBackend *blk = NULL;
2204     Error *local_err = NULL;
2205 
2206     ret = bdrv_create_file(filename, opts, &local_err);
2207     if (ret < 0) {
2208         error_propagate(errp, local_err);
2209         goto exit;
2210     }
2211 
2212     blk = blk_new_open(filename, NULL, NULL,
2213                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2214                        &local_err);
2215     if (blk == NULL) {
2216         error_propagate(errp, local_err);
2217         ret = -EIO;
2218         goto exit;
2219     }
2220 
2221     blk_set_allow_write_beyond_eof(blk, true);
2222 
2223     ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
2224 exit:
2225     if (blk) {
2226         if (pbb) {
2227             *pbb = blk;
2228         } else {
2229             blk_unref(blk);
2230             blk = NULL;
2231         }
2232     }
2233     return ret;
2234 }
2235 
2236 static int filename_decompose(const char *filename, char *path, char *prefix,
2237                               char *postfix, size_t buf_len, Error **errp)
2238 {
2239     const char *p, *q;
2240 
2241     if (filename == NULL || !strlen(filename)) {
2242         error_setg(errp, "No filename provided");
2243         return VMDK_ERROR;
2244     }
2245     p = strrchr(filename, '/');
2246     if (p == NULL) {
2247         p = strrchr(filename, '\\');
2248     }
2249     if (p == NULL) {
2250         p = strrchr(filename, ':');
2251     }
2252     if (p != NULL) {
2253         p++;
2254         if (p - filename >= buf_len) {
2255             return VMDK_ERROR;
2256         }
2257         pstrcpy(path, p - filename + 1, filename);
2258     } else {
2259         p = filename;
2260         path[0] = '\0';
2261     }
2262     q = strrchr(p, '.');
2263     if (q == NULL) {
2264         pstrcpy(prefix, buf_len, p);
2265         postfix[0] = '\0';
2266     } else {
2267         if (q - p >= buf_len) {
2268             return VMDK_ERROR;
2269         }
2270         pstrcpy(prefix, q - p + 1, p);
2271         pstrcpy(postfix, buf_len, q);
2272     }
2273     return VMDK_OK;
2274 }
2275 
2276 /*
2277  * idx == 0: get or create the descriptor file (also the image file if in a
2278  *           non-split format.
2279  * idx >= 1: get the n-th extent if in a split subformat
2280  */
2281 typedef BlockBackend *(*vmdk_create_extent_fn)(int64_t size,
2282                                                int idx,
2283                                                bool flat,
2284                                                bool split,
2285                                                bool compress,
2286                                                bool zeroed_grain,
2287                                                void *opaque,
2288                                                Error **errp);
2289 
2290 static void vmdk_desc_add_extent(GString *desc,
2291                                  const char *extent_line_fmt,
2292                                  int64_t size, const char *filename)
2293 {
2294     char *basename = g_path_get_basename(filename);
2295 
2296     g_string_append_printf(desc, extent_line_fmt,
2297                            DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
2298     g_free(basename);
2299 }
2300 
2301 static int coroutine_fn vmdk_co_do_create(int64_t size,
2302                                           BlockdevVmdkSubformat subformat,
2303                                           BlockdevVmdkAdapterType adapter_type,
2304                                           const char *backing_file,
2305                                           const char *hw_version,
2306                                           bool compat6,
2307                                           bool zeroed_grain,
2308                                           vmdk_create_extent_fn extent_fn,
2309                                           void *opaque,
2310                                           Error **errp)
2311 {
2312     int extent_idx;
2313     BlockBackend *blk = NULL;
2314     BlockBackend *extent_blk;
2315     Error *local_err = NULL;
2316     char *desc = NULL;
2317     int ret = 0;
2318     bool flat, split, compress;
2319     GString *ext_desc_lines;
2320     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
2321     int64_t extent_size;
