2 * Block driver for Connectix / Microsoft Virtual PC images
4 * Copyright (c) 2005 Alex Beregszaszi
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
25 #include "qemu-common.h"
26 #include "block_int.h"
28 #include "migration.h"
29 #if defined(CONFIG_UUID)
30 #include <uuid/uuid.h>
33 /**************************************************************/
35 #define HEADER_SIZE 512
45 // Seconds since Jan 1, 2000 0:00:00 (UTC)
46 #define VHD_TIMESTAMP_BASE 946684800
50 char creator[8]; // "conectix"
54 // Offset of next header structure, 0xFFFFFFFF if none
57 // Seconds since Jan 1, 2000 0:00:00 (UTC)
60 char creator_app[4]; // "vpc "
63 char creator_os[4]; // "Wi2k"
74 // Checksum of the Hard Disk Footer ("one's complement of the sum of all
75 // the bytes in the footer without the checksum field")
78 // UUID used to identify a parent hard disk (backing file)
81 uint8_t in_saved_state;
84 struct vhd_dyndisk_header {
85 char magic[8]; // "cxsparse"
87 // Offset of next header structure, 0xFFFFFFFF if none
90 // Offset of the Block Allocation Table (BAT)
91 uint64_t table_offset;
94 uint32_t max_table_entries; // 32bit/entry
96 // 2 MB by default, must be a power of two
100 uint8_t parent_uuid[16];
101 uint32_t parent_timestamp;
104 // Backing file name (in UTF-16)
105 uint8_t parent_name[512];
110 uint32_t data_length;
112 uint64_t data_offset;
116 typedef struct BDRVVPCState {
118 uint8_t footer_buf[HEADER_SIZE];
119 uint64_t free_data_block_offset;
120 int max_table_entries;
123 uint64_t last_bitmap_offset;
126 uint32_t bitmap_size;
129 uint8_t *pageentry_u8;
130 uint32_t *pageentry_u32;
131 uint16_t *pageentry_u16;
133 uint64_t last_bitmap;
136 Error *migration_blocker;
139 static uint32_t vpc_checksum(uint8_t* buf, size_t size)
144 for (i = 0; i < size; i++)
151 static int vpc_probe(const uint8_t *buf, int buf_size, const char *filename)
153 if (buf_size >= 8 && !strncmp((char *)buf, "conectix", 8))
158 static int vpc_open(BlockDriverState *bs, int flags)
160 BDRVVPCState *s = bs->opaque;
162 struct vhd_footer* footer;
163 struct vhd_dyndisk_header* dyndisk_header;
164 uint8_t buf[HEADER_SIZE];
167 int disk_type = VHD_DYNAMIC;
169 if (bdrv_pread(bs->file, 0, s->footer_buf, HEADER_SIZE) != HEADER_SIZE)
172 footer = (struct vhd_footer*) s->footer_buf;
173 if (strncmp(footer->creator, "conectix", 8)) {
174 int64_t offset = bdrv_getlength(bs->file);
175 if (offset < HEADER_SIZE) {
178 /* If a fixed disk, the footer is found only at the end of the file */
179 if (bdrv_pread(bs->file, offset-HEADER_SIZE, s->footer_buf, HEADER_SIZE)
183 if (strncmp(footer->creator, "conectix", 8)) {
186 disk_type = VHD_FIXED;
189 checksum = be32_to_cpu(footer->checksum);
190 footer->checksum = 0;
191 if (vpc_checksum(s->footer_buf, HEADER_SIZE) != checksum)
192 fprintf(stderr, "block-vpc: The header checksum of '%s' is "
193 "incorrect.\n", bs->filename);
195 /* Write 'checksum' back to footer, or else will leave it with zero. */
196 footer->checksum = be32_to_cpu(checksum);
198 // The visible size of a image in Virtual PC depends on the geometry
199 // rather than on the size stored in the footer (the size in the footer
200 // is too large usually)
201 bs->total_sectors = (int64_t)
202 be16_to_cpu(footer->cyls) * footer->heads * footer->secs_per_cyl;
204 if (bs->total_sectors >= 65535 * 16 * 255) {
209 if (disk_type == VHD_DYNAMIC) {
210 if (bdrv_pread(bs->file, be64_to_cpu(footer->data_offset), buf,
211 HEADER_SIZE) != HEADER_SIZE) {
215 dyndisk_header = (struct vhd_dyndisk_header *) buf;
217 if (strncmp(dyndisk_header->magic, "cxsparse", 8)) {
221 s->block_size = be32_to_cpu(dyndisk_header->block_size);
222 s->bitmap_size = ((s->block_size / (8 * 512)) + 511) & ~511;
224 s->max_table_entries = be32_to_cpu(dyndisk_header->max_table_entries);
225 s->pagetable = g_malloc(s->max_table_entries * 4);
