xref: /openbmc/qemu/block/parallels.c (revision 1141159c)
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 int64_t coroutine_fn GRAPH_RDLOCK
169 allocate_clusters(BlockDriverState *bs, 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 int coroutine_fn GRAPH_RDLOCK
265 parallels_co_flush_to_os(BlockDriverState *bs)
266 {
267     BDRVParallelsState *s = bs->opaque;
268     unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block);
269     unsigned long bit;
270 
271     qemu_co_mutex_lock(&s->lock);
272 
273     bit = find_first_bit(s->bat_dirty_bmap, size);
274     while (bit < size) {
275         uint32_t off = bit * s->bat_dirty_block;
276         uint32_t to_write = s->bat_dirty_block;
277         int ret;
278 
279         if (off + to_write > s->header_size) {
280             to_write = s->header_size - off;
281         }
282         ret = bdrv_co_pwrite(bs->file, off, to_write,
283                              (uint8_t *)s->header + off, 0);
284         if (ret < 0) {
285             qemu_co_mutex_unlock(&s->lock);
286             return ret;
287         }
288         bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1);
289     }
290     bitmap_zero(s->bat_dirty_bmap, size);
291 
292     qemu_co_mutex_unlock(&s->lock);
293     return 0;
294 }
295 
296 
297 static int coroutine_fn parallels_co_block_status(BlockDriverState *bs,
298                                                   bool want_zero,
299                                                   int64_t offset,
300                                                   int64_t bytes,
301                                                   int64_t *pnum,
302                                                   int64_t *map,
303                                                   BlockDriverState **file)
304 {
305     BDRVParallelsState *s = bs->opaque;
306     int count;
307 
308     assert(QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE));
309     qemu_co_mutex_lock(&s->lock);
310     offset = block_status(s, offset >> BDRV_SECTOR_BITS,
311                           bytes >> BDRV_SECTOR_BITS, &count);
312     qemu_co_mutex_unlock(&s->lock);
313 
314     *pnum = count * BDRV_SECTOR_SIZE;
315     if (offset < 0) {
316         return 0;
317     }
318 
319     *map = offset * BDRV_SECTOR_SIZE;
320     *file = bs->file->bs;
321     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
322 }
323 
324 static int coroutine_fn GRAPH_RDLOCK
325 parallels_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
326                     QEMUIOVector *qiov, int flags)
327 {
328     BDRVParallelsState *s = bs->opaque;
329     uint64_t bytes_done = 0;
330     QEMUIOVector hd_qiov;
331     int ret = 0;
332 
333     qemu_iovec_init(&hd_qiov, qiov->niov);
334 
335     while (nb_sectors > 0) {
336         int64_t position;
337         int n, nbytes;
338 
339         qemu_co_mutex_lock(&s->lock);
340         position = allocate_clusters(bs, sector_num, nb_sectors, &n);
341         qemu_co_mutex_unlock(&s->lock);
342         if (position < 0) {
343             ret = (int)position;
344             break;
345         }
346 
347         nbytes = n << BDRV_SECTOR_BITS;
348 
349         qemu_iovec_reset(&hd_qiov);
350         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
351 
352         ret = bdrv_co_pwritev(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
353                               &hd_qiov, 0);
354         if (ret < 0) {
355             break;
356         }
357 
358         nb_sectors -= n;
359         sector_num += n;
360         bytes_done += nbytes;
361     }
362 
363     qemu_iovec_destroy(&hd_qiov);
364     return ret;
365 }
366 
367 static int coroutine_fn GRAPH_RDLOCK
368 parallels_co_readv(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
369                    QEMUIOVector *qiov)
370 {
371     BDRVParallelsState *s = bs->opaque;
372     uint64_t bytes_done = 0;
373     QEMUIOVector hd_qiov;
374     int ret = 0;
375 
376     qemu_iovec_init(&hd_qiov, qiov->niov);
377 
378     while (nb_sectors > 0) {
379         int64_t position;
380         int n, nbytes;
381 
382         qemu_co_mutex_lock(&s->lock);
