xref: /openbmc/qemu/block/vmdk.c (revision d6e9098e10e82feeddb824d7c3d0cf61aff96c29)
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_int.h"
28 #include "module.h"
29 
30 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
31 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
32 
33 typedef struct {
34     uint32_t version;
35     uint32_t flags;
36     uint32_t disk_sectors;
37     uint32_t granularity;
38     uint32_t l1dir_offset;
39     uint32_t l1dir_size;
40     uint32_t file_sectors;
41     uint32_t cylinders;
42     uint32_t heads;
43     uint32_t sectors_per_track;
44 } VMDK3Header;
45 
46 typedef struct {
47     uint32_t version;
48     uint32_t flags;
49     int64_t capacity;
50     int64_t granularity;
51     int64_t desc_offset;
52     int64_t desc_size;
53     int32_t num_gtes_per_gte;
54     int64_t rgd_offset;
55     int64_t gd_offset;
56     int64_t grain_offset;
57     char filler[1];
58     char check_bytes[4];
59 } __attribute__((packed)) VMDK4Header;
60 
61 #define L2_CACHE_SIZE 16
62 
63 typedef struct BDRVVmdkState {
64     BlockDriverState *hd;
65     int64_t l1_table_offset;
66     int64_t l1_backup_table_offset;
67     uint32_t *l1_table;
68     uint32_t *l1_backup_table;
69     unsigned int l1_size;
70     uint32_t l1_entry_sectors;
71 
72     unsigned int l2_size;
73     uint32_t *l2_cache;
74     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
75     uint32_t l2_cache_counts[L2_CACHE_SIZE];
76 
77     unsigned int cluster_sectors;
78     uint32_t parent_cid;
79     int is_parent;
80 } BDRVVmdkState;
81 
82 typedef struct VmdkMetaData {
83     uint32_t offset;
84     unsigned int l1_index;
85     unsigned int l2_index;
86     unsigned int l2_offset;
87     int valid;
88 } VmdkMetaData;
89 
90 typedef struct ActiveBDRVState{
91     BlockDriverState *hd;            // active image handler
92     uint64_t cluster_offset;         // current write offset
93 }ActiveBDRVState;
94 
95 static ActiveBDRVState activeBDRV;
96 
97 
98 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
99 {
100     uint32_t magic;
101 
102     if (buf_size < 4)
103         return 0;
104     magic = be32_to_cpu(*(uint32_t *)buf);
105     if (magic == VMDK3_MAGIC ||
106         magic == VMDK4_MAGIC)
107         return 100;
108     else
109         return 0;
110 }
111 
112 #define CHECK_CID 1
113 
114 #define SECTOR_SIZE 512
115 #define DESC_SIZE 20*SECTOR_SIZE	// 20 sectors of 512 bytes each
116 #define HEADER_SIZE 512   			// first sector of 512 bytes
117 
118 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
119 {
120     BDRVVmdkState *s = bs->opaque;
121     char desc[DESC_SIZE];
122     uint32_t cid;
123     const char *p_name, *cid_str;
124     size_t cid_str_size;
125 
126     /* the descriptor offset = 0x200 */
127     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
128         return 0;
129 
130     if (parent) {
131         cid_str = "parentCID";
132         cid_str_size = sizeof("parentCID");
133     } else {
134         cid_str = "CID";
135         cid_str_size = sizeof("CID");
136     }
137 
138     if ((p_name = strstr(desc,cid_str)) != NULL) {
139         p_name += cid_str_size;
140         sscanf(p_name,"%x",&cid);
141     }
142 
143     return cid;
144 }
145 
146 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
147 {
148     BDRVVmdkState *s = bs->opaque;
149     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
150     char *p_name, *tmp_str;
151 
152     /* the descriptor offset = 0x200 */
153     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
154         return -1;
155 
156     tmp_str = strstr(desc,"parentCID");
157     pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
158     if ((p_name = strstr(desc,"CID")) != NULL) {
159         p_name += sizeof("CID");
160         snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
161         pstrcat(desc, sizeof(desc), tmp_desc);
162     }
163 
164     if (bdrv_pwrite(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
165         return -1;
166     return 0;
167 }
168 
169 static int vmdk_is_cid_valid(BlockDriverState *bs)
170 {
171 #ifdef CHECK_CID
172     BDRVVmdkState *s = bs->opaque;
173     BlockDriverState *p_bs = bs->backing_hd;
174     uint32_t cur_pcid;
175 
176     if (p_bs) {