227 s->bat_offset = be64_to_cpu(dyndisk_header->table_offset);
228 if (bdrv_pread(bs->file, s->bat_offset, s->pagetable,
229 s->max_table_entries * 4) != s->max_table_entries * 4) {
233 s->free_data_block_offset =
234 (s->bat_offset + (s->max_table_entries * 4) + 511) & ~511;
236 for (i = 0; i < s->max_table_entries; i++) {
237 be32_to_cpus(&s->pagetable[i]);
238 if (s->pagetable[i] != 0xFFFFFFFF) {
239 int64_t next = (512 * (int64_t) s->pagetable[i]) +
240 s->bitmap_size + s->block_size;
242 if (next > s->free_data_block_offset) {
243 s->free_data_block_offset = next;
248 s->last_bitmap_offset = (int64_t) -1;
251 s->pageentry_u8 = g_malloc(512);
252 s->pageentry_u32 = s->pageentry_u8;
253 s->pageentry_u16 = s->pageentry_u8;
254 s->last_pagetable = -1;
258 qemu_co_mutex_init(&s->lock);
260 /* Disable migration when VHD images are used */
261 error_set(&s->migration_blocker,
262 QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
263 "vpc", bs->device_name, "live migration");
264 migrate_add_blocker(s->migration_blocker);
271 static int vpc_reopen_prepare(BDRVReopenState *state,
272 BlockReopenQueue *queue, Error **errp)
278 * Returns the absolute byte offset of the given sector in the image file.
279 * If the sector is not allocated, -1 is returned instead.
281 * The parameter write must be 1 if the offset will be used for a write
282 * operation (the block bitmaps is updated then), 0 otherwise.
284 static inline int64_t get_sector_offset(BlockDriverState *bs,
285 int64_t sector_num, int write)
287 BDRVVPCState *s = bs->opaque;
288 uint64_t offset = sector_num * 512;
289 uint64_t bitmap_offset, block_offset;
290 uint32_t pagetable_index, pageentry_index;
292 pagetable_index = offset / s->block_size;
293 pageentry_index = (offset % s->block_size) / 512;
295 if (pagetable_index >= s->max_table_entries || s->pagetable[pagetable_index] == 0xffffffff)
296 return -1; // not allocated
298 bitmap_offset = 512 * (uint64_t) s->pagetable[pagetable_index];
299 block_offset = bitmap_offset + s->bitmap_size + (512 * pageentry_index);
301 // We must ensure that we don't write to any sectors which are marked as
302 // unused in the bitmap. We get away with setting all bits in the block
303 // bitmap each time we write to a new block. This might cause Virtual PC to
304 // miss sparse read optimization, but it's not a problem in terms of
306 if (write && (s->last_bitmap_offset != bitmap_offset)) {
307 uint8_t bitmap[s->bitmap_size];
309 s->last_bitmap_offset = bitmap_offset;
310 memset(bitmap, 0xff, s->bitmap_size);
311 bdrv_pwrite_sync(bs->file, bitmap_offset, bitmap, s->bitmap_size);
314 // printf("sector: %" PRIx64 ", index: %x, offset: %x, bioff: %" PRIx64 ", bloff: %" PRIx64 "\n",
315 // sector_num, pagetable_index, pageentry_index,
316 // bitmap_offset, block_offset);
318 // disabled by reason
321 if (bitmap_offset != s->last_bitmap)
323 lseek(s->fd, bitmap_offset, SEEK_SET);
325 s->last_bitmap = bitmap_offset;
327 // Scary! Bitmap is stored as big endian 32bit entries,
328 // while we used to look it up byte by byte
329 read(s->fd, s->pageentry_u8, 512);
330 for (i = 0; i < 128; i++)
331 be32_to_cpus(&s->pageentry_u32[i]);
334 if ((s->pageentry_u8[pageentry_index / 8] >> (pageentry_index % 8)) & 1)
337 lseek(s->fd, bitmap_offset + (pageentry_index / 8), SEEK_SET);
339 read(s->fd, &bitmap_entry, 1);
341 if ((bitmap_entry >> (pageentry_index % 8)) & 1)
342 return -1; // not allocated
350 * Writes the footer to the end of the image file. This is needed when the
351 * file grows as it overwrites the old footer
353 * Returns 0 on success and < 0 on error
355 static int rewrite_footer(BlockDriverState* bs)
358 BDRVVPCState *s = bs->opaque;
359 int64_t offset = s->free_data_block_offset;
361 ret = bdrv_pwrite_sync(bs->file, offset, s->footer_buf, HEADER_SIZE);
