xref: /openbmc/qemu/block/parallels.c (revision 03ff4f8d)
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 "migration/blocker.h"
45 #include "parallels.h"
46 
47 /**************************************************************/
48 
49 #define HEADER_MAGIC "WithoutFreeSpace"
50 #define HEADER_MAGIC2 "WithouFreSpacExt"
51 #define HEADER_VERSION 2
52 #define HEADER_INUSE_MAGIC  (0x746F6E59)
53 #define MAX_PARALLELS_IMAGE_FACTOR (1ull << 32)
54 
55 static QEnumLookup prealloc_mode_lookup = {
56     .array = (const char *const[]) {
57         "falloc",
58         "truncate",
59     },
60     .size = PRL_PREALLOC_MODE__MAX
61 };
62 
63 #define PARALLELS_OPT_PREALLOC_MODE     "prealloc-mode"
64 #define PARALLELS_OPT_PREALLOC_SIZE     "prealloc-size"
65 
66 static QemuOptsList parallels_runtime_opts = {
67     .name = "parallels",
68     .head = QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts.head),
69     .desc = {
70         {
71             .name = PARALLELS_OPT_PREALLOC_SIZE,
72             .type = QEMU_OPT_SIZE,
73             .help = "Preallocation size on image expansion",
74             .def_value_str = "128M",
75         },
76         {
77             .name = PARALLELS_OPT_PREALLOC_MODE,
78             .type = QEMU_OPT_STRING,
79             .help = "Preallocation mode on image expansion "
80                     "(allowed values: falloc, truncate)",
81             .def_value_str = "falloc",
82         },
83         { /* end of list */ },
84     },
85 };
86 
87 static QemuOptsList parallels_create_opts = {
88     .name = "parallels-create-opts",
89     .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head),
90     .desc = {
91         {
92             .name = BLOCK_OPT_SIZE,
93             .type = QEMU_OPT_SIZE,
94             .help = "Virtual disk size",
95         },
96         {
97             .name = BLOCK_OPT_CLUSTER_SIZE,
98             .type = QEMU_OPT_SIZE,
99             .help = "Parallels image cluster size",
100             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE),
101         },
102         { /* end of list */ }
103     }
104 };
105 
106 
107 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx)
108 {
109     return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier;
110 }
111 
112 static uint32_t bat_entry_off(uint32_t idx)
113 {
114     return sizeof(ParallelsHeader) + sizeof(uint32_t) * idx;
115 }
116 
117 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num)
118 {
119     uint32_t index, offset;
120 
121     index = sector_num / s->tracks;
122     offset = sector_num % s->tracks;
123 
124     /* not allocated */
125     if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) {
126         return -1;
127     }
128     return bat2sect(s, index) + offset;
129 }
130 
131 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num,
132         int nb_sectors)
133 {
134     int ret = s->tracks - sector_num % s->tracks;
135     return MIN(nb_sectors, ret);
136 }
137 
138 static int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
139                             int nb_sectors, int *pnum)
140 {
141     int64_t start_off = -2, prev_end_off = -2;
142 
143     *pnum = 0;
144     while (nb_sectors > 0 || start_off == -2) {
145         int64_t offset = seek_to_sector(s, sector_num);
146         int to_end;
147 
148         if (start_off == -2) {
149             start_off = offset;
150             prev_end_off = offset;
151         } else if (offset != prev_end_off) {
152             break;
153         }
154 
155         to_end = cluster_remainder(s, sector_num, nb_sectors);
156         nb_sectors -= to_end;
157         sector_num += to_end;
158         *pnum += to_end;
159 
160         if (offset > 0) {
161             prev_end_off += to_end;
162         }
163     }
164     return start_off;
165 }
166 
167 static int64_t allocate_clusters(BlockDriverState *bs, int64_t sector_num,
168                                  int nb_sectors, int *pnum)
169 {
170     int ret = 0;
171     BDRVParallelsState *s = bs->opaque;
172     int64_t pos, space, idx, to_allocate, i, len;
173 
174     pos = block_status(s, sector_num, nb_sectors, pnum);
175     if (pos > 0) {
176         return pos;
177     }
178 
179     idx = sector_num / s->tracks;
180     to_allocate = DIV_ROUND_UP(sector_num + *pnum, s->tracks) - idx;
181 
182     /* This function is called only by parallels_co_writev(), which will never
183      * pass a sector_num at or beyond the end of the image (because the block
184      * layer never passes such a sector_num to that function). Therefore, idx
185      * is always below s->bat_size.
