2 * Block driver for Parallels disk image format
4 * Copyright (c) 2007 Alex Beregszaszi
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/osdep.h"
31 #include "qapi/error.h"
32 #include "qemu-common.h"
33 #include "block/block_int.h"
34 #include "sysemu/block-backend.h"
35 #include "qemu/module.h"
36 #include "qemu/bswap.h"
37 #include "qemu/bitmap.h"
39 /**************************************************************/
41 #define HEADER_MAGIC "WithoutFreeSpace"
42 #define HEADER_MAGIC2 "WithouFreSpacExt"
43 #define HEADER_VERSION 2
44 #define HEADER_INUSE_MAGIC (0x746F6E59)
45 #define MAX_PARALLELS_IMAGE_FACTOR (1ull << 32)
47 #define DEFAULT_CLUSTER_SIZE 1048576 /* 1 MiB */
50 // always little-endian
51 typedef struct ParallelsHeader {
52 char magic[16]; // "WithoutFreeSpace"
62 } QEMU_PACKED ParallelsHeader;
65 typedef enum ParallelsPreallocMode {
66 PRL_PREALLOC_MODE_FALLOCATE = 0,
67 PRL_PREALLOC_MODE_TRUNCATE = 1,
68 PRL_PREALLOC_MODE__MAX = 2,
69 } ParallelsPreallocMode;
71 static QEnumLookup prealloc_mode_lookup = {
72 .array = (const char *const[]) {
77 .size = PRL_PREALLOC_MODE__MAX
80 typedef struct BDRVParallelsState {
81 /** Locking is conservative, the lock protects
82 * - image file extending (truncate, fallocate)
83 * - any access to block allocation table
87 ParallelsHeader *header;
91 unsigned long *bat_dirty_bmap;
92 unsigned int bat_dirty_block;
95 unsigned int bat_size;
98 uint64_t prealloc_size;
99 ParallelsPreallocMode prealloc_mode;
103 unsigned int off_multiplier;
104 } BDRVParallelsState;
107 #define PARALLELS_OPT_PREALLOC_MODE "prealloc-mode"
108 #define PARALLELS_OPT_PREALLOC_SIZE "prealloc-size"
110 static QemuOptsList parallels_runtime_opts = {
112 .head = QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts.head),
115 .name = PARALLELS_OPT_PREALLOC_SIZE,
116 .type = QEMU_OPT_SIZE,
117 .help = "Preallocation size on image expansion",
118 .def_value_str = "128M",
121 .name = PARALLELS_OPT_PREALLOC_MODE,
122 .type = QEMU_OPT_STRING,
123 .help = "Preallocation mode on image expansion "
124 "(allowed values: falloc, truncate)",
125 .def_value_str = "falloc",
127 { /* end of list */ },
132 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx)
134 return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier;
137 static uint32_t bat_entry_off(uint32_t idx)
139 return sizeof(ParallelsHeader) + sizeof(uint32_t) * idx;
142 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num)
144 uint32_t index, offset;
146 index = sector_num / s->tracks;
147 offset = sector_num % s->tracks;
150 if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) {
153 return bat2sect(s, index) + offset;
156 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num,
159 int ret = s->tracks - sector_num % s->tracks;
160 return MIN(nb_sectors, ret);
163 static int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
164 int nb_sectors, int *pnum)
166 int64_t start_off = -2, prev_end_off = -2;
169 while (nb_sectors > 0 || start_off == -2) {
170 int64_t offset = seek_to_sector(s, sector_num);
173 if (start_off == -2) {
175 prev_end_off = offset;
176 } else if (offset != prev_end_off) {
180 to_end = cluster_remainder(s, sector_num, nb_sectors);
181 nb_sectors -= to_end;
182 sector_num += to_end;
186 prev_end_off += to_end;
192 static int64_t allocate_clusters(BlockDriverState *bs, int64_t sector_num,
193 int nb_sectors, int *pnum)
195 BDRVParallelsState *s = bs->opaque;
196 int64_t pos, space, idx, to_allocate, i, len;
198 pos = block_status(s, sector_num, nb_sectors, pnum);
203 idx = sector_num / s->tracks;
204 to_allocate = DIV_ROUND_UP(sector_num + *pnum, s->tracks) - idx;
206 /* This function is called only by parallels_co_writev(), which will never
207 * pass a sector_num at or beyond the end of the image (because the block
208 * layer never passes such a sector_num to that function). Therefore, idx
209 * is always below s->bat_size.
