xref: /openbmc/qemu/block/parallels.c (revision 40d6ee94)
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 "qapi/error.h"
33 #include "block/block_int.h"
34 #include "block/qdict.h"
35 #include "sysemu/block-backend.h"
36 #include "qemu/module.h"
37 #include "qemu/option.h"
38 #include "qapi/qmp/qdict.h"
39 #include "qapi/qobject-input-visitor.h"
40 #include "qapi/qapi-visit-block-core.h"
41 #include "qemu/bswap.h"
42 #include "qemu/bitmap.h"
43 #include "migration/blocker.h"
44 #include "parallels.h"
45 
46 /**************************************************************/
47 
48 #define HEADER_MAGIC "WithoutFreeSpace"
49 #define HEADER_MAGIC2 "WithouFreSpacExt"
50 #define HEADER_VERSION 2
51 #define HEADER_INUSE_MAGIC  (0x746F6E59)
52 #define MAX_PARALLELS_IMAGE_FACTOR (1ull << 32)
53 
54 static QEnumLookup prealloc_mode_lookup = {
55     .array = (const char *const[]) {
56         "falloc",
57         "truncate",
58     },
59     .size = PRL_PREALLOC_MODE__MAX
60 };
61 
62 #define PARALLELS_OPT_PREALLOC_MODE     "prealloc-mode"
63 #define PARALLELS_OPT_PREALLOC_SIZE     "prealloc-size"
64 
65 static QemuOptsList parallels_runtime_opts = {
66     .name = "parallels",
67     .head = QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts.head),
68     .desc = {
69         {
70             .name = PARALLELS_OPT_PREALLOC_SIZE,
71             .type = QEMU_OPT_SIZE,
72             .help = "Preallocation size on image expansion",
73             .def_value_str = "128M",
74         },
75         {
76             .name = PARALLELS_OPT_PREALLOC_MODE,
77             .type = QEMU_OPT_STRING,
78             .help = "Preallocation mode on image expansion "
79                     "(allowed values: falloc, truncate)",
80             .def_value_str = "falloc",
81         },
82         { /* end of list */ },
83     },
84 };
85 
86 static QemuOptsList parallels_create_opts = {
87     .name = "parallels-create-opts",
88     .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head),
89     .desc = {
90         {
91             .name = BLOCK_OPT_SIZE,
92             .type = QEMU_OPT_SIZE,
93             .help = "Virtual disk size",
94         },
95         {
96             .name = BLOCK_OPT_CLUSTER_SIZE,
97             .type = QEMU_OPT_SIZE,
98             .help = "Parallels image cluster size",
99             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE),
100         },
101         { /* end of list */ }
102     }
103 };
104 
105 
106 static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx)
107 {
108     return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier;
109 }
110 
111 static uint32_t bat_entry_off(uint32_t idx)
112 {
113     return sizeof(ParallelsHeader) + sizeof(uint32_t) * idx;
114 }
115 
116 static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num)
117 {
118     uint32_t index, offset;
119 
120     index = sector_num / s->tracks;
121     offset = sector_num % s->tracks;
122 
123     /* not allocated */
124     if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) {
125         return -1;
126     }
127     return bat2sect(s, index) + offset;
128 }
129 
130 static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num,
131         int nb_sectors)
132 {
133     int ret = s->tracks - sector_num % s->tracks;
134     return MIN(nb_sectors, ret);
135 }
136 
137 static int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
138                             int nb_sectors, int *pnum)
139 {
140     int64_t start_off = -2, prev_end_off = -2;
141 
142     *pnum = 0;
143     while (nb_sectors > 0 || start_off == -2) {
144         int64_t offset = seek_to_sector(s, sector_num);
145         int to_end;
146 
147         if (start_off == -2) {
148             start_off = offset;
149             prev_end_off = offset;
150         } else if (offset != prev_end_off) {
151             break;
152         }
153 
154         to_end = cluster_remainder(s, sector_num, nb_sectors);
155         nb_sectors -= to_end;
156         sector_num += to_end;
157         *pnum += to_end;
158 
159         if (offset > 0) {
160             prev_end_off += to_end;
161         }
162     }
163     return start_off;
164 }
165 
166 static int64_t allocate_clusters(BlockDriverState *bs, int64_t sector_num,
167                                  int nb_sectors, int *pnum)
168 {
169     int ret;
170     BDRVParallelsState *s = bs->opaque;
171     int64_t pos, space, idx, to_allocate, i, len;
172 
173     pos = block_status(s, sector_num, nb_sectors, pnum);
174     if (pos > 0) {
175         return pos;
176     }
177 
178     idx = sector_num / s->tracks;
179     to_allocate = DIV_ROUND_UP(sector_num + *pnum, s->tracks) - idx;
180 
181     /* This function is called only by parallels_co_writev(), which will never
182      * pass a sector_num at or beyond the end of the image (because the block
183      * layer never passes such a sector_num to that function). Therefore, idx
184      * is always below s->bat_size.
