xref: /openbmc/qemu/block/parallels.c (revision 29b62a10)
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 int64_t block_status(BDRVParallelsState *s, int64_t sector_num,
140                             int nb_sectors, int *pnum)
141 {
142     int64_t start_off = -2, prev_end_off = -2;
143 
144     *pnum = 0;
145     while (nb_sectors > 0 || start_off == -2) {
146         int64_t offset = seek_to_sector(s, sector_num);
147         int to_end;
148 
149         if (start_off == -2) {
150             start_off = offset;
151             prev_end_off = offset;
152         } else if (offset != prev_end_off) {
153             break;
154         }
155 
156         to_end = cluster_remainder(s, sector_num, nb_sectors);
157         nb_sectors -= to_end;
158         sector_num += to_end;
159         *pnum += to_end;
160 
161         if (offset > 0) {
162             prev_end_off += to_end;
163         }
164     }
165     return start_off;
166 }
167 
168 static coroutine_fn int64_t allocate_clusters(BlockDriverState *bs,
169                                               int64_t sector_num,
170                                               int nb_sectors, int *pnum)
171 {
172     int ret = 0;
173     BDRVParallelsState *s = bs->opaque;
174     int64_t pos, space, idx, to_allocate, i, len;
175 
176     pos = block_status(s, sector_num, nb_sectors, pnum);
177     if (pos > 0) {
178         return pos;
179     }
180 
181     idx = sector_num / s->tracks;
182     to_allocate = DIV_ROUND_UP(sector_num + *pnum, s->tracks) - idx;
183 
184     /* This function is called only by parallels_co_writev(), which will never
185      * pass a sector_num at or beyond the end of the image (because the block
186      * layer never passes such a sector_num to that function). Therefore, idx
187      * is always below s->bat_size.
188      * block_status() will limit *pnum so that sector_num + *pnum will not
189      * exceed the image end. Therefore, idx + to_allocate cannot exceed
190      * s->bat_size.
191      * Note that s->bat_size is an unsigned int, therefore idx + to_allocate
192      * will always fit into a uint32_t. */
193     assert(idx < s->bat_size && idx + to_allocate <= s->bat_size);
194 
195     space = to_allocate * s->tracks;
196     len = bdrv_getlength(bs->file->bs);
197     if (len < 0) {
198         return len;
199     }
200     if (s->data_end + space > (len >> BDRV_SECTOR_BITS)) {
201         space += s->prealloc_size;
202         /*
203          * We require the expanded size to read back as zero. If the
204          * user permitted truncation, we try that; but if it fails, we
205          * force the safer-but-slower fallocate.
206          */
207         if (s->prealloc_mode == PRL_PREALLOC_MODE_TRUNCATE) {
208             ret = bdrv_co_truncate(bs->file,
209                                    (s->data_end + space) << BDRV_SECTOR_BITS,
210                                    false, PREALLOC_MODE_OFF,
211                                    BDRV_REQ_ZERO_WRITE, NULL);
212             if (ret == -ENOTSUP) {
213                 s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE;
214             }
215         }
216         if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) {
217             ret = bdrv_co_pwrite_zeroes(bs->file,
218                                         s->data_end << BDRV_SECTOR_BITS,
219                                         space << BDRV_SECTOR_BITS, 0);
220         }
221         if (ret < 0) {
222             return ret;
223         }
224     }
225 
226     /* Try to read from backing to fill empty clusters
227      * FIXME: 1. previous write_zeroes may be redundant
228      *        2. most of data we read from backing will be rewritten by
229      *           parallels_co_writev. On aligned-to-cluster write we do not need
230      *           this read at all.
