xref: /openbmc/qemu/block/vmdk.c (revision bc72ad67)
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     unsigned int cluster_sectors;
109 } VmdkExtent;
110 
111 typedef struct BDRVVmdkState {
112     CoMutex lock;
113     uint64_t desc_offset;
114     bool cid_updated;
115     uint32_t parent_cid;
116     int num_extents;
117     /* Extent array with num_extents entries, ascend ordered by address */
118     VmdkExtent *extents;
119     Error *migration_blocker;
120 } BDRVVmdkState;
121 
122 typedef struct VmdkMetaData {
123     uint32_t offset;
124     unsigned int l1_index;
125     unsigned int l2_index;
126     unsigned int l2_offset;
127     int valid;
128     uint32_t *l2_cache_entry;
129 } VmdkMetaData;
130 
131 typedef struct VmdkGrainMarker {
132     uint64_t lba;
133     uint32_t size;
134     uint8_t  data[0];
135 } QEMU_PACKED VmdkGrainMarker;
136 
137 enum {
138     MARKER_END_OF_STREAM    = 0,
139     MARKER_GRAIN_TABLE      = 1,
140     MARKER_GRAIN_DIRECTORY  = 2,
141     MARKER_FOOTER           = 3,
142 };
143 
144 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
145 {
146     uint32_t magic;
147 
148     if (buf_size < 4) {
149         return 0;
150     }
151     magic = be32_to_cpu(*(uint32_t *)buf);
152     if (magic == VMDK3_MAGIC ||
153         magic == VMDK4_MAGIC) {
154         return 100;
155     } else {
156         const char *p = (const char *)buf;
157         const char *end = p + buf_size;
158         while (p < end) {
159             if (*p == '#') {
160                 /* skip comment line */
161                 while (p < end && *p != '\n') {
162                     p++;
163                 }
164                 p++;
165                 continue;
166             }
167             if (*p == ' ') {
168                 while (p < end && *p == ' ') {
169                     p++;
170                 }
171                 /* skip '\r' if windows line endings used. */
172                 if (p < end && *p == '\r') {
173                     p++;
174                 }
175                 /* only accept blank lines before 'version=' line */
176                 if (p == end || *p != '\n') {
177                     return 0;
178                 }
179                 p++;
180                 continue;
181             }
182             if (end - p >= strlen("version=X\n")) {
183                 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
184                     strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
185                     return 100;
186                 }
187             }
188             if (end - p >= strlen("version=X\r\n")) {
189                 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
190                     strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
191                     return 100;
192                 }
193             }
194             return 0;
195         }
196         return 0;
197     }
198 }
199 
200 #define CHECK_CID 1
201 
202 #define SECTOR_SIZE 512
203 #define DESC_SIZE (20 * SECTOR_SIZE)    /* 20 sectors of 512 bytes each */
204 #define BUF_SIZE 4096
205 #define HEADER_SIZE 512                 /* first sector of 512 bytes */
206 
207 static void vmdk_free_extents(BlockDriverState *bs)
208 {
209     int i;
210     BDRVVmdkState *s = bs->opaque;
211     VmdkExtent *e;
212 
213     for (i = 0; i < s->num_extents; i++) {
214         e = &s->extents[i];
215         g_free(e->l1_table);
216         g_free(e->l2_cache);
217         g_free(e->l1_backup_table);
218         if (e->file != bs->file) {
219             bdrv_delete(e->file);
220         }
221     }
222     g_free(s->extents);
223 }
224 
225 static void vmdk_free_last_extent(BlockDriverState *bs)
226 {
227     BDRVVmdkState *s = bs->opaque;
228 
229     if (s->num_extents == 0) {
230         return;
231     }
232     s->num_extents--;
233     s->extents = g_realloc(s->extents, s->num_extents * sizeof(VmdkExtent));
234 }
235 
236 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
237 {
238     char desc[DESC_SIZE];
239     uint32_t cid = 0xffffffff;
240     const char *p_name, *cid_str;
241     size_t cid_str_size;
242     BDRVVmdkState *s = bs->opaque;
243     int ret;
244 
245     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
246     if (ret < 0) {
247         return 0;
248     }
249 
250     if (parent) {
251         cid_str = "parentCID";
252         cid_str_size = sizeof("parentCID");
253     } else {
254         cid_str = "CID";
255         cid_str_size = sizeof("CID");
256     }
257 
258     desc[DESC_SIZE - 1] = '\0';
259     p_name = strstr(desc, cid_str);
260     if (p_name != NULL) {
261         p_name += cid_str_size;
262         sscanf(p_name, "%x", &cid);
263     }
264 
265     return cid;
266 }
267 
268 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
269 {
270     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
271     char *p_name, *tmp_str;
272     BDRVVmdkState *s = bs->opaque;
273     int ret;
274 
275     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
276     if (ret < 0) {
277         return ret;
278     }
279 
280     desc[DESC_SIZE - 1] = '\0';
281     tmp_str = strstr(desc, "parentCID");
282     if (tmp_str == NULL) {
283         return -EINVAL;
284     }
285 
286     pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
287     p_name = strstr(desc, "CID");
288     if (p_name != NULL) {
289         p_name += sizeof("CID");
290         snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
291         pstrcat(desc, sizeof(desc), tmp_desc);
292     }
293 
294     ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
295     if (ret < 0) {
296         return ret;
297     }
298 
299     return 0;
300 }
301 
302 static int vmdk_is_cid_valid(BlockDriverState *bs)
303 {
304 #ifdef CHECK_CID
305     BDRVVmdkState *s = bs->opaque;
306     BlockDriverState *p_bs = bs->backing_hd;
307     uint32_t cur_pcid;
308 
309     if (p_bs) {
310         cur_pcid = vmdk_read_cid(p_bs, 0);
311         if (s->parent_cid != cur_pcid) {
312             /* CID not valid */
313             return 0;
314         }
315     }
316 #endif
317     /* CID valid */
318     return 1;
319 }
320 
321 /* Queue extents, if any, for reopen() */
322 static int vmdk_reopen_prepare(BDRVReopenState *state,
323                                BlockReopenQueue *queue, Error **errp)
324 {
325     BDRVVmdkState *s;
326     int ret = -1;
327     int i;
328     VmdkExtent *e;
329 
330     assert(state != NULL);
331     assert(state->bs != NULL);
332 
333     if (queue == NULL) {
334         error_set(errp, ERROR_CLASS_GENERIC_ERROR,
335                  "No reopen queue for VMDK extents");
336         goto exit;
337     }
338 
339     s = state->bs->opaque;
340 
341     assert(s != NULL);
342 
343     for (i = 0; i < s->num_extents; i++) {
344         e = &s->extents[i];
345         if (e->file != state->bs->file) {
346             bdrv_reopen_queue(queue, e->file, state->flags);
347         }
348     }
349     ret = 0;
350 
351 exit:
352     return ret;
353 }
354 
355 static int vmdk_parent_open(BlockDriverState *bs)
356 {
357     char *p_name;
358     char desc[DESC_SIZE + 1];
359     BDRVVmdkState *s = bs->opaque;
360     int ret;
361 
362     desc[DESC_SIZE] = '\0';
363     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
364     if (ret < 0) {
365         return ret;
366     }
367 
368     p_name = strstr(desc, "parentFileNameHint");
369     if (p_name != NULL) {
370         char *end_name;
371 
372         p_name += sizeof("parentFileNameHint") + 1;
373         end_name = strchr(p_name, '\"');
374         if (end_name == NULL) {
375             return -EINVAL;
376         }
377         if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
378             return -EINVAL;
379         }
380 
381         pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
382     }
383 
384     return 0;
385 }
386 
387 /* Create and append extent to the extent array. Return the added VmdkExtent
388  * address. return NULL if allocation failed. */
389 static int vmdk_add_extent(BlockDriverState *bs,
390                            BlockDriverState *file, bool flat, int64_t sectors,
391                            int64_t l1_offset, int64_t l1_backup_offset,
392                            uint32_t l1_size,
393                            int l2_size, uint64_t cluster_sectors,
394                            VmdkExtent **new_extent)
395 {
396     VmdkExtent *extent;
397     BDRVVmdkState *s = bs->opaque;
398 
399     if (cluster_sectors > 0x200000) {
400         /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
401         error_report("invalid granularity, image may be corrupt");
402         return -EINVAL;
403     }
404     if (l1_size > 512 * 1024 * 1024) {
405         /* Although with big capacity and small l1_entry_sectors, we can get a
406          * big l1_size, we don't want unbounded value to allocate the table.
