xref: /openbmc/qemu/block/vmdk.c (revision 9e6190ae)
1 /*
2  * Block driver for the VMDK format
3  *
4  * Copyright (c) 2004 Fabrice Bellard
5  * Copyright (c) 2005 Filip Navara
6  *
7  * Permission is hereby granted, free of charge, to any person obtaining a copy
8  * of this software and associated documentation files (the "Software"), to deal
9  * in the Software without restriction, including without limitation the rights
10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11  * copies of the Software, and to permit persons to whom the Software is
12  * furnished to do so, subject to the following conditions:
13  *
14  * The above copyright notice and this permission notice shall be included in
15  * all copies or substantial portions of the Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23  * THE SOFTWARE.
24  */
25 
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "block/block_int.h"
29 #include "sysemu/block-backend.h"
30 #include "qapi/qmp/qdict.h"
31 #include "qapi/qmp/qerror.h"
32 #include "qemu/error-report.h"
33 #include "qemu/module.h"
34 #include "qemu/option.h"
35 #include "qemu/bswap.h"
36 #include "qemu/memalign.h"
37 #include "migration/blocker.h"
38 #include "qemu/cutils.h"
39 #include <zlib.h>
40 
41 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
42 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
43 #define VMDK4_COMPRESSION_DEFLATE 1
44 #define VMDK4_FLAG_NL_DETECT (1 << 0)
45 #define VMDK4_FLAG_RGD (1 << 1)
46 /* Zeroed-grain enable bit */
47 #define VMDK4_FLAG_ZERO_GRAIN   (1 << 2)
48 #define VMDK4_FLAG_COMPRESS (1 << 16)
49 #define VMDK4_FLAG_MARKER (1 << 17)
50 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
51 
52 #define VMDK_EXTENT_MAX_SECTORS (1ULL << 32)
53 
54 #define VMDK_GTE_ZEROED 0x1
55 
56 /* VMDK internal error codes */
57 #define VMDK_OK      0
58 #define VMDK_ERROR   (-1)
59 /* Cluster not allocated */
60 #define VMDK_UNALLOC (-2)
61 #define VMDK_ZEROED  (-3)
62 
63 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
64 #define BLOCK_OPT_TOOLSVERSION "toolsversion"
65 
66 typedef struct {
67     uint32_t version;
68     uint32_t flags;
69     uint32_t disk_sectors;
70     uint32_t granularity;
71     uint32_t l1dir_offset;
72     uint32_t l1dir_size;
73     uint32_t file_sectors;
74     uint32_t cylinders;
75     uint32_t heads;
76     uint32_t sectors_per_track;
77 } QEMU_PACKED VMDK3Header;
78 
79 typedef struct {
80     uint32_t version;
81     uint32_t flags;
82     uint64_t capacity;
83     uint64_t granularity;
84     uint64_t desc_offset;
85     uint64_t desc_size;
86     /* Number of GrainTableEntries per GrainTable */
87     uint32_t num_gtes_per_gt;
88     uint64_t rgd_offset;
89     uint64_t gd_offset;
90     uint64_t grain_offset;
91     char filler[1];
92     char check_bytes[4];
93     uint16_t compressAlgorithm;
94 } QEMU_PACKED VMDK4Header;
95 
96 typedef struct VMDKSESparseConstHeader {
97     uint64_t magic;
98     uint64_t version;
99     uint64_t capacity;
100     uint64_t grain_size;
101     uint64_t grain_table_size;
102     uint64_t flags;
103     uint64_t reserved1;
104     uint64_t reserved2;
105     uint64_t reserved3;
106     uint64_t reserved4;
107     uint64_t volatile_header_offset;
108     uint64_t volatile_header_size;
109     uint64_t journal_header_offset;
110     uint64_t journal_header_size;
111     uint64_t journal_offset;
112     uint64_t journal_size;
113     uint64_t grain_dir_offset;
114     uint64_t grain_dir_size;
115     uint64_t grain_tables_offset;
116     uint64_t grain_tables_size;
117     uint64_t free_bitmap_offset;
118     uint64_t free_bitmap_size;
119     uint64_t backmap_offset;
120     uint64_t backmap_size;
121     uint64_t grains_offset;
122     uint64_t grains_size;
123     uint8_t pad[304];
124 } QEMU_PACKED VMDKSESparseConstHeader;
125 
126 typedef struct VMDKSESparseVolatileHeader {
127     uint64_t magic;
128     uint64_t free_gt_number;
129     uint64_t next_txn_seq_number;
130     uint64_t replay_journal;
131     uint8_t pad[480];
132 } QEMU_PACKED VMDKSESparseVolatileHeader;
133 
134 #define L2_CACHE_SIZE 16
135 
136 typedef struct VmdkExtent {
137     BdrvChild *file;
138     bool flat;
139     bool compressed;
140     bool has_marker;
141     bool has_zero_grain;
142     bool sesparse;
143     uint64_t sesparse_l2_tables_offset;
144     uint64_t sesparse_clusters_offset;
145     int32_t entry_size;
146     int version;
147     int64_t sectors;
148     int64_t end_sector;
149     int64_t flat_start_offset;
150     int64_t l1_table_offset;
151     int64_t l1_backup_table_offset;
152     void *l1_table;
153     uint32_t *l1_backup_table;
154     unsigned int l1_size;
155     uint32_t l1_entry_sectors;
156 
157     unsigned int l2_size;
158     void *l2_cache;
159     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
160     uint32_t l2_cache_counts[L2_CACHE_SIZE];
161 
162     int64_t cluster_sectors;
163     int64_t next_cluster_sector;
164     char *type;
165 } VmdkExtent;
166 
167 typedef struct BDRVVmdkState {
168     CoMutex lock;
169     uint64_t desc_offset;
170     bool cid_updated;
171     bool cid_checked;
172     uint32_t cid;
173     uint32_t parent_cid;
174     int num_extents;
175     /* Extent array with num_extents entries, ascend ordered by address */
176     VmdkExtent *extents;
177     Error *migration_blocker;
178     char *create_type;
179 } BDRVVmdkState;
180 
181 typedef struct BDRVVmdkReopenState {
182     bool *extents_using_bs_file;
183 } BDRVVmdkReopenState;
184 
185 typedef struct VmdkMetaData {
186     unsigned int l1_index;
187     unsigned int l2_index;
188     unsigned int l2_offset;
189     bool new_allocation;
190     uint32_t *l2_cache_entry;
191 } VmdkMetaData;
192 
193 typedef struct VmdkGrainMarker {
194     uint64_t lba;
195     uint32_t size;
196     uint8_t  data[];
197 } QEMU_PACKED VmdkGrainMarker;
198 
199 enum {
200     MARKER_END_OF_STREAM    = 0,
201     MARKER_GRAIN_TABLE      = 1,
202     MARKER_GRAIN_DIRECTORY  = 2,
203     MARKER_FOOTER           = 3,
204 };
205 
206 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
207 {
208     uint32_t magic;
209 
210     if (buf_size < 4) {
211         return 0;
212     }
213     magic = be32_to_cpu(*(uint32_t *)buf);
214     if (magic == VMDK3_MAGIC ||
215         magic == VMDK4_MAGIC) {
216         return 100;
217     } else {
218         const char *p = (const char *)buf;
219         const char *end = p + buf_size;
220         while (p < end) {
221             if (*p == '#') {
222                 /* skip comment line */
223                 while (p < end && *p != '\n') {
224                     p++;
225                 }
226                 p++;
227                 continue;
228             }
229             if (*p == ' ') {
230                 while (p < end && *p == ' ') {
231                     p++;
232                 }
233                 /* skip '\r' if windows line endings used. */
234                 if (p < end && *p == '\r') {
235                     p++;
236                 }
237                 /* only accept blank lines before 'version=' line */
238                 if (p == end || *p != '\n') {
239                     return 0;
240                 }
241                 p++;
242                 continue;
243             }
244             if (end - p >= strlen("version=X\n")) {
245                 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
246                     strncmp("version=2\n", p, strlen("version=2\n")) == 0 ||
247                     strncmp("version=3\n", p, strlen("version=3\n")) == 0) {
248                     return 100;
249                 }
250             }
251             if (end - p >= strlen("version=X\r\n")) {
252                 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
253                     strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0 ||
254                     strncmp("version=3\r\n", p, strlen("version=3\r\n")) == 0) {
255                     return 100;
256                 }
257             }
258             return 0;
259         }
260         return 0;
261     }
262 }
263 
264 #define SECTOR_SIZE 512
265 #define DESC_SIZE (20 * SECTOR_SIZE)    /* 20 sectors of 512 bytes each */
266 #define BUF_SIZE 4096
267 #define HEADER_SIZE 512                 /* first sector of 512 bytes */
268 
269 static void vmdk_free_extents(BlockDriverState *bs)
270 {
271     int i;
272     BDRVVmdkState *s = bs->opaque;
273     VmdkExtent *e;
274 
275     bdrv_graph_wrlock();
276     for (i = 0; i < s->num_extents; i++) {
277         e = &s->extents[i];
278         g_free(e->l1_table);
279         g_free(e->l2_cache);
280         g_free(e->l1_backup_table);
281         g_free(e->type);
282         if (e->file != bs->file) {
283             bdrv_unref_child(bs, e->file);
284         }
285     }
286     bdrv_graph_wrunlock();
287 
288     g_free(s->extents);
289 }
290 
291 static void vmdk_free_last_extent(BlockDriverState *bs)
292 {
293     BDRVVmdkState *s = bs->opaque;
294 
295     if (s->num_extents == 0) {
296         return;
297     }
298     s->num_extents--;
299     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
300 }
301 
302 /* Return -ve errno, or 0 on success and write CID into *pcid. */
303 static int GRAPH_RDLOCK
304 vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
305 {
306     char *desc;
307     uint32_t cid;
308     const char *p_name, *cid_str;
309     size_t cid_str_size;
310     BDRVVmdkState *s = bs->opaque;
311     int ret;
312 
313     desc = g_malloc0(DESC_SIZE);
314     ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
315     if (ret < 0) {
316         goto out;
317     }
318 
319     if (parent) {
320         cid_str = "parentCID";
321         cid_str_size = sizeof("parentCID");
322     } else {
323         cid_str = "CID";
324         cid_str_size = sizeof("CID");
325     }
326 
327     desc[DESC_SIZE - 1] = '\0';
328     p_name = strstr(desc, cid_str);
329     if (p_name == NULL) {
330         ret = -EINVAL;
331         goto out;
332     }
333     p_name += cid_str_size;
334     if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
335         ret = -EINVAL;
336         goto out;
337     }
338     *pcid = cid;
339     ret = 0;
340 
341 out:
342     g_free(desc);
343     return ret;
344 }
345 
346 static int coroutine_fn GRAPH_RDLOCK
347 vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
348 {
349     char *desc, *tmp_desc;
350     char *p_name, *tmp_str;
351     BDRVVmdkState *s = bs->opaque;
352     int ret = 0;
353 
354     size_t desc_buf_size;
355 
356     if (s->desc_offset == 0) {
357         desc_buf_size = bdrv_getlength(bs->file->bs);
358         if (desc_buf_size > 16ULL << 20) {
359             error_report("VMDK description file too big");
360             return -EFBIG;
361         }
362     } else {
363         desc_buf_size = DESC_SIZE;
364     }
365 
366     desc = g_malloc0(desc_buf_size);
367     tmp_desc = g_malloc0(desc_buf_size);
368     ret = bdrv_co_pread(bs->file, s->desc_offset, desc_buf_size, desc, 0);
369     if (ret < 0) {
370         goto out;
371     }
372 
373     desc[desc_buf_size - 1] = '\0';
374     tmp_str = strstr(desc, "parentCID");
375     if (tmp_str == NULL) {
376         ret = -EINVAL;
377         goto out;
378     }
379 
380     pstrcpy(tmp_desc, desc_buf_size, tmp_str);
381     p_name = strstr(desc, "CID");
382     if (p_name != NULL) {
383         p_name += sizeof("CID");
384         snprintf(p_name, desc_buf_size - (p_name - desc), "%" PRIx32 "\n", cid);
385         pstrcat(desc, desc_buf_size, tmp_desc);
386     }
387 
388     ret = bdrv_co_pwrite_sync(bs->file, s->desc_offset, desc_buf_size, desc, 0);
389 
390 out:
391     g_free(desc);
392     g_free(tmp_desc);
393     return ret;
394 }
395 
396 static int coroutine_fn GRAPH_RDLOCK vmdk_is_cid_valid(BlockDriverState *bs)
397 {
398     BDRVVmdkState *s = bs->opaque;
399     uint32_t cur_pcid;
400 
401     if (!s->cid_checked && bs->backing) {
402         BlockDriverState *p_bs = bs->backing->bs;
403 
404         if (strcmp(p_bs->drv->format_name, "vmdk")) {
405             /* Backing file is not in vmdk format, so it does not have
406              * a CID, which makes the overlay's parent CID invalid */
407             return 0;
408         }
409 
410         if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
411             /* read failure: report as not valid */
412             return 0;
413         }
414         if (s->parent_cid != cur_pcid) {
415             /* CID not valid */
416             return 0;
417         }
418     }
419     s->cid_checked = true;
420     /* CID valid */
421     return 1;
422 }
423 
424 static int vmdk_reopen_prepare(BDRVReopenState *state,
425                                BlockReopenQueue *queue, Error **errp)
426 {
427     BDRVVmdkState *s;
428     BDRVVmdkReopenState *rs;
429     int i;
430 
431     GLOBAL_STATE_CODE();
432     GRAPH_RDLOCK_GUARD_MAINLOOP();
433 
434     assert(state != NULL);
435     assert(state->bs != NULL);
436     assert(state->opaque == NULL);
437 
438     s = state->bs->opaque;
439 
440     rs = g_new0(BDRVVmdkReopenState, 1);
441     state->opaque = rs;
442 
443     /*
444      * Check whether there are any extents stored in bs->file; if bs->file
445      * changes, we will need to update their .file pointers to follow suit
446      */
447     rs->extents_using_bs_file = g_new(bool, s->num_extents);
448     for (i = 0; i < s->num_extents; i++) {
449         rs->extents_using_bs_file[i] = s->extents[i].file == state->bs->file;
450     }
451 
452     return 0;
453 }
454 
455 static void vmdk_reopen_clean(BDRVReopenState *state)
456 {
457     BDRVVmdkReopenState *rs = state->opaque;
458 
459     g_free(rs->extents_using_bs_file);