231      *        3. it would be good to combine write of data from backing and new
232      *           data into one write call */
233     if (bs->backing) {
234         int64_t nb_cow_sectors = to_allocate * s->tracks;
235         int64_t nb_cow_bytes = nb_cow_sectors << BDRV_SECTOR_BITS;
236         void *buf = qemu_blockalign(bs, nb_cow_bytes);
237 
238         ret = bdrv_co_pread(bs->backing, idx * s->tracks * BDRV_SECTOR_SIZE,
239                             nb_cow_bytes, buf, 0);
240         if (ret < 0) {
241             qemu_vfree(buf);
242             return ret;
243         }
244 
245         ret = bdrv_co_pwrite(bs->file, s->data_end * BDRV_SECTOR_SIZE,
246                              nb_cow_bytes, buf, 0);
247         qemu_vfree(buf);
248         if (ret < 0) {
249             return ret;
250         }
251     }
252 
253     for (i = 0; i < to_allocate; i++) {
254         s->bat_bitmap[idx + i] = cpu_to_le32(s->data_end / s->off_multiplier);
255         s->data_end += s->tracks;
256         bitmap_set(s->bat_dirty_bmap,
257                    bat_entry_off(idx + i) / s->bat_dirty_block, 1);
258     }
259 
260     return bat2sect(s, idx) + sector_num % s->tracks;
261 }
262 
263 
264 static coroutine_fn int parallels_co_flush_to_os(BlockDriverState *bs)
265 {
266     BDRVParallelsState *s = bs->opaque;
267     unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block);
268     unsigned long bit;
269 
270     qemu_co_mutex_lock(&s->lock);
271 
272     bit = find_first_bit(s->bat_dirty_bmap, size);
273     while (bit < size) {
274         uint32_t off = bit * s->bat_dirty_block;
275         uint32_t to_write = s->bat_dirty_block;
276         int ret;
277 
278         if (off + to_write > s->header_size) {
279             to_write = s->header_size - off;
280         }
281         ret = bdrv_co_pwrite(bs->file, off, to_write,
282                              (uint8_t *)s->header + off, 0);
283         if (ret < 0) {
284             qemu_co_mutex_unlock(&s->lock);
285             return ret;
286         }
287         bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1);
288     }
289     bitmap_zero(s->bat_dirty_bmap, size);
290 
291     qemu_co_mutex_unlock(&s->lock);
292     return 0;
293 }
294 
295 
296 static int coroutine_fn parallels_co_block_status(BlockDriverState *bs,
297                                                   bool want_zero,
298                                                   int64_t offset,
299                                                   int64_t bytes,
300                                                   int64_t *pnum,
301                                                   int64_t *map,
302                                                   BlockDriverState **file)
303 {
304     BDRVParallelsState *s = bs->opaque;
305     int count;
306 
307     assert(QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE));
308     qemu_co_mutex_lock(&s->lock);
309     offset = block_status(s, offset >> BDRV_SECTOR_BITS,
310                           bytes >> BDRV_SECTOR_BITS, &count);
311     qemu_co_mutex_unlock(&s->lock);
312 
313     *pnum = count * BDRV_SECTOR_SIZE;
314     if (offset < 0) {
315         return 0;
316     }
317 
318     *map = offset * BDRV_SECTOR_SIZE;
319     *file = bs->file->bs;
320     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
321 }
322 
323 static coroutine_fn int parallels_co_writev(BlockDriverState *bs,
324                                             int64_t sector_num, int nb_sectors,
325                                             QEMUIOVector *qiov, int flags)
326 {
327     BDRVParallelsState *s = bs->opaque;
328     uint64_t bytes_done = 0;
329     QEMUIOVector hd_qiov;
330     int ret = 0;
331 
332     qemu_iovec_init(&hd_qiov, qiov->niov);
333 
334     while (nb_sectors > 0) {
335         int64_t position;
336         int n, nbytes;
337 
338         qemu_co_mutex_lock(&s->lock);
339         position = allocate_clusters(bs, sector_num, nb_sectors, &n);
340         qemu_co_mutex_unlock(&s->lock);
341         if (position < 0) {
342             ret = (int)position;
343             break;
344         }
345 
346         nbytes = n << BDRV_SECTOR_BITS;
347 
348         qemu_iovec_reset(&hd_qiov);
349         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
350 
351         ret = bdrv_co_pwritev(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
352                               &hd_qiov, 0);
353         if (ret < 0) {
354             break;
355         }
356 
357         nb_sectors -= n;
358         sector_num += n;
359         bytes_done += nbytes;
360     }
361 
362     qemu_iovec_destroy(&hd_qiov);
