xref: /openbmc/qemu/block/vpc.c (revision 1de7afc9)
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/block_int.h"
27 #include "qemu/module.h"
28 #include "migration/migration.h"
29 #if defined(CONFIG_UUID)
30 #include <uuid/uuid.h>
31 #endif
32 
33 /**************************************************************/
34 
35 #define HEADER_SIZE 512
36 
37 //#define CACHE
38 
39 enum vhd_type {
40     VHD_FIXED           = 2,
41     VHD_DYNAMIC         = 3,
42     VHD_DIFFERENCING    = 4,
43 };
44 
45 // Seconds since Jan 1, 2000 0:00:00 (UTC)
46 #define VHD_TIMESTAMP_BASE 946684800
47 
48 // always big-endian
49 struct vhd_footer {
50     char        creator[8]; // "conectix"
51     uint32_t    features;
52     uint32_t    version;
53 
54     // Offset of next header structure, 0xFFFFFFFF if none
55     uint64_t    data_offset;
56 
57     // Seconds since Jan 1, 2000 0:00:00 (UTC)
58     uint32_t    timestamp;
59 
60     char        creator_app[4]; // "vpc "
61     uint16_t    major;
62     uint16_t    minor;
63     char        creator_os[4]; // "Wi2k"
64 
65     uint64_t    orig_size;
66     uint64_t    size;
67 
68     uint16_t    cyls;
69     uint8_t     heads;
70     uint8_t     secs_per_cyl;
71 
72     uint32_t    type;
73 
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")
76     uint32_t    checksum;
77 
78     // UUID used to identify a parent hard disk (backing file)
79     uint8_t     uuid[16];
80 
81     uint8_t     in_saved_state;
82 };
83 
84 struct vhd_dyndisk_header {
85     char        magic[8]; // "cxsparse"
86 
87     // Offset of next header structure, 0xFFFFFFFF if none
88     uint64_t    data_offset;
89 
90     // Offset of the Block Allocation Table (BAT)
91     uint64_t    table_offset;
92 
93     uint32_t    version;
94     uint32_t    max_table_entries; // 32bit/entry
95 
96     // 2 MB by default, must be a power of two
97     uint32_t    block_size;
98 
99     uint32_t    checksum;
100     uint8_t     parent_uuid[16];
101     uint32_t    parent_timestamp;
102     uint32_t    reserved;
103 
104     // Backing file name (in UTF-16)
105     uint8_t     parent_name[512];
106 
107     struct {
108         uint32_t    platform;
109         uint32_t    data_space;
110         uint32_t    data_length;
111         uint32_t    reserved;
112         uint64_t    data_offset;
113     } parent_locator[8];
114 };
115 
116 typedef struct BDRVVPCState {
117     CoMutex lock;
118     uint8_t footer_buf[HEADER_SIZE];
119     uint64_t free_data_block_offset;
120     int max_table_entries;
121     uint32_t *pagetable;
122     uint64_t bat_offset;
123     uint64_t last_bitmap_offset;
124 
125     uint32_t block_size;
126     uint32_t bitmap_size;
127 
128 #ifdef CACHE
129     uint8_t *pageentry_u8;
130     uint32_t *pageentry_u32;
131     uint16_t *pageentry_u16;
132 
133     uint64_t last_bitmap;
134 #endif
135 
136     Error *migration_blocker;
137 } BDRVVPCState;
138 
139 static uint32_t vpc_checksum(uint8_t* buf, size_t size)
140 {
141     uint32_t res = 0;
142     int i;
143 
144     for (i = 0; i < size; i++)
145         res += buf[i];
146 
147     return ~res;
148 }
149 
150 
151 static int vpc_probe(const uint8_t *buf, int buf_size, const char *filename)
152 {
153     if (buf_size >= 8 && !strncmp((char *)buf, "conectix", 8))
154 	return 100;
155     return 0;
156 }
157 
158 static int vpc_open(BlockDriverState *bs, int flags)
159 {
160     BDRVVPCState *s = bs->opaque;
161     int i;
162     struct vhd_footer* footer;
163     struct vhd_dyndisk_header* dyndisk_header;
164     uint8_t buf[HEADER_SIZE];
165     uint32_t checksum;
166     int err = -1;