185      * block_status() will limit *pnum so that sector_num + *pnum will not
186      * exceed the image end. Therefore, idx + to_allocate cannot exceed
187      * s->bat_size.
188      * Note that s->bat_size is an unsigned int, therefore idx + to_allocate
189      * will always fit into a uint32_t. */
190     assert(idx < s->bat_size && idx + to_allocate <= s->bat_size);
191 
192     space = to_allocate * s->tracks;
193     len = bdrv_getlength(bs->file->bs);
194     if (len < 0) {
195         return len;
196     }
197     if (s->data_end + space > (len >> BDRV_SECTOR_BITS)) {
198         space += s->prealloc_size;
199         if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) {
200             ret = bdrv_pwrite_zeroes(bs->file,
201                                      s->data_end << BDRV_SECTOR_BITS,
202                                      space << BDRV_SECTOR_BITS, 0);
203         } else {
204             ret = bdrv_truncate(bs->file,
205                                 (s->data_end + space) << BDRV_SECTOR_BITS,
206                                 PREALLOC_MODE_OFF, NULL);
207         }
208         if (ret < 0) {
209             return ret;
210         }
211     }
212 
213     /* Try to read from backing to fill empty clusters
214      * FIXME: 1. previous write_zeroes may be redundant
215      *        2. most of data we read from backing will be rewritten by
216      *           parallels_co_writev. On aligned-to-cluster write we do not need
217      *           this read at all.
218      *        3. it would be good to combine write of data from backing and new
219      *           data into one write call */
220     if (bs->backing) {
221         int64_t nb_cow_sectors = to_allocate * s->tracks;
222         int64_t nb_cow_bytes = nb_cow_sectors << BDRV_SECTOR_BITS;
223         QEMUIOVector qiov =
224             QEMU_IOVEC_INIT_BUF(qiov, qemu_blockalign(bs, nb_cow_bytes),
225                                 nb_cow_bytes);
226 
227         ret = bdrv_co_preadv(bs->backing, idx * s->tracks * BDRV_SECTOR_SIZE,
228                              nb_cow_bytes, &qiov, 0);
229         if (ret < 0) {
230             qemu_vfree(qemu_iovec_buf(&qiov));
231             return ret;
232         }
233 
234         ret = bdrv_co_pwritev(bs->file, s->data_end * BDRV_SECTOR_SIZE,
235                               nb_cow_bytes, &qiov, 0);
236         qemu_vfree(qemu_iovec_buf(&qiov));
237         if (ret < 0) {
238             return ret;
239         }
240     }
241 
242     for (i = 0; i < to_allocate; i++) {
243         s->bat_bitmap[idx + i] = cpu_to_le32(s->data_end / s->off_multiplier);
244         s->data_end += s->tracks;
245         bitmap_set(s->bat_dirty_bmap,
246                    bat_entry_off(idx + i) / s->bat_dirty_block, 1);
247     }
248 
249     return bat2sect(s, idx) + sector_num % s->tracks;
250 }
251 
252 
253 static coroutine_fn int parallels_co_flush_to_os(BlockDriverState *bs)
254 {
255     BDRVParallelsState *s = bs->opaque;
256     unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block);
257     unsigned long bit;
258 
259     qemu_co_mutex_lock(&s->lock);
260 
261     bit = find_first_bit(s->bat_dirty_bmap, size);
262     while (bit < size) {
263         uint32_t off = bit * s->bat_dirty_block;
264         uint32_t to_write = s->bat_dirty_block;
265         int ret;
266 
267         if (off + to_write > s->header_size) {
268             to_write = s->header_size - off;
269         }
270         ret = bdrv_pwrite(bs->file, off, (uint8_t *)s->header + off,
271                           to_write);
272         if (ret < 0) {
273             qemu_co_mutex_unlock(&s->lock);
274             return ret;
275         }
276         bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1);
277     }
278     bitmap_zero(s->bat_dirty_bmap, size);
279 
280     qemu_co_mutex_unlock(&s->lock);
281     return 0;
282 }
283 
284 
285 static int coroutine_fn parallels_co_block_status(BlockDriverState *bs,
286                                                   bool want_zero,
287                                                   int64_t offset,
288                                                   int64_t bytes,
289                                                   int64_t *pnum,