383         position = block_status(s, sector_num, nb_sectors, &n);
384         qemu_co_mutex_unlock(&s->lock);
385 
386         nbytes = n << BDRV_SECTOR_BITS;
387 
388         qemu_iovec_reset(&hd_qiov);
389         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes);
390 
391         if (position < 0) {
392             if (bs->backing) {
393                 ret = bdrv_co_preadv(bs->backing, sector_num * BDRV_SECTOR_SIZE,
394                                      nbytes, &hd_qiov, 0);
395                 if (ret < 0) {
396                     break;
397                 }
398             } else {
399                 qemu_iovec_memset(&hd_qiov, 0, 0, nbytes);
400             }
401         } else {
402             ret = bdrv_co_preadv(bs->file, position * BDRV_SECTOR_SIZE, nbytes,
403                                  &hd_qiov, 0);
404             if (ret < 0) {
405                 break;
406             }
407         }
408 
409         nb_sectors -= n;
410         sector_num += n;
411         bytes_done += nbytes;
412     }
413 
414     qemu_iovec_destroy(&hd_qiov);
415     return ret;
416 }
417 
418 
419 static int coroutine_fn GRAPH_RDLOCK
420 parallels_co_check(BlockDriverState *bs, BdrvCheckResult *res,
421                    BdrvCheckMode fix)
422 {
423     BDRVParallelsState *s = bs->opaque;
424     int64_t size, prev_off, high_off;
425     int ret;
426     uint32_t i;
427     bool flush_bat = false;
428 
429     size = bdrv_getlength(bs->file->bs);
430     if (size < 0) {
431         res->check_errors++;
432         return size;
433     }
434 
435     qemu_co_mutex_lock(&s->lock);
436     if (s->header_unclean) {
437         fprintf(stderr, "%s image was not closed correctly\n",
438                 fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR");
439         res->corruptions++;
440         if (fix & BDRV_FIX_ERRORS) {
441             /* parallels_close will do the job right */
442             res->corruptions_fixed++;
443             s->header_unclean = false;
444         }
445     }
446 
447     res->bfi.total_clusters = s->bat_size;
448     res->bfi.compressed_clusters = 0; /* compression is not supported */
449 
450     high_off = 0;
451     prev_off = 0;
452     for (i = 0; i < s->bat_size; i++) {
453         int64_t off = bat2sect(s, i) << BDRV_SECTOR_BITS;
454         if (off == 0) {
455             prev_off = 0;
456             continue;
457         }
458 
459         /* cluster outside the image */
460         if (off > size) {
461             fprintf(stderr, "%s cluster %u is outside image\n",
462                     fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
463             res->corruptions++;
464             if (fix & BDRV_FIX_ERRORS) {
465                 prev_off = 0;
466                 s->bat_bitmap[i] = 0;
467                 res->corruptions_fixed++;
468                 flush_bat = true;
469                 continue;
470             }
471         }
472 
473         res->bfi.allocated_clusters++;
474         if (off > high_off) {
475             high_off = off;
476         }
477 
478         if (prev_off != 0 && (prev_off + s->cluster_size) != off) {
479             res->bfi.fragmented_clusters++;
480         }
481         prev_off = off;
482     }
483 
484     ret = 0;
485     if (flush_bat) {
486         ret = bdrv_co_pwrite_sync(bs->file, 0, s->header_size, s->header, 0);
487         if (ret < 0) {
488             res->check_errors++;
489             goto out;
490         }
491     }
492 
493     res->image_end_offset = high_off + s->cluster_size;
494     if (size > res->image_end_offset) {
495         int64_t count;
496         count = DIV_ROUND_UP(size - res->image_end_offset, s->cluster_size);
497         fprintf(stderr, "%s space leaked at the end of the image %" PRId64 "\n",
498                 fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR",
499                 size - res->image_end_offset);
500         res->leaks += count;
501         if (fix & BDRV_FIX_LEAKS) {
502             Error *local_err = NULL;
503 
504             /*
505              * In order to really repair the image, we must shrink it.
506              * That means we have to pass exact=true.