167     int disk_type = VHD_DYNAMIC;
168 
169     if (bdrv_pread(bs->file, 0, s->footer_buf, HEADER_SIZE) != HEADER_SIZE)
170         goto fail;
171 
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) {
176             goto fail;
177         }
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)
180                 != HEADER_SIZE) {
181             goto fail;
182         }
183         if (strncmp(footer->creator, "conectix", 8)) {
184             goto fail;
185         }
186         disk_type = VHD_FIXED;
187     }
188 
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);
194 
195     /* Write 'checksum' back to footer, or else will leave it with zero. */
196     footer->checksum = be32_to_cpu(checksum);
197 
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;
203 
204     /* Allow a maximum disk size of approximately 2 TB */
205     if (bs->total_sectors >= 65535LL * 255 * 255) {
206         err = -EFBIG;
207         goto fail;
208     }
209 
210     if (disk_type == VHD_DYNAMIC) {
211         if (bdrv_pread(bs->file, be64_to_cpu(footer->data_offset), buf,
212                 HEADER_SIZE) != HEADER_SIZE) {
213             goto fail;
214         }
215 
216         dyndisk_header = (struct vhd_dyndisk_header *) buf;
217 
218         if (strncmp(dyndisk_header->magic, "cxsparse", 8)) {
219             goto fail;
220         }
221 
222         s->block_size = be32_to_cpu(dyndisk_header->block_size);
223         s->bitmap_size = ((s->block_size / (8 * 512)) + 511) & ~511;
224 
225         s->max_table_entries = be32_to_cpu(dyndisk_header->max_table_entries);
226         s->pagetable = g_malloc(s->max_table_entries * 4);
227 
228         s->bat_offset = be64_to_cpu(dyndisk_header->table_offset);
229         if (bdrv_pread(bs->file, s->bat_offset, s->pagetable,
230                 s->max_table_entries * 4) != s->max_table_entries * 4) {
231             goto fail;
232         }
233 
234         s->free_data_block_offset =
235             (s->bat_offset + (s->max_table_entries * 4) + 511) & ~511;
236 
237         for (i = 0; i < s->max_table_entries; i++) {
238             be32_to_cpus(&s->pagetable[i]);
239             if (s->pagetable[i] != 0xFFFFFFFF) {
240                 int64_t next = (512 * (int64_t) s->pagetable[i]) +
241                     s->bitmap_size + s->block_size;
242 
243                 if (next > s->free_data_block_offset) {
244                     s->free_data_block_offset = next;
245                 }
246             }
247         }
248 
249         s->last_bitmap_offset = (int64_t) -1;
250 
251 #ifdef CACHE
252         s->pageentry_u8 = g_malloc(512);
253         s->pageentry_u32 = s->pageentry_u8;
254         s->pageentry_u16 = s->pageentry_u8;
255         s->last_pagetable = -1;
256 #endif
257     }
258 
259     qemu_co_mutex_init(&s->lock);
260 
261     /* Disable migration when VHD images are used */
262     error_set(&s->migration_blocker,
263               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
264               "vpc", bs->device_name, "live migration");
265     migrate_add_blocker(s->migration_blocker);
266 
267     return 0;
268  fail:
269     return err;
270 }
271 
272 static int vpc_reopen_prepare(BDRVReopenState *state,
273                               BlockReopenQueue *queue, Error **errp)
274 {
275     return 0;
276 }
277 
278 /*
279  * Returns the absolute byte offset of the given sector in the image file.
280  * If the sector is not allocated, -1 is returned instead.
281  *
282  * The parameter write must be 1 if the offset will be used for a write
283  * operation (the block bitmaps is updated then), 0 otherwise.
