xref: /openbmc/qemu/block/vmdk.c (revision 0399a381)
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-common.h"
27 #include "block/block_int.h"
28 #include "qemu/module.h"
29 #include "migration/migration.h"
30 #include <zlib.h>
31 
32 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
33 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
34 #define VMDK4_COMPRESSION_DEFLATE 1
35 #define VMDK4_FLAG_NL_DETECT (1 << 0)
36 #define VMDK4_FLAG_RGD (1 << 1)
37 /* Zeroed-grain enable bit */
38 #define VMDK4_FLAG_ZERO_GRAIN   (1 << 2)
39 #define VMDK4_FLAG_COMPRESS (1 << 16)
40 #define VMDK4_FLAG_MARKER (1 << 17)
41 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
42 
43 #define VMDK_GTE_ZEROED 0x1
44 
45 /* VMDK internal error codes */
46 #define VMDK_OK      0
47 #define VMDK_ERROR   (-1)
48 /* Cluster not allocated */
49 #define VMDK_UNALLOC (-2)
50 #define VMDK_ZEROED  (-3)
51 
52 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
53 
54 typedef struct {
55     uint32_t version;
56     uint32_t flags;
57     uint32_t disk_sectors;
58     uint32_t granularity;
59     uint32_t l1dir_offset;
60     uint32_t l1dir_size;
61     uint32_t file_sectors;
62     uint32_t cylinders;
63     uint32_t heads;
64     uint32_t sectors_per_track;
65 } QEMU_PACKED VMDK3Header;
66 
67 typedef struct {
68     uint32_t version;
69     uint32_t flags;
70     uint64_t capacity;
71     uint64_t granularity;
72     uint64_t desc_offset;
73     uint64_t desc_size;
74     /* Number of GrainTableEntries per GrainTable */
75     uint32_t num_gtes_per_gt;
76     uint64_t rgd_offset;
77     uint64_t gd_offset;
78     uint64_t grain_offset;
79     char filler[1];
80     char check_bytes[4];
81     uint16_t compressAlgorithm;
82 } QEMU_PACKED VMDK4Header;
83 
84 #define L2_CACHE_SIZE 16
85 
86 typedef struct VmdkExtent {
87     BlockDriverState *file;
88     bool flat;
89     bool compressed;
90     bool has_marker;
91     bool has_zero_grain;
92     int version;
93     int64_t sectors;
94     int64_t end_sector;
95     int64_t flat_start_offset;
96     int64_t l1_table_offset;
97     int64_t l1_backup_table_offset;
98     uint32_t *l1_table;
99     uint32_t *l1_backup_table;
100     unsigned int l1_size;
101     uint32_t l1_entry_sectors;
102 
103     unsigned int l2_size;
104     uint32_t *l2_cache;
105     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
106     uint32_t l2_cache_counts[L2_CACHE_SIZE];
107 
108     int64_t cluster_sectors;
109     char *type;
110 } VmdkExtent;
111 
112 typedef struct BDRVVmdkState {
113     CoMutex lock;
114     uint64_t desc_offset;
115     bool cid_updated;
116     bool cid_checked;
117     uint32_t cid;
118     uint32_t parent_cid;
119     int num_extents;
120     /* Extent array with num_extents entries, ascend ordered by address */
121     VmdkExtent *extents;
122     Error *migration_blocker;
123     char *create_type;
124 } BDRVVmdkState;
125 
126 typedef struct VmdkMetaData {
127     uint32_t offset;
128     unsigned int l1_index;
129     unsigned int l2_index;
130     unsigned int l2_offset;
131     int valid;
132     uint32_t *l2_cache_entry;
133 } VmdkMetaData;
134 
135 typedef struct VmdkGrainMarker {
136     uint64_t lba;
137     uint32_t size;
138     uint8_t  data[0];
139 } QEMU_PACKED VmdkGrainMarker;
140 
141 enum {
142     MARKER_END_OF_STREAM    = 0,
143     MARKER_GRAIN_TABLE      = 1,
144     MARKER_GRAIN_DIRECTORY  = 2,
145     MARKER_FOOTER           = 3,
146 };
147 
148 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
149 {
150     uint32_t magic;
151 
152     if (buf_size < 4) {
153         return 0;
154     }
155     magic = be32_to_cpu(*(uint32_t *)buf);
156     if (magic == VMDK3_MAGIC ||
157         magic == VMDK4_MAGIC) {
158         return 100;
159     } else {
160         const char *p = (const char *)buf;
161         const char *end = p + buf_size;
162         while (p < end) {
163             if (*p == '#') {
164                 /* skip comment line */
165                 while (p < end && *p != '\n') {
166                     p++;
167                 }
168                 p++;
169                 continue;
170             }
171             if (*p == ' ') {
172                 while (p < end && *p == ' ') {
173                     p++;
174                 }
175                 /* skip '\r' if windows line endings used. */
176                 if (p < end && *p == '\r') {
177                     p++;
178                 }
179                 /* only accept blank lines before 'version=' line */
180                 if (p == end || *p != '\n') {
181                     return 0;
182                 }
183                 p++;
184                 continue;
185             }
186             if (end - p >= strlen("version=X\n")) {
187                 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
188                     strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
189                     return 100;
190                 }
191             }
192             if (end - p >= strlen("version=X\r\n")) {
193                 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
194                     strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
195                     return 100;
196                 }
197             }
198             return 0;
199         }
200         return 0;
201     }
202 }
203 
204 #define SECTOR_SIZE 512
205 #define DESC_SIZE (20 * SECTOR_SIZE)    /* 20 sectors of 512 bytes each */
206 #define BUF_SIZE 4096
207 #define HEADER_SIZE 512                 /* first sector of 512 bytes */
208 
209 static void vmdk_free_extents(BlockDriverState *bs)
210 {
211     int i;
212     BDRVVmdkState *s = bs->opaque;
213     VmdkExtent *e;
214 
215     for (i = 0; i < s->num_extents; i++) {
216         e = &s->extents[i];
217         g_free(e->l1_table);
218         g_free(e->l2_cache);
219         g_free(e->l1_backup_table);
220         g_free(e->type);
221         if (e->file != bs->file) {
222             bdrv_unref(e->file);
223         }
224     }
225     g_free(s->extents);
226 }
227 
228 static void vmdk_free_last_extent(BlockDriverState *bs)
229 {
230     BDRVVmdkState *s = bs->opaque;
231 
232     if (s->num_extents == 0) {
233         return;
234     }
235     s->num_extents--;
236     s->extents = g_realloc(s->extents, s->num_extents * sizeof(VmdkExtent));
237 }
238 
239 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
240 {
241     char desc[DESC_SIZE];
242     uint32_t cid = 0xffffffff;
243     const char *p_name, *cid_str;
244     size_t cid_str_size;
245     BDRVVmdkState *s = bs->opaque;
246     int ret;
247 
248     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
249     if (ret < 0) {
250         return 0;
251     }
252 
253     if (parent) {
254         cid_str = "parentCID";
255         cid_str_size = sizeof("parentCID");
256     } else {
257         cid_str = "CID";
258         cid_str_size = sizeof("CID");
259     }
260 
261     desc[DESC_SIZE - 1] = '\0';
262     p_name = strstr(desc, cid_str);
263     if (p_name != NULL) {
264         p_name += cid_str_size;
265         sscanf(p_name, "%x", &cid);
266     }
267 
268     return cid;
269 }
270 
271 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
272 {
273     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
274     char *p_name, *tmp_str;
275     BDRVVmdkState *s = bs->opaque;
276     int ret;
277 
278     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
279     if (ret < 0) {
280         return ret;
281     }
282 
283     desc[DESC_SIZE - 1] = '\0';
284     tmp_str = strstr(desc, "parentCID");
285     if (tmp_str == NULL) {
286         return -EINVAL;
287     }
288 
289     pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
290     p_name = strstr(desc, "CID");
291     if (p_name != NULL) {
292         p_name += sizeof("CID");
293         snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
294         pstrcat(desc, sizeof(desc), tmp_desc);
295     }
296 
297     ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
298     if (ret < 0) {
299         return ret;
300     }
301 
302     return 0;
303 }
304 
305 static int vmdk_is_cid_valid(BlockDriverState *bs)
306 {
307     BDRVVmdkState *s = bs->opaque;
308     BlockDriverState *p_bs = bs->backing_hd;
309     uint32_t cur_pcid;
310 
311     if (!s->cid_checked && p_bs) {
312         cur_pcid = vmdk_read_cid(p_bs, 0);
313         if (s->parent_cid != cur_pcid) {
314             /* CID not valid */
315             return 0;
316         }
317     }
318     s->cid_checked = true;
319     /* CID valid */
320     return 1;
321 }
322 
323 /* Queue extents, if any, for reopen() */
324 static int vmdk_reopen_prepare(BDRVReopenState *state,
325                                BlockReopenQueue *queue, Error **errp)
326 {
327     BDRVVmdkState *s;
328     int ret = -1;
329     int i;
330     VmdkExtent *e;
331 
332     assert(state != NULL);
333     assert(state->bs != NULL);