177         cur_pcid = vmdk_read_cid(p_bs,0);
178         if (s->parent_cid != cur_pcid)
179             // CID not valid
180             return 0;
181     }
182 #endif
183     // CID valid
184     return 1;
185 }
186 
187 static int vmdk_snapshot_create(const char *filename, const char *backing_file)
188 {
189     int snp_fd, p_fd;
190     int ret;
191     uint32_t p_cid;
192     char *p_name, *gd_buf, *rgd_buf;
193     const char *real_filename, *temp_str;
194     VMDK4Header header;
195     uint32_t gde_entries, gd_size;
196     int64_t gd_offset, rgd_offset, capacity, gt_size;
197     char p_desc[DESC_SIZE], s_desc[DESC_SIZE], hdr[HEADER_SIZE];
198     static const char desc_template[] =
199     "# Disk DescriptorFile\n"
200     "version=1\n"
201     "CID=%x\n"
202     "parentCID=%x\n"
203     "createType=\"monolithicSparse\"\n"
204     "parentFileNameHint=\"%s\"\n"
205     "\n"
206     "# Extent description\n"
207     "RW %u SPARSE \"%s\"\n"
208     "\n"
209     "# The Disk Data Base \n"
210     "#DDB\n"
211     "\n";
212 
213     snp_fd = open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 0644);
214     if (snp_fd < 0)
215         return -errno;
216     p_fd = open(backing_file, O_RDONLY | O_BINARY | O_LARGEFILE);
217     if (p_fd < 0) {
218         close(snp_fd);
219         return -errno;
220     }
221 
222     /* read the header */
223     if (lseek(p_fd, 0x0, SEEK_SET) == -1) {
224         ret = -errno;
225         goto fail;
226     }
227     if (read(p_fd, hdr, HEADER_SIZE) != HEADER_SIZE) {
228         ret = -errno;
229         goto fail;
230     }
231 
232     /* write the header */
233     if (lseek(snp_fd, 0x0, SEEK_SET) == -1) {
234         ret = -errno;
235         goto fail;
236     }
237     if (write(snp_fd, hdr, HEADER_SIZE) == -1) {
238         ret = -errno;
239         goto fail;
240     }
241 
242     memset(&header, 0, sizeof(header));
243     memcpy(&header,&hdr[4], sizeof(header)); // skip the VMDK4_MAGIC
244 
245     if (ftruncate(snp_fd, header.grain_offset << 9)) {
246         ret = -errno;
247         goto fail;
248     }
249     /* the descriptor offset = 0x200 */
250     if (lseek(p_fd, 0x200, SEEK_SET) == -1) {
251         ret = -errno;
252         goto fail;
253     }
254     if (read(p_fd, p_desc, DESC_SIZE) != DESC_SIZE) {
255         ret = -errno;
256         goto fail;
257     }
258 
259     if ((p_name = strstr(p_desc,"CID")) != NULL) {
260         p_name += sizeof("CID");
261         sscanf(p_name,"%x",&p_cid);
262     }
263 
264     real_filename = filename;
265     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
266         real_filename = temp_str + 1;
267     if ((temp_str = strrchr(real_filename, '/')) != NULL)
268         real_filename = temp_str + 1;
269     if ((temp_str = strrchr(real_filename, ':')) != NULL)
270         real_filename = temp_str + 1;
271 
272     snprintf(s_desc, sizeof(s_desc), desc_template, p_cid, p_cid, backing_file,
273              (uint32_t)header.capacity, real_filename);
274 
275     /* write the descriptor */
276     if (lseek(snp_fd, 0x200, SEEK_SET) == -1) {
277         ret = -errno;
278         goto fail;
279     }
280     if (write(snp_fd, s_desc, strlen(s_desc)) == -1) {
281         ret = -errno;
282         goto fail;
283     }
284 
285     gd_offset = header.gd_offset * SECTOR_SIZE;     // offset of GD table
286     rgd_offset = header.rgd_offset * SECTOR_SIZE;   // offset of RGD table
287     capacity = header.capacity * SECTOR_SIZE;       // Extent size
288     /*
289      * Each GDE span 32M disk, means:
290      * 512 GTE per GT, each GTE points to grain
291      */
292     gt_size = (int64_t)header.num_gtes_per_gte * header.granularity * SECTOR_SIZE;
293     if (!gt_size) {
294         ret = -EINVAL;
295         goto fail;
296     }
297     gde_entries = (uint32_t)(capacity / gt_size);  // number of gde/rgde
298     gd_size = gde_entries * sizeof(uint32_t);
299 
300     /* write RGD */
301     rgd_buf = qemu_malloc(gd_size);
302     if (lseek(p_fd, rgd_offset, SEEK_SET) == -1) {
303         ret = -errno;
304         goto fail_rgd;
305     }
306     if (read(p_fd, rgd_buf, gd_size) != gd_size) {
307         ret = -errno;
308         goto fail_rgd;
309     }
310     if (lseek(snp_fd, rgd_offset, SEEK_SET) == -1) {
311         ret = -errno;
