xref: /openbmc/qemu/block/parallels.c (revision 095859e5)
1 /*
2  * Block driver for Parallels disk image format
3  *
4  * Copyright (c) 2007 Alex Beregszaszi
5  * Copyright (c) 2015 Denis V. Lunev <den@openvz.org>
6  *
7  * This code was originally based on comparing different disk images created
8  * by Parallels. Currently it is based on opened OpenVZ sources
9  * available at
10  *     http://git.openvz.org/?p=ploop;a=summary
11  *
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:
18  *
19  * The above copyright notice and this permission notice shall be included in
20  * all copies or substantial portions of the Software.
21  *
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
28  * THE SOFTWARE.
29  */
30 
31 #include "qemu/osdep.h"
32 #include "qemu/error-report.h"
33 #include "qapi/error.h"
34 #include "block/block_int.h"
35 #include "block/qdict.h"
36 #include "sysemu/block-backend.h"
37 #include "qemu/module.h"
38 #include "qemu/option.h"
39 #include "qapi/qmp/qdict.h"
40 #include "qapi/qobject-input-visitor.h"
41 #include "qapi/qapi-visit-block-core.h"
42 #include "qemu/bswap.h"
43 #include "qemu/bitmap.h"
44 #include "qemu/memalign.h"
45 #include "migration/blocker.h"
46 #include "parallels.h"
47 
48 /**************************************************************/
49 
50 #define HEADER_MAGIC "WithoutFreeSpace"
51 #define HEADER_MAGIC2 "WithouFreSpacExt"
52 #define HEADER_VERSION 2
53 #define HEADER_INUSE_MAGIC  (0x746F6E59)
54 #define MAX_PARALLELS_IMAGE_FACTOR (1ull << 32)
55 
56 static QEnumLookup prealloc_mode_lookup = {
57     .array = (const char *const[]) {
58         "falloc",
59         "truncate",
60     },
61     .size = PRL_PREALLOC_MODE__MAX
62 };
63 
64 #define PARALLELS_OPT_PREALLOC_MODE     "prealloc-mode"
65 #define PARALLELS_OPT_PREALLOC_SIZE     "prealloc-size"
66 
67 static QemuOptsList parallels_runtime_opts = {
68     .name = "parallels",
69     .head = QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts.head),
70     .desc = {
71         {
72             .name = PARALLELS_OPT_PREALLOC_SIZE,
73             .type = QEMU_OPT_SIZE,
74             .help = "Preallocation size on image expansion",
75             .def_value_str = "128M",
76         },
77         {
78             .name = PARALLELS_OPT_PREALLOC_MODE,
79             .type = QEMU_OPT_STRING,
80             .help = "Preallocation mode on image expansion "
81                     "(allowed values: falloc, truncate)",
82             .def_value_str = "falloc",
83         },
84         { /* end of list */ },
85     },
86 };
87 
88 static QemuOptsList parallels_create_opts = {
89     .name = "parallels-create-opts",
90     .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head),
91     .desc = {
92         {
93             .name = BLOCK_OPT_SIZE,
94             .type = QEMU_OPT_SIZE,
95             .help = "Virtual disk size",
96         },
97         {
98             .name = BLOCK_OPT_CLUSTER_SIZE,
99             .type = QEMU_OPT_SIZE,
100             .help = "Parallels image cluster size",
101             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE),
102         },
103         { /* end of list */ }
104     }
105 };
106 
107 
108 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx)
109 {
110     return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier;
111 }
112 
113 static uint32_t bat_entry_off(uint32_t idx)
114 {
115     return sizeof(ParallelsHeader) + sizeof(uint32_t) * idx;
116 }
117 
118 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num)
119 {
120     uint32_t index, offset;
121 
122     index = sector_num / s->tracks;
123     offset = sector_num % s->tracks;
124 
125     /* not allocated */
126     if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) {
127         return -1;
128     }
129     return bat2sect(s, index) + offset;
130 }
131 
132 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num,
133         int nb_sectors)
134 {
135     int ret = s->tracks - sector_num % s->tracks;
136     return MIN(nb_sectors, ret);
137 }
138 
139 static uint32_t host_cluster_index(BDRVParallelsState *s, int64_t off)
140 {
141     off -= s->data_start << BDRV_SECTOR_BITS;
142     return off / s->cluster_size;
143 }
144 
145 static int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
146                             int nb_sectors, int *pnum)
147 {
148     int64_t start_off = -2, prev_end_off = -2;
149 
150     *pnum = 0;
151     while (nb_sectors > 0 || start_off == -2) {
152         int64_t offset = seek_to_sector(s, sector_num);
153         int to_end;
154 
155         if (start_off == -2) {
156             start_off = offset;
157             prev_end_off = offset;
158         } else if (offset != prev_end_off) {
159             break;
160         }
161 
162         to_end = cluster_remainder(s, sector_num, nb_sectors);
163         nb_sectors -= to_end;
164         sector_num += to_end;
165         *pnum += to_end;
166 
167         if (offset > 0) {
168             prev_end_off += to_end;
169         }
170     }
171     return start_off;
172 }
173 
174 static void parallels_set_bat_entry(BDRVParallelsState *s,
175                                     uint32_t index, uint32_t offset)
176 {
177     s->bat_bitmap[index] = cpu_to_le32(offset);
178     bitmap_set(s->bat_dirty_bmap, bat_entry_off(index) / s->bat_dirty_block, 1);
179 }
180 
181 static int64_t coroutine_fn GRAPH_RDLOCK
182 allocate_clusters(BlockDriverState *bs, int64_t sector_num,
183                   int nb_sectors, int *pnum)
184 {
185     int ret = 0;
186     BDRVParallelsState *s = bs->opaque;
187     int64_t pos, space, idx, to_allocate, i, len;
188 
189     pos = block_status(s, sector_num, nb_sectors, pnum);
190     if (pos > 0) {
191         return pos;
192     }
193 
194     idx = sector_num / s->tracks;
195     to_allocate = DIV_ROUND_UP(sector_num + *pnum, s->tracks) - idx;
196 
197     /*
198      * This function is called only by parallels_co_writev(), which will never
199      * pass a sector_num at or beyond the end of the image (because the block
200      * layer never passes such a sector_num to that function). Therefore, idx
201      * is always below s->bat_size.
