xref: /illumos-kvm-cmd/block/vpc.c (revision 68396ea9)
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     uint8_t footer_buf[HEADER_SIZE];
114     uint64_t free_data_block_offset;
115     int max_table_entries;
116     uint32_t *pagetable;
117     uint64_t bat_offset;
118     uint64_t last_bitmap_offset;
119 
120     uint32_t block_size;
121     uint32_t bitmap_size;
122 
123 #ifdef CACHE
124     uint8_t *pageentry_u8;
125     uint32_t *pageentry_u32;
126     uint16_t *pageentry_u16;
127 
128     uint64_t last_bitmap;
129 #endif
130 } BDRVVPCState;
131 
vpc_checksum(uint8_t * buf,size_t size)132 static uint32_t vpc_checksum(uint8_t* buf, size_t size)
133 {
134     uint32_t res = 0;
135     int i;
136 
137     for (i = 0; i < size; i++)
138         res += buf[i];
139 
140     return ~res;
141 }
142 
143 
vpc_probe(const uint8_t * buf,int buf_size,const char * filename)144 static int vpc_probe(const uint8_t *buf, int buf_size, const char *filename)
145 {
146     if (buf_size >= 8 && !strncmp((char *)buf, "conectix", 8))
147 	return 100;
148     return 0;
149 }
150 
vpc_open(BlockDriverState * bs,int flags)151 static int vpc_open(BlockDriverState *bs, int flags)
152 {
153     BDRVVPCState *s = bs->opaque;
154     int i;
155     struct vhd_footer* footer;
156     struct vhd_dyndisk_header* dyndisk_header;
157     uint8_t buf[HEADER_SIZE];
158     uint32_t checksum;
159 
160     if (bdrv_pread(bs->file, 0, s->footer_buf, HEADER_SIZE) != HEADER_SIZE)
161         goto fail;
162 
163     footer = (struct vhd_footer*) s->footer_buf;
164     if (strncmp(footer->creator, "conectix", 8))
165         goto fail;
166 
167     checksum = be32_to_cpu(footer->checksum);
168     footer->checksum = 0;
169     if (vpc_checksum(s->footer_buf, HEADER_SIZE) != checksum)
170         fprintf(stderr, "block-vpc: The header checksum of '%s' is "
171             "incorrect.\n", bs->filename);
172 
173     // The visible size of a image in Virtual PC depends on the geometry
174     // rather than on the size stored in the footer (the size in the footer
175     // is too large usually)
176     bs->total_sectors = (int64_t)
177         be16_to_cpu(footer->cyls) * footer->heads * footer->secs_per_cyl;
178 
179     if (bdrv_pread(bs->file, be64_to_cpu(footer->data_offset), buf, HEADER_SIZE)
180             != HEADER_SIZE)
181         goto fail;
182 
183     dyndisk_header = (struct vhd_dyndisk_header*) buf;
184 
185     if (strncmp(dyndisk_header->magic, "cxsparse", 8))
186         goto fail;
187 
188 
189     s->block_size = be32_to_cpu(dyndisk_header->block_size);
190     s->bitmap_size = ((s->block_size / (8 * 512)) + 511) & ~511;
191 
192     s->max_table_entries = be32_to_cpu(dyndisk_header->max_table_entries);
193     s->pagetable = qemu_malloc(s->max_table_entries * 4);
194 
195     s->bat_offset = be64_to_cpu(dyndisk_header->table_offset);
196     if (bdrv_pread(bs->file, s->bat_offset, s->pagetable,
197             s->max_table_entries * 4) != s->max_table_entries * 4)
198 	    goto fail;
199 
200     s->free_data_block_offset =
201         (s->bat_offset + (s->max_table_entries * 4) + 511) & ~511;
202 
203     for (i = 0; i < s->max_table_entries; i++) {
204         be32_to_cpus(&s->pagetable[i]);
205         if (s->pagetable[i] != 0xFFFFFFFF) {
206             int64_t next = (512 * (int64_t) s->pagetable[i]) +
207                 s->bitmap_size + s->block_size;
208 
209             if (next> s->free_data_block_offset)
210                 s->free_data_block_offset = next;
211         }
212     }
213 
214     s->last_bitmap_offset = (int64_t) -1;
215 
216 #ifdef CACHE
217     s->pageentry_u8 = qemu_malloc(512);
218     s->pageentry_u32 = s->pageentry_u8;
219     s->pageentry_u16 = s->pageentry_u8;
220     s->last_pagetable = -1;
221 #endif
222 
223     return 0;
224  fail:
225     return -1;
226 }
227 
228 /*
229  * Returns the absolute byte offset of the given sector in the image file.