312         goto fail_rgd;
313     }
314     if (write(snp_fd, rgd_buf, gd_size) == -1) {
315         ret = -errno;
316         goto fail_rgd;
317     }
318 
319     /* write GD */
320     gd_buf = qemu_malloc(gd_size);
321     if (lseek(p_fd, gd_offset, SEEK_SET) == -1) {
322         ret = -errno;
323         goto fail_gd;
324     }
325     if (read(p_fd, gd_buf, gd_size) != gd_size) {
326         ret = -errno;
327         goto fail_gd;
328     }
329     if (lseek(snp_fd, gd_offset, SEEK_SET) == -1) {
330         ret = -errno;
331         goto fail_gd;
332     }
333     if (write(snp_fd, gd_buf, gd_size) == -1) {
334         ret = -errno;
335         goto fail_gd;
336     }
337     ret = 0;
338 
339 fail_gd:
340     qemu_free(gd_buf);
341 fail_rgd:
342     qemu_free(rgd_buf);
343 fail:
344     close(p_fd);
345     close(snp_fd);
346     return ret;
347 }
348 
349 static void vmdk_parent_close(BlockDriverState *bs)
350 {
351     if (bs->backing_hd)
352         bdrv_close(bs->backing_hd);
353 }
354 
355 static int parent_open = 0;
356 static int vmdk_parent_open(BlockDriverState *bs, const char * filename)
357 {
358     BDRVVmdkState *s = bs->opaque;
359     char *p_name;
360     char desc[DESC_SIZE];
361     char parent_img_name[1024];
362 
363     /* the descriptor offset = 0x200 */
364     if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
365         return -1;
366 
367     if ((p_name = strstr(desc,"parentFileNameHint")) != NULL) {
368         char *end_name;
369         struct stat file_buf;
370 
371         p_name += sizeof("parentFileNameHint") + 1;
372         if ((end_name = strchr(p_name,'\"')) == NULL)
373             return -1;
374         if ((end_name - p_name) > sizeof (bs->backing_file) - 1)
375             return -1;
376 
377         pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
378         if (stat(bs->backing_file, &file_buf) != 0) {
379             path_combine(parent_img_name, sizeof(parent_img_name),
380                          filename, bs->backing_file);
381         } else {
382             pstrcpy(parent_img_name, sizeof(parent_img_name),
383                     bs->backing_file);
384         }
385 
386         bs->backing_hd = bdrv_new("");
387         if (!bs->backing_hd) {
388             failure:
389             bdrv_close(s->hd);
390             return -1;
391         }
392         parent_open = 1;
393         if (bdrv_open(bs->backing_hd, parent_img_name, 0, NULL) < 0)
394             goto failure;
395         parent_open = 0;
396     }
397 
398     return 0;
399 }
400 
401 static int vmdk_open(BlockDriverState *bs, const char *filename, int flags)
402 {
403     BDRVVmdkState *s = bs->opaque;
404     uint32_t magic;
405     int l1_size, i, ret;
406 
407     if (parent_open) {
408         /* Parent must be opened as RO, no RDWR. */
409         flags = 0;
410     }
411 
412     ret = bdrv_file_open(&s->hd, filename, flags);
413     if (ret < 0)
414         return ret;
415     if (bdrv_pread(s->hd, 0, &magic, sizeof(magic)) != sizeof(magic))
416         goto fail;
417 
418     magic = be32_to_cpu(magic);
419     if (magic == VMDK3_MAGIC) {
420         VMDK3Header header;
421 
422         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
423             goto fail;
424         s->cluster_sectors = le32_to_cpu(header.granularity);
425         s->l2_size = 1 << 9;
426         s->l1_size = 1 << 6;
427         bs->total_sectors = le32_to_cpu(header.disk_sectors);
428         s->l1_table_offset = le32_to_cpu(header.l1dir_offset) << 9;
429         s->l1_backup_table_offset = 0;
430         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
431     } else if (magic == VMDK4_MAGIC) {
432         VMDK4Header header;
433 
434         if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
435             goto fail;
436         bs->total_sectors = le64_to_cpu(header.capacity);
437         s->cluster_sectors = le64_to_cpu(header.granularity);
438         s->l2_size = le32_to_cpu(header.num_gtes_per_gte);
439         s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
440         if (s->l1_entry_sectors <= 0)
441             goto fail;
442         s->l1_size = (bs->total_sectors + s->l1_entry_sectors - 1)
443             / s->l1_entry_sectors;
444         s->l1_table_offset = le64_to_cpu(header.rgd_offset) << 9;