407          * Limit it to 512M, which is 16PB for default cluster and L2 table
408          * size */
409         error_report("L1 size too big");
410         return -EFBIG;
411     }
412 
413     s->extents = g_realloc(s->extents,
414                               (s->num_extents + 1) * sizeof(VmdkExtent));
415     extent = &s->extents[s->num_extents];
416     s->num_extents++;
417 
418     memset(extent, 0, sizeof(VmdkExtent));
419     extent->file = file;
420     extent->flat = flat;
421     extent->sectors = sectors;
422     extent->l1_table_offset = l1_offset;
423     extent->l1_backup_table_offset = l1_backup_offset;
424     extent->l1_size = l1_size;
425     extent->l1_entry_sectors = l2_size * cluster_sectors;
426     extent->l2_size = l2_size;
427     extent->cluster_sectors = cluster_sectors;
428 
429     if (s->num_extents > 1) {
430         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
431     } else {
432         extent->end_sector = extent->sectors;
433     }
434     bs->total_sectors = extent->end_sector;
435     if (new_extent) {
436         *new_extent = extent;
437     }
438     return 0;
439 }
440 
441 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent)
442 {
443     int ret;
444     int l1_size, i;
445 
446     /* read the L1 table */
447     l1_size = extent->l1_size * sizeof(uint32_t);
448     extent->l1_table = g_malloc(l1_size);
449     ret = bdrv_pread(extent->file,
450                     extent->l1_table_offset,
451                     extent->l1_table,
452                     l1_size);
453     if (ret < 0) {
454         goto fail_l1;
455     }
456     for (i = 0; i < extent->l1_size; i++) {
457         le32_to_cpus(&extent->l1_table[i]);
458     }
459 
460     if (extent->l1_backup_table_offset) {
461         extent->l1_backup_table = g_malloc(l1_size);
462         ret = bdrv_pread(extent->file,
463                         extent->l1_backup_table_offset,
464                         extent->l1_backup_table,
465                         l1_size);
466         if (ret < 0) {
467             goto fail_l1b;
468         }
469         for (i = 0; i < extent->l1_size; i++) {
470             le32_to_cpus(&extent->l1_backup_table[i]);
471         }
472     }
473 
474     extent->l2_cache =
475         g_malloc(extent->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
476     return 0;
477  fail_l1b:
478     g_free(extent->l1_backup_table);
479  fail_l1:
480     g_free(extent->l1_table);
481     return ret;
482 }
483 
484 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
485                                  BlockDriverState *file,
486                                  int flags)
487 {
488     int ret;
489     uint32_t magic;
490     VMDK3Header header;
491     VmdkExtent *extent;
492 
493     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
494     if (ret < 0) {
495         return ret;
496     }
497     ret = vmdk_add_extent(bs, file, false,
498                           le32_to_cpu(header.disk_sectors),
499                           le32_to_cpu(header.l1dir_offset) << 9,
500                           0,
501                           le32_to_cpu(header.l1dir_size),
502                           4096,
503                           le32_to_cpu(header.granularity),
504                           &extent);
505     if (ret < 0) {
506         return ret;
507     }
508     ret = vmdk_init_tables(bs, extent);
509     if (ret) {
510         /* free extent allocated by vmdk_add_extent */
511         vmdk_free_last_extent(bs);
512     }
513     return ret;
514 }
515 
516 static int vmdk_open_desc_file(BlockDriverState *bs, int flags,
517                                uint64_t desc_offset);
518 
519 static int vmdk_open_vmdk4(BlockDriverState *bs,
520                            BlockDriverState *file,
521                            int flags)
522 {
523     int ret;
524     uint32_t magic;
525     uint32_t l1_size, l1_entry_sectors;
526     VMDK4Header header;
527     VmdkExtent *extent;
528     int64_t l1_backup_offset = 0;
529 
530     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
531     if (ret < 0) {
532         return ret;
533     }
534     if (header.capacity == 0) {
535         uint64_t desc_offset = le64_to_cpu(header.desc_offset);
536         if (desc_offset) {
537             return vmdk_open_desc_file(bs, flags, desc_offset << 9);
538         }
539     }
540 
541     if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
542         /*
543          * The footer takes precedence over the header, so read it in. The
544          * footer starts at offset -1024 from the end: One sector for the
545          * footer, and another one for the end-of-stream marker.
