2 * Block driver for Parallels disk image format
4 * Copyright (c) 2007 Alex Beregszaszi
5 * Copyright (c) 2015 Denis V. Lunev <den@openvz.org>
7 * This code was originally based on comparing different disk images created
8 * by Parallels. Currently it is based on opened OpenVZ sources
10 * http://git.openvz.org/?p=ploop;a=summary
12 * Permission is hereby granted, free of charge, to any person obtaining a copy
13 * of this software and associated documentation files (the "Software"), to deal
14 * in the Software without restriction, including without limitation the rights
15 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
16 * copies of the Software, and to permit persons to whom the Software is
17 * furnished to do so, subject to the following conditions:
19 * The above copyright notice and this permission notice shall be included in
20 * all copies or substantial portions of the Software.
22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
25 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
26 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
27 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30 #include "qemu-common.h"
31 #include "block/block_int.h"
32 #include "qemu/module.h"
33 #include "qemu/bitmap.h"
34 #include "qapi/util.h"
36 /**************************************************************/
38 #define HEADER_MAGIC "WithoutFreeSpace"
39 #define HEADER_MAGIC2 "WithouFreSpacExt"
40 #define HEADER_VERSION 2
41 #define HEADER_INUSE_MAGIC (0x746F6E59)
43 #define DEFAULT_CLUSTER_SIZE 1048576 /* 1 MiB */
46 // always little-endian
47 typedef struct ParallelsHeader
{
48 char magic
[16]; // "WithoutFreeSpace"
58 } QEMU_PACKED ParallelsHeader
;
61 typedef enum ParallelsPreallocMode
{
62 PRL_PREALLOC_MODE_FALLOCATE
= 0,
63 PRL_PREALLOC_MODE_TRUNCATE
= 1,
64 PRL_PREALLOC_MODE_MAX
= 2,
65 } ParallelsPreallocMode
;
67 static const char *prealloc_mode_lookup
[] = {
74 typedef struct BDRVParallelsState
{
75 /** Locking is conservative, the lock protects
76 * - image file extending (truncate, fallocate)
77 * - any access to block allocation table
81 ParallelsHeader
*header
;
85 unsigned long *bat_dirty_bmap
;
86 unsigned int bat_dirty_block
;
89 unsigned int bat_size
;
92 uint64_t prealloc_size
;
93 ParallelsPreallocMode prealloc_mode
;
97 unsigned int off_multiplier
;
101 #define PARALLELS_OPT_PREALLOC_MODE "prealloc-mode"
102 #define PARALLELS_OPT_PREALLOC_SIZE "prealloc-size"
104 static QemuOptsList parallels_runtime_opts
= {
106 .head
= QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts
.head
),
109 .name
= PARALLELS_OPT_PREALLOC_SIZE
,
110 .type
= QEMU_OPT_SIZE
,
111 .help
= "Preallocation size on image expansion",
112 .def_value_str
= "128MiB",
115 .name
= PARALLELS_OPT_PREALLOC_MODE
,
116 .type
= QEMU_OPT_STRING
,
117 .help
= "Preallocation mode on image expansion "
118 "(allowed values: falloc, truncate)",
119 .def_value_str
= "falloc",
121 { /* end of list */ },
126 static int64_t bat2sect(BDRVParallelsState
*s
, uint32_t idx
)
128 return (uint64_t)le32_to_cpu(s
->bat_bitmap
[idx
]) * s
->off_multiplier
;
131 static uint32_t bat_entry_off(uint32_t idx
)
133 return sizeof(ParallelsHeader
) + sizeof(uint32_t) * idx
;
136 static int64_t seek_to_sector(BDRVParallelsState
*s
, int64_t sector_num
)
138 uint32_t index
, offset
;
140 index
= sector_num
/ s
->tracks
;
141 offset
= sector_num
% s
->tracks
;
144 if ((index
