]> git.proxmox.com Git - qemu.git/blame - block/vmdk.c
vmdk: Fix COW
[qemu.git] / block / vmdk.c
CommitLineData
ea2384d3
FB
1/*
2 * Block driver for the VMDK format
5fafdf24 3 *
ea2384d3 4 * Copyright (c) 2004 Fabrice Bellard
ff1afc72 5 * Copyright (c) 2005 Filip Navara
5fafdf24 6 *
ea2384d3
FB
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 */
5f4da8c0 25
faf07963 26#include "qemu-common.h"
ea2384d3 27#include "block_int.h"
5efa9d5a 28#include "module.h"
ea2384d3 29
ea2384d3
FB
30#define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
31#define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
32
33typedef 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
46typedef 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];
ff1afc72 59} __attribute__((packed)) VMDK4Header;
ea2384d3
FB
60
61#define L2_CACHE_SIZE 16
62
63typedef struct BDRVVmdkState {
5f4da8c0 64 BlockDriverState *hd;
ea2384d3 65 int64_t l1_table_offset;
ff1afc72 66 int64_t l1_backup_table_offset;
ea2384d3 67 uint32_t *l1_table;
ff1afc72 68 uint32_t *l1_backup_table;
ea2384d3
FB
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;
5f4da8c0 78 uint32_t parent_cid;
630530a6 79 int is_parent;
ea2384d3
FB
80} BDRVVmdkState;
81
630530a6
TS
82typedef 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
ea2384d3
FB
90static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
91{
92 uint32_t magic;
93
94 if (buf_size < 4)
95 return 0;
96 magic = be32_to_cpu(*(uint32_t *)buf);
97 if (magic == VMDK3_MAGIC ||
98 magic == VMDK4_MAGIC)
99 return 100;
100 else
101 return 0;
102}
103
5f4da8c0
TS
104#define CHECK_CID 1
105
3b46e624 106#define SECTOR_SIZE 512
5f4da8c0 107#define DESC_SIZE 20*SECTOR_SIZE // 20 sectors of 512 bytes each
5fafdf24 108#define HEADER_SIZE 512 // first sector of 512 bytes
5f4da8c0
TS
109
110static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
ea2384d3
FB
111{
112 BDRVVmdkState *s = bs->opaque;
5f4da8c0
TS
113 char desc[DESC_SIZE];
114 uint32_t cid;
7ccfb2eb 115 const char *p_name, *cid_str;
5f4da8c0
TS
116 size_t cid_str_size;
117
118 /* the descriptor offset = 0x200 */
119 if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
120 return 0;
121
122 if (parent) {
123 cid_str = "parentCID";
124 cid_str_size = sizeof("parentCID");
125 } else {
126 cid_str = "CID";
127 cid_str_size = sizeof("CID");
128 }
129
511d2b14 130 if ((p_name = strstr(desc,cid_str)) != NULL) {
5f4da8c0
TS
131 p_name += cid_str_size;
132 sscanf(p_name,"%x",&cid);
133 }
134
135 return cid;
136}
137
138static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
139{
140 BDRVVmdkState *s = bs->opaque;
141 char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
142 char *p_name, *tmp_str;
143
144 /* the descriptor offset = 0x200 */
145 if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
146 return -1;
147
148 tmp_str = strstr(desc,"parentCID");
363a37d5 149 pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
511d2b14 150 if ((p_name = strstr(desc,"CID")) != NULL) {
5f4da8c0 151 p_name += sizeof("CID");
363a37d5
BS
152 snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
153 pstrcat(desc, sizeof(desc), tmp_desc);
5f4da8c0
TS
154 }
155
156 if (bdrv_pwrite(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
157 return -1;
158 return 0;
159}
160
161static int vmdk_is_cid_valid(BlockDriverState *bs)
162{
163#ifdef CHECK_CID
164 BDRVVmdkState *s = bs->opaque;
b171271a 165 BlockDriverState *p_bs = bs->backing_hd;
5f4da8c0
TS
