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[]) {
76 .size = PRL_PREALLOC_MODE__MAX
79 typedef struct BDRVParallelsState {
80 /** Locking is conservative, the lock protects
81 * - image file extending (truncate, fallocate)
82 * - any access to block allocation table
86 ParallelsHeader *header;
90 unsigned long *bat_dirty_bmap;
91 unsigned int bat_dirty_block;
94 unsigned int bat_size;
97 uint64_t prealloc_size;
98 ParallelsPreallocMode prealloc_mode;
102 unsigned int off_multiplier;
103 } BDRVParallelsState;
106 #define PARALLELS_OPT_PREALLOC_MODE "prealloc-mode"
107 #define PARALLELS_OPT_PREALLOC_SIZE "prealloc-size"
109 static QemuOptsList parallels_runtime_opts = {
111 .head = QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts.head),
114 .name = PARALLELS_OPT_PREALLOC_SIZE,
115 .type = QEMU_OPT_SIZE,
116 .help = "Preallocation size on image expansion",
117 .def_value_str = "128M",
120 .name = PARALLELS_OPT_PREALLOC_MODE,
121 .type = QEMU_OPT_STRING,
122 .help = "Preallocation mode on image expansion "
123 "(allowed values: falloc, truncate)",
124 .def_value_str = "falloc",
126 { /* end of list */ },
131 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx)
133 return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier;
136 static uint32_t bat_entry_off(uint32_t idx)
138 return sizeof(ParallelsHeader) + sizeof(uint32_t) * idx;
141 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num)
143 uint32_t index, offset;
145 index = sector_num / s->tracks;
146 offset = sector_num % s->tracks;
149 if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) {
152 return bat2sect(s, index) + offset;
155 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num,
158 int ret = s->tracks - sector_num % s->tracks;
159 return MIN(nb_sectors, ret);
162 static int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
163 int nb_sectors, int *pnum)
165 int64_t start_off = -2, prev_end_off = -2;
168 while (nb_sectors > 0 || start_off == -2) {
169 int64_t offset = seek_to_sector(s, sector_num);
172 if (start_off == -2) {
174 prev_end_off = offset;
175 } else if (offset != prev_end_off) {
179 to_end = cluster_remainder(s, sector_num, nb_sectors);
180 nb_sectors -= to_end;
181 sector_num += to_end;
185 prev_end_off += to_end;
191 static int64_t allocate_clusters(BlockDriverState *bs, int64_t sector_num,
192 int nb_sectors, int *pnum)
194 BDRVParallelsState *s = bs->opaque;
195 int64_t pos, space, idx, to_allocate, i, len;
197 pos = block_status(s, sector_num, nb_sectors, pnum);
202 idx = sector_num / s->tracks;
203 to_allocate = DIV_ROUND_UP(sector_num + *pnum, s->tracks) - idx;
205 /* This function is called only by parallels_co_writev(), which will never
206 * pass a sector_num at or beyond the end of the image (because the block
207 * layer never passes such a sector_num to that function). Therefore, idx
208 * is always below s->bat_size.