507              */
508             ret = bdrv_co_truncate(bs->file, res->image_end_offset, true,
509                                    PREALLOC_MODE_OFF, 0, &local_err);
510             if (ret < 0) {
511                 error_report_err(local_err);
512                 res->check_errors++;
513                 goto out;
514             }
515             res->leaks_fixed += count;
516         }
517     }
518 
519 out:
520     qemu_co_mutex_unlock(&s->lock);
521     return ret;
522 }
523 
524 
525 static int coroutine_fn parallels_co_create(BlockdevCreateOptions* opts,
526                                             Error **errp)
527 {
528     BlockdevCreateOptionsParallels *parallels_opts;
529     BlockDriverState *bs;
530     BlockBackend *blk;
531     int64_t total_size, cl_size;
532     uint32_t bat_entries, bat_sectors;
533     ParallelsHeader header;
534     uint8_t tmp[BDRV_SECTOR_SIZE];
535     int ret;
536 
537     assert(opts->driver == BLOCKDEV_DRIVER_PARALLELS);
538     parallels_opts = &opts->u.parallels;
539 
540     /* Sanity checks */
541     total_size = parallels_opts->size;
542 
543     if (parallels_opts->has_cluster_size) {
544         cl_size = parallels_opts->cluster_size;
545     } else {
546         cl_size = DEFAULT_CLUSTER_SIZE;
547     }
548 
549     /* XXX What is the real limit here? This is an insanely large maximum. */
550     if (cl_size >= INT64_MAX / MAX_PARALLELS_IMAGE_FACTOR) {
551         error_setg(errp, "Cluster size is too large");
552         return -EINVAL;
553     }
554     if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) {
555         error_setg(errp, "Image size is too large for this cluster size");
556         return -E2BIG;
557     }
558 
559     if (!QEMU_IS_ALIGNED(total_size, BDRV_SECTOR_SIZE)) {
560         error_setg(errp, "Image size must be a multiple of 512 bytes");
561         return -EINVAL;
562     }
563 
564     if (!QEMU_IS_ALIGNED(cl_size, BDRV_SECTOR_SIZE)) {
565         error_setg(errp, "Cluster size must be a multiple of 512 bytes");
566         return -EINVAL;
567     }
568 
569     /* Create BlockBackend to write to the image */
570     bs = bdrv_co_open_blockdev_ref(parallels_opts->file, errp);
571     if (bs == NULL) {
572         return -EIO;
573     }
574 
575     blk = blk_co_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
576                              errp);
577     if (!blk) {
578         ret = -EPERM;
579         goto out;
580     }
581     blk_set_allow_write_beyond_eof(blk, true);
582 
583     /* Create image format */
584     bat_entries = DIV_ROUND_UP(total_size, cl_size);
585     bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size);
586     bat_sectors = (bat_sectors *  cl_size) >> BDRV_SECTOR_BITS;
587 
588     memset(&header, 0, sizeof(header));
589     memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic));
590     header.version = cpu_to_le32(HEADER_VERSION);
591     /* don't care much about geometry, it is not used on image level */
592     header.heads = cpu_to_le32(HEADS_NUMBER);
593     header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE
594                                    / HEADS_NUMBER / SEC_IN_CYL);
595     header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS);
596     header.bat_entries = cpu_to_le32(bat_entries);
597     header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE));
598     header.data_off = cpu_to_le32(bat_sectors);
599 
600     /* write all the data */
601     memset(tmp, 0, sizeof(tmp));
602     memcpy(tmp, &header, sizeof(header));
603 
604     ret = blk_co_pwrite(blk, 0, BDRV_SECTOR_SIZE, tmp, 0);
605     if (ret < 0) {
606         goto exit;
607     }
608     ret = blk_co_pwrite_zeroes(blk, BDRV_SECTOR_SIZE,
609                                (bat_sectors - 1) << BDRV_SECTOR_BITS, 0);
610     if (ret < 0) {
611         goto exit;
612     }
613 
614     ret = 0;
615 out:
616     blk_co_unref(blk);
617     bdrv_co_unref(bs);
618     return ret;
619 
620 exit:
621     error_setg_errno(errp, -ret, "Failed to create Parallels image");
622     goto out;
623 }
624 
625 static int coroutine_fn GRAPH_RDLOCK
626 parallels_co_create_opts(BlockDriver *drv, const char *filename,
627                          QemuOpts *opts, Error **errp)
628 {
629     BlockdevCreateOptions *create_options = NULL;
630     BlockDriverState *bs = NULL;
631     QDict *qdict;
632     Visitor *v;
633     int ret;
634 
635     static const QDictRenames opt_renames[] = {
636         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
637         { NULL, NULL },
638     };
639 
640     /* Parse options and convert legacy syntax */
641     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &parallels_create_opts,
642                                         true);
643 
644     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
645         ret = -EINVAL;
646         goto done;
647     }
648 
649     /* Create and open the file (protocol layer) */