186      * block_status() will limit *pnum so that sector_num + *pnum will not
187      * exceed the image end. Therefore, idx + to_allocate cannot exceed
188      * s->bat_size.
189      * Note that s->bat_size is an unsigned int, therefore idx + to_allocate
190      * will always fit into a uint32_t. */
191     assert(idx < s->bat_size && idx + to_allocate <= s->bat_size);
192 
193     space = to_allocate * s->tracks;
194     len = bdrv_getlength(bs->file->bs);
195     if (len < 0) {
196         return len;
197     }
198     if (s->data_end + space > (len >> BDRV_SECTOR_BITS)) {
199         space += s->prealloc_size;
200         /*
201          * We require the expanded size to read back as zero. If the
202          * user permitted truncation, we try that; but if it fails, we
203          * force the safer-but-slower fallocate.
204          */
205         if (s->prealloc_mode == PRL_PREALLOC_MODE_TRUNCATE) {
206             ret = bdrv_truncate(bs->file,
207                                 (s->data_end + space) << BDRV_SECTOR_BITS,
208                                 false, PREALLOC_MODE_OFF, BDRV_REQ_ZERO_WRITE,
209                                 NULL);
210             if (ret == -ENOTSUP) {
211                 s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
212             }
213         }
214         if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) {
215             ret = bdrv_pwrite_zeroes(bs->file,
216                                      s->data_end << BDRV_SECTOR_BITS,
217                                      space << BDRV_SECTOR_BITS, 0);
218         }
219         if (ret < 0) {
220             return ret;
221         }
222     }
223 
224     /* Try to read from backing to fill empty clusters
225      * FIXME: 1. previous write_zeroes may be redundant
226      *        2. most of data we read from backing will be rewritten by
227      *           parallels_co_writev. On aligned-to-cluster write we do not need
228      *           this read at all.
229      *        3. it would be good to combine write of data from backing and new
230      *           data into one write call */
231     if (bs->backing) {
232         int64_t nb_cow_sectors = to_allocate * s->tracks;
233         int64_t nb_cow_bytes = nb_cow_sectors << BDRV_SECTOR_BITS;
234         void *buf = qemu_blockalign(bs, nb_cow_bytes);
235 
236         ret = bdrv_co_pread(bs->backing, idx * s->tracks * BDRV_SECTOR_SIZE,
237                             nb_cow_bytes, buf, 0);
238         if (ret < 0) {
239             qemu_vfree(buf);
240             return ret;
241         }
242 
243         ret = bdrv_co_pwritev(bs->file, s->data_end * BDRV_SECTOR_SIZE,
244                               nb_cow_bytes, buf, 0);
245         qemu_vfree(buf);
246         if (ret < 0) {
247             return ret;
248         }
249     }
250 
251     for (i = 0; i < to_allocate; i++) {
252         s->bat_bitmap[idx + i] = cpu_to_le32(s->data_end / s->off_multiplier);
253         s->data_end += s->tracks;
254         bitmap_set(s->bat_dirty_bmap,
255                    bat_entry_off(idx + i) / s->bat_dirty_block, 1);
256     }
257 
258     return bat2sect(s, idx) + sector_num % s->tracks;
259 }
260 
261 
262 static coroutine_fn int parallels_co_flush_to_os(BlockDriverState *bs)
263 {
264     BDRVParallelsState *s = bs->opaque;
265     unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block);
266     unsigned long bit;
267 
268     qemu_co_mutex_lock(&s->lock);
269 
270     bit = find_first_bit(s->bat_dirty_bmap, size);
271     while (bit < size) {
272         uint32_t off = bit * s->bat_dirty_block;
273         uint32_t to_write = s->bat_dirty_block;
274         int ret;
275 
276         if (off + to_write > s->header_size) {
277             to_write = s->header_size - off;
278         }
279         ret = bdrv_pwrite(bs->file, off, (uint8_t *)s->header + off,
280                           to_write);
281         if (ret < 0) {
282             qemu_co_mutex_unlock(&s->lock);
283             return ret;
284         }
285         bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1);
286     }
287     bitmap_zero(s->bat_dirty_bmap, size);
288 
289     qemu_co_mutex_unlock(&s->lock);
290     return 0;
291 }
292 
293 
294 static int coroutine_fn parallels_co_block_status(BlockDriverState *bs,
295                                                   bool want_zero,
296                                                   int64_t offset,
297                                                   int64_t bytes,
298                                                   int64_t *pnum,
299                                                   int64_t *map,