210 * block_status() will limit *pnum so that sector_num + *pnum will not
211 * exceed the image end. Therefore, idx + to_allocate cannot exceed
213 * Note that s->bat_size is an unsigned int, therefore idx + to_allocate
214 * will always fit into a uint32_t. */
215 assert(idx < s->bat_size && idx + to_allocate <= s->bat_size);
217 space = to_allocate * s->tracks;
218 len = bdrv_getlength(bs->file->bs);
222 if (s->data_end + space > (len >> BDRV_SECTOR_BITS)) {
224 space += s->prealloc_size;
225 if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) {
226 ret = bdrv_pwrite_zeroes(bs->file,
227 s->data_end << BDRV_SECTOR_BITS,
228 space << BDRV_SECTOR_BITS, 0);
230 ret = bdrv_truncate(bs->file,
231 (s->data_end + space) << BDRV_SECTOR_BITS,
232 PREALLOC_MODE_OFF, NULL);
239 for (i = 0; i < to_allocate; i++) {
240 s->bat_bitmap[idx + i] = cpu_to_le32(s->data_end / s->off_multiplier);
241 s->data_end += s->tracks;
242 bitmap_set(s->bat_dirty_bmap,
243 bat_entry_off(idx + i) / s->bat_dirty_block, 1);
246 return bat2sect(s, idx) + sector_num % s->tracks;
250 static coroutine_fn int parallels_co_flush_to_os(BlockDriverState *bs)
252 BDRVParallelsState *s = bs->opaque;
253 unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block);
256 qemu_co_mutex_lock(&s->lock);
258 bit = find_first_bit(s->bat_dirty_bmap, size);
260 uint32_t off = bit * s->bat_dirty_block;
261 uint32_t to_write = s->bat_dirty_block;
264 if (off + to_write > s->header_size) {
265 to_write = s->header_size - off;
267 ret = bdrv_pwrite(bs->file, off, (uint8_t *)s->header + off,
270 qemu_co_mutex_unlock(&s->lock);
273 bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1);
275 bitmap_zero(s->bat_dirty_bmap, size);
277 qemu_co_mutex_unlock(&s->lock);
282 static int64_t coroutine_fn parallels_co_get_block_status(BlockDriverState *bs,
283 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
285 BDRVParallelsState *s = bs->opaque;
288 qemu_co_mutex_lock(&s->lock);
289 offset = block_status(s, sector_num, nb_sectors, pnum);
290 qemu_co_mutex_unlock(&s->lock);
296 *file = bs->file->bs;
297 return (offset << BDRV_SECTOR_BITS) |
298 BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
301 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
302 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
304 BDRVParallelsState *s = bs->opaque;
305 uint64_t bytes_done = 0;
306 QEMUIOVector hd_qiov;
309 qemu_iovec_init(&hd_qiov, qiov->niov);
311 while (nb_sectors > 0) {
315 qemu_co_mutex_lock(&s->lock);
316 position = allocate_clusters(bs, sector_num, nb_sectors, &n);
317 qemu_co_mutex_unlock(&s->lock);
323 nbytes = n << BDRV_SECTOR_BITS;
325 qemu_iovec_reset(&hd_qiov);
326 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
328 ret = bdrv_co_writev(bs->file, position, n, &hd_qiov);
335 bytes_done += nbytes;
338 qemu_iovec_destroy(&hd_qiov);
342 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
343 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
345 BDRVParallelsState *s = bs->opaque;
346 uint64_t bytes_done = 0;
347 QEMUIOVector hd_qiov;
350 qemu_iovec_init(&hd_qiov, qiov->niov);
352 while (nb_sectors > 0) {
356 qemu_co_mutex_lock(&s->lock);
357 position = block_status(s, sector_num, nb_sectors, &n);
358 qemu_co_mutex_unlock(&s->lock);
360 nbytes = n << BDRV_SECTOR_BITS;
363 qemu_iovec_memset(qiov, bytes_done, 0, nbytes);
365 qemu_iovec_reset(&hd_qiov);
366 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
368 ret = bdrv_co_readv(bs->file, position, n, &hd_qiov);
376 bytes_done += nbytes;
379 qemu_iovec_destroy(&hd_qiov);
384 static int parallels_check(BlockDriverState *bs, BdrvCheckResult *res,
387 BDRVParallelsState *s = bs->opaque;
388 int64_t size, prev_off, high_off;
391 bool flush_bat = false;
392 int cluster_size = s->tracks << BDRV_SECTOR_BITS;
394 size = bdrv_getlength(bs->file->bs);
400 if (s->header_unclean) {
401 fprintf(stderr, "%s image was not closed correctly\n",
402 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
404 if (fix & BDRV_FIX_ERRORS) {
405 /* parallels_close will do the job right */
406 res->corruptions_fixed++;
407 s->header_unclean = false;
411 res->bfi.total_clusters = s->bat_size;