2322     int64_t created_size = 0;
2323     const char *extent_line_fmt;
2324     char *parent_desc_line = g_malloc0(BUF_SIZE);
2325     uint32_t parent_cid = 0xffffffff;
2326     uint32_t number_heads = 16;
2327     uint32_t desc_offset = 0, desc_len;
2328     const char desc_template[] =
2329         "# Disk DescriptorFile\n"
2330         "version=1\n"
2331         "CID=%" PRIx32 "\n"
2332         "parentCID=%" PRIx32 "\n"
2333         "createType=\"%s\"\n"
2334         "%s"
2335         "\n"
2336         "# Extent description\n"
2337         "%s"
2338         "\n"
2339         "# The Disk Data Base\n"
2340         "#DDB\n"
2341         "\n"
2342         "ddb.virtualHWVersion = \"%s\"\n"
2343         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2344         "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2345         "ddb.geometry.sectors = \"63\"\n"
2346         "ddb.adapterType = \"%s\"\n";
2347 
2348     ext_desc_lines = g_string_new(NULL);
2349 
2350     /* Read out options */
2351     if (compat6) {
2352         if (hw_version) {
2353             error_setg(errp,
2354                        "compat6 cannot be enabled with hwversion set");
2355             ret = -EINVAL;
2356             goto exit;
2357         }
2358         hw_version = "6";
2359     }
2360     if (!hw_version) {
2361         hw_version = "4";
2362     }
2363 
2364     if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2365         /* that's the number of heads with which vmware operates when
2366            creating, exporting, etc. vmdk files with a non-ide adapter type */
2367         number_heads = 255;
2368     }
2369     split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2370             (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2371     flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2372            (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2373     compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2374 
2375     if (flat) {
2376         extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2377     } else {
2378         extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2379     }
2380     if (flat && backing_file) {
2381         error_setg(errp, "Flat image can't have backing file");
2382         ret = -ENOTSUP;
2383         goto exit;
2384     }
2385     if (flat && zeroed_grain) {
2386         error_setg(errp, "Flat image can't enable zeroed grain");
2387         ret = -ENOTSUP;
2388         goto exit;
2389     }
2390 
2391     /* Create extents */
2392     if (split) {
2393         extent_size = split_size;
2394     } else {
2395         extent_size = size;
2396     }
2397     if (!split && !flat) {
2398         created_size = extent_size;
2399     } else {
2400         created_size = 0;
2401     }
2402     /* Get the descriptor file BDS */
2403     blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2404                     opaque, errp);
2405     if (!blk) {
2406         ret = -EIO;
2407         goto exit;
2408     }
2409     if (!split && !flat) {
2410         vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2411                              blk_bs(blk)->filename);
2412     }
2413 
2414     if (backing_file) {
2415         BlockBackend *backing;
2416         char *full_backing =
2417             bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2418                                                          backing_file,
2419                                                          &local_err);
2420         if (local_err) {
2421             error_propagate(errp, local_err);
2422             ret = -ENOENT;
2423             goto exit;
2424         }
2425         assert(full_backing);
2426 
2427         backing = blk_new_open(full_backing, NULL, NULL,
2428                                BDRV_O_NO_BACKING, errp);
2429         g_free(full_backing);
2430         if (backing == NULL) {
2431             ret = -EIO;
2432             goto exit;
2433         }
2434         if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2435             error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2436                        blk_bs(backing)->drv->format_name);
2437             blk_unref(backing);
2438             ret = -EINVAL;
2439             goto exit;
2440         }