300                                                   BlockDriverState **file)
301 {
302     BDRVParallelsState *s = bs->opaque;
303     int count;
304 
305     assert(QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE));
306     qemu_co_mutex_lock(&s->lock);
307     offset = block_status(s, offset >> BDRV_SECTOR_BITS,
308                           bytes >> BDRV_SECTOR_BITS, &count);
309     qemu_co_mutex_unlock(&s->lock);
310 
311     *pnum = count * BDRV_SECTOR_SIZE;
312     if (offset < 0) {
313         return 0;
314     }
315 
316     *map = offset * BDRV_SECTOR_SIZE;
317     *file = bs->file->bs;
318     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
319 }
320 
321 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
322                                             int64_t sector_num, int nb_sectors,
323                                             QEMUIOVector *qiov, int flags)
324 {
325     BDRVParallelsState *s = bs->opaque;
326     uint64_t bytes_done = 0;
327     QEMUIOVector hd_qiov;
328     int ret = 0;
329 
330     assert(!flags);
331     qemu_iovec_init(&hd_qiov, qiov->niov);
332 
333     while (nb_sectors > 0) {
334         int64_t position;
335         int n, nbytes;
336 
337         qemu_co_mutex_lock(&s->lock);
338         position = allocate_clusters(bs, sector_num, nb_sectors, &n);
339         qemu_co_mutex_unlock(&s->lock);
340         if (position < 0) {
341             ret = (int)position;
342             break;
343         }
344 
345         nbytes = n << BDRV_SECTOR_BITS;
346 
347         qemu_iovec_reset(&hd_qiov);
348         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
349 
350         ret = bdrv_co_pwritev(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
351                               &hd_qiov, 0);
352         if (ret < 0) {
353             break;
354         }
355 
356         nb_sectors -= n;
357         sector_num += n;
358         bytes_done += nbytes;
359     }
360 
361     qemu_iovec_destroy(&hd_qiov);
362     return ret;
363 }
364 
365 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
366         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
367 {
368     BDRVParallelsState *s = bs->opaque;
369     uint64_t bytes_done = 0;
370     QEMUIOVector hd_qiov;
371     int ret = 0;
372 
373     qemu_iovec_init(&hd_qiov, qiov->niov);
374 
375     while (nb_sectors > 0) {
376         int64_t position;
377         int n, nbytes;
378 
379         qemu_co_mutex_lock(&s->lock);
380         position = block_status(s, sector_num, nb_sectors, &n);
381         qemu_co_mutex_unlock(&s->lock);
382 
383         nbytes = n << BDRV_SECTOR_BITS;
384 
385         qemu_iovec_reset(&hd_qiov);
386         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
387 
388         if (position < 0) {
389             if (bs->backing) {
390                 ret = bdrv_co_preadv(bs->backing, sector_num * BDRV_SECTOR_SIZE,
391                                      nbytes, &hd_qiov, 0);
392                 if (ret < 0) {
393                     break;
394                 }
395             } else {
396                 qemu_iovec_memset(&hd_qiov, 0, 0, nbytes);
397             }
398         } else {
399             ret = bdrv_co_preadv(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
400                                  &hd_qiov, 0);
401             if (ret < 0) {
402                 break;
403             }
404         }
405 
406         nb_sectors -= n;
407         sector_num += n;
408         bytes_done += nbytes;
409     }
410 
411     qemu_iovec_destroy(&hd_qiov);
412     return ret;
413 }
414 
415 
416 static int coroutine_fn parallels_co_check(BlockDriverState *bs,
417                                            BdrvCheckResult *res,
418                                            BdrvCheckMode fix)
419 {
420     BDRVParallelsState *s = bs->opaque;
421     int64_t size, prev_off, high_off;
422     int ret;
423     uint32_t i;
424     bool flush_bat = false;
425 
426     size = bdrv_getlength(bs->file->bs);
427     if (size < 0) {
428         res->check_errors++;
429         return size;
430     }
431 
432     qemu_co_mutex_lock(&s->lock);
433     if (s->header_unclean) {
434         fprintf(stderr, "%s image was not closed correctly\n",
435                 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
436         res->corruptions++;
437         if (fix & BDRV_FIX_ERRORS) {
438             /* parallels_close will do the job right */
439             res->corruptions_fixed++;
440             s->header_unclean = false;
441         }
442     }
443 
444     res->bfi.total_clusters = s->bat_size;