166 uint32_t cur_pcid;
167
168 if (p_bs) {
169 cur_pcid = vmdk_read_cid(p_bs,0);
170 if (s->parent_cid != cur_pcid)
171 // CID not valid
172 return 0;
173 }
174#endif
175 // CID valid
176 return 1;
177}
178
179static int vmdk_snapshot_create(const char *filename, const char *backing_file)
180{
181 int snp_fd, p_fd;
53c2e716 182 int ret;
5f4da8c0 183 uint32_t p_cid;
5fafdf24 184 char *p_name, *gd_buf, *rgd_buf;
5f4da8c0
TS
185 const char *real_filename, *temp_str;
186 VMDK4Header header;
187 uint32_t gde_entries, gd_size;
188 int64_t gd_offset, rgd_offset, capacity, gt_size;
189 char p_desc[DESC_SIZE], s_desc[DESC_SIZE], hdr[HEADER_SIZE];
7ccfb2eb 190 static const char desc_template[] =
5f4da8c0
TS
191 "# Disk DescriptorFile\n"
192 "version=1\n"
193 "CID=%x\n"
194 "parentCID=%x\n"
195 "createType=\"monolithicSparse\"\n"
196 "parentFileNameHint=\"%s\"\n"
197 "\n"
198 "# Extent description\n"
7ccfb2eb 199 "RW %u SPARSE \"%s\"\n"
5f4da8c0
TS
200 "\n"
201 "# The Disk Data Base \n"
202 "#DDB\n"
203 "\n";
204
205 snp_fd = open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE, 0644);
206 if (snp_fd < 0)
53c2e716 207 return -errno;
5f4da8c0
TS
208 p_fd = open(backing_file, O_RDONLY | O_BINARY | O_LARGEFILE);
209 if (p_fd < 0) {
210 close(snp_fd);
53c2e716 211 return -errno;
5f4da8c0
TS
212 }
213
214 /* read the header */
53c2e716
JQ
215 if (lseek(p_fd, 0x0, SEEK_SET) == -1) {
216 ret = -errno;
5f4da8c0 217 goto fail;
53c2e716
JQ
218 }
219 if (read(p_fd, hdr, HEADER_SIZE) != HEADER_SIZE) {
220 ret = -errno;
5f4da8c0 221 goto fail;
53c2e716 222 }
5f4da8c0
TS
223
224 /* write the header */
53c2e716
JQ
225 if (lseek(snp_fd, 0x0, SEEK_SET) == -1) {
226 ret = -errno;
5f4da8c0 227 goto fail;
53c2e716
JQ
228 }
229 if (write(snp_fd, hdr, HEADER_SIZE) == -1) {
230 ret = -errno;
5f4da8c0 231 goto fail;
53c2e716 232 }
5f4da8c0
TS
233
234 memset(&header, 0, sizeof(header));
235 memcpy(&header,&hdr[4], sizeof(header)); // skip the VMDK4_MAGIC
236
53c2e716
JQ
237 if (ftruncate(snp_fd, header.grain_offset << 9)) {
238 ret = -errno;
1640366c 239 goto fail;
53c2e716 240 }
5f4da8c0 241 /* the descriptor offset = 0x200 */
53c2e716
JQ
242 if (lseek(p_fd, 0x200, SEEK_SET) == -1) {
243 ret = -errno;
5f4da8c0 244 goto fail;
53c2e716
JQ
245 }
246 if (read(p_fd, p_desc, DESC_SIZE) != DESC_SIZE) {
247 ret = -errno;
5f4da8c0 248 goto fail;
53c2e716 249 }
5f4da8c0 250
511d2b14 251 if ((p_name = strstr(p_desc,"CID")) != NULL) {
5f4da8c0
TS
252 p_name += sizeof("CID");
253 sscanf(p_name,"%x",&p_cid);
254 }
255
256 real_filename = filename;
257 if ((temp_str = strrchr(real_filename, '\\')) != NULL)
258 real_filename = temp_str + 1;
259 if ((temp_str = strrchr(real_filename, '/')) != NULL)
260 real_filename = temp_str + 1;
261 if ((temp_str = strrchr(real_filename, ':')) != NULL)
262 real_filename = temp_str + 1;
263
363a37d5
BS
264 snprintf(s_desc, sizeof(s_desc), desc_template, p_cid, p_cid, backing_file,
265 (uint32_t)header.capacity, real_filename);
5f4da8c0
TS
266
267 /* write the descriptor */
53c2e716
JQ
268 if (lseek(snp_fd, 0x200, SEEK_SET) == -1) {
269 ret = -errno;
5f4da8c0 270 goto fail;
53c2e716
JQ
271 }
272 if (write(snp_fd, s_desc, strlen(s_desc)) == -1) {
273 ret = -errno;
5f4da8c0 274 goto fail;
53c2e716 275 }
ea2384d3 276
5f4da8c0
TS
277 gd_offset = header.gd_offset * SECTOR_SIZE; // offset of GD table
278 rgd_offset = header.rgd_offset * SECTOR_SIZE; // offset of RGD table
279 capacity = header.capacity * SECTOR_SIZE; // Extent size
