xref: /openbmc/qemu/block/vmdk.c (revision ccb23709)
1 /*
2  * Block driver for the VMDK format
3  *
4  * Copyright (c) 2004 Fabrice Bellard
5  * Copyright (c) 2005 Filip Navara
6  *
7  * Permission is hereby granted, free of charge, to any person obtaining a copy
8  * of this software and associated documentation files (the "Software"), to deal
9  * in the Software without restriction, including without limitation the rights
10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11  * copies of the Software, and to permit persons to whom the Software is
12  * furnished to do so, subject to the following conditions:
13  *
14  * The above copyright notice and this permission notice shall be included in
15  * all copies or substantial portions of the Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23  * THE SOFTWARE.
24  */
25 
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "block/block_int.h"
29 #include "sysemu/block-backend.h"
30 #include "qapi/qmp/qdict.h"
31 #include "qapi/qmp/qerror.h"
32 #include "qemu/error-report.h"
33 #include "qemu/module.h"
34 #include "qemu/option.h"
35 #include "qemu/bswap.h"
36 #include "migration/blocker.h"
37 #include "qemu/cutils.h"
38 #include <zlib.h>
39 
40 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
41 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
42 #define VMDK4_COMPRESSION_DEFLATE 1
43 #define VMDK4_FLAG_NL_DETECT (1 << 0)
44 #define VMDK4_FLAG_RGD (1 << 1)
45 /* Zeroed-grain enable bit */
46 #define VMDK4_FLAG_ZERO_GRAIN   (1 << 2)
47 #define VMDK4_FLAG_COMPRESS (1 << 16)
48 #define VMDK4_FLAG_MARKER (1 << 17)
49 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
50 
51 #define VMDK_EXTENT_MAX_SECTORS (1ULL << 32)
52 
53 #define VMDK_GTE_ZEROED 0x1
54 
55 /* VMDK internal error codes */
56 #define VMDK_OK      0
57 #define VMDK_ERROR   (-1)
58 /* Cluster not allocated */
59 #define VMDK_UNALLOC (-2)
60 #define VMDK_ZEROED  (-3)
61 
62 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
63 
64 typedef struct {
65     uint32_t version;
66     uint32_t flags;
67     uint32_t disk_sectors;
68     uint32_t granularity;
69     uint32_t l1dir_offset;
70     uint32_t l1dir_size;
71     uint32_t file_sectors;
72     uint32_t cylinders;
73     uint32_t heads;
74     uint32_t sectors_per_track;
75 } QEMU_PACKED VMDK3Header;
76 
77 typedef struct {
78     uint32_t version;
79     uint32_t flags;
80     uint64_t capacity;
81     uint64_t granularity;
82     uint64_t desc_offset;
83     uint64_t desc_size;
84     /* Number of GrainTableEntries per GrainTable */
85     uint32_t num_gtes_per_gt;
86     uint64_t rgd_offset;
87     uint64_t gd_offset;
88     uint64_t grain_offset;
89     char filler[1];
90     char check_bytes[4];
91     uint16_t compressAlgorithm;
92 } QEMU_PACKED VMDK4Header;
93 
94 typedef struct VMDKSESparseConstHeader {
95     uint64_t magic;
96     uint64_t version;
97     uint64_t capacity;
98     uint64_t grain_size;
99     uint64_t grain_table_size;
100     uint64_t flags;
101     uint64_t reserved1;
102     uint64_t reserved2;
103     uint64_t reserved3;
104     uint64_t reserved4;
105     uint64_t volatile_header_offset;
106     uint64_t volatile_header_size;
107     uint64_t journal_header_offset;
108     uint64_t journal_header_size;
109     uint64_t journal_offset;
110     uint64_t journal_size;
111     uint64_t grain_dir_offset;
112     uint64_t grain_dir_size;
113     uint64_t grain_tables_offset;
114     uint64_t grain_tables_size;
115     uint64_t free_bitmap_offset;
116     uint64_t free_bitmap_size;
117     uint64_t backmap_offset;
118     uint64_t backmap_size;
119     uint64_t grains_offset;
120     uint64_t grains_size;
121     uint8_t pad[304];
122 } QEMU_PACKED VMDKSESparseConstHeader;
123 
124 typedef struct VMDKSESparseVolatileHeader {
125     uint64_t magic;
126     uint64_t free_gt_number;
127     uint64_t next_txn_seq_number;
128     uint64_t replay_journal;
129     uint8_t pad[480];
130 } QEMU_PACKED VMDKSESparseVolatileHeader;
131 
132 #define L2_CACHE_SIZE 16
133 
134 typedef struct VmdkExtent {
135     BdrvChild *file;
136     bool flat;
137     bool compressed;
138     bool has_marker;
139     bool has_zero_grain;
140     bool sesparse;
141     uint64_t sesparse_l2_tables_offset;
142     uint64_t sesparse_clusters_offset;
143     int32_t entry_size;
144     int version;
145     int64_t sectors;
146     int64_t end_sector;
147     int64_t flat_start_offset;
148     int64_t l1_table_offset;
149     int64_t l1_backup_table_offset;
150     void *l1_table;
151     uint32_t *l1_backup_table;
152     unsigned int l1_size;
153     uint32_t l1_entry_sectors;
154 
155     unsigned int l2_size;
156     void *l2_cache;
157     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
158     uint32_t l2_cache_counts[L2_CACHE_SIZE];
159 
160     int64_t cluster_sectors;
161     int64_t next_cluster_sector;
162     char *type;
163 } VmdkExtent;
164 
165 typedef struct BDRVVmdkState {
166     CoMutex lock;
167     uint64_t desc_offset;
168     bool cid_updated;
169     bool cid_checked;
170     uint32_t cid;
171     uint32_t parent_cid;
172     int num_extents;
173     /* Extent array with num_extents entries, ascend ordered by address */
174     VmdkExtent *extents;
175     Error *migration_blocker;
176     char *create_type;
177 } BDRVVmdkState;
178 
179 typedef struct VmdkMetaData {
180     unsigned int l1_index;
181     unsigned int l2_index;
182     unsigned int l2_offset;
183     bool new_allocation;
184     uint32_t *l2_cache_entry;
185 } VmdkMetaData;
186 
187 typedef struct VmdkGrainMarker {
188     uint64_t lba;
189     uint32_t size;
190     uint8_t  data[];
191 } QEMU_PACKED VmdkGrainMarker;
192 
193 enum {
194     MARKER_END_OF_STREAM    = 0,
195     MARKER_GRAIN_TABLE      = 1,
196     MARKER_GRAIN_DIRECTORY  = 2,
197     MARKER_FOOTER           = 3,
198 };
199 
200 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
201 {
202     uint32_t magic;
203 
204     if (buf_size < 4) {
205         return 0;
206     }
207     magic = be32_to_cpu(*(uint32_t *)buf);
208     if (magic == VMDK3_MAGIC ||
209         magic == VMDK4_MAGIC) {
210         return 100;
211     } else {
212         const char *p = (const char *)buf;
213         const char *end = p + buf_size;
214         while (p < end) {
215             if (*p == '#') {
216                 /* skip comment line */
217                 while (p < end && *p != '\n') {
218                     p++;
219                 }
220                 p++;
221                 continue;
222             }
223             if (*p == ' ') {
224                 while (p < end && *p == ' ') {
225                     p++;
226                 }
227                 /* skip '\r' if windows line endings used. */
228                 if (p < end && *p == '\r') {
229                     p++;
230                 }
231                 /* only accept blank lines before 'version=' line */
232                 if (p == end || *p != '\n') {
233                     return 0;
234                 }
235                 p++;
236                 continue;
237             }
238             if (end - p >= strlen("version=X\n")) {
239                 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
240                     strncmp("version=2\n", p, strlen("version=2\n")) == 0 ||
241                     strncmp("version=3\n", p, strlen("version=3\n")) == 0) {
242                     return 100;
243                 }
244             }
245             if (end - p >= strlen("version=X\r\n")) {
246                 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
247                     strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0 ||
248                     strncmp("version=3\r\n", p, strlen("version=3\r\n")) == 0) {
249                     return 100;
250                 }
251             }
252             return 0;
253         }
254         return 0;
255     }
256 }
257 
258 #define SECTOR_SIZE 512
259 #define DESC_SIZE (20 * SECTOR_SIZE)    /* 20 sectors of 512 bytes each */
260 #define BUF_SIZE 4096
261 #define HEADER_SIZE 512                 /* first sector of 512 bytes */
262 
263 static void vmdk_free_extents(BlockDriverState *bs)
264 {
265     int i;
266     BDRVVmdkState *s = bs->opaque;
267     VmdkExtent *e;
268 
269     for (i = 0; i < s->num_extents; i++) {
270         e = &s->extents[i];
271         g_free(e->l1_table);
272         g_free(e->l2_cache);
273         g_free(e->l1_backup_table);
274         g_free(e->type);
275         if (e->file != bs->file) {
276             bdrv_unref_child(bs, e->file);
277         }
278     }
279     g_free(s->extents);
280 }
281 
282 static void vmdk_free_last_extent(BlockDriverState *bs)
283 {
284     BDRVVmdkState *s = bs->opaque;
285 
286     if (s->num_extents == 0) {
287         return;
288     }
289     s->num_extents--;
290     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
291 }
292 
293 /* Return -ve errno, or 0 on success and write CID into *pcid. */
294 static int vmdk_read_cid(BlockDriverState *bs, int parent, uint32_t *pcid)
295 {
296     char *desc;
297     uint32_t cid;
298     const char *p_name, *cid_str;
299     size_t cid_str_size;
300     BDRVVmdkState *s = bs->opaque;
301     int ret;
302 
303     desc = g_malloc0(DESC_SIZE);
304     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
305     if (ret < 0) {
306         goto out;
307     }
308 
309     if (parent) {
310         cid_str = "parentCID";
311         cid_str_size = sizeof("parentCID");
312     } else {
313         cid_str = "CID";
314         cid_str_size = sizeof("CID");
315     }
316 
317     desc[DESC_SIZE - 1] = '\0';
318     p_name = strstr(desc, cid_str);
319     if (p_name == NULL) {
320         ret = -EINVAL;
321         goto out;
322     }
323     p_name += cid_str_size;
324     if (sscanf(p_name, "%" SCNx32, &cid) != 1) {
325         ret = -EINVAL;
326         goto out;
327     }
328     *pcid = cid;
329     ret = 0;
330 
331 out:
332     g_free(desc);
333     return ret;
334 }
335 
336 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
337 {
338     char *desc, *tmp_desc;
339     char *p_name, *tmp_str;
340     BDRVVmdkState *s = bs->opaque;
341     int ret = 0;
342 
343     desc = g_malloc0(DESC_SIZE);
344     tmp_desc = g_malloc0(DESC_SIZE);
345     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
346     if (ret < 0) {
347         goto out;
348     }
349 
350     desc[DESC_SIZE - 1] = '\0';
351     tmp_str = strstr(desc, "parentCID");
352     if (tmp_str == NULL) {
353         ret = -EINVAL;
354         goto out;
355     }
356 
357     pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
358     p_name = strstr(desc, "CID");
359     if (p_name != NULL) {
360         p_name += sizeof("CID");
361         snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
362         pstrcat(desc, DESC_SIZE, tmp_desc);
363     }
364 
365     ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
366 
367 out:
368     g_free(desc);
369     g_free(tmp_desc);
370     return ret;
371 }
372 
373 static int vmdk_is_cid_valid(BlockDriverState *bs)
374 {
375     BDRVVmdkState *s = bs->opaque;
376     uint32_t cur_pcid;
377 
378     if (!s->cid_checked && bs->backing) {
379         BlockDriverState *p_bs = bs->backing->bs;
380 
381         if (strcmp(p_bs->drv->format_name, "vmdk")) {
382             /* Backing file is not in vmdk format, so it does not have
383              * a CID, which makes the overlay's parent CID invalid */
384             return 0;
385         }
386 
387         if (vmdk_read_cid(p_bs, 0, &cur_pcid) != 0) {
388             /* read failure: report as not valid */
389             return 0;
390         }
391         if (s->parent_cid != cur_pcid) {
392             /* CID not valid */
393             return 0;
394         }
395     }
396     s->cid_checked = true;
397     /* CID valid */
398     return 1;
399 }
400 
401 /* We have nothing to do for VMDK reopen, stubs just return success */
402 static int vmdk_reopen_prepare(BDRVReopenState *state,
403                                BlockReopenQueue *queue, Error **errp)
404 {
405     assert(state != NULL);
406     assert(state->bs != NULL);
407     return 0;
408 }
409 
410 static int vmdk_parent_open(BlockDriverState *bs)
411 {
412     char *p_name;
413     char *desc;
414     BDRVVmdkState *s = bs->opaque;
415     int ret;
416 
417     desc = g_malloc0(DESC_SIZE + 1);
418     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
419     if (ret < 0) {
420         goto out;
421     }
422     ret = 0;
423 
424     p_name = strstr(desc, "parentFileNameHint");
425     if (p_name != NULL) {
426         char *end_name;
427 
428         p_name += sizeof("parentFileNameHint") + 1;
429         end_name = strchr(p_name, '\"');
430         if (end_name == NULL) {
431             ret = -EINVAL;
432             goto out;
433         }
434         if ((end_name - p_name) > sizeof(bs->auto_backing_file) - 1) {
435             ret = -EINVAL;
436             goto out;
437         }
438 
439         pstrcpy(bs->auto_backing_file, end_name - p_name + 1, p_name);
440         pstrcpy(bs->backing_file, sizeof(bs->backing_file),
441                 bs->auto_backing_file);
442         pstrcpy(bs->backing_format, sizeof(bs->backing_format),
443                 "vmdk");
444     }
445 
446 out:
447     g_free(desc);
448     return ret;
449 }
450 
451 /* Create and append extent to the extent array. Return the added VmdkExtent
452  * address. return NULL if allocation failed. */
453 static int vmdk_add_extent(BlockDriverState *bs,
454                            BdrvChild *file, bool flat, int64_t sectors,
455                            int64_t l1_offset, int64_t l1_backup_offset,
456                            uint32_t l1_size,
457                            int l2_size, uint64_t cluster_sectors,
458                            VmdkExtent **new_extent,
459                            Error **errp)
460 {
461     VmdkExtent *extent;
462     BDRVVmdkState *s = bs->opaque;
463     int64_t nb_sectors;
464 
465     if (cluster_sectors > 0x200000) {
466         /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
467         error_setg(errp, "Invalid granularity, image may be corrupt");
468         return -EFBIG;
469     }
470     if (l1_size > 32 * 1024 * 1024) {
471         /*
472          * Although with big capacity and small l1_entry_sectors, we can get a
473          * big l1_size, we don't want unbounded value to allocate the table.