546          */
547         struct {
548             struct {
549                 uint64_t val;
550                 uint32_t size;
551                 uint32_t type;
552                 uint8_t pad[512 - 16];
553             } QEMU_PACKED footer_marker;
554 
555             uint32_t magic;
556             VMDK4Header header;
557             uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
558 
559             struct {
560                 uint64_t val;
561                 uint32_t size;
562                 uint32_t type;
563                 uint8_t pad[512 - 16];
564             } QEMU_PACKED eos_marker;
565         } QEMU_PACKED footer;
566 
567         ret = bdrv_pread(file,
568             bs->file->total_sectors * 512 - 1536,
569             &footer, sizeof(footer));
570         if (ret < 0) {
571             return ret;
572         }
573 
574         /* Some sanity checks for the footer */
575         if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
576             le32_to_cpu(footer.footer_marker.size) != 0  ||
577             le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
578             le64_to_cpu(footer.eos_marker.val) != 0  ||
579             le32_to_cpu(footer.eos_marker.size) != 0  ||
580             le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
581         {
582             return -EINVAL;
583         }
584 
585         header = footer.header;
586     }
587 
588     if (le32_to_cpu(header.version) >= 3) {
589         char buf[64];
590         snprintf(buf, sizeof(buf), "VMDK version %d",
591                  le32_to_cpu(header.version));
592         qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
593                 bs->device_name, "vmdk", buf);
594         return -ENOTSUP;
595     }
596 
597     if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
598         error_report("L2 table size too big");
599         return -EINVAL;
600     }
601 
602     l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
603                         * le64_to_cpu(header.granularity);
604     if (l1_entry_sectors == 0) {
605         return -EINVAL;
606     }
607     l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
608                 / l1_entry_sectors;
609     if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
610         l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
611     }
612     ret = vmdk_add_extent(bs, file, false,
613                           le64_to_cpu(header.capacity),
614                           le64_to_cpu(header.gd_offset) << 9,
615                           l1_backup_offset,
616                           l1_size,
617                           le32_to_cpu(header.num_gtes_per_gt),
618                           le64_to_cpu(header.granularity),
619                           &extent);
620     if (ret < 0) {
621         return ret;
622     }
623     extent->compressed =
624         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
625     extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
626     extent->version = le32_to_cpu(header.version);
627     extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
628     ret = vmdk_init_tables(bs, extent);
629     if (ret) {
630         /* free extent allocated by vmdk_add_extent */
631         vmdk_free_last_extent(bs);
632     }
633     return ret;
634 }
635 
636 /* find an option value out of descriptor file */
637 static int vmdk_parse_description(const char *desc, const char *opt_name,
638         char *buf, int buf_size)
639 {
640     char *opt_pos, *opt_end;
641     const char *end = desc + strlen(desc);
642 
643     opt_pos = strstr(desc, opt_name);
644     if (!opt_pos) {
645         return VMDK_ERROR;
646     }
647     /* Skip "=\"" following opt_name */
648     opt_pos += strlen(opt_name) + 2;
649     if (opt_pos >= end) {
650         return VMDK_ERROR;
651     }
652     opt_end = opt_pos;
653     while (opt_end < end && *opt_end != '"') {
654         opt_end++;
655     }
656     if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
657         return VMDK_ERROR;
658     }
659     pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
660     return VMDK_OK;
661 }
662 
663 /* Open an extent file and append to bs array */
664 static int vmdk_open_sparse(BlockDriverState *bs,
665                             BlockDriverState *file,
666                             int flags)
667 {
668     uint32_t magic;
669 
670     if (bdrv_pread(file, 0, &magic, sizeof(magic)) != sizeof(magic)) {
671         return -EIO;
672     }
673 
674     magic = be32_to_cpu(magic);
675     switch (magic) {
676         case VMDK3_MAGIC:
677             return vmdk_open_vmfs_sparse(bs, file, flags);
678             break;
679         case VMDK4_MAGIC:
680             return vmdk_open_vmdk4(bs, file, flags);
681             break;
682         default:
683             return -EMEDIUMTYPE;
684             break;
685     }
686 }
687 
688 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
689         const char *desc_file_path)
690 {
691     int ret;
692     char access[11];
693     char type[11];
694     char fname[512];
695     const char *p = desc;
696     int64_t sectors = 0;
697     int64_t flat_offset;
698     char extent_path[PATH_MAX];
699     BlockDriverState *extent_file;
700 
701     while (*p) {
702         /* parse extent line:
703          * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
704          * or
705          * RW [size in sectors] SPARSE "file-name.vmdk"
706          */
707         flat_offset = -1;
708         ret = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
709                 access, &sectors, type, fname, &flat_offset);
710         if (ret < 4 || strcmp(access, "RW")) {
711             goto next_line;
712         } else if (!strcmp(type, "FLAT")) {
713             if (ret != 5 || flat_offset < 0) {
714                 return -EINVAL;
715             }
716         } else if (ret != 4) {
717             return -EINVAL;
718         }
719 
720         if (sectors <= 0 ||
721             (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
722              strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
723             (strcmp(access, "RW"))) {
724             goto next_line;
725         }
726 
727         path_combine(extent_path, sizeof(extent_path),
728                 desc_file_path, fname);
729         ret = bdrv_file_open(&extent_file, extent_path, NULL, bs->open_flags);
730         if (ret) {
731             return ret;
732         }
733 
734         /* save to extents array */
735         if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
736             /* FLAT extent */
737             VmdkExtent *extent;
738 
739             ret = vmdk_add_extent(bs, extent_file, true, sectors,
740                             0, 0, 0, 0, sectors, &extent);
741             if (ret < 0) {
742                 return ret;
743             }
744             extent->flat_start_offset = flat_offset << 9;
745         } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
746             /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
747             ret = vmdk_open_sparse(bs, extent_file, bs->open_flags);
748             if (ret) {
749                 bdrv_delete(extent_file);
750                 return ret;
751             }
752         } else {
753             fprintf(stderr,
754                 "VMDK: Not supported extent type \"%s\""".\n", type);
755             return -ENOTSUP;
756         }
757 next_line:
758         /* move to next line */
759         while (*p && *p != '\n') {
760             p++;
761         }
762         p++;
763     }
764     return 0;
765 }
766 
767 static int vmdk_open_desc_file(BlockDriverState *bs, int flags,
768                                uint64_t desc_offset)
769 {
770     int ret;
771     char *buf = NULL;
772     char ct[128];
773     BDRVVmdkState *s = bs->opaque;
774     int64_t size;
775 
776     size = bdrv_getlength(bs->file);
777     if (size < 0) {
778         return -EINVAL;
779     }
780 
781     size = MIN(size, 1 << 20);  /* avoid unbounded allocation */
782     buf = g_malloc0(size + 1);
783 
784     ret = bdrv_pread(bs->file, desc_offset, buf, size);
785     if (ret < 0) {
786         goto exit;
787     }
788     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
789         ret = -EMEDIUMTYPE;
790         goto exit;
791     }
792     if (strcmp(ct, "monolithicFlat") &&
793         strcmp(ct, "vmfs") &&
794         strcmp(ct, "vmfsSparse") &&
795         strcmp(ct, "twoGbMaxExtentSparse") &&
796         strcmp(ct, "twoGbMaxExtentFlat")) {
797         fprintf(stderr,
798                 "VMDK: Not supported image type \"%s\""".\n", ct);
799         ret = -ENOTSUP;
800         goto exit;
801     }
802     s->desc_offset = 0;
803     ret = vmdk_parse_extents(buf, bs, bs->file->filename);
804 exit:
805     g_free(buf);
806     return ret;
807 }
808 
809 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags)
810 {
811     int ret;
812     BDRVVmdkState *s = bs->opaque;
813 
814     if (vmdk_open_sparse(bs, bs->file, flags) == 0) {
815         s->desc_offset = 0x200;
816     } else {
817         ret = vmdk_open_desc_file(bs, flags, 0);
818         if (ret) {
819             goto fail;
820         }
821     }
822     /* try to open parent images, if exist */
823     ret = vmdk_parent_open(bs);
824     if (ret) {
825         goto fail;
826     }
827     s->parent_cid = vmdk_read_cid(bs, 1);
828     qemu_co_mutex_init(&s->lock);
829 
830     /* Disable migration when VMDK images are used */
831     error_set(&s->migration_blocker,
832               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
833               "vmdk", bs->device_name, "live migration");
834     migrate_add_blocker(s->migration_blocker);
835 
836     return 0;
837 
838 fail:
839     vmdk_free_extents(bs);
840     return ret;
841 }
842 
843 static int get_whole_cluster(BlockDriverState *bs,
844                 VmdkExtent *extent,
845                 uint64_t cluster_offset,
846                 uint64_t offset,
847                 bool allocate)
848 {
849     int ret = VMDK_OK;
850     uint8_t *whole_grain = NULL;
851 
852     /* we will be here if it's first write on non-exist grain(cluster).