445         s->l1_backup_table_offset = le64_to_cpu(header.gd_offset) << 9;
446 
447         if (parent_open)
448             s->is_parent = 1;
449         else
450             s->is_parent = 0;
451 
452         // try to open parent images, if exist
453         if (vmdk_parent_open(bs, filename) != 0)
454             goto fail;
455         // write the CID once after the image creation
456         s->parent_cid = vmdk_read_cid(bs,1);
457     } else {
458         goto fail;
459     }
460 
461     /* read the L1 table */
462     l1_size = s->l1_size * sizeof(uint32_t);
463     s->l1_table = qemu_malloc(l1_size);
464     if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, l1_size) != l1_size)
465         goto fail;
466     for(i = 0; i < s->l1_size; i++) {
467         le32_to_cpus(&s->l1_table[i]);
468     }
469 
470     if (s->l1_backup_table_offset) {
471         s->l1_backup_table = qemu_malloc(l1_size);
472         if (bdrv_pread(s->hd, s->l1_backup_table_offset, s->l1_backup_table, l1_size) != l1_size)
473             goto fail;
474         for(i = 0; i < s->l1_size; i++) {
475             le32_to_cpus(&s->l1_backup_table[i]);
476         }
477     }
478 
479     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
480     return 0;
481  fail:
482     qemu_free(s->l1_backup_table);
483     qemu_free(s->l1_table);
484     qemu_free(s->l2_cache);
485     bdrv_delete(s->hd);
486     return -1;
487 }
488 
489 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
490                                    uint64_t offset, int allocate);
491 
492 static int get_whole_cluster(BlockDriverState *bs, uint64_t cluster_offset,
493                              uint64_t offset, int allocate)
494 {
495     uint64_t parent_cluster_offset;
496     BDRVVmdkState *s = bs->opaque;
497     uint8_t  whole_grain[s->cluster_sectors*512];        // 128 sectors * 512 bytes each = grain size 64KB
498 
499     // we will be here if it's first write on non-exist grain(cluster).
500     // try to read from parent image, if exist
501     if (bs->backing_hd) {
502         BDRVVmdkState *ps = bs->backing_hd->opaque;
503 
504         if (!vmdk_is_cid_valid(bs))
505             return -1;
506 
507         parent_cluster_offset = get_cluster_offset(bs->backing_hd, NULL,
508             offset, allocate);
509 
510         if (parent_cluster_offset) {
511             BDRVVmdkState *act_s = activeBDRV.hd->opaque;
512 
513             if (bdrv_pread(ps->hd, parent_cluster_offset, whole_grain, ps->cluster_sectors*512) != ps->cluster_sectors*512)
514                 return -1;
515 
516             //Write grain only into the active image
517             if (bdrv_pwrite(act_s->hd, activeBDRV.cluster_offset << 9, whole_grain, sizeof(whole_grain)) != sizeof(whole_grain))
518                 return -1;
519         }
520     }
521     return 0;
522 }
523 
524 static int vmdk_L2update(BlockDriverState *bs, VmdkMetaData *m_data)
525 {
526     BDRVVmdkState *s = bs->opaque;
527 
528     /* update L2 table */
529     if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
530                     &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
531         return -1;
532     /* update backup L2 table */
533     if (s->l1_backup_table_offset != 0) {
534         m_data->l2_offset = s->l1_backup_table[m_data->l1_index];
535         if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
536                         &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
537             return -1;
538     }
539 
540     return 0;
541 }
542 
543 static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
544                                    uint64_t offset, int allocate)
545 {
546     BDRVVmdkState *s = bs->opaque;
547     unsigned int l1_index, l2_offset, l2_index;
548     int min_index, i, j;
549     uint32_t min_count, *l2_table, tmp = 0;
550     uint64_t cluster_offset;
551 
552     if (m_data)
553         m_data->valid = 0;
554 
555     l1_index = (offset >> 9) / s->l1_entry_sectors;
556     if (l1_index >= s->l1_size)
557         return 0;
558     l2_offset = s->l1_table[l1_index];
559     if (!l2_offset)
560         return 0;
561     for(i = 0; i < L2_CACHE_SIZE; i++) {
562         if (l2_offset == s->l2_cache_offsets[i]) {