474          * Limit it to 32M, which is enough to store:
475          *     8TB  - for both VMDK3 & VMDK4 with
476          *            minimal cluster size: 512B
477          *            minimal L2 table size: 512 entries
478          *            8 TB is still more than the maximal value supported for
479          *            VMDK3 & VMDK4 which is 2TB.
480          *     64TB - for "ESXi seSparse Extent"
481          *            minimal cluster size: 512B (default is 4KB)
482          *            L2 table size: 4096 entries (const).
483          *            64TB is more than the maximal value supported for
484          *            seSparse VMDKs (which is slightly less than 64TB)
485          */
486         error_setg(errp, "L1 size too big");
487         return -EFBIG;
488     }
489 
490     nb_sectors = bdrv_nb_sectors(file->bs);
491     if (nb_sectors < 0) {
492         return nb_sectors;
493     }
494 
495     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
496     extent = &s->extents[s->num_extents];
497     s->num_extents++;
498 
499     memset(extent, 0, sizeof(VmdkExtent));
500     extent->file = file;
501     extent->flat = flat;
502     extent->sectors = sectors;
503     extent->l1_table_offset = l1_offset;
504     extent->l1_backup_table_offset = l1_backup_offset;
505     extent->l1_size = l1_size;
506     extent->l1_entry_sectors = l2_size * cluster_sectors;
507     extent->l2_size = l2_size;
508     extent->cluster_sectors = flat ? sectors : cluster_sectors;
509     extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
510     extent->entry_size = sizeof(uint32_t);
511 
512     if (s->num_extents > 1) {
513         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
514     } else {
515         extent->end_sector = extent->sectors;
516     }
517     bs->total_sectors = extent->end_sector;
518     if (new_extent) {
519         *new_extent = extent;
520     }
521     return 0;
522 }
523 
524 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
525                             Error **errp)
526 {
527     int ret;
528     size_t l1_size;
529     int i;
530 
531     /* read the L1 table */
532     l1_size = extent->l1_size * extent->entry_size;
533     extent->l1_table = g_try_malloc(l1_size);
534     if (l1_size && extent->l1_table == NULL) {
535         return -ENOMEM;
536     }
537 
538     ret = bdrv_pread(extent->file,
539                      extent->l1_table_offset,
540                      extent->l1_table,
541                      l1_size);
542     if (ret < 0) {
543         bdrv_refresh_filename(extent->file->bs);
544         error_setg_errno(errp, -ret,
545                          "Could not read l1 table from extent '%s'",
546                          extent->file->bs->filename);
547         goto fail_l1;
548     }
549     for (i = 0; i < extent->l1_size; i++) {
550         if (extent->entry_size == sizeof(uint64_t)) {
551             le64_to_cpus((uint64_t *)extent->l1_table + i);
552         } else {
553             assert(extent->entry_size == sizeof(uint32_t));
554             le32_to_cpus((uint32_t *)extent->l1_table + i);
555         }
556     }
557 
558     if (extent->l1_backup_table_offset) {
559         assert(!extent->sesparse);
560         extent->l1_backup_table = g_try_malloc(l1_size);
561         if (l1_size && extent->l1_backup_table == NULL) {
562             ret = -ENOMEM;
563             goto fail_l1;
564         }
565         ret = bdrv_pread(extent->file,
566                          extent->l1_backup_table_offset,
567                          extent->l1_backup_table,
568                          l1_size);
569         if (ret < 0) {
570             bdrv_refresh_filename(extent->file->bs);
571             error_setg_errno(errp, -ret,
572                              "Could not read l1 backup table from extent '%s'",
573                              extent->file->bs->filename);
574             goto fail_l1b;
575         }
576         for (i = 0; i < extent->l1_size; i++) {
577             le32_to_cpus(&extent->l1_backup_table[i]);
578         }
579     }
580 
581     extent->l2_cache =
582         g_malloc(extent->entry_size * extent->l2_size * L2_CACHE_SIZE);
583     return 0;
584  fail_l1b:
585     g_free(extent->l1_backup_table);
586  fail_l1:
587     g_free(extent->l1_table);
588     return ret;
589 }
590 
591 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
592                                  BdrvChild *file,
593                                  int flags, Error **errp)
594 {
595     int ret;
596     uint32_t magic;
597     VMDK3Header header;
598     VmdkExtent *extent;
599 
600     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
601     if (ret < 0) {
602         bdrv_refresh_filename(file->bs);
603         error_setg_errno(errp, -ret,
604                          "Could not read header from file '%s'",
605                          file->bs->filename);
606         return ret;
607     }
608     ret = vmdk_add_extent(bs, file, false,
609                           le32_to_cpu(header.disk_sectors),
610                           (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
611                           0,
612                           le32_to_cpu(header.l1dir_size),
613                           4096,
614                           le32_to_cpu(header.granularity),
615                           &extent,
616                           errp);
617     if (ret < 0) {
618         return ret;
619     }
620     ret = vmdk_init_tables(bs, extent, errp);
621     if (ret) {
622         /* free extent allocated by vmdk_add_extent */
623         vmdk_free_last_extent(bs);
624     }
625     return ret;
626 }
627 
628 #define SESPARSE_CONST_HEADER_MAGIC UINT64_C(0x00000000cafebabe)
629 #define SESPARSE_VOLATILE_HEADER_MAGIC UINT64_C(0x00000000cafecafe)
630 
631 /* Strict checks - format not officially documented */
632 static int check_se_sparse_const_header(VMDKSESparseConstHeader *header,
633                                         Error **errp)
634 {
635     header->magic = le64_to_cpu(header->magic);
636     header->version = le64_to_cpu(header->version);
637     header->grain_size = le64_to_cpu(header->grain_size);
638     header->grain_table_size = le64_to_cpu(header->grain_table_size);
639     header->flags = le64_to_cpu(header->flags);
640     header->reserved1 = le64_to_cpu(header->reserved1);
641     header->reserved2 = le64_to_cpu(header->reserved2);
642     header->reserved3 = le64_to_cpu(header->reserved3);
643     header->reserved4 = le64_to_cpu(header->reserved4);
644 
645     header->volatile_header_offset =
646         le64_to_cpu(header->volatile_header_offset);
647     header->volatile_header_size = le64_to_cpu(header->volatile_header_size);
648 
649     header->journal_header_offset = le64_to_cpu(header->journal_header_offset);
650     header->journal_header_size = le64_to_cpu(header->journal_header_size);
651 
652     header->journal_offset = le64_to_cpu(header->journal_offset);
653     header->journal_size = le64_to_cpu(header->journal_size);
654 
655     header->grain_dir_offset = le64_to_cpu(header->grain_dir_offset);
656     header->grain_dir_size = le64_to_cpu(header->grain_dir_size);
657 
658     header->grain_tables_offset = le64_to_cpu(header->grain_tables_offset);
659     header->grain_tables_size = le64_to_cpu(header->grain_tables_size);
660 
661     header->free_bitmap_offset = le64_to_cpu(header->free_bitmap_offset);
662     header->free_bitmap_size = le64_to_cpu(header->free_bitmap_size);
663 
664     header->backmap_offset = le64_to_cpu(header->backmap_offset);
665     header->backmap_size = le64_to_cpu(header->backmap_size);
666 
667     header->grains_offset = le64_to_cpu(header->grains_offset);
668     header->grains_size = le64_to_cpu(header->grains_size);
669 
670     if (header->magic != SESPARSE_CONST_HEADER_MAGIC) {
671         error_setg(errp, "Bad const header magic: 0x%016" PRIx64,
672                    header->magic);
673         return -EINVAL;
674     }
675 
676     if (header->version != 0x0000000200000001) {
677         error_setg(errp, "Unsupported version: 0x%016" PRIx64,
678                    header->version);
679         return -ENOTSUP;
680     }
681 
682     if (header->grain_size != 8) {
683         error_setg(errp, "Unsupported grain size: %" PRIu64,
684                    header->grain_size);
685         return -ENOTSUP;
686     }
687 
688     if (header->grain_table_size != 64) {
689         error_setg(errp, "Unsupported grain table size: %" PRIu64,
690                    header->grain_table_size);
691         return -ENOTSUP;
692     }
693 
694     if (header->flags != 0) {
695         error_setg(errp, "Unsupported flags: 0x%016" PRIx64,
696                    header->flags);
697         return -ENOTSUP;
698     }
699 
700     if (header->reserved1 != 0 || header->reserved2 != 0 ||
701         header->reserved3 != 0 || header->reserved4 != 0) {
702         error_setg(errp, "Unsupported reserved bits:"
703                    " 0x%016" PRIx64 " 0x%016" PRIx64
704                    " 0x%016" PRIx64 " 0x%016" PRIx64,
705                    header->reserved1, header->reserved2,
706                    header->reserved3, header->reserved4);
707         return -ENOTSUP;
708     }
709 
710     /* check that padding is 0 */
711     if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
712         error_setg(errp, "Unsupported non-zero const header padding");
713         return -ENOTSUP;
714     }
715 
716     return 0;
717 }
718 
719 static int check_se_sparse_volatile_header(VMDKSESparseVolatileHeader *header,
720                                            Error **errp)
721 {
722     header->magic = le64_to_cpu(header->magic);
723     header->free_gt_number = le64_to_cpu(header->free_gt_number);
724     header->next_txn_seq_number = le64_to_cpu(header->next_txn_seq_number);
725     header->replay_journal = le64_to_cpu(header->replay_journal);
726 
727     if (header->magic != SESPARSE_VOLATILE_HEADER_MAGIC) {
728         error_setg(errp, "Bad volatile header magic: 0x%016" PRIx64,
729                    header->magic);
730         return -EINVAL;
731     }
732 
733     if (header->replay_journal) {
734         error_setg(errp, "Image is dirty, Replaying journal not supported");
735         return -ENOTSUP;
736     }
737 
738     /* check that padding is 0 */
739     if (!buffer_is_zero(header->pad, sizeof(header->pad))) {
740         error_setg(errp, "Unsupported non-zero volatile header padding");
741         return -ENOTSUP;
742     }
743 
744     return 0;
745 }
746 
747 static int vmdk_open_se_sparse(BlockDriverState *bs,
748                                BdrvChild *file,
749                                int flags, Error **errp)
750 {
751     int ret;
752     VMDKSESparseConstHeader const_header;
753     VMDKSESparseVolatileHeader volatile_header;
754     VmdkExtent *extent;
755 
756     ret = bdrv_apply_auto_read_only(bs,
757             "No write support for seSparse images available", errp);
758     if (ret < 0) {
759         return ret;
760     }
761 
762     assert(sizeof(const_header) == SECTOR_SIZE);
763 
764     ret = bdrv_pread(file, 0, &const_header, sizeof(const_header));
765     if (ret < 0) {
766         bdrv_refresh_filename(file->bs);
767         error_setg_errno(errp, -ret,
768                          "Could not read const header from file '%s'",
769                          file->bs->filename);
770         return ret;
771     }
772 
773     /* check const header */
774     ret = check_se_sparse_const_header(&const_header, errp);
775     if (ret < 0) {
776         return ret;
777     }
778 
779     assert(sizeof(volatile_header) == SECTOR_SIZE);
780 
781     ret = bdrv_pread(file,
782                      const_header.volatile_header_offset * SECTOR_SIZE,
783                      &volatile_header, sizeof(volatile_header));
784     if (ret < 0) {
785         bdrv_refresh_filename(file->bs);
786         error_setg_errno(errp, -ret,
787                          "Could not read volatile header from file '%s'",
788                          file->bs->filename);
789         return ret;
790     }
791 
792     /* check volatile header */
793     ret = check_se_sparse_volatile_header(&volatile_header, errp);
794     if (ret < 0) {
795         return ret;
796     }
797 
798     ret = vmdk_add_extent(bs, file, false,
799                           const_header.capacity,
800                           const_header.grain_dir_offset * SECTOR_SIZE,
801                           0,
802                           const_header.grain_dir_size *
803                           SECTOR_SIZE / sizeof(uint64_t),
804                           const_header.grain_table_size *
805                           SECTOR_SIZE / sizeof(uint64_t),
806                           const_header.grain_size,
807                           &extent,
808                           errp);
809     if (ret < 0) {
810         return ret;
811     }
812 
813     extent->sesparse = true;
814     extent->sesparse_l2_tables_offset = const_header.grain_tables_offset;
815     extent->sesparse_clusters_offset = const_header.grains_offset;
816     extent->entry_size = sizeof(uint64_t);
817 