290                                                   int64_t *map,
291                                                   BlockDriverState **file)
292 {
293     BDRVParallelsState *s = bs->opaque;
294     int count;
295 
296     assert(QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE));
297     qemu_co_mutex_lock(&s->lock);
298     offset = block_status(s, offset >> BDRV_SECTOR_BITS,
299                           bytes >> BDRV_SECTOR_BITS, &count);
300     qemu_co_mutex_unlock(&s->lock);
301 
302     *pnum = count * BDRV_SECTOR_SIZE;
303     if (offset < 0) {
304         return 0;
305     }
306 
307     *map = offset * BDRV_SECTOR_SIZE;
308     *file = bs->file->bs;
309     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
310 }
311 
312 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
313                                             int64_t sector_num, int nb_sectors,
314                                             QEMUIOVector *qiov, int flags)
315 {
316     BDRVParallelsState *s = bs->opaque;
317     uint64_t bytes_done = 0;
318     QEMUIOVector hd_qiov;
319     int ret = 0;
320 
321     assert(!flags);
322     qemu_iovec_init(&hd_qiov, qiov->niov);
323 
324     while (nb_sectors > 0) {
325         int64_t position;
326         int n, nbytes;
327 
328         qemu_co_mutex_lock(&s->lock);
329         position = allocate_clusters(bs, sector_num, nb_sectors, &n);
330         qemu_co_mutex_unlock(&s->lock);
331         if (position < 0) {
332             ret = (int)position;
333             break;
334         }
335 
336         nbytes = n << BDRV_SECTOR_BITS;
337 
338         qemu_iovec_reset(&hd_qiov);
339         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
340 
341         ret = bdrv_co_pwritev(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
342                               &hd_qiov, 0);
343         if (ret < 0) {
344             break;
345         }
346 
347         nb_sectors -= n;
348         sector_num += n;
349         bytes_done += nbytes;
350     }
351 
352     qemu_iovec_destroy(&hd_qiov);
353     return ret;
354 }
355 
356 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
357         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
358 {
359     BDRVParallelsState *s = bs->opaque;
360     uint64_t bytes_done = 0;
361     QEMUIOVector hd_qiov;
362     int ret = 0;
363 
364     qemu_iovec_init(&hd_qiov, qiov->niov);
365 
366     while (nb_sectors > 0) {
367         int64_t position;
368         int n, nbytes;
369 
370         qemu_co_mutex_lock(&s->lock);
371         position = block_status(s, sector_num, nb_sectors, &n);
372         qemu_co_mutex_unlock(&s->lock);
373 
374         nbytes = n << BDRV_SECTOR_BITS;
375 
376         qemu_iovec_reset(&hd_qiov);
377         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
378 
379         if (position < 0) {
380             if (bs->backing) {
381                 ret = bdrv_co_preadv(bs->backing, sector_num * BDRV_SECTOR_SIZE,
382                                      nbytes, &hd_qiov, 0);
383                 if (ret < 0) {
384                     break;
385                 }
386             } else {
387                 qemu_iovec_memset(&hd_qiov, 0, 0, nbytes);
388             }
389         } else {
390             ret = bdrv_co_preadv(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
391                                  &hd_qiov, 0);
392             if (ret < 0) {
393                 break;
394             }
395         }
396 
397         nb_sectors -= n;
398         sector_num += n;
399         bytes_done += nbytes;
400     }
401 
402     qemu_iovec_destroy(&hd_qiov);
403     return ret;
404 }
405 
406 
407 static int coroutine_fn parallels_co_check(BlockDriverState *bs,
408                                            BdrvCheckResult *res,
409                                            BdrvCheckMode fix)
410 {
411     BDRVParallelsState *s = bs->opaque;
412     int64_t size, prev_off, high_off;
413     int ret;
414     uint32_t i;
415     bool flush_bat = false;
416     int cluster_size = s->tracks << BDRV_SECTOR_BITS;
417 
418     size = bdrv_getlength(bs->file->bs);
419     if (size < 0) {
420         res->check_errors++;