445     res->bfi.compressed_clusters = 0; /* compression is not supported */
446 
447     high_off = 0;
448     prev_off = 0;
449     for (i = 0; i < s->bat_size; i++) {
450         int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
451         if (off == 0) {
452             prev_off = 0;
453             continue;
454         }
455 
456         /* cluster outside the image */
457         if (off > size) {
458             fprintf(stderr, "%s cluster %u is outside image\n",
459                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
460             res->corruptions++;
461             if (fix & BDRV_FIX_ERRORS) {
462                 prev_off = 0;
463                 s->bat_bitmap[i] = 0;
464                 res->corruptions_fixed++;
465                 flush_bat = true;
466                 continue;
467             }
468         }
469 
470         res->bfi.allocated_clusters++;
471         if (off > high_off) {
472             high_off = off;
473         }
474 
475         if (prev_off != 0 && (prev_off + s->cluster_size) != off) {
476             res->bfi.fragmented_clusters++;
477         }
478         prev_off = off;
479     }
480 
481     ret = 0;
482     if (flush_bat) {
483         ret = bdrv_pwrite_sync(bs->file, 0, s->header, s->header_size);
484         if (ret < 0) {
485             res->check_errors++;
486             goto out;
487         }
488     }
489 
490     res->image_end_offset = high_off + s->cluster_size;
491     if (size > res->image_end_offset) {
492         int64_t count;
493         count = DIV_ROUND_UP(size - res->image_end_offset, s->cluster_size);
494         fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
495                 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
496                 size - res->image_end_offset);
497         res->leaks += count;
498         if (fix & BDRV_FIX_LEAKS) {
499             Error *local_err = NULL;
500 
501             /*
502              * In order to really repair the image, we must shrink it.
503              * That means we have to pass exact=true.
504              */
505             ret = bdrv_truncate(bs->file, res->image_end_offset, true,
506                                 PREALLOC_MODE_OFF, 0, &local_err);
507             if (ret < 0) {
508                 error_report_err(local_err);
509                 res->check_errors++;
510                 goto out;
511             }
512             res->leaks_fixed += count;
513         }
514     }
515 
516 out:
517     qemu_co_mutex_unlock(&s->lock);
518     return ret;
519 }
520 
521 
522 static int coroutine_fn parallels_co_create(BlockdevCreateOptions* opts,
523                                             Error **errp)
524 {
525     BlockdevCreateOptionsParallels *parallels_opts;
526     BlockDriverState *bs;
527     BlockBackend *blk;
528     int64_t total_size, cl_size;
529     uint32_t bat_entries, bat_sectors;
530     ParallelsHeader header;
531     uint8_t tmp[BDRV_SECTOR_SIZE];
532     int ret;
533 
534     assert(opts->driver == BLOCKDEV_DRIVER_PARALLELS);
535     parallels_opts = &opts->u.parallels;
536 
537     /* Sanity checks */
538     total_size = parallels_opts->size;
539 
540     if (parallels_opts->has_cluster_size) {
541         cl_size = parallels_opts->cluster_size;
542     } else {
543         cl_size = DEFAULT_CLUSTER_SIZE;
544     }
545 
546     /* XXX What is the real limit here? This is an insanely large maximum. */
547     if (cl_size >= INT64_MAX / MAX_PARALLELS_IMAGE_FACTOR) {
548         error_setg(errp, "Cluster size is too large");
549         return -EINVAL;
550     }
551     if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) {
552         error_setg(errp, "Image size is too large for this cluster size");
553         return -E2BIG;
554     }
555 
556     if (!QEMU_IS_ALIGNED(total_size, BDRV_SECTOR_SIZE)) {
557         error_setg(errp, "Image size must be a multiple of 512 bytes");
558         return -EINVAL;
559     }
560 
561     if (!QEMU_IS_ALIGNED(cl_size, BDRV_SECTOR_SIZE)) {
562         error_setg(errp, "Cluster size must be a multiple of 512 bytes");
563         return -EINVAL;
564     }
565 
566     /* Create BlockBackend to write to the image */
567     bs = bdrv_open_blockdev_ref(parallels_opts->file, errp);
568     if (bs == NULL) {
569         return -EIO;
570     }
571 
572     blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
573                           errp);
574     if (!blk) {
575         ret = -EPERM;
576         goto out;
577     }
578     blk_set_allow_write_beyond_eof(blk, true);
579 
580     /* Create image format */
581     bat_entries = DIV_ROUND_UP(total_size, cl_size);