818     ret = vmdk_init_tables(bs, extent, errp);
819     if (ret) {
820         /* free extent allocated by vmdk_add_extent */
821         vmdk_free_last_extent(bs);
822     }
823 
824     return ret;
825 }
826 
827 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
828                                QDict *options, Error **errp);
829 
830 static char *vmdk_read_desc(BdrvChild *file, uint64_t desc_offset, Error **errp)
831 {
832     int64_t size;
833     char *buf;
834     int ret;
835 
836     size = bdrv_getlength(file->bs);
837     if (size < 0) {
838         error_setg_errno(errp, -size, "Could not access file");
839         return NULL;
840     }
841 
842     if (size < 4) {
843         /* Both descriptor file and sparse image must be much larger than 4
844          * bytes, also callers of vmdk_read_desc want to compare the first 4
845          * bytes with VMDK4_MAGIC, let's error out if less is read. */
846         error_setg(errp, "File is too small, not a valid image");
847         return NULL;
848     }
849 
850     size = MIN(size, (1 << 20) - 1);  /* avoid unbounded allocation */
851     buf = g_malloc(size + 1);
852 
853     ret = bdrv_pread(file, desc_offset, buf, size);
854     if (ret < 0) {
855         error_setg_errno(errp, -ret, "Could not read from file");
856         g_free(buf);
857         return NULL;
858     }
859     buf[ret] = 0;
860 
861     return buf;
862 }
863 
864 static int vmdk_open_vmdk4(BlockDriverState *bs,
865                            BdrvChild *file,
866                            int flags, QDict *options, Error **errp)
867 {
868     int ret;
869     uint32_t magic;
870     uint32_t l1_size, l1_entry_sectors;
871     VMDK4Header header;
872     VmdkExtent *extent;
873     BDRVVmdkState *s = bs->opaque;
874     int64_t l1_backup_offset = 0;
875     bool compressed;
876 
877     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
878     if (ret < 0) {
879         bdrv_refresh_filename(file->bs);
880         error_setg_errno(errp, -ret,
881                          "Could not read header from file '%s'",
882                          file->bs->filename);
883         return -EINVAL;
884     }
885     if (header.capacity == 0) {
886         uint64_t desc_offset = le64_to_cpu(header.desc_offset);
887         if (desc_offset) {
888             char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
889             if (!buf) {
890                 return -EINVAL;
891             }
892             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
893             g_free(buf);
894             return ret;
895         }
896     }
897 
898     if (!s->create_type) {
899         s->create_type = g_strdup("monolithicSparse");
900     }
901 
902     if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
903         /*
904          * The footer takes precedence over the header, so read it in. The
905          * footer starts at offset -1024 from the end: One sector for the
906          * footer, and another one for the end-of-stream marker.
907          */
908         struct {
909             struct {
910                 uint64_t val;
911                 uint32_t size;
912                 uint32_t type;
913                 uint8_t pad[512 - 16];
914             } QEMU_PACKED footer_marker;
915 
916             uint32_t magic;
917             VMDK4Header header;
918             uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
919 
920             struct {
921                 uint64_t val;
922                 uint32_t size;
923                 uint32_t type;
924                 uint8_t pad[512 - 16];
925             } QEMU_PACKED eos_marker;
926         } QEMU_PACKED footer;
927 
928         ret = bdrv_pread(file,
929             bs->file->bs->total_sectors * 512 - 1536,
930             &footer, sizeof(footer));
931         if (ret < 0) {
932             error_setg_errno(errp, -ret, "Failed to read footer");
933             return ret;
934         }
935 
936         /* Some sanity checks for the footer */
937         if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
938             le32_to_cpu(footer.footer_marker.size) != 0  ||
939             le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
940             le64_to_cpu(footer.eos_marker.val) != 0  ||
941             le32_to_cpu(footer.eos_marker.size) != 0  ||
942             le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
943         {
944             error_setg(errp, "Invalid footer");
945             return -EINVAL;
946         }
947 
948         header = footer.header;
949     }
950 
951     compressed =
952         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
953     if (le32_to_cpu(header.version) > 3) {
954         error_setg(errp, "Unsupported VMDK version %" PRIu32,
955                    le32_to_cpu(header.version));
956         return -ENOTSUP;
957     } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
958                !compressed) {
959         /* VMware KB 2064959 explains that version 3 added support for
960          * persistent changed block tracking (CBT), and backup software can
961          * read it as version=1 if it doesn't care about the changed area
962          * information. So we are safe to enable read only. */
963         error_setg(errp, "VMDK version 3 must be read only");
964         return -EINVAL;
965     }
966 
967     if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
968         error_setg(errp, "L2 table size too big");
969         return -EINVAL;
970     }
971 
972     l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
973                         * le64_to_cpu(header.granularity);
974     if (l1_entry_sectors == 0) {
975         error_setg(errp, "L1 entry size is invalid");
976         return -EINVAL;
977     }
978     l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
979                 / l1_entry_sectors;
980     if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
981         l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
982     }
983     if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
984         error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
985                    (int64_t)(le64_to_cpu(header.grain_offset)
986                              * BDRV_SECTOR_SIZE));
987         return -EINVAL;
988     }
989 
990     ret = vmdk_add_extent(bs, file, false,
991                           le64_to_cpu(header.capacity),
992                           le64_to_cpu(header.gd_offset) << 9,
993                           l1_backup_offset,
994                           l1_size,
995                           le32_to_cpu(header.num_gtes_per_gt),
996                           le64_to_cpu(header.granularity),
997                           &extent,
998                           errp);
999     if (ret < 0) {
1000         return ret;
1001     }
1002     extent->compressed =
1003         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
1004     if (extent->compressed) {
1005         g_free(s->create_type);
1006         s->create_type = g_strdup("streamOptimized");
1007     }
1008     extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
1009     extent->version = le32_to_cpu(header.version);
1010     extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
1011     ret = vmdk_init_tables(bs, extent, errp);
1012     if (ret) {
1013         /* free extent allocated by vmdk_add_extent */
1014         vmdk_free_last_extent(bs);
1015     }
1016     return ret;
1017 }
1018 
1019 /* find an option value out of descriptor file */
1020 static int vmdk_parse_description(const char *desc, const char *opt_name,
1021         char *buf, int buf_size)
1022 {
1023     char *opt_pos, *opt_end;
1024     const char *end = desc + strlen(desc);
1025 
1026     opt_pos = strstr(desc, opt_name);
1027     if (!opt_pos) {
1028         return VMDK_ERROR;
1029     }
1030     /* Skip "=\"" following opt_name */
1031     opt_pos += strlen(opt_name) + 2;
1032     if (opt_pos >= end) {
1033         return VMDK_ERROR;
1034     }
1035     opt_end = opt_pos;
1036     while (opt_end < end && *opt_end != '"') {
1037         opt_end++;
1038     }
1039     if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
1040         return VMDK_ERROR;
1041     }
1042     pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
1043     return VMDK_OK;
1044 }
1045 
1046 /* Open an extent file and append to bs array */
1047 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
1048                             char *buf, QDict *options, Error **errp)
1049 {
1050     uint32_t magic;
1051 
1052     magic = ldl_be_p(buf);
1053     switch (magic) {
1054         case VMDK3_MAGIC:
1055             return vmdk_open_vmfs_sparse(bs, file, flags, errp);
1056             break;
1057         case VMDK4_MAGIC:
1058             return vmdk_open_vmdk4(bs, file, flags, options, errp);
1059             break;
1060         default:
1061             error_setg(errp, "Image not in VMDK format");
1062             return -EINVAL;
1063             break;
1064     }
1065 }
1066 
1067 static const char *next_line(const char *s)
1068 {
1069     while (*s) {
1070         if (*s == '\n') {
1071             return s + 1;
1072         }
1073         s++;
1074     }
1075     return s;
1076 }
1077 
1078 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
1079                               QDict *options, Error **errp)
1080 {
1081     int ret;
1082     int matches;
1083     char access[11];
1084     char type[11];
1085     char fname[512];
1086     const char *p, *np;
1087     int64_t sectors = 0;
1088     int64_t flat_offset;
1089     char *desc_file_dir = NULL;
1090     char *extent_path;
1091     BdrvChild *extent_file;
1092     BdrvChildRole extent_role;
1093     BDRVVmdkState *s = bs->opaque;
1094     VmdkExtent *extent;
1095     char extent_opt_prefix[32];
1096     Error *local_err = NULL;
1097 
1098     for (p = desc; *p; p = next_line(p)) {
1099         /* parse extent line in one of below formats:
1100          *
1101          * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
1102          * RW [size in sectors] SPARSE "file-name.vmdk"
1103          * RW [size in sectors] VMFS "file-name.vmdk"
1104          * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
1105          * RW [size in sectors] SESPARSE "file-name.vmdk"
1106          */
1107         flat_offset = -1;
1108         matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
1109                          access, &sectors, type, fname, &flat_offset);
1110         if (matches < 4 || strcmp(access, "RW")) {
1111             continue;
1112         } else if (!strcmp(type, "FLAT")) {
1113             if (matches != 5 || flat_offset < 0) {
1114                 goto invalid;
1115             }
1116         } else if (!strcmp(type, "VMFS")) {
1117             if (matches == 4) {
1118                 flat_offset = 0;
1119             } else {
1120                 goto invalid;
1121             }
1122         } else if (matches != 4) {
1123             goto invalid;
1124         }
1125 
1126         if (sectors <= 0 ||
1127             (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
1128              strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE") &&
1129              strcmp(type, "SESPARSE")) ||
1130             (strcmp(access, "RW"))) {
1131             continue;
1132         }
1133 
1134         if (path_is_absolute(fname)) {
1135             extent_path = g_strdup(fname);
1136         } else {
1137             if (!desc_file_dir) {
1138                 desc_file_dir = bdrv_dirname(bs->file->bs, errp);
1139                 if (!desc_file_dir) {
1140                     bdrv_refresh_filename(bs->file->bs);
1141                     error_prepend(errp, "Cannot use relative paths with VMDK "
1142                                   "descriptor file '%s': ",
1143                                   bs->file->bs->filename);
1144                     ret = -EINVAL;
1145                     goto out;
1146                 }
1147             }
1148 
1149             extent_path = g_strconcat(desc_file_dir, fname, NULL);
1150         }
1151 
1152         ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
1153         assert(ret < 32);
1154 
1155         extent_role = BDRV_CHILD_DATA;
1156         if (strcmp(type, "FLAT") != 0 && strcmp(type, "VMFS") != 0) {
1157             /* non-flat extents have metadata */
1158             extent_role |= BDRV_CHILD_METADATA;
1159         }
1160 
1161         extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
1162                                       bs, &child_of_bds, extent_role, false,
1163                                       &local_err);
1164         g_free(extent_path);
1165         if (local_err) {
1166             error_propagate(errp, local_err);
1167             ret = -EINVAL;
1168             goto out;
1169         }
1170 
1171         /* save to extents array */
1172         if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
1173             /* FLAT extent */
1174 
1175             ret = vmdk_add_extent(bs, extent_file, true, sectors,
1176                             0, 0, 0, 0, 0, &extent, errp);
1177             if (ret < 0) {