202      * block_status() will limit *pnum so that sector_num + *pnum will not
203      * exceed the image end. Therefore, idx + to_allocate cannot exceed
204      * s->bat_size.
205      * Note that s->bat_size is an unsigned int, therefore idx + to_allocate
206      * will always fit into a uint32_t.
207      */
208     assert(idx < s->bat_size && idx + to_allocate <= s->bat_size);
209 
210     space = to_allocate * s->tracks;
211     len = bdrv_co_getlength(bs->file->bs);
212     if (len < 0) {
213         return len;
214     }
215     if (s->data_end + space > (len >> BDRV_SECTOR_BITS)) {
216         space += s->prealloc_size;
217         /*
218          * We require the expanded size to read back as zero. If the
219          * user permitted truncation, we try that; but if it fails, we
220          * force the safer-but-slower fallocate.
221          */
222         if (s->prealloc_mode == PRL_PREALLOC_MODE_TRUNCATE) {
223             ret = bdrv_co_truncate(bs->file,
224                                    (s->data_end + space) << BDRV_SECTOR_BITS,
225                                    false, PREALLOC_MODE_OFF,
226                                    BDRV_REQ_ZERO_WRITE, NULL);
227             if (ret == -ENOTSUP) {
228                 s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
229             }
230         }
231         if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) {
232             ret = bdrv_co_pwrite_zeroes(bs->file,
233                                         s->data_end << BDRV_SECTOR_BITS,
234                                         space << BDRV_SECTOR_BITS, 0);
235         }
236         if (ret < 0) {
237             return ret;
238         }
239     }
240 
241     /*
242      * Try to read from backing to fill empty clusters
243      * FIXME: 1. previous write_zeroes may be redundant
244      *        2. most of data we read from backing will be rewritten by
245      *           parallels_co_writev. On aligned-to-cluster write we do not need
246      *           this read at all.
247      *        3. it would be good to combine write of data from backing and new
248      *           data into one write call.
249      */
250     if (bs->backing) {
251         int64_t nb_cow_sectors = to_allocate * s->tracks;
252         int64_t nb_cow_bytes = nb_cow_sectors << BDRV_SECTOR_BITS;
253         void *buf = qemu_blockalign(bs, nb_cow_bytes);
254 
255         ret = bdrv_co_pread(bs->backing, idx * s->tracks * BDRV_SECTOR_SIZE,
256                             nb_cow_bytes, buf, 0);
257         if (ret < 0) {
258             qemu_vfree(buf);
259             return ret;
260         }
261 
262         ret = bdrv_co_pwrite(bs->file, s->data_end * BDRV_SECTOR_SIZE,
263                              nb_cow_bytes, buf, 0);
264         qemu_vfree(buf);
265         if (ret < 0) {
266             return ret;
267         }
268     }
269 
270     for (i = 0; i < to_allocate; i++) {
271         parallels_set_bat_entry(s, idx + i, s->data_end / s->off_multiplier);
272         s->data_end += s->tracks;
273     }
274 
275     return bat2sect(s, idx) + sector_num % s->tracks;
276 }
277 
278 
279 static int coroutine_fn GRAPH_RDLOCK
280 parallels_co_flush_to_os(BlockDriverState *bs)
281 {
282     BDRVParallelsState *s = bs->opaque;
283     unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block);
284     unsigned long bit;
285 
286     qemu_co_mutex_lock(&s->lock);
287 
288     bit = find_first_bit(s->bat_dirty_bmap, size);
289     while (bit < size) {
290         uint32_t off = bit * s->bat_dirty_block;
291         uint32_t to_write = s->bat_dirty_block;
292         int ret;
293 
294         if (off + to_write > s->header_size) {
295             to_write = s->header_size - off;
296         }
297         ret = bdrv_co_pwrite(bs->file, off, to_write,
298                              (uint8_t *)s->header + off, 0);
299         if (ret < 0) {
300             qemu_co_mutex_unlock(&s->lock);
301             return ret;
302         }
303         bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1);
304     }
305     bitmap_zero(s->bat_dirty_bmap, size);
306 
307     qemu_co_mutex_unlock(&s->lock);
308     return 0;
309 }
310 
311 
312 static int coroutine_fn parallels_co_block_status(BlockDriverState *bs,
313                                                   bool want_zero,
314                                                   int64_t offset,
315                                                   int64_t bytes,
316                                                   int64_t *pnum,
317                                                   int64_t *map,
318                                                   BlockDriverState **file)
319 {
320     BDRVParallelsState *s = bs->opaque;
321     int count;
322 