363     return ret;
364 }
365 
366 static coroutine_fn int parallels_co_readv(BlockDriverState *bs,
367         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
368 {
369     BDRVParallelsState *s = bs->opaque;
370     uint64_t bytes_done = 0;
371     QEMUIOVector hd_qiov;
372     int ret = 0;
373 
374     qemu_iovec_init(&hd_qiov, qiov->niov);
375 
376     while (nb_sectors > 0) {
377         int64_t position;
378         int n, nbytes;
379 
380         qemu_co_mutex_lock(&s->lock);
381         position = block_status(s, sector_num, nb_sectors, &n);
382         qemu_co_mutex_unlock(&s->lock);
383 
384         nbytes = n << BDRV_SECTOR_BITS;
385 
386         qemu_iovec_reset(&hd_qiov);
387         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
388 
389         if (position < 0) {
390             if (bs->backing) {
391                 ret = bdrv_co_preadv(bs->backing, sector_num * BDRV_SECTOR_SIZE,
392                                      nbytes, &hd_qiov, 0);
393                 if (ret < 0) {
394                     break;
395                 }
396             } else {
397                 qemu_iovec_memset(&hd_qiov, 0, 0, nbytes);
398             }
399         } else {
400             ret = bdrv_co_preadv(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
401                                  &hd_qiov, 0);
402             if (ret < 0) {
403                 break;
404             }
405         }
406 
407         nb_sectors -= n;
408         sector_num += n;
409         bytes_done += nbytes;
410     }
411 
412     qemu_iovec_destroy(&hd_qiov);
413     return ret;
414 }
415 
416 
417 static int coroutine_fn parallels_co_check(BlockDriverState *bs,
418                                            BdrvCheckResult *res,
419                                            BdrvCheckMode fix)
420 {
421     BDRVParallelsState *s = bs->opaque;
422     int64_t size, prev_off, high_off;
423     int ret;
424     uint32_t i;
425     bool flush_bat = false;
426 
427     size = bdrv_getlength(bs->file->bs);
428     if (size < 0) {
429         res->check_errors++;
430         return size;
431     }
432 
433     qemu_co_mutex_lock(&s->lock);
434     if (s->header_unclean) {
435         fprintf(stderr, "%s image was not closed correctly\n",
436                 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
437         res->corruptions++;
438         if (fix & BDRV_FIX_ERRORS) {
439             /* parallels_close will do the job right */
440             res->corruptions_fixed++;
441             s->header_unclean = false;
442         }
443     }
444 
445     res->bfi.total_clusters = s->bat_size;
446     res->bfi.compressed_clusters = 0; /* compression is not supported */
447 
448     high_off = 0;
449     prev_off = 0;
450     for (i = 0; i < s->bat_size; i++) {
451         int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
452         if (off == 0) {
453             prev_off = 0;
454             continue;
455         }
456 
457         /* cluster outside the image */
458         if (off > size) {
459             fprintf(stderr, "%s cluster %u is outside image\n",
460                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
461             res->corruptions++;
462             if (fix & BDRV_FIX_ERRORS) {
463                 prev_off = 0;
464                 s->bat_bitmap[i] = 0;
465                 res->corruptions_fixed++;
466                 flush_bat = true;
467                 continue;
468             }
469         }
470 
471         res->bfi.allocated_clusters++;
472         if (off > high_off) {
473             high_off = off;
474         }
475 
476         if (prev_off != 0 && (prev_off + s->cluster_size) != off) {
477             res->bfi.fragmented_clusters++;
478         }
479         prev_off = off;
480     }
481 
482     ret = 0;
483     if (flush_bat) {
484         ret = bdrv_co_pwrite_sync(bs->file, 0, s->header_size, s->header, 0);
485         if (ret < 0) {
486             res->check_errors++;
487             goto out;
488         }
489     }
490 
491     res->image_end_offset = high_off + s->cluster_size;
492     if (size > res->image_end_offset) {
493         int64_t count;
494         count = DIV_ROUND_UP(size - res->image_end_offset, s->cluster_size);
495         fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
496                 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
497                 size - res->image_end_offset);
498         res->leaks += count;
499         if (fix & BDRV_FIX_LEAKS) {
500             Error *local_err = NULL;
501 
502             /*
503              * In order to really repair the image, we must shrink it.