280 /*
281 * Each GDE span 32M disk, means:
282 * 512 GTE per GT, each GTE points to grain
283 */
284 gt_size = (int64_t)header.num_gtes_per_gte * header.granularity * SECTOR_SIZE;
53c2e716
JQ
285 if (!gt_size) {
286 ret = -EINVAL;
5f4da8c0 287 goto fail;
53c2e716 288 }
5fafdf24 289 gde_entries = (uint32_t)(capacity / gt_size); // number of gde/rgde
5f4da8c0
TS
290 gd_size = gde_entries * sizeof(uint32_t);
291
292 /* write RGD */
293 rgd_buf = qemu_malloc(gd_size);
53c2e716
JQ
294 if (lseek(p_fd, rgd_offset, SEEK_SET) == -1) {
295 ret = -errno;
5f4da8c0 296 goto fail_rgd;
53c2e716
JQ
297 }
298 if (read(p_fd, rgd_buf, gd_size) != gd_size) {
299 ret = -errno;
5f4da8c0 300 goto fail_rgd;
53c2e716
JQ
301 }
302 if (lseek(snp_fd, rgd_offset, SEEK_SET) == -1) {
303 ret = -errno;
5f4da8c0 304 goto fail_rgd;
53c2e716
JQ
305 }
306 if (write(snp_fd, rgd_buf, gd_size) == -1) {
307 ret = -errno;
5f4da8c0 308 goto fail_rgd;
53c2e716 309 }
5f4da8c0
TS
310
311 /* write GD */
312 gd_buf = qemu_malloc(gd_size);
53c2e716
JQ
313 if (lseek(p_fd, gd_offset, SEEK_SET) == -1) {
314 ret = -errno;
5f4da8c0 315 goto fail_gd;
53c2e716
JQ
316 }
317 if (read(p_fd, gd_buf, gd_size) != gd_size) {
318 ret = -errno;
5f4da8c0 319 goto fail_gd;
53c2e716
JQ
320 }
321 if (lseek(snp_fd, gd_offset, SEEK_SET) == -1) {
322 ret = -errno;
5f4da8c0 323 goto fail_gd;
53c2e716
JQ
324 }
325 if (write(snp_fd, gd_buf, gd_size) == -1) {
326 ret = -errno;
5f4da8c0 327 goto fail_gd;
53c2e716 328 }
3829cb46 329 ret = 0;
5f4da8c0 330
3829cb46 331fail_gd:
5f4da8c0 332 qemu_free(gd_buf);
3829cb46 333fail_rgd:
5f4da8c0 334 qemu_free(rgd_buf);
3829cb46 335fail:
5f4da8c0
TS
336 close(p_fd);
337 close(snp_fd);
53c2e716 338 return ret;
5f4da8c0
TS
339}
340
341static void vmdk_parent_close(BlockDriverState *bs)
342{
343 if (bs->backing_hd)
344 bdrv_close(bs->backing_hd);
345}
346
b1d8e52e 347static int parent_open = 0;
5f4da8c0
TS
348static int vmdk_parent_open(BlockDriverState *bs, const char * filename)
349{
350 BDRVVmdkState *s = bs->opaque;
5fafdf24 351 char *p_name;
5f4da8c0
TS
352 char desc[DESC_SIZE];
353 char parent_img_name[1024];
354
355 /* the descriptor offset = 0x200 */
356 if (bdrv_pread(s->hd, 0x200, desc, DESC_SIZE) != DESC_SIZE)
357 return -1;
358
511d2b14 359 if ((p_name = strstr(desc,"parentFileNameHint")) != NULL) {
5f4da8c0
TS
360 char *end_name;
361 struct stat file_buf;
362
363 p_name += sizeof("parentFileNameHint") + 1;
511d2b14 364 if ((end_name = strchr(p_name,'\"')) == NULL)
5f4da8c0 365 return -1;
b171271a 366 if ((end_name - p_name) > sizeof (bs->backing_file) - 1)
b34d259a 367 return -1;
3b46e624 368
b171271a
KW
369 pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
370 if (stat(bs->backing_file, &file_buf) != 0) {
5f4da8c0 371 path_combine(parent_img_name, sizeof(parent_img_name),
b171271a 372 filename, bs->backing_file);
5f4da8c0 373 } else {
363a37d5 374 pstrcpy(parent_img_name, sizeof(parent_img_name),
b171271a 375 bs->backing_file);
5f4da8c0
TS
376 }
377
b171271a
KW
378 bs->backing_hd = bdrv_new("");
379 if (!bs->backing_hd) {
5f4da8c0
TS
380 failure:
381 bdrv_close(s->hd);
ff1afc72 382 return -1;
5f4da8c0 383 }
630530a6 384 parent_open = 1;
d6e9098e 385 if (bdrv_open(bs->backing_hd, parent_img_name, 0, NULL) < 0)
5f4da8c0 386 goto failure;
630530a6 387 parent_open = 0;
ff1afc72 388 }
5f4da8c0
TS
389
390 return 0;
391}
392
393static int vmdk_open(BlockDriverState *bs, const char *filename, int flags)
394{
