]> git.proxmox.com Git - mirror_qemu.git/blob - block/vpc.c
block: use fdatasync instead of fsync if possible
[mirror_qemu.git] / block / vpc.c
1 /*
2 * Block driver for Connectix / Microsoft Virtual PC images
3 *
4 * Copyright (c) 2005 Alex Beregszaszi
5 * Copyright (c) 2009 Kevin Wolf <kwolf@suse.de>
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 #include "qemu-common.h"
26 #include "block_int.h"
27 #include "module.h"
28
29 /**************************************************************/
30
31 #define HEADER_SIZE 512
32
33 //#define CACHE
34
35 enum vhd_type {
36 VHD_FIXED = 2,
37 VHD_DYNAMIC = 3,
38 VHD_DIFFERENCING = 4,
39 };
40
41 // Seconds since Jan 1, 2000 0:00:00 (UTC)
42 #define VHD_TIMESTAMP_BASE 946684800
43
44 // always big-endian
45 struct vhd_footer {
46 char creator[8]; // "conectix"
47 uint32_t features;
48 uint32_t version;
49
50 // Offset of next header structure, 0xFFFFFFFF if none
51 uint64_t data_offset;
52
53 // Seconds since Jan 1, 2000 0:00:00 (UTC)
54 uint32_t timestamp;
55
56 char creator_app[4]; // "vpc "
57 uint16_t major;
58 uint16_t minor;
59 char creator_os[4]; // "Wi2k"
60
61 uint64_t orig_size;
62 uint64_t size;
63
64 uint16_t cyls;
65 uint8_t heads;
66 uint8_t secs_per_cyl;
67
68 uint32_t type;
69
70 // Checksum of the Hard Disk Footer ("one's complement of the sum of all
71 // the bytes in the footer without the checksum field")
72 uint32_t checksum;
73
74 // UUID used to identify a parent hard disk (backing file)
75 uint8_t uuid[16];
76
77 uint8_t in_saved_state;
78 };
79
80 struct vhd_dyndisk_header {
81 char magic[8]; // "cxsparse"
82
83 // Offset of next header structure, 0xFFFFFFFF if none
84 uint64_t data_offset;
85
86 // Offset of the Block Allocation Table (BAT)
87 uint64_t table_offset;
88
89 uint32_t version;
90 uint32_t max_table_entries; // 32bit/entry
91
92 // 2 MB by default, must be a power of two
93 uint32_t block_size;
94
95 uint32_t checksum;
96 uint8_t parent_uuid[16];
97 uint32_t parent_timestamp;
98 uint32_t reserved;
99
100 // Backing file name (in UTF-16)
101 uint8_t parent_name[512];
102
103 struct {
104 uint32_t platform;
105 uint32_t data_space;
106 uint32_t data_length;
107 uint32_t reserved;
108 uint64_t data_offset;
109 } parent_locator[8];
110 };
111
112 typedef struct BDRVVPCState {
113 BlockDriverState *hd;
114
115 uint8_t footer_buf[HEADER_SIZE];
116 uint64_t free_data_block_offset;
117 int max_table_entries;
118 uint32_t *pagetable;
119 uint64_t bat_offset;
120 uint64_t last_bitmap_offset;
121
122 uint32_t block_size;
123 uint32_t bitmap_size;
124
125 #ifdef CACHE
126 uint8_t *pageentry_u8;
127 uint32_t *pageentry_u32;
128 uint16_t *pageentry_u16;
129
130 uint64_t last_bitmap;
131 #endif
132 } BDRVVPCState;
133
134 static uint32_t vpc_checksum(uint8_t* buf, size_t size)
135 {
136 uint32_t res = 0;
137 int i;
138
139 for (i = 0; i < size; i++)
140 res += buf[i];
141
142 return ~res;
143 }
144
145
146 static int vpc_probe(const uint8_t *buf, int buf_size, const char *filename)
147 {
148 if (buf_size >= 8 && !strncmp((char *)buf, "conectix", 8))
149 return 100;
150 return 0;
151 }
152
153 static int vpc_open(BlockDriverState *bs, const char *filename, int flags)
154 {
155 BDRVVPCState *s = bs->opaque;
156 int ret, i;
157 struct vhd_footer* footer;
158 struct vhd_dyndisk_header* dyndisk_header;
159 uint8_t buf[HEADER_SIZE];
160 uint32_t checksum;
161
162 ret = bdrv_file_open(&s->hd, filename, flags);
163 if (ret < 0)