369 * Allocates a new block. This involves writing a new footer and updating
370 * the Block Allocation Table to use the space at the old end of the image
371 * file (overwriting the old footer)
373 * Returns the sectors' offset in the image file on success and < 0 on error
375 static int64_t alloc_block(BlockDriverState* bs, int64_t sector_num)
377 BDRVVPCState *s = bs->opaque;
379 uint32_t index, bat_value;
381 uint8_t bitmap[s->bitmap_size];
383 // Check if sector_num is valid
384 if ((sector_num < 0) || (sector_num > bs->total_sectors))
387 // Write entry into in-memory BAT
388 index = (sector_num * 512) / s->block_size;
389 if (s->pagetable[index] != 0xFFFFFFFF)
392 s->pagetable[index] = s->free_data_block_offset / 512;
394 // Initialize the block's bitmap
395 memset(bitmap, 0xff, s->bitmap_size);
396 ret = bdrv_pwrite_sync(bs->file, s->free_data_block_offset, bitmap,
402 // Write new footer (the old one will be overwritten)
403 s->free_data_block_offset += s->block_size + s->bitmap_size;
404 ret = rewrite_footer(bs);
408 // Write BAT entry to disk
409 bat_offset = s->bat_offset + (4 * index);
410 bat_value = be32_to_cpu(s->pagetable[index]);
411 ret = bdrv_pwrite_sync(bs->file, bat_offset, &bat_value, 4);
415 return get_sector_offset(bs, sector_num, 0);
418 s->free_data_block_offset -= (s->block_size + s->bitmap_size);
422 static int vpc_read(BlockDriverState *bs, int64_t sector_num,
423 uint8_t *buf, int nb_sectors)
425 BDRVVPCState *s = bs->opaque;
428 int64_t sectors, sectors_per_block;
429 struct vhd_footer *footer = (struct vhd_footer *) s->footer_buf;
431 if (cpu_to_be32(footer->type) == VHD_FIXED) {
432 return bdrv_read(bs->file, sector_num, buf, nb_sectors);
434 while (nb_sectors > 0) {
435 offset = get_sector_offset(bs, sector_num, 0);
437 sectors_per_block = s->block_size >> BDRV_SECTOR_BITS;
438 sectors = sectors_per_block - (sector_num % sectors_per_block);
439 if (sectors > nb_sectors) {
440 sectors = nb_sectors;
444 memset(buf, 0, sectors * BDRV_SECTOR_SIZE);
446 ret = bdrv_pread(bs->file, offset, buf,
447 sectors * BDRV_SECTOR_SIZE);
448 if (ret != sectors * BDRV_SECTOR_SIZE) {
453 nb_sectors -= sectors;
454 sector_num += sectors;
455 buf += sectors * BDRV_SECTOR_SIZE;
460 static coroutine_fn int vpc_co_read(BlockDriverState *bs, int64_t sector_num,
461 uint8_t *buf, int nb_sectors)
464 BDRVVPCState *s = bs->opaque;
465 qemu_co_mutex_lock(&s->lock);
466 ret = vpc_read(bs, sector_num, buf, nb_sectors);
467 qemu_co_mutex_unlock(&s->lock);
471 static int vpc_write(BlockDriverState *bs, int64_t sector_num,
472 const uint8_t *buf, int nb_sectors)
474 BDRVVPCState *s = bs->opaque;
476 int64_t sectors, sectors_per_block;
478 struct vhd_footer *footer = (struct vhd_footer *) s->footer_buf;
480 if (cpu_to_be32(footer->type) == VHD_FIXED) {
481 return bdrv_write(bs->file, sector_num, buf, nb_sectors);
483 while (nb_sectors > 0) {
484 offset = get_sector_offset(bs, sector_num, 1);
486 sectors_per_block = s->block_size >> BDRV_SECTOR_BITS;
487 sectors = sectors_per_block - (sector_num % sectors_per_block);
488 if (sectors > nb_sectors) {
489 sectors = nb_sectors;
493 offset = alloc_block(bs, sector_num);
498 ret = bdrv_pwrite(bs->file, offset, buf, sectors * BDRV_SECTOR_SIZE);
499 if (ret != sectors * BDRV_SECTOR_SIZE) {
503 nb_sectors -= sectors;
504 sector_num += sectors;
505 buf += sectors * BDRV_SECTOR_SIZE;
511 static coroutine_fn int vpc_co_write(BlockDriverState *bs, int64_t sector_num,
512 const uint8_t *buf, int nb_sectors)
515 BDRVVPCState *s = bs->opaque;
516 qemu_co_mutex_lock(&s->lock);
517 ret = vpc_write(bs, sector_num, buf, nb_sectors);
518 qemu_co_mutex_unlock(&s->lock);
523 * Calculates the number of cylinders, heads and sectors per cylinder
524 * based on a given number of sectors. This is the algorithm described
525 * in the VHD specification.