323     assert(QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE));
324     qemu_co_mutex_lock(&s->lock);
325     offset = block_status(s, offset >> BDRV_SECTOR_BITS,
326                           bytes >> BDRV_SECTOR_BITS, &count);
327     qemu_co_mutex_unlock(&s->lock);
328 
329     *pnum = count * BDRV_SECTOR_SIZE;
330     if (offset < 0) {
331         return 0;
332     }
333 
334     *map = offset * BDRV_SECTOR_SIZE;
335     *file = bs->file->bs;
336     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
337 }
338 
339 static int coroutine_fn GRAPH_RDLOCK
340 parallels_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
341                     QEMUIOVector *qiov, int flags)
342 {
343     BDRVParallelsState *s = bs->opaque;
344     uint64_t bytes_done = 0;
345     QEMUIOVector hd_qiov;
346     int ret = 0;
347 
348     qemu_iovec_init(&hd_qiov, qiov->niov);
349 
350     while (nb_sectors > 0) {
351         int64_t position;
352         int n, nbytes;
353 
354         qemu_co_mutex_lock(&s->lock);
355         position = allocate_clusters(bs, sector_num, nb_sectors, &n);
356         qemu_co_mutex_unlock(&s->lock);
357         if (position < 0) {
358             ret = (int)position;
359             break;
360         }
361 
362         nbytes = n << BDRV_SECTOR_BITS;
363 
364         qemu_iovec_reset(&hd_qiov);
365         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
366 
367         ret = bdrv_co_pwritev(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
368                               &hd_qiov, 0);
369         if (ret < 0) {
370             break;
371         }
372 
373         nb_sectors -= n;
374         sector_num += n;
375         bytes_done += nbytes;
376     }
377 
378     qemu_iovec_destroy(&hd_qiov);
379     return ret;
380 }
381 
382 static int coroutine_fn GRAPH_RDLOCK
383 parallels_co_readv(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
384                    QEMUIOVector *qiov)
385 {
386     BDRVParallelsState *s = bs->opaque;
387     uint64_t bytes_done = 0;
388     QEMUIOVector hd_qiov;
389     int ret = 0;
390 
391     qemu_iovec_init(&hd_qiov, qiov->niov);
392 
393     while (nb_sectors > 0) {
394         int64_t position;
395         int n, nbytes;
396 
397         qemu_co_mutex_lock(&s->lock);
398         position = block_status(s, sector_num, nb_sectors, &n);
399         qemu_co_mutex_unlock(&s->lock);
400 
401         nbytes = n << BDRV_SECTOR_BITS;
402 
403         qemu_iovec_reset(&hd_qiov);
404         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
405 
406         if (position < 0) {
407             if (bs->backing) {
408                 ret = bdrv_co_preadv(bs->backing, sector_num * BDRV_SECTOR_SIZE,
409                                      nbytes, &hd_qiov, 0);
410                 if (ret < 0) {
411                     break;
412                 }
413             } else {
414                 qemu_iovec_memset(&hd_qiov, 0, 0, nbytes);
415             }
416         } else {
417             ret = bdrv_co_preadv(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
418                                  &hd_qiov, 0);
419             if (ret < 0) {
420                 break;
421             }
422         }
423 
424         nb_sectors -= n;
425         sector_num += n;
426         bytes_done += nbytes;
427     }
428 
429     qemu_iovec_destroy(&hd_qiov);
430     return ret;
431 }
432 
433 static void parallels_check_unclean(BlockDriverState *bs,
434                                     BdrvCheckResult *res,
435                                     BdrvCheckMode fix)
436 {
437     BDRVParallelsState *s = bs->opaque;
438 
439     if (!s->header_unclean) {
440         return;
441     }
442 
443     fprintf(stderr, "%s image was not closed correctly\n",
444             fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
445     res->corruptions++;
446     if (fix & BDRV_FIX_ERRORS) {
447         /* parallels_close will do the job right */
448         res->corruptions_fixed++;
449         s->header_unclean = false;
450     }
451 }
452 
453 /*
454  * Returns true if data_off is correct, otherwise false. In both cases
455  * correct_offset is set to the proper value.
456  */
457 static bool parallels_test_data_off(BDRVParallelsState *s,
458                                     int64_t file_nb_sectors,
459                                     uint32_t *correct_offset)
460 {
461     uint32_t data_off, min_off;
462     bool old_magic;
463 
464     /*
465      * There are two slightly different image formats: with "WithoutFreeSpace"
466      * or "WithouFreSpacExt" magic words. Call the first one as "old magic".
467      * In such images data_off field can be zero. In this case the offset is
468      * calculated as the end of BAT table plus some padding to ensure sector
469      * size alignment.