563             /* increment the hit count */
564             if (++s->l2_cache_counts[i] == 0xffffffff) {
565                 for(j = 0; j < L2_CACHE_SIZE; j++) {
566                     s->l2_cache_counts[j] >>= 1;
567                 }
568             }
569             l2_table = s->l2_cache + (i * s->l2_size);
570             goto found;
571         }
572     }
573     /* not found: load a new entry in the least used one */
574     min_index = 0;
575     min_count = 0xffffffff;
576     for(i = 0; i < L2_CACHE_SIZE; i++) {
577         if (s->l2_cache_counts[i] < min_count) {
578             min_count = s->l2_cache_counts[i];
579             min_index = i;
580         }
581     }
582     l2_table = s->l2_cache + (min_index * s->l2_size);
583     if (bdrv_pread(s->hd, (int64_t)l2_offset * 512, l2_table, s->l2_size * sizeof(uint32_t)) !=
584                                                                         s->l2_size * sizeof(uint32_t))
585         return 0;
586 
587     s->l2_cache_offsets[min_index] = l2_offset;
588     s->l2_cache_counts[min_index] = 1;
589  found:
590     l2_index = ((offset >> 9) / s->cluster_sectors) % s->l2_size;
591     cluster_offset = le32_to_cpu(l2_table[l2_index]);
592 
593     if (!cluster_offset) {
594         if (!allocate)
595             return 0;
596         // Avoid the L2 tables update for the images that have snapshots.
597         if (!s->is_parent) {
598             cluster_offset = bdrv_getlength(s->hd);
599             bdrv_truncate(s->hd, cluster_offset + (s->cluster_sectors << 9));
600 
601             cluster_offset >>= 9;
602             tmp = cpu_to_le32(cluster_offset);
603             l2_table[l2_index] = tmp;
604             // Save the active image state
605             activeBDRV.cluster_offset = cluster_offset;
606             activeBDRV.hd = bs;
607         }
608         /* First of all we write grain itself, to avoid race condition
609          * that may to corrupt the image.
610          * This problem may occur because of insufficient space on host disk
611          * or inappropriate VM shutdown.
612          */
613         if (get_whole_cluster(bs, cluster_offset, offset, allocate) == -1)
614             return 0;
615 
616         if (m_data) {
617             m_data->offset = tmp;
618             m_data->l1_index = l1_index;
619             m_data->l2_index = l2_index;
620             m_data->l2_offset = l2_offset;
621             m_data->valid = 1;
622         }
623     }
624     cluster_offset <<= 9;
625     return cluster_offset;
626 }
627 
628 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
629                              int nb_sectors, int *pnum)
630 {
631     BDRVVmdkState *s = bs->opaque;
632     int index_in_cluster, n;
633     uint64_t cluster_offset;
634 
635     cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
636     index_in_cluster = sector_num % s->cluster_sectors;
637     n = s->cluster_sectors - index_in_cluster;
638     if (n > nb_sectors)
639         n = nb_sectors;
640     *pnum = n;
641     return (cluster_offset != 0);
642 }
643 
644 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
645                     uint8_t *buf, int nb_sectors)
646 {
647     BDRVVmdkState *s = bs->opaque;
648     int index_in_cluster, n, ret;
649     uint64_t cluster_offset;
650 
651     while (nb_sectors > 0) {
652         cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
653         index_in_cluster = sector_num % s->cluster_sectors;
654         n = s->cluster_sectors - index_in_cluster;
655         if (n > nb_sectors)
656             n = nb_sectors;
657         if (!cluster_offset) {
658             // try to read from parent image, if exist
659             if (bs->backing_hd) {
660                 if (!vmdk_is_cid_valid(bs))
661                     return -1;
662                 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
663                 if (ret < 0)
664                     return -1;
665             } else {
666                 memset(buf, 0, 512 * n);
667             }
668         } else {
669             if(bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
670                 return -1;
671         }
672         nb_sectors -= n;
673         sector_num += n;
674         buf += n * 512;
675     }
676     return 0;