853      * try to read from parent image, if exist */
854     if (bs->backing_hd) {
855         whole_grain =
856             qemu_blockalign(bs, extent->cluster_sectors << BDRV_SECTOR_BITS);
857         if (!vmdk_is_cid_valid(bs)) {
858             ret = VMDK_ERROR;
859             goto exit;
860         }
861 
862         /* floor offset to cluster */
863         offset -= offset % (extent->cluster_sectors * 512);
864         ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain,
865                 extent->cluster_sectors);
866         if (ret < 0) {
867             ret = VMDK_ERROR;
868             goto exit;
869         }
870 
871         /* Write grain only into the active image */
872         ret = bdrv_write(extent->file, cluster_offset, whole_grain,
873                 extent->cluster_sectors);
874         if (ret < 0) {
875             ret = VMDK_ERROR;
876             goto exit;
877         }
878     }
879 exit:
880     qemu_vfree(whole_grain);
881     return ret;
882 }
883 
884 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data)
885 {
886     uint32_t offset;
887     QEMU_BUILD_BUG_ON(sizeof(offset) != sizeof(m_data->offset));
888     offset = cpu_to_le32(m_data->offset);
889     /* update L2 table */
890     if (bdrv_pwrite_sync(
891                 extent->file,
892                 ((int64_t)m_data->l2_offset * 512)
893                     + (m_data->l2_index * sizeof(m_data->offset)),
894                 &offset, sizeof(offset)) < 0) {
895         return VMDK_ERROR;
896     }
897     /* update backup L2 table */
898     if (extent->l1_backup_table_offset != 0) {
899         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
900         if (bdrv_pwrite_sync(
901                     extent->file,
902                     ((int64_t)m_data->l2_offset * 512)
903                         + (m_data->l2_index * sizeof(m_data->offset)),
904                     &offset, sizeof(offset)) < 0) {
905             return VMDK_ERROR;
906         }
907     }
908     if (m_data->l2_cache_entry) {
909         *m_data->l2_cache_entry = offset;
910     }
911 
912     return VMDK_OK;
913 }
914 
915 static int get_cluster_offset(BlockDriverState *bs,
916                                     VmdkExtent *extent,
917                                     VmdkMetaData *m_data,
918                                     uint64_t offset,
919                                     int allocate,
920                                     uint64_t *cluster_offset)
921 {
922     unsigned int l1_index, l2_offset, l2_index;
923     int min_index, i, j;
924     uint32_t min_count, *l2_table;
925     bool zeroed = false;
926 
927     if (m_data) {
928         m_data->valid = 0;
929     }
930     if (extent->flat) {
931         *cluster_offset = extent->flat_start_offset;
932         return VMDK_OK;
933     }
934 
935     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
936     l1_index = (offset >> 9) / extent->l1_entry_sectors;
937     if (l1_index >= extent->l1_size) {
938         return VMDK_ERROR;
939     }
940     l2_offset = extent->l1_table[l1_index];
941     if (!l2_offset) {
942         return VMDK_UNALLOC;
943     }
944     for (i = 0; i < L2_CACHE_SIZE; i++) {
945         if (l2_offset == extent->l2_cache_offsets[i]) {
946             /* increment the hit count */
947             if (++extent->l2_cache_counts[i] == 0xffffffff) {
948                 for (j = 0; j < L2_CACHE_SIZE; j++) {
949                     extent->l2_cache_counts[j] >>= 1;
950                 }
951             }
952             l2_table = extent->l2_cache + (i * extent->l2_size);
953             goto found;
954         }
955     }
956     /* not found: load a new entry in the least used one */
957     min_index = 0;
958     min_count = 0xffffffff;
959     for (i = 0; i < L2_CACHE_SIZE; i++) {
960         if (extent->l2_cache_counts[i] < min_count) {
961             min_count = extent->l2_cache_counts[i];
962             min_index = i;
963         }
964     }
965     l2_table = extent->l2_cache + (min_index * extent->l2_size);
966     if (bdrv_pread(
967                 extent->file,
968                 (int64_t)l2_offset * 512,
969                 l2_table,
970                 extent->l2_size * sizeof(uint32_t)
971             ) != extent->l2_size * sizeof(uint32_t)) {
972         return VMDK_ERROR;
973     }
974 
975     extent->l2_cache_offsets[min_index] = l2_offset;
976     extent->l2_cache_counts[min_index] = 1;
977  found:
978     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
979     *cluster_offset = le32_to_cpu(l2_table[l2_index]);
980 
981     if (m_data) {
982         m_data->valid = 1;
983         m_data->l1_index = l1_index;
984         m_data->l2_index = l2_index;
985         m_data->offset = *cluster_offset;
986         m_data->l2_offset = l2_offset;
987         m_data->l2_cache_entry = &l2_table[l2_index];
988     }
989     if (extent->has_zero_grain && *cluster_offset == VMDK_GTE_ZEROED) {
990         zeroed = true;
991     }
992 
993     if (!*cluster_offset || zeroed) {
994         if (!allocate) {
995             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
996         }
997 
998         /* Avoid the L2 tables update for the images that have snapshots. */
999         *cluster_offset = bdrv_getlength(extent->file);
1000         if (!extent->compressed) {
1001             bdrv_truncate(
1002                 extent->file,
1003                 *cluster_offset + (extent->cluster_sectors << 9)
1004             );
1005         }
1006 
1007         *cluster_offset >>= 9;
1008         l2_table[l2_index] = cpu_to_le32(*cluster_offset);
1009 
1010         /* First of all we write grain itself, to avoid race condition
1011          * that may to corrupt the image.
1012          * This problem may occur because of insufficient space on host disk
1013          * or inappropriate VM shutdown.