582     bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
583     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
584 
585     memset(&header, 0, sizeof(header));
586     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
587     header.version = cpu_to_le32(HEADER_VERSION);
588     /* don't care much about geometry, it is not used on image level */
589     header.heads = cpu_to_le32(HEADS_NUMBER);
590     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE
591                                    / HEADS_NUMBER / SEC_IN_CYL);
592     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
593     header.bat_entries = cpu_to_le32(bat_entries);
594     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
595     header.data_off = cpu_to_le32(bat_sectors);
596 
597     /* write all the data */
598     memset(tmp, 0, sizeof(tmp));
599     memcpy(tmp, &header, sizeof(header));
600 
601     ret = blk_pwrite(blk, 0, tmp, BDRV_SECTOR_SIZE, 0);
602     if (ret < 0) {
603         goto exit;
604     }
605     ret = blk_pwrite_zeroes(blk, BDRV_SECTOR_SIZE,
606                             (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
607     if (ret < 0) {
608         goto exit;
609     }
610 
611     ret = 0;
612 out:
613     blk_unref(blk);
614     bdrv_unref(bs);
615     return ret;
616 
617 exit:
618     error_setg_errno(errp, -ret, "Failed to create Parallels image");
619     goto out;
620 }
621 
622 static int coroutine_fn parallels_co_create_opts(BlockDriver *drv,
623                                                  const char *filename,
624                                                  QemuOpts *opts,
625                                                  Error **errp)
626 {
627     BlockdevCreateOptions *create_options = NULL;
628     BlockDriverState *bs = NULL;
629     QDict *qdict;
630     Visitor *v;
631     int ret;
632 
633     static const QDictRenames opt_renames[] = {
634         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
635         { NULL, NULL },
636     };
637 
638     /* Parse options and convert legacy syntax */
639     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &parallels_create_opts,
640                                         true);
641 
642     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
643         ret = -EINVAL;
644         goto done;
645     }
646 
647     /* Create and open the file (protocol layer) */
648     ret = bdrv_create_file(filename, opts, errp);
649     if (ret < 0) {
650         goto done;
651     }
652 
653     bs = bdrv_open(filename, NULL, NULL,
654                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
655     if (bs == NULL) {
656         ret = -EIO;
657         goto done;
658     }
659 
660     /* Now get the QAPI type BlockdevCreateOptions */
661     qdict_put_str(qdict, "driver", "parallels");
662     qdict_put_str(qdict, "file", bs->node_name);
663 
664     v = qobject_input_visitor_new_flat_confused(qdict, errp);
665     if (!v) {
666         ret = -EINVAL;
667         goto done;
668     }
669 
670     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
671     visit_free(v);
672     if (!create_options) {
673         ret = -EINVAL;
674         goto done;
675     }
676 
677     /* Silently round up sizes */
678     create_options->u.parallels.size =
679         ROUND_UP(create_options->u.parallels.size, BDRV_SECTOR_SIZE);
680     create_options->u.parallels.cluster_size =
681         ROUND_UP(create_options->u.parallels.cluster_size, BDRV_SECTOR_SIZE);
682 
683     /* Create the Parallels image (format layer) */
684     ret = parallels_co_create(create_options, errp);
685     if (ret < 0) {
686         goto done;
687     }
688     ret = 0;
689 
690 done:
691     qobject_unref(qdict);
692     bdrv_unref(bs);
693     qapi_free_BlockdevCreateOptions(create_options);
694     return ret;
695 }
696 
697 
698 static int parallels_probe(const uint8_t *buf, int buf_size,
699                            const char *filename)
700 {
701     const ParallelsHeader *ph = (const void *)buf;
702 
703     if (buf_size < sizeof(ParallelsHeader)) {
704         return 0;
705     }
706 
707     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
708            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
709            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
710         return 100;
711     }
712 
713     return 0;
714 }
715 
716 static int parallels_update_header(BlockDriverState *bs)
717 {
718     BDRVParallelsState *s = bs->opaque;
719     unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
720                         sizeof(ParallelsHeader));