470      */
471     old_magic = !memcmp(s->header->magic, HEADER_MAGIC, 16);
472 
473     min_off = DIV_ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
474     if (!old_magic) {
475         min_off = ROUND_UP(min_off, s->cluster_size / BDRV_SECTOR_SIZE);
476     }
477 
478     if (correct_offset) {
479         *correct_offset = min_off;
480     }
481 
482     data_off = le32_to_cpu(s->header->data_off);
483     if (data_off == 0 && old_magic) {
484         return true;
485     }
486 
487     if (data_off < min_off || data_off > file_nb_sectors) {
488         return false;
489     }
490 
491     if (correct_offset) {
492         *correct_offset = data_off;
493     }
494 
495     return true;
496 }
497 
498 static int coroutine_fn GRAPH_RDLOCK
499 parallels_check_data_off(BlockDriverState *bs, BdrvCheckResult *res,
500                          BdrvCheckMode fix)
501 {
502     BDRVParallelsState *s = bs->opaque;
503     int64_t file_size;
504     uint32_t data_off;
505 
506     file_size = bdrv_co_nb_sectors(bs->file->bs);
507     if (file_size < 0) {
508         res->check_errors++;
509         return file_size;
510     }
511 
512     if (parallels_test_data_off(s, file_size, &data_off)) {
513         return 0;
514     }
515 
516     res->corruptions++;
517     if (fix & BDRV_FIX_ERRORS) {
518         s->header->data_off = cpu_to_le32(data_off);
519         res->corruptions_fixed++;
520     }
521 
522     fprintf(stderr, "%s data_off field has incorrect value\n",
523             fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
524 
525     return 0;
526 }
527 
528 static int coroutine_fn GRAPH_RDLOCK
529 parallels_check_outside_image(BlockDriverState *bs, BdrvCheckResult *res,
530                               BdrvCheckMode fix)
531 {
532     BDRVParallelsState *s = bs->opaque;
533     uint32_t i;
534     int64_t off, high_off, size;
535 
536     size = bdrv_co_getlength(bs->file->bs);
537     if (size < 0) {
538         res->check_errors++;
539         return size;
540     }
541 
542     high_off = 0;
543     for (i = 0; i < s->bat_size; i++) {
544         off = bat2sect(s, i) << BDRV_SECTOR_BITS;
545         if (off + s->cluster_size > size) {
546             fprintf(stderr, "%s cluster %u is outside image\n",
547                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
548             res->corruptions++;
549             if (fix & BDRV_FIX_ERRORS) {
550                 parallels_set_bat_entry(s, i, 0);
551                 res->corruptions_fixed++;
552             }
553             continue;
554         }
555         if (high_off < off) {
556             high_off = off;
557         }
558     }
559 
560     if (high_off == 0) {
561         res->image_end_offset = s->data_end << BDRV_SECTOR_BITS;
562     } else {
563         res->image_end_offset = high_off + s->cluster_size;
564         s->data_end = res->image_end_offset >> BDRV_SECTOR_BITS;
565     }
566 
567     return 0;
568 }
569 
570 static int coroutine_fn GRAPH_RDLOCK
571 parallels_check_leak(BlockDriverState *bs, BdrvCheckResult *res,
572                      BdrvCheckMode fix, bool explicit)
573 {
574     BDRVParallelsState *s = bs->opaque;
575     int64_t size;
576     int ret;
577 
578     size = bdrv_co_getlength(bs->file->bs);
579     if (size < 0) {
580         res->check_errors++;
581         return size;
582     }
583 
584     if (size > res->image_end_offset) {
585         int64_t count;
586         count = DIV_ROUND_UP(size - res->image_end_offset, s->cluster_size);
587         if (explicit) {
588             fprintf(stderr,
589                     "%s space leaked at the end of the image %" PRId64 "\n",
590                     fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
591                     size - res->image_end_offset);
592             res->leaks += count;
593         }
594         if (fix & BDRV_FIX_LEAKS) {
595             Error *local_err = NULL;
596 
597             /*
598              * In order to really repair the image, we must shrink it.
599              * That means we have to pass exact=true.
600              */
601             ret = bdrv_co_truncate(bs->file, res->image_end_offset, true,
602                                    PREALLOC_MODE_OFF, 0, &local_err);
603             if (ret < 0) {
604                 error_report_err(local_err);
605                 res->check_errors++;
606                 return ret;
607             }
608             if (explicit) {
609                 res->leaks_fixed += count;
610             }
611         }
612     }
613 
614     return 0;
615 }
616 
617 static int coroutine_fn GRAPH_RDLOCK
618 parallels_check_duplicate(BlockDriverState *bs, BdrvCheckResult *res,
619                           BdrvCheckMode fix)
620 {
621     BDRVParallelsState *s = bs->opaque;
622     int64_t host_off, host_sector, guest_sector;
623     unsigned long *bitmap;
624     uint32_t i, bitmap_size, cluster_index, bat_entry;
625     int n, ret = 0;
626     uint64_t *buf = NULL;
627     bool fixed = false;
628 
629     /*
630      * Create a bitmap of used clusters.
631      * If a bit is set, there is a BAT entry pointing to this cluster.
632      * Loop through the BAT entries, check bits relevant to an entry offset.
633      * If bit is set, this entry is duplicated. Otherwise set the bit.
634      *
635      * We shouldn't worry about newly allocated clusters outside the image
636      * because they are created higher then any existing cluster pointed by
637      * a BAT entry.