209 * block_status() will limit *pnum so that sector_num + *pnum will not
210 * exceed the image end. Therefore, idx + to_allocate cannot exceed
212 * Note that s->bat_size is an unsigned int, therefore idx + to_allocate
213 * will always fit into a uint32_t. */
214 assert(idx < s->bat_size && idx + to_allocate <= s->bat_size);
216 space = to_allocate * s->tracks;
217 len = bdrv_getlength(bs->file->bs);
221 if (s->data_end + space > (len >> BDRV_SECTOR_BITS)) {
223 space += s->prealloc_size;
224 if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) {
225 ret = bdrv_pwrite_zeroes(bs->file,
226 s->data_end << BDRV_SECTOR_BITS,
227 space << BDRV_SECTOR_BITS, 0);
229 ret = bdrv_truncate(bs->file,
230 (s->data_end + space) << BDRV_SECTOR_BITS,
231 PREALLOC_MODE_OFF, NULL);
238 for (i = 0; i < to_allocate; i++) {
239 s->bat_bitmap[idx + i] = cpu_to_le32(s->data_end / s->off_multiplier);
240 s->data_end += s->tracks;
241 bitmap_set(s->bat_dirty_bmap,
242 bat_entry_off(idx + i) / s->bat_dirty_block, 1);
245 return bat2sect(s, idx) + sector_num % s->tracks;
249 static coroutine_fn int parallels_co_flush_to_os(BlockDriverState *bs)
251 BDRVParallelsState *s = bs->opaque;
252 unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block);
255 qemu_co_mutex_lock(&s->lock);
257 bit = find_first_bit(s->bat_dirty_bmap, size);
259 uint32_t off = bit * s->bat_dirty_block;
260 uint32_t to_write = s->bat_dirty_block;
263 if (off + to_write > s->header_size) {
264 to_write = s->header_size - off;
266 ret = bdrv_pwrite(bs->file, off, (uint8_t *)s->header + off,
269 qemu_co_mutex_unlock(&s->lock);
272 bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1);
274 bitmap_zero(s->bat_dirty_bmap, size);
276 qemu_co_mutex_unlock(&s->lock);
281 static int64_t coroutine_fn parallels_co_get_block_status(BlockDriverState *bs,
282 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
284 BDRVParallelsState *s = bs->opaque;
287 qemu_co_mutex_lock(&s->lock);
288 offset = block_status(s, sector_num, nb_sectors, pnum);
289 qemu_co_mutex_unlock(&s->lock);
295 *file = bs->file->bs;
296 return (offset << BDRV_SECTOR_BITS) |
297 BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
300 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
301 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
303 BDRVParallelsState *s = bs->opaque;
304 uint64_t bytes_done = 0;
305 QEMUIOVector hd_qiov;
308 qemu_iovec_init(&hd_qiov, qiov->niov);
310 while (nb_sectors > 0) {
314 qemu_co_mutex_lock(&s->lock);
315 position = allocate_clusters(bs, sector_num, nb_sectors, &n);
316 qemu_co_mutex_unlock(&s->lock);
322 nbytes = n << BDRV_SECTOR_BITS;
324 qemu_iovec_reset(&hd_qiov);
325 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
327 ret = bdrv_co_writev(bs->file, position, n, &hd_qiov);
334 bytes_done += nbytes;
337 qemu_iovec_destroy(&hd_qiov);
341 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
342 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
344 BDRVParallelsState *s = bs->opaque;
345 uint64_t bytes_done = 0;
346 QEMUIOVector hd_qiov;
349 qemu_iovec_init(&hd_qiov, qiov->niov);
351 while (nb_sectors > 0) {
355 qemu_co_mutex_lock(&s->lock);
356 position = block_status(s, sector_num, nb_sectors, &n);
357 qemu_co_mutex_unlock(&s->lock);
359 nbytes = n << BDRV_SECTOR_BITS;
362 qemu_iovec_memset(qiov, bytes_done, 0, nbytes);
364 qemu_iovec_reset(&hd_qiov);
365 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
367 ret = bdrv_co_readv(bs->file, position, n, &hd_qiov);
375 bytes_done += nbytes;
378 qemu_iovec_destroy(&hd_qiov);
383 static int parallels_check(BlockDriverState *bs, BdrvCheckResult *res,
386 BDRVParallelsState *s = bs->opaque;
387 int64_t size, prev_off, high_off;
390 bool flush_bat = false;
391 int cluster_size = s->tracks << BDRV_SECTOR_BITS;
393 size = bdrv_getlength(bs->file->bs);
399 if (s->header_unclean) {
400 fprintf(stderr, "%s image was not closed correctly\n",
401 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
403 if (fix & BDRV_FIX_ERRORS) {
404 /* parallels_close will do the job right */
405 res->corruptions_fixed++;
406 s->header_unclean = false;
410 res->bfi.total_clusters = s->bat_size;