1178                 bdrv_unref_child(bs, extent_file);
1179                 goto out;
1180             }
1181             extent->flat_start_offset = flat_offset << 9;
1182         } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
1183             /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
1184             char *buf = vmdk_read_desc(extent_file, 0, errp);
1185             if (!buf) {
1186                 ret = -EINVAL;
1187             } else {
1188                 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
1189                                        options, errp);
1190             }
1191             g_free(buf);
1192             if (ret) {
1193                 bdrv_unref_child(bs, extent_file);
1194                 goto out;
1195             }
1196             extent = &s->extents[s->num_extents - 1];
1197         } else if (!strcmp(type, "SESPARSE")) {
1198             ret = vmdk_open_se_sparse(bs, extent_file, bs->open_flags, errp);
1199             if (ret) {
1200                 bdrv_unref_child(bs, extent_file);
1201                 goto out;
1202             }
1203             extent = &s->extents[s->num_extents - 1];
1204         } else {
1205             error_setg(errp, "Unsupported extent type '%s'", type);
1206             bdrv_unref_child(bs, extent_file);
1207             ret = -ENOTSUP;
1208             goto out;
1209         }
1210         extent->type = g_strdup(type);
1211     }
1212 
1213     ret = 0;
1214     goto out;
1215 
1216 invalid:
1217     np = next_line(p);
1218     assert(np != p);
1219     if (np[-1] == '\n') {
1220         np--;
1221     }
1222     error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
1223     ret = -EINVAL;
1224 
1225 out:
1226     g_free(desc_file_dir);
1227     return ret;
1228 }
1229 
1230 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
1231                                QDict *options, Error **errp)
1232 {
1233     int ret;
1234     char ct[128];
1235     BDRVVmdkState *s = bs->opaque;
1236 
1237     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
1238         error_setg(errp, "invalid VMDK image descriptor");
1239         ret = -EINVAL;
1240         goto exit;
1241     }
1242     if (strcmp(ct, "monolithicFlat") &&
1243         strcmp(ct, "vmfs") &&
1244         strcmp(ct, "vmfsSparse") &&
1245         strcmp(ct, "seSparse") &&
1246         strcmp(ct, "twoGbMaxExtentSparse") &&
1247         strcmp(ct, "twoGbMaxExtentFlat")) {
1248         error_setg(errp, "Unsupported image type '%s'", ct);
1249         ret = -ENOTSUP;
1250         goto exit;
1251     }
1252     s->create_type = g_strdup(ct);
1253     s->desc_offset = 0;
1254     ret = vmdk_parse_extents(buf, bs, options, errp);
1255 exit:
1256     return ret;
1257 }
1258 
1259 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
1260                      Error **errp)
1261 {
1262     char *buf;
1263     int ret;
1264     BDRVVmdkState *s = bs->opaque;
1265     uint32_t magic;
1266 
1267     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
1268                                BDRV_CHILD_IMAGE, false, errp);
1269     if (!bs->file) {
1270         return -EINVAL;
1271     }
1272 
1273     buf = vmdk_read_desc(bs->file, 0, errp);
1274     if (!buf) {
1275         return -EINVAL;
1276     }
1277 
1278     magic = ldl_be_p(buf);
1279     switch (magic) {
1280         case VMDK3_MAGIC:
1281         case VMDK4_MAGIC:
1282             ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
1283                                    errp);
1284             s->desc_offset = 0x200;
1285             break;
1286         default:
1287             /* No data in the descriptor file */
1288             bs->file->role &= ~BDRV_CHILD_DATA;
1289 
1290             /* Must succeed because we have given up permissions if anything */
1291             bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1292 
1293             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
1294             break;
1295     }
1296     if (ret) {
1297         goto fail;
1298     }
1299 
1300     /* try to open parent images, if exist */
1301     ret = vmdk_parent_open(bs);
1302     if (ret) {
1303         goto fail;
1304     }
1305     ret = vmdk_read_cid(bs, 0, &s->cid);
1306     if (ret) {
1307         goto fail;
1308     }
1309     ret = vmdk_read_cid(bs, 1, &s->parent_cid);
1310     if (ret) {
1311         goto fail;
1312     }
1313     qemu_co_mutex_init(&s->lock);
1314 
1315     /* Disable migration when VMDK images are used */
1316     error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
1317                "does not support live migration",
1318                bdrv_get_device_or_node_name(bs));
1319     ret = migrate_add_blocker(s->migration_blocker, errp);
1320     if (ret < 0) {
1321         error_free(s->migration_blocker);
1322         goto fail;
1323     }
1324 
1325     g_free(buf);
1326     return 0;
1327 
1328 fail:
1329     g_free(buf);
1330     g_free(s->create_type);
1331     s->create_type = NULL;
1332     vmdk_free_extents(bs);
1333     return ret;
1334 }
1335 
1336 
1337 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
1338 {
1339     BDRVVmdkState *s = bs->opaque;
1340     int i;
1341 
1342     for (i = 0; i < s->num_extents; i++) {
1343         if (!s->extents[i].flat) {
1344             bs->bl.pwrite_zeroes_alignment =
1345                 MAX(bs->bl.pwrite_zeroes_alignment,
1346                     s->extents[i].cluster_sectors << BDRV_SECTOR_BITS);
1347         }
1348     }
1349 }
1350 
1351 /**
1352  * get_whole_cluster
1353  *
1354  * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1355  * to the cluster at @cluster_sector_num. If @zeroed is true, we're overwriting
1356  * a zeroed cluster in the current layer and must not copy data from the
1357  * backing file.
1358  *
1359  * If @skip_start_sector < @skip_end_sector, the relative range
1360  * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1361  * it for call to write user data in the request.
1362  */
1363 static int get_whole_cluster(BlockDriverState *bs,
1364                              VmdkExtent *extent,
1365                              uint64_t cluster_offset,
1366                              uint64_t offset,
1367                              uint64_t skip_start_bytes,
1368                              uint64_t skip_end_bytes,
1369                              bool zeroed)
1370 {
1371     int ret = VMDK_OK;
1372     int64_t cluster_bytes;
1373     uint8_t *whole_grain;
1374     bool copy_from_backing;
1375 
1376     /* For COW, align request sector_num to cluster start */
1377     cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1378     offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1379     whole_grain = qemu_blockalign(bs, cluster_bytes);
1380     copy_from_backing = bs->backing && !zeroed;
1381 
1382     if (!copy_from_backing) {
1383         memset(whole_grain, 0, skip_start_bytes);
1384         memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1385     }
1386 
1387     assert(skip_end_bytes <= cluster_bytes);
1388     /* we will be here if it's first write on non-exist grain(cluster).
1389      * try to read from parent image, if exist */
1390     if (bs->backing && !vmdk_is_cid_valid(bs)) {
1391         ret = VMDK_ERROR;
1392         goto exit;
1393     }
1394 
1395     /* Read backing data before skip range */
1396     if (skip_start_bytes > 0) {
1397         if (copy_from_backing) {
1398             /* qcow2 emits this on bs->file instead of bs->backing */
1399             BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1400             ret = bdrv_pread(bs->backing, offset, whole_grain,
1401                              skip_start_bytes);
1402             if (ret < 0) {
1403                 ret = VMDK_ERROR;
1404                 goto exit;
1405             }
1406         }
1407         BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1408         ret = bdrv_pwrite(extent->file, cluster_offset, whole_grain,
1409                           skip_start_bytes);
1410         if (ret < 0) {
1411             ret = VMDK_ERROR;
1412             goto exit;
1413         }
1414     }
1415     /* Read backing data after skip range */
1416     if (skip_end_bytes < cluster_bytes) {
1417         if (copy_from_backing) {
1418             /* qcow2 emits this on bs->file instead of bs->backing */
1419             BLKDBG_EVENT(extent->file, BLKDBG_COW_READ);
1420             ret = bdrv_pread(bs->backing, offset + skip_end_bytes,
1421                              whole_grain + skip_end_bytes,
1422                              cluster_bytes - skip_end_bytes);
1423             if (ret < 0) {
1424                 ret = VMDK_ERROR;
1425                 goto exit;
1426             }
1427         }
1428         BLKDBG_EVENT(extent->file, BLKDBG_COW_WRITE);
1429         ret = bdrv_pwrite(extent->file, cluster_offset + skip_end_bytes,
1430                           whole_grain + skip_end_bytes,
1431                           cluster_bytes - skip_end_bytes);
1432         if (ret < 0) {
1433             ret = VMDK_ERROR;
1434             goto exit;
1435         }
1436     }
1437 
1438     ret = VMDK_OK;
1439 exit:
1440     qemu_vfree(whole_grain);
1441     return ret;
1442 }
1443 
1444 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1445                          uint32_t offset)
1446 {
1447     offset = cpu_to_le32(offset);
1448     /* update L2 table */
1449     BLKDBG_EVENT(extent->file, BLKDBG_L2_UPDATE);
1450     if (bdrv_pwrite(extent->file,
1451                 ((int64_t)m_data->l2_offset * 512)
1452                     + (m_data->l2_index * sizeof(offset)),
1453                 &offset, sizeof(offset)) < 0) {
1454         return VMDK_ERROR;
1455     }
1456     /* update backup L2 table */
1457     if (extent->l1_backup_table_offset != 0) {
1458         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1459         if (bdrv_pwrite(extent->file,
1460                     ((int64_t)m_data->l2_offset * 512)
1461                         + (m_data->l2_index * sizeof(offset)),
1462                     &offset, sizeof(offset)) < 0) {
1463             return VMDK_ERROR;
1464         }
1465     }
1466     if (bdrv_flush(extent->file->bs) < 0) {
1467         return VMDK_ERROR;
1468     }
1469     if (m_data->l2_cache_entry) {
1470         *m_data->l2_cache_entry = offset;
1471     }
1472 
1473     return VMDK_OK;
1474 }
1475 
1476 /**
1477  * get_cluster_offset
1478  *
1479  * Look up cluster offset in extent file by sector number, and store in
1480  * @cluster_offset.
1481  *
1482  * For flat extents, the start offset as parsed from the description file is
1483  * returned.
1484  *
1485  * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1486  * offset for a new cluster and update L2 cache. If there is a backing file,
1487  * COW is done before returning; otherwise, zeroes are written to the allocated
1488  * cluster. Both COW and zero writing skips the sector range
1489  * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1490  * has new data to write there.
1491  *
1492  * Returns: VMDK_OK if cluster exists and mapped in the image.
1493  *          VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1494  *          VMDK_ERROR if failed.
1495  */
1496 static int get_cluster_offset(BlockDriverState *bs,
1497                               VmdkExtent *extent,
1498                               VmdkMetaData *m_data,
1499                               uint64_t offset,
1500                               bool allocate,
1501                               uint64_t *cluster_offset,
1502                               uint64_t skip_start_bytes,
1503                               uint64_t skip_end_bytes)
1504 {
1505     unsigned int l1_index, l2_offset, l2_index;
1506     int min_index, i, j;
1507     uint32_t min_count;
1508     void *l2_table;
1509     bool zeroed = false;
1510     int64_t ret;
1511     int64_t cluster_sector;
1512     unsigned int l2_size_bytes = extent->l2_size * extent->entry_size;
1513 
1514     if (m_data) {
1515         m_data->new_allocation = false;
1516     }
1517     if (extent->flat) {
1518         *cluster_offset = extent->flat_start_offset;
1519         return VMDK_OK;
1520     }
1521 
1522     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1523     l1_index = (offset >> 9) / extent->l1_entry_sectors;
1524     if (l1_index >= extent->l1_size) {
1525         return VMDK_ERROR;
1526     }
1527     if (extent->sesparse) {
1528         uint64_t l2_offset_u64;
1529 
1530         assert(extent->entry_size == sizeof(uint64_t));
1531 
1532         l2_offset_u64 = ((uint64_t *)extent->l1_table)[l1_index];
1533         if (l2_offset_u64 == 0) {
1534             l2_offset = 0;
1535         } else if ((l2_offset_u64 & 0xffffffff00000000) != 0x1000000000000000) {
1536             /*
1537              * Top most nibble is 0x1 if grain table is allocated.
1538              * strict check - top most 4 bytes must be 0x10000000 since max
1539              * supported size is 64TB for disk - so no more than 64TB / 16MB
1540              * grain directories which is smaller than uint32,
1541              * where 16MB is the only supported default grain table coverage.