677 }
678 
679 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
680                      const uint8_t *buf, int nb_sectors)
681 {
682     BDRVVmdkState *s = bs->opaque;
683     VmdkMetaData m_data;
684     int index_in_cluster, n;
685     uint64_t cluster_offset;
686     static int cid_update = 0;
687 
688     if (sector_num > bs->total_sectors) {
689         fprintf(stderr,
690                 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
691                 " total_sectors=0x%" PRIx64 "\n",
692                 sector_num, bs->total_sectors);
693         return -1;
694     }
695 
696     while (nb_sectors > 0) {
697         index_in_cluster = sector_num & (s->cluster_sectors - 1);
698         n = s->cluster_sectors - index_in_cluster;
699         if (n > nb_sectors)
700             n = nb_sectors;
701         cluster_offset = get_cluster_offset(bs, &m_data, sector_num << 9, 1);
702         if (!cluster_offset)
703             return -1;
704 
705         if (bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
706             return -1;
707         if (m_data.valid) {
708             /* update L2 tables */
709             if (vmdk_L2update(bs, &m_data) == -1)
710                 return -1;
711         }
712         nb_sectors -= n;
713         sector_num += n;
714         buf += n * 512;
715 
716         // update CID on the first write every time the virtual disk is opened
717         if (!cid_update) {
718             vmdk_write_cid(bs, time(NULL));
719             cid_update++;
720         }
721     }
722     return 0;
723 }
724 
725 static int vmdk_create(const char *filename, QEMUOptionParameter *options)
726 {
727     int fd, i;
728     VMDK4Header header;
729     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
730     static const char desc_template[] =
731         "# Disk DescriptorFile\n"
732         "version=1\n"
733         "CID=%x\n"
734         "parentCID=ffffffff\n"
735         "createType=\"monolithicSparse\"\n"
736         "\n"
737         "# Extent description\n"
738         "RW %" PRId64 " SPARSE \"%s\"\n"
739         "\n"
740         "# The Disk Data Base \n"
741         "#DDB\n"
742         "\n"
743         "ddb.virtualHWVersion = \"%d\"\n"
744         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
745         "ddb.geometry.heads = \"16\"\n"
746         "ddb.geometry.sectors = \"63\"\n"
747         "ddb.adapterType = \"ide\"\n";
748     char desc[1024];
749     const char *real_filename, *temp_str;
750     int64_t total_size = 0;
751     const char *backing_file = NULL;
752     int flags = 0;
753     int ret;
754 
755     // Read out options
756     while (options && options->name) {
757         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
758             total_size = options->value.n / 512;
759         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
760             backing_file = options->value.s;
761         } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) {
762             flags |= options->value.n ? BLOCK_FLAG_COMPAT6: 0;
763         }
764         options++;
765     }
766 
767     /* XXX: add support for backing file */
768     if (backing_file) {
769         return vmdk_snapshot_create(filename, backing_file);
770     }
771 
772     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
773               0644);
774     if (fd < 0)
775         return -errno;
776     magic = cpu_to_be32(VMDK4_MAGIC);
777     memset(&header, 0, sizeof(header));
778     header.version = cpu_to_le32(1);
779     header.flags = cpu_to_le32(3); /* ?? */
780     header.capacity = cpu_to_le64(total_size);
781     header.granularity = cpu_to_le64(128);
782     header.num_gtes_per_gte = cpu_to_le32(512);
783 
784     grains = (total_size + header.granularity - 1) / header.granularity;
785     gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
786     gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
787     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
788 
789     header.desc_offset = 1;
790     header.desc_size = 20;
791     header.rgd_offset = header.desc_offset + header.desc_size;
792     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
793     header.grain_offset =
794        ((header.gd_offset + gd_size + (gt_size * gt_count) +