334 
335     if (queue == NULL) {
336         error_setg(errp, "No reopen queue for VMDK extents");
337         goto exit;
338     }
339 
340     s = state->bs->opaque;
341 
342     assert(s != NULL);
343 
344     for (i = 0; i < s->num_extents; i++) {
345         e = &s->extents[i];
346         if (e->file != state->bs->file) {
347             bdrv_reopen_queue(queue, e->file, state->flags);
348         }
349     }
350     ret = 0;
351 
352 exit:
353     return ret;
354 }
355 
356 static int vmdk_parent_open(BlockDriverState *bs)
357 {
358     char *p_name;
359     char desc[DESC_SIZE + 1];
360     BDRVVmdkState *s = bs->opaque;
361     int ret;
362 
363     desc[DESC_SIZE] = '\0';
364     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
365     if (ret < 0) {
366         return ret;
367     }
368 
369     p_name = strstr(desc, "parentFileNameHint");
370     if (p_name != NULL) {
371         char *end_name;
372 
373         p_name += sizeof("parentFileNameHint") + 1;
374         end_name = strchr(p_name, '\"');
375         if (end_name == NULL) {
376             return -EINVAL;
377         }
378         if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
379             return -EINVAL;
380         }
381 
382         pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
383     }
384 
385     return 0;
386 }
387 
388 /* Create and append extent to the extent array. Return the added VmdkExtent
389  * address. return NULL if allocation failed. */
390 static int vmdk_add_extent(BlockDriverState *bs,
391                            BlockDriverState *file, bool flat, int64_t sectors,
392                            int64_t l1_offset, int64_t l1_backup_offset,
393                            uint32_t l1_size,
394                            int l2_size, uint64_t cluster_sectors,
395                            VmdkExtent **new_extent,
396                            Error **errp)
397 {
398     VmdkExtent *extent;
399     BDRVVmdkState *s = bs->opaque;
400 
401     if (cluster_sectors > 0x200000) {
402         /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
403         error_setg(errp, "Invalid granularity, image may be corrupt");
404         return -EFBIG;
405     }
406     if (l1_size > 512 * 1024 * 1024) {
407         /* Although with big capacity and small l1_entry_sectors, we can get a
408          * big l1_size, we don't want unbounded value to allocate the table.
409          * Limit it to 512M, which is 16PB for default cluster and L2 table
410          * size */
411         error_setg(errp, "L1 size too big");
412         return -EFBIG;
413     }
414 
415     s->extents = g_realloc(s->extents,
416                               (s->num_extents + 1) * sizeof(VmdkExtent));
417     extent = &s->extents[s->num_extents];
418     s->num_extents++;
419 
420     memset(extent, 0, sizeof(VmdkExtent));
421     extent->file = file;
422     extent->flat = flat;
423     extent->sectors = sectors;
424     extent->l1_table_offset = l1_offset;
425     extent->l1_backup_table_offset = l1_backup_offset;
426     extent->l1_size = l1_size;
427     extent->l1_entry_sectors = l2_size * cluster_sectors;
428     extent->l2_size = l2_size;
429     extent->cluster_sectors = flat ? sectors : cluster_sectors;
430 
431     if (s->num_extents > 1) {
432         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
433     } else {
434         extent->end_sector = extent->sectors;
435     }
436     bs->total_sectors = extent->end_sector;
437     if (new_extent) {
438         *new_extent = extent;
439     }
440     return 0;
441 }
442 
443 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
444                             Error **errp)
445 {
446     int ret;
447     int l1_size, i;
448 
449     /* read the L1 table */
450     l1_size = extent->l1_size * sizeof(uint32_t);
451     extent->l1_table = g_malloc(l1_size);
452     ret = bdrv_pread(extent->file,
453                      extent->l1_table_offset,
454                      extent->l1_table,
455                      l1_size);
456     if (ret < 0) {
457         error_setg_errno(errp, -ret,
458                          "Could not read l1 table from extent '%s'",
459                          extent->file->filename);
460         goto fail_l1;
461     }
462     for (i = 0; i < extent->l1_size; i++) {
463         le32_to_cpus(&extent->l1_table[i]);
464     }
465 
466     if (extent->l1_backup_table_offset) {
467         extent->l1_backup_table = g_malloc(l1_size);
468         ret = bdrv_pread(extent->file,
469                          extent->l1_backup_table_offset,
470                          extent->l1_backup_table,
471                          l1_size);
472         if (ret < 0) {
473             error_setg_errno(errp, -ret,
474                              "Could not read l1 backup table from extent '%s'",
475                              extent->file->filename);
476             goto fail_l1b;
477         }
478         for (i = 0; i < extent->l1_size; i++) {
479             le32_to_cpus(&extent->l1_backup_table[i]);
480         }
481     }
482 
483     extent->l2_cache =
484         g_malloc(extent->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
485     return 0;
486  fail_l1b:
487     g_free(extent->l1_backup_table);
488  fail_l1:
489     g_free(extent->l1_table);
490     return ret;
491 }
492 
493 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
494                                  BlockDriverState *file,
495                                  int flags, Error **errp)
496 {
497     int ret;
498     uint32_t magic;
499     VMDK3Header header;
500     VmdkExtent *extent;
501 
502     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
503     if (ret < 0) {
504         error_setg_errno(errp, -ret,
505                          "Could not read header from file '%s'",
506                          file->filename);
507         return ret;
508     }
509     ret = vmdk_add_extent(bs, file, false,
510                           le32_to_cpu(header.disk_sectors),
511                           le32_to_cpu(header.l1dir_offset) << 9,
512                           0,
513                           le32_to_cpu(header.l1dir_size),
514                           4096,
515                           le32_to_cpu(header.granularity),
516                           &extent,
517                           errp);
518     if (ret < 0) {
519         return ret;
520     }
521     ret = vmdk_init_tables(bs, extent, errp);
522     if (ret) {
523         /* free extent allocated by vmdk_add_extent */
524         vmdk_free_last_extent(bs);
525     }
526     return ret;
527 }
528 
529 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
530                                Error **errp);
531 
532 static char *vmdk_read_desc(BlockDriverState *file, uint64_t desc_offset,
533                             Error **errp)
534 {
535     int64_t size;
536     char *buf;
537     int ret;
538 
539     size = bdrv_getlength(file);
540     if (size < 0) {
541         error_setg_errno(errp, -size, "Could not access file");
542         return NULL;
543     }
544 
545     size = MIN(size, 1 << 20);  /* avoid unbounded allocation */
546     buf = g_malloc0(size + 1);
547 
548     ret = bdrv_pread(file, desc_offset, buf, size);
549     if (ret < 0) {
550         error_setg_errno(errp, -ret, "Could not read from file");
551         g_free(buf);
552         return NULL;
553     }
554 
555     return buf;
556 }
557 
558 static int vmdk_open_vmdk4(BlockDriverState *bs,
559                            BlockDriverState *file,
560                            int flags, Error **errp)
561 {
562     int ret;
563     uint32_t magic;
564     uint32_t l1_size, l1_entry_sectors;
565     VMDK4Header header;
566     VmdkExtent *extent;
567     BDRVVmdkState *s = bs->opaque;
568     int64_t l1_backup_offset = 0;
569 
570     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
571     if (ret < 0) {
572         error_setg_errno(errp, -ret,
573                          "Could not read header from file '%s'",
574                          file->filename);
575         return -EINVAL;
576     }
577     if (header.capacity == 0) {
578         uint64_t desc_offset = le64_to_cpu(header.desc_offset);
579         if (desc_offset) {
580             char *buf = vmdk_read_desc(file, desc_offset << 9, errp);
581             if (!buf) {
582                 return -EINVAL;
583             }
584             ret = vmdk_open_desc_file(bs, flags, buf, errp);
585             g_free(buf);
586             return ret;
587         }
588     }
589 
590     if (!s->create_type) {
591         s->create_type = g_strdup("monolithicSparse");
592     }
593 
594     if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
595         /*
596          * The footer takes precedence over the header, so read it in. The
597          * footer starts at offset -1024 from the end: One sector for the
598          * footer, and another one for the end-of-stream marker.