164 return ret;
165
166 if (bdrv_pread(s->hd, 0, s->footer_buf, HEADER_SIZE) != HEADER_SIZE)
167 goto fail;
168
169 footer = (struct vhd_footer*) s->footer_buf;
170 if (strncmp(footer->creator, "conectix", 8))
171 goto fail;
172
173 checksum = be32_to_cpu(footer->checksum);
174 footer->checksum = 0;
175 if (vpc_checksum(s->footer_buf, HEADER_SIZE) != checksum)
176 fprintf(stderr, "block-vpc: The header checksum of '%s' is "
177 "incorrect.\n", filename);
178
179 // The visible size of a image in Virtual PC depends on the geometry
180 // rather than on the size stored in the footer (the size in the footer
181 // is too large usually)
182 bs->total_sectors = (int64_t)
183 be16_to_cpu(footer->cyls) * footer->heads * footer->secs_per_cyl;
184
185 if (bdrv_pread(s->hd, be64_to_cpu(footer->data_offset), buf, HEADER_SIZE)
186 != HEADER_SIZE)
187 goto fail;
188
189 dyndisk_header = (struct vhd_dyndisk_header*) buf;
190
191 if (strncmp(dyndisk_header->magic, "cxsparse", 8))
192 goto fail;
193
194
195 s->block_size = be32_to_cpu(dyndisk_header->block_size);
196 s->bitmap_size = ((s->block_size / (8 * 512)) + 511) & ~511;
197
198 s->max_table_entries = be32_to_cpu(dyndisk_header->max_table_entries);
199 s->pagetable = qemu_malloc(s->max_table_entries * 4);
200
201 s->bat_offset = be64_to_cpu(dyndisk_header->table_offset);
202 if (bdrv_pread(s->hd, s->bat_offset, s->pagetable,
203 s->max_table_entries * 4) != s->max_table_entries * 4)
204 goto fail;
205
206 s->free_data_block_offset =
207 (s->bat_offset + (s->max_table_entries * 4) + 511) & ~511;
208
209 for (i = 0; i < s->max_table_entries; i++) {
210 be32_to_cpus(&s->pagetable[i]);
211 if (s->pagetable[i] != 0xFFFFFFFF) {
212 int64_t next = (512 * (int64_t) s->pagetable[i]) +
213 s->bitmap_size + s->block_size;
214
215 if (next> s->free_data_block_offset)
216 s->free_data_block_offset = next;
217 }
218 }
219
220 s->last_bitmap_offset = (int64_t) -1;
221
222 #ifdef CACHE
223 s->pageentry_u8 = qemu_malloc(512);
224 s->pageentry_u32 = s->pageentry_u8;
225 s->pageentry_u16 = s->pageentry_u8;
226 s->last_pagetable = -1;
227 #endif
228
229 return 0;
230 fail:
231 bdrv_delete(s->hd);
232 return -1;
233 }
234
235 /*
236 * Returns the absolute byte offset of the given sector in the image file.
237 * If the sector is not allocated, -1 is returned instead.
238 *
239 * The parameter write must be 1 if the offset will be used for a write
240 * operation (the block bitmaps is updated then), 0 otherwise.
241 */
242 static inline int64_t get_sector_offset(BlockDriverState *bs,
243 int64_t sector_num, int write)
244 {
245 BDRVVPCState *s = bs->opaque;
246 uint64_t offset = sector_num * 512;
247 uint64_t bitmap_offset, block_offset;
248 uint32_t pagetable_index, pageentry_index;
249
250 pagetable_index = offset / s->block_size;
251 pageentry_index = (offset % s->block_size) / 512;
252
253 if (pagetable_index >= s->max_table_entries || s->pagetable[pagetable_index] == 0xffffffff)
254 return -1; // not allocated
255
256 bitmap_offset = 512 * (uint64_t) s->pagetable[pagetable_index];
257 block_offset = bitmap_offset + s->bitmap_size + (512 * pageentry_index);
258
259 // We must ensure that we don't write to any sectors which are marked as
260 // unused in the bitmap. We get away with setting all bits in the block
261 // bitmap each time we write to a new block. This might cause Virtual PC to
262 // miss sparse read optimization, but it's not a problem in terms of
263 // correctness.