795          header.granularity - 1) / header.granularity) *
796         header.granularity;
797 
798     header.desc_offset = cpu_to_le64(header.desc_offset);
799     header.desc_size = cpu_to_le64(header.desc_size);
800     header.rgd_offset = cpu_to_le64(header.rgd_offset);
801     header.gd_offset = cpu_to_le64(header.gd_offset);
802     header.grain_offset = cpu_to_le64(header.grain_offset);
803 
804     header.check_bytes[0] = 0xa;
805     header.check_bytes[1] = 0x20;
806     header.check_bytes[2] = 0xd;
807     header.check_bytes[3] = 0xa;
808 
809     /* write all the data */
810     ret = qemu_write_full(fd, &magic, sizeof(magic));
811     if (ret != sizeof(magic)) {
812         ret = -errno;
813         goto exit;
814     }
815     ret = qemu_write_full(fd, &header, sizeof(header));
816     if (ret != sizeof(header)) {
817         ret = -errno;
818         goto exit;
819     }
820 
821     ret = ftruncate(fd, header.grain_offset << 9);
822     if (ret < 0) {
823         ret = -errno;
824         goto exit;
825     }
826 
827     /* write grain directory */
828     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
829     for (i = 0, tmp = header.rgd_offset + gd_size;
830          i < gt_count; i++, tmp += gt_size) {
831         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
832         if (ret != sizeof(tmp)) {
833             ret = -errno;
834             goto exit;
835         }
836     }
837 
838     /* write backup grain directory */
839     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
840     for (i = 0, tmp = header.gd_offset + gd_size;
841          i < gt_count; i++, tmp += gt_size) {
842         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
843         if (ret != sizeof(tmp)) {
844             ret = -errno;
845             goto exit;
846         }
847     }
848 
849     /* compose the descriptor */
850     real_filename = filename;
851     if ((temp_str = strrchr(real_filename, '\\')) != NULL)
852         real_filename = temp_str + 1;
853     if ((temp_str = strrchr(real_filename, '/')) != NULL)
854         real_filename = temp_str + 1;
855     if ((temp_str = strrchr(real_filename, ':')) != NULL)
856         real_filename = temp_str + 1;
857     snprintf(desc, sizeof(desc), desc_template, (unsigned int)time(NULL),
858              total_size, real_filename,
859              (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
860              total_size / (int64_t)(63 * 16));
861 
862     /* write the descriptor */
863     lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
864     ret = qemu_write_full(fd, desc, strlen(desc));
865     if (ret != strlen(desc)) {
866         ret = -errno;
867         goto exit;
868     }
869 
870     ret = 0;
871 exit:
872     close(fd);
873     return ret;
874 }
875 
876 static void vmdk_close(BlockDriverState *bs)
877 {
878     BDRVVmdkState *s = bs->opaque;
879 
880     qemu_free(s->l1_table);
881     qemu_free(s->l2_cache);
882     // try to close parent image, if exist
883     vmdk_parent_close(s->hd);
884     bdrv_delete(s->hd);
885 }
886 
887 static void vmdk_flush(BlockDriverState *bs)
888 {
889     BDRVVmdkState *s = bs->opaque;
890     bdrv_flush(s->hd);
891 }
892 
893 
894 static QEMUOptionParameter vmdk_create_options[] = {
895     {
896         .name = BLOCK_OPT_SIZE,
897         .type = OPT_SIZE,
898         .help = "Virtual disk size"
899     },
900     {
901         .name = BLOCK_OPT_BACKING_FILE,
902         .type = OPT_STRING,
903         .help = "File name of a base image"
904     },
905     {
906         .name = BLOCK_OPT_COMPAT6,
907         .type = OPT_FLAG,
908         .help = "VMDK version 6 image"
909     },
910     { NULL }
911 };
912 
913 static BlockDriver bdrv_vmdk = {
914     .format_name	= "vmdk",
915     .instance_size	= sizeof(BDRVVmdkState),
916     .bdrv_probe		= vmdk_probe,
917     .bdrv_open		= vmdk_open,
918     .bdrv_read		= vmdk_read,
919     .bdrv_write		= vmdk_write,
920     .bdrv_close		= vmdk_close,
921     .bdrv_create	= vmdk_create,
922     .bdrv_flush		= vmdk_flush,
923     .bdrv_is_allocated	= vmdk_is_allocated,
924 
925     .create_options = vmdk_create_options,
926 };
927 
928 static void bdrv_vmdk_init(void)
929 {
930     bdrv_register(&bdrv_vmdk);
931 }
932 
933 block_init(bdrv_vmdk_init);
934