block: add a CoMutex to synchronous read drivers
[qemu.git] / block / vpc.c
blob9155ee15e39f2a797417b5cbfd87a3e6d21a8cee
1 /*
2 * Block driver for Connectix / Microsoft Virtual PC images
4 * Copyright (c) 2005 Alex Beregszaszi
5 * Copyright (c) 2009 Kevin Wolf <kwolf@suse.de>
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:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
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.
25 #include "qemu-common.h"
26 #include "block_int.h"
27 #include "module.h"
29 /**************************************************************/
31 #define HEADER_SIZE 512
33 //#define CACHE
35 enum vhd_type {
36 VHD_FIXED = 2,
37 VHD_DYNAMIC = 3,
38 VHD_DIFFERENCING = 4,
41 // Seconds since Jan 1, 2000 0:00:00 (UTC)
42 #define VHD_TIMESTAMP_BASE 946684800
44 // always big-endian
45 struct vhd_footer {
46 char creator[8]; // "conectix"
47 uint32_t features;
48 uint32_t version;
50 // Offset of next header structure, 0xFFFFFFFF if none
51 uint64_t data_offset;
53 // Seconds since Jan 1, 2000 0:00:00 (UTC)
54 uint32_t timestamp;
56 char creator_app[4]; // "vpc "
57 uint16_t major;
58 uint16_t minor;
59 char creator_os[4]; // "Wi2k"
61 uint64_t orig_size;
62 uint64_t size;
64 uint16_t cyls;
65 uint8_t heads;
66 uint8_t secs_per_cyl;
68 uint32_t type;
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;
74 // UUID used to identify a parent hard disk (backing file)
75 uint8_t uuid[16];
77 uint8_t in_saved_state;
80 struct vhd_dyndisk_header {
81 char magic[8]; // "cxsparse"
83 // Offset of next header structure, 0xFFFFFFFF if none
84 uint64_t data_offset;
86 // Offset of the Block Allocation Table (BAT)
87 uint64_t table_offset;
89 uint32_t version;
90 uint32_t max_table_entries; // 32bit/entry
92 // 2 MB by default, must be a power of two
93 uint32_t block_size;
95 uint32_t checksum;
96 uint8_t parent_uuid[16];
97 uint32_t parent_timestamp;
98 uint32_t reserved;
100 // Backing file name (in UTF-16)
101 uint8_t parent_name[512];
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];
112 typedef struct BDRVVPCState {
113 CoMutex lock;
114 uint8_t footer_buf[HEADER_SIZE];
115 uint64_t free_data_block_offset;
116 int max_table_entries;
117 uint32_t *pagetable;
118 uint64_t bat_offset;
119 uint64_t last_bitmap_offset;
121 uint32_t block_size;
122 uint32_t bitmap_size;
124 #ifdef CACHE
125 uint8_t *pageentry_u8;
126 uint32_t *pageentry_u32;
127 uint16_t *pageentry_u16;
129 uint64_t last_bitmap;
130 #endif
131 } BDRVVPCState;
133 static uint32_t vpc_checksum(uint8_t* buf, size_t size)
135 uint32_t res = 0;
136 int i;
138 for (i = 0; i < size; i++)
139 res += buf[i];
141 return ~res;
145 static int vpc_probe(const uint8_t *buf, int buf_size, const char *filename)
147 if (buf_size >= 8 && !strncmp((char *)buf, "conectix", 8))
148 return 100;
149 return 0;
152 static int vpc_open(BlockDriverState *bs, int flags)
154 BDRVVPCState *s = bs->opaque;
155 int i;
156 struct vhd_footer* footer;
157 struct vhd_dyndisk_header* dyndisk_header;
158 uint8_t buf[HEADER_SIZE];
159 uint32_t checksum;
160 int err = -1;
162 if (bdrv_pread(bs->file, 0, s->footer_buf, HEADER_SIZE) != HEADER_SIZE)
163 goto fail;
165 footer = (struct vhd_footer*) s->footer_buf;
166 if (strncmp(footer->creator, "conectix", 8))