284  */
285 static inline int64_t get_sector_offset(BlockDriverState *bs,
286     int64_t sector_num, int write)
287 {
288     BDRVVPCState *s = bs->opaque;
289     uint64_t offset = sector_num * 512;
290     uint64_t bitmap_offset, block_offset;
291     uint32_t pagetable_index, pageentry_index;
292 
293     pagetable_index = offset / s->block_size;
294     pageentry_index = (offset % s->block_size) / 512;
295 
296     if (pagetable_index >= s->max_table_entries || s->pagetable[pagetable_index] == 0xffffffff)
297         return -1; // not allocated
298 
299     bitmap_offset = 512 * (uint64_t) s->pagetable[pagetable_index];
300     block_offset = bitmap_offset + s->bitmap_size + (512 * pageentry_index);
301 
302     // We must ensure that we don't write to any sectors which are marked as
303     // unused in the bitmap. We get away with setting all bits in the block
304     // bitmap each time we write to a new block. This might cause Virtual PC to
305     // miss sparse read optimization, but it's not a problem in terms of
306     // correctness.
307     if (write && (s->last_bitmap_offset != bitmap_offset)) {
308         uint8_t bitmap[s->bitmap_size];
309 
310         s->last_bitmap_offset = bitmap_offset;
311         memset(bitmap, 0xff, s->bitmap_size);
312         bdrv_pwrite_sync(bs->file, bitmap_offset, bitmap, s->bitmap_size);
313     }
314 
315 //    printf("sector: %" PRIx64 ", index: %x, offset: %x, bioff: %" PRIx64 ", bloff: %" PRIx64 "\n",
316 //	sector_num, pagetable_index, pageentry_index,
317 //	bitmap_offset, block_offset);
318 
319 // disabled by reason
320 #if 0
321 #ifdef CACHE
322     if (bitmap_offset != s->last_bitmap)
323     {
324 	lseek(s->fd, bitmap_offset, SEEK_SET);
325 
326 	s->last_bitmap = bitmap_offset;
327 
328 	// Scary! Bitmap is stored as big endian 32bit entries,
329 	// while we used to look it up byte by byte
330 	read(s->fd, s->pageentry_u8, 512);
331 	for (i = 0; i < 128; i++)
332 	    be32_to_cpus(&s->pageentry_u32[i]);
333     }
334 
335     if ((s->pageentry_u8[pageentry_index / 8] >> (pageentry_index % 8)) & 1)
336 	return -1;
337 #else
338     lseek(s->fd, bitmap_offset + (pageentry_index / 8), SEEK_SET);
339 
340     read(s->fd, &bitmap_entry, 1);
341 
342     if ((bitmap_entry >> (pageentry_index % 8)) & 1)
343 	return -1; // not allocated
344 #endif
345 #endif
346 
347     return block_offset;
348 }
349 
350 /*
351  * Writes the footer to the end of the image file. This is needed when the
352  * file grows as it overwrites the old footer
353  *
354  * Returns 0 on success and < 0 on error
355  */
356 static int rewrite_footer(BlockDriverState* bs)
357 {
358     int ret;
359     BDRVVPCState *s = bs->opaque;
360     int64_t offset = s->free_data_block_offset;
361 
362     ret = bdrv_pwrite_sync(bs->file, offset, s->footer_buf, HEADER_SIZE);
363     if (ret < 0)
364         return ret;
365 
366     return 0;
367 }
368 
369 /*
370  * Allocates a new block. This involves writing a new footer and updating
371  * the Block Allocation Table to use the space at the old end of the image
372  * file (overwriting the old footer)
373  *
374  * Returns the sectors' offset in the image file on success and < 0 on error
375  */
376 static int64_t alloc_block(BlockDriverState* bs, int64_t sector_num)
377 {
378     BDRVVPCState *s = bs->opaque;
379     int64_t bat_offset;
380     uint32_t index, bat_value;
381     int ret;
382     uint8_t bitmap[s->bitmap_size];
383 
384     // Check if sector_num is valid
385     if ((sector_num < 0) || (sector_num > bs->total_sectors))
386         return -1;
387 
388     // Write entry into in-memory BAT
389     index = (sector_num * 512) / s->block_size;
390     if (s->pagetable[index] != 0xFFFFFFFF)
391         return -1;
392 