460     g_free(rs);
461     state->opaque = NULL;
462 }
463 
464 static void vmdk_reopen_commit(BDRVReopenState *state)
465 {
466     BDRVVmdkState *s = state->bs->opaque;
467     BDRVVmdkReopenState *rs = state->opaque;
468     int i;
469 
470     GLOBAL_STATE_CODE();
471     GRAPH_RDLOCK_GUARD_MAINLOOP();
472 
473     for (i = 0; i < s->num_extents; i++) {
474         if (rs->extents_using_bs_file[i]) {
475             s->extents[i].file = state->bs->file;
476         }
477     }
478 
479     vmdk_reopen_clean(state);
480 }
481 
482 static void vmdk_reopen_abort(BDRVReopenState *state)
483 {
484     vmdk_reopen_clean(state);
485 }
486 
487 static int GRAPH_RDLOCK vmdk_parent_open(BlockDriverState *bs)
488 {
489     char *p_name;
490     char *desc;
491     BDRVVmdkState *s = bs->opaque;
492     int ret;
493 
494     desc = g_malloc0(DESC_SIZE + 1);
495     ret = bdrv_pread(bs->file, s->desc_offset, DESC_SIZE, desc, 0);
496     if (ret < 0) {
497         goto out;
498     }
499 
500     p_name = strstr(desc, "parentFileNameHint");
501     if (p_name != NULL) {
502         char *end_name;
503 
504         p_name += sizeof("parentFileNameHint") + 1;
505         end_name = strchr(p_name, '\"');
506         if (end_name == NULL) {
507             ret = -EINVAL;
508             goto out;
509         }
510         if ((end_name - p_name) > sizeof(bs->auto_backing_file) - 1) {
511             ret = -EINVAL;
512             goto out;
513         }
514 
515         pstrcpy(bs->auto_backing_file, end_name - p_name + 1, p_name);
516         pstrcpy(bs->backing_file, sizeof(bs->backing_file),
517                 bs->auto_backing_file);
518         pstrcpy(bs->backing_format, sizeof(bs->backing_format),
519                 "vmdk");
520     }
521 
522 out:
523     g_free(desc);
524     return ret;
525 }
526 
527 /* Create and append extent to the extent array. Return the added VmdkExtent
528  * address. return NULL if allocation failed. */
529 static int vmdk_add_extent(BlockDriverState *bs,
530                            BdrvChild *file, bool flat, int64_t sectors,
531                            int64_t l1_offset, int64_t l1_backup_offset,
532                            uint32_t l1_size,
533                            int l2_size, uint64_t cluster_sectors,
534                            VmdkExtent **new_extent,
535                            Error **errp)
536 {
537     VmdkExtent *extent;
538     BDRVVmdkState *s = bs->opaque;
539     int64_t nb_sectors;
540 
541     if (cluster_sectors > 0x200000) {
542         /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
543         error_setg(errp, "Invalid granularity, image may be corrupt");
544         return -EFBIG;
545     }
546     if (l1_size > 32 * 1024 * 1024) {
547         /*
548          * Although with big capacity and small l1_entry_sectors, we can get a
549          * big l1_size, we don't want unbounded value to allocate the table.
550          * Limit it to 32M, which is enough to store:
551          *     8TB  - for both VMDK3 & VMDK4 with
552          *            minimal cluster size: 512B
553          *            minimal L2 table size: 512 entries
554          *            8 TB is still more than the maximal value supported for
555          *            VMDK3 & VMDK4 which is 2TB.
556          *     64TB - for "ESXi seSparse Extent"
557          *            minimal cluster size: 512B (default is 4KB)
558          *            L2 table size: 4096 entries (const).
559          *            64TB is more than the maximal value supported for
560          *            seSparse VMDKs (which is slightly less than 64TB)
561          */
562         error_setg(errp, "L1 size too big");
563         return -EFBIG;
564     }
565 
566     nb_sectors = bdrv_nb_sectors(file->bs);
567     if (nb_sectors < 0) {
568         return nb_sectors;
569     }
570 
571     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
572     extent = &s->extents[s->num_extents];
573     s->num_extents++;
574 
575     memset(extent, 0, sizeof(VmdkExtent));
576     extent->file = file;
577     extent->flat = flat;
578     extent->sectors = sectors;
579     extent->l1_table_offset = l1_offset;
580     extent->l1_backup_table_offset = l1_backup_offset;
581     extent->l1_size = l1_size;
582     extent->l1_entry_sectors = l2_size * cluster_sectors;
583     extent->l2_size = l2_size;
584     extent->cluster_sectors = flat ? sectors : cluster_sectors;
585     extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
586     extent->entry_size = sizeof(uint32_t);
587 
588     if (s->num_extents > 1) {
589         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
590     } else {
591         extent->end_sector = extent->sectors;
592     }
593     bs->total_sectors = extent->end_sector;
594     if (new_extent) {
595         *new_extent = extent;
596     }
597     return 0;
598 }
599 
600 static int GRAPH_RDLOCK
601 vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent, Error **errp)
602 {
603     int ret;
604     size_t l1_size;
605     int i;
606 
607     /* read the L1 table */
608     l1_size = extent->l1_size * extent->entry_size;
609     extent->l1_table = g_try_malloc(l1_size);
610     if (l1_size && extent->l1_table == NULL) {
611         return -ENOMEM;
612     }
613 
614     ret = bdrv_pread(extent->file, extent->l1_table_offset, l1_size,
615                      extent->l1_table, 0);
616     if (ret < 0) {
617         bdrv_refresh_filename(extent->file->bs);
618         error_setg_errno(errp, -ret,
619                          "Could not read l1 table from extent '%s'",
620                          extent->file->bs->filename);
621         goto fail_l1;
622     }
623     for (i = 0; i < extent->l1_size; i++) {
624         if (extent->entry_size == sizeof(uint64_t)) {
625             le64_to_cpus((uint64_t *)extent->l1_table + i);
626         } else {
627             assert(extent->entry_size == sizeof(uint32_t));
628             le32_to_cpus((uint32_t *)extent->l1_table + i);
629         }
630     }
631 
632     if (extent->l1_backup_table_offset) {
633         assert(!extent->sesparse);
634         extent->l1_backup_table = g_try_malloc(l1_size);
635         if (l1_size && extent->l1_backup_table == NULL) {
636             ret = -ENOMEM;
637             goto fail_l1;
638         }
639         ret = bdrv_pread(extent->file, extent->l1_backup_table_offset,
640                          l1_size, extent->l1_backup_table, 0);
641         if (ret < 0) {
642             bdrv_refresh_filename(extent->file->bs);
643             error_setg_errno(errp, -ret,
644                              "Could not read l1 backup table from extent '%s'",
645                              extent->file->bs->filename);
646             goto fail_l1b;
647         }
648         for (i = 0; i < extent->l1_size; i++) {
649             le32_to_cpus(&extent->l1_backup_table[i]);
650         }
651     }
652 
653     extent->l2_cache =
654         g_malloc(extent->entry_size * extent->l2_size * L2_CACHE_SIZE);
655     return 0;
656  fail_l1b:
657     g_free(extent->l1_backup_table);
658  fail_l1:
659     g_free(extent->l1_table);
660     return ret;
661 }
662 
663 static int GRAPH_RDLOCK
664 vmdk_open_vmfs_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
665                       Error **errp)
666 {
667     int ret;
668     uint32_t magic;
669     VMDK3Header header;
670     VmdkExtent *extent = NULL;
671 
672     ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0);
673     if (ret < 0) {
674         bdrv_refresh_filename(file->bs);
675         error_setg_errno(errp, -ret,
676                          "Could not read header from file '%s'",
677                          file->bs->filename);
678         return ret;
679     }
680     ret = vmdk_add_extent(bs, file, false,
681                           le32_to_cpu(header.disk_sectors),
682                           (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
683                           0,
684                           le32_to_cpu(header.l1dir_size),
685                           4096,
686                           le32_to_cpu(header.granularity),
687                           &extent,
688                           errp);
689     if (ret < 0) {
690         return ret;
691     }
692     ret = vmdk_init_tables(bs, extent, errp);
693     if (ret) {
694         /* free extent allocated by vmdk_add_extent */
695         vmdk_free_last_extent(bs);
696     }
697     return ret;
698 }
699 
700 #define SESPARSE_CONST_HEADER_MAGIC UINT64_C(0x00000000cafebabe)
701 #define SESPARSE_VOLATILE_HEADER_MAGIC UINT64_C(0x00000000cafecafe)
702 
703 /* Strict checks - format not officially documented */
704 static int check_se_sparse_const_header(VMDKSESparseConstHeader *header,
705                                         Error **errp)
706 {
707     header->magic = le64_to_cpu(header->magic);
708     header->version = le64_to_cpu(header->version);
709     header->grain_size = le64_to_cpu(header->grain_size);
710     header->grain_table_size = le64_to_cpu(header->grain_table_size);
711     header->flags = le64_to_cpu(header->flags);
712     header->reserved1 = le64_to_cpu(header->reserved1);
713     header->reserved2 = le64_to_cpu(header->reserved2);
714     header->reserved3 = le64_to_cpu(header->reserved3);
715     header->reserved4 = le64_to_cpu(header->reserved4);
716 
717     header->volatile_header_offset =
718         le64_to_cpu(header->volatile_header_offset);
719     header->volatile_header_size = le64_to_cpu(header->volatile_header_size);
720 
721     header->journal_header_offset = le64_to_cpu(header->journal_header_offset);
722     header->journal_header_size = le64_to_cpu(header->journal_header_size);
723 
724     header->journal_offset = le64_to_cpu(header->journal_offset);
725     header->journal_size = le64_to_cpu(header->journal_size);
726 
727     header->grain_dir_offset = le64_to_cpu(header->grain_dir_offset);
728     header->grain_dir_size = le64_to_cpu(header->grain_dir_size);
729 
730     header->grain_tables_offset = le64_to_cpu(header->grain_tables_offset);
731     header->grain_tables_size = le64_to_cpu(header->grain_tables_size);
732 
733     header->free_bitmap_offset = le64_to_cpu(header->free_bitmap_offset);
734     header->free_bitmap_size = le64_to_cpu(header->free_bitmap_size);
735 
736     header->backmap_offset = le64_to_cpu(header->backmap_offset);
737     header->backmap_size = le64_to_cpu(header->backmap_size);
738 
739     header->grains_offset = le64_to_cpu(header->grains_offset);
740     header->grains_size = le64_to_cpu(header->grains_size);
741 
742     if (header->magic != SESPARSE_CONST_HEADER_MAGIC) {
743         error_setg(errp, "Bad const header magic: 0x%016" PRIx64,
744                    header->magic);
745         return -EINVAL;
746     }
747 
748     if (header->version != 0x0000000200000001) {
749         error_setg(errp, "Unsupported version: 0x%016" PRIx64,
750                    header->version);
751         return -ENOTSUP;
752     }
753 
754     if (header->grain_size != 8) {
755         error_setg(errp, "Unsupported grain size: %" PRIu64,
756                    header->grain_size);
757         return -ENOTSUP;
758     }
759 
760     if (header->grain_table_size != 64) {
761         error_setg(errp, "Unsupported grain table size: %" PRIu64,
762                    header->grain_table_size);
763         return -ENOTSUP;
764     }
765 
766     if (header->flags != 0) {
767         error_setg(errp, "Unsupported flags: 0x%016" PRIx64,
768                    header->flags);
769         return -ENOTSUP;
770     }
771 
772     if (header->reserved1 != 0 || header->reserved2 != 0 ||
773         header->reserved3 != 0 || header->reserved4 != 0) {
774         error_setg(errp, "Unsupported reserved bits:"
775                    " 0x%016" PRIx64 " 0x%016" PRIx64
776                    " 0x%016" PRIx64 " 0x%016" PRIx64,
777                    header->reserved1, header->reserved2,
778                    header->reserved3, header->reserved4);
779         return -ENOTSUP;
780     }
781 
782     /* check that padding is 0 */
783     if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
784         error_setg(errp, "Unsupported non-zero const header padding");
785         return -ENOTSUP;
786     }
787 
788     return 0;
789 }
790 
791 static int check_se_sparse_volatile_header(VMDKSESparseVolatileHeader *header,
792                                            Error **errp)
793 {
794     header->magic = le64_to_cpu(header->magic);
795     header->free_gt_number = le64_to_cpu(header->free_gt_number);
796     header->next_txn_seq_number = le64_to_cpu(header->next_txn_seq_number);
797     header->replay_journal = le64_to_cpu(header->replay_journal);
798 
799     if (header->magic != SESPARSE_VOLATILE_HEADER_MAGIC) {
800         error_setg(errp, "Bad volatile header magic: 0x%016" PRIx64,
801                    header->magic);
802         return -EINVAL;
803     }
804 
805     if (header->replay_journal) {
806         error_setg(errp, "Image is dirty, Replaying journal not supported");
807         return -ENOTSUP;
808     }
809 
810     /* check that padding is 0 */
811     if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
812         error_setg(errp, "Unsupported non-zero volatile header padding");
813         return -ENOTSUP;
814     }
815 
816     return 0;
817 }
818 
819 static int GRAPH_RDLOCK
820 vmdk_open_se_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
821                     Error **errp)
822 {
823     int ret;
824     VMDKSESparseConstHeader const_header;
825     VMDKSESparseVolatileHeader volatile_header;
826     VmdkExtent *extent = NULL;
827 
828     ret = bdrv_apply_auto_read_only(bs,
