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/bitmap.h"
37 #include "qapi/util.h"
39 /**************************************************************/
41 #define HEADER_MAGIC "WithoutFreeSpace"
42 #define HEADER_MAGIC2 "WithouFreSpacExt"
43 #define HEADER_VERSION 2
44 #define HEADER_INUSE_MAGIC (0x746F6E59)
46 #define DEFAULT_CLUSTER_SIZE 1048576 /* 1 MiB */
49 // always little-endian
50 typedef struct ParallelsHeader {
51 char magic[16]; // "WithoutFreeSpace"
61 } QEMU_PACKED ParallelsHeader;
64 typedef enum ParallelsPreallocMode {
65 PRL_PREALLOC_MODE_FALLOCATE = 0,
66 PRL_PREALLOC_MODE_TRUNCATE = 1,
67 PRL_PREALLOC_MODE__MAX = 2,
68 } ParallelsPreallocMode;
70 static const char *prealloc_mode_lookup[] = {
77 typedef struct BDRVParallelsState {
78 /** Locking is conservative, the lock protects
79 * - image file extending (truncate, fallocate)
80 * - any access to block allocation table
84 ParallelsHeader *header;
88 unsigned long *bat_dirty_bmap;
89 unsigned int bat_dirty_block;
92 unsigned int bat_size;
95 uint64_t prealloc_size;
96 ParallelsPreallocMode prealloc_mode;
100 unsigned int off_multiplier;
101 } BDRVParallelsState;
104 #define PARALLELS_OPT_PREALLOC_MODE "prealloc-mode"
105 #define PARALLELS_OPT_PREALLOC_SIZE "prealloc-size"
107 static QemuOptsList parallels_runtime_opts = {
109 .head = QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts.head),
112 .name = PARALLELS_OPT_PREALLOC_SIZE,
113 .type = QEMU_OPT_SIZE,
114 .help = "Preallocation size on image expansion",
115 .def_value_str = "128MiB",
118 .name = PARALLELS_OPT_PREALLOC_MODE,
119 .type = QEMU_OPT_STRING,
120 .help = "Preallocation mode on image expansion "
121 "(allowed values: falloc, truncate)",
122 .def_value_str = "falloc",
124 { /* end of list */ },
129 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx)
131 return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier;
134 static uint32_t bat_entry_off(uint32_t idx)
136 return sizeof(ParallelsHeader) + sizeof(uint32_t) * idx;
139 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num)
141 uint32_t index, offset;
143 index = sector_num / s->tracks;
144 offset = sector_num % s->tracks;
147 if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) {
150 return bat2sect(s, index) + offset;
153 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num,
156 int ret = s->tracks - sector_num % s->tracks;
157 return MIN(nb_sectors, ret);
160 static int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
161 int nb_sectors, int *pnum)
163 int64_t start_off = -2, prev_end_off = -2;
166 while (nb_sectors > 0 || start_off == -2) {
167 int64_t offset = seek_to_sector(s, sector_num);
170 if (start_off == -2) {
172 prev_end_off = offset;
173 } else if (offset != prev_end_off) {
177 to_end = cluster_remainder(s, sector_num, nb_sectors);
178 nb_sectors -= to_end;
179 sector_num += to_end;
183 prev_end_off += to_end;
189 static int64_t allocate_clusters(BlockDriverState *bs, int64_t sector_num,
190 int nb_sectors, int *pnum)
192 BDRVParallelsState *s = bs->opaque;
193 uint32_t idx, to_allocate, i;
196 pos = block_status(s, sector_num, nb_sectors, pnum);
201 idx = sector_num / s->tracks;
202 if (idx >= s->bat_size) {
206 to_allocate = (sector_num + *pnum + s->tracks - 1) / s->tracks - idx;
207 space = to_allocate * s->tracks;
208 if (s->data_end + space > bdrv_getlength(bs->file->bs) >> BDRV_SECTOR_BITS) {
210 space += s->prealloc_size;
211 if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) {
212 ret = bdrv_write_zeroes(bs->file->bs, s->data_end, space, 0);
214 ret = bdrv_truncate(bs->file->bs,
215 (s->data_end + space) << BDRV_SECTOR_BITS);
222 for (i = 0; i < to_allocate; i++) {
223 s->bat_bitmap[idx + i] = cpu_to_le32(s->data_end / s->off_multiplier);
224 s->data_end += s->tracks;