167 goto fail;
169 checksum = be32_to_cpu(footer->checksum);
170 footer->checksum = 0;
171 if (vpc_checksum(s->footer_buf, HEADER_SIZE) != checksum)
172 fprintf(stderr, "block-vpc: The header checksum of '%s' is "
173 "incorrect.\n", bs->filename);
175 // The visible size of a image in Virtual PC depends on the geometry
176 // rather than on the size stored in the footer (the size in the footer
177 // is too large usually)
178 bs->total_sectors = (int64_t)
179 be16_to_cpu(footer->cyls) * footer->heads * footer->secs_per_cyl;
181 if (bs->total_sectors >= 65535 * 16 * 255) {
182 err = -EFBIG;
183 goto fail;
186 if (bdrv_pread(bs->file, be64_to_cpu(footer->data_offset), buf, HEADER_SIZE)
187 != HEADER_SIZE)
188 goto fail;
190 dyndisk_header = (struct vhd_dyndisk_header*) buf;
192 if (strncmp(dyndisk_header->magic, "cxsparse", 8))
193 goto fail;
196 s->block_size = be32_to_cpu(dyndisk_header->block_size);
197 s->bitmap_size = ((s->block_size / (8 * 512)) + 511) & ~511;
199 s->max_table_entries = be32_to_cpu(dyndisk_header->max_table_entries);
200 s->pagetable = g_malloc(s->max_table_entries * 4);
202 s->bat_offset = be64_to_cpu(dyndisk_header->table_offset);
203 if (bdrv_pread(bs->file, s->bat_offset, s->pagetable,
204 s->max_table_entries * 4) != s->max_table_entries * 4)
205 goto fail;
207 s->free_data_block_offset =
208 (s->bat_offset + (s->max_table_entries * 4) + 511) & ~511;
210 for (i = 0; i < s->max_table_entries; i++) {
211 be32_to_cpus(&s->pagetable[i]);
212 if (s->pagetable[i] != 0xFFFFFFFF) {
213 int64_t next = (512 * (int64_t) s->pagetable[i]) +
214 s->bitmap_size + s->block_size;
216 if (next> s->free_data_block_offset)
217 s->free_data_block_offset = next;
221 s->last_bitmap_offset = (int64_t) -1;
223 #ifdef CACHE
224 s->pageentry_u8 = g_malloc(512);
225 s->pageentry_u32 = s->pageentry_u8;
226 s->pageentry_u16 = s->pageentry_u8;
227 s->last_pagetable = -1;
228 #endif
230 qemu_co_mutex_init(&s->lock);
231 return 0;
232 fail:
233 return err;
237 * Returns the absolute byte offset of the given sector in the image file.
238 * If the sector is not allocated, -1 is returned instead.
240 * The parameter write must be 1 if the offset will be used for a write
241 * operation (the block bitmaps is updated then), 0 otherwise.
243 static inline int64_t get_sector_offset(BlockDriverState *bs,
244 int64_t sector_num, int write)
246 BDRVVPCState *s = bs->opaque;
247 uint64_t offset = sector_num * 512;
248 uint64_t bitmap_offset, block_offset;
249 uint32_t pagetable_index, pageentry_index;
251 pagetable_index = offset / s->block_size;
252 pageentry_index = (offset % s->block_size) / 512;
254 if (pagetable_index >= s->max_table_entries || s->pagetable[pagetable_index] == 0xffffffff)
255 return -1; // not allocated
257 bitmap_offset = 512 * (uint64_t) s->pagetable[pagetable_index];
258 block_offset = bitmap_offset + s->bitmap_size + (512 * pageentry_index);
260 // We must ensure that we don't write to any sectors which are marked as
261 // unused in the bitmap. We get away with setting all bits in the block
262 // bitmap each time we write to a new block. This might cause Virtual PC to
263 // miss sparse read optimization, but it's not a problem in terms of
264 // correctness.