393     s->pagetable[index] = s->free_data_block_offset / 512;
394 
395     // Initialize the block's bitmap
396     memset(bitmap, 0xff, s->bitmap_size);
397     ret = bdrv_pwrite_sync(bs->file, s->free_data_block_offset, bitmap,
398         s->bitmap_size);
399     if (ret < 0) {
400         return ret;
401     }
402 
403     // Write new footer (the old one will be overwritten)
404     s->free_data_block_offset += s->block_size + s->bitmap_size;
405     ret = rewrite_footer(bs);
406     if (ret < 0)
407         goto fail;
408 
409     // Write BAT entry to disk
410     bat_offset = s->bat_offset + (4 * index);
411     bat_value = be32_to_cpu(s->pagetable[index]);
412     ret = bdrv_pwrite_sync(bs->file, bat_offset, &bat_value, 4);
413     if (ret < 0)
414         goto fail;
415 
416     return get_sector_offset(bs, sector_num, 0);
417 
418 fail:
419     s->free_data_block_offset -= (s->block_size + s->bitmap_size);
420     return -1;
421 }
422 
423 static int vpc_read(BlockDriverState *bs, int64_t sector_num,
424                     uint8_t *buf, int nb_sectors)
425 {
426     BDRVVPCState *s = bs->opaque;
427     int ret;
428     int64_t offset;
429     int64_t sectors, sectors_per_block;
430     struct vhd_footer *footer = (struct vhd_footer *) s->footer_buf;
431 
432     if (cpu_to_be32(footer->type) == VHD_FIXED) {
433         return bdrv_read(bs->file, sector_num, buf, nb_sectors);
434     }
435     while (nb_sectors > 0) {
436         offset = get_sector_offset(bs, sector_num, 0);
437 
438         sectors_per_block = s->block_size >> BDRV_SECTOR_BITS;
439         sectors = sectors_per_block - (sector_num % sectors_per_block);
440         if (sectors > nb_sectors) {
441             sectors = nb_sectors;
442         }
443 
444         if (offset == -1) {
445             memset(buf, 0, sectors * BDRV_SECTOR_SIZE);
446         } else {
447             ret = bdrv_pread(bs->file, offset, buf,
448                 sectors * BDRV_SECTOR_SIZE);
449             if (ret != sectors * BDRV_SECTOR_SIZE) {
450                 return -1;
451             }
452         }
453 
454         nb_sectors -= sectors;
455         sector_num += sectors;
456         buf += sectors * BDRV_SECTOR_SIZE;
457     }
458     return 0;
459 }
460 
461 static coroutine_fn int vpc_co_read(BlockDriverState *bs, int64_t sector_num,
462                                     uint8_t *buf, int nb_sectors)
463 {
464     int ret;
465     BDRVVPCState *s = bs->opaque;
466     qemu_co_mutex_lock(&s->lock);
467     ret = vpc_read(bs, sector_num, buf, nb_sectors);
468     qemu_co_mutex_unlock(&s->lock);
469     return ret;
470 }
471 
472 static int vpc_write(BlockDriverState *bs, int64_t sector_num,
473     const uint8_t *buf, int nb_sectors)
474 {
475     BDRVVPCState *s = bs->opaque;
476     int64_t offset;
477     int64_t sectors, sectors_per_block;
478     int ret;
479     struct vhd_footer *footer =  (struct vhd_footer *) s->footer_buf;
480 
481     if (cpu_to_be32(footer->type) == VHD_FIXED) {
482         return bdrv_write(bs->file, sector_num, buf, nb_sectors);
483     }
484     while (nb_sectors > 0) {
485         offset = get_sector_offset(bs, sector_num, 1);
486 
487         sectors_per_block = s->block_size >> BDRV_SECTOR_BITS;
488         sectors = sectors_per_block - (sector_num % sectors_per_block);
489         if (sectors > nb_sectors) {
490             sectors = nb_sectors;
491         }
492 
493         if (offset == -1) {
494             offset = alloc_block(bs, sector_num);
495             if (offset < 0)
496                 return -1;
497         }
498 
499         ret = bdrv_pwrite(bs->file, offset, buf, sectors * BDRV_SECTOR_SIZE);
500         if (ret != sectors * BDRV_SECTOR_SIZE) {
501             return -1;
502         }
503 
504         nb_sectors -= sectors;