504              * That means we have to pass exact=true.
505              */
506             ret = bdrv_co_truncate(bs->file, res->image_end_offset, true,
507                                    PREALLOC_MODE_OFF, 0, &local_err);
508             if (ret < 0) {
509                 error_report_err(local_err);
510                 res->check_errors++;
511                 goto out;
512             }
513             res->leaks_fixed += count;
514         }
515     }
516 
517 out:
518     qemu_co_mutex_unlock(&s->lock);
519     return ret;
520 }
521 
522 
523 static int coroutine_fn parallels_co_create(BlockdevCreateOptions* opts,
524                                             Error **errp)
525 {
526     BlockdevCreateOptionsParallels *parallels_opts;
527     BlockDriverState *bs;
528     BlockBackend *blk;
529     int64_t total_size, cl_size;
530     uint32_t bat_entries, bat_sectors;
531     ParallelsHeader header;
532     uint8_t tmp[BDRV_SECTOR_SIZE];
533     int ret;
534 
535     assert(opts->driver == BLOCKDEV_DRIVER_PARALLELS);
536     parallels_opts = &opts->u.parallels;
537 
538     /* Sanity checks */
539     total_size = parallels_opts->size;
540 
541     if (parallels_opts->has_cluster_size) {
542         cl_size = parallels_opts->cluster_size;
543     } else {
544         cl_size = DEFAULT_CLUSTER_SIZE;
545     }
546 
547     /* XXX What is the real limit here? This is an insanely large maximum. */
548     if (cl_size >= INT64_MAX / MAX_PARALLELS_IMAGE_FACTOR) {
549         error_setg(errp, "Cluster size is too large");
550         return -EINVAL;
551     }
552     if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) {
553         error_setg(errp, "Image size is too large for this cluster size");
554         return -E2BIG;
555     }
556 
557     if (!QEMU_IS_ALIGNED(total_size, BDRV_SECTOR_SIZE)) {
558         error_setg(errp, "Image size must be a multiple of 512 bytes");
559         return -EINVAL;
560     }
561 
562     if (!QEMU_IS_ALIGNED(cl_size, BDRV_SECTOR_SIZE)) {
563         error_setg(errp, "Cluster size must be a multiple of 512 bytes");
564         return -EINVAL;
565     }
566 
567     /* Create BlockBackend to write to the image */
568     bs = bdrv_open_blockdev_ref(parallels_opts->file, errp);
569     if (bs == NULL) {
570         return -EIO;
571     }
572 
573     blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
574                           errp);
575     if (!blk) {
576         ret = -EPERM;
577         goto out;
578     }
579     blk_set_allow_write_beyond_eof(blk, true);
580 
581     /* Create image format */
582     bat_entries = DIV_ROUND_UP(total_size, cl_size);
583     bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
584     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
585 
586     memset(&header, 0, sizeof(header));
587     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
588     header.version = cpu_to_le32(HEADER_VERSION);
589     /* don't care much about geometry, it is not used on image level */
590     header.heads = cpu_to_le32(HEADS_NUMBER);
591     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE
592                                    / HEADS_NUMBER / SEC_IN_CYL);
593     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
594     header.bat_entries = cpu_to_le32(bat_entries);
595     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
596     header.data_off = cpu_to_le32(bat_sectors);
597 
598     /* write all the data */
599     memset(tmp, 0, sizeof(tmp));
600     memcpy(tmp, &header, sizeof(header));
601 
602     ret = blk_co_pwrite(blk, 0, BDRV_SECTOR_SIZE, tmp, 0);
603     if (ret < 0) {
604         goto exit;
605     }
606     ret = blk_co_pwrite_zeroes(blk, BDRV_SECTOR_SIZE,
607                                (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
608     if (ret < 0) {
609         goto exit;
610     }
611 
612     ret = 0;
613 out:
614     blk_unref(blk);
615     bdrv_unref(bs);
616     return ret;
617 
618 exit:
619     error_setg_errno(errp, -ret, "Failed to create Parallels image");