265 if (write && (s->last_bitmap_offset != bitmap_offset)) {
266 uint8_t bitmap[s->bitmap_size];
268 s->last_bitmap_offset = bitmap_offset;
269 memset(bitmap, 0xff, s->bitmap_size);
270 bdrv_pwrite_sync(bs->file, bitmap_offset, bitmap, s->bitmap_size);
273 // printf("sector: %" PRIx64 ", index: %x, offset: %x, bioff: %" PRIx64 ", bloff: %" PRIx64 "\n",
274 // sector_num, pagetable_index, pageentry_index,
275 // bitmap_offset, block_offset);
277 // disabled by reason
278 #if 0
279 #ifdef CACHE
280 if (bitmap_offset != s->last_bitmap)
282 lseek(s->fd, bitmap_offset, SEEK_SET);
284 s->last_bitmap = bitmap_offset;
286 // Scary! Bitmap is stored as big endian 32bit entries,
287 // while we used to look it up byte by byte
288 read(s->fd, s->pageentry_u8, 512);
289 for (i = 0; i < 128; i++)
290 be32_to_cpus(&s->pageentry_u32[i]);
293 if ((s->pageentry_u8[pageentry_index / 8] >> (pageentry_index % 8)) & 1)
294 return -1;
295 #else
296 lseek(s->fd, bitmap_offset + (pageentry_index / 8), SEEK_SET);
298 read(s->fd, &bitmap_entry, 1);
300 if ((bitmap_entry >> (pageentry_index % 8)) & 1)
301 return -1; // not allocated
302 #endif
303 #endif
305 return block_offset;
309 * Writes the footer to the end of the image file. This is needed when the
310 * file grows as it overwrites the old footer
312 * Returns 0 on success and < 0 on error
314 static int rewrite_footer(BlockDriverState* bs)
316 int ret;
317 BDRVVPCState *s = bs->opaque;
318 int64_t offset = s->free_data_block_offset;
320 ret = bdrv_pwrite_sync(bs->file, offset, s->footer_buf, HEADER_SIZE);
321 if (ret < 0)
322 return ret;
324 return 0;
328 * Allocates a new block. This involves writing a new footer and updating
329 * the Block Allocation Table to use the space at the old end of the image
330 * file (overwriting the old footer)
332 * Returns the sectors' offset in the image file on success and < 0 on error
334 static int64_t alloc_block(BlockDriverState* bs, int64_t sector_num)
336 BDRVVPCState *s = bs->opaque;
337 int64_t bat_offset;
338 uint32_t index, bat_value;
339 int ret;
340 uint8_t bitmap[s->bitmap_size];
342 // Check if sector_num is valid
343 if ((sector_num < 0) || (sector_num > bs->total_sectors))
344 return -1;
346 // Write entry into in-memory BAT
347 index = (sector_num * 512) / s->block_size;
348 if (s->pagetable[index] != 0xFFFFFFFF)
349 return -1;
351 s->pagetable[index] = s->free_data_block_offset / 512;
353 // Initialize the block's bitmap
354 memset(bitmap, 0xff, s->bitmap_size);
355 bdrv_pwrite_sync(bs->file, s->free_data_block_offset, bitmap,
356 s->bitmap_size);
358 // Write new footer (the old one will be overwritten)
359 s->free_data_block_offset += s->block_size + s->bitmap_size;
360 ret = rewrite_footer(bs);
361 if (ret < 0)
362 goto fail;
364 // Write BAT entry to disk
365 bat_offset = s->bat_offset + (4 * index);
366 bat_value = be32_to_cpu(s->pagetable[index]);
367 ret = bdrv_pwrite_sync(bs->file, bat_offset, &bat_value, 4);
368 if (ret < 0)
369 goto fail;
371 return get_sector_offset(bs, sector_num, 0);
373 fail:
374 s->free_data_block_offset -= (s->block_size + s->bitmap_size);
375 return -1;
378 static int vpc_read(BlockDriverState *bs, int64_t sector_num,
379 uint8_t *buf, int nb_sectors)
381 BDRVVPCState *s = bs->opaque;
382 int ret;
383 int64_t offset;
384 int64_t sectors, sectors_per_block;
386 while (nb_sectors > 0) {