505         sector_num += sectors;
506         buf += sectors * BDRV_SECTOR_SIZE;
507     }
508 
509     return 0;
510 }
511 
512 static coroutine_fn int vpc_co_write(BlockDriverState *bs, int64_t sector_num,
513                                      const uint8_t *buf, int nb_sectors)
514 {
515     int ret;
516     BDRVVPCState *s = bs->opaque;
517     qemu_co_mutex_lock(&s->lock);
518     ret = vpc_write(bs, sector_num, buf, nb_sectors);
519     qemu_co_mutex_unlock(&s->lock);
520     return ret;
521 }
522 
523 /*
524  * Calculates the number of cylinders, heads and sectors per cylinder
525  * based on a given number of sectors. This is the algorithm described
526  * in the VHD specification.
527  *
528  * Note that the geometry doesn't always exactly match total_sectors but
529  * may round it down.
530  *
531  * Returns 0 on success, -EFBIG if the size is larger than ~2 TB. Override
532  * the hardware EIDE and ATA-2 limit of 16 heads (max disk size of 127 GB)
533  * and instead allow up to 255 heads.
534  */
535 static int calculate_geometry(int64_t total_sectors, uint16_t* cyls,
536     uint8_t* heads, uint8_t* secs_per_cyl)
537 {
538     uint32_t cyls_times_heads;
539 
540     /* Allow a maximum disk size of approximately 2 TB */
541     if (total_sectors > 65535LL * 255 * 255) {
542         return -EFBIG;
543     }
544 
545     if (total_sectors > 65535 * 16 * 63) {
546         *secs_per_cyl = 255;
547         if (total_sectors > 65535 * 16 * 255) {
548             *heads = 255;
549         } else {
550             *heads = 16;
551         }
552         cyls_times_heads = total_sectors / *secs_per_cyl;
553     } else {
554         *secs_per_cyl = 17;
555         cyls_times_heads = total_sectors / *secs_per_cyl;
556         *heads = (cyls_times_heads + 1023) / 1024;
557 
558         if (*heads < 4)
559             *heads = 4;
560 
561         if (cyls_times_heads >= (*heads * 1024) || *heads > 16) {
562             *secs_per_cyl = 31;
563             *heads = 16;
564             cyls_times_heads = total_sectors / *secs_per_cyl;
565         }
566 
567         if (cyls_times_heads >= (*heads * 1024)) {
568             *secs_per_cyl = 63;
569             *heads = 16;
570             cyls_times_heads = total_sectors / *secs_per_cyl;
571         }
572     }
573 
574     *cyls = cyls_times_heads / *heads;
575 
576     return 0;
577 }
578 
579 static int create_dynamic_disk(int fd, uint8_t *buf, int64_t total_sectors)
580 {
581     struct vhd_dyndisk_header* dyndisk_header =
582         (struct vhd_dyndisk_header*) buf;
583     size_t block_size, num_bat_entries;
584     int i;
585     int ret = -EIO;
586 
587     // Write the footer (twice: at the beginning and at the end)
588     block_size = 0x200000;
589     num_bat_entries = (total_sectors + block_size / 512) / (block_size / 512);
590 
591     if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE) {
592         goto fail;
593     }
594 
595     if (lseek(fd, 1536 + ((num_bat_entries * 4 + 511) & ~511), SEEK_SET) < 0) {
596         goto fail;
597     }
598     if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE) {
599         goto fail;
600     }
601 
602     // Write the initial BAT
603     if (lseek(fd, 3 * 512, SEEK_SET) < 0) {
604         goto fail;
605     }
606 
607     memset(buf, 0xFF, 512);
608     for (i = 0; i < (num_bat_entries * 4 + 511) / 512; i++) {
609         if (write(fd, buf, 512) != 512) {
610             goto fail;
611         }
612     }
613 
614     // Prepare the Dynamic Disk Header
615     memset(buf, 0, 1024);
616 
617     memcpy(dyndisk_header->magic, "cxsparse", 8);
618 
619     /*
620      * Note: The spec is actually wrong here for data_offset, it says
621      * 0xFFFFFFFF, but MS tools expect all 64 bits to be set.