>= s
->bat_size
) || (s
->bat_bitmap
[index
] == 0)) {
147 return bat2sect(s
, index
) + offset
;
150 static int cluster_remainder(BDRVParallelsState
*s
, int64_t sector_num
,
153 int ret
= s
->tracks
- sector_num
% s
->tracks
;
154 return MIN(nb_sectors
, ret
);
157 static int64_t block_status(BDRVParallelsState
*s
, int64_t sector_num
,
158 int nb_sectors
, int *pnum
)
160 int64_t start_off
= -2, prev_end_off
= -2;
163 while (nb_sectors
> 0 || start_off
== -2) {
164 int64_t offset
= seek_to_sector(s
, sector_num
);
167 if (start_off
== -2) {
169 prev_end_off
= offset
;
170 } else if (offset
!= prev_end_off
) {
174 to_end
= cluster_remainder(s
, sector_num
, nb_sectors
);
175 nb_sectors
-= to_end
;
176 sector_num
+= to_end
;
180 prev_end_off
+= to_end
;
186 static int64_t allocate_clusters(BlockDriverState
*bs
, int64_t sector_num
,
187 int nb_sectors
, int *pnum
)
189 BDRVParallelsState
*s
= bs
->opaque
;
190 uint32_t idx
, to_allocate
, i
;
193 pos
= block_status(s
, sector_num
, nb_sectors
, pnum
);
198 idx
= sector_num
/ s
->tracks
;
199 if (idx
>= s
->bat_size
) {
203 to_allocate
= (sector_num
+ *pnum
+ s
->tracks
- 1) / s
->tracks
- idx
;
204 space
= to_allocate
* s
->tracks
;
205 if (s
->data_end
+ space
> bdrv_getlength(bs
->file
->bs
) >> BDRV_SECTOR_BITS
) {
207 space
+= s
->prealloc_size
;
208 if (s
->prealloc_mode
== PRL_PREALLOC_MODE_FALLOCATE
) {
209 ret
= bdrv_write_zeroes(bs
->file
->bs
, s
->data_end
, space
, 0);
211 ret
= bdrv_truncate(bs
->file
->bs
,
212 (s
->data_end
+ space
) << BDRV_SECTOR_BITS
);
219 for (i
= 0; i
< to_allocate
; i
++) {
220 s
->bat_bitmap
[idx
+ i
] = cpu_to_le32(s
->data_end
/ s
->off_multiplier
);
221 s
->data_end
+= s
->tracks
;
222 bitmap_set(s
->bat_dirty_bmap
,
223 bat_entry_off(idx
) / s
->bat_dirty_block
, 1);
226 return bat2sect(s
, idx
) + sector_num
% s
->tracks
;
230 static coroutine_fn
int parallels_co_flush_to_os(BlockDriverState
*bs
)
232 BDRVParallelsState
*s
= bs
->opaque
;
233 unsigned long size
= DIV_ROUND_UP(s
->header_size
, s
->bat_dirty_block
);
236 qemu_co_mutex_lock(&s
->lock
);
238 bit
= find_first_bit(s
->bat_dirty_bmap
, size
);
240 uint32_t off
= bit
* s
->bat_dirty_block
;
241 uint32_t to_write
= s
->bat_dirty_block
;
244 if (off
+ to_write
> s
->header_size
) {
245 to_write
= s
->header_size
- off
;
247 ret
= bdrv_pwrite(bs
->file
->bs
, off
, (uint8_t *)s
->header
+ off
,
250 qemu_co_mutex_unlock(&s
->lock
);
253 bit
= find_next_bit(s
->bat_dirty_bmap
, size
, bit
+ 1);
255 bitmap_zero(s
->bat_dirty_bmap
, size
);
257 qemu_co_mutex_unlock(&s
->lock
);
262 static int64_t coroutine_fn
parallels_co_get_block_status(BlockDriverState
*bs
,
263 int64_t sector_num
, int nb_sectors
, int *pnum
)
265 BDRVParallelsState
*s
= bs
->opaque
;
268 qemu_co_mutex_lock(&s
->lock
);
269 offset
= block_status(s
, sector_num
, nb_sectors
, pnum
);
270 qemu_co_mutex_unlock(&s
->lock
);
276 return (offset
<< BDRV_SECTOR_BITS
) |
277 BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
280 static coroutine_fn
int parallels_co_writev(BlockDriverState
*bs
,
281 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
283 BDRVParallelsState
*s
= bs
->opaque
;
284 uint64_t bytes_done
= 0;
285 QEMUIOVector hd_qiov
;
288 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
290 while (nb_sectors
> 0) {
294 qemu_co_mutex_lock(&s
->lock
);
295 position
= allocate_clusters(bs
, sector_num