2441         ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2442         blk_unref(backing);
2443         if (ret) {
2444             error_setg(errp, "Failed to read parent CID");
2445             goto exit;
2446         }
2447         snprintf(parent_desc_line, BUF_SIZE,
2448                 "parentFileNameHint=\"%s\"", backing_file);
2449     }
2450     extent_idx = 1;
2451     while (created_size < size) {
2452         int64_t cur_size = MIN(size - created_size, extent_size);
2453         extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2454                                zeroed_grain, opaque, errp);
2455         if (!extent_blk) {
2456             ret = -EINVAL;
2457             goto exit;
2458         }
2459         vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2460                              blk_bs(extent_blk)->filename);
2461         created_size += cur_size;
2462         extent_idx++;
2463         blk_unref(extent_blk);
2464     }
2465 
2466     /* Check whether we got excess extents */
2467     extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2468                            opaque, NULL);
2469     if (extent_blk) {
2470         blk_unref(extent_blk);
2471         error_setg(errp, "List of extents contains unused extents");
2472         ret = -EINVAL;
2473         goto exit;
2474     }
2475 
2476     /* generate descriptor file */
2477     desc = g_strdup_printf(desc_template,
2478                            g_random_int(),
2479                            parent_cid,
2480                            BlockdevVmdkSubformat_str(subformat),
2481                            parent_desc_line,
2482                            ext_desc_lines->str,
2483                            hw_version,
2484                            size /
2485                                (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2486                            number_heads,
2487                            BlockdevVmdkAdapterType_str(adapter_type));
2488     desc_len = strlen(desc);
2489     /* the descriptor offset = 0x200 */
2490     if (!split && !flat) {
2491         desc_offset = 0x200;
2492     }
2493 
2494     ret = blk_pwrite(blk, desc_offset, desc, desc_len, 0);
2495     if (ret < 0) {
2496         error_setg_errno(errp, -ret, "Could not write description");
2497         goto exit;
2498     }
2499     /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2500      * for description file */
2501     if (desc_offset == 0) {
2502         ret = blk_truncate(blk, desc_len, PREALLOC_MODE_OFF, errp);
2503         if (ret < 0) {
2504             goto exit;
2505         }
2506     }
2507     ret = 0;
2508 exit:
2509     if (blk) {
2510         blk_unref(blk);
2511     }
2512     g_free(desc);
2513     g_free(parent_desc_line);
2514     g_string_free(ext_desc_lines, true);
2515     return ret;
2516 }
2517 
2518 typedef struct {
2519     char *path;
2520     char *prefix;
2521     char *postfix;
2522     QemuOpts *opts;
2523 } VMDKCreateOptsData;
2524 
2525 static BlockBackend *vmdk_co_create_opts_cb(int64_t size, int idx,
2526                                             bool flat, bool split, bool compress,
2527                                             bool zeroed_grain, void *opaque,
2528                                             Error **errp)
2529 {
2530     BlockBackend *blk = NULL;
2531     BlockDriverState *bs = NULL;
2532     VMDKCreateOptsData *data = opaque;
2533     char *ext_filename = NULL;
2534     char *rel_filename = NULL;
2535 
2536     /* We're done, don't create excess extents. */
2537     if (size == -1) {
2538         assert(errp == NULL);
2539         return NULL;
2540     }
2541 
2542     if (idx == 0) {
2543         rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2544     } else if (split) {
2545         rel_filename = g_strdup_printf("%s-%c%03d%s",
2546                                        data->prefix,
2547                                        flat ? 'f' : 's', idx, data->postfix);
2548     } else {
2549         assert(idx == 1);
2550         rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2551     }
2552 
2553     ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2554     g_free(rel_filename);
2555 
2556     if (vmdk_create_extent(ext_filename, size,
2557                            flat, compress, zeroed_grain, &blk, data->opts,