622      */
623     dyndisk_header->data_offset = be64_to_cpu(0xFFFFFFFFFFFFFFFFULL);
624     dyndisk_header->table_offset = be64_to_cpu(3 * 512);
625     dyndisk_header->version = be32_to_cpu(0x00010000);
626     dyndisk_header->block_size = be32_to_cpu(block_size);
627     dyndisk_header->max_table_entries = be32_to_cpu(num_bat_entries);
628 
629     dyndisk_header->checksum = be32_to_cpu(vpc_checksum(buf, 1024));
630 
631     // Write the header
632     if (lseek(fd, 512, SEEK_SET) < 0) {
633         goto fail;
634     }
635 
636     if (write(fd, buf, 1024) != 1024) {
637         goto fail;
638     }
639     ret = 0;
640 
641  fail:
642     return ret;
643 }
644 
645 static int create_fixed_disk(int fd, uint8_t *buf, int64_t total_size)
646 {
647     int ret = -EIO;
648 
649     /* Add footer to total size */
650     total_size += 512;
651     if (ftruncate(fd, total_size) != 0) {
652         ret = -errno;
653         goto fail;
654     }
655     if (lseek(fd, -512, SEEK_END) < 0) {
656         goto fail;
657     }
658     if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE) {
659         goto fail;
660     }
661 
662     ret = 0;
663 
664  fail:
665     return ret;
666 }
667 
668 static int vpc_create(const char *filename, QEMUOptionParameter *options)
669 {
670     uint8_t buf[1024];
671     struct vhd_footer *footer = (struct vhd_footer *) buf;
672     QEMUOptionParameter *disk_type_param;
673     int fd, i;
674     uint16_t cyls = 0;
675     uint8_t heads = 0;
676     uint8_t secs_per_cyl = 0;
677     int64_t total_sectors;
678     int64_t total_size;
679     int disk_type;
680     int ret = -EIO;
681 
682     /* Read out options */
683     total_size = get_option_parameter(options, BLOCK_OPT_SIZE)->value.n;
684 
685     disk_type_param = get_option_parameter(options, BLOCK_OPT_SUBFMT);
686     if (disk_type_param && disk_type_param->value.s) {
687         if (!strcmp(disk_type_param->value.s, "dynamic")) {
688             disk_type = VHD_DYNAMIC;
689         } else if (!strcmp(disk_type_param->value.s, "fixed")) {
690             disk_type = VHD_FIXED;
691         } else {
692             return -EINVAL;
693         }
694     } else {
695         disk_type = VHD_DYNAMIC;
696     }
697 
698     /* Create the file */
699     fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
700     if (fd < 0) {
701         return -EIO;
702     }
703 
704     /*
705      * Calculate matching total_size and geometry. Increase the number of
706      * sectors requested until we get enough (or fail). This ensures that
707      * qemu-img convert doesn't truncate images, but rather rounds up.