395 BDRVVmdkState *s = bs->opaque;
396 uint32_t magic;
397 int l1_size, i, ret;
398
f5edb014
NS
399 if (parent_open) {
400 /* Parent must be opened as RO, no RDWR. */
401 flags = 0;
402 }
630530a6 403
b5eff355 404 ret = bdrv_file_open(&s->hd, filename, flags);
5f4da8c0
TS
405 if (ret < 0)
406 return ret;
407 if (bdrv_pread(s->hd, 0, &magic, sizeof(magic)) != sizeof(magic))
ea2384d3 408 goto fail;
5f4da8c0 409
7143c62c 410 magic = be32_to_cpu(magic);
ea2384d3
FB
411 if (magic == VMDK3_MAGIC) {
412 VMDK3Header header;
5f4da8c0
TS
413
414 if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
ea2384d3
FB
415 goto fail;
416 s->cluster_sectors = le32_to_cpu(header.granularity);
417 s->l2_size = 1 << 9;
418 s->l1_size = 1 << 6;
419 bs->total_sectors = le32_to_cpu(header.disk_sectors);
ff1afc72
FB
420 s->l1_table_offset = le32_to_cpu(header.l1dir_offset) << 9;
421 s->l1_backup_table_offset = 0;
ea2384d3
FB
422 s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
423 } else if (magic == VMDK4_MAGIC) {
424 VMDK4Header header;
5f4da8c0
TS
425
426 if (bdrv_pread(s->hd, sizeof(magic), &header, sizeof(header)) != sizeof(header))
ea2384d3 427 goto fail;
bd6ea3c8
FB
428 bs->total_sectors = le64_to_cpu(header.capacity);
429 s->cluster_sectors = le64_to_cpu(header.granularity);
ea2384d3
FB
430 s->l2_size = le32_to_cpu(header.num_gtes_per_gte);
431 s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
432 if (s->l1_entry_sectors <= 0)
433 goto fail;
5fafdf24 434 s->l1_size = (bs->total_sectors + s->l1_entry_sectors - 1)
ea2384d3 435 / s->l1_entry_sectors;
ff1afc72
FB
436 s->l1_table_offset = le64_to_cpu(header.rgd_offset) << 9;
437 s->l1_backup_table_offset = le64_to_cpu(header.gd_offset) << 9;
5f4da8c0 438
630530a6
TS
439 if (parent_open)
440 s->is_parent = 1;
441 else
442 s->is_parent = 0;
443
5f4da8c0
TS
444 // try to open parent images, if exist
445 if (vmdk_parent_open(bs, filename) != 0)
446 goto fail;
447 // write the CID once after the image creation
448 s->parent_cid = vmdk_read_cid(bs,1);
ea2384d3
FB
449 } else {
450 goto fail;
451 }
5f4da8c0 452
ea2384d3
FB
453 /* read the L1 table */
454 l1_size = s->l1_size * sizeof(uint32_t);
455 s->l1_table = qemu_malloc(l1_size);
5f4da8c0 456 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, l1_size) != l1_size)
ea2384d3
FB
457 goto fail;
458 for(i = 0; i < s->l1_size; i++) {
459 le32_to_cpus(&s->l1_table[i]);
460 }
461
ff1afc72
FB
462 if (s->l1_backup_table_offset) {
463 s->l1_backup_table = qemu_malloc(l1_size);
5f4da8c0 464 if (bdrv_pread(s->hd, s->l1_backup_table_offset, s->l1_backup_table, l1_size) != l1_size)
ff1afc72
FB
465 goto fail;
466 for(i = 0; i < s->l1_size; i++) {
467 le32_to_cpus(&s->l1_backup_table[i]);
468 }
469 }
470
ea2384d3 471 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
ea2384d3
FB
472 return 0;
473 fail:
ff1afc72 474 qemu_free(s->l1_backup_table);
ea2384d3
FB
475 qemu_free(s->l1_table);
476 qemu_free(s->l2_cache);
5f4da8c0 477 bdrv_delete(s->hd);
ea2384d3
FB
478 return -1;
479}
480
630530a6
TS
481static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
482 uint64_t offset, int allocate);
5f4da8c0
TS
483
484static int get_whole_cluster(BlockDriverState *bs, uint64_t cluster_offset,
485 uint64_t offset, int allocate)
486{
5f4da8c0
TS
487 BDRVVmdkState *s = bs->opaque;
488 uint8_t whole_grain[s->cluster_sectors*512]; // 128 sectors * 512 bytes each = grain size 64KB
489
490 // we will be here if it's first write on non-exist grain(cluster).