638      */
639     bitmap_size = host_cluster_index(s, res->image_end_offset);
640     if (bitmap_size == 0) {
641         return 0;
642     }
643     if (res->image_end_offset % s->cluster_size) {
644         /* A not aligned image end leads to a bitmap shorter by 1 */
645         bitmap_size++;
646     }
647 
648     bitmap = bitmap_new(bitmap_size);
649 
650     buf = qemu_blockalign(bs, s->cluster_size);
651 
652     for (i = 0; i < s->bat_size; i++) {
653         host_off = bat2sect(s, i) << BDRV_SECTOR_BITS;
654         if (host_off == 0) {
655             continue;
656         }
657 
658         cluster_index = host_cluster_index(s, host_off);
659         assert(cluster_index < bitmap_size);
660         if (!test_bit(cluster_index, bitmap)) {
661             bitmap_set(bitmap, cluster_index, 1);
662             continue;
663         }
664 
665         /* this cluster duplicates another one */
666         fprintf(stderr, "%s duplicate offset in BAT entry %u\n",
667                 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
668 
669         res->corruptions++;
670 
671         if (!(fix & BDRV_FIX_ERRORS)) {
672             continue;
673         }
674 
675         /*
676          * Reset the entry and allocate a new cluster
677          * for the relevant guest offset. In this way we let
678          * the lower layer to place the new cluster properly.
679          * Copy the original cluster to the allocated one.
680          * But before save the old offset value for repairing
681          * if we have an error.
682          */
683         bat_entry = s->bat_bitmap[i];
684         parallels_set_bat_entry(s, i, 0);
685 
686         ret = bdrv_co_pread(bs->file, host_off, s->cluster_size, buf, 0);
687         if (ret < 0) {
688             res->check_errors++;
689             goto out_repair_bat;
690         }
691 
692         guest_sector = (i * (int64_t)s->cluster_size) >> BDRV_SECTOR_BITS;
693         host_sector = allocate_clusters(bs, guest_sector, s->tracks, &n);
694         if (host_sector < 0) {
695             res->check_errors++;
696             goto out_repair_bat;
697         }
698         host_off = host_sector << BDRV_SECTOR_BITS;
699 
700         ret = bdrv_co_pwrite(bs->file, host_off, s->cluster_size, buf, 0);
701         if (ret < 0) {
702             res->check_errors++;
703             goto out_repair_bat;
704         }
705 
706         if (host_off + s->cluster_size > res->image_end_offset) {
707             res->image_end_offset = host_off + s->cluster_size;
708         }
709 
710         /*
711          * In the future allocate_cluster() will reuse holed offsets
712          * inside the image. Keep the used clusters bitmap content
713          * consistent for the new allocated clusters too.
714          *
715          * Note, clusters allocated outside the current image are not
716          * considered, and the bitmap size doesn't change.
717          */
718         cluster_index = host_cluster_index(s, host_off);
719         if (cluster_index < bitmap_size) {
720             bitmap_set(bitmap, cluster_index, 1);
721         }
722 
723         fixed = true;
724         res->corruptions_fixed++;
725 
726     }
727 
728     if (fixed) {
729         /*
730          * When new clusters are allocated, the file size increases by
731          * 128 Mb. We need to truncate the file to the right size. Let
732          * the leak fix code make its job without res changing.
733          */
734         ret = parallels_check_leak(bs, res, fix, false);
735     }
736 
737 out_free:
738     g_free(buf);
739     g_free(bitmap);
740     return ret;
741 /*
742  * We can get here only from places where index and old_offset have
743  * meaningful values.
744  */
745 out_repair_bat:
746     s->bat_bitmap[i] = bat_entry;
747     goto out_free;
748 }
749 
750 static void parallels_collect_statistics(BlockDriverState *bs,
751                                          BdrvCheckResult *res,
752                                          BdrvCheckMode fix)
753 {
754     BDRVParallelsState *s = bs->opaque;
755     int64_t off, prev_off;
756     uint32_t i;
757 
758     res->bfi.total_clusters = s->bat_size;
759     res->bfi.compressed_clusters = 0; /* compression is not supported */
760 
761     prev_off = 0;
762     for (i = 0; i < s->bat_size; i++) {
763         off = bat2sect(s, i) << BDRV_SECTOR_BITS;
764         /*
765          * If BDRV_FIX_ERRORS is not set, out-of-image BAT entries were not
766          * fixed. Skip not allocated and out-of-image BAT entries.