620     goto out;
621 }
622 
623 static int coroutine_fn parallels_co_create_opts(BlockDriver *drv,
624                                                  const char *filename,
625                                                  QemuOpts *opts,
626                                                  Error **errp)
627 {
628     BlockdevCreateOptions *create_options = NULL;
629     BlockDriverState *bs = NULL;
630     QDict *qdict;
631     Visitor *v;
632     int ret;
633 
634     static const QDictRenames opt_renames[] = {
635         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
636         { NULL, NULL },
637     };
638 
639     /* Parse options and convert legacy syntax */
640     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &parallels_create_opts,
641                                         true);
642 
643     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
644         ret = -EINVAL;
645         goto done;
646     }
647 
648     /* Create and open the file (protocol layer) */
649     ret = bdrv_co_create_file(filename, opts, errp);
650     if (ret < 0) {
651         goto done;
652     }
653 
654     bs = bdrv_open(filename, NULL, NULL,
655                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
656     if (bs == NULL) {
657         ret = -EIO;
658         goto done;
659     }
660 
661     /* Now get the QAPI type BlockdevCreateOptions */
662     qdict_put_str(qdict, "driver", "parallels");
663     qdict_put_str(qdict, "file", bs->node_name);
664 
665     v = qobject_input_visitor_new_flat_confused(qdict, errp);
666     if (!v) {
667         ret = -EINVAL;
668         goto done;
669     }
670 
671     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
672     visit_free(v);
673     if (!create_options) {
674         ret = -EINVAL;
675         goto done;
676     }
677 
678     /* Silently round up sizes */
679     create_options->u.parallels.size =
680         ROUND_UP(create_options->u.parallels.size, BDRV_SECTOR_SIZE);
681     create_options->u.parallels.cluster_size =
682         ROUND_UP(create_options->u.parallels.cluster_size, BDRV_SECTOR_SIZE);
683 
684     /* Create the Parallels image (format layer) */
685     ret = parallels_co_create(create_options, errp);
686     if (ret < 0) {
687         goto done;
688     }
689     ret = 0;
690 
691 done:
692     qobject_unref(qdict);
693     bdrv_unref(bs);
694     qapi_free_BlockdevCreateOptions(create_options);
695     return ret;
696 }
697 
698 
699 static int parallels_probe(const uint8_t *buf, int buf_size,
700                            const char *filename)
701 {
702     const ParallelsHeader *ph = (const void *)buf;
703 
704     if (buf_size < sizeof(ParallelsHeader)) {
705         return 0;
706     }
707 
708     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
709            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
710            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
711         return 100;
712     }
713 
714     return 0;
715 }
716 
717 static int parallels_update_header(BlockDriverState *bs)
718 {
719     BDRVParallelsState *s = bs->opaque;
720     unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
721                         sizeof(ParallelsHeader));
722 
723     if (size > s->header_size) {
724         size = s->header_size;
725     }
726     return bdrv_pwrite_sync(bs->file, 0, size, s->header, 0);
727 }
728 
729 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
730                           Error **errp)
731 {
732     BDRVParallelsState *s = bs->opaque;
733     ParallelsHeader ph;
734     int ret, size, i;
735     QemuOpts *opts = NULL;
736     Error *local_err = NULL;
737     char *buf;
738 
739     ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
740     if (ret < 0) {
741         return ret;
742     }
743 
744     ret = bdrv_pread(bs->file, 0, sizeof(ph), &ph, 0);
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_size, s->header, 0);
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