721 
722     if (size > s->header_size) {
723         size = s->header_size;
724     }
725     return bdrv_pwrite_sync(bs->file, 0, s->header, size);
726 }
727 
728 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
729                           Error **errp)
730 {
731     BDRVParallelsState *s = bs->opaque;
732     ParallelsHeader ph;
733     int ret, size, i;
734     QemuOpts *opts = NULL;
735     Error *local_err = NULL;
736     char *buf;
737 
738     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
739                                BDRV_CHILD_IMAGE, false, errp);
740     if (!bs->file) {
741         return -EINVAL;
742     }
743 
744     ret = bdrv_pread(bs->file, 0, &ph, sizeof(ph));
745     if (ret < 0) {
746         goto fail;
747     }
748 
749     bs->total_sectors = le64_to_cpu(ph.nb_sectors);
750 
751     if (le32_to_cpu(ph.version) != HEADER_VERSION) {
752         goto fail_format;
753     }
754     if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
755         s->off_multiplier = 1;
756         bs->total_sectors = 0xffffffff & bs->total_sectors;
757     } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
758         s->off_multiplier = le32_to_cpu(ph.tracks);
759     } else {
760         goto fail_format;
761     }
762 
763     s->tracks = le32_to_cpu(ph.tracks);
764     if (s->tracks == 0) {
765         error_setg(errp, "Invalid image: Zero sectors per track");
766         ret = -EINVAL;
767         goto fail;
768     }
769     if (s->tracks > INT32_MAX/513) {
770         error_setg(errp, "Invalid image: Too big cluster");
771         ret = -EFBIG;
772         goto fail;
773     }
774     s->cluster_size = s->tracks << BDRV_SECTOR_BITS;
775 
776     s->bat_size = le32_to_cpu(ph.bat_entries);
777     if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
778         error_setg(errp, "Catalog too large");
779         ret = -EFBIG;
780         goto fail;
781     }
782 
783     size = bat_entry_off(s->bat_size);
784     s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
785     s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
786     if (s->header == NULL) {
787         ret = -ENOMEM;
788         goto fail;
789     }
790     s->data_end = le32_to_cpu(ph.data_off);
791     if (s->data_end == 0) {
792         s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
793     }
794     if (s->data_end < s->header_size) {
795         /* there is not enough unused space to fit to block align between BAT
796            and actual data. We can't avoid read-modify-write... */
797         s->header_size = size;
798     }
799 
800     ret = bdrv_pread(bs->file, 0, s->header, s->header_size);
801     if (ret < 0) {
802         goto fail;
803     }
804     s->bat_bitmap = (uint32_t *)(s->header + 1);
805 
806     for (i = 0; i < s->bat_size; i++) {
807         int64_t off = bat2sect(s, i);
808         if (off >= s->data_end) {
809             s->data_end = off + s->tracks;
810         }
811     }
812 
813     if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
814         /* Image was not closed correctly. The check is mandatory */
815         s->header_unclean = true;
816         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
817             error_setg(errp, "parallels: Image was not closed correctly; "
818                        "cannot be opened read/write");
819             ret = -EACCES;
820             goto fail;
821         }
822     }
823 
824     opts = qemu_opts_create(&parallels_runtime_opts, NULL, 0, errp);
825     if (!opts) {
826         goto fail_options;
827     }
828 
829     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
830         goto fail_options;
831     }
832 
833     s->prealloc_size =
834         qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
835     s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
836     buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
837     /* prealloc_mode can be downgraded later during allocate_clusters */
838     s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf,
839                                        PRL_PREALLOC_MODE_FALLOCATE,
840                                        &local_err);
841     g_free(buf);
842     if (local_err != NULL) {
843         error_propagate(errp, local_err);
844         goto fail_options;
845     }
846 
847     if (ph.ext_off) {
848         if (flags & BDRV_O_RDWR) {
849             /*
850              * It's unsafe to open image RW if there is an extension (as we
851              * don't support it). But parallels driver in QEMU historically
852              * ignores the extension, so print warning and don't care.