599          */
600         struct {
601             struct {
602                 uint64_t val;
603                 uint32_t size;
604                 uint32_t type;
605                 uint8_t pad[512 - 16];
606             } QEMU_PACKED footer_marker;
607 
608             uint32_t magic;
609             VMDK4Header header;
610             uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
611 
612             struct {
613                 uint64_t val;
614                 uint32_t size;
615                 uint32_t type;
616                 uint8_t pad[512 - 16];
617             } QEMU_PACKED eos_marker;
618         } QEMU_PACKED footer;
619 
620         ret = bdrv_pread(file,
621             bs->file->total_sectors * 512 - 1536,
622             &footer, sizeof(footer));
623         if (ret < 0) {
624             return ret;
625         }
626 
627         /* Some sanity checks for the footer */
628         if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
629             le32_to_cpu(footer.footer_marker.size) != 0  ||
630             le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
631             le64_to_cpu(footer.eos_marker.val) != 0  ||
632             le32_to_cpu(footer.eos_marker.size) != 0  ||
633             le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
634         {
635             return -EINVAL;
636         }
637 
638         header = footer.header;
639     }
640 
641     if (le32_to_cpu(header.version) > 3) {
642         char buf[64];
643         snprintf(buf, sizeof(buf), "VMDK version %d",
644                  le32_to_cpu(header.version));
645         error_set(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
646                   bs->device_name, "vmdk", buf);
647         return -ENOTSUP;
648     } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR)) {
649         /* VMware KB 2064959 explains that version 3 added support for
650          * persistent changed block tracking (CBT), and backup software can
651          * read it as version=1 if it doesn't care about the changed area
652          * information. So we are safe to enable read only. */
653         error_setg(errp, "VMDK version 3 must be read only");
654         return -EINVAL;
655     }
656 
657     if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
658         error_setg(errp, "L2 table size too big");
659         return -EINVAL;
660     }
661 
662     l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
663                         * le64_to_cpu(header.granularity);
664     if (l1_entry_sectors == 0) {
665         return -EINVAL;
666     }
667     l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
668                 / l1_entry_sectors;
669     if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
670         l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
671     }
672     if (bdrv_getlength(file) <
673             le64_to_cpu(header.grain_offset) * BDRV_SECTOR_SIZE) {
674         error_setg(errp, "File truncated, expecting at least %lld bytes",
675                    le64_to_cpu(header.grain_offset) * BDRV_SECTOR_SIZE);
676         return -EINVAL;
677     }
678 
679     ret = vmdk_add_extent(bs, file, false,
680                           le64_to_cpu(header.capacity),
681                           le64_to_cpu(header.gd_offset) << 9,
682                           l1_backup_offset,
683                           l1_size,
684                           le32_to_cpu(header.num_gtes_per_gt),
685                           le64_to_cpu(header.granularity),
686                           &extent,
687                           errp);
688     if (ret < 0) {
689         return ret;
690     }
691     extent->compressed =
692         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
693     if (extent->compressed) {
694         g_free(s->create_type);
695         s->create_type = g_strdup("streamOptimized");
696     }
697     extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
698     extent->version = le32_to_cpu(header.version);
699     extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
700     ret = vmdk_init_tables(bs, extent, errp);
701     if (ret) {
702         /* free extent allocated by vmdk_add_extent */
703         vmdk_free_last_extent(bs);
704     }
705     return ret;
706 }
707 
708 /* find an option value out of descriptor file */
709 static int vmdk_parse_description(const char *desc, const char *opt_name,
710         char *buf, int buf_size)
711 {
712     char *opt_pos, *opt_end;
713     const char *end = desc + strlen(desc);
714 
715     opt_pos = strstr(desc, opt_name);
716     if (!opt_pos) {
717         return VMDK_ERROR;
718     }
719     /* Skip "=\"" following opt_name */
720     opt_pos += strlen(opt_name) + 2;
721     if (opt_pos >= end) {
722         return VMDK_ERROR;
723     }
724     opt_end = opt_pos;
725     while (opt_end < end && *opt_end != '"') {
726         opt_end++;
727     }
728     if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
729         return VMDK_ERROR;
730     }
731     pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
732     return VMDK_OK;
733 }
734 
735 /* Open an extent file and append to bs array */
736 static int vmdk_open_sparse(BlockDriverState *bs,
737                             BlockDriverState *file, int flags,
738                             char *buf, Error **errp)
739 {
740     uint32_t magic;
741 
742     magic = ldl_be_p(buf);
743     switch (magic) {
744         case VMDK3_MAGIC:
745             return vmdk_open_vmfs_sparse(bs, file, flags, errp);
746             break;
747         case VMDK4_MAGIC:
748             return vmdk_open_vmdk4(bs, file, flags, errp);
749             break;
750         default:
751             error_setg(errp, "Image not in VMDK format");
752             return -EINVAL;
753             break;
754     }
755 }
756 
757 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
758                               const char *desc_file_path, Error **errp)
759 {
760     int ret;
761     char access[11];
762     char type[11];
763     char fname[512];
764     const char *p = desc;
765     int64_t sectors = 0;
766     int64_t flat_offset;
767     char extent_path[PATH_MAX];
768     BlockDriverState *extent_file;
769     BDRVVmdkState *s = bs->opaque;
770     VmdkExtent *extent;
771 
772     while (*p) {
773         /* parse extent line:
774          * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
775          * or
776          * RW [size in sectors] SPARSE "file-name.vmdk"
777          */
778         flat_offset = -1;
779         ret = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
780                 access, &sectors, type, fname, &flat_offset);
781         if (ret < 4 || strcmp(access, "RW")) {
782             goto next_line;
783         } else if (!strcmp(type, "FLAT")) {
784             if (ret != 5 || flat_offset < 0) {
785                 error_setg(errp, "Invalid extent lines: \n%s", p);
786                 return -EINVAL;
787             }
788         } else if (!strcmp(type, "VMFS")) {
789             if (ret == 4) {
790                 flat_offset = 0;
791             } else {
792                 error_setg(errp, "Invalid extent lines:\n%s", p);
793                 return -EINVAL;
794             }
795         } else if (ret != 4) {
796             error_setg(errp, "Invalid extent lines:\n%s", p);
797             return -EINVAL;
798         }
799 
800         if (sectors <= 0 ||
801             (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
802              strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
803             (strcmp(access, "RW"))) {
804             goto next_line;
805         }
806 
807         path_combine(extent_path, sizeof(extent_path),
808                 desc_file_path, fname);
809         extent_file = NULL;
810         ret = bdrv_open(&extent_file, extent_path, NULL, NULL,
811                         bs->open_flags | BDRV_O_PROTOCOL, NULL, errp);
812         if (ret) {
813             return ret;
814         }
815 
816         /* save to extents array */
817         if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
818             /* FLAT extent */
819 
820             ret = vmdk_add_extent(bs, extent_file, true, sectors,
821                             0, 0, 0, 0, 0, &extent, errp);
822             if (ret < 0) {
823                 return ret;
824             }
825             extent->flat_start_offset = flat_offset << 9;
826         } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
827             /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
828             char *buf = vmdk_read_desc(extent_file, 0, errp);
829             if (!buf) {
830                 ret = -EINVAL;
831             } else {
832                 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf, errp);
833             }
834             if (ret) {
835                 g_free(buf);
836                 bdrv_unref(extent_file);
837                 return ret;
838             }
839             extent = &s->extents[s->num_extents - 1];
840         } else {
841             error_setg(errp, "Unsupported extent type '%s'", type);
842             return -ENOTSUP;
843         }
844         extent->type = g_strdup(type);
845 next_line:
846         /* move to next line */
847         while (*p) {
848             if (*p == '\n') {
849                 p++;
850                 break;
851             }
852             p++;
853         }
854     }
855     return 0;
856 }
857 
858 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
859                                Error **errp)
860 {
861     int ret;
862     char ct[128];
863     BDRVVmdkState *s = bs->opaque;
864 
865     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
866         error_setg(errp, "invalid VMDK image descriptor");
867         ret = -EINVAL;
868         goto exit;
869     }
870     if (strcmp(ct, "monolithicFlat") &&
871         strcmp(ct, "vmfs") &&
872         strcmp(ct, "vmfsSparse") &&
873         strcmp(ct, "twoGbMaxExtentSparse") &&
874         strcmp(ct, "twoGbMaxExtentFlat")) {
875         error_setg(errp, "Unsupported image type '%s'", ct);
876         ret = -ENOTSUP;
877         goto exit;
878     }
879     s->create_type = g_strdup(ct);
880     s->desc_offset = 0;
881     ret = vmdk_parse_extents(buf, bs, bs->file->filename, errp);
882 exit:
883     return ret;
884 }
885 
886 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
887                      Error **errp)
888 {
889     char *buf = NULL;
890     int ret;
891     BDRVVmdkState *s = bs->opaque;
892     uint32_t magic;
893 
894     buf = vmdk_read_desc(bs->file, 0, errp);
895     if (!buf) {
896         return -EINVAL;
897     }
898 
899     magic = ldl_be_p(buf);
900     switch (magic) {
901         case VMDK3_MAGIC:
902         case VMDK4_MAGIC:
903             ret = vmdk_open_sparse(bs, bs->file, flags, buf, errp);
904             s->desc_offset = 0x200;
905             break;
906         default:
907             ret = vmdk_open_desc_file(bs, flags, buf, errp);
908             break;
909     }
910     if (ret) {
911         goto fail;
912     }
913 
914     /* try to open parent images, if exist */
915     ret = vmdk_parent_open(bs);
916     if (ret) {
917         goto fail;
918     }
919     s->cid = vmdk_read_cid(bs, 0);
920     s->parent_cid = vmdk_read_cid(bs, 1);
921     qemu_co_mutex_init(&s->lock);
922 
923     /* Disable migration when VMDK images are used */
924     error_set(&s->migration_blocker,
925               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
926               "vmdk", bs->device_name, "live migration");
927     migrate_add_blocker(s->migration_blocker);
928     g_free(buf);
929     return 0;
930 
931 fail:
932     g_free(buf);
933     g_free(s->create_type);
934     s->create_type = NULL;
935     vmdk_free_extents(bs);
936     return ret;
937 }
938 
939 
940 static int vmdk_refresh_limits(BlockDriverState *bs)
941 {
942     BDRVVmdkState *s = bs->opaque;
943     int i;
944 
945     for (i = 0; i < s->num_extents; i++) {
946         if (!s->extents[i].flat) {
947             bs->bl.write_zeroes_alignment =
948                 MAX(bs->bl.write_zeroes_alignment,
949                     s->extents[i].cluster_sectors);
950         }
951     }
952 
953     return 0;
954 }
955 
956 static int get_whole_cluster(BlockDriverState *bs,
957                 VmdkExtent *extent,
958                 uint64_t cluster_offset,
959                 uint64_t offset,
960                 bool allocate)
961 {
962     int ret = VMDK_OK;
963     uint8_t *whole_grain = NULL;
964 
965     /* we will be here if it's first write on non-exist grain(cluster).