225 bitmap_set(s->bat_dirty_bmap,
226 bat_entry_off(idx + i) / s->bat_dirty_block, 1);
229 return bat2sect(s, idx) + sector_num % s->tracks;
233 static coroutine_fn int parallels_co_flush_to_os(BlockDriverState *bs)
235 BDRVParallelsState *s = bs->opaque;
236 unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block);
239 qemu_co_mutex_lock(&s->lock);
241 bit = find_first_bit(s->bat_dirty_bmap, size);
243 uint32_t off = bit * s->bat_dirty_block;
244 uint32_t to_write = s->bat_dirty_block;
247 if (off + to_write > s->header_size) {
248 to_write = s->header_size - off;
250 ret = bdrv_pwrite(bs->file->bs, off, (uint8_t *)s->header + off,
253 qemu_co_mutex_unlock(&s->lock);
256 bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1);
258 bitmap_zero(s->bat_dirty_bmap, size);
260 qemu_co_mutex_unlock(&s->lock);
265 static int64_t coroutine_fn parallels_co_get_block_status(BlockDriverState *bs,
266 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
268 BDRVParallelsState *s = bs->opaque;
271 qemu_co_mutex_lock(&s->lock);
272 offset = block_status(s, sector_num, nb_sectors, pnum);
273 qemu_co_mutex_unlock(&s->lock);
279 *file = bs->file->bs;
280 return (offset << BDRV_SECTOR_BITS) |
281 BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
284 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
285 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
287 BDRVParallelsState *s = bs->opaque;
288 uint64_t bytes_done = 0;
289 QEMUIOVector hd_qiov;
292 qemu_iovec_init(&hd_qiov, qiov->niov);
294 while (nb_sectors > 0) {
298 qemu_co_mutex_lock(&s->lock);
299 position = allocate_clusters(bs, sector_num, nb_sectors, &n);
300 qemu_co_mutex_unlock(&s->lock);
306 nbytes = n << BDRV_SECTOR_BITS;
308 qemu_iovec_reset(&hd_qiov);
309 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
311 ret = bdrv_co_writev(bs->file->bs, position, n, &hd_qiov);
318 bytes_done += nbytes;
321 qemu_iovec_destroy(&hd_qiov);
325 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
326 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
328 BDRVParallelsState *s = bs->opaque;
329 uint64_t bytes_done = 0;
330 QEMUIOVector hd_qiov;
333 qemu_iovec_init(&hd_qiov, qiov->niov);
335 while (nb_sectors > 0) {
339 qemu_co_mutex_lock(&s->lock);
340 position = block_status(s, sector_num, nb_sectors, &n);
341 qemu_co_mutex_unlock(&s->lock);
343 nbytes = n << BDRV_SECTOR_BITS;
346 qemu_iovec_memset(qiov, bytes_done, 0, nbytes);
348 qemu_iovec_reset(&hd_qiov);
349 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
351 ret = bdrv_co_readv(bs->file->bs, position, n, &hd_qiov);
359 bytes_done += nbytes;
362 qemu_iovec_destroy(&hd_qiov);
367 static int parallels_check(BlockDriverState *bs, BdrvCheckResult *res,
370 BDRVParallelsState *s = bs->opaque;
371 int64_t size, prev_off, high_off;
374 bool flush_bat = false;
375 int cluster_size = s->tracks << BDRV_SECTOR_BITS;
377 size = bdrv_getlength(bs->file->bs);
383 if (s->header_unclean) {
384 fprintf(stderr, "%s image was not closed correctly\n",
385 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
387 if (fix & BDRV_FIX_ERRORS) {
388 /* parallels_close will do the job right */
389 res->corruptions_fixed++;
390 s->header_unclean = false;
394 res->bfi.total_clusters = s->bat_size;
395 res->bfi.compressed_clusters = 0; /* compression is not supported */
399 for (i = 0; i < s->bat_size; i++) {
400 int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
406 /* cluster outside the image */
408 fprintf(stderr, "%s cluster %u is outside image\n",
409 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
411 if (fix & BDRV_FIX_ERRORS) {
413 s->bat_bitmap[i] = 0;
414 res->corruptions_fixed++;
420 res->bfi.allocated_clusters++;
421 if (off > high_off) {
425 if (prev_off != 0 && (prev_off + cluster_size) != off) {