421         return size;
422     }
423 
424     qemu_co_mutex_lock(&s->lock);
425     if (s->header_unclean) {
426         fprintf(stderr, "%s image was not closed correctly\n",
427                 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
428         res->corruptions++;
429         if (fix & BDRV_FIX_ERRORS) {
430             /* parallels_close will do the job right */
431             res->corruptions_fixed++;
432             s->header_unclean = false;
433         }
434     }
435 
436     res->bfi.total_clusters = s->bat_size;
437     res->bfi.compressed_clusters = 0; /* compression is not supported */
438 
439     high_off = 0;
440     prev_off = 0;
441     for (i = 0; i < s->bat_size; i++) {
442         int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
443         if (off == 0) {
444             prev_off = 0;
445             continue;
446         }
447 
448         /* cluster outside the image */
449         if (off > size) {
450             fprintf(stderr, "%s cluster %u is outside image\n",
451                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
452             res->corruptions++;
453             if (fix & BDRV_FIX_ERRORS) {
454                 prev_off = 0;
455                 s->bat_bitmap[i] = 0;
456                 res->corruptions_fixed++;
457                 flush_bat = true;
458                 continue;
459             }
460         }
461 
462         res->bfi.allocated_clusters++;
463         if (off > high_off) {
464             high_off = off;
465         }
466 
467         if (prev_off != 0 && (prev_off + cluster_size) != off) {
468             res->bfi.fragmented_clusters++;
469         }
470         prev_off = off;
471     }
472 
473     ret = 0;
474     if (flush_bat) {
475         ret = bdrv_pwrite_sync(bs->file, 0, s->header, s->header_size);
476         if (ret < 0) {
477             res->check_errors++;
478             goto out;
479         }
480     }
481 
482     res->image_end_offset = high_off + cluster_size;
483     if (size > res->image_end_offset) {
484         int64_t count;
485         count = DIV_ROUND_UP(size - res->image_end_offset, cluster_size);
486         fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
487                 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
488                 size - res->image_end_offset);
489         res->leaks += count;
490         if (fix & BDRV_FIX_LEAKS) {
491             Error *local_err = NULL;
492             ret = bdrv_truncate(bs->file, res->image_end_offset,
493                                 PREALLOC_MODE_OFF, &local_err);
494             if (ret < 0) {
495                 error_report_err(local_err);
496                 res->check_errors++;
497                 goto out;
498             }
499             res->leaks_fixed += count;
500         }
501     }
502 
503 out:
504     qemu_co_mutex_unlock(&s->lock);
505     return ret;
506 }
507 
508 
509 static int coroutine_fn parallels_co_create(BlockdevCreateOptions* opts,
510                                             Error **errp)
511 {
512     BlockdevCreateOptionsParallels *parallels_opts;
513     BlockDriverState *bs;
514     BlockBackend *blk;
515     int64_t total_size, cl_size;
516     uint32_t bat_entries, bat_sectors;
517     ParallelsHeader header;
518     uint8_t tmp[BDRV_SECTOR_SIZE];
519     int ret;
520 
521     assert(opts->driver == BLOCKDEV_DRIVER_PARALLELS);
522     parallels_opts = &opts->u.parallels;
523 
524     /* Sanity checks */
525     total_size = parallels_opts->size;
526 
527     if (parallels_opts->has_cluster_size) {
528         cl_size = parallels_opts->cluster_size;
529     } else {
530         cl_size = DEFAULT_CLUSTER_SIZE;
531     }
532 
533     /* XXX What is the real limit here? This is an insanely large maximum. */
534     if (cl_size >= INT64_MAX / MAX_PARALLELS_IMAGE_FACTOR) {
535         error_setg(errp, "Cluster size is too large");
536         return -EINVAL;
537     }
538     if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) {
539         error_setg(errp, "Image size is too large for this cluster size");
540         return -E2BIG;
541     }
542 
543     if (!QEMU_IS_ALIGNED(total_size, BDRV_SECTOR_SIZE)) {