387 offset = get_sector_offset(bs, sector_num, 0);
389 sectors_per_block = s->block_size >> BDRV_SECTOR_BITS;
390 sectors = sectors_per_block - (sector_num % sectors_per_block);
391 if (sectors > nb_sectors) {
392 sectors = nb_sectors;
395 if (offset == -1) {
396 memset(buf, 0, sectors * BDRV_SECTOR_SIZE);
397 } else {
398 ret = bdrv_pread(bs->file, offset, buf,
399 sectors * BDRV_SECTOR_SIZE);
400 if (ret != sectors * BDRV_SECTOR_SIZE) {
401 return -1;
405 nb_sectors -= sectors;
406 sector_num += sectors;
407 buf += sectors * BDRV_SECTOR_SIZE;
409 return 0;
412 static int vpc_write(BlockDriverState *bs, int64_t sector_num,
413 const uint8_t *buf, int nb_sectors)
415 BDRVVPCState *s = bs->opaque;
416 int64_t offset;
417 int64_t sectors, sectors_per_block;
418 int ret;
420 while (nb_sectors > 0) {
421 offset = get_sector_offset(bs, sector_num, 1);
423 sectors_per_block = s->block_size >> BDRV_SECTOR_BITS;
424 sectors = sectors_per_block - (sector_num % sectors_per_block);
425 if (sectors > nb_sectors) {
426 sectors = nb_sectors;
429 if (offset == -1) {
430 offset = alloc_block(bs, sector_num);
431 if (offset < 0)
432 return -1;
435 ret = bdrv_pwrite(bs->file, offset, buf, sectors * BDRV_SECTOR_SIZE);
436 if (ret != sectors * BDRV_SECTOR_SIZE) {
437 return -1;
440 nb_sectors -= sectors;
441 sector_num += sectors;
442 buf += sectors * BDRV_SECTOR_SIZE;
445 return 0;
448 static int vpc_flush(BlockDriverState *bs)
450 return bdrv_flush(bs->file);
454 * Calculates the number of cylinders, heads and sectors per cylinder
455 * based on a given number of sectors. This is the algorithm described
456 * in the VHD specification.
458 * Note that the geometry doesn't always exactly match total_sectors but
459 * may round it down.
461 * Returns 0 on success, -EFBIG if the size is larger than 127 GB
463 static int calculate_geometry(int64_t total_sectors, uint16_t* cyls,
464 uint8_t* heads, uint8_t* secs_per_cyl)
466 uint32_t cyls_times_heads;
468 if (total_sectors > 65535 * 16 * 255)
469 return -EFBIG;
471 if (total_sectors > 65535 * 16 * 63) {
472 *secs_per_cyl = 255;
473 *heads = 16;
474 cyls_times_heads = total_sectors / *secs_per_cyl;
475 } else {
476 *secs_per_cyl = 17;
477 cyls_times_heads = total_sectors / *secs_per_cyl;
478 *heads = (cyls_times_heads + 1023) / 1024;
480 if (*heads < 4)
481 *heads = 4;
483 if (cyls_times_heads >= (*heads * 1024) || *heads > 16) {
484 *secs_per_cyl = 31;
485 *heads = 16;
486 cyls_times_heads = total_sectors / *secs_per_cyl;
489 if (cyls_times_heads >= (*heads * 1024)) {
490 *secs_per_cyl = 63;
491 *heads = 16;
492 cyls_times_heads = total_sectors / *secs_per_cyl;
496 *cyls = cyls_times_heads / *heads;
498 return 0;
501 static int vpc_create(const char *filename, QEMUOptionParameter *options)
503 uint8_t buf[1024];
504 struct vhd_footer* footer = (struct vhd_footer*) buf;
505 struct vhd_dyndisk_header* dyndisk_header =
506 (struct vhd_dyndisk_header*) buf;
507 int fd, i;
508 uint16_t cyls = 0;
509 uint8_t heads = 0;
510 uint8_t secs_per_cyl = 0;
511 size_t block_size, num_bat_entries;
512 int64_t total_sectors = 0;
513 int ret = -EIO;
515 // Read out options
516 total_sectors = get_option_parameter(options, BLOCK_OPT_SIZE)->value.n /
517 BDRV_SECTOR_SIZE;
519 // Create the file
520 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
521 if (fd < 0)
522 return -EIO;