426 res->bfi.fragmented_clusters++;
432 ret = bdrv_pwrite_sync(bs->file->bs, 0, s->header, s->header_size);
439 res->image_end_offset = high_off + cluster_size;
440 if (size > res->image_end_offset) {
442 count = DIV_ROUND_UP(size - res->image_end_offset, cluster_size);
443 fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
444 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
445 size - res->image_end_offset);
447 if (fix & BDRV_FIX_LEAKS) {
448 ret = bdrv_truncate(bs->file->bs, res->image_end_offset);
453 res->leaks_fixed += count;
461 static int parallels_create(const char *filename, QemuOpts *opts, Error **errp)
463 int64_t total_size, cl_size;
464 uint8_t tmp[BDRV_SECTOR_SIZE];
465 Error *local_err = NULL;
467 uint32_t bat_entries, bat_sectors;
468 ParallelsHeader header;
471 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
473 cl_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
474 DEFAULT_CLUSTER_SIZE), BDRV_SECTOR_SIZE);
476 ret = bdrv_create_file(filename, opts, &local_err);
478 error_propagate(errp, local_err);
482 file = blk_new_open(filename, NULL, NULL,
483 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_PROTOCOL,
486 error_propagate(errp, local_err);
490 blk_set_allow_write_beyond_eof(file, true);
492 ret = blk_truncate(file, 0);
497 bat_entries = DIV_ROUND_UP(total_size, cl_size);
498 bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
499 bat_sectors = (bat_sectors * cl_size) >> BDRV_SECTOR_BITS;
501 memset(&header, 0, sizeof(header));
502 memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
503 header.version = cpu_to_le32(HEADER_VERSION);
504 /* don't care much about geometry, it is not used on image level */
505 header.heads = cpu_to_le32(16);
506 header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE / 16 / 32);
507 header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
508 header.bat_entries = cpu_to_le32(bat_entries);
509 header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
510 header.data_off = cpu_to_le32(bat_sectors);
512 /* write all the data */
513 memset(tmp, 0, sizeof(tmp));
514 memcpy(tmp, &header, sizeof(header));
516 ret = blk_pwrite(file, 0, tmp, BDRV_SECTOR_SIZE);
520 ret = blk_write_zeroes(file, 1, bat_sectors - 1, 0);
531 error_setg_errno(errp, -ret, "Failed to create Parallels image");
536 static int parallels_probe(const uint8_t *buf, int buf_size,
537 const char *filename)
539 const ParallelsHeader *ph = (const void *)buf;
541 if (buf_size < sizeof(ParallelsHeader)) {
545 if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
546 !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
547 (le32_to_cpu(ph->version) == HEADER_VERSION)) {
554 static int parallels_update_header(BlockDriverState *bs)
556 BDRVParallelsState *s = bs->opaque;
557 unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
558 sizeof(ParallelsHeader));
560 if (size > s->header_size) {
561 size = s->header_size;
563 return bdrv_pwrite_sync(bs->file->bs, 0, s->header, size);
566 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
569 BDRVParallelsState *s = bs->opaque;
572 QemuOpts *opts = NULL;
573 Error *local_err = NULL;
576 ret = bdrv_pread(bs->file->bs, 0, &ph, sizeof(ph));
581 bs->total_sectors = le64_to_cpu(ph.nb_sectors);
583 if (le32_to_cpu(ph.version) != HEADER_VERSION) {
586 if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
587 s->off_multiplier = 1;
588 bs->total_sectors = 0xffffffff & bs->total_sectors;
589 } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
590 s->off_multiplier = le32_to_cpu(ph.tracks);
595 s->tracks = le32_to_cpu(ph.tracks);
596 if (s->tracks == 0) {
597 error_setg(errp, "Invalid image: Zero sectors per track");
601 if (s->tracks > INT32_MAX/513) {
602 error_setg(errp, "Invalid image: Too big cluster");
607 s->bat_size = le32_to_cpu(ph.bat_entries);