230  * If the sector is not allocated, -1 is returned instead.
231  *
232  * The parameter write must be 1 if the offset will be used for a write
233  * operation (the block bitmaps is updated then), 0 otherwise.
234  */
get_sector_offset(BlockDriverState * bs,int64_t sector_num,int write)235 static inline int64_t get_sector_offset(BlockDriverState *bs,
236     int64_t sector_num, int write)
237 {
238     BDRVVPCState *s = bs->opaque;
239     uint64_t offset = sector_num * 512;
240     uint64_t bitmap_offset, block_offset;
241     uint32_t pagetable_index, pageentry_index;
242 
243     pagetable_index = offset / s->block_size;
244     pageentry_index = (offset % s->block_size) / 512;
245 
246     if (pagetable_index >= s->max_table_entries || s->pagetable[pagetable_index] == 0xffffffff)
247         return -1; // not allocated
248 
249     bitmap_offset = 512 * (uint64_t) s->pagetable[pagetable_index];
250     block_offset = bitmap_offset + s->bitmap_size + (512 * pageentry_index);
251 
252     // We must ensure that we don't write to any sectors which are marked as
253     // unused in the bitmap. We get away with setting all bits in the block
254     // bitmap each time we write to a new block. This might cause Virtual PC to
255     // miss sparse read optimization, but it's not a problem in terms of
256     // correctness.
257     if (write && (s->last_bitmap_offset != bitmap_offset)) {
258         uint8_t bitmap[s->bitmap_size];
259 
260         s->last_bitmap_offset = bitmap_offset;
261         memset(bitmap, 0xff, s->bitmap_size);
262         bdrv_pwrite_sync(bs->file, bitmap_offset, bitmap, s->bitmap_size);
263     }
264 
265 //    printf("sector: %" PRIx64 ", index: %x, offset: %x, bioff: %" PRIx64 ", bloff: %" PRIx64 "\n",
266 //	sector_num, pagetable_index, pageentry_index,
267 //	bitmap_offset, block_offset);
268 
269 // disabled by reason
270 #if 0
271 #ifdef CACHE
272     if (bitmap_offset != s->last_bitmap)
273     {
274 	lseek(s->fd, bitmap_offset, SEEK_SET);
275 
276 	s->last_bitmap = bitmap_offset;
277 
278 	// Scary! Bitmap is stored as big endian 32bit entries,
279 	// while we used to look it up byte by byte
280 	read(s->fd, s->pageentry_u8, 512);
281 	for (i = 0; i < 128; i++)
282 	    be32_to_cpus(&s->pageentry_u32[i]);
283     }
284 
285     if ((s->pageentry_u8[pageentry_index / 8] >> (pageentry_index % 8)) & 1)
286 	return -1;
287 #else
288     lseek(s->fd, bitmap_offset + (pageentry_index / 8), SEEK_SET);
289 
290     read(s->fd, &bitmap_entry, 1);
291 
292     if ((bitmap_entry >> (pageentry_index % 8)) & 1)
293 	return -1; // not allocated
294 #endif
295 #endif
296 
297     return block_offset;
298 }
299 
300 /*
301  * Writes the footer to the end of the image file. This is needed when the
302  * file grows as it overwrites the old footer
303  *
304  * Returns 0 on success and < 0 on error
305  */
rewrite_footer(BlockDriverState * bs)306 static int rewrite_footer(BlockDriverState* bs)
307 {
308     int ret;
309     BDRVVPCState *s = bs->opaque;
310     int64_t offset = s->free_data_block_offset;
311 
312     ret = bdrv_pwrite_sync(bs->file, offset, s->footer_buf, HEADER_SIZE);
313     if (ret < 0)
314         return ret;
315 
316     return 0;
317 }
318 
319 /*
320  * Allocates a new block. This involves writing a new footer and updating
321  * the Block Allocation Table to use the space at the old end of the image
322  * file (overwriting the old footer)
323  *
324  * Returns the sectors' offset in the image file on success and < 0 on error
325  */
alloc_block(BlockDriverState * bs,int64_t sector_num)326 static int64_t alloc_block(BlockDriverState* bs, int64_t sector_num)
327 {
328     BDRVVPCState *s = bs->opaque;
329     int64_t bat_offset;
330     uint32_t index, bat_value;
331     int ret;
332     uint8_t bitmap[s->bitmap_size];
333 