411 res->bfi.compressed_clusters = 0; /* compression is not supported */
415 for (i = 0; i < s->bat_size; i++) {
416 int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
422 /* cluster outside the image */
424 fprintf(stderr, "%s cluster %u is outside image\n",
425 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
427 if (fix & BDRV_FIX_ERRORS) {
429 s->bat_bitmap[i] = 0;
430 res->corruptions_fixed++;
436 res->bfi.allocated_clusters++;
437 if (off > high_off) {
441 if (prev_off != 0 && (prev_off + cluster_size) != off) {
442 res->bfi.fragmented_clusters++;
448 ret = bdrv_pwrite_sync(bs->file, 0, s->header, s->header_size);
455 res->image_end_offset = high_off + cluster_size;
456 if (size > res->image_end_offset) {
458 count = DIV_ROUND_UP(size - res->image_end_offset, cluster_size);
459 fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
460 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
461 size - res->image_end_offset);
463 if (fix & BDRV_FIX_LEAKS) {
464 Error *local_err = NULL;
465 ret = bdrv_truncate(bs->file, res->image_end_offset,
466 PREALLOC_MODE_OFF, &local_err);
468 error_report_err(local_err);
472 res->leaks_fixed += count;
480 static int parallels_create(const char *filename, QemuOpts *opts, Error **errp)
482 int64_t total_size, cl_size;
483 uint8_t tmp[BDRV_SECTOR_SIZE];
484 Error *local_err = NULL;
486 uint32_t bat_entries, bat_sectors;
487 ParallelsHeader header;
490 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
492 cl_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
493 DEFAULT_CLUSTER_SIZE), BDRV_SECTOR_SIZE);
494 if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) {
495 error_propagate(errp, local_err);
499 ret = bdrv_create_file(filename, opts, &local_err);
501 error_propagate(errp, local_err);
505 file = blk_new_open(filename, NULL, NULL,
506 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
509 error_propagate(errp, local_err);
513 blk_set_allow_write_beyond_eof(file, true);
515 ret = blk_truncate(file, 0, PREALLOC_MODE_OFF, errp);
520 bat_entries = DIV_ROUND_UP(total_size, cl_size);
521 bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
522 bat_sectors = (bat_sectors * cl_size) >> BDRV_SECTOR_BITS;
524 memset(&header, 0, sizeof(header));
525 memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
526 header.version = cpu_to_le32(HEADER_VERSION);
527 /* don't care much about geometry, it is not used on image level */
528 header.heads = cpu_to_le32(16);
529 header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE / 16 / 32);
530 header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
531 header.bat_entries = cpu_to_le32(bat_entries);
532 header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
533 header.data_off = cpu_to_le32(bat_sectors);
535 /* write all the data */
536 memset(tmp, 0, sizeof(tmp));
537 memcpy(tmp, &header, sizeof(header));
539 ret = blk_pwrite(file, 0, tmp, BDRV_SECTOR_SIZE, 0);
543 ret = blk_pwrite_zeroes(file, BDRV_SECTOR_SIZE,
544 (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
555 error_setg_errno(errp, -ret, "Failed to create Parallels image");
560 static int parallels_probe(const uint8_t *buf, int buf_size,
561 const char *filename)
563 const ParallelsHeader *ph = (const void *)buf;
565 if (buf_size < sizeof(ParallelsHeader)) {
569 if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
570 !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
571 (le32_to_cpu(ph->version) == HEADER_VERSION)) {
578 static int parallels_update_header(BlockDriverState *bs)
580 BDRVParallelsState *s = bs->opaque;
581 unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
582 sizeof(ParallelsHeader));
584 if (size > s->header_size) {
585 size = s->header_size;
587 return bdrv_pwrite_sync(bs->file, 0, s->header, size);
590 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
593 BDRVParallelsState *s = bs->opaque;
596 QemuOpts *opts = NULL;
597 Error *local_err = NULL;
600 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
606 ret = bdrv_pread(bs->file, 0, &ph, sizeof(ph));
611 bs->total_sectors = le64_to_cpu(ph.nb_sectors);
613 if (le32_to_cpu(ph.version) != HEADER_VERSION) {
616 if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