767          */
768         if (off == 0 || off + s->cluster_size > res->image_end_offset) {
769             prev_off = 0;
770             continue;
771         }
772 
773         if (prev_off != 0 && (prev_off + s->cluster_size) != off) {
774             res->bfi.fragmented_clusters++;
775         }
776         prev_off = off;
777         res->bfi.allocated_clusters++;
778     }
779 }
780 
781 static int coroutine_fn GRAPH_RDLOCK
782 parallels_co_check(BlockDriverState *bs, BdrvCheckResult *res,
783                    BdrvCheckMode fix)
784 {
785     BDRVParallelsState *s = bs->opaque;
786     int ret;
787 
788     WITH_QEMU_LOCK_GUARD(&s->lock) {
789         parallels_check_unclean(bs, res, fix);
790 
791         ret = parallels_check_data_off(bs, res, fix);
792         if (ret < 0) {
793             return ret;
794         }
795 
796         ret = parallels_check_outside_image(bs, res, fix);
797         if (ret < 0) {
798             return ret;
799         }
800 
801         ret = parallels_check_leak(bs, res, fix, true);
802         if (ret < 0) {
803             return ret;
804         }
805 
806         ret = parallels_check_duplicate(bs, res, fix);
807         if (ret < 0) {
808             return ret;
809         }
810 
811         parallels_collect_statistics(bs, res, fix);
812     }
813 
814     ret = bdrv_co_flush(bs);
815     if (ret < 0) {
816         res->check_errors++;
817     }
818 
819     return ret;
820 }
821 
822 
823 static int coroutine_fn GRAPH_UNLOCKED
824 parallels_co_create(BlockdevCreateOptions* opts, Error **errp)
825 {
826     BlockdevCreateOptionsParallels *parallels_opts;
827     BlockDriverState *bs;
828     BlockBackend *blk;
829     int64_t total_size, cl_size;
830     uint32_t bat_entries, bat_sectors;
831     ParallelsHeader header;
832     uint8_t tmp[BDRV_SECTOR_SIZE];
833     int ret;
834 
835     assert(opts->driver == BLOCKDEV_DRIVER_PARALLELS);
836     parallels_opts = &opts->u.parallels;
837 
838     /* Sanity checks */
839     total_size = parallels_opts->size;
840 
841     if (parallels_opts->has_cluster_size) {
842         cl_size = parallels_opts->cluster_size;
843     } else {
844         cl_size = DEFAULT_CLUSTER_SIZE;
845     }
846 
847     /* XXX What is the real limit here? This is an insanely large maximum. */
848     if (cl_size >= INT64_MAX / MAX_PARALLELS_IMAGE_FACTOR) {
849         error_setg(errp, "Cluster size is too large");
850         return -EINVAL;
851     }
852     if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) {
853         error_setg(errp, "Image size is too large for this cluster size");
854         return -E2BIG;
855     }
856 
857     if (!QEMU_IS_ALIGNED(total_size, BDRV_SECTOR_SIZE)) {
858         error_setg(errp, "Image size must be a multiple of 512 bytes");
859         return -EINVAL;
860     }
861 
862     if (!QEMU_IS_ALIGNED(cl_size, BDRV_SECTOR_SIZE)) {
863         error_setg(errp, "Cluster size must be a multiple of 512 bytes");
864         return -EINVAL;
865     }
866 
867     /* Create BlockBackend to write to the image */
868     bs = bdrv_co_open_blockdev_ref(parallels_opts->file, errp);
869     if (bs == NULL) {
870         return -EIO;
871     }
872 
873     blk = blk_co_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
874                              errp);
875     if (!blk) {
876         ret = -EPERM;
877         goto out;
878     }
879     blk_set_allow_write_beyond_eof(blk, true);
880 
881     /* Create image format */
882     bat_entries = DIV_ROUND_UP(total_size, cl_size);
883     bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
884     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
885 
886     memset(&header, 0, sizeof(header));
887     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
888     header.version = cpu_to_le32(HEADER_VERSION);
889     /* don't care much about geometry, it is not used on image level */
890     header.heads = cpu_to_le32(HEADS_NUMBER);
891     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE
892                                    / HEADS_NUMBER / SEC_IN_CYL);
893     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
894     header.bat_entries = cpu_to_le32(bat_entries);
895     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
896     header.data_off = cpu_to_le32(bat_sectors);
897 
898     /* write all the data */
899     memset(tmp, 0, sizeof(tmp));
900     memcpy(tmp, &header, sizeof(header));
901 
902     ret = blk_co_pwrite(blk, 0, BDRV_SECTOR_SIZE, tmp, 0);
903     if (ret < 0) {
904         goto exit;
905     }
906     ret = blk_co_pwrite_zeroes(blk, BDRV_SECTOR_SIZE,
907                                (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
908     if (ret < 0) {
909         goto exit;
910     }
911 
912     ret = 0;
913 out:
914     blk_co_unref(blk);
915     bdrv_co_unref(bs);
916     return ret;
917 
918 exit:
919     error_setg_errno(errp, -ret, "Failed to create Parallels image");
920     goto out;
921 }
922 
923 static int coroutine_fn GRAPH_UNLOCKED
924 parallels_co_create_opts(BlockDriver *drv, const char *filename,
925                          QemuOpts *opts, Error **errp)
926 {
927     BlockdevCreateOptions *create_options = NULL;
928     BlockDriverState *bs = NULL;
929     QDict *qdict;
930     Visitor *v;
931     int ret;
932 
933     static const QDictRenames opt_renames[] = {
934         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
935         { NULL, NULL },
936     };
937 
938     /* Parse options and convert legacy syntax */
939     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &parallels_create_opts,
940                                         true);