650     ret = bdrv_co_create_file(filename, opts, errp);
651     if (ret < 0) {
652         goto done;
653     }
654 
655     bs = bdrv_co_open(filename, NULL, NULL,
656                       BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
657     if (bs == NULL) {
658         ret = -EIO;
659         goto done;
660     }
661 
662     /* Now get the QAPI type BlockdevCreateOptions */
663     qdict_put_str(qdict, "driver", "parallels");
664     qdict_put_str(qdict, "file", bs->node_name);
665 
666     v = qobject_input_visitor_new_flat_confused(qdict, errp);
667     if (!v) {
668         ret = -EINVAL;
669         goto done;
670     }
671 
672     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
673     visit_free(v);
674     if (!create_options) {
675         ret = -EINVAL;
676         goto done;
677     }
678 
679     /* Silently round up sizes */
680     create_options->u.parallels.size =
681         ROUND_UP(create_options->u.parallels.size, BDRV_SECTOR_SIZE);
682     create_options->u.parallels.cluster_size =
683         ROUND_UP(create_options->u.parallels.cluster_size, BDRV_SECTOR_SIZE);
684 
685     /* Create the Parallels image (format layer) */
686     ret = parallels_co_create(create_options, errp);
687     if (ret < 0) {
688         goto done;
689     }
690     ret = 0;
691 
692 done:
693     qobject_unref(qdict);
694     bdrv_co_unref(bs);
695     qapi_free_BlockdevCreateOptions(create_options);
696     return ret;
697 }
698 
699 
700 static int parallels_probe(const uint8_t *buf, int buf_size,
701                            const char *filename)
702 {
703     const ParallelsHeader *ph = (const void *)buf;
704 
705     if (buf_size < sizeof(ParallelsHeader)) {
706         return 0;
707     }
708 
709     if ((!memcmp(ph->magic, HEADER_MAGIC, 16) ||
710            !memcmp(ph->magic, HEADER_MAGIC2, 16)) &&
711            (le32_to_cpu(ph->version) == HEADER_VERSION)) {
712         return 100;
713     }
714 
715     return 0;
716 }
717 
718 static int parallels_update_header(BlockDriverState *bs)
719 {
720     BDRVParallelsState *s = bs->opaque;
721     unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs),
722                         sizeof(ParallelsHeader));
723 
724     if (size > s->header_size) {
725         size = s->header_size;
726     }
727     return bdrv_pwrite_sync(bs->file, 0, size, s->header, 0);
728 }
729 
730 static int parallels_open(BlockDriverState *bs, QDict *options, int flags,
731                           Error **errp)
732 {
733     BDRVParallelsState *s = bs->opaque;
734     ParallelsHeader ph;
735     int ret, size, i;
736     QemuOpts *opts = NULL;
737     Error *local_err = NULL;
738     char *buf;
739 
740     ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
741     if (ret < 0) {
742         return ret;
743     }
744 
745     ret = bdrv_pread(bs->file, 0, sizeof(ph), &ph, 0);
746     if (ret < 0) {
747         goto fail;
748     }
749 
750     bs->total_sectors = le64_to_cpu(ph.nb_sectors);
751 
752     if (le32_to_cpu(ph.version) != HEADER_VERSION) {
753         goto fail_format;
754     }
755     if (!memcmp(ph.magic, HEADER_MAGIC, 16)) {
756         s->off_multiplier = 1;
757         bs->total_sectors = 0xffffffff & bs->total_sectors;
758     } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) {
759         s->off_multiplier = le32_to_cpu(ph.tracks);
760     } else {
761         goto fail_format;
762     }
763 
764     s->tracks = le32_to_cpu(ph.tracks);
765     if (s->tracks == 0) {
766         error_setg(errp, "Invalid image: Zero sectors per track");
767         ret = -EINVAL;
768         goto fail;
769     }
770     if (s->tracks > INT32_MAX/513) {
771         error_setg(errp, "Invalid image: Too big cluster");
772         ret = -EFBIG;
773         goto fail;
774     }
775     s->cluster_size = s->tracks << BDRV_SECTOR_BITS;
776 
777     s->bat_size = le32_to_cpu(ph.bat_entries);
778     if (s->bat_size > INT_MAX / sizeof(uint32_t)) {
779         error_setg(errp, "Catalog too large");
780         ret = -EFBIG;
781         goto fail;
782     }
783 
784     size = bat_entry_off(s->bat_size);
785     s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs));
786     s->header = qemu_try_blockalign(bs->file->bs, s->header_size);
787     if (s->header == NULL) {
788         ret = -ENOMEM;
789         goto fail;
790     }
791     s->data_end = le32_to_cpu(ph.data_off);
792     if (s->data_end == 0) {
793         s->data_end = ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE);
794     }
795     if (s->data_end < s->header_size) {
796         /* there is not enough unused space to fit to block align between BAT
797            and actual data. We can't avoid read-modify-write... */
798         s->header_size = size;
799     }
800 
801     ret = bdrv_pread(bs->file, 0, s->header_size, s->header, 0);
802     if (ret < 0) {
803         goto fail;
804     }
805     s->bat_bitmap = (uint32_t *)(s->header + 1);
806 
807     for (i = 0; i < s->bat_size; i++) {