527 * Note that the geometry doesn't always exactly match total_sectors but
530 * Returns 0 on success, -EFBIG if the size is larger than 127 GB
532 static int calculate_geometry(int64_t total_sectors, uint16_t* cyls,
533 uint8_t* heads, uint8_t* secs_per_cyl)
535 uint32_t cyls_times_heads;
537 if (total_sectors > 65535 * 16 * 255)
540 if (total_sectors > 65535 * 16 * 63) {
543 cyls_times_heads = total_sectors / *secs_per_cyl;
546 cyls_times_heads = total_sectors / *secs_per_cyl;
547 *heads = (cyls_times_heads + 1023) / 1024;
552 if (cyls_times_heads >= (*heads * 1024) || *heads > 16) {
555 cyls_times_heads = total_sectors / *secs_per_cyl;
558 if (cyls_times_heads >= (*heads * 1024)) {
561 cyls_times_heads = total_sectors / *secs_per_cyl;
565 *cyls = cyls_times_heads / *heads;
570 static int create_dynamic_disk(int fd, uint8_t *buf, int64_t total_sectors)
572 struct vhd_dyndisk_header* dyndisk_header =
573 (struct vhd_dyndisk_header*) buf;
574 size_t block_size, num_bat_entries;
578 // Write the footer (twice: at the beginning and at the end)
579 block_size = 0x200000;
580 num_bat_entries = (total_sectors + block_size / 512) / (block_size / 512);
582 if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE) {
586 if (lseek(fd, 1536 + ((num_bat_entries * 4 + 511) & ~511), SEEK_SET) < 0) {
589 if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE) {
593 // Write the initial BAT
594 if (lseek(fd, 3 * 512, SEEK_SET) < 0) {
598 memset(buf, 0xFF, 512);
599 for (i = 0; i < (num_bat_entries * 4 + 511) / 512; i++) {
600 if (write(fd, buf, 512) != 512) {
605 // Prepare the Dynamic Disk Header
606 memset(buf, 0, 1024);
608 memcpy(dyndisk_header->magic, "cxsparse", 8);
611 * Note: The spec is actually wrong here for data_offset, it says
612 * 0xFFFFFFFF, but MS tools expect all 64 bits to be set.
614 dyndisk_header->data_offset = be64_to_cpu(0xFFFFFFFFFFFFFFFFULL);
615 dyndisk_header->table_offset = be64_to_cpu(3 * 512);
616 dyndisk_header->version = be32_to_cpu(0x00010000);
617 dyndisk_header->block_size = be32_to_cpu(block_size);
618 dyndisk_header->max_table_entries = be32_to_cpu(num_bat_entries);
620 dyndisk_header->checksum = be32_to_cpu(vpc_checksum(buf, 1024));
623 if (lseek(fd, 512, SEEK_SET) < 0) {
627 if (write(fd, buf, 1024) != 1024) {
636 static int create_fixed_disk(int fd, uint8_t *buf, int64_t total_size)
640 /* Add footer to total size */
642 if (ftruncate(fd, total_size) != 0) {
646 if (lseek(fd, -512, SEEK_END) < 0) {
649 if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE) {
659 static int vpc_create(const char *filename, QEMUOptionParameter *options)
662 struct vhd_footer *footer = (struct vhd_footer *) buf;
663 QEMUOptionParameter *disk_type_param;
667 uint8_t secs_per_cyl = 0;
668 int64_t total_sectors;
673 /* Read out options */
674 total_size = get_option_parameter(options, BLOCK_OPT_SIZE)->value.n;
676 disk_type_param = get_option_parameter(options, BLOCK_OPT_SUBFMT);
677 if (disk_type_param && disk_type_param->value.s) {
678 if (!strcmp(disk_type_param->value.s, "dynamic")) {
679 disk_type = VHD_DYNAMIC;
680 } else if (!strcmp(disk_type_param->value.s, "fixed")) {
681 disk_type = VHD_FIXED;
686 disk_type = VHD_DYNAMIC;
689 /* Create the file */
690 fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
696 * Calculate matching total_size and geometry. Increase the number of
697 * sectors requested until we get enough (or fail). This ensures that
698 * qemu-img convert doesn't truncate images, but rather rounds up.