2558                            errp)) {
2559         goto exit;
2560     }
2561     bdrv_unref(bs);
2562 exit:
2563     g_free(ext_filename);
2564     return blk;
2565 }
2566 
2567 static int coroutine_fn vmdk_co_create_opts(const char *filename, QemuOpts *opts,
2568                                             Error **errp)
2569 {
2570     Error *local_err = NULL;
2571     char *desc = NULL;
2572     int64_t total_size = 0;
2573     char *adapter_type = NULL;
2574     BlockdevVmdkAdapterType adapter_type_enum;
2575     char *backing_file = NULL;
2576     char *hw_version = NULL;
2577     char *fmt = NULL;
2578     BlockdevVmdkSubformat subformat;
2579     int ret = 0;
2580     char *path = g_malloc0(PATH_MAX);
2581     char *prefix = g_malloc0(PATH_MAX);
2582     char *postfix = g_malloc0(PATH_MAX);
2583     char *desc_line = g_malloc0(BUF_SIZE);
2584     char *ext_filename = g_malloc0(PATH_MAX);
2585     char *desc_filename = g_malloc0(PATH_MAX);
2586     char *parent_desc_line = g_malloc0(BUF_SIZE);
2587     bool zeroed_grain;
2588     bool compat6;
2589     VMDKCreateOptsData data;
2590 
2591     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
2592         ret = -EINVAL;
2593         goto exit;
2594     }
2595     /* Read out options */
2596     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2597                           BDRV_SECTOR_SIZE);
2598     adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
2599     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2600     hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
2601     compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false);
2602     if (strcmp(hw_version, "undefined") == 0) {
2603         g_free(hw_version);
2604         hw_version = NULL;
2605     }
2606     fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
2607     zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false);
2608 
2609     if (adapter_type) {
2610         adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup,
2611                                             adapter_type,
2612                                             BLOCKDEV_VMDK_ADAPTER_TYPE_IDE,
2613                                             &local_err);
2614         if (local_err) {
2615             error_propagate(errp, local_err);
2616             ret = -EINVAL;
2617             goto exit;
2618         }
2619     } else {
2620         adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE;
2621     }
2622 
2623     if (!fmt) {
2624         /* Default format to monolithicSparse */
2625         subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE;
2626     } else {
2627         subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup,
2628                                     fmt,
2629                                     BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE,
2630                                     &local_err);
2631         if (local_err) {
2632             error_propagate(errp, local_err);
2633             ret = -EINVAL;
2634             goto exit;
2635         }
2636     }
2637     data = (VMDKCreateOptsData){
2638         .prefix = prefix,
2639         .postfix = postfix,
2640         .path = path,
2641         .opts = opts,
2642     };
2643     ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum,
2644                             backing_file, hw_version, compat6, zeroed_grain,
2645                             vmdk_co_create_opts_cb, &data, errp);
2646 
2647 exit:
2648     g_free(adapter_type);
2649     g_free(backing_file);
2650     g_free(hw_version);
2651     g_free(fmt);
2652     g_free(desc);
2653     g_free(path);
2654     g_free(prefix);
2655     g_free(postfix);
2656     g_free(desc_line);
2657     g_free(ext_filename);
2658     g_free(desc_filename);
2659     g_free(parent_desc_line);
2660     return ret;
2661 }
2662 
2663 static BlockBackend *vmdk_co_create_cb(int64_t size, int idx,
2664                                        bool flat, bool split, bool compress,
2665                                        bool zeroed_grain, void *opaque,
2666                                        Error **errp)
2667 {
2668     int ret;
2669     BlockDriverState *bs;
2670     BlockBackend *blk;
2671     BlockdevCreateOptionsVmdk *opts = opaque;
2672 