966      * try to read from parent image, if exist */
967     if (bs->backing_hd) {
968         whole_grain =
969             qemu_blockalign(bs, extent->cluster_sectors << BDRV_SECTOR_BITS);
970         if (!vmdk_is_cid_valid(bs)) {
971             ret = VMDK_ERROR;
972             goto exit;
973         }
974 
975         /* floor offset to cluster */
976         offset -= offset % (extent->cluster_sectors * 512);
977         ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain,
978                 extent->cluster_sectors);
979         if (ret < 0) {
980             ret = VMDK_ERROR;
981             goto exit;
982         }
983 
984         /* Write grain only into the active image */
985         ret = bdrv_write(extent->file, cluster_offset, whole_grain,
986                 extent->cluster_sectors);
987         if (ret < 0) {
988             ret = VMDK_ERROR;
989             goto exit;
990         }
991     }
992 exit:
993     qemu_vfree(whole_grain);
994     return ret;
995 }
996 
997 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data)
998 {
999     uint32_t offset;
1000     QEMU_BUILD_BUG_ON(sizeof(offset) != sizeof(m_data->offset));
1001     offset = cpu_to_le32(m_data->offset);
1002     /* update L2 table */
1003     if (bdrv_pwrite_sync(
1004                 extent->file,
1005                 ((int64_t)m_data->l2_offset * 512)
1006                     + (m_data->l2_index * sizeof(m_data->offset)),
1007                 &offset, sizeof(offset)) < 0) {
1008         return VMDK_ERROR;
1009     }
1010     /* update backup L2 table */
1011     if (extent->l1_backup_table_offset != 0) {
1012         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1013         if (bdrv_pwrite_sync(
1014                     extent->file,
1015                     ((int64_t)m_data->l2_offset * 512)
1016                         + (m_data->l2_index * sizeof(m_data->offset)),
1017                     &offset, sizeof(offset)) < 0) {
1018             return VMDK_ERROR;
1019         }
1020     }
1021     if (m_data->l2_cache_entry) {
1022         *m_data->l2_cache_entry = offset;
1023     }
1024 
1025     return VMDK_OK;
1026 }
1027 
1028 static int get_cluster_offset(BlockDriverState *bs,
1029                                     VmdkExtent *extent,
1030                                     VmdkMetaData *m_data,
1031                                     uint64_t offset,
1032                                     int allocate,
1033                                     uint64_t *cluster_offset)
1034 {
1035     unsigned int l1_index, l2_offset, l2_index;
1036     int min_index, i, j;
1037     uint32_t min_count, *l2_table;
1038     bool zeroed = false;
1039 
1040     if (m_data) {
1041         m_data->valid = 0;
1042     }
1043     if (extent->flat) {
1044         *cluster_offset = extent->flat_start_offset;
1045         return VMDK_OK;
1046     }
1047 
1048     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1049     l1_index = (offset >> 9) / extent->l1_entry_sectors;
1050     if (l1_index >= extent->l1_size) {
1051         return VMDK_ERROR;
1052     }
1053     l2_offset = extent->l1_table[l1_index];
1054     if (!l2_offset) {
1055         return VMDK_UNALLOC;
1056     }
1057     for (i = 0; i < L2_CACHE_SIZE; i++) {
1058         if (l2_offset == extent->l2_cache_offsets[i]) {
1059             /* increment the hit count */
1060             if (++extent->l2_cache_counts[i] == 0xffffffff) {
1061                 for (j = 0; j < L2_CACHE_SIZE; j++) {
1062                     extent->l2_cache_counts[j] >>= 1;
1063                 }
1064             }
1065             l2_table = extent->l2_cache + (i * extent->l2_size);
1066             goto found;
1067         }
1068     }
1069     /* not found: load a new entry in the least used one */
1070     min_index = 0;
1071     min_count = 0xffffffff;
1072     for (i = 0; i < L2_CACHE_SIZE; i++) {
1073         if (extent->l2_cache_counts[i] < min_count) {
1074             min_count = extent->l2_cache_counts[i];
1075             min_index = i;
1076         }
1077     }
1078     l2_table = extent->l2_cache + (min_index * extent->l2_size);
1079     if (bdrv_pread(
1080                 extent->file,
1081                 (int64_t)l2_offset * 512,
1082                 l2_table,
1083                 extent->l2_size * sizeof(uint32_t)
1084             ) != extent->l2_size * sizeof(uint32_t)) {
1085         return VMDK_ERROR;
1086     }
1087 
1088     extent->l2_cache_offsets[min_index] = l2_offset;
1089     extent->l2_cache_counts[min_index] = 1;
1090  found:
1091     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1092     *cluster_offset = le32_to_cpu(l2_table[l2_index]);
1093 
1094     if (m_data) {
1095         m_data->valid = 1;
1096         m_data->l1_index = l1_index;
1097         m_data->l2_index = l2_index;
1098         m_data->offset = *cluster_offset;
1099         m_data->l2_offset = l2_offset;
1100         m_data->l2_cache_entry = &l2_table[l2_index];
1101     }
1102     if (extent->has_zero_grain && *cluster_offset == VMDK_GTE_ZEROED) {
1103         zeroed = true;
1104     }
1105 
1106     if (!*cluster_offset || zeroed) {
1107         if (!allocate) {
1108             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1109         }
1110 
1111         /* Avoid the L2 tables update for the images that have snapshots. */
1112         *cluster_offset = bdrv_getlength(extent->file);
1113         if (!extent->compressed) {
1114             bdrv_truncate(
1115                 extent->file,
1116                 *cluster_offset + (extent->cluster_sectors << 9)
1117             );
1118         }
1119 
1120         *cluster_offset >>= 9;
1121         l2_table[l2_index] = cpu_to_le32(*cluster_offset);
1122 
1123         /* First of all we write grain itself, to avoid race condition
1124          * that may to corrupt the image.
1125          * This problem may occur because of insufficient space on host disk
1126          * or inappropriate VM shutdown.