941 
942     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
943         ret = -EINVAL;
944         goto done;
945     }
946 
947     /* Create and open the file (protocol layer) */
948     ret = bdrv_co_create_file(filename, opts, errp);
949     if (ret < 0) {
950         goto done;
951     }
952 
953     bs = bdrv_co_open(filename, NULL, NULL,
954                       BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
955     if (bs == NULL) {
956         ret = -EIO;
957         goto done;
958     }
959 
960     /* Now get the QAPI type BlockdevCreateOptions */
961     qdict_put_str(qdict, "driver", "parallels");
962     qdict_put_str(qdict, "file", bs->node_name);
963 
964     v = qobject_input_visitor_new_flat_confused(qdict, errp);
965     if (!v) {
966         ret = -EINVAL;
967         goto done;
968     }
969 
970     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
971     visit_free(v);
972     if (!create_options) {
973         ret = -EINVAL;
974         goto done;
975     }
976 
977     /* Silently round up sizes */
978     create_options->u.parallels.size =
979         ROUND_UP(create_options->u.parallels.size, BDRV_SECTOR_SIZE);
980     create_options->u.parallels.cluster_size =
981         ROUND_UP(create_options->u.parallels.cluster_size, BDRV_SECTOR_SIZE);
982 
983     /* Create the Parallels image (format layer) */
984     ret = parallels_co_create(create_options, errp);
985     if (ret < 0) {
986         goto done;
987     }
988     ret = 0;
989 
990 done:
991     qobject_unref(qdict);
992     bdrv_co_unref(bs);
993     qapi_free_BlockdevCreateOptions(create_options);
994     return ret;
995 }
996 
997 
998 static int parallels_probe(const uint8_t *buf, int buf_size,
999                            const char *filename)
1000 {
1001     const ParallelsHeader *ph = (const void *)buf;
1002 
1003     if (buf_size < sizeof(ParallelsHeader)) {
1004         return 0;
1005     }
1006 
1007     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
1008            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
1009            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
1010         return 100;
1011     }
1012 
1013     return 0;
1014 }
1015 
1016 static int parallels_update_header(BlockDriverState *bs)
1017 {
1018     BDRVParallelsState *s = bs->opaque;
1019     unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
1020                         sizeof(ParallelsHeader));
1021 
1022     if (size > s->header_size) {
1023         size = s->header_size;
1024     }
1025     return bdrv_pwrite_sync(bs->file, 0, size, s->header, 0);
1026 }
1027 
1028 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
1029                           Error **errp)
1030 {
1031     BDRVParallelsState *s = bs->opaque;
1032     ParallelsHeader ph;
1033     int ret, size, i;
1034     int64_t file_nb_sectors, sector;
1035     uint32_t data_start;
1036     QemuOpts *opts = NULL;
1037     Error *local_err = NULL;
1038     char *buf;
1039     bool data_off_is_correct;
1040 
1041     ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
1042     if (ret < 0) {
1043         return ret;
1044     }
1045 
1046     file_nb_sectors = bdrv_nb_sectors(bs->file->bs);
1047     if (file_nb_sectors < 0) {
1048         return -EINVAL;
1049     }
1050 
1051     ret = bdrv_pread(bs->file, 0, sizeof(ph), &ph, 0);
1052     if (ret < 0) {
1053         goto fail;
1054     }
1055 
1056     bs->total_sectors = le64_to_cpu(ph.nb_sectors);
1057 
1058     if (le32_to_cpu(ph.version) != HEADER_VERSION) {
1059         goto fail_format;
1060     }
1061     if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
1062         s->off_multiplier = 1;
1063         bs->total_sectors = 0xffffffff & bs->total_sectors;
1064     } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
1065         s->off_multiplier = le32_to_cpu(ph.tracks);
1066     } else {
1067         goto fail_format;
1068     }
1069 
1070     s->tracks = le32_to_cpu(ph.tracks);
1071     if (s->tracks == 0) {
1072         error_setg(errp, "Invalid image: Zero sectors per track");
1073         ret = -EINVAL;
1074         goto fail;
1075     }
1076     if (s->tracks > INT32_MAX/513) {
1077         error_setg(errp, "Invalid image: Too big cluster");
1078         ret = -EFBIG;
1079         goto fail;
1080     }
1081     s->cluster_size = s->tracks << BDRV_SECTOR_BITS;
1082 
1083     s->bat_size = le32_to_cpu(ph.bat_entries);
1084     if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
1085         error_setg(errp, "Catalog too large");
1086         ret = -EFBIG;
1087         goto fail;
1088     }
1089 
1090     size = bat_entry_off(s->bat_size);
1091     s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
1092     s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
1093     if (s->header == NULL) {
1094         ret = -ENOMEM;
1095         goto fail;
1096     }
1097 
1098     ret = bdrv_pread(bs->file, 0, s->header_size, s->header, 0);
1099     if (ret < 0) {
1100         goto fail;
1101     }
1102     s->bat_bitmap = (uint32_t *)(s->header + 1);
1103 
1104     if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
1105         s->header_unclean = true;
1106     }
1107 
1108     data_off_is_correct = parallels_test_data_off(s, file_nb_sectors,
1109                                                   &data_start);
1110     s->data_start = data_start;
1111     s->data_end = s->data_start;
1112     if (s->data_end < (s->header_size >> BDRV_SECTOR_BITS)) {
1113         /*
1114          * There is not enough unused space to fit to block align between BAT
1115          * and actual data. We can't avoid read-modify-write...