2673     if (idx == 0) {
2674         bs = bdrv_open_blockdev_ref(opts->file, errp);
2675     } else {
2676         int i;
2677         BlockdevRefList *list = opts->extents;
2678         for (i = 1; i < idx; i++) {
2679             if (!list || !list->next) {
2680                 error_setg(errp, "Extent [%d] not specified", i);
2681                 return NULL;
2682             }
2683             list = list->next;
2684         }
2685         if (!list) {
2686             error_setg(errp, "Extent [%d] not specified", idx - 1);
2687             return NULL;
2688         }
2689         bs = bdrv_open_blockdev_ref(list->value, errp);
2690     }
2691     if (!bs) {
2692         return NULL;
2693     }
2694     blk = blk_new(bdrv_get_aio_context(bs),
2695                   BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | BLK_PERM_RESIZE,
2696                   BLK_PERM_ALL);
2697     if (blk_insert_bs(blk, bs, errp)) {
2698         bdrv_unref(bs);
2699         return NULL;
2700     }
2701     blk_set_allow_write_beyond_eof(blk, true);
2702     bdrv_unref(bs);
2703 
2704     if (size != -1) {
2705         ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2706         if (ret) {
2707             blk_unref(blk);
2708             blk = NULL;
2709         }
2710     }
2711     return blk;
2712 }
2713 
2714 static int coroutine_fn vmdk_co_create(BlockdevCreateOptions *create_options,
2715                                        Error **errp)
2716 {
2717     int ret;
2718     BlockdevCreateOptionsVmdk *opts;
2719 
2720     opts = &create_options->u.vmdk;
2721 
2722     /* Validate options */
2723     if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2724         error_setg(errp, "Image size must be a multiple of 512 bytes");
2725         ret = -EINVAL;
2726         goto out;
2727     }
2728 
2729     ret = vmdk_co_do_create(opts->size,
2730                             opts->subformat,
2731                             opts->adapter_type,
2732                             opts->backing_file,
2733                             opts->hwversion,
2734                             false,
2735                             opts->zeroed_grain,
2736                             vmdk_co_create_cb,
2737                             opts, errp);
2738     return ret;
2739 
2740 out:
2741     return ret;
2742 }
2743 
2744 static void vmdk_close(BlockDriverState *bs)
2745 {
2746     BDRVVmdkState *s = bs->opaque;
2747 
2748     vmdk_free_extents(bs);
2749     g_free(s->create_type);
2750 
2751     migrate_del_blocker(s->migration_blocker);
2752     error_free(s->migration_blocker);
2753 }
2754 
2755 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2756 {
2757     BDRVVmdkState *s = bs->opaque;
2758     int i, err;
2759     int ret = 0;
2760 
2761     for (i = 0; i < s->num_extents; i++) {
2762         err = bdrv_co_flush(s->extents[i].file->bs);
2763         if (err < 0) {
2764             ret = err;
2765         }
2766     }
2767     return ret;
2768 }
2769 
2770 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2771 {
2772     int i;
2773     int64_t ret = 0;
2774     int64_t r;
2775     BDRVVmdkState *s = bs->opaque;
2776 
2777     ret = bdrv_get_allocated_file_size(bs->file->bs);
2778     if (ret < 0) {
2779         return ret;
2780     }
2781     for (i = 0; i < s->num_extents; i++) {
2782         if (s->extents[i].file == bs->file) {
2783             continue;
2784         }
2785         r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2786         if (r < 0) {
2787             return r;
2788         }
2789         ret += r;
2790     }
2791     return ret;
2792 }
2793 
2794 static int vmdk_has_zero_init(BlockDriverState *bs)
2795 {
2796     int i;
2797     BDRVVmdkState *s = bs->opaque;
2798 
2799     /* If has a flat extent and its underlying storage doesn't have zero init,
2800      * return 0. */
2801     for (i = 0; i < s->num_extents; i++) {
2802         if (s->extents[i].flat) {
2803             if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2804                 return 0;
2805             }
2806         }
2807     }
2808     return 1;
2809 }
2810 
2811 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2812 {
2813     ImageInfo *info = g_new0(ImageInfo, 1);
2814 
2815     bdrv_refresh_filename(extent->file->bs);
2816     *info = (ImageInfo){