544         error_setg(errp, "Image size must be a multiple of 512 bytes");
545         return -EINVAL;
546     }
547 
548     if (!QEMU_IS_ALIGNED(cl_size, BDRV_SECTOR_SIZE)) {
549         error_setg(errp, "Cluster size must be a multiple of 512 bytes");
550         return -EINVAL;
551     }
552 
553     /* Create BlockBackend to write to the image */
554     bs = bdrv_open_blockdev_ref(parallels_opts->file, errp);
555     if (bs == NULL) {
556         return -EIO;
557     }
558 
559     blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
560     ret = blk_insert_bs(blk, bs, errp);
561     if (ret < 0) {
562         goto out;
563     }
564     blk_set_allow_write_beyond_eof(blk, true);
565 
566     /* Create image format */
567     ret = blk_truncate(blk, 0, PREALLOC_MODE_OFF, errp);
568     if (ret < 0) {
569         goto out;
570     }
571 
572     bat_entries = DIV_ROUND_UP(total_size, cl_size);
573     bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
574     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
575 
576     memset(&header, 0, sizeof(header));
577     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
578     header.version = cpu_to_le32(HEADER_VERSION);
579     /* don't care much about geometry, it is not used on image level */
580     header.heads = cpu_to_le32(HEADS_NUMBER);
581     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE
582                                    / HEADS_NUMBER / SEC_IN_CYL);
583     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
584     header.bat_entries = cpu_to_le32(bat_entries);
585     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
586     header.data_off = cpu_to_le32(bat_sectors);
587 
588     /* write all the data */
589     memset(tmp, 0, sizeof(tmp));
590     memcpy(tmp, &header, sizeof(header));
591 
592     ret = blk_pwrite(blk, 0, tmp, BDRV_SECTOR_SIZE, 0);
593     if (ret < 0) {
594         goto exit;
595     }
596     ret = blk_pwrite_zeroes(blk, BDRV_SECTOR_SIZE,
597                             (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
598     if (ret < 0) {
599         goto exit;
600     }
601 
602     ret = 0;
603 out:
604     blk_unref(blk);
605     bdrv_unref(bs);
606     return ret;
607 
608 exit:
609     error_setg_errno(errp, -ret, "Failed to create Parallels image");
610     goto out;
611 }
612 
613 static int coroutine_fn parallels_co_create_opts(const char *filename,
614                                                  QemuOpts *opts,
615                                                  Error **errp)
616 {
617     BlockdevCreateOptions *create_options = NULL;
618     Error *local_err = NULL;
619     BlockDriverState *bs = NULL;
620     QDict *qdict;
621     Visitor *v;
622     int ret;
623 
624     static const QDictRenames opt_renames[] = {
625         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
626         { NULL, NULL },
627     };
628 
629     /* Parse options and convert legacy syntax */
630     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &parallels_create_opts,
631                                         true);
632 
633     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
634         ret = -EINVAL;
635         goto done;
636     }
637 
638     /* Create and open the file (protocol layer) */
639     ret = bdrv_create_file(filename, opts, &local_err);
640     if (ret < 0) {
641         error_propagate(errp, local_err);
642         goto done;
643     }
644 
645     bs = bdrv_open(filename, NULL, NULL,
646                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
647     if (bs == NULL) {
648         ret = -EIO;
649         goto done;
650     }
651 
652     /* Now get the QAPI type BlockdevCreateOptions */
653     qdict_put_str(qdict, "driver", "parallels");
654     qdict_put_str(qdict, "file", bs->node_name);
655 
656     v = qobject_input_visitor_new_flat_confused(qdict, errp);
657     if (!v) {
658         ret = -EINVAL;
659         goto done;
660     }
661 
662     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
663     visit_free(v);
664 
665     if (local_err) {
666         error_propagate(errp, local_err);
667         ret = -EINVAL;
668         goto done;
669     }
670 
671     /* Silently round up sizes */