1014          */
1015         if (get_whole_cluster(
1016                 bs, extent, *cluster_offset, offset, allocate) == -1) {
1017             return VMDK_ERROR;
1018         }
1019 
1020         if (m_data) {
1021             m_data->offset = *cluster_offset;
1022         }
1023     }
1024     *cluster_offset <<= 9;
1025     return VMDK_OK;
1026 }
1027 
1028 static VmdkExtent *find_extent(BDRVVmdkState *s,
1029                                 int64_t sector_num, VmdkExtent *start_hint)
1030 {
1031     VmdkExtent *extent = start_hint;
1032 
1033     if (!extent) {
1034         extent = &s->extents[0];
1035     }
1036     while (extent < &s->extents[s->num_extents]) {
1037         if (sector_num < extent->end_sector) {
1038             return extent;
1039         }
1040         extent++;
1041     }
1042     return NULL;
1043 }
1044 
1045 static int coroutine_fn vmdk_co_is_allocated(BlockDriverState *bs,
1046         int64_t sector_num, int nb_sectors, int *pnum)
1047 {
1048     BDRVVmdkState *s = bs->opaque;
1049     int64_t index_in_cluster, n, ret;
1050     uint64_t offset;
1051     VmdkExtent *extent;
1052 
1053     extent = find_extent(s, sector_num, NULL);
1054     if (!extent) {
1055         return 0;
1056     }
1057     qemu_co_mutex_lock(&s->lock);
1058     ret = get_cluster_offset(bs, extent, NULL,
1059                             sector_num * 512, 0, &offset);
1060     qemu_co_mutex_unlock(&s->lock);
1061 
1062     ret = (ret == VMDK_OK || ret == VMDK_ZEROED);
1063 
1064     index_in_cluster = sector_num % extent->cluster_sectors;
1065     n = extent->cluster_sectors - index_in_cluster;
1066     if (n > nb_sectors) {
1067         n = nb_sectors;
1068     }
1069     *pnum = n;
1070     return ret;
1071 }
1072 
1073 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1074                             int64_t offset_in_cluster, const uint8_t *buf,
1075                             int nb_sectors, int64_t sector_num)
1076 {
1077     int ret;
1078     VmdkGrainMarker *data = NULL;
1079     uLongf buf_len;
1080     const uint8_t *write_buf = buf;
1081     int write_len = nb_sectors * 512;
1082 
1083     if (extent->compressed) {
1084         if (!extent->has_marker) {
1085             ret = -EINVAL;
1086             goto out;
1087         }
1088         buf_len = (extent->cluster_sectors << 9) * 2;
1089         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1090         if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK ||
1091                 buf_len == 0) {
1092             ret = -EINVAL;
1093             goto out;
1094         }
1095         data->lba = sector_num;
1096         data->size = buf_len;
1097         write_buf = (uint8_t *)data;
1098         write_len = buf_len + sizeof(VmdkGrainMarker);
1099     }
1100     ret = bdrv_pwrite(extent->file,
1101                         cluster_offset + offset_in_cluster,
1102                         write_buf,
1103                         write_len);
1104     if (ret != write_len) {
1105         ret = ret < 0 ? ret : -EIO;
1106         goto out;
1107     }
1108     ret = 0;
1109  out:
1110     g_free(data);
1111     return ret;
1112 }
1113 
1114 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1115                             int64_t offset_in_cluster, uint8_t *buf,
1116                             int nb_sectors)
1117 {
1118     int ret;
1119     int cluster_bytes, buf_bytes;
1120     uint8_t *cluster_buf, *compressed_data;
1121     uint8_t *uncomp_buf;
1122     uint32_t data_len;
1123     VmdkGrainMarker *marker;
1124     uLongf buf_len;
1125 
1126 
1127     if (!extent->compressed) {
1128         ret = bdrv_pread(extent->file,
1129                           cluster_offset + offset_in_cluster,
1130                           buf, nb_sectors * 512);
1131         if (ret == nb_sectors * 512) {
1132             return 0;
1133         } else {
1134             return -EIO;
1135         }
1136     }
1137     cluster_bytes = extent->cluster_sectors * 512;
1138     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1139     buf_bytes = cluster_bytes * 2;
1140     cluster_buf = g_malloc(buf_bytes);
1141     uncomp_buf = g_malloc(cluster_bytes);
1142     ret = bdrv_pread(extent->file,
1143                 cluster_offset,
1144                 cluster_buf, buf_bytes);
1145     if (ret < 0) {
1146         goto out;
1147     }
1148     compressed_data = cluster_buf;
1149     buf_len = cluster_bytes;
1150     data_len = cluster_bytes;
1151     if (extent->has_marker) {
1152         marker = (VmdkGrainMarker *)cluster_buf;
1153         compressed_data = marker->data;
1154         data_len = le32_to_cpu(marker->size);
1155     }
1156     if (!data_len || data_len > buf_bytes) {
1157         ret = -EINVAL;
1158         goto out;
1159     }
1160     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1161     if (ret != Z_OK) {
1162         ret = -EINVAL;
1163         goto out;
1164 
1165     }
1166     if (offset_in_cluster < 0 ||
1167             offset_in_cluster + nb_sectors * 512 > buf_len) {
1168         ret = -EINVAL;
1169         goto out;
1170     }
1171     memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512);
1172     ret = 0;
1173 
1174  out:
1175     g_free(uncomp_buf);
1176     g_free(cluster_buf);
1177     return ret;
1178 }
1179 
1180 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
1181                     uint8_t *buf, int nb_sectors)
1182 {
1183     BDRVVmdkState *s = bs->opaque;
1184     int ret;
1185     uint64_t n, index_in_cluster;
1186     uint64_t extent_begin_sector, extent_relative_sector_num;
1187     VmdkExtent *extent = NULL;
1188     uint64_t cluster_offset;
1189 
1190     while (nb_sectors > 0) {
1191         extent = find_extent(s, sector_num, extent);
1192         if (!extent) {
1193             return -EIO;
1194         }
1195         ret = get_cluster_offset(
1196                             bs, extent, NULL,
1197                             sector_num << 9, 0, &cluster_offset);
1198         extent_begin_sector = extent->end_sector - extent->sectors;
1199         extent_relative_sector_num = sector_num - extent_begin_sector;
1200         index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1201         n = extent->cluster_sectors - index_in_cluster;
1202         if (n > nb_sectors) {
1203             n = nb_sectors;
1204         }
1205         if (ret != VMDK_OK) {
1206             /* if not allocated, try to read from parent image, if exist */
1207             if (bs->backing_hd && ret != VMDK_ZEROED) {
1208                 if (!vmdk_is_cid_valid(bs)) {
1209                     return -EINVAL;
1210                 }
1211                 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
1212                 if (ret < 0) {
1213                     return ret;
1214                 }
1215             } else {
1216                 memset(buf, 0, 512 * n);
1217             }
1218         } else {
1219             ret = vmdk_read_extent(extent,
1220                             cluster_offset, index_in_cluster * 512,
1221                             buf, n);
1222             if (ret) {
1223                 return ret;
1224             }
1225         }
1226         nb_sectors -= n;
1227         sector_num += n;
1228         buf += n * 512;
1229     }
1230     return 0;
1231 }
1232 
1233 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num,
1234                                      uint8_t *buf, int nb_sectors)
1235 {
1236     int ret;
1237     BDRVVmdkState *s = bs->opaque;
1238     qemu_co_mutex_lock(&s->lock);
1239     ret = vmdk_read(bs, sector_num, buf, nb_sectors);
1240     qemu_co_mutex_unlock(&s->lock);
1241     return ret;
1242 }
1243 
1244 /**
1245  * vmdk_write:
1246  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
1247  *                if possible, otherwise return -ENOTSUP.