1542              */
1543             return VMDK_ERROR;
1544         } else {
1545             l2_offset_u64 = l2_offset_u64 & 0x00000000ffffffff;
1546             l2_offset_u64 = extent->sesparse_l2_tables_offset +
1547                 l2_offset_u64 * l2_size_bytes / SECTOR_SIZE;
1548             if (l2_offset_u64 > 0x00000000ffffffff) {
1549                 return VMDK_ERROR;
1550             }
1551             l2_offset = (unsigned int)(l2_offset_u64);
1552         }
1553     } else {
1554         assert(extent->entry_size == sizeof(uint32_t));
1555         l2_offset = ((uint32_t *)extent->l1_table)[l1_index];
1556     }
1557     if (!l2_offset) {
1558         return VMDK_UNALLOC;
1559     }
1560     for (i = 0; i < L2_CACHE_SIZE; i++) {
1561         if (l2_offset == extent->l2_cache_offsets[i]) {
1562             /* increment the hit count */
1563             if (++extent->l2_cache_counts[i] == 0xffffffff) {
1564                 for (j = 0; j < L2_CACHE_SIZE; j++) {
1565                     extent->l2_cache_counts[j] >>= 1;
1566                 }
1567             }
1568             l2_table = (char *)extent->l2_cache + (i * l2_size_bytes);
1569             goto found;
1570         }
1571     }
1572     /* not found: load a new entry in the least used one */
1573     min_index = 0;
1574     min_count = 0xffffffff;
1575     for (i = 0; i < L2_CACHE_SIZE; i++) {
1576         if (extent->l2_cache_counts[i] < min_count) {
1577             min_count = extent->l2_cache_counts[i];
1578             min_index = i;
1579         }
1580     }
1581     l2_table = (char *)extent->l2_cache + (min_index * l2_size_bytes);
1582     BLKDBG_EVENT(extent->file, BLKDBG_L2_LOAD);
1583     if (bdrv_pread(extent->file,
1584                 (int64_t)l2_offset * 512,
1585                 l2_table,
1586                 l2_size_bytes
1587             ) != l2_size_bytes) {
1588         return VMDK_ERROR;
1589     }
1590 
1591     extent->l2_cache_offsets[min_index] = l2_offset;
1592     extent->l2_cache_counts[min_index] = 1;
1593  found:
1594     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1595     if (m_data) {
1596         m_data->l1_index = l1_index;
1597         m_data->l2_index = l2_index;
1598         m_data->l2_offset = l2_offset;
1599         m_data->l2_cache_entry = ((uint32_t *)l2_table) + l2_index;
1600     }
1601 
1602     if (extent->sesparse) {
1603         cluster_sector = le64_to_cpu(((uint64_t *)l2_table)[l2_index]);
1604         switch (cluster_sector & 0xf000000000000000) {
1605         case 0x0000000000000000:
1606             /* unallocated grain */
1607             if (cluster_sector != 0) {
1608                 return VMDK_ERROR;
1609             }
1610             break;
1611         case 0x1000000000000000:
1612             /* scsi-unmapped grain - fallthrough */
1613         case 0x2000000000000000:
1614             /* zero grain */
1615             zeroed = true;
1616             break;
1617         case 0x3000000000000000:
1618             /* allocated grain */
1619             cluster_sector = (((cluster_sector & 0x0fff000000000000) >> 48) |
1620                               ((cluster_sector & 0x0000ffffffffffff) << 12));
1621             cluster_sector = extent->sesparse_clusters_offset +
1622                 cluster_sector * extent->cluster_sectors;
1623             break;
1624         default:
1625             return VMDK_ERROR;
1626         }
1627     } else {
1628         cluster_sector = le32_to_cpu(((uint32_t *)l2_table)[l2_index]);
1629 
1630         if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1631             zeroed = true;
1632         }
1633     }
1634 
1635     if (!cluster_sector || zeroed) {
1636         if (!allocate) {
1637             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1638         }
1639         assert(!extent->sesparse);
1640 
1641         if (extent->next_cluster_sector >= VMDK_EXTENT_MAX_SECTORS) {
1642             return VMDK_ERROR;
1643         }
1644 
1645         cluster_sector = extent->next_cluster_sector;
1646         extent->next_cluster_sector += extent->cluster_sectors;
1647 
1648         /* First of all we write grain itself, to avoid race condition
1649          * that may to corrupt the image.
1650          * This problem may occur because of insufficient space on host disk
1651          * or inappropriate VM shutdown.
1652          */
1653         ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1654                                 offset, skip_start_bytes, skip_end_bytes,
1655                                 zeroed);
1656         if (ret) {
1657             return ret;
1658         }
1659         if (m_data) {
1660             m_data->new_allocation = true;
1661         }
1662     }
1663     *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1664     return VMDK_OK;
1665 }
1666 
1667 static VmdkExtent *find_extent(BDRVVmdkState *s,
1668                                 int64_t sector_num, VmdkExtent *start_hint)
1669 {
1670     VmdkExtent *extent = start_hint;
1671 
1672     if (!extent) {
1673         extent = &s->extents[0];
1674     }
1675     while (extent < &s->extents[s->num_extents]) {
1676         if (sector_num < extent->end_sector) {
1677             return extent;
1678         }
1679         extent++;
1680     }
1681     return NULL;
1682 }
1683 
1684 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1685                                                    int64_t offset)
1686 {
1687     uint64_t extent_begin_offset, extent_relative_offset;
1688     uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1689 
1690     extent_begin_offset =
1691         (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1692     extent_relative_offset = offset - extent_begin_offset;
1693     return extent_relative_offset % cluster_size;
1694 }
1695 
1696 static int coroutine_fn vmdk_co_block_status(BlockDriverState *bs,
1697                                              bool want_zero,
1698                                              int64_t offset, int64_t bytes,
1699                                              int64_t *pnum, int64_t *map,
1700                                              BlockDriverState **file)
1701 {
1702     BDRVVmdkState *s = bs->opaque;
1703     int64_t index_in_cluster, n, ret;
1704     uint64_t cluster_offset;
1705     VmdkExtent *extent;
1706 
1707     extent = find_extent(s, offset >> BDRV_SECTOR_BITS, NULL);
1708     if (!extent) {
1709         return -EIO;
1710     }
1711     qemu_co_mutex_lock(&s->lock);
1712     ret = get_cluster_offset(bs, extent, NULL, offset, false, &cluster_offset,
1713                              0, 0);
1714     qemu_co_mutex_unlock(&s->lock);
1715 
1716     index_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1717     switch (ret) {
1718     case VMDK_ERROR:
1719         ret = -EIO;
1720         break;
1721     case VMDK_UNALLOC:
1722         ret = 0;
1723         break;
1724     case VMDK_ZEROED:
1725         ret = BDRV_BLOCK_ZERO;
1726         break;
1727     case VMDK_OK:
1728         ret = BDRV_BLOCK_DATA;
1729         if (!extent->compressed) {
1730             ret |= BDRV_BLOCK_OFFSET_VALID;
1731             *map = cluster_offset + index_in_cluster;
1732             if (extent->flat) {
1733                 ret |= BDRV_BLOCK_RECURSE;
1734             }
1735         }
1736         *file = extent->file->bs;
1737         break;
1738     }
1739 
1740     n = extent->cluster_sectors * BDRV_SECTOR_SIZE - index_in_cluster;
1741     *pnum = MIN(n, bytes);
1742     return ret;
1743 }
1744 
1745 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1746                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1747                             uint64_t qiov_offset, uint64_t n_bytes,
1748                             uint64_t offset)
1749 {
1750     int ret;
1751     VmdkGrainMarker *data = NULL;
1752     uLongf buf_len;
1753     QEMUIOVector local_qiov;
1754     int64_t write_offset;
1755     int64_t write_end_sector;
1756 
1757     if (extent->compressed) {
1758         void *compressed_data;
1759 
1760         /* Only whole clusters */
1761         if (offset_in_cluster ||
1762             n_bytes > (extent->cluster_sectors * SECTOR_SIZE) ||
1763             (n_bytes < (extent->cluster_sectors * SECTOR_SIZE) &&
1764              offset + n_bytes != extent->end_sector * SECTOR_SIZE))
1765         {
1766             ret = -EINVAL;
1767             goto out;
1768         }
1769 
1770         if (!extent->has_marker) {
1771             ret = -EINVAL;
1772             goto out;
1773         }
1774         buf_len = (extent->cluster_sectors << 9) * 2;
1775         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1776 
1777         compressed_data = g_malloc(n_bytes);
1778         qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1779         ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1780         g_free(compressed_data);
1781 
1782         if (ret != Z_OK || buf_len == 0) {
1783             ret = -EINVAL;
1784             goto out;
1785         }
1786 
1787         data->lba = cpu_to_le64(offset >> BDRV_SECTOR_BITS);
1788         data->size = cpu_to_le32(buf_len);
1789 
1790         n_bytes = buf_len + sizeof(VmdkGrainMarker);
1791         qemu_iovec_init_buf(&local_qiov, data, n_bytes);
1792 
1793         BLKDBG_EVENT(extent->file, BLKDBG_WRITE_COMPRESSED);
1794     } else {
1795         qemu_iovec_init(&local_qiov, qiov->niov);
1796         qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1797 
1798         BLKDBG_EVENT(extent->file, BLKDBG_WRITE_AIO);
1799     }
1800 
1801     write_offset = cluster_offset + offset_in_cluster;
1802     ret = bdrv_co_pwritev(extent->file, write_offset, n_bytes,
1803                           &local_qiov, 0);
1804 
1805     write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1806 
1807     if (extent->compressed) {
1808         extent->next_cluster_sector = write_end_sector;
1809     } else {
1810         extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1811                                           write_end_sector);
1812     }
1813 
1814     if (ret < 0) {
1815         goto out;
1816     }
1817     ret = 0;
1818  out:
1819     g_free(data);
1820     if (!extent->compressed) {
1821         qemu_iovec_destroy(&local_qiov);
1822     }
1823     return ret;
1824 }
1825 
1826 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1827                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1828                             int bytes)
1829 {
1830     int ret;
1831     int cluster_bytes, buf_bytes;
1832     uint8_t *cluster_buf, *compressed_data;
1833     uint8_t *uncomp_buf;
1834     uint32_t data_len;
1835     VmdkGrainMarker *marker;
1836     uLongf buf_len;
1837 
1838 
1839     if (!extent->compressed) {
1840         BLKDBG_EVENT(extent->file, BLKDBG_READ_AIO);
1841         ret = bdrv_co_preadv(extent->file,
1842                              cluster_offset + offset_in_cluster, bytes,
1843                              qiov, 0);
1844         if (ret < 0) {
1845             return ret;
1846         }
1847         return 0;
1848     }
1849     cluster_bytes = extent->cluster_sectors * 512;
1850     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1851     buf_bytes = cluster_bytes * 2;
1852     cluster_buf = g_malloc(buf_bytes);
1853     uncomp_buf = g_malloc(cluster_bytes);
1854     BLKDBG_EVENT(extent->file, BLKDBG_READ_COMPRESSED);
1855     ret = bdrv_pread(extent->file,
1856                 cluster_offset,
1857                 cluster_buf, buf_bytes);
1858     if (ret < 0) {
1859         goto out;
1860     }
1861     compressed_data = cluster_buf;
1862     buf_len = cluster_bytes;
1863     data_len = cluster_bytes;
1864     if (extent->has_marker) {
1865         marker = (VmdkGrainMarker *)cluster_buf;
1866         compressed_data = marker->data;
1867         data_len = le32_to_cpu(marker->size);
1868     }
1869     if (!data_len || data_len > buf_bytes) {
1870         ret = -EINVAL;
1871         goto out;
1872     }
1873     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1874     if (ret != Z_OK) {
1875         ret = -EINVAL;
1876         goto out;
1877 
1878     }
1879     if (offset_in_cluster < 0 ||
1880             offset_in_cluster + bytes > buf_len) {
1881         ret = -EINVAL;
1882         goto out;
1883     }
1884     qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1885     ret = 0;
1886 
1887  out:
1888     g_free(uncomp_buf);
1889     g_free(cluster_buf);
1890     return ret;
1891 }
1892 
1893 static int coroutine_fn
1894 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1895                QEMUIOVector *qiov, int flags)
1896 {
1897     BDRVVmdkState *s = bs->opaque;
1898     int ret;
1899     uint64_t n_bytes, offset_in_cluster;
1900     VmdkExtent *extent = NULL;
1901     QEMUIOVector local_qiov;
1902     uint64_t cluster_offset;
1903     uint64_t bytes_done = 0;
1904 
1905     qemu_iovec_init(&local_qiov, qiov->niov);
1906     qemu_co_mutex_lock(&s->lock);
1907 
1908     while (bytes > 0) {
1909         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1910         if (!extent) {
1911             ret = -EIO;
1912             goto fail;
1913         }
1914         ret = get_cluster_offset(bs, extent, NULL,
1915                                  offset, false, &cluster_offset, 0, 0);
1916         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1917 
1918         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1919                              - offset_in_cluster);
1920 
1921         if (ret != VMDK_OK) {
1922             /* if not allocated, try to read from parent image, if exist */
1923             if (bs->backing && ret != VMDK_ZEROED) {
1924                 if (!vmdk_is_cid_valid(bs)) {
1925                     ret = -EINVAL;
1926                     goto fail;
1927                 }
1928 
1929                 qemu_iovec_reset(&local_qiov);
1930                 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1931 
1932                 /* qcow2 emits this on bs->file instead of bs->backing */
1933                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1934                 ret = bdrv_co_preadv(bs->backing, offset, n_bytes,
1935                                      &local_qiov, 0);
1936                 if (ret < 0) {
1937                     goto fail;
1938                 }
1939             } else {
1940                 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1941             }
1942         } else {
1943             qemu_iovec_reset(&local_qiov);
1944             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1945 
1946             ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1947                                    &local_qiov, n_bytes);
1948             if (ret) {
1949                 goto fail;
1950             }
1951         }
1952         bytes -= n_bytes;
1953         offset += n_bytes;
1954         bytes_done += n_bytes;
1955     }
1956 
1957     ret = 0;
1958 fail:
1959     qemu_co_mutex_unlock(&s->lock);
1960     qemu_iovec_destroy(&local_qiov);
1961 
1962     return ret;
1963 }
1964 
1965 /**
1966  * vmdk_write:
1967  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
1968  *                if possible, otherwise return -ENOTSUP.