672     create_options->u.parallels.size =
673         ROUND_UP(create_options->u.parallels.size, BDRV_SECTOR_SIZE);
674     create_options->u.parallels.cluster_size =
675         ROUND_UP(create_options->u.parallels.cluster_size, BDRV_SECTOR_SIZE);
676 
677     /* Create the Parallels image (format layer) */
678     ret = parallels_co_create(create_options, errp);
679     if (ret < 0) {
680         goto done;
681     }
682     ret = 0;
683 
684 done:
685     qobject_unref(qdict);
686     bdrv_unref(bs);
687     qapi_free_BlockdevCreateOptions(create_options);
688     return ret;
689 }
690 
691 
692 static int parallels_probe(const uint8_t *buf, int buf_size,
693                            const char *filename)
694 {
695     const ParallelsHeader *ph = (const void *)buf;
696 
697     if (buf_size < sizeof(ParallelsHeader)) {
698         return 0;
699     }
700 
701     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
702            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
703            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
704         return 100;
705     }
706 
707     return 0;
708 }
709 
710 static int parallels_update_header(BlockDriverState *bs)
711 {
712     BDRVParallelsState *s = bs->opaque;
713     unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
714                         sizeof(ParallelsHeader));
715 
716     if (size > s->header_size) {
717         size = s->header_size;
718     }
719     return bdrv_pwrite_sync(bs->file, 0, s->header, size);
720 }
721 
722 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
723                           Error **errp)
724 {
725     BDRVParallelsState *s = bs->opaque;
726     ParallelsHeader ph;
727     int ret, size, i;
728     QemuOpts *opts = NULL;
729     Error *local_err = NULL;
730     char *buf;
731 
732     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
733                                false, errp);
734     if (!bs->file) {
735         return -EINVAL;
736     }
737 
738     ret = bdrv_pread(bs->file, 0, &ph, sizeof(ph));
739     if (ret < 0) {
740         goto fail;
741     }
742 
743     bs->total_sectors = le64_to_cpu(ph.nb_sectors);
744 
745     if (le32_to_cpu(ph.version) != HEADER_VERSION) {
746         goto fail_format;
747     }
748     if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
749         s->off_multiplier = 1;
750         bs->total_sectors = 0xffffffff & bs->total_sectors;
751     } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
752         s->off_multiplier = le32_to_cpu(ph.tracks);
753     } else {
754         goto fail_format;
755     }
756 
757     s->tracks = le32_to_cpu(ph.tracks);
758     if (s->tracks == 0) {
759         error_setg(errp, "Invalid image: Zero sectors per track");
760         ret = -EINVAL;
761         goto fail;
762     }
763     if (s->tracks > INT32_MAX/513) {
764         error_setg(errp, "Invalid image: Too big cluster");
765         ret = -EFBIG;
766         goto fail;
767     }
768 
769     s->bat_size = le32_to_cpu(ph.bat_entries);
770     if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
771         error_setg(errp, "Catalog too large");
772         ret = -EFBIG;
773         goto fail;
774     }
775 
776     size = bat_entry_off(s->bat_size);
777     s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
778     s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
779     if (s->header == NULL) {
780         ret = -ENOMEM;
781         goto fail;
782     }
783     s->data_end = le32_to_cpu(ph.data_off);
784     if (s->data_end == 0) {
785         s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
786     }
787     if (s->data_end < s->header_size) {
788         /* there is not enough unused space to fit to block align between BAT
789            and actual data. We can't avoid read-modify-write... */
790         s->header_size = size;
791     }
792 
793     ret = bdrv_pread(bs->file, 0, s->header, s->header_size);
794     if (ret < 0) {
795         goto fail;
796     }
797     s->bat_bitmap = (uint32_t *)(s->header + 1);
798 
799     for (i = 0; i < s->bat_size; i++) {
800         int64_t off = bat2sect(s, i);
801         if (off >= s->data_end) {
802             s->data_end = off + s->tracks;