617 s->off_multiplier = 1;
618 bs->total_sectors = 0xffffffff & bs->total_sectors;
619 } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
620 s->off_multiplier = le32_to_cpu(ph.tracks);
625 s->tracks = le32_to_cpu(ph.tracks);
626 if (s->tracks == 0) {
627 error_setg(errp, "Invalid image: Zero sectors per track");
631 if (s->tracks > INT32_MAX/513) {
632 error_setg(errp, "Invalid image: Too big cluster");
637 s->bat_size = le32_to_cpu(ph.bat_entries);
638 if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
639 error_setg(errp, "Catalog too large");
644 size = bat_entry_off(s->bat_size);
645 s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
646 s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
647 if (s->header == NULL) {
651 s->data_end = le32_to_cpu(ph.data_off);
652 if (s->data_end == 0) {
653 s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
655 if (s->data_end < s->header_size) {
656 /* there is not enough unused space to fit to block align between BAT
657 and actual data. We can't avoid read-modify-write... */
658 s->header_size = size;
661 ret = bdrv_pread(bs->file, 0, s->header, s->header_size);
665 s->bat_bitmap = (uint32_t *)(s->header + 1);
667 for (i = 0; i < s->bat_size; i++) {
668 int64_t off = bat2sect(s, i);
669 if (off >= s->data_end) {
670 s->data_end = off + s->tracks;
674 if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
675 /* Image was not closed correctly. The check is mandatory */
676 s->header_unclean = true;
677 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
678 error_setg(errp, "parallels: Image was not closed correctly; "
679 "cannot be opened read/write");
685 opts = qemu_opts_create(¶llels_runtime_opts, NULL, 0, &local_err);
686 if (local_err != NULL) {
690 qemu_opts_absorb_qdict(opts, options, &local_err);
691 if (local_err != NULL) {
696 qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
697 s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
698 buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
699 s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf,
700 PRL_PREALLOC_MODE_FALLOCATE,
703 if (local_err != NULL) {
707 if (!bdrv_has_zero_init(bs->file->bs)) {
708 s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
711 if (flags & BDRV_O_RDWR) {
712 s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
713 ret = parallels_update_header(bs);
719 s->bat_dirty_block = 4 * getpagesize();
721 bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block));
723 qemu_co_mutex_init(&s->lock);
727 error_setg(errp, "Image not in Parallels format");
730 qemu_vfree(s->header);
734 error_propagate(errp, local_err);
740 static void parallels_close(BlockDriverState *bs)
742 BDRVParallelsState *s = bs->opaque;
744 if (bs->open_flags & BDRV_O_RDWR) {
745 s->header->inuse = 0;
746 parallels_update_header(bs);
749 if (bs->open_flags & BDRV_O_RDWR) {
750 bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS,
751 PREALLOC_MODE_OFF, NULL);
754 g_free(s->bat_dirty_bmap);
755 qemu_vfree(s->header);
758 static QemuOptsList parallels_create_opts = {
759 .name = "parallels-create-opts",
760 .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head),
763 .name = BLOCK_OPT_SIZE,
764 .type = QEMU_OPT_SIZE,
765 .help = "Virtual disk size",
768 .name = BLOCK_OPT_CLUSTER_SIZE,
769 .type = QEMU_OPT_SIZE,
770 .help = "Parallels image cluster size",
771 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE),
773 { /* end of list */ }
777 static BlockDriver bdrv_parallels = {
778 .format_name = "parallels",
779 .instance_size = sizeof(BDRVParallelsState),
780 .bdrv_probe = parallels_probe,
781 .bdrv_open = parallels_open,
782 .bdrv_close = parallels_close,
783 .bdrv_child_perm = bdrv_format_default_perms,
784 .bdrv_co_get_block_status = parallels_co_get_block_status,
785 .bdrv_has_zero_init = bdrv_has_zero_init_1,
786 .bdrv_co_flush_to_os = parallels_co_flush_to_os,
787 .bdrv_co_readv = parallels_co_readv,
788 .bdrv_co_writev = parallels_co_writev,
790 .bdrv_create = parallels_create,
791 .bdrv_check = parallels_check,
792 .create_opts = ¶llels_create_opts,
795 static void bdrv_parallels_init(void)
797 bdrv_register(&bdrv_parallels);
800 block_init(bdrv_parallels_init);