853              */
854             warn_report("Format Extension ignored in RW mode");
855         } else {
856             ret = parallels_read_format_extension(
857                     bs, le64_to_cpu(ph.ext_off) << BDRV_SECTOR_BITS, errp);
858             if (ret < 0) {
859                 goto fail;
860             }
861         }
862     }
863 
864     if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_INACTIVE)) {
865         s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
866         ret = parallels_update_header(bs);
867         if (ret < 0) {
868             goto fail;
869         }
870     }
871 
872     s->bat_dirty_block = 4 * qemu_real_host_page_size;
873     s->bat_dirty_bmap =
874         bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block));
875 
876     /* Disable migration until bdrv_activate method is added */
877     error_setg(&s->migration_blocker, "The Parallels format used by node '%s' "
878                "does not support live migration",
879                bdrv_get_device_or_node_name(bs));
880     ret = migrate_add_blocker(s->migration_blocker, errp);
881     if (ret < 0) {
882         error_free(s->migration_blocker);
883         goto fail;
884     }
885     qemu_co_mutex_init(&s->lock);
886     return 0;
887 
888 fail_format:
889     error_setg(errp, "Image not in Parallels format");
890 fail_options:
891     ret = -EINVAL;
892 fail:
893     qemu_vfree(s->header);
894     return ret;
895 }
896 
897 
898 static void parallels_close(BlockDriverState *bs)
899 {
900     BDRVParallelsState *s = bs->opaque;
901 
902     if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) {
903         s->header->inuse = 0;
904         parallels_update_header(bs);
905 
906         /* errors are ignored, so we might as well pass exact=true */
907         bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS, true,
908                       PREALLOC_MODE_OFF, 0, NULL);
909     }
910 
911     g_free(s->bat_dirty_bmap);
912     qemu_vfree(s->header);
913 
914     migrate_del_blocker(s->migration_blocker);
915     error_free(s->migration_blocker);
916 }
917 
918 static BlockDriver bdrv_parallels = {
919     .format_name	= "parallels",
920     .instance_size	= sizeof(BDRVParallelsState),
921     .bdrv_probe		= parallels_probe,
922     .bdrv_open		= parallels_open,
923     .bdrv_close		= parallels_close,
924     .bdrv_child_perm          = bdrv_default_perms,
925     .bdrv_co_block_status     = parallels_co_block_status,
926     .bdrv_has_zero_init       = bdrv_has_zero_init_1,
927     .bdrv_co_flush_to_os      = parallels_co_flush_to_os,
928     .bdrv_co_readv  = parallels_co_readv,
929     .bdrv_co_writev = parallels_co_writev,
930     .is_format      = true,
931     .supports_backing = true,
932     .bdrv_co_create      = parallels_co_create,
933     .bdrv_co_create_opts = parallels_co_create_opts,
934     .bdrv_co_check  = parallels_co_check,
935     .create_opts    = &parallels_create_opts,
936 };
937 
938 static void bdrv_parallels_init(void)
939 {
940     bdrv_register(&bdrv_parallels);
941 }
942 
943 block_init(bdrv_parallels_init);
944