264 if (write && (s->last_bitmap_offset != bitmap_offset)) {
265 uint8_t bitmap[s->bitmap_size];
266
267 s->last_bitmap_offset = bitmap_offset;
268 memset(bitmap, 0xff, s->bitmap_size);
269 bdrv_pwrite(s->hd, bitmap_offset, bitmap, s->bitmap_size);
270 }
271
272 // printf("sector: %" PRIx64 ", index: %x, offset: %x, bioff: %" PRIx64 ", bloff: %" PRIx64 "\n",
273 // sector_num, pagetable_index, pageentry_index,
274 // bitmap_offset, block_offset);
275
276 // disabled by reason
277 #if 0
278 #ifdef CACHE
279 if (bitmap_offset != s->last_bitmap)
280 {
281 lseek(s->fd, bitmap_offset, SEEK_SET);
282
283 s->last_bitmap = bitmap_offset;
284
285 // Scary! Bitmap is stored as big endian 32bit entries,
286 // while we used to look it up byte by byte
287 read(s->fd, s->pageentry_u8, 512);
288 for (i = 0; i < 128; i++)
289 be32_to_cpus(&s->pageentry_u32[i]);
290 }
291
292 if ((s->pageentry_u8[pageentry_index / 8] >> (pageentry_index % 8)) & 1)
293 return -1;
294 #else
295 lseek(s->fd, bitmap_offset + (pageentry_index / 8), SEEK_SET);
296
297 read(s->fd, &bitmap_entry, 1);
298
299 if ((bitmap_entry >> (pageentry_index % 8)) & 1)
300 return -1; // not allocated
301 #endif
302 #endif
303
304 return block_offset;
305 }
306
307 /*
308 * Writes the footer to the end of the image file. This is needed when the
309 * file grows as it overwrites the old footer
310 *
311 * Returns 0 on success and < 0 on error
312 */
313 static int rewrite_footer(BlockDriverState* bs)
314 {
315 int ret;
316 BDRVVPCState *s = bs->opaque;
317 int64_t offset = s->free_data_block_offset;
318
319 ret = bdrv_pwrite(s->hd, offset, s->footer_buf, HEADER_SIZE);
320 if (ret < 0)
321 return ret;
322
323 return 0;
324 }
325
326 /*
327 * Allocates a new block. This involves writing a new footer and updating
328 * the Block Allocation Table to use the space at the old end of the image
329 * file (overwriting the old footer)
330 *
331 * Returns the sectors' offset in the image file on success and < 0 on error
332 */
333 static int64_t alloc_block(BlockDriverState* bs, int64_t sector_num)
334 {
335 BDRVVPCState *s = bs->opaque;
336 int64_t bat_offset;
337 uint32_t index, bat_value;
338 int ret;
339 uint8_t bitmap[s->bitmap_size];
340
341 // Check if sector_num is valid
342 if ((sector_num < 0) || (sector_num > bs->total_sectors))
343 return -1;
344
345 // Write entry into in-memory BAT
346 index = (sector_num * 512) / s->block_size;
347 if (s->pagetable[index] != 0xFFFFFFFF)
348 return -1;
349
350 s->pagetable[index] = s->free_data_block_offset / 512;
351
352 // Initialize the block's bitmap
353 memset(bitmap, 0xff, s->bitmap_size);
354 bdrv_pwrite(s->hd, s->free_data_block_offset, bitmap, s->bitmap_size);
355
356 // Write new footer (the old one will be overwritten)
357 s->free_data_block_offset += s->block_size + s->bitmap_size;
358 ret = rewrite_footer(bs);
359 if (ret < 0)
360 goto fail;
361
362 // Write BAT entry to disk
363 bat_offset = s->bat_offset + (4 * index);
364 bat_value = be32_to_cpu(s->pagetable[index]);
365 ret = bdrv_pwrite(s->hd, bat_offset, &bat_value, 4);
366 if (ret < 0)
367 goto fail;
368
369 return get_sector_offset(bs, sector_num, 0);
370
371 fail:
372 s->free_data_block_offset -= (s->block_size + s->bitmap_size);
373 return -1;
374 }
375