1127          */
1128         if (get_whole_cluster(
1129                 bs, extent, *cluster_offset, offset, allocate) == -1) {
1130             return VMDK_ERROR;
1131         }
1132 
1133         if (m_data) {
1134             m_data->offset = *cluster_offset;
1135         }
1136     }
1137     *cluster_offset <<= 9;
1138     return VMDK_OK;
1139 }
1140 
1141 static VmdkExtent *find_extent(BDRVVmdkState *s,
1142                                 int64_t sector_num, VmdkExtent *start_hint)
1143 {
1144     VmdkExtent *extent = start_hint;
1145 
1146     if (!extent) {
1147         extent = &s->extents[0];
1148     }
1149     while (extent < &s->extents[s->num_extents]) {
1150         if (sector_num < extent->end_sector) {
1151             return extent;
1152         }
1153         extent++;
1154     }
1155     return NULL;
1156 }
1157 
1158 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs,
1159         int64_t sector_num, int nb_sectors, int *pnum)
1160 {
1161     BDRVVmdkState *s = bs->opaque;
1162     int64_t index_in_cluster, n, ret;
1163     uint64_t offset;
1164     VmdkExtent *extent;
1165 
1166     extent = find_extent(s, sector_num, NULL);
1167     if (!extent) {
1168         return 0;
1169     }
1170     qemu_co_mutex_lock(&s->lock);
1171     ret = get_cluster_offset(bs, extent, NULL,
1172                             sector_num * 512, 0, &offset);
1173     qemu_co_mutex_unlock(&s->lock);
1174 
1175     switch (ret) {
1176     case VMDK_ERROR:
1177         ret = -EIO;
1178         break;
1179     case VMDK_UNALLOC:
1180         ret = 0;
1181         break;
1182     case VMDK_ZEROED:
1183         ret = BDRV_BLOCK_ZERO;
1184         break;
1185     case VMDK_OK:
1186         ret = BDRV_BLOCK_DATA;
1187         if (extent->file == bs->file && !extent->compressed) {
1188             ret |= BDRV_BLOCK_OFFSET_VALID | offset;
1189         }
1190 
1191         break;
1192     }
1193 
1194     index_in_cluster = sector_num % extent->cluster_sectors;
1195     n = extent->cluster_sectors - index_in_cluster;
1196     if (n > nb_sectors) {
1197         n = nb_sectors;
1198     }
1199     *pnum = n;
1200     return ret;
1201 }
1202 
1203 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1204                             int64_t offset_in_cluster, const uint8_t *buf,
1205                             int nb_sectors, int64_t sector_num)
1206 {
1207     int ret;
1208     VmdkGrainMarker *data = NULL;
1209     uLongf buf_len;
1210     const uint8_t *write_buf = buf;
1211     int write_len = nb_sectors * 512;
1212 
1213     if (extent->compressed) {
1214         if (!extent->has_marker) {
1215             ret = -EINVAL;
1216             goto out;
1217         }
1218         buf_len = (extent->cluster_sectors << 9) * 2;
1219         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1220         if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK ||
1221                 buf_len == 0) {
1222             ret = -EINVAL;
1223             goto out;
1224         }
1225         data->lba = sector_num;
1226         data->size = buf_len;
1227         write_buf = (uint8_t *)data;
1228         write_len = buf_len + sizeof(VmdkGrainMarker);
1229     }
1230     ret = bdrv_pwrite(extent->file,
1231                         cluster_offset + offset_in_cluster,
1232                         write_buf,
1233                         write_len);
1234     if (ret != write_len) {
1235         ret = ret < 0 ? ret : -EIO;
1236         goto out;
1237     }
1238     ret = 0;
1239  out:
1240     g_free(data);
1241     return ret;
1242 }
1243 
1244 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1245                             int64_t offset_in_cluster, uint8_t *buf,
1246                             int nb_sectors)
1247 {
1248     int ret;
1249     int cluster_bytes, buf_bytes;
1250     uint8_t *cluster_buf, *compressed_data;
1251     uint8_t *uncomp_buf;
1252     uint32_t data_len;
1253     VmdkGrainMarker *marker;
1254     uLongf buf_len;
1255 
1256 
1257     if (!extent->compressed) {
1258         ret = bdrv_pread(extent->file,
1259                           cluster_offset + offset_in_cluster,
1260                           buf, nb_sectors * 512);
1261         if (ret == nb_sectors * 512) {
1262             return 0;
1263         } else {
1264             return -EIO;
1265         }
1266     }
1267     cluster_bytes = extent->cluster_sectors * 512;
1268     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1269     buf_bytes = cluster_bytes * 2;
1270     cluster_buf = g_malloc(buf_bytes);
1271     uncomp_buf = g_malloc(cluster_bytes);
1272     ret = bdrv_pread(extent->file,
1273                 cluster_offset,
1274                 cluster_buf, buf_bytes);
1275     if (ret < 0) {
1276         goto out;
1277     }
1278     compressed_data = cluster_buf;
1279     buf_len = cluster_bytes;
1280     data_len = cluster_bytes;
1281     if (extent->has_marker) {
1282         marker = (VmdkGrainMarker *)cluster_buf;
1283         compressed_data = marker->data;
1284         data_len = le32_to_cpu(marker->size);
1285     }
1286     if (!data_len || data_len > buf_bytes) {
1287         ret = -EINVAL;
1288         goto out;
1289     }
1290     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1291     if (ret != Z_OK) {
1292         ret = -EINVAL;
1293         goto out;
1294 
1295     }
1296     if (offset_in_cluster < 0 ||
1297             offset_in_cluster + nb_sectors * 512 > buf_len) {
1298         ret = -EINVAL;
1299         goto out;
1300     }
1301     memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512);
1302     ret = 0;
1303 
1304  out:
1305     g_free(uncomp_buf);
1306     g_free(cluster_buf);
1307     return ret;
1308 }
1309 
1310 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
1311                     uint8_t *buf, int nb_sectors)
1312 {
1313     BDRVVmdkState *s = bs->opaque;
1314     int ret;
1315     uint64_t n, index_in_cluster;
1316     uint64_t extent_begin_sector, extent_relative_sector_num;
1317     VmdkExtent *extent = NULL;
1318     uint64_t cluster_offset;
1319 
1320     while (nb_sectors > 0) {
1321         extent = find_extent(s, sector_num, extent);
1322         if (!extent) {
1323             return -EIO;
1324         }
1325         ret = get_cluster_offset(
1326                             bs, extent, NULL,
1327                             sector_num << 9, 0, &cluster_offset);
1328         extent_begin_sector = extent->end_sector - extent->sectors;
1329         extent_relative_sector_num = sector_num - extent_begin_sector;
1330         index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1331         n = extent->cluster_sectors - index_in_cluster;
1332         if (n > nb_sectors) {
1333             n = nb_sectors;
1334         }
1335         if (ret != VMDK_OK) {
1336             /* if not allocated, try to read from parent image, if exist */
1337             if (bs->backing_hd && ret != VMDK_ZEROED) {
1338                 if (!vmdk_is_cid_valid(bs)) {
1339                     return -EINVAL;
1340                 }
1341                 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
1342                 if (ret < 0) {
1343                     return ret;
1344                 }
1345             } else {
1346                 memset(buf, 0, 512 * n);
1347             }
1348         } else {
1349             ret = vmdk_read_extent(extent,
1350                             cluster_offset, index_in_cluster * 512,
1351                             buf, n);
1352             if (ret) {
1353                 return ret;
1354             }
1355         }
1356         nb_sectors -= n;
1357         sector_num += n;
1358         buf += n * 512;
1359     }
1360     return 0;
1361 }
1362 
1363 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num,
1364                                      uint8_t *buf, int nb_sectors)
1365 {
1366     int ret;
1367     BDRVVmdkState *s = bs->opaque;
1368     qemu_co_mutex_lock(&s->lock);
1369     ret = vmdk_read(bs, sector_num, buf, nb_sectors);
1370     qemu_co_mutex_unlock(&s->lock);
1371     return ret;
1372 }
1373 
1374 /**
1375  * vmdk_write:
1376  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
1377  *                if possible, otherwise return -ENOTSUP.
1378  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1379  *                with each cluster. By dry run we can find if the zero write
1380  *                is possible without modifying image data.
1381  *
1382  * Returns: error code with 0 for success.