491 // try to read from parent image, if exist
b171271a 492 if (bs->backing_hd) {
c336500d 493 int ret;
5f4da8c0
TS
494
495 if (!vmdk_is_cid_valid(bs))
496 return -1;
5f4da8c0 497
c336500d
KW
498 ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain,
499 s->cluster_sectors);
500 if (ret < 0) {
501 return -1;
502 }
630530a6 503
c336500d
KW
504 //Write grain only into the active image
505 ret = bdrv_write(s->hd, cluster_offset, whole_grain,
506 s->cluster_sectors);
507 if (ret < 0) {
508 return -1;
630530a6
TS
509 }
510 }
511 return 0;
512}
513
514static int vmdk_L2update(BlockDriverState *bs, VmdkMetaData *m_data)
515{
516 BDRVVmdkState *s = bs->opaque;
517
518 /* update L2 table */
519 if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
520 &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
521 return -1;
522 /* update backup L2 table */
523 if (s->l1_backup_table_offset != 0) {
524 m_data->l2_offset = s->l1_backup_table[m_data->l1_index];
525 if (bdrv_pwrite(s->hd, ((int64_t)m_data->l2_offset * 512) + (m_data->l2_index * sizeof(m_data->offset)),
526 &(m_data->offset), sizeof(m_data->offset)) != sizeof(m_data->offset))
5f4da8c0
TS
527 return -1;
528 }
630530a6 529
5f4da8c0
TS
530 return 0;
531}
532
630530a6 533static uint64_t get_cluster_offset(BlockDriverState *bs, VmdkMetaData *m_data,
ff1afc72 534 uint64_t offset, int allocate)
ea2384d3
FB
535{
536 BDRVVmdkState *s = bs->opaque;
537 unsigned int l1_index, l2_offset, l2_index;
538 int min_index, i, j;
630530a6 539 uint32_t min_count, *l2_table, tmp = 0;
ea2384d3 540 uint64_t cluster_offset;
630530a6
TS
541
542 if (m_data)
543 m_data->valid = 0;
544
ea2384d3
FB
545 l1_index = (offset >> 9) / s->l1_entry_sectors;
546 if (l1_index >= s->l1_size)
547 return 0;
548 l2_offset = s->l1_table[l1_index];
549 if (!l2_offset)
550 return 0;
ea2384d3
FB
551 for(i = 0; i < L2_CACHE_SIZE; i++) {
552 if (l2_offset == s->l2_cache_offsets[i]) {
553 /* increment the hit count */
554 if (++s->l2_cache_counts[i] == 0xffffffff) {
555 for(j = 0; j < L2_CACHE_SIZE; j++) {
556 s->l2_cache_counts[j] >>= 1;
557 }
558 }
559 l2_table = s->l2_cache + (i * s->l2_size);
560 goto found;
561 }
562 }
563 /* not found: load a new entry in the least used one */
564 min_index = 0;
565 min_count = 0xffffffff;
566 for(i = 0; i < L2_CACHE_SIZE; i++) {
567 if (s->l2_cache_counts[i] < min_count) {
568 min_count = s->l2_cache_counts[i];
569 min_index = i;
570 }
571 }
572 l2_table = s->l2_cache + (min_index * s->l2_size);
5fafdf24 573 if (bdrv_pread(s->hd, (int64_t)l2_offset * 512, l2_table, s->l2_size * sizeof(uint32_t)) !=
5f4da8c0 574 s->l2_size * sizeof(uint32_t))
ea2384d3 575 return 0;
5f4da8c0 576
ea2384d3
FB
577 s->l2_cache_offsets[min_index] = l2_offset;
578 s->l2_cache_counts[min_index] = 1;
579 found:
580 l2_index = ((offset >> 9) / s->cluster_sectors) % s->l2_size;
581 cluster_offset = le32_to_cpu(l2_table[l2_index]);
630530a6 582
ff1afc72
FB
583 if (!cluster_offset) {
584 if (!allocate)
585 return 0;
630530a6
TS
586 // Avoid the L2 tables update for the images that have snapshots.