1248  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1249  *                with each cluster. By dry run we can find if the zero write
1250  *                is possible without modifying image data.
1251  *
1252  * Returns: error code with 0 for success.
1253  */
1254 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
1255                       const uint8_t *buf, int nb_sectors,
1256                       bool zeroed, bool zero_dry_run)
1257 {
1258     BDRVVmdkState *s = bs->opaque;
1259     VmdkExtent *extent = NULL;
1260     int n, ret;
1261     int64_t index_in_cluster;
1262     uint64_t extent_begin_sector, extent_relative_sector_num;
1263     uint64_t cluster_offset;
1264     VmdkMetaData m_data;
1265 
1266     if (sector_num > bs->total_sectors) {
1267         fprintf(stderr,
1268                 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
1269                 " total_sectors=0x%" PRIx64 "\n",
1270                 sector_num, bs->total_sectors);
1271         return -EIO;
1272     }
1273 
1274     while (nb_sectors > 0) {
1275         extent = find_extent(s, sector_num, extent);
1276         if (!extent) {
1277             return -EIO;
1278         }
1279         ret = get_cluster_offset(
1280                                 bs,
1281                                 extent,
1282                                 &m_data,
1283                                 sector_num << 9, !extent->compressed,
1284                                 &cluster_offset);
1285         if (extent->compressed) {
1286             if (ret == VMDK_OK) {
1287                 /* Refuse write to allocated cluster for streamOptimized */
1288                 fprintf(stderr,
1289                         "VMDK: can't write to allocated cluster"
1290                         " for streamOptimized\n");
1291                 return -EIO;
1292             } else {
1293                 /* allocate */
1294                 ret = get_cluster_offset(
1295                                         bs,
1296                                         extent,
1297                                         &m_data,
1298                                         sector_num << 9, 1,
1299                                         &cluster_offset);
1300             }
1301         }
1302         if (ret == VMDK_ERROR) {
1303             return -EINVAL;
1304         }
1305         extent_begin_sector = extent->end_sector - extent->sectors;
1306         extent_relative_sector_num = sector_num - extent_begin_sector;
1307         index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1308         n = extent->cluster_sectors - index_in_cluster;
1309         if (n > nb_sectors) {
1310             n = nb_sectors;
1311         }
1312         if (zeroed) {
1313             /* Do zeroed write, buf is ignored */
1314             if (extent->has_zero_grain &&
1315                     index_in_cluster == 0 &&
1316                     n >= extent->cluster_sectors) {
1317                 n = extent->cluster_sectors;
1318                 if (!zero_dry_run) {
1319                     m_data.offset = VMDK_GTE_ZEROED;
1320                     /* update L2 tables */
1321                     if (vmdk_L2update(extent, &m_data) != VMDK_OK) {
1322                         return -EIO;
1323                     }
1324                 }
1325             } else {
1326                 return -ENOTSUP;
1327             }
1328         } else {
1329             ret = vmdk_write_extent(extent,
1330                             cluster_offset, index_in_cluster * 512,
1331                             buf, n, sector_num);
1332             if (ret) {
1333                 return ret;
1334             }
1335             if (m_data.valid) {
1336                 /* update L2 tables */
1337                 if (vmdk_L2update(extent, &m_data) != VMDK_OK) {
1338                     return -EIO;
1339                 }
1340             }
1341         }
1342         nb_sectors -= n;
1343         sector_num += n;
1344         buf += n * 512;
1345 
1346         /* update CID on the first write every time the virtual disk is
1347          * opened */
1348         if (!s->cid_updated) {
1349             ret = vmdk_write_cid(bs, time(NULL));
1350             if (ret < 0) {
1351                 return ret;
1352             }
1353             s->cid_updated = true;
1354         }
1355     }
1356     return 0;
1357 }
1358 
1359 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num,
1360                                       const uint8_t *buf, int nb_sectors)
1361 {
1362     int ret;
1363     BDRVVmdkState *s = bs->opaque;
1364     qemu_co_mutex_lock(&s->lock);
1365     ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false);
1366     qemu_co_mutex_unlock(&s->lock);
1367     return ret;
1368 }
1369 
1370 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs,
1371                                              int64_t sector_num,
1372                                              int nb_sectors)
1373 {
1374     int ret;
1375     BDRVVmdkState *s = bs->opaque;
1376     qemu_co_mutex_lock(&s->lock);
1377     /* write zeroes could fail if sectors not aligned to cluster, test it with
1378      * dry_run == true before really updating image */
1379     ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true);
1380     if (!ret) {
1381         ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false);
1382     }
1383     qemu_co_mutex_unlock(&s->lock);
1384     return ret;
1385 }
1386 
1387 static int vmdk_create_extent(const char *filename, int64_t filesize,
1388                               bool flat, bool compress, bool zeroed_grain)
1389 {
1390     int ret, i;
1391     int fd = 0;
1392     VMDK4Header header;
1393     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
1394 
1395     fd = qemu_open(filename,
1396                    O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
1397                    0644);
1398     if (fd < 0) {
1399         return -errno;
1400     }
1401     if (flat) {
1402         ret = ftruncate(fd, filesize);
1403         if (ret < 0) {
1404             ret = -errno;
1405         }
1406         goto exit;
1407     }
1408     magic = cpu_to_be32(VMDK4_MAGIC);
1409     memset(&header, 0, sizeof(header));
1410     header.version = zeroed_grain ? 2 : 1;
1411     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1412                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1413                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1414     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1415     header.capacity = filesize / 512;
1416     header.granularity = 128;
1417     header.num_gtes_per_gt = 512;
1418 
1419     grains = (filesize / 512 + header.granularity - 1) / header.granularity;
1420     gt_size = ((header.num_gtes_per_gt * sizeof(uint32_t)) + 511) >> 9;
1421     gt_count =
1422         (grains + header.num_gtes_per_gt - 1) / header.num_gtes_per_gt;
1423     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
1424 
1425     header.desc_offset = 1;
1426     header.desc_size = 20;
1427     header.rgd_offset = header.desc_offset + header.desc_size;
1428     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
1429     header.grain_offset =
1430        ((header.gd_offset + gd_size + (gt_size * gt_count) +
1431          header.granularity - 1) / header.granularity) *
1432         header.granularity;