803         }
804     }
805 
806     if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
807         /* Image was not closed correctly. The check is mandatory */
808         s->header_unclean = true;
809         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
810             error_setg(errp, "parallels: Image was not closed correctly; "
811                        "cannot be opened read/write");
812             ret = -EACCES;
813             goto fail;
814         }
815     }
816 
817     opts = qemu_opts_create(&parallels_runtime_opts, NULL, 0, &local_err);
818     if (local_err != NULL) {
819         goto fail_options;
820     }
821 
822     qemu_opts_absorb_qdict(opts, options, &local_err);
823     if (local_err != NULL) {
824         goto fail_options;
825     }
826 
827     s->prealloc_size =
828         qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
829     s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
830     buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
831     s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf,
832                                        PRL_PREALLOC_MODE_FALLOCATE,
833                                        &local_err);
834     g_free(buf);
835     if (local_err != NULL) {
836         goto fail_options;
837     }
838 
839     if (!bdrv_has_zero_init(bs->file->bs)) {
840         s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
841     }
842 
843     if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_INACTIVE)) {
844         s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
845         ret = parallels_update_header(bs);
846         if (ret < 0) {
847             goto fail;
848         }
849     }
850 
851     s->bat_dirty_block = 4 * getpagesize();
852     s->bat_dirty_bmap =
853         bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block));
854 
855     /* Disable migration until bdrv_invalidate_cache method is added */
856     error_setg(&s->migration_blocker, "The Parallels format used by node '%s' "
857                "does not support live migration",
858                bdrv_get_device_or_node_name(bs));
859     ret = migrate_add_blocker(s->migration_blocker, &local_err);
860     if (local_err) {
861         error_propagate(errp, local_err);
862         error_free(s->migration_blocker);
863         goto fail;
864     }
865     qemu_co_mutex_init(&s->lock);
866     return 0;
867 
868 fail_format:
869     error_setg(errp, "Image not in Parallels format");
870     ret = -EINVAL;
871 fail:
872     qemu_vfree(s->header);
873     return ret;
874 
875 fail_options:
876     error_propagate(errp, local_err);
877     ret = -EINVAL;
878     goto fail;
879 }
880 
881 
882 static void parallels_close(BlockDriverState *bs)
883 {
884     BDRVParallelsState *s = bs->opaque;
885 
886     if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) {
887         s->header->inuse = 0;
888         parallels_update_header(bs);
889         bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS,
890                       PREALLOC_MODE_OFF, NULL);
891     }
892 
893     g_free(s->bat_dirty_bmap);
894     qemu_vfree(s->header);
895 
896     migrate_del_blocker(s->migration_blocker);
897     error_free(s->migration_blocker);
898 }
899 
900 static BlockDriver bdrv_parallels = {
901     .format_name	= "parallels",
902     .instance_size	= sizeof(BDRVParallelsState),
903     .bdrv_probe		= parallels_probe,
904     .bdrv_open		= parallels_open,
905     .bdrv_close		= parallels_close,
906     .bdrv_child_perm          = bdrv_format_default_perms,
907     .bdrv_co_block_status     = parallels_co_block_status,
908     .bdrv_has_zero_init       = bdrv_has_zero_init_1,
909     .bdrv_co_flush_to_os      = parallels_co_flush_to_os,
910     .bdrv_co_readv  = parallels_co_readv,
911     .bdrv_co_writev = parallels_co_writev,
912     .supports_backing = true,
913     .bdrv_co_create      = parallels_co_create,
914     .bdrv_co_create_opts = parallels_co_create_opts,
915     .bdrv_co_check  = parallels_co_check,
916     .create_opts    = &parallels_create_opts,
917 };
918 
919 static void bdrv_parallels_init(void)
920 {
921     bdrv_register(&bdrv_parallels);
922 }
923 
924 block_init(bdrv_parallels_init);
925