587 if (!s->is_parent) {
49140a77 588 cluster_offset = bdrv_getlength(s->hd);
630530a6
TS
589 bdrv_truncate(s->hd, cluster_offset + (s->cluster_sectors << 9));
590
591 cluster_offset >>= 9;
592 tmp = cpu_to_le32(cluster_offset);
593 l2_table[l2_index] = tmp;
ff1afc72 594 }
630530a6
TS
595 /* First of all we write grain itself, to avoid race condition
596 * that may to corrupt the image.
597 * This problem may occur because of insufficient space on host disk
598 * or inappropriate VM shutdown.
599 */
5f4da8c0
TS
600 if (get_whole_cluster(bs, cluster_offset, offset, allocate) == -1)
601 return 0;
630530a6
TS
602
603 if (m_data) {
604 m_data->offset = tmp;
605 m_data->l1_index = l1_index;
606 m_data->l2_index = l2_index;
607 m_data->l2_offset = l2_offset;
608 m_data->valid = 1;
609 }
ff1afc72 610 }
ea2384d3
FB
611 cluster_offset <<= 9;
612 return cluster_offset;
613}
614
5fafdf24 615static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
ea2384d3
FB
616 int nb_sectors, int *pnum)
617{
618 BDRVVmdkState *s = bs->opaque;
619 int index_in_cluster, n;
620 uint64_t cluster_offset;
621
630530a6 622 cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
ea2384d3
FB
623 index_in_cluster = sector_num % s->cluster_sectors;
624 n = s->cluster_sectors - index_in_cluster;
625 if (n > nb_sectors)
626 n = nb_sectors;
627 *pnum = n;
628 return (cluster_offset != 0);
629}
630
5fafdf24 631static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
ea2384d3
FB
632 uint8_t *buf, int nb_sectors)
633{
634 BDRVVmdkState *s = bs->opaque;
5f4da8c0 635 int index_in_cluster, n, ret;
ea2384d3 636 uint64_t cluster_offset;
5f4da8c0 637
ea2384d3 638 while (nb_sectors > 0) {
630530a6 639 cluster_offset = get_cluster_offset(bs, NULL, sector_num << 9, 0);
ea2384d3
FB
640 index_in_cluster = sector_num % s->cluster_sectors;
641 n = s->cluster_sectors - index_in_cluster;
642 if (n > nb_sectors)
643 n = nb_sectors;
644 if (!cluster_offset) {
5f4da8c0 645 // try to read from parent image, if exist
b171271a 646 if (bs->backing_hd) {
5f4da8c0
TS
647 if (!vmdk_is_cid_valid(bs))
648 return -1;
b171271a 649 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
5f4da8c0
TS
650 if (ret < 0)
651 return -1;
652 } else {
653 memset(buf, 0, 512 * n);
654 }
ea2384d3 655 } else {
5f4da8c0 656 if(bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
ea2384d3
FB
657 return -1;
658 }
659 nb_sectors -= n;
660 sector_num += n;
661 buf += n * 512;
662 }
663 return 0;
664}
665
5fafdf24 666static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
ea2384d3
FB
667 const uint8_t *buf, int nb_sectors)
668{
ff1afc72 669 BDRVVmdkState *s = bs->opaque;
630530a6 670 VmdkMetaData m_data;
5f4da8c0 671 int index_in_cluster, n;
ff1afc72 672 uint64_t cluster_offset;
5f4da8c0 673 static int cid_update = 0;
ff1afc72 674
630530a6
TS
675 if (sector_num > bs->total_sectors) {
676 fprintf(stderr,
92868412
JM
677 "(VMDK) Wrong offset: sector_num=0x%" PRIx64
678 " total_sectors=0x%" PRIx64 "\n",
630530a6
TS
679 sector_num, bs->total_sectors);
680 return -1;
681 }
682
ff1afc72
FB
683 while (nb_sectors > 0) {
684 index_in_cluster = sector_num & (s->cluster_sectors - 1);
685 n = s->cluster_sectors - index_in_cluster;
686 if (n > nb_sectors)
687 n = nb_sectors;
630530a6 688 cluster_offset = get_cluster_offset(bs, &m_data, sector_num << 9, 1);
ff1afc72
FB
689 if (!cluster_offset)
690 return -1;
630530a6 691
5f4da8c0 692 if (bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512) != n * 512)
ff1afc72 693 return -1;
630530a6
TS
694 if (m_data.valid) {
695 /* update L2 tables */