376 static int vpc_read(BlockDriverState *bs, int64_t sector_num,
377 uint8_t *buf, int nb_sectors)
378 {
379 BDRVVPCState *s = bs->opaque;
380 int ret;
381 int64_t offset;
382
383 while (nb_sectors > 0) {
384 offset = get_sector_offset(bs, sector_num, 0);
385
386 if (offset == -1) {
387 memset(buf, 0, 512);
388 } else {
389 ret = bdrv_pread(s->hd, offset, buf, 512);
390 if (ret != 512)
391 return -1;
392 }
393
394 nb_sectors--;
395 sector_num++;
396 buf += 512;
397 }
398 return 0;
399 }
400
401 static int vpc_write(BlockDriverState *bs, int64_t sector_num,
402 const uint8_t *buf, int nb_sectors)
403 {
404 BDRVVPCState *s = bs->opaque;
405 int64_t offset;
406 int ret;
407
408 while (nb_sectors > 0) {
409 offset = get_sector_offset(bs, sector_num, 1);
410
411 if (offset == -1) {
412 offset = alloc_block(bs, sector_num);
413 if (offset < 0)
414 return -1;
415 }
416
417 ret = bdrv_pwrite(s->hd, offset, buf, 512);
418 if (ret != 512)
419 return -1;
420
421 nb_sectors--;
422 sector_num++;
423 buf += 512;
424 }
425
426 return 0;
427 }
428
429
430 /*
431 * Calculates the number of cylinders, heads and sectors per cylinder
432 * based on a given number of sectors. This is the algorithm described
433 * in the VHD specification.
434 *
435 * Note that the geometry doesn't always exactly match total_sectors but
436 * may round it down.
437 *
438 * Returns 0 on success, -EFBIG if the size is larger than 127 GB
439 */
440 static int calculate_geometry(int64_t total_sectors, uint16_t* cyls,
441 uint8_t* heads, uint8_t* secs_per_cyl)
442 {
443 uint32_t cyls_times_heads;
444
445 if (total_sectors > 65535 * 16 * 255)
446 return -EFBIG;
447
448 if (total_sectors > 65535 * 16 * 63) {
449 *secs_per_cyl = 255;
450 *heads = 16;
451 cyls_times_heads = total_sectors / *secs_per_cyl;
452 } else {
453 *secs_per_cyl = 17;
454 cyls_times_heads = total_sectors / *secs_per_cyl;
455 *heads = (cyls_times_heads + 1023) / 1024;
456
457 if (*heads < 4)
458 *heads = 4;
459
460 if (cyls_times_heads >= (*heads * 1024) || *heads > 16) {
461 *secs_per_cyl = 31;
462 *heads = 16;
463 cyls_times_heads = total_sectors / *secs_per_cyl;
464 }
465
466 if (cyls_times_heads >= (*heads * 1024)) {
467 *secs_per_cyl = 63;
468 *heads = 16;
469 cyls_times_heads = total_sectors / *secs_per_cyl;
470 }
471 }
472
473 // Note: Rounding up deviates from the Virtual PC behaviour
474 // However, we need this to avoid truncating images in qemu-img convert
475 *cyls = (cyls_times_heads + *heads - 1) / *heads;
476
477 return 0;
478 }
479
480 static int vpc_create(const char *filename, QEMUOptionParameter *options)
481 {
482 uint8_t buf[1024];
483 struct vhd_footer* footer = (struct vhd_footer*) buf;
484 struct vhd_dyndisk_header* dyndisk_header =
485 (struct vhd_dyndisk_header*) buf;
486 int fd, i;
487 uint16_t cyls;
488 uint8_t heads;
489 uint8_t secs_per_cyl;
490 size_t block_size, num_bat_entries;
491 int64_t total_sectors = 0;
492
493 // Read out options
494 while (options && options->name) {
495 if (!strcmp(options->name, "size")) {
496 total_sectors = options->value.n / 512;
497 }
498 options++;
499 }
500
501 // Create the file
502 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
503 if (fd < 0)
504 return -EIO;
505
506 // Calculate matching total_size and geometry