334     // Check if sector_num is valid
335     if ((sector_num < 0) || (sector_num > bs->total_sectors))
336         return -1;
337 
338     // Write entry into in-memory BAT
339     index = (sector_num * 512) / s->block_size;
340     if (s->pagetable[index] != 0xFFFFFFFF)
341         return -1;
342 
343     s->pagetable[index] = s->free_data_block_offset / 512;
344 
345     // Initialize the block's bitmap
346     memset(bitmap, 0xff, s->bitmap_size);
347     bdrv_pwrite_sync(bs->file, s->free_data_block_offset, bitmap,
348         s->bitmap_size);
349 
350     // Write new footer (the old one will be overwritten)
351     s->free_data_block_offset += s->block_size + s->bitmap_size;
352     ret = rewrite_footer(bs);
353     if (ret < 0)
354         goto fail;
355 
356     // Write BAT entry to disk
357     bat_offset = s->bat_offset + (4 * index);
358     bat_value = be32_to_cpu(s->pagetable[index]);
359     ret = bdrv_pwrite_sync(bs->file, bat_offset, &bat_value, 4);
360     if (ret < 0)
361         goto fail;
362 
363     return get_sector_offset(bs, sector_num, 0);
364 
365 fail:
366     s->free_data_block_offset -= (s->block_size + s->bitmap_size);
367     return -1;
368 }
369 
vpc_read(BlockDriverState * bs,int64_t sector_num,uint8_t * buf,int nb_sectors)370 static int vpc_read(BlockDriverState *bs, int64_t sector_num,
371                     uint8_t *buf, int nb_sectors)
372 {
373     BDRVVPCState *s = bs->opaque;
374     int ret;
375     int64_t offset;
376     int64_t sectors, sectors_per_block;
377 
378     while (nb_sectors > 0) {
379         offset = get_sector_offset(bs, sector_num, 0);
380 
381         sectors_per_block = s->block_size >> BDRV_SECTOR_BITS;
382         sectors = sectors_per_block - (sector_num % sectors_per_block);
383         if (sectors > nb_sectors) {
384             sectors = nb_sectors;
385         }
386 
387         if (offset == -1) {
388             memset(buf, 0, sectors * BDRV_SECTOR_SIZE);
389         } else {
390             ret = bdrv_pread(bs->file, offset, buf,
391                 sectors * BDRV_SECTOR_SIZE);
392             if (ret != sectors * BDRV_SECTOR_SIZE) {
393                 return -1;
394             }
395         }
396 
397         nb_sectors -= sectors;
398         sector_num += sectors;
399         buf += sectors * BDRV_SECTOR_SIZE;
400     }
401     return 0;
402 }
403 
vpc_write(BlockDriverState * bs,int64_t sector_num,const uint8_t * buf,int nb_sectors)404 static int vpc_write(BlockDriverState *bs, int64_t sector_num,
405     const uint8_t *buf, int nb_sectors)
406 {
407     BDRVVPCState *s = bs->opaque;
408     int64_t offset;
409     int64_t sectors, sectors_per_block;
410     int ret;
411 
412     while (nb_sectors > 0) {
413         offset = get_sector_offset(bs, sector_num, 1);
414 
415         sectors_per_block = s->block_size >> BDRV_SECTOR_BITS;
416         sectors = sectors_per_block - (sector_num % sectors_per_block);
417         if (sectors > nb_sectors) {
418             sectors = nb_sectors;
419         }
420 
421         if (offset == -1) {
422             offset = alloc_block(bs, sector_num);
423             if (offset < 0)
424                 return -1;
425         }
426 
427         ret = bdrv_pwrite(bs->file, offset, buf, sectors * BDRV_SECTOR_SIZE);
428         if (ret != sectors * BDRV_SECTOR_SIZE) {
429             return -1;
430         }
431 
432         nb_sectors -= sectors;
433         sector_num += sectors;
434         buf += sectors * BDRV_SECTOR_SIZE;
435     }
436 
437     return 0;
438 }
439 
vpc_flush(BlockDriverState * bs)440 static int vpc_flush(BlockDriverState *bs)
441 {
442     return bdrv_flush(bs->file);
443 }
444 
445 /*
446  * Calculates the number of cylinders, heads and sectors per cylinder
447  * based on a given number of sectors. This is the algorithm described
448  * in the VHD specification.