412 res->bfi.compressed_clusters = 0; /* compression is not supported */
416 for (i = 0; i < s->bat_size; i++) {
417 int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
423 /* cluster outside the image */
425 fprintf(stderr, "%s cluster %u is outside image\n",
426 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
428 if (fix & BDRV_FIX_ERRORS) {
430 s->bat_bitmap[i] = 0;
431 res->corruptions_fixed++;
437 res->bfi.allocated_clusters++;
438 if (off > high_off) {
442 if (prev_off != 0 && (prev_off + cluster_size) != off) {
443 res->bfi.fragmented_clusters++;
449 ret = bdrv_pwrite_sync(bs->file, 0, s->header, s->header_size);
456 res->image_end_offset = high_off + cluster_size;
457 if (size > res->image_end_offset) {
459 count = DIV_ROUND_UP(size - res->image_end_offset, cluster_size);
460 fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
461 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
462 size - res->image_end_offset);
464 if (fix & BDRV_FIX_LEAKS) {
465 Error *local_err = NULL;
466 ret = bdrv_truncate(bs->file, res->image_end_offset,
467 PREALLOC_MODE_OFF, &local_err);
469 error_report_err(local_err);
473 res->leaks_fixed += count;
481 static int parallels_create(const char *filename, QemuOpts *opts, Error **errp)
483 int64_t total_size, cl_size;
484 uint8_t tmp[BDRV_SECTOR_SIZE];
485 Error *local_err = NULL;
487 uint32_t bat_entries, bat_sectors;
488 ParallelsHeader header;
491 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
493 cl_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
494 DEFAULT_CLUSTER_SIZE), BDRV_SECTOR_SIZE);
495 if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) {
496 error_propagate(errp, local_err);
500 ret = bdrv_create_file(filename, opts, &local_err);
502 error_propagate(errp, local_err);
506 file = blk_new_open(filename, NULL, NULL,
507 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
510 error_propagate(errp, local_err);
514 blk_set_allow_write_beyond_eof(file, true);
516 ret = blk_truncate(file, 0, PREALLOC_MODE_OFF, errp);
521 bat_entries = DIV_ROUND_UP(total_size, cl_size);
522 bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
523 bat_sectors = (bat_sectors * cl_size) >> BDRV_SECTOR_BITS;
525 memset(&header, 0, sizeof(header));
526 memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
527 header.version = cpu_to_le32(HEADER_VERSION);
528 /* don't care much about geometry, it is not used on image level */
529 header.heads = cpu_to_le32(16);
530 header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE / 16 / 32);
531 header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
532 header.bat_entries = cpu_to_le32(bat_entries);
533 header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
534 header.data_off = cpu_to_le32(bat_sectors);
536 /* write all the data */
537 memset(tmp, 0, sizeof(tmp));
538 memcpy(tmp, &header, sizeof(header));
540 ret = blk_pwrite(file, 0, tmp, BDRV_SECTOR_SIZE, 0);
544 ret = blk_pwrite_zeroes(file, BDRV_SECTOR_SIZE,
545 (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
556 error_setg_errno(errp, -ret, "Failed to create Parallels image");
561 static int parallels_probe(const uint8_t *buf, int buf_size,
562 const char *filename)
564 const ParallelsHeader *ph = (const void *)buf;
566 if (buf_size < sizeof(ParallelsHeader)) {
570 if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
571 !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
572 (le32_to_cpu(ph->version) == HEADER_VERSION)) {
579 static int parallels_update_header(BlockDriverState *bs)
581 BDRVParallelsState *s = bs->opaque;
582 unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
583 sizeof(ParallelsHeader));
585 if (size > s->header_size) {
586 size = s->header_size;
588 return bdrv_pwrite_sync(bs->file, 0, s->header, size);
591 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
594 BDRVParallelsState *s = bs->opaque;
597 QemuOpts *opts = NULL;
598 Error *local_err = NULL;
601 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
607 ret = bdrv_pread(bs->file, 0, &ph, sizeof(ph));
612 bs->total_sectors = le64_to_cpu(ph.nb_sectors);
614 if (le32_to_cpu(ph.version) != HEADER_VERSION) {
617 if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