808         int64_t off = bat2sect(s, i);
809         if (off >= s->data_end) {
810             s->data_end = off + s->tracks;
811         }
812     }
813 
814     if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) {
815         /* Image was not closed correctly. The check is mandatory */
816         s->header_unclean = true;
817         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
818             error_setg(errp, "parallels: Image was not closed correctly; "
819                        "cannot be opened read/write");
820             ret = -EACCES;
821             goto fail;
822         }
823     }
824 
825     opts = qemu_opts_create(&parallels_runtime_opts, NULL, 0, errp);
826     if (!opts) {
827         goto fail_options;
828     }
829 
830     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
831         goto fail_options;
832     }
833 
834     s->prealloc_size =
835         qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0);
836     s->prealloc_size = MAX(s->tracks, s->prealloc_size >> BDRV_SECTOR_BITS);
837     buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE);
838     /* prealloc_mode can be downgraded later during allocate_clusters */
839     s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf,
840                                        PRL_PREALLOC_MODE_FALLOCATE,
841                                        &local_err);
842     g_free(buf);
843     if (local_err != NULL) {
844         error_propagate(errp, local_err);
845         goto fail_options;
846     }
847 
848     if (ph.ext_off) {
849         if (flags & BDRV_O_RDWR) {
850             /*
851              * It's unsafe to open image RW if there is an extension (as we
852              * don't support it). But parallels driver in QEMU historically
853              * ignores the extension, so print warning and don't care.
854              */
855             warn_report("Format Extension ignored in RW mode");
856         } else {
857             ret = parallels_read_format_extension(
858                     bs, le64_to_cpu(ph.ext_off) << BDRV_SECTOR_BITS, errp);
859             if (ret < 0) {
860                 goto fail;
861             }
862         }
863     }
864 
865     if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_INACTIVE)) {
866         s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC);
867         ret = parallels_update_header(bs);
868         if (ret < 0) {
869             goto fail;
870         }
871     }
872 
873     s->bat_dirty_block = 4 * qemu_real_host_page_size();
874     s->bat_dirty_bmap =
875         bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block));
876 
877     /* Disable migration until bdrv_activate method is added */
878     error_setg(&s->migration_blocker, "The Parallels format used by node '%s' "
879                "does not support live migration",
880                bdrv_get_device_or_node_name(bs));
881     ret = migrate_add_blocker(s->migration_blocker, errp);
882     if (ret < 0) {
883         error_free(s->migration_blocker);
884         goto fail;
885     }
886     qemu_co_mutex_init(&s->lock);
887     return 0;
888 
889 fail_format:
890     error_setg(errp, "Image not in Parallels format");
891 fail_options:
892     ret = -EINVAL;
893 fail:
894     qemu_vfree(s->header);
895     return ret;
896 }
897 
898 
899 static void parallels_close(BlockDriverState *bs)
900 {
901     BDRVParallelsState *s = bs->opaque;
902 
903     if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) {
904         s->header->inuse = 0;
905         parallels_update_header(bs);
906 
907         /* errors are ignored, so we might as well pass exact=true */
908         bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS, true,
909                       PREALLOC_MODE_OFF, 0, NULL);
910     }
911 
912     g_free(s->bat_dirty_bmap);
913     qemu_vfree(s->header);
914 
915     migrate_del_blocker(s->migration_blocker);
916     error_free(s->migration_blocker);
917 }
918 
919 static BlockDriver bdrv_parallels = {
920     .format_name	= "parallels",
921     .instance_size	= sizeof(BDRVParallelsState),
922     .bdrv_probe		= parallels_probe,
923     .bdrv_open		= parallels_open,
924     .bdrv_close		= parallels_close,
925     .bdrv_child_perm          = bdrv_default_perms,
926     .bdrv_co_block_status     = parallels_co_block_status,
927     .bdrv_has_zero_init       = bdrv_has_zero_init_1,
928     .bdrv_co_flush_to_os      = parallels_co_flush_to_os,
929     .bdrv_co_readv  = parallels_co_readv,
930     .bdrv_co_writev = parallels_co_writev,
931     .is_format      = true,
932     .supports_backing = true,
933     .bdrv_co_create      = parallels_co_create,
934     .bdrv_co_create_opts = parallels_co_create_opts,
935     .bdrv_co_check  = parallels_co_check,
936     .create_opts    = &parallels_create_opts,
937 };
938 
939 static void bdrv_parallels_init(void)
940 {
941     bdrv_register(&bdrv_parallels);
942 }
943 
944 block_init(bdrv_parallels_init);
945