1433     /* swap endianness for all header fields */
1434     header.version = cpu_to_le32(header.version);
1435     header.flags = cpu_to_le32(header.flags);
1436     header.capacity = cpu_to_le64(header.capacity);
1437     header.granularity = cpu_to_le64(header.granularity);
1438     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1439     header.desc_offset = cpu_to_le64(header.desc_offset);
1440     header.desc_size = cpu_to_le64(header.desc_size);
1441     header.rgd_offset = cpu_to_le64(header.rgd_offset);
1442     header.gd_offset = cpu_to_le64(header.gd_offset);
1443     header.grain_offset = cpu_to_le64(header.grain_offset);
1444     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1445 
1446     header.check_bytes[0] = 0xa;
1447     header.check_bytes[1] = 0x20;
1448     header.check_bytes[2] = 0xd;
1449     header.check_bytes[3] = 0xa;
1450 
1451     /* write all the data */
1452     ret = qemu_write_full(fd, &magic, sizeof(magic));
1453     if (ret != sizeof(magic)) {
1454         ret = -errno;
1455         goto exit;
1456     }
1457     ret = qemu_write_full(fd, &header, sizeof(header));
1458     if (ret != sizeof(header)) {
1459         ret = -errno;
1460         goto exit;
1461     }
1462 
1463     ret = ftruncate(fd, le64_to_cpu(header.grain_offset) << 9);
1464     if (ret < 0) {
1465         ret = -errno;
1466         goto exit;
1467     }
1468 
1469     /* write grain directory */
1470     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
1471     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_size;
1472          i < gt_count; i++, tmp += gt_size) {
1473         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1474         if (ret != sizeof(tmp)) {
1475             ret = -errno;
1476             goto exit;
1477         }
1478     }
1479 
1480     /* write backup grain directory */
1481     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
1482     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_size;
1483          i < gt_count; i++, tmp += gt_size) {
1484         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1485         if (ret != sizeof(tmp)) {
1486             ret = -errno;
1487             goto exit;
1488         }
1489     }
1490 
1491     ret = 0;
1492  exit:
1493     qemu_close(fd);
1494     return ret;
1495 }
1496 
1497 static int filename_decompose(const char *filename, char *path, char *prefix,
1498         char *postfix, size_t buf_len)
1499 {
1500     const char *p, *q;
1501 
1502     if (filename == NULL || !strlen(filename)) {
1503         fprintf(stderr, "Vmdk: no filename provided.\n");
1504         return VMDK_ERROR;
1505     }
1506     p = strrchr(filename, '/');
1507     if (p == NULL) {
1508         p = strrchr(filename, '\\');
1509     }
1510     if (p == NULL) {
1511         p = strrchr(filename, ':');
1512     }
1513     if (p != NULL) {
1514         p++;
1515         if (p - filename >= buf_len) {
1516             return VMDK_ERROR;
1517         }
1518         pstrcpy(path, p - filename + 1, filename);
1519     } else {
1520         p = filename;
1521         path[0] = '\0';
1522     }
1523     q = strrchr(p, '.');
1524     if (q == NULL) {
1525         pstrcpy(prefix, buf_len, p);
1526         postfix[0] = '\0';
1527     } else {
1528         if (q - p >= buf_len) {
1529             return VMDK_ERROR;
1530         }
1531         pstrcpy(prefix, q - p + 1, p);
1532         pstrcpy(postfix, buf_len, q);
1533     }
1534     return VMDK_OK;
1535 }
1536 
1537 static int vmdk_create(const char *filename, QEMUOptionParameter *options)
1538 {
1539     int fd, idx = 0;
1540     char desc[BUF_SIZE];
1541     int64_t total_size = 0, filesize;
1542     const char *adapter_type = NULL;
1543     const char *backing_file = NULL;
1544     const char *fmt = NULL;
1545     int flags = 0;
1546     int ret = 0;
1547     bool flat, split, compress;
1548     char ext_desc_lines[BUF_SIZE] = "";
1549     char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX];
1550     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
1551     const char *desc_extent_line;
1552     char parent_desc_line[BUF_SIZE] = "";
1553     uint32_t parent_cid = 0xffffffff;
1554     uint32_t number_heads = 16;
1555     bool zeroed_grain = false;
1556     const char desc_template[] =
1557         "# Disk DescriptorFile\n"
1558         "version=1\n"
1559         "CID=%x\n"
1560         "parentCID=%x\n"
1561         "createType=\"%s\"\n"
1562         "%s"
1563         "\n"
1564         "# Extent description\n"
1565         "%s"
1566         "\n"
1567         "# The Disk Data Base\n"
1568         "#DDB\n"
1569         "\n"
1570         "ddb.virtualHWVersion = \"%d\"\n"
1571         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1572         "ddb.geometry.heads = \"%d\"\n"
1573         "ddb.geometry.sectors = \"63\"\n"
1574         "ddb.adapterType = \"%s\"\n";
1575 
1576     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX)) {
1577         return -EINVAL;
1578     }
1579     /* Read out options */
1580     while (options && options->name) {
1581         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1582             total_size = options->value.n;
1583         } else if (!strcmp(options->name, BLOCK_OPT_ADAPTER_TYPE)) {
1584             adapter_type = options->value.s;
1585         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1586             backing_file = options->value.s;
1587         } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) {
1588             flags |= options->value.n ? BLOCK_FLAG_COMPAT6 : 0;
1589         } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) {
1590             fmt = options->value.s;
1591         } else if (!strcmp(options->name, BLOCK_OPT_ZEROED_GRAIN)) {
1592             zeroed_grain |= options->value.n;
1593         }
1594         options++;
1595     }
1596     if (!adapter_type) {
1597         adapter_type = "ide";
1598     } else if (strcmp(adapter_type, "ide") &&
1599                strcmp(adapter_type, "buslogic") &&
1600                strcmp(adapter_type, "lsilogic") &&
1601                strcmp(adapter_type, "legacyESX")) {
1602         fprintf(stderr, "VMDK: Unknown adapter type: '%s'.\n", adapter_type);
1603         return -EINVAL;
1604     }
1605     if (strcmp(adapter_type, "ide") != 0) {
1606         /* that's the number of heads with which vmware operates when
1607            creating, exporting, etc. vmdk files with a non-ide adapter type */
1608         number_heads = 255;
1609     }
1610     if (!fmt) {
1611         /* Default format to monolithicSparse */
1612         fmt = "monolithicSparse";
1613     } else if (strcmp(fmt, "monolithicFlat") &&
1614                strcmp(fmt, "monolithicSparse") &&
1615                strcmp(fmt, "twoGbMaxExtentSparse") &&
1616                strcmp(fmt, "twoGbMaxExtentFlat") &&
1617                strcmp(fmt, "streamOptimized")) {
1618         fprintf(stderr, "VMDK: Unknown subformat: %s\n", fmt);
1619         return -EINVAL;
1620     }
1621     split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1622               strcmp(fmt, "twoGbMaxExtentSparse"));
1623     flat = !(strcmp(fmt, "monolithicFlat") &&
1624              strcmp(fmt, "twoGbMaxExtentFlat"));
1625     compress = !strcmp(fmt, "streamOptimized");
1626     if (flat) {
1627         desc_extent_line = "RW %lld FLAT \"%s\" 0\n";
1628     } else {
1629         desc_extent_line = "RW %lld SPARSE \"%s\"\n";