1969  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1970  *                with each cluster. By dry run we can find if the zero write
1971  *                is possible without modifying image data.
1972  *
1973  * Returns: error code with 0 for success.
1974  */
1975 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1976                        uint64_t bytes, QEMUIOVector *qiov,
1977                        bool zeroed, bool zero_dry_run)
1978 {
1979     BDRVVmdkState *s = bs->opaque;
1980     VmdkExtent *extent = NULL;
1981     int ret;
1982     int64_t offset_in_cluster, n_bytes;
1983     uint64_t cluster_offset;
1984     uint64_t bytes_done = 0;
1985     VmdkMetaData m_data;
1986 
1987     if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1988         error_report("Wrong offset: offset=0x%" PRIx64
1989                      " total_sectors=0x%" PRIx64,
1990                      offset, bs->total_sectors);
1991         return -EIO;
1992     }
1993 
1994     while (bytes > 0) {
1995         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1996         if (!extent) {
1997             return -EIO;
1998         }
1999         if (extent->sesparse) {
2000             return -ENOTSUP;
2001         }
2002         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
2003         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
2004                              - offset_in_cluster);
2005 
2006         ret = get_cluster_offset(bs, extent, &m_data, offset,
2007                                  !(extent->compressed || zeroed),
2008                                  &cluster_offset, offset_in_cluster,
2009                                  offset_in_cluster + n_bytes);
2010         if (extent->compressed) {
2011             if (ret == VMDK_OK) {
2012                 /* Refuse write to allocated cluster for streamOptimized */
2013                 error_report("Could not write to allocated cluster"
2014                               " for streamOptimized");
2015                 return -EIO;
2016             } else if (!zeroed) {
2017                 /* allocate */
2018                 ret = get_cluster_offset(bs, extent, &m_data, offset,
2019                                          true, &cluster_offset, 0, 0);
2020             }
2021         }
2022         if (ret == VMDK_ERROR) {
2023             return -EINVAL;
2024         }
2025         if (zeroed) {
2026             /* Do zeroed write, buf is ignored */
2027             if (extent->has_zero_grain &&
2028                     offset_in_cluster == 0 &&
2029                     n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
2030                 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
2031                 if (!zero_dry_run && ret != VMDK_ZEROED) {
2032                     /* update L2 tables */
2033                     if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
2034                             != VMDK_OK) {
2035                         return -EIO;
2036                     }
2037                 }
2038             } else {
2039                 return -ENOTSUP;
2040             }
2041         } else {
2042             ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
2043                                     qiov, bytes_done, n_bytes, offset);
2044             if (ret) {
2045                 return ret;
2046             }
2047             if (m_data.new_allocation) {
2048                 /* update L2 tables */
2049                 if (vmdk_L2update(extent, &m_data,
2050                                   cluster_offset >> BDRV_SECTOR_BITS)
2051                         != VMDK_OK) {
2052                     return -EIO;
2053                 }
2054             }
2055         }
2056         bytes -= n_bytes;
2057         offset += n_bytes;
2058         bytes_done += n_bytes;
2059 
2060         /* update CID on the first write every time the virtual disk is
2061          * opened */
2062         if (!s->cid_updated) {
2063             ret = vmdk_write_cid(bs, g_random_int());
2064             if (ret < 0) {
2065                 return ret;
2066             }
2067             s->cid_updated = true;
2068         }
2069     }
2070     return 0;
2071 }
2072 
2073 static int coroutine_fn
2074 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2075                 QEMUIOVector *qiov, int flags)
2076 {
2077     int ret;
2078     BDRVVmdkState *s = bs->opaque;
2079     qemu_co_mutex_lock(&s->lock);
2080     ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
2081     qemu_co_mutex_unlock(&s->lock);
2082     return ret;
2083 }
2084 
2085 static int coroutine_fn
2086 vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
2087                            uint64_t bytes, QEMUIOVector *qiov)
2088 {
2089     if (bytes == 0) {
2090         /* The caller will write bytes 0 to signal EOF.
2091          * When receive it, we align EOF to a sector boundary. */
2092         BDRVVmdkState *s = bs->opaque;
2093         int i, ret;
2094         int64_t length;
2095 
2096         for (i = 0; i < s->num_extents; i++) {
2097             length = bdrv_getlength(s->extents[i].file->bs);
2098             if (length < 0) {
2099                 return length;
2100             }
2101             length = QEMU_ALIGN_UP(length, BDRV_SECTOR_SIZE);
2102             ret = bdrv_truncate(s->extents[i].file, length, false,
2103                                 PREALLOC_MODE_OFF, 0, NULL);
2104             if (ret < 0) {
2105                 return ret;
2106             }
2107         }
2108         return 0;
2109     }
2110     return vmdk_co_pwritev(bs, offset, bytes, qiov, 0);
2111 }
2112 
2113 static int coroutine_fn vmdk_co_pwrite_zeroes(BlockDriverState *bs,
2114                                               int64_t offset,
2115                                               int bytes,
2116                                               BdrvRequestFlags flags)
2117 {
2118     int ret;
2119     BDRVVmdkState *s = bs->opaque;
2120 
2121     qemu_co_mutex_lock(&s->lock);
2122     /* write zeroes could fail if sectors not aligned to cluster, test it with
2123      * dry_run == true before really updating image */
2124     ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
2125     if (!ret) {
2126         ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
2127     }
2128     qemu_co_mutex_unlock(&s->lock);
2129     return ret;
2130 }
2131 
2132 static int vmdk_init_extent(BlockBackend *blk,
2133                             int64_t filesize, bool flat,
2134                             bool compress, bool zeroed_grain,
2135                             Error **errp)
2136 {
2137     int ret, i;
2138     VMDK4Header header;
2139     uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
2140     uint32_t *gd_buf = NULL;
2141     int gd_buf_size;
2142 
2143     if (flat) {
2144         ret = blk_truncate(blk, filesize, false, PREALLOC_MODE_OFF, 0, errp);
2145         goto exit;
2146     }
2147     magic = cpu_to_be32(VMDK4_MAGIC);
2148     memset(&header, 0, sizeof(header));
2149     if (compress) {
2150         header.version = 3;
2151     } else if (zeroed_grain) {
2152         header.version = 2;
2153     } else {
2154         header.version = 1;
2155     }
2156     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
2157                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
2158                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
2159     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
2160     header.capacity = filesize / BDRV_SECTOR_SIZE;
2161     header.granularity = 128;
2162     header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
2163 
2164     grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
2165     gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
2166                            BDRV_SECTOR_SIZE);
2167     gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
2168     gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
2169 
2170     header.desc_offset = 1;
2171     header.desc_size = 20;
2172     header.rgd_offset = header.desc_offset + header.desc_size;
2173     header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
2174     header.grain_offset =
2175         ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
2176                  header.granularity);
2177     /* swap endianness for all header fields */
2178     header.version = cpu_to_le32(header.version);
2179     header.flags = cpu_to_le32(header.flags);
2180     header.capacity = cpu_to_le64(header.capacity);
2181     header.granularity = cpu_to_le64(header.granularity);
2182     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
2183     header.desc_offset = cpu_to_le64(header.desc_offset);
2184     header.desc_size = cpu_to_le64(header.desc_size);
2185     header.rgd_offset = cpu_to_le64(header.rgd_offset);
2186     header.gd_offset = cpu_to_le64(header.gd_offset);
2187     header.grain_offset = cpu_to_le64(header.grain_offset);
2188     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
2189 
2190     header.check_bytes[0] = 0xa;
2191     header.check_bytes[1] = 0x20;
2192     header.check_bytes[2] = 0xd;
2193     header.check_bytes[3] = 0xa;
2194 
2195     /* write all the data */
2196     ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
2197     if (ret < 0) {
2198         error_setg(errp, QERR_IO_ERROR);
2199         goto exit;
2200     }
2201     ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
2202     if (ret < 0) {
2203         error_setg(errp, QERR_IO_ERROR);
2204         goto exit;
2205     }
2206 
2207     ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9, false,
2208                        PREALLOC_MODE_OFF, 0, errp);
2209     if (ret < 0) {
2210         goto exit;
2211     }
2212 
2213     /* write grain directory */
2214     gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
2215     gd_buf = g_malloc0(gd_buf_size);
2216     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
2217          i < gt_count; i++, tmp += gt_size) {
2218         gd_buf[i] = cpu_to_le32(tmp);
2219     }
2220     ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
2221                      gd_buf, gd_buf_size, 0);
2222     if (ret < 0) {
2223         error_setg(errp, QERR_IO_ERROR);
2224         goto exit;
2225     }
2226 
2227     /* write backup grain directory */
2228     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
2229          i < gt_count; i++, tmp += gt_size) {
2230         gd_buf[i] = cpu_to_le32(tmp);
2231     }
2232     ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
2233                      gd_buf, gd_buf_size, 0);
2234     if (ret < 0) {
2235         error_setg(errp, QERR_IO_ERROR);
2236     }
2237 
2238     ret = 0;
2239 exit:
2240     g_free(gd_buf);
2241     return ret;
2242 }
2243 
2244 static int vmdk_create_extent(const char *filename, int64_t filesize,
2245                               bool flat, bool compress, bool zeroed_grain,
2246                               BlockBackend **pbb,
2247                               QemuOpts *opts, Error **errp)
2248 {
2249     int ret;
2250     BlockBackend *blk = NULL;
2251 
2252     ret = bdrv_create_file(filename, opts, errp);
2253     if (ret < 0) {
2254         goto exit;
2255     }
2256 
2257     blk = blk_new_open(filename, NULL, NULL,
2258                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2259                        errp);
2260     if (blk == NULL) {
2261         ret = -EIO;
2262         goto exit;
2263     }
2264 
2265     blk_set_allow_write_beyond_eof(blk, true);
2266 
2267     ret = vmdk_init_extent(blk, filesize, flat, compress, zeroed_grain, errp);
2268 exit:
2269     if (blk) {
2270         if (pbb) {
2271             *pbb = blk;
2272         } else {
2273             blk_unref(blk);
2274             blk = NULL;
2275         }
2276     }
2277     return ret;
2278 }
2279 
2280 static int filename_decompose(const char *filename, char *path, char *prefix,
2281                               char *postfix, size_t buf_len, Error **errp)
2282 {
2283     const char *p, *q;
2284 
2285     if (filename == NULL || !strlen(filename)) {
2286         error_setg(errp, "No filename provided");
2287         return VMDK_ERROR;
2288     }
2289     p = strrchr(filename, '/');
2290     if (p == NULL) {
2291         p = strrchr(filename, '\\');
2292     }
2293     if (p == NULL) {
2294         p = strrchr(filename, ':');
2295     }
2296     if (p != NULL) {
2297         p++;
2298         if (p - filename >= buf_len) {
2299             return VMDK_ERROR;
2300         }
2301         pstrcpy(path, p - filename + 1, filename);
2302     } else {
2303         p = filename;
2304         path[0] = '\0';
2305     }
2306     q = strrchr(p, '.');
2307     if (q == NULL) {
2308         pstrcpy(prefix, buf_len, p);
2309         postfix[0] = '\0';
2310     } else {
2311         if (q - p >= buf_len) {
2312             return VMDK_ERROR;
2313         }
2314         pstrcpy(prefix, q - p + 1, p);
2315         pstrcpy(postfix, buf_len, q);
2316     }
2317     return VMDK_OK;
2318 }
2319 
2320 /*
2321  * idx == 0: get or create the descriptor file (also the image file if in a
2322  *           non-split format.