829             "No write support for seSparse images available", errp);
830     if (ret < 0) {
831         return ret;
832     }
833 
834     assert(sizeof(const_header) == SECTOR_SIZE);
835 
836     ret = bdrv_pread(file, 0, sizeof(const_header), &const_header, 0);
837     if (ret < 0) {
838         bdrv_refresh_filename(file->bs);
839         error_setg_errno(errp, -ret,
840                          "Could not read const header from file '%s'",
841                          file->bs->filename);
842         return ret;
843     }
844 
845     /* check const header */
846     ret = check_se_sparse_const_header(&const_header, errp);
847     if (ret < 0) {
848         return ret;
849     }
850 
851     assert(sizeof(volatile_header) == SECTOR_SIZE);
852 
853     ret = bdrv_pread(file, const_header.volatile_header_offset * SECTOR_SIZE,
854                      sizeof(volatile_header), &volatile_header, 0);
855     if (ret < 0) {
856         bdrv_refresh_filename(file->bs);
857         error_setg_errno(errp, -ret,
858                          "Could not read volatile header from file '%s'",
859                          file->bs->filename);
860         return ret;
861     }
862 
863     /* check volatile header */
864     ret = check_se_sparse_volatile_header(&volatile_header, errp);
865     if (ret < 0) {
866         return ret;
867     }
868 
869     ret = vmdk_add_extent(bs, file, false,
870                           const_header.capacity,
871                           const_header.grain_dir_offset * SECTOR_SIZE,
872                           0,
873                           const_header.grain_dir_size *
874                           SECTOR_SIZE / sizeof(uint64_t),
875                           const_header.grain_table_size *
876                           SECTOR_SIZE / sizeof(uint64_t),
877                           const_header.grain_size,
878                           &extent,
879                           errp);
880     if (ret < 0) {
881         return ret;
882     }
883 
884     extent->sesparse = true;
885     extent->sesparse_l2_tables_offset = const_header.grain_tables_offset;
886     extent->sesparse_clusters_offset = const_header.grains_offset;
887     extent->entry_size = sizeof(uint64_t);
888 
889     ret = vmdk_init_tables(bs, extent, errp);
890     if (ret) {
891         /* free extent allocated by vmdk_add_extent */
892         vmdk_free_last_extent(bs);
893     }
894 
895     return ret;
896 }
897 
898 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
899                                QDict *options, Error **errp);
900 
901 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
902 {
903     int64_t size;
904     char *buf;
905     int ret;
906 
907     size = bdrv_getlength(file->bs);
908     if (size < 0) {
909         error_setg_errno(errp, -size, "Could not access file");
910         return NULL;
911     }
912 
913     if (size < 4) {
914         /* Both descriptor file and sparse image must be much larger than 4
915          * bytes, also callers of vmdk_read_desc want to compare the first 4
916          * bytes with VMDK4_MAGIC, let's error out if less is read. */
917         error_setg(errp, "File is too small, not a valid image");
918         return NULL;
919     }
920 
921     size = MIN(size, (1 << 20) - 1);  /* avoid unbounded allocation */
922     buf = g_malloc(size + 1);
923 
924     ret = bdrv_pread(file, desc_offset, size, buf, 0);
925     if (ret < 0) {
926         error_setg_errno(errp, -ret, "Could not read from file");
927         g_free(buf);
928         return NULL;
929     }
930     buf[size] = 0;
931 
932     return buf;
933 }
934 
935 static int GRAPH_RDLOCK
936 vmdk_open_vmdk4(BlockDriverState *bs, BdrvChild *file, int flags,
937                 QDict *options, Error **errp)
938 {
939     int ret;
940     uint32_t magic;
941     uint32_t l1_size, l1_entry_sectors;
942     VMDK4Header header;
943     VmdkExtent *extent = NULL;
944     BDRVVmdkState *s = bs->opaque;
945     int64_t l1_backup_offset = 0;
946     bool compressed;
947 
948     ret = bdrv_pread(file, sizeof(magic), sizeof(header), &header, 0);
949     if (ret < 0) {
950         bdrv_refresh_filename(file->bs);
951         error_setg_errno(errp, -ret,
952                          "Could not read header from file '%s'",
953                          file->bs->filename);
954         return -EINVAL;
955     }
956     if (header.capacity == 0) {
957         uint64_t desc_offset = le64_to_cpu(header.desc_offset);
958         if (desc_offset) {
959             char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
960             if (!buf) {
961                 return -EINVAL;
962             }
963             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
964             g_free(buf);
965             return ret;
966         }
967     }
968 
969     if (!s->create_type) {
970         s->create_type = g_strdup("monolithicSparse");
971     }
972 
973     if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
974         /*
975          * The footer takes precedence over the header, so read it in. The
976          * footer starts at offset -1024 from the end: One sector for the
977          * footer, and another one for the end-of-stream marker.
978          */
979         struct {
980             struct {
981                 uint64_t val;
982                 uint32_t size;
983                 uint32_t type;
984                 uint8_t pad[512 - 16];
985             } QEMU_PACKED footer_marker;
986 
987             uint32_t magic;
988             VMDK4Header header;
989             uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
990 
991             struct {
992                 uint64_t val;
993                 uint32_t size;
994                 uint32_t type;
995                 uint8_t pad[512 - 16];
996             } QEMU_PACKED eos_marker;
997         } QEMU_PACKED footer;
998 
999         ret = bdrv_pread(file, bs->file->bs->total_sectors * 512 - 1536,
1000                          sizeof(footer), &footer, 0);
1001         if (ret < 0) {
1002             error_setg_errno(errp, -ret, "Failed to read footer");
1003             return ret;
1004         }
1005 
1006         /* Some sanity checks for the footer */
1007         if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
1008             le32_to_cpu(footer.footer_marker.size) != 0  ||
1009             le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
1010             le64_to_cpu(footer.eos_marker.val) != 0  ||
1011             le32_to_cpu(footer.eos_marker.size) != 0  ||
1012             le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
1013         {
1014             error_setg(errp, "Invalid footer");
1015             return -EINVAL;
1016         }
1017 
1018         header = footer.header;
1019     }
1020 
1021     compressed =
1022         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1023     if (le32_to_cpu(header.version) > 3) {
1024         error_setg(errp, "Unsupported VMDK version %" PRIu32,
1025                    le32_to_cpu(header.version));
1026         return -ENOTSUP;
1027     } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
1028                !compressed) {
1029         /* VMware KB 2064959 explains that version 3 added support for
1030          * persistent changed block tracking (CBT), and backup software can
1031          * read it as version=1 if it doesn't care about the changed area
1032          * information. So we are safe to enable read only. */
1033         error_setg(errp, "VMDK version 3 must be read only");
1034         return -EINVAL;
1035     }
1036 
1037     if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
1038         error_setg(errp, "L2 table size too big");
1039         return -EINVAL;
1040     }
1041 
1042     l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
1043                         * le64_to_cpu(header.granularity);
1044     if (l1_entry_sectors == 0) {
1045         error_setg(errp, "L1 entry size is invalid");
1046         return -EINVAL;
1047     }
1048     l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
1049                 / l1_entry_sectors;
1050     if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
1051         l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
1052     }
1053     if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
1054         error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
1055                    (int64_t)(le64_to_cpu(header.grain_offset)
1056                              * BDRV_SECTOR_SIZE));
1057         return -EINVAL;
1058     }
1059 
1060     ret = vmdk_add_extent(bs, file, false,
1061                           le64_to_cpu(header.capacity),
1062                           le64_to_cpu(header.gd_offset) << 9,
1063                           l1_backup_offset,
1064                           l1_size,
1065                           le32_to_cpu(header.num_gtes_per_gt),
1066                           le64_to_cpu(header.granularity),
1067                           &extent,
1068                           errp);
1069     if (ret < 0) {
1070         return ret;
1071     }
1072     extent->compressed =
1073         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1074     if (extent->compressed) {
1075         g_free(s->create_type);
1076         s->create_type = g_strdup("streamOptimized");
1077     }
1078     extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
1079     extent->version = le32_to_cpu(header.version);
1080     extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
1081     ret = vmdk_init_tables(bs, extent, errp);
1082     if (ret) {
1083         /* free extent allocated by vmdk_add_extent */
1084         vmdk_free_last_extent(bs);
1085     }
1086     return ret;
1087 }
1088 
1089 /* find an option value out of descriptor file */
1090 static int vmdk_parse_description(const char *desc, const char *opt_name,
1091         char *buf, int buf_size)
1092 {
1093     char *opt_pos, *opt_end;
1094     const char *end = desc + strlen(desc);
1095 
1096     opt_pos = strstr(desc, opt_name);
1097     if (!opt_pos) {
1098         return VMDK_ERROR;
1099     }
1100     /* Skip "=\"" following opt_name */
1101     opt_pos += strlen(opt_name) + 2;
1102     if (opt_pos >= end) {
1103         return VMDK_ERROR;
1104     }
1105     opt_end = opt_pos;
1106     while (opt_end < end && *opt_end != '"') {
1107         opt_end++;
1108     }
1109     if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
1110         return VMDK_ERROR;
1111     }
1112     pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
1113     return VMDK_OK;
1114 }
1115 
1116 /* Open an extent file and append to bs array */
1117 static int GRAPH_RDLOCK
1118 vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
1119                  char *buf, QDict *options, Error **errp)
1120 {
1121     uint32_t magic;
1122 
1123     magic = ldl_be_p(buf);
1124     switch (magic) {
1125         case VMDK3_MAGIC:
1126             return vmdk_open_vmfs_sparse(bs, file, flags, errp);
1127         case VMDK4_MAGIC:
1128             return vmdk_open_vmdk4(bs, file, flags, options, errp);
1129         default:
1130             error_setg(errp, "Image not in VMDK format");
1131             return -EINVAL;
1132     }
1133 }
1134 
1135 static const char *next_line(const char *s)
1136 {
1137     while (*s) {
1138         if (*s == '\n') {
1139             return s + 1;
1140         }
1141         s++;
1142     }
1143     return s;
1144 }
1145 
1146 static int GRAPH_RDLOCK
1147 vmdk_parse_extents(const char *desc, BlockDriverState *bs, QDict *options,
1148                    Error **errp)
1149 {
1150     ERRP_GUARD();
1151     int ret;
1152     int matches;
1153     char access[11];
1154     char type[11];
1155     char fname[512];
1156     const char *p, *np;
1157     int64_t sectors = 0;
1158     int64_t flat_offset;
1159     char *desc_file_dir = NULL;
1160     char *extent_path;
1161     BdrvChild *extent_file;
1162     BdrvChildRole extent_role;
1163     BDRVVmdkState *s = bs->opaque;
1164     VmdkExtent *extent = NULL;
1165     char extent_opt_prefix[32];
1166     Error *local_err = NULL;
1167 
1168     GLOBAL_STATE_CODE();
1169 
1170     for (p = desc; *p; p = next_line(p)) {
1171         /* parse extent line in one of below formats:
1172          *
1173          * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
1174          * RW [size in sectors] SPARSE "file-name.vmdk"
1175          * RW [size in sectors] VMFS "file-name.vmdk"
1176          * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
1177          * RW [size in sectors] SESPARSE "file-name.vmdk"
1178          */
1179         flat_offset = -1;
1180         matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
1181                          access, &sectors, type, fname, &flat_offset);
1182         if (matches < 4 || strcmp(access, "RW")) {
1183             continue;
1184         } else if (!strcmp(type, "FLAT")) {
1185             if (matches != 5 || flat_offset < 0) {
1186                 goto invalid;
1187             }
1188         } else if (!strcmp(type, "VMFS")) {
1189             if (matches == 4) {
1190                 flat_offset = 0;
1191             } else {
1192                 goto invalid;
1193             }
1194         } else if (matches != 4) {
1195             goto invalid;
1196         }
1197 
1198         if (sectors <= 0 ||
1199             (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
1200              strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") &&
1201              strcmp(type, "SESPARSE")) ||
1202             (strcmp(access, "RW"))) {
1203             continue;
1204         }
1205 
1206         if (path_is_absolute(fname)) {
1207             extent_path = g_strdup(fname);