2817         .filename         = g_strdup(extent->file->bs->filename),
2818         .format           = g_strdup(extent->type),
2819         .virtual_size     = extent->sectors * BDRV_SECTOR_SIZE,
2820         .compressed       = extent->compressed,
2821         .has_compressed   = extent->compressed,
2822         .cluster_size     = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2823         .has_cluster_size = !extent->flat,
2824     };
2825 
2826     return info;
2827 }
2828 
2829 static int coroutine_fn vmdk_co_check(BlockDriverState *bs,
2830                                       BdrvCheckResult *result,
2831                                       BdrvCheckMode fix)
2832 {
2833     BDRVVmdkState *s = bs->opaque;
2834     VmdkExtent *extent = NULL;
2835     int64_t sector_num = 0;
2836     int64_t total_sectors = bdrv_nb_sectors(bs);
2837     int ret;
2838     uint64_t cluster_offset;
2839 
2840     if (fix) {
2841         return -ENOTSUP;
2842     }
2843 
2844     for (;;) {
2845         if (sector_num >= total_sectors) {
2846             return 0;
2847         }
2848         extent = find_extent(s, sector_num, extent);
2849         if (!extent) {
2850             fprintf(stderr,
2851                     "ERROR: could not find extent for sector %" PRId64 "\n",
2852                     sector_num);
2853             ret = -EINVAL;
2854             break;
2855         }
2856         ret = get_cluster_offset(bs, extent, NULL,
2857                                  sector_num << BDRV_SECTOR_BITS,
2858                                  false, &cluster_offset, 0, 0);
2859         if (ret == VMDK_ERROR) {
2860             fprintf(stderr,
2861                     "ERROR: could not get cluster_offset for sector %"
2862                     PRId64 "\n", sector_num);
2863             break;
2864         }
2865         if (ret == VMDK_OK) {
2866             int64_t extent_len = bdrv_getlength(extent->file->bs);
2867             if (extent_len < 0) {
2868                 fprintf(stderr,
2869                         "ERROR: could not get extent file length for sector %"
2870                         PRId64 "\n", sector_num);
2871                 ret = extent_len;
2872                 break;
2873             }
2874             if (cluster_offset >= extent_len) {
2875                 fprintf(stderr,
2876                         "ERROR: cluster offset for sector %"
2877                         PRId64 " points after EOF\n", sector_num);
2878                 ret = -EINVAL;
2879                 break;
2880             }
2881         }
2882         sector_num += extent->cluster_sectors;
2883     }
2884 
2885     result->corruptions++;
2886     return ret;
2887 }
2888 
2889 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs,
2890                                                  Error **errp)
2891 {
2892     int i;
2893     BDRVVmdkState *s = bs->opaque;
2894     ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2895     ImageInfoList **next;
2896 
2897     *spec_info = (ImageInfoSpecific){
2898         .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2899         .u = {
2900             .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2901         },
2902     };
2903 
2904     *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2905         .create_type = g_strdup(s->create_type),
2906         .cid = s->cid,
2907         .parent_cid = s->parent_cid,
2908     };
2909 
2910     next = &spec_info->u.vmdk.data->extents;
2911     for (i = 0; i < s->num_extents; i++) {
2912         *next = g_new0(ImageInfoList, 1);
2913         (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2914         (*next)->next = NULL;
2915         next = &(*next)->next;
2916     }
2917 
2918     return spec_info;
2919 }
2920 
2921 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2922 {
2923     return a->flat == b->flat &&
2924            a->compressed == b->compressed &&
2925            (a->flat || a->cluster_sectors == b->cluster_sectors);
2926 }
2927 
2928 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2929 {
2930     int i;
2931     BDRVVmdkState *s = bs->opaque;
2932     assert(s->num_extents);
2933 
2934     /* See if we have multiple extents but they have different cases */
2935     for (i = 1; i < s->num_extents; i++) {