1383  */
1384 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
1385                       const uint8_t *buf, int nb_sectors,
1386                       bool zeroed, bool zero_dry_run)
1387 {
1388     BDRVVmdkState *s = bs->opaque;
1389     VmdkExtent *extent = NULL;
1390     int ret;
1391     int64_t index_in_cluster, n;
1392     uint64_t extent_begin_sector, extent_relative_sector_num;
1393     uint64_t cluster_offset;
1394     VmdkMetaData m_data;
1395 
1396     if (sector_num > bs->total_sectors) {
1397         error_report("Wrong offset: sector_num=0x%" PRIx64
1398                 " total_sectors=0x%" PRIx64 "\n",
1399                 sector_num, bs->total_sectors);
1400         return -EIO;
1401     }
1402 
1403     while (nb_sectors > 0) {
1404         extent = find_extent(s, sector_num, extent);
1405         if (!extent) {
1406             return -EIO;
1407         }
1408         ret = get_cluster_offset(
1409                                 bs,
1410                                 extent,
1411                                 &m_data,
1412                                 sector_num << 9, !extent->compressed,
1413                                 &cluster_offset);
1414         if (extent->compressed) {
1415             if (ret == VMDK_OK) {
1416                 /* Refuse write to allocated cluster for streamOptimized */
1417                 error_report("Could not write to allocated cluster"
1418                               " for streamOptimized");
1419                 return -EIO;
1420             } else {
1421                 /* allocate */
1422                 ret = get_cluster_offset(
1423                                         bs,
1424                                         extent,
1425                                         &m_data,
1426                                         sector_num << 9, 1,
1427                                         &cluster_offset);
1428             }
1429         }
1430         if (ret == VMDK_ERROR) {
1431             return -EINVAL;
1432         }
1433         extent_begin_sector = extent->end_sector - extent->sectors;
1434         extent_relative_sector_num = sector_num - extent_begin_sector;
1435         index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1436         n = extent->cluster_sectors - index_in_cluster;
1437         if (n > nb_sectors) {
1438             n = nb_sectors;
1439         }
1440         if (zeroed) {
1441             /* Do zeroed write, buf is ignored */
1442             if (extent->has_zero_grain &&
1443                     index_in_cluster == 0 &&
1444                     n >= extent->cluster_sectors) {
1445                 n = extent->cluster_sectors;
1446                 if (!zero_dry_run) {
1447                     m_data.offset = VMDK_GTE_ZEROED;
1448                     /* update L2 tables */
1449                     if (vmdk_L2update(extent, &m_data) != VMDK_OK) {
1450                         return -EIO;
1451                     }
1452                 }
1453             } else {
1454                 return -ENOTSUP;
1455             }
1456         } else {
1457             ret = vmdk_write_extent(extent,
1458                             cluster_offset, index_in_cluster * 512,
1459                             buf, n, sector_num);
1460             if (ret) {
1461                 return ret;
1462             }
1463             if (m_data.valid) {
1464                 /* update L2 tables */
1465                 if (vmdk_L2update(extent, &m_data) != VMDK_OK) {
1466                     return -EIO;
1467                 }
1468             }
1469         }
1470         nb_sectors -= n;
1471         sector_num += n;
1472         buf += n * 512;
1473 
1474         /* update CID on the first write every time the virtual disk is
1475          * opened */
1476         if (!s->cid_updated) {
1477             ret = vmdk_write_cid(bs, time(NULL));
1478             if (ret < 0) {
1479                 return ret;
1480             }
1481             s->cid_updated = true;
1482         }
1483     }
1484     return 0;
1485 }
1486 
1487 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num,
1488                                       const uint8_t *buf, int nb_sectors)
1489 {
1490     int ret;
1491     BDRVVmdkState *s = bs->opaque;
1492     qemu_co_mutex_lock(&s->lock);
1493     ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false);
1494     qemu_co_mutex_unlock(&s->lock);
1495     return ret;
1496 }
1497 
1498 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs,
1499                                              int64_t sector_num,
1500                                              int nb_sectors,
1501                                              BdrvRequestFlags flags)
1502 {
1503     int ret;
1504     BDRVVmdkState *s = bs->opaque;
1505     qemu_co_mutex_lock(&s->lock);
1506     /* write zeroes could fail if sectors not aligned to cluster, test it with
1507      * dry_run == true before really updating image */
1508     ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true);
1509     if (!ret) {
1510         ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false);
1511     }
1512     qemu_co_mutex_unlock(&s->lock);
1513     return ret;
1514 }
1515 
1516 static int vmdk_create_extent(const char *filename, int64_t filesize,
1517                               bool flat, bool compress, bool zeroed_grain,
1518                               Error **errp)
1519 {
1520     int ret, i;
1521     BlockDriverState *bs = NULL;
1522     VMDK4Header header;
1523     Error *local_err;
1524     uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1525     uint32_t *gd_buf = NULL;
1526     int gd_buf_size;
1527 
1528     ret = bdrv_create_file(filename, NULL, &local_err);
1529     if (ret < 0) {
1530         error_propagate(errp, local_err);
1531         goto exit;
1532     }
1533 
1534     assert(bs == NULL);
1535     ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1536                     NULL, &local_err);
1537     if (ret < 0) {
1538         error_propagate(errp, local_err);
1539         goto exit;
1540     }
1541 
1542     if (flat) {
1543         ret = bdrv_truncate(bs, filesize);
1544         if (ret < 0) {
1545             error_setg_errno(errp, -ret, "Could not truncate file");
1546         }
1547         goto exit;
1548     }
1549     magic = cpu_to_be32(VMDK4_MAGIC);
1550     memset(&header, 0, sizeof(header));
1551     header.version = zeroed_grain ? 2 : 1;
1552     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1553                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1554                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1555     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1556     header.capacity = filesize / BDRV_SECTOR_SIZE;
1557     header.granularity = 128;
1558     header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1559 
1560     grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1561     gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1562                            BDRV_SECTOR_SIZE);
1563     gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1564     gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1565 
1566     header.desc_offset = 1;
1567     header.desc_size = 20;
1568     header.rgd_offset = header.desc_offset + header.desc_size;
1569     header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1570     header.grain_offset =
1571         ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1572                  header.granularity);
1573     /* swap endianness for all header fields */
1574     header.version = cpu_to_le32(header.version);
1575     header.flags = cpu_to_le32(header.flags);
1576     header.capacity = cpu_to_le64(header.capacity);
1577     header.granularity = cpu_to_le64(header.granularity);
1578     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1579     header.desc_offset = cpu_to_le64(header.desc_offset);
1580     header.desc_size = cpu_to_le64(header.desc_size);
1581     header.rgd_offset = cpu_to_le64(header.rgd_offset);
1582     header.gd_offset = cpu_to_le64(header.gd_offset);
1583     header.grain_offset = cpu_to_le64(header.grain_offset);
1584     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1585 
1586     header.check_bytes[0] = 0xa;
1587     header.check_bytes[1] = 0x20;
1588     header.check_bytes[2] = 0xd;
1589     header.check_bytes[3] = 0xa;
1590 
1591     /* write all the data */
1592     ret = bdrv_pwrite(bs, 0, &magic, sizeof(magic));
1593     if (ret < 0) {
1594         error_set(errp, QERR_IO_ERROR);
1595         goto exit;
1596     }
1597     ret = bdrv_pwrite(bs, sizeof(magic), &header, sizeof(header));
1598     if (ret < 0) {
1599         error_set(errp, QERR_IO_ERROR);
1600         goto exit;
1601     }
1602 
1603     ret = bdrv_truncate(bs, le64_to_cpu(header.grain_offset) << 9);
1604     if (ret < 0) {
1605         error_setg_errno(errp, -ret, "Could not truncate file");
1606         goto exit;
1607     }
1608 
1609     /* write grain directory */
1610     gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1611     gd_buf = g_malloc0(gd_buf_size);