449  *
450  * Note that the geometry doesn't always exactly match total_sectors but
451  * may round it down.
452  *
453  * Returns 0 on success, -EFBIG if the size is larger than 127 GB
454  */
calculate_geometry(int64_t total_sectors,uint16_t * cyls,uint8_t * heads,uint8_t * secs_per_cyl)455 static int calculate_geometry(int64_t total_sectors, uint16_t* cyls,
456     uint8_t* heads, uint8_t* secs_per_cyl)
457 {
458     uint32_t cyls_times_heads;
459 
460     if (total_sectors > 65535 * 16 * 255)
461         return -EFBIG;
462 
463     if (total_sectors > 65535 * 16 * 63) {
464         *secs_per_cyl = 255;
465         *heads = 16;
466         cyls_times_heads = total_sectors / *secs_per_cyl;
467     } else {
468         *secs_per_cyl = 17;
469         cyls_times_heads = total_sectors / *secs_per_cyl;
470         *heads = (cyls_times_heads + 1023) / 1024;
471 
472         if (*heads < 4)
473             *heads = 4;
474 
475         if (cyls_times_heads >= (*heads * 1024) || *heads > 16) {
476             *secs_per_cyl = 31;
477             *heads = 16;
478             cyls_times_heads = total_sectors / *secs_per_cyl;
479         }
480 
481         if (cyls_times_heads >= (*heads * 1024)) {
482             *secs_per_cyl = 63;
483             *heads = 16;
484             cyls_times_heads = total_sectors / *secs_per_cyl;
485         }
486     }
487 
488     *cyls = cyls_times_heads / *heads;
489 
490     return 0;
491 }
492 
vpc_create(const char * filename,QEMUOptionParameter * options)493 static int vpc_create(const char *filename, QEMUOptionParameter *options)
494 {
495     uint8_t buf[1024];
496     struct vhd_footer* footer = (struct vhd_footer*) buf;
497     struct vhd_dyndisk_header* dyndisk_header =
498         (struct vhd_dyndisk_header*) buf;
499     int fd, i;
500     uint16_t cyls = 0;
501     uint8_t heads = 0;
502     uint8_t secs_per_cyl = 0;
503     size_t block_size, num_bat_entries;
504     int64_t total_sectors = 0;
505     int ret = -EIO;
506 
507     // Read out options
508     while (options && options->name) {
509         if (!strcmp(options->name, "size")) {
510             total_sectors = options->value.n / 512;
511         }
512         options++;
513     }
514 
515     // Create the file
516     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
517     if (fd < 0)
518         return -EIO;
519 
520     /* Calculate matching total_size and geometry. Increase the number of
521        sectors requested until we get enough (or fail). */
522     for (i = 0; total_sectors > (int64_t)cyls * heads * secs_per_cyl; i++) {
523         if (calculate_geometry(total_sectors + i,
524                                &cyls, &heads, &secs_per_cyl)) {
525             ret = -EFBIG;
526             goto fail;
527         }
528     }
529     total_sectors = (int64_t) cyls * heads * secs_per_cyl;
530 
531     // Prepare the Hard Disk Footer
532     memset(buf, 0, 1024);
533 
534     memcpy(footer->creator, "conectix", 8);
535     // TODO Check if "qemu" creator_app is ok for VPC
536     memcpy(footer->creator_app, "qemu", 4);
537     memcpy(footer->creator_os, "Wi2k", 4);
538 
539     footer->features = be32_to_cpu(0x02);
540     footer->version = be32_to_cpu(0x00010000);
541     footer->data_offset = be64_to_cpu(HEADER_SIZE);