2936         if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2937             return -ENOTSUP;
2938         }
2939     }
2940     bdi->needs_compressed_writes = s->extents[0].compressed;
2941     if (!s->extents[0].flat) {
2942         bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2943     }
2944     return 0;
2945 }
2946 
2947 static void vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
2948                                       bool backing_overridden)
2949 {
2950     /* No children but file and backing can be explicitly specified (TODO) */
2951     qdict_put(target, "file",
2952               qobject_ref(bs->file->bs->full_open_options));
2953 
2954     if (backing_overridden) {
2955         if (bs->backing) {
2956             qdict_put(target, "backing",
2957                       qobject_ref(bs->backing->bs->full_open_options));
2958         } else {
2959             qdict_put_null(target, "backing");
2960         }
2961     }
2962 }
2963 
2964 static QemuOptsList vmdk_create_opts = {
2965     .name = "vmdk-create-opts",
2966     .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2967     .desc = {
2968         {
2969             .name = BLOCK_OPT_SIZE,
2970             .type = QEMU_OPT_SIZE,
2971             .help = "Virtual disk size"
2972         },
2973         {
2974             .name = BLOCK_OPT_ADAPTER_TYPE,
2975             .type = QEMU_OPT_STRING,
2976             .help = "Virtual adapter type, can be one of "
2977                     "ide (default), lsilogic, buslogic or legacyESX"
2978         },
2979         {
2980             .name = BLOCK_OPT_BACKING_FILE,
2981             .type = QEMU_OPT_STRING,
2982             .help = "File name of a base image"
2983         },
2984         {
2985             .name = BLOCK_OPT_COMPAT6,
2986             .type = QEMU_OPT_BOOL,
2987             .help = "VMDK version 6 image",
2988             .def_value_str = "off"
2989         },
2990         {
2991             .name = BLOCK_OPT_HWVERSION,
2992             .type = QEMU_OPT_STRING,
2993             .help = "VMDK hardware version",
2994             .def_value_str = "undefined"
2995         },
2996         {
2997             .name = BLOCK_OPT_SUBFMT,
2998             .type = QEMU_OPT_STRING,
2999             .help =
3000                 "VMDK flat extent format, can be one of "
3001                 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
3002         },
3003         {
3004             .name = BLOCK_OPT_ZEROED_GRAIN,
3005             .type = QEMU_OPT_BOOL,
3006             .help = "Enable efficient zero writes "
3007                     "using the zeroed-grain GTE feature"
3008         },
3009         { /* end of list */ }
3010     }
3011 };
3012 
3013 static BlockDriver bdrv_vmdk = {
3014     .format_name                  = "vmdk",
3015     .instance_size                = sizeof(BDRVVmdkState),
3016     .bdrv_probe                   = vmdk_probe,
3017     .bdrv_open                    = vmdk_open,
3018     .bdrv_co_check                = vmdk_co_check,
3019     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
3020     .bdrv_child_perm              = bdrv_format_default_perms,
3021     .bdrv_co_preadv               = vmdk_co_preadv,
3022     .bdrv_co_pwritev              = vmdk_co_pwritev,
3023     .bdrv_co_pwritev_compressed   = vmdk_co_pwritev_compressed,
3024     .bdrv_co_pwrite_zeroes        = vmdk_co_pwrite_zeroes,
3025     .bdrv_close                   = vmdk_close,
3026     .bdrv_co_create_opts          = vmdk_co_create_opts,
3027     .bdrv_co_create               = vmdk_co_create,
3028     .bdrv_co_flush_to_disk        = vmdk_co_flush,
3029     .bdrv_co_block_status         = vmdk_co_block_status,
3030     .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
3031     .bdrv_has_zero_init           = vmdk_has_zero_init,
3032     .bdrv_get_specific_info       = vmdk_get_specific_info,
3033     .bdrv_refresh_limits          = vmdk_refresh_limits,
3034     .bdrv_get_info                = vmdk_get_info,
3035     .bdrv_gather_child_options    = vmdk_gather_child_options,
3036 
3037     .supports_backing             = true,
3038     .create_opts                  = &vmdk_create_opts,
3039 };
3040 
3041 static void bdrv_vmdk_init(void)
3042 {
3043     bdrv_register(&bdrv_vmdk);
3044 }
3045 
3046 block_init(bdrv_vmdk_init);
3047