696 if (vmdk_L2update(bs, &m_data) == -1)
697 return -1;
698 }
ff1afc72
FB
699 nb_sectors -= n;
700 sector_num += n;
701 buf += n * 512;
5f4da8c0
TS
702
703 // update CID on the first write every time the virtual disk is opened
704 if (!cid_update) {
705 vmdk_write_cid(bs, time(NULL));
706 cid_update++;
707 }
ff1afc72
FB
708 }
709 return 0;
ea2384d3
FB
710}
711
0e7e1989 712static int vmdk_create(const char *filename, QEMUOptionParameter *options)
8979b227
FB
713{
714 int fd, i;
715 VMDK4Header header;
716 uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
7ccfb2eb 717 static const char desc_template[] =
8979b227
FB
718 "# Disk DescriptorFile\n"
719 "version=1\n"
720 "CID=%x\n"
721 "parentCID=ffffffff\n"
722 "createType=\"monolithicSparse\"\n"
723 "\n"
724 "# Extent description\n"
7fd6d9fc 725 "RW %" PRId64 " SPARSE \"%s\"\n"
8979b227
FB
726 "\n"
727 "# The Disk Data Base \n"
728 "#DDB\n"
729 "\n"
ec36ba14 730 "ddb.virtualHWVersion = \"%d\"\n"
7fd6d9fc 731 "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
8979b227
FB
732 "ddb.geometry.heads = \"16\"\n"
733 "ddb.geometry.sectors = \"63\"\n"
734 "ddb.adapterType = \"ide\"\n";
735 char desc[1024];
736 const char *real_filename, *temp_str;
0e7e1989
KW
737 int64_t total_size = 0;
738 const char *backing_file = NULL;
739 int flags = 0;
1640366c 740 int ret;
0e7e1989
KW
741
742 // Read out options
743 while (options && options->name) {
744 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
745 total_size = options->value.n / 512;
746 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
747 backing_file = options->value.s;
748 } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) {
749 flags |= options->value.n ? BLOCK_FLAG_COMPAT6: 0;
750 }
751 options++;
752 }
8979b227
FB
753
754 /* XXX: add support for backing file */
5f4da8c0
TS
755 if (backing_file) {
756 return vmdk_snapshot_create(filename, backing_file);
757 }
8979b227
FB
758
759 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
760 0644);
761 if (fd < 0)
b781cce5 762 return -errno;
8979b227
FB
763 magic = cpu_to_be32(VMDK4_MAGIC);
764 memset(&header, 0, sizeof(header));
765 header.version = cpu_to_le32(1);
766 header.flags = cpu_to_le32(3); /* ?? */
767 header.capacity = cpu_to_le64(total_size);
768 header.granularity = cpu_to_le64(128);
769 header.num_gtes_per_gte = cpu_to_le32(512);
770
771 grains = (total_size + header.granularity - 1) / header.granularity;
772 gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
773 gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
774 gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
775
776 header.desc_offset = 1;
777 header.desc_size = 20;
778 header.rgd_offset = header.desc_offset + header.desc_size;
779 header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
780 header.grain_offset =
781 ((header.gd_offset + gd_size + (gt_size * gt_count) +
782 header.granularity - 1) / header.granularity) *
783 header.granularity;
784
785 header.desc_offset = cpu_to_le64(header.desc_offset);
786 header.desc_size = cpu_to_le64(header.desc_size);
787 header.rgd_offset = cpu_to_le64(header.rgd_offset);
788 header.gd_offset = cpu_to_le64(header.gd_offset);
789 header.grain_offset = cpu_to_le64(header.grain_offset);
790
791 header.check_bytes[0] = 0xa;
792 header.check_bytes[1] = 0x20;
793 header.check_bytes[2] = 0xd;
794 header.check_bytes[3] = 0xa;
3b46e624
TS
795
796 /* write all the data */
1640366c
KS
797 ret = qemu_write_full(fd, &magic, sizeof(magic));
798 if (ret != sizeof(magic)) {
b781cce5 799 ret = -errno;