542     footer->timestamp = be32_to_cpu(time(NULL) - VHD_TIMESTAMP_BASE);
543 
544     // Version of Virtual PC 2007
545     footer->major = be16_to_cpu(0x0005);
546     footer->minor =be16_to_cpu(0x0003);
547 
548     footer->orig_size = be64_to_cpu(total_sectors * 512);
549     footer->size = be64_to_cpu(total_sectors * 512);
550 
551     footer->cyls = be16_to_cpu(cyls);
552     footer->heads = heads;
553     footer->secs_per_cyl = secs_per_cyl;
554 
555     footer->type = be32_to_cpu(VHD_DYNAMIC);
556 
557     // TODO uuid is missing
558 
559     footer->checksum = be32_to_cpu(vpc_checksum(buf, HEADER_SIZE));
560 
561     // Write the footer (twice: at the beginning and at the end)
562     block_size = 0x200000;
563     num_bat_entries = (total_sectors + block_size / 512) / (block_size / 512);
564 
565     if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE) {
566         goto fail;
567     }
568 
569     if (lseek(fd, 1536 + ((num_bat_entries * 4 + 511) & ~511), SEEK_SET) < 0) {
570         goto fail;
571     }
572     if (write(fd, buf, HEADER_SIZE) != HEADER_SIZE) {
573         goto fail;
574     }
575 
576     // Write the initial BAT
577     if (lseek(fd, 3 * 512, SEEK_SET) < 0) {
578         goto fail;
579     }
580 
581     memset(buf, 0xFF, 512);
582     for (i = 0; i < (num_bat_entries * 4 + 511) / 512; i++) {
583         if (write(fd, buf, 512) != 512) {
584             goto fail;
585         }
586     }
587 
588 
589     // Prepare the Dynamic Disk Header
590     memset(buf, 0, 1024);
591 
592     memcpy(dyndisk_header->magic, "cxsparse", 8);
593 
594     dyndisk_header->data_offset = be64_to_cpu(0xFFFFFFFF);
595     dyndisk_header->table_offset = be64_to_cpu(3 * 512);
596     dyndisk_header->version = be32_to_cpu(0x00010000);
597     dyndisk_header->block_size = be32_to_cpu(block_size);
598     dyndisk_header->max_table_entries = be32_to_cpu(num_bat_entries);
599 
600     dyndisk_header->checksum = be32_to_cpu(vpc_checksum(buf, 1024));
601 
602     // Write the header
603     if (lseek(fd, 512, SEEK_SET) < 0) {
604         goto fail;
605     }
606 
607     if (write(fd, buf, 1024) != 1024) {
608         goto fail;
609     }
610     ret = 0;
611 
612  fail:
613     close(fd);
614     return ret;
615 }
616 
vpc_close(BlockDriverState * bs)617 static void vpc_close(BlockDriverState *bs)
618 {
619     BDRVVPCState *s = bs->opaque;
620     qemu_free(s->pagetable);
621 #ifdef CACHE
622     qemu_free(s->pageentry_u8);
623 #endif
624 }
625 
626 static QEMUOptionParameter vpc_create_options[] = {
627     {
628         .name = BLOCK_OPT_SIZE,
629         .type = OPT_SIZE,
630         .help = "Virtual disk size"
631     },
632     { NULL }
633 };
634 
635 static BlockDriver bdrv_vpc = {
636     .format_name    = "vpc",
637     .instance_size  = sizeof(BDRVVPCState),
638     .bdrv_probe     = vpc_probe,
639     .bdrv_open      = vpc_open,
640     .bdrv_read      = vpc_read,
641     .bdrv_write     = vpc_write,
642     .bdrv_flush     = vpc_flush,
643     .bdrv_close     = vpc_close,
644     .bdrv_create    = vpc_create,
645 
646     .create_options = vpc_create_options,
647 };
648 
bdrv_vpc_init(void)649 static void bdrv_vpc_init(void)
650 {
651     bdrv_register(&bdrv_vpc);
652 }
653 
654 block_init(bdrv_vpc_init);
655