700 total_sectors = total_size / BDRV_SECTOR_SIZE;
701 for (i = 0; total_sectors > (int64_t)cyls * heads * secs_per_cyl; i++) {
702 if (calculate_geometry(total_sectors + i, &cyls, &heads,
710 total_sectors = (int64_t) cyls * heads * secs_per_cyl;
712 /* Prepare the Hard Disk Footer */
713 memset(buf, 0, 1024);
715 memcpy(footer->creator, "conectix", 8);
716 /* TODO Check if "qemu" creator_app is ok for VPC */
717 memcpy(footer->creator_app, "qemu", 4);
718 memcpy(footer->creator_os, "Wi2k", 4);
720 footer->features = be32_to_cpu(0x02);
721 footer->version = be32_to_cpu(0x00010000);
722 if (disk_type == VHD_DYNAMIC) {
723 footer->data_offset = be64_to_cpu(HEADER_SIZE);
725 footer->data_offset = be64_to_cpu(0xFFFFFFFFFFFFFFFFULL);
727 footer->timestamp = be32_to_cpu(time(NULL) - VHD_TIMESTAMP_BASE);
729 /* Version of Virtual PC 2007 */
730 footer->major = be16_to_cpu(0x0005);
731 footer->minor = be16_to_cpu(0x0003);
732 if (disk_type == VHD_DYNAMIC) {
733 footer->orig_size = be64_to_cpu(total_sectors * 512);
734 footer->size = be64_to_cpu(total_sectors * 512);
736 footer->orig_size = be64_to_cpu(total_size);
737 footer->size = be64_to_cpu(total_size);
739 footer->cyls = be16_to_cpu(cyls);
740 footer->heads = heads;
741 footer->secs_per_cyl = secs_per_cyl;
743 footer->type = be32_to_cpu(disk_type);
745 #if defined(CONFIG_UUID)
746 uuid_generate(footer->uuid);
749 footer->checksum = be32_to_cpu(vpc_checksum(buf, HEADER_SIZE));
751 if (disk_type == VHD_DYNAMIC) {
752 ret = create_dynamic_disk(fd, buf, total_sectors);
754 ret = create_fixed_disk(fd, buf, total_size);
762 static void vpc_close(BlockDriverState *bs)
764 BDRVVPCState *s = bs->opaque;
765 g_free(s->pagetable);
767 g_free(s->pageentry_u8);
770 migrate_del_blocker(s->migration_blocker);
771 error_free(s->migration_blocker);
774 static QEMUOptionParameter vpc_create_options[] = {
776 .name = BLOCK_OPT_SIZE,
778 .help = "Virtual disk size"
781 .name = BLOCK_OPT_SUBFMT,
784 "Type of virtual hard disk format. Supported formats are "
785 "{dynamic (default) | fixed} "
790 static BlockDriver bdrv_vpc = {
791 .format_name = "vpc",
792 .instance_size = sizeof(BDRVVPCState),
794 .bdrv_probe = vpc_probe,
795 .bdrv_open = vpc_open,
796 .bdrv_close = vpc_close,
797 .bdrv_reopen_prepare = vpc_reopen_prepare,
798 .bdrv_create = vpc_create,
800 .bdrv_read = vpc_co_read,
801 .bdrv_write = vpc_co_write,
803 .create_options = vpc_create_options,
806 static void bdrv_vpc_init(void)
808 bdrv_register(&bdrv_vpc);
811 block_init(bdrv_vpc_init);