1116          */
1117         s->header_size = size;
1118     }
1119 
1120     opts = qemu_opts_create(&parallels_runtime_opts, NULL, 0, errp);
1121     if (!opts) {
1122         goto fail_options;
1123     }
1124 
1125     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1126         goto fail_options;
1127     }
1128 
1129     s->prealloc_size =
1130         qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
1131     s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
1132     buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
1133     /* prealloc_mode can be downgraded later during allocate_clusters */
1134     s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf,
1135                                        PRL_PREALLOC_MODE_FALLOCATE,
1136                                        &local_err);
1137     g_free(buf);
1138     if (local_err != NULL) {
1139         error_propagate(errp, local_err);
1140         goto fail_options;
1141     }
1142 
1143     if (ph.ext_off) {
1144         if (flags & BDRV_O_RDWR) {
1145             /*
1146              * It's unsafe to open image RW if there is an extension (as we
1147              * don't support it). But parallels driver in QEMU historically
1148              * ignores the extension, so print warning and don't care.
1149              */
1150             warn_report("Format Extension ignored in RW mode");
1151         } else {
1152             ret = parallels_read_format_extension(
1153                     bs, le64_to_cpu(ph.ext_off) << BDRV_SECTOR_BITS, errp);
1154             if (ret < 0) {
1155                 goto fail;
1156             }
1157         }
1158     }
1159 
1160     if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_INACTIVE)) {
1161         s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
1162         ret = parallels_update_header(bs);
1163         if (ret < 0) {
1164             goto fail;
1165         }
1166     }
1167 
1168     s->bat_dirty_block = 4 * qemu_real_host_page_size();
1169     s->bat_dirty_bmap =
1170         bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block));
1171 
1172     /* Disable migration until bdrv_activate method is added */
1173     error_setg(&s->migration_blocker, "The Parallels format used by node '%s' "
1174                "does not support live migration",
1175                bdrv_get_device_or_node_name(bs));
1176     ret = migrate_add_blocker(s->migration_blocker, errp);
1177     if (ret < 0) {
1178         error_setg(errp, "Migration blocker error");
1179         goto fail;
1180     }
1181     qemu_co_mutex_init(&s->lock);
1182 
1183     for (i = 0; i < s->bat_size; i++) {
1184         sector = bat2sect(s, i);
1185         if (sector + s->tracks > s->data_end) {
1186             s->data_end = sector + s->tracks;
1187         }
1188     }
1189 
1190     /*
1191      * We don't repair the image here if it's opened for checks. Also we don't
1192      * want to change inactive images and can't change readonly images.
1193      */
1194     if ((flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) || !(flags & BDRV_O_RDWR)) {
1195         return 0;
1196     }
1197 
1198     /*
1199      * Repair the image if it's dirty or
1200      * out-of-image corruption was detected.
1201      */
1202     if (s->data_end > file_nb_sectors || s->header_unclean
1203         || !data_off_is_correct) {
1204         BdrvCheckResult res;
1205         ret = bdrv_check(bs, &res, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1206         if (ret < 0) {
1207             error_setg_errno(errp, -ret, "Could not repair corrupted image");
1208             migrate_del_blocker(s->migration_blocker);
1209             goto fail;
1210         }
1211     }
1212 
1213     return 0;
1214 
1215 fail_format:
1216     error_setg(errp, "Image not in Parallels format");
1217 fail_options:
1218     ret = -EINVAL;
1219 fail:
1220     /*
1221      * "s" object was allocated by g_malloc0 so we can safely
1222      * try to free its fields even they were not allocated.
1223      */
1224     error_free(s->migration_blocker);
1225     g_free(s->bat_dirty_bmap);
1226     qemu_vfree(s->header);
1227     return ret;
1228 }
1229 
1230 
1231 static void parallels_close(BlockDriverState *bs)
1232 {
1233     BDRVParallelsState *s = bs->opaque;
1234 
1235     if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) {
1236         s->header->inuse = 0;
1237         parallels_update_header(bs);
1238 
1239         /* errors are ignored, so we might as well pass exact=true */
1240         bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS, true,
1241                       PREALLOC_MODE_OFF, 0, NULL);
1242     }
1243 
1244     g_free(s->bat_dirty_bmap);
1245     qemu_vfree(s->header);
1246 
1247     migrate_del_blocker(s->migration_blocker);
1248     error_free(s->migration_blocker);
1249 }
1250 
1251 static BlockDriver bdrv_parallels = {
1252     .format_name	= "parallels",
1253     .instance_size	= sizeof(BDRVParallelsState),
1254     .bdrv_probe		= parallels_probe,
1255     .bdrv_open		= parallels_open,
1256     .bdrv_close		= parallels_close,
1257     .bdrv_child_perm          = bdrv_default_perms,
1258     .bdrv_co_block_status     = parallels_co_block_status,
1259     .bdrv_has_zero_init       = bdrv_has_zero_init_1,
1260     .bdrv_co_flush_to_os      = parallels_co_flush_to_os,
1261     .bdrv_co_readv  = parallels_co_readv,
1262     .bdrv_co_writev = parallels_co_writev,
1263     .is_format      = true,
1264     .supports_backing = true,
1265     .bdrv_co_create      = parallels_co_create,
1266     .bdrv_co_create_opts = parallels_co_create_opts,
1267     .bdrv_co_check  = parallels_co_check,
1268     .create_opts    = &parallels_create_opts,
1269 };
1270 
1271 static void bdrv_parallels_init(void)
1272 {
1273     bdrv_register(&bdrv_parallels);
1274 }
1275 
1276 block_init(bdrv_parallels_init);
1277