1208         } else {
1209             if (!desc_file_dir) {
1210                 desc_file_dir = bdrv_dirname(bs->file->bs, errp);
1211                 if (!desc_file_dir) {
1212                     bdrv_refresh_filename(bs->file->bs);
1213                     error_prepend(errp, "Cannot use relative paths with VMDK "
1214                                   "descriptor file '%s': ",
1215                                   bs->file->bs->filename);
1216                     ret = -EINVAL;
1217                     goto out;
1218                 }
1219             }
1220 
1221             extent_path = g_strconcat(desc_file_dir, fname, NULL);
1222         }
1223 
1224         ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
1225         assert(ret < 32);
1226 
1227         extent_role = BDRV_CHILD_DATA;
1228         if (strcmp(type, "FLAT") != 0 && strcmp(type, "VMFS") != 0) {
1229             /* non-flat extents have metadata */
1230             extent_role |= BDRV_CHILD_METADATA;
1231         }
1232 
1233         extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
1234                                       bs, &child_of_bds, extent_role, false,
1235                                       &local_err);
1236         g_free(extent_path);
1237         if (!extent_file) {
1238             error_propagate(errp, local_err);
1239             ret = -EINVAL;
1240             goto out;
1241         }
1242 
1243         /* save to extents array */
1244         if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
1245             /* FLAT extent */
1246 
1247             ret = vmdk_add_extent(bs, extent_file, true, sectors,
1248                             0, 0, 0, 0, 0, &extent, errp);
1249             if (ret < 0) {
1250                 bdrv_graph_rdunlock_main_loop();
1251                 bdrv_graph_wrlock();
1252                 bdrv_unref_child(bs, extent_file);
1253                 bdrv_graph_wrunlock();
1254                 bdrv_graph_rdlock_main_loop();
1255                 goto out;
1256             }
1257             extent->flat_start_offset = flat_offset << 9;
1258         } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
1259             /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
1260             char *buf = vmdk_read_desc(extent_file, 0, errp);
1261             if (!buf) {
1262                 ret = -EINVAL;
1263             } else {
1264                 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
1265                                        options, errp);
1266             }
1267             g_free(buf);
1268             if (ret) {
1269                 bdrv_graph_rdunlock_main_loop();
1270                 bdrv_graph_wrlock();
1271                 bdrv_unref_child(bs, extent_file);
1272                 bdrv_graph_wrunlock();
1273                 bdrv_graph_rdlock_main_loop();
1274                 goto out;
1275             }
1276             extent = &s->extents[s->num_extents - 1];
1277         } else if (!strcmp(type, "SESPARSE")) {
1278             ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
1279             if (ret) {
1280                 bdrv_graph_rdunlock_main_loop();
1281                 bdrv_graph_wrlock();
1282                 bdrv_unref_child(bs, extent_file);
1283                 bdrv_graph_wrunlock();
1284                 bdrv_graph_rdlock_main_loop();
1285                 goto out;
1286             }
1287             extent = &s->extents[s->num_extents - 1];
1288         } else {
1289             error_setg(errp, "Unsupported extent type '%s'", type);
1290             bdrv_graph_rdunlock_main_loop();
1291             bdrv_graph_wrlock();
1292             bdrv_unref_child(bs, extent_file);
1293             bdrv_graph_wrunlock();
1294             bdrv_graph_rdlock_main_loop();
1295             ret = -ENOTSUP;
1296             goto out;
1297         }
1298         extent->type = g_strdup(type);
1299     }
1300 
1301     ret = 0;
1302     goto out;
1303 
1304 invalid:
1305     np = next_line(p);
1306     assert(np != p);
1307     if (np[-1] == '\n') {
1308         np--;
1309     }
1310     error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
1311     ret = -EINVAL;
1312 
1313 out:
1314     g_free(desc_file_dir);
1315     return ret;
1316 }
1317 
1318 static int GRAPH_RDLOCK
1319 vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf, QDict *options,
1320                     Error **errp)
1321 {
1322     int ret;
1323     char ct[128];
1324     BDRVVmdkState *s = bs->opaque;
1325 
1326     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
1327         error_setg(errp, "invalid VMDK image descriptor");
1328         ret = -EINVAL;
1329         goto exit;
1330     }
1331     if (strcmp(ct, "monolithicFlat") &&
1332         strcmp(ct, "vmfs") &&
1333         strcmp(ct, "vmfsSparse") &&
1334         strcmp(ct, "seSparse") &&
1335         strcmp(ct, "twoGbMaxExtentSparse") &&
1336         strcmp(ct, "twoGbMaxExtentFlat")) {
1337         error_setg(errp, "Unsupported image type '%s'", ct);
1338         ret = -ENOTSUP;
1339         goto exit;
1340     }
1341     s->create_type = g_strdup(ct);
1342     s->desc_offset = 0;
1343     ret = vmdk_parse_extents(buf, bs, options, errp);
1344 exit:
1345     return ret;
1346 }
1347 
1348 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
1349                      Error **errp)
1350 {
1351     char *buf;
1352     int ret;
1353     BDRVVmdkState *s = bs->opaque;
1354     uint32_t magic;
1355 
1356     GRAPH_RDLOCK_GUARD_MAINLOOP();
1357 
1358     ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
1359     if (ret < 0) {
1360         return ret;
1361     }
1362 
1363     buf = vmdk_read_desc(bs->file, 0, errp);
1364     if (!buf) {
1365         return -EINVAL;
1366     }
1367 
1368     magic = ldl_be_p(buf);
1369     switch (magic) {
1370         case VMDK3_MAGIC:
1371         case VMDK4_MAGIC:
1372             ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
1373                                    errp);
1374             s->desc_offset = 0x200;
1375             break;
1376         default:
1377             /* No data in the descriptor file */
1378             bs->file->role &= ~BDRV_CHILD_DATA;
1379 
1380             /* Must succeed because we have given up permissions if anything */
1381             bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1382 
1383             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1384             break;
1385     }
1386     if (ret) {
1387         goto fail;
1388     }
1389 
1390     /* try to open parent images, if exist */
1391     ret = vmdk_parent_open(bs);
1392     if (ret) {
1393         goto fail;
1394     }
1395     ret = vmdk_read_cid(bs, 0, &s->cid);
1396     if (ret) {
1397         goto fail;
1398     }
1399     ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1400     if (ret) {
1401         goto fail;
1402     }
1403     qemu_co_mutex_init(&s->lock);
1404 
1405     /* Disable migration when VMDK images are used */
1406     error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1407                "does not support live migration",
1408                bdrv_get_device_or_node_name(bs));
1409     ret = migrate_add_blocker_normal(&s->migration_blocker, errp);
1410     if (ret < 0) {
1411         goto fail;
1412     }
1413 
1414     g_free(buf);
1415     return 0;
1416 
1417 fail:
1418     g_free(buf);
1419     g_free(s->create_type);
1420     s->create_type = NULL;
1421     vmdk_free_extents(bs);
1422     return ret;
1423 }
1424 
1425 
1426 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1427 {
1428     BDRVVmdkState *s = bs->opaque;
1429     int i;
1430 
1431     for (i = 0; i < s->num_extents; i++) {
1432         if (!s->extents[i].flat) {
1433             bs->bl.pwrite_zeroes_alignment =
1434                 MAX(bs->bl.pwrite_zeroes_alignment,
1435                     s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1436         }
1437     }
1438 }
1439 
1440 /**
1441  * get_whole_cluster
1442  *
1443  * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1444  * to the cluster at @cluster_sector_num. If @zeroed is true, we're overwriting
1445  * a zeroed cluster in the current layer and must not copy data from the
1446  * backing file.
1447  *
1448  * If @skip_start_sector < @skip_end_sector, the relative range
1449  * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1450  * it for call to write user data in the request.
1451  */
1452 static int coroutine_fn GRAPH_RDLOCK
1453 get_whole_cluster(BlockDriverState *bs, VmdkExtent *extent,
1454                   uint64_t cluster_offset, uint64_t offset,
1455                   uint64_t skip_start_bytes, uint64_t skip_end_bytes,
1456                   bool zeroed)
1457 {
1458     int ret = VMDK_OK;
1459     int64_t cluster_bytes;
1460     uint8_t *whole_grain;
1461     bool copy_from_backing;
1462 
1463     /* For COW, align request sector_num to cluster start */
1464     cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1465     offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1466     whole_grain = qemu_blockalign(bs, cluster_bytes);
1467     copy_from_backing = bs->backing && !zeroed;
1468 
1469     if (!copy_from_backing) {
1470         memset(whole_grain, 0, skip_start_bytes);
1471         memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1472     }
1473 
1474     assert(skip_end_bytes <= cluster_bytes);
1475     /* we will be here if it's first write on non-exist grain(cluster).
1476      * try to read from parent image, if exist */
1477     if (bs->backing && !vmdk_is_cid_valid(bs)) {
1478         ret = VMDK_ERROR;
1479         goto exit;
1480     }
1481 
1482     /* Read backing data before skip range */
1483     if (skip_start_bytes > 0) {
1484         if (copy_from_backing) {
1485             /* qcow2 emits this on bs->file instead of bs->backing */
1486             BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_READ);
1487             ret = bdrv_co_pread(bs->backing, offset, skip_start_bytes,
1488                                 whole_grain, 0);
1489             if (ret < 0) {
1490                 ret = VMDK_ERROR;
1491                 goto exit;
1492             }
1493         }
1494         BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_WRITE);
1495         ret = bdrv_co_pwrite(extent->file, cluster_offset, skip_start_bytes,
1496                              whole_grain, 0);
1497         if (ret < 0) {
1498             ret = VMDK_ERROR;
1499             goto exit;
1500         }
1501     }
1502     /* Read backing data after skip range */
1503     if (skip_end_bytes < cluster_bytes) {
1504         if (copy_from_backing) {
1505             /* qcow2 emits this on bs->file instead of bs->backing */
1506             BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_READ);
1507             ret = bdrv_co_pread(bs->backing, offset + skip_end_bytes,
1508                                 cluster_bytes - skip_end_bytes,
1509                                 whole_grain + skip_end_bytes, 0);
1510             if (ret < 0) {
1511                 ret = VMDK_ERROR;
1512                 goto exit;
1513             }
1514         }
1515         BLKDBG_CO_EVENT(extent->file, BLKDBG_COW_WRITE);
1516         ret = bdrv_co_pwrite(extent->file, cluster_offset + skip_end_bytes,
1517                              cluster_bytes - skip_end_bytes,
1518                              whole_grain + skip_end_bytes, 0);
1519         if (ret < 0) {
1520             ret = VMDK_ERROR;
1521             goto exit;
1522         }
1523     }
1524 
1525     ret = VMDK_OK;
1526 exit:
1527     qemu_vfree(whole_grain);
1528     return ret;
1529 }
1530 
1531 static int coroutine_fn GRAPH_RDLOCK
1532 vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data, uint32_t offset)
1533 {
1534     offset = cpu_to_le32(offset);
1535     /* update L2 table */
1536     BLKDBG_CO_EVENT(extent->file, BLKDBG_L2_UPDATE);
1537     if (bdrv_co_pwrite(extent->file,
1538                        ((int64_t)m_data->l2_offset * 512)
1539                            + (m_data->l2_index * sizeof(offset)),
1540                        sizeof(offset), &offset, 0) < 0) {
1541         return VMDK_ERROR;
1542     }
1543     /* update backup L2 table */
1544     if (extent->l1_backup_table_offset != 0) {
1545         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1546         if (bdrv_co_pwrite(extent->file,
1547                            ((int64_t)m_data->l2_offset * 512)
1548                                + (m_data->l2_index * sizeof(offset)),
1549                            sizeof(offset), &offset, 0) < 0) {
1550             return VMDK_ERROR;
1551         }
1552     }
1553     if (bdrv_co_flush(extent->file->bs) < 0) {
1554         return VMDK_ERROR;
1555     }
1556     if (m_data->l2_cache_entry) {
1557         *m_data->l2_cache_entry = offset;
1558     }
1559 
1560     return VMDK_OK;
1561 }
1562 
1563 /**
1564  * get_cluster_offset
1565  *
1566  * Look up cluster offset in extent file by sector number, and store in
1567  * @cluster_offset.
1568  *
1569  * For flat extents, the start offset as parsed from the description file is
1570  * returned.
1571  *
1572  * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1573  * offset for a new cluster and update L2 cache. If there is a backing file,
1574  * COW is done before returning; otherwise, zeroes are written to the allocated
1575  * cluster. Both COW and zero writing skips the sector range
1576  * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1577  * has new data to write there.
1578  *
1579  * Returns: VMDK_OK if cluster exists and mapped in the image.
1580  *          VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1581  *          VMDK_ERROR if failed.