608 if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
609 error_setg(errp, "Catalog too large");
614 size = bat_entry_off(s->bat_size);
615 s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
616 s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
617 if (s->header == NULL) {
621 s->data_end = le32_to_cpu(ph.data_off);
622 if (s->data_end == 0) {
623 s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
625 if (s->data_end < s->header_size) {
626 /* there is not enough unused space to fit to block align between BAT
627 and actual data. We can't avoid read-modify-write... */
628 s->header_size = size;
631 ret = bdrv_pread(bs->file->bs, 0, s->header, s->header_size);
635 s->bat_bitmap = (uint32_t *)(s->header + 1);
637 for (i = 0; i < s->bat_size; i++) {
638 int64_t off = bat2sect(s, i);
639 if (off >= s->data_end) {
640 s->data_end = off + s->tracks;
644 if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
645 /* Image was not closed correctly. The check is mandatory */
646 s->header_unclean = true;
647 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
648 error_setg(errp, "parallels: Image was not closed correctly; "
649 "cannot be opened read/write");
655 opts = qemu_opts_create(¶llels_runtime_opts, NULL, 0, &local_err);
656 if (local_err != NULL) {
660 qemu_opts_absorb_qdict(opts, options, &local_err);
661 if (local_err != NULL) {
666 qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
667 s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
668 buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
669 s->prealloc_mode = qapi_enum_parse(prealloc_mode_lookup, buf,
670 PRL_PREALLOC_MODE__MAX, PRL_PREALLOC_MODE_FALLOCATE, &local_err);
672 if (local_err != NULL) {
675 if (!bdrv_has_zero_init(bs->file->bs) ||
676 bdrv_truncate(bs->file->bs, bdrv_getlength(bs->file->bs)) != 0) {
677 s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
680 if (flags & BDRV_O_RDWR) {
681 s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
682 ret = parallels_update_header(bs);
688 s->bat_dirty_block = 4 * getpagesize();
690 bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block));
692 qemu_co_mutex_init(&s->lock);
696 error_setg(errp, "Image not in Parallels format");
699 qemu_vfree(s->header);
703 error_propagate(errp, local_err);
709 static void parallels_close(BlockDriverState *bs)
711 BDRVParallelsState *s = bs->opaque;
713 if (bs->open_flags & BDRV_O_RDWR) {
714 s->header->inuse = 0;
715 parallels_update_header(bs);
718 if (bs->open_flags & BDRV_O_RDWR) {
719 bdrv_truncate(bs->file->bs, s->data_end << BDRV_SECTOR_BITS);
722 g_free(s->bat_dirty_bmap);
723 qemu_vfree(s->header);
726 static QemuOptsList parallels_create_opts = {
727 .name = "parallels-create-opts",
728 .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head),
731 .name = BLOCK_OPT_SIZE,
732 .type = QEMU_OPT_SIZE,
733 .help = "Virtual disk size",
736 .name = BLOCK_OPT_CLUSTER_SIZE,
737 .type = QEMU_OPT_SIZE,
738 .help = "Parallels image cluster size",
739 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE),
741 { /* end of list */ }
745 static BlockDriver bdrv_parallels = {
746 .format_name = "parallels",
747 .instance_size = sizeof(BDRVParallelsState),
748 .bdrv_probe = parallels_probe,
749 .bdrv_open = parallels_open,
750 .bdrv_close = parallels_close,
751 .bdrv_co_get_block_status = parallels_co_get_block_status,
752 .bdrv_has_zero_init = bdrv_has_zero_init_1,
753 .bdrv_co_flush_to_os = parallels_co_flush_to_os,
754 .bdrv_co_readv = parallels_co_readv,
755 .bdrv_co_writev = parallels_co_writev,
757 .bdrv_create = parallels_create,
758 .bdrv_check = parallels_check,
759 .create_opts = ¶llels_create_opts,
762 static void bdrv_parallels_init(void)
764 bdrv_register(&bdrv_parallels);
767 block_init(bdrv_parallels_init);