2323  * idx >= 1: get the n-th extent if in a split subformat
2324  */
2325 typedef BlockBackend *(*vmdk_create_extent_fn)(int64_t size,
2326                                                int idx,
2327                                                bool flat,
2328                                                bool split,
2329                                                bool compress,
2330                                                bool zeroed_grain,
2331                                                void *opaque,
2332                                                Error **errp);
2333 
2334 static void vmdk_desc_add_extent(GString *desc,
2335                                  const char *extent_line_fmt,
2336                                  int64_t size, const char *filename)
2337 {
2338     char *basename = g_path_get_basename(filename);
2339 
2340     g_string_append_printf(desc, extent_line_fmt,
2341                            DIV_ROUND_UP(size, BDRV_SECTOR_SIZE), basename);
2342     g_free(basename);
2343 }
2344 
2345 static int coroutine_fn vmdk_co_do_create(int64_t size,
2346                                           BlockdevVmdkSubformat subformat,
2347                                           BlockdevVmdkAdapterType adapter_type,
2348                                           const char *backing_file,
2349                                           const char *hw_version,
2350                                           bool compat6,
2351                                           bool zeroed_grain,
2352                                           vmdk_create_extent_fn extent_fn,
2353                                           void *opaque,
2354                                           Error **errp)
2355 {
2356     int extent_idx;
2357     BlockBackend *blk = NULL;
2358     BlockBackend *extent_blk;
2359     Error *local_err = NULL;
2360     char *desc = NULL;
2361     int ret = 0;
2362     bool flat, split, compress;
2363     GString *ext_desc_lines;
2364     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
2365     int64_t extent_size;
2366     int64_t created_size = 0;
2367     const char *extent_line_fmt;
2368     char *parent_desc_line = g_malloc0(BUF_SIZE);
2369     uint32_t parent_cid = 0xffffffff;
2370     uint32_t number_heads = 16;
2371     uint32_t desc_offset = 0, desc_len;
2372     const char desc_template[] =
2373         "# Disk DescriptorFile\n"
2374         "version=1\n"
2375         "CID=%" PRIx32 "\n"
2376         "parentCID=%" PRIx32 "\n"
2377         "createType=\"%s\"\n"
2378         "%s"
2379         "\n"
2380         "# Extent description\n"
2381         "%s"
2382         "\n"
2383         "# The Disk Data Base\n"
2384         "#DDB\n"
2385         "\n"
2386         "ddb.virtualHWVersion = \"%s\"\n"
2387         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
2388         "ddb.geometry.heads = \"%" PRIu32 "\"\n"
2389         "ddb.geometry.sectors = \"63\"\n"
2390         "ddb.adapterType = \"%s\"\n";
2391 
2392     ext_desc_lines = g_string_new(NULL);
2393 
2394     /* Read out options */
2395     if (compat6) {
2396         if (hw_version) {
2397             error_setg(errp,
2398                        "compat6 cannot be enabled with hwversion set");
2399             ret = -EINVAL;
2400             goto exit;
2401         }
2402         hw_version = "6";
2403     }
2404     if (!hw_version) {
2405         hw_version = "4";
2406     }
2407 
2408     if (adapter_type != BLOCKDEV_VMDK_ADAPTER_TYPE_IDE) {
2409         /* that's the number of heads with which vmware operates when
2410            creating, exporting, etc. vmdk files with a non-ide adapter type */
2411         number_heads = 255;
2412     }
2413     split = (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT) ||
2414             (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTSPARSE);
2415     flat = (subformat == BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICFLAT) ||
2416            (subformat == BLOCKDEV_VMDK_SUBFORMAT_TWOGBMAXEXTENTFLAT);
2417     compress = subformat == BLOCKDEV_VMDK_SUBFORMAT_STREAMOPTIMIZED;
2418 
2419     if (flat) {
2420         extent_line_fmt = "RW %" PRId64 " FLAT \"%s\" 0\n";
2421     } else {
2422         extent_line_fmt = "RW %" PRId64 " SPARSE \"%s\"\n";
2423     }
2424     if (flat && backing_file) {
2425         error_setg(errp, "Flat image can't have backing file");
2426         ret = -ENOTSUP;
2427         goto exit;
2428     }
2429     if (flat && zeroed_grain) {
2430         error_setg(errp, "Flat image can't enable zeroed grain");
2431         ret = -ENOTSUP;
2432         goto exit;
2433     }
2434 
2435     /* Create extents */
2436     if (split) {
2437         extent_size = split_size;
2438     } else {
2439         extent_size = size;
2440     }
2441     if (!split && !flat) {
2442         created_size = extent_size;
2443     } else {
2444         created_size = 0;
2445     }
2446     /* Get the descriptor file BDS */
2447     blk = extent_fn(created_size, 0, flat, split, compress, zeroed_grain,
2448                     opaque, errp);
2449     if (!blk) {
2450         ret = -EIO;
2451         goto exit;
2452     }
2453     if (!split && !flat) {
2454         vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, created_size,
2455                              blk_bs(blk)->filename);
2456     }
2457 
2458     if (backing_file) {
2459         BlockBackend *backing;
2460         char *full_backing =
2461             bdrv_get_full_backing_filename_from_filename(blk_bs(blk)->filename,
2462                                                          backing_file,
2463                                                          &local_err);
2464         if (local_err) {
2465             error_propagate(errp, local_err);
2466             ret = -ENOENT;
2467             goto exit;
2468         }
2469         assert(full_backing);
2470 
2471         backing = blk_new_open(full_backing, NULL, NULL,
2472                                BDRV_O_NO_BACKING, errp);
2473         g_free(full_backing);
2474         if (backing == NULL) {
2475             ret = -EIO;
2476             goto exit;
2477         }
2478         if (strcmp(blk_bs(backing)->drv->format_name, "vmdk")) {
2479             error_setg(errp, "Invalid backing file format: %s. Must be vmdk",
2480                        blk_bs(backing)->drv->format_name);
2481             blk_unref(backing);
2482             ret = -EINVAL;
2483             goto exit;
2484         }
2485         ret = vmdk_read_cid(blk_bs(backing), 0, &parent_cid);
2486         blk_unref(backing);
2487         if (ret) {
2488             error_setg(errp, "Failed to read parent CID");
2489             goto exit;
2490         }
2491         snprintf(parent_desc_line, BUF_SIZE,
2492                 "parentFileNameHint=\"%s\"", backing_file);
2493     }
2494     extent_idx = 1;
2495     while (created_size < size) {
2496         int64_t cur_size = MIN(size - created_size, extent_size);
2497         extent_blk = extent_fn(cur_size, extent_idx, flat, split, compress,
2498                                zeroed_grain, opaque, errp);
2499         if (!extent_blk) {
2500             ret = -EINVAL;
2501             goto exit;
2502         }
2503         vmdk_desc_add_extent(ext_desc_lines, extent_line_fmt, cur_size,
2504                              blk_bs(extent_blk)->filename);
2505         created_size += cur_size;
2506         extent_idx++;
2507         blk_unref(extent_blk);
2508     }
2509 
2510     /* Check whether we got excess extents */
2511     extent_blk = extent_fn(-1, extent_idx, flat, split, compress, zeroed_grain,
2512                            opaque, NULL);
2513     if (extent_blk) {
2514         blk_unref(extent_blk);
2515         error_setg(errp, "List of extents contains unused extents");
2516         ret = -EINVAL;
2517         goto exit;
2518     }
2519 
2520     /* generate descriptor file */
2521     desc = g_strdup_printf(desc_template,
2522                            g_random_int(),
2523                            parent_cid,
2524                            BlockdevVmdkSubformat_str(subformat),
2525                            parent_desc_line,
2526                            ext_desc_lines->str,
2527                            hw_version,
2528                            size /
2529                                (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2530                            number_heads,
2531                            BlockdevVmdkAdapterType_str(adapter_type));
2532     desc_len = strlen(desc);
2533     /* the descriptor offset = 0x200 */
2534     if (!split && !flat) {
2535         desc_offset = 0x200;
2536     }
2537 
2538     ret = blk_pwrite(blk, desc_offset, desc, desc_len, 0);
2539     if (ret < 0) {
2540         error_setg_errno(errp, -ret, "Could not write description");
2541         goto exit;
2542     }
2543     /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2544      * for description file */
2545     if (desc_offset == 0) {
2546         ret = blk_truncate(blk, desc_len, false, PREALLOC_MODE_OFF, 0, errp);
2547         if (ret < 0) {
2548             goto exit;
2549         }
2550     }
2551     ret = 0;
2552 exit:
2553     if (blk) {
2554         blk_unref(blk);
2555     }
2556     g_free(desc);
2557     g_free(parent_desc_line);
2558     g_string_free(ext_desc_lines, true);
2559     return ret;
2560 }
2561 
2562 typedef struct {
2563     char *path;
2564     char *prefix;
2565     char *postfix;
2566     QemuOpts *opts;
2567 } VMDKCreateOptsData;
2568 
2569 static BlockBackend *vmdk_co_create_opts_cb(int64_t size, int idx,
2570                                             bool flat, bool split, bool compress,
2571                                             bool zeroed_grain, void *opaque,
2572                                             Error **errp)
2573 {
2574     BlockBackend *blk = NULL;
2575     BlockDriverState *bs = NULL;
2576     VMDKCreateOptsData *data = opaque;
2577     char *ext_filename = NULL;
2578     char *rel_filename = NULL;
2579 
2580     /* We're done, don't create excess extents. */
2581     if (size == -1) {
2582         assert(errp == NULL);
2583         return NULL;
2584     }
2585 
2586     if (idx == 0) {
2587         rel_filename = g_strdup_printf("%s%s", data->prefix, data->postfix);
2588     } else if (split) {
2589         rel_filename = g_strdup_printf("%s-%c%03d%s",
2590                                        data->prefix,
2591                                        flat ? 'f' : 's', idx, data->postfix);
2592     } else {
2593         assert(idx == 1);
2594         rel_filename = g_strdup_printf("%s-flat%s", data->prefix, data->postfix);
2595     }
2596 
2597     ext_filename = g_strdup_printf("%s%s", data->path, rel_filename);
2598     g_free(rel_filename);
2599 
2600     if (vmdk_create_extent(ext_filename, size,
2601                            flat, compress, zeroed_grain, &blk, data->opts,
2602                            errp)) {
2603         goto exit;
2604     }
2605     bdrv_unref(bs);
2606 exit:
2607     g_free(ext_filename);
2608     return blk;
2609 }
2610 
2611 static int coroutine_fn vmdk_co_create_opts(BlockDriver *drv,
2612                                             const char *filename,
2613                                             QemuOpts *opts,
2614                                             Error **errp)
2615 {
2616     Error *local_err = NULL;
2617     char *desc = NULL;
2618     int64_t total_size = 0;
2619     char *adapter_type = NULL;
2620     BlockdevVmdkAdapterType adapter_type_enum;
2621     char *backing_file = NULL;
2622     char *hw_version = NULL;
2623     char *fmt = NULL;
2624     BlockdevVmdkSubformat subformat;
2625     int ret = 0;
2626     char *path = g_malloc0(PATH_MAX);
2627     char *prefix = g_malloc0(PATH_MAX);
2628     char *postfix = g_malloc0(PATH_MAX);
2629     char *desc_line = g_malloc0(BUF_SIZE);
2630     char *ext_filename = g_malloc0(PATH_MAX);
2631     char *desc_filename = g_malloc0(PATH_MAX);
2632     char *parent_desc_line = g_malloc0(BUF_SIZE);
2633     bool zeroed_grain;
2634     bool compat6;
2635     VMDKCreateOptsData data;
2636 
2637     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
2638         ret = -EINVAL;
2639         goto exit;
2640     }
2641     /* Read out options */
2642     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2643                           BDRV_SECTOR_SIZE);
2644     adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
2645     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2646     hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
2647     compat6 = qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false);
2648     if (strcmp(hw_version, "undefined") == 0) {
2649         g_free(hw_version);
2650         hw_version = NULL;
2651     }
2652     fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
2653     zeroed_grain = qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false);
2654 
2655     if (adapter_type) {
2656         adapter_type_enum = qapi_enum_parse(&BlockdevVmdkAdapterType_lookup,
2657                                             adapter_type,
2658                                             BLOCKDEV_VMDK_ADAPTER_TYPE_IDE,
2659                                             &local_err);
2660         if (local_err) {
2661             error_propagate(errp, local_err);
2662             ret = -EINVAL;
2663             goto exit;
2664         }
2665     } else {
2666         adapter_type_enum = BLOCKDEV_VMDK_ADAPTER_TYPE_IDE;
2667     }
2668 
2669     if (!fmt) {
2670         /* Default format to monolithicSparse */
2671         subformat = BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE;
2672     } else {
2673         subformat = qapi_enum_parse(&BlockdevVmdkSubformat_lookup,
2674                                     fmt,
2675                                     BLOCKDEV_VMDK_SUBFORMAT_MONOLITHICSPARSE,
2676                                     &local_err);
2677         if (local_err) {
2678             error_propagate(errp, local_err);
2679             ret = -EINVAL;
2680             goto exit;
2681         }
2682     }