524 /* Calculate matching total_size and geometry. Increase the number of
525 sectors requested until we get enough (or fail). */
526 for (i = 0; total_sectors > (int64_t)cyls * heads * secs_per_cyl; i++) {
527 if (calculate_geometry(total_sectors + i,
528 &cyls, &heads, &secs_per_cyl)) {
529 ret = -EFBIG;
530 goto fail;
533 total_sectors = (int64_t) cyls * heads * secs_per_cyl;
535 // Prepare the Hard Disk Footer
536 memset(buf, 0, 1024);
538 memcpy(footer->creator, "conectix", 8);
539 // TODO Check if "qemu" creator_app is ok for VPC
540 memcpy(footer->creator_app, "qemu", 4);
541 memcpy(footer->creator_os, "Wi2k", 4);
543 footer->features = be32_to_cpu(0x02);
544 footer->version = be32_to_cpu(0x00010000);
545 footer->data_offset = be64_to_cpu(HEADER_SIZE);
546 footer->timestamp = be32_to_cpu(time(NULL) - VHD_TIMESTAMP_BASE);
548 // Version of Virtual PC 2007
549 footer->major = be16_to_cpu(0x0005);
550 footer->minor =be16_to_cpu(0x0003);
552 footer->orig_size = be64_to_cpu(total_sectors * 512);
553 footer->size = be64_to_cpu(total_sectors * 512);
555 footer->cyls = be16_to_cpu(cyls);
556 footer->heads = heads;
557 footer->secs_per_cyl = secs_per_cyl;
559 footer->type = be32_to_cpu(VHD_DYNAMIC);
561 // TODO uuid is missing
563 footer->checksum = be32_to_cpu(vpc_checksum(buf, HEADER_SIZE));
565 // Write the footer (twice: at the beginning and at the end)
566 block_size = 0x200000;
567 num_bat_entries = (total_sectors + block_size / 512) / (block_size / 512);
569 if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE) {
570 goto fail;
573 if (lseek(fd, 1536 + ((num_bat_entries * 4 + 511) & ~511), SEEK_SET) < 0) {
574 goto fail;
576 if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE) {
577 goto fail;
580 // Write the initial BAT
581 if (lseek(fd, 3 * 512, SEEK_SET) < 0) {
582 goto fail;
585 memset(buf, 0xFF, 512);
586 for (i = 0; i < (num_bat_entries * 4 + 511) / 512; i++) {
587 if (write(fd, buf, 512) != 512) {
588 goto fail;
593 // Prepare the Dynamic Disk Header
594 memset(buf, 0, 1024);
596 memcpy(dyndisk_header->magic, "cxsparse", 8);
598 dyndisk_header->data_offset = be64_to_cpu(0xFFFFFFFF);
599 dyndisk_header->table_offset = be64_to_cpu(3 * 512);
600 dyndisk_header->version = be32_to_cpu(0x00010000);
601 dyndisk_header->block_size = be32_to_cpu(block_size);
602 dyndisk_header->max_table_entries = be32_to_cpu(num_bat_entries);
604 dyndisk_header->checksum = be32_to_cpu(vpc_checksum(buf, 1024));
606 // Write the header
607 if (lseek(fd, 512, SEEK_SET) < 0) {
608 goto fail;
611 if (write(fd, buf, 1024) != 1024) {
612 goto fail;
614 ret = 0;
616 fail:
617 close(fd);
618 return ret;
621 static void vpc_close(BlockDriverState *bs)
623 BDRVVPCState *s = bs->opaque;
624 g_free(s->pagetable);
625 #ifdef CACHE
626 g_free(s->pageentry_u8);
627 #endif
630 static QEMUOptionParameter vpc_create_options[] = {
632 .name = BLOCK_OPT_SIZE,
633 .type = OPT_SIZE,
634 .help = "Virtual disk size"
636 { NULL }
639 static BlockDriver bdrv_vpc = {
640 .format_name = "vpc",
641 .instance_size = sizeof(BDRVVPCState),
642 .bdrv_probe = vpc_probe,
643 .bdrv_open = vpc_open,
644 .bdrv_read = vpc_read,
645 .bdrv_write = vpc_write,
646 .bdrv_flush = vpc_flush,
647 .bdrv_close = vpc_close,
648 .bdrv_create = vpc_create,
650 .create_options = vpc_create_options,
653 static void bdrv_vpc_init(void)
655 bdrv_register(&bdrv_vpc);
658 block_init(bdrv_vpc_init);