1630     }
1631     if (flat && backing_file) {
1632         /* not supporting backing file for flat image */
1633         return -ENOTSUP;
1634     }
1635     if (backing_file) {
1636         BlockDriverState *bs = bdrv_new("");
1637         ret = bdrv_open(bs, backing_file, NULL, 0, NULL);
1638         if (ret != 0) {
1639             bdrv_delete(bs);
1640             return ret;
1641         }
1642         if (strcmp(bs->drv->format_name, "vmdk")) {
1643             bdrv_delete(bs);
1644             return -EINVAL;
1645         }
1646         parent_cid = vmdk_read_cid(bs, 0);
1647         bdrv_delete(bs);
1648         snprintf(parent_desc_line, sizeof(parent_desc_line),
1649                 "parentFileNameHint=\"%s\"", backing_file);
1650     }
1651 
1652     /* Create extents */
1653     filesize = total_size;
1654     while (filesize > 0) {
1655         char desc_line[BUF_SIZE];
1656         char ext_filename[PATH_MAX];
1657         char desc_filename[PATH_MAX];
1658         int64_t size = filesize;
1659 
1660         if (split && size > split_size) {
1661             size = split_size;
1662         }
1663         if (split) {
1664             snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s",
1665                     prefix, flat ? 'f' : 's', ++idx, postfix);
1666         } else if (flat) {
1667             snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s",
1668                     prefix, postfix);
1669         } else {
1670             snprintf(desc_filename, sizeof(desc_filename), "%s%s",
1671                     prefix, postfix);
1672         }
1673         snprintf(ext_filename, sizeof(ext_filename), "%s%s",
1674                 path, desc_filename);
1675 
1676         if (vmdk_create_extent(ext_filename, size,
1677                                flat, compress, zeroed_grain)) {
1678             return -EINVAL;
1679         }
1680         filesize -= size;
1681 
1682         /* Format description line */
1683         snprintf(desc_line, sizeof(desc_line),
1684                     desc_extent_line, size / 512, desc_filename);
1685         pstrcat(ext_desc_lines, sizeof(ext_desc_lines), desc_line);
1686     }
1687     /* generate descriptor file */
1688     snprintf(desc, sizeof(desc), desc_template,
1689             (unsigned int)time(NULL),
1690             parent_cid,
1691             fmt,
1692             parent_desc_line,
1693             ext_desc_lines,
1694             (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
1695             total_size / (int64_t)(63 * number_heads * 512), number_heads,
1696                 adapter_type);
1697     if (split || flat) {
1698         fd = qemu_open(filename,
1699                        O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
1700                        0644);
1701     } else {
1702         fd = qemu_open(filename,
1703                        O_WRONLY | O_BINARY | O_LARGEFILE,
1704                        0644);
1705     }
1706     if (fd < 0) {
1707         return -errno;
1708     }
1709     /* the descriptor offset = 0x200 */
1710     if (!split && !flat && 0x200 != lseek(fd, 0x200, SEEK_SET)) {
1711         ret = -errno;
1712         goto exit;
1713     }
1714     ret = qemu_write_full(fd, desc, strlen(desc));
1715     if (ret != strlen(desc)) {
1716         ret = -errno;
1717         goto exit;
1718     }
1719     ret = 0;
1720 exit:
1721     qemu_close(fd);
1722     return ret;
1723 }
1724 
1725 static void vmdk_close(BlockDriverState *bs)
1726 {
1727     BDRVVmdkState *s = bs->opaque;
1728 
1729     vmdk_free_extents(bs);
1730 
1731     migrate_del_blocker(s->migration_blocker);
1732     error_free(s->migration_blocker);
1733 }
1734 
1735 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
1736 {
1737     BDRVVmdkState *s = bs->opaque;
1738     int i, err;
1739     int ret = 0;
1740 
1741     for (i = 0; i < s->num_extents; i++) {
1742         err = bdrv_co_flush(s->extents[i].file);
1743         if (err < 0) {
1744             ret = err;
1745         }
1746     }
1747     return ret;
1748 }
1749 
1750 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
1751 {
1752     int i;
1753     int64_t ret = 0;
1754     int64_t r;
1755     BDRVVmdkState *s = bs->opaque;
1756 
1757     ret = bdrv_get_allocated_file_size(bs->file);
1758     if (ret < 0) {
1759         return ret;
1760     }
1761     for (i = 0; i < s->num_extents; i++) {
1762         if (s->extents[i].file == bs->file) {
1763             continue;
1764         }
1765         r = bdrv_get_allocated_file_size(s->extents[i].file);
1766         if (r < 0) {
1767             return r;
1768         }
1769         ret += r;
1770     }
1771     return ret;
1772 }
1773 
1774 static int vmdk_has_zero_init(BlockDriverState *bs)
1775 {
1776     int i;
1777     BDRVVmdkState *s = bs->opaque;
1778 
1779     /* If has a flat extent and its underlying storage doesn't have zero init,
1780      * return 0. */
1781     for (i = 0; i < s->num_extents; i++) {
1782         if (s->extents[i].flat) {
1783             if (!bdrv_has_zero_init(s->extents[i].file)) {
1784                 return 0;
1785             }
1786         }
1787     }
1788     return 1;
1789 }
1790 
1791 static QEMUOptionParameter vmdk_create_options[] = {
1792     {
1793         .name = BLOCK_OPT_SIZE,
1794         .type = OPT_SIZE,
1795         .help = "Virtual disk size"
1796     },
1797     {
1798         .name = BLOCK_OPT_ADAPTER_TYPE,
1799         .type = OPT_STRING,
1800         .help = "Virtual adapter type, can be one of "
1801                 "ide (default), lsilogic, buslogic or legacyESX"
1802     },
1803     {
1804         .name = BLOCK_OPT_BACKING_FILE,
1805         .type = OPT_STRING,
1806         .help = "File name of a base image"
1807     },
1808     {
1809         .name = BLOCK_OPT_COMPAT6,
1810         .type = OPT_FLAG,
1811         .help = "VMDK version 6 image"
1812     },
1813     {
1814         .name = BLOCK_OPT_SUBFMT,
1815         .type = OPT_STRING,
1816         .help =
1817             "VMDK flat extent format, can be one of "
1818             "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
1819     },
1820     {
1821         .name = BLOCK_OPT_ZEROED_GRAIN,
1822         .type = OPT_FLAG,
1823         .help = "Enable efficient zero writes using the zeroed-grain GTE feature"
1824     },
1825     { NULL }
1826 };
1827 
1828 static BlockDriver bdrv_vmdk = {
1829     .format_name                  = "vmdk",
1830     .instance_size                = sizeof(BDRVVmdkState),
1831     .bdrv_probe                   = vmdk_probe,
1832     .bdrv_open                    = vmdk_open,
1833     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
1834     .bdrv_read                    = vmdk_co_read,
1835     .bdrv_write                   = vmdk_co_write,
1836     .bdrv_co_write_zeroes         = vmdk_co_write_zeroes,
1837     .bdrv_close                   = vmdk_close,
1838     .bdrv_create                  = vmdk_create,
1839     .bdrv_co_flush_to_disk        = vmdk_co_flush,
1840     .bdrv_co_is_allocated         = vmdk_co_is_allocated,
1841     .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
1842     .bdrv_has_zero_init           = vmdk_has_zero_init,
1843 
1844     .create_options               = vmdk_create_options,
1845 };
1846 
1847 static void bdrv_vmdk_init(void)
1848 {
1849     bdrv_register(&bdrv_vmdk);
1850 }
1851 
1852 block_init(bdrv_vmdk_init);
1853