2683     data = (VMDKCreateOptsData){
2684         .prefix = prefix,
2685         .postfix = postfix,
2686         .path = path,
2687         .opts = opts,
2688     };
2689     ret = vmdk_co_do_create(total_size, subformat, adapter_type_enum,
2690                             backing_file, hw_version, compat6, zeroed_grain,
2691                             vmdk_co_create_opts_cb, &data, errp);
2692 
2693 exit:
2694     g_free(adapter_type);
2695     g_free(backing_file);
2696     g_free(hw_version);
2697     g_free(fmt);
2698     g_free(desc);
2699     g_free(path);
2700     g_free(prefix);
2701     g_free(postfix);
2702     g_free(desc_line);
2703     g_free(ext_filename);
2704     g_free(desc_filename);
2705     g_free(parent_desc_line);
2706     return ret;
2707 }
2708 
2709 static BlockBackend *vmdk_co_create_cb(int64_t size, int idx,
2710                                        bool flat, bool split, bool compress,
2711                                        bool zeroed_grain, void *opaque,
2712                                        Error **errp)
2713 {
2714     int ret;
2715     BlockDriverState *bs;
2716     BlockBackend *blk;
2717     BlockdevCreateOptionsVmdk *opts = opaque;
2718 
2719     if (idx == 0) {
2720         bs = bdrv_open_blockdev_ref(opts->file, errp);
2721     } else {
2722         int i;
2723         BlockdevRefList *list = opts->extents;
2724         for (i = 1; i < idx; i++) {
2725             if (!list || !list->next) {
2726                 error_setg(errp, "Extent [%d] not specified", i);
2727                 return NULL;
2728             }
2729             list = list->next;
2730         }
2731         if (!list) {
2732             error_setg(errp, "Extent [%d] not specified", idx - 1);
2733             return NULL;
2734         }
2735         bs = bdrv_open_blockdev_ref(list->value, errp);
2736     }
2737     if (!bs) {
2738         return NULL;
2739     }
2740     blk = blk_new_with_bs(bs,
2741                           BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | BLK_PERM_RESIZE,
2742                           BLK_PERM_ALL, errp);
2743     if (!blk) {
2744         return NULL;
2745     }
2746     blk_set_allow_write_beyond_eof(blk, true);
2747     bdrv_unref(bs);
2748 
2749     if (size != -1) {
2750         ret = vmdk_init_extent(blk, size, flat, compress, zeroed_grain, errp);
2751         if (ret) {
2752             blk_unref(blk);
2753             blk = NULL;
2754         }
2755     }
2756     return blk;
2757 }
2758 
2759 static int coroutine_fn vmdk_co_create(BlockdevCreateOptions *create_options,
2760                                        Error **errp)
2761 {
2762     int ret;
2763     BlockdevCreateOptionsVmdk *opts;
2764 
2765     opts = &create_options->u.vmdk;
2766 
2767     /* Validate options */
2768     if (!QEMU_IS_ALIGNED(opts->size, BDRV_SECTOR_SIZE)) {
2769         error_setg(errp, "Image size must be a multiple of 512 bytes");
2770         ret = -EINVAL;
2771         goto out;
2772     }
2773 
2774     ret = vmdk_co_do_create(opts->size,
2775                             opts->subformat,
2776                             opts->adapter_type,
2777                             opts->backing_file,
2778                             opts->hwversion,
2779                             false,
2780                             opts->zeroed_grain,
2781                             vmdk_co_create_cb,
2782                             opts, errp);
2783     return ret;
2784 
2785 out:
2786     return ret;
2787 }
2788 
2789 static void vmdk_close(BlockDriverState *bs)
2790 {
2791     BDRVVmdkState *s = bs->opaque;
2792 
2793     vmdk_free_extents(bs);
2794     g_free(s->create_type);
2795 
2796     migrate_del_blocker(s->migration_blocker);
2797     error_free(s->migration_blocker);
2798 }
2799 
2800 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2801 {
2802     BDRVVmdkState *s = bs->opaque;
2803     int i, err;
2804     int ret = 0;
2805 
2806     for (i = 0; i < s->num_extents; i++) {
2807         err = bdrv_co_flush(s->extents[i].file->bs);
2808         if (err < 0) {
2809             ret = err;
2810         }
2811     }
2812     return ret;
2813 }
2814 
2815 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2816 {
2817     int i;
2818     int64_t ret = 0;
2819     int64_t r;
2820     BDRVVmdkState *s = bs->opaque;
2821 
2822     ret = bdrv_get_allocated_file_size(bs->file->bs);
2823     if (ret < 0) {
2824         return ret;
2825     }
2826     for (i = 0; i < s->num_extents; i++) {
2827         if (s->extents[i].file == bs->file) {
2828             continue;
2829         }
2830         r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2831         if (r < 0) {
2832             return r;
2833         }
2834         ret += r;
2835     }
2836     return ret;
2837 }
2838 
2839 static int vmdk_has_zero_init(BlockDriverState *bs)
2840 {
2841     int i;
2842     BDRVVmdkState *s = bs->opaque;
2843 
2844     /* If has a flat extent and its underlying storage doesn't have zero init,
2845      * return 0. */
2846     for (i = 0; i < s->num_extents; i++) {
2847         if (s->extents[i].flat) {
2848             if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2849                 return 0;
2850             }
2851         }
2852     }
2853     return 1;
2854 }
2855 
2856 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2857 {
2858     ImageInfo *info = g_new0(ImageInfo, 1);
2859 
2860     bdrv_refresh_filename(extent->file->bs);
2861     *info = (ImageInfo){
2862         .filename         = g_strdup(extent->file->bs->filename),
2863         .format           = g_strdup(extent->type),
2864         .virtual_size     = extent->sectors * BDRV_SECTOR_SIZE,
2865         .compressed       = extent->compressed,
2866         .has_compressed   = extent->compressed,
2867         .cluster_size     = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2868         .has_cluster_size = !extent->flat,
2869     };
2870 
2871     return info;
2872 }
2873 
2874 static int coroutine_fn vmdk_co_check(BlockDriverState *bs,
2875                                       BdrvCheckResult *result,
2876                                       BdrvCheckMode fix)
2877 {
2878     BDRVVmdkState *s = bs->opaque;
2879     VmdkExtent *extent = NULL;
2880     int64_t sector_num = 0;
2881     int64_t total_sectors = bdrv_nb_sectors(bs);
2882     int ret;
2883     uint64_t cluster_offset;
2884 
2885     if (fix) {
2886         return -ENOTSUP;
2887     }
2888 
2889     for (;;) {
2890         if (sector_num >= total_sectors) {
2891             return 0;
2892         }
2893         extent = find_extent(s, sector_num, extent);
2894         if (!extent) {
2895             fprintf(stderr,
2896                     "ERROR: could not find extent for sector %" PRId64 "\n",
2897                     sector_num);
2898             ret = -EINVAL;
2899             break;
2900         }
2901         ret = get_cluster_offset(bs, extent, NULL,
2902                                  sector_num << BDRV_SECTOR_BITS,
2903                                  false, &cluster_offset, 0, 0);
2904         if (ret == VMDK_ERROR) {
2905             fprintf(stderr,
2906                     "ERROR: could not get cluster_offset for sector %"
2907                     PRId64 "\n", sector_num);
2908             break;
2909         }
2910         if (ret == VMDK_OK) {
2911             int64_t extent_len = bdrv_getlength(extent->file->bs);
2912             if (extent_len < 0) {
2913                 fprintf(stderr,
2914                         "ERROR: could not get extent file length for sector %"
2915                         PRId64 "\n", sector_num);
2916                 ret = extent_len;
2917                 break;
2918             }
2919             if (cluster_offset >= extent_len) {
2920                 fprintf(stderr,
2921                         "ERROR: cluster offset for sector %"
2922                         PRId64 " points after EOF\n", sector_num);
2923                 ret = -EINVAL;
2924                 break;
2925             }
2926         }
2927         sector_num += extent->cluster_sectors;
2928     }
2929 
2930     result->corruptions++;
2931     return ret;
2932 }
2933 
2934 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs,
2935                                                  Error **errp)
2936 {
2937     int i;
2938     BDRVVmdkState *s = bs->opaque;
2939     ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2940     ImageInfoList **next;
2941 
2942     *spec_info = (ImageInfoSpecific){
2943         .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2944         .u = {
2945             .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2946         },
2947     };
2948 
2949     *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2950         .create_type = g_strdup(s->create_type),
2951         .cid = s->cid,
2952         .parent_cid = s->parent_cid,
2953     };
2954 
2955     next = &spec_info->u.vmdk.data->extents;
2956     for (i = 0; i < s->num_extents; i++) {
2957         *next = g_new0(ImageInfoList, 1);
2958         (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2959         (*next)->next = NULL;
2960         next = &(*next)->next;
2961     }
2962 
2963     return spec_info;
2964 }
2965 
2966 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2967 {
2968     return a->flat == b->flat &&
2969            a->compressed == b->compressed &&
2970            (a->flat || a->cluster_sectors == b->cluster_sectors);
2971 }
2972 
2973 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2974 {
2975     int i;
2976     BDRVVmdkState *s = bs->opaque;
2977     assert(s->num_extents);
2978 
2979     /* See if we have multiple extents but they have different cases */
2980     for (i = 1; i < s->num_extents; i++) {
2981         if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2982             return -ENOTSUP;
2983         }
2984     }
2985     bdi->needs_compressed_writes = s->extents[0].compressed;
2986     if (!s->extents[0].flat) {
2987         bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2988     }
2989     return 0;
2990 }
2991 
2992 static void vmdk_gather_child_options(BlockDriverState *bs, QDict *target,
2993                                       bool backing_overridden)
2994 {
2995     /* No children but file and backing can be explicitly specified (TODO) */
2996     qdict_put(target, "file",
2997               qobject_ref(bs->file->bs->full_open_options));
2998 
2999     if (backing_overridden) {
3000         if (bs->backing) {
3001             qdict_put(target, "backing",
3002                       qobject_ref(bs->backing->bs->full_open_options));
3003         } else {
3004             qdict_put_null(target, "backing");
3005         }
3006     }
3007 }
3008 
3009 static QemuOptsList vmdk_create_opts = {
3010     .name = "vmdk-create-opts",
3011     .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
3012     .desc = {
3013         {
3014             .name = BLOCK_OPT_SIZE,
3015             .type = QEMU_OPT_SIZE,
3016             .help = "Virtual disk size"
3017         },
3018         {
3019             .name = BLOCK_OPT_ADAPTER_TYPE,
3020             .type = QEMU_OPT_STRING,
3021             .help = "Virtual adapter type, can be one of "
3022                     "ide (default), lsilogic, buslogic or legacyESX"
3023         },
3024         {
3025             .name = BLOCK_OPT_BACKING_FILE,
3026             .type = QEMU_OPT_STRING,
3027             .help = "File name of a base image"
3028         },
3029         {
3030             .name = BLOCK_OPT_COMPAT6,
3031             .type = QEMU_OPT_BOOL,
3032             .help = "VMDK version 6 image",
3033             .def_value_str = "off"
3034         },
3035         {
3036             .name = BLOCK_OPT_HWVERSION,
3037             .type = QEMU_OPT_STRING,
3038             .help = "VMDK hardware version",
3039             .def_value_str = "undefined"
3040         },
3041         {
3042             .name = BLOCK_OPT_SUBFMT,
3043             .type = QEMU_OPT_STRING,
3044             .help =
3045                 "VMDK flat extent format, can be one of "
3046                 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
3047         },
3048         {
3049             .name = BLOCK_OPT_ZEROED_GRAIN,
3050             .type = QEMU_OPT_BOOL,
3051             .help = "Enable efficient zero writes "
3052                     "using the zeroed-grain GTE feature"
3053         },
3054         { /* end of list */ }
3055     }
3056 };
3057 
3058 static BlockDriver bdrv_vmdk = {
3059     .format_name                  = "vmdk",
3060     .instance_size                = sizeof(BDRVVmdkState),
3061     .bdrv_probe                   = vmdk_probe,
3062     .bdrv_open                    = vmdk_open,
3063     .bdrv_co_check                = vmdk_co_check,
3064     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
3065     .bdrv_child_perm              = bdrv_default_perms,
3066     .bdrv_co_preadv               = vmdk_co_preadv,
3067     .bdrv_co_pwritev              = vmdk_co_pwritev,
3068     .bdrv_co_pwritev_compressed   = vmdk_co_pwritev_compressed,
3069     .bdrv_co_pwrite_zeroes        = vmdk_co_pwrite_zeroes,
3070     .bdrv_close                   = vmdk_close,
3071     .bdrv_co_create_opts          = vmdk_co_create_opts,
3072     .bdrv_co_create               = vmdk_co_create,
3073     .bdrv_co_flush_to_disk        = vmdk_co_flush,
3074     .bdrv_co_block_status         = vmdk_co_block_status,
3075     .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
3076     .bdrv_has_zero_init           = vmdk_has_zero_init,
3077     .bdrv_get_specific_info       = vmdk_get_specific_info,
3078     .bdrv_refresh_limits          = vmdk_refresh_limits,
3079     .bdrv_get_info                = vmdk_get_info,
3080     .bdrv_gather_child_options    = vmdk_gather_child_options,
3081 
3082     .is_format                    = true,
3083     .supports_backing             = true,
3084     .create_opts                  = &vmdk_create_opts,
3085 };
3086 
3087 static void bdrv_vmdk_init(void)
3088 {
3089     bdrv_register(&bdrv_vmdk);
3090 }
3091 
3092 block_init(bdrv_vmdk_init);
3093