, nb_sectors
, &n
);
296 qemu_co_mutex_unlock(&s
->lock
);
302 nbytes
= n
<< BDRV_SECTOR_BITS
;
304 qemu_iovec_reset(&hd_qiov
);
305 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, nbytes
);
307 ret
= bdrv_co_writev(bs
->file
->bs
, position
, n
, &hd_qiov
);
314 bytes_done
+= nbytes
;
317 qemu_iovec_destroy(&hd_qiov
);
321 static coroutine_fn
int parallels_co_readv(BlockDriverState
*bs
,
322 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
324 BDRVParallelsState
*s
= bs
->opaque
;
325 uint64_t bytes_done
= 0;
326 QEMUIOVector hd_qiov
;
329 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
331 while (nb_sectors
> 0) {
335 qemu_co_mutex_lock(&s
->lock
);
336 position
= block_status(s
, sector_num
, nb_sectors
, &n
);
337 qemu_co_mutex_unlock(&s
->lock
);
339 nbytes
= n
<< BDRV_SECTOR_BITS
;
342 qemu_iovec_memset(qiov
, bytes_done
, 0, nbytes
);
344 qemu_iovec_reset(&hd_qiov
);
345 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, nbytes
);
347 ret
= bdrv_co_readv(bs
->file
->bs
, position
, n
, &hd_qiov
);
355 bytes_done
+= nbytes
;
358 qemu_iovec_destroy(&hd_qiov
);
363 static int parallels_check(BlockDriverState
*bs
, BdrvCheckResult
*res
,
366 BDRVParallelsState
*s
= bs
->opaque
;
367 int64_t size
, prev_off
, high_off
;
370 bool flush_bat
= false;
371 int cluster_size
= s
->tracks
<< BDRV_SECTOR_BITS
;
373 size
= bdrv_getlength(bs
->file
->bs
);
379 if (s
->header_unclean
) {
380 fprintf(stderr
, "%s image was not closed correctly\n",
381 fix
& BDRV_FIX_ERRORS
? "Repairing" : "ERROR");
383 if (fix
& BDRV_FIX_ERRORS
) {
384 /* parallels_close will do the job right */
385 res
->corruptions_fixed
++;
386 s
->header_unclean
= false;
390 res
->bfi
.total_clusters
= s
->bat_size
;
391 res
->bfi
.compressed_clusters
= 0; /* compression is not supported */
395 for (i
= 0; i
< s
->bat_size
; i
++) {
396 int64_t off
= bat2sect(s
, i
) << BDRV_SECTOR_BITS
;
402 /* cluster outside the image */
404 fprintf(stderr
, "%s cluster %u is outside image\n",
405 fix
& BDRV_FIX_ERRORS
? "Repairing" : "ERROR", i
);
407 if (fix
& BDRV_FIX_ERRORS
) {
409 s
->bat_bitmap
[i
] = 0;
410 res
->corruptions_fixed
++;
416 res
->bfi
.allocated_clusters
++;
417 if (off
> high_off
) {
421 if (prev_off
!= 0 && (prev_off
+ cluster_size
) != off
) {
422 res
->bfi
.fragmented_clusters
++;
428 ret
= bdrv_pwrite_sync(bs
->file
->bs
, 0, s
->header
, s
->header_size
);
435 res
->image_end_offset
= high_off
+ cluster_size
;
436 if (size
> res
->image_end_offset
) {
438 count
= DIV_ROUND_UP(size
- res
->image_end_offset
, cluster_size
);
439 fprintf(stderr
, "%s space leaked at the end of the image %" PRId64
"\n",
440 fix
& BDRV_FIX_LEAKS
? "Repairing" : "ERROR",
441 size
- res
->image_end_offset
);
443 if (fix
& BDRV_FIX_LEAKS
) {
444 ret
= bdrv_truncate(bs
->file
->bs
, res
->image_end_offset
);
449 res
->leaks_fixed
+= count
;
457 static int parallels_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
459 int64_t total_size
, cl_size
;
460 uint8_t tmp
[BDRV_SECTOR_SIZE
];
461 Error
*local_err
= NULL
;
462 BlockDriverState
*file
;
463 uint32_t bat_entries
, bat_sectors
;
464 ParallelsHeader header
;
467 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
469 cl_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
470 DEFAULT_CLUSTER_SIZE
), BDRV_SECTOR_SIZE
);
472 ret
= bdrv_create_file(filename
, opts
, &local_err
);
474 error_propagate(errp
, local_err