1582  */
1583 static int coroutine_fn GRAPH_RDLOCK
1584 get_cluster_offset(BlockDriverState *bs, VmdkExtent *extent,
1585                    VmdkMetaData *m_data, uint64_t offset, bool allocate,
1586                    uint64_t *cluster_offset, uint64_t skip_start_bytes,
1587                    uint64_t skip_end_bytes)
1588 {
1589     unsigned int l1_index, l2_offset, l2_index;
1590     int min_index, i, j;
1591     uint32_t min_count;
1592     void *l2_table;
1593     bool zeroed = false;
1594     int64_t ret;
1595     int64_t cluster_sector;
1596     unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
1597 
1598     if (m_data) {
1599         m_data->new_allocation = false;
1600     }
1601     if (extent->flat) {
1602         *cluster_offset = extent->flat_start_offset;
1603         return VMDK_OK;
1604     }
1605 
1606     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1607     l1_index = (offset >> 9) / extent->l1_entry_sectors;
1608     if (l1_index >= extent->l1_size) {
1609         return VMDK_ERROR;
1610     }
1611     if (extent->sesparse) {
1612         uint64_t l2_offset_u64;
1613 
1614         assert(extent->entry_size == sizeof(uint64_t));
1615 
1616         l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
1617         if (l2_offset_u64 == 0) {
1618             l2_offset = 0;
1619         } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
1620             /*
1621              * Top most nibble is 0x1 if grain table is allocated.
1622              * strict check - top most 4 bytes must be 0x10000000 since max
1623              * supported size is 64TB for disk - so no more than 64TB / 16MB
1624              * grain directories which is smaller than uint32,
1625              * where 16MB is the only supported default grain table coverage.
1626              */
1627             return VMDK_ERROR;
1628         } else {
1629             l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
1630             l2_offset_u64 = extent->sesparse_l2_tables_offset +
1631                 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
1632             if (l2_offset_u64 > 0x00000000ffffffff) {
1633                 return VMDK_ERROR;
1634             }
1635             l2_offset = (unsigned int)(l2_offset_u64);
1636         }
1637     } else {
1638         assert(extent->entry_size == sizeof(uint32_t));
1639         l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
1640     }
1641     if (!l2_offset) {
1642         return VMDK_UNALLOC;
1643     }
1644     for (i = 0; i < L2_CACHE_SIZE; i++) {
1645         if (l2_offset == extent->l2_cache_offsets[i]) {
1646             /* increment the hit count */
1647             if (++extent->l2_cache_counts[i] == 0xffffffff) {
1648                 for (j = 0; j < L2_CACHE_SIZE; j++) {
1649                     extent->l2_cache_counts[j] >>= 1;
1650                 }
1651             }
1652             l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
1653             goto found;
1654         }
1655     }
1656     /* not found: load a new entry in the least used one */
1657     min_index = 0;
1658     min_count = 0xffffffff;
1659     for (i = 0; i < L2_CACHE_SIZE; i++) {
1660         if (extent->l2_cache_counts[i] < min_count) {
1661             min_count = extent->l2_cache_counts[i];
1662             min_index = i;
1663         }
1664     }
1665     l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
1666     BLKDBG_CO_EVENT(extent->file, BLKDBG_L2_LOAD);
1667     if (bdrv_co_pread(extent->file,
1668                 (int64_t)l2_offset * 512,
1669                 l2_size_bytes,
1670                 l2_table, 0
1671             ) < 0) {
1672         return VMDK_ERROR;
1673     }
1674 
1675     extent->l2_cache_offsets[min_index] = l2_offset;
1676     extent->l2_cache_counts[min_index] = 1;
1677  found:
1678     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1679     if (m_data) {
1680         m_data->l1_index = l1_index;
1681         m_data->l2_index = l2_index;
1682         m_data->l2_offset = l2_offset;
1683         m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
1684     }
1685 
1686     if (extent->sesparse) {
1687         cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
1688         switch (cluster_sector & 0xf000000000000000) {
1689         case 0x0000000000000000:
1690             /* unallocated grain */
1691             if (cluster_sector != 0) {
1692                 return VMDK_ERROR;
1693             }
1694             break;
1695         case 0x1000000000000000:
1696             /* scsi-unmapped grain - fallthrough */
1697         case 0x2000000000000000:
1698             /* zero grain */
1699             zeroed = true;
1700             break;
1701         case 0x3000000000000000:
1702             /* allocated grain */
1703             cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
1704                               ((cluster_sector & 0x0000ffffffffffff) << 12));
1705             cluster_sector = extent->sesparse_clusters_offset +
1706                 cluster_sector * extent->cluster_sectors;
1707             break;
1708         default:
1709             return VMDK_ERROR;
1710         }
1711     } else {
1712         cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
1713 
1714         if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1715             zeroed = true;
1716         }
1717     }
1718 
1719     if (!cluster_sector || zeroed) {
1720         if (!allocate) {
1721             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1722         }
1723         assert(!extent->sesparse);
1724 
1725         if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1726             return VMDK_ERROR;
1727         }
1728 
1729         cluster_sector = extent->next_cluster_sector;
1730         extent->next_cluster_sector += extent->cluster_sectors;
1731 
1732         /* First of all we write grain itself, to avoid race condition
1733          * that may to corrupt the image.
1734          * This problem may occur because of insufficient space on host disk
1735          * or inappropriate VM shutdown.
1736          */
1737         ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1738                                 offset, skip_start_bytes, skip_end_bytes,
1739                                 zeroed);
1740         if (ret) {
1741             return ret;
1742         }
1743         if (m_data) {
1744             m_data->new_allocation = true;
1745         }
1746     }
1747     *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1748     return VMDK_OK;
1749 }
1750 
1751 static VmdkExtent *find_extent(BDRVVmdkState *s,
1752                                 int64_t sector_num, VmdkExtent *start_hint)
1753 {
1754     VmdkExtent *extent = start_hint;
1755 
1756     if (!extent) {
1757         extent = &s->extents[0];
1758     }
1759     while (extent < &s->extents[s->num_extents]) {
1760         if (sector_num < extent->end_sector) {
1761             return extent;
1762         }
1763         extent++;
1764     }
1765     return NULL;
1766 }
1767 
1768 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1769                                                    int64_t offset)
1770 {
1771     uint64_t extent_begin_offset, extent_relative_offset;
1772     uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1773 
1774     extent_begin_offset =
1775         (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1776     extent_relative_offset = offset - extent_begin_offset;
1777     return extent_relative_offset % cluster_size;
1778 }
1779 
1780 static int coroutine_fn GRAPH_RDLOCK
1781 vmdk_co_block_status(BlockDriverState *bs, bool want_zero,
1782                      int64_t offset, int64_t bytes, int64_t *pnum,
1783                      int64_t *map, BlockDriverState **file)
1784 {
1785     BDRVVmdkState *s = bs->opaque;
1786     int64_t index_in_cluster, n, ret;
1787     uint64_t cluster_offset;
1788     VmdkExtent *extent;
1789 
1790     extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1791     if (!extent) {
1792         return -EIO;
1793     }
1794     qemu_co_mutex_lock(&s->lock);
1795     ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1796                              0, 0);
1797     qemu_co_mutex_unlock(&s->lock);
1798 
1799     index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1800     switch (ret) {
1801     case VMDK_ERROR:
1802         ret = -EIO;
1803         break;
1804     case VMDK_UNALLOC:
1805         ret = 0;
1806         break;
1807     case VMDK_ZEROED:
1808         ret = BDRV_BLOCK_ZERO;
1809         break;
1810     case VMDK_OK:
1811         ret = BDRV_BLOCK_DATA;
1812         if (!extent->compressed) {
1813             ret |= BDRV_BLOCK_OFFSET_VALID;
1814             *map = cluster_offset + index_in_cluster;
1815             if (extent->flat) {
1816                 ret |= BDRV_BLOCK_RECURSE;
1817             }
1818         } else {
1819             ret |= BDRV_BLOCK_COMPRESSED;
1820         }
1821         *file = extent->file->bs;
1822         break;
1823     }
1824 
1825     n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1826     *pnum = MIN(n, bytes);
1827     return ret;
1828 }
1829 
1830 static int coroutine_fn GRAPH_RDLOCK
1831 vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1832                   int64_t offset_in_cluster, QEMUIOVector *qiov,
1833                   uint64_t qiov_offset, uint64_t n_bytes,
1834                   uint64_t offset)
1835 {
1836     int ret;
1837     VmdkGrainMarker *data = NULL;
1838     uLongf buf_len;
1839     QEMUIOVector local_qiov;
1840     int64_t write_offset;
1841     int64_t write_end_sector;
1842 
1843     if (extent->compressed) {
1844         void *compressed_data;
1845 
1846         /* Only whole clusters */
1847         if (offset_in_cluster ||
1848             n_bytes > (extent->cluster_sectors * SECTOR_SIZE) ||
1849             (n_bytes < (extent->cluster_sectors * SECTOR_SIZE) &&
1850              offset + n_bytes != extent->end_sector * SECTOR_SIZE))
1851         {
1852             ret = -EINVAL;
1853             goto out;
1854         }
1855 
1856         if (!extent->has_marker) {
1857             ret = -EINVAL;
1858             goto out;
1859         }
1860         buf_len = (extent->cluster_sectors << 9) * 2;
1861         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1862 
1863         compressed_data = g_malloc(n_bytes);
1864         qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1865         ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1866         g_free(compressed_data);
1867 
1868         if (ret != Z_OK || buf_len == 0) {
1869             ret = -EINVAL;
1870             goto out;
1871         }
1872 
1873         data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1874         data->size = cpu_to_le32(buf_len);
1875 
1876         n_bytes = buf_len + sizeof(VmdkGrainMarker);
1877         qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1878 
1879         BLKDBG_CO_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1880     } else {
1881         qemu_iovec_init(&local_qiov, qiov->niov);
1882         qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1883 
1884         BLKDBG_CO_EVENT(extent->file, BLKDBG_WRITE_AIO);
1885     }
1886 
1887     write_offset = cluster_offset + offset_in_cluster;
1888     ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1889                           &local_qiov, 0);
1890 
1891     write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1892 
1893     if (extent->compressed) {
1894         extent->next_cluster_sector = write_end_sector;
1895     } else {
1896         extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1897                                           write_end_sector);
1898     }
1899 
1900     if (ret < 0) {
1901         goto out;
1902     }
1903     ret = 0;
1904  out:
1905     g_free(data);
1906     if (!extent->compressed) {
1907         qemu_iovec_destroy(&local_qiov);
1908     }
1909     return ret;
1910 }
1911 
1912 static int coroutine_fn GRAPH_RDLOCK
1913 vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1914                  int64_t offset_in_cluster, QEMUIOVector *qiov, int bytes)
1915 {
1916     int ret;
1917     int cluster_bytes, buf_bytes;
1918     uint8_t *cluster_buf, *compressed_data;
1919     uint8_t *uncomp_buf;
1920     uint32_t data_len;
1921     VmdkGrainMarker *marker;
1922     uLongf buf_len;
1923 
1924 
1925     if (!extent->compressed) {
1926         BLKDBG_CO_EVENT(extent->file, BLKDBG_READ_AIO);
1927         ret = bdrv_co_preadv(extent->file,
1928                              cluster_offset + offset_in_cluster, bytes,
1929                              qiov, 0);
1930         if (ret < 0) {
1931             return ret;
1932         }
1933         return 0;
1934     }
1935     cluster_bytes = extent->cluster_sectors * 512;
1936     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1937     buf_bytes = cluster_bytes * 2;
1938     cluster_buf = g_malloc(buf_bytes);
1939     uncomp_buf = g_malloc(cluster_bytes);
1940     BLKDBG_CO_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1941     ret = bdrv_co_pread(extent->file, cluster_offset, buf_bytes, cluster_buf,
1942                         0);
1943     if (ret < 0) {
1944         goto out;
1945     }
1946     compressed_data = cluster_buf;
1947     buf_len = cluster_bytes;
1948     data_len = cluster_bytes;
1949     if (extent->has_marker) {
1950         marker = (VmdkGrainMarker *)cluster_buf;
1951         compressed_data = marker->data;
1952         data_len = le32_to_cpu(marker->size);
1953     }
1954     if (!data_len || data_len > buf_bytes) {
1955         ret = -EINVAL;
1956         goto out;
1957     }
1958     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1959     if (ret != Z_OK) {
1960         ret = -EINVAL;
1961         goto out;
1962 
1963     }
1964     if (offset_in_cluster < 0 ||
1965             offset_in_cluster + bytes > buf_len) {
1966         ret = -EINVAL;
1967         goto out;
1968     }
1969     qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1970     ret = 0;
1971 
1972  out:
1973     g_free(uncomp_buf);
1974     g_free(cluster_buf);
1975     return ret;
1976 }
1977 
1978 static int coroutine_fn GRAPH_RDLOCK
1979 vmdk_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
1980                QEMUIOVector *qiov, BdrvRequestFlags flags)
1981 {
1982     BDRVVmdkState *s = bs->opaque;
1983     int ret;
1984     uint64_t n_bytes, offset_in_cluster;
1985     VmdkExtent *extent = NULL;
1986     QEMUIOVector local_qiov;
1987     uint64_t cluster_offset;
1988     uint64_t bytes_done = 0;
1989 
1990     qemu_iovec_init(&local_qiov, qiov->niov);
1991     qemu_co_mutex_lock(&s->lock);
1992 
1993     while (bytes > 0) {
1994         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1995         if (!extent) {
1996             ret = -EIO;
1997             goto fail;
1998         }
1999         ret = get_cluster_offset(bs, extent, NULL,
2000                                  offset, false, &cluster_offset, 0, 0);
2001         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
2002 
2003         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
2004                              - offset_in_cluster);
2005 
2006         if (ret != VMDK_OK) {
2007             /* if not allocated, try to read from parent image, if exist */
2008             if (bs->backing && ret != VMDK_ZEROED) {
2009                 if (!vmdk_is_cid_valid(bs)) {
2010                     ret = -EINVAL;
2011                     goto fail;
2012                 }
2013 
2014                 qemu_iovec_reset(&local_qiov);
2015                 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
2016 
2017                 /* qcow2 emits this on bs->file instead of bs->backing */
2018                 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
2019                 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
2020                                      &local_qiov, 0);
2021                 if (ret < 0) {
2022                     goto fail;
2023                 }
2024             } else {
2025                 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
2026             }
2027         } else {
2028             qemu_iovec_reset(&local_qiov);
2029             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
2030 
2031             ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
2032                                    &local_qiov, n_bytes);
2033             if (ret) {
2034                 goto fail;
2035             }
2036         }
2037         bytes -= n_bytes;
2038         offset += n_bytes;
2039         bytes_done += n_bytes;
2040     }
2041 
2042     ret = 0;
2043 fail:
2044     qemu_co_mutex_unlock(&s->lock);
2045     qemu_iovec_destroy(&local_qiov);
2046 
2047     return ret;
2048 }
2049 
2050 /**
2051  * vmdk_write:
2052  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
2053  *                if possible, otherwise return -ENOTSUP.