1640366c
KS
800 goto exit;
801 }
802 ret = qemu_write_full(fd, &header, sizeof(header));
803 if (ret != sizeof(header)) {
b781cce5 804 ret = -errno;
1640366c
KS
805 goto exit;
806 }
8979b227 807
1640366c
KS
808 ret = ftruncate(fd, header.grain_offset << 9);
809 if (ret < 0) {
b781cce5 810 ret = -errno;
1640366c
KS
811 goto exit;
812 }
8979b227
FB
813
814 /* write grain directory */
815 lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
816 for (i = 0, tmp = header.rgd_offset + gd_size;
1640366c
KS
817 i < gt_count; i++, tmp += gt_size) {
818 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
819 if (ret != sizeof(tmp)) {
b781cce5 820 ret = -errno;
1640366c
KS
821 goto exit;
822 }
823 }
3b46e624 824
8979b227
FB
825 /* write backup grain directory */
826 lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
827 for (i = 0, tmp = header.gd_offset + gd_size;
1640366c
KS
828 i < gt_count; i++, tmp += gt_size) {
829 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
830 if (ret != sizeof(tmp)) {
b781cce5 831 ret = -errno;
1640366c
KS
832 goto exit;
833 }
834 }
8979b227
FB
835
836 /* compose the descriptor */
837 real_filename = filename;
838 if ((temp_str = strrchr(real_filename, '\\')) != NULL)
839 real_filename = temp_str + 1;
840 if ((temp_str = strrchr(real_filename, '/')) != NULL)
841 real_filename = temp_str + 1;
842 if ((temp_str = strrchr(real_filename, ':')) != NULL)
843 real_filename = temp_str + 1;
7ccfb2eb 844 snprintf(desc, sizeof(desc), desc_template, (unsigned int)time(NULL),
7fd6d9fc
BS
845 total_size, real_filename,
846 (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
847 total_size / (int64_t)(63 * 16));
8979b227
FB
848
849 /* write the descriptor */
850 lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
1640366c
KS
851 ret = qemu_write_full(fd, desc, strlen(desc));
852 if (ret != strlen(desc)) {
b781cce5 853 ret = -errno;
1640366c
KS
854 goto exit;
855 }
8979b227 856
1640366c
KS
857 ret = 0;
858exit:
8979b227 859 close(fd);
1640366c 860 return ret;
8979b227
FB
861}
862
e2731add 863static void vmdk_close(BlockDriverState *bs)
ea2384d3
FB
864{
865 BDRVVmdkState *s = bs->opaque;
5f4da8c0 866
ea2384d3
FB
867 qemu_free(s->l1_table);
868 qemu_free(s->l2_cache);
5f4da8c0
TS
869 // try to close parent image, if exist
870 vmdk_parent_close(s->hd);
5cbdd273 871 bdrv_delete(s->hd);
ea2384d3
FB
872}
873
7a6cba61
PB
874static void vmdk_flush(BlockDriverState *bs)
875{
876 BDRVVmdkState *s = bs->opaque;
5f4da8c0 877 bdrv_flush(s->hd);
7a6cba61
PB
878}
879
0e7e1989
KW
880
881static QEMUOptionParameter vmdk_create_options[] = {
db08adf5
KW
882 {
883 .name = BLOCK_OPT_SIZE,
884 .type = OPT_SIZE,
885 .help = "Virtual disk size"
886 },
887 {
888 .name = BLOCK_OPT_BACKING_FILE,
889 .type = OPT_STRING,
890 .help = "File name of a base image"
891 },
892 {
893 .name = BLOCK_OPT_COMPAT6,
894 .type = OPT_FLAG,
895 .help = "VMDK version 6 image"
896 },
0e7e1989
KW
897 { NULL }
898};
899
5efa9d5a 900static BlockDriver bdrv_vmdk = {
e60f469c
AJ
901 .format_name = "vmdk",
902 .instance_size = sizeof(BDRVVmdkState),
903 .bdrv_probe = vmdk_probe,
66f82cee 904 .bdrv_file_open = vmdk_open,
e60f469c
AJ
905 .bdrv_read = vmdk_read,
906 .bdrv_write = vmdk_write,
907 .bdrv_close = vmdk_close,
908 .bdrv_create = vmdk_create,
909 .bdrv_flush = vmdk_flush,
910 .bdrv_is_allocated = vmdk_is_allocated,
0e7e1989
KW
911
912 .create_options = vmdk_create_options,
ea2384d3 913};
5efa9d5a
AL
914
915static void bdrv_vmdk_init(void)
916{
917 bdrv_register(&bdrv_vmdk);
918}
919
920block_init(bdrv_vmdk_init);