1612     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1613          i < gt_count; i++, tmp += gt_size) {
1614         gd_buf[i] = cpu_to_le32(tmp);
1615     }
1616     ret = bdrv_pwrite(bs, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1617                       gd_buf, gd_buf_size);
1618     if (ret < 0) {
1619         error_set(errp, QERR_IO_ERROR);
1620         goto exit;
1621     }
1622 
1623     /* write backup grain directory */
1624     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1625          i < gt_count; i++, tmp += gt_size) {
1626         gd_buf[i] = cpu_to_le32(tmp);
1627     }
1628     ret = bdrv_pwrite(bs, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1629                       gd_buf, gd_buf_size);
1630     if (ret < 0) {
1631         error_set(errp, QERR_IO_ERROR);
1632         goto exit;
1633     }
1634 
1635     ret = 0;
1636 exit:
1637     if (bs) {
1638         bdrv_unref(bs);
1639     }
1640     g_free(gd_buf);
1641     return ret;
1642 }
1643 
1644 static int filename_decompose(const char *filename, char *path, char *prefix,
1645                               char *postfix, size_t buf_len, Error **errp)
1646 {
1647     const char *p, *q;
1648 
1649     if (filename == NULL || !strlen(filename)) {
1650         error_setg(errp, "No filename provided");
1651         return VMDK_ERROR;
1652     }
1653     p = strrchr(filename, '/');
1654     if (p == NULL) {
1655         p = strrchr(filename, '\\');
1656     }
1657     if (p == NULL) {
1658         p = strrchr(filename, ':');
1659     }
1660     if (p != NULL) {
1661         p++;
1662         if (p - filename >= buf_len) {
1663             return VMDK_ERROR;
1664         }
1665         pstrcpy(path, p - filename + 1, filename);
1666     } else {
1667         p = filename;
1668         path[0] = '\0';
1669     }
1670     q = strrchr(p, '.');
1671     if (q == NULL) {
1672         pstrcpy(prefix, buf_len, p);
1673         postfix[0] = '\0';
1674     } else {
1675         if (q - p >= buf_len) {
1676             return VMDK_ERROR;
1677         }
1678         pstrcpy(prefix, q - p + 1, p);
1679         pstrcpy(postfix, buf_len, q);
1680     }
1681     return VMDK_OK;
1682 }
1683 
1684 static int vmdk_create(const char *filename, QEMUOptionParameter *options,
1685                        Error **errp)
1686 {
1687     int idx = 0;
1688     BlockDriverState *new_bs = NULL;
1689     Error *local_err;
1690     char *desc = NULL;
1691     int64_t total_size = 0, filesize;
1692     const char *adapter_type = NULL;
1693     const char *backing_file = NULL;
1694     const char *fmt = NULL;
1695     int flags = 0;
1696     int ret = 0;
1697     bool flat, split, compress;
1698     GString *ext_desc_lines;
1699     char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX];
1700     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
1701     const char *desc_extent_line;
1702     char parent_desc_line[BUF_SIZE] = "";
1703     uint32_t parent_cid = 0xffffffff;
1704     uint32_t number_heads = 16;
1705     bool zeroed_grain = false;
1706     uint32_t desc_offset = 0, desc_len;
1707     const char desc_template[] =
1708         "# Disk DescriptorFile\n"
1709         "version=1\n"
1710         "CID=%x\n"
1711         "parentCID=%x\n"
1712         "createType=\"%s\"\n"
1713         "%s"
1714         "\n"
1715         "# Extent description\n"
1716         "%s"
1717         "\n"
1718         "# The Disk Data Base\n"
1719         "#DDB\n"
1720         "\n"
1721         "ddb.virtualHWVersion = \"%d\"\n"
1722         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1723         "ddb.geometry.heads = \"%d\"\n"
1724         "ddb.geometry.sectors = \"63\"\n"
1725         "ddb.adapterType = \"%s\"\n";
1726 
1727     ext_desc_lines = g_string_new(NULL);
1728 
1729     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
1730         ret = -EINVAL;
1731         goto exit;
1732     }
1733     /* Read out options */
1734     while (options && options->name) {
1735         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1736             total_size = options->value.n;
1737         } else if (!strcmp(options->name, BLOCK_OPT_ADAPTER_TYPE)) {
1738             adapter_type = options->value.s;
1739         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1740             backing_file = options->value.s;
1741         } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) {
1742             flags |= options->value.n ? BLOCK_FLAG_COMPAT6 : 0;
1743         } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) {
1744             fmt = options->value.s;
1745         } else if (!strcmp(options->name, BLOCK_OPT_ZEROED_GRAIN)) {
1746             zeroed_grain |= options->value.n;
1747         }
1748         options++;
1749     }
1750     if (!adapter_type) {
1751         adapter_type = "ide";
1752     } else if (strcmp(adapter_type, "ide") &&
1753                strcmp(adapter_type, "buslogic") &&
1754                strcmp(adapter_type, "lsilogic") &&
1755                strcmp(adapter_type, "legacyESX")) {
1756         error_setg(errp, "Unknown adapter type: '%s'", adapter_type);
1757         ret = -EINVAL;
1758         goto exit;
1759     }
1760     if (strcmp(adapter_type, "ide") != 0) {
1761         /* that's the number of heads with which vmware operates when
1762            creating, exporting, etc. vmdk files with a non-ide adapter type */
1763         number_heads = 255;
1764     }
1765     if (!fmt) {
1766         /* Default format to monolithicSparse */
1767         fmt = "monolithicSparse";
1768     } else if (strcmp(fmt, "monolithicFlat") &&
1769                strcmp(fmt, "monolithicSparse") &&
1770                strcmp(fmt, "twoGbMaxExtentSparse") &&
1771                strcmp(fmt, "twoGbMaxExtentFlat") &&
1772                strcmp(fmt, "streamOptimized")) {
1773         error_setg(errp, "Unknown subformat: '%s'", fmt);
1774         ret = -EINVAL;
1775         goto exit;
1776     }
1777     split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1778               strcmp(fmt, "twoGbMaxExtentSparse"));
1779     flat = !(strcmp(fmt, "monolithicFlat") &&
1780              strcmp(fmt, "twoGbMaxExtentFlat"));
1781     compress = !strcmp(fmt, "streamOptimized");
1782     if (flat) {
1783         desc_extent_line = "RW %lld FLAT \"%s\" 0\n";
1784     } else {
1785         desc_extent_line = "RW %lld SPARSE \"%s\"\n";
1786     }
1787     if (flat && backing_file) {
1788         error_setg(errp, "Flat image can't have backing file");
1789         ret = -ENOTSUP;
1790         goto exit;
1791     }
1792     if (flat && zeroed_grain) {
1793         error_setg(errp, "Flat image can't enable zeroed grain");
1794         ret = -ENOTSUP;
1795         goto exit;
1796     }
1797     if (backing_file) {
1798         BlockDriverState *bs = NULL;
1799         ret = bdrv_open(&bs, backing_file, NULL, NULL, BDRV_O_NO_BACKING, NULL,
1800                         errp);
1801         if (ret != 0) {
1802             goto exit;
1803         }
1804         if (strcmp(bs->drv->format_name, "vmdk")) {
1805             bdrv_unref(bs);
1806             ret = -EINVAL;
1807             goto exit;
1808         }
1809         parent_cid = vmdk_read_cid(bs, 0);
1810         bdrv_unref(bs);
1811         snprintf(parent_desc_line, sizeof(parent_desc_line),
1812                 "parentFileNameHint=\"%s\"", backing_file);
1813     }
1814 
1815     /* Create extents */
1816     filesize = total_size;
1817     while (filesize > 0) {
1818         char desc_line[BUF_SIZE];
1819         char ext_filename[PATH_MAX];
1820         char desc_filename[PATH_MAX];
1821         int64_t size = filesize;
1822 
1823         if (split && size > split_size) {
1824             size = split_size;
1825         }
1826         if (split) {
1827             snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s",
1828                     prefix, flat ? 'f' : 's', ++idx, postfix);
1829         } else if (flat) {
1830             snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s",
1831                     prefix, postfix);
1832         } else {
1833             snprintf(desc_filename, sizeof(desc_filename), "%s%s",
1834                     prefix, postfix);
1835         }
1836         snprintf(ext_filename, sizeof(ext_filename), "%s%s",
1837                 path, desc_filename);
1838 
1839         if (vmdk_create_extent(ext_filename, size,
1840                                flat, compress, zeroed_grain, errp)) {
1841             ret = -EINVAL;
1842             goto exit;
1843         }
1844         filesize -= size;
1845 
1846         /* Format description line */
1847         snprintf(desc_line, sizeof(desc_line),
1848                     desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename);
1849         g_string_append(ext_desc_lines, desc_line);
1850     }
1851     /* generate descriptor file */
1852     desc = g_strdup_printf(desc_template,
1853                            (unsigned int)time(NULL),
1854                            parent_cid,
1855                            fmt,
1856                            parent_desc_line,
1857                            ext_desc_lines->str,