708      */
709     total_sectors = total_size / BDRV_SECTOR_SIZE;
710     for (i = 0; total_sectors > (int64_t)cyls * heads * secs_per_cyl; i++) {
711         if (calculate_geometry(total_sectors + i, &cyls, &heads,
712                                &secs_per_cyl))
713         {
714             ret = -EFBIG;
715             goto fail;
716         }
717     }
718 
719     total_sectors = (int64_t) cyls * heads * secs_per_cyl;
720 
721     /* Prepare the Hard Disk Footer */
722     memset(buf, 0, 1024);
723 
724     memcpy(footer->creator, "conectix", 8);
725     /* TODO Check if "qemu" creator_app is ok for VPC */
726     memcpy(footer->creator_app, "qemu", 4);
727     memcpy(footer->creator_os, "Wi2k", 4);
728 
729     footer->features = be32_to_cpu(0x02);
730     footer->version = be32_to_cpu(0x00010000);
731     if (disk_type == VHD_DYNAMIC) {
732         footer->data_offset = be64_to_cpu(HEADER_SIZE);
733     } else {
734         footer->data_offset = be64_to_cpu(0xFFFFFFFFFFFFFFFFULL);
735     }
736     footer->timestamp = be32_to_cpu(time(NULL) - VHD_TIMESTAMP_BASE);
737 
738     /* Version of Virtual PC 2007 */
739     footer->major = be16_to_cpu(0x0005);
740     footer->minor = be16_to_cpu(0x0003);
741     if (disk_type == VHD_DYNAMIC) {
742         footer->orig_size = be64_to_cpu(total_sectors * 512);
743         footer->size = be64_to_cpu(total_sectors * 512);
744     } else {
745         footer->orig_size = be64_to_cpu(total_size);
746         footer->size = be64_to_cpu(total_size);
747     }
748     footer->cyls = be16_to_cpu(cyls);
749     footer->heads = heads;
750     footer->secs_per_cyl = secs_per_cyl;
751 
752     footer->type = be32_to_cpu(disk_type);
753 
754 #if defined(CONFIG_UUID)
755     uuid_generate(footer->uuid);
756 #endif
757 
758     footer->checksum = be32_to_cpu(vpc_checksum(buf, HEADER_SIZE));
759 
760     if (disk_type == VHD_DYNAMIC) {
761         ret = create_dynamic_disk(fd, buf, total_sectors);
762     } else {
763         ret = create_fixed_disk(fd, buf, total_size);
764     }
765 
766  fail:
767     qemu_close(fd);
768     return ret;
769 }
770 
771 static void vpc_close(BlockDriverState *bs)
772 {
773     BDRVVPCState *s = bs->opaque;
774     g_free(s->pagetable);
775 #ifdef CACHE
776     g_free(s->pageentry_u8);
777 #endif
778 
779     migrate_del_blocker(s->migration_blocker);
780     error_free(s->migration_blocker);
781 }
782 
783 static QEMUOptionParameter vpc_create_options[] = {
784     {
785         .name = BLOCK_OPT_SIZE,
786         .type = OPT_SIZE,
787         .help = "Virtual disk size"
788     },
789     {
790         .name = BLOCK_OPT_SUBFMT,
791         .type = OPT_STRING,
792         .help =
793             "Type of virtual hard disk format. Supported formats are "
794             "{dynamic (default) | fixed} "
795     },
796     { NULL }
797 };
798 
799 static BlockDriver bdrv_vpc = {
800     .format_name    = "vpc",
801     .instance_size  = sizeof(BDRVVPCState),
802 
803     .bdrv_probe     = vpc_probe,
804     .bdrv_open      = vpc_open,
805     .bdrv_close     = vpc_close,
806     .bdrv_reopen_prepare = vpc_reopen_prepare,
807     .bdrv_create    = vpc_create,
808 
809     .bdrv_read              = vpc_co_read,
810     .bdrv_write             = vpc_co_write,
811 
812     .create_options = vpc_create_options,
813 };
814 
815 static void bdrv_vpc_init(void)
816 {
817     bdrv_register(&bdrv_vpc);
818 }
819 
820 block_init(bdrv_vpc_init);
821