);
479 ret
= bdrv_open(&file
, filename
, NULL
, NULL
,
480 BDRV_O_RDWR
| BDRV_O_PROTOCOL
, &local_err
);
482 error_propagate(errp
, local_err
);
485 ret
= bdrv_truncate(file
, 0);
490 bat_entries
= DIV_ROUND_UP(total_size
, cl_size
);
491 bat_sectors
= DIV_ROUND_UP(bat_entry_off(bat_entries
), cl_size
);
492 bat_sectors
= (bat_sectors
* cl_size
) >> BDRV_SECTOR_BITS
;
494 memset(&header
, 0, sizeof(header
));
495 memcpy(header
.magic
, HEADER_MAGIC2
, sizeof(header
.magic
));
496 header
.version
= cpu_to_le32(HEADER_VERSION
);
497 /* don't care much about geometry, it is not used on image level */
498 header
.heads
= cpu_to_le32(16);
499 header
.cylinders
= cpu_to_le32(total_size
/ BDRV_SECTOR_SIZE
/ 16 / 32);
500 header
.tracks
= cpu_to_le32(cl_size
>> BDRV_SECTOR_BITS
);
501 header
.bat_entries
= cpu_to_le32(bat_entries
);
502 header
.nb_sectors
= cpu_to_le64(DIV_ROUND_UP(total_size
, BDRV_SECTOR_SIZE
));
503 header
.data_off
= cpu_to_le32(bat_sectors
);
505 /* write all the data */
506 memset(tmp
, 0, sizeof(tmp
));
507 memcpy(tmp
, &header
, sizeof(header
));
509 ret
= bdrv_pwrite(file
, 0, tmp
, BDRV_SECTOR_SIZE
);
513 ret
= bdrv_write_zeroes(file
, 1, bat_sectors
- 1, 0);
524 error_setg_errno(errp
, -ret
, "Failed to create Parallels image");
529 static int parallels_probe(const uint8_t *buf
, int buf_size
,
530 const char *filename
)
532 const ParallelsHeader
*ph
= (const void *)buf
;
534 if (buf_size
< sizeof(ParallelsHeader
)) {
538 if ((!memcmp(ph
->magic
, HEADER_MAGIC
, 16) ||
539 !memcmp(ph
->magic
, HEADER_MAGIC2
, 16)) &&
540 (le32_to_cpu(ph
->version
) == HEADER_VERSION
)) {
547 static int parallels_update_header(BlockDriverState
*bs
)
549 BDRVParallelsState
*s
= bs
->opaque
;
550 unsigned size
= MAX(bdrv_opt_mem_align(bs
->file
->bs
),
551 sizeof(ParallelsHeader
));
553 if (size
> s
->header_size
) {
554 size
= s
->header_size
;
556 return bdrv_pwrite_sync(bs
->file
->bs
, 0, s
->header
, size
);
559 static int parallels_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
562 BDRVParallelsState
*s
= bs
->opaque
;
565 QemuOpts
*opts
= NULL
;
566 Error
*local_err
= NULL
;
569 ret
= bdrv_pread(bs
->file
->bs
, 0, &ph
, sizeof(ph
));
574 bs
->total_sectors
= le64_to_cpu(ph
.nb_sectors
);
576 if (le32_to_cpu(ph
.version
) != HEADER_VERSION
) {
579 if (!memcmp(ph
.magic
, HEADER_MAGIC
, 16)) {
580 s
->off_multiplier
= 1;
581 bs
->total_sectors
= 0xffffffff & bs
->total_sectors
;
582 } else if (!memcmp(ph
.magic
, HEADER_MAGIC2
, 16)) {
583 s
->off_multiplier
= le32_to_cpu(ph
.tracks
);
588 s
->tracks
= le32_to_cpu(ph
.tracks
);
589 if (s
->tracks
== 0) {
590 error_setg(errp
, "Invalid image: Zero sectors per track");
594 if (s
->tracks
> INT32_MAX
/513) {
595 error_setg(errp
, "Invalid image: Too big cluster");
600 s
->bat_size
= le32_to_cpu(ph
.bat_entries
);
601 if (s
->bat_size
> INT_MAX
/ sizeof(uint32_t)) {
602 error_setg(errp
, "Catalog too large");
607 size
= bat_entry_off(s
->bat_size
);
608 s
->header_size
= ROUND_UP(size
, bdrv_opt_mem_align(bs
->file
->bs
));
609 s
->header
= qemu_try_blockalign(bs
->file
->bs
, s
->header_size
);
610 if (s
->header
== NULL
) {
614 s
->data_end
= le32_to_cpu(ph
.data_off
);
615 if (s
->data_end
== 0) {
616 s
->data_end
= ROUND_UP(bat_entry_off(s
->bat_size
), BDRV_SECTOR_SIZE
);
618 if (s
->data_end
< s
->header_size
) {
619 /* there is not enough unused space to fit to block align between BAT