2054  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
2055  *                with each cluster. By dry run we can find if the zero write
2056  *                is possible without modifying image data.
2057  *
2058  * Returns: error code with 0 for success.
2059  */
2060 static int coroutine_fn GRAPH_RDLOCK
2061 vmdk_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2062              QEMUIOVector *qiov, bool zeroed, bool zero_dry_run)
2063 {
2064     BDRVVmdkState *s = bs->opaque;
2065     VmdkExtent *extent = NULL;
2066     int ret;
2067     int64_t offset_in_cluster, n_bytes;
2068     uint64_t cluster_offset;
2069     uint64_t bytes_done = 0;
2070     VmdkMetaData m_data;
2071 
2072     if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
2073         error_report("Wrong offset: offset=0x%" PRIx64
2074                      " total_sectors=0x%" PRIx64,
2075                      offset, bs->total_sectors);
2076         return -EIO;
2077     }
2078 
2079     while (bytes > 0) {
2080         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
2081         if (!extent) {
2082             return -EIO;
2083         }
2084         if (extent->sesparse) {
2085             return -ENOTSUP;
2086         }
2087         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
2088         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
2089                              - offset_in_cluster);
2090 
2091         ret = get_cluster_offset(bs, extent, &m_data, offset,
2092                                  !(extent->compressed || zeroed),
2093                                  &cluster_offset, offset_in_cluster,
2094                                  offset_in_cluster + n_bytes);
2095         if (extent->compressed) {
2096             if (ret == VMDK_OK) {
2097                 /* Refuse write to allocated cluster for streamOptimized */
2098                 error_report("Could not write to allocated cluster"
2099                               " for streamOptimized");
2100                 return -EIO;
2101             } else if (!zeroed) {
2102                 /* allocate */
2103                 ret = get_cluster_offset(bs, extent, &m_data, offset,
2104                                          true, &cluster_offset, 0, 0);
2105             }
2106         }
2107         if (ret == VMDK_ERROR) {
2108             return -EINVAL;
2109         }
2110         if (zeroed) {
2111             /* Do zeroed write, buf is ignored */
2112             if (extent->has_zero_grain &&
2113                     offset_in_cluster == 0 &&
2114                     n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
2115                 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
2116                 if (!zero_dry_run && ret != VMDK_ZEROED) {
2117                     /* update L2 tables */
2118                     if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
2119                             != VMDK_OK) {
2120                         return -EIO;
2121                     }
2122                 }
2123             } else {
2124                 return -ENOTSUP;
2125             }
2126         } else {
2127             ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
2128                                     qiov, bytes_done, n_bytes, offset);
2129             if (ret) {
2130                 return ret;
2131             }
2132             if (m_data.new_allocation) {
2133                 /* update L2 tables */
2134                 if (vmdk_L2update(extent, &m_data,
2135                                   cluster_offset >> BDRV_SECTOR_BITS)
2136                         != VMDK_OK) {
2137                     return -EIO;
2138                 }
2139             }
2140         }
2141         bytes -= n_bytes;
2142         offset += n_bytes;
2143         bytes_done += n_bytes;
2144 
2145         /* update CID on the first write every time the virtual disk is
2146          * opened */
2147         if (!s->cid_updated) {
2148             ret = vmdk_write_cid(bs, g_random_int());
2149             if (ret < 0) {
2150                 return ret;
2151             }
2152             s->cid_updated = true;
2153         }
2154     }
2155     return 0;
2156 }
2157 
2158 static int coroutine_fn GRAPH_RDLOCK
2159 vmdk_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
2160                 QEMUIOVector *qiov, BdrvRequestFlags flags)
2161 {
2162     int ret;
2163     BDRVVmdkState *s = bs->opaque;
2164     qemu_co_mutex_lock(&s->lock);
2165     ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
2166     qemu_co_mutex_unlock(&s->lock);
2167     return ret;
2168 }
2169 
2170 static int coroutine_fn GRAPH_RDLOCK
2171 vmdk_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes,
2172                            QEMUIOVector *qiov)
2173 {
2174     if (bytes == 0) {
2175         /* The caller will write bytes 0 to signal EOF.
2176          * When receive it, we align EOF to a sector boundary. */
2177         BDRVVmdkState *s = bs->opaque;
2178         int i, ret;
2179         int64_t length;
2180 
2181         for (i = 0; i < s->num_extents; i++) {
2182             length = bdrv_co_getlength(s->extents[i].file->bs);
2183             if (length < 0) {
2184                 return length;
2185             }
2186             length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
2187             ret = bdrv_co_truncate(s->extents[i].file, length, false,
2188                                    PREALLOC_MODE_OFF, 0, NULL);
2189             if (ret < 0) {
2190                 return ret;
2191             }
2192         }
2193         return 0;
2194     }
2195     return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
2196 }
2197 
2198 static int coroutine_fn GRAPH_RDLOCK
2199 vmdk_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
2200                       BdrvRequestFlags flags)
2201 {
2202     int ret;
2203     BDRVVmdkState *s = bs->opaque;
2204 
2205     qemu_co_mutex_lock(&s->lock);
2206     /* write zeroes could fail if sectors not aligned to cluster, test it with
2207      * dry_run == true before really updating image */
2208     ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
2209     if (!ret) {
2210         ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
2211     }
2212     qemu_co_mutex_unlock(&s->lock);
2213     return ret;
2214 }
2215 
2216 static int coroutine_fn GRAPH_UNLOCKED
2217 vmdk_init_extent(BlockBackend *blk, int64_t filesize, bool flat, bool compress,
2218                  bool zeroed_grain, Error **errp)
2219 {
2220     int ret, i;
2221     VMDK4Header header;
2222     uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
2223     uint32_t *gd_buf = NULL;
2224     int gd_buf_size;
2225 
2226     if (flat) {
2227         ret = blk_co_truncate(blk, filesize, false, PREALLOC_MODE_OFF, 0, errp);
2228         goto exit;
2229     }
2230     magic = cpu_to_be32(VMDK4_MAGIC);
2231     memset(&header, 0, sizeof(header));
2232     if (compress) {
2233         header.version = 3;
2234     } else if (zeroed_grain) {
2235         header.version = 2;
2236     } else {
2237         header.version = 1;
2238     }
2239     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
2240                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
2241                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
2242     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
2243     header.capacity = filesize / BDRV_SECTOR_SIZE;
2244     header.granularity = 128;
2245     header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
2246 
2247     grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
2248     gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
2249                            BDRV_SECTOR_SIZE);
2250     gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
2251     gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
2252 
2253     header.desc_offset = 1;
2254     header.desc_size = 20;
2255     header.rgd_offset = header.desc_offset + header.desc_size;
2256     header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
2257     header.grain_offset =
2258         ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
2259                  header.granularity);
2260     /* swap endianness for all header fields */
2261     header.version = cpu_to_le32(header.version);
2262     header.flags = cpu_to_le32(header.flags);
2263     header.capacity = cpu_to_le64(header.capacity);
2264     header.granularity = cpu_to_le64(header.granularity);
2265     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
2266     header.desc_offset = cpu_to_le64(header.desc_offset);
2267     header.desc_size = cpu_to_le64(header.desc_size);
2268     header.rgd_offset = cpu_to_le64(header.rgd_offset);
2269     header.gd_offset = cpu_to_le64(header.gd_offset);
2270     header.grain_offset = cpu_to_le64(header.grain_offset);
2271     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
2272 
2273     header.check_bytes[0] = 0xa;
2274     header.check_bytes[1] = 0x20;
2275     header.check_bytes[2] = 0xd;
2276     header.check_bytes[3] = 0xa;
2277 
2278     /* write all the data */
2279     ret = blk_co_pwrite(blk, 0, sizeof(magic), &magic, 0);
2280     if (ret < 0) {
2281         error_setg(errp, QERR_IO_ERROR);
2282         goto exit;
2283     }
2284     ret = blk_co_pwrite(blk, sizeof(magic), sizeof(header), &header, 0);
2285     if (ret < 0) {
2286         error_setg(errp, QERR_IO_ERROR);
2287         goto exit;
2288     }
2289 
2290     ret = blk_co_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false,
2291                           PREALLOC_MODE_OFF, 0, errp);
2292     if (ret < 0) {
2293         goto exit;
2294     }
2295 
2296     /* write grain directory */
2297     gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
2298     gd_buf = g_malloc0(gd_buf_size);
2299     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
2300          i < gt_count; i++, tmp += gt_size) {
2301         gd_buf[i] = cpu_to_le32(tmp);
2302     }
2303     ret = blk_co_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
2304                         gd_buf_size, gd_buf, 0);
2305     if (ret < 0) {
2306         error_setg(errp, QERR_IO_ERROR);
2307         goto exit;
2308     }
2309 
2310     /* write backup grain directory */
2311     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
2312          i < gt_count; i++, tmp += gt_size) {
2313         gd_buf[i] = cpu_to_le32(tmp);
2314     }
2315     ret = blk_co_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
2316                         gd_buf_size, gd_buf, 0);
2317     if (ret < 0) {
2318         error_setg(errp, QERR_IO_ERROR);
2319     }
2320 
2321     ret = 0;
2322 exit:
2323     g_free(gd_buf);
2324     return ret;
2325 }
2326 
2327 static int coroutine_fn GRAPH_UNLOCKED
2328 vmdk_create_extent(const char *filename, int64_t filesize, bool flat,
2329                    bool compress, bool zeroed_grain, BlockBackend **pbb,
2330                    QemuOpts *opts, Error **errp)
2331 {
2332     int ret;
2333     BlockBackend *blk = NULL;
2334 
2335     ret = bdrv_co_create_file(filename, opts, errp);
2336     if (ret < 0) {
2337         goto exit;
2338     }
2339 
2340     blk = blk_co_new_open(filename, NULL, NULL,
2341                           BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2342                           errp);
2343     if (blk == NULL) {
2344         ret = -EIO;
2345         goto exit;
2346     }
2347 
2348     blk_set_allow_write_beyond_eof(blk, true);
2349 
2350     ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
2351 exit:
2352     if (blk) {
2353         if (pbb) {
2354             *pbb = blk;
2355         } else {
2356             blk_co_unref(blk);
2357             blk = NULL;
2358         }
2359     }
2360     return ret;
2361 }
2362 
2363 static int filename_decompose(const char *filename, char *path, char *prefix,
2364                               char *postfix, size_t buf_len, Error **errp)
2365 {
2366     const char *p, *q;
2367 
2368     if (filename == NULL || !strlen(filename)) {
2369         error_setg(errp, "No filename provided");
2370         return VMDK_ERROR;
2371     }
2372     p = strrchr(filename, '/');
2373     if (p == NULL) {
2374         p = strrchr(filename, '\\');
2375     }
2376     if (p == NULL) {
2377         p = strrchr(filename, ':');
2378     }
2379     if (p != NULL) {
2380         p++;
2381         if (p - filename >= buf_len) {
2382             return VMDK_ERROR;
2383         }
2384         pstrcpy(path, p - filename + 1, filename);
2385     } else {
2386         p = filename;
2387         path[0] = '\0';
2388     }
2389     q = strrchr(p, '.');
2390     if (q == NULL) {
2391         pstrcpy(prefix, buf_len, p);
2392         postfix[0] = '\0';
2393     } else {
2394         if (q - p >= buf_len) {
2395             return VMDK_ERROR;
2396         }
2397         pstrcpy(prefix, q - p + 1, p);
2398         pstrcpy(postfix, buf_len, q);
2399     }
2400     return VMDK_OK;
2401 }
2402 
2403 /*
2404  * idx == 0: get or create the descriptor file (also the image file if in a
2405  *           non-split format.