507 if (calculate_geometry(total_sectors, &cyls, &heads, &secs_per_cyl))
508 return -EFBIG;
509 total_sectors = (int64_t) cyls * heads * secs_per_cyl;
510
511 // Prepare the Hard Disk Footer
512 memset(buf, 0, 1024);
513
514 memcpy(footer->creator, "conectix", 8);
515 // TODO Check if "qemu" creator_app is ok for VPC
516 memcpy(footer->creator_app, "qemu", 4);
517 memcpy(footer->creator_os, "Wi2k", 4);
518
519 footer->features = be32_to_cpu(0x02);
520 footer->version = be32_to_cpu(0x00010000);
521 footer->data_offset = be64_to_cpu(HEADER_SIZE);
522 footer->timestamp = be32_to_cpu(time(NULL) - VHD_TIMESTAMP_BASE);
523
524 // Version of Virtual PC 2007
525 footer->major = be16_to_cpu(0x0005);
526 footer->minor =be16_to_cpu(0x0003);
527
528 footer->orig_size = be64_to_cpu(total_sectors * 512);
529 footer->size = be64_to_cpu(total_sectors * 512);
530
531 footer->cyls = be16_to_cpu(cyls);
532 footer->heads = heads;
533 footer->secs_per_cyl = secs_per_cyl;
534
535 footer->type = be32_to_cpu(VHD_DYNAMIC);
536
537 // TODO uuid is missing
538
539 footer->checksum = be32_to_cpu(vpc_checksum(buf, HEADER_SIZE));
540
541 // Write the footer (twice: at the beginning and at the end)
542 block_size = 0x200000;
543 num_bat_entries = (total_sectors + block_size / 512) / (block_size / 512);
544
545 if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE)
546 return -EIO;
547
548 if (lseek(fd, 1536 + ((num_bat_entries * 4 + 511) & ~511), SEEK_SET) < 0)
549 return -EIO;
550 if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE)
551 return -EIO;
552
553 // Write the initial BAT
554 if (lseek(fd, 3 * 512, SEEK_SET) < 0)
555 return -EIO;
556
557 memset(buf, 0xFF, 512);
558 for (i = 0; i < (num_bat_entries * 4 + 511) / 512; i++)
559 if (write(fd, buf, 512) != 512)
560 return -EIO;
561
562
563 // Prepare the Dynamic Disk Header
564 memset(buf, 0, 1024);
565
566 memcpy(dyndisk_header->magic, "cxsparse", 8);
567
568 dyndisk_header->data_offset = be64_to_cpu(0xFFFFFFFF);
569 dyndisk_header->table_offset = be64_to_cpu(3 * 512);
570 dyndisk_header->version = be32_to_cpu(0x00010000);
571 dyndisk_header->block_size = be32_to_cpu(block_size);
572 dyndisk_header->max_table_entries = be32_to_cpu(num_bat_entries);
573
574 dyndisk_header->checksum = be32_to_cpu(vpc_checksum(buf, 1024));
575
576 // Write the header
577 if (lseek(fd, 512, SEEK_SET) < 0)
578 return -EIO;
579 if (write(fd, buf, 1024) != 1024)
580 return -EIO;
581
582 close(fd);
583 return 0;
584 }
585
586 static void vpc_close(BlockDriverState *bs)
587 {
588 BDRVVPCState *s = bs->opaque;
589 qemu_free(s->pagetable);
590 #ifdef CACHE
591 qemu_free(s->pageentry_u8);
592 #endif
593 bdrv_delete(s->hd);
594 }
595
596 static QEMUOptionParameter vpc_create_options[] = {
597 {
598 .name = BLOCK_OPT_SIZE,
599 .type = OPT_SIZE,
600 .help = "Virtual disk size"
601 },
602 { NULL }
603 };
604
605 static BlockDriver bdrv_vpc = {
606 .format_name = "vpc",
607 .instance_size = sizeof(BDRVVPCState),
608 .bdrv_probe = vpc_probe,
609 .bdrv_open = vpc_open,
610 .bdrv_read = vpc_read,
611 .bdrv_write = vpc_write,
612 .bdrv_close = vpc_close,
613 .bdrv_create = vpc_create,
614
615 .create_options = vpc_create_options,
616 };
617
618 static void bdrv_vpc_init(void)
619 {
620 bdrv_register(&bdrv_vpc);
621 }
622
623 block_init(bdrv_vpc_init);