1858                            (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
1859                            total_size /
1860                                (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
1861                            number_heads,
1862                            adapter_type);
1863     desc_len = strlen(desc);
1864     /* the descriptor offset = 0x200 */
1865     if (!split && !flat) {
1866         desc_offset = 0x200;
1867     } else {
1868         ret = bdrv_create_file(filename, options, &local_err);
1869         if (ret < 0) {
1870             error_setg_errno(errp, -ret, "Could not create image file");
1871             goto exit;
1872         }
1873     }
1874     assert(new_bs == NULL);
1875     ret = bdrv_open(&new_bs, filename, NULL, NULL,
1876                     BDRV_O_RDWR | BDRV_O_PROTOCOL, NULL, &local_err);
1877     if (ret < 0) {
1878         error_setg_errno(errp, -ret, "Could not write description");
1879         goto exit;
1880     }
1881     ret = bdrv_pwrite(new_bs, desc_offset, desc, desc_len);
1882     if (ret < 0) {
1883         error_setg_errno(errp, -ret, "Could not write description");
1884         goto exit;
1885     }
1886     /* bdrv_pwrite write padding zeros to align to sector, we don't need that
1887      * for description file */
1888     if (desc_offset == 0) {
1889         ret = bdrv_truncate(new_bs, desc_len);
1890         if (ret < 0) {
1891             error_setg_errno(errp, -ret, "Could not truncate file");
1892         }
1893     }
1894 exit:
1895     if (new_bs) {
1896         bdrv_unref(new_bs);
1897     }
1898     g_free(desc);
1899     g_string_free(ext_desc_lines, true);
1900     return ret;
1901 }
1902 
1903 static void vmdk_close(BlockDriverState *bs)
1904 {
1905     BDRVVmdkState *s = bs->opaque;
1906 
1907     vmdk_free_extents(bs);
1908     g_free(s->create_type);
1909 
1910     migrate_del_blocker(s->migration_blocker);
1911     error_free(s->migration_blocker);
1912 }
1913 
1914 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
1915 {
1916     BDRVVmdkState *s = bs->opaque;
1917     int i, err;
1918     int ret = 0;
1919 
1920     for (i = 0; i < s->num_extents; i++) {
1921         err = bdrv_co_flush(s->extents[i].file);
1922         if (err < 0) {
1923             ret = err;
1924         }
1925     }
1926     return ret;
1927 }
1928 
1929 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
1930 {
1931     int i;
1932     int64_t ret = 0;
1933     int64_t r;
1934     BDRVVmdkState *s = bs->opaque;
1935 
1936     ret = bdrv_get_allocated_file_size(bs->file);
1937     if (ret < 0) {
1938         return ret;
1939     }
1940     for (i = 0; i < s->num_extents; i++) {
1941         if (s->extents[i].file == bs->file) {
1942             continue;
1943         }
1944         r = bdrv_get_allocated_file_size(s->extents[i].file);
1945         if (r < 0) {
1946             return r;
1947         }
1948         ret += r;
1949     }
1950     return ret;
1951 }
1952 
1953 static int vmdk_has_zero_init(BlockDriverState *bs)
1954 {
1955     int i;
1956     BDRVVmdkState *s = bs->opaque;
1957 
1958     /* If has a flat extent and its underlying storage doesn't have zero init,
1959      * return 0. */
1960     for (i = 0; i < s->num_extents; i++) {
1961         if (s->extents[i].flat) {
1962             if (!bdrv_has_zero_init(s->extents[i].file)) {
1963                 return 0;
1964             }
1965         }
1966     }
1967     return 1;
1968 }
1969 
1970 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
1971 {
1972     ImageInfo *info = g_new0(ImageInfo, 1);
1973 
1974     *info = (ImageInfo){
1975         .filename         = g_strdup(extent->file->filename),
1976         .format           = g_strdup(extent->type),
1977         .virtual_size     = extent->sectors * BDRV_SECTOR_SIZE,
1978         .compressed       = extent->compressed,
1979         .has_compressed   = extent->compressed,
1980         .cluster_size     = extent->cluster_sectors * BDRV_SECTOR_SIZE,
1981         .has_cluster_size = !extent->flat,
1982     };
1983 
1984     return info;
1985 }
1986 
1987 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result,
1988                       BdrvCheckMode fix)
1989 {
1990     BDRVVmdkState *s = bs->opaque;
1991     VmdkExtent *extent = NULL;
1992     int64_t sector_num = 0;
1993     int64_t total_sectors = bdrv_getlength(bs) / BDRV_SECTOR_SIZE;
1994     int ret;
1995     uint64_t cluster_offset;
1996 
1997     if (fix) {
1998         return -ENOTSUP;
1999     }
2000 
2001     for (;;) {
2002         if (sector_num >= total_sectors) {
2003             return 0;
2004         }
2005         extent = find_extent(s, sector_num, extent);
2006         if (!extent) {
2007             fprintf(stderr,
2008                     "ERROR: could not find extent for sector %" PRId64 "\n",
2009                     sector_num);
2010             break;
2011         }
2012         ret = get_cluster_offset(bs, extent, NULL,
2013                                  sector_num << BDRV_SECTOR_BITS,
2014                                  0, &cluster_offset);
2015         if (ret == VMDK_ERROR) {
2016             fprintf(stderr,
2017                     "ERROR: could not get cluster_offset for sector %"
2018                     PRId64 "\n", sector_num);
2019             break;
2020         }
2021         if (ret == VMDK_OK && cluster_offset >= bdrv_getlength(extent->file)) {
2022             fprintf(stderr,
2023                     "ERROR: cluster offset for sector %"
2024                     PRId64 " points after EOF\n", sector_num);
2025             break;
2026         }
2027         sector_num += extent->cluster_sectors;
2028     }
2029 
2030     result->corruptions++;
2031     return 0;
2032 }
2033 
2034 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs)
2035 {
2036     int i;
2037     BDRVVmdkState *s = bs->opaque;
2038     ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2039     ImageInfoList **next;
2040 
2041     *spec_info = (ImageInfoSpecific){
2042         .kind = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2043         {
2044             .vmdk = g_new0(ImageInfoSpecificVmdk, 1),
2045         },
2046     };
2047 
2048     *spec_info->vmdk = (ImageInfoSpecificVmdk) {
2049         .create_type = g_strdup(s->create_type),
2050         .cid = s->cid,
2051         .parent_cid = s->parent_cid,
2052     };
2053 
2054     next = &spec_info->vmdk->extents;
2055     for (i = 0; i < s->num_extents; i++) {
2056         *next = g_new0(ImageInfoList, 1);
2057         (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2058         (*next)->next = NULL;
2059         next = &(*next)->next;
2060     }
2061 
2062     return spec_info;
2063 }
2064 
2065 static QEMUOptionParameter vmdk_create_options[] = {
2066     {
2067         .name = BLOCK_OPT_SIZE,
2068         .type = OPT_SIZE,
2069         .help = "Virtual disk size"
2070     },
2071     {
2072         .name = BLOCK_OPT_ADAPTER_TYPE,
2073         .type = OPT_STRING,
2074         .help = "Virtual adapter type, can be one of "
2075                 "ide (default), lsilogic, buslogic or legacyESX"
2076     },
2077     {
2078         .name = BLOCK_OPT_BACKING_FILE,
2079         .type = OPT_STRING,
2080         .help = "File name of a base image"
2081     },
2082     {
2083         .name = BLOCK_OPT_COMPAT6,
2084         .type = OPT_FLAG,
2085         .help = "VMDK version 6 image"
2086     },
2087     {
2088         .name = BLOCK_OPT_SUBFMT,
2089         .type = OPT_STRING,
2090         .help =
2091             "VMDK flat extent format, can be one of "
2092             "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2093     },
2094     {
2095         .name = BLOCK_OPT_ZEROED_GRAIN,
2096         .type = OPT_FLAG,
2097         .help = "Enable efficient zero writes using the zeroed-grain GTE feature"
2098     },
2099     { NULL }
2100 };
2101 
2102 static BlockDriver bdrv_vmdk = {
2103     .format_name                  = "vmdk",
2104     .instance_size                = sizeof(BDRVVmdkState),
2105     .bdrv_probe                   = vmdk_probe,
2106     .bdrv_open                    = vmdk_open,
2107     .bdrv_check                   = vmdk_check,
2108     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
2109     .bdrv_read                    = vmdk_co_read,
2110     .bdrv_write                   = vmdk_co_write,
2111     .bdrv_co_write_zeroes         = vmdk_co_write_zeroes,
2112     .bdrv_close                   = vmdk_close,
2113     .bdrv_create                  = vmdk_create,
2114     .bdrv_co_flush_to_disk        = vmdk_co_flush,
2115     .bdrv_co_get_block_status     = vmdk_co_get_block_status,
2116     .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2117     .bdrv_has_zero_init           = vmdk_has_zero_init,
2118     .bdrv_get_specific_info       = vmdk_get_specific_info,
2119     .bdrv_refresh_limits          = vmdk_refresh_limits,
2120 
2121     .create_options               = vmdk_create_options,
2122 };
2123 
2124 static void bdrv_vmdk_init(void)
2125 {
2126     bdrv_register(&bdrv_vmdk);
2127 }
2128 
2129 block_init(bdrv_vmdk_init);
2130