2406  * idx >= 1: get the n-th extent if in a split subformat
2407  */
2408 typedef BlockBackend * coroutine_fn GRAPH_UNLOCKED_PTR
2409     (*vmdk_create_extent_fn)(int64_t size, int idx, bool flat, bool split,
2410                              bool compress, bool zeroed_grain, void *opaque,
2411                              Error **errp);
2412 
2413 static void vmdk_desc_add_extent(GString *desc,
2414                                  const char *extent_line_fmt,
2415                                  int64_t size, const char *filename)
2416 {
2417     char *basename = g_path_get_basename(filename);
2418 
2419     g_string_append_printf(desc, extent_line_fmt,
2420                            DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
2421     g_free(basename);
2422 }
2423 
2424 static int coroutine_fn GRAPH_UNLOCKED
2425 vmdk_co_do_create(int64_t size,
2426                   BlockdevVmdkSubformat subformat,
2427                   BlockdevVmdkAdapterType adapter_type,
2428                   const char *backing_file,
2429                   const char *hw_version,
2430                   const char *toolsversion,
2431                   bool compat6,
2432                   bool zeroed_grain,
2433                   vmdk_create_extent_fn extent_fn,
2434                   void *opaque,
2435                   Error **errp)
2436 {
2437     int extent_idx;
2438     BlockBackend *blk = NULL;
2439     BlockBackend *extent_blk;
2440     Error *local_err = NULL;
2441     char *desc = NULL;
2442     int ret = 0;
2443     bool flat, split, compress;
2444     GString *ext_desc_lines;
2445     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
2446     int64_t extent_size;
2447     int64_t created_size = 0;
2448     const char *extent_line_fmt;
2449     char *parent_desc_line = g_malloc0(BUF_SIZE);
2450     uint32_t parent_cid = 0xffffffff;
2451     uint32_t number_heads = 16;
2452     uint32_t desc_offset = 0, desc_len;
2453     const char desc_template[] =
2454         "# Disk DescriptorFile\n"
2455         "version=1\n"
2456         "CID=%" PRIx32 "\n"
2457         "parentCID=%" PRIx32 "\n"
2458         "createType=\"%s\"\n"
2459         "%s"
2460         "\n"
2461         "# Extent description\n"
2462         "%s"
2463         "\n"
2464         "# The Disk Data Base\n"
2465         "#DDB\n"
2466         "\n"
2467         "ddb.virtualHWVersion = \"%s\"\n"
2468         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2469         "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2470         "ddb.geometry.sectors = \"63\"\n"
2471         "ddb.adapterType = \"%s\"\n"
2472         "ddb.toolsVersion = \"%s\"\n";
2473 
2474     ext_desc_lines = g_string_new(NULL);
2475 
2476     /* Read out options */
2477     if (compat6) {
2478         if (hw_version) {
2479             error_setg(errp,
2480                        "compat6 cannot be enabled with hwversion set");
2481             ret = -EINVAL;
2482             goto exit;
2483         }
2484         hw_version = "6";
2485     }
2486     if (!hw_version) {
2487         hw_version = "4";
2488     }
2489     if (!toolsversion) {
2490         toolsversion = "2147483647";
2491     }
2492 
2493     if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2494         /* that's the number of heads with which vmware operates when
2495            creating, exporting, etc. vmdk files with a non-ide adapter type */
2496         number_heads = 255;
2497     }
2498     split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2499             (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2500     flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2501            (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2502     compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2503 
2504     if (flat) {
2505         extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2506     } else {
2507         extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2508     }
2509     if (flat && backing_file) {
2510         error_setg(errp, "Flat image can't have backing file");
2511         ret = -ENOTSUP;
2512         goto exit;
2513     }
2514     if (flat && zeroed_grain) {
2515         error_setg(errp, "Flat image can't enable zeroed grain");
2516         ret = -ENOTSUP;
2517         goto exit;
2518     }
2519 
2520     /* Create extents */
2521     if (split) {
2522         extent_size = split_size;
2523     } else {
2524         extent_size = size;
2525     }
2526     if (!split && !flat) {
2527         created_size = extent_size;
2528     } else {
2529         created_size = 0;
2530     }
2531     /* Get the descriptor file BDS */
2532     blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2533                     opaque, errp);
2534     if (!blk) {
2535         ret = -EIO;
2536         goto exit;
2537     }
2538     if (!split && !flat) {
2539         vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2540                              blk_bs(blk)->filename);
2541     }
2542 
2543     if (backing_file) {
2544         BlockBackend *backing;
2545         char *full_backing =
2546             bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2547                                                          backing_file,
2548                                                          &local_err);
2549         if (local_err) {
2550             error_propagate(errp, local_err);
2551             ret = -ENOENT;
2552             goto exit;
2553         }
2554         assert(full_backing);
2555 
2556         backing = blk_co_new_open(full_backing, NULL, NULL,
2557                                   BDRV_O_NO_BACKING, errp);
2558         g_free(full_backing);
2559         if (backing == NULL) {
2560             ret = -EIO;
2561             goto exit;
2562         }
2563         if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2564             error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2565                        blk_bs(backing)->drv->format_name);
2566             blk_co_unref(backing);
2567             ret = -EINVAL;
2568             goto exit;
2569         }
2570 
2571         bdrv_graph_co_rdlock();
2572         ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2573         bdrv_graph_co_rdunlock();
2574         blk_co_unref(backing);
2575         if (ret) {
2576             error_setg(errp, "Failed to read parent CID");
2577             goto exit;
2578         }
2579         snprintf(parent_desc_line, BUF_SIZE,
2580                 "parentFileNameHint=\"%s\"", backing_file);
2581     }
2582     extent_idx = 1;
2583     while (created_size < size) {
2584         int64_t cur_size = MIN(size - created_size, extent_size);
2585         extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2586                                zeroed_grain, opaque, errp);
2587         if (!extent_blk) {
2588             ret = -EINVAL;
2589             goto exit;
2590         }
2591         vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2592                              blk_bs(extent_blk)->filename);
2593         created_size += cur_size;
2594         extent_idx++;
2595         blk_co_unref(extent_blk);
2596     }
2597 
2598     /* Check whether we got excess extents */
2599     extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2600                            opaque, NULL);
2601     if (extent_blk) {
2602         blk_co_unref(extent_blk);
2603         error_setg(errp, "List of extents contains unused extents");
2604         ret = -EINVAL;
2605         goto exit;
2606     }
2607 
2608     /* generate descriptor file */
2609     desc = g_strdup_printf(desc_template,
2610                            g_random_int(),
2611                            parent_cid,
2612                            BlockdevVmdkSubformat_str(subformat),
2613                            parent_desc_line,
2614                            ext_desc_lines->str,
2615                            hw_version,
2616                            size /
2617                                (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2618                            number_heads,
2619                            BlockdevVmdkAdapterType_str(adapter_type),
2620                            toolsversion);
2621     desc_len = strlen(desc);
2622     /* the descriptor offset = 0x200 */
2623     if (!split && !flat) {
2624         desc_offset = 0x200;
2625     }
2626 
2627     ret = blk_co_pwrite(blk, desc_offset, desc_len, desc, 0);
2628     if (ret < 0) {
2629         error_setg_errno(errp, -ret, "Could not write description");
2630         goto exit;
2631     }
2632     /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2633      * for description file */
2634     if (desc_offset == 0) {
2635         ret = blk_co_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, 0, errp);
2636         if (ret < 0) {
2637             goto exit;
2638         }
2639     }
2640     ret = 0;
2641 exit:
2642     if (blk) {
2643         blk_co_unref(blk);
2644     }
2645     g_free(desc);
2646     g_free(parent_desc_line);
2647     g_string_free(ext_desc_lines, true);
2648     return ret;
2649 }
2650 
2651 typedef struct {
2652     char *path;
2653     char *prefix;
2654     char *postfix;
2655     QemuOpts *opts;
2656 } VMDKCreateOptsData;
2657 
2658 static BlockBackend * coroutine_fn GRAPH_UNLOCKED
2659 vmdk_co_create_opts_cb(int64_t size, int idx, bool flat, bool split,
2660                        bool compress, bool zeroed_grain, void *opaque,
2661                        Error **errp)
2662 {
2663     BlockBackend *blk = NULL;
2664     BlockDriverState *bs = NULL;
2665     VMDKCreateOptsData *data = opaque;
2666     char *ext_filename = NULL;
2667     char *rel_filename = NULL;
2668 
2669     /* We're done, don't create excess extents. */
2670     if (size == -1) {
2671         assert(errp == NULL);
2672         return NULL;
2673     }
2674 
2675     if (idx == 0) {
2676         rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2677     } else if (split) {
2678         rel_filename = g_strdup_printf("%s-%c%03d%s",
2679                                        data->prefix,
2680                                        flat ? 'f' : 's', idx, data->postfix);
2681     } else {
2682         assert(idx == 1);
2683         rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2684     }
2685 
2686     ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2687     g_free(rel_filename);
2688 
2689     if (vmdk_create_extent(ext_filename, size,
2690                            flat, compress, zeroed_grain, &blk, data->opts,
2691                            errp)) {
2692         goto exit;
2693     }
2694     bdrv_co_unref(bs);
2695 exit:
2696     g_free(ext_filename);
2697     return blk;
2698 }
2699 
2700 static int coroutine_fn GRAPH_UNLOCKED
2701 vmdk_co_create_opts(BlockDriver *drv, const char *filename,
2702                     QemuOpts *opts, Error **errp)
2703 {
2704     Error *local_err = NULL;
2705     char *desc = NULL;
2706     int64_t total_size = 0;
2707     char *adapter_type = NULL;
2708     BlockdevVmdkAdapterType adapter_type_enum;
2709     char *backing_file = NULL;
2710     char *hw_version = NULL;
2711     char *toolsversion = NULL;
2712     char *fmt = NULL;
2713     BlockdevVmdkSubformat subformat;
2714     int ret = 0;
2715     char *path = g_malloc0(PATH_MAX);
2716     char *prefix = g_malloc0(PATH_MAX);
2717     char *postfix = g_malloc0(PATH_MAX);
2718     char *desc_line = g_malloc0(BUF_SIZE);
2719     char *ext_filename = g_malloc0(PATH_MAX);
2720     char *desc_filename = g_malloc0(PATH_MAX);
2721     char *parent_desc_line = g_malloc0(BUF_SIZE);
2722     bool zeroed_grain;
2723     bool compat6;
2724     VMDKCreateOptsData data;
2725     char *backing_fmt = NULL;
2726 
2727     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2728     if (backing_fmt && strcmp(backing_fmt, "vmdk") != 0) {
2729         error_setg(errp, "backing_file must be a vmdk image");
2730         ret = -EINVAL;
2731         goto exit;
2732     }
2733 
2734     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
2735         ret = -EINVAL;
2736         goto exit;
2737     }
2738     /* Read out options */
2739     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2740                           BDRV_SECTOR_SIZE);
2741     adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
2742     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2743     hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
2744     toolsversion = qemu_opt_get_del(opts, BLOCK_OPT_TOOLSVERSION);
2745     compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false);
2746     if (strcmp(hw_version, "undefined") == 0) {
2747         g_free(hw_version);
2748         hw_version = NULL;
2749     }
2750     fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
2751     zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false);
2752 
2753     if (adapter_type) {
2754         adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup,
2755                                             adapter_type,
2756                                             BLOCKDEV_VMDK_ADAPTER_TYPE_IDE,
2757                                             &local_err);
2758         if (local_err) {
2759             error_propagate(errp, local_err);
2760             ret = -EINVAL;
2761             goto exit;
2762         }
2763     } else {
2764         adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE;
2765     }
2766 
2767     if (!fmt) {
2768         /* Default format to monolithicSparse */
2769         subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE;
2770     } else {
2771         subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup,
2772                                     fmt,
2773                                     BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE,
2774                                     &local_err);
2775         if (local_err) {
2776             error_propagate(errp, local_err);
2777             ret = -EINVAL;
2778             goto exit;
2779         }
2780     }
2781     data = (VMDKCreateOptsData){