620 and actual data. We can't avoid read-modify-write... */
621 s
->header_size
= size
;
624 ret
= bdrv_pread(bs
->file
->bs
, 0, s
->header
, s
->header_size
);
628 s
->bat_bitmap
= (uint32_t *)(s
->header
+ 1);
630 for (i
= 0; i
< s
->bat_size
; i
++) {
631 int64_t off
= bat2sect(s
, i
);
632 if (off
>= s
->data_end
) {
633 s
->data_end
= off
+ s
->tracks
;
637 if (le32_to_cpu(ph
.inuse
) == HEADER_INUSE_MAGIC
) {
638 /* Image was not closed correctly. The check is mandatory */
639 s
->header_unclean
= true;
640 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
641 error_setg(errp
, "parallels: Image was not closed correctly; "
642 "cannot be opened read/write");
648 opts
= qemu_opts_create(¶llels_runtime_opts
, NULL
, 0, &local_err
);
649 if (local_err
!= NULL
) {
653 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
654 if (local_err
!= NULL
) {
659 qemu_opt_get_size_del(opts
, PARALLELS_OPT_PREALLOC_SIZE
, 0);
660 s
->prealloc_size
= MAX(s
->tracks
, s
->prealloc_size
>> BDRV_SECTOR_BITS
);
661 buf
= qemu_opt_get_del(opts
, PARALLELS_OPT_PREALLOC_MODE
);
662 s
->prealloc_mode
= qapi_enum_parse(prealloc_mode_lookup
, buf
,
663 PRL_PREALLOC_MODE_MAX
, PRL_PREALLOC_MODE_FALLOCATE
, &local_err
);
665 if (local_err
!= NULL
) {
668 if (!bdrv_has_zero_init(bs
->file
->bs
) ||
669 bdrv_truncate(bs
->file
->bs
, bdrv_getlength(bs
->file
->bs
)) != 0) {
670 s
->prealloc_mode
= PRL_PREALLOC_MODE_FALLOCATE
;
673 if (flags
& BDRV_O_RDWR
) {
674 s
->header
->inuse
= cpu_to_le32(HEADER_INUSE_MAGIC
);
675 ret
= parallels_update_header(bs
);
681 s
->bat_dirty_block
= 4 * getpagesize();
683 bitmap_new(DIV_ROUND_UP(s
->header_size
, s
->bat_dirty_block
));
685 qemu_co_mutex_init(&s
->lock
);
689 error_setg(errp
, "Image not in Parallels format");
692 qemu_vfree(s
->header
);
696 error_propagate(errp
, local_err
);
702 static void parallels_close(BlockDriverState
*bs
)
704 BDRVParallelsState
*s
= bs
->opaque
;
706 if (bs
->open_flags
& BDRV_O_RDWR
) {
707 s
->header
->inuse
= 0;
708 parallels_update_header(bs
);
711 if (bs
->open_flags
& BDRV_O_RDWR
) {
712 bdrv_truncate(bs
->file
->bs
, s
->data_end
<< BDRV_SECTOR_BITS
);
715 g_free(s
->bat_dirty_bmap
);
716 qemu_vfree(s
->header
);
719 static QemuOptsList parallels_create_opts
= {
720 .name
= "parallels-create-opts",
721 .head
= QTAILQ_HEAD_INITIALIZER(parallels_create_opts
.head
),
724 .name
= BLOCK_OPT_SIZE
,
725 .type
= QEMU_OPT_SIZE
,
726 .help
= "Virtual disk size",
729 .name
= BLOCK_OPT_CLUSTER_SIZE
,
730 .type
= QEMU_OPT_SIZE
,
731 .help
= "Parallels image cluster size",
732 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
),
734 { /* end of list */ }
738 static BlockDriver bdrv_parallels
= {
739 .format_name
= "parallels",
740 .instance_size
= sizeof(BDRVParallelsState
),
741 .bdrv_probe
= parallels_probe
,
742 .bdrv_open
= parallels_open
,
743 .bdrv_close
= parallels_close
,
744 .bdrv_co_get_block_status
= parallels_co_get_block_status
,
745 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
746 .bdrv_co_flush_to_os
= parallels_co_flush_to_os
,
747 .bdrv_co_readv
= parallels_co_readv
,
748 .bdrv_co_writev
= parallels_co_writev
,
750 .bdrv_create
= parallels_create
,
751 .bdrv_check
= parallels_check
,
752 .create_opts
= ¶llels_create_opts
,
755 static void bdrv_parallels_init(void)
757 bdrv_register(&bdrv_parallels
);
760 block_init(bdrv_parallels_init
);