618 s->off_multiplier = 1;
619 bs->total_sectors = 0xffffffff & bs->total_sectors;
620 } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
621 s->off_multiplier = le32_to_cpu(ph.tracks);
626 s->tracks = le32_to_cpu(ph.tracks);
627 if (s->tracks == 0) {
628 error_setg(errp, "Invalid image: Zero sectors per track");
632 if (s->tracks > INT32_MAX/513) {
633 error_setg(errp, "Invalid image: Too big cluster");
638 s->bat_size = le32_to_cpu(ph.bat_entries);
639 if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
640 error_setg(errp, "Catalog too large");
645 size = bat_entry_off(s->bat_size);
646 s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
647 s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
648 if (s->header == NULL) {
652 s->data_end = le32_to_cpu(ph.data_off);
653 if (s->data_end == 0) {
654 s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
656 if (s->data_end < s->header_size) {
657 /* there is not enough unused space to fit to block align between BAT
658 and actual data. We can't avoid read-modify-write... */
659 s->header_size = size;
662 ret = bdrv_pread(bs->file, 0, s->header, s->header_size);
666 s->bat_bitmap = (uint32_t *)(s->header + 1);
668 for (i = 0; i < s->bat_size; i++) {
669 int64_t off = bat2sect(s, i);
670 if (off >= s->data_end) {
671 s->data_end = off + s->tracks;
675 if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
676 /* Image was not closed correctly. The check is mandatory */
677 s->header_unclean = true;
678 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
679 error_setg(errp, "parallels: Image was not closed correctly; "
680 "cannot be opened read/write");
686 opts = qemu_opts_create(¶llels_runtime_opts, NULL, 0, &local_err);
687 if (local_err != NULL) {
691 qemu_opts_absorb_qdict(opts, options, &local_err);
692 if (local_err != NULL) {
697 qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
698 s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
699 buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
700 s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf,
701 PRL_PREALLOC_MODE_FALLOCATE,
704 if (local_err != NULL) {
708 if (!bdrv_has_zero_init(bs->file->bs)) {
709 s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
712 if (flags & BDRV_O_RDWR) {
713 s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
714 ret = parallels_update_header(bs);
720 s->bat_dirty_block = 4 * getpagesize();
722 bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block));
724 qemu_co_mutex_init(&s->lock);
728 error_setg(errp, "Image not in Parallels format");
731 qemu_vfree(s->header);
735 error_propagate(errp, local_err);
741 static void parallels_close(BlockDriverState *bs)
743 BDRVParallelsState *s = bs->opaque;
745 if (bs->open_flags & BDRV_O_RDWR) {
746 s->header->inuse = 0;
747 parallels_update_header(bs);
750 if (bs->open_flags & BDRV_O_RDWR) {
751 bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS,
752 PREALLOC_MODE_OFF, NULL);
755 g_free(s->bat_dirty_bmap);
756 qemu_vfree(s->header);
759 static QemuOptsList parallels_create_opts = {
760 .name = "parallels-create-opts",
761 .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head),
764 .name = BLOCK_OPT_SIZE,
765 .type = QEMU_OPT_SIZE,
766 .help = "Virtual disk size",
769 .name = BLOCK_OPT_CLUSTER_SIZE,
770 .type = QEMU_OPT_SIZE,
771 .help = "Parallels image cluster size",
772 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE),
774 { /* end of list */ }
778 static BlockDriver bdrv_parallels = {
779 .format_name = "parallels",
780 .instance_size = sizeof(BDRVParallelsState),
781 .bdrv_probe = parallels_probe,
782 .bdrv_open = parallels_open,
783 .bdrv_close = parallels_close,
784 .bdrv_child_perm = bdrv_format_default_perms,
785 .bdrv_co_get_block_status = parallels_co_get_block_status,
786 .bdrv_has_zero_init = bdrv_has_zero_init_1,
787 .bdrv_co_flush_to_os = parallels_co_flush_to_os,
788 .bdrv_co_readv = parallels_co_readv,
789 .bdrv_co_writev = parallels_co_writev,
791 .bdrv_create = parallels_create,
792 .bdrv_check = parallels_check,
793 .create_opts = ¶llels_create_opts,
796 static void bdrv_parallels_init(void)
798 bdrv_register(&bdrv_parallels);
801 block_init(bdrv_parallels_init);