2782         .prefix = prefix,
2783         .postfix = postfix,
2784         .path = path,
2785         .opts = opts,
2786     };
2787     ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum,
2788                             backing_file, hw_version, toolsversion, compat6,
2789                             zeroed_grain, vmdk_co_create_opts_cb, &data, errp);
2790 
2791 exit:
2792     g_free(backing_fmt);
2793     g_free(adapter_type);
2794     g_free(backing_file);
2795     g_free(hw_version);
2796     g_free(toolsversion);
2797     g_free(fmt);
2798     g_free(desc);
2799     g_free(path);
2800     g_free(prefix);
2801     g_free(postfix);
2802     g_free(desc_line);
2803     g_free(ext_filename);
2804     g_free(desc_filename);
2805     g_free(parent_desc_line);
2806     return ret;
2807 }
2808 
2809 static BlockBackend * coroutine_fn GRAPH_UNLOCKED
2810 vmdk_co_create_cb(int64_t size, int idx, bool flat, bool split, bool compress,
2811                   bool zeroed_grain, void *opaque, Error **errp)
2812 {
2813     int ret;
2814     BlockDriverState *bs;
2815     BlockBackend *blk;
2816     BlockdevCreateOptionsVmdk *opts = opaque;
2817 
2818     if (idx == 0) {
2819         bs = bdrv_co_open_blockdev_ref(opts->file, errp);
2820     } else {
2821         int i;
2822         BlockdevRefList *list = opts->extents;
2823         for (i = 1; i < idx; i++) {
2824             if (!list || !list->next) {
2825                 error_setg(errp, "Extent [%d] not specified", i);
2826                 return NULL;
2827             }
2828             list = list->next;
2829         }
2830         if (!list) {
2831             error_setg(errp, "Extent [%d] not specified", idx - 1);
2832             return NULL;
2833         }
2834         bs = bdrv_co_open_blockdev_ref(list->value, errp);
2835     }
2836     if (!bs) {
2837         return NULL;
2838     }
2839     blk = blk_co_new_with_bs(bs,
2840                              BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
2841                                 BLK_PERM_RESIZE,
2842                              BLK_PERM_ALL,
2843                              errp);
2844     if (!blk) {
2845         return NULL;
2846     }
2847     blk_set_allow_write_beyond_eof(blk, true);
2848     bdrv_co_unref(bs);
2849 
2850     if (size != -1) {
2851         ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2852         if (ret) {
2853             blk_co_unref(blk);
2854             blk = NULL;
2855         }
2856     }
2857     return blk;
2858 }
2859 
2860 static int coroutine_fn GRAPH_UNLOCKED
2861 vmdk_co_create(BlockdevCreateOptions *create_options, Error **errp)
2862 {
2863     BlockdevCreateOptionsVmdk *opts;
2864 
2865     opts = &create_options->u.vmdk;
2866 
2867     /* Validate options */
2868     if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2869         error_setg(errp, "Image size must be a multiple of 512 bytes");
2870         return -EINVAL;
2871     }
2872 
2873     return vmdk_co_do_create(opts->size,
2874                              opts->subformat,
2875                              opts->adapter_type,
2876                              opts->backing_file,
2877                              opts->hwversion,
2878                              opts->toolsversion,
2879                              false,
2880                              opts->zeroed_grain,
2881                              vmdk_co_create_cb,
2882                              opts, errp);
2883 }
2884 
2885 static void vmdk_close(BlockDriverState *bs)
2886 {
2887     BDRVVmdkState *s = bs->opaque;
2888 
2889     vmdk_free_extents(bs);
2890     g_free(s->create_type);
2891 
2892     migrate_del_blocker(&s->migration_blocker);
2893 }
2894 
2895 static int64_t coroutine_fn GRAPH_RDLOCK
2896 vmdk_co_get_allocated_file_size(BlockDriverState *bs)
2897 {
2898     int i;
2899     int64_t ret = 0;
2900     int64_t r;
2901     BDRVVmdkState *s = bs->opaque;
2902 
2903     ret = bdrv_co_get_allocated_file_size(bs->file->bs);
2904     if (ret < 0) {
2905         return ret;
2906     }
2907     for (i = 0; i < s->num_extents; i++) {
2908         if (s->extents[i].file == bs->file) {
2909             continue;
2910         }
2911         r = bdrv_co_get_allocated_file_size(s->extents[i].file->bs);
2912         if (r < 0) {
2913             return r;
2914         }
2915         ret += r;
2916     }
2917     return ret;
2918 }
2919 
2920 static int GRAPH_RDLOCK vmdk_has_zero_init(BlockDriverState *bs)
2921 {
2922     int i;
2923     BDRVVmdkState *s = bs->opaque;
2924 
2925     /* If has a flat extent and its underlying storage doesn't have zero init,
2926      * return 0. */
2927     for (i = 0; i < s->num_extents; i++) {
2928         if (s->extents[i].flat) {
2929             if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2930                 return 0;
2931             }
2932         }
2933     }
2934     return 1;
2935 }
2936 
2937 static VmdkExtentInfo * GRAPH_RDLOCK vmdk_get_extent_info(VmdkExtent *extent)
2938 {
2939     VmdkExtentInfo *info = g_new0(VmdkExtentInfo, 1);
2940 
2941     bdrv_refresh_filename(extent->file->bs);
2942     *info = (VmdkExtentInfo){
2943         .filename         = g_strdup(extent->file->bs->filename),
2944         .format           = g_strdup(extent->type),
2945         .virtual_size     = extent->sectors * BDRV_SECTOR_SIZE,
2946         .compressed       = extent->compressed,
2947         .has_compressed   = extent->compressed,
2948         .cluster_size     = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2949         .has_cluster_size = !extent->flat,
2950     };
2951 
2952     return info;
2953 }
2954 
2955 static int coroutine_fn GRAPH_RDLOCK
2956 vmdk_co_check(BlockDriverState *bs, BdrvCheckResult *result, BdrvCheckMode fix)
2957 {
2958     BDRVVmdkState *s = bs->opaque;
2959     VmdkExtent *extent = NULL;
2960     int64_t sector_num = 0;
2961     int64_t total_sectors = bdrv_co_nb_sectors(bs);
2962     int ret;
2963     uint64_t cluster_offset;
2964 
2965     if (fix) {
2966         return -ENOTSUP;
2967     }
2968 
2969     for (;;) {
2970         if (sector_num >= total_sectors) {
2971             return 0;
2972         }
2973         extent = find_extent(s, sector_num, extent);
2974         if (!extent) {
2975             fprintf(stderr,
2976                     "ERROR: could not find extent for sector %" PRId64 "\n",
2977                     sector_num);
2978             ret = -EINVAL;
2979             break;
2980         }
2981         ret = get_cluster_offset(bs, extent, NULL,
2982                                  sector_num << BDRV_SECTOR_BITS,
2983                                  false, &cluster_offset, 0, 0);
2984         if (ret == VMDK_ERROR) {
2985             fprintf(stderr,
2986                     "ERROR: could not get cluster_offset for sector %"
2987                     PRId64 "\n", sector_num);
2988             break;
2989         }
2990         if (ret == VMDK_OK) {
2991             int64_t extent_len = bdrv_co_getlength(extent->file->bs);
2992             if (extent_len < 0) {
2993                 fprintf(stderr,
2994                         "ERROR: could not get extent file length for sector %"
2995                         PRId64 "\n", sector_num);
2996                 ret = extent_len;
2997                 break;
2998             }
2999             if (cluster_offset >= extent_len) {
3000                 fprintf(stderr,
3001                         "ERROR: cluster offset for sector %"
3002                         PRId64 " points after EOF\n", sector_num);
3003                 ret = -EINVAL;
3004                 break;
3005             }
3006         }
3007         sector_num += extent->cluster_sectors;
3008     }
3009 
3010     result->corruptions++;
3011     return ret;
3012 }
3013 
3014 static ImageInfoSpecific * GRAPH_RDLOCK
3015 vmdk_get_specific_info(BlockDriverState *bs, Error **errp)
3016 {
3017     int i;
3018     BDRVVmdkState *s = bs->opaque;
3019     ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
3020     VmdkExtentInfoList **tail;
3021 
3022     *spec_info = (ImageInfoSpecific){
3023         .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
3024         .u = {
3025             .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
3026         },
3027     };
3028 
3029     *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
3030         .create_type = g_strdup(s->create_type),
3031         .cid = s->cid,
3032         .parent_cid = s->parent_cid,
3033     };
3034 
3035     tail = &spec_info->u.vmdk.data->extents;
3036     for (i = 0; i < s->num_extents; i++) {
3037         QAPI_LIST_APPEND(tail, vmdk_get_extent_info(&s->extents[i]));
3038     }
3039 
3040     return spec_info;
3041 }
3042 
3043 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
3044 {
3045     return a->flat == b->flat &&
3046            a->compressed == b->compressed &&
3047            (a->flat || a->cluster_sectors == b->cluster_sectors);
3048 }
3049 
3050 static int coroutine_fn
3051 vmdk_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3052 {
3053     int i;
3054     BDRVVmdkState *s = bs->opaque;
3055     assert(s->num_extents);
3056 
3057     /* See if we have multiple extents but they have different cases */
3058     for (i = 1; i < s->num_extents; i++) {
3059         if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
3060             return -ENOTSUP;
3061         }
3062     }
3063     bdi->needs_compressed_writes = s->extents[0].compressed;
3064     if (!s->extents[0].flat) {
3065         bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
3066     }
3067     return 0;
3068 }
3069 
3070 static void GRAPH_RDLOCK
3071 vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
3072                           bool backing_overridden)
3073 {
3074     /* No children but file and backing can be explicitly specified (TODO) */
3075     qdict_put(target, "file",
3076               qobject_ref(bs->file->bs->full_open_options));
3077 
3078     if (backing_overridden) {
3079         if (bs->backing) {
3080             qdict_put(target, "backing",
3081                       qobject_ref(bs->backing->bs->full_open_options));
3082         } else {
3083             qdict_put_null(target, "backing");
3084         }
3085     }
3086 }
3087 
3088 static QemuOptsList vmdk_create_opts = {
3089     .name = "vmdk-create-opts",
3090     .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
3091     .desc = {
3092         {
3093             .name = BLOCK_OPT_SIZE,
3094             .type = QEMU_OPT_SIZE,
3095             .help = "Virtual disk size"
3096         },
3097         {
3098             .name = BLOCK_OPT_ADAPTER_TYPE,
3099             .type = QEMU_OPT_STRING,
3100             .help = "Virtual adapter type, can be one of "
3101                     "ide (default), lsilogic, buslogic or legacyESX"
3102         },
3103         {
3104             .name = BLOCK_OPT_BACKING_FILE,
3105             .type = QEMU_OPT_STRING,
3106             .help = "File name of a base image"
3107         },
3108         {
3109             .name = BLOCK_OPT_BACKING_FMT,
3110             .type = QEMU_OPT_STRING,
3111             .help = "Must be 'vmdk' if present",
3112         },
3113         {
3114             .name = BLOCK_OPT_COMPAT6,
3115             .type = QEMU_OPT_BOOL,
3116             .help = "VMDK version 6 image",
3117             .def_value_str = "off"
3118         },
3119         {
3120             .name = BLOCK_OPT_HWVERSION,
3121             .type = QEMU_OPT_STRING,
3122             .help = "VMDK hardware version",
3123             .def_value_str = "undefined"
3124         },
3125         {
3126             .name = BLOCK_OPT_TOOLSVERSION,
3127             .type = QEMU_OPT_STRING,
3128             .help = "VMware guest tools version",
3129         },
3130         {
3131             .name = BLOCK_OPT_SUBFMT,
3132             .type = QEMU_OPT_STRING,
3133             .help =
3134                 "VMDK flat extent format, can be one of "
3135                 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
3136         },
3137         {
3138             .name = BLOCK_OPT_ZEROED_GRAIN,
3139             .type = QEMU_OPT_BOOL,
3140             .help = "Enable efficient zero writes "
3141                     "using the zeroed-grain GTE feature"
3142         },
3143         { /* end of list */ }
3144     }
3145 };
3146 
3147 static BlockDriver bdrv_vmdk = {
3148     .format_name                  = "vmdk",
3149     .instance_size                = sizeof(BDRVVmdkState),
3150     .bdrv_probe                   = vmdk_probe,
3151     .bdrv_open                    = vmdk_open,
3152     .bdrv_co_check                = vmdk_co_check,
3153     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
3154     .bdrv_reopen_commit           = vmdk_reopen_commit,
3155     .bdrv_reopen_abort            = vmdk_reopen_abort,
3156     .bdrv_child_perm              = bdrv_default_perms,
3157     .bdrv_co_preadv               = vmdk_co_preadv,
3158     .bdrv_co_pwritev              = vmdk_co_pwritev,
3159     .bdrv_co_pwritev_compressed   = vmdk_co_pwritev_compressed,
3160     .bdrv_co_pwrite_zeroes        = vmdk_co_pwrite_zeroes,
3161     .bdrv_close                   = vmdk_close,
3162     .bdrv_co_create_opts          = vmdk_co_create_opts,
3163     .bdrv_co_create               = vmdk_co_create,
3164     .bdrv_co_block_status         = vmdk_co_block_status,
3165     .bdrv_co_get_allocated_file_size = vmdk_co_get_allocated_file_size,
3166     .bdrv_has_zero_init           = vmdk_has_zero_init,
3167     .bdrv_get_specific_info       = vmdk_get_specific_info,
3168     .bdrv_refresh_limits          = vmdk_refresh_limits,
3169     .bdrv_co_get_info             = vmdk_co_get_info,
3170     .bdrv_gather_child_options    = vmdk_gather_child_options,
3171 
3172     .is_format                    = true,
3173     .supports_backing             = true,
3174     .create_opts                  = &vmdk_create_opts,
3175 };
3176 
3177 static void bdrv_vmdk_init(void)
3178 {
3179     bdrv_register(&bdrv_vmdk);
3180 }
3181 
3182 block_init(bdrv_vmdk_init);
3183